PROBABILISTIC FAULT DIAGNOSIS IN DISCRETE EVENT SYSTEMS WITH INCOMPLETE MODELS

Most model-based approaches to fault diagnosis require a complete model of the system. When the model is incomplete, system outputs may be inconsistent with the model, stopping the diagnosis process. In this paper, a probabilistic approach combined with model learning is used to diagnose a system with an incomplete discrete event system model. When an inconsistency arises, probabilistic abductive inference is used to learn improved models of the system. To improve the efficiency of model learning, a tabu search scheme is developed. The improved models are combined with the system outputs for probabilistic diagnosis.