Abstract: In the process of wind power generation, the normal operation of the bearing is related to the normal operation of the wind turbine. Aiming at the problem that the existing deep learning-based bearing fault diagnosis model has a complex structure and many parameters, which makes it difficult to train the model, an improved one-dimensional convolutional neural network rolling bearing fault diagnosis method based on the LeNet-5 model is proposed. First, in order to extract fault information to a greater extent, a short-time Fourier transform is introduced to preprocess the original vibration signal. Secondly, a one-dimensional network model is designed, which has a larger receptive field and faster calculation speed. At the same time, the Leaky-ReLU activation function is introduced to make the ability to process the details of the input signal stronger. The batch normalization layer and Dropout layer are added to improve the model generalization ability. Finally, the trained model is used to perform fault diagnosis experiments. The experimental results show that the diagnostic accuracy of this method can reach 99.98% in the classification of ten types of bearing faults, which has a good engineering application prospect for the fault diagnosis of wind turbine bearings.

Key words: wind turbine, rolling bearing, fault diagnosis, convolutional neural networks, short-time fourier transform, LeNet-5