Multiple Faults Model-Based Detection and Localisation in Complex Systems

Abstract

In this paper, we propose a general solution to detect and isolate multiple faults in complex systems. The proposed solution rests on the model-based diagnosis approach. A technique combining the adaptive thresholding and the fuzzy logic detection is proposed in the detection phase. To optimally adjust the fuzzy membership functions parameters, we rely on the PSO. The results of the detection module are presented as a colored causal graph representing the state of different system variables from the green nominal state to the red faulty state. In the isolation phase, we compared the results given by three classical methods: causal graphs technique, the signature matrix analysis method and the logical consistency method with faults models in order to choose the best one to localize efficiently multiple faults in complex systems. The experiments focused on a diagnosis benchmark: the three-tank hydraulic system.