Neural-network-based approach to dynamic hysteresis for circular and elliptical magnetization in electrical steel sheet

This paper presents a neural-network-based technique for the modeling of dynamic hysteresis in the special case of circular and elliptical magnetization patterns in laminated nonoriented electrical SiFe steels. The method employs the loss separation property to allow the separate treatment of quasi-static and dynamic hysteresis effects. Measurement results show that the magnetic coupling between the rolling and the transverse direction of the sheet can be neglected when describing dynamic effects as long as the induction level along the transverse direction is not too high. According to this property, the unidirectional hysteresis loops along the rolling and the transverse direction for different frequencies suffice to approximate the dynamic effects for all considered magnetization patterns. The corresponding neural-network models are described. Comparisons between simulation and measurement results show the good accuracy of the model and validate the proposed technique.