An OFSCoh-M-SSAE intelligent fault diagnosis method for rotating machinery under time-varying speed conditions

Abstract Fault diagnosis is critical in the maintenance of rotating machinery such as automobiles and wind turbines, particularly under time-varying speed conditions. However, existing fault diagnosis methods struggle to balance the ability to identify failure modes under time-varying speed conditions with robustness to speed fluctuations, and are prone to overfitting. To address this issue, an order-frequency spectral coherence stacked sparse autoencoder (OFSCoh-M-SSAE) intelligent fault diagnosis (IFD) method is proposed with the integration OFSCoh, mixup data augmentation, and SSAEs. OFSCoh is adopted to reveal the hidden angular cyclostationary behaviors in the vibration signal and construct discriminative feature maps that are insensitive to speed variations. Subsequently, the SSAE-based IFD model is established to build the mapping relation between learned features and labels, while mixup data augmentation is introduced to generate between-class samples to enhance the generalization ability of the model. Validation experiments were conducted on a bearing and a gearbox under time-varying speed conditions. The results show that the OFSCoh-M-SSAE method achieves 99.60% and 96.94% diagnosis accuracy, superior to existing methods. Furthermore, the role of mixed data augmentation in IFD models is analyzed to reveal the mechanisms by which it improves diagnostic accuracy under time-varying speed conditions.