High Frequency Transformer Design for Specific Static Magnetising and Leakage Inductances Using Combination of Multi-Layer Perceptron Neural Networks and FEM Simulations

Power industry is evolving toward high density power and therefore high frequency applications. Converters with low switching losses based on soft switching have been developed. In many of these converters, the transformers that are used, should have tuned magnetizing and leakage inductances. Hence, more accurate design methodologies should be developed. In this paper air-gap, winding overlap coefficient and core geometry are used for designing high frequency transformer with desired magnetizing and leakage inductances. Core geometry is a concept that is used by manufactures to select cores based on their power-handling capability. In this paper, instead of using complex analytical methods or time consuming and expensive FEM simulations, a methodology is proposed that uses few FEM simulations and multi-layer perceptron neural networks for designing specific magnetizing and leakage inductances, for a wide selection of cores. Moreover, the effect of core geometry on magnetizing and leakage inductances is explored. Detailed 3D-FEM simulations confirm the validity of proposed methodology. In addition, a prototype was tested and experimental results are consistent with Neural-network results.