31 / 2021-06-22 16:44:11

Intelligent Fault Diagnosis Based on Adversarial Domain Adaptive Network

With the wide application of deep learning in the field of fault diagnosis, a good model can be obtained through the training of deep network based on a large number of labeled data, which can correctly classify faults. However, the good diagnostic performance of the model mainly depends on the following two conditions: first, the training and testing data are from the same probability distribution, that is, the fault data are collected under the same working conditions; Second, the data needs to have a large number of available labeled data. However, it is difficult to meet the above two conditions in practical engineering. Therefore, in order to solve the above two problems, this paper proposes a new unsupervised domain adaptive method—Adversarial domain adaptive network; The network use the decision boundaries of fault diagnosis tasks to align the distribution of source domain and target domain. Target samples far from the source domain are detected by maximizing the output difference between the two classifier networks. The feature generator generates the corresponding target domain features near the source domain to minimize the difference. The generator and two classifiers are trained against each other so that they has a good classification effect in both target and source domains. Finally, this paper verifies on two datasets, and the diagnostic effect of this method in the target domain has been greatly improved, and this method is better than other domain adaptive diagnostic methods.