Vous êtes sur la page 1sur 11

copyright C 1998, Institut Franais du Ptrole

TRANSIENT SIMULATION
OF TWO-PHASE FLOWS IN PIPES

SIMULATION TRANSITOIRE DES COULEMENTS


J.-M. MASELLA
DIPHASIQUES DANS LES CONDUITES
Principia Recherche et Dveloppement1
La simulation transitoire des coulements diphasiques gaz-liquide
dans les conduites requiert des efforts de calcul considrables.
Q. H. TRAN, D. FERRE and C. PAUCHON Jusqu' rcemment, la plupart des codes de simulation
commercialement disponibles taient bass sur des modles
Institut franais du ptrole2 deux fluides mettant en uvre une quation de conservation de la
quantit de mouvement pour chaque phase. Toutefois, dans les
processus normaux de transport ptrolier dans les conduites, et
spcialement de transport de ptrole ou de gaz, la rponse
transitoire du systme se rvle tre relativement lente. Ainsi, il est
raisonnable de penser qu'une forme simplifie des quations de
transport pourrait tre suffisante pour reproduire de tels
phnomnes transitoires. De plus, ce type de modle pourrait tre
rsolu numriquement au moyen d'algorithmes moins coteux en
temps calcul.

TRANSIENT SIMULATION OF TWO-PHASE FLOWS


IN PIPES

Transient simulation of two-phase gas-liquid flow in pipes requires


considerable computational efforts. Until recently, most available
commercial codes are based on two-fluid models which include
one momentum conservation equation for each phase. However, in
normal pipe flow, especially in oil and gas transport, the transient
response of the system proves to be relatively slow. Thus, it is
reasonable to think that simpler forms of the transport equations
might suffice to represent transient phenomena. Furthermore,
these types of models may be solved using less time-consuming
numerical algorithms.

SIMULACIN TRANSITORIA
DE LOS FLUJOS DIFSICOS EN LOS CONDUCTOS.

La simulacin transitoria de los flujos difsicos de gas lquido en


los conductos, precisa esfuerzos de clculo considerables. Hasta
hace poco tiempo, la mayor parte de los cdigos de simulacin
comercialmente disponibles se fundaban en modelos de dos
fluidos que aplican una ecuacin de conservacin de la cantidad
de movimiento para cada fase. No obstante, en los procesos
normales de transporte de hidrocarburos por medio de conductos,
y especialmente al tratarse del transporte de petrleo y de gas, la
respuesta transitoria del sistema se manifiesta ser de forma
relativamente lenta. Por todo ello, parece razonable pensar que
una forma simplificada de las ecuaciones de transporte, podra ser
de forma suficiente para la reproduccin de tales fenmenos
(1) ZI Brgaillon, transitorios. Adems, este tipo de modelo se podra resolver por
83857 La Seyne-sur-Mer
va digital mediante algoritmos menos costosos en cuanto a
(2) 1 et 4, avenue de Bois-Prau, tiempos de clculo.
92852 Rueil-Malmaison Cedex - France

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

801
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

INTRODUCTION First, we seek to analyze the differences in response


due solely to the model equations. For the purpose of
Transient simulations of gas-liquid flow in pipelines our study, the pipeline is visualized as a 1D element of
involve elaborate computer codes, the design and use of length L. The coordinate along the pipe is called x. We
which demand tremendous effort. Several codes have also assume that the pipe properties such as inclination
been proposed so far, and supported by the oil and gas q with respect to the horizontal, diameter D, roughness,
industry. Historically, OLGA is the first of these. etc. are constant with respect to x. Temperature is
Developed in Norway (Bendiksen et al. 1987, 1991), constant as well, and no mass transfer occurs between
OLGA is a two-fluid model with an additional the two phases. The responses of the three transport
momentum equation for the droplet field. The PLAC models are compared to selected transient scenarios,
code (Black et al. 1990), developed in England a few which exemplify typical operational transient scenarios.
years later, is also a two-fluid model. At the Institut Then, we examine the three codes from the stand-
franais du ptrole, another approach based on drift point of computing efficiency. The latter reflects the
flux type models has been used. The resulting code trade-off between computing time and accuracy of the
TACITE is meant to become a commercial product model. As for accuracy, it is defined in terms of the
(Pauchon et al. 1993, 1994). Another drift flux model, operating variables which are most significant for the
TRAFLOW, developed by the Shell Oil Company, end user, that is, the peak in outlet liquid flowrate
should also be mentioned. subsequent to an increase in the inlet gas flowrate.
In the case of OLGA and PLAC, the model develop-
ment and solution algorithms had been initiated earlier
by the nuclear industry. In the nuclear industry, fast 1 THE TWO-FLUID MODEL
transients associated to Loss Of Coolant Accidents
(LOCA) are of major interest, while in the oil and gas In dispersed flow, the contrast between the two
industry, the interest often lies in relatively slow phasic velocities is small. Understandably, it is
transients, associated with the transport and subsequent anticipated that the added value of the two-fluid model
release of slugs at receiving facilities. Under these will be more apparent in the stratified regime. This is
conditions, one may consider the momentum equation why we shall emphasize the stratified configuration in
to be a steady state force balance, thus leading to our presentation of TFM.
simpler and less elaborate calculations. Taitel et al.
(1989) put forward a simplified model in which the gas 1.1 Transport Equations
mass flow rate is declared to be in steady state.
Unfortunately, this approach lacked the ability to TFM is governed by a set of four partial differential
account for the transport time along the line for the case equations, the first two of which express mass
of gas flow rate variation at the inlet. conservation:
In this paper, we wish to study the behavior of
various types of models under a few transient scenarios. [rG R G ] + [rG RG VG ] = 0 (1)
t x
Three different models were implemented, namely:
A Two Fluid Model (TFM), based on one momentum
[r L RL ] + [r L RL VL ] = 0 (2)
conservation equation for each phase. t x
A Drift Flux Model (DFM), based on one momentum
and the last two of which represent momentum balance:
conservation equation and an algebraic slip relation.
A No Pressure Wave (NPW) model, based on an alge-
[rG RG VG ] + [rG RG VG2 + RG DPG ]
braic relation for the pressure drop and an algebraic t x (3)
slip relation.
+ RG P = t G + t i - rG RG g sin q
It is to be expected that these three models have very x
different analytical properties. For instance, while TFM

and DFM are hyperbolic models, the NPW model is a [r L RL VL ] + [r L RL VL2 + RL DPL ] (4)
mixed parabolic/hyperbolic. Therefore, the numerical t x
schemes in use will be different from one model to
+ RL P = t L - t i - r L RL g sin q.
another. x

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

802
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

In Equations (3) and (4), P denotes the interface represents conservative variables, as a function of
pressure and depends on masses (rG RG, rL RL), while which the vectors:
Vk, rk and Rk are respectively the velocity, the density
f T = (rG RG VG ; r L RL VL ; rG RG VG2 + RG DPG ;
and the volume fraction of phase k {G, L}.The (9)
variables ti and tk are the interfacial and wall moment- r L RL VL2 + RL DPL )
um exchange terms, q is the flowline inclination to
the horizontal. Moreover we assume the constitutive (10)
r T = ( 0; 0; RG ; RL )
law RG + RL = 1.
The quantities DPG and DPL correspond to the static
head around the interface (De Henau and Raithby, q T = ( 0; 0; t G + t i - rG RG g sin q;
1995) defined as: t L - t i - r L RL g sin q) (11)
1 w
DPG = PG - P = -rG cos are computed. It is also possible to transform (7) into
2 2 the quasi-linear form:
w (5)
sin 3 gD cos q
1
+
3pRG 2 [ w ] + A( w ) [ w ] = q ( w ) (12)
t x
1 w
DPL = PL - P = r L - cos
Under appropriate conditions (Masella, 1997), e.g. if
2 2 (13)
VG - VL p cm
(6)
w
sin 3 gD cos q
1
+ where cm is a pseudo-sound-velocity of the mixture, the
3pRL 2
matrix A(w) has four real eigenvalues:
w being the wetted angle.
l1 ( w ) l 2 ( w) l 3 ( w) l 4 ( w )

as well as a base of eigenvectors. This property is


commonly known as hyperbolicity. This hyperbolicity is
Gas depends on the relation (13). Note that, should the static
head terms DPG and DPL be missing in Equations (3)
w and (4), then hyperbolicity would be lost as soon as
Liquid VG - VL.
From the standpoint of physics, the extreme
Scheme 1 eigenvalues l1 and l4 are associated with acoustic
Definition of wetted angle. waves and, therefore, can be very large, especially
when the mixture is mainly composed of liquid. The
P intermediate eigenvalues l2 and l3 are associated with
The densities are assumed to be given by rG = and void fraction waves, and their orders of magnitude,
aG2
P - PL0 which are close, are about those of fluid velocities.
r L = r0L + where ak is the speed of sound in
aL2
phase k, and (r0L , PL0 ) is a reference state of the liquid 1.2 Numerical Scheme
phase.
It is convenient to rewrite Equations (1)-(4) in a more The resolution algorithm is based on a finite volume
abstract way as: method, adapted to the non-conservative form (12). The
reader is referred to Masellas Thesis (1997) for a more
detailed description of the numerical implementation.
[ w ] + [f ( w )] + r( w ) [ P] = q( w ) (7)
t x x Let us divide the pipeline into a sequence of uniform
cells Mi = [xi 1/2, xi + 1/2], the length of each of which is
where the vector: Dx. The unknowns are located at the center of the cells.
w T = (rG RG ; r L R L ; rG RG VG ; r L RL VL ) (8) Let wi be the one associated to the cell Mi.

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

803
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

Discretization in space is based on a finite volume with velocities defined at the mesh boundaries and
formulation of non-linear system (12) and reads pressure and void fraction defined within the cells, alle-
d h - h i -1 / 2 P - Pi -1 / 2 viates the need to specify an extra boundary condition.
[ w i ] + i +1 / 2 + ri i +1 / 2 = qi (14) A careful investigation into different terms in
dt Dx Dx
Equations (3) and (4) reveals that, in slow transients, the
where hi+1/2 and Pi+1/2 are obtained via a linearized prevailing terms are the pressure drop and the source
Riemann problem at the interface xi+1/2. This simplified terms. When all remaining terms are neglected,
Riemann problem, defined from the surrounding states Equations (3) and (4) degenerate into:
wi and wi+1, involves a rough Godunov solver and is
referred to as the VFRoe solver (Masella et al., 1996). t G + t L - (rG RG + r L RL )g sin q - [ P] = 0 (17)
x
Let wi+1/2 be the solution at the interface of the rough
Riemann problem. Then hi+1/2 and Pi+1/2 are defined by: RG ( t L - t i - r L RL g sin q)
(18)
h i+1/ 2 = f ( w i +1 / 2 ) and Pi +1 / 2 = P( w i +1 / 2 ) (15) - RL ( t G + t i - rG RG g sin q)
This suggests to directly impose (18) as the third
The quantity ri can be put equal to r(wi). boundary condition at the inlet.
We can note that the discretization of the non- The next two figures illustrate what has just been
conservative system (12) is obtained without any non- said about boundary conditions. Two numerical runs
conservative reformulation like in Sainsaulieu 1995 and were performed for a pipe of length L = 10 000 m. At
Toumi et al., 1996, but by a direct integration of the the inlet, the flowrates are 10 kg/s (liquid) and 0.1 kg/s
non-conservative system on each cell. A similar (gas). At the outlet, the pressure is maintained at 1 bar.
approach was used by Louis, 1995. Since three eigenvalues are positive at the inlet, a third
Note that the VFRoe scheme has been tested boundary condition is required. We attempt to impose
(Gallouet et al., 1996 ; Masella et al., 1996) with RG = 0.6.
success on some hyperbolic conservative systems, such In the first run, the diameter is equal to D = 0.25 m,
as Eulers and has given good results. These are quite which corresponds to large friction terms. It can be seen
comparable with the well-known Roe scheme. from Figure 1 that the inlet value RG = 0.6 is immedia-
The time discretization is explicit with respect to the tely dissipated and as a consequence, does not exercise
intermediate eigenvalues and linearly implicit with any influence over the inner state of the mixture. This
respect to the extreme eigenvalues. phenomenon is insensitive to the mesh size.
One of the trickiest problems here is to deal with
the boundary conditions. Usually, two inlet mass flow
rates and one outlet pressure are imposed. However, if 0.64
l1(w) < 0 < l2(w) l3(w) l4(w) at the inlet, then a 25
further inlet datum must be supplied. At first sight, a 0.63 50
Gas vol fraction (0.1)

condition on the gas volumetric fraction RG or on the


100
gas mass fraction aG = rG RG / (rG RG + rL RL) seems to 0.62 200
be relevant. Nevertheless, it comes as a surprise that
such a choice does not always give rise to acceptable 0.61
results, and in reality, the source terms must be taken
into account. 0.60
As a matter of fact, most two-fluid type codes
impose a von Neuman type boundary condition (De 0.59
Henau and Raithby, 1995) such as: 0 1000 2000 3000 4000 5000
Position (m)

[ RL ] = 0 (16)
x Figure 1
Profile in space of the gas volumetric fraction for several
Others merely extrapolate the value of the liquid mesh sizes (25, 50, 100 and 200 cells), with D = 0.25 m
(zoom up to 5000 m).
fraction. The staggered mesh approach used in OLGA,

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

804
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

In the second run, the diameter is increased to The interfacial exchange term ti is no longer present
D = 0.75 m, so as to reduce the magnitude of friction in the above equation. This leads us to consider DFM, a
terms. From Figure 2, it is now obvious that the inlet new model which consists of three partial differential
value RG = 0.6 is perfectly compatible with the inner Equations, i.e. (1), (2) and (19). In order for DFM to be
state. cast into the strictly conservative form:

[ w ] + [f ( w )] = q( w ) (20)
0.67
t x

0.66 where the flux and the source:


Gas vol fraction (0.1)

0.65 f T = (rG RG VG ; r L RL VL ; rG RG VG2 + r L RL VL2 + P) (21)


0.64 q T = ( 0; 0; t G + t L - (rG RG + r L RL )g sin q) (22)
0.63 25
are to depend only on:
0.62 50
100 w T = (rG RG ; r L RL ; rG RG VG + r L RL VL ) (23)
0.61
200
0.60
it is necessary to introduce an algebraic relation called a
0 2000 4000 6000 8000 10 000 closure law or slip model.
In stratified flow, the slip relation is obtained by
Position (m)
Figure 2 combining the two momentum conservation Equations
Profile in space of the gas volumetric fraction for several mesh (3) and (4) in such a fashion that the pressure gradient
sizes (25, 50, 100 and 200 cells), with D = 0.75 m. vanishes. By neglecting derivatives with respect to time,
static head and inertia terms in this combination, we
end up with:
2 THE DRIFT-FLUX MODEL tG t L t
- + i + (rG - r L )g sin q = 0 (24)
RG RL RG RL
The Drift Flux Model is derived from the Two-Fluid
Model by neglecting the static head terms DPG and DPL For other flow regimes, the closure law may be much
in the last two Equations (3) and (4) and replacing the more sophisticated.
two momentum equations by their sum. The main It is convenient to rewrite (19) in the quasi-linear
advantages of this 3-equation model are: form:
The equations are in conservative form, which makes
it easier to discretize by finite volume methods; [ w ] + A( w ) [ w ] = q ( w ) (25)
t x
The interfacial friction term ti is cancelled out in the
momentum equation, although it appears in an add- in which A is the Jacobian matrix of f with respect
itional algebraic relation called slip law; to w. If hyperbolicity holds, there are three eigenvalues
One does not have to work out a third boundary l1 l2 l3 and a base of eigenvectors. As is the case
condition at the inlet; for TFM, the extreme eigenvalues are associated with
It is generally hyperbolic depending on the form of acoustic waves, whereas the intermediate eigenvalue is
the slip law. associated with a void fraction wave.

2.1 Transport Equations 2.2 Numerical Scheme

Adding (3) and (4) together yields: In this section, the basic ideas of the numerical
scheme are outlined. Further details are to be found in

[rG RG VG + r L RL VL ] + [rG RG VG2 + r L RL VL2 + P] (19) Faille and Heintz (1996).
t x
Once again, let us divide the pipeline into regular
= t G + t L - (rG RG + r L RL )sin q cells Mi = [xi1/2, xi+1/2], whose size is denoted by Dx.

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

805
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

The unknowns are located at the center of the cells. Let Let us introduce superficial velocities
wi be the one associated with the cell Mi. Discretization UG = RGVG (29)
in space reads:
UL = RLVL (30)
d h - h i -1 / 2
[ w i ] + i +1 / 2 = qi (26)
dt Dx Us = UG + UL (31)
with: and let us consider
1 1
h i +1 / 2 =
2
[ f ( w i ) + f ( w i +1 )] - Di +1 / 2 ( w i +1 - w i )
2
(27) vT = (RG, P, Us) (32)

as principal unknowns. This choice of variables turns


In this formula, Di+1/2 is a diffusion matrix that has to out to be very handy for tackling the limiting case of
be built up from wi and wi+1 to ensure sufficient incompressible flows. The equations of NPW can be
stability. Extension to second order accuracy is summarized as:
achieved via the MUSCL strategy. Scheme (26)-(27)

may appear to be more stable than standard Godunov- [e( v)] + [f (v)] = q(v) (33)
type schemes using approximate Riemann solvers t x
(Godlewski and Raviart, 1996) at sides xi+1/2 namely with:
when some discontinuity, such as pipe slope change, at eT = (rG RG; rL RL; 0) (34)
side xi+1/2 occurs.
The time discretization is explicit with respect to the fT = (rG RG VG; rL RL VL; P) (35)
intermediate eigenvalue and linearly implicit with
respect to the extreme eigenvalues. qT = (0; 0; tG + tL (rG RG + rL RL) g sin q) (36)

The slip law is formally expressed as:


3 THE NO-PRESSURE-WAVE MODEL
UG = Y (RG, P, Us) (37)
Experience with numerical simulations has shown
where the function Y may admit the pipe angle of
that, in most transients of interest to the oil and gas
inclination q as a parameter.
transport industry, pressure waves do not have a strong
effect on the initiation and transport of void waves. Viviand (1996) shows that this model is a good
Hence, in a bolder step toward simplification, we would approximation of the DFM as long as the phasic
like to rule out the very existence of acoustic waves velocities are small compared to the sound wave
from the model equations. velocities, which is true for most applications. He also
proves that NPW always has a single finite eigenvalue,
equal to:
3.1 Transport Equations

l(v) = [ Y( v)] (38)
Preliminary numerical investigations into the order RG
of magnitude of different terms in (18) suggest that,
in slow transients, it is legitimate to neglect the The characteristic equation associated to this
inertia terms. Doing so amounts to replacing the eigenvalue can be written as:
mixture momentum Equation (19) by a local static
force balance. l T ( v) [v] + l [v] = l T ( v) q( v) (39)
t x

[ P] = t G + t L - (rG RG + r L RL )g sin q (28) lT (v) being an appropriate left eigenvector. Besides,
x there exists an algebraically-double eigenvalue equal to
. On this ground, the model is qualified as mixed
The NPW model is made up of (1), (2) and (28). The hyperbolic/parabolic. The number of characteristic
three partial differential equations are complemented by equations associated with is usually one, i.e. (28), but
an algebraic slip law. may reach two for special thermodynamic laws.

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

806
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

3.2 Numerical Scheme Next, we compare the response of the three codes to
the experimental measurements. We also plot the
In this section, the basic ideas of the numerical computing time as a function of the accuracy for each
scheme are outlined. Further details are to be found in model response.
Patault and Tran (1996).
Finally, we look at the response of NPW and DFM in
The pipeline is divided into regular cells Mi = [xi1/2, a scenario giving rise to the severe slugging
xi+1/2, the size of which is denoted by Dx. The phenomenon. In this case, the differences in response
unknowns are located at the center of the cells. Let vi be are explained by arguments involving the transport
the one associated to the cell Mi. Unlike the two models and the numerical schemes.
previous schemes, the NPW equations are discretized in
a centered way at the sides xi+1/2. In other words:
4.1 Transient Response of Models
1 d f ( v i +1 ) - f ( v i )
[e( v i ) + e( v i +1 )] + for the Reference Case
2 dt Dx
1 (40) Let us start by describing the actual values of para-
= [q( v i ) + q( v i +1 )] meters for the reference case.
2
Figure 3 compares the transient response of DFM and
Time discretization of (40) is totally implicit. The NPW for Dx = 2 m. Below this value of the mesh size,
final scheme is first order accurate in time, and second no significant improvement in the accuracy of the solu-
order in space. tion for each model is observed. Therefore, we assimilate
Equation (40) has to be modified over a cell across the curves to the analytical solutions of the models.
which the sign of the eigenvalue l changes, to track
shocks and rarefaction waves. An algorithm, inspired
from Chattot and Mallet (1987), is set up to build the sys- 14 1.5
DFM
tem of equations to be solved at each time step, according 12 NPW 1.2
to the characteristic velocity sign configuration. Differences
10 0.9
QL at outler (m3/h)

8 0.6
4 COMPARISONS OF THE MODELS
6 0.3
To compare the response of the three models, we first 4 0
define a reference test case. This is done in Table 1. For
this test case, we display the theoretical response of 2 -0.3

each transport model. By theoretical response, we mean 0 -0.6


the limit response obtained by decreasing the mesh size 0 50 100 150 200 250 300
Dx to 0. Time (s)

TABLE 1 Figure 3
Comparison between DFM and NPW on outlet liquid flow rate.
Definition of the reference test case

Length 420 m
Pipe geometry Diameter 3 Note that the initial state is obtained solving each
Configuration Horizontal system TFM, DFM and NPW neglecting time differen-
Inlet gas flowrate From 0.0048 tial terms and keeping boundary conditions constant.
to 0.0354 kg/s in 20 s
To avoid discrepancies due to the slip model, we use
Transient scenario Inlet liquid flowrate 0.191 kg/s
Outlet pressure 1.68 bar
in both cases a Zuber and Findlay (1965) slip model,
which is an alternative to the slip model defined in
Composition Air/Kerosen
Gas density 2.418 kg/m3
Equation (24). The curves in Figure 3 exhibit very little
Fluid definition Liquid density 813 kg/m3 difference in the transient response of the two models.
and properties Gas viscosity 0.76 105 m2/s This corroborates the theoretical prediction by Viviand,
Liquid viscosity 0.22 105 m2/s according to which pressure waves play a minor role in
Superficial tension 0.03 N/m2 the propagation of the liquid.

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

807
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

4.5 0.35
DFM DFM x = 61 m
4.0
TFM TFM x = 61 m
0.30

Liquid vol fraction (0.1)


3.5 DFM x = 395 m
QL at outler (m3/h)

TFM x = 395 m
3.0
0.25
2.5

2.0 0.20

1.5
0.15
1.0

0.5 0.10
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time (s) Time (s)

Figure 4a Figure 4b
Comparison between DFM and TFM for outlet liquid flow Comparison between DFM and TFM for liquid volume
rate fraction.

Figure 4 compares the transient responses of DFM In Figure 5 we try to characterize the computing
and TFM, using a stratified slip model as specified in efficiency of the different codes. For this purpose, we
Equation (24). One must bear in mind the fact that define the computing efficiency as the relation
TFM includes a pressure difference term due to the between the computing time and the peak height in the
liquid fraction gradient, which is not included in DFM. liquid outlet flowrate. This maximum in the liquid
Although the transient imposed on the system is outlet flowrate is important for pipeline design because
quite rapid (a 20-second ramp in inlet gas flow rate), the it characterizes the outlet liquid flowrate surge into
responses of DFM and TFM are quite similar. This the separator following an increase in the inlet gas
result tends to indicate that TFMs decoupling of the flowrate. The estimation of this liquid surge will
phase velocities does not significantly affect the eventually control the outlet separator volume design
transient response of the model. In other words, it is the recommendation.
source terms that actually control the transient response. The diagram in Figure 5 shows that for a given
accuracy, the mixed explicit/implicit solution used in
the DFM code is faster than the implicit resolution in
the NPW code.
15
Maximum outlet liquid flow rate (m3/h)

14 4.2 Comparison with Experimental Data

13 Next we compare the code responses with


experimental data (Vigneron, 1995) taken for the same
12 test case. Figure 6 compares the experimental results
DFM with the transient response of the DFM using a slip
NPW model which takes into account the different flow
11
regimes which can be encountered.
10 Figure 6a. Comparison between measured and
10 100 1000 10000 100 000 computed liquid fraction using DF model.
Computing time (s) The large amplitude oscillations observed during the
Figure 5 transient period signal the occurrence of the slug flow
Comparison of computing efficiency between DFM and NPW regime. The recognition of this flow regime has
model. important implications for the pressure drop
calculation.

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

808
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

1.0 2.4
DFM x = 61 m
0.9 NPW x = 395 m 2.3 DFM x = 61 m
0.8 Expe. x = 61 m Expe. x = 61 m
Liquid vol fraction (0.1)

Expe. x = 395 m 2.2


0.7

Pressure (bar)
0.6 2.1

0.5 2.0
0.4 1.9
0.3
1.8
0.2
0.1 1.7

0.0 1.6
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time (s) Time (s)
Figure 6a Figure 6b
Comparison between measured and computed liquid fraction using Comparison between measured and computed pressure using the
DF model. DF model.

2.4 TABLE 2

2.3 Definition of the severe slugging test case


DFM
TFM
Pressure at 61 m (bar)

2.2 Experiment Length 60 m and 14 m


2.1 Pipe geometry Diameter 2
Configuration Horizontal and vertical
2.0
Inlet gas flowrate 0.000196 kg/s
1.9 Transient scenario Inlet liquid flowrate 0.07854 kg/s
1.8 Outlet pressure 1.0 bar

1.7 Composition Air/Kerosen


Gas density 1.0 kg/m3
1.6
0 200 400 600 800 1000 1200 Fluid definition Liquid density 1000 kg/m3
and properties Gas viscosity 1.5 105 m2/s
Time (s)
Liquid viscosity 1.5 106 m2/s
Figure 7 Superficial tension 0.07 N/m2
Comparison between experimental and computed pressure
using the DFM and TFM models with an imposed stratified
configuration. Slight differences are observed in the period and
amplitude of oscillations. However these can be
attributed to the differences in treatment of the outlet
Figure 7 compares the same experimental results boundary conditions. As can be seen, the DFM solution
with the transient response of the DFM and TFM with algorithm at the outlet does not allow for liquid going
the slip model assuming a stratified flow configuration. back into the pipeline during the reflux period. This
problem is resolved by introducing a fictitious cell at
From Figures 6b and 7, one can see the important
the outlet in which we assume single phase gas flow.
effect of the flow regime prediction and subsequent slip
relation on the pressure transient response of the model.
CONCLUSION
4.3 Severe Slugging Case
Extensive numerical tests have been carried out in
Lastly, we compare the DFM and the NPW models order to compare the analytical response of the three
for a severe slugging test case. following models:
Table 2 defines the severe slugging test case. Figure 8 A Two Fluid Model with one momentum equation
shows the comparison between the two codes. for each phase;

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

809
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

2.20 3.0
DFM DFM
2.15 NPW 2.5 NPW
Pressure at 61 m (bar)

2.10

QL at outlet (m3/h)
2.0
2.05
1.5
2.00
1.0
1.95
0.5
1.90

1.85 0.0

1.80 -0.5
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time (s) Time (s)

Figure 8a Figure 8b
Comparison between the DFM and the NPW models for Comparison between the DFM and the NPW models for
pressure at riser bottom for a severe slugging case. outlet liquid flow rate for a severe slugging case.

A Drift Flux Model with only one momentum REFERENCES


equation for the mixture;
Bendiksen K., Brandt I., Jacobsen K.A. and Pauchon C. (1987)
A No Pressure Wave model reducing the momentum Dynamic simulation of multiphase transportation systems.
equation to a force balance. Multiphase Technology and Consequences for Field Development
On one hand, different hydrodynamic slip laws have Forum, Stavanger, Norway.
been incorporated into these models. The results clearly Bendiksen K., Malnes D., Moe R. and Nuland S. (1991) The
dynamic two-fluid model OLGA: Theory and application. SPE
show that the choice of the former has an critical effect Production Engineering, 6, 171-180.
on the dynamic response of the latter. Black P.S., Daniels L.C., Hoyle N.C. and Jepson W.P. (1990)
On the other hand, comparison between the different Studying transient multiphase flow using the Pipeline Analysis
Code (PLAC). Journal of Energy Resources Technology, 112,
transport models confirms the fact that the source terms 25-29.
in the momentum equation do dominate the transient Chattot J.J. and Malet S. (1987) A box-scheme for the Euler
response of the model. This holds true for all models, but equations. In: Non-Linear Hyperbolic Problems, Eds. Carasso C.,
becomes overwhelmingly apparent in the case of TFM. Raviart P.A. and Serre D., Lecture Notes in Mathematics, 1270,
Springer-Verlag.
The computing efficiency as defined from the end De Henau V. and Raithby G.D. (1995) A transient two-fluid
user's perspective, that is, the peak in outlet liquid flow- model for the simulation of slug flow in pipelines, 1-Theory.
rate subsequent to an increase in the inlet gas flowrate, Int. J. Multiphase Flow, 21, 335349.
has been analyzed for DFM and NPW. For a given Gallout T , Masella J.M. (1996) On a rough Godunov scheme,
CRAS, Paris, I-323, 77-84.
accuracy, the DFM explicit/implicit solution is faster
Faille I. and Heintz E. (1996) Rough finite volume schemes for
than the implicit solution employed in the NPW code. modelling two-phase flow in a pipeline. In: Proceedings of the
CEA, EDF, INRIA Course: Mthodes numriques pour les
coulements diphasiques, INRIA, Rocquencourt, France.
ACKNOWLEDGMENTS Godlewski E. and Raviart P. A. (1996) Numerical Approximation
of Hyperbolic Systems of Conservation Laws, Springer-Verlag,
New York.
We gratefully acknowledge the contributions of Louis X. (1995) Doctoral Thesis, University of Pierre et Marie
H. Viviand for his analysis of the NPW model and of Curie, Paris, France.
S. Patault for its numerical implementation. We also Masella J.M. (1997) Doctoral Thesis, University of Pierre et
wish to thank J. Brill (TUFFP) for allowing us to use Marie Curie, Paris, France.
one of his experimental data sets as a reference test Masella J.M, Faille I. and Gallout T. (1996) On a Rough
Godunov Scheme, Submitted.
case. This work was performed as part of the TACITE
Patault S. and Tran, Q.H. (1996) Modle et schma numrique du
project supported by the Institut franais du ptrole, Elf code TACITE-NPW. Technical Report 42415, Institut franais du
Aquitaine, and Total Exploration Production. ptrole, France.

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

810
TRANSIENT SIMULATION OF TWO-PHASE FLOWS IN PIPES

Pauchon C., Dhulesia H., Lopez D. and Fabre J. (1993) TACITE: Viviand H. (1996) Modle simplifi d'coulement dans les
A comprehensive mechanistic model for two-phase flow. In: conduites ptrolires. Technical Report RSY, 43 069 02, Principia,
Proceedings of the 6th International Conference on Multiphase France.
Production, Cannes, France.
Vigneron F. (1995) Analysis of imposed two-phase flow transients
Pauchon C., Dhulesia H., Binh-Cirlot G. and Fabre J. (1994) in horizontal pipelines, Part 1-Experimental results. Research
TACITE: A transient tool for multiphase pipeline and well Report, Tulsa University Fluid Flow Project, USA.
simulation. SPE Annual Technical Conference, 28545, 25-28,
New Orleans, LA, USA. Toumi I., Kumbaro A. (1996) An approximate linearized Riemann
solver for a two-fluid model. J. Comput. Physics., 124, 286-300.
Sainsaulieu L. (1995); Thse d'habilitation, University of Pierre
et Marie Curie, Paris, France. Zuber N. and Findlay J.A. (1965) Average volumetric
concentration in two-phase flow systems. J. Heat Transfer Series
Taitel, Y, Shoham O. and Brill J. P. (1989) Simplified transient
solution and simulation of two-phase flow in pipelines. Chem. C 87, 453-458.
Eng. Sci., 44, 13531359. Final manuscript received in October 1998

REVUE DE LINSTITUT FRANAIS DU PTROLE


VOL. 53, N 6, NOVEMBRE-DCEMBRE 1998

811