With the increasing complexity of the modern engineering environment, diagnosis-bearing fault under the changeable engineering condition is of great significance to managing the equipment’s health state. Therefore, to solve the traditional method that is difficult to extract bearing fault features and lower diagnostic accuracy accurately, this paper presents a bearing fault diagnosis method based on a deep convolutional neural network. Firstly, the original data are pre-processed by data enhancement. The bearing fault features are extracted by alternately superimposed convolution layer and pooling layer, which enhances the nonlinear expression ability of the model and enlarges the range of high and low-frequency features captured by the model. Finally, based on fault feature extraction, bearing fault types are classified by using the softmax function. The validity of the method is verified by the Case Western Reserve University experimental platform’s fault data. The experimental results show that the proposed method’s classification accuracy in the standard bearing fault diagnosis data set of CWRU is over 99.6%, which is better than that of the Long Short-Term Memory(LSTM) neural network and other traditional classifiers.