Fault Detection and Isolation Based on Unknown Input Set-membership Observer

This paper proposes an actuator fault diagnosis method for a class of discrete-time linear dynamic systems subject to unknown but hounded disturbances and noises. The proposed unknown input observer has a new structure and the actuator faults of the system are taken as unknown inputs that provides more robust and more design degrees of freedom. A direct method based on zonotopes is used to estimate the threshold of residual generated by the unknown input observer. Based on the proposed method, a fault detection and isolation strategy using a set of unknown input set-membership observers are presented. The effectiveness of the proposed method is verified by numerical simulation of a flight control system.