A Fault Detection Scheme Based On Stochastic Qualitative Modeling

Abstract

This paper studies the problem of fault detection for nonlinear systems with parameter uncertainties. The kernel technique used is stochastic qualitative modeling. When only incomplete information about systems is available, a qualitative treatment might be an appropriate alternative for characterizing the dynamic behavior of systems. In addition, the introduction of the stochastic setting through randomized approximation enhances the ability of knowledge representation. The integration of stochastic and qualitative modeling techniques leads to a hidden Markov model which copes with the uncertainty propagation as time elapses. In the proposed approach, the new model is used to construct qualitative output observers for residual generation. Credibility indices are introduced to evaluate the qualitative estimate and the residual, and the detectability issue is discussed.