Adaptive PI Controller Based on a Reinforcement Learning Algorithm for Speed Control of a DC Motor.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Robotics

Background:

• Automated industrial processes rely on controllers for precise output signal generation.
• Classical PID controllers require retuning when system conditions change, impacting production.
• Existing control methods face challenges with dynamic or changing system parameters.

Purpose of the Study:

• To develop an adaptive PID controller for DC motor speed control.
• To utilize artificial intelligence, specifically reinforcement learning, for autonomous controller tuning.
• To eliminate the need for prior system knowledge in controller gain determination.

Main Methods:

• Implemented an adaptive PID controller for a DC motor speed plant.
• Employed a reinforcement learning algorithm with an actor-critic agent.
• Utilized the Deep Deterministic Policy Gradient with Twin Delayed (DDPG TD3) algorithm.
• Configured a neural network with 300 neurons for agent learning.

Main Results:

• The adaptive PID controller successfully generated appropriate gains without system identification.
• The DDPG TD3 algorithm facilitated policy optimization and critic training.
• Performance comparison demonstrated the effectiveness of the developed controller against classical methods.

Conclusions:

• The developed adaptive PID controller offers a robust solution for DC motor speed control in dynamic environments.
• Reinforcement learning, particularly DDPG TD3, provides an effective mechanism for automated controller tuning.
• This AI-driven approach enhances industrial process efficiency by eliminating manual retuning requirements.