Real-Time Interpretation Method of Aircraft Telemetry Based on Long Short-Term Memory Network

During spacecraft testing, a large amount of telemetry data needs to be interpreted in real-time. Current real-time data interpretation methods mainly rely on predefined rules, which make it difficult to comprehensively monitor telemetry data. These rule-based interpretations depend heavily on the accuracy and comprehensiveness of criteria defined by designers, as well as their experiential knowledge. While current rule-based interpretations perform well for Boolean and digital signals, they often have broad thresholds for analog signals, leading to imprecise interpretations and an inability to track data trends, making it difficult to encapsulate logical relationships between telemetry signals. To achieve more accurate and intelligent automatic interpretations while minimizing dependence on expert knowledge, this paper proposes an AI-driven telemetry monitoring model that enhances the precision of interpretations and facilitates correlations among multiple parameters. The proposed approach aims to accurately interpret the interdependent relationships between telemetry parameters to quickly detect anomalies and to monitor trends in continuously varying telemetry data, enabling the early prediction of anomalies to reduce the risk of missing critical alerts due to broad thresholds. This study utilizes the LSTM algorithm to address the challenges posed by inconsistent temporal sequences and missing data during testing phases. A framework is introduced that employs effective historical testing data for training machine learning models, applied in real-time telemetry data interpretation.