IFS Seminar

Abstract: 

The standard high performance operational scenarios in

tokamak fusion reactors are based on a formation of an edge transport

barrier and pressure pedestal at the edge of the plasma. This pedestal

region must enable sufficiently high overall fusion performance, while

also accommodating the necessary power exhaust measures to avoid

overheating of the reactor components. However, due to the multi-

ple relevant physical processes, and spatiotemporal scales, predicting

pedestals is very challenging, and an all-encompassing, high-fidelity 

pedestal model does not exist. Hence, models applying simplified 

assumptions about the characteristics of pedestal plasmas, such as 

primarily limited by features of  magnetohydrodynamics, provide 

presently the standard method for projecting performance between 

scenarios and devices. On the other hand, an integrated, high-fidelity 

model would be expected to be prohibitively expensive for agile reactor 

design, real-time prediction, or large-scale data-analysis applications. 

The development of scientific machine learning (ML) approaches has 

opened a pathway to overcome the trade-off between model fidelity and 

computational through put as well as to use experimental data to learn 

to represent the remaining reality gap in the underlying pedestal models. 

This  presentation gives an overview of the recent research conducted 

by the Fusion ML team at VTT in these ML methods to facilitate 

pedestal model development for tokamaks.