Answers are grounded in this paper and cite their sources.
Thrust and voltage models are integral to the design and evaluation of electric thrusters. Current semi-empirical models are thruster-specific and perform poorly outside their derived operating regimes, limiting their use in early-stage design. In this paper, we demonstrate the use of machine learning (ML) techniques to predict thrust and voltage for applied-field magnetoplasmadynamic thrusters (AF-MPDTs) using an updated version of the Princeton AF-MPDT database. The updated database incorporates data from the latest low-power tests, with operating points at applied magnetic flux densities ( B a ) of up to 1 T and input powers below 2 kW. Key features ( J , B a , m ̇ , R a , L a ) are identified using a combination of correlation analysis as well as embedded feature selection algorithm. Five ML models of varying complexity are trained and evaluated: linear regression, support vector regression, random forest, extreme gradient boosting, and neural networks. For thrust prediction, the random forest model achieves a mean relative error (MRE) of 17.28% on the test data and 18.64% on unseen data, compared to an MRE of 38.61% for the best performing thrust model from the literature. For voltage prediction, the support vector regression model achieves an MRE of 15.72% on test data and 14.20% on unseen data, compared to an MRE of 44.50% for the best performing voltage model from the literature. The predictive accuracy of these models suggests that ML techniques could be incorporated into future early-stage AF-MPDT design.
No takes yet. Share an insight, caveat, or question.
Parashar et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: