Study of Machine Learning Techniques for Performance Prediction in Low-Power Hall Effect Thruster Operating on Monoatomic and Molecular Propellants

<title>Abstract</title> <p> The transition to low-power Hall effect thrusters (HETs) and the increasing demand for alternative molecular propellants require predictive modeling techniques that transcend the limitations of traditional, monatomic empirical scaling laws. This study develops a robust, propellant-agnostic machine learning (ML) framework to predict critical performance metrics, namely thrust, beam current, and discharge current, in low-power regimes across varied propellants (Xe, Kr, Ar, CO ₂ , N ₂ ). Utilizing empirical data obtained from the CAMILA and CAM200 thrusters, a comprehensive data pre-processing pipeline is designed to mitigate the inherent sparsity of vacuum facility testing. This pipeline incorporates truncated Gaussian resampling for missing data imputation, ensuring preservation of natural variance, and a Monte Carlo augmentation to expand the dataset synthetically. Ten ML regression algorithms, spanning ordinary least squares, kernel, tree, and boosting architectures, were benchmarked. Ablation studies reveal that incorporating physics-informed theoretical input injection (injecting theoretical similarity parameters such as Hall parameter proxies directly into the feature space) alongside data augmentation systematically enhances model accuracy. Tree-based and boosting models, specifically XGBoost, demonstrated exceptional predictive capabilities, achieving R ² values exceeding 98% for thrust prediction. Ultimately, this data-driven methodology provides a highly adaptable tool for predicting stable operating points, accelerating the optimization of next-generation thruster configurations while reducing reliance on expensive experimental campaigns. </p>