Physics-Informed Neural Networks for Parametric Modeling of Permanent Magnet Synchronous Machines

The objective of this paper is to develop a physics-informed machine learning methodology for parametric modeling of permanent magnet synchronous machines (PMSMs). A deep neural network is trained to compute the magnetic field as a function of spatial coordinates and machine parameters, while enforcing physical properties such as Dirichlet boundary conditions and periodicities. Leveraging a DeepONet architecture, the network is trained in a data-free fashion by minimizing a physics-informed functional using a mesh-based coenergy evaluation. The methodology is demonstrated on a 15-dimensional PMSM problem, with model accuracy validated by comparing predictions with finite element analysis (FEA) results, focusing on coenergy, average torque, and total core loss. Computational cost is also assessed relative to FEA.