Neural network model of magnetic hysteresis

The classical Preisach model and its modifications are one of the most generally applied simulations to model the behavior of magnetic materials, to describe hysteresis phenomena and different properties, as noncongruent minor loops, frequency dependence, temperature dependence, accommodation, and so on. Artificial neural networks (NNs) are widely used in fields of research where the solution of problems with conventional methods on traditional computers is very difficult to work out, for example system identification, modeling and function approximation. NNs can be considered as universal approximation for functions based on the theorem of Kolmogorov‐Arnold. In this paper a new NN model of scalar hysteresis characteristics is introduced. The examined method is built on the function approximation and continuous interpolation capability of NNs. The anhysteretic magnetization curve and a set of the ascending and a set of the descending first order reversal branches can be stored in a system of three neural networks. Different properties of magnetic materials can be simulated by a simple knowledge‐based algorithm. Value of differential susceptibility can be expressed in analytical form. Finally hysteresis characteristics predicted by the introduced model are compared with the results of the Preisach simulation.