Plasma Electron Density Prediction AI Model using OES Data

AI model predicts plasma electron density from OES data. Uses CR and ID physics models to generate synthetic data

This project develops a high-precision deep learning AI model specifically designed for predicting plasma electron density (ne) using Optical Emission Spectroscopy (OES) data. The core innovation lies in its data generation methodology, which leverages sophisticated physics-based models: the Collisional-Radiative (CR) model and the Instrument Disturbance (ID) model.

The Collisional-Radiative (CR) model simulates ideal OES spectra by calculating atomic/ionic energy level populations within plasma, considering fundamental processes like electron collision excitation/de-excitation, ionization/recombination, and radiative transitions. This model provides the theoretical, pure OES spectrum intensities based on specific plasma states (ne, Te).

The Instrument Disturbance (ID) model then takes these ideal spectra and synthesizes realistic measurement effects. This includes simulating various noise sources (e.g., flicker noise, white noise), instrumental functions (e.g., spectral line broadening due to spectrometer resolution), baseline variations, and measurement errors. This crucial step ensures that the generated synthetic data closely mimics real-world OES measurements, making the AI model robust against experimental imperfections.

By training the AI model on this physically grounded and realistically perturbed synthetic dataset, the project ensures the model's high robustness and strong generalization performance. The AI model, a feedforward neural network, is optimized to accurately learn the complex relationship between OES spectral intensities (specifically, 17 distinct wavelength channels) and electron density. Electron density values are log-transformed during preprocessing to handle their wide dynamic range and align with fundamental plasma physics relationships.

The developed AI model offers significant advancements for real-time, non-invasive plasma diagnostics. Its capabilities extend to enabling precise in-situ process optimization, facilitating predictive maintenance by detecting plasma anomalies, and accelerating research and development cycles. Its applicability spans key industrial sectors such as semiconductor manufacturing, display production, and nuclear fusion energy, promising enhanced efficiency and productivity