Fault detection and diagnosis of satellite ACS base on Riemannian metric improved LPP and GLS

Fault detection and diagnosis is widely recognized as a crucial task for satellite attitude control systems (ACS) within the field. Any faults within the ACS can result in degraded system performance, disruption of satellite missions, and reduced lifespan. However, traditional fault feature extraction diagnosis faces challenges in the ACS due to the nonlinear characteristics of the data and the high dimensionality of the pattern features. Furthermore, the manifold structure of the data is unknown. In order to address these issues, a novel learning method that utilizes Riemannian metric is proposed,and Generalized Least Squares(GLS) is utilized to new sample extention, these methods preserve the intrinsic structure of the data and performs fault diagnosis within its feature space. Two statistical methods, T 2 and SPE, are employed, along with Kernel density estimation (KDE) to calculate the control limit. The effectiveness of the proposed method is validated through mathematical simulation on a fault simulation system for attitude control systems, demonstrating superior diagnostic capabilities compared to conventional approaches.