A fast hysteresis analysis based on neural network and play model

Abstract

This paper presents a fast hysteresis loss analysis method based on neural network, which predicts the hysteresis loss from the waveform of the magnetic flux density in a magnetic core. The hysteresis loss data used to train the neural network is computed using the play model which can accurately account for hysteresis characteristics such as minor loops and DC bias. The proposed method can perform loss analysis 2,000 times faster than conventional methods. The proposed method is highly effective when harmonic components are superimposed on the input waveforms and when loss analysis is performed for a large number of different waveforms.