Ho-Nien Shou
Dept. of Aviation & Communication Electronics, Air Force Institute of Technology, Gangshan, Kaohsiung, Taiwan (R.O.C.)
pp. 83–86
ABSTRACT
This paper proposes a new robust attitude control architecture for microsatellites. Based on deep learning fault detection method, Cerebellar Model Articulation Controller (CMAC) is used as fault-tolerant control. Using the image recognition function of Generation Adversarial Networks (GAN), the microsatellite actuator fault wavelet spectrum is used as the basis of training generator and discriminator for real-time fault diagnosis and classification. When the system fault diagnosis determines that the fault occurs, the cerebellar neural network participates in the fault-tolerant control. Using the Gan learning ability of generating confrontation network, the problems of insufficient sample data and insufficient sample labeling are solved respectively. As a kind of local learning network, CMAC has the advantages of strong generalization ability, fast convergence speed and simple hardware and software implementation. The simulation results show that, compared with the traditional methods, the fault detection and fault-tolerant control of GAN method combined with CMAC has higher accuracy and robustness.

ARTICLE INFO
Article History
Received 30 October 2020
Accepted 22 June 2021

Keywords
Deep learning
Fault-detection
Fault-tolerant control
Cerebellar model articulation controller
rGeneration adversarial networks
Wavelet

JAALR2207

Download article(PDF)