Abstract Details
Abstracts
Author: Xianzhu Tang
Requested Type: Poster
Submitted: 2024-04-12 17:16:13
Co-authors: Prashant Sharma, Mark Zammit, Chris Fontes, James Colgan, Xuping Xie, Qi Tang
Contact Info:
Los Alamos National Laboratory
T-5 Applied Math and Plasma Ph
Los Alamos, 87545
USA
Abstract Text:
Line radiation by impurity ions is anticipated to be primarily
responsible for plasma power exhaust in a fusion power reactor. In
current tokamak experiments, impurity radiation is also found to have
an important role in facilitating plasma detachment at the divertor
plate. Accurately modeling this radiative cooling physics requires the
solution of collisional-radiative (CR) models that are supported by
high-fidelity cross section data for various atomic and molecular
processes in a fusion plasma. Here we address two issues. The first
(i) is on the physics subtleties associated with both the CR models
and the cross section data. We will show the rather tortuous path to
get an accurate radiative power loss rate for even a pure hydrogen
plasma, at the low temperature range expected near the divertor/first
wall and after a tokamak thermal quench. The second (ii) is on the
possibility of constructing high-fidelity reduced CR model that can
serve as a low-cost surrogate in coupled radiative plasma modeling.
The current emphasis is the surrogate model for a dynamically evolving
CR as opposed to steady-state data tabulation. We will show that
machine learning (ML) can readily produce model reduction in the
latent space of an encoder-decoder deep neural network (DNN). In the
much reduced latent space variables, one can train a forward DNN flow
map for a one-time-step CR evolution. The end result is that we can
recover remarkably good whole-trajectory prediction of a highly
dynamical CR evolution in time.
This work was supported by Office of Fusion Energy Sciences through the
Base Theory Program and an ML/AI collaboration.
Comments:
I will withdraw this abstract for a poster if my other abstract for a review talk is selected by the program committee