Model-based reinforcement Learning-Based Control Concepts for Permanent Magnet Synchronous Machines

This thesis project focuses on implementing and evaluating model-based Reinforcement Learning concepts for the control of Permanent Magnet Synchronous Machines (PMSM). The main goal is to learn a policy to output optimal switching angles for torque/current tracking. Initially, steady-state operation is discussed to get a general insight into the system behavior and to compare the performance to state-of-the-art controllers. The concept is then extended to address transient operation. Optionally, there is a test bench validation using an actual PMSM to assess the controller's performance in a real-world environment.