A Reinforcement Learning-Based Fault Tolerant Control Design Approach Using the Double Q-Learning Algorithm

This paper presents a novel approach, the Reinforcement Learning-Based Fault-Tolerant Control strategy, designed to address sensor faults in control systems. Initially, a closed-loop system is established using the Conventional Proportional-Integral-Derivative controller. To investigate the system’s robustness, a simulated sensor fault is intentionally introduced, and the corresponding results are analyzed. In addition to the Reinforcement Learning-Based Fault-Tolerant Control strategy, we propose the utilization of the Double Q-Learning algorithm to tackle unknown faults in the control system. This enhanced approach aims to improve the performance of the controller under challenging conditions, ensuring reliable operation even in the presence of unexpected faults. To evaluate the effectiveness of both strategies, a comprehensive study is conducted on a Continuous Stirred Tank Reactor, a representative benchmark for control applications. Through extensive experimentation, the results demonstrate the superiority of the proposed Reinforcement Learning-Based Fault-Tolerant Control strategy compared to the Conventional Q-Learning algorithm, particularly in critical and sensitive systems like the Continuous Stirred Tank Reactor. The key contribution of this research lies in the ability of the Reinforcement Learning-Based Fault-Tolerant Control approach to enhance the speed of decision-making, an essential aspect in managing time-sensitive systems. By effectively addressing sensor faults and improving control performance, this method proves to be a promising solution for real-world applications. By combining theoretical analysis, simulation studies, and practical experiments on the Continuous Stirred Tank Reactor, this paper offers valuable insights into the development of robust and reliable control strategies, opening avenues for further research in the field of reinforcement learning-based fault-tolerant control.