To address the challenges of difficulty in extracting fault features and low diagnostic accuracy for intermediate bearings in aero-engines, a CWT-ECA-ResNet model was proposed for fault diagnosis. First, the Morlet wavelet was used to decompose signals into wavelet components of different frequencies for fault feature extraction. Second, an ECA attention mechanism was inserted into the ResNet network, and image feature data was extracted via the ReLu algorithm to establish the CWT-ECA-ResNet convolutional neural network model. Finally, the proposed model was trained and validated using datasets from the Case Western Reserve University and data collected from a self-built test bench. Comparative analysis with models such as CNN, VGG, and ResNet was conducted. The results demonstrated that the proposed model achieves an accuracy of 98.6% on the Case Western Reserve University validation set and 95.73% on the self-built intermediate bearing test bench validation set, both surpassing the traditional ResNet convolution algorithm.