Spacecraft 3-axis Controlled Attitude Determination and Control System Reaction Wheels Fault Detection, Isolation and Identification using Machine Learning Techniques

Spacecraft attitude control systems rely on reaction wheels as the primary means of precise three-axis attitude control. Faults in these reaction wheels might lead to system instability and, in severe cases, mission failure. This paper presents advanced machine learning-based techniques for the detection, isolation, and identification of reaction wheel faults in spacecraft.The proposed approach leverages advanced data analytics and machine learning algorithms to analyze sensor data from the reaction wheels, enabling early detection of faults and effective isolation of the faulty component and identify the types of faults detected, specifically, voltage, current and temperature faults.Three-axis controlled satellite high-fidelity models are simulated to generate data for both nominal and faulty states of RW. The simulated data is employed with the FDII approach. The generated data is passed into five different machine learning classifiers, the isolation and identification results are verified via cross-validation. The proposed techniques is tested on three defined datasets using the three-orthogonal RW configuration to verify its robustness. The results show that the system has higher isolation and identification accuracy when compared to other studies that used various methodologies.