Rolling bearing is one of the components with the high fault rate for rotating machinery. Big data-based deep learning is a hot topic in the field of bearing fault diagnosis. However, it is difficult to obtain the big actual data, which leads to a low accuracy of bearing fault diagnosis. WGAN-based data expansion approach is discussed in this paper. Firstly, the vibration signal is converted into the gray texture image by LBP to build the original data set. The small original data set is used to generate the new big data set by WGAN with GP. In order to verify its effectiveness, MMD is used for the expansion evaluation, and then the effect of the newly generated data on the original data expansion in different proportions is verified by CNN. The test results show that WGAN-GP data expansion approach can generate the high-quality samples, and CNN-based classification accuracy increases from 92.5% to 97.5% before and after the data expansion.

Ferrography analysis is one of main means to identify wear state of mechanical equipment, and its key is the intelligent recognition of wear debris ferrographic images. Ferrographic image acquisition is a complex and time-consuming work, so the direct deep learning cannot been carried out for the small tested samples. A virtual ferrographic image dataset is prepared firstly and then two-level transfer learning scheme is proposed to improve the identification rate of the tested samples based on the deep learning model trained by the virtual samples. A combined network of YOLOv3 and DarkNet53 is constructed, and the application effect of model is improved by two-level transfer learning of virtual dataset to open dataset and then open dataset to tested dataset, and the model errors before and after twice transfer learning are analyzed. The average identification accuracy of the model in the validation dataset is 86.1%, which is 44.5% higher than that without two-level transfer learning, and the average recall reaches 95.8%. The experimental results prove the proposed method have a high identification rate for the tested ferrographic images of an actual gearbox.

Rolling bearing is a key component of rotating machines, its working state directly affects the performance and safety of the whole equipment. Deep learning based on big data is a mainstream means of intelligent mechanical fault diagnosis. The key lies in enhancing fault feature and improving diagnosis accuracy. Different from the Convolution Neural Network (CNN) which relies on the convolution layer to extract the image features, the Vision Transformer (VIT) uses the multi-head attention mechanism to establish the relationship among the pixels in an image. In order to improve the accuracy of rolling bearing fault diagnosis, a new fault diagnosis method based on VIT is proposed. The vibration gray texture images to be input are divided into the patches according to the predetermined size and linearly mapped into input sequences, and the global image information is integrated through the self-attention mechanism to realize fault diagnosis. In order to enhance the expressiveness and generalization ability, the pooling layer is introduced into VIT. The tested results show that the fault diagnosis accuracy of VIT on the test set reaches 94.6%, and the corresponding classification indexes top-1 is 84.2% and top-5 is 95.0%. The accuracy of the new Pooling Vision Transformer (PIT) is 3.3% higher than that of the original VIT, which proves that the introduction to pooling layer can improve the image identification performance of VIT.