Scholarlink

<p>Artificial satellites are susceptible to harsh space environments, ageing effects, and thermal cycling, leading to anomalous behaviours in satellite subsystems’ health sensor data. This study proposes a unique approach for identifying four types of point anomalies (drift, stuck, out-of-limit, and spike anomalies) occurring in satellite health telemetry data by determining the most suitable machine-learning models. This novel machine-learning model-based approach reduces operator workload, minimises manual errors, and improves latency in health estimation by autonomously flagging anomalous behaviours in the satellite health data. Telemetry data from a Low Earth Orbit satellite was utilised, and synthetic anomalies were injected for evaluation. A supervised classification method was implemented, employing six different machine-learning algorithms: logistic regression, naïve Bayes, support vector machines, decision trees, k-NN, and deep learning using Keras. The F1 score was chosen as the figure of merit for selecting the best machine-learning model for each point anomaly type. This supervised learning approach has a low-computational footprint and efficiency, making it suitable for onboard satellite sensor circuitry, thereby introducing autonomy in fault detection during mission operations of large satellite constellations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MachineLearning models for identification of point anomalies in satellite telemetry data

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • M V Ramachandra Praveen,
  • Piyush Kuchhal,
  • Sushabhan Choudhury,
  • Neeraj Kumar Shukla

摘要

Artificial satellites are susceptible to harsh space environments, ageing effects, and thermal cycling, leading to anomalous behaviours in satellite subsystems’ health sensor data. This study proposes a unique approach for identifying four types of point anomalies (drift, stuck, out-of-limit, and spike anomalies) occurring in satellite health telemetry data by determining the most suitable machine-learning models. This novel machine-learning model-based approach reduces operator workload, minimises manual errors, and improves latency in health estimation by autonomously flagging anomalous behaviours in the satellite health data. Telemetry data from a Low Earth Orbit satellite was utilised, and synthetic anomalies were injected for evaluation. A supervised classification method was implemented, employing six different machine-learning algorithms: logistic regression, naïve Bayes, support vector machines, decision trees, k-NN, and deep learning using Keras. The F1 score was chosen as the figure of merit for selecting the best machine-learning model for each point anomaly type. This supervised learning approach has a low-computational footprint and efficiency, making it suitable for onboard satellite sensor circuitry, thereby introducing autonomy in fault detection during mission operations of large satellite constellations.