Fabrication-Specific Simulation of Mn-Zn Ferrite Core-Loss for Machine Learning-Based Surrogate Modeling With Limited Experimental Data

Mn-Zn ferrite is widely used as a core material in power electronic applications. However, core-loss modeling is challenging owing to the complexity of core-loss mechanisms and associated factors. The scarcity of experimental data is another significant impediment to the development of Mn-Zn ferrite...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on power electronics 2025-01, Vol.40 (1), p.1519-1531
Main Authors: Choi, Minwook, Park, Soyeon, Jang, Eunyoung, Ouk, Minae, Park, Kwangho, Lee, Sangwon, Noh, Gunwoo
Format: Article
Language:English
Subjects:
Accuracy
Core-loss
Data models
data-driven method
Fabrication
Ferrites
machine learning
Magnetic properties
Mathematical models
Mn-Zn ferrite
surrogate model
Transformer cores
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Mn-Zn ferrite is widely used as a core material in power electronic applications. However, core-loss modeling is challenging owing to the complexity of core-loss mechanisms and associated factors. The scarcity of experimental data is another significant impediment to the development of Mn-Zn ferrites. In this study, we propose a novel data-driven framework to construct an effective machine learning (ML) based surrogate model for estimating the core-loss of Mn-Zn ferrites. We developed a fabrication-specific finite element analysis model, simulating the experimental results of fabricated Mn-Zn ferrites, to generate simulation-driven data for expanding the training dataset. We considered various ML techniques for material property estimations in the fabrication-specific simulation and core-loss calculations. A case study using six limited experimental datasets with sample sizes of 3, 6, 8, 12, 20, and 30 showed that the proposed ML-based surrogate model can estimate core-loss with an accuracy of approximately 91.78%, a 17% increase in accuracy compared to the Steinmetz equation-based model. With the inclusion of the fabrication-specific simulation data, the accuracy of the ML-based surrogate model improved more rapidly as the amount of experimental data increased, and the models exhibited enhanced converged accuracy. Notably, the accuracy of each surrogate model was significantly enhanced for a sample size of less than 8, resulting in an improvement of approximately 35%.
ISSN:0885-8993
1941-0107
DOI:10.1109/TPEL.2024.3435366