High-Resistance Connection Fault Diagnosis of SRM Based on Multisensor Calibrated Transformer With Shifted Windows
Linghan Zhang, Juncai Song, Xiaoxian Wang, Jingfeng Lu, Siliang Lu
Abstract
A fault diagnosis and classification method based on a multisensor calibrated transformer with shifted windows (MSSwin-T) is proposed to classify the high-resistance connection (HRC) fault of a switched reluctance motor (SRM). First, an experimental platform of an SRM-based electric vehicle drive system is built. The nonintrusive acquisition method is used to collect the three-channel current signal of the SRM stator winding as the HRC fault signal. Second, MSSwin-T, a feature extraction framework with the ability to establish global dependencies, is proposed to extract features that fully contain fault information. Finally, experimental results show that the classification accuracy of the proposed method is as high as 100% in the HRC fault diagnosis of SRM. Compared with traditional deep learning and transformer-based models, namely, UniFormer, CrossViT, vision transformer (ViT), and ResNet-18, the proposed method effectively improves the accuracy of HRC fault diagnosis. The accuracy of the proposed method is improved by 1.85%, 4.23%, 7.69%, and 9.31%, respectively, which can complete the classification and identification of the corresponding fault severity. Finally, the classification experiment with added noise verifies that the proposed MSSwin-T method has good anti-noise ability and robustness, indicating that this method is effective and feasible.