A Review of Machine‐Learning‐Based Control Methods for Permanent Magnet Synchronous Machines

Conventional permanent magnet synchronous machine (PMSM) control methods often struggle to maintain satisfactory performance due to their dependency on parametric machine models. Parameter‐free control methods have garnered significant attention in addressing machine uncertainties; however, their continuous online optimisation increases computational demands and can degrade performance. Recently, machine learning (ML) approaches have emerged as a promising alternative, enabling control methods independent of machine models and parameters. These ML methods are trained offline to create computationally efficient control models without the need for continuous online optimisation. This article reviews and thoroughly investigates various ML‐based control approaches for PMSM drives, including reinforcement learning (RL) and supervised ML. The fundamentals and design principles of these methods are discussed with a focus on PMSM current control. Conventional RL, robust RL and classification‐type and regression‐type supervised ML are implemented and compared with traditional PMSM control methods. Simulation results, quantitative evaluations, and robustness analyses reveal that supervised ML‐based control methods outperform other approaches in the presence of uncertainties for electrical machines.