A Decision Fusion SWT-RF Method for Rolling Bearing Enhanced Diagnosis Under Low-Quality Data

Low-quality data, including insufficient samples and low signal-to-noise ratio (SNR), restrict the effective application of intelligent diagnostic methods based on deep learning. In this study, a new rolling bearing enhanced diagnostic method is proposed based on swin-transformer (SWT) and ReliefF (RF) combined with Dempster-Shafer (DS) evidence theory. First, SWT is improved for adaptive deep feature mining of the vibration signal. Meanwhile, to enhance the quality of features, RF is introduced to optimize the deep fault features output by the global average pooling layer, which helps improve the classifier performance. Then, the DS evidence theory-based decision fusion strategy is designed to realize the fusion of different axial signals at the decision level, which enhances the fault knowledge threshold and further improves the diagnostic ability. Finally, the bearing cases with data collected from the accelerated life degradation and different distributions are studied. The results reveal that the proposed method can adaptively mine fault features with low-quality data and realize efficient enhanced diagnosis.