Application of fuzzy inference systems to detection of faults in wireless sensor networks

In this paper we present a fault detection strategy for wireless sensor networks. The strategy is based on modeling a sensor node by Takagi–Sugeno–Kang (TSK) fuzzy inference system (FIS), where a sensor measurement of a node is approximated by a function of the sensor measurements of the neighboring nodes. We also model a node by recurrent TSK-FIS (RFIS), where the sensor measurement of the node is approximated value of the node itself. Temporary errors in sensor measurements and/or communication are overcome by redundancy of data gathering. A node which has developed a faulty sensor is not completely discarded because it is useful for relaying the information among the other nodes. Each node has its own fuzzy model that is trained with input of neighboring sensors’ measurements and an output of its actual measurement. A sensor is declared faulty if the difference between the outcome of the fuzzy model and the actual sensor measurement is greater than the prescribed amount depending on the physical quantity being measured. Simulations are performed using the fuzzy logic toolbox of Matlab. We also give a comparison of obtained results to those from a feed-forward artificial neural network, recurrent neural network and the median [1] of measured values of the neighboring nodes. & 2012 Elsevier B.V. All rights reserved.