Data-driven approaches for satellite SADA system health monitoring with limited data

The Solar Array Drive Assembly (SADA) system plays a critical role in managing satellite health by ensuring continuous power generation during orbital operations. Its operational dynamics are influenced by celestial phenomena involving the Sun, Earth, and Moon, particularly during eclipses. These dynamics produce complex, high-dimensional data patterns across different timescales and modes, necessitating advanced analytical approaches for effective health monitoring. This study focuses on comparing various data-driven methods to capture the multivariate, multiscale, and multimode nature of satellite operations, specifically for monitoring the SADA system. The methods employed include Principal Component Analysis (PCA), Long Short-Term Memory (LSTM), Dynamic Independent Component Analysis (DiCCA), and a scale-mode decoupled DiCCA framework. The latter is designed to uncover latent dynamics in orbital movements and satellite functionalities, using DiCCA as internal blocks for building prediction models. By comparing sensor observations with model predictions, the study tracks residuals to assess the SADA system’s health. Real-world datasets from a communication satellite SADA system validate the effectiveness of the scale-mode decoupled framework. This study not only enhances satellite anomaly detection capabilities but also advances understanding of SADA operations, contributing to more reliable satellite health management.