Machine Learning Surrogate Model of Ion Impact Energy-Angle Distributions on Plasma-Facing Components in presence of Thermal and Radio-Frequency Plasma Sheaths

In this work we obtained a fast and computationally efficient surrogate model for the determination of the ion impact distributions on plasma-facing components. Distributions are obtained from a large set of high-resolution particle-in-cell simulations with the hPIC2 code, which provide the distribution functions of the ions resolved in energy and angle with respect to the surface. Distributions are machined learned using a custom ANN/CNN technique, and the quality of the ML predictions are evaluated through a variety of error metrics (MAE, L2, etc.). Datasets for both thermal (quiescent) sheaths and phase-resolved radio-frequency sheaths, where RF sheath rectification is present, have been produced. The resulting ML surrogate model allows to considerably speed up wall erosion calculations on whole-device scale, still retaining PIC-level kinetic accuracy.