Electromagnetic Field Model of Tubular Permanent Magnet Synchronous Linear Motor Based on Deep Transfer Neural Network

During high-speed operation, tubular permanent magnet linear motors with slotted stators often experience large fluctuations in thrust, overheating of the coils and irreversible demagnetization of the permanent magnets. For the first two cases, this paper proposes a deep transfer neural network (DTNN)-based electromagnetic field model for TPMLM, the specific implementation steps of which include: (1) Using different geometric parameters of TPMLM as input and average thrust, thrust fluctuation and coil copper loss as output, finite element analysis (FEA) and analytical method (AM) are used to provide the electromagnetic parameter dataset of the motor respectively, and these two datasets are used as the source and target domains for transfer learning; (2) Based on the features of the sample dataset, the source tasks corresponding to the source domain are pre-trained using DTNN, and the target tasks corresponding to the target domain are fitted by fine-tuning the model parameters. Then, in order to verify the accuracy of the proposed model, this paper compares the output prediction results with some non-parametric models, such as Random Forest (RF), Support Vector Machine (SVM), and Deep Neural Network (DNN), by dividing the target domain dataset with different scales. The results show that the best prediction results are obtained from the DTNN model, which fully combines the accuracy of FEA and the efficiency of AM, and shows better generalization ability in the case of insufficient real training data.