Real-Time Magnetic Field and Performance Prediction of EV Motors Based on Physics-Informed Machine Learning

Accurate magnetic field analysis is essential for predicting motor performance. However, the complex operating conditions combined with core material nonlinearity due to magnetic saturation present significant computational challenges for obtaining both fast and precise field solutions. To address these challenges, this article proposes a physics-informed neural network (PINN) based reduced-order model for rapid motor performance calculation. First, the proper orthogonal decomposition (POD) method is employed to reduce the dimensionality of the magnetic field matrix extracted from the finite-element method. Subsequently, a neural network (NN) is trained to learn the matrix decomposition results across various operating points. To enhance accuracy, the proposed approach integrates the motor mechanism with a data-driven learning framework, incorporating a physics-guided loss function into the NN training process. This method enables the rapid computation of magnetic field characteristics for any motor operating point, facilitating dynamic simulation and real-time performance prediction. The computed magnetic field results demonstrate strong agreement with those obtained from the finite-element method. For further verification, a prototype motor has been manufactured. Experimental measurements of the back electromotive force (EMF) curves from the search coils (SCs) are compared with the calculated results, confirming the effectiveness and accuracy of the proposed method.