Author

Muhammad Azam

Date of Award

January 2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical Engineering

First Advisor

Hossein Salehfar

Abstract

Motor control plays a critical role in industrial automation, ensuring the precise regulation of speed, torque, and position across various applications. Traditional motor control techniques, such as open-loop, closed-loop, and feedback control, have been widely used but face limitations in handling complex and dynamic systems. Recent advancements in machine learning (ML) offer potential improvements in motor control by enabling systems to learn optimal control policies autonomously. This research investigates the integration of Reinforcement Learning (RL) with Field-Oriented Control (FOC) for the control of Permanent Magnet Synchronous Motors (PMSMs). The study aims to address a significant gap in reward function design, specifically the lack of emphasis on speed error in RL-based PMSM control. A modified reward function that penalizes speed deviations is proposed, incentivizing accurate speed tracking while maintaining energy efficiency and system stability. The TD3 algorithm is utilized to train the RL agent within the FOC framework, optimizing motor performance under varying conditions. Simulations compare the proposed reward function with traditional ones, evaluating key performance metrics such as speed tracking accuracy, energy consumption, and torque ripple. The results show that the modified reward function leads to improved motor control, with enhanced speed regulation and a more balanced optimization of performance. This research contributes to the growing body of knowledge on RL-based motor control, offering a novel approach to enhancing PMSM performance and providing insights into future applications in industrial automation.

Share

COinS