Robust fault detection and isolation in bond graph modelled processes with Bayesian networks

Abstract

The main objective of this paper is to present a new method for Fault Detection and Isolation (FDI) of non-linear uncertain parameters systems modelled by bond graphs (BGs) with Bayesian networks (BN). From the BG model of a process, residuals, which are fault detectors, are determined directly from the Diagnostic Bond Graph (DBG). In ideal conditions, those residuals are equal to zero. But in practice, owing to uncertainties, perturbations and measurement noises, residuals are different from zero. Classical approaches used thresholds to deduce whether a process is in normal operating mode or in faulty mode. In our approach, we generate a statistical decision procedure to detect the operating mode. For isolation, a Bayesian network is generated by covering the causal paths of the DBG, and the method proposed by Weber et al. is exploited. A simulation example on a three tanks system is provided to show the efficiency of the proposed FDI procedure.