The high correlation between single fault and composite fault samples, resulting in misclassification. Moreover, rotating machinery often works at different speeds, which further increases the difficulty of composite fault diagnosis of rotating machinery. Aiming at the above problems, a composite fault diagnosis method of rotating machinery at different speeds with one-dimensional depth subdomain adaptation was proposed. Firstly, frequency domain signals of composite faults of rotating machinery are used as the input of the network to get rid of the dependence on signal processing and professional knowledge; Secondly, a domain shared onedimensional convolutional neural network was built to learn the frequency domain signal characteristics of composite faults of rotating machinery at different speeds; Then, the local maximum mean difference is added to form the sub-domain adaptation layer, which aligns each pair of sub-domain distribution to avoid the feature mixing of single fault and compound fault, and reduces the feature distribution difference of the two subdomains by minimizing the local maximum mean difference to reduce the interference caused by different speeds. Finally, softmax classification layer is added after the sub-domain adaptation layer to realize fault state identification of the target data. The effectiveness of the proposed method is proved by the composite fault diagnosis experiments of rotating machinery at different speeds.