Using neural networks in the identification of Preisach-type hysteresis models

The identification process of the classical Preisach-type hysteresis model reduces to the determination of the weight function of elementary hysteresis operators upon which the model is built. It is well known that the classical Preisach model can exactly represent hysteretic nonlinearities which exhibit wiping-out and congruency properties. In that case, the model identification can be analytically and systematically accomplished by using first-order reversal curves. If the congruency property is not exactly valid, the Preisach model can only be used as an approximation. It is possible to improve the model accuracy in this situation by incorporating more appropriate experimental data during the identification stage. However, performing this process using the traditional systematic techniques becomes almost impossible. In this paper, the machinery of neural networks is proposed as a tool to accomplish this identification task. The suggested identification approach has been numerically implemented and carried out for a magnetic tape sample that does not possess the congruency property. A comparison between measured data and model predictions suggests that the proposed identification approach yields more accurate results