UNIVERSITY
PROCEEDINGS BOOK
Mechanical
Engineering
Proceedings
Editors
Ismail SARITAS
Omer Faruk BAY
Kemal TUTUNCU
Proceedings
Editors
Ismail SARITAS
Omer Faruk BAY
Kemal TUTUNCU
e-ISSN- 978-605-9119-74-0
www.aybilonline.com
ii
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
EDITORS :
Ismail SARITAS
Selcuk University, Turkey
Depertmant of Electric-Electronic Engineering, Faculty of Technology
Alaeddin Keykubat Campus 42031 Konya, Turkey
isaritas@selcuk.edu.tr
Kemal TUTUNCU
Selcuk University, Turkey
Depertmant of Electric-Electronic Engineering, Faculty of Technology
Alaeddin Keykubat Campus 42031 Konya, Turkey
ktutuncu@selcuk.edu.tr
ASSISTANT EDITORS :
Ilker Ali OZKAN
Selcuk University, Turkey
Depertmant of Computer Engineering, Faculty of Technology
Alaeddin Keykubat Campus 42031 Konya, Turkey
ilkerozkan@selcuk.edu.tr
Murat KOKLU
Selcuk University, Turkey
Depertmant of Computer Engineering, Faculty of Technology
Alaeddin Keykubat Campus 42031 Konya, Turkey
mkoklu@selcuk.edu.tr
Ali YASAR
Selcuk University, Turkey
Guneysinir Vocational School of Higher Education,
Computer Programming, Konya, Turkey
aliyasar@selcuk.edu.tr
iii
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
PREFACE
International Conference on Advanced Technology & Sciences (ICAT'16) has been organized in Konya,
Turkey on 01-03 September, 2016.
The aim of International Conference on Advanced Technology & Sciences is to provide a platform for
researchers and academics as well as practicing professionals from all over the world, to present their research
and professional development activities in Computer, Electric and Electronics, Energy and Mechatronics.This
conference provides opportunities for the different areas delegates to exchange new ideas and application
experiences face to face, to establish business or research relations and to find global partners for future
collaboration. We hope that the conference results constituted significant contribution to the knowledge in
these up to date scientific fields.
All full paper submissions have been double blind and peer reviewed and evaluated based on originality,
technical and/or research content/depth, correctness, relevance to conference, contributions, and readability.
Selected papers presented in the conference that match with the topics of the journals will be published in the
following journals:
International Journal of Intelligent Systems and Applications in Engineering (IJISAE)
International Journal of Applied Mathematics, Electronics and Computers (IJAMEC)
International Journal of Energy Applications and technology (IJEAT)
International Journal of Automotive Engineering and Technologies (IJAET)
At this conference, there are 700 paper submissions from 107 different universities. Each paper proposal
was evaluated by two reviewers and 416 of these were accepted for presentation. And finally, 366 papers will
be presented at our conference.
We are sure that, ICAT will be the flagship conference for researchers, students, and professionals in
the areas of Electrical and Computer Engineering, Biomedical Engineering, Energy and Manufacturing
Engineering and their applications from Turkey and around the world to disseminate their research
advancements and discoveries, to network and exchange ideas in order to strengthen existing partnerships and
foster new collaborations.
In particular we would like to thank Prof. Dr. Mustafa SAHIN, Rector of Selcuk University; Advanced
Technology and Sciences, Academic Publisher; International Journal of Intelligent Systems and Applications
in Engineering (IJISAE); International Journal of Applied Mathematics, Electronics and Computers
(IJAMEC); Konya Metropolitan Municipality, Province of Konya Culture and Tourism Directorate, Konya
Chamber of Industry, Konya Chamber of Commerce, Firdevs Patent and AYBIL Organization. They have
made a crucial contribution towards the success of this conference. Our thanks also go to the colleagues in our
conference office.
Looking forward to see you in next ICAT.
Ismail SARITAS - Omer Faruk BAY
Kemal TUTUNCU
Editors
iv
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
PROGRAMME COMMITTEES
HONORARY CHAIR :
Mustafa Sahin, Turkey
GENERAL CHAIR :
Ismail Saritas, Turkey
CO-CHAIR :
Omer Faruk Bay, Turkey
Necmettin Tarakcioglu, Turkey
I Unal Sert, Turkey
Murat Ciniviz, Turkey
Silyan Sibinov Arsov, Bulgaria
Kemal Tutuncu, Turkey
PUBLICATION CHAIR :
Murat Koklu, Turkey
Mehmet Akif Sahman, Turkey
Ali Yasar, Turkey
Mustafa Buber, Turkey
ORGANIZING COMMITTEE
Omer Faruk Bay, Turkey Alla Anohina Naumeca, Latvia
Necmettin Tarakcioglu, Turkey Hidayet Oguz, Turkey
Ismail Saritas, Turkey Sakir Tasdemir, Turkey
Kemal Tutuncu, Turkey Silyan Sibinov Arsov, Bulgaria
Ilker Ali Ozkan, Turkey Ridvan Saracoglu, Turkey
Murat Koklu, Turkey Murat Ciniviz, Turkey
Betim Cico, Macedonia Dimitris Dranidis, Greece
Leon Rothkrantz, Netherlands Mustafa Altin, Turkey
Angel Smrikarov, Bulgaria
vi
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TECHNICAL COMMITTEE
M Turan Demirci, Turkey Ali Yasar, Turkey
Selahattin Alan, Turkey Mehmet Akif Sahman, Turkey
Mustafa Nevzat Ornek, Turkey Esra Kaya, Turkey
Mustafa Buber, Turkey
vii
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Contents Pages
AN EFFICIENT RESOURCE MANAGEMENT IN CLOUD COMPUTING 1
A MODEL OF AUTOMATIC BLOCK REALLOCATION IN THE LAND CONSOLIDATION PROJECTS USING
7
ARTIFICIAL BEE COLONY ALGORITHM
A COMPARATIVE STUDY OF STATISTICAL AND ARTIFICIAL INTELLIGENCE BASED CLASSIFICATION
12
ALGORITHMS ON CENTRAL NERVOUS SYSTEM CANCER MICROARRAY GENE EXPRESSION DATA
ROUTE PLANNING FOR MULTIPLE UNMANNED AERIAL VEHICLES UAVS WITH PARALLEL GENETIC
16
ALGORITHM ON GPU USING CUDA
TRAINING ANFIS USING GENETIC ALGORITHM FOR DYNAMIC SYSTEMS IDENTIFICATION 23
DS-MAC: DEADLINE SENSITIVE MEDIUM ACCESS CONTROL PROTOCOL FOR DELAY TOLERANT
27
NETWORK BASED MULTIPLE UNMANNED AERIAL VEHICLE SYSTEMS
IMPORTANCE OF BUSINESS INTELLIGENCE SOLUTION ON DECISION-MAKING PROCESS OF
36
COMPANY’S
ELEVATOR SYSTEM A CASE STUDY OF COLOURED PETRI NETS 40
APPLICATION OF FUZZY LOGIC IN LAND CONSOLIDATION-CLASSIFICATION STUDIES 46
SIMULATION AND ANALYZES OF HETEROGENEOUS WSN CLUSTERING PROTOCOLS 51
ROBOTICS EDUCATION BASED ON AUGMENTED REALITY IN PRIMARY SCHOOLS 55
WAVELET BASED MEDICAL IMAGE WATERMARKING SCHEME FOR PATIENT INFORMATION
60
AUTHENTICITY
COMPARISON OF THE EFFECT OF UNSUPERVISED AND SUPERVISED DISCRETIZATION METHODS
65
ON CLASSIFICATION PROCESS
USING K MEANS AND K MEDOIDS METHODS FOR MULTIVARIATE MAPPING 70
AUTOMATIC VOICE AND SPEECH RECOGNITION SYSTEM FOR THE GERMAN LANGUAGE WITH
75
DEEP LEARNING METHODS
A SOFT-CONFIGURED MANAGEMENT SYSTEM FOR MICROCONTROLLER TRAINING KIT 81
COMPARISON OF IMAGE ROTATION AND FILTERING OPERATIONS IN TERMS OF IMAGE QUALITY
84
FACTORS
IMPLEMENTATION OF MAINLY USED EDGE DETECTION ALGORITHMS ON FPGA 88
INVESTIGATING THE EFFECTS OF FACIAL REGIONS TO AGE ESTIMATION 95
DETECTION AND CONDITION MONITORING APPROACH OF RAIL SWITCHES BASED ON IMAGE
99
PROCESSING IN RAILWAYS
TRUST MODELS IN WIRELESS SENSOR NETWORKS 106
A REVIEW OF AUTOMATIC TEXT SUMMARIZATION 110
A REVIEW OF MULTI-OBJECTIVE OPTIMIZATION 117
LONG RANGE WIRELESS POINT TO POINT LINK NETWORK ON 5 GHZ FREQUENCY BAND WITH
125
VOIP
CONDITION MONITORING APPROACH USING 3D MODELLING OF RAILWAY TRACKS WITH LASER
132
CAMERAS
DIFFERENT APPLE VARIETIES CLASSIFICATION USING KNN AND MLP ALGORITHMS 136
ROLE OF ETHICS IN INFORMATION SECURITY 141
KNOWLEDGE MINING APPROACH FOR HEALTHY MONITORING FROM PREGNANCY DATA WITH
145
BIG VOLUMES
ENERGY EFFICIENT RANDOM SELECTED CONSTANT CLUSTERING APPROACH FOR WIRELESS
149
SENSOR NETWORKS
NETWORK TRAFFIC CLASSIFICATION BY KERNEL BASED EXTREME LEARNING MACHINE 153
HYBRID BIOMETRIC SYSTEM USING IRIS AND SPEAKER RECOGNITION 158
MOBILE APPLICATION OF DRUG FOLLOW UP INFORMATION SYSTEM WITH DATA MATRIX
162
READER
A HYBRID ALGORITHM FOR AUTOMATED GUIDED VEHICLE ROUTING PROBLEM 166
THE USE OF INTELLIGENT WATER DROPS IWD FOR B SPLINE CURVE FITTING 169
MONITORING OF ANXIETY LIKE BEHAVIORS ON RATS WITH VIDEO TRACKING TECHNOLOGY 177
viii
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
A NEW APPROACH BASED ON IMAGE PROCESSING FOR DETECTION OF WEAR OF GUIDE RAIL
181
SURFACE IN ELEVATOR SYSTEMS
CHEMICAL ANALYSIS PROGRAM FOR THE TIME OF FLIGHT MASS SPECTROMETRY SYSTEM 187
A MODIFIED CUCKOO SEARCH USING DIFFERENT SEARCH STRATEGIES 191
VIDEO STREAM WITH WEBSOCKET ON RASPBERRY PI 2 196
COMPARISON OF SIMULATED ANNEALING AND GENETIC ALGORITHM APPROACHES ON
200
INTEGRATED PROCESS ROUTING AND SCHEDULING PROBLEM
TRANSMIT POWER CONTROL (TPC) ALGORITHM FOR LTE-A FEMTOCELL NETWORKS 204
CUSTOMER SATISFACTION USING DATA MINING APPROACH 208
A FAST DETECTION APPROACH FOR ROAD DEFECTS USING IMAGE PROCESSING 212
A VISION BASED DETECTION APPROACH FOR LEVEL CROSSING AND SWITCH IN RAILWAY 217
A HYBRID APPROACH FOR INDOOR POSITIONING 224
A NEW REAL TIME CONTROL APPROACH FOR TIME EFFICIENCY IN GROUP ELEVATOR CONTROL
228
SYSTEM
DEEP BELIEF NETWORK BASED BRAIN ACTIVITY CLASSIFICATION USING EEG FROM SLOW
233
CORTICAL POTENTIALS IN STROKE
ARRHYTHMIA CLASSIFICATION USING WAVEFORM ECG SIGNALS 240
EFFECTS OF SPECTRAL CLUSTERING ON DOCUMENT CATEGORIZATION USING DISTRIBUTED
246
TOOLS
THE CLASSIFICATION OF WHITE WINE AND RED WINE ACCORDING TO THEIR PHYSICOCHEMICAL
251
QUALITIES
ESTABLISHMENT OF FIBER OPTIC CABLING SYSTEM IN KIRKUK CITY BY USING ANT COLONY
255
OPTIMIZATION AND GENETIC ALGORITHM
PARALLELIZATION OF A HIERARCHICAL GRAPH BASED IMAGE SEGMENTATION USING OPENMP 260
A PERFORMANCE COMPARISON OF GRAPH COLORING ALGORITHMS 266
DESIGN OF AN INTERFACE FOR GENETIC ALGORITHM BASED OPTIMIZATION OF FUNCTIONS 274
A HYBRID GENETIC ALGORITHM FOR MOBILE ROBOT PATH PLANNING PROBLEM 278
DIFFERENT DUTY CYCLE RATIO AND BRIGHTNESS OF VISUAL STIMULI CHANGE TO STEADY STATE
282
VISUAL EVOKED POTENTIAL RESPONSE
TOPOLOGY CONTROL IN MOBILE WIRELESS SENSOR NETWORKS 286
BIG BANG-BIG CRUNCH OPTIMIZATION ALGORITHM FOR SOLVING THE UNCAPACITATED FACILITY
291
LOCATION PROBLEM
IMAGE MOSAICING BASED CONDITION MONITORING APPROACH FOR MULTI ROBOTS AT
297
PRODUCTION LINES IN INDUSTRIAL SYSTEMS
DEFINING CROWD MOVEMENT AS PARABOLA AND CLASSIFYING THESE DEFINITIONS 302
VIBRATION ANALYSIS FOR INDUCTION MOTORS WITH AN EXPERT SYSTEM 307
SECURITY EVALUATION OF IOS AND ANDROID 310
AN EVALUTION OF SOME INSTANT MESSAGING APPLICATIONS (SIGNAL, TELEGRAM, THREEMA)
314
IN TERMS OF SECURITY
PREPARING DIET LIST SUGGESTION WITH FUZZY EXPERT SYSTEM 321
SWITCHING ACCESS POINTS ON AND OFF FOR AN ENERGY EFFICIENT WIRELESS
326
COMMUNICATION
CLUSTERING OF MITOCHONDRIAL D-LOOP SEQUENCES USING SIMILARITY MATRIX, PCA AND K-
331
MEANS ALGORITHM
TEXTURE SEGMENTATION BASED ON GABOR FILTERS AND NEUTROSOPHIC GRAPH CUT 336
INTERNET OF THINGS A SURVEY 340
AN EXPERT SYSTEM FOR BORING TOOLS USED FOR MACHINING HOLES 348
MULTI-CORE COMPUTING APPLICATION FOR LYAPUNOV EXPONENTS ANALYSIS OF CHAOTIC
353
SYSTEMS
ESTIMATION OF TURKEY ELECTRIC ENERGY DEMAND UNTIL YEAR 2035 USING TLBO ALGORITHM 357
A REVIEW OF AUTOMATIC AUTHOR IDENTIFICATION TASK 363
ix
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
xi
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
xiii
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
EFFECT OF DENSITY AND BLEND RATES OF COMPOSITE PANELS REINFORCED WITH POLYESTER
1163
FIBRES TO TENSILE RESISTANCE
ELECTROCHEMICAL STUDIES ON POLY(3,4-ETHYLENEDIOXYTHIOPHENE) POLYMER AND ITS
1168
POTENTIAL APPLICATION IN ELECTROCHEMICAL CAPACITOR
THE EXPERIMENTAL AND THEORETICAL STUDIES OF THE CHARACTERISTICS OF LSPR PEAKS OF
METAL NANOPARTICLES CONTROLLED BY AR AMBIENT GAS PRESSURE FOR THE EFFICIENCY IN 1175
THE SOLAR CELL
INVESTIGATION OF THE EFFECT OF NANOCLAY INCLUSION ON CHARPY IMPACT BEHAVIOR OF
1179
THE GLASS FIBER REINFORCED COMPOSITE LAMINATES
THE EFFECT OF NANOCLAY ON TENSILE AND FLEXURAL BEHAVIOR OF GLASS FIBER REINFORCED
1183
COMPOSITE LAMINATES
DENSIFICATION AND MECHANICAL BEHAVIOR OF ALUMINA MATRIX NANO METAL COMPOSITES 1186
EFFECTS OF SI AND MN ON MACHINABILITY AND WEAR RESISTANCE OF AS91 AND AM90
1190
MAGNESIUM ALLOYS
INVESTIGATION OF METALLURGICAL AND MECHANICAL PROPERTIES OF WELDED REGION OF API
1195
5L X80 STEEL MERGED BY SUBMERGED ARC WELDING METHOD
CHARACTERISATION OF NITI - TI POWDERS PROCESSED BY MECHANICAL ALLOYING TECHNIQUE 1200
HYDROPHILICITY PROCEDURE OF AGAVA AMERICANA L FIBERS WITH ECOLOGIC METHODS 1205
TRADITIONAL PRODUCTION AND WEAR BEHAVIOR OF COMPOSITE MATERIALS REINFORCED
1209
WASTE
TO SELECT ALUMINUM AL ALLOYS FOR MANY APPLICATIONS WITH THE HELP OF AN EXPERT
1218
SYSTEM
MECHANICAL AND THERMO-MECHANICAL BUCKLING ANALYSES OF COMPOSITE CYLINDRICAL
1223
SHELLS
TERMO MECHANICAL VIBRATION OF FUNCTIONALLY GRADED NANO PLATES AND BEAMS BASED
1227
ON COUPLE STRESS THEORY
EXPERIMENTAL INVESTIGATION OF THE BENDING OF LOCALLY HEATED AZ91B MAGNESIUM
1232
ALLOY SHEET
DRYING OF IONIC HYDROGELS USING FLUORESCENCE TECHNIQUE 1237
WEIGHT REDUCTION STUDY ON BIW BY USING BORON ALLOYED HOT STAMPING SOLUTION OF
1242
FRONT FRAME RAIL INSTEAD CONVENTIONAL HSS OR AHSS APPLICATIONS
WELD CURRENT EFFECT ON THE TENSILE STRENGTH OF COLD ROLLED TWIP980 STEEL JOINTS
1247
WELDED BY RESISTANCE SPOT WELDING
THE PHASE COMPOSITION AND MICROSTRUCTURE OF CLINOPTILOLITE BY ADDITION OF AL
1252
POWDER
IMPROVING DIE FILLING UTILIZING BI-DIRECTIONAL FORGING PROCESS 1256
BURST FAILURE OF NANO SILICA ADDED 55 6 FILAMENT WOUND BFR EPOXY COMPOSITE PIPE 1261
POTENTIAL APPLICATIONS OF NANO-SILICA PARTICLES IN CIVIL ENGINEERING 1265
COMPARISON OF OPTICAL AND ELECTRICAL CONDUCTANCE AND REFRACTIVE INDEX VALUES OF
1270
THE PTCDI C5 SMALL MOLECULE FOR VARIOUS RELATIONS AND CONDITIONS
ANALYSIS OF REFRACTIVE INDEX DISPERSION AND RELATION BETWEEN OPTICAL BAND GAPS
1275
AND INDEXES OF REFRACTION OF PPTTPP NANOFIBER
ANALYZING THE EUTECTIC AL SI ALLOYS MODIFICATION USING AL10SR AND CUSN5 MASTER
1279
ALLOYS
DETECTION AND COUNTING OF EMBRYONIC STEM CELLS IN FLUORESCENCE MICROSCOPY
1283
IMAGES BY A FULLY AUTOMATIC METHOD
DEVELOPING SOME BIOLOGICAL SYSTEM OF NANO HYBRID PEEK/PVDF REINFORCED WITH NANO
1288
HA , ZRO2,Y2O3 FOR INTERNAL FIXATION OF BONE FRACTURE.
ACTIVE CONTOUR BASED DEVELOPMENTAL HIP DYSPLASIA DIAGNOSIS WITH GRAF METHOD 1299
STRUCTURAL ELEMENTAL AND MOLECULAR CHARACTERIZATION OF HUMAN ARTICULAR
1305
CARTILAGE
STIFFNESS ANALYSIS OF ABOVE KNEE PROSTHESIS 1310
xv
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
A VISUAL STIMULUS MODULE FOR P300 BASED BRAIN COMPUTER INTERFACES 1315
PULSE OXİMETER AND WIRELESS TELEMETRY FOR VENTILATION OXYGEN SUPPORT 1321
EVALUATION OF THE COMPOUND MUSCLE ACTION POTENTIAL IN DIAGNOSIS OF THE MILD
1327
CARPAL TUNNEL SYNDROME
FEATURE SELECTION ON MR IMAGES USING GENETIC ALGORITHM WITH SVM AND NAIVE BAYES
1333
CLASSIFIERS
SMART DEVICES BASED REMOTE PATIENT MONITORING SYSTEM 1338
DESIGN AND CONSTRUCTION OF A NOVEL MICRO-EXTRUSION SYSTEM FOR BIO-PRINTING
1342
APPLICATIONS
A GENUINE GLCM-BASED FEATURE EXTRACTION FOR BREAST TISSUE CLASSIFICATION ON
1347
MAMMOGRAMS
FOUR-HEADED ARROW SHAPED DUAL BAND PERFECT ABSORBERS FOR BIOSENSING
1353
APPLICATIONS
BIOMECHANICAL CONSIDERATIONS IN FRACTURE FIXATION USING ORTHOPAEDIC IMPLANTS 1357
THE ASSESSMENT OF TIME DOMAIN FEATURES FOR DETECTING SYMPTOMS OF DIABETIC
1362
RETINOPATHY
RECOGNITION OF COMMON LUNG SOUNDS USING PCA AND SUPPORT VECTOR MACHINES 1367
RFID TAG ANTENNA DESIGN IN DIFFERENT ISM BANDS FOR IMPLANT IDENTIFICATION 1372
CLASSIFICATION OF STRUCTURAL MRI FOR DETECTING ALZHEIMER S DISEASE 1376
EXAMINATION AND CLASSIFICATION OF ROBOTS USED AT THE HEALTH SECTOR, AND SOME
1380
RECOMMENDATIONS FOR DISADVANTAGED CIRCUMSTANCES
THE DIAGNOSIS AND ESTIMATE OF CHRONIC KIDNEY DISEASE USING THE MACHINE LEARNING
1384
METHODS
PID CONTROL PERFORMANCE IMPROVEMENT FOR A LIQUID LEVEL SYSTEM USING PARAMETER
1389
DESIGN
COMPARING THE THERMAL PERFORMANCE OF TRADITIONAL BUILDING AND REINFORCED
1396
CONCRETE BUILDING BASED ON TS 825 (THERMAL INSULATION REQUIREMENTS FOR BUILDINGS)
A MATLAB SIMULATION OF POSITIONING OF POLITICAL PARTIES IN THE OPINION SPACE 1403
ABOUT ONE APPROACH TO INTELLIGENT MANAGING OF HEALTH SPECIALISTS LABOR MARKET 1408
DETERMINING EFFECTS OF FLEXIBLE STRUCTURE AND NON-RANDOM ROAD IRREGULARITY ON
1413
PASSENGER COMFORT USING A QUARTER CAR MODEL
STATISTICAL PROCESS CONTROL OF ASH CONTENT FOR 10 0 5 MM COAL PRODUCT OF HEAVY
1418
MEDIUM CYCLONE
INVESTIGATION OF WINGLET SHAPE EFFECT ON THE DRAG AND LIFT FORCE OF AIRCRAFT WING 1424
BIOGEOGRAPHY-BASED OPTIMIZATION ALGORITHM FOR DESIGNING OF PLANAR STEEL FRAMES 1430
BLADE NUMBER EFFECT ON THE THRUST, TORQUE AND POWER OF PROPELLER 1435
GENERATION OF DIGITAL TERRAIN MODEL FROM UN-MANNED AERIAL VEHICLE IMAGE DATA 1439
USING NANOFIBERS IN FREQUENCY SELECTIVE SURFACES AS DIELECTRIC SUBSTRATE 1446
T0 EXTENDED PSEUDO-SEMI METRIC SPACES 1448
T0 REFLEXIVE SPACES 1452
REMOVAL OF CONGO RED ANIONIC DYE FROM AQUEOUS SOLUTIONS BY ADSORPTION ONTO
1455
OLIVE POMACE
COMPARISON OF ASSOCIATION ANALYSIS ALGORITHM IN SPSS CLEMENTINE AND AN
1458
APPLICATION
NARMA-L2 CONTROLLER BASED ON ONLINE SUPPORT VECTOR REGRESSION 1462
DEFINING THE PLANTATION ROLE TO MITIGATE THE URBAN HEAT ISLAND EFFECTS ON GLOBAL
1469
WARMING USING THERMAL SATELLITE SENSORS
AN INTEGRATED APPROACH FOR SUSTAINABLE SUPPLIER SELECTION IN FUZZY ENVIRONMENT 1473
THE MINIMIZATION OF TORQUE RIPPLES OF SEGMENTAL TYPE SWITCHED RELUCTANCE MOTOR
1479
BY PARTICLE SWARM OPTIMIZATION
A NEW SUBSPACE BASED SOLUTION TO BACKGROUND
1485
MODELLING AND CHANGE DETECTION
xvi
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
xvii
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Ascloud computing is gaining more recognition to the algorithmshave been designed so far, and the need for much
public, request for services to a given task within the virtual more efficient algorithm that, with other things, also fairly
environment of the cloud also increases.This paper proposed a allocate the loads across the system is high. Moreover, each
max-min algorithm liked technique with the aim ofdevelopinga developed algorithm has its own draw back and mostly
new framework that tends to balance the load that may be
performs better in one application than in others [12].
experienced due to the high demand of resources by a set of task
within the virtual environmentof the cloud computing ecosystem. Load balancing, the absence of which negatively rises some
issues (in performance, availability), is one of the primary
Keywords— Load balancing, Max-min algorithm, Makespan, challenges in cloud computing [12] [14]. For improvement
Min-min algorithm, Task Scheduling, Resource Allocation. upon the available solutions, of which min-max relatively
performs well [13], a modified and extensively improved
I. INTRODUCTION algorithm is formed.
The fast development in the area of computing gives users The other part of this paper is organized as follows: Section
of computer system the opportunity to have access as well as 2 discussed on some previous studies related to task
exploit the vast resources that are contained within the system. scheduling algorithms. In Section 3 and 4 we introduce some
Cloud Computing is seen as a new technology that adopt the concept with regard to the technique for task scheduling and
aspect of distributed computing and internet. The concept of resource allocation in cloud computing, section 4 discuss on
cloud is to allow a client or a customer to have access to the issues pertaining load balancing, mathematical
computing resources through the use of web services in a formulation, and the proposed algorithm.In section 5,
more efficient manner. Resources in the field of information mathematical simulation and results arediscussed, and lastly
and communicationtechnology are more or less the concluding remarks are given in Section 7.
fundamental elements that include some part of computer
systems, computer networks, software application and so on. II. RELATED WORK
Managing these resources involve controlling and limiting As research in cloud environment is increasing, task
access to the pool of resources that are been shared. This scheduling needs to be more scalable to the user demands.
brings the concept of an agreement between the resource [1][2] Uses a technique known as improved max-min
services providers and the clients otherwise known as Service algorithm and enhance max-min algorithm respectively with
Level Agreement (SLA), The idea behind the SLA is to the aim of distributing the load among the available resources.
restrict access to a given resources. [1] is a modification of [2] in which both uses the max-min
Cloud resource management may involve some policies task scheduling algorithm. This paper employs the technique
such as admission control, Quality of Service (QoS) which in [1] to propose another algorithm that will help in balancing
involves specification from SLA. Other policy which is the load across the virtual resources and to allow for scalability
main topic regarding this research is Load Balancing which when handing task with the aim of improving the performance
involves balancing the work load evenly among the cloud of the system. Load balancing over resources in the cloud
servers.The cloud services are provided to customers through environment is used toachieveminimum load when using
the use of virtual hardware, the services provided to the client resources, different methods are use to achieved such balance
such as IaaS, SaaS, and PaaS as shown in Fig. 1can be scale as stated in [7, 8, and 9]. Based on these methods, we take
up or down depending on the client’s level of usage and the interest in max-min algorithm and shows how load can be
SLA adopted . balanced across different resources in the cloud environment.
Revolutionary increase of users and demand for various
services parallel with the need for efficient resources usage III. CLOUD COMPUTING AND TASK SCHEDULING
reveal that load balancing should be done correctly and As the aspect of parallel and distributed computing
efficiently. The aimed of which is to optimize involves Cloud computingwhich is a collection of computers
resourceutilization, maximize throughput, minimize response that are interconnected and virtualized as one computing
time, and avoiding overload. Many load balancing
____________________________________________________________________________________________________________
1
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
resources, cloud client get access to the resources through the include storage, memory, network bandwidth, and virtual
SLA [11]. machine [3]. The resources can be virtualized and provisioned
As mention previously cloud computing offers software, from the existing physical resources in the cloud environment.
platform, and infrastructure as a service respectively. The The parameters that are virtualized include; the CPU, memory,
software as a service includes providing software such as Mail disk etc. The provisioning can be done by mapping these
(e.g. Gmail, Yahoo mail), social network sites, Google drive, virtualized resources to their corresponding physical ones.
and so on, to the customers or clients. The infrastructure as a Resource allocation in cloud computing is all about assigning
service deals with VM, storage, network, load balancer and so available resources to a needing cloud application. Dynamic
on as a service to the client and lastly the platform as a service resource management is seen as a very active research area in
deals with database like sql, oracle, web services, runtime (e.g. the field of cloud computing. The cloud computing resources
java) and so on as a service to the client. The clients get access costs vary depending upon the type of configuration for using
to these services through various devices as shown in the such resources. Therefore an efficient use of these resources is
figure below [3] [4]. considered as a prime interest for both the customer/client and
the cloud provider. Resource allocation in cloud computing
takes place in two levels [5];firstly load is balanced within the
physical machine whenever an application is uploaded and
secondly, request are assigned to a specific application if there
are multiple request for the resource.
Task
Min-Min Algorithm
Set of Task
scheduler This algorithm has all the relevant information needed in
advance. The algorithm uses some parameters to obtain the
R information it needs. Some of these parameters are; ETC
(Expected Time Compute), MET (Minimum Execution Time),
Fig. 2 -Task Scheduling for Different Resources MTC (Minimum Completion Time) etc. The Min-Min
algorithm selects a task with minimum completion time and
IV. VIRTUAL RESOURCES AND ALLOCATION maps it with a node with a minimum completion time [8].
As it is shown in Fig. 2 above, the users send task to the
cloud environment with different requirement to the cloud Max-Min Algorithm:
service providers. The requirement can be tasks with different Max-min algorithm chooses large task to be executed firstly
set of data size and processing power, the task scheduler will before executing small once [10]. This algorithm works
then match the tasks with available resources (virtual almost the same way as the Min-Min algorithm except in
resources) that are available. Max-Min the task with maximum value is selected from the
Resources in cloud computing cover all useful entities set of execution time of tasks and maps it to a node with
which can be use through the cloud platform. These resources
____________________________________________________________________________________________________________
2
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
minimum completion time. The ready time of the node is The Max-min algorithm allocated task tito resource vj such
updated by adding the execution time of the task [7, 8]. that large tasks have higher priority. For instance for a given
As the cloud users sends task to the cloud environment with large task, the max-min algorithm execute smaller task
different requirement to the cloud service providers. The concurrently while running large tasks. Therefore, the largest
requirement can be tasks with different set of data size and task determines the total makespan for other resources. The
processing power, the task scheduler will then match the tasks improved max-min algorithm is given below [2].
with available resources (virtual resources) that are available.
For all submitted tasks in Meta-task;
Some mathematical relations are given in [9] to analyze for all resources;
resources scheduling in cloud computing which are employed C = E +t
and used in this paper are given below. Find task t costs maximum execution time
Assign task t to resource v which gives minimum
completion time
The set of VMs V with their respective processing power is Remove task t from Meta-tasks set.
given as; Update t for selectedv .
Update c for all j.
While Meta-task not Empty
= | = 1,2, … (1) Find task t costs maximum execution time.
Assign task t to resource v which gives minimum
The set of tasks is also given as completion time
Remove Task t form Meta-tasks set.
Update t for selectedv .
= , | = 1,2, … (2) Update c for all j.
Where
____________________________________________________________________________________________________________
3
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
In the algorithm the total makespan is made to be a pivot 1 By given this criteria, the resources can be used in a
value for the first step and another pivot 2 value is assigned balanced manner and fewer resources can be used, the
during the second step of the execution. Then during the next remaining resources will not be involved to minimize the use
execution step the second pivot value and the completion time of such resources. The aim of the above algorithm when
of the current state are summed up together. If they are greater compared to the improved Max-Min algorithm is to make
than the first pivot value, then a new resource is allocated to effective used of the available resource during scheduling.
that task. The flowchart for the above pseudo code is given in the
figure (Fig. 3)
Start
Compute Expected
Execution Time of each
task
Delete task tk
Update EET of the meta-
task set
Meta-task No
not empty
Y N
Assign task to
If corresponding
Pivot2+EET<pivot resource
Stop
____________________________________________________________________________________________________________
4
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
T1 522 200
2
1 T4 T3
T2 1128 500
0
T3 430 300
R1 R2 R3
T4 340 410
T5 570 328 Fig. 4: Chart for Resource Allocation for Max-Min Algorithm
Table he table (table 2) below holds the processing speed In contrast to theproposed max-min scheduling algorithm
and the bandwidth of the resources on a network system. the figure fig. 4 below shows how the allocation is performed.
TABLE 2
RESOURCE PROCESSING SPEED AND BANDWIDTH
Processing speed
R Bandwidth (MbPS)
(MIPS)
R1 130 100
R2 266 120
R3 294 150
TABLE 3
EXPECTED EXECUTION TIME OF TASK
R1 R2 R3
T1 4.015 1.962 1.776 Fig. 5: Chart for Resource Allocation for proposed Max-Min Algorithm
T2 8.677 4.241 3.837 From the chart (fig. 4), the largest task T1 (the blue colour)
T3 3.308 1.617 1.463 has a maximum makespan of 8.031 and it’s scheduled to
T4 2.615 1.278 1.156 resource R1. The maximum makespan, is considered as the
maximum throughput for other resources the summation of
T5 4.389 2.143 1.938
the remaining task i.e. T4 (red), T3(green), T1 (purple) and T5
(dark purple) are not up to the maximum throughput therefore
From the tables above i.e. Table 3: Ti with maximum they are assigned to the second resource R2. This makes it
execution time is selected and then is assigned to the possible to balance different smaller tasks to run concurrently
corresponding resource Ri. Fig. 4 shows how the allocation on different resources across the system and also to use the
was performed based on the max-min idea, the task are resources wisely when needed. Another important factor
allocated to all the available resources within the scheduler, which is based on the on-demand characteristics of cloud
the processing time is measured in seconds. computing is that, the number of resources used is also
minimized and a resource can be put into use when there is a
demand for that resource. Based on the results obtained,
instead of assigning the load to the three resources, it’s
possible to assign the task to only two resources, thereby
increasing the efficiency of the system, thus we can have
many task running concurrently on some resources and other
resources can be put in use only when the need arise.
____________________________________________________________________________________________________________
5
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
VII. CONCLUSIONS
In conclusion Cloud Computing is an on-demand service,
therefore, efficient on-demand allocation of VM is needed. In
thispaper technique to handle on-demand allocation is
analysed and it proved to be effective. Allocation of resources
can be performed efficiently within a cloud environment by
balancing the load across the various virtual machine
resources, by employing an efficient technique for load
balancing such as the max-min algorithm that was used in this
paper.
The usage of max-min technique made it possible to handle
resources in an efficient and balanced manner. Thus, for a
better service to be experienced in a field of cloud computing,
a proper and efficient allocation techniques need to be adopted.
REFERENCES
[1] UpendraBhoi, Purvi N. Ramanuj. Enhance Max-Min Task Scheduling
Algorithm in Cloud Computing. International Journal of Application or
Innovation Engineering & Management. 2013.
[2] O. M. Elzeki, M. Z. ReshadandM. A. Elsoud. Improved Max-Min
Algorithm in Cloud Computing. International Journal of Computer
Applications (0975 – 8887) Volume 50 – No.12, July 2012.
[3] Y Yuan, W-Cai Liu. Efficient resource management for cloud
computing 2011.
[4] Ryan Knight, The new role of XML in cloud data integration Using
XML to integrate Salesforce data with enterprise applications. June
2009.
[5] R Shelke, R Rajani.Dynamic resource allocation in Cloud Computing.
2013.
[6] Ronak Pate, Sanjay Patel, Survey on Resource Allocation Strategies in
CloudComputing. International Journal of Computer Applications
(0975 – 8887) Volume 50 – No.12, July 2012
[7] S. SwaroopMoharana, D. Rajadeepan. Analysis of Load Balancer in
Cloud Computing. International Journal of Computer Science and
Engineering Vol.2 2013.
[8] D. Manan Shah, A. Amit Kariyani, L. Dipak Agrawal. Allocation of
Virtual Machines in Cloud Computing using Load Balancing
Algorithm. International Journal of Computer Science and Information
Technology & Security. Vol. 3 2013.
[9] Yichao Yang, Yanbo Zhou. Heuristic Scheduling Algorithms for
Allocation of Virtualized Network and Computing Resources. Journal
of Software Engineering and Application 2013.
[10] Pinal Salot , A survey of various scheduling algorithm in cloud
computing environment, IJRET | FEB 2013
[11] . Patel, Pankesh, Ajith H. Ranabahu, and Amit P. Sheth. "Service level
agreement in cloud computing." (2009).
[12] G. Gopinath, S. Vasudevan An in-depth analysis and study of Load
balancingtechniques in the cloud computing
environment.2ndInternational Symposium on Big Data and Cloud
Computing. 2015
[13] NayandeepSran, Navdeep Kaur. Comparative Analysis of Existing
Load Balancing Techniques in Cloud Computing.International Journal
of Engineering Science Invention.Vol. 2 Issue 1 2013. PP.60-63
[14] Garima Joshi, S.K. VermaA Review on Load Balancing Approach in
Cloud Computing.International Journal of Computer Applications
(0975 – 8887) Volume 119 – No.20, June 2015.
____________________________________________________________________________________________________________
6
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Equitably reallocating of blocks among land owners providing fewer, more compact, contiguous and larger land
has been one of the most important tasks in Land Consolidation parcels for individual owners [2,3]. Land consolidation (LC)
studies. This task has to be fairly solved among landholdings for is a highly complex for spatial planning process because it
a land. This complicated problem is difficult to solve using linear involves many tasks, actors (landowners, rural engineers, local
methods. Therefore, a method is needed to solve this non-linear
and regional administrations, consultants) [4]. Lands of
problem among land owners impartially. There are many
applications employing optimization algorithms for solving the landholdings in Turkey are jointly owned, fragmented, away
complicated and non-linear problems in literature. When we from each other and in irregular geometric shapes. As a
examine the literature, it is seen that Genetic Algorithm has been consequence of that, because of non-monolithic parcels,
only used to overcome the block reallocation problem. Artificial increasing in workforce demand, a great loss of time and un-
Bee Colony (ABC) algorithm is one of the optimization cultivated areas, agricultural landholdings can’t use modern
algorithms that have been used to solve the non-linear and farming methods. Therefore, agricultural productivity
complicated problems in literature. Furthermore, this method decreases constantly. Besides, agricultural parcels continue to
has better performance when it is compared with the other be fragmented because of inheritance provisions, sale, newly
optimization algorithms. In this study, we have aimed to fairly
opened roads and irrigation channels. This situation makes LC
reallocate the landholding areas to blocks in a land by developing
an algorithm using Artificial Bee Colony optimization method. projects necessary on agricultural areas [5]. One of the
When we develop the steps of the algorithm, we give priority to important objectives of studies in agricultural LC is to
landholdings preferences and places of fixed installations. Data increase agricultural cultivation and to eliminate the problems
tables have been arranged by taking land consolidation data of in the agricultural structure. Land reallocation is the most
DOT Village in Adiyaman, Turkey that into consideration. DOT important stage of LC studies and a tool to rearrange
Village land consolidation project includes 143 blocks and 225 proprietary rights. In reallocation of parcels and their
landholders. Consequently, we have introduced the steps of an emplacement on lands, there are some technical obligations
algorithm solving the block reallocation problem automatically such as farmer’s wishes, fixed installations, roads and
using ABC for a sample land. Also, we have observed the
irrigation network and the most suitable dimension, geometry,
applicability of the proposed method for automatic block
reallocation problem in this study. This study is a preliminary size and direction of parcels. These obligations can be easily
study helping us to develop software providing to automatically overcame in the LC projects by using new computer
solve complicated block reallocation problem in real time land technologies employing mathematical models [6].
consolidation process. LC projects consist of various steps. Within these steps,
land reallocation is the most important and a time-
Keywords— Artificial Bee Colony Algorithm, Land consuming stage since high number of criteria are considered.
Consolidation, Automatic Block Reallocation, Landholding, Land reallocation is crucial for social peace to conduct land
Optimization reallocation studies in such way to meet the demands of
I. INTRODUCTION landowners and also the principles of equity and justice
[7]. Many different solutions have been suggested for the land
Land consolidation (LC) is an important component of reallocation process. These are based on many mathematical
determining the country farmland rational development models; model based on block priority [8–10], special-purpose
utilizing and improving the sustainable use of land resources. interface program to perform land reallocation [11,12], a new
Traditionally, LC is the most favourable land management model approach in reallocation [13], application of fuzzy logic
approach for solving land fragmentation and it has been in land reallocation [7], using the analytic hierarchy process
applied in many countries around the World [1]. Land in land consolidation [14], using the planning and decision
consolidation aims to change land ownership by redistribution support system in land consolidation [1], theoretical
____________________________________________________________________________________________________________
7
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
framework and application of the land-redistribution modules blocks. To this end, optimization studies have been conducted
[15], using genetic algorithm to land reallocation of land for land reallocation process [7]. The remaining amount after
consolidation [16], a spatial genetic algorithm for automating deducting the share from the total land amounts of each
land partitioning [17], a preliminary study for design of landholding in land consolidation area is placed in blocks, if
automatic block reallocation algorithm with genetic algorithm possible, in a single piece and as parcels in proper geometrical
method [5], comparison of designed different land reallocation shapes. In this location process; preferences of farmers, fixed
models [6]. installations and provisions of related legislation must be
Although Artificial Bee Colony (ABC) algorithm has been taken into account. As for this situation, new parcels can be
recently improved, there are lots of studies using ABC in placed in blocks in many different ways. Therefore, multiple
literature. Karaboga and Basturk compared the Genetic options and solutions emerge in this process. The aim is to
Algorithm (GA), Particle Swarm Optimization (PSO), Particle find the most appropriate solution for our goals. For this
Swarm Inspired Evolutionary Algorithm (PS-EA) and reason, block reallocation problem can be defined as an
Artificial Bee Colony Algorithm (ABC) in the optimization of ‘optimization’ process [5].
some numerical functions and they reported that ABC was
better [18]. Aydin et al. were modified the ABC algorithm and B. Application Area
they developed the Incremental ABC algorithm in which The study area, Dot Village is located in the east of
population size was grown over time and local search was Adıyaman Province (Fig. 1). It is situated 12 km along from
applied [19]. Kahta district and 35 km along from Adıyaman city centre. Its
Heuristic methods like swarm intelligence have been altitude is 660 m. Settlement area is on the hill and Atatürk
employed in complicated problems that are difficult to solve. Dam Lake coast. Çataltepe east of the village, west Sarısu
These methods simulate the social behaviours of animals like village, north Arılı and Zeytin villages, on the south west
bird flocking or fish schooling. Artificial bee colony (ABC) Beşaltı hamlet is located in the south Büyükbey village.
and genetic algorithm (GA) are examples of the heuristic Settlement area is between 38.170 East longitude and 37.450
methods. Solutions of automatic block reallocation problems North latitudes. Dot village LC project area is 1605.84
in land consolidation studies are difficult and this problem has hectares. There are 225 landholding and 143 blocks in the LC
been solved by traditional methods such block priority based. project area. There are 352 cadastre parcels and 688 numbers
However, as artificial intelligent systems like the heuristic of shares in the LC project area. Preference of 3 different
methods develop, the solution of such complicated problems blocks is made for each share.
is getting easier. Genetic algorithm (GA) is one of
optimization algorithms. This optimization method was firstly
demonstrated for solving the automatic block reallocation
problem in land consolidation projects in some studies [5,20].
Moreover, it was concluded that the optimization methods
was more successful than traditional one to reallocate the
blocks in a land consolidation project [20]. Artificial bee
colony algorithm (ABC) is an optimization algorithm which
can solve non-linear and complicated problems. This method
was firstly proposed in 2005 by Derviş Karaboga [18,21].
When ABC is compared with the other optimization
algorithms, it is seen to produce good results for complicated
problems described in literature. [22]. In this study, we
propose an algorithmic model that can automatically solve the
block reallocation problem in consolidation of a land using
ABC. The steps of this model are described in results.
In this paper, we described the steps of an algorithm to Fig. 1 Study Area
solve the block reallocation problem among landholders for
the land found in DOT village, Adiyaman-Turkey using ABC. C. Data Tables and Constraints
ABC algorithm is preferred because it is swarm intelligence Two data has been prepared for block reallocation. The first
and said to be better than the other optimization algorithms. In data include areas of cadastral parcels, the first degree fields
Sect. 2, the proposed ABC algorithm and some materials are of cadastral parcels prepared according to the index map and
described; in Sect. 3 results and discussion are presented; in farmer preferences. According to the data; there are 225
Sect. 4 conclusion is given. landholdings, 352 cadastral parcels and 688 numbers of shares
in the LC project area. Total of the first degree areas in
II. MATERIALS AND METHODS cadastral parcels is 9,528,608.84 m2 in the LC project area.
A. Block Reallocation in Land Consolidation The second data include block areas and the first degree
block fields. According to the data; there are 143 blocks in the
The problem encountered in LC projects can be defined as LC project area. Total of the first degree areas of blocks is
reallocating “n” number of cadastral parcels to “m” number of
____________________________________________________________________________________________________________
8
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
9,528,608.84 m2 in the LC project area. Cadastral parcels and optimization problems. There are three groups of bees which
block plan of DOT village are displayed in Fig. 2 and Fig 3 are employed, onlookers and scouts bees in the artificial bee
respectively. colony model. Initially, colony is divided two parts. The first
The constraints of block reallocation process are in the part of the colony includes the employed bees and second part
following: consist the onlookers. In the artificial bee colony the numbers
1- The first degree areas of the parcels. of food sources are equal to the number of employed bees in
2- Three preferences for parcel shares of landholdings. the artificial colony. The employed bee becomes the scout
3- Fixed installations. after the related food source is exhausted by the bees. The
4- Defendant parcels and grasslands located in multiple pseudo-code of ABC algorithm is given in the following Fig.
blocks. 2 [18,21].
• INITIALIZE.
• REPEAT.
(Step-1)Place the employed bees on the food sources
in the memory;
(Step-2)Place the onlooker bees on the food sources
in the memory;
(Step-3)Send the scouts to the search area for
discovering new food sources.
• UNTIL (requirements are met).
Fig. 2 The main steps of ABC algorithm.
E. Application of Algorithm
Our model includes three external functions. These three
functions generate population according to the constraints
expressed in section 2.C (function 1), calculate the fitness of
population (function 2) and select best population (function 3)
respectively. A population in our model includes 100 rows
(ensembles) and columns as the number of landholder. Each
Fig. 2 Cadastral Parcels of DOT Village column stores block numbers in which one landholder area is
located. Each landholder areas are randomly distributed to the
blocks except fixed installations in each ensemble. Our model
proposes to select the best ensemble in the population.
In the algorithm process, we set the initial values such that
colony size (number of employed bees + number of onlooker
bees) was 100; maximum cycle number in order to terminate
the algorithm was 2000; number of parameters of the
objective function was 225; control parameter in order to
abandon the food source was 100 and number of the runs was
three.
Generating of population worked as an external function in
the model. This function was designed such as; areas of
landholders having installations were fixed to their blocks for
each ensemble in the population because the fixed
installations found in the block areas couldn’t be changed. So
this situation was a limitation in the block reallocation process.
Areas of landholders having no fixed installations were
reallocated to remaining blocks in the model. Areas of
landholders having no fixed installations were fixed to first
preferences and second preferences in the first and second
ensembles of populations respectively. Third ensembles of
Fig.3 Block Plan of DOT Village population were generated such as; areas of landholders
having second preferences but third preferences and having
D. Artificial Bee Colony Algorithm three preferences were jointly fixed to their preferred blocks
Artificial Bee Colony algorithm was firstly proposed as an in third ensembles. The rest of the ensembles of the
optimization algorithm by Karaboga in 2005 [18,21]. population were composed such that the blocks were
Karaboga artificially simulated the intelligent behaviours of
bees in a real bee colony to solve multidimensional
____________________________________________________________________________________________________________
9
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
randomly reallocated for landholders regardless of their Step 19. Control the amount of food source in a limit rate.
preferences. (Scout phase)
Fitness function worked as an external function in our Step 20. Go to step 8.
model too. When one ensemble in the population was tested Step 21. Go to step 4.
with the fitness function in our model, we calculated the
settled areas and assessed the remaining areas information in
m2 for each block. And, we calculated settlements status to
the first, second and third preferences for each landholder.
Lastly, arithmetic means of these outputs assessed the fitness
status for each ensembles of population.
____________________________________________________________________________________________________________
10
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
11
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— A variety of methods are used in order to classify The interest in working with the rapid advancement of DNA
cancer gene expression profiles based on microarray data. microarray technology is increasing day by day. These studies,
Especially, statistical methods such as Support Vector Machines is a comprehensive technology used in molecular biology and
(SVM), Decision Trees (DT) and Bayes are widely preferred to medicine. DNA microarray data analysis plays an important
classify on microarray cancer data. However, the statistical role the identification of genes associated with diseases such as
methods can often be inadequate to solve problems which are cancer. It can be calculated in high probability that any
based on particularly large-scale data such as microarray data. individual is ill or healthy by identifying disease-related genes.
Therefore, artificial intelligence-based methods have been used to
Therefore, high performance classification methods are very
classify on microarray data lately. We are interested in classifying
microarray cancer gene expression by using both artificial important for microarray data.
intelligence based methods and statistical methods. In this study, Statistical methods such as Support Vector Machines,
Multi-Layer Perceptron (MLP), Radial basis Function Network Decision Trees and Bayesian Network are the most frequently
(RBFNetwork) and Ant Colony Optimization Algorithm (ACO) used methods in microarray classification. However, these
have been used including statistical methods. The performances of methods can often be inadequate to solve problems which are
these classification methods have been tested with validation based on especially large-scale data. Therefore, it is important
methods such as v-fold cross validation. To reduce dimension of to develop methods that can be effective to solve such
DNA microarray gene expression has been used Correlation- problems.
based Feature Selection (CFS) technique. According to the results
In the last few decades, artificial Intelligence techniques are
obtained from experimental study, artificial intelligence-based
classification methods exhibit better results than the statistical methods which are commonly used and preferred to solve
methods. difficult problems. That’s why, we are interested in classifying
central nervous system microarray cancer gene expression by
Keywords— DNA microarray, classification, v-fold cross using artificial intelligence based classification methods
validation, feature selection, gene expression data including statistical methods in this study.
In the experimental analysis, the results of classification
I. INTRODUCTION obtained by artificial intelligence methods are compared with
Gene expression analysis of thousands of genes can be the results of classification obtained by statistical methods. The
performed with a technique called microchip thanks to rest of this paper is organized as follow: We describe the
innovations in technology and research [1]. DNA microarray is relevant methods in our comparison study part in Section 2. In
high intensity gene array and it makes possible to examine Section 3, we introduced our experimental dataset called
thousands of gene expression profile [2]. Microarray central nervous system cancer microarray gene expression
technology provides abundance of knowledge on expression dataset. We report the results of our experiments which are
levels of thousands genes that has been used for diagnostic and followed by statistical analysis and discussions in Section 4.
prognostic purposes for various types of diseases. Microarray Finally, we conclude the paper with an outlook to our future
technology is an invention that allows for detection of too many work about the control parameter optimization of classification
genes. This technique is also used in many fields including algorithms.
medicine primarily [3]. The activity of genes in patient and
II. METHODS
healthy cells that get same tissues can compare through DNA
microarray technology. This technique can help find genes A classification function consists of two parts. First part is
associated with disease. For example, it can be used to identify about selecting important features by using “feature selection”
a gene associated with a disease thanks to compare gene methods. Second part is about classifying data thanks to
expression level of healthy and diseased cells [3].
____________________________________________________________________________________________________________
12
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
classification methods. System component used in this study 2) One Rule (OneR): One-R or “One Rule (a Rule)” is a
are described below. simple algorithm proposed by Holt R.C [7]. This algorithm
produces a rule in training data for each feature and then rule
A. Feature Selection which has minimum error rate according to One-R is selected.
The number of features is usually very high in gene
3) J48 Decision Tree: J48 is the Weka implementation of
expression dataset. Therefore we need to reduce dimension on
the C4.5 algorithm, based on the ID3 algorithm. The main idea
dataset to make better classification. Feature selection is very
of this method is to generate decision trees by using the
important process to make classification with high accuracy on
information entropy. The method divides dataset by calculating
microarray cancer datasets. We examined various feature
the information gain of each attribute and attribute which
selection methods and preferred Correlation based Feature
provides the most benefit is used to make a decision [8].
Selection (CFS) because it made a successful choice among all
features. 4) Multi-Layer Perceptron (MLP): This model on which
Rumelhart and his friends worked together is called error
1) Correlation based Feature Selection (CFS): CFS is a
propagation model or the back-propagation model
simple filter algorithm that ranks feature subsets and discovers
(backpropagation of the network) [9]. There are one or more
the merit of feature or subset of features according to a
hidden layer excluding an input and output layer in this model.
correlation based function. According to this approach, subsets
Neurons in layers is associated with other layers. Information
which has the best attributes consist of attributes which have a
flow direction is forward and there is no feedback on the
high correlation with the corresponding class label and have
network in MLP. Therefore it is known as feed forward neural
low correlation with each other. The rest of features should be
network. Data is not processed in neurons in the input layer.
ignored. CFS feature subset evaluation function is shown as
The number of neurons in this layer depends on the number of
follows:
dimensions of the problem to be applied to the network. The
k r ci number of hidden layer and neuron are randomly determined.
G s
(1) The number of neurons in the output layer depends on the type
k k (k 1)r ii of problem [10].
5) Radial basis Function Network (RBFNetwork):
where G S
is the heuristic merit of a feature subset S RBFNetwork was revealed in 1988 by inspiring the behavior of
biological neurons [11]. Training of this model can be
containing k features, r ci
is the mean feature–class compared to curve fitting approach in multidimensional space
[12]. It is used radial basis activation functions the transition
correlation, and r ii
, is the average feature-feature
from the input layer to the intermediate layer unlike other
intercorrelation. This equation is, in fact, Pearson's correlation, neural network structure in RBFNetwork and a non-linear
where all variables have been standardized [4]. clustering (cluster) analysis is performed. There are three layers
called input, hidden and output on RBFNetwork like general
B. Classification ANN structure. The structure between the intermediate and
1) Support Vector Machine (SVM): SVM is a statistical output layer is also same in other types of Artificial Neural
algorithm found by V. Vapnik [5] in the late 1960s. It is a Networks, and training is performed among neurons which are
method which is used particularly in classification microarray intermediate and output layer.
gene expression levels. SVM is a supervised classification 6) Ant Colony Optimization Algorithm (ACO) : ACO is a
algorithm based on statistical learning theory. SVM initially proposed algorithm inspired by the behavior of real ants in
had designed for two-class classification of linear data and then
nature [13]. This algorithm was proposed by utilizing the ability
was generalized to classify non-linear and multi-class data. The
to find the shortest path between nest and food source of the
working principle of SVM is based on predicting the optimal
ants [14]. The use of this algorithm for solution of classification
decision function that can separate two classes from each other, problems has been just used and Ant-Miner has been proposed
in other words, the most appropriate way to define SVM is the by Parpinelli and his friends [15] to classify data recently. After
hyper-plane, which can separate two classes each other [6]. An
that, Ant-Miner was developed by Liu and his friends [16] by
infinite number of non-optimal hyper-plane can be drawn to
the same logic, but different heuristic and pheromone update
split the two sets from each other. However, SVM try to find
strategy. cAnt-Miner algorithm was also developed for data
optimal hyper-plane that provides the maximum margin to
which has continuous values. High-level pseudocode of cAnt-
separate the two sets from each other. There are linear and Miner algorithm is shown Figure 1. We used cAnt-Miner
nonlinear forms of classification based on kernel function. algorithm because central nervous system microarray gene
Radial basis kernel function is frequently used in classification
expression dataset we use in this study has continuous values.
applications. SVM is an important classification method
because it is fast and especially perform good results on high
dimensional attributes.
____________________________________________________________________________________________________________
13
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
120
100
80 85,29 97,06 94,12 92,5
60 73,53 73,53
40
20
Fig. 1 High level pseudocode of cAnt-Miner
0
III. DATA DESCRIPTION
Central Nervous System Cancer dataset, provided by
Pomeroy [17], contains the expression levels of 7129 genes.
Each sample was obtained from brain tissues and was analyzed
using Affymetrix microarrays. This dataset contains two
Fig. 2 Percentage accuracy of 10-fold cross validation of classification methods
subtypes of cancer, namely classic medulloblastomas (CMD) for all genes.
and desmoplastic medulloblastomas (DMD). After data
preprocessing, 857 genes remain. The source of the 857 gene According to Figure 2, Multi-Layer Perceptron (MLP) has
expression measurements is publicly available at [18]. Central the best performance among classification methods which
Nervous System Cancer dataset is available at Schliep lab performed on central nervous system microarray gene
bioinformatics Repository of Rutgers University contains 857 expression dataset with 97.06%. RBFNetwork and cAnt-Miner
genes with one class attribute. The dataset includes numeric have the best results respectively, with 94.12% and 92.50%
attributes. The class shows two subtypes of cancer named CMD after MLP. SVM and OneR have the worst performance, with
and DMD. The dataset contains 34 samples belonging to two 73.53% in the experimental study.
different target class. Table 1 shows the summary of the
characteristics of the dataset. Percentage Average Accuracy of Classification Methods
TABLE I
100
CENTRAL NERVOUS SYSTEM CANCER DATASET
94,56
80
Dataset Comparison Class Gene Sample 77,45
Central Nervous Tumor 2 857 34 60
System Cancer Subtypes
40
20
IV. EXPERIMENTAL RESULTS AND DISCUSSIONS
0
In this study, we compared the efficiency of the Statistical Methods Artificial Intelligence based
classification methods including; SVM, OneR, J48 Decision Methods
Tree, MLP, RBFNetwork and cAnt-Miner methods for the
prediction of cancer risks. We used improved and modified
Weka [19] software for applying classification on dataset. The Fig. 3 Percentage average accuracy of statistical and artificial intelligence-
classification accuracy is used as the performance measures. based classification methods on the dataset.
We transformed data into the format arff for Weka. At first, As it is shown in Figure 3, the average performance of
feature selection method was used to find relevant features in artificial intelligence-based methods is better than the average
the central nervous cancer dataset and then, classification performance of statistical methods. Approximately, the average
algorithms were applied to the selected features to evaluate the performance of artificial intelligence-based methods is 22.09%
algorithms. Thirty features (genes) were selected by the feature more than the average performance of statistical methods.
selection algorithm. To select an important subset of genes Artificial Intelligence-based classification methods performed
from thousands of genes is pretty arduous. That’s why, gene very well on Central Nervous System microarray cancer with,
selection becomes the most needed requirement for a diagnostic 94.56%. However, the average of statistical methods is less
classifying system. For this reason, many researchers have
____________________________________________________________________________________________________________
14
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
than the average of artificial intelligence-based methods, with [6] I. Guyon, J. Weston, S. Barnhill, and V. Vapnik, “Gene Selection for
Cancer Classification using Support Vector Machines,” Mach.
83.33%.
Learn., vol. 46, no. 1/3, pp. 389–422, 2002.
[7] J. Novakovic, M. Minic, and A. Veljovic, “Genetic Search for
V. CONCLUSION AND FUTURE EXTENSIONS Feature Selection in Rule Induction Algorithms,” pp. 1109–1112,
In this paper, we have conducted a comparative study of the 2010.
[8] C. Saylan, “Intelligent method based on new feature selection
classification methods for microarray data analysis on publicly algorithm on renal transplantation patients,” Kadir Has University,
available dataset including microarray central nervous system 2013.
cancer dataset. Firstly we applied Correlation-based Feature [9] E. Oztemel, Artificial Neural Network. Papatya Publishing, 2003.
Selection (CFS) which is method of dimension reduction on [10] E. Cetin, “The Applications of Artificial Intelligence,”
Ankara,Turkey: Seckin Publishing, 2007, pp. 379–401.
microarray dataset and then we compared the performances of [11] D. S. Broomhead and D. Lowe, “Radial Basis Functions, Multi-
six classification methods, namely Support Vector Machine Variable Functional Interpolation and Adaptive Networks,” 1988.
(SVM), OneR, J48 Decision Tree, Multi-Layer Perceptron [12] F. M. Ham and I. Kostanic, Principles of neurocomputing for science
(MLP), Radial Basis Function Network (RBFNetwork) and and engineering. McGraw Hill, 2001.
[13] A. C. Marco Dorigo, V. Maniezzo, Alberto Colorni, Marco Dorigo,
cAnt-Miner on central nervous system dataset by using control Marco Dorigo, Vittorio Maniezzo, Vittorio Maniezzo, Alberto
parameters most commonly used in the literature. In Colorni, “Positive Feedback as a Search Strategy,” 1991.
conclusion, the experimental results show that the artificial [14] B. Alatas and E. Akin, “The Discovery of Classification Rules by Ant
intelligence-based classification methods have higher accuracy Colony Algorithm,” 2004.
[15] R. S. Parpinelli, H. S. Lopes, and A. A. Freitas, “An ant colony
than the statistical classification methods. In the future, we will algorithm for classification rule discovery,” in Data mining: A
study on control parameter optimization of classification heuristic approach, 2002, pp. 191–208.
methods and then we will compare results of classification [16] B. Liu, H. A. Abbass, and B. Mckay, “Classification Rule Discovery
methods with parameters most commonly used in the literature with Ant Colony Optimization,” IEEE Comput. Intell. Bull. , vol. 3,
no. 1, pp. 31–35, 2004.
and with optimal control parameters. [17] S. L. Pomeroy, P. Tamayo, M. Gaasenbeek, L. M. Sturla, M. Angelo,
M. E. Mclaughlin, J. Y. H. Kim, L. C. Goumnerova, P. M. Black, C.
ACKNOWLEDGMENT Lau, J. C. Allen, D. Z. I, J. M. Olson, T. Curran, C. Wetmore, J. A.
This work was supported by the Research Fund of Erciyes Biegel, T. Poggio, S. Mukherjee, R. Rifkin, A. Califano, G.
Stolovitzky, D. N. Louis, J. P. Mesirov, E. S. Lander, and T. R. G. Ii,
University of Turkey, grant number: FYL-2015-6095. “Prediction of central nervous system embryonal tumour outcome
based on gene expression,” vol. 415, no. January, pp. 436–442, 2002.
REFERENCES [18] “Central Nervous System Cancer Dataset,” 2013. [Online]. Available:
[1] H. Liu, I. Bebu, and X. Li, “Microarray probes and probe sets.,” http://bioinformatics.rutgers.edu/Static/Supplements/CompCancer/
Front. Biosci. (Elite Ed)., vol. 2, pp. 325–38, 2010. Affymetrix/pomeroy-2002-v1/pomeroy-2002-v1_database.txt.
[2] H. U. Luleyap, The Principles of Moleculer Genetics. Izmir: Nobel [Accessed: 12-Jul-2016].
Bookstore, 2008. [19] E. Frank, M. Hall, L. Trigg, G. Holmes, and I. H. Witten, “Data
[3] K. Ipekdal, “Microarray Technology,” 2011. [Online]. Available: mining in bioinformatics using Weka,” Bioinformatics, vol. 20, no.
http://yunus.hacettepe.edu.tr/~mergen/sunu/s_mikroarrayandecolog 15, pp. 2479–2481, Oct. 2004.
y.pdf. [Accessed: 05-Jul-2016].
[4] M. a. Hall and L. a. Smith, “Practical feature subset selection for
machine learning,” Comput. Sci., vol. 98, pp. 181–191, 1998.
[5] V. Vapnik and V. Vapnik, Statistical learning theory. 1998.
____________________________________________________________________________________________________________
15
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Unmanned Aerial Vehicles (UAVs) are designed as unmanned aerial vehicles follow in this study. According to
alternative solutions to traditional aerial vehicles controlled by Scholer [11], autonomy consists of six parts that are mis-
human on-board. Recent advances in robotic and communica- sion distribution, hazardous area detection, route planning,
tion technologies have enabled UAVs to become convenient plat-
forms for various missions from aerial surveillance to combat avoidance of collision, generation of trajectory and vehicle
operations. Today, most of the efforts on UAV technologies from control. By increasing autonomy, the stress on the operator
scientific and industrial communities are focusing on autonomy decreases and it provides an environment that can be used
to provide self-decision skills to UAVs particularly on their by lots of unmanned aerial vehicles. Therefore, while data
mobility. In order to execute a certain mission effectively in obtained by using only one UAV takes a long period of
the shortest possible time, an optimal pre-defined route plan
which regulate the mobility of UAVs is required. Moreover, time, using multiple UAVs simultaneosuly may provide time
for multi-UAV systems, which consist of more than one UAVs efficiency.
perform their tasks in collaboration, this pre-defined route plan Using a big UAV at the duties is not economic. Besides,
should be updatable during the mission in order to make the using the smaller UAVs together that is equipped with
system fault tolerant. Route planning for multi-UAV systems payload increases the success of the duties. The problem is
is NP-hard problem which is also known as multiple traveling
salesman problem (mTSP). Given a set of waypoints, multi take- that calculate the routes of lots of UAVs both accurately and
off points for all UAVs, and a cost metric, the objective of the fastly. This problem is in the NP-Hard category according
problem is finding a set of routes for a given number of UAVs to the target points that will be gone. In this structure that
so as to minimize the total cost. It is obvious that increase in can be named as the traveling salesman problem, increasing
the number of waypoints or UAVs cause exponential increase the target points that form the field of problem means that
in search space. In order to find a solution with good quality
from this huge search space within a reasonable amount of time, increasing the calculating time exponentially that is required
evolutionary algorithms and parallel computing techniques can for route planning. So, genetic algorithm offers a good
be used. In this study, it is aimed to find near-optimal route solution in an acceptable time as the other evolutionary
plans for multi-UAV systems. The objective function used in this solution methods.
study provides that all individual routes of UAVs to be close each The time that genetic algorithm requires to find the so-
other as possible. The algorithm is implemented on GPUs using
NVIDIAs parallel computing platform, Compute Unified Device lution increases with increasing the number of target point.
Architecture (CUDA). The efficiency and the effectiveness of In this situation, the information of route that UAVs, which
the proposed parallel GA approach are demonstrated through is controlled simultaneously, need must be calculated in a
simulations under different scenarios. shorter time. To attain the requiring time we must ensure
Keywords-parallel genetic algorithm, CUDA, multi-uav sys- that the algorithm of finding route process in parallel. Briefly
tems, multi-TSP.
total process time can be shortened by calculating in parallel,
using the parts of program that don’t effect each other in lots
I. I NTRODUCTION
of operation simultaneously and gathering the results. On
The aviation is a discipline that doesn’t accept risks. Lots account of requiring equipment, using the GPU cores that
of R&D (Research and Development) activities has been works at low frequencies unlike the high frequency CPU
made to decrease the risk factor to the lowest level and cores and has lots of in its is eligible for parallel calculating.
eradicate the risk at the end. The top of this activities is In this study, taking advantage of the power of the
unmanned aerial vehicles (UAV). Technological investments NVIDIA’s CUDA architecture, it is aimed to adopt the
and autonomous systems demonstrate that conventional air genetic algorithm approach to solve route planning problem
platforms are going to replace with unmanned aerial vehicles. for multiple UAVs. By this way, it is intended to find near-
Thus, people leave the platform will form the core of these optimal route plans for multi UAV systems. All UAVs are
systems that control program and structure. assumed to be resided at different locations. One of the goals
It is aimed to improve the abilities of autonomous move- in this study is making all individual routes of UAVs to be
ment by calculating the routes fastly and effectively that close as possible. For this reason, the objective function used
____________________________________________________________________________________________________________
16
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
II. BASICS
A. Multiple Traveling Salesman Problem (mTSP)
In this study, it is observed that m number of UAV control
n number of target points and turn back to the initial point
again. This problem is defined as a generalization of widely
known travelling salesman problem which more than one
salesman is allowed. It is called as multiple traveling sales-
man problem (mTSP). Given a set of points, m depots where Fig. 1. Flow of The Genetic Algorithm
m salesmen are located, and a cost metric, the objective of
mTSP is to determine a route for each salesman such that
the total tour cost is minimized and that each city is visited III. R ELATED S TUDIES
exactly once by only one salesman. In mTSP, all of the routes Dantzig and Ramser [4] have provided an algorithmic
must start and end at the same depot and each city must be approach to Vehicle Routing Problem (VRP) which is rep-
visited exactly once by only one salesman. Sorting targets in resented as in mTSP model by considering transport costs
the best order is not enough to get the optimal solution. It of vehicles. The main objective here was the minimizing the
is also imported that which point is assigned to a salesmen total cost by using Vehicle Routing Problem (VRP). This
[1]. study is considered as the first study of VRP which is still
Kara and Bektas [7] have provided a number of variations agreed to be used in today’s vehicle systems.
on the mTSP by presenting a mathematical model. With this In a study conducted by Ercan and Gencer [5], route plans
model, sub-tour creation is prevented by detecting neighbors for UAVs have been accepted as VRP and their types have
counts to visit. been inspected and classified according to the needs.
Sahingoz [10] have proposed a flyable route which fits
B. Genetic Algorithm
to multiple UAV’s maneuver angle in his study. The route
Genetic algorithm can be defined as similar to the evo- is calculated with the genetic algorithm and the solution is
lution observed in nature. Because it provides an heuristic produced in a faster way. In order to make the route flyable,
search method, it does not necessitate to search all solution Bezier curves are used in his study.
space of the problem. The terms used in this algorithm are In another route planning study, Karakaya [8], has pro-
very similar to the ones in biology. Some of these terms vided multiple Traveling Salesmen Problem approach to
are chromosome, generation, choosing parents, crossover, detect minimum count of UAV to spy a single target and to
mutation, etc. obtain a near-optimal route. The Ant Colony Optimization
In GA, each candidate solution represents chromosomes. (ACO) Algorithm is used in his work. With a single vehicle
Each chromosome is organized as an array of genes. In ad- base, considering the range limitations and coordination’s
dition, a number of chromosome forms a generation. In this of targets 20% saving is achieved in the number of used
regard, chromosomes in the generation are sorted in terms unmanned vehicle number.
of their solution quality according to fitness function. Each Ergezer and Leblebicioglu [6] have studied on the most
generation forms a new generation using GA operators such efficient way that unmanned vehicles can obtain information
as crossover and mutation. As a result of new generations, from the desired locations. It is aimed to solve Traveling
the population becomes more qualified having individuals Salesmen Problem by staying away from the restricted
with better fitness values. regions. Matlab’s genetic algorithm tools have been used for
Fundamental phases of GA is shown in Figure 1. the calculation of the route.
____________________________________________________________________________________________________________
17
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Valero et al. [12] have made the route planning using grid with given dimension and size. The reason why threads
genetic algorithm for robots which are planned for the service form two hierarcy is as follows. Threads in a thread block
to the patients. The scenarios which have 2,3,4, and 5 robots can share data efficiently via shared memory. However, the
have been tested for the problem areas which have 30,40, maximum number of threads per block is limited to 1024. So,
and 50 targets. if more than 1024 threads are required, we have to partition
Cekmez [3], [2] and the others revealed that the route threads into several thread blocks with the same size.
planning of one UAVs can be calculated faster by operating
V. PARALLEL G ENETIC A LGORITHM
ACO and genetic algorithm on GPUs in parallel.
A parallel genetic algorithm was designed for multiple
IV. CUDA A RCHITECTURE UAV route planning in this paper. Each solution in the
This section describes the CUDA architecture through the genetic algorithm is named as chromosome. A cluster formed
example of the GeForce GTX 480 GPU card, which is used by these chromosomes is named as a population. After
in our experiments. crossover and mutation over the members, the population,
which is formed randomly at the beginning, has become
In terms of hardware, CUDA GPUs are regarded as two
a more acceptable situation. We express every candidate
level shared-memory machines. Processors in a GTX 480
solution with a chromosome in our algorithm. We evaluate
are grouped into streaming multiprocessors (SMs). Each
the quality with a fitness function which we determined. Al-
SM consists of 32 processors. The GTX 480 has 15 SMs.
gorithm stages, parallel design, and performance comparison
Therefore, the GTX 480 is equipped with total of 480
according to the results will be explained in detail at the next
processors. CUDA is highly multithreaded architecture that
sections.
allows a maximum of 1536 concurrent threads per SM.
Hence, the GTX 480 can run a maximum of 23040 threads A. Codification of The Chromosome
concurrently. Processors in a SM can exchange data via fast To make the route solution of multiple UAVs, it is required
shared memory. On the other hand, data exchange between that m number of UAV and n number of target must exist in
SMs can be performed via VRAM. VRAM is also like the candidate solution (chromosome). Therefore, we can see
main memory for processors. So, code and data in a CUDA that which UAV will follow which route. In our study on one
program are basically stored in VRAM. chromosome, m number of UAV is shown with the values
CUDA programming model is one of multi-threaded pro- between [0, m − 1] and n number of target is shown with
gramming models. In the following, we describe an overview the values between [m, m + n − 1] [12]. Every chromosome
of CUDA programming model using Fig. 2. must be started one value showing a UAV. Providing that
the index repetition does not occur, the sequence that comes
after index of UAVs shows the route that UAV follows.
For example, the sample chromosome structure provided
below shows route plans for 3 UAVs for 12 targets;
[2−5−7−8−11−0−4−6−10−3−9−1−12−14−13]
In the solution;
• UAV 0 : 4-6-10-3-9
• UAV 1 : 12-14-13
• UAV 2 : 5-7-8-11
route is followed.
B. Calculation of The Fitness Value of Chromosome
It is required that the total route distance has to be its
minimum value at standard mTSP. However, in our case,
it is different from the standard mTSP. For route planning
problem for multiple UAVs, It is required that duty execution
time must be its minimum level. So, we can reduce the duty
time by shortening UAV’s routes that has the maximum route
distance in the candidate solution.
We want to minimize the total route distance (1st term in
Fig. 2. CUDA Programming Model. Eq. 11) and maximum subtour distance (2nd term in Eq. 11)
in our study. Besides we provide that the route distance of
In a CUDA program, threads form two hierarchy: a grid every UAV to be close to each other by trying to increase
and thread blocks. A thread block is a set of threads. A thread the maximum base tour distance (3nd term in Eq. 11) [12].
block has dimensions 1, 2, or 3. A grid is a set of blocks
with the same size and dimension. A grid has dimensions 1 m−1
or 2. Each thread executes the same code specified by the
X
F itness(indi ) = +0.9∗max(Ri )−0.9∗min(Ri ) (1)
kernel function. A kernel function call generates threads as a i=0
____________________________________________________________________________________________________________
18
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
19
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE V
T EST PARAMETERS FOR THE EFFECTS OF OBJECTIVE FUNCTION ON
AVERAGE FLIGHT COST
Num. of UAV 4
Num. of Waypoint 100
Mutation rate %20
Ending Criteria 200
Num. of threads 5120
Model of the GPU card Geforce GTX480
Penalty Factors [0 , 0.3 , 0.6 , 0.9]
____________________________________________________________________________________________________________
21
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
22
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—In this study, the premise and consequent parameters the following section, the basic structure of ANFIS is
of ANFIS are optimized using Genetic Algorithm (GA) based on a introduced, in the third section, information about genetic
population algorithm. The proposed approach is applied to the algorithm is given, in the fourth section, information about the
nonlinear dynamic system identification problem. The simulation presented approach is given and the obtained results are given
results of the method are compared with the Backpropagation
in the section 5.
(BP) algorithm and the results of other methods that are available
in the literature. With this study it was observed that the
II. ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM (ANFIS)
optimisation of ANFIS parameters using GA is more successful
than the other methods. Adaptive Network-based Fuzzy Inference System (ANFIS)
is a network model which Sugeno-type fuzzy system is
Keywords— neuro-fuzzy, ANFIS, genetic algorithm, system combined with neural learning ability. The main aim of ANFIS
identification is to optimize the parameters of the equivalent fuzzy logic
system by using input-output data sets via a learning algorithm.
I. INTRODUCTION Parameter optimization is carried out in such a way that the
System identification is a model generating process that is error value between the actual output and the target output to
developed to predict the behaviour of a system between its be minimum.
input and output signals. The approaches based on the fuzzy ANFIS contains two parameters as antecedent and
neural networks and artificial neural networks are among the consequent parameters. Those two parameter types connect the
methods that are used commonly for the dynamic systems fuzzy rules to each other and training of the model is provided
identification. However, due to some of its superior features, with the optimization of these parameters. ANFIS basically
studies towards using ANFIS for the purpose of dynamic consists of five layers. A basic ANFIS structure consisting of
system identification, have started to increase gradually in the two inputs and one output is given in Fig. 1.
recent years [1-5].
Training the ANFIS model is basically the determination
process of the optimal values for the model’s premise and
consequent parameters. Derivative-based algorithms are
commonly used for training of ANFIS. But there is a problem
of getting stuck in a local minimum in derivative-based
algorithms. In this context, various methods have been
proposed in recent years in order to train ANFIS parameters.
Some of these methods are artificial intelligence optimization
algorithms and heuristic algorithms such as Genetic algorithm,
PSO and Differential Evolution Algorithm that are not
derivative-based [6]. Fig. 1 A basic structure of ANFIS [9]
Ghomsheh et al. [3] proposed Particle Swarm Optimization
(PSO) algorithm that is modified for the optimization of ANFIS
parameters. Carrano et al. [7] used GA for the training of A. Layer 1
ANFIS in the solution of multi-objective optimization
This layer is named as the fuzzification layer. The signal that
problems. Cus et al. [8] used Ant Colony Optimization (ACO)
is obtained from each node is transfered to the other layer. In
algorithm for training of ANFIS parameters. They tried the
this layer, for the outputs of the cells, ( O1i ), Equation (1) and
performance analysis of the developed method on the non-
linear systems. Equation (2) are given [10].
In this study, an approach is presented towards the training
ANFIS by using GA based on a population algorithm. The O1i Ai ( x) i 1,2 (1)
performance of the presented approach is tested on non-linear O1i Bi 2 ( x) i 3,4 (2)
system and the results are compared with different methods. In
____________________________________________________________________________________________________________
23
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
In here, Ai and Bi are any membership functions belong to chromosomes transform into the solution candidates that
the inputs, Ai and Bi are membership degrees that are represent better results. This process is maintained until
reaching an acceptable compliance value or until meeting the
calculated for this function. For the bell-shaped membership criteria such as pre-determined processing time or generation
function Ai is calculated with the equation below. number. Basic steps of the Genetic Algorithm are given in Fig.2.
1 x c 2
( )
Ai e 2 a
i 1,2 (3) 1. Generate random population of n chromosomes
2. Evaluate the fitness of each chromosome
3. Create a new population by repeating following steps
In here, a i and ci are sigma and central parameters of the until the new population is complete
membership function respectively. (i) Select two parent chromosomes from a
population according to their fitness
B. Layer 2 (ii) With a crossover probability cross over the
This layer is named as the rule layer. Each rule’s firing parents to form new offspring
strength is calculated with the membership degrees that are (iii) With a mutation probability mutate new
coming from the previous layer. offspring at each locus
(iv) Place new offspring in the new population
O2i wi Ai ( x).Bi ( y) i 1,2 (4) 4. Use new generated population for a further run of the
algorithm
C. Layer 3 5. If the end condition is satisfied, stop, and return the
best solution in current population
In this layer which is named as the normalisation layer, each
6. Go to step 2
rule’s normalised firing strength is calculated.
O5i f wi . f i
w .f i i
i 1,2 (7)
ANFIS parameters are optimized with GA. Block diagram
showing this structure is given in Fig. 3. RMSE error function
w i which is obtained by using Equation (8), is used for
determining the error value of the solution. 𝑦̅𝑖 is used in
III. GENETIC ALGORITHM Equation (8), is showing the output obtained by ANFIS at the
Genetic Algorithm (GA) which its fundamental principles time i and 𝑦𝑖 is showing the actual output of the system. N is
are set forth by John Holland in 1970’s, is implemented with showing the instance number used in the application.
success on many problem types [11]. GA is a heuristic
algorithm which is used for being able to find exact or
[∑𝑁
𝑖=1(𝑦𝑖 −𝑦
̅̅̅)
𝑖 ]
approximate results in optimization or search problem. This 𝑅𝑀𝑆𝐸 = √ (8)
𝑁
algorithm is developed by being inspired from the techniques
in the evolutionary biology such as inheritance, mutation,
The selection of the input array which will be used in training
selection and crossover. GA can be implemented quite
is important on the success of training. The elements of the
effortlessly even on the multidimensional problems with large-
input array are usually chosen from sinusoidal and random
size search space and also with too many number of variables.
input values [13]. In the carried out simulation studies within
GA is a population based optimization algorithm. Equivalent
this study, u(k) input signal array which has proper distribution,
of the candidate solutions that generate the population is
randomly generated in the range of [−2,2] is used.
chromosomes. Due to the various evolution processes, these
____________________________________________________________________________________________________________
24
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
25
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES [8] F. Cus, J. Balic and U. Zuperl, “Hybrid ANFIS-ants system based
optimisation of turning parameters,” Journal of Achievements in
[1] D. Karaboga and E. Kaya, “Training ANFIS using artificial bee colony Materials. vol. 36, no. 1, pp. 79-86, 2009.
algorithm for nonlinear dynamic system identification,” in: IEEE 22nd [9] S. Uzundurukan, “Determination and modeling of basic parameters that
Signal Processing and Communications Applications Conference (SIU),
affect the swelling properties of soils,” Ph.D. Thesis, Suleyman Demirel
2014, pp. 493-496. University, Turkey, 2006.
[2] P. Liu, W. Leng and W. Fang, “Training ANFIS model with an improved [10] J.S.R. Jang and C.T. Sun, “Neuro-Fuzzy modeling and control,” The
quantum-behaved particle swarm optimization algorithm,”
proceedings of the IEEE. vol. 83, no. 3, 378-406, 1995.
Mathematical Prob. in Eng. vol. 2013, 2013. [11] M. Mitchell, “An Introduction to Genetic Algorithms,” MIT Press, 1998.
[3] V.S. Ghomsheh, M.A. Shoorehdeli and M. Teshnehlab, “Training [12] S.Lek, M.Scardi, P.F.M. Verdonschot, J.-P. Descy and Y.-S. Park,
ANFIS structure with modified PSO algorithm,” In Control and
Modelling Community Structure in Freshwater Ecosystems. Berlin,
Automation, Med’07. Mediterranean Conference on IEEE, 2007, pp. 1- Germany: Springer-Verlag, 2005.
6. [13] D.E. Rumelhart and J.L. McClelland, “Explorations in the micro-
[4] M.A. Shoorehdeli, M. Teshnehlab, A.K. Sedigh and M.A. Khanesar,
structure of cognition,” parallel distrubuted processing 1. Cambridge,
“Identification using ANFIS with intelligent hybrid stable learning MA: MIT Press, 1986.
algorithm approaches and stability analysis of training methods,” [14] C.F. Juang, “A TSK-type recurrent fuzzy network for dynamic systems
Applied Soft Comput. vol. 9, no. 2, pp. 833–850, 2009.
processing by neural network and genetic algorithms,” IEEE
[5] M.A. Shoorehdeli, M. Teshnehlab and A.K. Sedigh, “Training ANFIS Transactions on Fuzzy Systems. vol. 10, no. 2, pp. 155–170, 2002.
as an identifier with intelligent hybrid stable learning algorithm based [15] A. Kalinli, “Training Elman network using simulated annealing
on particle swarm optimization and extended Kalman filter,” Fuzzy Sets
algorithm,” Journal of the Institute of Science and Technology of
and Systems. vol. 160, pp. 922–948, 2009. Erciyes University. vol. 19, pp. 28-37, 2003.
[6] D. Simon, “Training fuzzy systems with the extended Kalman Filter,” [16] A. Kalinli and D. Karaboga, “Training recurrent neural networks by
Fuzzy Sets Syst. vol. 132, pp. 189–199, 2002.
using parallel tabu search algorithm based on crossover operation,”
[7] E.G. Carrano, R.H.C. Takahashi, W.M. Caminhas and O.M. Neto, “A Engineering Applications of Artificial Intell. vol. 17, pp. 529-542, 2004.
genetic algorithm for multiobjective training of ANFIS fuzzy networks,”
IEEE Congress on Evolutionary Computation, pp. 3259–3265, 2008.
____________________________________________________________________________________________________________
26
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Fast moving electro-mechanical and wireless com- smaller UAVs can be produced at lower costs with the
munication technologies have made it possible to design and downsizing trend experienced in embedded systems today.
implement multiple unmanned aerial vehicle (multi-UAV) sys- The decrease in the size of UAVs brings out strict limitations
tems which promise more efficient and reliable ways to per-
form application-specific missions through their simultaneous over maximum take-off load capacity so that application-
execution capability and fault tolerant potential. In a typical specific payloads used in these small UAVs have more
surveillance-based multi-UAV application, each individual UAV limited capabilities. For this reason, required efficiency for
collects required data from its area of interest and transfer some UAV applications cannot be achieved by using a single
it to a ground station (GS) through a shared communication small UAV. In this context, it is considered that a multi-UAV
link. In some challenging cases, e.g. in a hostile environment
such as battlefield communication, which it is unable to es- system, which is composed of multiple small UAVs, may
tablish and maintain a reliable end-to-end communication link offer more efficient solutions by performing jointly missions
between UAVs and the GS, Delay Tolerant Network (DTN) such as surveillance, reconnaissance, monitoring etc. Using
paradigm which is based on store-carry-forward data delivery multiple UAVs allows establishing a fault tolerant structure
technique can be applied. However, in such system, link layer along with providing task completion in a shorter period of
packet collisions may occur because it is quite possible that
multiple UAVs may arrive at communication range of the time owing to the advantage of its capability of simulta-
GS at overlapping times and need to transfer their data neous task execution in multiple locations. Also, potential
simultaneously. This paper presents deadline-sensitive medium coordination capability of multiple autonomous UAVs may
access control (DSMAC) protocol designed for DTN based provide to accomplish missions that no individual UAV can
multi-UAV systems. The DSMAC protocol provides a non- accomplish on its own. However, multi-UAV approach brings
preemptive medium access scheme in a centralized manner to
make the medium free of collisions. The main objective is to many distinctive challenging design problems.
enable UAVs to transfer their large amount of surveillance data Communication is one of the most important design
as fast as possible to the GS by taking deadline-constrained data problems for multi-UAV systems. The main objective for a
delivery requirement into consideration. Our protocol differs multi-UAV application which performs a typical reconnais-
significantly from commercially available technologies such as sance/surveillance mission is increasing success rate of the
802.11, 3G or LTE in that deadline meeting rate is the main
performance metric in our case rather than providing general mission and providing time efficiency by distributing tasks
connectivity and fair medium sharing among large number of on multiple UAVs. It is required for each individual UAVs
UAVs. Simulation studies are conducted by using NS-2. Results in the system to collect application-specific data from its
show that DSMAC protocol provides considerable performance own area of interest and transfer it to a ground station (GS)
improvements in terms of average waiting time and deadline as soon as possible for further data processing phases. In
meeting rate when compared to PCF (Point Coordination
Function) mode of widely used IEEE 802.11 technology. such applications, it is difficult to setup a reliable end-to-
Keywords— medium access control, delay tolerant networks, end communication link between UAVs and the GS in cases
multi-uav systems, deadline sensitive. which locations to be visited for data collection remain out
of the communication range provided by the GS or, in more
I. I NTRODUCTION general terms, which there are some challenging conditions
Recent developments in micro electro-mechanical sys- in the communication environment such as battlefield com-
tems and wireless communication technologies have enabled munication, disaster rescue etc.
the realization of unmanned aerial vehicle (UAV) systems. For multi-UAV applications performing tasks which re-
Single-UAV systems, which consist of only one large UAV quire collecting data in a certain operation area and then
with powerful hardware, have produced at relatively high deliver it to a GS, it is obvious that there is no need for
costs and provided successful results in many application immediate or real-time data transmission in cases which
areas for about last twenty years. On the other hand, much the application-specific data has a certain degree of delay
____________________________________________________________________________________________________________
27
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
tolerance. Clearly, for such cases, it is required to collect (CBDTP) for underwater wireless sensor networks which is
delay tolerant data from the operation area, store it into based on predicating a value for a sensor node if its data
the on-board memory of UAVs, and carry it until reaching were not received at the sink node instead of having the
required wireless communication range for reliable con- sensor node retransmit its data. In another study, Zhong et al.
nection with the GS. In this context, it is considered that [9] has proposed UMACAW as another delay tolerant MAC
using Delay Tolerant Network (DTN) [1] model, which have protocol which utilize the rest time of RTS/CTS/DATA/ACK
been developed for similar needs, may provide considerable exchange period by taking advantage of long propagation
advantages for such multi-UAV applications. The DTN based delay in underwater channels. However, none of these efforts
data delivery technique considered in this study for multi- made on MAC protocols intended for DTN paradigm in the
uav systems is first proposed by Spyropoulos et al. [2] and literature take deadline-constrained data delivery requirement
it is also known as Direct Delivery. Fig.1 below illustrates a into consideration.
multi-uav system which performs its operation within the For DTN based multi-UAV applications considered in this
context of this simple single-copy DTN delivery scheme. study, the problem of link layer packet collisions may occur
UAVs, as source nodes, carry their application-specific data when multiple UAVs arrive at the GS at overlapping time
until they reach the GS. periods and attempt to transfer their data simultaneously, as
it is shown in Fig.1. In general, COTS technologies such as
IEEE 802.11, 3G, and LTE are used to overcome link layer
packet collision problem for such applications. However, the
main objective of these technologies is providing medium
sharing as fair as possible for large number of stations. Al-
though multi-UAV applications perform their tasks by using
DTN model, there may still be some deadline constraints
for collected data within certain scenarios. Therefore, the
most important need for multi-UAV systems which perform
their tasks under such scenarios is providing medium access
for UAVs by taking deadline constraints of their data into
account and data transmission without any interruption in
order to ensure UAVs to continue their tasks as soon as
possible. To the best of our knowledge, medium access in
Fig. 1. Direct delivery scheme applied to a multi-uav system. DTN has not been studied from this perspective before and
this is the first research effort focusing on deadline sensitivity
DTN paradigm has been developed for abnormal network of channel allocation for DTN based multi-uav systems.
conditions where a reliable communication and a complete To address these needs, in this paper, a MAC protocol for
end-to-end connectivity between the data source and desti- DTN based multi-UAV systems is proposed which eliminates
nation may not exist. Different from traditional networking the link layer packet collisions, ensures UAVs to return
approach, DTNs are characterized by long latencies, unstable their operation area as soon as possible by providing data
network topologies, and inconsistent connectivity. With these transmission without suspending it because of any other
characteristics, DTNs find places in such applications that possible transfer requests, and uses deadline information of
traditional networks cannot work effectively, for example, collected data as a priority measure for medium access.
data communications in rural area where communication Because the main criteria for medium access is deadline
infrastructure cannot be deployed or costly [3] , disaster for transmission of data to the GS, the proposed MAC
rescue [4], military/battlefield communications [5], etc. protocol is named as “Deadline-sensitive Medium Access
Most of the research efforts on DTN in the literature Control” (DSMAC). DSMAC protocol provides a centralized
focus on data delivery/routing schemes for different network medium access model using two different communication
scenarios assuming that mobile communication devices are channel. Tests for DSMAC protocol are conducted by using
distributed sparsely and network traffic is low. However, link NS2. DSMAC’s superiority over PCF (Point Coordination
layer packet collision problem in DTN-based applications Function) [10] mode of widely used IEEE 802.11 protocol
is not considered in these studies. Within this context, a are analyzed through performance metrics such as average
study conducted by Jindal et al. [6] shows that the packet waiting time and deadline meeting rate. Simulation results
collision problem should not be overlooked especially for show that channel allocation scheduling mechanism and
densely distributed communication devices are on subject. centralized MAC model provided by DSMAC protocol out-
In another study, Liang et al. [7] has proposed a DTN- performs IEEE 802.11 PCF in terms of average waiting time
friendly MAC protocol (DFMAC) which is developed for and deadline meeting rate for aforementioned multi-UAV
radio resource allocation for a DTN/WLAN integrated net- applications.
works. In order to address the problem of large delay The remainder of the paper is organized as follows. The
variance in underwater acoustic channel, Zhang et al. [8] protocol design issues and its operation details are described
has proposed a cluster based delay tolerant MAC protocol in Section II. Medium access scheduling mechanism of
____________________________________________________________________________________________________________
28
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
DSMAC is provided in Section III. Performance evaluation version, frame type/subtype, retry flag, etc. is inherited from
and simulation results are presented in Section III. Finally, the 802.11 protocol specifications due to its ease of NS2
the paper is concluded in Section IV. implementation.
II. DSMAC P ROTOCOL
In the following sections, we first present an overview of
proposed DSMAC protocol to discuss the main objectives
and then we provide the design issues of DSMAC with its
operation details.
A. Overview Fig. 2. START frame format.
For a DTN based multi-UAV system considered in this
study which performs a surveillance-based mission, UAVs As it is shown in Fig.3, UAVs arriving at the GS wait
are required to collect application-specific data from their for the START frame to start transfer their data to the GS.
area of interests, store data in their own on-board memory, The GS, as a coordinator, is responsible to decide which
and then deliver them to the GS by moving to it until they UAV is allowed to use the channel first. The details of
arrive at required communication range for reliable wireless this decision process is described in Section III. Because
connection. Clearly, there is a need for a MAC protocol in UAVs are not allowed to access the channel freely, an
order to regulate medium accesses of UAVs in such cases ongoing transmission is not interrupted by other UAVs.
that multiple UAVs arrive at GS for data transmission at Clearly, there is no contention among UAVs to gain the
overlapping time periods and attempt to access the medium access to the channel due to the centralized mechanism of
simultaneously. DSMAC. To this end, additional frame exchanges between
One of the most important limitations for UAVs is en- the communication nodes in order to avoid packet collisions
durance capability. In order to utilize limited flight duration such in CSMA/CA are not needed in DSMAC protocol.
of UAVs within the context of data transfer to the GS, it is
quite important to consider this fact at MAC level operations.
Besides that, certain applications may necessitate specific
amount of delay tolerance for data delivery rather than
completely disregarding it even they perform under DTN
paradigm. For these reasons, to assure effective usage of
limited flight duration and timely data delivering, deadline-
constraint data transmission to the GS as fast as possible
must be effectively handled by an efficient MAC protocol
mechanism. To this end, the main objectives of DSMAC
protocol we design for DTN based muti-uav systems are:
• creating a centralized channel allocation mechanism
in order to provide contention-free medium access for
UAVs,
• using two different communication channels for data
and control frames to utilize data channel effectively,
• making channel allocation scheduling non-preemptive
to avoid suspending ongoing transmission,
• taking record deadlines into consideration for channel Fig. 3. Centralized medium access scheme in DSMAC.
allocation to increase the deadline meeting rate as much
as possible In DSMAC protocol, it is assumed to be used two different
transceiver in UAVs with two antennas and their associated
B. Design communication circuitry. One of them performs on data
DSMAC protocol is designed to provide a centralized channel while the other uses control channel. It is also
MAC scheme for DTN based multi-UAV systems. Accesses assumed that possible interference between two transceivers
to the medium are controlled only by the GS. It performs are resolved using a kind of digital signal processing methods
its operations as a coordinator while UAVs in the system or physical precautions.
are its associated mobile nodes. None of the UAVs has a It is clear that the GS has to be informed by UAVs in
direct access permission to the medium. UAVs are allowed to some way to notify that they have records to be transferred
transmit their data only when the GS sends a special frame to to the GS. In this regard, we introduce a term meta-info,
them, which is called START. Fig.2 below shows the START which is a brief information of records to be transferred to
frame format. It is important to note that Frame Control the GS. Since the DSMAC provides coordinator initiated data
field which has a number of sub-fields such as protocol transmission, this information is used by the GS to schedule
____________________________________________________________________________________________________________
29
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the channel allocations properly and to send a START frame Thus, DATA-ACK exchanges start between the UAV#1 and
to a UAV. The control channel mentioned above is used by the GS. After a short time, as this frame exchange continues
UAVs to send their meta-info to the GS when they arrive. between the UAV#1 and the GS, UAV#2 arrives at GS and
Thus, UAVs are able to send their data transfer requests using sends its meta-info using control channel as it is shown as red
control channel without violating the data channel. Because frames in the figure. In this way, the meta-info transmission
the size of meta-info is quite small and the essential need for does not disturb the ongoing data transmission. In order to
control channel is to provide a reliable communication link notify the GS for last DATA frame, UAV#1 set a flag into the
between UAVs and the GS, it is considered as a contention- Frame Control field of DATA frame which is shown in the
based channel which can be implemented by using one of figure as a red marker on the end of the last DATA frame sent
the COTS technologies such as XBee 802.15.4 [11] or Wi-Fi by UAV#1. The GS notices that it can send START frame
[10] modules. Data channel, on the other hand, is the channel to allocate the data channel to the UAV#2. To assure its last
allocated by the GS to a UAV for reliable data transmission ACK frame as a feedback is received by the UAV#1, it waits
using ARQ. Contention-free medium access for data transfer for a tD period of time before sending a new START frame
is provided using the data channel. to the UAV#2. Although the details of channel allocation
scheduling performed by the GS is provided in Section III,
at this point, it is important to note that the GS does not stop
the current data transmission even that UAV#2 has a record
with higher priority than the UAV#1 has. This is because the
GS schedules channel allocations based on non-preemptive
Earliest Deadline First service discipline as described later.
____________________________________________________________________________________________________________
30
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
next data transfer. Unlike EDF, scheduling algorithm used can be defined as:
in DSMAC is non-preemptive. This means the GS does not • The Waiting Time is an amount of time a data transfer
suspend ongoing data transfer to start another one. Although is waiting in the ready queue or sum of time waiting in
a newly received meta-info has a record with higher priority, ready queue.
which means a closer deadline value, current data transfer is • The Turnaround Time is an amount of time to complete
waited for to end. The reason that we are using such non- a data transfer (time from first placed in ready queue to
preemptive mechanism for channel access service is limited transfer completion).
flight duration of UAVs. It is important to note that one • The Response Time is an amount of time it takes
of the major objectives of DSMAC is to allow UAVs to from when a transfer request is placed in ready queue
continue their operation as soon as possible in addition to until the first response is produced (not completion of
satisfy deadline constraints of the records. transfer).
It is better to present a case study to gain more insight • The Deadline Meeting Rate (DMR) is the rate of satis-
regarding different scheduling mechanisms. To this end, we fied deadline constraints of records.
provide a sample transfer requests made by three UAVs in
Table II.
TABLE II
S AMPLE T RANSFER R EQUESTS
Transfer Request
Record Record Deadline
UAV Id Time Time
Id Size (sec)
(sec) (sec)
1 1 D1 60 120 0
1 2 D2 40 50 0
2 1 D3 10 30 30
2 2 D4 20 110 30
Fig. 6. Results for round-robin scheduling.
3 1 D5 30 170 60
3 2 D6 50 130 60
If round-robin scheduling is applied to sample transfer
requests provided in Table II, as it is shown in Fig.6, trans-
Based on the transfer requests provided in Table II, mission of records is performed in a fragmental manner with
Gantt charts and results for some performance measures are the 10 sec quantum value. Because round-robin scheduling
provided in Fig.6 and Fig.7 according to round-robin and is a preemptive approach, it gives relatively high average
non-preemptive EDF scheduling algorithms corresponding to times compared to non-preemptive EDF (Fig.7). The reason
802.11 PCF and DSMAC, respectively. Gantt charts show that we prefer to use round-robin scheduling mechanism
the order of data transfers for corresponding scheduling to compare it with the scheduling algorithm used in DS-
algorithms. Performance measures used in this case study MAC (non-preemptive EDF) is because it is used in 802.11
____________________________________________________________________________________________________________
31
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
PCF which also provides a centralized and contention-free A. Effects of Average Inter-arrival Time
medium access as DSMAC we propose. One of the main factors which effects DMR is inter-
arrival times of UAVs to the GS. When real-world conditions
are considered, arrivals of UAVs to the GS are completely
random and independent from each other. However, when
these random occurrences of arrivals are observed long
enough, a certain occurrence rate can be obtained. For this
reason, random UAV arrivals to the GS are considered as
a Poisson process which is widely used random process to
model random points in time. In this context, inter-arrival
times between UAV arrivals to the GS are produced based
on exponential distribution which arise naturally from the
Poisson process.
Fig. 7. Results for non-preemptive EDF scheduling.
TABLE IV
S IMULATION PARAMETERS
As it can be seen from Fig.6 and Fig.7, non-preemptive
EDF provides considerable performance improvement in
Parameter Value(s)
terms of DMR. This is because it schedules data transfer Number of UAV 4
requests by considering deadlines of records even it does not Number of Record for each UAV 1, 2, 3, 4, 5
suspend ongoing transfer. The round-robin scheduling also Record Size [5, 10] MB
satisfies a number of deadline constraints, but this is made Deadline Range [tTx x 1.5, tTx x 5] s
Average Inter-arrival Times 50, 75, 100, 125, 150, 175 s
unconsciously.
TABLE III
G ENERAL S IMULATION PARAMETERS
Parameter Value(s)
Channel Channel/WirelessChannel
Antenna Antenna/OmniAntenna
Radio Propagation Model Propagation/TwoRayGround
Network Interface Phy/WirelessPhy
Interface Queue Type Queue/DropTail/PriQueue
Interface Queue Capacity 50
Link Layer LL
MAC Protocol DSMAC and 802.11 PCF Fig. 8. Change in DMR with respect to average inter-arrival times.
CBR Data Rate 2 Mbps
CBR Packet Size 512 byte
Simulation Area 140 m x 140 m As can be seen from the Fig.8, there are similar trends
Number of UAVs 4, 8, and 12 for DSMAC and PCF with respect to DMR. DSMAC out-
performs PCF for all average inter-arrival times used in
the simulation. When it is considered that a UAV in the
____________________________________________________________________________________________________________
32
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
system has 3 records and 7.5 MB each on average, average multi-UAV systems. Table VI shows corresponding average
total transfer time of these records to the GS is calculated inter-arrival times for three different multi-UAV systems used
approximately as 77 seconds according to parameters used in this simulation.
in the simulation. Performance gain obtained for average
TABLE V
inter-arrival times which are close to this average value
S IMULATION PARAMETERS
(50, 75, 100) is about 8% on average while more relaxed
average inter-arrival times provide performance gain up to
Parameter Value(s)
15% as it is shown in Fig.8. For shorter average inter- Number of UAV 4, 8, 12
arrival times, PCF, which do not take record deadlines into Number of Record for each UAV 1, 2, 3, 4, 5
account, may complete most parts of the records which have Record Size [5, 10] MB
Deadline Range [tTx x 1.5, tTx x 5] s
longer deadline early in the process owing to its round- Arrival Periods 5, 10, 15, 20 min.
robin based medium access scheduling mechanism. On the
other hand, even DSMAC considers record deadlines for
medium accesses, it may suffer from convoy effect problem
TABLE VI
for shorter average inter-arrival times because of its non-
C ORRESPONDING I NTER - ARRIVAL T IMES FOR D IFFERENT M ULTI -UAV
preemptive scheduling scheme. For these reasons, the diver-
S YSTEMS
sity on performance gain for different average inter-arrival
times is observed in the simulation.
Average
Multi-UAV Arrival Period
Inter-arrival
System (min.)
Time (sec)
5 100
10 200
4 UAV
15 300
20 400
5 42
10 85
8 UAV
15 128
20 171
5 28
10 55
12 UAV
15 82
20 110
____________________________________________________________________________________________________________
33
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
show that 4-UAV system has a DMR about 80% while 8- distribution. Based on the fact that UAVs are assumed to be
UAV and 12-UAV systems have about 20% and 10% DMR capable of making delivery time estimation for each record
respectively for DSMAC protocol. Also, it is observed that they have, deadlines for each records are randomly generated
DSMAC outperforms PCF with respect to DMR for all between 1.5 and i times of delivery time estimation of
arrival periods used in the simulation. In addition, multi-UAV corresponding record. The value of i indicates the maximum
systems consisting much more UAVs reach their maximum boundary of the delivery deadline range. Fig.12 shows the
DMR slowly while systems that contain fewer UAVs con- change in DMR with respect to deadline generation range
verges its maximum DMR more quickly because of varying for DSMAC and PCF. Parameters used in this simulation is
corresponding average inter-arrival times. provided in Table VIII.
B. Effects of Average Record Size on Average Waiting Time
TABLE VIII
Size of records UAVs have directly effects the AWT S IMULATION PARAMETERS
performance for both protocols due to its first-hand relation
to the transmission time. This simulation study is conducted
Parameter Value(s)
to show how AWT changes through the average size of Number of UAV 4
records. Simulation configuration is provided in Table VI Number of Record for each UAV 1, 2, 3, 4, 5
below. Record Size [5, 10] MB
Deadline Range [tTx x 1.5, tTx x 3, 5, 7, 9] s
TABLE VII Average Inter-arrival Times 75 s
S IMULATION PARAMETERS
____________________________________________________________________________________________________________
34
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. C ONCLUSION [11] Wireless Medium Access Control (MAC) and Physical Layer
(PHY) Specifications for Low-Rate Wireless Personal Area Networks
Research efforts on DTN in the literature mainly focus on (WPANs), IEEE Std. 802.15.4, 2006.
data delivery/routing schemes. However, link layer packet [12] M. Caccamo, L. Y. Zhang, L. Sha, and G. Buttazzo, “An implicit
prioritized access protocol for wireless sensor networks,” in Real-Time
collision problem should not be overlooked for DTN based Systems Symposium, 2002. RTSS 2002. 23rd IEEE, 2002, pp. 39–48.
applications, particularly, densely distributed nodes are on [13] A. Lindgren, A. Almquist, and O. Schelén, Quality of Service Schemes
subject. Additionally, certain applications may necessitate for IEEE 802.11 A Simulation Study. Berlin, Heidelberg: Springer
Berlin Heidelberg, 2001, pp. 281–287.
specific amount of delay tolerance for data delivery rather
than completely disregarding it even they perform under
DTN model. Based on such DTN-based applications under
multi-uav system scenario, Deadline Sensitive Medium Ac-
cess Control (DSMAC) protocol is presented in this paper.
The DSMAC protocol is a novel medium access solution
that seeks to achieve timely data transfer as possible with
contention-free and centralized access mechanism. It enables
UAVs in a DTN-based multi-UAV system to continue their
operation as fast as possible with non-preemptive scheduling
executed on channel access function in the GS. To the best of
our knowledge, medium access in DTNs has not been studied
from this perspective before and this is the first research
effort focusing on deadline sensitivity of channel allocation
for DTN based multi-uav systems.
To investigate the performance of DSMAC protocol, we
have developed an evaluation environment by implementing
it on NS2. In addition, DSMAC has been compared to PCF
mode of widely used IEEE 802.11 protocol in terms of
deadline meeting rate and average waiting time performance
measures. Simulation results show that DSMAC protocol
provides considerable performance improvements in terms
of average waiting time and deadline meeting rate.
R EFERENCES
[1] K. Fall, “A delay-tolerant network architecture for challenged inter-
nets,” in Proceedings of the 2003 Conference on Applications, Tech-
nologies, Architectures, and Protocols for Computer Communications,
ser. SIGCOMM ’03. New York, NY, USA: ACM, 2003, pp. 27–34.
[2] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Single-copy
routing in intermittently connected mobile networks,” in Sensor and
Ad Hoc Communications and Networks, 2004. IEEE SECON 2004.
2004 First Annual IEEE Communications Society Conference on, Oct
2004, pp. 235–244.
[3] A. Pentland, R. Fletcher, and A. Hasson, “Daknet: rethinking connec-
tivity in developing nations,” Computer, vol. 37, no. 1, pp. 78–83, Jan
2004.
[4] K. Fall, G. Iannaccone, J. Kannan, F. Silveira, and N. Taft, “A
disruption-tolerant architecture for secure and efficient disaster re-
sponse communications,” in Proceedings of ISCRAM, 2010.
[5] R. A. Nichols and A. R. Hammons, “Performance of dtn-based free-
space optical networks with mobility,” in MILCOM 2007 - IEEE
Military Communications Conference, Oct 2007, pp. 1–6.
[6] A. Jindal and K. Psounis, “Contention-aware performance analysis of
mobility-assisted routing,” IEEE Transactions on Mobile Computing,
vol. 8, no. 2, pp. 145–161, Feb 2009.
[7] H. Liang and W. Zhuang, “Dfmac: Dtn-friendly medium access con-
trol for wireless local area networks supporting voice/data services,”
Mobile Networks and Applications, vol. 16, no. 5, pp. 531–543, 2011.
[8] Z. Zhang, “A cluster based delay tolerant mac protocol for underwater
wireless sensor network,” in 2013 International Conference of Wireless
Networks (ICWN2013), July 2013, pp. 85–91.
[9] Y. Zhong, J. Huang, and J. Han, “A delay-tolerant mac protocol with
collision avoidance for underwater acoustic networks,” in 2009 5th
International Conference on Wireless Communications, Networking
and Mobile Computing, Sept 2009, pp. 1–4.
[10] Wireless LAN Medium Access Control (MAC) and Physical Layer
(PHY) Specification, IEEE Std. 802.11, 1997.
____________________________________________________________________________________________________________
35
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract - Nowadays , many companies meet the needs of data The number of studies in Turkey on business intelligence
from different data sources in different formats in order to in systems, which are recognized more and more by companies
line with changing business needs. Data is managed and stored in for their importance, is insufficient. The purpose of this study
different parts of the system. Business intelligence is the most is to emphasize the effect of business intelligence solutions
effective solution that allows to see big picture by integrating all
used in the finance sector, which has a direct relationship with
of the distributed data within a storage. Business intelligence has
emerged as a natural result of the previous system designed to money, on company strategies and explain their advantages
support the decision-making process. Over time, visual for companies.
deficiencies discovered in decision support systems, difficulties of
useage and mismatch between applications, is one of the major II. Strategic Decision-Making
factors in the rise of business intelligence technology. Such
solutions are up to date and integrated view of business Strategic decisions have long-term and profound impacts.
performance it offers the greatest benefits to decision makers., Access to complete and accurate information in line with
business needs on time has made decisions more structured.
Data warehouses store the data constantly cleaned and
By increasing centralization of data quality, control and
scheduling capabilities have allowed us to take quick and right
updated with ETL (Extract-Transform-Load) processes, which
decisions in the evolving competitive environment.The concept of are designed in a planned way, and fulfill the quality data
business intelligence is an important element of taking strategic element, which is one of the important factors in the decision-
decisons and implementation point in globalized world.This making process. Softwares that allows for the management of
study has designed by Oracle business intelligence tool and the business intelligence process contribute to the decision
results have been a key element of evaluation in decision making making process by bringing all the technical work to a level
processes of the companies. which the decision maker will understand using visual tools
such as display panels, pivot tables or graphs.
Keyword – Decision making, business intelligence, big data Decision-making and action-taking occur after the
intervention of many factors simultaneously. Considering
I. Introduction today’s conditions, competition between companies seem to
Rapid changes in the business life and complex situations be getting fiercer with each passing day with emerging
that come along with these changes force businesses to make technology. To be able to survive in the dynamic business life
fast and accurate decisions. This journey started with the use is not possible with adopting emerging technologies to the
of decision support systems which are being replaced by company alone. The cornerstone of fast and effective
business intelligence technologies due to changing business decision-making is to provide analyses designed in accordance
needs. with today’s needs to competent decision makers who are able
to accurately assess results of these analyses, which provides a
Instead of depending on their personal predictions and considerable competitive advantage for companies.
intuition, decision makers must adopt approaches based on
facts in order for companies to have competitive advantage. It has been found in recent studies that most companies
With the introduction of business intelligence tools, cannot use corporate data in decision-making processes
centralized data sources enhance data quality and control, as efficiently. The amount of collected data increases
well as shortening decision-making and action-taking time by exponentially with each passing day, which means that the
preventing decision makers from getting lost in different data competition between companies increases at the same pace. It
sources. Centralized data sources may sometimes appear in is of great importance to have different and well-integrated
the form of data warehouses and sometimes in the form of information sources for an effective decision-making process.
data marts containing department- or issue-specific solutions. At this point, business intelligence is provision of a powerful,
____________________________________________________________________________________________________________
36
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
low-cost, easy-to-use and sharable source, which can meet the installation and implementation by saying, "These solutions
needs. have made everything we do easier”. In addition, improved
query performance and more self-serve reporting across the
The resulting heaps of information with changing company are other positive outcomes of the project.
technology were not useful to upper management. A need for
management information systems allowing for the Considering success stories with business intelligence
presentation of integrated and summary data emerged. softwares, the following are observed ([7.8]):
Decision support systems were developed for this purpose.
However, visual inadequacies, usage difficulties and - It is possible to access the desired data at the desired place
integration problems between applications created the need for through visual graphics comfortably and understandably
a new system to achieve this purpose and established a ground without needing IT specialists,
for the introduction of Business Intelligence Systems. - It is possible to access appropriate data among million
lines of data in a shorter time and ensure a faster decision-
III. Business Intelligence making process,
- The huge amount of time spent for data tabulation and
Business intelligence as a term was used for the first time in evaluation can be channelized to customers and decision
1989 by Howard Dresner, who worked as an analyst and makers,
researcher at Gartner Group. Howard Dresner saw business - A lower number of key performance indicators are used,
intelligence as a decision support system based on facts and an - A higher level of report optimization is achieved,
umbrella covering all methods and ideas that enhance the - The simplification of data is much more comfortable.
ability of business-related decision-making.
IV. Experiments
Humanity’s endless search has been an important driving
force in the advance of technology, as in many areas of life. In terms of accounting and financial reporting, the
Findings and results obtained from research have been the information required by banks for analysis and interpretations
harbinger of new technologies of future. Business intelligence such as the following is prepared in a standard way:
designs and softwares are a new technology developed in
- Ensuring uniformity,
order to overcome deficiencies experienced by companies in
- Obtaining uniform balance sheets and income statements
relation to support the decision-making process and increase
directly,
administrative decision-making efficiency, which has began to
- Obtaining information required for auditing and
be used in practice.
supervision in a verifiable and auditable manner,
Business intelligence application softwares are considered - Reaching information required by authorities to monitor
to be the latest point that decision support systems have and direct the economy and other statistical data,
reached. Therefore, they have a wider scope compared to - Financial analysis,
decision support systems. They contain a higher level of - Risk analysis,
analysis and more estimation algorithms. In addition, offering - Efficiency analysis.
a powerful visual tool support is one of the important reasons
Oracle Business Intelligence 11.1.1.7.140527 was used in
why they are used widely today.
the implementation of the study. Within the scope of our
A well-designed technological infrastructure is not enough study, we held meetings with finance group employees in
for a new solution to be adopted widely. To correctly describe different periods in order to determine business needs both
and explain the philosophy behind the solution and its verbally and in written. As a result of these meetings, the
meaning is at least as important as the solution itself. Many study was shaped on a business model covering technical
definitions have been made for business intelligence to analysis documents and queries that end-users desire and may
emphasize its contribution to the decision-making process and desire in future.
effectiveness of administrative decision-making ([1-6]).
Database which was used
Kraft Foods, which has its head office in Illinois, USA,
- Oracle Database 11g Enterprise Edition Release 11.2.0.4.0
employs about 125.000, has a annual revenue of USD 110
- 64bit Production
million with its more than 70 brand, has 40 brands in the
- PL/SQL Release 11.2.0.4.0 - Production
market for more than 100 years and has an asset size of USD
- CORE 11.2.0.4.0 - Production
49 billion and is the second biggest food company in the
- TNS for Linux: Version 11.2.0.4.0 - Production
world, started to use SAP Business Objects BI 4.0 business
- NLSRTL Version 11.2.0.4.0 – Production
intelligence application in 2012. The main purpose of the
project was to make it possible to rapidly and effectively reach Fig. 1 shows a screenshot related to one of the steps of
the rich data in the ERP system of the company through user- creating the physical layer used in the study and Fig. 2 shows
friendly interfaces. Mike Walsh, the Business Intelligence a screenshot related to displaying data.
Director of Kraft Foods expressed his satisfaction with the
____________________________________________________________________________________________________________
37
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 1 Creating physical layer tables Fig. 4 Business model and mapping fact table design-2
____________________________________________________________________________________________________________
38
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Oracle Business Intelligence web interface is a reporting [1]. G.J. Miller, D. Bräutigam and S.V. Gerlach, Business Intelligence
Competency Centers, John Wiley & Sons, Inc., 2006
tool that allows us to reach the data in the system in line with [2]. (2015)http://www.lodestonemc.com/files/pdf/success_stories.
defined targets, perform analyses that make is possible to [3]. H. Ateş, “Karar Vermede İş zekasının Önemi Tekstil Sektöründe Bir
make healthy decisions and present these with visual richness. araştırma”, Yüksek Lisans Tezi, Dokuz Eylül Üniversitesi, 2008.
Once the Data Mart design, which is the first phase of the [4]. A.S. Yılmaz, “Esnek raporlama aracı ve iş zekası uygulamaları ile
bütünleştirilmesi”, Yüksek Lisans Tezi, Ege Üniversitesi, 2010.
study, is completed, we can go on with the next phase, which [5]. M. Gibson, D. Arnott, and I. Jagielska, “Evaluating the Intangible
is providing an Oracle web interface for end-users. This phase Benefits of Business Intelligence Review & Research Agenda”,
which emerges after needs analysis allows end-users to create Proceedings of the 2004 IFIP International Conference, 2004, s. 295-
reports using a drag and drop technology. The relevant 305.
[6]. J. Reinschmidt, and A. Francoise, IBM Business Intelligence
screenshot is given in Fig.7. Certification Guide, pp.3-4, 2000.
[7]. (2015) Hannon, http://sapinsider.wispubs.com/Assets/Case-
Studies/2012/April/Kraft-Foods-Recipe-For-Successful-Business-
Intelligence.
[8]. (2015) Crider, http://www.sap.com/customer-testimonials/high-
tech/dell.html.
V. Conclusion
Reports obtained as a result of the solution designed within
the scope of the study allows us to find customer details based
on the ledger account number.
____________________________________________________________________________________________________________
39
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— A fairly general model of the elevator system is II. THE ELEVATOR SYSTEM
presented. Coloured Petri Nets (CPN) and CPN tools are adopted Elevator systems are an integral aspect of most buildings.
as modeling tools. The model, which is independent of the number
Transportation between floors, often with heavy goods, is in
of floors and elevators, covers different stages of the elevator
system in substantial detail. The model assists simulation-based
many cases almost impossible just by using stairs, especially
analysis of different algorithms and rules which govern real for high-rise buildings. Quite often multiple elevators are
elevator systems, including calculating serving time and waiting required and such systems are usually very complex. Multiple
time. The results prove the compatibility and applicability of this elevators must be controlled by a centralized control
model in various situations and demonstrate the expressive power mechanism. The complexity of these elevator systems arises
and convenience of CPN. from factors such as scheduling needs, resource allocation, and
stochastic control, to name a few. Handling these jobs usually
results in systems behaving as discrete event systems [12].
Keywords— Formal Specification, Elevator System, Software Moreover, the differences among the types of buildings and
Specification Benchmarks, Coloured Petri Nets, Calculating their traffic patterns also add to the complexity of the elevator
Service Time, Waiting Time systems. For example, elevator passengers in residential,
institutional, or commercial buildings might face some mix of
popular traffic patterns as: up-peak traffic (also called incoming
I. INTRODUCTION traffic) where the traffic flows mostly from the first floor to
Coloured Petri Nets (CPN), first proposed in [7] and later other floors, down-peak traffic (also called outgoing traffic)
substantially modified and enhanced in [8] and [9], are an where the traffic flows mostly to the first floor from other floors,
extension of Petri Nets (c.f. [13]), and are often used to model and balanced traffic (also called random traffic) where none of
behaviours of large variety of complex systems. Nevertheless, the two previous patterns dominates [2], [14].
the question whether or not CPN are an effective technique for The elevator system is usually defined as follows [5]: An
modeling real world applications of interconnected elevator system is to be installed in a building with m floors and
communicating systems is still of interest for software n cars. The elevator and the control mechanisms are supplied
developers and modelers. In this paper we discuss a CPN model by the manufacturer. The internal mechanism of an elevator
of the elevator system. The elevator system is one of the system is assumed (given). The problem concerns the logistics
software engineering benchmarks that are frequently used to of moving cars between floors according to the following
test the expressive power, readability and convenience of constraints:
various formal specification techniques [5].
This paper is an extension and refinement of the results • Each elevator’s car has a set of buttons - one for each
presented in [1], however it can be read independently. The floor. These buttons illuminate when pressed and signal
main difference is calculating service time and waiting time. the elevator to move to the corresponding floor. The
Various types of Petri nets have been used to model the illumination is canceled when the corresponding floor
elevator system before, a detailed list of references can be found is visited by the car.
in [1], here we mention only [4], [6], [10] and [11]. • On the wall outside the elevator each floor has two
Nevertheless, all the previous models (except [1]) are either buttons (with the exception of the ground and the top
static, or the concept of colour as a data type was not fully floors). One button is pressed to request an upward
utilized, or other formalisms as UML were substantially moving elevator and another button is pressed to re-
involved. quest a downward moving elevator. If both buttons are
The model presented in this paper is independent on the pressed, then each direction is assigned to a different
number of floors and elevators, it covers in substantial detail car. These buttons illuminate when pressed. The
different stages of the elevator system. We believe our model illumination is canceled when the assigned car visits the
is flexible enough to be adapted to different algorithms and floor.
rules, and may eventually evolve in a ’standard’ formal model • When an elevator has not received any requests for
of the elevator system. service, it should be held at its parking floor with its
doors closed until it receives further requests.
____________________________________________________________________________________________________________
40
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
• All requests for elevators from floors (i.e. hall calls) have been used. For more details and theory of CPN, the reader
must be serviced eventually. The applied algorithm is referred to [9].
controls the priority of floors. Throughout this paper we will use ℝ (ℝ ) to denote Reals
• All requests for floors within elevators (i.e. car calls) (non-negative Reals), ℤ (ℤ ) to denote Integers (non- negative
must be serviced eventually, with floors usually serviced Integers), [x1, . . . , xn] to denote the list x1, . . . , xn and lists(X)
sequentially in the direction of travel. to denote finite lists built from the elements of X.
• Each elevator’s car has an emergency button which
when pressed causes a warning signal that is sent to the
site manager. The car is then deemed "out of service". IV. CPN-BASED MODELLING OF ELEVATOR SYSTEM
Each car has a mechanism to cancel its "out of service" Our model of the elevator system consists of four major
status.
interconnected but independent sub-models, namely: timing
car-structure, timing hall-call, timing car-call, and timing
system-cycle.
III. COLOURED PETRI NETS Their functions and interconnections are described as
Coloured Petri Nets are a discrete-event modelling language follows. The timing car-structure sub-model represents the
combining the capabilities of Petri Nets with the capabilities of elevator’s cars. It is at the centre of all other sub-models that
a high-level programming language. Petri Nets provide the concurrently control the elevator’s cars. Typically, an elevator
foundation of the graphical notation and the basic primitives for car is requested by either a hall-call or a car-call. A hall-call is
modeling concurrency, communication, and synchronization. placed by pressing a button located in the hallway of a given
Coloured Petri Nets allow tokens to have a data value attached floor while a car-call is placed by pressing a button inside the
to them. This attached data value is called token colour. car of the elevator. When a hall-call is placed, the timing hall-
Although the colour can be of any arbitrarily complex type, call sub-model will assign the hall-call to the appropriate car of
places in CPNs usually contain tokens of one type. This type is the timing car-structure sub-model (details depend on the
referred to as the colour set of the place. algorithms that are used). Similarly, the timing car-call sub-
A semi-formal definition can be given as follows: model coordinates the placed car-calls with the cars of the
A Coloured Petri Net is a tuple: timing car-structure sub-model. Finally, the timing system-
cycle sub-model operates the cars of the timing car-structure
N = (P,T,A,Σ,C,N,E,G,I) sub-model to service the requested calls.
____________________________________________________________________________________________________________
41
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
hall calls are used to transfer the tokens of assigned hall calls The condition (3) guarantees the balance between the
from the timing hall-call sub-model to the timing system-cycle producing process and the assignment process.
sub-model. The other elements of the colour set Timing Cars
are rather self-explanatory. TABLE III
THE COLOUR SET HALL BUTTONS
TABLE I Colour Sets Definitions
THE COLOUR SET TIMING CARS
{(hall call floor, status) | hall call floor
Hall Call
Colour Sets Definitions ∈ Range, status ∈ Status}
Car ID {i | i ∈ ℤ ∧ 1 ≤ i ≤ total number of cars} Hall Calls {[h1,...,hk] | hi ∈ Hall Call}
{x | x ∈ ℤ ∧ lowest floor ≤ x ≤ highest {(IB, USB, UUB) | IB ∈ Hall Calls,
Range Hall Buttons
floor} USB ∈ Hall Calls, UUB ∈ Hall Calls}
Status { up, down, emergency, idle, out of service}
Desired
{[l1,...,lk] | li ∈ Range} initialize_halls'_buttons()
Floors Hall's_Buttons
Hall's_Buttons
Call Issuer {request, system, non, reservation} HB
illuminate_hall_btn(HB,C)
DB
Timing Hall
{[h1,...,lk] | hi ∈ Timing Hall Call}
Calls hall_call(HB,C,DB)
upd_coord(HB,C) C
Timing_Cars
Colour Sets Definitions After firing the transition Release Hall Call, an appropriate
{(current floor, status, destinations,
tuple from either the USB list or the UUB list is removed and
car id, time) | current floor ∈ Range,
Timing Car’s Data placed into both the IB list and place Requested Hall Call. The
status ∈ Status, destinations ∈ Desired
choice between the USB list and UUB list is based on the
Floors, car id ∈ Car ID, time ∈ ℝ }
following rules:
Timing Database {[g1,...,gk] | gi ∈ Timing Car′s Data} 1) When one list is empty, the other list is always selected.
2) The difference between both lengths of lists is less or
B. Timing Hall-call equal to the value of parameter frequency of most
This sub-model executes an algorithm, giving by the user, requested floors.
that assigns a hall-call to the most appropriate car. It also can 3) The internal choice between tuples is sequential for
generate hall-calls from arbitrary floors, selected floors, or both. USB list and arbitrary (non-deterministic) for UUB
Invoking hall calls, which involves button illuminations, list.
starts from place Hall Buttons, which contains a single token The waiting time is calculated in three steps. First, when a hall
from the colour set defined in Table III. In principle, it is a list call is released from place Hall Buttons and put in place
of triples (IB, USB, UUB), where IB - illuminated buttons (IB), Requested Hall Call, the current times of all cars become
USB - unilluminated-specified buttons, and UUB - attached to this hall call (see Table V). Second, when the placed
unilluminated-unspecified buttons, are also (internal) lists. hall call is assigned to a car, it is removed from place Requested
Invocation of a Hall Call requires the firing of transition Hall Call and the lists: desired-floors and served- hall-calls (see
Release Hall Call. This transition is enabled if and only if the Table I) are appropriately modified.
following conditions are satisfied: Third, when the assigned car arrives at the floor of the placed
1) At least one of USB or UUB is not empty,
hall call, then the waiting time is calculated as the absolute
2) If the limit of producing calls is finite, then it has not value of the difference between the time of the car’s arrival and
been already reached, the registered time when the hall call was released. For more
3) The number of produced calls is less than the value of details and particular algorithms, the reader is refereed to [1].
parameter pause number (see Table IV).
____________________________________________________________________________________________________________
42
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE IV The transition Place Car Call is enabled if and only if the
THE PARAMETERS OF THE TIMING HALL-CALL SUB-MODEL
following conditions are satisfied:
Parameters Legal values
1) At least one of the lists UUCB and USCB is not empty,
Producing mode { finite, infinite } 2) The selected car has not reached its maximum number
Times of finite hall
of accepted calls,
{ y | y ∈ ℤ ∧0 ≤ y}
calls
3) If the limit of producing calls is finite, then it has not
Most requested
{[hc1,...,hck] | hci ∈ Hall Call}
been reached yet,
floors
Frequency of most A selected algorithm in Timing Hall-call sub-model may also
{d | d ∈ ℤ ∧0 ≤ d} affect car calls process. Firing the transition Place Car Call
requested floors
Algorithms for modifies two lists: most-desired floors (see Table VII) and
{minimum waiting, nearest, scope}
assigning hall calls illuminated car buttons (ICB), by placing appropriate floor
Travel time {t | t ∈ ℝ } number in both of them respectively.
Average stop times {s | s ∈ ℝ }
Production pause TABLE VI
{p | p ℤ }
number COLOUR SETS CAR BUTTONS AND CALLS COUNTERS
____________________________________________________________________________________________________________
43
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
initialize_counters () initialize_car_call_log ()
Hall Call Car Call
LOG Doors LOG
Hall_Call_LOG Car_Call_LOG
Doors
[arr_guard(CB,car,cnt)] initialize_tdatabase()
upd_arr(car,DB)
Arrive Timing
Database
Database
DB
Timing_Database
CB unilluminate_btn HB unilluminate_hall_btn
(CB,car) (HB,car)
initialize_car's_buttons ()
initialize_halls'_buttons()
Car's_Buttons
Car's_Buttons
Hall's_Buttons
Hall's_Buttons upd_trn(car,DB) DB
Car's_Buttons
Hall's_Buttons
arrive(car,cnt) car
HB return_hall_call
CB reset_car's_btns(CB)
(HB,car)
[trn_guard(car)]
[mnt_guard(car,c)] initialize_tcars()
car
Maintain Timing
CarsCars
car Transfer
transfer(car)
0 Timing_Cars
DB upd_susp(car,DB)
suspend(car)
initialize_tdatabase()
warning
out of service Timing
Database_
Database
to manager
Timing_Cars Timing_Database
car upd_rst
(car,DB) DB
[rst_guard(car)]
Restart
restart(car)
Buttons, and the served-hall-calls of place Timing Cars. fired. In such case the pending car’s token is transferred
Moreover, each delivered hall call or car call is returned to its temporarily to place out of service, which is not accessible by
original list after it was removed from the illuminated-buttons any other sub-models, all assigned hall calls and car calls of the
list or illuminated-car-buttons list, respectively. The doors pending car are returned to places Hall’s Buttons and Car’s
operations are represented by the colour set Doors (c.f. Table Buttons, respectively, and a warning message is sent to the site
VIII). manager, which is denoted by place warning to manager. A
TABLE VIII
suspended car may be restarted, either automatically or
COLOUR SETS OF THE ARRIVAL STAGE manually based on the value of parameter restart cars
automatically that controls the enabling of transition Restart.
Colour Sets Definitions Cars movements between floors are modeled in the
{(car id, current delivery number,
transition stage. Transitions Transfer and Maintain are
Doors deliveries’ total ) | car id ∈ Car ID, current
mutually exclusive. Enabling and firing the transition Transfer
delivery number ∈ ℤ , deliveries’ total ∈ ℤ }
requires that the car desired-floors list is not empty, and the car
Hall Call {(hall call, waiting time) | hall call∈
current floor matches no calls of the desired-floors list. After
LOG Hall Call, waiting time ∈ ℝ } firing transition Transfer, the car token is updated in the
Serving {[(f1, st1), . . . , (fk, stk)] | fi = floori ∈ following way. If the car desired-floor list has calls beyond the
times Range, sti = serving timei ∈ ℝ} car current floor, it continues moving in the same direction.
Car Call {(car id, serving time) | car id ∈ Car Otherwise its direction is reversed. In both cases, the token in
LOG ID,
serving time ∈ Serving Times} place Database is updated accordingly.
Restart cars Once a car reached its desired destination, it is in the arrival
{yes, no}
automatically stage. At this stage, transition Arrive is enabled provided that
transition Maintain is disabled and the car current floor matches
In the maintenance stage, when a car is suspended either by a requested call from the desired-floors list. After firing
an emergency case (i.e. emergency button was pressed) or an transition Arrive, car’s token is updated by dropping the
operation failure case, the transition Maintain, which has the requested floor from the car desired-floor list. Additionally, the
highest priority of firing upon enabling in the entire model, is car state is set to one of three cases. If the car desired-floor list
____________________________________________________________________________________________________________
44
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
has more calls, then it continues serving the requested calls. 140'
The'Average'Wai3ng'in'Seconds'
Otherwise, the car is set to idle, if the car current floor agrees 120'
80'
its parking floor with an appropriate direction.
60'
Nearest'Car'Algo.'
40'
Minimum'Time'Algo.'
20'
V. ANALYSIS 0'
(1 p)'
(1 p)'
(1 p)'
(1 p)'
(1 p)'
(1 p)'
(1 p)'
(1 p)'
(1 p)'
(1 p)'
(2 )'
, d )'
, d )'
, d )'
, d )'
, d )'
, d )'
, d )'
'
CPN models can be analyzed and evaluated by various
n)
up
(3 ,up
(4 wn
(5 wn
(6 wn
(7 wn
(8 wn
(9 wn
,u
u
u
u
u
u
u
u
u
u
ow
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
o
o
o
o
o
o
(1
techniques and tools [15]. In this paper, we used the simulation- Placed'Hall'Calls'
based performance analysis using tools from [3], with various
parameters including numbers of cars and floors, types of
150$
decision algorithms, specified floors, etc. The results of our
130$
simulation-based performance analysis has proved validity and
Average'Wai*ng'Time'in'Seconds'
110$
applicability of our model in various situations.
90$
Due to page limit, we mention only one case study: five cars 70$
serving a twenty-floor building. Among others, we obtained the 50$
Four$Cars$
Five$Cars$
following results. 30$
1) Car calls were produced by timing car-call sub-model 10$
(1 p)$
(1 p)$
d )$
d )$
d )$
d )$
do )$
(1 n)$
(1 p)$
(2 ) $
do $
d )$
d )$
d $
d )$
d )$
d $
d )$
do )$
, d )$
d )$
d )$
d $
d )$
d )$
d $
do )$
)$
!10$
1, p)
4, n)
7, n)
2, n)
5, n)
(1 1,up
(1 own
(1 own
(1 own
(1 own
up
(2 wn
(2 own
(2 own
(2 own
(2 own
(2 own
(3 wn
(3 own
(3 own
(3 own
(3 own
(3 own
wn
w
(2 ow
(2 ow
(3 ow
(3 ow
,u
(2 ,u
buttons were illuminated when released and
0,
8,
9,
(1
3,
4,
5,
6,
7,
2,
3,
5,
6,
8,
9,
0,
1,
3,
4,
6,
Placed'Hall'Calls'
unilluminated when the cars visited the corresponding
floors. Fig. 5: A comparison of Nearest Car and Minimum Time algorithms (top) and
2) Hall calls were produced by timing hall-call sub- a comparison of four cars with five cars (bottom).
model from all floors to request cars. Additionally, the
hall-call buttons were illuminated when released and REFERENCES
unilluminated when the cars visited the floors of the [1] M. Assiri, M. Alqarni and R. Janicki, Modeling Elevator System With
requested hall calls.
Coloured Petri Nets, Proc. of SERP’2015 (Software Engineering Research and
Practice), Las Vegas, Nevada, USA, July 27-30, 2015, pp. 183-189, CSREA
3) All requested hall calls were eventually served for all Press
algorithms used.
[2] G. Barney, Elevator Traffic Handbook: Theory and Practice. Taylor &
4) All requested car calls were served eventually and Francis, 2003.
[3] CPN Tools AIS Group, The University of Technology, Eindhoven, The
sequentially in the direction of car movements.
Netherlands, http://www.cpntools.org.
5) If no calls, all cars were held at their parking floors.
[4] E. S. Etessami and G. S. Hura, “Abstract Petri net based approach to
6) The maintenance stage worked as it supposed to.
problem solving in real time applications,” Fourth IEEE Region 10
The results of two particular experiments are presented in International Conference, pp. 234–239, 1989.
[5] C. Ghezzi, M. Jazayeri, and D. Mandrioli, Eds., Fundamentals of Software
Figure 5.
Engineering, 2nd ed. Pearson Prentice Hall, 2003.
[6] Y.-H. Huang and L.-C. Fu, “Dynamic scheduling of elevator systems over
VI. CONCLUSION hybrid Petri net/rule modeling,” in IEEE International Conference on Robotics
In this paper, a fairly general CPN-based model of the and Automation, vol. 2, 1998, pp. 1805–1810.
elevator system, one of specification benchmarks [5], is [7] K. Jensen, “Coloured Petri Nets and the Invariant Method,” Theoretical
Computer Science, vol. 14, no. 3, pp. 317–336, 1981.
proposed. The model consists of four separated but inter-
[8] K. Jensen, Coloured Petri Nets. Springer, 1994.
connected parts and it emphasis the expressive power and [9] K. Jensen and L. M. Kristensen, “Coloured Petri Nets Modelling and
convenience of Coloured Petri Nets. Our model is quite flexible
Validation of Concurrent Systems.” Berlin: Springer, 2009.
and it allows using different algorithms and different rules at [10] C.-H. Lin and L.-C. Fu, “Petri net based dynamic scheduling of an elevator
ease. Division into four sub-models allows easy tracking of system,” in IEEE International Conference on Robotics and
Automation, vol.
errors and faults. A thorough simulation-based performance 1, 1996, pp. 192–199.
[11] D. Liqian, Z. Qun, and W. Lijian, “Modeling and analysis of elevator
analysis by using the simulator provided in [3].
system based on timed-coloured Petri net,” in Fifth World Congress
on
Intelligent Control and Automation, vol. 1, 2004, pp. 226–230.
ACKNOWLEDGEMENTS
[12] P. J. Ramadge and W. M. Wonham, “The control of discrete event
The first author was supported by the Ministry of Education
systems,” in Proceedings of the IEEE 77.1, 1989, pp. 81–98.
of Saudi Arabia, the second author was supported by Prince [13] W. Reisig, “Petri nets, an introduction, 2nd ed.” Berlin: Springer
Berlin
Sattam bin Abdulaziz University, while the third author Heidelberg., 1991.
acknowledges partial support by NSERC Discovery Grant of [14] G. R. Strakosch and R. S. Caporale (eds), The Vertical Transportation
Handbook, Wiley 2010.
Canada.
[15] W. M. P. van der Aalst and C. Stahl, Modeling Business Processes:
A
Petri Net-Oriented Approach. The MIT Press, 2011.
____________________________________________________________________________________________________________
45
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Land classification is one of the most important stages Land classification, which is the process to estimate the ability
of consolidation projects. The success and timely completion of of the land yield, includes studies relied that the soil,
this project depends on that this classification is useful and fair topography and other features of the land are interpreted and
and are accepted by landowners. Different methods have been carried out in order to make comparisons between certain forms
developed for the classification. Effects on the success of the land
of evaluation [2].
consolidation of the results of these methods are being investigated.
Land classification is one of the most important stages of
In this study, fuzzy logic method has been used for land consolidation projects. The success and timely completion of
classification according to Law No. 5403. In Mamdani Type Fuzzy this project depends on that this classification is useful and fair
Logic, Values of soil index, productivity index and the location and are accepted by landowners. In order to classify lands in
index, which are used to determine the value of the parcel index, many countries of the world, various methods have been
have been defined as input, whereas the value of parcel index have developed and have been put into practice [2].
been defined as the output. Inputs and outputs have been Values of soil index, productivity index and location index is
converted to the linguistic terms (such as very efficient, inefficient, determined when the classification maps for land consolidation
somewhat efficient, remote, near) by creating membership
are generated, 70% of the index obtained the results of the soil
functions. Rule base has been created for calculating of the parcel
index. As a result of fuzzy inference and defuzzification process, etudes (TE) is taken. By adding the index scores determined for
the model formed by Mamdani Type Fuzzy Logic gives the value productivity (VE) and location (KE) to this value, value of
of parcel index. By giving random input values to test generated parcel index is determined. Obtained values of parcel index are
model, results has been compared with results obtained manually. evaluated in particular group ranges, and thus, map of
classification is formed.
Classifications of parcels according to determined the parcel
Keywords— fuzzy logic, land consolidation, land classification, index are made according to Table 1.
fuzzy systems, soft computing Parcel index (PE) = TE*70% + V + k
I. INTRODUCTION TABLE I Classifications of parcels
Land consolidation studies have a set of technical services in
order to increase productivity in agriculture regulating Classification Parcel Index Classification Parcel Index
agricultural space along with a variety of infrastructure services
(road-irrigation network, drainage, soil leveling, construction 1 91-100 6 41-50
of village settlements etc.) [1]. 2 81-90 7 31-40
To give lands equal to their previous lands to landowners after
3 71-80 8 21-30
the consolidation process in land consolidation area,
classification process based on specific criteria of their existing 4 61-70 9 11-20
plots must be made [2].
Land classification according to Soil Conservation and Land 5 51-60 10 0-10
Use Law (Law No. 5403) has been defined as values found
based on soil and productivity etudes and been basis to the
change of land with soil's natural and permanent features and
the distance to settlement or business center of land. The aim of Discovered fuzzy sets, logic and system concepts by Zadeh in
classification is obtained according to certain criteria of the 1965 have occured the result that much nonlinear equations
previous value from land consolidation of lands belonging to are used, method is complicated and solution becomes difficult
businesses. in order to working of this researcher for many years in the
control area and obtaining the control he wants [3].
____________________________________________________________________________________________________________
46
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The concept of fuzzy logic generally tries to modelling the 1) General Information Base Unit: It contains the input
thinking style of people. It is communicable that turbidity variables which are affected of event which will be examined
signify the tentative information as a concept, in other words and all informations about these. The reason which it is called
uncertainty. Fuzzy logic is defined as an artificial intelligence as the general data base is numerical and / or verbal of the
technique which run with uncertainties instead of certainties by informations here.
using verbal variables. Typical value of the verbal variable is 2) Fuzzifier: This is a processor that assigns to membership
expressed with words, such as “hot” or “cold”, also, these degrees in fuzzy sets that has been described as verbal to the
values are represented by membership functions of fuzzy sets. digital input values.
The difference from standart mathematical methods of fuzzy 3) Fuzzy Rule Base Unit: This includes all of rules that
logic is runnig with uncertainties instead of certainties and connect to output variables to inputs in the database and that
allowing the qualitative definition. That uncertainties are can be written in logical IF – IF type. While these rules are
expressed mathematically is evaluated as the greatest written, all intermediate (fuzzy set) connections which may be
convenience which is brought by fuzzy logic the modeling of only between input datas and outputs is considered. Thus, each
complex systems [4]. rule logically connects to the output space to a part of the input
In classical set theory, any object is a member of a cluster or space. All of this contexts creates a rule base.
not. This object only has the value of “0” or “1”. The medium 4) Fuzzy Inference Engine Unit: This is a mechanism that
of these values is not possible. In fuzzy logic unlike classical contains processes community that provide that the system
logic, members belong to partially the fuzzy cluster and is used behaves with output by collecting together all of the piece
in values between '0' and '1'. In the following Figure 1, the relationship established between the input fuzzy sets and the
difference of binary (classical) logic and fuzzy logic is output fuzzy sets. This engine benefits that How to give an
schematically shown. In Figure 1 it is shown that intermediate output under inputs of whole system is determined by huddling
values do not exist in classical logic. together the implications of each rule.
5) Defuzzifier: This converts to sharp digital output values to
The processes of fuzzy logic consists of stages which the results of fuzzy inference obtained result of fuzzy processes.
problem is analyzed and is defined, sets and logical 6) Output Unit: This unit specifies community of the output
relationships are formed, available information is converted values obtained by the interaction by means of fuzzy inference
into fuzzy sets and model is interpreted. By using a lot of engine of information and fuzzy rule bases [6].
prerequisites, it can be decided whether fuzzy logic solves the In this study, fuzzy logic method has been used for land
problem or not. Originally for the problem to be solved, it is classification according to Law No. 5403. In Mamdani Type
decided whether fuzzy logic approach is the right choice or not. Fuzzy Logic, Values of soil index, productivity index and the
If the system's behavior to be applied can be expressed with location index, which are used to determine the value of the
rules or it requires a complex mathematical operation, fuzzy parcel index, have been defined as input, whereas the value of
logic approach can be applied. Otherwise, The results obtained parcel index have been defined as the output. Inputs and outputs
by the fuzzy logic will not give the desired values most likely have been converted to the linguistic terms (such as very
[5]. efficient, inefficient, somewhat efficient, remote, near) by
creating membership functions. Rule base has been created for
calculating of the parcel index. In order to benefit from fuzzy
informations obtained, this informations have to be
defuzzification. Centroid method has been used as
defuzzification method. As a result of fuzzy inference and
defuzzification process, the model formed by Mamdani Type
Fuzzy Logic gives the value of parcel index. By giving random
-a- - b- input values to test generated model, results has been compared
Fig. 1 Representation of the difference between fuzzy logic (a) and classical with results obtained manually.
logic (b)
____________________________________________________________________________________________________________
47
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
curve etc.) will be used. In this study, the seven sub-regions for
soil index, the five sub-regions for productivity index and the
four sub-regions for location index have been determined and
triangular membership function has been used.
0; a≤10;
⎧
⎪ .
10<a≤12;
μ ( ) = 12<a<18;
⎨
⎪ a≥12;
⎩ 0;
____________________________________________________________________________________________________________
48
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
B.Rule Base
Rule TE VE KE PE
Number Fig 9. The relationship with PE of TE and VE
Rule 1 very inefficient far too
inefficient inefficient
Rule 2 very inefficient near too
inefficient inefficient
Rule 3 very inefficient very near very
inefficient inefficient
…
Rule 85 efficient somewhat far somewhat Fig. 10The relationship with PE of VE and KE
efficient efficient
Rule 86 efficient somewhat near very
efficient efficient
Rule 87 too efficient inefficient very near very III.CONCLUSION
efficient
The method developed in this study has been applied the first
…
time for the land classification. according to Land Law No.
Rule 118 too efficient inefficient near very
5403. As shown in Table 3, the results of Fuzzy Logic Model
efficient are compared with results of traditional method. It is seen in
Rule 119 too efficient inefficient very near too Table 3 that both the results obtained from the fuzzy logic
efficient model and the results obtained from the traditional method are
Rule 120 too efficient inefficient too near too
efficient
similar one another. The system is fast and is more accurate
than traditional methods. This system also has a high reliability.
Besides, accuracy of established fuzzy system has been shown
on stabilizing right graph that its angle is 45 degrees (Figure
11).
____________________________________________________________________________________________________________
49
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III Comparison of the results of Traditional Method [4] Gülbağ, A., 2006, Yapay Sinir Ağı ve Bulanık Mantık Tabanlı
Algoritmalar İle Uçucu Organik Bileşiklerin Miktarsal Tayini, PhD
and Fuzzy System
Thesis, Sakarya University, Graduate School of Natural Sciences,
Sakarya-Turkey.
TE V K PE According to PE According to [5] Elmas, Ç., 2007, Yapay Zeka Uygulamaları, Seçkin Yayıncılık, Ankara-
the Traditional the Fuzzy Logic Turkey.
Method System [6] Yılmaz, M. and Arslan, E., 2005, Bulanık Mantığın Jeodezik
Problemlerin Çözümünde Kullanılması, 2. Mühendislik Ölçmeleri
83 9 18 85 90 Sempozyumu, 23-25 Kasım İstanbul-Turkey.
74 8 15 75 73.2
60 5 8 55 59.5
71 7 13 70 70.2
55 8 17 63.5 67.4
95 9 12 87.5 90
80 5 14 75 79
78 6 12 73 77.7
10 2 3 12 13.6
50 5 10 50 53.5
61 4 7 53.7 52.8
40 3 11 42 47.4
15 3 4 17.5 17.6
30 2 6 29 23.6
23 6 4 26.1 23.5
78 7 10 71.6 73.4
100
Tradational Method
75
50
25
0
0 25 50 75 100
Fuzzy Logic Model
Fig. 11 Accuracy of Fuzzy Logic Model with the help of 45 degrees right
REFERENCES
____________________________________________________________________________________________________________
50
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract — Limited battery supply of a sensor node is one of the sensor nodes. Because WSNs often use broadcast based radio
most important factors that limit the lifetime of the WSNs. As a communication.
consequence, prolonging the lifetime of WSNs through energy Limited energy availability in sensor nodes makes network
efficient mechanisms has become a challenging research area. lifetime an important issue in WSN applications. To extend the
Previous studies have shown that instead of implementing direct
network lifetime, energy efficient wireless sensor network
transmission or multi-hop routing, clustering is a kind of key
technique used to reduce energy consumption. Clustering can protocols and algorithms have been devised in the literature.
increase the scalability, decrease the energy consumption and Node clustering, in-network data processing, data fusion and
extend the lifetime of the network. Also, energy-efficient clustering network coding are some of the measures taken to reduce the
protocols have been designed for the characteristic of amount of data that is processed, sensed or transmitted.
heterogeneous wireless sensor networks to obtain additional Minimization of energy spent in processing, sensing and
energy savings. In heterogeneous wireless sensor networks, some transmission of data allows sensor nodes to save energy. Such
of sensor nodes is equipped with additional energy resources. In energy savings help to extend the lifetime of WSN applications.
this paper, comparison of stable election protocol (SEP), The main goal of cluster-based routing protocol is to
distributed energy-efficient clustering (DEEC) scheme with
efficiently maintain the energy consumption of sensor nodes by
LEACH which has also advanced nodes is aimed. Same simulation
parameters are used for comparison. The nodes in all involving them in multi-hop communication within a cluster
heterogeneous algorithms are equipped with same total energy. and by performing data aggregation and fusion in order to
The protocols are compared in terms of number of alive nodes, decrease the number of transmitted messages to the sink and
lifetime and energy-efficiency in MATLAB. The results of the transmission distance of sensor nodes.
simulations are discussed in details. There are two kinds of clustering schemes. The clustering
algorithms applied in homogeneous networks are called
Keywords— Energy Efficient Routing, Clustering, Wireless Sensor homogeneous schemes, and the clustering algorithms applied
Networks. in heterogeneous networks are referred to as heterogeneous
clustering schemes. It is difficult to devise an energy-efficient
I. INTRODUCTION heterogeneous clustering scheme due to the complicated energy
A WSN is obviously formed by a collection of sensor nodes configure and network operation. Thus most of the current
and corresponding protocols for routing messages in this clustering algorithms are homogeneous schemes, such as low
network. A typical wireless sensor node consists of a sensory energy adaptive clustering hierarchy (LEACH) [1] and their
unit, a communication unit, a power unit and a processing unit. variants [2].
Sensory unit consists of a data acquisition component and ADC The most successful heterogeneous clustering protocols are
(converts the sensed real world data to the digital form). stable election protocol (SEP) [3] and distributed energy-
Communication unit has a radio transceiver and the power is efficient clustering scheme (DEEC) [4].
backed by a battery source. The WSN node can be equipped LEACH is one of the most popular distributed cluster-based
with a limited power source (0.5 Ah, 1.2 V). The node remains routing protocols in WSNs. LEACH randomly selects a few
active as long as the battery is alive and hence power saving is nodes as CHs which aggregate data arriving from nodes and
a crucial criterion in this domain of applications. Energy forward the aggregated data to BS and rotates this role to
consumption happens in three domains: sensing, data balance the energy consumption of the sensor nodes in the
processing and communications. The sensing, signal network.
processing parts operate at low sequential and consume less In LEACH protocol, the time is divided into parts called a
than 1mW. This is over an order of extent less than the energy round. Each round consists of two phases. The first phase is set-
consumption of the communication part. Therefore we have to up phase which is the phase of node formation. The second
assign less communication/data exchange between sensor phase is related to the normal function of the network and is
nodes. called the Steady-State phase. In the first phase, the CHs are
The traditional infrastructure based protocols cannot be used elected based on a probability function. This election is as
for WSNs because of the limited storage and power of the follows: each sensor node selects itself to be CH at any given
time with a certain probability. Any node in the network
____________________________________________________________________________________________________________
51
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
chooses a random number between 0 and 1. Then this number DEEC, there are advanced and normal nodes. m is the fraction
is compared with threshold limit, if the number is less than a of the advanced nodes and these nodes have a times more
threshold, the node becomes a CH for the current round. This energy than the normal ones. Therefore, DEEC network has am
probability function is designed in such a way that within a times more energy and virtually am more nodes. The results of
specific number of rounds each sensor becomes a CH only once the simulations of DEEC protocol indicate that, DEEC
and thus the energy consumption is distributed over the whole prolongs the time of first node death when compared with
network. After the set up phase of the round, where the CHs are LEACH variants and SEP in heterogeneous networks.
elected, each CH announces its election to other nodes and each
node chooses a suitable (nearest) CH for itself; and then it II. SIMULATION ENVIRONMENT AND PARAMETERS
announces this decision to the related CH and thus the clusters The simulations are conducted in MATLAB. 100 sensor
are formed and the network comes into the steady-state nodes are randomly deployed in a 100 m x 100 m field as shown
operation i.e data transmission. Then each CH creates a TDMA in Fig. 1 and BS is placed outside of the sensor field which have
schedule in each cluster to organize the communication among coordinates of (150,50). Same simulation parameters of
cluster members. When the non CH nodes receive the TDMA
LEACH are used for all simulations in this paper. Each
scheme, nodes send their data to the CH once per frame during
simulation is realized for 100 independent iterations to obtain
their allocated transmission TDMA slots. This allows the radio
components of each non-cluster-head node to put the sleep more scalable results. Table I summarizes the simulation
mode at all times except during its transmit time, thus environment parameters used for simulations.
minimizing the energy dissipated in the individual sensors.
After a certain amount of frames i.e. a TDMA round, the
network re-elects CHs and re-form clusters. Then CH, after
combining all the received data, will send the results to the BS.
Besides, to prevent the interaction of the transmissions which
occur in clusters at the same time, LEACH uses different
CDMA codes. Existing CH chooses randomly a unique code
from a list of spreading codes. The CH filters all received
energy using this spreading code. Consequently, the radio
signals of the neighbouring nodes are filtered out hence
interference of the transmission of the nodes is minimized.
While many homogeneous LEACH variants have been
developed as described above, there are also heterogeneous
cluster based WSN routing protocols. SEP is a fundamental
protocol for WSNs and contains advanced nodes which are
fitted with extra energy resources. SEP uses a weighted election
probability based approach to determine CHs according to the
residual energy of each node. In SEP, m corresponds to fraction Fig. 1. The randomly deployed sensor nodes in a WSN.
of the advanced nodes which are fitted with a times more
energy than the normal nodes. As a consequence, the total TABLE I. SIMULATION ENVIRONMENT PARAMETERS
initial energy of the WSN is increased by 1+a.m times. The
additional energy of the advanced nodes forces them to be
elected as CHs. Each node is informed the total energy of the Parameters Values
network in order to adjust its election probability to become a Network area 100 m x 100 m
CH according to its residual energy. The remaining energy Number of nodes 100
values of normal and advanced nodes are transmitted to the Base station coordinates (150,50)
CHs while members send data. The remaining energy values of Initial energy per node 2J
the nodes are delivered to the BS by using CHs. The results of Data packet size 6400 bits
the simulations of the SEP show that, SEP provides significant Control Packet Size 200 bits
energy savings, lifetime gains and throughput improvement Transceiver Energy (Eelec) 50 nJ/bit
when compared with LEACH for both homogeneous and
Aggregation Energy per Bit (EDA) 5 nJ/bit/signal
heterogeneous scenarios.
Free Space Amplifier Energy (εfs) 10 pJ/bit/m2
DEEC is another heterogeneous and distributed clustering
protocol where the CHs are determined by a probability based Multipath Amplifier Energy (εmp) 0.0013 pJ/bit/m4
on the ratio between residual energy of each node and the
average energy of the network. The nodes which have high
III. SIMULATION RESULTS
residual energy are more probable to become CHs. Adaptive
approach of DEEC provides for controlling the energy In this section, the performance comparisons of LEACH,
consumption of the nodes to provide the energy-efficiency. In SEP and DEEC protocols are made for residual energy, number
of alive nodes and lifetime. To obtain more scalable results, 100
____________________________________________________________________________________________________________
52
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
A. Residual Energy
Fig. 2 illustrates the total residual energy of the nodes in
terms of rounds. With advanced, it can be observed that DEEC
and SEP provide significant energy savings when compared to
LEACH. After 2000 rounds, while LEACH holds 15% of its
initial total energy, SEP and DEEC hold approximately 30% of
their initial total energies. When the network lifetime ends
under LEACH, SEP and DEEC still maintain approximately
10% of its total initial energy. For same simulation parameters,
DEEC and SEP have approximately same performance.
C. Network Lifetime
Fig. 3 shows the number of alive nodes thus the network LEACH 3478 4216 200 J
lifetime for LEACH, SEP and DEEC. Under SEP and DEEC,
the network lifetime increases from 3237 rounds to 5000
rounds, approximately 50% improvement compared to LEACH. IV. CONCLUSIONS
Although, node deaths start earlier under DEEC, since the CHs Nowadays, saving energy, extending lifetime and designing
are abused, the rate of the number of dead node increases is green networks [5, 6] for WSNs has become an essential
significantly less and this yields to a longer lifetime. One major research area. By placing or determining advanced nodes and
difference between LEACH and DEEC can also be observed providing more energy-efficient CH determination approaches,
from this figure. Right after the first node death under LEACH, DEEC and SEP decrease the energy consumption of cluster
the remaining node deaths will follow quickly. But with DEEC, formation. The results of the simulations demonstrate that, SEP
node deaths are distributed evenly for the network lifetime. SEP and DEEC algorithms provide significant energy efficiency,
provides to delay first node failure in the network because it lifetime gains when compared with LEACH. DEEC also delays
uses advanced nodes. It allows to select advanced nodes to the first node death and provides equal alive node distribution
become CHs more frequently thus it provides to delay the CHs in the WSN.
in the network. SEP also utilises CH determination approach
which provides to select CH nodes according to their residual
energy. References
[1] W. Heinzelman, C. Anantha, and B. Hari, “An application-specific
protocol architecture for wireless microsensor networks,” IEEE
Transactions on Wireless Communications, vol. 1, no. 4, pp. 660-670,
2004.
[2] M. Islam, et. al., “Survey of Extended LEACH-Based Clustering
Routing Protocols for Wireless Sensor Networks,” IEEE 9th
____________________________________________________________________________________________________________
53
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
54
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
mehmetalbayrak@sdu.edu.tr
volkan.altintas@cbu.edu.tr
³Graduate School of Nature and App. Sciences, Süleyman Demirel University, Isparta, Turkey
alimrts07@gmail.com
kilic.gizem@metu.edu.tr
Abstract— Robots are located at every stage of our lives with According to the research, it was determined that 4-5-6
developing technologies. Interest in the field of robotics is graders have more creative and original ideas, according to the
increasing every day. There isn’t a course on robotics in primary 7th and 8th grade students [3]. In elementary schools of Turkey,
school curriculum in Turkey. Facilities of educational institutions computers first come to mind when term of technology is said.
is limited and primary school teachers is inadequate in robotic. However, we see that take out different terms the mention of
Therefore, many educational institutions aren’t given to students technology in the World. Robotics comes at the beginning of
in robotics education. Shown in programming courses in private these terms.
schools, Lego Mindstorm robot module constitutes awareness in
children from early ages. This study aims to increase the interest
of children in Turkey to robotics. To improve the students' Robotics is an area that needs to be focused on in primary
creativity and motivation, a simple robot design supported with school age children. With applications of robot design, robot
Augmented Reality education related notes have been prepared. contests, robot projects, it is seen that children gain a lot of
A simple model for the implementation of robotics in primary property, such as problem-solving, finding practical solutions
education has been constituted. to problems, learning by doing and experiencing and critical
thinking [4].
Keywords— Robotic, Education, Augmented Reality, Primary This research aims to provide information on the use of
School robotics in the education of primary school age children, and to
help planning applications about robotics in Turkey. It also
I. INTRODUCTION aims to provide the students get closer with the robotic and
technology. In this context, a document prepared that supported
Today, in the face of rapidly evolving technology and by augmented reality robot construc- tion and programming for
increasing information needs to be trained individuals who can the 4th and 5th grade students who are studying in primary
adapt to change. Knowledge of age together with our ability to schools.
quickly increase, based on the technology and the training of
information technology, including new technologies, the II. RELATED WORK
question of how it illustrates the way in the field of educational When the studies related to robotics in schools are analyzed,
technology [1]. it is seen that students were affected in a positive way [5].
A new term in the field of technology is Augmented Reality In the study of Vollstedt, Lego Robotics learning program
(AR) which is some- how can be considered as a different has been performed on about 300 students in secondary schools
version of virtual reality. It allows users to get experience of [6]. As a result of the analysis made, it is seen that students'
using virtual objects in real world. So, Augmented Reality is interest and knowledge of science, technology, engineering and
rapidly growing and developing. This technology supplements mathematics has been increased with robotics [2]. Hacker has
the real world with composite 3D virtual objects that are developed “Robolab” project for his work, and observed the
integrated into the real world. This technology is very effect of robotics on learning basic science and mathematics
interesting and interactive and therefore there is a vast range of principles of 3-6 grade children. With the study, it is seen that
potential applications of Augmented Reality such as in the field the stu- dents were affected positively with learning by doing
of medicine, manufacturing, urban planning, architecture, and living practices [7]. Teixeira uses robotics in secondary
archaeology, education and many more [2].
____________________________________________________________________________________________________________
55
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Augmented reality applications related to education seem to Fig. 1. BMW Maintenance Repair Works
increase in the world and Turkey. Comparing with traditional Marshall company has developed applications for wall color
classes, it is seen that augmented reality applications are test in the house. Images of the application is shown in Figure
increasing the students learning rate [12][13]. As a result of 2 [19].
spreading usage of smart phones, the number of augmented
reality applications and researches on mobile devices are
increasing [14].
____________________________________________________________________________________________________________
56
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
III. METHOD
Turkey’s private schools use robotics kits, "PicoCricket" and
"MINDSTORMS® LEGO® NXT" which has developed by
MIT University and Lego Group, on their robotic courses.
Students produce original designs by using robotic kits. They
learn gears and simple machine mechanisms. They try the
mechanisms they produced using sensors and motors. Robotic
studies aims to gain skills that students use technology, design,
science and mathematical operations about the subject.
As one of the application's functionality, while video Document has benefited from the Lego’s instruction manual
viewing on tablet or phone device pointed on an image , if for the verbal section. The document consists of 3 pages. Each
the device removed from image video stops and if the device pages has and image supported by Augmented Reality. First
replaced back on the image, video will continue where it left. visual is made public information about the robotic kit. The
second visual was introduced kind of robots that can be made
For example, on Figure 4, the application executed by a using Lego blocks. There is also information in latest visual that
picture, an animation presented to the user's access. Simple sensors can be used. An example is shown in Figure 6 from the
images can be a gate to easily access different videos or document.
animations [22].
____________________________________________________________________________________________________________
57
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
58
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Somyürek, S. Öğrenme Sürecinde Z Kuşağının Dikkatini Çekme:
Arttırılmış Gerçekçilik.Eğitim Teknolojisi Kuram ve Uygulama, Cilt 4,
1 (2014)
[2] Android Mobile Augmented Reality Application Based on Different
Learning Theories for Primary School Children . Parhizkar , Behrang
(2012)
[3] İlköğretim Öğrencilerinin Bilim İnsanına Yönelik Düşüncelerinin
Değerlendirilmesi.Nuhoğlu, H. ve Afacan, Ö. : 16. Ulusal Eğitim
Bilimleri Kongresi (2007)
[4] Growing up with Robots. Costa, M. F. ve Fernandes, J.: Proceedings of
Hsci, (2004)
[5] Fen ve Teknoloji Eğitiminde Teknoloji Tabanlı Öğrenme: Robotik
Uygulamaları. Koç,Ayşe ve Böyük, Uğur. 1,: Türk Fen Eğitim Dergisi
(2013)
[6] Vollstedt, A. M.Using Robotics to Increase Student Knowledge an
Interest in Science,Technology,Engineering and Maths. University of
Nevada,Reno : A Master Thesis of Science in Mechanical Engineering
(2005)
[7] Hacker, L.Robotics in Education: ROBOLAB and Robotic Technolgy
as Tools for Learning Science and Engineering.: Tese de licenciatura
apresentada ao Department of Child Development da Tufts University
(2003)
[8] Teixeira, J. C.Aplicaçoes da Robotica no Ensino Secundario:o Sistema
Lego Mindstorms e a Fisica.: Tese de Mestrado (2006)
[9] Robotic Training to Bridge School Students with Engineering. Balaji,
M., ve diğerleri.: IEEE International Symposium on Robotics and
Intelligent Sensors (2015)
[10] Motivational design for learning and performance . Keller, J. M.:
Springer (2010)
[11] Teaching based on augmented reality for a technical creative design
course. Wei, Xiadong,ve diğerleri. 81: Computers and Education (2015)
[12] SMART: a System of Augmented Reality for Teaching 2 nd grade
students. Freitas, R. ve Campos, P.: In Proceedings of the 22nd British
HCI Group Annual Conference on People and Computers: Culture,
Creativity, Interaction (2008)
[13] Making it real: exploring the potential of augmented reality for teaching
primary school science. Kerawalla, L., ve diğerleri. 10,: Virtual Real
(2006)
[14] Mobil Artırılmış Gerçeklik Uygulamalarının İncelenmesi ve
Eğitimsel Açıdan Değerlendirilmesi. Demirer, Veysel, Erbaş, Çağdaş:
Mersin Üniversitesi Eğitim Fakültesi Dergisi (2015)
[15] Eğitimde Artırılmış Gerçeklik Uygulamaları: Google Glass Örneği.
Erbaş, Çağdaş ve Demirer, Veysel. 3: Journal of Instructional
Technologies & Teacher Education, Cilt 2 (2014)
[16] İngilizce Öğreniminde Artırılmış Gerçeklik: Öğrencilerin Başarı, Tutum
ve Bilişsel Yük Düzeyleri. Küçük, Sevda, Yılmaz, Rabia M. ve Göktaş,
Yüksel. 176,: Eğitim ve Bilim, Cilt 39 (2014)
[17] Eğitim Ortamlarında Artırılmış Gerçeklik Uygulamaları. Çetinkaya, H.
H. ve Akçay, M. :Akademik Bilişim-15 (2013)
[18] BMW Augmented Reality,
https://www.youtube.com/watch?v=P9KPJlA5yds
[19] Augmented Reality Nedir?, http://www.teknolo.com/augmented-
reality-nedir-hangi- alanlarda-kullanilabilir/
[20] Realidad Aumentada,
https://i.ytimg.com/vi/vDNzTasuYEw/maxresdefault.jpg
[21] BilimselDüşünme,http://www.bahcesehir.k12.tr/tr/egitim/detay/Bilisims
el-Dusunme/22/82/0
[22] 3D Augmented Reality,http://creativesatworkblog.com/wp-
content/uploads/2012/02/CAWAurasma.jpg
____________________________________________________________________________________________________________
59
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
60
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the original watermark, respectively. Then, the singular signal processing area. Image compression, data hiding,
values of the medical image is added to the singular values noise reduction and image watermarking are just some
of chaotic watermark in all sub-bands. Finally, by using applications of SVD [6].
invers SVD and inverse DWT the watermarked image is
obtained. Given the data matrix A of dimension N x N which has the
independent column (i.e. rank (A) = N), there are two
The paper is organized as follows. In section II, in order unitary matrices V and U such that
to better understand the proposed algorithm, DWT, SVD
and ACM are discussed. Watermark embedding and 𝐀 = 𝐔𝐒𝐕 𝐓 (1)
extracting algorithms are explained step by step in section Where S= diag(σ1, σ2, … , σW) is an diagonal matrix.
III. Section IV gives the experimental results and compares Singular values are ordered as σ1≥ σ2≥…≥σW>0. The
the proposed method with related studies. Finally, Section Equation (1) is mathematical the representation of SVD.
V concludes the paper.
SVD is an optimal decomposition method that
II. APPLIED TECHNIQUES concentrates the maximum signal energy into as few
coefficients as possible. From the image processing point of
In this section, the techniques used in the proposed method
view SVD has three main advantages: i) it can applied to
are explained briefly to better understand the watermark
any dimension of image matrix, ii) after applying SVD to
embedding and extraction steps. image the singular values S corresponds to the brightness of
A. Discrete Wavelet Transform (DWT) image while U and V denotes the geometric properties of
the image, iii) the slight variations of singular values of an
DWT is a common tool used in watermarking schemes
image may not affect the human visual perception. This
which transforms the image from spatial domain to stability property of SVD is the main reason why it is
frequency domain. It is shown in many papers that DWT is preferred for watermarking applications [7].
superior then other transform techniques in watermarking
algorithms. By applying DWT, 1-D original signal is
separated into the low-frequency and high frequency sub- C. Arnold’s Cat Map (ACM)
bands. This process is called as decomposition and gives us
DWT coefficients. The original signal can be reconstructed
by using DFT coefficients. This reconstruction process is Chaotic signals has very broad range of application in
referred as inverse DWT (IDWT) [5]. secure communications, signal processing and
cryptography because of their inherent properties that can
By performing DWT on image (2-D signal) one time, the be taken into account such as high complexity. There are
image is split into four sub-bands, LL1, LH1, HL1, and HH1. many chaotic maps in literature appropriate for image
processing applications. ACM is one of the most famous
LL1 is the low frequency sub-band and it contains the
chaotic map used for randomizing the pixel locations in the
maximum of energy while the other sub-bands are middle
(LH1, HL1) and high (HH1) frequency sub-bands .These image matrix. This randomizing provides high security for
the image watermarking schemes. ACM can be applied for
middle and high frequency sub-bands represent the edges,
only square matrix. For N x N square image matrix it can
outline, texture and other detail information of the original
be expressed as
signal. If we want to perform again DWT to signal, we
should apply the DWT to the LL1 sub-band. This process 𝑥𝑛+1 𝑥𝑛 𝑥𝑛
1 𝑎
can be repeated until we get the appropriate decomposition [𝑦 ] = [ ] [ ] 𝑚𝑜𝑑 𝑁 = 𝑨 [ 𝑦 ] 𝑚𝑜𝑑 𝑁. (2)
𝑛+1 𝑏 𝑎𝑏 + 1 𝑦𝑛 𝑛
level for our application.
where (xn,yn) and (xn+1, yn+1) are the locations of pixels
Watermark embedding process is done by changing the before and after iterations, respectively. In Equation (2) the
wavelet coefficients according to the embedding algorithm. coefficients a and b are selected such as det(A)=1 [5].
The watermark can be embedded either in low- frequency
sub-band or high- frequency sub-band with respect to the After applying a few iterations into original image the
purpose of the application. Embedding watermark to the locations of pixel will be scrambled but when the iterations
high- frequency sub-band provides high imperceptibility are repeated enough we will attain the original image again.
advantage but robustness and stability of the scheme will There is no any formula for exact number of iterations for
be decreased. Robustness can be improved by embedding given dimension of image, but for higher dimension we
watermark to the middle a frequency sub-band but this will expect higher iterations. ACM transformation for different
cause decrement of imperceptibility. iterations is illustrated in Fig.1. As a conclusion the number
B. Singular Value Decomposition (SVD) of iterations both in watermarking embedding and
In recent years, SVD has become an extremely valuable tool extracting must be selected carefully [10].
in the analysis and the solution of problems especially in
____________________________________________________________________________________________________________
61
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
62
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
63
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
64
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Most of the machine learning and data mining discretise using an entropy-based discrete method. At [3], the
algorithms use discrete data for the classification process. But, results of comparison between using discrete method and
most data in practice include continuous features. Therefore, a continuous method for six datasets showing that the
discretization pre-processing step is applied on these datasets performance of the classification accuracy is improved, when
before the classification. Discretization process converts
the features of datasets discretise. Studies for Cluster Algorithm
continuous values to discrete values. In the literature, there are
many methods used for discretization process. These methods are also can be used as discretise values. [4] proposed accounting
grouped as supervised and unsupervised methods according to the interdependencies among different attributes and discretise
whether a class information is used or not. In this paper, we used the data using minimum entropy with minimum description
two unsupervised methods: Equal Width Interval (EWI), Equal length as the stopping criteria, used K-mean clustering methods
Frequency (EF) and one supervised method: Entropy Based (EB) and shared nearest neighbour. While [5] used the K-means
discretization. In the experiments, a well-known 10 dataset from clustering algorithm for discretization.
UCI (Machine Learning Repository) is used in order to compare In this study, we applied supervised EB and unsupervised
the effect of the discretization methods on the classification. The EW and EF discretization methods on 10 UCI dataset with
results show that, Naive Bayes (NB), C4.5 and ID3 classification
continuous values. These datasets are applied to classification
algorithms obtain higher accuracy with EB discretization method.
algorithms that work on discrete features, and then the accuracy
of classification methods are compared. At the next section, the
Keywords— Discretization, Supervised and Unsupervised
discretization methods unsupervised EW & EF and supervised
Discretization, Continuous Features, Discrete Feature,
classification algorithms. EB are explained. At section 3, the classification algorithms are
given. The experimental results are shown at Section 4 and the
I. INTRODUCTION paper is concluded at Section 5.
Many Machine Learning and Data Mining classification
II. DISCRETIZATION AND CLASSIFICATION OF DISCRETIZATION
algorithms have application possibility only to the discrete data.
METHODS
But, most data in practice have continuous feature. The datasets
usually contains mixed forms of nominal, discrete, and Data discretization process is a method aims to reduce the
continuous data. Discrete values have intervals between a volume of distinct values of continuous variables through
continuous series of values. The number of continuous values dividing its range into limited set of unrelated intervals and then
for an attribute can be endless, but the number of discrete values relating these intervals by specific descriptive labels. Usually
may be few or have an end value [1]. An example to continuous discretization steps are sorting continuous values, finding cut
features is blood sugar content. Whereas an example of discrete points and finally applying conversion process [1].
features is gender. Process of converting continuous values to Discretization methods are categorized along different needs,
discrete values is called discretization. Discretization is discretization of continuous values to obtain higher accuracy
generally used to sort and reshape continuous variables of rate on handling data with high cardinality. Main classification
attributes into categorized features. However, there are endless of discretization is as supervised and unsupervised. Generally
possibilities of discretization methods depending on the the categorization of the supervised and unsupervised
intervals which exist within domain. The idea of discretization discretization techniques depends on class information, for
is to divide the range of a numeric or ordinal attribute into example if the discretization process uses class information,
intervals through user given or computed cut points. [2] made then we say it is supervised discretization. Otherwise, it is
a comparison between unsupervised discretization method called as unsupervised discretization [1], [2]–[3].
(Equal width, Equal frequency) and supervised discretization
method (Entropy-based, Purity-based) from classification
A. Unsupervised Discretization Methods
accuracy point of view. They found that the classification The simple discretization (equal-width and equal-frequency
accuracy of the classification algorithms (Naive-Bayes, C4.5) interval binning) is among the unsupervised discretization
significantly improved when features of the datasets were methods and binning does not use class information.
____________________________________________________________________________________________________________
65
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Continuous ranges are subdivided into smaller ranges through A = {a0, a1, a2, .... an-1, an} where n is a number of element in
user specified width or frequency. Usually in the unsupervised A. And A is set of data with array having continuous values [1],
discretization methods, the number of parts must be supplied [2], [5]–[7]. In proportional K–interval discretization method,
by the user [1], [2]–[6]. the data instances with identical value must be placed in the
same interval, so it is not always possible to generate exactly K
1) Equal Width Interval Discretization: The simplest
equal frequency intervals [7]. Equal Frequency discretization
discretization method is Equal-width interval discretization
steps as shown in the Fig. 2. And these steps are given as an
which divides the range of observed values for a feature into
example in Table II below. Where I is the instance, T is the
equal sized bins represented by k that is a parameter supplied
temperature, C is the classes value, A sorted value of T and D
by the user [1]–[2]. The process involves finding the minimum
is a discretised value of T.
and maximum observed values through sorting of a continuous
feature, A = {a0, a1, a2, .... an-1, an}, amin = a0 and amax = an. A is
Input: data is an array having continuous values, A = {a0,
a continuous value array. Computing the interval may be done
a1, a2, ….. an-1, an} and k is the number of parts.
by dividing the range of the observed values for the variable
Where k > 0;
into equally sized bins using the following formula: [1], [2]–[5].
Output: data having discrete values.
Step 1: Sort the value A in increasing order
𝑎𝑚𝑎𝑥 − 𝑎𝑚𝑖𝑛
𝐼𝑛𝑡𝑒𝑟𝑣𝑎𝑙 = 1 Step 2: Determine the split point for A and calculate the data
𝑘
instances in each interval by dividing the number of
Boundaries = amin + (i x interval) elements in the array by number of parts.
Step 3: The value of an attribute in the array must be placed
The boundaries can be formed by i = 1...k-1 using the above in the same boundaries
formula. Equal Width Interval discretization steps as shown in Fig. 2 Equal Frequency Discretization algorithms
Fig. 1. And these steps are given as an example in Table I below.
Where I is the instance, T is the temperature, C are the classes TABLE II
value, A is the sorted value of T and D which is the discretised EXAMPLE FOR EQUAL FREQUENCY DISCRETIZATION
value of T. I 1 2 3 4 5 6 7 8 9 10
T 85 90 86 96 80 70 65 95 75 91
Input: data is the array having continuous values of the C no no yes yes yes no yes no yes yes
attribute, A = {a0, a1, a2, ….. an-1, an} and k is the number of A 65 70 75 80 85 86 90 91 95 96
parts, where k > 0; Here amin= 65, amax= 96, n = 10, and let k= 3
Output: data having discrete values. So Interval = 10/3=3.3 and around (interval) Then interval = 3;
each bining contained approximately 3 element, the rise is added to
Step 1: Sort the value A in increasing order
the last part. 0 = [65, 80); 1 = [80, 90); 2 = [90, 96].
Step 2: Calculate the interval by equation 1 D 1 2 1 2 1 0 0 2 0 2
Step 3: Binning the data by boundaries formula and
determine the split point for A B. Supervised Discretization Methods
Step 4: The value of attribute in the array must be placed in This discretization method converts numerical data to their
the same boundaries categorical counterparts and use class information while
Fig. 1 Equal Width Interval Discretization algorithms choosing discretization nodes. Entropy based discretization is
an example of this method [1], [2]–[5].
TABLE I
EXAMPLE FOR EQUAL WIDTH DISCRETIZATION 1) Entropy Based Discretization: This method uses division
approach. Entropy (or information content) is calculated based
I 1 2 3 4 5 6 7 8 9 10 on class labels. Intuitively works towards purifying each data
T 85 90 86 96 80 70 65 95 75 91 group by finding the best cut points, so each data in the pure
C no no yes yes yes no yes no yes yes groups will have the same class label as much as possible. It is
A 65 70 75 80 85 86 90 91 95 96 characterized by finding intervals that gives the maximum
Here amin= 65, amax= 96 and let k= 3 information gained [1]–[8]. The formula to calculate Entropy is
So Interval = (96-65)/3=10.3 then Boundary = 65+10.3 = 75.3
as follows;
0 = [65, 75.3); 1 = [75.3, 85.6); 2 = [85.6, 96.2]
𝑛
D 1 2 2 2 1 0 0 2 0 2
𝐸(S, A) = − ∑ 𝑃𝑖 log 2 (𝑃𝑖 )
2) Equal Frequency Interval Discretization: In the equal- 𝑖=1
frequency discretization algorithm the minimum and maximum 𝑣𝑖⁄
values are determined for discretized attribute, and then all 𝑃𝑖 = 𝑣
values are sorted in ascending order, and sorted continuous Entropy means “impurity”. But the impurity here means:
values divided into k intervals in a way that each interval diversity, having too much data with different specifications in
contains approximately n/k data instances with adjacent values, one group. When calculating entropy, we will first calculate
____________________________________________________________________________________________________________
66
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
entropy for estimated class value E(S). Then we will calculate A. Decision Tree Algorithm
entropy for specified cut points E(S, A). Finally, the best cut Decision trees are frequently used data mining approaches
point by calculating information gained by the flowing equation for classification and estimation. Despite the fact that other
is determined [1]–[8]. methods such as ANN can be used for classification, decision
Information Gain = E(S) – E(S, A) trees provide advantage of easy interpretation and
Entropy based discretization pseudo code is shown in Fig. 3 comprehensibility to the decision makers [3]. Classification of
below. All these steps are given as an example in Table 3 below. the data using decision tree technique is a two step process
comprising learning and classification. A previously known
Input: Data is an array having continuous values, A = {a0, education data is analyzed by the classification algorithm in
a1, a2, ….. an-1, an} order to constitute a model during the learning phase. The
Output: Data has discrete values. constituted model is shown as classification rules or a decision
Step 1: sort the value A in increasing order tree. During the classification phase, on the other hand, test data
Step 2: calculate entropy for class data is utilized in order to determine the accuracy of the
Step 3: determine the best split point for A by classification rules or a decision tree. If the accuracy is in an
Calculate Entropy for each split point acceptable range then the rules are used for classification of
Calculate information gain for each split point new data. It should be determined that which domains will be
Step 4: select the split point with highest information gain used in what type of order to constitute the tree. Entropy metric
Step 5: recursively the partition on each split point when is the most widely used measurement for this purpose. The
terminate once information gain falls below a certain results obtained using the domain are uncertain and instable
threshold or once you reach a specified number of bins. proportionally to the entropy measurement in that domain.
Fig. 3 Entropy based Discretization algorithms Thus the minimum entropy or maximum information gain
TABLE III
measurement of the domains are used in the roots of the
EXAMPLE FOR ENTROPY-BASED DISCRETIZATION decision tree. Entropy can be mathematically defined as;
if (𝑝1 , 𝑝2 , … 𝑝𝑛 ) = ∑𝑛𝑖=1 𝑝𝑖 log 2 𝑝𝑖 . (𝑝1 , 𝑝2 , … 𝑝𝑛 ) which
I 1 2 3 4 5 6 7 8 9 10 states the probabilities, then the sum of all these probabilities
T 85 90 86 96 80 70 65 95 75 91 should be exactly 1. Finally the information gain is calculated
C no no yes yes yes no yes no yes yes for each p via the following equation [9]–[10].
A 65 70 75 80 85 86 90 91 95 96 𝑛
No = 4, Yes = 6 probability No Pno = 4/10 = 0.4 probability Yes, 𝑖𝑛𝑓𝑔𝑎𝑖𝑛 (𝐷, 𝑆) = 𝐻(𝐷) − ∑ 𝑃(𝐷𝑖 )𝐻(𝐷𝑖 )
Pyes = 6/10 = 0.6 𝑖=1
Entropy (Play) = E(4,6) = E(0.4, 0.6) = -0.4 * log2(0.4) – 0.6 *
log2(0.6) = 0.970. In a decision tree, all paths from the root node to the leaf
Let k = 2 (80 and 86) then.
node proceed by way of AND [11]. There are multiple modules
For 1. Split <=80 = {65, 70, 75, 80} and > 80 = {85, 86, 90, 91,
95, 96} of the decision tree algorithms [10], ID3 and C4.5 modules are
For 2. Split <=86 = {65, 70, 75, 80, 85, 86} and > 86 = {90, 91, used in this paper.
95, 96}
Calculate entropy and information gain for each split then chose B. Naive Bayes algorithm
the split with the highest information gain. Naive Bayes Classifier is an approach of probability, which
Entropy E = - Σ pi log2 (pi) can be used with a proposition that first seems to be very
Entropy for split 1 = 0.954 and Entropy for split 2 = 0.888. limiting in classification problems [10]. This proposition is the
Information Gain = E(S) – E(S, A) necessity of independency of each defining quality or
Info for split 1 = 0.970 - 0.954 = 0.016 and parameter to be used in pattern identification in respect of
Info for split 2 = 0.970 - 0.888 = 0.082 then
0 = [65, 86); 1 = [86, 96];
statistics. No matter how this proposition is limiting the
application field of Naïve Bayes Classifier, this approach yields
D 0 1 0 1 0 0 0 1 0 1
results comparable to more complex methods such as ANN by
III. CLASSIFICATION ALGORITHMS stretching statistical independency condition.
Naive Bayes Classifier is the simplified state of Bayes
Classification concept is basically the distribution of the data
theorem by means of the independency proposition. Bayes
among predefined various classes on a dataset [1].
theorem is expressed with the following equation:
Classification algorithms learn this distribution type from the
𝑃(𝐵|𝐴) 𝑃(𝐴)
given education cluster and they try to distribute data when 𝑃(𝐴|𝐵) =
unclassified test data are received. The values indicating these 𝑃(𝐵)
classes on the dataset are named as labels and they are used in 𝑃(𝐴|𝐵) ; The probability of incident A when incident B is
order to denote the classes either for education and test. Some realized
classification algorithms process according to categorical 𝑃(𝐵|𝐴) ; The probability of incident B when incident A is
values (Decision Tree, Naive Bayes) while others process with realized
numerical values (ANN) [1]. 𝑃(𝐴) and 𝑃(𝐵); Prior probabilities of incidents A and B.
____________________________________________________________________________________________________________
67
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Here, prior probabilities add subjectivity to the Bayes Glass Identification 68,69 55,61 59,81
theorem. In other words, for instance, P(A) is the information Iris 95,33 89,33 96,00
that is obtained prior to any obtained data about incident A. On Stat log Heart 82,22 58,15 79,63
the other hand P(B|A) is post probability because it gives Australian Credit 85,80 67,54 86,52
information about the realization probability of incident B German Credit 74,50 59,10 72,60
when incident A is realized after data collection [1]–[10]. E.coli 85,71 58,63 74,11
____________________________________________________________________________________________________________
68
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
69
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Multivariate mapping is the visual exploration of point symbols [8]-[15], method in which combining different
multiple attributes using a map or data reduction technique. The type of symbols is used for representing multivariate data [16]
simultaneous display of sometimes multiple features and their and method of separating different attributes from integral
respective multivariate attributes allows for estimation of the symbols [17] can be listed.
degree or spatial pattern of cross-correlation between attributes.
Different from the methods above, in order to represent
Multivariate mapping integrates computational, visual, and
cartographic methods to develop a visual approach for exploring many attributes in the same map, classification method based
and understanding spatiotemporal and multivariate patterns. on clustering methods in data mining can be used as well.
More than one attribute can be visually explored and symbolized With the use of clustering methods, similar aspects of
using numerous statistical classification systems or data different spatial objects can be revealed by considering more
reduction techniques. In this sense, clustering analysis methods than one attribute [18]-[19]. In this sense, spatial analyses that
can be used for multivariate mapping. k-means and k-medoids would make important contributions for risk analysis,
methods which are non-hierarchical clustering analysis methods planning etc. can be done.
were analyzed in this study. The aim of this study is to determine It is quite important to make use of more than one attributes
the success of the spatial analysis of the multivariate maps
of current spatial event and determine similarity of spatial
produced by these methods. For this aim, classes and
multivariate maps created with these methods from traffic events in order to detect future events. For this aim, different
accident data of two different years in Turkey were presented. Inspatial and non-spatial analysis methods are used for
addition usability of such maps in risk management and determining common properties of the spatial events. In this
planning was discussed. context, [20], [21], [22] and [23] used different clustering
methods for determining common properties of different
Keywords— multivariate mapping; data mining; cluster analysis; spatial events. Mapping of spatial events according with
visualization; cartography common properties is also significant for estimating types and
effects of future events. Multivariate maps of occurred spatial
I. INTRODUCTION events are more effective on determining common properties
Multivariate mapping is the graphic display of more than of spatial events and planning investments according to these
one variable or attribute of geographic phenomena. The properties. The aim of this study is to determine the success of
simultaneous display of sometimes multiple features and their the spatial analysis of the multivariate maps produced by k-
respective multivariate attributes allows for estimation of the means and k-medoids clustering analysis methods. For this
degree or spatial pattern of cross-correlation between aim, k-means and k-medoids methods are used for designing
attributes. Multivariate mapping integrates computational, multivariate mapping of traffic accidents in Turkey with data
visual, and cartographic methods to develop a visual approach of two different years, and result maps are compared. Success
for exploring and understanding spatiotemporal and of these methods is determined with comparison of maps
multivariate patterns [1]. designed in different years according to two different methods.
A fundamental issue in multivariate mapping is whether In this study, Cluster Analysis and k-means and k-medoids
individual maps should be shown for each attribute or whether methods in the second section, application in the third section
all attributes should be displayed on the same map ([2], p.327). will be explained in details, results and suggestions will be
Producing separate maps for each attributes would make it shared in last section.
difficult to compare two objects which have various attributes.
Therefore methods in which various attributes are shown in II. CLUSTER ANALYSIS
the same map are preferred more. In this sense, Trivariate Cluster analysis is the process of grouping information in a
Choropleth Map which is created by overlapping of two data set according to specific proximity criteria. Similarity of
colored choropleth map [3]-[6], Multivariate Dot Maps element in the same cluster should be high, similarity between
method in which specific color or symbol is used for each clusters should be low [24]. In the process of classification,
attribute in the map [7], Multivariate Point Symbol methods classes are determined before. In clustering method, classes
which are used when multivariate data can be shown with
____________________________________________________________________________________________________________
70
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
are not determined before. Data are separated different classes object in each cluster is taken as representative. This
according to the similarity of data obtained. representative object, called a medoid, is meant to be the most
Cluster methods are classified in different ways in various centrally located object within the cluster [25]. After k
resources. In general sense, cluster methods can be classified medoids chosen for k-clusters are determined, each remaining
as hierarchical and non-hierarchical methods [2]. object is clustered with the representative object to which it is
Non-hierarchical Methods: In non-hierarchical methods, the most similar [27].
n objects are divided into k clusters according k number Steps of k-medoids algorithm are summarized as below
(k<n) given before. This method divides data in a way [28]:
that there will be at least one object in each cluster and (1) Determination of k-cluster number.
each object will be included at least in one cluster [25]. (2) Choose of k objects as initial medoids
Hierarchical Methods: Hierarchical clustering methods (3) Assign the remaining objects to cluster which has the
group data objects in tree structure. Hierarchical most similar x medoid.
clustering methods are classified as agglomerative or (4) Calculate aim function (sum of distances of all objects
divisive according to hierarchical division being to the closest medoid)
bottom-up or top-down [26]. (5) Arbitrarily choose of y point which is not medoid.
(6) If change of x and y would minimize aim function,
In this study, k-means and k-medoids methods from non- change the place of these two points (x and y)
hierarchical methods will be analyzed. (7) The process is repeated between 3rd and 6th step until
there is no change.
A. K-Means Method
This algorithm which was introduced by Mac Queen for the III. APPLICATION
first time in 1967 is a cyclical algorithm in which clusters are Casualties, injuries and financial damages as a result of
continuously renewed until the most suitable solution is traffic accidents are among the most important problems of
attained. General logic of k-means algorithm is to divide a Turkey. When data of the last 5 years are analyzed, it is seen
data set composed of n data object to k clusters determined that there have been more than 1.000.000 traffic accidents,
depending on preliminary information and experience of 145.000 of them ended up with death and injury, and nearly
researcher. The aim is to provide the intracluster similarity is 1.060.000 of them results with financial damage. 4000 people
high, but the intercluster similarity is low. Similarity of lose their life on average in these accidents and nearly
clusters is calculated with the mean value of objects. 250.000 people are injured. It is quite important to make use
The k-means procedure is summarized as below [25]: of more than one current traffic accidents attributes and
Input: determine similarity of traffic data on city basis in order to
k: the number of clusters, detect measures to be taken for traffic security in Turkey and
D: a data set containing n objects. future investments to be done. For this aim, in this study,
Output: A set of k clusters. clustering analysis will be made in 3 different methods by
Method: using number of motor land vehicles based on city, number of
(1) arbitrarily choose k objects from D as the initial cluster traffic accidents resulting in death and injury, number of
centers; casualties and injuries (4 different values) for the years 2011
(2) (re)assign each object to the cluster to which the object and 2012 prepared by Turkish Statistical Institute (TUIK) and
is the most similar, multivariate maps will be produced according to analysis
based on the distance between the object and the cluster results. Maps designed for both years with 3 different methods
mean; were compared, it will be evaluated which method is suitable
(3) update the cluster means, i.e., calculate the mean value for the success of multivariate mapping and clustering.
of the objects for each cluster; In the application of clustering analysis methods,
(4) repeat until no change; (Fig.1) RapidMiner software developed in Dortmund Technology
University Artificial Mind Unit by Ralf Klinkenberg, Ingo
Mierswa and Simon Fischer was used. Multivariate maps
were designed by ArcGIS software developed by ESRI group.
____________________________________________________________________________________________________________
71
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE II
CENTROID VALUES OF CLUSTERS GENERATED WITH K-MEANS METHOD FOR
2012 DATA
Fig. 2 Multivariate maps designed with k-means method for the years 2011
(above) and 2012 (below)
____________________________________________________________________________________________________________
72
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III
CENTROID VALUES OF CLUSTERS GENERATED WITH K-MEDOIDS METHOD FOR
2011 DATA
TABLE IV
CENTROID VALUES OF CLUSTERS GENERATED WITH K-MEDOIDS METHOD FOR
2012 DATA
Fig. 3 Multivariate maps designed with K-medoids for the years 2011 (above)
and 2012 (below)
IV. CONCLUSIONS
In the scope of multivariate mapping, more than one
attribute can be displayed in separate maps or in the same map.
Designing separate maps for each attribute would make it
difficult to compare two objects which have various attributes.
Therefore methods in which various attributes are shown in
the same map are preferred more. One of the methods in
which various attributes are displayed in the same map is to
generate thematic map classes by determining the effect of
different attributes with clustering analysis. In this sense, in
this study, considering traffic accidents in 2011 and 2012 in
Turkey, number of vehicles in traffic in these years, number of
traffic accidents resulting in death and injuries, number of
casualties and injuries parameters, multivariate maps were
designed with 2 different cluster analysis method.
K-means and k-medoids non-hierarchical clustering
algorithms divide n objects into k clusters according to k input
parameter. They form the same cluster if objects resemble
each other but not with the objects in other clusters. The
greatest problem in applying these algorithms is determination
of k cluster number. This can be determined with some of the
practices experiences. When k=5 cluster number is given for
data sets used in the study, it is observed that better clustering
____________________________________________________________________________________________________________
73
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
results were obtained. Although clustering success of both [12] C.G.,Healey and J.T. Enns, “Large Datasets at A Glance: Combining
Textures And Colors In Scientific Visualization,” IEEE Transactions
algorithms are similar, when centroid tables of clusters formed
on Visualization and Computer Graphics, 1999, 5(2), 145-167.
with both methods (Table 1-4) are observed, it was detected [13] J.R. Miller, “Attribute Blocks: Visualizing Multiple Continuously
that clusters are separated better in k-medoids algorithm. Defined Attributes,” IEEE Computer Graphics and Applications, 2007,
Since the aim is to provide high intracluster similarity and low 27(3), 57-69.
[14] X. Zhang and M. Pazner, “The Icon Imagemap Technique for
similarity between different clusters, it can be said that k-
Multivariate Geospatial Data Visualization: Approach and Software
medoids method gives better results for these data. When the System,” Cartography and Geographic Information Science, 2004,
maps designed using k-medoids method are examined, it is 31(1), 29-41.
observed that 61 of 81 cities are located in the same cluster for [15] E. S. Nelson and P. P. Gilmartin, “An evaluation of multivariate,
quantitative point symbols for maps,” In Cartographic Design:
two years. On the other hand, using k-means method 58 of 81
Theoretical and Practical Perspectives edited by C. H. Wood and C. P.
cities are in the same cluster. These results exhibit that Keller, 1996, 191–203.
foresights using these methods with 2011 data are highly [16] D. DiBiase, “Designing Animated Maps for A Multimedia
consistent with 2012 data. Encyclopedia,” Cartographic Perspectives, 1994, 19, 3-7.
[17] E.S. Nelson, “Designing Effective Bivariate Symbols: The Influence of
With this study it was shown that by using clustering
Perceptual Grouping Processes,” Cartography and Geographic
methods, similar aspects of different spatial objects can be Information Science, 2000, 27(4), 261-78.
presented by considering more than one attributes. It is [18] A. T. Murray and T. H. Grubesic, “Exploring spatial patterns of crime
thought that by using multivariate maps designed with using non-hierarchical cluster analysis,” In Crime modeling and
mapping using geospatial technologies, 2013, 105-124, Springer
clustering methods, spatial analyses which have important
Netherlands.
contributions for practices such as risk management, planning [19] T. H. Grubesic, R. Wei and A.T. Murray, “Spatial Clustering Overview
etc. can be made. and Comparison: Accuracy, Sensitivity, and Computational Expense,”
Annals of the Association of American Geographers, 2014, 104(6),
REFERENCES 1134-1156.
[20] J. Weng, W. Qiao, X. Qu and X. Yan (2015) “Cluster-Based
[1] A. Buckley, “Multivariate mapping,” In Encyclopedia of Geographic
Lognormal Distribution Model For Accident Duration,”
Information Science edited by Kemp K., 2008, 300-303.
Transportmetrica A: Transport Science, 2015,11(4), 345-363.
[2] T.A. Slocum, R.B. McMaster, F.C. Kessler and H.H. Howard,
[21] F. Guo and Y. Fang, “Individual Driver Risk Analysis Using
Thematic Cartography and Geovisualization, Pearson Education Inc.
Naturalistic Driving Data”, 3rd International Conference on Road
Third Edition, USA, 2009.
Safety and Simulation, September 14-16, 2011, Indianapolis, USA.
[3] C.A. Brewer, “Color Use Guidelines for Mapping and Visualization”,
[22] S. Feng, Z. Li, Y. Ci and G. Zhang, “Risk Factors Affecting Fatal Bus
In Visualization in Modern Cartography edited by MacEachren A.M.
Accident Severity: Their Impact on different Types of Bus Drivers”,
and Taylor D.R.F., 1994, 123-147.
Accident Analysis and Prevention, 2016, 86, 29–39.
[4] G. Metternicht and J. Stott, “Trivariate Spectral Encoding: A Prototype
[23] L. M. Martinussen, M. Møller, C.G. Prato, “Assessing the Relationship
System for Automated Selection of Colours for Soil Maps Based on
Between the Driver Behavior Questionnaire and the Driver Skill
Soil Textural Composition,” in Proceedings of the 21st International
Inventory: Revealing Sub-groups of drivers”, Transportation Research
Cartographic Conference, Durban, CD, 2003.
Part F, 2014, 26, 82–91.
[5] J. R. Byron, “Spectral Encoding of Soil Texture: A New Visualization
[24] E.Ş. Dinçer, “Veri Madenciliğinde K-means Algoritması ve Tıp
Method,” in GIS/LIS Proceedings, Phoenix, Airz., 1994, 125-132.
Alanında Uygulanması,” M. Eng. Thesis, Kocaeli University Institute
[6] V. Interrante, “Harnessing Natural Textures for Multivariate
of Science, Kocaeli, 2006 (in Turkish).
Visualization,” IEEE Computer Graphics and Applications, 2000,
[25] J. Han, J.G. Lee and M. Kamber, “An Overview of Clustering Methods
20(6), 6-11.
in Geographic Data Analysis,” In Geographic Data Mining and
[7] G. F. Jenks, “Pointillism as a Cartographic Technique,” The
Knowledge Discovery edited by Miller H.J. and Han H., Taylor &
Professional Geographer, 1953, 5, 4–6.
Francis Group, LLC, 2009.
[8] D.J. Cox, “The Art of Scientific Visualization,” Academic Computing,
[26] J. Han and M. Kamber, Data Mining: Concepts and Techniques, San
1990, 4, 20-22, 32-34, 36-38.
Francisco, 2006
[9] R. Ellson, “Visualization at Work,” Academic Computing, 1990, 4(6),
[27] Silahtaroğlu, G. 2013. Veri Madenciliği (Kavram ve Algoritmaları).
26-28,54-56.
Papatya Publishing, İstanbul, (in Turkish).
[10] D. Dorling, “The visualization of local urban change across Britain,”
[28] Akın, Y.K. 2008. “Veri Madenciliğinde Kümeleme Algoritmaları ve
Environment and Planning B: Planning and Design, 1995, 22, 269 -
Kümeleme Analizi.” PhD diss., Marmara University Institute of Social
290.
Science, İSTANBUL, (in Turkish).
[11] G.Grinstein, , J.C.J. Sieg, S. Smith and M.G. Williams, “Visualization
for Knowledge Discovery,” International Journal of Intelligent
Systems, 1992, 7, 637–648.
____________________________________________________________________________________________________________
74
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In our age, technological developments are trade, industry and many other fields in the international sense.
accompanied by certain problems associated with them. Security For this reason German language is used in a quite widespread
takes the first place amongst such kind of problems. In manner. However, common usage of this language to such an
particular, such biometric systems as authentication constitute extent, brings security problems for biometric data in this field.
the significant fraction of the security matters. This is because
Accordingly this calls for the requirement of a secure, fast
sound recordings having connection with the various crimes are
required to be analyzed for forensic purposes. Authentication automatic voice and speaker recognition.
systems necessitate transmission, design and classification of German language comprises of roots (words) and suffixes
biometric data in a secure manner. In this study, analysis of & prefixes and included in the inflected languages if we
German language employed in the economy, industry and trade consider properties of the German language. German is
in a wide spread manner, has been performed. In the same vein, written using the Latin alphabet and there are 29 letters in its
the aim was to actualize automatic voice and speech recognition alphabet. An article appears before each noun in German.
system using Mel Frequency Cepstral Coefficients (MFCC), Words are pronounced as they are written. In addition, it is
MelFrequency Discrete Wavelet Coefficients (MFDWC) and distinguished from other languages with various developed
Linear. Prediction Cepstral Coefficient (LPCC) taking German
sound shifts and intonation.
sound forms and properties into consideration. Approximately
2658 German voice samples of words and clauses with differing Various studies have been carried out in order for voice and
lengths have been collected from 50 males and 50 females. speaker recognition. Jie-Fu et al. have collected voice samples
Features of these voice samples have been obtained using wavelet in Chinese from 7 males and 5 females whose ages were
transform. Feature vectors of the voice samples obtained have ranging between 25 and 45 [8]. Attempts have been made to
been trained with such methods as Boltzmann Machines and identify the owner of the voice by trying to analyze these
Deep Belief Networks. In the test phase, owner of a given voice voice samples by means of their tones, vowels, consonants
sample has been identified taking the trained voice samples into and syllables. Voice samples have been separated into four
consideration. Results and performances of the algorithms frequency groups, and each frequency band has been analyzed.
employed in the study for classification have been also
However, this study has not been tested for very big data. In
demonstrated in a comparative manner.
addition, intended success was not exactly achieved since it
was performed taking its similarities with the English
Keywords— Boltzmann Machines, Deep Belief Networks. language into consideration.
Tokuda et al. have developed English speech synthesis
system using Hidden Markov Model [4]. This system has been
I. INTRODUCTION developed for speaker recognition and specifies the structure
In our day, security problems have been unearthed along by changing the voice feature. However characteristic feature
with the developments in the technology. Certain studies have of the synthesized voice in the study, is pretty low.
been accomplished especially in order to prevent information Reynolds et al. have implemented SuperSID project in
belonging to some people from being transferred to other order to enhance performance of speaker recognition systems
people in commercial transactions. Some of such studies are [1]. Purpose of this project is to develop speaker recognition
hand script recognition, signature recognition, face systems and employ the most suitable features in order to
recognition, iris recognition and voice recognition [9]. increase its accuracy. However this study failed to completely
German language belongs to the Germanic branch of Indo- achieve the acoustic characteristics of the voice and removal
European language family. Approximately 120 million people of the noise.
are speaking German language in the world. In addition, Reynolds et al. have attempted to substantiate speaker
Germany has an important standing in respect of economy, identification and verification using Gaussian Mixture Model
____________________________________________________________________________________________________________
75
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(GMM) method [2]. Attempts have been made to determine scaled between actual frequency of voice signal and estimated
speaker verification in the system according to the probability voice frequency. For this reason total energy of every frame is
distribution. 11 different hypothesis have been developed for calculated. Classification success in speaker identification has
this probability distribution. Data used in the study has been been calculated on an individual basis for MFDWC-3,
extracted from telephone conversations. MFDWC-5 and MFDWC-9 vectors by training the feature
Speech has an important place in communication. Voice vectors obtained from voice signals by means of Boltzmann
recognition study has been carried out for this reason. In this Machines and Deep Belief Networks.
study, a simulation also has been performed in order to solve
the voice recognition problem related to the security risk.
However, certain difficulties have got in the way while Pre-
Filtering Framing
creating voice database. There was such difficulty ranked first Emphasis
among the others that words were vocalized at different
speeds and in different pronunciation by different persons. In
addition to that, such reasons as the noise occurred in the
environment and voice while recording the voice data, toning
effect and syllable stress make voice recognition process FFT
difficult [10].
Feature extraction and classification techniques used, were
given in the section 2 of the study performed, and
experimental study and results were given in the section 3. Feature Mel-Frequency
DWT Filter Bank
Vector
II. FEATURE EXTRACTION METHODS
German language is widely used in economy, industry and
trade. Therefore, examinations have been made on German
language in this study. The study has been realized on a Fig. 1 Feature extraction steps of MFDWC
unique data base, which have been formed from the German B. Mel Frequency Cepstral Coefficients (MFCC)
sound samples, taken from men and women. These sound
samples are trained by getting dispersed to various feature MFCC is a feature extraction method, that is used in sound
vectors with MFDWC, MFCC and LPCC. In the second stage, processing. It is used to extract important information and
the feature vectors of the recorded sound signals are trained features by dividing the sound data to its subsets. The steps of
with classification algorithms, such as Boltzmann Machines feature extraction technique of MFCC is indicated in Figure
and Deep Belief Networks. The gender of the speaker is 2[11].
decided by looking at sound signals at the test data and Pre-Emphasis
training data after the system is trained. Furthermore, the
classification success in recognizing the gender of speaker has
been calculated separately for 1, 3 and 5 and 9 feature
vectors and the success of the methods have been presented Framing
comparatively by training the feature vectors, obtained from
speaking signals with Boltzmann Machines and Deep Belief
Networks.
Windowing
A. Mel-frequency Discrete Wavelet Cofficients (MFDWC)
The study in question has been performed, based on a
unique database comprising German voice samples collected FFT Spektrum
from men and women. These voice samples were separated
into various feature vectors with MFDWC, and trained.
MFDWC is a feature extraction method employed in the
speech processing. It is used to extract significant information Mel Spektrum
and features by dividing voice data into subsets. Feature
extraction steps of MFDWC technique is shown in the Figure
1 [7]. Mel Cepstrum
Sample speech signal is shown between the 40-40000 Hz
range in the MFDWC feature extraction method. Speech
signal is divided into frames after the pre-processing step. Feature Extraction
Hamming window has been used in this study in order to
smoothen the transition of speech samples between the frames.
One Mel shows the frequency of voice tone. Mel-scale is
Fig. 2 Feature extraction steps of MFCC
____________________________________________________________________________________________________________
76
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
77
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
78
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I
ATTRIBUTES OF USED DATABASES
Success rates of speech samples obtained utilizing LPCC
Number of speaker
different feature vectors are given for Boltzmann Machines
Age Range and Deep Belief Networks in the Table 4. Success in the
Male Female speaker identification increases as the number of words used
increases in all techniques employed. Deep Belief Network
18-25 range speakers 15 19 gave more successful results when compared to Boltzmann
26-40 range speakers 23 26 Machines.
41 and more speakers 12 15
TABLE IV
SUCCESS IN CLASSIFICATION FOR LPCC
Feature vectors of voice components with different
quantities; have been extracted by means of MFDWC, MFCC
3 5 9
and LPCC feature extraction method. Voice samples have Feature
been tested by training them, using available feature vectors vectors/ Male Female Male Female Male Female
by means of Boltzmann Machines and Deep Belief Networks Methods
methods. used
Success rates of speech samples obtained utilizing MFDWC Boltzmann 93.27 94.84 95.61 94.38 98.83 97.62
different feature vectors are given for Boltzmann Machines Machines
and Deep Belief Networks in the Table 2. Success in the Deep 96.71 93.62 97.65 96.05 99.62 99.17
speaker identification increases as the number of words used Belief
increases in all techniques employed. Deep Belief Network Networks
gave more successful results when compared to Boltzmann
Machines.
Success rates of speech samples obtained employing
TABLE II MFDWC, MFCC and LPCC feature vector. Success rates of
SUCCESS IN CLASSIFICATION FOR MFDWC speech samples obtained employing 9 feature vector, for all
feature extraction techniques.
3 5 9 A unique and genuine German language database has been
Feature employed in this study. Names, family names, ages, speeches
vectors/ Male Female Male Female Male Female and genders of the persons were added to this database.
Methods
Feature vectors of voice components with different quantities;
used
Boltzmann 87.62 82.15 74.62 70.82 79.62 76.21
have been extracted by means of MFDWC, MFCC and LPCC
Machines feature extraction method. Voice samples have been tested by
Deep 85.03 83.72 56.75 84.62 87.56 81.62 training them, using available feature vectors by means of
Belief Deep Belief Networks and Boltzmann Machines. In the
Networks testing phase it was determined successfull classification
techniques as male or female by available testing example. It
Success rates of speech samples obtained utilizing MFCC has also presented and compared by calculating the success of
different feature vectors are given for Boltzmann Machines any method used.
and Deep Belief Networks in the Table 3. Success in the
speaker identification increases as the number of words used V. CONCLUSION
increases in all techniques employed. Deep Belief Network
gave more successful results when compared to Boltzmann Voice recognition plays an important role in our day due to
Machines. security and many other reasons. Person and speaker
identification systems have been developed, being based on an
TABLE III unique database obtained by utilizing German language in this
SUCCESS IN CLASSIFICATION FOR MFCC study. Classification success of the methods employed in the
study have been calculated separately for men and women,
3 5 9 and results are demonstrated in a comparative manner. Deep
Feature
vectors/ Male Female Male Female Male Female
Belief Networks provided more successful results compared
Methods to the Boltzmann Machines when the results are taken into
used consideration. Speaker recognition system is more successful
Boltzmann 90.87 91.72 94.67 93.21 97.85 96.94 for men compared to the women. 9 feature extraction is more
Machines successful compared to the results obtained utilizing for all
Deep 94.67 90.86 95.65 95.39 99.91 98.78 other feature extraction techniques.
Belief
Networks
____________________________________________________________________________________________________________
79
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
80
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— There are many microcontroller training kits are companies were emerged to market the educational kits for
available on the market. Most of them are equipped with DIP- institutions that have short budgets. Since then, several types
switched circuit connections that are configured mostly by of experimental kits have been marketed by different
students during the experiments. Intensive use of DIP-switch producers with diverse elements and components.
sockets, cables, and materials on the experimental kits leads to
Almost all experimental kits have sockets and cable pairs
circuit breaks and faulty connections easily. Therefore, many
institutions face undesired hurdles in maintenance and repairing for the connection of the desired experiment. Since each
procedures even though they have skilled technicians or experiment has different connection setup, the users must
instructors. configure all cables and connectors on the board before the
In this paper, we have designed and implemented a software- test run. This procedure should be done with great care and in
configured microcontroller training set to overcome adverse a proper manner. Despite diligent operations, the user can
conditions originated by conventional microcontroller kits such harm the sockets and cables during the insertion or remove
as time-consuming maintenance procedures and circuit faults procedures because of the mechanical forces. In addition, the
caused by improper configurations. In doing so, an analog switch user can connect the ports in the wrong way inadvertently.
matrix board has been designed to perform connections between
The potential problems mentioned can increase the time and
the microcontroller port pins and external peripherals. The
system eliminates the possibility of the user errors or electrical maintenance cost dramatically along with reduced level of
faults caused by improper wiring along with mechanical availability of the kits.
damages caused by forceful interventions. Hence, the students In the literature, there are many approaches to ease the
can concentrate on the technical aspects of the experiment other laboratory burdens. For example, an educational kit has been
than mechanical inconveniences. designed to configure the program of the microcontroller and
The software designed, which manages an analog matrix switch port connections in a straightforward fashion. The educational
board, has many scenarios and configuration files for each kit also incorporates a software interface to ease the procedure
experiment. The students decide the type of the experiment and of the experiments [1].
he/she configures both the microcontroller and the CPLD using
Another paper presents a microcontroller training set which
software interfaces provided. The CPLD configuration file is
fixed for each scenario but the microcontroller hex file must be is also supported by training sets and video courses along with
developed by the students. The CPLD and the 8051 development a virtual laboratory [2].
board are connected to the computer via a USB 2.0 port. In a master thesis, which is the origin of this study, an 8051
based microcontroller training kit and its peripheral modules
are connected each other with a matrix switch configured by a
Keywords— Microcontroller, Training Set, Experiments, CPLD,
CPLD. In that study, there is no software interface used to
Matrix Switches
configure both CPLD and microcontroller devices
I. INTRODUCTION respectively. However, the user must configure both the
microcontroller and CPLD boards manually using software
Since the time microprocessors and microcontrollers
tools provided by manufacturers [3].
courses first appeared at the universities and colleges, there
An 8-bit Motorola 68HC11 microcontroller training kit
has been always a strong demand for experimental sets to
employed with RS-232 serial communication has been
comprehend the backgrounds of the subject. Emerging
designed. The connections for each experiment are performed
technologies like flash programming, high-level language
by on-board DIP-switches and the hex code files are
compilers, and simulation tools increase the number of
downloaded into the chip through RS-232 protocol [4].
candidates who are willing to develop embedded systems. At
Virtual platforms used in embedded systems training are
the same time, many companies promote their chips and
examined and it is claimed that students prefer virtual
education or training kits to draw the attention of the
environments rather than hardware based platforms since they
engineers from the world. Initially, the price of the kits was
are considered more efficient compared to conventional
too high for educational institutions. So, many local
approaches [5].
____________________________________________________________________________________________________________
81
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
An MC68C11/12 based experiment kit has been designed controlled by 2-bit selection signals (IN1, IN2) provided by
for engineering laboratories. The system includes only CPLD outputs as shown in Fig. 2. In fact, the microcontroller
hardware based components and modules [6]. has four port sockets deployed with 10-pin IDC headers,
An MC68000 microprocessor based standalone educational however, in each port header two extra pins are used for
board has been designed. The board incorporates keypad, power pins i.e. GND and VDD. So, there is no need to
LEDs, LCD, DIP switches, 7-segment display, DC motor and redistribute those power pins and they are shared with all ports
a couple of I/O peripherals. All hardware connections are by default.
implemented by DIP-switches and cables [7].
A microcontroller (ATmega128) processor based The software interface developed incorporates following
educational kit, which is targeted for bachelor engineering features:
students, has been examined. Again, entire hardware
connections are implemented by DIP-switches and cables [8]. All configuration procedures have visual icons
In another study, a PIC 16F84 based training program and Experiment selection by the user
an accompanying experimental set have been designed. HEX file selection and experiment setup facilities
Programme includes some experiments related to basic The circuit schematic of the experiment is loaded on
operations for target microcontroller [9]. the screen when an experiment is selected.
‘Status’ pane shows the procedures step by step to
In this paper, we propose a new design and test inform the user
environment used in microcontroller training to eliminate the ‘Experiment summary’ pane informs the user on the
drawbacks originated from conventional microcontroller details of the experiment such as port routing
experimental kits. The method includes a matrix switch circuit schemes and the materials used
and a user interface to configure the experimental kit within a
couple of seconds without mechanical or physical
interventions.
____________________________________________________________________________________________________________
82
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSIONS
In this paper, we have designed and implemented a user-
friendly, low cost and low maintainability laboratory kit used
in embedded systems classes. The system includes hardware
and software modules and enables students to focus on
programming of the embedded system and its modules other
than the hardware connection burdens. Since there is no
student intervention on the experiment kits, the risk of the
mechanical failures or breaks. The system also provides time
advantages, thereby promising more experimental
opportunities for the students. The system has been tested with
a limited number of modules and it would be better if the
number of modules is increased.
____________________________________________________________________________________________________________
83
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, research was carried out in order to Calculating a new pixel value of each pixel which is known
compare the effects of basic imaging applications such as image as interpolation during image rotating application is an
rotation and image filtering on a noisy image and determine important operation. For this application, bilinear, bicubic,
whether image rotating application can be used instead of nearest interpolation methods were developed in the Matlab
image filtering application. In the study, it was aimed to environment and put into practice [8].
compare bilinear, bicubic and nearest methods, among The comparison of image rotating operation and the
commonly used interpolation methods of image rotating, with mentioned interpolation methods noise elimination based on
average, wiener and median filters, which are the linear filters
image rotating process idea and implementation with the
among image filtering methods, in terms of image quality factors.
PSNR (Peak to Signal Noise Ratio), MSE (Mean Square Error), mentioned filtering methods was a fundamental aim of this
SSIM (Structural Similarity Index) were determined as quality study [8].
factors to be used in the study, and the processed image was a The measurability and comparability of an application
baboon image, an 8-bit gray-scale image of 512*512 dimension performed are important in terms of science. Therefore,
that is the most common image used in image processing. image quality factors are important in image processing
applications in order to obtain the measurements of the
Keywords— Image rotating, Image filtering, PSNR, MSE,SSIM original image and the image obtained after the performed
operation [9].
I. INTRODUCTION II. METHOD
The noise which is one of the most common encountered A. Image And Noise Filtering
problems in image processing operations consists of noises The basic logic of the median filter algorithm that is one of
such as photon shot noise, dark current noise, read noise, the filters used in this study is realized by a mask and
quantization noise [1]. These noises can occur within the making ascending and descending order of the pixel values
digital systems due to the reasons such as data transmission within the neighborhood of this mask and appointing the
and data storage. The random noises such as Gaussian Noise mean value to the center pixel value of the mask determined
and Salt&Paper Noise which are the most commonly used [4].
images to model these noises and create algorithms based on The second filter used in this study is called average filter
this can be created in software environment[1]. and is one of the most basic linear and spatial filters. The
The filtering technique which is one of the most frequently average filter is found by taking the arithmetic average of the
used methods to eliminate noises on the image is divided into sum of pixel values in the whole neighborhood of a mask with
two: linear, non-linear and spatial-frequency domain filters a certain neighborhood [5].
[2,3].Linear filters are generally defined as low-pass filters, Another filter used in this study was determined as the
create blurring on the image and smooth the image while wiener filter, and the use of the mentioned filter structure was
nonlinear filters make the noise filtering process by preserving realized in the Matlab environment and is currently used in
the edge information defined as a high frequency [3]. academic studies [6].
Average filter and wiener filter from linear filters and median B. Image Rotating
filter from non-linear filters were used in this study [4, 5, 6]. The methods used in image rotating operation were realized
Image rotating operation is one of the most frequently used by using the available Matlab functions. The relevant methods
applications in program-based and real-time processing in are bilinear, bicubic, nearest interpolation methods. The
image processing applications [7]. comparison of image rotating operation with image filtering
____________________________________________________________________________________________________________
84
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
85
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Again, the change in the noise density which was added to the
image in this study and the PSNR values of the relevant image
following the application of the median filter were obtained.
The results are shown in Fig.7.
Fig.7 Baboon Image PSNR Value For Each Noise Density Value
____________________________________________________________________________________________________________
86
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] P.Zhijun, T.Qingqiao, W.Lina, Z.Jun,”A Median Filter Method For
Image Noise Variance Estimation”, Second International Conference on
Information Technology and Computer Science, 2010
[2] Q.Xie and H.Chen ,”Image Denoising Based on Adaptive and
Multiframe Averaging Filtering”, International Conference on Artificial
Intelligence and Computational Intelligence,2009
[3] D.Fu-guo,”A Novel Fast Algorithm For Image Filtering Based on
Improved Sign Test Method”,978-1-4244-7330-4/10/2010 IEEE
[4] Y.Zhu, C.Huang , L. Zhai,”A Median Image Filtering Algorithm Based
on Statistical Histogram”, Fifth Conference on Measuring Technology and
Mechatronics Automation, 2013
Fig.12 Baboon Image 10 Degree Rotation and SSIM Value
[5] G.Mrityunjay , M.Srabani,”Cache Oblivious Algorithm of Average
Filtering in Image Processing”, IEEE/OSA/IAPR International Conference on
Informatics, Electronics&Vision,2012
[6] D.Manoj, K.Manoj, ”CT Image Noise Reduction Base on Adaptive
Wiener Filtering with Wavelet Packet Thresholding”, International
Conference on Parallel, Distributed and Grid Computing, 2014
[7] A.Amir Hossein,N.Md Jan , K.Seyed Mostafa Mousavi ,”Double Line
Image Rotation”, IEEE Transactions On Image Processing,Vol.24, No.
11,2015
[8] C.Huifang,W.Yi ,”A New Image Rotation Approach Using Radial Basis
Functions”, 8th International Congress on Image and Signal Processing,2015
[9] H.Alain, Djemel Ziou,”Image Quality Metrics PSNR vs.SSIM”,
International Conference on Pattern Recognition,2010
[10] S. M. Elshoura and D. B. Megherbi,”A New Estimation and Rotation
Algorithm”, International Symposium on Computer, Communication, Control
and Automation, 2010
____________________________________________________________________________________________________________
87
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Edge detection has important applications area in derivative method caller has provided a transition to zero [2].
image processing field. Today, it is a fact that the image processing Canny, the implementation of the operator derived from the
used in many fields. Therefore, the applicability of edge detection Gaussian mask derivative based proposed a method [3].
process in the field is also has great importance. In this study, Roberts is the first method used in image processing, Prewitt
mainly used edge detection algorithms in the literature; İe. Sobel,
and Sobel operators are an example to the most appropriate
Prewitt and Canny algorithms is provided using the verification
and inspection on FPGA (Field Programmable Gate Arrays). operators [4]. Because image processing and edge detection
Program files required for FPGA is prepared by Xilinx System processing speed for applications that require processing
Generator DSP blocks, which can work integrated with typically computer-based applications, DSP-based methods
Matlab/Simulink. For this study; gray format images, which is and FPGA's are used.
stored on the computer has been sent to FPGA with USB We can sort out some work done using image processing
configuration port interface on FPGA. Edge detection process is FPGA said. Chris and his friends for the detection of tumors in
realized by moving subject images from the computer with the the brain images taken by MRI device, brands have benefited
same connection to FPGA and then, Sobel, Prewitt and Canny from Xilinx FPGA development board. Implemented on FPGA
algorihms are applied to the images on FPGA respectively. Edge
determined the tumor site with different filters. Create a system
detection process for the same images are performed by Simulink
and FPGA bord at the same time and then, edge detected images in which they work in a way that is integrated with Matlab,
obtained from these two environment are compared and also it has compared to other systems available in the literature, 50% less
been observed on the FPGA resource usage. use of resources has been realized [5]. Rodriguez and
colleagues while running real-time median filter design on
FPGA, have created a system that enables visual identification
Keywords— FPGA (Field Programmable Gate Arrays), Sobel,
of errors in production centers [6]. Nelson, a team using
Prewitt, Canny, Matlab/Simulink and Xilinx System Generator,
DSP (Digital Signal Processing) different FPGA hardware produced by the master's thesis on
Xilinx and Altera company that performs morphological
operations and convolution processing, receives the results
I. INTRODUCTION were compared with the results produced by using MATLAB
[7]. Nana and her friends while improving the resolution of an
Today, with the development of image processing
image taken using DSP with FPGA hardware have gained a
technology it is also used quite a lot. Most of our lives
clearer picture. In this process; They used the Xilinx Spartan-3
intertwined with a technological structure is wrapped in an
FPGA hardware configuration of the image and they have
image processing device is configured on. The algorithms used
developed the resolution with DSP [8].
in this field of technology with the speed proportional manner
This article made a thorough research on the FPGA
are developing very rapidly. Work is done in many areas, I.e.;
programmable logic devices and technologies and the basic
Security systems, traffic warning systems, microscopy and
operators used in edge detection Sobel, Prewitt and Canny
medical imaging systems, making use of image processing
operator of the implementation on FPGA and is provided to be
techniques such as remote sensing systems.
examined. In this study, Matlab / Simulink and Xilinx System
One of the most important areas in the image processing is
Generator can run Xlinx integrated DSP blocks in Simulink and
an edge detection process. Edge detection is one of the most
Simulink blocks with the tool is used. Due to prepared system
important task of finding the image. Edges is the most
256*256 gray formatted images on computer sent to Simulink
important information in the image. Human visual system is
and FPGA at the same time and Sobel, Prewitt and Canny
based on the direct detection of edges [1]. To distinguish
operators were applied to the images. In prepared system, data
objects in the images, the display of details which can not be
transmission between computer and FPGA is provided with
noticed with the naked eye, in situations such as edge detection
USB configuration port. FPGA's need to work * .bit extension
algorithms to compare the two different patterns are used quite
program file has been created with the Xilinx System Generator.
extensively. Haralick, found that the gradient value using the
In practice, Spartan3E XC3S1600 belonging to the company
derivative value interpolation equation is found through applied
Xilinx FPGA device is used.
on pixels in the direction of the gradient on the second
____________________________________________________________________________________________________________
88
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The following sections of this article consists of more than A. The First Derivative(Gradient) Based Methods
four sections. Scope of issues examined in this section are as Gradient based single variable is expressed as the first
follows: Section 2 in this article; edge detection, edge detection derivative of a function. This operation is a function which
methods and general information about the basic edge detection depends on two variables calculated by applying the process
operators are given. At Section 3; information about Xilinx according to the 1st derivative of each variable. As an image is
System Generator, which used for programming of FPGA expressed as a function of two changes, the necessary
board and Simulink DSP blocks are given. Section 4 in this information about the direction of the gradient edge angle as
article; implementation of Sobel, Prewitt and Canny edge can be seen by looking at the size gradient to be applied to the
detection operator on FPGA and Simulink side is explained in edges in an image display function will give us. f(x,y) as an
detail. Results and conclusion parts are shown in sections 5 and equation for an image display function, gradient magnitude and
6 respectively. gradient angle is calculated by equations below.
II. EDGE DETECTION AND OPERATORS ( ∆ , ) ( , )
lim
Visual system in humans is first trend detection edges in the ∆ → ∆
[ ( , )] = = = ( , ∆ ) ( , ) (1)
image of the object located on any image. This is because the lim
stationary location typically does not contain information. In an ∆ → ∆
−1 0 1 1 2 1
= −2 0 2 , = 0 0 0 (4)
−1 0 1 −1 −2 −1
____________________________________________________________________________________________________________
89
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
90
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
A. Image Pre-Processing
During this process the data entry format of the “Gateway In”
the blog entry block for FPGA card in gray format image on a
computer and converted into serial digital data format. Fig. 6 Hardware and software realisation of proposed design
Simulink blocks used in image pre-processing steps are given
in Fig. 4. D. Edge Detection with XSG
In order to implementation of edge detection on FPGA, XSG
DSP blocks on Simulink is used. For implementation of Sobel
operator, gradient calculation and thresholding operation is
done XSG DSP blocks, which is shown in Fig. 7 and Fig. 8
respectively. Gradient calculation blocks give the convolution
Fig. 4 Image pre-processing blocks of Sobel mask coefficient with original image pixels and
thresholding blocks give the binary edge detected image with
respect to a threshold value.
____________________________________________________________________________________________________________
91
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Implementation of Prewitt operator is done same as Sobel Finally, double thresholding is applied to image pixels and
operator. Because only the convolution coefficient of Prewitt binary edge detected image is obtained. Double thresholding
mask is different from Sobel operator. XSG blocks are shown in Fig. 11.
(a)
(b) (c)
____________________________________________________________________________________________________________
92
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
∑ , [ ( , ) ( , )]
= (10)
∗
= 10 log (11)
(f) (g)
For MSE value; I1 is pixel value of original image, I2 is pixel
Fig. 12 (a) Original image (b) Sobel Simulink result (c) Sobel FPGA value of edge detected image, M and N is horizontal and
result (d) Prewitt Simulink result (e) Prewitt FPGA result (f) Canny Simulink
result (g) Canny FPGA result vertical pixel size of original image. R is taken 255 in this work,
because original image data format is 8-bit unsigned integer.
FPGA resource usage of Sobel-Prewitt and Canny Table III shows quantitative results of MSE and PSNR values
algorithms on Spartan3E XC3S1600E is given Table I and for proposed works.
Table II respectively. Sobel and Prewitt resource usage is TABLE IIIII
given in one table, because they use the same resource usage QUANTITATIVE RESULTS OF MSE AND PSNR
on FPGA.
Algorithms / Platform MSE PSNR (dB)
TABLE I
FPGA RESOURCE USAGE FOR SOBEL OR PREWITT Simulink 29327.8 3.458
Sobel
FPGA 30149.5 3.338
Resource Utilization Summary (Sobel or Prewitt)
Simulink 31137.2 3.198
Resources Used Available Utilization Prewitt
FPGA 31345.8 3.169
Number of Slice Flip Flops 1,404 29,504 4%
Simulink 31447.1 3.155
Number of 4 input LUTs 2,201 29,504 7% Canny
FPGA 31497.8 3.148
Number of occupied Slices 1,401 14,752 9%
Total Number of 4 input LUTs 2,369 29,504 8% MSE gives mean square difference between original image
Number of bonded IOBs 17 250 6% pixel and edge detected image pixel. If this difference is high,
this means that similarity between images is low [14]. For edge
Number of BUFGMUXs 1 24 4%
detection process, high MSE means that edge detection
Number of MULT18X18SIOs 18 36 50% performance is better. Therefore low PSNR value gives better
TABLE III edge detection performance [15].
FPGA RESOURCE USAGE FOR CANNY
VI. CONCLUSION
Resource Utilization Summary (Canny) In this proposed work, edge detection algorithms used in
Resources Used Available Utilization literature is introduced. Implementation of Sobel, Prewitt and
Canny edge detection algorithms are done in Matlab/Simulink
Number of Slice Flip Flops 916 29,504 3% and FPGA board at the same time. When we look at the
Number of 4 input LUTs 603 29,504 2% resource usage on FPGA board, proposed designs used small
Number of occupied Slices 654 14,752 4% part of FPGA resources. If we analysis edge detection
performance due to Matlab/Simulink and FPGA platform, we
Total Number of 4 input LUTs 688 29,504 2% can see that MSE values of edge detected images from FPGA
Number of bonded IOBs 17 250 6% is greater than Simulink and therefore PSNR values of edge
Number of RAMB16s 13 20 65% detected images from FPGA is less than Simulink platform.
Because of this, results have shown edge detection performance
Number of BUFGMUXs 1 24 4% of FPGA is better than Matlab/Simulink. Another performance
Number of MULT18X18SIOs 5 36 14% investigation is that algorithms, which used for edge detection.
That is Prewitt operator gives better results than Sobel operator
The proposed designs use low resources on FPGA. Sobel due to edge detection performance because of its MSE and
and Prewitt resource usage is same value, because only their PSNR values. Canny algorithm has the best edge detection
convolution mask coefficient is different from each other. performance than Sobel and Prewitt operators. Because, it has
Resource usage of Canny algorithm is greater than Sobel or highest MSE and lowest PSNR values comparing with them.
Prewitt, because Canny algorithm includes more processing As a result, more complex system can be implemented on
steps from Sobel or Prewitt. FPGA, and advantages of edge detection implemented on
____________________________________________________________________________________________________________
93
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Marr, D., Hildreth E.,” Theory of edge detection”, Proc. R. Soc.
Lond. A, Math. Phys. Sci., B 207, 187–217, 1980.
[2] Haralick, R.M.,” Digital step edges from zero crossing of second
directional derivatives”, IEEE Trans. Pattern Anal. Mach. Intell 58–
68, 1984.
[3] Canny, J F.,“A Computational Approach to Edge Detection”, IEEE
Transactions on Pattern Analysis and Machine Intelligence, 679-698,
1986.
[4] Bovik A., “Handbook of Image anf Video Processing”,Academic
Press, 1384, 2010
[5] Christe, S.A., Vignesh, M., Kandaswamy, A., “An efficient FPGA
implementation of MRI image filtering tumour characterization
using Xilinx system generator”, International Journal of VLSI
Desing & Communication Systems, 2 (4): 95-109, 2011.
[6] Vega-Rodríguez., M.A., Sánchez-Pérez, J.M., Gómez-Pulido, J.A.,
“An FPGAbased implementation for median filter meeting the real-
time requirements of automated visual inspection systems”, 10th
Mediterranean Conference on Control and Automation, Lisbon-
Portugal, 2002
[7] Nelson, A.E., “Implementation of image processing algorithms on
FPGA hardware”, Master of Science Thesis, Faculty of the Graduate
School of Vanderbilt University, Nashville, TN-USA, 2000.
[8] Nana, L., Weixing, Z., Shiying, M., Wen, H., “Super resolution
video reconstruction in DSP+FPGA based on lifting wavelet”, The
Tenth International Conference on Electronic Measurement &
Instruments, China, 101-104, 2011.
[9] Davies, E. R. , Machine Vision, Academic Press, 1997.
[10] Aarnink, R. G., Rosette, J. M. C. H., Feitz, Wouter F. J., Debruyne,
F. M. ve Wijkstra, H., A Preprocessing Algorithm for Edge
Detection with Multiple Scales of Resolution, Europan Journal of
Ultrasound, 5, 113-126, 1997.
[11] Ziou, D. ve Tabbone, S., Edge Detection Techniques - An Overview,
Technical Report, No. 195, Dept. Math & Informatique, Universit de
Sherbrooke, 1997.
[12] Sobel, I., An Isotropic 3x3 Gradient Operator, Machine Vision for
ThreeDimensional Scenes, Freeman, H., Academic Pres, NY, 376-
379, 1990.
[13] Prewitt, J., Object Enhancemet And Extraction, Picture Processing
and Psychopictorics (B. Lipkin ve A. Rosenfeld, editör), NY,
Academic Pres, 1970.
[14] P. Vidya, S. Veni and K.A. Narayanankutty, “Performance Analysis
of Edge Detection Methods on Hexagonal Sampling Grid”,
International Journal of Electronic Engineering Research, Volume 1,
pp. 313–328, 2009.
[15] Peter Kellman and Elliot R. McVeigh, “Image Reconstruction in
SNR Units: A General Method for SNR Measurement”, Magn Reson
Med. Author manuscript, Magn Reson Med. Volume 54(6) pp.
1439–1447, 2005.
____________________________________________________________________________________________________________
94
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Aging process causes evident alterations on human estimation systems by using the facial regions emphasizing
facial appearance. Real world age progression on human face is the aging information is an important issue in this field.
personalized and related with many factors such as, genetics, In this paper we investigate the age estimation
living style, eating habits, facial expressions, climate etc. The performances of facial regions: eye, nose, mouth and chin,
wide degree of variations on facial appearance of different
cheek, sides of mouth. For this purpose we designed an age
individuals affects the age estimation performance. In
accordance with these facts discovering the aging information estimation system using the texture features of these facial
contained in facial regions is an important issue in automatic age regions to produce an estimate of the age of a subject. The
estimation. Thus the facial regions emphasizing the aging block diagram of the system is shown in Fig.1. In our method
information can be used for more accurate age estimation. In this the input images are normalized so that the shape variations
context, age estimation performances of facial regions (eye, nose, such as scale, head pose are eliminated and the facial texture
mouth and chin, cheeks and sides of mouth) are investigated in is aligned. Then Local Phase Quantization (LPQ) is used to
this paper. For this purpose, an age estimation method is extract holistic and spatial texture features from facial regions.
designed to produce an estimate of the age of a subject by using After feature extraction, dimensionality reduction is
the texture features extracted from facial regions. In this method
performed with Principal Component Analysis (PCA). Finally
the facial images are warped into the mean shape thus variations
of head pose and scale are eliminated and the texture an aging function is learned using Multiple Linear Regression
information of facial images are aligned. Then the holistic and (MLR) for age estimation. The feature extraction,
spatial texture features are extracted from facial regions using dimensionality reduction and aging function learning steps are
Local Phase Quantization (LPQ) texture descriptor, robust to performed separately for each of the facial regions to discover
blur, illumination and expression variations. After the low the amount of aging information contained in that region.
dimensional representation of these features, a linear aging
function is learned using multiple linear regression. In the
experiments FGNET and PAL databases are used to evaluate the
age estimation accuracies of facial regions i.e. eye, nose, mouth
and chin, cheek and sides of mouth, separately. The results have
shown that the eye region carries the most significant
information for age estimation. Also the mouth and chin, cheek
regions are effective in the prediction of age. The results also
have shown that, using the spatial texture features enhances the
discriminative power of the texture descriptor and thus increases
the estimation accuracy.
Keywords— Age estimation, Local Phase Quantization, Facial
Regions, Regression Fig. 1. System overview
I. INTRODUCTION The rest of the paper is organized as follows. A survey on
Age estimation is the process of estimating the age or age the age estimation methods are given in Section 2. The
group of an individual according to his/her facial information. proposed age estimation approach is explained in Section 3. In
During the aging process, evident alterations occur on human Section 4 the experimental results on various databases are
facial appearance. These variations are personalized and reported and analyzed. Finally, the conclusions are outlined in
affected by factors such as race, genetics, living conditions, Section 5.
eating habits, the frequency of facial expressions, etc. This
makes the age estimation problem much harder than other II. RELATED WORK
facial image processing problems. Therefore the accuracy of Age estimation systems generally consist of age image
the age estimation systems are insufficient, even the human representation and age estimation modules. The aim of age
skills about age estimation are limited. In this context, image representation module is to extract the shape or texture
discovering the amount of aging information contained in based features from facial images. Then classification
facial regions and thus improving the accuracy of age techniques are used to classify the facial images into multiple
____________________________________________________________________________________________________________
95
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
age groups or regression techniques are applied to estimate the the texture information is aligned. The result of warping
age of the subjects. process is given in Fig. 2-c. As the head poses of facial images
The age image representation techniques can be grouped varies, the warped images can be inclined, therefore rotation is
under five topics. The anthropometric models relays on the the last step in image normalization as shown in Fig.2-d.
facial geometry. In these models the distances and the ratios
of these distances are calculated using the fiducial points on
the facial images. As these geometrical features can only deal
with young ages, wrinkle features are used with geometric
features to strengthen the classification performance for older
ages [1-3]. Active Appearance Model (AAM) based age
(a) (b) (c) (d)
estimation methods incorporate shape and appearance
information together. For this reason AAM’s are frequently Fig. 2. Image normalization a) Facial image labeled with 68 landmark points
used in age estimation methods [4-6]. In some studies AAM b) Delaunay triangulation c) The image after warping into the mean shape d)
Rotated image
features are extracted as global facial features and fused with
local facial features for efficient age estimation [7]. In Aging B. Facial Regions
Pattern Subspace method, the sequence of an individual’s Investigating the aging information contained in facial
aging face images are used to model the aging process [8]. But regions is important for age estimation system’s design. In this
age manifold methods uses the images of different individuals way, the accuracies of age estimation systems can be
at different ages to learn the common aging pattern. They improved by using the facial regions emphasizing the aging
utilize manifold embedding techniques to discover the aging information. For this purpose the facial images are divided
trend in a low dimensional space [9, 10]. Appearance models into regions and the age estimation algorithm is applied to
are mainly focused on aging-related facial feature extraction these regions separately. In the study the age estimation
using various texture descriptors such as local binary patterns, performances of eye (19055), nose (5570), mouth and chin
Gabor filters, histograms of gradients etc. [7, 11-13].
(8570), cheek (6560) and side of mouth (4045) regions are
All the studies mentioned above generally uses the whole
determined. The facial regions used in the experiments are
face in feature extraction phases. Unfortunately the number of
shown in Fig. 3.
works examining the age estimation performances of facial
regions/parts is insufficient in the literature. Lanitis [14]
investigated the significance of facial parts in age estimation.
In the experiments, the age estimation performances of whole
(a)
face (including the hairline), internal face, upper part of the
face and lower part of the face are calculated. AAMs are used
to represent the shape and appearance of facial parts with
(b) (b)
model parameters. The results showed that the upper part of (c)
the face gives the minimum age estimation error than other
parts. El-Dib and Onsi [15] used bio-inspired features to
analyze the different facial parts: eye wrinkles (covering the (d) (d)
eyes and the area under the eyes), internal face and whole face. (e)
They built six support vector regression and one support
Fig. 3. Facial regions a) eye b) cheeks (left and right) c) nose d) side of mouth
vector machine model to estimate the age of a subject. Their
(left and right) e) mouth
results showed that the eye region covering the eyes and the
area under the eyes contains the most important aging features C. Feature Extraction with LPQ
when compared with others. LPQ is a blur insensitive texture descriptor based on the
blur invariance property of the Fourier phase spectrum [16]. In
III. PROPOSED METHOD
this method LPQ codes are computed in local image windows
A. Image Normalization using discrete Fourier transform (DFT) and the results are
In order to eliminate the shape variations such as head pose, presented as a histogram.
scale, size, etc. and to align the texture information, image The spatially invariant blurring of an image can be
normalization is applied on facial images. Image expressed by a convolution between the image and the point
normalization is performed by warping the facial images into spread function (PSF). In the frequency domain this is equal
the mean shape obtained from training set. The facial images to 𝐺(u) = 𝐹(u). 𝐻(u) , where 𝐺(u) , 𝐹(u) and 𝐻(u) are the
in the training set are labelled with 68 landmark points as DFTs of the blurred image, the original image and the PSF,
shown in Fig.2-a. Mean shape is obtained by taking the mean respectively. Considering the phase of the spectrum we have
of all coordinates of these points. Then all the images are ∠𝐺(u) = ∠𝐹(u) + ∠𝐻(u). If we assume that the blur PSF
warped into the mean shape using Delaunay triangulation as ℎ(x) is centrally symmetric, ℎ(x) = ℎ(−x) , its Fourier
shown in Fig. 2.-b and affine transformation, so the landmark transform is always real valued, 0 for ∠𝐻(u) ≥ 0 and 𝜋 for
points of all images are matched with the mean shape and also ∠𝐻(u) < 0. This means that ∠𝐺(u) = ∠𝐹(u) for all 𝐻(u) ≥
____________________________________________________________________________________________________________
96
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
0. In other words the phase of the observed image ∠𝐺(u) is E. Age Estimation
invariant to centrally symmetric blur, at the frequencies where After finding the low dimensional representation of features,
𝐻(u) is positive. the age estimation problem is recast as a multiple linear
If the NxN neighborhood around a pixel x is denoted as 𝒩𝑥 , regression as 𝐿 = 𝑌𝐵 + 𝑒 where 𝑌 is the data matrix, B is the
the two dimensional (2-D) DFT of 𝒩𝑥 is defined by, unknown parameter vector, L is the age label vector and e is
the error vector. In the learning phase the unknown parameters
𝑇y
𝐹(u, x) = ∑y∈ 𝒩𝑥 𝑓(x − y)𝑒 −𝑗2𝜋u = wu𝑇 fx (1) are estimated by means of least squares, or robust regression.
The regression function used in this study is a linear function
𝑇
where wu is the basis vector of the 2-D DFT at frequency u, given by 𝑙̂ = 𝛽̂0 + 𝛽̂1 𝑦 where 𝑙̂ is the estimate of age, 𝛽̂0 is
and fx is the vector containing all N2 pixels in 𝒩𝑥 . Only the the offset, 𝛽̂1 is the weight vector containing the coefficients
complex coefficients of u1=[a, 0]T, u2=[0, a]T, u3=[a, a]T, for each value in the feature vector and y is the low
u4=[a,−a]T are considered in LPQ. a is a scalar frequency dimensional representation of the extracted feature vector.
below the first zero crossing of 𝐻(u) that satisfies 𝐻(u) ≥ 0.
For each pixel position this results in a vector given by IV. EXPERIMENTS AND RESULTS
In this paper the FGNET and PAL databases are used to
Fx𝑐 = [𝐹(u1 , x), 𝐹(u2 , x), 𝐹(u3 , x), 𝐹(u4 , x)], (2) evaluate the age estimation performances of facial regions.
FG-NET database [17] is composed of 1,002 images that were
Fx = [Re{Fx𝑐 }, Im{Fx𝑐 }]𝑇 , (3) retrieved from real-life albums of 82 subjects, thus includes
variations of head pose, occlusion, illumination, facial
where Re{. } and Im{. } return real and imaginary parts of a expressions, etc. The age range in this database is 0-69 years,
complex number, respectively. Then Gx is computed for all but the images are not uniformly distributed according to the
image positions, i.e., x ∈ {x1, x2, . . . , xN}, and the resulting ages. This can be a disadvantage for the estimation accuracy.
vectors are quantized using a simple scalar quantizer, The PAL aging database [18] contains 580 images of
different persons taken under natural lighting conditions using
1, if 𝑔𝑗 ≥ 0 a digital camera. The images include various expressions such
𝑞𝑗 = { (4) as neutral faces, anger, sadness or smiling. The age
0, otherwise distribution of the images in this database is between 18 and
93 years old and also not uniform.
where 𝑔𝑗 is the jth component of Gx . The quantized The age estimation performance of the system is evaluated
coefficients are represented as integer values between 0-255 using n-fold cross validation method. In this method all the
using binary coding using (5), and the histogram of these samples are randomly partitioned into n equal sized
integer values is used as a feature vector. subsamples. Then one subsample is used as test set and n-1
subsamples are used as training set. This procedure is repeated
𝑏 = ∑8𝑗=1 𝑞𝑗 2𝑗−1 (5) n times until each of the subsamples are used once as a test set.
Then the average of all n estimations is considered as the
The LPQ texture descriptor represents the input image as a system performance. In our experiments we set n=3.
histogram of 256 bins. In this holistic representation, the The performance is measured using the Mean Absolute
histogram is produced without taking into account the spatial Error (MAE) metric given as,
information of the pixels. The discriminative power of the 𝑁
texture descriptor can be enhanced by using the spatial MAE = ∑𝑖=1
𝑡
|𝑙̂𝑖 − 𝑙𝑖 | /𝑁𝑡
histograms. The spatial histograms are produced by
concatenating the local histograms extracted from small image where 𝑙̂𝑖 is the estimated age value of ith test sample, 𝑙𝑖 is the
blocks. real age value of ith test sample, and 𝑁𝑡 is the total number of
test samples.
D. Dimensionality Reduction In the experiments the age estimation performances of
In order to find a lower dimensional subspace of the facial regions are calculated using the features extracted with
extracted features and to obtain the significant features for age LPQ texture descriptor and regression. LPQ is performed on
estimation, dimensionality reduction is performed using PCA. 5 × 5 local image windows. The LPQ histograms are first
PCA method finds the embedding that maximizes the extracted from the whole facial region resulting the holistic
projected variance given by 𝑊𝑜𝑝𝑡 = arg max‖𝑊‖=1 𝑊 𝑇 𝑆𝑊 description of that region. Then the region is divided into
where 𝑆 = ∑𝑛𝑖=1(𝑥𝑖 − 𝑥̅ ) (𝑥𝑖 − 𝑥)𝑇 is the scatter matrix, 𝑥𝑖 is 𝑛 × 𝑛 blocks, LPQ histograms are extracted from each block
and concatenated to obtain the spatial description of the region.
ith feature vector with 𝑥𝑖 ∈ 𝑅𝐷 and 𝑥̅ is the mean of the feature
The spatial features are used to enhance the discriminative
vectors. By solving this problem, a set of 𝑑 ≤ 𝐷 eigenvectors
power of the texture descriptor. The spatial LPQ histograms
associated to the d largest eigenvalues of S is obtained. Then
are extracted for n=2, 3, 4, 5, 6 and age estimation is
dimensionality reduction is performed by projecting all
performed using these features. The effect of using the spatial
samples on the projection subspace using 𝑦𝑖 = 𝑊 𝑇 𝑥𝑖 with
texture features of facial regions for age estimation on
𝑦𝑖 ∈ 𝑅𝑑 . FGNET and PAL databases are given in Fig. 4. We can see
from the figure that the holistic feature representation is not
____________________________________________________________________________________________________________
97
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
encouraging and increasing the number of blocks in spatial by dividing the region into number of blocks, extracting
feature representation generally increases the estimation features from these blocks and concatenating them. The age
accuracy of the facial regions. estimation accuracies of the facial regions are evaluated
separately and the experiments on FGNET and PAL databases
have shown that the most of the aging information is
contained in the eye region. Moreover the age estimation
accuracy of the eye region is close to the whole face. As the
estimation of the whole face is better than the regions alone,
determining the weights of these regions in age estimation and
thus achieving better estimation accuracy is our future work.
REFERENCES
[1] Y. H. Kwon and N. V. Lobo, “Age Classification from Facial Images”,
Computer Vision and Image Understanding, vol. 74, no. 1, pp. 1-21,
1999
[2] W. B. Horng, C. P. Lee and C. W. Chen, “Classification of Age Groups
(a) Based on Facial Features”, Tamkang Journal of Science and
Engineering, vol. 4, no. 3, pp. 183-192, 2001.
[3] M. M. Dehshibi and A. Bastanfard, “A new algorithm for age
recognition from facial images”, Signal Processing, vol. 90, no. 8, pp.
2431-2444, 2010.
[4] A. Lanitis, C. Taylor and T. Cootes, “Toward Automatic Simulation of
Aging Effects on Face Images”, IEEE Transactions on Pattern Analysis
and Machine Intelligence, vol. 24, no. 4, pp. 442-455, 2002.
[5] S. Kohli, S. Prakash and P. Gupta, “Hierarchical age estimation with
dissimilarity-based classification”, Neurocomputing, vol. 120, pp.
164-176, 2013.
[6] W. L. Chao, J. Z. Liu and J. J. Ding, “Facial age estimation based on
label-sensitive learning and age oriented regression”, Pattern
Recognition, vol. 43, pp. 628-641, 2013.
[7] S. E. Choi,Y. J. Le, S. J. Lee, K. R. Park and J. Kim, “Age estimation
using a hierarchical classifier based on global and local facial features”,
(b) Pattern Recognition, vol. 44, pp. 1262-1281, 2011.
Fig. 4. The effect of using the spatial texture features of facial regions for age [8] X. Geng, Z. H. Zhou and K. S. Miles, “Automatic Age Estimation
estimation a) FGNET database b) PAL database Based on Facial Aging Patterns”, IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. 29, no. 12, pp. 2234-2240,
The age estimation accuracies of the whole face and facial 2007.
[9] Y. Fu and T. S. Huang, “Human Age Estimation with Regression on
regions are tabulated in Table 1. We can see from the results Discriminative Aging Manifold”, IEEE Transactions on Multimedia,
that the age estimation accuracy of the whole face and the eye vol. 10, no. 4, pp. 578-584, 2008.
region are close to each other. These results indicate that the [10] G. Guo, Y. Fu, C. R. Dyer and T. S. Huang, “Image-Based Human Age
eye region carries the most significant information for age Estimation by Manifold Learning and Locally Adjusted Robust
Regression”, IEEE Transactions on Image Processing, vol. 17, no. 7,
estimation. Also the mouth and chin, cheek regions are more pp. 1178-1188, 2008.
effective than other regions in the prediction of age. [11] C. Chen, W. Yang, Y. Wang, K. Ricanek and K. Luu,”Facial Feature
TABLE I Fusion and Model Selection for Age Estimation”, IEEE International
MAE’S OF FACIAL REGIONS FOR FGNET AND PAL DATABASES
Conference on Automatic Face & Gesture Recognition and Workshops
(FG’11), pp. 200-205, 2011.
Regions FGNET PAL [12] G. Guo, G. Mu, Y. Fu and T. S. Huang, “Human Age Estimation Using
Bio-Inspired Features”, Proceedings of IEEE Conference on Computer
Whole face 5.83 6.94 Vision and Pattern Recognition, pp. 112-119, 2009.
Eye 6.30 7.83 [13] J. Liu, Y. Ma, L. Duan, F. Wang and Y. Liu, “Hybrid constraint SVR
Nose 8.04 10.24 for facial age estimation”, Signal Processing, vol. 94, pp. 576-582,
Mouth and chin 7.71 8.46 2014.
[14] A. Lanitis, “On the Significance of Different Facial Parts for
Cheek 8.21 10.04 Automatic Age Estimation”, 14th International Conference on Digital
Side of mouth 8.66 10.52 Signal Processing, vol. 2, pp. 1027-1030, 2002.
[15] M. Y. El Dib and H. M. Onsi, “Human age estimation framework using
V. CONCLUSION different facial parts”, Egyptian Informatics Journal, vol. 12, no. 1, pp.
53-59, 2011.
In this paper the age estimation performances of the facial [16] V. Ojansivu V and J. Heikkila, “Blur Insensitive Texture Classification
regions are investigated. For this purpose the facial image is Using Local Phase Quantization”, Image and Signal Processing, vol.
divided into five regions: eye, nose, mouth and chin, cheek, 5099, pp. 236-243, 2008.
side of mouth. Then LPQ is used to extract features from these [17] FG-Net aging database. Available: http://sting.cycollege.ac.cy
/~alanitis/fgnetaging.
regions. In the feature extraction phase holistic representation [18] M. Minear and D. C. Park, “A lifespan database of adult stimuli”,
of the region is obtained by applying the texture descriptor on Behavior Research Methods, Instruments and Computers, vol.36, no.4,
the whole region. Also the spatial representation is obtained pp.630-633, 2004.
____________________________________________________________________________________________________________
98
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Today, railway transportation is one of the transport for this transportation. The switch has been developed for
modes commonly used. Compared to other transport modes, more than one train can use the same railway line. Trains can
railway traffic is highly critical. Multiple railway vehicles run change their direction, give way to another train or proceed to
constantly on one or two lines. Rail switch passages are used to the standby points at the station due to the switch system. The
prevent locomotives from colliding with one another and avoid
development of the switch system has made the railway line
traffic disruptions. Through switch passages, locomotives pass
from one line to another. Friction between rail and wheels on quite utilizable. Therefore, the switch system has a great
switch passages is considerably high. This friction leads to importance for the railway transportation. Excessive friction
failures on switch passages. Unless these failures are diagnosed and wear occur during switching [4]. Such undesirable
early and remedied, significant accidents emerge. situations lead to major wear on the rail in time. Major
In this study, a new approach based on image processing has accidents may occur in the event of failure in the early
been presented for detection of rail switch passages on railway diagnosis of these wears. The accidents that may occur can
lines. A test vehicle has been created in order to test the proposed cause loss of life and injury for passengers. However, the
approach and apply it on a real-time system. Railway line is railway line hinders the traffic for a while during the accident,
monitored by digital cameras fixed on this test vehicle. Image-
and major financial losses are formed. Today, the railway line
processing approach is developed on the real-time images
captured from the railway line and the switch passages on the is inspected at regular intervals, and the faults which may
line are detected. The image-processing approach consists of arise are diagnosed early. There are many literature studies
three main parts including pre-processing, feature extraction and related to switch system. Johansson et al. [5] conducted a
processing of the features obtained. At the pre-processing stage, study for the detection of the corrosion faults caused by
the basic image processing methods are used. At the feature friction and wear during switching. In the study, Switches &
extraction stage, Canny edge extraction algorithm is used and Crossings components are monitored. They modeled the
hence the edges in the image are detected. Hough transform contact of the wheel with the rail by simulation during
method is used at the stage of processing of the extracted switching. As a result of modeling, observation was made
features. Following Hough transform stage, straight lines and
according to the loads that may occur on the rails. Bocciolone
angles of these lines are obtained on the image. Taking into
account the angle of each straight line, the junction points of the et all.[6] proposed a signal processing-based method for the
lines are calculated. Thus, rail switch passage and switch types detection of the faults that occur on the rail surface. In the
are detected. The proposed image-processing approach is highly proposed method, the vibration signals made by the train
fast and real time-based. Compared to the existing studies in the while going on the railway line were used. Fault detection is
literature, it is seen that the proposed method gives fast and performed by improving the signal processing methods
successful results. This study intends to diagnose the failures on through vibration signals. Palsson [7] completed a thesis study
switch passages early and prevent potential accidents. on optimization in the Switches and Crossing sections on the
railways. In this thesis study, Switches and Crossing
Keywords— Railway, Condition Monitoring, Fault Detection, components were monitored and the faults that occurred in
Image Processing, Railroad Switches. these components were examined. The changes in the
Switches and Crossing areas were observed during switch
transitions. The load distribution from one rail to another rail
I. INTRODUCTION was examined. In addition to these studies, many studies have
Railway transportation is commonly used in the freight and been conducted for the detection of the faults occurring on the
passenger transport. Railway vehicles provide transportation rail surface and in its components. Quingyong et al. [8] an
using a particular rail line [1,2]. These vehicles are preferred image processing based contactless method is proposed and
by people because they can carry quite heavy loads and large detected faults on rails in real-time. Rail surface is extracted
number of passengers. It is very important for rails to be by using railway images. Surface faults are determined by
robust because railway vehicles provide transportation on the contrast stretching of the obtained surface. This method, is
rails [3]. The railway lines are the most critical components used only railway surface to detect faults. Limin et al. [9],
____________________________________________________________________________________________________________
99
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
proposed a method that detects faults on the surface by using Point (Switch) Stock Rails Crossing Running Rails
machine vision techniques. The roughness and cracks are Blades Stretcher Check Rails
determined with image processing algorithms. Edge detection
Wing Rails
and feature extraction methods are used.
Ying et al. [10] proposed an automatic monitoring approach Running Rails
to detect and evaluate the railway components. The proposed Toe Heel
method is working on a train with 16km/h speed. Ray
component detection is carried out by applying Sobel edge
detection and Hough transform methods. Status and positions
of the components are observed and analysis of them is Fig. 2 Example of a switch crossing system and components
carried out. The general architecture of the proposed railway
inspection system is presented in Figure 1. As it is seen in Figure 2, the switch crossing system has
Track Ling-term
many components. The components of Switch Blades, Toe,
Rail
Exeption Predictive Stretcher and Heel serve as a switching. The components of
Detection and Assessment
Data Component
Temporal Crossing, Check Rails and Wing Rails allow train to switch to
Acquisition Detection and
Optimization
Condition another rail in a secure way. The components of Check Rails
Analysis
and Wing Rails are the most important components that
Maintenance
prevent the derailment of the train. The switch crossings are
Backend
Planning and
Database
very important on the railway line. Major faults may occur as
Scheduling
a result of the failures of the components on the switch
crossings. Therefore, the railway line especially the switch
Fig. 1 General architecture of the railway inspection system proposed in the crossings should consistently be inspected. It is required to
literature [10]
make the necessary maintenance even considering the small
faults. On switch crossings, to which direction the train will
In this study, an image processing-based method was go is determined using the components of Switch Blades, Toe,
proposed for the detection of the crossing points on the Stretcher and Heel. The changes that occur on the switch
railways. Images were taken for different types of switches crossings when a train goes straight or to the right direction on
using a camera fixed on the train. The switch crossings were an example railway line are seen in Figure 3.
determined using image processing algorithms on the images
taken. X Y
____________________________________________________________________________________________________________
100
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
e) f)
Fig. 4 Railway switch types a) Single left switch b) Single right switch c)
Symmetric switch d) Compound switch e) Cross switch (full cross) f) Determine rails with angle limitation
crossover [13-15]
____________________________________________________________________________________________________________
101
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
After the gauss masking process, edge extraction is the angle values of the lines obtained and the start and end
performed by using the canny masking matrix of the image points of the line. The method of the slope of a line two points
from X and Y direction. The image is formed according to of which are known was used in obtaining these lines. The
certain angle limits by calculating the gradient magnitude on lines representing the rails are detected by considering the
the image matrices obtained. slope ranges of the lines on the railway image. The slope of a
The edge extraction process is achieved more successfully line two points of which are known is given in equation 3.
because Canny edge extraction algorithm uses the Gauss filter
before performing edge extraction on the image. Canny edge B y Ay
extraction algorithm gives better results compared to other m (3)
Bx Ax
edge extraction algorithms.
Hough transform is used to obtain the lines on the image
after the edge extraction process [18]. Hough transform is In equation 3, m represents the slope, ( Ax , Ay ) represents
generally used in detecting geometric models on the image in the location information of the A point and ( Bx , B y ) represents
combination with the edge extraction methods. Hough
the location information of the B point. The switch crossings
transform is applied onto the images, the edge information of
on the railway line are detected according to the slope value
which are obtained. The method transforms the problem of
obtained by applying this equation on the gray image. Two
finding a shape into a problem of finding a density by
points of each line are known. These two lines intersect and
transferring the information in the image space to the
give the switch crossing zone. For the calculation of the
parameter space [19-21]. Hough transform does not require
crossing point, the required calculations were made for the
that the points on the image are side-by-side or connected to
junction point of the line two points of which are known. First
detect an image. The pseudo code of Hough transform used in
of all, the points of the line taken from the normal image are A,
the proposed method is given in Figure 7.
B, C and D. The junction point of these two lines is E. The
Image Edge detection image in binary format
Column Width of Image junction point of two lines two points of which are given is
Row Height of Image shown in the analytical plane as in Figure 8.
for i=1:Column
for j=1:Row
if (Image(i,j) is an edge detected image)
for θ=0:
.
A(x,y)
y
incline = m1
r = Column*cos(θ) + Row*sin(θ)
r = round(r)
H(θ,r) =H(θ,r)+1
F(x,y)
. incline = m2
D(x,y)
.
E(x,y)
end
end
end incline = m3
.
.
end
This equation has two parameters to specify any straight In order to calculate the values of the E point given in Figure
right. These parameters are slope ( m ) and junction point ( n ).A 8, another line passing from E,F and G point is drawn. The
point in the straight lines parameter space is shown as equation of the line the slope of which is m3 is given in
( m, n ).The straight lines are transformed into coordinate equation 4.
system and shown in equation 2.
G y Fy (4)
m3
r x cos( ) y sin( ) (2) G x Fx
In this equation, the nearest vector from the origin is By calculating the location information of F and G given in
represented by a pair of ( r , ).All lines with angles ranging equation 4, the final state of the equation 4 is given in
between 60 and 150 on the detected lines are taken into equation 5.
account. The junction points of the lines are calculated using
____________________________________________________________________________________________________________
102
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(( Dy C y ) * (Cx Ax )) ((C y Ay ) * ( Dx Cx ))
m3 (5)
(( Dy C y ) * ( Bx Ax )) (( By Ay ) * ( Dx Cx ))
E x Ax ( Bx Ax ) * m3
(6)
E y Ay ( B y Ay ) * m3
c) d)
Main Railway
c) d)
a) b)
e) f)
Fig. 11 Performing feature extraction by applying edge extraction on the
image
____________________________________________________________________________________________________________
103
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. CONSLUSIONS
e) f) ACKNOWLEDGMENT
Fig. 12 Detection of the switch crossings on the railway line by Hough This work was supported by the TUBITAK (The
transform Scientific and Technological Research Council of Turkey)
under Grant No: 114E202.
The switch crossings in Figure 12 were detected. The REFERENCES
image of a normal railway without switch crossing is [1] S. V. Sawadısavı, “Development of Machine-Vision Technology for
presented in Figure 12.f. No switch crossing was detected Inspectıon of Railroad Track,” Graduate College of the University of
there. The switch crossings were detected successfully in Illinois at Urbana-Champaign, 2010.
other images. Within the context of this study, many images [2] Y. Santur, M. Karaköse, I. Aydın, E. Akın, “IMU based adaptive blur
removal approach using image processing for railway inspection,” In
were obtained for the test procedure. There are 100 switch 2016 International Conference on Systems, Signals and Image
images on these images. These switch images belong to Processing (IWSSIP) (pp. 1-4), 2016.
different types. The numbers of successful and unsuccessful [3] O. Yaman, M. Karakose, E. Akin, “PSO Based Diagnosis Approach
images when the proposed method was applied on these for Surface and Components Faults in Railways,” International
Journal of Computer Science and Software Engineering (IJCSSE), vol.
images are given in Table 1. 5, pp. 89–96, May. 2016.
[4] L. Xin, V.L. Markine, I. Shevtsov, “Numerical analysis of rolling
contact fatigue crack initiation and fatigue life prediction of the railway
crossing,” In CM2015: 10th International Conference on Contact
Mechanics, Colorado Springs, USA, 30 August-3 September 2015.
____________________________________________________________________________________________________________
104
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[5] A. Johansson, B. Palsson, M., Ekh, J.C. Nielsen, M.K. Ander, J. [14] S. Jalili Hassankiadeh, “Failure analysis of railway switches and
Brouzoulis, E. Kassa, “Simulation of wheel–rail contact and damage in crossings for the purpose of preventive maintenance,” 2011.
switches & crossings,” Wear, 271(1), 472-481, 2011. [15] G. Schupp, C. Weidemann, L. Mauer, “Modelling the contact between
[6] M. Bocciolone, A. Caprioli, A. Cigada, A. Collina, “A measurement wheel and rail within multibody system simulation,” Vehicle System
system for quick rail inspection and effective track maintenance Dynamics, 41(5), 349-364, 2004.
strategy,” Mechanical Systems and Signal Processing, 21(3), 1242- [16] M. Karakose and M. Baygin, “Image processing based analysis of
1254, 2007. moving shadow effects for reconfiguration in pv arrays,” in IEEE
[7] B. Palsson, “Optimisation of railway switches and crossings,” International Energy Conference (ENERGYCON), pp. 724-728, May
Chalmers University of Technology, 2014. 2014.
[8] L. Qingyong, R. Shengwei, “A Real-Time Visual Inspection System [17] O. Yaman, M. Karakose, E. Akin, I. Aydin, “Image processing based
for Discrete Surface Defects of Rail Heads,” IEEE Transactions on fault detection approach for rail surface,” In Signal Processing and
Instrumentation and Measurement, Vol. 61, 2012, 2189-2199. Communications Applications Conference (SIU), 2015, pp. 1118-1121.
[9] C. Limin, L.Yin, W. Kaimin, “Inspection of rail surface defect based [18] L. Qingyong, R. Shengwei, “A Real-Time Visual Inspection System
on machine vision system,” 2nd International Conference on for Discrete Surface Defects of Rail Heads,” IEEE Transactions on
Information Science and Engineering (ICISE), 3793 - 3796, 2010. Instrumentation and Measurement, vol. 61, 2012, pp. 2189-2199.
[10] L. Ying, T. Trinh, N. Haas, C. Otto, S. Pankanti, “Rail Component [19] A. Bouchikhi, A.O. Boudraa, J.C. Cexus, T. Chonavel, “Analysis of
Detection, Optimization, and Assessment for Automatic Rail Track multicomponent LFM signals by Teager Huang-Hough transform,”
Inspection,” IEEE Transactions on Intelligent Transportation Systems, IEEE Transactions on Aerospace and Electronic Systems, 1222-1233,
Vol. 15, 2014, 760 – 770. 2014.
[11] P. Wang, J. Xu, K. Xie, R. Chen, “Numerical simulation of rail profiles [20] I. Aydin, E. Karaköse, M. Karaköse, M.T. Gençoğlu, E. Akın, “A new
evolution in the switch panel of a railway turnout,” Wear, 2016. computer vision approach for active pantograph control,” In
[12] E. Kassa, J.C. Nielsen, “Dynamic interaction between train and railway Innovations in Intelligent Systems and Applications (INISTA), 2013
turnout: full-scale field test and validation of simulation models,” IEEE International Symposium on (pp. 1-5). IEEE, 2013.
Vehicle System Dynamics, 46(S1), 521-534, 2008. [21] M. Karakose, “Sensor Based Intelligent Systems for Detection and
[13] W.J. Zwanenburg, “Modelling degradation processes of switches & Diagnosis,” Journal of Sensors, 2016.
crossings for maintenance & renewal planning on the Swiss railway
network,” 2009.
____________________________________________________________________________________________________________
105
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Istanbul, Turkey
{gulustan, kavincan}@yildiz.edu.tr
Abstract— Wireless sensor networks are used widely in daily II. D EFINITIONS : T RUST, T RUSTWORTHINESS , R ISK ,
life. With that increase the problems faced in WSN usage R EPUTATION
have become more interesting for the research community.
The problems that outstand the most are related to the trust Josang et al. [20] define trust and trustworthiness based
and energy issues. As wireless sensor networks have limited on the definitions of Gambetta [16]. Solhaug et al. [39]
energy and computing power, cryptographic algorithms are not define trustworthiness as objective probability that the trustee
suitable for these type of networks. Along with that line, in this
work we explore trust models developed for wireless sensor
performs a particular action on which interests of the trustor
networks. depend. Trust is a subjective probability varying from 1
KeywordsWireless Sensor Networks, Trust, Trust Models (complete trust) to 0 (complete distrust) [20].
As trust is the believed probability and trustworthiness is the
actual probability, there can be a difference between them.
I. I NTRODUCTION This difference introduces the risk factor [12]. Risk increases
if the trust is misplaced.
Trust is quite important for self-configurable and autonomous
Reputation is also a concept that is very related to trust.
systems such as WSNs [15]. WSNs are very vulnerable
Sometimes reputation and trust is used in the same context
environments due to computational limitations, energy con-
however they have different meanings. Reputation is the
straints and network attacks. In addition, WSNs are very
opinion of an entity about the other entity. However trust
open to physical world effects such as a person walking
is derivation of reputation of an entity.
on a field can step on a sensor and make it dysfunctional.
A trust management scheme can make a WSN tolerant to A. Information Trust
node failures and misbehavior by assisting decision making
process. For example, a node can decide to cooperate with There are different types of trust such as social trust,
a node or not based on the feedback it will receive from the cognitive trust, and communication trust. In this work, we
trust model. Trust research on WSN is very new, few systems are assessing the information trust of data items and sensor
have considered it [17], [45]. More research has been done on nodes.
Trust in Ad-hoc and P2P networks. Although these network Information trust or data trust refers to the trust placed on
types have many similarities to WSN, still a separate trust data produced by objects or processes. Information trust in a
management system has to be developed for WSN because of network is important because it can prevent erroneous data
their specific characteristics such as energy and computation to accumulate in the network. In a network, a node can (i)
constraints. create data (ii) process the data such as fusion (iii) pass the
Data collection is very important in the process of designing data along. The trust of data depends on the trust of the node
a trust management system. The system should be history that creates the data and the trust of the nodes the data has
aware, past behaviors should be taken into consideration visited. Information trust in a network can be categorized
[15].WSNs should be history aware as the trust calculation into three: (i) creator node’s subjective view of the trust (ii)
is considering past behaviors. Moreover, every node keeps objective trust assessment of the data by the neighboring
their past behavior statistics regarding the data they produce nodes (iii) changes in information trust as the data travels
such as average error of the created data in the past time along the network.
intervals. One of the biggest constraints is the overhead that B. Properties of Trust
can be caused by the trust model. Trust model should be as
lightweight as possible [15]. Cho et al. [12] list characteristics of trust as follows in their
There are many different data that can be used as input of survey paper.
the trust model. For example, a node that is not alive for a • dynamicity : Due to node failures and mobility, sen-
long time or a node that appears or disappears randomly may sor network is highly dynamic hence trust should be
not be trusted. On the communication layer, a node which is dynamic too.
misreporting will not be trusted. For instance, a node which • subjectivity : Nodes might decide to put different levels
is giving a fire alarm when conditions are calm should be of trust on same nodes due to dynamicity of the network
given a low trust value [15]. [1].
____________________________________________________________________________________________________________
106
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
• transitivity : Trust is not necessarily transitive. For of peers. The resurrecting duckling model in [40] and its
transitivity, we need two types of trust, trust in a trustee descendants [6] use out-of-band channels to authenticate key
and trust in recommendations of the trustee. exchange. The established trust between the nodes is binary,
• asymmetry : Trust is not necessarily transitive. A node either secure or not secure.
may trust a node however the trustee node may not trust There are other trust models for peer-to-peer systems which
the trustor. we do not want to go into details of as we are interested in
• context-dependency : Trust is context-dependent [7]. trust models for sensor networks. Other trust mechanisms
For instance, a node may trust the image data coming surveyed by Momani and Challa[27] are SECURE[11],
from a node but may not trust the audio data coming Distributed Trust Model[2], Bayesian Network Model [43],
from the same node. UniTec[22], BambooTrust[23], B-trust model[34].
assessing the trust of entities. Hence entities form their • Use of low-cost cryptography to protect the integrity of
____________________________________________________________________________________________________________
107
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
108
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[15] M. Fernández-Gago, R. Roman, and J. Lopez. A survey on the appli- [41] J. Walters, Z. Liang, W. Shi, and V. Chaudhary. Wireless sensor
cability of trust management systems for wireless sensor networks. In network security: A survey. Security in distributed, grid, mobile, and
Security, Privacy and Trust in Pervasive and Ubiquitous Computing, pervasive computing, page 367, 2007.
2007. SECPerU 2007. Third International Workshop on, pages 25–30. [42] Y. Wang, G. Attebury, and B. Ramamurthy. A survey of security
IEEE, 2007. issues in wireless sensor networks. IEEE Communications Surveys
[16] D. Gambetta. Can we trust trust. Trust: Making and breaking and Tutorials, 2006.
cooperative relations, pages 213–237, 2000. [43] Y. Wang and J. Vassileva. Bayesian network-based trust model. In
[17] S. Ganeriwal, L. Balzano, and M. Srivastava. Reputation-based International Conference on Web Intelligence, pages 372–378. IEEE,
framework for high integrity sensor networks. ACM Transactions on 2003.
Sensor Networks (TOSN), 4(3):15, 2008. [44] L. Xiong and L. Liu. A reputation-based trust model for peer-to-peer
[18] A. Jøsang, E. Gray, and M. Kinateder. Simplification and analysis e-commerce communities. In E-Commerce, 2003. CEC 2003. IEEE
of transitive trust networks. Web Intelligence and Agent Systems, International Conference on, pages 275–284. IEEE, 2003.
4(2):139–161, 2006. [45] Z. Yao, D. Kim, I. Lee, K. Kim, and J. Jang. A security framework
[19] A. Josang, R. Ismail, and C. Boyd. A survey of trust and reputation with trust management for sensor networks. In Workshop of the 1st
systems for online service provision. Decision Support Systems, International Conference on Security and Privacy for Emerging Areas
43(2):618–644, 2007. in Communication Networks, pages 190–198. IEEE, 2005.
[20] A. Jøsang and S. Presti. Analysing the relationship between risk and [46] D. Zhou. Security issues in ad hoc networks. In The handbook of ad
trust. Trust Management, pages 135–145, 2004. hoc wireless networks, pages 569–582. CRC Press, Inc., 2003.
[21] A. Jsang and R. Ismail. The beta reputation system. In Proceedings of [47] L. Zhou and Z. Haas. Securing ad hoc networks. Network, IEEE,
the 15th Bled Electronic Commerce Conference, pages 41–55, 2002. 13(6):24–30, 1999.
[48] T. Zia and A. Zomaya. Security issues in wireless sensor networks. In
[22] M. Kinateder, E. Baschny, and K. Rothermel. Towards a generic
International Conference on Systems and Networks Communications.,
trust model–comparison of various trust update algorithms. Trust
pages 40–40. IEEE, 2006.
Management, pages 119–134, 2005.
[23] E. Kotsovinos and A. Williams. Bambootrust: Practical scalable trust
management for global public computing. In Proceedings of the 2006
ACM symposium on Applied computing, pages 1893–1897. ACM,
2006.
[24] J. Lopez, R. Roman, I. Agudo, and C. Fernandez-Gago. Trust
management systems for wireless sensor networks: Best practices.
Computer Communications, 33(9):1086–1093, 2010.
[25] D. McKnight and N. Chervany. The meanings of trust. Technical
Report WP9604, University of Minnesota Management Information
Systems Research Center, 1996.
[26] R. Molva and P. Michiardi. Core: a collaborative reputation mechanism
to enforce node cooperation in mobile ad hoc networks. Institute
EurecomResearch Report RR-02-062, 2001.
[27] M. Momani and S. Challa. Survey of trust models in different network
domains. Arxiv preprint arXiv:1010.0168, 2010.
[28] J. Newsome, E. Shi, D. Song, and A. Perrig. The sybil attack in sensor
networks: analysis & defenses. In Proceedings of the 3rd international
symposium on Information processing in sensor networks, pages 259–
268. ACM, 2004.
[29] S. Ozdemir. Functional reputation based reliable data aggregation and
transmission for wireless sensor networks. Computer Communications,
31(17):3941–3953, 2008.
[30] P. Papadimitratos and Z. Haas. Securing mobile ad hoc networks.
CRC Press, 2002.
[31] A. Perrig, J. Stankovic, and D. Wagner. Security in wireless sensor
networks. Communications of the ACM, 47(6):53–57, 2004.
[32] A. Pirzada and C. McDonald. Establishing trust in pure ad-hoc
networks. In Proceedings of the 27th Australasian conference on
Computer science-Volume 26, pages 47–54. Australian Computer
Society, Inc., 2004.
[33] B. Przydatek, D. Song, and A. Perrig. Sia: Secure information
aggregation in sensor networks. In Proceedings of the 1st international
conference on Embedded networked sensor systems, pages 255–265.
ACM, 2003.
[34] D. Quercia, S. Hailes, and L. Capra. B-trust: Bayesian trust framework
for pervasive computing. Trust Management, pages 298–312, 2006.
[35] P. Resnick and R. Zeckhauser. Trust among strangers in internet
transactions: Empirical analysis of ebay’s reputation system. Advances
in Applied Microeconomics, 11:127 – 157, 2002.
[36] S. Ries, J. Kangasharju, and M. Mühlhäuser. A classification of trust
systems. In On the Move to Meaningful Internet Systems 2006: OTM
2006 Workshops, pages 894–903. Springer, 2006.
[37] B. Rothstein. Trust, social dilemmas and collective memories. Journal
of Theoretical Politics, 12(4):477–501, 2000.
[38] J. Scott. Social network analysis. Sociology, 22(1):109–127, 1988.
[39] B. Solhaug, D. Elgesem, and K. Stølen. Why trust is not proportional
to risk. In Proceedings of The 2nd International Conference on
Availability, Reliability and Security (ARES), pages 11–18, 2007.
[40] F. Stajano and R. Anderson. The resurrecting duckling: Security issues
for ad-hoc wireless networks. In Security Protocols, pages 172–182.
Springer, 2000.
____________________________________________________________________________________________________________
109
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract — Giant information is available on the digital of success. Another aim is the awareness of scientists who are
environments. Due to it has hard to get the information what you to study on this subject.
want to achieve it, fast and most efficiently. For example, if a This review consists of 3 parts. The definition of
researcher has a summary of the literature about a topic, this summarization and what kinds of strategies have been used
will be much shorter than the duration of the research. From this
will be explained in the first part. Information will be
perspective, automatically occurring the summary of any text, it
is very important for those who want to be informed. Text provided about summarising strategies in the second part.
summarization is the compression of large document. This paper Some strategies will be provided about how summarising
presents a review of automatic text summarizations’ approaches, affects the evaluation of success in third part.
studies, software, algorithms, and methods which is written by
English, Turkish and some other language in the last decades. II. SUMMARIZATION AND APPLICATIONS HELPFUL TO
Automatic text summarization is generally divided into two SUMMARISING
systems. There are extraction based summarizations and
abstraction based summarizations. Firstly, extractions A. Summarization
summarization approach involves selecting sentences of high Text summarization can be defined as a process, which
score from the document based on word and sentence features.
takes a text as input, includes the most important context and
In short, finding the most important sentences from text is
performed. Secondly, it is examined the semantic relationships replaces the original and is shorter [1]. One or more textual
between words and sentences in abstraction approaches. The equipment which are desired to be summarised as an input for
most important difference from the first approach, made system are provided and in the end a comparatively shorter
semantic analysis of words and sentences afterward new words text is provided as the summary. With this transaction a text is
are obtained. Mostly, optimization algorithms, genetic given to the computer programme and a summarily text which
algorithms, fuzzy logic systems, machine learning, graph trees, belongs to the input and includes its most important content as
statistics and probability techniques are used in automatic text the output.
summarization systems.. The emphasis of only the crucial content of a text to be
summarised is called summarization process. What should be
Keywords—Semantic analysis, text Summarization, natural paid attention here is to determine what the content is. The
language processing content which is relevant to the topic is regarded important.
I. INTRODUCTION Majority of the researchers dealing with text summarization
have focused on summarising the texts in English language.
The development of technology with increasing speed has Therefore these studies which have been done according to
brought about textual data in digital cases. Under these grammatical structures of English language cannot be directly
circumstances attaining the desired information in textual data applied to Turkish summarization methods due to the latter’s
has become in terms of time. The desired textual data will be grammatical and structural differences from of the former. For
regarded as successful in terms of time expense if it includes instance some structural methods where there are strong
fewer sentences alongside with useful information. This meanings of the text are free from language. Semantic
subject is crucial as for the usage of effective time consuming, analyse methods in which methods such as parsing and natural
has led to studies about the solution of the problem. language operation are employed however cannot be used in
The fact that a sentence has been cleared off unnecessary Turkish text summarization.
words and sentences can be possible by means of a successful A number of distinctions have been done in literature for
summary of that text. The aim of this review is to provide text summarization. For example, choosing the important
information about strategies developed and studies conducted sentences and translating them without making any difference
for successful summarization. What I paid importance to and evaluating this under another title is called summarising
while informing has been putting more emphasis on the based on extractive. Another one is to comment the content
studies carried on after 2000s. While technology and from a text and rewrite it in shorter and new sentences and this
communication tools are improving rapidly, researches have method is called summarising based on abstractive.
come to the point of communicating more easily has been the Summarising studies based on abstractive require the
reason for this. And at the end of this some standards about employment of natural language operations and detailed
summarising studies are being developed. They have examination of the equipments. In abstractive based methods
employed success measurement softwares and summary of various parsing algorithms are used. In Turkish the by adding
data sets so that they can be of help in evaluation and increase derivational affix to a word a parsing word is formed and by
____________________________________________________________________________________________________________
110
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
eliminating the inflection from a word and thus finding the summarization studies is of rather high importance as an
main word is called stemming [2]. In order to help the assistant. Researches use Porter Stemmer algorithms [5] to
methods used in summarization there are a great number of decode words in English texts.
other applications. Some of these are dividing a full sentence
2) Wordnet: For those researches who want to use the
into its components, analysis and morphologic analysis
semantic relation between words, Wordnet [6] synonym
(finding the stem of word, suffixes or prefixes) [3]. For
dictionary, which define the correlation between words is a
instance words can be eliminated to their stems and therefore
significant tool. Wordnet came about in 80s, was developed to
now words can be formed.
3.0 in 2006 and includes 155,287 words in English. For the
Researchers have developed softwares so as to determine
researches working on summarization subject of Wordnet, it is
the adaptive relations between words and division of words. In
of great importance that it includes a word group consisting of
this part brief information about some these softwares will be
synonyms. The synonyms are correlated in the dictionary.
provided.
Birant, in his study [7], has developed software which does
B. Species of Summarization summarization on its own for Turkish language as well as a
dictionary of the semantic relation which he employed in his
Contrary to the abstractive based summarization, extractive
study and he has also argued that he has improved some
based summarizations are handier and more common.
natural language transaction tools. The two dictionaries;
Summarising without being bound to any particular word and
synonym and close-meaning dictionary and opposite meaning
just using some key words is called general summarization
dictionary that he has provided in his study, have been
system. Summarizations limited to questioning the content is
approved by Turkish language constitution. The text
called inquiry based summary system. There are
summarization system based on rules which he has developed
classifications based on the number of the sources of the text.
has given out results similar to those of others. In another
In this case summarization is divided into two groups as either
study [8] in which sentence purification for Turkish text
one document source or multiple documents. Single document
summarization techniques were used, methods of finding the
based summarization, is the one whose input is a single
stem of word were used as well. In this rule-based study in
textual data, whereas the multiple documents based
which syntax features are used rather than structural features,
summarization is the one which gets multiple documents as
omitting letters from the end of words technique has been
input, evaluates them, deals with the important parts and
successfully used. In the same study the success of techniques
brings together the similar information in them and eventually
which shorten sentences and divide sentences into shorter and
summarises them on a single document. Alongside with all
multi-meaning sentences.
these groupings the document to be translated might be
written in one or more different languages. Some of these III. STUDIES ON SUMMARIZATION
approaches, represent documents to be translated with both
The majority of studies on summarization are in English
semantic features and some structural features which enable
language. The first studies regarding textual summarization
sentences to be stated through sentence scores functions
was around 50 years ago and for English texts. Although
belonging sentences’ importance level.
numerous newer techniques for various problems have been
1) Zemberek: Zemberek is open coded NLP (natural developed, still the traditional techniques are employed in the
language processing) libraries which can basic NLP such as present and most of the problems. The reasons for them to be
decoding, spell checking, back formation written for Turkish used are their being simple, successful, efficient and low cost.
and other Turkish based languages. Due to its being open- The methods requiring abstractive and natural language
coded and unignorable feat it possesses an important position processing, cannot enable the methods developed for English
among the natural language transactions. Many researches being employed for Turkish too due to Turkish language’s
have employed Zemberek while studying Turkish suffixes structure. We can list the studies done on
summarization methods [4]. Researches have quite often used summarization under two general headings [9].
Zemberek in their studies on decoding which reveal the Extractive summarization (summarization with sentence
parsing and stem of words. In terms of structure Zemberek, selection) approaches are to select sentences from a text
consists of two parts as detailed language transactions and according to various techniques and combining them into a
language structure information. The library mainly includes meaningful text so as to get the summarization. These
the softwares necessary for detailed language transactions as methods are also called extractive methods. Another approach
well as structures including stem and additions. It also is abstractive summarization. In which parsing, natural
contains XML libraries, which have extra information about language transaction techniques are used and analysing words
other Turkish based languages for extra information. The in details, brand new words are emerged contrary to sentence
additional function specially in accordance with algorithms selection method. For example instead of this sentence “Selim
those are sensitive for special cases. There is also a unit which checked the menu, asked for his meal and ate" the following
controls voice cases such as sound falls, softening and sentence can be used as "Selim enjoyed his meal". A rich
assimilation. To enable these graphs tree algorithms are used. symbolic list of words is necessary for this method to be used
Zemberek has the ability of processing 12000 words [10].
simultaneously [4]. Zemberek, used as a tool in Turkish
____________________________________________________________________________________________________________
111
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
A. Extractive summarization Studies Suanmali et al. [20], Babar and Patil [21], Kyoomarsi et al.
The main focus of sentence selection technique [22], Binwahlan et al. [23] used fuzzy logic technique for
summarization is the presence of the sentence which is combine these methods. Silla et al. [24], Witte et al.[25],
regarded of vital importance. How will be the decision of Berker and Güngör [26], Filatova et al. [27] and Mcdonald [28]
whether a sentence is important? The answer to this question have rated the important sentences with heuristic, genetic and
is significant in determining the success rate of the sentence Swarm-based optimization algorithms in their studies.
selection summarization technique. In studies prior to 2000s, Alguliev et al. [29] have tried to solve summarize problem
it has been decided that some sentences were far more with nonlinear programming method. Svore et al. [30], Wong
important than others and thus be secluded from others et al. [31] have used to learning machine, Bayesian classifier,
according to a grading key. In these techniques simple but support vector machines, artificial neural networks techniques.
effective deductions are present. The thing that is desired to be Agrawal and Gupta [32] have used the similarity of sentences
emphasised is used more often. For example if the innovations to each other. For this, the K-means method is used. In the
in the field of technology are mentioned, the word technology work of Seinberger et al. [33] and Orasan [34], utilizing the
will be used more often and the sentences in which it is semantic relationships between words cohesion-based
present need to be present in summarization. Here, the point techniques are used. They have benefited from the Wordnet.
which should be kept in mind is that the conjunctions such as - D'avanzo [35] has carried out a study that knows the name of
and- or – but - which are used in daily life conversations, don't the entity and finds multiple words and based on supervised
in fact represent a content. This method called word frequency learning in his study.
was used by Luhn in his study [11]. We have also seen event-based methods in summarization
The name of this method is frequency of the word. Luhn studies. Liu et al. [36], Filotova et al.[37] examine the events
used this method in his study. In addition that, Edmundson in the text with the help of causality techniques. An event-
claimed in his study [12] that a summary must contain term graph is obtained. The same event terms are grouped.
sentences which included some words such as "as a result", Uzundere et al. [38] have examined 13 different features in
"in summary" etc. Edmundson gives more points sentences their study. According to this;
like these. This method is known as clue words phrases. His Head: The words are listed in the titles and subtitles.
other idea is that give more point sentences that contains Sentences are checked whether they have these words.
introduction and conclusion paragraphs. This method is called High Frequency: Holding the number of repetition of all
as sentences location. Also, sentences that contain the words words in the text is created. List is sorted in
in a title and subtitle are important according of Edmundson’s descending direction. Would not include frequently
thought. The name of this method is title method. used words in everyday like "and", "but", "or" etc. in
Knowledge of date ,history, arithmetic ,numeric character, this list. Words are included in summary according to
location, titles words, positive and negative words phrase are determined by a percentage. Percentage is determined
show importance of a sentences. Yeh et al. [13], Hernandez by the genetic algorithm or heuristic methods.
and Ledeneva [14] and Quyang et al. [15] have used this Location: Paragraphs of sentences is determined like
feature in their studies. which sentence in the introductory paragraph or
Study of Altan [16] is the first study on Turkish texts. He conclusion paragraph.
has used for data sets which written 50 document in financial Key Words: Important words for the user entered
issues. Firstly, he has separated heading and paragraphs using previously. Sentences that contain these words are
by HTML tags. Then, he examined word frequency, sentences important. For example, rebound is important word
location, words in a title, positive and negative words phrase about a sport text.
techniques in his study. Uppercase: Special names are pure information. It
Kılcı et al. [17] have examined date words, key words should be included in the summary.
phrases, punctuation marks and proper nouns as addition to Positive Words: More points are given to sentence
working Altan. Weighting of the sentence has already been which contains some word phrase like "as a result”, "at
entered in their study. last" etc. in the text.
Pembe [18] has carried out query based summarization Negative Words: Sentences containing details are given
system for using web search. Then, she has given score to less points.
sentence and chapter. Collocation: Words that reinforce the meaning is
Distinguishes the work of alien Çığır et al. [19], scoring considered important.
was normalized. The position in the sentence of the text has Numbers: The score of sentences containing the
been examined in this study. It was also examined according numerical is high.
to central position of sentence.120 document is used as text Quotation Mark: Sentences including quote is important.
data set. Ending Mark: Punctuation mark at the end indicates the
We have explained structural feature of language up to now. importance of a sentence.
We have seen methods of finding some most important Average Length: The mean number of words in the text
sentences in a text. Then found sentences are rated. There are is calculated. Nearest to average sentences is considered
techniques combined these methods in the literature. important.
____________________________________________________________________________________________________________
112
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Day-Month: Sentences including knowledge of day and In a thesis published in 2014, Attokurov [43] has claimed
month are important. that the main purpose of text summarization systems is to
In this study firstly text is divided into sentences and prevent the repetition of information. Optimal Tree Pruning
paragraphs. Zemberek is used in the separation process. Then algorithm and HAC (Hierarchical Agglomerative Clustering)
allocated structures examined designated by 13 features. algorithm are used together to prevent the repeat of
Heuristic method has been used for scoring. Weighting of information.HAC was used for extraction to repetition.
sentences are shown in Table I. Optimal tree pruning algorithm has also been used to reduce
repetition.
TABLE I
To solve the problem of summarization have also been used
WEIGHTING OF FEATURE optimization algorithms. In a study [44], ant colony
optimization algorithms, particle Swarm optimization,
Features Weighting artificial bee colony algorithms and genetic algorithms are
Head 20 used to extract more effective summary.
High Frequency 10
Introduction 20 B. Abstractive Summarization Studies
Conclusion 2 To good determine the relationship of word-event and
Uppercase 3 grouping of terms in the event tag is very important to
Positive Words 15 emergence of the meaning in a text. Based on this idea,
Negative Words -20 success of summarization was attempted to increase with
Collocation 4 linear algebra methods. This method is briefly performed in
Numbers 3 the following;
Quotation Mark 2 Primarily text is divided into words and sentences. Then a
Ending Mark 2 matrix of term-sentence is obtained. Matrix is factorized.
Average Length 10 After that words and phrases are converted into linearly
Day-Month 5 independent vectors. Words and phrases are grouped as
Key Phrase 8 semantic by these vectors. Important words and phrases are
selected grouped with semantic matrix factorization. After this
For example, a sentence has two key words and three process, semantic analysis of texts is performed by methods of
numerical characters in text. Total score of the sentence is Latent semantic analysis or probabilistic semantic analysis.
( 2*8) + (3*3) = 25.Each sentence score is calculated. A rate Murray et al. [45], Steinberger et al. [46], Yeh et al. [13]
is asked by users. This is called summarization percentage rate. have used latent semantic analysis in their studies. Lee et al.
Sentences are select according to this rate. For example, if [47], Mashechkin et al. [48] and Güran et al. [49] have also
summarization rate select 35 percent, 35 percent of the used non-negative matrix factorization (NMF) algorithms.
sentences of highest scores would select. Then, have been Güran, in her study [50] has determined sequential words
tried at different rate. They have taken an average and with the help of Turkish Wikipedia links. In Güran's study
measured to success 55 percent in their study. [50], structural and semantic features are used together.
In another study [39], summary of Turkish web pages has A hybrid system that combines structural and semantic
been being studied. Sentence selection is determined by 12 feature has been proposed in order to determine important
different features. Heuristic method has been used for sentences. The results of hybrid systems have been more
weighting. Success rate was measured as %59 in this web- successful than each individual system. The new techniques
based study. used in stage of creating term-sentences matrix have been
In a study [40] that used a combination of text and sentence explained. These techniques are based on location the words
clustering techniques, similar words and documents are and importance of words in sentences to which they belong.
clustered. The highest frequency sentences have been selected Study has also produced to use two new data sets for next
in this cluster. studies on Turkish language. The first one is 130 news
Pointed out that in another study [41], the information in document, second corpora is 20 shorter text news.
the text is obtained in two stages. In the first stage, tagged Bhandari et al. [51] Chris, H. and Ding, Q. [52] have used
sentences are separated according to certain rules. Word probabilistic latent semantic analysis in these studies.
phrases, word frequency, and location methods were used In another study [53], Non-negative Matrix Factorization
with modal verbs in the second stage. This study was carried methods were used to determine the relationship between
out for English texts. It has benefited from the tense of the sentences. Clustering and clustering methods were used in text
sentence to filter the information. pre-processing. The study has also been highlighted to
Extractions summarization approach was used in a study increase quality of summarization of clustering methods.
carried out in 2014. Güran at al.[42] have argued that adding a We continue to examine studies that using method of
new criteria for Extractions summarization system. Name semantic analysis. Gong and Lui [54] have divided text to
entity recognition method was used for the first time. topic and they have chosen sentences from each topic. In
contrast, Murray et al. [45] has chosen multiple sentences
____________________________________________________________________________________________________________
113
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
from each topic. Özsoy et al. [55], have proposed to eliminate algorithms were used. Sentences are determined with the help
the noise. They have also chosen all topics in text. of similarities between sentences. Success was measured by
Özsoy et al. [55] have proposed two new methods in Rouge assessment package. Successful results were obtained.
studies of semantic analysis named as Cross and topics. Dependency grammars has been used some problems that
Topics method was used to separate main subjects from sub- finding similarities between sentences. The usage of
subjects and make sentence selections from main subjects. dependency grammars to compute sentence similarity has
Cross method was also used for observe whether sentences are been proposed in a study [62]. This study have been achieved
important. Cross-method has produced more successful results. that dependency grammar representations of sentences find
There is a method that shows the interconnectedness of out to better results in finding the similarities between
words named as word chains method in the literature. Berker sentences. Details can be reached on [62].
[56] has benefited relationship between words in her study. Most studies in the literature have been made for
Word chains were formed in this study. Then this deep summarization of English language. Studies for other
knowledge was combined with a higher level analysis. Finally, languages are less. Therefore, minority of studies for written
this information at different levels is combined with the help on Arabic language have been emphasized in a study [63]. It
of genetic algorithms. has also talked about the lack of Arabic language
In a study conducted in 2010 [57], a software has been summarization standards. It has been mentioned that Arabic
developed named as FRESA (Framework for Evaluating was included in TAC 2011 MultiLing Pilot and ACL 2013
Summaries Automatically).This software is content-based and MultiLing Workshop. Other methods of extraction
it make summary assessment. A system which generic single- summarization studied on Arabic in the literature has been
document summarization in French and Spanish and focus- presented and grouped on a table.
based multi-document summarization in English was
introduced in study. System performance is measured by IV. MEASURING SUCCESS
assessment package ROUGE. ROUGE Package will be In order to test the success must be we should have ideally
discussed later. summary. In study by Rath et al. [64], people have been
In another study called ARTEX (AnotheR TEXt requested to select sentences which could be important
summarizer) [58], document vectors and lexical vector are according to them from different at different times. Güran [50]
kept. In order to rank sentences, a simple inner product is has asked to select important sentences from some people.
calculated between each sentence, a document vector (text Thus Ideal summary was obtained. In another study [65]
topic) and a lexical vector. Highest ranked were combined. No Morris, Kasper and Adams have suggested a method named as
ruled-based linguistic post-processing and based in sentences GMAT (Graduate Management Admission Test). A technique
extraction on of answering questions which exercise of reading
Vector Space Model (VSM) techniques have used. Ranking comprehension is recommended.
were normalized. Each sentence is calculated in the feature Success is evaluated by various methods. To compare Ideal
matrix format. The most important aspect of this study is not summary with the summary created by the system is the most
requires linguistic knowledge or resources. Study is tested on known method. In order to measure success such as
French and Spanish document and it has produced successful information retrieval, text categorization and question
results. answering are used. The most common methods in the
In another project [59] that using latent semantic analysis literature are precision/recall/F-score, value of relative benefit,
techniques have presented it presented to comparison of cosine similarity and ROUGE.
different vector space model. Success was measured by
Ngram assessment methods. Study has proved binary models A. Value of Precision-Recall-F-Score(PRF)
more successful than complex term frequency models. The purpose of this method is to find the number of
Looking at statistical distribution of words as semantically, common sentences between ideal summary and measured
language processing problems can be solution. In a study [60] summary. The number of common sentences is divided to
carried out in this thought, firstly meaning the relationship number of measured sentences. This is called precision (P).
between words was measured. Meaning of integrity and key Recall value (R) is found by dividing number of common
phrase extraction using lexical chains has been used. This sentences to number of measured sentences. The harmonic
method requires more prior knowledge. In study, has been mean of P and R is F-score (F). Formulas are shown below.
investigated the effect of the use of lexical cohesion features
in key phrase extraction, with a supervised machine learning | ∩ | | ∩ | 2
= = =
algorithm. Studies have shown that lexical cohesion based | | | | +
features improve key phrase extraction . Achieved good
results, compared to some other lexical cohesion based B. Value of Relative Benefit
algorithms. This method has been proposed to correct deficiencies in
M. Cakir and E.Çelebi have presented a method that PRF method. In Dragomir's study [66], sentences are rated
includes all languages to summarize their works [61]. C3M heuristically. This is called utility score. Each sentence is
(Cover Coefficient-Based Clustering Methodology) rated. People are rated the sentences. "N" is number of people
____________________________________________________________________________________________________________
114
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
and n is number of sentences in the following formula. "i" is (EMNLP), pages 702-706, Doha, Qatar, October, 2014. Association for
Computational Linguistics.
shown i-th assessor and j is shown number of sentences."U" is
[9] http://people.dsv.su.se/~hercules/papers/FarsiSum.pdf
shown score of utility. Can be accessed for details of study [10] Gündoğdu Ö.E., Duru N. “Türkçe Metin Özetlemede Kullanılan
here [66]. Yöntemler", 2016 Akademik Bilişim.
∑ δ ∑ [11] Luhn, H.R., (1958). “The automatic creation of literature abstracts.”,
= IBM Journal of Research Development, 2(2):159–165.
2∑ ∈ j∑ [12] Edmundson, H.P., (1969). “New methods in automatic extracting.”,
Journal of the Association for Computing Machinery, 16(2): 264–285.
[13] Yeh, J.Y., Ke, H.R., Yang, W.P. and Meng, I.H., (2005). “Text
C. Cosine Similarity Summarization using a Trainable Summarizer and Latent Semantic
Formula of another measurement method is given Analysis”, Journal of Information Processing and Management, 41:75–
below.” ” is word frequency value at ideal summary.” y ” is 95, 2005.
[14] Hernandez, R.A.G and Ledeneva, Y., (2009). “Word Sequence Models
also word frequency value at measured summary. Can be for Single Text Summarization”, In Proceeding of Second International
accessed for details of study here [67]. confererence on Advances in Computer-Human Interactions, 1-7 Feb.
2009, Mexico.
∑ [15] Quyang, Y., Li, W., Lu, Q. and Zhang, R., (2010). “A study on
cos( , ) = Position Information in Document Summarization”, In Procceding of
∑( ) ∑ (y ) Coling 2010, Poster Volume, August 2010, Beijing.
[16] Altan, Z., (2004). "A Turkish Automatic Text Summarization System",
D. Ngram Statistics IASTED International Conference on AIA, 16-18 February 2004,
Innsbruck, Austria.
Rouge (Recall- Oriented Understudy for Gisting Evaluation) [17] Kılcı, Y. and Diri, B., (2008). Turkish Text Summarization System,
software which a Ngram based technique was proposed in Senior Project, Yıldız Technical University, Turkey.
2004 by Chin-Yew Lin [68].It is produced by the Perl [18] Pembe, C. (2011). Automated Query-Biased and Structure-Preserving
programming language and was used in DUC [69] (Document Document Summarization for Web Search Tasks, PhD Thesis,
Boğaziçi University, Turkey.
Understanding Conference) firstly. It is a method that based [19] Çığır, C., Kutlu, M. and Cicekli, I., (2009). “Generic Text
on comparison of the number of common words. There are Summarization for Turkish”, the Computer Journal, 53(8):1315-1323.
five different measurement methods. There are ROUGE-N, [20] Suanmali, L., Salim, N. and Binwahlan, M.S., (2009). “Fuzzy Logic
ROUGE-L, ROUGE-S, ROUGE-W, ROUGE-SU and Based Method for Improving Text Summarization”, International
Journal of Computer Science and Information Security, 2(1): 65-70.
ROUGE-N. Ngram is number of 5-length sequential word [21] S.A.Babar ,Pallavi D.Patil , "Improving Performance of Text
group. ROUGE-N is number of common Ngrams between Summarization" International Conference on Information and
ideal summary and measured summary. It is calculated by the Communication Technologies (ICICT 2014) GA–GP.”, In IEEE
following formula. Can be accessed for details here [68]. international conference on fuzzy systems, 16–21 July. Vancouver,
Canada.
∑ ∑ ( ) [22] Kyoomarsi, F., Khosravi, H., Eslami, E. and Dehkordy, P.K, (2009).
∈{ } ∈
“Optimizing Machine Learning Approach Based on Fuzzy Logic in
∑ ∈{ } ∑ ∈
( ) Text Summarization”, International Journal of Hybrid Information
Technology, 2(2): 105-116.
IV. CONCLUSIONS [23] Binwahlan, M.S., Salim N. and Suanmali, L., (2010). “Fuzzy Swarm
Diversity Hybrid Model for Text Summarization”, Information
In This review, we have presented definitions, methods, Processing and Management, 46(5):571-588.
[24] Silla C.N., Pappa, G.L., Freitas, A.A. and Celso A.A., (2004).
approaches and studies about of Automatic Text “Automatic Text Summarization with Genetic Algorithm-Based
Summarization. We have detected that studies was focused on Attribute Selection”, 9th Ibero-American Conference on AL, Lecture
English language. We have seen the success of Turkish Notes in Computer Science, 3315 (2004):305–314.
studies around 50%.We have concluded that combination of [25] Witte, R., Krestel, R. and Bergler, S., (2007). “Generating Update
Summaries for DUC 2007.” In the Document Understanding
techniques have been more successful. Workshop, 26-27 April 2007, Rochester, NewYork USA.
[26] Berker, M. and Güngör, T., (2012), “Using Genetic Algorithms with
REFERENCES Lexical Chains for Automatic Text Summarization”, 4th International
[1] Tülek M., “Türkçe İçin Metin Özetleme”, İTÜ Fen Bil. Ens., Yük. Lis. Conference on Agents and Artificial Intelligence, February 2012,
Tezi (2007). p.595-600, Vilamoura, Portugal.
[2] Kesgin, F., 2007. Türkçe metinler için konu belirleme sistemi, Yüksek [27] Filatova, E. and Hatzivassiloglou, V., (2004), “A formal model for
Lisans Tezi, İ.T.Ü. Fen Bilimleri Enstitüsü, İstanbul. information selection in multi-sentence text extraction”, 20th
[3] Altan Z., Orhan Z., “Anlam Belirsizliği İçeren Türkçe Sözcüklerin International Conference on Computational Linguistics, August 2004,
Hesaplamalı Dilbilim Uygulamalarıyla Belirginleştirmesi”, Geneva, Switzerland.
http://turkoloji.cu.edu.tr/DILBILIM/anlam_belirsizligi.pdf, (2015). [28] McDonald, R., (2007). “A study of global inference algorithms in
[4] Akın, A.A. and Akın, M.D., 2007. Zemberek, an open source NLP multi- document summarization”, 29th European Conference on IR
framework for Turkish languages. Research, 2-5 April 2007, Rome, Italy.
[5] Porter Stemmer, http://tartarus.org/martin/PorterStemmer/, 1 May 2010. [29] Alguliev, R.M., Aliguliyev, R.M., Hajirahimove, M.S. and Mehdiyev,
[6] WordNet Dictionary Page, http://wordnet.princeton.edu/, 1 Haziran C., (2011), “MCMR: Maximum coverage and minimum redundant text
2009. summarization model”, Expert Systems with Applications, 38(2011):
[7] Birant, Ç.C.,”Rule Based Text Summarization in Turkish”,Dokuz 14514-14522.
Eylül University,April 2015 [30] Svore, K., Vanderwende, L. and Burges, C., (2007). “Enhancing
[8] Nuzumlali, M.Y., Özgür, A., ‘’Analyzing Stemming Approaches for Single-Document Summarization by Combining RankNet and Third-
Turkish Multi-Document Summarization’’, Proceedings of the 2014 Party Sources”, In Proceedings of the Joint Conference on Empirical
Conference on Empirical Methods in Natural Language Processing Methods in Natural Language Processing and Computational Natural
Language Learning, 28–30 June 2007, Czech Republic.
____________________________________________________________________________________________________________
115
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[31] Wong, K.F., Wu, M. and Li, W., (2008). “Extractive Summarization [53] Dmitry Tsarev, Mikhail Petrovskiy and Igor Mashechkin, “Supervised
using supervised and Semi-Supervised Learning”, In Proceedings of and Unsupervised Text Classification via Generic Summarization,
the 22nd international Conference on Computational Linguistics, Moscow, Russia, 2013.
August 2008, Manchester. [54] Gong, Y. and Liu, Xin. (2001). “Generic Text Summarization Using
[32] Ayush Agrawal, Utsav Gupta Extraction based approach for text Relevance Measure and Latent Semantic Analysis”, In the proceeding
summarization using k-means clustering “,International Journal of of ACM SIGIR, September 2001, New Orleans, Louisiana, United
Scientific and Research Publications, Volume 4, Issue 11, November States.
2014 [55] Özsoy, M., Çiçekli, İ. and Alpaslan, F.N., (2010). Text Summarization
[33] Steinberger, J., Massimo, P. and Sanchez-Graillet, A., (2005). of Turkish Texts using Latent Semantic Analysis", In Proceedings of
“Improving the LSA based Summarization with Anaphora Resolution”, the 23rd International Conference on Computational Linguistics,
In Proc. of the conference on Human Language Technology and August 2010, Beijing, China.
Empirical Methods in Natural Language Processing, 2005, Vancouver. [56] Berker,M. “Using Genetik Algorithms with Lexical Chains For
[34] Orasan, C., (2007). “Pronominal Anaphora Resolution for Text Automatic Text Sumamrization, Graduate Program in Computer
Summarization”, In Proceedings of the Recent Advances in Natural Engineering,Bogazici University,2011.
Language Processing, 2007, Borovets, Bulgaria. [57] Saggion, H., Cunha, I., Manuel, J. And Moreno,T. Sanhuan,E. ,”
[35] D'Avanzo, E., Magnini, B. and Vallin, A., (2004)., “Keyphrase Multilingual Summarization Evaluation without Human Models”
Extraction for Summarization Purposes: The LAKE System at DUC- August 2010, Beijing,China.
2004”, In the Document Understanding Workshop ,Boston, USA. [58] Manuel, J. And Moreno, T., “Artex is AnotheR TEXt summarizer" CP.
[36] Liu, Maofu., Li, W., Wu, M. and Lu, Q., (2007). “Extractive 6128 succursale Centre-ville, Montréal, Québec, Canada.
Summarization Based on Event Term Clustering”, In Proceedings of [59] Padmakumar, A., Eswaran, D. "Extractive Text Summarization using
the ACL, June 2007, Prague. Latent Semantic Analysis Natural Language Processing (Fall 2014) -
[37] Filatova E. and Hatzivassiloglou, V., (2004). “Event-based Extractive Project".
summarization”, In Proceedings of ACL 2004 Workshop on [60] Gönenç, E.Lexical Cohesion Analysis for Topic Segmentation,
Summarization, July 2004, Barcelona, Spain. Sumamrization and Keyphrase Extraction, December 2012.
[38] Uzundere E., Dedja E., Diri B., Amasyali M. F., "Türkçe haber [61] Çakır M., Çelebi E., “Kapsama Katsayısı Tabanlı Kümeleme İle Belge
metinleri için otomatik özetleme ", Akıllı Sistemlerde Yenilikler ve Özetleme”, IEEE 19th Signal Processing and Communications
Uygulamaları Sempozyumu, (2008). Applications Conference (SIU 2011), 186-189, Berlin, (2011).
[39] M.Vecdi SAMi,Banu Diri ,”Web tabanlı otomatik özet çıkarma sistemi” [62] Bilgin, Ş.B.,"Multi-Document Sumamrization Suing Dependency
Akıllı Sistemlerde Yenilikler ve Uygulamaları Sempozyumu, 21-24 Grammars" Boğaziçi University, 2014.
Haziran 2010, Kayseri & Kapadokya,AYSU 2010 [63] Al-Saleh A.B., Menai, M.E.B., “Automatic Arabic text
[40] Dashpande, R.A., “Text Summarization using Clustering Technique”, summarization: a survey”, Springer Science, October, 2015.
International Journal of Engineering Trends and Technology (IJETT) - [64] Rath, G.J., Resnick A., and Savage, T.R. 1961. The formation of
Volume4 Issue8- August 2013. abstracts by the selection of sentences. American Documentation,
[41] Wadhvani, Rajesh. , Roy, D. “Text Summarization Using Tense 12(2), pp. 139-143. Reprinted in Mani, I., and Maybury, M., eds.,
Recognizing Identifier” (IJCSIS) International Journal of Computer Advances in Automatic Text Summarization, MIT Press, pp. 287-292.
Science and Information Security, Vol. 11, No. 6, June 2013. [65] Morris, Andrew H., George M. Kasper, and Dennis A. Adams. 'The
[42] Güran A., Arslan S. N., Kılıç E., Diri B., " Sentence selection methods Effects and Limitations of Automated Text Condensing on Reading
for text summarization", IEEE 22nd Signal Processing and Comprehension Performance', Information Systems Research, vol.
Communications Applications Conference (SIU 2014),(2014). 3/no. 1, (1992), pp. 17-35.
[43] Ulukbek Attokurov, “Multi Document Sumamrization Using [66] Dragomir, R., Hongyan J. and Malgorzata B., (2000). “Centroid based
Distortion-Rate Ratio”, M.Sc. Thesis, İstanbul Tecnical University, summarization of multiple documents”, In ANLP/NAACL Workshop
June 2004. on Automatic Summarization, Seattle, USA.
[44] Raj Kumar V.S, Chandrakala.D,”A survey on text summarization using [67] Salton, G., (1988). Automatic text processing: Automatic Text
oprimization algorithm”. ISSN: 2454-3047; ISSN: 2394-0441 (Online) Processing: The Transformation Analysis and Retrieval of Information
Volume 2 Issue 1 (2016). by Computer, Addison-Wesley Publishing Company.
[45] Murray, G., Renals, S. and Carletta, J., (2005). “Extractive [68] Lin, C.Y., (2004). “ROUGE: a Package for Automatic Evaluation of
summarization of meeting recordings”, In Proceedings of the 9th Summaries”, In Proceedings of the Workshop on Text Summarization
European Conference on Speech Communication and Technology, Branches Out, Barcelona, 25 – 26 July 2004, Spain.
September 2005, Lisbon, Portugal. [69] DUC Conference, http://duc.nist.gov/, 10 June 2012.
[46] Steinberger, J., (2007). Text Summarization within the LSA
Framework, PhD Thesis, University of West Bohemia in Pilsen, Czech
Republic, January 2007.
[47] Lee, J.H., Sun P., Chan-Min A. and Daeho K., (2009). “Automatic
generic document summarization based on non-negative matrix
factorization”, Inf. Process. Manage, 45(1): 20–34.
[48] Mashechkin, I.V., Petrovskiy, M.I., Popov, D. S. and Tsarev, D.V.,
(2011). “Automatic text summarization using latent semantic analysis”,
Programming and Computer Software, 2011: 299-305.
[49] Güran, A., Güler, N. and Bekar, E., (2011). Automatic summarization
of Turkish documents using non-negative matrix factorization ",
INISTA 2011, Istanbul, Türkiye.
[50] Güran,A. , "Otomatik Metin özetleme Sistemi", Yıldız Teknik
Üniversitesi Fen Bilimleri Enstitüsü ,Doktora Tezi 2013,
İstanbul,Türkiye.
[51] Bhandari, H., Shimbo, M., Ito, T., Matsumoto, Y. and Bhandari, Y.,
(2007). “Generic Text Summarization Using Probabilistic Latent
Semantic Indexing”, 3rd International Joint Conference on Natural
Language Processing.
[52] Chris, H. and Ding, Q., (2005)., “A probabilistic model for latent
semantic indexing”, In Journal of the American Society for
Information Science and Technology, 56(6): 597–608.
____________________________________________________________________________________________________________
116
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Merging systems, enhancing inter-disciplinary In this study, articles included in the literature in the last 15
relations and increasing needs require multi objectives rather years and referred more than other articles written in the same
than a single objective in the optimization problems nowadays. year are analysed in terms of subjects, methodologies and
However, the objectives are frequently conflicting. When an results for Multiobjective optimization. These concepts in the
objective is improved, the other objective(s) may deteriorate. In
literature are handled under separate headings. Multiobjective
the multi-objective optimization problems (MOOPs), the aim is
to come up with the best solutions that can be an alternative for optimization methods can be divided into two groups as
each other in terms of objective function values under the classical and heuristic methods. In this study, we explain the
constraints caused by various reasons. During the last two methods mostly used or referred in the articles. Each article is
decades, MOOPs and solution methods have been studied with analysed in terms of its methodology under the relevant
great interest. It is possible to come across a MOOP in almost heading. Other unexplained methods and new approaches
every discipline in the literature. MOOPs have been modelled related to multiobjective optimization are analysed under the
and solved not only in the fields with more applications such as same heading. Methods chosen in the articles are analysed
production, management, business administration, marketing, along with the reasons in the conclusion part. Success criteria
transportation and finance but also in the basic sciences such as
of the studies conducted in special areas are mentioned.
chemistry, maths and statistics. Solution of MOOPs requires the
simultaneous optimization of conflicting multi objectives. In Subjects are suggested for future studies to obtain more solid
MOOPs, an optimal solution set on which a compromise is results from multiobjective optimization.
reached among the conflicting objectives is obtained. In this
study, the articles on multi-objective optimization written in 2015 II. MULTIOBJECTIVE OPTIMIZATION
and later are analysed and 61 articles are chosen among them. The main difference between single objective problems and
Classical and heuristic methods implemented for the solution of multiobjective problems is that; there is a single optimal
MOOPs presented in these articles are mentioned. The articles solution or alternative optimal solutions in single objective
are classified according to their subject areas. The methodology
problems while there is not any solution simultaneously
used in each article is identified. According to their
implementation areas, the multi-objective optimization methods
delivering the best value for each objective in multiobjective
and the areas they are implemented the most are discussed. The problems. Since all objectives are important in these problems,
areas to be focused on in the future studies to obtain more robust looking for a solution only for one objective is wrong.
results in the optimization are identified. Improvements in an objective cause deterioration in at least
one of the other objectives due to conflicting objectives.
Keywords— multi-objective optimization, Pareto optimal set, Tradeoff amounts are obtained among identified solutions
genetic algorithm methods, particle swarm optimization depending on their function values because of conflicting
objectives. Acceptable tradeoff amounts depend on decision
I. INTRODUCTION makers. In this case, there is generally not a certain optimal
Optimization can be described as obtaining the best solution in multiobjective problems. Some of the solutions
solution in line with the constraints identified for objective(s) obtained and tradedoff are presented to the decision maker as
desired to be reached. First, decision variable is identified and the best solutions.
then, objective function of the problem is formed in the A general multiobjective optimization problem includes n
solution of an optimization problem. Constraint function of parameters (decision variable), k objective functions and a
the problem, if available, is described and the appropriate group of m constraints. Objective functions and constraints
solution method to be chosen is applied according to are written as the function of decision variables. Optimization
optimization classification. Although the majority of studies process can be formulated as
on optimization focus on a single objective, optimization min = ( )= ( ), ( ), … , ( )
problems we come across in practice are based on multi ( )= ( ), ( ), … , ( ) ≤ 0
objective optimization.
____________________________________________________________________________________________________________
117
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
118
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
slow in classical methods. Searching for the optimal solutions objectives. It is know that using genetic and evolutionary
for these is only possible with heuristic methods [7]. algorithms for the solution of multiobjective optimization
Genetic algorithm (GA) is the application of genetic related problems is advantageous.
mechanisms in optimization problems as an algorithm.
Reproduction in genetic generally occurs through mutation A. Multi-Objective Genetic Algorithm
and crossover. These are GA operators. The obtained Multiobjective Genetic Algorithm (MOGA) was
individual (solution) should have a good fitness value to developed by Fonseca and Fleming [13]. Groups are formed
maintain its participation. In this way, strong individuals starting with the nondominated solutions in the first group
succeed in survival and can transfer their genes to the next and solutions dominated for once in the second group. In each
generations. Therefore, better solutions can be achieved by group, only the proximity of solutions to each other in the
transferring the information on the solutions with good results same group is calculated. Following the calculation of
to the next solutions [8]. proximities, fitness parameter ( ) which should be a
To be able to use GA in a computer, the most compatible small number depending on the problem is identified. It is
coding method for the problems should be used. The mostly easy to calculate fitness values of the solutions in MOGA
used methods are binary coding, permutation coding and real method. Furthermore, MOGA can be applied in various
number coding. optimization problems with fitness parameter . However,
Selection of the solutions to be used in the next generation in this technique, giving the same fitness value to different
is one of the important steps in GA operation. Various solutions on the same front may cause the algorithm to look
selection methods were suggested as a result of research for the solutions on the wrong front. In practice, it has been
studies. Selection methods commonly used in practice include proved that MOGA is quite sensitive to the type of concerned
roulette wheel [9], selection sort [10] and tournament solution area. Besides, the algorithm does not guarantee that a
selection [11]. The common feature of these methods is that solution has always a better fitness value than a worse solution
solutions obtained are probabilistically selected according to [14].
their compatibility value. Probability of the solutions with Considering the studies using MOGA, Wang et al [25]
good compatibility value to be included in the next generation optimized the conflicting economic and environmental
is always high. objectives in environment friendly building designs with
Crossover and mutation are used to obtain new solutions MOGA. They presented a case study and demonstrated the
using the selected solutions. Crossover is more commonly effectiveness of their approach for identifying a number of
used compared to mutation operator. There are onepoint, Pareto optimal solutions for green building design. Tahara et
twopoint and multipoint crossover techniques. It is observed al [30] optimized various objective values in ship design with
that onepoint and twopoint crossover techniques are used the Sequential quadratic programming (SQP) and MOGA. They
most. Mutation is an operator applied on solutions obtained formulated a multiobjective optimization problem where
after crossover operation. It is applied on the solutions with a ship propulsion and maneuverability performances are
small mutation probability identified beforehand and various considered. Pierobon et al [72] realized Rankine cycles
solutions are obtained in this operation. Mutation operator optimization for waste heat recovery using MOGA. They
provides diversity in the formation of next generations. It is observed in the results that they could prevent heat loss.
possible to obtain bad results as well as good ones with
crossover and mutation; however, these solutions are B. Non-dominated Sorting Genetic Algorithm
eliminated during selection. In GA, population extent should Nondominated Sorting Genetic Algorithm (NSGA) was
also be determined in addition to crossover and mutation developed by Srivas and Deb [15]. The method starts with
control parameters. Most of the problems are quite sensitive to sorting the solutions obtained randomly according to their
these three parameters and the best solution is obtained only dominance. As in MOGA method, proximity among the
with the right selection of parameters [6] . solutions in each subpopulation is calculated. Solutions are
Considering the studies using GA, Rao and Patel [66] selected with roulette wheel method. In this method, the
presented a new approach for heat exchangers optimization chance of the solutions in the nondominated 1st sub
using modified teachinglearningbased optimization population ( ) to be selected is higher. New solutions are
algorithm. They compared the results obtained with GA obtained by applying crossover and mutation on the selected
results. Omkar et al [50] compared the results of the system solutions and the algorithm keeps looking for solutions until
designed for Vector Evaluated Artificial Bee Colony (VEABC) stopping criterion is formed. NSGA method classifies
and its multiobjective optimization with PSO, Artificial solutions according to their dominance and assigns fitness
Immune System (AIS) and GA. values. Various practices have shown that the algorithm is
especially sensitive to its parameter [15].
V. MULTIOBJECTIVE GENETIC ALGORITHMS
Considering the studies using NSGA, Ahmadi et al [70]
There may not be a single best solution for all objectives in aimed to obtain maximized power, thermal efficiency and
multiobjective cases. In this case, the decision maker is asked minimized pressure loss using NSGA in Stirling heat engine
to select a solution from a compromised finite set. Appropriate design. They observed the results using experimental data.
solution should perform at an acceptable level for all
____________________________________________________________________________________________________________
119
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
C. Elitist Non-dominated Sorting Genetic Algorithm from NSGAII and MOEA/D with the results obtained from
Elitist Nondominated Sorting Genetic Algorithm (NSGA two classical methods. Ahmadi et al [78] presented a new
II) was developed by Deb and Goel [18],[19]. This algorithm approach by using NSGAII in order to optimize each stage of
is similar to NSGA method; however, parameters used in biomass energy. They showed the Pareto front and observed
the results with statistical analyses. Song et al [80] suggested a
NSGA are not used in NSGA II. NSGA II does not
semiactive battery/supercapacitor (SC) hybrid energy storage
need the fitness parameter. The algorithm gives more chance
system (HESS) for electric vehicles. They realized parameter
to the prominent results and it is almost impossible to lose
optimization using NSGAII in their studies.
these solutions. Therefore, the algorithm never loses the
Pareto optimal solution found until the current step. Solution D. Strength Pareto Evolutionary Algorithm
selection mechanism is used to limit the population size;
however, in this case, the algorithm may lose its feature of Strength Pareto Evolutionary Algorithm (SPEA) was
proximity to the optimal solution. Since the number of developed by Zitzler and Thiele [20]. In this algorithm,
solutions in the subset where first nondominated solutions diversity in the selection of Pareto optimal solutions is
exist is not higher than the number of main population, all ensured with clustering analysis. Calculation of proximities is
solutions in this set are selected. Therefore, diversity among easy and additional parameters are not required. In the
the selected solutions is not ensured. algorithm, parameter , which is the size of populations
In the studies using NSGAII, Sarkar and Modak [23] where prominent solutions will be tracked, should be
developed a solution method for fedbatch bioreactors with identified. Furthermore, balance between the main population
NSGAII. They applied this method on two problems included size and should be ensured in order to obtain good results
in the literature and solved with many methods previously. from the algorithm. should be neither too big nor too small.
Atashkari et al [26] found out Pareto fronts of conflicting If it is too big, the algorithm loses much time with prominent
objective functions in the thermodynamic cycle of turbojet solutions and may not generate other solutions and converge
engines using εelimination and NSGA II. Further, they to optimal solutions. If it is too small, solutions in the
observed some interesting and important relationships among prominent set are not used enough and it may cause the
optimal objective functions and decision variables involved in algorithm to excessively look for the solutions outside the area
the thermodynamic cycle of turbojet engines. Deb et al [31] where optimal solutions exist. A criticised feature of the
developed a new robustness procedure in order to obtain more algorithm is that it gives a fitness value for dominance
realistic results instead of a general pareto front. They used criterion to the solutions in the main population during the
NSGA II to compare global and robust pareto fronts. They formation of populations. Furthermore, the power values
developed many limited and unlimited test problems to show obtained sometimes cannot keep the nondominated solutions
the obtained results through simulation. Goel et al [32] of the same importance at the same value [2].
realized approximations on the Pareto Front obtained with Considering the studies using SPEA, Farmani et al [24]
NSGA II. They showed that the approximated Pareto optimal focused on certain objectives such as capital, operation, life
front (POF) can help visualize balances and quantify values cycle, maintenance costs, system reliability and water quality
among objectives to select compromise designs. Mandal et al which should be simultaneously optimized in the design of
[34] obtained Pareto set by modelling electrical discharge water systems. Two examples related to design of water
machining (EDM) with NSGA II. They carried out systems were applied on each of MOGA, NSGA and SPEA
experiments over a wide range of machining conditions for algorithms and their Pareto fronts were compared. Ali et al
training and verification of the model. Testing results [57] amended Differential Evolution (DE) algorithm and
demonstrated that the model is suitable for predicting the transformed it into MultiObjective Differential Evolution
response parameters. Jia et al [40] worked on the optimization Algorithm (MODEA). They introduced new approaches to
of energy, cost and scope parameters in the use of wireless mutation and selection mechanisms in the meanwhile. The
network with NSGA II. Numerical and simulation results new approach obtained was applied on different multi
validate that the procedure to find the optimal balance point objective problems and the results were compared with
among the maximum coverage rate, the least energy algorithms NSGAII, SPEA and Pareto Archived
consumption, as well as the minimum number of active nodes Evolutionary Strategy (PAES) developed by Knowles ve
is fast and effective. Shokri et al [67] suggested a new method Corne [21].
by combining NSGAII and artificial neural networks (ANN)
in order to reduce time spent for the solution of multi
E. Strength Pareto Evolutionary Algorithm 2
objective optimization problems with evolutionary algorithms Strength Pareto Evolutionary Algorithm 2 (SPEA2) was
(EAs). The suggested method was applied on three standard developed by Zitzler et al. [22]. It is the developed version of
problems and one reallife problem. They were able to SPEA method. SPEA2 introduces a better scoring mechanism,
considerably reduce the time required to find Pareto optimal an intensity estimation technique and a developed archive
front compared to NSGAII solutions used without ANN. (secondary community) management compared to SPEA
CampomanesÁlvarez et al [74] realized surface method. SPEA2 uses finegrained fitness assignment strategy
simplification optimization for a 3D open model mesh using intensity information. Furthermore, size of the archive
simplification problem. They compared the results obtained externally storing the nondominated individuals is stable. If
____________________________________________________________________________________________________________
120
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the number of nondominated individuals is lower than the flexible jobshop scheduling problem using particle swarm
previously identified archive size, the archive is filled with optimization and local search. They compared the efficiency
nondominated individuals. Additionally; the clustering of the approach presented with other algorithms in the
technique used when nondominated surface is over the literature. Qasem and Shamsuddin [51] applied TVMOPSO
archive size was replaced with an alternative downsizing on radial basis function (RBF) used in the diagnosis of
method that has similar features but does not lose front points. medical disorders. They compared the results with MOPSO
Finally, another difference from SPEA is that only the archive and NSGA II. Yildiz and Solanki [59] presented a new
members are used in selection process. method for the multiobjective optimization of
Considering the studies using SPEAII, Durillo and Nebro crashworthiness of vehicles. The PSObased method was
[54] designed a developed optimization practice with NSGA applied on two optimization problems in the literature.
II and SPEA2. In this way, they realized metaheuristic and KhaliliDamghani et al [64] suggested dynamic selfadaptive
experimental practices in the solution of multiobjective multiobjective particles warm optimization (DSAMOPSO)
optimization applications. Xue et al [75] realized a feature for the solution of multiobjective reliability redundancy
selection optimization using PSO with the aim of reducing allocation problems (MORAPs). They compared the results
surplus and unnecessary features in the data sets and obtain obtained from test problems with TVMOPSO and NSGAII
more accurate results in a shorter time in classification. They results. Garg and Sharma [65] reformulated multiobjective
compared the results obtained with NSGAII, SPEAII and reliabilityredundancy allocation problem with fuzzy multi
PAES. objective optimization problem (FMOOP). They solved the
fuzzy MOOP obtained using PSO. The approach has been
VI. PARTICLE SWARM OPTIMIZATION demonstrated through the case study of a pharmaceutical plant
Particle Swarm Optimization is a swarmbased algorithm situated in the northern part of India. Zhang et al [68] realized
developed by Kennedy and Eberhart in 1995 with the an uncertain orbit optimization using risk and orbit distance
inspiration from fish schools and bird flocks in the nature [12]. parameters in the problems related to orbit planning for robots.
The algorithm was built on the behaviours of flock animals In their studies, they used PSO and observed the results from
practices in order to meet essential needs. Particle concept simulation. Taormina and Chau [82] created a multiobjective
used in PSO algorithm represents each individual in the problem using crossvalidation for PSOtrained neural
swarm. Each individual in the swarm has position information network river forecasting (NNRF). They solved this problem
indicating the position of the individual in dsize solution with multi objective fully informed particle swarm (MOFIPS)
space and speed information indicating its movement in dsize approach using actual data.
solution space. PSO is a repetitive algorithm. Therefore, speed
and position information of the particle is updated during each VII. OTHER METHODS AND NEW APPROACHES
repetition. While speed information of the particle is updated, New approaches presented for the solution of MOOPs and
not only the heuristic speed information from the previous methods developed for special systems are handled.
step but also cognitive and social experiences are used. For Mahapatra and Roy [27] developed a new fuzzy multi
the updated position information of the particle, position objective optimization method in order to increase system
information and updated speed information of the particle reliability. The main aim was to increase reliability while
from the previous step are used. [12, 16]. reducing the cost. Huang et al [28] developed a special
Considering the studies using PSO; Wang and Singh [35] interactive fuzzy multiobjective optimization model for
suggested fuzzified multiobjective particle swarm engineering designs. In this model, balance matrix was used
optimization (FMOPSO) for the compromising economic and and weight coefficients of objective functions were identified
environmental objectives in electricity transmission. The depending on preference. Igel et al [33] transformed
performance of the suggested approach was compared with covariance matrix adaptation evolution strategy (CMAES)
Weighted Clustering (WA) and evolutionary multiobjective delivering quite good results in singleobjective optimization
optimization algorithms. Tripathi et al [36] developed Time for multiobjective (MOCMAES) and elitists problems.
Variant MultiObjective Particle Swarm Optimization (TV Obtained results were compared with NSGA II. Jaeggi et al
MOPSO) procedure based on PSO multi objective [37] adapted Tabu Search algorithm for the solution of
optimization approaches and measured the performance by MOOPs. All results were compared with MOTS (Multi
comparing with algorithms such as NSGA II and PESA II. Objective Tabu Search), PRMOTS (Path Relinking Multi
Zhang and Liu [39] presented a new formulation with fuzzy Objective Tabu Search) and NSGA II. Thiele et al [41]
adaptive PSO (FAPSO) for multiobjective reactive power elaborated the obtained pareto optimal set with the
and voltage control. The proposed approach has been instantaneous preferences of the user in each repetition using
examined and tested with promising numerical results of the multiobjective evolutionary algorithm (MOEA). In this way,
IEEE 30bus and IEEE 118bus power systems. Zhang and they managed to include ignored but essential solutions in the
Xing [47] developed a construction method with PSO for the pareto optimal set. Kim et al [42] designed multiobjective
time–cost–quality tradeoff problem. They applied the populationbased incremental learning (MOPBIL) for the
suggested method with computerized analyses and verified. fuzzy path planning of robots. Simulation and experiment
Moslehi and Mahnam [49] presented a new approach for results show the effectiveness of the proposed MOPBIL from
____________________________________________________________________________________________________________
121
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the viewpoint of the proximity to the Paretooptimal set, size be more precise, the more the problem is complicated and
of the dominated space, coverage of two sets and diversity unknown, the more the probability of selecting heuristic
metric. Chaudhuri and Deb [43] developed a procedure to methods increases. Classical methods are chosen when
assist users to select a specific solution on the obtained pareto complete and certain results are desired to be obtained. In the
optimal set with multiobjective evolutionary optimization operations where solution performance is low due to the
(EMO). Huang et al [52] developed a new online multi complexity of the problem, pareto optimal set which can
objective optimization algorithm for membrane computing. satisfy the operator and show the approximate results is
They prevented the formation of undesired and wrong obtained. In the articles, finding the algorithms that can
solutions with online controls in this dynamic algorithm. Yang present the pareto optimal set most accurately and in the
[55] developed the multiobjective bat algorithm (MOBA). He shortest notice becomes prominent as the main aim.
applied the algorithm on the problems related to welded beam Another outstanding subject in the articles is NSGAII
design and observed its success in the results of simulation. which is one of the multiobjective genetic algorithm types
Fesanghary et al [58] presented a model based on harmony among heuristic methods. It was observed this algorithm
search algorithm (HS) in order to minimize life cycle cost delivers the most diversity in the solution of problems. This
(LCC) and carbon dioxide equivalent (CO2eq) and maximize algorithm either serves as the main method of many
energy efficiency of the buildings. To measure the efficiency operations or the developers use a variant of this algorithm.
of the proposed approach, they tested the performance of the The abundance of NSGAII use in the measurement of special
model on a typical single family house. Jiménez et al [69] algorithms developed is outstanding. In the studies using PSO,
conducted an optimization study on production planning using the most remarkable feature of PSO is its easy practice in
fuzzy multiobjective evolutionary algorithm. They compared most of the systems. Developers have experimented many
the results with NSGA II. Sindhya et al [71] attempted to variants of PSO on the same problem and compared the
eliminate the problems arising in optimization process with results.
multiobjective evolutionary algorithms (MOEA) using new Considering the studies in the special areas, it is observed
hybrid systems. NSGAII and MOEA types were transformed that multiobjective optimization methods in the literature are
into hybrid systems and the results were compared. Niu et al either used in the same way or their variants are applied. The
[73] developed a new multiobjective Bacterial Foraging success of algorithms developed in the studies are measured
Optimization (MBFO) algorithm and compared the obtained with the comparison of results from many methods applied on
results with NSGAII and MOPSO. Wang et al [76] presented the same problem.
an approach for costenergy optimization in the coalfired One of the biggest problems arisen during the studies is the
thermal power plants. In their approach, they used multi ignorance of essential solutions for users while finding the
objective differential evolution (MODE) and showed the pareto optimal front. All these observations bring this question
results with simulation. Huang [77] developed a design to the mind: Is it possible to develop a multiobjective
optimization for design exploration of threedimensional optimization method that can be applied on optimization
transverse jet in a supersonic crossflow and observed the problems in each area and provide the maximum performance
results with simulation. Wang et al [81] presented a multi in the solution? For the answer of this question, it is observed
objective optimization for combined cooling, heating and that studies on the performance of multiobjective
power system (CCHP). In the experiments conducted, it was optimization among the articles are quite important. The
found out that a more comprehensive solution set is required outstanding point in these studies is the interaction between
to obtain more successful results from optimization. application and user. If the future studies focus on increasing
the performance and robustness of multiobjective
VIII. CONCLUSIONS optimization, methods delivering results where most
In this study, 61 articles published in the last 15 years are approximate values to the solution are found in the shortest
selected and analysed. Each article is handled under separate notice and essential solutions for the operators are not ignored
headings in terms of their subjects, methodology and results. can be developed.
As a results of the analysis, it is observed that operations using
multiobjective optimization techniques are quite successful in REFERENCES
the solution of comprehensive problems which is a huge [1] Zitzler, E. 1999. Evolutionary Algorithms for Multiobjective
Optimization: Methods and Applications. Doktora Tezi. İsviçre Federal
challenge for operators in the special areas.
Zürih Teknik Bilimler Enstitüsü, Zürih(4Zitzler 1999)
The method used in the solution of MOOPs should be [2] Deb K., 2001, Multiobjective Optimization Using Evolutionary
chosen considering the objectives and constraints peculiar to Algorithms, Wiley & Sons, England. (15)
each area. Because classical and heuristic methods do not [3] Zitzler, E., Deb, K and Thiele, L., 2000, Comparison of Multiobjective
Evolutionary Algorithms: Empirical Results. Evolutionary
deliver the same results in the same operation. While
Computation, 8(2),pp.173195. (17)
determining the classical and heuristic methods to be used in [4] Mietttinen K., 1999, Nonlinear Multiobjective Optimization, Kluwer,
article analyses, the extent of information available on the Boston.(18)
handled problem is quite effective. If the importance of each [5] Haimes, Y.Y., Lasdon L. S., Wismer, D. A., 1971, On a bicriterion
formulation of the problems of integrated system identification and
objective and constraint is know in numbers for the problem
system optimization. IEEE Transactions on Systems, Man, and
aimed to be solved, mostly classical methods are preferred. To Cybernetics 1 (3), 296297.(19)
____________________________________________________________________________________________________________
122
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[6] Reeves, C.R., Modern Heuristic Techniques for Combinatorial [29] Altiparmak, F., Gen, M., Lin, L. & Paksoy, T. 2006, "A genetic
Problems, John Wiley & Sons, Inc., New York, NY, 1993.(13) algorithm approach for multiobjective optimization of supply chain
[7] Jaszkiewicz, A., 1998, How to Solve It : Modern Heuristics, Springer, networks", Computers and Industrial Engineering, vol. 51, no. 1, pp.
New York.(14) 196215.
[8] Deb,K., 1999, MultiObjective Genetic Algorithms: Problem [30] Tahara, Y., Tohyama, S. & Katsui, T. 2006, "CFDbased multi
Difficulties and Construction of Test Problems, Evolutionary objective optimization method for ship design", International Journal
Computation, 7(3), pp.205230.(122) for Numerical Methods in Fluids, vol. 52, no. 5, pp. 499527.
[9] Goldberg,D.E.,1989, Genetic Algorithms for Search, Optimizations [31] Deb, K. & Gupta, H. 2006, "Introducing robustness in multiobjective
and Machine Learning, Addison Wesley, Reading, MA.(117) optimization", Evolutionary computation, vol. 14, no. 4, pp. 463494.
[10] Baker, J.E.,1985,Adaptive Selection Methods for Genetic Algorithms, [32] Goel, T., Vaidyanathan, R., Haftka, R.T., Shyy, W., Queipo, N.V. &
In Proceedings of an International Conference of Genetic Algorithms Tucker, K. 2007, "Response surface approximation of Pareto optimal
and Their Applications,pp.101111.(123) front in multiobjective optimization", Computer Methods in Applied
[11] Joines J., Houck C., 1994,On the use of nonstaionary penalty Mechanics and Engineering, vol. 196, no. 46, pp. 879893.
functions to solve nonlinear constrained optimization problems with [33] Igel, C., Hansen, N. & Roth, S. 2007, "Covariance matrix adaptation
GAs, the Proceedings of the First IEEE Conference on Evolutionary for multiobjective optimization", Evolutionary computation, vol. 15,
Computation, pp. 98108.(124) no. 1, pp. 128.
[12] Kennedy, J. and Eberhart, R. 1995. Particle Swarm Optimization, [34] Mandal, D., Pal, S.K. & Saha, P. 2007, "Modeling of electrical
Proceedings of IEEE International Conference on Neural Networks IV, discharge machining process using back propagation neural network
4(1), 1942–1948 and multiobjective optimization using nondominating sorting genetic
[13] Fonseca, C.M., Fleming, P.J., 1993, Genetic Algorithms for algorithmII", Journal of Materials Processing Technology, vol. 186,
Multiobjective Optimization: Formulation, discussion and no. 13, pp. 154162.
generalization, In Proceedings of the Fifth International Conference on [35] Wang, L. & Singh, C. 2007, "Environmental/economic power dispatch
Genetics Algorithms, pp. 416423.(126) using a fuzzified multiobjective particle swarm optimization
[14] Deb, K., Goldberg, D.E., 1989, An Investigation Niche and Species algorithm", Electric Power Systems Research, vol. 77, no. 12, pp.
Formation in Genetic Function Optimization, In Proceedings of the 16541664.
Third International Conference on Genetic Algorithms, pp. 4250.(128) [36] Tripathi, P.K., Bandyopadhyay, S. & Pal, S.K. 2007, "MultiObjective
[15] Srinivas,N., Deb, K.,1994, Multiobjective Optimization Using Particle Swarm Optimization with time variant inertia and acceleration
Nondominated Sorting in Genetic Algorithms, Evolutionary coefficients", Information Sciences, vol. 177, no. 22, pp. 50335049.
Computation, 2(3),pp.221248.(129) [37] Jaeggi, D.M., Parks, G.T., Kipouros, T. & Clarkson, P.J. 2008, "The
[16] Altınöz, Ö. T. and Yılmaz, A. E. 2009. Parçacık sürüsü development of a multiobjective Tabu Search algorithm for
Optimizasyonunda Yeni Bir Birey DavranıĢ Biçimi Önerisi, EMO 13. continuous optimisation problems", European Journal of Operational
Ulusal Kongre Bildiriler Kitabı, 2326 Aralık 2009, Ankara Research, vol. 185, no. 3, pp. 11921212.
[17] Konak, A., Coit, D.W, Smith, E.A., 2006, Multiobjective [38] Diakaki, C., Grigoroudis, E. & Kolokotsa, D. 2008, "Towards a multi
Optimization using Genetic Algorithms: A tutorial, Reliability objective optimization approach for improving energy efficiency in
Engineering & System Safety, vol. 91, pp. 9921007. (131) buildings", Energy and Buildings, vol. 40, no. 9, pp. 17471754.
[18] Deb, K., Goel, T., 2001, Controlled Elitlist Nondominated Sorting [39] Zhang, W. & Liu, Y. 2008, "Multiobjective reactive power and
Genetic Algorithms for Better Convergence, In Proceedings of the First voltage control based on fuzzy optimization strategy and fuzzy
Conference on Evolutionary MultiCriterion Optimization (EMO adaptive particle swarm", International Journal of Electrical Power and
2001), pp. 6181. (132) Energy Systems, vol. 30, no. 9, pp. 525532.
[19] Deb, K., Goel, T., 2001, A Hybrid Multiobjective Evolutionary [40] Jia, J., Chen, J., Chang, G., Wen, Y. & Song, J. 2009, "Multiobjective
Approach to Engineering Shape Design, In Proceedings of the First optimization for coverage control in wireless sensor network with
Conference on Evolutionary MultiCriterion Optimization (EMO adjustable sensing radius", Computers and Mathematics with
2001), pp. 385399. (133) Applications, vol. 57, no. 1112, pp. 17671775.
[20] Zitzler, E.,Thiele,L.,1999, Multiobjective Evolutionary Algorithms: A [41] Thiele, L., Miettinen, K., Korhonen, P.J. & Molina, J. 2009, "A
Comparative Case Study and the Strength Pareto Approach, IEEE preferencebased evolutionary algorithm for multiobjective
Transactions on Evolutionary Computation, 3(4), pp.257271. (134) optimization", Evolutionary computation, vol. 17, no. 3, pp. 411436.
[21] Knowles, J.D., Corne, D.W., 2000, Approximating the Nondominated [42] Kim, J.., Kim, Y.., Choi, S.. & Park, I.. 2009, "Evolutionary multi
Front using the Pareto Archived Evolutionary Strategy, Evolutionary objective optimization in robot soccer system for education", IEEE
Computation Journal 8(2), pp. 149172. (136) Computational Intelligence Magazine, vol. 4, no. 1, pp. 3141.
[22] Zitzler, E., Laumanns, M., Thiele, L., 2001. SPEA2: Improving the [43] Chaudhuri, S. & Deb, K. 2010, "An interactive evolutionary multi
Strength Pareto Evolutionary Algorithm. TIKReport 103, Swiss objective optimization and decision making procedure", Applied Soft
Federal Institute of Technology, Zurich, Switzerland, 121.(4) Computing Journal, vol. 10, no. 2, pp. 496511.
[23] Sarkar, D. & Modak, J.M. 2005, "Paretooptimal solutions for multi [44] Xiang, Y., Arora, J.S., Rahmatalla, S., Marler, T., Bhatt, R. & Abdel
objective optimization of fedbatch bioreactors using nondominated Malek, K. 2010, "Human lifting simulation using a multiobjective
sorting genetic algorithm", Chemical Engineering Science, vol. 60, no. optimization approach", Multibody System Dynamics, vol. 23, no. 4,
2, pp. 481492. pp. 431451.
[24] Farmani, R., Savic, D.A. & Walters, G.A. 2005, "Evolutionary multi [45] NarimanZadeh, N., Salehpour, M., Jamali, A. & Haghgoo, E. 2010,
objective optimization in water distribution network design", "Pareto optimization of a fivedegree of freedom vehicle vibration
Engineering Optimization, vol. 37, no. 2, pp. 167183. model using a multiobjective uniformdiversity genetic algorithm
[25] Wang, W., Zmeureanu, R. & Rivard, H. 2005, "Applying multi (MUGA)", Engineering Applications of Artificial Intelligence, vol. 23,
objective genetic algorithms in green building design optimization", no. 4, pp. 543551.
Building and Environment, vol. 40, no. 11, pp. 15121525. [46] Marler, R.T. & Arora, J.S. 2010, "The weighted sum method for multi
[26] Atashkari, K., NarimanZadeh, N., Pilechi, A., Jamali, A. & Yao, X. objective optimization: New insights", Structural and Multidisciplinary
2005, "Thermodynamic Pareto optimization of turbojet engines using Optimization, vol. 41, no. 6, pp. 853862.
multiobjective genetic algorithms", International Journal of Thermal [47] Zhang, H. & Xing, F. 2010, "Fuzzymultiobjective particle swarm
Sciences, vol. 44, no. 11, pp. 10611071. optimization for timecostquality tradeoff in construction",
[27] Mahapatra, G.S. & Roy, T.K. 2006, "Fuzzy multiobjective Automation in Construction, vol. 19, no. 8, pp. 10671075.
mathematical programming on reliability optimization model", Applied [48] Ren, H., Zhou, W., Nakagami, K., Gao, W. & Wu, Q. 2010, "Multi
Mathematics and Computation, vol. 174, no. 1, pp. 643659. objective optimization for the operation of distributed energy systems
[28] Huang, H.., Gu, Y.. & Du, X. 2006, "An interactive fuzzy multi considering economic and environmental aspects", Applied Energy,
objective optimization method for engineering design", Engineering vol. 87, no. 12, pp. 36423651.
Applications of Artificial Intelligence, vol. 19, no. 5, pp. 451460.
____________________________________________________________________________________________________________
123
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[49] Moslehi, G. & Mahnam, M. 2011, "A Pareto approach to multi [68] Zhang, Y., Gong, D.. & Zhang, J.. 2013, "Robot path planning in
objective flexible jobshop scheduling problem using particle swarm uncertain environment using multiobjective particle swarm
optimization and local search", International Journal of Production optimization", Neurocomputing, vol. 103, pp. 172185.
Economics, vol. 129, no. 1, pp. 1422. [69] Jiménez, F., Sánchez, G. & Vasant, P. 2013, "A multiobjective
[50] Omkar, S.N., Senthilnath, J., Khandelwal, R., Narayana Naik, G. & evolutionary approach for fuzzy optimization in production planning",
Gopalakrishnan, S. 2011, "Artificial Bee Colony (ABC) for multi Journal of Intelligent and Fuzzy Systems, vol. 25, no. 2, pp. 441455.
objective design optimization of composite structures", Applied Soft [70] Ahmadi, M.H., Hosseinzade, H., Sayyaadi, H., Mohammadi, A.H. &
Computing Journal, vol. 11, no. 1, pp. 489499. Kimiaghalam, F. 2013, "Application of the multiobjective
[51] Qasem, S.N. & Shamsuddin, S.M. 2011, "Radial basis function optimization method for designing a powered Stirling heat engine:
network based on time variant multiobjective particle swarm Design with maximized power, thermal efficiency and minimized
optimization for medical diseases diagnosis", Applied Soft Computing pressure loss", Renewable Energy, vol. 60, pp. 313322.
Journal, vol. 11, no. 1, pp. 14271438. [71] Sindhya, K., Miettinen, K. & Deb, K. 2013, "A hybrid framework for
[52] Huang, L., Suh, I.H. & Abraham, A. 2011, "Dynamic multiobjective evolutionary multiobjective optimization", IEEE Transactions on
optimization based on membrane computing for control of time Evolutionary Computation, vol. 17, no. 4, pp. 495511.
varying unstable plants", Information Sciences, vol. 181, no. 11, pp. [72] Pierobon, L., Nguyen, T.., Larsen, U., Haglind, F. & Elmegaard, B.
23702391. 2013, "Multiobjective optimization of organic Rankine cycles for
[53] Giarola, S., Zamboni, A. & Bezzo, F. 2011, "Spatially explicit multi waste heat recovery: Application in an offshore platform", Energy, vol.
objective optimisation for design and planning of hybrid first and 58, pp. 538549.
second generation biorefineries", Computers and Chemical [73] Niu, B., Wang, H., Wang, J. & Tan, L. 2013, "Multiobjective bacterial
Engineering, vol. 35, no. 9, pp. 17821797. foraging optimization", Neurocomputing, vol. 116, pp. 336345.
[54] Durillo, J.J. & Nebro, A.J. 2011, "JMetal: A Java framework for multi [74] CampomanesÁlvarez, B.R., Cordón, O. & Damas, S. 2013,
objective optimization", Advances in Engineering Software, vol. 42, no. "Evolutionary multiobjective optimization for mesh simplification of
10, pp. 760771. 3D open models", Integrated ComputerAided Engineering, vol. 20, no.
[55] Yang, X.. 2011, "Bat algorithm for multiobjective optimisation", 4, pp. 375390.
International Journal of BioInspired Computation, vol. 3, no. 5, pp. [75] Xue, B., Zhang, M. & Browne, W.N. 2013, "Particle swarm
267274. optimization for feature selection in classification: A multiobjective
[56] Asadi, E., Da Silva, M.G., Antunes, C.H. & Dias, L. 2012, "Multi approach", IEEE Transactions on Cybernetics, vol. 43, no. 6, pp. 1656
objective optimization for building retrofit strategies: A model and an 1671.
application", Energy and Buildings, vol. 44, no. 1, pp. 8187. [76] Wang, L., Yang, Y., Dong, C., Morosuk, T. & Tsatsaronis, G. 2014,
[57] Ali, M., Siarry, P. & Pant, M. 2012, "An efficient Differential "Multiobjective optimization of coalfired power plants using
Evolution based algorithm for solving multiobjective optimization differential evolution", Applied Energy, vol. 115, pp. 254264.
problems", European Journal of Operational Research, vol. 217, no. 2, [77] Huang, W. 2014, "Design exploration of threedimensional transverse
pp. 404416. jet in a supersonic crossflow based on data mining and multiobjective
[58] Fesanghary, M., Asadi, S. & Geem, Z.W. 2012, "Design of low design optimization approaches", International Journal of Hydrogen
emission and energyefficient residential buildings using a multi Energy, vol. 39, no. 8, pp. 39143925.
objective optimization algorithm", Building and Environment, vol. 49, [78] Ahmadi, P., Dincer, I. & Rosen, M.A. 2014, "Thermoeconomic multi
no. 1, pp. 245250. objective optimization of a novel biomassbased integrated energy
[59] Yildiz, A.R. & Solanki, K.N. 2012, "Multiobjective optimization of system", Energy, vol. 68, pp. 958970.
vehicle crashworthiness using a new particle swarm based approach", [79] Ehrgott, M., Ide, J. & Schöbel, A. 2014, "Minmax robustness for multi
International Journal of Advanced Manufacturing Technology, vol. 59, objective optimization problems", European Journal of Operational
no. 14, pp. 367376. Research, vol. 239, no. 1, pp. 1731.
[60] Karande, P. & Chakraborty, S. 2012, "Application of multiobjective [80] Song, Z., Li, J., Han, X., Xu, L., Lu, L., Ouyang, M. & Hofmann, H.
optimization on the basis of ratio analysis (MOORA) method for 2014, "Multiobjective optimization of a semiactive
materials selection", Materials and Design, vol. 37, pp. 317324. battery/supercapacitor energy storage system for electric vehicles",
[61] Devarajan, R., Jha, S.C., Phuyal, U. & Bhargava, V.K. 2012, "Energy Applied Energy, vol. 135, pp. 212224.
aware resource allocation for cooperative cellular network using multi [81] Wang, M., Wang, J., Zhao, P. & Dai, Y. 2015, "Multiobjective
objective optimization approach", IEEE Transactions on Wireless optimization of a combined cooling, heating and power system driven
Communications, vol. 11, no. 5, pp. 17971807. by solar energy", Energy Conversion and Management, vol. 89, pp.
[62] Gjorgiev, B. & Cepin, M. 2013, "A multiobjective optimization based 289297.
solution for the combined economicenvironmental power dispatch [82] Taormina, R. & Chau, K.. 2015, "Neural network river forecasting
problem", Engineering Applications of Artificial Intelligence, vol. 26, with multiobjective fully informed particle swarm optimization",
no. 1, pp. 417429. Journal of Hydroinformatics, vol. 17, no. 1, pp. 99113.
[63] Najafi, M., Eshghi, K. & Dullaert, W. 2013, "A multiobjective robust [83] Sun, L., Depuy, G.W. & Evans, G.W. 2016, "Multiobjective
optimization model for logistics planning in the earthquake response optimization models for patient allocation during a pandemic influenza
phase", Transportation Research Part E: Logistics and Transportation outbreak", Computers and Operations Research, vol. 51, pp. 350359.
Review, vol. 49, no. 1, pp. 217249.
[64] KhaliliDamghani, K., Abtahi, A.. & Tavana, M. 2013, "A new multi
objective particle swarm optimization method for solving reliability
redundancy allocation problems", Reliability Engineering and System
Safety, vol. 111, pp. 5875.
[65] Garg, H. & Sharma, S.P. 2013, "Multiobjective reliabilityredundancy
allocation problem using particle swarm optimization", Computers and
Industrial Engineering, vol. 64, no. 1, pp. 247255.
[66] Rao, R.V. & Patel, V. 2013, "Multiobjective optimization of heat
exchangers using a modified teachinglearningbased optimization
algorithm", Applied Mathematical Modelling, vol. 37, no. 3, pp. 1147
1162.
[67] Shokri, A., Bozorg Haddad, O. & Mariño, M.A. 2013, "Algorithm for
Increasing the Speed of Evolutionary Optimization and its Accuracy in
Multiobjective Problems", Water Resources Management, vol. 27, no.
7, pp. 22312249.
____________________________________________________________________________________________________________
124
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— 802.11 Wi-Fi technology is commonly used for expensive and resources are wasted as well. Satellites
creating wireless access networks with a maximum range of one technologies are very efficient for broadcast traffic, but when
hundred meters. With careful planning and proper antennas, it comes to bidirectional internet and intranet access and
this same technology can also be used to make point to point interactive communications, they tend to be severely limited
links up to several kilometres. Since, it is not always feasible in throughput and costly. Traditional microwave links can
and wise to run cables over long distances to connect different scale in throughput but generally require licensing and hence
networks, therefore, wireless links may turn out to be cost
effective alternative to their counterpart wired links while
are expensive to be utilized. Networks with base station model
creating long distance networks and providing network such as Worldwide Interoperability for Microwave Access
scalability. Wi-Fi-based point to point links can therefore be (WiMAX) and cellular networks like General Packet Radio
used to connect two local area network (LAN) segments, which Service (GPRS) and Code Division Multiple Access (CDMA)
besides being cost-effective, provides network scalability and consist of expensive base stations, which do not cover enough
other advantages such as high speed, centralized and easier users in low-density regions and hence lead to the waste of
management and high throughput for line of sight (LOS) resources [2].
applications. A Wi-Fi-based point to point link can extend the
range of wireless LAN by a few hundred feet to few miles which Wireless mesh networks are generally established using the
can further be increased by using highly directional antennas 802.11 Wi-Fi technology and serve to provide internet access
for point to point links, while serving as a backup network in in high user density environments. These wireless mesh
different organizations. So, we have designed a soft private networks provide full coverage of the region using Omni-
branch exchange (PBX) system for a university campus or directional Access Points (APs) over a range of about one
organization, facilitating voice over internet protocol (VoIP) hundred meters. But mesh networks suffer from two major
calls and instant messaging, employing the idea of Wi-Fi- based
short comings when scaled to larger areas. First, an increase in
point to point links. Moreover, we have also analysed the
quality of service (QOS) of the given setup in terms of data rate number of APs in the growing network with Omni-directional
and connectivity using bandwidth test and ping test respectively, antennas leads to increased interference in over-lapping cells.
for both transmission control protocol (TCP) and user Second, the use of low gain Omni-directional antennas
datagram protocol (UDP) scenarios. increases the hop length, resulting in a decrease in throughput
[2]. Thus, it can be implied that for low density of users,
Keywords— Wireless Point to Point Bridge Link, LAN traditional approaches that provide full coverage or require
segments, Soft PBX, VoIP, Instant Messaging, QOS. wiring of sites are not feasible. The alternative is to cover only
those few places where connectivity is required, by employing
I. INTRODUCTION long distance wireless point to point links. Such links can be
Many rural regions in developing and developed countries employed using Wi-Fi for low cost and ease of configurability.
with low user densities do not have good connectivity Hence, the best solution for internet access and intranet
solutions with a low budget. In low density environments, communications in low density environments and some
people are usually clustered around small localities, with large organizations is using Wi-Fi in bridge mode providing point
distances among these clusters. The example of such to point and point to multipoint links [1].
environments is a large enterprise situated in premises, with Wi-Fi-based point to point links have proved to be the cost
different departments or buildings at a distance or possibly effective alternative to their counterpart wired networks to
villages or sub-urban areas. In such cases, traditional options provide communication over long distances, especially in
to provide connectivity are not economically viable [1]. rural areas and within an enterprise or premises. Wi-Fi-based
Fibre optics can be a very good choice for long distance point to point links are relatively cheap and provide a number
backbone networks. It provides good reliability but it is not of advantages such as network scalability, high speed,
suitable for network scalability. Besides its installation and centralized and easier management and high throughput for
maintenance costs are quite high. So in case of low density Line of sight (LOS) applications. Wi-Fi-based point to point
environments and intranet connectivity, they turn out to be links can extend the range of wireless local area network
____________________________________________________________________________________________________________
125
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(WLAN) by a few hundred feet to a few miles, by segmenting wireless bridge is much more cost effective than the
it, using highly directional antenna and adequate planning for installation and maintenance of a wired network whether it
these point to point links. They can also serve as a backup is fiber optics or copper, for the same purpose. Wireless
network in different organizations. bridges provide connectivity when it is difficult to wire the
Therefore, in this paper, we propose the idea of Wi-Fi-based sites. Moreover, with wireless bridges, network scalability
point to point links for any type of low density environment can be performed very easily by just providing another AP
with voice over internet protocol (VoIP) and instant working in the bridge mode at the place of interest [4].
messaging services in a soft private branch exchange (PBX)
system. We setup a server to enable these applications at one B. Wireless APs
end of the soft PBX system and configure the APs to enable Wireless access points (APs or WAPs) are specially con-
Wi-Fi-based point to point links for providing services at the figured nodes on WLANs. These are the network devices
other end of the soft PBX system. Furthermore, we also used to connect multiple wireless devices to wired LAN,
analyse the quality of service (QOS) of the given setup in
for accessing internet. APs act as a central transmitter and
terms of data rate and connectivity using bandwidth test and
receiver of WLAN radio signals. APs used in home or
ping test respectively, for both transmission control protocol
small business networks are generally small, dedicated
(TCP) and user datagram protocol (UDP) scenarios.
hardware devices featuring a built-in network adapter,
The rest of the paper is organized as follows. In Section II,
related terminology used in this paper is discussed and Section antenna, and radio transmitter. APs support 802.11 wireless
III puts light on the system model. Section IV explains the soft communication standards. Usually, WAPs operate in "root
PBX system and Section V presents observations. Section VI mode", a point to multipoint configuration in which the
finally concludes the paper. AP relays frames between many 802.11 stations and an
adjacent Ethernet LAN. However, WAPs also have
II. RELATED TERMINOLOGY bridging mode that can be configured for connecting LAN
A. Wireless Bridges segments in point to point link configuration. Wireless
bridges relay frames between LAN segments using the
We use wireless bridges in our system to connect network
same 802.11 wireless communication standards [4].
segments using point to point links. LAN would be flooded
with unnecessary traffic if the messages are broadcasted to C. 2.1 FCC Rules for 802.11 Standards
every destination in that network thus bridges are usually
Since, Wi-Fi 802.11 standard operates in an unlicensed
used for the segmentation of LAN and interconnection of
Industrial scientific and medical (ISM) band therefore, the
LAN segments. A bridge network solution does not
signals operating over that band can interfere with each
necessitate difficult configurations, like IP routing. The
other. So, there are some rules defined by FCC to cope up
bridge network manages LAN segments and creates a single
with it that put some limitations on the power or Effective
subnet for the entire network. Bridges work at the data link
Isotropic Radiated Power (EIRP) of the signals radiated on
layer of the Open System Interconnection (OSI) model.
this band. These rules are different for 2.4 GHz and 5 GHz
Bridges learn which addresses are on which network
operating frequencies of ISM bands as well as for point to
segment and develop a forwarding table so that subsequent
point and point to multipoint topologies. In this project, a
messages can be forwarded to the right network segment.
point to point link is established between LAN segments
Bridges examine the incoming packets and look up the
operating on the 5 GHz band within the limitations defined
forwarding table for their destination m e d i a a c c e s s
by FCC, which state that. “For maximum transmitted power
c o n t r o l ( MAC) address. If the destination MAC address is
of 30dBm, a directional antenna gain of 23dBi can be used
found in the forwarding table, the packet is forwarded to the
with no reduction of transmitter power output in point to
corresponding port and if the destination MAC address is not
point links. However, if directional antenna gain increases
found in the same segment, the bridge restricts the
greater than 23dBi, a 1dB reduction in peak transmitter
transmission [3].
power is required for each 1dBi increase in antenna gain
Wireless bridges are same in functionality and used to greater than 23dBi” [5].
connect two LAN segments via a wireless interface such as
a radio link, to facilitate connectivity and data transfer D. Fresnel Zone Clearance in Point to Point Links
between them. Wireless bridges are commonly used to We use 5 GHz ISM band to establish point to point link
interconnect wired network segments such as an Ethernet in our system that falls in the microwave region of
network via wireless link. In simple terms a wireless bridge electromagnetic spectrum. Microwave links are generally
is a device that allows two network segments of users to used for LOS communications as microwaves are highly
transparently communicate to one another over long directional and travel in straight lines. But the energy of
distances without wires. They can be used to connect microwaves is not pencil thin. They spread out the farther
areas that are geographically apart like a remote building to they get from the antenna. The area that the signal spreads
the main building up to 30 miles, using proper antennas out is called Fresnel Zone. If there is an obstacle in the
and LOS. Connecting two locations wirelessly through a Fresnel zone, part of the radio signal will bent away from
____________________________________________________________________________________________________________
126
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the straight line path which will result in the reduction of the enable the use of MIMO feature of 802.11n standard used in
amount of radio frequency (RF) energy reaching the the device. With HT chains 0 and 1 enabled at the
receiving antenna. Fresnel zones can also be viewed as the transmitter and receiver sides, 2 by 2 antenna diversity and
concentric ellipsoids surrounding the transmitter, receiver spatial multiplexing can be achieved which significantly
and the LOS between them. The first Fresnel zone is the increases the throughput of the device. With “HT guard
region where the microwave transmission energy is the most interval” option, the length of guard interval can be adjusted
intense and it is the closest to the direct line between to minimize the inter symbol interference while maintaining
transmitter and receiver, as shown in Fig. 1. the desired throughput [9].
____________________________________________________________________________________________________________
127
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Wireless point to point link is created at both AP and client (SIP) address, very much like extensions in conventional
by defining different interfaces with address resolution PBX, with the soft PBX server, which maintains a
protocol (ARP) enabled. The wireless characteristics of the database of all its users and their corresponding SIP
point to point link such as band, channel width, frequency addresses. The VoIP telephone calls are established modified
and wireless protocol are set to be 5 GHz, 20 MHz, 5180 and terminated using the IP telephony protocol SIP. To make
MHz and Nv2 respectively at both AP and client end of the a call, the server is requested to establish a connection.
LAN segment. The manual transmit power is set to 30dBi When the called party’s number is dialed, the IP address of
as per FCC rules requirement at the client end. All HT the phone is mapped to the corresponding SIP address and
chains are enabled at both AP and client ends to provide then it is sent to the destination. The calls to and from
MIMO feature and the guard interval is kept 800ns to avoid PSTN are routed via VoIP gateway. If the call is to be made
inter symbol interference and get a more stable link. After outside the soft PBX network, the server directs it to the
setting the wireless point to point link, it is necessary to gateways from where it is sent towards the PSTN. Similarly
make bridge configurations at both ends of the LAN the calls from PSTN are directed to the server via the
segment. At both AP and client side, rapid spanning tree gateway and the server then sends them to the appropriate
protocol (RSTP) is used to prevent bridging loops and two destination [8].
interfaces are defined, one for the point to point bridge link
and the other for connecting to the particular LAN segment. A. For VoIP calls
We use Dynamic host configuration protocol (DHCP) to For VoIP calls, we use Asterisk which is open source soft
provide automatic IP address to the devices, connected to PBX software that can use both traditional Time Division
each LAN segment for soft PBX call and instant messaging. multiplexing (TDM) technology and packet voice protocols
The configuration is made on AP device that is connected to (VoIP and Voice over Frame relay). Asterisk acts as a full
the soft PBX server. A static IP is defined for the gateway featured PBX, supporting virtually all conventional call
DHCP and the rest of the IP addresses are allocated features on SIP phones like Caller ID, Call waiting, Call
dynamically from the DHCP pool for a maximum lease time forward/busy, Call forward/no answer, Voice mail, Least
of 3 days. cost routing, Call conferencing and many more. Asterisk
supports three VoIP protocols, two industry standards and
IV. SOFT PBX one specifically for Asterisk. Inter-Asterisk exchange (IAX)
PBX is a telephone switching system owned by a is the de-facto standard for Asterisk networking. The other
company that manages incoming and outgoing calls for the protocol used by Asterisk is Session initiation Protocol (SIP)
company’s internal users. Companies lease only one line and which is an Internet Engineering Task Force (IETF) standard
have many people using it, with each one having a phone at for VoIP. The last one is H.323 which is an International
the desk with different number, called extension. PBX Telecommunication Unit (ITU) standard for VoIP. Asterisk
automatically sends incoming calls to the required provides seamless and transparent translation among so
extensions. A PBX hence can switch calls within the many codecs and file formats such as A-law, u-law, GSM
organization and can also connect to the Public switched 6.10, MP3, PCM, VOX and LCP-10 [9]. In order to
Telephone Network (PSTN) for calling outside the configure soft PBX, we download and install Asterisk10
organization. A conventional PBX can be costly as it along with its packages on Ubuntu Linux operating system.
requires copper wires and other hardware equipment. Soft SIP file is edited to define SIP clients, codecs used, mailbox
PBX (also called IP or virtual PBX) is a telephone system addresses and other options. We set different extensions for
used for the sending voice over the internet protocol (IP) each SIP client or user under the context of phones
network and is same in application as the conventional PBX and a shared password is used to authenticate each phone.
but functions differently. It is based on a computer PBX The user reachability checks are performed every 60seconds.
software and voice over internet protocol (VoIP) instead of Out of all codes available, we use only u-law, A-law and
relying on the traditional telephone hardware and copper GSM codes in our soft PBX system. Mailbox is used for the
circuits. A soft PBX system uses a single network for data voice mail messages such that the voice mails for the user
traffic and voice calls, both encapsulated in IP packets and having extension 101 will be saved at 101@default. Dial
then transferred over the network, unlike conventional PBX. plan file is created to handle incoming and outgoing calls
The major part of the soft PBX is handled by the software so such that the commands are executed in the following order:
they are relatively inexpensive. A soft PBX system consists when there is a call for some extension, direct the call to that
of three components; phones, soft PBX server and an extension. After 20s the called party is asked to leave their
optional VoIP gateway. Phones should capable of supporting message which will be saved at extension@phones. The
VoIP. Soft PBX software installed in PC can serve the server is then directed to play an audio file and then hang-up
purpose of a soft PBX server. VoIP gateways are optional the call. A separate file for voicemail is made with a separate
and are used to connect to the external land line PSTN. The extension such that the called party is directed to the main
users in soft PBX register their session initiation protocol voice mail system to check the voice mails if there are any.
____________________________________________________________________________________________________________
128
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
B. For Instant Messaging Bandwidth test is done to get the bandwidth and data rate
of the link in both directions that is transmit and receive, at
Open fire version 3.8.0 server is used in the project to
both AP and the client. Bandwidth test also gives the packet
facilitate instant messaging which uses the only widely
loss at an instant when some particular number of packets
adopted open protocol for instant messaging; that is
is sent. Bandwidth test is done in two different scenarios of
Extensible Messaging and Presence Protocol (XMPP) (also
TCP and UDP as they provide different results in terms of
called Jabber). Jabber is an open, Extensible Mark-up
data rate and packet loss. Bandwidth test performed at the
Language (XML)-based protocol for instant messaging
AP device gives the results shown in Fig. 3. Results are
written in Java. Open fire can support Instant Messaging
taken for UDP as well as TCP in both directions when the
Transports that provides connectivity to external Instant
transmitted and received packet sizes are 1500 bits. The
Messaging services such as Yahoo Instant Messaging,
QOS analysis for UDP shows that the data rate at the
Microsoft Network (MSN)/Windows Live Messaging and
transmission is 45 Mbps and at the reception is 54.5 Mbps
Google talk Messaging. The server is also compatible with
giving the total data rate of about 99.5 Mbps in both
the Asterisk 10 server used for VoIP calls in the system.
directions that can adequately make several voice calls and
Concisely, open fire is a freeware open source server that
data transmissions simultaneously. For a packet size of 1500
provides instant messaging, broadcast messaging, offline
bits per packet, almost 66,333 packets are transmitted per
messaging and group chat. Open fire also provides a higher
second in both directions at a bidirectional data rate of
degree of security to the end user as the clients are connected
99.5 Mbps, out of which 554 packets are lost at an instant.
to the server using the Secured Sockets Layer (SSL) secured
Therefore, the packet loss is never exceeding 1 percent
connection so; the traffic streams in both directions are
which is optimum for voice calls and data transmission. For
encrypted. The server also provides a way of authentication
TCP, it can be clearly seen that the packet loss is 0 since
to its clients for improved security using a different user ID
TCP is a connection oriented reliable protocol in which each
and password for each client. Whenever a client is registered
transmitted packet is acknowledged and none of the packet is
to the server, the user ID and password for that instant
lost. However, due to multiple acknowledgements, the data
messaging service are stored in encrypted form on the server
rate is significantly reduced to 9.1 Mbps at the transmission
and are used for client authentication. Open fire is based on
side, 24.3 Mbps at the reception side and a total data rate of
the protocol that runs on Java so open fire requires Java Run
39.4 Mbps in both directions. But the data rate is still enough
time Environment (JRE) and Java development Kit (JDK) to
to send multiple voice calls and data files simultaneously.
operate. Open fire also requires MySQL server to make its
Bandwidth test performed at the client device gives the
user accounts database [10]. Therefore, first the latest
results presented in Fig. 4. The results for UDP show that the
versions of Oracle JRE/JDK are installed on the Linux
data rate at the transmission side is 61.5 Mbps and at the
server and then MySQL database server and client packages
reception is 32 Mbps, giving a total data rate of 93.5 Mbps in
are installed. When the MySQL server and client packages
both directions that is optimum for making several voice
are installed, the database or user accounts are ready to be
calls at the same time and the packet loss is seen to be zero
set. MySQL has its own user accounts which are not related
even for UDP protocol. The results for TCP protocol at the
to the user accounts on the Linux machine for VoIP calls. By
client side show that the effective data rate is reduced to
default, the root account of MySQL server is empty so, root
25.8 Mbps at the transmission side and 11.3 Mbps at the
accounts along with the passwords are defined. A new
reception side summing up to a total data rate of 37.1 Mbps
MySQL database is created for Open fire with a new
in both directions. The packet loss is zero for the same
username and password. After that Open fire 3.8.0 is
reason of reliability.
downloaded, installed and configured. For the configuration
of Open fire, first administrator account is setup with email B. Ping Test
and password and then the administrator account is used to Ping is an acronym for the word “Packet internet gofer”.
create users/clients with their corresponding usernames, Ping works over Internet Control Message (ICMP) protocol
email addresses and passwords. that checks for the relative connectivity in the network. So,
ping test is performed to check the connectivity of the link
V. OBSERVATIONS
and an average delay that it takes for a packet to be sent
Once the point to point bridge link is established and it is to a destination and receive a packet in response from the
working properly, it becomes important to analyze the QOS destination. Ping test is performed at both the AP and client
provided by the link for the VoIP calls and instant messaging devices to check connectivity and delay at both ends. Ping
services. The key parameters to estimate the QOS of any test is performed at AP to check its connectivity with the
network are its delay, bandwidth or data rate and packet loss. client. Ping test performed at the AP device gives the
These parameters are tested one by one over the link to results, as depicted in Fig. 5. As seen from the ping
observe the QOS by the link for each service. statistics of the packets sent and ping replies, it is clear that
____________________________________________________________________________________________________________
129
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
VI. CONCLUSION
In this work, we have used wireless point to point bridge
link to connect LAN segments located separated 1km apart
and provide VoIP and instant messaging service using soft
PBX in both LAN segments. Although wireless bridge link
operating in an unlicensed band and can suffer from
Fig. 3. Bandwidth Test at AP for UDP and TCP interference and lack of reliability, but these shortcomings can
be adjusted with proper planning of the link. Wireless bridge
link can offer outstanding advantages such as cost
effectiveness, suitable data rate, easier installation, easier
trouble shooting and scalability for low density environments.
REFERENCES
[1] Wireless Revolution. (2016) Wireless Revolution on Economist.
[Online].Available: http://www.economist.com/node/9080024.
[2] Point to Point Broadband Wireless for Enterprise. (2016) White paper
on Point to Point Broadband Wireless for Enterprise.
[Online].Available:
http://www.motorolasolutions.com/content/dam/msi/docs/business/solu
tions/industry_solutions/education/motowi4/_documents/static_files/ne
_wb_enterprise_wp_us_r4_new.pdf.
[3] Jeffrey S. Beasley, Networking, 2nd ed., New Mexico State Univ.,
USA: Prentice Hall, 2008.
[4] Understanding Wireless Bridging and WDS. (2016) on Connect802.
[Online].Available:
http://www.connect802.com/wireless_bridging.html.
[5] FCC Rules. (2016) the fcc-rules on afar tutorials. [Online].Available:
http://www.afar.net/tutorials/fcc-rules.
[6] Fresnel Zone. (2016) Fresnel Zone on Digital Air Wireless.
[Online].Available:
http://www.digitalairwireless.com/wireless-blog/recent/fresnel-zones-
what-are-they-and-why-are-they-so-important.html.
Fig. 4. Bandwidth Test at client for UDP and TCP [7] Mikrotik. (2016) Wireless Workshop on Mikrotik. [Online].Available:
http://mum.mikrotik.com/presentations/US12/uldis.pdf.
____________________________________________________________________________________________________________
130
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
131
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Detecting the rail surface faults is one of the most Today, the rail inspection operation is basically divided into
important components of railway inspection process which should two groups as contact and contactless methods. The simplest
be performed periodically. Today, the railway inspection process method of inspecting is that an expert visually inspects the rail
is commonly performed using computer vision. Performing line with mechanical measuring instruments. This method is
railway inspection based on image processing can lead to false- limited by very slow, low accuracy specialized knowledge
positive results. The fact that the oil and dust residues occurring inspecting the rail line [4].
on railway surfaces can be detected as an error by the image
processing software can lead to loss of time and additional costs in
the railway maintenance process.
In this study, a hardware and software architecture are
presented to perform railway surface inspection using three-
dimensional laser cameras. In addition two-dimensional data that
Ccd/cmos cameras have on the x-y plane, laser cameras have
three-dimensional input data as they include precise distance
information on the z plane. Generally, three-dimensional data
acquiring processing is very commonly used in machine vision
applications such as mobile robots, image enhancement, medical
and fault diagnosis. The use of three-dimensional laser cameras in
railway inspection process provides high accuracy rates. The
reading rate of laser cameras to read 10,000-30,000 profiles per
second is another important advantage provided in real time
railway inspection.
Consequently, a computer vision-based approach in which
three-dimensional laser cameras that could allow for contactless
and fast detection of the railway surface defects such as fracture,
scouring and wear with high accuracy are used in the railway
inspection process was proposed in the study.
Fig. 1 Rail inspection with ultrasonic devices [5]
Keywords— Railway Inspection, Anomaly Detect, Computer
Vision, Laser Camera, Machine Learning
Another one of the contact methods is performing the rail
I. INTRODUCTION inspection with ultrasonic devices. Fault detection is conducted
In industrial applications, it is necessary to conduct activities by analyzing the data and graphics obtained by the friction of
such as maintenance, condition monitoring and fault detection the mechanical device which is moved along the rail line to the
as well as production and R & D activities. In recent years, railway rail line [4]. The inspection process is slow in this
these operations are commonly conducted using computer method shown in Fig. 1. Its most important disadvantage is that
vision [1]. Computer vision ensures that these operations are the mechanical device can increase the existing fault on the rail
conducted fast with high accuracy and without human line or new faults may occur because of the device's necessity
dependence. Due to all these advantages offered by it, the of friction to the rail line during inspection.
maintenance, monitoring and fault detection conducted using Although contact methods are low-cost, they have
computer vision are also commonly used in railway rail significant disadvantages because they do not have high
inspection as in other fields of the industry [2]. accuracy ratios and the inspection process is long. Today, rail
Rail transportation systems are widely used all around the inspection operation can be performed rapidly as contactless
world. The anomalies that could occur or are present on railway and with high accuracy using computer vision. The general
can lead to both troubles and financial losses and accidents in components of a computer vision-based rail inspection
the transportation system. Therefore, the rail line should be operation are seen in Fig. 2 [6].
checked periodically [3].
____________________________________________________________________________________________________________
132
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
133
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
∑ =∑ ( − ) +( − ) (1)
TABLE I I
Feature Value
Performance 10.000-35.000 Prof le/Second
Resolut on 1536 x 512 P xel
F eld of V ew 250 x1200 mm
3d resolut ons 0.2 mm
Interface G gab t ethernet b)
Encoder frequency 2Mhz
Fig. 4 a) Experimental setup, b) Block diagram of the proposed method
Other H gh-pass f lter, v vrat on reduct on
____________________________________________________________________________________________________________
134
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Y. Tai-shan, C. & Du-wu, “The method of intelligent inspection of
product quality based on computer vision”, In 2006 7th International
Conference on Computer-Aided Industrial Design and Conceptual
Design , pp. 1-6, IEEE, 2006.
[2] S. Zheng, X. Chai, X. An, L. Li, ”Railway track gauge inspection method
based on computer vision”, In 2012 IEEE International Conference on
Mechatronics and Automation, pp.1292-1296, IEEE, 2012.
[3] Q. Li, Z. Zhong, Z. Liang, Y. Liang, “Rail Inspection Meets Big Data:
Methods and Trends”, In Network-Based Information Systems (NBiS),
Fig. 6 Roc performance curves 2015 18th International Conference, pp.302-308, IEEE, 2015.
[4] D. F. Cannon, K. O. EDEL, S. L. Grassie, K. Sawley, ”Rail defects: an
overview”, Fatigue & Fracture of Engineering Materials &
The trained system was tested by measuring the operating Structures, 26(10), 865-886, 2003.
[5] www.cater-eu.com/gallery, “C.A.T.E.R”. [Online]. 2016
speed and accuracy performance on a rail line for a short time [6] Y. Santur, M. Karaköse, İ. Aydın and E. Akın, “IMU based adaptive blur
in the test phase. A total of 400 rail profiles were collected on removal approach using image processing for railway inspection”,
a 1 km-long rail line by the proposed system, and these profiles In 2016 International Conference on Systems, Signals and Image
were trained by labeling as healthy and faulty. 75% of the Processing (IWSSIP), pp.1-4, IEEE, May, 2016.
[7] E. Resendiz, L. F. Molina, J. M. Hart, J. R. Edwards, S. Sawadisavi, N.
dataset was used in training. The remaining 25% was used for Ahuja and C. P. L. Barkan, “Development of a machine-vision system
testing purposes. % 98 accuracy rate was achieved in the test for inspection of railway track components”, In 12th World Conference
phase. The receiver operating characteristic (ROC) analysis on Transport Research, 12.WCTR, Lisbon, Portugal, 2010.
was used to obtain the accuracy performance values shown in [8] R. Huber-Mörk, M. Nölle, A. Oberhauser, E. Fischmeister, “Statistical
Rail Surface Classification Based on 2D and 21/2D Image Analysis”,
detecting the "faulty" frames containing system anomalies and In Advanced Concepts for Intelligent Vision Systems, pp.50-61, 2010
the "healthy" frames having no anomalies [18]. The roc curve [9] T. Hackel, D. Stein, I. Maindorfer, M. Lauer and A. Reiterer, “Track
obtained by the proposed method is given in Fig. 6. detection in 3-D laser scanning data of railway infrastructure”, I2MTC,
2015 IEEE International (pp. 693-698), 2015.
V. CONCLUSIONS [10] M. Karakose and M. Baygin, “Image processing based analysis of
moving shadow effects for reconfiguration in pv arrays”, In Energy
Rail transportation systems are widely used around the world. Conference (ENERGYCON), IEEE Intern., pp. 683-687, May, 2014.
The rail lines should be inspected periodically and their [11] X. Peng, J Liang, “3D Detection Technique of Surface Defects for Steel
maintenance should also be performed to ensure the railway Rails Based on Linear Lasers”, Journal of Mech. Eng., 8, 003, 2010.
[12] M. Karakose and E. Akin, “Type-2 fuzzy activation function for
transportation safety. Today, this inspection process is multilayer feedforward neural networks”, In Systems, Man and
commonly carried out using CVS. In this study, the software Cybernetics, 2004 IEEE Int. Conf. on (Vol. 4, pp. 3762-3767), 2004.
architecture is presented for a rail inspection conducted using [13] Y. Santur, M. Karakose, E. Akin, "Random Forest Based Diagnosis
3-D laser cameras. Approach for Rail Fault Inspection in Railways", International
Conference on Electrical and Electronics Engineering (Eleco 2015),
In general, there are three criteria expected from a rail 9.th, pp.714-719, 2015.
inspection application. These are the high accuracy results, high [14] H. Misawa, S. Juodkazis, “3D laser microfabrication: principles and
operating speed and the cost of the system. applications”, John Wiley & Sons, 2006.
3-d laser cameras allow for obtaining higher accuracy rate in [15] G. Zhang, Z. Wei, “A novel calibration approach to structured light 3D
vision inspection”, Optics&Laser Technology, 34(5), pp.373-380, 2002.
rail inspection because they include both Rgb data and precise [16] Sick sensor intelligence, ”3d Cameras”, 8011438 datasheet, May.2009.
distance information that two-dimensional cameras have. [17] P. Dubath, L. Rimoldini, M. Süveges, J. Blomme, M. López, L. M. Sarro,
When they are compared with normal cameras, their another K. Nienartowicz, “Random forest automated supervised classification of
advantage is that they are more susceptible to false positive Hipparcos periodic variable stars”, , 414(3), pp.2602-2617, 2011.
[18] D. Lorente, N. Aleixos, J. Gómez-Sanchis, S. Cubero and J. Blasco,
result situations in oil and dust stains caused by image “Selection of optimal wavelength features for decay detection in citrus
processing algorithms because they use distance information. fruit using the ROC curve and neural networks”, Food and Bioprocess
Technology, 6(2), pp.530-541, 2013.
____________________________________________________________________________________________________________
135
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, three different apple varieties grown in Wu et al. (2016) proposed a novel fuzzy clustering model,
Karaman province are classified using kNN and MLP algorithms. called fuzzy discriminant c-means (FDCM) clustering, to
90 apples in total, 30 Golden Delicious, 30 Granny Smith and 30 classify apple varieties couple with near infrared spectroscopy.
Starking Delicious have been used in the study. DFK 23U445 USB The 200 apple samples of four apple varieties were made
3.0 (with Fujinon C Mount Lens) industrial camera has been used
experiments to collect the near infrared reflectance (NIR)
to capture apple images. 4 size properties (diameter, area,
perimeter and fullness) and 3 color properties (red, green, blue) spectra. The clustering accuracy of FDCM achieved 97%
have been decided using image processing techniques through which was higher than FCM, possibilistic c-means and
analysing each apple image. A data set which contains 7 physical Gustafson–Kessel (GK) clustering [3].
features for each apple has been obtained. Classification success Shahin et al. (2013) have proposed to detect bruises in apples
rates and error rates have been decided changing the neuron by using line–scan x–ray imaging method. Spatial and
numbers in the hidden layers in the classification using MLP transform features were evaluated for their discriminating
model and in different neighbour values in the classification made contributions to fruit classification based on bruise defects. Red
using kNN algorithm. It is seen that the classification using MLP delicious (RD) and golden delicious (GD) apples have been
model is much higher. While the success rate of classification made
used in this study. Separate artificial neural network (ANN)
according to apple type is 98.8889%.
classifiers were developed for old (one month) and new (24
hour) bruises. When an ANN classifier was used to sort apples
Keywords— Image processing, Apple classification, kNN, MLP
based on old bruises, it achieved an accuracy of 90% for RD
I. INTRODUCTION apples and 83% (93% after threshold adjustment) for GD
apples. For new bruises, the accuracy was approximately 60%
Apple production is an important economic activity in
for both RD and GD apples. New bruises were not adequately
Karaman. The dominant apple varieties produced in Karaman,
separated using this methodology [4].
are Golden Delicious, Granny Smith and Starking Delicious.
In this study, 90 apples grown in Karaman have been used.
During marketing these apples, the product is wanted to be
7 morphological attributes have been obtained from their
consisted of just one species of apples. The mixture rate of other
images by using image processing techniques. The apples have
apple varieties in the product is wanted to be at minimum level.
been classified into 3 group (Golden Delicious, Granny Smith
For this purpose many selection system have been proposed in
and Starking Delicious) with kNN and MLP algorithms by
the literature.
using the obtained attributes.
He Yong et al. (2007) have evaluated the potential of
visible/near-infrared (Vis/NIR) spectroscopy for its ability to II. MATERIALS AND METHODS
non-destructively differentiate apple varieties. The chemo
An interface presented in Figure 1, have been designed by
metrics procedures applied to the Vis/NIR data were principal
using GUIDE (Graphical User Interface Design) in MATLAB.
component analysis (PCA), wavelet transform (WT), and
The image of the apple under the camera is taken when the Take
artificial neural network (ANN). The highest level of correct
Button pressed on the GUI.
classification (100%) of the apple varieties have been achieved
The mean of R (red), G (green) and B (blue) pixel values of
with WT-ANN model [1].
the area where the apple is in the image have been calculated.
Ronald et al. (2016) have proposed to investigate the
These values have been used for classification of the apples for
applicability and performance of Naive Bayes algorithm in the
their colors. Then the colored images have been converted to
classification of apple fruit varieties. Apple classification
grey level image by using image processing techniques.
system prototype was built using MATLAB R2015a
Morphological operations have been applied onto bitwise
development platform environment. The results showed that
images. The image processing steps have been presented in
the averaged values of the estimated accuracy, sensitivity,
Figure 2.
precision and specificity were 91%, 77%, 100% and 80%
respectively [2].
____________________________________________________________________________________________________________
136
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
cluster either directly or by calling via Java code [6,7]. They are
also suitable for developing new machine learning algorithms.
1) Multilayer Perceptron: It is a feed forward type
artificial neural network model which maps input sets
onto appropriate output sets. A multilayer perceptron
(MLP) is composed of multiple layers of nodes where
each layer is connected to the next. Each node is a
processing element or a neuron that has a nonlinear
activation function except the input nodes. It uses a
supervised learning technique named back propagation
and it is used for training the network. The alteration of
the standard linear perceptron, MLP is capable of
distinguishing data which are not linearly separable [7].
2) K-Nearest Neighbour Algorithm: The k-NN is a
supervised learning algorithm that solves classification
problems. Classification is the examination of the
attributes of an image and the designation of this image
to a predefined class. The important point is the
determination of the features of each category in
advance [8]. According to the kNN algorithm used in
the classification, based on the attributes drawn from the
Fig. 2 The image processing steps
classification stage, the distance of the new individual
The physical properties of the apple like radius, perimeter, that is wanted to be classified to all previous individuals
area, volume, mass and eccentricity have been extracted. So is considered and the nearest k class is used. As a result
that creates 7 attributes by gathering 4 morphological and 3 of this process, test data belongs to the k-nearest
colour. neighbour category that has more members in a certain
class. The most important optimization problems in the
A. Software-WEKA kNN method are the identification of the number of
Developed by Waikato University in New Zealand, WEKA neighbours and the method of distance calculation
is an open-source data mining software with a functional algorithm. In the study, the identification of the
graphical interface which incorporates machine learning optimum k number is performed with experiments, and
algorithms [5]. WEKA includes various data pre-processing, the Euclidean Distance Calculations method is used as
classification, regression, clustering, association rules, and a distance calculation method.
visualization tools. The algorithms can be applied on the data Euclidean calculation method [9]:
____________________________________________________________________________________________________________
137
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I
III. RESULTS AND DISCUSSION THE SUCCESS RATE AND ERROR VALUES OBTAINED BY USING KNN
In the study, WEKA software was used in order to classify CLASSIFIER
apples by species (Golden Delicious, Granny Smith and
Starking Delicious). Using the kNN algorithm, the Neighbourliness Classification
MAE RMSE
classification success rates, mean absolute error (MAE) and Number (k) accuracy (%)
root mean square error (RMSE) were obtained for various k- 1 93.3333 0.0414 0.1886
neighbour values. The classification success rates obtained with
2 96.6667 0.2296 0.2854
kNN algorithm, and MAE and RMSE values can be seen in
Table 1. The success rate change versus number of neighbours 3 97.7778 0.0364 0.1283
for kNN method is presented in Figure 4. 4 90.0000 0.0512 0.1537
5 94.4444 0.0543 0.1541
6 93.3333 0.0539 0.1500
7 95.5556 0.0546 0.1503
8 94.4444 0.0644 0.1573
9 94.4444 0.0728 0.1670
10 91.1111 0.0803 0.1735
____________________________________________________________________________________________________________
138
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The number of
Classification
neurons in the MAE RMSE
accuracy (%)
hidden layer
1 94.4444 0.1975 0.2513
2 97.7778 0.033 0.0985
3 95.5556 0.0375 0.1350
4 96.6667 0.0352 0.1152
5 98.8889 0.1698 0.2225
6 97.7778 0.0355 0.1089 Fig. 6 The structure of Multilayer Perceptron
7 97.7778 0.0269 0.1049
IV. CONCLUSIONS
8 97.7778 0.0223 0.1073 In this study, classification of apples as Golden Delicious,
9 96.6667 0.0479 0.1234 Granny Smith and Starking Delicious have been proposed by
10 96.6667 0.0441 0.1214 using machine learning algorithms. The classification success
rates and error values of data mining algorithms were
20 95.5556 0.0746 0.1476
calculated for kNN and MLP. It was observed that the success
rate was higher for the classifications performed using the MLP
The diagram demonstrating the changes in classification algorithm. The highest classification success rate was achieved
success rate based on the number of neuron in the hidden layer when the number of neighborhood was 3 and the success rate
is demonstrated in Figure 5. was 97.7778 %. The MAE and RMSE error value were 0.0364
and 0.1283 respectively. For the classification success rates
obtained using MLP algorithm, the highest classification
success rate and minimum MAE error value was achieved for
5 neurons in the hidden layer and it was 98.8889%. For this
MLP structure, the MAE and RMSE error values were 0.1698
and 0.2225 respectively.
____________________________________________________________________________________________________________
139
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
ACKNOWLEDGMENT [4] Shahin, M.A., Tollner, E.W., McClendon, R.W., Arabnia, H.R. Apple
Classification Based on Surface Bruises Using Image Processing and
This study was supported by Karamanoglu Mehmetbey Neural Networks, Transactions of the ASAE, 2013, Vol. 45(5): 1619–
University scientific research project coordination with 01-M- 1627.
15 project number. [5] Witten I.H., Frank E., Hall M.A. (2011). Data mining: practical machine
learning tools and techniques. Elsevier, London.
[6] Patterson, D., Liu, F., Turner, D., Concepcion, A., Lynch, R., (2008).
Performance Comparison of the Data Reduction System. Proceedings of
REFERENCES the SPIE Symposium on Defense and Security, Mart, Orlando, FL, pp.
[1] He, Y., Li, X., Shao, Y. Fast Discrimination of Apple Varieties Using 27-34.
Vis/NIR Spectroscopy, Volume 10, Issue 1, 2007, pages 9-18. [7] Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P. & Witten,
[2] Ronald, M., Evans, M. Classification of Selected Apple Fruit Varieties I. H. (2009). The WEKA Data Mining Software: An Update, SIGKDD
Using Naive Bayes, Indian Journal of Computer Science and Explorations, Volume 11, Issue 1, pp. 10-18.
Engineering (IJCSE), Vol. 7 No. 1 Feb-Mar 2016, pages 13-19. [8] Wang, J., Neskovic, P., & Cooper, L. N., (2007). Improving nearest
[3] Wu, X., Wu, B., Yang, N. Classification of Apple Varieties Using Near neighbor rule with a simple adaptive distance measure, Pattern
Infrared Reflectance Spectroscopy and Fuzzy Discriminant C-Means Recognition Letters, 28(2):207-213.
Clustering Model, Journal of Food Process Engineering, doi: [9] Zhou, Y., Li, Y. & Xia, S., (2009). An improved KNN text classification
10.1111/jfpe.12355. algorithm based on clustering, Journal of computers, 4(3):230-237, 2009.
____________________________________________________________________________________________________________
140
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Information is power. Nowadays, main concern of light of a group’s cultural customs. But as in laws it’s different,
cyber community is to protect this valuable asset. Technical and many laws are universal across the world.
technological security measures are sometimes insufficient to Laws are the rules that all of the people must oblige to
protect an information system. Because there is a human factor follow. If they don’t follow, they will be punished. That’s not
in information system. Ethics are set of moral rules that guide
the case in ethics; in ethics all people have their own will.
people. With the help of ethics a better and robust security can
be achieved. In this paper role of ethics in information security is They can choose to follow or not follow code of ethics. In
discussed. First of all law, ethics and information security ethics, it depends on a moral decision, many people agree
concepts are briefly introduced. Later, some ethical concerns and upon ethical behaviors but some people can disagree on these.
perspectives in information security are given. To emphasize role Ethics can be seen from different approaches. These
of ethics in information security, several studies are reviewed. approaches are consequentialism and deontological [3].
Finally, mechanisms to make ethical rules effective in an Consequentialism claims that if actions have bad
organization/community are discussed with several case studies. consequences they are bad after all. Deontological approaches
don’t think like that. They assume that actions of people have
Keywords— Ethics, Ethical Issues, Information Security, Cyber neither good nor bad consequences they have only moral
Security, Information Systems. duties independent of their actions.
There exists a blurriness of ethics in individual’s mind but
I. INTRODUCTION
with the help of the laws and regulations security and safety
With the advance of Information Technology, new threats can be achieved. In a cyber world, law and ethics work
and unauthorized actions arise each day. To be able to protect together to form a security layer. Ethics can fill the void where
information assets against these threats and actions is one of laws cannot be applied.
the most important issues nowadays. But sometimes technical
and technological measures are not enough to protect an III. INFORMATION SECURITY
information asset. Additional measures must be employed Rapid growing of information technology enables many
because there are a lot of parameters when it comes to organizations, government and civilians exchange important
information security. One of these parameters is people. These information every day. So security issues arise when sharing
people can be system administrators, security professionals, information [4]. Nowadays relationship of organizations
employees and users. These are the people that interact with depends on computer and information systems, many
information system. In order to secure people parameter in an organizations are concerned about information security
information system, a measure that employs moral judgment because they use technologies like e-commerce, mobile and
must be introduced. Computer and information ethics are virtual private networks. With the increase of involvement of
studied by many researchers, scholars and practitioners [1]. To these technologies, number of threats to organization’s
include ethical layer to information security is very important valuable assets also increases [4]. Information security
because it can fill the gap that people create. protects confidentiality, integrity, and availability of
In this paper role of ethics in information security is information assets against various threats. Technical
discussed. First of all law, ethics and information security protection measures are not enough to provide information
concepts are briefly introduced. Some ethical concerns and security. There must be other measures. To develop a robust
perspectives of several researchers in information security are and good information security; in addition to technical
given. To emphasize role of ethics in information security, measures; operational, ethical, sociological and legal measures
several studies are reviewed. Mechanisms to make ethical must be considered [5].
rules effective in an organization/community are discussed. Organizations depend on processes, technology and people.
Finally, paper finishes with conclusions and references. Even if we have top of the line information systems and
security, there are people operating these information systems
II. LAW & ETHICS
[4]. They control daily activities and thus the entire
Laws are a form rules that prevent certain behavior and information system. They need to have moral and ethical
actions happen. They are created from ethical structures. The conducts, if they don’t have these conducts they will make
major difference between law and ethics is: law needs an information system vulnerable to threats. There is no efficient
authority to process and ethics don’t [2]. Ethical behavior technical protection of information security because people
comes from individual’s conscience. Ethics are defined in the are involved [4]. People are critical on ensuring a robust
____________________________________________________________________________________________________________
141
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
information security. So there comes ethics to rescue. People There is a control gap in information systems and
that act with ethical conscience in information systems ensure it is getting wider with every new technology. This
information security. control gap can cause problems in retaining
security in information systems.
IV. SOME ETHICAL CONCERNS OF INFORMATION SECURITY Ethics must be a common language among
A. Hacking and Computer Crime individuals from different expertise. Ethics should
Computer security mainly interested in protection of also understandable by individuals from other
computer assets and important data against leaks and communities besides computer society.
unauthorized access. These leaks and unauthorized actions are Nowadays information systems are getting so
called hacking. Hackers are computer users that gain large that it is getting harder to manage with only
unauthorized access to a system and share knowledge to other technological control mechanisms. So control
users or hackers in the process. As it sounds a negative action, mechanisms should be built on individuals control
hacking can be beneficial in some ways [3]. White hat hackers mechanism that can be achieved by ethical
or self-identified hackers suggest that their actions cause no frameworks.
harm to community. On the contrary they claim that they help
development of a better security system. Because they think If these problems don’t answered right there can be an
hacking can be used to release data to benefit all in a opening that leads to information security issues.
community [3]. They’ve even developed a code of ethics for As seen from given perspectives complete security of
hackers. They are suggesting that hackers should use a code of information can only be achieved by the help of ethics and
ethics when hacking. There is a gray area in this manner some ethical frameworks.
researchers think hacking can’t be ethical, some claim VI. FRAMEWORKS FOR ESTABLISHING AN ETHICAL BASELINE
otherwise [6]. This is an ethical issue that information security IN DIFFERENT ENVIRONMENTS
researchers argue on.
Many researchers stated that information security is lacking
B. Privacy & Ethics without an ethical concept. So in order to establish a better
security, they proposed frameworks that include ethical
There is no precise definition for privacy but privacy term principles. These frameworks exist in many information
mainly refers to “right to be left alone” [7]. Privacy doesn’t environments like biomedical, e-banking and health and etc.
only apply to personal data. It also concerns with human Example frameworks will be discussed in this chapter to
relationships, private belongings, actions and even our homes. emphasize the importance of ethical principles in information
Ethical issues arise in many areas of privacy including security.
surveillance, medical privacy, internet privacy and work France [10] discussed ethical violations in biomedical area
privacy [3]. that leads to violation of patient privacy and medical records.
He also stated some ethical codes and suggestions to
V. DIFFERENT ETHICAL PERSPECTIVES IN INFORMATION overcome this security issue. These ethical codes include
SECURITY doctor and hospital staff jurisdictions, patient record storage,
There are different ethical views in information security. etc.
Researchers proposed several models to achieve better Abreu and co-workers [11] discusses fraud (phishing) in e-
information security. banking services. They point out several threats,
Hartmann [8] stated that security levels couldn’t consist of vulnerabilities, incidents and impact of threats on e-banking
only a technical layer. It consists of technical, technological, services. Researchers suggested ethical rules and trainings on
organizational, legal, social and ecological levels. In addition clients and bank personnel to overcome frauds in e-banking
to these levels Hartmann suggests ethical scope must be services. They also stated that Public discussion on incidents
included in information security. The researcher also suggests will develop awareness and creates a security behavior among
that ethics is a so large question that developers, users and e-banking service users.
system admins cannot answer separately. Entire community Kluge [12] developed a code of ethics for health
should answer this question by discussing together. So information professionals. He discussed several problems in
Hartmann suggests that ethics should span an entire health informatics domain and proposed code of ethics for
community and ethical rules should be prepared with a several information security problems. He suggested that in
mixture of individuals from different areas. addition to technical security layer in a health information
Another ethical view in information security is proposed by system there must be an ethical layer that protects privacy of
Kowalski [9]. Kowalski stated several ethical problems and patients. He created this ethical layer on health information
also stated that information security is threatened if these professionals.
problems occur in information technology systems. These
problems are as follows; VII. MAKING ETHICAL RULES EFFECTIVE IN COMMUNITY
Making ethical rules effective in a community plays an
important role in security of information. Being able to give
____________________________________________________________________________________________________________
142
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
individuals a set of ethical rules to follow and provide an a standard there is a need for collaborative effort [16].
awareness of ethics will surely establish a better security layer Awareness training helps to establish this collaborative effort.
in organizations. To provide a better and robust security of
information, there are two ways: Developing a code of B. Awareness Training
conduct/ethics and provide trainings to individuals so they can
gain awareness in security and ethics. Awareness training and security training takes a major part
in establishing security. In order to provide security there must
A. Code of Conduct be awareness among individuals. This awareness spans to a
large definition including; ethics, threats and how to deal with
Nowadays many organizations and communities develop them, security incidents and so on.
code of conduct for its members to follow [13], [14]. Without Awareness training must provide an appropriate security
the development of these codes information security will have and ethics behavior to all individual in an organization in
gaps because of laws couldn’t fill some gaps and ethics can. order to full commitment to security policies [17]. There are
For example; in order to fill these gaps, computer ethics some research that focuses on awareness training and how it is
institute developed Ten Commandments of Computer Ethics effective in providing security. Stephanou and Dadaga [17]
[15]: have asked the question “to what extent does information
security awareness training influence information security
1. Thou Shalt Not Use A Computer To Harm Other behavior?” and they discuss existing awareness training
People. research and proposed a model to examine the impact of
training on security behavior among individuals.
2. Thou Shalt Not Interfere With Other People’s Aliyu and co-workers [18] conducted a study that inspects
Computer Work. security and ethics behavior among students. They conducted
several experiments among students to find out that is
3. Thou Shalt Not Snoop Around In Other People’s awareness actually affects security. The experiments showed
Computer Files. that the awareness training created a conscience of security
among students. Because of security awareness trainings and
4. Thou Shalt Not Use A Computer To Steal. ethics courses students are more aware of security concerns
and ethics than the students who receive no training.
5. Thou Shalt Not Use A Computer To Bear False So according to these studies that we mentioned above, we
Witness. can say that awareness training actually can affect the security
behavior of individuals. Also they are crucial in making
6. Thou Shalt Not Copy or Use Proprietary Software for ethical rules effective in community. Because this training
Which You have Not Paid. also gives a conscience of ethics to individuals and in our
opinion is crucial to establish information security.
7. Thou Shalt Not Use Other People’s Computer
Resources without Authorization or Proper Compensation. VIII. CONCLUSIONS
Ethics play an important role in our lives and also in cyber
8. Thou Shalt Not Appropriate Other People’s Intellectual technology domain. Ethics fill gaps in an information system
Output. that laws can’t be able to fill. In this paper the importance of
ethical principles in information security has discussed.
9. Thou Shalt Think About The Social Consequences Of Different ethical perspectives in literature are inspected to
The Program You Are Writing Or The System You Are show how ethical layer can be built on security layer. Also in
Designing. this paper, to provide a solid proof that ethics complete
information security, several ethical frameworks are inspected.
10. Thou Shalt Always Use A Computer In Ways That Finally, methods to make ethical rules effective in a
Insure Consideration And Respect For Your Fellow Humans. community are given. Several examined studies showed that
awareness training and code of conducts are effective in this
This code of ethics guides individuals when interacting manner.
with an information system. Code of ethics is applied to all of
the members in an information system including system REFERENCES
administrators, developers, users and security professionals. [1] H. Taherdoost , S. Sahibuddin, M. Namayandeh and N. Jalaliyoon,”
Instead of applying general code of ethics, organizations can Computer and Information Security Ethics’ Models”, International
Conference on Advanced Computer Science Applications and
develop their own code of conducts. In our opinion this can be Technologies, pp. 145-149, 2013.
prove to be most effective because to target and fill specific [2] M. E. Whitmann and H. J. Mattord, Principles of Information Security,
gaps in information security, specific ethical codes must be 4th edition, Course Technology Cengage Technology, 2012.
defined. But in order to make users abide code of ethics and to [3] P. Brey, “Ethical Aspects of Information Security and Privacy”, in:
Security, Privacy, and Trust in Modern Data Management, M.
____________________________________________________________________________________________________________
143
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
144
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The process for obtaining information that will The analysis and interpretation of large-scale data and
create value on a large-scale data stack is called data mining by its uncovering the relationships between data are difficult and long
general name. Data mining is commonly used in sales and processes by human labour and talent. Today, performing
marketing departments, in determining strategies and making knowledge retrieval using data mining is commonly used in
critical decisions for the future in many sectors. Similarly, data many fields such as marketing, banking, stock market,
mining is used in the determination of health policies, more
communication systems, education, health and engineering.
effective implementation of health services and in the management
of resources and institutions in the health sector. In the marketing field, an attempt to bring out user patterns
In this study, it was aimed to create a software architecture of using data mining was made by Ünal et al. in order to determine
data mining that will help the personal monitoring of the the customer portfolio and to describe the user's shopping
pregnancy process in a more effective way in the health sector. behaviors by using data of an internet retailer [3].
Many different types of data such as age, gender, location, In the banking field, Tosun et al. tried to determine the
education, physical characteristics, lifestyle habits and medical customer losses and the reasons of the losses by constructing
history of the people that could be used for this purpose are stored the profile of the 30,000 customers of Yapı Kredi Bank [4].
online by health institutions. The machine learning algorithms Between the years 2007-2013, Öcal et al. used data mining
have been created to determine classification, clustering and
to make a qualitative classification according to the financial
association rule on these data.
success or failure of the companies and the capacity situations
Keywords— Data Mining, Knowledge Mining, Healty of the companies by using the data of Borsa İstanbul [5].
Monitoring, Classification, Clustering, Machine Learning, A large telecommunication company's customers who
Association Rule showed a tendency to leave were determine by Gürsoy in 2010,
and new campaigns for these customers and marketing
I. INTRODUCTION strategies for the customer profile were developed by
classification method [6].
The data generated by the computer systems do not make a
In the research carried out by Gülen at al. on gifted children
sense by themselves, and their conventional analysis by human
aged 7 years and older in Ankara science and art center, the
intervention is also impossible. Today, the diversity of data
education for the children's individual needs was differentiated
generated in fields such as stock market, trade, education,
through educational data mining and the course schedules were
communication and e-commerce in addition to the fact that
organized according to children's field of interest [7].
these data are large by volume and continue to increase are the
In the engineering field, Kaya et al. compared the open
main reasons of this.
source data mining softwares Keel, Knime, Orange, R,
In today's world, obtaining valuable information by
RapidMiner (Yale) and Weka. In the study, which data mining
interpreting these data by means of intelligent algorithms is
method and which software would be more effective on which
called data mining. In other words, data mining is the discovery
data set was investigated [8].
and interpretation of the significant relationships through
Tantuğ et al. examined the techniques used in data mining
computer programs that could be effective in providing future
and especially investigated the clustering techniques. An
predictions from among large data stacks [1, 2].
efficient software architecture to reduce memory complexity
Data mining has a great importance in many aspects from
was designed in the study. An efficient system with a lower
sales strategy of businesses to determination of Research and
complexity capable of clustering on large-scale data was
Development activities.
developed by the software architecture designed [9].
____________________________________________________________________________________________________________
145
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
146
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
clusters and sub-clusters are generated based on the similarities Other: The personal data such as the person's income
and differences between variabilities. level and nutritional habit apart from these were used by
being converted into integer values by grouping once
3) Determining the Association Rule: It is a data mining
again.
method which is applied on the possibility of which situations
can occur simultaneously by identifying the relationships of the TABLE III
TRAINING SET DATA TYPES AND THEIR NORMALIZATION
data included in the database with each other.
Data Data Normalized Data Sample Data
III. PROPOSED METHOD Type 1 2 … N
In this study, expectant mothers' information shown in Fig. Age [17] Integer 0-19: Adolescent
2 were used and the process of knowledge retrieval was [0-2] 19-35: Adult 0 1 2
performed using data mining on these data. >35: Menopause
The software architecture of the proposed method, the block Body Float <18.4:
diagram of which is given in Fig. 2, works in four stages. In the mass [0-3] Underweight 1 1 2
index [17] 18.5-24.9: Normal
first step, a no-sql based database was created for storing data.
25-29.9:
In the second step, preprocessing was performed to extract Overweight
information on data set. In the third step, classification, >30-34.9: Obesity
clustering and association rule were identified by running three Number Integer
separate data extraction algorithms. In the fourth and final step, of babies [0,n] 2 1 2
estimation on new data, identifying the cluster to which it Number Integer
belongs and the process of finding the other data to which it is of [0,n] 1 0 … 1
related were performed. abortions
Number Integer
of natural [0,n] 0 0 2
birth
Number Integer
of [0,n] 2 1 0
cesarean
Prosperity Integer 50 25 60
Index [0,100]
Disease Boolean
{0,1} 0 1 0
…
Fig. 2 Block diagram of the proposed system Dietary String
[0,n] 2 1 3
In the study, expectant mothers' information shown in Fig. 2
were used and the process of knowledge retrieval was Matlab and python (scikit library) are mostly used for data
performed using data mining on these data. The sample training mining studies in computer environment. In addition to these,
set, main data types and the normalization of the data type into there are also various tools which are free and open source and
a certain range if necessary are given in Table II. also include sample data-sets [8, 19]. However, a data-set was
All values shown in Table II were converted to integer generated in a synthetic dataset matlab environment in
numbers by normalization, if needed, depending on the primary accordance with Table II because there was not any dataset
data types. The values of the sets obtained after normalization specific to this study, and it was saved in json format as no-sql
constituted the input data of the training algorithm. According based.
to what some values are normalized is explained as follows. In the sample application scenario, random forest for
classification and k-means for clustering were used [18]. The
Body mass index: It is a value calculated based on height
target class labels chosen for classification in the training
and weight, and it is divided into 3 groups in practice [17].
algorithm were identified as “healthy pregnancy” and “risky
Age: It is normalized to 3 different groups by considering
pregnancy”. In the test phase, the results were analyzed by
the fertility criteria [17].
achieving classification accuracy according to "1" based on the
Disease: Whether the person has a disease is discussed as confusion matrix given in Table III.
a Boolean value.
Pregnancy History: It includes the statistics of the
TABLE III
person's pregnancy history, if available. They are
EXPERIMENTAL RESULTS
numerical data such as the number of children if she had
a baby, whether the previous pregnancies were Actual Class
concluded with healthy results, cesarean and the number Prediction P N
of normal delivery. Class P TP FP
N FN TN
____________________________________________________________________________________________________________
147
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] S. Savaş, N. Topaloğlu and M. Yılmaz, “Veri madenciliği ve
Türkiye’deki uygulama örnekleri”, 2012.
[2] X. Wu, X. Zhu, G. Q. Wu and W. Ding, “Data mining with big data”,
IEEE transactions on knowledge and data eng., 26(1), 97-107, 2014.
[3] O. O. Ünal, “Internet Kullanım Analizi Ve Kullanıcı Betimleme
Konularında Veri Madenciliği Uygulamaları”, Doctoral dissertation,
Fen Bilimleri Enstitüsü, 2015.
[4] T. Tosun, “Veri Madenciliği Teknikleriyle Kredi Kartlarında Müşteri
Kaybetme Analizi”, Doctoral dissertation, Fen Bilimleri Enstitüsü, 2015.
[5] N. Ocal, M. K. Ercan and E. Kadioglu, “Predicting Financial Failure
Using Decision Tree Algorithms: An Empirical Test on the
Manufacturing Industry at Borsa Istanbul”. International Journal of
Economics and Finance,7(7), 189, 2015.
____________________________________________________________________________________________________________
148
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract — Nowadays, extending the lifetime of WSNs through The traditional infrastructure based protocols cannot be used
energy efficient mechanisms has become a challenging research for WSNs because of the limited storage and power of the
area. Previous studies have shown that instead of implementing sensor nodes. Because WSNs often use broadcast based radio
direct transmission or multi-hop routing, clustering is a kind of communication.
fundamental technique used to decrease energy consumption.
Limited energy availability in sensor nodes makes network
Clustering can increase the scalability, decrease the energy
consumption and prolong the lifetime of the network. In lifetime an important issue in WSN applications. To extend the
literature, LEACH and its variants aim to use clustering network lifetime, energy efficient wireless sensor network
mechanisms to provide energy-efficiency. However, most of the protocols and algorithms have been devised in the literature.
LEACH variants aim to form clusters in each round by changing Node clustering, in-network data processing, data fusion and
CHs randomly. These formations cause to consume large amount network coding are some of the measures taken to reduce the
of energy and bring additional network costs. Also, in some rounds amount of data that is processed, sensed or transmitted.
of LEACH variants none cluster formations may occur because of Minimization of energy spent in processing, sensing and
the probabilistic CH selection process of these protocols. In this transmission of data allows sensor nodes to save energy. Such
paper, an energy-efficient random selected constant clustering
energy savings help to extend the lifetime of WSN applications.
approach is proposed to solve the problems of LEACH based
protocols. The proposed approach uses constant clusters which The main goal of cluster-based routing protocol is to
are formed only once at the beginning of the algorithm. The efficiently maintain the energy consumption of sensor nodes by
formed clusters remain fixed until all nodes are dead in the involving them in multi-hop communication within a cluster
network. Proposed approach aims to select CHs in each cluster and by performing data fusion in order to decrease the number
randomly without changing the cluster formations. It aims to of transmitted packets to the base station (BS) and transmission
reduce cluster formation packet overhead in the network. In distance of sensor nodes.
addition, proposed approach aims to provide energy-efficiency by Low energy adaptive clustering hierarchy (LEACH) [1-2] is
using fixed clustering. The results of the simulations show that, one of the most popular distributed cluster-based routing
constant clustering approach saves extra energy and prolongs
protocols in WSNs. LEACH randomly selects a few nodes as
lifetime of the WSN when compared to LEACH and ModLEACH.
CHs which aggregate data arriving from nodes and forward the
aggregated data to BS and rotates this role to balance the energy
Keywords— Energy Efficient Routing, Fixed Clustering, Wireless
consumption of the sensor nodes in the network.
Sensor Networks.
In LEACH protocol, the time is divided into parts called a
I. INTRODUCTION round. Each round consists of two phases. The first phase is set-
up phase which is the phase of node formation. The second
A WSN contains sensor nodes and corresponding protocols
phase is related to the normal function of the network and is
for routing messages in this network. A typical wireless sensor
called the Steady-State phase. In the first phase, the CHs are
node includes a sensory unit, a communication unit, a power
elected based on a probability function. This election is as
unit and a processing unit. Sensory unit consists of a data
follows: each sensor node selects itself to be CH at any given
acquisition component and ADC (converts the sensed real
time with a certain probability. Any node in the network
world data to the digital form). Communication unit has a radio
chooses a random number between 0 and 1. Then this number
transceiver and the power is backed by a battery source. The
is compared with threshold limit, if the number is less than a
WSN node can be equipped with a limited power source (0.5
threshold, the node becomes a CH for the current round. This
Ah, 1.2 V). The node remains active as long as the battery is
probability function is designed in such a way that within a
alive and hence power saving is a crucial criterion in this
specific number of rounds each sensor becomes a CH only once
domain of applications. Energy consumption happens in three
and thus the energy consumption is distributed over the whole
domains: sensing, data processing and communications. The
network. After the set up phase of the round, where the CHs are
sensing, signal processing parts operate at low sequential and
elected, each CH announces its election to other nodes and each
consume less than 1mW. This is over an order of extent less
node chooses a suitable (nearest) CH for itself; and then it
than the energy consumption of the communication part.
announces this decision to the related CH and thus the clusters
are formed and the network comes into the steady-state
____________________________________________________________________________________________________________
149
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
operation i.e data transmission. Then each CH creates a TDMA networking and set-up costs of LEACH based protocols. The
schedule in each cluster to organize the communication among nodes in the network are static, thus transmission of position
cluster members. When the non CH nodes receive the TDMA information is realized only once.
scheme, nodes send their data to the CH once per frame during When the CHs are selected, they broadcast CH notification
their allocated transmission TDMA slots. This allows the radio messages to the WSN. Each cluster member joins nearest
components of each non-cluster-head node to put the sleep cluster. After this joining process, the clusters are formed. In
mode at all times except during its transmit time, thus each cluster, CH is responsible for receiving data from its
minimizing the energy dissipated in the individual sensors. members. When data collection is completed, each CH delivers
After a certain amount of frames i.e. a TDMA round, the collected data to the BS which is placed outside of the sensor
network re-elects CHs and re-form clusters. Then CH, after field. In EERCCA, each CH node selects randomly new CH
combining all the received data, will send the results to the BS. node from alive members in the cluster for every round. After
Besides, to prevent the interaction of the transmissions which this selection, new elected CH starts to collect data from cluster
occur in clusters at the same time, LEACH uses different members. Then this new selected CH transmits the collected
CDMA codes. Existing CH chooses randomly a unique code data to the BS.
from a list of spreading codes. The CH filters all received
energy using this spreading code. Consequently, the radio III. SIMULATION ENVIRONMENT AND PARAMETERS
signals of the neighbouring nodes are filtered out hence The simulations are conducted in MATLAB. 100 sensor
interference of the transmission of the nodes is minimized. nodes are randomly deployed in a 100 m x 100 m field as shown
In some studies, authors have tried to improve the in Fig. 1 and BS is placed outside of the sensor field which have
performance of LEACH. They have designed some protocols coordinates of (150,50). Same simulation parameters of
which made extensions to LEACH.
LEACH are used for all simulations in this paper. Each
Modified LEACH (ModLEACH) [3] includes an efficient
simulation is realized for 100 independent iterations to obtain
CH replacement scheme and dual transmitting power levels.
The amplification energy is set to be the same for all kinds of more scalable results. Table I summarizes the simulation
transmissions in LEACH. But with ModLEACH, low energy environment parameters used for simulations.
level is used for intra cluster communications. Multi power
levels are used to reduce the packet drop ratio, collisions and
interference from other signals. When a node becomes a CH,
the routing protocol in ModLEACH informs it to use high
power amplification and when a node becomes a regular cluster
member, the mode of that node becomes low level power
amplification mode. Threshold based CH changing mechanism
is used in ModLEACH to provide more efficient CH
replacement. If the energy of the existing CH is higher than the
threshold it continues to act as a CH if not a new CH for that
cluster is elected and the cluster is formed again. Simulations
of ModLEACH show that, ModLEACH outperforms LEACH
in terms of throughput, network lifetime and CH formation.
In this paper, Energy Efficient Random Selected Constant
Clustering Approach (EERCCA) is presented. It aims to use
fixed clustering approach. It determines CHs randomly at the
beginning of the algorithm. The formed clusters remain Fig. 1. The randomly deployed sensor nodes in a WSN.
constant until the batteries of all sensor nodes are drained. In
each round, the CHs are changed randomly but the clusters TABLE I. SIMULATION ENVIRONMENT PARAMETERS
remain fixed. EERCCA aims to solve the network cost issues
which occur in new cluster formations of LEACH based
Parameters Values
protocols. The performance of EERCCA algorithm is
Network area 100 m x 100 m
compared with LEACH and ModLEACH in terms of lifetime,
residual energy and throughput. Number of nodes 100
Base station coordinates (150,50)
II. EERCCA ALGORITHM Initial energy per node 2J
At the beginning of EERCCA algorithm, the sensor nodes Data packet size 6400 bits
are randomly deployed in a field as shown in Fig. 1. EERCCA Control Packet Size 200 bits
algorithm starts with the CH selection process. The CHs are Transceiver Energy (Eelec) 50 nJ/bit
selected randomly at the beginning of the algorithm. The Aggregation Energy per Bit (EDA) 5 nJ/bit/signal
clusters are formed only once in the clustering phase. The fixed Free Space Amplifier Energy (εfs) 10 pJ/bit/m2
clustering approach of EERCCA provides to decrease the Multipath Amplifier Energy (εmp) 0.0013 pJ/bit/m4
____________________________________________________________________________________________________________
150
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
A. Residual Energy
Fig. 2 illustrates the total residual energy of the nodes in
terms of rounds. With fixed clustering, it can be observed that
EERCCA provides significant energy savings when compared
to LEACH and ModLEACH. After 2500 rounds, while
LEACH holds 5% of its initial total energy, EERCCA holds
approximately 20% of their initial total energy. When the
network lifetime ends under LEACH, EERCCA still saves Fig. 3. Round Number vs Alive Nodes
approximately 5% of its total initial energy. EERCCA also
outperforms ModLEACH in terms of energy-efficiency. C. Network Lifetime
____________________________________________________________________________________________________________
151
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. CONCLUSIONS
Nowadays, providing energy-efficiency, prolonging battery
life span of the sensor nodes and designing green network
components [4, 5] for WSNs has become a promising research
area. By using constant cluster formations, EERCCA
minimizes the cluster formation overhead. When compared
with LEACH and other protocols significant improvements are
achieved in terms of energy usage, network lifetime and
throughput. EERCCA prolongs the lifetime of the WSN, while
reducing the energy usage. Also, the throughput performance
of EERCCA makes it more possible candidate to be used in
critical applications such as fire resque, eartquake
measurements and disaster relief.
References
____________________________________________________________________________________________________________
152
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The classification of data on the internet in order to for telnet traffic. Especially with the widespread use of point to
make internet use more efficient has an important place especially point (P2P) applications, this method has started to lose his
for network administrators managing corporate networks. functionality as some applications use non-standard port
Studies for the classification of internet traffic have increased numbers to escape from firewall and network security tools and
recently. By these studies, it is aimed to increase the quality of
some use port hiding and dynamic port methods [10-13].
service on the network, use the network efficiently, create the
service packages and offer them to the users. The first Payload based classification is based on the principle of
classification method used for the classification of the internet Internet traffic classification by analysing TCP/UDP packet
traffic was the classification for the use of port numbers. This loads. The analysis of loads is performed by determining
classification method has already lost its validity although it was whether the known applications contain characteristic
an effective and quick method of classification for the first usage signatures [10, 14].
times of the internet. Another classification method used for the When the packets are not encrypted, it works quite
classification of network traffic is called as load-based successfully. However, because of the following reasons, this
classification or deep packet analysis. This approach is based on classification technique is not much preferred today:
the principle of classification by identifying signatures on packets
• It causes privacy and security concerns,
flowing on the network. Another method of classification of the
internet traffic which is commonly used in our day and has been • Some applications communicate by using encrypted
also selected for this study is the kernel based on extreme learning packets,
machine based approaches. In this study, over 95% was achieved • It can only make assessment based on the signatures
accuracies using different activation functions. which are experienced by the previous classification methods,
• As it requires high processing and storage capacity, it
Keywords— machine learning, classification, extreme learning is not suitable for real-time classification [11-13].
machine, network classification, kernel activation function The network classification method via machine learning
algorithms is the most popular traffic classification method at
I. INTRODUCTION the present time. In the studies, the classification is usually
Traffic classification methods are used to provide the performed using supervised and unsupervised learning
efficient realization of the data traffic on network resources, to algorithms. Supervised learning algorithms perform
do user analysis by using network data, to manage and plan classification by using the classification analysis methods in
network resources, to detect the attacks and the abnormalities data mining; and unsupervised learning algorithms perform
on the network [1, 2]. Recently, the network traffic classification by using the clustering analysis methods.
classification has been frequently used in order to improve Machine learning algorithms perform the process of network
service quality in big networks [3, 4], use the network traffic classification in two steps. In the first step, it forms a
effectively, develop new service packets and perform internet classification model; and in the second step, it performs the
traffic analysis [5]. Internet traffic analysis can be done both classification. Statistical methods and calculations are usually
on-line and offline. In the on-line traffic analysis, each data utilized when performing the classification process.
packet on the network are captured and analysed [6, 7]. In the Machine learning based classification method uses the
offline traffic analysis, network traffic flow is firstly captured following TCP and UDP statistical attributes of the flow during
and stored; then the stored flow is analysed and classified [8]. the flow-based classification:
In this paper, offline traffic analysis was performed. In the • Total size statistics,
literature, three kind of classification technique, including port, • Total number of forward and backward packets,
payload and machine learning based, have been used. • Total amount of forward and backward byte,
Port based classification is performed by comparing the port • The transit time between packets,
information retrieved from flow data with the port numbers of • And flow time.
protocols determined by Internet Assigned Numbers Authority There are many studies on the field of Internet traffic
(IANA) [9]. For instance; port 80 is used for http, and port 23 classification by using machine learning. In their study, L.
____________________________________________________________________________________________________________
153
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
154
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
represents the output weight matrix, T is the matrix that f ( x) k ( x, x1 ), , k ( x, x N ) M T (14)
consists of output labels for the N data samples.
C
To train a SLFNs as mentioned in (2), it is equivalent to finding T
Where M HH and k ( x, y ) is the kernel function of
the least-square solution of linear system (5), that is:
hidden neurons of single hidden layer feed-forward neural
networks. There are many kernel functions satisfying the
|| Hˆ T || min || H T || (8) Mercer’s condition available from the existing literature, such
as linear kernel, polynomial kernel, Gaussian kernel, and
In the case that the number of hidden nodes is coequal to the exponential kernel. In this paper, we use Radial Basis (RBF)
number of different training samples N , it is possible to find a and Polynomial kernel function for performance analysis.
̂ such that the training error reaches zero. The hidden layer III. EXPERIMENTAL STUDIES
output H is an invertible square matrix. Hence the solution of Moore et al [34] used the data received from the Cambridge
the linear system can be given as: University campus in this study. The most important factor in
this selection is to enable to make comparison of studies
previously carried out by using these data with the methods
ˆ H 1T (9)
used in this study. One of the important factors is the use of data
having flows belonging to different classes. In addition, the use
In the case that the number of hidden nodes is less than the of existing data that everyone can reach will provide a basis to
number of distinct training samples N , to achieve the smallest get more reliable results.
training error || H T || , the solution of the linear system (5) 12 features were chosen from output data. The chosen
features and their explanations are illustrated in Table 1.
can be obtained as:
TABLE I
FEATURES AND EXPLANATION
ˆ H T †
(10)
No Explanation
1 Server port
Where H † is called the Moore-Penrose universalized inverse
[32]. 2 Client port
3 Actual data packets (from client to server)
The least squares solution of (8) based on Karush-Kuhn-Tucker 4 Pushed data packets (from client to server)
(KKT) conditions can be written as
5 Pushed data packets (from server to client)
1 6 Min segment size (from client to server)
1
ˆ H † HH T T (11) 7 Average segment size (from server to client)
C 8 Initial windows bytes (from client to server)
9 Initial windows bytes (from server to client)
where H is the hidden layer output matrix, C is the regulation
10 RTT samples (from client to server)
coefficient, and T is the expected output matrix of samples.
11 Median data IP(from client to server)
Then, the output function of the ELM learning algorithm is
12 Variable data wire (from server to client)
1
1
f ( x ) h( x ) H T HH T T (12) Among the classes belonging to the flows, 7 most
C commonly used ones were chosen; and 1000 learning processes
and 750 tests were applied to each classes. Totally, 7000 classes
If the feature mapping h(x) is unknown and the kernel matrix
were chosen for the learning process and 5250 for the test.
of ELM based on Mercer’s conditions can be defined as follows Chosen classes are illustrated in Table 2.
[33]:
____________________________________________________________________________________________________________
155
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE II
CLASSES AND DATA SETS
TABLE IV
No Class Explanation Training Test KERNEL POLYNOMIAL FUNCTION
1 Attack worm, virus 1000 750 Parameter1 Parameter2 Accuracy (%)
2 P2p bittorrent, 1000 750 1 1 75.14
3 Mail pop3,smtp 1000 750 0.1 0.1 64.88
4 Www http, https 1000 750 1 0.1 31.79
5 Services dns, ntp 1000 750 0.1 1 75.09
6 Bulk ftp, ssh 1000 750 0.01 0.1 60.84
7 Database mysql, 1000 750 0.001 1 75.45
10 10 92.69
The success measurement of classification by machine 100 100 18.63
learning algorithms can be examined according to the 1 100 21.14
evaluation table -in Fig. 2- including confusion matrix for
1 10 93.07
classification algorithms, and the evaluation metrics [7].
1 100 26.06
100 100 21.31
IV. CONCLUSIONS
The classification of data on the internet in order to make
Fig. 2. Evaluation Metrics internet use more efficient has an important place especially for
network administrators managing corporate networks. Studies
In Figure 6, lines indicate the actual value of the example;
for the classification of internet traffic have increased recently.
and columns of matrix indicate estimated values which were Machine learning methods of classification of the internet
classified or clustered. Accordingly, above-mentioned metrics traffic which is commonly used in our day and has been also
can be defined as follows: selected for this study is the kernel based on extreme learning
True Positive (TP): The number of examples which machine based approaches. In this study, over 95% was
actually belong to class X and are correctly estimated to be achieved accuracies using different activation functions. In this
in class X. study, the kernel based ELM function; Radial Basis and
False Positive (FP): The number of examples which don’t Polynomial functions are used. Functions classification
belong to class X but estimated to be in Class X. performance by changing the parameters used were observed.
True Negative (TN): The number of examples which RBF is used for the first parameter value. Polynomial function
actually don’t belong to class X and estimated not to be in for changing the classification is made and 2 parameter values.
class X. With a decrease of the parameter value used for RBF was found
False Negative (FN): The number of examples which that the accuracy increases. This parameter value is 0.01 and
belong to class X but estimated not to be in class X reached a value of 95.10% accuracy. With 0,001 of these
parameters have reached a value of 96.27% accuracy. But the
increase has been observed that a lot of work time. 2 parameter
TP + TN value for Polynomial function is used. The value of 1 and 10
Accuracy = (17) Accuracy rate of 93.07% was observed with the election.
TP + TN + FP + FN
REFERENCES
[1] K. Xu, F. Wang and L. Gu, Behavior analysis of internet traffic via
TABLE III
bipartite graphs and one-mode projections. Networking, IEEE/ACM
KERNEL RADIAL BASIS FUNCTION
Transactions on, 22(3), (2014) 931– 942.
Parameter Accuracy (%) [2] M. Roughan, S. Sen, O. Spatscheck and N. Duffield, Class-of-service
mapping for QoS: a statistical signature-based approach to IP traffic
100 74.57 classification, Proceedings of the 4th ACM SIGCOMM conference on
Internet measurement, (2004), 135 – 148.
10 79.06
[3] X. Zang, A. Tangpong, G. Kesidis and D.J. Miller, Botnet detection
1 88.55 through fine flow classification, Departments of CS&E and EE, The
Pennsylvania State University, University Park, PA, Report No. CSE11-
0.1 92.38 001. 2011.
0.01 95.10 [4] I. Ismail, M.N. Marsono and S.M. Nor, Detecting worms using data
mining techniques: learning in the presence of class noise. In Signal-
0.001 96.27
____________________________________________________________________________________________________________
156
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Image Technology and Internet-Based Systems (SITIS), 2010 Sixth [20] S. Agrawal and K. Singh, Feature Extraction based IP Traffic
International Conference on. (2010) pp. 187-194 Classification using Machine Learning, Proceeding of the International
[5] M. Soysal and E.G. Schmidt, Machine learning algorithms for accurate Conference on Advances in Computing and Artificial Intelligence,
flow-based network traffic classification: Evaluation and comparison. (ACM New York, NY, USA, 2011), pp. 208-212.
Performance Evaluation, 67(6) (2010), 451-467. [21] T. Karagiannis, K. Papagiannaki and M. Faloutsos, BLINC: Multilevel
[6] T. Karagiannis, K. Papagiannaki and M. Faloutsos, BLINC: multilevel Traffic Classification in the Dark, in Proceeding of the Special Interest
traffic classification in the dark. In ACM SIGCOMM Computer Group on Data Communication Conference, (Philadelphia, USA, 2005)
Communication Review 35(4) (2005), pp. 229-240. [22] L. Bernaille, R. Teixeira, I. Akodkenou, A. Soule and K. Salamatian,
[7] T.T. Nguyen and G.A. Armitage, survey of techniques for internet traffic Traffic Classification On The Fly, ACM Special Interest Group on Data
classification using machine learning. Communications Surveys & Communication Computer Communication Review, 36(2) (2006).
Tutorials, IEEE, 10(4) (2008), pp. 56-76. [23] G.B. Huang, Q.Y. Zhu and C.K. Siew, Extreme learning machine:
[8] A. Callado, C. Kamienski, G. Szabo, B. P. Gero, J. Kelner, S. Fernandes theory and applications. Neurocomputing, 70(1) (2006).
and D.Sadok, A Survey on Internet Identification, IEEE [24] G.B. Huang and L. Chen, Convex incremental extreme learning machine,
Communications Survey and Tutorials, 11(3) (2009), pp 37-52. Neurocomputing, 70(1) (2007), pp.3056–3062.
[9] IANA, Internet Assigned Numbers Authority [Online] Available: [25] G.B. Huang and L. Chen, Enhanced random search based incremental
http://www.iana.org/protocols extreme learning machine, Neurocomputing, 71(1) (2008), pp. 3460–
[10] T. Karagiannis, A. Broido, M. Faloutsos and K. Claffy, Transport Layer 3468.
Identification of P2P Traffic, IMC’04 Proceeding of the 4th SIGCOMM [26] G.B. Huang, Q.Y. Zhu and C.K. Siew, Extreme learning machine: a new
Conference on Internet measurement, (ACM New York, New York, learning scheme of feedforward neural networks. In Neural Networks,
U.S.A, 2004) pp. 121-134 Proceedings. IEEE International Joint Conference on 2(1) (2004), 985-
[11] F. Dehghani, N. Movahhedinia and M. R. Khayyambashi, Real-time 990.
Traffic Classification Based on Statistical and Payload Content Features, [27] E. Avci and R. Coteli, A new automatic target recognition system based
IEEE ISA, (Wuhan, China, 2010), pp. 1-4. on wavelet extreme learning machine, Expert Systems with Applications,
[12] J. Erman, A. Mahanti and M. Arlitt, Internet Traffic Identification Using 39(16) (2012), 12340-12348.
Machine Learning, IEEE GLOBECOM, (San Francisco, CA, 2006), pp. [28] G.B. Huang, H. Zhou, X. Ding and R. Zhang, Extreme learning machine
1-6. for regression and multiclass classification. Systems, Man, and
[13] L. Yingqiu, L. Wei, L. Yunchun, Network Traffic Classificaion Using Cybernetics, Part B: Cybernetics, IEEE Transactions on, 42(2) (2012),
K-means Clustering, IEEE IMSCCS, (Iowa City, IA, 2007), pp. 360-365. 513-529.
[14] A. W. Moore and K. Papagiannaki, Toward the Accurate Identification [29] G.B Huang, X. Ding and H. Zhou, Optimization method based extreme
of Network Applications, SpringerLink, PAM Lecture Notes in learning machine for classification. Neurocomputing, 74(1) (2010), 155-
Computer Science, 3431(1) 2005, pp. 41-54. 163.
[15] J. Erman, M. Arlitt and A. Mahanti, Traffic Classification Using [30] G.B. Huang, Q.Y. Zhu and C.K. Siew, Extreme learning machine: a new
Clustering Algorithms, MineNet'06 Proceedings of the 2006 learning scheme of feedforward neural networks. In Neural Networks,
SIGCOMM workshop on Mining network data, , (ACM New York, New Proceedings. IEEE International Joint Conference on, 2 (2004), pp. 985-
York, USA, 2006), pp. 281-286. 990.
[16] K. Singh and S. Agrawal, Comparative Analysis of Five Machine [31] J. Luo, C.M. Vong and P.K. Wong, Sparse Bayesian Extreme Learning
Learning Algorithms for IP Traffic Classification, International Machine for Multi-classification, Neural Networks and Learning
Conference on Emerging Trends in Networks and Computing Systems, 4(25) (2014), 836-843.
Communications (ETNCC), (Udaipur, Rajasthan, India, 2011), pp. 33- [32] K.S. Banerjee, Generalized inverse of matrices and its applications,
38. Technometrics, 1(15) (1973), 197-197
[17] N. Williams, S. Zender and G. Armitage, Evaluating Machine Learning [33] B. Li, X. Rong and Y. Li, An Improved Kernel Based Extreme Learning
Algorithms for Automated Network Application Identification, Centre Machine for Robot Execution Failures. The Scientific World Journal,
for Advanced Internet Architectures (CAIA), Technical Report 2014.
060410B, 2006. [34] A. Moore, D. Zuev and M. Crogan, Discriminators for use in flow-based
[18] N. Williams, S. Zender and G. Armitage, A Preliminary Performance classification (Queen Mary and Westfield College, Department of
Comparison of Five Machine Learning Algorithms for Practical IP Computer Science, 2005)
Traffic Flow Classification, ACM SIGCOMM Computer [35] Ertam, F., Avci E., Classification with Intelligent Systems for Internet
Communication Review, 36(5) (2006), pp. 5-16. Traffic in Enterprise Networks. Int'l Journal of Computing,
[19] S. Agrawal and K. Singh, Performance Evaluation of Five Machine Communications & Instrumentation Engg. (IJCCIE) Vol. 3, Issue 1
Learning Algorithms and Three Feature Selection Algorithms for IP (2016) ISSN 2349-1469 EISSN 2349-1477
Traffic Classification, IJCA Special Issue on Evolution in Networks and
Computer Communications, (2011), pp. 25-32.
____________________________________________________________________________________________________________
157
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—In this study, a hybrid security system is proposed. verification and identification. Verification refers the case
The proposed system is composed of two subsystems namely iris where the user desires to be recognized by using an identity
recognition system (IRS) and speaker recognition system (SRS). such as personal identification number (PIN), login name,
Pre-processing, feature extraction and feature matching are the smart card; and the system makes one-to-one comparison to
main steps of these systems. determine whether this claim is true or not. The main goal is
In IRS subsystem, Gaussian filter, Canny edge detector, Hough to prevent multiple people using the same identity.
transform, and histogram equalization is performed for pre- Identification refers the case where the system makes one-to-
processing, respectively. After that, by applying 4-level Discrete
many comparison of user to determine who is exactly the user.
Wavelet Transform (DWT) to pure iris image, the iris image is
decomposed into four sub-bands (LL4, LH4, HL4 and HH4). In The main goal is to prevent a single person from using
order to extract the feature vector from iris pattern, the LH4, multiple identities. One of the key advantages of biometrics
HL4 and HH4 sub-bands (matrices) are merged into one matrix. over the traditional methods such as knowledge based systems
Finally the matrix is transformed in vector to obtain the feature is that biometrics cannot be lost or forgotten; they are difficult
vector of iris image. For SRS subsystem, the pre-processing step for attackers to forge and for user to repudiate [6].
includes spectral arrangement, silence part removing and band Among biometric systems, iris recognition system is very
limitation operations. After pre-processing, frame blocking and popular due to significant reasons such as, uniqueness, non-
windowing are applied to the long-term speech samples and then invasiveness and high stability. Iris is part of human eye
Fast Fourier Transform (FFT) is performed for the each short-
which lies between pupil and sclera. Iris cannot be copied so
term speech segments (frames). Finally, the Mel Frequency
Cepstral Coefficients (MFCC) technique is performed in order to iris recognition system is the most reliable system in
obtain feature vector of the speech. comparison with other biometric systems [2].
The feature matching step of both IRS and SRS is implemented In recent years, speaker recognition also gained increased
with Dynamic Time Warping (DTW) which is an efficient significance. Searching or controlling based on speaker
algorithm to measure the distance between two vectors. identity is a growing interest in today’s technologies. Speaker
According to the DTW results, the false acceptance rate (FAR) is recognition is commonly used in the applications of personal
zero and false rejecting rate (FRR) is about 4 % for the proposed authentication, national security and general forensics. In this
hybrid system. study, speech analysis and decision process is performed by
computer analysis automatically. The main goal of the
Keywords— Biometric systems, iris recognition, speaker proposed speaker recognition system is to increase the
recognition, DTW, security. security of entire hybrid system [5].
I. INTRODUCTION Speaker recognition consists of two main tasks:
identification and verification. In speaker identification we
intend to identify a speaker from a closed or open set of
Nowadays, with the emerging of information technology speakers. For speaker verification, the objective is to verify
the security of digital data has gained a great importance. One the claim of an unknown speaker is true or not. This
way to protect the digital data from unauthorized persons is to verification is performed by comparing the speech samples of
use identification and access control mechanisms. In order to claimer and other speaker’s speech samples available in
provide these mechanisms one emerging technology that database. The characteristic of each speaker voice is typical
becomes more widespread is biometrics. and secret in her/his voice box. So, speaker recognition
Biometric systems are designed based on people’s systems are the reliable biometric systems, too.
physiological or behavioural characteristics. These systems In this study, by connecting iris and speaker recognition
can be defined as a pattern-recognition system which tries to systems in sequence, a hybrid high security biometric system
recognize a person based on feature vector derived from is obtained. Both IRS and SRS are designed by applying pre-
her/his specific physiological or behavioural characteristics. processing, feature extracting and feature matching (decision)
The physiological characteristics include fingerprint, face, steps. Used techniques in order to achieve these steps are
iris, DNA, hand geometry whether the behavioural explained in the corresponding sections. By performing
characteristics include gait, voice. Depending on application, computer simulations, it is shown that the proposed hybrid
a biometric system can operate for two main purposes:
____________________________________________________________________________________________________________
158
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
biometric system (HBS) is more robust against to the intruders the maximum energy of the signal, one more times until we
than separate biometric system. To evaluate the performance get the desired level of decomposition. The middle and high
of the system FAR and FRR values are also calculated in the frequency (LH1, HL1, and HH1) sub-bands represent the
simulations. According to the results, we can say that the details of image such as edges, outline and texture.
proposed systems ensure security requirements. In our study, we used 4-level DWT in order to extract the
The paper is organized as follows. In Section II, IRS and correct features from iris texture. By performing 4-level DWT
SRS are explained with all steps, respectively. A brief we get LL4, LH4, HL4 and HH4 sub-bands. Since the
information about used techniques in these steps are also distinctive features of iris are hidden in the high- frequency
given in Section II. HBS which consist of IRS and SRS is components, we constitute the feature matrix by merging LH4,
introduced in Section III. To evaluate the performance of the HL4, and HH4 sub-bands in one matrix. Finally the matrix is
system, simulation results are given in Section IV. Finally, transformed into a vector to obtain the feature vector of iris
discussion and conclusion part is stated in Section V. image.
II. SUBSYSTEMS
In this section, two subsystems of the HBS are explained in
details. Pre-processing, feature extraction and feature
matching are the main phases of both subsystems.
A. Iris Recognition System (IRS)
Pre-processing: In pre-processing, eye image is transformed
into an appropriate model to extract the distinctive features for
recognition. Iris localization and normalization are two basic (a) (c)
processes in pre-processing. In our study, before localization
and normalization 2-D Gaussian filter is applied for noise
reduction. Then, the edges are detected with Canny edge
detector.
Iris localization is crucial for iris recognition systems. As we
see, there are two contours in an eye. The first one is inner
contour separating the iris andpupil; second one is the outer
contour separating the iris and sclera. By using Circular
Hough Transform (CHT), we obtain the inner and outer
boundaries of iris. In other words, we estimate the centre and (b) (d)
Fig. 1 Pre-processing operations: (a) Original iris image, (b) inner and
radius of contours. outer boundaries of iris (c) Canny edge detector applied to iris image, (d) iris
Iris normalization is an operation converting iris region from template in polar form.
Cartesian coordinates to polar coordinates. Thus, circular
image turns into a rectangular form and now is suitable for
feature extraction. As a last step, histogram equalization is B. Speaker Recognition System (SRS)
used for enhancing the contrast of iris pattern. In Figure 1, In SRS, at the beginning we record the speech of the users
pre-processing is illustrated. under same conditions. Since the SRS is text dependent
Feature Extraction: In order to use iris signature for recognition system, the users say the same words for database.
recognition systems, the representing of iris in an appropriate The block diagram of the proposed SRS system is given in
way is very important. To extract the specific features from Figure 2. The steps of SRS can be explained as follows.
iris, Discrete Wavelet Transform (DWT) is used. DWT is an Pre-processing: In order to process a digital signal, first we
efficient tool, which transforms the signal from time domain must sample the analog signal by Nyquist frequency rate.
into frequency domain without losing time information. Then, pre-processing step including spectral arrangement,
In general, by performing DWT to 1-D signal, the original silence part removing and band limitation operations is
signal is divided in low and high frequency parts. This process applied to the digitized signal.
is called as decomposition and it is provided through special For spectral arrangement, to enhance the high-frequency
analysis filters. The outputs of filter are referred as DWT energy of the signal, we use a high-pass finite impulse
coefficients and the original signal can be reconstructed from response pre-emphasize filter. The transfer function of this
these coefficients by using appropriate synthesis filters filter is given as follows,
(inverse DWT, IDWT). For 2-D signal case, after performing
DWT, image will be divided into four sub-bands 𝐻(𝑧) = 1 − 𝜇𝑧 −1 , 0.95 ≤ 𝜇 ≤ 0.97 (1)
corresponding to low frequency (LL1), middle-frequency (LH1,
HL1) and high frequency (HH1) components, respectively. In silence part removing differentiation operation is
DWT can be applied to the LL1 sub-band which concentrates performed on speech samples from left-to-right and right-to-
____________________________________________________________________________________________________________
159
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
left to eliminate the non-variant (silence) parts. Then The HBS is designed by connecting IRS and SRS
accordance with the human audio system, frequency subsystems as demonstrated in Figure 4. The HBS works as
components outside of 1KHz-5KHz band is stopped. This can follows; at first the user tries to pass through IRS. If IRS
be referred as band limitation. denies his/her access, the system will automatically reject this
attempt. On the other hand, if the user passes through IRS
successfully, SRS is put into use. With the acceptance of SRS,
the user will be able to enter.
cp(X,Y) := ∑𝑈
𝑢=1 𝑐(𝑥𝑛𝑢 , 𝑦𝑚𝑢 ) (3)
____________________________________________________________________________________________________________
160
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
161
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The number of products that simplify people’s lives are Until 2030, health spending is expected to absorb %15 of the
increasing with the enormous development of the technology. gross domestic product of developed countries. These countries
Mobile devices have a great importance for the provision of such as Brazil, Russia, India and China anticipate that 175-200
communication which is one of the most significant need of human billion dollars will be saved in the management of chronic
beings. Mobile devices have gone beyond to be used originally as a
diseases using mobile health applications actively. This means
mobile phone purposes and they have begun to be used as
smartphones by taking in charge of computers. They are not only that the area of the mobile health will gain more importance in
used for communication but also they are used like camera, photo the future.
camera, notebook, television and reminder. Google's Android There are some systems for tracking up chronic illnesses
platform is a widely anticipated open source operating system for developed by mobile operators such as “Turkcell Health
mobile phones. Google’s Android Operating System (AOS) in mobile Monitoring Service” and “Avea Patient Status Tracking System
phones are still relatively new, however, AOS has been progressing with Chronic illnesses” in Turkey.
quite rapidly. The increasing number of smartphone users has In literature, there are numerous examples of mobile
prepared the ground for the emergence of new ideas to make life applications to make people’s life easier. Hacer and Kürşat[2]
easier. Recently, especially some applications in health sector reflect
developed a web based mobile application of child tracking
one of the most important samples for making life easier. Some of
mobile applications in this field used by humans are about hearing system. In their study, GPS (Global Positioning System)
test, vision test, diabetes, pregnancy, and doctor appointment. This technology was used to provide location data. Another mobile
paper focuses on following of drugs, taken by patients, through application developed by Yuce et. al [3] focused on asthma, a
mobile phones. The application running on the AOS provides the use chronic disease which can be control with patient-doctor
of drugs on time with the alarm system. In addition to this, the cooperation involving frequent medical review and regular
application gives information (time, dosage, name) about drugs by self-monitoring by basic indicators. In their study, they
reading data matrix located on the medicine box. Thanks to visual proposed to help patients keep their asthma under control and
and understandable interface and easy usage, many difficulties enable doctors reach patients. Finally, Kato and Tan[4]
experienced in drug receiption can be eliminated with this
reviewed the six 2D barcodes and then use an extra metric - a
application. Finally, the percentage of receiption of drugs on time
can be increased in the future. first-read rate - to quantitatively verify our earlier results and
better gauge reading reliability.
Keywords— Android, data matrix, web service, drug alarm,
prospectus, smartphones, mobile programming.
II. BARCODE AND DATA MATRIX TECHNOLOGIES
AND STANDARDS
I. INTRODUCTION Barcode technology is one of the information coding
standard. Barcodes are interface elements that provide access
Providing access to health records with the communication
to encoded data via the optical reader. As seen Fig. 1, barcodes,
networks and mobile devices from the desired location has been
consisting of vertical lines and spaces of different thicknesses,
suggested as a growing business value in the health field[1].
is a method used to transfer data automatically and error-free
The drug which is one of the most important thing in the field
manner to another environment.
of health disorder is one of the most effective methods used for
immunity. For this purpose, following of medicine becomes
more difficult and it reveals a dangerous situation in terms of
health for patients. All these problems have emerged the idea
of mobile drug tracking application. The main purpose of drug
tracking is to prevent disease by providing medicine taking in
the right time.
Positive impacts of the market of mobile applications in the
health sector are increasing day by day. Elderly population has Fig. 1 Sample barcode labels
led to increase the health expenditures in a remarkable way.
____________________________________________________________________________________________________________
162
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The use of barcode system has provided great advantages in provide easy way of moving SQL Server databases to Azure.
daily life and business. This technology provides fast and The database is designed to be moved to cloud in the future.
accurate data reading. In addition to this, it provides time saving SQL Server serves the majority of the database and
in the processes such as data update. management tasks. Because of such facilities SQL Server has
There are many types of barcode which have the serious been the preferred database server for our application.
areas and development process past to present. Generally, the
basic characteristic that distinguish species from each other are B. WCF Service
the amount of information that the code can hold. Fig. 2
demonstrates the types of barcode used in literature. The main objective of WCF is to enable the communication
In this paper, data matrix is preferred because of its use in between applications on the network or internet.
the pharmaceutical industry. Some basic information about data Representational State Transfer (REST) is an architecture to
matrix is given as follows: provide the data transfer between client and server via Hyper-
Text Transfer Protocol (HTTP). Since REST services are
Platform-independent, language-independent an easy to scale,
REST architecture is preferred in our system. The return types
BARCODES of REST Services can be used as Extensible Mark-up Language
(XML), JavaScript Object Notation (JSON), etc. based on the
need. JSON is preferred in our application because it can
2D BARCODES transfer smaller pieces of information while transferring data in
1D BARCODES
the application, and it is easier to use on the mobile platforms.
As an example, the barcode number and drug information
which is sent to the service is illustrated in Fig. 3 in JSON
ShotCode
Code 49 Visual Code format. Likewise the database system, the service is also
Code 11 Code 93 Code 16K Color Code designed to be moved to the cloud.
Code 25 Code Data matrix QR Code
Code 39 128/A/B/C Maxicode vs
Onecode EAN 2/5/8/13
MSI
A. Data Matrix
Data matrix code is a two-dimensional matrix barcode
consisting of black and white "cells" structured in either a
square or rectangular pattern. Encoded information on the data
matrix can be text or numeric. Data matrix which can save 3116
numerical or 2335 character can hold more data than one-
dimensional barcode. On data matrix, while each black cell
refers “1”, each white cell refers “0” mathematically.
A. Database For the mobile application, the open source operating system,
Android, is used, and Android Studio is preferred as an editor.
SQL Server Express is a database system, developed and The purpose of this mobile application is to let the users doing
distributed by Microsoft, which can be used to provide a basis medication tracking via Android mobile devices. One of the
for the applications running on desktop and small servers. most important features of the application is to notify the user
Microsoft's Azure cloud infrastructure and application platform, when it is medication time and to let the medication is taken on
time. The medication alarm can be set manually as well as via
____________________________________________________________________________________________________________
163
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
164
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSION
With the increased use of mobile devices, mobile
applications have made a quick entry into our lives in many
areas such as education and health. Health applications attract
the interest of both health workers and patients.
In this application that users can use on smart phones and
tablet computers, follow the prescription section, taking drugs
time and amount of dosage reminders for informative system.
Application makes possible to add up to the required
medication reminder record. The reminder of each drug is
shown separately and when desired a reminder notification can
be added for each drug. The most important benefit of this
application is that nearly almost problems about drug intake
(dosage, time interval …etc.) will disappear. As a result,
receiving time of the overall drug percentage will be raised by
this application.
____________________________________________________________________________________________________________
165
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Nowadays, automatic systems become a crucial in either online or offline. While offline planning requires prior
many factories to achieve some tasks such as minimizing cost, knowledge related to tasks (it can be nodes, duties and so on),
maximizing efficiency, quality, and reliability. The planning is
important for manufacturing systems to adopt changing online planning enable to give new tasks [2].
conditions. Also, manufacturers want to obtain fast, reliable, Actually, AGV scheduling and routing problems are very
qualified and economic products. Flexible Manufacturing complex in reality. However, this problem will be solved in
S ystems (FMS s) are used to meet this need. FMS s make this paper by simplifying and ignoring some conditions.
production fast, qualified, reliable and economic by using Details related conditions, assumptions and constraints will
computer-controlled structure that includes robots and be given in following parts.
transportation systems. Automated Guided Vehicle s (AGVs) and
The paper is organized as the follows. Related works in
FMS are thought to be integrated because FMS uses AGV as a
part of transportation in the factory. AGVs are used to carry literature for AGV routing is given in section 2. Section 3
loads, in other words products, in production areas, warehouses, focuses on problem statement, in other words, assumptions,
factories that use magnets, landmarks, laser sensors, lines to constraints and information about used algorithms. Test
know where they are. AGV scheduling and routing is NP-hard results for proposed algorithm are given in section 4. Finally,
and open-ended problems. In the literature, there are many conclusion part summarizes all written things in the paper
algorithms and methods are proposed to solve these problems. with short paragraph in section 5.
In this study, we present a hybrid algorithm that is composed of
simulated annealing (S A) and Dijkstra’s algorithm to solve the II. LIT ERAT URE REVIEW
routing problem. The hybrid algorithm is compared with S A
algorithm in terms of distance cost using benchmark problems As mentioned before, scheduling and routing of AGVs is
in the literature. NP-hard problem. Therefore, there are so many researches
and solutions in literature and also researchers have been still
Keywords— Automated Guided Vehicle, Flexible Manufacturing trying to deal with it. There are so many methods to solve
S ystems, S imulated Annealing Algorithm, Dijkstra, Routing, AGV or vehicle routing problem such as simulated annealing
Vehicle Route Planning, Automatic S ystems. algorithm, genetic algorithm, tabu search and so on. Also,
some hybrid algorithms that are composed of two or more
I. INT RODUCT ION
algorithms are available in the literature.
Nowadays, there are many factories that use automated Liu and Fu [3] dynamically solve the vehicle routing,
systems to improve quality while minimizing cost to meet scheduling and planning problem by using A* algorithm,
demands of customers in competitive market. There can be minimax criterion and some heuristic rules. They model the
uncertainties and changes due to this competition. The result problem as optimal routing assignment of „p‟ AGVs among
of these uncertainties and changes is decreasing performance „m‟ workstations in order to accomplish „n‟ tasks by using
and flexibility. Therefore, routing and scheduling is important AGV in FMS. Czech and Czarnas [4] investigate the vehicle
to prevent such an undesired condition. Flexible routing problem with time windows. Parallel-simulated
Manufacturing Systems are used to meet this need. FMSs annealing algorithm is proposed to solve routing problem
make production fast, reliable, qualified and economic by assuming that there is a central depot of cargo and n
using computer-controlled structure that includes robots and customers a particular distance far away from the depot. The
transportation systems. Coordination in FMSs is important effectiveness of the method is tested on some well-known
for production and transportation because it is required to instances of the problem. Lin et al. [5] apply an algorithm
schedule limited resources in industry. If FMSs are well- combining simulated annealing with local search approach to
planned, coming jobs are done on time. Therefore, production handle capacitated vehicle routing problem (CVRP) and test
and transportation is carried out smoothly. AGVs are the algorithm on fourteen benchmark problems having
indispensable parts of FMSs. Here, AGVs are responsible for different settings. Onoyama et al. [6] propose an algorith m to
transportation in FMSs. solve vehicle routing problem for a cooperative logistics
AGV control includes duty assignment, routing and network by using genetic algorithm. They use multistage GA
scheduling [1]. AGV routing and scheduling can be done and a method of multistage point evaluation to satisfy the
____________________________________________________________________________________________________________
166
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
167
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
T ABLE I
PROBLEM SET DESCRIPT ION
____________________________________________________________________________________________________________
168
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Kalinovcic, L., Petrovic, T., Bogdan, S., Bobanac, V. , “ Modified
Banker‟s Algorithm for Scheduling in Multi-AGV Systems”,
2011 IEEE International Conference on Automation, Science and
Engineering, pp.351-356, 2011
[2] Bobanac, V., Bogdan, S., “Routing and Scheduling in Multi-
AGV Systems Based on Dynamic Banker Algorithm”, 16th
Mediterranean Conference on Control and Automation Congress
Centre, pp. 1168-1173, 2008
[3] Liu, P., Fu, L., “Planning and Scheduling in a Flexible
Manufacturing System Using a Dynamic Routing Method for
Automated Guided Vehicles”, IEEE, pp. 1584-1589, 1989
[4] Czech, Z.J., Czarnas, P., “Parallel Simulated Annealing for
Vehicle Routing Problem with T ime Windows”, Proceeding of
10th Euromicro Workshop on Parallel, Distributed, and Network-
Fig. 2 VRP solution for hybrid SA using problem set A1. based Processing, pp. 1-8, 2002
[5] Lin, S.-W., Ying, K.C., Lee, Z.-J., Hsi F.-H., “ Applying
Fig. 1 gives the route of the AGVs as the following: Simulated Annealing Approach for Capacitated Vehicle Routing
Problems”, 2006 IEEE International Conference on Systems, Man
Route for AGV 1: 1-21-6-26-11-16-23-10-9-19-30-1 and Cybernetics, pp. 639-644, 2006
Route for AGV 2: 1-28-25-1 [6] Onoyama, T .,Maekawa, T ., Kubota, S., Komoda, N., “ GA
Route for AGV 3: 1-15-29-12-5-24-3-4-7-27-1
Applied VRP Solving Method for a Cooperative Logistics
Route for AGV 4: 1-8-14-18-20-32-22-1
Network”, 2006 IEEE, pp. 1101-1106, 2006
Route for AGV 5: 1-31-17-2-13-1
[7] Morihiro, Y., Miyamoto, T ., Kumagai, S., “An Initial T ask
Assignment Method for Autonomous Distributed Vehicle
Figure 2 gives the route of the AGVs as the following:
Systems with Finite Buffer Capacity”, 2006 IEEE, pp. 805-812,
Route for AGV 1: 1-21-6-26-11-16-10-23-30-1 2006
Route for AGV 2: 1-28-25-1 [8] Wu, N., Zhou, M., “Shortest Routing of Bidirectional Automated
Route for AGV 3: 1-19-9-12-5-29-24-3-4-27-1 Guided Vehicles Avoiding Deadlock and Blocking”, IEEE/ASME
Route for AGV 4: 1-15-7-18-20-32-22-14-1 TRANSACTION ON MECHATRONICS, VOL. 12, NO. 1, pp. 63-
Route for AGV 5: 1-31-17-8-2-13-1 72, 2007
[9] Bouhafs, L., Hajjam, A., Koukam, A., “A T abu Search and Ant
As seen above, routes for AGVs are generally different in Colony System Approach for the Capacitated Location-Routing
Fig. 1 and Fig. 2 but it is not important because algorithm is Problem”, Ninth ACIS International Conference on Software
probabilistic and can give different results at each run. It is Engineering, Artificial Intelligence, Networking, and
required to control whether total distance is getting smaller o r Parallel/Distributed Computing, pp. 46-50, 2008
not. Distance can be controlled with Table I and it can be said [10] Guan, C., Cao, Y., Shi, J., “T abu Search Algorithm for Solving
that SA is improved by using Dijkstra‟s algorithm. the Vehicle Routing Problem”, Third International Symposium on
Information Processing, pp. 74-77, 2010
[11] Li, H., Xu, Z., Zhou F., “A Study on Vehicle Routing Problem
V. CONCLUSION
with Fuzzy Demands Based on Improved T abu Search ”, 2012
So many factories use automatic systems due to developing Fourth International Conference on Computational and
technology to minimize cost and maximize efficiency, Information Sciences, pp. 73-76, 2012
quality, and reliability. FMSs are used to meet this need. [12] Zhu, C., T ang, L., Zhang, W., “Multi-vehicle Coordination and
AGVs play big role in this for the duty of transportation. It is Flexible Scheduling based on Simulated Annealing Algorithm”,
required to make route planning of AGVs to use them
2014 26th Chinese Control and Decision Conference (CCDC) ,
effectively due to costs. SA algorithm is the one of the
pp. 2686-2691, 2014
common methods in literature to solve the routing problem of
[13] Alcan, P., Başligil, H., “A literature Review for the Simulated
AGVs. In this study, it is combined with Dijkstra‟s algorithm
Annealing Method on the Parallel Machines Since 2003 ”, 14th
that is another common method in literature. Hybrid
International Research/Expert Conference, pp. 289-292, 2010
algorithm is tried on problem set by Augerat et al as
mentioned before. Obtained results show that the hybrid [14] Dijkstra, E. W. (1959). "A note on two problems in connexion
approach gives better distance results . In the future work, the with graphs" (PDF). Numerische Mathematik 1: 269 –271.
algorithms are planned to handle the problem for the doi:10.1007/BF01386390.
autonomous AGVs. [15] (2016) Vehicle Routing Dataset. [Online].http://www.coin-
or.org/SYMPHONY/branchandcut/VRP/data/index.htm.old#A
____________________________________________________________________________________________________________
169
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The use of B-spline curves has spreaded too many fields using B-spline curves. In this work, dominant points on the
such as computer aided design (CAD), data visualization, surface borders of the object calculated by Local Curvature Maximum
modeling, signal processing and statistics. The flexible and powerful (LCM) and control points are calculated by least square method.
mathematical properties of B-spline are the cause of being one of the Consequently, similar characters are determined by utilizing
most preferred curve in literature. They can represent a large variety
data set of sample characters. Valenzuela and Pasadas[2] used
of shapes efficiently. The curve behind of the model can be obtained
by doing approximation of control points, approximation of knot simulated annealing method for cubic spline approximation for
points or parameterization.. It is obvious that the selection of knot knot adjustment. Discrete curvature of data points was
points in B-spline curve approximation has an important and smoothed by low pass filters. Artificial immune system is used
considerable effect on the behavior of final approximation. In for optimization of the knots. Gálvez and Iglesias[3] calculated
addition to this, an unreasonable knot vector may introduce positions of the optimal knots using particle swarm
unpredictable and unacceptable shape. Recently, in literature, there optimization algorithm which was one of the most important
has been a considerable attention on the algorithms inspired from metaheuristic approach in literature. They highlighted that
natural processes or events to solve optimization problems such as parameters of the algorithm had an significiant effect on the
simulated annealing, ant colony optimization, particle swarm
performance of the curve fitting problem. Finally, Gálvez et
optimization, artificial bee colony optimization, and genetic
algorithms. This paper implements and analyzes a solution to al.[4] studied elistist clonal selection algorithm for selecting
approximate B-spline curves using Intelligent Water Drops (IWD) optimal free knots. In their study, the focused on artificial
algorithm. This algorithm is a swarm based optimization algorithm immune system for the problem of knot adjustment. They
inspired from the processes that happen in the natural river systems. mentioned that adjustment of computation time and parameters
The algorithm is based on the actions and reactions that take place were basic limitation of the method but it was a matter of tuning
between water drops in the river and the changes that happen in the parameters based on metaheuristic technical problems and they
environment that the river is flowing. Some basic properties of emphasized that this problem was inevitable. Standart
natural water drops are adopted in the algorithm here to solve B- problems solved by IWD algorithm are well known in literature
spline curve fitting problem. Optimal knots are selected through
such as vehicle routing problem, travelling salesman problem
IWD algorithm. The IWD algorithm was experimented by some
benchmark functions. The proposed algorithm convergences and robot routing problem. Differently, Dadaneh et.al[5]
optimal solutions and finds good and promising results. applied IWD algorithm to tackle the graph coloring problem.
This study focuses on the swarm based IWD algorithm
Keywords— Intelligent water drops, natural water drops, which is inspired from natural colonizing behaviour of water
evolutionary algorithms, B-Spline curves, knot points, optimization, drops in nature. Optimal knot points for drawing the curve are
reverse engineering. determined by applying proposed algorithm. We demonstrated
the effectiveness of the proposed method using numerical
studies on multiple representative functions.
I. INTRODUCTION
B-spline curves are usually used in computer aided design
(CAD), data visualization, surface modeling and many other II. B-SPLINE CURVES
fields. B-spline curve data fitting is a challenging problem
encountered in reverse engineering. However, B-spline curves B-spline curves can be determined for a collection of + 1
are most preferred approximation curve because they are very control points. The first and the last control points intersect with
flexible and have powerful mathematical characteristics. curve. The degree of the curve is and must be satisfy the
Because of this feature, they can offer a large variety of shapes equation 1 ≤ ≤ . B-spline curves use blending functions
efficiently. In literature, many approaches and methods have which have local domain areas to overcome restrictions and
been developed for B-spline curve approximation. Tirandaz et disadvantages. These functions are equal to zero outside of own
al.[1] studied on curve matching and character recognition
____________________________________________________________________________________________________________
170
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(1)
5. The error value is calculated by using data estimated
and data obtained . This approach is called Euclidean
where is the control points, is the curve degree, + 1 is the error sum which is expressed as follows:
number of control points and , is the blending function
which is calculated as follows:
(10)
(2)
(7) (11)
(8) (12)
(, )
3. Control points can be obtained by using the equation (, ) (, ) ≥0 (13)
= which is the B-spline curve formula. In this =
(, ) − (, )
equivalance, in order to obtain (control points)
matrix, at first, matrix should be obtained by using
blending functions. Matrix operations are applied over where the set ( ) represents the locations that the IWD
the equation = to left alone. Accordingly, should not visit to keep satisfied the constraints of the problem,
matrix is calculated using the equation = is a small positive number to prevent possible zero division,
4. As a result, B-spline curve can be drawn.
____________________________________________________________________________________________________________
171
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
where ∆ ( , ) is the soil which removes from the This paper focuses on the problem of selecting optimal knots
path between i and j and is a small positive number generated for the best B-spline curve fitting. Data points have been
from the interval [0,1]. In addition , and are constant soil accepted as a cities and all water drops start from the same point.
parameters and is a small positive number to prevent When a water drop reaches predetermined point, it is accepted
possible zero division (According to original algorithm for that this water drop completed its tour. At the end of the
finding the best tour we calculate the path with minimum length, algorithm, the cities on the best path are accepted optimal
however sometimes depending on the problem different points to fit data. Fig. 2 shows the optimal data points for the
measures can be used. For this study, we use error value B-spline curve fiiting.
between original curve and fitted curve instead of Euclidean In this study, some modifications have been made on the
error.) IWD method. The selection of some unnecessary points has
been prevented by adding the concept of the radius of curvature.
Global soil updating rule: At the end of the each iteration, the Neighborhood-based local search increases the probability of
amount of soil on the edges of the best tour (solution) ( ) is selecting better ways. Another concept that has been added to
updated via Eq. 19[7]: the steps of the algorithm is the diversity rate used in local
search step. In this way, paths that water drops may follow are
gone up. Experimental studies demonstrate that diversity rate
prevents hanging out the local minimum. In the following
section, some experimental studies were represented to
demonstrate the effectiveness of the proposed algorithm
____________________________________________________________________________________________________________
172
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. EXPERIMENTAL STUDIES
(22)
In this section, numerous numerical simulations are
performed to demonstrate the effectiveness of the proposed
optimization algorithm. The scope of B-spline curve fitting,
where is the number of data used in the approximation by
IWD algorithm is coded. On the knot prediction of curve in
cubic spline, is the number of interior knots used for
question, mentioned reverse engineering approaches and
construction of B-spline, is the order of the spline to be fitted
optimization process steps have been combined. Some
for the given data, and is sum of squared error.
parameters and values used in the problem are given in the
The reason why we used AIC and BIC as a fitness
following table:
function is that they do not use subjective parameters such as
Table 1 Algorithm parameters (*:all values in the range used in simulations ) error limits and smoothing functions. Therefore, these
functions provide simple and straightforward procedure for
Parameter Value identifying the best result[3]. According to these functions,
Number of knot >6 lower values denote better fitness values.
Proposed method is compared with the elitist clonal
Curve degree 3
selection algorithm. In that study, Galvez et. al[4] introduced
Number of iteration < 20 an adopted elitist clonal selection algorithm for automatic knot
adjustment of B-spline curves. In their study artificial immune
Number of IWD [3,10]∗
system paradigm was applied to solve knot adjustment problem
Tolerance error 0,0001 for B-spline curves. In this section, performance of the methods
The number of neighbors [1,10]∗ was compared by calculating AIC and BIC.
Our algorithm for B-spline curve fitting has been applied to
Diversity rate [10,100]∗ several test functions and we consider here three of them,
displayed in Fig. 1.
Static parameter values used in the proposed algorithm for
the solution of B-spline curve fitting problem are listed as
follows:
Parameter
Value 1 1000 100 [0.7-0.9]* 0.01 0.0001 Fig. 1 Test functions used in this paper
All experiments in this paper have been performed on a 2.4 Table 3 shows their corresponding mathematical definitions
GHz. Intel Core i7 processor with 12 GB. Of RAM. The source and associated domains. Each function is evaluated at
code has been implemented by the authors in the programming uniformly distributed values of t in its domain to generate a
language of the popular scientific program Matlab, version collection of 201 data points on the interval [0,1].
R2015a. The reason for selection of Matlab is that it is very
suitable tool for this task. Because it is fast, and it provides Table 3 Test functions used in this paper: mathematical definition[3]
reliable, well tested routines for efficient matrix manipulations. Equation Domain
For this study we compute two fitness functions: Akaike
Information Criterian (AIC) and Bayesian Information
Criterian (BIC). AIC and BIC which are used to measure
performance are defined as follows:
= + 2(2 + ) (20)
= + ( )(2 + ) (21)
____________________________________________________________________________________________________________
173
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 2 Selected data points of functions f1, f2 and f3 respectively. The red dots represent data points(true function);blue circles represent selected data points
for B-spline curve approximaton
____________________________________________________________________________________________________________
174
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 3 Approximated results of f1, f2 and f3 respectively. The green dashed line represent data points(true function);blue dashed line is the approximaton.
____________________________________________________________________________________________________________
175
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Table 4 summarizes our simulation results of B-spline 3. Galvez, A. and A. Iglesias, Efficient particle
curve fitting. Here, best results are highlighted in bold. Fig. swarm optimization approach for data fitting with
3 demonstrates the approximated results of all function that free knot B-splines. Computer-Aided Design,
we tested for the best solutions visually. Fig. 4 shows the 2011. 43(12): p. 1683-1692.
effect of the diversity rate on the AIC and BIC values of the 4. Galvez, A., et al., Elitist clonal selection
experimental result for function as an example. algorithm for optimal choice of free knots in B-
Our comperative results reported in Table 4 confirm that spline data fitting. Applied Soft Computing, 2015.
our method outperforms in terms of the fitting errors for the 26: p. 90-106.
knot selection problem. The number of iterations used to 5. Dadaneh, B.Z., H.Y. Markid, and A.
obtain best fitting is nearly ten times lower than other Zakerolhosseini, Graph coloring using intelligent
method. By using the AIC and BIC error functions we also water drops algorithm. Iranian Conference on
provide efficient procedure to determine the optimal Electrical Engineering, 2015: p. 595-600.
number of interior knots. 6. Ulker, E., Surface Modeling Using Artificial
Intelligence Techniques. Selcuk University–
Electronic Electrical Engineering Department,
VI. CONCLUSION 2007: p. 142.
7. Shah-Hosseini, H., Problem solving by intelligent
We proposed the use of this algorithm for solving curve water drops. 2007 Ieee Congress on Evolutionary
fitting problem. The points have been accepted as cities and Computation, Vols 1-10, Proceedings, 2007: p.
a number of IWDs were placed on the cities randomly. At 3226-3231.
the end of the algorithm, the best path followed by IWD was
determined. In this study, the points on the path were
accepted the best points that represent the curve optimally.
Metaheuristic approaches frequently used in optimization
problems are very strong but the convergence slows down
when the generations approach to optimal solutions.
Because of this, the cost of the problem rises up. Proposed
method solves this problem in the local search step of the
algorithm. In this step, diversity rate is used to prevent
hanging out the local minimum.
In this paper, performance of IWD algorithm is
demonstrated by representing numerous test functions and
simulations. As a result, the proposed algorithm converges
fast to optimal solutions and finds good, efficient and
satisfactory results.
ACKNOWLEDGEMENT
REFERENCES
____________________________________________________________________________________________________________
176
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Artificial sweeteners like MSG (MonoSodium Glutamate) The video tracking systems in open field tests (including
model has been used anxiety-like behaviors on rats. The tracking of mazes) has become a very popular technique for monitoring of
rat’s movements has broad applicability to questions in anxiety-like anxiety-like behaviors on rats. The automated observation
behaviors with different doses MSG injections (50 mg/kg/day, 100 systems, which always work in the same way therefore
mg/kg/day and 200 mg/kg/day) to rats. In this paper, in order to
behaviors are recorded more reliably, provide to researchers
measure three types locomotor activity (line crossing, rearing,
grooming), a video tracking software is used. The advantage of this many behavioural information about animals. In literature,
type of tracking software is that it provides to give locomotor activity various types of the techniques can be found, including infrared
of rats in real-time. The experimental results obtained in this study photobeam detectors [10], [11], [12], [13], vibration detection
have shown that learning and memorial functions negatively [14], radio telemetry [15], radar [16] and computerized video
affected in the brains of the rats an anxiety-like model. In addition, analysis [17], [18]. However, video-based tracking systems
the visual tracking results demonstrate that video tracking system using video cameras are currently the most common approach
provides an accurate monitoring of rat’s behavior. to achieve the task [19]. As video-camera technology has
preferred to obtain the desired outputs, many automated video
Keywords—Rat, locomotor activity, anxiety-like behaviors, MSG, tracking methods have been developed [20]. In those methods,
computerized video tracking, animal tracking. analog video signals are sent to video processing unit which
detects peaks of the signal. The detected signals indicate
contrast between the tracked animal and the background. Then
I. INTRODUCTION the coordinates of the tracked animal are determined. Modern
The etiology of the behavioral disorders might be genetic, video tracking systems generally digitize analog video signals
physical, developmental and also psychological or social as using frame grabbers. The digitized video frames is composed
well. Neurological disorder appears mostly with anxiety of pixels and each pixel has gray scale value [21]. The tracking
disorder and depression. In animals, behaviors likely to anxiety procedure analyzes digitized frame in order to pick out the
can be identified through experimental open field test, animal from the background. Then, the desired calculations are
light/dark test, elevated plus maze test, tail suspension test, performed in order to explain the meaning of animal’s
forced swim test, social interaction task [1]. Open field test behaviour.
enables the evaluation of parameters such as line crossing, In this study, the anxiety-like behaviors are investigated
rearing, grooming and the defecation number in rats [2]. using pharmacological activation model with MSG. The
Studies in rodents have shown that the administration of objective of this study is to evaluate the effects of neonatal
MSG at the neonatal stage alterations in locomotor activity exposure to MSG on locomotor activity with video tracking
have been shown [3], [4]. Grooming duration and rearing system.
counts can be used to assess anxiety-like and exploratory This paper is organized as follows: In Section 2, material
behaviors in rodents [5]-[7]. Following acute MSG and method, which are designed here, are explained in detail.
administration, locomotor activity reduced, grooming Section 3 presents the numerical results of the monitoring
increased, while rearing activity reduced at MSG doses [8]. behaviors on rats with video tracking technique. The discussion
Reduced locomotor activity can be considered a form of and conclusions about the observation results are given in
anxiety-like behavior [9]. Section 4.
1 Corresponding author
____________________________________________________________________________________________________________
177
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
II. MATERIAL AND METHOD scaling is the fastest. The threshold values of the video signals,
In this study, in order to measure various locomotor activity which are used to subtract rat’s image from the background, are
(such as line crossing, rearing and grooming) an automated set manually. Then, the coordinates of the animal are
video tracking mechanism is designed. determined and recorded. Then, the monitoring of animal
behavior is achieved in the open field maze.
A. Testing Procedure
B. Experimental Animal Group
The open field test setup consists of a square field with a
dimension of 45 × 45 cm and it is lightened by 3 fluorescent Rats that are included in the study, are subject to 8 sessions
bulbs. The top of the field is open to the ambient and the square of MSG injections in total, in the order of 50 mg/kg/day (group
field is bordered by white stripes each of 1 cm width, on a black number = 6), 100 mg/kg/day (group number = 6) and 200
colored background. Each of the experimental animals are mg/kg/day (group number = 6). Rats that are selected as the
brought to the center of the open field setup and released to control group (group number = 6), are subject to the
move freely. Duration of the test is determined as 3 minutes intraperitoneal serum physiologic.
during which the rats are observed closely and any of their
behaviors are recorded as line crossing, rearing and grooming.
Figure 1 shows the open field test area and the used video
C. Hardware and software equipment
tracking mechanism is illustrated in Figure 2. Video data were processed with a PC (Windows 10 Pro
platform, Intel Core i7-4600U CPU @ 2.10GHz 2.70 GHz
processor). Video frames were obtained by Flaxes FC 1604
camera (30fbs/15fbs). The data processing algorithms were
developed using MATLAB 2014b and Image Processing,
Computer Vision toolboxes.
D. Data Analysis
To analyze the data, we used an analysis of variance
(ANOVA) for multiple groups. More complex designs, such as
one-way ANOVA with repeated measures or n-way ANOVA
(additional factors: time, treatment, genotype, stress, etc), can
also be used on behavioral studies of rats.
Fig. 1 Open field test area
III. OBSERVATION RESULTS
In this section, the observed rat’s movements are illustrated
separately in certain moments as shown in Figure 3. In addition,
comparative tables related to locomotor activity are shown to
discuss the behaviors of rats.
The video camera records the open field setup from the top
where the rats are free in this area. Then, the camera signals
synchronously are sent to a PC. The signals are digitized by
frame grabbers. The video processing software analyzes each (a) 5th second
frame in order to distinguish the tracked objects from the
background on the basis of their gray scale since the gray
____________________________________________________________________________________________________________
178
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Poi
Rearing
3.1±1.04 4.0±1.09 2.6±0.4 3.1±0.6 0.740
time
(sec.)
pi: pre injection, poi: post injection.
(b) 20th second *p<0.05 (p-value is statistically significant).
Poi
Grooming 14.0±1.78 52.0±12.07 49.6±9.82 39.5±9.28 0.030*
time (sec.)
(c) 160th second
pi: pre injection, poi: post injection.
Fig. 3 Observed rat’s movements for several moments *p<0.05 (p-value is statistically significant).
Following the MSG injections, when the characteristics of
Following post injection of MSG, grooming is increased
anxiety resembling behaviors of the rats are monitored, a
significantly at doses of 50 mg/kg/day and 100 mg/kg/day,
statistically considerable decrease is observed (p<0.05) in the
respectively. It is known that the following acute MSG
line crossing of the animals in the open field (Table 1).
administration, locomotor activity reduced, grooming
Table 1. ANOVA values of line crossing and descriptive statistics, referred increased, while rearing activity reduced at MSG doses.
to the open field test values of the rats, pre and post injections of MSG Reduced locomotor activity is considered a form of anxiety-like
behavior.
Control MSG Dose Groups
Group MSG50 MSG100 MSG200
(n=6) (n=6) (n=6) (n=6)
p-value IV. DISCUSSION AND CONCLUSION
Pi In the literature, studies based on the anxiety model are
Line
crossing 51.5±3.21 53.8±9.54 60.0±5.06 0.043* achieved on animals through injections of artificial sweeteners
32.83±7.277
time such as MSG [22], [23], [24]. In the present study,
(sec.)
characteristics of anxiety like behaviors of rats are determined
Poi through the injections of MSG at repetitive dosages. Following
Line to the MSG injections, when we monitored the behavioral
crossing 30.83±7.068 23.5±4.11 18.1±3.99 23.8±2.96 0.343
time characteristics of rats resembling to anxiety, we determined a
(sec.) statistically considerable decrease in the roaming time of the
pi: pre injection, poi: post injection. animals in the open field and a decrease in the number of
*p<0.05 (p-value is statistically significant).
rearing behavior. However, the number of grooming is
increased in a statistically considerable scale. Similar results is
Descriptive statistics are point out Mean ± standard error in
provided that a decrease in the motor activity is argued in [25],
Table 1. In addition to online tracking of the rat, designed video
[26].
tracking software can automatically detect the other locomotor
____________________________________________________________________________________________________________
179
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[3] M. Dubovicky, D. Tokarev, I. Skultetyova, D. Jezova. Changes of [18] CV. Vorhees, KD. Acuff-Smith, DR. Minck, RE. Butcher, A method for
exploratory behavior and its habituation in rats neonatally treated with measuring locomotor behavior in rodents: contrast-sensitive computer-
monosodium glutamate, Pharmacol Biochem Behav, vol, 56, pp. 565- controlled video tracking activity assessment in rats. Neurotoxicol
569, 1997. Teratol, vol. 14, pp. 43–49, 1992.
[4] Z. Hlinak, D. Gandalovicova & I. Krejci. Behavioral deficits in adult [19] Q. Xu, C. Cai, H. Zhou, H. Ren, A video tracking system for limb motion
rats treated neonatally with glutamate, Neurotoxicol Teratol, vol. 27, pp. measurement in small animals. In Optoelectronics and Image
465–473, 2005. Processing, International Conference, vol. 1, pp. 181-184, 2010.
[5] N. Enginar, I. Hatipoglu, ˘ M. Firtina, Evaluation of the acute effects of [20] S. C. Fu, K. M. Chan, L. S. Chan, D. T. P Fong, P. Y. P Lui, The use of
amitriptyline and fluoxetine on anxiety using grooming analysis motion analysis to measure pain-related behaviour in a rat model of
algorithm in rats, Pharmacol. Biochem. Behav. Vol. 89, pp. 450–455, degenerative tendon injuries. Journal of neuroscience methods, vol.
2008. 179.2, pp. 309-318, 2009.
[6] T.S. Perrot-Sinal, A. Gregus, D. Boudreau, L.E. Kalynchuk, Sex and [21] JA. Endler, On the measurement and classification of colour in studies
repeated restraint stress interact to affect cat odor-induced defensive of animal colour patterns. Biol J Linn Soc, vol. 41, pp. 315–352, 1990.
behavior in adult rats, Brain Res., vol. 1027, pp. 161–172, 2004.
[22] BA. McCool, AM. Chappell. Chronic intermittent ethanol inhalation
[7] A. Takeda, H. Tamano, F. Kan, H. Itoh, N. Oku, Anxiety-like behavior increases ethanol self-administration in both C57BL/6J and DBA/2J
of young rats after 2-week zinc deprivation, Behav. Brain Res., vol. 177, mice. Alcohol., vol. 49, pp. 111-120, 2015.
pp. 1–6, 2007.
[23] CB. Quines, SG. Rosa, JT. Da Rocha, BM. Gai, CF. Bortolatto, MF.
[8] O.J. Onaolapo, A.Y. Onaolapo, M.A. Akanmu, G. Olayiwola, Foraging Duarte, CW. Nogueira. Monosodium glutamate, a food additive, induces
enrichment modulates open field response to monosodium glutamate in depressive-like and anxiogenic-like behaviors in young rats. Life
mice. Annals of neurosciences, vol. 22.3, pp. 162, 2015. Sciences, vol. 107, pp. 27-31, 2016.
[9] P. Muigg, S. Scheiber, P. Salchner, M. Bunck, R. Landgraf, N. [24] SG. Rosa, CB. Quines, EC. Stangherlin, CW. Nogueira. Diphenyl
Singewald, Differential stress-induced neuronal activation patterns in diselenide ameliorates monosodium glutamate induced anxiety-like
mouse lines selectively bred for high, normal or low anxiety, PLoS One behavior in rats by modulating hippocampal BDNF-Akt pathway and
vol. 4, e5346, 2009. uptake of GABA and serotonin neurotransmitters. Physiology &
Behavior, vol. 155, pp. 1-8, 2016.
[10] RJ. Beninger, TA. Cooper, EJ. Mazurski, “Automating the measurement
of locomotor activity”, Neurobehav Toxicol Teratol, vol. 7, pp. 79–85, [25] J. Sun, C. Wan, P. Jia, AH. Fanous, KS. Kendler, BP. Riley, et al.
1985. Application of systems biology approach identifies and validates GRB2
as a risk gene for schizophrenia in the Irish Case Control Study of
[11] RL. Clarke, RF. Smith, DR. Justesen, “An infrared device for detecting Schizophrenia (ICCSS) sample. Schizophrenia research., vol. 125, pp.
locomotor activity”, Behav Res Methods Instrum Comput, vol. 17, pp. 201–208, 2011.
519–525, 1985.
[26] NH. Woo, HK. Teng, CJ. Siao, et al. Activation of p75NTR by
[12] E. Robles, “A method to analyze the spatial distribution of behavior”, proBDNF facilitates hippocampal long-term depression. Nat Neurosci.,
Behav Res Methods Instrum Comput, vol. 22, pp. 540–549, 1990. vol. 8, pp. 1069-1077, 2005.
____________________________________________________________________________________________________________
180
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, a system based on image processing has failures in elevators are usually seen in the components such
been developed in order to prevent wear on guide-rail surface in as elevator guide-rail system, ropes and motors.
elevators. In the proposed method, real-time condition Today, when we take multi-storey buildings as an example,
monitoring is performed by cameras using built-in system. The the fact that the elevator becomes out of service negatively
images of elevator guide-rail surface are captured via four digital
affects the life of users and reduces the quality of life. In this
cameras fixed onto elevator cab. The image-processing methods
are applied on the images captured by cameras and hence the study, it is aimed to repair the small failures that may occur in
wears on the surface of guide-rails are detected. The surface of the elevator line without causing major failures through early
guide-rail is firstly detected in the proposed method. Then, image detection. Thus, elevator companies will bring quality service
segmentation and mathematical morphology are applied on the to the elevator systems by gaining customer satisfaction. A
image of guide-rail surface and the wears on the surface of rail elevator company with such a system will be preferred more
are detected. The failure extent of the wear failures detected are by its customers. In this way, the fact that users are stuck in
calculated. By processing the images captured by four cameras the elevator cages and the system becomes out of service will
during movement of elevator, the results for surface of guide- be prevented by eliminating the failures that may occur
rails are obtained. Using these results, reporting is performed.
suddenly in elevators.
An elevator prototype has been created in order to carry out
tests for development of the proposed method. The tests have Today, most of the multi-storey buildings have elevator
been conducted by fixing the built-in system and cameras onto systems. Elevator systems are excessively used in the daily
this elevator prototype. It is considerably advantageous to detect life. Elevators are used not only in buildings but also in
the failures on elevator guide-rails through image-processing dangerous places such as mines. In this study, a system has
methods. Following a literature review, it is seen that the been proposed for the early detection of failures that may
proposed method is a new approach. occur in the rail line, which is the most important component
of the elevator. In the proposed system, it is aimed to produce
Keywords— Elevator Systems, Fault Detection, Image accurate results by using digital cameras with image
Processing, Image Segmentation processing techniques without being affected by the
environmental conditions. Elevator rail is fixed to the elevator
shaft and is a unit which facilitates movement of the elevator.
I. INTRODUCTION In the elevator system, the guide rails are used for the
Elevators ensure transportation of people inside buildings purposes of guiding the cage and counter weight separately in
and increase their life quality. High-rise buildings whose the vertical movements, minimizing its horizontal movements,
number is increasingly going up today has one or more protecting the vertical directions of the cage and counter
elevator cabs to provide vertical transportation. A great weight and preventing its rotation. In this study, methods have
number of people use elevators in many buildings such as been developed for the early detection of failures that may
business centres, hotels, hospitals and shopping centres daily. occur in the rails during the movement of the elevator cage.
It is highly essential for the elevators used by many people The general structure showing the sample elevator system
daily to operate constantly. In the event of sudden failure of is given in Figure 1.
elevators during operation, people inside them face with a It is seen that similar studies are not available when a
tough situation. Also, people have difficulty during the literature review is performed for the proposed method. When
maintenance-repair period of elevators. Elevator system has national theses are analyzed, we see how important the
counterweight system in order to balance the weight of elevator guide rails are [1-4]. In the master's thesis, Serhat [1]
elevator cab. A guide-rail system has been developed to limit carried out a thesis study about the stress analysis of guide rail
the movements of elevator cab and counterweights on consoles by mentioning the importance of the elevator guide
horizontal axis. When an elevator system is operational, cab rails. When we analyze the studies which have been carried
and counterweight system move reversely. The common out until today, studies have usually been carried out by
elevator control and elevator component analysis [5].
____________________________________________________________________________________________________________
181
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
182
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
0 0 1 0 0
1 1 1 1 1
Cameras
1 1 1 1 1 (5)
1 1 1 1 1
0 0 1 0 0
The flow chart of the proposed method is seen in Figure 4. The image in Figure 5.b was obtained by applying the
Firstly, the rail surface is extracted as preprocessing on the opening operation to the normal image in Figure 5.a. In Figure
image taken from the camera. A section with a size of 5.b, it is seen that small asperities on the rail surface were
480x200 is taken from the image taken in 480x640 size. This disappeared and the real failures appeared more clearly. After
section taken involves the rail surface image. The rail surfaces the opening operation, the failure detection and the size of the
represent the same pixels on the continuous image because the failure are estimated by taking into account the number of
cameras are fixed over the elevator cage. Therefore, the pixels white pixels on the image. The code particle given in Figure 6
belonging to rail surface are extracted in 480x200 size from is used for the estimation of the failure detection and the size
the image. The obtained RGB rail surface image is of the failure.
transformed into gray format and then binary format [18,19].
The image was transformed into binary format by using a
specific threshold value. This threshold value was determined
through the tests performed. The opening operation was
performed as the mathematical morphology operation on the
binary image. The mathematical expressions of the basic Fig. 6 The code particle used for failure detection and the size of the failure
binary morphological operations are defined in equations
1,2,3 and 4 [20].
____________________________________________________________________________________________________________
183
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Induction
motor
Camera 2 Camera 3
Elevator
rope
Camera 1 Camera 4
Control panel
Elevator cabin
Raspberry pi
Guide rails Fig. 8 Fixing of the cameras and the preparation of the test environment
____________________________________________________________________________________________________________
184
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
presented in Figure 10.a. The proposed method was applied to IV. CONCLUSIONS
the failed rail surface image in Figure 9.b, and the obtained Elevator systems have become an indispensable part of
result is presented in Figure 10.b. people in our day. Elevators are the means of transport which
are beneficial to people regarding the transportation of loads
and people. The fact that elevators are failed while working is
one of the undesirable events. In particular, the fact that
elevators are failed while people are using them reduces the
quality of life. A new image processing-based method for the
prevention of the elevator failures has been proposed in this
study. The rail surface erosion failures that may occur on the
rails, one of the critical components in the elevator systems,
are detected. In this study, a sample elevator system was
created and 4 cameras were fixed on the elevator cage.
Whether the rail surface was robust or failed was detected and
a) b) the size of the failure was estimated if it was failed, by
Fig. 10 Application of the proposed method to the rail surface image a)
applying the proposed method on the images taken from the
Robust rail surface image b) Failed rail surface image cameras. When the proposed method is compared with the
studies in the literature, it is seen that it is a new study and it
can operate in real time. This study provides a basis for the
In Figure 10.a, it is seen that the rail surface is robust and future studies, and it is aimed to detect the erosion, cracking,
there is not any erosion failure. In Figure 10.b, it is seen that warping and corrosion on the elevator rails in the future
there is an erosion failure. The proposed method was tested studies.
for 4 cameras and the erosion failures on the rail surface were
detected. The speed of the developed elevator prototype is ACKNOWLEDGMENT
about 0.60 m/sec. When the speed of the elevator was taken This study was supported under the Teknogirisim project of
into account, the real time operation of the proposed method the Ministry of Science, Industry and Technology. Project No:
was evaluated as appropriate and it was seen that it was 0684.TGSD.2015.
working successfully as a result of the tests performed. The
real-time rail surface is monitored in the “Python Sheel” REFERENCES
environment while the elevator is working. As in Figure 11, [1] K. Serhat, "Asansör kılavuz ray konsollarının gerilme analizi," İstanbul
the size of the failure is monitored by performing failure Teknik Üniversitesi / Fen Bilimleri Enstitüsü / Makine Mühendisliği
detection in the “Python Sheel” environment while the Anabilim Dalı, Yüksek Lisans, 2009.
[2] S. Ekrem, "Asansör kılavuz ray konsollarının bilgisayar destekli
elevator is working on a sample rail with erosion failure on
gerilme analizi," Yıldız Teknik Üniversitesi / Fen Bilimleri Enstitüsü /
the rail surface. Makine Mühendisliği Anabilim Dalı / Konstrüksiyon ve İmalat Bilim
Dalı, Yüksek Lisans, 2011.
[3] A. Mehmet, "Asansör kılavuz ray bağlantı elemanlarının deneysel
gerilme analizi," İstanbul Teknik Üniversitesi / Fen Bilimleri Enstitüsü
/ Makine Mühendisliği Anabilim Dalı, Yüksek Lisans, 2012.
[4] E. Serkan, "Kılavuz ray bağlantı elemanlarının gerilme analizi,"
İstanbul Teknik Üniversitesi / Fen Bilimleri Enstitüsü / Makine
Mühendisliği Anabilim Dalı / Konstrüksiyon ve İmalat Bilim Dalı,
Yüksek Lisans, 2011.
[5] D. Zafer, "Elektrik kesintilerine karşı asansör kurtarma sistemi tasarımı
ve uygulaması," Selçuk Üniversitesi / Fen Bilimleri Enstitüsü, Yüksek
Lisans, 2006.
[6] J. Landaluze, I. Portilla, N. Cabezón, A. Martínez and R. Reyero,
“Application of active noise control to an elevator cabin,” Control
Engineering Practice, vol.11, pp. 1423-1431, 2002.
Fault Detection [7] W. Peiliang, H. Wuming, Y. Wenjun, “Fault Diagnosis of Elevator
Braking System Based On Wavelet packet algorithm and Fuzzy Neural
Network,” 9th International Conference on Electronic Measurement &
Instruments, pp. 4-1028 - 4-1031, 2009.
[8] G. Zhang, S. Huang, Y. Yuan, “The Study of Elevator Fault Diagnosis
Based on Multi-Agent System,” International Conference on
Computational Intelligence and Software Engineering, pp. 1 – 5, 2009.
[9] Z. Xi, Z. Guo-jun, W. Yuan-ping, W. Chao-rong, W. Jun-hui,
“Research on the elevator door control system based on the image
processing technology,” International Conference on Electrical and
Control Engineering (ICECE), pp. 1781 – 1784, 2010.
[10] Z. Chen-Guang X. Hong-Yu, J. Liang, “Research of elevator fault
diagnosis based on decision tree and rough set,” International
Fig. 11 Real-time operation of the proposed method and its monitoring in the Conference on Computer Science and Information Processing (CSIP),
“Python Sheel” environment pp. 1318 - 1322, 2012.
____________________________________________________________________________________________________________
185
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[11] J. Chen, X. Liu, “AR bispectrum in fault diagnosis of elevator application for energy and time saving,” The Scientific World Journal,
machinery,” IEEE Fifth International Conference on Advanced 2013.
Computational Intelligence (ICACI), pp. 1149 – 1152. [17] M. Baygin, M. Karaköse, A new intelligent group elevator control
[12] W. Hu, M. Schroeder, A.G. Starr, “A Knowledge-Based Real-Time approach. In MECHATRONIKA, 2012 15th International Symposium
Diagnostic System for PLC Controlled Manufacturing Systems,” IEEE pp. 1-6, 2012.
International Conference on Systems, Man, and Cybernetics, vol.4, pp. [18] O. Yaman, M. Karaköse, E. Akın, I. Aydın, “Image processing based
499 – 504, 1999. fault detection approach for rail surface,” In 2015 23nd Signal
[13] Z. Yimou, “Study to Elevator inspection market in Deyang city, Processing and Communications Applications Conference (SIU) pp.
China ,” Bachelor’s thesiss International business Degree programme, 1118-1121, 2015.
2012. [19] E. Karakose, M.T. Gencoglu, M. Karakose, O. Yaman, I. Aydin, E.
[14] G. Niu, S.S. Lee, B.S. Yang, S.J. Lee, ”Decision fusion system for fault Akin, “A new arc detection method based on fuzzy logic using S-
diagnosis of elevator traction machine,” Journal of Mechanical Science transform for pantograph–catenary systems,” Journal of Intelligent
and Technology, 22(1), 85-95, 2008. Manufacturing, 1-18, 2015.
[15] F. Zhao, X. Koutsoukos, H. Haussecker, J. Reich, P. Cheung, [20] M. Karhan, M.O. Oktay, Z. Karhan, H. Demir, “Morfolojik Görüntü
“Monitoring and fault diagnosis of hybrid systems,” IEEE Transactions İşleme Yöntemleri ile Kayısılarda Yaprak Delen (Çil) Hastalığı Sonucu
on Systems, Man, and Cybernetics, Part B (Cybernetics), 35(6), 1225- Oluşan Lekelerin Tespiti,” 2011.
1240, 2005. [21] P. Soille, “Morphological image analysis: principles and applications,”.
[16] M. Baygin, M. Karakose, “Immunity-based optimal estimation Springer Science & Business Media, 2013.
approach for a new real time group elevator dynamic control
____________________________________________________________________________________________________________
186
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this work, we present a computer program that repetitions for each experiment. Thus, dimension reduction
reads, manipulates, analyses and stores the mass spectra and selection of the data analysis methods (PCA, LDA, kNN
obtained by using Laser Time of Flight Mass Spectrometry (L- etc.) became an important problem. In order to overcome
TOF-MS) system described elsewhere. This program allows these problems, new software tools was created similar to
users to record the characteristic parameters of experimental
MCR-ALS (Multivariate Curve Resolution-Alternating Least
data such as vacuum pressure, voltages, laser power, sample
name etc. both manually and automatically. Also, the obtained Squares) toolbox programmed in MATLAB environment [16].
mass spectra can be used to investigate chemical substances A similar program (FlavonQ) was developed to
(NOx, SOx, organic molecules etc.) or can be used to real time simultaneously analyse ultra-high-performance liquid
identification of the sample differences for given material chromatography (UHPLC) high-resolution accurate mass
database (metal oxides, alloys, paintings, healthy and cancerous spectrometry −mass spectrometry (HRAM-MS) data and UV-
tissues etc.) by using different mathematical and statistical VIS spectrometry data in order to profile the flavonoids [17].
procedures (PCA, LDA, kNN etc.). We have recently presented a Recently, a user-friendly program (SIM-XL) was developed
simple chemical analysis procedure by starting from reading raw for structural protein characterization by using peptide cross-
MS data from oscilloscope using TCP/IP protocol and following
linking analysis method for tandem mass spectrometric data
data processing steps. We have a great success to reduce the
numbers of steps and time duration consumed for the following [18]. In proteomics studies, a GUI program (IPeak) also
procedure. presented for identification of the peptides data obtained from
LC-MS/MS method [19]. To identify TLC (Thin Layer
Chromatography) information from MSI (Mass Spectrometry
Keywords— Mass Spectrometry, Data Acquisition, Data
Imaging) data sets, a program (DetectTLC) was developed
Processing, Statistical Analysis, GUI, MATLAB
[20]. Homemade software or package programs were used for
biological tissue identification [21] or cancer detection [22] by
using mass spectrometry data.
I. INTRODUCTION
Mass spectrometry is an analytical method and has a wide III. EXPERIMENTAL SETUP
applications including; industrial quality control [1-3], Experimental setup consists of laser systems, time of flight
pharmacology [4-7], isotope ratio determination [7-10] and mass spectrometer and electronics. In our experiments, we
space exploration researches [11-13]. Mass spectrometers are using high-intensity femtosecond/nanosecond lasers. In
consist of three units, which are an ionisation unit, separation the scope of this work, we have used a femtosecond laser
unit and detector. The obtaining and interpretation of the system (Quantronix, Integra-C-3.5, NY, USA) as an ionisation
mass spectral data have a great importance to reveal the source. TOF-MS system was designed and built by our group;
molecular structure and laser ionisation/dissociation processes background pressure of the vacuum chamber can be pumped
[14, 15]. down to 10-8 mbar. Ion signals were collected by using an
Data analysis is the most time-consuming processes and MCP detector (El-Mul Technologies LTD., Israel). The
manual data processing can cause the unintended human error, details of the experimental setup were presented elsewhere [23,
such as, reading or writing ion intensities, workspace (time – 24].
mass to charge ratio) conversion errors. In this work, we have Mass spectra were recorded remotely via TCP-IP protocol
presented the reducing the numbers of steps and time duration and real time by using a four-channel oscilloscope (LeCroy
spend for the mass spectral analysis procedure. Corporation, WaveRunner 64 Xi, NY, USA).
II. BACKGROUND AND THEORY IV. RESULTS AND DISCUSSION
Recent years, multivariate data analysis methods have GUI of the mass spectrometry program consists of the
attracted great attention due to the large experimental data sets subpages that are Experiment, Setup, Connection, Laser
obtained from different experimental methods, especially in Diagnostics and Database Control, as shown in Fig. 1. In the
mass spectrometry, researchers obtain thousands of ion peaks experiment section, the user can write sample name, number
(mass to charge ratio) in different intensities for multiple of sampling and perform plotting mass spectra. Setup section
____________________________________________________________________________________________________________
187
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(given as an inset in Fig. 1) provides the user to recording MATLAB .mat file. After selecting and opening database file,
experimental parameters such as electrode voltages, vacuum the program record raw experimental data that contain mass
pressure and laser power. In the connection section, the user spectra as an ion intensity (intensity corresponds to voltage
defines IP number of the oscilloscope to connect using TCP- value read by the oscilloscope) versus the time of flight of the
IP protocol. In laser diagnostics section, the user can obtain corresponding ion peak. The user has to calibrate the system
real-time UV-VIS spectrum of the laser pulses (Fig. 2.) by by using equation ((m/z)=at2+b) to transform time of flight
using UV-VIS-NIR portable fiber-optic USB spectrometer spectrum to mass spectrum to use m/z ratio for further spectral
(0.1 nm FWHM optical resolution, USB 2000+, Ocean analysis [25]. For calibration, the user has to select two
Optics). In database control section, the user can load a known peaks (H+ and C+ or CH3+ peaks) to calculate constant
database file for given directory from local drivers. The user a and b to convert all time points to m/z ratio. The peak
can control and plot mass spectra for desired rows and each selection on the mass spectrum is shown in Fig. 4. The user
row corresponds to a mass spectrum (Fig. 3). select first hydrogen ion peak and then carbon ion peak in Fig.
4 by using mouse cursor (appearing crosshair).
Fig. 2 In laser diagnostic section the user can read UV-VIS spectrometer for
desired integration time in ms time scale and spectra are plotting real time in
the optical spectrum figure area.
Fig. 4 Reference peak selection on the read MS spectra to perform time of
In the database control section, the user firstly uses “Load flight to mass spectrum transformation.
Database” button to and then select database file stored as a
____________________________________________________________________________________________________________
188
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Following the time of flight spectrum transformation to time of flight space first and after calibration, spectra were
mass spectrum, the user can view all collected mass spectra by transformed into mass space.
using “MS Viewer” button, the opening window seen in Fig. 5. We have got a great success to reduce the numbers of steps
On the other hand, the user can investigate statistical results and time duration spend for the manual data processing
by using PCA visualization tool as seen in Fig. 6. procedures of the mass spectra analysis from weeks to
seconds. Thus, we can eliminate the possibility of the
unintended human errors appears in these manual procedures.
Within the context of this work, we built a backbone of the
flexible program, which can be manipulated by customs needs,
changing experimental setup and/or new unused statistical
procedures in laser TOF-MS.
Analysing fs-TOF-MS data using multivariate statistical
method, which suggests a new opportunity to use this program
in real-time cancer detection research.
ACKNOWLEDGMENT
This work was supported by the; Scientific and Technical
Research Council of Turkey (TUBITAK) under Grant No.
1649B031405880 and Scientific Research Projects
Coordination Unit of Selçuk University, Project No.
Fig. 5 MS Viewer window opens when the user clicks MS Viewer button seen 14201085. Program presented in this article was developed as
in Fig. 3. The user can select desired peaks to build a marker list into the left- a part of the PhD thesis being written by Abdullah Kepcoğlu.
hand side section.
REFERENCES
____________________________________________________________________________________________________________
189
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
190
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Cuckoo search (CS) is one of the recent population- the ability of applying the existing knowledge to look for
based algorithms used for solving continuous optimization better solutions [16]. In other words, exploration represents
problems. The most known problem for optimization techniques global search and the exploitation stands for local search. CS
is balancing between exploration and exploitation. CS uses two has two search techniques and uses Levy flight rather than
search strategies to updating the nest: local and global search.
standard random walks for the global search. Thus, CS can
Although cuckoo search are adequate for the exploration, it is
not well enough the exploitation. Only one search equation is explore the search space efficiently [17]. Although CS has
used for local search, this equation remains incapable and causes local and global search capabilities, it opens to improvement
some deficiencies about the exploitation. To enhance the ability about balancing between exploitation and exploration. In
of exploitation and to balance between global search and local addition, due to some deficiencies about exploitation,
search, different search strategies were implemented in CS strengthening the local search is more important for the CS.
algorithm. The proposed method was compared with basic CS on Many studies were performed to improve performance of
well-known unimodal and multimodal benchmark functions. cuckoo search. The cuckoo search parameters were properly
Experimental results show that the proposed method is more tuned by Valian et al. [18] to enhance accuracy and
successful than the basic CS in terms of solution quality.
convergence rate of the cuckoo search. Walton et al. [19]
proposed modified cuckoo search which involves the addition
Keywords— Cuckoo search, continuous optimization, search of information exchange between the top eggs, or best
strategies
solution. To improve refining ability and convergence rate of
I. INTRODUCTION cuckoo search, Zhang and Chen [20] proposed the cuckoo
search with adaptive method. This adaptive method was used
In the recent years, many novel nature inspired algorithm
to control the scaling factor and find probability to enhance
have been proposed to solve continuous optimization
the population diversity. A new cuckoo search algorithm
problems, alongside mostly known swarm intelligence
based on the idea of opposition (OCS) algorithm is proposed
algorithms such as particle swarm optimization [1], artificial
by Zhao and Li [21] to increase the exploration efficiency of
bee colony [2], ant colony optimization [3] etc. The recent
solution space. They merged the opposition-based learning in
nature inspired algorithm was inspired by behaviours of
the CS and the proposed algorithm fully used beneficial
animals in the nature. To illustrate, the firefly algorithm [4]
information of the best solutions. In order to balance the
was developed by flashing characteristic of fireflies, the bat
exploitation and exploration of the cuckoo search algorithm,
algorithm [5] was proposed by being inspired behaviour of
Li and Yin [22] proposed a new approach which uses two new
echolocation of bats. Yang and Deb [6] proposed the cuckoo
mutation rules based on the rand and best individuals among
search by being inspired breeding behaviour of some cuckoo
the entire population. In addition, this new rules were
species.
combined through a linear decreasing probability rule. CS was
Cuckoo search (CS) is based on the obligate brood
used in hybridization with the other algorithm such as fuzzy c
parasitism of some cuckoo species in combination with Levy
means [23], NEH heuristic algorithm [24].
flight random walk. Cuckoo search has been very popular in a
To improve and enhance the performance of cuckoo search,
short time and started to apply many engineering fields and
we focus on ability of exploitation and some deficiencies of
optimization with promising efficiency and few control
local search. CS uses only one search equation for the local
parameter settings. Due to these advantages, CS was used for
search and this equation remains incapable and causes some
clustering of web search results [7], hydrothermal scheduling
deficiencies about the exploitation. To overcome this problem
[8,9], economic dispatch [10], multilevel thresholding [11],
and balance between global and local search, different search
parameter estimation and optimization [12, 13] , redundancy
strategies were implemented in the CS. Modified cuckoo
allocation problems [14], forecasting solar radiation [15] and
search with the different search strategies (CSDSS) were
so on.
compared with the basic CS on twelve benchmark functions.
The main problem of the optimization techniques is
Experimental results show that the proposed method was more
balancing between exploration and exploitation. The
successful than the basic CS algorithm for both unimodal and
exploration is concerned the ability of autonomously seeking
multimodal functions. By virtue of different search strategies,
for the global optimum, whereas the exploitation is related to
____________________________________________________________________________________________________________
191
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the CSDSS enhanced the efficiency of local search and III. MODIFIED CUCKOO SEARCH WITH DIFFERENT SEARCH
improved the balance between exploration and exploitation. STRATEGIES (CSDSS)
The rest of the paper is divided as follows. In Section 2, the The CS can explore the search space efficiently thanks to
basic CS algorithm is presented. The proposed algorithm is Levy flights random walk. But one solution search equation is
detailed in Section 3. Section 4 gives the experimental results used for the local search and the best solution of nest is
and comparison of the methods. As a final, the paper is ignored by this equation. This causes lack of the exploitation.
concluded with the future works. To strengthen and increase the efficiency of local search
without losing the ability of exploration, different search
II. CUCKOO SEARCH
strategies are implemented in the local random walk phase of
Cuckoo search was inspired by the interesting breeding the cuckoo search.
behaviour such as brood parasitism of certain species of
cuckoos by laying their eggs in the nests of host birds [6]. In xit 1 xit rand (xrt xkt ) i 1,2,..., N, i k r (4)
the CS algorithm, there are mainly three principle rules as
t 1 t t t
follows [6, 17]: x x ( x x ) i 1,2,..., N, i k r
i r i k (5)
t 1 t t j
x x (x x )
i i i best i 1,2,..., N, (6)
– Each cuckoo lays one egg at a time, and dumps it in a t 1 t t t
randomly chosen nest; x x
i best ( x x ) i 1,2,..., N , i k r
k r (7)
– The best nests with high-quality eggs will be carried over
to the next generations; Equation (4) is the same as local search update rule of the
– The number of available host nests is fixed, and the egg basic CS algorithm [17].In (4), (5) [25] and (7) [26], xrt and
laid by a cuckoo is discovered by the host bird with a
probability pa ϵ (0, 1). In this case, the host bird can either xkt are solutions randomly selected from the population at
get rid of the egg, or simply abandon the nest and build a time step t, and k and r are not equal to each other and i. In (6)
completely new nest. t
and (7), xbest in the equations is the best solution obtained by
For convenience, this last rules can be approximated by the the population so far. Also, is a random number in range
fraction pa of the n nests are replaced by new nests (with new of [-1,1] and produced for each cuckoo which is updated at the
random solutions) [6]. Each egg in a nest stands for a solution time step t. To increase and protect the efficiency of
and each cuckoo can lay only one egg. For the simplest exploration, the random neighbours are used in Eq. (4) and Eq.
approach of algorithm, each nest has only one egg. Thus, there (5). Also, Eq. (6) and Eq. (7) are implemented to support the
is no difference between egg, nest and cuckoo. All of them local search around the global best solution of population and
represent one solution. enhance the ability of exploitation. The important point is how
The CS uses two search strategies and two solution search these equations are used together to obtain new solutions.
equation to generate new solutions. One of them is local Usage of these equations is explained by Fig.1.
random walk and other one is global random walk. The global
random walk is performed by the using Levy flights [17] for i=1:number of cuckoo
if rand < ls_rate
if( rand < 0.5)
xit 1 xit L( s, ), (1) Generate new cuckoo with Eq. (7)
else
( ) sin( / 2) 1 Generate new cuckoo with Eq. (6)
L( s, ) , (s s0 0) (2) end
s1 else
if( rand < 0.5)
where α > 0 is the step size of scaling factor and it is generally Generate new cuckoo with Eq. (5)
determined with α=O(L/10), where L is characteristic scale of else
the problem of interest. On the other hand, the local random Generate new cuckoo with Eq. (4)
walk can be written as [17]: end
end
end
xit 1 xit s H ( pa ) ( xtj xkt ) (3) Fig.1 Usage of equations for the proposed method
t t
where x and x are two different solution selected by
j k In Fig.1, ls_rate is the constant parameter in range of [0,1]
randomly, H(u) is Heaviside function, represents random that controls the level of local search for the proposed method.
number drawn from a uniform distribution, pa stands for If the ls_rate is high, the proposed method searches the
switching parameter of controlling the balance between local around of global best solution and makes more local search.
and global random walk and s is the step size. The new nests The proposed method uses the algorithm in Fig.1 instead of
are generated by the both of Eq.(1) and Eq.(3) in the every the Eq. (3) in the basic CS to generate new solutions.
iteration. Detailed information about the cuckoo search, please
refer to [6, 17].
____________________________________________________________________________________________________________
192
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
193
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. CONCLUSIONS
In the basic CS, one solution search equation is used for the
local search and the best solution of nest is ignored by this
equation. This causes lack of the local search. In order to
overcome these deficiencies of the CS algorithm in local
search, this paper proposed the CSDSS algorithm using the
different search strategies. Not only the proposed algorithm
strengthens the ability of local search but also it improves the
balance between exploration and exploitation. The proposed
algorithm was compared with the basic CS on twelve
benchmark functions. The CSDSS performs better results than
Fig.2 Convergence graph of Schwefel2.22 function
the basic CS for all benchmark functions except f3 and f8. The
proposed method is more efficient and more effective than the
In order to compare convergence rates of the methods, basic CS algorithm especially on the unimodal and some of
convergence graphs for CS and CSDSS algorithms are given multimodal functions.
in Fig. 2, Fig. 3 and Fig. 4 for Schwefel2.22, Rastrigin and As future work, the proposed method will be implemented
Griewank functions. The convergence graphs show that the
to optimization problems and different search strategies will
proposed method has rapidly convergence and also obtains
be used for the other nature-inspired algorithms. Furthermore,
successful results. Examining Fig.2, while the CS converges different search strategies can be increased and varied to
slowly, the CSDSS continues the improving solutions thanks improve the success of the proposed algorithm.
to ability of local search. For the Rastrigin function, the
proposed method has early convergence and then gets stuck ACKNOWLEDGMENT
the local minima in Fig.3. Although its early convergence, it
This study has been supported by No. 15611818 Scientific
obtains better result than the basic CS.
Research Project of Selcuk University.
____________________________________________________________________________________________________________
194
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
195
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
196
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
performance of the card. According to their intended use these in the Java Spring Framework, STOMP protocol support with
operating systems, you will need to the shaped and user SockJs library. In this study , it has been the preferred socket
programs have been included in the operating system. In IO library.
addition, 17 December 2012 Raspberry Pi Foundation, in
cooperation with IndieCity and Velocix companies open Pi A. Socket IO
Store for users download application programs. Users located The socket IO's with almost every browser in real time
in the official distribution operating system an application (including mobile devices) in spite of different
using it is possible to download various categories of communication protocols to communicate study is preferred.
applications. In addition, users can also be installed on the On servers that have been created on the node Js platform
application developed and can be made available to other Socket IO is used with ease. The socket IO library created
users after approval. Apart from these operating systems that from object is set to the specified port on the server is made
are not developed by the Foundation, Raspberry Pi Raspberry ready for use by the rest.
Pi announced operating systems are also available. Raspberry
Pi cards that can be installed on these operating systems is B. Real Time Event Model
mostly Linux-based operating systems. On the Windows The Socket IO creates web socket objects. Sockets are
platform is Windows 10 has started to be supported by Pi card created when request by client to the web server-side on the
IoT version of Raspberry. In this study; Raspberry Pi used port opens a channel between the client and service (Fig 2).
highly efficient and compatible running Debian Linux
[5]. Selected operating system is written directly using the
appropriate SD card program. Raspberry Pi 2 version does not
contain any key to open and close power is transferred to the
device SD card operating system searches for and finds the
bootloader want to run. The card slot SD card prepared with
Raspberry Pi is used.
B. Node Js
Node Js in 2009 Joyent company by Google's V8 Javascript
engine is produced with a typical run-time environment to be
run by the server. Node Js is scalable, event-driven event-
driven asynchronous, non-blocking I/O uses working model
[6]. This also allows you to run with high
performance. LinkedIn mobile server side after passing to
Node Js Server dropped the number from 3 to 30 and up to 20
times faster in some operations in masterminding. Node Js are Fig. 2 The socket IO client/server connection model
modular. As a module each Node Js library and shrink the size Connecting to a socket request pop-up web posted by all
of the nucleus so that it allows applications to run faster clients through data communications same channel (Fig 3).
[7]. Downloaded the file from the official website of Node Js
platform is easy to install. According to other programming
languages is the biggest advantage is the package manager
Node Js NPM. Ask if you want to do what you want probably
as module. NPM is located and easily the desired with the
search over the package is used by adding to the project. This
package is one of the Socket IO this work was used in the
acquisition of data with the web socket.
____________________________________________________________________________________________________________
197
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
1) Multicast: The message to be sent to clients only is Led used on pertinax plate gives information about web
desired. socket communications.
2) Unicast: Exchange of messages between client-
server only point-to-point.
3) Broadcast. The message is sent to all clients in the
channel.
____________________________________________________________________________________________________________
198
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. CONCLUSIONS
Raspberry Pi this system based on device that in terms of
cost cheap and simple founded, without need for any
additional software development ,it offers real-time video
which can be monitored from any web browser or your mobile
____________________________________________________________________________________________________________
199
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Today flexible manufacturing systems are highly Any operation j (j∈Oi) of job i (i∈N) can be processed by
popular due to their capability of quick response to customer a machine k of machine set Mij with process time tijk,
needs. Although the advantages of flexible manufacturing systems One machine can only process one operation of a job at a
cannot be denied, these systems also bring new issues on
production planning side. Especially assigning machines to
time and no preemption is allowed,
production operations and scheduling these operations with The precedence relationship between the operations of a
respect to machine constraints turn out to be an NP-Hard job should be satisfied.
problem. In this study, the integrated process routing and The objective is minimizing the completion time of the last
scheduling problem is explained, and the performance of two operation (makespan) of the schedule.
different meta-heuristic techniques, which are genetic algorithms By introducing the following additional variables and
and simulated annealing, are compared in terms of solution time parameters, this problem can be modeled as below:
and quality.
∶ ℎ
Keywords— Optimization, Integrated Process Planning and : ℎ
Scheduling, Job Scheduling, Simulated Annealing, Genetic ( ∈ , ∈ , ⊂ )
Algorithms 1, if operation j of job i is processed
= by machine k (i ∈ N, j ∈ O , k ∈ M , M ∈ Q
I. INTRODUCTION
0 otherwise.
Today, as the flexible manufacturing systems get more 1, if operation j of job i starts before
popular, scheduling production activities becomes more = operation of job ( , ∈ N, j ∈ O , m ∈
complex. The production managers have to make decisions 0 otherwise.
such as: “Identifying the production operations that the jobs ` ∶
should go through”, “Assigning machines to job operations”, ∈ , ∈ , ≥0
and “Identifying the sequence job operations on the machines”. ∶
Some of these decisions are related to the routing of the jobs on
( ∈ , ∈ , > 0)
the shop floor and some of them are related to machine
scheduling. These multiple aspects of production planning : ( > 0)
make the decision making much harder. Although it is possible ∶
make decisions sequentially, this approach leads to a sub
optimal solution. In order to find the best solution that Objective function:
minimizes the completion time of all jobs (makespan), the (1) Minimize
solution algorithm should take into account all aspects of the Subject to:
problem. This integrated production planning problem is (2) ∑ ∈ =1 ∀ ∈ ,∀ ∈
referred as integrated process routing and scheduling problem (3) = +∑ ∈ ∀ ∈ ,∀ ∈
in the literature. (4) ≥ ∀ ∈ ,∀ ∈ ,∀ ∈
It is possible to solve integrated process routing and (5) − ≥ −1 × ∀ ∈ , , ∈ N, j ∈
scheduling problem optimally using the following
O ,m ∈ : ∈ ∩
mathematical programming model given by Botsalı and Şeker
[1]. The model is described as follows: (6) + =1 ∀ ∈ , , ∈ N,
j ∈ O ,m ∈ : ∈ ∩
There is a set of jobs N and each job i∈ N has a set of (7) + − − ≤ 1 ∀ ∈ , , ∈ N, j∈
operations Oi to be completed, O ,m ∈ : ∈ ∩
____________________________________________________________________________________________________________
200
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
201
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Finally when two algorithms are compared in terms of the genetic algorithm a little bit by reducing the number of
solution time, the simulated annealing algorithm’s solution generation parameter value from 150 to a lower value. Yet, still
time is in general 90% shorter than the solution time of the it is not possible for the genetic algorithm to be as effective as
genetic algorithm. In fact this is something expected since the the simulated annealing algorithm in terms of computation time.
number of computations is higher in the genetic algorithm. TABLE III
TABLE I COMPARISON OF COMPUTATION TIME
COMPARISON OF MAKESPAN SOLUTIONS
% Improvement in Genetic
Algorithm Solution Time
%
Simulated
Genetic Improvement in
Annealing
Instance No
Algorithm Genetic
Algorithm
Algorithm
% Difference in Average
Instance No
% Difference in Best
Makespan Value
Makespan Value
Avg. Makespan
Avg. Makespan
Best Makespan
Best Makespan
____________________________________________________________________________________________________________
202
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSION
In this study, two different algorithms are compared for the
integrated process routing and scheduling problem. These two
algorithms are based on genetic algorithm and simulated
annealing algorithm techniques. The algorithms are tested over
the problem instances provided by Kim et. al. [2]. It is observed
that the simulated annealing algorithm has better performance
than the genetic algorithm in general.
It is possible to extend this study in various directions.
Different assumptions such as the allowance of pre-emption on
shop floor can be considered. Also the performance of different
heuristic algorithms such as particle swarm optimization or
tabu search can be compared.
REFERENCES
[1] A. R. Botsalı And A. Şeker, “A Scheduling and Rescheduling Algorithm
for Integrated Process Planning and Scheduling Problem”, Necmettin
Erbakan University, Konya, Turkey, Working paper, 2016.
[2] Y. K. Kim, K. Park, and J. Ko, “A symbiotic evolutionary algorithm for
the integration of process planning and job shop scheduling”, Computers
& Operations Research, vol. 30, pp 1151–1171, 2003.
[3] H. Lee and S. Kim, “Integration of process planning and scheduling
using simulation based genetic algorithms”, International Journal of
Advanced Manufacturing Technology, vol. 18, pp 586–590, 2001.
____________________________________________________________________________________________________________
203
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— 3GPP LTE-Advanced release-13 is now supporting to improve the broadband user experience but also increase the
the deployment of femtocells (HeNBs), which provide better coverage and capacity of the cell in cost effective manner [3].
connectivity to the users (UEs) at home, offices, shopping malls,
and dense urban areas, where the macrocell (eNB) has weak
signals strength or no signal. These femtocells cover short
distance (10 ~ 20 m) with high throughput and increased
capacity, while on the other hand the deployment of huge
MME / S-GW MME / S-GW MME / S-GW
number of femtocells create co-tier and cross-tier interference
S1
issues. To mitigate the co-tier interference problem we propose a S1 S1
Transmit Power Control (TPC) Algorithm, which will establish
and control the connection between femtocell and the users. In
S1
S1
S1
case of downlink and uplink scenarios the transmit power of
S1
S5
S1
S1
femtocells and UEs will be monitored accordingly. As a result HeNB GW
this data will help to manage the switching of users among X2 X2
eNB eNB X2 GW E-UTRAN
femtocells, macrocells to femtocells and vice versa. In the end we X2
S1
show the simulation results to compare the SIR and Throughput
S1
X2
X2
X2
of users.
X2
X2
eNB HeNB X2
Keywords— 3GPP, LTE-A, Femtocell, Macrocell, RRM, UE, X2
X2
eNB, HeNB, SIR, Throughput
HeNB HeNB
I. INTRODUCTION
Fig. 1. Source: 3GPP Release 13 [6]
3GPP LTE-A is a very promising 4G technology for the
thirsty mobile cellular network. With the advancement of Small cells can support wireless applications for homes and
mobile communication and technology, the expectations for enterprises as well as metropolitan and rural public spaces. Due
high data rates are also increased. Due to the huge growth of to the smaller coverage area, the same licensed frequency band
smart phones, tablets, and notebook etc. the quality of service can be efficiently reused multiple times within the small cells
(QoS) and ubiquitous connections are required. On the other in a HetNet (Figure 1), thus improving the spectral efficiency
hand the number of subscribers increasing every year per unit area (and hence the capacity) of the network. In a
dramatically. According to [1] Global LTE passed 1 billion HetNet, small cells are envisioned as traffic off-loading spots
subscriptions in 2015 and will reach up to 3.62 until 2020. Due in the Radio Access Network (RAN) to decrease the
to the explosive growth of users and multimedia services the congestion in macrocells, and enhance the users’ QoS
main challenge for next generation cellular networks is to experience [7].
oblige mobile data traffic by improving the capacity of the
networks [2]. The present LTE-A system is based on There are different types of small cells included like
homogeneous network, where every single eNB cover the femtocells, picocells, and microcells, in this paper we focus
whole cell, and each eNB use similar transmission power only on femtocells, which are denoted by Home evolved Node
levels, modulation techniques, access schemes, antenna Base-Station (HeNB) in LTE-A, this low power device can be
patterns to offer QoS to the UEs across cell [3, 4]. However, installed at Homes, Shopping Malls and Congested Urban
such deployment degrades the coverage and capacity of the Areas and managed by the user himself. It covers up to 20 to
cell-edge users. 30 meters distance of space; it is very useful for indoor
communication where macrocell (eNB) coverage is not enough
One approach to solve the above mentioned problem is to for customers. The main advantages of using HeNBs are low
shrink the cells, it provide the high data throughput and can cost, IP backhauling and frequency reuse; provide high
fulfil the increasing demand of cellular systems, which will capacity, Low battery usage and seamless transition between
also improve the interference and signal to noise ratio (SNR), wired and wireless connecting devices [8]. However, in order
but this approach is not economical and as a result its need lot to successfully deploy the femtocell architecture several
of expenses [5]. Therefore, the deployment of heterogeneous challenges need to be addressed. Interference management
network (HetNet) is more scalable and beneficial for both remains one of the major issues: femtocells which operate in
operators as well as users. This approach is expected not only the same spectrum of macrocell users produce a cross-tier
interference which degrades their transmissions. Moreover,
____________________________________________________________________________________________________________
204
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
neighbour femtocells which belong to same operators, which III. EXISTING NETWORK AND ITS PROBLEM
may interfere each other. The latter interference is known as According to [06], the users can install femtocells under a
co-tier interference [9]. In this article we propose an algorithm macrocell and share the same spectrum. In this paper, we
for co-tier interference management to avoid the interference studied the downlink and uplink co-channel interference when
between neighbour HeNBs and UEs. macrocell and femtocell share the same spectrum. The
In the preceding sections we discuss different approaches deployment configurations are CSG and co-channel.
and techniques use for RRM; in Section III we proposed an Specifically, the downlink transmissions from the femtocell
algorithm, in Section IV we have simulation results and will cause strong interference at a nearby macrocell’s UE and
Section V is the conclusion. may cause the received macrocell signal at the macrocell’s UE
degraded and become unacceptable. Hence, the co-channel
II. RELATED WORK interference is one of the major barriers for the successful co-
existence of macrocell and femtocell. We address the downlink
These days the Radio Interference Management (RRM) in and uplink interference problem by considering the transmit
LTE-A femtocell is the most challenging issue for cellular power at both the macrocell’s UEs and the femtocell’s UEs in
network operators. This is also a very hot topic for wireless terms of their received SINR.
communication researchers. Some of the works done by
researchers are discussed below.
IV. PROPOSED ALGORITHM
The authors of the article [11] explained the orthogonal and As we discussed in above section many techniques are
co-channel frequency allocation schemes, compared them and implemented in LTE-A femtocell networks to mitigate the
also show the simulations results. According to the distance interference problem, all these techniques are working very
between HeNB and eNBs which are two coverage areas, where good but they have some limitations as well. In this paper we
the spectrum of eNBs is different from the HeNBs. The [12] proposed Transmit Power Control (TPC) Algorithm, which
use the hybrid approach for both uplink and downlink would reduce the co-tier interference for uplink and downlink
interferences, in this article the use of shared spectrum is of the LTE-A femtocell system. The “Interference Avoidance”
exploited only when this approach is favourable both to eNB is used to reduce the system interference. Our proposed
and HeNB users; otherwise, the partitioned approach is chosen. algorithm will avoid the interference based on the threshold
A power control method are proposed in [13], the author value of Transmit Power of User Equipment (PTUE) and Home
explain the uplink and downlink scenario, which limits the co- evolved Node Base-station (HeNB). PTUE represents Transmit
tier interference in, while the encroachment of co-tier Power of UE while PTHeNB represents Transmit Power of
interference is ignored. In [14] two possible solutions are HeNB and their values are pre-calculated. The PTUE of one
highlighted, the first-one is to use the separate spectrum for specific HeNB is changing by the other HeNB which are
HeNB and eNB, this technique is said to be Dedicated Channel deployed near to it, and these power levels are expressed in
deployment, while the other approach is to use the same dbm. The uplink interference could be controlled and
spectrum for HeNB and NBs, this is kwon as Co-Chennel maintained by the HeNB Gateway Server, the UE under the
deployment technique. These solutions are good to sufficiently particular HeNB should transmit the assign power as soon as
increase the overall capacity of the system, but the Co-Channel possible, because the HeNB will connect on First Come First
interference issue remains the same, and the deployment is Serve (FCFS) basis. The HeNB uses pilot signal to detect the
costly for operators. Whenever HeNB and eNB share the same UE in its coverage area and set the Pilot bit (Pbit) to 1. After
part of spectrum, they will interrupt each other. Article [15] successfully established the connection the HeNB will instruct
proposed a systematic way to optimize the resource allocation the UE to communicate at assigned transmit power level. In
for public femtocell networks, including three schemes of case of downlink, the HeNB power levels are registered on
different stages: (1) long-term resource management, which is start-up in femtocell gateway (FG) center and the power within
to allocate spectrum resource between eNB and HeNB the power level assigned to each downlink channels of the
networks; (2) medium-term resource management, which is to particular HeNB it handles.
allocate radio resources to each HeNB; (3) fast resource
management, which is to further enable multiple HeNBs to
cooperate to improve the network’s coverage and capacity. As V. FLOW CHART
we know that cross-tier interference can also be minimize by The backhaul of LTE-A femtocell and connections are
interference cancellation, and this method is by [16] as the explained in [19], the HeNB is initially connected with FG
class of techniques that demodulate/decode desired Server through ADSL or fiber optic and then it will further
information, and then use this information along with channel connect to ISP. The figure 2, explaining the algorithm how it
estimates to cancel received interference from the received works.
signal. The key technique for interference avoidance is
Transmit Power Control, especially when many HeNBs are Step 1: When the femtocell (HeNB) is Switch ON. It will
installed very close to each other. One such approach is refresh/update its database and all values will set to “0”. At the
discussed in [17], where the transmit power of a HeNB can be same time the HeNB will be registered in FG, which will
utilized accordingly, while the outdoor eNB’s UE can be assign a unique identification number (ID) to the particular
protected sufficiently. In article [18], the authors use a simple HeNB to recognize with its Transmit Power (PTHeNB).
power control method to avoid the interference cause nearby The HeNB will adjust the UE to set the specific transmit
eNB’s users. power level.
____________________________________________________________________________________________________________
205
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Step 2: The HeNB will set PTHeNB to 1, while in FG the interference of the signals. To avoid the co-tier interference
(PTHeNB) will set on First Come First Serve (FCFS) basis. problem, we proposed an algorithm and can see the result of
UEs and HeNB which are using different level of transmission
Step 3: For Uplink there are many UEs i.e i=1,2,3,…,N and power. There are four different users (UE 1, UE 2, UE 3 and
single HeNB. If the set value of PTHeNB is less than PTUEi, UE 4) and they are using different transmission power. In the
the connection will establish, otherwise the PTHeNB will be simulation we also use three different groups according to
set according to new PTUE. populations of the UEs, which are served by macrocell i.e
Step 4: If the new PTUEj is greater than PTUEi, then the (eNB). The three different groups are divided by Dense
PTHeNB will set according to new value, otherwise rejected. Populated (DP Group), Medium Populated (MP Group), and
Less Populated (LP Group).
Step 5: Downlink side, the FG will set the threshold value of
HeNB in its database. If the PTHeNB is less than the threshold
PTTH value, the connection will establish, otherwise rejected.
____________________________________________________________________________________________________________
206
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
see in this figure, that the UE 1, UE 3 and UE 4 have high [4] HetNets with cognitive small cells: User offloading and distributed
throughputs, while the UE 2 has low throughput. From these channel allocation techniques by H. ElSawy, E. Hossain, and D. I. Kim.
results we can say, that high transmit power has high [5] Achieving High Capacity with Small Cells in LTE-A by Yufei W.
Blankenship, Member, IEEE Research In Motion - Advanced
throughput; while low transmit power has low throughput. Technology.
Similarly DP Group has low throughput, while LP Group has [6] 3rd Generation Partnership Project; Technical Specification Group
high throughput. On the other hand if there is high transmit Radio Access Network; Evolved Universal Terrestrial Radio Access (E-
power and LP Group the delay would be quite low, and vice UTRA) and Evolved Universal Terrestrial Radio Access Network (E-
versa. UTRAN); Overall description; Stage 2 (Release 13). 3GPP TS 36.300
[7] Channel Assignment and Opportunistic Spectrum Access in Two-tier
Cellular Networks with Cognitive Small Cells by Hesham ElSawy and
Ekram Hossain Published by Globecom 2013 - Wireless Networking
Symposium.
[8] Enhanced Cognitive Radio Resource Management for LTE Systems by
Ismail AlQerm, Basem Shihada, and Kang G. Shin Dept. of Electrical
Engineering and Computer Science, University of Michigan, USA.
[9] A Radio Resource Management Scheduling Algorithm for Self-
Organizing Femtocells by Antonio De Domenico and Emilio Calvanese
Strinati, Published 2010 IEEE 21st International Symposium on
Personal, Indoor and Mobile Radio Communications Workshops.
[10] WiMAX Femtocell, Requirements, challenges and solutions by Ronny
YK and Jin Sam Kwak, Kamran E
[11] Resource allocation for orthogonal and co-channel femtocells in a
hierarchical cell structure Kyongkuk Cho , Wooju Lee , Dongweon
Yoon , Kwangmin Hyun , and Yun-Sung Choi, Published in The 13th
IEEE International Symposium on Consumer Electronics (ISCE2009.
[12] Hybrid spectrum sharing for coexistence of macrocell and femtocell by
Y. Bai, J. Zhou, and L. Chen, Published in IEEE International
Fig. 5. HeNB-3, Throughput and Delay Conference on Communications Technology and Applications, 2009.
[13] Performance of Macro and Co-Channel Femtocells in a Hierarchical
From the above simulation we can describe, that transmit Cell Structure by Holger Claussen Bell Laboratories, Alcatel-Lucent,
power of femtocell (HeNB) is very important and will decide Published in the 18th Annual IEEE International Symposium on
the commucation between HeNB to HeNB and UEs. The Personal, Indoor and Mobile Radio Communications.
transmit power of both UE and HeNB will decide the [14] Interference Management and Performance Analysis of UMTS/HSPA+
connections between them. In our proposed algorithm, we use Femtocells by Mehmet Yavuz, Farhad Meshkati, and Sanjiv Nanda,
the transmit power of femtocell (HeNB) as a decider for both Qualcomm Inc. Akhilesh Pokhariyal and Nick ohnson, ip.access Ltd.
Balaji Raghothaman and Andy Richardson, Airvana Inc. White Paper.
Uplink and Downlink scenarios. Higher the tranmit power has
[15] Radio resource management for public femtocell networks by Yizhe Li,
higher chance of connectivity; lower transmit power will be Zhiyong Feng, Shi Chen, Yami Chen, Ding Xu, Ping Zhang and Qixun
rejected. Zhang, Published in EURASIP Journal on Wireless Communications
and Networking 2011
VII. CONCLUSION [16] Interference cancellation for cellular systems: A contemporary overview
by J. G. Andrews, White Paper, IEEE Wireless Commun., vol. 2, no. 3,
The evolution in LTE-A, the deployment of femtocells pp. 19–29, April. 2005.
(HeNBs) is of great improvement due its efficiency and low [17] Optimization of femtocell network configuration under interference
power consumption. In this article we proposed a Transmit constraints Kwanghun Han, Youngkyu Choi, Dongmyoung Kim,
Power Control (TPC) algorithm, which can work very Minsoo Na, Sunghyun Choi, and Kiyoung Han.
intelligently to decide the connection between HeNB and UEs. [18] A downlink power control scheme for interference avoidance in
femtocells by Talha Zahir, Kamran Arshad, Youngwook Ko and Klaus
Using this algorithm the co-tier interferences will eliminate and Moessner
provide better performance. In the simulation section we
[19] A Power Control Algorithm for 3G WCDMA System by Loutfi Nuaymi,
discussed the importance of transmitting power of UEs and Xavier Lagrange, Philippe Godlewski
HeNBs in LTE-A femtocell networks; if the power is high the [20] Copyright 2009-2013 The MathWorks, Inc.
throughput is also good. In future we can propose some
techniques (algorithms) for cross-tier interference which also
affect the performance of femtocell network.
REFERENCES
[1] Source: AVUM, Reported by Mark Newman 27th January 2016
http://www.ovum.com/press_releases/global-lte-subscriptions-pass-1-
billion.
[2] Book: Radio Resource Management in Multi-Tier Cellular Wireless
Networks by Ekram Hossain, Long Bao Le and Dusit Niyato.
[3] On Cognitive Small Cells in Two-Tier Heterogeneous Networks by
Hesham ElSawy and Ekram Hossain Published in 9th International
Workshop on Spatial Stochastic Models for Wireless Networks 2013,
May 13, 2013.
____________________________________________________________________________________________________________
207
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Customers and products are the main assets for every way that an operator can react (including ‘do nothing’) in
business. Companies make their best to satisfy customers terms of retention [1]. Previous studies have indicated that the
because of coming back to their companies. After sales service cost of gaining new customers is much expensive than the cost
related to different steps that make customers are satisfied with of retention the existing customers [1], [2], [3], [4], [5], [6]
the company service and products. After sales service covers
Loss of customer causes negative effects on firms reputations
different many activities to investigate whether the customer is
satisfied with the service, products or not? Hence, after sales and income reduction [7]. Companies are increasingly focused
service is acting very crucial role for customer satisfaction, on accurate customer churn forecasting models. These models
retention and loyalty. If the after sales service customer and should be enhanced to identify the factors to churn and the
services data is saved by companies, this data is the key for developed needs to retain customers. Churn of customer is
growing companies. Companies can add value their brand value defined as the tendency of customers to interrupt doing
with the managing of this data. In this study, we aim to business with a firm in the time process. Customer churn has
investigate effect of 6 factors on customer churn prediction via become a significant problem today. In worldwide, one of the
data mining methods. After sale service software database is the main challenges is determined to propensity of churn of
source of our data. Our data source variables are Customer
customers. Geppert (2003) suggest another list of causes of
Type, Usage Type, Churn Reason, Subscriber Period and Tariff
The data is examined by data mining program. Data are churn: Price, Service quality, Fraud, Lack of carrier
compared 8 classification algorithm and clustered by simple K responsiveness, Brand disloyalty, Privacy concerns, Lack of
means method. We will determine the most effective variables on features, New technology introduced by competitor, New
customer churn prediction. As a result of this research we can competitor enter the market, Billing or service disputes [8] .
extract knowledge from international firms marketing data. High customer churn risk is one of the stringent challenges
Keywords— Data Mining, Customer Churn Prediction, Customer in telecommunication companies. Pareek has classified
Satisfaction, Knowledge Discovery in Database telecommunication industry challenges in four groups;
Consolidation, Competition, Commoditization, Customer
I. INTRODUCTION
service. Telecommunication data which is very complex has
Telecommunication companies maintain and store needed many prepocessing for analyse. Data mining which is
tremendous amount of data about the customer information, a useful tool, is extracting and exploring information from
phone calls and the operations of their networks. Due to the data. In this paper, we will provide an application to
improvement of computer systems and telecommunication telecommunication companies. We want to have knowledge
technologies, this industry has expanded rapidly. Data mining about customers that who are existing and loyal to company,
helps to improve quality of service, detect the customer who are going to leave or quitting from company
communication type, determine deceitfulness activities and products/services.
make better use of resource in telecommunication industry. Customer churn probability is predicted by data mining
One of the widely applied areas of data mining is Customer algorithms analysing historical data. Data mining techniques
Relationship Management. Customer Relationship used for this purposes typically utilize the average number of
Management approach is focuses on retention, relationship calls, billing data, the change in the average number of calls,
development and increase satisfaction. From a business call detail data, subscription and customer information [10].
intelligence perspective, churn management process under the In this paper, we investigate features of applying the data
customer relationship management (CRM) framework mining in customer churn prediction of a telecommunication
consists of two major analytical modelling efforts: predicting data. First, we identify the variable of effecting customer
those who are about to churn and assessing the most effective
____________________________________________________________________________________________________________
208
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
churn prediction of a telecommunication system. These Data sampling randomly selects a set of customers with the
variables are examined in data mining algorithms. Data required information, according the definition of churn in
mining algorithms are compared with each other. Proposed designated telecommunication company.
algorithm is also discussed and we conclude the paper with Evaluate factors and sub factors distribution as shown in
conclusions. Fig. 1.
A. Data Set
The dataset that we used obtained from a Turkey’s one of
the biggest telecommunication company. Real time data was
obtain on May 2016. Untreated data consisted of 6 factors, 66
sub factors and 498,866 instances. 38,36% of data were
churners.
Evaluate factors is following;
TABLE I
FACTORS
According to (Acker, et al., 2013), data pre-processing can In the study we used some of pre-processing techniques. In
be up to 80% of the total analysis work and analyzing the data, data cleaning step we clean some of the data, which are
once joined, cleaned and transformed consumes just about identifying incorrect information or incorrect. At the end of
20%. The raw world data is usually noisy, inconsistent and this step, we have 498.356 data. Ordered sub factors are
incomplete due to their typically large size and their likely converted to numeric values. We create some of groups of
origin from multiple and heterogeneous sources. Dimensions subscribe period for subscriber time.
of data quality; Accuracy, Completeness, Consistency
Timeliness, Believability, Value added, Interpretability B. Data Mining Techniques in WEKA
Accessibility. Data mining technique in CRM is usually apply in real
world case because CRM on data mining have attracted both
the practitioners and academicians. In CRM there are several
____________________________________________________________________________________________________________
209
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
different functionalities, techniques and applications. as The results from TABLE III have been analyzed the
shown in TABLE II. classifiers work better. The classifiers J48, Random Forest and
PART have performed better in data set (up to %78). Errors
TABLE II
and Kappa statistic seem to be same among 3 classifiers and
DATA MINING FUNCTIONALITY, TECHNIQUE AND APPLICATION are based on the accuracy of the prediction.
Customer clustering is very important issues in data mining
Functionality Technique Application
Set Theory
methodologies for customer relationship management (CRM).
Association Statistics Cross Sell To segment customers by Customer Type, Usage Type,
Bayesian Classification Churn Reason, Subscriber Period and Tariff as variables and
Neural network used K-Means to model the customers into five clusters. To
Estimation Statistics Exchange rate estimation generate roughly the same number of subscribers in each of
Time series
the seven clusters. Table IV is summarized the clustering
Decision Tree
Fuzzy Credit embezzle results. Simple K-Means method is a common and effective
Classification method for clustering.
Neural network Market segmentation
Genetic Algorithm
TABLE IIIV
Regression
Churn Prediction CLUSTER OF CUSTOMER SEGMENTATION
Prediction Neural network
Fraudster prediction
Decision Tree Full
Cluster 0 1 2 3 4 5 6
Neural network Data
Statistics
Segmentation Market segmentation Attribute 498356 85008 109671 65748 86170 67292 62389 22078
Genetic Algorithm
Decision Tree % 100 17% 22% 13% 17% 14% 13% 4%
Active_
Active Active Active Churn Churn Active Active Churn
For churn prediction is used most popular ones are: Churn
• Decision trees H_B H H H B H H H B
• Artificial Neural networks Period 6-10 0-5 6-10 6-10 6-10 16- 11-15 0-5
• Regression.
Weka (Waikato Environment for Knowledge Analysis) Tariff 7 4 14 1 14 7 7 3
____________________________________________________________________________________________________________
210
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSIONS [10] G.M. Weiss. “Data Mining in Telecommunications’, Data Mining and
Knowledge Discovery”, Springer Science, New York, 2005, pp.1189-
1201
With the strongly development of telecommunication [11] O. Acker,., Blockus, A. and Pötscher, F., 2013. Benefiting from Big
industry, the service providers needs more knowledge of the Data: A New Approach for the Telecom Industry, s.l.: Booz &
Company.
subscriber. In today’s very hard competitive environment, [12] R.I. Jony, Preprocessing Solutions for Telecommunication Specific Big
holding of existing customers has become an enormous Data Use Cases, AALTO UNIVERSITY School of Electrical
challenge. Churn analysis is used to predict of customer Engineering Department of Communications and Networking, Master's
behaviours that are most likely to change provided service and Thesis.
to compose special marketing tools for them.
Churn prediction in customer relationship management is
critical issues in telecommunication industry. For the purpose
of competitive in this industry, service providers must be able
to predict probability of churners and take proactive approach
to retain existing customers. In this research, we use WEKA
data mining algorithms for purpose of different classification
techniques and clustering for churn prediction.
This study is inspected determinants of customer churn in
the Turkish telecommunications industry service market using
a sample of have 498.357 actual and churn customer data. The
efficiency and the performance is compared of Naive Bayes
and Bayes net from Bayes Classifier, Multilayer Perceptron
from functions, JRip, PART, OneR from Rules Classifier and
Random Forest and J48 from Trees.
The goal of the paper is define and explain the related
factors in active and churn prediction modelling. On the other
hand this study aims to cluster the application data with data
mining simple k means clustered technique. Classified and
clustered that uses for churn prediction in Data Mining.
ACKNOWLEDGMENT
The dataset were obtained one of the biggest
telecommunication companies in Kayseri, Turkey.
REFERENCES
[1] S.Y. Hung, D.C. Yen and H.Y. Wang, “Applying data mining to
telecom churn management”. Expert Systems with Applications, vol. 31,
pp.515–524, 2006.
[2] C. Kirui, L. Hong, C. Wilson, H. Kirui, “Predicting Customer Churn in
Mobile Telephony Industry Using Probabilistic Classifiers in Data”
Mining. IJCS, 10, 2013.
[3] R. Mattison, The Telco Churn Management Handbook, Oakwood Hills,
Illinois: XiT Press, 2005.
[4] C. P. Wei and I.-T. Chiu, "Turning telecommunication calls details to
churn prediction: a data mining approach," Expert Systems with
Applications, vol. 31, pp. 103-112, 2002.
[5] B. Huang, M. T. Kechadi and B. Buckley, "Customer churn prediction
in telecommunications," Expert Systems with Applications, vol. 39, p.
1414–1425, 2012.
[6] Basiri, Taghiyareh and Moshiri. A Hybrid Approach to Predict Churn.
Services Computing Conference (APSCC), 2010 IEEE Asia-Pacific,
2010, pp.485 – 491
[7] Saradh and Palshikar. Employees churn prediction. Expert Systems
with Applications, vol. 38,2011, pp. 1999-2006
[8] K. Geppert. Customers churn management. KPMG International, A
SWISS Association, 2002.
[9] H. Edelstein. Building Profitable Customer Relationships with Data
Mining, Two Crows Corporations, 2000. [Online].Available:
https://books.google.com.tr/
____________________________________________________________________________________________________________
211
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Road defect is one of the most important factor for The crack in the image taken and the contrast between the
traffic accident. Therefore, this defects should be corrected as road are reduced due to the light on the road surface and the
soon as possible. It usually occurs cracks, rutting, and potholes changes in weather conditions [3]. It is difficult to perform
in road surface. There are various methods in the literature for feature extraction from the road surfaces using image
the road defect detection. Traditionally these defects is tried to processing methods and to detect feature due to the plurality
detected by the human eye are performed using image processing of the colour changes of the road surfaces [4]. Edge detection
in recent years. However, there are deficiencies such as inability methods are very efficient for feature extraction in the other
of real time application, slow work, and inability to identify with
detection problems [5-7] ,but performing feature extraction
high accuracy in addition to being shortage of studies using
image processing. from the road surface is not only edge detection or texture
analysis problem [8]. Color changes and noise in received
Aim of this study is defect detection at road by using image image make difficult to defect detection. In this case,
processing algorithm with images taken from a camera on a improvement of the image becomes an important duty in order
vehicle. In first step of this study, preprocessing is performed by
to make an accurate prediction [9]. The road cracks constitute
utilizing median filter algorithm. Then, in second step essentially
threshold values to detection process are obtained by performing the continuous and darkest region of the image [10]. In the
feature extraction with mathematically morphological study carried out, the cracked structure was brought to the
techniques. In final step whether road defect exist or not, and forefront to improve the image by using a median filter and
which level of defect are obtained by being classified in real time. collecting with an image on which no processing was
To reveal the accuracy and performance of the proposed performed. In the literature, productive results were also
approach, comparative results are given by examining images obtained by using median filter [12,13] to improve image. In
obtained as experimentally. addition to median filter, histogram equalization [8], calculate
the threshold value [10], etc. were used for image improving.
Keywords— Road defect detection, image processing, Feature extraction defines the relevant shape information
morphological operation, decision tree, median filter in the image [11]. In the literature, edge control [14] and
various parameters such as slope, inertia [10] which can be
removed through image were used for feature extraction. For
I. INTRODUCTİON instance, Guanqun Bao [12], proposed feature mapping
method. This method was based on from vertical histogram
and horizontal histogram values.
Roads are the transportation method which is most widely
used around the world at the present time. One of the most The usability of road defect detection applications depends
important causes of accidents on the roads is the road surface on being able to achieve fast and good results. In accordance
faults. The improper operations in the construction phase and with this purpose, an attempt to achieve faster and better
the reasons such as pressure caused by heavy vehicles on the results was made in the study carried out. If deep cracks on
roads and weather condition lead to the deformations on the the road are not repaired in a short time, the damage on the
road surfaces over time. These deformations on the road road is further increased. Therefore, it is necessary to make
surfaces are called Alligator Cracking, Block cracks, Linear the distinction of superficial and deep cracks. In the study
Cracking, Transverse Cracking, Joint Reflection Cracks etc. carried out, roads were classified as without crack, superficial
crack, crocodile cracking, transverse cracking and longitudinal
The elimination of the road defects is both a big work load (linear) cracking.
and an organizational operation [1]. Today, road defect
detection based on the fact that people see these defects. This
method prevents the elimination of defects in a short time.
Many researchers have proposed different methods about how II. PROPOSED METHOD
to detect the defects in a faster way. The features such as
being cheap, being fast and and applicable determine the The proposed method is based on feature extraction and
effectiveness of a method [2]. Among these methods, one of classification through the image obtained by bringing cracked
the cheapest and fastest method is the algorithms developed structure into the forefront. The flow chart of the study carried
using image processing methods. out is presented in Fig. 1.
____________________________________________________________________________________________________________
212
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(a) (b)
(c)
____________________________________________________________________________________________________________
213
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Transverse
Y -45 < Gradient of Crack < 45 N Linear Crack
Crack
____________________________________________________________________________________________________________
214
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
215
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
216
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
217
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
218
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Railway Images
ROI Rail Track Switch In the next step, Hough transform is carried out to binary
Segmentation Extraction Detection image Binary(i,j) in order to detect rail lines. Hough transform
returns lines that have different lengths. Hough transform is
Figure 8. Block diagram of switch detection algorithm very helpful in order to detect straight lines. Lines returned
include their end-point coordinates without line length
A. ROI Segmentation Method information. The length of a straight line between its two end
The flowchart of ROI segmentation algorithm is shown in points (𝑥 1 , 𝑦1 ) and ( 𝑥 2 , 𝑦2 ) is computed by Equation 2.
Fig. 9. To extract region of interest area, we compute to the
𝑙𝑒𝑛𝑔𝑡ℎ = √ ( 𝑥 2 − 𝑥 1 )2 + ( 𝑦2 − 𝑦1 ) 2 (2)
railroad vanishing point p and track width parameter 𝑊𝑚𝑎𝑥
from an input image. Rail tracks are aligned parallel to each This operation specifies many lines including rail track lines.
other and their width decrease linearly in the image from bottom But the lines expect for rail track lines are also available. Thus
to up. Vanishing point term refers to the intersection of the track we extract the rail lines which are the longest lines on the left
lines defined as in Liang [18]. It is firstly performed color and right. Then the longest line on the left and right can be
conversion from RGB to HSV color space in this algorith m. founded by calculating line lengths by Equation 3.
HSV color space consists of 3 channels. These channels are Hue
(H), Saturation (S), and Value (V). Hue channel defines the 𝑦 = 𝑚𝑥 + 𝑐 (3)
dominant wavelength of the color such as green, blue, yellow, After rail track lines are detected by using Hough transform
and etc. And this channel has a value between 0-360 angular. [19, 20], the intersection of two straight rail lines in the image
Saturation channel identifies the vividness of the color. High can be computed. To compute vanishing point, common form
saturation reasons to vivid colors and low saturation ensures of a linear equation can be used. This equation is described in
that the color approaches to shades of gray. While value channel Equation 3. This equation is slope-intercept form, where m is
defines white rate of within color. So it identifies the bright of the slope and c is the y-intercept of the line. The slope equation
color. In this study, ROI segmentation steps are performed on of a line which passes through two different points 𝑃1 (𝑥1 , 𝑦1 )
V channel image. and 𝑃2 (𝑥 2 , 𝑦2 ). This equation is shown in Equation 4.
In gradient computing step, the gradient of input image is 𝑚 = (𝑦2 − 𝑦1 )/(𝑥 2 − 𝑥 1 ) (4)
computed by using Sobel operator and thereby gradient
magnitude image Gradient(i,j) is obtained. Then a given The y intercept of a line which passes through two different
threshold value is applied to the bottom 1/3 of the gradient points 𝑃1 (𝑥 1 , 𝑦1 ) and 𝑃2 (𝑥 2 , 𝑦2 ). This equation is shown in
image due to the fact that this part of the image includes Equation 5.
sufficient information to extract the features. Furthermore rail 𝑐 = (𝑥 2 𝑦1 − 𝑥1 𝑦2 )/(𝑥 2 − 𝑥1 ) (5)
tracks are straight and their gradient magnitudes are high in the
bottom of the image. This threshold process eliminates the The y coordinate of vanishing point p is computed by
points that have lower gradient magnitude from the threshold Equation 6.
____________________________________________________________________________________________________________
219
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
220
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
1 0 -1 1 1 1
1 0 -1 0 0 0
Figure 12. Sample representation of connection points in rail track
1 0 -1 -1 -1 -1
____________________________________________________________________________________________________________
221
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(a) (b)
Figure 16. (a) Example railway input image (b) output image for
ROI segmentation method
____________________________________________________________________________________________________________
222
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
crossing, obstacle detection process can be done in the image. IEEE Transactions on Vehicular Technology, Vol. 53, pp. 1636-1648,
2004.
Avarage elapsed time of the proposed level crossing detection [19] L. Chandrasekar, and G. Durga, , “Implementation of Hough Transform
method has been 0.45 seconds and accuracy rate of the for Image Processing Applications”, In International Conference on
proposed level crossing detection method has been 87 %. Comm unication and Signal Processing, pp. 843-847, India, 2014
[20] C. T astimur, M. Karaköse, Y. Celik, and E. Akin, “ Image Processing
Based T raffic Sign Detection and Recognition with Fuzzy Integral”, In
A CKNOWLEDGMENT 2016 International Conference on Systems, Signals and Image Processing
This work has been supported by TUBİTAK (The Scientific (IWSSIP), pp. 1-4, doı: 10.1109/IWSSIP.2016.7502715, 2016.
[21] Rafael C. Gonzalez, and Richard E. Woods, “Digital Image Processing
and Technological Research Council of Turkey). Project (3rd Edition)”, Prentice Hall, 2007.
Number: 114E202 [22] Y. Fang, Y. Yuan, L. Xu, and W. Lin, “A benchmark for robustness
analysis of visual tracking algorithms”, In 2016 IEEE International
REFERENCES Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.
1120-1124, doı : 10.1109/ICASSP.2016.7471850, 2016.
[1] J. C. Espino, B. Stanciulescu and P. Forin, “Rail and turnout detection [23] Y. Wu, J. Lim, and M.-H. Yang, “Online object tracking: a benchmark”,
using gradient information and template matching”, In Intelligent Rail In IEEE International Conference on Computer Vision and Pattern
Transportation (ICIRT), 2013 IEEE International Conference on, Recognition, pp. 2411 – 2418, doı: 10.1109/CVPR.2013.312, 2013.
Beijing, pp. 233-238, 2013. [24] A. W. M. Smeulders, D. M. Chu, R. Cucchiara, S. Calderara, A. Dehghan,
[2] Y. Santur, M. Karaköse, I. Aydın, and E. Akın, ”IMU based adaptive blur and M. Shah, “Visual tracking: an experimental survey”, In IEEE
removal approach using image processing for railway inspection”, In Transactions on Pattern Analysis and Machine Integillence, Vol. 36, No.
2016 International Conference on Systems, Signals and Image Processing 7, pp. 1442-1468, 2014.
(IWSSIP) IEEE, pp. 1-4, 2016. [25] T . Siriteerakul and R. Gullayanon, “Robust T racking Algorithm with
[3] C. T astimur, E. Akın, M. Karaköse, and, I. Aydin “Detection of rail faults Designed Marker for Limited-Power Computer”, In 2016 8th
using morphological feature extraction based image processing”, In 2015 International Conference on Knowledge and Smart Technology (KST),
23nd Signal Processing and Communications Applications Conference pp. 245-248, doı: 10.1109/KST .2016.7440529, 2016.
(SIU), pp. 1244-1247,doı: 10.1109/SIU.2015.7130063, 2015. [26] S. Dotenco, F. Gallwitz,, and E. Angelopoulou, “Autonomous Approach
[4] L. Shen, X. Wei and L. Jia, “ Surface Defects Detection of Railway and Landing for a Low-Cost Quadrotor Using Monocular Cameras”. In
T urnouts”, In Control Conference (CCC), 2015 34th Chinese, China, pp. Computer Vision-ECCV 2014 Workshops, pp. 209-222, 2014.
6285 – 6290, 2015. [27] S. Lange, N. Snderhauf, and P. Protzel, “A vision based onboard approach
[5] Y. Chen and H. Zhao, “Fault detection and diagnosis for railway for landing and position control of an autonomous multirotor UAV in
switching points using fuzzy neural network”, In Industrial Electronics GPS-denied environments”, In Advanced Robotics, 2009. ICAR 2009.
and Applications (ICIEA), 2015 IEEE 10th Conference on, pp. 860-865, International Conference on, pp. 1-6, 2009.
2015. [28] T . Merz, S. Duranti, and G. Conte, “Autonomous landing of an unmanned
[6] F. Kaleli and Y.S. Akgul, “Vision-Based Railroad Track Extraction Using helicopter based on vision and inertial sensing”, In Experimental Robotics
Dynamic Programming”, In 12th International IEEE Conference on IX Springer Berlin Heidelberg, pp. 343-352, 2006.
Intelligent Transportation Systems, pp. 1-6, 2009. [29] H. B. Kekre and K. Sonawane, “Comparative study of color histogram
[7] Q. Li, J.Shi and C. Li, “Fast line detection method for Railroad Switch based bins approach in RGB, XYZ, Kekre's LXY and L′X′Y′ color
Machine Monitoring System”, In 2009 International Conference on spaces”, In Circuits, Systems, Communication and Information
Image Analysis and Signal Processing, pp. 61-64, 2009. Technology Applications (CSCITA), 2014 International Conference on,
[8] Z. Qi, Y. T ian and Y. Shi, “Efficient railway tracks detection and turnouts pp. 364-369, doı: 10.1109/CSCIT A.2014.6839288, 2014.
recognition method using HOG features”, In Neural Computing and [30] J. Zhang, L. Zhang, and T . Xu, “Image Segmentation Using a Hybrid
Applications 23.1, pp. 245-254, 2013. Gradient Based Watershed T ransform”, In 2013 International Conference
[9] J. Wohlfeil, “ Vision based rail track and switch recognition for self- on Mechatronic Sciences, Electric Engineering and Computer (MEC), pp.
localization of trains in a rail network”, In Intelligent Vehicles Symposium 1408-1412, doı: 10.1109/MEC.2013.6885288, Shengyang, 2013.
(IV), 2011 IEEE. IEEE, p. 1025-1030, 2011. [31] P. Kaur and A. Gupta, “Contour Detection of Gradient Images Using
[10] F. Kaleli and Y.S. Akgul,”Vision-based railroad track extraction using Morphological Operator and T ransform Domain Filtering”, In
dynamic programming”, In 2009 12th International IEEE Conference on Computational Intelligence & Communication Technology (CICT), 2015
Intelligent Transportation Systems, pp. 1-6, 2009. IEEE International Conference on, pp. 107-111, doı:
[11] J. C. Espino and B. Stanciulescu, “ Turnout detection and classification 10.1109/CICT .2015.106, Ghaziabad, 2015.
using a modified HOG and template matching”, In 16th International [32] G. M. H. Amer and A. M. Abushaala, “Edge detection methods”, In Web
IEEE Conference on Intelligent Transportation Systems (ITSC 2013) pp. Applications and Networking (WSWAN), 2015 2nd World Symposium on,
2045-2050, 2013. pp. 1-7, doı: 10.1109/WSWAN.2015.7210349, 2015.
[12] R. Ross, “ Vision-based track estimation and turnout detection using [33] G. N. Chaple, R. D. Daruwala, and M. S. Gofane, “ Comparisions of
recursive estimation”, In Intelligent Transportation Systems (ITSC), 2010 Robert, Prewitt, Sobel operator based edge detection methods for real
13th International IEEE Conference on, pp. 1330-1335, 2010. time uses on FPGA”, In Technologies for Sustainable Development
[13] Mainline, (2014, June), “Deliverable 3.3: Rail Switches and Crossings. (ICTSD), 2015 International Conference on , pp. 1-4, doı:
Development of new technologies for replacement”. [Online]. Available: 10.1109/ICT SD.2015.7095920, 2015.
http://www.mainline-project.eu/IMG/pdf/ml-d3.3-f- [34] V. Katkar, S. Kulkarni, and D. Bhatia, “Traffic Video Classification using
methods_for_switches__-crossings_replacement.pdf edge detection techniques”, In 2015 International Conference on Nascent
[14] O. Yaman, M. Karakose, and E. Akin, “ PSO Based Diagnosis Approach Technologies in the Engineering Field (ICNTE-2015), pp. 1-6, doı:
for Surface and Components Faults in Railways”, In International 10.1109/ICNT E.2015.7029907, 2015.
Journal of Computer Science and Software Engineering (IJCSSE), [35] N. Gajjar, V. Patel, and A. J. Shukla, “ Implementation of edge detection
Volume 5, Issue 5, pp.89-96, 2016. algorithms in real time on FPGA”, In 2015 5th Nirma University
[15] Railway T echnical Web Pages, (2016, January), “Infrastructure”. International Conference on Engineering (NUiCONE), pp. 1-4, doı:
[Online]. Available: http://www.railway-technical.com/track.shtml 10.1109/NUICONE.2015.7449635, 2015.
[16] Arrive Alive, (2016), “Road Safety and Rail Crossings/ Level Crossings”. [36] İ. Aydin, E. Karaköse, M. Karaköse, M. T. Gençoğlu, and E Akın, “A
[Online]. Available: https://arrivealive.co.za/Road-Safety-and-Rail- New Computer Vision Approach for Active Pantograph Control”, In
Crossings-Level-Crossings Innovations in Intelligent Systems and Applications (INISTA), 2013 IEEE
[17] Level Crossing Installations, (2016), “ Level Crossing Installations”. International Symposium on, pp. 1-5, doı:
[Online]. Available: http://www.levelcrossinginstallations.co.uk/level- 10.1109/INIST A.2013.6577665, 2013.
crossing-installations/ [37] M. Karakose and M. Baygin, “Image processing based analysis of moving
[18] Y.M. Liang, H.R. T yan, S.L. Chang, H.Y.M. Liao, and S.W. Chen, shadow effects for reconfiguration in PV arrays”, In Energy Conference
“ Video stabilization for a camcorder mounted on a moving vehicle”, In (ENERGYCON), 2014 IEEE International, pp. 683-687, 2014.
____________________________________________________________________________________________________________
223
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Positioning systems have wide range of applications computational cost and also cause deterioration in the
with the developing technology. Global Positioning Systems accuracy of position. For this purpose, the redundant APs are
(GPS) is an efficient solution for outdoor applications but it gives eliminated using feature selection methods. During the online
poor accuracy in indoor environment. And, various methods are phase, the radio map is used to estimate the location of a
proposed in the literature such as geometric-based, WiFi
mobile device using a new RSS measurement using machine
fingerprint-based, etc. In this study, a hybrid approach that uses
both clustering and classification is developed for Wi-Fi learning algorithms such as Decision Tree (DT), K-nearest
fingerprint-based. Information gain based feature selection neighbour (KNN), or Expectation Maximization (EM) etc. [4]
method is used for selection of the most appropriate features In this study, a hybrid approach that combines clustering
from the WiFi fingerprint dataset in the initial step of this and classification algorithms is applied. InfoGain based
approach. Then, Expectation Maximization (EM) algorithm is feature selection is firstly used to removed redundant APs in
applied for clustering purpose. Then, decision tree algorithm is the radio map. Then, EM clustering algorithm is utilized to
used as a classification task for each cluster. Experimental results divide radio map into distinct groups or clusters. Finally, after
indicate that applied algorithms lead to a substantial assigning each test data to correct cluster, DT classier
improvement on localization accuracy.
algorithm is used as a classification task. In the experiments,
DT and applied hybrid approach is compared in terms of
Keywords— WiFi fingerprint-based, indoor positioning, access positioning accuracy. In the experiments, WEKA open source
point selection, clustering, classification, feature selection,
machine learning toolbox and RFKON database [5] are used.
expectation maximization, decision tree, received signal strength,
WLAN. The paper is organized as the follows. Related works in
literature for fingerprinting based indoor positioning is given
I. INTRODUCTION in section 2. Section 3 focuses on applied hybrid approach
Positioning systems are becoming very important with the introducing the algorithms using in this approach.
developing technology. These systems are used to determine Experimental area is given in section 4. Test results for
the position of the mobile device and it is used by location- proposed algorithm are given in section 5. As a result of this
based services for navigating, or tracking etc. They are paper, conclusion part summarizes all written things in the
classified as outdoor and indoor positioning systems. Global paper with a short paragraph including future works in section
Positioning System (GPS) is used for outdoor in our daily life. 6.
But, it cannot be used for indoor environment due to lack of
GPS signals including multipath and signal blockage [1]. II. RELATED WORKS
Therefore, different kinds of indoor positioning systems (IPSs) A large number of studies that adopt fingerprinting as the
have been developed. WiFi-based indoor positioning systems position estimation method are proposed in the literature.
have become an attractive solution in indoor area since WiFi Fingerprint-based positioning algorithms can be categorized
access points (APs) can be found widely in indoor buildings into two groups: deterministic algorithms [2, 6-8] and
such as airports, shopping malls, or office buildings, etc. probabilistic algorithms [9-13].
WiFi-based indoor positioning is an inexpensive solution Deterministic algorithms are used to find the minimum
because they have not need any additional installation cost [2]. signal distance between the newly measured RSS vector and
Fingerprinting technique is the most accurate technique that pre-measured fingerprints which are vectors of RSSs from
is based on Received Signal Strength (RSS) in WiFi-based detectable APs in the region (radio map). Each fingerprint in
indoor positioning systems [3]. In this technique, the RSS the radio map is associated with a known RP. In [2], KNN is
obtained from WiFi APs is used to determine the position of a used to estimate the position of a mobile WiFi-enabled device.
mobile device. It contains two phases named as offline and A feature scaling based KNN (FS-kNN) algorithm is proposed
online phases. In offline phase, RSS values are collected at in [6] to improve localization accuracy. In FS-kNN, different
known reference points (RPs) in the experimental area and weights are assigned to signal differences at different RSS
then are used to construct radio map. The dimension of the levels when estimating the similarity between two RSS vector.
radio map is grown while including all the detectable APs in Artificial neural network (ANN) classifier is utilized to
the region. But all these APs are not contributed positively to classify link quality patterns for each location in [7]. In [8],
the accuracy of position. Therefore, the redundant APs will be DT algorithm is used as the classification algorithm.
removed from the radio map, since they increase the
____________________________________________________________________________________________________________
224
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
In probabilistic algorithms, the probability of the mobile of expectation iteratively. In this study, EM algorithm in
device position to be each reference point is calculated and WEKA is used for clustering purpose.
then maximum probability is returned as a mobile device
position. In [9], particle filter, a Bayesian based method, is C. Decision Tree Classifier Algorithm
employed. Observed signal strengths are obtained using Decision Tree predicts an output by tracking the decisions
Bayesian inference in [10] and the estimated position is in the tree from the root node down to a leaf node according to
determined as the highest probability in the resulting the outcome of the tests along the path [19]. In this study,
distribution. An extended Kalman filter based approach is C4.5 that is a benchmark tree (J48 in WEKA) is applied in the
presented in [11], where the intra cell position of a cellular classification step.
device is estimated using RSS readings from base stations.
This estimate, movement pattern data and velocity vectors are IV. EXPERIMENT
combined in order to predict the next cell crossing. In [12], a Data were collected to construct our database for Eskisehir
Bayesian filter based approach is proposed. In this study, a Osmangazi University Teknopark. It has two floors of area of
posterior probability distribution over the target’s location is 800 . This area was broken into grid squares (each of size
obtained by inverting Bayesian belief network. In [13], subset 2.4 × 2.4 ). We collect the data from the first floor of
of the strongest APs is considered instead of all APs and Teknopark and the center of each grid square was noted as
target location is predicted using Bayesian estimate. seen in Fig. 1.
There are also various algorithms are proposed in the
literature to reduce the computational cost by eliminating
redundant APs in the radio map. In [14], according to strength
of the signal a weight is assigned to each AP and then the APs
with minimum weights are dropped from each fingerprint.
Various AP significance measures such as average RSS,
entropy, variance, maximum RSS are examined in [15]. Fast
Orthogonal Search (FOS) and modified FOS (mFOS)
algorithms are implemented in [16] in order to reduce the
dimensionality of the radio map in an IPS.
____________________________________________________________________________________________________________
225
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III
COMPARISON OF ACCURACY RESULTS
____________________________________________________________________________________________________________
226
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
localization accuracy. In addition to this, by the help of the [17] A. G.Karegowda, A. S. Manjunath, and M. A. Jayaram, “ Comparative
study of attribute selection using gain ratio and correlation based
clustering phase of the applied approach and constructing
feature selection”, International Journal of Information Technology and
cluster specific DT models reduce the size of the tree Knowledge Management, vol. 2,pp. 271-277, Dec. 2010.
significantly. [18] I.H. Witten, E Frank, MA Hall. Data Mining: Practical Machine
In a future work, different clustering and classification Learning Tools and Techniques: Practical Machine Learning Tools and
Techniques. Elsevier, 2011.
algorithms will be evaluated in a hybrid approach to get better
[19] L. Rokach, O. Maimon, Decision Trees. In The Data Mining and
results in terms of accuracy and computational time. Knowledge Discovery Handbook. Springer, pp. 165–192, 2005.
ACKNOWLEDGMENT
This work is supported by The Scientific and Technological
Research Council of Turkey (TUBITAK) under grant number
1130024.
REFERENCES
[1] G. M. Djuknic, and R. E. Richton, “Geolocation and Assisted GPS,”
IEEE Computer, vol. 2, pp. 123–125, Feb. 2001.
[2] P. Bahl and V. N. Padmanabhan, “RADAR: An InBuilding RF-based
User Location and Tracking System,” in Proc. IEEE INFOCOM, 2000,
pp. 775–784.
[3] A. Abusara and M. Hassan, “Enhanced fingerprinting in wlan-based
indoor positioning using hybrid search techniques,” in International
Conference on Communications, Signal Processing, and their
Applications (ICCSPA), 2015, pp. 1–6, Feb. 2015.
[4] H. Liu, H. Darabi, P. Banerjee, and J. Liu, “Survey of wireless indoor
positioning techniques and systems,” Systems, Man, and Cybernetics,
Part C: IEEE Transactions on Applications and Reviews, vol. 37, pp.
1067–1080, Nov. 2007.
[5] S. Bozkurt Keser, U. Yayan, A. Yazici, S. Gunal, "A priori verification
and validation study of RFKON database", International Journal of
Computer Science: Theory and Application, vol. 5, 20-27, 2016.
[6] D. Li, B. Zhang, Z. Yao and C. Li, "A feature scaling based k-nearest
neighbor algorithm for indoor positioning system," 2014 IEEE Global
Communications Conference, Austin, TX, 2014, pp. 436-441.
[7] Y. Ha, E. Ae-cheoun, and B. Yung-cheol, "Efficient sensor localization
for indoor environments using classification of link quality patterns",
International Journal of Distributed Sensor Networks, 2013.
[8] S. Eisa, J. Peixoto, F. Meneses, and A. Moreira, "Removing useless
APs and fingerprints from WiFi indoor positioning radio maps",
International Conference on Indoor Positioning and Indoor Navigation
(IPIN), pp.1-7, Oct. 2013.
[9] V. Seshadri, V. G. Zaruba, and M. Huber, "A Bayesian sampling
approach to in-door localization of wireless devices using received
signal strength indication", Third IEEE International Conference on
Pervasive Computing and Communications (PERCOM 2005), pp.75-
84, March 2005.
[10] X. Chai, and Q. Yang, "Reducing the calibration effort for location
estimation using unlabeled samples", Third IEEE International
Conference on Pervasive Computing and Communications (PERCOM
2005), pp. 95-104, March 2005.
[11] T. Liu, P. Bahl, and I. Chlamtac, “A hierarchical position-prediction
algorithm for efficient management of resources in cellular networks”,
Global Telecommunication Conference (GLOBECOM ’97), IEEE, vol.
2, pp. 982-986, Nov. 1997.
[12] M. Isard, and A. Blake, “Contour tracking by stochastic propagation of
conditional density”, Computer Vision (ECCV ’96), pp. 343-356, 1996.
[13] M. A. Youssef, A. Agrawala, A. U. Shankar, and S. H. Noh, “A
probabilistic clustering-based indoor location determination system”,
Tech. Report, University of Maryland at College Park, CS-TR 4350,
March 2002.
[14] J. Ledlie, “Method and apparatus for on-device positioning using
compressed fingerprint archives,” June 2011.
[15] E. Laitinen, E. Lohan, J. Talvitie, and S. Shrestha, “Access point
significance measures in WLAN-based location,” in 2012 9th
Workshop on Positioning Navigation and Communication (WPNC), pp.
24–29, Mar. 2012.
[16] A. Abusara, M. S. Hassan, and M. H. Ismail, "RSS fingerprints
dimensionality reduction in WLAN-based indoor positioning", IEEE
2016 Wireless Telecommunications Symposium (WTS), April 2016.
____________________________________________________________________________________________________________
227
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In parallel with the increase seen in the number of a system of 16 floors and 4 cabins was preferred. Average
high-rise buildings, vertical transport systems are progressing. waiting period, long waiting percent and process time was
One of the results of this progress is the emergence of group taken into account in the study. Proposed approach overcame
elevator systems and their primary aim is to transport its the deficiencies of particle batch optimization algorithm and
passengers to the target floor the fastest way possible. Studies on
offered a new solution for timing group elevator systems [9].
this field are generally simulation and optimization based and
they have an aim of minimizing the passengers’ waiting and In another study on group elevator systems, three different
traveling periods. In this study, a real time group elevator optimization techniques were used for process of timing.
experimental setup was created and an optimization algorithm Genetic algorithm, artificial immune algorithm and DNA
was applied on the setup. Genetic algorithm was chosen as algorithm were used in this study and an estimation algorithm
optimization algorithm and this method was tested in an elevator was proposed. In simulation processes, a 20 floor and 5 cabin
prototype of 10 floors and 5 cabins. The results obtained revealed system was considered and approximately 500 directional
efficiency, performance and accuracy of proposed method. calls were created. In the system in which all methods are
tested in line with these calls, a decrease in people’s average
Keywords— group elevator systems, optimization, genetic waiting periods was observed [10]. In another study for
algorithm, average waiting time, real time control average waiting time, a simulation environment consisting of
16 floors and 4 floors was created. In this study, a numeric
keypad was used for people to enter their destination floors.
I. INTRODUCTION Therefore target floors could be known beforehand. In
Elevator systems are of utmost importance in our daily addition to this, an artificial immune optimization algorithm
lives. These systems are used in nearly all fields and they was used in the study in order to for the cabins to detect the
provide ease especially in vertical transport [1, 2]. In addition optimal way [11]. In another study on the same field, a fuzzy
to this, when we take rising number of high-rise buildings of group elevator control algorithm was proposed. In this study,
today’s world into account, importance of elevator systems there are five different control algorithms based on FPGA. An
increases [3]. example of block diagram summarizing this structure which
In parallel with increasing number of floors in buildings, consists of four different modules is as seen in Fig. 1 [12].
the population in the buildings also increases and single cabin
elevator systems fail to meet the needs of this population [4-6].
Traffic Module Fuzzy Module
Group elevator systems, which have emerged for satisfying
this needs, are seen in many buildings and are in active use Traffic Current Suitable
today. In general sense, group elevator systems are active Characteristics Traffic Elevator
usage of two or more elevator systems in the same building.
Main aim of these systems is to ensure that people’s average Measure Module Mapping Module
waiting and traveling period is reduced [5-7].
In one of the studies in line with these objectives, group
elevator systems are optimized using genetic algorithms. In Suitable Algorithm
the study, a system of 20 floors and 4 cabins were created and Calls Status of Elevator Selection
Each
reducing people’s average waiting and traveling period is Elevator
aimed. Directional calls were used and these calls were
previously uploaded into the system. Obtained simulation Fig. 1 The proposed approach in literature [12]
results show efficiency of the proposed approach [8]. In In another study which proposes a different point of view to
another study on the same subject, hybrid optimization analyse group elevators, the aim was to create double cabin
technique was proposed. In the study in which particle batch elevator systems. In this approach, which was applied on
optimization and simulated annealing algorithms are used and group elevator systems, genetic network programming
____________________________________________________________________________________________________________
228
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
technique was used and it was revealed that the proposed II. GENETIC ALGORITHM FOR ELEVATOR CONTROL SYSTEM
approach diminished waiting and traveling periods in different Control and optimization of a group elevator system is an
building traffics. The block diagram to summarize the system important issue. Using optimization techniques in these
used in this study is given in Fig. 2 [13]. Another issue to take systems for usage of elevator cabins ideally and for the
into account in group elevator systems is energy efficiency. In elevator to provide its services to the people is a must. In the
another study, in addition to minimizing cabins’ average study performed with this objective, genetic algorithm was
waiting and traveling times, an adaptive control system with used and minimizing people’s waiting period was aimed.
the aim of diminishing energy use was proposed. In this Genetic algorithm is an optimization technique which is
approach, genetic algorithm, clonal selection algorithm, DNA commonly used in literature and applied in many fields [15,
algorithm and estimating algorithm was used. Besides, in this 16]. Pseudo code of this approach which aims best ideal
system, a fuzzy structure was created to make effective use of solution in big solution space is shown step by step in Fig. 4.
these algorithms. The proposed system calculates average
waiting and traveling periods and energy amounts to be used
by the cabins to detect the optimal solution. An example block Genetic Algorithm Pseudo Code
diagram of this paper is presented in Fig. 3 [14]. Initial population of randomly generated sequences of
Step 1:
In this study, a genetic algorithm was used for time binary numbers
efficiency optimization of group elevator systems. Within the
scope of the study, an experimental setup of 10 floors and 5 Step 2: A certain amount of element is selected for the solution
cabins was established and optimization algorithm was tested
on this system. On the established experimental setup, people Step 3: Crossing is applied to new population
could enter their destination floors thanks to numeric keypad.
Directional floor call keys were also used. Also, optimization
Step 4: Mutation process is applied for the same population
algorithm was applied on both call types and system
behaviour was studied. In line with these objectives, details of
the optimization algorithm was given in part two. In part three, Step 5: Affinity values of elements of these populations are found
work patterns of the elevator control system was referred to
2. step is repeated until it reaches the maximum number of
and in part four, experimental results were included. In the Step 6:
transactions
fifth and the last parts of the study, results were presented.
Top
Floors One of the most important difficulties encountered in
organization methods is specifying objective function.
Objective function, which is of utmost importance in problem
solving, is seen as in Equation 1 for group elevator system.
(1)
Base
Floors
Determine traffic times Genetic algorithms are one of the most frequently used
• Up-peak, Down-peak, Inter-floor optimization techniques. One of the principal reasons for this
is that genetic algorithms are able to offer really fast and
Optimization Module effective solutions in big solution spaces. Therefore, this
• AIS, GA, DNA, EA method was chosen for the study and used for control and
optimization of group elevators.
Fuzzy Module
Optimal Path
____________________________________________________________________________________________________________
229
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
III. PROPOSED ELEVATOR CONTROL APPROACH system with the outer world is established with an RS-232
In order to put the proposed approach into practice, an port via serial communication systems. Data is both
elevator prototype was developed which worked in line with transferred and read through this network. Various data polls
the real elevators. This system has 10 floors and 5 cabins. Dial were created on main computer line for measuring
calls are designed both as directional and as in a way which performance of the system and these calls are transferred to
destination floor can be entered. For this purpose, a numeric the processor controlling the elevator. Therefore performance
keypad is integrated onto the system. An example block tests could be made and statistical information could be
diagram on the designed system can be seen in Fig. 5. obtained.
On the system, the tests are made through three different
ways. These are elevator control approach using classical
directional floor call, elevator control approach based on
10cm
1
2,5cm
15cm
calls from call pool, floor and cabin numbers, cabin position
17cm
1 2 3
4 5 6
7 8 9
* 0 #
9cm 13cm
4
7
*
5
8
0
6
9
#
13cm 13cm 9cm as seen in Fig. 6. In addition to this, a block diagram to
summarize the way the system works is shown in Fig. 7.
Randomly
Directions
Positions
Numbers
1 2 3
4 5 6
7 8
Floor,
9
Cabin
Cabin
Cabin
0 #
Calls
*
1 2 3
4 5 6
7 8 9
* 0 #
1 2 3
Genetic Algorithm
180cm
4 5 6
7 8 9
* 0 #
Optimal Path
1 2 3
4 5 6
7 8 9
* 0 #
Microcontroller
1 2 3
Average Journey Average Waiting
Time Time
4 5 6
7 8 9
* 0 #
PC
1 2 3
4 5 6
7 8 9
* 0 #
1 2 3
4 5 6
1 2 3
7 8 9
Microcontroler
* 0 #
4 5 6
1 2 3
4 5 6
7 8 9
7 8 9
* 0 #
Elevator
Control
110cm * 0 #
____________________________________________________________________________________________________________
230
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I
BUILDING AND ELEVATOR FEATURES
Element Values
Number of floors 10
Number of elevators 5
Floor height (cm) 17
Speed (cm/s) 5
Number of calls 500
Number of up direction calls 289
Number of down direction calls 211
TABLE III
FLOOR AND DIRECTION OF CABINS
____________________________________________________________________________________________________________
231
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
20
REFERENCES
18
[1] J. R. Fernandez and P. Cortes, “A Survey of Elevator Group Control
%16.2 Systems for Vertical Transportation: A Look at Recent Literature,”
16 IEEE Control Systems, vol. 35, no. 4, pp. 38-55, Aug. 2015.
[2] J. Fernandez, P. Cortes, J. Munuzuri and J. Guadix, “Dynamic Fuzzy
14 %47.3
Logic Elevator Group Control System With Relative Waiting Time
%26.8 Consideration,” IEEE Transactions on Industrial Electronics, vol. 61,
12 no. 9, pp. 4912-4919, Nov. 2014.
[3] J. Jamaludin, N. A. Rahim and W. P. Hew, “An elevator group control
system with a self-tuning fuzzy logic group controller,” IEEE
10
0 10 20 30 40 50 60 70 80 90 100 Transactions on Industrial Electronics, vol. 57, no. 12, pp. 4188–4198,
Iteration Dec. 2010.
[4] A. Valdivielso and T. Miyamoto, “Multicar-Elevator group control
Fig. 10 Average waiting times for three different test condition algorithm for interference prevention and optimal call allocation”,
As it can be seen in Fig. 10, the ideal solution of three IEEE Transactions on Systems, Man and Cybernetics Part A: Systems
and Humans, vol. 41, no. 2, pp. 311-322, Mar. 2011.
different tests turned out to be the one in which numeric [5] P. E. Utgoff and M. E. Connell, “Real-time combinatorial optimization
keypad and optimization were used together. Besides, for elevator group dispatching,” IEEE Transactions on Systems, Man,
classical method offered provided the worst performance. In and Cybernetics A, vol. 42, no. 1, pp. 130–146, Jan. 2012.
the light of this data, Table 3 shows average waiting periods [6] J. Sun, Q. C. Zha, and P. B. Luh, “Optimization of group elevator
scheduling with advance information,” IEEE Transactions on
and system performance assessment. Automation Science and Engineering, vol. 7, no. 2, pp. 352–363, Apr.
2010.
[7] B. Bolat and P. Cortés, “Genetic and tabu search approaches for
TABLE IIIII optimizing the hall call—car allocation problem in elevator group
PERFORMANCE OF METHODS systems,” Applied Soft Computing, vol. 11, no. 2, pp. 1792–1800, Mar.
2011.
Classic Optimization Keypad
[8] E. O. Tartan, H. Erdem and A. Berkol, “Optimization of waiting and
AWT (s) 18.01 16.67 14.54 journey time in group elevator system using genetic algorithm,” IEEE
~CPU (%) 11 63 35 International Symposium on Innovations in Intelligent Systems and
Applications (INISTA) Proceedings, pp. 361-367, Jun. 2014,
Alberobello.
As it can also be seen on Table 3, the systems in which [9] L. Fei, Z. Xiaocui and X. Yuge, “A new hybrid elevator group control
numeric keypads were used turned out 23,8% more efficient, system scheduling strategy based on Particle Swarm Simulated
which means these systems provided a 14,6% more Annealing Optimization algorithm,” 8th World Congress on Intelligent
Control and Automation (WCICA), pp. 5121-5124, Jul. 2010, Jinan.
performance level when compared to the approach in which [10] M. Baygin and M. Karakose, “A New Intelligent Group Elevator
the optimization technique is used. When CPU usage data is Control Approach,” in Proceedings of the 15th International
analysed, the reason that classical method is the most Symposium (MECHATRONIKA), pp. 1–6, Dec. 2012, Prague.
advantageous one is that it steers the cabins according to the [11] X. Yuge, L. Fei and L. Xiaolan, “Hybrid destination registration
elevator group control system with artificial immune optimization
closest floor principle. When optimized system and the system algorithm,” 8th World Congress on Intelligent Control and Automation
with numeric keypad are compared by means of CPU (WCICA), pp. 5067-5071, Jul. 2010, Jinan.
performance, numeric keypad system is on step ahead. [12] D. M. Munoz, C. H. Lianos, M. A. Rincon and R. H. van Els,
“Distributed approach to group control of elevator systems using fuzzy
V. CONCLUSIONS logic and FPGA implementation of dispatching algorithms,”
Engineering Applications of Artificial Intelligence, vol. 21, no. 8, pp.
Group elevators were optimized with the study’s proposed 1309-1320, Dec. 2008.
approach. With this objective, a prototype of a group elevator [13] L. Yu, J. Zhou, S. Mabu, K. Hirasawa and J. Hu, “A Double-deck
Elevator Group Supervisory Control System with Destination Floor
system of 10 floors and 5 cabins was set up and tests Guidance System using Genetic Network Programming,” SICE-ICASE
progressed on this experimental setup in real time. A numeric International Joint Conference, pp. 5989-5994, Oct. 2006, Busan.
keypad system was integrated into the elevator, different from [14] M. Baygin and M. Karakose, “Immunity-Based Optimal Estimation
classical elevator systems, and the people were allowed to Approach for a New Real Time Group Elevator Dynamic Control
Application for Energy and Time Saving,” The Scientific World
enter their destination floor before they get on the elevator.
Journal, vol. 2013, pp. 1-12, Jun. 2013.
In this approach, genetic algorithm was chosen for the [15] M. Karakose, M. Baygin and N. Baygin, “An analysis approach for
method of organization. This algorithm is fast and efficient, optimization based reconfiguration in photovoltaic arrays,” IEEE 23rd
paving the way to its preferability. Three different approaches International Symposium on Industrial Electronics (ISIE), pp. 954-959,
Jun. 2014, Istanbul.
were taken in testing process. As a result of the studies, the [16] H. Yetis, M. Baygin and M. Karakose, “A New Micro Genetic
control system with numeric keypad worked better. Thanks to Algorithm Based Image Stitching Approach in Camera Arrays,” The
this new approach, a 20% more efficiency could be get and a 5th International Conference on Manufacturing Engineering and
14% gain could be managed when compared to the systems in Process (ICMEP), May. 2016, Istanbul.
which only optimization techniques were used.
____________________________________________________________________________________________________________
232
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— An electroencephalogram (EEG) is an electrical patterns using the peaks and valleys on the EEG for all age
activity which is recorded from the scalp over the sensorimotor groups. Seizure disorders (such as epilepsy), a head injury,
cortex during vigilance or sleeping conditions of subjects. It can encephalitis, a brain tumour, encephalopathy, memory
be used to detect potential problems associated with brain problems, sleep disorders, stroke, dementia, and etc. are the
disorders. The aim of this study is assessing the clinical
diagnosable brain disorders using an EEG. The normal
usefulness of EEG which is recorded from slow cortical
potentials (SCP) training in stroke patients using Deep belief electrical activity of the brain is disrupted in many
network (DBN) which has a greedy layer wise training using neurological disorders and brain potentials [2]. The kind of the
Restricted Boltzmann Machines based unsupervised weight and disorder can be stated by evaluating the shape and the interval
bias evaluation and neural network based supervised training. of the disruptions on the EEG. EEG carries characteristic
EEGs are recorded during eight SCP neurofeedback sessions features that are helpful in the early diagnosis and early
from two stroke patients with a sampling rate of 256 Hz. All treatment processes of psycho-physiological and neurological
EEGs are filtered with a low pass filter. Hilbert-Huang disorders [11].
Transform is applied to the trails and various numbers of
Instinct Mode Functions (IMFs) are obtained. High order
statistics and standard statistics are extracted from IMFs to
create the dataset. The proposed DBN-based brain activity
classification has discriminated positivity and negativity tasks in
stroke patients and has achieved high rates of 90.30%, 96.58%,
and 91.15%, for sensitivity, selectivity, and accuracy,
respectively.
I. INTRODUCTION
An electroencephalogram (EEG) is a biomedical signal that
records the electrical activity in the brain [1]. The neurons Fig. 1 EEG electrode placements on head
connect to each other by the dendrites and axons in the brain.
The communication between the neurons is provided by the Slow cortical potential (SCP) is gradual changes in the
electrical impulses over the dendrites and axons. The cortical layer and the regulation of cortical excitability in
electrical activity in the brain can be recorded and monitored cortical neuronal networks [12]. The SCP has a duration that
placing the electrodes to specific areas on the head [2]. The varies between 300ms and 10s [13]. The SCP is a low-
placement of the electrodes and the channel names are seen in frequency EEG component and a non-invasive method. The
Figure 1. The EEG is used in several areas such as providing electrodes are placed on the top center of the head and
the interaction between human and machine [3], [4], conditioned responses method used in giving feedback. Due to
determining the response of the brain to the visual and these characteristics, the SCP may include the contingent
auditory signals [5], [6], allowing the diagnosis of psycho- negative variation, readiness potential, movement-related
physiological and neurological disorders [7]–[11], and helping potentials, P300 and N400 potentials [1]. Negative SCP is
doctors to make a quick assessment of normal and abnormal assumed to depolarization of cortical neuronal cells and
____________________________________________________________________________________________________________
233
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
positive SCP shows neuronal complication [13]. The SCP has In this section, the SCP training database, the pre-
been correlated with a large number of cognitive processes in processing of the database, Hilbert-Huang Transform, and
a systematic and topographically ways and has been statistical feature extraction methods would be explained.
determinately utilized in psychophysiological experiments to
dissociate cognitive functions and motor performance of the A. Database
brain [14], [15]. Ergenoglu et al. worked on determining the The EEG is an electrical activity which is recorded from
relationship between SCP and P300 amplitude [12], Khader et the scalp over the sensorimotor cortex during vigilance or
al. analysed the relations between the SCP and Blood-oxygen- sleeping conditions of subjects. It can be used to detect
level dependent (BOLD) signal changes [16], Devrim et al. potential problems associated with brain disorders. EEG data
investigated the detection of visual stimuli at sensory were recorded during eight SCP neurofeedback sessions from
threshold using the SCP [6], Kotchoubey et al. used the SCP two chronic stroke patients [21]. Neurofeedback sessions were
training in the research on epilepsy with analysis of conducted with an approximate interval of one week between
influencing factors [7], [8]. Strehl et al. used functional sessions.
magnetic resonance imaging and the BOLD signal in the SCP The EEG was recorded with a sampling rate of 256 Hz
to reduce epileptic seizure frequency [17], Siniatchkin et al. from channel Cz using a Nexus-10 MKII DC amplifier
evaluated the analysis of migraine [9], Schneider et al. (Mindmedia, Herten, The Netherlands). Each neurofeedback
determined the efficiency of the SCP training in psychiatric session included trials in which cortical positivity had to be
patients with alcohol dependency[10], Cosch et al. associated increased and trials in which cortical negativity had to be
the SCP with the event-related potentials such as object, increased. The durational presentation of the trials is 8s and is
spatial, and verbal information [4], Hinterberger et al. seen in Figure 2. The feedback includes a circle with the
suggested a robust and steady communication method different size and colour related with the subject set
between computer and brain for amyotrophic lateral sclerosis successfully to baseline activity. The brain activities are
patients using the SCP and studied on developing a tough labelled successful as positivity, and success indicated to the
translation device [3], Pham et al. developed an auditory participant as negativity, if the trials are evaluated according
brain-computer stimuli for paralysed patients using the SCP to the task correctly, or not, respectively.
[5].
Stroke is a brain disorder that causes when the blood supply Baseline (0-2s) Active Phase (2s-8s)
of brain cells is either cut off or reduced and the cells began to
die. The abilities controlled by dead cells such as memory and
muscle control are lost. The disorder affects the EEG and can Fig. 2 Timeline of a trial from neurofeedback session
be declared from the SCP and more potential.
Deep Learning (DL) is an invasive, effective and machine All EEGs were filtered with a low pass filter (10 Hz). The
learning algorithm that has a growing popularity and attempts negativity trials in SCP training sessions more frequent than
to model high-level abstractions [18]. In recent years, the DL the positivity trials. A total number of 8000 trials (500 trials
is a new method of Machine Learning researches, which has for each session and each patient) with 2400 data points were
been recognized with image processing, character recognition, segmented from 2 of EEGs. Distribution of the trials
speech recognition, frequently. Convolutional Neural according to positivity and negativity situation in the database
Networks, Stacked Auto-encoders, Deep Boltzmann Machines is given in Table I.
and Deep Belief Networks (DBN) are the most effective DL TABLE I
algorithms [19]. The biggest advantage of the DL is STIMULI TRIAL DISTRIBUTION OF EACH STROKE PATIENT
representing handcrafted features with the efficient algorithms
for unsupervised or supervised feature learning and # #
Sessions Negativity Positivity
hierarchical feature extraction [20]. In this study, the DBN Trials Trials
algorithm is utilized as the classifier. The DBN is a robust and Session 1-3 250 250
simple type of the DL algorithms which is comprised of both Session 4-8 375 125
supervised and unsupervised learning stages. The Total 2625 1375
experimental results that are obtained with the same number
of hidden units and same structure of the DBN are compared
with artificial neural network (ANN). B. Hilbert-Huang Transform
In the following section, the SCP training database, Hilbert- Hilbert-Huang Transform is a method that has a frequently
Huang Transform (HHT), the HHT-based statistical feature use for feature extraction, filtering the signals, and similar
extraction processes, and the DBN classifier are described in processes on nonlinear and non-stationary signals. The
detail. The proposed SCP training classification system is growing trend on analysis of non-stationary and non-linear has
explained. The experimental results that are obtained using the led to opportunities of the analysis to expand the technical
DBN classifier are presented. requirements [22]. The HHT is one of the adaptive and
extensive methods that can be considered as relatively recent.
II. MATERIAL AND METHODS
____________________________________________________________________________________________________________
234
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The HHT has been applied to lots of fields such as biomedical 2) Hilbert Transform: The foremost characteristic feature
signal processing and geophysics [23]. of the non-linear signals is the internal wave frequency
Due to the flexibility of the stoppage criteria of the HHT modulation that shows the instantaneous frequency oscillation
algorithm, mathematical description of the HHT algorithm in a single period. The signal characteristics are indicated
could not be defined precisely and clearly [24]. The HHT has straightforwardly by the instantaneous frequency distribution.
a two-step analysis. The first step is Empirical Mode The HT is a decomposition which determines the amplitude-
Decomposition (EMD), and the second step is Hilbert frequency-time distribution of the signal [23], [26]. The HT of
Transform (HT), respectively. The EMD is pre-treatment of a signal is defined as:
the original data and extracts n number of Intrinsic Mode
Functions (IMF) and a residual signal. Each IMF is a signal ( )
which is based on a frequency modulation of the original data.
( )=ℜ ( )
The HT performs to obtain instantaneous frequency and
amplitude values of each IMFs in the time-frequency domain
[11], [24]. The HHT can perform more precise, distinctive and indicates the amplitude function of the signal,
clear results than other methods in the presentation of time- represents for the instantaneous frequency function. The
frequency-energy for non-stationary and nonlinear signals frequency-time distribution of the amplitude is called as
[25]. Hilbert Spectrum and is shown as ( , ).
1) Empirical Mode Decomposition: The EMD is a flexible ℎ( ) = ( , )
analysis method which is used for extracting characteristic
information obtained from non-linear and non-stationary The above equations show each IMF is amplitude and
processes. The most important characteristic of the EMD that frequency modulated signals. The EMD proves its impact by
is separated from other transformation algorithms is producing analysing the non-stationary signals in the variable amplitude
self-distinct oscillation of the original data by assuming that and frequency scales.
the signal consists of self-oscillations modes at the different
frequency bands [26]. Each oscillation is symmetrical relative C. Deep Belief Networks
to local mean of local extrema. Each different oscillation in The DBN is a generative DL algorithm that consists of both
the signal is indicated by an IMF. The IMFs are extracted unsupervised training and supervised training phases. The
from the signal by following two basic conditions[24], [25]: most important feature of the DBN is the idea of pre-training
The number of the local extremas and zero-crossing the weights, biases, and the other parameters using an
must be either same or difference must be equal to one unsupervised training algorithm such as Sparse Autoencoder
The mean of the upper and lower envelopes which are or Restricted Boltzmann Machines (RBM) [18], [27]. In this
obtained by the combination of the local maximum and study, the RBM is selected as pre-training algorithm in
local minimum must be equal at any t time unsupervised training phase. The RBM is a stochastic ANN
The stated IMF extraction conditions are used for type that calculates the weights of the units according to the
preventing negative frequency and keeping the instantaneous probability distribution over a set of inputs. The stacking of
frequency of narrow-band signals in the frequency band while the RBMs can build up the DBN with gradient descent
calculating the instantaneous frequency information [11]. algorithm or contrastive divergence algorithms [28]. In the
The local mean is calculated taking the average of the RBM, visible units that represent input data are connected to
lower envelope defined by local minimums and the upper hidden units that learn to represent features using undirected
envelope defined by local maximums. The local mean is weighted connections [19], [28].Considering the RBM with
subtracted from the original signal and the new form of the input layer activations (for visible units) and hidden layer
signal is controlled verifying if it is an IMF. If the new form activations ℎ (hidden units), bias of the visible unit , bias of
of the signal does not satisfy the IMF extraction conditions, the hidden unit :
the local mean is recalculated using the local maximums and
local minimums of new signal. This calculation is performed ( , ℎ) = −ℎ − − ℎ
until an IMF is extracted. The residual signal is obtained
subtracting the IMFs from last form of the residual signal. ( , )
When the residual signal is a monotonic function or the ( , ℎ) = ( , )
∑
residual signal has only one local extreme, it is not possible to
extract another IMF and the EMD process ends [11], [23], ( , ℎ) indicates the joint distribution of the RBM and
[24]. ( ) represents for the last residual signal, ( ) indicates ( , ℎ) represents for the energy function of the distribution.
the original signal, and is the number of the IMF. The DBN used latent variables in the deepest layer easy to
understand the deepest features using at least two hidden
( )= ( )+ ( ) layers in network. Each adjacent two layers have a connection
to evaluate the greedily layer-wise pre-training [18]. The
parameters such as the weights and the biases obtained in
____________________________________________________________________________________________________________
235
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
unsupervised learning phase are unfolded to a neural network and the importance of the SCP training and classification of
structure. The pre-trained parameters are updated and fine- the brain activities from the SCP training such as negativity
tuned in supervising training phase. The whole network can be trials or positivity trials in stroke patients.
optimized by gradient descent algorithm in this learning phase. The HHT is an effective method on non-linear and non-
The detailed formulation about the DBN is presented in [18], stationary EEG signals. Huang et al. [25] proposed a
[29]. communication between human and computer. They used the
HHT and wavelet transform for extracting the features from
III. EXPERIMENTAL RESULTS the steady-state visual evoked potential and indicated the HHT
The brain computer interfaces allow setting a connection is more accurately expressing the time and frequency
between the machines and the brain to control devices with characteristics ability than the wavelet transform. Li et al. [30]
the SCP and more. The SCP training neurofeedback sessions proposed a sleep stage classification method using the EEG.
are used to make a quick assessment of normal and abnormal They achieved a classification mean accuracy of 81.7% using
patterns in the diagnosis of psycho-physiological and the HHT, Fourier transform and wavelet transform features
neurological disorders using the peaks and valleys from the and discussed that the HHT is more successful and faster
EEG. The aim of this study is classifying the brain activities response for extracting the EEG features and tracking the
from the SCP training in stroke patients and pointing out the rapid changes. Ozdemir et al. [11] and Oweis et al [22]
negativity and positivity trails in the neurofeedback. applied the HHT to EEG and used intra wave frequency
The SCP training trials have been related with a large modulation on the different frequency bands to diagnose
number of cognitive functions, motor performance of the epileptic seizure prediction and classification and achieved
brain and many neurological disorders in literature. Ergenoglu accuracy rates of 89.66% and 94%, respectively. Considering
et al. worked with the positivity and negativity SCP trials to these achievements of the HHT on EEG, we decided to utilize
getting a connection between the SCP trainings and P300 the HHT in the feature extraction of the brain activity
amplitude and experienced the P300 amplitudes of trials with classification from the SCP training in stroke.
the negative SCPs are significantly higher in comparison with The proposed method consists of preparing EEG signal,
the positive SCPs at Cz, Pz, Fz, P3, and P4 channels under analysing EEG using the HHT and the HHT-based statistical
normal and abnormal conditions of the brain [12], Khader et feature extraction stage and the DBN-based brain activity
al. investigated the relations between the SCP and Blood- classification stage. To provide the different definitive and
oxygen-level dependent (BOLD) signal changes and proposed descriptive situations in patient and recording process of the
the similar topographical specificity of the SCP trials and the SCP training, the dataset with eight of the SCP neurofeedback
BOLD signal under cognitive experiments [16], Devrim et al. sessions that were conducted with an approximate interval of
investigated the detection of visual stimuli at sensory one week from two chronic stroke patients is selected. A low
threshold using the SCP and the negative SCP trials have a pass filter is applied to remove the potential noises and peaks
better separation ability than the positive SCP trials at Oz, Pz, in the EEG. 8000 of the SCP trials with 2400 data points were
Cz, and Fz channels [6]. Kotchoubey et al. analysed the extracted from the stroke patient EEGs. The extracted EEG
influencing factors in epilepsy using twenty sessions SCP trials are labelled as cortical negativity and positivity in the
training and detected the positive trials are more important dataset.
than the negative trials [7], [8]. Strehl et al. used functional The EMD is applied to extracted trials with 2400 of data
magnetic resonance imaging and the BOLD signal in the SCP points. IMFs were obtained vary in the number from 5 to 8.
and reduced epileptic seizure frequency [17], Siniatchkin et al. Figure 3 depicts a random trial and the IMFs decomposed
evaluated the analysis of individual differences in migraine from the random trial. HT is applied to each IMF. The
and experienced the negative trials from the SCP differed amplitude-frequency-time distribution of each IMF is
significantly between the individual habitat [9], Schneider et determined. Statistical features such as standard deviation,
al. determined the efficiency of the SCP training with self- correlation co-efficient, skewness, kurtosis, minimum,
regulation task utilizing biofeedback and instrumental maximum, covariance, mode and mean, energy based features,
conditioning in psychiatric patients with alcohol dependency high order statistical features such as moment and cumulants
[10], Hinterberger et al. developed a tough translation device were calculated for the each HT applied IMF and the feature
for ALS patients using the SCP trials [3], Pham et al. set is created.
developed an auditory brain-computer stimuli for paralysed The half of the trials (1750 of the positivity trials and 2250
patients using the SCP and discussed auditory stimuli of the negativity trials) from all sessions of the neurofeedback
characteristics may have to be adapted to optimize brain- is used for training of the DBN classifier model and the
computer interface performances [5]. The studies based on remaining trials are reserved for testing the proposed DBN
cognitive functions, motor performance of the brain and many system.
neurological disorders in literature are indicated the efficiency
____________________________________________________________________________________________________________
236
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Predicted Trials
Labels
Positivity Negativity
____________________________________________________________________________________________________________
237
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
238
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
vol. 10, p. 38, 2011. Sci., vol. 454, no. 1971, pp. 903–995, 1998.
[23] N. E. Huang and Z. Wu, “a Review on Hilbert-Huang Transform : [27] G. Hinton, G. Hinton, T. Sejnowski, and T. Sejnowski, Learning
Method and Its Applications,” October, vol. 46, no. 2007, pp. 1–23, and relearning in Boltzmann machines, vol. 1. 1986.
2008. [28] Y. Bengio and O. Delalleau, “Justifying and generalizing
[24] Y. Hou and H. Tian, “An automatic modulation recognition contrastive divergence,” Neural Comput., vol. 21, no. 6, pp. 1601–
algorithm based on HHT and SVD,” in Proceedings - 2010 3rd 1621, 2009.
International Congress on Image and Signal Processing, CISP [29] N. Allahverdi, G. Altan, and Y. Kutlu, “Diagnosis of Coronary
2010, 2010, vol. 8, pp. 3577–3581. Artery Disease Using Deep Belief Networks,” 2. Int. Conf. Eng.
[25] M. Huang, P. Wu, Y. Liu, L. Bi, and H. Chen, “Application and Nat. Sci., Nat. Sci., Sarajevo, Bosnia-Herzegovina, The Book of
contrast in brain-computer interface Between hilbert-huang Abstracts, p.9, 2016
transform and wavelet transform,” in Proceedings of the 9th [30] Y. Li, F. Yingle, L. Gu, and T. Qinye, “Sleep stage classification
International Conference for Young Computer Scientists, ICYCS based on EEG hilbert-huang transform,” in 2009 4th IEEE
2008, 2008, pp. 1706–1710. Conference on Industrial Electronics and Applications, ICIEA
[26] N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, H. H. Shih, Q. Zheng, 2009, 2009, pp. 3676–3681.
N.-C. Yen, C. C. Tung, and H. H. Liu, “The empirical mode
decomposition and the Hilbert spectrum for nonlinear and non-
stationary time series analysis,” Proc. R. Soc. A Math. Phys. Eng.
____________________________________________________________________________________________________________
239
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— An electrocardiogram (ECG) is a non-linear and non- Arrhythmia is a problem concerning the abnormal rhythm
stationary diagnostic biomedical signal that has a great and rate of heartbeats. The heart can beat too fast, too slowly,
importance for cardiac disorders. The computer-assisted analysis or inconsistently in different types of arrhythmias, which may
of biomedical signals has become an essential tool in recent years. feel like antagonism affection or fluttering. Arrhythmia may
This study introduces a deep learning application in automatic
be classified by rate of heartbeats, mechanism (automaticity,
arrhythmia classification. The proposed model consists of a
multi-stage classification system in raw ECG using a Deep belief re-entry, triggered) or duration of the heartbeats [4]. Several
network (DBN) which has a greedy layer wise training phase. types of arrhythmia are harmless, but some of them refer the
The multistage DBN model classified the MIT-BIH Arrhythmia cardiac disorders that may cause death. The ECG is a popular
Database heartbeats into 5 main groups defined by ANSI/AAMI diagnosis tool which is of the primary importance for
standards. All ECGs are filtered with median filters to remove cardiologists [5].
the baseline wander. ECG waveforms were segmented from There are many studies that are used for detecting
long-term ECGs using a window with a length of 501 data points arrhythmias, classifying them and diagnosing cardiac diseases
(R peak of the wave is located at the centre of the window). The that occur as a result of arrhythmias. These studies can be
proposed DBN-based multistage arrhythmia classification has
incorporated into two basic feature extractions: fiducial and
discriminated five types of heartbeats with a high accuracy rate
of 95.05%. non-fiducial methods. The fiducial methods contain the local
features such as temporal, morphological, amplitude, duration,
interval and segments between two selected waves which are
Keywords— Arrhythmia, Deep Belief Networks, Deep Learning, extracted from ECG waveforms. These methods are based on
AAMI, Raw ECG Waveform the time-domain features on the ECG [6]. The non-fiducial
methods are based on the frequency-domain features such as
wavelet transformations, and the other digital signal
processing techniques that extract new signal forms, sub-
I. INTRODUCTION bands and coefficients from ECG waveforms [7].
According to the World Health Organization surveys [1], Deep learning (DL) is an effective and high-performance
heart diseases are one of the most important reasons which machine learning algorithm which is gaining popularity.
cause death. Heart disease symptoms depend on what type of Frequently used analyses of the DL are used in image
heart disease you have. An electrocardiogram (ECG) is a non- processing, speech and natural language processing processes.
linear and non-stationary diagnostic signal that is important Actually, DL is a neural network structure which addresses
for cardiac disorders [2]. It is hard to assess a cardiac disorder the deeper feature levels using more hidden layers [8]. In this
using ECG because of long processes that need a control in study, Deep Belief Networks (DBN), which is an adaptable
detail and infrequent arrhythmias. In order to overcome these DL algorithm, is utilized to classify the heartbeats from
challenges, the computer-assisted analysis of biomedical different classes of arrhythmia using ECG waveform as input
signals has become an essential method in recent years. The of the structure.
computer assisted diagnosis and analysis systems achieve The remainder of the paper is structured in the following
rapid and advanced assessments in long, and hard to identify manner. The database and the arrhythmia types are defined by
processes. Arrhythmia and many cardiac disorders usually AAMI standards, pre-processing, and feature extraction from
need to use long-term ECG in inspection controls [3]. arrhythmia heartbeats are described in detail. The proposed
Therefore, computer-based methods and diagnosis systems multistage classification system is explained. The
provide major simplicity and reliability in the diagnosis and experimental results that are obtained using the DBN classifier
treatment of the diseases for cardiologists. are presented.
____________________________________________________________________________________________________________
240
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
II. MATERIALS AND METHODS baseline wander [11]. 6077 short-term ECGs were segmented
The general management of medical treatment and from long-term ECGs using a window with a length of 501
assessment systems has become effective and convenient data points (R peak of the wave is located at the center of
processes because of the recent technological developments in window). All data points are normalized to a [0, 1] range.
integrated circuit systems and computer-aided intelligent B. ECG Waveform
monitoring and diagnosis systems. In this section,
information about ECG waveforms and the DBN classifier are The ECG is a method that finds out the regularity or
described in detail. irregularity of heart beats and heart rates using the electrical
activity of the heart. The recorded electrical activities of the
A. Database heart represent for a waveform on the clinical assessments.
There are several arrhythmia databases in the literature. In These waveforms may have different forms according to the
this study, the MIT-BIH arrhythmia database (MADB) is lead of the ECG [11], [12]. The use of the ECG in medical
utilized [9]. This database has been used for evaluating assessment processes is very important in detecting the
arrhythmia detection and classifying the arrhythmia types. different waveforms and various cardiovascular heart diseases.
MADB contains 48 long-term ECGs from 25 men aged 32–89 In the entire body only the heart muscle has the ability to
years, and 22 women aged 23–89 years; each has 11-bit contract spontaneously. Polarity is the event of discharge of
resolution with 360 Hz sampling frequency. The heartbeats electrical charge of heart tissue. Depolarization is the positive
are labelled as five main arrhythmia types defined by the charging case of electrical activation in heart tissue [13].
Association for the Advancement of Medical Instruments The ECG has 0 mV to 5 mV amplitude and a frequency
(AAMI) standard. AAMI standardizations provide an band between 0.5 Hz and 100 Hz [2], [13]. P, Q, R, S, T and
objective, understanding, and dividual assessments and U waves appear over the baseline in the signal, respectively. If
monitoring processes of the arrhythmia types for clinical the amplitude of Q, R and S waves is less than 5 mV, the
treatments and an increased capability of testing and training wave is referred to using small capitals (q, r, and s). The
abilities for supervised learning phases [10]. AAMI classifies remaining portion between the waves is a segment; the
heartbeats into normal beats (N), supraventricular ectopic distance between the waves is an interval [12].
heartbeats (S), ventricular ectopic heartbeats (V), fusion
heartbeats (F), and unknown heartbeats (Q). The testing and
training dispersions of the heartbeats from the MADB are
seen in Table I.
TABLE I
DISPERSION OF MADB ACCORDING TO AAMI STANDARDS AND
QUANTITIES OF TEST AND TRAINING SETS
____________________________________________________________________________________________________________
241
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
exceed 0.11 seconds and has an amplitude value up to 2-3 mV classification of the arrhythmia types, a computer-aided
[2]. classification of the 5 arrhythmia types is implemented using a
The T-wave occurs as a result of ventricular re-polarization. DBN-based multistage classification. Figure 2 depicts the
The T-wave may have a pointed or flat view and positive, structure of the arrhythmia classification model.
negative or biphasic value on various leads. The duration of
the T wave that belongs to a normal subject is between 0.10
and 0.25 seconds. It takes place after about 300ms from the
QRS complex. The positions of these waves vary according to
the heart rhythm. The T wave is closer to the QRS complex
when the heart rhythm accelerates [14].
C. Deep Belief Networks
This study introduces a deep learning (DL) application for
automatic arrhythmia classification. The proposed model
consists of a multi-stage classification system of raw ECG
using DL algorithms. The DBN is one of the most effective
DL algorithms which has a greedy layer wise training phase
[15]. The DBN is composed of both Restricted Boltzmann
Machines (RBM) or an autoencoder based layer-by-layer
unsupervised pre-training procedure and neural network based
supervised training [8], [16]. Considering RBM with input
layer activations (for visible units) and hidden layer
activations ℎ (hidden units), bias of the visible unit , bias of Fig. 2 Structure of proposed Arrhythmia Classification Method
hidden unit :
Two median filters are applied to remove the noise and the
( , ℎ) = −ℎ baseline wanders to raw ECGs. Analysis of the long-term
− − ℎ (1)
ECGs is a demanding process for clinicians and also for
( , ) computer-aided systems. Considering this situation, ECG
( , ℎ) = ( , ) (2) waveforms were extracted from long-term ECGs using the
∑
moving window analysis technique. The R peak centred
( , ℎ) represents the joint distribution of the RBM and window with 501 data points was moved to extract ECG
( , ℎ) represents the energy function of the distribution. waveforms. 6077 of ECG waveforms were obtained from
RBM is used for calculating the conditional distribution of the long-term ECGs. ECG waveforms with 501 data points were
visible and hidden units. Each adjacent two layers create an directly used as features. Having a great number of the feature
RBM. The first visible unit is the input feature vector and the dimensionality causes long and deceiving training processes
other RBM parameters = ( , , ) are denoted by for the supervised machine learning algorithms. Feature
depending on the first visible unit [17]. dimensionality reduction for the provides for the extraction of
In the unsupervised training phase, the sub-network's more meaningful classification rules, the elimination of the
hidden layer serves as the visible layer for the next adjacent pointless feature vector for machine learning algorithms, the
layer applying contrastive divergence and the probabilistically improvement of generalization capabilities using fewer
reconstruction of the shared weights is implemented [8]. In the parameters and reduced complexity and run-time and for the
supervised training phase of the DBN, the calculated shared evaluation and prediction of accuracy for classifiers [18]. The
weights and the structure of the DBN are unfolded to a neural sequential forward feature selection algorithm is utilized in the
network structure for fine-tuning all the parameters of the proposed method to reduce feature dimensionality [19]. The
deep structure such as the weights and the biases [15]. The algorithm selects a subset of features which are not yet
DBN consists of at least two hidden layers (latent variables) in selected from 501 data points and the best predict the
the neural network. The number of the hidden layers is related arrhythmia types by sequentially selecting features until there
to the deep analysis of the input features in detail [15], [17]. is no improvement in the prediction. The highest accuracy is
achieved using 106 features from the ECG waveforms. The
III. EXPERIMENTAL RESULTS reduced feature vector is normalized to 0-1. The proposed
The morphological features are the ones most used in DBN-based multistage classifier was trained using 106 data
clinical trials for the diagnosis of the arrhythmia types. The points. Selected data points on the ECG waveform are seen in
robust and steady detection of arrhythmia is a common need Figure 3 with the red asterisk.
for all the cardiac diseases. Each arrhythmia type can be
related to different types of cardiac and pulmoner diseases.
That’s why detection and classification of the arrhythmia
types are so important in the early diagnosis and early
treatment processes. Considering the importance of the
____________________________________________________________________________________________________________
242
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
243
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Proposed ECG Waveform DBN 95.05% 3rd editio. Humana Press, 2007.
CWT: Continuous Wavelet Transform, LD: Linear Discriminant, ELM: Extreme Learning [13] Dr Patrick Davey, “ECG (electrocardiogram),” NetDoctor, pp. 1–4,
Machines, PNN: Probabilistic Neural Network, DWT: Discrete Wavelet Transform 2011.
[14] S. Kara, “Sensing of ECG signals and Imaging at the Computer in Real
It is hard to compare the studies in a stable way, because of Time,” Erciyes University, 1991.
[15] Y. Bengio, P. Lamblin, D. Popovici, and H. Larochelle, “Greedy Layer-
reasons such as the different number of subjects, different Wise Training of Deep Networks,” Adv. Neural Inf. Process. Syst., vol.
number of the arrhythmia types, different subjects, different 19, no. 1, p. 153, 2007.
databases and different classification types. High [16] Y. Yan, X. Qin, Y. Wu, N. Zhang, J. Fan, and L. Wang, “A restricted
classification performances are reported in the literature. In Boltzmann machine based two-lead electrocardiography classification,”
2015 IEEE 12th International Conference on Wearable and Implantable
this study, the efficiency of the DL algorithms has been Body Sensor Networks (BSN). pp. 1–9, 2015.
proven with high classification performances of 95.05%, [17] G. E. Hinton, S. Osindero, and Y.-W. Teh, “A fast learning algorithm
93.87%, and 94.51% for accuracy, sensitivity, and selectivity, for deep belief nets.,” Neural Comput., vol. 18, no. 7, pp. 1527–54, 2006.
respectively. [18] H. S. Lee, Q. lan Cheng, and N. V. Thakor, “ECG waveform analysis by
significant point extraction. I. Data reduction,” Comput. Biomed. Res.,
vol. 20, no. 5, pp. 410–427, 1987.
IV. CONCLUSIONS [19] D. Ververidis and C. Kotropoulos, “Sequential forward feature selection
PQRS complexes and T waves plots a regular form in with low computational cost,” in Signal Processing Conference 2005
normal sinus rhythm. Any obvious changes occurring in the 13th European, 2005, vol. 13, pp. 1–4.
[20] Z. Zhang, J. Dong, X. Luo, K. S. Choi, and X. Wu, “Heartbeat
PQRST lines indicate the irregularity or arrhythmia in classification using disease-specific feature selection,” Comput. Biol.
heartbeats. Since the determination of the features such as Med., vol. 46, no. 1, pp. 79–89, 2014.
intervals, segment measurements, heart rate, and the [21] P. Melin, J. Amezcua, F. Valdez, and O. Castillo, “A new neural
frequency of R waves have great benefits for clinicians to network model based on the LVQ algorithm for multi-class
classification of arrhythmias,” Inf. Sci. (Ny)., vol. 279, pp. 483–497, Sep.
identify the cardiac diseases and arrhythmias, the physiology 2014.
and the morphology of the ECG waveforms have frequently [22] M. Thomas, M. K. Das, and S. Ari, “Automatic ECG arrhythmia
been used in clinical trials [12], [14]. The meaningful data classification using dual tree complex wavelet based features,” AEU -
points for arrhythmia classification are thickened between S-T Int. J. Electron. Commun., vol. 69, no. 4, pp. 715–721, 2015.
[23] A. Batra and V. Jawa, “Classification of Arrhythmia Using Conjunction
waves and P-Q waves for the proposed DBN-based multistage of Machine Learning Algorithms and ECG Diagnostic Criteria,” Int. J.
classification model. Biol. Biomed., vol. 1, pp. 1–7, 2016.
The proposed DBN-based multistage arrhythmia [24] H. Leutheuser, S. Gradl, P. Kugler, L. Anneken, M. Arnold, S.
classification has discriminated five types of heartbeats with a Achenbach, and B. M. Eskofier, “Comparison of real-time classification
systems for arrhythmia detection on Android-based mobile devices,”
high accuracy rate of 95.05%. The achievements prove the IEEE Eng. Med. Biol. Soc. Annu. Conf., vol. 2014, pp. 2690–2693, 2014.
success and efficiency of the DBN algorithm in raw ECG [25] N. Alajlan, Y. Bazi, F. Melgani, S. Malek, and M. A. Bencherif,
signals. “Detection of premature ventricular contraction arrhythmias in
electrocardiogram signals with kernel methods,” Signal, Image Video
REFERENCES Process., vol. 8, no. 5, pp. 931–942, Jul. 2014.
[26] M. M. Al Rahhal, Y. Bazi, H. AlHichri, N. Alajlan, F. Melgani, and R.
[1] WHO, “The top 10 causes of death.” [Online]. Available: R. Yager, “Deep Learning Approach for Active Classification of
http://www.who.int/mediacentre/factsheets/fs310/en/. [Accessed: 07- Electrocardiogram Signals,” Inf. Sci. (Ny)., vol. 345, pp. 340–354, Feb.
Aug-2016]. 2016.
[2] J. G. Webster, Medical Instrumentation, Application and Design, 4th ed. [27] M. Huanhuan and Z. Yue, “Classification of Electrocardiogram Signals
Boston: Houghtoon Mifflin Company, 1978. with Deep Belief Networks,” Computational Science and Engineering
[3] Y. C. Yeh, C. W. Chiou, and H. J. Lin, “Analyzing ECG for cardiac (CSE), 2014 IEEE 17th International Conference on. pp. 7–12, 2014.
arrhythmia using cluster analysis,” Expert Syst. Appl., vol. 39, no. 1, pp. [28] N. Allahverdi, G. Altan, and Y. Kutlu, “Diagnosis of Coronary Artery
1000–1010, 2012. Disease Using Deep Belief Networks,” 2. Int. Conf. Eng. Nat. Sci., The
[4] P. Hamilton, “Open source ECG analysis,” Comput. Cardiol., vol. 29, Book of Abstracts p.9, 2016.
pp. 101–104, 2002. [29] M. I. Owis, A. H. Abou-Zied, a. B. M. Youssef, and Y. M. Kadah,
[5] D. Ge, N. Srinivasan, and S. M. Krishnan, “Cardiac arrhythmia “Study of features based on nonlinear dynamical modeling in ECG
classification using autoregressive modeling.,” Biomed. Eng. Online, vol. arrhythmia detection and classification,” IEEE Trans. Biomed. Eng., vol.
1, p. 5, 2002. 49, no. 7, pp. 733–736, 2002.
[6] S. A. Israel, J. M. Irvine, A. Cheng, M. D. Wiederhold, and B. K. [30] R. J. Martis, U. R. Acharya, C. M. Lim, and J. S. Suri, “Characterization
Wiederhold, “ECG to identify individuals,” Pattern Recognit., vol. 38, of ECG beats from cardiac arrhythmia using discrete cosine transform in
no. 1, pp. 133–142, 2005. PCA framework,” Knowledge-Based Syst., vol. 45, pp. 76–82, 2013.
[7] K. N. Plataniotis, D. Hatzinakos, and J. K. M. Lee, “ECG Biometric [31] J. Kim, S. D. Min, and M. Lee, “An arrhythmia classification algorithm
Recognition Without Fiducial Detection,” in 2006 Biometrics using a dedicated wavelet adapted to different subjects.,” Biomed. Eng.
Symposium: Special Session on Research at the Biometric Consortium Online, vol. 10, no. 1, p. 56, 2011.
Conference, 2006, pp. 1–6. [32] P. Tadejko and W. Rakowski, “Hybrid wavelet-mathematical
[8] Y. Bengio and O. Delalleau, “Justifying and generalizing contrastive morphology feature extraction for heartbeat classification,” in
divergence,” Neural Comput., vol. 21, no. 6, pp. 1601–1621, 2009. EUROCON 2007 - The International Conference on Computer as a
[9] G. B. Moody and R. G. Mark, “The impact of the MIT-BIH arrhythmia Tool, 2007, pp. 127–132.
database.,” IEEE Eng. Med. Biol. Mag., vol. 20, no. 3, pp. 45–50. [33] M. Llamedo and J. P. Martinez, “Heartbeat classification using feature
[10] Y. Kutlu and D. Kuntalp, “A multi-stage automatic arrhythmia selection driven by database generalization criteria,” IEEE Trans.
recognition and classification system,” Comput. Biol. Med., vol. 41, no. Biomed. Eng., vol. 58, no. 3 PART 1, pp. 616–625, 2011.
1, pp. 37–45, 2011. [34] A. S. Alvarado, C. Lakshminarayan, and J. C. Príncipe, “Time-based
[11] U. Rajendra Acharya, J. S. Suri, J. A. E. Spaan, and S. M. Krishnan, compression and classification of heartbeats,” IEEE Trans. Biomed.
Advances in cardiac signal processing. 2007. Eng., vol. 59, no. 6, pp. 1641–1648, 2012.
[12] M. Gabriel Khan, Rapid ECG Interpretation(Contemporary Cardiology), [35] C. Ye, B. V. K. Vijaya Kumar, and M. T. Coimbra, “Heartbeat
____________________________________________________________________________________________________________
244
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
245
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Amount of text data being generated each day is Leading motivation behind collecting and analysing data
so huge that it is impossible to keep track of. Clustering text is reaching a higher level of profit. Effective analysis of pro-
documents is an important task to make use of information duced data enables marketing products in a more proactive
carried by plain text. In many text document collections,
clusters do not form easily separable compact shapes. Spectral way and governing customers effectively. Another purpose
clustering is an efficient when clustering such datasets because of analysis of data is enhancing performance of system.
algorithm considers connectedness of clusters rather than For example, in a search application, it is important to
assuming a specific statistical model and optimizing parameters reduce search space and prevent search operation in whole
of that model. On the other hand, distributed data processing data store. Many search engines crawl and index whole
tools have been becoming more and more widespread. Running
clustering algorithms with distributed tools is getting essential pages existing in world wide web. Performing a user search
in big data environments. In this research, spectral clustering operation on a related subset of web document collection
is executed on text data using big data analysis tools. Effects can provide a quicker search to user and less overhead to the
of algorithm on document clustering is discussed with respect vendor. This concern gets apparent in most of applications
to cluster quality measures. Moreover, existing works about providing services using big data.
scalability of spectral clustering is investigated. Results show
that spectral clustering algorithm categorizes documents more Recently, users of the software applications tend to search
accurately than k-means. In experiments, text documents in specific information with several keywords. They expect
20-newsgroups are used as dataset, Apache Mahout is used search system to be intelligent enough to examine requested
as distributed machine learning library, Rand Index and information and fetch relevant results even data entities are
Mutual Information measures are used to evaluate clustering not free text documents. Moreover, important part of the
performance.
Keywords: Text mining, document clustering, distributed data produced by world consist of text documents such as
systems, spectral clustering, big data news, blog posts, comments, product descriptions, and so on.
Mining text data is important because of these facts.
I. I NTRODUCTION In text mining, there are several issues where discussions
As information technologies become more widespread are focused. First of all, converting text documents into
in many aspects of the life, amount of data produced is vectors. It is necessary to express instances in dataset to
growing gradually. Applications producing data includes run learning algorithms in data mining. Data contained by
both enterprise and public applications. Stock management, text documents consists of terms and number of terms in
enterprise resource planning, customer relationship manage- common text data sets are so high that they can be expressed
ment applications are examples of enterprise applications. with tens of thousands. Second issue is distance / similarity
Social network applications, forums, news portals and blogs measure to use. It is important to choose distance best fitting
are examples of public applications. One of the most impor- to data domain in order to get better results. Another issue is
tant common characteristics of these two separate software algorithm to classify / cluster dataset. Characteristics of data
application groups is the fact that data produced using these domain and shape of data are important factors to determine
applications should be stored and analysed. Storing data algorithm.
is important when auditing and tracking past transactions. This study focuses on clustering of text documents using
Unless these are required, storing data without analysing is spectral clustering algorithm [1], which is one of the
nothing other than wasting storage units of machines. effective methods for clusters that can hardly be separated
Most of content & service providers analyse data gener- linearly. It regards affinity matrix of data. Entries of this
ated by users of systems they provide. For example, CRM matrix are pairwise affinities between data instances. Algo-
application logs are analysed for segmentation or labelling of rithm is composed around eigenvectors of this matrix and
customer, user actions in an e-commerce application will be has successful applications in image segmentation, that is
logged for promoting right product to right customer, social another domain where clusters frequently do not seem in
network application logs are analysed for providing people linearly separable compact shapes [2]. Moreover, in big data
and content that are closely related to users, search engine environments distribution of computation is usually required.
logs are analysed for listing web pages relevant to search Therefore, when clustering big datasets, accuracy of the
terms in an acceptable time amount. used method is not sufficient alone. It must also scalable in
____________________________________________________________________________________________________________
246
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
order to be applied for such datasets. Therefore, distributed calculations are mentioned in this paper. The article prepared
tools are used in the work in scope of this study. Methods by Datta et. al. provides an overview of distributed data
and algorithms are evaluated based parallel and distributed mining algorithms and applications in peer-to-peer networks
execution. [16].
In section 2, previous work on spectral clustering and
text mining is introduced. Then, in the 3rd section, spectral III. S PECTRAL C LUSTERING O N T EXT DATA
clustering algorithm is described in detail and some issues
about how to apply algorithm are discussed. In section 4, ex- A. Representing text data
periments are explained including methods, used tools, data
Data instances are required to be expressed in vectors in
set and experimental results. Then, final remarks, concluding
learning applications. In text documents, data information
words and possible future work are expressed in section 5.
is carried by terms. Therefore, as well as size of document
II. R ELATED W ORK vectors, algebraic value of entries in document vectors is
determined by terms that occurs in documents. Basically,
Text document clustering is a main branch of data mining
document x can be converted to vectors as follows:
so there exist some surveys about it. M. W. Berry published
an extensive survey of text clustering [3]. In that study, effec-
x = {t1 : tf1 , t2 : tf2 ..., td : tfd } (1)
tive vectorization of text passages, dimensionality reduction
methods and term-to-term similarities are mentioned. In the and where x is vector representation of document, ti s are
survey prepared by Sathiyakumari et al. [4], pros and cons of terms falling into document x, and tfi s are the frequency
several clustering algorithms on text data are revealed and (number of occurrence) of ti in x. This vector is called term
they are compared. In the study of Anna [8], similarity frequency vectors. A more advanced and more efficiently
measures are compared using experiments on several data used vector model for text document is tf-idf model:
sets. Aggarwal and Zhao made a comprehensive study about
issues that should be overcome in text clustering [10]. In x = {t1 : w1 , t2 : w2 ..., td : wd } (2)
that study, feature selection, dimensionality reduction and
algorithm choice are discussed topics. and
Spectral clustering algorithm is based on spectral graph
theory. The theory includes analysis of graphs using eigen- |S|
wi = tfi · (3)
vectors of the matrix representing pairwise similarities of |{xj ∈ S|ti ∈ xj }|
vertices in the graph. Spielman’s lecture notes are one of
where S is data set, xi s are documents in data set. This
the resources explaining theory well [15]. On the usage of
model is called tf-idf model. In this model, lower scores
spectral clustering for text data, Vempala and Wang assert
are assigned to terms that are not distinctive for document.
spectral projection enhances clustering results on document
This model is one of the most common techniques used
collections [5]. Their assertion is based on experimental
to transform text documents into vectors. Derivation of tf-
results. Chen et al. address scalability problem of spectral
idf vectors from plain text is done by Apache Mahout’s
clustering [11]. They offer methods for approximating dense
seq2sparse utility tool [7].
similarity matrix and parallelize computations in spectral
clustering. Chennubhotla and Jepson proposes a hierarchical Learning algorithm executions require a distance/similarity
algorithm for efficiently compute largest eigenvectrs of large measure to cluster dataset into groups. When deciding dis-
matrices [12]. A parallel eigensolver algorithm is proposed tance measure characteristics of the dataset domain is an
by Bientinesi et al. [13] and empirical success of their important factor. There is no global right distance measure
algorithm is expressed by comparing with other well-known but there are effective distance measures for data type. Cosine
eigensolver algorithms. Strakova and Gansterer introduce a distance measure is one of the best fitting ones when working
distributed eigensolver [14]. Their proposed method con- on text data so it is used in experiments of this study. With
siders only nearest neighbour similarities and is efficient in this measure, distance between two data instances xi and xj
loosely coupled networks that are distributed arbitrarily. is calculated in following way :
As the notion of ”big data” become more widespread, x~i · x~j
more studies about distributed tools and algorithms to pro- dij = (4)
|xi | · |xj |
cess big datasets are being done. Judith and Jayakumari
prepared a review about recent work on distributed document As the name implies, cosine distance dij is equals to the
clustering algorithms [17]. In this study, existing distributed cosine of the angel between two data vectors xi and xj . Cal-
solutions are compared in terms of speed, accuracy, cluster culating affinity from distance is another issue. Calculation
quality and scalability. In another work done by Li et al. of cosine similarity between text documents is as follows:
[18], distributed k-means algorithm executing on MapReduce
model is evaluated. Possible enhancements approaches like aij = 1 − dij (5)
partitioning dataset with an effective hashing algorithm, ef-
fective center initialization and pruning unnecessary distance where dij is cosine distance.
____________________________________________________________________________________________________________
247
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
B. Spectral Clustering Algorithm approach when creating affinity matrix is using a graph
Spectral clustering algorithm is an effective way of data construction method. In that way, many weak connections
clustering because it considers pairwise affinities of data between instances are pruned in order to relax calculation
instances without making any statistical model assumptions. and promote strong local neighbourhoods. Two techniques
When dataset is composed of complex-shaped connected are used for graph construction:
clusters rather than compact clusters, ordinary clustering • -neighbourhood graph : Affinity dij between xi
algorithms like k-means are hard to perform well. and xj is considered only if its value is higher than
a predefined constant value . Choosing parameter is
important to determine which connections will exist and
which will be pruned.
• k-nearest neighbourhood graph : For each data
instance, connections to only k-nearest neighbours are
considered. Other connections are ignored. Affinity ma-
trix is created using k highest affinity values for each
xi .
Another issue in spectral clustering algorithm is calcula-
tion of graph Laplacian. When calculating Laplacian matrix,
degree matrix is used besides affinity matrix. In fact, it is
derived using affinity matrix. Degree matrix is a diagonal
matrix whose ith diagonal entry is the sum of affinities
Fig. 1 Example of compact and connected clusters calculated for instance i.
____________________________________________________________________________________________________________
248
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
score
talk.politics.guns
sci.med 0.48
talk.politics.mideast
sci.space
0.39
talk.religion.misc 0.4
alt.atheism misc.forsale 0.3
soc.religion.christian
0.2
IV. E XPERIMENTS
0 k-means spectral
A. Dataset algorithm
20-newsgroups dataset hosted by UCI Machine Learning Fig. 2 Adjusted Rand Index and Adjusted Mutual Informa-
Repository [6] is used in tests executions. It contains 18846 tion scores of k-means and spectral clustering algorithms
documents of 20 different groups. There are 93,560 distinct
terms in dataset. No NLP method such as stemming or
stop-word elimination is used. In other words, ”play” and
Figure 2 shows clustering results of k-means and spectral
”playing” are accepted different terms for our algorithm.
clustering algorithms. Spectral clustering apparently per-
Topics in 20-newsgroups dataset are listed in Table I.
forms better than k-means algorithm in seen in the figure.
Enhancement of score is obvious for both adjusted rand index
B. Tools Used and adjusted mutual information measures. Then, it can be
Apache Hadoops distributed file system is used for storing said that spectral clustering has a more accurate clustering
data and MapReduce is used for processing data. MapReduce result on text data than k-means.
is a distributed processing model that fits into many real V. C ONCLUSION AND F UTURE W ORK
world problems [9]. Hadoop is one of the most common
scalable MapReduce vendors and it has proven success in Clustering text documents is an important because text
many real time systems. Hadoop is run in pseudo-distributed data constitutes an important part of the data produced each
mode in our experiments. day. When running learning algorithms, data sets may not
Machine learning library used in experiments is Apache be easily separable. Text document collections usually exist
Mahout [7]. It works on top of Hadoop. Machine learning in hardly-separable forms. In scope of this study, we focus
tasks are submitted by Mahout to Hadoop and they are our effort to apply spectral clustering algorithm on text
executed as MapReduce jobs. Mahout does not only provide data. Effects of spectral clustering algorithm observed on
machine learning algorithms but also perform some utility 20-newsgroups data set. Results show that accuracy of the
tasks such as creating tf-idf vectors from plain text docu- spectral clustering algorithm is better compared to k-means.
ments or calculating distance between vectors. Solutions are prepared using distributed tools in order to
make use it in big data problems.
C. Experimental Results For future work, a dimensionality reduction technique may
As stated earlier, k-means and spectral clustering algo- be applied in order to get rid of high number of features in tf-
rithms are executed to cluster 20-newsgroup dataset in scope idf vector form. Another enhancement may be using a graph
of experiments in this research. Experiments are run 10 times construction method in order to prune unnecessary distance
consecutively using Mahout’s command line interface. Some and affinity calculations. Experiments in this study are done
parameters passed required to clustering commands are as without using any graph construction method.
follows:
• convergence delta : 0.001
R EFERENCES
• max. number of iterations : 100 [1] Ng, Andrew Y. and Jordan, Michael I. and Weiss, Y. (2001) On Spec-
• initial clustering : Canopy tral Clustering: Analysis and an Algorithm. Advances in Neural In-
formation Processing Systems, Vancouver, British Columbia, Canada,
• distance measure : Cosine December 4-6, pp. 849–856. MIT Press, Cambridge, Massachusetts,
• num. of clusters : 20 London, England.
[2] Bach, Francis R. and Jordan, Michael I. (2006), Learning Spectral
After executions of clustering algorithms, found cluster re- Clustering, With Application To Speech Separation, Journal of Ma-
sults are evaluated in terms of two cluster quality measures chine Learning Research, Vol. 7, pp. 1963–2001
: adjusted rand index and adjusted mutual information. [3] Berry, M.W. (2011), Survey of Text Mining: Clustering, Classification
and Retrieval, Springer, New York.
____________________________________________________________________________________________________________
249
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[4] Sathiyakumari, K. and Manimekalai, G. and Preamsudha, V. (2011), A [15] Spielman, D. (2009). Spectral graph theory. Lecture Notes, Yale
Survey on Various Approaches in Document Clustering, International University, 740-0776.
Journal of Computer Technplogy and Applications (IJCTA), Vol. 2, [16] Datta, S., Bhaduri, K., Giannella, C., Wolff, R., & Kargupta, H. (2006).
pp. 1534–1539. Distributed data mining in peer-to-peer networks. IEEE Internet Com-
[5] Vempala, S. and Wang, G. (2005), On the Benefit of Spectral Pro- puting, 10(4), 18-26.
jection for Document Clustering, SIAM Conference on Data Mining [17] Judith, J. E., & Jayakumari, J. (2015). Distributed document cluster-
(SDM’05), California, USA ing algorithms: a recent survey. International Journal of Enterprise
[6] Mitchell, Tom, UCI Machine Learning Repository: Network Management, 6(3), 207-221.
Twenty Newsgroups Data Set (1999), available at [18] Li, Q., Wang, P., Wang, W., Hu, H., Li, Z., & Li, J. (2014, April). An
https://archive.ics.uci.edu/ml/datasets/Twenty+Newsgroups efficient K-means clustering algorithm on MapReduce. In International
Conference on Database Systems for Advanced Applications (pp. 357-
[7] Owen, S. and Anil, R. and Dunning, T. and Friedman, E. (2011),
371). Springer International Publishing.
Mahout in Action, Manning Publications Co., Greenwich, CT, USA
[8] Huang, Anna (2008), Similarity measures for text document clustering
Proceedings of the sixth new zealand computer science research
student conference (NZCSRSC2008), Christchurch, New Zealand, pp.
49–56
[9] Dean, J. and Ghemawat, S (2004), MapReduce: Simplified Data Pro-
cessing on Large Clusters, OSDI’04: Sixth Symposium on Operating
System Design and Implementation, San Francisco, CA, December,
2004.
[10] Aggarwal, C. C., Zhao, P. (2013). Towards graphical models for text
processing. Knowledge and information systems, 36(1), 1-21.
[11] Chen, W. Y., Song, Y., Bai, H., Lin, C. J., & Chang, E. Y. (2011).
Parallel spectral clustering in distributed systems. IEEE transactions
on pattern analysis and machine intelligence, 33(3), 568-586.
[12] Chennubhotla, C., & Jepson, A. (2005). Hierarchical eigensolver for
transition matrices in spectral methods. Advances in Neural Informa-
tion Processing Systems, 17, 273-280.
[13] Bientinesi, P., Dhillon, I. S., & Van De Geijn, R. A. (2005). A parallel
eigensolver for dense symmetric matrices based on multiple relatively
robust representations. SIAM Journal on Scientific Computing, 27(1),
43-66.
[14] Strakov, H., & Gansterer, W. N. (2013, February). A distributed
eigensolver for loosely coupled networks. In 2013 21st Euromicro
International Conference on Parallel, Distributed, and Network-Based
Processing (pp. 51-57). IEEE.
____________________________________________________________________________________________________________
250
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—The main purpose of this study is to predict wine quality physicochemical variables. Another work of wine classification
based on physicochemical data. In this study, two large separate depended on the physicochemical information. This
data sets which were taken from UC Irvine Machine Learning information involved in wine aroma chromatograms as
Repository were used. These data sets contain 1599 instances for measured with a Fast GC Analyser [6]. In the latter study, three
red wine and 4898 instances for white wine with 11 features of
classification methods such as Linear Discriminant Analysis,
physicochemical data such as alcohol, chlorides, density, total
sulfur dioxide, free sulfur dioxide, residual sugar, and pH. First, Radial Basis Function Neural Networks and Support Vector
the instances were successfully classified as red wine and white Machines (SVM) are compared according to their performance
wine with the accuracy of 99.5229% by using Random Forests in a two-staged architecture.
Algorithm. Then, the following three different data mining Some have proposed a few applications of data mining
algorithms were used to classify the quality of both red wine and techniques to wine quality assessment. Cortez et al. [1]
white wine: k-nearest-neighbourhood, random forests and proposed a taste prediction technique. Their taste prediction
support vector machines. There are 6 quality classes of red wine technique, a support vector machine, multiple regression, and a
and 7 quality classes of white wine. The most successful neural network were applied to chemical analysis of wines.
classification was obtained by using Random Forests Algorithm.
Shanmuganathan’s technique was about prediction the effects
In this study, it is also observed that the use of principal
component analysis in the feature selection increases the success of season and climate on wine yields and wine quality [7]. The
rate of classification in Random Forests Algorithm. Wineinformatics system according to Chen et al. [8] idealized
the flavor and characteristics of wine from natural-language
reviews. They used hierarchical clustering and association rules.
Keywords— Classification, Random Forests, Support Vector
Machines, k-Nearest-Neighbourhood. II. MATERIALS AND METHODS
I. INTRODUCTION A. Wine Data
Today, varied consumers enjoy wine more and more. Wine The data set is a wine quality dataset that is publicly
industry is researching new technologies for both wine making available for research purposes from
and selling processes in order to back up this growth [1]. https://archive.ics.uci.edu/ml/datasets/Wine+Quality [9]. Both
Physicochemical and sensory tests are used for evaluating dataset contains 1599 instances with 11 features for red wine
wine certification [2]. The discrimination of wines is not an and 4898 instances and the same 11 features for white wine.
easy process owing to the complexity and heterogeneity of its The inputs include objective tests (e. g. PH values) and the
headspace. The classification of wines is very important output is based on sensory data (median of at least 3 evaluations
because of different reasons. These reasons are economic value made by wine experts). Each expert graded the wine quality
of wine products, to protect and assure the quality of wines, to between 0 (very bad) and 10 (very excellent). The two datasets
forbid adulteration of wines, and to control beverage processing are related to red and white variants of the Portuguese “Vinho
[3]. Verde” wine.
Data mining technologies have been applied to classification The features include fixed acidity, volatile acidity, citric acid,
of wine quality. The aim of machine learning methods similar residual sugar, chlorides, free sulfur dioxide, total sulfur
to other applications is to create models from data to predict dioxide, density, pH, sulphates, and alcohol. pH describes how
wine quality. acidic or basic a wine is on a scale from 0 (very acidic) to 14
In the year of 1991, a “Wine” data set which contains 178 (very basic). Most wines are between 3-4 on the pH scale.
instances with measurements of 13 different chemical Chloride is the amount of salt in the wine. Alcohol is the percent
constituents such as alcohol, magnesium was donated into UCI alcohol content of the wine.
repository to classify three cultivars from Italy [4]. For new The goal of the data set is to predict the rating that an expert
data mining classifiers this data set has been majorly used as a will give to a wine sample, using a range of physicochemical
benchmark because it is very easy to discriminate. For wine properties, such as acidity and alcohol composition. Due to
classification according to geographical region; principal privacy and logistic issues, only physicochemical (inputs) and
component analysis (PCA) was carried out and reported [5]. sensory (the output) variables are available (e.g. There is no
The data they used in their study includes 33 Greek wines with data about grape types, wine brand, wine selling price, etc.).
____________________________________________________________________________________________________________
251
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
252
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
253
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The most successful classification result of red wine sample [10] J. Han, M. Kamber, and J. Pei, “Classification: Basic Concepts,” in Data
Mining Concepts and Techniques, 3rd ed., Waltham, MA, USA: Morgan
qualities after applying PCA was obtained by using Random
Kaufmann, 2012, pp. 327-393.
Forest algorithm for both test modes. The accuracy of each [11] J. Han, M. Kamber, and J. Pei, “Classification: Advanced Methods,” in
cross-validation and percentage split mode with this algorithm Data Mining Concepts and Techniques, 3rd ed., Waltham, MA, USA:
is 71.232%, and 73.4375% respectively. Morgan Kaufmann, 2012, pp. 393-443.
[12] W. L. Martinez, A. R. Martinez, “Supervised Learning” in
IV. CONCLUSIONS Computational Statistics Handbook with MATLAB, 2nd ed., Boca Raton,
FL, USA: Chapman & Hall/CRC, 2007, pp. 363-431.
For each classification model, we analyzed how the results
vary whenever test mode is changed. The study includes the
analysis of classifiers on both red and white wine data set. The
results are described in percentage of correctly classified
instances, precision, recall, F measure, and ROC after applying
the cross- validation or percentage split mode.
Different classifiers like k-nearest-neighborhood, random
forests, and support vector machines are evaluated on datasets.
Results from the experiments lead us to conclude that
Random Forests Algorithm performs better in classification
task as compared against the support vector machine, and k-
nearest neighbourhood.
After applying PCA, the success rate of quality classification
for white wine has decreased from 70.3757% to 69.9061% for
cross validation mode. The success rate of quality classification
for white wine has decreased from 68.6735% to 67.449% for
percentage split mode.
After applying PCA, the success rate of quality classification
for red wine has increased from 69.606% to 71.232% for cross
validation mode. The success rate of quality classification for
red wine samples has increased from 71.875% to 73.4375% for
percentage split mode.
REFERENCES
[1] P. Cortez, A. Cerderia, F. Almeida, T. Matos, and J. Reis, “Modelling
wine preferences by data mining from physicochemical properties,” In
Decision Support Systems, Elsevier, 47 (4): 547-553. ISSN: 0167-9236.
[2] S. Ebeler, “Linking Flavour Chemistry to Sensory Analysis of Wine,” in
Flavor Chemistry, Thirty Years of Progress, Kluwer Academic
Publishers, 1999, pp. 409-422.
[3] V. Preedy, and M. L. R. Mendez, “Wine Applications with Electronic
Noses,” in Electronic Noses and Tongues in Food Science, Cambridge,
MA, USA: Academic Press, 2016, pp. 137-151.
[4] A. Asuncion, and D. Newman (2007), UCI Machine Learning
Repository, University of California, Irvine, [Online]. Available:
http://www.ics.uci.edu/~mlearn/MLRepository.html
[5] S. Kallithraka, IS. Arvanitoyannis, P. Kefalas, A. El-Zajouli, E.
Soufleros, and E. Psarra, “Instrumental and sensory analysis of Greek
wines; implementation of principal component analysis (PCA) for
classification according to geographical origin,” Food Chemistry, 73(4):
501-514, 2001.
[6] N. H. Beltran, M. A. Duarte- MErmound, V. A. S. Vicencio, S. A. Salah,
and M. A. Bustos, “Chilean wine classification using volatile organic
compounds data obtained with a fast GC analyzer,” Instrum.
Measurement, IEEE Trans., 57: 2421-2436, 2008.
[7] S. Shanmuganathan, P. Sallis, and A. Narayanan, “Data mining
techniques for modelling seasonal climate effects on grapevine yield and
wine quality,” IEEE International Conference on Computational
Intelligence Communication Systems and Networks, pp. 82-89, July
2010.
[8] B. Chen, C. Rhodes, A. Crawford, and L. Hambuchen,
“Wineinformatics: applying data mining on wine sensory reviews
processed by the computational wine wheel,” IEEE International
Conference on Data Mining Workshop, pp. 142-149, Dec. 2014.
[9] UCI Machine Learning Repository, Wine quality data set, [Online].
Available: https://archive.ics.uci.edu/ml/datasets/Wine+Quality.
____________________________________________________________________________________________________________
254
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract: Connection-oriented network routing (CONR) is one of In literature, different researchers dealt with the subjects of
the NP-hard problems. In the recent years, to solve the problem of
network routing and finding the shortest route. These
CONR it was using the heuristic methods such as minimum researchers applied different kinds of heuristic algorithms and
spanning tree methods (Kruskal and Prim Algorithms) and metaheuristic algorithms to solve the problem of CONR.
shortest path methods (A Star and Dijkstra Algorithms) and also
Karakas [1] applied Dijkstra and Prim Algorithms to decrease
metaheuristic methods (Ant Colony Optimization (ACO), Genetic
Algorithm (GA) and Artificial Bee Colony (ABC)). But as the the cost of creating network across Turkey to the half. Hardi
complexity of the problem increase, the guarantee to achieve the [2] used the Genetic Algorithms (GA) to determine the
shortest route that could be followed by the mineral water
best results by heuristic algorithms decreases. Therefore, in large
scale problems metaheuristic algorithms are preferred. In this company in order to enable the distributors to get to their
destinations through the shortest distance of all available paths.
study, it was applied CONR to provide internet and telephone lines
for Kirkuk city and the near regions in Iraqi by using the fiber Mahia et al. [3] proposed new hybrid method to optimize
optical systems. Firstly, the locations of 56 fiber optic panels parameters that affect performance of the ACO algorithm
previously determined with particle swarm intelligence algorithm using Particle Swarm Optimization (PSO). PSO is preferred to
(PSO) are placed Kirkuk city. Then, it was aimed to wire all the
detect optimum values of parameters which are used for city
panels by fiber optic cables in minimum cost, to raise the data
selection operations in the ACO algorithm and determines
transfer speed, to decrease the excavation process. As a result, two
different metaheuristic algorithms (ACO and GA) were applied to significance of inter-city pheromone and distances.
achieve these targets and compared the results of them. After theExperimental results showed that the performance of proposed
experiments, the optimum distance between the panels were method by using fewer ants than the number of cities for the
obtained with ACO and GA as 366100 km and 353400 km, TSPs was better than the performance of compared methods
respectively. The results show that GA is the best optimization in most cases in terms of solution quality and robustness.
algorithm for this problem. Kuzu et al. [4] used eight of the most common kinds of
metaheuristic algorithms to solve the TSP problem which is
Keywords— Connection-oriented network routing,Ant Colony
considered one of the NP-hard problems. Dikmen et al [5]
Optimization,Genetic Algorithm,fiber optical systems,CONR applied the artificial intelligence techniques to solve the
travelling salesman problem (TSP) in order to find the
shortest ways to connect the Turkish cities. The results
showed that the ACO method is better than the GA
concerning the distance and time. Fong and Yuan [6] applied
I. INTRODUCTION
ACO to solve the TSP problem and reached the best results of
When transmitting data through a network from one source network routing solution. İşçi and Korukoğlu [7] applied
to another the data follows a special path. In its way towards heuristic algorithms (prim & kruskal) and metaheuristic
the target area the data passes through a chain of panels, this algorithm (GA) to decrease the cost of connecting network
process is called network routing [1,2]. cables. The results showed that GA is better than prim &
Whether the distribution path is reasonable, it has large kruskal concerning this issue [7]. Kaya [8] applied Minimum
impact on speed, cost and effectiveness in the logistics system, spanning tree algorithms (Floyd &kruskal) to find the shortest
so it is very important to use a reasonable method to path to provide gas pipeline to a particular area.
determine a reasonable distribution route. The exact solution In this study, it was proposed CONR to provide internet and
is very difficult to study due to the NP-Hard nature of the telephone lines for Kirkuk city and its near region in Iraqi
problem, so heuristics may well be a viable direction [1,2]. using the fiber optical systems. The CONR consists a number
One of the NP-hard problems is connection-oriented network of Panels, where each pair of panels has a definite distance.
routing (CONR). The purpose of this study is to visit all the panels such that the
____________________________________________________________________________________________________________
255
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
total distance will be minimized. Also, it was aimed to wire all In order to solve this problem, it was determined the
the panels by fiber optic cables in minimum cost, to raise the distances between these panels in a two-dimensional matrix as
data transfer speed and to decrease the excavation process. As shown in Table 1, firstly.
a result, two different metaheuristic algorithms, ACO and GA,
were applied to achieve these targets and compared the results TABLE. 1. DISTANCES BETWEEN THE PANELS
of them.
Merkez PanelPanel 1 Panel 2 Panel 3 Panel 4 Panel 5 Panel 6 Panel 7 Panel 8 Panel 9 Panel 10 Panel 11 Panel 12 Panel 13 Panel 14 Panel 15 Panel 16 Panel 17 Panel 18 Panel 19 Panel 20
Merkez Panel 0,0 27,7 24,6 30,2 31,7 25,6 25,3 25,8 27,4 27,0 26,7 26,2 31,8 29,8 29,4 27,8 32,6 20,7 19,0 20,7 20,8
Panel 1 29,3 0,0 4,2 3,1 4,2 4,7 4,5 2,3 2,8 2,6 2,4 3,2 4,3 5,2 2,5 2,6 5,4 9,8 11,1 10,6 9,4
Panel 2 24,8 3,2 0,0 3,5 5,0 1,6 1,6 1,0 1,5 2,5 1,8 1,7 5,1 6,0 5,7 4,3 5,9 6,7 8,0 7,5 6,3
Panel 3 31,0 2,3 4,4 0,0 1,6 6,1 5,8 3,7 5,2 1,9 6,0 5,6 1,6 2,4 4,7 5,0 2,5 10,7 12,0 11,5 10,3
II. MATERIAL AND METHODS Panel 4 29,8
Panel 5 23,3
3,3
5,5
5,4
2,3
1,9
5,8
0,0
7,3
7,1
0,0
6,8
1,0
4,7
3,6
6,2
5,2
3,6
4,8
7,0
3,4
6,6
1,5
2,6
7,4
2,3
8,3
5,7
6,1
6,0
3,9
1,7
8,2
11,7
5,2
13,0
6,5
12,5
6,0
11,3
4,7
Panel 6 24,0 5,7 1,6 5,0 6,6 1,0 0,0 2,9 4,4 4,0 2,2 1,1 6,6 7,6 5,3 3,3 7,5 5,9 7,2 6,7 5,4
Panel 7 25,6 2,3 1,0 2,7 4,2 2,8 2,6 0,0 1,0 1,7 2,1 3,3 4,3 5,2 4,9 5,1 5,1 7,5 8,8 8,3 7,1
A. Problem Description Panel 8 26,2
Panel 9 29,2
2,8
1,6
1,5
3,7
3,3
1,9
4,8
4,5
3,4
6,4
3,2
6,1
1,0
1,7
0,0
5,5
2,3
0,0
1,3
5,6
2,6
5,9
4,9
3,6
5,8
4,0
5,0
5,2
3,7
5,3
5,7
5,4
8,1
12,4
9,4
12,3
8,9
11,8
7,7
11,9
Panel 10 25,8 3,9 1,8 4,2 5,5 2,7 2,6 2,6 1,3 4,1 0,0 1,1 5,5 6,5 3,4 2,1 6,3 7,7 9,0 8,5 7,3
Panel 11 24,6 4,9 1,7 5,1 6,7 1,5 1,1 2,3 1,9 4,1 1,1 0,0 6,8 7,7 5,4 3,5 7,6 6,6 7,8 7,3 6,1
Panel 12 29,9 4,4 6,4 3,0 2,8 8,4 7,8 5,9 7,4 4,7 7,8 7,6 0,0 14,0 6,8 8,3 4,5 13,5 14,7 14,2 13,6
In this study, connection-oriented network routing application Panel 13 30,9
Panel 14 29,8
4,5
5,1
6,5
6,8
3,0
4,2
2,3
5,0
8,2
6,6
7,9
6,3
5,8
6,2
7,3
7,7
4,8
5,1
7,5
4,9
7,7
5,1
14,0
5,0
0,0
6,0
6,8
0,0
4,5
4,5
5,4
5,4
12,9
11,7
14,1
12,9
13,6
12,5
12,4
11,2
____________________________________________________________________________________________________________
256
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Mutation Operator
The initial population generation
Mutation operator is also considered of the effective
When composing the initial population generation for elements in the performance of GA. The aim of mutation
CONR problem, there are 57 genes for each chromosome and operator is maintaining the diversity within the chromosome
this number differs from one problem to another. These 5 6 that exists in the population. The mutation operator cannot be
genes represent one chromosome and each chromosome is applied on all children (chromosomes), but only on a small
considered as path for the fiber optical cable. percentage of them. In CONR problem, the mutation operator
will swap the places of the panels in a single chromosome in
random form[11.14].
Fitness Function
____________________________________________________________________________________________________________
257
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
crossover and mutation operations, respectively. If the new TABLE 5. RESULTS OBTAINED FROM (GA) ALGORITHM.
population is better than the previous one, so the new
population will take its place. But if the previous one is better Search Information Result
than the new one, the new population will be neglected[11]. Number of iteration 3000
Found route distance 353400
1-9-51-54-21-55-43-42-
III. RESULTS AND DISCUSSION 45-25-23-24-17-18-46-
19-20-40-41-39-44-5-6-
In this study, it was investigated CONR problem with
11-2-7-8-10-15-37-33-
minimum cost and excavation process and maximum speed to Shortest route
52-36-38-35-34-30-31-
provide internet and telephone lines for Kirkuk city and the 32-29-53-28-50-27-26-
near regions in Iraqi using GA and ACO algorithms and the 49-56-57-48-16-4-13-
results of these algorithms were compared. To solve the 12-3-22-14-1
Kirkuk’s CONR problem, the C#2013 platform was preferred
for programming GA and ACO. The experiments showed that ACO algorithm gave the
The number of iteration was selected as 3000 times for both most optimal performance when the number of ants was equal
GA and ACO. Other parameter values for ACO and GA were to the number of fibre optic panels. Therefore, the shortest
given in Table 2 and Table 3, respectively. route obtained as 366100 km while the number of ants was
The obtained results from GA and ACO algorithms were selected as 56. Table 4 illustrates the optimum route distance
showed in Table 4 and Table 5. and the shortest route for ACO.
TABLE 2. As for GA when applying the same number of iterations
PARAMETER VALUES FOR ACO
3000, it was seen that the results was better than ACO
algorithm and the shortest route was obtained as 353400 km.
Parameters Value
From this results its concluded that the optimum solution for
Number of ants 56
CONR problem is using the GA algorithm with value out of
Pheromone relatively importance 1
353400 km. Furthermore, the shortest route obtained with GA
Relative importance of heuristic factor 3
Pheromone evaporation coefficient 0.1
was directly drew using actual locations of panels and this
route is shown in Fig. 5 which has been taken by the fibre
TABLE 3. optical cables in Kirkuk city.
PARAMETERS VALUES FOR (GA)
Parameters Value
Population size 3000
TABLE 4.
RESULTS OBTAINED FROM ACO ALGORITHM. Fig. 5. Shortest Route for Fibre Optics Cables in Kirkuk City
____________________________________________________________________________________________________________
258
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES [8] B. Kaya, “An Algorithm For Economically Efficient Layout” Deu
Engineering Faculty Of Science And Engineering Journal Volume: 1
[1] M.Karakaş “Internet Service Providers Cost Optimization”, Number: 1 p. 73-85 January 1999.
Department of Computer Engineering TOBB University of Economics [9] M. Dorigo, V. Maniezzo, A. Colorni, “Ant system optimization by a
and Technology in Ankara, Turkey, ¸2015. colony ofcooperating agents”, IEEE Trans. Syst. Man Cybern. B 26
[2] R. Hardi, “Genetic Algorithm in Solving the TSP on These Mineral (1996) 29–41.
Water”, Department of Informatics Engineering Sekolah Tinggi [10] M. Dorigo, T. Stutzle, “Ant Colony Optimization”, Massachusetts
Teknologi Bontang, Juanda No.73 RT. 36 Bontang, International Institute of Technology, 2004.
Seminar on Intelligent Technology and Its Applications, 2015. [11] K. Rani, V. Kumar “Solving Travelling Salesman Problem Using
[3] M. Mahi, Ö. K. Baykan, H.Kodaz,“A new hybrid method based on Genetic Algorithm Based On Heuristic Crossover And Mutation
Particle Swarm Optimization, Ant Colony Optimization and 3-Opt Operator” IMPACT: International Journal of Research in Engineering
algorithms for Traveling Salesman Problem” Applied Soft Computing & Technology (IMPACT: IJRET) ISSN(E): 2321-8843; ISSN(P):
30: 484-490 ,(2015).. 2347-4599 ,Vol. 2, Issue 2, 27-34, Feb 2014.
[4] S. Kuzu,, O. Önay,U. Şen, M.Tunçer, B.Yıldırım, T. Keskintürk, [12] H. Thomas, E. Leiserson, L. Rivest, “Introduction to Algorithms
“Solution with metaheuristic for the traveling salesman problem”, Second Edition”, PHI Learning.
Istanbul University Faculty of Management Journal, 2014. [13] Goldberg, D.E. (1989) “Genetic Algorithms in search, optimization
[5] H. Dikman,A. Albir , Z. Ekşi, F.Genlik, “Traveling Salesman Problem machine learning”, Pearson Education.
of Ant Colony and Genetic Algorithms optimization and Comparison” [14] G. Üçoluk ,“ Genetic Algorithm Solution of the TSP Avoiding Special
Süleyman Demirel University Journal of the Institute of Science and Crossover and Mutation” Department of Computer
Technology, Suleyman Demirel University Journal of Natural and Engineering ,middle East Technical University ,06531 Ankara, Turkey
applied Science , 18(1), 8-13, 2014.
[6] Y.Zhu Ju, F.LI Qing, “Solving the Travelling Salesman Problem by
theProgram of Ant Colony Algorithm” Chinese Academy of Surveying
& Mapping, CASM Chinese University of Mining & Technolegy 2009.
[7] Ö. İşçi, S. Korukoğlu, 06.12.2005, “Steiner Using Genetic Algorithm
for Solving the Minimum Spanning Tree Problem”, Pamukkale
University, Enginering College 13, Journal Of Engineering Sciences:
429-437.
____________________________________________________________________________________________________________
259
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In many image-processing applications, image seg- that consider local features are mostly preferred in real-time
mentation is an essential stage. In this stage, an image is applications [2].
partitioned into several regions according to the similarity of its Graph-based methods are some of the segmentation meth-
pixels. In addition to the accuracy of the image segmentation,
the speed is also very important for real-time image processing ods. Because of their representation relevance and ease of
applications. Many computer applications take advantages retention of images, the graph theory tools are utilized
of the multi-processor architecture to up to their running as prominent tools in computer vision [2]. In graph-based
performance. However, to run an algorithm as parallel is very segmentation methods, the nodes (or vertices) of the graph
difficult in many cases. Due to using the same memory blocks, refer the pixels of the image and the edges define the
many conflicts might be happened between the processors.
Moreover, each process of one processor may depend on connections between the adjacent (or neighboring) vertices.
those of another processor. For this reason, the algorithm to Each of these edges generally has a numerical value, which
be parallelized must be suitable to parallel. In addition, the is called weight [4]. The weights of the edges are considered
processing traffic that is pursued by the processors must be the difference of the gray level values or color values (if the
controlled within some parallel directives. In this paper, we segmentation is performed on a color image) of the vertices
provide a parallel implementation to a hierarchical graph-based
image segmentation method by using its hierarchical processing that exists at the ends of the edge. The difference is generally
steps. To achieve this goal, we utilize the OpenMP (Open Multi- computed by using Euclidean distance measurement [5].
Processing) Library to run the segmentation process as parallel Spanning trees are sub-graphs of a graph; such that they
on images of different sizes from the INRIA Holidays dataset. do not contain any loop and it is easy to make process on
The experimental results show that the parallel implementation the graph via the spanning tree structure. The spanning tree
of the algorithm is more effective than the serial type according
to processing time. of a graph that has the least total edge weight among the
Keywords— Parallel programming; Image segmentation; other spanning trees of the graph is called minimum spanning
Graph; OpenMP tree (MST) [4]. Image segmentation algorithms based on
minimum spanning tree take advantage of the features of
I. I NTRODUCTION the MST, because it takes no account of most of the edges
For many image processing application, the image seg- on the image-graph. Therefore, the cost of the operation is
mentation is one of the low-level and mid-level image reduced significantly [6].
processing stages. During the image segmentation stage, The segmentation process in MST-based segmentation
the pixels of an image are grouped into regions according algorithms comes true in two ways. The first way is the
to the similarity in the regions and differences among the cutting procedure in which the edges to be cut are removed
regions. This stage is one of the most difficult stages in the on the MST that covers all of the pixels of the image. The
image processing applications [1]. The consequence of this edges to be cut represent the borders on the image. By this
stage affects all of the following stages and their results. way, a tree is separated into several sub-trees. The sub-trees
Image segmentation reduces the large amounts of data for the represent regions in the image [5], [7], [8], [9]. The second
analysis operations. In an image, the segmentation process way is the merging procedure in which the sub-trees that
inherently focuses on the homogeneous regions and the bor- satisfies a homogeneity criterion are merged [10]. If two sub-
ders between them. It performs this task by using brightness, trees do not satisfy the criterion, they are not merged, and
color, pattern (texture), or density differences, etc. [2], [3]. in this case, the border between the two regions that the two
Many pattern recognition and computer vision applications sub-trees represent occurs. In this paper, we parallelize the
take advantage of the global features of the images for more merging procedure that is performed hierarchically.
accurate segmentation. However, in this way, the processing Efficient Graph-Based Image Segmentation algorithm,
time exceeds the allocated time for the input frame periods of which is proposed by Felzenszwalb and Huttenlocher, de-
the used camera in many real-time computer visions system. fines a merging criterion [10]. The algorithm employs the
Due to the computational complexity of the methods that Kruskal’s algorithm [12], which is one of the most popular
take into account the global features of the images, those MST algorithms. In this algorithm, all of the edges of
____________________________________________________________________________________________________________
260
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the graph structured an image is sorted in ascending order out the similarity, the one-dimensional gray-scale values or
according to their weight. Then, the edges are added to the multi-dimensional color values of the pixels can be used,
MST structure by starting the shortest edge, such that the and to compute the difference for both ways, the Euclidean
MST is empty at first. If the trees that are at the ends of the distance measurement can be used. The edge (u, v) connects
edge to be added do not satisfy the merging criterion, the the vertices u and v, and w(u, v) is the weight value of the
edge is not added. By the way, the MST may be divided edge. Let xu be the one-level gray-scale value, and xu,i be
into more than one sub-trees. Haxhimusa and Kropatsch the ith value of the z-level color vector of the vertex u where
[13] used this merging criterion on Boruvka’s hierarchical i = {1, 2, . . . , z}. The Euclidean distance as the weight value
MST algorithm [14]. In this algorithm, each node (vertex) of (u, v) is computed as in Eq. 1.
assumed a sub-tree, and each merging process of these sub- z
1
X
trees actualizes independently from others. For this reason, w(u, v) = ( (xu,i − xv,i )2 ) 2 (1)
this makes it suitable for parallelization. i=1
Nowadays, the multi-core processors are prevalent; and
A. Boruvka’s Minimum Spanning Tree Algorithm
therefore, the parallel programming has been a very popular
issue. Scientific and engineering programs such as image- Boruvka’s MST algorithm constructs an MST structure
processing applications need to be fast to overcome the heavy from a graph, such that the graph must be a weighted,
processing load in limited times. In order to make use of undirected, and complete graph (Fig. 1). According to the
this technology, parallelization of the applications reduces algorithm, each vertex is assumed as a tree at first stage
the process load. Obtaining the same or similar results with and no one is connected to any other one. The union of
sequential execution and ensuring to not overlapping of the multiple trees is called forest. At first, all of the vertices
parallel operations is mostly important and difficult matters compose the forest F0 , because each vertex is a tree such
for parallel programming issue. In this paper, we organize that F0 = T1 ∪ T2 ∪ . . . ∪ Tn , such that n refers to the
the hierarchical graph-based image segmentation algorithm number of the vertices on the graph G. Ti refers to the ith
as appropriate for parallel execution [15]. tree (Ti equals to the ith vertex at first stage) and each tree Ti
chooses the smallest edge (u, v) where u ∈ Ti and v ∈ / Ti ,
II. T HE H IERARCHICAL G RAPH -BASED I MAGE then the edge is added to the new forest F1 that is being
S EGMENTATION composed for the next stage. At the begin of the next stage,
Let the graph G = (V, E) be the image-graph that is the number of the trees will have been reduced by at least
constructed from an image. V represents the set of the one-half, i.e., t will have been decreased by at least one-half.
vertices in the graph G, such that each vertex refers a If any edge is not added in the current stage, on the other
pixel in the image and V contains all of the pixel-vertices. words, the last forest Flast has one tree, the algorithm ends.
E represents the set of the edges that connects each ver-
tex to its neighboring vertices. In this paper, 8-connected B. The Graph-Based Segmentation Algorithm
neighborhood is used as a neighborhood system. The edges In the MST of a graph representation of an image, the
are generally weighted and undirected. The weight of each highest weighted edges represent the boundaries of the image
edge is a numerical and positive value that indicates the (Fig. 1). Felzenszwalb and Huttenlocher have segmented the
degree of similarity between two vertices at the ends of images by detecting the edges of the boundaries, which
the edge. As the degree of similarity two vertices is lower, are called boundary edges [10], [11]. For the purpose of
the weight value is higher(Ters orantllar hocam). To figure to detect the boundary edges to be excluded from the
a. b. c.
Fig. 1. (a) A simple black and white image , (b) the graph representation of the image according to 4-connection neighborhood, (c) the MST of the
image extracted from its graph representation and an example of boundary edge in the MST
____________________________________________________________________________________________________________
261
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
MST, Felzenszwalb and Huttenlocher have used the merging of trees t is equal to the number of vertices |V | initially.
criterion Eq. 2 in the Kruskal’s MST algorithm. According For each tree Ti , the smallest edge (u, v), where u ∈ Ti and
to the Kruskal’s algorithm, each vertex in the graph G that is v ∈/ Ti , is found. If the weight value w(u, v) satisfies the
representation of the image seems as a tree. Firstly, all of the criterion in Eq. 2, the tree Tu to which the vertex u belongs
edges in G are sorted in order to their weight in ascending and the tree Tv to which the vertex v belongs are merged,
order. The two trees at the ends of the smallest weight edge and thus, the number of tree is reduced to one. After these
in the sorted edge queue are merged if the two trees are processes are complete for each tree, this cycle is repeated
not the same tree, and the edge is removed from the queue. iteratively. If no merging is happened, the algorithm ends.
At last, only one single tree remains such that it is called
III. T HE PARALLEL I MPLEMENTATION OF T HE
MST [12]. If the merging criterion in Eq. 2 is applied on the
A LGORITHM
algorithm, more than one MSTs might remain at the end of
this process [10]., because the edges that does not satisfy the A. OpenMP
criterion are not added to the MST to be formed, and thus, Multi-core processors have become quite prevalent nowa-
the MST that covers all of the vertices might not be created. days. Scientific and engineering programs such as image-
processing applications need to be fast to overcome the
Dif f (T1 , T2 ) < min(Int(T1 ) + k/|T1 |, Int(T2 ) + k/|T2 |) high processing load in limited times. In order to make
(2) use of the multi-core processor technology, such applications
In Eq. 2, T1 and T2 are the sub-trees to be merged. need to be parallelized. Therefore, some libraries and tools,
Dif f (T1 , T2 ) defines the weight value of the shortest which are developed to take advantage of the multi-core
weighted edge that connects the trees T1 and T2 . Int(T1 ) processor architecture, should be used and the programs
and Int(T2 ) defines the internal difference of T1 and T2 that will be parallelized must be coded suitable for paral-
respectively. An internal difference of a tree is the weight lelization. OpenMP (Open Multi-Processing) is a run-time
value of the highest weighted edge in the tree. |T1 | and program library that has routines and environment variables
|T2 | defines the number of vertices in the trees T1 and T2 and includes a set of compiler directives. The directives
respectively. The constant value k refers to a threshold value can be embedded into C/C++ and Fortran programming
that controls how much greater Dif f (T1 , T2 ) must be than languages using shared-memory platforms to run codes on
the internal differences of the trees T1 and T2 . more processors [17]. On the platforms like these, OpenMP
In this paper, Eq. 2 is utilized as a merging criterion in provides an easier programming model. However, getting
the Boruvka’s algorithm, because the Boruvka’s algorithm high performance from OpenMP sometimes might be a very
progresses in a hierarchical order. Due to this reason, this difficult task. The acceleration rate of the program that
algorithm is more suitable for parallelization than the other use OpenMP directives depends on many factors, including
popular MST algorithms [16]. The hierarchical graph-based the problem how the variables are used in the code, data
image segmentation algorithm is sketched in Fig. 2. layout, workload balancing and so on. OpenMP is especially
In Fig. 2, the constructed graph G = (V, E) is obtained useful for partitioning a loop and running the iterations of
from the image used, and then it is taken to the process as an the loop on more processors simultaneously. For that, the
input data. Firstly, each vertex seems as a tree. The number loop dependencies need to be configured in accordance to
parallelism [17].
B. Implementation of The Parallel Directives
The parallel computation is put into practice in three ways:
general-purpose computing on graphics processing units, via
a message processing interface, or using shared memory
architecture (OpenMP) [18]. In a shared memory model
system, threads, which are processes such that each runs
as parallel, are communicated each other by reading and
writing directly to the same memory. Since the OpenMP
specification was introduced as a standard shared-memory
programming model in 1998, there have been many studies to
Fig. 2. The steps of the hierarchical segmentation algorithm Fig. 3. A simple presentation of a shared memory architecture
____________________________________________________________________________________________________________
262
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
implement the OpenMP directives to their programs to run as expression omp unset lock by the processor that has set
parallel [19]. By using the OpenMP instructions, it is easy to the lock.
solve the high processing load problems of the classic single Another problem for synchronizing the processes is the
pipeline program model, because OpenMP execute some mutual exclusion problem. In cases when two clusters
calibrations automatically, and we do not need to consider chose each other as nearest cluster on different threads
which core should run in which part of the calculation [20]. simultaneously, a crash may happen. To solve this problem,
OpenMP uses a fork-join model, which is a standard way we get the related code into a critical block by the directive
to execute a parallel program on shared-memory systems critical. By this way, only one processor employs the codes
(Fig. 3). Firstly, a parallel program begins to run on a single that are in the critical block as seen below:
thread called master thread. Once the process flow encounters
a parallel block, the process flow is separated to several #pragma omp critical
threads (known as the fork) to run on additional processor {
cores [21]. omp set lock(mylock1);
OpenMP provides several mechanisms to run the code as omp set lock(mylock2);
parallel. Additionally, OpenMP provides an extra translator }
that analyzes the parallel loop and generate executor
code before the OpenMP translation [19]. The translator There is also another directive for getting the code into
determines the number of cores to run, separates the loop a critical block, which is called atomic such that it is used
into sub-loops automatically according to the number of the for only one variable to perform faster processing [17]. For
cores, and employs each sub-loops to different threats. The example:
example of its use is as below:
#pragma omp atomic
#pragma omp parallel f or a = a − 1;
f or(i = 0; i < 1000; i + +)
{
x[i] = y[i];
}
____________________________________________________________________________________________________________
263
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The parallel implemented version of the algorithm in Fig. In Fig. 5, the original versions of some images from
2 is sketched in Fig. 4. INRIA Holidays dataset [25], their segmentation results and
which processor has performed on which segments of the
IV. E XPERIMENTAL R ESULTS AND D ISCUSSION
images in last iteration are presented.
The system that has been used for the values in Table I has
32 bit Intel i5-2400 3.10-GHz Quad Core processor with 4 We applied the algorithm on some images from INRIA
GB RAM. We used C++ programming language. To import Holidays dataset [25] in the different versions of the images
and illustrate the images, we used OpenCV (Open Source as the sizes of 128×96, 256×192, 512×384, and 1024×768.
Computer Vision) library [24]. We executed the program five times for each elapsed time
Fig. 5. Visualization of the segmentation results and the task sharing of the processors for some images
____________________________________________________________________________________________________________
264
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I
T HE E LAPSED T IMES ( S ) OF S OME I MAGES FROM T HE DATASET WITH T HE D IFFERENT I MAGE S IZES AND T HE N UMBER OF C ORES
value in Table I, and put the smallest one of the five values [12] J. B. Kruskal, On the shortest spanning subtree of a graph and the
to the table. The time unit is given as seconds (s) in Table I. traveling salesman problem, Proc. Am. Math. Soc., vol. 7, no. 1, pp.
48−48, 1956.
[13] Y. Haxhimusa and W. Kropatsch, Segmentation graph hierarchies,
V. C ONCLUSIONS Lect. Notes Comput. Sci. 3138, pp. 343−351, 2004.
As seen in the results, parallel programming provides an [14] O. Borvka, O jistm problmu minimlnm, Prce Morav. prodovdeck
spolenosti, no. 3, pp. 37−58, 1926.
important acceleration in processing speed. However, the [15] A. Marongiu, P. Burgio, and L. Benini, Supporting OpenMP on a
processing time decreases no more in all cases, as the number multi-cluster embedded MPSoC, in Microprocessors and Microsys-
of cores increases. The speedup of the execution relies on tems, 2011, vol. 35, no. 8, pp. 668−682.
[16] M. Tepper, P. Mus, A. Almansa, and M. Mejail, Boruvka meets
many factors, including compiler optimizations, runtime sup- nearest neighbors, in Lecture Notes in Computer Science (including
port, data layout, operating system noise, workload balancing subseries Lecture Notes in Artificial Intelligence and Lecture Notes in
and so on. Additionally, it may be depend on the regions Bioinformatics), 2013, vol. 8259 LNCS, no. PART 2, pp. 560−567.
[17] L. Dagum, E. Rameshm, and R. Menon, OpenMP: An Industry-
to merge in the structure of the graph of the image. In Standard API for Shared- Memory Programming, Comput. Sci. & Eng.
a parallel loop, even if the threads complete their process IEEE, vol. 5, no. 1, pp. 46−55, 1998.
except one, they will have to wait for it. In these cases, [18] S. Zhang, Z. Xia, R. Yuan, and X. Jiang, Parallel computation of a
dam-break flow model using OpenMP on a multi-core computer, J.
the operating system might have executed another task on Hydrol., vol. 512, pp. 126−133, 2014.
the thread in the meantime. Nevertheless, the parallelization [19] W. Jeun, Y. Kee, and S. Ha, Improving performance of OpenMP for
saved a noticeable time for us. SMP clusters through overlapped page migrations, in International
Workshop on OpenMP (IWOMP06, 2006.
[20] B. Chapman, G. Jost, and R. Van Der Pas, Using OpenMP: Portable
R EFERENCES Shared Memory Parallel Programming, vol. 10. 2008.
[1] R. Nikhil and K. Scansar, A review on image segmentation techniques, [21] Vikas, N. Giacaman, and O. Sinnen, Multiprocessing with GUI-
Pattern Recognition, vol. 26, no. 9. pp. 1277−1294, 1993. awareness using OpenMP-like directives in Java, Parallel Comput.,
[2] B. Peng, L. Zhang, and D. Zhang, A survey of graph theoretical vol. 40, no. 2, pp. 69−89, 2014.
approaches to image segmentation, Pattern Recognit., vol. 46, no. 3, [22] R. Van Der Pas, An Introduction Into OpenMP, Eur. J. Surg., vol. 160,
pp. 1020−1038, 2013. no. 3, pp. 145−151, 2005.
[3] W. Tao, H. Jin, and Y. Zhang, Color image segmentation based on [23] E. Ayguad, N. Copty, A. Duran, J. Hoeflinger, Y. Lin, F. Massaioli, X.
mean shift and normalized cuts., IEEE Trans. Syst. Man. Cybern. B. Teruel, P. Unnikrishnan, and G. Zhang, The design of OpenMP tasks,
Cybern., vol. 37, no. 5, pp. 1382−1389, 2007. IEEE Trans. Parallel Distrib. Syst., vol. 20, no. 3, pp. 404−418, 2009.
[4] J. A. Bondy, Graph Theory With Applications. Oxford, UK, UK: [24] G. Bradski, The OpenCV Library, Dr Dobbs J. Softw. Tools, vol. 25,
Elsevier Science Ltd., 1976. pp. 120−125, 2000.
[5] C. T. Zahn, Graph-Theoretical Methods for Detecting and Describing [25] H. Jegou, M. Douze, and C. Schmid, Hamming embedding and weak
Gestalt Clusters, IEEE Trans. Comput., vol. C−20, no. 1, pp. 68−86, geometric consistency for large scale image search, in Lecture Notes
1971. in Computer Science (including subseries Lecture Notes in Artificial
[6] O. J. Morris, M. D. J. Lee, and A. G. Constantinides, Graph Theory Intelligence and Lecture Notes in Bioinformatics), 2008, vol. 5302
for Image Analysis: An Approach Based on The Shortest Spanning LNCS, no. PART 1, pp. 304−317.
Tree, Commun. Radar Signal Process. IEE Proc. F, vol. 133, no. 2,
pp. 146−152, 1986.
[7] Y. X. V. Olman, D. Xu, Solving data clustering problem as a
string search problem, in Proc. Conf. Stat. Data Mining Know. Dis.,
Chapman & Hall CRC, 2003, pp. 417−434.
[8] J. Shi and J. Malik, Normalized Cuts and Image Segmentation, IEEE
Trans. Pattern Anal. Mach. Intell., vol. 22, no. 8, pp. 888−905, 2000.
[9] A. Saglam and N. A. Baykan, Sequential image segmentation
based on minimum spanning tree representation, Pattern Recog-
nit. Lett., Available online 16 June 2016, ISSN 0167-8655,
http://dx.doi.org/10.1016/j.patrec.2016.06.001.
[10] P. F. Felzenszwalb and D. P. Huttenlocher, Efficient graph-based image
segmentation, Int. J. Comput. Vis., vol. 59, no. 2, pp. 167−181, 2004.
[11] A. Saglam, Minimum Yayilan Agac Tabanli Sirali Goruntu Bolutleme
(Minimum Spanning Tree-based Sequential Image Segmentation),
Master thesis, Selcuk University, Turkey, 2016.
____________________________________________________________________________________________________________
265
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Graph coloring problem (GCP) is getting more vertices. Therefore, these vertices are not adjacent vertices.
popular to solve the problem of coloring the adjacent regions in a However, the other vertices in the graph are adjacent because
map with minimum different number of colors. It is used to solve there is at least one edge between each other. In the context of
a variety of real-world problems like map coloring, timetabling this study we described vertex coloring problem in graphs. If
and scheduling. Graph coloring is associated with two types of
coloring as vertex and edge coloring. The goal of the both types
an undirected 𝐺 = (𝑉, 𝐸 ) graph is examining; 𝑉 is the set of
of coloring is to color the whole graph without conflicts. vertices and 𝐸 is the set of edges. The graph which is given in
Therefore, adjacent vertices or adjacent edges must be colored Fig. 1, the set of vertices are 𝑉 = {𝑣1 , 𝑣2 , 𝑣3 , 𝑣4 } and the set
with different colors. The number of the least possible colors to of edges are 𝐸 = {𝑒1 , 𝑒2 , 𝑒3 , 𝑒4 , 𝑒5 } . In addition, R =
be used for GCP is called chromatic number. As the number of {1,2, … … … , 𝑘} is the set of colors which are used for coloring
vertices or edges in a graph increases, the complexity of the the vertices. In this case, if the whole graph colored without
problem also increases. Because of this, each algorithm can not conflicts is performed by utilizing the minimum number of
find the chromatic number of the problems and may also be different colors, it’s called “k-coloring graph” [4]. This
different in their executing times. Due to these constructions,
minimum number of different colors is known as chromatic
GCP is known an NP-hard problem. Various heuristic and
metaheuristic methods have been developed in order to solve the number. Chromatic number is indicated by χ(G) [5,6].
GCP. In this study, we described First Fit (FF), Largest Degree
Ordering (LDO), Welsh and Powell (WP), Incidence Degree
Ordering (IDO), Degrees of Saturation (DSATUR) and
Recursive Largest First (RLF) algorithms which have been
proposed in the literature for the vertex coloring problem and
these algorithms were tested on benchmark graphs provided by
DIMACS. The performances of the algorithms were compared as
their solution qualities and executing times. Experimental results
show that while RLF and DSATUR algorithms are sufficient for Figure 1. A graph with four vertex and five edge
the GCP, FF algorithm is generally deficient. WP algorithm finds
out the best solution in the shortest time on Register Allocation,
CAR, Mycielski, Stanford Miles, Book and Game graphs. On the
Graph coloring problem is mostly used for solving
other hand, RLF algorithm is quite better than the other computer based applications and problems. Graph coloring
algorithms on Leighton, Flat, Random (DSJC) and Stanford algorithms are usable for solving the many engineering
Queen graphs. applications and real-world problems [2]. Some of the these
problems are Map Coloring [7], Timetabling and Scheduling
Keywords — Chromatic number, Graph coloring algorithms. problems [8,9], Register Allocation problems [10,11], Sudoku
problem [12] and Frequency Assignment problems [13].
I. INTRODUCTION As the number of vertices or edges in a graph increases, the
Graph theory is a problem represented with vertices (nodes) complexity of the problem also increases. Because of this,
and edges (arcs) [1]. Otherwise, graph coloring problem each algorithm can not find the chromatic number of the
(GCP) is a problem where adjacent vertices or edges in graph problems and may also be different in their executing times.
must be colored by using different colors [2]. GCP was Due to these conditions, GCP is known an NP-hard problem
proposed by Francis Gutrie as the four color problem. Four [14]. Hence, for getting a better solution for GCP many
color problem has described by F. Gutrie to solve the problem hueristic and meta-huestistic algorithms are developed.
of coloring the adjacent regions in a map using the minimum Huerictic algorithms generally can be used for a problem with
number of diffrent colors [3]. fewer numbers of vertices. On the other hand, for the complex
Graph coloring is associated with two types of coloring as graphs meta-hueristic algorithms can find better solutions [15].
vertex and edge coloring [2]. The goal of the both types of Tabu Search (TS) Algorithm [16], Simulated Annealing
coloring is to color the whole graph without conflicts. (SA) Algorithm [17], Genetic Algorithm (GA) [7], Ant
Therefore, adjacent vertices or edges must be colored with Colony (ACO) Algorithm [18], Cuckoo (COA) Algorithm
different colors. If there is at least one link (edge) between [15] are some of the meta heueristic algorithms used for graph
two nodes, it is called adjacent vertices. If the Fig. 1 examines, coloring problem. When the vertices in a graph 𝐺 are colored
it can be seen that there is no edge between V2 and V4 by means of the greedy algorithms, the coloring issue is
____________________________________________________________________________________________________________
266
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III
THE RESULT OF THE COLORED GRAPH USING THE FF ALGORITHM
____________________________________________________________________________________________________________
267
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
B. Welsh Powell Algorithm (WP) Table V shows the result for the graph shown in Fig. 2.
For the given graph 𝐺, the set of the vertices is described as Selected order of the vertices and their colors are given in
𝑉 = {𝑣1 , 𝑣2 , … . . , 𝑣𝑛 } and the set of the colors of the vertices is Table V.
described as 𝑅 = {𝑟1 , 𝑟2 , … , 𝑟𝑘 } . The steps of the WP TABLE V
algorithm: THE RESULT OF THE COLORED GRAPH USING THE LDO ALGORITHM
C. Largest Degree Ordering Algorithm (LDO) Table VI shows the result for the graph shown in Fig. 2.
Selected order of the vertices and their colors are given in
For the given graph 𝐺, the set of the vertices is described as
Table VI.
𝑉 = {𝑣1 , 𝑣2 , … . . , 𝑣𝑛 } and the set of the colors of the vertices is
described as 𝑅 = {𝑟1 , 𝑟2 , … , 𝑟𝑘 } . The steps of the LDO TABLE VI
algorithm: THE RESULT OF THE COLORED GRAPH USING THE IDO ALGORITHM
____________________________________________________________________________________________________________
268
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Step 2: The uncolored vertex that has the largest degree in the the number adjacent vertices which are in the set of U
degree set 𝐷𝑒𝑔(𝑣𝑖 ) is selected for coloring. The selected for every vertex in set of V′. After that, the uncolored
vertex is colored with first color. vertex whose has maximum adjacent vertices (which
Step 3: Firstly, calculate the number adjacent vertices which are in the set of U ) in the set of 𝑉′ is selected for
are colored with different colors for every uncolored vertex. coloring. The selected vertex is colored with active
After that, the uncolored vertex whose number of adjacent color.
vertices colored with different colors is the maximum is The colored vertex and the adjacent vertices of the
selected for coloring. If more than one vertex provide this colored vertex are deleted from 𝑉 ′ and added to the
condition, the vertex which has the largest degree among them set of U.
is selected. If the set of V′ is not empty, it is returned to the step 2.
Step 4: Firstly, the selected vertex is tried to color with the Otherwise move to step 3.
colors in the color set. If the colors in the color set are not
appropriate to color the vertex, a new color is defined. The new Step 3: If the uncolored vertex exists, next color in the color
color is added to the color set and appointed to the selected set is selected as active color. Otherwise the program is
vertex. terminated.
Step 5: If the uncolored vertex exists, it is returned to the step Step 4: Calculate the number adjacent vertices for every
3. Otherwise the program is terminated. uncolored vertex. After that, the uncolored vertex whose has
maximum adjacent vertices is selected for coloring process.
For DSATUR algorithm; Table VII shows the result for the If more than one vertex provide this condition, the vertex
graph shown in Fig. 2. which has the largest degree among them is selected. Then, it
is returned to the step 2.
TABLE VII
THE RESULT OF THE COLORED GRAPH USING THE DSATUR ALGORITHM
For RLF algorithm; Table VIII shows the result for the
A B C D E F G graph shown in Fig. 2. Selected order of the vertices and their
Selected order 4 3 1 5 2 6 7
colors are given in Table VIII.
A B C D E F G
F. Recursive Largest First Algorithm (RLF)
Selected order 4 7 1 5 6 2 3
RLF is used a recursive structure for coloring the vertices
in graph. This recursive structure is the most important feature Vertex color 𝑟2 𝑟3 𝑟1 𝑟2 𝑟2 𝑟1 𝑟1
of the RLF algorithm [20]. According to this recursive
structure, whole graph is colored with minimum different
colors. For the given G graph, the set of the vertices is IV. EXPERIMENT RESULT
described as 𝑉 = {𝑣1 , 𝑣2 , … . . , 𝑣𝑛 } and the set of the colors is FF, LDO, WP, IDO, DSATUR and RLF algorithms were
described as 𝑅 = {𝑟1 , 𝑟2 , … , 𝑟𝑘 } . The steps of the RLF tested on benchmark graphs provided by DIMACS [22]. The
algorithm: reason of preferring the DIMACS graph, it’s given a standard
for performance comparison of the algorithms. The
Step 1: Vertex degree is calculated for each vertex and the performances of the algorithms were compared as their
degrees of vertices added to the set of 𝐷𝑒𝑔(𝑣𝑖 ). Initially, the solution quality and executing times. The edge density (D) of
first color in the color set is selected as the active color. Select the benchmark graphs which are used in this study are
the uncolored vertex which has the largest degree from set of calculated from equation 2. E represents the number of the
𝐷𝑒𝑔(𝑣𝑖 ) for coloring. edges and V represents the vertices number of the graph [23].
Step 2: The selected vertex is colored with active color.
Adjacent vertices of the selected vertex can not color with 2∗𝐸
active color. But the uncolored vertices which are not adjacent 𝐷= ( 2)
V ∗ (V − 1)
vertices of the colored vertex can be colored with active color.
So RLF uses a recursive structure for select the uncolored In this study we used Mycielski, CAR, Stanford Graph Base
vertices to color with active color. During this process the (SGB), Register Allocation, Leighton, Flat, Random (DSJC)
below steps should be followed: and Random geometric (DSJR and R250) DIMACS graphs. V
Adjacent vertices of the selected vertex vi are found represents the number of the vertices, E is the number of the
from adjacency matrix. Adjacent vertices are added edges, Den. represents density, Best/χ(G) means chromatic
to the adjacent set U. ( 𝑈 = {𝑢1 , 𝑢2 , … … . , 𝑢𝑡 } ) number or the best known number, R represents the number of
The vertices which are not adjacent vertices of the colors that algorithms are found, T is computation time in
selected vertex vi are found from adjacency matrix. seconds. The algorithms are written in the programming
These vertices are added to the set of 𝑉′ . Calculate language Matlab R2010a. For experiments we used a Laptop
____________________________________________________________________________________________________________
269
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
computer. It has Intel Core i5 2.20 GHz processor and 8 GB (anna), Homer’s Iliad (homer) and Mark Twain's Huckleberry
DDR3 RAM. Finn (huck). For miles graphs the vertices represents some of
Mycielski graphs are triangle free graphs. It’s mean that the the United States cities and if there is a road between two cities
edge connections in the graph must be free of triangle. For which provides the conditions, there is an edge generated
mycielski graph, if the vertices in the graph increases, the between them. For game graph, any vertex in the graph
number of the colors for coloring the graph increases too [24]. represents a college team. There is an edge generated between
Stanford GraphBase (SGB) graphs are created from Donald for every two teams when the teams played to each other
Knuth. SGB graph can be divided to books, miles, game and during the season. Table IX shows the results for algorithms
queen graphs [22]. For books graphs, a character in the book which are used for this study. Except the FF algorithm all other
represents a vertex. So the books graphs are created for holds algorithms find out the best/χ(G) results. In addition, FF
to relationship between characters. If the characters in the book algorithm finds out the best/χ(G) results for graphs which are
have relationship to each other, an edge is generated between used for this study except miles graphs and anna, David, homer
two vertex which characters run across in the book. These graph. On the other hand, if the algorithms compared to each
books are Charles Dicken's David Copperfield (david), Victor other about their computation times, WP algorithm is reached
Hugo's Les Misêrables (jean), Lev Tolstoy's Anna Karenina these conclusions in a quite short time.
TABLE IX
THE RESULTS AND COMPUTATION TIMES FOR MYCIELSKI AND SGB GRAPHS
Queen graphs are 𝑛𝑥𝑛 dimensional chessboard graphs. If graphs. Experimental results show that while RLF and
two queens on the chessboard are in the same row, column, or DSATUR algorithms are sufficient for the queens graphs, but
diagonal, there is an edge generated between them. So, if two the other algorithms are generally deficient. RLF algorithm
queens placed in same row, column or diagonal, one queen can finds out just only the chromatic number of queen5_5 graph
eat the other one. Because of this, there is an edge between and also it finds out quite better results for the other queen
them for they don’t eat each other. For queens graph if only the graphs. DSATUR algorithm generally finds out good results,
graph is colored with minimum number n, two queens can but it is very slow according to the RLF.
move on chessboard. Table X shows the results for queens
TABLE X
THE RESULTS AND COMPUTATION TIMES FOR QUEEN GRAPHS
____________________________________________________________________________________________________________
270
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
queen10_10 100 2940 0,59 11 13 0,0215 14 0,4351 17 0,0009 17 0,0052 17 0,1876 16 0,0044
queen11_11 121 3960 0,54 11 14 0,0345 15 0,6300 17 0,0009 17 0,0072 18 0,2964 17 0,0063
queen12_12 144 5192 0,50 13 15 0,0550 16 0,9163 19 0,0010 19 0,0100 20 0,4604 20 0,0092
queen13_13 169 6656 0,47 13 16 0,0800 17 1,3226 23 0,0013 23 0,0134 22 0,6869 21 0,0125
queen14_14 196 8372 0,44 16 17 0,1227 19 1,8408 25 0,0015 25 0,0170 24 1,0488 23 0,0169
R: Result of the algorithm, T: Computation time (in second)
CAR graphs are created from inspiration of the mycielski best/χ(G) results. Furthermore the IDO algorithm generally
graphs. After the some new vertices inserted graph, the graph finds out the best/χ(G) results. But FF algorithm is generally
size increases, but the density of the graph is unchanging [25]. deficient. If the algorithms compare to each other about their
The CAR graphs are more difficult than the mycielski graphs. computation times, the best algorithm for CAR graphs is WP
Table XI shows the results for CAR graphs. According to the algorithm. Because WP algorithm reaches the best/χ(G) results
Table XI; RLF, DSATUR, WP and LDO algorithms reach the shortest computation times.
TABLE XI
THE RESULTS AND COMPUTATION TIMES FOR CAR GRAPHS
Random (DSJ) graphs which are created from David algorithms generally find out the best/χ(G) results. For the
Johnson and R250_5 graph are difficult to solve benchmark R250_5 graph, DSATUR algorithm’s computation time is
graphs [15]. Flat graphs are created from Culberson. [26]. further than RLF’s. But DSATUR finds out quite better result
First parameter of the Flat graphs represents the number of the than RLF for R250_5 graph. On the other hand, RLF finds
vertices and the second parameter represents the chromatic out quite better results for the other graphs and RLF is faster
number. Table XII shows the results for Random and Flat than the other algorithms.
graphs. According to the Table XII, RLF and DSATUR
TABLE XII
THE RESULTS AND COMPUTATION TIMES FOR RANDOM AND FLAT GRAPHS
____________________________________________________________________________________________________________
271
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
DSJC250_1 250 3218 0,10 8 10 0,0665 10 0,4791 11 0,0011 11 0,0142 12 0,2086 13 0,0097
DSJC250_5 250 15668 0,50 28 35 0,4661 37 4,9399 41 0,0025 41 0,0371 40 3,0145 43 0,0394
DSJR500_1 500 3555 0,03 12 12 0,2863 13 0,5829 13 0,0023 13 0,0237 13 0,2609 15 0,0199
R250_5 250 14849 0,48 65 71 0,7803 68 4,6108 70 0,0034 70 0,0439 69 2,7790 79 0,0478
flat300_20 300 21375 0,48 20 38 0,7199 42 8,5125 44 0,0032 44 0,0554 45 5,3253 47 0,0609
flat300_26 300 21633 0,48 26 39 0,8435 41 8,5990 45 0,0030 45 0,0566 48 5,5501 45 0,0610
flat300_28 300 21695 0,48 28 38 0,8624 42 8,6611 45 0,0030 45 0,0564 48 5,5324 46 0,0613
R: Result of the algorithm, T: Computation time (in second)
fpsol2*, inithx*, zeroin* and mulsol* graphs are computer All algorithms which are used for this study reach the χ(G)
register allocation problem graphs which are generated from results for computer register allocation graphs. If the
Gary Lewandowski [24]. These graphs are real-world algorithms compare to each other about their computation
problem’s graphs. The computer registers and the operations times, the best algorithm for register allocation graphs is WP
are defined as vertices. If a register and an operation have a algorithm and also FF algorithm reaches the χ(G) results a
relationship, there is an edge generated between them. Table quite short times. The slowest algorithm for these graphs is
XIII shows the results for computer register allocation graphs. DSATUR algorithm.
TABLE XIII
THE RESULTS AND COMPUTATION TIMES FOR REGISTER ALLOCATION GRAPHS
In the Leighton graphs, each graph consists of 450 vertices. algorithm finds out quite better results for Leighton graphs.
First parameter of the Leighton graphs represents the number Just for le450_25b graph, WP algorithm finds out the χ(G)
of the vertices and the second parameter represents the result the better computation time. The other algorithms are
chromatic number [20]. Table XIV shows the results for generally deficient.
Leighton graphs. Experimental results show that RLF
TABLE XIV
THE RESULTS AND COMPUTATION TIMES FOR LEIGHTON GRAPHS
____________________________________________________________________________________________________________
272
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. CONCLUSION
Experimental results show that while RLF and DSATUR [15] S. Mahmoudi, S. Lotfi, Modified cuckoo optimization algorithm
algorithms are sufficient for the GCP, FF algorithm is (MCOA) to solve graph coloring problem, Applied soft computing
generally deficient. WP algorithm finds out the best solution in Volume 33, 48–64, 2015.
the shortest time on Register Allocation, CAR, Mycielski, [16] A. Hertz, D. de. Werra, Using tabu search techniques for graph
Stanford Miles, Book and Game graphs. On the other hand, coloring, Computing 39, 345–351, 1987.
RLF algorithm is quite better than the other algorithms on [17] M. Chams, A. Hertz and D. de. Werra, Some experiments with
Leighton, Flat, Random (DSJC) and Stanford Queen graphs. simulated annealing for coloring graphs, European Journal of
As shown in the study, firstly it should be decided that the Operational Research, 32(2):260-266, 1987.
[18] S. H. Ahn, S. G. Lee, T. C. Chung, Modified Ant Colony System for
problems which we want solve with graph coloring algorithms
Coloring Graphs, Information, Communications and Signal
is similar to what benchmark graphs. After that, the optimum
Processing, 2003 and Fourth Pacific Rim Conference on Multimedia.
graph coloring algorithms must be applied to the problem for
Proceedings of the Joint Conference of the Fourth International
finds out the the best solution. Thus, it can be avoided to waste
Conference on, IEEE, 1849 – 1853, 2003.
of times and it can be reached the best results a quite short [19] H. A. Omari and K.E. Sabri,New Graph Coloring Algorithms,
time. American Journal of Mathematics and Statistics 2 (4): 739-741,
2006.
REFERENCES
[20] F.T. Leighton, A graph coloring algorithm for large scheduling
[1] D. B. West, Introduction to Graph Theory, Prentice Hall, U. S. A., problems, J. Res.Nat. Bur. Stand. 84 489–506, 1979.
588 pp, 2001. [21] D. Brélaz, New methods to color the vertices of a graph, Commun.
[2] J. L. Gross And J. Yellen, Graph Theory and Its Applications, CRC ACM 22 251–256, 1979.
Press, Mathematics, 600 pages, 1998. [22] DIMACS graph coloring instances. (2016) instances homepage on
[3] R. Fritsch, and G. Fritsch, The Four-Color Theorem: History, CMU. [online]. Available:
Topological Foundations and Idea of Proof, Newyork: Springer, http:mat.gsia.cmu.edu/COLOR/instances.html.
pages 260, 1998. [23] I. C. R. Ruiz, “Gravitational Swarm for Graph Coloring”, PHD
[4] F. Ge, Z. Wei, Y. Tian, Z. Huang, Chaotic Ant Swarm for Graph thesis, The University of the Basque Country Donostia - San
Coloring , Intelligent Computing and Intelligent Systems (ICIS), Sebastian, 2012.
IEEE International Conference, 512-516 p., 2010. [24] D. S. Johnson and M. A. Trick, Cliques, Coloring, and Satisfiability:
[5] D. J. A. Welsh, and M. B. Powell, An upper bound for the chromatic Second DIMACS Implementation Challenge, 1993, vol. 26 of
number of a graph and its application to timetabling problems, The DIMACS Series in Discrete Mathematics and Theoretical Computer
Computer Journal 10 (1): 85–86, 1967. Science, American Mathematical Society. 1996.
[6] I. M. Díaz and P. Zabala, A Generalization of the Graph Coloring [25] M. Caramia and P. Dell'Olmo. A lower bound on the chromatic
Problem, Departamento de Computacion, Universidad de Buenes number of mycielski graphs. Discrete Mathematics, 235(1-3):79_86,
Aires, 1999. 2001.
[7] B. H. Gwee, M. H. Limand and J. S. Ho, Solving fourcolouring map [26] B. Yılmaz, A novel meta-heuristic for graph coloring problem:
problem using genetic algorithm. In Proceedings of First New simulated annealing with backtracking, Yeditepe Üniversitesi Fen
Zealand International Two-Stream Conference on Artificial Neural Bilimleri Enstitüsü, Istanbul, 2011.
Networks and Expert Systems, 332-333, 1993.
[8] N. Chmait, and K. Challita, Using Simulated Annealing and Ant-
Colony Optimization Algorithms to Solve the Scheduling Problem,
Computer Science and Information Technology 1(3), 208-224, 2013.
[9] K. Dowsland, J. Thompson, Ant colony optimization for the
examination scheduling problem, J. Oper. Res. Soc. 56, 426–438,
2005.
[10] G.J. Chaitin, M.A. Auslander, A.K. Chandra, J. Cocke, , M.E.
Hopkins, P.W. Mark-stein, Register allocation via coloring, Comput.
Lang. 6 47–57, 1981.
[11] F.C. Chow, J.L. Hennessy, The priority based coloring approach to
register allocation, ACM Trans. Program. Lang. Syst, 501–536, 1990.
[12] S. Ono, R. Miyamoto, S. Nakayama, and K. Mizuno, "Difficulty
estimation of number place puzzle and its problem generation
support." ICCAS-SICE, 2009. IEEE, 2009.
[13] W. K. Hale, Frequency assignment: Theory and applications.
Proceedings of the IEEE 12: 1497-1514, 1980.
[14] M.R. Garey, D.S. Johnson, Computers and intractability, in: A Guide
to the Theoryof NP-Completeness, W.H. Freeman & Co., New York,
NY, U.S.A., 1979.
____________________________________________________________________________________________________________
273
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
274
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I
GENETIC ALGORITHM FLOWCHART, STEPS AND DESCRIPTION
4. Substitution: It is used to
Step 4. A constant population is provided by
operate new population
removing the old chromosomes.
algorithm again.
Y
5. Situation: If the new
population is enough, it stops.
This is the best solution found Step 5. The success of the new population is
ever because of saving the best found by calculating all the chromosomes.
individuals during the
calculating population.
6. Loop (circuit): Turned back to Step 6. Many populations are generated and
step 2. calculated by operating Genetic algorithm
many times.
Common Applications in Business different problems. The steps for solutions of Genetic
Finance Algorithms and description of them are given in Table I [9].
Marketing
Production
Flow of Genetic Algorithm is quite easy. Many parameters
and adjustments are accomplished in different ways for
____________________________________________________________________________________________________________
275
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III
FUNCTIONS USED IN THE SIMULATION
Name of
Description
Function
Rosenbrock ∗( − ) +(1- )
Function 1 − + ∗ + ∗ ∗ + )
Function 2 ( + ∗ )+ + ∗
Fig. 1 The first working appearance
Function 3 ∗ ( ∗ )+ ∗ ( ∗ )
Function 4 ( − ) + ( − ∗ )
Function 5 ∗ +
Function 6 ∗ + ∗ − ∗ + ∗ +19
Function 7 ∗( − . ) + ∗( − . )
Function 8 ∗ + ∗ + ∗ ∗ +2
Function 9 ∗ + ∗ ∗ +
Function 10 + − ∗ − ∗ ∗ − ∗
Function 11 + ( − ) + ( − )
∗ + * ∗ − − ∗ ∗
Himmelblau
+
Function 12 ∗ + ∗ ∗ +5*
. + . ∗ +((0.1+0.1* )/ )+
Function 12
((0.1∗ .∗ + )/( ∗ ) Fig. 2 Main screen of programme
In calculation with genetic algorithm, the number of optimization was occurred (Fig. 3).In the second simulation
variables, lower and upper limits of root intervals should be
entered by keyboard. Also, parameters of genetic algoritm can y= x1.*exp(-x1.^2- x2.^2) (2)
be changed and then applied. The changable parameters of
genetic algorithm are size of population, scaling the suitability roots of an exponentinal function were optimized (Fig. 3).
(fitness) function (rank, proportional, hill, linear changee), These functions have two different roots. Research space was
selection function (Stochastic uniform, increasing, uniform, formed between [1, 10] interval to find the roots.
____________________________________________________________________________________________________________
276
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
Parameters of genetic algorithms were adjusted as below: [1] Ikotun Abiodun M, Lawal Olawale N. and Adelokun Adebowale
P. ” The Effectiveness of Genetic Algorithm in Solving
Simultaneous Equations” , International Journal of Computer
Population size: 20
Applications 14(8):38–41, Feb. 2011
Scaling suitability functions: Rank [2] Harsh Bhasin, Surbhi Bhatia, “Use of Genetic Algorithms for
Finding Roots ofAlgebraic Equations”, International Journal of
Selection function: Stochastic uniform Computer Science and Information Technologies, Vol. 2 (4) , 1693-
Mutation function: Gaussian 1696, May. 2011
[3] Punam S M. “Genetic Algorithm For Linear And Nonlinear
Crossing-over function: Distributed Equation”, International Journal of Advanced Engineering
Crossing-over rate: 0.8 Technology , Vol.III, Issue II, 114-118, April-June, 2012
[4] Nayak T., Dash T. “ Solution to Quadratic Equation Using Genetic
Algorithm”, Proceedings of National Conference on AIRES-2012,
The change in suitability function and roots can easily be Andhra University
detected in Simulator by changing parameters of genetic [5] Vatansever F., Batık Z., “Equation Solutions Based On Genetic
algorithm. The effect of the change in crossing-over rates on Algorithm”, International Advanced Technologies Symposium
suitability value can be seen in Fig. 3. To obtain the values (IATS’09), Karabük, Turkey, May 2009
that would give the best result, the crossing-over rates were [6] Jebari K., Madiafi M., Moujahid A.,” Solving Poisson Equation by
Genetic Algorithms”, International Journal of Computer
changed and its effects were investigated during the iteration. Applications (0975 – 8887) Volume 83 – No 5, Dec. 2013
The best individuals are in the interval of 0.7 ve 0.9 of [7] Cormen T., Leiserson C., Rivest R., Stein C., Introduction to
crossing-over rate. Algorithms, MIT Publisher, 1332, 2009
[8] MELANİE, M., An Introduction to GeneticAlgorithms, MIT
Publisher, 162, 1998
[9] ŞEN, Z., Genetik Algoritmalar ve En İyileme Yöntemleri, Su Vakfı
Publisher, 142, 2004
[10] Matlab Com Builder User’s Guide, The Mathworks Inc., Oct. 2008
IV. CONCLUSIONS
____________________________________________________________________________________________________________
277
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— This paper proposes an algorithm to solve the The GA was introduced by Holland [9] in the early 1970s.
problem of path planning for a mobile robot in a static It imitates the evolutionary processes such as inheritance,
environment with obstacles. The proposed algorithm is a Hybrid mutation, selection and crossover. Also for centuries, there is
Genetic algorithm (HGA) which includes Genetic and Dijkstra a biological evolution in the world and GA simulates it by
algorithms together. The genetic algorithm (GA) is preferred
using iterative process. The GA has a wide using area in many
since the structure of robot path planning problem is very
convenient to apply genetic algorithm’s coding and operators types of problem thanks to easy adaptation and high high-
such as permutation coding, crossover and mutation. GA performance computing.
provides diversification while searching possible global solutions,
but Dijkstra algorithm makes more and more intensification in In this paper, a Hybrid Genetic algorithm (HGA) is
local solutions. The simulation results show that the mobile robot proposed to solve RYP problem. The aim is to combine the
can plan a set of optimized path with an efficient algorithm. global search capability of genetic algorithm and local exact
solution capability of the Dijkstra algorithm to get faster
Keywords— Robot path planning, genetic algorithm, Dijkstra solution for shortest path problem. In numerical simulations,
algorithm. 20x20 and 50x50 environments with obstacles are solved to
get shortest distances and, Dijkstra, genetic and proposed
I. INTRODUCTION hybrid genetic algorithms are compared in Table for iteration
Robot path planning (RPP) is a popular issue in mobile number and required time, respectively.
robotics. An intelligent robot must be able to move by itself
from a start location to a target location without collision with The paper is organized as follows. Section 2 describes the
obstacles. Therefore an important research subject in this field method of grid modeling, Dijkstra algorithm and genetic
is navigation of autonomous mobile robots, which is to find a algorithm. Section 3 describes the proposed hybrid genetic
globally optimal path from a starting point to a target in a algorithm for the robot path planning problem. Section 4
given environment and at the same time avoid collisions. presents the numerical simulations on robot path planning.
Moreover, the ”optimal” here means that the path must satisfy Section 5 presents concluding remarks.
some criterions like length of the path is the shortest, or
energy consumption of robot is the lowest or needed time to II. PRELIMINARY WORKS
reach a target is minimum etc.
A. Grid modeling
There are several search algorithms for the shortest path The method was proposed by the Howden in 1968, and its
problem such as the A* algorithm, the Dijkstra algorithm and primary task is to construct a path grid map according to the
the Bellman–Ford algorithm [1], [2], [3]. These algorithms environment. Given environment is divided into small units
ensure to find a shortest path. However, these algorithms are and a robot goes one unit with one move. Two kind of grid
not effective for large dimensioned problems because of with can be described in method which are free and obstacle grid.
low searching efficiency and high computational time. In black grid’s mean is obstacle, white grid’s mean is free
namely movable area. In real environment sizes of obstacles
In recent years, many researchers have studied the robot can be different and they can occupy lower one or more grids.
path planning on various intelligent methods, such as particle A robot cannot enter an obstacle even if the obstacle is very
swarm algorithms [4], immune algorithm [5], neural networks small.
and ant colony algorithm [6], differential evolution [7], and so
on. Additionally, genetic algorithm is one of the popular The grid can be divided into intermediate grid and
metaheuristic algorithm for RPP [8] problem. boundary grid. For intermediate grid, robot may have eight
directions for the next motion. Such as up (7), down (2), left
____________________________________________________________________________________________________________
278
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(4), right (5), right-up (8), right-down (3), left-up (6) and left- Select: Select operation can select adaptable individual in
down (1). Fig. 1 shows the motion direction of the robot and current population, so that adaptable individual has the
Fig. 2 shows a 10x10 grid map. And for boundary grid, it has opportunity to breed the next generation as a parent.
to subtract inaccessible directions. Robot must avoid obstacles Crossover: Crossover operation can produce new
to select an optimization motion path moving to the target individual, the new individual combines the characteristics of
position [10]. parents.
Mutation: mutation operation can change one or a number
of genes in a fixed probability. The purpose of mutation is to
create new chromosomes for next generation.
The process of genetic algorithm: determine the encoding
rules, generate a population randomly, calculate fitness
function and selection probability, select, crossover, mutation,
loop all the operation above until it meet the terminate
conditions [11].
Fig. 1 The motion direction of the robot.
III. PROPOSED HYBRID GENETIC ALGORITHM
In this section, proposed algorithm steps are explained in
detail.
A. Problem Description
A road from starting node (S) to target node T is an (S, T)
path, the length of a path is the sum of the length of the edges
Fig. 2 10x10 grid map
(|pipi+1|) in one path. Ro bot path problem is to find the path
from S to T, where the length of the path is minimum one at
the same time avoid obstacles. A candidate path for robot path
planning is denoted by
B. Dijkstra Algorithm P= { }
Dijkstra algorithm is introduced by W. Dijkstra in 1959[1]. where pi denotes the i th node of the planned path P, n
It is one of the most common used algorithm for solving the denotes the number of without obstacle nodes in a given
shortest path problem. It is possible to find a shortest path environment and p1=S and pn=T.
from starting point to any node by Dijkstra Algorithm. If the Total length of the path can be obtained by equation (1):
performance criterion is different from distance, it is also
useful for measuring other criteria such as time, cost and
energy. Dijkstra algorithm guarantees the shortest path which ( ) ∑| | ( )
is generated. Dijkstra algorithm is a useful method for not
only robot path planning but also network optimization,
transportation, logistics, electronics and other fields. However, where | | is the length of the path segment
it takes much time while solving large problems and this is a which is computed by Euclidean distance and it can
disadvantage for this method. take value √ or 1. Additionally ( ) is objective function of
given path which we try to minimize.
C. Genetic Algorithm
Genetic algorithm combines genetics and computer science. B. Chromosome encoding and initial population generation
GA is biology inspired, population based, has memory,
stochastic and greedy algorithm. For a given graph, each node will be coded as a gene sorted
Here are a few basic concepts of genetic algorithm: by the node number. A chromosome includes path
Chromosome and Coding: In order to use genetic algorithm, information and sequence of genes. In this paper, permutation
feasible solution of the problem needs to be coded into symbol coding is used for encoding. Nodes take place as
string which have fixed structure. The string is called combinatorics in a solution if a node is visited as an example
chromosome. Each bit of the string represents a gene. shown below:
Population: the total number of chromosomes in each [1→9→16→17→…→71→79]
generation is called population. A population contains a set of
feasible solutions in current generation. The first and end nodes must be in the solution, indicating
Fitness: Each chromosome corresponds to a feasible the starting node S and target node T. There is no need to the
solution, and each solution corresponds to a function value. test chromosome if the path is connected. Producing initial
This function value is used to measure the environmental solution strategy guarantees that all the nodes which are
adaptability of a chromosome. included solution are connected. In other word, length of a
____________________________________________________________________________________________________________
279
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
chromosome is equal to number of passed nodes. Therefore, tournament selection for reproduction. In this work M is
there is a remarkable decreasing usage of memory for a assumed 2 for the tournament selection.
chromosome and computation time. Cross-over and mutation Crossover: For the given chromosome x1, x2 which are
operators are modified because of different length of parent, their intersected nodes (intersect(x1,x2)) are found
chromosome which are shown next subsections. except for starting and target nodes (1,79) and one of the
While generating initial solution, a node is selected (initial found nodes is selected randomly. This node is called as
gene of solution is starting node) and other neighbors are crossover point and single point crossover is used. A
found which are connected to the selected node. Randomly crossover is done at the selected node.
selected a neighbor node is added to solution. This procedure
is repeated until reached target nodes. Additionally, For example:
generating strategy checks if a node is visited before. Table 1 x1 = 1,2,9,17,18,19,27,36,44,52,61,68,76,77,78,79
shows generating a chromosome for given map in Fig. 3. x2 = 1,9,16,25,33,41,42,43,44,53,54,62,71,79
TABLE I
GENERATING A CHROMOSOME (intersect(x1,x2)) = 9,44 (except for 1,79)
[1,2,9,17,18,19,27,36,44,53,54,62,71,79]
27→19,20,26,28,34,35,36
44→34,35,36,43,52,53
____________________________________________________________________________________________________________
280
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. NUMERICAL SIMULATIONS shorter time. Note that when the problem size increased, the
Computer simulation environment includes a notebook Dijkstra algorithm requires larger time to solve which is seen
computer with Intel® Core™ i7-4712 MQ CPU @ 2.30 GHz, in given Table.
8 GB RAM. The planned path is shown in Fig. 4. TABLE III
PERFORMANCES OF ALGORITHMS
The simulations are conducted with 20x20 and 50x50
Method/ 20x20 50x50
environments with random obstacles. The 20x20 grid map
Performanc Iteration Time Iteration Time
example, which includes 400 nodes and 318 of total nodes are
e number (sec) number (sec)
free and 82 nodes are obstacles. At the same time the 50x50
Dijkstra Alg. - 0.352 - 17.104
grid map example, which includes 2500 nodes and 2080 of
Genetic Alg. 102 4.097 281 11.768
total nodes are free and 420 nodes are obstacles. The resulting
paths with HGA are shown in Fig. 4 and Fig. 5, respectively. Proposed HGA 69 2.811 153 7.246
V. CONCLUSION
REFERENCES
Fig. 4 Optimized path with HGA for 20x20.
[1] E. W. Dijkstra, “A note on two problems in connexion with graphs”,
Numerische Mathematik, vol.1, pp. 269-271, 1959.
[2] R. Bellman, “On a routing problem”, Quarterly Applied Mathematics,
vol.16, pp. 87– 90, 1958.
[3] P. E. Hart, N.J. Nilsson and B. Raphael, “A Formal Basis for the
Heuristic Determination of Minimum Cost Paths”, IEEE Transactions
on Systems Science and Cybernetics, vol.4, 2, pp. 100–107, 1968.
[4] X. Chen and Y. M. Li, “Smooth path planning of a mobile robot using
stochastic particle swarm optimization”, In Proceedings of IEEE
International Conference on Mechatronics and Automation, pp. 1722–
1727, 2006.
[5] X. Z. Hu and Q. G. Xu. “Robot path planning based on artificial
immune network”,In Proceedings of the 2007 IEEE International
Conference on Robotics and Biomimetics, pp. 1053-1058, 2007.
[6] Y. Huang, "Intelligent Technique for Robot Path planning Using
Artificial Neural Network and Adaptive Ant Colony Optimization",
JCIT, vol.7, 9, pp. 246 - 252, 2012.
[7] H. W. Mo and Z. Z. Li, “Biogeography based differential evolution for
robot path planning”, In Proceedings of International Conference on
Information and Automation, pp. 1- 6, 2012.
[8] J. C. Mohanta, D. R. Parhi and S. K. Patel “Path planning strategy for
autonomous mobile robot navigation using Petri-GA optimization”,
Fig. 5 Optimized path with HGA for 50x50. Computers&Electrical Engineering, vol.37, 6, pp. 1058-1070, 2011.
[9] J. Holland, Adaptation in Natural and Artificial Systems, Ann Arbor:
University of Michigan Press, 1975.
The performances of the algorithms are given in Table II. [10] J. Cao, ”Robot Global Path Planning Based on anImproved Ant Colony
Algorithm”, Journal of Computer and Communications, vol.4, pp. 11-
These algorithms produce same fitness value for 20x20 and 19,2006
50x50 environments as 59.8701 and 72.8112, respectively. [11] B. K. Oleiwi, H. Roth and B. I. Kazem, "Modified Genetic Algorithm
However, the iteration number and required time change with based on A* Algorithm of Multi Objective Optimization for Path
the algorithm therefore they are given in Table II. According Planning," Jounal of Automation and Control Engineering, vol. 2, no.
4, pp. 357-362, December, 2014.
to the performances, the proposed HGA is always better than
the conventional genetic algorithm. However, for small size
problem, the Dijkstra algorithm provides the solution in a
____________________________________________________________________________________________________________
281
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Stimuli types are very crucial for the performance of The Steady state visual evoked potential (SSVEP) is a brain
electroencephalogram (EEG) based brain computer interface response modulated by the frequency of repetitive visual
(BCI) systems. This study aims to investigate methods for stimulus [8]. SSVEP is known to be most prominent at the
obtaining higher information transfer rate (ITR) through duty occipital location over the visual cortex.
cycle and brightness variation of visual stimuli which have high
The SSVEP-based BCI has been preferred due to high
frequency for steady state visual evoked potential-based BCI.
Although previous studies were concentrated on either duty cycle signal to noise ratio and high information transfer rate (ITR)
or brightness of stimuli separately, our study focused on the according to other BCI approaches [9]. Moreover, the SSVEP
change of duty cycle ratio and brightness of stimuli at the same has a short training and response time [10].
time. Duty cycle values of 40%, 50%, and 60% were used. The term of Information transfer rate is a very significant
During the experiment, 16 flickering stimuli were used on liquid unit that has been used to evaluate overall performance of BCI
crystal display. Participants gazed to the flicker which had systems [11]. Types of stimulus have very important role for
frequency of 15 Hz. Canonical correlation analyses (CCA) was high SSVEP response that related on ITR that can be
used for channel selection and frequency detection. According to described as information transmitted in bits per minute (bpm).
the CCA, the maximum average accuracy of the experiment was
Thus, there are many studies for investigate best stimuli
92.54% when the frequency of flicker was in beta band and its
duty cycle was 40% with a brightness tuning wave. Under the type for SSVEP based BCI systems. Wu (2010) [12] proposed
same conditions stated above, average ITR was improved 16.1% that %40 duty cycle ratio when the flicker was 10 Hz, SSVEP
according to the most commonly used flicker model which is response reached highest value. Shyu et al. (2013) [13] made a
square wave and has 50% duty cycle. SSVEP-based BCI system that using FPGA, which based on
Keywords— Brain Computer Interface, EEG, Human-Computer duty cycle ratio. Manyakov et al. (2013) [14] showed that
Interaction and Systems, Steady State Visual Evoked Potential sampled sinusoidal stimulation enabled for encoding more
targets and elicited SSVEP response with a significantly better
I. INTRODUCTION phase stability. Moreover, Chen et al. (2014) [15] presented
Brain computer interface systems gives chance users to sampled sinusoidal stimuli for SSVEP response. According to
communicate with external devices without using peripheral result of this study, ITR value was reached the highest one
neuromuscular systems. There are many brain imaging when the flicker was at low and high frequencies.
methods, such as electroencephalography (EEG), functional In recent years, researchers presented many approaches to
magnetic resonance imaging(fMRI), magnetoencephalography make and develop SSVEP based BCI systems.
(MEG), and and near-IR spectroscopy (NIRS) have been In this study we aimed to investigate effect the change of
adopted BCI systems [1], [2]. In BCI systems, the commands duty cycle ratio and brightness of stimuli at the same time,
are identified by extracting Electroencephalograms (EEG) although previous studies were concentrated on either duty
which measures the electric field generated from brain. EEG cycle or brightness of stimuli separately.
based BCI systems is preferred for other BCI system because
of practical use, low cost, good response, portability [3]. II. MATERIAL AND METHOD
There are some most common approaches to EEG based Five healthy participants (males, mean age 29 years,
BCI systems that are P300 potentials, event-related corrected to normal vision) joined experiments in this study.
(de)synchronization (ERD/ERS), slow cortical potentials, Participants didn’t have any history of epilepsy and
oscillatory activity, and visual evoked potentials (VEPs) [4-7]. psychiatric disorders.
____________________________________________________________________________________________________________
282
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
A. Experimental Structure
C. Experimental Steps
Fig. 1 The SSVEP stimulator software setting menu
B. Stimulation Types
In this study, we focused on the change of duty cycle ratio
and brightness of stimuli at the same time. Duty cycle values
of 40%, 50%, and 60% were used. As shown in Figure 2, we
used a square pulse that was presented as Type I signal.
𝑇𝑝𝑒𝑟𝑖𝑜𝑎𝑑 is denoted the pulse cycle duration which is total of
𝑇𝑜𝑛 and 𝑇𝑜𝑓𝑓 that are defined stimuli on and off cycle
durations respectively. 𝑅𝑑𝑢𝑡𝑦 is denoted duty cycle ratio which Fig. 4 The 16 Flickers on LCD
is presented in Equation 1. 16 stimuli were 150 pixels x 150 pixels arranged in a 4 x 4
matrix.
𝑅𝑑𝑢𝑡𝑦 =𝑇𝑜𝑛 +𝑇𝑜𝑓𝑓 (1) Participants gazed to the flicker which had frequency of 15
Hz. During the experiment, participant gazed to flicker 6
seconds, after that 4 seconds for resting time. 10 trials were
____________________________________________________________________________________________________________
283
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
examined by each participants for every duty cycle ratio and 𝐵𝑖𝑡𝑠 (1 − 𝑃)
signal types that include type I and type II signals. = 𝑙𝑜𝑔2 𝑁 + 𝑃𝑙𝑜𝑔2 𝑃 + (1 − 𝑃)𝑙𝑜𝑔2 [ ]
𝐶𝑜𝑚𝑚𝑎𝑛𝑑 (𝑁 − 1)
D. Data Analysis 𝐵𝑖𝑡𝑠 60
𝐼𝑇𝑅 = × (4)
Canonical correlation analysis (CCA) is a multivariate 𝐶𝑜𝑚𝑚𝑎𝑛𝑑 𝐶𝑇𝐼
analysis that presented by Lin[16]. According to the Equation
2 CCA investigates the maximum correlation between the two Where P denotes accuracy of classification, N denotes total
variables. stimuli, CTI denotes command transfer intervals.
In this study, CTI is 6 seconds because of gaze time during
𝐸[𝑥 𝑇 𝑦] in one trial.
𝑚𝑎𝑥 𝜌 (𝑥, 𝑦) =
√𝐸 [𝑥 𝑇 𝑥]𝐸[𝑦 𝑇 𝑦] III. RESEULT AND DISCUSSION
Experiments were completed without any problem. Table 2
shows that outcome of the experiment average accuracy and
𝐸[𝑊𝑋𝑇 𝑋𝑌 𝑇 𝑊𝑌 ] (2)
= the average ITR.
√𝐸[𝑊𝑋𝑇 𝑋𝑋 𝑇 𝑊𝑋 ]𝐸[𝑊𝑌𝑇 𝑌𝑌 𝑇 𝑊𝑌 ]
TABLE I
The weight vectors 𝑊𝑥 and 𝑊𝑦 to maximize the correlation RESULT OF EXPERIMENT ACCORDING TO THE CCA BASED
between, linear combinations, x and y by Equation 2. METHOD
𝑚𝑎𝑥 𝜌 (𝑥, 𝑦) denoted as the maximum canonical
correlation of the 𝑊𝑥 and 𝑊𝑦 . As shown in Figure 5, shows the Average
Duty Cycle Average ITR
CCA model was presented. According to the model, X Ratio
Signal Type Accuracy
(bpm)
denoted EEG signal, 𝑌𝑓 denoted the reference signal. (%)
sin(2𝜋𝑓𝑡)
cos(2𝜋𝑓𝑡)
1 2 𝑁𝑡 Average ITR (Bpm)
. (3)
𝑌𝑓 = , 𝑡 = , ,…, 34
. 𝐹𝑠 𝐹𝑠 𝐹𝑠
sin(2𝜋𝐻𝑓𝑡) 32
16.1% 32,89
( cos(2𝜋𝐻𝑡) )
30
Sixth-order Butterworth band pass filter with 5 and 20 Hz 28,32
28
cut-off frequencies was applied. After that CCA method was
used. 26
C𝑊𝑥 showed that brain frequency map. According to the 50% TYPE I 4 0 % TYPE II
result, 𝑂𝑧 , 𝑂1 , 𝑂2 EEG channels were used for data accuracy.
ITR is a very significant unit that has been used to evaluate Fig. 5 The difference of average ITR and improved ratio
overall performance of BCI systems. As written before, ITR is a very important metric to show
ITR can be described by following Equation. the overall performance of BCI systems. There are different
methods to find out a high ITR; (1) increasing total stimuli, (2)
____________________________________________________________________________________________________________
284
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
improving the accuracy, and (3) decreasing the command [10] Z. Oralhan and M. Tokmakçı. "The Effect of Duty Cycle and
Brightness Variation of Visual Stimuli on SSVEP in Brain Computer
transfer intervals. In this study, we aimed to improve ITR
Interface Systems." IETE Journal of Research (doi:
value by increasing the accuracy with a new created signal 10.1080/03772063.2016.1176543), 1-9, 2016.
type for stimuli. [11] B. Dal Seno, M. Matteucci, and L.T. Mainardi, “The utility metric: A
We applied statistical test to the results shown in Table 2 to novel method to assess the overall performance of discrete
braincomputer interfaces,” IEEE Trans. Neural Syst. Rehabil. Eng., vol.
examine whether the differences in the classification accuracy
18(1), pp. 2028, Feb. 2010.
were significant depending on the duty cycle and flicker type. [12] Z. Wu, “The difference of SSVEP resulted by different pulse duty-
A one-way analysis of variance (ANOVA) was conducted on cycle,” in Communications, Circuits and Systems ICCCAS 2009
the classification of accuracy for each flicker models, International Conference, Milpitas, CA, Jul. 2009, pp. 605607.
[13] K. K. Shyu, Y. J. Chiu, P. L. Lee, J.M. Liang, and S. H. Peng,
indicating that the difference was considered significant at the
“Adaptive SSVEP-based BCI system with frequency and pulse duty-
p-value < 0.05. cycle stimuli tuning design,” IEEE Trans. Neural Syst. Rehabil. Eng.,
vol. 21(5), pp. 697703, Sept. 2013.
CONCLUSION [14] N. V. Manyakov, N. Chumerin, A. Robben, A. Combaz, M. Van Vliet,
and M. M. Van Hulle, “Sampled sinusoidal stimulation profile and
It can be concluded that previous studies mainly focused on multichannel fuzzy logic classification for monitor-based phase-coded
the effects of either duty cycle pulses ratio or brightness of SSVEP braincomputer interfacing,” J. Neural Eng., vol. 10(3), pp. 115,
stimuli separately. But, we aimed to investigate the effects of Apr. 2013.
both parameters at the same time. We optimized signal type [15] X. Chen, Z. Chen, S. Gao, and X. Gao, “A high-ITR SSVEP-based
BCI speller,” Taylor Francis Brain-Comput. Interfaces, vol. 1(34), pp.
and duty cycle ratio. 181191, Sept. 2014.
Despite the fact that there is not enough knowledge for [16] Z. Lin, C. Zhang, W. Wu, and X. Gao, “Frequency recognition based
characteristic attitude of brightness and duty cycle parameters on canonical correlation analysis for SSVEPbased BCIs,” IEEE Trans.
which are very important for ITR and accuracy in SSVEP- Biomed. Eng., vol. 53(12), pp.26102614, Jun. 2007.
based BCI systems, in this study, for these parameters
remarkable results have been accomplished.
ACKNOWLEDGMENT
The experiments were performed according to the rules of
the Ethics Committee of Erciyes University, Kayseri, Turkey,
under registration number 2014/352.
This work is supported by Department of Scientific
Research Projects of Erciyes University, Turkey [grant
number 2014-5204 FDK].
REFERENCES
[1] N.Weiskopf, K.Mathiak, S.W. Bock, F. Scharnowski, R. Veit,W.
Grodd, R. Goebel, and N. Birbaumer, “Principles of a brain–computer
interface (BCI) based on real-time functionalmagnetic resonance
imaging (fMRI),” IEEE Trans. Biomed. Eng., vol. 51, no. 6, pp. 966–
970, Jun. 2004.
[2] R. Sitaram, A. Caria, and N. Birbaumer, “Hemodynamic brain–
computer interfaces for communication and rehabilitation,” Neural
Netw., vol. 22, pp. 1320–1328, Nov. 2009.
[3] G. R. Muller-Putz and G. Pfurtscheller, “Control of an electrical
prosthesis with an SSVEP-based BCI,” IEEE Trans. Biomed. Eng., vol.
55, no. 1, pp. 361–364, Jan. 2008.
[4] E. Donchin, K. V. Spencer, and R. Wijesinghe, “The mental prosthesis:
Assessing the speed of a P300-based braincomputer interface,” IEEE
Trans. Rehab. Eng., vol. 8(2), pp. 174179, Jun. 2000.
[5] N. Birbaumer, A. Kubler, N. Ghanayim, T. Hinterberger, J.
Perelmouter, J. Kaiser, I. Iversen, B. Kotchoubey, N. Neumann, and H.
Flor, “The thought translation device (TTD) for completely paralyzed
patients,” IEEE Trans. Rehab. Eng., vol. 8(2), pp. 190193, Jun. 2000.
[6] G. Pfurtscheller and C. Neuper, “Motor imagery and direct brain
computer communication,” Proc. IEEE, vol. 89(7), pp. 11231134, Jul.
2001.
[7] Y. Zhang, G. Zhou, J. Jin, X. Wang, and A. Cichocki, “Optimizing
spatial patterns with sparse filter bands for motor-imagery based brain-
computer interface,” J. Neurosci. Methods, vol. 255, pp. 8591, Nov.
2015.
[8] Y. Wang, R. Wang, X. Gao, B. Hong, and S. Gao, "A practical VEP-
based brain-computer interface," IEEE Transactions on Neural Systems
and Rehabilitation Engineering, vol. 14, pp. 234-240, 2006.
[9] Y. Wang, Y. T. Wang, and T. P. Jung, “Visual stimulus design for
high-rate SSVEP BCI,” Electron Lett., vol. 46 (15) pp. 10571058, Jul.
2010
____________________________________________________________________________________________________________
285
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Topology Control in
Mobile Wireless Sensor Networks
Umut Can Çabuk*, Gökhan Dalkılıç+
*
Department of Electrical and Electronics Engineering, Erzincan University
Yalnizbag Campus, Erzincan, Turkey
umutcancabuk@hotmail.com
+
Department of Computer Engineering, Dokuz Eylul University
Tinaztepe Campus, Buca, Izmir, Turkey
dalkilic@cs.deu.edu.tr
Abstract—Topology control is a set of strategies and actions those about topology control [3] and maintenance [4] in wireless
aim to improve energy efficiency, lifetime and reliability of the sensor networks. And of course we used valuable information
networks. One of the most effective methodology is topology provided in these works. But we couldn’t find a comparative
reduction, which is a part of the overall topology control process. study to refer, which focuses directly on mobility case. For
Topology control process is usually considered under two phases:
instance, in [5] authors present a comparative study similar to
construction and maintenance. This hard work would even be
harder for mobile ad-hoc sensor networks, those consist of devices ours, however they used fixed nodes and totally excluded the
with limited capabilities and continual independent movement. In mobility case, which is our main focus. In their work, Latré et
al. [6] introduce a protocol (Cascading information retrieval by
this work, we made our efforts to find out effects of mobility of the
nodes in an ad-hoc wireless sensor network on network controlling access with distributed slot assignment - CICADA)
parameters, by using different pre-defined topology control designed for mobility cases and present a detailed analysis, but
protocols. Thus, we made simulations using an experimental their study is not comparative. Reluctantly, we could not
simulator, called Atarraya. Furthermore, we measured the include CICADA and CICADA-S (a secure version proposed
performance of each protocol and compared the outcomes in later [7]) in our work because of the limitations of the tool we
order to find the best topology control strategy. Lastly, possible
have used.
solutions to the uncovered problems are proposed.
The reason of that scarcity is that wireless sensor networks
is a quite new area, and mobility in these networks is even a
Keywords—Ad-hoc wireless sensor networks, mobility, topology
newer aspect. Most of the previous works in this area focused
control, wireless sensor Networks, simulation
on higher layer architecture, like routing and forwarding
I. INTRODUCTION protocols. Here, we worked on a lower level relationship
between nodes by reducing the number of links between them.
After conventional wide area and local area networks (LAN),
The simulator we used, namely Atarraya, was also
nowadays, personal area (PAN) and body area networks (BAN)
introduced in a previous work [8] and still under development
are becoming more and more popular in applications of
with new features. A side note: Mobility wasn’t even included
industry, commerce and science [1]. Along with their abilities
in our simulation tool Atarraya’s first version, here we used
and advantages, these networks also have their own challenges
version 1.3.
[2]. For instance, they usually include nodes on the move or at
least nodes that can be relocated. Hence, the newest solutions III. ASSUMPTIONS & DECISIONS
in the interconnecting network technologies have to deal with
the mobility cases as well. To find key points, we had to choose optimum preferences
From that starting point, in order to build the rest of the work and had to limit our angle of sight. Our very first assumption
on it, we’ve presumed a set of use cases, called as scenario. about the concept is a scenario as described in introduction.
Although the results wouldn’t be directly related to the scenario, Before going on, we assume the undetermined smallest integer
it would help us to make some decisions. Our scenario based time unit in Atarraya as milliseconds.
on an indoor working environment including unsynchronized A. Algorithms
moving people, who are representing BAN or PAN networks
by holding or even wearing wireless sensor devices. Most For the evaluation, we wanted to choose topology control
applicable realization is hospitals, but also mines, schools, and maintenance algorithms, which have different approaches
prisons, sport fields and in a wider point of view, zoos. as possible and tried to avoid similar algorithms. So, we have
Additionally, we suppose a continuous communication is chosen A3, energy efficient connected dominating set (EECDS)
required in our scenario. and Kneigh tree protocols for topology construction. Detailed
descriptions of these protocols can be found in [9, 10]. Then we
II. PREVIOUS WORKS considered dynamic global time-based topology recreation
While building our background, we found out several studies (DGTTRec), static global time-based topology rotation
(SGTTRot), and HGTTRec and rotation (HGTTRecRot) for
____________________________________________________________________________________________________________
286
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
topology maintenance. All these algorithms are chosen from Obviously, this connectivity metric is a function of time (and
the set that is predefined in the tool [8]. Adding custom movement), so it will also represent robustness and
protocols is not possible for end-users but the source of the tool sustainability of the protocols. By the way, this metric can be
must be edited, which could be done in another work. the most important information to determine the usefulness of
During topology maintenance evaluation we used A3 as a a topology control protocol depending on the scenario.
base construction algorithm, in order to make results
comparable. C. Movement Model
Energy efficiency is always a fundamental aim, but in We designed a movement model in order to provide mobility
mobile networks geographic distribution of nodes is dynamic, to nodes in the simulation. There are two main limitations to
this means physical link between nodes are not stable, hence it consider: First, movement model should be realistic, thus
is more accurate to use time triggered maintenance algorithms results will be useful in further works and industrial
than energy threshold triggered ones, to be sure about the applications. Second, simulation tool Atarraya is an (very
connectivity. useful but) experimental tool and has limited capabilities in this
manner [8].
B. Test Scene, Parameters & Metrics So, our movement model consists of these features: Simple
We created a 1000 x 1000 m2 observation field, but didn’t random walk mechanism for each node with sleep time 100 ms
use this entire field for initial node deployment. We placed 30 and step size 1 m, without surrounding limits (fences) on the
nodes in a 500 x 500 m2 area in the center of the observation field. This makes 10 m/s and is 5 to 10 times faster than the
field mentioned above, using Normal distribution with sigma = average human walking speed, which is around 1 to 2 m/s [12].
100. When simulation starts, nodes can leave this initial We used that increased value to see the effects of mobility more
deployment area and can move in the whole observation field, clearly. By using randomness, movement of nodes is non-
or even may leave that area. deterministic to each other and no predictions can be made
Our test field represents a surrounded area like a hospital, a besides the step size.
factory or an office environment. It can also be an open-air field Another characteristic of the model is spreading. All nodes
but we mostly focused on an indoor environment. Hence, we in a very long period of time, eventually, will leave the test field
placed 30 nodes (+1 initializer sink), which makes a sparse since there is no fence. This is not an intended behavior but due
network. This number may be increased until around 100 to program limitations, it is not possible to prevent this. Only
without distorting the test outcomes. Numbers bigger than a the simulation time can be limited at an optimal time period.
few hundred may cause the simulator to slow down, stutter or However, this behavior allows us to see what happens when
act inconsistently. some nodes leave the area, which can be a realistic scenario.
We used a realistic communication range of 100 m, which is
typical for ZigBee, Bluetooth Class 1 and Wi-Fi b/g/n [11]. We IV. RESULTS
set sensing area to 20 m but, in this work, it wasn’t important We first evaluated the topology construction protocols, by
in our point of view, so we didn’t take care of sensing range starting the simulation without a time limit. 30 Nodes and 1 sink
and sensed data. node are deployed in the field as described above. Simulation
As total simulation time for topology maintenance tests, we ended up after a successful topology creation. All values given
considered 60 seconds. It was enough to see effects of mobility, in this section are average of 10 measurements.
plus we had some limitations about our movement model (see
below). This interval theoretically provides 600 m of moving
70
range to each node (see below). Thus, we also decided re- 55,886 58,42
60
setting the topology time interval 1 s.
We assigned 1000 mJ total energy stock to each node and 50
used simple energy model of the software. We also used simple 40
26,188
communication model. Another assumption was 0 bit error rate 30
(BER) value, means communication link is perfect. Effect of 20
7,08
BER is subsequently analyzed. 10
Finally, a connectivity metric is defined to measure the 0
performance of the protocols in the sense of physical and Topology construction time (s)
logical connectivity. This metric is considered as a percentage
of nodes (directly or as multi-hop) connected to the sink node A3 EECDS KNEIGH-T CDS-K
among all the nodes in the coverage of the core network.
Basically, the nodes out of the core network coverage are Fig. 1 Durations of the initial topology construction period
excluded from that ratio. Here, core network means the network
initialized by the sink.
Fig. 1 gives the time passed until the successful creation for
# ( )−1 each protocol. Here, it is very clear that, tree based A3 and
(%) = Kneigh tree protocols are much faster than dominating set
# ℎ ( )−1
based EECDS and k-connected k-dominating set (CDS-K).
____________________________________________________________________________________________________________
287
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Due to the mobility of the nodes, it would be better to use faster So, information given until here, are not enough to uncover
protocols. the effects of the mobility in that network. Because, if there
were no mobility, results would be very close, and also network
2000 1786
operating time should be much higher than construction time.
1535 Over and above, we have to check topology maintenance
1500 simulation results.
1000 4000
3362,55 3277,53
496
500 256 244 228 3000 2665,21
98 31
0 2000
A3 EECDS KNEIGH-T CDS-K
1000 542,45 716,36
Messages sent Messages received
197,23
0
Fig. 2 Total message traffics during the initial topology construction EECDS+DGTT CDS-K+DGTT A3+DGTT A3+ELP DSR A3+SGTT A3+HGTT
Fig. 3 Total energy consumptions in the whole network during topology 40000
construction (for 31 nodes)
20000
0
Fig. 3 shows total energy consumed among whole network. EECDS+DGTT CDS-K+DGTT A3+DGTT A3+ELP DSR A3+SGTT A3+HGTT
These values are some of the consumptions of all 31 nodes. As
expected from the message traffic results, A3 and Kneigh tree Messages sent Messages received
consume much less energy from EECDS and CDS-K. While
Kneigh tree consumes least energy, EECDS becomes the last Fig. 5 Total message traffics during topology maintenance in 1 minute run
preferable with the highest consumption, which is 4 times more
than A3 and 30 times more than Kneigh tree.
In Fig. 5, total message traffic of the network is given. There
A very important statement; all these protocols complete is a huge difference between DGTT and the others. Actually,
their work by constructing the topology much faster than the the number of sent messages is very similar, just HGTT is
movement speed of mobile nodes. Our movement model
relatively higher. But, the number of received messages is very
proposes a 5 m step size per 500 ms. But, here the slowest
high in DGTT. It is 4 times higher than the runner-up HGTT.
protocol, CDS-K, requires 58.42 ms. Even when we change our
SGTT has the lowest total traffic in general. In fact, the reason
movement model (unless we set an unrealistic super speed like
of this big difference between the number of received and sent
100 m step size per 10 ms or so), the results are (almost) the messages is broadcast style transmission and mass reception of
same as stationary ad-hoc wireless sensor networks’ results.
____________________________________________________________________________________________________________
288
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the broadcasted messages by all nodes in the range. This boundary, connectivity metric is very high; 97% to 100%. And
behavior artificially boosts the network traffic. until 10^-3 values are acceptable. But one shouldn’t forget that,
Probably the most important data is presented in Fig. 6: sensor data transfer is not included in this communication. It
connectivity metric. It shows reliability and sustainability of the just represents topology control messagings. So, for higher
protocols via 2 different time intervals. Here, all DGTT level decisions about the architecture, level of the BER should
combinations together get the highest value by %100 be maintained no lower than 10^-5 if the data sensed is time
connectivity rate in both 1-minute and 5-minutes tests. Mobility critical.
(in the coverage area of course) does not cause impairment
while using DGTT. Distinctly, if a node is in the 100%
communication range, then it will be included to the network
via DGTT. The others have relatively lower values in 1-minute 80%
test and they become much worse by the time, as can be seen
60%
in 5-minutes test. In a 10-minutes test (which is not shown in
the figure) ratios of SGTT and HGTT are below 25%. 40%
20%
100%
0%
80% 0 10^-5 10^-4 10^-3 10^-2
60% A3+DGTT CDS-K+DGTT EECDS+DGTT
40%
Fig. 7 Effect of the bit error rate on the connectivity ratio
20%
0%
EECDS+DGTT CDS-K+DGTT A3+DGTT A3+ELP DSR A3+SGTT A3+HGTT 100% 14000
12000
Connectivity in 1 min Connectivity in 5 mins 80%
10000
60% 8000
Fig. 6 Connectivity levels during topology maintenance in 1 minute and 5 6000
minutes of run 40%
4000
20%
2000
Manifestly, here, we discovered a clear tradeoff between 0% 0
energy spent plus message traffic and connectivity performance. 0,1 0,5 1 2 5 10
There is no linearity, but the correlation is clear. Before
evaluating the values given in Fig. 6, it wasn’t possible to say Connectivity Energy spent (mJ)
A3 or CDS-K or even EECDS can be a good choice. But after,
we can say, for that type of a network, DGTT has the best Fig. 8 Effect of topology recreation time interval on connectivity and energy
results and preferable with A3, from the previous results. consumption based on 1 minute run using A3 + DGTT
So-called energy efficiency and traffic emptiness of SGTT
and HGTT can be very deceptive. This is highly probable a
paralogy, caused by the low connectivity and the link loss. Less We discovered a clear correlation between connectivity and
links mean, less message receptions and less message energy consumption, but this does not mean there is nothing to
receptions lead less energy consumption. Even so, this do to save energy. Figure 8 shows the effect of topology
conjecture is valid for specific scenarios and could be different recreation interval during maintenance on both connectivity
under variety of conditions. and energy consumption. First, please note that we used 500 ms
All in all, although these values can give very valuable movement sleep time and 1 s refreshing interval in previous
information and can provide guidance for related applications, experiments. Here we kept movement model as defined, but
the optimum decision is always application dependent. For tested different time intervals. Hence, we exposed that, a
instance, in a scenario that continuous communication is not as refreshing frequency which is much higher than the movement
important as node lifetime (energy efficiency), DGTT should speed (this also means physical deployment change frequency)
be avoided, while A3 could still be preserved. causes extraordinary energy consumption and highly
We also evaluated the effect of BER on DGTT with A3, ineffective. We can also see that a frequency much lower,
CDS-K and EECDS respectively. In Fig. 7, the relation causes loss of communication links. Thus, a balance should be
between BER and connectivity is analyzed. It can be clearly preserved to save energy and to protect connectivity at the same
seen that, a higher error rate lowers the connectivity metric time. An interval 2 to 4 times longer than basic movement
significantly, but this relation is roughly logarithmical. While period looks acceptable in our case. As a final comment,
the error rate is 10^-5, which is a de facto industry standard
____________________________________________________________________________________________________________
289
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
statistical information about the deployment environment and [3] P. M. Wightman Rojas, “Topology Control in Wireless Sensor
Networks,” PhD thesis, Computer Science and Engineering
movement should be obtained to find a (sub-) optimal solution.
Department, University of South Florida, Tampa, Florida, USA, 2010.
[4] P. M. Wightman, M. A. Labrador, “Topology Maintenance: Extending
V. CONCLUSIONS & FUTURE WORKS the Lifetime of Wireless Sensor Networks,” IEEE Latin America
In this work, we aimed to show possible effects of the Transactions, vol. 8, no. 4, pp. 469-475, Aug. 2010.
[5] G. Pachnanda, R. Chaudhary, “Comparative Study of A3, EECDS,
mobility to ad-hoc network performance and reliability in CDS Rule K and KNEIGH Tree Protocols in a Grid Manner,” Advances
different aspects. Furthermore, we wanted to propose basic in Electronic and Electric Engineering, vol. 3, no. 4, pp. 509-514, 2013.
solutions. [6] B. Latré, et al., “A low-delay protocol for multihop wireless body area
In spite of the application dependability, we got some solid networks,” in Proc. 4th Annual International Conference on Mobile
and Ubiquitous Systems: Networking & Services, 2007, pp. 1-8.
outcomes; first, unlike DGTT, SGTT and HGTT are not [7] D. Singelée, et al., “A Secure Low-Delay Protocol for Wireless Body
suitable for that type of a mobile network. One should better Area Networks,” Ad Hoc & Sensor Wireless Networks, vol. 9, pp. 53-
prefer a dynamic protocol, at least in a similar scenario like our 72, 2010.
presumption. Because of the connectivity level provided, we [8] P. M. Wightman, M. A. Labrador, “Atarraya: A Simulation Tool to
Teach and Research Topology Control Algorithms for Wireless Sensor
are not satisfied with the performance of static and hybrid Networks,” in Proc. Simutools '09, 2009, Article No. 26.
methods. But in a different scenario, significant energy saving [9] P. M. Wightman, M. A. Labrador, “A3: A Topology Construction
(we also found out that this is deceptive and caused by physical Algorithm for Wireless Sensor Networks,” in Proc. GLOBECOM, 2008,
link loss) of these protocols can be exploited in a way. pp. 1-6.
[10] G. Pachnanda, K. Singh, L. Gangwar, “Comparative Analysis of A3,
Additionally, between the construction algorithms, A3 EECDS and KNEIGH Tree Protocols in Wireless Sensor Networks,”
provides better values from the rest apart from Kneigh tree. But International Journal of Electronics and Computer Science
in long term, we showed that this advantage gained from Engineering, vol. 2, no.3, pp. 987-991, 2013.
construction phase becomes less significant by the time. Even [11] H. Fornazier, A. Martin, S. Messner (2012) “Wireless Communication:
Wi-Fi, Bluetooth, IEEE 802.15.4, DASH7,” [Online]. Available:
so, A3 could be a reasonable choice for our scenario. http://rose.eu.org/2012/wp-content/uploads/2012/03/Wireless-
An impromptu (but predictable) observation (or solution) to communication.pdf
the “connectivity-energy efficiency” problem is also given. [12] R. C. Browning, E. A. Baker, J. A. Herron, R. Kram, "Effects of obesity
We got a strong impression about; building a cell-like and sex on the energetic cost and preferred speed of walking," Journal
of Applied Physiology, vol. 100, no. 2, pp. 390-398, Feb. 2006.
structure by placing several stationary nodes in the deployment
field could help us to reduce redundant traffic and to increase
energy efficiency without downgrading the link quality. This
hybrid network structure may give better results. This strategy
should be researched and it can be a good extension to this work.
We still have no overall optimum choices, so search for a
topology maintenance protocol that offers a better balance
between energy consumption (or message traffic) and
connectivity should go on.
Another point is super-speed mobility. We created a realistic
human-like movement model; it can also be used for animals,
or robotic structures. But super-speed mobility is not supported;
in that case it is not even possible to construct a network. And
even we deploy a cell structure, handovers would be a big
problem. However, super-speed moving objects, including
racing cars, satellites, missiles or so, rarely move randomly.
They follow a track or orbit. This feature can be exploited. So,
that aspect can also be another topic to investigate.
ACKNOWLEDGMENTS
We want to thank Dr. Qi Zhang from Aarhus University,
Department of Engineering, for her valuable guidance. Most
work is done using the simulator Atarraya and we would like to
present our sincerest respect and recognition to creators of the
program: Pedro M. Wightman and Miguel A. Labrador.
REFERENCES
[1] S. Gonzalez-Valenzuela et al., Autonomous Sensor Networks:
Collective Sensing Strategies for Analytical Purposes, D. Filippini, Ed.
Berlin, Germany: Springer-Verlag, pp. 17-38, 2012.
[2] M. A. Hanson, et al., “Body area sensor networks: Challenges and
opportunities,” Computer, vol. 42, no. 1, pp. 58-65, Jan. 2009.
____________________________________________________________________________________________________________
290
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The big bang–big crunch (BB–BC) algorithm has They do not require the assumptions as done in standard
been proposed as a new optimization method based on the big algorithm.
bang and big crunch theory, one of the theories of the evolution They do not need the change on the given problem unlike
of the universe. The BB-BC algorithm has been firstly presented the usual algorithms. They adapt themselves for solving
to solve the optimization problems with continuous solutions different types of optimization problems [2].
space. If the solution space of the problem is binary-structural, In the last decades, different swarm-based evolutionary
the algorithm must be modified to solve this kind of the algorithms have been proposed for solving this optimization
problems. Therefore, in this study, the BB-BC method, one of the
problem including Genetic Algorithms [3-6], Tabu Search
population-based optimization algorithms, is modified to deal
with binary optimization problems. The performance of the algorithm [7, 8], Ant Colony Optimization [9], Particle Swarm
proposed methods is analyzed on uncapacitated facility location Optimizer [10-12] and Artificial Bee Colony [13-15]. The
problems (UFLPs) which are one of the binary problems used in random selection process and the information attained at the
literature. The well-known small and medium twelve instances of end of each iteration (cycle) are utilized in order to discover
UFLPs are used to analyze the performances and the effects of more optimal solutions in the subsequent iterations [16].
the control parameter of the BB-BC algorithm. The obtained Big-Bang Big-Crunch (BB-BC) algorithm which is one of
results are comparatively presented. According to the the swarm intelligence algorithms has been proposed by Erol
experimental results, the binary version of the BB-BC method and Eksin in 2006 for numerical optimization problems and
achieves successful results in solving UFLP in terms of solution
was based on the big bang and big crunch theory, one of the
quality.
theories of the evolution of the universe [17]. While in the Big
Bang phase, the BB–BC method similarly produces haphazard
Keywords— Big Bang-Big Crunch Algorithm, Population-based
points in solution space, in the Big Crunch phase it shrinks all
optimization algorithms, Binary optimization, UFLP, Modulo
function of the points in the search space to a single agent point due to
a centre of mass. It has shown that The BB–BC method
I. INTRODUCTION outperformed the enhanced classical Genetic Algorithm on
Many swarm intelligence methods have been recently many benchmark problems.
proposed in order to solve compelling optimization problems According to the literature review, the basic BB-BC
by reason of their simple structures and creation of influential algorithm is a competitive algorithm in solving optimization
solutions for problems [1]. Heuristics algorithms are the problems with continuous solution space. If the solution space
algorithms that employ a simple approach to produce an of the problem is binary-structured, the basic BB-BC
acceptable solution to search and are recently becoming algorithm must be modified in order to solve this type of
powerful and getting more common. The reasons of that can optimization problems. Using modulo function that is one of
be shortly given as follows: the main mathematical operators, we propose a binary version
In case of different decision variables, objective functions of the BB-BC method for obtaining the reasonable solutions
and constraints, they provide strategies of the general solution for binary optimization problems. The proposed method is
that can be practiced to the problem. investigated on a standard binary optimization problem by
They independently operate from the type of solution utilizing the uncapacitated facility location problem (UFLP).
space, the number of decision variables and constraints. The UFLP is one of the most commonly used problems in
They utilize probabilistic random searches. They do not combinatorial optimization. In this problem, the main
need excessive computation time because their computation objective is to minimize the total cost by providing the
power is in reasonable level. demand of customers under the given conditions that are a
The processes of their transformation and adaptation for constant cost of setting up a facility and a shipping cost of
different type of problems are simple. satisfying the customer demand for the corresponding facility
They dictate fewer mathematical requirements and in [10].
addition, they do not need very well defined mathematical
models.
They present impressive solutions for the large-scale
combinatorial and non-linear problems.
____________________________________________________________________________________________________________
291
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
II. BASIC VERSION OF BIG BANG BIG CRUNCH ALGORITHM The steps of the BB-BC method are presented in Fig. 1.
(BB-BC)
The BB-BC algorithm was firstly introduced for solving Description
continuous optimization problems in 2006 by Erol and Eksin. To initialize algorithm, the size of population,
Initialization number of iteration, fitness function and error value
This method is constructed on two main steps: the first phase Stage
is the Big Bang phase in which all of the candidate solutions are determined.
are randomly distributed into the search space and the next A population which consists of N individuals in
Step 1:
phase is the Big Crunch where a centre of mass is calculated search space is randomly generated.
considering individuals in the whole population [18, 19]. The The values of all the candidate solutions (individuals)
Step 2:
initial population is randomly produced over the search space are calculated by using fitness function.
as done in the other swarm-based algorithms. All subsequent The center of mass or the fittest individual is
Step 3:
Big Bang phases are randomly distributed about the centre of determined as Big Bang point by the help of Eq. (1).
mass or the candidate with the best fitness value in a similar A new population is generated about the center of
way. The most important feature of the algorithm is that it has Step 4:
mass or the best-fit individual.
high convergence speed but low computational complexity. Go to Step 2 until the stopping criterion (number of
Step 5:
For instance, while many evolution-based algorithms in the iteration or error value) is met.
literature present near-optimal solutions at the end of too Fig. 1. The steps of the BB-BC method
much iteration, BB-BC algorithm obtains solutions very close
to the optimal solution of this problem on far less number of III. PROPOSED BINARY VERSION OF BB-BC
iterations in general.
Modulo base 2 is used to convert the continuous solutions
After the Big Bang, a contraction procedure is implemented to binary version. This conversion is presented in Eq.4.
during the Big Crunch. In this stage, the contraction operator
holds the available locations of each candidate solution in the
, = ( , , 2) (4)
population and its associated cost function value and
calculates a centre of mass. The centre of mass is presented as
follow: where, , is binary solution obtained from , , ⌊ ⌋ is
rounding operation to down, abs is a function used in order to
∑ obtain absolute value of , . When , is computed, first of
= all, rounding operator is applied to , , then absolute value of
∑ (1)
, is obtained. As for last process, the modulo base two is
,where Xc = position of the centre of mass; Xi = position of applied to the obtained value.
candidate solution; fi = value of cost function of candidate i;
and N = population size. The best fittest candidate solution , = round (| , |) mod 2
can also be utilized as the starting point instead of the position
of the centre of mass. An example of the solving Eq. (4) is presented as follows:
After the second stage completes, new individuals are once
again calculated for Big Bang stage according to the formula , = round (| − 12.24|) mod 2
below. , = round (12.24) mod 2
r , =12 mod 2
X =X + =0
k (2) ,
( ) = min ∑ ∑ , , +∑ (5)
subject to :
∑ , =1 ∀ (6)
____________________________________________________________________________________________________________
292
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Problem name Problem size Cost of the optimal solution Cap133 977059,899 17914,583 934917,063 1012152,913
Cap71 16x50 932,615.75 Cap134 1047398,447 30954,550 982799,250 1097979,288
Cap72 16x50 977,799.40
Cap73 16x50 1,010,641.45 The reporting results obtained using center of mass
Cap74 16x50 1,034,976.98 function for value l=2 are given in Table III. When examined
Cap101 25x50 796,648.44 the mean values, it could not be reached to the optimal value
Cap102 25x50 854,704.20 for any problem under the given parameters. However, when
Cap103 25x50 893,782.11
the best values are examined, the optimal results could not be
found for small and medium-sized problems problem groups
Cap104 25x50 928,941.75
except for large-sized problems. In addition, while the
Cap131 50x50 793,439.56
standard deviation values obtained from small-sized problems
Cap132 50x50 851,495.33
got lower, those of medium-sized and large-sized problem sets
Cap133 50x50 893,076.71 got higher. Therefore, it can be said that this method is more
Cap134 50x50 928,941.75 robust for small-sized problems than the other problem
groups.
In order to make a clear and consistent comparison each
other, the population size is taken as 40. For the entire
experiments, the termination condition is selected as the
maximum number of function evaluations (Max_FEs) and it is
set as 80.000 and each experimental study performed by using
center of mass and best fit individual instead of Eq. 1 and
different upper limit (l), which is one of the control parameter
of the BB-BC, is run 30 times in order to solve UFLP. The
mean, best and worst values and the standard deviations
____________________________________________________________________________________________________________
293
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III conditions, any optimal value for any problem could not be
EXPERIMENTAL RESULTS OBTAINED USING CENTER FUNCTION FOR L= 2
found. However, when analyzed the best values, it is seen that
the proposed method reached to the optimal solution for
Mean Std Best Worst
small-sized and medium-sized problems problem groups but
Cap71 934166,882 1993,341 932615,750 939177,513 not large-sized problems. Besides, when analyzed the mean
Cap72 979944,208 2584,446 977799,400 984829,450 values in this table while this method obtained very near-
optimal results for both of small-sized and medium-sized
Cap73 1012865,261 1802,325 1010641,450 1015589,325
problems, it did not obtain good solution for large-sized
Cap74 1040141,943 5846,441 1034976,975 1053912,938 problems. Therefore, it can be specified that this method is
Cap101 804429,968 5385,079 796648,438 812834,288 relatively more effective for small-sized and medium-large
problem sets than large-sized problems.
Cap102 863982,102 6550,753 854704,200 875894,200
Cap103 904765,695 8869,370 893782,113 917587,900 TABLE V
EXPERIMENTAL RESULTS OBTAINED USING BEST FIT INDIVIDUAL FOR L= 1
Cap104 952661,198 17286,486 928941,750 976991,075
Cap131 821229,431 18868,929 795883,238 851320,750 Mean Std Best Worst
Cap132 912322,125 21859,908 854704,200 938541,638 Cap71 933189,717 1389,929 932615,750 938122,238
Cap133 980653,598 16274,168 948982,750 1006973,775 Cap72 978581,522 1488,624 977799,400 982711,600
Cap134 1056551,117 25561,281 1005104,075 1105009,113 Cap73 1011318,064 1011,356 1010641,450 1014253,438
Cap74 1036464,014 3242,674 1034976,975 1052187,150
The experimental results attained utilizing center of mass
Cap101 798149,257 1574,117 796648,438 801947,025
function for value l=3 are given in Table IV. When examined
the mean values, the optimal value could not be found for any Cap102 858319,158 2494,585 854704,200 862792,900
problem under the available parameters. However, when Cap103 897041,810 3883,140 894008,138 911215,038
examined the obtained best values, it is seen that this method
Cap104 934136,345 5617,622 928941,750 955362,300
obtained optimal results for small-sized and medium-sized
problems except for large-sized problems. In addition, while Cap131 807064,141 8882,906 794956,113 835703,763
the standard deviation values obtained from small-sized Cap132 872173,273 10397,055 857605,800 896433,588
problems got lower, those of large-sized problems groups got
higher. Therefore, it can be said that this method is relatively Cap133 924420,926 18208,083 896522,713 962142,413
more effective for small-sized problems rather than large- Cap134 996889,000 31413,774 942813,938 1058213,813
sized problems.
The final results obtained using best fit individual instead
TABLE IV of center of mass function for value l=2 are given in Table VI.
EXPERIMENTAL RESULTS OBTAINED USING CENTER FUNCTION FOR L= 3
When examined the mean values under these conditions, it
could not be reached to any optimal result for any problem.
Mean Std Best Worst
However, when examined the best values, it is seen that this
Cap71 934130,279 2157,679 932615,750 939626,575 method reached to the optimal values for small-sized and
Cap72 981160,794 3084,162 977799,400 987144,550 medium-sized problems problem groups but not large-sized
problems. Besides, when looked at the mean values while this
Cap73 1012575,554 2165,025 1010641,450 1017544,888
method obtained very near-optimal results for small-sized and
Cap74 1041407,615 5023,067 1034976,975 1048963,413 medium-sized problems problem groups, it did not obtain any
Cap101 801790,383 4102,594 796648,438 810451,925 sufficient solution for large-sized problems. Therefore, for
small-sized and medium-sized problems, it can be stated that
Cap102 863112,714 6553,889 854704,200 874035,350
this method is relatively more competitive.
Cap103 902364,675 9815,888 893782,113 922215,888
Cap104 947196,955 18341,711 928941,750 987505,475
Cap131 823123,130 18485,441 799291,000 852911,088
Cap132 910805,458 21907,858 859028,188 943963,650
Cap133 972870,255 15372,560 942524,650 1009722,438
Cap134 1055681,507 26056,078 989378,975 1103422,350
____________________________________________________________________________________________________________
294
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
295
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[5] M. Maric, "An Efficient Genetic Algorithm for Solving the [23] J. Barcelo, A. Hallefjord, E. Fernandez, and K. Jornsten,
Multi-Level Uncapacitated Facility Location Problem," "Lagrangean Relaxation and Constraint Generation Procedures
Computing and Informatics, vol. 29, pp. 183-201, 2010. for Capacitated Plant Location-Problems with Single
[6] H. Topcuoglu, F. Corut, M. Ermis, and G. Yimaz, "Solving the Sourcing," Or Spektrum, vol. 12, pp. 79-88, 1990.
uncapacitated hub location problem using genetic algorithms," [24] L. L. Gao and E. P. Robinson, "A Dual-Based Optimization
Computers & Operations Research, vol. 32, pp. 967-984, Apr Procedure for the 2-Echelon Uncapacitated Facility Location
2005. Problem," Naval Research Logistics, vol. 39, pp. 191-212, Mar
[7] K. S. Al-Sultan and M. A. Al-Fawzan, "A tabu search approach 1992.
to the uncapacitated facility location problem," Annals of [25] J. E. Beasley, "Or-Library - Distributing Test Problems by
Operations Research, vol. 86, pp. 91-103, 1999. Electronic Mail," Journal of the Operational Research Society,
[8] M. H. Sun, "Solving the uncapacitated facility location vol. 41, pp. 1069-1072, Nov 1990.
problem using tabu search," Computers & Operations
Research, vol. 33, pp. 2563-2589, Sep 2006.
[9] F. Altiparmak and E. Caliskan, "An Ant Colony Optimization
Algorithm for the Uncapacitated Facility Location Problem,"
Proceedings of the 38th International Conference on
Computers and Industrial Engineering, Vols 1-3, pp. 553-560,
2008.
[10] M. Sevkli and A. R. Guner, "A continuous particle swarm
optimization algorithm for uncapacitated facility location
problem," Ant Colony Optimization and Swarm Intelligence,
Proceedings, vol. 4150, pp. 316-323, 2006.
[11] D. Z. Wang, C. H. Wu, A. Ip, D. W. Wang, and Y. Yan,
"Parallel Multi-Population Particle Swarm Optimization
Algorithm for the Uncapacitated Facility Location Problem
using OpenMP," 2008 Ieee Congress on Evolutionary
Computation, Vols 1-8, pp. 1214-1218, 2008.
[12] S. Saha, A. Kole, and K. Dey, "A Modified Continuous
Particle Swarm Optimization Algorithm for Uncapacitated
Facility Location Problem," Information Technology and
Mobile Communication, vol. 147, pp. 305-311, 2011.
[13] Y. Watanabe, M. Takaya, and A. Yamamura, "Fitness Function
in ABC Algorithm for Uncapacitated Facility Location
Problem," Information and Communication Technology, vol.
9357, pp. 129-138, 2015.
[14] M. S. Kiran, "The continuous artificial bee colony algorithm
for binary optimization," Applied Soft Computing, vol. 33, pp.
15-23, Aug 2015.
[15] M. H. Kashan, N. Nahavandi, and A. H. Kashan, "DisABC: A
new artificial bee colony algorithm for binary optimization,"
Applied Soft Computing, vol. 12, pp. 342-352, Jan 2012.
[16] A. Kaveh and S. Talatahari, "Size optimization of space trusses
using Big Bang-Big Crunch algorithm," Computers &
Structures, vol. 87, pp. 1129-1140, Sep 2009.
[17] O. K. Erol and I. Eksin, "A new optimization method: Big
Bang Big Crunch," Advances in Engineering Software, vol. 37,
pp. 106-111, Feb 2006.
[18] A. Kaveh and S. Talatahari, "Optimal design of Schwedler and
ribbed domes via hybrid Big Bang-Big Crunch algorithm,"
Journal of Constructional Steel Research, vol. 66, pp. 412-419,
Mar 2010.
[19] C. V. Camp, "Design of space trusses using big bang-big
crunch optimization," Journal of Structural Engineering-Asce,
vol. 133, pp. 999-1008, Jul 2007.
[20] M. S. Daskin, L. V. Snyder, and R. T. Berger, "Facility
location in supply chain design," Lehigh University, Working,
pp. 03-010, 2003.
[21] G. Cornuejols, G. L. Nemhauser, and L. A. Wolsey, "The
uncapacitated facility location problem," Lecture Notes in
Artificial Intelligence, vol. 1865, pp. 171, 1990, 1990.
[22] K. Holmberg, "Exact solution methods for uncapacitated
location problems with convex transportation costs," European
Journal of Operational Research, vol. 114, pp. 127-140, Apr 1
1999.
____________________________________________________________________________________________________________
296
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In today industry, manufacturing become big and same time. Thus the robots in the facility can be observed in
serial as it never been before thanks to the autonomy robots. one screen and machines which could lead to accidents can be
Hitches on such autonomy systems used in industrial production noticed early. The stitched image can also be used for image
may cause production delaying. In this study, it is aimed to obtain processing in cyber layer.
alive bird's eye view map of full system in order to monitor
There are studies about image mosaicing in literature [1-5].
manufacturing robots at production facilities that are big and
impossible to be monitored with only one camera. Finding the One of the studies, Divya et al. [1], implemented weighted
similar scenes of input images, estimation of homography, average method in addition to the methods used in image
warping and blending operations are applied respectively in order mosaicing such as feature extraction, cluster analysing,
to mosaic the images by twos. Thus the robots in the facility can dynamic programming to obtain wide angle panorama.
be observed in one screen. With observation of the obtained Although the study is effective for illuminated images, it is not
images, faults on cyber-physical systems that may cause damage useful for dark images. Another study about this subject is Lin’s
in machines which are not cheap can be handled beforetime. study [2]. Lin et al. use image mosaicing in order to obtain
cylindrical panoramic image. Scale Invariant Feature
Keywords— Condition monitoring, Multi robots, Production lines, Transform (SIFT) and Random Sample Consensus (RANSAC)
Image processing, Image mosaicing are used in this study. Patel et al. [3] employ Speeded-Up
Robust Features (SURF) for feature extraction and Sum of
I. INTRODUCTION
Squared Differences (SSD) for feature matching. They also
Accuracy and security of cyber-physical systems become used RANSAC to eliminate incorrect matchings. At the end of
crucial with growing up cyber-physical systems in industry. In their study, they compare Harris and SURF and decide that
today industry, manufacturing become big and serial as it never SURF is more effective for multi-imaged image mosaicing
been before thanks to the autonomy robots whose cyber and problems. Huang et al. effort to mosaic frames in videos [4].
physical layers are diverse. Monitoring physical layer from Instead of using all frames in a video, they used the frames that
cyber layer is necessary for link the layers and integrate them contain small piece of other frame to enhance video mosaicing
more tightly. Furthermore, it will ease the control of big performance. In their study, Lee et al. [5] employ SURF for
facilities. Hitches on cyber-physical systems used in industrial feature extraction and Histogram of Gradient (HoG) for
production may cause production delaying. Furthermore, these matching. They also utilize from Approximate Nearest
hitches may cause damage in hardware which are not cheap. So Neighbour (ANN), RANSAC algorithms.
observation of the system and determination of possible
accident are required for deduction the cost of delay in
production and repair. With processing the obtained image by Images that will be stitched
algorithms in cyber layer, the cyber-physical system would be
more integrated and secure.
In this study, it is aimed to obtain alive bird's eye view map Sort the images with the aid of extracted features
of full system in order to monitor manufacturing robots at
production facilities that are big and impossible to be monitored
with only one camera. With the created alive map, it will be Form the stitched image
possible to monitor positions of all robots instant and extract
more detailed information about the facility. It is need that
locate cameras each of whose angel of vision contain the small Apply methods such as noise reduction and
balancing in order to improve quality
piece of neighbour one’s. Finding the similar scenes of input
images, estimation of homography, warping and blending
operations will be applied respectively in order to mosaic the
Obtained image that has high resolution
images by twos. After all steps done, only one stitched image
will be obtained from all these images taken by cameras at the Figure 1. General Image Mosaicing Steps in Literature [6]
____________________________________________________________________________________________________________
297
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
298
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the correct result within the tolerance defined before, the datum
YES
is marked as in-liner. If in-liner data is sufficient to verify the NO
model, the model is added to solution set. This operation is Add the model to solution set
actualized pre-determined iteration times. At the end of the
iterations best model that fits the majority of data with
minimum error is selected as solution. The diagram belongs to Does iteration exceeded
RANSAC algorithm is given by Fig. 4. max iteration number
We aimed to fit a model whose input is the pixel coordinate Terminate the process and
of target interest points and output is the pixel coordinate of evaluate the models in solution set
corresponding reference interest point. The model that help us
to calculate the new position of target image pixels’ in order to Figure 4. Block Diagram of RANSAC Algorithm [22]
superpose these images is called homography matrices.
Homography matrices is 3x3 sized matrices and it is used to
move the points in homogeneous coordinate system. The III. EXPERIMENTAL RESULTS
equation for calculating new position of point and and its In this paper, the image stitching operation is implemented
extended version are given by respectively (1) and (2). on images that belongs to robots at production line. The
diagram of desired system given by Fig. 5. Although the sample
8. : = :′ (1) system shown by Fig. 5 includes 4 cameras with the same
ℎ<< ℎ<= ℎ<> ?@ ?C characteristics, the count of cameras can be increased. A
ℎ=< ℎ== ℎ=> A@ = AC (2) computer with intel i7 6020M CPU, 6 GB ram, and Windows
7 64bit home premium is used to perform the proposed
ℎ>< ℎ>= ℎ>> B@ BC
approach on the sample images belong to production
environment.
where H is homography matrices, P and P’ are former and new
coordinates of pixel in homogenous coordinate system
respectively. To transform 2 sized vector to homogenous
coordinate system, B@ is selected as 1; to transform
homogenous coordinate system to X and Y values (3) is used.
?C AC
?D = , A D = (3)
BC BC
____________________________________________________________________________________________________________
299
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(a) 1st camera of first row (b) 2st camera of first row
(c) 1st camera of second row (d) 2st camera of second row
Fig. 6. Images that are stitched
In order to test the effect of the images order, the four image
given by Fig. 6 are stitched respectively first. The result of this (i) Stitching of (c) and (d)
process is given by Fig. 7. Then the same images are subjected
to the process of stitching by twos. In Fig. 8, e is obtained by
stitching of a and b, f is obtained by stitching of c and d, and
finally g is obtained by stitching of e and f.
____________________________________________________________________________________________________________
300
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE II. THE PERFORMANCES OF STITCHING BY ORDER [2] M. Lin, G. Xu, X. Ren, and K. Xu, “Cylindrical Panoramic Image
WITH FULL SEARCH AREA Stitching Method Based On Multi-cameras,” The 5th Annual IEEE
International Conference on Cyber Technology in Automation, Control
Process in and Intelligent Systems, Shenyang, China, pp. 1091-1096, June 2015.
Size Time (ms)
Fig. 7. [3] A. Laraqui, A. Baataoui, A. Saaidi, A. Jarrar, M. Masrar, and K. Satori,
e 560 x 189 65 “Image mosaicing using voronoi diagram,” Multimedia Tools and
f 560 x 310 92 Applications, pp. 1-27, 2016.
[4] C. M. Huang, S. W. Lin, and J. H. Chen, “Efficient Image Stitching of
g 560 x 310 123 Continuous Image Sequence with Image and Seam Selections,” IEEE
Total 560 x 310 280 Sensors Journal, vol. 15, pp. 5910-5918, 2015.
[5] S. Lee, Y. Park, and D. Lee, “Seamless Image Stitching Using Structure
Deformation with HoG Matching,” International Conference on
Information and Communication Technology Convergence (ICTC), Jeju,
TABLE III. THE PERFORMANCES OF STITCHING BY TWOS South Korea, pp. 933-935, Oct. 2015.
WITH LIMITED SEARCH AREA [6] R. Abraham, and P. Simon, “Review on Mosaicing Techniques in Image
Processing,” International Conference on Advanced Computing &
Process in Communication Technologies, Rohtak, India, pp. 63-68, April 2013.
Size Time (ms)
Fig. 8. [7] Z. Qui, P. Shi, X. Jiang, D. Pan, C. Feng, and Y. Sha, “Image Stitching
h 560 x 189 15 and Ghost Elimination Based on Shape-Preserving Half-Projective
i 560 x 189 16 Warps,” International Conference on Information and Automation,
Lijiang, China, pp. 2610-2615, Aug. 2015.
j 560 x 310 28 [8] A. Laraqui, A. Baataoui, A. Saaidi, A. Jarrar, M. Masrar, and K. Satori,
Total 560 x 310 59 “Image mosaicing using voronoi diagram,” Multimedia Tools and
Applications, pp. 1-27, 2016.
[9] M. Baygin, and M, Karakose, “A new image stitching approach for
resolution enhancement in camera arrays,” 9th International Conference
IV. CONCLUSION on Electrical and Electronics Engineering (ELECO), Bursa, Turkey, pp.
It is very crucial to monitor physical elements especially for 1186-1190, Nov. 2015.
[10] P. M. Jain, and V. K. Shandliya, “A Review Paper on Various
cyber-physical systems that is trend of the industry of today, Approaches for Image Mosaicing,” International Journal of
which is called industry 4.0. In this paper, it is proposed that Computational Engineering Research, vol. 3, pp. 106-109, 2013.
use image mosaicing to monitor conditions of multi robots at [11] H. Joshi, and K. Sinha, “A Survey on Image Mosaicing Techniques,”
production lines and the proposed method is tested by International Journal of Advanced Research in Computer Engineering
& Technology (IJARCET), vol. 2, pp. 365-369, 2013.
experimental. In order to test the method 4 images that belongs [12] D. Ghosh, and N. Kaabouch, “A Survey on Image Mosaicing
the same science are used, and no one of them contains more Techniques,” Journal of Visual Communication and Image
than 1/3 of another one in spite of the possible deflections. With Representation, vol. 32, pp. 1-11, 2016.
the knowledge of this, only a part of image is used to search [13] I. K. Sarangi, and S. Nayak, “Image Mosaicing of Panoramic Images,”
Bachelor Thesis, National Institute of Technology, Rourkela, 2014.
interest points instead of full image. Finding interest points,
[14] M. Karakose, M. Baygin, “Image processing based analysis of moving
feature point extraction, matching, warping and blending shadow effects for reconfiguration in PV arrays,” IEEE International
operations are used respectively to stitch images. Since the Energy Conference (ENERGYCON), Cavtat, Croatia, pp. 683-687, May.
order of images that are stitched are certain, sorting of images 2014.
is not required for such camera array applications. The images [15] V. K. S. Prathap, S. A. K. Jilani, and P. R. Reddy, “A Critical Review
on Image Mosaicing,” International Conference on Computer
are stitched by order and by twos respectively. The effect of Communication Informatics (ICCCI), Coimbatore, India, pp. 1-8, Jan.
image order is demonstrated by results given by table 2 and 2016.
table 3. The results show that stitching by twos outperforms [16] J. Krizaj, V. Struc, and N. Pvesic, “Adaptation of SIFT Features for
stitching by order for the camera arrays which occurred from Robust Face Recognition,” International Conference on Image Analysis
and Recognition (ICIAR), Povoa de Varzim, Portugal, pp. 1-10, June
NxN camera, N is even number. Furthermore, if N is not even
2010.
or the size of camera array is not NxN, a sub array that provides [17] H. Yetis, M. Baygin, and M. Karakose, “A New Micro Genetic
the constraints and the operation is applied for sub array. After Algorithm Based Image Stitching Approach for Camera Arrays at
the remains may be stitched by orders. Little time consumption Production Lines,” 5th International Conference on Manufacturing
of this approach make it possible to be used in camera array Engineering and Process (ICMEP), Istanbul, Turkey, in press, 2016.
[18] D. G. Lowe, “Distinctive Image Features from Scale-Invariant
problems. The feature-based method makes the solution more Keypoints,” International Journal of Computer Vision, vol. 60, pp. 91-
reliable in case of changing in position, angle or illumination. 110, 2004.
Consequently, it is possible to get high quality reliable image [19] S. Mistry, and A. Patel, “Image Stitching using Harris Feature Detection,”
using low quality cameras thanks to the method. Monitoring the International Research Journal of Engineering and Technology (IRJET),
vol. 3, pp. 1363-1369, 2016.
condition of robots at production line, which is a requirement [20] A. Levin, A. Zomet, S. Peleg, and Y. Weiss, “Seamless Image Stitching
for today’s industry facilities, in a screen is achieved by the in the Gradient Domain,” Computer Vision (ECCV), pp. 377-389, 2004.
study. [21] Y. Santur, M. Karakose, I. Aydin, E. Akin, “IMU based adaptive blur
removal approach using image processing for railway inspection,”
International Conference on Systems, Signals and Image Processing
(IWSSIP), Bratislava, Slovakia, pp. 1-4, May 2016.
REFERENCES [22] D. K. Jain, G. Saxena, and V. K. Singh, “Image Mosaicing Using Corner
[1] G. Divya, and C. Chandrasekhar, “Image Mosaicing Technique for Wide Techniques,” International Conference on Communication Systems and
Angle Panorama,” TELKOMNIKA Indonesian Journal of Electrical Network Technologies, Rajkot, India, pp. 79-84, May 2012.
Engineering, vol. 15, pp. 420-429, 2015.
____________________________________________________________________________________________________________
301
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
302
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
303
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I
B. Particle Advection
L UCAS -K ANADE O PTICAL F LOW A LGORITHM
Optical flow provides the data related to where a pixel will
Input: Set of input image frames S be heading to on the next frame. Particle Advection, on the
Output: Optical flow vectors [u,v] other hand, is a method used in order to observe (understand)
how a pixel moves throughout the video.[19][20][21]
Begin:
A= set 2x2 zero matrix At this level, particles that are located on every pixel in the
b= set 2x1 zero matrix first frame of the video are being directed by the time domain
d= set 2x1 zero matrix pixel level 3D movement data that is obtained by optical flow.
NWS= floor(Neighbor Window Size)-1
(Ix,Iy)=gradient(1th frame) At the end of the process, the time domain 3D optical flow
It=1th frame-2th frame data is obtained as 2D movement data. This process is called
particle advection and the distance covered by each particle
for i=1+NWS to X size of(1th frame)
for j=1+NWS to Y size of(1th frame) is called particle trajectories. [22][23][24][25][26][27]
A grid is located on 1th frame in video sequence, namely
A= set 2x2 zero matrix each pixel in the first frame and the particles are moved
b= set 2x1 zero matrix
according to optical flow data. Ideally, a grid is made up
for m=i-NWS to i+NWS as N=Width X Height. Particle advection can be formulized
for n=j-NWS to j+NWS according to notation expressed above.
A(1,1)=A(1,1)+Ix(m,n)*Ix(m,n)
A(1,2)=A(1,2)+Ix(m,n)*Iy(m,n) xi (t + 1) = xi (t) + u[xi (t), yi (t)] (6)
A(2,1)=A(2,1)+Iy(m,n)*Ix(m,n)
A(2,2)=A(2,2)+Iy(m,n)*Iy(m,n)
yi (t + 1) = yi (t) + v[xi (t), yi (t)] (7)
b(1,1)=b(1,1)+ Ix(m,n)*It(m,n)
b(2,1)=b(2,1)+ Iy(m,n)*It(m,n) C. Clustering
end Differentiating multiple actions on an image from each
end other successfully is directly related to the success of the
cluster process. The particles animated with the help of
d=PseudoInverse(A)*(-b)
u(i,j)=d(1,1) Particle advection process infact carry the motion data and
v(i,j)=d(2,1) at the end of the process particles representing any action
gathers at a cluster point. It is expected from the system to
end
end differentiate these clusters from each other. We can think that
each cluster represents an action. In this study, we tried to
obtain clusters using mean-shift method.[28][29]
Some places on the image might not move and therefore
yet the bottlenecks or small movements might not be noticed. the particles on these pixels do not move. In this case,
There is a trade off in question. Therefore, the size of the particles not moving or moving at a value lower than the
neighbouring window is another problem to think about. threshold value and clusters emerged under a definite number
When the equation (4) is solved using minimum least of particles are accepted as noise and they are not included
squares method: in the evaluation. By this way, only clusters that have the
capability to represent motion or define a significant motion
can be obtained.
(AT A)d = AT b
D. Define as Parabola
2x2 2x1 2x1
In this step, the moving areas in the image will be defined
as parabola. In this way, these results will be helpful in terms
ΣIxIx ΣIxIy u ΣIxIt of determining moving areas and understanding the crowd
=− (5)
ΣIxIy ΣIyIy v ΣIyIt behaviour.
There are four basic steps to be taken in obtaining optical Determining as moving parabola can be as follows; par-
flow using Lucas-Kanade method. ticles representing an action or behaviour move as a group
• Noise reduction is applied in frames with particle advection. This information can be expressed as
• Ix, Iy, and It are calculated a curve using least squares method according to the location
• u and v values are found by calculating the equation of each particle has along the motion.
(5).
• Due to the fact that optical flow values are independent f (x) = ax2 + bx + c (8)
from each other and that they might show differences
from the general flow, the resulting values (u,v) are
line if a u 0
applied a median filter. a=
circle others
____________________________________________________________________________________________________________
304
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 3. Every color show different event in the image and will be defined as parabola. These results are a (8th equation) for every quadratic equation
a)blue = 0.000786, green = -0.033997 b)blue = 0.001927, green = -0.010793, red = 0.000681 c)blue = 0.013857, green = 0.040634 d)blue = 0.105957,
green = -0.092312, red = 0.032337 e)blue = -0.003065, green = 0.002836, red = -0.079450 cyan = 0.024176, magenta = 0.006983 f)blue = 0.019147,
green = 0.000424
____________________________________________________________________________________________________________
305
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[6] B.T.Morris, M.M.Trivedi, A Survey of Vision-Based Trajectory Learn- density function, with applications in pattern recognition,Information
ing and Analysis for Surveillance IEEE Transactions on Circuits and Theory, IEEE Transactions on, vol. 21, no. 1, pp. 32–40, 1975.
Systems For Video Technology [29] D.Comaniciu, P Meer, Mean Shift: A Robust Approach Toward Feature
[7] M. Hu, A. Saad and M. Shah,Learning Motion Patterns in Crowded Space Analysis, IEEE Transactions on Pattern Analysis And Machine
Scenes Using Motion Flow Field, in 19th International Conference on Intelligence, vol.24, no.5, May 2002.
Pattern Recognition, ICPR, 2008.
[8] A.Dehghan M.M.Kalayeh, Understanding Crowd Collectivity: A
Meta-Tracking ApproachIEEE International Conference on Computer
Vision and Pattern Recognition Workshop(CVPRW) 2015
[9] Bolei Zhou, Xiaoou Tang, and Xiaogang Wang, Measuring Crowd
Collectiveness, IEEE Conference on Computer Vision and Pattern
Recognition (CVPR) 2013.
[10] B. Solmaz, B. E. Moore and S. Mubarak, Identifying Behaviors in
Crowd Scenes Using Stability Analysis for Dynamical Systems,IEEE
Transactions on Pattern Analysis And Machine Intelligence, 2012.
[11] M. Rodriguez, J. Sivic, I. Laptev and J.-Y. Audibert, Data-driven
Crowd Analysis in Videos, International Conference on Computer
Vision (ICCV), 2011.
[12] R. Mehran, A. Oyama and M. Shah, Abnormal Crowd Behavior
Detection using Social Force Model, in Conference on Computer
Vision and Pattern Recongition,CVPR, 2009.
[13] B. Zhou, X. Wang and X. Tang, Understanding Collective Crowd
Behaviors: Learning a Mixture Model of Dynamic Pedestrian-Agents,
IEEE Conference on Computer Vision and Pattern Recognition
(CVPR) 2012.
[14] J.Shao, C.C.Loy X.Wang, Scene-Independent Group Profiling in
Crowd, IEEE Conference on Computer Vision and Pattern Recognition
(CVPR), 2014
[15] S. S. Beauchemin and J. L. Barron, The Computation of Optical Flow,
London, Ontario, Canada: University of Western Ontario, 1995.
[16] J. L. Barron, D. J. Fleet and S. S. Beauchemin, Performance of Optical
Flow Techniques, International Journal of Computer Vision,IJCV, vol.
12:1, pp. 43-77, 1994.
[17] K. R. Aires, A. M. Santana and A. A. D. Medeiros, Optıcal Flow
Usıng Color Informatıon:Prelımınary Results, in ACM Symposium
on Applied Computing, 2008.
[18] JY. Bouguet, Pyramidal Implementation of the Lucas Kanade Feature
Tracker Description of the algorithm, Intel Corporation, Microproces-
sor Research Labs, 2000.
[19] S. Wu, B. E. Moore and M. Shah, Chaotic Invariants of Lagrangian
Particle Trajectories for Anomaly Detection in Crowd Scenes, IEEE
International Conference on Computer Vision and Pattern Recognition
(CVPR), 2010.
[20] P. Sand and S. Teller, Particle Video: Long-Range Motion Estimation
using Point Trajectories, IEEE International Conference on Computer
Vision and Pattern Recognition (CVPR), 2006.
[21] R. Mehran, B. E. Moore and M. Shah, A Streakline Representation
of Flow in Crowded Scenes, in European Conference on Computer
Vision (ECCV), 2010.
[22] A. Saad and M. Shah, A Lagrangian Particle Dynamics Approach for
Crowd Flow Segmentation and Stability Analysis, IEEE International
Conference on Computer Vision and Pattern Recognition (CVPR),
2007.
[23] S. Wu, O. Oreifej and M. Shah, Action Recognition in Videos Acquired
by a Moving Camera Using Motion Decomposition of Lagrangian
Particle Trajectories, International Conference on Computer Vision
(ICCV), 2011.
[24] K. K. Reddy, Action Recognition Using Particle Flow Fields, Orlando,
Florida: Phd Thesis, 2012.
[25] B. Zhou, X. Tang, and X. Wang, Coherent Filtering :Detecting
Coherent Motions from Crowd Clutters, 12th European Conference
on Computer Vision (ECCV) 2012.
[26] B.Zhou, X. Wang, and X.Tang, Random Field Topic Model for
Semantic Region Analysis in Crowded Scenes from Tracklets, IEEE
Conference on Computer Vision and Pattern Recognition (CVPR)
2011.
[27] B.T.Morris, M.M.Trivedi, Trajectory Learning for Activity Under-
standing:Unsupervised, Multilevel, and Long-Term Adaptive Ap-
proach, IEEE Transactions on Pattern Analysis And Machine Intel-
ligence, VOL. 33, NO. 11, NOVEMBER 2011.
[28] K Fukunaga and L. Hostetler, The estimation of the gradient of a
____________________________________________________________________________________________________________
306
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— This paper presents an expert system for induction motor increased stresses, energy losses, cause added wear, increase
fault detection based on vibration analysis by using corvid expert bearing loads, induce fatigue, create passenger discomfort in
system. Vibration signals of induction motors on four different vehicles, and absorb energy from the system.
actuating mechanism are collected with a specific vibration The amplitude of the vibration can be measured as
measuring device. The device evaluates the values with three
displacement, acceleration or velocity. Displacement is a
harmonics in frequency domain. Expert system provides the
recommendations as maintenance activity or the reason of the distance for moved object. Its measurement unit is mm
vibration by using vibration values. This system is tested and (millimeter). Acceleration is a difference the rate of the velocity
validated on four type of actuating mechanisms. Obtained results and is measured as mm/s².
show that this system can detect faults in early stages with high The amplitude of the vibration can be measured as
accuracy and reliability. Thus, it provides malfunction and failure displacement, acceleration or velocity. Displacement is a
prevention and improves overall performance and efficiency of distance for moved object. Its measurement unit is mm
industrial systems. (millimeter). Acceleration is a difference the rate of the velocity
and is measured as mm/s².
Keywords— fault detection, induction motor, vibration analysis, data
analyse, expert systems.
I. INTRODUCTION
Induction motors play an important role as prime movers in
manufacturing, process industry and transportation due to their
reliability and simplicity in construction. Although induction
motors are reliable, the possibility of unexpected faults. The
issue of robustness and reliability is very important to guarantee
the effective operational condition. Therefore, condition
monitoring of induction motors has received considerable
attention in recent years. Early fault diagnosis and condition
monitoring can reduce the consequential damage with
unexpected production stops. Faults can occur in the stator,
rotor, bearing, or the external systems connected to the
induction motor. Because of this reason, fault diagnosis of Figure 1 : Amplitude – Frequency Chart
engineering plant has increased recently.
Displacement accentuates the low frequencies. Acceleration
Expert Systems contain specialized knowledge elicited
accentuates the higher frequencies. Velocity is consistent
from a domain expert usually in the form of IF-THEN rules.
across a larger range of frequencies as presented Figure 1.
Expert systems are widely used in fault detection and diagnosis
Vibration data can be collected through the vibration sensors.
applications [1]. The knowledge engineer usually extracts rules
Vibration sensors can be classified as accelerometers, velocity
for the given domain, in this case, fault diagnosis system for
sensors, capacitive and eddy current sensors and laser
induction motors, and the inputs of expert system are the
displacement sensors. All these sensors have some advantages
measurements taken by sensitive instruments, i.e. sensors. The
and disadvantages according to the application. After the values
expert system then proceeds on to reasoning with rules and
are collected vibration analysis should take place. Vibration
infer meaningful recommendations. The present research is
analysis consist of some logical steps. These are detection,
based on using motor vibration data measured with a
analysis and correction. Detection should be implemented by
professional device and use these results obtained from the
the devices or sensors then analysis occurs by software program
device for diagnosing faults in an induction motor.
or just like ours with an expert system. The last step is the
II. VIBRATION ANALYSIS correction through the analysis. . A systematic vibration
analysis can then be carried out to identify the more common
Vibration is the cyclic or oscillating motion of a machine or machinery problems, including: Unbalance, Misalignment,
machine component from its position of rest. Vibration is a Looseness, Defective Bearings, Resonance, Eccentricity, Worn
repeated action and is continuous. Most vibrations are Gears, Motor Electrical Problems, Drive Belt Problems,
undesirable in machines and structures because they produce
____________________________________________________________________________________________________________
307
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Distortion (Soft-Foot & Piping Strain). Diagnostic Charts are When the IF part is true, the statements in the THEN part are
implemented to make rules for expert system. also considered true.
A sample dialog for Motor – Belt – Fan Mechanism is as
follows:
IF:
Choose the actuating mechanism: Motor – Belt – Fan
AND : Choose measuring point : Horizontal Vibration
AND : Is there any pick vibration value before the
[Vibration Value1X], Evet
AND : [Vibration Value2X] = 0
Fig. 1 : Unbalance Diagnostic Chart for Fan Mechanism AND : Vibration value3X] = 0
Mass unbalance will be in phase and steady as shown in THEN:
Figure 2. Amplitude due to unbalance will increase by the ‘’There is a misalignment at the center of motor side.
square of the speed below first rotor critical (a 3X speed Maintenance is necessary. ‘’
increase = 9X higher vibration). 1X RPM always present and
IV. EXPERIMENTS AND RESULTS
normally dominates spectrum.
Expert system for induction motor fault detection presented
III. EXPERT SYSTEM DESIGN AND IMPLEMENTATION in this paper implements all previously stated phases of the
Expert systems are computer programs that emulate the detection process. The measurements have been taken on the
interaction that a person would have asking a human expert for Motor- Belt – Fan mechanism by vibration device as presented
advice or a recommendation. Most decision-making processes in Figure 3. There is a peak value on the first harmonic as
can be broken down into many small parts. The human expert shown in Figure 4 that means there is a misalignment on motor
often makes decisions almost automatically and does not side as recommended at expert system forward chaining rule
consciously think about each small step to solve problems or structure. The system has recommended to the maintenance
reach conclusions. However, they become apparent when the team before the problem has occurred that means the system
reason for the decision is explained to someone else, or the has prevented the production loss and has guaranteed
decision process is taught to others. A decision may be based production stability.
on various facts that individually would not be conclusive, but
when combined together lead to a specific decision or diagnosis.
The kernel of an expert system has two main components.
Namely the knowledge base and the inference engine. The
knowledge base contains knowledge about the expert’s domain.
It may be represented by simple facts, or by more complex
representations like frames. There are also rules which
explicitly represent the expert’s skills or knowledge about the
domain under consideration. The expert system uses this
knowledge by exploiting the second main component, that is
the inference engine which has several roles including
determining how the system reasons using the IF-THEN rules
in the knowledge base. Once the knowledge base is built, the
ES can begin making inferences. The most common forms of
inferencing are forward and backward chaining. The process of
moving forwards from known facts to conclusions that follow Figure 3 : Motor – Belt – Fan Mechanism
them is called forward chaining. Alternatively, the process of
working backwards from an hypothesis to known facts that
support it, is called backward chaining.
Various expert system building tools or shells exist to
facilitate and speed up the development process. In the present
study, the new generation Exsys Corvid tool has been utilized
[3]. Exsys tools have traditionally been used a pure rule-based
approach to logic and this has proven to be extremely effective
with heuristic rules [2]. There are many ways of describing the
heuristics for a decision-making process, but the one that has
proven the most effective and efficient is the IF/THEN rule.
This is a rule where there is an IF part that can be tested to be
true or false based on the data for a specific case or situation. Figure 4 : Vibration Value of Motor – Belt – Fan Mechanism
____________________________________________________________________________________________________________
308
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
309
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In the race of smartphone operation systems, IOS and growth from 2015 [2]. Also, worldwide smartphone sales to
Android seem to have the big part of pie. Both platforms have end-users by operating system in 2014 and 2013 are shown in
grown more enterprise-friendly since about one decade. Their Table I [3].
adaptable functionalities make people' lives easy and give them a
reputation in the current competitive technology world. We all
have our personal thoughts it comes to features, usability and
design. However, what about security? Mobile devices,
smartphone operation systems run on, hold valuable, sensitive
and classified information or content. So, that increases their TABLE I
attractiveness as targets for cybercriminals. The security of these SMARTPHONE SALES IN 2014 AND 2013
devices is a growing concern and focus point for smartphone
users. Eventually, the security technology of the smartphones OS 2014 (Units) 2013 (Units)
becomes one of the prime research and these smartphone Android 1,004,675 761,288
vendors have increasingly focused on security in their design IOS 191,426 150,786
efforts. In this paper, factors that influence security within IOS Windows 35,133 30,714
and Android are studied to promote the discussion. Security BlackBerry 7,911 18,606
technologies of IOS and Android are briefly presented. And, Other OS 5,745 8,327
varies factors are considered such as methods of application Total 1,244,890 969,721
distribution, reduced attach surface, privilege separation,
permission-based access control, sandboxing, data encryption, As mobile platforms become increasingly popular, so do
data execution prevention and address space layout the incentives for attackers, especially when mobile payment
randomisation, geo-location and auto-erase. Then, brief
transactions are projected to reach almost US$630 billion by
information is given about malicious apps. Lastly, discussion is
concluded to answer that tight question in the light of security 2014. Recent security surveys describe the rapidly increasing
models investigation and evidences collected from current life. number and sophistication of mobile attacks. Mobile devices’
prevalence and mobile threats’ rapid growth have resulted in a
Keywords— mobile security, mobile device, mobile OS, IOS, shortage of mobile-security personnel [4].
Android
The storing of sensitive personal data in the smartphones
I. INTRODUCTION and the increasing popularity of smartphones, lead them to
Mobile devices are rapidly becoming the platform of choice attract the hackers' attention [5]. Cybercriminals have found
for consumers and businesses. There is a wide range of many sophisticated ways to perform malicious activities on
functionalities provided to the users by the mobile devices mobile devices. Cybercrimes such as identity theft,
such as fast connection to social platforms, personal data information theft, distribution of malwares, and financial
storing, financial processes, web browsing and tons of fraud have become a real threat to individuals, organizations,
services. Therefore, the smartphones sales achieve high and service providers. Nevertheless, there are defences that
number of records worldwide and going on to suppress other can be put in place to minimize and mitigate these threats [6].
electronic technologies like notebooks and tablets [1]. On the
other hand, while the number of mobile devices increased, the Although academic research in mobile device and
security threats such as privacy violations, malicious code and smartphone security has been conducted for several years now,
exploits in smart environment increased. According to a it’s only recently— due to the explosion of mobile device
recent report from Gartner, global smartphone sales are deployment in the enterprise environment —that those
estimated to reach 1.5 billion units in 2016, a 7 per cent
____________________________________________________________________________________________________________
310
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
research activities have truly focused on solving real-world minimizing the amount of code an attacker can access,
security challenges [7]. especially remotely [10].
For example, flash has a history of security vulnerabilities
In this study, Security technologies of IOS and Android are and not available on IOS. Android is still going on support it.
briefly presented and considered. Then, brief information is In addition, there are vulnerability occasions reported about
given about malicious apps and exploits. Lastly, discussion is PDF viewer on Android such as some script in PDF performs
concluded to answer that tight question in the light of security dropper functionality. But, IOS render PDF files natively,
models investigation and evidences collected from current life. only some features of it parsed. That reduces the number of
potential vulnerabilities.
II. EVALUATION OF SECURITY FACTORS Also, security problem of WebView has been studied
In this part of study, security factors that have chosen for before and still exists [11]. Interestingly, research has shown
comparison are evaluated. that the WebView technology can be further exploited to
break the sandbox protection mechanism in the underlying
A. Application Distribution mobile OS. More mobile applications that integrate WebView,
App Store is the only place to distribute apps. Therefore, the broader the attack surface will become [7]. This problem
the apps must be uploaded that official marketplace for public. is closely pursuing by these tow big companies.
Before this process, developer must register with Apple and
gets aware of license conditions. After registration, digital C. Privilege Separation
certificate can be obtained to sign app for distribution. Aim of Both IOS and Android kernel implements a privilege
digitally signing is to guaranty that both the identity of the separation model like on UNIX system. In privilege
app's developer and the app are not modified, and that app is separation model, mobile operation system requires every
belongs to intended developer. Second important step to application run with its own group and users. Thus, this
publish app is vetting process. Vetting process is process for ensures that applications have no permission to access other
checking application for privacy and security violations [8]. applications. The most important system processes run as a
That process can be taken one or more weeks. Also, some root that is most important privilege. Usually, phone
other issues that may get developer into trouble are explored manufacturer use this root right for system apps. Applications
during this process. For example, the app that works with you are downloading from store will run as mobile. An
public web site in real time needs to be confirmed its own End attacker who gets full control of the app such as the PDF
User License Agreement (EULA) by users and web site viewer will be constrained by the fact the code she is
moderator must check and decide the content that send over executing will be running as mobile.
that app to publish or not.
For Android, users need to create account on Google Play D. Permission-Based Access Control
as a publisher like Apple. But signing and vetting processes in IOS permission mechanism is quite different from Android.
Google Play are different! Android does not require the When application requires use of some other application or
applications’ developers to sign apps with Google-issued service feature at runtime, IOS asks the user about allowing
signing certificates. Actually, developers who want to post requested access to intended resources. In IOS, predefined
their apps on Google are able to create as many self-signing limited sets of permissions are given to third party apps. And,
certificates as they want without being monitored by Google accessing to many of devices’ subsystems is blocked. In that,
[9]. That leads to some security vulnerability such as attackers IOS don’t leave some important security sensitive decisions to
can use famous company name to fasten their malicious app user.
distribution. Also, vetting process in android is shallow and For Android, inside the application manifest there is a static
automatic means not checked by human beings. This situation list of permissions that is shown the user during installation
gives easy way to put malicious apps on this trusted domain process. Then, users decide to accept or deny of these
play.google.com. Also, there is more important issue here permission requests. By the way, users cannot partially select
Google Apps (APK Files) can be published over anonymous permissions, in that all or not. If users don’t allow this
web site or over e-mail. That totally leaves the decision to permission request, installing will be stopped.
consumers for installation. But, one central problem is the
inability of users to make good security choices and
awareness [7]. E. Sandboxing
IOS and Android sandboxing mechanism sometimes called
isolation almost similar where each application is separated
B. Reduced Attach Surface from other applications and system’s kernel. Isolation limits a
The attack surface is the code that processes attacker- process’s ability to access sensitive data or system resources
supplied input. If app has vulnerability in some code, and from another process. Still, there is slight difference between
either the attacker can't reach it, an attacker cannot base an them. In Android, each application is given permissions to
exploit on this vulnerability. Therefore, a key practice is access certain resources, and the isolation system prevents
____________________________________________________________________________________________________________
311
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
accessing resources beyond the approved permissions. In IOS, prevent data from execution. There are one way to bypass
developer himself can’t define permission mechanism. It is DEP is to use ROP (Return-Oriented Programming). In ROP,
mean that there are certain access rules defined by IOS that attackers harness existing valid code in memory to perform
takes over some responsibility from end users shoulders. IOS desired actions. However, to do this, they need to know where
isolates some apps from the in-out email boxes and the SMS the code segments they want to reuse are located. Address
of a device. For instance, apps cannot send SMS without the space layout randomization (ASLR) makes this difficult by
users' involvement. Of course, apps can still access some randomizing the location of objects in memory. Also, writing
resources on the system. Thus, an attacker can use a malicious large payloads in ROP is very time-consuming and complex.
application to steal private information such as the device's This makes exploitation of IOS more difficult than just about
unique ID, sending email spam, or performing an off service any other platform [10].
(DoS) attack on the device [8]. In Android, some ways are In Android, before version 4.0 there is not this kind of
also open for attackers. For example, app list can be retrieved special security layer. On the other hand, Android leaned on
by another app using certain permissions. And, files on SD code signing and memory management unit to restrain
(Secure Digital) card can be read and modified by an app attackers from reaching intendant memory location. But, after
without any restriction. version 4.0 android implemented DEP and ASLR in its system
core.
F. Data Encryption
The features of this dimension interest about how to protect
the user data. In IOS, generally there are two-protection H. Geo-Location and Auto-Erase
mechanisms that are hardware security and data protection. Geo-Location is very useful feature to locate your phone in
Every IOS device has a dedicated AES 256 crypto engine case of lost. Apple as a feature of its operating system and
implemented in silicon using UID and GID as key. Integrating accompanying online service provides this feature. Our
these keys into the silicon helps prevent them from being smartphones carry with them lots of sensitive data that, in the
bypassed or accessed. The UID allows data to be wrong hands, is capable of being used for identity theft and
cryptographically tied to a particular device. For example, the fraud. For this situation, auto erase come to help. If your
key hierarchy protecting the file system includes the UID, so phone is stolen or lost you can wipe your personal sensitive
the files cannot be accessed if someone moves the memory data from your phone. When this feature enabled 10-failed
chips from one device to another. In addition to the hardware passcode attempts will automatically erase all data on the
encryption features built into IOS devices, Data Protection device.
technology is used to further protect data stored in flash For Android, there is not native solution. But there is third
memory on the device. This feature use Hardware Key and party apps on market.
Passcode Key. Data Protection is automatically enables by
setting up a device passcode. In addition to unlocking the
device, a passcode provides entropy for certain encryption III. MALICIOUS APPS
keys. This means an attacker in possession of a device can’t The cybercriminal motivations behind mobile malware may
get access to data in specific protection classes without the vary from collecting confidential data to financial gain. The
passcode. Besides, IOS keychain securely store passwords and three main motivations behind mobile malware include
other short but sensitive bits of data, such as keys and login obtaining financial gain, collecting sensitive data without a
tokens. Keychain data is protected using a class structure user’s knowledge or approval, leave a security hole in the
similar to the one used in file Data Protection [12]. device and accessing private networks. Also, mobile-malware
For Android, data encryption likes IOS with a few missing. functions include activity monitoring and data retrieval,
Android encryption mechanism is not powerful and system modification, and unauthorized dialling [4].
sophisticated as IOS. Basically, there is no hardware level As given in evaluation of security factor, Android is the
encryption and for file protection it uses 128-bit AES. Also, it most susceptible OS for threats and attacks. The authors in [14]
is hard to guarantee that the file system encryption is enabled. stated three foremost explanations aspects for that: the
So, this feature selected by the user. shortage in reviewing for applications in Android official
market, the openness and the compatibility with other
smartphones Apps [5]. On the contrary, in Apps Store, there is
G. Data Execution Prevention and Address Space Layout a strict process to review and sign Apps before accepting it.
Randomization Also, IOS is less openness and less compatible with other
DEP (Data Execution Prevention) is the another layer for third parties Apps.
security in IOS where a processor can distinguish which parts Malware has significantly increased, and writers of mobile
of memory can hold executable code and which parts are data. malware are targeting mostly the Android platform. The most
In that, DEP just allow execution of the code not the data. frequently targeted mobile platform in 2013 was Android with
That hardens attacker mission. For example, when an exploit 79%, compared to 0% threats with IOS [15]. According to
tries to run a payload that is injected into the process, DEP McAfee, a 30% increase in the number of attacks targeting the
____________________________________________________________________________________________________________
312
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Android operating system was detected. Moreover, out of a [2] (2016) Gartner Report Press Release. [Online]. Available: http://www.
gartner.com/newsroom/id/3270418
total of 8,000 mobile malware, Android threats make up 7,000
[3] (2015) Gartner Report Press Release. [Online]. Available: http://www.
[16]. Of course, this statistic belongs to 2013-2014 years. As gartner.com/newsroom/id/2996817
we consider the previous section, IOS and Android have been [4] P. Bhattacharya, L. Yang, M. Guo, K. Qian and M. Yang, "Learning
enhancing their security technology and adapting against new Mobile Security with Labware," in IEEE Security & Privacy, vol. 12,
no. 1, pp. 69-72, Jan.-Feb. 2014.
threats.
[5] F. Al-Qershi, M. Al-Qurishi, S. Md Mizanur Rahman and A. Al-Amri,
In sum, it can be said that most important malware attacks "Android vs. iOS: The security battle," Computer Applications and
are related with social engineering, because user does not Information Systems (WCCAIS), 2014 World Congress on,
know what is really happening when installing new software. Hammamet, 2014, pp. 1-8.
[6] I. Mohamed and D. Patel, "Android vs. IOS Security: A Comparative
For this reason, Apple is making an effort to get this risky and
Study," Information Technology - New Generations (ITNG), 2015 12th
heavy duty from user’s shoulders according to Android. International Conference on, Las Vegas, NV, 2015, pp. 725-730.
[7] Q. Li and G. Clark, "Mobile Security: A Look Ahead," in IEEE
Security & Privacy, vol. 11, no. 1, pp. 78-81, Jan.-Feb. 2013.
[8] C. Nachenberg, “A window into mobile device security,” Symantec
IV. DISCUSSION AND CONCLUSION Security Response, Symantec, 2011, pp.4-9.
[9] Z. Kazmi, F. Toni, J. A. Vila, and M. M. Marcos, “TASAM-Towards
The growing popularity and sophistication of mobile the Smart Devices App-Stores Applications Security Management
platforms have made security and privacy a major issue for Related Best Practices,” in New Technologies, Mobility and Security
(NTMS), 2012 5th International Conference on, IEEE, 2012, pp. 1-5.
everyone, since these devices welcome every need that a PC [10] C. Miller,D. Blazakis,D. DaiZovi,S. Esser,V. Iozzo and R. Weinmann,
offers. To cope with this concern, mobile operation system “iOS Hacker's Handbook,”, John Wiley & Sons, 2012.
vendors have been improving new techniques and [11] T. Luo, X. Jin, A. Ananthanarayanan, and W. Du, "Touchjacking
technologies. Majority of these are evaluated in this study. Attacks on Web in Android, iOS, and Windows Phone?", Syracuse
University, Syracuse NY, USA, 2012
Operation system of two giant technology providers Apple [12] IOS Security (2016) in Apple Docs. [Online]. Available:
and Google are almost same in case of technology. But, in https://www.apple.com/business/docs/iOS_Security_Guide.pdf
approach manner Android have a few problem. Which are [13] N. Penning, M. Hoffman, J. Nikolai and Y. Wang, "Mobile malware
signing process, permission system drawback in case of social security challenges and cloud-based detection," Collaboration
Technologies and Systems (CTS), 2014 International Conference on,
engineering and quick vetting process. For IOS, these Minneapolis, MN, 2014, pp. 181-188.
concerns are handled very well. Besides, App store is only [14] Oh, Tae, Bill Stackpole, Emily Cummins, Carlos Gonzalez, Rahul
place to download app. Ramachandran, and Shinyoung Lim "Best security practices for
In this study, frankly it is not tried to answer which android, blackberry, and iOS." In Enabling Technologies for
Smartphone and Internet of Things (ETSloT), 2012 First IEEE
platform is more secure. Main goal here is to open a Workshop on, pp. 42-47. IEEE, 2012.
discussion for future works. [15] Symantec, (2014, April), “Internet security threat report 2014,”
vol.19,[Online],Available: http://www.symantec.com/content/en/us/
enterprise/other_resources/b- istr_main_report_v19_21291018.en-
REFERENCES us.pdf.
[16] McAfee, (2013, November). “McAfee 3rd Quarter 2013 Threat
[1] (2013) Canalys Report. [Online]. Available: http://www.canalys.com/ Report,”[Online],Available: http://malwarelist.net/2013/11/20/mcafee-
newsroom/mobile-device-market-reach-26-billion-units-2016 3rd-quarter-threat-report- released/
____________________________________________________________________________________________________________
313
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—With the development of technology, the usage of this study we compare Signal, Telegram and Threema among
the internet via mobile devices are increasingly widespread. At some popular secure IM applications in terms of security. Our
the same time people generally use the internet to contact with main purpose is to detect which application is the most secure
their environment. So instant messaging applications operating against the MitM attack. MitM is type of an attack based on
with internet have became very popular,therefore they are in the principle of seeing unencrypted data or data capture by
competition with each other. In this paper, Signal, Threema and
entering between the victim computers and the other network
Telegram among the most popular commercial instant messaging
applications that allow users to messaging each other safely tools [7]
are handled in terms of security. Although these applications
use telephone numbers as contact list, calls and messages use II. INSTANT MESSAGING APPLICATIONS
data connection, therefore both of the two users talking to
each other must connect to internet. Thus, users can send A. Signal
messages without expose to fee of SMS or MMS. Although Signal application supported by Android and IOS operation
these applications are practice and fast, they have some security
systems is free and open source software application allows
problems arising from the usage of the internet. So unauthorized
users (attackers) can access user conversations by using fake users to send message to each other, to do encrypted calls, to
identity. To prevent this situation, some authentication techniques send photograph and video with end-to-end encryption. After
are used in these applications. In this study, these authentication the application installed from the store of the device, in first
techniques are examined and when the user’s public key changed, step the screen of registration and phone number confirmation
what changes have occured in the application are tested. Also are seen as shown in Fig. 1.
whether these applications have vulnerabilities against the MitM
(Man in the Middle) attack or not is detected, and also the
shortcomings of these applications are evaluated from an ordinary
user perspective
Keywords—instant messaging applications, Signal, Threema,
Telegram, MitM, Authentication
I. INTRODUCTION
Instant messaging (IM) is kind of an online messag-
ing which offers users to communicate each other in real-
time. IM has become very popular recently. It is used more
than SMS worldwide. The number of IM accounts were
over 3.2 bilion in 2015 [1]. It is estimated that this figure
will reach over 3.8 bilion over the next four years [1].
When the topic comes to internet, IM applications bring
some security problems. The users of IM applications don’t
want anyone to read or modify their private conversations.
There are some possible threats for these applications, the
most important are government intelligence, curious family
or friends or neighbors etc., analyzing messages for targeted
advertisements etc. So, the instant messaging applications Fig. 1: The registration screen and confirmation code in Signal
are divided into non-secure (not end-to-end) and end-to-end application
encrypted. The non-secure IM examples are Skype, Viber,
Line, Telegram (it has both non-secure and secure form) etc.
The end-to-end encrypted IM examples are Signal, Threema,
Telegram (secret chat), Whats App (recently added) etc. In
____________________________________________________________________________________________________________
314
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
315
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
B. Telegram
Telegram is a free application and it has both Android
and IOS support. For it is faster than other instant messaging
applications, Telegram has come to forefront when compared
with others. Because it works cloud-based it is given right of
access to last correspondences in seconds on any platform. As
well as opportunity of setting up a group with 200 person by
starting secure session, it enables to sharing videos in size of 1
GB. Telegram application has common usage because of these
features is installed in a device to test. After the installation
is completed, the user is verified with phone number and
verification code (Fig. 8).
____________________________________________________________________________________________________________
316
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
317
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
318
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
319
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
verification. Another drawback is the lack of encrypted or [8] https://threema.ch/press-files/cryptography_whitepaper.pdf. Access date:
unencrypted calls. The user’s public key exchange can be 20.06.2016
done by removing the application or reset the application with [9] http://tr.androidpro.net/category/communication/threemadownload-
Delete ID option. 22785.html. Access date: 22.06.2016
[10] https://threema.ch/pressfiles/cryptography_whitepaper.pdf. Access date:
20.06.2016
IV. CONCLUSION [11] https://telegram.org/blog/crowdsourcingamoresecurefuture. Access
date: 26.06.2016
In this study, we focus on Signal, Telegram and Threema
[12] http://thoughtcrime.org/blog/telegramcryptochallenge/. Access date:
among secure instant messaging applications, we evaluate 25.06.2016
usage patterns of these applications and we also test the [13] http://www.ibtimes.co.uk/nsawhistlebloweredwardsnowdenmakespitch-
applications by secure signing in or without adding security. androidsignalprivatemessenger1526909. Access date: 28.06.2016
Also when someone’s public key changed from two people [14] http://www.tripwire.com/stateofsecurity/latestsecuritynews/snowden-
during the communication, we examine occuring changes and endorsedsignalprivatemessengerappcomestoandroid/. Access date:
make inferences about whether the application is safe against 01.07.2106
MitM attacks or not. [15] https://www.os3.nl/_media/20132014/courses/ssn/projects/threema_report.pdf.
Access date: 07.07.2016
In Signal and Threema, thanks to QR codes, imitating any
client is prevented but this precaution is not exist in Telegram.
One of the disadvantages of Telegram is that uses symmetric
key structure and any authentication is not performed between
clients. If users want to share their encrypted keys with each
other, they send the screenshot of their barcodes. The result of
changing user’s public key Signal and Threema has taken secu-
rity precaution by changing the user’s QR code. The informa-
tion of the user’s public key exchange is delivered to opponent
side in Signal, but such a warning message is not given and
the messages of the user’s changing public key are transmitted
from insecure channel in Threema. Also if the other user wants,
the user’s new public key can be verified without re-matching
QR code. Although this feature provides convenience in terms
of the user, it can create sense of danger against copies phone
numbers. For there is no such right in Threema, users need to
meet face-to-face or via e-mail to re-verification their public
keys of each other. They have to send their QR codes to each
other. Due to this feature, there is slightly lack of user inter-
active in Threema. The balance between adding security and
being user interactive is very important for such applications.
As a result, if we sort these applications according to secu-
rity level, a free software Signal is evaluated in first rank when
compared to the other two applications in terms of security
and user interaction because of it’s secure group messaging,
encrypted conversations and practice solutions presented to op-
ponent user when another user’s public key changed. Threema
is located in second rank,due to the situations of being paid
software, the lack of calling feature and not giving a warning
message to user as result of changing the public key. And
finally Telegram is located in last rank in terms of security
due to the lack of mutual authentication system and the use
of symmetric key. Although the symmetric key is much faster
than the public key, it can be broken more easily.
R EFERENCES
[1] http://www.radicati.com/, Instant Messaging Statistics Report, 2015-
2019. Access date: 15.06.2016
[2] https://telegram.org/. Access date: 15.06.2016
[3] https://threema.ch/en. Access date: 15.06.2016
[4] https://en.wikipedia.org/wiki/Signal_(software). Access date: 16.06.2016
[5] https://en.wikipedia.org/wiki/Telegram_(software). Access date:
17.06.2016
[6] https://en.wikipedia.org/wiki/Threema. Access date: 18.06.2016
[7] https://en.wikipedia.org/wiki/Maninthemiddle_attack. Access date:
10.06.2016
____________________________________________________________________________________________________________
320
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Proportion of disease is growing due to the system using fuzzy logic modeling method for boarding
malnutrition and sedentary life. In this work, a diet list is schools in Croatia. Daily menu price, meal preference,
proposed to the user by calculating the necessary amount of nutritional and energy value of foods considered input
calorie according to gender, weight, height, age and activity level. parameters. They take meal preference as input since nutrition
Diet list are prepared for seven days of a week as divided by six
habits are different in various places of Croatia. Their
meals. Parameter assessment of users and offering
recommendations are made via fuzzy expert system. Prepared software is suitable for planning of menu which has optimal
diet list are constituted considering calories of nutrients and cost and adequate nutrient intake. Kurozumi et al. [4]
based on the dieticians’ general diet list proposals. Developed proposed a Japanese diet evaluation system that uses Fuzzy
software also includes some functions such as nutrition advices, Markup Language and food exchange list published by Japan
calculation of ideal weight, information about benefits of several Diabetes Society. Dietary healthy level is evaluated by
nutrients and calorie evaluation of some daily activities. Thus, calculating the nutrient percentages of foods selected for menu
nutrition suggestion software carried out against for growing and proposed method according to the Japanese dietician
obesity and healthy eating problems in order that people would database. Another region specific work is presented by
educate themselves about wellness.
Şahiner et al. [5]. They examined how to prepare a sample
weekly menu for a young person who has a business
Keywords— Diet list, fuzzy expert systems, nutrition, nutritional environment that requires intensive physical activity and his
value. revenue determined by Türk-İş (Confederation of Turkish
Trade Unions) on the edge of hunger.
I. INTRODUCTION Faiz et al. [6] implemented an approach that gives diet and
European office of World Health Organization (WHO) exercise recommendations for diabetes by using Web
pointed out in ―European Food and Nutrition Action Plan Ontology Language and Semantic Web Rule Language as a
2015–2020‖ that malnutrition, obesity and unhealthy diets specific human case application. Similarly, Li et al. [7]
lead to noncommunicable diseases (NCDs). Also, these cause proposed automated food ontology for diabetes that cares
economic and social costs for individuals, families and necessary nutrient values and limitations. They used
countries. In the same report, about %57.4 of people over the Hierarchical Clustering Algorithms, Class Naming by
age of 20 are reported as overweight or obese. Action plan intersection naming and instance ranking by granular ranking
that further analyzes given in this report is about and positioning so as to construct various food groups (diet
implementation of healthy diet and controlling the food lists). Another work considers the diseases were presented by
productions with the aid of governments [1]. Besides all these, Chen et al. [8]. An diet recommendation was offered by using
diabetes, hypertension and high cholesterol are among the top Fuzzy Rules and Knapsack Method that utilizes users’ height,
10 causes of death in the world [2]. weight, activity level, renal function, hypertension, high
Every nutrition plays an important role for supporting daily cholesterol and preference data. Mamat et al. [9] proposed
activities of the people. Therefore, a balanced diet should be Fuzzy Multi-Objective Linear Programming application that
in proportion to the values of carbohydrate, protein, fat, provides an optimum diet list includes enough nutrients for a
vitamins and minerals. Values of these nutrients should be day with reasonable price. Decision Support System also used
calculated by taking into account taken and consumed energy. in this work to recognize the type of disease according to the
Nevertheless, people do not prefer allowing time to go symptoms obtained from the users.
dietitian because of their habits and busy schedule. To Main method used in the diet assessment systems is Fuzzy
overcome this problem, web or pc based expert systems that Logic and its derivatives. Lee et al. [10, 11] produced Type-2
have input parameters such as current diseases, activity level Fuzzy Ontology based on Type-2 Fuzzy Food Ontology and
and age should be used. Fuzzy Markup Languege-2(FML2). Similarly, they presents to
There are many studies which are focused on diet planning the user healthy level of the foods in diet with semantic
for humans. Some of them are focused on a specific region or expressions by developing FML2 based agent. Mamat et al.
a situation. Kljusurić et al. [3] presented a nutrition planning [12] made a balanced diet planning using Fuzzy Linear
____________________________________________________________________________________________________________
321
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Programing approach. In this planning, assessment of follows this group. We should consume fat and sugar group at
carbohydrate and fat amounts in the food that eaten every day, least [14].
food prices and the amount of nutrients that required to take a We can group the animal and plant nutrients according to
day are for 30 years old sedentary woman considered. Mák et the chemical properties as follows:
al. [13] have presented an equation that define dietary and Carbohydrates
physical activity advisory problem. Proposed equation Proteins
expressed as ―Hierarchical Multi-Objective Dietary Menu Fat
Planning Problem with Harmony‖ allows the formalization of Vitamin
any diet and physical activity planning problem. Minerals
When we interpret the proposed works we deduced that Water
there is a necessity for a software support to the diet list There are similar or different features of each food group
preparation. Another argument is usefulness of fuzzy logic which are necessary for the body. Healthy eating achieved by
based expert system since input parameters for diet and taking these nutrient groups adequately and balanced as
outcome presentation are verbal. Also, diet is a multi- mentioned before.
parameter process should vary according to the individual and 1) Carbohydrates
cultural characteristics. In this paper, we calculated adequate Carbohydrates are used as primary energy source of body
calories in proportion to the person’s gender, weight, height, and it is important for brain functions and nervous systems.
age and activity level parameters by using fuzzy logic. After The building blocks of carbohydrates are glucoses. It can be
calorie determination software presents one diet list that divided as monosaccharaides, disaccharides and
previously prepared. This diet list arranged according to the polysaccharides. Also while the most common dietary
food culture and presence in related area. disaccharides can be divided as are maltose, lactose, and
II. NUTRITION AND BASAL METABOLISM sucrose, the most common forms of digestible polysaccharides
are amylose and amylopectin. Monosaccharaides and
A. Nutrition disaccharides are found in such food as fruits, vegetables,
Nutrition is a conscious activity, taking nutrients needed by milk, milk productions, sugar and sugar products. And, as for
the body adequately and on time, to protect the health and polysaccharides, they are found in such food as cereals, bread,
improve the quality of life. Taking every item of the energy pasta, rice and potatoes. Glucose is stored as the form of
and nutrients that essential for growing-regeneration and glycogen in the liver and mostly skeletal muscle [15, 16].
working of the body and proper use of them in the body is 2) Proteins
called ―Adequate and Balanced Nutrition‖ [14]. Proteins are related with all forms of plant and animal life
Nutrients that we need for healthy and balanced diet can be and 20 percent of our body is composed of protein. Proteins
collected in 5 main groups as shown in Fig. 1. form the structure of each cell of our body tissue, muscle,
When we examine the pyramid, we should say that grains internal organs, tendons, skin, hair and nails. And it also used
are most preferred nutrient group. Hereinafter vegetables and for energy productions as a fuel in such metabolic formation
fruits come. Milk and dairy products with meat group foods activations as growth, formation of new tissue and repairment
of tissue damaged. Besides it is necessary for the production
Fats, Oil,
Desserts
Milk, Yogurt, Meat, Giblets,
Cheese Fish, Legumes, Eggs,
2-3 Portion Nuts
2-3 Portion
Vegetables Fruits
3-5 portion 2-4 portion
Bread, Rice, Pasta
6-11 portion
____________________________________________________________________________________________________________
322
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
of many enzymes, hormones (adrenaline and insulin) and element for life and it helps to control of body temperature,
neurotransmitters. In addition to that it has important role in use of
the protection of fluid balance in tissues, the transition of TABLE I
nutrients into cell sand out and making the blood. Protein is CONTENT AND CALORIES OF SOME NUTRIENTS FOR 100 GR [19]
composed of 20 amino acids and these amino acids are
grouped as essential amino acids and non-essential amino Food Calorie Protein Fat Carbohydr
ate
acids [15, 16].
Trout 168 18,3 10 0
3) Fat Perch 93 19,2 1,2 0
Fat can provide more than twice the energy by weight Beef Meat (Low 156 19,7 8 0
compared to protein and carbohydrate and it is essential Fat)
macronutrients which have most energy dense. Fat provides Sausage 452 21,4 40,8 0
energy production, insulation against the cold, transportation White Chicken 114 23,2 1,7 0
of fat-soluble vitamins in the body and making the body tissue. Broad Bean 338 25,1 1,7 58,2
Fat is grouped as saturated fat (butter, margarine), Haricot Bean 340 22,3 1,6 61,3
monounsaturated fat (olive oil, soybean oil, fish oil) and poly Lentil 340 24,7 1,1 60,1
unsaturated fat (sunflower, corn). Especially, saturated fat Chickpea 360 20,5 4,8 61
increases the amount of fat in the body. There are a lot of fat Pea 348 24,2 1 62,7
in such food as meat, fish, milk, cream, eggs, nuts, chocolate, Sunflower Seed 560 24 47,3 19,9
olive, coconut, cottonseed, corn, soybean, flax and peanut [15, Almond 598 18,6 54,2 19,5
16]. Walnut 651 14,8 64 15,8
4) Vitamins Hazelnut 634 12,6 62,4 16,7
Pumpkin seeds 610 30 50 10
Vitamins, cannot be synthesized by the body, are required
Pear 61 0,7 0,4 15,3
to support health and well-being. Vitamins are found in small
Strawberry 37 0,7 0,5 8,4
amounts in foods and are designated as nutrients. They help Fresh Figs 80 1,2 0,4 20,4
catalyzing the numerous biochemical reactions and aren’t Melon 33 0,8 0,3 7,7
used for direct source of energy. Vitamins are grouped as Banana 85 1,1 0,2 22
Water-Soluble Vitamins and Fat-Soluble Vitamins. B group Okra 36 2,4 0,3 7,6
vitamins and vitamin C are the water-soluble vitamins. Unlike Potato 76 2,1 0,1 17,1
the fat-soluble vitamins, they leave the body after completing Fat White Cheese 289 22,5 21,6 0
their task; they don’t accumulate in the body [15, 17]. Cheddar cheese 404 27 31,7 1,4
5) Minerals Whole Milk 61 3,3 3,3 4,7
Minerals, occur naturally, are inorganic solid with a Oil Yogurt 61 3,5 3,3 4,7
specific chemical composition. The properties of minerals can Egg 158 12,1 11,2 1,2
be listed as follows: Honey 315 0,3 0 78,4
It occurs naturally. Chocolate 528 4,4 35,1 57,9
Candy 385 0 0 99,5
It has the characteristics of any part of the whole.
Bulgur 357 10,3 1,2 78,1
It has a specific chemical formula. Wheat Bread 276 9,1 0,8 56,4
It is solid but rarely fluid. Macaroni 369 12,5 1,2 75,2
Inorganic. Rice 363 6,7 0,4 80,4
The chemical compounds obtained synthetically in the Sunflower oil 884 0 100 0
laboratory do not considered as mineral with the technological Margarine 736 0,6 81 0,4
progress. These types of solid materials are named as artificial Olive oil 884 0 100 0
minerals. Each mineral have different features for the body Butter 717 9 81,1 0,1
and the features of some minerals are described. Iron is the Tomato paste 98 2,7 0,4 21,3
necessary element for the delivery of oxygen to tissues and the Black olive 207 1,8 21 1,1
use of oxygen at the cellular and subcellular levels.
Magnesium is required in a variety of cellular activities that nutrients in the body, digestion and absorption in stomach.
support diverse physiologic systems. Zinc is necessary for Fluid intake must be increased to prevent dehydrations.
more than 300 enzymes activity. Chromium is the essential Thirsty feel can be seen late in children and recommended
mineral for regulation of glucose, lipid and protein amount of fluid are as follows to avoid of fluid loss for
metabolism [17, 18]. children:
6) Water Up to 10 kg bodyweight 100 ml / kg / day
The water is abundant in the world, indispensable for life, Up to 20 kg bodyweight 50 ml (added for each kg) / kg /
odorless and tasteless compound. The amount of fluid we day
need to take daily is associated the calorie we spend. 1 ml Up to 30 kg bodyweight 25 ml (added for each kg) / kg /
water should be taken per calorie. Water is the essential day [15, 18].
____________________________________________________________________________________________________________
323
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Inputs Outputs
(Crisp Values) Fuzzy Inputs Fuzzy Outputs (Crisp Values)
Rules
Input Output Membership
Membership Functions Functions
Questions
____________________________________________________________________________________________________________
324
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] World Health Organization. European food and nutrition action
plan 2015-2020. WHO Regional Office for Europe 2014 [cited
2016 15 July]; Available from:
http://www.euro.who.int/__data/assets/pdf_file/0008/253727/64w
d14e_FoodNutAP_140426.pdf.
[2] World Health Organization. Global Health Risks. 2009 [cited
2016 July, 15]; Available from:
http://www.who.int/healthinfo/global_burden_disease/GlobalHealt
hRisks_report_full.pdf.
[3] Kljusurić, J.G. and Ž. Kurtanjek, Fuzzy logic modelling in
nutrition planning-application on meals in boarding schools, in
Current Studies of Biotechnology, Vol. III-Food. 2003, Medicinska
naklada.
[4] Kurozumi, K., et al. FML-based Japanese diet assessment system.
in Fuzzy Systems (FUZZ), 2013 IEEE International Conference on.
2013. IEEE.
[5] ŞAHİNER, A. and G. BUZKAN, An Application of Fuzzy Goal
Programming to a Specific Diet Problem. Journal of Natural &
Applied Sciences, 2010. 14(2).
[6] Faiz, I., et al. A semantic rules & reasoning based approach for
Fig. 3 Interface of developed software Diet and Exercise management for diabetics. in Emerging
Technologies (ICET), 2014 International Conference on. 2014.
IEEE.
[7] Li, H.-C. and W.-M. Ko. Automated food ontology construction
mechanism for diabetes diet care. in 2007 International
Conference on Machine Learning and Cybernetics. 2007. IEEE.
[8] Chen, R.-C., et al. Constructing a diet recommendation system
based on fuzzy rules and knapsack method. in International
Conference on Industrial, Engineering and Other Applications of
Applied Intelligent Systems. 2013. Springer.
[9] Mamat, M., et al., Fuzzy Multi-Objective Linear Programming
Method Applied in Decision Support System to Control Chronic
Disease. Applied Mathematical Sciences, 2013. 7(2): p. 61-72.
[10] Lee, C.S., et al., Diet assessment based on type‐2 fuzzy ontology
Fig. 4 Membership functions example for height and fuzzy markup language. International Journal of Intelligent
Systems, 2010. 25(12): p. 1187-1216.
[11] Lee, C.-S., et al., Type-2 fuzzy set and fuzzy ontology for diet
and body mass index. Membership functions of female height application, in Advances in Type-2 Fuzzy Sets and Systems. 2013,
which is one of the diet system inputs is shown in Fig. 4. Springer. p. 237-256.
[12] Mamat, M., et al., Fuzzy linear programming approach in balance
Diet list are prepared for seven days of week as divided by diet planning for eating disorder and disease-related lifestyle.
six meals. Diet list that previously prepared taking into Applied Mathematical Sciences, 2012. 6(103): p. 5109-5118.
account the calorie values of the foods and these are classified [13] Mák, E., et al. A formal domain model for dietary and physical
according to the fuzzy inference outputs. After required activity counseling. in International Conference on Knowledge-
Based and Intelligent Information and Engineering Systems. 2010.
calorie calculation, relevant list proposed to the user. Springer.
[14] Hacettepe Üniversitesi Beslenme ve Diyetetik Bölümü.
V. CONCLUSION TÜRKIYE’YE ÖZGÜ BESLENME REHBERI. 2004 [cited 2016
Balanced diet should help users to prevent eating disorder July, 15]; Available from:
http://beslenme.gov.tr/content/files/yayinlar/kitaplar/diger_kitaplar
such as anorexia nervosa, diabetes, and obesity. This study /beslenme_rehberi.pdf.
shows that application of fuzzy logic can be used to represent [15] Pehlivan, A. Çocuk ve Genç Futbolcuda Beslenme. 2009 [cited
recommended energy and nutrient intake adequately, as well 2016 July, 15]; Available from:
as to present acceptable price and preferences of menu https://www.tff.org/Resources/TFF/Documents/2009DK/FGM/yay
inlar/aylik-kitapcik/Sporcuda-Beslenme(Temmuz).pdf.
selection for people in a specific region. Diet list is proposed [16] Litchford, M.D., Macronutrient recommendations: protein, carbohydrate,
to the user by calculating the necessary amount of calorie and fat, in The ASMBS Textbook of Bariatric Surgery. 2014, Springer. p.
according to gender, weight, height, age and activity level. 101-109.
[17] Lukaski, H.C., Vitamin and mineral status: effects on physical performance.
These lists are prepared for seven days as divided by six meals Nutrition, 2004. 20(7): p. 632-644.
and constituted considering calories of nutrients and based on [18] Mineral. 2016 [cited 2016 July, 15]; Available from:
the dieticians’ general diet list proposals. https://tr.wikipedia.org/wiki/Mineral.
[19] Besinlerin Değer Cetveli. 2011 [cited 2016 July, 15]; Available from:
http://www.antrenmanyap.com/tr/blog.asp?ID=6&CID=7&AID=20&do=sh
ACKNOWLEDGMENT ow.
[20] Milli Eğitim Bakanlığı. ENERJİ HESAPLAMALARI. 2011 [cited 2016
I would like to thank my undergraduate student Mustafa July,15]; Available from:
http://www.megep.meb.gov.tr/mte_program_modul/moduller_pdf/Enerji%2
Oğuzhan MUSARA because of his valuable study in this 0Hesaplamalar%C4%B1.pdf.
project. [21] Calories Burned During Exercise, Activities, Sports and Work. 2016 [cited
2016 July, 15]; Available from:
http://www.nutristrategy.com/caloriesburned.htm.
____________________________________________________________________________________________________________
325
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— IEEE 802.11 wireless network standard has become is activated according to the received signal strength (RSS),
one of the most used wireless networking technologies for smart such that stations select an AP that has the strongest RSSI. If
devices as it offers mobility support and low cost deployment. AP selection is activated only by observing the RSS, it results
However, these devices deeply rely on the energy provided by their in many stations associate with one of the APs as shown in Fig.
batteries, which results in limited running time. In addition,
1. As a result, throughput of the heavily loaded AP decreases
condensed deployment of Access Points (APs), which is essential
to handle increasing demand of performance and coverage, has due to the congestion, while there are other APs available but
also been causing rise of network-side power consumption. In this not used efficiently. Therefore, a new algorithm which balances
context, this paper proposes an energy-aware Access Point the load among APs is required to choose possibly the best AP
switching procedure for smart mobile devices to increase overall from an available AP list.
energy saving of both APs and smart devices. In the proposed
method, each channel is investigated, in terms of channel
utilization, signal quality, probability of collision and deployed
traffic types, making use of local and IEEE 802.21-based
management frames. With the help of reputation and context-
aware computations, Access Points inform the stations that are
associated with themselves to maintain their connection or to
handover to another network. The aim of the proposed scheme is
to maintain desired QoS with minimum number of APs and
optimal energy consumption. Widespread simulations have been
executed to validate the efficiency of the proposed method. The
results demonstrate that the proposed method dramatically
increases overall throughput and reduces power consumption of Figure 1. (a) Unbalanced assignment (b) Balanced assignment
stations over IEEE 802.11 WLANs.
Current energy-aware handover approaches succeed energy
Keywords— IEEE 802.11 WLANs, IEEE 802.21, Load balancing, savings either by diminishing channel scanning time or by
Energy efficiency, Handover connecting to the most energy-efficient AP according to the the
traffic type. Yet, an energy-aware handover approach must
I. INTRODUCTION contain more local and network related metrics, such as battery
Nowadays, wireless local area networks (WLAN) and smart and interface information, application-specific QoS
mobile devices have an enormous amount of usage in daily life. requirements, user preferences, number of stations connected
The IEEE 802.11 WLANs [1] are widely deployed all around to each AP, channel utilization ratio per AP, and etc. [3].
the world. However, due to design choices and requirements,
communication range of the IEEE 802.11 WLAN standard is In line with the above-mentioned clarifications, this paper
limited and large wireless network coverage is provided using addresses an energy-aware AP switching on/off procedure for
several APs [2]. smart mobile devices. In the proposed method, each channel is
investigated in terms of channel utilization, signal quality,
Since smart devices are mobile, stations on the move may probability of collision and deployed traffic types by making
lose their signals from their associated APs. In this case, use of local and IEEE 802.21-based management frames. With
stations attempt to retain their wireless connectivity by the assist of reputation and context-aware computations, APs
connecting to other APs in vicinity. Therefore, AP selection, inform stations that are associated with themselves to maintain
a.k.a. handover operation plays a key role for the optimization their connection or to handover to another network.
of the wireless connectivity over the WLAN environment. Widespread simulations have been executed to validate the
performance of the suggested method. The results demonstrate
Handover is the procedure of shifting an ongoing call or a that it dramatically increases overall throughput and reduces
data session from one AP to another. Conventionally, handover power consumption of stations over IEEE 802.11 WLANs.
____________________________________________________________________________________________________________
326
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
327
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
scanning. Nevertheless, throughout the handover procedure, RSSI values obtained in channel scanning are used to compute
connecting to an AP that is expected to consume the least the expected amount of power consumption of the device for
amount of energy among all APs is also as important as the fast every possible association scenario with available APs in
handover. Therefore, we also propose an energy-efficient vicinity. Subsequently, the proposed method allows mobile
handover scheme, which calculates the expected amount of devices to associate with APs that are expected to consume the
power consumption per AP, by utilizing the Channel Busy least amount of energy among all APs.
Times (CBTs) and the number of stations deployed in each AP.
Since unsaturated (e.g. UDP) delay sensitive stations have
CBT is the fraction of time that the wireless channel is busy different characteristics and requirements, E-model
within a given interval. CBT can be calculated by each mobile conversational audio quality factor R is also used to estimate
device distributedly through summing the time utilized by the the perceived voice quality for real-time VoIP applications.
transmission of all data, management, and control frames in the Association of a new station to the AP affects the voice quality
network. Consequently, CBT of a data frame that has a size of of other stations associated to the same AP. For an unsaturated
S bytes and a transmission rate R can be computed as follows, station, there are many factors that may affect the selection of
the right AP. R value, capacity utilization, transmission rate,
= + ( )( ) fairness, hidden nodes and the number of associated stations to
an AP are some of these factors. Existing beacon or probe
CBT of an ACK frame is computed as follows, response frames do not contain all of this information. A new
= + algorithm that also considers all of these factors explained
above is required for the best AP selection for unsaturated
CBT of a beacon frame is computed as follows, stations. Therefore, a few bytes of extra status information have
to be added to the beacon and probe response frames. Since it
= + is only a few bytes of information, it does not affect the
overhead significantly but it improves the voice quality and
Finally, the total channel busy time CBTTotal within a overall throughput.
specific interval t can be calculated as,
In this context, APs will transmit a beacon or probe response
( )=( ( )× )+ ( )× frame which contains three additional items; an average R value
( )
of the stations in AP, a new estimated average R value if the
+ ( )( ) station selects the same AP and the possible R value of the
station which is trying to associate with the same AP. Then, the
station will receive the enhanced beacon or probe response
where a(t), b(t) and c(t) are the numbers of ACK, beacon and frames from APs and consider the best AP selection by looking
data frames that are encountered within a specific interval t, at the RSSI value as well. Besides, an AP which has less station
respectively. Accordingly, channel utilization percentage U(t) due to different traffic types and frame sizes can be more loaded
within a specific interval t can be computed as, than other APs which have more stations. That’s why fairness is
( ) not only about the number of stations associated to the APs. The
( )= × 100 main point of this study is selecting an AP of a station to have a
higher voice quality. Improving the voice quality also means
With this computation, mobile devices can compute better throughput, load balancing and fairness.
channel utilization rates of their APs within a specific interval
distributedly. Nonetheless, this computation is impractical for In the proposed algorithm, central algorithm first checks if
devices as it requires continuous processing overhead, prevents there is a new association request from a station. If there is a
devices to be in the doze state and rises the power consumption. new request, algorithm increases its request count by 1 to
However, in the proposed method, APs are responsible for prevent hidden node effect. Then algorithm gathers MAC layer
computing the CBT values within a specific interval. For information and RTCP periodic frames to calculate channel
instance, Atheros AR5212 chipset maintains 32-bit register utilization rate and R values of VoIP flows. RTCP feedback can
counters to track the “medium busy time” and the “cycle time”. be easily acquired from RTCP sender and receiver reports [13].
These counters give the fraction of time that the wireless The proposed algorithm uses two one-second fast feedbacks.
channel is busy (CBT) per second [12]. Likewise, in the Our algorithm maintains a multi-dimensional control array that
proposed method, we assume each AP computes and updates contains caller_ID, Tdelay, Loss, R1, R2, Rmean values to assist
its CBT periodically with one-second intervals and then, rapid computations and responses. After calculations,
delivers this information with the IEEE 802.21 IS. Whenever a algorithm defines its Rk, Ravr (old) and Ravr (new) values and
device initializes its network discovery phase, the IS informs responds this additional information with beacon frames.
the station about the number of stations in available APs and
their CBTs along with the connectivity graph. The information Up until now, we have addressed energy-efficient wireless
of the CBT and the number of stations per AP along with the communication for mobile-side. In order to increase overall
____________________________________________________________________________________________________________
328
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Name Value Unit Throughout the simulations, we also evaluated the impact of
Transmitting state 1.3 W minimum number of AP deployment on the power
Idle state 0.74 W
Receiving state 0.9 W consumption, throughput, delay and jitter. In this context, we
MinChTime 30 ms initiated an environment that has 6 APs, each of which has one
MaxChTime 50 ms
ChSwitching 2 ms TCP and one UDP station. When the proposed algorithm was
Slot time 20 µs
Bit rate 11 Mbps off, 6 APs maintained their communications. However, when
Table 1. Timing unit and values for IEEE 802.11b
the proposed algorithm was on, each channel was investigated
in terms of channel utilization, signal quality, probability of
Name Length (bits) Bit rate (Mbps) Time (µs) collision and deployed traffic types. Afterwards, the proposed
DIFS 50 1 50 algorithm informed 8 of these 12 stations to handover. This way,
Preamble 192 1 192
Header 240 11 21,81 only two of six APs maintained their operations. Other four APs
Data (TCP) 8000 11 727,27 were switched off for the energy efficiency. In this context,
Voice (UDP) 1280 11 116,36
SIFS 10 1 10 Table 4 shows the comparison of power consumption,
ACK 112 2 56 throughput, delay, jitter and number of active APs when the
Table 2. IEEE 802.11b parameter values used at simulations proposed algorithm is on and off.
____________________________________________________________________________________________________________
329
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[6] Gong, Huazhi, Kitae Nahm, and Jongwon Kim. "Distributed fair access
Description Alg. Is off Alg. Is on point selection for multi-rate IEEE 802.11 WLANs." 2008 5th IEEE
Avr. Power Consumption 1120 mw 740 mw Consumer Communications and Networking Conference. IEEE, 2008.
Avr. Throughput 2340 Kbps 2215 Kbps [7] Vasudevan, Sudarshan, et al. "Facilitating access point selection in IEEE
Avr. Delay 72 ms 77 ms 802.11 wireless networks." Proceedings of the 5th ACM SIGCOMM
Avr. Jitter 21 ms 23 ms
Nr. of active APs 6 2 conference on Internet Measurement. Usenix Association, 2005.
Table 4. Comparision of power consumption, throughput, delay, jitter and [8] Nicholson, Anthony J., et al. "Improved access point
number of active APs when the proposed algorithm is on and off selection." Proceedings of the 4th international conference on Mobile
systems, applications and services. ACM, 2006.
[9] Haratcherev, Ivaylo, Michele Fiorito, and Carine Balageas. "Low-power
As it is seen from Table 4, the proposed scheme increases sleep mode and out-of-band wake-up for indoor access
the energy efficiency dramatically by switching some of the APs points." GLOBECOM Workshops. 2009.
off. However, since the remaining APs have denser channel [10] Abusubaih, Murad, et al. "On access point selection in IEEE 802.11
utilization, average throughput is slightly decreased due to wireless local area networks." Proceedings. 2006 31st IEEE Conference
increased probability of collision. Besides, delay and jitter are on Local Computer Networks. IEEE, 2006.
increased due to frequent contention compared to the scenario [11] Taniuchi, Kenichi, et al. "IEEE 802.21: Media independent handover:
Features, applicability, and realization." IEEE Communications
of the “proposed algorithm is off”. Magazine 47.1 (2009): 112-120.
[12] Jardosh, Amit P., et al. "Understanding congestion in IEEE 802.11 b
V. CONCLUSIONS wireless networks." Proceedings of the 5th ACM SIGCOMM
In order to reduce the overall energy consumption of both conference on Internet Measurement. USENIX Association, 2005.
Access Points and smart devices, this paper proposes an [13] Tüysüz, M. F., and Mantar, H.A., "Minimizing the Effects of Multi-rate
WLANs by Adapting Link Adaptation and Call Admission Interfaces.
energy-aware Access Point switching procedure for smart "International Conference on Networked Digital Technologies. Springer
mobile devices. In the proposed method, each channel is Berlin Heidelberg, 2010.
investigated, in terms of channel utilization, signal quality,
probability of collision and deployed traffic types, making use
of local and IEEE 802.21-based management frames. With the
help of reputation, historical and context-aware computations,
Access Points inform the stations that are associated with
themselves to maintain their connection or to handover to
another network.
ACKNOWLEDGMENT
This work was supported by The Scientific and Technological
Research Council of Turkey (TÜBİTAK) under Grant No:
114E075.
REFERENCES
[1] IEEE Std 802.11, “Wireless LAN Medium Access Control (MAC) and
Physical Layer (PHY) specifications”, 1999 Edition.
[2] Tuysuz, M. F., and Haci Ali Mantar. "Minimizing communication
interruptions using smart proactive channel scanning over IEEE 802.11
WLANs." Wireless Personal Communications 82.4 (2015): 2249-2274.
[3] Tuysuz, M. F., "An energy-efficient QoS-based network selection
scheme over heterogeneous WLAN–3G networks." Computer
Networks, 75 (2014): 113-133.
[4] Marsan, Marco Ajmone, et al. "A simple analytical model for the
energy-efficient activation of access points in dense
WLANs." Proceedings of the 1st International Conference on Energy-
Efficient Computing and Networking. ACM, 2010.
[5] Fukuda, Yutaka, Takamitsu Abe, and Yuji Oie. "Decentralized access
point selection architecture for wireless LANs." Wireless
Telecommunications Symposium, 2004. IEEE, 2004.
____________________________________________________________________________________________________________
330
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—In this study, mitochondrial displacement-loop (D- can be applied to nucleotide substitution in alphabet (A, C, G,
loop) sequences isolated from different hominid species are T) [14].
clustered using similarity matrix, Principal Component Analysis In data mining, the field of clustering has received
(PCA) and K-means algorithm. Firstly, the mitochondrial D-loop significant attention in recent years and has become one of the
sequence data are retrieved from the GenBank database and
important parts of machine learning research. Clustering is the
copied into MATLAB. Pairwise distances are computed using p-
distance and Jukes-Cantor methods. A phylogenetic tree is process of categorizing a finite number of objects into groups
created and then a similarity matrix is generated according to where all members have common properties. Data mining is
the pairwise distances. Furthermore, the clustering is performed the process of using clustering algorithms in order to analyse
using only K-means algorithm. After that PCA and K-means are data for patterns and relationships. K-means clustering
used together in order to cluster mitochondrial D-loop sequences. algorithm [15] is one of the most used and popular clustering
algorithms [16]. Furthermore, K-means is a simple
Keywords—clustering; p-distance; PCA; Jukes-Cantor; K-means unsupervised learning algorithm used to solve well known
algorithm; similarity matrix. clustering problems.
The rest of the paper is organized as follows. In section II,
the methods used for clustering of mitochondrial D-loop
I. INTRODUCTION sequences are explained. Section III investigates the results of
the methods used in this study. Finally, conclusions being
Mitochondrial DNA (mtDNA) sequences of mammals
under study are summarized in section IV.
evolve more rapidly than nuclear DNA sequences [1], [2].
This fast rate of evolution generates more change between
sequences. In order for research of closely related species and II. METHODS
populations, this rate is a benefit [1], [3]. In animal mtDNA,
there are four principal kinds for sequence changes. These are In this paper, in order for clustering mitochondrial D-loop
sequence rearrangements, additions, deletions and nucleotide sequences isolated from different hominid species, similarity
substitutions [4], [5]. Nucleotide substitutions are the most matrix, PCA and K-means algorithm are used. To calculate
important principal for the derivation of phylogenetic pairwise distances, p-distance and Jukes-Cantor methods are
relationships [4], [6]. The fastest evolving part of the utilized. K-means algorithm is used alone and then it is
mitochondrial genome is the mitochondrial control region utilized with PCA for clustering. Clustering of mitochondrial
(Displacement or D-loop) [4], [7]–[9]. D-loop sequences using Jukes-Cantor method is shown in Fig.
Principal Component Analysis (PCA) is a classical feature 1. As seen in the flowchart, firstly, the mitochondrial D-loop
extraction and data representation method [10], [11]. In sequence data are retrieved from the GenBank database.
addition, it can be used to reduce the dimension of similarity Secondly, retrieved sequence data are copied into MATLAB.
matrix generated according to the pairwise distances and After that pairwise distances are calculated using Jukes-
simplify the mitochondrial D-loop sequence data structure. Cantor method and then a phylogenetic tree is created. A
The main features of the mitochondrial D-loop sequences can similarity matrix is generated according to the pairwise
be extracted using PCA by means of mapping high distances. In addition, a dendrogram is drawn. Clustering is
dimensional space data into low dimensional space. performed using K-means algorithm. Finally, K-means
In phylogenetic analysis, distance measure is a significant algorithm is used after applying PCA.
matter [12]. Using p-distance and Jukes-Cantor methods, Clustering of mitochondrial D-loop sequences using p-
pairwise distances are calculated. Jukes-Cantor method is the distance method is shown in Fig. 2. As seen in the flowchart,
simplest nucleotide substitution model which estimates the unlike the first method, pairwise distances are calculated using
evolutionary distance between two sequences. Besides, it is p-distance method. Following steps are performed in the same
called one parameter model in the literature [13]. This model manner.
331
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
End clustering
End clustering
332
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 4. Creating similarity matrix and drawing dendrogram (Jukes-Cantor). Fig. 7. Performing PCA on square form (Jukes-Cantor).
Fig. 5. Converting pairwise distances to square form (Jukes-Cantor). Fig. 8. Clustering using K-means algorithm after applying PCA (Jukes-
Cantor).
333
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 9. Phylogenetic tree (p-distance). Fig. 12. Clustering using K-means algorithm (p-distance).
Fig. 10. Creating similarity matrix and drawing dendrogram (p-distance). Fig. 13. Performing PCA on square form (p-distance).
Fig. 11. Converting pairwise distances to square form (p-distance). Fig. 14. Clustering using K-means algorithm after applying PCA (p-distance).
334
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
As seen in Fig. 3–Fig. 14, clustering of mitochondrial D- [5] W. M. Brown, “The mitochondrial genome of animals,” MacIntyre RJ
(ed) Molecular Evolutionary Genetics, Plenum Press, New York, pp.
loop sequences isolated from different hominid species is
95−130, 1985.
successfully performed using similarity matrix, PCA and K- [6] A. C. Wilson, R. L. Cann, S. M. Carr, M. George, U. B. Gyllensten, K.
means algorithm. It is observed that both Jukes-Cantor and p- M. Helm-Bychowski, R. G. Higuchi, S. R. Palumbi, E. M. Prager, R. D.
distance methods are practical for computing pairwise Sage, and M. Stoneking, “Mitochondrial DNA and two perspectives on
evolutionary genetics,” Biological Journal of the Linnean Society, vol.
distances.
26, no. 4, pp. 375−400, December 1985.
[7] W. B. Upholt and I. B. Dawid, “Mapping of mitochondrial DNA of
individual sheep and goats: rapid evolution in the D loop region,” Cell,
IV. CONCLUSION vol. 11, no. 3, pp. 571−583, July 1977.
[8] M. W. Walberg and D. A. Clayton, “Sequence and properties of the
In recent years, clustering has become a significant research human KB cell and mouse L cell D-loop regions of mitochondrial
topic in the area of machine learning. In this paper, when DNA,” Nucleic Acids Research, vol. 9, no. 20, pp. 5411−5421, October
clustering mitochondrial D-loop sequences isolated from 1981.
different hominid species, similarity matrix, PCA and K- [9] D. Chang and D. A. Clayton, “Priming of human mitochondrial DNA
replication occurs at the light-strand promoter,” Proceedings of the
means algorithm are used. First of all, K-means algorithm is National Academy of Sciences of the United States of America, vol. 82,
used alone and then it is utilized with PCA in order for no. 2, pp. 351−355, January 1985.
extracting features of the pairwise distances located in the [10] C. Eyupoglu, “Implementation of Color Face Recognition Using PCA
similarity matrix. Besides, pairwise distances are calculated and k-NN Classifier,” 2016 IEEE NW Russia Young Researchers in
Electrical and Electronic Engineering Conference (ElConRusNW), pp.
using Jukes-Cantor and p-distance methods. According to the 199−202, St. Petersburg, Russia, 2–3 February 2016.
study results, it is seen that the mitochondrial D-loop [11] X. Xiang, J. Yang, and Q. Chen, “Color face recognition by PCA-like
sequences are successfully clustered using similarity matrix, approach,” Neurocomputing, vol. 152, pp. 231−235, March 2015.
PCA and K-means algorithm. [12] D. Wei and Q. Jiang, “A DNA Sequence Distance Measure Approach
for Phylogenetic Tree Construction,” 2010 IEEE Fifth International
Conference Bio-Inspired Computing: Theories and Applications (BIC-
TA), pp. 204−212, Changsha, 23–26 September 2010.
REFERENCES [13] P. Bhambri and O. P. Gupta, “Development of Phylogenetic Tree
[1] H. Zischler, H. Geisert, A. Von Haeseler, and S. Pääbo, “A nuclear Based on Kimura's Method,” 2012 2nd IEEE International Conference
'fossil' of the mitochondrial D-loop and the origin of modern humans,” on Parallel Distributed and Grid Computing (PDGC), pp. 721–723,
Nature, vol. 378, no. 6556, pp. 489–492, November 1995. Solan, 6–8 December 2012.
[2] W. M. Brown, E. M. Prager, A. Wang, and A. C. Wilson, [14] S. S. Patil, V. Kumar, V. R. Pai, and A. K. Patil, “Constructing
“Mitochondrial DNA sequences of primates: tempo and mode of phylogenetic tree and analysis using information retrieval approach for
evolution,” Journal of Molecular Evolution, vol. 18, no. 4, pp. 225–239, MYB tfr's of rice genome,” 2015 IEEE International WIE Conference
July 1982. on Electrical and Computer Engineering (WIECON-ECE), pp. 523–
[3] D. R. Maddison, M. Ruvolo, and D. L. Swofford, “Geographic origins 529, Dhaka, 19–20 December 2015.
of human mitochondrial DNA phylogenetic inference from control [15] J. MacQueen, “Some methods for classification and analysis of
region sequences,” Systematic Biology, vol. 41, no. 1, pp. 111−124, multivariate observations,” Proceedings of 5th Berkeley Symposium on
1992. Mathematical Statistics and Probability, pp. 281–297, Berkeley,
[4] A. R. Hoelzel, J. M. Hancock, and G. A. Dover, “Evolution of the University of California Press, 1967.
Cetacean Mitochondrial D-Loop Region,” Molecular Biology and [16] W. K. Daniel Pun and A. B. M. Shawkat Ali, “Unique Distance
Evolution, vol. 8, no. 3, pp. 475−493, 1991. Measure Approach for K-means (UDMA-Km) Clustering Algorithm,”
2007 IEEE Region 10 Conference (TENCON), pp. 1–4, Taipei, 30
October–2 November 2007.
335
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Image segmentation is the first step of image is associated with each edge based on some property of
processing and image analysis. Texture segmentation is a chal- the pixels that it connects. Normalized-cut, Minimal-cut and
lenging task in image segmentation applications. Neutrosophy Mean-Cut can be seen in some of popular graph cut methods
has a natural ability to handle the indeterminate information. In
this work, we investigate the texture image segmentation based [14–16]. Neutrosophic set (NS) was proposed as an extension
on Gabor filters (GFs) and neutrosophic graph cut (NGC). We of fuzzy set [17]. In NS theory, a member of a set has
proposed an image segmentation approach, which applies GFs a degree of truth, a falsity degree and an indeterminacy
to gray-level images to extract image features matrix, and it degree [18]. Therefore, it has an ability to deal with the
segments them into regions. First, color images are transformed indeterminacy information, and has attracted much attention
to gray level images as input images. Then, input parameters of
GFs are adjusted, and GFs are performed on the input images almost in all engineering communities and subsequently a
to extract features. The NGC is employed for classification of great number of works have been studied [19], [20].
input images. Finally, experiments are conducted on various In this paper, we propose a texture image segmentation
natural images to evaluate the approach. Experimental results algorithm which uses GFs, NS and graph cut method. Firstly,
show that the proposed approach achieves desired performance various GFs are considered which constructs a texture feature
of texture segmentation. However, it cannot segment the texture-
free images as well as texture images. In future works, we will volume. Constructed volume is then projected into 2d by
try to segment both texture images and texture-free images at using principal component analysis (PCA). The projected
the same time. feature image is transformed into NS domain and indeter-
Keywords— Image segmentation, texture segmentation, Ga- minacy degree is calculated accordingly. Then an indeter-
bor filters, Neutrosophic Graph Cut. minacy filter is constructed using the indeterminacy value
on feature image which is defined by combining the spatial
I. I NTRODUCTION
information and global texture feature information. The
The goal of the image segmentation considered as dividing indeterminacy filter is employed to remove the indeterminacy
an input image into several sub-images. To do this, some in the intensity and spatial information respectively. A graph
predefined criterions should be accepted where the sub- is defined on the image and the weight for each pixel is
images are disjoint, homogeneous and meaningful. Many represented using the value after indeterminacy filtering, and
researchers have been working on image segmentation, and the energy function is also redefined using the neutrosophic
a variety of works have been published [1]. Texture seg- value. Finally, the segmentation results are obtained using a
mentation is a challenging and important task in image maximum-flow algorithm on the graph.
segmentation applications. For the texture feature extraction, The organisation of the paper is as following; in Section
there exist various texture descriptors such as gray level co- 2 we give the methodology of the proposed method. In
occurence matrix, Gabor filters (GFs) and Markov random Section 3, some experimental works are conducted. Finally,
fields [2]. GFs have been successfully applied to many image we concluded the paper in Section 4.
applications such as texture segmentation [3–5], document
analysis [6], edge detection [7], retina identification [8], II. M ETHODOLOGY
fingerprint processing [9], image coding [10], and image Briefly speaking, the proposed approach composed of 3
representation [11]. main building blocks. Extracting texture features from input
Graph based methods are important for image segmen- image by using GFs is the first building block. This block
tation applications [12]. A graph G can be defined as constructs a texture feature volume according to the number
G = (V, E) where V and E are a set of vertices and edges, of considered GFs. The second block employes the NS and
respectively. On an image, vertices can be either pixels or calculates the indeterminacy degree accordingly. Then an
regions, and edges connect the neighbouring vertices [13]. indeterminacy filter is constructed using the indeterminacy
A weight, which is a non-negative measure of dissimilarity, value on feature image which is defined by combining the
____________________________________________________________________________________________________________
336
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
337
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 3. Top-left: Lady image, top-right: result image after GF and PCA,
bottom-left: segmentation of NGC, bottom-right: border of segmentation.
Fig. 2. Top-left: Cheetah image, top-right: result image after GF and PCA,
bottom-left: segmentation of NGC, bottom-right: border of segmentation.
____________________________________________________________________________________________________________
338
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
339
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Internet of Things (IoT) is a global infrastructure governmental organizations [6]. Internet of things is a new
worldwide which links objects and enables data generation and world for connecting every object of the real world with the
sharing of this data. IoT is considered as one of the most virtual space in the computer world. IoT is a type of network
important areas of future technology and gets attention that links physical objects that surround us with the virtual
considerably by researchers and practitioners in recent years.
world to perform information exchange [7].
Applications presented by IoT makes possible a large number of
developments, but very few can be used currently. With rapid When the literature is viewed, many of identification of
advances in this system technology, successful implementations internet of thing are seen;
will continue to emerge to improve the quality of life in many The Internet of Things is the connection – via the internet –
areas. In this study, IoT is examined in general terms and its of objects from the physical world that are equipped with
applications, advantages and disadvantages are reviewed. sensors, actuators and communication technology [8].
The Internet of Things is a novel paradigm that is rapidly
Keywords— Internet of Things, RFID (radio frequency gaining ground in the scenario of modern wireless
identification), Sensor, Smart objects. telecommunications. The basic idea of this concept is the
pervasive presence around us of a variety of things or objects
I. INTRODUCTION – such as Radio-Frequency Identification tags, sensors,
Internet has made a significant impact on the economy and actuators, mobile phones, etc. – which, through unique
society bringing exceptional network infrastructure and addressing schemes, are able to interact with each other and
communication. It has been more persistent with the advent of cooperate with their neighbors to reach common goals [9].
the low cost wireless connection. With emerging technologies, The Internet of things is a things connected network, where
billions of people are connected to the Internet via portable things are wirelessly connected via smart sensors; IoT is able
computers or mobile devices. After this stage, the expected to interact without human intervention [4].
big step is that interconnected computers exchange The internet of things is a network of objects equipped with
information with interconnected objects. radio frequency identification chips and similar technologies
IoT is considered as a part of the Internet of the future and so that the objects could communicate and interact with each
will comprise billions of intelligent communicating ‘things’. other [10].
You can think of all the things / objects as from cars to books, The internet of things is a common worldwide network
from electrical appliances to foods, from refrigerators to water which in a unique way addressable things / objects created by
heaters, from intelligent buildings to shoes that are connected them and the objects in the network are in contact each other
with each other will be a development waiting for us in the with a specific protocol [2].
future. Products previously consisting of only mechanical and The impact of the IoT has influence on many areas in a
electrical components will return to hardware, sensors, short time. Especially, ability of IoT which adapts to any
electronics and complex devices which are interconnected in a environment, in any conditions without restriction of place,
variety of formats via the internet and even some platforms time and object has made it popular [11]. IoT is currently
[1]-[4]. going through a phase of rapid growth. Gartner, Inc. forecasts
It is becoming more apparent, we stand on the threshold of that 6.4 billion connected things will be in use worldwide in
new computer applications era that will radically affect our 2016, 30 percent more from 2015, and will reach 20.8 billion
lives. But the oncoming wave of technological revolution will by 2020. In 2016, 5.5 million new things will get connected
affect us more from all directions; it seems that our near future every day. Gartner estimates that the Internet of Things will
will be filled with small processors which communicate with support total services spending of $235 billion in 2016, 22
each other simultaneously and these will be integrated into the percent more from 2015. Services are dominated by the
majority of everyday objects when their small sizes and low professional category (in which businesses contract with
costs are considered [5]. external providers in order to design, install and operate IoT
Internet of Things has emerged as new platforms in the systems), however connectivity services (through
Information and Communication Technologies revolution of communications service providers) and consumer services
the 21st century. This platform is used to provide (on demand) will grow at a faster pace [12]. The evolvement of IoT can be
services and resources in different domains including illustrated by several phases as shown in Fig. 1 [4].
education, commerce, healthcare, public sector and various
____________________________________________________________________________________________________________
340
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
341
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Dijkman et al. [8] presented a business model framework Internet of Things is an interplay of smart objects and smart
for Internet of Things applications. Sicari et al. [23] presented communication networks and include own system storage,
the main research challenges and the existing solutions in the computing, visualization and interpretation tools for data
field of IoT security. Tao et al. [16] introduced the concept, analytics [15], [22].
characteristics, applications of IoT and analyzed energy When objects equipped with sensors and electronic circuits,
consumption in product life cycle. Then they summarized the they begin to acquire properties "thinking", "feeling" and
existing applications of IoT in product life-cycle energy "speaking". Thus, they have access the ability to update their
management and analyzed the potential applications and own status information by contacting us [2].
challenges of IoT in product life-cycle energy management. Smart objects embedded device of IoT;
Verdouw et al. [24] analyzed concept of virtual food supply 1. It has a unique identity.
chains with Internet of Things and proposed an architecture to 2. It has the ability to communicate effectively with
implement enabling information systems. Then they verified environment.
the proposed architecture with a case study of a fish supply 3. It has obtained and storage capability data related itself.
chain. Fang et al. [25] presented an integrated three-stage 4. It deploys a language to show properties, production
model based on IoT technology for the optimization of demands etc.
procurement, production and product recovery, pricing and 5. It has capable of making decisions about its own fate
strategy of return acquisition and proposed a novel particle [27].
swarm optimization algorithm based on two heuristic methods They can communicate and interact either among
to solve the problem. themselves, building networks of interconnected objects, or
Aktaş et al. [3] suggested a system based IOT which allows with end-users or other entities in the network [20].
viewing physiological data received from hospitalized babies Think of a cup that can access the Internet. It is able to
in infant intensive care unit targeted realization within the detect what is the liquid put into it. Not only as a species; also
scope of this technology both from the hospital environment in terms of the content-forming composition. This
and outside the hospital environment, referring to the IoT information can be transferred to a desired application or
technology applications in the health field. environment on the internet. Thus, you can examine the
The aim of this study is to examine the concept of "Internet knowledge of fluid you drink throughout the day, for example
of Things" widely used in recent years and to determine the through an application on your smart phone, at the end of the
existing situation of IoT. Thus, it was strived to obtain more day. Cups, on top of that, can keep statistics. It can map your
information about IoT that is a technology revolution. liquid consumption by determining what you drank in what
Following the introduction, the main components of IoT time of day. It can suggest that you drink coffee when your
described in second section and current applications of IoT coffee time. Even it can order instead of you at a nearby
were discussed in the third section. Then the potential coffee shop [28].
difficulties beside the opportunities of IoT were mentioned in Actualization of IoT into the real world is possible by the
the fourth section. And study is terminated with conclusion integration of several effective technologies [9]. These
section. technologies are provided below.
____________________________________________________________________________________________________________
342
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the operations such as each maintenance, improvement. analytics is required. Storage and calculating tools for data
Dynamic information such as environmental conditions analytics are included in middleware. The data have to be
occurring during use of the product, working time and stored and used intelligently for smart monitoring and
frequency of the product is recorded with the help of sensors actuation. Many IoT applications require massive data storage,
[13], [15], [21]. Fig. 4 shows building blocks of IoT. huge processing speed to enable real time decision making,
and high-speed broadband networks to stream data. As of
2012, Cloud based storage solutions are becoming
increasingly popular. Cloud computing provides an ideal
back-end solution for handling huge data streams and
processing them for the numerous number of IoT devices and
humans in real time [1], [15], [21].
D. Presentation
Presentation is related to visualization and interpretation
tools. These tools can be accessed on any platform and can be
designed for a large variety of applications. Presentation
allowed the interaction of the user with the object in real time
anywhere, anytime is important for IoT systems [1], [15]. For
example, IoT-enabled home appliances and devices can be
monitored and controlled outside the user’s home through a
computer, tablet, or smart phone due to presentations function
of IoT [21].
____________________________________________________________________________________________________________
343
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
344
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
345
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
346
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[26] B. Yılmaz, 2015, “Internet of Things (IoT) Nedir?” [online], Available: [30] B. Dolay, (2016) [online], Available: http://e-bergi.com/y/Kablosuz-
https://burkanylmz.wordpress.com/2015/10/12/internet-of-things-iot- Sensor-Aglari/
nedir/ [31] B. Atakul, (2015), “Nesnelerin İnterneti ve Kullanım Alanları” [online],
[27] D. C. McFarlane, V. Giannikas, A. Wong and M. G. Harrison, Available: http://www.teknolo.com/nesnelerin-interneti-kullanim-
“Intelligent products in the supply chain-10 years on,” Information alanlari/
Control Problems in Manufacturing, vol. 14(1), pp. 655-660, May. [32] E.S. Bayrak Meydanoğlu, “IoT ve tedarik zinciri”, Lojistürk Dergisi,
2012. vol. 34, pp. 28, 2013.
[28] (2016) [online], Available: [33] M. He, C. Ren, Q. Wang, B. Shao, and J. Dong, “The Internet of
https://ooofoffline.wordpress.com/tag/nesnelerin-interneti/ Things as an enabler to supply chain innovation,” In e-Business
[29] (2016) [online], Available: http://www.smallcapnetwork.com/Small- Engineering (ICEBE), IEEE 7th International Conference on IEEE, pp.
Cap-Atmel-Corporation-ATML-Has-the-Internet-of-Things-Covered- 326-331, 2010.
AMBA-FLEX-CSCO/s/via/3414/article/view/p/mid/1/id/1803/ [34] O. Köroğlu, "Nesnelerin Interneti, algılayıcı ağları ve medya."
____________________________________________________________________________________________________________
347
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Boring is a finish operation that widely used to ES is a technique used in artificial intelligence discipline.
machine the holes on the mechanical parts in manufacturing Moreover, there are a lot of techniques such as neural
industry. The Boring also is a method that is usually preferred to networks, genetic algorithm, fuzzy logic, image processing, in
size holes that can meet the appropriate tolerance. There are artificial intelligence discipline. Especially, ES [1], [2] and
many tools for boring process in the world wide. Generally,
neural networks [3] well as various programs are used in the
Boring tools were parted two groups that are milling and
turning. The boring tools especially for milling were packed in process sequence. Also, ES is an intelligence program that
many set that include head, bars, etc. Decision of using any set in uncovers an expert thinking and behaviour for a specific area
these packed is not easy exercise. For this reason, selection [4]. ES can calculate the data and make inferences something
suitable head, bars, cutting tools and cutting conditions need long by using about information in the knowledge base. Thus, they
preparation time and good expertise for boring process in provide fast and reliable solutions for special problems. ES is
milling. Because of the quality of holes is depended by an interactive system for experts in many areas.
determining these parameters rightly Expert systems were improved with different capabilities
In this study, an expert system that is called BT_expert was for many areas since computers were invented. Especially,
developed for using the boring tools in milling. BT_expert was
these systems rapidly climbed high level in 20th century.
built up by using an expert system shell that is called Kappa PC.
Kappa PC is preferred because of programming with C++. However at present time, new expert systems have been
BT_expert has a friendly user interface that is designed visual implementing to solve problems and select optimal parameters
objects. A lot of rules about 70 are written for BT_expert. The (boring tools, material, cutting tools and cutting conditions) in
system asks a few simple questions to user about boring process. machining industry in the world. Some studies carried out in
And, BT_expert can make a decision by using forward chaining this direction are summarized below.
mechanism. Finally, BT_expert system determines the boring Tan, et al. [1] have designed a selection of cutting tools of
bars, cutting tools and ideal cutting parameters automatically. expert and have checked the accuracy of the system for CNC
As a result, BT_expert system makes easy to select boring and turning. An expert system called EXTOOL developed by D.
cutting tools, and cutting parameters correctly in many set
Mookherjee and Bhattacharyya [2]. The program stated that
without any expert. And, the system decrease the long
preparation time of boring process. the process for the customer's choice of material and types for
cutting lathes and milling enables automatic determination
Keywords—Boring, Boring tools, Milling, Expert System, Kappa based on geometry. An expert system for the selection of
PC cutting tools and turning conditions were developed by
Arezoo et al. [5]. Sapuan, et al. [6] were developed a system
I. INTRODUCTION of specialist material election for tools and components for
Nowadays, computers’ hardware and software use easily ceramic matrix composite. Er and Dias [7] were improved an
for many applications in many areas. For example, machining expert system engineering components, based on which the
is a very important area that is widely used computers and selection rule for the production of casting system. Kayir, et
programs. Computers and programs provide a lot of al. [8] were built up an expert system by using Leonardo shell
advantages, such as speed, accuracy, lower costs, etc. program for selection optimal cutting tools and operation for
Generally, computer programs have an algorithm. So, drilling holes. Poyrazoglu, et al. [9] were implemented an
solutions of these programs are not enough for all application. expert program to choose right operation from non-traditional
Because, an expert decision require for some application. In machining methods.
addition, many years need to be an expert of a person in a In this study, an expert system called BT_expert was
sector. For this reason, a lot of expert systems were developed developed for machining of various size holes. BT_expert
on the world. Expert Systems (ES) have very special selects the optimal boring tools that are commonly used for
programs. These systems solve a problem according to rules machining holes desired with geometrical tolerances. The
obtained from the knowledge of the experts. In other word, the system was built up with KAPPA-PC expert shell program.
ES evaluate knowledge, the classical programs that have BT_expert system contains 4 modules. They are database
algorithm evaluate evident data. Thus, ES provide to use module, reasoning (inference engine) module, user interface
expertise at any time without an expert person. module and knowledge base module. BT_expert makes a
decision for selecting an optimum boring cartridge, cutting
____________________________________________________________________________________________________________
348
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
tool (insert type and radius), cutting parameters (cutting speed, The other subject is that the length of the selected tool is
feed rate). It has user interface that includes screen pages not too longer than the hole’s length. In this study by looking
formed menus, buttons. User can input or change data on L/D rate, the determination of boring cartridge is done. L
these pages. The system can be used easily by any man shows deepness of the hole and D shows the diameter of the
(engineer, worker, etc.). BT_expert reduces planning time and final hole. Programme offers a higher level boring cartridge
improves product quality because of choice right boring when L/D rate is higher than 2,5. If this case is not paid
cartridge and the insert and cutting parameters for boring attention in Boring tool that will be used, chatters are
operations. observed over the hole surface. Abele et al., in their study,
placed a damping system to shaft of boring tool for preventing
II. BORING TOOLS chatters [11].
Boring machining can be referred as one of the difficult
process in machining. Boring process is widely performed on
the lathe and milling machine tools. Boring process is process
of completing finished final dimensions and tolerances by
upsizing parts of which first bore is drilled. The machining
process is performed as a finish work. After performing the
boring process, additional boring operations cannot be
performed on the hole. Hole is ready to use. Boring process is
similar to the hole turning on the lathe, but in this process
boring tool rotates not workpiece. Hole is machined by
performing an eccentric movement of the boring tool.
In boring process a lot of boring tools are used. Boring
tools exist in different sets that include a lot of parts in the
market. In addition to different types of boring tools in the
market, boring tools are classified according to specialities of Fig. 1 Boring tool set [14]
hole which will be processed. (rough and finish etc.).
Classified boring sets (Fig.1) are capable of machining Processing range of used boring set is given at Fig.2. If
specific diameter hole in specific circumstances. It is required given dimensions are not paid attention, surface defects and
that a boring cartridge is determined according to first hole dimension faults stemming from tool deflection are seen in
and final hole shown at Fig.1. hole surface. Apart from these, it is required to determine
One of the subjects which have to be considered is that cutting speed, feed rate and insert type and radius according to
diameter of the selected boring cartridge is not too smaller workpiece material.
than the first bore of workpiece.
____________________________________________________________________________________________________________
349
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
So, there are a lot of parameters correctly determined for Moreover, name of the rules and task, and numbers of
successful operation in boring machining. Especially, boring each task about 72 rules written in the BT_expert rule base are
tools, insert cutting tools and cutting parameters must be shown at Table 1.
chosen correctly to get precision and desired tolerance.
Mistakes made in their selection, cause faults on machined
Table 1. Rules name and numbers included in BT_expert system
holes and used tools. In a study, Rao et al. try to estimate the
occurred vibration, the surface roughness and the tool wear Sequence Rule name Task Number
during the machining hole on AISI 316 material by using 1 Barasecimi To choose for boring tool 22
2 Baralamaoranı To determine boring cartridge 10
artificial neural networks [10]. according to rate of bore diameter
/ boring diameter
III. BT_EXPERT SYSTEM AND BORING TOOL SELECTION 3 Radyüssecimi To determine insert 3
KAPPA-PC expert shell program was used to develop a 4 Ucparametreleri To determine cutting parameters 24
for insert
system that is called BT_expert system for holes machining.
5 Uyarı To redirect user according to enter 13
BT_expert system was designed for two main objects that are data and values
boring tools, cutting tools and cutting parameters. There are Total: 72
several factors for preference the KAPPA_PC shell program
to develop BT_expert. KAPPA PC expert shell program is Rules defined in the BT_expert rule base automatically
suitable program to improve any application in field of associated with each other. For example the connection of
machining sector [12]. In addition, KAPPA-PC is an object Barasecimi 1 with other rules in the BT_expert rule base is
oriented program. In KAPPA-PC program all objects are shown at figure 5. Generally, two methods are used to
connected each other with a tree structure form (Fig. 3.). evaluate rules for solution of any problem in expert shell
programs. These are forward chaining and backward chaining
method. The BT_expert rule base was designed to use forward
chaining method. The solution of the forward chaining
method starts from condition (IF) sentence and ends
conclusion (THEN) sentence. If the conditions are satisfied,
conclusion is true in “THEN” part.
Fig. 3 The tree schema of KAPPA ES developed for boring tool selection
Barasecimi 1
IF
(Bara:Soncap >= 6 ) And
(Bara:Soncap <= 8 ) And
(Bara:Delikboyu < 21 ) And
(Bara:Radius #= 0.4 );
THEN
SetValue( Bara:Baratipi, B3.06 );
Fig. 4 A sample rule used in BT_expert system Fig. 5 The connection of a rule (Barasecimi 1) with other rules
____________________________________________________________________________________________________________
350
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
351
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
352
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The Lyapunov exponents analysis is one of the In this study which differs from studies in the literature, the
stability analyses of nonlinear systems. In addition, this analysis Lyapunov exponent spectrum obtained faster via parallel
method is also used for obtaining information about chaos that is computing application in a PC with multi-core CPU. For this
a behavior of nonlinear systems. In this study, it is explained by application MATLAB Parallel Computing Toolbox was used.
an example application that the chaotic behavior analysis of a
The performance of the application was assessed according to
nonlinear chaotic system based on the Lyapunov exponents can
be performed faster by a Multi-Core CPU. For this application, speed and efficiency.
MATLAB parallel processing toolbox has been used and the
parallel computing performance of the application has been In the next section of this article, the algorithm that uses
analysed by using obtained results. Lyapunov exponent is introduced. In the third part, Lorenz
chaotic system and its Lyapunov exponent spectrum is
Keywords— Chaotic systems, Parallel processing, Lyapunov presented as an example. In the fourth part, the realized
exponents parallel computing application is mentioned. In the fifth part
the performance of the parallel computing is assessed. In the
I. INTRODUCTION last part, discussion and evaluation is given.
The models of nonlinear systems are used for modelling
and assessing the behavior of the systems that we encounter.
One of the behavior types of the nonlinear systems is the II. LYAPUNOV EXPONENTS ANALYSIS
chaotic behavior. For a nonlinear system to show chaotic
behavior, the system’s parameters and initial values must be The Lyapunov exponents are calculated analytically as in
appropriate to the chaotic behavior. The nonlinear systems Equation 1 [6].
which show chaotic behavior for the appropriate values are
called chaotic systems. With the Lyapunov exponents analysis 1 pi (t )
it can be determined that whether the system exhibits chaotic i lim ln (1)
behavior or not for the given parameters values.
t t pi (0)
____________________________________________________________________________________________________________
353
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
this, many algorithms for calculating the Lyapunov exponents In this study, faster obtainment of the Lyapunov exponents
from time series have been proposed. [5-9]. In this study, the spectrum with data parallel technique in a multi-core CPU PC
MATLAB code that is proposed by [11] which employs the was realized. The realized parallel computing application is
algorithm in [6] is used. seen in Fig. 2.
i 1
i The parallel computing application in Fig. 2, to obtain the
df j (2) Lyapunov exponents spectrum n core processor is used.
j 1 Moreover the system parameter take m number values
j between a and b. Each processor calculates Lyapunov
Here the term j is determined according to
i 1
i 0 and exponents for m/n number of system parameter values.
j 1
After this, these results are brought together to obtain the
i 1
i 0. Lyapunov exponent spectrum. The parallel computing
application mentioned above was realized with parfor loop in
MATLAB Parallel Computing Toolbox.
III. PARALLEL COMPUTING APPLICATION OF THE
A. MATLAB Parallel computing toolbox and Parfor Loop
LYAPUNOV EXPONENT ANALYSIS
____________________________________________________________________________________________________________
354
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
355
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
4 REFERENCES
Speedup
2
[1] A.M. Lyapunov, Probleme general de la stabilite du mouvment,
0 Annales Mathematical Study, 17, 1947.
2 4 6 8 [2] V.I., Oseledec, A multiplicative ergodic theorem: Lyapunov
Number of cores characteristic numbers for dynamical systems, Trans. Moscow Math.
1 Soc., 19, 197-231, 1968.
Efficiency
2003), pp.116-117.
2000
[6] A. Wolf, J. B. Swift, H. L. Swinney, and J. A. Vastano, Determining
Lyapunov Exponents from a Time Series, Physica D 16, 285-317
0
2 4 6 8 (1985).
Number of cores [7] M. T. Rosenstein, J. J. Collins and C. J. De Luca, A practical method
for calculating largest Lyapunov exponents from small data sets,
Fig. 5. Graphs of speedup, efficiency and cost analysis Physica D, 1993.
[8] Hai-Feng Liu, Zheng-Hua Dai, Wei-Feng Li, Xin Gong, Zun-Hong Yu,
Noise robust estimates of the largest Lyapunov exponent,Physics
When Fig. 5 is examined, the result is obtained faster with Letters A 341, 119–127, 2005.
parallel computing. As the number of the core increases the [9] X. Zeng, R. Eykholt, and R. A. Pielke, Estimating the Lyapunov-
speed of the calculation is increased. However, as the number Exponent Spectrum from Short Time Series of Low Precision, Physical
Review Letters, Vol. 66, Number 25, 1991.
of the core increases the cost and efficiency for the obtainment [10] http://sprott.physics.wisc.edu/chaos/lespec.htm
of the Lyapunov exponents spectrum with parallel computing [11] http://www.mathworks.com/matlabcentral/fileexchange/4628-
does not improve like speedup. In fact, the efficiency calculation-lyapunov-exponents-for-ode/content/lyapunov.m
decreases dramatically and the cost increases as the number of [12] http://fmwww.bc.edu/repec/bocode/l/lyaprosen.m
[13] http://www.mpipks-dresden.mpg.de/~tisean/Tisean_3.0.1/index.html
the core increases. This situation is encountered in most of the [14] http://www.scientio.com/Products/ChaosKit
application in PCs. The reason for this the increase in the [15] S. Kaçar, Z. Ekşi, A. Akgül, F. Horasan, Computation of Shannon
parallel overhead with the number of the core used and Entropy with MATLAB Parallel Computing Toolbox, ISITES 2013,
reaching the maximum capacity of PC’s data transmission rate. 765-773.
[16] T. Rauber, G. Rünger, Parallel Programming For Multicore and
This leads to negative effect on both cost and efficiency. Cluster Systems, Springer (2010).
Another reason for low efficiency and high cost is the CPU
has not real 8 cores but virtual 8 cores. Thus, the increase in
speedup and efficiency is much more in the computing
application with 2 and 4 cores.
IV. CONCLUSIONS
In the proposed study, an application of parallel computing
for faster obtainment of Lyapunov exponents spectrum is
realized with a multi-core PC. In the application, as the
number of core used increases the speed of the computing
application increases, but the cost increases and efficiency
decreases dramatically because of the PC’s limitations such as
parallel overhead and data transmission rate .However,
employing parallel computing for obtainment of the Lyapunov
exponents spectrum is a sound approach when considering the
speed of the computing. For more complex systems, the
number of values to be calculated and the dimension of the
problem increases, hence the efficiency of the parallel
computing increases. In a future study, the Lyapunov
exponents analysis of more complex systems will be
examined.
____________________________________________________________________________________________________________
356
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Estimation of Turkey Electric Energy Demand Until
Year 2035 Using TLBO Algorithm
M. Fatih TEFEK*, Harun UĞUZ+
*
Ahi Evran University,
Kırşehir/Turkey,
mftefek@ahievran.edu.tr
+
Selcuk University,
Konya/Turkey,
harun_uguz@selcuk.edu.tr
Abstract— In this study, the estimation of Turkey primary electric agreements between countries. The excessive deviation leads to
energy demand until 2035 is tried to estimate by using Teaching- the economical deficits.
Learning Based Optimization (TLBO) Algorithm. Two models are Although demand studies on energy have been studies in
proposed which are based on economic indicators TLBO algorithm Western countries for a long time, the studies have accelerated
linear energy demand (TLBOEDL) and TLBO algorithm quadratic
at the end of 1990s in our country. State Planning Organization,
energy demand (TLBOEDQ). In both of these two models the
indicators used are Gross Domestic Product (GDP), population, State Statistical Institute and Ministry of Energy and Natural
importation and exportation. After a comparison of these two Sources have been used the mathematical models for the energy
models with real values between 1979 and 2005 years, it is applied demand at the last of 1970s [4].
to the estimation of Turkey electric energy demand until 2035 by Recently, many studies on the Turkey primary energy demand
three different scenario. The estimation results are suitable with have been done. In these studies, different methods, different
the estimation of Turkey total primary energy supply of 2013 scenarios with various data and variables for many terms have
Energy Report of World Energy Council Turkish National been used [3]. Some studies on this research area are the energy
Committee (WEC-TNC ). demand by Genetic Algorithm [3-7], energy consumption
estimation by Artificial Neural Networks (ANN) [8-10], energy
Keywords— Teaching Learning Based Optimization (TLBO)
demand by Ant Colony Optimization (ACO) [11] and energy
Algorithm, Energy Demand Estimation, TLBOEDL Model,
TLBOEDQ Model, Turkey Energy Report 2013 demand estimation by hybrid Algorithm [12].
Generally, any event which is wanted to model in terms of
I. INTRODUCTION estimation by mathematically, as the numbers of variables are
getting increased, correct estimation possibility decreases [3]. In
Energy is the most important need of developed and this study, the estimation of Turkey energy demand until 2035 is
developing countries. Energy is an indispensable fact which its try to estimate by using some economical variable indicators
importance has been growing gradually. It is the most important such as Gross Domestic Product (GDP), population, importation
means in the development of countries. Energy consumption has and exportation.
been increased with the growth in world population. The
economic growth leads to energy consumption and energy II. TEACHING LEARNING BASED OPTIMIZATION (TLBO)
consumption leads to economic growth [1]. Therefore the ALGORITHM
energy demand has been increasing despite important nuclear
power plant accidents and financial crisis. Foreign energy TLBO algorithm is a social based optimization algorithm
dependency of Turkey as % 70 percentage obliges the special which depends on interaction between students and teachers in a
policies and special behavior styles as society on energy subject class. At every step of algorithm, successful students are elected
[2]. and the best students have been determined [13].
To meet the electrical energy of society economically, supply, It has three parameters which is the number of students,
demand, transmission, contribution and pricing policies should number of classes and iteration number. TLBO algorithm has
be constituted effectively. Moreover, ease of use of electrical two phases which are teacher and student phases.
energy for consumers, non-storability of electrical energy At teacher phase of the algorithm, students learn from the
increases the importance of demand studies on electrical energy teacher by imitating. Teacher is the most experienced and the
area [3]. Energy demand estimation has a critical role in energy most informed person so the best student can learn as much as
the teacher.
____________________________________________________________________________________________________________
1
357
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Between teacher and student’s learning capacity, there is an Table I: Energy demand, GDP, population, import and export data of Turkey
[17,18].
average difference called difference mean and it is defined as
Eq. 1, Energy GDP Population Import Export
Difference_Mean j,i = ri X j,kbest,i -Tf M j,i (1) Year Demand
( $109 ) ( 106 ) ( $109 ) ( $109 )
(MTOE)
where ri is a random number between 0 and 1, X j ,kbest ,i is the 1979 30,71 82,00 43,53 5,07 2,26
1980 31,97 68,00 44,44 7,91 2,91
result of teacher (the best result) and T f is teaching factor 1981 32,05 72,00 45,54 8,93 4,70
between 1 and 2. 1982 34,39 64,00 46,69 8,84 5,75
1983 35,70 60,00 47,86 9,24 5,73
T f is defined as Eq. 2,
1984 37,43 59,00 49,07 10,76 7,13
Tf = round 1+rand 0,1 1,2 (2) 1985 39,40 67,00 50,31 11,34 7,95
1986 42,47 75,00 51,43 11,10 7,46
If difference mean is better than present result, Eq. 2 is 1987 46,88 86,00 52,56 14,16 10,19
arranged as Eq. 4, the best function result is given by Eq. 3, 1988 47,91 90,00 53,72 14,34 11,66
1989 50,71 108,00 54,89 15,79 11,62
X 'j,k,i X j,k,i Difference _ Mean j,k,i (3) 1990 52,98 151,00 56,10 22,30 12,96
' 1991 54,27 150,00 57,19 21,05 13,59
where X j , k ,i is the best function result accepted. 1992 56,68 158,00 58,25 22,87 14,72
After teacher’s phase, all best function values are kept to use 1993 60,26 179,00 59,32 29,43 15,35
1994 59,12 132,00 60,42 23,27 18,11
at student’s phase. At this stage, students learn the knowledge by
1995 63,68 170,00 61,53 35,71 21,64
interacting and by discussing between them. If a student is 1996 69,86 184,00 62,67 43,63 23,22
more knowledgeable, the other is updating himself by 1997 73,78 192,00 63,82 48,56 26,26
interaction. 1998 74,71 207,00 65,00 45,92 26,97
P and Q are the random students are given in Eq. 4, 1999 76,77 187,00 66,43 40,67 26,59
X total P,i X 'total Q,i (4)
'
2000 80,50 200,00 67,42 54,50 27,78
2001 75,40 146,00 68,37 41,40 31,33
where X 'total P,i and X total
'
Q ,i
are updated values of X total P,i 2002 78,33 181,00 69,30 51,55 36,06
2003 83,84 239,00 70,23 69,34 47,25
and X total Q,i
2004 87,82 299,00 71,15 97,54 63,17
If X 'total P,i X 'total Q,i X ''j,P,i is obtained as Eq. 5
The linear model proposed by TLBO algorithm as in Eq. 7,
X ''j,P,i X 'j,P,i ri ( X 'j,P,i X 'j,Q,i ) (5)
and if X 'total Q,i X 'total P,i X ''j,P,i is obtained as Eq. 6 ETLBOEDL =w1.X1+w2.X2 +w3.X3 +w4.X4 +w5 (7)
X ''j,P,i X 'j,P,i ri ( X 'j,Q,i X 'j,P,i ) (6)
Quadratic model equation is given in Eq. 8,
____________________________________________________________________________________________________________
2
358
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
respectively. The objective function in energy demand The teacher will try to shift the mean from M, D towards
estimating is given by Eq. 9,
W,teacher which will act as a new mean for the iteration. So,
m
min f(v)= (Eobserved
r -E predicted
r ) 2 (9)
M_new, D =Wteacher,D (14)
r=1
____________________________________________________________________________________________________________
3
359
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TLBOEDQ
TLBOEDQ
TLBOEDL
TLBOEDL
f (v)TLBOEDQ is obtained as 19,32921. energy
Years demand
(MTOE)
E TLBOEDQ = 0,01833664807X1 + 0,1271526286X 2
0,7303814145X 3 + 1,5173167741 X 4 2006 99,59 95,96 95,19 -3,78 -4,62
+ 0,0057624768 X1X 2 + 0,0118978356X1X 3 2007 107,63 98,90 98,62 -8,83 -9,14
2008 106,34 101,94 102,21 -4,32 -4,04
0,0079494952X1X 4 0,0013078678X 2 X 3 2009 106,14 105,09 105,97 -1 -0,16
0,0149756126X 2 X 4 + 0,0365796927X 3X 4 (18) 2010 109 108,34 109,91 -0,61 0,83
0,0017245010X12 + 0,0096564193X 22 2011 115 111,71 114,01 -2,95 -0,87
2012 121 115,19 118,30 -5,04 -2,28
0,0234475391X 23 0,0197717121X 24
2,0471885591 Table IV: Energy demand estimation of TLBOED between 2006 and 2012
years according to scenario 2.
TLBOEDL and TLBOEDQ estimation models were executed Estimated energy Relative errors
for 50 times and best results were considered by given demand (MTOE) (%)
parameters in Eq. 17 and Eq. 18.
TLBOEDQ
TLBOEDQ
Observed
TLBOEDL
TLBOEDL
TLBOEDL and TLBOEDQ models which are formed by energy
taking the data from Table II are effective and successful Years demand
according to data between 1996 and 2005 years. (MTOE)
In Table II, the biggest decrease in energy demand is in 2001. 2006 99,59 96,14 95,32 -3,59 -4,48
In Table I, a considerable decrease is seen for the same year 2007 107,63 99,29 98,91 -8,4 -8,82
2001. This situation results from the economic crisis in 2001. 2008 106,34 102,55 102,7 -3,7 -3,54
2009 106,14 105,93 106,68 -0,2 0,51
Table II: Energy demand estimation of TLBOED models between 1996 and 2010 109 109,44 110,86 0,4 1,68
2005 years.
2011 115 113,08 115,26 -1,7 0,23
Estimated energy Relative errors 2012 121 116,87 119,86 -3,53 -0,95
Observed demand (MTOE) (%)
energy
demand Table V: Energy demand estimation of TLBOED between 2006 and 2012 years
Linear Quadratic Linear Quadratic
Years (MTOE) according to scenario 3.
1996 69,86 69,71 69,85 -0,21 -0,02
Estimated energy Relative errors
1997 73,78 72,32 72,71 -1,99 -1,45
demand (MTOE) (%)
1998 74,71 73,30 74,28 -1,89 -0,58
TLBOEDQ
TLBOEDQ
Observed
TLBOEDL
TLBOEDL
1999 76,77 74,18 75,27 -3,37 -1,95 energy
2000 80,50 80,71 80,96 0,27 0,57
____________________________________________________________________________________________________________
4
360
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TLBOEDQ
TLBOEDQ
TLBOEDL
TLBOEDL
TLBOEDL
165
150
135
Years 120
105
2013 118,8 122,8 120,8 124,7 114 116,1 90
2014 122,5 127,4 124,9 129,7 116,9 119,6
2006
2007
2008
2009
2010
2011
2012
2015
2020
2023
2025
2030
2035
2015 126,4 132,2 129,1 135 120 123,3
2016 130,4 137,2 133,5 140,5 123,2 127,1
2017 134,5 142,4 138,1 146,1 126,5 131,1 Figure 1 : Energy demand estimations realized between 2006 and 2035 years
2018 138,8 147,7 142,9 152 129,8 135,1 according to scenario 1 in MTOE
2019 143,3 153,1 147,9 158,1 133,3 139,3
2020 147,9 158,7 153,1 164,3 136,9 143,6 TLBOEDL TLBOEDQ WEC-TNC
2021 152,6 164,5 158,5 170,7 140,7 148
2022 157,6 170,3 164,1 177,3 144,5 152,5 285
2023 162,7 176,2 169,9 183,9 148,5 157,1 270
2024 168,1 182,2 176 190,5 152,6 161,7 255
240
2025 173,6 188,1 182,4 197,2 156,8 166,5 225
2026 179,4 194,1 189 203,8 161,2 171,2 210
195
2027 185,3 200 196 210,3 165,7 176 180
2028 191,5 205,8 203,2 216,6 170,4 180,9 165
2029 198 211,4 210,7 222,5 175,3 185,7 150
135
2030 204,7 216,8 218,6 228,1 180,3 190,4 120
2031 211,6 221,8 226,8 233,1 185,5 195,1 105
90
2032 218,9 226,4 235,4 237,4 190,9 199,6
2006
2007
2008
2009
2010
2011
2012
2015
2020
2023
2025
2030
2035
____________________________________________________________________________________________________________
5
361
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] T. Lorde, K. Waithe, B. Francis, “ The importance of electrical energy for
economic growth in Barbados”, Energy Economics, vol. 32(6), pp. 1411-
1420, Nov.2010.
[2] Turkey Energy Report 2013, World Energy Council Turkish National
Committee (WEC-TNC), ISSN: 1301-6318, Ankara, Jan. 2014.
[3] V. Yiğit, “Estimation Of Turkey Net Electric Energy Consumption Until
to Year 2020 Using Genetic Algorithm”, International J. of Engineering
Res. and Development, Vol. 3, pp. 37-41, Jun. 2011.
[4] Haldenbilen, S., Ceylan, H., “Genetic algorithm approach to estimate
transport energy demand in Turkey”, Energy Policy, Vol. 33(1), pp. 89-
98, Jan. 2005.
[5] O. E. Canyurt., H. Ceylan, H.K. Öztürk., A. Hepbaşlı, “Energy demand
estimation based on two-different genetic algorithm approaches”, Energy
Sources, vol. 26, pp. 1313–20, Feb. 2004.
[6] H. Ceylan, H. K. Öztürk, “Estimating energy demand of Turkey based on
economic indicators using genetic algorithm approach”,
Energy Convers Manag, vol. 45(15–16), pp. 2525–37, Sep. 2004.
____________________________________________________________________________________________________________
6
362
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract — Nowadays, the author of the distinguish, and made studies on the presence of the
recognition process, the development of technology and is author. language-dependent studies, language syntax
made to find solutions to some problems emerged with the structure that makes identifying the author examines
spread of knowledge. Some of these problems, author how writers use. We follow a columnist, without
unknown documents, e-mail the threat, and the author of name and even when the article was published on a
the criminal investigation is to determine the author of the different page, we can guess who wrote it in writing.
text can’t be sure who. Authors identification is the process We have become used to the author's writing style
of determining who is the author of a text written in any because we follow. We could do this to separate, we
language. This document, based on the language made the can make the computer automatically? The answer
last 15 years and regardless of the language, the author to this question is yes, of course. Every writer has a
presents a review about the identification their work. The unique style. This style is converted to a
authors recognize the language-dependent work in the
mathematical function author recognition process is
language of the syntax of specific words in the uses of the
realized. Turkish studies done on this subject is
language by the author and the word of their arrangement
limited [1]. Author identification is classified into
with the order in which to do that by paying attention to
three different research areas which include
the author of the text is the process of estimating different
authorship identification, authorship verification,
algorithms starting from here. The author of the
and authorship characterization [2]. Authorship
independent study of the language of the unknown text
with the character n-gram methods of machine learning is
attribution determines the most probable author of an
the process after the calculation of weight and estimating anonymous document by comparing it with the
Based on the text of the previous authors using artificial known available documents. Authorship
neural networks. Authors frequently used in the characterization is helpful to determine the
identification, used in the selected article, methods have characteristics like gender, age etc. [3]. Author
been investigated and their authors compared the accuracy identification of physical and electronic documents,
of the identification. Authors alone can’t fully it is close to an area of research, and many
identification the many methods used to give accurate researchers are currently working in this research
results and language used by the author using some field [4].
combination of these methods and it was emphasized that This paper consists of three parts. the first part
the words have to look at trends in the author's writings. will be released in the definition of author
identification. In the second part, the author
Keywords— Author identification, Writer Identification, identification the field, made different studies. In the
Language-Dependent Studies third part of the study measurements it is included.
I. INTRODUCTION
Document classification and verification is a II. AUTHOR IDENTIFICATION AND
matter of in natural language studies. If our goal is to AUTHOR ANALYSIS
separate the types of documents. We use properties,
properties that are used to find the author of a author identification the unknown author of the
document, may be different from each other. Author document the author is no process. This process,
identification work is a subject studied for many artwork, criminal investigation, online messages and
years, artificial neural networks, statistical methods handwritten etc. It is working in the fields. However,
and machine learning, such techniques are used for the online documentation for writing verification can
author identification. With the advancement of be used to solve various crime cases such as
technology, who written by the unknown online text, extortion and terrorist activities [9]. online messages,
author identification work has increased. The there are difficulties to implement the writer
authors identify a lead in the other study authors are identification. these challenges are; Length of Online
handwritten identification process [5]. In recent document is short, an online message, there is less
years, terrorist acts in the use of online messages, regular font style, knows the right words to be
these events may be subject to messages, to
____________________________________________________________________________________________________________
363
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
written, Online documentation style and structure is Found nearest feature vectors, Euclidean distance is
different from the normal text format documents. used. The vector k found that most of what class, the
The different languages spoken in the world, making class label is defined as a class of data to be sorted.
it difficult to the fight against international
cybercrime. along with the challenges mentioned, IV. STUDIES ON AUTHOR IDENTIFICATION
crime analysis of online posts by author analysis it
can be used for various applications. [2]. authorship Document classification and author
analysis has been applied to various cybercrimes. for identification studies so far, various features are
example, malware [6,7], it is used in the web spam used. Burrows [10] based on the word wealth
measures [8]. characteristics, Stamatelatos and friends [11]
The author analysis two different machine various combinations of syntactic features of style,
learning algorithm is used as the basis. they are Fürnkranz [12] n-grams, Authorship analysis is the
supervised and unsupervised machine learning. The study of linguistic and computational characteristics
supervised learning algorithms are also suitable of the written documents of individuals [14,15].
when the information of both training and testing special article features extracted from text written in
data that is the data’s label is known beforehand [3]. advance of individuals can be used to distinguish
another person [16]. writing styles can be
III. AUTHOR CLASSIFICATION categorized in five different types, these namely
lexical, syntactic, structural, content-specific, and
Today, a lot of sources of information is held in idiosyncratic features [4,13]. Authorship analysis
the form of unstructured text documents in electronic has been very successful for resolving authorship
form. In this case, users can find what they are identification disputes over literary and conventional
looking for them, an increasing number of writings [17]. online document analysis is more
documents from the documents research, analysis, difficult. [18]. There are many applications for
revealed the deal requires the grouping. plagiarism detection and similarity determination.
automatically parses documents to predetermined advancing technology, facilitate their replication and
class system being developed to find solutions to this spread [19]. Similarly, taking different user names of
situation. Recently, Support Vector Machine, Naive customers, protect the reputation of the company
Bayes, K-Nearest a machine learning method is used [20]. have developed techniques for detecting aliases
in the neighbour-hood, such as text classification in online systems (e.g., eBay) by analyzing the users’
work. Artificial Immune Systems is a method has not feedback. Abbasi et al [21]. The first approach
been tested before, and when appropriate work in conducts verification using classifica-tion; three
text classification feature vectors used in the trials, different classifiers are investigated, namely,
particularly in species recognition system has shown Adaboost.M1, Bayesian Network, and
that successful results. They are frequently used Discriminative Multinomial Naive Bayes (DMNB).
classification methods; The second approach conducts verification by
Naïve Bayes (NB). Classic Naïve Bayes regression; three different classifiers were studied in-
algorithm often used in the determination of a cluding linear regression, SVM with Sequential
document with the word class and the class of the Minimum Optimization (SMO), and SVM with RBF
joint probability [35]. kernel. The feature set was composed of 292
Support Vector Machines (SVM). Class biggest attributes, which included lexical (collected either in
margin separating from each other, aims to set a terms of characters or words), syntactic (punctuation
distinctive linear function. Linear inseparable as and function words), idiosyncratic (spelling and
examples for examples, moved to another higher- grammatical mistakes) and content-specific
dimensional space could be divided and linear (keywords commonly found in a specific domain).
classification that is done in space. Experimental evaluation of the proposed approach,
Random Forests (RO). Breiman to produce a using the Enron e-mail corpus and by analyzing200
single decision tree Instead, each trained a large e-mails per author, yielded EER ranging from 17.1%
number of different training sets, He proposed the to 22.4% [22]. Canales et al. trained a K-Nearest
merging of the multivariate decision tree. Different Neighbor (KNN) classifier with 82 stylistic features
sets of training boot (bootstrap) and random by including 49 character-based, 13word-based, and 20
selecting the feature created from the original syntactic features [23]. In addition, they combined
training set. Very variable decision trees obtained by stylometry and keystroke dynamics analysis for the
CART algorithm [36]. Before any decision tree gives purpose of authenticating online test takers. They
its decision, forest in-class maximum votes, which experimented with 40 students with sample
considered the final decision it is and incoming test document size rang-ing between 1710 and 70300
data is included in that class. characters, and obtained as performances
K-Nearest Neighboring (K-NN). class of data (FRR=20.25%, FAR=4.18%) and (FRR=93.46%,
which will feature vector value is based on the FRR=4.84%) when using separately keystroke and
principle of finding the closest feature vector k. stylometry, respectively. The combination of both
____________________________________________________________________________________________________________
364
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
types of features yielded EER=30%. They concluded investigator wants to precisely extract the writing
that the feature set must be extended and certain type styles of each suspect from the sample messages, use
of punctuations may not neces-sarily represent the such patterns to identify the author of the anonymous
style of students when taking online exams [23]. For message, and present such patterns as evidence to
the author analyzes explored the use of stylometry support the finding. Most of the previous works on
for online posts [24]. The technique was based on a authorship identification [4,33,34,13,27] assume
combination of supervised learning and n-gram
that every suspect has only one writing style. We
analysis. The experimental evaluation yielded an
argue that a person’s writing style may vary
EER of 14.35%for 87users for message blocks of
500 characters. The evaluation used real-life dataset depending on the recipients or the topics. For
from Enron, where the e-mails were combined to example, when a student writes an e-mail, his writing
produce a single long message per individual, and style to a professor is probably different from his
then divided into smaller blocks used for authorship writing style to a friend. The challenge is how to
verification. Advanced Information Networking and precisely identify such stylistic variations and utilize
Applications (ANE-2014) on 28 IEEE presented at the variations to further improve the accuracy of
the International Conference on the current paper, in authorship identification.
an earlier version, feature set is expanded and
Information Gain (IG) was used to rank the metric B. Authorship identification with small
good features. also used for SVM classification. training samples: Given a collection of anonymous
Corpus based on a data set used to evaluate messages from a group of suspects, a cybercrime
experimentally. Mutual information (MI) called for investigator wants to determine the author of each
an extra filter is added in the process of feature anonymous message in the collection. Unlike the
selection to take high correlation properties. SVM previous problem, this problem assumes that the
classification technique of the authors only, but also investigator has access to only a few training
L (SVM-LR) with SVM as a hybrid method that
samples written by the suspects. In real-life
combines research. The approach proposed two
investigation, the investigator can ask a suspect to
different data sets, based on a twitter feed, and other
email-based Enron 140, consisting of 280 blocks and produce a sample of his writing by listening to a
500 characters of text are evaluated using a text story or watching a movie, then reproducing the
message [25]. verbal elements extracted from a text played scene in his own writing. Clearly, the number
(word or character) consists of a series [26]. This of samples is very limited. The major challenge is
series of uppercase letters, lowercase letters, vowels, how to identify the author of the anonymous
white space, digits, including the frequency of messages when there are insufficient training data to
characters includes special characters [27]. build a classifier [4,27] or to extract any significant
automatic recognition task writer, author patterns.
attributions, and has various practical applications
for plagiarism detection. the author is a challenging C. Authorship characterization: Given a
area of recognition [28]. The structure of a document collection of anonymous text messages, a
can be easily interpreted in the machine side. Each cybercrime investigator sometimes has no clues
author's style, interpretation, natural language, is about who the potential suspects are and, therefore,
subjective general; it is more difficult to find an has no training samples of the suspects. Yet, the
explanation that allows the machine to automatically investigator would like to infer characteristics, such
tell a writer from another. Is reported in, a literature as gender, age group, and ethnic group, of the
review on modern authorship attribution methods,
author(s) based on the writing styles in the
usually coming from the fields of machine learning
anonymous messages. We assume the investigator
and statistical analysis [28,29,30,31,32]. To resolve
the problem of identifying the author, with Canadian has access to some external source of text messages
law enforcement agencies were identified in a joint such as blog postings and social network websites
project of writing analysis, three issues have been that disclose the authors’ public profiles. The
resolved [8]. challenge is how to utilize such external sources to
A. Authorship identification with large infer characteristics of the authors of the anonymous
training samples: A cybercrime investigator wants to messages.
identify the most plausible author of an anonymous Author identification studies, they often use
text message from a group of suspects. We assume the n-gram methods before learning machine. N-
that the investigator has access to a large collection gram choices are important. Examples of studies in
of messages that are previously written by suspects. [37]. To extract the dominant character n-grams in a
In real-life investigation, the sample text messages corpus modified the algorithm LocalMaxs
introduced in [38]. Then, the dominant n-grams are
can be obtained from the suspects’ e-mail archives
selected according to the following rules:
and chat logs on the seized personal computer(s), or
if (C.length>3)
from e-mail service providers with warrants. An
____________________________________________________________________________________________________________
365
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
366
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
367
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Neighborhood topologies are extensively used in Artificial bee colony algorithm (ABC) is an another pop-
Particle Swarm Optimization (PSO). The structure of the ulation based technique [10]. ABC uses individual which
selected topology may affect the PSO algorithm behavior. is given name as food source. Neighborhood structure of
Therefore, neighborhood topology plays a crucial role on the
performance of PSO algorithms when the population moves food sources in ABC algorithm is similar to gbest or fully-
with the guidance of the best particles. However, in Artificial connected topology. It means, all food sources communicate
Bee Colony (ABC) algorithms, new population members are with each others. In this study, we have adopted PSO topol-
generated based on information exchange between the base ogy mechanism to original ABC and global-best ABC al-
food source and a selected member among all population. gorithms. We have tested fully-connected, ring, four-cluster,
From this point of view, ABC algorithms use fully-connected
neighborhood topology defined as in PSO algorithms. In this square and random topologies on the original ABC and
paper, we will investigate the contributions of some well-known global-best ABC (GABC) algorithms on the 19 benchmark
neighborhood topologies used in PSO on the performance functions suite (SOCO) from a special issue of the Soft
of ABC algorithms. We have tested fully-connected, ring, Computing journal.
four-cluster, square and random topologies on the original
ABC and global-best ABC (GABC) algorithms on the 19 II. BACKGROUND
benchmark functions suite (SOCO) from a special issue of the
Soft Computing journal. SOCO benchmark suite consists of A. ABC Algorithms
7 shifted unimodal and 4 shifted multi-modal and 8 shifted 1) Original Artificial Bee Colony: Artificial Bee Colony
hybrid functions. Experimental results reveal that population
neighborhood topologies influence the performance of ABC
Algorithm is proposed for numerical optimization. It imitates
algorithms related to the tackling problem type. smart behaviors of the real honey bee. According to pseudo-
Keywords Artificial bee colony; Particle Swarm Optimization; code of ABC which is listed in Algorithm 1, ABC has
Swarm intelligence; Single objective optimization; Neighbor- one initialization step and three repeating steps. Until a
hood topology. termination condition is met, Employed bees, onlooker bees
I. I NTRODUCTION and scout bees steps continue. Further details about this
regarding steps are given bellow.
Particle swarm optimization (PSO) which is a population ABC algorithm in initialization steps which uses equation
based algorithm for global optimization, is introduced by 1 to create uniformly randomly initial population of SN (SN
Kennedy and Eberhart [1]. It tries to find a solution with is the number of food sources) solutions where each solution
individuals called particles. xi (i = 1, 2, . . . , SN ) is a D-dimensional vector (D is the
In Standard PSO, particles communicate with each other number of parameters).
based on social network structure called topology [2], [3].
Accordingly, each particle in the swarm can exchange infor- xij = xmin + rand(0, 1)(xmax − xmin ) (1)
j j j
mation with other individuals. But some particles have infor-
mation that may be useless, or some particle of information where i = 1, 2, . . . , SN , j = 1, 2, . . . , D; i is the index of
may be more important. For instance, the best-so-far particle food source and j is the index of dimension. xij represents
which called as global best particle (gbest) communicates a solution, xmin
j is lower bound value, xmax j is upper bound
with all particles in the neighborhood structure. In literature value of the jth dimension.
several research have been proposed for topology structure In the employed bee steps, algorithm searches new
and various type of topology structure is recommended [4], promised candidate solution. By using equation 2; ABC
[5], [6], [7]. Neighborhood Topology has a significant impact generates new food source and evaluates its fitness value.
on the performance of the algorithm [3]. For example, gbest Between old and new solution, greedy selection is applied.
or fully connected topology converge more quickly but it If the new is better than the old one, the new source position
may easily trapped in a local optima [8]. On the other is memorized and old one is forgotten. If not, position of the
hand; another topology which is lbest or the ring have slow old food source is kept.
convergence speed, but it can escape from the local optima
[9]. vi,j = xi,j + φi,j (xi,j − xr1,j ), (2)
____________________________________________________________________________________________________________
368
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
369
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
370
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I
D ESCRIPTION OF THE SOCO BENCHMARK FUNCTION SUITE .
TABLE II
PARAMETERS SETTING OF THE ALGORITHMS
[8] J. J. Liang, A. K. Qin, P. N. Suganthan, and S. Baskar, “Compre- multimodal functions,” Evolutionary Computation, IEEE Transactions
hensive learning particle swarm optimizer for global optimization of on, vol. 10, no. 3, pp. 281–295, 2006.
____________________________________________________________________________________________________________
371
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III
G IVEN ARE MEAN FUNCTION VALUES OF ABC WITH TOPOLOGY FOR EACH FUNCTIONS OF DIMENSION 50 FROM THE SOCO BENCHMARK SET
Rank 2 1 4 3
[9] J. Kennedy and R. Mendes, “Population structure and particle swarm “Memetic algorithms based on local search chains for large
performance,” Proceedings of the 2002 Congress on Evolutionary scale continuous optimisation problems: MA-SSW-Chains,”
Computation, CEC 2002, vol. 2, pp. 1671–1676, 2002. Soft Computing, vol. 15, no. 11, pp. 2201–2220, nov
[10] D. Karaboga, “An idea based on honey bee swarm for numerical 2011. [Online]. Available: http://www.mitpressjournals.org/doi/
optimization,” Technical report-tr06, Erciyes university, engineering abs/10.1162/evco.2010.18.1.18102{\#}.Va48PVM-P3Ahttp://www.
faculty, computer engineering department, Tech. Rep., 2005. mitpressjournals.org/doi/abs/10.1162/evco.2010.18.1.18102http:
[11] G. Zhu and S. Kwong, “Gbest-guided artificial bee colony algorithm //link.springer.com/10.1007/s00500-010-0647-2
for numerical function optimization,” Applied Mathematics and Com- [13] T. Liao, D. Aydın, and T. Stützle, “Artificial bee colonies for contin-
putation, vol. 217, no. 7, pp. 3166–3173, 2010. uous optimization: Experimental analysis and improvements,” Swarm
[12] D. Molina, M. Lozano, A. M. Sánchez, and F. Herrera, Intelligence, vol. 7, no. 4, pp. 327–356, 2013.
____________________________________________________________________________________________________________
372
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE IV
G IVEN ARE MEAN FUNCTION VALUES OF GABC WITH TOPOLOGY FOR EACH FUNCTIONS OF DIMENSION 50 FROM THE SOCO BENCHMARK SET. T HE
VALUES IN PARENTHESIS ARE THE AVERAGE NUMBER OF FUNCTION EVALUATIONS TO REACH OPTIMUM VALUE .
Rank 3 1 4 2
____________________________________________________________________________________________________________
373
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In many different fields, there is a high demand for re-utilize this information, it would very difficult to read or
storing information to a computer storage disk from the data query text or other information from these image files.
available in printed or handwritten documents or images to later Therefore a technique to automatically retrieve and store
re-utilize this information by means of computers. One simple information, in particular text, from image files is needed. Of
way to store information to a computer system from these
course, this is not a very trivial task. Some major challenges
printed documents could be first to scan the documents and then
store them as image files. But to re-utilize this information, it need to be laid out and handled in order to achieve a
would very difficult to read or query text or other information successful automation. The font characteristics of the
from these image files. Therefore a technique to automatically characters in paper documents and quality of images are only
retrieve and store information, in particular text, from image some of the recent challenges. Due to these challenges,
files is needed. Optical character recognition is an active research characters sometimes may not be recognized correctly by
area that attempts to develop a computer system with the ability computer system. Thus there is a need of mechanisms of
to extract and process text from images automatically. The character recognition to perform Document Image Analysis
objective of OCR is to achieve modification or conversion of any (DIA) which overcomes these challenges and produces
form of text or text-containing documents such as handwritten
electronic format from the transformed documents in paper
text, printed or scanned text images, into an editable digital
format for deeper and further processing. Therefore, OCR format.
enables a machine to automatically recognize text in such Similarly, Optical Character Recognition (OCR) is the
documents. Some major challenges need to be recognized and process of modification or conversion of any form of text or
handled in order to achieve a successful automation. The font text-containing documents such as handwritten text, printed or
characteristics of the characters in paper documents and quality scanned text images, into an editable digital format for deeper
of images are only some of the recent challenges. Due to these and further processing. Optical character recognition
challenges, characters sometimes may not be recognized technology enables a machine to automatically recognize text
correctly by computer system. In this paper we investigate OCR in such documents. In real world example, it is like
in three different ways. We first review the general phases of an
combination of mind and eye of human body. An eye can
OCR system such as pre-processing, segmentation,
normalization, feature extraction, classification and post- detect, view and extract the text from the images but
processing. Then, we give a detailed overview of the challenges absolutely the human’s brain processes that detected or
that might emerge in these OCR stages. Finally, we highlight extracted text read by eye [1]. Of course OCR technology has
development and main applications and uses of OCR. Therefore, not advanced enough to compete with human’s ability. The
this discussion provides a very comprehensive review of the performance and accuracy of OCR is directly dependent upon
state-of-the-art of the field. the quality of input documents. Again, when we think of
human’s ability to recognize text, the performance of brain’s
Keywords— OCR; OCR Challenges; OCR Phases; OCR process directly depends upon the quality of the input read by
Applications eye.
Today, to solve the text recognition problem several
I. INTRODUCTION different types of OCR software exist such as Desktop OCR,
Optical character recognition is an active research area that Server OCR, web OCR and so on. Since the OCR research is
attempts to develop a computer system with the ability to an active and important field in general pattern recognition
extract and process text from images automatically. These problems, due to its fast progress, comprehensive reviews of
days there is a huge demand for storing information to a the field are needed on a regular basis to keep track of the new
computer storage disk from the data available in printed or advancements. One such review was published to discuss the
handwritten documents to later re-utilize this information by challenges with text recognition in scene imagery [2]. This
means of computers. One simple way to store information to paper attempts to elaborate on these kinds of studies by
computer system from these paper documents could be to first providing a comprehensive literature review of optical
scan the documents and then store them as image files. But to character recognition research. We discuss major challenges
____________________________________________________________________________________________________________
374
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
and main phases of optical character recognition such us pre- D. Blurring and Degradation
processing, segmentation, normalization, feature extraction, Since working over a variety of distances are intended to
classification and post processing in detail which needs to be numerous digital cameras, an important factor is the digital
considered during implementing any application related to the camera’s focusing. For the best accuracy of character
OCR, and in the last section of our paper some OCR recognition and character segmentation, character sharpness is
applications are discussed. required. At large apertures and short distances, uneven focus
II. OCR CHALLENGES can be observed when a small point of view changes. For the
most part connected with photography, there are two kinds of
For good quality and high accuracy character recognition, obscure which is: out of focus obscure and movement obscure
OCR techniques expect high quality or high resolution images [3]. At the point for catching a moving item, when the shade
with some basic structural properties such as high rate of the camera is not sufficiently high, the sensor gets
differentiating text and background. The way images are presented to a continually changing scene. Accordingly,
generated is an important and determining factor in the blurring will observed in parts in motion.
accuracy and success of OCR, since this often affects the
quality of images dramatically. Usually OCR with images E. Aspect Ratios
produced by scanners gives high accuracy and good
Text has different aspect ratios. Text may be brief such as
performance. In contrast, images produced by cameras usually
traffic signs, while other text may be much longer, such as
are not as good of an input as scanned images to be used for
video captions. Location, scale and length of text need to be
OCR due to the environmental or camera related factors.
considered with search procedure to detect text, which
Numerous errors might emerge, which are clarified as follow.
introduces high computational complexity.
A. Scene Complexity F. Tilting (Perspective Distortion)
In a regular environment, we can see large numbers of
Document images obtained by scanners is constantly
man-made objects which are included in camera taken images
parallel to the plane of sensor, but this cannot be observed all
such as paintings, buildings, and symbols. These objects have
times for recorded picture obtained by a portable camera, that
comparative structures and appearances to text which makes
may not generally be parallel to the form plane. Accordingly,
text recognition very challenging in the processed image. Text
lines of text that distant from the camera seem littler than
itself is regularly laid out to encourage decipherability. The
those that nearer to the camera which seems greater. This
challenge with scene intricacy is that the surrounding scene
situation causes tilted pictures. Observation of a perspective
makes it hard to segregate text from non-text [2].
distortion is clear if the recognizer is not perspective intolerant,
B. Conditions of Uneven Lighting which causes lower recognition rate and accuracy [4].
Cell phones have an advantage with orientation sensors.
Oftentimes, taking images in natural environments results They can recognize whether the device is tilted and when
in uneven lighting and shadows. This poses a challenge for twisting happens they can forbid clients to take pictures.
OCR as it degrades the desired characteristics of the image Permitting the user to align plane of the form with that of the
and hence causes less accurate detection, segmentation and camera is also provided by this feature. Orientation sensors
recognition results [2]. therefore may assure that produced pictures satisfy a certain
This condition with uneven lighting is what distinguishes a degree of evenness.
scanned image form one that is produced with a camera. The
lack of such disparities in lighting and shadows makes G. Fonts
scanned images preferred over camera images for their better
Italic style and script fonts of characters might overlap each
characteristics and quality. Although using an on-camera flash
other, making it difficult to perform some of the main OCR
may eliminate such problems with uneven lighting, it
processes such as segmentation. Characters of various fonts
introduces new challenges.
have large within-class variations and form many pattern sub-
C. Skewness (Rotation) spaces, making it difficult to perform accurate recognition
when the character class number is large.
For optical character recognition systems, the point of
view for the input image that taken from camera of hand-held H. Multilingual Environments
device or other gadgets that used for taken image is not fixed
Although most of the Latin languages have tens of
like a scanner input, which skewing of text lines from their
characters, languages such as Chinese, Japanese and Korean,
unique orientation might be observed. Great degree poor
have thousands of character classes. Arabic has connected
results will be observed when such a skewed image is fed to
characters, which change shape according to context. In Hindi
the OCR classifier. Many techniques available for the purpose
syllables represent by combining alphabetic letters into
of deskew the image documents, such as Projection Profile,
thousands of shapes [5]. In multilingual environments, OCR
RAST algorithm, Hough transform, methods of Fourier
in scanned documents remains as a main research problem [5],
transformation, etc.
sice text recognition in complex imagery is more difficult.
____________________________________________________________________________________________________________
375
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
376
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
that utilizing the traditional vertical and horizontal projection OCR systems broadly utilize the methodologies of pattern
profile method makes text easily segmented into lines and recognition, which assigns each example to a predefined class.
words. They reported experimental results with 98% accuracy Classification is the procedure of distributing inputs with
of line and word segmentation. respect to detected information to their comparing class in
order to create groups with homogeneous qualities, while
C. Normalization Phase segregating different inputs into different classes.
As a result of segmentation process isolated characters Classification is conveyed out on the premise of put away
which are ready to move through feature extraction phase are features in the feature space, for example, structural features,
obtained, hence the isolated characters are minimized to a global features and so forth. It can be said that classification
particular size depending on the algorithms used. The isolates the feature space into several classes taking into
segmentation process is crucial as it converts the image in the account the decision rule. Choosing classifier depends on
form of m*n matrix. These matrices are then commonly several agents, such as, number of free parameters, available
normalized by minimizing the size and eliminating the training set and so forth. Various procedures for OCR are
unnecessary information from the image without missing any explored by the scientists.
influential information [12]. Techniques of OCR classification can be categorized [19]
as Statistical Techniques, Neural Networks, Template
D. Feature Extraction Phase Matching, Support Vector Machine (SVM) algorithms, and
Feature extraction is the operation of extracting the Combination of classifier.
pertinent features from objects or alphabets to build feature 1) Template matching: This is the least complex method for
vectors. These feature vectors are then utilized by classifiers character recognition, in view of matching the stored models
to identify the input unit with objective output unit. It against the word or character to be perceived. By gathering of
becomes effortless for the classifier to classify between shapes, pixels, curvature and so forth, the operation of
dissimilar classes by glancing at these features as it becomes matching decides the level of similitude between two vectors.
fairly easy to determine [13]. A gray-level or binary input character is contrasted with a
Several techniques are proposed for extracting features standard arrangement of stored models. The recognition rate
from the segmented characters in literature. U. Pal et al [14] of this strategy is extremely delicate to noise and input
have proposed directional chain code features and zoning and disfigurement.
for handwritten numeral recognition considered a feature
vector of length 100 and have presented a high level of 2) Statistical Techniques: Hypothesis of Statistical decision is
recognition accuracy. But, the feature extraction process is treating with statistical decision capacities and an arrangement
time consuming and complex [13]. of optimality criteria, which for a given model of a specific
Dinesh et al [15] have proposed end points as the potential class can amplify the likelihood of the observed pattern. The
features for recognition and used horizontal/vertical strokes main statistical methods that are performed in the area of OCR
and for handwritten Kannada numerals obtained a recognition [19] are Nearest Neighbor (NN), Likelihood or Bayes
accuracy of 90.50%. But, this method uses the thinning classifier, Clustering Analysis, Hidden Markov Modelling
process which results in some loss of features [13]. (HMM), Fuzzy Set Reasoning, and Quadratic classifier.
E. Srinivasan et al [13] for handwritten alphabets 3) Neural Networks: Character classification issue is
recognition system have proposed diagonal based feature identified with heuristic rationale as people can perceive
extraction utilizing neural network. characters and records by their learning and experience. Thus
Sharma, Om Prakash et al [16] for handwritten alphabets neural networks which are pretty much heuristic in nature are
recognition proposed an improved zone based hybrid feature greatly appropriate for this type of issue [19]. A neural
extraction model utilizing Euler number. network is an ascertaining architecture that includes
Following to Suen [17], there are two major classes of enormously parallel interconnection of flexible node
features: statistical features and structural features. In a processors. Output from one node is reinforcing to the next
character matrix statistical features are obtained from one in the network and an official choice relies on the
statistical distribution of every point such as zoning, moments, complicated collaboration of all nodes. As a result of its
crossings, fourier transforms and projection histograms [18]. similar character, it can apply calculations at a rate higher
Statistical features are also notable as global features as they contrasted with the traditional strategies. Feed-forward neural
are usually averaged and extracted in sub-images such as networks and feedback neural networks can be thought as
meshes. Initially, statistical features are supplied to recognize categorization of neural network architectures. Table II,
machine printed characters. On the other hand, structural or compares and discusses some recent proposed OCR
topological features are concern to the geometry of the applications based on Neural Network.
character set to be contemplated. Some of these features are
convexities and concavities in the characters, number of holes TABLE III
ACCURACY COMPARISON AMONG RECENT PROPOSED OCR SYSTEMS BASED
in the characters, number of end points etc. ON NEURAL NETWORK
____________________________________________________________________________________________________________
377
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
378
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
G. Automatic Number Plate Recognition [9] Basu, Subhadip, et al. "Text line extraction from multi-skewed
handwritten documents." Pattern Recognition 40.6 (2007): 1825-1839.
Automatic number plate recognition [29] is utilized as a [10] Khandelwal, Abhishek, et al. "Text line segmentation for
mass observation method making utilization of optical unconstrained handwritten document images using neighborhood
connected component analysis." International Conference on Pattern
character recognition on pictures to recognize vehicle Recognition and Machine Intelligence. Springer Berlin Heidelberg,
registration plates. ANPR has additionally been made to store 2009.
the pictures caught by the cameras including the numbers [11] Shinde, Archana A., and D. G. Chougule. "Text Pre-processing and
caught from license plate. ANPR innovation own to plate Text Segmentation for OCR." International Journal of Computer
Science Engineering and Technology (2012): 810-812.
variety from place to place as it is an area particular [12] Trier, Øivind Due, Anil K. Jain, and Torfinn Taxt. "Feature extraction
innovation. They are utilized by different police forces and as methods for character recognition-a survey." Pattern recognition 29.4
a technique for electronic toll accumulation on pay-per-use (1996): 641-662.
streets. [13] Pradeep, J., E. Srinivasan, and S. Himavathi. "Diagonal based feature
extraction for handwritten character recognition system using neural
network." Electronics Computer Technology (ICECT), 2011 3rd
V. CONCLUSIONS
International Conference on. Vol. 4. IEEE, 2011.
Numerous algorithms, methods and techniques have been [14] Bishnu, Arijit, et al. "A pipeline architecture for computing the Euler
proposed to optical character recognition in scene imagery, number of a binary image." Journal of Systems Architecture 51.8
(2005): 470-487.
yet there are not enough literature surveys in this field. In this [15] Dinesh Acharya, U., and N. V. Subbareddy. "Krishnamoorthy: Isolated
paper, we have proposed an organization of these methods, Kannada Numeral Recognition Using Structural Features and K-
algorithms and techniques. It is hoped that this comprehensive Means Cluster." Proc. of IISN (2007): 125-129.
survey will provide insight into the concepts involved, and [16] Sharma, Om Prakash, M. K. Ghose, and Krishna Bikram Shah. "An
improved zone based hybrid feature extraction model for handwritten
perhaps provoke further advances in the area. Firstly, we alphabets recognition using euler number." International Journal of
discussed major challenges of OCR, then we discussed in Soft Computing and Engineering 2.2 (2012): 504-58.
great detail the main important phases, architecture, proposed [17] Suen, Ching Y. "Character recognition by computer and applications."
algorithms and techniques of OCR, we highlight that for Handbook of pattern recognition and image processing (1986): 569-
586.
designing any application related to the OCR, one must pay [18] Rehman, Amjad, and Tanzila Saba. "Neural networks for document
great attention to each phase to obtain high accurate character image preprocessing: state of the art." Artificial Intelligence Review
recognition rate, but still we cannot propose comprehensive 42.2 (2014): 253-273.
algorithms for each phase because it depends upon datasets, [19] Dongre, Vikas J., and Vijay H. Mankar. "A review of research on
Devnagari character recognition." arXiv preprint arXiv:1101.2491
application specifics, and parameter specifics. Finally major (2011).
applications related to the OCR are discussed. [20] Shah, Parul, et al. "OCR-based chassis-number recognition using
Although the state-of-the art OCR enables text recognition artificial neural networks." Vehicular Electronics and Safety (ICVES),
with high accuracy, we think that there could be many more 2009 IEEE International Conference on. IEEE, 2009.
[21] Zhai, Xiaojun, Faycal Bensaali, and Reza Sotudeh. "OCR-based neural
practical applications of OCR. As a future work we are network for ANPR." 2012 IEEE International Conference on Imaging
planning to use OCR for such practical applications for daily Systems and Techniques Proceedings. IEEE, 2012.
personal use. We are planning to incorporate mobile devices [22] Shamsher, Inam, et al. "OCR for printed urdu script using feed forward
with OCR in one OCR system. An automated book reader or a neural network." Proceedings of World Academy of Science,
Engineering and Technology. Vol. 23. 2007.
receipt tracker constitutes some of our future OCR based [23] Yetirajam, Manas, Manas Ranjan Nayak, and Subhagata
applications. Chattopadhyay. "Recognition and classification of broken characters
using feed forward neural network to enhance an OCR solution."
REFERENCES International Journal of Advanced Research in Computer Engineering
[1] Patel, Chirag, Atul Patel, and Dharmendra Patel. "Optical character & Technology (IJARCET) Volume 1 (2012).
recognition by open source OCR tool tesseract: A case study." [24] Verma, Rohit, and Dr Jahid Ali. "A-Survey of Feature Extraction and
International Journal of Computer Applications 55.10 (2012). Classification Techniques in OCR Systems." International Journal of
[2] Ye, Qixiang, and David Doermann. "Text detection and recognition in Computer Applications & Information Technology 1.3 (2012).
imagery: A survey." IEEE transactions on pattern analysis and machine [25] Jain, Anil K., Robert P. W. Duin, and Jianchang Mao. "Statistical
intelligence 37.7 (2015): 1480-1500. pattern recognition: A review." IEEE Transactions on pattern analysis
[3] Jain, Atishay, et al. "Fundamental challenges to mobile based ocr." vol and machine intelligence 22.1 (2000): 4-37.
2 (2013): 86-101. [26] Matei, Oliviu, Petrica C. Pop, and H. Vălean. "Optical character
[4] Moravec, Kimberly. "A Grayscale Reader for Camera Images of Xerox recognition in real environments using neural networks and k-nearest
DataGlyphs." BMVC. 2002. neighbor." Applied intelligence 39.4 (2013): 739-748.
[5] Smith, Ray, Daria Antonova, and Dar-Shyang Lee. "Adapting the [27] Ganis, M. D., Charles L. Wilson, and James L. Blue. "Neural network-
Tesseract open source OCR engine for multilingual OCR." based systems for handprint OCR applications." IEEE Transactions on
Proceedings of the International Workshop on Multilingual OCR. Image Processing 7.8 (1998): 1097-1112.
ACM, 2009. [28] Gossweiler, Rich, Maryam Kamvar, and Shumeet Baluja. "What's up
[6] Ulges, Adrian, Christoph H. Lampert, and Thomas M. Breuel. CAPTCHA?: a CAPTCHA based on image orientation." Proceedings
"Document image dewarping using robust estimation of curled text of the 18th international conference on World Wide Web. ACM, 2009.
lines." Eighth International Conference on Document Analysis and [29] Gao, Jianjun, et al. "Automatic vehicle license plate recognition with
Recognition (ICDAR'05). IEEE, 2005. color component texture detection and template matching." SPIE
[7] Kaur, Sukhvir, P. S. Mann, and Shivani Khurana. "Page Segmentation Defense, Security, and Sensing. International Society for Optics and
in OCR System-A Review." (IJCSIT) International Journal of Computer Photonics, 2013.
Science and Information Technologies, Vol. 4 (3) , 2013, 420-422.
[8] Saha, Satadal, et al. "A Hough transform based technique for text
segmentation." arXiv preprint arXiv: 1002.4048 (2010).
____________________________________________________________________________________________________________
379
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
+
Süleyman Demirel University, Faculty of Technology, Isparta, TURKEY
tuncayaydogan@sdu.edu.tr
*
Mehmet Akif Ersoy University, Vocational School of Technical Sciences, Burdur, TURKEY
imuammer@mehmetakif.edu.tr
Abstract— Today, every kind of text, audio and visual data, the data which can be regarded as Big Data. It was observed
which are thought to be transformed into pieces of that Big Data tools and techniques that are used in processing
information, are stored for long periods of time for processing. this data provided better performance results in comparison
The concept of Bid Data is not only associated with the data with traditional tools and techniques [3].
stored, but also with the system involving hardware and
software that collects, processes, stores, and analyzes the data. One of these tools is Solr which is an open source, Java-
As the data grows bigger, their physical storage options must based tools that uses Apache Lucene search library and it can
be provided in a distributed architecture. be used on servers such as Tomcat, Jetty, etc. [4]. According
Solr and Elasticsearch are among the most preferred tools to July, 2016 figures of DB-Engines Ranking of Search
which makes this storage process easier. As a part of Apache Engines, Solr is the second most popular search engine
Lucene project, Solr is a software which was started to be software [5]. With its 5th version, Solr has become a separate
developed in 2004 with the searching features of full text, software instead of being a Java package [6].
multiple search, dynamic clustering, database-integrated, open
source and elasticity. Similarly, Elasticsearch is a new open- Elasticsearch is one of the other tools that is used in
source tool for real-time, full-text and distributed search, analyzing and process Big Data. It is an open-source, Lucene-
which was launched in 2010 using the Lucene library. based content search and analysis tool that was developed in
Although Solr and Elasticsearch have similar features, there Java. This tool, which began to be developed in 2010, is used
are many parameters that differentiates one from the other by many domestic/foreign organizations. It is also a tool that
such as intended use, type of use, and query and indexing can be set up in Cloud servers purchased through Google For
performances. This study researches and analyzes the Works, and it is in a sense supported by Google [7].
differences between Solr and Elasticsearch with regards to According to July, 2016 figures of DB-Engines Ranking of
their query and indexing speeds, ease and difficulties of use, Search Engines, Elasticsearch is the most popular search
configuration forms, and architectures in light of the literature, engine software [5].
and the results are discussed regarding these tools’
performances. Because of the fact that both of Solr and Elasticsearch
tools are based on Apache Lucene library, and they have
Keywords— big data, elasticsearch, solr similarities between the intended use and methods, supported
protocols, and coding languages, today they are compared and
I. INTRODUCTION contrasted by users in different categories. Users try to decide
Huge amount of data stack which cannot be stored nor which one is the best tool for them based on their own needs
processed by traditional methods is called Big Data [1]. Based and system requirements.
on the fact that most of the smart devices communicate on the
Internet and the technological devices and computers will This study, therefore, collects comparative data conducted
communicate through cloud computing, it is easy to predict so far between Solr and Elasticsearch. These collections
that much bigger data is to be produced in the future [2]. It is involve technical configurations, performance tests, etc. This
impossible to store, process, and analysis this big data by study is thought to contribute to software developers in using
conventional tools. Therefore, many tools were developed for Solr and Elasticsearch.
____________________________________________________________________________________________________________
380
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
A sample schema file involves the following data: - Lists all data
- wt=json Listed in JSON format
<field name="id" type="string" indexed="true"
stored="true" required="true" multiValued="false" /> Request:
<field name="sku" type="text_en_splitting_tight"
indexed="true" stored="true" omitNorms="true"/> http://localhost:8983/solr/select?q=id:VS1GB400C3&wt=json
<field name="name" type="text_general" indexed="true" &fl=id,name
stored="true"/>
<field name="manu" type="text_general" indexed="true" - Lists data with VS1GB400C3 id parameter value
stored="true" omitNorms="true"/>. - wt=json Listed in JSON format
- fl=id,name Lists only id and name information in
results page.
____________________________________________________________________________________________________________
381
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
382
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. TECHNICAL COMPARISON BETWEEN SOLR AND C. Data and Distributed Architecture
ELASTICSEARCH
A. Overview Data storage, data processing, and distributed architecture
Table IV provides a general overview for the features of features of Lucene-based Solr and Elasticsearch are found in
Solr and Elasticsearch tools. Table VI.
TABLE IV TABLE VI
OVERVIEW DATA AND DISTRIBUTED ARCHITECTURE
Solr Elasticsearch Solr Elasticsearch
Inital Release 2004 2010 Distributed Architecture Supports Supports
Current Release 6.1.0 June 2016 2.3.3, May 2016 Sharding Yes Yes
Licence Open Source Open Source Replication Yes Yes
Database as a Shard Splitting Yes No
No No Changing Number of Yes No
Service
SQL No No Shards
Implementation Relocating Yes Yes
language Java Java Routing Yes No
Schemaless Yes (After 4.4) Yes
All OS with a Map Reduce No No
Java VM and a Automatic Shard No Yes
Server operating All OS with a Java Rebalancing
servlet container
systems VM Distributed Group By Yes No
(Tomcat, Jetty,
etc)
Based Apache Lucene Apache Lucene
Plugin Needed D. Searching and Indexing
(Bigdesk, Kopf,
Embedded
Admin Interface ElasticHQ, Solr and Elasticsearch have different indexing and searching
Available
Paramedic, Marvel, features based on their basic structures and the plugins
etc) developed by users (see Table VII).
____________________________________________________________________________________________________________
383
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. PERFORMANCE COMPARISON BETWEEN SOLR regarding Solr and Elasticsearch nodes. As seen in Figure 1, it
AND ELASTICSEARCH was found out that most of the Elasticsearch queries were
completed between 0.10 sec and 0.22 sec as a result of 1000
One of the issues that Solr and Elasticsearch tools are search inquiries sent to nodes. The same test provided the
compared most is their performance figures. Even though results of 0.12 sec and 0.54 sec for Solr tool. [13].
these tools are quite similar to one another in terms of being
Lucene-based, different figures can be obtained related to
their performance. In studies on performance comparisons, it
was found out that Solr and Elasticsearch do not greatly have
superiority over each other [12][13]. However, sometimes
Solr had better performance over Elasticsearch based on
intended use, place of use, and changes in some parameters,
and vice versa. In this part of this study, some of the
performance results conducted so far on Solr and
Elasticsearch are listed.
A. Flax.co.uk Tests
Flax.co.uk design, build and support open source powered
search applications based in Cambridge U.K. Flax has
completed many performance tests on Solr and Elasticsearch.
Cluster specifications used in a test are as follows: Fig. 1 Search Test [15]
Two machines, each with 96GB RAM C. Search Test with Indexing Load
Two instances of SolrCloud or Elasticsearch on each
Each instance has 24GB JVM heap In the second test, servers were requested to operate search
Four shards inquiries while engaged in indexing process. These tests were
No replicas also conducted with 20 million entries on servers with similar
features. As seen in Figure 2, Elasticsearch had results varying
In the first stage of this test, a data consisting of 40 million between 0.14 sec and 0.34 sec, Solr completed search inquiry
pieces with 5 to 20 words was created. The results of Solr and between 0.24 sec and 0.68 sec. Elasticsearch was slightly
Elasticsearch indexes are listed below: better than Solr in search tests with indexing load.
Based on these results, it can be concluded that Elasticsearch Fig. 2 Search Test with Indexing Load [15]
performs better with short documents, while Solr had better
performance with long documents.
____________________________________________________________________________________________________________
384
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
D. QPS Test
TABLE VIII
Queries per second (QPS) is a common measure of the
DB-ENGINES JULY, 2016 FIGURES
amount of search traffic an information retrieval system, such
as a search engine or a database, receives during one second Jul Jul Jul Jun Jul
DBMS
[14]. 2016 2015 2016 2016 2015
88.6
64.6
46.6
____________________________________________________________________________________________________________
385
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
386
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The User Knowledge Modelling dataset in the UCI measures taken by the university, and the ‘high-risk’ students,
machine learning repository was used in this study. The students who have a high probability of failing (or dropping out). The
were classified into 4 class (very low, low, middle, and high) due to fact that the most significant variables correlated to academic
the 5 performance data in the dataset. 258 data of 403 data in the success have been provided among all the questions asked to
dataset were used for training and 145 of them were used for tests.
533 first-year university students during the month of
The Weka (Waikato Environment for Knowledge Analysis)
software was used for classification. In classification Multilayer November of academic year 2003-04. Finally, it presents the
Perceptron (MLP), k Nearest Neighbour (kNN), J48, NativeBayes, results of the application of discriminant analysis, neural
BayesNet, KStar, RBFNetwork and RBFClassifier machine networks, random forests and decision trees aimed at predicting
learning algorithms were used and success rates and error rates those students' academic success [3]. Vera et al. (2012) have
were calculated. In this study 8 different data mining algorithm proposed a genetic programming algorithm and different data
were used and the best classification success rate was obtained by mining approaches for solving these problems using real data
MLP. With Multilayer perceptron neural network model the about 670 high school students from Zacatecas, Mexico. They
classification success rates was calculated when there are different also propose to use a genetic programming model versus
number of neurons in the hidden layer of MLP. The best
different white box techniques in order to obtain both more
classification success rate was achieved as 97.2414% when there
was 8 neurons in the hidden layer. MAE and RMSE values were comprehensible and accuracy classification rules. They have
obtained for this classification success rate as 0.0242 and 0.1094 shown that some approaches such as selecting the best
respectively. attributes, cost-sensitive classification, and data balancing can
also be very useful for improving accuracy [4]. Sen et al. (2012)
Keywords— Machine learning, Weka, MLP, kNN, J48 developed models to predict secondary education placement
test results, and using sensitivity analysis on those prediction
I. INTRODUCTION models using a large and feature rich dataset from Secondary
Data mining is an interdisciplinary subfield of computer Education Transition System in Turkey. The Support Vector
science. It is the practice of automatically searching large stores Machines, Artificial Neural Networks and Logistic Regression
of data to discover patterns and trends that go beyond simple models have been used for prediction and results were
analysis. Data mining uses sophisticated mathematical presented [5].
algorithms to segment the data and evaluate the probability of In this study, the User Knowledge Modelling dataset
future events. Data mining is also known as Knowledge obtained from UCI Machine Learning Repository have been
Discovery in Data (KDD) [1]. used for classification. In this dataset, there are 5 attributes that
Educational data mining (EDM) is defined as the area of used for determining the students' educational status in 4 class
scientific inquiry centered around the development of methods as very low, low, middle, high. For classification 8 machine
for making discoveries within the unique kinds of data that learning algorithm have been used. The obtained success rates
come from educational settings, and using those methods to and error values by those algorithms, have been compared.
better understand students and the settings which they learn in
II. MATERIAL AND METHODS
[2]. Many studies in the literature have been proposed to
explore the relationship between successes of the students and Software-WEKA
his/her culture, habits, life style, family structure etc.
Weka (Waikato Environment for Knowledge Analysis)
Superby et al. (2006) have search a relation between
written in Java, developed at the University of Waikato, New
Academic failures and increase the number of debates among
Zealand [6]. Weka supports several standard data mining tasks,
first-year university students. They aims to classify, as early in
more specifically, data pre-processing, clustering,
the academic year as possible, students into three groups: the
classification, regression, visualization, and feature selection.
‘low-risk’ students, who have a high probability of succeeding,
All techniques of Weka's software are predicated on the
the ‘medium-risk’ students, who may succeed thanks to the
____________________________________________________________________________________________________________
387
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
assumption that the data is available as a single flat file or Given symptoms, the network can be used to compute the
relation, where each data point is described by a fixed number probabilities of the presence of various diseases [6].
of attributes (normally, numeric or nominal attributes, but some
other attribute types are also supported) [7]. Naive Bayes
Naive Bayes classifiers are highly scalable, requiring a
K-Nearest Neighbour Algorithm number of parameters linear in the number of variables
kNN is the supervised learning algorithm, solving the (features/predictors) in a learning problem. Maximum-
classification problems. Classification is to analyse of features likelihood training can be done by evaluating a closed-form
of image, and to belong it to the class which named in advanced. expression which takes linear time, rather than by expensive
The important thing is determination the features of each iterative approximation as used for many other types of
category in advance [8]. According to kNN algorithm that was classifiers [6].
used in classification, characteristics which extracted from
classification process, viewing the entire distance among new J48
individual which should be classified and earlier individuals J48 algorithm of Weka software is a popular machine
and then the nearest k category is used. As a result of that learning algorithm based upon J.R. Quilan C4.5 algorithm. All
process test data belongs to the nearest k neighbour category data to be examined will be of the categorical type and therefore
that has more members in certain class. The most important continuous data will not be examined at this stage. The
optimization problems in kNN method are identification of algorithm will however leave room for adaption to include this
neighbour’s number and method of distance calculation capability [6, 7].
algorithm. At the study identification of the optimum k number
is done by experiments and Euclidean Distance Calculations KStar
method was used as distance calculations method. K-star or K* is an instance-based classifier. The class of a
Euclidean calculation method [9]: test instance is based on the training instances similar to it, as
determined by some similarity function. It differs from other
instance-based learners in that it uses an entropy-based distance
function. [10].
xi and xj are two different points, and need distance calculation III. RESULTS AND DISCUSSION
process in between. 403 sample data at User Knowledge Modelling dataset were
processed by using Weka program. The classification success
Multilayer Perceptron rates at different neurons numbers in the hidden layers and
A multilayer perceptron (MLP) is a feed forward artificial MAE, RMSE error rates were determined. The diagram
neural network model that maps sets of input data onto a set of demonstrating the changes in MAE and RMSE error values
appropriate outputs. A MLP consists of multiple layers of nodes based on the number of neighbours in the hidden layer is
in a directed graph, with each layer fully connected to the next demonstrated in Figure 1.
one. Except for the input nodes, each node is a neuron (or TABLE I
processing element) with a nonlinear activation function. MLP THE SUCCESS RATE AND ERROR VALUES OBTAINED BY USING MLP
utilizes a supervised learning technique called back CLASSIFIER
propagation for training the network. MLP is a modification of
the standard linear perceptron and can distinguish data that are
not linearly separable [6].
RBF Network
It is an artificial neural network that uses radial basis
functions as activation functions. The output of the network is
a linear combination of radial basis functions of the inputs and
neuron parameters. Radial basis function networks have many
uses, including function approximation, time series prediction,
classification, and system control [6].
BayesNet
It is probabilistic graphical model (a type of statistical
model) that represents a set of random variables and their Fig. 1 Variation of error rate based on the number of neurons in hidden layer
conditional dependencies via a directed acyclic graph (DAG). The classification success of the state of eye same was
For example, a Bayesian network could represent the obtained by using kNN algorithm with the same dataset.
probabilistic relationships between diseases and symptoms. Classification success for different k neighbour values with
kNN algorithm in the state of the eye to be open or closed were
____________________________________________________________________________________________________________
388
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
obtained. In addition, the root mean square error (RMSE) and IV. CONCLUSION
the mean absolute error (MAE) values were found. The graph In this study, students have been classified about their
showing the variation of MAE and RMSE error values success in the school as very low, low, middle and high by
depended on number of neighbourhood in the classification using their daily habits. By this intention, popular data mining
made by kNN algorithm is shown in Figure 2. algorithms like kNN, MLP, J48, NaiveBayes, RBFNetwork,
RBFClassifier, BayesNet and Kstar, have been used and
compared with each other. At obtained classification success
rates, success rates made with kNN algorithm is seen to be
higher. At classification success rates obtained with kNN
algorithm, highest classification success was obtained for 3
neighbour values and this value is 87.5862%. At this neighbour
value, MAE error value is 0.1037 and RMSE error value is
0.2324. It is seen that classification success rates obtained by
using Multilayer perceptron model is too low for kNN classifier.
While the number of neurons in the hidden layer is 8, highest
classification success was obtained and this value is 97.2414%.
At this neurons number in the hidden layer, MAE error value is
0.0242 and RMSE error value is 0.1094.
The success rates obtained using J48, NaiveBayes,
Fig. 2 Variation of error rate based on the number of neighborhood RBFNetwork, RBFClassifier, BayesNet and Kstar
classification algorithms were found as 91.0345%, 84.8276%,
Then the same data was processed using J48, NaiveBayes,
93.7931%, 71.0345%, 86.8966% and 81.37935.respectively.
RBFNetwork, RBFClassifier, BayesNet, Kstar machine
learning algorithms and classification success rates and MAE
and RMSE error values of different tree types in the forest were REFERENCES
obtained. The success and error rates obtained using 8 different
[1] Clifton, Christopher (2010). "Encyclopædia Britannica: Definition of
classification algorithms (Multilayer Perceptron, kNN, J48, Data Mining". Retrieved 2010-12-09.
Naïve Bayes, Bayes Net, KStar) can be seen in Table I. The [2] B., Peterson, P., Baker, E. (Eds.) International Encyclopedia of
diagram demonstrating the error values obtained based on Education (3rd edition). Oxford, UK: Elsevier.
[3] Superby J. F., Vandamme J. P., Meskens N. Determination of factors
different machine learning algorithms can be seen in Figure 3.
influencing the achievement of the first-year university students using
data mining methods. In International conference on intelligent tutoring
TABLE II systems, Educational Data Mining Workshop, Taiwan, 2006:1 – 8.
SUCCESS RATE OBTAINED BY USING VARIOUS MACHINE LEARNING [4] Márquez-Vera, C., Cano, A., Romero, C., Ventura, S. Predicting student
ALGORITHMS failure at school using genetic programming and different data mining
approaches with high dimensional and imbalanced data, Applied
Intelligence, April 2013, Volume 38, Issue 3, pp 315-330.
Machine Learning Classification [5] Sen B., Ucar E., Delen D. Predicting and analyzing secondary education
Algorithms Success MAE RMSE placement-test scores: A data mining approach, Expert Systems with
(%) Applications, Vol. 39, No. 10, pp. 9468-9476, 2012.
MLP 97.2414 0.0242 0.1094 [6] WEKA, http://www.cs.waikato.ac.nz/~ml/weka/ Last access:
10.04.2015.
kNN 87.5862 0.1037 0.2324 [7] Rohit Arora and Suman, Comparative Analysis of Classification
J48 91.0345 0.0481 0.2092 Algorithms on Different Datasets using WEKA, International Journal of
NaiveBayes 84.8276 0.1173 0.2582 Computer Applications (0975 – 8887) Volume 54– No.13, September
RBFNetwork 93.7931 0.0411 0.1567 2012.
RBFClassifier 71.0345 0.2143 0.2988 [8] J. Wang, P. Neskovic, and L. N. Cooper, “Improving nearest neighbor
rule with a simple adaptive distance measure”, Pattern Recognition
BayesNet 86.8966 0.1312 0.2485
Letters, 28(2):207-213, 2007
Kstar 81.3793 0.1122 0.2695 [9] Y. Zhou, Y. Li, and S. Xia, “An improved KNN text classification
algorithm based on clustering”, Journal of computers, 4(3):230-237,
2009.
[10] John G. Cleary, Leonard E. Trigg: “K*: An Instance based Learner
Using an Entropic Distance Measure”, 12th International Conference on
Machine Learning, 108-114, 1995.
____________________________________________________________________________________________________________
389
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The Default of Credit Card Clients dataset in the UCI habits of two different culture may be investigated for
machine learning repository was used in this study. The credit similarity [3].
card customers were classified if they would do payment or not Data mining methods can be divided into three groups due
(yes=1 no=0) for next month by using 23 information about them. to their function.
Totally 30000 data in the dataset’s 66% was used for training and
1. Classification and Regression
rest of them as 33% was used for tests. The Weka (Waikato
Environment for Knowledge Analysis) software was used for 2. Clustering
estimation. In estimation Multilayer Perceptron (MLP) and k 3. Association Rules
Nearest Neighbour (kNN) machine learning algorithms was used Data mining methods are used to classifying the data set. In
and success rates and error rates were calculated. With kNN classification, training examples are used to learn a model that
estimation success rates for various number of neighbourhood can classify the data samples into known classes. The
value was calculated one by one. The highest success rate was classification process involves following these steps: creating a
achieved as 80.6569% when the number of neighbour is 10. With training data set, identifying class attributes and classes,
MLP neural network model the estimation success rates was identifying useful attributes for classification, relevance
calculated when there are different number of neurons in the
analysis, learning a model using training examples in the
hidden layer of MLP. The best estimation success rate was
achieved as 81.049% when there was only one neuron in the training set and using the model to classify the unknown data
hidden layer. MAE and RMSE values were obtained for this [4].
estimation success rate as 0.3237 and 0.388 respectively. In this study, an estimation about whether the payment for
next month is going to be done or not by the credit card clients
Keywords— Data mining, Weka, MLP, kNN in the default credit card clients data set with 23 attributes
obtained from the UCI Machine Learning Repository, have
I. INTRODUCTION been done. For estimation kNN and MLP algorithms have been
Credit card is a payment card issued to users (cardholders) used. The success rates and error values have been presented
as a method of payment. It allows the cardholder to pay for and compared with each other.
goods and services based on the holder's promise to pay for
II. MATERIAL AND METHODS
them [1]. The issuer of the card (usually a bank) creates a
revolving account and grants a line of credit to the cardholder, Dataset
from which the cardholder can borrow money for payment to a
merchant or as a cash advance. For the banks the most In this study the default of credit card clients data set
important thing during credit card marketing is the payment obtained from UCI Machine Learning Repository have been
capability of customers. In this study a payment status used. This data set have been obtained from Credit cart
estimation have been proposed for credit card customers. For customers’ default payments in Taiwan. In this data set there
this purpose data mining algorithms have been used. are 23 attributes and a binary type class. These attributes and
Data mining is a computational process that reveals patterns descriptions are as follow [5].
in data sets by using such methods like artificial intelligence, X1: Amount of the given credit (NT dollar): it includes both
machine learning, statistics etc [2]. The methods used in data the individual consumer credit and his/her family
mining are investigated in two groups as predictive and (supplementary) credit.
descriptive. In predictive methods, a model is created by using X2: Gender (1 = male; 2 = female).
a dataset whose results are known. For Example in a bank, the X3: Education (1 = graduate school; 2 = university; 3 = high
properties of customers who pay their credits back can be school; 4 = others).
revealed and a model can be created by using previous data sets X4: Marital status (1 = married; 2 = single; 3 = others).
about funding of them. Afterward this model can be used on X5: Age (year).
new customers for determining the possibility of pay their X6 - X11: History of past payment from April to September,
credits back. In descriptive methods, a relationship can be 2005. The measurement scale for the repayment status is: -1 =
searched between two data sets. For example, the shopping pay duly or number of the delayed month.
____________________________________________________________________________________________________________
390
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
X12-X17: Amount of bill statement (NT dollar). III. RESULTS AND DISCUSSION
X18-X23: Amount of previous payment (NT dollar).
X24: Class (0 = No; 1 = Yes) In the study, WEKA software was used in order to estimate
the payment for next month is going to be done or not by the
Software-WEKA credit card. Using the kNN algorithm, the estimation success
rates of payment were obtained for different k-neighbour values.
Developed by Waikato University in New Zealand, WEKA Also, root mean square error (RMSE) and mean absolute error
is an open-source data mining software with a functional (MAE) values were obtained. The estimation success rates
graphical interface which incorporates machine learning obtained with kNN algorithm, and MAE and RMSE values can
algorithms [6]. WEKA includes various data pre-processing, be seen in Table 1. The diagram demonstrating the changes in
classification, regression, clustering, association rules, and MAE and RMSE error values based on the number of
visualization tools. The algorithms can be applied on the data neighbours in the estimation performed with the kNN algorithm
cluster either directly or by calling via Java code [7][8]. They is shown in Figure 1.
are also suitable for developing new machine learning
algorithms.
TABLE I
THE SUCCESS RATE AND ERROR VALUES OBTAINED BY USING KNN
Machine learning algorithms CLASSIFIER
____________________________________________________________________________________________________________
391
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
taking the learning rate value as 0.3, momentum value as 0.2 IV. CONCLUSION
and iteration number as 500. The estimation success rates, and In this study, credit card clients’ behaviours about payment
MAE and RMSE values obtained using the MLP model can be have been estimated. For this purpose machine learning
seen in Table 2. The diagram demonstrating the changes in algorithms like kNN and MLP have been used. The estimation
MAE and RMSE error values based on the number of success rates and error values of kNN and MLP were calculated.
neighbours in the hidden layer is demonstrated in Figure 2. It was observed that the success rate was higher for the
estimation performed by using the MLP algorithm. The highest
TABLE III estimation success rate was achieved when the number of
SUCCESS RATE OBTAINED BY USING MLP CLASSIFIER ERROR VALUES neurons in the hidden layer was 1 and the success rate was
81.049%. The MAE error value was 0.3237 and the RMSE
The number error value was 0.388 when there is only one neuron in the
Estimation
of neurons in
Success MAE RMSE hidden layer. For the estimation success rates obtained using K-
the hidden
layer
(%) Nearest Neighbour Algorithm, the highest estimation success
rate was achieved for 10 neighbourhood values, and it was
1 81.049 0.3237 0.388
80.6569%. For this neighbourhood value, the MAE error value
2 79.1961 0.2982 0.3837 was 0.2743 and the RMSE error value was 0.3897.
3 79.8137 0.2978 0.3838
REFERENCES
4 80.3922 0.2878 0.38 [1] O'Sullivan, A., Sheffrin, S. M., (2003). Economics: Principles in action.
5 80.3922 0.2926 0.3811 Upper Saddle River, New Jersey 07458: Pearson Prentice Hall. p. 261.
ISBN 0-13-063085-3.
6 80.3137 0.3005 0.3807 [2] Chen, M. S., Han, J., & Yu, P. S., (1996). Data mining: an overview
from a database perspective. Knowledge and data Engineering, IEEE
7 80.1765 0.2934 0.3833 Transactions on, 8(6), 866-883.
8 80.2451 0.3021 0.3838 [3] Özekes, S., (2003). Data mining methods and application areas (Veri
madenciliği modelleri ve uygulama alanları). Istanbul Ticaret
9 80.5686 0.3002 0.3821 Universitesi Dergisi, vol 3, 65-82 (Turkish).
[4] Sharma, T. C., & Jain, M. (2013). WEKA approach for comparative
10 80.3137 0.3033 0.3838 study of classification algorithm. International Journal of Advanced
Research in Computer and Communication Engineering, 2(4), 1925-
15 80.2353 0.3066 0.3836 1931.
20 80.098 0.2973 0.3881 [5] Yeh, I. C., & Lien, C. H. (2009). The comparisons of data mining
techniques for the predictive accuracy of probability of default of credit
25 80.1569 0.3107 0.386 card clients. Expert Systems with Applications, 36(2), 2473-2480.
[6] Witten I.H., Frank E., & Hall M.A. (2011). Data mining: practical
30 80.1176 0.3066 0.3853 machine learning tools and techniques. Elsevier, London.
40 80.4804 0.3038 0.3866 [7] Patterson, D., Liu, F., Turner, D., Concepcion, A., & Lynch, R., (2008).
Performance Comparison of the Data Reduction System. Proceedings of
50 80.6961 0.2936 0.3855 the SPIE Symposium on Defense and Security, Mart, Orlando, FL, pp.
27-34.
[8] Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P., &
Witten, I.H. (2009). The WEKA Data Mining Software: An Update.
ACM SIGKDD Explorations Newsletter, 11(1), 10–18.
[9] Wang, J., Neskovic, P., & Cooper, L. N., (2007). Improving nearest
neighbour rule with a simple adaptive distance measure, Pattern
Recognition Letters, 28(2):207-213.
[10] Zhou, Y., Li, Y. & Xia, S., (2009). An improved KNN text classification
algorithm based on clustering, Journal of computers, 4(3):230-237, 2009.
Fig. 2 Variation of error rate based on the number of neurons in hidden layer
____________________________________________________________________________________________________________
392
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Blind Source Separation (BSS) is one of the most the microphones without any delay which known as stan-
important and challenging problem for the researchers in audio dard blind source separation. Then Bell and Sejnowski in
and speech processing area. In the literature, many different 1995 developed the Independent Component Analysis (ICA)
methods have been proposed to solve BSS problem. In this
study, we have compared the performance of three popular method to solve BSS problem when the sources are mixed
BSS methods based on Independent Component Analysis (ICA) simultaneously [3]. Also some different algorithms based on
and Independent Vector Analysis Models, which are Fast-ICA, ICA such as Fast-Fixed Point ICA [4], the Jade-ICA [5], the
Kernel-ICA and Fast-IVA. We collected experimental data by EGLD-ICA, the MS-ICA [6], and the Kernel-ICA [7] were
recording speech from 13 people. Three different scenarios proposed in literature.
are proposed to compare the performance of BSS methods
effectively. Experimental results show that the Fast-IVA has As a result, BSS problem becomes more complicated for
better performance than the ICA based methods according real room environment and this speech propagation problem
to performance metrics of Source-to-Artifact Ratio, Source-to- is called convolutive blind source separation (CBSS) [8].
Distortion Ratio and Source-to-Noise Ratio. But ICA methods
give better results than Fast-IVA according to the Source-to- In the literature, some solutions were proposed in the
Interference Ratio. time domain. Due to the complicated calculation caused by
Keywords— Blind source separation, Independent component convolution, Parra et al [15] suggested another method based
analysis, Independent vector analysis on frequency domain. In frequency domain, the convolution
is replaced with multiplication to have low cost in terms
I. I NTRODUCTION of execution time. However, the frequency based methods
still have scaling and permutation ambiguities.To prevent per-
Blind Source Separation (BSS) is one of the most impor-
mutation problem, an advanced method named Independent
tant problems in speech processing area. The better descrip-
Vector Analysis (IVA) was proposed by Kim et al [9].
tion of this problem can be represented by this question: How
can we accurately determine what a particular person talks In this study we have compared the performance of two
among several speakers at the same time?. Figure 1 shows an ICA based algorithms and Fast-IVA according to perfor-
illustration of BSS problem. This problem describes the sit- mance measurement metrics commonly used for BSS prob-
uation of focusing on one speaker in case of several persons lem. This paper is organized as follow: Section II contains
talking simultaneously in same room. To separate the mixed the standard ICA method and its properties, ICA based
speech signals to obtain just a speech signal which belongs algorithms (Fast-ICA and Kernel-ICA). Section III explains
to a particular speaker is very challenging and complicated details of Fast Fixed Point IVA algorithm. Section IV gives
problem [1]. In the literature, many different methods based brief information about the commonly used performance
on signal processing and statistics were proposed to solve measurement metrics. Experimental results obtained by using
BSS problem. different proposed scenarios are represented in Section V.
Finally the conclusion is presented in the last section.
____________________________________________________________________________________________________________
393
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
394
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
vergence between two functions as the measure of depen- artifacts error terms are represented by where, sinterf , snoise
dence is employed in IVA. These two functions are the joint and sartif , respectively.
probability density function p = 1 , , n and the product of The second step involves computing the energy ratios in
probability density functions of the individual source vectors order to estimate the relative amount of each of these four
Πq(ŝi ). This function can be defined as follow: terms either on the local frames of the signal or the whole
signal duration. The way of decomposing into four terms are
J = kL(p(ŝ1 , , ŝn )||Πq(ŝi )) given in [12] in detail. Relevant energy ratios between these
K n terms are defined.
X X (7)
= const − log|det(W )| − E[logq(ŝi )] After the decomposition of j following the procedures
k=1 k=1 given in [21]. Numerical performance criteria was defined by
We can keep the dependency between the components of
computing energy ratios expressed in decibels. Definition of
each vector, and remove the dependency between the source
source-to-distortion ratio (SDR), the source-to-interference
vectors if the cost function is minimized [11].
ratio (SIR), source-to-artifact ratio(SAR) and source-to-noise
In literature, there are different version of IVA such as
ratio(SNR) are given below, respectively
NG-IVA, Fast-IVA and Aux-IVA [11]. In this study Fast-IVA
algorithm is employed for BSS. ||starget ||2
SDR := 10 log10 (12)
A. FAST FIXED-POINT INDEPENDENT VECTOR ANAL- ||einterf + enoise + eartif ||2
YSIS ||starget ||2
SIR := 10 log10 (13)
This algorithm utilizes Newtons method to update the ||einterf ||2
||starget + einterf + enoise ||2
original IVA method, which converges quadratically and SAR := 10 log10 (14)
select an efficient learning rate. In order to apply Newtons ||eartif ||2
2
method in the update rules, polynomial approximation of ||starget + einterf ||
SN R := 10 log10 (15)
a quadratic Taylor series is produced in the notations of ||enoise ||2
complex variables. In this way, it can be used for a contrast V. EXPERIMENTAL RESULTS
function of complex-valued variables [11]. The contrast The performance of Fast-ICA, Kernel-ICA and Fast-IVA
function used by Fast IVA is as follows: methods for separating mixing speech signals was compared.
n K K
X X X (K) We collected 13 speech signals recorded in real room, each
J= (E[G |ŝi (k)|2 ] λi (wi (k)wi (k) − 1) (8)
record long 10sec in Arabic language with 16 kHz. These
i=1 k=1 k=1
where, λi is the ith Lagrange multiplier, and w( i) denotes records are mixed by using random parameters.
the ith row of the unmixing matrix W, G () is the non- In our experiment, three different scenarios are proposed
linearity function, which can take on several different forms to compare performance more effectively. The first scenario
as discussed in [11]. The learning rule can be defined as includes measuring and comparing the performance of Fast-
follow with normalization: ICA, Kernel-ICA and Fast-IVA for separating mixing speech
(k) X (K) 2 X (K) 2 00 X (K) 2 signals without noise, as shown in Figure 2. Figure 3
(Wi )H ← E G |ŝi | + |ŝi | G |ŝi | illustrates the second scenario which shows the performance
K K K of these methods for separating mixed speech signals with
|ŝi |2 xK Gaussian noise added to signals before mixing. In the third
(k) (k) X (K)
x(Wi )H − E ŝi ∗ G0
K scenario, we add Gaussian noise to the signals after mixing
(9) as shown in Figure 4. Since Gaussian noise is added to
0 00
where G () and G () represent the first and second the sources or mixtures in the second and third scenarios,
derivative of G(), respectively. And if we use that for all Savitzky-Golay smoothing filter [13]] is performed to en-
sources, we can construct an unmixing matrix W(k) to be hance the signals before the separation.
decorrelated with
T −1/2
W (K) ← W (K) W (K) W (K) (10)
IV. PERFORMANCE MEASUREMENT
There are several performance measurement metrics to
Fig. 2. Illustration of Scenario 1 for the experiment
evaluate the quality of estimated signals obtained by BSS
methods. The performance of BSS algorithms can be mea-
sured by comparing each estimated source j to a given true
source j . The measuring processing includes two successive
steps [12].
The first step involves decompose j as:
ŝj = starget + einterf + enoise + eartif (11)
Where starget = f (sj ) denotes the version of sj modified Fig. 3. Illustration of Scenario 2 for the experiment
by an allowed distortion, and the interferences, noise, and
____________________________________________________________________________________________________________
395
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE IV
E XECUTION TIME OF THE ALGORITHMS
____________________________________________________________________________________________________________
396
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this work, note recognition from monophonic The last parameter is not taken into account in the existing
musical signals is studied. It is aimed to achieve a musical note systems. But making recordings with people's ideas will give
from the fundamental frequency. Artificial Neural Networks healthier and more productive results. These peoples are
(ANN) is used in order to estimate the fundamental frequency. musical instruments producers, users, recorders and listeners
Firstly, around 6-10 seconds audio recordings for each musical
The frequency content of a note is made up of a
notes are taken with the flute. Certain number of frames are
taken from these audio recordings and used for training of the fundamental frequency and its exact multiples. As in Figure 1;
ANN. After training, any parts that are played on the flute are if fundamental frequency is f0 then first harmonic is 2f0
tested. The musical notes was found correctly on the tested parts second harmonic is 3f0 and so on. [9].
that are played on the flute. Because of overlapping of fundamental frequency and
harmonic content, sound produced by a group of artists or
Keywords — Artificial neural networks (ANN), Signal instruments is percepted as same note.
processing, Monophonic music signals.
Amplitude
I. INTRODUCTION Fundamental
frequency
Music note recognition is one of the most important part of Harmonics
some music applications. This fact is emphasized on [1] and
in more specific works such as music transcription [2,3], song
retrieval[4,5] and others[6,7,8].
ANN is one the useful techniques to overcome the
drawbacks of classical signal processing and filtering
f0 2f0 3f0 4f0 Frequency
techniques. It is claimed [3] that no work that refers to ANN
for music note recognition has ever existed prior to their work Fig. 1 Frequency content of monophonic musical signals [11]
[1].
Monophonic means the production of musical signal from
The fundamental frequency relationship of two consecutive
single source in a time interval. Sound produced by a variety
notes f0 and f1 is given in Equation 1.
of sources in different time intervals does not disrupt the
monophony. Sound received in the same structure from the /
group consisting of multiples wires can be considered as = =2 ≈ 1.05946 (1)
monophonic at the stringed instrument. But as in the Turkish
instrument “bağlama”, produced sound by a thick (bam) wire A4 (440 Hz) is accepted La note other fundamental
and one or two thin wires cannot be regarded as monophonic. frequency values is calculated according to the Equation 1.
In the polyphony, multiple voices are produced in the same This means that each octave is divided into 12 equal intervals
time range [9]. [9].
Based on the physiological relationship all the sounds in
music can be handled four variables in a very simple way
[10]:
II. METARIAL AND METOD
1. Repetition rate of the audio signal is related to the
fundamental frequency. In this work, due to the simplicity of the harmonic structure
2. The amplitude of the audio signal associated with recordings from the musical instrument called "block flute” is
rotundity studied. Recordings are obtained by Sound Recorder in
3. The shape of the audio signal associated with the timbre Windows Operating System, in 21.05 kHz sampling
of musical instruments frequency and .wav format.
4. According to the sensors location of the sound source are Flut is a musical tool designed to produce sound in equal
associated with the position of the sensors different detection. order. Samples notes (do, re, mi, fa, sol, la, si, do) are taken of
____________________________________________________________________________________________________________
397
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the notes that are played on the flute. For each note 6-10 to understand it. Also simple elements may be gathered to
second audio file is saved. The audio recording of file belongs produce a complex system [12]. Networks are one approach
to do note, and its zooming part around 200 samples are for achieving this. There are a large number of different types
shown in Figure 2. Periodic structure in the zoomed image can of networks, but they all are characterized by the following
be seen very clearly. Frames containing 200 samples are components: a set of nodes, and connections between nodes.
obtained from these recordings performed separately for each One type of network sees the nodes as ‘artificial neurons’.
note. Values are normalized to the range -1 to 1 and placed on These are called artificial neural networks (ANN). An
a column of the data matrix. This procedure is repeated 400 artificial neuron is a computational model inspired in the
times for each note. As a result, 200x400 sized matrix is natural neurons [13]. In other Words, Artificial neural
created for a note. For 8 notes, by concatenating these 8 networks are an attempt at modeling the information
matrices, finally 200x3200 sized data matrix is formed. processing capabilities of nervous systems [14]. As shown in
Figure 4, Biological Natural neurons receive signals through
synapses located on the dendrites or membrane of the neuron.
When the signals received are strong enough, the neuron is
activated and emits a signal though the axon. This signal
might be sent to another synapse, and might activate other
neurons [13].
Fig. 2 Recording of the Do note (below) and a section that is backed zoom for
samples in the range 58200-58400 (above)
Fig. 3 Data samples at 200 samples length for each note (left) and target
column vectors generated for every 200 example window (right)
Fig. 5 Structure of artificial neural neuron[15]
A. Artificial neural networks (ANN)
One efficient way of solving complex problems is The sum function netj computed according to the Equation 2.
following the lemma “divide and conquer”. A complex system
may be decomposed into simpler elements, in order to be able
____________________________________________________________________________________________________________
398
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
= ∗ (2) Trained ANN was tested Twinkle Twinkle Little Star song
for a second test. This song is an international song. The
melody of the song and even music are taught in basic
The output of the processing element is calculated by netj value education institutions by many people. The song notes are
passed through the activation function (Equation 3). given in Figure 9. Figure 10 shows very clearly that the values
are assigned to the notes correctly. The persons who know the
= ( )= ( ∗ ) (3) melody can see the accuracy of the results.
0
0 500 1000 1500
6
Fig. 10 The waveform for Twinkle Twinkle Little Star song (below) and ANN
4 findings (above).
2
During the creation of the data matrix for ANN training, the
0
0 100 200 300 400 500 used preprocessing methods, such as avoid repeating each
window and take into account phase information can improve
The notes played by the flute
1.5 the performance.
1 It is observed that ANN is more suitable techniques than
0.5
classical signal processing and filtering techniques [3, 9]. In
0
-0.5
future, the workings comparing classical and ANN techniques
-1
are adviced. These workings are especially meaningful for
-1.5 generally Eastern music and Turkish music and musical
0 2 4 6 8 10
instruments because of more detailed structure of their
musical temperament.
Fig. 6 The waveform for note which played by the flute(below) and ANN
findings(above)
____________________________________________________________________________________________________________
399
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
____________________________________________________________________________________________________________
400
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The predictable and deterministic world of the past has response analysis based on the decision-maker’s intuition,
been replaced by the uncertain, random, and disorderly world of judgement, and experience. However, when the complexity of
today. Different attributes represent different dimensions of a system increases, the ability of human beings to make precise,
alternatives, which may not be easily represented on a quantitative but significant statements diminishes [4]. Therefore, the human
scale, may not be directly measurable, and may be stochastic or
reasoning usually gives results in approximate ranges [5]. In
fuzzy. Wide range applications of building performance
evaluations are based on numerical expert grading, which have addition, the decision making in real-world situations is a
been conducted for comprehensive building analysis. However, complex subject with uncertainty and vagueness [2]. In order
expert evaluation, as a nature of human knowledge, tends to be to insert human reasoning in building performance evaluation
vague or imprecise. Moreover, human decision-making needs a systems, fuzzy logic system is used as a decision making and
quick-response analysis based on the decision-maker’s intuition, evaluation method. Fuzzy logic helps to integrate uncountable,
judgement, and experience. In this study, in order to insert undefined, and uncertain information into the decision-making
uncertainty and human knowledge to building performance process by converting hidden information into workable
evaluation systems, fuzzy logic method is proposed for algorithms.
vulnerability evaluation of building characteristics. For the first
part of evaluation, fire safety performance evaluation model is II. PERFORMANCE EVALUATION OF BUILDINGS
proposed, and critical building components in terms of passive fire
safety are identified including numerical and linguistic input The evaluation methods used in traditional occupational
variables. hazards in the construction industry were mainly qualitative or
statistical approaches. In addition, construction experts tend to
Keywords— fuzzy logic, fuzzy expert system, construction project use a single crisp value during risk assessments, which may
evaluation lead to inaccurate assessment results [6]. By the time
knowledge is being transformed into their digital
I. INTRODUCTION representations in all fields, performance evaluation of building
From the earliest periods of architecture and building, parameters imported into automated code compliance checking
performance evaluations have been conditioned by rules, systems. Computerized checklist systems have advantages for
regulations, standards, and governance practices The idea of detecting building vulnerabilities faster, however, in terms of
implementing rules and regulations by architects is a reflecting the diversity of evaluation methods, they are
contradictive issue since some architects see these rules as dependent to deterministic [7]. The deterministic approach of
contradicting the creative basis of their practice [1]. On the evaluation is the disadvantage of accurate performance
other hand, new construction methods, resulting from new and evaluation in the uncertain nature of construction. Hence, the
unfamiliar materials, new organizations of functions and single-criterion evaluation prevents the integration of design
occupancy classes, new equipment in the buildings, or different objectives.
activities facilitated within building occupancy introduce new In integrated building design, architects and engineers
complex interactions. As the interactions get more complex, develop building specifications together from an early stage of
existing building performance guidelines remain too generic to the project. Therefore, various key design objectives as
understand the relationships and they become less competent in performance evaluation determinants are needed to be assessed
evaluating performance of buildings. Moreover, rapid increase to take prevention measures in preliminary design stages. The
in the complexity and embedded information in building multi-objective structure of building performance evaluation
systems drastically limits the time available for making need to be analyzed and prioritized with respect to multiple
decisions. The inherent nature of construction projects tends to attributes. The decision-making process consists activities of:
be uncertain and depends on a subjective judgment of decision ● studying the situation,
makers [2]. Therefore, the single-criterion and simple decision- ● organizing multiple criteria,
making requirements of the guidelines have given way to ● assessing multiple criteria,
highly complex decision problems involving multitudes of ● evaluating alternatives based on the assessed criteria,
variables, which may be stochastic, fuzzy, or at worst unknown ● ranking the alternatives,
[3]. On the other hand, human decision-making needs a quick- ● incorporating the judgements of multiple experts.
____________________________________________________________________________________________________________
401
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
In multi- attribute evaluation system attributes are referred In order to integrate the uncountable, undefined, and uncertain
to as goals; and different attributes represent different information in decision-making process, fuzzy expert systems
dimensions of evaluating the alternatives [3]. Due to the are used. Expert systems use domain specific knowledge to
dependency of evaluation system to expert opinion, the multi- simulate the reasoning of an expert in the field in order to
objective performance evaluation systems are suitable to be perform intelligent tasks [11]. Based on experts’ dependency
expressed by linguistic variables. When people are asked to do on common sense when solving problems by using vague and
subjective decision making on a subject in verbal categories ambiguous terms, the expert system is developed with five
rather than numerical categories, the consistency is improved main steps [12]:
and more accurate results are achieved [8]. Most words are ● specify the problem and define linguistic variables,
inherently vague and depend on some arbitrary qualification for ● determine fuzzy sets,
crisp application. Considering the operational and stochastic ● construct fuzzy rules,
characteristics of construction projects and the fuzziness of ● encode the fuzzy sets and fuzzy rules to perform fuzzy
multiple-objectives need to be considered for appropriate inference,
evaluation model [9]. Fuzzy logic, with less information and ● evaluate and tune the system.
workability of linguistic terms, can absorb human knowledge
without having to translate it into a complex mathematical Fuzzification Inference Defuzzification
model. In this research, the advantages and the challenges of interface engine interface
Crisp Fuzzy Fuzzy Crisp
evaluating building performance by using fuzzy logic is input input output output
discussed. The method is proposed for quick-response
performance assessment in the preliminary design phase, with Fuzzy
rule base
an intelligible model for each project member. In the following
parts of the study, the definition of fuzzy logic with fuzzy
Fig. 1. Fuzzy expert system model (Nilashi et al., 2011)
expert systems and hybrid fuzzy systems which combine other
methodologies with fuzzy logic are explained. In the fuzzification interface, the input variables are
structured and fuzzified through membership functions, and the
A. Fuzzy Logic degree of belonging of input variables is determined. In fuzzy
Fuzzy logic is developed by by Zadeh (1965), on the idea rule base module, if-then rules are set by experts by using
that all things admit of degrees. The classical (Boolean) logic linguistic values. Fuzzy if-then rules and fuzzy reasoning are
was expanded to consider the partial truth and all values the basis of fuzzy expert systems to structure the decision-
between “completely true (1)” and “completely false (0)” making system. On the other hand, the consequence of each
(Zimmermann). In general, the design of fuzzy controllers rule is computed by inference engine. Finally, in
consists of defining three parameters, the domain for the input defuzzification phase, fuzzy output is converted to crisp output
and output of linguistic variables, the set and the type of and the performance result is calculated by using membership
membership function for each linguistic value. A linguistic degree of the output value [13].
variable differs from numerical variable since its values are not On the other hand, for more complicated systems and
numbers but words or sentences in a natural or artificial decision processes fuzzy sets are combined with neural
language [2]. In the fuzzy logic system, the linguistic variables networks with learning and generalization capacity. Neural
describe the vague concept of decision-making (e.g. good networks are strong in pattern recognition and automatic
performance), while the membership functions define the shape learning while fuzzy logic are strong in modeling of
of the fuzzy set. Several input parameters with corresponding uncertainties.
subdivisions are combined with expert opinion to give an
output value [10]. III. APPLICATIONS OF FUZZY LOGIC IN BUILDING
Fuzzy systems used in many areas such as data classification, PERFORMANCE EVALUATION
imperfections detection and to support decision making. Fuzzy Performance requirements for building systems such as
systems have advantages to evaluate performances by serviceability, safety, security, comfort and functionality are
representing uncertainties. However, the extension of the often affected by various uncertainties. The definition of these
system is done through the addition of new rules, which may uncertainties can be hardly described by traditional
be the disadvantage since the rules depend on the existence of probabilistic methods. Moreover, translation of human needs,
experts. Therefore, for more complicated models, the fuzzy especially those concerning occupancy comfort to provide
systems combined with other systems such as fuzzy expert performance requirements often results in a vague and
systems, fuzzy neural network, fuzzy reasoning and fuzzy imprecise definition of the technical requirements. Thus, in
clustering to improve the decision-making capacities. addition to the natural randomness of basic variables,
performance requirements may be considerably affected by
B. Hybrid Fuzzy Systems fuzziness [14].
In construction management research, it is observed that Fuzzy techniques have been increasingly applied to the
hybrid fuzzy techniques are widely adopted since they can research area of construction management discipline over the
better solve problems that fuzzy logic alone may not solve [2]. last decade [4]. Fuzzy logic and hybrid fuzzy techniques are
____________________________________________________________________________________________________________
402
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
403
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
parameters for both architectural and fire safety design can be designers and engineers, entering the project data in terms of
expanded according to performance objectives and the detail numerical and linguistic values and to detect imperfections in
level of assessment. The evaluation factors and the input the form of severity levels to take necessary precautions. For
variables for each parameter defines the database set for the multi-objective building performance evaluation model,
evaluation. The input variables of the building objectives that using hybrid fuzzy systems provides advantages with learning
are generated in categorical functions with crisp numbers, and and generalization capacity, in order to define interdependency
fuzzy membership functions with triangular membership of parameters. Moreover, further research is required to test the
functions form the database. The membership functions are applicability of method in terms of quick response assessment
need to be generated for each sub-parameter under the input analyses.
parameter, such as resistance, integrity, insulation of structural
members and providing fire stop joints in concealed and REFERENCES
intersection spaces express structure separation performance [1] R. Imrie, & E. Street. Architectural Design and Regulation. John Wiley
assessment. The output membership function for each sub- & Sons Ltd., Oxford, 2011.
[2] D. Singh, & R.A. Tiong, Fuzzy Decision Framework for Contractor
parameter, parameter and building objective is required to have Selection. Journal of Construction Engineering and Management,
same number of severity levels to interrelate final linguistic 131(1), 62–70, 2005.
value. The example of five scale risk assessment levels from [3] N. Bushan, & K. Rai, The Analytic Hierarchy Process. Strategic
the least severe level “low”, to the most severe level to “high”, Decision Making, pp.11–21, 2004.
[4] A.P.C. Chan, et al., Overview of the Application of “Fuzzy Techniques”
and the intermediate levels “low-to-medium”, “medium”, and in Construction Management Research, 135 (November), pp.1241–1252,
“medium to high” are represented in Figure 4. 2009.
[5] V. Kecman, Learning and Soft Computing, Support Vector Machines,
Neural Networks, and Fuzzy Logic Models, The MIT Press, Cambridge,
MA, USA, [608] p, 2001.
[6] H. Liu, & Y. Tsai. A fuzzy risk assessment approach for occupational
hazards in the construction industry. Safety Science, 50(4), 1067–1078,
2012.
[7] E. Ayçın, & O. Özveri, Bulanık Modelleme İle Tedarik Zinciri
Performansının Değerlendirilmesi ve İmalat Sektöründe Bir Uygulama.
Akü İibf Journal, (1), pp.51–60, 2015. (in Turkish)
[8] M. Kochen, A.N. Badre. Recognızıng and Formulating Problems
Learning to Comprehend and Organizing Knowledge into Structures.
National Center for Educational Research and Development.
Washington, D.C. Regional, 1973.
[9] D.X.M. Zheng, and T. Ng, “Stochastic time-cost optimization model
incorporating fuzzy sets theory and non-replacement.” Journal of
Construction Engineering Management, 131(2) 176–186, 2005.
[10] G. Ozyurt, Fuzzy Vulnerability Assessment Model of Coastal Areas to
Fig. 4. Membership functions for risk assessment (Tah & Carr, 2000) Sea Level Rise, 2010.
[11] M.R. Mcgartland, C.T. Hendrickson, & A.M. Asce, Expert systems for
The fuzzy system to assess building performance is rule- construction project monitoring, I (3), 293–307, 1986
based model, so that the domain knowledge contains the rules. [12] M. Negnevitsky, Fuzzy expert systems. Artificial Intelligence: A Guide
Rules are generated in the If-Then form to represent the to Intelligent Systems, 2011.
[13] M. Nilashi, et al., An Application Expert System for Evaluating
expert’s reasoning process. For example, If “the interior lining Effective Factors on Trust in B2C Websites Trust, Security, ANFIS,
are A1”; Then “the fire risk severity is Low”. The literature Fuzzy Logic, Rule Based Systems, Electronic Commerce. Engineering,
reviews are conducted to develop sub-parameter evaluation 03(11), pp.1063–1071, 2011.
tables and the structured interviews are needed to include [14] H. Milan, Fuzzy probabilistic optimization of building performance.
Automation in Construction, 8(4), 437–443, 1999.
expert opinion. The success of the system relies on the ability [15] O. Taylan, A.O. Bafail, R.M.S. Abdulaal., & M.R. Kabli, Construction
to formalize and represent the knowledge within the applied projects selection and risk assessment by fuzzy AHP and fuzzy TOPSIS
field. The collected knowledge is subjective, incomplete, and methodologies. Applied Soft Computing Journal, 17, 105–116, 2014.
ill-defined, based on expert experience, judgement and [16] J.H.M. Tah, & V. Carr, A proposal for construction project risk
assessment using fuzzy logic A proposal for construction project risk
intuition. assessment using fuzzy logic, 6193. 2000.
[17] K. Knight, A.R. Fayek, “Use of fuzzy logic for predicting design cost
IV. CONCLUSIONS overruns on building projects.” J. Constr. Eng. Manage., 128 (6), 503–
Fuzzy performance evaluation of building objectives 512, 2002.
[18] S. S. Leu, A.T. Chen, and C.H. Yang, C. H. “A GA-based fuzzy optimal
provides an alternative method to regulation-based model for construction time-cost trade off.” Int. Journal of Project
deterministic evaluation methodologies. In this paper, the Management,19(1), 47–58, 2001.
limitation of studies for comprehensive fuzzy evaluation in [19] P. Bonnal, D. Gourc, and G. Lacoste “Where do we stand with fuzzy
construction field is criticized and uncertain characteristics of project scheduling?” J. Constr. Eng. Manage., 130(1), 114–123, 2004.
[20] A.V.O. Oliveros, and A.R: Fayek “Fuzzy logic approach for activity
building objectives are emphasized. As an example, fuzzy delay analysis and schedule updating.” Journal of Construction
methodology is described in relation to fire safety objective, by Engineering Management, 131(1), 42–51, 2005.
means of confining fire by construction. The performance [21] C. M. Tam, T.K.L. Tong, G.C.W. Chiu., & I.W.H. Fung, Non-Structural
evaluation model enables the project team members such as Fuzzy Decision Support System for Evaluation of Construction Safety
____________________________________________________________________________________________________________
404
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
405
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Faruk Buğra Canşe*, Derya Uluğ+, Yusuf Özgün Özdemir#, Ümit Çiğdem Turhal**
*
Bilecik Şeyh Edebali University Insttitute of Science and Technology
*
farukcanse@gmail.com, +derya.ulug@bilecik.edu.tr, #yoozdemir@yahoo.com
**
Bilecik Şeyh Edebali University Engineering Faculty
Electrics and Electronics Engineering Department
ucigdem.turhal@.bilecik.edu.tr
Abstract— The mining industry is a sector that hosts on site Turkey [1]. In mining sector due to the occupational accidents
many of occupational health and safety risks from the search each year many people come to a state that can not work and
phase to the production and shipping. In case of failure to many deaths ocur. It can be seen from the accident statistics in
prevent these risks it is encountered high occupational accidents mining across the country, especially in underground coal
and diseases. In mining, the main thing is to produce according
mine methane gas explosion and uncontrolled caving are the
to constantly changing environmental conditions that’s why it
differs from other business branches. In this situation in mining major accidents [2].
sector in determination of risks and obtaining projective Source of hazards in mines can be given as methane
measures correctly, the statisticall evaluation of the previous content in coal seam, uncontrolled caving due to lack of
accidents is very important. For this purpose there can be found support, spontaneous combustion in coal seam, gas poissoning
many learning systems that learn from the past and make or gas suffocation, coal or metal dust explosion, haulage,
estimations for the future. Therefore in this paper a classification drilling and blasting, and sub-contract system [3]. In this
model is constructed that makes estimation of the sum of injured situation in mining sector it is very important to be aware of
employees in the mines according to some attributes. In this the possible risks and to take precausition because these can
estimation model the amount of production, amount of
save lifes. In determination of risks and obtaining projective
production yield, total number of employees, amount of
dynamite in kilos, amount of capsules, mast consumption and measures correctly, the statisticall evaluation of the previous
total number of injured are used as the attributions. Thus with accidents is foremost. For this purpose there can be found
this model estimation of injured employees is performed many learning systems that learn from the past and make
according to these attributions using MATLAB platform. For estimations for the future ([2],[4]-[6]).
this study the database is used obtained from Turkish Hard Coal In this paper a classification model is constructed that
Enterprise that belongs to the years between 2010-2014. As the makes estimation of the sum of injured employees in the
classification algorithms Linear Regression and Artificial Neural mines according to some attributes. In this estimation model
Netwworks (ANN) are used and the comparative results are the amount of production, amount of production yield, total
evaluated.
number of employees, amount of dynamite in kilos, amount of
capsules, mast consumption and total number of injured are
Keywords— Learning System, Classification Model, Mine used as the attributions. For this study the database is used
Industry,Machine Learning Algorithms obtained from Turkish Hard Coal Enterprise that belongs to
the years between 2010-2014. The rest of the paper is
organized as follows: In the second section hard coal mining
I. INTRODUCTION in Turkey is considered. In the third section material and
As occupational safety is a concept of engineering, all steps methods are given, in the fourth section experimental results
for prevention of occupational accidents carry within itself the are given and in the final part conclusion is given.
stages that require engineering solutions. However although
there is a rapid development of science, technology, II. HARD COAL MINING IN TURKEY
development and industrialization it can not be said these Hard Coal is a fossil fuel and it is used as a source of
developments are exactly reflected to the process of working energy for hundreds of years. As an energy source coal has
life, occupational health and safety. Within industrial sectors many advantages and is being used for electricity production.
mining industry is a sector that hosts on site many of It is one of the most abundant sources of energy, more so than
occupational health and safety risks from the search phase to oil and natural gas. According to the 2010 BP Statistical
the production and shipping. According to the data obtained Energy Survey, Turkey had end 2009 coal reserves of 1814
from Social Security Administration of Turkish Republic in million tonnes, 0.21% of the world total. The most important
2012, 11.79% of work accidents occurred in coal mining in
____________________________________________________________________________________________________________
406
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
coal reserves of our country are located in Zonguldak Basin values are used to test the performance of the constructed
[7]. Hard coal reserves in Turkey are as given in Table 1. model.
Zonguldak Basin where deep in underground coal mining TABLE III
has been made has a complex geological structure. That’s why A SAMPLE DATABASE FOR 2010
instead of full mechanization coal production is mainly carried
out on the basis of the labor-intensive manpower. Attributes
Total injured
Mounts
Years
(Class)
Production
Production
TABLE I
Amount of
Amount of
Employee
Explosive
Mast cost
Capsule
number
Yield
Total
(kg)
COAL RESERVES IN TURKEY (TON)
Semi Coking
Jan
Noncoking
Coking
TTK
Total Feb 245.52 3.35 10.80 22.41 80.36 8.95 314
Mar 283.98 3.48 10.78 25.02 91.23 10.25 341
Apr 259.74 3.49 10.88 24.00 91.56 9.96 292
Reserve
Type May 231.59 3.42 10.68 21.94 78.38 9.31 245
2010
Jun 230.46 3.32 10.64 23.12 86.35 8.87 328
Armutçuk
Karadon
Üzülmez
Amasra
Kozlu
____________________________________________________________________________________________________________
407
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE IIIII
ONE VARIABLE REGRESSION ANALYSIS RESULTS
Attribute
of
used
Production yield
Total Employee
Production
capsules used
amount
amount
Mast costs
Explosive
Number
(kilos)
R2
Value 0.6936 0.0858 0,0106 0,0106 0,0106 0.6192
Fig. 1 Working process of a simple neuron
IV. EXPERIMENTAL RESULTS In multivariate regression model the six attributes such that
In this study three different experimental studies are the amount of production, amount of production yield, total
performed. Two of them are regression analysis and one of number of employees, amount of dynamite in kilos, amount of
them is ANN application. Two different regression analyses capsules and mast consumption are used. Then using these
are performed. One of them is simple one variable regression attribute values as the variables multivariate regression
analysis and the other one is multivariate regression analysis. analysis is performed and according to the obtained regression
model the linear line is given in Eq. 3
A. Linear Regression
Two different regression analyses are performed such as (3)
simple one variable regression analysis and multivariate
regression analysis. In one variable regression analysis each number of injured employees are estimated. The R2 for this
attribute in the database given in Table II is related with the analysis is determined as 0.733. With respect to the regression
class information that is the total number of injured. With this equation given in Eq. 3 when there is an increase in the
analysis it is aimed to show the relationships between each amount of production, total number of employees, amount of
attribute with the total number of injured. Thus it can be seen dynamite, amount of capsules an increase is also observed in
that which attribute most affects the estimation accuracy. The the number of injured. In contrast when an increase is
MATLAB simulation of this regression analysis can be seen observed in the amount of production yield and the mast
in Fig. 2 and the results are given in Table III. consumption it is observed a decrease in the number of injured.
Residual plots help to discover errors, outliers, or
correlations in the model [10]. Among the several residual
plots the histogram plot in Fig. 3 that is the simplest one and
the probability plot in Fig. 4 are preferred.
____________________________________________________________________________________________________________
408
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 4 The probability plot of residuals for the multivariate linear regression
model
Fig. 6 The ANN configuration of the classification.
The placement of the data dots as shown in the graph in
Fig.4 is not a linear line but it can be approached in such a According to the results best performance is obtained
linear line shows that the model is significant and the data has during the 26 th epoch (Fig. 7).
a positive and high correlation on the total number of injured.
B. Artificial Neural Networks (ANN)
In this classification model MATLAB ANN Toolbox (Fig.
5) is used. The six attributes given in Table II are all used as
the inputs and the class is used as the output of the ANN
configuration given in Fig. 6.
(a)
(b)
Fig. 5 ANN toolbox MATLAB a) Training part of the toolbox, b) Validation Fig. 8 Evaluation of the network in ANN Toolbox.
and testing part of the toolbox
____________________________________________________________________________________________________________
409
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] SGK (2012). Social Security Administration of Turkish Republic,
Work Accidents of 2012 (In Turkish).
[2] M.K. Özfırat Regression Modeling of Work Accidents in Underground
Mines in Turkey. Journal of Selcuk University Natural and Applied
Science pp. 37-45, 2015.
[3] Regulations of work healthy and safety in Mines (2013). Report no:
28770.
[4] H. Aydin Evaluation of the risk of coal workers pneumoconiosis
(CWP): a case study for the Turkish hardcoal mining. Scientific
Research and Essays5.21 pp. 3289-3297, 2010.
[5] Ö. D. Eratak Analysis and modeling for risk management for
underground coal mines’ safety Doctor of Philosophy in Mining
Engineering, Middle East Technical University, January 2014.
[6] A.D. Morantz Coal Mine Safety: Do Unions Make a Difference?.
Industrial & Labor Relations Review 66.1 pp. 88-116, 2013.
[7] TTK Hard Coal Report
http://www.enerji.gov.tr/File/?path=ROOT%2F1%2FDocuments%2F
Sayfalar%’F2012+Y%C4%B11%C4%B1+Sekt%C3%B6r+Raporu+(T
TK).pdf
[8] J. Fox Applied regression analysis, linear models, and related methods.
Sage Publications, Inc, 1997.
[9] J.E. Dayhoff and M.D. James Artificial neural network Cancer91.S8
pp. 1615-1635, 2001.
[10] http://www.mathworks.com/help/stats/generalized-linear-
regression.html#btdanb8-1
Fig. 9 The plot of regression for the ANN configuration.
V. CONCLUSIONS
In this study using the data obtained from Turkish Hard
Coal Enterprise that belongs to the years between 2010-2014
three different classification models are constructed. The two
of them are linear regression models and the other one is the
AAN model. In each model amount of production, amount of
production yield, total number of employees, amount of
dynamite in kilos, amount of capsules, mast consumption are
used as the attributes and the total number of injured is used as
the class information (the estimated parameter).
According to the simple linear regression model results
given in Table III the amount of production is the most
effective factor in determining the total number of the injured
and then the mast consumption are the second important
factor. The other attributes that are amount of production yield,
total number of employee amount of dynamite and amount of
capsules are not effective in determining the total number of
injured.
According to the multivariate linear regression model
results when there is an increase in the amount of production,
total number of employees, amount of dynamite, amount of
capsules an increase is also observed in the number of injured.
In contrast when an increase is observed in the amount of
production yield and the mast consumption it is observed a
decrease in the number of injured. According to this
histogram it can be seen that production levels, total number
of employee, explosive used in kilos and number of capsules
used have peaks. Thus it can be understood that these
attributes are the most effective ones in the model. And also
the probability plot in Fig.4 shows that the model is
significant and the data has a positive and high correlation on
the total number of injured. The ANN classification model
results are also significant.
____________________________________________________________________________________________________________
410
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, a solution to the Monk’s problem activation function. Since t neuron model stems from
(Monk’s 2- M2) employing a single neuron dependent on rules multiplicative neuron model, then several multiplicative
which use either modified translated multiplicative (πm) neuron neurons which have been proposed [9, 10, 11] can be
or McCulloch-Pitts neuron model, is proposed. Since M2 problem
is similar to N-bit parity problem, first N-bit parity problem is
examined to comprehend t neuron model. The model is
examined with translated multiplicative (πt) neuron. Then this defined as follows:
architecture is modified for M2 problem. Also, McCulloch-Pitts N
neuron model is used to increase classification performance. v = b ( xi ti ) y=fth(v) (2)
When the result of proposed only one πm neuron model that is
i 1
not required in any training stage and hidden layer is compared
with the other approaches, it shows satisfactory performance.
where, b R and ti R (i = 1,…,N) which are the
neuron’s adjustable parameters, are bias and weights,
Keywords— Machine learning; Translated multiplicative neuron
model; Modified translated multiplicative neuron model; N-bit respectively. The neuron’s output is y.
parity problem; Monk’s problems
The threshold activation function fth: R {0, 1} is
I. INTRODUCTION defined as follows:
Translated multiplicative neuron (t- neuron) is primarily
1, v 0
used to the N-bit parity problem. N-bit parity problem is an f th
approach to test neural network architectures and learning 0, v 0 (3)
algorithms. The N-bit parity problem is considered as a very
hard problem to be solved by neural networks, because a In fact, t- neuron model which is shown in Figure 1 is
single ‘flip’ of a bit in the input string requires a inspired from McCulloch-Pitts. McCulloch-Pitts neuron
complementary classification. The N-bit parity problem is a model is given by the following equation:
generalization of the ‘eXclusive-OR’ (XOR) problem. N-bit
N
x w . y=fth(vm)
parity problem can be explained as follows. Let x =
[x1,…,xN]T is N-bit binary vector and xi {0,1} (i = 1, …, vm = w0 + i i
N). The parity generator function which is stated as shown in i 1 (4)
Eq.1, can be determined the parity as follows:
where, w0 is bias and wi are the weights.
N
411
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
translated multiplicative t neuron model can solve the N-bit In the following section, Monk’s problem is presented.
parity problem for N 1 [1]. Previous studies in Monk’s problem are given in Section 3.
The πm neuron model and results obtained from the
application of either one πm neuron model or one McCulloch-
Pitts neuron model to Monk’s M2 problem are given in
Section 4.
412
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
M2 problem with this study are given. One significant error. The result of the Cascade Correlation algorithm for M2
characteristic of this comparison is that it was performed by a problem: After 82 epochs, 1 hidden unit: 0 Errors on training
collection of researchers, each of whom was an advocate of set and 0 Errors on test set. Elapsed real time: 7.75 seconds.
the technique they tested. Here some algorithms that have
recognition rate more than this study are explained in this IV. MODIFIED TRANSLATED MULTIPLICATIVE (ΠM) NEURON
section. Since Thrun et al. give brief description about their MODEL
studies [12], special properties of the algorithms e.g. rules that Only M2 problem is similar to parity problem among these
are used in their algorithms or other things which are based on three Monk’s problems. So, πm neuron model that is formed
neural networks will reemphasize in the following: by the algorithm of πt is applied to M2 problem. When πm
AQ17 Algorithms. AQ17-DCI algorithm is based on AQ neuron model is used stand alone, no good classification
learning programs. Here is a brief description of the AQ performance is obtained. Therefore, πm or McCulloch-Pitts
algorithm: neuron models alternatively are used according to the rules.
1. Select a seed example from the set of training examples for Data matrix that has size of 169x7 is obtained from ftp
a given decision class. server of University of California, Irvine [13]. According to
2. Using the extend against operator, generate a set of robot’s attributes, 64 of the data produced the output 1 while
alternative most general rules (a star) that cover the seed the rest produced output 0. The first experiment is done for
example, but do not cover any negative examples of the class. examining the t neuron model using 169 data matrix. The b
3. Select the “best" rule from the star according to a multi- and ti parameters of t neuron model are chosen -1 and 0.5,
criteria rule quality function (called LEF – the respectively (b=-1, t1,...,tN = 0.5, where N=24). Since the
Lexicographical Evaluation Function), and remove the robot has 6 attributes and each of them is represented by 4-bit
examples covered by this rule from the set of positive binary number, the model has 24 inputs. For the first
examples yet to be covered. classification, 105 of the data have been correctly classified
4. If this set is not empty, select a new seed from it and go to with 62.130% success. While 115 out of 169 data are already
step 2. Otherwise, if another decision class still requires rules 0, the 62.130% system performance is not satisfying for the
to be learned, return to step 1, and perform it for the other classification given above. If all the outputs of the model are
decision class. assumed to be 0, anyway 68.047% performance is obtained.
AQ17-DCI uses 2 rules for Class 0 and 1 rule for Class1. The input data are examined to get a better solution than
Backpropagation and Backpropagation with weight above. When any of x1, x2, x4, and x5 has the value “3” in
decay. There were 17 input units, all having either value 0 or decimal number system, it is observed that t neuron model is
1 corresponding to which attribute-value was set. All input not good in classifying according to N-bit parity rule. So,
units had a connection to 2 hidden units, which itself were some changes in algorithm are made by adding rules to
fully connected to the output unit. An input was classified as multiplicative t neuron model. This neuron model is named
class member if the output, which is naturally restricted to (0; as modified translated multiplicative (m) neuron model.
1), was 0.5. Training took between ten and thirty seconds on Here, b parameter in m neuron model is chosen different
a SUN Sparc Station. On a parallel computer, namely the
from t that is used for N-bit parity problem.
Connection Machine CM-2, training time was further reduced
The following rules and threshold activation function
to less than 5 seconds for each problem. The following results
given in Eq.3 are used for both m and McCulloch-Pitts
are obtained by the plain, unmodified backpropagation
neuron models:
algorithm. After 90 training epochs, the system performance
Rule 1: IF (x1=3 or x2=3 or x4=3 or x5=3) THEN b=2
was reached to 100% accuracy. Weight decay widely used
use Eq.2 ELSE b=-2 use Eq.4
technique often prevents backpropagation nets from
If only Rule 1 is used, the 125 of 169 data are correctly classified.
overfitting the training data and thus improves the
The system performance is 73.964%.
generalization. With weight decay α=0.01 Thrun improved the
classification accuracy on this third set for M3 problem Rule 2: IF x5=4 THEN b=-2 use Eq.2
significantly and, moreover, the concept learned was the same If Rule 1 and Rule 2 are used together, 143 of 169 data are
for all architectures he tested (i,e, 2, 3, or 4 hidden units). correctly classified. The system performance is 84.615%.
The Cascade Correlation Algorithm. Cascade Rule 3: IF x5=4 and ((x1 = x2 = x3 = 1 and x4 ≠ 1) or
Correlation is a supervised neural network learning (x1=3 and x2 ≠1 and x3 ≠1 and x4 ≠1 and x6 ≠ 1) or
architecture that builds a near-minimal multi-layer network (x1=2, 3 and x2 =2, 3 and x3 =2, 3 and x4 =2, 3 and
topology in the course of training. Initially the network x6=2)) THEN b=2 use Eq.2 ELSE b=-2 use Eq.2
contains only inputs, output units, and the connections If Rule 1, Rule 2 and Rule 3 are used together, 149 of 169 data
between them. This single layer of connections is trained are correctly classified. The system performance is 88.166%.
(using the Quickprop algorithm) to minimize the error. When Rule 4: IF (x1=3 or x2=3 or x4=3 or x5=3) and ((x1=
no further improvement is seen in the level of error, the x2= x3 = x6 =1) or (x1=3 and x2 = 2,3 and x3 = 2,3
network's performance is evaluated. If the error is small and x4 = 2,3 and x5 = 2,3 and x6=2) or (x1 = 2,3 and
enough, training stage stops. Otherwise a new hidden unit is x2 ≠ 2 and x3 ≠ 2 and x4 = x5 = x6 =1) or (x1 = 2 and
added to the network in an attempt to reduce the residual x2 = 3 and x3 = 2 and x4 ≠1 and x6 ≠ 1) or (x1 ≠ 3
and x2 ≠ 2 and x3 = 1 and x4 = 1 and x6 ≠ 2) or (x1 =
413
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
3 and x2 = x5 = x6 = 1)) THEN b=-2 use Eq.2 ELSE In addition to performance sequence of previous studies on
b=2 use Eq.2 Monk’s problem, m and McCulloch-Pitts neuron models
proposed in this study are given in Table 4. The results
If Rule 1, Rule 2, Rule 3 and Rule 4 are used together, 154 of obtained in this paper have higher performance when
169 data are correctly classified. The system performance is compared to the some of the studies given in Table 4. Studies
91.124%. supplying 100% performance for M2 problem are already
known. This paper proposes a new approach which is called
Rule 5: IF (x1≠3 or x2≠3 or x4 ≠3 or x5≠3) THEN m neuron model. Moreover, 6 individual rules can be defined
b=-2 use Eq.4 for the remaining 6 data, which are not correctly classified to
make system performance 100%.
If Rule 1, Rule 2, Rule 3, Rule 4 and Rule 5 are used together,
163 of 169 data are correctly classified. The system Table 4. The performance sorting for Monk’s M2 problem of different
performance is 96.45%. In the all rules, weights of the neuron methods
models are chosen as 0.5.
System
Perfor
Perfor
A study is carried out to examine the performance of m mance Method and Reference
neuron model parameters b and ti as shown in Table 3. To get mance
Sequence
the best performance, the parameters b and ti are to be chosen (%)
in M2 problem as follows: 1 AQ17-DCI / Bala et al. 100.00
2 Backpropagation / Thrun 100.00
b: ± 1 and ti: 0.3
3 Backpropagation with weight decay / Thrun 100.00
b: ± 2 and ti: [0.5-0.6]
b: ± 3 and ti: [0.8-0.9] 4 Cascade Correlation / Fahlman 100.00
5 m & McC.-P. neuron models / our study 96.45
Table 3. Performances due to different b and ti values 6 AQ17-HCI / Bala et al. 93.10
414
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Iyoda, E. M., Nobuhara, H. and Hirota, K.: A Solution for the N-bit
Parity Problem Using a Single Translated Multiplicative Neuron,
Neural Processing Letters, vol.18, pp. 213-218, 2003.
[2] Arslanov, M.Z., Ashigaliev, D.U. and Ismail, E. E.: N-bit Parity
Ordered Neural Network, Neurocomputing 48 (2002), 1053-1056
[3] Al-Rawi, M.: A Neural Network to Solve the Hybrid N-parity:
Learning with Generalization Issues, Neurocomputing, vol.68, pp. 273-
280, 2005.
[4] Hohil, M. E., Liu, D., Smith, S. H., Solving the N-bit parity problem
using neural networks, Neural Networks, vol.12, pp.1321-1323, 1999.
[5] Li, D., Hirasawa, K., Hu, J., Murata, J., Studying the effects on
multiplication neurons for parity problem, 41st Society of Instrument
and Control Engineers-SICE,2002.
[6] Kim, K., Kim, S., Joo, Y., Oh, A.S.: Enhanced fuzzy single layer
perceptron, Advances in Neural Networks, vol.3496,pp. 603-608, 2005.
[7] Setiono, R.: On the solution of the parity problem by a single hidden
layer feedforward neural network, Neurocomputing, vol.16 (3), pp.
225-235, 1997.
[8] Setiono, R., Hui, L. C. K.: Some N-bit parity problems are solvable by
feed-forward networks with less than n hidden units, Int. Joint Conf. on
Neural Networks, 1993, pp. 305-308.
[9] Schmitt, M.: On the complexity of computing and learning with
multiplicative neurons, Neural Computation, vol.14(2), pp. 241-301,
2002.
[10] Zhang, B.-T.: A Bayesian Evolutionary Approach to The Design and
Learning of Heterogeneous Neural Trees, Integrated Computer-Aided
Engineering, vol. 9(1), pp. 73-86, 2002.
[11] Bas, E., Uslu, V. R. and Egrioglu, E.: Robust learning algorithm for
multiplicative neuron model artificial neural networks, Expert Systems
with Applications, vol.56, pp. 80-88, 2016.
[12] Thrun, S.B., Bala, J., Bloedorn, E., Bratko, I., Cestnik, B., Cheng, J.,
De Jong, K., Dzeroski, S., Fahlman, S.E., Fisher, D., Hamann, R.,
Kaufman, K., Keller, S., Kononenko, I., Kreuziger, J., Michalski, R.S.,
Mitchell, T., Pachowicz, P., Reich, Y., Vafaie, H., Van de Welde, W.,
Wenzel, W., Wnek, J., and Zhang, J.: The Monk’s Problems: A
Performance Comparison of Different Learning Algorithms, A Report,
Carnegie Mellon University CMU-CS-91-197, 1991.
[13] University of California, Irvine Dataset: [Online]. Available:
ftp://ftp.ics.uci.edu/pub/machine-learning-databases/monks
problems/monks-2.train
415
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Facility layout which increases productivity and healthy come emergency department. While taking care of
provides the most effective way to run the organization, is used patients, hospital staff, materials and flow of information
both determine the location of departments, assistant services, between emergency service departments must be quick. In this
facilities and edit this sections. There are numerous studies in the
regard layout is important.
literature for the solution of facility layout’s problems. We have
In this study after receiving the necessary data from the
observed that the heuristics are preferred in some of these
solutions. We used data mining in this study. Data mining information systems department of the Emergency
continues to offer new horizons, perspectives and new methods in Department in Kirikkale University Faculty of Medicine
a new area every day. The basic objective of this study is to Hospital, data correction and data categorization work was
develop a model of the layout for the purpose of improving the done. Then the data table was converted into text file format
emergency department's functionality, the quality of patient that accepted by the apriori software format. After the data
care, the satisfaction of patient and employee. Firstly in this preparation stage , apriori algorithm was applied on the data
study data warehouse was created by using data of the past year which was categorical structure according to the
in the hospital information system. Data warehousing is the entry predetermined support and confidence value and association
of the association analysis model. Then a system will be develop
rules was obtained by using SPSS Clementine software. Then
to solve the problem using association analysis with Apriori
association analysis which is one of the methods of data mining. the desired rules were generated from large item sets and then
The application of the study is carried out in Kirikkale these association rules were used as a reference during the
University Faculty of Medicine Hospital. The goal of this study to decision-making process for developed algorithm. This
develop a model proposal through utilizing the method of algorithm was applied on apriori algorithm’s results. Finally
Facility layout techniques for emergency department which will alternative facility layout planning was created.
be designed according to way of intervention on patients. Data mining algorithms have been used in various Facility
Layout Problem in literature review. K-means clustering
algorithm is applied to evaluate supermarket’s layout Larson
Keywords— Apriori Algorithm, Association Analysis, Data
et al.[2]. Ay and Çil [3], in their study, used apriori algorithm
Mining, Emergency Department, Facility Layout Problem
and multidimensional scaling to acquire relational database.
I. INTRODUCTION Kundu and Dan [4] used genetic algorithm so as to cope up
with facility layout problems. The aim of Altuntas and Selim’s
Data mining is process that various advance data analysis
study [5] is to propose new weighted association rule-based
based on statistics and artificial intelligence are used by way
data mining approaches for facility layout problem. In Cil[6]’s
of visual programming interface in order to find out hidden
study, different weighted association rule-based data mining
patterns and relations in huge data stacks. Data mining
approaches was applied to facility layout problem. Lee [7]
algorithms include statistical algorithms, mathematical
offers an integrated model that forecasts the total walking time
algorithms and artificial intelligence algorithms (such as
of a passenger by simulation and searches for a near-optimal
neutral networks, decision trees, association rules) together.
layout by ant colony optimization. Aiello et al. [8] offers in
Emergency department where data mining technique was
their study a new multi objective genetic algorithm for dealing
applied is the first line in hospitals to face emergency patients.
with unequal area facility layout problems. In Ozcan and
As a major function of emergency department, when a patient
Esnaf’s study[9], a mathematical model is propose based on
comes to the emergency department, the emergency
association rule mining for store layout problem which
department personnel perform a triage procedure and then
involves the identification of the position of products and
transfer the patient to associated departments for treatment [1].
product categories which are shown in-store shelves. Yener
For patients waiting in the emergency services, delay in the
[10] aims to find an interior design of a warehouse that is
medical examinations decrease the quality of service and
minimizing the costs occurs while orders gathering within the
customer satisfaction. At the emergency services these factors
main distribution warehouse which is performing in retailer
are of much more importance since patients who have risky
____________________________________________________________________________________________________________
416
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
sector. Performances of different warehouse designs obtained Several techniques and algorithms such as Classification,
via different methods are examined with the aid of simulation Clustering, Regression, Artificial Intelligence, Neural
techniques. In the workshop, data mining techniques which Networks, Association Rules and Decision Trees are used for
have association rules are also benefited from. Gonçalves and information exploration from databases [19]. One of the
Resende [11], in their study propose a biased random key techniques is clustering which is used to describe the object of
genetic algorithm for the unequal are a facility layout problem similar classes. The clustering technique defines the classes
where a set of rectangular facilities with given are a and assigns each object to a particular class. Classification
necessities has to be placed, without aliasing, on a rectangular widely used data mining technique, to develop a class and
floor space. Arulkumar and Saravanan used to solve the layout learning. Most frequent item in the large data set is found by
problem Artificial Bee Colony algorithm and Particle Swarm association analysis. The main aim is to release interesting
Optimization algorithms in their paper [12]. In Aktepe et al.’s correlation and association between a big data set [17].
study [13], association analysis which is one of the data Regression analysis can be used to model the relationship
mining technique and VIP-Plant Optimization was used to between one or more independent variables and dependent
develop a new proposal of facility layout for a shopping center. variables. In data mining independent variables are attributes
In paper of Lin and Guan [14] a hybrid algorithm based on which is known and response variables are what we want to
variable neighborhood search and ant colony optimization is forecast. Neural network implies a set of connected input or
pro-posed to solve the single row facility layout problem. output units and each connection has a weight present with it.
Studies implemented FLP and Data mining in hospital are Neural networks have the significant ability to derive meaning
very little in literature review. The contribution of paper of from complicated or indefinite data and can be used to acquire
Yeh [15] is a new framework which is named annealed neural patterns and find trends that are too complicated to be noticed
network. Aim of framework is to find vying solutions for the by either humans or other computer techniques [18]. Decision
FLP efficiently. This framework take place the integration of tree is one of the most important classification methods.
Hopfield neural networks and simulated annealing. A Decision tree is a method used to divide data sets including
Hopfield neural network is a representation model of the more registries into smaller data sets via an array of decision
layout problem and simulated annealing is a search algorithm rules. Artificial intelligence is capability that a computer or a
for finding the optimum or near optimum solutions. In Cakar’s robot controlled by computer carries out various activities like
PhD thesis [16] genetic algorithm is used to optimize the intelligence living creature.
emergency department's layout.
In the second section, methodology which is used in this
study was mentioned. Studies which were done, developed B. Association Analysis in Data Mining
algorithm, new facility layout which was generated were
Association analysis, one of the most important and well
explained under application title that was explained in the
researched techniques of data mining, was first introduced in
third section. In the fourth section, summary of study was
study of Agrawal et al. [19]. The association rule algorithm is
explained and improvement acquired was expressed
commonly used to identify relationships between items or
numerically.
features that occur synchronously in a database. The Apriori
II. METODOLOGY algorithm [20] is widely used to mine frequent item sets and
learning association rules in a datamining field. It seeks to
generate the desired rules from large item sets and then uses
A. Data Mining these association rules as a reference during the decision-
making process. The Apriori algorithm is usually divided into
Data mining is the process of finding of information which two separate steps. First, minimum support is applied to find
is unseen from a big data. Data mining, analyzing the data all of the frequent item sets in a database. Second, these
from different source and transform it meaningful information frequent item sets and the minimum confidence constraint are
[17]. Data mining is also named as knowledge discovery used to form rules [21]. Association rules are rules presenting
process, knowledge extraction, data analysis or knowledge association or correlation between item sets.
mining from data.
C. Apriori Algorithm used for Association Analysis
Apriori algorithm the most popular algorithm which is used
for the association rules discovery. The algorithm's name
comes from the word "before (prior)".Because this algorithm
receives information from a previous step. Apriori algorithm’s
steps are given below:
Determination of the number of minimum support
and the value of minimum confidence.
Finding the value of support of the each element in
Fig. 1 Information discovery process [18].
the element sets.
____________________________________________________________________________________________________________
417
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Information and
Material Handling
System Design
____________________________________________________________________________________________________________
418
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSIONS
____________________________________________________________________________________________________________
419
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
420
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
{volkan.gangal, gokcehacioglu}@ktu.edu.tr
Abstract—This study contains a simulation of such as traffic jams, road length, and average delay time for
dynamic road navigation system for intelligent transport ant colony algorithm. In this way, the proposed approach can
system. The proposed system uses wireless sensor easily find optimum route between two points based on
networks in roadsides. Through a vector map of Trabzon current data in Trabzon city.
and ant colony algorithm, the optimum route is calculated
in vehicles. Wireless sensor networks provide some B. ANT COLONY ALGORITHM
coefficients such as traffic jams, road length, and average
delay time for ant colony algorithm. A small area in Ants live in colonies and they make division of labor
Trabzon is considered for the proposed dynamic road between them to overcome the difficulties. When an ant finds
navigation. In literature search, it has not found any the food source, the other ants can reach the food source easily
realistic study that contains intelligent transportation by using the most convenient way. The chemical which causes
system for Trabzon city. that phenomenon known as pheromone. Ants leave pheromone
chemical on the roads which they pass. In addition, they prefer
the ways that have much pheromone. Also the pheromone
Keywords— Ant Colony Algorithm, Wireless Sensor substance has evaporation property. Therefore the amount of
Networks, Dynamic Road Navigation, Intelligent pheromone is much more where they frequently pass. Assume
Transport System that the ants have same speed and they drop pheromone with
same rate. Therefore, there will be much pheromone on the
shortest path and the ants will prefer that route to go to food
source [5,6].
I. INTRODUCTION
In recent years, there is a growing interest to navigation There are two main parameters in the Ant Colony
works. However, many of these worksdon’t contain actual Algorithm. The first one is selection possibility of the route
information about the road or roadside. The proposed method and the second one is evaporation rate of pheromone.
uses Ant Colony Algorithm and coefficients which come from
WSN to determine the reliable and short route to the target.
____________________________________________________________________________________________________________
421
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
422
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
423
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The selection of the appropriate pattern for palmprint cost of these methods makes difficult to apply them on the
recognition systems depends on the success of the segmentation palmprint recognition system.
techniques used. Segmentation can be difficult due to free In this study in order to eliminate alignment problems, a
movement of the hand in unrestricted environment. Also holding stereo camera device which has two webcam is used. Stereo
the hand at different angles causes perspective distortions. In
camera systems are more advantageous in terms of cost
order to correct these distortions, active scanning systems have
been proposed to find the 3D position of the hand. However, as the compared to other 3D techniques. However, the computational
cost of these systems is high, it seems that they are not possible to complexity of these systems makes it difficult to precise 3D
spread. scanning. In the proposed method rather than precise 3D
In this study, a stereo camera system which is more cost effective, scanning, it has been tried to dermine number of landmarks that
have been used. With the help of this system, 3D coordinates of the may help finding the orientation and location of the hand. Thus,
landmarks falling on hand are calculated. A new transformation factors affecting the recognition in a negative way emerged in
is defined based on the main axis of the distribution of these points. contactless systems (rotation, translation and scaling like) are
Thus perspective distortions are substantially corrected. The eliminated to a great extent. After determining hand’s pose in
success of the proposed approach has been tested on our dataset
3D space, perspective correction is performed using geometric
consisting of hand images taken from 138 people. As a result of the
experiments, recognition rate even in the cases reached over 90%. transformations. In order to evaluate the performance of the
proposed approach, a dataset that contains the stereo images of
138 different people is used. Pose corrections on the images in
Keywords— Palmprint Recognition, Stereo Camera, Pose
Correction the dataset are made and necessary patterns for recognition are
produced. Hand segmentation on the corrected images and the
I. INTRODUCTION detection of the palm area (ROI) are made using Active
In the first devices developed for palmprint validation [1, 2] Appearance Method (AAM). Feature vector is produced by
cameras which have own lighting system are used. At the time Gabor based Kernel Fisher Discriminant Analysis method
of taking image, a flat surface is used to keep the hand at the (Gabor-based KFA). Cosine Mahalanobis (cosmah) distance is
same distance. The movement hand is significantly restricted used for similarity measure between the vectors.
by the pegs in the device. Despite the successful results
II. MATERIAL AND METHODS
obtained in these studies, it is mentioned that the pegs cause
distress for some users (especially those with physical problems A. Hand Acquisition Device
in their fingers) [3-6]. Then unrestricted systems without pegs
An open access dataset of stereo-based palm images was not
are proposed. [7-10] However, they are contact systems and
found in the literature. Therefore, in order to test the proposed
cause hygiene problems especially in a multiuser evironments
approach in this study, a new dataset was created consisting of
[11].
stereo image pairs. In the device prepared to generate this
Recent studies about hand biometric systems has focused on
dataset, two CMOS cameras having the same physical
unrestricted and contactless techniques that do not need user
properties were used (Fig. 1). The cameras were placed
cooperation. In the unrestricted and contactless systems, in
vertically in the system (up and down cameras). The camera
order to overcome alignment problems, hand’s pose and the
system has its own light source to avoid exposure of ambient
distance to the camera should be known. It is very difficult to
light sources. 12 V led strips are used for lighting. The height
obtain these informations in 2D image. For this reason, some
of the protective box was 25 cm. The distance from the central
researchers have proposed 3D hand acquation systems with
focus of the two cameras was approximately 3 cm. In the stereo
different characteristics (digital laser scanner [12], structured
camera system, the internal values and the detection of relative
ligth [13], the depth camera [4], ultrasound [14] etc.). 3D
positions of the camera used are required to be gathered the 3D
information offered by these systems raise recognition rates as
information. Thereby, pre-treatment (camera calibration) is
well as presenting some security advantages. However, extra
____________________________________________________________________________________________________________
424
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
available in the system. For calibration, 32 chess images The pose correction process was conducted according to the
consisting of 7x5 squares with edge length of 25 mm are used. distribution of the 3D point cloud detected on the hand. As
shown in Fig. 3a, these points are approximately in the form of
an ellipse. This ellipse (𝒆𝒆����⃗, ����⃗,
𝟏𝟏 𝒆𝒆 ����⃗)
𝟐𝟐 𝒆𝒆 𝟑𝟑 and its main axis, which are
perpendicular to each other, can be found using Principle
Component Analysis (PCA). The eigenvector
����⃗)corresponding
(𝒆𝒆 𝟏𝟏 to the maximum value detected by PCA is
the primary main axis. This axis is in line with the fingers and
′
usually falls on the middle finger. If it is assumed that 𝑃𝑃3𝐷𝐷 and
𝑃𝑃3𝐷𝐷 are new and current 3D points respectively, then the
transformation is represented as:
Fig. 1 Hand Acquisition System ′
𝑃𝑃3𝐷𝐷 = 𝑅𝑅𝑃𝑃3𝐷𝐷 (1)
B. Dataset where R is the 3 × 3 transformation matrix and is defined as
In the dataset, there were a total of 2760 images (640 × 480 follows:
pixels) taken from 138 (47 female and 91 male) individuals −1
based on 10 different scenarios (S1, S2, ..., S10). Images were 𝑒𝑒���⃗1
taken according to previously identified scenarios by giving 𝑅𝑅 = ����⃗
𝑒𝑒2 � (2)
instructions to the users, but no intervention was made during 𝑒𝑒3
���⃗
the making of the images. In six of the specified scenarios (S1,
The 3D points detected on the hand and the transformation
S2, S3, S4, S5 and S10), it has been asked from user to keep
results of their centralized coordinate values are shown in Fig
their hands parallel to cameras. In the others, it has been asked
3b. It can be seen from the figure that the center of the
from users to keep their hands in different directions with an
transformed points is O(0,0,0) and the main axis is along the X,
angle of about 30 degrees (S6, S7, S8 and S9).
Y and Z axes.
C. Pose Correction
Perspective distortions arises when the hand is not parallel (a) (b)
to the camera plane. Kanganhad et al. [3] apply Iterative
Reweighted Least Squares approaches on the point cloud Fig. 3 (a) The ellipse form of 3D points; (b) original points (black) and
obtained from digital 3D scanner to make pose corrections and transformed points (green)
have determined the normal vector of the hand plane. 3x3 In order to apply the proposed transformation to the image,
transformation matrix is defined with the help of this vector. the 𝑃𝑃2𝐷𝐷 constant matrix was created:
Another approach can be seen in [15]. In this study, the relative
𝑦𝑦1 − 𝑦𝑦𝑐𝑐 𝑦𝑦2 − 𝑦𝑦𝑐𝑐 … 𝑦𝑦𝑛𝑛 − 𝑦𝑦𝑐𝑐
coordinates of 7x5 pieces light points directed from light sourse
to the hand is calculated. Projective matrix is created with the 𝑃𝑃2𝐷𝐷 = �𝑥𝑥1 − 𝑥𝑥𝑐𝑐 𝑥𝑥2 − 𝑥𝑥𝑐𝑐 … 𝑥𝑥𝑛𝑛 − 𝑥𝑥𝑐𝑐 � (3)
help of the normal vector of the plane formed by points. 0 0 … 0
Geometric correction was performed with this transformation where 𝑥𝑥𝑖𝑖 and 𝑦𝑦𝑖𝑖 are image coordinates, 𝑥𝑥𝑐𝑐 and 𝑦𝑦𝑐𝑐 are the
matrix. On the correction step of these experiments, rather than image's center points and 𝑛𝑛 is the number of pixels in the image.
finger positions and the distance between hand and the camera, In this representation, the image coordinates are assumed as a
midpoint of the hand is taken as reference and and the point cloud along the axis of 𝑍𝑍 = 0 in 3D space. The new pose
correction process is made in that position. As the background corrected coordinates can be found with the following formula:
is monochrome, segmentation is performed without problem.
′
In our study, first landmarks on each stereo cameras are 𝑃𝑃2𝐷𝐷 = 𝑅𝑅𝑃𝑃2𝐷𝐷 (4)
detected by SURF[16] method. Depth information of these
′
points is obtained with the help of epipolar geometry 𝑃𝑃2𝐷𝐷 , represents a transformed new surface in the 3D space
′
techniques by matching. Assuming that the hand is the nearest and the 𝑍𝑍 values are different from 0. 𝑃𝑃2𝐷𝐷 points are
object to the camera plane, all points over a certain distance transformed to the homogeneous coordinates, and as in
from the plane have been eliminated. Thus, only point clouds equation 5, the perspective projection of the axis 𝑍𝑍 = 𝑑𝑑 is
falling on hand was detected. obtained.
____________________________________________________________________________________________________________
425
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
2
1 0 0 0 ��𝐼𝐼�𝑊𝑊(𝑥𝑥; 𝑝𝑝)� − 𝐴𝐴0 (𝑊𝑊(𝑥𝑥; Δ𝑝𝑝))� (9)
′
𝑃𝑃2𝐷𝐷 = �0 1 0 0� 𝑃𝑃2𝐷𝐷
′
ℎ
(5) 𝑥𝑥
𝑓𝑓
0 0 1� 1 where 𝐴𝐴0 is an average image, 𝑊𝑊(. ) represents the warping
𝑑𝑑
process, 𝑝𝑝 is the necessary parameter for the warp operations,
where 𝑑𝑑 is the average distance of the points on the hand to and Δ𝑝𝑝 shows the parameter change in the next iteration.
the camera plane. The new image coordinates were determined In this study, 51 points on 70 randomly selected images in
′
by adding 𝑥𝑥𝑐𝑐 and 𝑦𝑦𝑐𝑐 values to 𝑥𝑥𝑓𝑓𝑖𝑖 and 𝑦𝑦𝑓𝑓𝑖𝑖 in the matrix 𝑃𝑃2𝐷𝐷 𝐼𝐼
. the stereo database are marked for the training of AAM model
(Fig. 5). Shape and appearance models of annotated images
𝑥𝑥𝑓𝑓′ = 𝑥𝑥𝑓𝑓𝑖𝑖 + 𝑥𝑥𝑐𝑐 𝑎𝑎𝑎𝑎𝑎𝑎 𝑦𝑦𝑓𝑓′ = 𝑦𝑦𝑓𝑓𝑖𝑖 + 𝑦𝑦𝑐𝑐 (6) have been created by training with AAM. AAM points on other
𝑖𝑖 𝑖𝑖
images has been tried to be determined using this model. 3 level
Intensity values corresponding to each coordinate were image pyramid has been created to perform the search process
found by making an interpolation of the values in the original faster.
image. With the help of the transformation, the middle finger
was arranged along the Y axis in the new image. Thus, the pose
of the hand in 3D space was corrected and the fingers in the
new image were aligned to face upward (Fig. 4).
E. ROI Extraction
Detection of the ROI is one of the most important stages of
palmprint recognition and affects recognition performance
significantly. As AAM-based hand segmentation uses texture
(d) (e) (f) information, it provides a significant advantage in the detection
Fig. 4 Pose correction: (left column) rectified images; (middle column)
of the valley points. In this study, the ROI extraction method
resampled images obtained using pose correction method; (right column) used is similar to that proposed by Zhang et al. [1], with an
interpolated images approach allowing a wide area to be selected (Fig. 6c). The
valley points required by this method are selected from the
D. Hand Segmentation control points determined by the AAM. Here, the distance (d)
The AAM [17] is based on the philosophy of finding the model between the two valley points was via a proportional
parameters that minimize the difference between the target relationship. The selected valley points were P15 and P37 (Fig.
image and the learned average image. Training images 5). The ROI was determined to be a square with each side 1.4
previously annotated by an expert are used to create an average d and 0.2 d from the mid-point of the valleys. The obtained
image model. Points marked on each image are aligned with ROIs were scaled to 128 × 128 pixels.
Procrustes Analysis and the shape model of the AAM is created.
Variations of the shape model are determined by PCA.
According to the average shape image, each image in the
training set is subjected to a warp process. The texture model is
obtained by transforming textures within the delaunay
triangulation established between points of the warped image
to a vector [18, 19].
𝑠𝑠 = 𝑠𝑠0 − 𝜙𝜙𝑠𝑠 𝑝𝑝 𝑝𝑝 = 𝜙𝜙𝑠𝑠𝑇𝑇 (𝑠𝑠 − 𝑠𝑠0 ) (7)
(a) (b) (c)
𝑔𝑔 = 𝑔𝑔0 − 𝜙𝜙𝑡𝑡 𝑞𝑞 𝑞𝑞 = 𝜙𝜙𝑡𝑡𝑇𝑇 (𝑔𝑔 − 𝑔𝑔0 )
(8)
Fig. 6 Various RIO extraction methods: (a) Zhang et al. [1]; (b) Connie et al.
where 𝑠𝑠 and 𝑔𝑔 are the shape vector and synthesized shape- [11]; (c) proposed method. Skin color segmentation is used in (a) and (b); AAM
free texture vector; 𝑠𝑠0 and 𝑔𝑔0 are mean shape and mean texture; segmentation is used in (c).
𝜙𝜙𝑠𝑠 and 𝜙𝜙𝑡𝑡 are the eigenvectors of shape and texture; 𝑝𝑝 and 𝑞𝑞
ROI extraction approach is applied on the image before the
are shape and texture parameter vectors, respectively.
pose correction actually the importance of correction emerges
It has been recommended that the inverse compositional
clearly. Related results of the same images are shown in Fig. 7.
image alignment approach be the adapted AAM [20]. This
algorithm tries to minimize equation 9.
____________________________________________________________________________________________________________
426
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
427
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
used for training did not match with itself. The cumulative
match curves (CMCs) of the first comparisons are shown in Fig.
8, and the verification rates are given in Table 2.
(a) (b)
Fig. 8 CMCs for the unangled scenarios: (a) Up camera; (b) Down camera Fig. 9 ROC curve for the angled scenarios before and after transformation
(a) (b)
Fig. 10 Results of fusion models: (a) ROC curves of fusion models; (b) FAR-
FRR graph of Fusion 2.
____________________________________________________________________________________________________________
428
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSIONS [6] Michael, G. K. O., Connie, T. and Teoh, A. B. J. (2012) A contactless
biometric system using multiple hand features. Journal of Visual
In this study, the pose information for the palm was obtained Communication and Image Representation. 23(7) 1068-1084.
from stereo-based hand images in a 3D environment and the [7] Connie, T., Jin, A. T. B., Ong, M. G. K. and Ling, D. N. C. (2005) An
images were converted with the help of the proposed geometric automated palmprint recognition system. Image and Vision Computing.
23(5) 501-515.
transformations. Thus, a pre-processing operation was [8] Ribaric, S. and Fratric, I. (2005) A biometric identification system based
produced for the powerful 2D recognition techniques. It is on eigenpalm and eigenfinger features. Ieee Transactions on Pattern
evident that the existing 3D systems are more advantageous for Analysis and Machine Intelligence. 27(11) 1698-1709.
feature extraction than 2D. However, the extra cost of these [9] Lin, C. L., Chuang, T. C. and Fan, K. C. (2005) Palmprint verification
using hierarchical decomposition. Pattern Recognition. 38(12) 2639-
systems and their nonportable devices reduce their applicability. 2652.
In this study, 3D information was used only to make geometric [10] Oliveira, H. P. and Magalhaes, F. (2012) Two Unconstrained Biometric
corrections with a different perspective because the web Databases. Image Analysis and Recognition, Pt Ii. 7325 11-19.
cameras, which were used to keep the recognition system’s cost [11] Morales, A., Ferrer, M. A. and Kumar, A. (2011) Towards contactless
palmprint authentication. Iet Computer Vision. 5(6) 407-416.
low, were incapable of producing very detailed 3D point [12] Kanhangad, V., Kumar, A. and Zhang, D. (2011) Contactless and Pose
information. Detection of depth in high-resolution stereo Invariant Biometric Identification Using Hand Surface. Ieee
images does not seem advantageous for real time applications. Transactions on Image Processing. 20(5) 1415-1424.
The most important advantage of this study is that very detailed [13] Li, W., Zhang, D., Lu, G. and Luo, N. (2012) A Novel 3-D Palmprint
Acquisition System. Ieee Transactions on Systems Man and Cybernetics
3D information is not needed. Also miscalculations of some Part a-Systems and Humans. 42(2) 443-452.
points because of noise in the images can be tolerated at [14] Iula, A., Savoia, A. S. and Caliano, G. (2014) An ultrasound technique
determination of hand plane equation. for 3D palmprint extraction. Sensors and Actuators a-Physical. 212 18-
One of the major advantages of the study was that two palm 24.
[15] Sato, T., Aoyama, S., Sakai, S., Yusa, S., Ito, K., Aoki, T. and Ieee.
patterns could be produced from two different cameras at the (2013) A Contact less Palm Recognition System Using Simple Active
same time and used by the proposed simple fusion models to 3D Measurement with Diffraction Grating Laser. 2013 Second Iapr
improve the chances of success. Asian Conference on Pattern Recognition (Acpr 2013) 542-546.
In this study, in order to make the evaluations needed, a [16] Bay, H., Ess, A., Tuytelaars, T. and Van Gool, L. (2008) Speeded-Up
Robust Features (SURF). Computer Vision and Image Understanding.
stereo-based dataset was compiled using images taken from 110(3) 346-359.
138 different people. As a result, the recognition rate ratios of [17] Cootes, T. F., Edwards, G. J. and Taylor, C. J. (2001) Active appearance
about 62% obtained by each camera evaluated individually models. Ieee Transactions on Pattern Analysis and Machine Intelligence.
were increased to 93.17% by employing a simple stereo fusion 23(6) 681-685.
[18] Aykut, M. (2013) Esnek Kullanımlı Avuç İzi Bölgesine Dayalı
approach. Doğrulama Sistemlerinin Tasarım Çalışması. Computer Engineering.
This study revealed the need for perspective correction on Karadeniz Technical University.
the images taken parallel to the camera plane in contactless and [19] Bingol, O., Ekinci, M., Gokdogan, A. and Ieee. (2014) A New Approach
unconstrained systems. Although after the evaluations, Stereo Based Palmprint Extraction in Unrestricted Postures. 2014 Ieee
Workshop on Biometric Measurements and Systems for Security and
significant improvements were observed for all scenarios, the Medical Applications (Bioms) Proceedings 44-49.
obtained results could be further improved upon. In particular, [20] Matthews, I. and Baker, S. (2004) Active appearance models revisited.
on the high-angle images, pattern distortions can arise, International Journal of Computer Vision. 60(2) 135-164.
depending on the 3D structure of the hand. Furthermore, some [21] Ekinci, M. and Aykut, M. (2013) Kernel Fisher Discriminant Analysis
Of Gabor Features For Online Palmprint Verification. Turkish Journal
users are unable to completely extend their fingers, in which of Electrical and Computer Engineering. 23(5) 1-18.
case, the inner surface of the hand resembles a curved structure. [22] Lee, T. S. (1996) Image representation using 2D gabor wavelets. Ieee
This is one of the factors that negatively affected the success of Transactions on Pattern Analysis and Machine Intelligence. 18(10) 959-
the system. In the future, studies must further examine the 971.
[23] Mika, S., Ratsch, G., Weston, J., Schölkopft, B. and Müller, K.-R. (1999)
problem of correcting the curves of the palm, and develop Fisher Discriminat Analysis with Kernels. Neural Networks for Signal
techniques to improve the quality of the images. Processing. X(1) 41-48.
[24] The CSU face indentification evaluation system user’s guide: Version 5,
2003, University, C. S., Denver.
REFERENCES
[1] Zhang, D., Kong, W. K., You, J. and Wong, M. (2003) Online palmprint
identification. Ieee Transactions on Pattern Analysis and Machine
Intelligence. 25(9) 1041-1050.
[2] Han, C. C. (2004) A hand-based personal authentication using a coarse-
to-fine strategy. Image and Vision Computing. 22(11) 909-918.
[3] Kanhangad, V., Kumar, A. and Zhang, D. (2011) A Unified Framework
for Contactless Hand Verification. Ieee Transactions on Information
Forensics and Security. 6(3) 1014-1027.
[4] Wang, C., Liu, H. and Liu, X. (2014) Contact-free and pose-invariant
hand-biometric-based personal identification system using RGB and
depth data. Journal of Zhejiang University-Science C-Computers &
Electronics. 15(7) 525-536.
[5] Han, Y., Sun, Z., Wang, F. and Tan, T. (2007) Palmprint recognition
under unconstrained scenes. Computer Vision - Accv 2007, Pt Ii,
Proceedings. 4844 1-11.
____________________________________________________________________________________________________________
429
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
430
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
different methods like regression models, artificial There are many advantageous aspects of ANN such as being
intelligence-based (ANN, fuzzy logic, support vector able to do parallel processing, ability to learn, making
machines, etc.) used in different estimates based operations. generalizations and ease of analysis and design. One of the
In this study the ANN and RM which are the most widely most important features is that different information can be
used methods have been used. derived without requiring help from the new knowledge
In this study, at first an experimental study has been acquired through learning.
carried out to determine the fire resistance of fire doors. In this study, for the predict the in-room temperature (oC),
Thermocouples were placed at the surface of door and were an ANN structure with 7 inputs and 1 output was designed
connected to data logger to record the data. System was and application was performed using the designed structure
modeled by an artificial neural network with 7 inputs and 1 (Fig. 2). As input parameters, Top Left, Top Right, Middle
output which is an artificial intelligence techniques and RM Left, Middle Right, Bottom Left, Bottom Right Temperature
using non-linear experimental data and fire room interior (oC) and Time (minute), as output parameter In-room
temperature values were estimated. These two models were Temperature (oC) were taken. The performance of the ANN
compared with each other and the model that give the closest was presented by examining the consistency between the
result to experiments have been identified. values obtained through the ANN approach and the
experimental data [10].
____________________________________________________________________________________________________________
431
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Generally, multiple regression analysis is used to find mean In this study, modelling approach is developed with ANN
output values relating to dependent variable [9, 11]. by utilizing the data obtained experimentally [10]. Using
Functional form of simple linear regression model can be statistical analysis of the same data, the system of single and
written in the form of equation (1) for a mass [9]. multiple regression equations was created with the SPSS
Y = β0 + β1X + ε program. Temperature values were found with the estimation
(1) of in-room temperatures (°C), by using the mathematical
In this model; expressions obtained with RM. RM results were compared
Y: Value of dependent variable, separately with the values obtained by both the experiments
X: Value of independent variable, and ANN model. The descriptive statistics 136 values for the
β0: Sectional of Y for mass (constant term), in-room temperature can be seen in Table I.
β1: Slope of mass regression line,
ε: Error term
Table I
Descriptive statistics of measurements
std. error
range minimum maximum mean
of mean
Time (minute) 136.0 1.0 136.0 69.03 3.42
Top Left (°C) 284.1 3.2 387.3 163.72 9.13
Top Right (°C) 335.5 3.0 338.5 186.78 10.27
Middle Left (°C) 353.5 4.1 357.7 208.36 11.16
Middle Right (°C) 330.3 3.4 333.7 184.76 10.37
Bottom Left (°C) 275.1 3.6 278.7 134.71 9.12
Bottom Right (°C) 220.7 3.3 224.0 122.45 7.21
In-Room Temperature (°C) 944.0 17.0 961.0 720.09 25.70
RM models and temperature (°C) determines dependent way. In addition, when in-room temperature (°C) is predicted
variable In-Room Temperature (°C) and independent’s with multiple regression equation, determination coefficient
variables affecting this situation, Top Left, Top, Right, is found as (R2=0.973) (Table II-model summary). The best
Middle Left, Middle Right, Bottom Left, Bottom Right result in the regression models is seen in Table III. It
Temperatures (°C) and Time (minute) based on regression suggests that the values are meaning and compatible
equations were generated using seven parameters. The statistically. According to correlation values found, it is seen
regression equations generated by the SPSS software, the in the r column in Table II that there is a general strong
status of each variable were calculated correlation relation in the same direction between in-room temperature
coefficients obtained and determination. (°C)’s calculated with RM and in-room temperature (°C)
The highest correlation (R=0.986) is obtained by linear measured in the experiment (Table III).
multiple regression equation by using these seven variables.
And, the closest results to the experiment are obtained in this
Table II
Model Summary
Change Statistics
R (r) R Square Adjusted R Square Std. Error of the Estimate R Square Change F Change df1 df2 Sig. F Change
____________________________________________________________________________________________________________
432
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Table III
Linear effects of input parameters and in-room temperature (°C) predict equation with regression analysis
____________________________________________________________________________________________________________
433
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
VI. REFERENCES
____________________________________________________________________________________________________________
434
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
+
Rıdvan Saraçoğlu, Assoc.
Yüzüncü Yıl University Zeve Campus,
Engineering Faculty, Electric-Electronic Department,
ridvansaracoglu@yyu.edu.tr, Turkey
Abstract— Traffic accidents are one of the most important and of ambulances are examined, various causes listed below are
serious problems, our country encounters. As the percentage of observed [2].
injured and casualties in traffic accidents are huge in numbers, it
causes thousands of people injured or dead in a year. In the - Distance of the accident scene to the location of
situation of coming by a traffic accident that we face commonly ambulances
in daily life, generally people directly use their mobile phones - Ambulances receiving wrong directions
and call the related units. However, as the required, right and - Ambulance drivers not being able to locate accidents
adequate information is not given to these related units, the cases due to lack of knowledge of exact latitudes and
may sometimes end with casualties. One of the most important longitudes of accidents.
reasons of casualties in car accidents is the wrong, senseless first - Directing ambulances to false locations (pranks etc.)
aids performed to the victims of injuries by surrenders. In this - Deploying un-operational ambulances from faraway
study, the aim is to determine the exact location of the accident
centers to the accident scene, rather than nearest ones,
with its latitude and longitude that is determined automatically
via the application present in the responsive citizen who tries to
due to lack of communication
help the victims of the accident. - Inadequate number of ambulances when the accident
With this system named as “Mobile Accident Notification” occurred
the users are aimed to send accident notification in online or via - Traffic jams
sending message when there is no internet access with the help of - Ambulances not being able to pull up close to
mobile applications that process in Android Operating System. accident scenes due to irregularly parked cars on
These accident data kept in remote server are monitored on web streets and avenues.
pages or Android mobile phones. In this thesis, a web-based,
open-source accident filtration system was developed and the
The objective of this thesis is to avail responsive citizens to
accidents were provided to be monitored by the authorized
people in related units. notify accidents with an application downloaded into their
personal mobile phones and to directly locate the scene of the
accident on the map by automatically retrieving latitude and
Keywords— Mobile accident, Accident notification, Mobile
accident tracking, Android mobile accident statement longitude values of the informant. Additionally, by filling a
form included with the application, appropriate number of
ambulances and police will be diverted to the accident scene
Preventing traffic accidents is a shared concern of the since the exact number of people injured will be learned or
whole world. Since the ratio of deaths and injuries possibly life saving information will be given to healthcare
significantly high, thousands of people die and hundreds of professionals since the nature of the injuries will be identified.
thousands of people get injured in considerably high number Further objectives of this thesis are grouped under general
of accidents in each year. When an accident occurs in daily headlines below.
life, the common practice is to call relevant units and
institutions. However, since sometimes correct information is ACCIDENT NOTIFICATION
not given to relevant units, many lives are lost. Minutes after AND DETECTION SYSTEMS
the accident occurs (the time from the injury until the hospital
care) are crucial for resuscitating accident victims and Accident notification systems are usually the process of
reducing physical effects of the accident. Improving first aid calling the phone numbers of related units and informing
services is anticipated to reduce casualties up to 20% [1]. about an accident. With this notification method generally
Among the reasons of injuries in traffic accidents resulting in people either call the police and request other units to be
casualties and permanent disabilities are late arriving dispatched or call the emergency services to request an
ambulances and series of unapprised interventions of ambulance to be sent to the scene of the accident.
bystanders to accident victims. When reasons of late arrivals In case of accident detection systems, usually the detection
is realized by reading the data from sensors attached to
____________________________________________________________________________________________________________
435
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
vehicles. Another method is a study conducted by Amin et User prefers this method if there is internet
al[3] where data from GPSs of smart phones are read in 0.1 connection and when he/she does not want to enter
second intervals and the location of the vehicle is located by too much information. When the user clicks on fast
map matching algorithms and then the previous speed values notification (GPS must be active), a record of the
collected in 0.1 second intervals are subtracted from the accident is created in the database immediately.
subsequent speed values where the difference is then The other information on the form is updated with
compared to the threshold values to determine if an accident the data entered by the user. Since the location of
occurred or not. Additionally there are other solutions such as the accident and the location where the user filled
detecting accidents without interaction with Electronic the form may significantly differ, a record is
Control Units, false positive preventions and WreckWatch immediately created when fast notification is
client-server architecture. One final distinctive study is done clicked. For that reason the record is created
by placing noise and image recording cameras on junctions immediately after the user clicked on the fast
and start recording exactly when the accident occurs by notification; other information can be updated at a
recognizing the noises from the cars involved in an accident later time.
and by not recording continuously but only when the accident
took place [4].
- Detailed Notification
FILTERING NOTIFICATIONS
____________________________________________________________________________________________________________
436
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
different notifications by taking the estimated in the shortest possible time with the help of the support of a
distance of the informant to the accident scene. mobile application has been tried to be conveyed in this thesis.
With this study, automatically obtaining latitudes and
SENDING RELATED ACCIDENT MESSAGE TO longitudes of exact locations of user/users who have witnessed
RELEVANT UNIT WORKERS the accident through GPS, either via the internet connection or
when there are no internet connections, via SMS text
This is the process of sending messages according messaging method has been achieved. With numbers of
to the notifications of various informants to the closest people injured as well as other pertinent data entered in the
Police Headquarters, Health Institutions and Fire form, dispatching both adequate number of ambulances and
Department workers. diverting relevant police units to the accident scene have been
targeted. Depending on the nature of injuries, determining and
transferring probably life saving information to healthcare
M- ACCIDENT NOTIFICATION APPLICATION professionals has been proposed. At the present time when an
accident occurs, among reasons of late arrivals of relevant
With this application; it is targeted to create awareness units to the accident scene are; receiving wrong directions or
among end-users by involving latest technologies to intervene not being able to find the correct address. With this
accidents in a timely manner. This study came forth with the application, actions such as; direct determination of the exact
idea of ensuring fast and productive interference to accidents location of the accident on the map by taking the information
and with minimizing the response period during when of the closets users was targeted. In summary, with this thesis
accidents occur until first aid services reached to victims, is actions such as; contacting emergency services of nearest
developed by using many integrated web and mobile hospitals, contacting nearest Police Headquarters, totally
technologies altogether. eliminating the risk of giving wrong directions to ambulances,
dispatching ambulances to the correct location by extracting
In the thesis, two of them mobile, one of them web based, exact latitudes and longitudes of accident scenes on the map,
three different applications are developed. A single database is by collecting data from the informant such as his/her exact
used for every each application. The application is consisted latitude and longitude, number of vehicles involved, number
of three modules. These are; SMS Filtering module, of injured, and the nature of injuries, dispatching adequate
Notification Sending module and administrator follow-up number of ambulances with appropriate equipment, when tens,
module. Detailed explanations and screen shots from these sometimes hundreds of people inform an accident, giving only
modules will be given in the upcoming sections. the notification of the relevant accident will be taken. When
this mobile application is used, it is our belief that the time
CONCLUSION passed from the moment the accident occurs until the first
intervention to accident victims will be minimized, adequate
One of the major concerns of our country as well as the number of equipment and vehicles will be dispatched to the
world is traffic accidents. In the name of preventing these accident scene, assistance will arrive on the shortest time
accidents which result in hundreds of thousands of casualties possible by utilizing maps, closest relevant units will be
each year and cause many of them left with permanent dispatched to the location and many similar advantages will
disabilities, science people and researches have conducted be provided. No such study has been conducted ever before.
many widespread studies. In the light of these studies, they Therefore a study like this is believed to immensely contribute
have proposed several accident detection and accident to the literature as well.
tracking systems underlying how these accidents can be
prevented. In one of those studies, a three-step protection
method has been offered by Kraus et al[5]. The first of these REFERENCES
protection methods is the detection of causes of accidents and [1] K. Sangeetha, P. Archana, M. Ramya, and P. Ramya, “Automatic
taking necessary protections against them; the second method ambulance rescue with intelligent traffic light system, “, IOSR Journal
of Engineering(IOSRJEN), vol. 4, pp. 53-57, 2014.
is things that need to be done when an accident occurs and [2] Ch. R. Keerthi, G. Shanmukh, and R. Sivaram, “Various accident
measures that need to be taken to reduce severe effects of detection technologies and recovery systems with victim analysis,
accidents; and the final method is what needs to be done after “, International Journal of Advanced Trends in Computer Science and
accidents occur and necessary precautions to minimize effects Engineering (IJATCSE), vol. 2, pp. 07-12, 2013.
[3] MD. S. Amin, M. A. S. Bhuiyan, M. B. I. Reaz, and S. S. Nasir, “GPS
of injuries[5]. and map matching based vehicle accident detection system, “ IEEE
Student Conference on Research and Development (SCOReD),
Taking inspiration from what needs to be done when an Putrajaya, Malaysia, Dec. 2013.
accident occurs, as pointed out in the second method of the [4] P. J. Lagassey, “Advanced automobile accident detection, data
recordation and reporting system,“ U.S. Patent 7 348 895, 2008.
study by Kraus et al above, and many other similar studies, [5] J. Kraus, C. Peek-Asa, D. Vimalachandra, “Injury control: The Public
actions to be taken in order to minimize the time from the Health Approach, “ Maxcy- Rosenau-Last, Public Health&Preventive
moment the accident occurs until the first intervention to the Medicine, Wallace Robert B., MD, Msc, vol.2, pp. 1209-1210, 1998.
accident victim (the golden hour) and to treat accident victims
____________________________________________________________________________________________________________
437
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Determination of a blood type has a crucial importance determination of blood type is called Slide method in which
for blood transfusion. Therefore it is mandatory doing tests to three blood drops are dripped on separate lams then A, B, D
determine blood type before the transfusion. In order to prevent the antigens are dripped respectively on lams and blood type is
errors in determining blood type and to save time these tests are determined by angulation [3].
carried out by the automatic devices. However these devices are
In this study gel test method is used for determination of
very expensive and it is necessary to develop cheaper alternative
systems. In this study, we designed a basic device which will be a blood type which is widely used in Tukey. The gel card used
first step for a cheap and fast prototype. It utilizes the image in this method is shown in Fig 1. Gel card has a special gel
processing techniques and gel test method for real time blood and blood cells which don‟t agglutinate passes through this
type determination on embedded system. During the tests, fifty gel gel and accumulate at the bottom of the tuple [4].
test cards data were used and, it is found that the proposed system Looking at literature there are several studies for
can process each gel test card in 2 seconds with 99% accuracy on determination of blood type. Zarifi and co-workers used plate
average. method for detection of blood type on FPGA [5]. Feraz
developed a prototype system for emergency situations and
Keywords— Image Processing, Blood Types, Embedded Systems, used slide method in this system [6]. Fatima determined blood
OpenCv, Gel Test Method type using slide method microscope images by using image
processing technics [7]. Dolmashk and colleagues using laser
I. INTRODUCTION
light radiation and image processing techniques together have
Determination of blood type is a vital process before the shown that the detection of blood group is possible [8]. Swarp
blood transfusion. Otherwise fatal consequences may occur. and colleagues compared conventional tube method and gel
It must mandatory doing blood type determination tests to test method [9]. Toz and colleagues developed a software for
avoid this situation. International Society of Blood reading gel test cards by using image processing techniques
Transfusion (ISBT) had recognized a total number of 30 [10]. Also there are several devices for blood type
human blood group systems, with ABO and Rh as the two determination. Some of them are Technicon Auto Analyzer
most common groups [1]. The most important blood group is [11-13], Auto-Grouper [12], Technicon Auto Analyzer II
the ABO group. The ABO group system has A, B, AB and O [14,15].
blood types. Type A, B and AB have A, B and AB antigens
respectively on the surface of the red blood cells (RBCs), Gel test method needs three devices. These are gel test
whereas type O has none of these antigens [2]. The second centrifuge, gel test incubator and gel test reader. These
most significant blood-group system is the Rh system. The Rh devices are very expensive. In general, technicians in hospitals
system determines blood type from the presence or absence of are detecting blood type visually instead of gel test reader.
a „„Rhesus factor‟‟ surface on red blood cells. If Rh factor is Therefore errors in reading or interpreting cause hundreds of
present on surface of the RBC it is positive otherwise it is fatal blood transfusions over a year. Furthermore, analysing a
called negative. According to the ABO and Rh grouping large number of samples consumes a lot of time and requires
systems there are 8 blood groups. These are A Rh+, A Rh-, B special care for the sampler. In this study, our aim is to
Rh+, B Rh-, AB Rh+, AB Rh-, O Rh+, O Rh- . A, B, D (Rh) develop a cheap, fast and efficient system for gel test reader
antigens tests are carried out for determination of blood type device. For this purpose a system is designed on Beagle Bone
[3]. Black (BBB) which is a mini computer card which can run
Linux operating systems. It is useful, stronger and cheaper
Until recently different methods have been developed to alternative for image processing applications. Using an USB
determine the blood group types. Most of the techniques web cam gel card images were taken and image processing
applied today are based on the principle of interaction techniques are applied on BBB for detection of blood type.
between antigen and antibody. As a result of this interaction,
results are interpreted as positive if agglutination of RBCs
occurs and negative if otherwise. The first method used for
____________________________________________________________________________________________________________
438
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
439
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
formula given below. Fig 5 shows grey level gel card image
example.
Figure 9. Dilation
____________________________________________________________________________________________________________
440
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
second. This time is very efficient value for a real time system [2] G. Daniels, Human Blood Groups, 2nd ed. Blackwell Science, 2002.
[3] Malomgre W, and Neumeister B, “Recent and Future Trends in Blood
[5].
Group Typing”, Anal Bioanal Chem, 393, 1443-1451, (2009)
[4] Mujahid A, Dickert F.L, “Blood Group Typing: From Classical
Strategies to the Application of Synthetic Antibodies Generated by
Molecular Imprinting”, Sensors, 16-51, 2016
[5] Zarifi T, Malek M. , ”FPGA implementation of image processing
technique for blood samples characterization”. Computers and
Electrical Engineering (2013)
[6] Ferraza, A., Carvalhoa, V., and Soaresa, F., (2010). “Development of a
Human Blood Type Detection Automatic System”, Procedia
Engineering, 5, 496–499.
[7] Fathima, S. M. N., (2013). “Classification of Blood Types by
Microscope Color Images”, International Journal of Machine Learning
and Computing, 3, 376-379.
[8] Dolmashkin, A., Dubrovskii, V. A., and Zabenkov, I. V., ”Blood
Group Typing Based on Recording the Elastic Scattering of Laser
Radiation Using the Method of Digital Imaging”, Quantum
Electronics, 42, 409-416, (2012)..
[9] Swarup, D., Dhot, P. S., Kotwal, J., and Verma A. K, (2008).
“Comparative Study of Blood Cross Matching Using Conventional
Tube and Gel Method”, Medical Journal Armed Forces India, 64, 129-
130.
[10] Toz G, Erdogmus P, Donmez K, “A Software Desıgn For Blood
IV. CONCLUSIONS AND DISCUSSIONS
Groupıng By Usıng Gel Test Method”, Journal of Advanced
The Blood type determination test is one of the most Technology Sciences Vol 2, No 3, 100-109, 2013
important test performed in hospitals, blood banks, and other [11] Blood policy and technology, congress, office of technology
assessment. Washington, DC, USA; FAS
health organizations. Gel test method is one of the blood type <http://www.fas.org/ota/reports/8505.pdf>;January 1985.
determination methods which is widely used in Turkey. In this [12] Sturgeon P.,” Automation: its introduction to the field of blood group
study our aim is to develop a cheaper and efficient alternative serology”. Immunohematol J Blood Group Serol Educ 2001;17(4)
system for blood type determination. During the tests, fifty gel [13] W. A. Coakly, Handbook of Automated Analysis, Mercel Dekker,
1981 pp. 6l.
test cards data were used and, it is found that the proposed [14] Ewing GW, Analytical instrumentation handbook 2nd ed. New York:
system can process each gel test card in 2 seconds with 99% Marcel Dekker; 1997. p. 152
accuracy on average. [15] Autoanalyzer,<http://www.sealanalytical.com/Products/AutoAnalyzerI
I/tabid/108/language/en-US/Default.aspx> ,(Accessed 2016)
[16] Dia Pro Jel Kart, ,<http://www.diapro.com.tr/tr/urunlerimiz-across-
gel.html>, (Accessed 2016).
REFERENCES [17] <http://www.gokselmedical.com/WEB/DIAMED.doc>,
[1] Lögdberg L, Reid ME, Zelinski T. , ”Human blood group genes 2010: (Accessed,2016).
Chromosomal locations and cloning strategies” revisited. Transfus
Med Rev,25:36–46,2011.
____________________________________________________________________________________________________________
441
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Timber is used widely in construction industry lately mechanical properties of the materials used in structures. In
due to its some advantages such as neutrality, lightness, addition, the safety of element composed from timber are
environmentally-friendly, resistant against earthquakes, determined by whether it is strong enough to withstand against
manufacturing flexibility, and having a good compatibility with internal stresses which occurs as a result of external forces [1-
other construction materials like steel, concrete, and adobe-like
2].
materials. Fire retardant materials are used in order to protect
wood from the fire. The main task of the fire retardant material is Because of the many advantages of timber such as strength
to protect timber against fire, besides it is very important to know and versatile ease of use, it has been widely used as a
whether the retardant material has any effect on the compressive construction material for many years from the past till today [3].
strength of the material or not and it would help the designer to Timber is a natural material which is used in today’s
make a choice based on material sections. Today computer construction sector and it has been widely used in the past and
technology is widely used in the construction industry as well as in will continue to be used. Moreover, timber can be renewed in
all sectors. Various mathematical and statistical methods are used terms of planting new one and can be accepted as a non-
for computer-aided models. In this study samples with different consumable resource [4]. The properties like naturalness &
dimensions and different fire-retardant material has been
temperature, lightness, being an eco-friendly and, earthquake
prepared from most commonly used wood materials in the
construction sector. The compressive strength test has been resistant material, providing manufacturing flexibility,
applied on the samples after certain operations. The data obtained compliance with other construction materials such as steel,
from experiments have been used to apply regression analysis concrete and adobe, aesthetical and acoustic properties have a
(RA) which is a statistical approach and regression models (RM) huge role in the preference of timber, in spite of increasing
have been established to predict compressive strength (CS). Data acceleration in technological developments in the construction
like wood sample weight, cross-section, volume and breaking load sector in recent years [5]. Historical structures where wooden
were used to create regression equations for estimating the material usage providing good examples in many fields, can be
tension. The data obtained from the regression model developed encountered in both our country and many other countries.
by experimental data were statistically compared, it is seen that
Because necessary measures weren’t/aren’t taken in the
the outcome of the designed model is successful. When the
correlation coefficient between predicted compressive strength past/future when wooden material is used, lots of wooden
values and generated models was calculated, it has been seen that structures catch on fire they are completely destroyed or
the methods applied can be used safely. The data obtained from substantially damaged [6].
regression models is close to experimental data and such models With the aim of protecting wooden materials from fire, fire
can be used to estimate the compressive strength wood. retardant polishes, boron-based impregnated materials, fire
retardant paints, etc. are used. An experimental study was
Keywords— Regression analysis, Species of wood, Resistance of performed about whether coating materials which are fire
wood against pressure retardant at specific rates, decrease the strength of wooden
material or not. 3 different dimensions used with different aims
I. INTRODUCTION in construction affairs, 3 different types of wood which is easily
Since ancient times, trees and timber have been used as supplied in our country and 2 different coating materials which
construction materials. In that sense, the importance and are widely used and a raw material which is coated as proof bar
preservation of the forests is crucial. The fact that wood can be are used in experimental study [7-8].
used in different fields of life, is closely related to strength of In engineering studies, statistical approaches, regression
timber-type materials in the field where it is used. Therefore, it analysis, mathematical modelling, fuzzy-set-based technical,
is important to determine the applicability of timber, namely artificial neural nets and artificial intelligence approaches etc.
mechanical properties of it should be determined. In design can be widely used to estimate one or more than one parameter
stage, it is very crucial to determine the physical and and for different modellings. Regression analysis is a linear or
____________________________________________________________________________________________________________
442
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
non-linear statistical field, where it is used with the aim of being III. REGRESSION ANALYSIS
able to determine the cause-effect relation between two or more One of the most widely used area of utilization of the
variables that can be used to estimate or approximate about that statistical studies is regression analysis. In case there are more
subject. Therefore, one of the most suitable methods of the than one factor affecting a case, researching of cause-effect
process of determining wooden compressive strength is to use relation becomes possible with regression analysis. A
regression model. In this process, the fact that parameters can regression model consists of dependent and independent
be included in statistical model positively affects the accuracy variables. Independent variable is the variable which is not
of result [9]. coincidental and used to express dependent variable. Generally,
In this study, samples are prepared at different dimensions multiple regression analysis is used to find mean output values
and with different fire retardant from different types of wooden relating to dependent variable [9].
materials which are widely used in the sector of construction. Functional form of simple linear regression model can be
Tension stress tests are applied to these samples after specific written in the form of equation (1) for a mass [11].
processes. Regression analyses (RA), which are a statistical Y = β0 + β1X + ε (1)
approach by using data in these experimental studies, are made In this model;
and regression models (REM), which can estimate compressive Y: Value of dependent variable,
strength (CS), are composed. Results are compared considering X: Value of independent variable,
coherence between REM and experimental results. β0: Sectional of Y for mass (constant term),
β1: Slope of mass regression line,
II. EXPERIMENTAL STUDY
ε: Error term
For the experimental studies, primarily, the types of fire For it, the specifications of wood materials are created with
retardant materials which are most frequently available in the single and multiple regression equations of the system with
markets were analysed. For wood, an imported fire retardant SPSS program by statistical analysis method by taking
which are in conformity with UNI EN ISO 14001: 2004 (ISO advantage of data obtained experimentally [8]. The values of
14001: 2004) [10] standards and boron impregnated material in compressive strength (CS) are found and compared to
conformity with (ISO 14001, 2004) standards were preferred. experimental results by estimation Tension (kgf/cm2) with
Three types of wood were chosen to use in the experimental mathematical notation obtained with RM. The descriptive
studies. One of them is Scotch Pine (Pinus silvestris L.) with statistics 27 values for compressive strength can be seen in
the density of 620-780 kg/m³ frequently preferred in the Table I.
buildings, the other is fir tree (Abies sp.) which has the density In this study, regression equations are composed using these
of 430-520 kg/m³ and poplar tree (Populus sp.) with the density four parameters by getting dependent variable Tension (Tn-
of 310-400 kg/m³ were preferred. kgf/cm2) which determines compressive strength and
For pressure tests, a computerized full automatic testing independent variables, Test Material Weight (TMW-gr),
device with the capacity of 200 tons (2000 kN) which was Material Cross-Section (MCS-mm2), Material Volume (MV-
produced by Ankara originated Yüksel Kaya Machinery in mm3) and Breaking Load (BL-kgf) which affect this situation.
accordance with the standards of ISO9001(2000) Quality With the regression equations composed in the software SPSS,
Management System and Press 5.4 version prepared by the the situation of each of the variables is calculated with
company and the experiments were conducted. The pressing correlation and determination coefficients obtained.
machine and computer program which the experiments were The best result in the regression models is seen in Table II . The
conducted are seen in Fig. 1. highest correlation (R=0.926) is obtained by linear multiple
The prepared wooden samples were located cantering the regression equation by using these four variables. And, the
pressing machine a gradually increasing pressure stress (0.3 - closest results to the experiment are obtained in this way. In
N/s) was applied from bottom and top surfaces until the addition, when Tn is guessed with multiple regression equation,
samples were broken. The pressure values applied to the determination coefficient is found as (R2=0.858). It suggests
wooden samples were automatically given by the computer that the values are meaning and compatible statistically.
program. In the experiment, the averages of each 3 samples According to correlation values found, it is seen in the r column
were taken and the 27 results were obtained [8]. in Table II that there is a general strong relation in the same
direction between Tn’s calculated with RM and Tn measured
in the experiment.
Linear multiple regression equation obtained in this way is
expressed and the system is guessed (equation 2).
= 401.929 + (0.017 ∗ TMW) + (0 ∗ MCS) +
(−0.0000936983624273154 ∗ MV) +(0.006 ∗ BL) (2)
Mathematical formula can be created without a need for
MCS variable in RM developed like Table IIIMCS.
Fig. 1 The image of the pressing machine and the computer program used in
the experiments
____________________________________________________________________________________________________________
443
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Table I
Descriptive statistics of measurements
Table II
Linear effects of input parameters and Tn guess equation with regression analysis
Table III
Excluded Variables
____________________________________________________________________________________________________________
444
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Uluata A. R. (1987), Factors which affect mechanical specifications of
wooden materials, Journal of the Faculty of Agriculture, 18, 1, 113-124.
[2] Efe H., Çağatay K. (2011). Determination of Some Physical and
Mechanical Specifications of Various Massive Wooden Materials.
Politeknik Magazine, 14, 1, 55-61.
[3] Stevens R., Es van S.D., Bezemer R., Kranenbarg A., (2006), The
Structure Activity Relationship of Fire Retardant Phosphorus
Compounds in Wood, Polymer Degradation and Stability, vol. 91, 832-
841.
[4] Hunt D., (2012), Properties of wood in the conservation of historical
wooden artifacts, Journal of Cultural Heritage. vol. 13, 3, 10-15
[5] Guler S.B., (2011), Analysis of the protection and repair methods of
wood within the scope of its life cycle, İstanbul Technical University,
Graduate School Of Natural And Applied Sciences, Department of
Architecture, Master's thesis, İstanbul, 1-10.
[6] http://www.bianet.org/bianet/kent/157033-itfaiye-koskte-yangin-
cikmasi-icin-neden-yok
[7] Erol S.S., (2007), Effects of fire retardant dye primer (nanosön) on some
properties of wood and wood based materials, Süleyman Demirel
University, Graduate School Of Natural And Applied Sciences,
Department of Forest Engineering, Master's thesis, Isparta.
[8] Altin M., Tasdemir S., Pehlivan G., Erkis S., Altin S. The Experimental
and Statistical Analysis of the Wood Strength against Pressure
According to Different Wood Types, Sizes, and Coatings, 17th
International Conference on Materials Engineering and Technology
(ICMET 2015), 20602064 pp., Jeddah, Saudi Arabia, January 26-
27,2015.
[9] Taşdemir Ş., (2011). Determination and Comparison of Surface
Roughness of Artificial Neural Network and Regression Models, Journal
of Technical-Online, 10,3, 215-226.
[10] ISO 14001. (2004), Environmental management systems - Requirements
with guidance for use.
[11] Huang L., Chen J. C. A (2001). Multiple regression model to predict in-
process surface roughness in turning operation via accelerometer.
Journal of Industrial Technology, 17, 1-8.
____________________________________________________________________________________________________________
445
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Although wind energy at certain intervals and of radiation. While producing energy from the wind; first, the
random in nature, today it is one of the commonly utilized energy turns into kinetic energy, then into electric energy. The
alternative energy source in the world. Because of sustainability velocity of the wind is one of the most important parameters
and environmentally-friendly energy source, countries in the production of energy from the wind. For this reason, it
increasingly benefit from wind energy. Several estimation
is a must to measure both the velocity and potential of the
methods are applied in the determination of a region's wind
energy potential. Today, one of the most commonly used wind in a certain area before launching a wind power plant
prediction methods is artificial neural network (ANN) method. [2]. For this reason, the researchers developed methods to
In this study, Estimation of wind power in Osmaniye district was predict the wind potential in recent years. These prediction
investigated in method with artificial neural network (ANN) methods aim at measuring the wind power of a certain area or
using data from meteorological measurement stations from the predicting it by using certain inputs. As it is shown in (1) the
meteorological measurement device at the campus of Osmaniye power of the wind is stated as a mathematical function in
Korkut ATA University. In order to give the best values of which it equals the cube of the wind velocity [3].
prediction results, several methods increasing the impact on
output of different models for the input variables were 1
investigated. Power = ρAv3 (1)
2
Keywords— Wind Power, Prediction, Artificial Neuron Network Equation (1) according to the formula, it is seen that the
calculated or predicted value of the wind velocity grows
exponentially. Hence, the wind velocity acts as a defining
factor in power calculation more than other variables in the
I. INTRODUCTION formula. This velocity variant is taken into account in the
In recent years, the importance of the renewable energy detection of establishing wind power plant into a certain place
sources grows worldwide. The reasons of this growth are that and of the identification of their performance [3].
the sources named as fossil fuel are exhaustible, and they have When we have a look at the recent studies, it can be seen
negative effects on environment. Especially as a result of the that the methods used to predict the velocity and power of the
gradual increase in greenhouse gases like Carbon dioxide and wind are artificial intelligent based (ANN, Fuzzy Logic,
Methane which have direct impact on global warming, our Support Vector Machine) methods. The commonly used one
environment is affected negatively [1]. Although there are is the ANN method. The studied conducted by using this
some saving studies worldwide considering the issue of fossil method are indigenous to a definite area. A model can create
fuels in energy usage, these are not enough. Because the using the data chosen such as the latitude, longitude,
energy need of people and corporations increase gradually. To elevation, and average of wind velocity (yearly, monthly,
avoid from this deficiency, renewable energy sources are used daily, hourly), their minimum and maximum velocity values,
in recent years. Renewable sources are becoming one of the distance to the sea level, air pressure, density and temperature.
leading actors all around the world in the production of These chosen data are defined as input variants in the model.
electric energy. In a short time, they are foreseen to become These input variants are chosen by trying as many different
alternatives for power plants running fossil fuels, and they will methods as possible to do the best predictions. There are many
become more widespread. layers in these models. The input and output are the main
When the renewable energy types are considered, the first layers. There are one or more hidden layers between them.
ones to remember are wind and sun. They are also the energy There are a few neurons in each layer. When a model has been
sources on which the highest number of research and created, some of the data is used for education (learning)
scientific studied are conducted. The popular method to use to while some of it is used for test (prediction). The aim is to
produce electric energy by using solar energy is to make use predict the wind velocity and power belonging to a certain
____________________________________________________________________________________________________________
446
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
area as output. Besides, by making use of Box-Jenkins be paralleled, generalized, and as it is easy to analyze and
statistical model (ARIMA-Autoregressive Integrated Moving design. One of the most important features of it is its capacity
Average) which is used to predict seasonal or non-seasonal to derive as much information as possible without the need to
time periods, fuzzy logic methods (ANFİS-Adaptive any assistance. A mathematical model of an artificial neural
Network-Based Fuzzy Inference System), Genetic Algorithm network and its working principle are given in Fig. 2. This
based models; and by creating a hybrid or a different model structure is the process of imitating the human brain and the
with ANN, it can be used both while predicting the wind process of transferring learning to the computer based systems
potential and while comparing the predictions. [3-11]. [12].
In our study, the short term wind power prediction with
ANN is aimed using the data taken from the meteorological
station which was mounted at the Campus of University of
Osmaniye Korkut Ata. The data from the first half of 2013 is
used (between January and June). With the aim of evaluating
the prediction performances of a model created with ANN,
three different statistical indicators are used such as
coefficient of determination (R2), Mean Absolute Percentage
Error (MAPE) and Root Mean Square Error (RMSE). The
power plant potential of Osmaniye province for the first six
months has been estimated using these indicators.
Fig. 2 ANN Mathematical Model. (P) İnput Vectors, (W) Weight Matrix, (n)
II. MATERİALS AND METHODS Total Net Entry, (a) Output of Cell, (f) Activation Function.
X - Xmin
Xn =
Xmax - Xmin
____________________________________________________________________________________________________________
447
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
n
(P - Oi )2
RMSE = (4)
____________________________________________________________________________________________________________
448
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] H. Mituharu, and B. Kermanshahi, "Application of artificial neural
network for wind speed prediction and determination of wind power
generation output," Proceedings of ICEE, 2001.
Fig. 6 MSE Performance Graph [2] F. O. Hocaoğlu, M. Kurban, and Ü. B. Filik. "Wasp Yazılımı ile
Rüzgar Potansiyeli Analizi ve Uygulama, IV," Yenilenebilir Enerji
Kaynakları Sempozyumu, 2007.
Generally speaking when compared statistically, obtained [3] Y. Noorollahi, M. A. Jokar and A. Kalhor, "Using artificial neural
MATLAB results are approximately same. Comparison networks for temporal and spatial wind speed forecasting in Iran,"
between actual calculation of 24-hour power estimation Energy Conversion and Management, vol. 115, pp. 17-25, May. 2016.
measurements of an arbitrary day (11th) in June and power [4] M. Lei, L. Shiyan, J. Chuanwen, L. Hongling, and Z. Yan, “A review
on the forecasting of wind speed and generated power,” Renewable and
rating with estimation of ANN model is given in Fig.7.
Sustainable Energy Reviews, vol. 13, pp. 915-920, May. 2009.
[5] R. Velo, P. López, and F. Maseda, “Wind speed estimation using
multilayer perceptron,” Energy Conversion and Management, vol. 81,
pp. 1-9, May. 2014.
____________________________________________________________________________________________________________
449
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
450
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
nihan.kazak@bilecik.edu.tr
+
Department of Electrical and Electronics Engineering, Bilecik Şeyh Edebali University
Gulumbe Campus, Bilecik, Turkey
mehmet.koc@.bilecik.edu.tr
Abstract— In many texture recognition problems, Local Binary improved versions of original LBP. In [9], authors present
Patterns (LBP) method is used for feature extraction. This rotation invariant and uniform LBP which include
method is based on comparison of each centre pixel and its fundamental information about texture. In [10], multi-
neighbours’ intensity value in image. Due to its simplicity of dimensional LBP was proposed to using multi-dimensional
calculation, LBP has become one of the most popular feature
extraction techniques. In literature, different neighbourhood
histograms for different radii. Several versions of LBP were
topologies of LBP structure are given such as circle, square, studied in [8].
ellipse, parabola, hyperbola, and Archimedean spiral. This paper The basic LBP calculates histograms with circular
focuses on the use of uniform and basic LBP that have spiral neighbourhood topology. Using this topology is advantage for
topology in texture classification. We first derive basic and rotation invariant problem in the texture classification
uniform LBP features based on spiral topology. Then the application but anisotropic structural information may
performances of several classification methods such as linear important features for some problems. An elliptical
discriminant analysis (LDA), linear regression classifier (LRC), neighbourhood topology was used to exploit this anisotropic
support vector machines (SVM), Chi-square test, and G-test are structural information for a face recognition system in [11].
compared using these features in UIUC texture database.
Other variants of LBP are proposed by [12]. Different
shapes such as ellipse, circle, parabola, hyperbola and
Keywords— Local binary patterns, texture recognition, feature Archimedean spiral for the neighbourhood calculation are
extraction, classification methods, spiral topology.
tested. Different encodings to create LBP code are described
I. INTRODUCTION for the evaluation of the local grayscale difference. According
to this work, proposed quinary encoding and using elliptic
Texture can be defined as a repeating pattern of local
neighbourhood performs the best with the medical image
variations in image intensity. Texture contains information
databases.
about the spatial arrangement of the colours or intensities in
In the classification phase, performance of subspace based
an image. Texture classification basically means that
classification method is better than the statistical tests based
assignment of unknown image to one of known classes.
on similarity measures [13]. We employed Chi-square test, G-
Texture classification process consists of two main operations.
test, support vector machines (SVM), linear discriminant
The first operation is extracting features of the image and the
analysis (LDA) and linear regression classifier (LRC) for the
second one is classification phase.
classification. LRC is one of the subspace methods [14] that is
The best description of image is a very important step for
generally used for face recognition problems [15-17]. The
classification phase of texture classification. There are a lot of
experimental results of our study show that using LRC with
feature extraction methods in literature for the best description
SLBP features gives better results in texture recognition.
of image. In [1], Raju and Durai presents texture classification
The remainder of the paper is organized as follows. Section
techniques based on feature extraction in two categories
2 introduces the LBP operator. Our approach for computing
according the local and global features.
the LBP codes is explained in Section 3 and classification
Local Binary Pattern (LBP) is one of the methods which
methods are presented in Section 4. The experimental study
used local features. It is gray-scale invariant local descriptor
performed in this paper is given in Section 5 and finally
which was proposed by Ojala [2] and is a powerful descriptor
conclusions are presented in Section 6.
for texture recognition. The descriptor is used various fields of
image processing problems such as shape localization [3], II. LBP
interest region description [4], gender classification [5], palm-
LBP proposed by Ojala et al. is one of the most famous
print recognition [6], digital image stabilization [7] and
texture descriptor. LBP operator is the grayscale invariant
especially face recognition problems [8]. There are a lot of
____________________________________________________________________________________________________________
451
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
texture measure. This operator works with the gray value of bin is allocated for the all non uniform patterns so histogram
image pixels. consists of 59 bins.
For the each pixels of an image which is defined as centre
pixel, binary code is generated by comparing it and its III. CLASSIFIERS
neighbours. If the difference between compared neighbouring
pixel value and centre pixel value is higher or equal to 0, A. G-test
compared neighbouring pixel value is coded as a binary 1, in G-test is defined as the following equation:
the other case it is coded 0. Fig. 1 shows an example of LBP 𝑆𝑆𝑏𝑏
method. 𝐺𝐺(𝑆𝑆, 𝑀𝑀) = 2 ∑𝐵𝐵𝑏𝑏=1 𝑆𝑆𝑏𝑏 𝑙𝑙𝑙𝑙𝑙𝑙 (2)
𝑀𝑀𝑏𝑏
____________________________________________________________________________________________________________
452
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
𝑗𝑗
where 𝑥𝑥𝑖𝑖 represents the jth feature vector of the ith class. If y
belongs to the ith class, it should be represented as a linear
combination of the training images from the same class and
can be defined as
𝑦𝑦 = 𝑊𝑊𝑖𝑖 𝛽𝛽𝑖𝑖 + 𝜀𝜀 (7)
where 𝛽𝛽𝑖𝑖 is the vector of parameters that is estimated using
least squares estimation. Hence the estimation of 𝛽𝛽𝑖𝑖 is
𝛽𝛽�𝚤𝚤 = (𝑊𝑊𝑖𝑖𝑇𝑇 𝑊𝑊𝑖𝑖 )−1 𝑊𝑊𝑖𝑖𝑇𝑇 𝑦𝑦 (8)
The estimation of the vector y is calculated as follow:
𝑦𝑦�𝚤𝚤 = 𝑊𝑊𝑖𝑖 (𝑊𝑊𝑖𝑖𝑇𝑇 𝑊𝑊𝑖𝑖 )−1 𝑊𝑊𝑖𝑖𝑇𝑇 𝑦𝑦 (9) Fig. 3 Spiral neighbourhood topology
____________________________________________________________________________________________________________
453
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Performance of proposed SLBP is higher than the other [2] T. Ojala, M. Pietikainen ve D. Harwood, A Comparative Study of
feature extraction methods for all classification methods. The Texture Measures with Classification Based on Feature Distributions, Pattern
Recognition, vol. 29, no. 1, pp. 51-59, 1996.
statistical tests based on similarity measures perform worse [3] X. Huang, S. Li ve Y. Wang, Shape localization based on statistical
from subspace based methods. LRC is the best classification method using extended local binary pattern, Proc.Int.Conf.Image Graph.,
method in UIUC texture database as a result of experiments. 2004.
[4] M. Heikkila, M. Pietikainen ve C. Schmid, Decription of interest
TABLE I regions with local binary patterns, Pattern Recognition, vol. 42, pp. 425-436,
CLASSIFICATION ACCURACIES (%) 2009.
LBP ULBP SLBP SULBP [5] S. C., Learning local binary patterns for gender classification on real-
world face images, Pattern Recognition Letters, vol. 33, pp. 431-437, 2012.
%33.40 %33.00 %33.72 %31.04 [6] M. G. K. Ong, T. Connie ve T. A. B. Jin, Touch-less palm print
G-test
±1.98 ±1.92 ±2.73 ±3.00 biometrics: Novel design and implementation, Image and Vision Computing,
%63.24 %61.76 %63.88 %60.76 vol. 26, pp. 1551-1560, 2008.
Chi-square [7] B. Kir, M. Kurt ve O. Urhan, Local Binary Pattern Based Fast Digital
±2.35 ±3.01 ±2.81 ±2.40
Image Stabilization, IEEE Signal Processing Letters, vol. 22, pp. 341-345,
%63.24 %74.2 %80.84 %69.52 2015.
SVM
±2.35 ±3.87 ±2.55 ±3.59 [8] S.-M. Huang ve J.-F. Yang, Linear Discriminant Regression
%79.96 %80.16 %83.32 %83.72 Classification for Face Recognition, IEEE Signal Processing Letters, vol. 20,
LDA no. 1, pp. 91-94, 2013.
±3.50 ±3.20 ±2.60 ±3.04
[9] T. Ojala, M. Pietikainen ve T. Maenpaa, Multiresolution Gray-Scale
%84.44 %83.28 %85.64 %84.36
LRC and Rotation Invariant Texture Classification with Local Binary Patterns,
±2.34 ±1.45 ±1.91 ±2.19
IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 24, no. 7, pp.
971-987, 2002.
VI. CONCLUSIONS [10] G. Schaefer ve N. P. Doshi, Multi-dimensional Local Binary Pattern
Descriptors for Improved Texture Analysis, 21st International Conference on
In this paper, we have presented uniform and basic Local Pattern Recognition (ICPR 2012), Tsukuba, Japan, 2012.
Binary Pattern (LBP) with Archimedean spiral neighbourhood [11] S. Liao ve A. Chung, Face recognition by using enlongated local binary
topology to texture recognition. Firstly, we derived basic and patterns with average maximum distance gradient magnitude, Computer
uniform LBP features based on spiral topology. Then the Vision – ACCV 2007. Lecture Notes in Computer Science, vol. 4844, Berlin,
Springer, 2007, p. 672–679.
performances of several classification methods such as linear [12] L. Nanni, A. Lumini ve S. Brahnam, Local binary patterns variants as
discriminant analysis (LDA), linear regression classifier texture descriptors for medical image analysis, Artificial Intelligence in
(LRC), support vector machines (SVM), Chi-square test, and Medicine, vol. 49, pp. 117-125, 2010.
G-test are compared using these features. Experimental results [13] N. Kazak, M. Koç, B. Benligiray ve C. Topal, A comparison of
classification methods for local binary patterns, 24th Signal Processing and
obtained from UIUC texture database clearly demonstrate that Communication Application Conference (SIU), Zonguldak, 2016.
the best recognition rate is obtained with using SLBP features [14] I. Naseem, R. Togneri ve M. Bennamoun, Linear Regression for Face
and LRC method. In the future, we plan to focus on improving Recognition, IEEE Trans. Pattern Analysis and Machine Intelligence, pp.
spiral topology to obtain more meaningful information. 2106-2112, 2010.
[15] I. Naseem, R. Togneri ve M. Bennamoun, Robust regression for face
recognition, Pattern Recognition, vol. 45, pp. 104-118, 2012.
REFERENCES [16] M. Koc ve A. Barkana, Application of Linear Regression Classification
[1] J. Raju ve C. Durai, A Survey on Texture Classification Techniques, to Low Dimensional Datasets, Neurocomputing, vol. 131, pp. 331-335, 2014.
Information Communication and Embedded Systems (ICICES), Chennai, [17] A. Suruliandi, Local binary pattern and its derivatives for face
2013. recognition, Computer Vision, IET, vol. 6, pp. 480-488, 2012.
____________________________________________________________________________________________________________
454
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
455
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
product transport systems to be overloaded, the time for the abovementioned program has been given in this study and
preparation to increase and product queues to be longer [5]. it has been aimed to shed light on many problems’ solutions.
Therefore, the solution to PLP is a strategical stage for all
II. PLANT LAYOUT WITH VIP-PLAN OPTIMIZATION
enteprises.
PROGRAM
Generally, two different approaches are being used to
VIP-PLANOPT is an important solution instrument for
solve PLP. One of them is the qualitative approach which tries
plant designing problems in engineering, industrial planning
to maximize the total amount of the proximity scores among
and plant designing. VIP-PLANOPT (Visually Interfaced
the sectors. Systematic Layout Planning procedure developed
Package of PLANOPT) is a package software program
by Muther [6] is one of the most important studies on this
developed to produce optimum layouts for blocked or modular
issue. The second one is the quantitative approach which tries
problems that have small, medium or large scaled unequal
to minimize the total amount of the product transport costs
rectengular dimensions. The term “optimum layouts” means
among sectors. The study made by Armour and Buffa [7] can
locating the certain number of modules in the Euclidean space
be shown as a good example fort his approach.
in the best way without any extensions out.
Many studies in the literature has used the qualitative and
A. Optimization Algorithm
the quantitative approaches seperately so as to solve the PLP.
However, many researchers questioned whether it is logical to The PlanOPT optimization algorithm is a research product.
consider only one criterion to solve the PLP. Both of the It is a robust, registered and hybrid algorithm. The Algorithm
approaches have some advantages and disadvantages [8]. occured as a result of the development of a technic that was
However, those models which are based on only one criterion producing better results and developed by two university
is not explanatory, because there are too many factors to affect professors to solve the KAP in unequal sizes in 1995.
the plant layout. In fact, PLP should include both the VIP PlanOPT 2006 has reinforced its superiority over the
qualitative and the quantitative criteria. For this reason, it is in other algorithms by making the following additions to this
the category of multi- functional plant planning projection algotihm.
problem. [9].
In order to bring the variability of the algorithm observed
According to Malakooti, the aim of the multi- functional in some circumstances, double- precision arithmetic has been
plant layout planning is to produce efficient layout plan added.
choices to be given to the decision maker. Thus, the decision
maker can prefer the best layout plan option by considering In order to decrease the dependance on the
the conflictive aims. There are three different methods to starting point determined by the user, new technics
solve this problem. One of the is to find out the proactive has been added.
cluster of the layout plan options and to submit them to the In order to provide the optimal layout within
decision maker. The second one is that the preferences of the boundries with complicated and mixed shapes and
decision maker are found out first and then the optimum simple rectengular shapes determined by the user,
layout plan option is indicated according to these preferences. punishment functions has been added.
The third one is to use an interactive way to find out the The user interface screen of the program is shown in
optimum layout plan option. [10]. Attach-1.
In order to solve the multi- functional plant layout B. Terms Used in the Program
planning problem, many studies have been carried out. In
many of these studies, predetermined loads has been given to Modul: The term “modul” used fort he rectengular blocks
each goal and combined and only one goal has been acquired stands for the functional units like sectors, machines, rooms
in order to indicate the optimum layout plan option. The and areas. The size of a module along with the x line stands
quadratic assignment problem formulation has been used to for its lenght and the size along with the y line stands for its
model the multi- functional plant layout planning problem. width, and they are shown with the symbols Li and Wi.
In this study, an optimal solution has been targeted by The Modul width-lenght rate: This rate is described as
using VIP-Plant Optimization program for plant layout the rate of the width of a module to its lenght. If the width-
problem. Different from the methods mentioned in the lenght rate for a i modul is shown as Ri, it can be shown as the
literature, the convenient plant layout has been found out by followig equality:
using a package program. In addition, some information about
____________________________________________________________________________________________________________
456
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
457
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSION AND EVALUATION [5] Jajodia, S., Minis, I., Harhalakis, G. and Proth, J.M.
CLASS: computerized layout solutions using simulated
It has been seen in the literature research and
annealing, International Journal of Production Research, 1992,
implementation that the optimum layout for the factory has
Cilt 30, No 1, 95–108.
been found with the method used, but the important points are
that the solution should be convenient for the operation of the [6] Muther, R., Systematic Layout Planning (SLP), 2nd
factory and the real life. In other words, it should be feasible. Edition, Cahners Books, Boston, 1974.
In this study, studies about the plant layout planning problem
[7] Armour, G.C. and Buffa, E.S. A heuristic algorithm
which is one of the most important and basic issues in
and simulation approach to relative allocation of facilities,
production systems has been reviewed and then an optimum
Management Science, 1963, Cilt 9, 294–309.
layout plan has been acquired through real data taken from a
real production system by using VIP-PLANOPT software. In [8] Houshyar, A. Computer aided facilities layout: an
the implementation stage, 8 shelves with different features and interactive multi-goal approach, Computers and Industrial
the relations with each other has been indicated and the Engineering, 1991, Cilt 20, No 2, 177- 186.
convenient layout plan has been emerged. [9] Chen, C.-W. and Sha, D.Y. Heuristic approach for
REFERENCES solving the multi-objective facility layout problem,
International Journal of Production Research, 2005, Cilt 43,
[1] Simchi-Levi, D., Kaminsky, P. ve Simchi-Levi, E.
No 21, 4493-4507.
Designing and Managing the Supply Chain: Concepts,
Strategies, and Case Studies. McGraw- Hill/Irwin, 2003, sayfa [10] Malakooti, B., “Multiple objective facility layout: a
133-136. heuristic to generate efficient alternatives”, International
Journal of Production Research, 1989, Cilt 27, No 7, 1225–
[2] Ertay, T., Ruan, D., Tuzyaka, U. R. Integrated Data
1238.
Envelopment Analysis and Analytic Hierarchy for the facility
layout design in manufacturing systems, Information
Sciences, 2006, 176, 237-262. ss.
[3] Baysakoğlu, A., Dereli, T., Sabuncu, I. An ant colony
algorithm for solving budget constraint and unconstraint
dynamic facility layout problems, Omega, 2006, 34, 385-396.
ss.
[4] El-Baz, M.A. A genetic algorithm for facility layout
problems of different manufacturing environments,
Computers & Industrial Engineering, 2006, 47, 233-146, ss.
____________________________________________________________________________________________________________
458
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
ATTACHMENTS
____________________________________________________________________________________________________________
459
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
460
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
461
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
462
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— One of the widespread image processing applications algorithms, a fitness function is used to define the problem. In
is image filtering with two dimensional convolution. Determining the present case, fitness functions is formed according to mean
the weights of image filters are of importance for the success of squared value of original and noisy images. For the
filtering operation. Heuristic algorithms such as genetic computation of fitness function, intense multiplication and
algorithms provide an efficient way of training these types of addition operations are carried out to obtain fitness value.
filters. Due to the high computational cost of repetitive image Furthermore, computation time depends on the number of
filtering operations, this process may take hours to implement weights as well as the image size. During computations,
using single core computing. OpenMP (Open Multi Processing)
fitness function is called at each iteration of the genetic
provides an efficient library for utilizing the computing power of
multicore processors. In this study, OpenMP accelerated algorithm. This significantly slow down the process and make
training of separable filters that are a subclass of convolution the applications impractical. A method for the acceleration of
filters has been implemented based on genetic algorithms. the process is to utilize the computational power of multicore
Comparative speed-up results for various sizes of images using processor. For this purpose, OpenMP provides a useful tool
various sizes of filtering kernels were presented. Also the effect of for efficient use of the cores of a multicore processor.
population size of genetic algorithm and the number of working OpenMP helps distribute the computational load to defined
cores have been investigated. number of threads.
In the present study, OpenMP is utilized to accelerate the
Keywords— OpenMP, separable filters, image processing, computation of fitness function. Fitness function is computed
genetic algorithms for all individuals in the population and these operation can be
realized independently on processor cores. In the experiments,
an eight core computer is used and the results are obtained
I. INTRODUCTION against the number of cores to see the effect of the number of
Image filters are widespread operators in image processing cores. Also various filter kernel sizes, image sizes and the
applications such as image enhancement, image smoothing, population sizes used in the experiments to show the
edge detection and noise elimination [1]. Linear filtering using efficiency of the OpenMP based acceleration.
two dimensional convolution or correlation is one of the main
filtering operations. This is realized by applying the filtering II. SEPARABLE IMAGE FILTERS
kernel to each pixel of input image where the kernel is a Separable image filters are used in a slightly different way
matrix of weights. The size of the matrix can be 3×3 or 5×5 from the non-separable filters. An example of non-separable
larger such as 21×21. If the kernel has symmetric properties, it image filter is shown by Fig. 1 which has size 3×3. The filter
can be expressed as the multiplication of a row and column kernel has a total of 9 weights. This means that the image
vectors. This form is called as separable filter and it reduces filtering process train weights number of genetic algorithms is
the number of multiplication/addition operations. 9. When the size of the filter kernel grows, it is increasing
The values of the filter weights are determined according to training time of genetic algorithms. For instance, 25 weights
the desired behaviour of the filter. Weights can readily be for the filter kernel with 5×5, 49 weights for the filter kernel
obtained using different analytical techniques [2], [3]. In with 7×7, 81 weights for the filter kernel with 9×9, etc. The
another approach, the kernel weights can be trained using the growth of genetic algorithms filter kernel increases the
original and noisy image samples [4]–[7]. Heuristic training time. The 3×3 filter kernel used in separable image
algorithms provide an efficient way for the computation of the filter is shown in Fig 1. This filter kernel which horizontal
filter kernel weights [8]–[10]. One of the well-known heuristic and vertical vectors as shown in Fig. 2 in the separable image
algorithms is Genetic algorithm which provided its efficiency filter is used.
in various researches. Genetic algorithm is selected to train the
weights of the separable filter. In the application of genetic
____________________________________________________________________________________________________________
463
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Complier Environment
1 2 1 1 * 1 2 1 Directives Variables
2 4 2 = 2 Runtime Library
1 2 1 1
Threads in Operating System
Fig. 2 Separable image filter is used filter kernel
Separable image filtering process can be divided into two OpenMP Architecture is shown
Fig. 3 OpenMP figure 4.
architecture OpenMP, compiler
[16].
stages. The first stage is to filter noisy image using one of the
vectors. The second stage is to filter the resulting image from OpenMP Architecture is shown figure 4. OpenMP,
the first stage using the other vector. Therefore, image compiler directives, runtime library and environment variables
filtering process is completed in two stages. Separable image are comprised from. Programmers write the code to run
filter has the advantage of reduced number of weights over concurrently by putting special comments in that codes.For
non-separable filter. For example, the number of filter weights instance “#pragma omp parallel”. This study was
to be trained in separable image filter for 5×5 is 10, while it is parallelization of the fitness value calculating for “for” block.
25 for non-separable image filter.
Process 0
Thread 2 Thread 5
Initial population
Fitness
While (termination criteria is not met) Fig. 4 OpenMP hierarchy of 2D array [15].
Selection
Crossover
Mutation OpenMP’s operation diagram is shown in Fig 5. OpenMP is
Fitness identified one of the threads as main thread. Tasks are
End while distributed in equal amounts other threads by the main thread.
Due to the fact that this study was developed with C
Fig. 3 Genetic algorithm steps programming language, to use the OpenMP function
“#include <omp.h>” as is included in the project.
Genetic algorithm involves applying selection, crossover, Fig. 6 shows the area of the genetic algorithms parallelization
mutation and Fitness calculations on candidate population process on the flow chart. This block is calculated fitness
which are initially formed randomly. A pseudo code function value. This block contains computationally intensive
illustrating the operation of genetic algorithms is shown in Fig. mathematical operations. Therefore parallelization process is
3. In the present case, computationally most intensive part is performed here.
calculating the value of the fitness function due to the image
filtering operations.
____________________________________________________________________________________________________________
464
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE 2
V. EXPERIMENTAL RESULTS AND DISCUSSION
COMPUTATION TIMES FOR 200 POPULATION AND 3X3 FILTER KERNEL.
Experimental studies on Windows Server 2012
Essentials™ 64-bit operating system, Quad-Core AMD 200 Population, 3x3 Filter Kernel
Opteron™ 2378 2.40GHz dual processor, 18GB Ram, have Core Numbers (TIME (Minute))
been working on computer servers. The algorithm is written in Image Size
1 Core 2 Core 4 Core 6 Core 8 Core
C programming language. In the experiments, 256×256,
256×256 6.74 3.20 1.70 1.58 0.94
512×512 and 1024×1024 with a pixel size images are used
[15]. 3×3, 5×5 and 7×7 sizes filter masks are used for images 512×512 25.93 14.60 8.10 6.20 5.59
filtering process. The developed algorithm was running 10 1024×1024 107.02 55.85 34.34 24.66 23.07
times for each image and at the end of working, these 10
average MSE (Mean Squared Error) value and average TABLE 3
training time are taken. The number of population for
analysing the impact of population on training time while 100 COMPUTATION TIMES FOR 100 POPULATION AND 5X5 FILTER KERNEL.
and 200 were obtained results determined separately. The 100 Population, 5×5 Filter Kernel
number of iterations has been fixed at 400 for all calculations.
Core Numbers (TIME (Minute))
Mutation rate 0.005 and crossover rate 0.3 is defined as the Image Size
constant. Termination criteria of genetic algorithms are 1 Core 2 Core 4 Core 6 Core 8 Core
defined as the number of iterations. Noisy image used were 256×256 3.97 2.02 1.13 1.00 0.62
obtained by adding Gaussian noise on the original image. 512×512 16.16 8.79 4.86 3.61 3.03
1024×1024 68.75 34.85 19.67 14.03 12.42
TABLE 1
COMPUTATION TIMES FOR 100 POPULATION AND 3X3 FILTER KERNEL. TABLE 4
100 Population, 3×3 Filter Kernel COMPUTATION TIMES FOR 200 POPULATION AND 5X5 FILTER KERNEL.
____________________________________________________________________________________________________________
465
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE 5 Table 1 and Table 2 shows the computational times for 3×3
window using 100 and 200 populations respectively. Table 3
COMPUTATION TIMES FOR 100 POPULATION AND 7X7 FILTER KERNEL.
and Table 4 shows the computational times for 5×5 window
100 Population, 7×7 Filter Kernel using 100 and 200 populations respectively. Table 5 and Table
Core Numbers (TIME (Minute)) 6 shows the computational times for 7×7 window using 100
Image Size and 200 populations respectively. All results show that as the
1 Core 2 Core 4 Core 6 Core 8 Core
number of cores increased, the computational times reduces
256×256 4.03 2.29 1.27 1.14 0.69 significantly. Figure 8a to 8f show the graphical comparison
512×512 19.27 9.65 5.68 3.86 3.00 of the results. Best acceleration rates are obtained for 256×256
1024×1024 77.81 38.79 21.85 15.63 12,29 image.
TABLE 6
COMPUTATION TIMES FOR 200 POPULATION AND 7X7 FILTER KERNEL.
(a) (b)
(c) (d)
____________________________________________________________________________________________________________
466
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(e) (f)
REFERENCES
____________________________________________________________________________________________________________
467
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—The number of devices connected to the Internet is With the rising of popularity of the IoT concept recently,
increasing, data centers are growing continuously and computer the number of devices connected to the internet increases
networks are getting more complex. Traditional network every day. IoT offers an environment that interact the objects
management approach is becoming more difficult and we use in daily life with the other objects. All of devices that
insufficient. Software-Defined Networks (SDN) is a new support networking technology such as computers, smart
generation networking approach which is expected to take place phones, tablets, air conditioners, refrigerators, cars etc.
of the traditional computer networks. SDN architecture provides continuously produce data and this data is growing each day.
effective management of the large and complex networks.
As a result, big data concept emerges that represents high-
Although SDN have benefits from the network security
perspective, it also brings new attack vectors. We believe that the volume, complex and irregular data. Big data can not be
network security problems in SDN architecture need more processed, stored and managed by traditional methods. Big
advanced solutions. In this work, a survey on the SDN security data which becomes more valuable requires meaningful
problems is presented, challenges are discussed. In this context, results. So, big data should be processed. Big data needs more
security threats and attack surfaces in SDN are described, the bandwidth for processing. Nowadays, smart cities and smart
significant SDN security solution examples in the literature are management concepts come to the fore more and SDN
given. management, security and optimization topics will need more
advanced mechanisms.
Keywords—Software-Defined Networks, SDN, SDN Security
In the second section of this study, the basic concepts of
I. INTRODUCTION SDN, the benefits of SDN architecture are explained. In the
There are many network devices in computer networks third section, security threats and attack surfaces in SDN are
such as router, switch, firewall, Intrusion Detection System described. In the fourth section, significant SDN security
(IDS) and Intrusion Prevention System (IPS). Complex and solution examples in the literature are given. In the last section,
different protocols are running on these devices. Every day, this study is summarized and future works are presented.
new technologies such as Internet of Things (IoT), smart cities
and smart management are emerging, data centers are growing II. SDN & BENEFITS
and the number of devices in computer networks is increasing. There are three planes/layers including application, control
Computer networks are getting more complex and and data and two interfaces including application-control and
heterogeneous and management of the network is becoming control-data. Control plane decides where frames/packets will
more difficult. Traditional network management approach is be forwarded/routed. The data plane forwards the traffic to the
insufficient in large-scale computer networks. There is a need destination. Routers and switches that we used in today,
for a better network management approach and new methods. includes control plane and data plane are integrated on the
Software-Defined Networks (SDN) is an emerging concept same hardware. SDN concept is based on the idea of the
that bring a new generation network management approach separation of these planes. Control plane in other words
which is expected to take place of the traditional computer network intelligence is moved to a high performance server
networks. SDN promise administrative convenience, and network management is performed with centralized
hardware-independent, dynamic, scalable and flexible controller software. The data plane is left on OpenFlow-
networking architecture. SDN provide a centralized network enabled router or switch and is responsible for forwarding of
management and a global perspective on the network. So, this packets only. SDN architecture is shown in Figure 1. This
enables effective management of the large and complex architecture provides ability to directly programming the
networks. network and enables underlying infrastructure layer to be
abstracted for network services and applications [1]. So this
provides more dynamic, flexible, scalable platform and easy
____________________________________________________________________________________________________________
468
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
management of the network compared to traditional network target the controller software, communication between the
infrastructures. control plane and the data plane (control-data interface) and
the communication between the control plane and the
application plane (controller-application interface). All those
threat vectors have a potential effect on the operation of the
entire network. Attacks against SDN planes and interfaces and
the targeted security services are given in Table I.
____________________________________________________________________________________________________________
469
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Controller hijacking or unauthorized controller access model, allows eavesdropping or modifying the traffic flow
attacks target the confidentiality principle of the security. between network resources such as server, router or switch
Vulnerabilities in the controller may have consequences and endpoint on the network. In this case, the attacker may
which can put the entire network at danger [3]. The attacker modify flows on the switches or be able to add new flow rules
can take over the management of misconfigured, vulnerable [4]. Communication channel can be made more secure by the
controller and also the management of the network. Then, the use of TLS which is a cryptographic protocol [6]. OpenFlow
attacker can programme switches in the data plane to drop the protocol supports TLS connection by default. Mutual
traffic coming to the controller can use to launch attacks on authentication can be done by exchanging certificates between
other targets [5]. controllers and switches which are responsible for the
transmission of network packets. The attacker cannot view or
In the absence of mechanisms to ensure security in the modify the contents of messages when encrypted protocols are
communication between the control plane and the application used. However, TLS/SSL and Public Key Infrastructure (PKI)
plane, malicious applications can insert fraudulent rules into has some weaknesses and these vulnerabilities could be
the flow table of switches. This will cause the conflict rules in exploited [7]. If an attacker can access control plane by
the network. Therefore, reliable connection must be benefiting the protocol vulnerabilities in communication,
established by creating authorization and authentication switches under its control can be used to launch DDoS attacks
mechanisms between controllers and applications [4]. [3].
Man-in-the-middle (MITM) attack, which occurs between Unauthorized and unauthenticated applications target
control plane and the data plane communication targets confidentiality and integrity principle of the security. There
confidentiality and integrity principle of security. Both the are many third-party applications which run at the application
control plane and data plane will be affected by this attack layer. Controller provides abstraction for the SDN
type. The data modification between the control plane and the applications and this enables the applications to read and write
data plane is one of the most important problems in SDN. network state [4]. This situation poses a problem for the
SDN architecture brings discrete planes and using control of the network. The attacker can use the applications
unencrypted protocols in the communication between these that cause unauthorized access to hide himself and access
discrete planes can cause serious consequences. MITM attack, network resources and manipulate the operation of the
which performs in the second layer of the OSI reference network [4].
TABLE I
SUMMARY OF ATTACKS AGAINST SDN
Control Plane
Interception Controller Hijacking, Confidentiality
Unauthorized Controller Access
____________________________________________________________________________________________________________
470
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Administrative computers that are directly connected to the Connection Migration module decreases the interaction
controller can cause an entry point into the network. If these between control plane and data plane. The Actuating Trigger
computers have some vulnerability the attacker can use these module provides the installation of the necessary flow rules. It
vulnerabilities to get control of the computers to access the is stated that this plug-in is also effective against network
controller easily. scanning attacks but does not provide any protection against
DoS attacks in application-level and UDP or ICMP protocols-
Additionally, not quickly detecting an error or a problem in based attacks. It is expressed that after the attack is detected,
the network and a failure to obtain a reliable recovery point of control plane should be able to respond quickly. Therefore,
the network is considered as a threat vector [3]. quick access to the statistics belonging network traffic from
data plane is of great importance. Within the scope of study,
IV. SDN SECURITY SOLUTIONS the statistics are collected from data plane and sent to the
There are some comprehensive surveys [4], [8]-[11] on control plane. Accordingly, the behaviors detected as attack
SDN and SDN security. In these papers, concept, architecture, are prevented.
core components, advantages, current challenges of SDN and
SDN specific security threats and solutions are discussed in IDS and Anomaly Detection System (ADS) are used for the
detail. purpose of providing protection against threats in traditional
network infrastructures. These security systems are generally
Despite the advantages provided by SDN architecture, it is located in Internet Service Providers (ISP) or backbone
necessary to review network security issues. In this context, devices. This approach changes in SDN. These systems can be
there have been some studies carried out in literature. In these brought to the endpoints with SDN. In a study [13], ADS is
studies, prevention and mitigation of DoS and DDoS attacks proposed for home and Small Office/Home Office (SOHO)
are mainly considered. Furthermore there are also the studies networks using OpenFlow. Accordingly, an application which
on authentication and authorization mechanisms, development runs in NOX [14] controller is developed and implementation
of network security applications such as IDS/IPS and firewall. of four anomaly detection algorithms are showed. This
In this section, some of these studies on SDN security are solution offers more efficient anomaly detection in home and
described. SOHO networks than the ISP.
Some security measures are offered in [3] where also SDN- In another study [15], usage of central controller is
specific threat vectors are discussed. In case if only one proposed for the detection of DDoS attacks. Random
controller is used and the controller is collapsed, there will not distribution of incoming packets to the network is calculated
be a fault tolerance of the network and whole network may in this method that runs on POX [16] controller. Entropy is
have collapsed which is called Single Point of Failure (SPOF). used in order to calculate this probability. There are two
To avoid this situation creation of replication of controllers components in DDoS detection. One of them is windows size
and applications is proposed. Against software bugs usage of and the other one is threshold. The window size is depended
diversity of controllers is recommended. Furthermore, it is on time period and number of packets. Entropy calculates
stated that the switches in data plane must be able to keep in random distribution of incoming packet depending on window
touch with another controller in case the controller is size. If the entropy value exceeds the predefined threshold
collapsed. In such a case, dynamic device association value, traffic is determined as an attack. The proposed method
mechanism which provides the connection of switches with can detect the attack within the first 250 packets of harmful
multiple controllers dynamically would tolerate the faults in traffic by using destination IP addresses. It is stated that attack
the network. detection rate for the predefined threshold value is %96.
Furthermore, such parameters as destination IP addresses,
Use of OpenFlow protocol leads to some security issues window size and threshold value can be set to the desired
with it. For example, an attacker may send too many values in real time according to the requirements of the
OpenFlow request and expose the control plane to DoS attack. controller. In this respect, it offers a flexible solution. Also,
This case will cause a bottleneck between control plane and tests are performed and same results are obtained for TCP and
data plane. So, the network will be unmanageable. Therefore, UDP packets. It is stated that detecting DDoS in its early
the central controller must be protected from DoS and DDoS stages is depended on tolerance of the server used as a
attacks which could affect the entire network. A framework controller and traffic properties.
which is called AVANT-GUARD [12] has been developed for
the purpose of enhancing security in OpenFlow networks. In a recent work [17], a simple DoS prevention system is
This framework is located in the data plane and consists of performed in SDN. A solution is offered against DNS DoS
two modules named Connection Migration and Actuating attacks using flow information obtained from each network
Trigger. Intelligence is added to the data plane in Connection device. Therefore, anomaly detection can be performed on
Migration and control plane is being more resistant against each switch in the local area network. Also, the advantage of
DoS attacks such as TCP SYN. This is carried out by central management which is brought by SDN architecture is
analyzing the TCP sessions opened in the data plane. discussed.
____________________________________________________________________________________________________________
471
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Controller acts like a firewall in OpenFlow-based networks. packet is sent to the NOX controller for source address
The traffic is passed on the controller and analyzed. validation. If an IP spoofing is detected then controller adds a
According to its result, passing of packets is allowed or rule to the switch in order to stop incoming traffic from this
rejected. In [18], it is discussed that a design acts like a source address. VAVE provides protection against data plane
firewall of each switch in data plane and sending of packets to DoS attacks such as flow table flooding.
the controller is not necessary. A flow-based distributed
firewall prototype is developed in this work for developing a OpenFlow and sFlow is combined for anomaly detection
simple packet filtering firewall in SDN. The rule set is and mitigation in [24]. This solution consists of collector,
installed on each network device as flow entries. This firewall anomaly detection and anomaly mitigation modules. Flow
prototype creates a firewall object for each network device statistics are gathered by using OpenFlow and sFlow protocols
connected to the controller. Firewall object is connected to the in the collector module. The statistics are analyzed and
related device without any delay. Each firewall object has an anomalies are identified in anomaly detection module. Flow-
index number and stored in a list in order. The functions of entries are inserted in the flow table of switches in order to
each firewall object can be accessed through command line by neutralize malicious traffic in anomaly mitigation module.
index number in the list. Firewall can control the traffic by Flow-entries which are inserted have higher priority than any
modifying flow tables of switches in data plane. existing flow-entry in the flow table. These modules act as a
feedback control loop. This architecture supports various
In [19], more than one controller usage is suggested for the algorithms such as statistical anomaly detection, machine
purpose of protecting control plane from unauthorized access learning-based anomaly detection and data mining-based
in SDN. Each switch in data plane can be managed by more anomaly detection according to preferred design. In this study,
than one controller using byzantine fault tolerance algorithm. entropy-based algorithm is used. DDoS attacks, worm
propagation and port scan attacks are detected successfully.
A security application kernel which is called FortNOX [20]
is proposed for preventing fraudulent rule insertion that may SDN, Network Functions Virtualization (NFV) and cloud
be caused by malicious applications. FortNOX provides computing technologies will play important role to meet the
prioritizing the flow rules by performing role-based requirements of future mobile networks. In [25], multi-tier
authorization. Furthermore, it detects a new flow rule which security architecture is presented to solve the security
conflicts with an existing flow rule. problems in the future of Software-Defined Mobile Networks
(SDMN). This architecture consists of four components.
Security Enhanced Floodlight (SE-Floodlight) which is an Security is provided between the control and data plane
extension of OpenFlow controller Floodlight is introduced in communication by using Host Identity Protocol (HIP) and
[21]. SE-Floodlight which is an improvement of FortNOX IPSec tunneling techniques. Rule-based approach is used to
offers a Security Enforcement Kernel (SEK). It provides role- protect the network to unwanted access, source address
based authorization between control plane and OpenFlow spoofing and DoS attacks. Software-Defined Monitoring
applications in application plane. SE-Floodlight has a dijital (SDM) is used to detect and prevent security threats on the
signature validation for each rule insertion. OpenFlow network. SDM uses Deep Packet Inspection (DPI) and traffic
application is digitally verified by the SEK at runtime. After monitoring techniques. Synchronizing network security with
the application is signed and validity is verified, it is permitted the network traffic provides real-time information and
for the purpose of making query, modification on the network necessary flow rules are installed to the flow tables of
or creating traffic flow rule. switches in the data plane.
In [22], assignments of full priveleges for each OpenFlow V. CONCLUSION AND FUTURE WORK
application which cause unauthorized access problems are SDN bring a new generation networking approach.
discussed. In this context, an isolation mechanism is proposed. Traditional and cumbersome network architectures transform
The system which is called PermOF provides permissions into the dynamic network architecture with SDN. Although
with minimum priveleges for applications. PermOF enforce to SDN provides an open and programmable platform there are
perform these permissions in Application Programming many problems to be solved in topics such as network security,
Interface (API) entry of the controller. It is stated that this routing algorithms, virtualization and load balancing. SDN
solution protects the network from unauthorized controller must be designed in a good manner from the security
attacks. perspective.
In [23], a solution which is called Virtual Address At this stage, we believe it is too early that we can say SDN
Validation Edge (VAVE) is proposed for IP spoofing in provide a secure network infrastructure. There is much work
OpenFlow/NOX architecture. VAVE is an application that needs to be done and more effort should be spent on SDN
runs on the controller. It performs source address validation security for SDN potential.
against IP spoofing attacks. If any incoming packet does not
match a rule in the flow table of OpenFlow switch then first
____________________________________________________________________________________________________________
472
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Cognitive Networks (CN) which exhibit intelligent [15] S. M. Mousavi and M. St-Hilaire. “Early Detection of DDoS Attacks
against SDN Controllers”. IEEE International Conference on
behaviours will probably come out in the near future, we
Computing, Networking and Communications, Communications and
believe AI-based solutions in SDN security would be much Information Security Symposium, pp. 77-81, 2015.
more useful. Studies on integration of AI techniques with [16] S. Ramadona, B. A. Hidayatulloh, D. F. Siswanto and N. Syambas.
SDN, 4G/5G networks, Heterogeneous Networks (HetNets) “The Simulation of SDN Network Using POX Controller: Case in
Politeknik Caltex Riau”. 9th International Conference on
and mobile networks can play an important role in the creation
Telecommunication Systems, Services and Applications (TSSA), pp.
of CN. Processing big data with using AI techniques such as 1-6, 2015.
machine learning will allow the development of CN. CN will [17] G. Akin, E. Karaarslan, O. Buk and E. Ucar. “SDN Architecture
depend on information and be learning-based, exhibiting Fundamentals & DoS Prevention Basics: A Case Study with
OpenFlow”. International Scientific Conference (UNITECH 2015),
intelligent behaviors. With CN, it may be possible to develop
Gabrovo, 2015.
networks which can learn from past data and decide [18] J. G. V. Pena and W. E. Yu. “Development of a Distributed Firewall
automatically about encountered in future events. It will be Using Software Defined Networking Technology”. IEEE 4th
possible to give specific services to the users with this International Conference on Information Science and Technology, pp.
449-452, 2014.
intelligent network architecture. As a future work, we plan to
[19] H. Li, P. Li, S. Guo and S. Yu. “Byzantine-Resilient Secure Software-
represent these issues and propose new solutions in the Defined Networks with Multiple Controllers”. IEEE International
following publications. Conference on Communications (ICC 2014) - Communication and
Information Systems Security Symposium, pp. 695-700, 2014.
REFERENCES [20] P. Porras, S. Shin, V. Yegneswaran, M. Fong, M. Tyson and G. Gu. “A
Security Enforcement Kernel for OpenFlow Networks”. First ACM
[1] Open Networking Foundation, “Software-Defined Networking: The
SIGCOMM Workshop on Hot Topics in Software Defined Networks,
New Norm for Networks”. White Paper, 2013.
pp. 121-126, 2012.
[2] N. McKeown, T. Anderson, H. Balakrishnan, G. Parulkar, L. Peterson,
[21] P. Porras, S. Cheung, M. Fong, K. Skinner and V. Yegneswaran.
J. Rexford, S. Shenker and J. Turner. "OpenFlow: Enabling Innovation
“Securing the Software-Defined Network Control Layer”. Network and
in Campus Networks". ACM SIGCOMM Computer Communication
Distributed System Security Symposium (NDSS), pp. 1-15, 2015.
Review, vol. 38, no. 2, pp. 69-74, 2008.
[22] X. Wen, Y. Chen, C. Hu, C. Shi and Y. Wang. “Towards a Secure
[3] D. Kreutz, F. M. V. Ramos and P. Verissimo. “Towards Secure and
Controller Platform for OpenFlow Applications”. Second ACM
Dependable Software-Defined Networks”. Second ACM SIGCOMM
SIGCOMM Workshop on Hot Topics in Software Defined Networking,
Workshop on Hot Topics in Software Defined Networking, pp. 55-60,
pp. 171-172, 2013.
2013.
[23] G. Yao, J. Bi and P. Xiao. “Source Address Validation Solution with
[4] S. Scott-Hayward, S. Natarajan and S. Sezer. “A Survey of Security in
OpenFlow/NOX Architecture”. 19th IEEE International Conference on
Software Defined Networks”. IEEE Communication Surveys &
Network Protocols (ICNP), pp. 7-12, 2011.
Tutorials, vol. 18, no. 1, pp. 623-654, 2016.
[24] K. Giotis, C. Argyropoulos, G. Androulikadis, D. Kalogeras and V.
[5] M. Dabbagh, B. Hamdaoui, M. Guizani and A. Rayes. “Software-
Maglaris. “Combining OpenFlow and sFlow for an Effective and
Defined Networking Security: Pros and Cons”. IEEE Communications
Scalable Anomaly Detection and Mitigation Mechanism on SDN
Magazine - Communication Standards Supplement, pp. 73-79, 2015.
Environments”. Computer Networks, vol. 62, pp. 122-136, 2014.
[6] Open Networking Foundation, “OpenFlow Switch Specification”,
[25] M. Liyanage, I. Ahmad, M. Ylianttila, J. L. Santos, R. Kantola, O. L.
Version 1.5.1, 2015.
Perez, M. U. Itzazelaia, E. M. de Oca, A. Valtierra and C. Jimenez.
[7] R. Holz, T. Riedmaier, N. Kammenhuber and G. Carle. “X.509
“Security for Future Software Defined Mobile Networks”. IEEE 9th
Forensics: Detecting and Localising the SSL/TLS Men-in-the-Middle”.
International Conference on Next Generation Mobile Applications,
17th European Symposium on Research in Computer Security
Services and Technologies (NGMAST 2015), pp. 256-264, 2015.
(ESORICS 2012), pp. 217-234, 2012.
[8] D. Kreutz, F. M. V. Ramos, P. E. Verissimo, C. E. Rothenberg, S.
Azodolmolky and S. Uhlig. "Software-Defined Networking: A
Comprehensive Survey". Proceedings of the IEEE, vol. 103, no. 1, pp.
14-76, 2015.
[9] Y. Jarraya, T. Madi and M. Debbabi. "A Survey and a Layered
Taxonomy of Software-Defined Networking". IEEE Communication
Surveys & Tutorials, vol. 16, no. 4, pp. 1955-1980, 2014.
[10] S. Scott-Hayward, G. O’Callaghan and S. Sezer. “SDN Security: A
Survey”. IEEE SDN for Future Networks and Services (SDN4FNS
2013), pp. 1-7, 2013.
[11] I. Ahmad, S. Namal, M. Ylianttila and A. Gurtov. “Security in
Software Defined Networks: A Survey”. IEEE Communication
Surveys & Tutorials, vol. 17, no. 4, pp. 2317-2346, 2015.
[12] S. Shin, V. Yegneswaran, P. Porras and G. Gu. “AVANT-GUARD:
Scalable and Vigilant Switch Flow Management in Software-Defined
Networks”. 20th ACM SIGSAC Conference on Computer &
Communications Security (CCS 2013), pp. 413-424, 2013.
[13] S. A. Mehdi, J. Khalid and S. A. Khayam. “Revisiting Traffic Anomaly
Detection using Software Defined Networking”. 14th International
Conference on Recent Advances in Intrusion Detection (RAID 2011),
pp. 161-180, 2011.
[14] N. Gude, T. Koponen, J. Pettit, B. Pfaff, M. Casado, N. McKeown and
S. Shenker. “NOX: Towards an Operating System for Networks”.
ACM SIGCOMM Computer Communication Review, vol. 38, no. 3,
pp. 105-110, 2008.
____________________________________________________________________________________________________________
473
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Bilgisayar bilimlerindeki gelişmeler ve tıp Birçok alanda kullanılan görüntü işleme segmentasyonu, tıp
dünyasındaki paradigma değişimleri sonucunda artık günümüzde alanında da yoğun bir şekilde kullanılmakta olup, birçok
bilgisayarsız tıp uygulamaları düşünülemez hale gelmiştir. hastalığın tanı ve tedavisinde büyük bir önem arz etmektedir
Bilgisayar destekli görüntüleme yöntemlerindeki gelişmeler ve (Altuntaş, 2015). Kapilleroskopi cihazı ile yapılan tetkikler de
hesaplama kapasitesindeki artışlar sayesinde organizmaya
bunlardan bir tanesidir. Kapilleroskopi, bir cihaz yardımıyla
müdahale etmeden içsel durumlar gözlenebilmekte ve ilgili
tedaviler planlanabilmektedir. Bu gözlemlemeleri yapabilmek kılcal damarların resim ya da video görüntülerinin alınması
için öncelikle sağlam, güvenilir ve hızlı bir bölütlemenin yapılması işlemidir. Alınan bu görüntülerle kılcal damarlar incelenir ve
gerekmektedir. Bölütleme, genelde görüntü işleme ve özelde de damarlardaki değişiklikler bazı hastalıkların habercisi olabilir.
tıbbi görüntü işlemede, görüntü çözümlemesi ve yorumlaması için Özellikle tırnak dibi kapilleroskopisi romatolojik hastalıkların
gerekli temel önişlemlerden biridir. Görüntülerin sağlıklı ve hızlı tanısının koyulmasında kolaylık sağlamaktadır (Elmas, 2015).
bir şekilde bölütlenebilmesi, tıbbi uygulamalarda tanı ve tedavi
planlamasının daha iyi ve hızlı yapılabilmesini sağlamaktadır. II. MATERIALS AND METHODS
Bu çalışmada matched filter (Eşleştirilmiş filtre) kullanarak
Matched filter (Eşleştirilmiş filtre) kullanarak
videocapillaroscopic görüntülerinde damarları bölütleyip ve bu
bölütlenmiş damarları bir veri tabanındaki bölütlenmiş
videocapillaroscopic görüntü damarları bölütlüyüp ve bu
damarların üzerinde karşılaştırılmıştır. Eşleştirilmiş filtre bölütlenmiş damarları bir veri tabanındaki bölütlenmiş
yöntemi ilk kez dünyada bu görüntülerin üzerinde uygulanmıştır. damarların üzerinde karşılaştırılmıştır.
Eşleştirilmiş filtre aşağıdaki denklemde gösterilmiştir.
Keywords—Görüntü bölütleme, Kapiller video, Segmentasyon,
Matched filter x2
1 (
2s2
)
f ( x, y ) e m (1)
I. INTRODUCTION 2s
Görüntüyü çeşitli öğelere ayırma işlemi olan segmentasyon
(bölütleme); görüntü analizi, örüntü tanıma ve görüntü işleme Burada s filtrenin skalasını ifade eder. m aşağıdaki
için başlıca öneme sahiptir. Segmentasyon, görüntüdeki denklemde gösterilmiştir.
ts x2
bölgelerin belirlenmesi ya da görüntüdeki objelerin 1 ( )
2s2
kenarlarının belirlenip, görüntünün parçalara ayrılmasını m ( e dx ) /( 2ts ) (2)
sağlayan düşük seviye işlemdir. Segmentasyonun temel amacı ts 2s
görüntüyü farklı özellikte anlamlı bölgelere ayırmaktır. Birçok
görüntü işleme uygulamaları içerisinde ilk adım Eşleştirilmiş filtre arka plana karşı kan damarlarının
segmentasyondur. Çünkü tanıma ve sorgulama sistemlerinin kontrastını arttırmak için kullanılmaktadır. İlk olarak, bir
kalitesini, segmentasyonun kalitesi belirler. Hiçbir eşleştirilmiş filtre hesaplama ardından damarları geliştirmek
segmentasyon yöntemi mükemmel değildir ve için GLCM istatistiksel bir özelliği hesaplandığı, bir eşik değer
segmentasyonlar genel-bütün görüntüler için kullanılmaz olarak hareket ederek kullanılır.
(Kızılkaya 2008).
Geçmişten günümüze kadar görüntü işlemede gri seviyeli
görüntü kullanılmıştır. Gri seviyeli görüntünün işlenme süresi
kısadır. Gelişen teknoloji ile gri seviyeli görüntü yerini yavaş
yavaş renkli görüntü işlemeye bırakmaya başlamıştır. Renkli
görüntü işleme, griye göre daha fazla bilgi aktarımı sağlar.
Renkli görüntü işlemenin, bilgilerin üç boyutlu renk uzaylarıyla
gösterimi esnasında renklerin ayırt edilememesi sonucu
segmentasyonun zor olması gibi olumsuz yönleri vardır
(Karakoç, 2012).
____________________________________________________________________________________________________________
474
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Videocapillaroscopic görüntüsü
____________________________________________________________________________________________________________
475
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Bahsedilen kriterlere ek olarak, “Hassaslık” olarak [4] M. Wegmuller, J. P. von der Weid, P. Oberson, and N. Gisin, “High
resolution fiber distributed measurements with coherent OFDR,” in Proc.
adlandırılan diğer bir görüntü sınıflandırma kriteri vardır ve
ECOC’00, 2000, paper 11.3.4, p. 109.
denklemi şu şekildedir; [5] R. E. Sorace, V. S. Reinhardt, and S. A. Vaughn, “High-speed digital-
to-RF converter,” U.S. Patent 5 668 842, Sept. 16, 1997.
[6] (2002) The IEEE website. [Online]. Available: http://www.ieee.org/
= (4) [7] M. Shell. (2002) IEEEtran homepage on CTAN. [Online]. Available:
http://www.ctan.org/tex-
archive/macros/latex/contrib/supported/IEEEtran/
[8] FLEXChip Signal Processor (MC68175/D), Motorola, 1996.
[9] “PDCA12-70 data sheet,” Opto Speed SA, Mezzovico, Switzerland.
REFERENCES [10] A. Karnik, “Performance of TCP congestion control with rate feedback:
[1] S. M. Metev and V. P. Veiko, Laser Assisted Microtechnology, 2nd ed., TCP/ABR and rate adaptive TCP/IP,” M. Eng. thesis, Indian Institute of
R. M. Osgood, Jr., Ed. Berlin, Germany: Springer-Verlag, 1998. Science, Bangalore, India, Jan. 1999.
[2] J. Breckling, Ed., The Analysis of Directional Time Series: Applications [11] J. Padhye, V. Firoiu, and D. Towsley, “A stochastic model of TCP Reno
to Wind Speed and Direction, ser. Lecture Notes in Statistics. Berlin, congestion avoidance and control,” Univ. of Massachusetts, Amherst,
Germany: Springer, 1989, vol. 61. MA, CMPSCI Tech. Rep. 99-02, 1999.
[3] S. Zhang, C. Zhu, J. K. O. Sin, and P. K. T. Mok, “A novel ultrathin [12] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)
elevated channel low-temperature poly-Si TFT,” IEEE Electron Device Specification, IEEE Std. 802.11, 1997.
Lett., vol. 20, pp. 569–571, Nov. 1999.
____________________________________________________________________________________________________________
476
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Gastric cancer is one of the most common reflections and intestinal juices cover the surface
and the most common fatal cancers in the world. In tissue with color and texture patterns, rendering
this study, our aims to detect the cancerous regions these image parts unsuited for cancer detection based
semi-automatically according to the endoscopy on color and texture analysis. The black image
images obtained. In this process, semi-automatic border and the lumen do not show any tissue, which
diagnosis of cancer is done according to the makes them irrelevant for our detection system.
endoscopic images received from the patients using Since we aim to use color and texture information for
image processing techniques and the results are the detection of early cancer, we define a region of
compared with the cancerous region determined interest for two reasons: (1) decreasing the number
from the endoscopic images by specialized doctors. of false positives that are caused by other texture-
The Gastric cancerous region obtained semi rich or color-deviating parts of the image and (2)
automatically using image processing techniques reducing the computation time by only analyzing the
and determined by the specialized doctors are relevant image area[3].
compared. 96.1641% accuracy rate is determined
II. MATERYAL VE METOTLAR
from the comparison results. Considering the results
obtained, it can be seen that the suspected region In this study, image processing techniques and
can be determined via software by the specialized Image Processing Toolbox of Matlab R2016a has
doctors. been used for this study.
Keywords— Gastric Cancer, Image Processing , Endoscopic 2.1. Digital Image Processing:
Image Image processing is a rapidly growing area of
computer science. Its growth has been fueled by
technological advances in digital imaging, computer
I. INTRODUCTION
processors and mass storage devices. Fields which
Gastric cancer is the fourth most commonly traditionally used analog imaging are now switching
diagnosed cancer and the second leading cause of to digital systems, for their flexibility and
cancer-related deaths worldwide [1]. Symptoms affordability. Important examples are medicine, film
associated with gastric cancer often seen in advanced and video production, photography, remote sensing,
stage tumors. Therefore, it is not easy to diagnose the and security monitoring. These and other sources
tumors at an early stage of the cancer. Endoscopic produce huge volumes of digital image data every
examination and biopsy is the most important tool in day, more than could ever be examined manually [4].
the diagnosis of the gastric cancer. In the
examination, all parts of the gastric can be seen and The comparison between semi-automatically
the cardia and fundus evaluation can be provided. detected cancerous region and the detected region by
The accuracy rate of diagnosis by biopsy is around specialized doctors is done in this article.
80% - 85% [2]. The detection of cancerous regions
using image processing techniques is aimed by In computer vision, image segmentation is the
benefitting from endoscopy images obtained from process of partitioning a digital image into multiple
the patients in the Selcuk University Medical Faculty segments (sets of pixels, also known as super-pixels).
Hospital Gastroenterology Unit. Endoscopic images The goal of segmentation is to simplify and/or
generally contain elements that are unsuited or change the representation of an image into
irrelevant for our analysis. For example, specular
____________________________________________________________________________________________________________
477
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig.2
The representation of images, which is the
combination of discrete areas, applied on the
endoscopy image is shown in Fig.3.
____________________________________________________________________________________________________________
478
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig.3
At this stage, after the segmentation process, the
specialized doctor activates the software by clicking Fig.5
on the suspected cancerous regions. During the
execution of the software, the region determined IV. CONCLUSIONS
after the segmentation process and the region
determined by the specialized doctor is compared In this study, success is obtained in the semi-
and the accuracy rate is calculated. automatic detection of gastric cancer with image
processing techniques. In our study, values such as
The cancerous region determined by the software is accuracy rate of 96.1641% and 0.8160 Area Under
shown in Fig. 4. Curve (AUC) value indicates our success rate. Roc
Analysis chart belonging to AUC is shown in Fig.6.
Fig6
Fig.4
On the other hand, the region determined as ACKNOWLEDGEMENTS
cancerous by the specialized doctor is shown in Fig.
This study is supported by Selcuk University
5.
Scientific Research Projects Unit (BAP) under the
project number of 15101020. Appreciations to
Selcuk University Scientific Research Unit (BAP)
due to its support.
____________________________________________________________________________________________________________
479
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. REFERENCES
[1] Jemal A, Center MM, DeSantis C, Ward EM. Global
patterns of cancer incidence and mortality rates and trends.
Cancer Epidemiol Biomarkers Prev 2010;19:1893-1907.
____________________________________________________________________________________________________________
480
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract– Data mining is used for the extraction of secret, body and they help the disposal of waste through the urinary
valuable and usable data from the big data and to provide tract. Sometimes some of this waste might cumulate in the
strategic decision support. It created a new perspective for the kidneys. Renal calculi are the solid pieces that are formed
use of the data in healthcare in addition to finding the answers of
with the crystallization of calcium, phosphate and other
unexplored questions. It has gained wider usage as a method.
minerals in the urine.
The aim of this study is to develop a decision tree and a list of
Drinking inadequate amount of water, obesity, consuming
rules by data mining for the early diagnosis of renal calculi. A
data set including blind and retrospective data for 150 people can
too much sugar and salt might cause nephrolithiasis as well as
diagnose with 6 attributes. A decision support system analysis genetic factors. Most of the renal calculi are calcium stones
was developed for the diagnosis of the patients with suspected resulting from the accumulation of the calcium in the kidneys.
renal calculi. Based on the results obtained and the analysis In addition to the environmental factors in the formation of
developed, a decision tree and list of rules were created to renal calculi, genetic factors may be the primary cause. Not
determine the factors that affect renal calculi. Weka program having enough water is one of the biggest risk factors for renal
and J48 algorithm were used to create the decision tree and the
calculi. It is often recommended to drink about 2 liters of
list of rules and it was found to be 74.63% successful.
water daily. Besides, it is good for our well-being in general
Keywords– Data Mining, Decision Tree, Renal Calculi Diagnosis,
not to consume too much sugar and salt, to follow a balanced
J48
diet, to stay away from convenience food and not to be
overweight. Renal calculi may also stay in the kidneys without
I. INTRODUCTION
The aim of medical informatics is to use computer and showing any symptoms or doing any harm to the kidneys.
communication technologies that are interacting with other Following the general information given for renal calculi, we
sciences in order to utilize, analyze and reconstruct medical are going to talk about data mining.
information in an effective way. Medical informatics helps to Data mining is the extraction of secret, valuable and usable
obtain valid, detailed and reliable results on a global scale as it information from a big amount of data. Data mining which is
exponentially improves the data collection, process and used to provide strategical decision support aims to construct
evaluation capacities in medical centers [1]. decision making models based on analysis methods.
A renal calculus is among the most common diseases in It creates directive models for decision making techniques
Turkey. About 15% of women and 5% of men in Turkey have for medical institutions where the data is collected for
been diagnosed with nephrolithiasis. It generally occurs in analyses [2]. It is important to emphasize the difference
thirties for the first time. Kidneys are like filters in the human between flaws and misappropriations in health services, to
minimize the risks and to take necessary precautions based on
____________________________________________________________________________________________________________
481
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
this distinction for patients’ security and well-being [3] [4]. A II. RECENT STUDIES
large amount of information can be retrieved from the various İlkuçar examined the chronic kidney disease with the help of
data sets used in daily life [5, 6]. Data mining has made life artificial neural networks in his study. He emphasized what is
easier by enabling us to access to more data in a shorter time to be done and which tests to be run for early diagnosis of
span [7]. In the simplest term, data mining is the extraction of chronic kidney disease [21]. Danacı et al. focused on the
imprecise, valid and applicable data from data stacks with a diagnosis of breast cancer cells using data mining methods
dynamic process [8, 9]. It is possible to obtain valuable [22]. In the study conducted by Yurtay et al. they did a data
information from the big data stacks using data mining and mining research for anaemia diagnosis. In a study on iron
statistical analysis techniques. This information helps doctors deficiency anaemia, the system is run with decision trees [23].
during the decision process with computer-aided diagnosis Özkan et al. tried to improve the diagnosis accuracy of the
work and contributes to the improvement of health care laboratory tests used for the diagnosis of fibromyalgia
Although potent devices have been produced by expert response parameters. SSR parameters and laboratory tests
systems, they have not gained common use because of the which were calculated by Matlab were analyzed with artificial
field-related data changing rapidly and the diversity of views neural networks and the percentages of accuracy were found
among experts [11]. Data warehouse is a large store of data [24]. Kökver et al. probed the factors affecting hypertension
accumulated from a wide range of sources under the same with data mining methods. They developed a diagnosis system
roof [12]. The steps in data mining include the definition of that will estimate whether the patients have hypertension or
the problem to be solved, the obtaining of previous data about not [25].
the problem, the selection of usable data, housekeeping of the Kusiak et al. work through a decision support system which
data, the evaluation of analysis results, and the utilization of will determine whether the lung tumor is benign or malignant
these results [13, 14]. The time saving and accelerating effect [26]. In their research, Topaloğlu & Sur created a decision
of the data selection will be obvious in the later stages [15]. support system to diagnose hepatitis and to minimize the
A decision tree is an approach commonly used in data number of wrong diagnoses. It will help the doctors with the
mining for categorization and estimations. Decision trees diagnosis of hepatitis [27].
____________________________________________________________________________________________________________
482
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The Importance of the Study The Formation of the Decision Tree with Weka
Urinary analysis is one of the most common methods used J48, which is a decision tree algorithm, was used in the
for the diagnosis of renal calculi. Especially substances like program. The last one of the training set with seven attributes
uric acid and calcium present in normal urine become is the “RISK” attribute where the diagnosis is made. The
crystallized and form the structures called renal calculi. These others are as follows; Leucocyte (LOW, MILD, HIGH), Urine
formations can have great negative effects on a patient’s daily Color (COLORLESS, STRAW-COLORED, YELLOW), PH
life. The diagnosis of the disease of which treatment process is (LOW, NORMAL, HIGH), Bilirubin (NEGATIVE,
quite hard is also of utmost importance. POZITIVE), Appearance (CLEAR, BLURRED,
Data Set SLIGHTLYBLURRED, HEAVILYBLURRE), Erythtocyte
For this study, a data set including retrospective and blind (TRACE AMOUNT, RARE, NORMAL). 55% of the data
data from 150 people. Six attributes were used for the obtained were used for training while the rest was used for
diagnosis of the disease. The seventh attribute is the testing. Classification model provides information about the
comments “yes” and “no” for the diagnosis. The six attributes structure and the size of the tree. This model belongs to the
used are as follows; learning set.
Leucocyte,
Urine color, Full Urine Analysis and Value Ranges
PH, If Leucocyte value is between; 0 and 3 Low, 3 and 5
Appearance, If PH value is between; 4,5 and 5,5 Low, 5,5 and 6,5
The Decision Tree Model and Algorithm to Be Applied to the If Erythrocyte value is between; 0 and 4 Rare, 4 and 9
processed in csv format and the data was converted to arff shown below;
format and transferred to Weka Program. J48 algorithm was @relation kidney disease
used to determine the factors contributing to the formation of @attribute LEUCOCYTE string
renal calculi and for the diagnosis. RISK was selected as the @attribute COLOR string
root node. The results were transferred with the decision tree @attribute PH string
____________________________________________________________________________________________________________
483
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Evaluation module gives the classification error and the 6. If ERITROSIT = Normal and Color = Straw-colored
Kappa (=0.482) statistic both. The mean absolute error and the and LEUCOCYTE = High, then Risk = Yes
root mean squared error of the category probability 7. If ERITROSIT = Normal and Color = Straw-colored
estimations assigned by the tree are found 0.3049 and 0.4345, and LEUCOCYTE = Medium, then Risk = Yes
respectively. Classification performance was calculated as
V. CONCLUSIONS
74.63%. This value shows that the decision tree is 74.63%
Full urine analysis values were examined with the method
successful.
used in this study. It is possible to diagnose renal calculi
The algorithm excludes meaningless variables
without a surgical examination of the patient. Thus, it helps to
automatically and it makes the selection of the variables itself
take the necessary precautions for nephrolithiasis.
during the new learning process [28]. Here, the gender was
According to the design of the decision tree, the most
considered to be a meaningless variable and it was excluded
important attribute to take into account is the amount of
from the decision tree variable order. Activity was chosen as
Erythrocyte. If the amount of the Erythrocyte is above 10,
the root node. The decision tree created based on J48
then a pathological test needs to done with ultrasonography.
algorithm is presented in Figure-1.
Again, if the amount of the Erythrocyte is between 4 and 9,
then the color of the urine, which is among the attributes,
should be examined. If the color is “Yellow”, it means that
you do not have any renal calculi. If the color is “Straw-
yellow”, then the amount of the Leucocyte should be
examined. If the amount of Leucocyte is more than 3, then
you are very likely to have renal calculi. If it is below 3, then
you do not have it. If the color of the urine is “Colorless”, then
you need to have and pathological test. If the amount of the
Erythrocyte is below 4, it means that you are less likely to
have kidney problems.
____________________________________________________________________________________________________________
484
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
485
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, the banknote authentication data set in clustering process using ANN approach on the pictures
the UCI machine learning repository was used as classification belonging to our dataset to determine if the banknotes are
data set. Four features obtained from banknote images that were genuine or counterfeit. Four input parameters, one hidden layer
taken from genuine and counterfeit banknotes were used for with 10 neurons and one output has been used for the ANN.
classifying them as genuine or counterfeit. 906 of 1372 data in the
Wavelet Transform tool are used to extract features from
dataset were assigned for training and rest of them were assigned
for testing. Weka (Waikato Environment for Knowledge Analysis) images. It is determined that the training regression is 0.99914,
software was used as classification environment. The classification testing regression is 0.99786 and the validation regression is
success rates were calculated by using data mining algorithms like 0.9953, respectively. Mariana et al. (2013) have used Raman
Multilayer Perceptron, RBFNetwork, RBFClassifier, kNN, J48, spectroscopy and Partial least squares discriminant analysis
RandomForest, RandomTree, NaiveBayes, BayesNet, OneR, (PLS-DA) for determining the authentic and counterfeit
DecisionTable and Kstar. The best classification success rate was banknotes. The classification method PLS-DA is employed to
achieved by using Multilayer Perceptron model. The classification discriminate authentic and counterfeit banknotes, as well as the
success rates for various number of neurons in the hidden layer counterfeit type. In the proposed method all fake and not fake
were obtained in Multilayer Perceptron model. The best success
banknotes used to validate the analysis are correctly classified.
rate was obtained as 100 % when the model has 4 neuron in the
hidden layer. Hinwood et al. (2006) have designed a system that helps the
blind Australians to recognition of banknotes. Each note has the
same feel, with no Braille markings, irregular edges or other
Keywords— Data mining, Weka, MLP, kNN
tangible features in Australia. Their portable system called
I. INTRODUCTION MoneyTalker scans and speaks about the banknote. The
accuracy, weighted by the percentage breakdown of total notes
By the improvement in banking services the Asynchronous
in issue is determined as 98.9%.
Transfer Mode (ATM) machines have been grown up
In this study the banknotes have been classified as genuine
worldwide. Due to this fact the classification of banknotes as
or counterfeit by using 4 attributes created from their images
genuine or counterfeit become an important issue. Image
that obtained from banknote authentication dataset. In
processing occupies an important role during feature extraction.
classification process 12 machine learning algorithm have been
After the feature extraction operation data mining algorithms
used. The classification success rates and error values obtained
have to be used for classification.
from each classification algorithm have been compared.
Shan et al. (2013) used quaternion wavelet transform for
feature extraction. In their study, Artificial Neural Network
(ANN) has been used as classifier. The lowest classification II. MATERIAL AND METHODS
error rate is determined as 0.58 %. Ali et al. (2004) have
proposed a method that rejects counterfeit banknotes. The
improvement of reliability in banknote neuro-classifier is Dataset
investigated and a reject option is proposed based on the
probability density function of the input data. In their study, for In this study, images that were taken from genuine and
modelling the non-linear correlation among the data variables counterfeit banknote-like specimens have been used to extract
and extracting the features, a local principal components four attributes. For digitization, an industrial camera usually
analysis (PCA) is applied. The method is tested with a learning used for print inspection, was used. The final images have
vector quantization (LVQ) classifier. The results show that by 400x400 pixels. Due to the object lens and distance to the
taking a suitable reject threshold value and also a proper investigated object gray-scale pictures with a resolution of
number of regions for the local PCA, the reliability of system about 660 dpi were gained. Wavelet Transform tool were used
can be improved significantly. Esra et al. (2015) has performed to extract features from images. The attributes that have were
____________________________________________________________________________________________________________
486
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
variance of Wavelet Transformed image, skewness of Wavelet III. RESULTS AND DISCUSSION
Transformed image, curtosis of Wavelet Transformed image
and entropy of image. The data set was obtained from UCI During classification process the WEKA software have been
machine learning repository [6]. used for application of algorithms. Classification success rates
and error values like MAE and RMSE have been determined
Software-WEKA by changing the neuron numbers in the hidden layers in the
classification by using same dataset on MLP model. The
Developed by Waikato University in New Zealand, WEKA success rates, MAE and RMSE values have been presented on
is an open-source data mining software with a functional Table 1. The highest success rate have been obtained when the
graphical interface which incorporates machine learning number of neuron in the hidden layer is 4 for MLP model. The
algorithms [7]. WEKA includes various data pre-processing, MLP model for best situation has been presented in Figure 2.
classification, regression, clustering, association rules, and
visualization tools. The algorithms can be applied on the data
cluster either directly or by calling via Java code [8][9]. They
TABLE I
are also suitable for developing new machine learning SUCCESS RATE OBTAINED BY USING MULTILAYER PERCEPTRON CLASSIFIER
algorithms. ERROR VALUES
Then the same data set have been used for classification with
the RBFNetwork, RBFClassifier, kNN, J48, RandomForest,
RandomTree, NaiveBayes, BayesNet, OneR, DecisionTable
and Kstar machine learning algorithms. The classification
Fig. 1 The block diagram of the classification process success rates, MAE and RMSE error values for these
algorithms have been obtained. The success and error rates
obtained using 11 different classification algorithms
(Multilayer Perceptron, kNN, RBFNetwork, RBFClassifier,
kNN, J48, RandomForest, RandomTree, NaiveBayes,
____________________________________________________________________________________________________________
487
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
RBFClassifier 99.5708 0.07 0.115
kNN 99.5708 0.0054 0.0654 [1] Gai, S., Yang, G., Wan, M. Employing quaternion wavelet transform for
banknote classification, Neurocomputing, Volume 118, 22 October
J48 96.3519 0.0389 0.1801 2013, Pages 171–178.
[2] Ahmadia, A., Omatua, S., Fujinakaa, T., Kosakab, T. Improvement of
RandomForest 98.927 0.0227 0.0903 reliability in banknote classification using reject option and local PCA,
Information Sciences, Volume 168, Issues 1–4, 3 December 2004, Pages
277–293.
RandomTree 97.2103 0.0279 0.167 [3] Kaya, E., Yasar, A., Saritas, I. Banknote Classification Using Artificial
Neural Network Approach, International Journal of Intelligent System
Naive Bayes 85.6223 0.1842 0.3158 and Applications in Engineering, Volume 4, No 1 2016, Pages 16-19.
[4] Almeida, M.R., Correa, D.N., Rocha, W.F.C., Scafi, F.J.O., Poppi, R.J.
Discrimination between authentic and counterfeit banknotes using
Bayesnet 93.3476 0.1272 0.2422
Raman spectroscopy and PLS-DA with uncertainty estimation,
Microchemical Journal, Volume 109 (2013) 170–177.
OneR 85.8369 0.1416 0.3763 [5] Hinwood, A., Preston, P., Suaning, G.J., Lovell, N. H. Bank note
recognition for the vision impaired, Australasian Physical &
DecisionTable 95.279 0.0898 0.1869 Engineering Sciences in Medicine, Volume 29, Number 2, 2006 Pages
229-233.
[6] UCI Machine Learning Repository, Banknote authentication Data Set,
Kstar 99.5708 0.0089 0.0504 https://archive.ics.uci.edu/ml/datasets/banknote+authentication,
15.06.2016
[7] Witten I.H., Frank E., & Hall M.A. (2011). Data mining: practical
machine learning tools and techniques. Elsevier, London.
[8] Patterson, D., Liu, F., Turner, D., Concepcion, A., & Lynch, R., (2008).
Performance Comparison of the Data Reduction System. Proceedings of
the SPIE Symposium on Defense and Security, Mart, Orlando, FL, pp.
27-34.
[9] Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P., & Witten,
I.H. (2009). The WEKA Data Mining Software: An Update. ACM
SIGKDD Explorations Newsletter, 11(1), 10–18.
____________________________________________________________________________________________________________
488
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
I. INTRODUCTION
Direct Current (DA) motors are frequently utilized in open
and closed circuit applications in the industry. Majority of these where the parameters [8] of the DC motor are shown in Table
applications require control to ensure high performance [1]. 1.
Acquisition of high performance in industrial applications is TABLE 1 DEFINITION OF PARAMETERS
crucial. In the present study, chopper direct current drivers
which display high efficiency, flexible control capability and No Definition Symbols Unit
small and light-weight physical characteristics in number of 1 Rotor speed ω(t) rad s-1
2 Armature resistance R Ω
applications were employed [2-3]. Speed and load fluctuations
3 Armature inductance L H
on DC motors are undesired circumstances. Difficult hardware 4 Armature current i(t) A
infrastructure of control methods and their dependency to 5 Load torque TL Nm
mathematical modeling allow formation of control methods 6 Rotor inertia J kg m2
through various artificial intelligence-based control systems. 7 Is torque constant Kt V s rad-1
Type-1 fuzzy logic systems are not capable of directly 8 Back EMF constant Kp V s rad-1
tolerating some uncertainties because they use Type-1 fuzzy. 9 Viscous friction coefficient D Nm s rad-1
Type-2 fuzzy set concept was entered by Zadeh as an extension 10 Coulomb friction torque TF Nm
of traditional fuzzy set concept known as Type-1 fuzzy sets [4]. 11 Armature voltage V(t) v
On the other hand, Type-2 fuzzy logic systems are useful
III. TYPE-2 FUZZY LOGIC
especially when it is difficult to determine certain membership
function for a fuzzy set. Thus, Type-2 fuzzy logic systems Fuzzy logic (BM) and Fuzzy Set (BK) Theories were first
could be utilized to eliminate uncertainties related with rules introduced by PhD. Lotfi A. Zadeh in 1965 [9]; and embraced
and even with measurements [5,6]. quite extensive practice area. Fuzzy logic is applied in
In the present study, a driver circuit application controlled numerous research areas such as Fuzzy Math, Fuzzy Systems,
by the armature current GTO (Gate Turn Off) thyristor was Fuzzy Decision Making and Artificial Intelligence. In the
taken into consideration through the MATLAB/Simulink and meantime, the Type-2 Fuzzy Set Concept was also introduced
electro-mechanical system was controlled by the control by Zadeh [10]. Then, Mendel and his students [11,12], and
system. Regarding the speed control of the DC motor, results Turksen [13,14] have improved the Type-2 Fuzzy System
were acquired by harnessing the Type-2 fuzzy logic within the Theory on the basis of the Fuzzy Set Concept. Similar to the
Matlab/Simulink environment. Type-1 Systems, the Type-2 Fuzzy Systems are comprised of
“If-Then” rules as well. However, entry and exit membership
sets are the Type-2 Sets in these. When certain membership
function is difficult to find for Fuzzy Sets, utilization from the
____________________________________________________________________________________________________________
489
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
490
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
VI. RESULTS
Figure 6 exhibits the plotted graph displaying the change in
speed measured at the exit of the system in which Type-2 fuzzy
controller was used. It is possible to observe with the MATLAB
SimPowerSystem that suggested system was able to reach the
reference value of 120 rad/s without exceeding.
Figure 8. Change in Va, Ia and Motor Speed in a Control System over the
time
____________________________________________________________________________________________________________
491
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Flyback converter circuit is a type of DC-DC design flyback converter with parasitic components in [3].
converter circuits widely-used in industry. This paper presents The authors realize the small-signal analysis of flyback
analysis and modelling of flyback converter circuit using state- converter. Ref.[4] gives analysis of different flyback converter
space technique. Switching elements in the converter are topologies: Active clamp converter, converter with RCD
modelled with ideal switch concept. The model of flyback
snubber, interleaved converter, two switch converter, Dual AC
transformer and system equations concerning every topology are
given. The circuit is realized. Experimental waveforms are and DC output converter. Functionalities of these converters
compared with the analysis to show the correctness of the are simulated and their results are compared. Active clamp
modelling and analysis. flyback converters, the one of the most popular flyback
converters, are analysed, designed and implemented in [5-9].
Keywords— Flyback DC-DC converter, state-space model, The resonant flyback converters [10] were proposed to
analysis increase the system efficiency and reduce the switching losses.
In Ref.[11], based on the mechanism that the primary and
I. INTRODUCTION secondary currents of flyback converter do not appear at the
Flyback converter is a type of DC-DC converter circuits. It same time, two different equivalent models of flyback
is widely used in industry because of some advantages: (i) it transformer are brought forward correspondingly. In Ref. [12,
provides an complete isolation between input and output, (ii) 13], the authors analysed the influence of transformer parasitic
One can get multiple and high voltage outputs, (iii) it is the to low power high voltage output flyback converter by
simplest one if compared with other isolated DC/DC converter utilizing the simplified transformer model.
types, (iv) it is cheap and practical to implement. Besides The objective of this paper proposes an effecitve
these advantages, it requires a protection for leakage modelling and analysis method for Flyback converter. The
inductance and because of the B-H curve limitations it is structure of the paper is as follows. Section 2 explains
designed for maximum 150W power applications. Depending fundamental properties and working stages of flyback
on the applications, the flyback DC-DC converter can be converter. In Section 3, first, the flyback transformer is
operated in two different modes. These modes are continuous modelled with an ideal transformer and a magnetizing
conduction mode (CCM) and discontinuous conduction mode inductor, then, system equations concerning every topology
(DCM). The proposed model in this study is suitable for both are obtained. Experimental and simulation results are given
continuous conduction mode (CCM) and discontinuous and compared in Section 4. The paper closes with conclusions
conduction mode (DCM). in Section 5.
Electronic circuits which use integrated-circuits need a
standard DC voltage of fixed magnitude. In some electronic II. BASIC FLYBACK CONVERTER CIRCUIT
equipment, one may need multiple output power supplies. The basic flyback converter circuit is shown in Fig.1. It
These DC power supplies are generally provided from the has a transformer, a main switch, (Q), a diode (D), a filter
standard power source i.e. AC voltage. For low power capacitor, a load (R), an input (E). In the circuit, Ip and Is
applications, the most preferred converter is flyback. The state primer current and secondary current of the transformer,
circuit topology of flyback converter is the simplest if we respectively. The converter operates in two stages according
compare it with other isolated DC-DC converters used in the to the states of the main switch (Q). Mode 1: when the main
switch mode power supplies (SMPS) [1]. The input given to switch (Q) is on-state and the diode is off-state, and Mode 2:
the converter is generally unregulated DC voltage which is when the main switch (Q) is off-state and the diode is on-state.
obtained by rectifying the AC voltage followed by a capacitor Stage 1: In this mode, the main switch (Q) is turned on-
filter. Flyback converter uses a transformer so that they can state, the primary winding of flyback transformer is connected
give single or multiple isolated output voltages. We talk about to the supply. The current and magnetic flux in the primary
energy efficiency, the flyback converter is not good as side are increased and energy is stored in it. While the
compared with other SMPS circuits though its topology is potential induced in the secondary winding has opposite
very simple [2]. It is presented a simple methodology to polarity than the primary winding, it causes the diode (D) to
____________________________________________________________________________________________________________
492
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
get reverse biased and as a result the diode doesn’t allow After adding the equivalent circuit concerning flyback
current through it. The output filter capacitor (C) supplies transformer into the converter, the exact equivalent circuit
energy to the load. relating to the flyback converter is obtained as in Fig. 3.
Stage 2: It starts when the main switch is turned off-state Although primary and secondary currents of flyback converter
after conducting. Since the main switch (Q) is open, the do not appear at the same time, only one equivalent circuit
current and magnetic flux in the primary side are decreased. model for transformer is used during a period. The model is
According to laws of magnetic induction, the voltage valid for both continuous conduction mode (CCM) and
polarities across the windings get reversed. So now, the discontinuous conduction mode (DCM).
positive voltage is induced in secondary side and the diode is
forward biased. Energy stored in the transformer is delivered Ip Is D
Q
to output capacitor and the load.
Ip Is I1 I2
Q D
+
E + U1 J2 R
Lm C
.
E + Lp Ls C R
. İdeal transformer
Flyback transformer
İdeal transformer
The state-space equations concerning the equivalent circuit
Fig. 2 Model of flyback transformer in Fig 4 are obtained in Eq.(3).
____________________________________________________________________________________________________________
493
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I
= =− (1) PARAMETERS OF THE EXPERIMENT CONVERTER
U1 + J2 C
Lm R
= = =− =− (4)
= =− − =− − =− + (5)
−1 Fig. 6 Experimental waveform of output voltage
0
= + (6)
− 0 0
____________________________________________________________________________________________________________
494
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] T. Chen, W. Lin and C. Liaw, “Dynamic modeling and controller
design of flyback converter”, Aerospace and Electronic Systems, IEEE
Transactions on, vol. 35, no. 4, pp. 1230–1239, 1999.
[2] B. Singh and G. D. Chaturvedi, “Comparative performance of isolated
forward and flyback AC-DC converters for low power applications” in
Power System Technology and IEEE Power India Conference,
POWERCON. pp. 1–6, 2008.
[3] S.K. Pandey, S.L.Pati, V.S. Rajguru, “Isolated Flyback Converter
Designing, Modeling and Suitable Control Strategies”, Int. Conf. on
Advances in Power Electronics and Instrumentation Engineering,
ACEEE, 2014.
[4] R.Kanthimathi, .J.Kamala, “Analysis of different Flyback Converter
Topologies”, Int. Conf. on Industrial Instrumentation and Control
Fig. 9 Experimental waveform of secondary current of transformer (IClC), India, 2015.
[5] N.P. Papanikolaou and E.C.Tatakis, “Active voltage clamp in flyback
converters operating in CCM mode under wide load variation", IEEE
The flyback converter circuit with the design parameters in Transactions on industrial electronics, vol. 51, no. 3, june 2004.
[6] N.S. Aruna, D.S.G. Srivani, P. Balaji, “Active clamp circuit to achieve
Table I are simulated with MATLAB using state-space ZVS in multiple output flyback converter”, International Journal of
equations (3) and (6). The result is illustrated in Figure 10. Emerging Technology and Advanced Engineering, vol. 4, Issue 7, pp
Since the input of converter has a ripple voltage, the output 596-601, July 2014.
voltage in Fig.10 has a wide range oscillation. As shown from [7] P. Jailyn, A.A. Dheeraj, and V. Rajini, “Analysis of active clamp
flyback converter”, Modern Applied Science, Vol. 9, No. I, pp.12-24,
Fig. 10, the min. and max. simulation values of output voltage 2014.
are: 30V-49V. [8] D. Wang, “Analysis, design and implementation of an active clamp
flyback converter”, International Conference on Power Electronics
and Drives Systems, pp.424-429, 2005.
[9] B.R. Lin, H.K. Chiang, K.C. Chen, D. Wang, “Analysis, design and
implementation of an active clamp flyback converter”, Int. Conf. on
Power Electronics and Drives Systems”, 2015.
[10] P. Alou, 0. Garcia, J. A. Cobos, J. Uceda, M. Rascon, “Flyback with
active clamp: a suitable topology for low power and very wide input
voltage range applications”, IEEE APEC conference, vol. 1, pp. 242-
248, 2002.
[11] H. Kewei, L. Jie , F. Ningjun, L. Yuebin, H. Xiaolin, W. Luo,
“Modeling Analysis and Simulation of High-voltage Flyback DC-DC
Converter”, IEEE International Symposium on Industrial Electronics
(ISlE), Korea July 2009.
[12] L. Jie, H. Kewei, F. Ningjun, “An Isolated Feedback Circuit for a
Flyback Charging Circuit”, Proceedings of the WSEAS Int. Conference
on Circuits, Systems , Signal and Telecommunications, Gold Coast,
Australia, pp.35-39, Jan. 2007.
Fig. 10 Output voltage [13] L. Jie, H. Kewei, Z. Shengxian, F. Ningjun, H. Guanglin, “Design and
Simulation of a New Isolated Feedback Circuit for Flyback Charging
The simulation and the experimental results concerning Circuit”, WSEAS Transactions on Circuits and Systems, vol.6, No.2,
the output voltage are compared. These experimental results pp.208-214, 2007.
can verify the modeling and analysis in this paper very well. [14] E. Thomas and A. J. Rosa, The Analysis and Design of Linear Circuits,
5th ed., John Wiley & Sons, New York, 2006.
[15] A.B.Yildiz, “A MNA-Based Unified Ideal Switch Model for Analysis
V. CONCLUSION of Switching Circuits”, Journal of Circuits, Systems and Computers,
Flyback Converter is used in various applications due its Vol. 22, No. 6, pp.1-12, July 2013.
less complexity, cost effectiveness, multiple output and
isolation. In this paper, the flyback transformer, the most
____________________________________________________________________________________________________________
495
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
cceyhunyildiz@gmail.com
Abstract — The correction of power factor in electric power mathematical models when a synchronous motor is used for
systems is called reactive power compensation. A synchronous power factor correction, a fuzzy logic based compensation
motor is used as a capacitive reactive power generator in control system was used in this study.
compensation systems. It is less costly for an enterprise to use a
synchronous motor as both mechanical power generator and II. POWER FACTOR
power factor corrector, which increases their efficiency. There are
various studies on increasing the efficiency, capacity and stability Reactive power compensation plays an important role in the
of a power system using power factor correction under different improvement of efficiency and capacity of electric power
operating conditions. This study focuses on the power factor systems. The current of inductive power consumers is supplied
correction of the system by controlling the excitation current of by two components of the current. The former is active current
the synchronous motor via fuzzy logic thanks to the asynchronous converted to work while the latter is reactive current which
motor connected to the system. creates the necessary magnetic field for electric machinery and
devices.
Keywords — Power factor, Fuzzy Logic, Synchronous Motor Types of power corresponding to these currents are:
S = Apparent Power (VA),
I. INTRODUCTION P = Active Power (W),
As a result of technological improvements, energy Q = Reactive Power (On), then
consumption has increased in recent years due to the increasing
use of inductive loads in industrial applications. Besides active
power, inductive loads also absorb reactive power from the grid S P 2 Q2 (1)
[1].Although reactive power absorbed from the transmission
line loads it, it cannot be converted to the energy [2]. Therefore,
reactive power absorbed from the grid causes losses in electric the equation above can be obtained, which can be geometrically
power systems, and these losses must be minimized. The defined as a power triangle as shown in Fig.1.
minimization of energy losses will reduce cable and other
measurement and protection costs, thus creating a more cost-
effective electric power system. This can only be achieved
when reactive power needed by the inductive loads, which the
transmission line feeds, is supplied to the load as closely as
possible. Reactive power needed by the loads is supplied
statically by a capacitor or reactor and dynamically by a
synchronous motor [3-5]. Reactive power compensation via a
synchronous motor can be achieved by changing the excitation Fig. 1 Power triangle [11].
current of the motor if the motor operates in a capacitive or
inductive character [6]. In addition, the amount of reactive Here, φ is defined as the phase angle. Cos φ is the power
power that a synchronous motor absorbs from the grid can be factor, which is defined as follows [12].
adjusted thanks to the excitation current. An efficiently P
compensated system will improve the power factor, minimize cos (2)
S
losses and become efficient [7-9].
This study focuses on the power factor correction of an III. SYNCHRONOUS MOTOR
electric power system using a fuzzy logic based and excitation
Synchronous motor is an alternative current motor in which
current controlled synchronous motor. Because it is difficult to
rotor rotational speed is equal to the rotational speed of the
determine control parameters via methods requiring
stator rotating field and the rotation speed does not vary in
____________________________________________________________________________________________________________
496
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
497
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Input variables of the fuzzy logic control unit designed for this
study were determined as error (e), Q (reactive power) and
output variable (IE). Error (e) is found when power factor value
is subtracted from input reference power factor value as given
in equation 4.
Fig.5 Changes in the excitation current and reactive power that the Fig.7 Membership functions designed for error
synchronous motor absorbs from the grid
____________________________________________________________________________________________________________
498
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Table II. Asynchronous motor parameters As shown in Fig.10, the synchronous motor reached a
synchronous speed after it moved for 0.9 seconds. Afterwards,
Parameter Value an excitation current of 5 percent is applied to the synchronous
Nominal Power(P) 3 kW motor via fuzzy controller, which supplied reactive power in a
Nominal Revolution[n] 1430 rpm capacitive character to the power system of the synchronous
motor. Thus, the synchronous motor compensated itself and
Nominal Voltage(V) 400 V
fixed the power factor of the system at 0.98, which was 0.8 at
Nominal Current(I) 6.7 A the beginning.
Nominal Load Torque(M) 19 Nm
Pole pairs(p) 2 VII. CONCLUSION
Frequency(f) 50 Hz In the present study, a fuzzy logic control method was used in
Rotor Type Squirrel-cage order to create a more sensitive power system performance and
Stator Resistance[Rs] 1.45 ohm increase system efficiency instead of a conventional control
Stator Inductance[L] 12.2 mH system requiring mathematical models for the calculation of
controller parameters. During the operation of control system,
Rotor Resistance[Rr’] 1.93 ohm
the excitation current of the synchronous motor, the reactive
Rotor Inductance[Lr’] 2.66 mH power that it absorbs from the grid and its power factor were
Mutual Inductance[Lm] 187.8mH constantly observed. These values were evaluated in the rule
Friction Factor[F] 0.03 N.m.s table, and it was aimed to more effectively apply the change in
Mechanical Inertia[J] 0.03 kg.m2 the excitation current of synchronous motor to the system.
Contrary to previous studies, this new fuzzy logic control
____________________________________________________________________________________________________________
499
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Bayram M., Reactive Power Compensation in Power Installations ,August
2000, Birsen Publisher.(Original in Turkish).
[2] Mekhamer, S.F., El-Hawary, M.E., Mansour, M.M., Moustafa, M.A., &
Soliman, S.A., “State of the art in optimal capacitor allocation for
reactive power compensation in distribution feeders”, IEEE Large
Engineering Systems Conference on Power Engineering,
LESCOPE 02, 2002, 61-75.
[3] Bal G., Colak İ., “Reactive power compensator using constant capacitor
and thyristor controlled reactor”, Gazi University, J. of Institute of
Science and Technology, vol. 8, No. 2, 1995, p. 9-15.
[4] P. W. Sauer, “What is Reactive Power?,” Power Systems Engineering
Research Center, PSERC Background Paper, Department of Electrical
and Computer Engineering, University of Illinois at Urbana-Champaign,
September 16, 2003.
[5] Miller TJE., Reactive Power Control in Electric Systems, October 1982,
A Viley-Interscience Publication, New York, p.182-222.
[6] Al-Hamrani, M.M., Von Jouanne, A., & Wallace, A.,“Power factor
correction in industrial facilities using adaptive excitation control of
synchronous machines”, Pulp and Paper Industry Technical Conference,
Conference Record of the 2002 Annual, 148–154.
[7] Bayındır R., Sagiroglu S., Colak I, “Artificial neural network-based
reactive power compensation”, Gazi University Faculty of Technical
Education, Journal of Politeknik, 2007, 10(2),129-135.(Original in
Turkish).
[8] Colak, I., Bayindir R., Bay, O.F.: ‘Reactive Power Compensation Using
A Fuzzy Logic Controlled Synchronous Motor’, Energy Conversion and
Management, 2003, 44 (13), 2059-2215..
[9] Sagiroglu S., Colak I, Bayındır R., “Power Factor Correction Technique
Based on Artificial Neural Networks”, Energy Conversion and
Management, 2006, Vol. 47/(18-19), 3204-3215.
[10] Şekkeli M.,“Using the basic components in a account of the required
power for precise reactive power compensation”, KSÜ, Journal of
Science and Engineering, 2006, 9(2), 66-70.(Original in Turkish).
[11] http://www.bilgiustam.com/What is the reactive power compensation?/
(Date of access: December 22, 2015). (Original in Turkish).
[12] Şekkeli M.,Tarkan M.,“The minimum number of switching and optimal
reactive power selection in the reactive power control relay”, İTÜ,
İTÜ/d Dergisi, 2005, 4(6), 15-22. (Original in Turkish).
[13] Bayındır, R. Görgün, A., “A compensator application with PIC-based a
PI controller using a synchronous motor”, Pamukkale University
Engineering Faculty, Journal of Engineering Science,2009, 15(1), 87-99.
(Original in Turkish).
[14] Çolak İ., Bayındır R., Görgün, A., Kundakoğlu H., “Observing the
different work situations in synchronous motors”, IV. Automation
Symposium2007, Samsun. (Original in Turkish).
[15] Sekkeli, M., Açikgoz H., Gani A., Keçecioglu O.F., “Simulation study
of permanent magnet synchronous motor with self-tuning fuzzy-PID
controller and speed control.” KSÜ Journal of Engineering Science,
2015, 18(1), 17-22. (Original in Turkish).
[16] Ozcalik, H.R., Kiliç, E., Yılmaz, S., Gani, A., “Investigation of control
performance of different membership functions in fuzzy logic based
liquid level control”,TOK2013, 26-28 September 2013, Malatya, S.243-
247. (Original in Turkish).
[17] Gani,A., Açikgoz, H., Kiliç, E., Sit, S., 2014. “Fuzzy logic based control
of inverted pendulum”, ELECO 2014, Bursa. (Original in Turkish).
[18] Çolak İ., Bayındır R., Sefa, İ., “Experimental study on reactive power
compensation using a fuzzy logic controlled synchronous motor” Energy
Conversion and Management 45 (2004) 2371–2391.
[19] Bayındır, R., (2002), Reactıve Power Compensatıon Using A Fuzzy Logic
Controlled Synchronous Motor. Ph.D.Thesis, Gazi University, Institute
of Science and Technology. (Original in Turkish).
____________________________________________________________________________________________________________
500
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Delta and Sigma-Delta modulation methods have been software-based generators (e.g., linear congruential generators
getting a great interest recently due to the great progress in [26]), chaotic systems are very simple to realize and offer a
analog-digital very large scale integration technology. Since the hybrid structure with the features of hardware and software
outputs of these methods are digital, the data can be securely based approaches [27]–[30]. The number of the chaotic
encrypted using very simple standard hardware. In this work, a
systems have been increased over time in the literature, which
chaotic random bit generator based approach for encrypting
digital data of the delta and delta-sigma modulators is studied. allows us to benefit from chaotic dynamics for generating
The chaotic bit generation can easily be implemented in the efficient chaotic random bits for use in cryptographic
digital hardware of the modulators due to simplicity of the applications [31]–[35].
chaotic dynamics. The randomness of the generated chaotic bits In this work, a chaotic random bit generator is developed
are proved with visual and statistical tests. The security of the and integrated into the Δ and ΔΣ modulators for data
proposed approach is evaluated via key space estimation based encryption. The goal is to provide security in such systems
attacks. The efficiency of the methods is validated with during data transmissions. The Δ modulation systems offer
simulations. low cost solutions with a strong immunity against crosstalk
and noise in the transmission line, and integration of the chaos
Keywords—Chaos, delta modulation, delta-sigma modulation, into these systems will enhance reliability and security.
random bits, cryptography, communication In the following sections, the Δ modulation methods are
overviewed and a chaotic map based encryption scheme is
applied to digital outputs of the Δ and ΔΣ encoders.
I. INTRODUCTION
The delta (Δ) and delta-sigma (ΔΣ) modulators offer simple, II. CHAOTIC DATA ENCRYPTION FOR ∆ AND ΔΣ MODULATION
efficient methods for telecommunication and signal Chaotic systems are able to provide diffusion and confusion,
processing applications. The Δ modulation systems have i.e., hiding and spreading plaintext over the ciphertext, and for
gained significance in recent years due to their very simple this reason have potential applications in some functional
hardware structure, digital transmission and easy to add blocks of communication systems including encryption,
adaptive features. In addition ΔΣ modulators have noise modulation and compression. By considering a delta
shaping feature that makes them well-suited for low- modulation scheme, the chaotic random bits can easily be
frequency, high-accuracy measurements. There are many used for encrypting digital plaintext for secure
applications of Δ modulators including reliable voice communications. Figure 1 shows a chaos based digital data
communications, analog-to-digital signal conversion, encryption and decryption approach for Δ modulators.
performing audio delay lines, telemetry systems and feedback
power control in code-division multiple-access radio
u x vi wi vi' u'(t)
communication systems [1]–[5]. The Δ modulation systems, a XOR XOR LPF
type of variable structure control, are also getting a special +
- hi hi
interest in the control community [6]–[10]. Other recent n Binary Binary
studies on Δ modulators have been focused on multibit Convertor Convertor
modulation, chaotic modulation, chaotification and tone θi θi
suppression in communications [11]–[17]. θ0 Chaotic Chaotic
Map Map θ0
The digital output of the Δ modulation systems can be
encrypted by using chaotic systems. Since chaotic dynamics
Fig. 1 Chaos based encryption for delta modulation system
have strong similarities with the cryptography, e.g.
aperiodicity, deterministic dynamics, ergodicity and
sensitivity to initial conditions, they have recently been Similarly, the chaotic bits can also be incorporated into
utilized in cryptosystems [18]–[23]. To encrypt the Δ digital output of the ΔΣ modulator and demodulator systems
modulation systems, while the required random bits can be for securing the data as illustrated in Fig. 2. In data encryption
generated from a hardware-based generator (e.g., using the exclusive-or (XOR) logical function is used. The Δ
thermal noise [24] and radioactive decay [25]) or from modulated signals are easily demodulated at the receiver by
____________________________________________________________________________________________________________
501
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
using a low pass filter, but now with the chaotic encryption, it
is not possible to extract message without correct chaotic L xx x u sign( x ) u x (3)
decrypter with correct initial conditions and parameter values.
Thus, the modulator is stable if the following condition holds
u x vi wi vi' u'(t)
XOR XOR LPF max u (t ) (4)
+
- hi hi
Binary Binary
Convertor Convertor The equivalent condition for the discrete-time (sampling)
θi θi implementations is given by
Chaotic Chaotic
θ0 θ0
Map Map f s max u (t ) (5)
Fig. 2 Chaos based encryption for delta-sigma modulation system
where is the sampling frequency. Equation (5) shows that
the Δ modulator produces a binary coded output from the
A. Δ and ΔΣ Modulators time-derivative of the analog input signal.
For ΔΣ modulators seen in Fig. 2, since the input signal
Today, digital techniques are dominating signal processing. first passes through an integrator, then the governing equation
The Δ modulation systems have an important place in the can be written as
digital signal processing area with simple, efficient solutions.
For example the ΔΣ snalog-to-digital converters are ideal for
x(t ) u(t ) sign( x(t )) (6)
many applications whose signal frequencies vary from dc to
several hundred megahertz. These approaches are composed
of an oversampling modulator followed by a digital filter that Similar to (2), if the Lyapunov stability is applied, one can
together generates a high-resolution digital data streams. easily obtain the following stability condition
Typically, a Δ modulator with one bit quantization requires a
resolution on the order of 14–20 bits (e.g., around 100 kbit/s f s max u (t ) (7)
for a voice bandwidth of 4 kHz). The main working principle
of the Δ modulator is illustrated in Fig. 3. The modulator is Therefore, the ΔΣ modulators have noise suppression
simply a sampled data system employing a negative feedback advantage compared to Δ modulators because the quantization
loop via integration. A one-bit quantizer (or comparator) level is proportional to the amplitude of the input signal (while
senses if the instantaneous level of the analog input is greater is proportional to the derivative of the input signal in Δ
or less than the feedback signal and produces a continuous modulators). This noise-shaping feature of the ΔΣ modulators
non-return-to-zero digital data stream. The negative feedback is well suited to signal processing applications, e.g.
loop integrates the digital data to form an approximation of communication and digital audio.
the input signal. It is also quite simple to demodulate the input
signal by using an identical integrator and a low pass filter
(LPF). The ΔΣ modulator has a simpler structure at the
demodulator by having only a low pass filter. u(t)
By considering Fig. 1, the equations of the delta modulator
are given by n(t)
____________________________________________________________________________________________________________
502
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
satisfied for the related modulator type. There exist adaptive The randomness features of (10) can be evaluated with the
algorithms for adjusting quantization step size to eliminate visual and test statistics based methods. Visualization is a
quantization errors of the Δ modulator systems. On the other quick way to get rough information about the chaotic random
hand, the performance of these modulators is dependent on the sequences. The bifurcation diagram and the histogram plot are
quantization and channel noise. The quantization noise used for visual evaluations. Figure 5 displays a histogram plot
averages to zero and can be defined by its root mean square of the chaotic map for 100 categories. The histogram plot
(rms) value for a dynamic input signal by considering its shows a uniform distribution over the ±1 range. This means
limits ± /2. Thus, the quantization error ( ) is given by that the chaotic map with the selected parameter value
generates a uniformly distributed random sequence.
/2
1
xrms x 2 (t )dx(t ) (8)
/2 12
u urms
SNR(dB) 20 log rms 20 log 12 (9)
xrms
____________________________________________________________________________________________________________
503
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
frequency within a block (or block frequency) tests [38]–[40]. eavesdropping system is assumed to be estimated with a very
These tests are used to determine whether the chaotic random small initial condition error, e.g., = − = 1x10 , and
bits are unbiased and uncorrelated. That is to say, the chaotic it is seen in Fig. 6e that the message cannot be recovered. The
random bits ℎ with security bound S bits should include chaotic maps in the transmitter and receiver must be exactly
unbiased bits (probability of 0 and 1 must be equal) and the same with the same parameters and initial conditions to
undistinguished bits without performing at least 2 operations generate the correct key sequence and to decrypt the correct
[41]. Even though the statistical tests alone are not enough for message.
such evaluations, it is nice to see that the chaotic bits pass all
these statistical tests. Note that for practical applications, 1.5
application specific tests are usually carried out for 1
randomness analysis. The statistical test results are given in 0.5
ui
Table I. It is clear that all tests are successfully passed, and for 0
this reason, the robust chaotic map (10) with the binary -0.5
0 1 2 3 4 5 6
converter algorithm (11) produces a highly-satisfactory (a) Time (s)
random bits for use in cryptosystems. 1.5
1
0.5
vi
TABLE I
STATISTICAL EVALUATION OF THE CHAOTIC RANDOM BITS 0
-0.5
Test Name Test Values Statistics Result 0 1 2 3 4 5 6
Monobit < 3.8415 1x10-6 success (b) Time (s)
Block frequency > 0.01 0.886 success 1.5
0.5
0
-0.5
III. SIMULATION RESULTS 0 1 2 3 4 5 6
(d) Time (s)
The Matlab/Simulink based numerical simulation results 1.5
are given in Figs. 6-7. Figure 6 shows the chaotic data 1
encryption in Δ modulator based data transmission results.
ui´
0.5
The message signal which includes ASCII codes of “chaos” is 0
seen in Fig. 6a. The digital output of the Δ modulator is -0.5
displayed in Fig. 6b, where there are some windows in the 0 1 2 3 4 5 6
(e) Time (s)
modulated signal. This digital signal can easily be
Fig. 6 (a) Message bits. (b) Digital delta modulator output. (c) Chaotic bits
demodulated with an integrator and low-pass filter. To encrypt based encrypted message (transmitted bits). (d) Recovered message bits. (e)
the modulator output bits, the XOR logic function is used for Recovered message bits for a wrong chaotic key sequence.
the chaotic bit sequence and the modulator bits, namely,
Similarly, the numerical simulation results for the ΔΣ
wi vi hi (12) modulator based data transmission are given in Fig. 7. The
waveform of the first-order ΔΣ modulator is illustrated in Fig.
Figure 6c shows the encrypted modulator output bits, 7a when the input signal is a sinusoid as given in Fig. 7a. It is
which does not have any visual pattern. The same chaotic bit should be noted that the modulator performs both the
generator is used in the demodulator to decrypt the original sampling and the quantization operation in this example,
modulator output bits and then Δ demodulator is employed to which is typical in practical circuit implementations. The ΔΣ
extract the original message signal. The recovered message modulator output shows that the output is either plus or minus
signal is seen in Fig. 6d. which shows a perfect recovery. The full scale and when the sinusoidal input to the modulator is
security of the scheme is tested with the use of Δ demodulator, close to the full scale, the output is either positive or negative
low pass filtering and estimated key sequence based tests. In during the cycle. It is seen that the local average of the
Fig. 6e, the test result for a wrong chaotic key sequence is modulator output follows the input signal. When the input
illustrated. In this test, the same chaotic map and binary signal is around zero, the modulator output changes fast
convertor algorithm is used. The initial condition of this between ± with nearly zero mean. Therefore, the input
signal can easily be recovered by using a low pass filter as
____________________________________________________________________________________________________________
504
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
demodulator. On the other hand, the chaotic bits based chaos theory. The randomness of the robust chaotic map based
encryption of the modulator output removes all the modulator random bits are evaluated with qualitative statistical tests.
patterns as seen in Fig. 7c. The recovered input signal is seen The security of the delta modulation systems under chaotic
in Fig. 7d. which shows an excellent recovery. To evaluate encryption is tested with low pass filtering and key space
security of the approach, the test result for a wrong chaotic estimation based attacks, and it is shown that the methods are
key sequence is shown in Fig. 7e. In this test, the same chaotic highly secure and reliable. The use of chaos in securing delta
map and binary convertor algorithm is used with a very small modulation approaches provides a good option to be
initial condition error, e.g., = − = 1x10 . It is clear considered as a framework for the next generation
from Fig. 7e that the message cannot be recovered with such a communication and data transmission systems.
very small estimation error.
ACKNOWLEDGMENT
1
This work was supported by Research Fund of the
0 Abdullah Gül University under project number FAB-2015-4.
ui
-1
REFERENCES
0 1 2 3 4 5 6
(a) Time (s) [1] D. G. Zrilic, Circuits and Systems Based on Delta Modulation: Linear,
Nonlinear and Mixed Mode Processing. Springer, 2006.
1.5
[2] J. Liberti and T. S. Rappaport, Smart Antennas for Wireless
1 Communications: IS-95 and Third Generation CDMA Applications, 1
0.5 edition. Upper Saddle River, NJ: Prentice Hall, 1999.
vi
0.5
0 prediction for access control in integrated voice/data CDMA systems,”
IEEE J. Sel. Areas Commun., vol. 18, no. 1, pp. 112–122, Jan. 2000.
-0.5
0 1 2 3 4 5 6 [6] H. Sira-Ramírez, Sliding Mode Control: The Delta-Sigma Modulation
(c) Time (s) Approach, 1st ed. Basel: Birkhäuser, 2015.
[7] T. Hu, Z. Lin, and L. Qiu, “Stabilization of exponentially unstable
1 linear systems with saturating actuators,” IEEE Trans. Autom. Control,
vol. 46, no. 6, pp. 973–979, Jun. 2001.
ui´
____________________________________________________________________________________________________________
505
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[18] I. Hussain and M. A. Gondal, “An extended image encryption using [30] G. Alvarez and S. Li, “Some Basic Cryptographic Requirements for
chaotic coupled map and S-box transformation,” Nonlinear Dyn., vol. Chaos-Based Cryptosystems,” Int J Bifurc Chaos Appl Sci Eng, vol. 16,
76, no. 2, pp. 1355–1363, Jan. 2014. no. 8, p. 2129, 2006.
[19] V. Lynnyk, N. Sakamoto, and S. Čelikovský, “Pseudo random number [31] S. H. Strogatz, Nonlinear Dynamics and Chaos: With Applications to
generator based on the generalized Lorenz chaotic system,” IFAC-Pap., Physics, Biology, Chemistry, and Engineering, Second Edition, Second
vol. 48, no. 18, pp. 257–261, 2015. Edition edition. Boulder, CO: Westview Press, 2014.
[20] M. Park, J. C. Rodgers, and D. P. Lathrop, “True random number [32] J. C. Sprott, Chaos and Time-Series Analysis, 1 edition. Oxford ; New
generation using CMOS Boolean chaotic oscillator,” Microelectron. J., York: Oxford University Press, 2001.
vol. 46, no. 12, Part A, pp. 1364–1370, Dec. 2015. [33] G. Ablay, “Chaotic map construction from common nonlinearities and
[21] I. Cicek, A. E. Pusane, and G. Dundar, “A novel design method for microcontroller implementations,” Int. J. Bifurc. Chaos, vol. to be
discrete time chaos based true random number generators,” Integr. appear, 2016.
VLSI J., vol. 47, no. 1, pp. 38–47, Jan. 2014. [34] G. Ablay, “Novel chaotic delay systems and electronic circuit
[22] L. Kocarev and S. Lian, Chaos-based cryptography theory, algorithms solutions,” Nonlinear Dyn., vol. 81, no. 4, pp. 1795–1804, May 2015.
and applications. Berlin: Springer, 2011. [35] G. Ablay, “Chaos in PID Controlled Nonlinear Systems,” J. Electr. Eng.
[23] R. F. Martínez-González, J. A. Díaz-Méndez, L. Palacios-Luengas, J. Technol., vol. 10, no. 4, pp. 1843–1850, 2015.
López-Hernández, and R. Vázquez-Medina, “A steganographic method [36] D. J. W. Simpson, “On the relative coexistence of fixed points and
using Bernoulli’s chaotic maps,” Comput. Electr. Eng., 2016. period-two solutions near border-collision bifurcations,” Appl. Math.
[24] D. C. Ranasinghe, D. Lim, S. Devadas, D. Abbott, and P. H. Cole, Lett., vol. 38, pp. 162–167, Dec. 2014.
“Random numbers from metastability and thermal noise,” Electron. [37] A. S. Mansingka, M. Affan Zidan, M. L. Barakat, A. G. Radwan, and
Lett., vol. 41, no. 16, pp. 13–14, Aug. 2005. K. N. Salama, “Fully digital jerk-based chaotic oscillators for high
[25] J. Walker, “HotBits: Genuine Random Numbers,” 2016. [Online]. throughput pseudo-random number generators up to 8.77 Gbits/s,”
Available: https://www.fourmilab.ch/hotbits/. [Accessed: 12-Mar- Microelectron. J., vol. 44, no. 9, pp. 744–752, Sep. 2013.
2016]. [38] G. Marsaglia, “Random Number Generators,” J. Mod. Appl. Stat.
[26] D. P. Kroese, T. Taimre, and Z. I. Botev, Handbook of Monte Carlo Methods, vol. 2, no. 1, May 2003.
Methods, 1 edition. Hoboken, N.J: Wiley, 2011. [39] A. J. Menezes, P. C. van Oorschot, and S. A. Vanstone, Handbook of
[27] İ. Öztürk and R. Kılıç, “A novel method for producing pseudo random Applied Cryptography. CRC Press, 1996.
numbers from differential equation-based chaotic systems,” Nonlinear [40] D. Naccache, Cryptography and Security: From Theory to Applications:
Dyn., vol. 80, no. 3, pp. 1147–1157, Feb. 2015. Essays Dedicated to Jean-Jacques Quisquater on the Occasion of His
[28] N. Romero, J. Silva, and R. Vivas, “On a coupled logistic map with 65th Birthday. Springer, 2012.
large strength,” J. Math. Anal. Appl., vol. 415, no. 1, pp. 346–357, Jul. [41] J. Katz and Y. Lindell, Introduction to Modern Cryptography:
2014. Principles and Protocols, 1 edition. Boca Raton: Chapman and
[29] X. Wang and X. Bao, “A novel block cryptosystem based on the Hall/CRC, 2007.
coupled chaotic map lattice,” Nonlinear Dyn., vol. 72, no. 4, pp. 707–
715, Jan. 2013.
____________________________________________________________________________________________________________
506
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— This work presents a performance comparison of phase The PSFB-PWM converter can operate with SS providing
shifted full bridge (PSFB) converter and inductor-inductor- phase delay for the control signals of the cross-conducted
capacitor (LLC) resonant converter for on-board electrical primary switches of the converter. However, this soft switching
vehicle (EV) battery charge applications. In the comparison, process is affected by load conditions and dead time [12], [13].
lithium-ion battery cells are taken into consideration to evaluate
The resonant converters operates naturally with soft switching
the performance of converters because of their high power
density. In the performance evaluation, for both converter, soft and their soft switching operation doesn’t depend on load
switching operation of the primary semiconductors is discussed to condition. However, the resonant converter can regulate the
provide output requirements of the battery. The performance of output voltage with the change of switching frequency.
both converter are also validated experimentally by two Moreover, they have also high conduction loss due to increased
prototypes, operating with 810 W maximum output power. rms currents. The LLC resonant converter topology provides
Experimental results are obtained in good match with theoretical the narrow switching frequency compared to series, parallel
results. According to obtained results, PSFB converter works with and series-parallel resonant converter topologies [14].
soft switching from 50 V to 54 V output voltage, with constant Therefore, it is usually preferred for battery charge applications
output current, at 200 kHz switching frequency. LLC resonant
[9]-[11], [15] and [16]. It seems both PSFB PWM converter and
converter works with soft switching from 42 V to 54 V but with
changing switching frequency between 150 kHz and 200 kHz. LLC resonant converter have some advantages and
disadvantages. Therefore, the performance of PSFB converter
and LLC resonant converter needs to be discussed for battery
Keywords— On-board battery charge, PSFB converter, LLC
resonant converter, li-ion battery charge. charge applications.
The lithium-ion battery charge process has wide range load
I. INTRODUCTION and output voltage condition. Therefore, this work discusses the
The use of electrical vehicle (EV) increases greatly every performance of PSFB PWM converter and LLC resonant
day since the fossil fuels run out, rapidly. Therefore, EV converter based on their soft switching operation according to
provides benefits in the area of the economy for the most lithium-ion battery charge characteristic. For this purpose, two
country in the world. The battery charge of vehicles can be experimental setup are built to validate presented theoretical
possible from a battery charge station or directly a power outlet evaluations. The built setups were tested for 42-54 V output
with on-board chargers. On-board battery chargers can voltage range with 810 W maximum output power.
encourage the user to use EV giving flexibility to charge
II. THE PRINCIPLES OF LLC RESONANT AND PSFB PWM DC-
vehicle’s battery from any power point [1]. However, the
DC CONVERTERS
volume, weight and the cost of the EV increase with on-board
chargers [2]-[6]. A. LLC Resonant Converter
The lithium-ion battery cells have high voltage and current
A circuit diagram of LLC resonant converter is shown in Fig.
rate so their power density is higher compared to the other
1. In the diagram, resonant tank composed of Lr, Lm and Cr. S1-
battery cells [7], [8]. However, larger voltage range of the
S4 are the primary MOSFETS, DR1 and DR2 are the rectifier
battery cell brings the need of wide range output voltage
diodes producing dc output voltage. The output voltage is v0,
regulation for the converter. An on-board battery charger is
the output capacitor filter is Co, the load resistor is Ro, the input
composed of two stages which are AC –DC conversion stage
voltage is vin and TR represents the transformer with N turns
and DC-DC conversion stage. This work focuses on the second
ratio.
stage of a battery charger. The second stage usually uses soft
The voltage gain characteristic of the converter based on
switching converters to provide high efficiency and high power
normalized frequency, fs/fr1, and quality factor, Q, variation is
density. The phase shifted full bridge (PSFB), pulse width
given in Fig. 2 [17].
modulated (PWM) converter and resonant converters are good
candidates, and they are mostly used for the second stage of the
battery charger [1], [9]-[11].
____________________________________________________________________________________________________________
507
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
MOSFET (S3 or S2) are switched on. During the output current
commutation, both rectifier diode stays on condition and the
secondary side of the transformer is short circuited. In the PSFB
PWM DC-DC converter, the switching losses are reduced
significantly by allowing a delay time between the control states
of two diagonal switches. During this delay time, the parasitic
capacitors of MOSFETs in one leg are charged/discharged, and
ZVS turn-on of primary MOSFETs is achieved.
Fig. 1. The circuit diagram of LLC resonant converter.
3
Q=0.2
Open Circuit
Q=0.25
Gain
Q=0.3
2.5
Q=0.35
Region 2 Region 1
ZVS ZVS Q=0.4
Q=0.5
2 Q=0.68
Q=0.82
Gain=nVo/Vin
Q=1.5
1.5 Q=2.5
Q=4.82
Region 3
ZCS
1
____________________________________________________________________________________________________________
508
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
III. PERFORMANCE COMPARISON The PSFB PWM converter was operated at 200 kHz
For the comparison study, design of each converter is switching frequency to test its performance. The SS operation
evaluated based on wide range output voltage regulation and of leading leg MOSFETs is usually achieved with the reflected
soft switching operation for the second stage of a battery load current to the primary side. The lagging leg MOSFETs
charger. At the output, series connection of 14 lithium-ion usually have problem due to their dependence on load and dead
battery cell is assumed for both converter. Thus, the output of time condition. Therefore, the measured results are given in Fig.
the battery package has 43.4-53.9 V / 15 A with that assumption. 5 for lagging leg MOSFET’s switching. Fig. 5 (a) shows the SS
The input of converters are fed with 385 V dc voltage obtained operation of S2 MOSFET with 54 V / 15 A output. The
from a dc voltage source. The battery charge control has two converter can operate with SS until 48 V with full load due to
modes which are constant current mode and constant voltage no optimized dead time then primary switches work with HS as
mode. In this work, performance of each converter is evaluated shown in Fig. 5 (b).
based on constant current mode with changing battery voltage. TABLE II
The operation analysis of the converters are performed by THE COMPONENTS USED IN PSFB PWM CONVERTER AND LLC RESONANT
CONVERTER
Matlab and obtained results are summarized in Table I.
TABLE I PSFB PWM LLC Resonant
Components
PERFORMANCE COMPARISON OF LLC RESONANT AND PSFB PWM Converter Converter
CONVERTERS.
E/65/32/27 E/65/32/27
Transformer
Operation of Np=12, Ns=2 Np=16, Ns=2
Turns
Switching Primary
Converter Ratio Snubber Lr=23.7µH, Cr=32.9nF,
Frequency MOSFETs in Ls=20 µH
Topology of Components Lm=75 µH
Range Constant Current
TR, N
Mode Lo=1.1 µH,
LLC Resonant Output Filters Co=12x22 µF
Co=12x22 µF
DC-DC 8 150-200 kHz SS From 42-54V
Converter Primary
C2M0080120
PSFB PWM MOSFETs S1-S4
HS From 42-48 V
DC-DC 6 200 kHz Secondary
SS From 50-54 V DSS2x101-015A
Converter Rectifiers DR1-DR2
____________________________________________________________________________________________________________
509
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] B. Whitaker, A. Barkley, Z. Cole, B. Passmore, D. Martin, T. R. McNutt,
A. B. Lostetter, J. S. Lee, and K. Shiozaki, “A high-density, high-
efficiency, isolated on-board vehicle battery charger utilizing silicon
carbide power devices”, IEEE Transactions on Power Electronics, vol.
29, pp. 2606-2617, May 2014.
[2] M. Yilmaz and P. T. Krein, “Review of battery charger topologies,
charging power levels, and infrastructure for plug-in electric and hybrid
vehicles”, IEEE Transactions on Power Electronics, vol. 28, no. 5, May
(b) 2013.
[3] M. Grenier, M. H. Aghdam, and T. Thiringer, “Design of on-board
Fig. 5 The measured results for lagging leg switch of PSFB converter. (a) The charger for plug-in hybrid electric vehicle,” in Proc. Power Electronics,
output is 54 V/15 A. (b) The output is 48 V /15 A. Machine and Drives, pp. 1–6, 2010.
[4] S. Haghbin, K. Khan, S. Lundmark, M. Alak¨ula, O. Carlson, M.
Leksell, and O. Wallmark, “Integrated chargers for EV’s and PHEV’s:
The efficiency values as function of battery voltage were Examples and new solutions,” in Proc. Int. Conf. Electrical Machines,
pp. 1–6, 2010.
measured and obtained results are shown in Fig. 6. The [5] A. Emadi, Y. J. Lee, and K. Rajashekara, “Power electronics and motor
obtained results show that the efficiency of PSFB PWM drives in electric, hybrid electric, and plug-in hybrid electric vehicles,”
converter is higher than the LLC resonant converter although IEEE Trans. Ind. Appl., vol. 55, pp. 2237–2245, Jun. 2008.
the soft switching feature of PSFB PWM converter is not [6] A. Emadi, S. S.Williamson, and A. Khaligh, “Power electronics
intensive solutions for advanced electric, hybrid electric, and fuel cell
maintained below 50 V battery voltage. Because the conduction vehicular power systems,” IEEE Trans. Power Electron., vol. 21, pp.
loss of LLC resonant converter is higher due to its increased 567–577, May 2006.
rms currents. The maximum efficiency of PSFB converter and [7] M. Chen and G. A. Rinc´on-Mora, “Accurate, compact and power-
LLC resonant converter is 94.4 % and 93.6%, respectively. efficient Li-ion battery charger circuit,” IEEE Trans. Circuits Syst. II,
Exp. Briefs, vol. 53, pp. 1180–1184, Nov. 2006.
[8] S. Dearborn, “Charging Li-ion batteries for maximum run times,” Power
Electron. Technol. Mag., vol. 31, pp. 40–49, Apr. 2005.
[9] F. Musavi, M. Craciun, D. S. Gautam, W. Eberle and W. G. Dunford,
“An LLC resonant DC–DC converter for wide output voltage range
battery charging applications”, IEEE Transactions on Power
Electronics, vol. 28, pp. 5437-5445, December 2013.
[10] J. Deng, S. Li, S. Hu, C. C. Mi and R. Ma, “Design methodology of LLC
resonant converters for electric vehicle battery chargers” IEEE
Transactions on Vehicular Technology, vol. 63, pp. 1581-1592, May
2014.
[11] Z. Fang, T. Cai, S. Duan, and C. Chen, “Optimal design methodology
for LLC resonant converter in battery charging applications based on
time-weighted average efficiency”, IEEE Transactions on Power
Electronics, vol. 30, pp. 5469-5483, October 2015.
[12] A. F. Bakan, N. Altintas, and I. Aksoy, “An improved PSFB PWM DC–
DC converter for high-power and frequency applications,” IEEE Trans.
Power Electron., vol. 28, pp. 64-74, Jan. 2013.
[13] S. Cetin, “High efficiency design considerations for the self-driven
synchronous rectified phase shifted full bridge converters of server
power systems, Journal of Power Electronics, vol.15, pp.634-643, May
Fig. 6 The compared efficiency values of PSFB and LLC resonant converter as 2015.
function of battery voltage. [14] L. R. Steigerwald, “A comparison of half bridge resonant converter
topologies”, IEEE Trans. on Power Electronics, vol. 3, pp. 174-182,
V. CONCLUSIONS April 1988.
[15] Y. S. Dow 1, H. I. Son 1, and H. D. Lee, “A study on half bridge LLC
In this work, soft switching operation of PSFB PWM resonant converter for battery charger on board, IEEE 8th International
converter and LLC resonant converter was discussed based on Conference on Power Electronics and ECCE Asia (ICPE & ECCE), pp.
lithium-ion battery charge profile for on board EV battery 2694 – 2698, 2011.
____________________________________________________________________________________________________________
510
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[16] Y.S. Dow, H.H. Kim, Y.I. Kwon, B.Y. Kim, J.C. Kim, “A study of 6.6 [19] Y. Fang, D. Xu, Y. Zhang, F. Gao, L. Zhu, “Design of high power
kW on board charger for electric vehicle”, EVS28 International Electric density LLC resonant converter with extra wide input range”, Applied
Vehicle Symposium and Exhibition, pp. 1-8, 3-6 May 2015. Power Electronics Conference, pp. 976-981, 2007.
[17] S. Cetin, “High efficiency design approach of a LLC resonant converter [20] D. Huang, D. Gilham, W. Feng, P. Kong, D. Fu, and F.C. Lee, “High
for on-board electrical vehicle battery charge applications”, Pamukkale Power Density High Efficiency DC-DC Converter”, Energy Conversion
University Journal of Engineering Science, Congress and Exposition (ECCE), pp. 1392-1399, 2011.
DOI:10.5505/pajes.2016.56198. [21] J. Biela, U. Badstuebner, and J.W. Kolar, “Design of a 5-kW, 1-U, 10-
[18] B. Lu, W. Liu, Y. Liang, F. C. Lee, J. D. van Wyk, “Optimal Design kW/dm3 Resonant DC–DC Converter for Telecom Applications”, IEEE
Methodology for LLC Resonant Converter”, Applied Power Electronics Trans. on Power Electronics, vol. 24, pp. 1701-1710, July 2009.
Conference and Exposition, APEC '06, pp. 533-538, 2006.
____________________________________________________________________________________________________________
511
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
512
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 2. Circuit implementation of the proposed CMIA using CCII+ and DDCC elements
V X ( DDCC ) VY 1 VY 2 VY 3 , (3.a)
____________________________________________________________________________________________________________
513
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
1 1 .
f HP (5)
2 2R3 C1
R4 2 R1
VOUT (Vin Vin ) . (6)
R3 R 2
____________________________________________________________________________________________________________
514
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
one of the driving operational trans-conductance feedback loop, i.e. DC servo loop. The current-mode
amplifiers (OTA) are in the off state, related output integrator provides grounded capacitor implementation
transistor MON or MOP gate input floats and there exists for the integrator, which is advantageous for various
extreme current consumption at the output stage. Using implementations such as capacitance multiplication for
the proposed scheme, both of the output transistors are large capacitance values, etc. The integrator is employed
active for the full swing and provide class AB operation using a CCII+ with a capacitor shown as C1 in Fig. 2.
properly. Adjustment of the output biasing can be done by Filter high-pass cut-off frequency can be calculated using
sizing of the diode connected transistors MD1 and MD2. (5).
Selecting larger length (L) for the diode connected The internal structure of the DDCC element is shown in
transistors limits the output biasing current which is Fig. 5. The DDCC element is based on two equivalent
important for low-power operation. Self-biasing scheme differential input stage, one output stage with feedback for
for driving of OTAs and output stages provides stable the output X and current copying circuit from X to Z to
operation over large temperature range. Whenever bias generate Z output. The circuit is the improved version of
currents of OTAs increase by temperature, bias currents the DDCC proposed in [8]. The aspect ratios of the circuit
Ib_out also increase, then quiescent drain currents ( ID) of are given in Table 2. In the second stage of the CMIA, i.e.
MON and MOD is limited. So that low power operation is DDCC circuit, a rail-to-rail input is not required since
provided over a large temperature range. Full circuit large common-mode voltage component is removed at the
diagram with self-biasing scheme is shown in Fig. 4. first stage. Here, low-voltage cascode output is employed
Miller compensation is also employed between node X to at Z node, so that rail-to-rail output is possible with high
drain of M8 and node X to drain of M4 is also employed. precision the output of the proposed CMIA. Current
source Ib is set to 2µA with cascode current sources in the
circuit. Moreover, the circuit is operational under wide
TABLE I. TRANSISTOR ASPECT RATIOS OF FIG. 4. variety of current values. Bias voltages Vbias1 and Vbias2 are
W/L set to VDD˗1V and VSS+1V, respectively. In the proposed
Transistor
(µm/ µm) scenario, desired gain is possible by adjusting the R3 and
M1 - M2 40/2 R4 resistances if R1 and R2 are internally adjusted to a
constant value in Fig. 2 and using (6). DC servo loop can
M3 - M4, M7 – M8 20/2
also be de-activated by connecting Y3 input to GND.
M5 - M6 40/4 Since X node of the DDCC is constructed using voltage
MOP1 – MOP2 30/2
followers and Z node is implemented using cascade
stages, output gain can be precisely adjusted.
MON1 – MON2 20/2 For DC coupled implementations, Y3 node can also be
MB1, MB4 10/2 used for offset removal by removing the integrator and
applying a voltage divider network at node Y3. So that a
MB2, MB3 3/3
simple rail-to-rail input and output operation with
MB5, MB6, MB7, MB10 40/2 precision gain and various configurations is possible using
MB8, MB9 7/2 the proposed CMIA structure. Output can also be supplied
by a unity gain amplifier.
As explained before, the proposed instrumentation
amplifier is composed of two amplifier stages. First stage
of the CMIA is based on the basic structure shown in Fig.
1. Second stage is also a differential amplifier constructed
using DDCC element. DDCC element is a rather new
current conveyor, providing wide variety of algebraic
operations. However, DDCC element has lower common-
mode range. Since common-mode voltage is mostly
removed at the first stage, only residual common-mode
voltage appears at the second stage. So that, DDCC
element can be effectively employed at the second stage.
Moreover, various input algebraic operations provide also
an optional DC removal input at node Y3 as shown in Fig.
2. DC removal procedure is important especially for bio-
potential recordings since a low-offset voltage at the input
saturates the instrumentation amplifier and prevents
proper operation. So that after pre-amplification at first Fig. 6. Frequency response of the proposed CMIA.
stage, remaining DC offset is removed using an integrator
____________________________________________________________________________________________________________
515
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. SIMULATION RESULTS AND DISCUSSIONS inputs applied to Y1 and Y2 nodes of the DDCC input at
The proposed CMIA is configured for 40 dB gain. The the second stage as the outputs from the CCII+ elements
first stage gain is selected to be 20, the second stage gain are subtracted using the second stage amplifier. If
is set to 5. Using (6), R1 and R2 are selected to be 5kΩ and mismatch effect is not included the CMRR of the
500Ω, respectively. However, X nodes of the CCIIs have proposed structure is as high as 150 dB which is not
non-zero resistive input impedance. Due to input possible in reality. By applying geometry mismatch to the
resistance of X nodes, using simulations, R2 is replaced by differential stages of the CCII+ elements in the
485Ω. As a result the X node resistance of each CCII+ is simulations, the CMRR is recorded as 105 dB. Therefore,
estimated as (500-485) Ω /2 = 7.5 Ω. Second stage X and matching is very critical for the pairs of M1-M2; M3-M4;
Z nodes of the DDCC element are close to ideal, since M5-M6; and, M7-M8 of Fig. 4. CMRR calculation in [13,
better output stage is designed for the DDCC element. For 14] also shows that if two of the differential inputs have
the second stage, R3 and R4 are selected as 100kΩ and same gain error, the common-mode effect is minimized.
500kΩ, respectively. Capacitor C1 is selected as 1µF. Larger L values are employed for the transistors at the
Using (5) high-pass frequency for DC removal is 1.5 Hz, first stage of the CCII+ in Fig. 4, NMOS transistor lengths
which is suitable for electro-cardioghraphy (ECG) signals. especially kept larger to reduce flicker noise at the first
The frequency response of the amplifier is shown in Fig. stage. As a result, 850 nV/√Hz input referred noise at 1
6. The gain of 100 with cutoff frequency of 1.5 Hz is Hz is recorded. Noise level is reduced at higher
achieved using the given component values. frequencies, as expected. The results are summarized in
Table 1. The CMRR recording of the propsed CMIA is
comparable with the implementations in [14-16].
TABLE II. TRANSISTOR ASPECT RATIOS OF FIG. 5.
____________________________________________________________________________________________________________
516
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
____________________________________________________________________________________________________________
517
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Signal levels of GSM900, GSM1800 and UMTS bands division multiple access technology (FDMA). The maximum
were measured as mobile on the main streets in the city center of output power of a radio transmitter used in GSM is 10-40 watt
Diyarbakır at the same time and route during a week. The and frequency of power control is about 2 Hz [7]. UMTS uses
measurements were performed by using high precision and code division multiple access technology (CDMA). The
portable spectrum analyzer with an isotropic electric field
maximum output power of a radio transmitter used in UMTS
antenna and a laptop. The high precision spectrum analyzer is a
device that measures the high frequency electromagnetic fields. is 20-60 watt and frequency of power control is 1500 Hz [7].
The isotropic electric field antenna allows a three axis Power control and discontinuous transmission are two smart
measurements. The laptop was used to record and analyze the techniques [8] which are used in GSM and UMTS networks to
measurement samples. Electric field strengths were measured as reduce the EMF signal levels in the environment. Thus, these
signal levels in this study. Some statistical parameters such as smart techniques enable to save the power of mobile phone
standard deviation and variance were calculated with 95% and base station.
confidence interval for measurement samples. The highest signal The aim of this study was to measure signal levels of
levels were measured as 3.8580 V/m for GSM 900 band, 2.9440 GSM900, GSM1800 and UMTS bands as mobile in the city
V/m for GSM1800 band and 6.0900 V/m for UMTS band during
center of Diyarbakır in Turkey at the same time and route
a week. Similarly, mean electric field strength values as average
signal levels were calculated as 0.4985 V/m for GSM 900 band, during a week. Then, signal levels of them were compared
0.2350 V/m for GSM1800 band and 0.6281 V/m for UMTS band. with each other and statistically analyzed. Some statistical
According to the mean electric field strength values, the average values such as standard deviation and variance were
signal levels of three bands were ranked as UMTS, GSM900 and calculated with 95% confidence interval. The average signal
GSM1800 from the largest to the smallest, respectively. The levels of three bands were evaluated.
highest variation in standard deviation and variance was
observed on Wednesday for GSM900 and UMTS bands and
Thursday for GSM1800 band. II. MATERIAL AND METHOD
The measurements were performed between 10 November
Keywords— Signal level; Electric field strength; GSM; UMTS; 2014 and 16 November 2014 as mobile on the main streets
Statistics
where people spend time and have fun. Communication way
I. INTRODUCTION from base station to mobile phone is called downlink.
Likewise, communication way from mobile phone to base
Advanced technological developments in mobile station is called uplink. Output power of the base station is
communication systems have influenced people's lives in much higher than output power of the mobile phone.
many ways. Electromagnetic field (EMF) signal levels of Therefore, downlink band was only taken into account during
mobile communication systems such as Global System for the measurements. Downlink frequency bands of GSM900,
Mobile Communications (GSM) and Universal Mobile GSM1800 and UMTS as shown in Table 1 were measured
Telecommunications System (UMTS) are very important for during the whole week between 17:00 and 18:20 hours at the
epidemiological studies [1-3]. The effects of EMF signal same route.
levels on human health were examined by many scientific
studies [4, 5]. Basic restrictions for EMF signal levels were TABLE I
EMF BANDS AND THEIR FREQUENCY RANGE
recommended by the International Commission on Non-
Ionizing Radiation Protection (ICNIRP) [6]. Turkey has RF EMF Band Frequency Range (MHz)
adopted the ICNIRP's reference levels for EMF exposure. GSM900 DL 935 - 960
Mobile communication systems such as GSM and UMTS GSM1800 DL 1805 - 1880
use different technologies. For this reason, signal levels of UMTS DL 2110 - 2170
them may be different in the environment. GSM uses time
division multiple access technology (TDMA) and frequency DL: Downlink
____________________________________________________________________________________________________________
518
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The measurement set consisted of selective radiation meter measurement samples at an interval of 10 second during the
SRM-3006 (Narda Safety Test Solutions, USA) connected to mobile measurements. Then, the number of measurement
isotropic antenna by a radiofrequency cable and a laptop for samples for each EMF band was become equal to 472 samples
data collection. SRM-3006 is a high precision and portable per a day and 3304 samples per a week. The program written
spectrum analyzer which measures the high frequency by us got information (e.g., latitude, longitude, the maximum
electromagnetic fields. The isotropic antenna is the three axis instantaneous electric field strength values) from the SRM-
electric field antenna frequency range from 27 MHz to 3 GHz, 3006 and recorded measurement samples for statistical
dynamic range 0.2 mV/m - 200 V/m and maximum extended analysis. Descriptive statistics such as standard deviation and
measurement uncertainty +2.6/-3.8 dB for 85-2200 MHz. variance were calculated with 95% confidence interval.
Cable contains ferrite to reduce the effects of the external Measurement samples were analyzed using IBM SPSS
fields and its length is 1.5 meter [9]. Statistics software, version 21 [12].
Channel bandwidth for GSM900 [10] and GSM1800 [10] is
200 kHz, and for UMTS [11] is 5 MHz. Resolution bandwidth
of SRM-3006 for GSM900 and GSM1800 was set to 50 kHz III. RESULTS AND DISCUSSION
and for UMTS was set to 1000 kHz for better measurement. Currently, there are three mobile network operators in
SRM-3006 equipped with three axis electric field antenna Turkey. GSM frequency band of two mobile network
which was placed at 1.70 meter height above ground was operators is 900 MHz but the other is 1800 MHz. UMTS
mounted on the top of the car. Average speed of the car in the frequency band of three mobile network operators is 2100
measurement route was about 40 km/h. MHz. As seen in Table 2, Table 3 and Table 4,_10, _11, _12,
In this study, electric field strength values of three bands _13, _14, _15, _16 denote measurement days (Monday,
were measured as signal levels. The worst case was Tuesday, Wednesday, Thursday, Friday, Saturday and
considered during the performance of measurements. In this Sunday).
way, the maximum instantaneous electric field strength values
of GSM900, GSM1800 and UMTS bands were chosen as
TABLE III
DAILY STATISTICS FOR GSM900 DL
EMF band N Range Minimum Maximum Sum Mean Std. Deviation Variance
GSM900_10 472 2.8594 0.0316 2.8910 225.7753 0.478338 0.4312111 0.186
GSM900_11 472 3.0889 0.0321 3.1210 230.1624 0.487632 0.4306846 0.185
GSM900_12 472 3.8250 0.0331 3.8580 235.3006 0.498518 0.4636433 0.215
GSM900_13 472 2.1772 0.0308 2.2080 221.1566 0.468552 0.4148358 0.172
GSM900_14 472 2.4504 0.0336 2.4840 229.4507 0.486124 0.4238145 0.180
GSM900_15 472 2.8609 0.0342 2.8950 221.1493 0.468537 0.4340354 0.188
GSM900_16 472 3.1351 0.0329 3.1680 222.0962 0.470543 0.4346786 0.189
N: number of samples Std: Standard
TABLE IIIII
DAILY STATISTICS FOR GSM1800 DL
EMF band N Range Minimum Maximum Sum Mean Std. Deviation Variance
GSM1800_10 472 1.6957 0.0533 1.7490 101.6737 0.215410 0.2206387 0.049
GSM1800_11 472 2.7386 0.0535 2.7920 100.0886 0.212052 0.2668446 0.071
GSM1800_12 472 2.6898 0.0532 2.7430 104.4067 0.221201 0.2548263 0.065
GSM1800_13 472 2.8910 0.0530 2.9440 110.9306 0.235023 0.2713217 0.074
GSM1800_14 472 1.7811 0.0519 1.8330 101.6694 0.215401 0.2271515 0.052
GSM1800_15 472 2.0164 0.0536 2.0700 105.7467 0.224040 0.2543026 0.065
GSM1800_16 472 2.1617 0.0524 2.2140 104.4510 0.221294 0.2443795 0.060
N: number of samples Std: Standard
____________________________________________________________________________________________________________
519
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE IV
DAILY STATISTICS FOR UMTS DL
EMF band N Range Minimum Maximum Sum Mean Std. Deviation Variance
UMTS_10 472 3.1661 0.0729 3.2390 280.1896 0.593622 0.4958261 0.246
UMTS_11 472 4.6792 0.0748 4.7540 293.7641 0.622382 0.5356311 0.287
UMTS_12 472 6.0150 0.0750 6.0900 296.4812 0.628138 0.5753035 0.331
UMTS_13 472 3.9406 0.0724 4.0130 290.9976 0.616520 0.5179273 0.268
UMTS_14 472 3.9470 0.0741 4.0210 291.0592 0.616651 0.5403461 0.292
UMTS_15 472 4.7784 0.0786 4.8570 285.9496 0.605825 0.5459556 0.298
UMTS_16 472 4.3507 0.0753 4.4260 287.3403 0.608772 0.5040396 0.254
N: number of samples Std: Standard
TABLE V
A WEEKLY STATISTICS FOR MEASURED EMF BANDS
EMF band N Range Minimum Maximum Sum Mean Std. Deviation Variance
GSM900 3304 3.8272 0.0308 3.8580 1585.091 0.479749 0.4332338 0.188
GSM1800 3304 2.8921 0.0519 2.9440 728.9667 0.220632 0.2489943 0.062
UMTS 3304 6.0176 0.0724 6.0900 2025.782 0.613130 0.5309375 0.282
N: number of samples Std: Standard
____________________________________________________________________________________________________________
520
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[3] J.F. Viel, E Cardis, M. Moissonnier, et al., “Radiofrequency exposure base station antenna masts, Radiation Protection Dosimetry, 158, 263-
in the French general population: Band, time, location and activity 275, 2014.
variability”, Environment International, 35, 1150-1154, 2009. [8] Mishra, A. R. (ed.). “Advanced Cellular Network Planning and
[4] M. Vrijheid, E. Cardis, B.K. Armstrong, et al., “Validation of short Optimisation: 2G/2.5G/3G Evolution to 4G”, JohnWiley & Sons, 2007.
term recall of mobile phone use for the interphone study”, [9] Narda Safety Test Solutions, SRM-3006 selective radiation meter
Occupational and Environmental Medicine, 63, 237-243, 2006. operating manual, Pfullingen, Germany, 2012.
[5] F. Samkange-Zeeb, G. Berg, M. Blettner, “Validation of self-reported [10] Narda Safety Test Solutions, Application Note, GSM measurements
cellular phone use”, Journal of Exposure Analysis and Environmental with the Selective Radiation Meter SRM-3000, Pfullingen, Germany,
Epidemiology, 14, 245-248, 2004. 2007.
[6] ICNIRP,” Guidelines for limiting exposure to time-varying electric, [11] [Narda Safety Test Solutions, Application Note, UMTS measurements
magnetic and electromagnetic fields (up to 300 GHz), International with the Selective Radiation Meter SRM-3000, Pfullingen, Germany,
Commission on Non-Ionizing Radiation Protection”, Health Physics, 2008.
74, 494-522, 1998. [12] IBM Corp. Released 2012. IBM SPSS Statistics for Windows, Version
[7] M. Koprivica, N. Neskovic, A. Neskovic, et al., Statistical analysis of 21.0. Armonk, NY: IBM Corp., 2012.
electromagnetic radiation measurements in the vicinity of gsm/umts
____________________________________________________________________________________________________________
521
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Microwave imaging technology has attracted many Furthermore, there is a need for compact sized, low cost and
interests nowadays. This imaging system has been used in a high efficiency antennas. Consequently, microwave imaging
variety of applications such as: non-destructive testing and systems require wide-band, linearly polarized, high gain and
evaluation, through-the-wall imaging, concealed weapon compact antennas as their scattered field sensors [2].
detection at security check points, structural health monitoring
For microwave imaging, researchers are interested in the
and medical imaging. Microwave imaging techniques have in
recent decades experienced strong growth as a research topic in 1-11 GHz frequency range because it appears to balance the
biomedical diagnostics. The basic idea of using microwave conflicting demands of better spatial resolution (higher
imaging system is to transmit electromagnetic waves from a frequencies) and better penetration depth (lower frequencies)
transmitting antenna to the target material and receive the [3].
scattered waves at a receiving antenna. Thus, antenna choice In this study; the microstrip antenna which has circular
plays an important role for the system. There is a need for shaped patch with slot is selected due to their ease of
compact sized, low cost and high efficiency antennas which can fabrication, properties, small size and other advantages of
radiate ultra-wideband signal to transmit short pulses. microstrip technology. Ansoft HFSS software is used to
Especially, ‘large bandwidth’ is vital to attain good time domain
design the proposed structure and optimize the radiation
characteristics which is important for microwave imaging
systems. In this study; we selected the circular patch antenna pattern and return loss.
with a slot, due to their ease of fabrication, properties, small size The paper is organized as follows. In Section II, the
and other advantages of microstrip technology. The schematic of schematic of the proposed antenna is presented and the
the proposed antenna is presented and the dimensions are dimensions are summarized. In Section III, simulation results
summarized in the paper. Ansoft HFSS software is used to design are demonstrated and discussed.
the proposed structure and optimize the radiation pattern and
return loss. Simulation results such as directivity and return loss II. ANTENNA DESIGN
are given and discussed in the result part.
Ultra-wideband (UWB) technology has its frequency
defined in the range of 3.1 GHz to 10.6 GHz by the Federal
Keywords— Microstrip Antenna, Utra-Wide Band Antenna, Communication Commission (FCC) for UWB radio
Return Loss, Microwave Imaging
applications [4]. This technology has become very popular in
I. INTRODUCTION recent years and has found widespread applications in
communications landmine detection and biomedical
Microwave imaging techniques have shown excellent applications. Another important application is the UWB
capabilities in various fields such as civil engineering, technology in early breast cancer detection by using
nondestructive testing, industrial applications, and have in microwave imaging system. For this system, there is a need
recent decades experienced strong growth as a research topic for compact sized, low cost and high efficiency antennas
in biomedical diagnostics. Many research groups throughout which can radiate ultra-wideband signal to transmit short
the world work on prototype systems for producing images of pulses. Large bandwidth is vital to attain good time domain
human tissues in different biomedical applications, characteristics which is important for microwave imaging
particularly breast tumor detection [1]. The basic idea of using systems [5]. Recent research is on UWB antenna is planar
microwave imaging system is to transmit electromagnetic technology as they are more practical in term of
waves from a transmitting antenna to the target material and manufacturing and integration with the entire system.
receive the scattered waves at a receiving antenna. Thus, There are several compact UWB antennas are proposed in
antenna choice plays an important role. planar technology but they exhibit omini directional radiation
Microwave imaging systems require certain preprocessing property. However, in microwave imaging system, antenna's
steps, which accept only a single polarization of the incident HPBW is one of the main parameters determining the
field as input. Also, the designed antenna must have higher resolution; finer details can be resolved by using a narrower
gain levels together with a higher front-to-back ratio level. beam.
____________________________________________________________________________________________________________
522
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
In this study, first omini directional planar UWB antenna is III. ANTENNA SIMULATION
designed, then we try to improve it to directional planar UWB. For simulation findings, high frequency structure simulator
The proposed antenna is a modification of the antennas given (HFSS) [12] based on full-wave finite elements method and
in [6], which didn’t use slot on patch. widely used in the analysis of electromagnetic structures was
Microstrip antennas produced with printed circuit utilized. Presentation of the proposed ultra-wide band antenna
technology includes a thin layer of low loss nonconductive with HFSS is given in Figure 2.
material with a radiating conductive layer on one side and
conductive ground layer that completely covers the non-
conductive. Microstrip antennas are basically fed by the micro
strip or a coaxial line. Along with technological advances,
researchers have focused on different feeding structures and
developed three different feeding techniques [7,8,9]. In this
study a microstrip line feeding technique that is easy to design
and produce is presented.
The purpose of our study is to propose an antenna which
has the impedance bandwidth between 5 and 10.5 GHz for use
in near field near surface imaging applications. According to
this span, center frequency will be 7.5 GHz and wavelength (λ)
is equal to 40 mm. For the optimum results, radius of the
circular disk should be λ/4 =10 mm [10]. Also we added a slot
to circular patch which has 4 mm radius and return loss
Fig. 2 Proposed antenna with HFSS
graphic has been improved by using this slot. To achieve
directivity in designed circular monopole antenna, generally When simulation results are analyzed based on 10 dB
L– shaped reflector is used. In our study, ground parameters return loss criterion, it was observed that the impedance
were changed manually to obtain better results. Due to this bandwidth is between 5 and 10.5 GHz. At maximum
modification, front-to-back ratio is higher than the antenna operating frequency of the proposed antenna Return Loss
which presented in [11]. value was obtained as -33.98 dB. RL characteristic (S11) of
Figure 1 shows the layout view of designed circular disc the antenna is presented in Figure 3.
monopole antenna. Dimension of ground plane is shown in
figure. Width and length of microstrip line feed is 3mm and
17.2 mm respectively to match with 50Ω impedance line.
Center point of slot is same with center of circular patch and
diameter of circular disc is 20 mm.
As nonconductor material “FR4_epoxy” was selected for
the constructed antenna, dielectric constant (εr) was taken as
4.4 and dielectric height (h) was 1.6 mm.
____________________________________________________________________________________________________________
523
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] P. Nikola, “Measurement System for Microwave Imaging Towards a
Biomedical Application” Doctoral Thesis, Mälardalen University,
School of Innovation, Design and Engineering., Sweden, 2014.
[2] Abbak, M., “Antenna and Measurement System for Microwave
Imaging of Breast Tumors” Doctoral Thesis, Istanbul Technical
University Informatics Institute, Istanbul, Turkey, 2015.
[3] L. Xu, D. Shakti, H. Susan “Microwave Imaging via Space–Time
Beamforming: Experimental Investigation of Tumor Detection in
Multilayer Breast Phantoms” IEEE Transactions on Microwave Theory
and Techniques, vol. 52, no.8, pp. 1856-1865, August 2004.
[4] J. Breckling, Federal Communications Commission, “First report and
Fig. 4 Radiation pattern of proposed antenna at 7 Ghz
order", Revision of Part 15 of the Commission's Rules Regarding
Ultra-Wideband Transmission Systems, 2002, http://www.fcc.gov
IV. CONCLUSIONS [5] Golezanı, J. “Directional Wide Band Printed Monopole Antenna for
Size, directivity and a good reflection coefficient are the most Use in Microwave Breast Cancer Imaging” M.Sc. Thesis, Istanbul
Technical University, Turkey, 2012.
important characteristics of the antenna design requirements [6] Dixena, P., “Study and Design of Directional Ultra Wideband Antenna
for use in near field near surface measurement applications in Planar Technology”, ECE Dept.,IISc Bangalore
such as the microwave breast cancer imaging system. In this [7] O. Çolak, “Tek Darbe Mikroşerit Dizi Anten Tasarımı”, Yüksek Lisans
study, compact and small size of the antenna is presented. Tezi, İstanbul 2013.
[8] Volakis, J.L., ve Eibert, T.F., (2007), “Antenna Engineering
Simulation results confirm that reflection coefficient is under - Handbook”, The McGraw-Hill Companies, Ohio/ABD.
10dB throughout 5-10.5 GHz. The properties of compact and [9] Zhi, N. C. ve Michael Y. W. Chia, (2006) “Broadband Planar Antennas
UWB antennas have been ensured. Radiation pattern shows Design And Applications”, John Wiley & Sons, New York/ABD.
that, front-to-back ratio is 14 dB which is the good value for [10] K.P. Ray, “Design aspects of printed monopole antennas for ultra-wide
band applications", Hindawi Publishing Corporation, International
directivity feature. Unfortunately, it seems that there are Journal of Antennas and Propagation, Article ID 713858, 2008.
unwanted side loop. This problem can be resolved by making [11] Dixena, P., “Study and Design of Directional Ultra Wideband Antenna
some modification on ground plane. in Planar Technology”, ECE Dept.,IISc Bangalore
From simulation we can conclude that circular monopole [12] ANSYS HFSS (ver. 15), Ansys Corporation. Canonsburg, PA, USA,
2014.
antenna is favorable for microwave imaging systems.
____________________________________________________________________________________________________________
524
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— An annular ring patch antenna (ARPA) constructed by pin/wall and slot-loading on the antenna structure, have been
loading a circular slot in the center of the circular patch antenna introduced in recent years [1]. Several slot loaded CPA
is a popular microstrip antenna due to its favourable properties. configurations such as C [3], E [4-5], H [3, 6], L [7], rectangular
In this paper, an application of artificial neural network (ANN) ring [3] and annular ring [8] shapes have been presented in the
using bayesian regularization (BR) learning algorithm based on
literature as an alternative and effectively method to reduce the
multilayer perceptron (MLP) model is presented for computing
the operating frequency of annular ring ARPAs in UHF band. size of antenna.
Firstly, the operating frequencies of 80 ARPAs having varied Annular ring patch antennas (ARPAs) are miniaturized
dimensions and electrical parameters were simulated with antenna constructed by loading a circular slot in the center of
IE3DTM packaged software based on method of moment (MoM) in the circular patch. The size of the ARPA is substantially smaller
order to generate the data set for training and testing processes of than circular patch antenna (CPA) at the same operating
the ANN model. Then ANN model was built with data set and frequency [8]. It can be appreciated that the average path length
while 70 simulated ARPAs and remaining 10 simulated ARPAs travelled by the current in the annular-ring patch is much longer
were employed for ANN model training and testing respectively. than the corresponding circular patch for the lowest order mode
The proposed ANN model were confirmed by comparing with the
[8]. Also, by choosing the inner and outer radius of the ring
suggestions reported elsewhere via measurement data published
earlier in the literature. These results show that ANN model with properly, both bandwidth broadening [8] and controlling the
BR learning algorithm can be successfully used to compute the separation of resonant modes can be managed [9]. Due to these
operating frequency of ARPAs. useful properties, it is the one of the most studied PAs. In the
literature, the ARPA was theoretically investigated its resonator
Keywords— Annular ring patch antenna, operating frequency, model in [10-13]. The mathematical tools such as vector
artificial neural network, bayesian regularization learning Hankel transform, Galerkin’s method and Green functions were
algorithm greatly utilized in the analysis of the ARPAs [14-19]. Methods
based on cavity model and transmission line model was
I. INTRODUCTION presented to investigate some parameters such as the operating
The wireless communication systems are also moving frequency, input impedance and bandwidth [20-26]. The
towards the miniaturization very rapidly [1-2]. All these experimental studies concerning the ARPA were also
requirements of today’s age wireless communication systems performed to confirm the theoretical calculations in [9-10, 15-
have led the antenna researchers to work on the various aspects 16, 26-30]. It can be seen from the literature that these methods
of antenna designing with distinctive ideas of manufacturing include rigorous calculation of Hankel and Fourier transforms
and synthesis. Present portable communication and handheld and Bessel functions.
devices inherently need miniaturized patch antennas (PAs). By Analytical methods seems to be easier but they result in
using the substrate materials with high dielectric constant, the accurate solutions only for regular shapes of the patch, whereas
smaller antennas can be achieved but this gives rise to decrease the numerical electromagnetic computation methods are
the bandwidth and efficiency performances [1-2]. Thus, it is not suitable for all shapes of the PA. However, the numerical
an easy task to design a small PA managing the requirements methods require much more time in solving Maxwell’s
of mobile communication devices. To cope such shortcomings equations including integral and/or differential computations.
of the traditional PAs, compact patch antennas (CPAs) which So, it becomes time consuming since it repeats the same
are formed by applying some modification such as shorting- mathematical procedure even if a minor change in geometry is
____________________________________________________________________________________________________________
525
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
carried out. On the other hand, antenna designers prefer the TABLE I
DIMENSIONS AND DIELECTRIC CONSTANTS OF THE SIMULATED ARPAS
easier approaches without requiring much rigorous
computations and consuming time. Patch dimensions (mm)
Over a last decade artificial neural network (ANN) adopts Number of ao
remarkable importance in field of wireless communication due simulations Outer ai h εr
to its fast and accurate modelling, simulation, and optimization. radius Inner radius
In ANN model, we can use measured, simulated, and calculated
15 2, 4, 6, 8, 10 3.175 2.2
data for training. Trained ANN model predicts accurate
operating frequency for every small variation in the geometry 20 3, 6, 9, 12, 15 3.175 2.2
both for thin and thick substrates. The purpose behind the 25 4, 8, 12, 16, 20 3.175 2.2
training of ANN model is to minimize the error between actual 30 5, 10, 15, 20, 25 3.175 2.2
output and reference output. The use of ANN in computing 15 2, 4, 6, 8, 10 2.5 9.8
resonant frequency of CPAs have been proposed in [5-6].
20 3, 6, 9, 12, 15 2.5 9.8
In this study, a method of feed forward back propagation
ANN (FFBP-ANN) with bayesian regularization (BR) learning 25 4, 8, 12, 16, 20 2.5 9.8
algorithm [31] based on multilayered perceptron (MLP) model 30 5, 10, 15, 20, 25 2.5 9.8
4x20
has been applied to compute the resonant frequencies of 15 2, 4, 6, 8, 10 1.57 2.33
ARPAs. The operating frequency values of 80 ARPAs 20 3, 6, 9, 12, 15 1.57 2.33
corresponding most of UHF band covering GSM, LTE, WLAN
25 4, 8, 12, 16, 20 1.57 2.33
and WiMAX applications were determined by the
electromagnetic simulator IE3DTM using method of moment 30 5, 10, 15, 20, 25 1.57 2.33
(MoM) [32]. The simulation parameters of 70 ARPAs 15 2, 4, 6, 8, 10 0.64 4.5
representing the overall problem space were used to training 20 3, 6, 9, 12, 15 0.64 4.5
and the remaining 10 were then employed to test the accuracy.
25 4, 8, 12, 16, 20 0.64 4.5
The results of the ANN model obtained in this study was
confirmed by comparing with the measurement results 30 5, 10, 15, 20, 25 0.64 4.5
published earlier in the literature [9, 10, 15, 16, 26–30].
In the simulations, the antennas were supposed to a probe
II. ANNULAR RING PATCH ANTENNA AND SIMULATION PHASE feed with 50 Ω. For meshing process, cell/wavelength rate
As shown in Fig. 1, ARPA has an annular ring patch formed values were assumed as 40 in limit of 4 GHz. The built in
by loading a circular slot with radius ai on a circular patch of optimization module of the IE3DTM was utilized to determine
radius ao on the substrate having relative dielectric constant εr the feed point which gives the best return loss value with the
overall on the ground plane. objective function S11(dB) < −10 for the operating frequencies
at TM11 mode.
____________________________________________________________________________________________________________
526
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 3 The comparative results of the simulation and ANN model for training
phase
Antenna parameters
TABLE II Antenna
THE ANN PARAMETERS number Patch dimensions (mm)
ao ai h εr h/λd
Parameters Value 1 15 4 2.5 9.8 0.045
Number of input 4
2 15 6 1.57 2.33 0.027
Number of output 1
Epochs 500 3 15 10 3.175 2.2 0.056
Seed value 662862703 4 20 3 0.64 4.5 0.009
Minimum gradient descent 10-10 5 20 6 1.57 2.33 0.021
Momentum parameter (μ) 0.0001 6 20 15 2.5 9.8 0.027
μ increment 4 7 25 4 3.175 2.2 0.035
μ decrement 0.1 8 25 12 1.57 2.33 0.015
Maximum μ 1010
9 30 10 0.64 4.5 0.005
10 30 20 3.175 2.2 0.024
____________________________________________________________________________________________________________
527
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE IV model learning with BR has been compared with those of the
THE OPERATING FREQUENCIES AND APE VALUES FOR TEST PROCESS
methods reported elsewhere [12-13, 25-26] over several
Operating Frequencies (GHz) measurement data of ARPAs published earlier in literature [9-
fANN Calculated by 10, 15-16, 26-30]. Table V gives the operating frequency values
fIE3D predicted with the ANN model with BR and calculated with the
BR [12] [13] [25] [26]
methods given in [12-13, 25-26], and also the their respective
1.734 1.744 1.735 1.814 1.813 2.888
APE values according to the measurement parameters [9-10,
3.323 3.329 3.265 3.181 3.181 3.749 15-16, 26-30]. It should be noted that these measurement
3.563 3.498 2.917 2.843 2.844 2.396 parameters were not employed for training the ANN model. As
2.006 1.965 1.995 2.025 2.025 5.234 can be seen from the Table V, the methods proposed for
2.594 2.600 2.576 2.534 2.535 3.700 operating frequency of the ARPA yield the comparable results,
1.033 1.035 1.041 1.027 1.026 0.804 however, some calculations are in good agreement with some
2.258 2.300 2.135 2.354 2.355 5.748 measured data, and others are far off. But, the operating
1.833 1.819 1.832 1.785 1.785 1.854 frequency results of the ANN providing the least APE value
1.189 1.190 1.193 1.160 1.159 1.564 within 1% are closer to the measurement ones for the most
1.547 1.601 1.420 1.395 1.395 1.175 cases. The results achieved here show that the ANN can
APE (%) 1.118 3.595 5.207 5.217 54.910 effectively be employed to estimate the operating frequency of
ARPAs.
In order to verify the accuracy and validity for the crosscheck,
the operating frequency results obtained by means of the ANN
TABLE V
THE MEASURED AND CALCULATED OPERATING FREQUENCIES FOR ARPAS
As an important matter of the fact that, although the method results with ANN for training and test data are in a good
of ANN seems more complicated as compared to the other ones, agreement with the simulation results. This method achieved
it provides the more accurate and relatively simple way since it the more accurate results as compared to those of the methods
requires neither sophisticated functions of mathematical proposed in the literature. This ANN approach is simple and
transformations nor rigorous expertness to determine the fast modeling which produces more accurate results for the
unknown parameters in any problem including highly operating frequency of the ARPAs with less computational
nonlinearity. The training process is once completed in a few time and least errors. The most important advantages ANN
minutes by properly choosing the network parameters and the model is accuracy and easy to implement for the engineering
learning method, one can easily compute any parameters of problems which include the high nonlinearity.
interest in microseconds.
ACKNOWLEDGMENT
IV. CONCLUSIONS The heading of the Acknowledgment section and the
In this paper, an application of ANN model which have been References section must not be numbered.
used BR learning algorithm is successfully implemented for the
prediction of accurate operating frequency of ARPAs. IE3D™ REFERENCES
simulation software based on MoM was used to define [1] K. Wong, Compact and broadband microstrip antennas, John Wiley &
operating frequency of 80 ARPAs. ANN model, physically and Sons, Inc., 2002.
[2] G. Kumar and K. P. Ray, Broadband microstrip antennas, Norwood:
electrical parameters of 70 ARPAs were utilized training data, Artech House, 2003.
10 ARPAs were utilized for the test. It was seen that computed
____________________________________________________________________________________________________________
528
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[3] A. A. Deshmukh and G. Kumar, “Formulation of resonant frequency for [24] V. Sathi, C. H. Ghobadi and J. Nourinia, “Optimization of circular ring
compact rectangular microstrip antennas,” Microwave and Optical microstrip antenna using genetic algorithm,” International Journal of
Technology Letters, vol. 49, no. 2, pp. 498–501, 2007. Infrared and Millimeter Waves, vol. 29, pp. 897–905, 2008.
[4] A. A. Deshmukh, N. V. Phatak, S. Nagarbovdi and R. Ahuja, “Analysis [25] I. J. Bahl and S. S. Stuchly, “Closed-form expressions for computer-
of Broadband E-shaped Microstrip Antennas,” International Journal of aided design of microstrip ring antennas,” International Journal of
Computer Applications, vol. 80, no. 7, pp. 24– 29, 2013. Infrared and Millimeter Wave Computer-Aided Engineering, vol. 2, pp.
[5] A. Akdagli, A. Toktas, A. Kayabasi and I. Develi, “An application of 144–154, 1992.
artificial neural network to compute the resonant frequency of E-shaped [26] R. Kumar and D. C. Dhubkarya, “Design and analysis of circular ring
compact microstrip antennas,” Journal of Electrical Engineering- microstrip antenna,” Global Journal of Researches In Engineering, vol.
Elektrotechnicky Casopis, vol. 64, no. 5, pp. 317–322, 2013. 11, no. 1, 2011.
[6] A. Kayabasi, M. B. Bicer, A. Akdagli and A. Toktas, “Computing [27] J. S. Dahele and K. F. Lee, “Characteristics of annular-ring microstrip
resonant frequency of H-shaped compact microstrip antennas operating antenna,” Electronics Letters, vol. 18, pp. 1051–1052, 1982.
at UHF band by using artificial neural networks,” Journal of the Faculty [28] K. F. Lee, J. S. Dahele and K. Y. Ho, “Annular-ring and circular-disc
of Engineering and Architecture of Gazi University, vol. 26, pp. 833– microstrip antennas with and without air gaps,” 13th European
840, 2011. Microwave Conference, pp. 389–394, 1983.
[7] Z. N. Chen, “Radiation pattern of a probe fed L-shaped plate antenna,” [29] J. S. Row, “Dual-frequency circularly polarized annular-ring microstrip
Microwave and Optical Technology Letters, vol. 27, pp. 410– 13, 2000. antenna,” Electronics Letters, vol. 40, pp. 153–154, 2004.
[8] W. Chew, “A broad-band annular-ring microstrip antenna,” IEEE [30] J. Shinde, P. Shinde, R. Kumar, M. D. Uplane and B. K. Mishra,
Transactions on Antennas and Propagation, vol. 30, no. 5, pp. 918–922, “Resonant frequencies of a circularly polarized nearly circular annular
1982. ring microstrip antenna with superstrate loading and airgaps,” In:
[9] J. S. Dahele, K. F. Lee and D. Wong, “Dual-frequency stacked annular- Kaleidoscope: Innovations for Future Networks and Services, pp. 1–7
ring microstrip antenna,” IEEE Transactions on Antennas and 2010.
Propagation, vol. 35, pp. 1281–1285, 1987. [31] M. Zandieh, A. Azadeh, B. Hadadi and M. Saberi, “Application of neural
[10] I. J. Bahl, S. S. Stuchly and M. A. Stuchly, “A new microstrip radiator networks for airline number of passenger estimation in time series state,”
for medical applications,” IEEE Transactions on Microwave Theory and Journal of Applied Science, vol. 9, no. 6, pp.1001–1013, 2009.
Techniques, vol. 28, pp. 1464–1469, 1980. [32] R. F. Harrington, Field computation by moment methods, Piscataway.
[11] I. Wolff and N. Knoppik, “Microstrip ring resonator and dispersion NJ, IEEE Press., 1993.
measurement on microstrip lines,” Electronics Letters, vol. 7, pp. 779– [33] D. J. C. Mackay, “Bayesian interpolation,” Neural Computation, vol. 4,
781, 1971. pp. 415–447, 1992.
[12] S. G. Pintzos and R. Pregla, “A simple method for computing the
resonant frequencies of microstrip ring resonators,” IEEE Transactions
on Microwave Theory and Techniques, vol. 26, pp. 809–813, 1978.
[13] Y. S. Wu and F. J. Rosenbaum, “Mode chart for microstrip ring
resonators,” IEEE Transactions on Microwave Theory and Techniques,
vol. 21, pp. 487–489, 1973.
[14] S. M. Ali, C. Weng and J. Kong, “Vector Hankel transform analysis of
annular-ring microstrip antenna,” IEEE Transactions on Antennas and
Propagation, vol. 30, pp. 637–644, 1982.
[15] Z. Fan and K. F. Lee, “Hankel transform domain analysis of dual-
frequency stacked circular-disk and annular-ring microstrip antennas,”
IEEE Transactions on Antennas and Propagations, vol. 29, pp. 867–870,
1991.
[16] H. Liu and X. F. Hu, “An improved method to analyze the input
impedance of microstrip annular-ring antennas,” Journal
Electromagnetic Waves and Applications, vol. 10, pp. 827–833, 1996.
[17] H. Liu and X. F. Hu, “Input impedance analysis of microstrip annular
ring antenna with thick substrate,” Progress In Electromagnetic
Research, vol. 12, pp. 177–204, 1996.
[18] C. S. Gurel and E. Yazgan, “Resonant frequency analysis of annular ring
microstrip patch on uniaxial medium via Hankel transform domain
immittance approach,” Progress In Electromagnetic Research, vol. 11,
pp. 37–52, 2010.
[19] A. Motevasselian, “Specteral domain analysis of resonant characteristics
and radiation patterns of a circular disk and annular ring microstrip
antenna on uniaxial substrate,” Progress In Electromagnetic Research,
vol. 21, pp. 237–251, 2011.
[20] W. F. Richards, O. Jai-Dong and S. Long, “A theoretical and
experimental investigation of annular, annular sector, and circular sector
microstrip antennas,” IEEE Transactions on Antennas and Propagations,
vol. 32, pp. 864–867, 1984.
[21] A. K. Bhattacharyya and R. Garg, “Input impedance of annular ring
microstrip antenna using circuit theory approach,” IEEE Transactions
on Antennas and Propagations, vol. 33, pp. 369–374, 1985.
[22] S. E. El-Khamy, R. M. El-Awadi and E. B. A. El-Sharrawy, “Simple
analysis and design of annular ring microstrip antennas,” Microwaves,
Antennas and Propagation, IEE Proceedings H, vol. 133, pp. 198–202,
1986.
[23] J. Gomez-Tagleand and C. G. Christodoulou, “Extended cavity model
analysis of stacked microstrip ring antennas,” IEEE Transactions on
Antennas and Propagations, vol. 45, pp. 1626–1635, 1997.
____________________________________________________________________________________________________________
529
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— There are two regions in the orbital path of the the speed. Therefore, high speed is important to store energy
satellite such as dark and bright region. The energy is provided in flywheel. The BLDC motors are extensively used in the
by solar panels in the bright region and by flywheel energy FESS besides industrial application such as automotive
storage system (FESS) in the dark region. Brushless dc motor industry, medical, automation system and aircraft applications.
(BLDC) are extensively used in satellite power system as motor
BLDC motors are used due to high power density, low weight,
and generator. Also, these motors are preferred in the FESS due
to the low weight, high power density, high efficiency and high high efficiency and high speed in the satellite power system
speed. The voltage is obtained by BLDC motor in the generator [10]. The high speed BLDC motors are used as motor in
mode and generated voltage is must be constant a value. The charge mode and generator in discharge mode.
FESS speed and the BLDC motor phase voltage are reduced in Three phase converters is widely used in motor driver,
the dark region. Therefore the voltage regulation is provided by energy systems and other power electronic applications. Many
using the boost converter in the generator mode. Boost converter works related to the three phase direct AC-DC boost rectifier
is designed by using the three phase inverter which used as are presented such as resistive input behavior of converter,
BLDC motor driver in the bright region. Information of the hall- low voltage and low power applications, boost rectifier for
effect sensors are used to the voltage regulation of BLDC
high-speed permanent magnet generator, energy regenerative
generator. In this paper, sensorless control method was proposed
in order to regulate the voltage of the satellite power system. The damping [11]–[14] and optimal energy harvesting [15]. Three-
performance of the proposed method has been demonstrated by phase boost rectifier topology is extensively used in low
using the Matlab/SimPowerSystem blocks. power applications.
In generator mode, the flywheel speed with the BLDC
motor back-emf is decreased during the dark region time.
Keywords— BLDC generator, Flywheel, Energy storage, Three-
Therefore, voltage regulation cannot achieved in power
phase boost rectifier
system of satellite in dark region. In this applications, voltage
regulation is presented by using three-phase boost rectifier in
I. INTRODUCTION discharge mode of the FESS.
Phase back-emf of BLDC motor is corresponded position
FESS has been used power regulated in grid power system, of hall-effect sensors which used boost converter switching
electrical vehicle, satellite power system and uninterrupted [16]. Phase back-emf of BLDC motor is measured to
power supplies in the last two decades. The FESS is preferred determine phase conduction order in proposed method.
according to electrochemical battery in satellite power system Therefore, sensor placement errors are eliminated with
[1]–[9]. Because, the FESS is advantages such as longer life sensorless control algorithm. Lower switches of the inverter
time, bigger depth of discharge and be integrated at moment are made ON by using the BLDC motor of the phase back-
control gyroscope. The energy is provided by the FESS in the emf, respectively. Upper switches of the inverter are
dark region. The mechanical energy is stored by the solar continually made OFF and their parallel diode used as boost
panels in the bright region and the stored energy is used to converter diode.
provide the satellite's power needs in the dark region. Another Current reference is obtained by using the DC bus voltage
task of the solar panels is provide the energy of the satellite in which measured by voltage sensor. Generator current can be
the bright region. controlled by using current reference. Flywheel speed and DC
The FESS is occurred from some mechanical parts such as bus voltage is presented by using MATLAB/SimPowerSystem
flywheel, vacuumed environment, electric machinery and high blocks in generator mode with conventional method and
speed mechanic or magnetic bearing. The magnitude of proposed method.
energy stored in the flywheel is proportional to the square of
____________________________________________________________________________________________________________
530
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
1
fc (4)
2 L f C f
____________________________________________________________________________________________________________
531
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the mass center and dm is the average diameter of bearing. When Q4 is turned to the OF, circuit of boost
Viscos friction of bearing grease have been not calculated due converter consist of two and two series inductors. Current
to use lean full-ceramic bearing. flow is realized through phase-a and phase-b by following the
path D1-C-D4. In this situation, the inductor energy is
III. THREE-PHASE BOOST CONVERTER transferred to the DC bus capacitor and load.
Speed of the flywheel is reduced according to demand
energy of the satellite during dark region. Therefore voltage of
BLDC generator is decrease but the voltage of satellite power
system must be constant a value. In generator mode, three-
phase inverter and LC filter inductance are used as three-
phase boost converter and boost inductor, respectively.
Figure 2 shows the general control technique of a three-
phase boost rectifier. Generator current and the DC link
voltage are controlled by the PI controller as shown in Fig.2.
Current of BLDC generator is calculated as given in Eq. (14).
Ia Ib Ic
Is (14)
2
where Ia , Ib and Ic are current of BLDC generator phase, Is
is current of generator.
DC link voltage is measured to compare by reference
voltage. Voltage controller output is obtained as reference to
current controller. Upper legs of the three-phase inverter have
been continually made OFF and their diodes have been used
as boost converter diode. PWM has been respectively applied
in lower legs by using the BLDC generator phase voltage.
Conduction switch is determined by using the phase voltage in
generator mode.
1
Fig. 2 The general control technique of a three-phase boost rectifier
0.5
____________________________________________________________________________________________________________
532
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
1 (15).
Peak region of the positive phase voltage was used in order
0.5
to do ON lower switches. And it was obtained as shown in
0 figure (7).
-0.5
1
-1
5 5.5 6 6.5 7
V pa , pb , pc
2
Va ,b ,c Va ,b ,c (15)
t [s] -3
x 10
20
Phase Voltage [V]
10
-10
-20
5 5.5 6 6.5 7
t [s] -3
x 10
f0 :1.3;
f1 :0.00037;
dm :18 mm; Fig. 7 Detection peak region of phase voltage in proposed method
m :0.670 kg;
g :9.8 m/s2;
In proposed method, state of boost converter switches and
mr *e :10 gmm;
phase voltages of BLDC generator are the similar to
J :0.008 kgm2;
conventional method as shown in Fig 8. Phase voltage,
positive region of the phase voltage and switching sequence is
Maxon EC-25 high-speed BLDC motor is used in
shown in Fig 8, respectively.
simulation. Phase voltage waveform obtained by using the
hall-effect sensors and phase voltage are shown in Figure (4)
Phase Voltage [V]
20
and Figure (5), respectively.
Q2, Q4 and Q6 are done respectively ON at the positive
0
region of this voltage as shown in Fig. (6). Thus hall-effect
sensors have been synchronized by the phase voltage of
BLDC generator. -20
5 5.5 6 6.5 7
t [s] x 10
-3
20
10
0
5 5.5 6 6.5 7
t [s] x 10
-3
1
ON-Switch
0
5 5.5 6 6.5 7
t [s] x 10
-3
Fig. 8 Sensorless Control Method. Top: Phase voltage of the BLDC generator,
Middle: Positive region of the phase voltage, Bottom: Switching sequence of
the three-phase boost converter.
Fig. 6 ON position of lower switch is obtained
Full ceramic bearing has been used in simulation of the
The phase voltage of BLDC generator was measured by FESS. The bearing losses has been calculated to obtain load
using the star point of the LC filter capacitors in the proposed
____________________________________________________________________________________________________________
533
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
40 40 6
4
DC Bus Voltage [V]
____________________________________________________________________________________________________________
534
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
535
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— this paper presents low noise, high voltage [3] presented The strategy for minimizing the flicker noise in
gain(Av), and wide bandwidth(GW) CMOS folded cascade Op- the folded-cascode amplifier topology and the inter-
Amp. The strategy used for improving the input referred noise relationship of design parameters for optimum design. The
of the proposed CMOS operation amplifier based-on proposed solution offers good tradeoff on the conflicting
maximizing the value of the small-signal transconductance (gm)
performance parameters such as noise, silicon area, bandwidth
of MOSFET transistors. This strategy can be achieved using a
large DC bias current. For test the operation of the proposed and power consumption. Yamu Hu, M. Sawan, in 2000 [4]
CMOS op-amps, Electrocardiogram (ECG) signal used as input described a low-noise, low-power and low voltage amplifier
signal with 2mV peak to peak value and 200Hz frequency. dedicated to very low amplitude signal acquisition from
PSPICE (version 16.6) used for simulation the proposed CMOS implantable electronic devices. Low noise and low DC offset
OP-AMP with 0.18µm channel length. The simulation results are realized by means of Chopper Stabilization (CHS)
with 10µA DC biasing current showed that, the value of DC technique. Reid R. Harrison, Cameron Charles, in 2003 [5]
power consumption is 0.296mW, the input referred noise is designed and tested a novel bio amplifier that uses a MOS-
67.003nV/ √Hz, THE Av=13.6dB, 203KHz. The simulation bipolar pseudo resistor element to amplify low-frequency
results of the proposed CMOS op-amp with 100µA DC biasing
signals down to the milli hertz range while rejecting large dc
current consumes DC power around 1.002mW and exhibits a
lower input referred noise of 16.119nV/√Hz, the voltage gain is offsets, and they derived the theoretical inversion Pedram
51.4dB and GW is 10MHz. The simulation results confirmed the Mohseni, and Khalil Najafi, in 2004 [6] addressed low noise,
theoretical calculations and verified very good noise low power fully integrated operational amplifier for a variety
performance of the proposed CMOS OP-AMP. of biomedical neural recording applications. They
demonstrated full functionality of this amplifier through in
Index Terms— CMOS analog integrated circuit, design
CMOS OP-AMP, low-noise OP-AMP, ECG signals data vitro measurements in saline using a polyimide sieve electrode
acquisition circuit. and a micro-machined penetrating silicon probe. 4
____________________________________________________________________________________________________________
536
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
II. METHODOLOGY having a large width to length (W/L) ratio for the
device [9].Two different values of the biasing current
Our objective is the simulation and the design are used to improve noise performance of the proposed
analysis of a high voltage gain, low noise, and low power folded cascode CMOS OP-AMP. Table (1)
folded cascode CMOS OP-AMP using 0.18μm CMOS summarized the gate dimensions for MOSFETs of the
technology so as to use it in many applications such as in proposed folded cascode CMOS OP-AMP.
Bio-potential signal acquisition systems [7].
____________________________________________________________________________________________________________
537
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig.3. Frequency Response (phase) of the proposed folded cascode CMOS (b)
OP-AMP with 10µA Biasing Current
Simulation results of the input referred noise, and noise Fig. 4.Noise performance of the folded cascade CMOS Op-Amp using
spectral density of the proposed folded cascode CMOS biasing current 100µA (a) Input Referred noise (b) Output noise
OP-AMP with 10µA basing current are shown in
Figure (4) (a) and Figure (4)(b) respectively. We note
that, the value of input referred noise is 67.003nV/√Hz
and the value of spectral density is 319.62µV.
(a)
(a)
____________________________________________________________________________________________________________
538
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE 2. PERRONAMCE PARAMETERS OF THE PROPOSED CMOS Meanwhile, the recording electrodes might pick up
OP-AMP USNG TWO DIFFERENT BIASING CURRENTS
much other unneeded interference. However, the
Performance Folded cascade Folded
biomedical electronics might be unable to detect small
parameters CMOS OP-AMP cascade biomedical signals. Therefore, we need a high gain,
using basing CMOS OP- accurate, and high CMRR amplifier to reduce the
current (100µA) AMP using common mode noise and to amplify the biomedical
basing signal only [11, 7].
current
(10µA) For testing the designed folded cascode CMOS OP-
Supply voltage ±1.8 V ±1.8 V AMP, a closed loop voltage gain of figure 7, with
Biasing current 100μA 10μA feedback resistor and input resistor are 500kΩ, 1kΩ are
Gain 51.475 dB 13.603 dB
used.
Bandwidth 10.453 MHz 203.392 KHz
(GW) 8.6 mV/μs 5.48 mV/μs
Slew Rate 116.13µs 179.84 µs
Settling time 1.002mW 0.296mW
Power
consumption 16.119nV/√Hz 67.003nV/√Hz
Input referred
noise 5.5 E-12m² 0.368E-12 m²
Area 78.6 dB 91.548 dB
Signal to noise
ratio
IV. BIOMEDICAL APPLICATION (ECG SIGNAL) Fig. 7. Closed loop voltage gain of proposed CMOS OP-AMP with ECG as
input signal
BASED ON THE DESIGNED CMOS OP-AMP
____________________________________________________________________________________________________________
539
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] F. Mohd-Yasin, M.T. Yap, "CMOS Instrumentation Amplifier
with Offset Cancellation Circuitry and High PSRR for Low
Power Application", Trieste, Italy, 2009, pp. 1-9.
[2] P.E. Allen and D.R. Holberg, "CMOS Analog Circuit Design",
Oxford University Press, 2002.
[3] P.K.Chan, L.S. Ng, L. Siek, and K.T. Lau, "Designing CMOS
folded-cascode operational amplifier with flicker noise
minimisation", Microelectronics Journal vol. 32, 2001.p.p. 69–
73.
____________________________________________________________________________________________________________
540
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The residual LSF vector quantization yields bit rate pronunciation the velum is always closed and the sound wave
reduction in the vocoders. In this work, a residual LSF vector proceeds only through the oral tract. So the influence of the
quantization obtained from Auto Regressive Moving Average nasal tract is ignored in this assumption. There is no big
(ARMA) prediction is proposed for designing codebooks at very problem when non-nasal sounds are processed but in case of
low bit rates. This residual quantization method is applied to
nasal sounds the mismatch of the LP model becomes severe.
multi stage vector quantization method and codebooks are
designed. For each codebook, the effectiveness and quality are The zeros during nasal sounds supress the peaks in mid-
investigated by calculating the spectral distortion and outliers. frequency by flattening the spectrum there but this effect
The proposed quantization method reduced the distortion cannot easily fitted by all pole modelling. In order to include
without any additional complexity. the effect of both oral and nasal tracts, it is necessary to
modify all pole modelling into a pole zero modelling [3]. In
order to obtain more efficient speech coding algorithms
Index Terms – very low bit rate, speech processing,
especially for transmission over noisy channels, differential
residual vector quantization, formant tracking, arma
quantization or predictive quantization of spectrum parameters
prediction
are used. There are some pole zero modelling approaches in
the literature [5-11] but they usually uses nonlinear equations
or approximations. While all pole modelling is simple, pole-
zero modelling requires complex nonlinear calculations.
1. INTRODUCTION Although a pole-zero algorithm based on adaptive kalman
Speech coding refers to process of reducing the bit rate of filtering presented in [11] linearize the nonlinear components
digital speech representations for transmission or storage, by dividing the frequency range of each formant into four
while maintaining a speech quality that is acceptable for the bands, this nonlinear approximation method also requires too
application. Most of the speech coders reported in the many calculations causing a complexity in the pole zero
literature are based on linear prediction (LP) analysis [1]. For modelling.
the LP based vocoders, the bit rate reduction is strongly tied to In this work, we propose an ARMA prediction model for
efficient quantization of the LPC filter coefficients {aj}. The predictive quantization of spectrum parameters. This ARMA
Line Spectral Frequencies (LSF) –an equivalent representation prediction method combines the good features of AR and MA
of {aj}, more suitable for quantization and interpolation– can prediction methods while eliminating their some drawbacks.
alternatively be used. In this sense, the Multi-Stage Vector
In section 2, basic formulas of MSVQ are given and then in
Quantization (MSVQ) of LSF parameters presented in [2] has
section 2.2, residual vector quantization method using ARMA
an efficient quantization performance at 22-24 bits per 20 ms
prediction is described. In section 3, designed codebook
frames. Furthermore, the multi stage structure has more
results using proposed method are presented.
flexibility than a single stage VQ in terms of search
complexity, codebook storage and channel error protection.
Very low rate speech communication systems require efficient
fixed-rate and low delay coding methods which operate at
lower bit rates. 2. RESIDUAL LSF QUANTIZATION
Generalized vocal tract model consists of the oral tract and In this section, a brief description of the MSVQ method
nasal tract. On the other hand, the linear predictive coding, presented in [2] is outlined. The definitions presented in this
which has been widely used in the speech analysis and section are introductory information for the residual MSVQ.
synthesis, uses all pole type digital filters. Speech signals are The training technique we used in designing the
assumed to be produced by filtering glottal excitation with codebooks is the joint design technique [2]. Representative
these filters. This all pole type filter model approximates the results of the residual LSF scheme with the joint design
true physical configuration of the human vocal tract, but with technique are presented in section 3.
the nasal tract left out. The most crucial and well known
shortcoming in this assumption that during any voiced
____________________________________________________________________________________________________________
541
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
2. 1. Notation and Definition where the superscript n identifies the nth vector from the
The MSVQ codebooks are designed using the Generalized training sequence and the subscript r represents the iteration
Lloyd Algorithm (GLA) to minimize average Weighted Mean number during the training of a codebook. For details of
designing a codebook in MSVQ, the reader is referred to [2].
Square Error (WMSE) based on a sufficiently rich training
sequence. The training sequence is first partitioned into
decision regions or cells for a given set of centroids (or 2.2. Residual LSF vector Quantization with ARMA
codevectors). Then, for the given partitioning, the codebooks prediction
are re-optimized to minimize the distortion over the particular
decision regions. In the R_MSVQ method, the residual LSF parameters of
current frame are predicted from the quantized LSF
In the MSVQ system [2], the parameter vector x consisting parameters of the previous frames using interframe correlation
of p LSF parameters is approximated as a quantized parameter feature of spectrum parameters [5-8] and then residual LSF
vector x̂ using the minimum distortion rule1 vectors are coded with a MSVQ codebook. Firstly, the LSF
parameter vector is obtained by transforming the 10th order
xˆ y0(l0 ) y1(l1 ) ... yK(lK11 ) LPC parameter vector. Next, the average LSF vector of the
B0(l0 ) c0 B1(l1 ) c1 ... BK(lK11 ) cK 1 (1) training set x DC is subtracted from the LSF vector x(i)
belonging to the ith frame. By defining mean removed LSF
Bc (i )
vectors ( z x (i ) x DC ) and its quantized version
where superscripts denote codevector indices from each stage, (i ) (i )
subscripts denote the stage numbers and K is the number of ( zˆ xˆ xDC ), the residual LSF vector e(i) is calculated
stages. cj is the codebook vector for the jth stage and created using
by stacking the codevectors,
e (i ) z (i) r (i) (6)
cj y ( 0 )T
j y (1 ) T
j .... y
( L j 1)T
j
T
L j p 1
(2) where i=1,2,…and r
(i )
(0)
= 0. The quantized residual vector
eˆ is found by quantizing e(i) with a VQ codebook.
(k )
where y j (p 1) is the kth codevector for the jth stage and Lj Depending on how r ( i ) is computed, various prediction
is the size of codebook for that stage. The column vector c is
schemes can be proposed. If r ( i ) eˆ ( i 1 ) r ( i 1 ) , a
constructed by stacking all codebook vectors from all stages first order Auto Regressive (AR(1)) predictor is obtained [5].
and referred to as the ‘codebook’ where, When r ( i ) eˆ ( i 1 ) we have a first order Moving Average
( 0 )T (1)T ( L 1) T (MA(1)) predictor [5]. The research in the literature have
c0 [ y 0 y 0 ... y 0 0 ]T focused on these two schemes [6-8] which show that
c ( 0 )T (1)T ( L 1) T T
codebooks designed with using AR predictors produce lower
1 [ y1 y1 ... y1 1 ]
c (3) distortion than codebooks with MA predictors, however the
... ... .... ... .... use of an alternative ARMA model in residual LSF prediction,
which is untouched in the literature, can be more
c
K [ y K ( 0)T (1) T
y K ... y K
( LK 1) T T
] Lp1 advantageous. The ARMA(1,1) predictor is
____________________________________________________________________________________________________________
542
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
1 T
SD d SD 2 (e j 2n / N ) dB 2
T i 1
(16)
____________________________________________________________________________________________________________
543
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(b) SD and outliers in residual LSF joint codebook design with Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing, Vol.1,
pp.41-44, May 1998.
ARMA prediction
[6] J. Skoglund, J. Lindén, “Predictive VQ for noisy channel spectrum coding
: AR or MA?” , Proc. IEEE Int. Conf. on Acoustics, Speech, Signal
Processing, May, 1996
[7] J.R. de Marca, “An LSF quantizer for the North-American Half-Rate
REFERENCES Speech Coder”, IEEE Transactions on Vehicular Technology, Vol.43,
No.3, August, 1994.
[1] A.V. McCree and T.P. Barnwell III, “ A Mixed Excitation LPC Vocoder
Model for Low Bit Rate Speech Coding”, IEEE Transactions on Speech [8] H. Ohmuro, T. Moriya, K. Mano, and S. Miki, “Vector quantization of
and Audio Processing, Vol.3, No.4, pp.242-250, July 1995 LSP parameters using moving average interframe prediction”,
Electronics and Communications in Japan, Part 3, Vol.77, pp.12-26,
[2] W.P. LeBlanc, B.Bhattacharya, S.A. Mahmoud, “Efficient Search and 1994.
Design Procedures for Robust Multi Stage Vector Quantization of LPC
Parameters for 4 kbps Speech Coding”, IEEE Trans. on Speech and [9] B. Wahberg, “ARMA spectral estimation of narrow band processes via
Audio Processing, Vol.1, No.4, pp.373-385, Oct., 1993. model reduction”, IEEE Transactions on Acoustics, Speech and Signal
Processing, vol. 38, July, 1990
[3] I.T.Lim, B.G. Lee, “Lossless pole-zero modelling of speech signals”,
IEEE Transactions on Speech, Signal and Audio processing, Vol.1, [10] M.G. Kang, B.G. Lee, “A generalized vocal tract model for pole-zero
No.3, pp.269-276, July, 1993 type prediction”, Proc. Int.Conf. on ASSP, S14, 10, 1988
[4] S. Ozaydin, B. Baykal, “Matrix quantization based linear predictive [11] L. Deng, L.J. Lee, etc, “Adaptive Kalman filtering and smoothing for
speech coding at very low bit rates”, Speech Communication, Vol. 41, tracking vocal tract resonances using a continuous valued hidden
Issues 2-3, pp:381-392, Oct. 2003 dynamic model”, IEEE Transactions on Audio, Speech and Language
Processing, Vol.15, No.1, 2007
[5] S. Nandkumar, K. Swaminathan, U. Bhaskar, “Robust Speech Model
based LSF Vector Quantization for Low Bit Rate Speech Coders”, in
____________________________________________________________________________________________________________
544
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Environment monitoring systems have gained a great types. The works in [2-4] propose wireless data logger
attraction to maintain human health and safety in human systems for monitoring temperature using ZigBee
working areas. To assure appropriate working conditions communication protocol. For building a wireless data logger
especially in safety-critical areas, wireless communication for thermal validation systems for pharmaceutical industry is
systems are mainly preferred. As a main component of such
given in [5]. Another data logger system is developed for
monitoring systems, wireless data loggers are required to collect,
store and display sensor data at high speed, lower cost and high industrial cooling applications [6]. The work in [7] aims to
efficiency. This paper presents the results of a design process propose a data logging system that store large numbers of
regarding a wireless data logger. The system is designed to data from analog signal over an extended long time. In [8], a
detect released toxic gases along with conventional video wireless sensor network that uses data logger for a real-time
monitoring of safety-critical areas. The system provides air quality monitoring system is presented. The study in [9]
measurement time and date stamped sensor data onto video introduces a wireless data logger system for real-time
signal for real-time monitoring of the environment. The system monitoring with remote control capability via PC and tablet.
also offers instant access to stored data using a mobile Although it has been successfully developed, today’s
application through Android devices. This provides cost
industrial market demands more cost-effective systems.
effective solution from end-user perspective.1
In this work, wireless data logger system is designed to
Index Terms—data logger, android, wireless communication, monitor real-time measurements, which is the combination of
remote control, microcontroller sensed and video recorded data, in a specified area.
Furthermore, this system enables end-users to control the
stored data with all Android smartphones. Hence, it provides
I. INTRODUCTION cost effective solution for environment monitoring system.
Proposed wireless data logger system may be beneficial in
Data logger is an electronic device that records several detecting unsafe or hazardous conditions for human health in
types of measurements over the time. It stores all data into the working or common areas. This forms the foundation for
memory storage and informs user at any time required. Data a security system with a view to protecting human health.
loggers, which use digital technologies, are commonly Besides, sensor modules that used for gas detecting are
equipped with sensor, microcontroller, external memory, real protected from external effects now that these modules are
time clock, and wireless communication module. However, not depended on external resources. Therefore, data
structure of the data logger may be varied. Typically, data transmission could not be interrupted from external effects
loggers work together with sensors to take input data. This which is the another advantage of the developed data logger
data is converted to binary data and then analyzed by full- system.
featured software to store for further processes. With This work is developed as an industry sponsored
emerging technologies, functionality of the data loggers has undergraduate project. In fact, it reports the results of the
been enhanced by generic programming techniques. system developed by undergraduate students. All
Using data logger has been considered in many development stages of the system, such as PCB layout,
applications for different purposes. A survey on data logger software, hardware and prototyping, were completed at
system is presented in [1]. In [1], several studies in data Atılım University laboratories.
logger field are reviewed. More recently, developed systems The paper starts with describing the concept of the
are proposed to monitor sensed data in various environment developed system with a block diagram as given in the next
section. In section III, hardware development of data logger,
software architecture, development of end-user application,
____________________________________________________________________________________________________________
545
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
and implementation of the system is presented. Conclusions as usual, in that these sensors have an ability to obtain wide
and discussions are drawn at the end of the paper. range of measurements, from temperature to carbon dioxide.
After selecting gas sensors, an ADC converter is required to
II. CONCEPT OF THE SYSTEM read binary data. Due to the fact that low power consumption
and flexible connectivity applications play important role in
The system that presented in this article comprises of three
measurement systems, a 16-bit Microcontroller Unit (MCU)
main modules which are measurement, data management and
is chosen which contains 8-bit ADC. Digitalized sensor data
On Screen Display (OSD). The block diagram of the
are stamped with date and time by using Real Time Clock
proposed data logger system is shown in Fig. 1. (RTC) that is connected to MCU with I2C (Inter-Integrated
Circuit) protocol. Another connection with MCU is Bluetooth
Measurement Data Management module. To access into MCU from Bluetooth module, UART
(Universal Asynchronous Receiver/Transmitter) protocol is
used. Bluetooth modules are designed to address the needs of
Sensors ADC Sorting Storage new wireless applications. In this work, Bluetooth module
enables end-users to control the stored data through a
developed application for smartphones. It works on the baud
RTC Bluetooth Module rate of 9600 and has approximately the range of 12 meter. In
the last stage of data management module, data is stored in a
On Screen Display (OSD) Multi Media Card (MMC) which is connected to MCU with
Serial Peripheral Interface (SPI) protocol as it provides
Camera synchronous data transmission.
OSD module is the last part of the system contains another
MCU for retrieving data from data management module. To
LCD Overlaying
provide data transmission between management module and
OSD, a RF (Radio Frequency) transmitter in management
Fig. 1 Block diagram of the system. module and a RF receiver in OSD module are used. Both RF
modules use ISM (Industrial, Scientific and Medical)-433
In measurement module, the amount of gases released in a MHz band. These RF modules work on the baud rate of 600
specified area is measured by means of gas sensors. and have approximately 300 meter outdoor range. In OSD
Measured data is gathered and stored in data management module, MCU is connected to a single-channel monochrome
module, which is the core part of the data logger system. OSD generator (MAX7456). It is used to create on-screen
Since used sensors are analog sensor, an Analog-to-Digital menus and other video overlays. It is NTSC and PAL
Converter (ADC) is required to read measured data. For this compatible and displays up to 16 rows × 30 characters. SPI
purpose, the data management module first converts analog protocol is utilized for maintaining data transmission from
data to digital one. Then, it stamps sensed data with date and MCU to OSD generator. The main working principle of OSD
time. Data storing is the last phase of the data management generator is overlaying received data from MCU onto video
module. The module is able to send data to Bluetooth signals recorded by a wireless surveillance camera. As a
module, as well. This enhances the data logger to transmit the consequence, the security monitor is linked to output of the
data directly to end user. The last part of the system is OSD OSD generator for displaying real-time recordings with date
module which retrieves time and date stamped data in order and time stamped sensor data. The whole system is
to put them on the video signals of an area that inclined to represented in Fig. 2.
observe. It should be noted that a wireless surveillance
camera is used for the purpose of observing an area. The
details of the system are given in the next section.
____________________________________________________________________________________________________________
546
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
ADC
Data Storage
Data request:
set time and
date
Is data
request
received
No
Fig. 3 Working principle of data logger Fig. 5 Application main panel and received data log
____________________________________________________________________________________________________________
547
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The other feature of the application is “Old Records” that system can detect released toxic gases in working places. It
lists the former requests of the end-users and corresponding also merges time and date stamped sensor data with
responses from data logger. It can also send these logs to surveillance records for real-time monitoring. Thus,
users as an e-mail or SMS (Short Message Service) at any propagation of toxic gases can be prevented by using
time needed. Fig. 6 shows the recorded the requests and developed wireless data logger system. Moreover, an Android
responded data logs. application is developed for the data logger system using
Eclipse software. Hence, the sensed environmental data can
be accessed and controlled by smartphones. The system
enables end-users to run data through the system without
using wireless remote controller or control unit. This provides
low cost solution and instant inquiry transaction.
As a result, it is expected that presented wireless data
logger system could fulfill industrial demands. In this
context, it can be adapted to various applications such as
mining, defense and agriculture system applications.
Therefore, in future, the system that is presented in this article
may be developed for commercial purposes.
ACKNOWLEDGMENT
____________________________________________________________________________________________________________
548
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—in this study, used in GSM1800, Wi-Fi and WiMAX diagram due to impedance discord [2]. In the present study, a
applications, high gain microstrip line feed 2 x 1 patch array microstrip line feeding technique that is easy to design and
antenna design work was done. Antenna simulation of this produce is presented. Due to certain planar shortages of
antenna design made using HFSS 3D gain and directivity microstrip line feeding [5], an inset fed microstrip antenna
characteristics were simulated. Since antenna impedance is
was designed. The side radiating microstrip inset fed antenna
greater than line impedance, impedance matching has been made
with inset – fed microstrip line and the quarter-wave is presented in Figure 1.
transformer. The results show that the proposed antenna
parameters are sufficient for GSM1800, Wi-Fi and WiMAX
applications.
I. INTRODUCTION
As a result of advances in wireless systems in new
generation communications and increasing demand for
wireless applications promoted the significance of low profile,
integrated and high gain antenna designs. Microstrip antennas
used in wireless network services, satellite and missile
navigation, radar systems, biomedical field and in diverse
applications have a seamless record in providing for these
requirements [1]. There are studies that attempt to improve Fig. 1 The side radiating microstrip inset fed antenna
Microstrip antennas by selection of special conductors and
modification of geometric structures (array of antennas) Various different methods were developed to analyze
because of their disadvantages such as low gain and narrow microstrip antennas. The most popular models among these is
bandwidth. Microstrip antennas produced with printed circuit transmission line model, cavity model and full-wave model.
technology includes a thin layer of low loss insulator material Although transmission line model is the simplest method
with a radiating conductive layer on one side and conductive among analysis methods, it is proficient in analyzing the
ground layer that completely covers the non-conductive. physical structure. However, its accuracy is lower than the
Microstrip antennas are basically fed by the microstrip or a other methods and it is inadequate in modeling multilayer
coaxial line. Along with technological advances, researchers structures. When compared to transmission line model, cavity
have focused on different feeding structures and developed model has a better accuracy, however it also entails a complex
three different feeding techniques [2, 3, 4]. Several factors structure. In addition, the cavity model is effective in physical
play a role in feeding technique selection. The most analysis similar to the transmission line model, but this model
significant of these is the effective transfer of antenna power is also not adequate for modeling multilayer structures. Full-
between the feed line and radiation structure with the feeding wave model is the most accurate model among these methods.
technique. This occurs thanks to impedance compliance However, this model is quite complex and has low capability
between the layers. Unwanted radiation could cause increases for physical analysis [6]. The present antenna model was
in side lobe level and cross polarization amplitude in radiation designed with transmission line model. Application interface
____________________________________________________________________________________________________________
549
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
displayed in Figure 2 was developed in NetBeans IDE 8.0.2 calculated using equations 1 - 6 used in the transmission line
editor [7] using JAVA programming language. model [1 – 11] and exhibited below.
( r 1) (1)
W c / 2 fo
2
c
L 2 (2)
2 fo r
r 1 r 1 1 12 h (3)
e
2 2 W
W
( e 0 . 3 )( 0 . 264 )
L 0 . 412 h h (4)
W
( e 0 . 258 )( 0 .8 )
h
Fig. 2 Transmission Line Mode developed using JAVA L R in (5)
yo arccos
Microstrip antennas usually have low gain. In some cases, a ( ) R ino
singular antenna could not provide the desired requirements c 4 . 65 x10 12
(e.g. gain, shifting the radiation to the desired direction). Thus, xo (6)
2 eff fo
to improve the performance of the antenna, an array of
antennae that are formed by placing more than one antennas
geometrically are utilized [2 – 8]. Due to simplicity of design Rectangular microstrip feed antenna input and output
and their geometrical structures, microstrip antennas are one parameters are given in Table 1 below:
of the most frequently used antenna arrays [9]. Each
TABLE I
component of an antenna array could be selected from RECTANGULAR MICROSTRIP FEED ANTENNA INPUT AND OUTPUT PARAMETERS
elements with different structures, but to calculate the total
radiation pattern of the array, similar elements are preferred. Inputs
To obtain a radiation pattern in the desired direction, radiation Solution Frequency (fo) 2.45GHz
emitted from each antenna should be added to each other in Substrate Thickness (h) 3.175mm
the desired direction and should eliminate one another in Loss Tangent (tanδ) 0.0009
undesired directions. Demand for multi-frequency antenna 2.17
Dielectric Constant (εr)
terminals that have the capacity to receive multiple services
Conductor thickness (t) 0.035mm
provided by different wireless technology networks today is
increasing along with the available high gain antennas. In the Outputs
present study, GSM1800 technology that contains 1.8GHz Patch Width (W) 48.63mm
GSM frequency, Wi-Fi that contains 2.4GHz ISM frequency, Patch Length (L) 39.72mm
and WiMAX that contains 2.9GHz ISM frequency were Microstrip Line Width (Wf) 2.868mm
referenced [8]. Inset Distance (yo) 12.375mm
Inset/Notch Gap (xo) 4.71mm
II. ANTENNA ARRAY DESIGN Micro strip line impedance (Zin) 100Ω
Theoretical calculations were calculated based on the
transmission model. Since numerical values obtained with Where εr is the dielectric constant for the insulator material
theoretical calculations and obtained simulation results were
used in the designed antenna, εe is the effective value of the
not at the desired level, certain parameters were changed
manually to obtain adequate results. For simulation findings, insulator material, c is the speed of light, f0 is the resonance
High Frequency Structure Simulator (HFSS) [10] based on frequency, h is the height of the insulator material and ∆L is
full-wave finite elements method and widely used in the the line expansion. Wf depicts the width of the feed line, y0
analysis of electromagnetic structures was utilized. As depicts the position of the feeding point, xo depicts the space
insulator material “Rogers Ultralam1217™” was selected for between the feeding point and the patch, Rin0 depicts the
the constructed antenna and dielectric constant (εr) was taken antenna input impedance before the feed was added, and Rin
as 2.17, dielectric thickness (t) was 0.035 mm and dielectric depicts the target input impedance. In general applications,
height (h) was 3.175 mm. Central frequency of the antenna antenna input impedance is selected as 50Ω. Microstrip
was designed as 2.45GHz and physical parameters were antenna obtained with transmission line model was multiplied
to create a 2 x 1 antenna array. Microstrip lines where
____________________________________________________________________________________________________________
550
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
551
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
G e D (8)
Fig. 8. 3D gain (top) and directionality (bottom) graphs for the proposed
IV. CONCLUSIONS
antenna In the present study, a 2x1 antenna array suitable for
GSM1800, Wi-Fi and WiMAX applications [15-16], with a
Analysis of the above graphs shows that the main lobe was central frequency of 2.45GHz and 10dB directionality was
at the direction of z-axis, maximum gain was 10.09dB and designed and simulation values were presented. The small-
maximum directionality was 10.14dB. The difference in gain sized and high gain microstrip antenna had a suitable structure
was due to the imperfection of the utilized dielectric insulator for the communication devices used today. S-parameters of
material. Since loss tangent of the material was 0.0009, it the developed antenna array demonstrate that it could serve as
caused a loss, albeit small. Antenna output values at the end of a multi-frequency (GSM1800+Wi-Fi+WiMAX) functional
the simulation are given in Table 2 below. antenna when adjusted for communication systems that
TABLE II operate in GSM1800, Wi-Fi, WiMAX bandwidths in outdoors.
ANTENNA OUTPUT VALUES AT THE END OF THE SIMULATION
ACKNOWLEDGMENT
Antenna No S11 (dB) Directivity Gain (dB)
This research was financially supported by the Researching
(dB) Projects Committee of the University of Dicle (DUBAPK)
Single -15.78 8.03 7.93 with project number 14–MF–71. We are grateful to DUBAPK
2x1 Array -31.98 10.14 10.09 for financial assistance.
____________________________________________________________________________________________________________
552
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES [10] ANSYS HFSS (ver. 15), Ansys Corporation. Canonsburg, PA, USA,
2014
[1] C.A. Balanis, Antenna Theory Analysis and Design, 3rd ed. Hoboken, [11] Mohammad Abdul Matin, “A design rulefor inset-fed rectangular
NJ, USA: Wiley- Interscience, 2005, p. 816. microstrip patch antenna”, January 2010, Universiti Teknologi Brunei
[2] O. Çolak, “Tek Darbe Mikroşerit Dizi Anten Tasarımı”, Yüksek Lisans [12] İ. Ataş, M.B. Kurt, and M. Ataş, “Açıklık Kuplajlı Mikroşerit Yama
Tezi, İstanbul 2013 Antenler için Yapay Sinir Ağ Modeli”, Dicle Üniversitesi Mühendislik
[3] Volakis, J.L., ve Eibert, T.F., (2007), “Antenna Engineering Dergisi, Cilt:4, Sayi:2, Sayfa: 69-75, 2013.
Handbook”, The McGraw-Hill Companies, Ohio/ABD. [13] M. S. Garba, “Design of Tri-Band Microstrip Patch Antenna”,
[4] Zhi, N. C. ve Michael Y. W. Chia, (2006) “Broadband Planar Antennas International Journal of Science and Research (IJSR) ISSN (Online):
Design And Applications”, John Wiley & Sons, New York/ABD. [6] 2319-7064
[5] Garg, ., Bhartia, P., Bahl, I., Ittipiboon, A., “Microstrip Antenna [14] Shu, P. and Q. Feng, “Design of a compact quad-band hybrid antenna
Design Handbook”, Artech House Antennas and Propagation Library, for compass/WiMAX/WLAN applications,” Progress In
1-2, 2001 Electromagnetics Research, Vol. 138, 585–598, 2013.
[6] Kumar, G. and Ray, “Broadband Microstrip Antennas”, Artech House. [15] Azini, A. S., M. R. Kamarudin, T. A. Rahman, H. U. Iddi, A. Y.
451p., K.P. 2003, USA. Abdulrahman, and M. F. B. Jamlos, “Transparent antenna design for
[7] NetbeansIDE 8.0.2. https://netbeans.org/ WiMAX application,” Progress In Electromagnetics Research, Vol.
[8] 2.4 GHz’de Yüksek Kazançlı Dairesel Polarizasyonlu Yama Anten 138, 133–141, 2013.
Dizisi Tasarımı, Alican Uysal, Osman Ceylan, Selçuk Paker, H. Bülent [16] MOBILE BROADBAND MOBILE ANTENNA BROCHURE,
Yağcı İstanbul Teknik Üniversitesi Elektronik ve Haberleşme PCTEL Inc, www.antenna.
Mühendisliği Bölümü Sarıyer, İstanbul, URSI-TÜRKİYE’2014 VII.
Bilimsel Kongresi, 28-30 Ağustos 2014, ELAZIĞ
____________________________________________________________________________________________________________
553
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Wireless technology is the key technology to In this cognitive radio driven hospitals, devices are
eliminate the dense wire ropes from hospitals and far access to categorized as primary and secondary devices . Primary devices
medical devices. In order to overcome the problem of bandwidth are the ones that are vial and have higher priority to have
scarcity, cognitive radio driven hospitals are introduced and communication. Secondary devices are the ones which have
devices are divided in two categories. The first category is primary
low priority and thy can wait until the frequency band get
devices and the second one is secondary devices. Primary devices
has very high priority and their communication is vital for the vacant. So that, primary devices should not be interference by
hospital and patients, so that no interference should be made with other devices as their data are so valuable. In cognitive radio
such devices. Secondary devices are the ones which has lower technology, the secondary users sense the bandwidth and in the
priority and they can wait until the primary devices do their case of vacancy, the begin communication, otherwise, they wait
communication and then, they begin to use the allocated spectrum. for predetermined moment and sense the frequency band again
One of the key functions to assure that there will be no [1]. There are many sensing algorithms including ED, matched
interference is a reliable spectrum sensing method. This method filtering, wavelet based detection, cyclostationary detection [8]
should be a simple one to be able to implement it in the secondary and covariance based detection [7] methods with their different
devices. Among all the sensing methods, energy detection (ED)
required parameters, advantages and disadvantages. For
based spectrum sensing is very popular. In order to improve the
performance of ED, double threshold ED (DTED) method is instance, cyclostationary sensing methods needs the data about
the cyclic frequency of the primary user and in matched filter
introduced in literature. In this paper, a new algorithm is intruded
for DTED considering previous sensing period results in detection sensing method, the waveform information is needed [4]. As
procedure by using a memory stick. Memoryful DTED (MDTED) mentioned before, each sensing method has its own advantages
improves the performance of DTED considerably by only the cost and disadvantages that makes them unique in their kind, but in
of delay in secondary devices communication which has less value practice, the most important factor is the simplicity of the used
than the improvement of the detection method performance. method. Among all the methods mentioned, ED method is the
most simplest one to use in devices. ED method is based on
Keywords— wireless technology hospitals, cognitive radio comparison of the energy of collected samples with a threshold
driven hospitals, spectrum sensing, energy detection, double calculated before [3] – [5]. In order to improve the performance
threshold energy detection, memory. of ED sensing method, cooperative sensing method and DTED
are introduced in literature [6]. Cooperative ED spectrum
sensing is not practical solution for CogMed as there are many
I. INTRODUCTION devices and in the case that each device use multiple sensing
It has been a long while that biomedical and e-health experts nodes, too complicated and busy environment will be the result
are trying to use wireless technologies in their field. The main of this process. The second solution is using double threshold
purpose of this matter, instead of eliminating the dense wire instead of single one. In double threshold the detection is made
ropes from the hospital environment, is to provide access from comparison of the energy of the signal with two pre-calculated
distance to the devices in the environment [1]. To meet this thresholds [4]. This process has improved the performance of
requirement, the number of wireless technology based hospitals ED method considerably, but for CogMed usage, as the inter
grow so fast that after a while the problems began to appear. symbol interference (ISI) with primary users could cost
One of the most important problems was the scarcity of somebodies life, the performance should be much higher with
bandwidth. The bandwidth of the medical wireless little additional computational complexity. In this paper we
communication is limited because of static requncy allocation recommend to use memory for the secondary user devices, so
which is done by governmental and non-governmental that, in the detection procedure, they could consider the result
commissions. In united states, the responsibility of spectrum of the previous sensing period. By MDTED, in the case of any
allocation is with Federal Communications Commission (FCC). error in the sensing period, the previously gathered information
In order to mitigate the scarcity of the spectrum, cognitive radio will help the algorithm to check the status of the channel
driven hospitals (CogMed) are introduced by researchers [2]. previously.
____________________________________________________________________________________________________________
554
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The rest of the paper is organized as follows. In section II, which is called probability of detection (Pd) can be shown as
The system model, ED method and some background follow:
information are provided. In section III, DTED and MDTED λ λ
Pfa = prob(E(t)> λ|H0) = Γ(u, )/ Γ(u)=Q( ) (5)
methods are introduced. In Section IV simulation results and /
analysis is done, followed by concluding remarks in section V.
Pd= prob(E(t)> λ|H1) = Qu(√2 , √ ) =
II. ENERGY DETECTION, SYSTEM MODEL AND λ (| | )
Q( ) (6)
BACKGROUND (| | )/
____________________________________________________________________________________________________________
555
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
As mentioned earlier, reliability is the most important factor Pd = p(M(t)> |H1) = p(E(ri)> |, E(ri-1), … , E(ri-c)> |H1)
in CogMed and having multiple detectors for having = p (E(ri)> |H1).p(E(ri-1)> |H1) … p(E(ri-c)> |H1)
cooperative detection is not possible. MDTED is a novel (| | ) T+1
= Q( ) (16)
approach to DTED which improves the detection performance (| | )/
of it. This algorithm is based on saving the energy of last Т
previously sensed signal samples and detection is made P0 = p(λ1< M(t)<λ2|H0) = (Q( ) −Q( ))T+1 (17)
/ /
considering these energies. Collection of Т consecutive energy
samples can be shown as follow: (| | )
P1=p(λ1< M(t)<λ2|H1) =( Q( )-
Ϻi(yi) = { E(yi-Т), E(yi-Т+1), E(yi-Т+2), … , E(yi)} (| | )/
In the case that the maximum of these decision energies fall (| | )
below the threshold λ1 the detection process decides the Q( ))T+1 (18)
(| | )/
hypothesis H0 and in the case that the maximum is higher than
λ2 the detection process decides the hypothesis H1. If the MBED detection performance is analysed and compared
maximum is between λ1 and λ2 in any way, the next maximum with DTED method in section IV.
value is going to be used until it doesn’t be in the restricted gap.
Because of environment in real life doesn’t change rapidly and IV. SIMULATION RESULTS AND ANALYSIS
in high order by time, this method will help our detection In this section the performance of MDTED is studied,
system to not only have the results of the time, it will use analysed and compared with very well-known DTED. All
previously gathered information also that will help decision simulations are done in MATLAB software using QPSK
process in big range. Considering the improvement of detection modulated random signals and i.i.d. noise samples with
process performance and reliability, T-1 sensing period delays Gaussian random variables. It is assumed that the channel
can be bearable. Also by choosing T small enough, it is possible doesn’t change while sampling in each sampling period. Based
to make the delay minimum. The algorithm of MDTED is as on IEEE 802.22, Pfa should be less or equal to 0.1 and Pd be
follow: more than 0.9. In the simulations Pfa is chosen as 0.1 and 105
test signals are used and averaged. DTED and MDTED is
Algorithm 3. MDTED spectrum sensing method studied with RAC equal to 0.5 and in memory section, T is
Input : Т, λ1, λ2, η equal to 2 unless it is told to be different. T equal to 2 mean the
Output : Yi detection method needs 2 previously sensed normalized
1: for each sensing period do energies that should be saved in memory and be used in the
2: for ζ=1:N samples do detection procedure.
3: if λ1< sample energy < λ2 Figure 1 shows the performance of ED, DTED and MDTED
4: do nothing in Gaussian channel models. Gaussian channel is the simplest
5: else channel model among all communication channel models.
6: ℮ζ(t) Energy of sample ζ , ζ ϵ [1,2,…,N] MBDTED with a memory equal to 2, has a much better
7: end for performance compared to DTED and ED in all SNR ranges.
8: ℮(t) = normalized energy of the selected With only saving 2 previously sensed signals energy in
total ℮ζ(t) samples detection process, a high range of performance can be achieved.
9: Ϻi(yi) Energy of previous Т received Unlike DTED, the performance of MBDTED doesn’t diminish
signals and ℮(t) suddenly in low SNRs like DTED.
10: M(t) = MAX { Ϻi(yi) } 1
0.5
16: Yi H0 0.4
18: Yi H1 0.2
____________________________________________________________________________________________________________
556
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
ED also in all SNR ranges by only the delay of 2 sensing Choosing the value of memory, based on the communication
periods. channel estimated in the cognitive radio driven hospitals and
1
acceptable delay in the sensing procedure, we can have the best
0.9
ED
DTED
performance needed in such areas.
MBDTED
0.8
0.7
0.6
V. CONCLUSION
Pd
0.5
0.4
Memory based double threshold energy detection or in other
0.3
words, double threshold energy detection with memory sensing
0.2
algorithm is introduced in this paper. It is shown that using
0.1
memory and saving the previously received signals energy in
0
the defined algorithm can affect the sensing performance and
-20 -15 -10 -5 0 5
SNR in dB improve the detection performance of the energy detection
Figure 2. ED, DTED and MBDTED sensing methods performance over based spectrum sensing in very high range. DTED was defined
Rayleigh channel. in literature to improve the performance and reliability of ED
method. We compared the MBDTED with ED and DTED
Figure 3 shows the performance of MBDTED whith methods in Gaussian and Rayleigh channels by using only 2
different memory values in Gaussian channels. In MBDTED previously saved signals and we saw the performance of this
without memory is the original DTED. By only using 1 method is much higher than both ED and DTED in wide SNR
previously sensed signal energy the detection performance range. Disadvantage of this method is the time delay after the
gets much higher. This value is getting better when we choose channel is free but considering the improvement of the sensing
the T value equal to 2. performance, T sensing period delay for the secondary user in
1 order to detect the vacancy of the signal could be bearable in
0.9
No Memory
T=1
cognitive radio driven hospitals. Also, by choosing T and
0.8
T=2 periodic sensing distance small enough, we can make the delay
0.7 minimum. This research findings help to understand the effect
0.6 of the memory in a sensing method and advantage and
disadvantages of that which helps find optimal solutions to
Pd
0.5
0.1
REFERENCES
0
-20 -15 -10 -5 0 5
SNR in dB [1] I. A. Mamoon, A. K. M Muzahidul-Islam, Sabariah
Figure 3. MBDTED sensing method’s performance without memory and Baharun, Shozo Komaki, Ashir Ahmed, “Architecture and
memory of saving 1 and 2 previously sensed signals energy over Gaussian communication protocols for cognitive radio network enabled hospital”
channel. IEEE International Symposium on Medical Information and
Communication Technology (ISMICT), 2015, pp. 170-174.
[2] I. A. Mamoon, A. K. M Muzahidul Islam, Ahmad
In figure 4, the performance of MBDTED is evaluated in
Shahrizal Sani, Sabariah Baharun, Shozo Komaki Toshio Wakabayashi,
different memory values in Rayleigh channel. This can be seen “Definition, design and priority management of a cognitive radio driven
from the figure that in this kind of channel model, again the hospital: CogMed”, IEEE Conference on Biomedical Engineering and
performance is getting better as the T value gets higher Sciences (IECBES), 2014, pp. 373-378.
[3] A. Eslami, S. Karamzadeh, “Performance Analysis of
1 Energy Based Spectrum Sensing over Nakagami-m Fading Channels
No Memory
0.9 T=1 with Noise uncertainty”, International Journal Of Electronics,
0.8
T=2 Mechanical Aand Mechatronics Engineering (IJEMME), 2016, vol:6, pp:
1101-1106.
0.7
[4] A. Eslami, S. Karamzadeh, “Performance analysis of
0.6
double threshold energy detection-based spectrum sensing in low SNRs
over Nakagami-m fading channels with noise uncertainty”, IEEE
Pd
0.5
____________________________________________________________________________________________________________
557
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[8] N. Han, S. H. Shon, J. O. Joo, and J. M. Kim, “Spectral [10] Z.Ye, G.Memik, and J.Grosspietsch, "Energy Detection
correlation based signal detection method for spectrum sensing in IEEE using Estimated Noise Variance for Spectrum Sensing in Cognitive
802.22 WRAN systems," Intern. Conf. Advanced Commun. Technology, Radio Networks, " in Wireless Communications and Networking
Korea, Feb. 2006. Conference 2008.Las Vegas, USA: IEEE Communications Society,
[9] Y. Zeng and Y.-C. Liang , "Eigenvalue-based spectrum Mar.-Apr.2008.
sensing algorithms for cognitive radio" , IEEE Trans. Commun. , vol.
57 , no. 6 , pp.1784 -793 , 2009.
____________________________________________________________________________________________________________
558
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The aim of this paper is to compare the performance is simple and their maintenance is easy [6]. There is also no
evaluation in terms of the efficiency, cogging torque and copper loss on rotor. However, one of the biggest
manufacturing cost for a permanent magnet synchronous disadvantages is their cogging torque [7]. Because of the high
generator (PMSG) with interior rotor of N35 and N42 NdFeB cogging torque, they do not commence electrical energy
permanent magnets (PMs) having same geometry separately
generation at low wind speed [2].
used in a micro wind turbine (MWT). Firstly N35 type PMs have
been used on an interior rotor, and then its performance has SWTs are designed as horizontal and vertical axis. Also,
been measured and calculated by a MWT setup. After that, N42 PMSGs inserted in SWTs are manufactured two types as outer
type PMs have been used on other interior rotor. According to and interior rotors [2, 8]. Although the PMSGs with interior
the obtained results, the PMSG with N42 type presented better rotor have more size than the PMSGs with outer rotor, their
efficiency than the PMSG with N35 type. But the PMSG with power density, moment, and efficiency are high. But, when
N42 type has induced a bit more cogging torque than the PMSG considering in terms of manufacturing cost and labor,
with N35 type. When comparing in terms of efficiency, a higher manufacturing cost and labor of the PMSGs with interior rotor
efficiency has been obtained by the PMSG with N42 type. are higher than the PMSGs with outer rotor. In addition to
these, PMSGs are constructed as axial and radial flux [9-10].
Keywords— PMSG, PM, efficiency, NdFeB, N42, N35 Output voltage and torque of the PMSGs with radial flux are
greater than the axial flux [11].
I. INTRODUCTION
Nowadays, the usage of renewable energy resources such II. PERMANENT MAGNETS
as wind, geothermal, tidal and solar in electricity production is Permanent magnet (PM) forms and elements that are used
gradually increasing. Wind energy is the most popular of these in PMSGs have a great impact on performance of PMSGs [12-
renewable energy resources. In order to convert wind energy 14]. PM forms, which are placed on an interior rotor with four
into electrical energy, wind turbines are employed [1]. poles, are illustrated in Fig. 1. A surface radial, a parallel, a
Wind turbines are divided into two classes as large wind breadloaf and a ring PM form are presented in Fig. 1. These
turbines (LWTs) and small wind turbines (SWTs) [2]. While are classified as the PMSGs with surface PM rotor.
LWTs usually are operated as grid-on, SWTs are utilized as Additionally, there are also the types of PMSG with PMs that
both grid-on and grid-off. Furthermore, as given in Table I, are embedded in a rotor.
SWTs are separated into three different classes [3]. According Magnetization profiles also are important during the PM
to the power of SWTs in Table I, there are SWT types usage. These are sine angle or sine direction, parallel, radial
employed by direct-drive and gear mechanism systems [1]. sine and radial profile. In addition, these are illustrated in Fig.
Depending on the powers of SWTs, induction generators 1 and the most popular of which is the parallel magnetization.
(IGs) and permanent magnet synchronous generators (PMSGs) PMs are made up of Aluminium-Nickel-Cobalt (AlNiCo),
are widely preferred [4]. PMSGs used SWTs are more Ferrite, Samarium-Cobalt (SmCo), and Neodymium-Iron-
advantageous than IGs because they have more power density, Boron (NdFeB) materials. Their operating temperatures and
higher torque and direct-drive ability [5]. Structure of PMSG magnetic flux density-magnetic field intensity (BHmax) are
TABLE I
GENERATION POWERS OF SWTS
Category Power (kW) Annual energy production (kWh) Tower height (m)
Micro wind turbine < 1.5 < 1,000 10–18
Small wind turbine 1.5-50 < 200,000 15–35
Small–medium wind turbine 50-500 < 1,800,000 25–55
____________________________________________________________________________________________________________
559
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
different from each other. The values are given in Table II. Temperature coefficient, βHci (% / °C) -0.6 -0.6
When Table II is examined carefully, although NdFeB PMs
have the highest BHmax value, their operating temperatures are III. APPLICATION
lower than others. Recently, NdFeB PMs are widely preferred Two rotors with surface magnet having N35 and N42
in PMSG design. [15-16]. NdFeB PMs were designed and manufactured. The dimension
The remanence magnetization value (Br) of AlNiCo of PMs was sized as 20 × 20 × 6 mm. The usage of the rotor
magnets among them is quite high. Due to the fact that they together with a PMSG stator is demonstrated in Fig. 2. A
are expensive and also can easily lose their magnetization designed MWT, the stator and the rotor dimensions are
property, they are not preferred lately. presented in Table IV. To reduce the cogging torque, the used
Ferrite magnets being another type of PMs that are PMs were selected in a breadloaf form. Moreover, to reduce
relatively lower cost and more commonly usage than AlNiCo the cogging torque, their placement was fulfilled according to
PMs. They are resistant to the loss of remanence a pole shifting method [7].
magnetization. BHmax values of both AlNiCo and Ferrite PMs After the rotors that manufactured two prototypes had
are lower than SmCo and NdFeB PMs. inserted in the PMSGs, they were mounted in the MWT, as
SmCo and NdFeB PMs are known as rare-earth magnets. given in Fig. 3. Lastly, their performance tests were carried
Their BHmax values are quite high compared the AlNiCo and out by means of a truck test [2]. During the truck test, the
Ferrite PMs. One of their biggest disadvantages is that they generated power from the MWT was transferred to a battery
are fragile. The remanence and demagnetization values are group, which was consisted of four batteries of 12 V and 60
high. Ah. The battery group was charged until 13.5 V. After the
In recent years, the remanence values of the sintered NdFeB charge, two load resistances of 600 W were used to discharge
PMs are increased above 1.0 T. In Table III, properties of N35 to the battery group. In order to measure the generated power
and N45 type NdFeB PMs that are sintered at different grade from the MWT with PMSGs used N35 and N42 NdFeB PMs,
are listed. Their performance analysis two type PMs were a data acquisition system, which was described and embedded
carried out in a micro wind turbine (MWT). The sintered PMs formulas in detail [2], was operated. The measurement of
offer up to 150°C operation temperatures. After the operation input power of the MWT was realized by two different
temperatures, SmCo PMs are preferred. SmCo PMs also have calibrated anemometers – one was Prova AVM-03, the other
different temperatures. Their operation temperatures rise at was Kestrel 3000 model. In addition to these measurements,
250°C. Studies continue to increase the operation wind speed was also checked by the truck display. The
temperatures of NdFeB and SmCo PMs. ambient temperature was registered by both the data
____________________________________________________________________________________________________________
560
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 2 The designed PMSG Fig. 3 The designed and constructed MWT
TABLE IV
THE DESIGNED MWT, STATOR AND ROTOR DIMENSIONS AND PROPERTIES
Fig. 4 The obtained powers as a function of wind speed Fig. 5 The obtained power coefficient as a function of wind speed
acquisition system and Extect HD200 measurement In the MWT having the PMSG with the surface magnet
instrument. To define the cogging torque, Crane rotor used N35 NdFeB PMs, the cut-in was measured 2.7 m/s.
Electronics/UTA-451-0020-OP 5 Nm brand/model transmitter In other, which was the MWT having the PMSG with the
and Crane Electronics/TO-890-01CR-0-EUR brand/model surface magnet rotor used N42 NdFeB PMs, the cut-in was
display, which was described in detail [7], were utilized. taken 4.1 m/s.
Depending on wind speed, the generated powers from the
IV. RESULTS AND DISCUSSION MWTs and their power coefficients are presented in Fig. 4
The truck test was carried out by using of the PMSGs and 5. When Fig. 4 was examined, the nominal power of the
having two rotors with surface magnet separately used N35 MWT that has N35 NdFeB PMs at 11 m/s wind speed was
and N42 NdFeB PMs in the MWT. The air pressure was taken 177 W. Also, the nominal power of the MWT that has N42
as 1020 hPa to calculate the input power of the MWT. The air NdFeB PMs at 11 m/s wind speed was 210 W. In Fig. 5, the
density was obtained 1.221 kg/m3. power coefficients of the MWTs having N35 and N42 NdFeB
____________________________________________________________________________________________________________
561
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
ACKNOWLEDGMENT
The author would like to acknowledge that this work was
supported by TUBITAK under No. 114E419.
REFERENCES
[1] H. Polinder, J. A. Ferreira, B. B. Jensen, A. B. Abrahamsen, K. Atallah
and R. A. McMahon, “Trends in wind turbine generator systems,”
IEEE Journal of Emerging and Selected Topics in Power Electronics,
vol. 1, pp. 174–185, Sep. 2013.
Fig. 6 The obtained cogging torque as a function of rotor position [2] H. Mamur, “Design, application, and power performance analyses of a
micro wind turbine,” Turk J Electr Eng Co, vol. 23, pp. 1619–1637,
2015.
PMs were defined 0.277 and 0.388, respectively. [3] H. Mamur, M. Ari, F. Korkmaz, I. Topaloglu, A. Cicek and E. Bektas,
The highest cogging torques of the MWTs having N35 and “Micro wind turbines and power performance analyses,” in Proc.
UNITECH 2015 International Scientific Conference, pp. 59-62, vol. 1,
N42 NdFeB PMs were measured as 0.1 Nm and 0.17 Nm, Gabrovo, Bulgaria, 20-21 November 2015.
respectively, as shown in Fig. 6. In the MWT with N42 [4] H. Mamur, “Comparison of cogging torque reduction methods for
NdFeB PMs, the cogging torque was occurred a quite high micro wind turbine using permanent magnet synchronous generator,”
value. In order to reduce the value, it is necessary that other in Proc. UNITECH 2015 International Scientific Conference, pp. 63–
67, vol. 1, Gabrovo, Bulgaria, 20-21 November 2015.
cogging torque reduction methods have to be used together [5] M. M. Koo, S. M. Jang, Y. S. Park, H. I. Park and J. Y. Choi,
with the pole shifting method. “Characteristic analysis of direct-drive wind power generator
When examining the literature regarding the study, Lee et considering permanent magnet shape and skew effects to reduce torque
al. [8] executed a study with outer rotor PMSG. The cogging ripple based on analytical approach,” IEEE Transactions on Magnetics,
vol. 49, pp. 3917–3920, Jul. 2013.
torque of about 0.18 Nm was measured in their application [6] N. Madani, A. Cosic and C. Sadarangani, “A permanent magnet
used PMs of Br 0.43 T. Also, Jang et al. [10] manufactured an synchronous generator for a small scale vertical axis wind turbine,” in
MWT of 1.5 kW employed both NdFeB of 1.26 T and Ferrite Proc. 2015 IEEE International Electric Machines & Drives
type PMs. They tried on the usability of Ferrite PM instead of Conference (IEMDC), pp. 48-52, Coeur d’Alene, ID, 10–13 May 2015.
[7] H. Mamur, I. Topaloglu, F. Korkmaz, M. Ari and I. Atacak, “Design
NdFeB PM. Although the PMSGs generated same power, the and experimental analysis for reduction of cogging torque by pole
PMSG with Ferrite type PM had larger volume than the shifting in permanent magnet synchronous generator,” Elektron
PMSG with NdFeB PM in reference to their results. They saw Elektrotech, vol. 20, pp. 39-43, 2014.
an increase of manufacturing cost. At low speed conditions, [8] G. C. Lee, S. M. Kang and T. U. Jung, “Permanent magnet structure
design of outer rotor radial flux permanent magnet generator for
Ani et al. [17] compared six different SWT in terms of the reduction cogging torque with design of experiment,” in Proc. 2013
energy yield and the generated electricity cost. Their cut-in International Conference on Electrical Machines and Systems, pp.
speeds were observed between 2.5 and 4.0 m/s. The 315–319, Susan, Korea, 26–29 October 2013.
manufactured MWT with N35 type PM was compared with [9] A. A. Pop, F. Gillon and M. M. Radulescu, “Modeling and permanent-
magnet shape optimization of an axial-flux machine,” in Proc. 2012
regard to cut in, manufacturing cost, energy yield and XXth International Conference on Electrical Machines (ICEM), pp.
generated electricity cost in [2] in detail. Because of the usage 357–363, Marseille, France, 2–5 September 2012.
of N42 type PMs, a small increase of 20 € occurred in the [10] S. M. Jang, H. J. Seo, Y. S. Park, H. I. Park and J. Y. Choi, “Design
manufacturing cost of MWT. The value was below 1%. and electromagnetic field characteristic analysis of 1.5 kW small scale
wind power generator for substitution of Nd-Fe-B to ferrite permanent
According these obtained results, these were quite acceptable magnet,” IEEE Transactions on Magnetics, vol. 48, pp. 2933–2936,
when the carried out study was compared the other studies in Nov. 2012.
literature. [11] A. A. Pop, F. Jurca, C. Oprea, M. Chirca, S. Breban and M. M.
Radulescu, “Axial-flux vs. radial-flux permanent-magnet synchronous
V. CONCLUSIONS generators for micro-wind turbine application,” in Proc. 2013 15th
European Conference on Power Electronics and Applications (EPE),
In this study, the performance analyses of both N35 and pp. 1–10, Lille, France, 2–5 September 2013.
N42 type PMs for interior rotor PMSGs were carried out. For [12] Z. Chen, C. Xia, Q. Geng and Y. Yan, “Modeling and analyzing of
this purpose, two surface-mounted rotors that had N35 and surface-mounted permanent-magnet synchronous machines with
optimized magnetic pole shape,” IEEE Transactions on Magnetics, vol.
N42 NdFeB PMs were designed and manufactured. These 50, pp. 1–4. Nov. 2014.
were tried on an MWT with PMSG inserted in the rotors [13] Z. S. Du and T. A. Lipo, “Improved use of rare earth permanent
separately. According to the obtained data, the generated magnet materials and reduction of torque pulsation in interior
nominal power of the MWT having N35 NdFeB PMs was permanent magnet machines,” in Proc. 2015 IEEE Energy Conversion
Congress and Exposition (ECCE), pp. 1773–1780, Montreal, QC, 20–
obtained 177 W. For the other MWT having N42 NdFeB PMs, 24 September 2015.
____________________________________________________________________________________________________________
562
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[14] Z. Chen, Z. Li, H. Ma and Y. Guo, “Optimization and analysis of [16] R. Lacal-Arántegui, “Materials use in electricity generators in wind
permanent-magnet synchronous machine with eccentric magnetic pole turbines–state-of-the-art and future specifications,” Journal of Cleaner
shape,” in Proc. 18th International Conference on Electrical Machines Production, vol. 87, pp. 275–283, 2015.
and Systems (ICEMS), pp. 1181–1186, Pattaya, Tailand, 25–28 [17] S. O. Ani, H. Polinder and J. A. Ferreira, “Comparison of energy yield
October 2015. of small wind turbines in low wind speed areas,” IEEE Transactions on
[15] M. J. Kramer, R. W. McCallum, I. A. Anderson and S. Constantinides, Sustainable Energy, vol. 4, pp. 42–49, (2013).
“Prospects for non-rare earth permanent magnets for traction motors
and generators,” JOM, vol. 64, pp. 752–763, 2012.
____________________________________________________________________________________________________________
563
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Hodgkin-Huxley (HH) neuronal model has been widely squad. In fact, it is a five state model that are the membrane
accepted neuronal model in neuroscience. The variation of the ionic potential and four ion channel currents modeled as an electrical
currents in neuron cell causes the variations in the membrane circuit. Based on the HH model, there are developed electrical
potential. The level of membrane potential indicates the activation circuits to realize its behavior and stimulate under different
and inactivation dynamics. In this paper, in order to observe the
conditions. At the same time, in order to produce practical and
unmeasurable states and parameters of HH neuron accurately,
Runge-Kutta discretization based nonlinear observer is designed. In implementable neuron model, different neuron models are
numerical simulations, the membrane potential is measured and the developed based on the HH model such as namely FitzHugh
ionic currents are estimated. The numerical results provide accurate Nagumo model (FHN), Morris-Lecar model, Hindmarsh-Rose
estimation results that can be used both in monitoring and control model and etc. [1]. In order to analyse the behaviour of neuron
of neuron dynamics. models, some of models are realized using electronic
components [3]-[5].
Keywords—Nonlinear observer, state estimation, Hodgkin-Huxley Nonlinear observers have been used for considerably
neuronal model, discretization based gradient observer, sliding- amount of applications in the subject of such as state estimation,
mode observer, extended Kalman filter, RMSE. parameter estimation, fault detection and isolation, disturbance
estimation, unknown input estimation and other applications.
Therefore, in literature, various types of the nonlinear observers
I. INTRODUCTION can be found for a specific application. As a leading work on
Mathematical model of a biological system presents an state observers have been first published for linear systems [6],
understanding for the behavior of the system. Recently, and then extended for nonlinear systems [7]. With the
mathematical models of the bacteria populations, diseases, requirements on the state estimation, there have introduced
microbiological organisms and nerve cells have increasing several nonlinear observers such as extended-Luenberger
applications for biology, medicine, biomedical, neuroscience observer [8], extended-Kalman filter [9], sliding-mode
fields. In this work, we have interested on Hodgkin-Huxley observer [10], [11], high-gain observer [12], Takagi-Sugeno
neuronal model. In general, nervous systems have very fuzzy observers [13], Runge-Kutta observer [14]-[16] etc. The
complex structures with largely interconnected neuron cells. nonlinear observers mentioned above are based on the
These neurons carry out the computational and communication mathematical model of the nonlinear system.
tasks with electrical potentials. The charge ionic distribution of In this study, some of nonlinear observers are chosen to
the neuron cell generates of the electrical potential on the estimate the states of HH neuronal model. For the HH neuronal
membrane potential. When a membrane has enough action model, the membrane potential is assumed to be measurable but
potential, then firing threshold can be reached then there exist the currents or auxiliary variables are assumed to unmeasurable.
spikes on the membrane voltage [1]. The communication with The aim is to estimate the unmeasurable states of the HH
neighboring neurons or the information patterns stored on the neuron model for an observer based future work such as in
neurons are occurred based on this membrane potential such neuroscience medicine; there exist some mental diseases to
that the control of the membrane potential by ionic currents is treat using external stimulus of the nerve cells. The designed
an important subject of the neuroscience and biomedical observers are selected purposely such that extended Kalman
engineering. filter (EKF), sliding-mode observer (SMO), and discretization
In neuroscience, Hodgkin-Huxley (HH) model is a first based gradient observer (DBGO). The chosen and designed
introduced model of the neuron cell [2]. The mathematical nonlinear observers are applied to estimate the states of the HH
model of the HH model explaining the relations between neuron model then estimation results are plotted and root-mean
membrane potential and ion channels is experimentally squared errors (RMSE) are given in Table. As a general result,
constructed using the electrical behavior of the giant axon in state estimation results are accurately obtained for the HH
1 Corresponding author
____________________________________________________________________________________________________________
564
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
neuronal model of future applications in neuroscience, where it is called as the discretized model of the continuous-
biomedical engineering and medicine. time system. For a Lipschitz nonlinear system the stability of
This paper is organized as follows: In Section 2, nonlinear RK discretization is shown using sufficiently small step size
observers, which are designed here, are explained in detail. [17].
Section 3 presents the Hodgkin-Huxley neuron model with
mathematical dynamics. The Section 4 illustrates the state A. Extended Kalman Filter
estimation results of the HH neuronal model using designed The EKF uses a recursive algorithm consisting of two parts
nonlinear observers. The discussions about the design namely prediction and measurement correction [9]. Consider
conditions and application results of the observers are given in the following state-space model of a discretized nonlinear
Section 5. dynamic system of (1) the states of which are to be estimated,
𝐱[𝑛 + 1] = 𝐟(𝐱[𝑛], 𝐮[𝑛]) + 𝐰[𝑛]
II. NONLINEAR OBSERVERS
𝐲[𝑛 + 1] = 𝐠(𝐱[𝑛], 𝐮[𝑛]) + 𝐯[𝑛]
Consider a n𝑡ℎ order continuous-time nonlinear multi-input 𝐰 ~ 𝒩(0, 𝐐) (6)
multi-output (MIMO) system: 𝐯 ~ 𝒩(0, 𝐑)
𝐱̇ = 𝐟(𝐱, 𝐮) where 𝐱[𝑛] is a N-dimensional state vector, 𝐮[𝑛] ∈ ℜR is
𝐲 = 𝐠(𝐱, 𝐮) the vector of input signals and 𝐲[𝑛] ∈ ℜ𝑄 is the vector of
𝑥1 (𝑡) ∈ 𝑋1 , … , 𝑥𝑁 (𝑡) ∈ 𝑋𝑁 , ∀𝑡 ≥ 0 (1) output signals. In (6), 𝐟(. ) is the discrete model of the nonlinear
𝑢1 (𝑡) ∈ 𝑈1 , … , 𝑢𝑅 (𝑡) ∈ 𝑈𝑅 , ∀𝑡 ≥ 0 system. The random variables 𝐰 and 𝐯 represent the process
where 𝐱(𝑡) ∈ X ⊂ ℜN is the state vector, 𝐮(𝑡) ∈ U ⊂ ℜR is and measurement noises, respectively, which are assumed to
the vector of measured control inputs and 𝐲(𝑡) ∈ ℜ𝑄 is the have independent and normal probability distributions with
vector of outputs measurements. Nonlinear dynamics are zero mean. Moreover, they have uncorrelated 𝐐 and 𝐑 noise
subject to state and input constraints written as covariance matrices. In the measurement correction stage a
posteriori state estimates 𝐱̂[𝑛] and the a posteriori error 𝐏[𝑛]
𝑋𝑖 = {𝑥𝑖 ∈ ℜ | 𝑥𝑖𝑚𝑖𝑛 ≤ 𝑥𝑖 ≤ 𝑥𝑖𝑚𝑎𝑥 },
are calculated using current measurements and the observer
𝑈𝑖 = {𝑢𝑖 ∈ ℜ | 𝑢𝑖𝑚𝑖𝑛 ≤ 𝑢𝑖 ≤ 𝑢𝑖𝑚𝑎𝑥 }. (2)
model. Thus, error covariance of the estimator is minimized.
It is assumed that the functions 𝐟𝑖 and 𝐠 𝑗 [𝑖 = Time update equations for prediction stage are as follows,
1, … , 𝑁 , 𝑗 = 1, … , 𝑄] are known and continuously 𝐱̂ − [𝑛] = 𝐟̂(𝐱̂ − [𝑛 − 1], 𝐮[𝑛 − 1]),
differentiable with respect to the control inputs and the state 𝐏 − [𝑛] = 𝐀[𝑛]𝐏[𝑛 − 1]𝐀𝑻 [𝑛] + 𝐐. (7)
variables and also the state variables are not available for Similarly, measurement update equations related to
measurement. The problem is to get the estimates 𝐱̂(𝑡) of the correction stage are,
unmeasured states of the system (1) by using only the available
input and output measurements. The nonlinear system is given 𝜰[𝑛] = 𝐏− [𝑛]𝐇𝐓 [𝑛](𝐇[𝑛]𝐏− [𝑛]𝐇𝐓 [𝑛] + 𝐑),−1
in Eq. (1) is discretized to get samples from the time instants. ̂[𝑛] = 𝐱̂ − [𝑛] + 𝚼[𝑛](𝐲[𝑛] − 𝐠(𝐱̂ − [𝑛], 𝐮[𝑛 − 1])),
𝒙
(8)
The Runge-Kutta (RK) discretization method is adopted for 𝐏[𝑛] = (𝐈 − 𝚼[𝑛]𝐇[𝑛])𝐏− [𝑛],
obtaining discretized models of the continuous-time nonlinear where 𝐱̂[𝑛] is state estimation vector, 𝚼[𝑛] and 𝐏[𝑛]
system. The states and the output values of the system which matrices are Kalman gain and error covariance matrices,
belong to the next sampling time as in compact form can be respectively. It is assumed that the functions 𝐟̂(. ) and 𝐠(. ) are
predicted as differentiable with respect to 𝐱 and 𝐮 parameters where 𝐀[𝑛]
𝐱̂[𝑛 + 1] = 𝐟̂(𝐱̂[𝑛], 𝐮[𝑛]) = 𝐱̂[𝑛] + 𝐤[𝑛] and 𝐇[𝑛] matrices are the Jacobian matrices as
𝐲̂[𝑛] = 𝐠(𝐱̂[𝑛], 𝐮[𝑛]) (3) ∂𝐟̂ ∂𝐠
𝐀[𝑛] = | , 𝐇[𝑛] = |
where ∂𝐱 𝐱 = 𝐱̂[n − 1] ∂𝐱 𝐱 = 𝐱̂[n − 1] (9)
𝐮 = 𝐮[n − 1] 𝐮 = 𝐮[n − 1]
𝑘11 + 2𝑘12 + 2𝑘13 + 𝑘14
1 𝑘21 + 2𝑘22 + 2𝑘23 + 𝑘24 These matrices, which are updated at every sampling time,
𝒌[𝑛] = [ ]
6 ⋮ are used recursively in prediction and correction update
𝑘𝑁1 + 2𝑘𝑁2 + 2𝑘𝑁3 + 𝑘𝑁4 (4) equations of the EKF.
1
= (𝐤 𝟏 + 2𝐤 𝟐 + 2𝐤 𝟑 + 𝐤 𝟒 ) B. Sliding-Mode Observer
6
𝑘𝑖𝑗 variables with 𝑖 = 1, … ,4 and 𝑗 = 1, … , 𝑁 are explicitly The sliding-mode observers are known for robustness to
defined as uncertainties and finite-time convergence properties [10], [11].
SMOs construct a sliding-motion of output estimation error
𝐤𝟏 = 𝑇𝑠 𝐟(𝐱̂, 𝐮) between the measured system output and observer output.
𝐤𝟐 = 𝑇𝑠 𝐟(𝐱̂ + 𝟎. 𝟓𝐤 𝟏 , 𝐮) There are some applications of the SMOs for state and
𝐤𝟑 = 𝑇𝑠 𝐟(𝐱̂ + 𝟎. 𝟓𝐤 𝟐 , 𝐮) (5) parameter estimation of nonlinear systems [18], [19]. One of
𝐤𝟒 = 𝑇𝑠 𝐟(𝐱̂ + 𝐤 𝟑 , 𝐮) the states is measured and the estimated states are corrected
using measurement error in a switching function and
multiplying by a feedback constant. For the state estimation of
____________________________________________________________________________________________________________
565
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
566
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V(mV)
𝛽ℎ (𝑉) = 1⁄(1 + 𝑒𝑥𝑝(− (𝑉 + 35)⁄10)) , (20) -20
and 1 means that the gate is open. The parameters of the single a) Membrane potential (𝑉) estimate (mV)
neuron based on the HH model are given in TABLE I.
TABLE I: Model variables and parameters 1.2 XReal
XDBGO
Symbol Units Description 1
𝐶 1𝜇𝐹/𝑐𝑚2 Membrane capacitance
𝑔𝑁𝑎 120𝑚𝑆/𝑚2 Maximal conductance of sodium current 0.8
m
𝑉𝑁𝑎 50𝑚𝑉 Reversal potential of sodium current 0.4
𝑉𝐾 −77𝑚𝑉 Reversal potential of potassium current
𝑉𝐿 −54𝑚𝑉 Reversal potential of leak current 0.2
0
An example of chaotic dynamics of the HH neuronal model 0 50 100 150
time(ms)
is shown in Fig. 1 for initial conditions 𝑉 = 0𝑚𝑉, 𝑚 = 0.0529,
ℎ = 0.5961 and 𝑛 = 0.3177 with sampling period 𝑇𝑠 = 0.01𝑠. b) Activation of the sodium flow current (𝑚) estimate
1
XReal
0.9 0.6
XDBGO
0.8
0.5
0.7
0.6 0.4
m state
0.5
h
0.3
0.4
0.3 0.2
0.2
0.1
0.1
0 0
-80 -60 -40 -20 0 20 40 60 0 50 100 150
V state time(ms)
Fig. 1 Membrane potential 𝑉 vs sodium flow current 𝑚. c) Inactivation of the sodium flow current (ℎ) estimate
In this section, the state estimation results are plotted 0.8 XDBGO
0.5
provide accurate state estimation results for HH neuronal model
using different nonlinear observers. At the same time, we have 0.4
tuned to get best estimation results for each designed observer. time(ms)
____________________________________________________________________________________________________________
567
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
XReal
-20 40
XEKF
-40 20
-60 0
V(mV)
-80 -20
0 50 100 150
time(ms)
-40
0.6 1.2
m
XReal
XEKF
0.4 1
0.2 0.8
0 0.6
m
0 50 100 150
time(ms)
0.4
0
XReal 0 50 100 150
0.6
XSMO time(ms)
0.5
b) Activation of the sodium flow current (𝑚) estimate
0.4
h
XReal
0.3 0.6
XEKF
0.2 0.5
0.1 0.4
h
0 0.3
0 50 100 150
time(ms)
0.2
XReal 0
0 50 100 150
0.8 XSMO time(ms)
XReal
0.5
0.8 XEKF
0.4
0.7
0.3
0.6
0.2
n
0.5
0 50 100 150
time(ms)
0.4
0 50 100 150
SMO based state estimation results are shown in Fig. 3, time(ms)
respectively. The estimated states are following to the real d) Activation of the potassium flow current (𝑛) estimate
states slightly late compared to the DBGO estimation results.
However, after a few milliseconds, the real states are estimated Fig. 4 EKF estimation results for Hodgkin-Huxley neuronal model.
____________________________________________________________________________________________________________
568
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
569
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Trabzon, TURKEY
kadir@ktu.edu.tr
Abstract— Polar code is one of the major breakthroughs in methods, WIB method uses only a small cluster of infor-
information theory field by its theoretically proven capacity mation bits to detect successful decoding. In this paper we
achieving error correction property and low encoding, decod- propose simplified minLLR method using this idea of WIB
ing complexities. Since Arikan submitted his original paper,
researchers have made many improvements on both decoding method.
and encoding sections. Successive cancellation (SC) and belief
propagation (BP) is widely used decoding algorithms for polar II. SMS BP AND E ARLY S TOPPING C RITERIA
codes. To reduce the complexity of BP decoder, early stopping
methods are studied in literature. In this paper we simplify early A. SMS BP Polar Code Decoder
stopping method for BP decoder by using channel polarization
phenomena and we reduce the complexity of early stopping BP is a commonly used algorithm for decoding of block
section by observing only a small cluster of information bits codes [5]. SMS BP is basically the same algorithm with
which are polarized to the highest error probabilities. Sim- min-sum approach and a scaling factor. SMS BP has a
ulation results shows that early stopping detection algorithm similar performance with BP while having a significantly
needs to observe only n/8 bits instead of n for considered code reduced complexity and a hardware friendly structure. For a
length and rates without any performance loss and faulty early
stopping detection.
(n, k) polar code there is (n = 2m ) m-stage factor graph
Keywords— Polar code, belief propagation decoder, early representation. Factor graph representation of (8, 4) polar
stopping detection code and single processing element (PE) showed in Fig. 1
and Fig. 2, respectively. Here, n is the code length and k
I. I NTRODUCTION is the amount of information bits and m is the number of
As the first theoretically proven capacity-achieving error factor graph layer and each layer of factor graph includes
correction code, polar code [1], draw serious attention by its n/2 PE. Update equations for decoder are given in Eqn. (1-
low encoding and decoding complexities. After polar code 4). In Eqn. (1-4) log-likelihood ratio(LLR) values for right
is introduced researches are focused on developing more and left direction referred as R and L LLRs, i is bit index,
efficient encoding and decoding algorithms. j is layer index and t is iteration number.
There are two main decoding algorithms for polar code,
belief propagation (BP) and successive cancellation (SC) as Layer 1 Layer 2 Layer 3
a special form of BP [1], [2]. Both decoders have their (1,1) (1,2) (1,3) (1,4)
own advantages and disadvantages. SC decoder has a serial
+ + +
(2,1) (2,2) (2,3) (2,4)
structure which increases the latency with low decoding com- + +
plexity [2], [3]. On the other hand BP has parallel structure (3,1) (3,2) (3,3) (3,4)
+ +
which is suitable for parallel hardware implementation with (4,1) (4,2) (4,3) (4,4)
high complexity [2]. Scaled min-sum (SMS) BP decoder +
proposed in [4] offers a great deal of complexity reduction
(5,1) (5,2) (5,3) (5,4)
over original BP decoder. + +
As all the other iterative decoding algorithms without early (6,1) (6,2) (6,3) (6,4)
stopping detection SMS BP decoder needs fixed number of +
(7,1) (7,2) (7,3) (7,4)
iterations for decoding [5]. With the help of early stopping +
methods some of the redundant iterations might be avoided. (8,1) (8,2) (8,3) (8,4)
In [3], two early stopping methods, G-matrix and minLLR,
for BP based polar code decoder are proposed. Another
simplified early stopping method called worst of information Fig. 1. Factor Graph Representation for an (8, 4) Polar Code Decoder
bits (WIB) is proposed in [6]. Unlike G-matrix and minLLR
____________________________________________________________________________________________________________
570
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
( ) ( ) ( )
Lti,j = s ∗ sign Lt−1 ∗ t−1
∗ Lt−1 , Lt−1 j
i+n/2 ,j
t t
i,j+1 sign L j
i+n/2 ,j+1 + R j
i+n/2 ,j min i,j+1 i+n/2 ,j+1 + R j (1)
( t−1 ) ( t ) ( t−1 t )
i+n/2j ,j+1 + s ∗ sign Li,j+1 ∗ sign Ri,j ∗ min Li,j+1 , Ri,j
Lti+n/2j ,j = Lt−1 (2)
( t ) ( ) ( )
t t−1
t
Ri,j+1 = s ∗ sign Ri,j ∗ sign Lt−1
i+n/2j ,j+1 + R t
i+n/2j ,j ∗ min R i,j , L i+n/2j ,j+1 + R i+n/2j ,j
t
(3)
( t−1 ) ( t ) ( t )
t
Ri+n/2 j ,j+1 = Ri+n/2j ,j + s ∗ sign Li,j+1 ∗ sign Ri,j
t
∗ min Lt−1
i,j+1 , Ri,j
(4)
࢚
ࡾ࢚ǡ ࡾǡ called WIB early stopping criterion proposed in [6] and the
ሺǡ ሻ + ሺǡ ሻ
method uses only small cluster of information bits which are
ࡸ࢚ǡ ࢚െ
ࡸǡ
polarized to the highest error probabilities called WIB. As
suggested in [6] for n up to 2048 it is enough to use n/8
࢚
ࡾȀ ǡ
ࡾ࢚Ȁǡ bits to detect successful decoding. Method checks for LLR
ሺ Ȁ ǡ ሻ ሺ Ȁ ǡ ሻ values sign alterations of WIB. If there is no sign change
࢚
ࡸȀ ǡ
ࡸ࢚െ
Ȁ ǡ
of WIB LLR values during last M iterations, decoding is
assumed successful.
Fig. 2. Processing Element In this study we combine the idea of WIB method with
the structure of minLLR method.
____________________________________________________________________________________________________________
571
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
!"
! "
! "
=>)
)*"+&,-./012' 3
! " )*"+%,-./012' 3
)*%+&,-./012' 3
)*"+&24.566)27.8925+:
)*"+%24.566)27.8925+:
! " )*%+&24.566)27.8925+:
)*"+&24.566)27.8925+"(
)*"+%24.566)27.8925+"(
)*%+&24.566)27.8925+"(
! "
# !$ !$ "!$ %!$ &!$ '!$ $!$ (!$
;<),1=3
Fig. 4. BER-SNR Results n = 1024 and R = (1/3, 1/2, 2/3) for 40 fixed iterations, minLLR (nW IB = n/8, n/16)
TABLE I
P ROBABILITIES OF M INIMUM LLR B EING O UTSIDE OF WIB AND AVERAGE LLR VALUE D IFFERENCE WHEN M INIMUM LLR IS O UTSIDE WIB
Probability of minimum LLR
Average LLR value difference
being outside of WIB
WIB n/16 n/8 n/16 n/8
RATE
1/3 1/2 2/3 1/3 1/2 2/3 1/3 1/2 2/3 1/3 1/2 2/3
SNR(dB)
−0.5 0.20 - - 0 - - 1.88 - - 0 - -
0.0 0.66 - - 0 - - 2.00 - - 0 - -
0.5 0.86 - - 0 - - 2.11 - - 0 - -
1.0 0.87 - - 0 - - 4.59 - - 0 - -
1.5 0.97 0.14 - 0 0 - 4.71 1.60 - 0 0 -
2.0 - 0.62 - - 0 - - 1.74 - - 0 -
2.5 - 0.89 - - 0 - - 3.33 - - 0 -
3.0 - 0.93 0.09 - 0 0.02 - 3.74 1.35 - 0 0.43
3.5 - 0.96 0.19 - 0 0.06 - 4.19 1.39 - 0 0.52
4.0 - 0.98 0.62 - 0 0.26 - 4.53 1.52 - 0 0.59
4.5 - - 0.86 - - 0.39 - - 3.17 - - 0.65
5.0 - - 0.49 - - 0.14 - - 3.43 - - 1.14
5.5 - - 0.80 - - 0.39 - - 3.73 - - 1.20
6.0 - - 0.94 - - 0.53 - - 3.97 - - 1.24
Table I along with Fig. 4 tell us if nW IB = n/8 there signal to noise ratio (SNR) over 10000 trials.
is no need to look for minLLR value outside WIB cluster.
This allows us to simplify minLLR early stopping criterion As illustrated in Fig. 4 there is no performance loss with
by reducing computational need. proposed method along with complexity reduction for early
stopping section which is also compared with Table II. Fig.
III. S IMULATION R ESULTS 4 also tells us that nW IB = n/16 is not enough to cover the
semi-polarized bits.
For simulation study we use the same methodology as in
[3], [6]. During simulations we consider binary-input additive The most important contribution of an early stopping
white Gaussian noise (BI-AWGN) channel. We choose code criterion to an iterative decoder is iteration reduction perfor-
length as 1024 and we provide simulation results for various mance. Since proposed simplified minLLR method has the
code rates (1/3, 1/2, 2/3). Scaling parameter for SMS BP same iteration reduction performance with original minLLR
algorithm is s = 0.9375. In Fig. 4, we provide bit error rate method we have not provided these results casue these results
(BER) curves for various code rate and WIB number versus clearly stated in [3], [6].
____________________________________________________________________________________________________________
572
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE II
C OMPUTATIONAL C OMPLEXITIES OF E ARLY S TOPPING M ETHODS FOR
S INGLE I TERATION
IV. C ONCLUSIONS
In this study we proposed a simplification to an existing
early stopping criterion (minLLR) for BP based polar code
decoders by using a theory from another early stopping
method (WIB). Simulation results show that there is no
performance loss with simplified minLLR method compared
to original minLLR or fixed iteration amount. Although this
method has the lowest computational complexity for early
stopping section, overall performance would be better if it
is used with G-Matrix method because of average iteration
numbers.
R EFERENCES
[1] E. Arikan, “Channel polarization: A method for constructing capacity
achieving codes for symmetric binary-input memoryless channels,
IEEE Trans. Inf. Theory, vol. 55, no. 7, pp. 3051-3071, 2009.
[2] K. Nai et al. “Polar Codes: Primary Concepts and Practical Decoding
Algorithms,” IEEE Communications Magazine, vol. 52, no. 7, pp. 192-
203, Jul. 2014.
[3] B. Yuan and K. K. Parhi, “Early Stopping Criteria for Energy-Efficient
Low-Latency Belief-Propagation Polar Code Decoders,” IEEE Trans.
Signal Process., vol. 62, no. 24, pp. 6496-6506, Dec. 2014.
[4] B. Yuan and K. K. Parhi, “Architecture optimizations for BP polar
decoders,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Proces.
(ICASSP), May 2013.
[5] T. Richardson and R. Urbanke, Modern Coding Theory, UK: Cam-
bridge University Press, 2008.
[6] C. Simsek and K. Turk, “Simplified Early Stopping Criterion for
Belief-Propagation Polar Code Decoders” in IEEE Communications
Letters, Early Access Articles, DOI: 10.1109/LCOMM.2016.2580514
____________________________________________________________________________________________________________
573
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
I. INTRODUCTION In the same vain, the middle feeder carries the loads on the
Distribution networks (DNs) carry electrical power right rest of the feeders if S1 and S2 are open while S7 and S8 are
from transmission system to the consumption sites. Being the closed. In either case, the radial configuration of network is
last stage of the of power delivery, DN contributes at least 40% always maintained.
of the total power loss occurring in the entire system [1]. Most However, this network modification is a non-linear, multi-
DNs are built as weakly meshed but operated radially, for objective, highly constrained and combinatorial optimization
effective coordination of their protection systems [2, 3, 4]. This problem [8]. The complexity of the problem arises from the fact
is done with aid of switches within a feeder or across different that distribution network topology has to be radial and power
feeders as shown in Fig. 1. They are either normally closed flow constraints are nonlinear in nature.
(NC) sectionalizing switches; S1-S6 existing within each feeder The early studies on the network reconfiguration were
and normally open (NO) tie switches; S7 and S8 connecting directed to the planning stage [9, 10]. The sole objective these
sections of different feeders [5]. DN is vulnerable to number of researches is to minimize the cost of construction, giving no
outages due its radial nature, as a fault on a single line can result concern on the quality of the supply. The first recognized work
in blackout to many customers [6]. As a result, many efforts which attempts to solve the problem of distribution system
have been made to suppress these complications and improve reconfiguration for loss reduction was presented by Merlin and
the reliability and quality of the supply. Among the techniques Back [11]. Henceforth, many algorithms have been developed
proposed is distribution network reconfiguration (DNR). to solve this problem. In the proposed method, all switches in
____________________________________________________________________________________________________________
574
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the DN were closed and then the switches are opened iteration. and are positive real numbers, called learning
successively to eliminate the loops. factors or acceleration constants, which are used to weight the
Recently, some researches have integrated both the DNR particle individual knowledge and the swarm social knowledge,
and DG placement problems to improve the effectiveness of the respectively. rand1,i and rand2,i are real random numbers
DN. Thuan Thanh Nguyen et al. applied adaptive cuckoo uniformly distributed between 0 and 1 that make stochastic
search for DNR for DG allocation [1]. The objective of this changes in the ith particle trajectory. Finally, is the inertia
technique is minimizing power loss and improve voltage weight factor, which represents the weighting of a particle’s
stability. Similarly, in [12], DNR is applied to distribution previous velocity [17].
feeder to allow increasing the penetration of plug-in electric
vehicles and minimizing network cost. Power loss and voltage II. PROBLEM FORMULATION
deviation minimization using DNR is presented in [13] The problem to be solved involves determining the optimal
In this paper, binary particle swarm optimization (B-PSO) DNR, taking into account different technical constraints. The
is applied for optimal DNR to minimize the power loss and objective of the proposed technique is to minimize power loss
voltage deviation in primary DN. The outstanding performance and voltage deviation as formulated in objective function (OF)
of the proposed technique is its ability to solve DNR problem below.
on single feeder or multi-feeder DNs.
Where
B. Binary Particle Swarm
= + (3)
Due the non-linearity and combinatorial nature of the DNR
optimization problem, various techniques were applied for its The first part of the OF is the index of power loss obtained by
solution. It can either be a single or multiple objective problem normalizing the power loss with its initial value as shown in (4)
[6]. Both classical and heuristic optimization methods were
.
applied for DNR problems. Mixed-integer conic programming
is formulated in [5] to minimize power loss. In a similar work = (4)
presented in [14], selective PSO is employed. However, in
these studies less concern was giving to voltage deviation. .
Where is the power loss after reconfiguration and
The particle swarm optimization (PSO) was first introduced
by Kennedy and Eberhart in 1995 [15]. It was developed is the power loss before reconfiguration calculated as;
through simulation of many simplified work, it has been found
to be robust in solving continuous nonlinear optimization +
= = 1,2, … , (5)
problems, PSO is attractive because very few parameters are
require for its applicability [16].
The algorithm used in this paper; B-PSO, has an advantage The second segment of the OF, f2 is the index of the total
of handling both discrete and continuous parameters. voltage deviation. It is similarly obtained by normalizing its
.
It involves initializing the search space with swarm particles, xi reconfigured value, . with its initial value, . .
randomly. The search space in this problem is the set of all the
sectionalizing and tie switches. At each iteration, the position .
∆
of ith particle is updated by its previous position in the velocity = (6)
∆
vector, vi according (1)
The total inherent voltage deviation is computed as;
= + ; = 1,2, … , (1)
=[ ≤ , = 1,2, … , (9)
Where , , , ,…, , ] is the
best solution achieved for the ith particle at the (t−1)th iteration, Where
and =[ , ,…, ] is the best = + −2 cos − sin (10)
position found for all particles in the swarm at the (t−1)th
____________________________________________________________________________________________________________
575
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
= + + (13)
2
= (14)
2
= (15)
Fig. 3 IEEE 33 bus test system
____________________________________________________________________________________________________________
576
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III
33 BUS TEST SYSTEM RESULTS
TABLE I
16 BUS TEST SYSTEM RESULTS Fig. 7 Optimized Configuration of 33 bus test system
respectively.
Voltage (p.u)
0.97
0.96
1 0.95
Before Reconfiguration
After Reconfiguration
0.94
0.995
0.93
0.99
0.92
Voltage(pu)
0.985 0.91
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
Bus Number
0.98
____________________________________________________________________________________________________________
577
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
578
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[8] H. Li, W. Mao, A. Zhang, C. Li, "An improved distribution network Optimization," International Journal Of Multidisciplinary Sciences And
reconfiguration method based on minimum spanning tree algorithm and Engineering, vol. 3, no. 6, pp. 16-21, June 2012.
heuristic rules," Electrical Power and Energy Systems, vol. 82, pp. 466- [15] R. Eberhart, "Particle swarm optimization," Proc. IEEE Int. Conf. on
473, 2016. Neural Networks, November 1995.
[9] RN. Allan, R Billinton, A. M. Breipohl and C.H. Grigg, "Bibliography [16] R. Eberhart, "Empirical study of particle swarm optimization," in Proc.
On The Application Of Probability Methods In Power System Reliability IEEE Int. Conf. Evol. Comput., Washington, DC, July 1999.
Evaluation" IEEE Transactions on Power Systems, vol. 9, no. 1, pp. 41-
49, February 1994. [17] F. J. and M. Gomez-Gonzaleza, "Optimization of distributed generation
systems using a new discrete PSO and OPF," Electric Power Systems
[10] G. T. Gonen, A. A. Mahmoud and H. W. Golbum, "Bibliography of Research, vol. 84, no. 1, p. 174–180, March 2012.
Power Distribution System Planning," IEEE Transactions on Power
Apparatus and Systems, vol. vol. PAS 102, pp. 1178-1787, June 1983. [18] M. A. Kashem, M. Negnevitsky and G. Ledwich, "Maximising Voltage
Support in Distribution Systems by Distributed Generation," in
[11] A. Merlin and H. Back, "Search for a minimal loss operating spanning TENCON 2005 - 2005 IEEE Region 10 Conference, Melbourne, Qld.,
tree configuration in an urban power distribution system," in Proceeding Nov. 2005.
5th power syst comput conf (PSCC), Cambridge, UK, vol. 1–18, 1975.
[19] N. R. Srinivasa Rao, K. Ravindra, K. Satish, and S. V. L. Narasimham,
[12] A. Kavousi-Fard, A. Abbasi, M.-A. Rostami and Abbas Khosravi, "Power Loss Minimization in Distribution System Using Network
"Optimal distribution feeder reconfiguration for increasing the Reconfiguration in the Presence of Distributed Generation," IEEE
penetration of plug-in electric vehicles and minimizing network costs," Transactions on Power Systems, vol. 28, no. 1, pp. 317-325, Feb. 2013.
Energy, vol. 93, pp. 1693-1703, 2015.
[20] L. Hou, J. Yin, Y. Yu and L. Zhang, "Distribution Network
[13] B. C. Lee, S. Mehrotra, and Z. Bie, "Robust Distribution Network Reconfiguration Based on Load Forecasting," Intelligent Computation
Reconfiguration," IEEE Transactions on Smart Grid, , vol. 6, no. 2, pp. Technology and Automation (ICICTA), 2008 International Conference
836-842, March 2015. on, Hunan, Oct. 2008.
[14] Tamer M. Khalil and A. V. Gorpinich, "Reconfiguration for Loss
Reduction of Distribution Systems Using Selective Particle Swarm
____________________________________________________________________________________________________________
579
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Today, the using of permanent magnet synchronous and the differential evolution algorithm were run to provide
motors has increased in industrial fields and these motors have the optimal efficiency values. As input parameters the
many different structures. The outer rotor structure is structural parameters of the motor were taken. The
remarkable in terms of ease of design and torque density. This optimization and the analytical results were compared and
study proposes to design optimization of the outer rotor with
eventually the results are sufficiently.
genetic algorithm and differential evolution algorithm. The
geometric sizes were selected as design parameters and the motor
II. DESIGN OPTIMIZATION STAGES OF THE PMSM
efficiency is used as objective function. The obtained results by
using of each algorithm are compared and the industrial
conclusions were presented. As a result, the results are
A. The Winding Layout
satisfactory and the study is useful for academic and industrial In this study, the investigated PMSM are the surface
technical staff. mounted with outer rotor and the number of stator slots and
pole numbers are 12/10 respectively. The concentrated/double
Keywords— design optimization, differential evolution algorithm, layer winding layout is shown in Figure 1. The winding factor
genetic algorithm, outer rotor, permanent magnet synchronous is a good value for performance and it is 0.933 [5], [6].
motor
I. INTRODUCTION
Electric motors consume a large portion of the electrical
Fig. 1 The winding layout
energy in the world. Therefore, academic and industrial works
related to energy consumption and high efficiency motors are B. The Design Parameters and the Objective Function
more important.
The design of the permanent magnet synchronous motors is
Nowadays permanent magnet synchronous motors
nonlinear study and it has different design parameters. Some
(PMSMs) have been preferred because of their high efficiency
of them are variables, invariables and constants. Therefore,
and torque density. Motor structures affect these merits. Here
the effects of these parameters on the objective are different in
the outer rotor permanent magnet synchronous motors are
terms of ease and accuracy of the design. The design
remarkable. Generally these motors have common advantages
optimization parameters are the rotor diameter for the inner
as other permanent magnet synchronous motors for example
rotor or stator diameter for outer rotor Drc, magnet thickness lm,
gearless [1]. In low-speed applications, the surface mounted
air gap length δ, slot wedge height hsw, stator tooth width bts,
PMSM with inner rotor is most popular but in high-speeds the
stator slot height hss, ratio of the slot opening over the slot
outer rotor is more appropriate for the permanent magnets can
width kopen. The two dimensional geometry of the motor is
be protected from centrifugal forces in outer rotors. Moreover
shown in Figure 2 [7].
their torque density is greater due to the wide diameter of the
air gap and the production costs of the outer rotor are suitable.
The stochastic methods have been used for electrical
machine design optimization. These methods can give
effective results to the very complex problems like motor
designs [2], [3]. Some of the most famous of these methods
are the genetic algorithm and the differential evolution
algorithm. These methods do not require the initial solution or
population. The objectives of the design optimization of the
PMSMs based on stochastic methods are reduce the torque
harmonics or to increase the efficiency of the motor [4].
This study was conducted to optimize the design of the
outer rotor PMSM for high efficiency. The genetic algorithm Fig. 2 Two dimensional geometries of the outer rotor PMSM
____________________________________________________________________________________________________________
580
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
problem [8].
Genetic algorithms do not guarantee to investigate the 96
X: 3
optimal result. The GAs may converge to a local solution but Y: 96.06
optimal efficiency values (%)
also they do not require the solution of the initial solution. 95.5
96.15
similar to genetic algorithm. Furthermore the main feature of 96.1
the DEA is the use of differential operator. Differential action 96.05
shall be taken to evaluate fitness values on the parents’ genes 96
by the differential operator, thus the quality of the population
95.95
is tried to be increased. DEA shows superior performance
95.9
because of very small number of parameters to be set and the
95.85
understandable actual code sequence [8].
0 5 10 15 20 25 30 35 40 45 50
iteration number
IV. THE DESIGN APPLICATION AND THE EVALUATION OF THE
RESULTS Fig. 3 The convergence graphics of the algorithms
In this study some assumptions were made about the motor
design such as neglect of the mechanical strength, thermal
The analytical, the GA, and the DEA efficiency results are
loadability. Obviously, industrial manufacturability of the
95.09%, 96.06% and 96.22% respectively. The ohmic losses
PMSM was neglected in the design optimization results. The
have more decreased than the iron losses. The genetic
saturation on the stator teeth and stator and rotor yokes were
____________________________________________________________________________________________________________
581
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
algorithm may be fallen in a local solution it is absolutely of the Nordic Workshop on Power and Industrial Electronics (NORpie),
2004.
weakness. The graphics of the GA and the DEA were shown
[2] B. N. Cassimere, S. D. Sudhoff, “Population-based design of surface-
in Figure 4. The DEA is more sensitive and strong. mounted permanent-magnet synchronous machines,” IEEE
Transactions on Energy Conversion, vol. 24, no. 2, June 2009.
V. CONCLUSIONS [3] M. Mutluer, O. Bilgin, “Design optimization of PMSM by particle
swarm optimization and genetic algorithm,” in International
The design optimization of the outer rotor PMSM was Symposium on Innovations in Intelligent Systems and Applications,
investigated by the genetic algorithm and the differential 2012, Pages 1-4.
evolution algorithm. The obtained results show that the [4] S. D. Sudhoff, J. Cale, B. Cassimere, M. Swinney, “Genetic algorithm
differential evolution algorithm has better results than the based design of a permanent magnet synchronous machine,” IEEE
International Conference on Electric Machines and Drives, 2005, Pages
genetic algorithm. The ohmic losses are more important in 1011-1019.
low frequency application. Eventually, the design [5] J. A. Güemes, A. M. Iraolagoitia, M. P. Donsión, J. I. Del Hoyo,
optimization study of the outer rotor permanent magnet “Analysis of torque in permanent magnet synchronous motors with
synchronous motor and the results are acceptable. fractional slot windings cogging torque,” in Proceedings of The
International Conference on Electrical Machines, 2008, Pages 1-4.
[6] F. Libert, J. Soulard, “Investigation on Pole-Slot Combinations for
ACKNOWLEDGMENT Permanent-Magnet Machines with Concentrated Windings,” in
This study is supported by The Coordinator of Scientific Proceedings of the International Conference on Electrical Machines
Research Projects of Necmettin Erbakan University. (ICEM), Cracow, Poland, 2004, Pages 530-535.
[7] Flux2D Version 10.3, Tutorial: Brushless Embedded Magnet Motor,
2010.
REFERENCES [8] S. S. Rao, “Engineering optimization theory and practice,” Wiley, 2009.
[1] F. Libert, J. Soulard, “Design study of different direct-driven
permanent-magnet motors for a low speed application,” in Proceedings
____________________________________________________________________________________________________________
582
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In order to cover wide spectrum with high antenna this situation, a sensing antenna should be inserted into the
gain, the reconfiguration process may be proposed as an system so as to scan and figure out the suitable frequency
alternative solution instead of using several antennas which is not bands [6]. All these process address the frequency
efficient considering the size restriction situation. In this study, a reconfiguration. On the other hand, it should be noted that
microprocessor based antenna reconfiguration controller circuit
pattern or polarization reconfiguration may be needed for the
is designed and fabricated. Designed circuit can be used to
reconfigure the antenna frequency, pattern or polarization for future communication standards.
the possible 5G communication requirements. Proposed antenna The paper is organized as follows. In Section 2, the
reconfiguration controller has the ability of switching selected fabricated Antenna Reconfiguration Controller (ARC) is given.
PIN diodes in order to cover the desired frequency bands. Section 3 gives and discusses the results. Section 4 concludes
the paper.
Keywords— Antenna reconfiguration, 5G communication
II. ANTENNA RECONFIGURATION CONTROLLER (ARC)
I. INTRODUCTION The fabricated microprocessor based ARC is given in Fig.1.
The necessity of high speed communication drives Parallel cables are connected to the one port of the 8 bit
researches to figure out new designs and algorithms. These microprocessor. LEDs are inserted for testing the signalization
developments may be in hardware or software side. In order to correctly. In order to test the ARC, a monopole antenna is
meet the high speed communication requirements, these designed with one high speed PIN diode as an electronic
developments should move on together. For example, switch. The aim of this testbed is to use a monopole antenna
considering a mobile phone, suitable software is loaded into for two different frequencies. The monopole size can be
different type microprocessors which manage several selected to obtain the desired frequency band. However the
functions in the phone. On the other hand, electromagnetic bandwidth of the antenna does not change is this method.
properties of the communication device are also important.
One of them is the antenna which converts electrical signals
into electromagnetic waves and vice versa. The antenna in a
mobile phone or base station should send and receive
electromagnetic waves in different bands of the
electromagnetic spectrum. In order to meet the
communication standards, different types of antennas have
been designed so far. Some electronics devices have more
than one antenna to cover different bands. Another option is to
use reconfigurable antennas of which the resonant bands can
be rearranged by the operator. Moreover, beside frequency
reconfiguration, different properties of the antennas can be
reconfigured such as antenna pattern or polarization [1,2].
Several reconfiguration methods are used in literature [3-5].
However, electronic and optical reconfigurations are very
topical these days. In this study, a microprocessor based
reconfiguration tool is designed fabricated and tested. The
designed antenna reconfiguration controller (ARC) can be
applied for controlling 64 electronic or optical switches on the
antenna surface. For the sake of simplicity, a monopole
antenna is used in the testbed. However, considering a
cognitive radio process which may be a possible network in
5G, the instant channel state information (CSI) will be Fig.1. Fabricated Antenna Reconfiguration Controller
necessary in order to focus on the desired frequency band. In
____________________________________________________________________________________________________________
583
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
584
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSION
An antenna reconfiguration controller (ARC) is designed,
fabricated and tested. A monopole antenna which is controlled
by the ARC has been fabricated and operated for two different
frequencies using a PIN diode. This antenna reconfiguration
controller is inevitable part of the cognitive radio which is
suggested as possible 5G networks in literature. Using this
fabricated ARC, reconfiguration process can be applied up to
64 swithes. This number can be increased but considering the
complexity it is not suggested.
REFERENCES
[1] Aboufoul T., Alomainy A., Parini C., Polarization Reconfigurable
Ultrawideband Antenna for Cognitive Radio Applications,
Microwave and Optical Technology Letters, 55,3 (2013) 501-506.
[2] Aboufoul T., Shoaib I., Alomainy A., Xiaodong C. Pattern
Reconfigurable Planar UWB Antenna Array for Future Cognitive
Radio Portable Devices, Antennas and Propagation Conference
(LAPC), Loughborough, 2012.
[3] Tawk Y., Albrecht A. R., Hemmady S., Balakrishnan G., Christos
G., Christodoulou C. G., Optically Pumped Frequency
Reconfigurable Antenna Design, IEEE Antennas and Wireless
Propagation Letters, 9 (2010) 280-283.
[4] Cai Y., Guo Y. J. , Bird T. S., A Frequency Reconfigurable Printed
Yagi-Uda Dipole Antenna for Cognitive Radio Applications, IEEE
Transactions on Antennas and Propagation, 60,6 (2012) 2905-
2912.
[5] Yazgan A., Kaya H., Çavdar İ.H., "Optically Reconfigurable
Sierpinski Fractal Antennas for RoF Based Communication
Systems", Telecommunication Systems, vol.59, no.4, pp.453-461,
2015
[6] Zhang R., Lim T. L., Liang Y. C., Zeng Y., Multi-Antenna Based
Spectrum Sensing for Cognitive Radios a Glirt Approach, IEEE
Transactions on Communication, 58,1 pp. 84-88, 2015.
[7] Yazgan A., Çavdar İ.H., Kaya H., "The Effect of PIN Diode
Characteristics on the Antenna Reconfiguration Part of Cognitive
Radio", 38th International Conference on Telecommunications and
Signal Processing (TSP), Prag, CEK CUM., 9-11, pp.191-194,
2015.
____________________________________________________________________________________________________________
585
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract – Brain is composed by neurons using electricity to Brainwaves change regard to what we’re doing and feeling. In
communicate to each other. There is a great amount electrical the case of that slower brainwaves are dominant; we can feel
activity in the brain due to collection of numerous neurons tired, slow, sluggish, or dreamy. On the contrary in the case of
sending signal. Neurons send brainwaves detected by sensitive that higher frequencies are dominant, we feel hyper-alert.
equipment like electroencephalogram or electroencephalograph Brainwaves are classified into groups with respect to
(EEG). Beta waves emanates from brains responsible for frequencies in Hertz, like gamma, beta, alpha, theta and delta.
problem solving or decision making. Some equipment like
neurosky biosensor intercepts these brain waves. In this study,
Frequency of delta waves changes between 0.5 and 3 Hz.
neurosky biosensor is used in measuring the meditation level of
students in the physics course. Developed program process the Delta waves slowest whereas loudest brainwaves. Delta waves
data of neurosky biosensor sent and make decision about generated in deepest meditation and dreamless sleep.
meditation level of student in the course. Program says that
“your meditation level is low to pass another subject or solve the Frequency of theta waves changes between 3 and 8 Hz. Theta
exam”. waves generated in deep meditation and sleep.
Key words – Brainwaves, neurosky biosensor, EEG, education, Frequency of alpha waves changes between 8 and 12 Hz.
neurolearning. Alpha waves responsible for now and no quite meditation.
Alpha waves generated in calmness and learning.
I. INTRODUCTION
Our brain is made up of billions of neurons using electricity to Frequency of beta waves changes between 12 and 38 Hz.
communicate to each other. There is a great amount of These waves are dominant in waking state of consciousness
electrical activity in the brain thank to sending signals. This and engaged with problem solving and decision making.
electricity can be detected by equipment such as
electroencephalograph (EEG). Electrical activity in the brain Frequency of gamma waves, fastest changes between 38 and
composes brainwaves. Brainwaves are produced by 42 Hz [1, 2].
synchronised electrical pulses from masses of neurons
communicating with each other. Brainwaves are divided into Neurosky EEG biosensor, which is a non-invasive, dry and
bandwidths to describe their functions as indicated in Fig. 1, low cost, can intercept the brainwaves, measures the attention
however are best thought of as a continuous spectrum of and meditation level is illustrated in Fig. 2. [3].
consciousness.
____________________________________________________________________________________________________________
Fig. 1 Brainwaves
586
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
different room colour. By changing the room colour these Thirdly, lesson names, required average attention and
meditation and attention rates can be increased [5]. In order to meditation rate for subjects are determined as illustrated in Fig.
observe how the rates changes, a lot of game is developed [6]. 7.
Neurosky EEG biosensor can be used for identifying the
emotion while playing game [7]. In this study, neurosky EEG
biosensors are used in measuring the meditation level of
students in the course. Developed program process the data of
neurosky EEG biosensors sent and make decision about
meditation level of student in the exam. Program says that
“your meditation level is low to solve the exam or pass to
another subject”.
Fig. 7 Lessons of subjects
II. EDUCATION WITH NEUROSKY BIOSENSOR Fourthly, course materials are uploaded into developed
We have developed a program by means of using neurosky program as demonstrated in Fig. 8 a-b for text and video
biosensor. Firstly, instructors and students have to be sign up upload respectively.
this program as demonstrated in Fig.3.
Fig. 7 (a)
Fig. 3 Sign up
Instructors sign in the program and load the course materials (a)
in the form of text, power point, image and video. Main menu
of the developed program is illustrated in Fig.4.
Finally, after course materials are loaded, student can enter the
system and study the subjects. During studying period,
attention and meditation rates are recorded. When the students
Fig. 5 Offering the courses want to pass another lesson in the case of lower meditation
and attention rate, the system gives the warning like “your
Secondly, subjects of courses are entered as depicted in Fig. 6. attention and meditation level is lower than predetermined
value for that reason you cannot pass another lesson”. Also
instructor can see the statistical mean of the meditation and
attention rate as shown in Fig. 9.
____________________________________________________________________________________________________________
587
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
enter the system with their user name and password. Then
study the first subject. If meditation and attention rates are
high students pass to second subject. Otherwise, students have
to study the subject again. Whenever these rates are high,
students pass to second subject.
ACKNOWLEDGMENT
This work is supported partially by Bursa Technical
University Scientific Research Projects Unit under the grant
no. 2016-01-021.
Fig. 9 Statistical mean of the recorded values REFERENCES
III. CONCLUSION
Brain emanates brainwaves in doing or feeling something.
These brainwaves can be intercepted and processed by
equipment like neurosky biosensor. In this study, neurosky
biosensor used for measuring meditation and attention rates of
student. We have developed a computer program. Via the
program, we load the course materials into system. Students
____________________________________________________________________________________________________________
588
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Level control of liquids in coupled tanks is a basic fails; it will stay at the faulty mode during the rest running
requirement in many industrial processes. Liquid levels in tanks process. In other words, the number of actuator failures is
must be controlled accurately regardless of environmental finite. However, in practice, especially for safety-critical
circumstances. Minor faults in sensors, actuators or other system systems, the failed actuator may recover itself by a self-
components that take place in processes where liquid level
repairing control system and the actuator may fail more than
control is required can result in catastrophic consequences. In
once during the operation process, that is, the number of
this case, a fault tolerant control system is needed. The controller
must be either robust (passive) or in reconfigurable (active) typeactuator failures is infinite. The actuator failures are
in order to compensate for the effect of actuator faults and practically stochastic in nature. For instance, a normally open
maintain system reliability and performance. In this study, a electrical contact in electrical switches, relays or breakers is
water tank level control system and possible valve actuator faultsrequired to be open when it is in a de-energized or relaxed
are modelled. By designing different controllers and using state. However, at some moment, the contact may undergo
modelled failures a simulation is constructed. To test the abrupt disturbances, which turns it to be closed, and it may
reconfigurable type controller performance against
recover to be open again, that is, the contact may jump
faults/failures, a model reference adaptive control system is
between closed mode (failure) and open mode (failure-free) in
implemented and compared with PI-controlled system.
a stochastic manner. Several studies design compensators with
Keywords— Actuator faults, fault modelling, fault tolerant
stochastic actuator failure [17-25]. In [17-19], actuator failure
control, model reference adaptive control, liquid tank
model output represented with a failure or faulty matrix. This
diagonal matrix takes ‗1‘ and ‗0‘ values, which correspond to
total failure and normal operation conditions. Also values
I. INTRODUCTION
between ‗1‘ and ‗0‘ are considered which represent loss of
Main industries where liquid level and flow control are effectiveness. The diagonal matrix is stochastic with a known
essential include petro-chemical industries, paper production distribution. Generally Bernoulli distributed sequences and
plants, water treatment industries, food processing industries, Gaussian distribution are used to model failure behaviours. In
pharmaceutical industries, nuclear power generation plants, [20-25], the aforementioned abrupt changes are modeled as a
and automatic water drainage systems. Due to increasing Markovian process. Markovian process is a stochastic model
demands on system performance and product quality as well that can be used to model a random system that changes states
as economic benefits, modern technical processes have according to a transition rule that only depends on the current
become more complicated. These complicated systems are state. [26] shows that for given adequate historical data, the
vulnerable to faults in sensors and actuators. Minor faults in abrupt changes can be modeled as Markovian process. In the
sensors, actuators or other system components that take place studies cited above, the stochastic functions related to
in processes in which fluid level control is a basic requirement Markovian variables are employed to denote the failure
can result in catastrophic consequences. Control system in this scaling factors for each actuator.
case must compensate for the effect of actuator faults and It is needed to design control methods capable of ensuring
must maintain system reliability and performance. nominal performance considering the occurrence of failures.
Failures are difficult to accurately predict in time and can This control is referred to as fault tolerant control (FTC)
be abrupt, incipient and intermittent [1]. They are classified which has become of significant importance in the last
according to their location of occurrence in the system. The decades. There are lots of approaches in FTC area and
impact of a fault can be a small reduction in efficiency but individual research has been carried out extensively. However
could also lead to overall system failure and system instability. systematic concepts, design methods and even terminology
Therefore, in the design phase of compensators a good are still not standardized [2, 3]. In FTC systems, the
actuator failure model that covers all possible cases is crucial achievable system performance depends on the availability of
and in the application phase, more practical. Different types of redundancies in the control system as well as the design
actuator models are available in both deterministic and approaches used in synthesis of fault-tolerant controllers.
stochastic manners for linear or nonlinear systems. Most of Recently, FTC systems are classified into two categories,
related studies consider deterministic actuator fault models [2]. namely, active FTCs, and passive FTCs. These two
They use various approaches in deterministic manner [3-15] approaches use different design methodologies for the same
and generally consider a specific case that once the actuator control objective. However, due to the distinctive design
____________________________________________________________________________________________________________
589
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
approaches used, each method can result in some unique that i-th actuator operates normally ( ua (t ) u (t ) ). The
properties. A general requirement for both methods is the quantities ia can also take values in between 0 and 1, making
existence of system redundancies which can be analytical or
applied on hardware and difference is how the redundancy is it possible to represent partial actuator faults. For different
utilized. types of fault conditions the above model can be specified in
In passive approach, a list of potential malfunctions is detail as:
assumed to be known a priori as design basis faults, and all u i (t ) ai 1; f ai 0 t t0 Fault free
failure modes as well as the normal system operating i i
conditions are considered at the design stage. An active FTC au (t ) 0 ai 1; f ai 0 t tF Partial fault
reacts to system component malfunctions (including actuators, a f ai u i (tF )
i
ai 0 t tF Stuck fault (2)
i ai 0 t tF Hard over fault
fa u
system itself, and sensors) by reconfiguring the controller i
ua (t ) A u (t ) (I A ) f a (t )
diag 1a , 2a , ..., ma
A (1)
____________________________________________________________________________________________________________
590
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
According to Bernoulli equation, the rate of flow of water plant numerator polynomial is monic, plant degree is known,
through the pipe is given by: plant relative degree is known; reference model numerator and
denominator are monic Hurwitz polynomials and relative
qo a 2 gh (4) degree is the same as that of plant. In other words, reference
model must be chosen to be strictly positive definite (SPR) to
where h is the level of the tank, a is the pipe cross-section
obtain a stable controller. Following the certainty equivalence
area and g is the acceleration of gravity. Conservation of mass
principle, the controller given below can be used [27]:
yields the equation:
( s) ( s)
dh u 1*T u 2*T y p 3*T y p c0* r (10)
A qi qo qi a 2 gh (5) ( s ) ( s )
dt
where A is the bottom area of the tank, qi is the inflow rate where ( s ) n 2 ( s) [ s n 2 , s n 3 ,...s,1] for n 2 , ( s ) 0
through the valve. for n=1. 1 , 2 n 1 ; c0 , are constant parameters to
*
3
If we linearize the system around an operating point and be designed and ( s ) is an arbitrary monic Hurwitz
ignore the pressure of the tank outlet, the system can be
polynomial of degree n-1. The controller given in (10) has
assumed to be first order but by taking valve dynamics into
account, system dynamics become second order. The first known parameters ( * ) i.e., the controller is designed for a
order system transfer function between inflow rate and water known plant transfer function therefore it is named as model
level is: reference controller (MRC). However, by using certainty
equivalence, we can define an update law and use the same
H (s) R controller with unknown parameters. We can write the
(6)
Qi ( s) ARs 1 equation in compact form as follows:
____________________________________________________________________________________________________________
591
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
e Gm (s)(s p0 ) p* (u f *T ) (17)
1 1
with u f u , , which implies that the
s p0 s p0
control law in compact form is:
u T T (18)
____________________________________________________________________________________________________________
592
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
593
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Y. Mo and B. Sinopoli, ―Secure control against replay attacks,‖ in
Communication, Control, and Computing, 2009. Allerton 2009. 47th
Annual Allerton Conference on, 2009, pp. 911–918.
[2] W. Wang and C. Wen, ―Adaptive compensation for infinite number of
actuator failures or faults,‖ Automatica, vol. 47, no. 10, pp. 2197–2210,
Oct. 2011.
[3] Y. Zhang, ―Actuator fault-tolerant control for discrete systems with
strong uncertainties,‖ Computers & Chemical Engineering, vol. 33, no.
11, pp. 1870–1878, Nov. 2009.
[4] X.-Z. Jin and G.-H. Yang, ―Robust Adaptive Fault-tolerant
Compensation Control with Actuator Failures and Bounded
Disturbances,‖ Acta Automatica Sinica, vol. 35, no. 3, pp. 305–309,
Mar. 2009.
[5] X.-J. Li and G.-H. Yang, ―Robust adaptive fault-tolerant control for
Fig. 12 Controlled system response in intermittent lock-in-place mode uncertain linear systems with actuator failures,‖ IET Control Theory &
Applications, vol. 6, no. 10, pp. 1544–1551, Jul. 2012.
[6] P. Hajiani and J. Poshtan, ―Abrupt and incipient fault detection and
compensation for a 4-tank system benchmark,‖ Turkish Journal Of
Electrical Engineering & Computer Sciences, vol. 22, pp. 1287–1297,
2014.
[7] Y. Niu, Y. Liu, and T. Jia, ―Reliable control of stochastic systems via
sliding mode technique: Reliable Control of Stochastic Systems via a
____________________________________________________________________________________________________________
594
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Smc Technique,‖ Optimal Control Applications and Methods, vol. 34, failure: Reliable Control For Systems With Actuator Saturation &Amp;
no. 6, pp. 712–727, Nov. 2013. Failure,‖ Optimal Control Applications and Methods, vol. 33, no. 6, pp.
[8] A. Mihankhah, F. R. Salmasi, and K. Salahshoor, ―Partial and total 739–750, Nov. 2012.
actuator faults accommodation for input-affine nonlinear process [18] G. Wei, L. Wang, and F. Han, ―A gain-scheduled approach to fault-
plants,‖ ISA Transactions, vol. 52, no. 3, pp. 351–357, May 2013. tolerant control for discrete-time stochastic delayed systems with
[9] D. Shin, G. Moon, and Y. Kim, ―Design of Reconfigurable Flight randomly occurring actuator faults,‖ Systems Science & Control
Control System Using Adaptive Sliding Mode Control: Actuator Fault,‖ Engineering, vol. 1, no. 1, pp. 82–90, Dec. 2013.
Proceedings of the Institution of Mechanical Engineers, Part G: Journal [19] J. Hu, J. Liang, and D. Chen, ―Reliable guaranteed-cost control for
of Aerospace Engineering, vol. 219, no. 4, pp. 321–328, Jun. 2005. networked systems with randomly occurring actuator failures and
[10] J. Jiang and X. Yu, ―Fault-tolerant control systems: A comparative fading performance output,‖ International Journal of General Systems,
study between active and passive approaches,‖ Annual Reviews in vol. 44, no. 2, pp. 129–141, Feb. 2015.
Control, vol. 36, no. 1, pp. 60–72, Apr. 2012. [20] M. Mahmoud, J. Jiang, and Y. Zhang, ―Optimal control law for fault
[11] Y. Zhang and J. Jiang, ―Bibliographical review on reconfigurable fault- tolerant control systems,‖ in Decision and Control, 2000. Proceedings
tolerant control systems,‖ Annual Reviews in Control, vol. 32, no. 2, of the 39th IEEE Conference on, 2000, vol. 4, pp. 4126–4131.
pp. 229–252, Dec. 2008. [21] H. Fan, B. Liu, and Y. Shen, ―Fault tolerant control for uncertain
[12] P. Millán, L. Orihuela, C. Vivas, F. R. Rubio, D. V. Dimarogonas, and systems with actuator stochastic failures,‖ in Control Automation
K. H. Johansson, ―Sensor-network-based robust distributed control and Robotics & Vision (ICARCV), 2012 12th International Conference on,
estimation,‖ Control Engineering Practice, vol. 21, no. 9, pp. 1238– 2012, pp. 70–75.
1249, Sep. 2013. [22] B. Chen, Y. Niu, and Y. Zou, ―Adaptive sliding mode control for
[13] D. Papageorgiou, M. Blanke, H. H. Niemann, and J. H. Richter, ―Fault stochastic Markovian jumping systems with actuator degradation,‖
tolerance for industrial actuators in absence of accurate models and Automatica, vol. 49, no. 6, pp. 1748–1754, Jun. 2013.
hardware redundancy,‖ in Control Applications (CCA), 2015 IEEE [23] H. Fan, B. Liu, W. Wang, and C. Wen, ―Adaptive fault-tolerant
Conference on, 2015, pp. 1887–1894. stabilization for nonlinear systems with Markovian jumping actuator
[14] X.-Z. Jin and G.-H. Yang, ―Robust Adaptive Fault-tolerant failures and stochastic noises,‖ Automatica, vol. 51, pp. 200–209, Jan.
Compensation Control with Actuator Failures and Bounded 2015.
Disturbances,‖ Acta Automatica Sinica, vol. 35, no. 3, pp. 305–309, [24] C. Cheng, Q. Zhao, and F. Tao, ―Stability and performance of the
Mar. 2009. stochastic fault tolerant control systems,‖ in Decision and Control,
[15] K. H. Johansson, ―The quadruple-tank process: a multivariable 2003. Proceedings. 42nd IEEE Conference on, 2003, vol. 3, pp. 2484–
laboratory process with an adjustable zero,‖ Control Systems 2489.
Technology, IEEE Transactions on, vol. 8, no. 3, pp. 456–465, 2000. [25] O.L.V., Costa, M.D., Fragoso, R.P, Marques. ―Discrete-Time Markov
[16] C. Peng, E. G. Tian, and T. C. Yang, ―Robust fault-tolerant control of Jump Linear Systems‖.Springer, London. 2005.
networked control systems with stochastic actuator failure,‖ IET [26] M., Verhaegen, S., Kanev, R., Hallouzi. ―Fault Tolerant Flight Control
Control Theory & Applications, vol. 4, no. 12, pp. 3003–3011, Dec. - A Survey‖, Lecture Notes in Control and Information Sciences, 2010,
2010. vol. 399, pp. 47-89.
[17] Z. Gu, J. Liu, C. Peng, and E. Tian, ―Reliable control for interval time- [27] P., Ioannou, B. Fidan. ―Adaptive Control Tutorial‖, Society for
varying delay systems subjected to actuator saturation and stochastic Industrial and Applied Mathematics (SIAM). 2006.
____________________________________________________________________________________________________________
595
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Wireless Sensor Networks (WSN’s) have been finding which are unknown. Location information of the sensor nodes
to itself new applications continuously. Many of these carry vital weight for many WSN apps. For instance;
applications need location information of nodes. The localization monitoring and surveillance of volcano activities, tsunami risk
of nodes can be made by range based or range free localization in oceans, melting regime of the poles, wild animals under
methods conventionally. Angle-of-Arrival (AoA), Time-
risk of extinction, mobility of civilian and military vehicles
Difference-of-Arrival (TDoA), Received Signal Strength
Indicator (RSSI), Time-of-Arrival (ToA) are well known range and agricultural fields require that.
based methods. Therefore AoA, ToA and TDoA have some In WSN’s, there are two typical node types. Anchor node
hardware and software difficulties for nodes which have limited (also is called Beacon) is one of them. Anchor node knows
processing and power sources. However RSSI based localization location information of itself and can send data to the adjacent
doesn’t cost high processing resources or complex hardware nodes. Other one is called non-anchor node which estimates
modifications. Most of the WSN nodes already have RSSI physical coordinates of itself by various methods through
measurement capability. However RSSI measurements is location information of the anchor nodes that are in coverage
vulnerable to noise and environmental effects. Therefore error of area of it.
RSSI based localization can be over to an acceptable level.
In literature, various localization methods have been
Centroid, APIT, DV-Hop and Amorphous are some of the range
free localization methods. Range free methods can only give classified conventionally into two categories named as range-
location information approximately but they don’t need any based and range-free methods. Range-based methods require
extra hardware or high processing capability. additional hardware. Naturally, additional hardware makes it
In this study WSN nodes are assumed randomly or regularly costly.
distributed on a certain area. Some of the nodes are beacon Angle-of-Arrival (AoA), Time-Difference-of-Arrival
nodes. The beacon nodes are assumed as having higher power (TDoA), Received Signal Strength Indicator (RSSI), Time-of-
resources and GPS receivers. The locations of nodes are assumed Arrival (ToA) are well known range based methods. AoA
as fixed. The beacon nodes send their location information needs angle of received signal, time based methods; ToA
sequentially. Localization of nodes are made through RSSI and
needs exact synchronization between nodes, TDoA needs
location information of beacon nodes. The mean of RSSI is
calculated to reduce effect of noise on it. A rough location multiple receivers and synchronization between these
estimation made by weighted centroid. A probabilistic based receivers [3]. Therefore AoA, ToA and TDoA have some
location estimation and flower pollination algorithm (FPA) are hardware and software difficulties for nodes which have
used separately to make final decision about the location. Rough limited processing and power sources.
estimates are used to limit search area of flower pollination Range-free methods generally use hop-counting and local
algorithm in order to reduce convergence time. techniques [4]. As a feature Range-free methods don’t require
additional hardware but location estimations are rougher than
Keywords— RSSI, FPA, WSN, optimization, probabilistic. range-based methods. Centroid, APIT, DV-Hop and
Amorphous are some of the range-free localization methods
I. INTRODUCTION [5]-[6]-[7].
Recently, WSN has become popular study in If RSSI based localization is taken into account, it is seen
communication. The apps like health, business, military and that it doesn’t require high processing resources or complex
habitat are the factors which made WSN has been so popular hardware modifications. Even it can be considered the least
[1]. Nodes which are the elements of the WSN have complex among the range-based methods. Most of the WSN
advantages as operating with low-power, having small nodes already have RSSI measurement capability. But on the
physical structure, low-cost and communication capability other hand RSSI measurements are under effects of noise and
with adjacent nodes in limited range [2]. These are the other environment. Therefore error of RSSI based localization
reasons why the WSN is so popular. should be over to an acceptable level. For the RSSI based
One of the main objective about WSN is acquiring localization noise and environmental effects are considerable
location information of the sensor nodes. The localization problems. To reduce these effects, there are many methods
simply means determining physical coordinates of the nodes have been proposed in literature. Some studies are based on
____________________________________________________________________________________________________________
596
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
reducing of distance error by estimating of path loss be fixed locations. In this conditions, path loss is under effect
parameters dynamically [8]-[9]. In another study, error is of Gaussian noise as seen in equation (2).
reduced by making a mapping database between RSSI and
distance information [10]. On the other hand, particle filter A. The Improvement of RSS Quality
usage may has been another solution for RSSI error. For
instance, in reference [11] a new particle filter with a First phase of proposed system is about improvement of
hardware-free initialization phase is presented. In other RSS quality. Absolute uncalibrated RSSI/CCA accuracy is
particle filter approach, multiple antenna arrays and particle ±4dB in CC2538 catalogue [15]. This accuracy means ±8m
filter are used together for reducing of RSSI error [12]. location error at 25m distance under certain conditions. For
In this study, locations of nodes are assumed as fixed. some apps that accuracy is acceptable but not always. To
Namely, environment is assumed as static. Therefore, path reduce noise and improve the accuracy, anchor nodes should
loss attenuation is also assumed a fixed value. In simulation send signal sequences to its non-anchor nodes. Each non-
scenario, the beacon nodes send their location information anchor node gets RSS sequences from the anchor nodes that
sequentially. Location estimations of the nodes are made are at coverage area of it. Due to nodes are static, distances
through RSSI and location information of beacon nodes. The don’t change. In that case, mean of each RSS sequence is an
mean of RSSI measurements are calculated to reduce effect of improved RSS information because of Gaussian noise reduced.
noise on it. A rough location estimation made by weighted This is expressed mathematically in equation (3).
centroid. In the next phase, a probabilistic based location
estimation and flower pollination algorithm (FPA) are used k
separately to make final decision about the location. Rough
estimates are used to limit search area of flower pollination RSS
i 1
i
algorithm in order to reduce convergence time. RSS IMP (3)
k
II. RECEIVED SIGNAL STRENGTH
RSSI method is based on location estimations of the non- Where, RSS IMP is improved RSS value that is acquired
anchor nodes by using signal strength of the anchor nodes from anchor node, k is the number of sequence, RSSi is the ith
where are in coverage area of the non-anchor nodes. Received
Signal Strength (RSS) can be formulated simply shown as (1). RSS value that is acquired from anchor node. Improved RSS
information is important for the distance information. For the
next phase of the proposed system, distance can be calculated
RSS (dB) PTX (dB) PLOSS (dB) (1) shown as equation (4).
Where, RSS (dB) is received signal strength acquired at ( PTX RSS IMPi PLOSS ( d 0 ))
____________________________________________________________________________________________________________
597
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
N N
RSS measurement is done from received packets,
wi .xi wi .yi
RSS is mapped to the one dimensional pdf acquired
xr i 1N , yr
i 1
N (6) from the first phase, and a pdf constraint is generated
wi wi
i 1 i 1 which is function of the 2D coordinate.
Old pdf is intersected with generated constraint and
Where, xi is abscissa of the ith anchor node, yi is ordinate update is done.
of the ith anchor node, xr is abscissa of the estimated rough At last, non-anchor node sends updated pdf estimation
coordinates and yr is ordinate of the estimated rough to all its adjacent nodes.
coordinates. Even if, the coordinates that are estimated
through weighted centroid might be sufficient for many WSN Consequently, final estimations for the non-anchor node
apps, sometimes it might not be sufficient. Hence at the last coordinates are made according to the maximum probability
step, probabilistic and FPA methods are used for the evaluation.
improvement.
V. FLOWER POLLINATION ALGORITHM (FPA)
IV. PROBABILISTIC APPROACH
Pollination process in the nature consists of two different
In this approach non-anchor nodes estimate their own forms called biotic and abiotic forms. Biotic pollination is
location through probability density function (pdf). We can carried out by pollinator creatures as bird, bat and bee. 90% of
express the probability of non-anchor node location in limited all pollination events take place in biotic form. And remaining
area at two dimensional space as in (7). 10% happens in abiotic form that is occurred by wind or water
diffusion. It does not include any pollinators. Pollinators travel
to long distances to reach plants that they wish. They
y n y x n x
maximize the pollination probability of the same-species via
PR x , y
n n
y n y
x n x
f X ,Y ( x, y )dxdy
(7) flying over other species. Pollination process happens in two
main types as self-pollination and cross-pollination. Self and
cross-pollination are shown in Fig. 1.
Where, PR (xn, yn) is the probability of the non-anchor node,
x can change between xmin and xmax and y can change between
ymin and ymax which are bounding coordinates. Δx and Δy are
arbitrary small values [17].
The first phase of this approach is calibration phase. RSS
measurements are collected at different distances between
anchor and non-anchor node. The mean value of RSS
measurements µRSS (d) and standard deviation σRSS (d) can be
calculated from measurements, where d defines distance. In
reference [17] it is mentioned that in theory and experiments
σRSS doesn’t vary significantly with the distance. If X0 is 0,
RSS probability of any distance d’ is equal to 1.
Fig. 1 Pollination types
( PTX RSS PLOSS ( d 0 ))
10 Cross-pollination is defined as pollination that occurs
p PR d' 10 1 (8) among different plants in the same species. Thus, pollinators
are very effective for the cross-pollination. On the other hand,
if the pollination occurs in the same plant, this is defined as
self-pollination. While self-pollination might occur in
Practically, channel is under effect of shadowing. Thereby, different flowers of the same plant, it might also occur in same
distribution of distance for a fixed RSS value is log-normal. flower of the same plant [18].
Consequently in calibration phase, log-normal mappings of Pollinators like bird, bat and bee show Lévy Fligt
the RSS measurements are acquired [17]. behaviour and flying steps show Lévy distribution [19].
Second phase of the probabilistic approach is localization Pseudo algorithm of the FPA can be expressed as below.
with positive constraints. In this phase, each non-anchor node
estimates its pdf position through the log-normal mappings of Rule (1): Biotic and cross-pollination is accepted as global
the RSS measurements. pollination process. Pollinators who carry pollens show Lévy
At first, each non-anchor sets initial estimation for entire Flight behaviour.
network area. Then, anchor nodes send information which Rule (2): Abiotic and self-pollination is accepted as local
includes their own position information and updated pdf pollination process.
estimations of both their own and non-anchors to the adjacent Rule (3): Flower pollination constancy is proportional
nodes. Non-anchor receives the information and executes the with the probability of breeding between two similar species.
following algorithm;
____________________________________________________________________________________________________________
598
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Rule (4): Global and local pollination probability is minimizing the value of sum of dependent errors of the every
controlled by a switching probability which is defined as p ∈ pollen.
[0, 1]. The effects like physical proximity, wind and local START Produced rough
pollination are considerable part of the global pollination. As a solutions, di (improved
result, they are also considerable for the switching probability. Define Objective function range datas)
In other words, these effects can be changed by controlling
Define Dimension of search variables by
switching probability. Weighted Centroid
In global pollination phase, the most convenient
Population size (n), max. iter., switch
pollination can be achieved by pollinators who can travel to probability (p)
long distances. While the most convenient pollination
Find the best gb according to initial conditions
parameter is gb, flower pollination constancy can be expressed
mathematically as in (9). Define a switch probability p [0,1]
i=1,t=1
i
t 1
it L it gb (9)
rand<p ?
N
Local
pollination
process
Y
i
t 1
it tj kt (10) i++
N
i==n?
Where tj and kt defines pollens where they come from Y
Find the current gb
different flowers of the same plant species. If tj and kt t++
comes from the same species or selected from the same
population, it shows random walk characteristics with uniform Y
t<max.iter. ?
distrubition as defined ϵ [0, 1] [18]. Consequently, flow chart N
of FPA can be drawn as shown in Fig. 2. Output the best
In this study unlike the conventional FPA algorithm, rough sol.
solutions and improved data are integrated to the algorithm.
Objective function of this optimization phase is expressed in END
____________________________________________________________________________________________________________
599
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
600
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
increasing of the RSS samples to 50 improved the location [18] X. S. Yang, In Unconventional computation and natural computation
(Flower pollination algorithm for global optimization), J. Durand-
accuracy approximately 50% for the FPA.
Lose, N. Jonoska (Eds.), Springer Berlin Heidelberg, 2012
TABLE II [19] I. Pavlyukevich,”Lévy flights, non-local search and simulated
RESULTS FOR RSS SAMPLE=50 annealing.” Journal of Computational Physics, vol. 226(2), pp. 1830-
1844, 2007.
Mean Error
10m 15m 30m
Weighted Centroid 5,038320472 5,145930371 5,583851882
Probabalistic 3,845946492 3,25200935 4,824532778
FPA 0,843777096 0,901263359 1,522088242
VII. CONCLUSIONS
In this study, two different approaches are analysed which
used weighted centroid rough results. Both the methods
showed improvement according to the weighted centroid
localization. When comparison is made between two methods,
it shows us that FPA has more accurate results and lower
processing load. As a future work, RSSI based location
accuracy improvement for mobile nodes will be examined.
REFERENCES
[1] D. Puccinelli & M. Haenggi, “Wireless sensor networks: applications
and challenges of ubiquitous sensing.” IEEE Circuits and systems
magazine, vol. 5(3), pp. 19-31, 2005.
[2] M. Guoqiang, B. Fidan, and BDO Anderson. "Wireless sensor network
localization techniques." Computer networks, vol. 51(10), pp. 2529-
2553, 2007.
[3] J. Bachrach. and C. Taylor, Handbook of sensor networks: Algorithms
and Architectures 1, I. Stojmenovic, New Jersey, John Wiley & Sons,
Inc , Nov. 2005.
[4] F. Liu, et al., Wireless Sensor Networks and Applications, Y. Li, M.T.
Thai, W. Wu, US, Springer, 2008, pp. 175-193.
[5] S.P Singh, and S. C. Sharma. "Range free localization techniques in
wireless sensor networks: A review," Procedia Computer Science, vol.
57, pp. 7-16, 2015.
[6] R. Stoleru, T. He, and J.A. Stankovic, Secure Localization and Time
Synchronization for Wireless Sensor and Ad Hoc Networks, R.
Poovendran, C. Wang and S. Roy, US, Springer, 2007, pp. 3-31.
[7] T. He et al. "Range-free localization schemes for large scale sensor
networks." Proceedings of the 9th annual international conference on
Mobile computing and networking, ACM, 2003, pp. 81-95.
[8] J. Zheng, et al. "An improved RSSI measurement in wireless sensor
networks." Procedia engineering, vol. 15, pp. 876-880, 2011.
[9] M. Botta, and M. Simek. "Adaptive distance estimation based on RSSI
in 802.15. 4 network." Radioengineering, vol. 22(4), pp. 1162-1168,
December 2013.
[10] H. Wang, J. Wan and R. Liu. "A novel ranging method based on RSSI",
Energy Procedia, vol. 12, pp. 230-235, 2011.
[11] Z. Wu, et al. "Improved particle filter based on WLAN RSSI
fingerprinting and smart sensors for indoor localization." Computer
Communications, vol. 83, pp. 64-71, 2016.
[12] J. Svečko, M. Malajner and D. Gleich. "Distance estimation using
RSSI and particle filter.", ISA transactions, vol. 55, pp. 275-285, 2015.
[13] M. Chen and H. Liu , “Enhance Performance of Centroid Algorithm in
Wireless Sensor Networks”, Fourth International Conference on
Computational and Information Sciences, 17-19 Aug. 2012,
Chongqing, pp. 1066-1068.
[14] L. Tan, F. Luo and K. Liu, “Weighted Centroid Location Algorithm in
Wireless Sensor Network”, Wireless Mobile and Computing
(CCWMC), 2011, pp. 414-418.
[15] “CC2538 data sheet”, Texas Intruments, Texas, US.
[16] J. Zhao et al. "An improved Weighted Centroid Localization algorithm
based on difference of estimated distances for Wireless Sensor
Networks." Telecommunication Systems, vol. 53, pp. 25-31, 2013.
[17] R. Peng, and M. L. Sichitiu. "Robust, probabilistic, constraint-based
localization for wireless sensor networks." SECON, 2005, pp. 541-550.
____________________________________________________________________________________________________________
601
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— in this paper a cascade inverter based on photovoltaic of off grid which serves the regions far from transmission
system is proposed and the control method which is used in this lines and generation units [4] .
inverter is the charge balance control method in order to
equalize the amount of power which drawn from each h Several methods are used to control the cascade h bridge
bridge(except the first one ). As a result the batteries life of the
multilevel inverter .Some of these control methods are : [6]:
inverter which is consider the most important component of the
inverter will increase .Also the magnitude of the dc sources is Fundamental frequency control method
selected in such a way that the number of levels which are Sinusoidal pulse with modulation method
produced is more than symmetrical cascade inverter .All of the Space victor pulse with modulation method
expressed theoretical results are confirmed by simulation and
experimental results. The Fundamental frequency control method is considered
the most applicable method. In this method every
Keywords— cascade multilevel inverter, photovoltaic, full wave semiconductor device is turned on one time per
charge balance control method, buck – boost dc to dc converter, fundamental cycle and that leads to reduce the
perturb and observe maximum power point method. conduction losses and the electromagnetic
interference [6]. On the other hand it has the
drawback related to the amount of power which
I. INTRODUCTION drawn from one DC source which is different from
Today, cascade H bridge multilevel inverters have got more the other sources, so the battery life will be different
interest according to their essential advantages such as: high from each other and that leads to high maintenance
voltage and high power ratings applications, excellent power cost [2] .In order to overcome the disadvantage of
quality, lower order harmonics, reduction in switching losses this method the charge balance control method is
[1-2] .The output waveform of these inverters is synthesized proposed in this work to equalize the amount of
from several dc voltage sources as inputs. power drawn from each Dc source.
Generally the multilevel inverters are divided into three In this paper, firstly we interview the basics of
types according to the topology structure, diode clamped cascade H bridge multilevel inverter, secondly brief
multilevel inverter, flying capacitor multilevel inverter, description of the pv system and then the simulation
cascade H bridge multilevel inverter [3].There is more focus and practical results will submit the proposed
on cascade H bridge multilevel inverter because of its simple cascade inverter.
design ,simple maintenance, the ability of generating large
number of output levels by using the same number of the
semiconductor devices as compared with the other topologies II. CASCADE MULTI-LEVEL INVERTERS
[2-3].Depending on the above advantages, the losses and the
cost will be reduced, on the other hand the efficiency will be
improved [2] .The sub classifications of cascade multilevel Generally, the key component of cascaded inverter is the
inverter are the symmetrical and asymmetrical inverters if one full bridge unit which is shown in Fig .1.[11].it is used to
of the input dc sources is different and then we can recognize produce three level output waveform (vdc, 0and -vdc) with
that inverter as asymmetrical inverter[2-5] which can produce different duty cycle. It consists of four switches which can
more numbers of output levels and amplitude than conduct in bidirectional power flow and can block
symmetrical type[6] .The most distinct feature of cascade unidirectional voltage. This reason is behind the connecting
multilevel inverter is that it need separate dc sources as an of anti-parallel diode with the switching devices and in case of
input so it is considered the most suitable way to use with pure resistive loads we can use the semiconductor switch
renewable energy sources such as pv panels which increases without the anti-parallel diode due to unique current
interest because of the Global increasing demand , reduces the direction .Each switching device has to switch on in a
negative impact of the power generations on the environment, complementary manner with another switch in the same arm
needs no fuel cost and reduces the transmission losses in case to avoid the short circuit across the source so all the control
____________________________________________________________________________________________________________
602
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
strategies are built according to this fact , e.g when s1 is on s3 DC sources and H bridges ,in [9 ] and [10 ] the manner of
is already off and when s2 is on s4 is already off . selecting the Dc source voltages is binary progressing in
These H bridges are connected to distinct Dc sources in the [9]and ternary progressing in [10] and the output levels (n)is
supply side while the diode clamped and flying capacitor given as(3) and (4) .
requires common DC source and connected in series in the
output side. We can call it as ac side Fig.2. It Show the single
phase diagram of cascaded H bridge inverter[11]according to = 2 -1 in binary progress ratio (3)
cascaded inverter configuration, the control and modulation
and maintenance are modular.
1= (5)
, =3 1 m = 2,3,N (6)
n= 6N-3 (7)
Fig.2. Single phase diagram of cascaded inverter Where Nis the number of the series connected H bridges
.The reason behind the selection of this manner is according to
the fact that we cannot use the charge balance control method
Where vo represent the output voltage of the cascaded in [9 ]and [10 ],because of different voltages of the batteries .
inverter and it can be calculated by adding the amounts of The main advantages of this method as compared with the
vo1(t),vo2(t),voN(t). Where N is the number of the cascaded fundamental frequency control method is to equalize the
bridges. amount of power which drawn from different batteries. As a
The relationship between the produced number of output results battery life will be equil which will reduce the
voltage levels (n) with respect to the number of connected maintenance cost .Several modulation patterns can be used to
cells N in symmetric converters is : implement the charge balance control method, in fullwave
pattern the angles will route continously after one cycle while
= 1+2 (1) in halfwave pattern the routating angle will routate after half
cycle.If any pattern is used we will make all the dc sources get
charges and be balanced after a specific time of operation
and the maximum output voltage vo is given as afunction expect the battery which is used in the first h bridges so the
with the dc sources that cascaded .as : amount of power will be given as :
= (2)
From the above two relations, we can recognize the most
related disadvantage of this topology which is increased of , = ∫ . (8)
the DC sources and the semiconductor switches when we aim
to increase the output levels, so to reduce the number of DC
sources and also the semiconductor switches and the Where is the time duration to equalize the power drawn
connecting cables the asymmetrical cascade h bridges from the bridges .
multilevel inverter is proposed. This produces maximum In Fig.3. a seven level symmetric cascade inverter is
numbers of levels and by utilizing the same number of the implemented by fundamental frequency control method , full
____________________________________________________________________________________________________________
603
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
wave charge balance control method and half wave charge operate in its maximum point so that one of the maximum
balance control method [11]. power point tracing algorithm have to be used with the dc-dc
converter to maintain the operating point of the panel near the
maximum power point area[4] .some of these techniques are
Constant Voltage MPPT Algorithm, Perturb and Observe
(P&O)MPPT Algorithm, Incremental Conductance (INC)
MPPT Algorithm[8] .and the other important component is
the dc-dc converter such as :buck ,boost ,buck boost
converter to be merged with the mppt algorithm to ensure that
the module will operate in it is maximum power point tracing
and to construct the solar charge controller [7], in this paper
we will use the following components :
____________________________________________________________________________________________________________
604
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Where D is the duty ratio which will feed the controlling observe method according to its simplicity and accuracy .In
switch and vo in both modes in average values . the Fig .5.bellow it shows the flow chart of this technique [8].
In this method, the sign of the last perturbation and the sign
Io is the current through the load . of the last increment in the power are used to decide what the
next perturbation should be. On the left of the MPP
incrementing the voltage increases the power whereas on the
In continuous conduction mode (ccm) when D < 0.5 it will right decrementing the voltage increases the power.
operate in buck mode while when D > 0.5 it will be in boost
mode .
____________________________________________________________________________________________________________
605
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Figures(6-7-8) bellow show the duty cycles of the first, second Vdc, 2,3 = 41.5 -42 v
and third bridges respectively . We put this range according to the control which is merged
with the mppt algorithm to control the battery voltages.
The voltage of the batteries is shown in the figures (13-14-15):
Fig .7. Second H bridge duty cycle. Fig . 14. Second battery voltage
Fig .8. Third H bridge duty cycle. Fig. 15. Third battery voltage
From the above figures it is clear that the first pv module
operates in buck mode and its duty cycle around 0.36( less
than 0.5) according to the buck –boost dc – dc converter Figures (16-17-18) show the output of each H bridge and the
characteristic while the second and the third H bridges final waveform (15 levels) is shown in fig 19 which represents
operate in boost mode and its duty cycle = 0.56(greater than the summation of the three H Bridges voltages.
0.5) .
The battery voltage of the first, second and the third is as
follows vdc1 = 12.5 -14 v
____________________________________________________________________________________________________________
606
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V-EXPERIMENTAL RESULTS
The laboratory setup consists of a single phase 15 level
cascaded asymmetrical multilevel inverter. This inverter
consists of three H bridges which is able to produce 15 level
output with a maximum output voltage equal to 7 vdc .
Each H bridge consists of six semiconductor switch. Four
of them build the H bridge frame and the type of this
transistor of the prototype are tip142 (npn bjt) for the upper
Q1,Q3 and tip147(pnpbjt) Q2,Q4 for the lower while the
other two transistors are used for controlling and the type of
them are 2n3904(npn) Q5,Q6 .Fig . 20 shows one H bridge
diagram in hardware configuration. Fig .21 . First H bridge output(hardware).
____________________________________________________________________________________________________________
607
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig .22 . Second H bridge output(hardware). Fig .24 . Output wave form fifteen level.
VI-Conclusions
In this work an improvement in the cascade inverter is done .
This improvement includes two points: the first is related to
the number of the output levels which is increased by using
the asymmetrical topology with minimum number of dc
sources and semiconductor switches and the second concern
with the amount of power which is drawn from each DC
source (except the first source) is equalized by using the
charge balance control method .This inverter works perfectly
with direct connected batteries and pv panels as shown in the
simulation result, also experimental work shows the
performance of the proposed inverter .
The output wave form of the inverter (15 level ,7 positive ,7 negative ,1 zero )
Is shown in Fig.24 .
____________________________________________________________________________________________________________
608
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] K.W.E. Cheng and Y.P. Zou, s,2 K. Ding, "Analysis of an asymmetric
modulation method for cascaded multilevel inverter," IET Power
Electronic Journal, vol. 5, pp. 74-85, 2012.
[2] E. & Hosseini, S. H. Babaei, "Charge balance control methods for
asymmetrical cascade multilevel converters ," in electrical machines and
systems , seoul, 2007, pp. 74-79.
[3] H. Fathi, N. Yousefpoor and M.K. Bakhshizadeh N. Farokhnia,
"“Minimisation of total harmonic distortion in a cascaded multi
levelinverter by regulating of voltages dc sources.," IET Power
ElectronicJournal, vol. 5, no. no. 1, pp. pp. 106-114, 2012
[4] J.Tadeo, F., Prada, C., & Palacin Salazar, "Simulation and Control of a
PV System connected to a Low Voltage Network.," in Days of
Automatic, Jaén, 2010, pp.1-6.
[5] K. Ramani and A. krishan, "New hybrid multilevel inverter fed,"
proc.IREE, vol. 5, no.11.6,part, pp. 2562-2569, Dec 2010.
[6] E.Babaei, & Hosseini, S. H. (2009).., 50(11), , "New cascaded multilevel
inverter topology with minimum number of switches," Elsevier journal
of Energy Conversion and Management, vol. 50, no.11, pp. 2761-2767,
Nov 2009.
[7] J. Ijaz, U., Niazi, M. H. K., Amer, M. S., Younis, M. M., & Hassan, W.
(2015, June). Ali, "Simulation and implementation of solar power battery
charger using Perturb & Observe algorithm. In Power Generation
System and Renewable Energy Technologies (PGSRET)," in IEEE, june
2015, pp 1-5M.
[8] U. Sahu, M. D., & Tirkey, D. (2013, September). Patel, "Maximum
power point tracking using preturb and observe algorithum and compare
with anathor algorithm ," international journal of digital application
&contemporary research conference , vol. 2, no.2, Sep 2013.
[9] M. Manjrekar and T. A. Lipo, "“A hybrid Multilevel Inverter Topology
for Drive application," in APEC 98, 1998, pp. 523-529..
[10] M. Veenstra and A. Gopakumar,“ A. Rufer, "Asymmetric Multilevel
Converter for high resolution voltage phasor generation," in EPE99,
lausanne, 1999.
[11] S.H Hosseni E.babaei, "charge balanced control method for cascade
multilevel inverter ," icems power electronic journals , 2007.
[12] S.Abbaszadeh, K., & Lesani, H. Laali,".A new algorithm to determine
the magnitude of dc voltage sources in asymmetric cascaded multilevel
converters capable of using charge balance control methods," in
electrical machines and systems , 2010.
[13] H. Rashid, "Power electronics: circuits, devices,and applications.," in
IEEE education india , pearson, 2009, pp. 56-61.
____________________________________________________________________________________________________________
609
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Conventional data driven process monitoring The remainder of this paper is organized as follows. Section
algorithms are limited to Gaussian process data for principal II describes PCA and ICA algorithms. How to obtain SVDD
component analysis (PCA) algorithm and non-Gaussian process algorithm is given in section III. Section IV gives monitoring
data for independent component analysis (ICA) algorithm. This results of the application to TE process. Finally, section V
paper provides a comparison study between the conventional data
provides a concluding summary of this paper.
driven methods and support vector data description (SVDD)
algorithm for fault detection (FD). Different from the traditional
II. PROCESS MONITORING BASED ON PCA AND ICA
methods, SVDD algorithm has no Gausssian assumption. Thus the
distribution of process data is not important for SVDD method. In The basic model considered in PCA and ICA is
order to compare their FD performances of the proposed methods 𝒙 = 𝑨𝒔 (1)
from the application viewpoint, Tennessee Eastman (TE)
benchmark process is utilized to compare the results of all the where 𝒔 = [𝑠1 , … , 𝑠𝑑 ]𝑻 is a vector of unknown source signals
discussed methods. Simulation results on TE process show that which are independent, and 𝒙 is obsevation vector which is
ICA and SVDD methods perform better for false faults than the mixtures of the source signals via unknown mixing matrix 𝑨.
PCA method. The objective of PCA is to make variables uncorrelated through
orthogonal rotation. The orthogonal rotation matrix is obtained
Keywords— Process Monitoring, Fault Detection, Support Vector by using eigenvalue decomposition on covariance matrix
Data Description, Independent Component Analysis, Principal 𝑿𝑻 𝑿 (2)
Component Analysis, Statistical Process Control 𝚺= = 𝑽𝚲𝑽𝑻
𝑚−1
I. INTRODUCTION
where 𝚲 is the diagonal eigenvalue matrix with its diagonal
In statistical process control (SPC) systems there exist many
elements in decreasing order, 𝑽 is orthogonal eigenvector
variables need to be monitored. These measurements provide
useful information about the status of the systems. So applying matrix . The loading matrix 𝑷 is formed by a first vector which
the univariate SPC methods on that systems may produce false possesses acceptable percent of data variance. The
alarms. Using multivariate SPC (MSPC) methods can transformation of 𝑿 matrix is called as the score matrix and
overcome that problem. MSPC has several advantages over calculated as follows
univariates, for example, showing relationships between 𝑻 = 𝑿𝑷 (3)
variables which cannot be detected with univariate statistics, The transformation into the original space is achieved by using
and help to understand the interaction between variables. PCA (3)
and ICA algorithms have been widely used as a multivariate ̂ = 𝑻𝑷𝑻
𝑿 (4)
statistic intended to find latent variables in the FD field [1-6].
PCA is a well-known algorithm, and depends only on the The residual matrix 𝑬 is calculated as
second order statistics which means that the latent variables 𝑬=𝑿−𝑿 ̂ = 𝑿 − 𝑻𝑷𝑻 (5)
capture the most variance of the source signals. Unlike PCA, PCA algorithm divides data space into two parts. First part is
the goal of ICA algorithms is to minimize the statistical determined by a first major component and has the greatest data
dependence between the basis vectors, and there is no closed variance. The other part is defined by a small percentage of
form expression for ICA algorithms. data variance and shows noise. 𝑇 2 statistic can be used to
Another efficient FD algorithm is SVDD algorithm measure the variation of PCA model for a loading vector.
proposed by Tax and Duin [7]. SVDD algorithm aims at finding 𝑇 2 = 𝒙𝑻 𝑽𝚲−𝟏 𝑻
𝒂 𝑽 𝒙 (6)
spherically shaped boundary around a data set. In this boundary,
a hypersphere enclosing most of the data set belonging to the where 𝚲𝒂 is the first a rows and columns of 𝚲. In PCA the 𝑿
class of interest and rejecting the outliers. SVDD is a new data is assumed to follow a multivariate normal distribution so
method in the FD area, but it has been used in a wide range of 𝑇 2 follows an F distribution with degrees of freedom 𝐴 and
FD applications [8-10]. In this paper, it is employed the one- 𝐴 − 𝑀, and confidence limits can be calculated as follows
class SVDD to find the separating boundary the normal data set 𝐴(𝑀2 − 1) (7)
T𝛼2 = 𝐹
and faulty data set. In this respect, SVDD is implemented and 𝑀(𝑀 − 𝐴) 𝐴,𝑀−𝐴,𝛼
compared with standard MSPC methods for FD.
____________________________________________________________________________________________________________
610
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
3. Normalize 𝒘 ← 𝒘⁄‖𝒘‖.
IV. COMPARISION STUDY BASEN ON TE
4. If not converged, go back to step 2. In this section PCA, ICA and SVDD methods will be
applied to TE process for a comparison study. TE simulator was
Separating matrix W is obtained by assembling all the vector w, developed by Downs and Vogel which produces two products
and the demixing sources are calculated as from four reactants [15]. It has five major units, i.e. reactor,
𝒔̂ = 𝑾𝒙 (13) condenser, compressor, separator, and stripper. There are 41
After obtaining W, it is divided into two parts, dominant part measured variables and 12 manipulated variables. In TE
(𝑾𝒅 ), excluded part (𝑾𝒆 ), and using these parts three statistics process, 20 process faults and an additional valve faults were
are calculated as follows defined by Downs and Vogel [15]. The sample length of the
training data set is 500 under normal operating conditions. Each
𝐼2 = ̂ 𝑺 𝒅 = 𝒙𝑻 𝑾 𝒅 𝑻 𝑾 𝒅 𝒙
𝑺𝑻𝒅 ̂ (14)
testing data set for one fault mode consist of 960 samples, and
𝐼𝑒2 = ̂ 𝑺𝒆 = 𝒙𝑻 𝑾𝒆 𝑻 𝑾𝒆 𝒙
𝑺𝑻𝒆 ̂ (15) fault was introduced at sample 160 for each data set. All the
data were autoscaled prior to application of the algorithms. To
̂)𝑻 (𝒙 − 𝒙
𝑆𝑃𝐸 = (𝒙 − 𝒙 ̂) (16) obtain better effect of simulation, 52 variables are selected for
−𝟏 analysis. The confidence limit of kernel density estimation is
where 𝒙̂ = (𝚲−𝟎.𝟓 𝑽𝑻 ) 𝑩𝒅 𝑾𝒅 𝒙. In ICA, the latent variables selected as 99%. Fault detection rate (FDR) and false alarm rate
are assumed to be non-Gaussian distributed, hence the (FAR) are considered to evaluate FD performance [16,17].
confidence limits of tree statistics are calculated by kernel FDR and FAR rates are defined as
density estimation method [13]. 𝐹𝐷𝑅 = No. of samples (𝐽 > 𝐽𝑡ℎ |𝑓 ≠ 0) (22)
𝐹𝐴𝑅 = No. of samples (𝐽 > 𝐽𝑡ℎ |𝑓 = 0) (23)
III. PROCESS MONITORING BASED ON SVDD
____________________________________________________________________________________________________________
611
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE I TABLE II
DESCRIPTION OF PROCESS FAULTS, AND FDRS OF ALGORITHMS DESCRIPTION OF FARS
Fault ALGO- Fault Detection Rate Fault Fault ALGO- False Alarm Rate
Mode RITHMS (FDR) Type Mode RITHMS (FAR)
If one of the test statistics exceeds threshold, a fault can be According to Table I, all the tested methods give similar
detected successfully. The FD performances of the methods are FDRs but different FARs (Table II). However FARs of ICA
investigated for 11 typical fault modes in TE process, and the and SVDD have almost the same, and are lower than PCA.
results are calculated and tabled in Table I. Also, in Table 1 the Therefore, ICA and SVDD algorithms can be preferred to PCA
fault types and fault modes can be seen. As it can be seen in to obtain lower FARs, and may provide the process operator
Table 1, FDRs are almost close each other. For fault 15 all with more reliable information. The drawback of the ICA
algorithms show poor FD performance. FARs for fault 6 is very algorithm is that using the kernel density estimation method is
low for FastICA and SVDD in Table II. computationally expensive. Unlike the ICA, SVDD does not
According to FARs given in Table II, PCA has higher FARs suffer from a high computational load, so using SVDD is more
than FastICA and SVDD, which means PCA produces more appropriate for FD applications.
false alarms than the others. For fault 12 they have almost the
same FARs, but for fault 8 only FastICA approach gives the
best FAR. REFERENCES
[1] M. B. Shams, H. M. Budman, and T. A. Duever, “Fault detection,
identification and diagnosis using CUSUM based PCA,” Chemical
Engineering Science, 66(20), 4488-4498, 2011.
V. CONCLUSIONS [2] T. Villegas, M. J. Fuente, and M. Rodríguez, “Principal component
In this paper, PCA, ICA and SVDD methods have been analysis for fault detection and diagnosis. experience with a pilot plant,”
In CIMMACS'10 Proceedings of the 9th WSEAS international
introduced, which was based on linear (PCA, ICA) and
conference on computational intelligence, man-machine systems and
nonlinear process monitoring techniques (SVDD). Then, all the cybernetics , 2010, December, pp. 147-152.
methods were implemented on TE process to compare the
sensivity of the algorithms quantified by calculating the FARS.
____________________________________________________________________________________________________________
612
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
613
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The knowledge of Line of Sight (LoS) probability is points in two or three dimensional space. LOS is an ideal
curicial to estimate signal attenuation correctly in mobile wireless condition for mobile ground communication because
communication. Especially in built-up areas, more accurate LoS transmitted signal is received with best possible signal strength
probability determination helps to obtain more realistic without exposing any obstacles [1]. In NLOS, there are many
propagation models or path loss models. Geographic Information
Systems (GIS) and City Information Systems (CIS) are used to
obstacles such as buildings, trees, forest, poles, tunnels and
provide a necessary data to calculate the LoS probability. In this traffic lights. In this case, transmitted signal reaches the
study, LoS analyses are made via Arcgis software for the most receiver weakened. Apart from these, transmitted signal also
well-known streets in Trabzon, Turkey. For these analyses, the get attenuated due to rain. In urban type environment, the signal
Earth''s surface is accepted as flat and a simple geometrical gets attenuated especially by buildings. Urban type
approach is used for calculations in this paper. A Matlab environment is covered dense buildings which are closely
algorithm was created to calculate LoS probability depending on located to each other and streets. Therefore, transmitted signal
the elevation angle which is an important parameter for satellite gets attenuated as a result of diffraction and reflection caused
services. LoS probability vs. elevation angle is presented for by these buildings.
interested streets. As a result, LoS probability for Trabzon
dependent elevation angle is estimated and presented.
In urban type environment, when examined it was observed
that signal quality in mobile communication systems is
significantly affected on the streets for LOS and NLOS cases
Keywords— Line of Sight, Line of Sight Probability, LOS,
Wireless Communication, GIS, CIS, Arcgis, Arcmap, Elevation
[1]. Therefore, the height of the buildings, transmitter height,
angle, Matlab receiver height, elevation angle from receiver to transmitter,
street widths and lengths are most important factors to
I. INTRODUCTION determine the LOS and NLOS conditions created by buildings.
In terrastial mobile communications, above mentioned
The increasing demand for higher data rate in wireless characteristics of the building and the street can be obtained via
mobile communications services has expedited the need for digital maps which is used in geographic information
more innovative and flexible communications infrastructures. systems(GIS) [3]. Transferring the entire city to digital map is
Terrestrial ground-based systems and satellite systems are used a difficult and sensitive process. This has led to the emergence
for providing mobile communications services [1]. To of City Information Systems(CIS) [4]. In Turkey, all
overcome some of the disadvantages of both terrestrial ground- municipalities are gradually creating their City Information
based systems and satellite systems, high altitude plAtform Systems(CIS). Some of these cities are İstanbul, İzmir,
stations (HAPs) technology can be used to provide cellular Eskişehir, Konya, Kayseri, and Trabzon. In this study, City
communication. So, there is plenty of research in this area [2]. Information Systems of Trabzon Municipality is used to obtain
In [1], researchers obtain propagation models and performance LOS probability values [5].
analyses for HAPs systems. It is obvious that elevation angle A lot of software is available in the GIS and many of them
is the dominant parameter on the propagation models in [1]. are paid. ARCGIS [6] and MAPINFO [7] could be given as an
The simplest definition of Line of Sight (LOS) is the straight example for most commonly used paid software. Also,
path between two fixed point in two or three dimensional space NETCAD [8] can be given as an example national paid
when unobstructed by the horizon. On the other hand, non-line- software. Some of the most commonly used free software is
of-sight (NLOS) is indirect path from one point to the other GRASS [9], QGIS [10] ve SAGA GIS [11]. Urban Information
System of Trabzon Municipality was created by the NETCAD
____________________________________________________________________________________________________________
614
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
software. In this study, GIS data sets received from Trabzon use the data from existing GIS to calculate the LOS coverage
Municipality is adapted to ArcGIS software. probability in Prague capital city of the Czech Republic [20].
In her study [1], Hasırcı used direct sight (LOS) and non- Oestges and Janvier have represented the physical and statical
direct sight (NLOS) probability values from Holis and Pchac model of LOS coverage probability calculation for HAPs [21].
[12]. Depending on this probability values, propagation models Saunders and Evans has provided a physical model which
have created and performance analyses of communication calculates the probability LOS for land mobile satellite systems
channels are made in [1]. derived from a simple geometric model [22]. RamaSarma, in
In this study, the most well-known streets and avenues for his research, has calculated the probabilities of LOS for
the city of Trabzon were considered. For each street and avenue interested regions and stated it analytically [23]. RamaSarma is
LOS and NLOS probabilities were calculated depending on the used two models in his research namely CRABS and
elevation angle through a code written in Matlab. After that, PLEXTEK which estimate visibility [23]. Both models have a
calculated LOS probabilities for streets were combined with structure that will be an alternative to the ray tracing method in
Bayes conditional probability theorem [24] by performing the literature. In both models, required parameters are produce
simulation to create Trabzon city LOS possibilities. by statistical methods from existing GIS data. Using these
Realistic LOS probability values calculated in this study will parameters, the radio propagation models were estimated.
be put in LOS probability equation (1) from [12] and HAPs
propagation models and performance analyses will be obtained II. LINE-OF-SIGHT PROBABILITY MODELLING APPOACH
for Trabzon. Also, these calculated real probability values can
be used to estimate propagation and coverage area for Trabzon. In this section, propagation modelling and statistical study
This LOS modelling idea is not new. Actually, LOS analyses for channel performance analysis on HAPs systems in [1] were
were used in the military fields. First LOS analyses were made investigated. In above mentioned study, the analytical LOS
at the beginning of the 18 century by the French military probability expression depending on elevation angle was
engineer Prestre de Vauban (1603-1707). LOS analyses slowly obtained. In her research, Hasırcı emphasizes propagation
began to gain importance over the following years because of modelling of HAPs and gives performace analysis [1]. All
increasing military activities as a result of wars [13]. LOS possible propagation environments were divided into four
analysis has played an important role in military especially for groups: suburban (SU), urban (U), dense urban (DU) and urban
the placement of an optimum observation tower, selection of high rise (UHR) area. These propagation environments were
military areas, and preparation of war plans. modeled using well-known statistical models with a
ARCGIS software one of most commonly used GIS software. dependence on elevation angle. Focusing only urban areas in
ArcGIS ™ technology, developed by ESRI, is a scalable this study provides a more detailed examination. With rapidly
integrated Geographic Information System(GIS) software. growing metropolitan areas, more detailed studies that have
ArcObjects is a development environment of the ArcGIS become much more important in signal attenuation will help to
family of applications [14]. In this study, ARCMAP which is solve this problem. Statistical models were combined with free
one of the module of ARCGIS software is used. ARCMAP has space path loss, and full formulations of total path loss for the
provided visualisation of available graphical and verbal data, four possible HAPs propagation environments and different
data update, query and analysis, charting and reporting tools conditions at 2-6 GHz frequency band were obtained [1].
and, it also has high quality cartographic production functions. Elavation angle and propagation environment are the most
In ArcGIS 3D analysis tool, there are functions such as LOS important parameters to determine propagation characteristics
(places visible and not visible among a line), Viewshed (visible for HAPs systems.
areal places and not visible areal places for desired specific First, using ITU-R Rec. P.1410 [17] statistical model, LOS
point of view) and drawing LOS profile [14]. and NLOS probability distributions corresponding to each
Height of buildings and the distances between the buildings elevation angle in each propagation environment were
are important parameters for radio propagation models and calculated [12], the data has been created to be used in the
LOS probability calculations [1]. In the literature, it was shown produced model.
that building height distributions have similarities with some Probability distribution for LOS and NLOS corresponding to
known probability distributions. each elevation angle have been created on the following
When these studies were analyzed, it is clear that rayleigh geometry for four propagation environment [see Fig. 1]. This
distribution is most commonly used distribution. Log-normal geometry is essential scenario for simulation result in this paper.
distribution is also used in some studies. Some references state
that building height densities are similar to log-normal
distribution rather than Rayleigh distribution while creating
radio propagation models [16,17]. The distributions related to
the height of the building are also investigated thoroughly in
[15], [17] and [18].
Cheng and Wu have represented the LOS calculation model
which determines security and route for vehicles traveling at
highway [19]. Led and Pechac, using the ITU-R P.1410 model,
____________________________________________________________________________________________________________
615
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
t n
PLOS ( ) t (1)
k p
1 ( )
l
TABLE 1
PARAMETERS FOR LOS PROBABILITY CALCULATION [12].
Propagation t n k l p
Medium
Suburban(SU) 101.6 0 0 3.25 1.241
Fig. 1. Geometry of basic LOS and NLOS [12]. Urban(U) 120 0 0 24.3 1.229
A 2 x 2 km size city area was considered in analyses and the Dense 187.3 0 0 82.1 1.478
layout of the building sampled every meter in simulation setup. Urban(DU)
Simulation were separated in two group. In first group Urban High- 352.0 -1.37 -53 173.8 4.670
simulations, LOS probability in the streets was analyzed as a Rise(UHR)
function of elevation angle for different propagation
environment filled with buildings. A much simpler geometry Where PLOS ( ) is the probability of LOS in percent, is an
was shown in Fig.1 for LOS and NLOS conditions of area. elevation angle in degrees and t,n,k,l,p are the empirical
Calculations were made every 9 degree increments for azimuth parameters given in Table I for the four typical environments.
angles in a range of from 0 to 360 degrees. Initially, buildings
were randomly generated by the statistical models. After that,
to implement and analyze a large number of scenarios, HAPs III. OUR EMPIRICAL LOS PROBABILITY
location were determined separately for each sampled point of MODELLING APPROACH
the street for given elevation and azimuth angles. LOS
probability for a given elevation angle were set to the median Flat unobstructed path from the transmitter to the receiver is
value of data obtained for all azimuth angles. Thus, the results called free line of sight. Line of Sight is very important for
have become independent from the azimuth angle because transmitted signal in wireless communication. Even there is not
buildings in the real world are usually not located regularly. line of sight in the signal path, the signal can still reach its target.
These simulations were applied from 1 degree to 90 degrees for Actually, if all transmitted signals move through free Line of
all elevation angles and a distribution was obtained as in Fig. 2 sight path, the best propagation occurs. The high-frequency
signals need more free line of sight path than the low frequency
signals. Infrared communication is particularly sensitive to
obstacles which is in the free LOS fields.
In this study, firstly, it was opened Trabzon data set which
have binalar.shp, yollar.shp and mahalleler.shp file. After that,
they were compiled via ARCMAP module of ARCGIS
software. The most well-known streets and avenues were
selected in the city center of Trabzon, Turkey. After that, the
buildings which is on the north side of this streets and at a
certain distance from these streets were selected. Finally,
various operations which use some ARCMAP tools were
applied for selected buildings and streets. As a result of these
processes, the knowledges such as the height of the buildings,
the area occupied by the buildings, the edge lengths of the
buildings and the distance to the road from buildings are
extracted from the attribute tables which are obtained from
Trabzon GIS dataset. It was created a Matlab algorithm that
calculates LOS probability depending on the elevation angle for
interested streets. Input variables of this algorithm are the
Fig. 2 LOS probability in the streets as a function of the elevation angle for height of the building, the edge lengths of the building and the
selected environments [12].
distance from building to the road. Firstly, LOS probability was
Equation (1) depending on the elevation angle was obtained calculated for every building on the interested street. After that,
with the help of the data in Fig. 2. Additive LOS probability which consists also building spaces
____________________________________________________________________________________________________________
616
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
was calculated and this additive LOS probability was added to Fig. 4 illustrates a simple LOS scenario in Urban city. When
pre-calculated LOS probability created by building. As a result Fig.4 is examined, it can be seen that there are buildings, trees
of this, Elevation angle dependent LoS probability for and cars that prevent direct sight (LOS) between the transmitter
interested street were estimated. When it was examined LOS and the receiver. They are common obstacles that may exist in
probability graphs in this study, it has been seen that they are all the typical urban-type settlements.
similar to some known probability distributions in the literature
[25]. This study has been no analysis of similarity to known
probability distributions in the literature. LOS probability
values depending on the elevation angle for Trabzon was
obtained by utilizing Bayes' conditional probability theorem
[24].
____________________________________________________________________________________________________________
617
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
As seen in Fig. 5, Hi is observation height, Li is distance of angle, and L is total length of N number streets (given by
LOS, e is elevation angle, Gi is distance to the receiver from equation (9)).
The clearest expression of equation (7) is given by equation
building edge which is parallel to road, ‘i’ is building index (8).
which helps to differentiate buildings from each other in the L1 L2 LN
street and ‘e’ index represents elevation angle ranging from 1 Ploscity(i) * Plos(L1 \ i) * Plos(L2 \ i) .. * Plos(LN\ i) (8)
L L L
- 90 . Relationship between elevation angle, e and distance of
LOS, Li can be stated as shown in equation (2). L in equation (7) and (8) can be calculated as below:
Hi L=L1+L2+L3……..+LN (9)
tan(e ) and Li= Hi 2 Gi 2 (2)
Gi
In above equations, Li and L are easily obtained by using
First, NLOS probability was calculated in this study. After ARCGIS attribute tables.
that, LOS probability was calculated. Relationship between D. Our LOS Probability Graphics for Trabzon street and city
these probabilities was shown below.
In this section, LOS probability figures vs. elevation angle
Sall=Snlos+Slos (3) were presented for four streets. The LOS probability figures
were plotted in Matlab software. LOS probability of four streets
where Sall is total area of the interested region in square were given this section (Fig. 6, Fig 7, Fig. 8 and Fig. 9). After
meters, Snlos is area of NLOS region in square meters, and Slos that, LOS probability figure of Akif Saruhan street was
is area of LOS region in square meters. Relation between area examined and discussed. Fig.6 illustrates LOS probability
of regions is was given in equation (3). NLOS probability can figure vs. elevation angle of Akif Saruhan street.
be calculated by using equation (4). Fig. 6 illustrates LOS probability figure vs. elevation angle
of Akif Saruhan street. Fig. 7 illustrates LOS probability figure
Pnlos=Ss/Sall (4) vs. elevation angle of Deliklitaş street. Fig. 8 illustrates LOS
probability figure vs. elevation angle of Ahmet Barutçu
where Pnlos is NLOS probability. Kütüphanesi street. Fig. 9 illustrates LOS probability figure vs.
elevation angle of Moloz street. Fig. 10 illustrates LOS
Substituting Pnlos in equation (4), LOS probability can be probability figure vs. elevation angle of Trabzon city.
calculated using equation (6).
Plos+Pnlos=1 (5)
Plos=1-Pnlos (6)
____________________________________________________________________________________________________________
618
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Z.HASIRCI, “Yüksek rakımlı platform istasyonlarda (haps) yayılım
Modellemesi ve başarım analizi,” MSc.thesis, Electrical and electronics
Fig. 9. LOS probability of Moloz street engineering department, Blacksea Technıcial University institute of
Science, Trabzon,Turkey, May.2011.
[2] S. Ohmori, Y.Yamao and N.Nakajima, “The future generations of
mobile communications based on broadband access technologies,” IEEE
communications magazine, vol.38, pp. 134-142, Dec.2000.
____________________________________________________________________________________________________________
619
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[3] Geographic information systems. (2014) Wikipedia homepage. [Online]. [16] R.M.Barts and W.L.Stutzman, (1992). “Modeling and simulation of
Available: http://tr.wikipedia.org/wiki/Co%C4%9Frafi_bilgi_sistemi/ mobile satellite propagation,” IEEE transactions on antennas and
[4] Kent Bilgi Sistemleri. (2014). CBS homepage on CBS. [Online]. propagation, vol.40, pp. 375-382, Apr.1992.
Available: http://www.csb.gov.tr/gm/cbs/ [17] ITU-R P.1410-2, Propagation data and prediction methods required for
[5] 3-D Trabzon Kent Bilgi Sistemi. (2014). Netgis Harita Servis İşlemleri design of terrestrial broadband millimetric radio access system operating
homepage on Keos. [Online]. Available: www.trabzon.bel.tr/ in a frequency range about 20–50 GHz, ITU, 2005.
[6] ArcGIS software manual. (2014). [Online]. Available: [18] C.Cheon, G.Liang and H.L.Bertoni, “Simulating radio channel statistics
http://www.arcgis.com/home/index.html for different building environments,” IEEE Journal on Selected Areas in
[7] Mapinfo software manual. (2014). [Online]. Available: Communications, vol.19, pp. 2191-2200, Nov.2001.
www.mapinfo.com/ [19] L.Cheng and H.Wu, “A study on rush hour vehicular communication
[8] Netcad software manual. (2014). [Online]. Available: blockage.” IEEE International Symposium on Antennas and
www.netcad.com.tr/ Propagation (APSURSI), July 2011, pp. 829-831.
[9] Grass GIS Software manual. (2014). [Online]. Available: [20] P.LEDL and P. PECHAC, “Area Coverage Simulations for Millimeter
http://grass.osgeo.org/ Point-to-Multipoint Systems Using Building Blockage,”
[10] QGIS software manual. (2014). [Online]. Available: Radioengineering , vol.11, pp.43-47, Dec.2002.
http://www.qgis.org/en/site/ [21] C.Oestges and D.Vanhoenacker-Janvier, “Coverage modelling of high-
[11] Saga-GIS software manual. (2014). [Online]. Available: altitude platforms communication systems,” Electronics Letters, vol.37,
http://www.saga-gis.org/en/index.html 12.15 05.09.2014 pp.119-121, Jan.2001.
[12] J.Holis, J and P.Pechac, “Elevation dependent shadowing model for [22] S.R.Saunders and B.G.Evans. “Physical model of shadowing probability
mobile communications via high altitude platforms in built-up areas,” for land mobile satellite propagation,” Electronics Letters, vol.32, pp.
IEEE Transactions on Antennas and Propagation, vol.56, pp. 1078-1084, 1548-1549. Aug.1996.
Apr.2008. [23] V.RamaSarma, “A Coverage Area Estimation Model for Interference-
[13] D. R.Caldwell, Studies in military geography and geology. Cluver Limited Non-Line-of-Sight Point-to-Multipoint Fixed Broadband
academic publishers, Dordrecht, 2004. Wireless Communication Systems,” MSc.thesis, Electrical and
[14] Çiğdem KOL and Serdar KÜPCÜ, ArcGIS 3D Analiz eğitim computer engineering, Virginia Polytechnic Institute and State
dökümanları homepage on EsriTurkey. (2014). [Online]. Available: University, Virginia , USA, Apr.2002.
http://www.esriturkey.com.tr/egitim/egitim-dokumanlari/ [24] [2014]. The Wikipedia website. [Online]. Available:
[15] J.Holis and P.Pechac, “Penetration loss measurement and modeling for http://tr.wikipedia.org/wiki/Bayes_teoremi/
HAP mobile systems in urban environment,” EURASIP Journal on [25] [2014]. The Wikipedia website. [Online]. Available:
Wireless Communications and Networking, pp. 1-7, Apr.2008. http://en.wikipedia.org/wiki/List_of_probability_distributions
____________________________________________________________________________________________________________
620
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, an estimation algorithm is proposed to proposed method gives the approximate solution of the
estimate the state variables of linear time-invariant multi input- estimation vector as a function of time in the interval
multi output systems. The proposed recursive algorithm is based [0,t]. Consequently, computation of the state integral
on the orthogonal Taylor series approximation and uses some equations for each t is eliminated. The Taylor series are
important properties of the Taylor series. It has an analog
solution which can be obtained easily by a computer program. It
defined on the interval t[0,1] and have the orthogonality
is not affected by rounding errors. When the number of elements property like the Walsh, Chebyshev and Legendre series
of series is sufficiently large, the Taylor series approximation [7,8]. The proposed algorithm uses some important
gives results very close to exact solution. properties such as the operational matrix of integration for
Taylor vector [9,10]. The algorithm consists of four steps.
Keywords -- State Estimation, Taylor series, State Observers, In the first step, the feedback gain matrix G, which will
Curve fitting force the estimation error to go to zero in a short time, is
determined by using a suitable method [4]. In the second
I. INTRODUCTION step, the observer state equation is converted into integral
State variables that determine a system’s dynamics should equation by integrating the terms on either side of the
be known for analysis and control of dynamical systems equation. After some algebraic manipulations, the time
[1,2]. Specifically, dynamics feedback for pole placement is dependent terms on either side of the integral equation are
required. Furthermore, estimation of state variables in real removed. Hence, the problem is reduced to a set of
time is a very important problem in adaptive control nonlinear equations with constant coefficients. System
applications [3]. Unfortunately, all of the state variables outputs are used by the observer equations. Therefore, we
cannot be measured in practice. As a result, use of a suitable have to calculate it’s as the function of time. They can
state observer or estimator is unavoidable in order to obtain obtained from plant output measurement by using curve
immeasurable state variables. There exit a variety of state fitting methods such as Linear Least Squares, Levenberg-
observers in the literature [4,5]. Implementation of state Marquardt and Gauss-Newton [11]. Finally, in the last step,
observers that use only input and output measurements of nonlinear equations for unknown state vector are converted
the systems are carried out via solution of the observer state into a recursive form whose solution can be obtained easily
integral equations pertinent to the observer. There are by a computer program. The proposed estimation algorithm
several numerical solution algorithms for a solution of the was implemented in MATLABTM and it was applied to
observer state integral equations in the literature [6]. Even different cases. Results obtained by the proposed algorithm
though the Runge-Kutta numerical integration algorithm is are in harmony with the real results.
frequently used for this purpose, it has several drawbacks This document represents a template for ICAT’14. It can
that depend on the step-size h. First, accuracy gets poorer as be downloaded from the conference website, and used as a
h increases. Second, computation time becomes an issue if reference in the typesetting of the final paper to be included
h is too small. Third, round-off errors may become in the conference proceedings. Extra information regarding
important for small values of h because the number of the submission procedure is available at the conference
cycles required to cover the desired time interval [0,t] website. Any question regarding the template or paper
increases. Note that equations are evaluated for each t in the guidelines must be directed to info@atscience.org.
interval [0,t] in all of the above mentioned algorithms.
In this study, a simple general algorithm is proposed for II. PROPERTIES OF THE TAYLOR POLYNOMIALS
state variables estimation of linear, time-invariant multi-
input multi output systems. The proposed algorithm is The Taylor series expansion of an analytic function f(t) in
based on Taylor series approximation and has an analog the neighborhood of the point t=t0 can be written as
solution. The solution that results from the proposed
algorithm gets closer to the true solution when more and f (t) f(t0 ) a1 (t-t0 ) a2 (t-t0 )2 ... ak (t-t0 )k .... (1)
more terms are kept in the Taylor series. Finally, the
____________________________________________________________________________________________________________
621
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
where
t T
k f() dt a Pλ ( t ) (12)
1 d f ( t0 ) 0
ak ( ) (2)
k! dt k where the square matrix P is called the Taylor series
operational matrix of integration, and is given by
When t0 =0, the Maclaurin Formula given in Eq. (1) is
obtained: 0 1 0 0 ...... 0 0
0 0 1/2 0 ..... 0 0
f (t) a
k0
k λk ( t ) (3)
P=
0 0 0 1/3 ..... 0 0
.... ...................................... = Prxr (13)
where 0 0 0 0 .... 1/(r-2) 0
0 0 0 0 ..... 0 1/(r-1)
λk ( t ) t
k 0 0 0 0 ..... 0 0
(4)
By truncating the series given in Eq. (3) after the (r+1) th in the unique form. It can easily be shown that k-fold
term, the approximate analytic function f(t) can be written integration of P is Pk [16, 17]. The matrix P of size rxr in
as Eq. (13) is called the integration operation
r 1 matrix.
f (t) a
k 0
k λ k ( t ) a λ (t)
T
(5)
III. THE PROPOSED ESTIMATION ALGORITHM
where a and λ (t) are the Taylor polynomial coefficient The proposed estimation algorithm can be considered as
state observer and its simulation diagram is given in Fig. 1.
vector and the Taylor series basic vector, respectively
defined as
t k 1
t 1
0 k ()d k 1 ( t ) . (9)
k 1 k 1
t r 1 t
State and error equations for the state observer shown in
f() dt
k 0
a k k (t) dt Figure 1 are
0 0
r 1
ak r
xˆ (t) Mxˆ (t) Bu(t) Gy (t) (14)
( ) k 1 ( t ) bk k ( t ) (10)
k 0 k 1 k 1 e (t) Me(t), e(0) x(0) xˆ (0) (15)
where bk, k=1,2,…, r are real numbers. From the last where
relation the following equalities can easily be obtained M (A GC) (16)
b1=a0, b2=(a1)/2 ,…., br=(ar-1)/r. (11)
In Eq. (14) and (15) , x̂(t) , u(t) , e(t) and y(t) are the nx1
Thus, substituting Eq. (11) in Eq. (10) leads to estimation vector, mx1 input vector nx1 error vector and
____________________________________________________________________________________________________________
622
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
px1 output vector , respectively. A, B, C and G are nxn u10 u11 ... u1,r-1
state matrix, nxm input matrix, pxn output matrix and nxp u
gain matrix, respectively. The gain matrix G is effective 20 x 21 ... u 2,r-1
only if x(0) x̂(0) and it should be chosen such that the u (t)= . . . λ (t) Uλ (t) (23)
estimation error goes to zero in a short period of time. The ...
elements of the gain matrix G can be determined form the . . .
characteristic equation given in (17) by using arbitrary u m0 u m1 ... u m,r-1
eigenvalues of (A-GC) denoted α1 ,α 2 ,....,α n (note that
eigenvalues are chosen such that e(t) goes to zero as where
quickly as possible) (Kailath,1980): U [u0 u1 ... u r-1 ] ,
T
αI (A GC ) = 0 (17) u k u1k u 2k ... u mk , k=0,1,2,…,r-1
we obtain
where
t t t
T
ˆ (t)dτ+ Bu (τ)dτ+ Gy (τ )dτ
ˆx (t)-ˆx (0)= Mx (18) yk y1k y 2k ... y pk , k=0,1,2,…,r-1.
0 0 0
1 n
T ˆx =( (mi,j ˆx j,k-2 ))+fi,k-1 (28)
ˆx (0) ˆx1 (0) ˆx 2 (0) ... ˆx n (0) i,k-1
k-1 j=1
i=1,2,…,n , k=2,3,…,r
____________________________________________________________________________________________________________
623
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. NUMERICAL APPLICATIONS Y=[ 0.5 -0.5 0.7499 -0.5827 0.3089 -0.1184 0.02664]
0 1 0
x (t)
2 3 x (t) 1 u(t)
y (t) 1 0 x(t)
y(t)=0,0021x6-0,0123x5+0,0347x4-,0585x3+0,043x2
+0.0487x +0,3838, for 1≤t<2
y(t)=0,0021x6-0,0013x5+0,0029x4-0,0008x3
-0,0155x2+0,0493x+0,4415, for 2≤t<3
y(t)=-0,0002x4+0,0025x3-0,0099x2+0,0224x+0,4764,for
3≤t<4
____________________________________________________________________________________________________________
624
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
-1 -1 1 1
x (t)
2 0 x(t) 1 0 u(t)
y (t) 1 2 x (t)
____________________________________________________________________________________________________________
625
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
626
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
____________________________________________________________________________________________________________
627
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— the requirement of power is very much essential for 1) Determination of short-circuit duties on switching
the development of any country. The simulation and short circuit devices, that is, high-, medium-, and low-voltage
study of an electrical power system is performed on Electrical circuit breakers and fuses
Transient Analyzer Program (ETAP). The output of short circuit 2) Calculation of short-circuit currents required for
studies are helpful in order to determine system configuration,
protective relaying and coordination of protective
protection equipment, system voltage levels, cables size , and
switchgears, transformers and grounding. ETAP short circuit devices
analysis software analyzes the effect of 3-phase, 1-phase, line-to- 3) Evaluations of adequacy of short-circuit withstand
ground, line-to-line, and line-to-line-to-ground fault currents on ratings of static equipment like cables, conductors, bus
electrical power systems. In this research, selected a city where a bars, reactors, and transformers
national network connected with four distributed generation. 4) Calculations of fault voltage dips and their time-
This distributed generation was operated depending upon solar dependent recovery profiles [2].
energy, wind power, waterpower and natural gas. In the normal In the power stations and the industrial facilities and the
case, the national network operates in parallel with the solar public buildings the designers take in consideration the safety
station and the wind station to feed all loads in the city. But, in
procedures to maintain the equipment’s and the labour, and
the case of national network outage from city, feed only task
loads in the city in island mode through these four distributed provide the systems high reliability in the control of systems
generation. With priority given to the work of the stations, the parts.
stations operates at the least cost of operating condition. Short The urgency short circuits in the systems leading to big
circuit analysis can be done for the two cases mentioned above. losses also some person’s loss these lives or at less power
supplies loss.
Keywords— Short circuit, ETAP, Distributed generation, Island The short circuits caused the isolation collapse in the
mode, Grid. electrical parts and in some cases can see sparks in the short
circuits points and these sparks caused fires in the locations.
In first time of the short circuits occur the protection devices
I. INTRODUCTION will work in the various parts and these devices distributed on
the system or buildings parts.
The short circuits happened in the electrical grid by the
The late response when the short circuits appear lead to
reasons of the urgency high voltage that over from the grid.
raising risks to the systems and huge losses [3]. In the
Lot accidents appear in the general electric grid by this
electrical power systems the short circuits happened between
current for example (damage in the grid parts, firing or
the phases itself or between the phases and the ground in this
explosion in the breakers and in the lines).
paper explain the four types of short circuits as following:
One of practices to minimize the short circuits in the grid
1) Three phase faults
determine the fault region and treatment the problem without
2) Line to ground
stopping the lines and to avoid the problem development.
3) Line to line
The high sensitivity protection devices and the high
4) Line to line to ground
response from the technicians help us to maintain the power
In the Electrical Transient Analyzer Program (ETAP)
supply [1].
software determined the total short circuit currents in the
In this paper using the ETAP software to illustrate the short
generators, transformer, lines and the other parts of electrical
circuits analysis and treatment it by simulation the single line
systems.
diagram of the general grid connected with four distributed
ETAP depending on two standards the ANSI and IEC
generation (wind, hydroelectric, photovoltaic energy and
standards [4].
natural gas).
The major aims from applied the short circuits to select the
appropriate equipment’s to avoid occur the accidents in the
equipment’s and the employees with continuity power supply.
And it collected in:
____________________________________________________________________________________________________________
628
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 3 Three Phase Short Circuit (island mode) Fig. 5 Line to Ground Short Circuit (island mode)
The Line to line short circuit analysis shown in figure 4. The Line to line to ground short circuit analysis shown in
figure 6.
Fig. 4 Line to Line Short Circuit (island mode) Fig. 6 Line to Line to Ground Short Circuit (island mode)
The Line to ground short circuit analysis shown in figure 5.
____________________________________________________________________________________________________________
630
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Definition of Terms This is the rms value of the short-circuit current which
IEC standards use the following definitions, which are remains after the decay of the transient phenomena
relevant in the calculations and outputs of PowerStation.
REFERENCES
[1] Saravanan, C. and K. Sathiyasekar, IMPACT OF DISTRIBUTED
GENERATION ON LOSS, VOLTAGE PROFILE, EQUIPMENT
LOADING AND SHORT CIRCUIT LEVEL WITH SVC BY USING
ETAP. Int J Adv Engg Tech/Vol. VII/Issue II/April-June, 2016.
782: p. 788.
[2] Das, J., Power system analysis: short-circuit load flow and
harmonics. 2016: CRC press.
[3] IEEE, Recommended Practice for Calculating AC Short-Circuit
Currents in Industrial and Commercial Power Systems. 2006,
Published by the Institute of Electrical and Electronics Engineers,
Inc.: New York, NY 10016-5997, USA.
[4] Hairi, M., et al., An Investigation of Short Circuit Analysis in
Komag Sarawak Operations (KSO) Factory. World Academy of
Science, Engineering and Technology, International Journal of
Electrical, Computer, Energetic, Electronic and Communication
Engineering, 2009. 3(11): p. 1946-1951.
[5] Operation Technology, I., ETAP 12.6, User Guide , Chapter 9-
One-Line Diagram. 2014: Southern California.
[6] Khan, R.A.J., M. Junaid, and M.M. Asgher. Analyses and
monitoring of 132 kV grid using ETAP software. in Electrical and
Electronics Engineering, 2009. ELECO 2009. International
Conference on. 2009. IEEE.
[7] Operation Technology, I., ETAP 12.6, User Guide , Chapter 15-
Short-Circuit Analysis. 2014: Southern California.
[8] Pujhari, T., Islanding detection in distributed generation, in
Department of Electrical Engineering. 2009, National Institute of
Technology Rourkela: Rourkela.
____________________________________________________________________________________________________________
631
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The geometrical dependence of metallic nano-patterns antenna arrays operating in the mid-infrared regime. The
provides adjustable resonance frequencies. Especially, aperture- transmittance spectra of the structure are investigated by using
based nanostructures exhibit resonant behaviors in transmission the finite-difference time-domain (FDTD) method [7]. We
phenomena with easily accessible ultra-high-field localization obtain the field distributions at the resonant wavelength,
characteristics and aperture dimensions influence the spectral
which indicates the physical origin of the resonant modes. We
response of the structure. We report the transmission and
reflection characteristics of fourfold rectangular shaped analyze the parameters that can enable fine control of the
nanoaperture antennas that exhibit high near field distributions resonance frequencies of the stucture. High SEIRA (surface
in the mid-infrared regime. Numerical analyses are carried out enhanced infrared absorption) enhancement [8] at resonant
by using finite-difference-time-domain method to analyze the wavelength and rezonant behavior in mid-infrared regime
transmittance and reflectance spectra of the structure. We ensure that the proposed antenna can be utilized for infrared
investigate the geometrical parameters that can enable fine detection applications.
control of the resonance frequencies and tunable optical
characteristics are concluded by the parameter sweeps. High
SEIRA (surface enhanced infrared absorption) enhancement at II. NUMERICAL ANALYSES
resonant wavelength and resonant behavior in mid-infrared
regime ensure that the proposed antenna can be utilized for
infrared detection applications.
I. INTRODUCTION
Dependence on the geometrical parameters of metallic Fig. 1 The schematic view of the unit cell: (a) Top view (aperture width W,
nano-patterns provides adjustable resonance frequencies as aperture length H, the width the central square aperture C, the geometrical
observed from the theoretical and experimental studies. interference of the rectangles with central square S, periodicities Px and Py).
and (b) cross-section view.
Recently, particle- and aperture-based nanoantennas with
different geometries have been studied to capture, control and
Fig. 1 shows a schematic structure of the unit cell of the
manipulate the light at the preferred wavelengths [1-6].
proposed nanoaperture antenna array. In this figure, W is
Especially, aperture-based nanostructures has been the subject
rectangular aperture width, H is rectangular aperture length, C
of several researches due to their resonant behavior in
is the width the central square aperture, S is the geometrical
transmission phenomena with easily accessible ultra-high-
interference of the rectangles with central square, and Px and
field localization characteristics and it has been showed
Py are periodicities of the structure. During the simulations,
aperture dimensions influence the spectral response of the
the unit cell of the proposed resonator antenna array is
structure [2-6].
modeled on a free-standing 100-nm-thick silicon nitride
In this study, we present the transmission characteristics of
(SiNx) substrate. For the metal overlay, a 30-nm-thick Au
a fourfold rotationally symmetric rectangular nanoaperture
layer and a 5-nm-thick Ti adheson layer are added onto the
____________________________________________________________________________________________________________
632
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
dielectric substrate. Excitation the surface plasmon modes on interference of the rectangles with central square S ensures
the resonator by Ex-polarized normally incident only a significant frequency shift (Fig. 2d). Increasing the
electromagnetic field is simulated. The dielectric constants of rectangular aperture width W leads to frequency red-shift as
the metals used for simulation are taken from Ref. [9]. observed from Fig. 2e.
Periodic boundary conditions are choosen as x- and y-axes,
and perfectly matched layers are used along the z-axis. Unless
otherwise indicated, the analyses are executed under the
x-polarization and geometrical parameters are chosen as
C = W = 200 nm, H=800 nm, P = Px = Py = 2000 nm, and
S = 50 nm.
Fig. 3 Reflectance spectra of the structure (a) C variation, (b) H variation, (c)
P variation, (d) S variation, and (e) W variation.
____________________________________________________________________________________________________________
633
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] E. Aslan and M. Turkmen, “Novel dual-band resonator nanoantenna
array for infrared detection applications,” Sensors and Materials, vol.
Fig. 4 Distributions of SEIRA enhancements |E|2/|Eint|2, (a) at air - metal
25, no. 9, pp. 689-696, 2013.
interface (z = 30 nm) from the horizontal field monitor, and (b) y = 50 nm and
[2] M. Turkmen, “Characterization of x-shaped nano-aperture antenna
(c) y = 100 nm from the vertical field monitors (C = W = 200 nm, H=800 nm,
arrays operating in mid-infrared regime,” Ch. Opt. Let., vol. 11,
S = 50 nm and P = 2000 nm).
070501, 2013.
[3] E. Aslan and M. Turkmen, “Refractive index sensing characteristics of
dual resonances in rectangular fractal nano-apertures,” Opt. Mater., vol.
III. CONCLUSIONS 46, pp. 1-8, 2015.
[4] A.E. Cetin, S. Kaya, A. Mertiri, E. Aslan, S. Erramilli, H. Altug, and M.
In conclusion, we present the transmission and reflection Turkmen, “Dual-band plasmonic resonator based on Jerusalem cross-
characteristics of a resonator plasmonic nano-antenna based shaped nanoapertures, ” Phot. Nano. Fund. Appl., vol. 15, pp. 73-80,
2015.
on fourfold rotationally symmetric rectangular shaped [5] A.E. Cetin, M. Turkmen, S. Aksu, D. Etezadi, H. Altug, “Multi-
nanoaperture array which can be useful for infrared detection resonant compact nanoaperture with accessible large nearfields, ” Appl.
applications. We determined the behavior of the spectral Phys. B: Lasers Opt., vol. 118, pp. 29-38, 2015.
response towards geometrical parameter variations. We [6] A.E. Cetin, S. Aksu, M. Turkmen, D. Etezadi, and H. Altug,
“Theoretical and experimental analysis of subwavelength bowtie-
revealed the physical provenance of the resonant behavior by shaped antennas,” J Electromagnet Wave, vol. 2, pp. 1-8, 2015.
SEIRA enhancements distribution analysis. Such [7] (2015) Lumerical FDTDTM website. [Online]. Available:
nanoapertures with tunable spectral response can be capable https://www.lumerical.com/
for comprehensive consequences for infrared detection [8] L. V. Brown, X. Yang, K. Zhao, B. Y. Zheng, P. Nordlander, and N.J.
Halas, “Fan-shaped gold nanoantennas above reflective substrates for
applications. The proposed nanoaperture design can support a surface-enhanced infrared absorption (SEIRA),” Nano Lett., vol. 15, pp.
high quality resonant mode with strong near-field 1272-1280, 2015.
enhancement. [9] E. D. Palik, Handbook of Optical Constants of Solids, Orlando:
Academic, 1985.
____________________________________________________________________________________________________________
634
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Channel estimation algorithms can be grouped into obtained than pilot based channel estimation because of less
three categories as pilot based, blind and semi-blind. In pilot based pilot usage.
channel estimation, some of data symbols are used to estimate
channel. In blind channel estimation statistical properties of
channel are used. In semi-blind channel estimation information II. MIMO-OFDM CHANNEL
from both data symbols and statistical properties is utilized. In this
study, pilot based and semi-blind channel estimation are used to 2x2 MIMO-OFDM communication system using Alamouti
estimate the channels with various frequency and time selectivity illustrated in Fig. 1.
in multiple input multiple output-orthogonal frequency division
Channel
multiplexing (MIMO-OFDM) systems. Semi-blind channel Estimation
estimation is done by using independent component analysis
(ICA). Simulation results show that if the channel to be estimated
Alamouti Decoding
is frequency selective, semi-blind channel estimation technique Alamouti Coding
____________________________________________________________________________________________________________
635
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
x*0 (k)y11 (k) x1 (k)y 21 (k) 2) Whitening: A simple method to perform the whitening
Ĥ11 (k) (6) transformation is to use the eigenvalue decomposition
x 0 (k) x1 (k)
2 2
Ĥ12 (k) 0
(8)
x 0 (k) x1 (k)
2 2
the diagonal matrix of eigenvalues. The observation vector can
x (k)y12 (k) x 0 (k)y 22 (k)
* be whitened by the following transformation:
Ĥ 22 (k) 1
(9)
x 0 (k) x1 (k)
2 2
y w VD 1/ 2 V T y (13)
V. RESULTS
frequency
y Hx (10)
____________________________________________________________________________________________________________
636
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
637
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Elazığ, TURKEY
{oatila, ksengur}@firat.edu.tr
Abstract—Automatic Modulation Classification (AMC) is accordingly and then the data can be extracted. AMC,
carried out as a basic structure between signal detection and however, can recognize the modulation types without need to
demodulation types. Detection of modulation type of a received this prior information.
signal is a challenging task in communication environment.
Recently, AMC has gained a paramount importance especially in In a communication system, transmitted signal is exposed
cognitive radio applications. Most of the AMC methods assume to the disruptive effects of the communication channel. One of
that additive white Gaussian noise contaminates the received the disruptive effects is noise in the environment. Most of the
signal. There might be degradation in the performance of the modulation classification algorithms are implemented under
traditional modulation classification methods in the impulsive additive white Gaussian noise (AWGN) condition for the
noise condition. In this study, firstly the digital modulation performance analysis. However, noise types in the
signals are transformed into t-f domain by Short Time Fourier
communication channel occur as a mix of AWGN and
Transform (STFT) and then by the Spectrograms of STFT are
used to obtain the t-f images of digital modulation signals. We
impulsive noise characteristic [8]. The impulsive noise occurs
then use Gray-Level Co-occurrence Matrix (GLCM) method as various natural and man-made sources caused by
which includes some statistical texture recognition techniques as electromagnetic interference such as automotive ignitions,
a feature based AMC. Finally, we use the Support Vector neon lights and etc. in communication channels and this
Machines (SVMs) classification technique for classifying the situation decreases the performance of the classifiers used for
digital modulation types. We use 0, 5 and 10 dB noise levels, AMC [5].
respectively. We implement the proposed scheme on MATLAB.
QAM, 16-QAM, 32-QAM, 64-QAM, BPSK and 8-PSK signals There are usually two methods called the statistical texture
are considered in the experiments. The method we propose recognition and the decision theoretic approach for grouping
achieves the classification with having very little performance the modulation classification schemes. In order to formulate
loss in impulsive noise condition compared to the Gaussian noise the classification problem, probabilistic and hypothesis testing
condition.
arguments are used in decision theoretic approach and with
this way the classification rule can be formed. For it keeps the
Keywords—Automatic Modulation Classification, Statistical
error rate classification in a minimum level, generally the
Texture Recognition, Support Vector Machines.
solution is optimal. Need for prior information can be
I. INTRODUCTION accepted as the disadvantage of this approach [6], [8].
With the rapid development of wireless communication Texture recognition methods, however, do not need such
technologies such as cognitive radio, spectrum surveillance, careful treatment, although choosing the right features set is
threat evaluation, interference identification and etc. use of still an issue. This classification method consists of two steps:
automatic modulation classification (AMC) has gained much the feature extraction subsystem and recognition subsystem.
attention for both civil and military applications [1], [3]. AMC The feature extraction subsystem works as extracting
provides detection of the correct modulation type from the distinctive characteristics that are called features from the
received waveforms having different distinctive receiver signal. Example of features used are higher-order
characteristics [9]. In a general communication system, cumulates, signal spectral is prepared from wavelet transform,
receiver knows the prior information including modulation Fourier transform, constellation shape, signal peaks, power
type, carrier frequency, bandwidth, symbol rate, carrier phase, moments etc. [2], [4], [6]- [8].
and etc. therefore, the received signal can be demodulated
____________________________________________________________________________________________________________
638
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
classifying the received signal based on the features extracted. X (n, ) x[m]w[n m]e
m
j n
(4)
It can be applied in many ways such as maximum likelihood
classifier (MLC), K-nearest neighborhood classifier (KNN), The discrete STFT is defined as;
artificial neural network (ANN), SVM, Genetic Algorithm
(GA) etc. [7], [10]- [12]. X (n, k ) X (n, ) 2 k
N (5)
In this study we propose feature based automatic
modulation classifier using SVM for classification of where the window function w[m] centered at time n is
modulation BPSK, 8-PSK, QAM, 16-QAM,32-QAM in the multiplied with the signal x[m] before the Fourier transform.
mix of Gaussian noise and Impulsive noise environment. We The window function is viewing the signal just close to the
use time-frequency texture feature based on the presented time n and the Fourier transform will be an estimate locally
GLCM. We also present probability of correct classification around n. The usual way of finding the STFT is to use a fixed
for different SNR value between -5 dB to 20 dB. positive even window, w[m], of a certain shape, which is
centered around zero and has unity power. Similar to the
This paper is organized as follows: we focus on signal ordinary Fourier transform and spectrum we can formulate the
models and problem statement for digital modulation signals spectrogram as;
in Section II. STFT which transformed into t-f domain is
S(n, k ) X (n, )
2
explained in Section III. Feature extraction based on GLCM
including specific distinctive statistic information is presented (6)
in Section IV. SVM for digital modulation classification is which is used very frequently for analyzing time-varying
explained in Section V. Simulation result is represented in and non-stationary signals.
Section VI. Finally, conclusions are drawn in Section VII.
IV. FEATURE EXTRACTION
II. SIGNAL MODELS AND PROBLEM STEMENT
GLCM is known as a popular texture encoder where
In digital communication, a modulated signal can be directional patterns with a specific distance and angle between
generally represented as; neighboring image pixel pairs are countered [15]. In other
𝑠(𝑡) = 𝐴𝑚 𝐺𝑇 cos(2𝜋𝑓𝑚 (𝑡) + 𝜑𝑚 (𝑡)) , (1) words, GLCM is a square matrix where the number of rows
and columns is equal to the number of gray levels of input
where Am, 𝐺𝑇 , fm and 𝜑m are the message amplitude, pulse image [13].
shaping function, message frequency and message phase, Instead of using the GLCM directly, 20 popular statistical
respectively. features are extracted from GLCM of input images. These
The received signal can be represented as; features are autocorrelation, contrast, correlation, cluster
prominence, cluster shade, dissimilarity, energy, entropy,
𝑟(𝑡) = 𝑠(𝑡) + 𝑛(𝑡) , (2) homogeneity, maximum probability, sum of squares variance,
sum average, sum variance, sum entropy, difference variance,
received signal is sum of the transmitted signals and noise
difference entropy, information measure of correlation,
signals. where s(t), r(t), n(t) transmitted signal, received
inverse difference is homogeneity, inverse difference
signals and noise signal respectively.
normalized, Inverse difference moment normalized,
The noise signal can be represented as; respectively.
noise signal n(t) in our noise signal model occurs noise signal The SVM is a machine-learning algorithm, which uses a
flexible representation of the class boundaries in order to
w(t) is the additive white Gaussian noise and i(t) is the
impulsive noise added. The AMC algorithm needs to achieve solve the classification problems [14]. Given a set of training
samples, each marked samples for belonging to one of two
the classification without being effected by the noise so that
the signal coming to the receiver can recognize the classes, a SVM training algorithm builds a model that assigns
modulation type correctly. new examples into one class or the other. SVM model is a
representation of the samples as points in space, mapped so
III. Spectrogram that the samples of the separate classes are divided by a gap
that is as wide as possible. New examples are then mapped
The discrete time STFT of x[m] is defined as; into that same space and predicted to belong to a class based
on which side of the gap they fall on [12].
____________________________________________________________________________________________________________
639
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Parameters Values
Sampling Frequency, fs 1Mhz
Carrier frequency, fm 2Khz
Baud Rate, Rb 1000 baud
No symbol, Ns 64/128/256 symbols
After constructing the gray scale sub-images for digital Fig. 2. (a) transmitter 32-QAM modulated signal (b) noise signals occurred by
modulation signals, the texture descriptors were computed. mix of Gaussian noise in 20 dB and impulsive noise (c) received modulated
signal which is corrupted by noise signal.
For computing the GLCM, the distance parameter set to 1 and
the angle parameter set to the values ranging from 0 o to 135o
within a 45o increment. Thus, 4 GLCMs were obtained and by
computing the contrast, correlation, energy and homogeneity
features, a 20-dimensional feature vector was constructed for
each sub-image. After concatenation procedure, 80-
dimensional feature vector was obtained for each digital
modulation signals. For SVM classifier, we experimented
with all kernels, and the best result was obtained with radial
basis function kernel. The C parameter was set to 10000.
____________________________________________________________________________________________________________
640
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE III. CONFUSION MATRIX FOR SVM CLASSIFIER IN GAUSSIAN NOISE CONDITION
TABLE IV. CONFUSION MATRIX FOR SVM CLASSIFIER IN MIX OF GAUSSIAN AND IMPULSIVE NOISE CONDITIONS
VII.CONCLUSIONS [3] Adjemov, S. S., Klenov, N. V., Tereshonok, M. V., & Chirov, D.
S. "Methods for the automatic recognition of digital modulation of
In this paper, we supervised signal feature based signals in cognitive radio systems," Moscow University Physics
Bulletin, vol.70.6, p. 448-456, 2015.
automatic modulation classification by using SVM classifier [4] Ho, K. C., W. Prokopiw, and Y_T Chan. "Modulation identification
under mix of Gaussian noise and impulsive noise condition. of digital signals by the wavelet transform." IEE Proceedings-Radar,
Sonar and Navigation vol.147.4, p.169-176, 2000.
For different modulation techniques including BPSK, 8-PSK,
[5] Blackard, K. L., Rappaport, T. S., & Bostian, C. W. “Measurements
QAM, 16-QAM, 32-QAM,64-QAM, the simulation results and models of radio frequency impulsive noise for indoor wireless
and confusion matrix are presented under different SNR communications,” IEEE Journal on selected areas in
communications, 1993, vol.11(7), p.991-1001
values and magnitudes of impulse noises. We propose a [6] Wong, ML Dennis, and Asoke K. Nandi. "Automatic digital
method which achieves the classification in an accuracy modulation recognition using artificial neural network and genetic
similar to the performance in the Gaussian noise conditions algorithm." Signal Processing, vol. 84.2, p. 351-365, 2004.
[7] Hazar, M. A., Odabaşioğlu, N., Ensari, T., & Kavurucu, Y.
without being effected by the impulsive noise. The AMC “Evaluation of Machine Learning Algorithms for Automatic
method we propose has an approximately 1% lower Modulation Recognition,” In: International Conference on Neural
performance in impulsive noise condition compared to the Information Processing, Springer International Publishing,2015, p.
208-215.
Gaussian noise condition. The proposed system provides a [8] Mohanty, M., Satija, U., Ramkumar, B., & Manikandan, M. S.
classification accuracy of 83 percent at 0 dB SNR under mix ”Digital modulation classification under non-Gaussian noise using
of Gaussian and Impulsive noise conditions. We have also sparse signal decomposition and maximum likelihood,” In:
Communications (NCC), 2015 Twenty First National Conference
observed that when we use more symbol number for on. IEEE, 2015. p. 1-6.
classification, we have better results. [9] Hazza, A., Shoaib, M., Alshebeili, S. A., & Fahad, A.”An overview
of feature-based methods for digital modulation classification,” In:
We will investigate how the performance of AMC in Communications, Signal Processing, and their Applications
(ICCSPA), 2013 1st International Conference on.IEEE, 2013. p. 1-6.
frequency selective channels can be improved in our further [10] AlOtaiby, T. N., Shoaib, M., Saleh, A., & Hazza, A. “Support
studies. vector machine based classifier for digital modulations in presence
of HF noise,” In: Electronics, Communications and Photonics
Conference (SIECPC), 2013 Saudi International. IEEE, 2013. p. 1-4.
REFERENCES [11] Haq, K. N., Mansour, A., & Nordholm, S. “Classification of digital
modulated signals based on time frequency representation,” In:
[1] D. L. Guen, A. Mansour, “Automatic Recognition Algorithm for Signal Processing and Communication Systems (ICSPCS), 2010 4th
Digitally Modulated Signals”, In: International Conference on Signal International Conference on. IEEE, 2010. p. 1-5.
Processing, Pattern Recognition, and Applications, Crete, 2002, p. [12] Mustafa, H., & Doroslovacki, M. “Digital modulation recognition
25-28. using support vector machine classifie,”. In: Signals, Systems and
[2] M.L.D. Wong, A.K. Nandi, “Automatic digital modulation Computers, Conference Record of the Thirty-Eighth Asilomar
recognition using spectral and statistical features with multi-layer Conference on. IEEE, 2004. p. 2238-2242.
perceptrons,” in: Proceedings of International Symposium of Signal [13] Bajaj, V., Guo, Y., Sengur, A., Siuly, S., & Alcin, O. F. “A hybrid
Processing and Application (ISSPA), Kuala Lumpur, Vol. II, 2001, method based on time–frequency images for classification of alcohol
p. 390–393. and control EEG signals,” Neural Computing and Applications,
2016, p.1-7.
____________________________________________________________________________________________________________
641
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
642
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
*
Department of Electronic and Automation, Vocational School of Technical Sciences, Gaziantep,
Gaziantep University, Gaziantep, Turkey
tokur@gantep.edu.tr
+
Department of Electrical and Electronic Engineering, Faculty of Engineering, Erciyes University, Kayseri, Turkey
taspinar@erciyes.edu.tr
Abstract— Partial Transmit Sequence (PTS) scheme is an effective these methods, the PTS is one of the popular distortionless
peak-to-average power ratio (PAPR) lessening tool for techniques and provides remarkable PAPR reduction
Orthogonal Frequency Division Multiplexing (OFDM) system. performance. However, the search and combination of phase
However, computational complexity for the optimum phase factors which requires an exhaustive search procedure is the
factors searches of PTS scheme entails huge computational
major drawback of PTS technique. The replacement of
requirements and limits its applicability to practical applications
especially for high-speed data transmissions. This study proposes optimum phase factor search steps of PTS scheme with a
a PAPR reduction method with a low computational complexity suboptimal search technique offers reduction in computational
based on a combination of Cuckoo search optimization algorithm complexity of PTS scheme. Many suboptimal PTS schemes,
with PTS scheme in OFDM system. In terms of PAPR and including the harmony search (HS) [15], parallel tabu search
computational complexity reductions, the performance of the (parallel-TS) [16], artificial bee colony (ABC) [17], di erential
Cuckoo-PTS scheme is comparatively investigated by performing evolution (DE) [18], particle swarm optimization (PSO) [19]
a set of simulations with different PTS schemes. and random search [20], have been presented in the literature
and it is reported that PAPR reductions with low computational
Keywords— PAPR, OFDM, partial transmit sequence (PTS), complexities are successfully obtained.
Cuckoo search, phase factor In this paper, a suboptimal PTS scheme based on Cuckoo
Search Optimization Algorithm is proposed for the PAPR
I. INTRODUCTION
reduction in OFDM systems. The Cuckoo Search Optimization
Over the past few decades, due to key features as, high Algorithm, proposed by Xin-she Yang and Suash Deb, is a
bandwidth e ciency, narrowband interference, efficient relatively new optimization method that mimics the breeding
implementation and robustness to frequency selective fading behavior of some cuckoo species.
[1–2], orthogonal frequency division multiplexing (OFDM)
systems have received increasing attention and become one of II. MODEL OF OFDM SYSTEM BASED ON PTS SCHEME
the promising techniques for high data rate transmissions in The OFDM system model based on PTS scheme is given in
digital communication. However, OFDM systems are Fig. 1. First, the incoming bit sequences are interleaved to
susceptible to high peak-to-average power ratio (PAPR) [5] transform burst errors into random errors. The interleaved
which causes significant reductions in radio frequency power signal is modulated with 16-QAM modulation. To recover the
amplifiers and yields high complexity in digital-to-analog and original signal at the receiver, side information is transferred.
analog-to-digital converters. High PAPR also leads to The signal is passed through a high power amplifier and then a
implementation problems for the systems with large numbers cycle prefix (CP) is applied to the signal to combat with
of subcarriers. The PAPR can be reduced with the applications intersymbol interference (ISI) induced by communication
of constitutional arrangements and restrictions which in turn channel. The CP is removed from the signal at the receiver. The
adversely affect the operation of the power amplifier and the phase of original signal is acquired with the use of phase
spectral efficiency of OFDM systems. rotation and side information following the realization of fast
Numerous methods with different advantages and Fourier transform (FFT). After 16-QAM demodulation, each
drawbacks have been appeared in the literature for the demodulated symbol is conveyed to the original place in the bit
reduction of PAPR in OFDM systems such as selected mapping sequence by means of the deinterleaver [1].
[4], clipping and filtering [5], clipping [6], partial transmit
sequence (PTS) [7-8], coding [9], tone injection [10], tone
reservation [11], peak windowing [12], interleaving [13] and
active constellation extension [14] have been appeared. Among
____________________________________________________________________________________________________________
643
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Add N-point x1
QAM x1
Interleaver IFFT PTS HPA Cyclic IFFT
Modulation Prefix
Input bit stream
Input data
Remove +
…
…
QAM Phase
Deinterleaver FFT Cyclic
Demodulation Rotate
Prefix
V. SIMULATIONS
The randomly generated OFDM signals with = 256 sub-
carriers with 16 − modulation were used in computer
simulations. The number of the population, , number of the
____________________________________________________________________________________________________________
644
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
generation, number of sub-blocks, , and the number of the 10 for = 25 , = 50 , = 75 , and = 100,
phase factor, , was selected as 20, 50, 16, 4 respectively. respectively.
The complementary cumulative density functions (CCDFs) of
10 0
the PAPR were obtained with the use of 10000 random OFDM G=25
symbol generations. The solid state power amplifier (SSPA) G=50
G=75
with input back-off factor, = 9 , and smoothness factor, G=100
= 2 was used as an amplifier. The additive white Gaussian
noise (AWGN) channel was chosen as communication channel. 10 -1
10 -1
10 -2
Figure 5. Comparison of the PAPR0 (dB) versus CCDF for CS-PTS
with different values for = 4, = 16, = 20, and = 25.
VI. CONCLUSION
10 -3
4 5 6 7 8 9 10 11 12 13 14 15 In this paper, PTS based on cuckoo search optimization
PAPR 0 [dB]
algorithm in OFDM system is proposed for the PAPR reduction
Figure 3. Comparison of the PAPR0 (dB) versus CCDF of the original, with less computational load. In order to evaluate the PAPR
CS-PTS, RS-PTS, PSO-PTS and optimum-PTS of the PAPR0 (dB)
reduction performance of the proposed CS-PTS scheme, the
versus CCDF of the OFDM system.
CCDF simulations are performed. Also, its performance is
compared with the performances of original PTS, O-PTS, PSO-
Figure 4, the variations of CCDFs of PAPR for different
PTS and RS-PTS in the OFDM systems. Simulation results
number of generations, , with = 20 are shown.
show that the PAPR reduction performance of the CS-PTS in
According to Figure 4, = 25 yields the worst performance
the OFDM system is better than that of O-PTS, PSO-PTS and
while = 100 yields the best performance for OFDM. The
RS-PTS in the OFDM system.
PAPR is 7.32 , 7.10 , 7.00 and 6.86 at =
____________________________________________________________________________________________________________
645
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES [12] G. Chen, R. Ansari, and Y. Yingwei, “Improved peak windowing for
PAPR reduction in OFDM”, In: IEEE Vehicular Technology Conference,
[1] S.H. Müller, and J.B. Huber JB. “OFDM with reduced peak-to-average 26-29 April 2009, Barcelona, Spain: pp. 1-5.
power ratio by optimum combination of partial transmit sequences”, [13] A.D.S. Jayalath, and C. Tellambura, “Reducing the peak-to- average
Electron Lett, vol. 33, pp. 368–369, 1997. power ratio of an OFDM signal through bit or symbol interleaving”,
[2] A. Chini, and Y. Wu, M. El-Tanny, and S. Mahmoud, “Hardware Electron Lett., vol. 36, pp. 1161-1163, 2000.
nonlinearity’s in digital TV broadcasting using OFDM modulation”, [14] B.S. Krongold, and D.L. Jones, “PAR reduction in OFDM via active
IEEE T Broadcast, vol. 44, pp. 12–21, 1998. constellation extension”, IEEE T Broadcast, vol. 49, pp. 258-268, 2003.
[3] T. Jiang, and Y. Wu, “An overview: peak-to-average power ratio [15] E.M. Kermani, H. Salehinejad, and S. Talebi, “PAPR reduction of
reduction techniques for OFDM signals”, IEEE T Broadcast, vol. 54, pp. OFDM signals using harmony search algorithm”, In: 18th International
257-268, 2008. Conference on Telecommunications, 8-11 May 2011, Ayia Napa,
[4] Y. Jie, C. Lei, L. Quan, C. De, “A modified selected mapping technique Cyprus, pp 90-94.
to reduce the peak-to-average power ratio of OFDM signal”, IEEE T [16] N. Taşpınar, A. Kalınlı, M. Yıldırım, “Partial transmit sequences for
Consum Electr., vol. 53, pp. 846-851, 2007. PAPR reduction using parallel tabu search algorithm in OFDM systems”,
[5] X. Li, and L.J. Cimini, “Effect of clipping and filtering on the IEEE Commun Lett., vol. 15, pp. 974-976, 2011.
performance of OFDM”, IEEE Commun Lett, vol. 2, pp. 131-133, 1998. [17] N. Taşpınar, D. Karaboğa, M. Yıldırım, and B. Akay, “PAPR reduction
[6] J. Wang, Y. Guo, and X. Zhou, “PTS clipping method to reduce the using artificial bee colony algorithm in OFDM systems”, Turk J Elec
PAPR in ROF-OFDM system”, IEEE T Consum Electr., vol. 55, pp. Eng& Comp Sci., vol. 19, pp. 47-58, 2011.
356-359, 2009. [18] H.L. Hung, and Y.F. Huang, “Peak-to-average power ratio reduction in
[7] Y Zhao, T Wang, R Tao, P Shi, ZG Jiang, “Peak to average power ratio orthogonal frequency division multiplexing system using differential
reduction with low computational complexity in FrFT-OFDM system”, evolution-based partial transmit sequences scheme”, Commun IET, vol.
J Elec & Info Tech, vol. 36, pp. 246-249, 2014. 6, pp. 1483-1488, 2012.
[8] C. Ye, Z. Li, T. Jiang, C. Ni, and Q. Qi, “PAPR reduction of OQAM- [19] J. Gao, J. Wang, and B. Wang, “Improved particle swarm optimization
OFDM signals using segmental PTS scheme with low complexity”, for PAPR reduction of OFDM systems”, In: International Conference
IEEE T Broadcast, vol. 60, pp. 141-147, 2014. on Networking, Sensing and Control; 2010; Chicago, IL: pp. 621-624.
[9] A.E. Jones, T.A. Wilkinson, and S.K. Barton, “Block coding scheme for [20] L.J. Jr Cimini, Sollenberger NR. “Peak-to-average power ratio reduction
reduction of peak to mean envelope power ratio of multicarrier of an OFDM signal using partial transmits sequences”, IEEE Commun
transmission scheme”, Electron Lett., vol. 30, pp. 2098-99, 1994. Lett; vol. 4, pp. 86-88, 2000.
[10] J.C. Chen, and C.K. Wen, “PAPR reduction of OFDM signals using [21] X. Yang, and S. Deb, “Cuckoo Search via Lévy flights”, Proc. of World
cross-entropy- based tone injection schemes”, IEEE Signal Proc Let., Congress on Nature & Biologically Inspired Computing, 9-11 Dec 2011,
vol. 17, pp. 727-730, 2010. pp. 210-214.
[11] J.C. Chen, and C.P. Li, “Tone reservation using near- optimal peak
reduction tone set selection algorithm for PAPR reduction in OFDM
systems”, IEEE Signal Proc Let.; vol. 17, pp. 933-936, 2010.
____________________________________________________________________________________________________________
646
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Recently, utilization of mobile robots has increased and a software framework for fast validation of concepts. The
substantially. Accordingly, wireless communication is preferred existing robotic platforms are presented for different
in mobile robots. QBot 2 is an autonomous ground robot which is applications [8-9]. The significance of developed software
a new product of Quarc. The QBot 2 utilizes an onboard data algorithms on computers is the increase in robotic control.
acquisition card and a wireless embedded computer to measure
Quanser has developed Quarc real time control software for
the onboard sensors and drive motors. In this study, QBot 2
mobile robot is evaluated in terms of its ability to maneuver. In MATLAB® Simulink [10].
this way, controlling QBot 2 mobile robot in real time is planned The QBot 2 is ideally suited for research applications such
while a command recognition system is developed. Connection as differential drive kinematics, forward and inverse
with the QBot 2 is carried out in a wireless environment.
A Simulink model is developed in MATLAB® environment. The
kinematics, dead reckoning and odometric localization, path
created model is built with Quarc control software. Compiled planning and obstacle avoidance, 2D mapping and occupancy
model is downloaded with TCP/IP connection to QBot 2 and the grid map, image acquisition, processing and reasoning,
application is carried out on an embedded computer. The QBot 2 simultaneous localization and mapping (SLAM), high level
mobile platform consists of two central drive wheels mounted on control architecture of mobile robots and, vision-guided
a common axis. This drive configuration is known as differential vehicle control. Applications of QBot which is previous
drive. The two drive wheels are independently driven forward generation of Quanser ground robots are mentioned in the
and backward in order to actuate the robot. Motion of the wheels literature [11-13].
is realized using high performance DC motors. When the results
are analyzed, 13 different motion types are observed in total. The The QBot 2 mobile platform consists of two central drive
observed motion types could be used as references in future wheels mounted on a common axis that bisects the robot. The
works since many practical applications, such as the remote two drive wheels are independently driven forward and
control of QBot 2 mobile robot via the human voice, require the backward in order to actuate the robot [14]. This drive
availability of different motion types.
configuration is known as differential drive [15]. There are
different applications of wheeled mobile robots [16].
Keywords—Intelligent robotic systems, Mobile robots; Wi-Fi
based remote control. In this study, QBot 2 mobile robot is evaluated to explore
its ability to maneuver. In this way, controlling QBot 2 mobile
I. INTRODUCTION robot in real time is planned while a command recognition
system is developed.
The first major mobile robot development effort was
Shakey, developed in the late 1960s [1]. Recently, utilization Interacting with mobile robots using speech is of particular
of mobile robots has increased substantially. Mobile robot interest to researchers since verbal communication is the most
systems are designed for research, education, commercial, natural way of communication for human beings. Speech has
industrial, space missions or different purpose. Today been previously used to perform different tasks, such as
researchers are focusing on mobile robots owing to their command based voice teleoperation of a mobile robot [17-18].
potential applications in hazardous environments [2], house- The paper is organized as follows. Section II. introduces
hold tasks [3], in the field of agriculture [4], medical [5], and how to be teleoperated mobil robot QBot 2. Then, differential
military applications [6-7]. drive kinematics is described in Section III. The simulation
The basic technologies involved in these applications are a results and analysis are described in details in Section IV.
suitable robotics platform equipped with enhanced sensors Finally, Section V. concludes the paper.
____________________________________________________________________________________________________________
647
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
II. THE TELEOPERATED MOBILE ROBOT Q BOT 2 hardware interface for the QBot 2. ‘’HIL Read/Write’’ block
set is utilized to read from sensors and/or write to outputs.
The QBot 2 utilizes an onboard data acquisition card and a
‘’Kinect Initialize’’ block set is used to initialize the Kinect
wireless embedded computer to measure the onboard sensor
sensor. ‘’Kinect Get Image and Depth Sensor’’ blocks are
and drive motors. Connection with the QBot 2 is carried out in
used captures RGB and depth data from the Kinect sensor.
a wireless environment A Simulink model is developed in
‘’Display Image’’ block set is used to transmit the input data
MATLAB® environment. The created model is built with
(RGB or depth) from QBot 2 to the PC and displays them on
Quarc control software. Compiled model is downloaded with
the monitor. Other than these blocks, the ‘’Host Initialize’’
TCP/IP connection to QBot 2 and the application is carried out
block can be used to make use of external input devices such
on an embedded computer.
as a keyboard or joystick.
A. The Hardware Platform III. TWO WHEELED DIFFERENTIAL ROBOT MODEL
The QBot 2 is an autonomous ground robot which is a new The QBot 2 mobile platform consists of two central drive
product of Quanser. The QBot 2 is comprised of a Yujin Robot wheels mounted on a common axis. The two drive wheels are
Kobuki platform, a Microsoft Kinect RGB camera and depth independently driven forward and backward in order to actuate
sensor and a, data acquisition card (DAQ) with a wireless the robot. Motion of the wheels is realized using high
embedded computer. The embedded computer system mounted performance DC motors. This drive configuration is known as
on the mobile robot uses the Gumstix DuoVero [19] to run differential drive [22-23]. Modeling of the QBot 2 is shown in
QUARC and interface with the QBot 2 data acquisition card Fig. 2 (a) and top view of the QBot 2 is shown in Fig. 2 (b).
[20].
B. Establishing Wireless Connection
Between the host computer and the QBot 2 connection can
be over wired or wireless communications. In this study, the
QBot 2 communicates with the host computer by utilizing an
ad-hoc peer-to-peer wireless TCP/IP connection. The network
established is called GSAH. The teleoperated mobile robot
QBot 2 is shown in Fig.1.
Fig. 2. Left: Illustration of the QBot 2; Right: The QBot 2– top view
____________________________________________________________________________________________________________
648
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
𝐴 𝐴 Rotation
-A 0 (− ) CW (− ) around
2 𝐿 left wheel
Turning
𝐴−𝐵 𝐴+𝐵 around
-A B (− ) CW (− )
2 𝐿 right
wheel
𝐴 𝐴 Rotation
When –A, –B, 0, B, A speed values are set to the left and A 0 ( ) CCW ( ) around
right wheels in accordance with the equality 0 ≤ B ≤ A ≤ 0,65 2 𝐿 left wheel
m⁄s, 13 different motion types are observed in total. Observed
Rotation
maneuvers information of the QBot 2 is presented in Table I. 𝐴 𝐴 around
When both wheels rotate at the same speed, we observe linear 0 -A (− ) CCW ( )
2 𝐿 right
motion in a straight line as, shown in the first two rows in wheel
Table I. When both wheels rotate at the same speed but in the Turning
opposite direction, we observe rotation about the midpoint of 𝐴−𝐵 𝐴+𝐵
A -B ( ) CCW ( ) around
the wheel axis and rotate in place, as shown in the third and 2 𝐿 left wheel
fourth rows in Table I. When only one wheel speed of the
QBot 2 is set, we observe rotation around the non-moving 𝐴+𝐵 𝐴−𝐵 Left arc
A B ( ) CCW ( )
wheel as shown in the fifth, sixth, ninth and tenth rows in 2 𝐿 turning
Table I. When a wheels turns faster than the other, QBot 2
turns in an arc towards the slower wheel as shown in the 0 0 0 0 0
Initial
seventh, eighth, eleventh and twelfth rows in Table I. When ω Position
has a positive value (ω > 0), it creates a counter clockwise
____________________________________________________________________________________________________________
649
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. CONCLUSION [7] Hill,Susan G., Daniel Barber, and Arthur W. Evans III. "Achieving the
vision of effective Soldier-robot teaming: Recent work in multimodal
In this study, QBot 2 mobile robot maneuvers were communication."Proceedings of the Tenth Annual ACM/IEEE
analyzed. Determined left and right wheel speed values are International Conference on Human-Robot Interaction Extended
entered with a model created in MATLAB Simulink using Abstracts.ACM, 2015.
[8] http://www.irobot.com
QUARC control software. When the results are analyzed, 13 [9] http://www.mobilerobots.com
different motion types are observed in total. [10] Quarc, http://www.quanser.com/quarc.
The difference between the calculated velocities and the [11] Luo, Jing, Cheng-Lin Liu, and Fei Liu. "A leader-following formation
measured ones as a result of measurement error are caused by control of multiple mobile robots with obstacle." Information and
Automation, 2015 IEEE International Conference on. IEEE, 2015.
load, friction and disturbances to the wheels. [12] Yu, Ningbo, et al. "A haptic shared control approach to teleoperation of
The observed motion types could be used as references in mobile robots." Cyber Technology in Automation, Control, and
future works since many practical applications, such as the Intelligent Systems (CYBER), 2015 IEEE International Conference on.
remote control of QBot 2 mobile robot via the human voice, IEEE, 2015.
[13] Sharifi, Farid, et al. "A distributed deployment strategy for a network of
require the availability of different motion types. cooperative autonomous vehicles." Control Systems Technology, IEEE
Transactions on 23.2 (2015): 737-745.
REFERENCES
[14] J. C. Alexander and J. H. Maddocks, “On the kinematics of wheeled
[1] Nilsson, Nils J. A mobile automaton: An application of artificial mobile robots,” Int. J. of Robotics Research, vol. 8, no. 5, pp. 15–27,
intelligence techniques. SRI INTERNATIONAL MENLO PARK CA 1989.
ARTIFICIAL INTELLIGENCE CENTER, 1969. [15] Quanser Inc., QBOT 2 Workbook – Student, 2015
[2] G. Kantor, S. Singh1, R. Peterson, D. Rus, A. Das,V. Kumar, G. Pereira, [16] M. Hadi Amoozgar, "Development of Fault Diagnosis and Fault Tolerant
and J. Spletzer, “Distributed search and rescue with robot and sensor Control Algorithms with Application to Unmanned Systems", Concordia
teams,” in Field and Service Robotics. 2006, vol. 24, pp. 529– University, Canada, 2012.
538,Springer Berlin / Heidelberg. [17] Shrivastava, Kunal, et al. "A speech recognition client-server model for
[3] M. Hans, B. Graf, and R.D. Schraft, “Robotic home assistant care-Obot: control of multiple robots." Proceedings of Conference on Advances In
Past-present-future,” in Proc. IEEE International Conference on Robot Robotics. ACM, 2013.
and Human Interactive Communication, 2001, [18] Poncela, Alberto, and Leticia Gallardo-Estrella. "Command-based voice
[4] B. Astrand and A.J. Baerveldt, “An agricultural mobile robot with vision- teleoperation of a mobile robot via a human-robot
based perception for mechanical weed control,” Autonomous Robots, interface." Robotica 33.01 (2015): 1-18.
vol. 13, no. 1, pp. 21–35, July 2002. [19] Gumstix, http://www.gumstix.com.
[5] J. Eriksson, M.J. Mataric, and C.J.Winstein, “Hands-off assistive robotics [20] Quanser Inc.,QBOT 2 - User Manual, 2015
for post-stroke arm rehabilitation,” in Proc. IEEE International [21] Quanser Inc., Q BALL 2 Workbook, 2015
Conference on Rehabilitation Robotics, June 2005, pp. 21–24. [22] R. Siegwart, I.R. Nourbakhsh, D. Scaramuzza. 2011. Introduction to
[6] ALPER, İrfan H., “Ulusal Tank ve Zırhlı Araç Tasarımlarına Yönelik Autonomus Mobile Robots. MIT Press, Cambridge, Massachusetts, USA.
Silah Sistemleri Mühendisliği Yaklaşımları”, SAVTEK 2006, 3. [23] Gregory Dudek and Michael Jenkin. 2000. Computational Principles of
Savunma Teknolojileri Kongresi, Orta Doğu Teknik Üniversitesi, Mobile Robotics. Cambridge University Press, New York, NY, USA
Ankara, Haziran 2006, pp. 137-143
____________________________________________________________________________________________________________
650
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Use of landmines in soils is a significant a variety of different fields in industry such as nondestructive
international threat facing the world today. There are no safe testing, condition monitoring, and predictive maintenance,
and highly reliable methods or inspection systems capable of reducing energy costs of processes and buildings, detection of
detecting landmines in all situations. The use of infrared gaseous species, and many more [1]. IR thermal imaging has
thermography is one of the promising methods for mine detection
also been widely used for landmine detection.
tasks. In infrared thermography, the investigation is done in
either way: actively or passively. In this study, thermal signatures The detection of landmines and clearance is still a time
of the active infrared time difference images of buried mines and consuming and unsafe task. Additionally, efficient and
sand are investigated for different camera angles. It is aimed to accurate detection of buried mines is still a challenging
find the effects of shot angles of the thermal camera on the
problem [2].
performance of landmine detection. The experiments are
performed at a sandbox emplaced in an indoor laboratory There is no universal technique capable of detecting
environment. A metal and a plastic antipersonnel mine are landmines in all conditions. Infrared thermography is a
buried at 2 cm depth in sandbox. The sand surface is initially
promising technique in the detection and discrimination of the
heated homogeneously by an infrared heater (2400 W) for 10
minutes on different days. During the cooling phase of the
landmines. The detection principle is based on the variation of
surface, a sequence of images are captured with an LWIR (8-12 the ground temperature due to the presence of buried or
µm band) camera (FLIR T 650 SC), which is 280 cm away from surface landmines.
the detection area at different angles (90°, 60° and -60°). Images
Sand and mines have different thermal properties and this
of the size of 480×640 pixels are taken at 15 seconds intervals
during one hour. “Thermal signatures” of the buried mines and difference can be observed on the surface through the thermal
soil in three viewing angles are compared in MATLAB® sensors. Since thermal property is a dynamic behavior driven
environment. The results show that the locations of landmines by radiation from the external heating system, it can change
are easily detected from the captured images during the cooling with the temperature in a few minutes and it can be observed
phase of the surface since observable differences develop between better in thermal image time series than a single image.
temperature signatures of landmines and sand, but the
observation angle of camera has little affects on the detection Active thermograpy technique can be applied to enhance
performance. In addition, it is found that one hour measurement the contrast between the possible targets and the background.
period is adequate for the detection of landmines at 2 cm depth in This contrast stems from the difference in radiant
active thermography. characteristics between the landmines and the background
soil.
Keywords— Landmine detection; Thermal infrared imagery;
Active thermography. Emissivity value, which is one of these characteristics,
plays a significant role in the determination of correct
I. INTRODUCTION temperature of an object surface [3]. The source of uncertainty
Infrared (IR) thermal imaging, also often briefly called in temperature measurement with the infrared camera can be
thermography, is a very rapidly evolving field in science as listed as followings: emissivity (ℰ) of objects, infrared
well as industry owing to the enormous progress made in the detector-to-surface angle and distance. Those are mainstream
last two decades in micro system technologies of IR detector technical factors for an accurate infrared thermographic
design, electronics, and computer science. Thermography measurement.
nowadays is applied in research and development as well as in
____________________________________________________________________________________________________________
651
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Some authors [4] [5] [6] have used analytical, numerical III. EXPERIMENTS
and experimental methods to model the thermal signatures of
The equipment required for the experiments consists of
land mines. But, in our humble opinion, how angle of
model anti-personal landmines (plastic and metal), thermal
observation influences the results has not been investigated in
camera, a tripod of camera stand or framework, sandbox and a
detail yet.
heating unit.
Therefore, the main objective of this paper is to evaluate
The landmines can be categorized mainly into two sections
the effect of change in camera angle on detection performance
according to their materials, metal or plastic mines. The
of buried landmines.
surrogate model mine used in the experimental study is Plastic
DM-11 (partly filled with wax whose thermal properties are
similar to TNT) and Metal M-16 anti-personnel mines
II. BACKGROUND
(APM’s) as shown in Fig. 1. DM-11 has a diameter of 0.08 m
A. Mine Problem and a height of 0.035 m. M-16 has a diameter of 0.12 m and a
height of 0.20 m.
Landmines are explosive devices hidden just below the
surface designed to be detonated by contact of people or The mines were buried in a sandbox which has a height of
vehicles, as they pass over or near them. Still some 60 1.85 m length, a width of 1.55 m and a depth of 0.225 m, filled
countries around the world are contaminated by landmines and with sand. The thermographic imaging was performed with a
thousands of people continue living with a risk of losing their portable infrared camera (FLIR T 650 SC) equipped with an
life or limb. In addition, emplaced landmines hinder the uncooled micro bolometer, a focal plane array infrared
cultivation of large productive areas. Besides, they maintain a detector with a spectral range between 7.5 – 13.0 μm and
sense of insecurity long after conflicts end, delay peace 480×640 pixels.
processes and impede countries’ development for years [7]. The camera was attached to a framework, consisting of a
Because of these horrendous effects, a lot of research and fixed attachment point for the camera. The sandbox was
technological developments are needed to solve the detection located at the base of the framework and the camera was
and clearing of landmines problem. One widely searched focused on the center of the sandbox.
solution for this problem is IR thermal imaging.
B. Thermal imaging
Each material shows a characteristic thermal response to a
given stimulus, also known as the thermal signature. Thus, the
cooling or heating process affects buried objects and the
surrounding soil in a different way. This difference is due to
the fact that the mines are better insulators than the soil. The
general concept of using infrared thermography for mine
detection is based on the fact that mines may have different
thermal properties from the surrounding material [8]. Thermal
imaging devices measure the emissivity of surfaces in an area
at various temperature ranges.
Uniquely, IR can work in either way, actively or passively.
It can work by accepting only the natural radiation from the
object called as passive thermography, or it can provide an
extra heat source and receive the artificial radiation created by
that heat source dubbed as active thermography [9].
Fig. 1. Experimental setup for an assessment of angular variation of
Most thermal detection concepts involve single snapshot of emissivity. The FLIR T 650 SC was rotated about the sample and fixed at 30-
the region of interest. The soil over a mine has different degree intervals. Plastic APM DM 11 ( up) and Metal APM M 16 (down)
thermal dynamics than homogeneous soil and, as a result, a Positions.
time sequence of images can often produce better detection Infrared Heater (UFO-L2300:2400 Watt Power) was
than a single image [10]. used as a heat source. The heating phase (active
Consequently, the soil temperature on the ground above thermography) is shown in Fig. 2.
the mines is often different from that of the background. This
temperature can be measured by an IR imaging system placed
above the soil area.
____________________________________________________________________________________________________________
652
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. RESULTS
a) Viewing at 90°.
Three types of experiments performed at the viewing
angles of 90°, 60° and -60° as it is shown in Fig.3 during three
days. The acquired thermal images are stored in a personal
computer. The model mines were buried into a depth of 0.02
m beneath the sand surface.
b) Viewing at 60°.
Fig. 3. An IR image sequence of a minefield; images taken at (a) 90°, (b) 60°
and (c) -60°.For images obtained at the same distance(2.80-m )
____________________________________________________________________________________________________________
653
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
b) Viewing at 60°.
V.CONCLUSION
In our study the effect of viewing angle is investigated by
combining the analysis of temporal IR image sequences,
showing the dynamic scene behavior during time variant
heating by an infrared heater. The results show that different
viewing angles (for 90°, 60° and -60°) have relatively little
effect on the temperature at the surface. The emissivity is
approximately constant at the viewing angles between 60° and
-60°.
Experiments have shown that active thermography has
strong positive effect on thermal signature in a short time. It
can be especially useful in military tasks.
____________________________________________________________________________________________________________
654
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
VI. REFERENCES
[1] M. Vollmer and K.-P. Möllmann, Infrared Thermal Imaging
Fundamentals, Research and Applications, Weinheim,GERMANY:
WILEY-VCH Verlag GmbH & Co., 2010.
[2] A. Ajlouni and A. Sheta, "A Novel Landmine Detection Process Using
KL Transform and Marker based Watershed Segmentation in IR
Images," International Journal of Signal and Imaging Systems
Engineering, 2010.
[3] T. Nunak, K. Rakrueangdet, N. Nunak and T. Suesut, "Thermal Image
Resolution On Angular Emissivity Measurements using Infrared
Thermography," Proceedings of the International MultiConference of
Engineers and Computer Scientists 2015 Vol I,, 2015.
[4] J. Deans, J. Gerhard and L. Carter, "Analysis of a thermal imaging
method for landmine detection,using infrared heating of the sand
surface," Infrared Physics & Technology , vol. 48 , p. 202–216, 2006.
[5] R. L. Dam, B. Borchers, J. M. Hendrickx and S. Hong, "Soil effects on
thermal signatures of buried nonmetallic landmines," Detection and
Remediation Technologies for Mines and Minelike Targets VIII, vol.
5089, 2003.
[6] K. P. ,. Lamorski, P., W. Swiderski, D. Szabra, R. Walczak and B.
Usowicz, "Thermal signatures of land mines buried in mineral and
organic soils––modelling and experiments," Infrared Physics &
Technology, vol. 43, p. 303–309, 2002.
[7] "International Campaign to Ban Landmines," the ICBL , [Online].
Available: http://www.icbl.org/en-gb/problem/why-landmines-are-still-
a-problem.aspx. [Accessed 2016].
[8] C. P. Lee, Landmine detection techniques using multiple sensors, M. Sc.
Thesis, Electrical, 2004.
[9] L. Kempen, A. Katarzin, Y. Pizurion, C. Corneli and H. Sahli, "Digital
Signal/Image Processing for Mine Detection," in Proceedings Euro
Conference on Sensor Systems and Signal Processing Techniques
applied to the Detection of Mines and Unexploded Ordnance, 1999.
[10] J. MacDonald, J. Lockwood, J. McFee and T. Altshuler,
"Infrared/Hyperspectral Methods," in Alternatives for Landmine
Detection, Rand, 2003, pp. 93-110.
____________________________________________________________________________________________________________
655
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—With the rapid development of industrial technology, such as energy, water, gas, cement, climatization,
industrial automation systems will provide safe, quality, fast and transportation, security and banking [3]. Usage of PLC and
competitive product manufacturing. Programmable Logic SCADA systems has many advantages. For example, quality
Controller (PLC) is preferred in industrial automation systems increases, process monitoring is easier, failure caused by
due to its features such as process speed, quality and reliability,
human factor and occupational accidents can be minimized [4].
small physical size and easy installation. As PLC systems
develop, SCADA systems are began to use in a wide area. In this study, prototype and application of mortar mixing
Industrial plants can be monitored remotely can be controlled by automation system which is based on PLC and SCADA was
time and past data can be stored using SCADA systems with realized. The system was controlled by the S7-200 PLC and
PLC. WinTr SCADA interface. PLC and SCADA interface has
In this study, prototype and implementation of SCADA based worked simultaneously. Thus, system can be monitored and
mortar mixing automation system with PLC was carried out. controlled in real time.
The system was controlled by S7-200 PLC and WinTr SCADA
interface. All parameters can be entered, monitored and II. PROTOTYPE DESIGN OF SYSTEM
controlled by means of SCADA interface. In the system, mortar
mixture were comprised of cement, aggregate, gravel and water. A. Plc Properties used in the System
Each material was located in separate hoppers. When material
values for mixture were entered, hoppers cover was opened In this study, Siemens S7-200 CPU 224 was used as PLC.
respectively. The amount of entered value and weighed value S7-200 includes a compact microprocessor to create a
were compared and when their values were equal, hopper covers powerful Micro PLC, integrated power supply, input and
were closed thanks to loadcells located at the bottom of the output circuits. CPU 224 model has 14 inputs and 10 outputs
hoppers. Materials were transferred to mixing hopper by means [5, 6]. The block diagram of system is shown in Figure 1.
of band system after sand, aggregate and gravel were added.
Then, water was pumped from the water hopper to the mixing
hopper and all materials were mixed by mixer.
Keywords— PLC, SCADA, mortar mixing automation
I. INTRODUCTION
Industrial automation systems will provide qualified, safe,
fast and competitive way for product manufacturing with the
rapid development of industrial technology. With the use of
automation systems, production desired and planned in plants
can be made and also systems can be monitored remotely.
Furthermore, automation technology also provides fast and
reliable production [1].
Mortar mixed systems work under the control software
because they are completely subject to the automation system.
Nowadays, Programmable Logic Controller (PLC) is used in
almost every area of the industry but it is used especially in
automation systems. PLC has a wide range of application in Figure1. Block diagram of system
all areas of industry such as air conditioning, elevator,
ventilation, cooling, packaging, storage and transportation B. Design of System
plants, automobile industry, petroleum filling and washing Prototype of the mortar mixing automation system is shown
plants, cement industry, lighting plants, all kinds of machines, in Figure 2 and prototype material list is shown in Table I.
electropneumatic and hydraulic systems [2].
SCADA systems are used in wide area as long as PLC
systems evolve. They enable to remote communication. Thus,
they provide easier intervention to long distance system.
Nowadays, SCADA systems are widely used in various fields
____________________________________________________________________________________________________________
656
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
657
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
III. CONCLUSION
Nowadays, automation systems are used for safe, quality
and fast industrial production with the rapid development of
industrial technology. PLC systems are widely used in
automation systems. Usage of SCADA systems has increased
with the expansion of the application areas of PLC. Industrial
plants can be remotely monitored, controlled using SCADA
with the PLC. Moreover, historical data of plants can be
stored.
In this study, prototype and application of mortar mixing
automation systems were realized. The system was controlled
by a PLC and SCADA interface. Furthermore, the system can
be remotely monitored and can be controlled with SCADA
interface.
REFERENCES
[1] A. Lakshmi, B. Sangeetha, A. Naveenkumar, B. Ganesh, and N.
Bharathi, “Experimental validation of PID based cascade control
system through SCADA-PLC-OPC interface,” In Computer
Communication and Informatics (ICCCI), International Conference
on IEEE, 2012, pp. 1-4.
[2] O. Bingöl, T. Aydoğan, H.R. Dıdın, and A.S. Yalçiner, “PLC
Kontrollü Otomatik Katlı Otopark Sistemi,” SDU International
Journal of Technological Science, vol. 2, pp. 65-76, Feb. 2010.
[3] R. Bayındır, O. Kaplan, C. Bayyiğit, Y. Sarıkaya, and M. Hallaçlıoğlu,
“PLC ve SCADA kullanılarak bir endüstriyel sistemin otomasyonu,”
Erciyes Üniversitesi Fen Bilimleri Enstitüsü Dergisi, vol.27, pp.107-
Figure5. Flowchart of the program 115, Jan. 2011.
[4] O. Bilgin, Y. Altun, and M. Mutluer, “Remote monitoring and
E. Expression of the Program with Ladder Diagram diagnostic system of PLC controlled an elevator using SCADA,” In
Electrical, Electronics and Computer Engineering (ELECO), 2010, pp.
Addresses of the input and output terminal of the PLC are 212-216.
[5] S7-200 Programmable Controller System Manual, Siemens, 2005.
shown in Table III. [6] T.Aydoğan, M.E.Özgün and S.Karaahmetoğlu, “SIEMENS SIMATIC
S7-200 CPU 224 Model PLC Kontrollü Asansör", SDU International
Technologic Science, vol.1, pp. 1-9, 2009.
____________________________________________________________________________________________________________
658
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this paper, real time application of a sliding mode controllers have been developed to apply the systems that
control (SMC) is used for level control of experimental setup of have nonlinear system dynamics. As an example, the
liquid level system due to its properties such as robustness backstepping controller has been presented to control the
against large parameter variation and disturbances rejection. A systems where system dynamics can be written in a lower
well-tuned conventional proportional integral (PI) controller is
triangular form [4-7]. Also, it has a recursive structure and its
also applied to the two coupled tank system for comparison with
the SMC controller. Experimentation of the coupled tank system stability analysis can be proven based on Lyapunov stability
is realized in two different configurations, namely configuration theorem, the backstepping controller mostly preferred. On the
#1 and configuration #2 respectively. In configuration #1, the other hand, the all states of the system must be measured and
water level in the top tank is controlled by a pump. In in complex systems, more mathematical processes are needed
configuration #2, the water level in the bottom tank is controlled to calculate the control input signal. Moreover, fuzzy
by the water flow coming out of the top tank. The performance controller is proposed where the system is not well defined
of controllers is analyzed according to their tracking and does not have an exact mathematical model [8]. Apart
performance and error elimination capability for different from these advantages, the fuzzy controller is not good
references applied to the system. Experimental results prove that
enough to stabilize the system where the quick response is
the SMC shows better trajectory tracking performance than PI
controller in that the plant transient responses to the desired needed and the controller’s sensivity need to be very high.
output changes have shorter settling time and smaller magnitude Also, it is a rule based controller and there is no theorem to
overshot/undershoot. Robustness of the SMC with respect to proof its stability criteria. As an another solution, SMC is
water level variation and capability to eliminate external proposed to control nonlinear systems. The SMC is a type of
disturbances are also achieved. nonlinear controller and preferred for the systems where the
controller must be very robust to model uncertainties and
Keywords— Liquid Level Control, Sliding Mode Control external disturbances [9-15]. In addition to this advantages
mentioned on SMC, it has a simple structure and its stability
I. INTRODUCTION criteria can be proven using Lyapunov stability theorem.
In this study, the liquid level control of coupled tank system
Liquid level control is one of the essential problems of
is examined in experimental application using PI and SMC
process industry. In nuclear power plants, wastewater
controllers for different reference inputs. Both controllers are
treatment plants, filling-disposal plants etc., high sensitive
compared in term of error elimination capability and
liquid level systems must be used to do complex and
robustness against to parameter uncertainty. The results prove
dangerous processes that are not done by humans. To achieve
that the SMC shows better trajectory tracking performance
correct mixture ratio, adjust liquid flow between tanks
than PI controller. Also, the results show that SMC is really
precisely and to get high level performance from this systems,
robust to model uncertainties occur in the system and good at
well-designed controllers must be also used. Due to their
eliminating the errors compared with the PI controller.
nonlinear system dynamics and can be effected disturbances
easily, the control of liquid level systems are difficult. To II. MODELLING AND CONTROL OF THE COUPLED
overcome these difficulties and control this kind of systems in TANK SYSTEM
a desired efficiency, many control algorithms have been
proposed by the researchers. Due to their simple structures, A. Single Tank Model (Configuration #1)
easy parameter adjustment and basic calculation knowledge Single tank system which consisting of the top tank is
needed P, PI, PID controllers are used to control liquid level shown in Fig. 1. It is reminded that in configuration #1, the
of the tanks [1-3]. In the experimental application, classical P, pump feeds into tank 1 and that tank 2 is not considered at all.
PI, PID controllers are not sufficient for a precise control Therefore, the input to the process is the voltage to the pump
where sudden changes occur, parameter adjustment and fast and its output is the water level in tank 1. The model of the
response are needed. For these reasons, conventional P, PI and single tank system determined by relating the volumetric
PID controllers are not a good choice for a nonlinear system. inflow rate into tank and the outflow rate leaving
In control engineering and relevant areas, some nonlinear through the hole at the tank bottom.
____________________________________________________________________________________________________________
659
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
660
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
y z1 (22) If we take the derivative of both sides of the Eq. (32) the
following equation obtained.
The dynamics model of the coupled tank system is nonlinear.
Therefore, in order to define the dynamics of the coupled tank s e1 e1 (35)
system, the dynamic model subject to a transformation
Eq. (35) is employed as the expression written instead of the
equation as follows.
second order derivative of the error, following equations can
x1 be written,
x , and x T (z ) if we make the transformation,
x2 s e1 ( x1r x1 ) (36)
____________________________________________________________________________________________________________
661
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
s e e e (58)
k1 k 2 k (44) dt
If writing to these assumptions in place of the coupled tank Where n is the order of the system to be controlled, is a
system dynamic model equations are obtained from the positive constant, e is the error. Error we take as a reference-
following equation. measured, and we have obtained the following equation.
z 2 k z 2 u (45) e2 L2 r L2 (59)
Thus, the dynamics of the combined form of coupled tank Ideally, we want to sliding surface to be equal to zero. If
system can be written as follows, sliding surface is zero than derivative of the sliding surface
will be zero as following equations.
x1 x2 (50)
0 e2 ( x2 r x2 ) (63)
k k
x 2 z 2 z1 (51) 0 e2 f ( x) ( x)u eq (64)
2 z2 2 z1
If u eq in Eq. (64) left alone than Eq. (65) can be obtained.
k k
x 2
2 z2
k z 2 u 2 z1
k
z 2 k z1 (52)
f ( x ) e2
u eq (65)
( x) ( x)
ku k 2 z2
x 2 (53) Considering the system parameters and disruptive we can get
2 z2 2 z1 our control signal as follows.
Thus, the dynamics of the combined form of coupled tank u u eq k x sgn(s ) (66)
system can be written as follows,
____________________________________________________________________________________________________________
662
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Substituting u eq into Eq. (66) the following equation which is The experimental results given for configuration #2, are
realized using the step + trapezoidal reference signal for PI
control signal of the tank 2 can be obtained. and SMC controllers shown in Fig. 6 and Fig. 8 respectively.
In this section, the cascaded tank system is used to realize
kL1 2 L2 e level control for different reference signals. For PI controller,
u k x sgn( s ) (67)
L2 k it has longer rising and settling time. Also, PI controller has
more overshoot than SMC controller. At the same time, SMC
shows better performance for position tracking and also has
Where, k x sgn(s ) is switching control function. k x , is the
given fast response the parameter variations.
switching gain and sgn(s ) can be expressed given below [18]. 15
since as is positive.
10
V. EXPERIMENTAL RESULTS Reference
In the first experiment of configuration #1, the step+sinusoidal 5 Measured
reference signal is applied for PI and SMC controllers given in 0
Fig. 3 and Fig. 4 respectively. For the step reference signal 0 10 20 30 40 50
part, PI controller has more overshoot and has longer settling Time (sec)
time compared with the SMC. The time varying part of the 10
reference signal, both controllers show the same performance 5
in term of tracking the reference signal, under/overshooth and
0
error elimination capability etc. On the other hand, SMC is
generated the pump voltage with more chattering phenomena -5
than the pump voltage generated by PI controller. 0 10 20 30 40 50
Time (sec)
For the second experiment of configuration #1, the
step+sawtooth reference signal is applied to show the
responses and performance of both controllers shown in Fig. 5 20
and Fig. 7 respectively. The sawtooth reference signal is
10
important due to consisting of the sudden and continuously
changing part in the same period. For PI controller, it has 0
more overshoot and longer settling time compared with SMC. 0 10 20 30 40 50
Also, PI controller has continued to make overshoot through Time (sec)
the continuously changing part of the sawtooth signal. Besides, Fig. 4. SMC experimental results for step + sin reference in configuration #1
SMC has no overshoot while tracking the step+sawtooth
signal and has better performance in term of error elimination
and robustness to parameter variations occur in the system.
____________________________________________________________________________________________________________
663
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
10 10
Reference 5 Reference
5
Measured Measured
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Time (sec) Time (sec)
15 15
10 10
5 5
0 0
-5 -5
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Time (sec) Time (sec)
25 25
15 15
5 5
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Time (sec) Time (sec)
Fig. 5. PI experimental results for step + sawtooth reference in Fig. 7. SMC experimental results for step + sawtooth reference in
configuration configuration #1
#1
15 25 15 25
Reference
Tank 1 Level (cm)
Reference
Tank 2 Level(cm)
20
20
10 Measured Measured
15 10
15
10 10
5 Reference 5 Reference
Measured 5 5
Measured
0 0 0 0
0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150
Time (sec) Time (sec) Time (sec) Time (sec)
10 25 10 25
Tank 2 Level Error (cm)
20
Pump Voltage (V)
20
Pump Voltage (V)
5 15 5 15
10 10
0 5 0 5
0 0
-5 -5 -5 5
0 50 100 150 50 0 100 150 0 50 100 150 0 50 100 150
Time (sec) Time (sec) Time (sec) Time (sec)
Fig. 6. PI experimental results for step + trapezoidal reference in Fig. 8. SMC experimental results for step + trapezoidal reference in
configuration #2 configuration #2
the levels that required to be kept at a constant level and
VI. CONCLUSIONS following time varying reference signals very well.
In this study, SMC has been designed and applied to two
tank liquid level system in order to maintain a stable reference REFERENCES
tracking for different reference signals for both configurations. [1] S. Tunyasrirut, T. Suksri, A. Numsomran, S Gulpanich, K.
The experimental results indicate that the proposed controller Tirasesth, “The Auto-Tuning PID Controller for Interacting
has better performance in term of error elimination capability, Water Level Process,” Proceedings of World Academy of
good reference signal tracking, robustness to parameter Science, Engineering and Technology, vol.12,134-138, 2006.
variations. It can be also noticed from the results obtained [2] W. Gou., Research of Intelligent PID and Its Application in the
Water Level Control, Heifei: Anhui Agricultural University,
that SMC is reasonably prior to the classical PI controller for
2008.
____________________________________________________________________________________________________________
664
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[3] K. Liu., Advanced PID Control and Matlab Simulation. [11] B. Moshiri, M. Jalili-Kharaajoo, F. Besharati, "Application of
Beijing: Publishing House of Electronics Industry, 2004. Fuzzy Sliding Mode Based on Genetic Algorithms to Control
[4] H.Gouta, S.H.Said and F.M’sahli, "Model-Based Predictive of Robotic Manipulators," Emerging Technologies and Factory
and Backstepping Controllers Four-Tank System With Automation, Vol. 2, pp. 169 – 172, 16-19,2003.
Bounded Control İnputs: A Comparative Study" Journal of the [12] R .Benayache, L. Chrifi-Alaoui, P. Bussy and J.M.
Franklin Institute, vol.352, pp. 4864-4889, Nov. 2015. Castelain, “Design and Implementation of Sliding model
[5] B. Nail, B. Bekhiti, A. Hafaifa, A. Kouzou and N. Hadroug Controller with Varying Boundary Layer for a Coupled Tanks
"Feedback Linearization and Backstepping Controlllers for System,” 17th Mediterranean Conference on Cont. & Aut., pp.
Coupled Tanks," Informatics Engineering, an International 1215-1220, 2009.
Journal (IEIJ), Vol.3, No.4, December 2015. [13] N.B. Almutairi, M. Zribi, “Sliding Mode Control of Coupled
[6] R.Benayache, L. Chıfı-Alaouı, .Dovıfaaz and Pascal Bussy Tanks,” Mechatronics, pp.427-441, 2006.
"Real-time Nonlinear Adaptive backstepping Liquid Level [14] A.Levant, “Chattering Analysis,” IEEE Transactions on
Control for a State Coupled Three Tank System," in Automatic Control, Vol.55, pp. 1380-1389, 2010.
Proceedings of the European Control Conference, 2009. [15] T. Floquet, S. K. Spurgeon and C. Edwards, “An Outp
[7] H. Gouta, S. H. Said, N. Barhoumi and F. M’Sahli "Observer- Feedback Sliding Mode Control Strategy for MIMO Systems of
Based Backstepping Controller for a State-Coupled Two-Tank Arbitrary Relative Degree,” International Journal of Robust
System,"IETE Journal of Research,Vol.61:3, pp:259-268, 2015. and Nonlinear Control, Vol. 21, No. 2, 2010.
[8] A. Başçi and A. Derdiyok “Implementation of an adaptive [16] Quanser-Two Tank Manuel, 2005.
Fuzzy Compensator for coupled tank liquid level control [17] S. R. Mahapatro, “Control Algorithms for a Two Tank Liquid
system,” Measurement, vol. 91, pp. 12-18, 2016. Level System: An Experimental Study,” M. Eng. Thesis,
[9] A. Derdiyok and A. Başçi "The Application of Chattering-free National Institute of Technology, Odisha, India, 2014.
Sliding Mode Controller in Coupled Tank Liquid-level Control [18] Y. Longand, L. Li, “Fuzzy Fractional Order Sliding Mode
System," Korean J. Chem. Eng., vol. 30(3), pp.540-545, 2013. Control for Automatic Clutch of Vehicle AMT” International
[10] Ng KC, Li Y, Murray-Smith DJ, Sharman KC. "Genetic Journal of Smart Home Vol. 9, No. 2 (2015), pp. 53-68, 2015.
Algorithms Applied to Fuzzy Sliding Mode Controller
Design," Presented at First International Conference on
Genetic Algorithms in Engineering Systems: innovations and
applications, 1995.
____________________________________________________________________________________________________________
665
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The design of the system used for brushless DC logic approach for BLDC motor controller in variable speeds
(BLDC) motor control in speed and position control is difficult and the fuzzy logic tuner is used to adjust the gains of the PI
due to the non-linear structure. Therefore, the designed controller and the results obtained in the simulation study
controller is required to respond to these challenges and need showed less ripple under variation in system parameters with
high-efficiency operation. This paper presents the experimental
fast response times.Yu and Hwang have proposed an optimal
validation of a robust speed control structure of a BLDC motor
based on continuous sliding mode (CSM) and fractional sliding PID controller and controller parameters are determined by
mode (FSM) controllers. The controllers have been tested for lowlinear quadratic regulator. The successes of the proposed
and medium speed reference signals and amplitude values. Then, method were compared with conventional PID controller and
both controllers have been compared in term of tracking simulation and experimental results were given. Navidi et al.
performance and error elimination and the results have been [4] proposed a method determined by ant colony search
shown graphically. Experimental results prove that the FSM algorithm for PID controller parameters. They have
controller shows better trajectory tracking performance than demonstrated success with the simulation results of the
CSM controller with high precision as well as good robustness proposed method in improving the step response
against changes of references.
characteristics such as reducing the steady-states error, settling
and rise time, and maximum overshoot in speed control. Chen
Keywords— BLDC motor, Continuous sliding mode control, and Tang proposed a sliding mode current control scheme for
Fractional sliding control, Speed control
pulsewidth modulation (PWM) brushless dc motor drives in
I. INTRODUCTION their study [5]. In this scheme, an improved “equivalent
control” method is used and they stated that the validity of this
scheme is achieved by simulation and experimental results.
Parallel to the developments in the control areas, brushless
Moshiri et al. proposed an approach that has the merit to
DC motors are used in computers, automated office
determine the optimal structure and the inference rules of
equipment, robotic applications, electro-mechanical systems
fuzzy sliding mode controller simultaneously [6]. The success
and many precision machines. Brushless DC motors can be
of the proposed controller is provided with the simulation
controlled more simply than other direct current motors andit
results. Wang et al. offered a stable hierarchical sliding-mode
has advantages such as high torque, high efficiency and small
control method for a class of second-order under actuated
size. In addition, problems such as mechanical wear occur in
systems [7]. They consider the system as two subsystems and
the brushes and commutator by changing the position of the
defined a first-level sliding surface for each part. Also they
stator and rotor in the DC motor. And also maintenance of the
defined a second-level sliding surface for these two first-level
brush takes a long time. But instead of brushes and
sliding surfaces and simulation results have shown the success
commutator, BLDC motors use Hall Effect sensors [1]. Today,
of the proposed method and adaptive abilities for all kinds of
drivers have developed high processing ability and therefore
extraneous disturbances.
robust control of BLDC motors are successfully carried out.
The main advantages of sliding mode control are
The desired control algorithm is mathematically analysed for
demonstrated in numerous examples and simulations. The
robust operation of the controller designed in accordance with,
history of this control structure; In 1977, after V. Utkin [8]
are tested in various computer programs and R & D work
compiler work, Sliding Mode Controller - SMC and Variable
done. In the literature, various studies have been made for
Structure Control - VSC methods are widely used in control
speed control of BLDC motors. In [2], is described a fuzzy
____________________________________________________________________________________________________________
666
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
applications until today [9]-[19]. The aim of sliding mode number of turns and flux linkage/conductor, ‘ ’ is the
control methods, especially in the real dynamic system; are angular speed of the motor, ‘ ’ is mechanical angle of rotor,
designed to drive the system states onto a particular surface in ‘ ’ is electrical angle of rotor, ‘ ’ is number of pole on rotor,
the state space, named sliding surface. Once the sliding = L – M, L is the self inductance of the winding per phase,
surface is reached, sliding mode control keeps the states on M is the mutual inductance per phase. is written in the (6)
the close neighbourhood of the sliding surface. There are two and it can be rearranged in the following form for each motor;
main advantages of sliding mode control. First is that the
dynamic behaviour of the system may be tailored by the
particular choice of the sliding function. Secondly, the closed =( + ) +( + ) + ( ̇ )(8)
loop response becomes totally insensitive to some particular
uncertainties. The disadvantage of sliding mode control BLDC motor state equations are written in the following form;
method is a control signal that changes direction too much and
it is called chattering. Also chattering causes some problems ̇( ) = ( )+ ( ) (9)
in practice such as damaging the component parts of the fast
moving systems, to causes fatigue in the controlled ( )= ( ) (10)
system,significantly reduce the life of the system andto cause
unnecessary energy consumption. Various methods are where the states and input vector are chosen as;
available to reduce this negative effect of chatteringsuch as
filtering, discontinuous approach, saturation function and ( )= [ ωφ] (11)
fuzzy control. But this time the robustness of the sliding mode
control functionality is lost [12-14]. ( )= [ ] (12)
In this study, a fractional order sliding mode controller is
designed to reduce the effect of chattering and also to the system matrices are given below,
maintain the high robustness and high accuracy features
sliding mode control.
( )
II. BLDC MOTOR MODEL ⎡ − 0 0 0⎤
⎢ ⎥
The electrical and mechanical mathematical equation of ⎢ ( ) ⎥
BLDC motor can be expressed as, 0 − 0 0
⎢ ⎥
⎢ ( ) ⎥
1 1 =⎢ 0 0 − 0⎥ (13)
=− − + (1) ⎢ ⎥
⎢ ( ) ( ) ( )
− 0⎥
⎢ ⎥
1 ⎢ ⎥
= ( − − ) (2)
⎣ 0 0 0
2
0⎦
= (3)
2 1
⎡ 0 0 0 ⎤
( ) ⎢ ⎥
⎢0 1
= ( ) (4) 0 0 ⎥
⎢ ⎥
( ) =⎢ ⎥ (14)
1
⎢0 0 0 ⎥
= + + (5) ⎢ ⎥
⎢0 1⎥
0 0 −
⎢ ⎥
= + + (6) ⎣0 0 0 0 ⎦
̇
= + + | ̇ |
(7) = [0 0 0 1 0 ] (15)
____________________________________________________________________________________________________________
667
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The goal is to drive states of the system given (9),(10) in the ε̇ = −Dε = τ̇ − ξ̇ (27)
set Sdefined by;
τ̇ − ( )+ = −Dε (28)
S = {x: τ(t) − ξ(x) = ε(x, t) = 0} (16)
and the result of the short algebra can be written as
where τ(t)is the time dependent part of the sliding function,
containing reference inputs to be applied to the controller. = +( ) Dε (29)
ξ(x) denotes the state dependent part of the sliding
function, ε(x, t). The derivation of the control involves the Third, from time derivative of the sliding function
selection of a Lyapunov function V(ε) and a desired form of
derivative of the Lyapunov function such that closed-loop ε̇ = τ̇ − ( ( )+ ) (30)
system is stable. The selected Lyapunov function is [20-22]
multiplying both sides with( )
1
V= ε ε (17)
2 ( ) ε̇ = ( ) (τ̇ − ( ) − (31)
which is positive definite, and its derivative is
and by using (25)
V̇ = ε ε̇ (18)
( ) ε̇ = − (32)
The solution ε(x, t) = 0 will be stable if time derivative of the
Lyapunov function can be expressed as [22] and finally when this equation is substituted in (29) the control
is found as
V̇ = −ε Dε (19)
( )= ( )+( ) (ε̇ + Dε) (33)
where D is a positive definite matrix. Thus, the derivative of
= + ∆ , ∆→ 0
the Lyapunov function will be negative definite and this will
ensure the stability. Eq. (18) and (19) lead to
The value of the control at the instant is calculated from the
value at the time + ∆ and the weighed sum of the control
ε (Dε + ε̇ ) = 0 (20)
error ε and its time derivative. Control (33) is continuous
function everywhere except in the points of discontinuity of
A solution for this equation is
the function ( , ) . When these equations are adapted for
BLDC motor control system shown in Fig.1, the following
Dε + ε̇ = 0 (21)
equation can be written for the control loop as
The expression for derivative of the sliding function is
( )= ( )+( ) (ε ̇ + Dε ) (34)
d d d
ε= τ− ξ (22)
dt dt dt
where,
ξ = Gx(t) (23)
____________________________________________________________________________________________________________
668
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Measured
for ( − 1 < < ), (. )is the Euler’s gamma function, is
the initial time and parameter is used when the differential 0
and integral are taken.
-10
C. Fractional Order Sliding Mode Controller 0 5 10 15 20 25
In (34), if the derivative term expressed as a fractional order, Time (sec)
0.2
( )= ( )+( ) + (38) Command
V (volt)
10 Reference
Speed (rpm)
Measured
Fig. 2. Fractional sliding mode controller block diagram 0
____________________________________________________________________________________________________________
669
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Speed (rpm)
Measured
0 0
Reference
Measured
-1000 -1000
0 5 10 15 20 25 0 5 10 15 20 25
Time (sec) Time (sec)
10 10
Command
V (volt)
V (volt)
0 0
Command
-10 -10
0 5 10 15 20 25 0 5 10 15 20 25
Time (sec) Time (sec)
Fig. 5. Continuous sliding mode 1000 rpm sinus ref. Fig. 8. Fractional sliding mode 1000 rpm sinus ref.
10 10 Reference
Speed (rpm)
Speed (rpm)
Reference
Measured
Measured
0 0
-10 -10
0 5 10 15 20 25 0 5 10 15 20 25
Time (sec) Time (sec)
0.2 0,2
Command
Command
V (volt)
V (volt)
0 0
-0.2 -0,2
0 5 10 15 20 25 0 5 10 15 20 25
Time (sec) Time (sec)
Fig. 6. Continuous sliding mode 10 rpm trapezoidal ref. Fig. 9. Fractional sliding mode 10 rpm trapezoidal ref.
Reference
Measured
Measured
0 0
-1000 -1000
0 5 10 15 20 25 0 5 10 15 20 25
Time (sec) Time (sec)
10 10
Command Command
V (volt)
V (volt)
0 0
-10 -10
0 5 10 15 20 25 0 5 10 15 20 25
Time (sec) Time (sec)
Fig. 7. Continuous sliding mode 1000 rpm trapezoidal ref. Fig. 10. Fractional sliding mode 1000 rpm trapezoidal ref.
____________________________________________________________________________________________________________
670
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. CONCLUSIONS [9] Derdiyok, A., Guven, M. K., Inanc, N., Rehman, H. U., & Xu, L. “A
DSP-based indirect field oriented induction machine control by using
In this paper, an experimental study on the application of chattering-free sliding mode,” In National Aerospace and Electronics
CSM and FSM controllers to a BLDC motor under the Conference, NAECON 2000. pp. 568-573, 2000
different speed references was presented. The experimental [10] Nguyen, D., Sliding-Mode Control: Advanced Design Techniques, Phd
Thesis, University of Techology, Sydney, 1998
results show that the FSM controller shows better steady state [11] Hung, J.Y., Gao, W., ve Hung, J.C. “Variable structure control: A
performance with better rise time, smaller speed error and survey”, IEEE Transactions on Industrial Elect., Vol 40, No 1, 2–22,
having less overshoot when it compared to the responses of 1993
CSM. To conclude, the applied FSM controller results in [12] Eker, İ.,“Sliding Mode Control with PID Sliding Surface and
Experimental Application to An Electromechanical Plant”, ISA
better responses than CSM controller to control the speed of Transactions,vol.45,pp.109-118,Number 1, January 2006
the BLDC motor under changing references. [13] Özdal, O.,“Model Dayanaklı Kayan Kipli Denetim”, Master Thesis,
Hacettepe Üniversitesi FBE, Ankara ,2008.
REFERENCES [14] Kızmaz, H., “Asılı Sarkacın Kayma Kipli Kontrolü”, Master Thesis,
[1] Allan R. Hambley, Electrical Engineering: Principles and Application, SAU FBE, Sakarya , June 2009
Prentice Hall, New Jersey 1997. [15] R. Benayache, L. Chrifi-Alaoui, P. Bussy and J.M. Castelain, “Design
[2] Lee, C. K., and W. H. Pang. "A brushless DC motor speed control and implementation of sliding mode controller with varying boundary
system using fuzzy rules," Power Electronics and Variable-Speed layer for a coupled tanks system”. 17thMediterranean Conference on
Drives. Fifth International Conference on. IET, pp. 101-106, 1994 Cont. & Aut., p:1215-1220, 2009.
[3] Yu, G-R and Hwang, R-C. “Optimal PID speed control of brush less [16] A.Levant, “Chattering Analysis,” IEEE Transactions on Automatic
DC motors using LQR approach,” In: Systems, Man and Cybernetics, Control,Vol.55, pp. 1380-1389, 2010.
2004 IEEE International Conference on. IEEE, 2004. p. 473-478. [17] T. Floquet, S. K. Spurgeon and C. Edwards, “An Output Feedback
[4] Navidi, N., M. Bavafa, and S. Hesami. "A new approach for designing Sliding Mode Control Strategy for MIMO Systems of Arbitrary
of PID controller for a linear brushless DC motor with using ant colony Relative Degree,” International Journal of Robust and Nonlinear
search algorithm," 2009 Asia-Pacific Power and Energy Engineering Control, Vol. 21, No. 2, 2010
Conference. IEEE, pp.1-5, 2009. [18] Derdiyok, A. and Başçi, A. , “The application of chattering-free sliding
[5] Chen, Jessen, and Pei-Chong Tang. "A sliding mode current control mode controller in coupled tank liquid-level control system,” Korean
scheme for PWM brushless DC motor drives," IEEE transactions on Journal of Chemical Engineering, 30(3), pp.540-545, 2013
Power Electronics14.3 pp. 541-551, 1999 [19] Soysal, Birol. "Real-time control of an automated guided vehicle using
[6] B. Moshiri, M. Jalili-Kharaajoo, F. Besharati, "Application of fuzzy a continuous mode of sliding mode control." Turkish Journal of
sliding mode based on genetic algorithms to control of robotic Electrical Engineering & Computer Sciences 22.5, pp.1298-1306, 2014
manipulators", Emerging Technologies and Factory Automation, Vol. [20] K. Jezernik, M. Rodic, R. Safaric and B. Curk, B., “Neural network
2, pp. 169 – 172, 2003 sliding mode robot control,” Robotica, 15(1), pp. 23-30, 1997
[7] Wang, W., et al. "Design of a stable sliding-mode controller for a class [21] A. Sabanovic, K. Jezernik and K. Wada,” Chattering-free sliding
of second-order underactuated systems." IEE Proceedings-Control modes in robotic manipulators control.” Robotica, 14, 17, 1996.
Theory and Applications 151.6, pp. 683-690, 2004 [22] A. Derdiyok and M. Levent, "Sliding mode control of a bioreactor."
[8] Vadim, I. Utkin. "Survey paper variable structure systems with sliding Korean J. Chem. Eng., 17(6), 619 ,2000.
modes,"IEEE Transactions on Automatic control 22.2, pp. 212-222, [23] I. Podlubny, Fractional Differential Equations, Academic Press, San
1977 Diego, California, 1999.
[24] I. Podlubny, “Fractional-order systems and PI D controllers”, IEEE
Transactions on Automatic Control, vol. 44(1), pp. 208–214, 1999
____________________________________________________________________________________________________________
671
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— The purpose of this study is to design a 3 axis It is clear that there are high expectations as to the future
automated programming machine for programming devices such potential of robotics, even euphoric ones and somewhat
as microcontrollers, EPROMs and CPLDs. While programming unrealistically utopic [3]. On the other side, there are sceptical
numerous devices, an automated system is needed. For his views, seeing robotics as one of the most powerful
purpose, a prototype system is designed with 3 axis. The system
technologies of the 21st century, together with genetic
consists of electronics, mechanics and vacuum parts. All of these
are controlled by ATMEL’s ATMEGA128 8 bit microcontroller engineering and nanotech [4] threatening to make humans an
using C language. In this study, microcontrollers are endangered species. A more moderate and realistic, but still
programmed which have 44 pins with TQFP package. With some fascinating approach has been taken by a study group,
modifications on hardware and software, the designed machine is comprising of experts from engineering, medical,
able to program the devices which have different packages. philosophical and legal sciences, discussing the provoking
Thanks to the user interface software, programming results question whether humans could be substituted by robots [5].
which of the devices are programmed and which are not) can be
seen on PC.
II. LITERATURE REVIEW
Keywords— Microcontroller, Robotic, Automated programming The work shows the application of the manipulator to a
printed-circuit-board driller system. The manipulator uses DC
I. INTRODUCTION motors with permanent magnets. The design of the
manipulator is based on microcontrollers to control the speed
Robotics is a science that combines the technology and and position of the motors by using full bridge DC-DC
knowledge. Owing to the fast development of technology, the converters with a pulse-width-modulation technique. The
term robot, as an automatic machine that replaces humans, is designed manipulator includes a PC that works as a user
not so clearly defined anymore. A robot is no longer just a interface to load the file with the numeric control format for
humanoid robot, robotic hand at an auto assembly line, the driller. From this information, the PC generates the
autopilot in aircraft, artificial intelligence built of living required position and speed commands for each
neurons or simple cleaning robot; it is also computer software servomechanism. The works describes in detail the software
that completes works. meant for humans. It is known that and hardware designed to measure and control the speed and
robots are built to replace humans at certain works [1]. the position of each servomechanism [6].
The word “robot” was created in the twenties by Czech poet, This work focuses on hardware and software interfacing for
Karel Capek, in one of his plays, a play that ended tragically. robot arm controller application. The aim of this study is to
In the forties, another writer, Isaac Asimov, made robots the design and build a control system for position control for
leading figures in his utopian novels. Since these times, robots robot arm with an FPGA chip. This is closed loop control
have been subjects of imagination. The reality of industrial system. In which controller, drive circuits and the sensor
robots only came in the sixties when Joseph F. Engelberger circuit plays important role.
introduced the PUMA robot as a freely programmable, Plus Width Modulation (PWM) is used to control the speed
universal, handling device. With it came automation in of DC motor. The hardware functional block is to be design in
manufacturing industry, economic issues, and social concern software module with the help of VHDL coding. This
about human labour replaced by machines. The versatility of hardware and software co-design is for five axes OWI 535
these robot machines has been increasing, largely due to their arm robot model for the movement of robot arm for pick and
continuously increasing ability of information processing. The place application. The interfacing of different hardware blocks
main goal was the autonomous robot. But, as the application with software module for real time application is very
field for robots is widening, and the robot is coming out of the challenging and demanding in every field to achieve better
factory halls, new challenges are seen, and even a change of command over control through software only without
paradigm is taking shape [2]. disturbing the hardware. It reduced the manufacture time,
performance, accuracy and lifetime of the particular system
[7].
____________________________________________________________________________________________________________
672
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
673
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
674
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[4] Joy, B., “Why the future doesn’t need us”, Wired Archive 8.04,
http://www.wired.com/wired/archive/8.04/joy.html, 1999.
[5] Christaller, T., Cognitive robotics. A new approach to artificial
intelligence, Artificial Life and Robotics, 3, 1999.
[6] Zavala, S. R., & Cardiel, X. A microcontroller-based 3-axis
manipulator applied to a printed-circuit-board driller system. Paper
presented at the Industrial Electronics, 2000. ISIE 2000. Proceedings of
the 2000 IEEE International Symposium on, 2000.
[7] Meshram, U., Bande, P., & Harkare, R. Hardware and Software Co-
design for Robot Arm Position Control Using VHDL and FPGA. Paper
presented at the Advances in Recent Technologies in Communication
and Computing, 2009. ARTCom'09. International Conference on, 2009.
[8] Hace, A., Jezernik, K., Curk, B., & Terbuc, M. Robust motion control
of XY table for laser cutting machine. Paper presented at the Industrial
Electronics Society, 1998. IECON'98. Proceedings of the 24th Annual
Conference of the IEEE, 1998.
[9] ATMEL. ATMEGA128 Datasheet Retrieved 21.04.2015, from
http://www.atmel.com/Images/Atmel-8151-8-bit-AVR-
ATmega128A_Datasheet.pdf
[10] Barnett, R., Cox, S., & O'Cull, L. Embedded C programming and the
Atmel AVR: Cengage Learning, 2006.
[11] Mazidi, M. A., Naimi, S., & Naimi, S. AVR Microcontroller and
Embedded Systems: Using Assembly and C: Prentice Hall Press, 2010
[12] Microchip. PICkit 3 Programmer/Debugger User's Guide Retrieved
21.04.2015, from
http://ww1.microchip.com/downloads/en/DeviceDoc/PICkit_3_User_
Guide_51795A.pdf
[13] National-Semiconductor. LM2596S Datasheet Retrieved 21.04.2015,
from
http://pdf.datasheetcatalog.com/datasheet/nationalsemiconductor/DS01
2583.PDF
____________________________________________________________________________________________________________
675
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— A coplanar waveguide (CPW) fed patch antenna rectenna with compact structure for communication systems.
array is proposed in this paper. The resonance frequency of An aperture-coupled dual polarization patch antenna is
antenna is 2.47 GHz and is inside of ISM 2.45 GHz band. The utilized as the receiving antenna. The max gain of antenna has
dimension of proposed antenna is 230 mm x 91.5 mm. For been reported 7.0dBi @6.1 GHz [12].
substrate the material used is Roger 5870. The relative
The proposed antenna has been designed and successfully
permittivity of the substrate material is 2.33. For increasing the
gain, an array antenna has been designed. The antenna gain achieved outstanding performance within operating frequency
reachs up to 12 dB. The working has been simulated in the HFSS of 2.45 GHz. Moreover, such antenna manage to achieve
Ansoft. The reflection coefficient has been measured by Agilent minimum reflection coefficient of S11 < -10dB which is
Network Analyzer. The proposed antenna can be used in suitable for industrial, scientific and medical (ISM) band
applications necessaried high gain such as wireless power applications. The length of proposed antenna is 230 mm and
transmission the width is 91.5 mm
Keywords— Antenna, High gain, Antenna array, Microstrip II. GEOMETRY OF DESIGNED ANTENNA
patch antenna The antenna has been fabricated on a 1.6mm thick
RT/Duroid 5870 substrate with a dielectric constant of 2.33
I. INTRODUCTION
and loss tangent of 0.0012. Fig.1 is the proposed high gain
Antenna is the transitional structure between free space and microstrip patch antenna array. For antenna simulation has
a guiding device. The guiding device or transmission line may been used Ansoft HFSS.
take the form of a coaxial line or a waveguide, and it is used
to transport electromagnetic energy from the transmitting
source to the antenna, or from the antenna to the receiver [1]. a
Antennas are very important element of electromagnetic
communication. An antenna performance can be described
with some parameters. The resonance frequency describe the
antennas working frequency. The researches of antenna are
intensive in ISM Bands. Many of ISM Band centre b
frequencies are, 13.56MHz [2], 433.92MHz [3], 2.45GHz [4],
5.8GHz [5]. In other hand these antennas have different
structures.
In the some application bandwidth is an important
parameter [6] but in the other applications antenna gain may
be more important parameter [7]. An application where
antenna gain is important may be done array antenna working
[8, 9].
Huang et al. has proposed an antenna design for triple-band
biotelemetry with data telemetry (402 MHz), wireless
powering transmission (433 MHz), and wake-up controller First
(2.45 GHz). The prototype of the rectenna was fabricated on a Second Feeding Part Impedance
Roger 3210 substrate. The antenna type was PIFA and Impedance
antenna gains were -7dB @402MHz, -11 dB @433MHz and -
Matching
Matching Part
15dB @2450MHz [10]. Ali et al. has proposed a circularly
polarized (CP) microstrip patch antenna that is function as a Part
WLAN antenna in the 5.15–5.35 GHz and as a rectenna at 5.5
GHz. The antenna is 14.8 mm by 14.8 mm by 3 mm in size Fig. 1 Coplanar feeded patch antenna array
which can be readily integrated within a sensor circuit board
[11]. Yang et al. presented two-port printed microstrip The antenna is detailed in 3 parts. Fig. 1 present the feeding
part, fig. 2 present the first Impedance matching part and fig. 3
____________________________________________________________________________________________________________
676
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
present the second Impedance matching part of prosed The parameters in fig. 1, fig.2, fig. 3 and fig 4 given is
antenna. bellow descripted; a=39mm, b=48mm, c=2,5mm, d=4,25mm,
e=12,5mm, f=6mm, g=3mm, h=4mm,i=7mm,j=7mm, k=4mm,
c l=3mm, m=5mm. The general dimension of proposed antenna
is 230 mm x 91.5 mm. The Photograph on fabricated coplanar
d feeded patch antenna array is given in fig.5.
III. RESULTS
The high gain microstrip patch antenna array has been
fabricated and the S11 (return loss) parameter has been
measured with Agilent Technologies Field Fox Microwave
Fig. 2 Feeding part of coplanar feeded patch antenna array
Vector Network Analyser N9926A. Fig.6 shows the S11
results comparatively of the designed antenna as simulated
e values with Ansoft HFSS and measured values network
analyser. The S11 value approximates to -14dB for 2.48 GHz.
i f The Bandwidth is from 2.48GHz to 2.51GHz about 30MHz.
-3
j hg
S11(dB)
-6
-9
Fig. 3 First Impedance Matching part of coplanar feeded patch antenna array HFSS
-12
Meas
-15
m 2,35 2,4 2,45 2,5 2,55
Freq (GHz)
Fig. 6 Simulated and measured return loss value of coplanar feeded patch
antenna array
k
l
Fig. 4 Second Impedance Matching part of coplanar feeded patch antenna
array
____________________________________________________________________________________________________________
677
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSION
In this work, high gain microstrip patch array antenna has
been designed and fabricated that can operates at 2.45 GHz
for ISM applications. Measured and simulated results show
that the designed antenna has good performance over a
bandwidth of 2.47–2.5 GHz. The gain has been measured a
successful value such as 12 dBi for the designed antenna. The
antenna has had an excellent performance of high gain. The
proposed antenna can be used for energy harvesting and
wireless power transfer applications.
REFERENCES
[1] C. A. Balanis, Antenna theory: analysis and design: John Wiley &
Sons, 2016.
[2] V. Fiore, P. Battiato, S. Abdinia, S. Jacobs, I. Chartier, R. Coppard, G.
Klink, E. Cantatore, E. Ragonese, and G. Palmisano, "An integrated 13.56-
MHz RFID tag in a printed organic complementary TFT technology on
flexible substrate," IEEE Transactions on Circuits and Systems I: Regular
Papers, vol. 62, pp. 1668-1677, 2015.
[3] M. Kahriman, "Tangent circles antenna design for RFID applications in
ISM 433 MHz," OPTOELECTRONICS AND ADVANCED MATERIALS-
RAPID COMMUNICATIONS, vol. 8, pp. 127-130, 2014.
[4] A. Kaya, I. Kaya, and H. E. Karaca, "Radio Frequency U‐Shape Slot
Antenna Design with Niti Shape Memory Alloys," Microwave and Optical
Technology Letters, vol. 55, pp. 2976-2984, 2013.
[5] C. Baytöre, M. Palandöken, A. Kaya, and Y. Zoral, "Compact
multiband MIMO antenna designs and analysis for 2.4/5.2/5.8 GHz WLAN
MIMO systems," in 2015 23nd Signal Processing and Communications
Applications Conference (SIU), 2015, pp. 2082-2085.
____________________________________________________________________________________________________________
678
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
+
Computer Engineering, Faculty of Technology
Konya, Turkey
ilkerozkan@selcuk.edu.tr
Abstract— Wheat is the main food source of the humankind. After eliminate all those adverse conditions mentioned. With rapidly
its harvest, it goes through many procedures from its separation developing computer technologies, machine vision systems and
from chaff to its packaging. With the development in technology, image processing techniques have become one of the most
many of these procedures are realized with automatic systems popular research areas in wheat inspection and classification,
which saves the manufacturer the cost of labour, time and
because they have the ability to visually characterize wheat
provides the customer with more quality food. One of the main
concerns of quality food production is to provide a customer with grains by their physical attributes and the process is objective,
the product in its purest form which means the product must be speedy, most efficient, cheap, repeatable and harmless to wheat
separated from all foreign matters. In this study, type-1252 durum seeds [2], [3].
wheat seeds have been separated from junk using the Through the last years, many researchers have evaluated
morphological properties of wheat seeds through the machine vision and image processing techniques if they really
uncompressed video image taken with the camera Prosilica meet the expectations for the inspection and classification of
GT2000c. The main references for the quality measurement of the quality of wheat. There have been many studies about the
wheat seeds are the shape and the dimensions of a wheat seed. determination of the properties of single wheat seed, separation
Aiming for high quality wheat grain storage with no junk, this
of one type wheat from another or identification of damaged
article has adopted various image processing techniques from
image preprocessing to feature extraction. The image processing wheat seeds, but there have not been many researches about
has been realized in a computer environment and the results show separating the wheat seeds from non-wheat seeds [1]. In a study
that the image processing is successful and the detection of wheat conducted by Pourreza et al., nine different wheat classes
seeds from junk was accurate. growing in Iran have been classified according to their textural
properties extracted from Gray Level, GLCM (Gray Level Co-
Keywords— BlobAnalysis, Feature Extraction, Image Processing, occurrence Matrix), GLRM (Gray Level Run-length Matrix),
Junk, Morphological Properties, Segmentation, Wheat Seed, LBP (Local Binary Pattern), LSP (Local Similarity Pattern) and
LSN (Local Similarity Numbers) matrices and classified using
LDA (Linear Discriminate Analysis) [4]. Xia et al. classified a
I. INTRODUCTION single type wheat with regards to its quality by accounting its 7
Food supply is one of the most essential materials for the morphological properties and 6 colour properties [3]. In an
survival of human beings. Wheat, which is the most article written by Güneş et al., it is explained the varieties of
fundamental material of amylaceous products like bread and wheat growing in Turkey are classified according to its textural
pasta, is also one of the main food supply for human beings. It analysis using GLCM and LBP methods and k-Nearest
is important for the food industry to provide quality goods Neighbour type classifier [5]. In a study conducted by Babalik
which includes wheat grains. After its harvest, wheat seeds go et al., variety of wheat classes are identified with 9
through many procedures from its separation from chaff to its morphological and 3 colour features using M-SVM (Multiclass
packaging and they are stored in warehouses to be sold at Support Vector Machines) and BPSO (Binary Particle Swarm
specified intervals [1]. The inspection and the classification of Optimization algorithm [6]. In another study, Farahani tried to
good quality wheat grains can be done by manually through a determine the best potential morphologic features to classify 5
series of instrumental or chemical analysis. Obtaining good different types of durum wheat [7]. Manickavasagan et al. tried
quality wheat product through these tests and analysis is to measure the ability of a machine vision system with a
subjective, time consuming, less efficient, costly and the safe monochrome camera to classify the different types of western
inspection of food without damaging its structure is nearly Canadian wheat types by using bulk sample analysis [8].
impossible [2], [3]. Williams et al. have evaluated two different digital image
In recent years, it has become imperative to use automatic analysis (DIA) approaches to quantifying wheat seed shape for
systems in the inspection and classification of wheat seeds to exploring trait correlations and QTL (Quantitative Trait Loci)
____________________________________________________________________________________________________________
679
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
mapping [9]. All these studies generally have been made with The camera views a 8 cm x 10 cm area inside the box and
the purpose of classifying different types of wheat. However, the 5 sec uncompressed video obtained from the camera is
there is also a small number of studies about purification of transmitted to the image processing computer software with an
wheat from its chaff and other impurities. One of these studies ethernet cable. The system used for this study is shown in Fig.2.
is made by Ebrahimi et al. to propose a machine vision Also, a frame sample of the video acquired is shown in Fig. 3.
automatic grading system which separates the wheat from the
impurities within [1]. The objective of another study conducted
by Paliwal et al. is to develop an algorithm which classifies 5
different types of Canadian wheats and also differentiates the
wheats from non-wheat materials [10]. In another study, FN
Chen et al. have developed an image processing algorithm
which determines black germ wheat [11]. It can be seen from
these examples that so few of the studies are about purification
of wheat from non-wheat materials. Also, all of these studies
are conducted in stationary environment and the possible
outcomes are not known for a non-stationary environment.
In this study, our objective is to obtain good quality type-
1252 durum wheat grains by separating the impurities from
wheat grains in a dynamic system using image processing
techniques. The process stages we used in this study are shown
in Fig.1.
A. Image Acquisition
The system used consists of a band-wheel system where the
durum wheat grains are moving upon, image acquisition
camera Prosilica GT2000C, and illumination apparatus with a
shady box stand for preventing the shadow formation on the
background.
Prosilica GT2000C camera is a 2.2 megapixel, RGB camera
with 2048 x 1088 resolution, CMOS type sensor, 53.7fps
maximum frame rate at full resolution and efficient operation
temperature range between -20 °C and +65 ° C [12]. The Fig. 2 Dynamic system used for acquiring video sample of durum wheat
camera has been placed atop the illumination apparatus and grains with impurities
daylight coloured powerleds have been placed around the
camera. Under the illumination apparatus there is a ground
glass around the camera lens so that the light can refract
uniformly. The camera and the light illumination apparatus has
been placed above a shady box so that the outside light does not
affect the wheat grains and shadow formation can be prevented.
Fig. 3 Frame sample of the video visualizing durum wheat grains with
impurities
____________________________________________________________________________________________________________
680
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
B. Image Preprocessing and Segmentation After adjusting the colormap, the images are turned into gray
The obtained video has been separated into frames and all images and subjected to a 5 x 5 median filter. Median filtering
the frames have been subjected to same processes. First the is a non-linear filtering process which offers an effective noise
colormap of the images adjusted suitably after several trials so reduction without blurring the image [3]. After the filtering
that the image segmentation is successful. The image with process, the background is subtracted from the image with “top-
adjusted colormap is shown in Fig.4. hat” operation so the background noise does not affect the
image processing. The median filtered image with background
subtraction is shown in Fig. 5. Then, morphologically “open”
and “close” operations have been applied to the images to
smooth the boundaries of the objects and the images have their
contrast values adjusted to reduce edge detection errors. The
frame sample after several morphological operations is shown
in Fig. 6. Lastly, they were converted to binary images using an
appropriate threshold value. The binary image of the frame
sample is shown in Fig.7.
Durum wheat grains and other non-wheat materials reside on
the band in a random order so the grains can touch other grains
or the other non-wheat materials. While the morphological
operation “erode” can shrink the objects, thus maybe separate
them. However, if the objects are touching each other they can
still be connected despite the touching area is really small. In
such conditions, a watershed algorithm is needed. The
watershed transform is a segmentation algorithm that can
divide the image into multiple regions where the darker colours
represent a region with lower altitude and lighter colours
represent a region with higher altitude topologically [13]. Thus,
a watershed segmentation algorithm used for properly segment
all the objects from each other. To apply watershed transform
efficiently for segmentation, the first step is to apply the
distance transform which labels each existing pixel according
to the distance with the nearest boundary pixel in a binary
Fig. 4 The frame sample of the video with adjusted colormap image [13]. Although the distance transform is successful for
segmenting round and simple touching objects, it is not
successful for kernels with irregular shapes [13]. The reason
why it is not successful to segment kernels with irregular shapes
is because large number of local minima occur due to large
clusters of objects which causes over-segmentation [14]. So, in
our study, we used a watershed segmentation algorithm with
extended-minima transform to segment the durum wheat grains
where the extended-minima transform can produce markers for
objects where local minima with greater depth than h are
marked and where local minima with less depth than h are
eliminated. By using minima imposition technique, we create
minima at the specific locations associated with the markers
[14]. After this procedure, watershed segmentation is applied
which results in perfect segmentation for touching grains. The
edge detected binary sample image after watershed
segmentation with extended-minima transform is shown in
Fig.8.
C. Feature Extraction
The blob is the region of connected pixels of an object and
blob analysis is the method which investigates these regions
and provides us the information about the regions’ several
properties. It is used for feature extraction processes [15]. The
blob analysis algorithm distinguishes the pixels according to
their values and labels them into 2 categories. The foreground
Fig. 5 Median filtered image with background subtraction
____________________________________________________________________________________________________________
681
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
category is generally the pixels with a non-zero value and the TABLE I
FEATURES USED AND THEIR DEFINITIONS
other category is the background where the pixels with a zero
value belong. Thus, the blob analysis algorithm measures
several properties of the object regions in the image [15]. Feature Definition
After the segmentation process, blob analysis algorithm is Area Number of pixels in the region
used for the feature extraction process to measure the properties The length(in pixels) of the
Major Axis
of wheat grain regions. In our study, 9 morphological features major axis of the ellipse
are used for the separation of durum wheat grains from The length(in pixels) of the
Minor Axis
impurities. These features which are extracted directly by blob minor axis of the ellipse
analysis are area, major axis, minor axis, perimeter equivalent The length(in pixels) around
Perimeter
diameter, eccentricity. Features which can be calculated from the boundary of the region
the others are roundness, shape factor, compactness [6]. The The diameter of a circle with
Equivalent Diameter
features used for this study and their definitions are shown in the same area as the region
the Table I [6], [7], [16]. The ratio of the distance
Eccentricity between the foci of the ellipse
and its major axis length
Fig. 8 The edge detected binary image of the frame sample after watershed
segmentation with local-minima transform
Fig. 7 Binary image of the frame sample
____________________________________________________________________________________________________________
682
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
III. RESULTS AND DISCUSSIONS future studies, we will use different and more efficient
After feature extraction, mean and standard deviation classification models to obtain perfect classification results.
values of the properties have been calculated. Durum wheat
ACKNOWLEDGMENT
grains are separated from the impurities by a simple algorithm
which uses the calculated results as ranges. The mean and the Appreciations to Selcuk University Scientific Research
standard deviation values of the properties is given in Table II. Projects Unit (BAP) for its support of this project.
The algorithm determines the particles which are accepted
REFERENCES
as durum wheat grains by comparing the properties of every
single object within a range specified using the mean values and
[1] Ebrahimi, E., K. Mollazade, and S. Babaei, Toward an automatic
standard deviation values of the properties. If all the properties wheat purity measuring device: A machine vision-based neural
of a single object is within the specified range, then the object networks-assisted imperialist competitive algorithm approach.
is a durum wheat grain. If not, then the object is non-wheat Measurement, 2014. 55: p. 196-205.
material. The obtained result is 73.52% accuracy for the [2] Fengnong Chen, F.C. Wheat image correction for feature
segmentation Based on color Linear CCD. in Digital Manufacturing
separation of non-wheat materials from durum wheat grains and Automation (ICDMA). 2010. ChangSha: IEEE.
which is an optimistic result for image processing of a dynamic [3] Xia, X.F., Chao; Lu, Shu-Jie; Hou, Li-Long. The Analysis of Wheat
environment, because it is much more complex than image Appearance Quality Based on Digital Image Processing. in
processing of a single image of a stationary environment. If a Environmental Science and Information Application Technology.
2010. Wuhan: IEEE.
more complex and more efficient classification algorithm is [4] Pourreza, A., et al., Identification of nine Iranian wheat seed varieties
used for the separation of durum wheat grains from impurities, by textural analysis with image processing. Computers and
than the accuracy result will be improved. Electronics in Agriculture, 2012. 83: p. 102-108.
[5] Ece Olcay Güneş, S.A., Mürvet Kırcı, Amir Kalateh, Yüksel Çakır.
Determination of the varieties and characteristics of wheat seeds
TABLE II
grown in Turkey using image processing techniques. in Third
MEAN AND STANDARD DEVIATION VALUES OF THE FEATURES
International Conference on Agro-geoinformatics 2014. Beijing:
IEEE.
Features Mean Standard Deviation [6] Babalik, A., et al., Effects of Feature Selection Using Binary Particle
Area 151,151 33,585 Swarm Optimization on Wheat Variety Classification. Advances in
Information Technology, 2010. 114: p. 11-17.
Major Axis 88,590 17,901 [7] Farahani, L., Discrimination of some cultivars of durum wheat
Minor Axis 38,357 7,077 (Triticum durum Desf.) using image analysis. International Research
Journal of Applied and Basic Sciences, 2012. 3(7): p. 1375-1380.
Perimeter 172,485 35,456 [8] Manickavasagan, A., et al., Wheat class identification using
Equivalent monochrome images. Journal of Cereal Science, 2008. 47(3): p. 518-
192,451 42,761 527.
Diameter [9] Williams, K., J. Munkvold, and M. Sorrells, Comparison of digital
Eccentricity 0,892 0,056 image analysis using elliptic Fourier descriptors and major
Roundness 0,026 0,011 dimensions to phenotype seed shape in hexaploid wheat (Triticum
aestivum L.). Euphytica, 2013. 190(1): p. 99-116.
Shape [10] Paliwal, J., et al., Cereal grain and dockage identification using
0,068 0,023 machine vision. Biosystems Engineering, 2003. 85(1): p. 51-57.
Factor
[11] FN Chen, F.C., YB Ying. Detect black germ in wheat using machine
Compactness 0,160 0,027 vision. in International Conference on Computer Distributed Control
and Intelligent Environmental Monitoring (CDCIEM). 2011.
Changsha: IEEE.
IV. CONCLUSIONS [12] Technologies, A.V. Prosilica_GT_TechMan.pdf. 2016 [cited 2016 1
Aug 2016]; Available from:
As it is shown in literature, previous studies conducted are https://cdn.alliedvision.com/fileadmin/content/documents/products/c
not about the purification of wheat from non-wheat materials ameras/Prosilica_GT/techman/Prosilica_GT_TechMan.pdf.
generally but about classification of wheat varieties. Also, the [13] Qin, Y.B., et al., Extended-Maxima Transform Watershed
Segmentation Algorithm for Touching Corn Kernels. Advances in
studies conducted in a dynamic environment are almost non- Mechanical Engineering, 2013.
existing. In this study, the objective has been to obtain good [14] Kamran Ali, A.J., M. Umair Gull, Mustansar Fiaz. Medical Image
quality type-1252 durum wheat by separating the impurities Segmentation Using H-minima Transform and Region Merging
from wheat grains in a dynamic system using image processing Technique. in Frontiers of Information Technology (FIT), 2011. 2011.
Islamabad: IEEE.
techniques. These processes are image preprocessing for image [15] R.Kiruthika, S.M., Azha Periasamy, Matching Of Different Rice
enhancement, watershed transform with extended-minima Grains Using Digital Image Processing. International Journal of
transform for segmentation and blob analysis for feature Advanced Research in Electrical, Electronics and Instrumentation
extraction. Engineering, 2013. 2(7): p. 2937-2941.
[16] MathWorks. regionprops. 2016 [cited 2016 02.08.2016]; Available
In this study, after segmentation and feature extraction from: http://www.mathworks.com/help/images/ref/regionprops.html.
processes, a simple algorithm is used for the separation of non-
wheat materials from durum wheat grains. The reason for using
a simple algorithm is because it was a start for dynamic
environments which is not the case for many researches in
literature. However, the obtained result is not bad for a start. In
____________________________________________________________________________________________________________
683
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—A design of ultra-wideband monopole patch antenna is Therefore the printed forms of that designs can be
presented in this study. The design consists of a radiating successfully exploited for ultra-wideband systems. Over the
triangular patch antenna (TPA) fed through a 50 Ohm past years, several monopole antenna structures, such as
microstrip transmission line (MTL) and defected ground rectangular [10], circular [11], elliptical [12], conical [13],
structure (DGS). The triangular monopole is matched to MTL
tapered [14, 15], slot-loaded [16-18] and defected ground [19-
with inserting a thin strip line. The performance of the proposed
antenna is numerically investigated using method of moments 22] configurations, have been suggested for ultra-wideband
(MoM) and verified through measurements. Based on the applications.
measured results, the antenna operates over large frequency The popularity of defected ground structure (DGS) in
range of 2.6-18.2 GHz at -10 dB. The proposed antenna has well antenna study has increased immensely [19-22]. The DGS is
characteristics in terms of radiation pattern, peak gain and total basically obtained by properly modifying the ground plane of
efficiency across the operating ultra-wideband frequency range. a patch antenna. Thus compact designs with high
Furthermore, the antenna system is suitable for near-millimetre characteristics of especially wideband and gain can be
wave applications, since the operating band reaches to 18.2 GHz. achieved.
In this study, an ultra-wideband triangular patch antenna
Keywords—Antennas, patch antennas, antenna designing, ultra- (TPA) that consists of a triangular monopole is designed. The
wideband (UWB), defected ground structure (DGS) triangular monopole is positioned on a rectangular DGS. The
I. INTRODUCTION triangular radiating monopole patch is fed through a
microstrip transmission line (MTL). The TPA is modelled in
The ultra-wideband systems have received considerable HyperLynx 3D EM simulation software based on method of
attention in recent years due to allowing signal transmission in moments (MoM) and fabricated for measurements. The TPA
a large bandwidth with a low energy level. Radar imaging, with a small size of 23x31 mm2 operates across 2.6 GHz to
mobile and satellite communications, multimedia streaming, 18.2 GHz based on measured results.
biomedical imaging are the most important application areas
of ultra-wideband systems [1-5]. The ultra-wideband systems II. THE PROPOSED DGS-BASED TPA STRUCTURE
have been further studied especially since Federal The 3D geometry of the TPA model is depicted in Fig. 1 and
Communications Commission (FCC) approved the unlicensed its dimensions are given in Table I. The TPA structure is
use of the frequency range of 3.1–10.6 GHz for low power composed of a 14x23 mm2 triangular monopole patch fed by
emitting implementations [6]. 50 Ohm MTL and a 17x23 mm2 rectangular DGS. The
Patch antennas are widely used in the wireless systems due radiating plane is placed over a 1.6 mm height substrate with
to various advantages such as low profile, small size, light 4.4 relative dielectric constant on the DGS. Therefore, the
weight, conformability to mounting host and capability of TPA has a total dimension of 23x31x1.6 mm3. In order to
integration into the arrays. Owing to these advantages, the improve the impedance matching, a 1 mm thin strip line is
printed antennas have been integrated to the latest wireless inserted between triangular radiating patch and the MTL.
communications of wireless local area network (WLAN), long
term evolution (LTE) and worldwide interoperability for
microwave access (WiMAX) standards, and ultra-wideband
systems. Designing an appropriate antenna is the most
important part of ultra-wideband applications because it
effects the whole system performance.
There are several antenna structures that can be used for
ultra-wideband systems such as the Vivaldi antenna [7], log
periodic antenna [8], spiral antenna [9] and monopole antenna
[10-18]. Monopole designs show inherently wideband
behaviour as well as omni-directional radiation pattern. (a)
____________________________________________________________________________________________________________
684
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(b) (c)
____________________________________________________________________________________________________________
685
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(b)
____________________________________________________________________________________________________________
686
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
VII. CONCLUSIONS
In this work, we presented a compact ultra-wideband
monopole patch antenna design operating over a large
frequency range of 2.6–18.2 GHz. The monopole TPA fed via
a MTL is built on a half-length DGS. The whole antenna
system has a small size of 23x31 mm2. A thinner strip line is
inserted between the triangular monopole and the MTL to
improve the impedance matching. The proposed TPA is
verified by simulation and validated through measurement.
The TPA has superior characteristics of larger bandwidth than
(c) (d) the regulated ultra-wideband of 3.1–10.6 GHz, near omni-
directional pattern, uniform gain and high efficiency. In
conclusion, since the TPA operates up to 18.2 GHz it can be
also utilized for near-millimeter wave applications.
ACKNOWLEDGMENT
This work is supported by the Scientific Research Fund
Department of Karamanoglu Mehmetbey University under
grant no: 12-M-15.
REFERENCES
(e) (f) [1] D. W. Winters, J.D. Shea, E. L. Madsen, G. R. Frank, B. D. Van Veen,
and S. C. Hagness, “Estimating the breast surface using UWB
Fig. 7 Measured and simulated radiation gain patterns of TPA
microwave monostatic backscatter measurements,” IEEE Transactions
at: a) 2.7 GHz on x-z plane, b) 2.7 GHz on y-z plane, c) 8.1 on Biomedical Engineering, vol. 55, pp. 247–256,2008.
GHz on x-z plane, d) 8.1 GHz on y-z plane, e) 11.9 GHz on x-z [2] S. Nag, M. A. Barnes, T. Payment, and G. W. Holladay, “An ultra-
plane, f) 11.9 GHz on y-z plane (solid line measured, dot line wideband through-wall radar for detecting the motion of people in real
time”, Proc. SPIE, Radar Sensor Technology and Data Visualization,
is simulated)
2002, vol. 4744.
Fig. 8 indicates the peak gain together with the total [3] D. L. Sostanovsky, A. O. Boryssenko, and E. S. Boryssenko, “UWB
efficiency of the TPA. It is known that total efficiency of an radar imaging system with two-element receiving array antenna”, Proc.
antenna is the product of the efficiencies relevant to radiation, IEEE 5th International Conference on Antenna Theory and Techniques,
2005, pp. 357–360.
mismatch, conduction and dielectric. From the figure, the
[4] W. Cui, P. Ranta, T. A. Brown, and C. Reed, “Wireless video
peak gain changes between 2.2 dBi and 4.5 dBi levels across streaming over UWB”, Proc. IEEE International Conference on Ultra-
the interested frequency band, whereas a minimum gain Wideband 2007, pp. 933–936.
occurs 2.2 dBi at 2.7 GHz and maximum gain take places 4.5 [5] D. Valente, E. Cianca, S. Mukherjee, T. Rossi, M. Ruggieri, and R.
Prasad, “Comparison of UWB approaches applied to ehf satellite
dBi at 16 GHz. On the other hand, the total efficiency keeps
communications”, Proc. IEEE Aerospace Conference 2012, pp. 1–6.
higher than 66%. The highest and second highest total [6] Federal Communications Commission, “FCC report and order on ultra
efficiencies are 94% and 93% at 2.7 GHz and 5 GHz, wideband technology”, Washington DC, 2002.
respectively. [7] S. G. Kim and K. Chang, “Ultra wideband exponentially-tapered
antipodal vivaldi antennas”, Proc. IEEE Antennas and Propagation
Society Symposium, 2004, pp. 2273–2276.
[8] S. Y. Chen; P. H. Wang; and P. Hsu, “Uniplanar log-periodic slot
antenna fed by a CPW for UWB applications”, IEEE Antennas and
Wireless Propagation Lett., vol. 5, pp. 256–259, 2006.
[9] B. A. Kramer, M. Lee, C. C. Chen, and J. L. Volakis, “Design and
performance of an ultra-wideband ceramic-loaded slot spiral”, IEEE
Transactions on Antennas and Propagation, vol. 53, pp. 2193–2199,
July 2005.
[10] Q. Wu, R. Jin, J. Geng, M. Ding, “printed omni-directional UWB
monopole antenna with very compact size”, IEEE Transactions on
Antennas and Propagation, vol. 56, pp. 896–899, 2008.
[11] J. Liang, C. C. Chiau, X. Chen, C. G. Parini, “Study of a printed
circular disc monopole antenna for UWB systems”, IEEE Transactions
on Antennas and Propagation, vol. 53, pp. 3500–3504, July 2005.
[12] K. G. Thomas and N. Lenin, “Ultra-Wideband Printed Monopole
Antenna”, Microwave and Optical Technology Lett., vol. 49, pp. 1082-
1085,2007.
[13] S. Y. Suh, W. Stutzman, and W. Davis, “A new ultra-wideband printed
monopole antenna: the planar inverted cone antenna (PICA)”, IEEE
Transactions on Antennas and Propagation, vol. 52, pp. 1361–1364,
2004.
Fig. 8 Measured peak gain and total efficiency of TPA
____________________________________________________________________________________________________________
687
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
688
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—Signal denoising approaches on data of any dimension proposed. Therefore, brief methodology of other well-known
largely relies on the assumption that data and the noise subspace is given for comparisons.
components and the noise itself are somewhat uncorrelated. Subspace methods rely on the expectation that the noisy
However, any denoising process heavily depending on this data can be separated into two or more component where
assumption retreats when the signal component is adversely
noise can be handled more efficiently within. A Singular
affected by the operation. Therefore, several proposed
algorithms try to separate the data into two or more parts with Value Decomposition (SVD) based approach, proposed by
varying noise levels so that denoising process can be applied on Dendrinos et al. [1], uses the expectation that, after
them with different parameters and constraints. In this paper, factorization of the data into sub-data, noise energy
the proposed method separates the speech data into magnitude concentrates in vectors corresponding to smaller singular
and phase where the magnitude part is further separated into values. In the simplest denoising approach these are zeroed
common and difference parts using common vector analysis. It is and voice data is recomposed. This technique is improved by
assumed that the noise largely resides on difference part and Jensen et al. [2] for colored noise on which the former method
therefore denoised by a known algorithm. The speech data is has somewhat failed to reduce. On the other hand, their
reconstructed by combining common, difference and phase
method with high computational complexity had several
parts. Using Linear Minimum Mean Square Error Estimation
algorithm on the difference part, excellent denoising results are constraints for controlling residual noise. Ephraim et al. [3]
obtained. Results are compared with that of the state of the art in aimed to optimize the estimator that minimizes distortion
well-known speech quality measures. caused by residual noise. Noisy signal is separated into noise
and signal subspaces using Karhunen Loeve Transform (KLT)
Keywords— speech, denoising, cva, subspace, frequency domain whereby zeroing the components in noise subspace and
restructuring the signal subspace using a gain function.
I. INTRODUCTION Components in subspaces are recombined again to obtain
For decades, as cognitive science penetrated into denoised signal through inverse KLT. Mittal et al. [4] and
automation systems, voice automated applications like voice Rezayee et al. [5] expanded this work for colored noise. They
directed banking, voice signatures, intelligent homes; voice obtained better results by employing different KLT matrices
recognizing mobile-phone apps and such became both and converging covariance matrix of the noise vectors to a
possible and popular. All voice applications somewhat diagonal matrix respectively.
necessitate high quality voice signals, mostly in digital form, Common Vector Approach (CVA) is a subspace method
impetrating voice denoising algorithms. All naturally used in recognition applications. In CVA, training data
collected signals carry some noise energy weather from representing each subject to be discriminated are used to form
electronics/transmission or unrelated background signals. its own class. In a speech recognition application, ambient
Voice denoising aims to improve signal quality, voice noise, ages and genders of speakers result in differences
intelligibility or do both, achieving that with minimal loss in within a class [6]. CVA is based on the common component
signal energy. of those, basically by removing these differences within the
Although the voice denoising methods can be separated class. This component is called the common vector. It has
into two as single and multiple channel algorithms, been employed in speaker recognition [7], speech recognition
researchers mostly focused on single channel because of [8]-[10], face recognition [11], fault detection in electrical
incomparably higher number of encounters of such. Spectral motors [12], spam e-mail detection [13]. CVA has also been
subtraction, noise estimation, Wiener deconvolution/filtering, used in image denoising [14].
statistical and subspace based methods are considered as
II. COMMON VECTOR APPROACH
mainly single channel, notwithstanding the fact that they can
be and are also employed in multichannel systems. In this When differences between feature vectors in a class are
paper, a method based on common vector analysis (CVA) is removed, the remaining vector which consists of features
invariant within the class is called the common vector. A
____________________________________________________________________________________________________________
689
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
feature vector is then, presumed to be a sum of common and classes can be constructed by searching similar vectors within
difference components. If the number of feature vectors ( m ) a reasonable time range.
is greater than the dimension ( n ) of the vectors, then this is Since the raw vectors in speech data are selected to be
called a sufficient case ( m n ). In the sufficient case, the sample frames of length n , the words vector and frame are
common vector is the mean vector. The insufficient case used interchangeably used within this paper.
occurs when the vector dimension is larger than or equal to
III. PROPOSED ALGORITHM
the number of vectors ( n m ), which is the case in most
practical applications where, for example, too few image The denoising algorithm proposed in this paper relies on
blocks with many pixels each exist. In general, it covers the the intuition that the spectral content of the speech does not
setups where number of samples is less than the sample abrubtly change and changes are mostly noise related.
dimension. Since we are interested in such setups, only Approach is similar to time averaging of Fast Fourier
insufficient case of CVA is described in the following. Transform (FFT) data in digital spectrum analyzers. On the
Let the feature vectors in a pattern class be linearly other hand, the overlapping ratio is the highest in the proposed
algorithm. As illustrated in the Fig. 1, frames are picked from
independent a1 , a2 ,..., am . Each vector can be written as
original speech data stream by a sliding Hamming window of
ai ai ,diff acomm , i 1, 2,..., m (1) width n
where acomm and ai ,diff are common vector of the class wi 0.54 0.46cos(2 i /( n 1)) , i 1, 2,..., n (5)
which slides 1 sample for each subsequent frame. Although
and difference vector per ai respectively. The covariance not required, it is logical to select n to correspond
matrix for the class is approximately 4 ms speech data since spectral characteristics
m of speech may greatly change for longer intervals. We
(ai aavg )(ai aavg )T (2) conducted some tests for determining optimal frame length, as
i 1 explained in the following section and concluded that 4 ms is
m adequate.
where aavg 1
m a i denotes the mean of the vectors in For each frame to be denoised, a class is constructed by
i 1 picking m most spectrally similar Hamming windowed frames
the class. In insufficient case ( n m ), n m 1 of the within its neighbourhood. Obviously entire data stream can be
eigenvalues will be zero and corresponding eigenvectors ( u j ) used for selecting the frames and constructing the class. In our
experiments, it is determined that a neighbourhood size that
span the indifference subspace B , while the remaining contains 2 n 1 frames, xi , including the current frame to
eigenvectors span the difference subspace B where B and
be denoised, is sufficient for both required number of vectors
B are orthogonal. The common vector of the class can be for the class and reasonable computational complexity. FFT of
found by projecting any feature vector onto the indifference these 2 n 1 frames are calculated and their magnitude and
subspace B using phase components are separated as
bk FFT {xk }
acom P ai , i 1, 2,..., m . (3) . (5)
pk FFT {xk }
The projection matrix P is calculated using eigenvectors
m 1 magnitude frames that are most similar to the
u j that correspond zero valued eigenvalues (spans B ) via magnitude frame of the current frame (the one to be denoised)
n m 1 are picked and the class is constructed with a total of m
P uu j
T
j . (4) frames. When distances to the current magnitude frame
j 1 calculated using the Euclidean distance
The subspace methods other than CVA requires the inverse n
of the covariance matrix [15]. However, n m inhibits b
2
dk k ,i bcur ,i . (6)
the calculation of inverse of whereas CVA does not have i 1
this problem. where bcur ,i and bk ,i are i th dimension components of the
It is expected that noise mainly reside within the difference
components when it is uncorrelated between class vectors. current and k th
magnitude frames, and are sorted from
Therefore, it is imperative to construct classes as correctly as smallest to largest, the m frames with the smallest d k are
possible. When class information of the vectors is not
available, classes should be constructed by collecting similar selected into the class member set A . Since bcur would have
vectors into a data set matrix for each evaluated vector whose zero distance to itself, it will be assigned index 1 and called
common vector is to be found. When the input data is a stream a1 as indicated in Fig. 1.
or can be handled as a stream with bounds, for example,
____________________________________________________________________________________________________________
690
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
as described in the previous section. aavg should be added common component, which is considered almost noise-free.
After denoising a1,dif and obtaining a1,dif , the denoised
back after the denoising.
difference frame, current denoised magnitude frame is
reconstructed via a1 acom a1,dif . Time domain speech
... ... ... ... frame is reconstructed by adding phase information and
t applying inverse FFT.
x1 After applying the described algorithm on each frame,
first frame last frame denoised time frames are combined to build the denoised
start of second frame xi x2 n 1
x2 current frame speech data. Since the frames are overlapping, there are
several options at the recombining stage. Just adding them
onto the appropriate time location is one of them. Here a
x1 x1 weighting window can be used to increase the weight of the
x2 x2 center of the frame. In our experiments, we noted that just
... ...
xi Hamming
xi adding the frames (flat window) is sufficient and have least
... ... complexity.
x2 n 1 x2 n 1 It should be noted here that there are several parameters
FFT that can be optimized for the best performance on the speech
n
~4 ms data to be denoised; m (number of frames in classes), n
Mag Phase (frame size), PCA parameters, recombining options. However,
. since we intended an algorithm that requires no data
... ... dependent optimisation parameters, these optimizations are
keep phase performed for a large training speech data set and best logical
... ... parameter set is kept for all.
IV. EXPERIMENTS
For the experimental work on the proposed algorithm,
collect most similar a1 NOISEUS (Hu, and Loizou, 2007) speech database is used.
m mag. frames a2 NOISEUS is composed of 30 English sentences spoken by 3
male and 3 female speakers. Recordings are sampled by 8 kHz
... 16 bits with approximately 2 seconds in length. 8 different
am
a1,diff noise type (airport, crowd, car, exhibition hall, restaurant,
CVA train station, street and train) are added onto each speech data
difference aavg to obtain 4 SNR levels (0 dB, 5 dB, 10 dB, 15 dB). Noise data
denoise
acom is itself taken from AURORA database. In addition, the
using PCA common database is extended by adding 4 levels of white noise onto
the data.
denoised dif Initial experiments are conducted to determine best or
reasonable parameter values for CVA. These are; frame size,
overlap ratio and neighbourhood size from which the class
member candidates are picked. Fig. 2 shows the performance
graphics for various frame sizes and input noise levels. From
add phase
these tests, it is determined that frame size of 40 samples
IFFT next denoised frame (corresponding to 4 ms) and highest overlap ratio are adequate
for both performance and complexity. Neighbourhood size
denoised frame
... ... ... ... tests, on the other hand, were inconclusive for widths greater
t
than three frame sizes. It is seen that algorithm becomes data
denoised frames are combined using overlap & add dependent for larger search areas. In the following tests, we
Fig. 1. Flow of the proposed algorithm.
____________________________________________________________________________________________________________
691
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
used NW 3N as the neighbourhood size, where N is the 1. Perceptually motivated subspace algorithm [16], will
shortly be called as sub from now on.
frame size in samples. 2. Psychoacoustically motivated statistical method [17],
will be called as stat.
3. Wiener filtering algorithm based on wavelet thresholding
multi-taper spectra [18], will be called as wien.
4. A variant of minimum controlled recursive average
algorithm [19], will be named as rec.
5. Continuous spectral tracking [20], will be named as spec
in the following sections.
Performance measures used in comparisons are Perceptual
Evaluation of Speech Quality (PESQ), Log Likelihood Ratio
(LLR) and Euclidean Distance in Cepstral Domain (CEP),
most used measures in the literature.
TABLE I
COMPARISON OF 6 METHODS ON PESQ MEASURE
airport
5 dB 1.87 1.68 1.84 1.77 1.82 2.05
10 dB 2.23 2.09 2.24 2.20 2.26 2.40
15 dB 2.63 2.59 2.68 2.72 2.73 2.74
0 dB 1.37 1.20 1.34 1.13 1.34 1.64
5 dB 1.84 1.73 1.82 1.62 1.78 2.02
crowd 10 dB 2.23 2.16 2.25 2.24 2.24 2.39
15 dB 2.63 2.60 2.65 2.69 2.70 2.74
0 dB 1.46 1.27 1.41 1.23 1.40 1.68
5 dB 1.91 1.92 1.89 1.67 1.86 1.97
car
____________________________________________________________________________________________________________
692
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
of the average disturbance value Davg and the average [23]. Lower LLR values mean higher quality speech signal. In
Table II, LLR values for CVA and other 5 methods are
asymmetrical disturbance values Aavg where a0 =4.5, a1 =- compared with boldfaces indicating the best/lowest LLR value
for a test input. It is notable that CVA is superior to the
01 and a2 =-0.0309 are the optimized for speech processed
compared methods since it generated the lowest LLR for all
through networks [21]. Table 1 gives the results for CVA background noise tests. However, for white noise cases CVA
along with sub [16], stat [17], wien [18], rec [19], spec [20] in failed to be the best even though the scores are close.
PESQ measure. Best PESQ values are marked with boldfaced CVA performed best among all methods in all PESQ tests
characters. Each number in the table is the average of PESQ except 5. That indicates about 83% success. In white noise
values for 180 recordings (30 sentences spoken by 6 cases, although not the best, CVA performed close to the best.
individuals).
TABLE IIIII
TABLE III COMPARISON OF 6 METHODS ON CEP MEASURE
COMPARISON OF 6 METHODS ON LLR MEASURE
sub stat wien rec spec cva
sub stat wien rec spec cva 0 dB 7,27 6,29 7,44 6,11 5,85 5,27
airport
0 dB 1.29 1.09 1.37 1.03 0.96 0.85 5 dB 6,47 5,57 6,54 5,42 5,27 4,52
airport
5 dB 1.10 0.92 1.16 0.86 0.83 0.66 10 dB 5,39 4,62 5,55 4,59 4,62 3,73
10 dB 0.86 0.70 0.94 0.67 0.69 0.49 15 dB 4,41 3,90 4,45 3,91 4,02 3,02
15 dB 0.66 0.56 0.72 0.54 0.56 0.35 0 dB 7,26 6,47 7,32 6,22 6,04 5,37
0 dB 1.30 1.14 1.35 1.08 1.03 0.89 5 dB 6,43 5,55 6,58 5,50 5,42 4,60
crowd
5 dB 1.10 0.92 1.17 0.90 0.88 0.70 10 dB 5,47 4,63 5,50 4,61 4,69 3,76
crowd
10 dB 0.88 0.71 0.93 0.69 0.72 0.50 15 dB 4,43 3,89 4,47 3,92 4,08 3,00
15 dB 0.67 0.56 0.72 0.55 0.59 0.36 0 dB 7,46 6,23 7,77 6,15 5,96 5,43
0 dB 1.35 1.10 1.47 1.08 1.03 0.94 5 dB 6,49 5,40 6,74 5,30 5,31 4,66
car
5 dB 1.11 0.89 1.22 0.86 0.86 0.73 10 dB 5,92 4,76 6,19 4,50 4,62 3,87
car
10 dB 0.98 0.75 1.10 0.67 0.71 0.54 15 dB 4,63 3,90 4,88 3,74 3,94 3,06
15 dB 0.71 0.57 0.82 0.51 0.57 0.37 0 dB 7,44 6,59 8,02 6,62 6,43 5,95
ex. hall
0 dB 1.34 1.19 1.52 1.17 1.14 1.06 5 dB 6,60 5,76 7,12 5,87 5,78 5,16
ex. hall
5 dB 1.13 0.98 1.31 0.98 0.97 0.85 10 dB 6,08 5,23 6,61 5,03 5,09 4,33
10 dB 1.02 0.86 1.19 0.78 0.81 0.64 15 dB 5,11 4,61 5,70 4,21 4,39 3,53
15 dB 0.81 0.74 1.01 0.61 0.66 0.47 0 dB 7,50 6,45 7,58 6,07 5,89 5,27
restrnt
0 dB 1.35 1.12 1.39 1.02 0.98 0.85 5 dB 6,19 5,42 6,21 5,29 5,18 4,52
restrnt
5 dB 1.04 0.88 1.08 0.84 0.82 0.66 10 dB 5,46 4,74 5,55 4,58 4,57 3,73
10 dB 0.88 0.73 0.94 0.68 0.68 0.49 15 dB 4,57 4,06 4,66 4,05 4,16 3,08
15 dB 0.70 0.60 0.77 0.56 0.60 0.36 0 dB 7,20 6,27 7,46 6,14 5,91 5,35
0 dB 1.28 1.10 1.39 1.06 1.00 0.89 5 dB 6,37 5,28 6,68 5,25 5,18 4,51
station
5 dB 1.20 1.06 1.28 0.99 1.01 0.88 10 dB 5,99 5,31 6,26 5,25 5,37 4,63
train
10 dB 1.01 0.86 1.10 0.82 0.86 0.68 15 dB 5,05 4,47 5,26 4,43 4,60 3,81
15 dB 0.80 0.66 0.87 0.64 0.69 0.50 0 dB 8,00 8,45 8,60 7,78 8,66 8,45
0 dB 1.44 1.50 1.64 1.36 1.49 1.43 5 dB 7,36 7,64 8,02 7,32 8,29 7,97
white
5 dB 1.30 1.27 1.52 1.22 1.37 1.26 10 dB 6,57 7,01 7,36 6,69 7,79 7,36
white
10 dB 1.11 1.12 1.36 1.03 1.23 1.07 15 dB 5,65 6,25 6,48 5,91 7,09 6,61
15 dB 0.91 0.96 1.17 0.84 1.06 0.88
____________________________________________________________________________________________________________
693
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
V. CONCLUSIONS IEEE Trans. On Speech and Audio Processing, vol. 9( 6), pp. 655-662,
2001.
Tests conducted on 30 sentences spoken by 6 individuals [10] M. B. Gülmezoglu, V. Dzhafarov and A. Barkana, "The Common
with added 4 levels of 8 structured background noise Vector Approach and its Comparison with Other Subspace Methods in
recordings (total of 5760 recordings per method per quality Case of Sufficient Data," Computer Speech and Language, vol. 21, pp.
266-281, 2007.
measure) let us safely conclude that the proposed CVA [11] H. Çevikalp, M. Neamtu, M. Wilkes and A. Barkana, "Discriminative
method is superior against other 5 methods. In additional tests Common Vectors for Face Recognition," IEEE Trans. on Pattern
using white noise, on the other hand, CVA has failed to be the Analysis and Machine Intelligence, vol. 27(1), pp. 4-13, 2005.
best (total of 720 recordings per method per quality measure). [12] M. B. Gülmezoglu and S. Ergin, "An Approach for Bearing Fault
Detection in Electrical Motors," European Transactions on Electrical
However, in most of the tests that CVA was not the best, its Power, vol. 17(6), pp. 628-641, 2007.
scores were close to the best. [13] S. Günal, S. Ergin, M. B. Gülmezoglu and Ö. N. Gerek, "On Feature
Extraction for Spam E-Mail Detection," Lecture Notes in Computer
ACKNOWLEDGMENT Science, vol. 4105, pp. 635-642, 2006.
[14] K. Özkan and E. Seke, "Image Denoising Using Common Vector
This research has been funded by Eskisehir Osmangazi Approach, "IET Image Processing, vol. 9(8), pp. 709-715, 2015.
University Scientific Research Commission with project no [15] L. Zhang, W. S. Dong, D. Zhang, and G. M. Shi, "Two-stage image
2015-758. denoising by principal component analysis with local pixel grouping,"
Pattern Recognition, vol. 43, pp. 1531-1549, 2010.
REFERENCES [16] F. Jabloun and B. Champagne, "Incorporating the human hearing
properties in the signal subspace approach for speech enhancement,"
[1] M. Dendrinos, S. Bakamidis and G. Carayannis, "Speech enhancement IEEE Trans. on Speech and Audio Processing, vol. 11(6), pp. 700-708,
from noise: a regenerative approach," Speech Comm., vol. 10 (1), pp. 2003.
45-57, 1991. [17] Y. Hu and P. Loizou, "Incorporating a psychoacoustical model in
[2] S. H. Jensen, P. C. Hansen, S. D. Hansen and J. A Sorensen, frequency domain speech enhancement," IEEE Signal Processing
"Reduction of broadband noise in speech by truncated QSVD," IEEE Letters, vol. 11(2), pp. 270-273, 2004.
Trans. Speech Audio Processing, vol. 3, pp. 439-448, 1995. [18] Y. Hu and P. Loizou, "Speech enhancement based on wavelet
[3] Y. Ephraim and H. L. Van Trees, "A signal subspace approach for thresholding the multitaper spectrum," IEEE Trans. on Speech and
speech Enhancement," IEEE Trans. on Speech Audio Processing, 3, Audio Processing, vol. 12(1), pp. 59-67, 2004.
251. 166, 1995. [19] S. Rangachari and P. Loizou, "A noise estimation algorithm for highly
[4] U. Mittal and N. Phamdo, "Signal/noise KLT based approach for nonstationary environments," Speech Communication, vol. 28, pp.
enhancing speech degraded by colored noise," IEEE Trans. Speech and 220-231, 2006.
Audio Proc., vol. 8 (2), pp. 159-167, 2000. [20] G. Doblinger, "Computationally Efficient Speech Enhancement By
[5] A. Rezayee and S. Gazor, "An Adaptive KLT Approach for Speech Spectral Minima Tracking in Subbands," Proc. EuroSpeech, vol. 2, p.
Enhancement," IEEE Trans. Speech and Audio Proc., vol. 9 (2), pp. 1513-1516, 1995.
87-95, 2001. [21] Y. Hu and P. Loizou, "Evaluation of objective quality measures for
[6] S. Ergin, "The Improvement and Recognition of the Noisy Speech speech enhancement," IEEE Trans. Speech Audio Processsing, vol. 16
Parameters," M. Eng. Thesis, Eskisehir Osmangazi University, (1), pp. 229-238, 2008.
Eskisehir, Turkey, 2004, [Online]. Available: [22] S. Quackenbush, T. Barnwell and M. Clements, "Objective measures
http://ulusaltezmerkezi.com/gurultulu-ses-parametrelerinin- of speech quality," NJ: Prentice-Hall, Eaglewood Cliffs, 1988.
iyilestirilmesi-ve-taninmasi/ . [23] J. H. L. Hansen and B. L. Pellom, "An effective quality evaluation
[7] M. B. Gülmezoglu. and A. Barkana, "Text-dependent speaker protocol for speech enhancement algorithms," Int'l. Conf. on Spoken
recognition by using Gram-Schmidt orthogonalization method," Proc. Language Processing, vol. 7, pp. 2819–2822 , 1998.
of IASTED Int. Conf. on Sig. Proc. & App., 438-440, 1998. [24] N. Kitawaki, H. Nagabuchi and K. Itoh, "Objective quality evaluation
[8] M. B. Gülmezoglu, V. Dzhafarov, M. Keskin and A. Barkana, "A for low bitrate speech coding systems," IEEE J. Select. Areas in
novel approach to isolated word recognition," IEEE Trans. on Acoustic Comm., vol. 6(2), pp. 262-273, 1988.
Speech and Signal Processing, vol. 7(6), pp. 620-628, 1999.
[9] M.B.Gülmezoglu, V. Dzhafarov and A. Barkana, "The Common
Vector Approach and its relation to Principal Component Analysis,"
____________________________________________________________________________________________________________
694
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Owing to the theory depend heavily on mathematical reconfigurable software using digital signal processing (DSP)
models in communication lessons, instructors find students lost techniques [3].
in all the equations and notations that given. Setting a lab There are several examples of SDR based education
component for students to put together the theory with practical courses. Domain analyses such as time, frequency, joint time-
implementations is not very easy in many times because it may
frequency, code, and modulation discussed in [4]. M-ary PSK
cost a large number of equipment that is not low-priced. Yet
there are still some solutions which reduce the cost in designing a modulation and demodulation implemented in [5]. AM
real-life communication laboratory and effective in teaching. We implementation presented in [6]. Five week communication
present the usage of NI-USRP Software Defined Radio (SDR) applications programme discussed in [7].Therefore, multi-
and Emona DATEx Add-on Board on digital modulation purpose usage of SDR in communication lessons will be
techniques that take part in communication engineering educatory and preliminary for real world.
education. Most commonly used digital modulation techniques; In this study’s first part we use NI-USRP 2922 supporting
Amplitude Shift Keying (ASK), Phase Shift Keying (PSK) and frequency range of 400 MHz to 4.4 GHz with SBX
Frequency Shift Keying (FSK) are analyzed in both devices. Both daughterboard. It’s omni-directional vertical antenna provides
modulations and demodulations of mentioned techniques are
3 dBi gain for two-way high bandwidth communication. A
done by experimentally.
Gigabit Ethernet connectivity provide data streaming to host
processors enables simultaneously sending up to 50 MHz RF
Keywords— Software Defined Radio, Communication
signal in and out of the USRP device [8].
Engineering Education, Digital Modulation Techniques, USRP,
DATEx Another way of using computer hosted device in
communication education is Emona DATEx. By using it
I. INTRODUCTION students can find the opportunity to experiment the real world
In general most of the students at communication signals in lessons.
engineering start learning communication systems with the The Emona DATEx, Digital Analog Telecommunications
theory and the principles. However learning the theory and Experimenter, is an add-on which could serve as a basic
principles solely does not help students to design training board in various concepts of communication lessons
communication systems properly. They need to acquire [9]. This add-on is fully integrated with the NI ELVIS II
experimental skills beside the theoretical domain. In order to platform and is controllable through the NI LabVIEW
bridge this gap instructors use several equipments such as development environment. All DATEx knobs and switches
function generators, oscilloscopes, spectrum analyzers that is are manually controllable or programmable from LabVIEW.
not affordable many times. The NI ELVIS II is an all-in-one electronics workstation
With the development of microprocessors and software that performs the functions of instruments which are
technology, traditional equipments began to transform into conventionally found in an electrical engineering laboratory.
computer hosted devices which are low-cost, flexible and The NI ELVIS II holds a removable prototyping board on
multi-functional. As an example, SDR, offers large number of which developers can build electrical circuits which connect
tools to teach signals and communications [1]. Limits of the different ELVIS's instruments to achieve a particular
traditional equipments, students being remained at the level of experimental goal.
pre-digital area and the motivation for SDR based In communication education, examples of DATEx usage
communication education is stated clearly in [2]. SDR refers limited. Practical aspects of Direct Sequence Spread Spectrum
that analog-to-digital (A/D) conversion is performed at the (DSSS) communication discussed in [10]. AM/FM/BPSK
antenna connector, with all following signal processing signal generating experiments over the internet is presented in
(upconversion and/or downconversion, intermediate- [11]. Time and frequency domain demonstrations discussed in
frequency (IF) filtering, demodulation, etc.) performed in [12].
____________________________________________________________________________________________________________
695
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
696
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
be used in an implementation are; 100 kHz sine/cosine, 2 kHz Depending the value of modulation index distortion on the
sine, 2-8 kHz digital. On the contrary in USRP students can received signal can be seen on the eye diagram easily.
use all the frequencies supported by daughterboard (400 MHz- As in Fig. 6 by changing symbol rate, alpha, filter type etc.
4.4.GHz). you can easily discover conversions in the modulated signal.
After mathematical theory, showing their importance on a real
III. FSK MODULATION APPLICATIONS life signal would be very beneficial for students.
FSK refers to a type of frequency modulation that assigns
bit values to discrete frequency levels. In other words
frequency does not fluctuate over symbol time but fluctuates
in discrete amounts in response to message. Amplitude
remains unchanged. This is a desirable characteristic for
improving the power efficiency of transmitters. In FSK
modulated signal consist of two different carriers;
____________________________________________________________________________________________________________
697
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
As seen clearly from the block diagram, setting FSK constellation will spread around the circle. If the problem is
demodulation is more difficult in LabVIEW than the one in related to magnitude, the constellation will spread off the
DATEx. As we mentioned before some of the functions used circle, either towards or away from the origin. These graphical
in block diagram are standart while designing real signals. techniques assist in isolating problems much faster than when
And their connections seems little bit complicated. But it is using other techniques. With the help of Modulation Toolkit
obvious that USRP gives us chance to radiate our signal in the Examples in the LabVİEW library, PSK transmitter
air and demodulating/recovering a signal with additive noise application can be implemented easily as shown in Fig. 10.
is not so simple.
These two signals have the same frequency and energy but
they have a correlation coefficient of −1.
A. PSK Implementation with USRP Fig. 10 Front Panel of PSK Transmitter Application
Our laboratory have two USRP radios, one for B. PSK Implementation with DATEx
transmission and the other one for reception. There are two
Switching halves of the PSK signal’s envelopes have the
possible configurations:
same shape as the message. So PSK generation and the
Both radios connected to single computer (running
recovery of the data can be handled by conventional Double
both the transmitter and receiver VIs) with a dual-port Gigabit
Side Band Suppressed Carrier (DSBSC) modulation and
Ethernet interface, or using a data switch.
demodulation techniques. The Sequence Generator module is
Each radio connected to a different computer; one
used for modelling the digital data and its SYNC output
running the transmitter VI, and the other, the receiver VI.
triggering the scope to provide a stable screen. With the help
of Multiplier module, PSK signal generated as shown in Fig.
We preferred the first one for our applications because it is
11.
better and faster for managing all the process in one screen.
Our experimental set up with USRP can be seen on Fig. 9.
Connection between the two USRP is provided with the
Multiple Input Multiple Output (MIMO) cable.
____________________________________________________________________________________________________________
698
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
699
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—In this study MUltiple SIgnal Classification (MUSIC) choose to reduce the effect of interference in FMCW channel
algorithm were used to reduce the effect of interference in the data.
frequency modulated continuous wave (FMCW) channel data.
Since the interference affects a very wide band including the II. MUSIC METHOD
frequency band of interest, the classical band pass filtering did not
reduce the noise level. The challenge was do develop an
The MUSIC algorithm is a frequency estimation technique
interference reduction algorithm that reduces the noise floor in the like Pisarenko’s method. In Pisarenko, only a single
frequency band of interest. MUSIC algorithm was chosen for this eigenvector is used and taken to be a set of autoregressive
purpose. The MUSIC algorithm is the one of the subspace method coefficients, whose zeros can be found analytically or with
which is frequency estimation technique and it is more useful for polynomial root finding algorithms. In contrast, MUSIC
reducing the in-band interference than band pass filter. assumes that several such functions have been added together,
so zeros may not be present. Instead there are local minima,
Keywords— In-band interference, MUSİC algorithm, band pass which can be located by computationally searching the
filter.
estimation function for peaks. MUSIC estimates the frequency
I. INTRODUCTION content of a signal or autocorrelation matrix using an
eigenspace method [8].
An in-band interference occurs when a part of transmission
band is simultaneously used by another application. Because of We assume the received data, x(n) is a random process
interfering signals and desired signal to be transmitted are in consisting of p complex exponentials in white noise with a
the same frequency band, the interfering signals aren’t variance of 2w . Rx is the MxM autocorrelation matrix of x(n)
suppressed in the receiver. The effect of in-band interference in with M>p+1. If the eigenvalues of Rx are arranged in
chirp modulated frequency modulated continuous wave decreasing order ( 1 2 ... M ) and if v1, v2, …,vM are
(FMCW) data is to raise the noise floor in the average power
delay profiles (APDP) [1, 2]. In channel sounding, although the the corresponding eigenvectors, then these eigenvectors can be
signal processing can be performed off-the-line, measurement divided into two groups: the p signal eigenvectors
campaigns are expensive and difficult to repeat. Since every corresponding to the p largest eigenvalues, and the M-p noise
channel data is valuable for a sound channel model, a technique eigenvectors that, ideally, have eigenvalues equal to 2w .
that alleviates the effect of in-band-interference in FMCW
The white noise variance by averaging the M-p smallest
mobile radio channel data is desirable. The basic method
eigenvalues as follow:
proposed to reduce the effect of in band interference have
included clipping the level of interference or put zero to data M
1
which has in band interference [1,3,4]. Also Minimum Norm 2w
M p k p 1
k (1)
Method (MN) [5] and Prony modelling [6] of data are used to
reduce the effect of in-band interference.
The MUSIC method frequency estimation function as
Noise subspace methods are used as a frequency detection
follows [7]:
methods. For example; MUSIC, Pisarenko harmonic
1
decomposition, minimum norm and Eigen Vector (EV) [7]. The P̂MU (e jw ) M (2)
methods that based on an eigendecomposition of the
H 2
e v i
autocorrelation matrix into two subspaces, a signal subspace i p 1
and a noise subspace. In this study MUSIC algorithm was
____________________________________________________________________________________________________________
700
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(b)
____________________________________________________________________________________________________________
701
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] Salous S., Gokalp H., “Medium and Large Scale Characterization of
UMTS Allocated Frequency Division Duplex Channels”, IEEE
Transactions on Vehicular Technology, 56, 5, 2831-2843-2483, 2007.
[2] Shearman E. D. R., Unsal R. R., “Compatibility of High Frequency
Radar Remote Sensing With Communication” Proc. Of the international
conf. on Radio spectrum conservation techniques, London, 1980.
[3] Brooker G. M., “Mutual Interference of Millimeter Wave Radar
Systems”, IEEE Transactions on Electromagnetic Compatibility, 49, 1,
170-181, 2007.
[4] Schipper T., Harter M., Zwirello L., Mahler T., Zwick T., “Systematic
approach To Investigate and Counteract Interference-Effects in
Automatiive Radars”, 9th European Radar Conference (EuRAD), 2012.
Fig. 4 Band pass filtering in the frequency domain [5] Gokalp H., Taflan G.Y., Ustun S., “Min-norm Method for Estimating
Delay and Doppler Profiles from FMCW Channel Data With In-band
Interference”, IET Electronic Letters, 40, 11, 799-800, 2010.
As it seen, after the band-pass filtering in a frequency [6] Gokalp H., Taflan G.Y., Salous S., “In-Band Interference Reduction in
domain, we saw that the crude data noise floor did not decrease. FMCW Channel Data Using Prony Modelling”, IET Electronic Letters,
Because when a desired data is taken by band pass filter, 45, 2, 132-133, 2009.
[7] Hayes M: H., “Statistical Digital Signal Processing and Modelling”.
interference is also taken.
John Wiley&Sons, Inc., 1996.
The reason of appear to have been cleaned signal in a time [8] https://en.wikipedia.org/wiki/MUSIC_(algorithm)
[9] Gokalp H., “Characterization of UMTS FDD Channels”, Science and
domain is the suppression of the high frequency component of Technology for the degree of Doctor of Philosophy, Department of
the signal which is out of the desired band. Electrical Engineering and Electronics, Manchester, United Kingdom,
2001.
After the band pass filter application, the MUSIC algorithm
is used to reduce in-band interference. MUSIC algorithm is
applied different values of p and M. The range of appropriate p
values were between 300-330 and M values were between 400-
450. The best result is taken by p=320 M=440 and so noise floor
is reduced 8 dB approximately with that p and M values (e.g.
Fig. 5).
Relative power (dB)
Crude data
Cleaned data
____________________________________________________________________________________________________________
702
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—There has been a substantial growth in the use of mobile will be around 130.000. Because there is a growing number of
communication services over the last few years. As a result of this base stations being placed into crowded places; measuring,
growth increase in the number of base stations, and exposed evaluating, the levels of EMR and controlling their compliance
electromagnetic radiation (EMR) levels have become inevitable. with standards/limit values has become more crucial than
Considering the public debate about possible health hazards
before. Therefore detrimental effects of EMR on human health
caused by EMR, in this study, evaluating the effect of
establishment of 4.5G systems on existing EMR levels is aimed. have been the subject of many researches [1-9] in the last
For this purpose; the total 536 EMR measurements were decade.
conducted in Samsun before and after 1 April 2016 using PMM There are international standards and limits on effects of
8053 EMR meter. It is seen from the measurements that maximum
electric field strength (Emax) is 6.32 V/m while the maximum
EMR on human health. The limits are recommended by an
average electric field strength (Eavg) is 3.26 V/m which are both international commission ICNIRP which is recognized by
below the limits (28V/m, 21V/m the minimum values for 100kHz- World Health Organization (WHO). The limits of electrical
3GHz band) that are determined by the International Commission field are shown in Table I based on ICNIRP and ICTA
on Non-Ionizing Radiation Protection (ICNIRP) and Information guidelines on exposure limits [10-11].
and Communication Technologies Authority of Turkey (ICTA).
The mean of Emaxs and Eavgs are 1.2111V/m, 0.5892/m before 4.5G,
and 1.8449V/m, 0.6628 V/m after 4.5G respectively. The results Table I Reference levels for general public exposure to time-varying
demonstrate that the deployment of 4.5G yielded up to 52.3% electric fields for ICNIRP and ICTA
increase in Emax while 12.5% in Eavg values. Even though the
measured E levels are below the limits; there is a significant
increase in them in parallel with 4.5G. It can be also predicted Electric field strength (V/m)
from the measurement results that E levels are likely to increase Frequency range
over forthcoming years, and exceed the lower limits (4 V/m) (MHz)
applied by some countries e.g. Switzerland and Italy. ICNIRP ICTA
0.010 – 0.15 87 65.25
Keywords—Electric field strength; electromagnetic (EM)
measurement; EM pollution; statistical analysis. 0.15 - 1 87 65.25
1 - 10 87/f 1/2 65.25/f 1/2
I. INTRODUCTION
10 - 400 28 21
Electromagnetic (EM) waves are radiated from many
sources, both natural and man-made, that produce 400 - 2000 1.375f 1/2 1.03f 1/2
electromagnetic pollution. In parallel with technological 2000 - 60000 61 45.75
developments, there has been a dramatic growth in the use of f is frequency in MHz
cellular systems which based on base stations. Because each
base station is an electromagnetic radiation (EMR) source, with
the increase in the number of base stations, exposed EMR
The limits are given for exposure averaged over a six
levels have also increased. Since each base station works within
minute interval. Each country has its own determined limits.
a limited geographical region and for limited number of users,
The USA and some European countries use limits determined
mobile system operators must install more base stations in
by the ICNIRP, while other European countries like
order to meet growing demand for multimedia services, and
Switzerland and Italy use 1/10 of the ICNIRP’s values as a
communicating from any place. There are around 100,000 base
limit. Turkey applies limits that are 75% lower [11] than those
stations in Turkey right now, but with the establishment of 4.5G
suggested by the ICNIRP [10]. There are currently three mobile
systems on 1 April 2016, the estimated number of base stations
____________________________________________________________________________________________________________
703
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
communication operators in Turkey and they use 2G (second III. MEASUREMENT RESULTS
generation), 3G (third generation) and 4.5G (fourth generation) The maximum E (Emax) and the average E (Eavg) which were
systems. According to the [11], the limits are 30.9 (V/m) for measured at 67 different locations are given in Fig. 2.a and Fig.
900 MHz base station, 43.7 (V/m) for 1800 MHz base station, 2.b respectively. Reasons of various E levels may be: the
45.75 (V/m) for 3G systems, which is 2100 MHz, and also number of base stations in measurement area, output powers of
45.75 (V/m) for 2600MHz base station. base stations, distances from the base stations, whether there is
line of sight (LOS) between the units or not, and the number of
In this study, to evaluate the effects of newly established users. As seen from the figure that the Emax is 6.32 V/m which is
4.5G system on EMR levels, extensive EMR measurements measured at 23th location (measurement time 1). The maximum
were performed before and after 1 April 2016 in Samsun city Eavg is also obtained at 23th location (measurement time 4) as
center, at 67 different locations. Statistical properties of the 3.26 V/m accordingly.
measured data are determined. The changes in EMR levels
which occurred with the deployment of 4.5G systems are
visualized through maps.
____________________________________________________________________________________________________________
704
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Table II Statistical properties of the measured data In order to have better visualization changes in EMR levels
Emax (V/m) are transferred on a map using MapInfo and Emaxs and Eavgs are
shown in Fig. 3 and Fig. 4 before and after 4.5G respectively.
Max. Mean Std.
-a-
M1 6.32 1.2001 0.9167
M2 4.49 1.2221 0.8349
M3 5.59 2.1607 1.2592
M4 5.11 1.5291 0.9691
Eavg (V/m)
Max. Mean Std.
M1 2.7800 0.5718 0.6488
M2 3.0600 0.6066 0.6034
-b-
M3 2.8800 0.6940 0.5460
M4 3.2600 0.6316 0.6263
-b-
Fig.3 Probability density function for a) Emax, b)Eavg Fig.5 MapInfo picture of Eavg for a) before b) after 4.5G
____________________________________________________________________________________________________________
705
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
IV. CONCLUSIONS
In this study, in order to determine the effect of deployment
of 4.5G systems on EMR levels, extensive measurements were
performed at 67 different locations in Samsun city centre before
and after 1 April 2016. According to 536 measurements the
maximum total electric field strength is 6.32 V/m while the
maximum average electric field strength is 3.26 V/m. The
results demonstrate that the deployment of 4.5G yielded up to
52.3% increase in Emax while 12.5% in Eavg values.
REFERENCES
[1] Ç. Kurnaz, “An Emprical Modelling of Electromagnetic Pollution on an
University Campus”, The Applied Computational Electromagnetic
Society Express Journal, vol.1, no.2, pp.76-79, 2016.
[2] B. K. Gül, Ç. Kurnaz, and B. K. Engiz, “Measurement and Evaluation
of Electromagnetic Pollution in Ondokuz Mayıs University Kurupelit
Campus in Samsun, Turkey”, Third International Conference on
Advances in Information Processing and Communication Technology,
pp.80-84, Rome, Italy, 2015.
[3] A. Mousa, “Electromagnetic radiation measurements and safety issues
same cellular base stations in Nablus”, Journal of Engineering Science
and Technology Review, vol. 4, no. 1, pp. 35-42, 2011.
[4] O. Genç, M. Bayrak, and E. Yaldız, “Analysis of the effects of GSM
bands to the electromagnetic pollution in the RF spectrum”, Progress in
Electromagnetics Research, vol.101, pp.17-32, 2010.
[5] S. Miclaus, and P. Bechet, “Estimated and Measured values of the
Radiofrequency Radiation Power Density around Cellular Base
Stations”, Environment Physics, vol. 52, no. 3–4, pp. 429–440, 2007.
[6] L. Seyfi, “Measurement of electromagnetic radiation with respect to the
hours and days of a week at 100kHz–3GHz frequency band in a Turkish
dwelling”, Measurement, vol.46, no.9, pp.3002-3009, 2013.
[7] P. Baltrenas, and R. Buckus, “Measurements and analysis of the
electromagnetic fields of mobile communication antennas”,
Measurement, vol.46, no.10, pp.3942-3949, 2013.
[8] M. Koprivica, V. Slavkovic, N. Neskovic, A. Neskovic, “Statistical
Analysis of Electromagnetic Radiation Measurements in the Vicinity of
GSM/UMTS Base Station Installed on Buildings in Serbia”, Radiation
Protection Dosimetry, pp.1-14, 2015, doi: 10.1093/rpd/ncv372.
[9] T. Karadag, M. Yüceer, and T. Abbasov, “A Large-Scale Measurement,
Analysis And Modelling of Electromagnetic Radiation Levels in the
Vicinity of GSM/UMTS Base Stations in an Urban Area”, Radiation
Protection Dosimetry (2015), pp. 1–14, doi:10.1093/rpd/ncv008.
[10] ICNIRP Guidelines, “Guidelines for Limiting Exposure to Time-
Varying Electric, Magnetic, and Electromagnetic Fields (up to
300GHz)”, International Commission on Non-Ionizing Radiation
Protection, Health Physics vol.74, no.4, pp.494-522, 1998.
[11] Information and Communication Technologies Authority of Turkey,
“Ordinance change on By-Law on Determination, Control and
Inspection of the Limit Values of Electromagnetic Field Force from The
Electronic Communication Devices According to International
Standards”, Law no.29497, 9 October 2015.
[12] www.pmm.it/docs/8053en1001.pdf
____________________________________________________________________________________________________________
706
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, in order to determine the signal strengths In mobile communication systems, modulated
of the existing cellular system operators for 2G, 3G and 4G electromagnetic waves arrive at the receiver either Line of
services drive test measurements were conducted on a specific Sight (LOS) path or through different paths. Along these paths
route on Ondokuz Mayis University (OMU) Kurupelit Campus. the transmitted signal undergoes reflection, scattering,
The synchronized measurements were performed using nine same diffraction because of the surrounding objects (buildings,
brand and model smartphones with “Android” operating system mountains, trees etc.), and the received electric field strength
and “Netmonitor” application. It is seen from the measurements varies as a result. This phenomenon is known as multipath
that the received signal strength can change depending on the fading, and if the receiver position moves the destructive effect
location, frequency, line of sight, and base stations’ output power.
s of multipath becomes more complicated [5].
2G signal strengths are significantly higher than 3G/4G.
Considering all operators; the signal strengths vary from -51dBm In Turkey, currently 2G (second generation), 3G (second
to -87dBm for 2G, while from -51dBm to -103dBm, and from - generation) and 4G (fourth generation) cellular systems are
67dBm to -130dBm for 3G and 4G respectively. The average signal used, by means of three different Cellular System (CS)
strengths of 2G, 3G and 4G are -65.97dBm, -80.01dBm and -
operators. In order to maintain customers’ loyalty and gain new
94.52dBm. The quality of signal is higher than the minimum limit
(<-100dBm) determined by Information and Communication customers the CS operators mostly emphasize on “coverage
Technologies Authority (ICTA) of Turkey for 2G at all area”, “signal power”, and “signal quality” in commercials.
measurement locations. However, for 3G lower than the limit at Some recent customer surveys on CS operators show that a high
8.33% of all measurement locations for operator C while for 4G level of satisfaction is strongly correlated with signal strength.
69.71% for operator A. Using these results help to have foresight Therefore CS operators must improve the provided signal
on needed improvements/enhancements. quality in order to meet costumers’ demands and fulfill the
requirements determined by ICTA. For this reason, in this study,
Keywords—Cellular System; Signal Strength; Drive Test measuring the signal strengths of existing three operators
Measurement; Netmonitor. (named as A,B,C) in Turkey for 2G, 3G and 4G systems on a
specific route on OMU Kurupelit Campus, transferring them on
a map are aimed.
I. INTRODUCTION
Communication has an important role in human life, and has II. MATERIAL AND METHOD
been developing each day in parallel with the technological In this study, the signal strengths of 2G, 3G and 4G services
improvements. As a result of these improvements many new of the three CS operators were measured, and the results were
generation communication devices have been developed and transferred on a map. In measurements nine same brand and
presented. Mobile communication systems especially mobile
model smartphones with “Android” operating system and
phones have become the most common tool of communication
“Netmonitor” application, and nine sim cards whose three are
over recent years.
for each CS operator were used. After the determination of
In cellular systems, the coverage area is divided into a measurement route, required mobile network settings were
number of elementary areas called cells, and a base station is adjusted then Netmonitor application was started. According
placed center of the cell. Each active mobile unit within a cell the type of connection mode measured, the network mode of
uses assigned frequencies to communicate with the serving the each smartphone was adjusted to “2G only”, “3G only”, or
base station. With the movement of mobile user the cell “LTE only”. In order to determine the exact measurement
boundaries may be crossed from time to time. In such case new locations GPSs’ of all phones were turned on. Drive test
channels are assigned, and call could continue without any measurements were conducted the busiest times of a day, at a
interruption. Theoretically each cell is hexagonal; and the
speed of 30km/h, during 25min. with 5sec. intervals. The signal
design of a cellular system begins with the classification of the
quality of each operator was measured in terms of dBm.
coverage area [1-2]. With the growth in the capacity of mobile
communications, the size of a cell is becoming smaller and Fig.1 shows an image for a measurement conducted with
smaller: from macro cell to microcell and to picocell [3-4]. Netmonitor application. Netmonitor application which is an
____________________________________________________________________________________________________________
707
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Android based network software, shows the information of TABLE I Signal strength equivalency table
connected base station, signal strength, and location instantly,
and save these in CLF (list) as well as KML (map) format. Fig.2 dBm equivalent Classification
a and b illustrate examples of screen shots of 2G and 3G
measurements at a measurement location respectively. In figure, -101 or less Very weak
Operator shows the connected CS operator’s MCC-MNC
number and name, Type represents the connection type, LAC -100…-91 weak
defines the connected cell number while CID indicates
connected sector number. RNC also indicates radio network -91…-81 average
controller.
-80 or more Good
III. RESULTS
Drive test measurements that performed for 2G, 3G and 4G
systems using Netmonitor are given in Fig.3.a, b, c for the
operators A, B, C respectively. As seen from the figure that 2G
signal strengths are significantly higher than those for 3G and
4G. It is seen from the 2G measurements that neither very weak
nor weak signals concurred, and the operator C has worst signal
strength. The main reason of this case may be; the change in the
signal strength inversely proportional to frequency (Friis
transmission equation [5]), and using 1800 MHz for 2G while
Fig. 1. Netmonitor application and its operation the others use 900 MHz. It is also seen from the measurements
that the received signal strength can change depending on the
-a- location, frequency, line of sight, and base stations’ output
power for all operators.
The overall assessments for the measurements are given in
Table II. As seen from the Table that the average signal
strengths of 2G are -63.36 dBm for operator A, while
64.88dBm and -69.69dBm for operator B and C respectively.
Corresponding signal strengths of 3G are measured as
-85.27dBm, -76.67dBm and -78.33dBm. As a result of newly
establishment of 4G; very weak signals are recorded at many
location whose averages are -104.21dBm, -89.55dBm, and
-89.80dBm for the operators A, B and C respectively.
-b-
TABLE II Overall assessments of measurement results
____________________________________________________________________________________________________________
708
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
-b-
-c-
-b-
-c-
____________________________________________________________________________________________________________
709
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
-a- -a-
-b- -b-
-c- -c-
Fig. 5. Netmonitor measurement results of operator B for a) 2G, b) 3G, c) 4G Fig.6. Netmonitor measurement results of operator C for a) 2G, b) 3G, c) 4G
____________________________________________________________________________________________________________
710
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
711
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, the mathematical equations of five and rotor position. That's why SRM’s need a strong controller
phase segmental switched reluctance motor (SARM) was in order to produce maximum torque [5].
established by using basic electrical motor model. This five The difference of new 5-phase bipolar excitation SARM
phases SARM has a different rotor and excitation structure than from classic SRM is that the rotor structure has been altered
classical ARM, and its phase currents and magnetic field completely and contains variations in excitation method. In
variations with respect to inductance were calculated by means case of being used of the same resource, flux distributions
of magnetic equivalence circuit. The value of inductances was produced by the winding of both engines per phase and torque
calculated by stating the visible inductance profiles of different
amounts produced depending on this show changes due to
phases instantaneously in order to emphasise the difference
between two motors. In addition, the momentum equation of inductances of the windings and the mutual inductance value
SARM was obtained by determining the situations of different occurs between phases [6]–[7].
phases. The inductance values were calculated by assuming the Besides the successful results obtained in motor structure
current applied to phases created from an ideal current source and performance of classic switched reluctance, in also the
and by determining the situations of different phases. It was studies on switched reluctance motors (SARM) based on
revealed that the SARM produced much more momentum than segmental and two-phase energizing essence forming the basis
the classical ARM having the same phase number by using a of this study important progresses have been made. In order to
mathematical model. model the non-linear structure of SARM and in case of the
As a result of this study, it was understood that the magnetic flow
installation of motor, different structure were suggested by
equations depend not only on the function of rotor position, but
also the changing current many researchers in motor and actuator circuit design in order
to eliminate the fluctuations in the current varying due to time.
The first study about this subject was conducted by
Keywords— Switched Reluctance Motor, Mathematical Model, Lawrenson and Agu in literature in 1964 [8]. Mecrow et al.
State Equation examined the performances of SARM’s with various polarity
I. INTRODUCTION and winding structure [9]. Uygun et al. performed the analysis
of linear model of magnetic circuit by designing U-type
Switched reluctance motors (SRM) are electrical machines segmental type SRM [10]. Belhandi and Kolli showed the
with simple structure but superior properties which converts comparative simulation results of speed and torque parameters
electrical energy into mechanical energy with changing of the classic SRM and SARM by using the finite element
reluctance effect changes and which are in need of convertor method [11]. In other studies, the flux distribution topologies
power electronics while performing this action. Although of SARM was developed on magnetic structure by changing
SRMs first put forward in the 1800s, the driving motor pole, rotor and winding structures and engine driver
depending on developments in power electronics in the early optimizations were arranged.
1960s has become possible. Although it is the engine type In this study, a control algorithm including PI control for a
based on the old, the examination and investigation of these new segmental switched reluctance motor with bipolar
engines are new. In particular, the studies on this motor are excitation, 5 phases and 10/8 pole configuration, which was
seen to be focused after 1980 [1] [2] [3]. designed as segmental rotor (SARM), and entered in the
ARMs are increasingly being used in changing-rate and literature has been developed and presented.
servo-type industrial applications due to mechanical structures Also in this study, the actuator circuit including PI control
and low rotor losses [4]. These motors have many different option developed for controlling bipolar driven segmental
application areas such as electric cars, spacecraft, household rotor and computer-controlled management, monitoring
appliances, radars, automatic door systems and water pumps. interface will be discussed.
The most important characteristic of SRM’s is having
magnetic circuit in non-linear form depending on phase circuit
____________________________________________________________________________________________________________
712
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
II. STRUCTURAL FEATURES OF BIPOLAR EXCITED Thus, torque value produced per Nm/kg weight taken into
SEGMENTAL ROTOR 5-PHASE AND 10/8 SWITCHED account in the engines, Nm / kg will be automatically
RELUCTANCE MOTOR increased. In the engine, the two-phase energizing method was
In this section, it has been mentioned briefly the structural used instead of the single phase energization method used in a
features of the new design segmental type of engine belonging conventional SRM. The reason for using such a method is to
to SRM. see that the magnetic poles of the engine follow the short flux
In Fig. 1. (a), newly designed bipolar excited segmental paths when the windings are properly installed and
type SRM using short flux paths and the cross sectional view simultaneously two phases of SARM is energized. This case
of 5-phase classical switched reluctance motor with 10/8 (reduction of flux path) will lead the reduction in both sheet
configuration to be used in performance comparisons were losses and the temperature of motor sheets.
given. The case that occurs in case of being energized of this As it can be understood in Fig. 2 (b), segmental SRM
phase is shown in Fig. 2 (b). operates by using short current paths instead of long flow
paths unlike conventional SRM. Since in every time, two
phases remain with power, it is clear that common inductance
occurred between the phases and this case will create
complexity for the finding of non-linear models of the engine.
Ga Gb Gc Gd Ge
+ + + + +
Φa ua Φb ub Φc uc Φ d ud Φ e ue
- - - - -
up
Fig. 3. Magnetic Equivalent of Segmental ARM
(a) (b)
Fig. 2. (a) Cross-section representation of new SARM In Fig. 3, magnetic equivalent circuit belonging to 5-phase
(b) Only when A and B phases are energized to form suitable polarity, the engine is given. Here, e represents opposite electromotor
state of flux paths formed by new SARM power, ϕ represents flux, G represents magnetic permeability
of phases and Up represents voltage drop. By using this
____________________________________________________________________________________________________________
713
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
magnetic equivalence and basic electric machines, the Flux connections belonging to each phase can be
following processes have been realized.. expressed as follows;
According to ARM basic knowledge, when angular change
is considered, it is possible to write the equivalent moment x N x (8)
expression for the moment produced by the engine depending
on the co-energy produced by the motor as seen in Fig. 1: If it is used for the Equity 4 flux connection, it is written as
follows;
( ,)
( , )=− ( = ) (1)
If the moment produced by SARM is calculated from
a b c d e 0 (9)
general moment equity by using Equity 1, the following is If the magnetic permeability of the phases is expressed as
obtained; a function of phase inductances, (as being x=a, b, c, d, e), it is
obtained as follows;
1 dL dL dL dL dL dM dM dM
T aa ia2 bb ib2 cc ic2 dd ib2 ee ie2 iaib ab iaic ac iaid ad
2 d d d d d d d d Lx
Gx (10)
dM dM dM dM dM dM
iaie ae ibic bc ibid bd ibie be icid cd icie ce idie de
dM N2
d d d d d d d When flux connections are widened depending on this
(2) expression (as being x=a, b, c, d, e), it becomes;
Here, represents the produced moment (Nm), L
represents overlapped location phase inductance (mH),
represents rotor location angle (degree), M represents ex u p
x Lx ( ) (11)
common inductance (mH), i phase flow (amper).
N
The following equity can be written according to Kirşof’s
flux rule: up
An arbitrary flow is described as ip , (as being x=a,
a b c d e 0 (3) N
b, c, d, e);
x (as being x=a, b, c, d, e) represents phase fluxes.
x Lx ( ) (ix i p ) (12)
Depending on magnetic conductivity, flux change belonging
is obtained. If this equity obtained is places in the Equity
to each phase depending on conductivity (as being x (x=a, b, 9., it becomes;
c, d, e )) is expressed as in Equity 4.:
La ()(ia ip ) Lb ()(ib ip ) Lx ()(ix ip ) Lc ()(ic ip )
x Gx ( )ux (4)
Ld ()(id ip ) Le ()(ie ip ) 0
Here, the thing expressed is Gx (as being x=a, b, c, d, e)
(13)
represents magnetic permeability of phases while ux (as being If the value of i p is taken from this equity, it is obtained as
x=a, b, c, d, e) is representing voltage drop. Besides, the follows
values of the sources are calculated as in Equity 6.:
La ( )ia Lb ( )ib Lc ( )ic Ld ( )id Le ( )ie
ex Nix (5) ip
La ( ) Lb ( ) Lc ( ) Ld ( ) Le ( )
ex (as being x=a, b, c, d, e) represents the value of the (14)
resource, N represents the wrapping number belonging to each If the Equity 10 is rearranged according to Equity 12,
Equity 15 is obtained for phase A:
phase and ix (as being x=a, b, c, d, e) represents flow
belonging to each phase. If it is considered as the value of up La2()
potential; a La() ia
La () Lb () Lc () Ld () Le ()
ux ex u p (6)
L ()L ()i L ()L ()i L ()L ()i L ()L ()i
is obtained. With the widened expression, phase fluxes can be a b b a c c a d d a e e
calculated as in the Equity 7.; La () Lb () Lc () Ld () Le ()
(15)
x G x ( ) ex u p (7)
____________________________________________________________________________________________________________
714
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
This expression can be written as the same for the other TABLE I
THE RUNNING STATUS OF PHASES DEPENDING ON THE
phases. As it is easily understood from the expression, core
ROTATION ANGLES OF PHASES
inductance value belonging to phase A is seen as follows;
0-18 18-36 36-54 54-72 72-90
La2 ( ) A Phase 0 + 0 0 -
Laa La ( ) (16)
La ( ) Lb ( ) Lc ( ) Ld ( ) Le ( ) B Phase 0 0 - 0 +
Similarly, common inductance values are found between C Phase - 0 + 0 0
phases A and B as;
D Phase + 0 0 - 0
La ( ) Lb ( ) E Phase 0 - 0 + 0
M ab (17)
La ( ) Lb ( ) Lc ( ) Ld ( ) Le ( )
A. Visible Inductance Profile of SARM
In order to be comprehended of running of the Segmental
ARM more clearly, it is required to examine visible
inductance profiles. In this method, thinking that the phases of
segmental ARM are run with ideal flow source and by
revealing the status of different phases at each moment, the
values of inductances will be calculated. In this calculation
method, phase switches were ignored and phases were
connected as star like in Fig. 4.
Ia A Fazı
Ib B Fazı
Ic C Fazı Fig. 5. When only B and C phases are energized, the status of equivalent
circuit
Equivalent with the simple form of this circuit is seen in
the circuit in Fig. 6.
Id D Fazı
Ie E Fazı
Fig. 4. When ideal flow sources are used, the equivalent circuit of Fig. 6. Ideal equivalent circuit of SSRM when phase B and C are
segmental ARM excited
At any running moment of the designed engine, the phases When B and C phases are energized, the simplified status
given in Table I will run together. Accordingly, if only B and of the equivalent circuit, the following equity group is
C phases are thought to be energized at the first sight, the obtained;
circuit in Fig. 5 will be obtained.
____________________________________________________________________________________________________________
715
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
716
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] R. Krishnan, Switched Reluctance Motor Drives Modeling, CRC Press,
2001, pp. 1–3.
[2] E. Hamid, K. S. Virendra, A. Chandra, A. Kamal., "Performance
Simulation of Switched Reluctance Motor Drive System Operating With
Fixed Angle Control Scheme", 7 th International Conference on
Modeling and Simulation of Electric Machines, Converters and Systems,
Electrimacs, August 18-21, Montreal Canada, 2002.
[3] V. Horacio, K.P. Joey, "A New Simplified Mathematical Model for a
Switched Reluctance Motor in a Variable Speed Pumping Application",
Science Direct Mechatronics, vol. 14, pp.1055–1068, Nov. 2004.
[4] O. F. Bay, C. Elmas, "Modeling of the Inductance Variation and Control
of the Switched Reluctance Motor Based on Fuzzy Logic", Intelligent
Automation and Soft Computing, vol. 10(3), pp. 233-246, 2004.
[5] R. Mutlu, "Modelling of the Short Flux Path Mutually Coupled
Switched Reluctance Machine", Doctorate Thesis, The Graduate
Faculty of Rensselaer Polytechnic Institute, USA, 2004.
[6] D. Uygun, "Switched Reluctance Motor Design and Control", Doctorate
Thesis, Institute of Science and Technology, Gazi University, Turkey,
2012.
[7] G. Bal, D. Uygun, "A Novel Switched Reluctance Motor with U-type
Segmental Rotor Pairs: Design, Analysis and Simulation Results",
ICCESSE 2010 International Conference on Computer, Electrical, and
Systems Science, and Engineering, April 28-30, pp.686-690 Rome,
Italy, 2010.
[8] P. J. Lawrenson, L. A. Agu, "Theory and performance of polyphase
reluctance machine", Proceedings of the. Institution of Electrical.
Engineering, vol. 111(8), pp.1435–1445, 1964.
[9] B. C. Mecrow, E. A. El-Kharashi, J. W. Finch, A. G. Jack,
"Preliminary Performance Evaluation of Switched Reluctance Motors
With Segmental Rotors" IEEE Trans. Energy Conversion, vol. 19, pp.
679–686, 2004.
[10] D. Uygun, G. Bal, I. Sefa, "Linear Model of a Novel 5-Phase Segment
Type Switched Reluctance Motor (Bimonthly)", Elektronika Ir
Elektrotechnika, vol. 20(1), pp. 3–7, 2014.
http://dx.doi.org/10.5755/j01.eee.20.1.6159
[11] M. Belhadi, A. Kolli, G. Krebs, C. Marchand, "Evaluation of Torque-
Speed Curve of Switched Reluctance Motor with Segmantal Rotor",
2012 IEEE Electrical Machines (ICEM), 2012 XXth International
Conference, pp. 250–255
____________________________________________________________________________________________________________
717
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— This paper aimed to illustrate the use of the Lean Six practices of the Lean Six Sigma was going ahead to obtain
Sigma (LSS) methodology in power plants maintenance. LSS has goodness and generation improvement in power plants. The
become the indeed method for process performance studying of LSS and practice it gives us more opportunities to
improvement in many industrial companies and LSS has been improvement the equipment operation. These opportunities to
gradually increasing its acceptance in electric power generating be noticed for everything for improvement and confirm the
that confronts challenges, such as, how to decrease maintenance
requirement for study driving to improvement and placing
time, increase quality, better use resources etc. Consequently, the
quality of any process may be enhanced by utilizing Lean Six methodologies that like LSS [5]. Now the question is when
Sigma in any climate and it has the potential to offer a useful deciding the maintenance procedure when the lean six sigma
approach in term of improving quality and reducing costs. Lean start? And lean six sigma application starts. All the current
shows us an exclusive method that helps identify possible studies are leading to when power plants want to minimize
improvement areas. The study includes the plant maintenance these equipment's maintenance time and cost, it must apply
cycle time from request writing to work finished. Improvement the lean six sigma methodology. In the lean six sigma
opportunities were identified from a high–level value road map. methodology waste is a very important factor affect on the
The Define, Measure, Analyses, Improve and Control maintenance process of improvement equipment operation [6].
(DMAIC) approach was applied to a treat the identified
opportunities for improvement. The result shows that the
maintenance time was reduced from 180 days to 95 days. II. SIX SIGMA STEPS (DMAIC)
to improve and increase the quality the maintenance practices
Keywords— Lean Six Sigma, maintenance Time, DMAIC, CTQs,
DPMO
to obtain a high level of power plant operation, must arrive at
the standard specification by using six sigma (Define,
I. INTRODUCTION Measure, Analyses, Improve, Control) as shown in Fig. 1.
We are sense the huge development of the maintenance These steps are[7, 8]:
programs. It is leading to mend raising power plant reliability 1) Define: First of all, the problem should be defined
and assuring plant capacity by applied the deferent and requirements of the customers need to be
maintenance kinds. The LSS methodology was firstly identified. This step is set the guesses to development
developed and implemented by Motorola in 1985[1, 2]. the maintenance requirements and concentrate on the
Motorola knew that there was a modality of betterment applications of six sigma methodology but not forget
which would be classification to the five phases of problems the customer demands.
treatment applied by manufacturers, i.e., DMAIC. DMAIC is 2) Measure: This step is illustrating the errors in
the mainly five steps for every maintenance planners of the maintenance applications and schedules the
power plant. information to use it to confirm the improvement
Lean Six Sigma is a controlled, depends on the data targets.
methodology used to remove and minify the practices hence 3) Analysis: this step is important in every program to
the job faults and squandering (time and sources) [3]. minimize the errors, and studying the information
To obtain Six Sigma goodness and practices of a job must collected and put it in tables. And in next period set
be no more than 3.4 defect per million opportunities (DPBO). the targets and planning to reach the wanted level.
The researchers submit new ways to monitor data and analyse 4) Improvement: in this step planners confirm the
to appear (DPMO). Six Sigma essential offer amount is that changes in the maintenance practices and measured
rules for maintenance practices improvement, statistical it's affecting on the equipment quality (CTQ), and
process, alertness in the work, and maintenance management find acceptable rules to minimize change errors and
framework central on high-return improvement power plant measure affecting on (CTQ) to find the best solutions.
generation result in persistent improvement and considerable 5) Control: this is the final step of six sigma
economic earning [4]. When the power plants applied the Six applications, in it the planners find the advantages of
Sigma, it can discover the factors affect on the maintenance in applied six sigma and schedule the results of the
power plant operation, recognize the optimal scale of practices.
allowance and betterment chances. In this paper, the DMAIC
____________________________________________________________________________________________________________
718
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Today Six Sigma methodology applied widespread in the Baiji thermal power plant in Iraq through practice lean six
world such us (engineering, markets, industrial, management). sigma steps from taking the maintenance requests during one
Particularly, the widespread applications of Six Sigma were year (2012) and the requests rate about 180 day. The purpose
possible because organizations were able to articulate the of maintenance requests in the power stations is too useful to
benefits of Six Sigma presented. confirm the maintenance efficiency and productivity.
Essentially the request helps maintenance technical to obtain
vital power plant data and set every equipment work. The
Problem request avoids an incompatible utilization of oral statement.
define definition The principles of request system aimed to make a regular
Test stream
shape for data collected and work to equipment rehabilitation.
For this work, we suggest integrated methodology of lean
six sigma. In first we must put the road map for this study to
define the opportunities amendment and progressing it to the
Decision
measure control maximum level. And we must fix all the changes in the road
map without pay attention to defects type. Six sigma
methodologies must apply in this process on these chances.
Result Demand
Six sigma methodologies is a strategic way to keep the
view maintenance on the right road. In this paper, we can use lean
six sigma methodology practice or one of them depending on
the errors type[10].
finished maintenance work in the set time. The six sigma 60,00%
methodology sight have the measurement to decrease errors in 40,00%
2000
the operation changes and confirm it's in maintenance works 20,00%
quality and spenders. Lean six sigma is represents the strong
0 0,00%
sets to increase the speed maintenance works and decrease the 1 2 3 4 5 6
waste to obtain the better solution of maintenance engineering
Seri1 3100200014501100 867 500
management [3].
Seri2 49,3 69,4 80,5 91,5 95,7 100,
V. CASE STUDY
In this paper, we will apply the lean six sigma methodology to Fig. 2 Pareto chart of maintenance days(Seri 1. maintenance process
decrease the maintenance request time. Our work executed in Seri 2. request numbers)
____________________________________________________________________________________________________________
719
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
2) Step 2: Measure. Each CTQ’s of measurable enable study the reformative process are scheduled. It is
us to confirm work basis. In this step, to confirm the controlled maintenance works and monitor the
maintenance requirements we must collect the progress of work by the maintenance departments
information around the happened process in the chiefs and submit periodic reports on the progress of
maintenance wait and anticipation and ultimately business to the maintenance head manager in the
valiantly the measurement by using modern technical power plant
to measure and detect the errors such as Pareto chart.
3) Step 3: Analyze. The concentrate in an analysis for VII. RESULTS
the practice holes in every procedure in CTQs. the These papers focus on the major faults in the mechanical,
intention in this steps dividing the defects to "do it, electrical and other taking into account the resulting acts of
not do it ". We seek for the style of lean behavior also disparate maintenance process to get the desired results of the
the style of best behavior in power plant right request planning procedure. This work and collect data
maintenance. Such must in this style the hard work started in 2012; after 6 months showed signs of success
continued on effort confirm and navigable. As well, by %50 as we see in Fig. 3. As in Fig. 3, we were able to
we take all the input and output variation between the reduce the average time required for maintenance from 180
input and output of CTQs. moreover, we tighten to days to 105! Furthermore, we have improved the response to
concentrate to some subjects and find an appropriate urgent requests for maintenance at the station and overcome
mechanism for data analysis. Also, we found the the difficulties related in terms of adjusted a time for
different reasons of the troubles. When using the maintenance, as it is shown in fig 4. Follow-up development
method of statistical analysis at this stage, depending and success of our work to develop a system to identify
on the type of items will wait a few days. actions that need further scientific studies.
4) Step 4: Improve. If we wish to examine closely a
strong improvement in the process must VIII. CONCLUSION
underestimate the ability gaps in CTQ steps. Also, Lean Six Sigma is not an enchantment solution that you can
we support every supposition of cause root to the use for quick success. It is a methodology, which has to be
errors and confirm the substitution solutions and in a slowly integrated into the culture of the organization. It can be
parallel way obtain the better solution. applied to any process and we as Maintenance Plants need to
Work on these securities began based on the data make the best out of this methodology.
obtained in 2012. (Seri 1 the requests life Improvement .)
The recommendations, as described below:
• Resettlement the planners and managers offices to 250
maintenance location in power plants.
• Identify the people responsible for managing the 200
stores that contain maintenance materials. 150
• Give adequate space for planners to management
raw Material. 100
• Writing a periodic report to the requests of Seri1
maintenance in order to be classified. 50
• Prevent acts of direct equipment maintenance 0
without a request order. May.09 Ağu.09 Ara.09 Mar.10 Haz.10
• Continuous work development courses for planners
to work in order to stay within the set in plans.
• Emergency notices that are written by the Fig. 3 Improvement in the requests life
operations shift leader.
5) Step 5: Control. We desire to control the better
setting of each better solution. The behaviors of all
the CTQs able do steady over time to a normal 2009
nonrandom low point in maintenance troubles 6
concentrating. Furthermore, we have to assure the 21
reformative process are taken and moving the control 1
of the hot process to the maintenance holder. We are
also in parallel using special techniques to maintain 2
73
the performance of maintenance and monitoring 3
work, this is done by scheduling maintenance
requests that require more than 3 months in key
process indicator (KPI) and submit them to the
maintenance manager. Furthermore, for the next Fig.4 a. Improvement on emergency request in2009
____________________________________________________________________________________________________________
720
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[6] Jacobsen, J., Ford Team Uses Six Sigma to Reduce Costs While
2011
12
1
50
38 2
3
2012 10
13
1
77 2
3
REFERENCES
____________________________________________________________________________________________________________
721
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, permanent magnet axial-flux coreless and light weight compared to other applications. Main
synchronous generator is designed as single stators and double purpose of this type of machine design is to obtain the best of
rotors and its electromagnetic and structural characteristics are output power [4]. Furthermore, axial-flux permanent magnet
analyzed. Core is not been used in the stator of the machine generator with coreless stator is considered as machines
intended to be designed. Aim of this study is to provide reduction
having high power density for energy-generating systems [5].
of iron loss. Moreover, easiness in the production stage of the
machine is provided. Three-dimensioned electromagnetic What has been aimed to reveal within this study is that iron
analysis of the designed machine has been done through finite losses have reduced using coreless stator. Moreover, the
element method and transient solutions are suggested based on machine has been made lighter. As core has not been used in
this. Within this study, arrangements have been made depending the stator windings of the designed machine with coreless
on certain standards in order that permanent magnets and coils stator, stator iron losses have been eliminated completely. As
obtain direct alternating current. The designed new axial-flux there has not been used of the core in the stator, stator
generator move as permanent speed of 500 rpm and so maximum windings will not be influenced by core warm-ups. As the
voltage of approximately 100 V per phase is obtained. surfaces of the windings are in contact with air, it will be able
Furthermore, this machine does not need a gear system due to its
to take the heat on the surface out more quickly. As the core is
design structure.
not used for the stator windings in this type of axial machine,
production difficulty to arise from applications in such type of
Keywords— Axial-flux generator, Permanent magnet, machines has been reduced. Though copper losses have
Generator, Single stator, Double rotor increased here as the core has not been used in the stator, this
newly designed machine provides production convenience.
I. INTRODUCTION
This design can be used to obtain alternating current without
Electrical machines have started to be used almost within need for gear system during wind turbine applications.
all areas nowadays. Every passing day, new studies take place Besides, it is intended to be used in electric vehicles both as
in literature. New models are developed in these studies done. in-wheel motor and generator. Finite element method has been
Within these developed models, increases in power density of used for analysis in this study. Finite element method is a
the machine, change in the shape of design as well as size numerical method used in solving linear and nonlinear partial
reductions and studies done depending on excitation types differential equation [6]. Design features of the designed
have gained speed. Studies have centred on axial-flux machine have been given within the second chapter. The third
permanent magnet machines as different excitation types chapter consists of electromagnetic analysis results of the
recently. Permanent magnet axial machines have been aimed machine. Consequences related to simulation depending on
to be analyzed due to reasons such as being highly efficient time have been given within the fourth chapter. The final
and economic and having ability to produce energy at low cost chapter includes the conclusion.
[1]. Micro wind turbine application has been conducted in a
study by Pop et al. Here it has been found that after a II. THE DESIGN FEATURES OF THE MACHINE
comparison of axial and radial-flux permanent magnet Main purpose of the machine design is to obtain the best of
generator, axial-flux permanent magnet has shown the best output power. Estimated value of the output power belonging
result. It has been discovered that axial-flux generator is less to the designed machine can be calculated using the equality
costly [2]. Reducing no-load momentum of the generator has below;
been possible using skewed magnets in axial-flux permanent
magnet [3]. It is a huge advantage that axial flux permanent
magnet generator (AFPMG) has high efficiency, compact size
____________________________________________________________________________________________________________
722
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
15 2 f
P k k (1 k )(1 k 2 ) B D3 (1)
out 8 p i p d d g o
In order that output power of the machine reaches maximum,
rate between its dimensions should be as in k 1 3 [7].
d
Statement related to the relationship between the dimension
and power in axial flux permanent magnet synchronous
generator has been given within equality 1 [8]. The k p here
is electric wave form factor and calculated as 0,5 in
sinusoidal designs where k i is current wave form and
calculated as 2 for sinusoidal wave forms [9]. B g refers to
maximum flux density within the air gap. P refers to dipole
number and f refers to the frequency. D0 and Di refers to Fig. 2 Geometry of side rotors of axial flux permanent magnet generator
inner and outer diameter respectively where k d refers to the
ratio of the inner diameter to the outer diameter. The designed Rotor steel used for the machine has been made up of M19.
axial-flux coreless permanent magnet synchronous generator There are 12 magnets in rotor steel. Magnets placed upon the
is given in Figure 1. There are one coreless stators and two rotors have been placed between each other with 30 degrees
rotors within this design. each. Do and Di values are 145 mm and 85 mm.
There exist two rotor discs in Figure 1. 12 permanent magnets Geometric structure of the magnets is as shown in Figure 3.
have been placed into the single end of the lower and upper These types of designed magnets are more advantageous to
rotor discs. The stator winding has been placed between two prevent the heat and more suitable to obtain a straight
rotor discs. The stator winding consists of 9 coils in total. sinusoidal wave. Permanent magnet used in axial flux
Each phase in windings corresponds to three coils. permanent generator is Neodymium. Length of the
neodymium magnet is 50mm and its thickness is 8mm.
Representation of flux paths of the magnets have been given
in Figure 4.
____________________________________________________________________________________________________________
723
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 6 Distribution of magnetic flux density for axial flux permanent magnet
generator.
____________________________________________________________________________________________________________
724
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Within the Fig. above, coils belonging to each phase and their been given. Within this study, due to the form of design there
relationships with each other have been given. There are three will be much convenience in production. Though the designed
coils per phase for each stator winding here. 3 coils in total machine is coreless, a high power density has been obtained.
have been used for each phase. Resistance has been That the output wave forms are close to sinusoidal reveals that
determined as 1,72 ohm in total for one phase of stator. The design data related to the machine have been properly chosen.
permanent magnets used are trapezoidal, phase voltage and 9 coils and 24 magnets have been used in the machine and
current wave form to be obtained time-dependently have been according to the simulation results at 500 rpm, the machine
demonstrated in Figure 10, 11 respectively. generates approximately 750 W power. The designed
generator within this study offers a solution for both in-wheel
motor-generator in electric vehicles and wind power
applications.
REFERENCES
[1] C. C. Chan, "Axial-Field Electrical Machines - Design And
Applications" IEEE Transactions on Energy Conversion. no. 2, pp.
294–300, 1987.
[2] A. A. Pop, F. Jurca, C. Oprea, M. Chirca, S. Breban, and M. M.
Radulescu, “Axial-Flux Vs. Radial-Flux Permanent-Magnet
Synchronous Generators For Micro-Wind Turbine Application", Power
Electronics and Applications, 15th European Conference, Lille, 2013.
[3] G. Lee and T. Jung “Design Of Dual Structural Axial-Flux Permanent
Fig. 10 Voltage value for A-B-C phase of axial-flux permanent magnet Magnet Generator For Small Wind Turbine”, IEEE Tencon. pp. 90–94,
generator. 2013.
[4] N. F. Lombard and M. J. Kamper, “Analysis And Performance Of An
Ironless Stator Axial-Flux PM Machine”, IEEE Trans. Energy
Convers., vol. 14, no. 4, pp. 1051–1056, 1999.
[5] F. Caricchi, F. Crescimbini, O. Honorati, G. Lo Bianco, and E. Santini,
“Performance Of Coreless-Winding Axial-Flux Permanent-Magnet
Generator With Power Output At”, IEEE Trans. Ind. Appl., vol. 34, no.
6, pp. 1263–1269, 1998.
[6] V.K. Desai, “Scope of Small Wind Turbines in INDIA”, Bulletion on
Energy Efficiency, Rajkot, India, 7 :1-2., 2006.
[7] J. F. Gieras, R. J., Wang, M. J., Kamper, “Axial-Flux Permanent
Magnet Brushless Machines”, Kluwer Academic Publisher, 32-341 p.,
2004.
[8] L. Soderlund, A. Koski, H. Vihriala, J.-T. Eriksson, and R. Perala,
“Design Of An Axial-Flux Permanent Magnet Wind Power Generator”,
1997 Eighth Int. Conf. Electr. Mach. Drives, (Conf. Publ. No. 444), no.
444, pp. 224–228, 1997.
[9] A. Mahmoudi, S. Kahourzade, N.A. Rahim and W.P. Hew, “Design,
Analysis, And Prototyping Of An Axial-Flux Permanent Magnet Motor
Based On Genetic Algorithm and Finite-Element Analysis”, Magnetics,
Fig. 11. Current value for A-B-C phase of axial-flux permanent magnet IEEE Transactions on, vol.49, no.4, pp.1479,1492, April 2013.
generator
V. CONCLUSION
____________________________________________________________________________________________________________
725
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(1.1)
____________________________________________________________________________________________________________
726
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
727
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
interpolated separately for each. The equations of the slope III. MATLAB RESULTS
and approximated are There are three main objectives while designing a DDFS.
These are maximizing the SFDR, minimizing the ROM size
(2.7) and increasing the maximum operating frequency as much as
possible. In this paper, the SFDR and ROM size
(2.8) considerations are investigated.
(3.2)
(3.3)
(2.9) TABLE I
BTM RESULTS IN MATLAB
The study of De Caro and his friends shows that the ROM
number of TOs is not proportional to SFDR. The study also Resolution SFDR
size
(bits) (dB)
indicates that SFDR of the DDFS strongly depends on (bits)
numbers. From the point of this view, it is obvious that an 5 2 8 56.22 352
optimization is essential to get best results. If the number of 5 3 8 58.14 448
TOs is more than two then the optimization requires more 5 4 8 56.25 640
complex algorithms and calculations. Moreover, the higher 5 2 10 70.45 960
number of TOs does not ensure smaller physical area in 5 3 10 69.93 1600
compression with two or three TOs for the SFDR levels less 5 4 10 71.80 2880
5 2 12 78.76 3968
than 90 dBc [9]. Thus, this paper is focused on the table based
5 3 12 82.68 7552
methods with one or two TOs.
5 4 12 80.88 14720
____________________________________________________________________________________________________________
728
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The BTM design results are given in Table I and allow two TABLE II
2 TOS MTM DESIGN RESULTS IN MATLAB
significant deductions. Firstly, it is obvious that increasing the
length of the phase word does not improve the SFDR and it ROM
SFDR
makes the ROM size bigger. So, determining the best size
(dB)
decomposition of the phase word has a great importance. (bits)
Secondly, increasing the amplitude resolution provides better 5 3 2 2 2 57.09 304
SFDR but causes a notable increment in the ROM size. 4 4 2 2 2 55.96 256
4 3 2 3 2 55.32 208
P=10
R=8
B. MTM Design 4 3 3 3 2 59.26 272
MTM is originally based on BTM and requires smaller 4 3 4 3 2 57.02 400
ROM size comparing to it with a negligible decrease in SFDR. 4 3 2 3 4 57.09 256
It is generally preferred when higher SFDR is required 3 3 2 4 4 55.64 400
because BTM provides the same SFDR with a very high ROM 7 3 3 3 3 77.18 1568
size. In this part of the paper, some MTM design results are 7 4 3 2 3 80.34 1664
investigated and are compared with BTM. 6 4 3 3 3 81.98 1056
R=11
P=13
The TIV size calculation in MTM is similar to BTM. Using 5 5 3 3 3 75.20 1152
more than one TO makes the difference in total ROM size. 5 4 3 4 2 78.25 864
The phase word decomposition is given in Fig. 7 where 5 4 5 4 2 73.20 1952
. 4 4 5 5 2 69.04 2160
determine SFDR and ROM size. Some design results are 75.20 928 388.41:1
(2 TOs)
given in Table II. As clearly seen from the table it is possible
MTM
to reach better SFDR by just rearranging the decomposition of 75.30 496 726.71:1
(3 TOs)
the phase word. The results indicate that minimizing α value BTM 84.24 16384 208:1
allows making total ROM size smaller. It is also clear that MTM
R=13
P=16
increasing and values causes a remarkable rise in SFDR. 84.50 1984 1717.68:1
(2 TOs)
While closer α, i and i+1 values provide the best MTM
84.66 984 3463.29:1
compression, it is more complicated to offer any condition for (3 TOs)
the best SFDR. Although there are some algorithms proposed
to find out the best decomposition of the phase word for a
specific SFDR level, there is not any multi-objective IV. CONCLUSIONS
algorithm to optimize both the SFDR and the ROM size. In The DDFS structure has been investigated in this study.
this paper, any algorithm is not used and the results given in The phase accumulator, phase to amplitude converter and
tables show the effect of the parameters. digital to analog converter blocks of the structure has been
mentioned in section I. Later on, CORDIC, BTM and MTM
____________________________________________________________________________________________________________
729
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
____________________________________________________________________________________________________________
730
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
731
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
All electrically powered devices radiate electromagnetic As measurement probe we used WPT mobile frequency
waves. Additionally, some communication devices uses probe. This probe measures only total electric field value of
electromagnetic waves for their purpose and increases 2G, 3G and 4G frequencies. Measurement device and probe
electromagnetic waves they were radiated. are shown in Fig. 2.
World Health Organization (WHO) researches effects of
electromagnetic waves in last three decades. There are a lot of
studies about health effect of electromagnetic waves. But most
of them could not show health effect of electromagnetic
waves clearly. International Commission on Non-Ionizing
Radiation Protection (ICNIRP) determines limit values for
non-ionizing electromagnetic waves. In Table I, ICNIRP limit
values are shown.[4-5]
TABLE I
ICNIRP LIMIT VALUES
II. MATERIALS AND METHODS Fig. 2 SMP2 EMR Measurement Device and WPT Probe
In this paper we measured electric field of mobile phone B. Measurement Mode and Standards
frequencies (GSM900, GSM1800, UMTS, LTE) in city center
of Mus. Measurements are made in main streets of the city. Measurements are made in mobile measurement type.
With GPS data we showed measurement results on the map. Measurement device was put on a car and while car was
travelling 30 km/h fixed speed, device measured electric field
A. Measurement Device and Probe of the medium. Measurement instances are taken in every 5
We used Wavecontrol SMP2 electromagnetic radiation seconds.
measurement device. Device specifications are shown below; Measurement results were shown on a map according to
1) 1Hz to 18 GHz Broadband Measurements: With RMS GPS data and field strength.
and Field probes can measure up to 18 GHz. III. MEASUREMENT RESULTS
2) Spectrum Analysis: With a specific probe, spectrum
analysis 1 Hz to 400 kHz. According to measurement results;
3) Weighted Peak Method (WPM): Real time comparison Maximum measurement value is 5.85 V/m and
in percentage with selected limits. measured in middle of the Ataturk Boulevard.
4) Assessments According to European Directive Average measurement value is 1.098 V/m during the
2013/35/EU: In accordance with international whole measurement
standards such as ICNIRP, IEC, EN, IEEE, etc. Comparison of electric field strength values in main
5) GPS Support: GPS module integrated in the instrument. streets from the highest to the lowest: Ataturk
6) Connection Options: USB and Fibre Optic connection Boulevard, Istasyon Street and Cumhuriyet Street.
supported by a PC software. Measurement results are shown in Fig. 3. Green points
show 0-2 V/m electric field strength value. Yellow points
____________________________________________________________________________________________________________
732
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
show 2-4 V/m electric field strength value and Red points
show 4-6 V/m electric field strength value.
____________________________________________________________________________________________________________
733
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
ahmetsolak@selcuk.edu.tr
sherdem@selcuk.edu.tr
____________________________________________________________________________________________________________
734
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
voltage. I-V characteristic of memristor is shown in In 2008, 37 years after the definition of
Fig.2. Memristor is also a frequency dependent memristor, Stanley Williams and his team from HP
circuit element which was mentioned before in (Hewlett-Packard) announced that the memristor is
Chua’s paper [2]. And Chua gave frequency physically implemented [3]. The team used TiO2
dependent I-V characteristic of memristor in this which is a semiconductor material in this study.
paper which is shown in Fig.3. According to this Pure TiO2 has high resistance and doped TiO2 with
figure, when applied frequency is increased, oxygen vacancies has lower resistance. These two
hysteresis curve of characteristic narrows. And kinds of TiO2 sandwiched between two Pt layers
when frequency goes to infinity, I-V characteristic and so element is obtained. Fig.4 shows HP
becomes linear as resistor characteristic. memristor model. In this model, there is a thin film
between pure and doped sides. Position of this film
i changes depending on energy flows from where.
When energy flows from doped side to pure side,
memristance value decreases. Otherwise, when
energy flows from pure side to doped side,
memristance value increases. These changes consist
of depending on ion mobility. So, HP memristor
model also called as linear ion drift model.
____________________________________________________________________________________________________________
735
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
736
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
0.5
0 1
Comparison of I-V Characteristics
-0.5 2
-1 1.5
4
-1.5 1
Current (mA)
-2 0.5
-4 -3 -2 -1 0 1 2 3 4
Voltage (V) 0
-1 Vourkas&Sirakoulis
Joglekar & Wolf
-1.5
M-V Characteristic of Proposed Model Proposed Model
8 -2
-4 -3 -2 -1 0 1 2 3 4
Voltage (V)
7
Fig.10 Comparison of I-V characteristics
6
Memristance (k)
3 8
Memristance (k)
2
6
1
-4 -3 -2 -1 0 1 2 3 4
Voltage (V) 4
____________________________________________________________________________________________________________
737
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
738
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this paper, Autoregressive with exogenous input In this study, particular type DNN and AutoRegressive
(ARX) and dynamic neural network (DNN) based generalized with eXagenous input (ARX) model based GPC are per-
predictive control (GPC) methods are designed to control of formed. ARX is a linear model while both linear and
nonlinear systems. ARX and DNN models adaptively approxi-
mate the plant dynamics and predict the future behavior of the nonlinear dynamics exists in the DNN model. Models are
nonlinear system. While control process goes on, the poles of trained by extended Kalman filter (EKF). In addition to
the ARX and DNN models are constrained in a stable region adaptation of the parameters, poles are adapted such that
using a projection operator for structural stability. Simulation they are bounded to maintain stability of the model. Thus,
results are given to compare the tracking performances of the quick and stable online identification is obtained. In nu-
methods. ARX-GPC and DNN-GPC both yield good tracking
performances while keeping the changes in control signal as merical simulations, bioreactor and continuously-stirred tank
low as possible. The simulation results show that even though reactor (CSTR) benchmark systems are controlled using two
ARX is a linear model, it provides acceptable tracking results methods and results are compared.
as well as DNN model. Rest of the paper is organized as follows. In Section II,
Keywords: Generalized predictive control, ARX, dynamic DNN and ARX models are presented, model stability and
neural network, Kalman filter and extended Kalman filter,
nonlinear systems and adaptive learning rate. EKF-based training are introduced. GPC and ARX and DNN
based GPC are detailed in Section III. Simulations are given
I. I NTRODUCTION in Section IV. Finally, paper is concluded in Section V.
____________________________________________________________________________________________________________
739
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
yk = h(xk ) + vk (14)
n m
Here, x ∈ R is state vector, y ∈ R is system output,
f(.) is nonlinear process function vector, h(.) is nonlinear
measurement function vector, w ∈ Rn is process noise and
Fig. 2. Diagonal dynamic model. v ∈ Rm is measurement noise, which both are Gaussian
white noises. Extended Kalman filter can be summarized
____________________________________________________________________________________________________________
740
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
in two steps given Q and R are covariance matrices cor- In (22), P ∈ Rt×t , t = 2(n + n × m) is estimation error
responding to process and measurement noises respectively. covariance matrix of the parameters in the vector θ and Jh
Time Update is updated as follows.
x̂− ∂ŷ(θ̂k− )
k =f(x̂k−1 ) Jh (θ̂k− ) = (23)
P− T
(15) ∂ θ̂k−
k =Jf (x̂k−1 )Pk−1 Jf (x̂k−1 ) + Qk−1
T − − − T −
Kk =P−
k Jh (x̂k )(Jh (x̂k )Pk Jh (x̂k ) + Rk )
−1 Projection onto the constraint space is taken as in (20) to
satisfy the constraints in (9) for ARX model while in (12)
Measurement Update for DNN model. For the DNN case, states are updated as
in (3) using the parameters which are updated satisfying the
x̂k =x̂− −
k + Kk (yk − h(x̂k ))
(16) constraints. Thus, states are not the ones which are estimated
Pk =(I − Kk Jh (x̂− −
k ))Pk optimally but parameters in the vector θ are.
In (15) and (16), P ∈ Rn×n is state estimation error III. ARX AND DNN BASED G ENERALIZED P REDICTIVE
covariance matrix and Kk is Kalman gain. Jf (.) and Jh (.) C ONTROL
are given in (17) and (18). Section III-A briefly introduces GPC while Section III-
B and III-C detail the ARX-based and DNN-based GPC
∂f(x)
Jf (x̂k−1 ) = (17) respectively.
∂x x=x̂k−1
A. GPC
− ∂h(x)
Jh (x̂k ) = (18) Let us have a NARX (Nonlinear Auto-Regressive with
∂x x=x̂−
k eXogeneous input) data model for a nonlinear system [9].
Let us consider there exist inequality constraints on states as yn = f (un , un−1 , . . . , un−nu , yn−1 , . . . , yn−ny ) (24)
given below in state estimation by ICEKF.
As ỹn is the reference signal and ŷn is the model output, GPC
Mx̂ ≤ N (19) aims to have the model output track the reference signal with
Parameteres must be within the constraint space to satisfy minimum tracking error possible while keeping the changes
the constraints. By this reason, projection is taken onto the in the control signal as low as possible. Constraints exist on
constraint space [17]. the control signal.
____________________________________________________________________________________________________________
741
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
vector u ∈ RKu +1 is generated and first element in the vector Thus, partial derivative expression will be
is applied to the system. To predict the future behaviour of ny
the system whose output function is unknown, an appropriate δ ŷk+l X δ ŷk+l−i
= θk,nu +1+i δ1 (k − i − 1)+
model to approximate that system is used. Also that model δuk+h i=1 δuk+h
is utilised for the gradient information which is necessary to nu
( (34)
update the control signal at each time instant. This update is
X θk,i+1 δk−i,h k − Ku < i
expressed as follows. i=0
θk,i+1 δKu ,h k − Ku ≥ i
ŷk+l = ck+l θkT , l = 1, . . . , Ky (32) δ1 is the unit step function. Obtained partial derivative is
substituted in (29) and (30) to obtain the gradient vector
It can be written explicitly and Hessian matrix.
ny
X
ŷk+l = θk,nu +1+i ŷk+l−i + IV. C OMPUTER S IMULATIONS
i=1
nu
( (33) ARX-based and DNN-based GPC controllers were tested
X θk,i+1 uk+l−i k − Ku < i on bioreactor [5] and continuously stirred tank reactor [23]
i=0
θk,i+1 uk+Ku k − Ku ≥ i systems.
____________________________________________________________________________________________________________
742
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
0.1
produced control signal by ARX-based GPC controller while
0.05
Figure 4 shows time evolution of parameters. Similarly, 0
0 50 100 150 200 250 300
t
Ky=10 | Ku=2 | l=0.01 | umin=0 | umax=2 | dumax=0.2
0.2 yref 2
y
0.15 1.5
u(t)
0.1 1
0.05 0.5
0
0 50 100 150 200 250 300 0
0 50 100 150 200 250 300
t
t
2
1.5
Fig. 5. Bioreactor reference tracking and control signal produced by DNN-
u(t)
1
based GPC controller.
0.5
0
0 50 100 150 200 250 300
t
D parameters A parameters
0.3 0.3
Fig. 3. Bioreactor reference tracking and control signal produced by ARX-
0.2 0.2
based GPC controller.
0.1 0.1
0 0
0.6
0.2 0.2
0.4 0 0
−0.2
0 50 100 150 200 250 300
t
Fig. 6. Time evolution of parameters in control of bioreactor by DNN-
based GPC controller.
Fig. 4. Time evolution of parameters in control of bioreactor by ARX-based
GPC controller.
____________________________________________________________________________________________________________
743
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(43) 1
0.8
u(t) is the control signal and d2 is time-varying parameter
0.6
of the process which has a nominal value of 1. Controllers
u(t)
0.4
were employed to make the reactor output track desired 0.2
reference value with minimum tracking error possible. GPC 0
0 20 40 60 80 100 120
parameters are Ky = 10, Ku = 2, λ = 0.1, umin = 0, t
0.6 1
Ky=10 | Ku=2 | l=0.1 | umin=0 | umax=1 | dumax=0.1 yref
y 0.4 0.5
0.6
0.2 0
0.5
0 −0.5
0 50 100 150 0 50 100 150
0.4 t t
B parameters C parameters
0.3 1.5 0.3
0 20 40 60 80 100 120
t
1 0.2
1
0.5 0.1
0.8
0.6 0 0
u(t)
____________________________________________________________________________________________________________
744
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
ARX and DNN models are adopted succesfully in GPC [22] S. Wu and M. J. Er, “Dynamic fuzzy neural networks-a novel approach
scheme for both systems. to function approximation,” IEEE Transactions on Systems, Man, and
Cybernetics, Part B (Cybernetics), vol. 30, no. 2, pp. 358–364, Apr
2000.
ACKNOWLEDGMENT [23] W. Wu and Y.-S. Chou, “Adaptive feedforward and feedback control
This work is supported by Pamukkale University Scientific of non-linear time-varying uncertain systems,” International Journal
of Control, vol. 72, no. 12, pp. 1127–1138, 1999.
Research Projects (BAP) Department. [24] S. J. Yoo, Y. H. Choi, and J. B. Park, “Generalized predictive control
based on self-recurrent wavelet neural network for stable path tracking
R EFERENCES of mobile robots: adaptive learning rates approach,” IEEE Transactions
on Circuits and Systems I: Regular Papers, vol. 53, no. 6, pp. 1381–
[1] K. Astrom, “Theory and applications of adaptive control survey,”
1394, June 2006.
Automatica, vol. 19, no. 5, pp. 471 – 486, 1983.
[25] W. Yu, “State-space recurrent fuzzy neural networks for nonlinear
[2] S. Beyhan and M. Alcı, “Extended fuzzy function model with sta-
system identification,” Neural Processing Letters, vol. 22, no. 3, pp.
ble learning methods for online system identification,” International
391–404, 2005.
Journal of Adaptive Control and Signal Processing, vol. 25, no. 2, pp.
168–182, 2011.
[3] D. W. Clarke, C. Mohtadi, and P. S. Tuffs, “Generalized predictive
control—part ii. extensions and interpretations,” Automatica,
vol. 23, no. 2, pp. 149–160, Mar. 1987.
[4] D. Clarke, C. Mohtadi, and P. Tuffs, “Generalized predictive
controlpart i. the basic algorithm,” Automatica, vol. 23, no. 2, pp.
137 – 148, 1987.
[5] K. O. Efe M O, Abadoglu E, “A novel analysis and design of a neural
network assisted nonlinear controller for a bioreactor,” International
Journal of Robust and Nonlinear Control, vol. 9, pp. 799–815, 1999.
[6] M. Ghiassi, H. Saidane, and D. Zimbra, “A dynamic artificial neural
network model for forecasting time series events,” International
Journal of Forecasting, vol. 21, no. 2, pp. 341 – 362, 2005.
[7] S. Hu and J. Wang, “Global stability of a class of discrete-time
recurrent neural networks,” Circuits and Systems I: Fundamental
Theory and Applications, IEEE Transactions on, vol. 49, no. 8, pp.
1104–1117, Aug 2002.
[8] P. A. Ioannou and J. Sun, Robust Adaptive Control. Upper Saddle
River, NJ, USA: Prentice-Hall, Inc., 1995.
[9] S. Iplikci, “A support vector machine based control application to the
experimental three-tank system,” {ISA} Transactions, vol. 49, no. 3,
pp. 376 – 386, 2010.
[10] L. Jin, P. N. Nikiforuk, and M. M. Gupta, “Absolute stability
conditions for discrete-time recurrent neural networks.” IEEE
Transactions on Neural Networks, vol. 5, no. 6, pp. 954–964, 1994.
[11] K. Kim, J. Park, and Y. Choi, “The adaptive learning rates of extended
kalman filter based training algorithm for wavelet neural networks,” in
MICAI 2006: Advances in Artificial Intelligence, ser. Lecture Notes in
Computer Science, A. Gelbukh and C. Reyes-Garcia, Eds. Springer
Berlin / Heidelberg, 2006, vol. 4293, pp. 327–337.
[12] J. Korbicz, K. Patan, and A. Obuchowicz, “Dynamic neural networks
for process modelling in fault detection and isolation systems,” Inter-
national Journal of Applied Mathematics and Computer Science, vol.
Vol. 9, no 3, pp. 519–546, 1999.
[13] G. J. Kulawski and M. A. Brdy, “Stable adaptive control with
recurrent networks,” Automatica, vol. 36, no. 1, pp. 5 – 22, 2000.
[14] C. J. LI Li-Juan, SU Hong-Ye, “Generalized predictive control with
online least squares support vector machines,” Acta Automatica Sinica,
vol. 33, no. 11, p. 1182, 2007.
[15] K. Patan, “Stability analysis and the stabilization of a class of discrete-
time dynamic neural networks,” Neural Networks, IEEE Transactions
on, vol. 18, no. 3, pp. 660 –673, May 2007.
[16] G. A. Rovithakis and M. A. Christodoulou, “Adaptive control of
unknown plants using dynamical neural networks,” IEEE Transactions
on Systems, Man, and Cybernetics, vol. 24, no. 3, pp. 400–412, Mar
1994.
[17] D. Simon, “Kalman filtering with state constraints: a survey of
linear and nonlinear algorithms,” Control Theory & Applications,
IET, vol. 4, no. 8, pp. 1303–1318, Aug. 2010.
[18] D. Simon, “Training fuzzy systems with the extended kalman filter,”
Fuzzy Sets Syst., vol. 132, pp. 189–199, December 2002.
[19] N. K. Sinha, M. M. Gupta, and D. H. Rao, “Dynamic neural networks:
an overview,” in Industrial Technology 2000. Proceedings of IEEE
International Conference on, vol. 1, Jan 2000, pp. 491–496 vol.2.
[20] Y. Song, Z. Chen, and Z. Yuan, “New chaotic pso-based neural
network predictive control for nonlinear process,” IEEE Transactions
on Neural Networks, vol. 18, no. 2, pp. 595–601, March 2007.
[21] P. Srensen, M. Nrgaard, O. Ravn, and N. Poulsen, “Implementation of
neural network based non-linear predictive control,” Neurocomputing,
vol. 28, no. 13, pp. 37 – 51, 1999.
____________________________________________________________________________________________________________
745
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this paper, Cuckoo search optimization algorithm Optimization and differential evolution etc., heuristic
is used to selectthe standard component values for active filter algorithms can be used for such a complex and burdensome
design by means of fitness function and constraints in order to problems[1]. These algorithmsgenerally include objective
facilitate design tasks of designers. Conventionally after function expressing the problem and solution space. Thus,
designers solve circuit, they select the nearest
suitable results can be obtained with such difficult problems.
standardcomponent values, which are available in the market,
instead of calculated one. However, this process causes design
errors which have to be recalculated. From this perspective, In the literature, David et al. performed a low pass active
unlike conventional method, Cuckoo Search uses a set of filter design using genetic algorithms. They examined the
solution, which is discreet component values and fitness function effect of the changing crossover and mutation rates to the
to minimize the error. As a result of this work, Cuckoo Search results in the genetic algorithm. They also obtained quite
algorithm has been successfully implemented to this filter design successful results in the standard component selection
process for searching the optimum discreet component values of compared with the conventional methods[2]. Turgay and et al.
the solution space. have made the design, using genetic algorithms, for the
Sallen-Key band pass filter. They made solutions for the
desired frequency and filter gain to find out component values
Keywords— Cuckoo search, filter design, optimization regardless of standard components[3]. Adem Kanali made
solution to find out standard values of low pass filter and
Wien-Robinson filter using Ant Colony algorithm and
I. INTRODUCTION conventional methods. He compared these algorithms each
In the conventionalanalogue circuit design methods, other and demonstrated that Ant Colony algorithm is more
component values are generally considered ideal and limitless. efficient rather than conventional methods[4]. Gandomi and
However, electronic components are produced in specific etc. showed that Cuckoo Search algorithm can be used at the
standard values in order to reduce the design costs. solution of the basic civil engineering problems[5]. Karagul
Consequently, the analogue circuit components to be used are made plastic waste collection application using different
selected from the specific standard components which are algorithms. He saw that the cuckoo search algorithm provided
available in the market. In the traditional methods, designers better results than others[6].
makesome assumptions when they know the status of the
resolution of the analogue circuit in order to facilitate the In this study, classical method and cuckoo search
circuit, to reduce the solution space and to work in the desired algorithm are used to find out standard component values in
circuit status. In these assumptions, some components are the active filter design. Finally, results are compared with
fixed or selected in multiples of one another. This process each other.
allows us to search the solution in a smaller area by narrowing
the solution space. But this gives rise to potential better
solution would be ignored. Also these assumptions restrict the II. CUCKOO SEARCH ALGORITHM
designer and cause some components remain constant value.
These acceptances cause undesirable product errors to be Cuckoo Search Algorithm is proposed by Yang and Deb in
fixed. 2009. Such algorithms typically used in global optimization
In the design of analogue circuits, to reduce the problems[7-8]. Cuckoo Search Algorithm is based on natural
mathematical operation load and loss of time, it has been behaviour of some species of cuckoo called brood parasitism.
proposed different methods correspond to the classical method. Such types of cuckoo that lays its eggs into the appropriate
Inspired by nature as an alternative calculation methods nests and throws eggs from the nests with the possibility of
developed with Genetic Algorithm, Ant Colony Algorithm, hatching. When host bird arrive the nest, it may not recognize
Tabu Search Algorithm, Simulated Annealing, Bee Colony the foreign egg. Therefore, large number of hosts exposed to
Algorithm, Firefly Algorithm, Cuckoo Search, Particle Swarm brood parasitism. But if host bird recognises the foreign egg, it
____________________________________________________________________________________________________________
746
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
may throw the egg from nest or leave the nest to build new
one. Yang and Deb used there idealised rules for this fiction
[7-8]:
Each bird can lay one egg and drop one egg in a
random nest at a time.
Quality solutions are transferred to the next
generations.
The number of nests is constant and host can
discover foreign egg by probability pa∈ (0,1). If it
happens, host can either throws the egg and leaves in
the same nest or abandons the nest to build new one.
= + ⊗ ( −∈) ⊗ ( − ) (1)
Local random walk, which is expressed the equation (1), Fig 1.Cuckoo Search Algorithm[7]
is determined by a random permutation of x and x . H(u)
Heaviside is step function. While ∈ is a parameter that comes
from gaussian distribution, s is step length. III. ACTIVE FILTER DESING
Global random walk is carried out by Levy Flight as Filter circuit is defined as passing circuit of specified part
following equation; of sent or received signal[9]. While filters are used to be
passed the specific band of signal in Communication,
( , ) Biomedical and Defence Industries, for example in Power
= + (2)
Electronic, they are used to maximize active power by
reducing Total Harmonic Distortion. Filters are classified to
Levy Flight is performed in order to produce new
high-pass filter, low-pass filter, band-pass filter and band-stop
generations. Moreover, not only α > 0 is selected but also
filter. According to the used components, Passive filters
levy flight can be scaled according to problem size.
typically are made of the inductor, capacitor and resistance,
whereas active filters are made of not only resistors,
When looked over the equations, deduced that next status
capacitors, but also amplifiers[10].
depends on current status and transition probability.Best
solution is obtained by keeping good solutions in the solution
set. A. State Variable Active Filter
Cuckoo Search algorithm is given in Fig 1. Firstly, In this work, state variable active filter, as known Kerwin-
objective function, which defines problem, is determined. Huelsman-Newcomb, is used and its schematic image is given
Subsequently, random and limited initial values of solution set in Fig 2. This low pass filter is totally specified by band
of population are assigned and stopcriterion is determined to passgain(H), band pass cut-off frequency ( = 2πf ) and
make the running algorithm stopped. Loop is generated to selectivity factor (Q). These parameters determine the desired
search until the suitable solution is obtained. In the loop, new operating status of filter. Kerwin-Heulsman-Newcomb Active
potential solution is generated by Levy Flight and if the Filter is used and its parameters are expressed as below [2- 4-
potential solution was better than old ones, new potential 10]
solution was kept by evaluating their fitness. After that, as a ( + )
= ,
part of this algorithm feature, new potential solution is ( + )
generated by probability of pa and better ones kept again. 1
After the iteration accomplished, stop criteria checked again. = , (3)
If stop criteria was not suitable, then iteration would continue
again. Otherwise iteration would be finished and results would ( )
be visualised. =
( )
____________________________________________________________________________________________________________
747
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
B. Implementation TABLE I
RESULTS OF CUCKOO SEARCH AND CONVENTİOANAL METHOD
In this work, pa, which is one of the control parameter of Conventional Method Cuckoo Search
Cuckoo Search Algorithm, is determined as 0.25[11]. Later, Algorithm
size of population is determined as 80 for each component H 1.2713 1.3656
variable. 8 different component variables are determined to Q 1.3723 1.4158
solve the problem. Standard of resistor was selected E24 and 1773.05 rad/sn 1579.5 rad/sn
standard of capacitor was selected E12 according to Electronic 4700Ω 100Ω
Industries Association (EIA) norms. Subsequently, solution 8200Ω 510000Ω
set was generated by random values within boundary. 4700Ω 820000Ω
Afterwards, fitness function (4) and constraint (5) were 4700Ω 300000Ω
determined. Fitness function was determined as percentage 4700Ω 24000Ω
difference of desired values and results, for selectivity and 4700Ω 13000Ω
frequency. Also, constraint was determined that both half gain, 1.2x10-7 F 1x10-11 F
which is desired, must be smaller than filter gain and one and 1.2x10-7 F 4.7x10-5 F
half gain, which is desired, must be greater than filter gain. ( , ) %14.36 %0.86
____________________________________________________________________________________________________________
748
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[4] A. Kalinli, "Karınca Koloni Algoritması Kullanarak Aktif Filtre [8] B. Premalatha and S. Umamaheswari, "Cuckoo Search
Tasarımı," Elektrik-Elektronik-Bilgisayar Mühendisliği 10. Ulusal Optimization Algorithm based Hardware Task Placement and
Kongresi, pp. 267-270, 2003. Routing in CAD of FPGAs Design Flow," Indian Journal of
[5] A. H. Gandomi, X.-S. Yang, and A. H. Alavi, "Cuckoo Search Science and Technology, vol. 9, 2016.
algorithm: a metaheusristic approch to solve structural [9] A. Demirkol, "Sinyaller ve Sistemler Frekans Cevabı," Sakarya,
optimization problems," Springer-Verlag, vol. 29, p. 18, 2013-01. Turkey, September 2013.
[6] K. Karagül, "Guguk Kuşu Algoritması: Bir Plastik Atık Toplama [10] M. KAÇMAZ, "Aktif Filtreler," Elektrik Mühendisleri Odası Aylık
Uygulaması," 15th International Symposium on Econometrics, Yayın Organı, vol. 181/5, pp. 21-37, Ocak 1972.
Operations Research and Statistic, Isparta, Turkey, vol. 15, pp. [11] X.-S. Yang, "Nature-Inspried Optimization Algorithms, 1st
775-784, 22-25 May 2014. Edition," Elsevier, 20 Feb 2014.
[7] X.-S. Yang and S. Deb, "Cuckoo Search via L´evy Flights,"
Nature & Biologically Inspired Computing, pp. 210 - 214, 9-11
Dec. 2009.
____________________________________________________________________________________________________________
749
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
750
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
751
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(b)
Fig. 4 General PN models of (a) working modes and (b) program modes of
appliances
Fig. 3 PN models of (a) True/False sensors and (b) sensors with wider range
measurement capability.
Working modes and program modes of any appliance are
monitored by the cyber part of the building and controlled
For example, explanation of places and transitions of the PN according to the predefined algorithms and scenarios. The
model of an occupancy sensor, which is a kind of True/False cyber part of the building uses the information gathered from
sensor, are given in Table 1. the sensors and responds quickly to the changes inside the
building.
____________________________________________________________________________________________________________
752
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE II
PLACES OF PN MODEL OF LIGHTING SYSTEM
____________________________________________________________________________________________________________
753
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
754
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Short term load forecasting is a subject about useful for determining maintenance scheduling, hydrothermal
estimating future electricity consumption for a time interval from coordination and fuel purchasing strategies [1].
one hour to one week and it has a vital importance for the Short-term load forecasting is a subject on which electric
operation of a power system and smart grids. This process is utility industry and researchers are studied for several decades
mandatory for distribution companies and big electricity
and is an important process for the operations of electric power
consumers, especially in liberalized energy markets,. Electricity
generation plans are made according to the amount of electricity systems and smart grids. It has a critical role on generation
consumption forecasts. If the forecast is overestimated, it leads to scheduling, unit commitment and maintenance plans for
the start-up of too many units supplying an unnecessary level of generators. Short term load forecast has been made in a time
reserve, therefore the production cost is increased. On the period from one minute to one hour.
contrary if the forecast is underestimated, it may result in a risky The concept of smart grid has become a global trend in the
operation and consequently power outages can occur at the power last years and there is not any definition of smart grid
system. In this study, a hybrid method based on the combination acknowledged by all institutions. For example, The Smart
of Artificial Bee Colony (ABC) and Artificial Neural Network Grids European Technology Platform defines smart grid as
(ANN) is developed for short term load forecasting. ABC
‘‘electricity networks that can intelligently integrate the
algorithm is used in ANN learning process and it optimizes the
neuron connections weights of ANN. Historical load, temperature behavior and actions of all users connected to it – generators,
difference and season are selected as model inputs. While three consumers and those that do both – in order to efficiently
years hourly data is selected as training data, one year hourly data deliver sustainable, economic and secure electricity supplies’’
is selected as testing data. The results show that the application of [2]. Smart grid is an advanced electrical network that uses
this hybrid system produce forecast values close to the actual control, information and communication technologies to
values. monitoring the power system. In smart grids, energy
management is performed with using some forecasts which use
Keywords— artificial bee colony, artificial neural network, hybrid real time data. These forecasts are renewable production
method, short term load forecasting, smart grids forecast, load forecast and electric price forecast. The power
network can be managed efficiently by making these forecasts.
I. INTRODUCTION Short term load forecast is also crucially important for smart
Today existing power system needs to be enlarged so as to grids. Recently, the demand side management becomes an
meet the future demand for electrical energy. The general emerging emphasis in the electric industry with the
purpose of the planning of electric power system is delivering development of smart grid technologies. Especially in micro
the electric power to end user with some desired features such grids, short term load forecast helps to meet the objectives of
as continuous, safe, reliable and minimum price. Various plans demand side management issues like determining the electric
should be made to achieve these features and load forecasting price, determining the amount of the purchases and sales of
is one of the most important steps of these plans. In literature, energy and reducing the peak load value. It answers the
load forecasting is divided into three categories according to the questions of load needs such as when, how much, why and
forecasting time horizons. Long term load forecast consists which factor effect [3].
yearly load demand forecasts and it is used for planning future There are many studies under the umbrella of short term load
electric generation, transmission and distribution. Generally forecasting. Studies can be split into different groups according
gross domestic product, population, export, import, etc. are to the methods, data analyzing forms, data sets, the input
considered as inputs in the long term load forecasting studies. variable and hourly/half-hour forecast. But, generally in review
Forecasting which is made in a time period ranging from one studies, they are grouped according to the methods. The used
week to one year is called midterm load forecasting and it is methods can be divided into three as statistical methods,
____________________________________________________________________________________________________________
755
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
756
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
1
= ( − ) (2)
2
The data is divided into two as test and learning data for the
execution of the algorithm. The weights obtained at the end of
the training have to include the entire training data. The training
will be completed when the total error reaches its minimum
value.
Fig. 3 Linear activation function
B. Artificial Bee Colony(ABC)
Artificial Bee Colony (ABC) algorithm is used as an
optimization technique with relation between food and bees.
There are three types of artificial bees and these are onlooker
(observer) bees, employed bees and scout bees. The onlooker
bees give information to the employed bees about food source
and distance by dancing on the dance area. The employed bees
visit the food source and collect nectar from these sources. The
scout bees look for better resources by making a random search
[19]. The working principle of the algorithm is shown in Fig. 6.
____________________________________________________________________________________________________________
757
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
758
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Get Inputs,
Weights and
Use transfer Train ABC
ANN Init Food
Inputs function and Get Best
structure Source for
produce solution Solution
ABC
MW MW
32000 32000
28000 28000
24000 24000
20000 20000
1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23
hour hour
Actual Forecasted Actual Forecasted
(a) Best Result for Monday Day Type - 03.12.2012 (b) Worst Result for Monday Day Type - 31.12.2012
MW MW
34000 34000
30000 30000
26000 26000
22000 22000
1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23
hour hour
Actual Forecasted Actual Forecasted
(c) Best Result for Weekday Day Type - 16.02.2012 (d) Worst Result for Weekday Day Type - 03.01.2012
MW
MW 35000
30000
28000 30000
26000
25000
24000
22000 20000
1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23
hour hour
Actual Forecasted Actual Forecasted
(e) Best Result for Saturday Day Type - 01.12.2012 (f) Worst Result for Saturday Day Type - 16.06.2012
MW MW
32000 32000
28000 28000
24000 24000
20000 20000
1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23
hour hour
Actual Forecasted Actual Forecasted
(g) Best Result for Sunday Day Type - 29.01.2012 (h) Worst Result for Sunday Day Type - 02.09.2012
Fig. 9 Actual and Forecasted values
____________________________________________________________________________________________________________
759
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
760
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— In this study, a fractional order PID (FOPID) fast charging-discharging capacity [4]. Integration of the UC
controller is designed and used to control a DC-DC non-inverting to WEC with and without power electronic devices can be
buck-boost converter (NIBBC) for a wave/ultra-capacitor (UC) found in literature [5, 6, 7]. UCs can be directly connected in
energy system. Because of the energy discontinuities encountered parallel with an energy source, especially low voltage
in wave energy conversion systems (WECS), an UC is integrated
applications [8]. Also, direct integration of an UC to energy
to the WECS. In order to obtain the best controller performance,
particle swarm optimization (PSO) is employed to find the the system increases the overall system efficiency by
optimum controller parameters. Integral of time weighted eliminating the converter losses.
absolute error (ITAE) criteria is used as an objective function. Fractional calculus has become highly popular in
Also, an optimized PID controller is designed to test the engineering applications, especially in control systems. One of
performance of the FOPID controller. The whole system is the fractional order control method is the fractional order PID
developed in Matlab/Simulink/SimPower environment. The controller which is the extension of the PID controller and
simulation results show that the FOPID controller provides proposed by Podlubny in 1999 [9]. A FOPID consist of 5
lower value performance indices than the PID controller in terms parameters: the proportional gain (KP), integral gain (KI),
of reducing the output voltage sags and swells.
order of integral (λ), derivative gain (KD) and order of
derivative (µ). These parameters increases the complexity of
Keywords— Wave energy, ultra-capacitor, non-inverting buck- the controller and they should be optimized in order to obtain
boost converter, fractional order PID controller, particle swarm the best controller performance.
optimization. One of the tuning method of controller parameters is to use
meta-heuristic algorithms with error-based objective functions.
I. INTRODUCTION
One of these algorithms is the PSO algorithm. PSO is
Alternative energy sources such as renewable energy are developed by Kennedy and Eberhart in 1995 [10]. The
highly considered in order to meet increasing energy demands algorithm is an iterative optimization method based on the
all over the globe. Among the renewable energy sources, stochastic movements of the swarms such as fishes and birds
wave energy (WE) is a promising energy source. Its energy and it is suitable for global optimization problems.
density is more available than either solar and wind energy [1]. In this study, irregular wave effects seen on generated
Whereas, it has some design challenges because of the power in WEC are regulated by using a UC unit and NIBBC
irregular sea or ocean wave characteristics and extreme in order to obtain a reliable and sustainable load voltage. The
weather conditions [2] parallel connected UC to the WEC provides energy to the load
A wave energy converter (WEC) used for harvesting when the WEC output is insufficient. The converter is
energy form the waves. It basically includes a turbine and a controlled by both of FOPID and PID controllers for
floating buoy moves up and down on the sea surface. The comparison. Also, PSO algorithm is used to tune the
power production rate of a WEC is mostly depending on both parameters of the both controllers by minimizing the ITAE
of the wave height and frequency in a direct-driven WEC. So, performance index. Both of the system results with and
the amplitude and frequency of the WEC output voltage without UC connection are discussed.
fluctuates chaotically. Due to the output power of the WEC This paper is organized as follows. Wave energy
has a wide variations, it cannot directly connected to a load or conversion system with the subtitles wave energy converter,
grid. To overcome this problem, WECs are integrated with NIBB converter and UC modelling are presented in Section II.
power electronic devices and storage units [3]. FOPID controller is described in Section III. PSO algorithm is
An ultra-capacitor (UC) is an electrochemical capacitor given in Section IV. In Section V, the simulation results are
with high-capacitance value. Compared to batteries, it has a discussed. Finally, conclusion is stated in Section VI.
very long life cycle, high efficiency, high power density and
____________________________________________________________________________________________________________
761
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
NIBBC
S1 D2
DUC L
Cin
R
+ Cout
UC S2
- D1
Passive
WEC rectifier
KP, KI, λ KD, µ
u e - Vout
c d Sat. FOPID
PWM
+ VRef
generator
PSO ITAE
____________________________________________________________________________________________________________
762
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE II
L ULTRA-CAPACITOR SPECIFICATIONS
S1 D2
+ + Rated capacitance (F) 83
VL - Vo
+
Cout ESR (mΩ) 10
D1 S2 RL Rated voltage (V) 48
Vi
I2 Power density (W/kg) 2700
I1
- Max. energy density (Wh/kg) 2.6
-
(a)
III. FRACTIONAL ORDER PID CONTROLLER
L
S1 D2 Fractional-order PID controller based on fractional calculus
+ +
VL - Vo includes more than two additional parameters (λ, µ) to the
+
- VUC
+
U (s) K
GFOPID ( s ) K P I K D s (7)
Resr E (s) s
C IC Where U(s) and E(s) are the control and error signals,
IUC respectively. In this study, FOPID controller software are
performed by FOMCON Toolbox [19].
Repr Iepr
IV. PARTICLE SWARM OPTIMIZATION
Fig. 3 Equivalent circuit of the RC modelled UC Particle swarm optimization (PSO) is a population based
evolutionary algorithm developed from the simulations of
where VUC is the terminal voltage of the UC, Resr is the bird-flocking. The first step in algorithm is that the specified
equivalent series resistance used to simulate internal number of particles are placed randomly in the d-dimensional
resistance of the UC, Repr is the equivalent parallel resistance search space and objective function of the each particle is
used to simulate leakage currents and C is the capacitance of calculated and saved at their current position [10].
the UC. The mathematical description between the UC Associated coordinates with the best solution is called pbest .
voltage and current are given in (6). Then, the obtained best solution is compared to each other to
find the global best solution which is called as gbest. The
t movements of the particles are updated by using the position
1
C 0
VUC ( final ) (t ) IUC (t ) R e sr I C (t )dt VUC (initial ) (t ) (6) (X) and velocity (V) equations given in (8) and (9),
respectively [20].
____________________________________________________________________________________________________________
763
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
t
ITAE t e(t ) dt (10)
0
V. SIMULATION RESULTS
The WECS illustrated in Fig 1. schematically is developed
in Matlab/Simulink/SimPower environment. Because the
small-scale waves are considered in this study, current
drawing from the generator is limited by duty cycle (d) in the
range of [0, 0.65]. Simulations are performed for 30 seconds.
The initial charge voltage of the UC is 12 V and the load
resistor is 10 Ω.
The DC-DC NIBB converter used for load voltage
regulation is controlled by both of FOPID and classical PID Fig. 5 WEC phase to phase voltage and current with (a, b) and without (c, d)
controllers. PSO algorithm is employed to tune controller UC unit
parameters and ITAE performance measure is used as an
objective function. Also, other error-based performance
indexes : integral of squared error (ISE), integral of absolute
error (IAE) and integral of time weighted squared error (ITSE)
are considered for a better comparison of the designed
controllers. The optimized parameters of the controllers and
calculated performance measures are given in Table III and
Table IV, respectively.
TABLE III
OPTIMIZED CONTROLLER PARAMETERS
Controller parameters
Controller
KP KI λ KD µ
FOPID 13.42 27.78 1.41 0.0153 0.34
PID 12.11 29.78 - 0.0794 -
TABLE IV
PERFORMANCE MEASURES
Performance measures
Controller
ITAE IAE ITSE ISE
FOPID 51.39 3.535 7.129 1.287
PID 61.10 4.196 10.77 1.659
____________________________________________________________________________________________________________
764
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 6 shows that the parallel connected UC unit to the aimed to regulate the energy irregularities encountered in
WEC output eliminates the voltage drops encountered in a WEC output.
WEC because of the irregular wave effects. As a result, a The designed WECS is tested under irregular wave
more stable input voltage is obtained for the NIBBC input. condition. The rectified variable WEC/UC output voltage is
WEC provides energy to the load when the induced voltage applied to the NIBBC. DC-DC converter is controlled by a
value is greater than UC charge voltage. FOPID controller. Also, a classical PID controller is used for
comparison. In tuning process of the both controller
parameters, a well-known optimization algorithm PSO is
employed. The ITAE is used as an objective function. The
simulation results show that the FOPID provides lower value
objective function value than PID controller. This means the
higher quality output voltage without overshoot is obtained by
using the FOPID controller.
Since the small-scale waves are considered for this study,
WEC is alone inefficient to charge the UC unit. A solar or
grid charged UC can be considered to increase the overall
system sustainability.
ACKNOWLEDGMENT
This study was supported by Karadeniz Technical
University Scientific Research Projects Unit. Project No:
FBA-2014-5168.
REFERENCES
Fig. 7 Load voltage and current with and without UC unit [1] Muetze, Annette, and J. G. Vining, “Ocean wave energy conversion-a
survey,” in 41st IAS Ann. Meet. Conf., 2006, pp. 1410-1417.
[2] B. Czech and P. Bauer, “Wave energy converter concepts: design
challenges and classification,” IEEE Trans Ind Electron, vol.6, pp. 4-
16, 2012.
[3] Y. Hong, et al., “Review on electrical control strategies for wave
energy converting systems,” Renew. Sustainable Energy Rev. vol. 31,
pp. 329-342, 2014.
[4] B.E. Conway, Electrochemical supercapacitors: scientific
fundamentals and technological applications, Kluwer Academic
Publishers/Plenum Publisher, New York, pp. 11 -15, 1999.
[5] S. Hazra, and S. Bhattacharya. “Short time power smoothing of a low
power wave energy system,” in 38th IECON, pp. 5846-5851, 2012.
[6] Murray, B. Donal, et al., “Supercapacitor testing for power smoothing
in a variable speed offshore Wave Energy Converter,” IEEE J. Ocean.
Eng. vol. 37, pp. 301-308, 2012.
[7] Kovaltchouk, Thibaut, et al., “Enhanced aging model for
supercapacitors taking into account power cycling: Application to the
sizing of an Energy Storage System in a Direct Wave Energy
Converter,” IEEE Trans. Ind. Appl., vol. 51, pp. 2405-2414, 2015.
[8] M. Uzunoglu, and M. S. Alam, “Dynamic modeling, design, and
simulation of a combined PEM fuel cell and ultracapacitor system for
stand-alone residential applications,” IEEE Trans. Energy Convers.,
vol. 21, pp.767-775, 2006.
Fig. 8 FOPID and PID controlled load voltage with UC [9] I. Podlubny, Fractional Differential Equations, Academic Press, San
Diego, 1999.
Fig. 7 shows the load voltage and current with and without [10] R. C. Eberhart, , and J. Kennedy, “A new optimizer using particle
UC bank. The WEC is alone inefficient to feed the load. The swarm theory” in Proceedings of the 6th international symposium on
comparison of the PSO optimized FOPID and PID controllers micro machine and human science, New York, NY, vol. 1, pp. 39–43,
are given in Fig. 8. The FOPID controller provides a stable 1995.
[11] Y. Hong, et al., “Linear generator-based wave energy converter model
voltage with increased quality than the PID controller. Also, with experimental verification and three loading strategies,” IET Renew
the system response without overshoot is obtained with Power Gen, vol. 10, pp. 349-359, 2015.
FOPID controller. [12] H. Sira-Ramirez, and R. Silva-Ortigoza, Control design techniques in
power electronic devices, Springer Science and Business Media, 2006.
VI. CONCLUSIONS [13] L. Zubieta, and R. Boner, “Characterization of double-layer capacitors
for power electronics applications,” IEEE Trans. Ind. Appl., vol. 36,
In this study, an initially charged UC unit is connected in pp.199-205, 2000.
parallel with a WEC output. A NIBBC is used as an interface [14] R. de Levie, “On porous electrodes in electrolyte solutions,”
circuit between the WEC/UC side and the load. Thus, it is Electrochimica Acta, vol. 8, pp. 751-780, 1963:
____________________________________________________________________________________________________________
765
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[15] R. L. Spyker and R. M. Nelm, “Classical Equivalent Circuit Parameters [20] R. Poli, J. Kennedy, T. Blackwell, “Particle swarm optimization,”
for a Double-Layer Capacitor,” IEEE Trans. Aerosp. Electron. Syst., Swarm Intelligence, vol. 1, pp. 33–57, 2007.
vol. 36, pp. 829–836, 2000. [21] Z. L. Gaing, “A particle swarm optimization approach for optimum
[16] R. M. Nelms, D. R. Cahela, and J. Bruce Tatarchuk, “Modelling design of pid controller in avr system. IEEE Trans. Energy Convers.,
double-layer capacitor behaviour using ladder circuits,” IEEE Trans. vol. 19,pp. 384–391, 2004.
Aerosp. Electron. Syst., vol. 39, pp. 430-438, 2003. [22] W. C. Schultz and V. C. Rideout, “Control system performance
[17] Maxwell Technologies, BMOD0083-P048 ultracapacitor datasheet, measures: Past, present and future,” IRE Trans. on Automatic Control,
[Online]. Available: www.maxwell.com vol. 6, pp. 22-35, 1961.
[18] I. Podlubny, “Fractional-order systems and PI λDμ controllers,” IEEE [23] I. J. Nagrath, Control systems engineering, New Age International,
Trans. Autom. Control, vol. 44, pp. 208–213, 1999. 2006.
[19] A. Tepljakov, E. Petlenkov, and J. Belikov, “FOMCON: Fractional-
Order Modeling and Control Toolbox for MATLAB,” in 18th MIXDES,
pp. 684–689, 2011.
____________________________________________________________________________________________________________
766
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Determination of the appropriate feature vector from the EEG signal for
epilepsy diagnosis
Sema YILDIRIM, Hakan ISIK, Hakan EKMEKCI and Gulay TEZEL
Graduate School of Natural Sciences, Selcuk University, semayildirim@selcuk.edu.tr
Electrical and Electronics Engineering, Faculty of Technology, Selcuk University, hisik@selcuk.edu.tr
Department of Neurology, School of Medicine, Selcuk University, hakanekmekci@selcuk.edu.tr
Department of Computer Engineering, Faculty of Engineering, Selcuk University, Konya, Turkey, gtezel@gmail.com
Abstract- Epilepsy that occurs suddenly and repeatability of using SVM. Moreover, Valderrama et al. [13] identified via
seizure is a chronic neurological disorder and it is estimated that SVM an increased risk of epileptic seizures using a multi-
1% of the world population suffering from epilepsy. Therefore, feature EEG-ECG classification. A comparative study of
many studies are carried out for the diagnosis of epilepsy. In this wavelet families for EEG signal classification was performed
study, we performed to determine the most convenient features
by Gandhi et al. using Probabilistic Neural Network (PNN)
by using electroencephalography signals (EEGs). EEG
recordings that obtained from 10 normal subjects and 10 with SVM. Whereas Siuly et al. [14] proposed Clustering
epileptic patients were conducted from Selcuk University Faculty Technique-based Least q-Square Support Vector Machine
of Medicine. Different features extraction methods that include (CT-LS-SVM) classification using some feature vectors, Lima
Discrete Wavelet Transform, statistical methods were applied for et al. [15] used some SVM methods such as Lyapunov the
identifying highest success features vectors. Different standard SVM. We may reach the conclusion that classifies
classification methods were performed for determining success of EEG signal was performed by using different classification
classifiers. High classification accuracy (CA) was obtained when and feature extraction methods.
the statistical methods was used in this new dataset. K-Nearest
Neighbors (kNN) method achieved higher CA with 98% than the EEG is a complicated and non-stationary signal and its
other classification methods. characteristics are spatio-temporal dependent. They were not
Keywords-Epilepsy, electroencephalography, feature extraction, considered using statistical properties or sub-bands. The
discrete wavelet transform. suitable feature extraction and accurate classifiers are needed
for evaluation of these signals. Therefore, in this study, EEGs
have been evaluated with different classifiers by extracting
I. INTRODUCTION different features from EEGs. We carried out various
Epilepsy is a type of neurological disorder disease classification and feature extraction methods in the same work
stemming from temporary abnormal discharges of the brain for observing and detecting feature vectors that provide the
electrical activity [1] and [2]. The significant characteristic of highest success. At the same time, we observed which feature
epilepsy is recurrent seizures. Sometimes seizures may go vectors on the not suitable for this dataset. On the other hand,
unnoticed, depending on their presentation [3]. Therefore, we aim to increase the success by using determined features
automatic seizure detection, particularly if it is performed on- by means of new classification methods or hybrid systems in
line, can be very helpful to identify segments of future work.
electroencephalography (EEG) likely to have seizures [4].
However, there still exists a need to further improve existing There is no doubt that the number of inputs is very crucial
detection methods published in the literature. Elman Classifier in the system since the classification method should be
was designed to acquire EEG and psycho physiological selected very carefully. The number of unnecessarily high
signals by Hosseini et al.[5]. Kumar et al. [6] highlighted the inputs might decrease performance of the system or the
importance of entropy-based features in recognizing EEGs number of unnecessarily low inputs may not give the result
with recurrent Elman Networks. Orhan et al. [7] achieved accurately and confidently. Due to the fact that in this paper,
Multi-Layer Perceptron Neural Network (MLPNN) for we used the various feature extraction methods that include
classification of EEG signals and feature vectors were statistical features and power of sub-bands features via
clustered by using K-means clustering method. Aydın et al. Discrete Wavelet Transform for detecting epilepsy from
[8] developed a classification using Multilayer Neural EEGs.
Network (MLNN) architecture with EEG series.
Lee [9] proposed the Neural Network with Weighted Fuzzy II. MATERIALS AND METHODS
Membership (NEWFM) to classify EEG signals. Fu [10] 1. DATA ACQUISTION
performed the classification of EEG signals using Hilbert–
Huang transform (HHT) and Support Vector Machine (SVM) In this study, EEG data recorded in the Department of
Neurology is used with decision of Selcuk University Faculty
with Radial Basis Function (RBF). Joshi et al. [11] carried out
of Medicine Hospital (Non-Invasive Clinical Research Ethics
classification of EEG signals using fractional linear
Committee No. 2014/423). We have used a total of 20
prediction. Musselman et al. [12] proposed time-frequency subjects (10 epileptics, 10 non-epileptics) that collected
distributions in the classification of epileptic activities by between 2012 and 2015. All of EEGs was obtained from
____________________________________________________________________________________________________________
767
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
routine EEG recording and all subjects were awake. Their TABLE I.
mean age of 10 epileptic patients was 39 (7 male, 3 female) FEATURE VECTORS BY USING DIFFERENT FEATURE EXTRACTING
and mean age of 10 non-epileptic patients was 42 (6 male, 4 METHODS
female) and all EEG tracing has been converted
electronically. Meanwhile, EEG recordings in different dates Name Description
FV1_PB Average power of β in each sub-band
of two patients who are epilepsy and undergo a treatment FV1_PA Average power of α in each sub-band
were included to our work. FV1_PT Average power of θ in each sub-band
FV1_PD Average power of δ in each sub-band
EEG recording consist of 20 channels that include 18 FV2_Min Minimum of the coefficients in each channel
channels for EEG, 1 channel for electrocardiography (EKG) FV2_Max Maximum of the coefficients in each channel
and 1 channel for photic stimulation. However, we used 18 FV2_Var Variance of the coefficients in each channel
channels in this study because of including brain waves. FV2_Std Standard deviation of the coefficients in each channel
Sampling frequency is 200 Hz and each epoch consists of 15 FV3_Skw Skewness of the coefficients in each channel
FV3_Krt Kurtosis of the coefficients in each channel
seconds. FV3_Iqr Interquartile range of the coefficients in each channel
Recording times is different from each other due to the fact FV3_Mad Mean absolute deviation of the coefficients in each channel
that the situation of each subject in record time is different.
Moreover, dataset showing epileptic activity that is 72 (18x4) features were obtained due to the fact that 4
determined by specialists are selected. These dataset was features were extracted every channel (18). Thus, each dataset
created EEGs that include various waves such as sharp wave, consists of 100 segments and 72 features.
spike, spike and slow wave, multiple spike and slow wave
complex. 5 epochs are used in this study both epileptic and 3. CLASSIFICATION METHODS
non-epileptic patients and the length of each epoch are
composed of 3000 samples. 100 segment data for 20 subjects SVM and different Neural Network gorithms are often used
was obtained as a result of calculations and adjustments. in literature for the epilepsy diagnosis. On the other hand, in
Whereas a value of 1 is given for 50 epileptic data, a value of this study, we carried out different methods that include kNN,
0 is given for 50 non-epileptic data in order to classify. Classification Tree, Random Forest, Naïve Bayes in addition
to these methods. The parameters of classifier were optimised
in order to determine the highest success by various trials. The
2. FEATURE EXTRACTION parameters of all classifier that provide the highest success is
Feature extraction process is an important factor affecting shown in Table II.
the success of classification. Features should be determined TABLE II
for achieving results faster and more accurately with fewer
LEARNING PARAMETERS OF EACH CLASSIFIER
data. Therefore, the feature vectors were computed by the
usage of the Matlab (Version 7.11, R2010b) software Classification Parameters
package. Feature Vectors (FV) are divided into 3 groups Methods
Multilayer Hidden layer neurons: 20
respectively, Feature Vector 1 (FV1), Feature Vector 2 (FV1) Perceptron (MLP) Regularization factor: 1.0
and Feature Vector 3 (FV3) as a result of feature extraction as Max iterations: 1000
shown in Table I. The number of decomposition levels is Naïve Bayes Probability estimation: Relative Frequency
LOESSa window size: 0.5
chosen based on the dominant frequency components of the Number of points in LOESS: 100
signal. The levels are chosen such that those parts of the Classification Attribute selection: ReliefF
signal that correlate well with the frequencies required for Tree Relieff settings: 100 reference examples, 5 neighbours
Pruning: 6 instances in leaves
classification of the signal are retained in the wavelet Recursively merge leaves with same majority class: Yes
coefficients [16] . In this study, the EEG signals are filtered Pruning with m-estimate: m=2
with band-pass Butterworth filter to extract four common SVM Kernel: Sigmoid, tanh (0.0000*x.y+0.0000)
Cost (C): 512.0
frequency bands, beta (β), alpha (α), and delta (δ), theta (θ). Numeric precision: 0.001
The power of subbands has been found by using the Discrete Estimate class probabilities: Yes
Wavelet Transform (DWT) for FV1. The wavelet function Normalize data: Yes
Random Forest Number of trees: 10
selected was Daubechies with order 4, which was also proven Seed for random generator: 2
to be the best suitable wavelet function for epileptic EEG Minimal number of instances in a leaf: 5
signal analysis [17]. Eight statistical features are extracted kNN Metrics: Manhattan
Unknown values ignored: Yes
from each EEG channel data as they are the most Number of neighbours: 6
representative values to describe the original signals. Weighting: By distances
Experiments have shown that computing speed of classifiers a
LOESS: Local weighter regression (kernel smoothing procedure)
and CA decreased with using of combination of all statistical
features. Therefore, whereas some statistical features consists
of minimum, maximum, standard deviation and variance have 4. STATISTICAL CALCULATION OF PERFORMANCE
been found for FV2, skewness, kurtosis, interquartile range The performance of the classifier methods is assessed by
and mean absolute deviation have been found for FV3. the following measures.
Consequently, statistical features were divided into two
groups as FV2 and FV3. 1) True positive (TP): It identifies an input as a patient with
epileptic diagnosed by the expert clinicians.
____________________________________________________________________________________________________________
768
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
2) True negative (TN): It identifies an input as a normal that true, 10 false for 50 non-epileptic data, 29 true, 21 false for
was labeled as a normal by the expert clinicians. epilepsy group in FV1. On the other hand, kNN has been
found 49 true and 1 false both non-epileptic and epileptic data
3) False positive (FP): The detection of epileptic that was
in FV2. 49 true and 1 false for non-epileptic data and 47 true
labeled as a normal by the expert. and 3 false was found for FV3 by means of kNN. Confusion
4) False negative (FN): The detection of a normal that was matrix in Table IV gives number of true and false in detail
labeled as an epileptic patient by the expert. The that this table is also confirmed in Table II.
performance of the classifier is also assessed in terms of
sensitivity and specificity as follows.
TABLE IV
Additionally, we can calculate sensitivity, specificity and CONFUSION MATRIX FOR ALL FEATURE VECTORS BY USING
accuracy values by using TP, TN, FP and FN as shown below. DIFFERENT CLASSIFIER
(1) Sensitivity (Sens): A measure of the ability of the classifier Feature Vectors
to detect epileptic signal. Classification
FV1 FV2 FV3
Methods
0 1 0 1 0 1
TP 0 40 21 45 3 45 4
Sensitivit y % (1) Naïve Bayes
(TP FN ) 1 10 29 5 47 5 46
0 46 31 46 6 45 4
MLP
(2) Specificity (Spec): A measure of the ability of the 1 4 19 4 44 5 46
0 39 22 46 4 46 7
classifier to specify normal signal. Random Forest
1 11 28 4 46 4 43
TN 0 47 35 46 5 47 5
SVM
Specifity % (2) 1 3 15 4 45 3 45
(TN FP) 0 30 16 42 6 42 3
Classification Tree
1 20 34 8 44 8 47
TP TN 0 38 20 49 1 49 3
Accuracy % (3) kNN
1 12 30 1 49 1 47
(TN TN FP FN )
Note: column represent predictions, row represent true classes
III. EXPERIMENTS AND RESULTS In spite of the fact that SVM and MLP are providing the
In this study, we tried to find which feature vector that highest achievement many studies in the literature, the results
represents better the EEG signal using various classification show that kNN provides the highest CA in our work.
methods such as MLP, Random Forest, SVM, kNN, Meanwhile, the accuracy of classification has been finding
Classification Tree and Naïve Bayes. First of all, data was very high with FV2.
organized and then extracted some feature vectors by using Receiver operating characteristics (ROC) graphs that are
power of subbands and some statistical functions. 5-fold commonly used in medical decision is a plot of class
cross-validation is performed on all data when determining sensitivity against specificity as a threshold parameter is
the success of classification methods. varied. In this study, the epilepsy/non-epilepsy decision
It can be seen that kNN is the most successful methods with threshold is varied in the range 0–1 as shown in Fig 1.
98% CA for FV2 and 96% CA for FV3 as shown in Table III.
Meanwhile, Naïve Bayes is the most successful method for
FV1 with 69%. However, at the same time kNN is more
successful for FV1 than other methods except for Naïve
Bayes. Consequently, we can say that whereas kNN is the best
classifier for this dataset, Classification Tree is the lowest
classifier.
TABLE III
THE RESULTS OF TEST (EXAMPLES: 100)
Classification Methods
Naïve Random Classification
MLP SVM kNN
FV Bayes Forest Tree
CA 0.6900 0.6500 0.6700 0.6200 0.6400 0.6800
Sens 0.8000 0.9200 0.7800 0.9400 0.6000 0.7600
FV1 Spec 0.5800 0.3800 0.5600 0.3000 0.6800 0.6000
CA 0.9200 0.9000 0.9200 0.9100 0.8600 0.9800
Sens 0.9000 0.9200 0.9200 0.9200 0.8400 0.9800
FV2 Spec 0.9400 0.8800 0.9200 0.9000 0.8800 0.9800
CA 0.9100 0.9100 0.8900 0.9200 0.8900 0.9600
Sens 0.9000 0.9000 0.9200 0.9400 0.8400 0.9800
FV3 Spec 0.9200 0.9200 0.8600 0.9000 0.9400 0.9400
Confusion matrix that is true and false predictions in each
feature vectors including FV1, FV2 and FV3 are given in
Table IV. Whereas Naïve Bayes classifier has been found 40 Fig. 1. ROC curve for FV2 - epileptic data
____________________________________________________________________________________________________________
769
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The ROC area is an effective method in order to compare Ann. Biomed. Eng., vol. 37, no. 12, pp. 2626–2630, 2009.
the performance of different features. As is known, the area [9] S.-H. Lee, J. S. Lim, J.-K. Kim, J. Yang, and Y. Lee,
under the 1 value in ROC curve shows the excellent test “Classification of normal and epileptic seizure EEG signals using
results, whereas the area under the 0.5 value in ROC curve wavelet transform, phase-space reconstruction, and Euclidean
distance.,” Comput. Methods Programs Biomed., vol. 116, no. 1,
shows the unsuccessful test results. The gathered results pp. 10–25, Aug. 2014.
showed that statistical features for this new dataset could be
[10] K. Fu, J. Qu, Y. Chai, and Y. Dong, “Classification of seizure based
tried on similar dataset that includes larger data in order to test
on the time-frequency image of EEG signals using HHT and
and improve. In the next study, it is planned to develop a SVM,” Biomed. Signal Process. Control, vol. 13, pp. 15–22, Sep.
system that allows distinguishing the differences and 2014.
similarities between seizures and artifacts because even [11] V. Joshi, R. B. Pachori, and A. Vijesh, “Classification of ictal and
experts can give wrong decisions besides the computer seizure-free EEG signals using fractional linear prediction,”
systems. Biomed. Signal Process. Control, vol. 9, pp. 1–5, 2014.
REFERENCES
[1] E. R. Kandel, J. H. Schwartz, and T. M. Jessell, Principles of neural
science, vol. 4. McGraw-Hill New York, 2000.
[2] H. Adeli, Z. Zhou, and N. Dadmehr, “Analysis of EEG records in
an epileptic patient using wavelet transform,” J. Neurosci. Methods,
vol. 123, no. 1, pp. 69–87, 2003.
[3] L. D. Iasemidis, D.-S. Shiau, W. Chaovalitwongse, J. C.
Sackellares, P. M. Pardalos, J. C. Principe, P. R. Carney, A. Prasad,
B. Veeramani, and K. Tsakalis, “Adaptive epileptic seizure
prediction system,” Biomed. Eng. IEEE Trans., vol. 50, no. 5, pp.
616–627, 2003.
[4] J. Gotman, “Automatic recognition of epileptic seizures in the
EEG,” Electroencephalogr. Clin. Neurophysiol., vol. 54, no. 5, pp.
530–540, 1982.
[5] S. A. Hosseini and M. A. Khalilzadeh, “Emotional stress
recognition system using EEG and psychophysiological signals:
Using new labelling process of EEG signals in emotional stress
state,” in Biomedical Engineering and Computer Science
(ICBECS), 2010 International Conference on, 2010, pp. 1–6.
[6] S. P. Kumar, N. Sriraam, P. G. Benakop, and B. C. Jinaga,
“Entropies based detection of epileptic seizures with artificial
neural network classifiers,” Expert Syst. Appl., vol. 37, no. 4, pp.
3284–3291, 2010.
[7] U. Orhan, M. Hekim, and M. Ozer, “EEG signals classification
using the K-means clustering and a multilayer perceptron neural
network model,” Expert Syst. Appl., vol. 38, no. 10, pp. 13475–
13481, Sep. 2011.
[8] S. Aydın, H. M. Saraoğlu, and S. Kara, “Log energy entropy-based
EEG classification with multilayer neural networks in seizure,”
____________________________________________________________________________________________________________
770
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
murat.toren@erdogan.edu.tr
+
Department of Electric and Electronic Engineering, Ataturk University, Erzurum, Turkey
mcelebi@atauni.edu.tr
Abstract— Electrical machines technology has become In essence, real criterion is considered to be torque calculation.
too dependent on finite element method (FEM) programs in our This is because magnetic flux distributions of this machine,
time. In this study, three programs that are sold commercially which has been produced commercially, have already been
are compared over a 4-poles asynchronous machine of 1.1 kW. designed within the acceptable limits. In all simulations, the
With this comparison, it is investigated how close the values material is used with a BH characteristic given in Fig. 1 The
obtained by convenience, appearance, integration and, in computer system that is used for simulations is given in Table
particular, torque calculations with the real values will be. For I and specifications of the test machine in Table II.
this, the mechanical torque value of the machine is obtained first
by using the machine’s foreknown geometrical parameters and
electrical parameters found by tests. Then, the magnetic flux line
and density distribution and the obtained reluctance / real torque
values are compared between FEM programs.
I. INTRODUCTION
In recent years, the FE software package has been widely
used in applications. The programs using these applications
are Ansoft, Ansys and Cedrat. Ansoft is the most common
program among these publications although all types of
machine and analysis are available in Ansys. Ansys is the
world's most widely used FE software package program.
Cedrat is not a very popular program.
In this study, FEM software packages were compared starting Fig.1. BH curve of material used in simulations
from real values on a 4-pole asynchronous machine of 1.1 kW
TABLE I
[1]. As a result, it is proposed to give an idea for electrical
machine designers using finite element technology. Although SYSTEM CONFIGURATION
this test was performed only for the asynchronous machine, it Computer type Laptop
is generally useful for demonstrating the advantages and Manufacturer Dell
disadvantages of these programs. As a matter of fact, as a CPU AMD Turion 64x2 TL-50 1.60 Ghz
result of the variations and difficulties between theoretical
Physical RAM 2 GB
calculation and reality due to the working manner of the
Graphic Card ATI Radeon Xpress / Shared system
asynchronous machine, the closer a simulation program for
memory
such a type of machine to the real value the more acceptable
OS Windows 7 Ultimate 64 bit
performance given by the simpler direct-current machines.
____________________________________________________________________________________________________________
771
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
TABLE II
TEST MACHINE SPECIFICATIONS
h j2 (mm) (4)
Power / Frequency / 1,1 kW / 50 Hz / 3-Y 2k fe LB j 2
Phase-Connection Type
Pole number / Stator 4 / 2,7 A / 1380 rpm In here, the rotor yoke height is taken as 1.2 T. Leakage
current / Speed reactance of stator winding is,
Efficiency / Nominal & 78 % / 7,7 & 16,9 Nm
Starting Torque 2
Stator & Rotor slot 36 & 28 / 65 mm X 1 0.5 w121 (/phase) (5)
f
numbers / Core Length
(L) where, 1 = 12.46 is the leakage conductance of stator winding.
Number of turns (w1)/ 66 / 0,51x2 mm Then, the leakage reactance of rotor bars reduced to stator is
wire diameter obtained from (Eq (6)).
Winding type / Coil 1-layer / 8
pitch (AAAZZZBBBXXXCCCYYY) X 21 1,6u1 22 (/phase) (6)
Stator (bda) / Rotor tooth 3,76 / 4,79 mm
widths where, 2 = 19.63 is the leakage conductance of stator
Stator (Da)& Rotor / 135 & 79,1 / 30 mm winding and u1 = 6.67x104 is the conversion rate. The leakage
Shaft diameters factor of the rotor is obtained as,
Stator & Rotor slot teeth 2,35 & 1 / 0,5 mm
gap / air gap X 21
Stator / Rotor slot 15,55 / 15 mm 2 1 (7)
X1
heights
Then, nominal and startup moments found from the equivalent
Initially analytical expressions are obtained from the values
circuit parameters are obtained as follows, respectively:
determined by the company. After calculating emf in the
stator as 192 V by using the air gap magnetic flux density as
T1 1 1
0.7 T and heyland leakage factor, the number of conductors (8)
per slot is found, by taking the equivalent pole length factor X k T1 X 1 Tl 2 X 21
= 0.76, as follows,
By using the above relationships, the nominal moment,
2m E1
z0 (1)
N1 4,44 fk w 60 U
2
R21 / s
M dn m (Nm)
(1) (9)
2ns 3 R1 T1R21 / s 2 X k2
In here, z0 is found as 69 and N1 is the number of stator slots;
kw = 0.96 is the winding factor and = 0.0022 Wb is the air
gap flux. The stator slot pitch induction is found from Eq (2): and the startup moment,
2
0 60 U R21
Bda (T) (2) M dy m (Nm) (10)
k febda B0 (1 0,6667 ) 2ns 3 R1 T1R21 2 X k2
(2)
where, Bda is stator slot pitch induction (T); 0, stator slot pitch; are found as 7 and 16.84 Nm, respectively. Here, U is the
bda, stator tooth width; kfe = 0.95 is the iron packing factor; voltage at motor end; ns is synchronous speed; m is the phase
B0 is the air gap induction (T) and = 0.0115 is heyland number; T1 is the conversion rate; R1 and X1 are resistance and
leakage factor. The height of the yoke is found by using the reactance of stator winding, respectively, and Xk is the short
below equation. circuit reactance.
____________________________________________________________________________________________________________
772
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
visual quality, performance, advantages and disadvantages of The graphical interface of Maxwell v2.6 used for 2D
this program in 2D and 3D applications were investigated and, applications is similar to the DOS programs.
information was given about them. - Direct AutoCAD integration is weak:
Integration is not possible for direct intervention in
Advantages of Ansoft / Maxwell; the AutoCAD files.
- Moderate accuracy;
- Electrostatic analysis and a geometry module are Magnetic flux density distribution is shown in Figure
available for electrical machines: 2. Although the reluctance torque values obtained as
The geometry of the machine is generated with 5.53 Nm is not the realistic value, Ansoft can easily
rmxprt integrated as a module by using Maxwell bring the simplicity together with the mastership.
v11.1 2D/3D program and this can be used with the (Torque value is -5.53 Nm).
versions 2.4 or 2.6 for 2D analysis. - Inadequate help menus:
Almost every electrical machine geometry in 2D or Help files achieved by F1 button on the program
3D can be generated easily by v11.1 developed for menu are inadequate.
Maxwell 3D applications and the rmxprt module
integrated into it. In conclusion, opportunities offered by Ansoft[14] will be
- A translator module is available for AutoCad file: very useful for the researchers who will begin Magnetic field
A useful module, called a translator, is available in static analysis with finite elements method for the first time.
Maxwell control panel that converts Autocad DXF
files, etc. to sm2. B. CEDRAT
- Very easy-to-use: Preflu2d v10.3.2 program [15-18] of the Cedrat group is
The geometry part can be easily skipped in this analyzed here. Cedrat classifies its products as 2D and 3D.
program by converting the AutoCAD files to sm2 The visual quality, performance, advantages and
format with a translator. In addition, any electrical disadvantages of this program in 2D and 3D applications were
machine generated with rmxprt program can easily investigated and, information was given about them.
be exported in 2D. The BH curve can easily be
obtained directly from txt file. Advantages of Cedrat / Preflu:
- Winding tool is available: - Electrical machines module is available
Winding tool availability in the program provides The design part of the electrical machine is the result
convenience to the designer. of an original design. In Cedrat’s design module, the
- Graphical performance is high: part that is moving or the part which is fixed can be
Maxwell 11.1 provides an excellent graphics entered and a good machine geometry can be created.
performance in 2D/ 3D applications. Drawing can be The meaning of the parameters in the designed
easily managed with handy shortcuts using the asynchronous machine geometry can be displayed in
mouse. The size of the shape does not cause much the figures (with animations) available on the ready
decrease in performance. Even a synchronous overlayers.
generator with an internal diameter of 2 m, an outer - Armature reaction is taken into account:
diameter of 2.8m and a depth of 30 cm takes up the a In Figure 3, it can be seen that the magnetic flux lines
space of approximately 600 MB and offers a superior are different from the Ansoft and Ansys programs.
graphics performance. This is Cedrav taking the armature reaction into
- Graphical plotting specifications are sound: account and this really seems to be a significant
Plotting can be performed easily and a plotting form advantage by itself.
can be obtained including the number of lines. - Winding tool is available:
- High solution speed: The winding tool available in the program provides
Solution part progress is very fast convenience to the designer.
- Memory and CPU performance is good - Mechanical torque calculation is available:
Program memory is suitable as drawings do not A mechanical torque calculation by Cedrat is a good
occupy too much space and the processor specification although it can hardly approach the real
performances provide the results in a short time. torque value as can be seen in Table 3.
- Electrical circuit module is available:
Disadvantages; The electrical circuit diagram of the machine is
handled separately by the Cedrat. The values of
- Static Analysis: circuit elements cannot be seen with these
It does not allow professional solutions with respect specifications offering the possibility of making any
to the static analysis. kind of connections. Furthermore, one has to enter
- DOS view for 2D applications: resistance and inductance values of each phase into
____________________________________________________________________________________________________________
773
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
the program, as external data requires the use of other in the help menus. Incomprehensible points and
programs. inconsistencies exist between the sample
- Graphical interface is good asynchronous machine in the help file and the present
The graphical interface is designed well although it is version and, there is no information regarding the use
not as good as Ansoft. of the over layers.
- Graphics performance is satisfactory: - Intensive memory and disk usage:
Graphical result unit of Cedrat generates values When a file is opened in Preflux2D, it occupies more
suitable with what to do. Understanding of the results than 1 GB (ram+virtual memory) according to status
takes a short time with the help files. Graphics are of the computer hardware. A number of programs
found as desired and features of the results part make running at once, poses a problem even in today’s
a good impression. computers. Moreover, with the addition of solution
- Graphical plotting specifications are very handy program block, results in a memory of 2 GB. It is
It is possible to obtain a variety of parameters difficult to open two files together in Cedrat. It is a
graphically and animation features are fairly time consuming process to open your project.
advanced in graphic illustrations. Harmonic analysis
can be represented graphically in spectral analysis In conclusion, even if you believe you are doing
and this makes Preflu of the Cedrat one-step ahead. everything correctly in Cedrat/ Preflu[19], the odd
In addition, all graphics can be shown in the same bugs may be found(such as node error). Technical
axis in a single window. support is needed for this program throughout the
- Solution speed is moderate time that you are inexperienced in using it.
It is observed that the solution speed generates
solutions at normal speed compared to other C. ANSYS MULTI PHYSICS
programs when it is considered to make the memory In this part, the Ansys version v11 and the Ansys Multi
stacked. Physics program [20-29]are analyzed.
____________________________________________________________________________________________________________
774
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
c) Cedrat
____________________________________________________________________________________________________________
775
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
b) Cedrat Disadvantages:
____________________________________________________________________________________________________________
776
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
777
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 9 The slipping of the regions of rotor bars after pattern process. Notice
the location on axis.
TABLE III
Fig. 10. 3D geometry in Workbench after the import process
COMPARISON OF TORQUE VALUES (NM)
TABLE IV
COMPARISON OF FEM SOFTWARE PACKAGES
Electric
Machine Memory /
ry Auto CAD file Graphic Disk Overall
Software Version Usage Module integration Performance performance performance Accuracy
Ansoft /
2.6 / 11.1 Easy Exist Poor Very Good Good Good Average
Maxwell
Cedrat /
10.3.2 Difficult Exist - Good Average Good Average
Preflu
Ansys /
multiphysic 11 Hard No Excellent Very Good Good Very Good Very Good
s
Ansys /
11 Average No Good Bad Poor Poor -
Workbench
____________________________________________________________________________________________________________
778
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
779
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
780
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—In this study, modelling and simulation studies of DG in abc/dq-SRF. That is because PR controller is to transform
based fuel cell (FC) and solar cell (SC), connected to the common the dc type regulator network into an equivalent ac regulator,
which could directly achieve zero steady-state error in
direct current (DC) bus system to provide energy for various stationary frame [14]. Moreover, PR controllers are used in
loads. In order to synchronising of energy sources with grid, DG systems due to their superior behaviour over the
using three-phase phase locked loop (PLL) technique based park traditional PI controllers, when filtering sinusoidal signals.
Due to PR controller could achieve a very high gain around
transformation which is one of the most common methods in the given resonance frequency, thus being capable to
literature. Power control methods based dq/abc synchronous eliminate the steady-state error when this controller tracking
reference frame (SRF) or proportional resonant (PR) are used in
sinusoidal reference, which could be suitable for the using of
stationary reference frame control strategy. The advantages of
the DG systems. The performances of two controllers are the Proportional Resonant (PR) controller in opposition to PI+
discussed in this paper. Simulation results are obtained for dq/abc synchronous reference frame controller are a less
complexity of the control structure and a good capability to
various scenarios at the designed and created simulation model
compensate harmonics. Compared with Proportional Integral
of DG+Grid+Load system. The system is modeled and simulated + synchronous reference frame (PI+SRF) controller, the PR
by using PSCAD/EMTDC software package. controller eliminates abc-to-dq and dq-to-abc frame
transformations which significantly reduces computation
Keywords— Distributed Generation, Renewable Energy Sources, burden and PLL errors. However PI+SRF controller could
PLL, Synchronous Reference Frame, PR controller improve the dynamic response to power change and
unbalanced faults [4]-[5].
In this paper, DG system based on RES, such as
I.INTRODUCTION characteristic of fuel cell and solar cell with maximum power
point (MPP) are investigated. It is analysed the performances
Nowadays, environmental issues and growing demand for of PI+SRF (abc/dq) and PR based Power Controllers in grid
lower cost electricity with reliability and security, eventuate in tied distributed generation systems. Reliable performances of
an increased use of small electrical power generation located controllers under grid voltage drop and references power
close to end users which are called distributed generation change cases are discussed. Simulation results are provided to
(DG). There is an increasing use of renewable power sources illustrate the effectiveness of the proposed system. The
(RES) in DG systems because the growing demands for validity of proposed topology and control methods are verified
electrical energy [1]-[20]. The DG systems utilize RES, such with PSCAD/EMDTC software package.
as solar cells, fuel cells, wind turbines and micro turbines [2].
Grid connected inverters as one of the most importantly parts
in DG systems, their current and power loop controllers are II.DG SYSTEM CONFIGURATION
always the investigated [4].
In The grid tied DG systems, should be carefully designed The implemented DG system in PSCAD/EMTDC is shown
and controlled in order to get an optimal power under fault in Fig. 1 where FC/Solar cell (SC) combined form is the main
conditions [21]. The reason to apply PR in DG system design form of DG system. DG system is connected to DC bus
is to eliminate steady-state error [16]. The traditional system. Voltage source inverter (VSI) connected AC side with
Proportional Integral (PI) controller in abc/dq- synchronous grid and loads. The output voltage from the inverter contains
reference frame (SRF) is the common solution to solve this high order harmonics which must be eliminated and thus, a
problem. But the complexity of displacement the ac SRF back low LC filter and PR current controller are used to reduce the
to dc SRF where the PI controller can be used is also brought harmonic contents in the phase current and the output voltage.
into the control system design. In this case, a transformed The technical details of each part of the DG system are given
regulator known as PR controller is preferred considering the in the following subsections.
advantage of avoiding modulation and demodulation process
____________________________________________________________________________________________________________
781
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
pll
____________________________________________________________________________________________________________
782
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Fig. 2 Under variable DC loads; (a) DC Bus Voltage, (b) Boost Converter Current and (c) Boost Converter Power
____________________________________________________________________________________________________________
783
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
b. abc/dq-SRF
III.ANALYSIS OF CONTROLLERS and 3p-PLL
PI controller cannot eliminate steady-state error for the
a. SRF-PLL alternative current (AC) control as in three-phase grid tied DG
system. To solve this problem, a mathematical transformation
A PLL is frequency control system that sycronize for grid abc/dq-SRF has been universally applied system. Abc/dq-SRF
voltage and phase voltage [19]. A synchronous Reference transformation is short for direct quadrature zero
Frame PLL (SRF-PLL) is mainly used for tracking the phase transformation which is usually applied to the three-phase AC
angle in case of 3-phase signals [18]. SRF-PLL is widely used circuit analysis. The main reason to apply this method is for
in three-phase grid tied DG system for its simple simplification of analysis. In balanced three-phase systems,
implementation and fast and accurate estimation of the abc/dq-SRF transformation transform three AC quantities to
phase/frequency in ideal grid conditions. But, the three-phase two directive current (DC) quantities into two-phase frame
voltage vector in abc stationary frame need to be transformed such as direct axis (d) and quadrature axis (q) [17]. Simplified
to dq rotating reference frame in park's transformation as calculations can then be carried out on these imaginary DC
shown in eq.(6) and (7). Conventional SRF-PLL allows fast quantities. In this case, the application of dq0 transformation
and accurate estimation of phase angle and grid-voltage not only reduces the complexity of system analysis but also
frequency in ideal condition [15]. The basic structure of SRF- provides an available DC operation condition for PI controller
PLL based park and clarke transformation is illustrated in Fig. [14].
7. Park's transformation-which is given in eq. (6) and (7);
2π 2π
⎡ cos(θ) cos(θ − ) cos(θ + ) ⎤
⎢ 3 3 ⎥
2 2π 2π I
I = ⎢−sin(θ) −sin(θ − ) −sin(θ + )⎥ I (6)
3⎢ 3 3 ⎥
I
⎢ √2 √2 √2 ⎥
⎣ 2 2 2 ⎦
⎡ cos(θ) √2⎤
−sin(θ)
⎢ 2⎥ I
pll 2⎢ 2π 2π √2⎥ I
I = (7)
3 ⎢cos(θ − 3 ) − sin(θ − 3 ) 2 ⎥
⎢ ⎥ I
⎢ 2π 2π √2⎥
⎣cos(θ + 3 ) − sin(θ + 3 ) 2 ⎦
Fig. 5 Basic structure of SRF-PLL. Modelled and simulated the PI+SRF controller based park
transformation in PSCAD/EMTDC are given in Fig.7.
SRF-PLL is an important technique for grid
synchronization along with control inverter. Fig. 6 shows that
____________________________________________________________________________________________________________
784
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
L
L pll
pll
c. The Proportional Resonant (PR) Current Controller adequate to eliminate harmonic influence caused by grid
voltage harmonics [7]-[8].
Although its popularity, a PI controller is not able to track a Fig. 8 shows that ω is grid fundamental angular frequency
sinusoidal reference without the steady-state error (magnitude and ki is a constant which is carefully selected to shift the
and phase). PI controller such as lack of tracking a sinusoidal controller’s magnitude response vertically [6].
reference with zero steady-state error and poor disturbance
rejection capability [6]. To overcome the mentioned problems kωs
of conventional PI controllers in SRF, PR current controller is G(s) = k + (8)
s + 2ω s + ω
proposed. The PR controller provides a high gain and zero
steady-state error. Besides its simplicity, the system dynamics And kp and ki are the proportional and integral gains of PR
is almost not affected by the harmonic compensation terms. controller respectively, and ω is cut-off angular frequency.
However, the PR controller is difficult to implement in reality. The transfer function shows a no-ideal PR controller which
It has two problems. Firstly, the infinite gain introduced by a has lower gain and wider bandwidth than ideal PR controller
PR controller leads to an infinite quality factor which cannot at the resonant frequency [6]. The power/current controller of
be achieved in an analog or a digital system processor (DSP). the DG system with PR controller is illustrated in Fig. 9.
Secondly, the gain of the PR controller is much decreased at
other frequencies and it is not
kp
2wc
ki 2wc
2
wo
Fig. 8 PR Current Controller Block Diagram [10]
____________________________________________________________________________________________________________
785
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
pll
pll pll
IV.SIMULATION RESULTS
Parameters Values
R filter 0.005 Ω
L filter 0.5 [mH]
C filter 16 [μF]
DC bus voltage 800 V
Grid voltage 380 VL-L rms
____________________________________________________________________________________________________________
786
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
787
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
1
Ege University, Ege Vocational High School, Bornova, İzmir-Türkiye
seyhan.coskun@.ege.edu.tr
2
Ege University, Electric-Electronics division, Faculty of Engineering, Bornova, İzmir-Türkiye
yavuz.oztürk@ege.edu.tr
alpkahraman@ieee.org
Abstract-Optical fibers which have been in human life for 30-40 is guided with principle of photonic band gap. Light with
years, were used in long haul telecommunication links in first some wave length does not propagate in the fiber. Guidance
decade. But today, fiber optic is used every distance of of light in the fiber occures in certain bands[14].
communications. Beside communications, fiber optic
Technologies are spreading out in optical sensor technologies and
networks, industry and medical in today. Classical optical fibers
II. MODES OF PCF
is standardized and has come to limit of refinement. As a new In core and cladding of PCFs, various modes ocur due to
generation of optical fiber, with feature of very flexible design, guidance conditions determined with parameters of geometry
photonic crystal fibers (PCF) are promising new horizones. PCFs and material of PCF and light. A number of modelling
are designed as endlessly single mod fiber in very wide range of methods have been offered to analyse modes, mode
wavelengths. Dispersion, birefringence and nonlinearity can be properties and propagation characteristics of PCFs[15-23].
obtained in desired characters by changing geometrical Considerable modelling methods are effective-index, the
parameters of PCFs. plane-wave expansion, localized basis function, finite-
In this simulation study, fundamental and high order core
element, finite difference, finite difference time domain
modes, cladding modes, fundamental space filling mode (FSM)
and high order space filling modes were found. Effective
(FDTD), beam propagation and multipole method.
refractive indices of the modes were calculated. Modelling and after modelling processes give patterns of
modes, effective refractive indices of modes, mode areas,
Keywords-Fiber optic, Photonic Crystal Fiber, Fiber optic confinement losses, dispersion characteristics. Effective
sensor, Effective refractive index refractive indices of modes are very important parameters for
especially fiber optic sensor systems. Cladding modes are
I. INTRODUCTION very critical for PCF sensor in respect to analyse interaction
Photonic crystal fibers (PCF) are new generation fiber liquid or/and gas materials in holes of the PCFs and guided
optic waveguides which have different structures and features light [24,25].
from classical fiber optics[1,2]. PCFs are produced from a Fundamental Space Filling Mode (FSM), is fundamental
single material. They have air holes around core which are mode which propagates in cladding[3]. FSM is used effective
parallel to core and run along the fiber. Air holes are cladding index and the mode with highest refractive index in
positioned in many rings around core. PCFs whith various cladding[26].
geometric features (hole diameter, distance of holes, number nFSM and βFSM are supposed to effective refractive index
of rings, etc.) of cross section show different guiding features. and propagation constant of FSM, respectively.
PCFs have many advantages over classical fibers. PCFs have nFSM = βFSM/k0
wide range of wavelength of single mod guidance, dispersion, neff is valid for index-guided mode in core. Refraction
non-linearity and birefringence control and endurances to index of silica glass in core of PCF is up to nco This inequality
bending[1-7]. is valid:
PCFs have two dimensions periodicity. At least one nco > neff = β/k0 > nFSM
characteristic dimension is about to magnitude of wavelength
of guided light[8]. PCFs are called as crystal fibers, holey
fibers and microstructure optical fibers (MOF), too. Since air
holes arrayed in two dimensions in periodic structure, PCFs is
a special tip of MOFs[9]. Core of PCF may be solid or air
hole (Fig. 1) [10 -13]. Photonic crystal fibers with air hole
core is also called hollow core fibers.
Guided light propagate with different mechanism in 2
holey fibers which have different basic structures. In fibers
with solid core, light propagate with modified total internal
reflection. Cladding that is the section with air holes shows
lower refraction index than core and provides guiding of light.
Fig. 1 Holey fibers: a) Solid core PCF[28], b) Hollow core fiber [6].
That is also called index guiding. In hollow core fibers, light
1
____________________________________________________________________________________________________________
788
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
(e)
(a)
(f)
2
____________________________________________________________________________________________________________
789
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
REFERENCES
[1] P. Russell, 2003, “Photonic Crystal Fibers”, Science, 299, 358.
[2] J. Broenga, D. Mogilevstev, S. E. Barkou, A. Bjarklev, “Photonic Crystal
Fibers: A New Class of Optical Waveguides”, Optical Fiber Technology 5,
305-330, 1999.
[3] T. A. Birks, J. C. Knight, and P. St. J. Russell, “Endlessly single-mode
photonic crystal Fiber”, Opt. Lett., Vol. 22, No. 13, 1997
[4] J. C. Knight, T. A. Birks, P. St. J. Russel, and D. M. Atkin, ‘‘All-silica
single-mode optical fiber with photonic crystal cladding,’’ Opt. Lett., vol. 21,
no. 19, 1996, 1547
[5] J.C. Knight, T.A. Birks, R.F. Cregan, P.St.J. Russell, J.-P. de Sandro,
(g) “Photonic crystals as optical fibres, physics and applications”, Optical
Materials, 11, (1999), 143 – 151
[6] P. S. J. Russell, “ Photonic-Crystal Fibers”, Journal of Lightwave Tech.
Vol. 24, No 12, 2006, 4729
[7] K. Saitoh, M. Koshiba, T. Hasegawa, and E. Sasaoka, “Chromatic
dispersion control in photonic crystal fibres: application to ultra –flattened
dispersion,” Optics Express, Vol. 11, 2003, pp. 843-852.
[8] R. M. De La Rue, S. A. De La Rue, “ Introduction to Photonic Crystals
and Photonic Band-Gaps”, in “ Photonic Crystals: Physics and Technology””,
Editors: C. Sibilia, T.M. Benson, M. Marciniak, T. Szoplik, Springer-Verlag,
Italia, 2008
[9] C. M. B. Cordeiro, M. A. R. Franco, G. Chesini et al., “Microstructured-
core optical fibre for evanescent sensing applications,” Optics Express, vol.
14, no. 26, pp. 13056–13066, 2006.
[10] P. Russell, “Photonic Crystal Fibers”, Science 299, 358. , 2003
[11] R. F. Cregan, B. J. Mangan, J. C. Knight, T. A. Birks, J. Russell, J.
(h) Roberts, D. C. Allan, “Single-Mode Photonic Band Gap Guidance of Light in
Air”, Science, V. 285 3, 1999
Fig. 2 Over simulations for wavelength of 600 nm, obtained electric field [12] C. Sibilia, T.M. Benson, M. Marciniak, T.Szoplik ( Editors), “Photonic
Crystals: Physics and Technology”, Springer, Verlag, Italia, 2008
intensities of various modes: a) LP01 core fundamental mode, b) LP01 at FSM
[13] H. Ademgil, M. Toycan, S. Haxha, “Fotonik Kristal Fiber Yapılarının
layer, c) LP11 at FSM layer, d) LP31 at FSM layer, e) LP11 cladding mode, f) Kıvrılma Kaybı Analizi”, SIU, 2012
LP02 cladding mode, g) LP31 cladding mode, h) LP32 cladding mode. [14] N. M. Litchinitser, S. C. Dunn, P. E. Steinvurzel, B. J. Eggleton, T. P.
White, R. C. McPhedran and C. M. de Sterke, “Application of an ARROW
model for designing tunable photonic devices”, Opt. Express 12, 1540–1550,
IV. CONCLUSION 2004
In this simulation study, Comsol Multiphysics which is [15] R. Buczynski, “Photonic Crystal Fibers”, Acta Physica Polonica A, Vol.
based on FEM, one of the PCF modelling method, is used. 106, No. 2, 2004
[16] K. Saitoh, and M. Koshiba, “Numerical Modeling of Photonic Crystal
The simulation was implemented for PCF of NKT Photonics Fibers”, Journal of Lightwave Tech.,Vol. 23, No. 11, 2005
LMA-20. Lowest wavelength at which core fundamental [17] Y. Li, Y. Yao, M. Hu, L. Chai, and C. Wang, “Improved fully vectorial
guided mode was detected have been ascertained as 560 nm effective index method for photonic crystal fibers: evaluation and
(cut off wavelength). The modes were searched for 560 -2300 enhancement”, Applied Optics Vol. 47, No. 3, 2008
[18] Y. Li, C. Wang, M. Hu, “A fully vectorial effective index method for
nm of wavelength range. Modes and their effective refractive Photonic crystal fibers: application to dispersion calculation”, Optics
indices of core, cladding and FSM were obtained. For this Communications, 238, 29–33, 2004
PCF, it has been seen that maximum effective refractive [19] A. D. Varshney and R. K. Sinha, “Propagation Characteristics of
indices of guided modes are at the modes of FSM layer Photonic Crystal Fiber: Scalar Effective Index Method & Fully Vectorial
Effective Index Method”, Adv. Studies Theor. Phys., Vol.1, no. 2, 75 – 85,
(figure-3). 2007
[20] D. Mogilevtsev, T. A. Birks and P. St. J. Russell, “Localized Function
Refractive indices of PCF modes Method for Modeling Defect Modes in 2-D Photonic Crystals”, Journal of
1.46
n
coreLP01 Lightwave Tech., Vol. 17, No. 11, 1999
fsmLP01
1.455
fsmLP11
cladLP11
[21] B. Kim, H. Kim, K. Naeem, L. Cui and Y, Chung, T. Kim, “Design,
cladLP02
cladLP21
Fabrication, and Sensor Applications of Photonic Crystal Fibers”, Journal of
cladLP31
1.45 cladLP03 the Korean Physical Society, Vol. 57, No. 6, 1937-1941, 2010
[22] Z. Zhu and T. G. Brown, “Full-vectorial finite-difference analysis of
Effective refractive index
1.445
microstructured optical fibers”, Optics Express, Vol. 10, No. 17, 853, 2002
1.44
[23] F. J. Marcou, D. Pagnoux, and P. Roy, “Complete Analysis of the
Characteristics of Propagation into Photonic Crystal Fibers, by the Finite
1.435 Element Method”, Optical Fiber Technology 6, 181191, 2000.
[24] H. C. Park, I. K. Hwang, D. I. Yeom, and B.Y. Kim, “Analyses of
1.43
cladding modes in photonic crystal fiber,” Opt. Express, vol. 15, 15154-
15160, 2007
600 800 1000 1200 1400 1600 1800 2000 2200 2400
Wavelength [nm] [25] N. Khurram, L. V. Nguyen, K. Alameh and Y. Chung, "Cladding Modes
Analysis of Photonics Crystal Fiber for Refractive Index Sensors Using Finite
Element Method", CLEO-QELS, (2010/05)
3
____________________________________________________________________________________________________________
790
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
[26] Y. Li, C. Wang, Y. Chen, M. Hu, B. Liu and L. Chai, “Solution of the
fundamental space-filling mode of photonic crystal fibers: numerical method
versus analytical approaches”, Appl. Phys. B:Lasers and Optics, vol. 85, 597–
601, 2006.
[27] A. G. Prabhakar, B. A. Peer, C. A. Kumar and D. V. Rastogi, “Finite
Element Analysis of Solid-Core Photonic Crystal Fiber”, SCES 2012
[28] F. Tian, J. Kanka, B. Zou, K. S. Chiang, and H. Du, “Long period
gratings inscribed in photonic crystal fiber by symmetric CO2 laser
irradiation”, Opt. Express, Vol. 21, No. 11, 13208, 2013
4
____________________________________________________________________________________________________________
791
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract— Tracking mobile objects has always attracted people’s In case of GPRS or generally internet connection loss, there
interest and over time it has become a necessity to keep records of might be data loss if the device on the vehicle does not have a
them. Various solutions have been developed to meet this need. In local storage unit.
particular, the development of GPS and digital mapping system There might some cases that real time tracking of the
has contributed greatly to this solutions. The developed methods
vehicles is not essential and having past records might be
can be divided into two main categories as online and offline
systems. Although the online systems provide advanced features, sufficient. For example, for a school bus debates about
production and operating costs are quite high and have regular following route, timing and obeying speed limits can be
expenses. In this study, we propose a microcontroller based offline illuminated by examining previous records. Similarly, a
vehicle tracking system. The proposed system offers a practical, company can use recorded past data to check its vehicles routes,
low cost and high capacity approach for vehicle tracking systems. departure and arrival times and to estimate amount of fuel
The proposed system calculates the vehicle’s position and speed consumption. This data may also be used as evidence in case of
according to the signals gathered from GPS satellites and records accident or a crime. For such cases, there is no need for real
them to an SD card continuously. The stored data, then, can be time or online tracking systems and having just an offline
transferred to a computer easily whenever needed and the routes
tracking record is sufficient.
that the vehicle followed can be tracked on a digital map using
google earth. Besides, some visualization technics like coloring In this study, a practical and low cost solution is proposed as
routes with regards to time or speed intervals can be used for an alternative to real time or online tracking systems. For this
advanced analyses of the data. The proposed system allows the purpose, an offline tracking system which records the data
GPS data to be recorded as long as 3 months on a 4 GB SD card received from GPS satellites to a pluggable SD card is
continuously. developed. Then, the recorded data can be transferred easily to
any computer when needed and can be monitored on a digital
Keywords— GPS; Digital Map; Offline Tracking; Vehicle map using google earth. Since the GPS data is recorded directly
Tracking; GPS Data Visualizing on SD card as standard NMEA format, there is not a
dependency to any special or commercial software to monitor
I. INTRODUCTION the data on a digital map. Any software able to monitor GPS
The free, open, and dependable nature of GPS has led to the data can be used. In this study google earth is used to visualize
development of hundreds of applications in such areas as the users route on digital maps.
marine, aviation, agriculture, environment, space, surveying,
mapping, public safety, disaster relief, recreation, timing, rail, II. GLOBAL POSITIONING SYSTEM
roads and highways. Although tracking systems was first The Global Positioning System (GPS) is a satellite system
developed for fleet management and shipping systems, today it that provides users with positioning, navigation, and timing
is used in many areas of daily life [1], [2]. information. The GPS system was created by US. Department
Vehicle tracking systems used today are generally online of Defense and has been available to civilians since 1980.
systems. That is, the information like location and speed of the Generally, it is used to track and navigate the mobile objects.
vehicle are sent to a management center in real time. Online The GPS system consists of three segments: The space
systems depend on GPRS or SMS technology to communicate segment, the control segment, and the user segment.
with a management center or a central server. Despite the many
advantages of this system, they have some regular expenses The space segment basically consists of 24 satellites that
beside the initial setup costs. These expenses involves monthly travel around the earth two times a day. The
subscription fee per vehicle for GPRS services, server and arrangements of the satellites on 6 planes allow at least
management center operating and maintenance fees, domain four of them to be visible by receivers from any point on
name fees and server hosting costs. Moreover, if these services the Earth.
are not owned by self, it is needed to subscribe such services
and pay subscription fee monthly for each vehicle to be tracked.
____________________________________________________________________________________________________________
792
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
The control segment consists of ground stations which GPS module transfers data using NMEA 0183 standard.
track the satellites, monitor their transmissions and send Although there would be device specific communication
command and data to keep them in their orbits. protocols, NMEA is a standard protocol supported by all GPS
The user segment is composed of military and civilian receivers. There are several NMEA sentences as GGA, GLL,
applications. The military services use L2 frequency and GSA, GSV, RMC and VTG. These sentences include some
have a more precise positioning and timing capability common and some sentence specific information. [4]
than the civilians. [3], [4] In this study, RMC sentence is used because it includes all
the information required for this application. A sample RMC
III. SYSTEM DESIGN AND IMPLEMENTATION sentence and message format is as follows:
The designed system is described in detail in this section. $GPRMC,053740.000,A,2503.6319,N,12136.0099,E,2.69,
Block diagram of the designed offline tracking system is shown 79.65,100106,,,A*53
in Figure 1. The three main parts of the system are GPS module, Description of the sample RMC sentence is seen in Table 1.
microcontroller and data storage. The GPS receiver module’s TABLE I
function is to calculate the vehicle position according to DESCRIPTION OF AN EXAMPLE RMC SENTENCE
information received from GPS satellites. The microcontroller
Name Example Description
processes the data obtained from GPS receiver module and
RMC protocol
selects desired NMEA sentences. This data is stored to a Message ID $GPRMC
header
pluggable SD card. Whenever needed, SD card can be plugged UTC Time 053740.000 hhmmss.sss
in a computer to monitor the data on a digital map. All A=data valid or
components in the circuit require a 3.3V voltage source. The Status A
V=data not valid
device includes a voltage regulator on board and vehicle’s Latitude 2503.6319 ddmm.mmmm
battery or a separate battery can be used as a power supply. N/S Indicator N N=north or S=south
Longitude 12136.0099 dddmm.mmmm
E/W Indicator E E=east or W=west
Speed over ground 2.69 TRUE
GPS MODULE Course over ground 79.65
Date 100106 ddmmyy
Not Available, Null
Magnetic Variation
CONTROL UNIT Field
E=east or
Variation Sense
W=west(Not Shown)
STORAGE UNIT
A=autonomous,
Mode A
IN VEHICLE UNIT
D=DGPS, E=DR
Checksum *53
Fig 1. Block diagram of the designed offline vehicle tracking system End of message
<CR> <LF>
termination
A. GPS Module
The GPS module is used to get the current position and speed
of vehicle. In this study, LINX RXM-GPS-SR receiver module B. Microcontroller
(shown in Figure 2) is used for getting latitude, longitude, In this project, PIC18F4620 microcontroller is used.
altitude, date and time information. Small dimensions PIC18F4620 has built in UART and SPI hardware for serial
(25.2mm-12.4mm-11.9mm) and low cost of this device is the communication and has a wide range of operating voltage (2-
major reasons for selecting the module. It has an integrated 5.5V) [5]. In this study, both GPS receiver and SD card module
antenna and direct serial interface to communicate with host requires 3.3V power supply. A microcontroller capable of
device. The update rate of the module is 1Hz and position operating at 3.3V simplifies the circuit design and reduces the
accuracy is less than 10 meters. The module is based on the number of power supplies required.
SiRFstar III low power chipset. The module’s high- The microcontroller has two main tasks. First one is to get
performance RF architecture allows it to receive GPS signals NMEA sentences from the GPS receiver. The microcontroller
that are as low as -159dBm. [4] communicates with GPS receiver through RS232 serial
communication protocol. The GPS receiver sends 6 different
NMEA sentences every second. Microcontroller receives all
these sentences and selects the desired ones.
The second task of the microcontroller is to store the selected
NMEA sentences to an SD card. The microcontroller uses build
in SPI hardware to communicate with SD card. The data is
written to a text file in a FAT formatted SD card.
Fig. 2 LINX RXM-GPS-SR receiver module
____________________________________________________________________________________________________________
793
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Is the N There are many digital maps used today. But most well-
character $ ? known and widely used maps are google map and google earth.
Y Google maps are mostly used for navigational purposes while
google earth is used exploring the world in a more detailed 3D
maps. Although there are many common features between
Receive a character
Save it in a string array google earth and google maps, the main difference is that
google earth is an desktop application and google maps is an
online web application. Google earth is used in this study, since
Is it end of N
it can examine the GPS data directly.
line?
IV. RESULTS AND DISCUSSION
Y
The prototype device prepared for offline tracking is given
N
Is the string array
in Figure 5. The dimensions of the device are 45mm x 65mm x
a RMC sentence? 25mm.
Y
N
Is the data
valid?
Is the file N
Create a file
exist?
Y
Record data to
text file
____________________________________________________________________________________________________________
794
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
795
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
796
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
Abstract—This paper presents a practical comparison study on the Speed at dynamic performance and steady-state can be
speed estimation methods for induction motors by using Direct calculated properly if parasitic effects are wiped away [9].
Torque Control in terms of their estimation performances Using only motor parameters is easy way to estimate actual
especially at startup. Simulation and experimental test studies on speed of the motor. However, the method can be applied in a
various operations characteristics are achieved and presented
limited range. In a study, a hybrid mathematical model, which
together with the evaluations. The estimation techniques are
compared in terms of their capabilities of speed tracking and sustains a low-cost proper design for the industrial applications,
sensitivities for each operation condition. It is concluded that at is presented [10-13]. Also, flux, speed and rotor time-constant
startup operation and in steady state behavior, dynamic can be estimated by using a sliding functions. It guaranteed an
performance of Model Reference Adaptive System (MRAS) is accuracy of the parameters when the difference between the
more stable and effective than sliding mode observers and open- actual current and the observed current is zero [14-18].
loop estimators. However, open-loop speed estimation method has Moreover, model reference adaptive system (MRAS) based
better performance than the others. Dynamic performance of the speed sensorless algorithm can be used for controlling an
speed estimators in DTC control algorithm are tested in a induction machine. In this method, a proportional-integral (PI)
laboratory setup. System configuration includes a low cost
controller is employed in the adaption mechanism instead of
STM32F407VG Discovery board used for the implementation of
the control algorithm, there-phase power inverter which is pure integral because of initial problems [19-23].
designed for this purpose and industrial type of 11 kW induction In this study, some of these methods used for control of the
motor. induction motors with direct torque control (DTC) without any
speed sensor, such as open-loop speed estimator, sliding mode
Keywords— Speed Estimation, Sensorless Control, Direct Torque observer based speed estimator and MRAS based speed
Control, AC Drives, Sliding Mode Estimator, MRAS, Open-Loop estimator are compared with each other on various conditions.
Estimator, Performance Investigation, Induction Motors. And, the results obtained by the experiments are presented and
introduced a discussion on the results accompanied with the
I. INTRODUCTION evaluations.
In recent years, importance of the adjustable-speed drives
used in industrial and electric vehicle applications has been II. INDUCTION MACHINE MODEL
increased. Modern AC motor drives produced in last decades Dynamic model equations of the induction motor used in
are composed of microcontrollers, electronic components, many control algorithm are given by the equations 1 to 5. The
digital or analog controllers and sensors. However, many equations include not only steady state characteristics but also
control techniques consisted of more complex algorithms based starting behavior. The model is represented by differential
on the sensorless control system are preferred due to high prices equations referred to d-q reference frame rotating at the
of the sensors, their low safety operations and drawbacks of the synchronous speed with a view to simplify the analysis of three-
noise effects. The enhancement in processor technologies, phase circuits.
advanced microcontrollers have made it possible to carry out The parameters used in the equations are stator winding
more complex control algorithms with high efficiency. resistance and inductance ( , ) , rotor bar resistance and
Due to avoiding of sensor drawbacks and reducing the cost inductance ( , ) and mutual inductance( ).
of the inverters, estimation of the actual speed instead of the
measurements is more available for the researchers and =β λ + βω λ −k i +k V (1)
engineers. In literature, control algorithms designed with
feedback are used generally to estimate the speed of induction
motors with high accuracy [1-8]. Most of the estimation =β λ − βω λ −k i +k V (2)
techniques work on the magnetic flux waves, such as open-loop
estimators and close-loop observers used for this purpose.
____________________________________________________________________________________________________________
797
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
where,
σ = 1− k =
T = β=
k =k R + -
Fig. 2 Space Vectors and sectors
____________________________________________________________________________________________________________
798
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
ψ = −u sign(s ) (12)
=− λ −ω λ + i (22)
where,
=− λ +ω λ + i (23)
s = ı̂ −i (13)
____________________________________________________________________________________________________________
799
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
B. Reverse Operating
The experimental operation is carried out at half of the
synchronous speed of the induction motor. At 5.5th second,
speed is the same, but rotation is reversed. It is shown that open-
loop estimators has the best time for reaching to the reference
Fig. 4 Estimated speed from Open-Loop, Sliding-Mode point. For the reverse directed speed and braking mode, all
and MRAS base estimators at nominal speed reference algorithms have nearly almost the same decelerating time. But
MRAS has the most change in speed at positive speed side (e.g.
Fig. 7).
B. Ramp Reference Operation
On ramp operation, speed estimators performance observed
during 2 second are compared at synchronous speed. The ramp
value reaches the synchronous speed at 1 second. Minor
changes are observed on the starting characteristics. First one,
an open-loop initial point is changed. Also, sliding-mode has
more oscillation with respect to the previous operation. Also,
MRAS based estimation has no significant changes (e.g. Fig.
5).
VI. CONCLUSION
Open-loop estimation method is more stable at the large
part of the operations with the exception of starting point
problem. On the other hand, sliding-mode method has more
Fig. 5 Estimated speed from Open Loop, Sliding-Mode and MRAS
based estimators at nominal speed reference with one second ramp ripples than others. The method, MRAS, is minimized ripples
at steady-state and its initial condition is the best on all working
operations.
VII. APPENDIX
Rated Power: 11 kW
Rated Voltage: 380 V
Rated Frequency: 50 Hz
Rated Speed: 1470 rpm
Stator Resistance: 1.1 Ω
Rotor Resistance: 1.6 Ω
Magnetizing Inductance: 0.351 H
Stator Leakage Inductance: 0.3615 H
Fig. 6 Estimated speed from Open Loop, Sliding-Mode and Rotor Leakage Inductance: 0.3615 H
MRAS based estimators at different speed reference changing in time Inertia: 0.077 kg m2
____________________________________________________________________________________________________________
800
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
____________________________________________________________________________________________________________
801
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
*
Yuzuncu Yil University, Ercis Technical Vocational School of Higher Education, Electronic and Communication Technologies,
Ercis 65080, VAN, Turkey
+
Yuzuncu Yil University, Electrical-Electronic Engineering, 65080 VAN, Turkey
Abstract— After the memristor, which passive electrical fractional order, the behavior of this element has begun to be
component relating electrical charge and magnetic flux linkage the subject for new researches [20-22].
and it was known as lost element since 1971, has been found by
Modeling of fractional order systems known as the
researchers in Hp Lab, it has been the focus of attention of
researchers. For analysis of the memristor, analog circuit model systems having system degree's any real number except whole
of memristors is generally used. In this paper, using fractional numbers [20-23] has practices in many fields such as control
order a memristor model, fractional order adaptive PI controller systems [24-25], effects on chaos behavior [26-27],
has been designed. For performance analysis of the controller, it communication systems [28]. The studies exercised fractional
has applied to control of nonlinear system, and compared order systems on circuit models have proved that system
classical PI controller. degree could be changed according to circuit topology made
by using passive circuit elements [29]. One of the most
1. Introduction common practicing field of fractional order systems is control
system. Having wider stability domain at control systems and
Memristors are known as the materials representing the performing in a good way at controling nonlinear systems
link between the charge and flux suggested as the missing have increased the interest in this type of controllers [30]. The
element by Chua. It has been known as the missing element behavior of a controller designed with fractional order
since 1971, however it has been found this material's entity at memristors which is a fractional order nonlinear element still
the result of a research in Hp laboratory in 2008 [1-3]. remains uncertain.
Because of memristor's feature to hold the last resistance
In this study, PI type controller has been designed by
value as a passive circuit element (likely to be used as a
passive memory element) has increased interest in these using fractional order memristor from model of memristor in
materials. Studies on the memristors have risen more and literature. The performance of a nonlinear system on control
more after 2008. These studies have generally spread into has been measured with this controller designed by using
many different scientific fields like; memristor emulator analog circuit designation method. To evaluate controller
circuits [4-8], production methods [8-14], applications on performance, it has been compared with performance of PI
communication systems [15], nonlinear circuit design [16-18]. controller occurred with classic memristors. The effect of
In some of these studies, it has been observed that the usage of controller parameters on control system has been investigated.
memristor affects system performance positively. Besides the The study was organized like above. The structure and
model of memristor developed in Hp laboratory, the idea of mathematical model of classic memristors and fractional order
using the emulator circuit in practice is still widespread. One memristors have been investigated in II. section. In III.
of these studies is the model put through by Muthuswamy section, the applications of classic memristors and fractional
[19]. It has been observed that Chua circuit, composed by order memristors on control systems has been investigated. In
using this model, demonstrates chaotic behavior. IV. section, simulation results have been given. In conclusion,
Memristor are defined as the ratio of the change in raids the results have been investigated.
over the last load changes. It has been known that the change 2. Memristors
of flux in terms of current is inductance, the change of current 2.1. Classic Memristors
in terms of voltage is resistance, and the change of voltage in
terms of load is capacitor. According to these definitions, as Memristors have been found at HP laboratory in
the element representing to the change rate between flux and 2008 after Chua suggested this as an idea. According to this
load as the fourth relation had not been found, it was called as model, the relation between voltage and current applied to
missing element. After finding this element, studies has circuit has been given at Equation 1. [1-2].
determined the relation between load and flux as a function of
____________________________________________________________________________________________________________
802
_____________________________________________________________________________________________________________
International Conference on Advanced Technology & Sciences (ICAT’16) September 01-03, 2016 Konya-Turkey
v(t ) .R .R 1 d
mt
f ( )
M (q ) Ron v 2 on q(t ) Roff (1 v 2 on q(t )) Dt f (t ) (t )
1 ( m )
d
(m ) dt
a
i(t ) D D (4)
a
1
A N
1
s C z ,a
C 3