Abstract: To address the problem of inaccurate characterization of fault features due to differences in the data distribution of rotating machinery under different working conditions, a joint adaptation method of attention to depth migration was proposed for fault diagnosis of rotating machinery under different working conditions. Firstly, the frequency domain features are input to the deep convolutional neural network in the form of a unilateral spectrum, which preserves the original signal features while reducing the input dimension of the network and effectively improves the training efficiency of the network; then, the sample features of the two domains are mined to form a distributed feature representation of the corresponding domain, and the attention mechanism of small channels is used to focus on the intrinsic connection between the feature channels formed in the two domains and the essential features of the faults in the two domains; further, the minimum mean difference Finally, the adapted classification information is integrated through the fully connected layer to achieve fault diagnosis of rotating machinery under different working conditions. The proposed method is proved to have high diagnostic accuracy and generalization capability through two sets of rotating machinery fault diagnosis tests under different working conditions.