A motor bearing fault diagnosis model based on multi-adversarial domain adaptation

In response to the weakened capability of feature transfer and parameter distribution alignment across domains due to significant differences in data distribution collected by different devices, this paper constructs a motor bearing fault diagnosis model based on multi-adversarial domain adaptation. Initially, an improved residual network is employed as the feature extraction module to enhance feature extraction capabilities. It then incorporates a Selective Kernel Network (SKNet) to implement attention mechanisms on convolutional kernels, and a Global Context Network (GCNet) to effectively utilize global contextual information for re-weighting across different channels. Additionally, the model uses a multi-kernel maximum mean discrepancy to measure the distribution between domains and classes, establishing a dynamic adjustment factor in conjunction with multiple domain discriminators to modulate the importance of marginal and conditional distributions. Ultimately, the proposed model was applied to transfer experiments across different operating conditions and devices, demonstrating excellent diagnostic results.