Sie sind auf Seite 1von 11

Chemical Engineering Science 56 (2001) 68516861

www.elsevier.com/locate/ces
A numerical model of stratied wavy gasliquid pipe ow
C. H. Newton, M. Behnia

School of Mechanical and Manufacturing Engineering, The University of New South Wales, Sydney, NSW 2052, Australia
Abstract
A numerical model of stratied wavy gasliquid ow in a horizontal tube has been developed and tested with experimental data.
The model solves the steady axial momentum equation together with a low Reynolds number kc model of turbulence for the eddy
viscosity. The turbulence model is modied to accommodate the irregular geometry of stratied pipe ow, and the computations
are performed in the bipolar coordinate system for convenient mapping of the physical domain. A set of boundary conditions
for the wavy gasliquid interface was devised from empirical considerations. Calculations of pressure drop and liquid holdup are
observed to compare favourably with experimental measurements, and predicted ow quantities such as velocity and wall shear
stress are also satisfactory, provided that the ow conditions at the gasliquid interface remain steady or quasi-steady. ? 2001
Published by Elsevier Science Ltd.
Keywords: Modelling; Multiphase ow; Numerical analysis; Pipe; Stratied; Turbulence
1. Introduction
The simultaneous ow of liquid and gas in a pipe is
commonly found in the petroleum and chemical process-
ing industries, in steam generation equipment, and in nu-
clear reactors. Flows of this nature are governed by a
wide variety of parameters including the phase ow rates,
thermo-physical properties of the uids, and pipeline ge-
ometry. The pipes may be horizontal, vertical or inclined.
Further, the in situ conguration of the uids or ow
pattern is a complicated function of these variables, and
is strongly dependent on the pipe inclination angle. Con-
sequently, these ows are often unsteady, or even random
or chaotic.
Despite several decades of research into gasliquid
pipe ow phenomena, no reliable analytical methods
for the calculation of the primary pipeline design vari-
ables such as pressure drop and liquid holdup (or void
fraction) presently exist. Historically, approaches to
modelling such ows have been in the form of either
generalised empirical data correlations or simple mech-
anistic models of individual ow patterns. In the case
of the former, the majority of these empirical equations

Corresponding author. Tel.: +61-2-9385-4253; fax: +61-2-


9663-1222.
E-mail address: m.behnia@unsw.edu.au (M. Behnia).
originated from the oil and gas industries (e.g. Baxendall
& Thomas, 1961; Beggs & Brill 1973). Correlations have
also emerged from the power generation and chemical in-
dustries (e.g. Chisholm, 1973; Olujic, 1985). However, it
is widely accepted that empirical correlations for pressure
drop and liquid holdup perform poorly when applied to
data outside the range on which they are developed (e.g.
Battara, Mariani, Gentilini, & Giachetta, 1985; Gregory
& Fogarasi, 1985).
More recently attention has turned to so-called mech-
anistic models to formulate semi-analytical calculation
methods for ow patterns with specic geometries. In
horizontal gasliquid pipe ow where gravity separation
is most likely to occur, the pioneering studies have been
those of Taitel and Dukler (1976) for the relatively sim-
ple stratied ow conguration, and Dukler and Hubbard
(1975) for the more complicated slug ow geometry.
However, the greater majority of mechanistic models rely
on empirical data for closure, with correlations for wall
and interfacial shear stresses the most commonly used.
Unfortunately, with so little known about the distribution
of wall shear in gasliquid ows, for expediency recourse
is often made to relationships established in single-phase
pipe ows, with a resulting loss in calculation accuracy.
More recent attempts at mechanistic modelling of strat-
ied ow have been focussed at developing more ac-
curate relationships for wall and interfacial shear stress
0009-2509/01/$ - see front matter ? 2001 Published by Elsevier Science Ltd.
PII: S 0009- 2509( 01)00322- 0
6852 C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861
(e.g. Andritsos & Hanratty, 1987; Spedding & Hand,
1997). However, the basic problemstill remains mech-
anistic models use virtually the same amount of empirical
information to obtain the same outcome as purely em-
pirical data correlations, even for the simplest ow ge-
ometries. For more complex geometries such a slug ow,
additional inputs such as translational velocity and fre-
quency are also required.
For stratied gasliquid ows some attempts at using
computational uid dynamics (CFD) to predict pressure
drop and holdup have been made. A variety of solutions
for ows in horizontal rectangular ducts have been pre-
sented, including those of Akai, Inoue, and Aoki (1981),
Issa (1988) and Srichai, Jayanthi, and Hewitt (1995).
In pipe ow, where the geometry is more complicated,
Shoham and Taitel (1984) solved a two-dimensional ax-
ial momentum equation in the liquid region in the bipolar
coordinate system and a zero-equation turbulence model.
The gas region was treated as a bulk ow, and the phases
coupled with an explicit denition of interfacial shear
stress, using an empirical friction factor to model the ad-
ditional interfacial shear generated by a wavy interface.
Solutions were obtained for owin horizontal and slightly
upwardly inclined pipes with a diameter of 25.4 mm.
However, several rst order viscosity gradient terms were
missing from the momentum equation published in their
paper, casting doubts over the validity of their results.
Issa (1988) also modelled stratied ow using the
bipolar coordinate system and solved the axial momen-
tum equation, coupled with a two-equation model of tur-
bulence, in both the liquid and gas phases. The results
showed good agreement with predictions from the mech-
anistic model of Taitel and Dukler (1976), but were con-
ned to horizontal stratied smooth ow in a relatively
small 25 mm diameter pipe. Further, wall functions were
used at all boundaries, requiring additional empirical in-
put, and precluding any inferences to be drawn about the
distribution of wall shear stress in each phase. The cases
presented by Shoham and Taitel (1984) were recalcu-
lated, and major dierences in the computed ow elds
were observed. Issa (1988) attributed these discrepancies
to the more rigorous solution of the ow in each phase.
Recently, Newton and Behnia (2000) extended the
work of Issa (1988) to the prediction of stratied smooth
gasliquid ow in a larger, 50 mm, diameter pipe. To
limit the empirical input into the solution, a low Reynolds
number kc model of turbulence was used in each phase.
This model allowed solution of the governing equations
without the use of empirical wall functions. Hence, dis-
tributions of the calculated wall and interfacial shear
stresses could be presented. The resulting pressure drop
and holdup calculations were favourable when compared
with the mechanistic model of Taitel and Dukler (1976)
and experimental data.
Stratied smooth conditions are dicult to achieve in
practice, especially with liquids having relatively low vis-
cosity such as water. A model in which wavy stratied
ows can be predicted will have far more practical value
in the prediction of pressure drop and liquid holdup un-
der design conditions. Further, such a model will allow
insight into the relative distributions of wall and interfa-
cial shear stresses as a function of the condition of the
gasliquid interface. The construction of such a model is
the focus of the present article.
We aim to extend our previous model of smooth strat-
ied ow (Newton & Behnia, 2000) to allow predictions
for ows where the gasliquid interface contains rough-
ness caused by the presence of surface disturbances or
waves. We retain the same underlying assumptions as
those in the commonly used stratied momentumbalance,
i.e. uni-directional ow, at interface, constant uid prop-
erties, and solve momentum and turbulence equations in
both phases using special boundary conditions at the in-
terface. The only directly empirical information required
by the model is an average interfacial friction factor to
account for the interfacial roughness caused by the pres-
ence of the waves. The model is then used to evaluate
the distribution of wall shear stresses from the computed
ow elds, and the resulting axial pressure gradient and
holdup. The results are compared with predictions from
a variety of mechanistic models for stratied gasliquid
ows that have appeared in the literature over the past
three decades.
2. Numerical modelling of stratied gas--liquid pipe
ow
2.1. Governing equations and coordinate system
We consider turbulent, unidirectional ow, in an ar-
bitrarily shaped domain mapped by an orthogonal r 0
coordinate system. The ow can be described in compact
form by the following general equation:
1.
_
P
_
c
2
q
cr
2
+
c
2
q
c0
2
_
+Q
_
cj
t
cr
cq
cr
+
cj
t
c0
cq
c0
_
+R(E
2
+F
2
)
_
S =0, (1)
where q is a vector describing axial velocity w, turbulent
kinetic energy k and dissipation c respectively, and j
t
is the turbulent viscosity. The coecients dening the
three respective transport equations are given in Table 1.
The parameter 1 describes the transformation from the
Cartesian x, coordinate system to the generalized r 0
system, and is given by
1 =
1
(cx}cr)
2
+ (cx}c0)
2
. (2)
C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861 6853
Table 1
Coecients of the generalized transport equation given by Eq. (1).
For the present calculations o
k
=1.0, o
c
=1.3, c
1
=1.92 and c
2
=1.3
q P Q R E F S
w j + j
t
1 0 0 0
dP
d:
k
1
j
_
j +
j
t
o
k
_
1
o
k
j
t
cw
cr
cw
c0
jc
c
1
j
_
j +
j
t
o
c
_
1
o
c
c
1
[
1
j
t
c
k
cw
cr
cw
c0
c
2
[
2
j
c
2
k
Fig. 1. Schematic representation of stratied gasliquid ow.
The non-circular liquid and gas domains in stratied
pipe ow, as shown schematically in Fig. 1, are conve-
niently modelled with the bipolar coordinate system (eg.
see Shoham & Taitel, 1984), described analytically by:
x =
c.sinh(r)
cosh(r) cos(0)
, , =
c.sin(0)
cosh(r) cos(0)
(3)
and illustrated graphically in Fig. 2. The mapping given
by Eq. (3) is orthogonal, yielding an innitely long, rect-
angular computational domain with the bounds:
0 , r for the gas,
0 +, r for the liquid, (4)
where is equal to half the angle subtended by the centre
of the pipe and the gasliquid interface, and is given by:
=cos
1
_
1
2h
L
D
_
. (5)
Dierentiation of Eq. (3) gives the transformation pa-
rameter for the bipolar scheme:
1 =
[cosh(r) cos(0)]
2
c
2
. (6)
Evidently, it is necessary to place a nite limit on the
maximum value of r in Eq. (4) for the purposes of the
numerical computations. In all the results presented here
this maximum was set to r =6, which corresponds to a
Fig. 2. The bipolar coordinate system.
circular boundary (AB in Fig. 2) with a radius of less
than 0.5% of c.
2.2. Turbulence modelling
The widely used kc model of turbulence is used to
provide a closure relationship for the turbulent viscosity,
which is coupled to the solution of the transport equations
for turbulent kinetic energy and dissipation, and is dened
by
j
t
=
[
j
C
j
jk
2
c
. (7)
For the standard kc formulation in which wall
functions are used at the boundaries, [
j
=[
1
=[
2
=1.
However, to model the ow right up to the pipe walls
without standard wall functions we use the Low
Reynolds Number turbulence model, in which these
damping factors are designed to inuence the dissipation
equation in the wall region, where the viscous forces are
dominant. For the present study the choice of the wall
damping functions [
j
, [
1
and [
2
is due to Patel, Rodi,
and Scheurer (1985) who conducted an extensive review
of near-wall turbulence models. Their results suggested
that the functions of Lam and Bremhorst (1981) were
amongst the best at predicting the general characteristics
of the laminar sub-layer in turbulent pipe ow. Dening p
as the normal distance from the wall, being the smaller of
6854 C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861
(refer to Fig. 2 for the coordinate system):
p =
D
2

x
2
+
_
,
D
2
+h
L
_
2
, p =,, (8)
the damping functions of Lam and Bremhorst (1981) are
given by
[
j
=[1 exp(0.0165j

kp}j)]
2
(1 +[jc}jk
2
),
[
1
=1 + (0.05}[
j
)
3
,
[
2
=1 exp[ {(jk
2
}jc)
2
}]. (9)
These functions have been adopted here, subject to
some minor modication to account for the non-circular
ow geometry in each phase. This involves a correlation
for [, which is given as [ =20.5 for turbulent pipe ow
by Lam and Bremhorst (1981). The origins of this cor-
relation, given as
[ =20.5[1 1.45 exp(1.4 10
4
Re
G
)] (10)
are described in Newton and Behnia (2000), and are not
repeated here.
2.3. Boundary conditions
The wall boundary conditions in both the liquid and
gas regions are given by
w=k =j
t
=0, (11)
cc
w
}cp =0. (12)
To model the wavy interface, more complex boundary
conditions are required. This task is complicated by the
unsteadiness of the wavy motion itself, which may be pe-
riodic or totally chaotic, depending on the relative ow
rates of the liquid and gas. However, the steady, unidirec-
tional nature of the model developed thus far precludes
the incorporation of unsteadiness or developing ow, and
hence the shear stresses generated by the wavy interface
must be accounted for using approximate methods. Given
that the solution of the ow eld will be for average ve-
locity and turbulence quantities, it seems reasonable to
assume that for modelling purposes the concept of an av-
erage interfacial shear stress can also be used.
The most common method of introducing wall rough-
ness into a numerical model is to use empirical wall func-
tions, such as those employed by Srichai et al. (1995)
in a study of stratied gasliquid ow with roll waves.
However, such functions require that a constant friction
velocity and various aspects of the wave structure be
known a priori. Consequently these functions are of no
help here, since we can reasonably assume that the in-
terfacial shear stress is not constant across the gasliquid
interface. Other investigators have attempted to treat the
presence of surface waves as a source of turbulent kinetic
energy, but have found a basic incompatibility between
the turbulence levels in both phases. Akai et al. (1981)
have concluded that since the production of large-scale
eddies via the separation of the gas ow from the liq-
uid, continuity of turbulence energy at the interface is not
necessary. However, they were not able to nd an ap-
propriate relationship for the relative magnitudes of these
energies, and were thus forced to revert to wall functions
to dene the interface. Issa (1988) also used wall func-
tions in numerical calculations of wavy gasliquid ow
in a rectangular duct.
In the present model the concept of continuity of shear
stress at the interface is used, and the shear stress itself
is imposed on the solution via empirical means. Newton
and Behnia (2000) have found for smooth interfaces that
the distribution of shear stress across the interface is not
uniform, but rather a function of the distance from the
wall. Presumably a similar situation exists for wavy in-
terfaces, and hence this distribution is approximated by
the following power law relation:
t
i
(x) =
_
1 +
1
n
_
t
i
_
1
x
c
_
1}n
, (13)
where n =6.6, and the average interfacial shear stress t
i
is given by
t
i
=[
i
j
G
u
2
G
2
, (14)
where [
i
is the empirical friction factor. Aside from the
empirical wall damping functions and constants in the
turbulence model, the empirical friction factor is the only
additional empirical information required for the closure
of the wavy ow model. In terms of the numerical cal-
culations it is entered at the start of the simulation and
remains constant throughout.
All that remains is to provide interfacial boundary con-
ditions for the turbulence quantities. For wind-induced
shear stress at a free surface, Rodi (1984) has proposed
the following denition of interfacial turbulent kinetic
energy:
k
i
(x) =
t
i
(x)
j
_
C
j
. (15)
Evidently, when Eq. (13) is substituted in Eq. (15) the
interfacial kinetic energy is maximum at the centre of the
pipe, and will reduce as the damping inuence of the wall
is increased (i.e. as x c).
Since it is assumed that the shear stresses at the inter-
face are equal in both phases, then the boundary condi-
tions for k in each phase can be written as:
Liquid : k
iL
(x) =
t
i
(x)
j
L
_
C
j
, Gas : k
iG
(x) =
t
i
(x)
j
G
_
C
j
.
(16)
C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861 6855
Fig. 3. Distribution of grid points in the physical domain. Adapted
mesh for )
L
=0.03 ms
1
, )
G
=5.5 ms
1
, and h
L
=8.6 mm.
Thus, the level of turbulent kinetic energy at the interface
will dier by the ratio of the uid densities, as proposed
by Issa (1988). The dissipation at the interface is assumed
to be given by Eq. (12).
The treatment of the interface as a source of turbulence
generation would seem to preclude the use of the turbu-
lence damping functions given by Eqs. (9)(10) near the
gasliquid boundary. In the gas region the following are
applied when the normal distance to the interface is less
than the normal distance to the wall:
[
j
=[
1
=[
2
=1. (17)
In the liquid region, however, it may be expected that the
wall will have some eect on the turbulence structures,
since the liquid height will be reduced under the action
of the increased pressure gradient. Consequently, for the
damping functions given by Eq. (8), the distance to the
solid wall is taken as the characteristic distance in this
region, irrespective of the location in the ow region.
2.4. Grid generation
To adequately resolve the large gradients of k and c
near the wall and interface a nominal grid spacing of
p
+
=p

jt
w
}j 1 was used in these regions up to a
distance of approximately p
+
=40. The remainder of the
ow region was carefully meshed to ensure that the grid
growth ratio between adjacent cells did not exceed 1.5,
except in the centre of each ow region where all gradi-
ents are quite small. A typical grid is shown in Fig. 3.
The construction of such a detailed grid made the use
of second order nite dierencing schemes virtually re-
dundant for the ows considered here. In solving the gov-
erning ow equations both rst and second schemes were
employed, but the results were suciently similar to al-
low perseverance with the rst order scheme, which was
also benecial in reducing solution times.
2.5. Numerical solution procedure
Solution of Eq. (1) in both phases follows a procedure
similar to that described in Newton and Behnia (2000),
but adapted slightly to handle the wavy gasliquid inter-
face. The transport equations are solved consecutively in
each phase using an iterative nite dierence technique.
The input variables are pipe diameter, supercial liquid
and gas velocities, )
L
and )
G
, and average interfacial fric-
tion factor. Since for wavy ow the turbulent viscosity is
not zero at the interface, continuity of shear stress gives:
t
i
=(j
t
+j)
G
dw
dn

p=0
=(j
t
+j)
L
dw
dn

p=0
. (18)
The ow eld in the liquid region solved rst using the
imposed shear stress from Eq. (13). The calculations pro-
ceed until the liquid ow continuity requirement is met.
The solution in the liquid region then provides the interfa-
cial velocity boundary condition via the use of Eq. (18),
for the solution of the gas region ow eld. Thus the so-
lution of the gas region does not directly inuence the
liquid region, which has completely independent bound-
ary conditions of its own. Once continuity of the gas ow
is achieved the calculated average pressure gradients in
each phase are compared and the liquid height adjusted.
The procedure is repeated until the pressure gradients in
each phase are matched.
The calculations were performed on a 200 MHz PII PC.
Even with the rst order nite dierencing scheme the
computations were quite slow because of the ne mesh.
In the worst case a single liquid height iteration could
take up to 7000 CPU seconds to complete. Generally 34
height iterations were required to obtain a solution. Future
solution schemes involving the simultaneous solution of
the liquid and gas regions with a VOF model would most
likely lead to a considerable reduction in solution times.
3. Experimental data
For the purpose of validating the numerical calcula-
tions, a series of experiments involving stratied air
water ow in a 50 mm diameter, 16 m long horizontal
pipe were conducted. Aside from basic measurements of
axial pressure gradient and mean liquid holdup, the gas
wall shear stress distribution was measured by travers-
ing a roving Preston tube around the circumference of
the pipe wetted by the gas ow. Further, the centre-line
velocity prole in the gas region was measured with a
traversing Pitot tube. The experimental data encompassed
6856 C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861
smooth, rippled and completely wavy interfacial condi-
tions. Details of the experimental setup and results are
available in Newton and Behnia (1996), and are not re-
peated.
4. Numerical results
4.1. Basis of the calculations
Stratied wavy gasliquid pipe ow is characterised
by a variety of interfacial conditions depending on the
relative velocities of the liquid and gas phases, the liquid
holdup, and the uid properties. For the purposes of this
analysis the interfacial wave structures can be broadly
classied into rippled, two-dimensional (2D) wave and
breaking roll wave ow. Rippled ow is a low ampli-
tude, high frequency disturbance normally observed at
relatively low liquid holdups and higher gas rates. At ex-
tremely low holdups (approx. H
L
0.05) the interface
takes on substantial curvature, rendering the present geo-
metrical analysis inadequate. Hence, in the present study
all measured data presented here are taken at H
L
0.05.
The 2D interfacial disturbance is characterized by large
amplitude, lowfrequency ( 10 Hz) sharp crested waves
with relatively long troughs, that form across the width
of the interface and propagate down the pipe with a rel-
atively steady velocity. This ow occurs at lower gas
velocities and higher liquid holdups. Under the correct
conditions, instability at the interface can cause the sur-
face waves to be suciently large such that they bridge
the pipe and initiate slug ow.
The nal class of disturbance is the breaking roll wave,
which occurs at very high gas velocities and low liquid
holdups. This regime is highly unsteady, and precedes the
intermittent slug}spray ow if the gas velocity is su-
ciently high. Generally speaking, the present calculations
were more successful in cases where the observed ow
was characterized by steady or quasi-steady interfacial
conditions, as will be shown in the following sections.
4.2. Selection of the interfacial friction factor
For closure of the numerical model the average inter-
facial friction factor used in the numerical calculations
was obtained from the experimental data by calculating a
one-dimensional momentum balance in each phase. An-
dritsos and Hanratty (1987) have used this technique in
the analysis of their own experimental data, except that
they were compelled to use an empirical correlation for
the gas wall shear, whereas we were able to use directly
measured values. The eect of our somewhat arbitrary
choice of [
i
can be seen in Table 2, which shows the
sensitivity of the numerical calculations to variations in
this parameter.
Table 2
Eect of interfacial friction factor on numerical predictions for
)
L
=0.03 ms
1
, )
G
=3.64 ms
1
. The measured values of pressure
gradient and liquid height were 7.9 Pa m
1
and 0.0136 m, respec-
tively
Average [
i
Calculated d}d: Calculated h
L
(Pa m
1
) (m)
0.011 9.92 0.0137
0.009 9.64 0.0141
0.007 9.20 0.0143
0.005 9.14 0.0152
For the case under consideration the measured pressure
gradient and liquid height were 7.9 Pa m
1
and 13.6 mm,
respectively. The value of [
i
obtained from the measured
data was 0.01. Evidently the choice of interfacial friction
factor has an eect on the calculated results, but this ef-
fect is not as dramatic as one might have expected. In
the present example, a 50% increase in [
i
results in an
increase of 7% in the calculated pressure gradient and a
decrease of 9% in the calculated liquid height, suggest-
ing that the model is relatively insensitive to choices of
interfacial friction factor. This insensitivity was observed
over the range of data presented here.
4.3. Calculation of pressure gradient and liquid holdup
Predictions of axial pressure gradient and liquid holdup
fromthe present model are compared with the experimen-
tal data in Fig. 4a. Generally speaking, the agreement was
very good, given the uncertainty in the experimental mea-
surements. The model generally gave over-predictions of
the pressure gradient data, with an absolute average error
of 20%. Although the agreement in the pressure gradient
calculations can be partially attributed to the optimal
choice of friction factor, the success in the calculation
of mean liquid holdup is notable, the model generally
slightly under-predicting the measured data with an ab-
solute average error of 19%. An additional feature of the
results was the increase in predictive accuracy at higher
gas rates, i.e. at higher pressure drops and lower liquid
holdups, conditions which are more consistent with in-
dustrial ows.
The present numerical calculations can also be com-
pared with results from a variety of mechanistic models
of stratied ow in Fig. 4. The three models selected are
illustrative of the development of the mechanistic mod-
elling technique over the past three decades. In the orig-
inal model of Taitel and Dukler (1976) it was assumed
that [
i
=[
G
, and that the liquid and gas wall friction
factors could be well approximated by the Blasius equa-
tion for single-phase pipe ow. Since then various in-
vestigators, including Andritsos and Hanratty (1987) and
Spedding and Hand (1997), have progressively modied
the so-called stratied ow momentum balance to more
C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861 6857
accurately model the wall and interfacial shear stresses,
with correlations originating directly fromtwo-phase pipe
ow experiments quite similar to those reported here.
Generally speaking, the present numerical calculations
compare quite favourably with commonly used mecha-
nistic techniques, despite the fact that these models use
a considerable amount of empirical data tailored to suit
the ows modelled here. The absolute average predic-
tive errors for each calculation technique over the present
experimental data set are shown in Table 3. Not sur-
prisingly, the recent mechanistic model of Spedding and
Hand (1997) was highly successful in predicting the ex-
perimental data.
4.4. Calculated velocity eld
A series of calculated centre-line velocity proles are
compared with Pitot tube measurements in Fig. 5, where
agreement is seen to be quite reasonable. It can be ob-
served, however, that the model slightly over-predicts the
location of the maximum in the gas velocity prole. Note
Fig. 4. Comparison of prediction and experiment for calculations of axial pressure gradient and liquid holdup (a) present model, (b) Taitel &
Dukler (1976), (c) Andritsos & Hanratty (1987), (d) Spedding & Hand (1997).
also the highly turbulent liquid velocity prole. A se-
lection of calculated centre-line turbulence quantities are
shown in Fig. 6. Note the magnitude of the gas turbulent
kinetic energy at the interface, which is approximately
60% higher than the corresponding maximum value near
the wall.
The upward shift in the predicted gas velocity prole
is expected as a consequence of the increase in interfa-
cial roughness. The shift occurs purely to properly bal-
ance the competing gas wall and interfacial shear stresses
against the axial pressure gradient; the interface draws
increasingly more momentum from the gas region as the
Reynolds number is increased, with a consequent de-
crease in the friction factor at the gas wall.
Numerical experimentation indicated that the turbu-
lence boundary conditions at the interface caused this
upwards shift to be more pronounced as the gas phase
Reynolds number was increased. When compared with
the experimental data a strong correlation between this
eect and increasingly unsteady interfacial conditions
was observed. For such ows an upwards shift in the
6858 C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861
Fig. 4. (Continued)
Table 3
Comparison of average absolute errors of present calculations with
various mechanistic models for stratied gasliquid pipe ow
Model Average absolute error (%)
Pressure Liquid
gradient holdup
Present model 27.3 15.5
Taitel & Dukler (1976) 30.4 10.0
Andritsos & Hanratty (1987) 27.3 17.1
Spedding & Hand (1997) 15.0 16.6
gas velocity prole, coupled with signicant increases
in the calculated turbulence quantities, suggested that
the calculated (i.e. imposed) turbulent kinetic energy
boundary condition on the gas side of the interface was
too high. Hence, despite the rather promising pressure
drop and holdup calculations over the range of data pre-
sented here, it would appear that a numerical model of
the type described here is eective for ows with steady
or quasi-steady interfacial conditions only.
Fig. 5. Comparison of measured and calculated centre-line velocity
proles with )
L
=0.03 ms
1
.
4.5. Gas and liquid wall shear stress
Further evidence for the need to reduce the turbulence
levels in the gas region for unsteady wavy ows comes
C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861 6859
Fig. 6. Calculated centre-line turbulence quantities, )
L
=0.03 ms
1
, )
G
=5.5 ms
1
, and h
L
=8.75 mm.
Fig. 7. Wall shear stress distribution, rippled ow with a ten-
dency to slightly unsteady interfacial conditions, with )
L
=0.03 ms
1
,
)
G
=7.5 ms
1
. The mean wall shear stresses given by the Blasius
equation for these ow conditions are 0.28 and 0.54 Pa for the gas
and liquid domains, respectively.
from a consideration of the calculated gas and liquid wall
shear stresses. A comparison of the calculated wall shear
stress distribution with the experimental measurements at
a relatively high gas rate (corresponding to rippled ow
conditions tending to a transition to unsteady roll waves),
is shown in Fig. 7. The relaxation in the wall shear stress
distribution near the gasliquid interface, observed exper-
imentally by Kowalski (1987) and Newton and Behnia
(1996), is clearly evident. However, the calculated local
gas wall shear stresses exceed the measured values by up
to 30%. The liquid wall shear stresses also appear higher
than would normally be expected. In these ow condi-
tions the numerical model appears to transfer too much
energy from the interface to each of the phases, the net
result being an increase in the wall shear required to bal-
ance the eect of increased interfacial roughness.
Calculations of the wall shear stress for ows with
steady interfacial conditions were more successful, a sam-
ple being shown in Fig. 8, which represents a ow with
2D interfacial waves. Generally speaking, the predicted
Fig. 8. Wall shear stress distribution, 2D wavy ow, with
)
L
=0.07 ms
1
, )
G
=3.5 ms
1
. The mean wall shear stresses given
by the Blasius equation for these ow conditions are 0.09 and 0.22 Pa
for the gas and liquid domains, respectively.
gas wall shear stresses for ows with steady interfa-
cial conditions closely matched those given by the Bla-
sius equation for single-phase pipe ow, as observed by
Kowalski (1987). Consequently, the wall and interfacial
stresses were adequately balanced, leading to a more re-
alistic calculation of the ow eld.
4.6. Phase momentum correction factors
An additional by-product on a numerical solution of
the ow eld is the ability to estimate the momentum
correction factor in each phase, dened by
MCF=
_
A
w(x, ,)
2
dA
u
2
A
, (19)
where u is the average velocity, and A the phase
cross-sectional area. The results for the present calcula-
tions are shown in Fig. 9. Additional data for stratied
smooth ow, obtained from an earlier study (Newton &
Behnia, 2000) are also shown in the gure. In the gas
6860 C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861
Fig. 9. Calculated phase momentum correction factors (a) liquid
region (b) gas region.
region it can be seen that the data approach a constant
value of approximately 1.06 as the Reynolds number is
increased. This value is higher than those normally ob-
tained in turbulent pipe ows. More data is needed in the
liquid region to identify a denite trend, but the momen-
tum correction factors also seem higher than would be
expected in pipe ow. For the present ow conditions,
further increases in the liquid phase Reynolds number
would have resulted in a transition from the stratied
ow regime.
5. Conclusions
A numerical model of stratied wavy gasliquid pipe
ow has been developed and tested with experimental
data. Predictions of axial pressure drop and liquid holdup
showed acceptable agreement with the measured data,
and compared favourably with mechanistic models de-
veloped specically for stratied ow conditions. The
pressure drop and holdup calculations were observed to
be relatively insensitive to the choice of interfacial fric-
tion factor, but assumptions regarding the distribution of
turbulence quantities at the interface appeared to become
less valid as the interfacial ow conditions tended to un-
steady behaviour. Predictions for ows with more steady
interfacial conditions were quite satisfactory.
Although the present formulation is rather complex and
slow, due to the nature of the turbulence model, and the
ne grids required for its implementation, it does appear
to demonstrate that the CFD technique can be success-
fully applied to stratied wavy ows. If more general
boundary conditions for the turbulence quantities at the
gasliquid interface can be found then there is potential,
using increased computer power coupled with more ad-
vanced turbulence models, to obtain faster and more accu-
rate engineering solutions for a wide range of ow rates,
uid properties and pipe diameters, without the usual re-
liance on an array of empirical input.
Notation
c focal length in bipolar coordinate system, m
D pipe diameter, m
h
L
liquid height, m
H
L
liquid holdup
) supercial velocity, ms
1
k local turbulent kinetic energy, m
2
s
2
MCF momentum correction factor
u average phase velocity, ms
1
w local axial velocity, ms
1
Greek letters
c local rate of turbulence dissipation, m
2
s
3
half-angle subtended by liquid at centre of
pipe
p distance to wall, m
j viscosity, Pa s
j density, kg m
3
t shear stress, Pa
Subscripts
G gas
i interface
L liquid
t turbulent
W wall
References
Akai, M., Inoue, A., & Aoki, S. (1981). The prediction of
stratied two-phase ow with a two-equation model of turbulence.
International Journal of Multiphase Flow, 7, 2139.
Andritsos, N., & Hanratty, T. J. (1987). Inuence of interfacial waves
in stratied gasliquid ows. A.I.Ch.E. Journal, 33, 444454.
Battara, V., Mariani, O., Gentilini, M., & Giachetta, G. (1985).
Condensate line correlations for calculating holdup, friction
compared to eld data. Oil and Gas Journal, December, 148152.
Baxendall, P. B., & Thomas, R. (1961). The calculation of pressure
gradients in high-rate owing wells. Journal of Petroleum
Technology, October, 10231028.
C. H. Newton, M. Behnia / Chemical Engineering Science 56 (2001) 68516861 6861
Beggs, H. D., & Brill, J. P. (1973). A study of two-phase ow in
inclined pipes. Journal of Petroleum Technology, May (5), 607
617.
Chisholm, D. (1973). Pressure gradients due to friction during the ow
of evaporating two-phase mixtures in smooth tubes and channels.
International Journal of Heat Mass Transfer, 16, 347358.
Dukler, A. E., & Hubbard, M. G. (1975). A model for gasliquid slug
ow in horizontal and near horizontal tubes. Industrial Engineering
and Chemistry Fundamentals, 14, 337347.
Gregory, G. A., & Fogarasi, M. (1985). A critical evaluation of
multiphase gasliquid pipeline calculation methods. Proceedings
of the second international conference on multiphase ow, London
(pp. 93108).
Issa, R. I. (1988). Prediction of turbulent, stratied, two-phase ow in
inclined pipes and channels. International Journal of Multiphase
Flow, 14, 141154.
Kowalski, J. E. (1987). Wall and interfacial shear stress in stratied
ow in a pipe. A.I.Ch.E. Journal, 33, 274281.
Lam, C. K. G, & Bremhorst, K. (1981). A modied form of the kc
model for predicting wall turbulence. ASME Journal of Fluids
Engineering, 103, 456460.
Newton, C. H., & Behnia, M. (1996). Estimation of wall shear stress
in horizontal gasliquid stratied ow. A.I.Ch.E. Journal, 42,
23692373.
Newton, C. H., & Behnia, M. (2000). Numerical calculation of
turbulent stratied gasliquid pipe ows. International Journal of
Multiphase Flow, 26, 327337.
Olujic, Z. (1985). Predicting two-phase ow in horizontal pipes.
Chemical Engineering, 24, 4550.
Patel, V. C., Rodi, W., & Scheurer, G. (1985). Turbulence models
for near-wall and low Reynolds number ows: A review. AIAA
Journal, 23, 13081319.
Rodi, W. (1984). Turbulence models and their application in
hydraulics. IAHR Monograph Series.
Shoham, O., & Taitel, Y. (1984). Stratied turbulentturbulent gas
liquid ow in horizontal and inclined pipes. A.I.Ch.E. Journal,
30, 377385.
Spedding, P. L., & Hand, N. P. (1997). Prediction in stratied
gasliquid co-current ow in horizontal pipelines. International
Journal of Heat Mass Transfer, 40, 19231935.
Srichai, S., Jayanti, S., & Hewitt, G. F. (1995). CFD techniques for
separated two-phase ows. Proceedings of the rst Asian CFD
Conference, Hong Kong (pp. 10251031).
Taitel, Y., & Dukler, A. E. (1976). A model for predicting ow
regime transitions in horizontal and near-horizontal gasliquid
ow. A.I.Ch.E Journal, 22, 4755.

Das könnte Ihnen auch gefallen