Toward Reliable Nanosatellite Missions: A Novel Graph Neural Network Framework for Fault and Anomaly Detection via Single Event Effect

As nanosatellites play a critical role in scientific research, Earth observation, and communication, their vulnerability to space radiation is a substantial risk. Specifically Single Event Effect, caused by high-energy particles, can lead to data corruption, hardware malfunctions, and potentially mission-critical failures. Traditional SEE mitigation techniques such as radiation-hardened electronics and redundancy systems, although effective, are often infeasible for nanosatellites due to limitations in size, weight, and cost. Our proposed GNNsee algorithm addresses this gap by leveraging the power of GNNs to model complex interactions between nanosatellite subsystems, enabling real-time detection and prediction of SEE-induced faults and anomalies. Unlike conventional machine learning models that struggle to capture the interconnected nature of nanosatellite components, GNNs inherently represent these relationships, allowing for more accurate prediction of potential failures. Our approach integrates a fault detection mechanism and an anomaly detection system, which monitor the behavior of subsystems to proactively identify and respond to SEEs.Through extensive testing, GNNsee achieved a prediction accuracy of 75%, demonstrating its effectiveness in mitigating SEE-related disruptions in space environments. This research contributes a scalable and cost-effective framework to improve nanosatellite resilience, reduce mission risks, and extend satellite lifespans. By advancing the predictive capabilities for SEEs, GNNsee sets a new standard in ensuring data integrity and operational reliability for nanosatellites operating in harsh space conditions.