Development of Machine Learning-based Design Platform for Permanent Magnet Synchronous Motor Toward Simulation Free

This paper proposes an approach that combines machine learning (ML) and equivalent magnetic circuit (EMC) analysis for the design of surface-mounted permanent magnet synchronous motors (SPMSMs). This is aimed at building a service platform for non-professional users who need motor designs. The developed method can quickly obtain PMSM designs and parameters with a certain level of accuracy without using finite element (FE) simulation. Therefore, the users can take advantage of the platform and obtain the motor designs in a few seconds. The users only need to input key specifications, such as the torque required, speed, and voltage available, and the ML-based platform can predict and output a design that satisfies the specifications. In this paper, an EMC model is first developed, and FE is employed to validate its accuracy. With the EMC, more than 6000 motor models are produced as the data pool for the ML. The ML algorithms are trained by making use of this motor design data pool so that the design platform can be built. Finally, the FE simulations validate the accuracy of the proposed method.