Efficient machine learning based techniques for fault detection and identification in spacecraft reaction wheel

Abstract

Space exploration demands robust spacecraft(SC) subsystems to endure the harsh conditions of space and ensure mission success. Attitude determination and control subsystems (ADCS), as a significant subsystem within SC, are essential for providing the necessary pointing accuracy and stability, and failures in the ADCS can lead to mission failure. Therefore, robust design, thorough testing, and Fault Detection, Isolation and Identification(FDII) techniques are crucial for spacecraft operations. This paper focuses on developing advanced FDII techniques for reaction wheels(RW) within ADCS, evaluating the Prony-based FDII technique for RW, considering its accuracy, time complexity, and memory usage, and Additionally, it introduces new machine learning-based FDII techniques, including enhancements to the Prony-based FDII technique, to manage single faults more effectively. The new proposed techniques used to explore the novel area of multiple faults within the same subsystem. Results indicate that the proposed FDII techniques significantly improve fault detection accuracy, isolation time, and memory efficiency compared to traditional techniques. These advancements enhance the reliability and longevity of spacecraft missions, ensuring that critical subsystems like ADCS operate effectively in the challenging conditions of space. The contributions presented in the paper are introducing three different FDII machine learning-based techniques that support identifying five types of single faults in spacecraft ADCS RW, outperform the Prony-based FDII technique for spacecraft ADCS RW in terms of time and memory complexity, and Finally, improves the fault tolerance of the spacecraft system by detecting Multiple fault combinations that may occur at the same time in one system.