Spacecraft are exceptionally complex engineering systems that generate vast amounts of multivariate telemetry data during pre-launch environmental and functional testing. Traditional quality assurance processes rely heavily on manual analysis and static limit checking to detect anomalies and non-conformances. However, these methods often fail to capture contextual anomalies — situations where individual parameters remain within prescribed thresholds yet exhibit unnatural temporal or correlational behaviors. This study proposes a machine learning-based decision-making mechanism to automate the analysis of time-series datasets generated during satellite testing phases. By utilizing predictive algorithms, the model learns the nominal behavioral patterns of subsystems from historical test data to establish dynamic, context-aware thresholds. The performance of these data-driven models is evaluated on test datasets containing various contextual anomaly scenarios. The findings indicate that the machine learning approach effectively identifies complex discrepancies that conventional static limit checks fail to capture, thereby significantly reducing false positive rates. Ultimately, this framework serves as a robust decision-support tool for quality engineers, optimizing the anomaly detection process and ensuring higher reliability in space mission deployments.

Keywords

Space Systems Testing, Contextual Anomaly Detection, Deep Learning, Quality Assurance, Decision Support Systems