Magnetic Hysteresis Simulation by Using a Deep Neural Network for Non-sinusoidal Excitations

Abstract

Here we present an effective and performing hysteresis model, based on a deep neural network, with the ability to reproduce the evolution of the magnetization processes under arbitrary excitation waveforms. The proposed model consists of an autonomous multilayer feed-forward neural network, with input neurons reserved for the past values of both input (H) and output (M), aimed at reproducing the memorization mechanism typical of hysteretic systems. The training set was suitably prepared starting from a set of simulations, carried out using the Preisach hysteresis model. The optimized training procedure, based on multi-stage control of the model performance, will be extensively discussed. The comparative analysis between the neural network-based model, implemented at a low level of abstraction, and the Preisach model covers further hysteresis processes, different from those involved in the training, will be also presented.