Towards Digital Twin for Spacecraft Components: a Delta Learning method to assess Degradation and estimate Remaining Useful Life

TheEuropeanSpace Operations Center (ESOC) oversees spacecraft from launch to decommissioning, supporting interplanetary exploration and long-term operations. Over time, spacecraft components degrade due to extended use and harsh space conditions. In this light, stimating the Remaining Useful Life (RUL) of components is crucial for maintenance planning, end-of-life operations, and extending the lifespan of aerospace systems. However, degradation models are often incomplete or imprecise due to environmental variability, missing physics, proprietary constraints, and component-specific differences, leading to discrepancies between predictions and actual performance. This thesis proposes a novel method, based on Delta Learning, to enhance the RUL estimation by leveraging past prediction errors derived from Analytical Models. Using AI/ML algorithms, Delta Learning adjusts future predictions, generating a Confidence Interval for RUL that spans best- and worst-case scenarios. The proposed method outperformed previous models, delivering lower error rates and ensuring predictions consistently fell within the Confidence Interval. Validation was conducted on real-world failure data from space missions, demonstrating its effectiveness and reliability.