HMRF-based distributed fault detection for wireless sensor networks

In the practical applications of wireless sensor networks, it is almost inevitable that some sensors become faulty during running. The faulty measurement values will cause a burden to the limited energy of sensor networks. Furthermore, wrong judgement might be deduced because of the faulty data when they reach base station. Therefore, proper fault detection especially for long-term large-scale systems is crucial and challenging. Motivated by the requirement of practical applications, we propose a distributed fault detection approach for wireless senor networks. Firstly, Hidden Markov Random Field (HMRF) model is introduced to characterize the correlations between measurement values and real values of sensor nodes. Then, an errors-in-variables estimation method is presented to obtain the parameters in the HMRF model. Finally, a distributed fault detection algorithm is proposed based on the HMRF model. Both theoretical analysis and simulation results show that the proposed HMRF-based fault detection achieves considerable high detection accuracy and low false alarm rate simultaneously.