Elsevier

Computers & Fluids

Volume 96, 13 June 2014, Pages 322-337
Computers & Fluids

A mass-conserving volume-of-fluid method: Volume tracking and droplet surface-tension in incompressible isotropic turbulence

https://doi.org/10.1016/j.compfluid.2013.12.018Get rights and content

Highlights

  • We present a volume-of-fluid method that is locally and globally mass conserving.

  • The method is wisp-free, consistent, and second-order accurate.

  • Surface tension is accounted for using the continuum-surface-force method.

  • We simulate droplet-laden incompressible flows.

  • DNS of droplet-laden isotropic turbulence with 7000 droplets and 10243 mesh.

Abstract

We have developed a volume of fluid (VoF)/projection method for simulating droplet-laden incompressible turbulent flows with uniform density and viscosity. The method is mass-conserving, wisp-free, and consistent (i.e., the VoF function, C, satisfies the condition 0  C  1).

First, we present the results of the VoF method for tracking volumes of initially spherical shape and with zero surface tension in analytical velocity fields (linear translation, solid-body rotation, single-vortex flow, and Taylor–Green vortex) and in incompressible isotropic turbulence at Reλ0 = 75 and 190. These numerical tests show that (i) our VoF method is mass-conserving, consistent, and wisp-free; (ii) for a CFL number of 0.1, the VoF geometrical error has almost a second-order convergence rate for a mesh resolution with more than 10 grid points per diameter; (iii) in the isotropic turbulence case, a resolution of about 32 grid points per diameter of the sphere is required in order to limit the VoF geometrical error below 1%. Then, in order to simulate droplet-laden flows, we have adopted the continuum surface force (CSF) model to compute the surface tension force. We have modified the sequence of the VoF advection sweeps, and show that, in the case of droplet in a translating reference frame, the r.m.s. of the spurious currents is about 1% of the translating velocity. Finally, we present DNS results of fully-resolved droplet-laden incompressible isotropic turbulence at Reλ0=75 using a computational mesh of 10243 grid points and 7000 droplets of Weber number Werms=0.5, and initial droplet diameter equal to the Taylor length-scale of turbulence.

Introduction

The computational methods to perform fully-resolved simulations of multiphase flows fall into two main categories: 1. interface capturing methods (ICM), such as the volume of fluid (VoF) [1] and the level-set [2] methods; 2. interface tracking methods (ITM), such as the front tracking [3] and the immersed boundary [4] methods. Both ICM and ITM are designed to compute (capture or track) sharp interfaces, and can be used on a fixed Cartesian mesh. Mass conservation, the ability to compute strong deformations of the interface, and interface topology changes due to break-up and coalescence are highly desirable features in the simulations of multiphase flows, e.g. gas–liquid and liquid–liquid. Interface topology changes need ad hoc modeling in ITM [3], whereas they are directly captured by the ICM [5].

Among the ICM, the level-set is a widely used method where a distance function to the interface is advected with the flow throughout the computational domain in place of the interface itself, thus avoiding the need to advect a discontinuous function. The drawback of the level-set method is that it does not conserve mass even when coupled with a particle tracking method [6]. In the VoF method, instead, the advection equation of the volume fraction is directly solved for, thus, the VoF method could potentially conserve mass exactly. Limits to this potential may only come from the numerics adopted. A color function representing the volume fraction of the reference phase is advected geometrically and the interface is reconstructed from this function typically with a piecewise linear representation.

VoF advection schemes can be broadly classified as either direction split or unsplit schemes. Split advection schemes consist of a sequence of one-dimensional advection and reconstruction steps in each coordinate direction, thus they are algorithmically straightforward to implement compared to multidimensional (unsplit) schemes. On the other hand, unsplit methods have the advantage of only requiring one advection and reconstruction step per time step, however the advection step is often algorithmically complex. This is because unsplit methods require either the computation of a flux polyhedron for each cell face [7] or calculation of polyhedra volumes with non-planar surfaces requiring triangulation [8]. In our experience, these geometric calculations are the computational bottleneck of the VoF advection scheme. Thus, the computational savings of the unsplit versus the split method (if any) are likely to be small. Furthermore, unsplit methods do not necessarily conserve mass to zero machine precision [8]. For these reasons, we chose to adopt a split mass-conserving VoF advection approach.

This paper presents a numerical methodology to perform DNS of droplet-laden incompressible turbulent flows with uniform density and viscosity. This is a necessary step to verify that the numerical method is accurately simulating the motion of finite-size droplets, i.e., volume tracking and surface tension computations, before we can develop a coupled flow solver where density and viscosity variations occur between the droplet and the surrounding fluid (such a flow solver has been developed in Ref. [9]). We adopted the volume of fluid (VoF) method because of its potential to conserve mass with zero-machine accuracy. The VoF advection is performed through a spatially split approach, i.e., the Eulerian implicit - Eulerian algebraic - Lagrangian explicit (EI-EA-LE) algorithm which was originally proposed by Scardovelli et al. [10]. We chose EI-EA-LE over the EILE-3D algorithm [11] because EI-EA-LE requires half the number of advection and reconstruction steps and does not require the calculation of three two-dimensional divergence-free velocity fields. Thus, EI-EA-LE is at least two times computationally faster than EILE-3D. The original EI-EA-LE algorithm [10] is globally mass-conserving but generates wisps, and does not conserve the mass of the individual volumes tracked in the flow. We have improved this method with the addition of a redistribution and a wisp suppression algorithm. Our method is consistent (i.e., the volume of fluid function, C, satisfies the condition 0C1) and wisps-free and, thus, conserves mass both globally and locally within each volume tracked. We also present and analyze the numerical treatment of the surface tension force within a projection method combined with our VoF method. The surface tension effects are treated with the continuum surface force (CSF) approach proposed by Brackbill et al. [12] and adapted to the VoF method by Francois et al. [13].

We present the governing equations for droplet-laden flows in Section 2, the projection method in Section 3, the VoF interface reconstruction and advection algorithms in Section 4, and the method to compute the interface curvature in Section 5. In Section 6, we present the numerical results for the volume-tracking test-cases (zero surface tension), and for the coupled droplet-flow cases (non-zero surface tension). In Section 7, we give the concluding remarks.

Section snippets

Governing equations

The non-dimensional governing equations for a droplet-laden incompressible flow are the continuity and momentum (Navier–Stokes) equations,·u=0,tu+·(uu)=1ρ-p+1Re·μu+(u)T+1Wefσ,where u(x,t) is the fluid velocity, p(x,t) is the pressure, ρ(x,t) is the density, μ(x,t) is the viscosity, and fσ(x,t) is the force per unit volume due to the surface tension,fσ=σκnδ(x-xs).In Eq. (3), σ is the surface tension coefficient, κ is the curvature of the interface between the two fluids (i.e., droplet and

Projection method

The Navier–Stokes equations, (6), (7), are solved numerically using the projection method. Time integration of (7) from time tn to tn+1(tn+1=tn+Δt) is performed using the second-order Adams–Bashforth scheme,u-unΔt=32RUn-12RUn-1.In Eq. 8, u is a non-divergence-free fluid velocity approximating un+1, andRUn=-·(unun)+1Re2un+1Wefσn+1.The divergence-free condition on the updated fluid velocity is imposed by solving the Poisson equation for pressure,2pn+1=1Δt·u,and by updating the fluid

Volume-of-fluid (VoF) method

This section describes the numerical method for tracking the droplets in space and time with the volume-of-fluid method. In the volume of fluid (VoF) method, the sharp interface between the two phases (e.g. liquid and gas) is determined using the VoF function, C, that represents the volume fraction of the reference phase in each computational cell, e.g. C=0 in the gas, and C=1 in the liquid (0C1). The VoF method requires two steps, the VoF interface reconstruction and the VoF advection which

Curvature computation

This section describes the numerical method for computing the curvature of the interface between the two fluids that is needed for calculating the surface tension force, fσ (Eqs. (7), (12)). In order to compute the interface curvature from the VoF function, C, the height function technique by Cummins et al. [19] has been adopted and investigated. A correction based on the local orientation of the interface normal is performed to minimize the error on a spherical interface [20]. The technique is

Results

In this section, we first report the numerical results on the accuracy of the VoF interface reconstruction (Section 6.1) and VoF advection (Section 6.2), described in Section 4, for several test-cases where volumes of initially spherical shapes are tracked without surface tension in specified velocity fields. This assesses the properties of the VoF method independently from the two-way coupling between the droplet and the surrounding fluid by setting the surface tension force to zero. Then, in

Concluding remarks

We have presented a coupled volume-of-fluid/projection method to simulate droplet-laden incompressible turbulent flows with uniform density and viscosity. We adopted a three-dimensional direction-split volume of fluid (VoF) method for tracking volumes accurately in incompressible velocity fields. Our VoF method relies on the MYC interface reconstruction [11] and the EI–EA–LE advection scheme [10] to which we added a redistribution and wisp suppression algorithms (Section 4.2). EI–EA–LE has the

Acknowledgments

This work was supported by the National Science Foundation CAREER Award, Grant No. OCI-1054591. The numerical simulations were performed in part on Hyak, high-performance computer cluster at the University of Washington, Seattle, and in part on the XSEDE computational resources provided by the National Institute for Computational Sciences (NICS) at the Oak Ridge National Laboratory, under XRAC Grant No. TG-CTS100024. We specifically acknowledge the assistance of the XSEDE Extended Collaborative

References (26)

Cited by (50)

  • Shear-thinning stimulative fluid breakup in 3D pore-throat

    2024, Chemical Engineering Research and Design
View all citing articles on Scopus
1

Ph.D. student, William E. Boeing Department of Aeronautics & Astronautics, University of Washington, Seattle, USA.

2

Assistant Professor, William E. Boeing Department of Aeronautics & Astronautics, University of Washington, Seattle, USA.

View full text