Abstract: Rolling bearing is one of the main components of rotating machinery, timely and accurate fault diagnosis plays an important role in the reliability and safety of modern industrial systems. However, most of the existing fault diagnosis methods are based on balanced datasets. Aiming at the problem of poor fault diagnosis ability and generalization ability, which are caused by class-imbalance of bearing data. A fault diagnosis model based on secondary data enhancement and deep convolution is proposed. In the first place, different datasets are constructed to study the impact of class-imbalance on the performance of fault diagnosis. Secondly, the datasets are reconstructed into balanced datasets based on resampling, and data enhancement are performed on the reconstructed datasets again, which can improve the utilization of sample points. Moreover, the deep one-dimensional convolutional network(1-DCNN) is used to extract the signal features and fully characterize the fault information of rolling bearing. Finally, the fault diagnosis is performed through the ensemble learning voting method. The experiments use t-SNE and a variety of indicators to evaluate results. At the same time, it is compared with other methods for verification. The results show that the proposed model has higher diagnostic accuracy and diagnostic speed, which has good robustness and versatility, and can be well applied to fault diagnosis of rolling bearing under class-imbalance scenes.

Key words: fault diagnosis, class-imbalance, resample, data enhancement, voting classification algorithm