Neural-network-based model for dynamic hysteresis in the magnetostriction of electrical steel under sinusoidal magnetisation

A model is presented to calculate the dynamic hysteresis behaviour of the magnetostriction in electrical steel under 1-D sinusoidal magnetisation. The input of the model is one period of the 1-D induction wave. On this induction wave, a fast Fourier transform (FFT) is performed, of which the frequency components are fed to the input of a neural network (NN).This neural network calculates the frequency components (amplitude and phase) of the magnetostriction, on which an inverse fast Fourier transform (IFFT) is performed, thus obtaining the magnetostriction wave complementary to the input induction wave. The filtering technique was used to model the magnetostriction of a grain oriented electrical steel. The amplitude of the induction in the sample ranged between 0.6T and 1.8T and the frequency ranged between quasi-static and 200Hz.