A General Failure Detection, Isolation and Accommodation System with Model Uncertainty and Measurement Noise

Abstract

The problem of failure detection, isolation and accommodation has received high interest recently. This task is achieved using the information of plant's model and input/output measurements. It is therefore important to study the effects of model uncertainty and measurement noise. Among the two effects, the former is much more severe and its analysis is very difficult. So far, only a threshold of nonzero occurrence (caused by these effects) of a normally zero-valued error signal is established in [3], but only after a very elaborate derivation and only for a single Kalman filter based failure detection system. Recently, such a failure detection and isolation system (achieved by the observance of the nonzero occurrence of some error signals) is derived in (6], which requires a multiple of robust failure detection observers. This paper establishes the nonzero threshold for model uncertainty and output measurement noise, for this system. This paper also proposes an adjustable state feedback control for failure accommodation, which is uniquely enabled by the failure isolation capability of this system.