Erlangen 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
P: Fachverband Plasmaphysik
P 6: Plasma Wall Interaction I
P 6.4: Talk
Tuesday, March 17, 2026, 12:00–12:15, KH 01.020
Vision transformer based model regression for plasma exposed surface structures — •Torben Schmitz1,2, Dirk Reiser1, Jose Ignacio Robledo3, and Sebastijan Brezinsek1,2 — 1Forschungszentrum Jülich GmbH, Institute of Fusion Energy and Nuclear Waste Management Plasma Physics, Partner of the Trilateral Euregio Cluster (TEC), 52425 Jülich, Germany — 2Faculty of Mathematics and Natural Sciences, Heinrich Heine University Düsseldorf, 40225 Düsseldorf, Germany — 3Forschungszentrum Jülich GmbH, Jülich Supercomputing Centre, 52425 Jülich, Germany
Exposing a surface to an ion beam or plasma leads to erosion and the development of surface structures on the nanoscale. Such nanostructures have been observed on tungsten samples exposed to plasma in magnetic confinement devices. A convenient description of the evolution of these structures is possible using a Kuramoto-Sivashinsky (KS) type model whose parameters we aim to infer for given experimental data. There exist previous approaches to this problem, one training a regression model on the Fourier transform of the surfaces, another using pretrained convolutional neural networks, finetuned on the regression task. We propose a different approach using the vision-transformer architecture and including additional input features to the training process. We show that training the model on our KS dataset leads to good predictive performance. We present details of the method and results on our synthetic (simulated) dataset. The results show the capability of our architecture to understand and extract information from fusion relevant surface structures.
Keywords: Plasma wall interaction; Partial Differential equations; Machine Learning; Parameter Regression