Principles of Model-Based Fault Detection

Abstract

The paper outlines the principles of fault detection and isolation (FDI) in dynamic systems using a mathematical model of the system. Implemented on a digital computer, these model-based algorithms can efficiently be applied to signal validation, on-line detection of abrupt faults and early fault diagnosis in a long-term system supervision. The basic concept of the model-based approach to fault detection and isolation is described, and the different methods of residual generation developed in the past two decades are reviewed. Among the methods considered are the parameter identification approach, the different observer-based concepts, the parity space approach and the knowledge-based strategy. It is also shown how structured residuals for the localization of the faults can be generated with the aid of observer schemes. As an issue of great practical relevance the robustness with respect to modeling errors is taken into consideration. Here emphasis is placed upon the unknown input observer approach and its application to instrument, actuator and component fault detection.