Development of Machine Learning Model for Prediction of Plasma Characteristics via Multi-Parameter Argon Collsional-Radiative Model and Comparison with Experimentally Produced Verification Data

A machine learning approach has been introduced to extract plasma characteristics from a multi-parameter argon collisional-radiative (CR) model. The multi-parameter CR model generates spectral emission intensity profiles from the plasma under various conditions of plasma density, electron temperature, wall neutralization, quenching ratio, and radiation trapping. A random forest-based machine learning model was developed, employing the various parameter conditions as input features and utilizing plasma density and electron temperature as target labels for supervised learning. The developed prediction model was validated against experimental data obtained through simultaneous diagnostics employing Langmuir probe measurements and optical emission spectroscopy (OES). The experimental data for model validation were also verified for plasma uniformity using 2-dimensional fluid simulations and for Maxwellian energy distribution through I-V curve analysis. The prediction model showed results comparable to Langmuir probe diagnostics for validation data within the range of sufficient reliability.