CASPEX: An end-to-end causal–physics explainable framework for trustworthy spacecraft health monitoring

Future spacecraft increasingly depend on onboard autonomy to ensure reliable operation under limited communication, complex subsystem interactions, and strict computational constraints. Effective health management must therefore provide not only early anomaly detection but also physically consistent root-cause diagnosis at the subsystem level. However, existing approaches either lack fault propagation awareness or produce diagnostic outputs that are not consistent with spacecraft dynamics. This paper presents CASPEX, a real-time causal–physics framework for autonomous spacecraft health monitoring. The proposed architecture integrates anomaly detection, dynamic causal graph learning, and physics-consistent attribution within a unified onboard pipeline to identify fault origin, propagation pathways, and subsystem-level impacts. A dual-module design supports low-latency intervention-aware diagnosis for isolated faults and detailed causal–physics analysis for complex multi-fault conditions. The framework is validated using high-fidelity multivariate telemetry generated in the Basilisk simulation environment, covering reaction wheel, power, and attitude control subsystems across single- and multi-satellite scenarios. Results show reliable identification of upstream fault sources, a 38–52% reduction in non-physical diagnostic attributions, and improved consistency with system behaviour. The operational module executes within 47–69 ms and the detailed analysis within 150–210 ms on a Cortex-A72 processor, demonstrating real-time onboard feasibility. The proposed approach provides physically grounded, interpretable, and computationally efficient fault diagnosis, supporting trustworthy autonomous operation for next-generation spacecraft and distributed satellite systems.