A Dual-Channel Decision Fusion Framework Integrating Swin Transformer and ResNet for Multi-Speed Gearbox Fault Diagnosis

In practical industrial operations, rotating machinery often operates under multiple rotational speed conditions. The vibration signals corresponding to faults exhibit significant variations across different rotational speeds, which substantially increases the complexity of fault diagnosis tasks. This paper proposes a dual-channel fault diagnosis fusion model integrating Swin Transformer and ResNet architectures. First, time-frequency representations of fault signals are generated through Continuous Wavelet Transform (CWT). Subsequently, Variational Mode Decomposition (VMD) is employed to decompose the original one-dimensional vibration signals into distinct Intrinsic Mode Functions (IMFs), from which statistical features are extracted to form feature vectors. These processed vectors are then concatenated with the original vibration signals to construct one-dimensional feature samples. Finally, the one-dimensional feature samples and wavelet time-frequency images are respectively fed into the ResNet and Swin Transformer models for training, followed by decision-level fusion of both model outputs. Experimental evaluations on the MCC5-THU dataset demonstrate that the proposed fusion model effectively extracts deep-level features from gearbox fault data samples across varying rotational speeds, successfully combining the global modeling capabilities of Swin Transformer with the local feature extraction strengths of ResNet. The integration achieves superior diagnostic performance compared to individual model implementations.