On-Line Fault Inference for Large-Scale Event-Driven Systems Based on Bayesian Network and Timed Markov Model

This paper presents an on-line fault inference, diagnosis ,and detection strategy for large-scale event-driven controlled systems. First of all, the controlled plant is decomposed into some subsystems, and the global diagnosis is formulated using the Bayesian Network (BN), which represents the causal relationship between the fault and observation in subsystems. The graph structure of the BN is constructed based on the control law adopted in the system. Second, the local diagnoser is developed using the conventional Timed Markov Model, and the local diagnosis results are used to specify the conditional probability assigned to each arc in the BN. By exploiting this decentralized architecture, the computational burden for the diagnosis can be distributed in the subsystems. As the result, the diagnosis for large scale practical system can be realized on-line. Finally, the usefulness of the proposed strategy is verified through some experimental results of an automatic transfer line.