Performance Degradation Monitoring and Self‐Healing Control for Industrial Control Systems via Reinforcement Learning Aided Optimal Feedback Compensation

作者：Feng Gao, Xu Yang, Jingjing Gao, Jian Huang · 发表于：International Journal of Adaptive Control and Signal Processing · 年份：2025 · DOI：10.1002/acs.4032 · 被引用次数：1 · 研究领域：Fault Detection and Control Systems、Adaptive Dynamic Programming Control、Advanced Control Systems Optimization

ABSTRACT The control performance of the automation system significantly influences the reliability of industrial processes and product quality. This article focuses on developing an integrated architecture for monitoring control performance, and recovering from degradation in industrial control systems under abnormal operating conditions. Initially, an online performance evaluation matrix is identified, which demonstrates variations of control performance in dynamic processes. It is produced by parameterizing the performance value function using the recursive Bellman equation, and it provides more full‐dimensional data than standard scalar indicators. On this basis, a performance degradation detection method is proposed by measuring the deviation between the performance evaluation matrix identified in normal and abnormal conditions. Furthermore, a redundancy‐based self‐healing control approach using dynamic feedback compensation allows the closed‐loop system to recover from performance degradation without changing the predesigned controller. The control gain of the proposed self‐healing controller is obtained by model‐free reinforcement learning, avoiding the requirement for an accurate representation of complex industrial systems. Finally, the effectiveness of the proposed approach is verified by a benchmark study on the three‐tank system.