Modeling turbulence by offline subgrid-scale simulations: the Pseudo-Direct Numerical Simulation
Turbulence remains a paramount unresolved challenge in classical physics. Its chaotic and multiscale nature is governed by the Navier-Stokes equations, which describe the dynamics of incompressible fluid flows. Although Direct Numerical Simulation (DNS) provides detailed insights into canonical turbulent flows, its high computational cost restricts its application to practical, real-world problems.
Various modeling strategies have been developed to efficiently represent mean flow features without reaching DNS grid requirements. Reynolds-Averaged Navier-Stokes (RANS) and Large Eddy Simulation (LES) are the predominant methodologies, which reduce the range of scales that must be resolved using closure models to account for the influence of unresolved scales on the resolved flow fields. These terms are typically derived from analytical formulae that quantify interactions between large-scale structures and smaller vortices, that are not captured by coarse grids.
Building on scale separation, the Pseudo-Direct Numerical Simulation (P-DNS) method offers an alternative by integrating data from detailed simulations with coarse-scale modeling techniques. P-DNS employs simulations within Representative Volume Elements (RVEs) to accurately capture essential small-scale turbulent dynamics, thereby constructing a comprehensive database that enhances closure models. This approach necessitates simulating multiple canonical configurations across a broad parameter space to encapsulate the diverse behaviors inherent in turbulent flows.
This presentation proposes a strategy to expand the RVE-based database through GPU-accelerated implementations of the Basilisk software. By simulating additional canonical configurations, we aim to refine P-DNS closure models and broaden their applicability to multiphase flow modeling.
