Anda di halaman 1dari 46

J Sci Comput

DOI 10.1007/s10915-015-0053-0

Full 3D Simulations of Two-Phase CoreAnnular Flow


in Horizontal Pipe Using Level Set Method

Byungjoon Lee1 Myungjoo Kang1

Received: 12 December 2014 / Revised: 3 May 2015 / Accepted: 30 May 2015


Springer Science+Business Media New York 2015

Abstract Simulating coreannular flow is an important task in the oil industry. Many
attempts have been made to simulate vertical upflow or downflow. However, in the case
of horizontal pipes, simulation is likely to succeed when the effect of gravity is ignored,
because the motion of the fluid is no longer axisymmetric. This gravity-ignoring
simulation is acceptable when the density of oil is almost similar to that of water, but in
most cases, density-matching does not occur in reality. A few attempts have been made to
simulate flow in horizontal pipes (Ooms et al. in Phys Fluids 25:032102, 2013), but these
attempts used 2D cross-sectional simulation or 3D simulation of a specific moment. This
paper discusses several results for coreannular flow in horizontal pipes with gravity
effect, using full 3D simulation. Using a level set method, we express the interface of
coreannular flow and compare different cases with different parameters.

Keywords Coreannular flow NavierStokes equation Level set methods

1 Introduction

The transportation of heavy crude oil through a pipe is a naturally important problem in oil
industry. Among several efficient modes of transportation, lubricating a pipe with a low-
viscosity fluid, such as water, is a well-used technique. The primary idea behind
lubricating pipes is the property of low-viscosity fluid to encapsulate high-viscosity fluid.
This type of flow is designated as coreannular flow, which features one fluid at the core
and another fluid in the annulus.

B Myungjoo Kang
mkang@snu.ac.kr

1
3
J Sci Comput
Byungjoon Lee
asone30@snu.ac.kr
1 Department of Mathematical Sciences, Seoul National University, Seoul, Korea

1
3
Many attempts have experimented on [1] and simulated coreannular flow [2,3].
However, in many cases, a vertical or parallel pipe without gravity is used because it is
difficult to analyze parallel pipes owing to the effect of gravity. Additionally, in previous
simulations, many researchers assumed that the given flow was axisymmetric in a stable
regime, which is not true for a parallel pipe with gravity. Recently, one study [4] attempted
to simulate a horizontal coreannular flow by using the VOF method [2], but this
simulation examined a 2D cross-sectional plane and a 3D simulation of a specific moment.
The purpose of this study is to perform a full 3D simulation on coreannular flow in a
horizontal pipe while considering the gravity effect. To describe the interface between oil
and water, we selected a level set method [5]. In many simulations on incompressible fluid
[6,7], level set formulation is well-suited for visualizing the interface. Because oil and
water have very different viscosities, we primarily attempted to perform the simulation
based on the boundary capturing method [7,8]. This is viewed as an extension of Kangs
work [9] in terms of using a level set method. However, in 3D simulation, axisymmetric
coordinates must be transformed into regular grid coordinates with irregular boundary
domain for a cylindrical pipe, which complicates the simulation.
In this paper, we will provide numerical results on coreannular flow in a parallel pipe
in full 3D domain. This paper discusses the following: governing equation and parameters
of simulation are given in Sect. 2. In Sect. 3, information is given about the numerical
method based on level set. This section also provides a discussion on boundary conditions
when a domain has an irregular boundary. Finally, we give numerical results in Sect. 4.

2 Governing Equations of Motion

2.1 The Equation of Motion

Coreannular flow, as part of a two-fluid flow, can be expressed with the NavierStokes
equation:
u 1
+ u u = ( P + (2S)) + F (1)
t
where is the density, the viscosity, and S the viscous stress tensor,
. u i
1 u j . (2)
+
Sij = xi x j
2
and F is the source term, such as acceleration exerted by gravity and surface tension. In
coreannular fluid, P = f x + p for the driven pressure gradient vector f and the axis
coordinate vector x. Because oil and water are liquids, the velocity u must be subject to
the
incompressibility condition:

u = 0 (3)

In this paper, oil is designated as fluid 1 and water as fluid 2. Using a level set method, we
represent the interface between oil and water, which is moved by the following equation:

+ u = 0 (4)
t
Here, is a signed distance function to the interface between oil and water.
J Sci Comput

2.2 Dimensionless Form of NavierStokes Equation

For convenience, we use the dimensionless form of Eq. (1). The following variables are
used for expressing dimensionless form:
x = Rx , u = V 0u , t = R t
V0
(5)
P = Poil(V0)2, = oil, = oil
where dimensionless variables are denoted as *; R is the undisturbed reference interface
position; and V0 is the centerline velocity. Transforming Eq. (1) into dimensionless form
and dropping *, we have
px . .
1 (2u x )x + (u y + vx ) y + ((u z + . .
1 R
wx ))z
ut + u u = + f1
Re oil V02
+

. . (6)
py 1 ((u y + vx ))x + (2v y ) y + (vz + w y ) z
R
=
vt + u v Re +
+ V02
. g (7)
1 R
.
+ f2
oil V02
pz . .
1 ((u z + wx ))x + (vz + w y ) y + (2wz . .
)z 1 R
wt + u w = + 2 f3
Re oil V0
+

(8)
RV
where the Reynolds number, Re = oiloil 0 , f = ( f1, f2, f3), is driven by the pressure
gradient.
Because there is discontinuity at the interface between oil and water, smoothed density
and viscosity are used for this equation:
() = 1 + ( 1) H (), (9)
() = 1 + ( 1) H (), (10)
where = water /oil , = water /oil and H () is the Heaviside function, as follows:

0, if < 0,
H () = 1 , if = 0, (11)
2
1, if > 0.

2.3 Parameters for Equations


J Sci Comput
The pipe radius is represented as R2 and the interface position is r = R1. Regarding the
interfacial tension , P2 P1 = /R1. Constant pressure gradient in the axial direction is
imposed: d P/dx = f . This constant gradient is viewed as the external force. The four
primary dimensionless parameters are given as follows:
m = 2 /1 , a = R2 /R1 , = 2 /1 , K = ( f + 1 g)/( f + 2 g), (12)
where K measures the ratio of driving forces in the core and annulus. The centerline
velocity is chosen to be
R2
V0(0) = ( f + 2g) 1 A, A = mK + a2 1 + 2(K 1) log a. (13)
4 2
The dimensionless base velocity field is V (r), where
. 1
V (r) = 2 r 2(K 1) log(r/a)], 1 r a,
A [a
2
1 (13)
1 A (mr 2 K ), r < 1.

where r is the distance from the axis of the axial direction to the interface. The velocity is
imposed only for the axial direction.
An interfacial tension parameter is J = R11/12. Reynolds numbers, Rei , are defined
by Rei = i V0(0) R1/i , i = 1, 2, where Re 1/Re 2 = m/ .

3 Numerical Formulation

3.1 Level Set Description

In this simulation, we introduce two different level set functions: one for the interface
between oil and water, and the other for the cylinderical pipe. For convenience, f is set
for the former function, and c for the latter.

3.1.1 Level Set Function to Describe the Interface

The level set technique is used to describe the interface [10]. We initialize f to be the
signed distance function from the interface between oil and water. Then, the interface can
be expressed as the zero level set of f as

= {x | f (x, t) = 0} (14)

The oil and water regions will be described by f < 0 and f > 0. Then, velocities in
each phase can be written as
.
u oil , f 0
u= (15)
uwater , f > 0.
where u is the velocity of fluid. When using the level set method, the level set function is
set to move along with the given velocity through the following equation:

( f )t + u f = 0. (16)

The unit normal on the interface and curvature of the interface can be computed as

f (17)
n=
| f |

= n (18)

where the unit normal is drawn from the oil into the water.
In this study, we select the initial function of the interface as follows:
f (x(r, x )) = R1 + C cos(x ) (19)
where C is the amplitude and is the wave number.
3.1.2 Level Set Function for Boundary Condition

Because the current simulation is a full 3D simulation in the pipe, the boundary for com-
putation is not axis-aligned. To address the boundary conditions, we introduce the level set
function for the pipe:
c(x(r, x )) = R2 (20)
During the entire simulation, the boundary pipe is never changed, so we only use this level
set function for determining the boundary conditions of the simulation.

3.2 Formulation of Smoothed Heaviside Function: Density and Viscosity

In Eqs. (9) and (10), density and viscosity are defined discontinuously near the interface,
which causes numerical difficulties in implementation. For this reason, we present a
numerical Heaviside function with a fixed width, which is proportional to the spatial mesh
size, as follows:
1, if > (21)
;
H () =
0, if <
;
.
.
1 x
1 .
1+ + sin , otherwise
.
2

Here, is the prescribed width. Now, we can replace () and () in Eqs. (9) and (10)
with smoothed density and viscosity functions, which are denoted as () and ().

() = + (1 ) H (), (22)
() = + (1 ) H (), (23)
We set = 1.56x .
3.3 Projection Method

This method was initially proposed by Chorin [11], whereas an explicit version of the
method was presented by Fortin et al. [12]. This explicit method is a fractional step method
with first-
order accuracy in time. Let
= (un, vn, wn). In the first step, we explicitly compute a
Vn
provisional value, V , with
n
. . n 1
V V
n
V +
6t
V (2S) F = 0 (24)
Re
which is the momentum equation with no pressure gradient. Only the discretization in time
equations
is considered here. Then, in the second step, we correct V by considering the
n+1
V
V 1 0 (25)
pn+1
+ =
6tn+1
V =0 (26)

By using the divergence of Eq. (24) and by making use of Eq. (26) which states that V
n+1
must be a divergence-free vector, .we gain the.Poisson equation:
1 n+1 1
= V (27)
p

6t
The boundary condition for p is obtained by projecting vector Eq. (26) on the outward
normal unit N to the boundary . Thus, we obtain the Neumann condition:
. . . N (28)
pn+1 .
n+ V
=
N 1
V
6t

where V is the value of V on . To summarize, Eq.(24) produces V ; the
solution of

the Neumann problem [Eq. (27)] produces p n+1 ; finally, the velocity, V n+1, is
computed
from Eq. (25). More details are provided in a previous study [13].

3.4 Numerical Procedure

We use a MAC grid to solve the NavierStokes equation as usual. All vector quantities,
such as un , are stored at cell face; scalar quantities, suchf as n , are located at the cell
center.
In one time step, we initially solve for un+1, then update fn+1. The numerical procedure
is as follows:
Step 1. Initialize f (x, t) as a signed distance function to the initial front.
Step 2. Solve the NavierStokes
n+1/2
equations [Eqs. (6),(7), and (8)] to obtain un+1.
Step 3. Propagate n to by using level set equation (16) with the velocity obtained
f f
at Step 2. n+1/2
Step 4. Reinitialize and obtain n+1.
f f

3.4.1 Advection Terms

The advection terms in Eqs. (1) and (4) can be discretized as follows:
.
.. u. ui +1/2, j,k ui +1, j,k ui,
V = .
i +1/2, j,k
. j,k

6x. .
vi +1/2, j,k ui +1/2, j +1/2,k ui +1/2, j 1/2,k
+
. 6y
wi +1/2, j,k ui +1/2, j,k+1/2 ui +1/2, (29)
.
j,k1/2
+
6z
. .
.. . . u i, j,k i +1/2, j,k i 1/2, v.i, j,k i, j +1/2,k i, j 1/2,k
V i, j,k = .
j,k +
6x 6y
.
wi, j,k .i, j,k+1/2 i, (30)
j,k1/2
+
6z

Velocity advection for v and w are treated similarly. In Eqs. (29) and (30), ui +1, j,k ,
i +1/2, j,k must be obtained by using proper approximation. Here, we use ENO 3rd for
velocity and WENO 5th for level set approximation [1416].

3.4.2 Viscous Terms

The viscous term (2S) in Eq. (1) can be discretized by using central differencing. For
example, in x directional velocity u,
. . . . (31)
1 (2u x )x + (u y + vx ) y + ((u z + wx ))z
Re
i +1/2, j,k
. . .
we have (u y + (u y + vx )
.
= vx )
. .
((u + v )) i +1/2, j +1/2,k i +1/2, j 1/2,k
y x y i +1/2, j,k 6y
where ui +1/2, j +1,k ui +1/2, j,k vi +1, j +1/2,k ui, j
6y +1/2,k
(u y + vx)i +1/2, j +1/2,k 6x
= +

1 .
i +1/2, j +1/2,k = i +1, j,k + i, j,k + i +1, j +1,k + i, j,k
4
.

The other terms in Eq. (31) and the y, z directional velocities v, w can be discretized
similarly.

3.5 Semi-Implicit Viscous Solver

Because the difference in viscosity between oil and water is relatively large, viscous terms
must be updated in a strictly small time step to ensure stability. This causes the entire sim-
ulation to be inefficient in terms of CFL time restrictions. To modify this, we used the
unconditionally stable semi-implicit technique [3] for updating viscosity terms.
In the semi-implicit method, each velocity is updated only as an implicit or explicit term.
That is,
. . . . ..
2u x + . .u + v + u + wn z
n
..
u un . n . 1 y x z x
n
V u = x y + F1
6t
n
(32)
Re
+
+ +
. . .. . . . ..
n .
n
v vn .n . n 1 uy + 2vy
vz + z
x y
V v = v w + F2
6t x y
Re (33)
+
. .
. . n ..x . .v + + 2w z
u + w .. n
w wn + .n
1
6t
V w =
. Re
n
z+ wx z y z (34)
y
+ F3

This modification provides the following systems:

u-velocity . . ...
.
. . .
t
.

6
I 2 + + u = explicit term x
Re x y y z z
(35)
x

v-velocity . . ...
.
. . .
t
.

6
I + 2 + v = explicit term y
Re x y y z z
(36)
x
w-velocity
. .
6t . . .
+ + w = explicit term z
. . ...
I 2 (37)
Re x y y z z
where
x
. 6t . . .
n n
ex plicit term x = u 6t . . n + wn + 6t F1
un + Re v
. x y
V . x z
. 6t . . n. . .
n .
ex plicit term y = v n 6t . vn + u y n + 6t F2
Re w
x
+ y
V z
. 6t ..
n ..
ex plicit term z = w n 6t . wn + Re n + vn + 6t F3
u
.
V z x . z y

These systems are symmetric positive definite (SPD). Hence, we can solve these by
using a fast algorithm such as the preconditioned conjugate gradient (PCG) method. A
previous study [3] recommended a factorization scheme to solve these systems. When we
simulate this element, we found that the semi-implicit solver requires <30 iterations,
except for the first step because of the time step. This implies that the efficiency of the
semi-implicit step is not degraded when we solve this part directly. Hence, we solve the
semi-implicit part directly rather than using a factorization scheme.

3.6 Poissons Equation

Solving Poissons equation is crucial for the incompressible NavierStokes equation


equipped with a projection method. The Poissons equation for this problem is
reformulated as follows: . .
1 n+1 1
= V (38)
p

6t
with the Neumann boundary condition

. .n+1 . . N (39)
p
= V
n+1
V

N 6t

Because the solution
of this Neumann problem is independent with the choice of V
, we can select V=
n+1 V for simplicity; that is,
. .n+1
p
=0 (40)

N
The condition of compatibility for the Neumann problem is
.
, . , n+1 .
p d = N d
.n+1

0= V V
N 6t .
, n+1

.


that is, = V V d (41)
6t

, ,

n
+1
V d = V d = 0 (42)

It is important for the discretization with respect to space to conserve the preceding
compat- ibility condition.
This is the homogeneous Neumann Poisson problem, which is a singular system, Ax =
b, in discretized form. Generally, when we solve singular system with PCG, some
special
treatment must be needed. By noting that e, the vector of all ones, is a basis of null space
of A, we solved ( AN + 1 ee t )y = b, where N is the number of row, and orthogonalize y
with respect to e to have real solution x . This modification guarantees null space of A is
shifted
away and we can find the solution (unique upto constant) on the restricted space of the
Krylov sequences of CG Span.

3.6.1 Variable Coefficient Poissons Equation with Jump Condition: Surface Tension
Effect Considered

Instead of using a continuous surface tension (CSF) model [17], we use a boundary
condition capturing (BCC) model for solving pressure [7,8].
In the case of a smoothed out viscosity, the jump condition in pressure at interface is
known to be [ p] = , where

[ p] = pwater poil

is the coefficient of surface tension and is the curvatureoilofRVthe


2 interface. In dimensionless
form, this can be modified, [ p] = 1 , where We = 0
= 1 ( Re1)2 is the Weber
We J
number. Then, Eq. (38) can be written as follows:

. pn+1. =

1
V
6t
. .n+1 0, [ p] = (43)
p =

N
where = 1 . Because the BCC method can be extended in dimension-by-dimension
fashion, we present only one case to solve this. Assume that interface is located between
xi, j,k and xi 1, j,k , i, j,k < 0, and i 1, j,k > 0. Then, the discretized version of Eq. (43)
is as follows:
. . . ..
. p

+ ,
p
i +1/2, i +1, j,k pi, i, j,k pi 1, j,k
i 1/2, x
j,k j,k 6
j,k
6x
6x
. .
pi, j +1,k pi,
+ i, j .
+1/2,k j,k

6y
. . .
i, j pi, j,k pi, j 1,k pi, j,k+1 pi,
.. , 6y + i, j,k+1/2 .
1/2,k j,k
. 6y .. , 6z
pi , j ,k pi , j ,k 1 1 . .
i, j,k1/2 6z = V (44)
6z 6t i,
j,k
p
In the preceding formula, ( ) has discretization with mixed sign on pressure, p. This
x x
mixing sign degrades the numerical solution of the PDE. Hence, the sign is unified by using
the jump condition on p, p = p+ ( ) .
J Sci Comput

. . . ..
. p + ,
p
i +1/2, i +1, j,k pi, i, j,k pi 1, j,k
i 1/2, x
j,k j,k 6
j,k
6x
6x
. . . ..
. p ,
p
i +1/2, i +1, j,k pi, i, j,k pi 1, j,k
i 1/2, x
j,k j,k 6
j,k
6x
6x .
. .
. p p
+
i +1, j,k pi, i, j,k pi, j,k () ,
j,k
i +1/2, j,k i 1/2, j,k
6x 6x
6x

Here, ( ) can be obtained by using interpolation with subcell


resolution.
(45)
( )i, j,k |i 1, j,k | + ( ) i 1, j,k |i, j,k |
( )
=
|i 1, j,k | + |i, j,k |
Substituting this into Eq. (44) and reorganizing the equation, we have

. . . .. ,
pi +1, j,k pi, pi, j,k pi 1, j,k
i +1/2, . i 1/2, x
j,k j,k j,k 6
6x
. . 6x
pi, j +1,k pi,
+ i, j .
+1/2,k j ,k
6y
. . .
i, j pi, j,k pi, j 1,k pi, j,k+1 pi,
.. , 6y + i, j,k+1/2 .
1/2,k j ,k
. 6y 6z
j,k pi, j,k1
..pi,, 1 . i 1/2, j,k ()

.
i, j,k1/2 6z = V +
6z 6t i, j,k (6x)2
(46)

It is important for this modification to never degrade the approximated solution or


change the matrix; that is, it should preserve the SPD system. Thus, we can apply the PCG
method for solving this formula.

3.7 Reinitialization

Although Eq. (16) will move the level set = 0 at the correct velocity, will no longer
be a distance function (i.e., | | /= 1); can become irregular after a certain period of
time. Maintaining as a distance function is essential for providing the interface with a
width
J Sci Comput
fixed in time. Computation of surface tension is difficult near a steep gradient in the
distance function. The values for (), especially for large density ratios, will be greatly
distorted if
| | is far from 1.
Conventional routines for reinitializing a distance function have to explicitly find the
contour, = 0, and reset at all points close to the front. An iteration method for
reinitializing is described in the following.
Given a region, + , with 0 on + and = 0 on , evolve the equation t =
1 | | until reaches steady state. If is already close to a distance function, then
one should not have to evolve too far in time.
Unfortunately, one still has to prescribe boundary conditions on +, which requires
explicitly finding the interface. However, we can eliminate the problem of finding the
inter- face. Consider the following function, 0(x), whose zero level set is the air-liquid
interface: 0(x) need not to be a distance function. We will construct a function, (x),
with the proper- ties that its zero level set is the same as 0(x) and that is the signed
normal distance to the interface. This is achieved by solving the following problem to
steady state:
t = S(0 ) .1 , 2 + 2 + 2 . (47)
x y z
(x, 0) = 0(x) (48)
where S is the signed function [6]. For numerical purposes, it is useful to smooth the sign
function as
0 (49)
S ( ) =
0 ,
02 +
2

However, Russo and Smereka [18] remarked that this approach can cause the zero levelset
to move into the closest grid node because of inconsistency with upwinding approximation
near interface. In order to fix this problem, they suggested the following modified version
of Eq. (47).
. n 6t . .
0
. , n if (i, j, k)
n+1 |i, j,k | Di, j,k
6x
i, j,k sgn i, 6x
0 j,k
i, j,k = . . G ()i, j,k , otherwise (50)
n
6t sgn
i, j,k i, j,k

where . . . . . .
,max a 2 , b2 + max c2 , d 2 + max e2 , f 2 1, if 0 > 0
i,
, + + + (51)
G()i, j,k = j,k
max a2 , b .

2
. . 2 . . .
+ max c , d 2 + max e2 , f 2 1, if 0
<0
+ + + i, j,k
0
26i, j,k
Di, j,k = . .2 .
1

.. 0 . 0 0 .2 . 2

0 0
2
i +1, j,k i 1, + i, j +1,k i, j 0 + i, j,k+1 i,
(52)
j,k 1,k j,k1

i, j,k i 1, i +1, j,k i,


with a Dx i, j,k = j,k , b D+x i, j,k = j,k
6x 6x
c Dy i, j,k i, j,k i, j , d D+yi, j,k = i, j +1,k i,
1,k j,k

z D i,
= e 6x , f D+z i, j,k = 6x
i, j,k i, i, j,k+1 i,
j,k = j ,k 1 j,k
6x 6x
0 0 0 0 0 0
6x = {(i, j, k)|i, j,ki 1, j,k < 0 or i, j,ki +1, j,k < 0 or i, j,ki, j +1,k < 0
0 0 0 0 0 0
or i, j,k i, j < 0 or i, j,k i, < 0 or i, j,k i, < 0}
1,k j,k+1 j,k1
Instead of using Eq. (52) we used a more robust formula based on Eq. (17) in [18] to
avoid the denominator in Eq. (52) becoming very small. Since this approximation gives
only first-order accurate in space, we chose the technique in [19] to apply 3rd ENO
approximation on spatial derivatives.

3.8 Boundary Conditions

Because the computational domain is not a regular grid, it is necessary to establish the
boundary conditions. This is the primary contribution of this work. We use a subcell
resolution technique for each case:
J Sci Comput

Fig. 1 Points near the boundary of the domain. Blue circles are related only to one point inside the domain,
whereas red squares are related to two points (Color figure online)

3.8.1 Advection/Viscous Terms

Define the interpolated boundary level set (c)i +1/2, j,k , (c)i, j +1/2,k , (c)i, j,k+1/2 as
fol- lows:
1 . .
(c )i +1/2, j,k = (c )i, j,k + (c )i 1, j,k (50)
2
1 . .
(c )i, j +1/2,k = (c )i, j,k + (c )i, j 1,k (51)
2
1. .
(c )i, j,k+1/2 = (c )i, j,k + (c )i, j,k1 (52)
2
According to the sign of c, we can determine whether a given domain is inside or
outside. For example, consider the case when (c)i +1/2, j,k < 0, (c)i +3/2, j,k > 0,
which is represented by blue circles in Fig. 1.
For velocity advection and viscous terms, define ui +3/2, j,k as
2u + (2 2)ui +1/2, j,k + ( 2 + 1)ui 1/2, j,k
ui +3/2, j,k = (53)
2+

where
|(c)i +1/2, j,k | (54)

=
|(c)i +1/2, j,k | + |(c)i +3/2,
j,k |

This setup provides the velocity boundary condition on the pipe at second-order
accuracy, which is consistent with our seond-order accurate time integration. In the ENO
3rd method, we require more than one stencil point. Establish points on the stencil outside
D
J Sci Comput
the domain as large values with large variation. For example, in the
(i, j,preceding
k) case,
establish ui +5/2, j,k = (10t)10,
where t = ((i + 5/2) (i + 1/2)).
In red squares, two different values can be assigned for one point. Instead of using these
two values independently, we use bilinear interpolation to save memory in the CPU.
p8 p7
Fig. 2 Cell near the boundary C
region s p6
p5 t

sl sp
sf

p3
A p4
sb
p1 B p2

3.8.2 Projection Term

In projection step, we need to solve the Poisson equation with Neumann boundary
condition:
. .
p =0 (55)
n

Here, we extended the direct discretization in [20] to three dimensional space. Assume that
lies in the cell Ci, j,k . Now, applying divergence theorem to the integral form of (43) on
Ci, j,k results in
, . , ,
.
pn+1 dV = pn+1 nd 1 VdV

Ci, (Ci, j,k) S = Ci, j,k 6t
j,k (55)

where n is outward normal to interface.


As one example, consider a cell cut by the boundary pipe as shown in Fig. 2. In this figure,
polyhedron with bases p5C D, p1 AB is Ci, j,k . Approximating bases as 6 p5C D, 6
p1 AB and pipe region as rectangle ABC D and applying boundary condition of no flow
normal to pipe surface s p , p n = 0, we can have the following discretization of
lefthand side of Eqn (55).
,
p n+1 n d S , plsl +i, j + 1 ptst + i, j 1 pbsb +i,
j,k
1p fs f
(Ci, j,k) 2
,k ,k j,k+
i 1
2 2

(56)

where pi 1, 6x
pl = pi, j 1,k pi, j,k
j,k pi,
pb = j,k
6y
J Sci Comput

pi, j +1,k pi,


, pt
j,k
=
6y
, pf pi, j,k+1 pi, j,k
= 6z
and sl, st, sb, s f , sp are the area of each surface designated in Fig. 2. For the discretization
of righthand side of (55), we can assume that numerical quantities of 16t V is cell-wise
constant. Since general finite difference discretization is used for cells in , this
cell-wise constant assumption guarantee the consistency of discretization in whole domain.
Hence the righthand side of (55) will be
,
1 1 .

Ci, j,k 6t V dV ( V ) . V olume( Ci, j,k) (56)


6t i, j,k

This boundary treatment technique preserves symmetry positivity of the corresponding


matrix so we can apply efficient solver to solve poisson equation.

3.9 Time Discretization and CFL Condition

3.9.1 Time Discretization

For time discretization, we select the second-order TVD Runge Kutta method. This can be
written as a convex combination of velocity update procedure. Let us define the update
step as

n+1= U . nV (61)
. V
Then, the second-order TVD Runge Kutta method is
(62)
1 .
n+1 1
V
+ U U .
n 2
= V V
2
n ..

It is important that all time advancing steps, including the level set update, reinitialization,
and velocity update, are approximated by the second-order TVD Runge Kutta method.
Otherwise, the order of the entire simulation can be degraded, which causes severe errors
in the final result. More details are provided in a previous study [16].

3.9.2 CFL Condition

Except for the implicit term, which is unconditionally stable, a CFL time restriction may
be needed for stability. For this study, the time condition is presented in dimensionless
terms [7].
1. Advection Term
Dimensional Form
|u| |v| |w|max
A +
max 6z
max +
c= 6y
Dimensionless 6x
Form
|u|max |v|max |w|max
Ac = + +
6x 6y 6z

2. External Force Term


Dimensional
Form
,
.. .|g|+
.
|f|
1 ,oil)
min(water
.
Fc =
6x
Dimensionless Form
., . .
.R 1
|g|+ min( |f|
. V0 2 water ,oil )

Fc = 6x

3. Surface Tension Term
| |
Dimensional
Form

Sc = oil
min( , wat )(min(6x 6y, 6z))2
er ,
Dimensionless Form
1
|
We |
Sc =
min(, 1)(min(6x , 6y, 6z))2
From the semi-implicit viscous solver, the CFL condition for the viscous term
. .. .
oil water 2
Vc = max ,
oil water (6x)2 + (6y)2 + (6z)2
is released.
In our simulation, a CFL restriction of 12 is used. Hence,
. , .
( A )2 + 4(F )2 + 4(S )2 1
+ c c c
6t A c
2 2

Remark Note that we limit upper bound of curvature with


1
| |
min(6x, 6y, 6z)
so that prevent large curvature near the principal axis from contributing erroneous large
surface tension forces. This ensures that CFL restriction on surface tension term is O(h3/2)
in uniform grid which is simlar as in [17].

4 Numerical Experiments

We experimentally examine the simulation of the behaviors of coreannular flow. In partic-


ular, we focus on the simulation of parallel pipe cases, which are not axisymmetric.

4.1 Upflow Case for Validation

We examine an upflow example for validation of our non-axisymmetric based simulation.

4.1.1 Example 1

We simulate the case with [Q w , Q o ] = [200, 429]cm3min1. We set a = 1.28 according


to equation (3.4) in a previous study [2], with experimental hold-up ratio h = 1.39. The
experiments are performed using a pipe of radius R2 = 0.47625, with undisturbed
interface, R1 = R2/a. By fixing a and Vo, where Vo = Qo/( R 2) = 10.034 cm s1, we
can use the other parameters from equation (18.15) in Li and Renardy [21] as follows:
2
Re1 = 0.9498, m = 0.00166, = 1.1, J = 0.07961, K = 0.4552
C = 0.001, = 2.0, V
0 = 16.9531
(a) t= 0 (b) t = 20 (c) t = 40 (d) t = 105 (e) t = 140

Fig. 3 Up-flow with Re1 = 0.94938, a = 1.28, m = 0.00166, = 1.1, J = 0.07961, K = 0.4552,
V0 = 10.034 cm s1, = 2.0, C = 0.001

Fig. 4 The contour of pressure field on cross sectional plane including centerline corresponding to the
interface at t = 140

In dimensionless form, the perturbed amplitude will be C = C/R 1 . This simulation is


performed on a 51 51 51 mesh over one spatial period on domain [1.57, 0,
1.57] [1.57, 3.14, 1.57]. Our results for this case are shown in Fig. 3. For
comparison with previous work [2,9], the pressure profile corresponding to the interface
on a cross-sectional plane,
including the centerline, is presented in Fig. 4.
(a) t= 0 (b) t = 20 (c) t = 40 (d) t = 95 (e) t = 130
Fig. 5 Up-flow with Re1 = 3.73754, a = 1.61, m = 0.00166, = 1.1, J = 0.063354, K = 2.030303,
V0 = 83.91 cm s1, = 2.4, C = 0.1

Fig. 6 The contour of pressure field on cross sectional plane including centerline corresponding to the
interface at t = 20

4.1.2 Example 2

In this example, we establish an amount of water to be relatively large compared with that
in example 1, a = 1.61. Corresponding parameters for this simulation are as follows:

Re1 = 3.73754, m = 0.00166, = 1.1, J = 0.063354, K = 2.030303


C = 0.1, = 2.4, V
0 = 83.91
Fig. 7 Schematic representation of the oilwater loop, (o) oil reservoir, (w) water reservoir, (l) laminar
flowme- ter, (e) electromagnetic flow meter, (t) thermocouple, (m) two-phase mixer, (d) flow development
region, (f ) fully developed flow, (c) capacitance pressure transducer, (s) separator tank

Fig. 8 Sotgia et al. [22]. The periodicity of wave is varying depending on the oil and water velocity

This simulation is performed on a 71 51 71 mesh over one spatial period on domain


[1.8326, 0, 1.8326] [1.8326, 2.618, 1.8326]. The results are shown in Figs. 5
and 6. Including the asymmetric aspect of the crest, these results agree with those in
previous studies
[2,9].

4.2 Horizontal Flow with Gravity Effect

Now, we discuss the cases of horizontal flow in a pipe. In Sotgia et al. [22], they
performed experiments based on their test facility with 10-m long pipeline in Fig. 7. This
is quite long campared with its cross-sectional raidus which give us a limitation of
computation. Also, It is known that there exist parameters which make fluid motion to be
periodic for coreannular
(a) t=0 (b) t = 10

(c) t = 25 (d) t = 40

(e) t = 55 (f) t = 65

Fig. 9 Horizontal flow with Re1 = 3.73754, a = 1.61, m = 0.00166, = 1.1, J = 0.063354, K =
2.030303, V0 = 83.91 cm s1, = 2.4, C = 0.01

flow by experiments in [22]. In this paper, we are focusing on periodic regime for
comparing with recent results in [4] by choosing parameters as the same with the one of
periodic regime case. We assume that the flow moves from right to left by negative
pressure gradient.

4.2.1 Example 3

In this example, we add more water than in the vertical case, and let the core-centerline be
sufficiently fast for choosing similar parameter with [4,22]. Corresponding parameters are
as follows:

Re1 = 3.73754, m = 0.00166, = 1.1, J = 0.063354, K = 2.030303


C = 0.01, = 2.4, V
0 = 83.91

Except for the amplitude of initial wave, parameters are as same with as those in example
2. Because water is relatively heavier than oil, a large amount levitates the oil. We can
verify this effect from Figs. 9 and 10 and we can see the flow pattern quitely agrees on
experimental case, Fig. 8.

4.2.2 Example 4

We choose similar parameters to those in example 1, except K .

Re1 = 0.9498, m = 0.00166, = 1.1, J = 0.07961, K = 0.9993


C = 0.001, = 2.0, V
0 = 53.4918
In this case, the density difference between two fluids is relatively large (1 2 =
0.090), otherwise the centerline velocity will be relatively low. Hence, we can expect that
the oil will
Fig. 10 The contour of pressure field on cross sectional plane including centerline corresponding to the
interface at t = 65

(a) t=0 (b) t = 20

(c) t = 40 (d) t = 60

(e) t = 80

Fig. 11 Horizontal flow with Re1 = 0.94938, a = 1.28, m = 0.00166, = 1.1, J = 0.07961, K =
0.9993, V0 = 53.4918 cm s1, = 2, C = 0.001

first move downward due to the gravity. However, a few seconds later, as the total velocity
increases, oil begins to levitate. Additionally, we can verify that the total amount of water
is increased as time passes. The results are shown in Figs. 11 and 12.

4.2.3 Example 5: Effect of Centerline Velocity

To verify the effect of changing the centerline velocity, we change K from the value in
example 4 into K = 0.4552 and K = 2.030303. Figures 6 and 7 show each result.
We
Fig. 12 The contour of pressure field on cross sectional plane including centerline corresponding to the
interface at t = 80

(a) t=0 (b) t = 20

(c) t = 40 (d) t = 60

(e) t = 80
Fig. 13 Horizontal flow with Re1 = 0.94938, a = 1.28, m = 0.00166, = 1.1, J = 0.07961, K =
0.4552, V0 = 10.034 cm s1, = 2, C = 0.001

Fig. 14 The contour of pressure


field on cross sectional plane
including centerline
corresponding to the interface at
t = 80
(a) t=0 (b) t = 20

(c) t = 40 (d) t = 60

(e) t = 80

Fig. 15 Horizontal flow with Re1 = 0.94938, a = 1.28, m = 0.00166, = 1.1, J = 0.07961, K =
2.030303, V0 = 83.91 cm s1, = 2, C = 0.001

Fig. 16 The contour of pressure


field on cross sectional plane
including centerline
corresponding to the interface at
t = 80

can conclude that the centerline velocity of the core flow affects the wavy crest and
thickness of the core flow and the height of levitation. Because the amount of water is
relatively low compared to example 3, the height of levitation is low. Additionally, we can
verify that when core flow moves around the pipe, a wavy interface occurs because of the
buoyancy force from the water near the pipe. Figs. 13, 14, 15, and 16 give results.

4.2.4 Example 6: Effect of Finite Amplitude on Core Flow

In this example, we vary the finite amplitude from that in example 4 to C = 0.01 and
C = 0.05. Figures 17 and 19 show each result and Figs. 18 and 20 show pressure. The
perturbed amplitude for the initial pipe can affect the pressure around the pipe wall. The
pipe
with relatively large amplitude (Fig. 19) tends to be smoother than the other, and carries
more of the core flow due to the initial pressure profile.
(a) t=0 (b) t = 20

(c) t = 40 (d) t = 60

(e) t = 80

Fig. 17 Horizontal flow with Re1 = 0.94938, a = 1.28, m = 0.00166, = 1.1, J = 0.07961, K =
0.9993, V0 = 53.4918 cm s1, = 2, C = 0.01

Fig. 18 The contour of pressure field on cross sectional plane including centerline corresponding to the
interface at t = 80

4.2.5 Example 7: Effect of Radius Ratio

Next, we verify the effect of changing the amount of water by controlling a. Let us change
a to a = 1.61. The amount of water plays an important role in the levitation of the core
flow. As the amount of fluid in the annulus grows, the buoyancy force from the annulus
strengthens.
This makes the core fluid levitate from the pipe wall. Flows and pressure profiles can be
verified in Figs. 21 and 22.
(a) t=0 (b) t = 20

(c) t = 40 (d) t = 60

(e) t = 80

Fig. 19 Horizontal flow with Re1 = 0.94938, a = 1.28, m = 0.00166, = 1.1, J = 0.07961, K =
0.9993, V0 = 53.4918 cm s1, = 2, C = 0.05

Fig. 20 The contour of pressure field on cross sectional plane including centerline corresponding to the
interface at t = 80

4.2.6 Example 8: Effect of Reynolds Number

Finally, we investigate the effects of several different Reynolds numbers, Re1, by varying
the Reynolds number in example 3 to Re1 = 1.5 and Re1 = 2.5. As the Reynolds number
grows, the viscosity effect of the core flow decreases. Hence, we can expect that the core
flow will levitate to a relatively high height when the Reynolds number is small. Figures
23 and 24 show that this expectation is correct.
(a) t=0 (b) t = 10

(c) t = 25 (d) t = 40

(e) t = 55 (f) t = 65

Fig. 21 Horizontal flow with Re1 = 0.94938, a = 1.61, m = 0.00166, = 1.1, J = 0.07961, K =
0.9993, V0 = 53.4918 cm s1, = 2, C = 0.01

Fig. 22 The contour of pressure field on cross sectional plane including centerline corresponding to the
interface at t = 60

4.3 Time for Computing

For simulations, we use a computer with a 3.5 Ghz Intel(R) i7-2700K processor.
Simulation is coded by using Visual C++, OpenGL as a graphical tool, and Boost as a
multithreading tool. We use eight cores for multithreading. Total and simulation times for
one step of each simulation are represented on the right edge of each figure.
(a) t=0 (b) t = 25

(c) t = 40 (d) t = 65

Fig. 23 Horizontal flow with Re1 = 1.5, a = 1.61, m = 0.00166, = 1.1, J = 0.063354, K =
2.030303,
V0 = 83.91 cm s1, = 2.4, C = 0.01

(a) t=0 (b) t = 25

(c) t = 40 (d) t = 65

Fig. 24 Horizontal flow with Re1 = 2.5, a = 1.61, m = 0.00166, = 1.1, J = 0.063354, K =
2.030303,
V0 = 83.91 cm s1, = 2.4, C = 0.01

5 Conclusion and Future Work

We present simulations of coreannular flow with a large viscosity ratio in a horizontal


pipe equipped with a level set method. Comparing this result with recent work [4], we
provide a full 3D description of coreannular flows in parallel pipes with several
examples. By selecting different parameters, we can ensure that the core fluid flows
downward due to the relatively large density difference. Additionally, in our simulation,
the oil-water interface is described accurately by means of the level set function. To the
best to our knowledge, full 3D simulations on horizontal pipes have not been conducted
before. Such simulations can anticipate real phenomena in transporting viscous liquid such
as crude oil.
Because analysis on horizontal pipes is not fully established, we have no choice but to
select parameters from the vertical case. In reality, the assumption of periodicity will not
be accurate for some regime, which is a limitation of our simulation. In future work, we
will consider finding a proper parameter for efficient transportation based on the foregoing
analysis. Additionally, we will extend the given simulation to long pipes without
periodicity assumption, and differently located pipes (such as inclined or curved).
Acknowledgments This work was supported by Basic Science Research Program through the National
Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT and Future Planning
(2014R/A2A/A10050531).

References

1. Bai, R., Chen, K., Joseph, D.D.: Lubricated pipelining: stability of coreannular flow. Part 5.
Experiments and comparison with theory. J. Fluid Mech. 240, 97142 (1992)
2. Li, J., Renardy, Y.Y., Renardy, M.: Direct simulation of unsteady axisymmetric coreannular flow with
high viscosity ratio. J. Fluid. Mech. 391, 123149 (1998)
3. Li, J., Renardy, Y.Y., Renardy, M.: A numerical study of periodic disturbances on two-layer Couette
flow. Phys. Fluids. 10(12), 30563071 (1998)
4. Ooms, G., Pourquie, M.J.B.M., Beerens, J.C.: On the levitation force in horizontal coreannular flow
with a large viscosity ratio and small density ratio. Phys. Fluids 25, 032102 (2013)
5. Osher, S., Setian, J.A.: Fronts propagating with curvature dependent speed: algorithms based on
Hamilton Jacobi formulations. J. Comput. Phys. 79, 1249 (1988)
6. Sussman, M., Smereka, P., Osher, S.: A level set approach for computing solutions to incompressible
two-phase flow. J. Comput. Phys. 114, 146159 (1994)
7. Kang, M., Fedkiw, R.P., Liu, X.D.: A boundary condition capturing method for multiphase
incompressible flow. J. Sci. Comput. 15, 323360 (2000)
8. Liu, X.D., Fedkiw, R.P., Kang, M.: A boundary condition capturing method for Poissons equation on
irregular domains. J. Comput. Phys. 160, 151178 (2000)
9. Kang, M., Shim, H., Osher, S.: Level set based simulations of two-phase oilwater flows in pipes. J.
Sci. Comput. 31, 153 (2006)
10. Osher, S., Fedikew, R.: Level Set Methods and Dynamic Implicit Surfaces. Springer, New York (2003)
11. Chorin, A.J.: Numerical solution of the NavierStokes equations. Math. Comput. 22, 745 (1968)
12. Fortin, M., Peyret, R., Temam, R.: Rsolution numrique des quations de NavierStokes pour un
fluide incompressible. J. Mcanique 10, 357390 (1971)
13. Peyret, R., Taylor, T.D.: Computational Methods for Fluid Flow. Springer, New York (1982)
14. Osher, S., Shu, C.W.: High-order essentially non-oscillatory schemes for HamiltonJacobi equation.
SIAM J. Numer. Anal. 28, 907922 (1991)
15. Jiang, G.S., Peng, D.: Weighted ENO schemes for HamiltonJacobi equations. SIAM J. Sci. Comput.
21, 21262143 (2000)
16. Shu, C.W., Osher, S.: Efficient implementation of essentially non-oscillatory shock capturing schemes.
J. Comput. Phys. 83, 32 (1988)
17. Brackbill, J.U., Kothe, D.B., Zemach, C.: A continuum method for modeling surface tension. J.
Comput. Phys. 100, 335354 (1992)
18. Russo, G., Semereka, P.: A remark on computing distance functions. J. Comput. Phys. 163, 5167 (2000)
19. Chn, A., Min, C., Gibou, F.: Second-order accurate computation of curvatures in a level set
framework using novel high-order reinitialization schemes. J. Sci. Comput. 35, 114131 (2008)
20. Purvis, J.W., Burkhalter, J.E.: Prediction of critical mach number for store configurations. AIAA J. 17,
11701177 (1979)
21. Joseph, D.D., Renardy, Y.Y.: Fundamentals of Two-Fluid Dynamics, Part II, Lubricated Transport,
Drops and Miscible Liquids, Springer, New York (1993)
22. Sotgia, G., Tartarini, P., Stalio, E.: Experimental analysis of flow regimes and pressure drop reduction
in oilwater mixtures. Int. J. Multiph. Flow 34, 1161 (2008)