Magnetics Design Optimization for LLC Converter employing Machine Learning

This paper proposes an intelligent and machine-learning based optimization method that targets to optimal windings layer setup for LLC converter transformer with small number of optimizing iterations. The research utilizes Bayesian optimization (BO) to guide designers toward winding layers setup that achieves the maximum efficiency at nominal operating point of LLC converters. Furthermore, a new multiphysics simulation platform is proposed for feeding efficiencies to optimizing iterations. This simulation platform is able to provide entire losses estimation, high simulation accuracy is verified by measurement. An LLC converter prototype with 33V input and 400V/250W output is implemented to demonstrate effectiveness of proposed optimization method.