Physics Informed Neural Networks for Electromagnetic Analysis

• Published in: IEEE transactions on magnetics Vol. 58; no. 9; pp. 1 - 4
• Main Authors: Khan, Arbaaz, Lowther, David A.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.09.2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; Boundary conditions; Computational electromagnetics; Deep learning; Electrostatics; Feasibility studies; Finite element method; Magnetic domains; Magnetism; Magnetostatics; Mathematical analysis; Mathematical models; Neural networks; neural networks (NNs); numerical analysis; Partial differential equations; partial differential equations (PDEs); Training
• ISSN: 0018-9464; 1941-0069
• DOI: 10.1109/TMAG.2022.3161814

Deep learning has achieved remarkable success in diverse applications; however, its use in solving partial differential equations (PDEs) has emerged only recently. Here, we present a feasibility study of applying physics-informed deep learning methods for solving PDEs related to the physical laws of electromagnetics. The methodology uses automatic differentiation, and the loss function is formulated based on the underlying PDE and boundary conditions. The feasibility of the method is shown using three electromagnetic problems of varying complexity and the results show close agreement with the ground truth from a finite-element analysis solver. The application of transfer learning is also explored and results in faster training. Furthermore, a hybrid approach involving physics-based governing equations and labeled data is also introduced to improve the accuracy of the results.