Agent-Based Architecture for Proactive Fault Tolerance and Management in Small Satellite Missions

The paper introduces an agent-based Proactive Fault Tolerance and Management (PFTM) architecture for small satellite missions, designed to improve reliability and autonomy. The architecture focuses on two key functions: Predictive Diagnostics and Prognostics (PDP) for fault analysis and prediction, and Proactive Fault Management (PFM) for isolating faults and executing recovery procedures. A novel aspect is the emphasis on predicting and mitigating faults before they occur, improving traditional Fault Detection, Isolation, and Recovery (FDIR) methods. A trade-off analysis is performed on five criteria: communication overhead, power consumption, maintainability, response time, and resiliency to select the best architecture for small satellite constellations. The paper reviews existing agent-based FDIR approaches and their applicability to PFTM, focusing on managing unresponsive reaction wheels, crucial for satellite attitude control. The effectiveness of the architecture is demonstrated through simulations using the Basilisk simulator, which shows that it can maintain operations despite reaction wheel failure. Both supervised and unsupervised learning methods are explored for PDP. Although unsupervised methods only require normal data for training, they are vulnerable to failure isolation. To address interpretability issues, Explainable AI (XAI) is integrated into the architecture, helping operators and agents understand the causes of anomalies. Regarding PFM, we first demonstrated the impact of unresponsive reaction wheel on the satellite's performance and attitude by analysing parameters such as attitude error, reaction wheel torque, and angular velocity. Then, the proactive decision-making module performs proper recovery action to recover from the fault and ensures that the spacecraft maintains its desired orientation. The results validate the architecture's potential to enhance mission success rates in small satellite constellations, making it a promising solution for future space missions.