Abstract: The mechanical faults of circuit breakers are common types of faults in Gas Insulated Switchgear (GIS). Monitoring and diagnosing these faults can help eliminate potential accidents in advance, ensuring the safe and stable operation of the power system. To improve the diagnostic accuracy of mechanical faults in GIS circuit breakers, this study proposes an intelligent diagnostic method based on deep learning. Four types of faults were simulated on a fault simulation test platform, and vibration feature signals were collected. The wavelet transform was applied to obtain the wavelet scale coefficient spectrograms of the feature signals, which were then fused. The fused spectrograms were further augmented using the Wasserstein Generative Adversarial Network (WGAN) to increase the sample size. The VGG16 network was employed for the diagnosis of GIS mechanical faults. The research results demonstrate that the augmented samples and the original data samples exhibit similar features for the same fault type. By using the proposed deep learning algorithm, the diagnostic accuracy of GIS mechanical faults can reach 97%. The augmented sample set obtained through data augmentation effectively addresses the issue of insufficient sample size and improves the generalization ability of the deep learning classifier.