Intelligent Fault Diagnosis Method of Rolling Bearing Based on Active Learning Under Unbalanced Samples

For the problem of low fault diagnosis accuracy of rolling bearings under unbalanced samples, we propose an intelligent fault diagnosis method of rolling bearings based on active learning. Firstly, data features are extracted from rolling bearing signals in both the time domain and time frequency domain according to the running sample to realize the feature representation of sample. Secondly, a Gaussian mixture model undersampling method with sample characteristics is proposed, and the weight coefficient of the model is obtained through training to realize the same distribution extraction of samples in multiple types, and solve the problem of sample imbalance. Thirdly, an active learning classifier is established based on density peak clustering method, which is then utilized for fault diagnosis of unlabelled samples. Finally, the proposed method is validated by the simulation test sample of rolling bearing fault. The results show that, compared with the comparison methods, the proposed method can effectively improve the problem of low fault diagnosis accuracy caused by sample imbalance.