Robust model-based fault detection using adaptive robust observers

A goal in many applications is to combine a priori knowledge of the physical system with experimental data to detect faults in a system at an early enough stage as to conduct preventive maintenance. The mathematical model of the physical system is the information available before hand. One of the key issues in the design of model-based fault detection schemes is the effect of the model uncertainties such as severe parametric uncertainty and unmodeled dynamics on their performance. This paper presents the application of a nonlinear model-based adaptive robust observer design to help in the detection of faults in sensors for a class of nonlinear systems. The observer is designed by explicitly considering the nonlinear dynamics of the system under consideration. Robust filter structures are used to attenuate the effect of the model uncertainty which combined with controlled parameter adaptation helps in reducing the extent of model uncertainty and in increasing the sensitivity of the fault detection scheme to help in the detection of incipient failure. This continuous monitoring of the state estimates enables us to detect any off-nominal system behavior and detect faults even in the presence of model uncertainties.