Prediction of Abnormal Plasma Discharge through Machine Learning

High-temperature plasma discharge experiments have been conducted at the National Institute for Fusion Science. Visible light emission is observed after the plasma ignition is started, which has contributed to monitoring the plasma discharge condition for guaranteeing the safe operation in the experiments. The future plasma discharge condition has to be prospected in order to avoid dangerous situations in which the plasma can destroy the devices inside of the machine. In this paper, we aimed to predict the abnormal emission in plasma videos. This paper proposes methods to predict an abnormal visible light emission inside of the machine using support vector machine (SVM) and neural network (NN). Consequently, we confirmed that the abnormal emission can be predicted with an accuracy rate of 85.8% by using NN, and 96.5% by using SVM.