GPGPU-based rising bubble simulations using a MRT lattice Boltzmann method coupled with level set interface capturing. (2nd January 2016)

• Record Type: Journal Article
• Authors: Safi, Mohammad Amin; Turek, Stefan
• Is Part Of: Computers & fluids. Volume 124(2016)
• Page Start: 170
• Page End: 184
• Publication Date: 2016-01-02
• Subjects: Rising bubbles -- Lattice Boltzmann method -- Level set method -- High density ratio -- GPGPU implementation; Fluid dynamics -- Data processing -- Periodicals; 532.050285
• DOI: 10.1016/j.compfluid.2015.06.001
• Languages: English
• ISSNs: 0045-7930

Abstract

Highlights: A consistent coupling of LBM and level set methods for multiphase flows. Robust and isotropic discretization is applied to guarantee the stability and accuracy. Successfully repeats the established finite element solutions for high density ratios. Higher GPGPU performance as compared to phase-field models. Brute-force reinitialization is computationally superior to PDE-based methods when coupled with LBM. Abstract: A multiphase Lattice Boltzmann (LB) scheme coupled with a level set interface capturing model is used for the simulation of multiphase flows, and in particular, rising bubbles under moderate and high density and viscosity ratios. We make use of consistent time integration and force discretization schemes in particular for pressure forces along with using multiple relaxation time (MRT) form of the collision in the LB equation which enables us to preserve stability and accuracy for high density and critical Eo numbers. We first present the solution for the standard test of a static bubble in order to show the accuracy of the solution with respect to the Laplace law for pressure and also the spurious velocity level. We present quantitative benchmark computations and error analysis for the 2D rising bubble test cases being further validated against high precision finite element solutions in Hysing et al. (2009). Furthermore, by applying efficient multi-core and many core general purpose GPU (GPGPU) implementations outlines, we demonstrate that the desired parallel scaling characteristics of general LBM solutions are well preserved for the proposed coupled computations. The presented implementations are shown to outperform the available GPU-based phase-field LBM solvers in terms of computational time, turning the scheme into a desirable choice for massive multiphase simulations in three dimensions.