Description
This conference contribution demonstrates an accelerated workflow for large-scale analysis of tokamak pedestals. The standard pedestal prediction tools, such as EPED, Europed, IPED, and IMEP, combine reduced transport assumptions with the overall magnetohydrodynamic (MHD) stability envelope [1 – 4]. Computing the latter dominates the overall computational cost of these models and limits their utility in large-scale analysis studies.
By deploying a recently published surrogate model for pedestal MHD stability evaluations, KARHU [5], within an analysis workflow, this work demonstrates capability to conduct large-scale pedestal analysis within a fraction of the cost of using a conventional MHD stability solver. With such an accelerated workflow, it is possible to efficiently scope through the JET Pedestal database with an EPED-like model and identify the parameter spaces within which the standard model assumptions do not hold, as published in previous studies [7, 8]. Furthermore, the surrogate accelerated workflow allows fast generation of the overall peeling-ballooning stability diagram, such that degree to which the experimental point is within proximity of the stability boundary can be evaluated nearly real-time with the workflow. The present focus of the work is to extend the operational space of the workflow to cover multiple tokamaks as well as to increase the fidelity of the surrogate model with resistive MHD features through targeted CASTOR simulations [9]. The results demonstrate the usefulness of machine learning surrogate models supporting pedestal stability analysis, integrated modelling, large-scale physics studies, and future real-time assessment for experiment planning and control.