Machine Learning for Determining Plasma Characteristics from Atomic Emission Spectroscopy

This study introduces a framework that integrates Optical Emission Spectroscopy (OES) with machine learning (ML) to provide quantitative diagnostics and characterization of plasma parameters. A MATLAB application was developed to filter and extract atomic emission lines from experimental broadband OES spectra. These identified lines were used to build, train, and optimize a four-layer feed-forward neural network that predicted synthetically assigned plasma parameters. Two models, predicting electron temperature and electron density, were produced and evaluated using new OES spectra. Both models accurately captured trends in the synthetically assigned target values. Further development of OES-ML frameworks, with benchmarking against advanced diagnostic techniques, could extend the applications of broadband OES to determine absolute values of electron temperature, electron density, and atomic species populations, and may even enable real-time determination of plasma parameters.