A cross-domain fault diagnosis method for planetary gear-boxes based on multi-channel information fusion

Cross-domain fault diagnosis of planetary gearboxes remains a significant challenge due to complex operating conditions and pronounced domain-specific distribution discrepancies. To address this issue, this study proposes a novel cross-domain fault diagnosis method based on multi-channel information fusion and domain-invariant representation learning. First, the synchrosqueezing S-transform (SSST) is employed to fuse and transform raw multi-channel vibration signals collected under varying working conditions into discriminative three-channel time–frequency representations, effectively enhancing fault-related feature expression. To mitigate domain shift, a global–local domain discrepancy metric strategy is introduced, which simultaneously measures and minimizes global distribution differences and local subdomain discrepancies, thereby promoting more effective domain confusion. Subsequently, a unified diagnostic framework is constructed based on the ResNet-50 architecture, enabling joint feature extraction and domain adaptation in an end-to-end manner. Experiments conducted on two planetary gearbox datasets demonstrate that the proposed method outperforms existing methods in terms of cross-domain diagnostic accuracy and robustness.