Building Data-Driven Satellite Digital Twins
- Paper number
IAC-24,D1,IP,31,x88427
- Author
Mr. Filipe Cravidão, Instituto Superior Técnico, Portugal
- Coauthor
Mr. José Pedro Figueiredo, Instituto Superior Técnico, Portugal
- Coauthor
Mr. João Paulo Monteiro, LAETA, IDMEC, Instituto Superior Técnico, Universidade de Lisboa, Portugal
- Coauthor
Prof. Paulo J.S. Gil, LAETA, IDMEC, Instituto Superior Técnico, Universidade de Lisboa, Portugal
- Coauthor
Prof. Rodrigo Ventura, Instituto Superior Técnico, Portugal
- Year
2024
- Abstract
In this paper, we propose a framework for developing data-driven satellite Digital Twins. The evolving nature of a Digital Twin allows for more accurate forecasts compared to what is achievable with traditional simulation tools, which rely on pre-defined static models of the physical entity. Increasing the accuracy of forecasts is particularly important for space traffic management, given the increased risk of collisions stemming from the growing number of satellites in orbit. Current research on the application of Digital Twin to satellite operations focuses on model-driven approaches where developers create a unique Digital Twin for each satellite, and continuously update the model using telemetry data. Model-driven Digital Twin approaches allow for accurate representations of the real spacecraft, but building the models is a time-consuming activity and requires detailed knowledge of each physical entity, usually only available to its developer. This may hinder the use of Digital Twins by most satellite operators --- particularly, those working with small spacecraft. A data-driven approach, proposed here, enables the development of Digital Twins with minimal effort from operators. To achieve this, we propose a framework to create a data-driven Digital Twin using machine learning techniques on telemetry and tracking data, as well as exogenous parameters (i.e. parameters which can be determined from ground), without any knowledge of the satellite internals. We discuss some opportunities to be leveraged, such as the increasing use of common subsystem technologies by different operators across a growing number of satellites, leading to wider availability of comparable telemetry data. We also discuss the challenges of a data-driven model, such as the sparsity, irregularity, and imprecision of telemetry data, as well as the influence of operator commands on the satellite state. We then propose a data standardization approach for building a multi-satellite telemetry database, and a set of metrics to evaluate the suitability of different machine learning techniques for predicting different kinds of data. Finally, we demonstrate the data-driven satellite Digital Twin framework in a small case study.
- Abstract document
- Manuscript document
IAC-24,D1,IP,31,x88427.pdf (🔒 authorized access only).
To get the manuscript, please contact IAF Secretariat.