Implementation of a xenon collisional radiative model with neural network for non-invasive determination of plasma parameters in Hall effect thrusters

The increasing number of in-orbit deployment of satellites is driven by the lowered costs brought by the advances in miniaturizing satellites' sub-systems, in particular propulsion systems. Xenon based thrusters are today widely used for their high specific impulse and low-energy consumption. However, estimation of their performance and lifetime is still performed through time-consuming stress campaigns. This work attempts to predict the operating conditions of a mid-power Hall thruster from the plume's emission spectrum, using neural network (NN) techniques. To this end, a xenon collisional radiative (CR) model based on the 5s and the highly radiative 6p states of xenon was developed. Insights from a sensitivity analysis of the xenon CR, using Morris method, backed by experimental observations allowed to identify a list of electron temperature-sensitive neutral lines relevant for diagnostic and machine learning purposes. Using these lines, a neural network model was trained and tested based on the output of the CR. Then the NN was used to make predictions over a new experimental dataset. Preliminary results show the potential of the method for applications that require a fast estimation of the plasma parameters of the thrusters, however, improvements of the CR are still required.