You are on page 1of 9

SPE 109868

A Basic Approach to Wellbore Two-Phase Flow Modeling


A.R. Hasan, SPE, University of MinnesotaDuluth; C.S. Kabir, SPE, Chevron ETC; and M. Sayarpour, SPE,
University of Texas at Austin
Copyright 2007, Society of Petroleum Engineers
This paper was prepared for presentation at the 2007 SPE Annual Technical Conference and
Exhibition held in Anaheim, California, U.S.A., 1114 November 2007.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than
300 words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, Texas 75083-3836 U.S.A., fax 01-972-952-9435.

Abstract
This study presents a simplified two-phase flow model using
the drift-flux approach to well orientation, geometry, and
fluids. For estimating the static head, the model uses a single
expression for liquid holdup, with flow-pattern-dependent
values for flow parameter and rise velocity. The gradual
change in the parameter values near transition boundaries
avoids discontinuity in the estimated gradients, unlike most
available methods. Frictional and kinetic heads are estimated
using the simple homogeneous modeling approach.
We present a comparative study involving the new model
as well as those that are based on physical principles, also
known as semimechanistic models. These models include
those of Ansari et al, Gomez et al., and OLGA. Two other
widely used empirical models, Hagedorn and Brown and PE2, are also included. The main ingredient of this study entails
the use of a small but reliable dataset, wherein calibrated PVT
properties minimizes uncertainty from this important source.
Statistical analyses suggest that all the models behave in a
similar fashion and that the models based on physical
principles appear to offer no advantage over the empirical
models. Uncertainty of performance appears to depend upon
the quality of data input, rather than the model characteristics.
Introduction
Modeling two-phase flow in wellbores is routine in every-day
applications. The use of two-phase flow modeling throughout
the project life cycle with an integrated asset modeling
network has rekindled interest in this area. Plethora of models,
some based on physical principles and others based on pure
empiricism, often beg the question which one to use in a given
application. Although a few comparative studies (Ansari et al.
1994; Gomez et al. 2000; Kaya et al. 2001) attempt to answer
this question, often reliability of the data base has left this
issue unsettled.
One of the main objectives of this paper is to present a
simplified two-phase flow model, which is rooted in drift-flux

approach. The drift-flux approach (Hasan and Kabir, 2002, pp.


21-62, Shi et al. 2005a, 2005b) has served the industry quite
well, as exemplified by its simplicity, transparency, and
accuracy in various applications. The second objective is to
show a comparative study with a few models using a small but
reliable data base to get a perspective on relative performance.
Here, data reliability stems from two elements: rate and fluid
PVT properties. Pressure data are typically gathered with
permanent downhole and wellhead sensors while rate data are
measured with surface flow meters or test separators. In each
case, the black-oil fluid PVT model was conditioned with
laboratory data to ensure reliability and consistency.
Proposed Model
Total pressure gradient during any type of fluid flow is the sum
of the static, friction, and kinetic gradients, the expressions for
which are given in Eq. A-1 in the Appendix. For most vertical
and inclined wells, the static head componentwhich directly
depends on the volume average-mixture densitydominates.
Thus, in simple terms, two-phase flow modeling boils down to
estimating density of the fluid mixture or gas-volume fraction.
Because gas-volume fraction depends on whether the flow
is bubbly, slug, churn, or annular, we individually model each
flow regime. However, for all flow regimes the gas (or lighter)
phase moves faster than the liquid (or heavier) because of its
buoyancy and its tendency to flow close to the channel center.
This allows us to express in-situ gas velocity as the sum of
bubble rise velocity, v, and channel center mixture velocity,
Covm. However, in-situ velocity is the ratio of superficial
velocity to volume fraction. Therefore, the generalized form of
gas-volume fraction relationship with measured variables
superficial velocity of gas and liquid phasescan be written as

fg

v sg
C o v m v

(1)

For downflow, buoyancy acts in the direction opposite to


flow. Thus, in Eq. 1, the negative sign in front of the rise
velocity applies to downflow and the positive sign is meant for
upflow.
While Eq. 1 is universal in its application, the values of the
flow parameter Co, and rise velocity v, are dependent on the
type of flow and flow pattern. Table 1 presents these values.
In the following, we briefly explain the rationale for these
values; greater detail is provided in Appendix A.

SPE 109868

Table 1 Parameters for flow type and flow pattern.


Flow Parameter, Co

Flow
Pattern

Upward

Countercurrent

Rise
Downward

Velocity
v

cocurrent

Bubbly/
D.Bubbly
Slug

1.2

2.0

1.2

vb

1.2

1.2

1.12

Churn

1.15

1.15

1.12

Annular

1.0

1.0

1.0

For turbulent flow, prevelant in most production situations,


velocity profile across the channel is relatively flat and the
velocity at the center is about 1.2 times the average velocity.
In bubbly and slug flow, most of the bubbles move through
this central portion, which explains the value of 1.2 for Co in
these two flow regimes. Exceptions are countercurrent bubbly
flow where vmax = 2.0vm and downward slug flow where
vmax = 1.12vm. In churn flow, the high mixture velocity that
breakes up the large bubbles also circulates the gas more
throughly across the channel, resulting in slightly lower value
of Co. Our recent work stongly suggests no slippage between
the two phases in annular flow; consequently, Co = 1 and v =
0 in this flow regime (Kabir and Hasan, 2006; Hasan and
Kabir, 2007) are well justified.
For rise-velocity of small bubbles in bubbly flow, vb, we
use the Harmathy correlation, which applies even to
liquid/liquid flow
v b

[(

= 1.53 g L g / L 2

1/ 4

(2)

Comparative Study of Six Models


As pointed out earlier, we used a small but high-quality
dataset to test the models performance relative to those
commonly used. These models include those of Hagedorn and
Brown (1965), PE-2 (2006), steady-state OLGA (1991),
Ansari et al. (1994), Gomez et al. (2000), and the one
presented here, designated as the HK model. Table 2
summarizes the range of variables used in the study.
Overall, all the models behave very similarly. For instance,
Figs. 1 and 2, respectively, show that the data are well
correlated by both the popular Hagedorn-Brown method and
the one proposed here. Put another way, when we crossplot
computed bottomhole pressures between different models,
good correlations are apparent. Figs. 3 and 4 are cases in
point.
Table 2 Range of variables used in the comparative study.

Variable

Maximum

Minimum

Liquid Rate, STB/D


Gas/Liquid Ratio, scf/STB
Watercut, %
Measured Depth, ft
Bottomhole Pressure, psig
Wellhead Pressure, psig

27,760
5,615
98
30,650
13,755
10,218

100
104
0
1,635
108
90

To put the performance of various models in proper


perspective, we resorted to statistical analyses. Table 3
showing correlation coefficients of 212 test points suggest that
all the models are highly correlated with data. In fact, the
significance of the correlation coefficient, r, is presented in
terms of the t values (shown in the last row) given by the
following expression (Blalock 1972, p. 400):

t = r (n 2) / (1 r 2 )

The Taylor bubble-rise velocity, vT, has been investigated


by many researchers. We use the following relation for its
estimation in vertical cylindrical channels; modification for
other flow directions and flow types are detailed in Appendix
A:

v T = 0.35 gD l g / L

(3)

Appendix A presents this model in detail, including the


procedure for calculating the average-rise velocity, v , of the
small and Taylor bubble in slug and churn flow.
Overall, presentation of this model differs in three
important ways from pervious studies. First, we show unified
formulation of the model in various flow types, such as
cocurrent upward (gas/oil and oil/water), countercurrent
(gas/oil), and downward (steam/water). Second, transition
from one flow regime into another is smoothed by a weighting
scheme. Third, annular flow is modeled with the no-slip
approach, for both simplicity and accuracy.

(4)

where n indicates the number of data points. We point out that


for any of the models to assure existence of correlation with
data, we needed a t value of only 2.6 within the 99%
confidence interval. As the second last row shows, all the t
values are well over 100, thereby indirectly suggesting that
sufficient data points exist for all these models to apply
equally within the confidence interval. Similarly, the F-test
values displayed in the last row exceed the critical value of
6.63, thereby attesting the t-test results.
To explore the level of similarity amongst the models, we
did another statistical t-test in accord with the following
relation (Blalock 1972, p. 407):

{(

)}

t xz = (rxy rzy ) {(n 3)(1 + rxz )} / 2 1 rxy 2 rxz 2 rzy 2 + 2rxy rxz rzy

(5)
In Eq. 5, y is the data set array and x and z represents the
model estimation arrays. Table 4 compares t values between
the correlations to show likelihood between any model pairs
for the same dataset. Those values colored in blue suggest that
no significant differentiation between the performances of the
model pair can be made because they are below the t value of
2.6 within the confidence interval of 99%. For those pair of

SPE 109868

models shaded black we conclude that models are


significantly different, but not necessarily inferior to others
because the previous test showed their reliability. Put simply,
four of the models correlate with each other better than the
other two.

input data, entering typical top-down pressure-traverse


calculations, perhaps plays a larger role in any comparative
study. Let us explore this point.
100,000
Gomez pwf, psig

HB/Gomez

Computed pwf, psig

15,000
12,000
9,000
6,000

p wfc = 1.0416p wfm

10,000

R 2 = 0.9854

3,000

y = 1.0039x
R2 = 0.9969

1,000

100
100

1,000

0
0

3,000

6,000

9,000

Fig. 4 Performance comparison of HB and Gomez models.

Table 3 Correlation coefficient for different models.

Fig. 1 Performance of Hagedorn-Brown (HB) correlation.

Computed pwf, psig

15,000

HB
1

HB

12,000
9,000
6,000
y = 1.0663x
R2 = 0.9865

0
0

3,000 6,000 9,000 12,000 15,000


Measured p wf , psig

HK pwf, psig

y = 1.0223x
R2 = 0.9966

100
100

1,000

10,000

100,000

HB p wf , psig
Fig. 3 Performance comparison of HB and HK models.

Uncertainty Analysis
Over the years, many authors (Aggour et al. 1996, Ansari et al.
1994, Pucknell et al. 1993, and Kabir and Hasan 1990, to
name a few) have attempted comparative studies to discern
reliability of two-phase flow models for diverse applications.
However, no model has emerged as single most reliable tool
for routine applications, although Hagedorn-Brown and PE-2
models enjoy widespread popularity. Based on the data
analysis of the previous section, we think that reliability of

Ansari

Gomez

HK

0.9998

0.9994

0.9984

0.9986
0.9989

0.9998

0.9998

0.9990

0.9980

0.9998

0.9998

0.9993

0.9983

0.9992

Ansari

0.9994

0.9990

0.9993

0.9994

0.9982

Gomez

0.9984

0.9980

0.9983

0.9994

0.9973

HK

0.9986

0.9989

0.9992

0.9982

0.9973

Data

0.9941

0.9941

0.9939

0.9931

0.9922

0.9937

132.52

132.87

130.75

122.42

114.98

128.42

24,849

26,738

26,323

21,961

19,657

27,722

Table 4 t-test values for different models.


HB

10,000

OLGA

0.9998

OLGA

100,000

PE-2

PE-2

Fig. 2 Performance of Hasan-Kabir (HK) model.

1,000

100,000

HB p wf , psig

12,000 15,000

Measured p wf , psig

3,000

10,000

HB
NA

PE-2

OLGA

Ansari

Gomez

0.211

0.965

3.938

4.570

0.975

HK

PE-2

0.211

NA

1.192

3.161

4.076

1.195

OLGA

0.965

1.192

NA

2.924

3.980

0.740

Ansari

3.938

3.161

2.924

NA

3.247

1.362

Gomez

4.570

4.076

3.980

3.247

NA

2.735

HK

0.975

1.195

0.740

1.362

2.735

NA

Consider a highly deviated 31,000-ft well from our dataset


producing at 6,394 STB/D. Fig. 5 shows that the computed
pressures from all the six models are clustered together and
suggest higher than that measured downhole with a permanent
sensor. Domination of single-phase liquid and bubbly flow
explains why all the models behave so similarly. The reported
GOR of 703 scf/STB translates into a maximum spread of
only 73 psi or 1.27% of the minimum pressure of 5,717 psig,
obtained from the HB model.
However, a test reported four days later suggested a
considerable increase in GOR. The new GOR value of 1,250
scf/STB is consistent with breakthrough of free gas from the
reservoir. When the new GOR was used, we observed an
acceptable match as Fig. 6 suggests. In Fig. 6, A/G implies
model results of Ansari et al. and Gomez et al., which are
indistinguishable traces for the scale used. Both dispersed
bubbly, owing to high-liquid velocity, and slug flow

SPE 109868

dominated the results in this increased GOR scenario. We


surmise that misreporting of GOR triggered the earlier
mismatch, because failure of all models is highly unlikely.
No attempt was made to seek corrections of this nature in
the database. This example merely highlights uncertainty of
input parameters in routine calculations even in wellinstrumented environment.
Pressure, psig
1,000

2,000

3,000

4,000

5,000

6,000

Table 4 Uncertainty variables for Plackett-Burman design.


Oil Rate
Watercut
GOR
Oil
rror
Gravity
qo
fw
Rgo
p
STB/D
scf/STB
o

psi

0
Well Depth, ft

We note that GOR is the largest contributor to error, followed


by the oil rate itself. Because gas is flared in many business
settings, the measurement accuracy suffers; the same comment
holds for watercut measurements. Of course, the impact of
uncertainty of each independent variable on the dependent
variable (error) is case specific; this example is intended for
illustrative purpose only.

21,764

14.45

732

36

-143

5,000

21,764

19.6

732

32

-231

10,000

21,764

19.6

990

32

-28

17,807

19.6

990

36

145

15,000

21,764

14.45

990

36

41

20,000

17,807

19.6

732

36

-88

17,807

14.45

990

32

157

17,807

14.45

732

32

-68

19,786

17

861

34.5

25,000
30,000
35,000

Fig. 5 Higher pwf calculation of all models suggest suspect data.

Well Depth, ft

2,500

3,500

qo 4

Pressure, psig
1,500

R go 5

4,500

5,500

fw 3

5,000

95% Confidence
interval

10,000
15,000

A/G

20,000

HB

25,000

PE-2
OLGA

30,000

Curvature 1

10

15

20

Effects Estimate (Absolute Value)

HK

35,000

Fig. 7 Pareto chart shows dominance of GOR and oil rate on pwf
error estimation.

Fig. 6 Good performance of all models after GOR correction.

Let us explore how measurement errors may potentially


cause significant uncertainty in pwf estimation. Consider a well
exhibiting all four flow regimes from this dataset. Table 4
presents the uncertainty values, which are assigned relative to
those presented in the last row displaying the true values. For
instance, we assumed the oil-rate measurement uncertainty to
be 10%, GOR to be 15%, and so on.
Fig. 7 displays the Pareto chart summarizing the results of
the four-variable Placket-Burman (1943) design of
experiments. The positive sign associated with each
independent variable suggests that any increase is reflected as
an increase in the dependant variable, while the opposite is
true for those associated with negative values. Note that
nonlinearity in this comparative picture is small as signified by
the small curvature value. Here, the dependent variable is
error, defined as the difference in pressure with respect to the
true value, corresponding to the input values of the last row.

Discussion
This study consolidates two-phase flow models that were
presented previously, based on the drift-flux approach. This
unified approach applies to a variety of well orientation,
geometry, and fluids with minimal changes in parameter
values. This simplified approach allows for ease of
implementation, understanding of mechanics of flow, and
above all, transparency.
When comparison of this model with others was sought,
we learned that their collective performance is comparable in a
statistical sense. Perhaps this outcome owes largely to the
high-quality dataset that we used. Even in this set, data were
far from being perfect; Fig. 5 illustrated this point. Our view is
that improvement in two-phase flow modeling has probably
reached a point of diminishing returns. Focusing on data
gathering and quality control will perhaps pay much more
dividends than improving the models. To that end, the current

SPE 109868

practice of production rate validation through real-time


integrated asset modeling is a step in the right direction.
Uncertainty in input data stems from allocated oil rates,
and low-quality measurements of both the gas and water
phases. Changes in oil properties also occur because of
changes in reservoir pressure and compositional grading,
particularly in thick oil columns. Therefore, periodic
calibration of oil properties with a PVT model is definitely in
order whenever pressure-traverse computations are sought.
We also note that the common practice entails the use of a
single overall-heat-transfer coefficient in pressure-drop
calculations. Although this approach often produces
satisfactory pwf estimation, the resultant temperature profile
may not necessarily reflect the reality, especially in deepwater
and high-pressure/high-temperature environments. In other
words, good accuracy in pwf calculations does not guarantee
the same in the temperature profile, as discussed elsewhere
(Hasan et al. 2007). The implication is that the simplifiedheat-transfer modeling may provide a temperature trace that is
unworthy of use for combating production of thermodynamic
solids, such as hydrates, paraffins, and asphaltenes. Solids
buildup or corrosion may also significantly contribute to the
uncertainty of pipe ID, thereby impacting results.
Conclusions
1. This study presents a simplified two-phase flow model
encompassing various types of fluid (oil/gas, oil/water)
and flow direction (cocurrent, countercurrent, and
downflow).
2. Staistical analysis show that all the six models used in this
study perform similarly. Models based on physical
principles appear to offer no improved accuracy than the
widely used Hagedorn-Brown and PE-2 correlations.
3. Input data accuracy is the key to a models performance.
Uncertainty stemming from rate measurements and PVT
parameters has significant influence on the performance
of one or a group of two-phase flow models.
Acknowledgment
We are grateful to our Chevron colleague Maria Portillo for
her timely contributions to make large number of
computations possible. Cooperation of many individuals
contributing to the database is gratefully acknowledged. We
thank Chevron management for permission to publish this
work.
Nomenclature
Ax= cross-sectional area for flow, ft2
Co= flow parameter, dimensionless
Cob= flow parameter for fully developed bubbly flow,
dimensionless
Cos= flow parameter for fully developed slug flow,
dimensionless
d = tubing ID, ft
fm = Moody friction factor, dimensionless
fg = in-situ gas volume fraction, dimensionless
fL = liquid holdup, dimensionless
fo = in-situ oil volume fraction, dimensionless
Fa = annulus factor, given by Eq. A-11, dimensionless

F = well-deviation factor, given by Eq. A-12,


dimensionless
g = gravitational acceleration, ft/sec2
gc = conversion factor, 32.17 (lbm-ft)/lbf/sec2
n = numer of data points, dimensionless
p = pressure, psig
pwf = flowing bottomhole pressure, psig
dp/dz = total pressure gradient, psi/ft
(dp/dz)A= accelerational pressure gradient, psi/ft
(dp/dz)F= frictional pressure gradient, psi/ft
(dp/dz)H=hydrostatic pressure gradient, psi/ft
r = correlation coefficient, dimensionless
rxy = correlation coefficient between data and model 1,
dimensionless
rxz = correlation coefficient between model 1 and model
2, dimensionless
rzy = correlation coefficient between data and model 2,
dimensionless
Rem= mixture Reynolds number, dimensionless
t = statistical measure for model comparison,
dimensionless
T = absolute fluid temperature, oR.
vgb = superficial gas velocity needed for transition from
bubbly to slug flow, ft/sec
vgc = superficial gas velocity needed for transition from
churn to annular flow, ft/sec
vg = in-situ velocity of gas, ft/sec
vot, = superficial oil velocity needed for transition to oildominated flow, ft/sec
vsL = superficial velocity of liquid, ft/sec
vsg = superficial velocity of gas, ft/sec
vm = velocity of gas/liquid mixture, ft/sec
vb = small bubble rise velocity, ft/sec
vT = Taylor bubble rise velocity, ft/sec
dv/dz = velocity gradient, ft/sec
x = mass fraction [= vsgg/(vsgg + vsLL)]
z = vertical well depth, ft
= parameter
for
friction-factor
calculation,
dimensionless
= pipe-roughness factor, ft
g = gas density, lbm/ft3
L = liquid density, lbm/ft3
m = mixture density, lbm/ft3
L = liquid viscosity, cp
m = mixture viscosity, cp
g = gas viscosity, cp
= surface tension, lbm/sec2
= well inclination with horizontal, degrees
References
Ansari, A.M., Sylvester, N.D., Sarica, C., Shoham, O., and Brill, J.P.
1994. A Comprehensive Mechanistic Model for Upward TwoPhase Flow in Wellbores. SPEPF 9 (2): 143151.
Aggour, M.A., Al-Yousef, H.Y., and Al-Muraikhi, A.J. 1996.
Vertical Multiphase Flow Correlations for High Production
Rates and Large Tubulars. SPEPF 11 (1): 41-48.
Bendiksen, K.H., Maines, D., and Moe, R. 1991. The Dynamic TwoFluid Model OLGA: Theory and Application. SPEPE 6 (3):
171180.
Blalock, H. M., Jr. 1972. Social Statistics. 2nd Edition, McGraw-Hill
Book Co., New York, NY, 406-407
Caetano, E.F., Shoham, O., and Brill, J.P. 1992. Upward Vertical
Two-Phase Flow Through an Annulus, Part II: Modeling

SPE 109868

Bubble, Slug, and Annular Flow. J. Energy Resources Tech. 114


(1): 1430.
Chen, N.H. 1979. An Explicit Equation for Friction Factor in Pipe.
Ind. Eng. Chem. Fundamentals 18 (3): 296.
Gomez, L.E., Shoham, O., Schmidt, Z., Chokshi, R.N., and Northug,
T. 2000. Unified Mechanistic Model for Steady-State TwoPhase Flow: Horizontal to Vertical Upward Flow. SPEJ 5 (3):
339350.
Hasan, A. R. 1995. Void Fraction in Bubbly and Slug Flow in
Downward Two-Phase Flow in Vertical and Inclined Wellbores.
SPEPF 10 (3): 172176..
Hasan, A.R. and Kabir, C.S. 2002. Fluid Flow and Heat Transfer in
Wellbores. SPE, Richardson, Texas.
Hasan, A.R. and Kabir, C.S. 2007. A Simple Model for Annular
Two-Phase Flow in Wellbores. SPEPO 22 (2): 168175.
Hasan, A.R., Kabir, C.S., and Wang, X. 2007. A Robust Steady-State
Model for Flowing-Fluid Temperature in Complex Wells. Paper
SPE 109765 presented at the SPE Annual Technical Conference
and Exhibition, Anaheim, California, 1114 November.
Kabir, C.S. and Hasan, A.R. 2006. Simplified Wellbore Flow
Modeling in Gas/Condensate Systems. SPEPO 21 (1): 8997.
Kabir, C.S. and Hasan, A.R. 1990. Performance of a Two-Phase
Gas/Liquid Flow Model in Vertical Wells. J. Pet. Sci. &
Engineering 4 (3): 273289.
Kaya, A.S., Sarica, C., and Brill J.P. 2001. Mechanistic Modeling of
Two-Phase Flow in Deviated Wells. SPEPF 16 (3): 156165.
Plackett, R.L. and Burman, J.P. 1943. The Design of Optimum
Multifactorial Experiments. Biometrika XXXIII, University
Press, Cambridge, England, 305.
Prosper Users Manual. 2006. Petroleum Experts Ltd., Edinburgh,
Scotland, UK, EH7 4GB.
Pucknell, J.K., Mason, J.N.E., and Vervest, E.G. 1993. An Evaluation
of Recent Mechanistic Models of Multiphase Flow for
Predicting Pressure Drops in Oil and Gas Wells. Paper SPE
26682 presented at the Offshore European Conference,
Aberdeen, Scotland, 710 September.
Shi, H., Holmes, J.A., Durlofsky, L.J., Aziz, K., Diaz, L.R., Alkaya,
B., and Oddie, G. 2005a. Drift-Flux Modeling of Two-Phase
Flow in Wellbores. SPEJ 10 (1): 2433.
Shi, H., Holmes, J.A., Diaz, L.R., Durlofsky, L.J., Aziz, K. 2005b.
Drift-Flux Parameters for Three-Phase Steady-State Flow in
Wellbores. SPEJ 10 (2): 130137.
Shoham, O. Flow Pattern Transition and Characterization in GasLiquid Two-Phase Flow in Inclined Pipes. 1982.
PhD Disertation, U. of Tel Aviv, Tel Aviv, Israel.
Steen, D.A. and Wallis, G.B. 1964. Pressure Drop and Liquid
Entrainment in Annular-Two-Phase Flow. AEC Report NYO3114-2.
Taitel, Y., Barnea, D. and Dukler, A.E. 1980. Modeling Flow Pattern
Transition for Steady Upward Gas-Liquid Flow in Vertical
Tubes. AIChE J 33 (2): 345354.
Wallis, G.B. 1969. One-Dimensional Two-Phase Flow, McGrawHill, New York. 93.
Zuber, N. and Findlay, J. 1965. Average Volumetric Concentration in
Two-Phase Flow Systems. Trans., ASME J. Heat Transfer 87,
453.

Appendix A Two-Phase Flow Model


Total pressure gradient during fluid flow is the sum of the static,
friction, and kinetic gradients, which is given by
-

f v 2m m
dp
dv
= g m+ m
+ m vm m
dz
2d
dz

(A-1)

The mixture density, m, is the volumetric-weighted average of


the two phases, L and g

g f g + L (1 f g )

(A-2)

In Eq. A-2, fg is in-situ volume fraction of the gas phase or


lighter phase in oil/water flow. The primary difficulty in using
Eq. A-1 to estimate pressure gradient during two-phase flow
arises from the fact that volume fraction is often not equal to
the ratio of the superficial gas velocity to the mixture velocity,
that is, fg vsg/vm. In addition to the need to estimate fg, a twophase flow model also requires estimating two-phase friction
factor, fm.
We use the homogeneous model to estimate frictional
pressure gradient. The friction factor in all flow regimes is
estimated from mixture Reynolds number, Rem = Dvm m/m,
using a mass-average mixture viscosity

g x + l (1 x)

(A-3)

and the Chen correlation (1979) for friction factor in rough


pipes
fm =

/ d 5.0452

log
4 log

3.7065 Re m

(A-4)

In Eq. A-4, is pipe roughness, and the dimensionless


parameter is given by

( / d )1.1098
2.8257

7.149

Re m

0 .8981

(A-5)

To estimate volume fraction needed to calculate mixture


density, we individually model the behavior of each flow
regime. However, for all flow regimes the gas (or lighter)
phase moves faster than the liquid (or heavier) because of its
buoyancy and its tendency to flow close to the channel center,
where the velocity is higher than the average mixture velocity.
Therefore, the in-situ gas velocity, vg, can be expressed as the
sum of the bubble-rise velocity and Co times the average
mixture velocity
vg

Co vm v

(A-6)

For cocurrent upflow, the mixture velocity, vm, is sum of


the gas and liquid superficial velocities, vm = vsg + vsL.
However, during countercurrent flow, when the liquid flows
downward as gas flows up, the appropriate mixture velocity is
given by, vm = vsg vsL. Similarly, when both the phases flow
downward, buoyancy acts in the direction opposite to the flow
and the rise velocity, v, is subtracted from Covm to obtain the
gas in-situ velocity, vg.

SPE 109868

Noting that in-situ velocity is the ratio of superficial


velocity vsg to volume fraction fg (vg vsg/ fg), we have a
simple relation between volume fraction and phase velocities
in these flow regimes
=

fg

v sg

(A-7)

C o v m v

Eq. A-7 is the same as Eq. 1 in the text.

Parameters. The values of the flow parameter, Co, and the rise
velocity, v, depend upon the flow pattern, well deviation, flow
direction (upward/downward; co- or counter-current), and phases
(gas-liquid, liquid-liquid). In the following, we present the model
parameters that we have either developed or adopted for various
flow regimes. For brevity and clarity, we have not discussed
alternative values/expressions for some of these parameters that
others may have found more acceptable. However, much of that
discussion appears in the text of Hasan and Kabir (2002).
The rise velocity for small bubbles in all flow directions
and well deviations is best represented by the Harmathy
equation, which is same as Eq. 2 in the text
v b

[(

= 1.53 g l g / l2

1/ 4

(A-8)

The rise-velocity of Taylor bubbles, characteristics of slug


flow, is influenced by well deviation and annular geometry.
We estimate the Taylor bubble-rise velocity from the
following expression

v T = 0.35 gD l g / L (F ) (Fa )

(A-9)

In Eq. A-9, the well-deviation factor F, in terms of deviation


angle from vertical , is given by,
F

cos (1 + sin )1.2

(A-10)

and the annulus factor Fa, in terms of the annulus inside/outside


diameter ratio di/do (Caetano et al. 1992; Hasan and Kabir 2002,
pp 55-56), is given by
Fa

0.29d i
1 +

do

(A-11)

For oil/water flow, the expression for F is slightly differeent


F

cos (1 + sin )2.0

In slug flow, liquid slugs that separate the Taylor bubbles


also contain small bubbles whose rise velocities are best
estimated with Eq. A-8. Therefore, for slug flow we need a
rise velocity that is an average of bubble-rise velocities,
represented by Eqs. A-8 and A-9. We use the following
averaging process that depends on the magnitude of the
superficial-gas velocity compared to that needed for transition
from bubbly to slug flow, vgb

(A-12)

Hasan (1995), analyzing data from several sources, showed that


Eqs. A-8 through A-11 are applicable even for downward twophase flow. Along with Eq. A-7 (Eq. 1 in the text), these
equations allow accurate estimateion of volume fraction in
bubbly and slug flows prevalent in downward flow.

= vb 1 e

+ vT e

0.1v gb /( v sg v gb )

0.1v gb /( v sg v gb )

(A-13)

We use Eq. A-13 for rise velocity in churn flow as well. Note
that Eq. A-13 is completely empirical in nature. Many
researchers have shown that at the high velocity associated with
annular flow, the effects of well orientation, geometry, and flow
direction are negligible. The parameter values for annular flow in
Table 1 reflect that fact.
The values of Co depend on the flow pattern, well
deviation, and flow direction. Table 1 in the text presents the
values of Co for fully developed flow patterns. Although
transition from one flow regime to another could be sharp,
they are rarely abrupt. The actual values of Co used with
Eq. A-7 are modified along the line of Eq. A-13 to account for
smooth transition from one flow regime to another.
Pattern Transition Criteria. The use of our unified approach
with separate values of parameters for estimating volume
fraction requires establishing the existing flow pattern.
Transition from bubbly to slug flow occurs at the volumefraction must exceed a value of 0.25 in vertical systems.
Experimental data of Zuber and Findlay (1965), Hasan and
Kabir (2002, p. 21), among others, support this contention. For
inclined systems, buoyancy concentrates the gas phase near the
upper wall, causing local volume fraction to exceed the
needed value of 0.25 before the average volume fraction,
causing an earlier transition from bubbly flow. Using a
geometrical approach (Hasan and Kabir 2002, p. 42) were able
to use the same criterion (fg = 0.25) for transition from bubbly
flow for all inclined systems by replacing vsg by vsg/cos. Eq.
A-7 can then be used, with vgb for the superficial gas velocity
needed for transition from bubbly flow, to obtain
v gb

C o v sL v
cos
4C o

(A-14)

Therefore, for cocurrent flow in both upward and downward


inclined systems for which Co = 1.2, one obtains
vgb

(0.43vsL 0.36v )cos

(A-15)

The absolute value of deviation must be used for


downflow. Because the negative sign applies to downflow, at
some flow conditions given by 0.43vsL 0.36 v, the
transitional superficial-gas velocity vgb becomes negative,
indicating an unstable system.

SPE 109868

For countercurrent flow, as shown by Hasan and Kabir


(2002, pp. 57-58), Co = 2.0 and
vgb

(0.5v vsL )cos

(A-16)

However, when v > vT, as would be the case for smalldiameter channels, bubbly flow cannot exist. Also, when
mixture velocity is higher than vms, given by the following
expression (Taitel et al. 1980; Shoham 1982):
f
2v 1ms.2

2d
=

0.4

0.725+ 4.15

0.6

0.4
g ( L g )

vm

(A-18)

Transition from slug to churn flow occurs when vm exceeds


vms (given by Eq. A-17) and vsg exceeds that given by Eq. A18. Transition from churn to annular flow occurs when vsg
exceeds the value given by the following expression (Taitel et
al. 1980)

[ (

) ]1 / 4

3.1 g L g / g2

(A-19)

The presence of g in the denominator of Eq. A-19


suggests that at very high pressures, annular flow can exist at
low-gas velocities. However, a little reflection about the
characteristic of this flow regimea gas core surrounded by a
thin liquid filmsuggests that enough gas must remain in the
core for the liquid not to be able to bridge the channel crosssection. To allow for this reality, we arbitarily use a minimum
fg of 0.7 as an additional requirement for the existence of
annular flow.
Transition smoothing. For Co, we use the values reported in
Table 1 for bubbly flow. For example, for cocurrent bubbly flow
in either upward or downward direction, Co = 1.2; Co = 2.0 for
countercurrent bubbly flow. For all other flow patterns we use an
exponential weighted-average value of Co derived from the fully
developed Co values of the adjoining flow regimes. Therefore,
for slug flow
Co

= C ob 1 e

0.1vbs /( vsg vbs )

) + C (e
os

(A-21)

Similarly, for upward co- and counter-current annular flow, we


have
= 1.15 1 e

0.1v gc /( vsg vc )

+ 1.0 e

0.1v gc /( vsg vc )

(A-22)

(A-17)

v sg

v sg >1.08v sL

0.1vms /( vm vms )

For downward flow, the value of Co is 1.12 for churn flow.

Co

Dispersed bubbly flow persists until the gas velocity is high


enough for transition to churn flow occurs. Transition to churn
flow from dispersed bubbly flow occurs when

v gc

(
+ 1.15 (e

= 1.2 1 e 0.1vms /(vm vms )

Co

0.1vbs /( vsg vbs )

) (A-20)

For cocurrent upflow, the parameter values are the same


for bubbly and fully developed slug flow; that is, Co equals 1.2
for cocurrent slug flow. However, the values are different for
countercurrent flow and for downward flow, and are
dependent on superficial gas velocities and the transition
velocity, vbs. Similarly, for upward co- or counter-current
churn flow, Co is given by the following expression

Oil/Water Flow. The dynamics of simultaneous flow of two


immiscible fluids, such as oil and water, is very similar to that
of gas/liquid flow. As discussed by Hasan and Kabir (2002,
pp. 58-59), Eqs. A-1 through A-7 apply to oil/water two-phase
flow. The rise-velocity of the oil bubbles in any system,
however, is given by the following expression
v o

[(

= 1.53(1 f o ) 2 g l g / l2

1/ 4

(A-23)

where
F

cos (1 + sin )2

(A-24)

Note that vo is influenced by oil-volume fraction, making the


expression for fo implicit.
The flow behavior is best described by recognizing two
flow regimes, water-dominated flow and oil-dominated flow.
When in-situ volume fraction of oil, fo, is below 0.7, flow is
water-dominated and fo is given by Eq. 7 (repeated here for
clarity)
fo

vso
Co vm vo

(A-25)

Co assumes a value of 1.2 in Eq. A-7. Above an oil fraction of


0.7, flow is oil-dominated and homogeneous, just as in the
case of gas/liquid annular flow. For fully developed cases of
oil-dominated flow, fg is given by Eq. A-7 with Co = 1.0 and
vo = 0. Assuming fo of 0.7, the superficial-oil velocity, vot, at
which transition to oil-dominated flow occurs, is given by
v ot

= 5.25v sw + 4.38v o

(A-26)

The value of Co for vso greater than that given by Eq. A-26 is
Co

= 1.2 1 e 0.1vot /(vso vot ) + 1.0 e 0.1vot /(vso vot )

(A-27)

Computational Algorithm. In spite of simplicity of the model


presented here, pressure-traverse computation can get quite
involved. The primary source of complications is in determining
the prevalent flow regime, because the diverse set of criteria
(equations) used for this task can lead to an apparently
conflicting logic. For example, at moderate and low pressures,

SPE 109868

annular flow regime can only exist at very high vsg values that
preclude existence of all other flow regimes, thereby making the
choice of annular flow regime easy. However, at very high
pressures, Eq. A-19 shows that annular flow can exist at
relatively low vsg, when slug, churn, and dispersed bubbly flows
are also possibilities. Thus, in top-down calculations, one can
possibly establish annular flow at the wellhead, and then
determine the occurrence of churn flow in the next cell because
of decrease in gas velocity. However, farther down the well, if
the pressure remains high the existing vsg could be higher than
that given by Eq. A-19, causing one to surmise that annular flow
exists at that cell. Such a situation where churn flow exists on top
of annular flow is, of course, physically unrealistic.
To avoid such anomalous situations, we adhered to the
following algorithm for assigning flow regimes to any
computational cell in the top-down calculation approach. We
assume that at the wellhead, if vsg is greater than that given by
Eq. A-19 and fg > 0.7, then annular flow exists, even if various
criteria allow other flow regimes to also exist. We follow this
logic down the wellbore untill Eq. A-19 (and/or fg > 0.7) is no
longer satisfied, at which point annular flow can no longer
exist anywhere farther down the wellbore. We use this same
approach of flow regime hierarchyannular, churn, slug, and
bubblyfor the entire wellbore.

You might also like