CubeSats orbital dynamics study using machine learning

Abstract

CubeSats are a young technology capable to perform complex and ambitious missions at considerably lower costs than larger spacecraft. A CubeSat has 1 kg of mass and a cubic shape with dimensions of 0.1 X 0.1 X 0.1 m. Several units of CubeSat can be combined to make a cluster, able to efficiently support many applications including Earth observation, telecommunications, astronomy, etc. Characterizing the Orbital and Dynamical State of such a nano-satellite system is challenging for traditional methods of analysis, due to the volume and complexity of the problem taking into account gravitational and non-gravitational perturbation models such as the effect of the aerodynamic or Solar Radiation Pressure. In this project, we will take advantage of Machine Learning techniques to approach different dynamical problems related to CubeSat constellations, such as resonances, trajectories, stability, and chaos. Our study will concentrate on a CubeSat in low, medium, and high orbits (LEO, MEO, and HEO), considering all the possible perturbations. Our Machine Learning algorithms will be implemented in libraries available in Python, such as Keras and TensorFlow to train a neural network. for instance, genetic algorithms can be useful to introduce an indicator to classify orbital behaviors, time series analysis can help to characterize chaos in the system. Our data will be created using our own integrator, which we intend to develop in python and validate it with other wildly used ones such as REBOUND. (AU)