Robust Fault Detection for Nonlinear Systems in the Presence of Unstructured Uncertainties

Abstract

For model-based fault detection and isolation (FDI), the residual plays an important role in indicating the occurrence of faults. The robust residual generation for nonlinear systems is in general a difficult task and the presence of unstructured uncertainties may lead to an NP-hard problem. This paper presents an alternative approach to residual generation for non linear systems with unstructured uncertainties by stochastic qualitative techniques. The proposed method combines stochastic modeling with the qualitative technique, forming a framework to cope with the uncertainty propagation into the system behavior as time elapses. The arising problem of spurious solutions in qualitative modeling is solved by a qualitative observer. The generated residual is normalized and less conservative compared to the adaptive threshold method.