基于无监督特征表示深度Q学习的智能故障诊断方法

Aiming at the problem of low accuracy and weak generalization of rail transit rolling bearing intelligent fault diagnosis under the condition of seriously insufficient samples, a fault diagnosis method based on unsupervised feature representation and deep Q-learning was proposed. Firstly, the training set was expanded by data augmentation algorithm, and the expanded data set was constructed without parameters. Secondly, through unsupervised comparison of feature representation, the deep reinforcement learning ResNet was pre trained to optimize the feature representation ability of the network, so as to improve the problem of long training time of deep reinforcement learning. Finally, the fault classification was realized by using the pre-trained ResNet and deep reinforcement learning algorithm. The results show that compared with the existing intelligent diagnosis algorithms, the average classification accuracy of the proposed method is significantly improved. The proposed method can effectively extract signal features and accurately identify fault types. In addition, the establishment of experience replay function can eliminate the correlation between samples, and make up for the impact of the serious shortage of training samples on the diagnosis results.