Journal: IEEE Transactions on Industry Applications
Authors: Ji-Hyeon Lee, Soo-Hwan Park, Du-Ha Park, Jae-Hoon Jeong, Myung-Seop Lim
DOI: 10.1109/TIA.2025.3585098
This article proposes a method for estimating the 3-D FEA-based demagnetization ratio (DR), which serves as a key measure for assessing irreversible demagnetization, using deep transfer learning. The complex configuration of the LOA, such as the segmented outer stator and permanent magnets (PMs), reduces the accuracy of 2-D axisymmetric finite element analysis (FEA). While 3-D FEA provides a more precise DR estimation, its substantial computational cost poses a significant challenge. Therefore, we propose a deep transfer learning-based demagnetization analysis method that improves computational efficiency while preserving high accuracy. This approach takes into account the permeance in the stator core and circumferential leakage flux. By leveraging deep transfer learning, knowledge acquired from a large-scale 2-D axisymmetric FEA-based DR dataset is transferred to a limited 3-D FEA-based DR dataset, effectively enhancing deep neural network performance. The DR predicted by the proposed method was compared with the results obtained from 3-D FEA. Using the proposed method, the analysis time is significantly reduced compared to employing only 3-D FEA, while maintaining high accuracy. This demonstrates its potential for accurate and efficient demagnetization analysis, positioning it as a viable solution for LOA design optimization. The proposed method is validated through computer simulations and experiments under various demagnetization conditions.