부산대학교 도서관

This paper proposes an Artificial Neural Network (ANN) as a method to thermally analyze a three-phase medium frequency transformer (MFT). This transformer is part of a 50kW three-phase Dual Active Bridge (DAB). After choosing the most suitable architecture for the ANN, it is trained with the results of finite element method (FEM) simulations that model its behavior. The ANN is thus able to calculate temperature increments of the core and of each winding with high reliability. This performance of the ANN is studied by comparing its results with some other from widely used state-of-the-art theoretical thermal models that emulate the transformer’s thermal behavior. The neural network-based solution is proved to be as accurate as the FEM simulations ground truth, and as fast to apply as the theoretical thermal model, therefore fusing the advantage of using any of these two in power converter design optimization.