Transfer Learning Methods for Magnetic Core Loss Modeling
Evan Dogariu, Haoran Li, Diego Serrano, Shukai Wang, Min Luo, Minjie Chen
Abstract
Transfer learning is a machine learning technique where a model developed for one task is reused as the starting point to initiate the model for another. This paper applies transfer learning to magnetic core loss modeling to reduce the amount of data needed to achieve improved performance for a variety of tasks. Leveraging a recently developed magnetic core loss dataset - MagNet - we demonstrate that a neural network trained for modeling the core losses of a certain group of magnetic materials under certain excitations can be retrained to model the core loss of other magnetic materials under similar excitations, with a reduced set of measurement data. This approach can also be applied to model the core loss of the same magnetic material under different excitations. Experiments are designed and compared to verify the effectiveness of material-to-material transfer learning and waveform-to-waveform transfer learning.