Abstract: In order to make the bearing fault diagnosis more accurate and intelligent, aiming at the problem of bearing fault feature extraction, a CNN(Convolutional Neural Network) based on residual structure and Inception structure is constructed, and a new bearing fault diagnosis method is proposed. First, the STFT ( Short-Time Fourier Transform) is used to transform the original one-dimensional signal of the rolling bearing into a two-dimensional time-frequency graph, which is divided into a training set, a validation set and a training set. Then the training set is used to iterate the built Inception-residual network model. The network parameters are constantly updated, and the verification set is used to check whether the model has over-fitting phenomenon. Finally, the trained model is applied to the test set, and the diagnosis result is output through the classifier of the output layer.The experiments proved the feasibility of the proposed method, and the average accuracy of bearing fault classification reached 99. 98% +-0. 02% , which has a higher accuracy and stability than other methods.

Key words: fault diagnosis； , convolutional neural network ( CNN )； , short-time Fourier transform； , Inception-residual