Enhancing BLDC Motor Control via Reinforcement Learning-Based Online Tuning of SS-PID Parameters

This study addresses the performance degradation of Brushless direct current motors caused by parameter drift and sensor noise. Conventional proportional-integral-derivative controllers struggle with these issues, as they amplify noise and cannot adapt to system changes. We propose a state-space proportional-integral-derivative controller with an online tuning framework that uses a deep deterministic policy gradient reinforcement learning agent. This approach avoids noise amplification by using observer-based state estimation instead of a direct derivative path, while the reinforcement learning agent adapts to parameter drift in real-time. Simulations confirm our method provides superior speed-tracking and reduced error compared to conventional methods, ensuring robust, long-term performance even under degraded conditions.