학술논문
DCDAN-Based Incipient Fault Diagnosis for Satellite ACS Under Variable Operating Conditions
Document Type
Periodical
Source
IEEE Transactions on Industrial Informatics IEEE Trans. Ind. Inf. Industrial Informatics, IEEE Transactions on. 20(3):3115-3123 Mar, 2024
Subject
Language
ISSN
1551-3203
1941-0050
1941-0050
Abstract
This article proposes a new distributed–collaborative domain adversarial network (DCDAN)-based incipient fault diagnosis method for satellite attitude control system under variable operating conditions. The designed DCDAN contains a new distributed domain classifier and collaborative domain classifier to provide the features of incipient faults for fault classifier. In the distributed domain classifier, the designed relative importance weight between the global distribution of all data and the conditional distribution of different faults can be adaptively adjusted by the contributions of different distributions. For the collaborative domain classifier, the weight constraint factor is introduced to deal with the loss of incipient fault information in the process of network forward propagation as the increasing of network layers. The experiments in a ground semiphysical platform are carried out, and the results show that the solution achieves over 95% accuracy for incipient faults and over 98% accuracy for the total test samples.