Decentralized Fault Detection in Wireless Sensor Networks

This paper introduces a model-free and reference-free decentralized sensor fault identification algorithm ideally suited for wireless sensor networks. Since wireless sensor networks (WSNs) are adopted in many large-scale applications, fault detection of sensors becomes an important subject, especially for low-cost nodes. The common energy-limited characteristic of WSNs, however, makes centralized or complex fault detection algorithms unfavorable. We developed an energy efficient algorithm based on pair-wise comparison. The linear relationship between the outputs of sensor pairs is used to reveal the faulty sensors. The input of the system and knowledge of reference sensors are not required for this method. We also present how to pair sensors in order to minimize power consumption. The performance and limitations of the algorithm are discussed and verified by simulations. Results show that even when error power is comparable to the output signal power and the observation noise level is high, the algorithm is still able to detect most of the faults. Moreover, the decentralized approach reduces the power consumption of the WSN significantly when compared with the traditional centralized method.