A fast self-tuning method for PID parameters of AMBs based on Deep Deterministic Policy Gradient algorithm

Active magnetic bearing(AMB) is an advanced contactless bearing system for high rotation speed equipment but cannot work stably without closed-loop control. Normally multiple proportional integral derivative (PID) controllers are used to control the multi-axis AMB system. However, existing AMB PID controller tuning methods mainly focus on optimization and cannot turn an AMB system from unstable to stable. In this article, a model free AMB PID self-tuning method based on the Deep Deterministic Policy Gradient (DDPG) algorithm is proposed to solve this problem. This approach enables the autonomous tuning of PID controller parameters via the DDPG algorithm. Additionally, a rapidly convergent reward function design method applicable to large action spaces is introduced to enhance the proposed algorithm's general applicability and validated. For an unknown and untuned AMB system, the proposed tuning method provides a viable parameter solution within a short timeframe while delivering acceptable performance. Furthermore, the designed DDPG-based self-tuner demonstrates acceptable robust capabilities under sinusoidal, step, impulse, and composite disturbances. The proposed self-tuning method not only reduces controller design complexity but also provides a foundation for the engineering application of AMB system.