Advancing Remaining Useful Lifetime Estimation for Spacecraft Components

Accurate Remaining Useful Lifetime (RUL) estimation is paramount for extending the operational lifespan and ensuring the success of long-duration spacecraft missions, particularly as in-orbit repairs are often impossible. This thesis addresses the critical challenge of RUL prediction for the aging gyroscopes
of the European Space Agency’s Mars Express mission. The gyroscopes are a core system component vital for spacecraft attitude control. The research was motivated by limitations in existing RUL methodologies, underscored by the unexpectedly early failure of a Mars Express gyroscope, which highlighted
the need for enhanced predictive accuracy to support crucial end-of-mission planning. The primary objective of this work was to develop and validate a more robust RUL estimation framework and to create an intuitive user interface to make these advanced predictions accessible to mission operators.
This research introduces the Delta Learning Algorithm, a novel hybrid approach that combines an analytical degradation model with a data-driven error model to capture complex aging behaviors. A sophisticated multi-stage data preprocessing pipeline, including data cleaning, combination, a crucial
temperature compensation procedure, and dedicated outlier removal was implemented to transform raw spacecraft telemetry into high-quality input for the algorithm. Furthermore, an interactive webbased user interface was developed to streamline the RUL estimation workflow end-to-end, from parameter
configuration to visualization of results. The effectiveness of the proposed methodology was rigorously validated using historical data from three previously failed Mars Express gyroscopes. The validation demonstrated that the Delta Learning Algorithm significantly outperformed previous methods, accurately predicting the failure of one critical gyroscope up to three years in advance. Applied to the currently operational and mission-critical gyroscope 5 and assuming continued low dutycycle operations, the algorithm forecasts a remaining lifetime extending into the early 2030s. This thesis delivers not only an advanced algorithmic solution for RUL estimation, but also a practical, deployable software tool planned for operational use at the European Space Operations Centre (ESOC), from which Mars Express is controlled. The key contributions include the enhanced predictive accuracy offered by the Delta Learning Algorithm, a systematic data handling methodology, and a user-centric tool that empowers informed decision-making for extending mission lifetimes and managing operational risks. This work represents a tangible advancement in prognostic health management for critical spacecraft components.