Neural Network-Based Orbit Control Method via Aerodynamic Force for Formation Flying with Variable Shape Function

Abstract

Conventional formation flying Earth-orbiting satellites control their orbits in order to properly perform their missions using thrusters, but the amount of thruster propellant loaded into a satellite is limited. Thus, the use of aerodynamic force as a control force for orbit control has been attracting attention particularly in low Earth orbit. The orbit control by the aerodynamic force can be achieved by appropriately changing the state of satellite, such as attitude and shape, to meet the aerodynamic requirements. In modeling the relationship between the state and aerodynamic forces, conventional methods assumes convexity for the satellite appearance. However, the appearance is often non-convex, which degrades control performance when conventional methods are applied. In order to solve this problem, we propose a neural network-based modeling method of aerodynamic forces acting on non-convex appearances. We train a neural network using data from an aerodynamics simulator. We then develop an algorithm for calculating the state to meet aerodynamic requirements while minimizing the amount of temporal change in the state to account for mechanical constraints. We conduct numerical simulations on the test cases of formation reconfiguration and maintenance. The results show convergence and continuous stable control of the deputy satellite to the ideal orbit.