Real-Time Machine Learning Control of a Hall Thruster Discharge Plasma

Abstract Facility effects are known to alter Hall thruster performance between ground and space.Furthermore, discharge current oscillations are linked to impacting both the lifetime andthrust of Hall thrusters. In this work, we present a machine learning–based control frameworkdesigned for nonlinear, time-varying systems with hysteresis that can mitigate breathing modeoscillations and adapt to environmental changes. Unlike reactive approaches such as PIDcontrol, our method is predictive and flexible across a range of control objectives. Specifically,we employ echo state networks (ESNs), a class of recurrent neural networks well-suited fortime-series prediction. To train and test the model, we apply a diverse set of discharge voltageperturbations—including sine, triangle, square, ramp, chirp, bandlimited gaussian noise,pseudorandom binary sequence, and PID—on a 6-kW H6 Hall thruster operated at 400 Vas well as 250 V at 4.3 A, and measure the resulting discharge current as a part of a processknown as system identification. Using extended convergent cross mapping, we confirm causalrelationships between the applied perturbations and discharge current response. The ESNachieves 90% prediction accuracy on previously unseen time-series data based on variance. Wethen integrate the ESN into a nonlinear model predictive control (NMPC) framework, creatinga machine learning controller, to compute optimal control trajectories that suppress dischargeoscillations. NMPC simulations indicate that discharge current oscillations can be reducedby over 90%. Finally, we employ a deployment strategy in which the learned control policy isembedded as a multilayer perceptron on a field-programmable gate array, enabling real-timeexperimental implementation. The machine learning controller reduces oscillations by 1.33%while PID increases oscillations by 4.17% at the 250 V operating point. The machine learning(ML) controller has the potential to be applicable to other operating points by following astandard procedure of system identification and control optimization. Online and adaptive MLcontrol can be used in the future to improve oscillation reduction.