Using Machine Learning to Generate a Surrogate Model for Plasma-Surface Interactions

Abstract

Plasma-surface interactions are an important effect in laboratory plasmas, but too complicatedto be modelled directly in plasma simulations. However, plasma surface interactions can be modelled by Transport and Range of Ions in Matter (TRIM) simulations based on a binary collision approximation of energetic ions impinging on a stationary material. In this work artificial neural networks are used to generate a model of TRIM simulations for the energy-angular distribution of ions observed at the boundary of a five-moment simulation of the sheared-flow-stabilized (SFS) z-pinch fusion experiment at the University of Washington for graphite and tungsten walls. The trained network then approximates plasma-surface interactions for conditions relevant to the SFS z-pinch fusion experiment. Connecting this model to a plasma simulation as a boundary condition promises to account for plasma-surface interactions for minimal computational expense. To this end boundary conditions representing graphite and carbon walls have been developed for the 5N moment plasma model using the trained models.