A Diagnosis Method for Linear Stochastic Systems With Parametric Failures

Abstract

A method of failure diagnosis for discrete-time stochastic systems with some parametric failures is proposed. It is assumed that the time instances of failure occurrence and the dynamic behaviors of failure parameters cannot be estimated in advance. The idea of Two-Filters is utilized as the filter construction for the diagnosis system. The new feature of the approach is to use a modified adaptive extended Kalman filter which estimates system parameters and states simultaneously under the failure mode. The adaptive mechanism of this filter is based on hypothesis tests for the innovation and for the estimation errors of system parameters, and modifies the filter gains if the hypothesis is rejected. In addition, a discriminator, which uses the trace value of the approximate local sensitivity, is developed in order to help the operator’s tasks. Experimental results indicate that the proposed method on the failure diagnosis is effective for several numerical examples.