Evaluation of Feasibility for Optimal Motor Design using Deep Q-Network

Abstract

The use of the reinforcement learning algorithm DQN (Deep Q-Network) can increase the design variables and offers the advantage of enabling more versatile motor optimization design. This paper evaluates the potential and advantages of reinforcement learning-based motor optimization design, which is still in its early stages globally. To demonstrate the performance of the proposed method, the geometry and drive conditions for reducing torque ripple of a FWSM (Field-Wound Synchronous Motor) using DQN were optimized. The proposed method exhibited excellent optimization performance, and it is expected that a more versatile optimization model can be developed by increasing the number of analysis cases.