Fault Diagnosis of Linear Synchronous Motor Based on Time-Frequency Representation and CNN

Long-stator Linear Synchronous Motor (LSLSM) is the core component of maglev trains. But the strong background noise generated by its complex operating environment often hides the stator winding damage characteristics, which makes it difficult to detect the stator turn-to-turn insulation damage status in time. This paper presents a method based on Iterative Adaptive Multiple Synchronous Compression of Transform (IAMST) and Deep Residual Convolutional Neural Network (DRS-CNN) for LSLSM stator winding fault detection. Firstly, the coupling model of a normal-conducting high-speed magnetic levitation machine-electricity-magnetism is constructed, and the accurate diagnosis of faults is realized through the analysis of multi-physical field coupling. Second, IAMST is utilized for energy aggregation to enhance the time-frequency (TF) resolution, and a TF representation is constructed to facilitate feature representation. Finally, DRS-CNN combines deep residual learning with Channel Attention Mechanism (ECA), which is used to further mine advanced features to identify the degree of turn-to-turn short-circuit faults. The experimental results show that this method effectively overcomes the limitation of traditional time-frequency analysis methods, which have difficulty extracting fault features in a noisy environment. It has solved the problems such as poor accuracy caused by directly adding signals to the network, the increase in the number of network layers, and the poor interpretability. It can accurately identify the severity of different short-circuit faults, providing an efficient and reliable solution for the detection of stator winding faults in the LSLMS in maglev trains.