Detecting and Estimating Intermittent Actuator Faults in Linear Stochastic Systems

Intermittent faults (IFs) in practical processes are characterized by non-determinability, repeatability and unpredictability, which brings an enormous challenge to the diagnosis of IFs. In this paper, the detection and estimation problem of intermittent actuator faults for a class of linear stochastic systems is investigated. In order to determine the appearing (dis-appearing) time before the subsequent disappearing (appearing) time, an observer-type detection filter is designed by utilizing a geometric approach. Based on the moving horizon technique, the output of the observer is applied to generate a novel residual, which is more sensitive to the appearing (disappearing) time of the IF. Then, two hypothesis tests are proposed to determine all the appearing time and disappearing time, respectively. Moreover, an estimation algorithm is provided for the magnitude of the IF. Finally, a simulation example on an unmanned arial vehicle system is given to illustrate the effectiveness of the proposed scheme.