Neural Network Prediction of Discharge Current using Plume Shape and Operational Parameters in Hall Thrusters

Abstract

A neural network for prediction of discharge current, which shows nonlinearity and hysteresis dependent on coil current, has been developed to build auto control system of Hall thrusters. The prediction accuracy dependence on training data sets composed of operational parameters (previous work), 250 images of plume shape and both, operational parameters and images, are investigated. The network using only plume images can describe the non-linear mode hop jump and hysteresis that the network using only operational parameters cannot describe. The predicted discharge current, however, is fluctuated up and down, while that observed in experiment shows smooth curve. The prediction using both operating parameters and plume images as the training data, can describe mode hop jump and hysteresis with 0.8% difference between prediction current and that observed in experiment.