Physics-Informed Neural Networks for Solving Two-Dimensional Magneto-Static Fields

A physics-informed neural network (PINN) for solving two-dimensional (2D) magneto-static fields is proposed. The magnetic field intensity and the magnetic vector potential are solved using a neural network. The computation of the spatial derivatives of media constitutive parameters, which negatively impacts the training of PINN, is eliminated. The proposed PINN is verified in a 2D magneto-static case study by comparing its results with those of a finite element method. It is expected that this work will promote applications of PINN in the numerical analysis of electromagnetic devices and systems.