chemical engineering research and design 9 0 ( 2 0 1 2 ) 507–513
Contents lists available at ScienceDirect
Chemical Engineering Resear
r .co
A me an
level -ph
predic
E. Perey Kou
a McDougall 741
b Thermal S ela
c Departmen 04, U
d Chevron E
a
A l in t
p colle
d ons o
model/method can be used), and the confidence level in the predictions is quantified. Also, gaps in the data base are
identified and future studies required in this are discussed.
© 2011 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
Keywords: Flow pattern; Two phase flow; Mechanistic modeling; Confidence level
1. Int
The term “
the phases
pipes. Whe
two phases
figurations
the interfa
teristics.
Determi
in two-pha
namely, ph
and mass
and rate of
the existin
flow patter
of two-pha
tion of pre
effective er
ical additiv
Also, segre
∗ Correspon
E-mail a
Received
0263-8762/$
doi:10.1016/
roduction
flow pattern” refers to the spatial distribution of
, which occur during gas–liquid two-phase flow in
n gas and liquid flow simultaneously in a pipe, the
can distribute themselves in a variety of flow con-
. The flow configurations differ from each other in
ce distribution, resulting in different flow charac-
nation of flow patterns is a fundamental problem
se flow analysis. Indeed all the design variables,
ase velocity, pressure drop, liquid holdup, heat
transfer coefficients, residence time distribution,
chemical reaction, are all strongly dependent on
g flow pattern. Thus, knowledge of the existing
n can help the industry carry out a better design
se flow systems. These include accurate predic-
ssure drop and liquid inventory in pipe flow, and
osion corrosion planning, utilizing properly chem-
es, such as corrosion inhibitors and demulsifies.
gated flow patterns are often desired for phase
ding author. Tel.: +1 918 740 8543; fax: +1 918 631 2059.
ddress: ep@utulsa.edu (E. Pereyra).
9 February 2010; Received in revised form 6 August 2011; Accepted 9 August 2011
separation efficiency improvement. Nowadays, a downward
inclined inlet section may be installed upstream of the sep-
arator, for promoting stratification and pre-separation of the
phases. This can be designed utilizing flow pattern prediction
to ensure stratified flow at the inlet section. Finally, the trans-
port and deposition of solid particles, e.g., hydrates, paraffins
and waxes, is an important flow assurance issue, which is
strongly affected by the different flow patterns.
In designing the above applications risky decisions can be
made based on the predicted flow pattern, which can result in
severe economical losses. Thus, it is very important to deter-
mine the confidence level in the prediction of the existing flow
pattern. However, no past studies have attempted to address
the confidence level in such predictions.
Fig. 1 presents the different transition boundaries occur-
ring in gas–liquid flow, as well as the different existing flow
patterns. The physical mechanisms and respective mod-
els of the different transition boundaries can be found in
Shoham (2006). Following is a summary of the commonly
accepted flow patterns, for the entire range of inclination
angles.
– see front matter © 2011 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
j.cherd.2011.08.009
j ourna l ho me page: www.elsev ie
thodology and database to qu
of methods for gas–liquid two
tion
raa,∗, C. Torresb, R. Mohanc, L. Gomeza, G.
School of Petroleum Engineering, The University of Tulsa, Tulsa, OK
cience Department, University of Los Andes, Merida, 5101 - Venezu
t of Mechanical Engineering, The University of Tulsa, Tulsa, OK 741
nergy Technology Company, Houston, TX 77002, United States
b s t r a c t
novel methodology is presented to quantify the confidence leve
atterns in pipes. An experimental flow pattern data base has been
ata points). The experimental data are compared with the predicti
ch and Design
m/locate /cherd
tify the confidence
ase flow pattern
bad, O. Shohama
04, United States
nited States
he prediction of gas–liquid two-phase flow
cted, consisting of 12 studies (a total of 9029
f the unified Barnea (1987) model (any other
508 chemical engineering research and design 9 0 ( 2 0 1 2 ) 507–513
A
0.01
0.1
1
10
0.01 0.1 1 10 100
VSG, m/s
VS
L,
m
/s
Annu lar
(A )
Slug Flow (SL)
Dispersed Bubble (DB )
A A
J
F
G
Stratifi ed
Smoth (SS)
Stratifi ed
Wavy
(SW)
C
B
0.01
0.1
1
10
0.01 0.1 1 10 10 0
VSG, m/s
VS
L,
m
/s
Bubble (B)
Annula r
(A)Slug Flo w (SL)
Dispe rsed Bubble (DB)
E
J
J
F
G
Fig. 1 – Transition boundaries and existing flow patterns in gas–liquid two-phase flow.
1.1. Ho
The existin
sified as st
stratified-w
and elonga
flow (DB).
1.2. Ver
In this rang
disappears
churn flow
around the
considered
flow (CH), a
1.3. Do
For downw
stratified-w
inclination
As observe
tical flow, d
all inclinat
uid and hig
the annula
f fal
ow i
ually
pipe
ding
.
Ex
erim
mos
by Ta
was
air w
over
o +9
flow
g the
(198
onta
1987
orizo
m, w
out
ta fo
Table 1 –
Authors
Shoham
Lin
Kouba
Kokal
Wilkens
Meng
Manabe
Van Dresar
Mata et al.
Abduvayt e
Gokcal
Omebere-I
rizontal and near-horizontal flow
g flow patterns in these configurations are clas-
ratified flow, including stratified-smooth (ST) and
avy (SW), intermittent flow (I), which includes slug
ted-bubble, annular flow (A) and dispersed-bubble
tical and sharply inclined flow
e of inclination angles, the stratified flow regime
and a different flow pattern is observed, namely,
. Usually, the flow patterns are more symmetric
pipe axis, and are less dominated by gravity. The
flow patterns are bubble flow (B), slug flow (I), churn
nnular flow (A), and dispersed-bubble flow (DB).
wnward inclined and vertical flow
ard inclined flow, the dominant flow pattern is
avy flow, occurring over a wide range of downward
angles, namely, from horizontal flow up to −80◦.
d in horizontal, upward inclined, and upward ver-
ispersed-bubble flow and annular flow occur for
ion angles (including downward flow) at high liq-
form o
ward fl
that us
off the
depen
phases
2.
An exp
of the
given
which
lizing
study c
−90◦ t
zontal
varyin
Kouba
a horiz
Kokal (
near h
76.3 m
carried
his da
h gas flow rates, respectively. For downward flow,
r regime exists also at low gas flow rates, in the
present stu
sure on flow
Summary of studies in the data base.
Year Variables range
1982 Air water, ID = 2 and 1 in., from � = −90◦ to +90
1982 Air water, ID = 25.4 and 95.4 mm, � = 0◦
1986 Air-kerosene, � = 0◦, ID = 3 in.; �L = 814 kg/m3; �
1987 � = 0◦, ±1◦, ±5◦ and ±9◦; ID = 25.8; 51.2 and 76.3
�L = 858 kg/m3; �L = 7 cP
1997 Salty water and oil; P = 40 psi; � = 0◦, 1◦ and 90◦
2001 Annular stratified flow transition for � = 0◦, ±1
2001 P = 209.3 psi and 464.8 psi; � = 0◦, 1◦ and 90◦
and Siegwarth 2001 Nitrogen and hydrogen, � = 1.5◦; ID = 8.43 mm
2002 ID = 2 in., � = 0◦, �L = 480 cP
t al. 2003 P = 592 and 2060 kPa; � = 0◦, 1◦ and 3◦; ID = 54.9
nitrogen and water
2005 � = 0◦; ID = 50.8; �L = 889 kg/m3; �L = 181–587 cP
yari et al. 2007 P = 2000 kPa, 9000 kPa; � = 90◦; ID = 189 mm; nit
ling-film. The slug flow pattern in vertical down-
s similar to that occurring in upward flow, except
the Taylor bubble is unstable, located eccentrically
axis. The Taylor bubble may either rise or descend,
on the relative flow rates of the gas and liquid
perimental data base
ental data base has been collected, which consists
t relevant studies on flow pattern prediction, as
ble 1. The earliest set of data is Shoham (1982),
acquired in 50.8 and 25.4 mm pipe diameters, uti-
ater at atmospheric conditions. This was the first
ing systematically all the inclinations angles, from
0◦. At the same time, Lin (1982) carried out hori-
experiments in 25.4 and 95.4 mm diameter pipes,
superficial gas velocity from 0.8 to 200 m/s. Later,
6) carried out an experimental study on slug flow in
l 3 in. diameter pipe, using air-kerosene. Following,
) studied two-phase flow patterns in horizontal and
ntal flow, utilizing pipe diameters of 25.8, 51.2 and
ith air and light oil as working fluid. Wilkens (1997)
studies on gas–liquid flow at 0◦, 1◦ and 90◦. Only
r salty water and air have been considered in the
dy. Later, Manabe (2001) studied the effect of pres-
patterns for 0◦, 1◦ and 90◦, using oil and natural
School Points
◦ TelAviv Univ. 5676
Univ. Illinois 141
L = 1.9 cP Univ. Tulsa 53
mm; Univ. Alberta, Canada 1668
Univ. Ohio 204
◦, ±2◦ Univ. Tulsa 153
Univ. Tulsa 247
NASA 116
Intevep, Venezuela 80
and 106.4 mm; Waseda Univ., Japan 443
Univ. Tulsa 183
rogen and water SINTEF 98
chemical engineering research and design 9 0 ( 2 0 1 2 ) 507–513 509
Fig. 2 – Dat
inclination
gas. Van Dr
gen and ni
upward inc
Mata et al.
high viscos
Recently, G
50.8 mm di
uid viscosit
(2007) prese
flow patter
The dat
points. Fig.
variables o
the superfi
resent the
a base variables histograms: (A) superficial velocities, (B) densiti
angle.
esar and Siegwarth (2001) used a mixture of hydro-
trogen to study gas–liquid flow patterns in a 1.5◦
lined pipe of 8.43 mm diameter. In a later study,
(2002) presented a flow pattern map for oil with
ity (480 mPa s) and air, in a 50.8 mm horizontal pipe.
okcal (2005) carried out an experimental study in a
ameter horizontal pipe, utilizing two different liq-
ies (181 and 587 mPa s). Finally, Omebere-Iyari et al.
nted an analysis on the effect of high pressure on
ns at 90◦ using larger diameter pipe of 189 mm.
a base consists of a total 9029 experimental data
2 presents the distributions of the most important
f the data base. Fig. 2(A) shows the histogram of
cial gas and liquid velocities, where the bars rep-
percentage of the total number of experimental
points. As
are concen
gas velocit
0.1 and 10 m
carried out
gas and liq
the numbe
intervals. F
out mainly
the liquid-
water (6464
using oils a
variation o
of the data
sponding to
es, (C) liquid viscosity, (D) pipe diameter and (E)
can be seen, the superficial liquid velocity points
trated between 0.01 and 10 m/s. For the superficial
y, the largest part of the experiments is between
/s. Also, a significant number of tests have been
in the interval of 10–100 m/s. Fig. 2(B) contains the
uid density histograms, where the bars represent
r of points with densities occurring between the
or the gas density, experiments have been carried
with air at atmospheric conditions (0–5 kg/m3). For
phase, the most commonly used working fluid is
points); the rest of the tests have been acquired
nd kerosene with densities of 750–900 kg/m3. The
f the liquid viscosity is presented in Fig. 2(C). Most
points fall in the first and second intervals, corre-
water and kerosene. There is a lack of information
510 chemical engineering research and design 9 0 ( 2 0 1 2 ) 507–513
Table 2 – Comparison of model prediction and entire data base.
(A) Successful (B) Unsuccessful
Total No. [%] DB SS SW A I B
DB 523 409 78.20 409a 0 3 0 111 0
SS 420 232 55.24 0 232a 70 14 104 0
SW 1522 832 54.66 54 319 832a 99 218 0
A 1907 1315 68.96 4 15 150 1315a 423 0
I 4537 3860 85.08 332 16 38 236 3860a 55
B 120 98 81.67 17 0 0 0 5 98a
Total 9029 6746 74.71 816 582 1093 1664 4721 153
a Diagonal elements are successful predictions.
for the inte
mention th
transition s
also includ
liquid visco
tal flow. Fig
The majori
50.8 mm pi
pipe diame
ation of th
tests were
3. Co
A compute
on the uni
tern maps
(−90◦ to +9
software p
phase flow
FLOPATN d
different fl
liquid supe
culated tran
flow-patter
coordinate
compiled a
in an Excel
following p
The inp
consists of
cosities and
inclination
VBA subrou
link library
perature ar
a gi
ing s
pe ge
ser
rpos
turn
noth
atter
and p
Co
ectio
tter
n pre
ation
tio b
xper
sults
ted
re gi
f 902
redic
itten
h 81.
ed-w
le 2(B
mns
lum
flow
redic
f the
itten
betw
r ana
ing
Table 3 –
DB
ST
A
I
B
Total
a Diagonal
rval 7–30 mPa s and 40–170 mPa s. It is important to
at there have been no experimental flow pattern
tudies for high viscous liquids (� > 7 mPa s), which
e the effect of inclination angle. All the studies for
sities greater than 7 mPa s correspond to horizon-
. 2(D) presents the histogram for the pipe diameter.
ty of the studies have been carried out in 25.4 and
pes, with only few experiments conducted in large
ters (>150 mm). Finally, Fig. 2(E) presents the vari-
e inclination angle, where the largest parts of the
conducted between 0◦ and −5◦.
mputer library
r program, FLOPATN, has been developed based
fied Barnea (1987) model, for generating flow pat-
applicable for the entire range of inclination angle
0◦), as well as for flow pattern predictions. This
rovides important information needed for two-
design. For a given set of inlet flow conditions,
etermines the transition boundaries between the
ow-pattern regions, as a function of the gas and
rficial velocities, vSG and vSL, respectively. The cal-
sition boundaries are then plotted in the form of a
n map using vSG and vSL as coordinates in a log–log
system. The library is written in Fortran 90 and is
s a dynamic link library (DDL). The user interface
Visual Basic (VBA) worksheet “FLOPATN.xls”. The
resents a description of the main program features.
ut data interface is implemented in Excel, which
: (1) gas and liquid fluid properties: densities, vis-
surface tension; (2) pipe geometry: pipe diameter,
and roughness. Once the input data are entered, a
tine validates the input and executes the dynamic
(DLL). Note that the operating pressure and tem-
e implicit in the fluid properties.
For
operat
and pi
Excel u
this pu
and re
tern. A
flow p
erties
4.
This s
flow pa
patter
observ
the ra
total e
son re
presen
tions a
total o
fully p
interm
(B) wit
stratifi
Tab
in colu
first co
bubble
were p
most o
interm
region
Simila
remain
Comparison result for (ST = SW + SS).
(A) Successful
Total No. [%] DB
523 409 78.20 409a
1942 1453 74.82 54 1
1907 1315 68.96 4
4537 3860 85.08 332
120 98 81.67 17
9029 7135 79.02 816 1
elements are successful predictions.
ven set of pipeline flow conditions, including the
uperficial gas and liquid velocities, fluid properties,
ometry, it is possible to predict the flow pattern. An
defined subroutine (FPPredIn) has been created for
e, which receives the input data described above,
a code, which represents the predicted flow pat-
er subroutine (FP) is capable to generates and plots
n maps based on the given input data (fluid prop-
ipe geometry).
mparison study
n describes a comparison between the predicted
n and the experimental data base. A particular flow
diction is successful if it agrees with experimental
. Thus, the model performance is quantified by
etween number of successful predictions and the
imental data points. Table 2 presents the compari-
for the entire data base. Successful predictions are
in part (A) of the table, while unsuccessful predic-
ven in part (B). As can be seen in Table 2(A), from the
9 experimental points, 6746 (74.7%) were success-
ted. The flow pattern with the best success rate is
t flow (I), at 85.1%, which is followed by bubble flow
6% success. The lowest success rate corresponds to
avy flow (SW), at 54.7%.
) provides the unsuccessful predictions, presented
for each of the flow pattern. For example, the
n presents unsuccessful predictions for dispersed-
(DB). As can be seen, 54 “DB” experimental points
ted as “SW”, 4 as “A”, 332 as “I” and 17 as “B”. Thus,
unsuccessful predictions of DB flow correspond to
t flow “I”. These data points occur in the transition
een “DB” and “I”, which is hard to identify visually.
lysis is presented for the other flow pattern in the
columns of Table 2(B).
(B) Unsuccessful
ST A I B
3 0 111 0
453a 113 322 0
165 1315a 423 0
54 236 3860a 55
0 0 5 98a
675 1664 4721 153
chemical engineering research and design 9 0 ( 2 0 1 2 ) 507–513 511
Table 4 – = SW + SS + A) and (dispersed = DB + B).
(B) Unsuccessful
ispersed Segregated Intermittent
Dispersed 524a 3 116
Segregated 58 3046a 745
Intermitten 387 290 3860a
Total 969 3339 4721
a Diagonal
Table 2
and “SW” d
For this cas
most of th
between an
be attribut
experimen
Similarl
the segrega
patterns (D
to 82.3%. Th
forward di
flow patter
5. Ide
Tables 2–4
where the
ried out in
data base
fluid prope
analysis of
map. When
the transiti
and presen
different flo
is presente
5.1. Dim
For horizon
namely, th
and Martin
pressure gr
between st
annular flo
F =
√
�G
(�L −
The second
X2 = −(dp/−(dp/
Fig. 3 show
the observe
for horizon
non-stratifi
originally b
transition
“B”) corresp
ig. 3, the empty squares represent the all the points
stratified flow (“ST”) was predicted and a non-stratified
ttern was observed. The filled squares represent all the
where a non stratified flow pattern was predicted and
as been observed. As can be seen in the figure, all the
s cor
arou
dict
emp
tion
visco
�L >
inter
ed “I
on p
may
roug
w pa
ht of
is a f
s liqu
ect o
Str
rizon
dicte
1
− h˜L)
h˜L i
ed fl
sionl
s are
Results for Shoham (1982) flow pattern definition (segregate
(A) Successful
Total No. [%] D
643 524 81.49
3849 3046 79.14
t 4537 3860 85.08
9029 7430 82.29
elements are successful predictions.
is transformed into Table 3 by combining the “SS”
ata points into stratified flow “ST” (ST = SS + SW).
e, the total success rate increases to 79.0%, whereby
e failed predictions are located in the region
nular and intermittent flow. This discrepancy can
ed to the different criteria used by the different
talists to classify the flow patterns.
y, Table 3 is transformed into Table 4 by combining
ted flow patterns (ST + A) and the dispersed flow
B + B). The total success rate for this case increases
is improvement is owing to the clear and straight-
stinction between the two considered combined
ns.
ntification of discrepancy regions
do not provide information about the conditions
discrepancies occur. Thus, further analysis is car-
this study to identify these regions. The acquired
presents a wide range of operational conditions,
rties and pipe geometry, which complicate the
the data utilizing a simple vSG vs. vSL flow pattern
the analysis is carried out in dimensionless form,
on lines and discrepancy regions can be identified
ted in a generalized form, which can be extended to
w conditions with more confidence. This analysis
d next.
ensionless horizontal flow pattern map
tal flow, two dimensionless groups (given below),
e modified Froude number (F) and the Lockhart
elli parameter (X) (which represents the superficial
adient ratio) determine the transition boundaries
ratified to non-stratified flow, and annular to non-
w (originally proposed by Taitel and Dukler, 1976).
�G)
vSG√
d g cos(�)
(1)
In F
where
flow pa
points
“ST” h
failure
spread
the pre
The
predic
liquid
cosity,
while
observ
sificati
points
flow th
this flo
the rig
of “A”
viscou
the eff
5.2.
For ho
be pre
F2
[
(1
where
stratifi
dimen
less ga
1
10
parameter is the Lockhart and Martinelli given by:
dL)SL
dL)SG
, Y = (�L − �G) g sin(�)−(dp/dL)SG
(2)
s the discrepancy between the predicted (Pred) and
d (Obv) flow pattern on a dimensionless F vs. X map
tal flow. The solid line represents the stratified to
ed transition boundary (transition “A”) proposed
y Taitel and Dukler (1976). The dashed line is the
between annular to non-annular flow (transition
onding to Barnea (1987).
0.001
0.01
0.1
0.01
F
"ST
"A"
"I"
Tra
Fig. 3 – Pre
pattern ma
responding to the stratified to non-stratified are
nd transition “A”, which attests to the accuracy of
ion of this transition.
ty triangles correspond to observations of “I” and
of “A”. On the other hand, the filled triangles (low
sity, �L < 7 cP) and the filled rhombs (high liquid vis-
150 cP) correspond to observations of annular flow
mittent flow was predicted. It is believed that the
” to the left of transition “B” is a flow pattern clas-
roblem. As explained by Shoh