Leibniz MMS Days 2024 - Abstract

Rudolph, Martin

Machine Learning of optical emission spectra for the efficient extraction of plasma parameters

Optical emission spectroscopy is a non-invasive technique for diagnosing plasma discharges. Emission spectra contain information on the chemical composition of a plasma, the properties of the electron population as well the evolution of the plasma in time. Yet, these spectra are rarely used to gain physical insights about a plasma, mostly due to the difficulty in interpreting the spectra. This is because the intensity of an emission line depends on the population density of a specific atom energy level, which itself depends on the population density of neighboring energy levels. Analytical descriptions of spectra are therefore rather limited when it comes to the extraction of physical parameters. Here, machine learning (ML) is used to predict electron density and temperature from emission spectra. The prediction performance of the model is evaluated based on the choice of the various hyperparameters. This may pave the way to fully exploit emission spectra in the future.