A support vector regression method for efficiently determining neutral profiles from laser induced fluorescence data.

Area of Science:

• Plasma Physics
• Atomic Physics
• Machine Learning

Background:

• Helicon plasmas are crucial in various applications, but accurately diagnosing their properties, like atomic profiles, remains challenging.
• Collisional Radiative (CR) models are essential for understanding plasma behavior but are sensitive to input parameter uncertainties.
• Laser-Induced Fluorescence (LIF) is a powerful diagnostic tool, but its interpretation can be complex.

Purpose of the Study:

• To develop an efficient method for determining Ar i (Argon ion) profiles in helicon plasmas.
• To integrate a Support Vector Regression (SVR) machine learning model with a CR model for enhanced plasma diagnostics.
• To improve the robustness of CR models by incorporating measurement errors from LIF data.

Main Methods:

• A Support Vector Regression (SVR) model was trained using data from a Helicon-Cathode (HelCat) linear plasma device.
• The SVR model was integrated with a Collisional Radiative (CR) model to interpret metastable-pumped Laser Induced Fluorescence (LIF) measurements.
• Radial points from plasma measurements served as input features for the SVR, mapping to parameters of a sigmoidal-type function as output.

Main Results:

• The integrated SVR-CR model successfully determined Ar i profiles from LIF measurements.
• The machine learning approach allowed for efficient exploration of the CR model's input parameter space.
• The method inherently incorporated uncertainties from probe and LIF measurements, improving reliability.

Conclusions:

• The developed SVR-CR method offers an efficient and robust approach for determining atomic profiles in plasmas.
• This technique can be adapted for other LIF pumping schemes and plasma devices.
• The method holds potential for validating electron temperature and density profiles when neutral or ion profiles are known.