Erlangen 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

P: Fachverband Plasmaphysik

P 16: Poster Session Plasma Physics

P 16.38: Poster

Thursday, March 19, 2026, 13:45–15:45, Redoutensaal

A dual machine learning framework for electron density profile reconstruction — •Christos Vagkidis, Mirko Ramisch, Günter Tovar, and Alf Köhn-Seemann — IGVP, University of Stuttgart, Germany

Machine learning algorithms can be used either as surrogate models to replace high-fidelity codes or as tools to provide some physical information. In this work, a random decision forest is applied to replace a 3D full-wave model and a deep neural network (DNN) to reconstruct the electron plasma density profile.

The COMSOL Multiphysics software is used for the 3D modeling. A microwave is propagating through an axially symmetric plasma. The spatial power distribution of the wave is measured after the interaction with the plasma. The random forest is trained on these data and is able to predict the wave power for a given electron density profile.

A new dataset is created with the random forest and is used to train the DNN, which is used as an inverse model. By using the beam power as input, the DNN predicts the electron density. The final step is to test the DNN on actual experimental conditions. For this purpose, an atmospheric plasma torch will be used.

Keywords: machine learning; random forest; neural network; comsol multiphysics