Introducing Machine-Learning in Spectroscopy for Plasma Diagnostics and Predictions

Abstract

Artificial Intelligence (AI) and data science techniques are increasingly introduced in physics including plasma physics where Machine Learning (ML) is applied to emission spectroscopy for plasma parameter determination. Recently, the open-access python-based Sickit-Learn ML platform was used to analyze line intensities in the order to infer the plasma electron densities and temperatures for conditions relevant to tokamak divertors. In this paper, we discuss the application of deep-learning (DL) to synthetic line spectra for conditions of magnetic fusion plasmas with hydrogen-deuterium (H-D) mixtures. The idea will be illustrated through application of Artificial Neural Networks (ANN) to spectra of the Balmer-α line emitted by H-D mixtures, the aim being to obtain the isotopic ratios. The objective of our approach is to provide a new method to infer the hydrogen isotopic ratio sufficiently fast that can be exploited for real-time applications. We will demonstrate the proof-of-principle of our method through the application of a TensorFlow DL regression algorithm to theoretical line spectra generated with predetermined parameters.