Fixed final time three-axis satellite attitude control with thrusters based on dynamic programming and neural networks

TL;DR

This paper presents a novel fixed final time attitude control method for three-axis satellites using dynamic programming and neural networks to optimize thruster switching, ensuring robustness and low cost for micro-satellites.

Contribution

It introduces a dynamic programming-based training method combined with neural networks for optimal thruster switching in satellite attitude control.

Findings

Successfully achieves fixed final time attitude maneuver.

Demonstrates robustness against system uncertainties.

Potential for low-cost micro-satellite control units.

Abstract

This paper studies the attitude control of a satellite in three-axis by thrusters. The mathematical model of attitude dynamics and kinematics of the satellite is represented as a switched system with sub-systems. Each sub-system is defined according to on/off thrusters state. A training method based on dynamic programming is utilized which can find the appropriate switching between sub-systems such that a cost function is optimized. Furthermore to extend the solution for a specific domain of interest neural network is used for approximating the cost function with basis functions. The training method is offline and it finds the optimal weights of basis functions which can be used to find optimal switching. It is shown that the proposed method can execute a maneuver in fixed final time and bring the attitude to final desired condition. Moreover, the proposed method is robust against uncertainties in system modeling. Finally, it is shown that the control scheme can be used to design low cost attitude control unit for micro-satellite or as a backup unit.