Rathore, P. (Corresponding author)FZJ* ; Liu, X.FZJ* ; Cifuentes, L.FZJ* ; Sarma, R.FZJ* ; Lintermann, A.FZJ*

2026

Abstract: As we move towards the Exascale era, the computational cost of Direct Numerical Simulations (DNS) remains a primary bottleneck for high-fidelity fluid flow analysis. Large Eddy Simulations (LES) provide a more affordable alternative, but they inherently lack sub-grid scale (SGS) details required to precisely describe turbulence. The current work presents an alternative for the above-mentioned problem by using techniques from computer vision as data-driven approaches for closure models in low-resolution simulations. Super Resolution Networks (SRN) have been used to model the smallest length and time scales present in a turbulent flow. Current work also aims to build an SRN using an open-source library, i.e., AI4HPC, which has already been tested on a different use case for a similar task.

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)