Housekeeping telemetry analysis for spacecraft health monitoring and predictive diagnosis using machine learning

Petr Mukhachev, Tagir Sadretdinov, Natalia Lebedeva, Sergey Soloviev, Anton Ivanov

Research output: Contribution to journalConference articlepeer-review


In this work we will explore housekeeping telemetry of thermal control subsystem in a large satellite. The subsystem is known to have periodic abnormal behaviour with sudden emergency shutdowns. Having a large amount of normal behaviour data, we will attempt to predict emergency shutdowns and detect any other unexpected behaviour, including their precursors using data-driven techniques. In this work we will explore anomaly detection techniques based on state reconstruction and prediction. Specifically, we will employ principal component analysis, denoising stacked autoencoder for state reconstruction, and random forest and lstm rnn as state prediction models. We will see that although the performance of PCA and SAE is similar on normal data, anomaly detection capabilities of SAE is superior. Although random forest predictive model will not give good results, we will be able to reliably predict normal operation of the subsystem for one hour, while detecting abnormal behaviour prior to emergency shutdown with LSTM RNN. We will conclude our work with some recommendations for building datasets for anomaly detection problems.

Original languageEnglish
JournalProceedings of the International Astronautical Congress, IAC
Publication statusPublished - 2020
Event71st International Astronautical Congress, IAC 2020 - Virtual, Online
Duration: 12 Oct 202014 Oct 2020


  • Anomaly detection
  • Machine learning
  • Prognostics and health management
  • Satellite operation
  • Spacecraft


Dive into the research topics of 'Housekeeping telemetry analysis for spacecraft health monitoring and predictive diagnosis using machine learning'. Together they form a unique fingerprint.

Cite this