Neural Network as Datasheet: Modeling B-H Loops of Power Magnetics with Sequence-to-Sequence LSTM Encoder-Decoder Architecture
Diego Serrano, Haoran Li, Thomas Guillod, Shukai Wang, Min Luo, Charles R. Sullivan, Minjie Chen
Abstract
This paper presents the concept of “Neural Network as Datasheet” for B- H loop modeling of power magnetics with sequence-to-sequence machine learning. Instead of directly pre-senting the measured characteristics of magnetic core materials, we employ a neural network to capture the B-H loop mapping relationships of magnetic materials under different excitation waveforms at different temperatures. The training and inference process of the neural network are fully automated to minimize the impact of human error. Neural networks are also effective in compressing the information contained in the raw database to avoid data search or interpolation. The neural network can be used to rapidly predict B- H loops under different operating conditions and support circuit simulations. Based on a recently developed large-scale magnetic core loss database - MagNet - we demonstrate that a neural network datasheet can effectively compress and release information about power magnetics and can play important roles in power electronics converter design.