An Implementation of Dowell model with Neural Network to Foil Winding Transformer

Abstract

This research investigates the accuracy of Dowell model (DM) and builds an accurate and practically useful semi-analytical method for leakage inductance modeling of foil winding transformers. As a widely used model for AC resistance and leakage inductance, DM is well known for its high applicability and high accuracy compared to the other modeling methods. However, it is found that transformers with certain geometrical conditions are used in most works of literature for implementing DM. Although some analyses about the modeling accuracy were conducted, the applicability of DM for some geometry is still unclear. Therefore, in this research, the accuracy of the modeling is analyzed, focusing on the frequency and geometry dependency of foil winding. Some results show the modeling error of DM has frequency dependency. Furthermore, modeling using DM with Artificial Neural Network (ANN) is implemented to achieve more practically usable modeling. As a result, the utility could be proved with accurate modeling results of the transformer samples. Some advantages are also discussed to show more possibility of developing this method.