Data-Efficient Machine Learning Methods for Electric Motor Surrogate Models

Typical electric motor design process involves time-consuming finite-element simulations. In recent years, machine learning and deep learning techniques have been investigated for the development of surrogate models which provide rapid evaluation of motor designs. One drawback of these techniques is the requirement of large dataset in order to achieve reasonable prediction accuracy. In this paper, we present strategies in developing data-efficient machine learning and deep learning surrogate models for electric motors: reducing input dimensions, utilizing physics knowledge for hybrid models, and applying feature extraction methods using geometrical and topological data analysis tools.