A Two - Stage Artificial Neural Network (ANN) - Based Design and Optimization of High Frequency Transformers for Dual Active Bridge Converter

With the increase in high-frequency applications such as electric vehicles, finding a transformer that can handle high frequency, high power density, and high efficiency while ensuring effective thermal management is crucial. This paper presents a novel two-stage model-based design optimization methodology for high-frequency transformers based on Artificial Neural Networks (ANN), focusing on electromagnetic (EM) and thermal modelling. This methodology accelerates the design process using FEM simulations, enabling the selection of the most suitable transformer design from 70,784 design points in 5 minutes instead of 246 days. The relative errors between FEM simulations and ANN models for 99.7% of the dataset are less than 2.5% for EM modelling and 7% for thermal modelling. First, the methodology based on FEM simulations was verified experimentally with a 100 kHz Litz wire E core transformer and a 75 kHz flat bar E core transformer. Then, a 95 kHz U core-based transformer for a dual active bridge converter, selected from the Pareto front using the ANN optimization method, was built, with results matching both the electromagnetic and thermal comparisons.