Description
Gyrokinetic (GK) stability affects the performance of high confinement mode pedestals in spherical tokamak devices [1 – 3]. High-fidelity gyrokinetic (GK) models, such as GENE [4], capture features of pedestal turbulence, but the required computational resources and time consumption of such simulations prevent the routine applicability of these models in integrated modelling workflows. Machine learning surrogate models offer an alternative for approximating GK outputs with a fraction of the computational costs, potentially enabling, for example, routine use and faster exploration of operational space. To achieve reliable predictions, such surrogates require a representative dataset that covers the experiment-relevant parameter space consistently.
In previous work, pedestal profiles around a single MAST-U shot were fitted and parameterized, equilibria were generated with the Grad-Shafranov solver HELENA [5], and the resulting geometry was passed to run local, linear GK scans across the pedestal, generating a small, self-consistent training set [6]. While useful as a proof of principle, a one-shot dataset is too narrow to support a broadly applicable surrogate.
To address this, the work focuses on expanding coverage by developing a pipeline that curates and processes experimental data from recent MAST-U campaigns, making it usable both as model inputs and for defining experiment-informed sampling spaces. The pipeline identifies relevant H-mode discharges, automates extraction, fitting and standardisation of edge profiles, and applies quality and consistency checks with recorded provenance. Its outputs are batch inputs that can be given to the
existing GK data-generation workflow at scale.
The work contributes to a curated database of MAST-U H-mode pedestals. The resulting dataset and workflow provide a reproducible foundation for training and validating fast GK surrogates over the experiment-relevant operating space of MAST-U.
[1] P.-Y. Li, et al. Nucl. Fusion 64 (2024) 016040.
[2] D Dickinson, et al. Plasma Phys. Control. Fusion 55 (2013) 074006.
[3] J.F. Parisi, et al. Nucl. Fusion 64 (2024) 054002.
[4] F. Jenko, et al. Phys. Plasmas 7 (2000) 1904-1910.
[5] G.T.A. Huysmans, et al. Proc. Int. Conf. Comp. Phys. (1991) 371.
[6] A.M. Niemelä, et al. Proc. EPS Conf. Plasma Phys. (2025).