Data-Driven Modeling of Miniature Hall Thrusters: A Machine Learning Approach

New data-driven and physics-based modeling approaches have been applied synergistically to advance the design of sub-kilowatt Hall thrusters. An extensive literature database was compiled and cleaned using a custom program to impute missing values resulting from patents and confidentiality restrictions. Generative adversarial networks (GANs) augmented the limited dataset, covering various thruster geometries and operating conditions. A correlation analysis identified the most influential performance parameters, which served as inputs to a surrogate artificial neural network (ANN) model for rapid design prediction. The ANN predictions were closely aligned with conventional linear regression models, thus validating the approach. Numerical simulations of the ANN-generated designs demonstrated accurate performance projections, highlighting the potential of these new machine-learning techniques in the design of electric propulsion systems. This framework synergized data resources, machine learning models, and high-fidelity simulations to realize next-generation, low-power Hall thrusters.