Machine learning techniques for the design of spinning tether system

Abstract

Two spacecraft can be joined by a tether to create a new, combined spacecraft. This kind of spacecraft is called the Spinning Tether System (STS). This technology finds use in various applications, including orbital maneuvers, aerobraking, and artificial gravity. In this work, we propose STS for payload transfer and orbit acquisition. The model uses gravitational capture, defined as the transition from a hyperbolic orbit to a temporary elliptical orbit, and a controlled rupture of the tether. In this process, one spacecraft loses orbital energy, while the other one has a gain in orbital energy. The one that loses energy has its temporary elliptical orbit turned into a permanent one, whereas the other is spelled from the orbit. In the present study, machine learning strategies are used to metamodel the relationship between design variables and performance index. Simulation data are ingested in the training phase of machine learning algorithms to build a metamodel that maps the relation between input and output quantities. The performance of two machine learning strategies is compared through the average error obtained in predicting new designs. The numerical results show the viability of the proposed methodology.