Abstract:In order to improve the accuracy of rolling bearing fault diagnosis and avoid tedious spectrum analysis, a bearing fault diagnosis model based on GAF and GoogLeNet was proposed. The vibration signals of normal rolling bearing,inner ring failure,outer ring fault and rolling fault were collected in the laboratory,and the EMD was used to decompose the vibration signal and extract the components contributed 90% of the total; based on the overlapping sampling principle, the selected EMD component and the original vibration signal were processed into a two-dimensional picture by using the Gram algorithm, and these pictures were divided into training set, verification set and test set; the GoogLeNet model was used to learn from the training set, and the trained GoogLeNet model was used to test bearing failure samples. The results show that the bearing fault samples can be better identified by using the GoogLeNet model under the data set constructed by GAF.