Author(s): Atul Jaiswal; Minh Duc Bui; Peter Rutschmann

Linked Author(s): Minh Duc Bui, Peter Rutschmann

Keywords: Particle-laden turbulent flow; RANS-DEM; Dispersion model; OpenFOAM

Abstract: Our work attempts to provide a comprehensive analysis of modelling two-phase flows (fluid-particle) with regard to computational requirements, available models, challenges and limitations. The adopted case (particle-laden backward facing step (BFS) flow) is numerically simulated in the framework of Eulerian-Lanrangian method (RANS-DEM) using OpenFOAM. The continuum phase (fluid) is resolved using Reynolds averaged Navier Stokes equation (RANS) and discrete phase (particles) is tracked using Newton’s second law of motion (discrete element method; DEM). Firstly, the case is simulated as single-phase (fluid without particles) using pimpleFOAM in order to define the simulation parameters and meshing requirement, giving good agreement with the observed fluid velocity profiles in the experiment. Single-phase results verify that fluid velocity profiles and flow re-attachment point are correctly predicted. Later on, particles are inserted into the system and two-phase simulations are performed using two different solvers namely DPMFoam (already available in standard OpenFOAM) and pimpleLPTFoam (self-compiled). The simulation results obtained from these solvers demonstrate almost no difference due to small concentration of particles. Thus, by not considering void fraction in the governing equations for fluid flow one can considerably save computational resources when particle concentration is small. We have also investigated the influence of level of coupling and different boundary conditions for particle initial velocity (exact data was not available for particle initial velocity). We observed almost no difference in fluid and particle velocity profiles corresponding coupling regime (one- and two-way coupling), as a small number of particles in each CFD cell are unable to modify flow fields significantly. Analysis on different initial velocities of particles shows that by proving zero initial velocity to the particles, they get the opportunity to attain the real velocity depending upon flow around them and particle response time (Stokes number) and should be the approach when the exact data is not available. Our RANS-DEM simulations give good agreement with the experiment data concerning fluid and particle velocity profiles but the particle dispersion is considerably underpredicted at all measurement locations. This can be explained as (1): RANS gives only the mean flow statistics and (2): the fluctuating components (turbulent effects on particles) are incorporated by simple dispersion models, which are simplistic models based on the turbulent kinetic energy thus unable to capture particle dispersion efficiently. We think, using large eddy simulation (LES) or direct numerical simulation (DNS) to resolve the fluid flow statistics might improve particle dispersion considerably, as unlike RANS-DEM, it does not require an extra dispersion model but at the cost of huge computational requirements. In the framework of RANS-DEM, dispersion models such as the continuous random walk (CRW) method, which considers anisotropic behavior of fluid flow, might also be able to capture particle dispersion correctly.

DOI: https://doi.org/10.3850/IAHR-39WC2521711920221242

Year: 2022