Subspace-based Fault Detection Using Performance Metrics

This paper investigates subspace-based identification for fault detection using performance metrics. Utilizing performance metrics to unify the design of different types of robust residuals enables robustness to unknown inputs and sensitivity to sensor or actuator faults. An optimized moving window algorithm is employed to address the challenge of detecting minor faults. The effectiveness of the algorithm is validated through simulation studies on high-speed train traction system.