Rathore, P.FZJ* ; Cifuentes, L. (Corresponding author)FZJ* ; Sarma, R.FZJ* ; Abdelsamie, A. ; Chi, C. ; Thévenin, D. ; Lintermann, A.FZJ*

2026

Abstract: High-fidelity computational fluid dynamics (CFD) simulations of turbulent flows are increasingly vital for biomedical and engineering applications. While direct numerical simulation (DNS) resolves all relevant flow scales, its computational requirements are often prohibitive for biomedical applications. Large-eddy simulation (LES) provides a more computationally efficient alternative by modeling the effects of unresolved sub-grid scales (SGS), albeit at the expense of losing fine-scale flow information. This work bridges these two CFD methods by using a super-resolution network (SRN) approach to reconstruct SGS turbulent quantities from coarse LES fields. A high-fidelity database of human laryngeal flows and the AI4HPC library are used to train the SRN, which is based on a novel convolutional defiltering model architecture.

1. 5111 - Domain-Specific Simulation & Data Life Cycle Labs (SDLs) and Research Groups (POF4-511) (POF4-511)
2. SDLFSE - SDL Fluids & Solids Engineering (SDLFSE) (SDLFSE)