Multi-Fidelity Machine Learning Approach for Fast Determination of Ion Impact Energy-Angle Distributions and Material Sputtering in Radio-Frequency Plasma Sheaths

High-fidelity predictions of impurity emission from the plasma-facing components surrounding RF actuators, such as ICRH antennas, often require a large number of computationally-expensive coupled Particle-in-Cell and Binary Collision Approximation (PIC-BCA) simulations. To address this challenge, we developed a Machine Learning (ML) model that integrates a Convolutional Neural Network (CNN) and a Multi-Layer Perceptron (MLP) to reconstruct the ion impact energy-angle distributions (IEAD) in RF plasma sheaths. In order to train the ML model, we generated large training data sets using the hPIC2 code. The simulations covered up to 14-dimensional parameter space, encompassing a wide range of expected plasma conditions. To further enhance the accuracy of the model, we employed a Multi-Fidelity ML architecture. For computing impurity emission on the ICRH antenna, we then integrated the IEAD-ML model with sputtering data generated by RustBCA. The resulting approach significantly reduces by several orders of magnitude the computational time necessary to estimate impurity emission from RF sheaths. The ML results maintain the same accuracy as the original PIC-BCA simulations, with bounded error across the whole training interval.