Belief consensus–based distributed particle filters for fault diagnosis of non-linear distributed systems

Abstract

This article presents a study on distributed fault diagnosis of spatially distributed systems composed of physically interconnected subsystems. To handle the complexity of such complicated systems, the overall state space model is decomposed to its constituent subsystems. Moreover, hybrid modelling is considered for the fault diagnosis problem. In this case, each subsystem is modelled as a hybrid one. Two different modes (healthy and faulty) are considered as discrete modes of the hybrid sub systems. A particle filtering–based algorithm is then proposed for the purpose of fault diagnosis of generally non-linear and non-Gaussian systems. The proposed method is applicable to a large class of distributed systems that undergo autonomous transitions from healthy to faulty modes. To implement the algorithm in a decentralized architecture, we employ embedded systems, called agents. Agents cooperate in the process of fault diagnosis by fusing local information to make the result of the decentralized approach asymptotically equivalent to the corresponding centralized one. To this end, belief (conditional probability) consensus algorithm is exploited. Simulation results are applied to a strongly interconnected four-tank system (as a benchmark) to demonstrate the efficiency of the proposed method and how it improves the previous ones.