Distributed fault detection in sensor networks via clustering and consensus

In this paper we address the average consensus problem in a Wireless Sensor-Actor Network with the particular focus on autonomous fault detection. To this aim, we design a distributed clustering procedure that partitions the network into clusters according to both similarity of measurements and communication connectivity. The exploitation of clustering techniques in consensus computation allows to obtain the detection and isolation of faulty nodes, thus assuring the convergence of the other nodes to the exact consensus value. More interestingly, the algorithm can be integrated into a Kalman filtering framework to perform distributed estimation of a dynamic quantity in presence of faults. The proposed approach is validated through numerical simulations and tests on a real world scenario dataset.