AI Detects Satellite Anomalies to Reduce In-Orbit Collision Risk

Researchers from the Alan Turing Institute Defence AI Research Centre, in collaboration with the University of Strathclyde and a multinational consortium, have developed an artificial intelligence model capable of monitoring satellites and predicting anomalous orbital behavior. Published in the journal Expert Systems, the research forms part of the AI4 Space Safety and Sustainability initiative, funded by the UK Space Agency and involving academic and industry partners across Australia, Canada, New Zealand, the United Kingdom, and the United States. Led by Professor Massimiliano Vasile of the University of Strathclyde Aerospace Centre of Excellence, the project addresses the escalating congestion in low Earth orbit, where satellite deployments have surged from 159 in 2000 to an estimated 4000 in 2025. The AI system analyzes light curves, or brightness readings collected by ground-based telescopes, to establish baseline motion patterns and flag deviations. Once trained on extensive observational data and refined with high-fidelity simulations from the Aerospace Centre of Excellence and industry partner GMV, the model processes real-time inputs from observatories to automatically identify irregularities. Testing demonstrates an 88 percent accuracy rate in detecting unexpected light curve patterns, alongside the ability to differentiate between rotational states such as stable spinning versus chaotic tumbling. These capabilities are critical for space traffic management, collision avoidance maneuvers, and in-orbit servicing operations. By automating the initial screening process, the tool reduces the manual burden on human operators and accelerates response times to potential hazards. Professor Vasile emphasized that AI-driven behavioral analysis represents a significant advancement in environmental awareness and long-term space safety. The system capacity to extract meaningful orbital insights from minimal visual data marks a step toward comprehensive, systematic monitoring of resident space objects. Future development phases will focus on integrating multimodal datasets, including radar telemetry, hyperspectral imagery, and precise orbital parameters, to enhance predictive accuracy and provide deeper operational context for space situational awareness networks.