Researchers at the UL Faculty of Mechanical Engineering have developed a machine-learning approach that significantly accelerates complex physical simulations of plasma in fusion reactors. By using advanced surrogate models, researchers can predict plasma behaviour much faster, contributing to more efficient development of fusion energy as one of the key sustainable energy sources of the future.
Understanding plasma behaviour in fusion reactors is essential for developing stable and efficient fusion devices. Simulations of plasma processes—particularly in the plasma edge region known as the Scrape-Off Layer—are computationally very demanding.
Traditional approaches rely on Particle-in-Cell simulations, which enable a highly accurate description of physical phenomena but require substantial computational power and time. This limits research scenarios where multiple operating conditions must be analysed.
Researchers Assoc. Prof. Nikola Vukašinović, PhD; Assoc. Prof. Leon Kos, PhD; Asst. Prof. Uroš Urbas, PhD; and Asst. Prof. Ivona Vasileska, PhD from the LECAD Laboratory for Engineering Design at the UL Faculty of Mechanical Engineering have developed an approach that applies machine-learning methods to accelerate these simulations. In the article Accelerating Particle-in-Cell simulations in Tokamak Scrape-off Layer using segmented surrogate models, published in the journal Engineering Applications of Artificial Intelligence (IF = 8.0), they present segmented surrogate models based on the XGBoost algorithm.
These models learn the relationship between input parameters and simulation outcomes and can rapidly predict plasma properties, such as the electric potential at the divertor target.
The results demonstrate that such models can significantly reduce simulation time while maintaining high accuracy.
The article is available at: https://doi.org/10.1016/j.engappai.2026.114332
“By applying machine-learning methods, demanding physical simulations can be significantly accelerated, opening new possibilities for analysing complex plasma processes,” emphasises Assoc. Prof. Nikola Vukašinović, the first author of the scientific article.
“Understanding plasma behaviour in the edge region of fusion devices is crucial for the development of future fusion reactors,” adds Asst. Prof. Ivona Vasileska, the lead author of the research.
In the future, such approaches could contribute to faster development of fusion reactors and broader use of artificial intelligence in physical simulations.