Reconstructing EEDFs from Emission Spectra in Capacitively Coupled Plasmas via Neural Networks Trained on PIC-CR Simulations

This work proposes a neural network-based diagnostic method for reconstructing electron energy distribution functions (EEDFs) in low-temperature capacitively coupled plasmas (CCPs) from optical emission spectra. One-dimensional particle-in-cell/Monte Carlo collision (PIC/MCC) simulations model CCP discharge behavior in argon under various discharge conditions at low gas pressure, yielding steady-state EEDFs. A global collisional-radiative (CR) model calculates excited-state densities and synthetic emission spectra based on electron collision cross sections. The resulting EEDF-spectrum dataset trains a neural network for inverse mapping from spectral line intensities to EEDF parameters. We particularly investigate the model's generalization capability at higher pressures, testing whether low-pressure training data can reliably predict EEDF features at elevated pressure conditions typical in industrial CCP applications. This method is expected to offer a non-intrusive diagnostic tool complementary to conventional probe methods for CCP characterization.