Machine Learning Approach for Magnetic Field Calculations

This paper presents a machine learning approach for 2D static and harmonic magnetic field calculations based on Poisson and Helmholtz equations for Dirichlet boundary problems. The machine learning approach is implemented on multilayer convolutional neural network trained over a benchmark problem dataset. Different neural network architectures are evaluated to estimate the efficiency in that particular application. Implementation is suitable for domains with non-homogeneous magnetic properties and distributed field sources. The accuracy is estimated in comparison with Finite Element Method model results of the same magnetic problem.