SPE 89754

Simplified Wellbore Flow Modeling in Gas-Condensate Systems

C. S. Kabir, SPE, ChevronTexaco Overseas Petroleum; A. R. Hasan, SPE, U. of Minnesota-Duluth

Copyright 2004, Society of Petroleum Engineers Inc.

remains unclear. Take the upper limits of condensate/gas
This paper was prepared for presentation at the SPE Annual Technical Conference and (CGR) ratio of 50 STB/MMscf or flow-string diameters of 3.5
Exhibition held in Houston, Texas, U.S.A., 26–29 September 2004.
in., for instance. Questions arise whether one should use a
This paper was selected for presentation by an SPE Program Committee following review of
information contained in a proposal submitted by the author(s). Contents of the paper, as
different model when one of these criteria, among others, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to set by Gray is not met.
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 Boundaries of applicability often get violated beyond a
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
correlation’s original intent; Gray’s correlation is no exception
for commercial purposes without the written consent of the Society of Petroleum Engineers is in this regard. Practicality demands that a user specifies one
prohibited. Permission to reproduce in print is restricted to a proposal of not more than 300
words; illustrations may not be copied. The proposal must contain conspicuous computational approach for flow in pipes when long-term
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O. integrated reservoir/wellbore/flowline performance is sought
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
over a field’s producing life. Declining CGR and increasing
water production with time have the potential for complicating
Abstract
Predicting long-term reservoir performance with realistic any modeling effort. What also remains unclear is how to treat
wellbore models is fraught with uncertainty owing to the the multicomponent fluid mixture that enters the
complexity of two-phase flow. That is because even a wellbore/flowline system, after undergoing compositional
calibrated two-phase flow model departs from its expected calculations in the reservoir.
performance trend when changes in flow conditions occur. Besides the two-component gas/liquid Gray correlation,3
These inevitable changes include gas/liquid ratio, wellhead other approaches have emerged for modeling gas-condensate
pressure, and flowline pressure with time, among others. flow. The semimechanistic model of Govier and Fogarasi4
Influx of water further exacerbates the prediction problem. represents the multicomponent approach with flash
This study explores the possibility of using simplified calculations. In contrast, the wet-gas concept offered by
approaches to computing bottomhole pressure (BHP) from Peffer et al.5 suggests extreme pseudoization with single-
wellhead pressure (WHP), measured rates, gravity of component gas. Nonetheless, the simplified approach of Peffer
producing fluids, and tubular dimensions. BHP computations et al. with good accuracy is appealing. A minor drawback of
on three independent data sets comprising 167 gas-condensate both methods is the neglect of accelerational term, which may
well tests suggest that the no-slip homogeneous model applies be significant in wells producing fluids at high gas/liquid ratio
quite well. Statistical results show that the homogeneous (GLR).
model compares quite favorably with mechanistic two-phase This paper advocates the use of two-component
flow models. However, the main advantage of the simplified homogeneous model to circumvent issues with any rigorous
model is that its recalibration with field data is not required two-phase flow modeling, such as delineating flow-pattern
because the gas/oil ratio increases with time, thereby making boundaries, estimating slip between phases, and doing flash
the model more robust. calculations. We show that Gray’s correlation is essentially a
Most field datasets suggest random error in BHP homogeneous model, and that of Ansari et al.7 also simplify to
calculations; uncertainty in rate measurements appears to be homogeneous model when mist flow occurs in gas-condensate
the most probable cause. High-GLR systems can tolerate large wells. The steady-state version of the transient simulator
errors in rate measurements, but low-GLR wells demand OLGA8 also lends support to the notion of homogeneous
greater accuracy. modeling.

Introduction Computational Approach and Results

Two-phase flow modeling for gas-condensate wells has not We used data from the literature4,5 and those from a few West
received as much attention as that for oil wells. Recent SPE African fields with medium to high-CGR production. In all,
books1,2 on this topic make very little mention of this flow we examined 167 independent tests. The methods used
condition, presumably because modeling is supposed to involved those of Gray, Aziz et al., homogeneous, Ansari et
conform to that offered for oil wells. This study probes this al., steady-state OLGA, and the wet-gas approach, advanced
premise, among other issues. by Peffer et al. In two datasets, we used the wet-gas concept
The popular Gray correlation3 appears to do a good job in but including the accelerational term. In other words, the
most gas-condensate wells. However, applicability of this accelerational term, besides friction and hydrostatic, is implicit
correlation outside the bounds of its specified parameters in all methods reported in this study.
2 SPE 90754

Example Applications 5
We considered three datasets in this study. The first set is
obtained from West Africa and the other two include those
4
reported by Govier and Fogarasi4 and by Peffer et al.5

West Africa Dataset. Some 25 datasets are available from 3

% Error
drillstem tests (DSTs). Typically, three flow tests were
conducted during each DST. The CGR ratio varied between 2
50 and 137 STB/MMscf, whereas pressure ranged from 5,100
and 7,300 psig. The overall statistical results for the wellbore 1
pressure-drop (pwf – pwh) calculations are presented in Table 1.
All the methods show comparable results, with Aziz et al. 0
method outperforming others. Note the striking similarity

63

66

87

86

00

28

27

71
between the homogeneous model and the Ansari model. We

,3

,6

,0

,6

,9

,3

,3

,8
11

14

16

17

17

19

25

35
show later that both the Gray and Ansari methods behave
close to the homogeneous model during mist flow, GOR, scf/STB
encountered in a typical gas-condensate well. Now, let us Fig. 1 – Random error is indicated for the subset of West Africa
explore some issues related to measurement uncertainty as dataset, North Apoi field.
they relate to pressure-traverse computation in each well,
intercepting multiple reservoirs. These tests were conducted
5,800
each reservoir at a time, however.
5,600

Calculated pwf, psig

Table 1 – Error Analysis: West Africa Dataset
Parameter Gray Aziz Homo Ansari OLGA Wet
Gas 5,400
Av.
% -6.46 0.29 -5.93 -4.92 -4.54 -4.43 5,200
Error
Abs. Av.
% 6.89 9.15 7.44 7.54 6.21 8.30 5,000
Error p wfc = 0.9766p wfm
% Stand
4,800
Deviation 20.91 0.14 17.63 12.18 10.36 34.52 R2 = 0.9688
Abs. Av. 4,600
Error, psi -100.54 24.2 -71.74 -53.38 -56.24 -38.84
4,600 4,800 5,000 5,200 5,400 5,600 5,800
Av. Error, Measured p wf , psig
psi 109.51 161.1 127.46 129.88 101.46 152.03

Fig. 2 – Comparison of measured and computed data with

homogeneous model, N. Apoi field.
Fig. 1 shows that random error, determined by the
difference in the measured and computed BHPs for the wet-
gas method, is apparent. The three-fold change in GOR values Only five tests were available from the F and H sands in
is not expected to occur in these reservoirs, given the the Sonam field. The lower pressures reflect the shallower
similarity in PVT properties. Laboratory-measured PVT data F-sand tests and the higher pressures are those from the deeper
and EOS modeling show only a minor increase in liquid H-sand. Errors larger than those observed previously are
content with increasing depth. In other words, we have apparent in Fig. 3. Let us explore this point.
uncertainty in rate measurements, which is probed further in a Any errors in liquid-rate measurement can translate into
later section. large error in BHP computation, because Rg controls the wet-
Intuitively, we expected the BHP error to decrease with gas gravity. Potentially, liquid measurements are subject to
increasing GOR because the assumption of single-phase flow larger errors compared to gas owing to small volumes
gets robust. Fig. 2 compares the BHPs directly. In either case, measured in a typical gas-condensate system. For instance, if
good agreement exists, with maximum error being about 4%. we arbitrarily assumed 15% error in GOR measurement,
significant reduction in the estimated BHP calculations occurs,
as Fig. 4 testifies. As expected, no changes occur for high-
GOR tests because we are dealing with light-fluid columns to
begin with. But, at low-GORs large improvement occurs with
the column being less dense than before. The implication is
that we probably have some uncertainty in the measurement of
gas, liquids, or both in the low-GOR tests. Put it another way,
low-GOR (< 8,000 scf/STB) tests demand greater precision in
rate measurements than their high-GOR counterparts.
SPE 90754 3

For the Govier-Fogarasi dataset, Fig. 5 exhibits the expected

trend; that is, the BHP error tends to decline with increasing
7,000
GOR. The solid line indicates the running average. As
discussed earlier, increasing GOR will inevitably lead toward
Calculated BHP, psig

the single-phase situation, thereby driving down the error.

6,000 Fig. 6 shows the goodness of the quality of computed results.

20
18
5,000 p wfc = 1.0563p wfm
R2 = 0.9663 16
14
12

% Error
4,000
10
4,000 5,000 6,000 7,000 8
Measured BHP, psig 6
Fig. 3 – Comparison of measured and computed data with the 4
wet-gas method, Sonam field. 2
12 0

12 0
00

14 0
16 0

18 0
24 0

26 0

29 0
34 0
00
0
0

11 0
10

0
0

0
0

0
90
10

82
,5
,5

,1

,5
,6

,5
,1

,9

,8
,6
GOR-15%

3,
6,

8,

13
8 increase GOR, scf/STB
GOR Fig. 5 – Error decreases with increasing GOR, Govier-Fogarasi
6 3
% Error

data.

2
4,500
0
6,630 7,102 8,008 13,667 15,001
Calculated BHP, psig

-2 3,500
-4
GOR, scf/STB 2,500
Fig. 4 – Error decreases with increasing GOR, Sonam field.
p wfc = 1.006p wfm
Govier-Fogarasi Dataset.5 We used 96 tests out of the 102 1,500 R2 = 0.9658
reported by Govier and Fogarasi. As discussed later, issues
with field data reporting, particularly with regard to
condensate gravity, presented some computational challenges. 500
Therefore, some discretion was used to leave out a few tests 500 1,500 2,500 3,500 4,500
from our analysis. Table 2 presents the overall results for
pressure-drop, showing comparable performance amongst Measured BHP, psig
methods used. Fig. 6 – Comparison of measured and computed BHP, Govier-
Fogarasi data.
Table 2 – Error Analysis: Govier-Fogarasi Dataset
Parameter Gray Aziz Homo Ansari OLGA Texas Railroad Commission (TRRC) Dataset.4 This data set
lends itself to an in-depth study because of large variations in
Av. GOR (5,180 to 780,000 scf/STB), WHP (3,107 to 12,313
-0.091 20.83 -3.13 -0.013 9.31
%Error psig), and BHP (4,125 to 11,250 psig). When we searched for
Abs. Av. a trend by graphing the percent error in BHP with GOR,
%Error 23.46 62.70 24.96 24.00 31.33
WHP, and rate, none emerged. Figs. 7 through 9 show that
% Stand. random errors abound. Perhaps uncertainty in rate
Deviation 0.0093 85.96 1.97 0.01 17.27
measurements is responsible for a lack of trend. Nonetheless,
Abs. Av. the overall error is rather small as testified by Fig. 10.
Error, psi -3.89 71.53 -4.14 11.93 65.44

Av. Error,
140 200 145 142 162
psi
4 SPE 90754

30 was rarely checked before the conduct of a test. Given this

reality, close agreement between a model and measurement
20 may be hard to achieve when old datasets are examined.

10 Table 3 – Error Analysis: TRRC Dataset

% Error

Parameter Gray Aziz Homo Ansari OLGA Wet

Gas
0 Av.
% 3.92 26.03 -0.228 1.37 4.25 5.60
0

13 1

14 1

15 8

21 1

31 6

38 9

57 4
18 27

0
Error
09

99

92

00
-10
,0

,3

,8

,5

,6

,6

,4
6,

8,

9,

0,
Abs. Av.
% 13.89 31.98 12.16 12.82 13.47 17.33
Error
-20
% Stand
Deviation 7.60 331.88 0.094 0.99 8.90 15.73
-30
Abs. Av.
GOR, scf/STB Error, psi 254.61 500.07 236.44 241.25 249.43 99.91

Fig. 7 – Error bears no correlation with GOR, TRRC data. Av. Error,
psi 27.72 366.92 47.13 18.13 42.10 356.3

30

20 15,000

Calculated BHP, psig

10
% Error

12,000
0
7

-10
10

40

32

52

96

55

19

51

15

45

21

47

9,000
3,

3,

4,

4,

4,

5,

6,

6,

7,

7,

8,

8,

-20
p wfc = 1.0173p wfm
6,000
-30 R2 = 0.9562

WHP, psig
3,000
Fig. 8 – Error bears no correlation with WHP, TRRC data. 3,000 6,000 9,000 12,000 15,000
Measured BHP, psig
25
Fig. 10 – Comparison of measured and computed BHP, TRC data.

15
Liquid rate or CGR trend/separator efficiency. Liquid
production is relatively small in a typical gas-condensate well.
% Error

5 Consequently, separation of this liquid from large volumes of

produced gas may not be very efficient. Issues with collecting
-5 representative fluid samples are well-known and are discussed
5

by McCain and Alexander.10

0.

0.

1.

1.

2.

2.

3.

4.

5.

7.

-15 Let us examine this issue for wells producing moderate

amount of condensate during a transient test. Fig. 11 shows
-25 that liquid rate measurement and/or separation of phases
introduces uncertainty. That is because the gas rate remains
Gas Rate, MMscf/D quite stable at each choke setting, but the liquid rate does not,
Fig. 9 – Error bears no correlation with rate, TRRC data. as signified by increasing CGR.
Note that the BHP stabilizes rapidly in concert with the
Uncertainty of Input Data for BHP Calculations gas-rate behavior while the WHP does not. Fig. 12 makes this
Absolute accuracy of two gauges. Any pressure-traverse point amply clear. The increasing WHP trend is expected to
computation presupposes that the wellhead gauge has the occur owing to thermal storage11 and is not an anomaly. We
same characteristics as the one placed downhole, especially think that inadequate separation and/or shearing of liquid at
with regard to absolute accuracy. In practice, this condition is low rates owing to excessive choking cause opposing trends
rarely satisfied. Typically, the downhole gauge is of quartz when the two fluid streams are measured at the separator.
crystal variety with good accuracy and resolution. In contrast, We ensured that all the reported rates were above the
in the past, the one at the wellhead has been of the bourdon- critical unloading rate for the West Africa dataset. The use of
tube variety, with far less accuracy. Their relative performance Li et al.12 model shows that the critical rate varied between
1.77 and 2.61 MMscf/D, thereby leaving no doubt about
SPE 90754 5

continuous lift of the condensate. These tests produced only the new wells in the West Africa dataset, but wells in the other
trace amounts of water with the lowest rate being 5 MMscf/D. two datasets have unknown character.

140 5,600 4,000

5,500
CGR, STB/MMscf

110 3,800

WHP, psia
BHP, psia
5,400

80 5,300
3,600
5,200
50
5,100 3,400
20 0 2 4 6 8 10
0 7 14 21 28 35 Elapsed Time, hr
Gas Rate, MMscf/D Fig. 12 – Pressure response show stable flow during a transient
test, West Africa data.
Fig. 11 – CGR trends underscore the difficulty of obtaining quality
rate data, West Africa data.
Justification of Homogeneous Modeling Approach
90 10 The results presented clearly show that the estimates of the
simple homogeneous model are very comparable to those
80 8
CGR.STB/MMscf

obtained from more complicated models of Ansari, Gray,

qg , MMscf/D

OLGA, and Aziz et al. To explain the performance of the

70 6 homogeneous model, vis-à-vis the sophisticated models, let us
explore the significance of various terms in these models.
60 4 Comparison with Gray Correlation. The Gray correlation
uses a pseudohomogenous approach for computing frictional,
50 2 static, and accelerational head losses. However, Gray noted
that even for condensate wells, the liquid droplets may not be
40 0 moving as fast as the gas phase; that is, there is some liquid
0 2 4 6 8 10 holdup. To account for the holdup he proposed the following
Elapsed Time, hr expression for in-situ gas-volume fraction, fg
Fig. 12 – Difficulty with liquid rate measurement is apparent
during a transient test, West Africa data. ⎡ ⎛ ⎛ 205 ⎞ ⎞ B ⎤
⎢1 − e
− 2.314 ⎜⎜ N v ⎜⎜ 1+
⎝ ⎝ ND ⎠ ⎠ ⎥
⎟⎟ ⎟
⎟ ⎛ vsg ⎞
fg = ⎜ ⎟⎟ ,
⎥⎜⎝ vm
(1)
⎢
12
Condensate gravity range. Moses reported that the liquid
⎠
gravity should range from 40 to 60 oAPI. Others have ⎣⎢ ⎦⎥
suggested 70 oAPI as the upper limit. But, the reported values
from the field may not adhere to this range. For instance, of
where
the 102 tests, Govier and Fogarasi reported that 56 tests had
o
API greater than 70 and five less than 40.
We surmise that fluid contamination is responsible for ⎧⎪ ⎛ 730v sL / v sg ⎞⎫⎪
B = 0.0814⎨1 − 0.0554 ln⎜1 + ⎟⎬ , (2)
this anomalous reporting. Presence of dissolved gas at
⎪⎩ ⎜ 1+ v / v ⎟⎪
conditions other than the stock tank will lighten the liquid. By ⎝ sL sg ⎠⎭
contrast, production of some liquids from the associated oil
rim will make the condensate heavier than the lower limit of and Nv and ND are dimensionless velocity and diameter
about 40 oAPI. numbers defined in the Nomenclature. Note that for the
Clearly, contamination can be a major source of density homogeneous model, fg = vsg / vm. In other words, Gray’s
error if used literally, particularly in the wet-gas approach correlation becomes identical to the homogeneous model
where densities are used explicitly. For our computations, we when the exponential term in Eq. 1 vanishes. Our
set the gravity values to the upper and lower limits when they computations on the field examples showed that the bracketed
were violated. In contrast to this dataset, the other two did not term varies between 0.004 and 0.01. Therefore, one can
have this issue. logically conclude that the Gray correlation collapses into the
Wall roughness. This item is another potential source for homogeneous model for most gas-condensate wells. Fig. 13
error. We used 0.0018-in. pipe-wall roughness for all wells illustrates this point by showing data from two wells with
reported here. In so doing, we presupposed all tubulars are high- and medium-CGRs.
made of smooth commercial steel with no provisions for Fig. 14 compares performance of the Gray correlation with
corrosion or in-line coating. This assumption is very good for that of the homogeneous model for all three datasets used in
6 SPE 90754

this study. The excellent correlation between the two methods obtained from using the ffg alone. Note that the homogeneous
suggest that the Gray correlation can be extended beyond its model uses average mixture property values of the gas and
original bounds and also sheds light on its popularity. liquid phases to calculate a Reynolds number, which is lower
Exponential Term in Gray Equation than dvsgρg/µ.. Therefore, the calculated friction factor for the
homogeneous model is very similar to that of the Ansari
0.001 0.006 0.011 0.016
model. In fact, the difference in the calculated pwf from the two
0 methods rarely exceeded 1 psi. Fig. 15 presents computational
2,000 results of δ for two wells producing different CGRs.
Well Depth, ft

GF-32 GF-61
4,000 164 STB/MMscf Dimensionless Film Thickness
78 STB/MMscf
6,000 1.E-06 1.E-05 1.E-04 1.E-03
8,000 0
2000 GF-61
10,000

Well Depth, ft
4000 GF-32 164 STB/MMscf
12,000 78 STB/MMscf
6000
Fig. 13 – Minimal departure of Gray correlation from the 8000
homogeneous model.
10000
12000
15,000
14000
pwf by homogeneous

12,000 p wfH = 0.991p wfG Fig. 15 – L-M model shows negligible film thickness in high- and
R2 = 0.998 medium-CGR wells.
model, psig

9,000

6,000 15,000
pwf by homogeneous

∆ Govier-Fogarasi p wfH = 0.996 p wfA

12,000
3,000 O West Africa
R2 = 0.999
model, psig

+ Texas Railroad 9,000

0
0 3,000 6,000 9,000 12,000 15,000 6,000 ∆ Govier-Fogarasi
3,000 O West Africa
p wf by Gray model, psig
+ Texas Railroad
Fig. 14 – Homogeneous model correlates well with Gray model. 0
0 5,000 10,000 15,000
Comparison with Ansari et al. model. In annular flow, the p wf by Ansari model, psig
only difference between the Ansari8 and the homogeneous
models is in the way friction factor is computed. In the Ansari Fig. 16 – Homogeneous model correlates well with Ansari model.
model, friction factor for the gas phase, ffg, flowing alone in
the well is initially computed from the usual Reynolds number The near-perfect agreement between the Ansari and the
(dvsgρg/µg)–friction factor relationship. Thereafter, following homogeneous model, shown in Fig. 16, establishes their
Wallis’ suggestion, this friction factor is multiplied by a factor equivalence in mist flow. This agreement owes largely to the
that accounts for the roughness, which the gas phase fact that the liquid-film thickness is negligible even for rich
experiences in flowing against a wavy film of liquid along the condensate wells. That a part of the well may experience
wall of a pipe. This multiplier depends on δ, the dimensionless single-phase flow becomes a non-issue so long the friction
thickness of the liquid film (film thickness/well ID) and is calculations are done appropriately in a gas-flow-dominated
given by (1+300δ). In other words, system.
Comparison with Aziz et al. model. Aziz et al.7 used the
Duns and Ros correlation15 for mist flow. Duns and Ros
f fgA = f fg (1 + 300δ ) . (3)
assumed that the liquid droplets do not influence pressure
gradient at all; that is, the flow is modeled as simple gas flow.
Ansari et al. proposed a method for estimating film Thus, it uses the same friction factor, ffg, that Ansari model
thickness by equating shear forces acting on the liquid film does, except without the correction factor represented by
and the gas phase. They suggested the method of steepest Eq. 3.
descent for computing film thickness from the resulting In other words, the Aziz model should result in slightly
equation. However, we chose the simpler Lockhart-Martinelli lower estimates of pressure loss in condensate wells compared
approach14 to estimate film thickness, because it is expected to to the Ansari model. We found this lower pressure loss
be very small for most condensate wells. A sample calculation estimates by the Aziz model for most of our field data. For
is shown in Appendix B. some data, however, Aziz model predicted churn flow regime
Our computations show that δ is always less than 0.001 for and, therefore, calculated a higher static head and a greater
all the field examples. Nonetheless, the Ansari et al. method pressure loss than the other models. Fig. 17 compares the
computes a friction factor that is slightly higher than that
SPE 90754 7

performance of Aziz model with that of homogeneous model.

Although a good correlation exists, the quality of agreement is 12,000 p wfH = 0.9952p wfm
somewhat less than that indicated by Figs. 14 and 16.

Computed pwf, psig

R2 = 0.9936
9,000
15,000
p wfH = 0.959 p wfAz
12,000 6,000
R2 = 0.983
pwf by homogeneous

9,000 ∆ Govier-Fogarasi
model, psig

3,000
O West Africa
6,000 + Texas Railroad
∆ Govier-Fogarasi 0
3,000 O West Africa 0 3,000 6,000 9,000 12,000
+ Texas Railroad
0 Measured p wf , psig
0 3,000 6,000 9,000 12,000 15,000 Fig. 18 – Overall performance of the homogeneous model.
p wf by Aziz model, psig
dv m /dz , 1/sec
Fig. 17 – Homogeneous model correlates with Aziz model. 0 0.002 0.004 0.006 0.008
0
Discussion
The main motivation for this study was to establish a reliable
and yet simple model for computation of pressure-drop in 3,000
wellbores for gas-condensate systems. Identification of this Depth, ft
model, in turn, allows generation of reliable flow tables for 6,000
two- and three-phase flows in reservoir-performance
forecasting, doing systems or nodal analysis, and monitoring 9,000
WHP for on-time reservoir management.
Note that the condition of homogeneity improves with 12,000
time as a condensate system becomes leaner. Therefore, the
use of homogeneous model in an integrated model involving
15,000
reservoir/wellbore/pipe network is very appealing because of
its simplicity and robustness. Fig. 19 – Velocity gradient in Sonam J-02 well.
Comparison of performance of homogeneous model with
others suggests that it is a reliable tool. Indeed, both Gray and Conclusions
Ansari models converge to homogeneous model when mist 1. In gas-condensate wells, mist flow supporting no-slip
flow occurs in gas-condensate systems. Occurrence of mist, a flow condition appears pertinent in the entire
condition of homogeneity of the two phases, appears to hold wellbore. Consequently, homogeneous modeling is
for a wide range of bottomhole pressure, CGR, well depth, and robust for a wide range of well-operating conditions.
flow rate for the three independent datasets. Fig. 18 exhibits a This approach is therefore suitable for long-term
very good correlation between the homogeneous model and reservoir performance prediction when the integrated
the measured data. Despite the uncertainty in various elements reservoir/wellbore/flowline network modeling is
of field data discussed earlier, we think that this agreement is sought.
remarkable. 2. The mechanistic model of Ansari et al. and the
Motivated by the ease of calculation before the computer empirical correlation of Gray converge to
era, some authors, Cullender and Smith6 for example, had homogeneous model formulation because mist flow
chosen not to include accelerational gradient (dvm/dz) in the occurs with marginal liquid-film thickness on pipe
overall pressure-drop calculations. This assumption is often wall. These methods provide very accurate solutions.
quite reasonable in many cases. However, our experience 3. No single method provided consistent superiority
showed that neglecting this component is not prudent as over any other for all three datasets. The single-
Fig. 19 demonstrates. phase, wet-gas approach proposed by Peffer et al. is a
good alternative to the homogeneous model.
However, accurate gravity inputs are required.
4. Computation errors decrease with increasing GLR.
Experiences with low-GOR fluids, such as those in
the Sonam field, suggest that good accuracy in BHP
calculation is directly related to the reliability of
measured liquid rates. In contrast, high-GLR systems,
such as those in N. Apoi fields, can tolerate large
errors in liquid rates.
8 SPE 90754

5. Most field datasets suggest random error in BHP Nomenclature

calculations, meaning flow rates have uncertainty. B = factor used in Gray correlation, dimensionless
Only the Govier-Fogarasi dataset show the expected dvm/dz = velocity gradient, ft/sec
trend; that is, declining error with increasing GOR. ffg = In-situ gas volume fraction, dimensionless
ffg = Moody friction factor for the gas phase,
Acknowledgments dimensionless
We thank ChevronTexaco management for permission to ffgA = Moody friction factor for the gas phase in annular
publish this work. Some computational assistance was flow, dimensionless
provided by summer intern Charles Odumah. One of us (CSK) fM = Moody friction factor, dimensionless
benefited from discussions with Hycal’s Brent Thomas and g = gravitational acceleration, ft/sec2
colleague Frank Ting on fluid sampling and surface metering, gc = conversion factor, 32.17 (lbm-ft)/lbf/s2
ND = dimensionless diameter number {=g(ρL – ρg)d2/σL}
respectively.
Nv = dimensionless velocity number {=ρn2vm4/g σL
(ρL – ρg)}
References Nvg = Duns-Ros dimensionless gas velocity number
1. Brill, J.P. and Mukherjee, H.: Multiphase Flow in Wells,
{=vsg(ρL /g σL)1/4}
Society of Petroleum Engineers, Monograph Vol. 7,
Richardson, Texas (1999).
NvL = Duns-Ros dimensionless liquid velocity number
2. Hasan, A.R. and Kabir, C.S.: Fluid Flow and Heat Transfer {=vsL(ρL /g σL)1/4}
in Wellbores, Society of Petroleum Engineers, Richardson, pwf = flowing bottomhole pressure, psig
Texas (2002). pwh = flowing wellhead pressure, psig
3. User’s Manual for API 14B, SSCSV Sizing Computer pwf A = pwf for Ansari et al. model, psig
Program, second edition, API (1978) Appendix B, 38. pwf Az = pwf for Aziz et al. model, psig
4. Govier, G.W. and Fogarasi, M.: “Pressure Drop in Wells pwf c = computed pwf with a given method, psig
Producing Gas and Condensate,” Cdn. J. Pet. Tech., 14, No. pwf m = measured pwf , psig
44 (October-December 1975) 28. pwf H = pwf for homogeneous model, psig
5. Peffer, J.W., Miller, M.A., and Hill, A.D.: “An Improved q = gas rate, MMscf/D
Method for Calculating Bottomhole Pressures in Gas Wells vsL = velocity of liquid, ft/sec
With Liquid Present,” SPEPE (November 1988) 643. vsg = velocity of gas, ft/sec
6. Cullender, M.H. and Smith, R.V.: “Practical Solutions vm = velocity of gas/liquid mixture, ft/sec
of Gas-Flow Equation for Wells and Pipelines with z = vertical well depth, ft
Large Temperature Gradients,” Trans., AIME, 207 Z = gas-law deviation factor, dimensionless
(1956) 281. γg = gas gravity (air = 1)
7. Aziz, K., Govier, G.W., and Fogarasi, M.: “Pressure Drop in ρn = no-slip mixture density, lbm/ft3
Wells Producing Oil and Gas,” Cdn J. Pet. Tech. (July- pwfc = computed flowing BHP, psig
September 1972) 38. pwfm = measured flowing BHP, psig
8. Ansari, A.M., et al.: “A Comprehensive Mechanistic Model qg = gas rate, MMscf/D
for Upward Two-Phase Flow in Wellbores,” SPEPF (May
qo = condensate rate, STB/D
1994) 143; Trans., AIME, 297.
9. Bendiksen, K.H. et al.: “The Dynamic Two-Fluid Model Rg = gas/oil ratio, scf/STB
OLGA: Theory and Application,” SPEPE (May 1991) 171. Rw = gas/water ratio, scf/STB
10. McCain, W.D., Jr., and Alexander, R.A.: “Sampling Gas- γg = dry-gas gravity (air = 1.0)
Condensate Wells,” SPERE (August 1992) 358. γo = condensate gravity (water = 1.0)
11. Hasan, A.R. and Kabir, C.S.: “Analytic Wellbore γwg = wet-gas gravity (air = 1.0)
Temperature Model for Gas-Well Testing,” paper SPE 84288 ρg = gas density, lbm/ft3
presented at the 2003 SPE Annual Technical Conference and ρL = liquid density, lbm/ft3
Exhibition, Denver, Colorado, 5-8 October. δ = dimensionless thickness of the liquid film (=film
12. Moses, P.L.: “Engineering Applications of Phase behavior of thickness/well ID)
Crude Oil and Condensate Systems,” JPT (July 1986) 715.
13. Li, M., et al.: “New View on Continuous-Removal Liquids
From Gas Wells,” SPEPF (February 2002) 42. Appendix A ⎯ Statistical Measures and Results
14. Lockhart, R.W. and Martinelli, R.C.: “Proposed Correlation Several statistical measures are used to compare relative
of Data for Isothermal Two-Phase Two Component Flow in performance of various models for three independent data sets.
Pipes,” Chem. Eng. Prog. (1949) 39. These measures include average percent error, absolute
15. Duns, H. and Ros, N.C.J.: “Vertical Flow of Gas and Liquid average percent error, percent standard deviation, average
Mixtures in Wells,” Proc., 6th World Petroleum Congress, error, and absolute average error. Using the nomenclature used
Tokyo (1963) 451. by Ansari et al.,8 let us define these measures. Average
16. Chen, N.H.: “An Explicit Equation for Friction Factor in percent error (E1), which indicates the overall performance
Pipe,” Ind. Eng. Chem., Fundamentals (1979) 18, No. 3, 296.
trend, is given by

⎛1 n ⎞
E1 = ⎜ ∑ eri ⎟ × 100 (A-1)
⎝ n i =1 ⎠
where
SPE 90754 9

∆p c − ∆p m q wg = q g + 3,004(1.03 − γ o )qo (B-2)

eri = (A-2)
∆p m
Once, the two basic ingredients are estimated from
Absolute average percent error (E2) indicates how large the Eqs. B-1 and B-2, single-phase flow calculations can be done
errors are on the average and is given by by considering hydrostatic, frictional, and accelerational
components. Note that we do not recommend the calculation
procedure of Cullender and Smith,2 suggested by Peffer et al.,
⎛1 n ⎞
E 2 = ⎜ ∑ eri ⎟ × 100 (A-3) because it does not include accelerational pressure drop. With
⎝ n i =1 ⎠ a commercial package, one can use the Gray or any other two-
phase flow correlation, but run in single-phase mode.
Percent standard deviation (E3) shows the degree of scattering
Appendix C ⎯ Example Calculations With Various
of errors about their average value and is represented by
Models
The Gray Correlation.3 Well #61 in the Govier-Fogarasi
⎛ n (e − E ) 2 ⎞ dataset is a 11,912-ft vertical well producing 70o API
E3 = ⎜ ∑ ri 1 ⎟ (A-4) (reported value of 85o API) dry oil at 635 STB/D through a
⎜ i =1 n −1 ⎟
⎝ ⎠ 2.992-in. ID tubing. The GOR is 6,100 scf/STB, and the gas
gravity is 0.712. The following property values and computed
Average error (E4) indicates the overall trend independent of fluid parameters are available at the wellhead where pressure
the measured pressure drop, which is given by is 1,169 psig:

vsL = 1.08 ft/sec, vsg = 10.4 ft/sec

⎛1 n ⎞
E 4 = ⎜ ∑ ei ⎟ , (A-5) vm = vsL + vsg = 11.48 ft/sec
⎝ n i =1 ⎠ ρL = 44.02 lbm/ft3, ρg = 4.0 lbm/ft3, d = 0.25 ft

where µL = 0.30 cp, µg = 0.0135 cp, σ = 15.68

dynes/cm = 0.0346 lbm/sec2
ei = ∆p c − ∆p m . (A-6) Quality, x = {vsg ρg /( vsg ρg + vsL ρL)} = 0.465.
Homogeneous gas-volume Fraction = vsg / vm = 0.906 =
Absolute average error (E5) measures the magnitude of
average error and is independent of measured pressure drop. 1/(R+1)
The E5 relation is given by Homogeneous mixture density,
ρm = 0.906ρg + 0.094ρL = 7.793 lbm/ft3
⎛1 n ⎞
E 5 = ⎜ ∑ ei ⎟ (A-7) Dimensionless Numbers
⎝ n i =1 ⎠
Nv = [vm4ρm2/{(gσ)(ρL- ρg)}] = {(11.48)4 (7.793)2
Appendix B ⎯ The Wet-Gas, Single-Phase Approach
Peffer et al.5 suggested a simple approach of weighing the gas /(32.2×0.0346x40.04)} = 23,381
gravity by considering the density and ratios of the producing Nd = (d2g(ρL - ρg)/ σ ) = (0.252)(40.04×32.2)/(0.0346) =
gas and liquids. The dry-gas gravity so modified lends itself to
single-phase flow treatment. 2318.
One can calculate mixture or weight-gas gravity with the ⎛ 730 R ⎞
following expression C = ln ⎜1 + ⎟ = 4.2468
⎝ R +1 ⎠
⎛ γo 1 ⎞⎟
γ g + 4,584⎜⎜ + B = 0.0814 (1 − 0.0554C ) = 0.0622
⎝ Rg Rw ⎟⎠
γ wg = (B-1) Correction to homogeneous gas-volume fraction yields,
⎛ 1.03 − γ o 1 ⎞⎟
1 + 132,800⎜ + ⎛ ⎧ ⎛ 205 ⎞⎫ ⎞⎟
B

⎜ 44.29 R 18 R ⎟ ⎜
ξ = exp − 2.314⎨ N v ⎜1 + ⎟⎬ = 0.01293
⎝ g w ⎠ ⎜
⎝ ⎩ ⎝ Nd ⎠⎭ ⎟⎠
fg, the two-phase, gas-volume fraction, is given by
where γ wg represents the wet-gas gravity, γ g the dry-gas 1− ξ
fg = = f g hom o (1 − ξ )
gravity, γ o the condensate gravity, Rg the producing gas/oil R +1
ratio, and Rw the producing gas/water ratio. The total wet-gas
rate is estimated from the following relation
10 SPE 90754

Therefore, for this particular case, the two-phase gas voulume NL = µ L 4 g / σ 3 ρL = 0.0023
fraction is 1.3% less than the one calculated using the
homogeneous model. ρg
The Ansari et al. model.8 To calculate liquid-film thickness, δ, Nw = (vsg µ L / σ ) 2 = 0.000332
Ansari et al. noted that the pressure gradient for the gas core ρL
must equal that for the liquid film. They wrote the expressions
for these two gradients in terms of liquid film thickness and For Nw < 0.005, ε/d = σ/(ρgvsg2)/d = 0.000323
solved for δ by equating the two expressions. The expression for However, the minimum value for ε/d is 0.001; therefore, we
δ so generated is an implicit one requiring numerical solution. set ε/d equal to 0.001. Note that this ε/d value is higher than
We use a simple approach to estimating δ by noting that 0.0006, used in the homogeneous model. We observed
geometrical considerations allow us to write δ = (1–fg)/4. To consistently higher ε/d ratio by the Aziz et al. model for all the
estimate gas-volume fraction, fg, we used the well-known wells used in this study.
Lockhart-Martinelli (L-M) correlation14 for the annular-flow The Reynolds number (= Dρgvsg/µg), works out to
regime, as follows: 1,183,910. Using the Chen16 explicit correlation, one can
compute the friction factor as
L-M Parameter: X2 = {(1–x)/x}1.8 (ρg / ρL) (µL/µg)0.2
1
The value of x, quality, to be used in the above expression is f = 2
= 0.0199
⎡ ⎛ ε / d 5.0452 ⎞⎤
the mass fraction of the vapor phase. Because we are ⎢4 log ⎜ - logΛ ⎟⎥
interested in estimating the film thickness of the liquid on the ⎣ ⎝ 3.7065 Re ⎠⎦
channel wall, the liquid droplets flowing as an entrained phase
along with the gas core needs to be counted as part of the core Therefore, the pressure gradients are
vapor phase. This calculation step requires an estimate of the
fraction, FE, of the total liquid that is flowing as entrained – (dp/dz)F = f vsg2ρg /(2gcd) = 0.0088 psi/ft.
droplets, which is given by – (dp/dz)H = ρm g/gc = 0.0541 psi/ft.

FE [
= 1 − exp − 0.125(vsgc − 1.5) ] – dp/dz = 0.0632 psi/ft total pressure gradient at wellhead.
For this particular example, the Aziz et al. method for
The critical gas velocity is given by calculating frictional (and hence the total) pressure gradient
turned out to be less than that of the homogeneous model. This
lower estimate is attributed to the use of gas density and gas
ρL / ρg
v sgc = 10 4 v sg µ g = 86.8 superficial velocity instead of the corresponding values for the
σ two-phase mixture, used in the homogeneous model.
Therefore, FE = 0.99999 However, in gas-condensate wells with low-CGR, one can
approximate vsg ~ vm and ρg ~ ρm. In those cases, the use of a
The “modified” quality for this case then higher roughness value for the liquid film in the Aziz et al.
method will result in a higher estimate of the pressure drop
xmod = {(vsg ρg + FE* vsL ρL)/( vsg ρg + vsL ρL)} = 0.99999. owing to friction. We observed a general tendency of the Aziz
et al. method to overestimate the pressure drop in the majority
The L-M Parameter: X2 = {(1–xmod)/xmod}1.8 (ρg / ρL) (µL/µg)0.2 of the condensate wells that we investigated.

= {(1–0. 99999)/ 0. 99999}1.8 (4.0/44.12) (1/0.0135)0.2

= 2.16 ×10–10
Therefore,

fg = (1+X0.8) –0.378 = 0.99995

δ/d = (1–fg)/4 = 1.3 ×10-5
The Aziz et al. Model. Aziz et al.7 adopted the Duns and Ros
model15 for the mist flow, appropriate for gas-condensate
wells. Let us work through the same example to understand
how pressure-drop is computed. The dimensionless variables
are
NVL = vsL (ρL/gσ)1/4 = 2.719
NVg = vsg (ρL/gσ)1/4 = 25.98
Nd = d g( ρL - ρg)/σ = 50.49