On the use of Machine Learning to Improve the Analytical NVH Predictions of Electric Machines

In this paper, Machine Learning (ML) is used to improve the accuracy of the traditional analytical approaches that predict the Noise Vibration and Harshness (NVH) behavior of electric machines. A parametric model of a Permanent Magnet Synchronous Machine (PMSM) is utilized to calculate the structural behavior a thousand designs using analytical and Finite Element (FE) methods. Then four ML algorithms are investigated to use this data and to translate the accuracy of the latter into the analytical approach, taking natural frequencies and unitary force responses as objectives. The most promising regression model developed manages to capture more accurately the main resonance peaks and the overall unitary force response amplitudes.