AI Classification of Low-frequency FRC Plasma Instabilities using Neural Networks

Abstract

One of the major challenges of plasma physics is the difficulty of identifying instabilities, which is the first step in learning to control or eliminate them. This thesis applies several machine learning techniques to identify instabilities in labeled datasets of fast camera and interferometer data from the PFRC-2 device (Princeton Field-Reversed-Configuration- second generation). The machine learning techniques used include a convolutional neural network and a triplet network. A combined final classification model outputs correlation values for each measuring instrument to an instability. This thesis provides a proof of concept for similar frameworks for lightweight, fast analysis of data to be developed across experimental physics.