Anda di halaman 1dari 5

A Method for Reliability Evaluation of Industrial Wireless Network

Name: Wang Quan Department of Electrical and Computer Engineering The University of Western Ontario, London, Ontario, N6A 5B9, Canada Email: qwang249@uwo.ca
AbstractWireless sensors have started being utilized for industrial application which is typical harsh environment due to complicated layout and plentiful stationary/moving obstacles. As two hot standards for wireless industrial, WirelessHART and ISA100.11a, both adopt the centralized wireless sensor network with mesh topology to enable the reliable monitoring of a variety of environments. This kind of network gives a chance to evaluate the successful communication probability between two nodes through multi-hop or multi-graph. In order to utilize radios for industrial applications, their channel characteristics and link quality must be investigated for optimal design and reliability assessment. This paper presents a finite-Markov model for reliability prediction of this typical industrial wireless network. This finite Markov model for industrial wireless system could be used as a routing schedule algorithm, and also a good way to insure the reliability of industrial wireless system in the field. Keywords-Industrial wireless Reliability Evaluation; network; Finite-state Markov;

I.

INTRODUCTION

It is well known that wireless communication in industry will deliver cost advantages and increase the flexibility in plants, allow a rapid and easy plant setup and reconfiguration, and also allow the reduction of cable breakage when dealing with movable components. Wireless Technologies have existed within industrial and commercial facilities for many years. Many Automatic Meter Reading, Automatic Metering Infrastructure systems are being installed with ZigBee or various proprietary solutions [2]. Since 2006, Hart foundation began to establish an international wireless industrial standard, which named for Wireless HART(Simplified as WirelessHART in this paper) [4], in order to make sure the new industrial wireless system will be more suitable for industrial requirements, such as reliability, real-time, and determination, etc. At the same time, ISA100.11a standard [5], developed by ISA wireless committee was finished in 2009, also focus on industrial wireless application. Compared with other wireless technologies currently used in industrial fields, such as Zigbee, Bluetooth and WiFi, etc, both WirelessHART and ISA100.11a have many special technical advantages which make them more suitable for industrial application. Despite the obvious benefits of wireless technologies for industrial settings, there are still some important constraints to be solved. One of the most important issues is the

unpredictable behavior and the higher error rate in wireless channels. Unlike the office networks, the industrial environment for wireless sensor networks are harsher due to the unpredictable variations in temperature, pressure, humidity, present of heavy equipments etc[6]. This restriction is also caused by limitations of the solutions on most of the current WirelessHART and ISA100.11a system. It is possible to mitigate these limitations by using reliability evaluation and prediction technology in Industrial Wireless Network. If success probability of communication graphs could be evaluated, it would certainly be helpful for routing schedule in the industrial system. By means of calculating the success probability of communication graph, the system manager in the WirelessHART or ISA100.11a system can choose the highest success probability graphs for each node. Moreover, if the success probability of some communication links is below the expected value causing by unpredictable variations in industrial environment, the wireless system can alarm the operator. The operator can take action to make the wireless system reliable as expected. Aim to solve this problem, this paper presents a Finite-state Markov Model to evaluate and predict the communications reliability of industrial wireless network. The rest of this paper is organized as follows. Section 2 introduces briefly the WirelessHART system, which is a typical industrial wireless network. In order to evaluate reliability of communications graph, Section 3 present a finitestate Markov model and relevant algorithm of calculating the successful communication probability between two wireless nodes through multi-hops or multi-graph. Section 4 takes node H as example to describe its communication success probability with the gateway through three different graphs with different orders. Finally, conclusion is drawn. II. BACKGROUND OF WIRELESSHART

The two hottest standards for wireless industrial based on IEEE 802.15.4-2006, WirelessHART and ISA100.11a, both adopt the centralized network structure with mesh topology. Only the WirelessHART is introduces in this section since the ISA100.11a system is similar with the WirelessHART. In such kind of wireless networks, the routing paths are all decided by the system manager.

The WirelessHART network supports a wide variety of devices from many manufactures. Fig.1 illustrates the basic network device types, which include: (1) Field Devices that are attached to the plant process, (2) Handheld which is a portable

Zigbee can be comparatively considered as dynamic mesh network. In other words, the route path is frequently changed in Zigbee network; the route path is comparatively fixed after scheduled and assigned by the system manager in the WirelessHART or ISA100.11a network. Due to this special feature of the WirelessHART or ISA100.11a network, it gives a chance to evaluate successful communication probability between two far nodes through multi hops or multi graphs. So, a finite-state Markov model is derived to evaluate successful communication probability between any nodes in WirelessHART or ISA100.11a network. III. LINK ESTIMATION

Fig 1.The typical network of the Wireless HART system

WirelessHART-enabled computer used to configure devices, run diagnostics, perform calibrations, and manage network information inside each device, (3) A gateway that connects host applications with field devices, and (4) A network manager that is responsible for configuring the network, scheduling and managing communication between WirelessHART devices. Generally, the network manager is integrated in the Gateway. To support the mesh communication technology, each WirelessHART device is required to be able to forward packets on behalf of other devices. Although the graph routing and source routing are two routing protocols defined in WirelessHART, main route method is generally the graph routing: A graph is a collection of paths that connect network nodes. The paths in each graph is explicitly created by the network manager and downloaded to each individual network device. To send a packet, the source device writes a specific graph ID in the network header. All network devices on the way to the destination must be pre-configured with graph information that specifies the neighbors to which the packets may be forwarded. In a properly configured network, all devices generally have at least two devices in the graph by which they may send packets. This redundancy is necessary for transmission retries. The network manager is responsible for the overall management, scheduling, and optimization of the WirelessHART Network. As part of its duties, the network manager initializes and maintains network communication parameter values. It is also responsible for devices joining and leaving the network, managing dedicated and shared network resources, and collecting and maintaining diagnostics about the overall health of the network [7]. To perform its functions the network manager needs information from the devices, information about how the network is to be used, and feedback from the network on how well the network is performing. Generally speaking, the WirelessHART or ISA100.11a network is typical static network with mesh topology, and the

Due to limitation of experimental condition, the link estimation model is refer to the paper [28].In this paper, a pair of MicaZ platform with CC2420 radio chip is used in the experiment. Based on the statistical analysis made for a great deal of sampling values, a link evaluation model in terms of the Gaussian distribution is developed to infer the link quality or its condition between nodes from LQI(1ink quality indicate) values which the model is able to recognize. The relationship between the packet reception rate (PPR) with the mean of LQI is defined by equation (1). The value range of CC2420 radio chips RSSI is between [-100, 0]dbm, similarly the value range of CC2420 radio chips LQI is from 0 to 110 [29].If the LQI is under 70, it is experimentally observed that the PPR is almost zero.
1 106 ( E LQI ) 3 E Prr ( E LQI ) = +0.0656( E LQI ) 4.1948, 0 70 < E LQI 110; 54 E LQI 70.

(1)

The

relationship

between

the

standard

error

Std Prr ( E LQI ) with the mean of LQI is defined by


equation (2).

Std Prr ( E LQI ) = 0.001 ( E LQI )2 + 0.1704 ( E LQI ) 6.5652

(2)

The packet reception rate with LQI value is described by the following Gaussian distribution, where Std Prr ( E LQI ) is standard error.
x 2 Std P ( E LQI ) 2 1 ) (3) p( x) = e ( rr dx 0 2 Std Prr ( E _ LQI ) So, the packet reception rate of link can be approximatively estimated using those equations when LQI value of link is given. Another issue of link estimation is link symmetry problem. Link symmetry indicates the similarity of link quality in both directions. Specifically, if the link quality differs in uplink and downlink, it is classified as an asymmetric link. The differences between uplink and downlink may be attributed to radio hardware differences and environment, etc. To simplify, the link is considered as symmetrical so that the LQI values of two directions are same in this paper. ( x E Prr ( E LQI )2

IV.

A FINITE-STATE MARKOV MODEL

To complete the functions of system schedule, the information, such as their neighbor nodes and each link quality indication parameter and so on, need to be received by the network manager from the devices. The devices will update their relevant information periodically. As soon as it gets the information, the network manager will perform its responsibility for the overall management, scheduling, and optimization of the network. A typical industrial wireless network is supposed as shown in Fig 2. The cycle stands for wireless devices from A to K and a gateway, the line between two wireless nodes represents a communication link of these two nodes, the percentage number under the line means the packet reception rate (PPR) between these two nodes which this line connected. After getting enough information from nodes A to H, the system manager stores all the information. In order to describe the network and related information, a packet reception rate matrix PPR_Matrix is established by the system manager. This N N PPR_Matrix not only indicates the connection relationship of wireless nodes but also the PPR of each connected link. N is the number of nodes in the network. Each column and row of the matrix is allocated to each link. An entry at (i, j) is nonzero if node i can communicate with node j directly, otherwise the entry is zero. The nonzero number in the matrix stands for the PPR in term of percentage.
Plant Gateway 75% Graph 1 A 80% 82% D 50% 75% 65% I 82% E 88% 50% 82% B 84% 78% C 70% System Manager

state transition probability of each links is calculated by equations (1). ( g 1) (1 TPH )PPR_Matrix HE (g) (1) TPHE = 100 Where TP is the state transition matrix, g is the order of graph which link belongs to. For example, the g is equal to 2 if the state transition probability of link EB is calculated because the link EB belongs to the Graph 2. Similarly, the g is equal to 3 if the state transition probability of link FC is computed. The one-step state transition probability from one node to itself is calculated according to equation (2).
N TPii = 1 TPij j =1 0 , if TP = 1 ii

(2)

Where

TPii is denoted as the one-step state transition

probability from node i to itself. The probability of going from state i to state j in n time steps is (n) (3) pij = TP( X n = j X 0 = i) = TPij n In order to calculate the success rate of communication from node i to node j in n time steps, the initial distribution is also needed. The initial distribution is obtained by means of equation (4).

1, chosen node Pini = 0, others

(4)

So, the probability of going from state H to the gateway in n time steps is calculated by equation (5), where N is the number of wireless nodes in current network. (n) n (5) TPi j = Pini TP N N This probability is also considered as the probability of successful communication from node i to node j in n time steps. Thus, the accumulation probability distribution of successful communication from node i to node j in n time steps can easily be given by equation (6).
(n) n ATPi j = Pini TP NN k =1
(n) The accumulation probability distribution ATPi denotes j the success rate of node i sending its data to node to node j within n hops.

78% Graph 2 60% Graph 3

73% F 68% J 55% 72%

G 60% K

(6)

Fig.2 A typical industrial wireless network

As shown in Fig.2, many route paths can be chosen by any node in order to communicate with other nodes. According to the approach of the graph routing defined in the WirelessHART standard, at least two different route graphs will be scheduled and assigned to each node during the setup time of the entire industrial wireless system. Graph 1, Graph 2 and Graph 3 are supposedly assigned to node H for uploading its data to the gateway. The Graph 1 with red color means the first-choose communication paths between the node H and the gateway, and the Graph 2 and 3 are second and third redundant path respectively. In the case of communication failure in the Graph 1, the graph 2 will be chosen for retransmission stead of the graph 1. Therefore, the finite-state Markov model could be set up based on each links packet success rates and graphs priority. Communication reliability evaluation from the node H to gateway would be calculated according to the Markov model and related algorithm. The

V.

SIMULATION

In this section, the finite-state Markov model, which used for evaluating the successful communication probability of graphs in the industrial wireless network, is simulated and calculated by Matlab. In the industrial network, the uppermost communication way is generally from nodes to gateway so that the host can monitor each nodes status and show their information to the operator. Therefore, communication between node H with the gateway is typically chosen as an example in this section. Supposed the network and relevant parameters illustrated in Fig.1, the task is to calculate the successful communication probability from node H to the gateway is shown as follows. First of all, a finite-state Markov model shown in Fig.3 is obtained. Three different graphs are

assigned for node H to communicate with the gateway: Graph 1, Graph 2 and Graph 3. Three sequences of each graph are , and Graph 1 = [ H , D, A, GT ] Graph 2 = [ H , E , B, GT ]
Graph 3 = [ H , E , F , C , GT ]

time step is equal to 4 stands for that the probability is 0.289 for node H successfully sends its data to the gateway through 4 hops.
The accumulation probability distribution of successful communication from node H to the gateway 1

respectively.

According

to

the

equation (4) and (5), the one-step state transition probability matrix TPH GT for node H communicated with the gateway can be figured out.
Probability

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

4 The time step(n)

Fig 5. The Accumulation Probability Distribution

Fig 3. Finite-state Markov Model

Secondly, according to equation (5), the initial distribution is defined as Pini = ( 0, 0, 0, 0, 0, 0, 0, 0,1, 0, 0, 0 ) when node H is
chosen. So, the probability of successful communication going from state H to the gateway in n time steps is calculated by equation (7). (n) n TPH GT = Pini TP N N (7) This probability is also considered as the probability of successful communication from node H to the gateway in n time steps. Fig.4 is the result of the probability distribution of successful communication from node H to the gateway in n time steps. The accumulation probability distribution of successful communication from node H to the gateway in n time steps is shown in Fig.5. In the Fig.5, The both zero values when the time steps are 1 and 2 respectively are reasonable because node H can not send its message to the gateway by one hop or two hops; the highest point means that the probability is 0.565 for node H successfully sends its data to the gateway through 3 hops; the second high point when the
The probability distribution of successful communication from node H to the gateway 0.7

As shown in the Fig.5, the probability of the points when the time step is 4 is 0.853, which means that the probability is 0.853 for node H =successfully sends its data to the gateway within 4 hops. Similarly, the probability is 0.956 for node H successfully sends its data to the gateway within 5 hops. However, probability with 4 hops is meaningful because the longest of those three assigned graphs is 4 hops. The probability with 5 or more means some failure or delays happen.
The probability distributions with different graph orders 0.7

= [ 0, 0, 0, 0, 0, 0, 0, 0,1, 0, 0, 0] TPNn N

0.6

Graph order: 123 Graph order: 312 Graph order: 321

0.5

0.4 Probability 0.3 0.2 0.1 0 1

5 The time step(n)

Fig 6. The Probability Distribution


The accumulation probability distribution with different graph orders 1

0.9

Graph order: 123 Graph order: 312 Graph order: 321

0.8

0.7

0.6

0.6

Probability

0.5

0.4

0.5
0.3

Probability

0.4

0.2

0.1

0.3

5 The time step(n)

0.2

Fig 7. The Accumulation Distribution

0.1

4 The time step(n)

Fig 4. The Probability Distribution

If the graph order from node H to the gateway is changed, the success rate will also change despite the sequence of each graph is not changed. In order to evaluate the effect of graph order for communicate rate, three different graph orders are

compared: 123, 312, 321. The graph order 312 represents that Graph 3 = [ H , E , F , C , GT ] is the first graph for H with the gateway, the Graph 1 = [ H , D, A, GT ] and Graph 2 = [ H , E, B, GT ] are the second and third graph respectively. In the Fig 5, which shows the probability distribution of different graph order, the biggest communication success rate in time step 3 belongs to graph order 312, communication success rate is the least one at the same step when graph order 321 is chosen. Fig 6 is figure of the accumulation distribution with three different graph orders. For accumulation distribution view, the graph order 312 is the best choose, and the performance of graph order 321 is worst. So, though PPR of the first link (75%) in the graph [ H , D, A, GT ] is bigger than that of the first link (65%) in the graph [ H , E , B, GT ] , the graph [ H , E , B, GT ] would better be the first graph in stead of the graph [ H , D, A, GT ] .Therefore, using this finite Markov model and relevant algorithm, the system manager in the wireless industrial network can choose the most reliable routing graph by means of not only comparing single link but also considering in whole network view. VI. CONCLUSION

Deji Chen, Mark Nixon, Aloysius Mok. WirelessHART : RealTime Mesh Network for Industrial Automation. 1st Edition., 2010, XXII, 276 p. 40 illus., 20 in color. Hardcover ISBN: 9781-4419-6046-7 [8] ZHUJian,ZHAO Hai,etc. LQI-Based Evaluation Model of Wireless Link. Journal of Northeastern University (Natural Science). V01,29. Sep, 2008. Pp:1262-1265.
[7]

In the same way, all communication reliability among each nodes and gateway could be evaluated as long as communication graphs are identified. Using this finite Markov model and relevant algorithm, the system manager in the wireless industrial network can choose the most reliable routing graph by means of not only comparing single link but also considering in whole network view. In other words, this finite Markov model for industrial wireless system could be used as a routing schedule algorithm. Whats more, if the success probability of some communication links is below the expected value causing by unpredictable variations in industrial environment, the wireless system can alarm the operator. The operator can take action to make the wireless system reliable as expected. Therefore, this finite Markov model and its relevant algorithm also give a good way to insure the reliability of industrial wireless system in the field. REFERENCES
[1]

[2]

[3]

[4] [5]

[6]

A.Willig. Recent and Emerging Topics in Wireless Industrial Communications: A Selection. IEEE Transactions on Industrial Informatics, 4(2):102124, 2008. Baker, N. ZigBee and Bluetooth: Strengths and weaknesses for industrial applications, IEE Computing & Control Engineering, vol. 16, no. 2, pp 20-25, April/May 2005. Gianluca Cena, Ivan Cibrario Bertolotti,Adriano Valenzano, and Claudio Zunino. Evaluation of Response Times in Industrial WLANs. IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 3, NO. 3, AUGUST 2007 HART Foundation, www.hartcomm.org. The ISA100 Standards - Overview and Status. http://www.isa. org/MSTemplate.cfm?Section=Home1034&Site=SP100,_Wirel ess_Systems_for_Automation&Template=/ContentManagement/ MSContentDisplay.cfm&ContentID=80821. Low, K.S.; Win, W.N.N.; Er, M.J. Wireless Sensor Networks for Industrial Environments. International Conference on Computational Intelligence for Modelling, Control and Automation, 2005 and International Conference on Intelligent Agents, Web Technologies and Internet Commerce (CIMCAIAWTIC05). Volume 2, 28-30 Nov. 2005. Page(s):271 276.

Anda mungkin juga menyukai