Intelligent hybrid robust fault detection and isolation of reaction wheels in satellite attitude control system

• Published in: 2022 IEEE 9th International Workshop on Metrology for AeroSpace (MetroAeroSpace) pp. 441 - 446
• Main Authors: Castaldi, P., Nozari, H. A, Sadati-Rostami, Jalil, Banadaki, H. D., Simani, S.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.06.2022
• Subjects: Attitude control; Blended learning; Fault detection; Fault diagnosis; Fault Ensemble Classifier; Robust Fault detection and isolation; Satellite Reaction Wheels; Satellites; Series and Parallel FDI forms; Simulation; Uncertain systems; Wheels
• ISSN: 2575-7490
• DOI: 10.1109/MetroAeroSpace54187.2022.9855938

This paper presents a novel hybrid scheme for robust detection and isolation of faults affecting control torques of reaction wheel motors in the satellite control system. The proposed fault diagnosis scheme consists of a residual generation module relying on a bank of residual filters, followed by an intelligent residual evaluation module. The residuals are designed to be decoupled from aerodynamic disturbance and maneuvers by exploiting a nonlinear geometric approach. The residual evaluation module is then implemented via two separate schemes arranged in series and parallel forms. In particular, in the series form the detection module detects the occurrence of a fault, whilst the isolation module identifies the occurred fault in cascade. On the other hand, the parallel form exploits a single module carrying out these tasks simultaneously. Furthermore, an ensemble classification scheme, defined as blended learning, is exploited along with geometric approach for the first time in this work. This strategy blends heterogeneous classification schemes to improve the fault classification performances. Extensive assessments on the performances and robustness properties of the presented methods are performed by a high-fidelity satellite simulator with respect to parameter uncertainties, attitude maneuvers, disturbances, and measurements errors. The results document that the suggested hybrid fault detection and isolation outperforms the classic nonlinear geometric approach.