Convolutional Physics- Informed Neural Networks for Fast Prediction of Core Losses in Axisymmetric Transformers

We adapt a convolutional approach on Physics-Informed Neural Networks (PINNs) for solving the magnetostatic Maxwell's Equation on parametric axisymmetric transformer geometries. The trained PINN is capable of solving the physical equations in a matter of milliseconds for arbitrary geometries constrained by a total of 18 degrees of freedom (DoF). The results are analysed with respect to a numerical reference solution based on derived inductances and squared magnetic flux densities as an indication for resulting core losses. The combination of inference speed and accuracy enables new orders of magnitude for the optimization of transformer designs going forward.