Machine Learning Based Approach for Electric Machine NVH Modelling

In this paper, a surrogate modelling approach based on Machine Learning (ML) is investigated to estimate the natural frequencies of electric machine stators. The ML models are trained using a large dataset containing a thousand stator designs, to which an automatic labelling workflow is developed to identify three hundred modes within a wide frequency band up to 20 kHz. The most accurate regressor allows to estimate the majority of the natural frequencies with an error below 1%, given eight input parameters. Additionally, explainable Artificial Intelligence (AI) is explored to describe the importance of each feature to each of the three hundred modes.