You are on page 1of 23

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN FLUIDS

Int. J. Numer. Meth. Fluids 2013; 71:960982


Published online 17 July 2012 in Wiley Online Library (wileyonlinelibrary.com/journal/nmf). DOI: 10.1002/d.3692
Numerical simulation of a single rising bubble by VOF with
surface compression
J. Klostermann
*
,
, K. Schaake and R. Schwarze
Institute of Mechanics and Fluid Dynamics, TU Bergakademie Freiberg, Freiberg, Germany
SUMMARY
The capability of the direct volume of uid method for describing the surface dynamics of a free two-
dimensional rising bubble is evaluated using quantities of a recently published benchmark. The model
equations are implemented in the open source computational uid dynamics library OpenFOAM

. Here,
a main ingredient of the numerical method is the so-called surface compression that corrects the uxes near
the interface between two phases. The application of this method with respect to two test cases of a bench-
mark is considered in the main part. The test cases differ in physical properties, thus in different surface
tension effects. The quantities centre of mass position, circularity and rise velocity are tracked over time and
compared with the ones given in the benchmark. For test case one, where surface tension effects are more
pre-eminent, deviations from the benchmark results become more obvious. However, the ow features are
still within reasonable range. Nevertheless, for test case two, which has higher density and viscosity ratios
and above all a lower inuence of the surface tension force, good agreement compared with the benchmark
reference results is achieved. This paper demonstrates the good capabilities of the direct volume of uid
method with surface compression with regard to the preservation of sharp interfaces, boundedness, mass
conservation and low computational time. Some limitation regarding the occurrence of parasitic currents,
bad pressure jump prediction and bad grid convergence have been observed. With these restrictions in
mind, the method is suitable for the simulation of similar two-phase ow congurations. Copyright 2012
John Wiley & Sons, Ltd.
Received 1 September 2010; Revised 16 April 2012; Accepted 30 April 2012
KEY WORDS: VOF; surface tension force; parasitic/spurious currents; direct method; free surface; rising
bubble benchmark
1. INTRODUCTION
Interfacial ows of two or more immiscible uids are widely used in many industrial processes,
for example, in casting operations, food processing, bre coating, emulsication, spraying and so
on. The kinematics and dynamics of the uid interfaces play an important role in many of these
processes. For example, the interface between the covering slag and the liquid steel in a continuous
casting mould can strongly affect the quality of the nal product, when interfacial instabilities are
induced by the ow [1].
Fast, robust and accurate numerical models are necessary for the investigation of the interfacial
uid dynamics by means of computational uid dynamics (CFD) on xed computing grids. Promi-
nent approaches are marker and cell [2], volume of uid method (VOF) [3] and level set method
(LS) [4, 5]. A detailed discussion of these approaches is beyond the scope of this paper; interested
readers are referred to review papers, for example, by Rider and Kothe [6], Scardovelli and Zaleski
[7] or Sethian and Smereka [8].
*Correspondence to: J. Klostermann, Institute of Mechanics and Fluid Dynamics, TU Bergakademie Freiberg,
Lampadiusstr. 4, 09599 Freiberg, Germany.

E-mail: jens.klostermann@imfd.tu-freiberg.de
Copyright 2012 John Wiley & Sons, Ltd.
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 961
There are numerous papers available which demonstrate that marker and cell , VOF and LS are
indeed fast and robust tools. However, as recently noticed by Hysing et al. [9], the accuracy of
the numerical models is only seldom checked. Validation of two-phase ow simulations is often
performed by qualitative comparison of interface locations in CFD simulations with pictures from
corresponding experiments, see, for example, Ferziger and Peri [10] or Yeoh and Barber [11].
Therefore, Hysing et al. [9] have proposed a quantitative benchmark case, the two-dimensional
rising bubble problem for two different set-ups. The benchmark is deduced from the numerical
results of independent CFD simulations with three different codes. Two of them are based on the LS
approach on xed grids. The third one resolves the two-phase ow with an arbitrary Lagrangian
Eulerian approach where the computing mesh follows the phase interface. This is often referred to
as interface tracking approach. The results of the three different codes are very close to each other,
which suggests that the modelling errors should be very small. Hysing et al. [9] deduced measures
from their results, which can be used to allow a quantitative comparison with results from other
two-phase ow models.
The benchmark of Hysing et al. [9] has been already employed by [12] to check the accuracy of
a two-uid approach. However, a quantitative comparison of the VOF model data with the LS and
arbitrary LagrangianEulerian results in Hysing et al. [9] is missing up to now.
In the VOF approach, a phase transport equation is used to capture the interphase. There are
basically two methods for evaluating phase transport equation:
v direct methods: which discretise the phase transport equations directly by using high res-
olution discretisation schemes for the convection parts. Schemes specically developed for
interface phenomena such as, for example, compressive interface capturing scheme for arbi-
trary meshes [13] or high-resolution interface-capturing scheme [14]. There are no limitations
for the underlying computing mesh for the direct methods.
v reconstruction methods: where the phase transport equation is approximated typically in two
steps, a geometric interface reconstruction step and an interface propagation step. Methods
following this approach include donoracceptor method [3], simple line interface calculation,
piecewise linear interface construction [15] or efcient least-squares volume of uid inter-
face reconstruction algorithm [16]. All these methods are restricted to hexahedral (or quadratic
in two dimensions) shaped mesh cells. The recently presented moments of uids method by
Dyadechko and Shashkov [17, 18] overcomes this limit and extends the reconstruction methods
to arbitrary cell shapes. However, these methods are more computational demanding.
zkan et al. [19] evaluated the two different VOF methods for a three-dimensional bubble-train ow.
Their study reveals that there are some deciencies for the evaluated direct methods (implemented
in the commercial CFD-solvers STAR-CD, CFX and FLUENT).
In this paper, we evaluate the VOF approach implemented in the CFDlibrary OpenFOAM

version
1.5.1 through version 2.1.0 [20, 21]. In OpenFOAM

, the phase transport equation is solved directly,


which is described in detail in Section 2.
The rest of the paper is outlined as follows: A short summary of the test cases and the numerical
set-up of the present investigation is given in Section 3. The results of the VOF model applied to the
two test cases are presented, compared with the ndings of Hysing and co-workers [9] and discussed
in detail in Section 4.
2. MATHEMATICAL MODEL AND COMPUTATIONAL METHODS
2.1. Governing equations
The rising bubble problem investigates the ascending motion of a bubble of uid 2 that is surrounded
by uid 1. To track the two-phase ow of the rising bubble, the VOF method of Hirt and Nichols [3]
is adopted. Here, a volume fraction ;
1
is used to mark uid 1 with the denition
;
1
=
_
_
_
1 in uid 1
0 in uid 2
0 < ;
1
< 1 at the interface
(1)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
962 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
The governing equations of the unsteady, laminar and incompressible ow are the equation of
continuity (Equation (2)) and the NavierStokes equation (Equation (3))
V u =0 (2)
d
dt
(j u) V (j u u) =V 2V
_
j

D
_
j g okV;
1
(3)
Here, u is the velocity, is the pressure and g is the gravitational acceleration. The mixed uid
properties density j and viscosity j are weighted by the volume fractions ;
1
and ;
2
=1;
1
of the
two uids
j =;
1
j
1
(1 ;
1
)j
2
, j =;
1
j
1
(1 ;
1
)j
2
(4)
The rate of strain tensor is dened by

D =
1
2
_
V u
_
V u
_
T
_
(5)
Because of the continuous surface force (CSF) approach of Brackbill et al. [22], the surface tension
force is included as an additional source term okV;
1
in the NavierStokes equation. The curvature
k is dened as
k =V ( n) (6)
the divergence of the normal vector n of the interface. Its denition is given in Section 2.2.2.
Solving the phase transport equation by the surface compression approach. The transport equation
of each volume fraction ;
1
and ;
2
in a incompressible two-uid system is given by
d;
i
dt
V ( u
i
;
i
) =0, i =1, 2 (7)
For the derivation of the surface compression approach, it is sufcient to consider the transport
equation of the volume fraction ;
1
only
d;
1
dt
V ( u
1
;
1
) =0 (8)
To solve this transport equation, the velocity u
1
of uid 1 is needed. In the widely used origi-
nal VOF method by Hirt and Nichols [3], the velocity u
1
is assumed to be equal to the mixed
velocity u = u
1
d;
1
dt
V ( u;
1
) =0 (9)
This is only valid if ;
1
is maintained as a step function throughout the domain, for example,
numerical diffusion at the interface is not allowed. Furthermore, the upper boundedness ;
1
6 1
for Equation (9) is not guaranteed because this formulation is not conservative [23].
According to Rusche [23], it was Weller who rst developed and implemented a conservative
form in the CFD library OpenFOAM

[20, 21]. He dened the mixed u and the relative velocities


u
i
between phases ;
1
and ;
2
,
u =;
1
u
1
;
2
u
2
=;
1
u
1
(1 ;
1
) u
2
(10)
u
i
= u
1
u
2
(11)
Then, the addition of Equations (10) (multiplied by ;
1
) and (11) (multiplied by 1 ;
1
) yields
;
1
u
1
=;
1
u ;
1
(1 ;
1
) u
i
(12)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 963
Finally, the relative velocity formulation of the transport of ;
1
in the surface compression approach
is obtained by inserting Equation (12) in Equation (8)
d;
1
dt
V ( u;
1
) V u
i
;
1
(1 ;
1
)| =0 (13)
with the explicitly xed relative velocity u
i
, which is dened in Section 2.2.2. The surface compres-
sion termV u
i
;
1
(1;
1
)| contributes only in the region of the interface (0 < ;
1
< 1) and limits the
smearing of the interface because of the compensation of the diffusive uxes. The convective term in
the surface compression approach (Equation (13)) differ from the ones used in the other direct VOF
methods (e.g. compressive interface capturing scheme for arbitrary meshes [13] and [14]) where the
third term on the left-hand side (LHS) of Equation (13) is neglected.
2.2. Discretised model equations
Equations (2), (3) and (13) are integrated with the CFD library OpenFOAM

[20, 21]. OpenFOAM

uses the nite volume method in cell-centred formulation to solve systems of partial differential
equations on three-dimensional block-structured or unstructured meshes consisting of arbitrarily
shaped convex cells. With Gausss theorem applied to the convective and diffusive

terms and
using the Euler implicit time scheme

of Equation (3), the following semi-discretised system of
equations is derived
_
V
P
_
j
^t
( u
n
u
o
)
_
JV

(
j
(
u
n
(
=

(
j
(

S V
?
(
u
n

_
_
1
V
]

(
u
n1

S
_
_

(
j
n1
(

_
V
P
(V
n
g .Vj okV;)JV (14)
where = u
(


S represents the volumetric ux through the cell face S. The subscript ()
(
denotes
face values, which are interpolated from the cell values indicated by ()
]
. The superscript ()
o
indi-
cates previous (old) time step. The superscript ()
n
denotes the value of the actual iteration within
the current time step, thus indicating implicit treatment. Respectively, the superscript ()
n1
gives the
value of the previous iteration, thus indicating explicit treatment. Note that is always a face value,
thus the subscript is omitted here. For simplication, the subscript of the volume fraction is also
omitted, thus ; =;
1
. The operator V
?
(
denotes a surface normal gradient, which is the component
normal to the cell face of the gradient. This term is part of the Laplacian term and is discretised as
follows V
?
(
u
n
=
E u
n
N
E u
n
P
jdj
. The index 1 denotes the cell of interest, the index N the neighbouring
cell and [J[ the distance between both neighbouring cells. The pressure has been replaced by the
modied pressure
n
= j g . with . as the position vector. Its derivation is given by Rusche
[23]. He states that this treatment enables an efcient numerical treatment of the steep density jump
at the interface by including the hydrostatic term g .Vj into the RhieChow correction.
2.2.1. Segregated pressure-based solver (velocitypressure coupling). For the description of the
velocitypressure coupling, we follow the procedure of Rusche [23] and Medina [24].
Equations (2) and (3) are solved by the segregated pressure correction method, which will be
outlined in this section. If we take only the terms containing the velocities u from Equation (14), the
following system of linear algebraic equations can be dened.
A:=
_
V
P
j
^t
(U
n
)JV

(
j
(
U
n
(

r(pE uE u)

(
j
(
V
?
(
U
n
(

(
U
n1
(

(
j
n1
(
[

S[

r(rE uC(rE u)
T
)
(15)

By using the product rule, the diffusive term of the momentum equation (Equation (3) with Equation (5)) can be
expanded to: r (
_
rE uC
_
rE u)
T
__
Dr (rE u) Cr (rE u)
T
Cr (rE u)
T
Dr (rE u) CrE u r.

This time scheme is used for the demonstration of the discretisation process; for the simulations later on, the
CrankNicolson time scheme is used.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
964 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
This (A) represents the full system of equations AU = b, with A being the momentum coefcient
matrix, U the vector of the unknown velocity vectors u and b the vector containing the source terms.
The following operators acting on A can be dened and are implemented in OpenFOAM

: the
D-operator A
D
gives the diagonal part of A, the N-operator A
N
representing the lower and
upper triangle (off-diagonal) part of A, the S-operator A
S
extracting the source vector b and
the H-operator A
H
dened by A
H
:= A
S
A
N
U = A
D
U. With the operators given earlier,
the equation system A can be rewritten as (A
D
A
N
)U = A
S
, which allows us to rewrite
Equation (14) as
A
D
U
n
=A
H

_
V
P
_
V
n
g .Vj okV;
j
^t
U
o
_
JV (16)
Assembling a block matrix system of Equations (16) and (2)
_
A
D
V
V 0
_ _
U
n

n
_
=
_
A
H

_
V
P
_
V
n
g

z Vj okV;
p
Zt
U
o
_
JV
0
_
(17)
and taking the Schur complement of it

gives the pressure equation:
V
_
(A
D
1
)
(
V
n
|
_
=V

(18)
with ux predictor

=(A
D
1
A
H
)
(


S (A
D
1
)
(
_
( g .)
(
[

S[V
?
(
j (ok)
(
[

S[V
?
(
;
_

source term
(19)
(A
D
1
)
(
_
j
^t
U
o
_
(
[

S[

extra source from time derivative
The ux is corrected by
=

(A
D
1
)
(

S V
?
(

n
(20)
after the calculation of the pressure with Equation (18).
This method gives an oscillation-free velocity eld in line with the RhieChow correction [25],
even though there is no explicit RhieChow correction.
To obey the conservation of the uxes on the right-hand side of equations ((18), (19) and (20)),
one has to assure consistent treatment of all terms because they are assembled explicit.
2.2.2. Surface compression. In OpenFOAM

[20, 21], the transport equation for the volume frac-


tion ; is given by Equation (13). By default, a relative velocity u
i
cannot be determined in the VOF
method because only a single velocity u for both uids is considered in the whole domain. Thus,
u
i
or rather the cell face ux
i
velocity has to be approximated as given in Equation (25). The
semi-discretised form of Equation (13) is
;|
nC1
i
=;|
n
i
^t {V (; u) u
i
;(1 ;)|} (21)
Gausss theorem and integration over a control volume V
]
gives
_
V
P
V (; u) u
i
;(1 ;)|JV =

(
{(;)
(
(;
i
(1 ;))
(
} (22)
Introducing
;
as

;
:={(; u) u
i
;(1 ;)|}
(


S =(;)
(
(;
i
(1 ;))
(
(23)

This is the same as taking the divergence of Equation (15) with the continuity equation Equation (2) r
E
U
n
D0.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 965
which results in the discretised transport equation for the phase indicator
_
V
P
;|
nC1
i
JV =
_
V
P
;|
n
i
JV ^t

;
(24)
The compression ux,
i
in Equation (23), is dened by the magnitude of the face ux velocity
in the transition region (u
c
)
(
multiplied with the normal face vector of the interface (n
i
)
(
, which
yields only the contribution normal to the interface.

i
:=(u
c
)
(
(n
i
)
(
(25)
The face ux velocity in the transition region (u
c
)
(
is then evaluated by
(u
c
)
(
=min
_
c
;

S[

S[

max
_
(26)
which is bounded to the maximum face ux velocity


j
E
Sj

max
in the ow eld. The coefcient c
;
controls the weight of the compression ux
i
and should be in the range of unity [24]. By choosing
c
;
= 0, the compression ux can be forced to
i
= 0. The contribution to the interface, which is
normal at the cell face, is derived by the scalar projection (n
i
)
(
of the normal vector of the interface
pointing from uid 2 (; =0) to uid 1 (; =1) onto the cell face vector

S. This term is determined
by the following inner product:
(n
i
)
(
= n

S =
V
?
(
;
[V
?
(
; b[


S with b =
10
S
3
_

N
V
P
1
(27)
Algorithm 1. Explicit two-phase numerical solution procedure.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
966 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
to prevent division by zero with V
1
denoting the cell volume and N the number of cells. The scalar
(n
i
)
(
is also used for the determination of the curvature in Equation (6); here, the divergence term
is rewritten with Gausss theorem:
k =V ( n) =

(
(n
i
)
(
(28)
2.2.3. Solution algorithm. The overall solution cycle is given in Algorithm 1. Here, the distribu-
tion of the volume fraction eld ( ;-subcyle at step 4) is calculated ahead of the well-known PISO
algorithm [26] (step 8) that updates the velocity and pressure elds.
The key element of the ;-subcyle is the solution of Equation (24). To guarantee a sharp and
bounded solution, a numerical scheme designed for the multi-dimensional advection equation
should be employed (e.g. [27, 28]). For this study, the explicit multidimensional universal limiter
with explicit solution (MULES) implemented in OpenFOAM

[20, 21] is used to integrate the


;-equation Equation (24) explicitly. Therefore, Equations (23) and (24) are solved together with
the mass ux j
(

i
j
(

i
:=j
(
u
(


S =(;j
1
)
(
(1 ;)j
2
|
(
=
;
(j
1
j
2
)
(
(j
2
)
(
(29)
for a number of N
;corr
corrector steps. During this correction,
i
stays constant, see Algorithm 1.
Finally, the mass ux is calculated according to Equation (24) as
j
(
=
1
sub

iD1
t
i
^t
{j
(

i
} (30)
with t
i
= ^t ,N
;sub
. Equations (4), (6), (25) and (30) are iterated inside the so-called ;-sub
cycle. This sub cycle consists of solving the equations N
;sub
-times (sub time-stepping; here, also

i
is updated, see in Algorithm 1) and then averaging the mass ux.
During the PISO loop, the mass ux j
(
in the advection term is maintained constant. Once
the pressurevelocity equations are solved, the volumetric ux is updated. The overall numerical
procedure is sketched in Algorithm 1.
2.2.4. Discretisation schemes and used parameters. The applied discretisation schemes and the
parameters of the numerical model are summarised in Table I. For convenience, the corresponding
terminology of OpenFOAM

is also given. The transient term in the momentum equation,


Equation (14), is discretised by the CrankNicolson scheme [29]. Different interpolation schemes
are applied to determine cell face values in the convective terms. For the convective part in the
momentum equation, a total variation diminishing scheme with the ux limiter function [(r) =
max0, min(2 r, 1)| is used. The smoothness parameter r is dened by the ratio of consecutive gra-
dients. The van Leer [30] scheme with the ux limiter function [(r) =(r [r[),(1[r[) is used for
the discretisation of the V( u;) termin Equation (13). Here, the ux is bounded between 0 and 1. For
the second convective part V ( u
i
(1 ;);) in Equation (13), the so-called interfaceCompression
scheme is used. Here, the limiter function reads
[(
1
,
1
) =min
_
max
_
1 max
_
_
1 4
1
(1
]
),
_
1 4
1
(1
1
)
_
, 0
_
, 1
_
(31)
which is bounded between 0 and 1, with the ux
1
at the evaluated cell and the ux
1
at the
neighbour cell. In our study, we included the optional momentum predictor step that brought small
but noticeable improvements in the results. This will be discussed later.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 967
Table I. Used discretisation and interpolation schemes, parameters of the numerical model.
Discretisation scheme
Term (OpenFOAM

terminology) Method
J
Jt
(j u) CrankNicolson 1 CrankNicolson scheme [29]
V (j u u) Limited linearV 1 TVD scheme, limiter function see text
V ( u;) vanLeer 01 TVD scheme, van Leer limiter see text
V ( u
i
(1 ;);) InterfaceCompression Bounded limited scheme, limiter see text
V u, V; Linear CDS
V
?
(

Linear corrected Surface normal gradient: CDS from the neighbouring


cell values with explicit correction on nonorthogonal
meshes [31]
V (jV
i;
) Linear corrected Face values (j) approximated by CDS, and the resulting
surface normal gradient is evaluated as given earlier
by CDS with nonorthogonal correction [31]
Term Interpolation scheme Method
()
(
Linear Default interpolation scheme for getting face values
from cell values, for example, j, j and CDS from
the neighbouring cell values
Parameter Value Notes
c
;
1 Interface compression valid values between 0 and 1
N
;corr
1 Number of ; corrector steps
N
;sub
2 Number of ; -sub cycles
N
nonortho-corr
0 Only used for nonorthogonal meshes
N
corr
15 Max. no. of PISO loops
momPred true Momentum prediction step
TVD, total variation diminishing; CDS, central differencing scheme.

The symbol stands for an arbitrary scalar or vector.


3. NUMERICAL SET-UP OF THE SIMULATIONS
3.1. Computational domain
On the basis of the benchmark denition of Hysing et al. [9], a two-dimensional computational
domain with an aspect ratio . : , = 1 : 2 is employed, see Figure 1. The bubble is initially centred
at (., ,) = (0.5, 0.5) with r
b0
= 0.25 as the initial radius. The viscosity and density of uid 2
(bubble C
2
) are smaller than those of the surrounding uid 1 (C
1
). The domain is fully enclosed by
no-slip walls ( u = (0, 0)) at the top and the bottom and free slip walls (u
x
= 0 and V
?
(
u = 0) on
the left and the right. The gravity vector g points towards the bottom of the domain.
Hysing et al. [9] distinguished two different set-ups of the numerical experiment: ellipsoidal bub-
ble (TC1) and skirted bubble (TC2). They differ by the density ratio j
1
,j
2
and the ratio of the
viscosity j
1
,j
2
between both phases (index 1 for heavy and index 2 for light uid). Physical
parameters of TC1 and TC2 are given in Table II.
All parameters in Table II are made dimensionless with characteristic scales for the length
1 =2r
b0
, for the time t =1,U

and for the rising velocity U

=
_
g2r
b0
. The Reynolds number
1e, the Etvs number 1o and the capillary number Ca are dened as
1e =
j
1
U

1
j
1
, 1o =
j
1
U
2

1
o
, Ca =
j
1
U

o
=
1o
1e
(32)
Thus, the surface tension force is more prominent for TC1 in comparison with TC2, see Table II.
The simulations of TC1 and TC2 are performed with four different grid spacings h =
1,40, 80, 160, 320|, respectively. The simulation time is t
nal
=3 with a grid size-dependent time
step of ^t =h,2.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
968 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
Figure 1. Simulation domain, boundary conditions and initial conguration of the rising bubble problem
due to Hysing et al. [9], length ratio x:y=1:2, gravity g, C
1
and C
2
regions of heavy and light phase,
respectively, in green free surface (; =0.5), bubble centre at (., ,) =(0.5, 0.5) and initial bubble diameter
2 r
b0
=0.5. The boundary conditions are slip (blue) and no-slip walls (red).
Table II. TC1 and TC2: Physical properties and similarity parameters.
Case j
1
j
2
j
1
j
2
g o 1e 1o Ca j
1
,j
2
j
1
,j
2
TC1 1000 100 10 1 0.98 24.5 35 10 0.286 10 10
TC2 1000 1 10 0.1 0.98 1.96 35 125 3.571 1000 100
Table III. TC1 and TC2: Simulation statistics, grid renement 1,h, number of elements
n
elements
, total number of time steps n
t
and computation time CPU on one processor
(dual-core AMD Opteron 2.4 GHz).
1,h n
elements
n
Zt
CPU TC1 CPU TC2
40 3 200 240 1 min 1 min
80 12 800 480 10 min 10 min
160 51 200 960 2.5 h 2.5 h
320 204 800 1 920 35 h 37 h
The resulting number of elements n
elements
and time steps n
Zt
along with computing times CPU
are given in Table III. The computing times are the measured wall clock times for serial simu-
lations on a single 2.4 GHz dual-core AMD Opteron processor (AMD Corporation, Sunnyvale,
California, USA). The computing times for both test cases differ only marginally and are compa-
rable with the ones given by Hysing and co-workers [9] indicating that OpenFOAM

has a similar
computing performance.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 969
3.2. Benchmark quantities
The following specic values of the benchmark quantities are used for the comparison of our
simulation results with the reference data of Hysing et al. [9]:
3.2.1. Final vertical position of the centre of mass Y
c
(t = 3). The nal position of the centre of
mass is obtained from the centre of mass position

X
c
=(X
c
, Y
c
)

X
c
=
_
D
1
\D
2
; .
c
J
_
D
1
\D
2
; J
(33)
Here, .
c
=(.
c
, ,
c
) is the centre of mass position of an individual cell in the computational mesh.
3.2.2. Minimum circularity C
min
with corresponding incidence time. The minimum circularity is
deduced from the circularity
C =
2 r
b0
1
b
(34)
where 2 r
b0
is the perimeter of the initial round bubble and 1
b
is the actual perimeter of the
deformed bubble. The actual perimeter depends strongly on the position of the interface. Never-
theless, within the VOF method, the interface is dened somewhere in the 0 < ; < 1 region.
Theoretically, this region should tend to zero; in practice, the region is spread over a few cells so
that the surface position can vary within this region. To pinpoint the surface, we chose a value
of ; =0.5.
3.2.3. Maximum rise velocity V
max
with corresponding incidence time. The maximum rise velocity
is deduced from the rise velocity V , which is evaluated for each time step as follows.
V =
_
D
1
\D
2
; J
_
D
1
\D
2
; J
(35)
Here, is the velocity in ,-direction in an individual cell in the computational mesh.
(a) domain after patching (b) domain after relaxation
Figure 2. Distribution of both phases in the computational domain of the coarsest grid (h = 1,40), yellow
corresponds to ; = 1, blue to ; = 0 and green to 0 < ; < 1, situation after patching and relaxation.
(a) Domain after patching and (b) relaxation.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
970 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
3.3. Initialisation
At the beginning of the simulations, the cell values for ; are patched to either ; =1 for the bubble
or ; = 0 for the surrounding uid. An example of the staircase prole resulting from the patch-
ing process can be seen in Figure 2(a); no cells with 0 < ; < 1 have been dened. A transient
simulation with surface tension but without gravity is employed to relax the unphysical staircase
prole of the bubble surface. The results of ; from the transient simulation with the zero gravity
condition at t =3 serve as initial values for the rising bubble simulation, see Figure 2(b). However,
after the relaxation, the interface is smeared across four cells no matter which grid spacing was used.
(a) (b)
(c) (d)
Figure 3. The nondimensional velocity eld during initialisation and zero gravity condition indicating the
patterns of the parasitic currents for the conditions of TC1. The elds are nearly symmetric. For visualisation
purposes, the pictures have been divided: the left sides show the ow patterns indicated by vector arrows,
and the right sides show the coloured magnitude of the velocity.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 971
4. RESULTS OF THE BENCHMARK TEST CASES
4.1. Zero gravity condition
After the initialisation with the zero gravity condition, some rst errors and uncertainties of the
numerical model can be estimated. Because of the action of the viscous forces, the momentum of
the ow eld should be damped out, and a nal velocity of u =0 is expected in C
1
and C
2
. Thus,
the bubble should keep its initial position

X
i
c
= (0.5, 0.5). However, small deviations from these
assumed values are found in the zero gravity simulations at t = 3. Spurious velocities are found
on both sides of the interface, which we interpret as parasitic currents. These observations are in
agreement with the ndings in, for example, [12, 3234] for a static viscous drop in equilibrium.
The shape and the magnitude of the parasitic currents are shown in Figure 3 (for the parameters of
TC1 at t = 3). For all mesh resolutions, 16 counterrotating vortices can be identied in the ow
eld of these parasitic currents. These parasitic currents arise from small deviations in the curvature
because of the implementation of the CSF model, see, for example, Lafaurie et al. [32].
In Figure 4, the inuence of c
;
on the development of max[u
]
[ (local parasitic current maximum)
over time is shown. Obviously, the parasitic currents are damped for both cases with and without sur-
face compression until a small but nite value max[u
]
[ > 0. In detail, the development of max[u
]
[
shows small dependence on the grid spacing and on c
;
. In the case of
i
=0 (c
;
=1), a convergent
behaviour for a resolution ner than h 6 1,40 is noticeable. Note that the parasitic current is pro-
portional to the kinetic energy in the system. To obey energy conservation, the kinetic energy and
thus the parasitic currents should approach zero because of the inclusion of viscous forces.
A snapshot of the pressure eld for TC1 under zero gravity condition at t = 3 is depicted in
Figure 5. The pressure is normalised by the pressure jump across the surface given by the Young
Laplace equation ^ = 2 o,r
b0
resulting in + = ,^ = r
b0
,(2o) . In the case of a static
bubble, the normalised pressure should read in C
1
: + = 0 and C
2
: + = 1 with a sharp pressure
jump at the surface. This is not the case because the values range from + = 0.83 to + = 0.70
(in C
2
) for the mesh resolution ranging from h =1,40 to h =1,320, respectively. Furthermore, at
the interface, some unphysical spikes (overshots and undershots) in the pressure eld can be found.
To reduce these spikes, we varied the interpolation scheme (least squares and cell-limited linear
schemes) [20, 21] for the V; term, which gave no improvements.
The resulting errors of the zero gravity simulations for TC1 and TC2 on the coarsest and the
nest grids are given in Table IV. The order of magnitude of Ca
]
remains nearly constant in both
test cases. No grid convergence was found. Therefore, Ca
]
~ 10
4
is determined as a measure to
estimate the inuence of the parasitic currents in our simulations.
However, the parasitic currents only have a marginal inuence on the centre of mass position
because this error is much smaller than the nominal displacement length of the parasitic currents,
(a) (b)
Figure 4. Inuence of compression ux
r
on the parasitic currents. (a) c

=0 and (b) c

=1.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
972 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
(a) (b)
(c) (d)
Figure 5. Pressure eld after relaxation for initialisation of TC1. (a) h =1,40, (b) h =1,80, (c) h =1,160
and (d) h =1,320.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 973
Table IV. The spatial errors in the zero gravity simulation at t =3 for TC1 and TC2: centre of mass position
X
c
, Y
c
and difference [^

X
c
=

X
c


X
i
c
[ with respect to the initial position

X
i
c
, maximum of the parasitic
current velocities max[u
p
(t =3)[ and their capillary number Ca
p
=j [u
p
[
max
,o.
1,h X
c
Y
c
[^

X
c
[ max[u
]
(t =3)[ Ca
TC1 40 0.500011 0.499637 3.6 10
4
0.0112 4.6 10
4
320 0.500002 0.499948 5.2 10
4
0.00522 2.1 10
4
TC2 40 0.5 0.499776 2.2 10
4
0.00252 1.3 10
4
320 0.500007 0.499908 9.0 10
5
0.00575 2.9 10
4
(a) t = 0 (b) t = 1.5 (c) t = 3
Figure 6. TC1: shapes of the rising bubble at different times. The bubble surface is indicated by a value of
; =0.5 and ne grid h =1,320. (a) t =0, (b) t =1.5 and (c) t =3.
(a) overall situation (b) detail
Figure 7. TC1: nal shape of the bubble at t = 3, volume of uid simulations on the ne h = 1,320 (red)
and the coarse grids h = 1,40 (blue) and reference solutions by Hysing et al. [9] black (TP2D) and grey
(FreeLIFE). (a) Overall situation and (b) detail.
[^

X
c
[ max[u
]
(t = 3)[ t . The effects of [^

X
c
[ and max[u
]
(t = 3)[ on the results of the rising
bubble simulations are discussed later.
4.2. TC1
4.2.1. Simulation results. In Figure 6, typical results of the computed interface of the rising bub-
ble for TC1 are shown. As illustrated for the stages (t = 0, t = 1.5 and t = 3), the bubble gets
deformed, while staying compact as one connected region throughout the simulation. This is due to
the low viscosity and density ratios and low 1o and Ca numbers indicating the high inuence of
the surface tension force.
Figure 7 compares the bubble shape at the nal time (t = 3) in our simulations for the
coarsest (h = 1,40) and the nest grids (h = 1,320) with the corresponding results for the codes
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
974 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
(a) complete simulation period (b) detail with benchmark point
Figure 8. TC1: temporal evolution of centre of mass position Y
c
, volume of uid simulations on grids
h = 1,40, 80, 160, 320| (coloured) and reference solutions from Hysing et al. [9] in black (TP2D) and
grey (FreeLIFE). (a) Complete simulation period and (b) detail with benchmark point Y
c
(t =3).
(a) complete simulation period (b) detail with benchmark point
Figure 9. TC1: temporal evolution of circularity C, volume of uid simulations on grids h =
1,40, 80, 160, 320| (coloured) and reference solutions in black (TP2D) and grey (FreeLIFE). (a) Complete
simulation period and (b) detail with benchmark point C
min
(t
min
).
TP2D/MooNMD and FreeLIFE
||
given by Hysing et al. [9]. Note that the results of the TP2D and
the MooNMD codes are indistinguishable for TC1; thus, we use only TP2D to refer both of them
in this test case. Obviously, the VOF solutions on the coarsest and the nest grids are very close to
each other, but there is a small but noticeable difference to the nal bubble shape of the reference
results, see Figure 7 (left). A detailed view at a typical section reveals that congruence between
the VOF simulations and the reference results seems to be unreachable, not even with much ner
grid resolutions.
Figures 810 illustrate the temporal development of the quantities centre of mass position Y
c
,
circularity C and rising velocity V from which the benchmark quantities are deduced. Again, the
results of the VOF simulations for the four-grid resolutions h =1,40, 80, 160, 320| are compared
with the corresponding results of the codes TP2D and FreeLIFE given by Hysing et al. [9].
The temporal evolution of Y
c
in our simulations is nearly identical to the reference results, see
Figure 8 (left), but the close-up showing the range t =2.53 in Figure 8 (right) shows that complete
convergence towards the benchmark results is again not achieved. On the contrary, the temporal evo-
lution of C in Figure 9 (left) and V in Figure 10 (left) shows more apparent differences between our
ndings and the corresponding benchmark data. Here, the close-ups near the minimum circularity
C
min
in Figure 9 (right) and the maximum rising velocity V
max
in Figure 10 (right) reveal that the
||
The reference results have the following grid resolutions h D1{640 (TP2D) and h D1{160 (FreeLIFE) respectively.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 975
(a) complete simulation period (b) detail with benchmark point
Figure 10. TC1: temporal evolution of rising velocity V , volume of uid simulations on grids h =
1,40, 80, 160, 320| (coloured) and reference solutions in black (TP2D) and grey (FreeLIFE). (a) Complete
simulation period and (b) detail with benchmark point V
max
(t
max
).
Table V. TC1: the benchmark quantities minimum circularity C
min
with corresponding incidence times
t (C
min
), maximum rising velocity V
max
with t (V
max
) and nal position Y
c
of centre of mass at t = 3.
1,h 40 80 160 320 Benchmark Hysing et al. [9] trubelj et al. [12]
C
min
0.8985 0.8966 0.8999 0.9044 0.9012 0.0001 0.8876
t (C
min
) 1.8500 1.9625 1.9375 1.9625 1.9 1.8915
V
max
0.2364 0.2373 0.2365 0.2348 0.2419 0.0002 0.2457
t (V
max
) 0.9250 0.9250 0.9219 0.9516 0.927 0.011 0.9235
Y
c
(t =3) 1.0616 1.0655 1.0668 1.0696 1.081 0.001 1.0679
results of our simulations do not show any convergence towards the reference. Instead of this result,
our simulations scatter slightly around the results of the benchmark.
Table V summarises the benchmark quantities for TC1. Minimum circularity C
min
and maximum
rising velocity V
max
, both with the corresponding incidence times t (C
min
) and t (V
max
), respectively,
and the nal position Y
c
of the centre of mass at t = 3 for TC1 are given. For convenience, the
results of trubelj et al. [12] for TC1 are also included. The values of h =1,320 serve as the basis.
The maximum relative errors in our simulations for the grids h = 1,40, 80, 160| are about 1%
for C
min
, V
max
and Y
c
(t = 3), 3% for t (V
max
) and 6% for t (C
min
). Comparing our results of grid
h = 1,320 with the benchmark data of Hysing et al. [9], the relative errors are in the order of 1%
for C
min
and Y
c
(t =3) and 3% for V
max
, t (V
max
) and t (C
min
).
As explained in the previous section, the parasitic currents induce some uncertainties in the VOF
simulations. From the ndings of the zero gravity simulations for TC1, we estimate the magnitude
of the parasitic currents to be in the order of u
1
~0.01. Therefore, we expect a relative uncertainty
in the ow eld near the bubble interface in the order of u
1
,[V [ ~ 0.01,0.25 = 4%. Indeed, this
is an upper limit of all the relative errors between our data and the benchmark results. So, we con-
clude that the performance of our VOF model for TC1 is mainly dominated by the existence of the
parasitic currents.
4.2.2. Inuence of the model parameters.
Compression ux. For comparison, some simulations with zero compression ux
i
=0 have been
carried out to reveal its inuence on the sharpness of the surface and the temporal evolution of
the circularity.
In Figure 11, the inuence of
i
on the surface is depicted. Two grid resolutions (h =
1,320 and h = 1,40) are investigated. For both grid resolutions, a clear smearing of the surface
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
976 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
(a) = 1/40
(b) = 1/320
Figure 11. Inuence of compression ux
r
on the surface,
r
= 0 (left) and
r
= 1 (right), for two
different grid instances: (a) h =1,40 and (b) h =1,320.
Figure 12. Inuence of compression term on C.
can be identied in the case of
i
=0. In the case of
i
=1, the smearing is clearly reduced for the
coarse (h =1,40), and for the ne grid (h =1,320), it is invisible.
The inuence on the evolution of C shows exemplarily the inuence of the compression ux,
see Figure 12. In the case of the coarse mesh (h = 1,40), the case of
i
= 0 leads to a diverging
behaviour at t = 3. For the ne mesh case (h = 1,320), a larger discrepancy to the reference solu-
tions between t =1 and t =2.5 can be distinguished, whereas for t >2.5, the circularity converges
towards the reference solutions.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 977
(a) Crank-Nicolson scheme (b) backward scheme
Figure 13. Inuence of time step and time discretisation scheme on the evolution of C for the grid
h =1,160. (a) CrankNicolson and (b) backward schemes.
Figure 14. Inuence of the momentum predictor on the evolution of C.
Time schemes and time step. The time step (Jt = 1,160, 1,320, 1,640 ) and the time discretisa-
tion scheme (CrankNicolson and backward) have been varied for TC1, with h = 1,160 to check
for their inuence. The temporal evolution of C was chosen exemplarily to show the negligible
inuence of these variations, see Figure 13. The choice of the time discretisation scheme had no
inuence on the evolution of C.
Momentum predictor. In Figure 14, the inuence of the momentum predictor step (see
Section 2.2.3) on the temporal evolution of C for the case TC1, h = 1,40 and 1,320 is depicted.
In the low-resolution case (h = 1,40), inuence of an additional momentum predictor step
(momPred = 1) on the temporal evolution is negligible. For the ner resolution (h = 1,320),
we see that omitting the momentum predictor leads to larger deviations; hence, all simulations are
carried out with the additional momentum predictor step.
4.3. TC2
Figure 15 shows typical results for the rising bubble in TC2 at three different stages (t = 0, t =
1.5 and t =3). The distortion of the interface in TC2 is considerably larger than in TC1. The bubble
is deformed to a skirted bubble and eventually forms thin laments with smaller satellite bubbles
that are detached. Compared with TC1, lower surface tension forces indicated by the higher 1o
and Ca numbers as well as higher viscosity and density ratios are the reasons for the formation
of a skirted bubble. The correct simulation of these physics is especially challenging as there is
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
978 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
(a) = 0 (b) = 1.5 (c) = 3
Figure 15. TC2: shapes of the rising bubble at different times, bubble surface indicated by a value of ; =0.5
and simulation on grid h =1,320. (a) t =0, (b) t =1.5 and t =3.
(a) overall situation (b) in detail
Figure 16. TC2: nal shape of the bubble at t = 3, volume of uid simulations on ne h = 1,320 (red)
and coarse grids h = 1,40 (blue) and reference solutions from Hysing et al. [9] in black (TP2D) and grey
(FreeLIFE). (a) Overall situation and (b) in detail.
(a) complete simulation period (b) detail with benchmark point
Figure 17. TC2: temporal evolution of centre of mass position Y
c
, volume of uid simulations on grids
h = 1,40, 80, 160, 320| (coloured) and reference solutions from Hysing et al. [9] in black (TP2D) and
grey (FreeLIFE). (a) Complete simulation period and (b) detail with benchmark point Y
c
(t
3
).
no unique numerical result for t > 2 available. For Y
C
and V , the difference is less pronounced
as for C.
In Figure 16, the bubble shapes at the nal time (t = 3) of our simulations for the coarsest
(h = 1,40) and the nest grids (h = 1,320) with the corresponding results of the codes TP2D
and FreeLIFE are compared. Note that we omit the results of MooNMD code because the applied
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 979
(a) complete simulation period (b) detail with benchmark point
Figure 18. TC2: temporal evolution of circularity C, volume of uid simulations on grids h =
1,40, 80, 160, 320| (coloured) and reference solutions in black (TP2D) and grey (FreeLIFE). (a) Complete
simulation period and (b) detail with benchmark point C
min
(t
min
).
(a) complete simulation period (b) with benchmark points
Figure 19. TC2: temporal evolution of rising velocity V , volume of uid simulations on grids h =
1,40, 80, 160, 320| (coloured) and reference solutions in black (TP2D) and grey (FreeLIFE). (a) Complete
simulation period and (b) with benchmark points V
max1
(t
max1
), V
max2
(t
max2
).
arbitrary LagrangianEulerian method does not allow for bubble breakup [9]. However, our solu-
tions on the nest grid are very close to the nal bubble shape of the reference calculations, see
Figure 16 (left). The close-up at a range between . = 0.65 and . = 0.75 reveals that congru-
ence between our VOF simulations and the reference result of FreeLIFE is achieved at this grid
resolution, see Figure 16 (right).
The temporal development of the quantities Y
c
, C and V is shown in Figures 1719.
Similar to TC1, the evolution of Y
c
in our simulations is nearly identical to the reference results,
see Figure 17 (left). The close-up of Figure 17 (right) conrms the good agreement between our
results on grid h =1,320 and the reference results; even so, convergence, that is a grid-independent
solution, is again not achieved.
Contrary to the ndings for TC1, the evolution of C in Figure 18 (left) and V in Figure 19 (left) of
our simulations are also closer to the corresponding results of FreeLIFE. Please note the differences
in the benchmark results: in TP2D, there is a separation of satellite bubbles, whereas in FreeLIFE,
the bubble remains skirted with elongated laments. This difference between TP2D and FreeLIFE
is documented by the very different temporal evolution of C.
In our simulations, the rising bubble also remains skirted, although the separation of the satellite
bubbles is found close to the nal situation at t = 3. The close-ups near the minimum circularity
C
min
in Figure 18 (right) and the maximum rising velocities V
max1
and V
max2
in Figure 19 (right)
conrm that the results of our simulations approximate the results of FreeLIFE, although complete
convergence is again not achieved.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
980 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
Table VI. TC2: the benchmark quantities minimum circularity C
min
with corresponding incidence times
t (C
min
), maximum rising velocities V
max1/2
with t (V
max1/2
) and nal position Y
c
of centre of mass at t =3.
Benchmark
1,h 40 80 160 320 (FreeLIFE) Hysing et al. [9]
C
min
0.5000 0.4999 0.5051 0.4945 0.4647
t (C
min
) 3.0000 2.8125 2.9875 2.9500 3.0000
V
max1
0.2388 0.2441 0.2431 0.2474 0.2514
t (V
max1
) 0.6875 0.7000 0.7250 0.7156 0.7281
V
max2
0.2134 0.2240 0.2302 0.2353 0.2440
t (V
max2
) 1.7000 1.8688 1.9594 2.0047 1.9844
,
c
(t =3) 1.0862 1.1014 1.1089 1.1218 1.1249
Table VI summarises the benchmark quantities minimum circularity C
min
and maximum ris-
ing velocities V
max1
, V
max2
, with the corresponding incidence times t (C
min
), t (V
max1
) and t (V
max2
),
respectively, and the nal position Y
c
of the centre of mass at t = 3 for TC2. The relative errors of
our simulations decrease for an increasing grid resolution
**
up to the grid resolution of h =1,160,
if h =1,320 is taken as a basis. Nevertheless, complete convergence has not been achieved with the
nest grid. Comparing our results of grid h = 1,320 with the benchmark data of FreeLIFE in [9],
the relative errors are below 1% for Y
c
(t =3), around 2% for t (C
min
), V
max1
, t (V
max1
) and t (V
max2
),
around 4% for V
max2
and around 6% for C
min
.
From the ndings of the zero gravity simulations for TC2, we estimate the magnitude of the para-
sitic currents to be in the order of u
1
~0.005. Therefore, we expect a relative uncertainty in the ow
eld near the bubble interface in the order of u
1
,[V [ ~ 0.005,0.25 = 2%. We see that this value
is again an upper limit of most of the relative errors between our data and the benchmark results.
Similar to TC1, we conclude that the performance of our VOF model for TC2 is also dominated by
the existence of the parasitic currents.
5. CONCLUSIONS
We present a numerical model for the rising bubble problem. The numerical model is based on the
volume of uid method. The CSF approach is employed to describe the inuence of the surface
tension. Surface compression is used to sharpen the resolution of the bubble interface. The model
equations are solved with the open-source CFD library OpenFOAM

. The implementation of the


surface compression algorithm into OpenFOAM

is described in detail.
Two different realisations of the rising bubble problem are investigated with the numerical model
with four different grid resolutions. Both test cases differ in the Reynolds number 1e, the Etvs
number 1o, the density and the viscosity ratios. Therefore, the nal shape of the rising bubbles is
ellipsoidal in one and skirted in the other test case. These two test cases have been also investigated
by several other groups in the past. Their data for the circularity, the centre of mass and the rising
velocity of the rising bubbles, serve as benchmark values for the VOF simulations.
The comparison of our results with the benchmark data shows that the proposed model can
resolve the bubble behaviour in reasonable agreement with the benchmark. In the case of the ellip-
soidal bubble (TC1), our results deviate from the reference results of the benchmark data. It is very
important to note that TC1 results do not show any grid convergence. This is especially critical
because a grid-converged solution is commonly used if there is no benchmark (or experimental)
result available.
In the case of the skirted bubble (TC2), physics and the associated bubble shape are more com-
plex. Here, correct simulation of the formation of the shape and detachment of satellite bubbles is
**
Nevertheless, the relative error is not decreasing monotonously, see, for example, V
max1
(h D 1{S0 and h D 1{160).
A reason for this might be the nonlinear temporal evolution of the parasitic currents. Here, much more research
is necessary.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
NUMERICAL SIMULATION OF A SINGLE RISING BUBBLE BY VOF WITH SURFACE COMPRESSION 981
especially critical and is still a challenge. As seen in the paper by Hysing et al. [9], a unique bench-
mark solution does not exist for TC2. For this test case, we were able to show grid convergence
competing with the results given by [9].
In both cases, parasitic currents are observed in the ow eld. These currents result from the
numerical realisation of the CSF approach. We show that the magnitude of the parasitic currents
denes an upper measure for the uncertainty of the VOF simulations. The presented VOF approach
shows deciencies with cases that are surface force dominated (Ca < 1) as was the case for TC1.
The study of zkan et al.[19] indicates that such problems occur especially for VOF methods that
solve the phase transport equations directly (direct methods). However, we did not experience these
problems, such as the mentioned solver divergence problems. Although in our study, the capillary
number is one order of magnitude higher, the deviations are much smaller compared with their
study. We conclude that the surface compression approach is a more improved method compared
with high-resolution interface-capturing scheme approach investigated by zkan et al. [19].
The advantages of the presented VOF approachare are
1. boundedness of the volume fraction 0 6; 61,
2. mass conservation without any reinitialisation step (as in LSs),
3. maintenance of sharp interfaces through the surface compression approach,
4. the surface compression approach gives better results compared to the ones without
surface compression.
Nevertheless, on the basis of the results presented here, we propose that the used surface com-
pression approach should be further improved in future versions of the VOF solver. It remains an
open question whether the source of these deviations for surface-tension-dominated ows comes
from the direct VOF method, from the CSF method or from the interaction of both methods. For
further improvements, we suggest the following:
1. The implementation of the CSF approach should be rened to minimise the parasitic currents.
Or alternative methods, such as the continuous surface stress method, should be tested.
2. The improvement of the determination of the curvature (approximation of the interface
position) and the correct determination of the pressure jump needs to be addressed.
Here, a reconstruction method, for example, moments of uid method [17, 18] could give
some improvements.
The results presented in this paper should raise the awareness to use the direct VOF method,
currently implemented in OpenFOAM

version 1.5.1 through version 2.1.0 [20, 21], cautiously


for surface-tension-dominated ows (e.g. free surface liquid metal ows). Here, we recommend
careful validation with appropriate reference solutions (results from experiments and published
benchmark results).
ACKNOWLEDGEMENTS
This work presents a part of the results obtained under contract number SFB 799 C1 sponsored by the
Deutsche Forschungsgemeinschaft (DFG), Germany.
REFERENCES
1. Maiwald A, Scheller P, Brcker C, Schwarze R. Flow-induced emulsication of mold powder slag into liquid steel.
STEELSIM2009 3rd International Conference Simulation and Modelling of Metallurgical Processes in Steelmaking,
Leoben, 8.-10. September, 2009; 162167.
2. Harlow F, Welch J. Numerical calculation of time-dependent viscous incompressible ow of uid with free surface.
Physics of Fluids 1965; 8(12):21822189.
3. Hirt CW, Nichols BD. Volume of uid (VOF) method for the dynamics of free boundaries. Journal of Computational
Physics 1981; 39:201225.
4. Osher S, Sethian JA. Fronts propagating with curvature-dependent speedalgorithms based on HamiltonJacobi
formulations. Journal of Computational Physics 1988; 79(1):1249.
5. Sussman M, Smereka P, Osher S. A level set approach for computing solutions to incompressible two-phase ow.
Journal of Computational Physics 1994; 114(1):146159.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d
982 J. KLOSTERMANN, K. SCHAAKE AND R. SCHWARZE
6. Rider WJ, tracking K, DB. Reconstructing volume. Journal of Computational Physics 1998; 141(2):112152.
7. Scardovelli R, Zaleski S. Direct numerical simulation of free-surface and interfacial ow. Annual Review of Fluid
Mechanics Jan 1999; 31(1):567603.
8. Sethian J, Smereka P. Level set methods for uid interfaces. Annual Review of Fluid Mechanics 2003; 35:341372.
9. Hysing S, Turek S, Kuzmin D, Parolini N, Burman E, Ganesan S, Tobiska L. Quantitative benchmark computations
of two-dimensional bubble dynamics. International Journal for Numerical Methods in Fluids 2008; 60:12591288.
10. Ferziger JH, Peri c M. Computational Methods for Fluid Dynamics. Springer: Berlin, Germany, 2002.
11. Yeoh GH, Barber T. Assessment of interface capturing methods in computational uid dynamics (CFD) codesa
case study. The Journal of Computational Multiphase Flows Jun 2009; 1(2):201215.
12. trubelj L, Tiselj I, Mavko B. Simulations of free surface ows with implementation of surface tension and interface
sharpening in the two-uid model. International Journal of Heat and Fluid Flow 2009; 30:741750.
13. Ubbink O. Numerical prediction of two uid systems with sharp interfaces. PhD Thesis, Imperial College, University
of London, 1997.
14. Muzaferija S, Peric M. Computation of free-surface ows using the nite-volume method and moving grids.
Numerical Heat Transfer Part B-Fundamentals December 1997; 32(4):369384.
15. Youngs D. Time-dependent Multi-material ow with Large Fluid Distortion. Academic Press Inc:
(London) Ltd, 1982. http://www.scopus.com/inward/record.url?eid=2-s2.0-0019936463&partnerID=40&md5=
f1dbae4707c3493b06e31cc4da893bb7, cited By (since 1996) 19.
16. Popinet S. An accurate adaptive solver for surface-tension-driven interfacial ows. Journal of Computational Physics
2009; 228(16):58385866. http://www.scopus.com/inward/record.url?eid=2-s2.0-67649472402&partnerID=
40&md5=6fcd38f15413e444be2e8a6c13465c45, cited By (since 1996) 35.
17. Dyadechko V, Shashkov M. Moment-of-uid interface reconstruction. Technical Report, Los Alamos National
Laboratoty, P.O. Box 1663, Los Alamos, NM 87545, 2005.
18. Dyadechko V, Shashkov M. Reconstruction of multi-material interfaces from moment data. Journal of Computa-
tional Physics 2008; 227(11):53615384. http://www.sciencedirect.com/science/article/pii/S0021999107005748.
19. zkan F, Wrner M, Wenka A, Soyhan H. Critical evaluation of CFD codes for interfacial simulation of bubble-
train ow in a narrow channel. International Journal for Numerical Methods in Fluids 2007; 55(6):537564.
DOI: 10.1002/d.1468. http://www.scopus.com/inward/record.url?eid=2-s2.0-35248859711&partnerID=40&md5=
b1e4b01a78a0402c575caf2d0f551eae, cited By (since 1996) 10.
20. Jasak H, Gschaider B, Nilson H, Beaudoin M, et al. Openfoam-1.5-dev extend project on sourceforge. http:
//openfoam-extend.svn.sourceforge.net/viewvc/openfoam-extend/trunk/Core/OpenFOAM-1.5-dev/;2009.
21. OpenCFD Limited. User guide openfoam 1.6. Technical Report. http://www.opencfd.co.uk/openfoam;2009.
22. Brackbill JU, Kothe DB, Zemach C. A continuum method for modeling surface tension. Journal of Computational
Physics 1992; 100(2):335354.
23. Rusche H. Computational uid dynamics of dispersed two-phase ows at high phase fractions. PhD Thesis, Imperial
College of Science, Technology and Medicine, London, England, 2002.
24. de M, P B R. Study and numerical simulation of sediment transport in free-surface ow. PhD Thesis, University of
Mlaga, July 2008.
25. Rhie CM, Chow LW. Numerical study of the turbulent ow past an airfoil with trailing edge separation. AIAA 1983;
21:15251532. http://ci.nii.ac.jp/naid/80001467460/en/.
26. Issa RI. Solution of the implicitly discretised uid ow equations by operator-splitting. Journal of Computational
Physics 1986; 62(1):4065.
27. LeVeque RL. Finite Volume Methodes for Hyperbolic Problems. Cambridge University Press: Cambridge, UK, 2002.
28. Hirsch C. Numerical Computation of Internal and External Flows: The Fundamentals of Computational Fluid
Dynamics, 2nd ed. Butterworth-Heinemann: Oxford, UK, 2007.
29. Crank J, Nicolson P. A practical method for numerical evaluation of solutions of partial differential equations of the
heat-conduction type. Advances in Computational Mathematics December 1996; 6(1):207226. http://dx.doi.org/10.
1007/BF02127704.
30. van L, B. Towards the ultimate conservative difference scheme. IV. monotonicity andconservation combined in a
second order scheme. Journal of Computational Physics 1974; 14:361370.
31. Jasak H. Error analysis and estimation for the nite volume method with applications to uid ows. PhD Thesis,
Imperial College, University of London, 1996.
32. Lafaurie B, Nardone C, Scardovelli R, Zaleski S, Zanetti G. Modelling merging and fragmentation in multiphase
ows with surfer. Journal of Computational Physics 1994; 113(113):134147.
33. Harvie D, Davidson M, Rudman M. An analysis of parasitic current generation in volume of uid simula-
tions. Applied Mathematics Model October 2006; 30(10):10561066. http://www.sciencedirect.com/science/article/
B6TYC-4HC6M40-2/2/aba2563fc39451c85a3adf331f9c1816.
34. Sussman M, Smith KM, Hussaini MY, Ohta M, Zhi-Wei R. A sharp interface method for incompressible two-phase
ows. Journal of Computational Physics 2007; 221(2):469505.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2013; 71:960982
DOI: 10.1002/d

You might also like