Distributed Self Fault Diagnosis in Wireless Sensor Networks using Statistical Methods

Abstract

Wireless sensor networks (WSNs) are widely used in various real life applications where the sensor nodes are randomly deployed in hostile, human inaccessible and adversarial environments. One major research focus in wireless sensor networks in the past decades has been to diagnose the sensor nodes to identify their fault status. This helps to provide continuous service of the network despite the occurrence of failure due to environmental conditions. Some of the burning issues related to fault diagnosis in wireless sensor networks have been addressed in this thesis mainly focusing on improvement of diagnostic accuracy, reduction of communication overhead and latency, and robustness to erroneous data by using statistical methods. All the proposed algorithms are evaluated analytically and implemented in standard network simulator NS3 (version 3.19).

A distributed self fault diagnosis algorithm using neighbor coordination (DSFDNC) is proposed to identify both hard and soft faulty sensor nodes in wireless sensor networks. The algorithm is distributed (runs in each sensor node), self diagnosable (each node identifies its fault status) and can diagnose the most common faults like stuck at zero, stuck at one, random data and hard faults. In this algorithm, each sensor node gathered the observed data from the neighbors and computes the mean to check the presence of faulty sensor node. If a node diagnoses a faulty sensor node in the neighbors, then it compares observed data with the data of the neighbors and predicts its probable fault status. The final fault status is determined by diffusing the fault information obtained from the neighbors. The accuracy and completeness of the algorithm are verified based on the statistical analysis over sensors data. The performance parameters such as diagnosis accuracy, false alarm rate, false positive rate, total number of message exchanges, energy consumption, network life time, and diagnosis latency of the DSFDNC algorithm are determined for different fault probabilities and average degrees and compared with existing distributed fault diagnosis algorithms.

To enhance the diagnosis accuracy, another self fault diagnosis algorithm is proposed based on hypothesis testing (DSFDHT) using the neighbor coordination approach. The Newman-Pearson hypothesis test is used to diagnose the soft fault status of each sensor node along with the neighbors. The algorithm can diagnose the faulty sensor node when the average degree of the network is less. The diagnosis accuracy, false alarm rate and false positive rate performance of the DSFDHT algorithm are improved over DSFDNC for sparse wireless sensor networks by keeping other performance parameters nearly same. The classical methods for fault finding using mean, median, majority voting and hypothesis testing are not suitable for large scale wireless sensor networks due to large devi- ation in transmitted data by faulty sensor nodes. Therefore, a modified three sigma edit test based self fault diagnosis algorithm (DSFD3SET) is proposed which diagnoses in an efficient manner over a large scale wireless sensor networks. The diagnosis accuracy, false alarm rate, and false positive rate of the proposed algorithm improve as compared to that of the DSFDNC and DSFDHT algorithms. The algorithm enhances the total number of message exchanges, energy consumption, network life time, and diagnosis latency, because the proposed algorithm needs less number of message exchanges over the algorithms such as DSFDNC and DSFDHT.
In the DSFDNC, DSFDHT and DSFD3SET algorithms, the faulty sensor nodes are considered as soft faulty nodes which behave permanently. However in wireless sensor networks, the sensor nodes behave either fault free or faulty during different periods of time and are considered as intermittent faulty sensor nodes. Diagnosing intermittent faulty sensor nodes in wireless sensor networks is a challenging problem, because of inconsistent result patterns generated by the sensor nodes. The traditional distributed fault diagnosis (DIFD) algorithms consume more message exchanges to obtain the global fault status of the network. To optimize the number of message exchanges over the network, a self fault diagnosis algorithm is proposed here, which repeatedly conducts the self fault diagnosis procedure based on the modified three sigma edit test over a duration to identify the intermittent faulty sensor nodes. The algorithm needs less number of iterations to identify the intermittent faulty sensor nodes. The simulation results show that, the performance of the HISFD3SET algorithm improves in diagnosis accuracy, false alarm rate and false positive rate over the DIFD algorithm