Modeling of the hall thruster plasma with one-dimensional hybrid particle-in-cell and neural network methods

Abstract

In this study, a one-dimensional hybrid Particle-In-Cell numerical simulation code and neural network model were developed to investigate the plasma characteristics in the Hall thruster and predict the thruster performance without an expensive computational cost. The developed numerical code used Xe as a working gas and calculated neutrals and ions with a kinetic method to capture non-equilibrium dynamics. On the other hand, electrons were assumed as a fluid to reduce the computational cost. To verify the numerical simulation, experimental data of 50 W-class and 300 W-class Hall thrusters developed in KAIST were utilized. The calculated thrust and discharge current were agreed with experimental measurements within 15% and 5%, respectively, under four operating conditions. Furthermore, calculated electric potential profiles were agreed with experiments. Lastly, the neural network method was utilized to generalize the input and output relationship of the developed 1D hybrid-PIC numerical simulation. Output parameters were thrust and discharge current and input parameters consisted of anode flow rate, voltage drop, channel area and length, and magnetic field structure. The discrepancy between neural network prediction results and experimental measurements of the untrained 700 W-class Hall thruster performance was within 10%. Furthermore, it was shown that the input-output trend of numerical simulation was generalized with the neural network model.