Model-Free Fault Detection Based on Performance Residual for Feedback Control Systems

Abstract

This brief proposes a control performance-oriented fault detection (PFD) strategy for closed-loop linear time-invariant systems. Firstly, the performance index with the summation of quadratic form over the infinite time interval is investigated in virtue of a data-driven residual-centered closed-loop model, which can be constructed utilizing input and output data instead of system models. Then the adaptive residual generator is presented to update the parameters of the data-driven model in case of system changes caused by anomalies, serving the information of the faulty plant for PFD. Further, the performance residual, which represents the performance degradation, is defined based on the Bellman equation. By embedding the predesigned controller parameters in the performance residual generator, the performance degradation caused by system faults and controller mismatches can be detected. Additionally, the proposed scheme can be applied in various operating points without revising the detector parameters. Finally, a randomized algorithm-aided threshold design is developed to handle unknown input. Case studies on a circuit model are conducted to show the superiority of PFD.