Anda di halaman 1dari 14

Ir thermography

1. 1. ELECRICAL SYSTEM FAULT DIAGNOSIS USING INFRARED THERMOGRAPHY A


PHASE-I PROJECT REPORT Submitted by PRIYADHARSINI. S (090107126063) SHIYAM
DHARSAN.R.P (090107126077) SUJITH KUMAR.B (090107126084) YOGAPRIYA.M
(090107126097) In partial fulfilment for the award of the degree of BACHELOR OF
ENGINEERING in ELECTRONICS AND COMMUNICATION ENGINEERING SRI SHAKTHI
INSTITUTE OF ENGINEERING AND TECHNOLOGY, COIMBATORE-641062 ANNA
UNIVERSITY : CHENNAI-600 025
2. 2. TABLE OF CONTENTS CHAPTER NO. TITLE PAGE NO. LIST OF FIGURES
ACKNOWLEDGMENT i ii ABSTRACT iii 1. 2. INTRODUCTION INFRARED
THERMOGRAPHY 1 2.1 Introduction 3 2.2 Thermography 4 2.3 Advantages of
thermography 5 2.4 Limitations and disadvantage of thermography 6 2.5 2.5 Spectral
distribution of radiation intensity of black Body using Planck’s equation 6 2.6 2.6 Infrared
Thermography in Condition Monitoring of Electrical Equipment 7 2. 2.7 Thermographic
camera 8 2.8 Most common errors in thermographic measurement 11 2.9 Finding Hot Spots
11 2.10 Challenges Faced During Infrared Thermography Of Electrical Equipment 11 3.
LITERATURE SURVEY 17 4. PROJECT DESCRIPTION 4.1 Block diagram of the system
used for inspection of equipment 20 4.2 Input Image 4.3 SIFT Algorithm 21 5. SIMULATION
RESULT AND CONCLUSION
3. 3. 5.1Simulation result 24 REFERENCES 26
4. 4. NAME OF THE FIGURE 1.Thermogram of loose connection 2.Thermal image of electrical
contact 3.Electromagnetic spectrum 4.FLIR i7 camera 5.Block diagram of the system used
for inspection of equipment Output PAGE NO. 5 9 10 11 12 24
5. 5. ABSTRACT An automatic diagnosis system is proposed in this project for a more and
more important issue, preventive maintenance. Every year, various workplace accidents
happen due to undesirable maintenance. No matter how stringent the rules governing the
maintenance of electrical equipment may be, it is always a challenge for the power industry
due to the large number of electrical equipment and the shortage of manpower. In this
project, an automatic diagnosis system for testing electrical equipment for defectsis
proposed. Based on non-destructive inspection, infrared thermography is used to automate
the diagnosis process. Thermal image processing based on statistical methods and
morphological image processing technique are used to identify hotspots, the reference
temperature and cause for defects. The problematic area is captured using thermal camera.
The repeated objects are detected by using feature detection algorithm like SIFT. Once the
repeated objects are identified, their real time temperature are measured using thermal
software. By comparing the real temperature of repeated objects, the fault area is localised
and the reason for fault is identified using Artificial Neural Network. Here we used Levenberg
– Marquardt Algorithm which comes under Back Propagation technique. The thermal
diagnosis system to be implemented by this project can be used at the various power
facilities to improve inspection efficiency as the reason for the defect can be identified.
6. 6. CHAPTER 1 INTRODUCTION
7. 7. CHAPTER-1 INTRODUCTION In the power system, there are many kinds of electrical
equipments such as circuit- breaker, transformer, lightningarrester, capacitor, current and
potential transformer, bushing, and insulator and so on. These equipment play animportant
role in power-supply system. Unfortunately, some nasty faults often happen to the electrical
equipment because of a variety of reasons, and they are seriously imperilling the safe
operation of electric power production.Therefore, a great cost is paid for preventive test to
remove hidden dangers in power system. In recently years, withthe fast development of
infrared technology, infrared thermography technique is in great advantages of diagnosing
electrical equipment with faults. By means of getting the thermography of electrical
equipment withouttouching, one can judge whether the equipments is in good or not by
analyse the thermal distribution of these equipment. It is proved by practice that infrared
thermography diagnosis has become the necessary andeffective supplementary measure of
the preventive test to the electrical equipment. It can be divided into two kinds of faults
attribute according to the location of the faults of electricalequipment, the external and
internal faults. Asfar as infrared thermography diagnosis, the external faults showmainlythe
overheatof connectors, and they are easy to be discerned. However, the internal faults are
difficult to bepenetrated because internal faults are much more complex. To find out the
internal faults, one must know the law of the internal faults attribute to the relation of their
infrared thermography characteristics. The internal faults of the electrical equipment can be
divided into loose connection or contact of internal conductors and inferiority in insulation and
other faults. In order to know the law of infrared thermography diagnosis to electrical
equipment with internal faults, on the basis of the electrical equipment simulation test and the
experience by infrared thermography diagnosis to a great many substations, we sum up a
few technique problems of infrared thermography diagnosis to the electrical equipment, with
internal faults and show some typical examples thermographies of electrical equipment with
internal faults.
8. 8. By automating the inspection process, the problem of time wastage and manual
involvement in the inspection process can also be done away with. Thus, this project
proposes an automatic diagnosis system fo1r testing electrical equipment for defects. For
this purpose Scale invariant feature transform (SIFT) algorithm is used. The defective parts
are detected by determining which of these areas on the infrared images are with higher
temperatures than the normal prescribed levels. Inspection results are classified into different
categories depending on the levels of temperatures detected that tell the power companies
the seriousness of each situation in each of these areas. Thermal inspection of the electrical
equipment can reveal various types of problems in electrical installations. In recent years, the
use of thermal imaging or infrared thermography has become an important tool in preventive
and predictive maintenance. It is a useful method for inspecting the condition of electrical
equipment. Thermal imaging inspection is well known as a non-contact measurement
technique where the inspection can be done without interrupting or shutting down the
operation of a system. It is a safe, reliable and very cost-effective approach for a
maintenance programme.
9. 9. CHAPTER 2 INFRARED THERMOGRARHY
10. 10. CHAPTER 2 INFRARED THERMOGRAPHY 2.1.Introduction In 1800, astronomer Sir
William Herschel discovered infrared, and thus began the exploration of the science of
thermography. Sir William designed and created his own telescopes - becoming very familiar
with lenses, mirrors and light refraction. His thermography research began with the
knowledge that sunlight was made up of all the colours of the spectrum, and that it was also
a source of heat, so he set out to determine which colours were responsible for heating
objects. The first thermography experiment utilized a prism, paperboard, and thermometers
with blackened bulbs where the temperatures of the different colours were measured. As
sunlight passed through the prism, Sir William observed an increase in temperature as he
moved the thermometer from violet to red in the rainbow created by the light. Herschel noted
that the hottest temperature was actually beyond red light, and that the radiation causing this
heating was invisible. He called this invisible radiation “calorific rays." Today, we refer to the
light/energy as infrared, and the measuring of the heat emitted as thermography. Infrared
Thermography is simply a picture of heat. All the bodies emit energy from their surface as
electromagnetic waves, which magnitude is directly related to their temperature. The hotter
the object is, the more energy it tends to radiate. Such temperature settles the wavelength of
the emitted energy, the colder the object is, the higher its wavelength will be, whereas the
hotter it is, the lower its wavelength will be. This last case, is the one of the infrared energy,
non visible to the human eye, but visible by means of an infrared camera.
11. 11. The radiation measured by the infrared camera depends not only on the temperature of
the object but also on its emissivity. The radiation coming from the surrounding area and
reflected on the object also influences the measuring. Therefore, to measure the temperature
accurately, besides the effects of different sources of radiation that interact with the object,
other variables such as emissivity, distance between the camera and the object scanned,
environment temperature and humidity, must also be considered. In addition, due to the
characteristics of the infrared radiation, to detect any overheating by IR scans, the heat
generated must be “directly” in sight of the thermographer. 2.2 Thermography Thermography
is one of the most powerful tools available for electrical maintenance. With professional
training and some experience a thermographer can quickly locate high resistance
connections,load imbalance and overloads while the system is in operation. This can all be
accomplished without direct contact to the energized system. Electrical inspections have
typically produced remarkable returns, with documented returns of 30 to 1 on the part of a
major industrial insurer. Prevention of catastrophic failure and unscheduled outages often
results in cost savings far in excess of the cost of the test equipment and program. Today’s
economic climate, however, demands even greater assurances for reliability from
maintenance thermographers from past. Experience can reveal the inspection program’s
successes and limitation. Some limitations to thermographic tests of electrical equipment are
quite obvious. Some problems are inherent to laws of physics and must be lived with or
worked around. Others are related to environmental or operating conditions. The latest
infrared test equipment is no longer a limiting factor; it will do more than usually needed. But
inadequate data collection procedures and a poor understanding of how to use the
information gathered are very much limiting factors.
12. 12. IR film is sensitive to infrared (IR) radiation in the 250°C to 500°C range, while the range
of thermography is approximately -50°C to over 2,000°C. So, for an IR film to show
something, it must be over 250°C or be reflecting infrared radiation from something that is at
least that hot. Night vision infrared devices image in the near-infrared, just beyond the visual
spectrum, and can see emitted or reflected near-infrared in complete visual darkness.
Starlight-type night vision devices generally only magnify ambient light. Figure:1
Thermogram of loose connection Infrared thermography is generally classified in two types,
passive and active thermography, In passive thermography, the temperature gradients are
present in the materials and structures under tests naturally. One of the applications of
passive thermography is for preventive and predictive maintenance. In active infrared
thermography, the sample is heated by an external controlled heat sourceand its surface
temperature is monitored as a function of time through changes of emittedinfrared radiation.
2.3 Advantages of thermography  It shows a visual picture so temperatures over a large
area can be compared  It is capable of catching moving targets in real time  It is able to
find deteriorating, i.e., higher temperature components prior to their failure
13. 13.  It can be used to measure or observe in areas inaccessible or hazardous for other
methods  It is a non-destructive test method  It can be used to find defects in shafts, pipes,
and other metal or plastic part.  It can be used to detect objects in dark areas 2.4
Limitations and disadvantages of thermography  Quality cameras often have a high price
range (often US$6,000 or more)  Images can be difficult to interpret accurately when based
upon certain objects, specifically objects with erratic temperatures, although this problem is
reduced in active thermal imaging  Accurate temperature measurements are hindered by
differing emissivities and reflections from other surfaces  Most cameras have ±2% accuracy
or worse in measurement of temperature and are not as accurate as contact methods  Only
able to directly detect surface temperatures 2.5 Spectral distribution of the radiation intensity
from a black body using Planck equation Planck derived the law as in equation (1), which
describes the spectral distribution of theradiation intensity from a black body where the
emissivity of the surface, ε is equal to 1 (Holst, 2000). 𝜀𝜆𝑏 = 𝐶1 𝜆5(𝑒 𝐶2 𝜆𝑇 −1) 𝑊 𝑚2−𝜇𝑚 (1)
Where 𝜀𝜆𝑏 is the black body monochromatic radiation intensity, C1 (3.7411 x108 W-
μm4/m2) and C2 (1.4388 x104 μm-K) are the first and second radiation constants
respectively; λ is the wavelength of the radiation being considered and T is the absolute
14. 14. temperature of the blackbody. By integrating Planck’s law over the entire spectrum (λ = 0
to ∞), the total hemispherical radiation intensity is obtained. 𝜀 𝑏 = 𝜎𝑇4 (2) where ζ is the
Stefan–Boltzmann constant (5.67051 x 10-8 W/m2K). It has to be pointed out that equation
(2) describes the radiation emitted from a black body which is the maximumvalue radiated by
a body at a given temperature. Real objects almost never comply with thislaw although they
may approach the behaviour of a black body in certain spectral intervals.A real object
generally emits only a part ελ of the radiation emitted by a black body at thesame
temperature and at the same wavelength. By introducing the quantity, 𝜺 = 𝜺 𝝀 𝜺 𝝀𝒃 (3) which
is called the spectral emissivity coefficient, equation (2) can be rewritten for real bodies by
simply multiplying its second term by ελ. When averaged over all wavelengths, the total
power density for a non-black body object is [1] 𝑒𝑚𝑖𝑠𝑠𝑖𝑣𝑖𝑡𝑦 = 𝜀𝜎𝑇4 (4) 2.6 Infrared
Thermography in Condition Monitoring of Electrical Equipment All electrical devices are
usually rated for power, which indicates the amount of energy that the devices can conduct
without being damaged. If the device is operated at a power above its specifications, the
excess power can reduce the device's life cycle and efficiency. Basically, faults in electrical
power system can be classified into few categories, i.e., poor connection, short circuit,
overloading, load imbalance and improper component installation. In most cases, the major
cause of overheating in utility components is the change in resistance due to loose
connection. The loose connection causes electricity to use smaller area of the defective
connection than required for proper current flow and therefore, increases the resistance and
temperature of the connection. Any problem, which accompanies a change in resistance of
the equipment, causes it to consume morepower than the intended load.
15. 15. According to a thermographic survey, it was found that 48% of the problems were found
in conductor connection accessories and bolted connections. This is mainly resulted from the
loose connection, corrosion, rust and non-adequate use of inhibitory grease. On the other
hand, 45% of the thermal anomalies appear in disconnectors contacts. Mostof the anomalies
are due to deformations, deficient pressure of contact, incorrect alignment of arms and
dirtiness. Only 7% of the problems were found in electrical equipment. Another major cause
of overheating in electrical components within the structure is overloading. Through IRT
camera, the sign of overloading can be seen clearly even if the cable was located deep into
the concrete where the red region which has high temperature value covered all parts of the
components or cables. By utilizing IRT technology, the thermal image will clearly indicates
the problematic area. The suspected area can be easily identified and interpreted.
Nevertheless, in some cases, the interpretation of thermographic image cannot be done
directly except for an experienced and qualified thermographers. 2.7 Thermographic
cameras A thermographic camera or infrared camera is a device that forms an image using
infrared radiation, similar to a common camera that forms an image using visiblelight. Instead
of the 450–750 nanometer range of the visible light camera, infrared cameras operate in
wavelengths as long as 14,000 nm (14 µm). In 1929, Hungarian physicist Kalman Tihanyi
invented the first infrared-sensitive (night vision) electronic television camera for anti-aircraft
defense in Britain. The first conventional IR camera, the "Evaporograph", was declassified
around 1956. Thermal imaging cameras convert the energy in the infrared wavelength into a
visible light display. All objects above absolute zero emit thermal infrared energy, so thermal
cameras can passively see all objects, regardless of ambient light. However, most thermal
cameras only see objects warmer than -50°C.
16. 16. Figure 2: Thermal image of electrical contact Theory of operation All objects (even cold
ones) radiate heat in the form of infrared energy. As an object heats up, it radiates more
energy, and the wavelength gets shorter. Infrared radiation, visible light and ultraviolet light
are all forms of energy in the electromagnetic spectrum. The only difference is their
wavelength. Figure 3:Electromagnetic spectrum The human eye can only see a small range
of colours in the electromagnetic spectrum. These light waves range in length from 0.4 to 0.7
microns. If an object gets hot enough, the energy will reach the visible range and the object
will be “glowing” red, like the burner on an electric stove. Fortunately, infrared imaging
systems can detect infrared energy long before it reaches the visible stage. The camera-like
device then converts these invisible light waves into a graphic image that is displayed on a
monitor. Modern infrared cameras also provide actual temperature readings, and store the
data, so that the information can be later used to
17. 17. produce a report. However, gathering the information is the easy part. The real work and
value is what the thermographer can do with the data that is gathered. As in any form of non-
destructive testing, the interpretation of the finding takes both education and experience.
FLIR camera The FLIR i7 thermal imaging camera is an extraordinary tool for detecting
structural problems (or potential problems) with heavy equipment, motors, pumps, buildings,
circuit breakers, cooling systems, and much more . When equipment and/or structural
integrity begins to malfunction, all heck can break loose, often resulting in a loss of
productivity or worse, bodily harm. Professionals and homeowners alike can take comfort in
knowing that a powerful diagnostic tool like the FLIR i7 thermal imaging camera can save
them from this fate by detecting miniscule temperature variations.These variations can
quickly and easily identify any pipe or duct leaks, cracked or loose seals, electrical failures,
mechanical breakdowns, moisture issues, insulation scarcity, and so much more. With the
help of FLIR i7 infrared camera , the problem in contractor repairs, heating and cooling
systems, pumps, circuit breakers, transformers, motors, building structures can be identified.
Figure 4: FLIRi7 camera The FLIR i7 thermal imaging camera is at the high end of entry level
IR cameras. With a resolution of 120 x 120 pixels, this particular camera can be used to
identify a variety of structural and mechanical problems before they escalate into full-blown
18. 18. crises. The compact size of the FLIR i7 camera, along with its light 12-ounce body and
ultra powerful infrared capabilities, make it among the most sought-after thermal imaging
units in today's marketplace. The unit boasts a visually appealing 2.8" LCD color display, and
can store up to 5000 high-quality JPEG images. Additionally, the i7 is incredibly accurate. It
has an accuracy of plus or minus 2 percent, along with a temperature measurement
sensitivity of 0.1 degrees C. This is essential for monitoring the condition of a thermally
sensitive target. 2.8 The most common errors in thermographic measurements To accurately
measure the temperature distribution on the body of electrical machines, it is necessary to
separate the influence of wished sources from disturbances that have to be compensated. In
order to do this automatically, the camera must be informed of the following parameters: 
the atmospheric temperature,  the distance between the object and the camera,  the
relative humidity,  the emissivity of the object. 2.9 Finding the "Hot Spots" Infrared
thermography is a non-destructive technique for detecting “hot spots,” which are temperature
differentials that may indicate problems such as loose electrical connections or excessive
friction in machinery and mechanical systems. Other uses for this technique include finding
defective or leaky steam traps and clogged steamed lines, as well as many other conditions
which can lead to failure or energy loss. Infrared thermography utilizes a camera-like device
which views a large area at a time, senses infrared emissions and converts the emissions
into a visual display. Large quantities of equipment are scanned while in operation, so
production need not be
19. 19. interrupted. Trouble spots can be pinpointed quickly, saving labour and cost and
targeting plant maintenance resources where they are needed. 2.10 Challenges Faced
During Infrared Thermography of Electrical Equipment When performing an infrared
inspection of an electrical system it is important to realize that all of the radiation leaving a
surface is not due solely to the temperature of the surface. Unless knowledge, understanding
and caution are applied during the analysis portion of the inspection, documentation and
interpretation may result in the false conclusion that a fault does or does not exist.Wesum up
a few technique problems of infrared thermographydiagnosis to the electrical equipment, with
internalfaults andshow some typical examples thermographies of electrical equipment with
internal faults. They are loose connection or contact of Internal conductors,Inferiority in
Internal Insulation of Electrical Equipment, etc., Loose connection or contact of internal
conductors It will lead to the resistance increasing and overheat in the action of the current
when some internal conductors arein loose connection or contact. The heat power conforms
to the low of P=I’R (where P is the heat power, I is thecurrent passes through the conductor,
and R is the contact resistance of the conductors). Although the heat source ofthis kind of
faults is in the internal of equipment, the external of equipment will show overheat
somewhere because of the action of the thermal transmission. Therefore, compared with the
thermography of normal equipment, the thermography of this kind of faults will show the
characteristic of local overheat somewhere of the equipment. This kind of faults often take
place in the contacts of circuit-breaker and primary internal connection of current transformer
and bushing and cable splice in internal connection of conductors. this kind of faults can be
found out by the Characteristic of their thermographies a) Loose contact of internal contacts
of short coil circuit The loose contact of internal contacts of short coilcircuit-breaker shows
mainly the loose contact of the uppercontact or intermediate contact. Either of them will lead
20. 20. to overheat of the circuit-breaker, but their pattern of theoverheat are different. When the
upper contact is in loosecontact, the infrared thermography shows the temperature of its
header is the highest, then the basal stump flange,and the intermediate porcelain bushing is
the lowest. b) Loose connection of primary internal connection of current transformer When
the primary internal connection of the currenttransformer is in loose connection, it will lead to
currenttransformeroverheat under the action of the current. Thethermography of the current
transformer shows thecharacteristic of the header of the current transformer isoverheat while
the porcelain body is almost normally. c) Loose connection of the internal outlet terminal of
high- voltage bushing header When the internal outlet terminal of bushing header isin loose
connection, it will lead to the bushing headeroverheat under the action of the current. The
thermography of the bushing shows the characteristic of the bushingheader is the heating
center while the body of the porcelainbushing is basically in normal. d) Loose connection of
internal conductors of cable splice When a phase of cable splice is in loose connection of
internal conductor, the cable splice will be overheat. Thethermography of the cable splice will
show the characteristic of local overheat, and the heat centre is in the forked of the phase.
Inferiority in Internal Insulation of Electrical Equipment Under the action of high-voltage,
some electrical equipment will overheat because of their inferiority of internal insulation; the
heat power conforms to the law of 𝑃 = 𝑈2 . 𝜔. 𝐶. 𝑡𝑔. 𝛿 (Where P is the heat power, U is the
voltage, 𝜔is the angle frequency, C is the capacitance of the equipment, and the 𝛿is the
angle of dielectric loss), the electric test of this kind of fault generally shows the dielectric
loss increasing. These kinds of faults often happento the
21. 21. equipment such as potential and current transformer. Coupling condenser, cable splice
deliquescence and deviation of post insulator. a) Inferiority in internal insulation of potential
transformer When a phase of the potential transformer is inferior in internal insulation, its
thermography will show thecharacteristic of whole body overheat in comparison withthe other
phases moreover, there is not a remarkableoverheat center on its body but the temperature
of its headeris a little higher than that of the porcelain body. b) Inferiority in internal insulation
of current transformer heating and there When a phase of the current transformer is inferior
in internal insulation, its thermography will show thecharacteristic of whole body overheat in
comparison with the other phases, there is not a remarkable overheat Centreon its body but
the temperature of the header is higher thanthat of its porcelain body. c) Inferiority in internal
insulation of coupling condenser When a phase of the coupling condenser is inferior
ininternal insulation, its thermography will show thecharacteristic of whole body overheat in
comparison with the other phases moreover, its whole body is almost homogeneous heating
and there is not a overheat centre. d) Insulation deliquescence of the cable splice When the
cable splice is deliquesced because of local damage or poor sealed, the thermography of the
cable splice will show the characteristics of whole body overheat or local overheat. When
whole insulation of the cable splice is deliquesced, the thermography of the cable splice
shows characteristics of whole body overheat. e) Deviation of post insulator Sometimes the
post insulator may deviate from insulation and give out abnormal heat because of fail in
manufacturing and technology or porcelain aging in the long time. The thermography of the
post insulator shows the characteristic of whole body overheat.
22. 22. Other faults Some faults of electrical equipment are rather particular. For example, some
electrical equipment with oilfor insulator dielectric may lack of oil or low oil level, sometimes
the oil level is pseudo or false oil level. Thiskind of fault is difficult to be found by common
electrical test, but it is very effective for infrared thermography tofind out. As electrical
equipment are with normal heat because of power consumption and the oil of electrical
equipment is the carrier of heat. Therefore, when the oil level is lower, the thermography of
the equipment willshow characteristic of the mark of oil level separatrix with temperature
lower above and higher below. This kind of fault often happens to the equipment such as
coupling condenser and potential transformer and oil bushing of transformer. In addition,
some equipment give out unbalanced or abnormal heat distribution because of their internal
components being damped, this kind of faults often happen to the equipment such as all
sorts of arrester. a) Lack of oil in coupling condenser The oil level of coupling condenser with
lack of oil is often lower for several skirts, this kind of fault can be seen from its thermography
when the high sensitivity of the thermo vision is selected. Moreover, there is a clear
temperature gradient in oil level. b) Lack of oil in potential transformer When a potential
transformer is lack of oil or low oillevel, its thermography will show the characteristic
ofseparatrix with dimer above and brighter below, theseparatrix indicates the true oil level.
This kind of fault canbe seen from its thermography when the high sensitivity ofthe thermo
vision is selected. c) Lack of oil in bushing of transformer When a bushing of transformer is
lack of oil or low oillevel, its thermography will show the characteristic ofseparatrix with dimer
above and brighter below, theseparatrix indicates the true oil level. This kind of faultscan be
seen from its thermography when the highsensitivity of the thermo vision is selected.
Moreover, thereis a clear temperature gradient in oil level.
23. 23. d)Dampness in internal components of arrester When the internal components of
arrester is dampedbecause of poor sealed or porcelain damage, the resistance of the
internal components will be abnormal, thecharacteristic of its thermography is subject to the
structure of the arrester. Generally speaking, when wholecomponents of the arrester are
damped, its thermographywill show the characteristic of whole body overheat bycomparison
with other phases. When local components ofthe arrester are damped, its thermography will
show thecharacteristic of local overheat and local dimer (lower temperature) by comparison
with other phases, it shouldtake notice that in the latter case, the location of the dampis just
in the dimer of the thermography. CHAPTER 3 LITERATURE REVIEW
24. 24. Soib Taib , Mohd Shawal Jadin,Shahid Kabir-“Thermal Imaging For Enhancing
Inspection Reliability : Detection and Characterization” The role of non-destructive testing
(NDT) is to ensure integrity, and in turn, reliability of equipment or structure. Besides, NDT
can also monitor in-service degradation and to avoid premature failure of the
equipment/structures and prevent accidents as well as savehuman life. Up to now, NDT has
been used in various fields of applications such as the inspection of electrical power plant,
substation, storage tanks, bridges, aircraft, pressurevessel, rail, pipeline and so on. Human
eyes can only see light in the visible spectrum, ranging from about 400 nm to a littleover 700
nm. The electromagnetic spectrum is a band of all electromagnetic waves
arrangedaccording to frequency and wavelength. As shown in Fig. 1, the wavelength
spectrum ofinfrared light ranges from about 1 mm down to 750 nm. All objects emit energy
proportionalto its surface temperature. However, the energy radiated can only be detected by
an infrareddetector that depends on the emissivity coefficient of the surface under
measurement. The Stefan-Boltzmann law describes the total maximumradiation that can be
released from a surface. Since thermal imaging systems only respondto a small portion of
the spectrum, it is necessary to introduce Planck’s blackbody law. Infrared thermography is
generally classified in two types, passive and active thermography. In active thermography,
the relevant thermalcontrasts are
25. 25. induced by an external stimulus. The passive method has beenwidely applied in diverse
areas such as production, predictive maintenance, medicine,detection of forest fire, thermal
efficiency survey of buildings, road traffic monitoring,agriculture and biology, detection of gas
and in NDT. Early prevention of electrical power failures is veryimportant since power
interruptions can have serious impacts on the social and economic activities of a country. To
ensure acontinuous power supply, the reliability of electrical powerequipment must be
checked regularly. Abnormalities in the equipment will occur when their internal temperatures
exceed their limits. Consequently, the overheating of electrical equipment can lead to
subsequent failure of the equipment and can potentially result in unplanned outages, injury
and fire hazard. In addition, the efficiency of an electrical grid reduces prior to failure; thus
energy is spent generating heat, causing unnecessary loss. The common problems
regarding thermal anomalies in electrical installations are loose or poor connections,
unbalanced loads, short circuits, overloading andcracks or defects in the equipment body.
Basically, there are two ways to analyse the thermal characteristics in electrical equipment.
The first is thequantitative evaluation, which is to measure the exact real temperature value
of an object. However, these measurements are relatively difficult to obtain; in order to
determine the real and accurate temperature value, the true emissivity value mustbe
identified by considering the effects of ambient conditions and atmospheric attenuation. The
second is the qualitative measurement, which considers the relative temperature valueof a
particular hotspot with respect to other equipment in a similar environment. This method,
which employs the ΔT criteria, is widely used to evaluate electrical equipment. The
qualitative measurement is the most suitable method for evaluating the thermal condition of
electrical equipment; therefore, all similar and identical structures within the thermal image
should be grouped together. For detecting the regular structure of electrical equipment, the
tasks can be broken down into three separate steps: a) Detecting the interesting feature
points in the thermalimage with specific descriptors.
26. 26. b) Comparing all the features and matching them withthe most similar and closest
distance points. c) Segmenting the regions of interest. Ying-Chieh Chou , Leehter Yao-
“Automatic diagnostic system of electrical equipment using infrared thermography." The
repair and maintenance of equipment at important facilities has been a primary area of
concern. Of these facilities, the repair and maintenance of equipment at power transmission
facilities is listed as a task with the highest priority because our abilities to continue to enjoy
the quality of life we are enjoying now depend solely on the continuous operation of this
equipment in the future. Power installations are usually located in every corner of small
villages and big cities. This is because electricity has to be provided to wherever the
consumers are conducting their indoor or outdoor activities, leading to tens of thousands of
such facilities. The repair and maintenance of a facility can be classified under three different
categories: when equipment malfunction, time–based, and condition-based maintenance.
The most popular one is condition-based maintenance, also known as preventive
maintenance. Infrared thermography technique is widely used in preventive maintenance for
the advantage of carrying out quick, accurate, and wide area inspections by telemetry. With
this technique, defective parts can be detected through simple observation of infrared
images and there is no need to shut down the operation of a facility to look inside the
equipment for inspection. Infrared thermography technology is a technology that uses
infrared sensors and optical lenses in a constructed electrical circuitry to capture images of
thermal objects based on temperature variations. Infrared thermal camera stores the infrared
pictures of thermal objects as thermal images that the human can see in order to understand
the inside conditions of the objects. With the images, inspectors can analyse the temperature
variations of thermal objects to lookfor defective parts. Infrared thermography technology is a
non-destructive inspection technique. The inspection can be conducted efficiently by keeping
a distance from the inspected equipment. There is no need to halt equipment operation while
an inspection
27. 27. is going on. Since the collection of information for inspection is by telemetry, hazardous
operations can be avoided . For these reasons, Infrared thermography is widely used for
many applications involving preventive maintenance. J.Rantala.D.Wu,A.Salerno,G.Busse –
“Thermal Imaging for Qualitative Based Measurements of Thermal Anomalies in Electrical
Components” Temperature change in materials can be induced with mechanical vibration
where thethermo-elastic effect and hysteresis effect are involved. The thermo-elastic effect is
the dominating mechanism in metals. In polymers, however, the hysteresis loss
dominatesalready at low amplitudes, as is obvious from their high acoustic or mechanical
damping. The enclosed area corresponds to the dissipated energy which is converted into
heat. Delamination can occur in the manufacturing process of wood-based panels with
coatingmaterials such as veneer. These defects have to be detected in an early stage. A
sample shown in was investigated Between the 0.5 mm veneer layer and substrate there are
several holes and two embedded sheets: a piece of teflon film in the middle and an
aluminium film at the right bottom corner. All these defects can be clearly detected in the
phase and amplitude image. The quality of polymer materials may suffer from boundary
effects. Therefore one is interested to detect boundaries or their changes. One example is
welding of polymers. Areas of disband are inherent sources of weakness which result in
failure under load. In the middle of the sample there is an air gapbetween two bonded parts.
Defects were shown up in the phase and amplitude image whenultrasonic waves are
coupled into the sample. The contrast of phase and amplitude image ismuch better than in a
thermographic image. Intact boundaries are also essential for coated materials where the
outer layers are supposed to have some special functions, e.g. improved resistance to wear,
corrosion, or heat.In these cases one is interested to monitor local variations of thickness or
to detect areas of disband where the loss of adhesion may later on result in failure. A metal
sample with ceramic coating on a metal substrate was inspected by using lock-in
28. 28. vibrothermography. Adelamination area on the upper edge of the sample is detected. In
the phase image the beginning of the delamination can be seen clearly. CHAPTER 4
PROJECT DESCRIPTION
29. 29. CHAPTER 4 PROJECT DESCRIPTION 4.1 Block Diagram of the System Used for
Inspection of Equipment Process Under Diagnostic Data Acquisition Using IR Camera
Finding Repeated Objects Fault Classification Using ANN/SVM Data Bank Type Of Fault
Normal Abnormal
30. 30. 4.2 Input Image The input image is the infrared image that is captured on the Thermal
camera during inspection. When the thermal images are captured inappropriately, the image
colour can appear too bright or too dark and monotonous. These thermal images with flaws
tend to increase the amount of mistakes inspectors made by making wrongful judgements
during an inspection In order to avoid this, it is suggested that thermal images that do not
show vivid distinction between the main objects and their backgrounds must not be included
in an inspector report. Thermal images, or thermograms, are actually visual displays of the
amount of infrared energy transmitted and reflected by an object. As there are many sources
of infrared, it is difficult to get the accurate temperature of an object using this method. The
inputs are detected by the feature extraction technique like SIFT, SURF, MSER etc., out of
which sift is most commonly used. The SURF algorithm is derived from the SIFT algorithm.
The SURF algorithm is more efficient than SIFT because the number of iterations are
reduced in SURF. The sift algorithm is used to find the repeated objects in a system. The
input is obtained from the images of a thermal camera which gives of a thermogram of the
image being captured. Thermographs are converted from colour domain
31. 31. to gray scale domain in order to reduce the computational complexity. The defect region
in thermographs is highlighted by Image enhancement which is done with averaging
Gaussian filters. Image segmentation by edge detection is performed on those thermographs
with suitable filters. After which post processing (dilation, region growing and erosion) is
done to remove the undesirable region. Dilation is performed on the edge detected image to
bridge the gap between each pixel; region growing is performed on the dilated image to
differentiate the region of interest pixels from the background pixels; erosion is performed on
the region filled image to isolate the defect or the region of interest. The input image is given
to the neurons in the artificial neural network. The input layers pass on the node input to the
hidden layer where the processing is done. After the processing, the output is obtained in the
output layer with errors eliminated to a maximum extent. 4.3 SIFT Algorithm The detection
step of SIFT algorithm is based on difference of mean (without interpolation). This algorithm
is used to find the repeated objects and the stable points by diluting the image. Following are
the major stages of computation used to generate the set of image features: 1. Scale-space
extrema detection: The first stage of computation searches over all scales and image
locations. It is implemented efficiently by using a difference-of-Gaussian function to identify
potential interest points that are invariant to scale and orientation. 2. Key point localization:
At each candidate location, a detailed model is fit to determine location and scale. Key points
are selected based on measures of their stability. 3. Orientation assignment: One or more
orientations are assigned to each keypoint location based on local image gradient directions.
All future operations are performedon image data that has been transformed relative to the
assigned orientation, scale, andlocation for each feature, thereby providing invariance to
these transformations.
32. 32. 4. Keypoint descriptor: The local image gradients are measured at the selected scalein
the region around each keypoint. These are transformed into a representation thatallows for
significant levels of local shape distortion and change in illumination. This approach has
been named the Scale Invariant Feature Transform (SIFT), as it transforms image data into
scale-invariant coordinates relative to local features. 4.4 SURF Algorithm: SURF (Speeded
Up Robust Feature) is a robust local feature detector, first presented by Herbert Bay et al. in
2006, that can be used in computer vision tasks like object recognition or 3D reconstruction.
It is partly inspired by the SIFT descriptor. The standard version of SURF is several times
faster than SIFT and claimed by its authors to be more robust against different image
transformations than SIFT. SURF is based on sums of 2D Haar wavelet responses and
makes an efficient use of integral images. It uses an integer approximation to the Hessian
blob, which can be computed extremely quickly with an integral image (3 integer operations).
For features, it uses the sum of the Haar wavelet response around the point of interest.
Again, these can be computed with the aid of the integral image. The speed of SURF
algorithm is same as that of SIFT but the quality is less when compared to SIFT. Matching
Features: Repeating structures are present in almost every image of electrical installations.
This is due to the fact that electrical installations often are made in a very structured and
symmetrical way. Another factor is the presence of a three-phase system, where there will
be at least three similar copies of most components. Such repeating structures can be
considered as an indication of the existence of multiple similar objects in the image. If
working properly, similar objects ought to have the same temperature. If not, you can
conclude that the function of the installation is not optimal. Thus, by comparing temperature
properties of regions with the same geometric appearance, conclusions on the function of
the installation can be made.
33. 33. To be able to find a repeating pattern in an image one approach is to identify distinctive
features in the image, describe the features and compare them with each other to find similar
regions within the image. If a number of such matches are found in the image you can further
investigate the mutual properties between the matches to strengthen the hypothesis of a
repeating structure. An indication of a repeating structure from matching feature points is if
pairs of matching points have the same relative translation. To be able to use the detected
SIFT features to repeating structures in the image the descriptors of the points must be
compared to detect similar image regions. Although the SIFT descriptors are scale invariant
only features of the same scale are compared. If the detail in the image is detected as a
feature at a certain scale the corresponding detail in a similar object will be detected at the
same scale. This cannot be true always because of the discreteness of the concept of scale
nearly identical details in an image can appear as a feature of adjacent scales. The strategy
of considering pairs of features with a sufficiently small mutual distance as being similar is
not very reliable. It will generate many false matches if the threshold is chosen too high and
will miss many true matches if the threshold is chosen too low. 4.5 Artificial Neural Network
Artificial neural network (ANN) is an artificial model of real neural networks which refers to
the vast networks of interconnected neural cells that exist in the brains of animals and
humans. These neural cells act as summations, collecting the sum of the inputs to the neural
cells and sending a corresponding output onwards to the next neural cells. The connections
between neurons, or neural cells can vary in size. The difference in size of the connections
leads to varying levels of the signals being transmitted among cells, giving varying relevance
to the signals transmitted. The neurons of the brain are replaced by nodes, and are
connected by weights.
34. 34. A network can consist of any number of nodes, arranged in any kind of pattern. In our
project, a feed-forward back propagation network is used, which consist of three or more
layers of nodes. The first layer is the input layer, which has nodes depending upon the
number of input. The last layer is the output layer which consists of as many nodes as there
are outputs from the system. The hidden layer lies in between these two layers which consist
of an arbitrary number of nodes. The number of nodes in the hidden layers must be chosen
for individual network as it affects the behaviour of the network. The weight of each node is
updated during a training phase and this updating of the nodes can be done in many
different ways using training algorithms. Common for all training algorithm that uses
supervised learning is that the output of the system is compared to a specified target output,
and the error between these two is used to update the weights. 4.6.Applications of artificial
neural networks The utility of artificial neural network models lies in the fact that they can be
used to infer a function from observations and also to use it. Unsupervised neural networks
can also be used to learn representations of the input that capture the salient characteristics
of the input distribution and more recently, deep learning algorithms, which can implicitly
learn the distribution function of the observed data. Learning in neural networks is particularly
useful in applications where the complexity of the data or task makes the design of such
functions by hand impractical. The tasks to which artificial neural networks are applied tend
to fall within the following broad categories:  Function approximation, or regression analysis,
including time series predictionand modelling.  Classification, including pattern and
sequence recognition, novelty detection and sequential decision making.  Data processing,
including filtering, clustering, blind signal separation and compression. Application areas of
ANNs include system identification and control (vehicle control, process control), game-
playing and decision making (backgammon, chess, racing),
35. 35. pattern recognition (radar systems, face identification, object recognition), sequence
recognition (gesture, speech, handwritten text recognition), medical diagnosis, financial
applications, data mining (or knowledge discovery in databases, "KDD"), visualization and e-
mail spam filtering. 4.7. Neural Network Structure Neural networks are models of biological
neural structures. The starting point for most neural networks is a model neuron. This neuron
consists of multiple inputs and a single output. Each input is modified by a weight, which
multiplies with the input value. The neuron will combine these weighted inputs and, with
reference to a threshold value and activation function, use these to determine its output. This
behaviour follows closely our understanding of how real neurons work. Figure 6: A Model
Neuron The description of the structure is given using a network known as the back
propagation network. To build a back propagation network, proceed in the following W1j W2j
Wij sigmoidX2 Neuron j X1 yi xi uj Ti
36. 36. fashion. First, take a number of neurons and array them to form a layer. A layer has all its
inputs connected to either a preceding layer or the inputs from the external world, but not
both within the same layer. A layer has all its outputs connected to either a succeeding layer
or the outputs to the external world, but not both within the same layer. Next, multiple layers
are then arrayed one succeeding the other so that there is an input layer, multiple
intermediate layers and finally an output layer. Intermediate layers, that is those that have no
inputs or outputs to the external world, are called hidden layers. Back propagation neural
networks are usually fully connected. This means that each neuron is connected to every
output from the preceding layer or one input from the external world if the neuron is in the
first layer and, correspondingly, each neuron has its output connected to every neuron in the
succeeding layer. Figure 7: Backpropagation Network
37. 37. Generally, the input layer is considered a distributor of the signals from the external
world. Hidden layers are considered to be categorizers or feature detectors of such signals.
The output layer is considered a collector of the features detected and producer of the
response. 4.8. Neural Network Operation The output of each neuron is a function of its
inputs. In particular, the output of the jth neuron in any layer is described by two sets of
equations: (5) (6) For every neuron, j, in a layer, each of the i inputs, Xi, to that layer is
multiplied by a previously established weight, wij. These are all summed together, resulting
in the internal value of this operation, Uj. This value is then biased by a previously
established threshold value, tj, and sent through an activation function, Fth. This activation
function is usually the sigmoid function, which has an input to output mapping as shown in
Figure 4. The resulting output, Yj, is an input to the next layer or it is a response of the neural
network if it is the last layer. Neuralyst allows other threshold functions to be used in place of
the sigmoid described here. Figure 8: Sigmoid Function
38. 38. Equation 1 implements the combination operation of the neuron and Equation 2
implements the firing of the neuron. With a predetermined set of weights, a predetermined
set of threshold values and a description of the network structure (that is the number of
layers and the number of neurons in each layer), it is possible to compute the response of
the neural network to any set of inputs. 4.9. Back propagation Network The difference
between the desired response and the actual response, the erroris determined and a portion
of it is propagated backward through the network. At each neuron in the network the error is
used to adjust the weights and threshold values of the neuron, so that the next time, the error
in the network response will be less for the same input. This corrective procedure is called
backpropagation and it is applied continuously and repetitively for each set of inputs and
corresponding set of outputs produced in response to the inputs. This procedure continues
so long as the individual or total errors in the responses exceed a specified level or until
there are no measurable errors. At this point, the neural network has learned the training
material and you can stop the training process and use the neural network to produce
responses to new input data. W1j W2j Wij Neuron j Error Sigmoid X1 X2 Xi Yj Dj Tj
39. 39. Figure 9: Neuron Weight Adjustment Back propagation starts at the output layer with the
following equations: (7) and (8) For the ith input of the jth neuron in the output layer, the
weight wij is adjusted by adding to the previous weight value, w'ij, a term determined by the
product of a learning rate, LR, an error term, ej, and the value of the ith input, Xi. The error
term, ej, for the jth neuron is determined by the product of the actual output, Yj, its
complement, 1 - Yj, and the difference between the desired output, dj, and the actual output.
Once the error terms are computed and weights are adjusted for the output layer, the values
are recorded and the next layer back is adjusted. The same weight adjustment process,
determined by Equation 7, is followed, but the error term is generated by a slightly modified
version of Equation 8. This modification is: (9) Here, the difference between the desired
output and the actual output is replaced by the sum of the error terms for each neuron, k, in
the layer immediately succeeding the layer being processed times the respective pre-
adjustment weights. The learning rate, LR, applies a greater or lesser portion of the
respective adjustment to the old weight. If the factor is set to a large value, then the neural
network may learn
40. 40. more quickly, but if there is a large variability in the input set then the network may not
learn very well or at all. In many cases, it is useful to use a revised weight adjustment
process. This is described by the equation: (10) This is similar to Equation 7, with a
momentum factor, M, the previous weight, w'ij, and the next to previous weight, w''ij, included
in the last term. This extra term allows for momentum in weight adjustment. Momentum
basically allows a change to the weights to persist for a number of adjustment cycles. The
magnitude of the persistence is controlled by the momentum factor. If the momentum factor
is set to 0, then the equation reduces to that of Equation 7. If the momentum factor is
increased from 0, then increasingly greater persistence of previous adjustments is allowed in
modifying the current adjustment. This can improve the learning rate in some situations, by
helping to smooth out unusual conditions in the training set. As you train the network, the
total error, that is the sum of the errors over all the training sets, will become smaller and
smaller. Once the network reduces the total error to the limit set, training may stop. You may
then apply the network, using the weights and thresholds as trained. It is a good idea to set
aside some subset of all the inputs available and reserve them for testing the trained
network. 4.10.Back Propagation Rule (Or) Generalised Delta Learning Rule : The total
squared error of the output computed by net is minimised by a gradient descent method
known as Back Propagation or Generalised Delta rule. Derivation : Consider an arbitrary
activation function f(x). The derivation of activation function is denoted by F(x). Let
41. 41. 𝑦−𝑖𝑛𝑘 = 𝑧𝑖 𝑤𝑗𝑘 𝑖 (11) 𝑧−𝑖𝑛𝐽 = 𝑣𝑖𝑗 𝑥𝑖 𝑖 𝑌𝑘 = 𝑓(𝑦−𝑖𝑛𝑘 ) (12) The error to be minimised is 𝐸 =
0.5 [𝑡 𝑘 − 𝑦 𝑘 ] 𝑘 ² (13) By use of chain rule we have 𝜕𝐸 𝜕𝑤𝑗𝑘 = 𝜕 𝜕𝑤𝑗𝑘 0.5 [𝑡 𝑘 − 𝑦 𝑘] 𝑘 ² (14)
= 𝜕 𝜕𝑤 𝑗𝑘 0.5[𝑡 𝑘 − 𝑡 𝑦−𝑖𝑛𝑘 ]² = -[𝑡 𝑘 − 𝑦 𝑘] 𝜕 𝜕𝑤 𝐽𝑘 𝑓(𝑦−𝑖𝑛𝑘 ) =-[𝑡 𝑘 − 𝑦 𝑘] 𝑓(𝑦−𝑖𝑛𝑘 ) 𝜕 𝜕𝑤 𝐽𝑘
(𝑦−𝑖𝑛𝑘 ) =-[𝑡 𝑘 − 𝑦 𝑘]𝑓¹(𝑦−𝑖𝑛𝑘 )𝑍𝑗 (15) Let us define δk=-[𝑡 𝑘 − 𝑦 𝑘]𝑓¹(𝑦−𝑖𝑛𝑘 ) (16) Weights on
connections to the hidden unit zj 𝜕𝐸 𝜕𝑣𝑖𝑗 = − 𝑡 𝑘 − 𝑦 𝑘 𝜕𝐸 𝜕𝑣𝑖𝑗 𝑦 𝑘 𝑘 (17)
42. 42. = − 𝑡 𝑘 − 𝑦 𝑘 𝑓(𝑦𝑖𝑛𝑘 ) 𝜕 𝜕𝑣𝑖𝑗 𝑦−𝑖𝑛𝑘 𝑘 = − δ 𝑘 𝜕 𝜕𝑣𝑖𝑗 𝑦−𝑖𝑛𝑘 𝑘 (18) Rewriting the equation
and substituting the values of y-ink = − δ 𝑘 𝜕 𝜕𝑣𝑖𝐽 ( 𝑧𝑗 − 𝑤𝐽𝑘 ) 𝑘 (19) = − δ 𝑘 𝑤𝐽𝑘 𝜕 𝜕𝑣𝑖𝐽 𝑧𝐽 𝑘 =
− δ 𝑘 𝑤𝐽𝑘 𝜕 𝜕𝑣𝑖𝐽 𝑓(𝑧𝑖𝑛𝐽 ) 𝑘 = − δ 𝑘 𝑤𝐽𝑘 𝑓´(𝑧𝑖𝑛𝐽 )( 𝑥𝑖) 𝑘 δj= − δ 𝑘 𝑤𝐽𝑘 𝑓´(𝑧𝑖𝑛𝐽 )𝑘 (20) The weight
updation for output unit is given by Δwjk=−α ∂E ∂wjk (21) =α[𝑡 𝑘 − 𝑦 𝑘 ]f¹(𝑦−𝑖𝑛𝑘 )𝑧𝑗 =αδk 𝑧𝑗
The weight updation for the hidden unit is given by Δ𝑣𝑖𝑗 = −𝛼 𝜕𝐸 𝜕𝑣𝑖𝑗 = 𝛼𝑓1 𝑧−𝑖𝑛𝑗 𝑥𝑖 𝛿 𝑘 𝑘
𝑤𝑗𝑘 (22) =αδj 𝑥𝑖(23)
43. 43. This is a generalised Delta Rule used in the Back Propagation network during training.
4.11.Back Propagation Learning Algorithm based on Levenberg Marquardt Algorithm (LM)
Levenberg – Marquardt algorithm is specifically designed to minimize sum-of-square
errorfunctions of the form. 𝐸 = 1/2 𝐾(𝑒 𝐾)2 =1/2 𝑒 2 (24) Where ek is the error in the kth
exemplar or pattern and e is a vector with element ek. If thedifference between the pervious
weight vector and the new weight vector is small, the errorvector can be expanded to first
order by means of a Taylor series. 𝑒 𝑗 + 1 = 𝑒 𝑗 + 𝜕𝑒 𝑘 𝜕𝑤𝑖 𝑤 𝑗 + 1 − 𝑤 𝑗 25 The error function
can be expressed as 𝐸 = 1/2||𝑒(𝑗) + 𝜕𝑒 𝑘 𝜕𝑤𝑖 𝑤 𝑗 + 1 − 𝑤 𝑗 ||2 (26) Minimizing the error
function with respect to the new weight vector, gives 𝑤(𝑗 + 1) = 𝑤(𝑗) − (𝑍 𝑇 𝑍)−1 𝑍 𝑇 𝑒 (𝑗)(27)
Where (z)ki = 𝜕𝑒 𝑘 𝜕𝑤 𝑖 Since the Hessian for the sum-of-square error function is 𝐻𝑖𝑗 = 𝜕2
𝐸/𝜕 𝑤𝑖 𝜕 𝑤𝑗 = {(𝜕𝑒𝑘 /𝜕 𝑤𝑖 )(𝜕𝑒𝑘 /𝜕 𝑤𝑖 ) + 𝑒 𝑘 𝜕²𝑒 𝑘 𝜕 𝑤𝑖 𝜕 𝑤𝑗 } (28) Neglecting the second
term, the Hessian can be written as H= ZT Z
44. 44. Updating of the weights therefore involves the inverse Hessian or an approximation there
of for nonlinear networks. The Hessian is relatively easy to compute, since it is based on first
order derivatives with respect to the network weights that are easily accommodated by back
propagation. Although the updating formula could be applied iteratively to minimize the error
function, this may result in a large step size, which would invalidated the linearapproximation
on which the formula is based. In the Levenberg-Marquardt algorithm, the error function is
minimized, while the step size is kept small in order to ensure the validity of the linear
approximation. This is accomplished by use of a modified error function of the form. 𝐸 =
1/2||𝑒(𝑗) + 𝜕𝑒 𝑘 𝜕𝑤𝑖 𝑤 𝑗 + 1 − 𝑤 𝑗 ||2 + 𝜆||𝑤(𝑗 + 1) − 𝑤(𝑗)||² (29) where l is a parameter
governing the step size. Minimizing the modified error with respect tow(j+1) gives 𝑤 𝑗 + 1 = 𝑤
𝑗 − (𝑍 𝑇 𝑍 + 𝜆𝐼)−1 𝑍 𝑇 𝑒 𝑗 (30) very large values of l amount to standard gradient descent,
while very small values l of amountto the Newton method.
45. 45. CHAPTER 5 RESULT AND CONCLUISON
46. 46. 5.1.SIMULATION AND RESULT: These are the image in which the repeated objects in
an electrical equipment is tested and its corresponding vector points are found out.
47. 47. 4.4 CONCLUSION: The repeated objects in an electrical equipment is found by using the
SIFT and SURF algorithms. A new software using Artificial neural network, Fuzzy logic,
etc…, will be developed in the next phase .

Anda mungkin juga menyukai