Abstract
Wireless sensor networks (WSNs) comprise a collection of sensors used to collect data, which allow knowing the state of a zone or a monitored system. Sensor are usually deployed in harsh environments, where failures are common. In this work, we propose a distributed fault detection (DFD) algorithm based on the nonlinear automatic regression recurrent non-linear autoregressive with exogenous inputs (NARX) for failure detection. Defective sensors are identified by comparisons between series of actually measured values and their predicted values. The proposed method, in the first step, divides the network into a set of cliques, forming different areas. In the next step, the damaged cliques are identified using the Gaussian distribution theorem. Finally, the NARX approach is applied only to the damaged cliques to determine the defective nodes. The comparisons of simulation results to other existing algorithms show that the proposed method reaches the best results.