Abstract: The mechanical defect of gas insulated metal enclosed switchgear (GIS) equipment has become an important hidden danger of power grid security. Aiming at the insufficient accuracy of existing defect diagnosis methods due to limited feature information, in combination with deep learning theory, we proposed a GIS mechanical defect diagnosis method based on multi-modal combined vibration images and stacked sparse autoencoder. Firstly, the modal spectrum component of GIS original vibration signal was obtained by using variational mode decomposition algorithm, and the multi-modal combined vibration information image was constructed. Then, the support vector machine algorithm was used to construct a load classification model, and a double-layer stacked sparse autoencoder (DC-SSAE) was proposed to establish mechanical defect identification and severity assessment model under a large range current. Finally, based on the 550 kV GIS equipment mechanical defect test platform, vibration simulation tests under different currents were carried out to verify the effectiveness of the method. The results show that the feature representation of multi-modal combined vibration image is better than the traditional image, and the diagnostic model can fully mine the image information, overcoming the subjectivity of the traditional machine learning algorithm feature selection. The DC-SSAE model combined with load classification and defect matching can effectively diagnose GIS mechanical defects, and the overall accuracy of defect identification and severity evaluation is 99.38% and 99.44%, respectively. The method proposed in this paper has a good defect diagnosis effect, which can provide strong technical support for the safe and stable operation of GIS.