Fast Three-dimensional Optimization of Magnetic Cores for Loss and Volume Reduction.

Power electronic devices have been widely applied such as in consumer devices and electrical vehicles, and improvement of their efficiency is strongly required. One of the main factors of energy loss in the devices is the magnetic core loss in inductors and transformers. Therefore, simulation technology that realizes accurate energy loss estimation based on magnetic characteristics of the materials has been highly demanded. For this reason, we have been developing large-scale magnetic field simulation technology by combining micromagnetic models and the finite element method (FEM) [1]. One the other hand, structural design optimization for devices with complicated shapes requires huge man-hours, and it is difficult to judge whether the designed shape is sufficiently optimized. Genetic algorithm (GA) is widely used in various optimal design problems for its excellent global search capabilities. However, GA generally requires large numbers of simulations, and it takes a long time to search the optimal solution. To solve this problem, neural networks (NN) are applied as surrogate models instead of physics models to reduce computing time in some previous work [2]. In this work, we perform shape parameter optimization of an EI-shaped inductor using GA and NN surrogate models designed from magnetic simulation results. Magnetic loss and volume of the cores are chosen to be objective functions. Few previous works have been reported on optimization studies with magnetic core loss as an objective function [3]. Here we solve multi-objective optimization problems, and Pareto optimal fronts for various inductance values are obtained. We also show that experiment that employs one of the optimal design parameters presents good agreement within 10% with the simulation results.