Anisotropic vector hysteresis model applying Everett function and neural network

This paper deals with a simulation technique based on neural networks and an identification method to approximate the behavior of vector hysteresis characteristics of ferromagnetic materials. The identification procedure is based on theoretical measured vector Everett functions using Fourier expansion to deal with angle dependence of the measured scalar Everett functions and of the vector Everett functions in the 2D or in the 3D space. Computing afterwards the theoretical measured vector Everett functions for some given directions, the corresponding hysteresis models are approximated by neural networks and are used to build up the vectorial hysteresis model both in isotropic and anisotropic case. The properties of the anisotropic model has been analyzed and shown in figures. For some examples, the first order reversal curves determined from the vectorial model are compared with the corresponding measured curves that have been used to compute the measured scalar Everett functions being the input for the identification procedure.