Data based construction of Bayesian Network for fault diagnosis of event-driven systems

This paper presents a construction strategy of Bayesian Network (BN) structures in decentralized fault diagnosis of event-driven systems based on probabilistic inference. In the proposed decentralized diagnosis method, a fault is identified using the BN and Timed Markov Model (TMM). The BN represents the causal relation between the faults and the observed event sequences in subsystems, and the structure of the BN plays an essential role since the computational complexity and the fault diagnosis performance depend on it. This paper particularly focuses on a construction strategy of the BN based on an importance indicator of the arc, which expresses independence properties between faults and observations, in fault diagnosis of event-driven systems. Finally, the usefulness of the proposed strategy is verified through some experimental results of the automatic transfer line simulated on a PC.