Novel Approach to Model Order Reduction for Nonlinear Eddy-Current Problems

Abstract

A novel model order reduction approach is proposed for nonlinear eddy-current problems in which the B-H curves are approximated by neural networks. Such approach stems from rewriting the eddy-current equations in an equivalent way in which only quadratic nonlinearities occur and allows to directly construct compact models by projection. The numerical results show that, using such compact models, the whole space-time distribution of the electromagnetic field can be accurately approximated at low computational cost.