Speaker
Description
Reactor-scale divertor design demands rapid yet physics-consistent exploration of exhaust operational spaces. Fusion power plants (FPPs) beyond ITER require fully integrated scenario development that ensures both robust core-plasma performance and compliance with engineering limits on plasma-facing components (PFCs), particularly with respect to heat fluxes and wall erosion. Central to this challenge is accelerating the evaluation of key divertor quantities - target fluxes, detachment behaviour, neutral dynamics, and pumping efficiency - while retaining the predictive capability of SOLPS-ITER for upstream conditions, i.e., separatrix density, impurity concentration and plasma fuelling. We present an AI-enabled workflow designed to meet this need by combining SOLPS-ITER model databases with data-driven surrogate modelling.
The workflow employs the SOLPS-NN surrogate model trained on an extensive set of wide-grid SOLPS-ITER simulations incorporating the advanced fluid neutral (AFN) model. These surrogates reproduce the dominant plasma–neutral physics governing the transition from attached to detached conditions and provide orders-of-magnitude speedup relative to full SOLPS-ITER simulations including EIRENE neutral kinetics. A key enhancement is the use of transfer learning to increase surrogate fidelity: AFN-based SOLPS-NN models are retrained on a smaller set of high-fidelity SOLPS-ITER simulations with EIRENE, enabling the surrogate to capture kinetic-neutral effects - such as non-Maxwellian transport, molecular processes, and detailed ionisation–recombination balance - without requiring full kinetic simulations across the entire parameter space. In principle, active-learning methods can further refine the training set by automatically targeting regions where the physics becomes strongly nonlinear or where surrogate uncertainties are elevated.
Using the low-aspect-ratio EU-DEMO (DEMO-LAR) configuration as a test case, we demonstrate how the workflow efficiently maps the exhaust operational space and identifies regimes compatible with power-handling constraints. Comparisons with reduced exhaust models indicate where simplified approaches remain valid and where kinetic-neutral physics leads to significant deviations. Post-processing of SOLPS-NN 2D plasma and neutral profiles enables rapid generation of heat- and particle-flux maps on all PFCs (e.g., for wall-erosion assessments across many scenarios) and supports fast synthetic diagnostics suitable for digital-twin environments (DTE). Finally, preliminary integration of SOLPS-NN into the JINTRAC modelling framework is underway, enabling initial tests of computationally efficient core–edge coupling schemes with dynamic boundary-condition exchange, also suitable for future Pulse Design Tools (PDT).