Use of DL Model to Capture Noise Variance of EV Motor by Eccentricity

Abstract

In the era of mass-production of EVs, the expectations of performance and NVH performance required for power electric(PE) motors are gradually increasing. In order to satisfy the noise requirement and to prepare countermeasures, it is necessary not only to verify the nominal design value, but also to analyse the characteristics due to manufacturing variance. However, it was not easy to thoroughly examine the tilting and eccentricity issues that affect the noise characteristics of the axial type motor due to time and computational costs. To overcome this limitation, a machine learning model was developed to predict the electromagnetic(EM) excitation force of the motor eccentric condition. The training set of 2D EM FE results were generated with various eccentricity condition and RPMs. Using spatial-temporal frequency analysis and unbalanced magnetic pull analysis, the tendency of the excitation forces due to eccentricity was analysed. The distribution of electromagnetic force by time was predicted using random forest model. It showed R2 value of 0.999. The prediction model reduces the computational time within 5 minutes from 31 hours and enables the realistic analysis of tilting and eccentric condition. The 3D EM force distribution as Vibro-Acoustic FE model input could be predicted by putting the eccentricity value into the machine learning model.