SMuK 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
P: Fachverband Plasmaphysik
P 17: Complex Plasmas and Dusty Plasmas/Codes and Modeling II
P 17.6: Vortrag
Donnerstag, 23. März 2023, 17:00–17:15, CHE/0089
Towards Machine-Learned Poisson Solvers for Low-Temperature Plasma Simulations — •Ihda Chaerony Siffa1,2, Markus M. Becker1, and Jan Trieschmann2 — 1Leibniz Institute for Plasma Science and Technology (INP), Felix-Hausdorff-Straße 2, 17489 Greifswald, Germany — 2Kiel University, Kaiserstraße 2, 24143 Kiel, Germany
In multi-dimensional self-consistent low-temperature electrostatic plasma simulations, the computational effort for solving the Poisson equation can represent a large part of the overall evaluation runtime. Recently, it has been shown that by using machine learning (ML) techniques, in particular artificial neural networks (ANN), one can arrive to solutions of the Poisson equation faster (and with promising accuracy) than using the conventional numerical methods. However, the currently proposed ML-based Poisson solvers still fall short for being widely applicable in low-temperature plasma simulations, which may employ complex geometries, mixed boundary condition, etc. In this work, the requirements for making ML-based Poisson solvers applicable in low-temperature plasma simulations are discussed. Furthermore, a machine-learned Poisson solver that attempts to tackle these requirements is presented, with examples from dielectric barrier discharge (DBD) geometries. First results suggest that supervised training of an ANN with spatially dependent simulation properties and corresponding ground truth electric potential solutions allows for a machine-learned Poisson solver that generalizes well to various geometric and material configurations.