Research on demagnetization fault diagnosis of permanent magnet flux-switching linear motors based on few-shot learning

Abstract

Aiming at the problems of few samples of local demagnetization faults and high difficulty in air-gap magnetic field detection for permanent magnet flux-switching linear motors (PMFSLMs), this paper adopts the primary backplate magnetic flux leakage signal of PMFSLMs to characterize demagnetization fault, which effectively avoids air-gap magnetic field detection. A few-shot learning method based on a Siamese convolutional neural network (CNN) is adopted to diagnose and identify the demagnetization faults in PMFSLMs. Firstly, a finite element model (FEM) of the PMFSLM is established to analyze the electromagnetic characteristics of the PMFSLM, so as to obtain the backplate magnetic flux leakage data under different demagnetization conditions. Then, a dual-channel dataset of backplate magnetic flux leakage signals is constructed and thus the signals of magnetic flux density difference are reconstructed under different demagnetization faults. Finally, a few-shot learning method based on a Siamese CNN is used to establish a classification model for identifying different demagnetization faults. The finite element simulation results show that the proposed method can accurately identify different demagnetization locations and demagnetization combinations of the PMFSLM, and exhibits high accuracy under limited-sample conditions.