A dual-experience pool deep reinforcement learning method and its application in fault diagnosis of rolling bearing with unbalanced data

A dual-experience pool deep reinforcement learning (DEPDRL) model is proposed for rolling bearing fault diagnosis with unbalanced data. In this method, a dual-experience pool structure is designed to store the sample data of majority and minority classes. A parallel double residual network model is established to extract deep features of the majority and minority input samples, respectively. In the process of training, the proposed balanced cross-sampling technique is used to randomly select samples from dual-experience pool in a certain proportion to realize the training of a double residual network model. We show the effectiveness of our method on three standard data sets, and compared with Resnet18, DCNN, DQN and DQNimb methods, the results show that DEPDRL has the best performance. Finally, with wavelet time-frequency graph as input, DEPDRL is applied to rolling bearing fault diagnosis with unbalanced test data. The results show that on a variety of unbalanced data sets, both the diagnostic accuracy and the G-means value of the DEPDRL are more than 5 % higher than other algorithms, which fully indicates that the DEPDRL has a very high fault diagnosis ability of rolling bearing with unbalanced data.