Distributed Fault Detection Using Consensus of Markov Chains

Abstract

We propose a fault detection procedure appropriate for use in a variety of industrial engineering contexts, which employs consensus among a group of agents about the state of a system.Markov chains are used to model subsystem behaviour, and consensus is reached by way of an iterative method based on estimates of a mixture of the transition matrices of these chains. To deal with the case where system states cannot be observed directly, we extended the procedure to accommodate Hidden Markov Models. Index TermsFault detection, consensus algorithm, mixtures of Markov chains, the EM algorithm, Hidden Markov Model (HMM), multi-agent systems.