Channel-Weighted Adaptive Cross-Domain Fault Diagnosis of Rolling Bearings Under Varying Operating Conditions

In practical engineering applications, single-channel information often suffers from weak directionality and limited representational capability. To address these challenges, this paper proposes a channel-weighted adaptive cross-domain method for fault diagnosis of rolling bearings under different operating conditions. Firstly, through multi-channel feature fusion, multi-channel samples of the same category are concatenated along the channel dimension to construct channel-representational data with multi-angle, multi-dimensional key features, thereby enhancing the sample's representational capability. Secondly, adaptive channel dynamic weighting mechanism is designed, which utilizes a weight generator to dynamically learn the importance coefficients of each channel. By fusing multi-channel features through weighting, the interference of high-difference channels is effectively suppressed, reducing the risk of negative transfer. Finally, to avoid the problem of loss accumulation in traditional multi-source cross-domain methods, a unified subdomain alignment loss is constructed in the manifold space of the multi-channel fused samples. With the help of A-distance, unsupervised inter-domain distribution adaptation is achieved, and the feature distributions of the source and target domains are aligned to overcome domain shift. Experiments on two sets of rolling bearing fault diagnosis tasks under different operating conditions show that the proposed method improves the average accuracy by 8.12% compared to classical multi-source transfer models, with a significant reduction in parameter count. This verifies method's advantages in structural simplicity, difference robustness, and engineering applicability.