TMCnet News
A Fault-Event Detection Model Using Trust Matrix in WSN [Sensors & Transducers (Canada)](Sensors & Transducers (Canada) Via Acquire Media NewsEdge) Abstract: Sensor nodes and wireless links are prone to failure, while the network is also open to various malicious attacks. When detecting faults, one of the most important tasks is to distinguish an event from fault with discarding irrelevant data by computing over the data as it flows through the sensors. We investigate a method of fault and event detection using trust model in WSN based on similarity matrix. We use similarity matrix to distinguish groups from each other in one cluster. Then cluster head sends the result to its parent where event nodes will be selected from abnormal set according to members in other sets from other cluster heads at the same level. Copyright © 2013 IFSA. Keywords: Fault, Event, Similarity, Matrix, Trust model. (ProQuest: ... denotes formulae omitted.) 1. Introduction Wireless sensor networks comprising of thousands of inexpensive sensor nodes can be set up with relative ease by placing the nodes in predefined locations manually or through the use of robots, as well as by random deployment of self-organizing nodes. A large amount of applications ranging from health, home, environmental to military and defense make use of sensor nodes for collection of appropriate data. The sensor nodes comprising of data collecting, processing, and transmitting units are very small in size and can be densely deployed owing to their low cost. Meanwhile, the sensor nodes and wireless links are prone to failure, while the network is also open to various malicious attacks. So development of fault tolerance in this highly volatile scenario remains an interesting open research issue. Conventional fault tolerance and intrusion tolerance protocols do not translate well to the sensor network domain due to its large scale and the resource constraints on the sensor nodes. Then serials of new tolerance protocols have been proposed to detect various kinds of faults. When detecting faults, one of the most important tasks is aggregating data from multiple sensor nodes to decide if an event has occurred and determining the location of the event, with discarding irrelevant data by computing over the data as it flows through the sensors. The rest of the paper is organized as follows. The detailed trust model is depicted in Section 3. The mechanism of getting partition and merging sets is depicted in Section 4. The comparison and evaluation of our trust model with K-means and TAG are given in Sections 5. The related work and our conclusions are presented in Sections 2 and 6. 2. Related work As in any sensor networks problems, we require a great deal of related material to ensure that our model is helpful and necessary to data aggregation for fault detection. For data aggregation, in [1], authors present a distributed stochastic formulation of the KMeans clustering algorithm which is fully decentralized and intrinsically fault tolerant. And [2] propose the global k-means clustering algorithm, which constitutes a deterministic effective global clustering algorithm for the minimization of the clustering error that employs the k-means algorithm as a local search procedure. But the complexity of kmeans algorithm is o (n*k*t) where n is the number of nodes, k is the number of groups and t is the iteration times. Except for the above method deduced form traditional clustering algorithm using in data aggregation, there are also new methods derived from dataset technology such as in [3]. TAG allows users to express simple, declarative queries and have them distributed and executed efficiently in networks of low-power, wireless sensors. They discuss various generic properties of aggregates, and show how those properties affect the performance of the approach. [4] puts forward a robust aggregation algorithm RAA (robust aggregation algorithm). RAA improves traditional aggregation query based on [3] using reading vector to judge whether a faulty or outlier has happened. With the application and development of trust, fresh methods to data aggregation and fault detection have been introduced in recent years. Ganeriwal, Balzano and Srivastava proposed a reputation-based framework for high integrity WSN named RFSN [5]. In the RFSN framework, a Bayesian formulation is employed to update reputation matrix. Trust Voting [7] is an efficient in-network voting algorithm to determine faulty sensor readings based on Sensor Rank by exploring Markov Chain. Xiao Wei proposed a fault tolerant scheme for data aggregation based event clustering called EFSA [6]. The event cluster was generated before the weighted average data were extracted as approximate event value by way of k-means algorithm. The confident rate of every node in the event cluster was computed and adjusted by iterative algorithm and the CR (trust value) functioned as the weighted value in aggregating data and the indicator of data fault from nodes. Obviously, there have been works to combine traditional method with trust model, but as in [6], Xiao computes nodes' trust rank only based on its own value. In this condition, when event occurs, nodes detecting event may be considered as not trust so that their trust rank will decreased according to the definition TI = e-Av where TI is the trust value of node and A is a proportionality constant that is application dependent. Variable v is maintained for each node at the CH. Each time a node makes a report deemed faulty by the CH its v is incremented by the expression 1-fr. Each time a node makes a report deemed to be correct by the CH its v is decreased by fr (natural error rate) if v is larger than zero. So it is reasonable to consider both nodes' single trust and neighborhood trust. In this paper, we propose a model to compute neighborhood trust and a method to delete fault data and obtain event data. The contribution of our work is: 1. Create a neighborhood trust matrix according to neighbor data similarity in common cluster. 2. Get partition of the cluster with using the trust matrix. 3. Merge sets to determine event data and delete fault data. 4. Compare our model with TAG and K-means. 3. A Trust Model Based on Similarity 3.1. System Model All nodes in the network are identical and arranged into disjoint clusters, each with a set of cluster heads. The CH serves as the data sink for its particular cluster. The nodes in a cluster are within one hop communication of the CH. The CH is rotated over time and CH election is based on energy-related parameters of the constituent nodes. In each cluster, the node that is chosen to be the CH knows the topology of the cluster. Nodes that are within the detection range of an event are called event neighbors for that event. When an event occurs, all the event neighbors are expected to report the occurrence of the event to each other. Then, each node sends its neighbor's information to CH. The CH clusters neighbor's information and create a matrix according to similarity between node and its neighbors. After the creation of matrix, we compute the indirect similarity among all nodes in the cluster to makes a decision on whether the event has occurred and where are the faults. Traditionally, the process of CH making decision is as follows: When an event occurs, all the event neighbors are expected to report the occurrence of the event to the CH. The CH can create a matrix based on similarity between each pair nodes. But, there are some special cases such as shown in Fig. 1. It assumes that node A detects fire in region 1 and B detects fire in region 2. Since A and B are in the same cluster, their detected data is similar and the similarity value in the matrix is close to 1. When locate fire region, it may be considered that the region in border A to B is event occurrence. Here, A and B detect the same event in different regions. 3.2. Trust Matrix When consider a cluster, we get a G = (V, E, s) consists of vertexes V, edges E and similarity weight s. Each vertex is a node and each edge is the connection of two neighbors. We compute the similarity among sensor nodes as: ... where node i and node j is adjacent in location. Assume there is a G as shown in Fig. 2, its trust matrix is like Fig. 3. In order to make decision which node detect event and whether there are fault nodes in this cluster, different types of nodes must be divide into different sets. With conference to trust matrix, we only know trust character between neighbors but not all nodes. Next, an algorithm described in Fig. 4 is proposed to compute indirect trust value. In this algorithm, we define operator T as: ... where t is the threshold of trust. It is considered that if node A and node B is adjacent and their sensor data are highly similar, we may decide they trust each other. Otherwise, we make the trust value decline sharply by plus them. As an example, the indirect trust value of node nlwith four other nodes in figure 2 is (0.9, 0.09, 0.95, 0.98). Then the nodes can be divided into two sets {ni, n2, n4, n5} and {n3}. When data in one set is close to normal value, the set will be regarded as normal set. Otherwise, it should be regarded as abnormal set in which the node may be event node or fault node. 4. Faulty-Event Detection Model 4.1. Detection After partition, how to prove whether a node is event node or fault is a crucial step. Both the above two sets are maintained by cluster head and will be sent to the parent node as a combination. It should be emphasized that the minority set includes the abnormal node's information and its neighbors' list in its cluster. At the next iterating, abnormal data will search for its approximate data from other cluster head's value. Once there are approximate companies and the companies are in its neighbor list, abnormal node will be identified as event node. Finally, sink node will get event nodes' information and delete fault nodes in its topology. The algorithm is described in Fig. 5. For example, there is a network as shown in Fig. 6, since node A and node B are event nodes in different regions that are no fire, they will be considered as abnormal and set in Fi. Then Fi including A and F2 including B are sent to chl-1, meanwhile, E3 from ch3 is sent to chl-1 too. In chl-1, algorithm of fault detecting is run to merge A and B into E3. The followed levels are acted as so on till sink. Finally, there may be more than one event set since there are probably multi events in regions that are not adjacent. Moreover, the rest nodes in Fi are fault nodes that will be deleted from the network. 5. Simulation Our experiment uses ns3 to design. Fifty sensor nodes are distributed in a space of 500 * 700, and the communication radius is set as 60. Each node has two to five neighbors in the experiment and the node's location is already known. We only consider the communication overhead with ignoring calculation cost. It is also assumed that the route is reliable without considering the case of route failure. We consider the energy consumption and error detection rate for simulation. It is shown in Fig. 7 that, when data error rate is changed, trust model maintains an average of packets by 1000. But TAG algorithm gets an increased average of packets as the data error rate growing. This is because that trust model completes the process of aggregation and abnormality detection at the same time. While for TAG, the node that is abnormal need to be confirmed by neighbors' voting. With the fault rate increasing, the number of votes also increased which resulting in more energy consumption. K-means algorithm maintains a little variable average of packets, as its energy mainly consumed on the information exchange when grouping and iterating. And the iteration number is deduced from data but not the fault rate. Except for energy consumption, error detection rate is another important merit to measure a detection algorithm. We define error detection rate as fs/f, where fs is the number of fault nodes that have been detected and f is the total number of fault nodes. Simulation result is shown in Fig. 8 which indicates that the detection rate of matrix model is higher than TAG. Since TAG aggregates data according to its SQL sentences in which a parameter must be set to group, different parameters may deduce different result. So the selection of parameter is crucial to TAG which influence the detection rate. Matrix model uses similarity to group without any parameter and compares data in sets with normal data to decide the belonging of each set. Trust model is more precise than TAG in grouping so as to make higher detection rate. 6. Conclusions In this paper, we investigate a method of faulty and event detection using trust model in WSN based on similarity matrix. We use similarity matrix to distinguish groups from each other in clusters. Then cluster head send the result to its parent where event nodes are selected from abnormal set according to members in other sets from other cluster heads at the same level. This model can successfully detect fault nodes and event nodes. We did a series of simulations to test the performance of our proposed model. The simulation results have showed that fault detection rate is sensitive to node density, and the performance of our model is better than TAG [3] in terms of fault detection rate and energy consumption. Acknowledgements This work is supported by the National HighTech Research & Development Program of China (Grant No. 2011AA010101), the National Basic Research Program of China (Grant No. 2011CB302802) and the National Natural Science Foundation of China (Grant Nos. 610211004, 61370100). Chen acknowledges the part support by Shanghai Knowledge Service Platform Project (No. ZF1213). The authors would like to thank the referees for their invaluable comments and suggestions. References [1]. Giuseppe Di Fatta, Francesco Blasa, Simone Cafiero, Giancarlo Fortino, Fault tolerant decentralised K-Means clustering for asynchronous large-scale networks, Journal of Parallel and Distributed Computing, Vol. 73, Issue 3, March 2013, pp. 317-329. [2]. Aristidis Likas, Nikos Vlassis, Jakob J. Verbeek, The global k-means clustering algorithm, Pattern Recognition, Vol. 36, Issue 2, February 2003, pp. 451-461. [3]. S. R. Madden, M. J. Franklin, J. M. Hellerstein, W. Hong, TAG: A tiny aggregation service for ad-hoc sensor networks, in Proceedings of the 5th Symposium on Operating Systems Design and Implementation, Vol. 36, Issue SI, 2002, pp. 131-146. [4]. Zhong-Bo Wu, Chong-Sheng Zhang, Robust aggregation algorithm in sensor networks, Journal of Software, Vol. 20, No. 7, July 2009. [5]. S. Ganeriwal, L. Balzano, M. Srivastava, Reputationbased framework for high integrity sensor networks, ACM Transactions on Sensor Networks, Vol. 4, No. 3, May 1,2008. [6]. Wei XIAO, Min Xu, Fault tolerant scheme for data aggregation in event cluster over wireless sensor networks, Journal on Communications, Vol. 31, No. 6, June 2010, pp. 112-118. [7]. Guoxing Zhana, Weisong Shi, Julia Deng, SensorTrust: A resilient trust model for wireless sensing systems, Pervasive and Mobile Computing, Vol. 7, No. 4, August 2011, pp. 509-522. [8]. Yanli Yu, Keqiu Li, Wanlei Zhou, Ping Li, Trust mechanisms in wireless sensor networks: Attack analysis and countermeasures, Journal of Network and Computer Applications, Vol. 35, No. 3, May 2012, pp. 867-880. [9]. Mark Krasniewski, Padma Varadharajan, Bryan Rabeler, Saurabh Bagchi, TIBFIT: trust index based fault tolerance for arbitrary data faults in sensor networks, in Proceedings of the International Conference on Dependable Systems and Networks, 2005, pp. 672-681. [10]. Yan-Ming Chen, Yong-Jun Xu, Qiu-Guang Wang, Lei Xie, An adaptive fault-tolerant scheme for wireless sensor networks, in Proceedings of the WRI International Conference on Communications and Mobile Computing (CMC 2009), Vol. 2, 2009, pp. 32-36. 1,2 Na WANG,1 Yi-Xiang CHEN 1 Shanghai Key Laboratory of Trustworthy Computing, East China Normal University, Shanghai, 200062, China 2 School of Computer and information, Shanghai Second Polytechnic University, Shanghai, 201209, China Tel: 086+02150214252 E-mail: [email protected] Received: 18 September 2013 /Accepted: 25 October 2013 /Published: 30 November 2013 (c) 2013 International Frequency Sensor Association |