Supervised machine learning-based Hall thruster scaling
Alfredo Marianacci, Stéphane Mazouffre
Abstract
Abstract The scaling methodology described in this paper to find the geometry and working parameters of Hall Thrusters is based on algorithms of supervised Machine Learning. The approach considers the determination of the geometrical sizes, propellant mass flow rate and discharge voltage taking thrust and specific impulse as requirements. The magnetic field is also considered. The Gradient Boosting Regression is found as the most suitable algorithm for our purpose. Scaling relies on a specific database of 54 thrusters for the determination of all parameters. The database includes measurements carried out with xenon, krypton and argon as propellant. A unique analytical approach based on the GBR algorithm has been developed and validated to determine the suitable design for a Hall thruster according to space mission requirements.