Autoencoder-Based Features Extraction for the Health Monitoring in the Space domain
- Resource Type
- Conference
- Authors
- Onofri, Silvia; Enttsel, Andriy; Manovi, Livia; Marchioni, Alex; Cognetta, Salvatore; Corallo, Francesco; Ciancarelli, Carlo; Mangia, Mauro; Rovatti, Riccardo; Setti, Gianluca
- Source
- 2023 IEEE 66th International Midwest Symposium on Circuits and Systems (MWSCAS) Circuits and Systems (MWSCAS), 2023 IEEE 66th International Midwest Symposium on. :458-462 Aug, 2023
- Subject
- Components, Circuits, Devices and Systems
Degradation
Satellites
Estimation
Feature extraction
Data models
Time-domain analysis
Monitoring
Remaining Useful Life (RUL)
Prognostics and health management (PHM)
Telemetry data
Autoencoder
Deep Learning
- Language
- ISSN
- 1558-3899
In the satellite operation domain, the accurate pre-diction of the Remaining Useful Life (RUL) of satellite subsystems and components is fundamental for an effective management of the mission. The accuracy of the RUL estimation depends not only on the predictive model but also on the quality and quantity of degradation features to monitor and predict. This paper proposes the use of an Autoencoder to extract time-domain features of a complex multi-sensor satellite sub-system. The Autoencoder is tested on a real-world satellite dataset for condition monitoring. The data is injected with increasing drifts to emulate degradation phenomena. The obtained features are compared with traditional time-domain features such as mean and standard deviation.