Abstract
Fault diagnosis is an important issue in wireless sensor networks (WSNs) because the node/link failures not only decrease the accuracy of surveillance result, but also shorten the lifetime of network. In order to minimize damages caused by node/link failures, the node/link failures must be detected and recovered by system supervisors as soon as possible. Although many diagnosis methods for WSNs are proposed, these methods are works either in a centralized manner or in the networks with bidirectional links. The centralized methods often need all sensors in the network to periodically report the node/link information to a speci c node, resulting in heavy burdens of sensors and generating a large amount of overhead in networks. In addition, in actual practice, the links in WSNs are either unidirectional or bidirectional. In this thesis, a distributed cooperative diagnosis method, termed CDM, is proposed to work in networks with unidirectional links and minimize the message overhead required for diagnosis. Using NS-2 simulations, we evaluate the performance of the proposed method (CDM) and a well-known diagnosis method (TinyD2) in terms of detection ratio, fault-alarm ratio, and diagnosis overhead. The simulations demonstrate that our diagnosis method has good performance in terms of detection ratio and diagnosis overhead and provides comparable fault-alarm ratio to TinyD2.