Neural-Network-Based Sensor Fault Detection Isolation for Nonlinear Hybrid Systems

摘要

This paper investigates the problem of designing a sensor fault detection, isolation and identification scheme for nonlinear hybrid systems subject to actuator disturbances, sensor noises and plant un-modeled dynamics. In the proposed strategy, the neural network serve as both residuals generator and sensor fault estimator. A sufficient condition for stability of hybrid system is then presented. Based on this condition, stability of the overall system in the presence of actuator and sensor uncertainties as well as unknown faults is guaranteed. Finally the strategy is applied to fault detection, isolation and identification in a nonlinear hybrid chemical process. Simulation results better highlight and validate the performance capabilities and effectiveness of the proposed strategy.