Speaker
Description
Plasma detachment via impurity seeding is the most common and widely explored solution for the heat exhaust problem to date. The physics of this process has been usually studied with dedicated 2D edge transport codes such as SOLPS-ITER. However, scenario development and testing of control schemes with rapid pulse design simulators on future devices requires reduced models that run on much shorter timescales, yet provide reliable predictions for key quantities such as heat flux and radiation profile. This can be achieved, for example, by extracting a scaling law from a database of higher fidelity simulations [1] or training a neural network [2][3].
Here, we report on an ongoing effort to constitute a database of SOLPS-ITER simulations for KSTAR operating with a lower tungsten divertor with the main wall armoured with carbon PFCs. The database, comprising more than 700 simulations (all from coupled fluid/kinetic neutral code runs), a representative vertical-target magnetic equilibrium of discharge #34560 ($I_p$ = 500 kA, $B_T$ = 1.8 T, and $q_{95}$ = 4.5). The power crossing the core boundary ($P_{SOL}$) in the database varies from 4 to 7 MW. Three different impurity species, nitrogen, neon, and argon, are considered and the ratio of radiative power dissipated in the SOL to PSOL ranges from 10% to 40%.
We first use the database to demonstrate that the existing semi-empirical scaling law [4], linking effective charge ($Z_{eff}$) to the line averaged electron density and total radiation power ($P_{rad,tot}$), is in promising agreement with both simulation and experimental data. This agreement validates the database and gives physical insights for the impurity leakage of different impurities. In a second step, a deep neural network has been trained to predict the 2D radiation distribution and heat flux profiles at both divertor targets from 0D input parameters, including $P_{SOL}$, $P_{rad,tot}$, impurity species, electron temperature and density at the outer midplane. The resulting surrogate model reproduces the target heat flux profiles an order of magnitude faster than a full SOLPS-ITER simulation, with a relative error below 10% on the test set. These results indicate that the proposed approach is a promising predictor of the divertor–SOL plasma environment.
References
[1] H.D. Pacher et al., J. Nucl. Mater. 463 591–595 (2015)
[2] S. Dasbach et. al., Nucl. Mater. and Energy 34 101396 (2023)
[3] Ben Zhu et. al., Phys. Plasmas 32, 062508 (2025)
[4] G. F. Matthews et al., J. Nucl. Mater. 241 450-455 (1997)