Abstract: Aiming at the problems of difficult manual extraction of motor rolling bearing signal features and poor fault classification effect, an improved GoogLeNet convolutional neural network is proposed by combining the traditional GoogLeNet model unit and the dense connection idea. The proposed model is applied to the fault diagnosis test of motor rolling bearings. After grouping and labeling the original data, it is directly input into the improved model for training, and finally the test set is input into the trained model to test its classification accuracy rate. The entire diagnosis process does not require manual feature extraction, which avoids the difficulties and errors caused by manual fault extraction, greatly simplifing the fault identification process, and proving the feasibility of the improved GoogLeNet model in fault diagnosis. Finally, the proposed model is compared with the traditional GoogLeNet model and other typical models. The comparison results show that the improved GoogLeNet convolutional neural network model has the characteristics of higher accuracy, strong feature extraction ability, fast convergence speed, and stable performance than the traditional GoogLeNet model and other comparison models. 

Key words: deep learning,  , fault diagnosis of motor rolling bearing,  , convolutional neural networks ( CNN),  , GoogLeNet,  , dense connection