Adversarial Multi-Target Domain Adaptation with Multi-Scale Convolutional Neural Networks for Cross-Domain Fault Diagnosis

Bearing fault diagnosis is vital for the smooth operation of rotating machines. In order to solve the problem of domain migration in cross-domain fault diagnosis, this paper proposes a multi-scale convolutional neural network based adaptive approach to counter multi-objective domain (MSCNN-AMDA). The traditional method is usually limited to the single source single target scenario, which is difficult to adapt to the dynamic changes of the multi-target domain in the actual industrial environment. In this paper, the model can adapt to multiple target domains at the same time by sharing the weights of multi-target domain feature extractors and combining adversarial training strategies. In addition, multi-scale convolution structure is introduced to enhance feature extraction capability and capture fault features of different scales, thus improving cross-domain generalization performance. Experiments have verified the effectiveness of the proposed method on the dataset of Case Western Reserve University (CWRU), and its average diagnostic accuracy is significantly better than that of existing methods, especially in scenarios with large domain differences. The results show that MSCNN-AMDA significantly improves the robustness and scalability of multi-target domain fault diagnosis while reducing the cost of repeated model training.