Automatic Design System With Generative Adversarial Network and Vision Transformer for Efficiency Optimization of Interior Permanent Magnet Synchronous Motor
Yuki Shimizu
Abstract
Interior permanent magnet synchronous motors are becoming increasingly popular as traction motors in environmentally friendly vehicles. These motors, which offer a wide range of design options, require time-consuming finite-element analysis to verify their performance, thereby extending design times. To address this problem, in this article, we propose a deep learning model that can accurately predict the iron loss characteristics of different rotor topologies under various speed and current conditions, resulting in an automatic design system for the internal permanent magnet synchronous motor rotor core. Using this system, the computation time for efficiency maps is reduced to less than 1/3000 of the time required for finite-element analysis. The system also shows efficiency optimization results similar to the best results of previous research while reducing the computational time for optimization by one or two orders of magnitude.