A Neural-Network Based Model of the Magnetic Nonlinearity of a DC Electromagnet

Abstract

This paper shows how an Artificial Neural Network model (ANN) can be used to fit the nonlinear magnetic behavior of a DC electromagnet. An ANN model is trained to obtain a generalized function of the B2-?r curve, which is commonly used in an electromagnetic model. Once the generalized function and its derivative are obtained, they are used to solve a magnetostatic nonlinear problem of a DC device using the finite element method and the Newton-Raphson algorithm.