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

Summary form only given. Feed-forward neural networks have been successfully applied to model quasi-static and dynamic unidirectional magnetization, as well as quasi-static vector magnetization for circular and elliptical magnetization patterns in electrical steel sheets. The magnetic state of the material is thereby the input of the neural network. In the case of quasi-static circular and elliptical magnetization, the magnetic state is determined by the value of the magnetic induction vector B(t) (amplitude and phase), together with the maximum amplitude and axis ratio of the considered magnetization pattern. Dynamic hysteresis is treated based on the loss separation property of magnetic materials. The dynamic field can be determined the same way as in the case of unidirectional excitation, with a neural network. Consequently, only the unidirectional sinusoidal loops for saturation along the x- and the y- axis for different frequencies need to be measured. This approach reduces the required amount of measurement data substantially, while retaining good accuracy for the particular case of low and moderate induction levels. Standard neural network techniques are used. The numerical investigation of the accuracy of the proposed approach is described.