Abstract: Health assessment of satellite key components is essential for ensuring on-orbit reliability and supporting mission planning. Performance degradation often involves several parameters that change together and show coupled dynamics. A single parameter fails to capture the full evolution of the state, while existing fusion methods rely on expert weighting and cannot represent the dynamic shift in inter-parameter dependence. This study proposed a health assessment method based on coupled deviations in multivariate distributions. We first built a Multivariate Distribution Deviation (MDD) metric and used the Sliced Wasserstein distance to quantify high-dimensional distribution shifts efficiently. The metric captured degradation from two views: marginal variation and coupling-structure change, and avoided the computational burden of the classical Wasserstein distance in high dimensions. We then introduced a Degradation Momentum (DM) metric that accumulated positive increments in distribution-shift rates to describe deviation magnitude and trend strength. MDD and DM jointly formed a two-dimensional state-feature space. Spectral clustering on this space achieved adaptive grading and continuous tracking over the full life cycle. Validation on real data from momentum wheels and gyroscopes showed that the method detected weak early-stage changes, revealed the state-evolution process, and produced stable and interpretable assessments. The approach provides a theoretically sound and practically deployable multivariate strategy for health management of satellite components.

Key words: satellite components, health assessment, joint-distribution deviation, Sliced Wasserstein distance, degradation momentum, spectral clustering