Robust Failure Detection and Isolation for Input-Output Non-Linear Systems

Abstract

The Unknown Input Observer approach is one of the well-established method for robust failure detection and isolation in linear systems. Although some extensions to non-linear systems have been reported, these extensions are limited to rather restrictive classes of systems. This paper presents an extension of the Unknown Input Observer approach valid for any non-linear system which can be expressed in a NARMAX form. It is shown that the disturbance and failure distribution matrices are composed of the partial derivatives of the model with respect to the corresponding variables. For non-linear systems, these matrices are time-varying, and therefore perfect failure-disturbance decoupling requires time-varying observers. The design procedure of such observers is detailed, and the method is illustrated on simple examples.