Power loss and permeability prediction, sensitivity analysis on toroidal transformer cores using artificial neural networks

Abstract

In this investigation a multi-layer perception with a feed-forward neural network model was used. The input parameters included the outer and inner diameters of the toroidal core, the strip width and thickness o electrical steel, the induction frequency and the peak magnetic flux density. The output parameters were power loss and permeability. Experimental data were collated different combination of core dimensions over 179 samples. A total of 3451 input vectors were available in the training set. The best output results were obtained for models formed by tanh+sig and sig only functions for power loss and permeability respectively. A self-organising feature map neural network model was also formed for sensitivity analysis. The proposed model was in good agreement with experimental data and can be used for estimation of power loss. © 2007 American Institute of Physics.