Artificial Intelligence-based short-term satellite health forecasting
- Paper number
IAC-24,B6,IP,36,x88664
- Author
Dr. Gabriele De Canio, European Space Agency (ESA-ESOC), Germany
- Coauthor
Ms. Alisa Krstova, Airbus DS GmbH, Germany
- Coauthor
Mr. Florian Hegwein, Airbus DS GmbH, Germany
- Coauthor
Mr. Jonas Hansen, Airbus DS GmbH, Germany
- Coauthor
Mr. Patrick Fleith, Solenix GmbH, Germany
- Coauthor
Mr. Jose Martinez, Solenix GmbH, Germany
- Coauthor
Mr. Jens Lerch, European Space Agency (ESA-ESOC), Germany
- Year
2024
- Abstract
Daily monitoring of satellite telemetry (TM) behavior and identification of deviations from nominal operations is a complicated and time-consuming task for spacecraft flight control teams (FCTs). The number of TM parameters whose data is continuously being downlinked to ground is usually in the range of a few thousand, resulting in large amount of data to be analyzed by engineers to determine the spacecraft health. Since an exhaustive manual analysis is infeasible, FCTs often resort to periodically review the most recent historical data of a few selected telemetry parameters and use simple approaches to make future projections. Existing prediction systems at ESA’ European Space Operations Centre that aim to automate this process have not achieved high adoption rates across missions due to: 1) their reliance on manually labelled nominal periods in past telemetry data, and 2) high rates of false positives that can further increase the workload of the FCTs. To address these challenges, a novel AI-based system for short-term satellite health forecasting called 4caster has been developed as part of ESOC’s Artificial Intelligence for Automation (A²I) Roadmap. The system uses data-driven time series forecasting techniques to predict the future behavior of a large set of TM parameters and offers the possibility to report deviations from nominal conditions as an effect of a possible anomaly. 4caster uses a machine learning model based on the transformer architecture that jointly processes several hundreds of parameters, resulting in a novel foundation model for spacecraft telemetry forecasting. In this paper we describe the developed solution, its software architecture, functionalities, and underpinning technology. ESA’s Cryosat-2 mission was used to build the solution and guide its development. We also present the challenges associated with bringing AI-powered systems into production in the context of on-ground data processing and spacecraft operations for daily use by the Cryosat-2 FCT. Additionally, we discuss initial findings on user satisfaction and benefits brought to the existing workflows. We hope that our work will stimulate future research in the domain of AI for short-term satellite health forecasting as well as development and deployment of our solution to other ESA missions and beyond.
- Abstract document
- Manuscript document
IAC-24,B6,IP,36,x88664.pdf (🔒 authorized access only).
To get the manuscript, please contact IAF Secretariat.