Fault Diagnosis Algorithms for Wireless Sensor Networks

Abstract

The sensor nodes in wireless sensor networks (WSNs) are deployed in unattended and hostile environments. The ill-disposed environment affects the monitoring infrastructure that includes the sensor nodes and the links.
In addition, node failures and environmental hazards cause frequent topology change, communication failure, and network partition. This in turn adds a new dimension to the fragility of the WSN topology. Such perturbations are far more common in WSNs than those found in conventional wireless networks. These perturbations demand efficient techniques for discovering disruptive behavior in
WSNs. Traditional fault diagnosis techniques devised for wired interconnected networks, and conventional wireless networks are not directly applicable to WSNs due to its specific requirements and limitations.
System-level diagnosis is a technique to identify faults in distributed networks such as multiprocessor systems, wired interconnected networks, and conventional wireless networks. Recently, this has been applied on ad hoc networks and WSNs. This is performed by deduction, based on information in the form of results of tests applied to the sensor nodes. Neighbor coordination-based system-level diagnosis is a variation of this method, which exploits the spatio-temporal correlation between sensor measurements. In this thesis, we present a new approach to diagnose faulty
sensor nodes in a WSN, which works in conjunction with the underlying clustering protocol and exploits spatio-temporal correlation between sensor measurements. An advantage of this method is that the diagnostic operation constitutes real work performed by the system, rather than a specialized diagnostic task. In this way, the normal operation of the network can be used for the diagnosis and resulting less time and message overhead. In this thesis, we have devised and evaluated fault diagnosis algorithms for WSNs considering persistence of the faults (transient,
intermittent, and permanent), faults in communication channels and in one of the approaches, we attempt to solve the issue of node mobility in diagnosis.
A cluster based distributed fault diagnosis (CDFD) algorithm is proposed where the diagnostic local view is obtained by exploiting the spatially correlated sensor measurements. We derived an optimal threshold for effective fault diagnosis in sparse networks. The message complexity of CDFD is O(n) and the number of bits exchanged to diagnose the network are O(n log2 n). The intermittent fault diagnosis is formulated as a multiobjective optimization problem based on the inter-test interval and number of test repetitions required to diagnose the intermittent faults. The two objectives such as detection latency and energy overhead are taken into consideration with a constraint of detection errors. A high level (> 95%) of detection accuracy is achieved while keeping the false alarm rate low (< 1%) for sparse networks. The proposed cluster based distributed intermittent fault diagnosis (CDIFD) algorithm is energy efficient because in CDIFD, diagnostic messages are sent as the output of the routine
tasks of the WSNs. A count and threshold-based mechanism is used to discriminate the persistence of faults. The main characteristics of these faults are the amounts of time the fault disappears. We adopt this state-holding time to discriminate transient from intermittent or permanent faults. The proposed cluster based distributed fault
diagnosis and discrimination (CDFDD) algorithm is energy efficient due to the improved network lifetime which is greater than 1150 data-gathering rounds with transient fault rates as high as 20%.
A mobility aware hierarchal architecture is proposed which is to detect hard and soft faults in dynamic WSN topology assuming random movements of nodes in the WSN. A test pattern that ensures error checking of each functional block of a sensor node is employed to diagnose the network. The proposed mobility aware cluster based distributed fault diagnosis (MCDFD) algorithm assures a better packet delivery ratio (> 80%) in highly dynamic networks with a fault rate as high as 30%. The network lifetime is more than 900 data-gathering rounds in a highly dynamic network with a fault rate as high as 20%.