On the Capability of Wall-Modeled Large Eddy Simulations to Predict Particle Dispersion in Complex Turbulent Flows

Predicting particle transport in turbulent flows has a plethora of applications, some of which are: the transport of atmospheric aerosols, the deposition of blood cells in the arteries of human bodies and the atomization of fuel droplets in combustion chambers of propulsion systems. Today this is mostly done using Computational Fluid Dynamics (CFD) methods. In this light, the main impetus for the present research is the assessment of computational methodologies to simulate the transport and deposition of droplets and fission particles simulants in various components of nuclear reactors, which is considered an issue of high safety relevance. In order to accurately describe particle dispersion in a medium, one has to first properly compute the carrier fluid field, which is a rather challenging task, especially in complex 3D wall-bounded turbulent flows. While Reynolds-Averaged Navier-Stokes (RANS) approaches are generally unsatisfactory and Direct Numerical Simulations (DNS) are computationally prohibitive, the Large Eddy Simulation (LES) stands as the most adequate tool to address complex flows at reasonably high turbulence levels. Particulate flows require that the wall boundary layer be accurately resolved, since it is near the wall that particle physics is the most complex due to turbulence anisotropy and inhomogeneity. However, wall-bounded LES which resolves the boundary layer has stringent spatial resolution requirements in all directions. This translates into large CPU needs, which grow exponentially with the Reynolds (Re) number. To address this bottleneck, recent research has proposed the so-called Wall Modeled LES (WMLES), which is a promising alternative to dramatically reduce the dependency of conventional LES on Reynolds number. Our investigation aims to take the WMLES methodology one step further by modeling the dispersion of inertial particles in an Euler/Lagrange framework under simplified conditions. As a first step in this project, a Lagrangian Particle Tracking (LPT) algorithm was implemented in T-Flows code to simulate the dispersed phase. The particulate flow is assumed to be dilute enough to justify a one-way coupling treatment. The LPT algorithm was tested against the commercial code ANSYS Fluent through canonical turbulent flows in a verification step. Then, the algorithm was validated against the reference experimental data in 90-degree bend flow. To qualify a suitable WMLES model for later solving complex configurations, two recent WMLES methods were investigated; the Algebraic WMLES (AWMLES) model by Shur et al., 2008, and the Elliptic Relaxation Hybrid RANS/LES (ER-HRL) model (Hadziabdic and Hanjalic, 2020). Both models were assessed with scrutiny in a turbulent channel flow where mean flow and Root Mean Square (RMS) values obtained by each model were compared to DNS data. To account for the effect of the unresolved scales on particle dispersion, two promising particle Sub-Grid Scale (SGS) approaches which have been proposed recently were investigated (Fukagata et al., 2004, Sayed et al., 2021-b). The predictions of both models have been validated against DNS data in periodic channel flow at two Reynolds numbers i.e. Re_tau = 150, 590. To check model performance in complex flows, three benchmark configurations have been carefully assessed, namely: Differentially Heated Cavity (DHC), Gas Cyclone Separator (GCS) and the swirl vane (droplet separator).