You are on page 1of 10

Chemical Engineering Science 65 (2010) 5160–5169

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage:

Interfacial friction factor determination using CFD simulations in a horizontal

stratified two-phase flow
Kamel Sidi-Ali a,b,, Renée Gatignol a
Institut Jean Le Rond d ’Alembert, Université Pierre et Marie Curie et CNRS 4, Place Jussieu, Case courrier 162, 75252 Paris Cedex 05, France
Commissariat a l’Energie Atomique, Centre de Recherche Nucléaire de Draria, BP 43, Sebbala, Draria Alger 16003, Algeria

a r t i c l e in f o a b s t r a c t

Article history: Stratified two-phase gas–liquid flow in three-dimensional (3D) is challenging because of the difficulty
Received 15 December 2009 in determining the shear stresses at the walls and at the gas–liquid interface. The study is simplified by
Received in revised form considering only the gas and the gas–liquid interface. The gas flows between the fixed wall of the pipe
6 May 2010
and the interface, which is regarded as wall moving at liquid velocity. It is shown, using a CFD code, that
Accepted 14 June 2010
one can determine the correct horizontal and vertical profiles of the longitudinal velocity of the gas.
Available online 1 July 2010
Seven experiments done by Strand (1993) for a horizontal stratified gas–liquid flow are simulated. The
Keywords: simulation results are compared with experimental results, and also to the numerical results of
Two-phase flow Meknassi et al. (2000). We deduce from these comparisons that the approach suggested in this work is
Stratified flow
very valuable. Then expressions for calculating the gas–wall friction factor, and the gas–liquid interface
Gas–liquid interface
friction factor are presented. These factors are crucial for evaluating the shear stresses. It appears, after
Velocity profile
Shear stress comparison with the experimental values that the correlation proposed for calculating the friction
Friction factor factor at the interface, in the present work, agree with the experimental values better than that given by
Taitel and Duckler (1976a, b).
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction SG twG pG ts I þ rG SG gsind ¼ 0 ð2Þ
Multiphase flows are met in chemical industry, space industry, Fig. 1represents a stratified gas–liquid flow in a pipe of circular
nuclear and oil plants. Multiphase flows are very complex to cross section slightly tilted downward in the gravity field. In this
model, there is no unique model that can satisfy the multitude of figure, the entries of the two fluids are shown, where the gas
configurations that a two or more phase flow can take. Two-phase velocity is UG and the liquid velocity is UL. In the cross-section, the
flow configurations are described in detail for example, in Ishii interface is denoted I, the perimeter of the pipe wetted by gas is
(1975), Delhaye et al. (1981), Nigmatulin (1991), and Ishii and pG , the one wetted by the liquid is pL , and the height of the liquid
Hibiki (2006). The stratified two-phase gas–liquid flow is one of is hL. In Eqs. (1) and (2), SL and SG are areas of the slots,
these flow configurations. respectively, for liquid and gas, dp/dx is the pressure drop per unit
Among the first works on stratified flows gas–liquid, there are length of the pipe and d is the angle of inclination of the pipe,
those of Taitel and Duckler (1976a), who have studied these flows which is zero when the pipe is horizontal, Taitel and Dukler
in the case of an inclined cylindrical pipe (Taitel and Duckler, (1976b). Finally, twG , twL and ts are, respectively, the wall–gas, the
1976a) (Fig. 1), and in the case of a horizontal pipe (Taitel and wall–liquid, and the interface shear stresses.
Dukler, 1976b). Their model is based on the resolution of a system The analysis of the model of Taitel and Duckler (1976a, b) shows
of two equations, one equation written for the liquid phase (1) that it is based on data related to the geometry of the flow and
and another written for the gas phase (2). This model takes into the wall–fluid and the interface shear stresses. The shear stresses are
account the geometry of the flow, and different fluid–wall and the only quantities that may differ from one study to another. The
interface shear stresses: difference is in the choice of the fluid–wall friction factor appearing in
  the fluid–wall shear stress expression given by (3). The first
dp correlation giving fluid–wall friction factors according to the Reynolds
SL twL pL þ ts I þ rL SL gsind ¼ 0 ð1Þ
dx number for one-phase flow was established by Blasius (1912). For a
two-phase flow, Agrawal et al. (1973) introduce a new definition of
 Corresponding author at: Commissariat a l’Energie Atomique, Centre de the hydraulic diameter with which the Reynolds number is
Recherche Nucléaire de Draria, BP 43, Sebbala, Draria Alger 16003, Algeria.
calculated. For a laminar or a turbulent flow, Taitel and Duckler
Tel.: + 213 21 31 01 75; fax: + 213 21 31 03 32. (1976a, b), Cheremisinoff and James Davis (1979), and Shoham and
E-mail addresses:, (K. Sidi-Ali). Taitel (1984) use two different friction factor expressions, one for the

0009-2509/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169 5161

Fig. 1. Stratified gas–liquid flow in an inclined pipe, Taitel and Duckler (1976a).

liquid–wall region and another for the gas–wall one. Spedding and Table 1
Hand (1997) introduce the superficial Reynolds number, as defined Interface gas–liquid friction factors fs.
below, and the liquid fraction when the flow is laminar. Petalas and
Authors Interface gas–liquid friction factor
Aziz (1998) propose a laminar friction factor according to the
superficial Reynolds number and use the Blasius equation as a Agrawal et al. (1973) fs ¼ 2[0.804(ResG)  0.285]2
turbulent friction factor. The expressions for the shear stresses, Taitel and Dukler (1976b) fs ¼ 0.046 (ReG)  0.2
twL , twG , and ts are given Cheremisinoff and James Davis (1979)
Low amplitude waves fs ¼ 0.00355
2 2 2 Large amplitude waves fs ¼ 0.0080+ 2  10  5 (ReL)
L UL r
G UG rG ðUG UL Þ Shoham and Taitel (1984) fs ¼ 0.0142
twL ¼ fwL , twG ¼ fwG , ts ¼ fs ð3Þ
2 2 2 Kowalski (1987)
Low amplitude waves fs ¼ 0.96 (ResG)  0.52
Large amplitude waves fs ¼ 7:5  105 a0:25 ðReG Þ0:3 ðReL Þ0:83
In these expressions, the friction factors fwL, fwG and fs are, Crowley et al. (1992) fs ¼ 0.005
" #
respectively, the liquid–wall, the gas–wall and the interface Petalas and Aziz (1998) rL dg
fs ¼ ð0:004 þ 0:5106 ResL ÞFr 1:335
friction factors. In the case of turbulent–turbulent gas–liquid flow, rG UG2
two of these factors apply (Newton et al., 1999; Kowalski, 1987):

fwL ¼ 0:046ðReL Þ0:2 and fwG ¼ 0:046ðReG Þ0:2 ð4Þ

ReL and ReG are, respectively, the Reynolds numbers of liquid and
gas given by: ReL ¼ UL dhL =nL and ReG ¼ UG dhG =nG . Superficial
Reynolds numbers for liquid and gas are defined by:
ResL ¼ UsL dhL =nL and ResG ¼ UsG dhG =nG , where UsL and UsG are the
superficial velocities of liquid and gas based on the liquid fraction
a and are calculated by: UsL ¼ aUL and UsG ¼ ð1aÞUG . The liquid
fraction in these last two expressions is given by a ¼ SL =S, where
S is the area of total cross-section of the pipe. The quantities nL
and nG are the kinematic viscosities of the liquid and the gas, and
dhL and dhg are, respectively, the hydraulic diameters of liquid and
gas, given by: dhL ¼ 4SL =pL and dhG ¼ 4SG =ðpG þ IÞ.
While the gas–wall and the liquid–wall friction factors give
generally good results, it is not the same about the interface friction
factor. Different expressions, for calculating the interface friction
factor proposed in the literature, are presented in Table 1. Among
the earliest expressions established, that of Agrawal et al. (1973) is
Fig. 2. Gas–liquid stratified flow.
written as a function of Reynolds number ResG based on the
superficial gas velocity. Taitel and Duckler (1976a, b) propose an
expression depending on the Reynolds number of gas ReG.
Cheremisinoff and James Davis (1979) propose two expressions. present here a numerical method, based on the transposition of
The first expression is equal to a constant and is valuable for a the geometry of the flow and using the CFD code (Fluent, 2006).
stratified flow with low amplitude waves, and the second one The stratified two-phase gas–liquid flow, as shown in Fig. 2, is
depends on the Reynolds number of liquid ReL as a linear function assumed stationary. The flow takes place in a cylindrical pipe
for a stratified flow of large wave amplitudes. Other authors use disposed horizontally. The objective is the CFD simulation, using
constant friction factors as that given by Shoham and Taitel (1984) the CFD code Fluent, of seven experiments conducted by Strand
or the other proposed by Crowley et al. (1992). Kowalski (1987) also (1993). The last part will be dedicated to determine an expression
proposes two expressions, the first one for a stratified flow with low for the gas–liquid interface friction factor.
amplitude waves, depending on the Reynolds number based on the
superficial gas velocity. The second expression for a stratified flow 1.1. Basic principles of the method
with large wave amplitude is a function of Reynolds numbers of gas
and liquid and the liquid fraction a. Finally, we quote the expression The flow is divided into two parts, one with the gas, the other
given by Petalas and Aziz (1998); written as a function of several with the liquid. For gas, we assume that it flows through a pipe of
quantities including the Froude number of liquid FrL. same section as the actual flow and the interface is considered as
The various methods for determining correlations giving the a moving wall. The problem then reduces to a single phase flow in
friction factors can be theoretical, experimental or numerical. We the upper part of the pipe (Fig. 3), with an interface moving with
5162 K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169

where r is the gas density, U the velocity vector of gas, p
the pressure, m the shear viscosity and g denotes the gravity
For turbulence, the k2e model of Launder and Spalding (1972)
is used. The kinetic energy of turbulence k and its dissipation rate
e are obtained by solving the two following transport equations:
@ @ @ m @k
ðrkÞ þ ðrkU i Þ ¼ mþ t þ Gk re ð7Þ
@t @xi @xj sk @xj
@ @ @ mt @e e e2
ðreÞ þ ðreUi Þ ¼ mþ þ C1e Gk C2e r ð8Þ
@t @xi @xj se @xj k k
where Gk is the production term of turbulence and mt the
turbulent viscosity given by
mt ¼ rCm ð9Þ
The model constants of Launder and Spalding (1972) are,
Fig. 3. Studied configuration.
C1e ¼ 1:44, C2e ¼ 1:92, Cm ¼ 0:09, sk ¼ 1:0 and se ¼ 1:3.

liquid velocity. The study is done in 3D. In this study, Gambit is 2.1. Boundary conditions
used to mesh the considered computational domain and the CFD
code Fluent for solving the equations governing the flow. The gas For the gas, a uniform velocity profile at the entry is used:
is turbulent air and we consider that there is no heat or mass U ¼ U0 ð10Þ
transfer between gas and the moving wall, and between gas and
A non-slip condition is imposed on the upper wall of the pipe:
pipe wall. Gravity is taken into account.
The validation of this work will be done in comparison with two UG ¼ 0 at the wall ð11Þ
studies, one experimental and one numerical. The experimental work
is that of Strand (1993), who made several experiments on a water– The model of two-layers is used for numerical computation
air flow and provides a layered set of horizontal and vertical profiles of turbulent flow in region located near the wall. In this approach,
of longitudinal velocities. In these experiments, Strand (1993) the whole domain is subdivided into a viscosity-affected
observes three types of profiles, the first type for small gas velocities region and a fully turbulent region. The demarcation of the two
where the profile has a maximum at the center of the flow of the gas regions is determined by a wall-distance-based, turbulent
phase, the second type on average velocities of gas where the Reynolds number, Rey, defined as Rey  ry k=m where y is the
maximum velocity is shifted toward the bottom of the flow and the distance of the considered point to the wall, as presented
third type for a high gas velocity where the maximum velocity is by Wilcox (1993) The first numerical grid near the wall is located
shifted toward the top of the pipe, this latter type is not studied in this far enough from the wall, in order for the turbulent Reynolds
work. The numerical work is that of Meknassi et al. (2000), who number to be greater than unity, this means that turbulence
developed a tool for numerical calculation based on the finite volume forces are more important than viscous forces. The two expres-
method to simulate the experiments of Strand (1993). In Meknassi sions of the velocity given by (12), and used in this work, were
et al. (2000), the pressure gradient, the phase distribution and the introduced by Prandtl (1927, 1932) and are universal laws of the
equivalent roughness at the interface were imposed and constant wall:
along the pipe. The equivalent roughness, expressed as height of 8 þ þ
< y ,y ryv
roughness was extracted from the experimental work of Strand >
(1993), and the shear stress at the gas–liquid interface was written ¼ 1 ð12Þ
U > : LnðEy þ Þ, y þ Z yvþ
according to the height of roughness. The authors had performed kVK
iterations in the gas phase to approach the gas–wall shear stress and
where the local friction velocity U* is calculated thanks to
the interface shear stress. Then they resumed the iterations in the 1=4
U  ¼ Cm k1=2 , and where y þ  rU  y=m is the dimensionless
liquid phase with the interface shear stress already calculated to
position of y, and yv+ the dimensionless thickness of the viscous
approximate the wall–liquid shear stress. Once the calculations
sub-layer equal to 11.225. The constant E is equal to 9.793 and the
converged, they draw the horizontal and vertical profiles of long-
Von Karman constant kVK is equal to 0.4187.
itudinal velocities to compare with the profiles obtained by Strand
The diffusion fluxes for all quantities in the exit direction are
(1993). The relative differences between experimental values and
equal to zero as proposed by Chung (2002):
those of Meknassi et al. (2000) for wall–gas and interface shear
stresses vary from 1% to 22%. @Ui @k @e
¼ 0, ¼ 0, ¼0 ð13Þ
@n @n @n
with ð@=@nÞ the derivative with respect to the normal direction to
2. Equations set-up the exit surface.
The interface is supposed to have a constant velocity which is
For the gas phase, the laws of mass and momentum balance equal to the liquid velocity, and the gas velocity at the interface is
are written as equal to the interface velocity:
@r ! UG ¼ Us ¼ UL at the interface ð14Þ
þ r  ðr U Þ ¼ 0 ð5Þ
At the interface, the model of two-layers is used for numerical
@ ! !! ! ! ! computation of turbulent flow in regions located near the
ðr U Þ þ r  ðr U U Þ ¼ rp þ r g þ r  ½mðr U þ ðr U ÞT Þ ð6Þ
@t interface. The expression of the velocity given in (12) is used.
K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169 5163

3. Numerical method and meshing of the computational


3.1. Numerical method

The governing equations of the flow are solved with the finite
volume method. This method is based on the integration of the
balance laws, written as integrals on elementary volumes, as
explained by Hirch (1991) and Leveque (2002).
The discretization of the equations is classical and can be
illustrated on the steady-state conservation equation for the
transport of a scalar quantity f. This equation, in integral form for
an arbitrary control volume V, is given by
! ! !
rf U  d A ¼ Gf rf  d A þ Sf dV ð15Þ
where A is the surface vector, Gf the diffusion coefficient, rf the
gradient of f and Sf the source term per unit volume. Fig. 5. Meshing of the gas entry.
Eq. (15) is applied to each control volume, or cell, in the
calculation domain, and is discretized as given by the liquid height hL, it varies between 42.9 and 26 mm (Table 2). The
X Nfaces
X 3D area is meshed with Gambit, using ‘‘wedges’’. The total number
! ! !
rf ff U f  A f ¼ Gff rff  A f þSf V ð16Þ of meshes is about 300,000. In the near wall, four rows are taken, the
f f first at 10  4 m, with a growth factor of 1.2 for the following rows.
Several different meshes were tested. Some, creating negative
where Nfaces is the number of faces surrounding the cell, ff is the
! ! volumes which do not allow Fluent to run, and others which make
value of f convected through the surface f, rf ff U f  A f the f flux
! the computation diverge. The number of grids used varies, depending
through the face f, A f the surface vector of the face f, rff the
on the height of the liquid that is the height of the moving wall from
gradient of f at the face f and V the volume of the cell.
the bottom of the pipe, between 300,000 and 140,000 grids.
This equation leads to a discretized equation of the form:
ap fp ¼ af ff þS ð17Þ
4. Simulations
In Fig. 4, is shown a cell where ap fp is calculated, in the center of
the cell at the point ‘‘p’’, as in (17). The index ‘‘f’’ is relative to the The simulations are conducted under steady-state. Data on the
faces ‘‘N’’ for north, ‘‘S’’ for south, ‘‘E’’ for east and ‘‘W’’ for west. geometry of the pipe are related to the height of the liquid hL. The
For more details on the calculations, see Patankar (1980). heights of the liquid are assumed constant along the pipe. Their
values, for different simulations, are taken from Strand (1993) and
3.2. Meshing of the computational domain correspond to the first column in Table 2. The gas velocity in the
inlet section is denoted UGCFD and the interface velocity Us (equal
The computational domain is shown on Fig. 5. The pipe is to the liquid velocity at the entrance). These two velocities are
cylindrical and parallel to the x axis, the inlet section being at x ¼ 0. found by iterations, as is explained in the following paragraph.
In order to simulate the experiences of Strand (1993), the diameter The simulation procedure is initiated by choosing the values of
of the pipe is d ¼ 0.1 m, and x ¼ 10 m is the length of the pipe. As for the velocities UGCFD and Us, for a given height of liquid hL. When the
calculations are converged, the results are analyzed for the
longitudinal velocity in the sections of the pipe. We check for all
the numerical experiments, and for x 41 m, that all the velocity
profiles are identical. In the cross section at x ¼ 5 m, the vertical
profile of the longitudinal velocity, in the plane of symmetry of the
flow, is drawn on (Fig. 7). Then, one can make an estimate of the area
bounded by the profile and compare this area with the area obtained
experimentally. The experimental area is obtained by extrapolating
the four experimental points at the bottom and the top of the
experimental profile to the walls. Iterations on UGCFD and Us will
continue until the two areas are as close as possible. The values of
UGCFD and Us presented in Table 2 are the velocities for which the two
areas are the closest, thus ensuring the correct mass conservation.
The governing equations are solved sequentially. The fact that
these equations are coupled makes it necessary to perform several
iterations of the solution loop before convergence can be reached.
The solution loop (Fig. 6) consists of seven steps that are
performed in sequential order.
The momentum equations for all directions are each solved
using boundary conditions values, in order to update the velocity
field. The obtained velocities may not satisfy the continuity
equation locally. Using the continuity equation and the linearized
momentum equation, an equation for pressure correction is
Fig. 4. Two-dimensional discretized cell. derived. Using this pressure correction the pressure and velocities
5164 K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169

Table 2
Position of the peak and corresponding velocity.

Exp. hL (mm) UGCFD (m/s) Us (¼ UL) (m/s) y peak exp y peak CFD Upeak exp (m/s) Upeak CFD (m/s)

a 42.9 2.30 0.243 0.47 0.45 3.05 3.04

b 41 3.15 0.24 0.44 0.43 4 4.09
c 38.8 5.02 0.285 0.42 0.44 6.3 6.44
d 34.5 6.09 0.331 0.45 0.47 7.6 7.72
e 32.5 6.95 0.354 0.39 0.49 8.6 8.54
f 30 8.10 0.410 0.38 0.47 10.1 10.03
g 26 8.38 0.488 0.44 0.43 10.5 10.47

Fig. 7. Vertical profile.

Fig. 6. Flowsheet of the calculation procedure.

are corrected to achieve continuity. After this, turbulence

equations are solved with corrected velocity field. Then, the fluid
properties and additional source terms are updated. In the last
step, a check for convergence is performed.

5. Computed velocity profiles

The normalized height of the pipe (Fig. 7) is defined by: Fig. 8. Horizontal profile.

y ¼ y=ðdhL Þ, where d is the pipe diameter. The normalized width

of the pipe (Fig. 8) is given by z ¼ ðz þ ‘Þ=2‘, where 2‘ is the chord
at y ¼65 mm. by simulation is y peak CFD , the experimental velocity is Upeak exp and
finally, the velocity peak obtained by simulation is Upeak CFD.
5.1. Vertical profile of the longitudinal velocity The positions of the peak velocities obtained for simulations
(a, b, c, d, g) show differences ranging from 2.2% to 4.7%. For the
We are situated in the cross section, at a distance of x ¼ 5 m. positions of peak velocities of experiments (e and f), this difference
The profiles of the longitudinal velocity in the vertical plane of is important because the experimental profiles do not allow peaks
symmetry are plotted in Fig. 9. but constant values on a vertical line. Strand (1993) then defined
In Table 2 is presented for each liquid height hL, the velocity the peak as being at the middle of the vertical line. The velocity
UGCFD and Us, with which the simulations were carried out. The peaks obtained by simulations are very acceptable, the differences
dimensionless position of the experimental velocity peak is vary between 0.28% and 2.2% for all simulations. The differences in
y peak exp , the dimensionless position of the velocity peak obtained the positions of the peaks obtained by Meknassi et al. (2000) are
K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169 5165

Fig. 9. Vertical profiles of longitudinal velocities for different heights of liquid and different superficial gas velocities. (a) hL ¼ 42.9 mm and UsG ¼ 1.2 m/s; (b) hL ¼41 mm and
UsG ¼1.7 m/s; (c) hL ¼ 38.8 mm and UsG ¼ 3.1 m/s; (d) hL ¼34.5 mm and UsG ¼ 4.3 m/s; (e) hL ¼ 32.5 mm and UsG ¼ 4.9 m/s; (f) hL ¼30 mm and UsG ¼ 5.5 m/s; (g) hL ¼ 26 mm and
UsG ¼6.6 m/s.
5166 K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169

greater than ours. The positions of the peaks obtained by them are differences to the experimental values of the velocity is 7.3% for
halfway through the height of the gas. In addition to these the first, and 2.5% for the second. The velocities used in
differences, we note that our simulations yield profiles covering simulations (d, f, and g) show a relative difference from 0.8% to
the entire height from the top wall to the interface, which is not 3.5%, but are identical to those of Meknassi et al. (2000). The
the case of the profiles given by Meknassi et al. (2000). relative differences between the velocities of Meknassi et al.
(2000), and the experimental ones, on experiments (a, b, c, and e),
vary between 0.8% and 9.4%. It seems clear that the velocities we
5.2. Horizontal profile of longitudinal velocity
used, are closer to the experimental values for all experiments but
(b), than those of Meknassi et al. (2000) are.
We are situated in the cross section, at a distance of x ¼ 5 m
from the duct entrance. The streamwise velocity profiles, in the
horizontal plane at y ¼ 65 mm from the bottom of the duct, which
is of diameter d (Fig. 8), are plotted in Fig. 10. 6. Study of the correlations
In Fig. 10, six horizontal profiles of streamwise velocity are
plotted. According to the authors, the experimental profile of In Table 3 is presented, for each height of the liquid, the
experiment (f) in Strand (1993), is erroneous. Indeed, the peak superficial gas velocity UsG, and the gas velocities used for the
velocity in the horizontal plane at y ¼ 65 mm, does not correspond experimental work, for the numerical one and for the present
to the value displayed on the vertical profile. The profiles simulations are denoted, respectively, by UG(exp), UG(num) and
computed, thanks to the simulations, are symmetric with respects UG(CFD). The liquid velocities are defined in a similar manner and
to the z ¼ 0.5 line, and velocity is at its maximum halfway through are denoted, respectively, by UL(exp), UL(num) and UL(CFD). The
the duct width. The computed profiles are correct, and agree with experimental gas–wall shear stress is given by twGðexpÞ and the one
Strand’s experimental values, near the centre and near the duct obtained by simulation twGðCFDÞ . The experimental interface shear
walls. Furthermore, these calculations are carried out throughout stress is denoted tsðexpÞ and the interface shear stress obtained by
the entire duct width, unlike in the works of Meknassi et al. simulations is denoted tsðCFDÞ . Knowing all these quantities, one
(2000), which are not extended to the walls. We note that the
simulation of experiment (g) was not presented by these authors.

5.3. Analysis of the gas velocities

Fig. 11 plots the different gas velocities UG imposed at the duct

entrance in our simulations, in those by Meknassi et al. (2000),
and the experimental values of Strand (1993). Note that our
values are in good agreement with the experimental ones, in the
seven experiments. For experiments (c, d, e, and g), the relative
difference with the experimental values vary between 0.8% and
4%. For experiments (a, b, and f), these relative differences vary
between 10% and 14%. The velocities used by Meknassi et al.
(2000), are close to the experimental values, on the entire set of
experiments, with a relative difference with regards to the
experimental data, which varies between 0.4% and 6%.
In Fig. 12, we have plotted the velocities of the liquid at the
interface, which we used in our simulations, along with those of
Meknassi et al. (2000), and Strand’s experimental values. The
velocities used in simulations (a and e) are identical to the
experimental values. For simulations (b and c), the relative Fig. 11. Gas velocity used in the three approaches.

Fig. 10. Horizontal profiles of longitudinal velocities for different heights of liquid and different superficial gas velocities.
K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169 5167

can deduce the expressions of the friction factors near the wall An expression for the calculation of the interface shear stress
and at the interface. can be derived likewise. It depends on the velocity of the gas, and
is obtained from the numerical results:

6.1. Correlations fwG and fs

rG UG2
ts ¼ fs ð21Þ
The friction factors fwG and fs are determined using the values where the friction factor fs is given by the following expression:
of the shear stresses. Their expression is sought under the form:
fs ¼ 0:94Re0:427
G ð22Þ
f ¼ aRe ð18Þ We can thus derive another expression of the interface shear
stress, as a function of the relative velocities of the gas and the
The Reynolds number depends on the gas velocity, and is defined liquid:
as ReG ¼ UG dhG =nG . The friction factor is then found thanks to the rG ðUG UL Þ2
expression of the shear stress t ¼ ð1=2Þf rU 2 . The values of the ts ¼ fs ð23Þ
constants a and b can now the calculated.
where the coefficient fs is given by the following expression:
The least square approximation is used to find a function
which corresponds to the form given in (18), and which fits best fs ¼ 2:82Re0:522
G ð24Þ
the values of our simulations. The correlations are then obtained.
The gas–wall friction factor is given by
The correlations suggested here are valid in the following ranges
rG UG2 for the Reynolds number of the liquid and gas:
twG ¼ fwG ð19Þ
2 9000o ReG o49 000 and 21 000 o ReL o30 000 ð25Þ
where the friction factor fwG is written as follows: These correspond to the numerical simulations of this study.
fwG ¼ 1:14Re0:45
G ð20Þ
6.2. Comparisons of our correlations with those found in literature

This friction factor can then be used to calculate the gas–wall To compare the results given by the expressions of friction
shear stress, in any experiment. factors and shear stresses obtained by the present work and those
of Taitel and Duckler (1976a), we consider the experimental
velocities UGexp and ULexp (which is the interface velocity) given by
Strand (1993). With these velocities we can calculate the values of
the friction factors and the shear stresses obtained in the present
work (expressions (19)–(22)). The same experimental velocities
are used to calculate the friction factors and the shear stresses
given by the expressions of Taitel and Duckler (1976a). Our values
and those of Taitel and Dukler are compared with each other and
also with the experimental shear stresses given by Strand (1993).
In Table 4, are presented, the friction factors fWGTD given by
Taitel and Duckler (1976a) and fWGPW obtained by the present
work. The gas–wall shear stresses tWGTD , tWGPW and tWGexp are,
respectively, those of Taitel and Duckler (1976a), those obtained
by the present work and those obtained experimentally. The same
indices (TD, PW and exp) are used for the interfacial friction
factors and shear stresses.
In Fig. 13, we compare the values given by our correlation, for
the calculation of the gas-wall shear stress, with the one of Taitel
and Duckler (1976a). The experimental results of Strand (1993)
are also plotted. The Taitel and Dukler correlation is the one which
fits best the experimental data, as stated in the Introduction.
Our correlation is precise for the first experiment only, carried
Fig. 12. Liquid velocity used in the three approaches. out with an important liquid height. As the liquid height

Table 3
Experimental results (exp) Strand (1993), numerical (num) Meknassi et al., and present work (CFD).

Exp. hL UsG UG (exp) UG (num) UG (CFD) UL (exp) UL (num) UL (CFD) twG (exp) twG (CFD) ts (exp) ts (CFD)
(mm) (m/s) (m/s) (m/s) (m/s) (m/s) (m/s) (m/s) (Pa) (Pa) (Pa) (Pa)

a 42.9 1.2 2.03 1.91 2.30 0.243 0.22 0.243 0.020 0.056 0.041 0.061
b 41 1.7 2.76 2.68 3.15 0.259 0.247 0.24 0.035 0.089 0.083 0.095
c 38.8 3.1 4.83 4.85 5.02 0.278 0.287 0.285 0.098 0.191 0.182 0.191
d 34.5 4.3 6.19 6.06 6.09 0.327 0.331 0.331 0.141 0.250 0.217 0.273
e 32.5 4.9 6.82 7.13 6.95 0.354 0.357 0.354 0.167 0.290 0.282 0.325
f 30 5.5 7.35 7.80 8.10 0.396 0.410 0.410 0.189 0.380 0.470 0.425
g 26 6.6 8.31 8.75 8.38 0.484 0.488 0.488 0.232 0.434 0.552 0.457
5168 K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169

Table 4
Friction factors and shear stresses calculated with the experimental velocities.

Exp. hL (mm) UGexp ULexp fWGTD fWGPW tWGTD (Pa) tWGPW (Pa) tWGexp (Pa) fsTD fsPW tsTD (Pa) tsPW (Pa) tsexp (Pa)
(m/s) (m/s)

a 42.9 2.03 0.243 0.00410 0.0182 0.019 0.047 0.020 0.0041 0.016 0.015 0.040 0.041
b 41.0 2.76 0.259 0.00386 0.0157 0.033 0.074 0.035 0.00386 0.0133 0.027 0.064 0.083
c 38.8 4.83 0.278 0.00348 0.0120 0.088 0.175 0.098 0.00348 0.0102 0.079 0.149 0.182
d 34.5 6.19 0.327 0.00337 0.0105 0.137 0.252 0.141 0.00337 0.0089 0.123 0.214 0.217
e 32.5 6.82 0.354 0.00334 0.0099 0.162 0.289 0.167 0.00334 0.0085 0.146 0.246 0.282
f 30.0 7.35 0.396 0.00333 0.0095 0.185 0.321 0.189 0.00333 0.0081 0.165 0.273 0.470
g 26.0 8.31 0.484 0.00333 0.0088 0.229 0.381 0.232 0.00333 0.0075 0.203 0.324 0.552

experimental data. The correlation, we have suggested, gives

results which are close to the experimental values. Taitel and
Dukler’s correlation agrees less with the experimental values, as
the height decreases. This is not the case with the correlation we
suggested. Its results are in an excellent agreement, with the first
five experiments (very high liquid heights), and these results are
better than Taitel and Dukler (1976b) for the last two experiments
(small liquid heights).
Taitel and Dukler use a fluid–wall friction factor expression as
the Blasius one. This friction factor is determined for a one-phase
flow in a pipe. They use the same friction factor for the non-
moving wall and for the interface. The shear stresses calculated
with these friction factors give good results at the non-moving
wall but not good ones at the interface. The gas–wall shear stress
expression proposed by the present work is not the best, it was
calculated iteratively within the program according to the friction
velocity t ¼ rU 2 . This shear stress is overestimated at the non-
moving wall but well estimated at the interface. The k2e
turbulence model and the mesh used affect the friction velocity
at the non-moving wall more than at the interface (moving wall),
Fig. 13. Comparison of different values of the wall-gas shear stress in the 7
experiments. this is well explained in Djenidi et al. (2002), Orlandi and Borello
(2009), and Jiang and Lai (2009).
The interface friction factor expression developed in this work
and the gas–wall friction factor given by Taitel and Dukler can be
used in the expressions giving the shear stresses at the interface
and at the gas–wall (Eq. (3)), to study a two-phase flow. The two
expressions are complementary.

7. Concluding remarks

The geometry of a gas–liquid stratified flow has been

transposed then studied using the CFD code Fluent. The velocity
profiles of the gas, which have been computed using this analogy,
are very close to the experimental profiles of Strand. Some profiles
are in a better agreement, than those of Meknassi et al. are. This
confirms the validity of this method. The vertical profiles are
complete and cover the entire width of the duct, follow the
downwards shift of the profile, and provide peak velocities which
are similar to those given by the experiments. Thanks to this
successful approach, the expressions of two friction factors have
been derived: one for the gas–wall, and another for the interface.
We can confirm that Taitler and Dukler’s correlation, which is
used to derive the gas–wall friction factor, remains the most valid.
Fig. 14. Comparison of different values of the interface shear stress in the 7
experiments. However, the correlation suggested in the present work, to
calculate the interface friction factor, is the best.

decreases, our correlation agrees less and less with the

experimental data. References
In Fig. 14, we compare the values given by our correlation for
the calculation of the interface shear stress, with the values given Agrawal, S.S., Gregory, G.A., Govier, G.W., 1973. An analysis of horizontal stratified
by the correlation of Taitel and Duckler (1976a), and Strand’s two phase flow in pipes. Can. J. Chem. Eng. 51, 280–286.
K. Sidi-Ali, R. Gatignol / Chemical Engineering Science 65 (2010) 5160–5169 5169

Blasius, H., 1912. Das aehnlichkeitsgesetz bei reibungsvorgangen. Z. Ver. Dtsch. Newton, C.H., Behnia, M., Reizes, J.A., 1999. The effect of liquid viscosity on gas
Ing. 56 (16), 639–643. wall and interfacial shear stress in horizontal two-phase pipe flow. Chem. Eng.
Cheremisinoff, N.P., James Davis, E., 1979. Stratified turbulent–turbulent gas– Sci. 54, 1071–1079.
liquid flow. A.I.Ch.E. J. 25, 48–56. Nigmatulin, N.I., 1991. Dynamics of Multiphase Media. Hemisphere Publishing
Chung, T.J., 2002. Computational Fluid Dynamics. Cambridge University Press. Corporation, New York.
Crowley, C.J., Wallis, G.B., Barry, J.J., 1992. Validation of a one dimensional wave Orlandi, P., Borello, D., 2009. DNS and zf closure in turbulent channels with a
model for the stratified-to-slug flow regime transition, with consequences for moving wall. Technical report WT-090414-URS-2.
wave growth and slug frequency. Int. J. Multiphase Flow 18, 249–271. Patankar, S.V., 1980. Numerical Heat Transfer and Fluid Flow. Hemisphere
Delhaye, J.M., Giot, M., Riethmuller, M.L., 1981. Thermohydraulics of Two Phase Publishing Corporation.
Systems for Industrial Design and Nuclear Engineering. Mc-Graw Hill. Petalas, N., Aziz, Kh., 1998. A mechanistic model for multiphase flow in pipes. In:
Djenidi, L., Savill, A., Antonia, R.A., 2002. Calculation of a low-shear turbulent 49th Annual Technical Meeting of the Petroleum Society of the Canadian
boundary layer using a second-moment order closure. Engineering Turbulence Institute of Mining, Metallurgy and Petroleum, Paper 98-39, Alberta, Canada,
Modelling and Experiments 5. Mallorca, Spain. June 8–10.
Fluent users guide, 2006. Version 6.4, Fluent Inc., Lebanon, NH. Prandtl, L., 1927. Über den reibungswiderstand strömer luft. Ergebnisse, AVA,
Hirch, C., 1991. Numerical Computation of Internal and External Flows, Vol. 1, Göttingen, 3. Liefg, pp. 1–5, 4. Liefg, pp. 18–29.
Fundamentals of Numerical Discretisation. John Wiley and sons. Prandtl, L., 1932. Zur turbulenten strömung in rohren, und längs platen. AVA,
Ishii, M., 1975. Thermo-fluid Dynamic Theory of Two-phase Flow. Eyrolles, Paris. Göttingen, vierte serie.
Ishii, M., Hibiki, T., 2006. Thermo-fluid Dynamics of Two-phase Flow. Springer, USA. Shoham, O., Taitel, Y., 1984. Stratified turbulent–turbulent gas–liquid flow in
Jiang, X., Lai, C.H., 2009. Numerical Techniques for Direct and Large-eddy horizontal and inclined pipes. A.I.Ch.E. J. 30, 377–385.
Simulations. CRC Press, Taylor and Francis New edition. Spedding, P.L., Hand, N.P., 1997. Prediction in stratified gas–liquid co-current flow
Kowalski, J.E., 1987. Wall and interfacial shear stress in stratified flow in a in horizontal pipelines. In. J. Heat Mass Transf. 40, 1923–1935.
horizontal pipe. A.I.Ch.E. J. 33, 274–281. Strand, Ø., 1993. An experimental investigation of stratified two-phase flow in
Launder, B.E., Spalding, D.B., 1972. Lectures in Mathematical Models of Turbulence. horizontal pipes. Ph.D. Thesis, University of Oslo, Norway.
Academic Press, London, England. Taitel, Y., Duckler, A.E., 1976a. A model for predicting flow regime transitions in
Leveque, R.J., 2002. Finite Volume Method for Hyperbolic Problems. Cambridge horizontal and near horizontal gas–liquid flow. A.I.Ch.E. J. 22, 47–55.
University Press. Taitel, Y., Dukler, A.E., 1976b. A theoretical approach to the Lockhart–Martinelli
Meknassi, F., Benkirane, R., Line, A., Masbernat, L., 2000. Numerical modeling of correlation for stratified flow. Int. J. Multiphase Flow 2, 591–595.
wavy stratified two-phase flow in pipes. Chem. Eng. Sci. 55, 4681–4697. Wilcox, D.C., 1993. Turbulence Modeling for CFD. DCW Inductries Inc.