Predictive, data-driven model for the anomalous electron collision frequency in a Hall effect thruster

Abstract

Symbolic regression is applied to find a data-driven model for the anomalous cross-field electron transport in a Hall effect thruster. This model is formulated in terms of an anomalous electron collision frequency that is related to the cross-field electron transport through a generalized Ohm’s law. Empirically determined estimates of this anomalous collision frequency as a function of local plasma parameters from three 1–6 kW class Hall effect thrusters form the training dataset for this investigation. A commercially-available, evolutionary genetic algorithm is applied to regress this dataset and identify models for the anomalous collision frequency that are expressed as symbolic functions of the local plasma properties. It is found that these data-driven models not only fit the training dataset but that they can predict anomalous collision frequency values for a test dataset taken from a fourth thruster not used in the initial regression. Five existing models for the anomalous collision frequency derived from first-principles are applied to the same training and test datasets used for the data-driven model. The estimates of the anomalous collision frequency as a function of local plasma parameters from the data-driven models are shown to exhibit improved quantitative agreement with both datasets compared to the analytical models. These findings are discussed in terms of the physical insight they yield for identifying dominant physical processes that govern electron transport as well as the practical application of using this technique for creating predictive Hall thruster models.