Multilayer perceptron-based inductor iron loss predictions with loss map by a measurement kit

Accurately predicting high-frequency magnetic iron losses poses a significant challenge while optimizing the design of high-frequency (HF) magnetic components and devices, such as inductors, transformers, and filters. Existing models fail to account for the complex interplay between different magnetic materials, geometry, and temperature factors, all of which have discrete and substantial effects on magnetic iron loss. In this paper, we propose a multilayer perceptron (MLP) model with techniques against overfitting for predicting iron losses based on measured data. The proposed model considers crucial parameters such as frequency, flux density, magnetic material size, as well as various materials and temperature factors. By incorporating these factors, the MLP model offers enhanced accuracy in iron loss predictions. Additionally, we introduce an automatic data measurement box specifically designed for the efficient collection of massive measurement data. These datasets serve as the training data for the proposed MLP model. Experimental validation demonstrates that the proposed model effectively predicts iron losses while considering all the aforementioned factors, thereby supporting its application in the design of HF magnetic devices in the future.