Electric Machine Two-dimensional Flux Map Prediction with Ensemble Learning

Numerous finite-element simulations are required to evaluate the performance of an electric machine at different operating points, posing a great challenge to the design optimization of such electric machines. Surrogate modeling approaches have been investigated in recent years to speed up the analysis and optimization process, including machine learning and deep learning models. In particular, various convolutional neural network based deep learning models have been proposed and trained to predict motor performances for a given motor design. However, larger dataset and relatively long training time are required for such deep models. In this paper, we present a method for the rapid prediction of 2d flux maps using ensemble learning technique, with multiple relatively simple regression models. We show that the technique is much faster to train compared with deep convolutional networks, while achieving improved accuracy.