Innovative Bearing Fault Diagnosis Method: Combining Swin Transformer Deep Learning and Acoustic Emission Technology

Abstract

Wind power generation, as a paragon of clean energy, places great importance on the reliability of its equipment. Bearings, in particular, as the core components of wind turbines, have a direct correlation with the stable operation and economic benefits of the entire system. Against this backdrop, addressing the core challenges in the field of bearing fault diagnosis, an innovative fault diagnosis method has been proposed. For the first time, the Swin Transformer deep learning model is combined with acoustic emission (AE) technology, and through advanced signal processing techniques, bearing signals are transformed into filter banks (FBank) feature inputs for the model, effectively achieving precise fault detection in low-speed, heavy-load bearings. With extensive validation on laboratory data of low-speed, heavy-load bearings and the Case Western Reserve University (CWRU) bearing dataset, this method has achieved significant results in identifying four main damage categories. In-depth comparative analysis shows that (1) the improved Swin Transformer achieved an accuracy of 98.6% on the acoustic emission signal laboratory dataset, performing well under data imbalance conditions. (2) It achieved an accuracy of 95.63% on the vibration signal CWRU dataset, demonstrating good generalization capabilities.