Sunteți pe pagina 1din 19

Numerical Heat Transfer, Part B: Fundamentals

An International Journal of Computation and Methodology

ISSN: 1040-7790 (Print) 1521-0626 (Online) Journal homepage: http://www.tandfonline.com/loi/unhb20

A consistent balanced force refined moment-of-


fluid method for surface tension dominant two-
phase flows

H. K. Zinjala & J. Banerjee

To cite this article: H. K. Zinjala & J. Banerjee (2018): A consistent balanced force refined
moment-of-fluid method for surface tension dominant two-phase flows, Numerical Heat Transfer,
Part B: Fundamentals, DOI: 10.1080/10407790.2018.1495423

To link to this article: https://doi.org/10.1080/10407790.2018.1495423

Published online: 16 Oct 2018.

Submit your article to this journal

Article views: 7

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=unhb20
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS
https://doi.org/10.1080/10407790.2018.1495423

A consistent balanced force refined moment-of-fluid method


for surface tension dominant two-phase flows
H. K. Zinjala and J. Banerjee
Mechanical Engineering Department, Sardar Vallabhbhai National Institute of Technology, Surat, India

ABSTRACT ARTICLE HISTORY


Inconsistent implementation of surface tension in an interfacial solver for Received 8 May 2018
incompressible two-phase flow results in spurious currents near the inter- Accepted 27 June 2018
face. The balanced force algorithm (BFA) enforces consistent implementa-
tion of surface tension force and pressure gradient at the cell faces
thereby generating conservative face center velocity field, even though the
cell center velocity may not be divergence free and spurious current free.
The implication of this inconsistency is directly reflected in interface
smoothness while using a Lagrangian–Eulerian interface evolution in con-
trary to Eulerian advection where divergence free face velocities are used.
This paper describes a consistent balanced force algorithm (CBFA) in collo-
cated grid for consistent implementation of surface tension force. The
interface evolution is tracked by refined moment-of-fluid (RMOF) method
using Lagrangian–Eulerian advection scheme. In CBFA, additional Poisson
equation is solved to obtain cell face corrections for conservative cell cen-
ter velocity computation. The effectiveness of the proposed method is
demonstrated in this article for several test cases.

1. Motivation and introduction


Accurate material interface representation is crucial for two-phase flow computation. In volume-
of-fluid method (VOF) [1], the interface is captured using volume fraction field. The VOF
method involves interface reconstruction and volume fraction advection. Piecewise Linear
Interface Calculation (PLIC) method proposed by Youngs [2] is a popular method for interface
reconstruction. In PLIC method, the interface is assumed to be a line in 2D or a plane in 3D and
the interface is advected by the volume fraction field. Traditionally, volume fraction field is
advected by Eulerian advection scheme [3–10] where volume fraction equation (which is a pure
advection equation) is solved using the face fluxes in a Eulerian grid. Recently, Zinjala and
Banerjee [11] reported a Lagrangian–Eulerian advection scheme (LEAS) for volume fraction
advection. Lagrangian–Eulerian advection is performed in three steps namely back tracking of cell
vertices along the stream lines passing through the cell vertices, volume fraction remapping, and
volume fraction repair. The implementation details for each step of LEAS ares described in [11].
In [11], it is inferred that LEAS performs comparable with published Eulerian schemes. Further
improvement in accuracy of volume tracking can be obtained by improving interface reconstruc-
tion rather than volume fraction advection. In [12], moment-of-fluid (MOF) interface reconstruc-
tion [13] is coupled with LEAS. MOF method is an extension of existing VOF methods. In MOF
interface reconstruction, in addition to volume fraction, material centroid is incorporated for

CONTACT J. Banerjee jbaner@med.svnit.ac.in Department of Mechanical Engineering, Sardar Vallabhbhai National


Institute of Technology, Surat, 395007, India
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/unhb.
ß 2018 Taylor & Francis Group, LLC
2 H. K. ZINJALA AND J. BANERJEE

Nomenclature
Roman symbols Superscripts
f volume fraction n; n þ 1 old and new time levels
t time a advected velocity
u velocity vector  predicted quantity
x Centroid of the material ref reference material
XC computational cell
XL Lagrangian precell Subscripts
r surface tension coefficient ref reference material
d Dirac delta function c cell center qunatity
j curvature f face centre quantity
l viscosity nb neighbor cell qunatity
q density f !c face to cell center interpolation

interface reconstruction. Physically, material centroid represents the location of the material
inside a region. Therefore, interface reconstruction based on centroid is more accurate compared
to its VOF counterpart. This is demonstrated in [13]. However, MOF is highly sensitive to cen-
troid location. Slight error in centroid location may lead to wrong prediction of interface normal
and hence the quality of interface reconstruction deteriorates. Particularly, in dynamic interface
reconstruction, these errors accumulate with time and generate wisps inside pure material region
[12]. In [12], an accurate centroid advection scheme based on barycenter is presented. It was
found in [12] that barycenter based centroid advection is more accurate compared to
Runge–Kutta integrator. This is because, in the Lagrangian motion, barycentric coordinates of the
point remain constant. This fact is utilized for centroid advection. Standard advection tests reveal
that barycenter based centroid advection performs more accurately than Runge–Kutta integrator
for prescribed velocity field. In real flow problems where velocity field is obtained from an inter-
facial flow solver, MOF performs poorly if the velocity field contains spurious currents.
In addition to property jump across material interface, surface tension plays a vital role in
two-phase flow. The surface tension force is present only at the interface between the two-fluids.
Due to this singular nature of surface tension force, discretization is a challenge. Surface tension
modeling is thus a key element for robustness and accuracy of the two-phase flow solver. There
are two approaches for incorporating surface tension force in momentum equation. In the first
approach, the surface tension force is modeled by continuum surface force (CSF) of Brackbill
et al. [14]. In the CSF approach, the surface tension force is assumed to be continuous across the
interface. Brackbill et al. [14] proposed a Navier–Stokes solver with surface tension force discre-
tized at cell center and interface captured using the VOF method. The curvatures are computed
at cell centers by convolution of volume fraction field. This cell centered discretization of surface
tension force term in momentum equation does not balance the pressure gradient. Since pressure
gradient is discretized at cell faces, surface tension must be discretized at cell faces. This formula-
tion is known as balanced force algorithm (BFA) [15]. In BFA, surface tension term is discretized
at the same location where pressure gradient is discretized with the same numerical operator.
Thus, in BFA, the surface tension force is balanced by the pressure gradient. In [16], the authors
have extended the BFA of [15] to arbitrary meshes with a compressive scheme for volume frac-
tion advection in fully-coupled flow solver on collocated grid arrangement. This balanced-force
VOF framework provides a strong pressure–velocity coupling and ensures a discrete balance
among the pressure gradient, surface tension and gravity force. Waters et al. [17] presented the
implementation of surface tension in a finite element method (FEM). Since volume fraction near
the interfacial region is smooth in FEM representation, surface tension estimation is straight for-
ward. In the second approach, the surface tension force is applied at the interface as jump
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 3

condition. Fedkiw [18] proposed Ghost Fluid Method to incorporate surface tension force such
that surface tension force is balanced by pressure field in a sharp manner. Jump conditions of
pressure across interface were applied directly in the discretization of pressure gradient term in
momentum equation using Level set field.
Typically, face velocity computed in a flow solver is conservative since pressure gradient and
surface tension are discretized at cell faces. Therefore, the interface advection scheme based on
Eulerian discretization, which uses face velocities, works well even though spurious currents are
present near the interface [19,20]. Cao et al. [21] used Lagrangian–Eulerian advection for volume
fraction by combining with level set method. In their approach, the level set function is used to
compute interface curvature. The LEAS for interface evolution uses cell corner velocity for back
tracking and it suffers due to velocity interpolation from cell centers. The situation becomes diffi-
cult when the MOF method is used for interface reconstruction since the MOF is not only
dependent on volume fraction but also on material centroids. This difficulty is overcome in the
present paper by incorporating consistent projection method of Ming-Jiu Ni [22]. In the consist-
ent projection method, cell centered velocity is computed by conservative face velocities. Linear
interpolation of face velocities to compute cell center velocity introduces numerical dissipation in
the velocity field. This dissipation is regarded here as a new face velocity. These face velocities are
used to compute new face fluxes which are not divergence free. In order to make the new face
velocities divergence free, additional pressure field is established by solving a pressure Poisson
equation. Thus, in consistent projection method, the pressure is evaluated in two parts. The first
part is based on predicted velocity and surface tension force while the second part is based on
newly constructed velocity fluxes. The final conservative cell center velocity is computed by linear
averaging of cell face velocity and new conservative cell face velocity. In [22], it was demonstrated
that the consistent projection method shows reduced spurious currents when surface tension is
considered in context to Level set method.
In the present article, a consistent balanced force refined moment-of-fluid (CBFA-RMOF)
method is proposed for incompressible two-phase flow computation. In consistent formulation
presented here, the interface is tracked by RMOF method. The performance of CBFA-RMOF
method is demonstrated through series of test problems like inviscid, viscous equilibrium drop,
and bubble rise simulations in buoyancy driven field.
This paper is organized as follows: Section 2 discusses governing equations for incompressible
two-phase flow. The discretization of governing equations and numerical methodology utilized for
each term is detailed in Section 3. Section 4 focuses on numerical results obtained by comparing each
algorithm discussed in Section 3. The conclusions derived from the results are presented in Section 5.

2. Governing equations
The governing equations of motion for incompressible two-phase flows are written as follows:
Continuity equation:
ru¼0 (1)
and momentum equation
@u 1 1   1
þ r  ðuuÞ ¼  rp þ r  lðru þ ruT Þ þ Fg^j (2)
@t q q q
where u is the velocity vector, p is the pressure, F is the surface tension force, and g is gravita-
tional acceleration. The density and viscosity are computed from the volume fraction as follows:
q ¼ q1 f þ ð1f Þq2 (3)
4 H. K. ZINJALA AND J. BANERJEE

l ¼ l1 f þ ð1f Þl2 (4)


Here the subscript 1 and 2 are material 1 and material 2, respectively. The volume fraction of
pure cell is either f ¼ 0 or f ¼ 1 depending on the material present inside the cell. For mixed cells,
the volume fraction value lies in between 0 and 1, 0 < f < 1.

3. Numerical methodology
In this section, brief description of interface evolution based on RMOF method is presented first.
This is followed by a discussion on coupling of RMOF with interfacial flow solver. The BFA for
incompressible two-phase flow computation is described along with the implementation details.
Finally, consistent balanced force algorithm (CBFA) is derived from the standard BFA.

3.1. Refined moment-of-fluid method


In the present paper, interface evolution is tracked by RMOF method [23] in which all interfacial
cells are refined to a fixed level of refinement for interface reconstruction and advection. The
grid for the flow solver is not refined. Since both the grids are separate, refinement for interface
can be arbitrary in order to ensure grid converged interface. Most computationally expensive
component of MOF method is the interface reconstruction. Thus an upper limit for refinement is
specified in RMOF as opposed to adaptive mesh refinement moment-of-fluid (AMR-MOF)
method in which refinement proceed to upper limit as discussed in detail in [23]. The intermedi-
ate interface reconstructions in AMR-MOF are time consuming [23] which is not the case with
RMOF. Therefore, RMOF is less time consuming as compared to AMR-MOF.

3.1.1. MOF interface reconstruction


The MOF interface reconstruction method is an extension of VOF methods. In MOF interface
reconstruction, in addition to material volume fractions, fref, (zeroth moment of material), mater-
ial centroids, xref (first moment of material) is used for obtaining the interface in a cell. Since
centroid location is the average location of the material inside the cell, MOF method is second
order accurate. Thus linear interfaces are exactly reconstructed by MOF method. The MOF inter-
face reconstruction problem can be stated as follows: “for given material moments, reconstruct
material interface such that reconstructed volume fraction is exactly conserved while recon-
structed centroid is the best possible approximation to reference material centroid.” Eventually
the MOF reconstruction problem can be cast as a constrained optimization problem given in
[12]. It is to be noted that in [12], reference centroid alone is not approximated, both centroids,
i.e., centroids of material 1 and material 2, are used for interface reconstruction as shown in
Figure 1. This strategy is termed as symmetric MOF method [24] in which the effect of both
centroids in the interface reconstruction is considered in a pairwise manner. The optimization
problem is solved using multi-dimensional BFGS method. The implementation details can be
found in [12].

3.1.2. Material moments advection for refined mesh


LEAS presented in [11,12] is used in the present work for material moments advection. A similar
scheme is used in [25] for AMR-MOF method and further used in [23] for RMOF. Typical steps
in any LEAS are as follows:

1. Lagrangian back tracking of cell vertices.


NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 5

Figure 1. Moment-of-fluid interface reconstruction.

Figure 2. Material moments advection.

2. Material remapping (Polygon intersection).


3. Forward tracking of material centroids i.e., Material centroid advection.

In the case of localized refinement as in AMR-MOF or RMOF, the two neighbor cells, which
have different level of refinement, are not well connected when they are back tracked. Due to this
uneven refinement of mesh, there are overlaps or gaps in the back tracked cells. This leads to
deviation of total volume of the reference material. This require an additional volume repair algo-
rithm in addition to the standard volume fraction repair algorithm [12,26]. In the present work,
the advection scheme presented in [23] is modified to eliminate volume error due to simplified
6 H. K. ZINJALA AND J. BANERJEE

back tracking. In the modified scheme, each possible mixed cell is back tracked, similar to the
advection scheme for unrefined mesh, to obtain Lagrangian precell, XL of Eulerian cell, Xc as
shown in Figure 2. This is possible since the change in velocity field between two time steps is
small in an interfacial flow solver. The material volume fraction and centroids are computed by
material remapping using polygon–polygon intersection. If the cell turns out to be a mixed cell
after remapping then the cell is refined to first level. This refinement is carried out in both
Lagrangian precell as well as Eulerian cell unlike the algorithm presented in [23] in which refined
cells are back tracked. The advantage of this strategy is that it eliminates the velocity interpolation
at vertex of the refined cells. Hence, simplified back tracking approach utilized in AMR-MOF
presented in [23,25] is not necessary. Then material remapping is repeated for each subcell. After
first level of refinement and remapping, each mixed cell and precell are further refined to next
level of refinement. This process is repeated for each level of refinement till the maximum speci-
fied level of refinement is reached. Once final level of refinement is reached, material centroid for
each mixed cells is advected using barycenter based advection scheme [12].
Due to approximate integration in back tracking of cell vertices, cell volume in Eulerian mesh
and Lagrangian mesh does not remain constant. Therefore, there is always a possibility of discrep-
ancy in volume fraction estimation. In the present work, we eliminate this discrepancy in volume
fraction directly and compute total deviation in volume from initial volume. If the final volume is
greater than the initial volume then this deviation is subtracted in equal proportion from each
mixed cell. On the other hand, if the final volume is less than the initial volume then this deviation
in volume is added equally to each mixed cell. During this repair process, it should be kept in
mind that the local bound of volume fraction (f < 0 or f > 1) for each cell should not break.

3.2. Interfacial flow solver


In the present work, governing equations are discretized on collocated grid. All flow variables like
velocity, pressure, and volume fraction are calculated at cell centers of control volumes. In this
section, the numerical methodology for solution of governing equations is discussed in detail.
The balanced force projection algorithm [15] is discussed first. This is followed by discussion on
consistent balanced force projection algorithm [22].

3.2.1. Balanced force algorithm


In the BFA, first the velocity is predicted excluding pressure gradient and surface tension force
terms in the momentum equations. This predicted velocity field is not divergence-free. The diver-
gence of this velocity field is computed in which predicted face velocity is obtained by momen-
tum interpolation. The pressure field is established such that the velocity field is divergence free
and surface tension is balanced by the pressure gradient. Therefore, surface tension term is
included in the pressure Poisson equation. The final correct velocity field is obtained by adding
pressure gradient and surface tension into the momentum equations. In the present implementa-
tion, the continuity equation is discretized implicitly and all terms, except pressure gradient and
surface tension, are treated explicitly in momentum equations. Discretized governing equations at
n þ 1th time level are
r  unþ1 ¼ 0 (5)
 
1 n 1 ^
unþ1 ¼ u þ Dt nþ1 r  ðlruÞ  nþ1 rp
a nþ1
þF nþ1
 gj (6)
q q
where ua is the velocity after advection. The predicted velocity is obtained without adding pres-
sure gradient and surface tension in the momentum equation,
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 7

 
1 ^j
u ¼ ua þ Dt nþ1
r  ð lru Þ n
 g (7)
q
Since the predicted velocity field does not satisfy continuity Eq. (5), this velocity must be cor-
rected such that pressure field at n þ 1th time level satisfy continuity equation. Subtracting Eq.
(7) from Eq. (6) gives

unþ1 u 1
¼  nþ1 rpnþ1 þ Fnþ1 (8)
Dt q

Taking divergence on both side of Eq. (8) and using Eq. (5)
 nþ1 
rp
Dtr  ¼ r  u þ r  Fnþ1 (9)
qnþ1

The discretized pressure Poisson equation is solved to compute new time level pressure field.
The pressure Poisson equation is solved by GS-SUR iterative technique. Once pressure field at
new time level is determined, divergence-free velocity field at cell faces is computed by incorpo-
rating pressure gradient and surface tension in momentum equation,
!
 1 1
uf nþ1
¼ uf þ Dt  rpf nþ1 þ Ff nþ1 (10)
qf qf

and the cell center velocity is computed by


!
 1 1
uc nþ1
¼ uc þ Dt  rpf nþ1 þ Ff nþ1 (11)
qf qf
f !c

where cell face predicted velocity in Eq. (10) is computed from momentum interpolation. f ! c
represents the quantities inside braces are interpolated from cell faces to cell centers. The face
density that is to be incorporated in the discretization of pressure gradient and surface tension at
cell faces are computed by linear interpolation as follows:
qc þ qnb
qf ¼ (12)
2
The balanced force solution algorithm is summarized as follows:

1. Advect the material interface using RMOF method to compute new volume fraction
field, f nþ1 .
2. Calculate density and viscosity using Eq. (3).
3. Solve for predicted velocity using Eq. (7).
4. Compute pressure field by solving pressure Poisson equation.
5. Compute new face velocity and cell center velocity using Eq. (10) and Eq. (11) respectively.
6. Calculate RMS for velocity components; if it satisfies a prescribed residual limit then the
solution is obtained else go to step 1 and repeat till the limiting criterion is satisfied.

3.2.2. Consistent balanced force algorithm


The cell-centered velocity field obtained in the BFA discussed in the previous section is not con-
servative, i.e., discretely divergence-free, since the cell centered pressure gradient and surface
8 H. K. ZINJALA AND J. BANERJEE

tension force are interpolated from the cell faces. This interpolation leads to non-physical velocity
currents especially when surface tension is dominant. In [22], a consistent projection algorithm
for incompressible two-phase flow is proposed. The same implementation is followed here in con-
text to VOF method used in the present work in place of level set method considered in [22].
In the consistent formulation, cell centered velocity is obtained using conservative face velocities
given in Eq. (10) as
uf þ nþ1 þ uf  nþ1
uc nþ1 ¼ (13)
2
where f þ and f  denote the opposite face of a cell under consideration. Even though this simple
interpolation gives conservative velocity field, it introduces numerical dissipation in the velocity
field as discussed in [22]. This numerical dissipation must be eliminated to obtain smooth and
conservative velocity field by computing new velocity fluxes as defined below:
    !
2  2 
1 @ u @ u
gf  ¼  qc þ qnb (14)
8qf @x2 c @x2 nb
However, this velocity field is not divergence free. In order to obtain divergence free new face
velocity, gf nþ1 , a scalar p0 is obtained by solving the following Poisson equation:
 0
rp
Dtr  ¼ r  g (15)
q
This newly computed pressure field is used to obtain new face velocity which is to be used for
final cell center velocity computation.
!
 1 0
gf nþ1
¼ gf Dt rpf (16)
qf

The conservative cell center velocity is computed by the following interpolation


1  nþ1  1 
uc nþ1 ¼ uf þ þ uf  nþ1 þ gf þ nþ1 þ gf  nþ1 (17)
2 2
The additional steps required in CBFA are as follows:

1. Compute new face velocity field using Eq. (14).


2. Compute pressure field by solving pressure Poisson equation Eq. (15).
3. Compute divergence free new face velocity field by using Eq. (16).
4. Compute divergence free cell center velocity field by using Eq. (17).

3.2.3. Discretization of advection and diffusion terms


The advected velocity, uc a , is computed by solving only the advection terms in the momentum
equation. First order upwind scheme is used for discretization of the advection terms in the pre-
sent work. The viscous terms present in momentum equations are included in flow solver by con-
tinuous approach of [14]. They are discretized by second-order central difference scheme in an
explicit manner. In the present work, the face viscosity is obtained through the harmonic mean
of cell center viscosities as suggested in [27].

3.2.4. Surface tension model


The surface tension force is expressed by the continuum surface tension force of Brackbill et al.
[14]
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 9

F ¼ rj^
nd (18)
^ is the interface normal
where r is the surface tension coefficient, j is the interfacial curvature, n
vector, and d is the Dirac delta function. In case of continuous approach, the product of interface
normal and Dirac delta function is simply taken as the gradient of volume fraction. In BFA, due
to discretization of pressure gradient at cell faces, the surface tension force is also computed at
the corresponding cell face. The surface tension force at cell face is
F ¼ rjf ðrf Þf (19)

Therefore, surface tension is computed when the gradient of volume fraction is non-zero. The
face-centered curvature is computed by linear averaging of curvatures from cell centers. Height
function technique [15] is used for cell-centered curvature. Curvature calculation using height
function technique is followed from [15].

4. Results
The proposed CBFA-RMOF method is compared with the balanced force refined moment-of-
fluid (BFA-RMOF) method in this section. CBFA-RMOF and BFA-RMOF are compared for
standard test cases namely static drop, inviscid equilibrium drop, and viscous equilibrium drop.
Further demonstration is also provided for bubble rise simulation in buoyancy driven field.
Unless otherwise stated, RMOF method is used for interface evolution.

4.1. Static drop in the absence of gravity


Static drop in the absence of gravity is a benchmark test problem for the balance of the forces
due to surface tension and pressure. Theoretically, the jump in pressure across the drop is equal
to the surface tension given by
Dpexact ¼ rj (20)
where the exact curvature is expressed as:
jexact ¼ 1=R (21)
and R is the radius of the drop. In the present work, the test setup for static drop is taken from
[15]. The computational domain is a square of side lengths of eight units with the drop of radius
R ¼ 2 is placed at the center of the domain. The surface tension coefficient is considered as 73,
densities of the fluid inside the drop is q1 ¼ 1 and outside the drop is varied such that density
ratios vary from 1 to 105. The exact pressure difference across the drop is 36.5 from Eq. (20).
Two-dimensional simulations are carried out for both the balanced force algorithms (CBA and
CBFA). The relative errors in pressure jump ares measured in three different ways as suggested
by Francois et al. [15]:

 Total relative pressure jump error is given by


jDptotal Dpexact j
EðDpÞtotal ¼ (22)
Dpexact
where Dptotal is total pressure difference between inside the drop (r  R)and outside the
drop (r>R).
10 H. K. ZINJALA AND J. BANERJEE

Table 1. Spurious currents and relative errors in pressure jump for static drop test after single time step with Dt ¼ 106
q1 =q2 jujmax jujavg EðDptotal Þ EðDppartial Þ EðDpmax Þ
(a) BFA
1 7:80  108 9:44  109 2:45  102 4:84  103 6:11  103
101 1:69  107 2:03  108 2:45  102 4:83  103 6:14  103
103 2:53  107 2:70  108 2:45  102 4:83  103 6:12  103
105 2:55  107 2:73  108 2:45  102 4:83  103 6:12  103
(b) CBFA
1 7:80  108 9:44  109 2:45  102 4:84  103 6:11  103
101 1:69  107 2:03  108 2:45  102 4:83  103 6:14  103
103 2:53  107 2:70  108 2:45  102 4:83  103 6:12  103
105 2:54  107 2:73  108 2:45  102 4:83  103 6:12  103

 Partial relative pressure jump error is given by


jDppartial Dpexact j
EðDpÞpartial ¼ (23)
Dpexact
where Dppartial is partial pressure difference between inside the drop (r  R=2)and outside the
drop (r  3R=2).

 Maximum relative pressure jump error is given by


jDpmax Dpexact j
EðDpÞmax ¼ (24)
Dpexact
where Dpmax is maximum pressure difference in the entire domain.
In addition to the error in pressure jump, spurious currents are computed in terms of norms
of the velocity as maximum velocity
umax ¼ L1 ðuÞ ¼ maxðjjujjÞ (25)
and average velocity
sP
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n;m jjujj
uavg ¼ L 2 ðuÞ ¼ (26)
nm
The curvature is estimated using height function technique. The mesh size is taken as h ¼ 0.2
for density ratios varying from 1 to 105 and the time step size of Dt ¼ 106 is considered fixed
for the simulations. The pressure Poisson convergence criterion is set to 1012 and simulations
are carried out for single time step.
Table 1 summarizes the results for the maximum velocity and pressure jump after one time
step for BFA and CBFA. It can be observed that spurious currents in both the formulation are of
the order of Oð108 107 Þ. A slight increase in spurious current is observed as the density ratio
increases. This is similar to the observations reported in [15]. However, the same magnitude of
spurious current is found for both the BFA and CBFA. This is due to the fact that simulations
are carried out for single time step.

4.2. Long-time evolution of the equilibrium inviscid drop


In order to consider the effect of consistent implementation, long time evolution of inviscid drop
is considered. It is possible that spurious currents developed in static drop may increase with
time. For the purpose of comparison of spurious currents, simulations are carried out for 1,000
time steps with time step size of 103 . The surface tension coefficient is taken as r ¼ 73.
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 11

Table 2. Spurious currents for balanced force and consistent balanced force algorithms after 500 and 1,000 time steps
with Dt ¼ 103
q1
q2 ! 1 101 103 105
After 500 time steps
umax
(a) 4:12  103 4:23  103 6:13  103 6:71  103
(b) 1:95  103 2:80  103 4:42  103 4:83  103
uavg
(a) 5:34  104 6:44  104 1:10  103 1:13  103
(b) 2:69  104 3:79  104 8:18  104 9:05  104
After 1,000 time steps
umax
(a) 3:69  103 5:07  103 3:74  103 1:08  102
(b) 2:20  103 3:18  103 2:11  103 2:83  103
uavg
(a) 6:50  104 9:56  104 3:79  103 1:13  103
(b) 3:08  104 4:46  104 2:84  104 4:15  104
The density ratio is varied from 1 to 105. (a) Balanced force algorithm and (b) Consistent balanced force algorithm.

ρ1 ρ1
(a)
ρ2
= 100 (b)
ρ2
= 101
-3 -3
10 10
BF BF
-4
10 CBFA 10-4 CBFA
-5 -5
10 10

10-6 10-6
TKE

TKE

-7 -7
10 10

10-8 10-8
-9 -9
10 10
-10 -10
10 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t t

ρ1 ρ1
(c) ρ2
= 103 (d) ρ2
= 105
-3 -3
10 10
BF BF
-4 -4
10 CBFA 10 CBFA

10-5 10-5
-6 -6
10 10
TKE

TKE

-7 -7
10 10

10-8 10-8
-9 -9
10 10

10-10 10-10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t t
Figure 3. Total Kinetic Energy versus time for 1,000 time steps for inviscid drop in equilibrium for balanced force algorithm
(BFA) and consistent balanced force algorithm (CBFA) at various density ratios.

The density ratio is varied from 1 to 105 with density inside drop being fixed to 1 while outside
the drop density is varied.
Table 2 shows the results of spurious currents in term of maximum and average velocities after
500 and 1,000 time steps. The spurious currents are reduced in case of consistent implementation
compared to balanced force formulation. For complete understanding on spurious current,
12 H. K. ZINJALA AND J. BANERJEE

average velocity norm is also included in Table 2. Average velocity in consistent implementation
is lower than the balanced force formulation. It can be observed that spurious currents increase
with the density ratio. For given density ratio, spurious currents reduce with time for CBFA com-
pared to balanced force formulation. For further investigation, evolution of total kinetic energy
(TKE) of the flow is established as a function of time. The evolution of the TKE is an indication
of the growth of the spurious velocities with time [15]. The total kinetic energy is defined as
1X
TKE ¼ q jXc juc  uc (27)
2 cells c

where qc and jXc j are density and volume of the cell, respectively. Figure 3 shows TKE as a func-
tion of time for 2D inviscid static drop for various density ratios. The results are plotted for both
balanced force and consistent balanced force formulations with RMOF for interface evolution.
The integration time step is kept constant at Dt ¼ 103 and is integrated for time upto 1.0 unit.
The mesh resolution is such that R=h ¼ 10. As observed in [15], total kinetic energy oscillates
with time for height function technique of curvature estimation as shown in Figure 3. On the
other hand, in consistent implementation, the amplitude of TKE decreases with time. This shows
reduced spurious currents in CBFA. The decay in TKE is faster in CBFA compared to BFA for
large density ratio.

4.3. Long-time evolution of the viscous drop in equilibrium


The performance of the proposed method for viscous drop is compared with BFA of [28] in
which interface was tracked using refined level set method [28] and front tracking method of
[29]. The simulations are carried out for long time evolution of viscous drop in equilibrium. A
drop of diameter D ¼ 0.4 is placed at the center of the computational domain of size 1  1. The
domain is discretized with mesh size of 32  32. The viscosity and density ratios are kept constant
as one and surface tension coefficient is set to r ¼ 1. The same problem was considered in the
earlier study of [29] using front tracking method and in [28] using refined level set method with
BFA. The spurious currents are studied in both these studies [28,29] in terms of non-dimensional
spurious currents given by Capillary number, Ca ¼ jumax jl=r with Laplace number being varied.

Table 3. Spurious currents in terms of Capillary number for balanced force and consistent balanced force at t ¼ 250 for vari-
ous Laplace number
La ! 12 120 1,200 12,000
(a) 6:76  106 5:71  106 5:99  106 8:76  106
(b) 0:10  106 0:11  106 0:12  106 1:44  106
(c) 2:47  108 6:58  108 1:90  107 7:95  106
(d) 2:44  108 6:58  108 1:80  107 4:15  106
Results are compared with front tracking method [29] and Level set method [28] (a) Front tracking [29] (b) Level set method
[28] (c) Balanced force algorithm and (d) Consistent balanced force algorithm. Bold numerical data represents result of the
present work.

Table 4. Spurious currents in terms of Capillary number under grid refinement for balanced force and consistent balanced
force at t ¼ 250 for Laplace number ¼12,000
h ! 1/16 1/32 1/64 1/128
(a) 8:76  106 6:68  106 1:07  106 1:20  107
(b) 4:92  106 1:44  106 3:4  107 5:0  108
(c) 5:34  104 7:95  106 6:24  106 2:53  106
(d) 3:25  104 4:15  106 4:15  106 2:61  106
Results are compared with front tracking method [29] and Level set method [28] (a) Front tracking [29] (b) Level set method
[28] (c) Balanced force algorithm and (d) Consistent balanced force algorithm. Bold numerical data represents result of the
present work.
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 13

The Laplace number La ¼ 1=Oh2 ¼ rqD=l2 is varied by changing the fluid density while keeping
the density ratio constant. The time step size is chosen such that capillary time step restriction
is satisfied.
Table 3 shows spurious currents in terms of capillary number with varying Laplace number at
non-dimensional time t ¼ tr=ðDlÞ ¼ 250. Results of CBFA-RMOF method shows capillary
numbers which are two-orders of magnitude smaller than the results reported in [29] and [28].
Compared to results of balanced force algorithm refined moment of fluid (BFA-RMOF) method,
results obtained by CBFA-RMOF shows the least spurious currents. At large Laplace numbers
spurious currents increases by an order of magnitude. The grid convergence results are shown in
Table 4 for Laplace number of 12,000. Spurious currents in terms of Capillary number are com-
pared with published literature [28,29]. The spurious currents are higher for the same grid size
compared to level set and front tracking method. The reason for this is the fact that curvature

(a) (b)
2 2
t = 0.25 t = 0.25

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2

2.5 2.5
t = 0.5 t = 0.5

2 2

1.5 1.5

1 1

0.5 0.5
0 0.5 1 1.5 2 0 0.5 1 1.5 2

2.5 2.5
t = 0.75 t = 0.75

2 2

1.5 1.5

1 1

0.5 0.5
0 0.5 1 1.5 2 0 0.5 1 1.5 2

Figure 4. Air bubble rise in water with low surface tension at various time for grid resolution of 50  100 for (a) Standard MOF
and (b) RMOF.
14 H. K. ZINJALA AND J. BANERJEE

(a) (b)
4 4
t = 1.3 t = 1.3

3 3

2 2

1 1

0 0
0 1 2 0 1 2
Figure 5. Air bubble rise in water with low surface tension at time t ¼ 1:3s for grid resolution of 50  100 for (a) Standard MOF
and (b) RMOF.

calculation from volume fraction is not as accurate as in refined level set method discussed in
[28]. Therefore, second order convergence behavior is not obtained for the present method
(CBFA-RMOF).

4.4. Rise of air bubble in water


The performance of CBFA-RMOF method is demonstrated using the rise of air bubble in water
due to buoyancy for low surface tension and high surface tension. In this problem an air (low
density fluid) bubble is placed inside water (high density fluid). Due to buoyancy force acting on
the bubble, air bubble rises inside the dense fluid. The fluid properties for air and water
are q1 ¼ 1:226; q2 ¼ 1000; l1 ¼ 1:137  103 ; l2 ¼ 1:78  105 .

4.4.1. Low surface tension case


The discusion in this subsection is for low surface tension with complex topological deformations.
The performance of RMOF is compared with the standard MOF method. This problem, reported
earlier in [30] for slightly different air density and viscosity, demonstrates the robustness and
applicability of the RMOF for large deformation. The computational domain is of size 2  4 with
an air bubble of radius R ¼ 0.5 placed at (1, 1). The grid size used is 50  100 with constant time
step of Dt ¼ 0:001s. The surface tension coefficient is chosen as r ¼ 0:072N=m. Figure 4 shows
interface evolution in terms of fluid polygons at different times for both standard MOF and
RMOF. The difference between standard MOF and RMOF is negligibly small for initial time lev-
els since interface is well connected. However, small improvement in terms of interface connectiv-
ity is observed for RMOF. Interface between two neighbor cell is resolved easily in case of
RMOF. This fact is also emphasized in [30]. The present results are almost identical to the results
reported in [30] for time upto 0.75 s. The performance of the algorithms is compared when the
interface breaks and small structures are generated. As time progresses, tails of the air bubble
becomes thinner and at some point in time air droplets get separated from the main fluid body.
These small interface structures are extremely difficult to capture using PLIC and QUASI [30].
However, MOF method captures some small structures compared to VOF methods as shown
in Figure 5. Despite being accurate interface capturing method, inconsistency in volume and
centroid leads to wrong prediction of fluid interfaces using MOF as depicted in Figure 5.
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 15

(a) (b)
2.2 2.2
t = 0.2 t = 0.2

2 2

1.8 1.8

1.6 1.6

1.4 1.4

1.2 1.2
0.4 0.6 0.8 1 1.2 1.4 1.6 0.4 0.6 0.8 1 1.2 1.4 1.6

2.4 2.4
t = 0.35 t = 0.35

2.2 2.2

2 2

1.8 1.8

1.6 1.6

1.4 1.4
0.4 0.6 0.8 1 1.2 1.4 1.6 0.4 0.6 0.8 1 1.2 1.4 1.6

2.4 2.4
t = 0.5 t = 0.5

2.2 2.2

2 2

1.8 1.8

1.6 1.6

1.4 1.4
0.4 0.6 0.8 1 1.2 1.4 1.6 0.4 0.6 0.8 1 1.2 1.4 1.6
Figure 6. Bubble shapes in terms of fluid polygons at time t ¼ 0:2s; t ¼ 0:35s and t ¼ 0:5s for grid resolution of 40  60 (a) bal-
anced force RMOF and (b) consistent balanced force RMOF.

This actually is due to the fact that the size of separated fluids is smaller than the cell size. These
under resolved thin structures are however better captured by RMOF method.

4.4.2. High surface tension case


The problem under consideration is the rise of a large bubble reported in [31] using ghost fluid
method with level set interface evolution. Francois et al. [15] earlier reported these simulations
using balanced force formulation with VOF method for interface evolution. The computational
domain is a rectangular region of size ½0; 2  ½0; 3 . The initial position of bubble is such that
the center of the bubble is placed at the center of the domain. The bubble radius is R ¼ 1/3.
16 H. K. ZINJALA AND J. BANERJEE

(a) (b)
2.2 2.2
t = 0.2 t = 0.2

2 2

1.8 1.8

1.6 1.6

1.4 1.4

1.2 1.2
0.4 0.6 0.8 1 1.2 1.4 1.6 0.4 0.6 0.8 1 1.2 1.4 1.6

2.4 2.4
t = 0.35 t = 0.35

2.2 2.2

2 2

1.8 1.8

1.6 1.6

1.4 1.4
0.4 0.6 0.8 1 1.2 1.4 1.6 0.4 0.6 0.8 1 1.2 1.4 1.6

2.4 2.4
t = 0.5 t = 0.5

2.2 2.2

2 2

1.8 1.8

1.6 1.6

1.4 1.4
0.4 0.6 0.8 1 1.2 1.4 1.6 0.4 0.6 0.8 1 1.2 1.4 1.6
Figure 7. Bubble shapes in terms of fluid polygons at time t ¼ 0:2s; t ¼ 0:35s and t ¼ 0:5s for grid resolution of 80  120 (a)
balanced force RMOF and (b) consistent balanced force RMOF.

The surface tension coefficient is taken as r ¼ 728 N=m. The gravitational acceleration is g ¼
9:8 m=s2 and is acting in the downward direction. Grid resolutions used for the simulation are
40  60 and 80  120. Comparisons are presented here for simulations obtained with BFA and
CBFA. The interface is tracked using RMOF method for accurate interface representation.
Computed bubble shapes are shown in Figure 6 in terms of fluid polygons for grid resolution of
40  60 at time t ¼ 0:2s; t ¼ 0:35s and t ¼ 0:5s for BFA-RMOF and CBFA-RMOF. Interface
shape captured by both methods is almost similar. However a wavy interface is observed for
BFA. This waviness in interface is due to spurious currents at the interface. On the other hand,
consistent implementation maintains interface sharpness and smoothness thereby demonstrating
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 17

that the spurious currents are eliminated. It is important to note that the shape of the interface is
wavy only because of LEAS for volume fraction and centroid advection. This is not observed for
Eulerian advection schemes where face centered velocities are used for flux computation which
are anyway conservative [15].
On grid refinement to 80  120, the BFA-RMOF method is not able to maintain the shape of
the bubble as shown in Figure 7 while CBFA-RMOF exactly produces cap bubble. This shows
excellent capability of the CBFA to capture interface dynamics for surface tension dominant two-
phase flows.

5. Conclusions
A consistent balanced force RMOF method is presented for Cartesian grid. The material interface
is tracked using RMOF method. The localized refinement in interfacial cells is introduced for
accurate interface tracking. The material advection in RMOF is performed using LEAS. A modi-
fied advection strategy similar to standard MOF method is presented where Lagrangian precell of
Eulerian cell is refined for remapping.
In BFA, the cell center velocity is computed from cell center pressure gradient and cell center
surface tension force. These cell center quantities are interpolated from cell faces. This interpol-
ation does not generate conservative velocity field and generates spurious currents near the inter-
facial band. These spurious velocities are removed in the present work by implementing CBFA of
[22]. In consistent implementation, cell center velocity is computed from cell face velocity.
However, this direct interpolation results in large numerical dissipation added to cell center vel-
ocity field. This numerical dissipation is removed by correcting this velocity field. The velocity
field required for correction is computed by considering dissipative velocity fluxes. Since these
velocity fluxes are not divergence free, additional pressure Poisson equation is solved. This newly
computed pressure field is then used to estimate divergence free velocity field. This velocity field
is applied as correction for cell center velocity computation. This velocity field is conservative
and generates least spurious currents.
The proposed method is verified for variety of test cases for spurious currents by considering
equilibrium drop in the absence of gravity. From these tests, it is concluded that spurious cur-
rents are smaller in magnitude for CBFA compared to BFA. Finally, both the algorithms are
tested for rise of air bubble inside water. For low surface tension, the interface breakup of large
bubble was captured by RMOF. MOF fails to capture the broken interfaces. For high surface ten-
sion case, performance of CBFA-RMOF is compared with BFA-RMOF. The waviness in interface
is observed for BFA-RMOF while CBFA-RMOF exactly produces cap bubble. This waviness in
interface is attributed to spurious currents near the interfacial region. Therefore, it is concluded
that spurious currents are eliminated by the CBFA.

References
[1] C. W. Hirt and B. D. Nichols, “Volume of fluid (VOF) method for the dynamics of free boundaries,” J.
Comput. Phys., vol. 39, pp. 201–225, 1981.
[2] D. L. Youngs, “Time-dependent multimaterial flow with large fluid distortion,” in Numerical Methods for
Fluid Dynamics, K. W. Morton and M. J. Baines, Eds. New York, NY: Academic Press, 1982, pp. 273–285.
[3] W.J. Rider and D.B. Kothe, “Reconstructing volume tracking,” J. Comput. Phys., vol. 141, pp. 112–152,
1998.
[4] J. Lopez, J. Hermandez, P. Gomez, and F. Faura, “A volume of fluid method based on multidimensional
advection and spline interface reconstruction,” J. Comput. Phys., vol. 195, pp. 718–742, 2004.
[5] E. Aulisa, S. Manservisi, R. Scardovelli, and S. Zaleski, “A geometrical area-preserving volume-of-fluid
advection method,” J. Comput. Phys., vol. 192, 355–364, 2003.
[6] R. Scardovelli and S. Zaleski, “Interface reconstruction with least-square fit and split Eulerian Lagrangian
advection,” Int. J. Numer. Methods Fluids, vol. 41, pp. 251–274, 2003.
18 H. K. ZINJALA AND J. BANERJEE

[7] E. Aulisa, S. Manservisi, R. Scardovelli, and S. Zaleski, “Interface reconstruction with least-squares fit and
split advection in three-dimensional Cartesian geometry,”. J. Comput. Phys., vol. 225, pp. 2301–2319, 2007.
[8] A. Cervone, S. Manservisi, R. Scardovelli, and S. Zaleski, “A geometrical predictorcorrector advection
scheme and its application to the volume fraction function,” J. Comput. Phys., vol. 228, pp. 406–419, 2009.
[9] D.J.E. Harvie and D.F. Fletcher, “A new volume of fluid advection algorithm: the stream scheme,” J. Comp.
Phys., vol. 162, pp. 1–32, 2000.
[10] D.J.E. Harvie and D.F. Fletcher, “A new volume of fluid advection algorithm: the defined donating region
scheme,” Int J. Numer. Methods Fluids, vol. 35, pp. 151–172, 2001.
[11] H. K. Zinjala and J. Banerjee, “A Lagrangian-Eulerian volume tracking with linearity-preserving interface
reconstruction,” Numer. Heat Trans. Part B, vol. 68, pp. 459–478, 2015.
[12] H. K. Zinjala and J. Banerjee, “A Lagrangian-Eulerian advection scheme with moment-of-fluid interface
reconstruction,” Numer. Heat Trans. Part B, vol. 69, pp. 563–574, 2016.
[13] V. Dyadechko and M. Shashkov, “Reconstruction of multi-material interfaces from moment data,”
J. Comput. Phys., vol. 227, pp. 5361–5384, 2008.
[14] D. B. Kothe J. U. Brackbill and C. Zemach, “A continuum method for modelling surface tension,”
J. Comput. Phys., vol. 100, pp. 335–354, 1992.
[15] M. M. Francois et al., “A balanced-force algorithm for continuous and sharp interfacial surface tension
models within a volume tracking framework,” J. Comput. Phys., vol. 213, no. 1, pp. 141–173, 2006.
[16] F. Denner and B. G. M. van Wachem, “Fully-coupled balanced-force VOF framework for arbitrary meshes
with least-squares curvature evaluation from volume fractions,” Numer. Heat Trans. Part B, vol. 65, pp.
218–255, 2014.
[17] J. Waters, D. B. Carrington, and M. M. Francois, “Modeling multiphase flow: spray breakup using volume
of fluids in a dynamics LES FEM method,” Numer. Heat Trans. Part B, vol. 72, no. 4, pp. 285–299, 2017.
[18] B. Merriman, R. P. Fedkiw, T. Aslam, and Stanley Osher, “A non-oscillatory Eulerian approach to interfaces
in multimaterial flows (the ghost fluid method),” J. Comput. Phys., vol. 152, pp. 457–492, 1999.
[19] B. Pulvirenti M. Magnini, and J.R. Thome, “Characterization of the velocity fields generated by flow initial-
ization in the CFD simulation of multiphase flows,” J. Appl. Math. Model., vol. 40, pp. 6811–6830, 2016.
[20] M. R. Davidson, D. J. E. Harvie, and M. Rudman, “An analysis of parasitic current generation in volume of
fluid simulations,” J. Appl. Math. Model., vol. 30, pp. 1056–1066, 2006.
[21] Z. Cao, D. Sun, B. Yu, and J. Wei, “A coupled volume of fluid and level set method based on analytic
PLIC for unstructured quadrilateral grids,” Numer Heat Trans. Part B, vol. 73, no. 4, pp. 189–205, 2018.
[22] M.-J. Ni, “Consistent projection methods for variable density incompressible Navier Stokes equations with
continuous surface forces on a rectangular collocated mesh,” J. Comput. Phys., vol. 228, pp. 6938–6956,
2009.
[23] H. K. Zinjala and J. Banerjee, “Refined moment-of-fluid method,” Numer. Heat Trans. Part B, vol. 71, pp.
574–591, 2017.
[24] R. N. Hill and M. Shashkov, “The symmetric moment-of-fluid interface reconstruction algorithm,”
J. Comput. Phys., vol. 249, pp. 180–184, 2013.
[25] H.T. Ahn and M. Shashkov, “Adaptive moment-of-fluid method,” J. Comput. Phys., vol. 228, pp.
2792–2821, 2009.
[26] S. Saincher and J. Banerjee, “A redistribution-based volume-preserving PLIC-VOF technique,” Numer. Heat
Trans. Part B, vol. 67, pp. 338–362, 2015.
[27] M. Rudman, “A volume-tracking method for incompressible multifluid flows with large density variations,”
Int. J. Numer. Methods Fluids, vol. 28, pp. 357–378, 1998.
[28] M. Herrmann. A balanced force refined level set grid method for two-phase flows on unstructured flow
solver grids. J. Comput. Phys., 227:2674–2706, 2008.
[29] S. Popinet and S. Zaleski, “A front-tracking algorithm for accurate representation of surface tension,” Int. J.
Numer. Methods Fluids, vol. 30, pp. 775–793, 1999.
[30] S. K. Das, S. V. Diwakar, and T. Sundarajan, “A quadratic spline based interface (QUASI) reconstruction
algorithm for accurate tracking of two-phase flows,” J. Comput. Phys., vol. 228, pp. 9107–9130, 2009.
[31] R. Fedkiw M. Kang, and X. D. Liu, “A boundary condition capturing method for multiphase incompress-
ible flow,” J. Sci. Comput., vol. 15, pp. 323–360, 2000.

S-ar putea să vă placă și