Reinforcement Learning-based Direct Torque Control of Externally Excited Synchronous Motors: a Proof of Concept

In externally excited synchronous motors (EESMs), the controllable rotor excitation provides an additional degree of freedom that is not available in permanent magnet synchronous motors. The drawback of this degree of freedom is the more intricate system behavior and the cross-coupling between the rotor and stator circuits, complicating the application of control strategies. Detailed system identification is required to describe a model for the design process of a conventional controller. To circumvent the necessity for a system model for control strategies such as field-oriented control and model predictive control, the application of model-free control methods like reinforcement learning (RL) is an interesting alternative. By leveraging a data-driven approach, the RL controller can learn to handle these coupling effects between rotor and stator circuit with minimum a priori knowledge and without requiring an explicit system model or plant parameters. This proof of concept demonstrates that an RL-based torque controller can effectively tackle the system dynamics of an EESM, while adhering to operational constraints such as current and voltage limits.