Machine Learning-based Prediction of Operation Conditions in Hall Effect Thruster Systems Using Optical Emission Spectra

This work investigates the application of machine learning (ML) algorithms in predicting the operational conditions of a Hall Effect Thruster (HET) system based on the plasma emission. The primary focus is on forecasting the anode voltage, flow rate, discharge current, and coil current from the intensity of an ensemble of xenon neutral and ionic spectral lines. An experimental setup was developed allowing to explore a wide range of operating condition. A substantial optical database containing the intensity of 28 spectral lines was built and optimized to account for 6000 different operating conditions of the HET. Fifteen statistically significant lines were then carefully selected from the dataset to train an articificial neural networks (ANNs). The Evaluation of the selected ANN model yielded promising results, with the highest accuracy achieved for the discharge current and flow rate predictions at 99% and the lowest for the prediction of the coil current at 84%.