International Journal of Heat and Fluid Flow 33 (2012) 156167

Contents lists available at SciVerse ScienceDirect

International Journal of Heat and Fluid Flow

journal homepage: www.elsevier.com/locate/ijhff

Two-phase ow structure in large diameter pipes

T.R. Smith, J.P. Schlegel, T. Hibiki, M. Ishii
School of Nuclear Engineering, Purdue University, 400 Central Dr, West Lafayette, IN 47907-2017, United States

a r t i c l e

i n f o

Article history:
Received 24 May 2011
Received in revised form 8 September 2011
Accepted 22 October 2011
Available online 15 December 2011
Keywords:
Large pipe
Interfacial area
Void fraction
Flow regime

a b s t r a c t
Flow in large pipes is important in a wide variety of applications. In the nuclear industry in particular,
understanding of ow in large diameter pipes is essential in predicting the behavior of reactor systems.
This is especially true of natural circulation Boiling Water Reactor (BWR) designs, where a large-diameter
chimney above the core provides the gravity head to drive circulation of the coolant through the reactor.
The behavior of such reactors during transients and during normal operation will be predicted using
advanced thermalhydraulics analysis codes utilizing the two-uid model. Essential to accurate twouid model calculations is reliable and accurate computation of the interfacial transfer terms. These
interfacial transfer terms can be expressed as the product of one term describing the potential driving
the transfer and a second term describing the available surface area for transfer, or interfacial area concentration. Currently, the interfacial area is predicted using ow regime dependent empirical correlations; however the interfacial area concentration is best computed through the use of the onedimensional interfacial area transport equation (IATE). To facilitate the development of IATE source
and sink term models in large-diameter pipes a fundamental understanding of the structure of the
two-phase ow is essential. This understanding is improved through measurement of the local void fraction, interfacial area concentration and gas velocity proles in pipes with diameters of 0.102 m and
0.152 m under a wide variety of ow conditions. Additionally, ow regime identication has been performed to evaluate the existing ow regime transition criteria for large pipes. This has provided a more
extensive database for the development and evaluation of IATE source and sink models. The data shows
the expected trends with some distortion in the transition region between cap-bubbly and churn-turbulent ow. The ow regime map for the 0.102 m and 0.152 m diameter test sections agree with the existing ow regime transition criteria. It may be necessary to perform further experiments in larger pipes and
at higher gas ow rates to expand the range of conditions for which models can be developed and tested.
2011 Elsevier Inc. All rights reserved.

1. Introduction
Two-phase ows occur in a wide variety of common industrial
applications. Many of these applications involve large diameter
pipes. This is especially true of the chemical and petroleum industries, where bubble column chemical reactors and large pipe
pumping systems are quite common. In the nuclear industry, two
phase ows often occur in large channels. For this reason a lack
of fundamental knowledge in this area can have signicant ramications for nuclear safety. In next-generation BWR systems, for
example, the ow through the reactor is driven by natural circulation. This requires a large diameter chimney section above the core
to provide the necessary gravity head (Ishii et al., 1998). This
region is very sensitive to variations in the two-phase ow, especially during reactor startup. Flow in large pipes has several significant differences from ow in small pipes. Once the ow channel
Corresponding author. Tel.: +1 765 494 5759; fax: +1 765 494 9570.
E-mail address: ishii@purdue.edu (M. Ishii).
0142-727X/$ - see front matter 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.ijheatuidow.2011.10.008

diameter is larger than the maximum cap bubble size, which is

dened by Kataoka and Ishii (1987) as

DH
DH q  30
r

g Dq

a variety of fundamental changes to the ow occur. Here, DH is the

hydraulic diameter, r is the surface tension, g is gravitational acceleration, and Dq is the density difference between the liquid and gas
phases. First slug bubbles bridging the entire pipe cross-section can
no longer be sustained due to Taylor instability, which causes the
upper surface of larger bubbles to distort and collapse, breaking
the large bubble into two or more daughter bubbles. This results
in signicant three-dimensional recirculatory behavior as the liquid
ows around the cap bubbles rather than being forced out of the
way, as is the case with slug bubbles. This causes signicant
changes to the void fraction and velocity proles and can result in
very different behavior from ow in smaller pipes, where slug bubbles can be sustained. For reactor safety it is vitally important that

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

157

Nomenclature
Latin Characters
ai
interfacial area concentration (1/m)
C
constant
D
diameter (m)
d
diameter (m)
F
fraction of eddies causing breakup ()
g
gravitational acceleration (m/s2); breakup frequency
(s1)
h
enthalpy (J/kg)
Kg
constant ()
L
length (m)
M
interfacial momentum transfer (kg/m2 s2)
Nlf
viscosity number ()
n
concentration (m3)
p
pressure (kPa)
q
heat transfer (W/m2)
r
radial location of measurement (m)
R
pipe radius (m)
r
radius (m)
S
collision cross-sectional area (m2)
t
correlated time (s)
tb
breakup time (s)
u
velocity (m/s)
V
volume (m3)
v
velocity (m/s)
We
Weber number ()
Greek Characters
a
void fraction ()
b
PDF of daughter particle size ()
e
turbulent dissipation rate (m2/s3)
h
collision frequency (s1)

the capability to accurately model and predict two-phase ows in

such systems be developed.
These models will be integrated into existing thermalhydraulic analysis codes for use in predicting system behavior. The most
accurate way of predicting system behavior is full-scale testing,
however in the nuclear industry full-scale tests are expensive
and often impractical. In place of full-scale tests, a variety of scaled,
separate effect, and local phenomena studies are used to develop
mathematical models for the prediction of ow behavior under a
wide variety of conditions. These models are then solved numerically using a computer. For this approach, reliable models with
appropriate constitutive relations are essential for accurate predictions of the behavior of two-phase ow systems.
Most of these analysis codes make use of the two-uid model,
which is currently the most practical model for two-phase ow because it is more detailed than other models while using fewer resources than DNS or LES. This two-uid model is the most
detailed two-phase ow model currently used in system analysis
codes. This model treats each phase separately, resulting in two
sets of balance equations for mass, momentum and energy. The
one drawback to this model is its complexity, which is largely
introduced by the terms representing the transfer of mass,
momentum and energy across the gasliquid interface. Mathematically, the one-dimensional version of the two-uid model is given
as (Ishii and Hibiki, 2010):

@hak iqk @
hak iqk hhv zk ii hCk i
@z
@t

C
k

l
U

u
Dq
DT

s
q
r
rv2
s

interfacial mass transfer (kg/m3 s)

coalescence efciency ()
viscosity (Pa s)
energy source due to turbulent dissipation (W/m3)
source or sink term for IATE (1/ms)
density difference between phases (kg/m3)
time interval (s)
characteristic time (s)
density (kg/m3)
surface tension (N/m)
variance
shear force (N/m3)

Superscripts and Subscripts

non-dimensional value
b
bubble
c
critical value
h
hydraulic
i
interfacial value; bubble index
j
bubble index
k
value for phase k
T
value due to turbulence
t
turbulent uctuation
ted
value for turbulent Eddy
v
value for velocity
w
value at the wall
z
denotes axial direction
P
Operators
summation
hi
area-averaged quantity
hhii
void-weighted area-averaged quantity

time-averaged quantity



@hak iqk hhv k ii @
C v k hak iqk hhv zk ii2
@t
@z
@hhpk ii @
4akw skw
hak ihhskzz sTkzz ii 
hak i
hak iqk g z
@z
@z
D


@ ak
hhv ki iihCk i hMik  rak  si iz pki  pk
@z

@hak iqk hhhk ii @

C hk hak iqk hhv kz iihhhk ii
@t
@z
Dk
@
n
hak i hhpk ii  hak ihhqk qTk ii h akw q00kw hhhki iihCk i
@z
Dt
A
hq00k ai i hUik i

X
hCk i 0

X
hM ik  rak  si iz 0

X


hCk ihhhki ii hq00k ai i 0:

Here, Ck, M ik , si, q00ki , and uk are the mass generation, generalized
interfacial drag, interfacial shear stress and interfacial heat ux,
which are key parameters in the interfacial transfer of mass,
momentum and energy. Denitions of other quantities can be found
in the nomenclature.

158

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

The interfacial transfer terms can be thought of as being composed of two components, one being the amount of interface available for transfer, or interfacial area concentration, and the other
being the driving potential for the transfer, with the form being

Interfacial l Transfer Terms  a1  Driv ing Force

Therefore in order to close the two-uid model accurate constitutive relations must be developed for the driving forces and the
interfacial area concentration.
Traditionally the interfacial area concentration has been specied using static, ow-regime dependent criteria. This approach
has some shortcomings however, as they are limited by the accuracy of the ow regime transition criteria and the experimental
range for which they have been validated. This static nature limits
the ability of the models to predict truly dynamic features of twophase ow during transient events and in developing ow, especially in the transition regions between ow regimes. This method
can also lead to numerical instabilities and bifurcations that can result in degraded convergence or prevent convergence altogether.
Further, the majority of these models have been developed for
small pipes rather than for large-diameter channels.
For these reasons a more dynamic approach to the prediction of
interfacial area concentration has been proposed by developing a
transport equation for the uid particle number density. Integrating such an equation over the entire range of bubble sizes resulted
in a number density transport equation. Multiplying the number
density transport equation by the average surface area of a bubble
an interfacial area transport equation (IATE) was developed which
was later rened and is given by Ishii and Hibiki (2010) as,

@ai
2 ai  @ a
vi
v g  gph
r  ai~
r  a~
3 a
@t
@t
 2
1 a X
Rj pD2bc Rph

3w ai
j

where ai is the average interfacial area per unit volume of uid and
~
v i is the interfacial velocity. On the right side of Eq. (9) Rj represents
the source and sink reaction rates of interfacial area due to the
interaction of uid particles by breakup and coalescence processes,
while Rph represents the source and sink reaction rate of interfacial
area concentration due to evaporation or condensation.
This was followed by the development of the two-group IATE, in
which the bubble number density equation was averaged twice,
once for small spherical and distorted Group 1 bubbles and once
for cap-shaped
p Group 2 bubbles with diameter larger than
Db 4= r=g Dq based on the bubble drag properties and the work
of Ishii and Zuber (1979). Additional details regarding this modeling effort can be found in Ishii and Hibiki (2010). This renement of
the IATE is shown in Eqs. (10) and (11) for a one-dimensional, steady-state system with no phase change. In the equations, C is a constant and Dc1 is the ratio of the Group 1 Sauter mean diameter to
the maximum distorted bubble size. This allowed improved accuracy in the slug and churn-turbulent ow regimes, however both
of these models were developed for small pipes.



X
d
2
hai1 i d
hai1 ihhv i1 ii
 CDc1 2
ha1 ihhv g1 ii
huj1 i
dz
3
ha1 i dz
j
d
2hai2 i d
hai2 ihhv i2 ii
ha2 ihhv g2 ii
dz
3ha2 i dz
X
hai1 i d
ha1 ihhv g1 ii
h/j2 i;
CDc1 2
ha1 i dz
j

10

diameter channels kept in mind. To validate and benchmark these

models it is necessary to have a fundamental understanding of
the ow structure, which requires a signicant database of interfacial area concentration data covering a wide range of ow conditions and pipe diameters. To expand the existing database an
experiment has been undertaken to make measurements of the
interfacial area concentration and void fraction proles in round
pipes with diameters of 0.102 m and 0.152 m.
2. Previous studies of interfacial area concentration in large
pipes
An early experiment measuring interfacial area concentration
for large pipes was reported by Sun et al. (2002). This experiment
reported data collected in a 0.102 m diameter test section for the
purposes of evaluating one-group IATE models. As the data was
to be used for the one-group model, which performs well only in
bubbly ows, the data for this experiment was limited to void fractions less than 0.3.
In 2003, Sun et al. reported additional data including void fraction and interfacial area concentration proles as well as bubble
number frequency data. The detailed structure of the two-phase
ow was investigated to determine the development of the interfacial structure along the ow direction and provide a limited database for the future development of the two-group IATE. Void
fractions for this data were as high as 0.45 and included several
conditions in the cap bubbly ow regime as well as the bubbly ow
regime.
Later Shoukri et al. (2003) studied the structure of two-phase
ow in a test section with 0.2 m diameter. The radial distributions
of void fraction, bubble velocity, bubble size and interfacial area
concentration were measured using dual-sensor optical probes.
Experimental conditions were limited to void fractions smaller
than 0.04 as the experiment was intended to measure the hydrodynamics at low void fractions. The authors noted that wall-peak
void proles occur only when the void fraction was very small
and that the bubble sizes for ows in large pipes were smaller than
the bubble sizes reported in the literature for ow in small pipes.
This results in a somewhat higher interfacial area concentration
for ows in large pipes under similar ow conditions.
Shen et al. (2006) also studied two-phase ows in a 0.2 m diameter facility using dual-sensor optical probes. Void fractions for this
study were as high as 0.4, however interfacial area concentration
data was only reported for some ow conditions as dual-sensor
probes cannot accurately measure interfacial area when capshaped Taylor bubbles are present, at void fractions higher than
about 0.25.
Additionally Prasser (2007) performed experiments using air
and water in a 0.195 m diameter test facility. A double wire-mesh
sensor was used to calculate the void fraction and bubble velocities. The use of the wire-mesh sensor allowed direct reconstruction
of the three-dimensional structure of the two-phase ow so that
the total interfacial area concentration could be calculated directly
from the measured bubble shapes. This allows rapid calculation of
the two-phase ow parameters, however the resolution of the
wire-mesh sensors is only about 3 mm.
3. Experimental facility

11

In order to complete the transport equation, these source and sink

processes must be modeled mechanistically with the effect of the
hydrodynamic differences between small diameter and large

3.1. Experimental loop

A schematic of the experimental loop used in this study is
shown in Fig. 1. The system pressure can be as high as 500 kPa,
with liquid supercial velocities up to 10 m/s and gas supercial
velocities up to 20 m/s in the 0.102 m diameter test section and

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

159

Fig. 1. Experimental loop.

4 m/s liquid velocity and 9 m/s gas velocity in the 0.152 m diameter test section. Water is stored in a stainless steel reservoir. A centrifugal pump establishes the liquid ow rate, while compressed
air is supplied by a compressor system. The water ow rate is measured using a magnetic ow meter with an accuracy of 1% while
the air ow rate is measured by rotameters or Venturi mass ow
meters, also with an accuracy of 1%. The air and water mix in a
mixing chamber at the base of the test section before owing
through the test section and returning to the reservoir, which also
serves as a separator. The mixing chamber consists of three stainless steel porous tubes, with pore sizes of 10 lm. Rotameters are
used to maintain a constant liquid ow velocity over the porous
tubes so that the size distribution of the bubbles sheared off the
tubes remains relatively constant for all experiments. This also ensures that the inlet ow condition remains as bubbly ow for all
possible test conditions.

Two vertical test sections, one with diameter of 0.102 m and

one with diameter of 0.152 m, are used. Each is tted with a variety
of instrumentation including differential pressure transmitters and
three mounting anges for conductivity probes and impedance
meters, which are located at z/DH = 5.0, 20 and 30 for the
0.102 m section and z/DH = 4.0, 11 and 18 for the 0.152 m section.
The differential pressure cell measures the pressure drop along the
test section. For ow visualization, a high-speed movie camera
(Kodak Motioncorder Analyzer model 1000/SR) with frame rates
up to 10,000 frames per second is used to capture images for the
study of particle motions. These images were also used to verify
the classication of ow conditions in ow regime identication.
The experimental conditions for the measurement of interfacial
area concentration are given in Fig. 2 with the ow regime boundaries modeled by Schlegel et al. (2009) for large diameter pipes. As
the gures show, the tests cover a wide range of ow conditions

Fig. 2. Test conditions with ow regime transitions proposed by Schlegel et al. (2009), (a) 0.102 m test section, and (b) 0.152 m test section.

160

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

Fig. 3. Current design for four-sensor conductivity probe (Kim et al., 2000).

from bubbly ow to churn ow, so that the resulting data will be

representative of ows encountered in many typical industrial
settings.
3.2. Conductivity probes
The electrical conductivity probe is the key instrument in this
study. First proposed by Neal and Bankoff (1963) it has been one
of the most widely used instruments for local measurements in
gasliquid two-phase ow experiments. A diagram of a conductivity probe used in this study is shown in Fig. 3 (Kim et al., 2000).
Variation in the impedance between the sensors, measured as a
change in voltage, shows when the sensor tips are surrounded by
liquid or by gas. A multi-sensor conductivity probe is capable of
measuring the local interfacial velocity of individual bubbles and
thereby determine the local time-averaged interfacial area concentration. The development of the miniaturized probe used in this
study is found in the work of Kim et al. (2000). The cross-sectional
area of the four sensor probe is about 0.5 mm2.
The uncertainty of the conductivity probe can be generally divided into two sources. The rst source is that due to the probe
structure itself, which is related to the interface velocity, probe
geometry, and measurement rate. This uncertainty gives the
uncertainty in the velocity of a single bubble interface based on
the uncertainty in measuring the time taken for the interface to
travel between sensors. For this experiment fteen separate probes
were used. The same fteen probes were used for all of the experiments, and the maximum uncertainty from this source is 9.8%
using a measurement frequency of 20 kHz. Based on the work of
Kim et al. (2000), who compared the results of probe measurements to the results of image processing techniques, there is a second uncertainty due to the deformation of the interfaces of large
bubbles on contact with the probe. Based on the measurement
uncertainty of the probes and the reported disagreement between
the two measurement techniques, the additional uncertainty due
to this source is approximately 6%. Using error propagation techniques gives a total experimental uncertainty of 11.5%. This uncertainty applies to interfacial area concentration measurements as
well as the individual group void fraction measurements, as the
bubble group is determined by the bubble chord length. The error
in the total void fraction, however, is governed by statistical
parameters and measurement duration and in this experiment is
approximately 5%. This uncertainty is based on the uncertainty in
when a bubble interface crosses the sensor due to the discreet
measurement method, and is computed based on the total number
of bubble signals, measurement frequency, and time-averaged void
fraction.

These experimental uncertainties can be evaluated using other

one-dimensional approaches. The void fraction measurements
made by conductivity probes is converted into an area-averaged
value using a weighted sum, then compared to the void fraction
computed from differential pressure measurements. Using the 5%
uncertainty in the conductivity probe measurement and the 1%
uncertainty in the differential pressure measurement gives an expected difference of 5.5%. Analysis of the differential pressure data
shows an average difference of 5.26%, within the expected discrepancy. The interfacial area concentration is computed from the measured interfacial velocity, so this measure can be validated against
the gas ow rate measurement by computing the area-averaged
gas ux from the local data. The gas ow rate measurement has
a 3% uncertainty, meaning that the expected discrepancy between
the two measurement methods should be less than 14.5%. Comparison of the two measurement methods shows an average difference of 13.6%, which is less than the expected value.
3.3. Flow regime identication
Several experiments were also performed using the electrical
impedance void meter. The electrical impedance void meter consists of two conducting electrodes on opposing sides of the test
section. A high-frequency alternating current is passed through
the two-phase mixture. As water conducts electricity rather readily
but air does not, the variation in the impedance of the mixture as
the ow passes through the measurement volume allows the void
fraction to be measured. This impedance is converted into a voltage output and measured by a computer. This provides a measurement of the time-dependent area-averaged void fraction. The
pattern of variations in the void fraction signal is characteristic of
the ow structure, or ow regime.
To determine the ow regime from the impedance meter signal,
a neural network classication system is used. A neural network is
a simulated group of interconnected articial neurons intended to
simulate the function of the human brain on a limited scale. In the
self-organized approach, the neural network is given a number of
sample data sets and told to group them into several categories.
The network then trains itself using these sample sets to recognize
the most signicant features of the data sets and correctly categorize the inputs. This training is accomplished by altering the weight
values of the inputs for each of the various neurons.
The inputs to the neural network are the mean and standard
deviation of the void fraction signal from the electrical impedance
meter. The neural network system used for ow regime identication in this study is a committee of three neural networks. Each
neural network is trained using one-third of the experimental data.

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

Each network then is used to categorize the remaining data into

bubbly, slug, churn and annular ow. In this way, each data point
is classied twice to conrm the categorization. The result of the
neural network ow regime identication process is a number
associated with each input that corresponds to the category number which that data point was assigned to. These numbers are then
associated with the ow conditions for each experiment to create
the ow regime map (Mi et al., 1998).

4. Results and discussion

4.1. Proles of local data
Local proles of void fraction and interfacial area concentration
can provide a great deal of insight into the behavior and structure
of two phase ows. For this reason, the proles for several sets of
conditions have been measured and are shown in Figs. 47. Fig. 4
shows the proles for all tests with liquid velocity of 1.0 m/s in
the 0.102 m diameter test section. This data was collected at axial
location z/DH = 20. Each data set represents a different gas supercial velocity, allowing evaluation of the effect of the gas velocity on
the void and interfacial area concentration proles. When considering the void fraction proles, it is interesting to note that for lower Group 1 void fractions, a wall peak in the Group 1 void prole
exists even for higher gas velocity conditions while the prole is
nearly at for higher Group 1 void fractions. This behavior is reasonable based on the expected bubble behavior given the behavior
of the lift force at lower void fractions and due to Group 1 bubbles
being crowded out of the center of the pipe by larger Group 2 bubbles at higher gas ow rates (Hibiki and Ishii, 2007). As expected,
the Group 2 void fraction shows distinct center peaking for all ow
conditions. The total void fraction plot shows that the transition
from wall peaking to center peaking occurs at area-averaged void

161

fraction of about 15%, where the void prole is nearly at. This
corresponds roughly to the void fraction at which Group 2 bubbles
begin to be seen.
The local proles show signicant differences from those in
smaller pipe ows. Typically small pipe data is either centerpeaked or wall-peaked depending on the ow conditions or development length (Hibiki et al., 2001). In large pipes however the
available development length is much shorter, with L/D of 30 for
the 0.102 m diameter pipe and 18 for the 0.152 m diameter pipe.
These types of ows are thus generally developing ows. The turbulence in two-phase ows plays an important role in determining
the proles and wall-peaking phenomenon. In particular, the turbulence intensity may be attenuated at high liquid volumetric
uxes or for very small bubble sizes (Elgholbashi and Abou-Arab,
1983; Kataoka et al., 1992). Large particles such as Group 2 bubbles, however, can enhance the turbulence in the ow. Using simple models, Kataoka and Serizawa (1995) demonstrated that
interfacial transfer of turbulence to small-scale bubbles decreases
turbulence due to the energy required to move the interface, while
drag on large bubbles results in a source of turbulence due to interfacial drag. Turbulent mixing near the wall region especially is enhanced by these larger bubbles, resulting in minimization of the
wall-peaking phenomenon. The experimental data indicates that
wall-peaking in large pipes is over-predicted when using smallpipe models (Ohnuki and Akimoto, 1998), which means that either
the lift force is smaller or turbulent dispersion forces near the wall
are larger in large pipe ows. The newly acquired data for the
0.102 m diameter test loop conrms this data.
The interfacial area concentration proles also show interesting
behavior. Unsurprisingly the magnitude of the interfacial area concentration follows the void fraction magnitude rather closely, with
the average value peaking under similar ow conditions. Wall
peaking is even more prevalent for the interfacial area concentration, however. This indicates a denite change in bubble size across

Fig. 4. Effect of changing gas velocity on void fraction and interfacial area concentration.

162

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

Fig. 5. Effect of changing liquid velocity on void fraction and interfacial area concentration.

Fig. 6. Effect of pipe diameter on void fraction and interfacial area concentration.

the pipe with bubbles near the pipe wall having a much smaller
average size than Group 1 bubbles near the center of the pipe. As

expected, the total interfacial area concentration peaks at void

fractions between 50% and 60%. At higher void fractions, the in-

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

163

Fig. 7. Gas velocity and bubble size for varied ow conditions and pipe diameters.

creased Group 2 void fraction and increased bubble size results in

reduced interfacial area concentrations.
Fig. 5 shows similar data, but for varying liquid supercial
velocities and approximately constant gas velocities in the
0.102 m diameter test section. As the gure shows, the effect of
changing the liquid velocity on the total void fraction prole is very
small and the effect on the shape of the Groups 1 and 2 void fraction proles is small. The largest effect of liquid velocity is that at
higher liquid velocity the increased turbulence results in a greater
proportion of Group 1 bubbles. This leads to a higher Group 1 void
fraction and higher total interfacial area concentration at higher
liquid velocities.

The void and interfacial are concentration proles for similar

ow conditions in the 0.102 m and 0.152 m diameter test sections
are shown in Fig. 6. The similarly shaded lines indicate that the
ow conditions are similar, and the two different symbols for that
shade indicate the pipe diameter for that test condition. The data
for the 0.102 m diameter test section was collected at axial location z/DH = 20, while the data for the 0.152 m test section was collected at axial location z/DH = 18. The gure shows several things.
First, that the transition from wall peak to center peak void prole
occurs at lower void fraction in the 0.152 m diameter pipe. Second,
the data from the larger test section shows signicantly increased
void fraction in both bubble groups and greatly increased interfa-

Fig. 8. Axial development of void prole in 0.102 m diameter test facility.

164

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

Fig. 9. Axial development of void prole in 0.152 m diameter test facility.

cial area concentration for Group 1 bubbles. This is likely due to increased turbulence as the pipe diameter increases, resulting in an
increase in the rate of bubble breakup. It should be noted that
the Group 2 interfacial area concentration remains relatively
constant between pipe sizes, which indicates that the increased
turbulence has much smaller effect on larger bubbles. The exception to this at the lowest gas velocity condition, where Group 2
bubbles exist for the 0.152 m test section but not for the 0.102 m
test section. Based on the interfacial area concentration proles
these Group 2 bubbles are relatively small. This difference is likely
due to slight differences in the experimental conditions.
Fig. 7 shows the gas velocity and Sauter mean diameter for each
bubble group for the same ow conditions as shown in Fig. 6. As
expected, the bubble diameter tends to decrease near the pipe
walls. The larger test section shows a trend for larger bubbles than
the 0.102 m test section at low gas velocities, but smaller bubbles
at higher velocities. Also, the bubble velocities tend to be smaller in
the larger test section, explaining the increase in the void fraction
seen in Fig. 6. The velocity proles all show that the gas velocity is
relatively constant across the pipe, though for higher gas velocities
there is a slight decrease in the gas velocity as one moves away
from the pipe center.
Figs. 8 and 9 show the axial development of the Groups 1 and 2
void fraction proles for the 0.102 m and 0.152 m diameter test
facilities for selected ow conditions. Each gure shows identical
ow conditions to allow comparison of the ow development in
each test section. As both gures show, little ow development occurs for L/D greater than 45. In this case, ow development is
meant to refer to changes in the shape of the phase distribution.
Some increase in the void fraction occurs over the length of both
test sections, however this has very little effect on the void distribution, as evidenced by the gures, or on the velocity distribution.
Of note in the two gures are the distinct differences in the proles
for the two pipe diameters, as has been noted for previous gures.
The Group 1 void fraction tends to be signicantly higher while the
Group 2 void fraction is signicantly lower for the larger diameter

test facility, and the Group 2 void prole in the 0.152 m diameter
facility is much more linear from pipe center to pipe wall than
the prole in the 0.102 m facility.
4.2. Area-averaged data and ow development
As most thermalhydraulic analysis codes are one-dimensional,
the area-averaged data is of specic interest here due to its applicability to code testing and evaluation. Fig. 10 shows a selection of
the area-averaged void fraction data. The void fraction data is plotted with the void fraction prediction using the model developed by
Hibiki and Ishii (2007) for liquid velocities of 0 m/s and 1.0 m/s.
The data for high liquid velocity is predicted well, however there
is some scatter in the data for lower liquid velocities. This may
be the result of the increased recirculation seen in large diameter
pipe ows at higher void fractions. As the liquid ow direction
oscillates due to the presence of large bubbles, some bubbles move
downward. These bubbles cannot be measured accurately using
existing conductivity probe methods. This may result in void fraction measurements which are less than the actual void fraction for
low liquid velocities and void fractions over about 0.2.
Additionally, Fig. 11 shows the measured change in interfacial
area concentration as compared to the predicted change if only
gas expansion contributed, n = haiimeas/haiipred. To calculate the
expansion contribution to the interfacial area concentration, the
expansion term from the interfacial area transport equation was
used. Accordingly, the change in interfacial area concentration
can be described as






d
2
hai1 i ha1 ihhv g1 ii dP
hai1 ihhv g1 ii
 CDc1
dz
3
ha1 i
P
dz



 2 hai2 i ha2 ihhv g2 ii dP
d
hai2 ihhv g2 ii
dz
3 ha2 i
P
dz



ha
i
h
a
ihh
v
i1
1
g1 ii dP
CDc1
ha1 i
P
dz

13

14

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

165

Fig. 10. Area-averaged void fraction data.

Fig. 11. Comparison of measured change in interfacial area concentration with predicted change due to bubble expansion.

for Groups 1 and 2 respectively, where P is the system pressure.

This then provides the change in interfacial area concentration
per unit length of the test section at each of the three measurement
locations. To approximate the total change due to expansion, this is
then multiplied by the total length between one measurement location and the next, giving expected values at the second and third
location. The data in the gure is then the ratio of this value to
the measured interfacial area concentration, with values less than

unity indicating that interfacial area concentration sinks are dominant and values greater than unity indicating that interfacial area
concentration sources are dominant. As the gure shows, for the
0.102 m diameter test section bubble coalescence is dominant at
nearly all of the conditions except for those at very low gas ow
rates, where the bubbles are too dispersed for many coalescence
interactions to occur. The trend is similar for the 0.152 m diameter
test section, however in this case coalescence is dominant even at

166

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167

Fig. 12. Flow regime maps with ow regime transitions proposed by Schlegel et al. (2009).

lower gas ow rates. Overall, much of the data is quite far from
unity. This emphasizes the need for mechanistic bubble interaction
models for the IATE to accurately predict the behavior of two-phase
ows.
4.3. Flow regime identication
Currently the interfacial area concentration is calculated by
using static, ow-regime dependent correlations based on existing
experimental data. For this reason it is important that, until the
development of the IATE is complete and has been incorporated
into the most recent thermalhydraulic analysis codes, the ow regime transitions be accurately determined in order to improve the
accuracy and applicability of the correlations. Therefore the results
of the ow regime identication as described in Section 3.3 have
been presented in Fig. 12. The ow regime transitions are those
predicted by Schlegel et al. (2009). As the map for the 0.102 m test
section shows, the existing ow regime transition criteria perform
reasonably well with the ow regime transitions as given by the
neural network agreeing well with those given by the model. The
map for the 0.152 m diameter test section also shows very good
agreement with the model. The map generated by the neural network shows that the transition from bubbly to cap-bubbly ow occurs at slightly lower void fraction than in the 0.102 m test section,
however this is likely due to the injection method. The increased
gas velocity required to achieve a given ow condition in the larger
test section requires signicantly increased gas ow rates, which
may result in larger cap bubbles being formed at the test section
inlet for some conditions where this would not be the case in the
smaller test section. This would articially decrease the void fraction at which the transition is seen for the 0.152 m test section and
would account for the trend seen in the gure.
5. Conclusions
Flow in large pipes is signicant in many industries from pharmaceuticals to nuclear energy. In the nuclear industry, an understanding of ow in large pipes is essential for predictions of
reactor safety and performance. These predictions will be developed using thermalhydraulics analysis codes, all of which use
the two-uid model to calculate the behavior of two-phase uid
ows. The two-uid model is the most practical model for twophase ow available, however it is also quite complex. The most
signicant source of complexity is the interfacial transfer terms,
which can be decomposed into two components: one dening
the available surface area for transfer and another dening the potential driving the transfer. For the two-uid model to be accurate
both of these components must be correctly modeled. The most

advanced model for the interfacial area concentration is the interfacial area transport equation (IATE), however this model has not
been well-developed in large pipes. To facilitate the development
of IATE models, experiments have been performed in large pipes
of diameters 0.102 m and 0.152 m, with liquid supercial velocities
up to 2 m/s and gas supercial velocities up to 8 m/s in the 0.102 m
diameter pipe. Measurements of local void fraction, interfacial area
concentration, and interface velocity were made at three axial
locations along the test section using electrical conductivity probes
with cross sections of 0.2 mm2. The resulting local proles have
been presented along with the axial development of the area-averaged quantities and the results of ow regime identication performed using electrical impedance void probes.
This has resulted in the development of a more extensive database for the development of interfacial area source and sink terms
for the two-group IATE. The trends shown by the data are as expected. At low void fractions, the void prole displays a wall peak.
This transitions to a center peak near the transition from bubbly
ow to cap-bubbly ow. The data in the larger test section shows
similar total void fraction values to the smaller test section but
higher Group 1 void fraction and therefore a higher interfacial area
concentration. These effects are likely due to increased turbulent
mixing in the larger test section, as the smaller test section is near
the boundary of the transition to large pipe behavior and may
therefore still exhibit some of the stabilizing effects of the pipe
wall on the ow. The area-averaged data also shows the expected
trends. Further, the data indicates that the inlet conditions may
have strong effects on the ow pattern. The ow regime map
developed for the 0.102 m diameter test section conrms the existing ow regime transition criteria for large pipes as does the map
developed for the 0.152 m test section, but the map for the 0.152 m
test section shows that the inlet conditions can result in induced
changes to the ow pattern.
To improve the database of interfacial area concentration data
in large pipes it may be necessary to perform experiments in larger
test sections and at signicantly higher gas ow rates so that the
performance of IATE models can be validated for the entire range
of ow conditions that may be seen in reactor systems.
Acknowledgments
This work was performed at Purdue University under the auspices of the US Nuclear Regulatory Commission, Ofce of Nuclear
Regulatory Research, through the Institute of Thermal Hydraulics.
References
Elgholbashi, S., Abou-Arab, T., 1983. Two-equation turbulence model for two-phase
ows. Phys. Fluids 26, 931938.

T.R. Smith et al. / International Journal of Heat and Fluid Flow 33 (2012) 156167
Hibiki, T., Ishii, M., Xiao, Z., 2001. Axial interfacial area transport of vertical upward
two-phase ow. Int. J. Heat Mass Transfer 44, 18691888.
Hibiki, T., Ishii, M., 2007. Lift force in bubbly ow systems. Chem. Eng. Sci. 62, 6457
6476.
Ishii, M., Zuber, N., 1979. Drag coefcient and relative velocity in bubbly, droplet or
particulate ows. AIChE Journal 25, 843855.
Ishii, M., Revankar, S.T., Leonardi, T., Dowlati, R., Bertodano, M.L., Babelli, I., Wang,
W., Pokharna, H., Ransom, V.H., Viskanta, R., Han, J.T., 1998. The three-level
scaling approach with application to the Purdue University Multi-Dimensional
Integral Test Assembly (PUMA). Nucl. Eng. Des. 186, 177211.
Ishii, M., Hibiki, T., 2010. Thermo-uid dynamics of two-phase ow, second ed.
Springer, New York.
Kataoka, I., Ishii, M., 1987. Drift-ux model for large diameter pipe and new
correlation for pool void fraction. Int. J. Heat Mass Transfer 30, 19271939.
Kataoka, I., Besnard, D., Serizawa, A., 1992. Basic equation of turbulence and
modeling of interfacial transfer terms in gasliquid two-phase ow. Chem. Eng.
Commun., 221236.
Kataoka, I., Serizawa, A., 1995. Modeling and prediction of turbulence in bubbly
two-phase ow. In: Proc. 2nd Int. Conference on Multiphase Flow. April 37, pp.
1116.
Kim, S., Fu, X.Y., Wang, X., Ishii, M., 2000. Development of the miniaturized four
sensor conductivity probe and the signal processing scheme. Int. J. Heat Mass
Transfer 43 (22), 41014118.

167

Mi, Y., Ishii, M., Tsoukalas, L.H., 1998. Vertical two-phase ow identication using
advanced instrumentation and neural networks. Nucl. Eng. Des. 184, 409420.
Neal, L.G., Bankoff, S.G., 1963. A high resolution resistivity probe for determination
of local void properties in gasliquid ow. AIChE J. 9, 490494.
Ohnuki, A., Akimoto, H., 1998. Prediction of phase distribution under bubbly ow in
a large vertical pipe by multidimensional two-uid model. In: Third Int.
Conference on Multiphase Flow, ICMF 98, Lyon, France. June 812, 1998.
Prasser, H.M., 2007. Evolution of interfacial area concentration in a vertical air
water ow measured by wire-mesh sensors. Nucl. Eng. Des. 237, 16081617.
Schlegel, J.P., Sawant, P., Paranjape, S., Ozar, B., Hibiki, T., Ishii, M., 2009. Void
fraction and ow regime in adiabatic upward two-phase ow in large diameter
pipes. Nucl. Eng. Des. 238, 28642874.
Shen, X., Saito, Y., Mishima, K., Nakamura, H., 2006. A study on the characteristics of
upward airwater two-phase ow in a large diameter pipe. Exp. Therm. Fluid
Sci. 31, 2136.
Shoukri, M., Hassan, I., Gerges, I., 2003. Two-phase bubbly ow structure in largediameter vertical pipes. Can. J. Chem. Eng. 81, 205211.
Sun, X., Smith, T.R., Kim, S., Ishii, M., Uhle, J., 2002. Interfacial area of bubbly ow in
a relatively large diameter pipe. Exp. Therm. Fluid Sci. 27, 97109.
Sun, X., Smith, T.R., Kim, S., Ishii, M., Uhle, J., 2003. Interfacial structure of airwater
two-phase ow in a relatively large pipe. Exp. Fluids 34, 206219.