International Journal of Multiphase Flow 47 (2012) 1–16

Contents lists available at SciVerse ScienceDirect

International Journal of Multiphase Flow

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / i j m u l fl o w

Mechanistic modeling of interfacial area transport in large diameter pipes

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

a r t i c l e i n f o a b s t r a c t

Article history: Flow in large pipes is important in a wide variety of applications. In the nuclear industry in particular,
Received 15 February 2012 understanding of flow in large diameter pipes is essential in predicting the behavior of reactor systems.
Received in revised form 12 June 2012 This is especially true of natural circulation Boiling Water Reactor (BWR) designs, where a large-diameter
Accepted 15 June 2012
chimney above the core provides the gravity head to drive circulation of the coolant through the reactor.
Available online 23 June 2012
The behavior of such reactors during transients and during normal operation will be predicted using
advanced thermal-hydraulics analysis codes utilizing the two-fluid model. Essential to accurate two-fluid
Keywords:
model calculations is reliable and accurate computation of the interfacial transfer terms. These interfacial
Large diameter
Interfacial area transport
transfer terms can be expressed as the product of one term describing the potential driving the transfer
Void fraction and a second term describing the available surface area for transfer, or interfacial area concentration. Cur-
Coalescence rently, the interfacial area is predicted using flow regime-dependent semi-empirical correlations; how-
Breakup ever the interfacial area concentration is best computed through the use of the one-dimensional
interfacial area transport equation (IATE). To be useful in practical applications the IATE requires mech-
anistic models for the change in interfacial area concentration due to interactions between bubbles.
Therefore using previously collected data and proper scaling methods, current models for bubble inter-
actions for the IATE are briefly discussed and new models applicable to large pipes are developed and
compared with the existing data.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction cap-shaped bubbles. The maximum size of a stable Taylor bubble

is (Kataoka and Ishii, 1987)
Two-phase flows are an important component in a wide variety rffiffiffiffiffiffiffiffiffiffi
of industrial applications. In the nuclear industry, the ability to r
Db;max ¼ 40 ð1Þ
understand the mechanics of two-phase flows under a wide range g Dq
of condition is essential for the prediction of steady state behavior
in Boiling Water Reactor (BWR) systems and accident scenarios in which corresponds to approximately 0.10 m for air and water at
many types of reactors including Pressurized Water Reactors atmospheric conditions or 0.07 m for steam and water under typical
(PWRs) and liquid-metal-cooled reactors. The prediction of all of BWR conditions. In larger pipes, Taylor bubbles formed with diame-
these systems is very sensitive to the accuracy of the models used, ters larger than this value have the upper surface collapse, resulting
but for steady-state behavior in new natural-circulation BWR de- in the large bubble breaking into two or more smaller Taylor bub-
signs accurate models are essential for prediction of the natural cir- bles. The fact that stable gas slugs can no longer exist results in many
culation flow rate and liquid inventory in the Reactor Pressure changes to the hydrodynamics of the flow. The added mobility and
Vessel (RPV). These quantities are both significantly influenced smaller size of the Taylor bubbles within the pipe results in in-
by the predicted void fraction in the region above the reactor, as creased bubble-induced turbulence in large diameter systems. This
this section provides the gravity head to drive natural circulation enhances turbulent diffusion of small bubbles and the passage of
and contains a significant portion of the liquid volume in the large bubbles, due to their relatively high relative velocity, can pro-
RPV. This region behaves as a large diameter channel, as do many duce strong secondary recirculating flow patterns. This behavior is
other components in reactor systems. especially prevalent at low volumetric fluxes and result in large dif-
Large diameter channels differ from small diameter systems in ferences in the physical mechanisms governing the transport of the
that stable Taylor bubbles which occupy the entire pipe diameter gas, liquid, and interface. This means that models developed and ap-
cannot be sustained due to instability in the upper surface of the plied to two-phase flows in small diameter channels may not be
used to accurately predict flows in large diameter channels.
Typically, such two-phase flows are predicted using the two-
⇑ Corresponding author. fluid model. This is a general model that utilizes balance equations
E-mail address: jschlege@purdue.edu (J.P. Schlegel). for the mass, momentum and energy of each phase. For this reason

0301-9322/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijmultiphaseflow.2012.06.009
2 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16

it is also often called the six-equation model. Short of DNS or LES the particle density distribution function. This distribution func-
the two-fluid model is the most accurate model for two-phase flow tion describes the probably number density of fluid particles with
prediction. It also requires much shorter computational time than given properties at a given time and location. It is assumed that the
either of these prediction methods. Further, the two-fluid model equation is continuous. This results in the equation
has an advantage in that it can be derived in a one-dimensional   X
@f @ dV
form which is easy to use in thermal-hydraulic analysis codes. vpÞ þ
þ r  ðf ~ f ¼ Sj þ Sph ð2Þ
@t @V dt
The major disadvantage of the two-fluid model is in modeling j

the interfacial transfer terms which arise due to time-averaging

which is the basis for the derivations of the transport equations for
of the instantaneous transport equations and represent the ex-
fluid particle number n, the volume fraction a, and the interfacial
change of mass, momentum and energy between the phases (Ishii
area concentration ai. Here f is the bubble number density function,
and Hibiki, 2010).
t is time, vp is the particle velocity, V is the particle volume, Sj is the
These interfacial transfer terms are often expressed as the prod-
particle number density change due to mechanism j and Sph is the
uct of the driving potential for a given type of transfer, for example
particle number density due to phase change.
the drag coefficient and relative velocity in interfacial momentum
For applications in two-phase flows, Eq. (2) is too detailed to be
transfer, and a term representing the geometry of the interface.
practical. Therefore it is integrated over the entire range of possible
Current advanced system analysis codes typically use a two-step
particle volumes, from Vmin to Vmax. With appropriate definitions,
method in which an experimentally determined flow regime map
this gives the bubble number density transport equation. The void
is used to determine the flow geometry, then empirical correla-
fraction, a, is one of the most important parameters used in
tions are used to predict the interfacial transfer. Such flow regime
describing two-phase flows and is defined as the fraction of the
maps have already been developed for large diameter channels
volume occupied by gas particles. This void fraction can be ob-
(Schlegel et al., 2009), and constitutive models for each flow re-
tained from the bubble number density transport equation by mul-
gime have been developed and recommended (Hibiki and Ishii,
tiplying by the particle volume and integrating over the entire
2003). This approach has two significant difficulties however. The
particle volume range. The details of this derivation can be found
first is that uncertainties exist in the flow regime boundaries due
in the work of Ishii and Hibiki (2010), however the net result is
to the continuous nature of two-phase flows. The second is the er-
Eq. (3):
ror that exists in the predictions by the empirical models used,  
which are typically developed under steady-state conditions and @a a ðCg  gph qg Þ dqg
þ r  ða~vgÞ  
may not be readily applicable to highly transient scenarios such @t qg a dt
as accident analysis. Z V max "X #
X
In order to mitigate these effects, it has been proposed that an ¼ Sj þ Sph VdV ¼ gj þ gph ð3Þ
interfacial area transport equation be utilized to predict the inter- V min j j
facial geometry. In this approach, a transport equation is used to
Here, the first two terms on the left-hand side of Eq. (3) represent
predict the change in the amount of interfacial area per unit vol-
the total change in void fraction, while the third term represents
ume of the mixture, or interfacial area concentration, in the two-
the volume change due to change in density caused by change in
phase system. This is done by accounting for the interactions of
pressure. In this equation vg is the gas velocity, qg is the gas density,
bubbles with each other and the continuous phase as well as mod-
eling the expansion and compression of bubbles due to changes in
Cg is the mass source due to phase change, gph is the volume source
due to phase change and gj is the volume source or sink due to
pressure. First proposed by Kocamustafaogullari and Ishii (1995),
mechanism j. The right-hand side represents the change in void
the approach has been refined for one-group approaches including
fraction due to particle interactions and phase change.
only spherical or distorted bubbles (Wu et al., 1998) as well as two-
The one-group interfacial area transport equation is obtained in
group approaches which also account for the behavior of Taylor
a similar manner as the previous equation, multiplying the number
bubbles (Fu and Ishii, 2003a,b). This method eliminates the need
density transport equation by the particle surface area Ai(V), inte-
for empirical flow regime maps, reducing numerical oscillations
grating and simplifying. To close the IATE, the source and sink rates
and bifurcations. In addition the transport equation is dynamic,
for the void fraction and interfacial area concentration must be
allowing transient conditions to be predicted well given appropri-
specified using constitutive relations. In this process, the bubble
ate boundary conditions and bubble interaction models. Further,
number sources and sinks Rj should be modeled separately based
the use of the IATE allows accurate calculation of the average bub-
on the type of particle interaction. Combining these equations
ble size. This can further improve the predictive ability of advanced
and rearranging gives
analysis codes.
  X
Currently reasonable bubble interaction models have been @ai 2 ai  @ a
developed for small diameter pipes (Wu et al., 1998; Fu and Ishii,
v iÞ ¼
þ r  ðai~ v g  gph þ /j þ /ph
þ r  a~ ð4Þ
@t 3 a @t j
2003a,b) and for narrow rectangular channels (Sun et al., 2002),
however the applicability of these models to large diameter chan- Here, the first two terms on the left side of Eq. (4) represent the
nels may be somewhat questionable due to the hydrodynamic dif- total change in interfacial area concentration. The third term
ferences. It is therefore necessary to develop and test bubble represents the change in ai due to change in pressure, while the
interaction models that accurately account for the unique features right-hand side represents the change in ai due to coalescence
present in large diameter channels. and breakup mechanisms and phase change. Here vi is the interfa-
cial velocity, /j is the interfacial area concentration source or sink
due to mechanism j and /ph is the interfacial area concentration
2. Previous work source or sink due to phase change.

2.1. Formulation of the IATE 2.1.2. Two-group formulation

For cap-bubbly, slug, or churn-turbulent flow patterns a wide
2.1.1. One-group formulation range of bubbles which vary significantly in both size and shape
The foundation of the IATE is the Boltzmann transport equation, can exist. In this case, the model must account for differences in
an integro-differential equation describing particle transport using the transport characteristics of the various types of bubbles. In
 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16 3

most two-phase flow conditions, bubbles can be categorizes as distribution is weighted to larger bubbles the value will be higher.
spherical, distorted, cap, Taylor or churn-turbulent. Based on their The value of this constant has been shown through previous exper-
drag and transport characteristics these five categories can be iments (Fu and Ishii, 2003a,b) to have a value of between 0 and 2.
reclassified into two groups, with Group 1 consisting of spherical Additionally, Dc is the critical bubble size between groups as given
and distorted bubbles and Group 2 consisting of cap, Taylor and in Eq. (5) and Dm12 is the mass transfer between groups due to bub-
churn-turbulent bubbles. Using this categorization, Fu and Ishii ble coalescence and breakup.
(2003a) developed a two-group approach to the interfacial area
transport equation. 2.1.3. Major coalescence and breakup mechanisms
In the two-group formulation the integration limit for each To close the transport equations, the fluid particle interaction
transport equation, that is the boundary between the two bubble mechanisms must be identified and modeled. These mechanisms
groups Dd,max, must be determined. This is the boundary between have been analyzed and identified by Fu and Ishii (2003a) and
distorted and cap bubbles, which is defined by Ishii and Zuber can be summarized as
(1979) as
rffiffiffiffiffiffiffiffiffiffi 1. Random Collision (RC): collision driven by turbulent eddies,
r
Dd;max ¼ 4  1 cm ð5Þ 2. Wake Entrainment (WE): collision due to acceleration of the
g Dq
following particle in the wake region of the preceding particle,
where g is the gravitational acceleration and Dq is the density dif- 3. Turbulent Impact (TI): disintegration upon impact of turbulent
ference between phases. Bubbles larger than this size tend to take eddies,
on a cap shape, and the drag properties of the bubble begin to 4. Shearing-Off (SO): shearing off of small bubbles around the base
change. This then gives us a bubble volume at the group boundary rim of the cap bubbles.
for use as the integration limit. Some simplifying assumptions are 5. Surface Instability (SI): break-up of a large cap bubble due to
made regarding the distribution function for each bubble group, surface instability at the interface.
namely that the bubbly number distribution is independent of bub-
bly volume so that they can be reasonably described by a group Based on these major mechanisms, the possible source and sink
average. The resulting integration of the Boltzmann transport equa- terms are given in Table 1 and illustrated in Fig. 1. These include
tion results in two-group transport equations for the bubble num- intra-group processes and inter-group processes. The intra-group
ber density, void fraction: processes for Group 1 bubbles are largely similar for many geome-
  try types and have been established by Wu et al. (1998) however
1 @ a1 qg mechanistic modeling of Group 2 bubble behavior and inter-group
v g1 Þ  Cg1 þ Dm12 þ gph
þ r  ða1 qg ~
qg @t transport mechanisms must be performed with careful consider-
 3   X ation of the effect of channel geometry on Group 2 bubble behavior.
Dc @ a1
¼ C þ r  ða1~v g1 Þ  gph þ gj1 þ gph ð6Þ
Dsm1 @t j
2.2. Other coalescence and breakup models
 
1 @ a2 qg The IATE is probably the best-developed method for predicting
þ r  ða2 qg ~v g2 Þ  Cg2 þ Dm12
qg @t interfacial area concentration currently being used in two-phase
 3   X flow computation, however many other modeling efforts have at-
Dc @ a1
¼C þ r  ða1 ~v g1 Þ  gph þ gj2 ð7Þ tempted to account for the effects of bubbly coalescence and
Dsm1 @t j breakup. Millies et al., 1996 developed a first-order relaxation
and interfacial area concentration: model by modeling the population balance for bubbly flows in
  similar fashion to the one-group IATE, performing a perturbation
@ai1 2 a @a analysis on the bubble coalescence and breakup kernels and retain-
v i1 Þ ¼ i1 1 þ r  ða1~
þ r  ðai1~ v g1 Þ  gph
@t 3 a1 @t ing only first-order terms. Later Millies and Mewes (1999) used a
 3   similar approach to analyze the dependence of bubble breakup
Dc ai1 @ a1
C þ r  ða1 ~ v g1 Þ  gph and coalescence on the dissipation rate and existence of impurities
Dsm1 a1 @t
X in the liquid phase. In 2001, Lehr and Mewes modified these pop-
þ /j1 þ /ph ð8Þ ulation balance equations and developed improved bubble coales-
j
cence and breakup kernels for application to bubble columns.
   3 These kernels are quite complex and have been tested using
@ai2 2 a @a D commercial CFD codes. Other researchers (Cheung et al., 2007; Sari
v i2 Þ ¼ i2 2 þ r  ða2~
þ r  ðai2~ v g2 Þ þ C c
@t 3 a2 @t Dsm1 et al., 2009) have also developed one-group approaches using the
  X
ai1 @ a1 population balance method.
 v g1 Þ  gph þ /j2
þ r  ða1~ ð9Þ
Yao and Morel (2004) developed additional models for the one-
a1 @t j
group IATE including for random collision, turbulent-induced
Here, the subscripts 1 and 2 represent the values for small spherical breakup and wall nucleation. While this is not a two-group IATE
distorted bubbles (Group 1) and large cap bubbles (Group 2) respec- study and is limited to bubbly flows, its efficacy in predicting wall
tively. In addition, Dm12 represents the mass transfer between bub- nucleation may merit further investigation.
ble groups due to inter-group transport mechanisms, Dc is the Several other researchers have developed beakup and coales-
critical bubble size at the group boundary equal to Dd,max and Dsm cence kernels for the MuSiG (Multiple Size Group) model (Huh
is the Sauter mean diameter. These equations are similar in form et al., 2006; Liao et al., 2011) using the population balance ap-
to the one-group equations, however there are some differences. proach. These models are typically developed for CFD codes. Cur-
Most notably the constant C in the void transport and interfacial rently these models do not predict the behavior of large bubbles
area transport equations. This constant affects the transfer of bub- (Group 2 bubbles in the two-group IATE analysis) very well and
bles across the group boundary by expansion or contraction and may be limited to lower void fraction flows. Additionally the com-
rises from the effect of the bubbly size distribution. For uniform putational cost increases greatly as the number of bubble groups is
bubble size distributions the value should be unity, while if the increased. These may limit the applicability of these methods.
4 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16

Table 1
IATE sources and sinks.

Notation Mechanism Contribution Effect on ai

ð1Þ
RRC Random collision (1) + (1)) = (1) Sink in Group 1
ð11;2Þ
RRC Random collision (1) + (1)) = (2) Sink in Group 1 Source in Group 2
ð12;2Þ
RRC Random collision (1) + (2)) = (2) Sink in Group 1 Source in Group 2
ð2Þ
RRC Random collision (2) + (2)) = (2) Sink in Group 2
ð1Þ
RWE Wake entrainment (1) + (1)) = (1) Sink in Group 1
ð11;2Þ
RWE Wake entrainment (1) + (1)) = (2) Sink in Group 1 Source in Group 2
ð12;2Þ
RWE Wake entrainment (1) + (2)) = (2) Sink in Group 1 Source in Group 2
ð2Þ
RWE Wake entrainment (2) + (2)) = (2) Sink in Group 2
ð1Þ
RTI Turbulent impact (1)) = (1) + (1) Source in Group 1
ð2;11Þ
RTI Turbulent impact (2)) = (1) + (1) Source in Group 1 Sink in Group 2
ð2;12Þ
RTI Turbulent impact (2)) = (2) + (1) Source in Group 1 Sink in Group 2
ð2Þ
RTI Turbulent impact (2)) = (2) + (2) Source in Group 2
RSo Shearing-off (2)) = (2) + n  (1) Source in Group 1 Sink in Group 2
RSI Surface instability (2) = (2) + (2) Source in Group 2

Fig. 1. Source and sink mechanisms, (a) random collision, (b) wake entrainment, (c) turbulent impact, (d) shearing-off, (e) surface instability.

2.3. Characteristics of flows in large diameter channels and developed a new theoretical criterion for the transition from
cap-turbulent to churn-turbulent flow.
Two-phase flows in large diameter channels are much different Two-phase flows in small diameter channels also exhibit wall-
than flows in small diameter pipes. For flows in small diameter peaking profiles in the radial void fraction distribution. Several
pipes, slug flows in which Taylor bubbles occupy the entire studies have noted that the wall-peaking effect is much weaker
cross-section of the pipe exist. In large diameter channels however, in large diameter channels (Shen et al., 2010) and occurs only at
such bubbles are unstable and quickly collapse due to surface extremely low void fractions. This may be due to reduced velocity
instability as shown in Fig. 1. Because of this the typical flow re- gradients near the wall and increased bubble-induced turbulence
gimes noted for pipe flows are not applicable for large diameter as noted by Ohnuki and Akimoto (2001). Further, the formation
channels. This phenomenon has been confirmed by Ohnuki and of large-volume cap bubbles with no wall restrictions tends to re-
Akimoto (2000), who identified undisturbed bubbly, agitated bub- sult in increased relative velocity for Group 2 bubbles in large
bly, churn bubbly, churn slug and churn froth flow regimes in a diameter channels (Shen et al., 2010). The effect of bubble disinte-
pipe with diameter of 0.2 m. Schlegel et al. (2009) classified the gration by surface instability is much more prevalent in large
flow into bubbly, cap bubbly, churn turbulent and annular flow diameter channels, where coalescence of Group 2 bubbles can
based on experiments in 0.152 m and 0.203 m diameter facilities readily form bubbles larger than the stability limit. In addition to
 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16 5

where

k(V  V0 ,V0 ) coalescence efficiency of bubbles with volume

V  V0 and V0
h(V  V0 ,V0 ) coalescence frequency of bubbles with volume
V  V0 and V0
f(V  V0 ) number density distribution function of bubbles
with volume V  V0 or V0

For bubble breakup processes, the change in bubble number

density is given by
Z V max
0
SðVÞ ¼ bðV 0 ; VÞnðV 0 ÞgðV 0 Þf ðV 0 ÞdV ð11Þ
Fig. 2. Experimental studies regarding interfacial area transport in large diameter V
pipes. (See above-mentioned references for further information.)
where
this, shearing off of small bubbles around the skirt of a large cap
bubble may be much more significant in large diameter channels b(V0 ,V) Distribution of daughter bubbles from parent bubble
due to the larger number of cap bubbles. V0
While a more universal approach for all geometries would be n(V0 ) Number of daughter bubbles produced from parent
ideal, current modeling efforts are limited by the state of knowl- bubble V0
edge of two-phase flows. Many aspects of bubble coalescence g(V0 ) Breakup frequency of bubble with volume V0
and breakup simply cannot be measured or predicted using current
techniques. This inability to measure or model every single aspect
of the flow gives rise to the use of empirical constants to represent Also, in order to develop mechanistic models, some assumption
flow properties such as the critical film rupture thickness in bubble regarding the shape of the number density distribution function
coalescence. Additionally, many of these models are highly depen- must be made as the actual shape of the function is difficult to
dent on the turbulence characteristics while models for predicting determine. For simplicity, f is assumed to take on a constant value
the turbulent dissipation are questionable in terms of accuracy and of f1 over the Group 1 bubble region and f2 over the Group 2 bubble
do not describe every important aspect of turbulence behavior. region.
This is especially true for two-phase flows where bubble-induced
turbulence must be accounted for. All of these effects can vary sig- 3.1. Bubble coalescence mechanisms
nificantly between different geometries in addition to the differ-
ences in the physical mechanisms of bubble interaction for 3.1.1. Coalescence due to random collision
geometries of different scales. This results in the necessity for One of the major coalescence mechanisms in two-phase flows is
developing models and determining the values of the different random collision between bubbles driven by turbulent eddies. In
constants for each type of geometry. this case, only binary collisions are assumed to simplify the prob-
lem. From a transport perspective, this may result in four signifi-
2.4. Experimental data available for large diameter channels cant types of collision: (1) coalescence of two Group 1 bubbles to
form a Group 1 bubble, (2) coalescence of two Group 1 bubbles
In order to benchmark the models developed for the IATE in to form a Group 2 bubble, (3) coalescence of one Group 1 and
large diameter channels, an experimental database is required. A one Group 2 bubble and (4) coalescence of two Group 2 bubbles.
review of the literature has found several efforts to measure quan- Case (1) has already been modeled by Wu et al. (1998) for bubbly
tities relevant to interfacial area transport, which are noted in flows based on this approach and this model should be applicable
Fig. 2. Previous experiments have resulted in a great deal of data to large diameter channels, however cases (2), (3) and (4) require
for bubbly flow conditions, however for cap-bubbly and churn-tur- new models in order to be predicted accurately. These three cases
bulent flow very little data is available. Previous experiments by are modeled using Eq. (10) as a basis.
Smith et al. (2011) and Lucas et al. (2010) have collected data in For case (1) and case (2), the major driving force for collisions is
this region using the electrical conductivity probe and wire mesh turbulent velocity fluctuation in the flow. Thus the collision fre-
sensor, respectively. The data collected by Lucas et al. (2010), how- quency has been expressed by Prince and Blanch (1990). Consider-
ever, is largely intended for CFD code development and interfacial ing the cross-sectional area of Group 1 bubbles, assuming isotropic
area concentration measurements are not reported. Thus the mod- turbulence (Batchelor, 1951), accounting for the spatial probability
els developed will be compared and benchmarked against the data of collision and assuming that (V  V0 ) > V0 , this can be simplified to
collected by Smith et al. (2011). " !#  2=3
a1=3 1=3
1;max a1 a1
hðV  V 0 ; V 0 Þ  1:84 1  exp C RC1 e1=3
3. Mechanistic models for interfacial area sources and sinks 1=3 1=3
ða1;max  a1 Þ a1;max
0 7=9
ðV  V Þ ð11Þ
Due to the complexity of bubble interactions, Kocamustafaogul-
lari and Ishii (1995) have developed a basic formulation for model- where CRC1 is a proportionality constant related to the turbulent
ing coalescence and breakup processes. According to this eddy size (Wu et al., 1998) and a1,max is the maximum possible
formulation, the change in bubble number density for coalescence Group 1 void fraction. The efficiency of the bubble collision also de-
processes is given by pends on the spatial and temporal collision probabilities. Based on
Z V=2 the work of Coulaloglou and Tavlarides (1977) and accounting for
0
SðVÞ ¼ kðV  V 0 ; V 0 ÞhðV  V 0 ; V 0 Þf ðV  V 0 Þf ðV 0 ÞdV ð10Þ the time required for coalescence and bubble contact time in turbu-
V min
lent flows, this efficiency can be written as
6 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16

!
5=6
DSm qf
1=2 1=3
e where Dc1 is given by
kRC ¼ exp C RC0 ð12Þ
r1=2 Dc
Dc1 ¼ : ð19Þ
Dsm1
where CRC0 is a coefficient accounting for the initial film thickness
and critical film thickness for coalescence (Wu et al., 1998). Further, Case (3) describes the collision of one bubble from each group.
qf is the liquid density, e is the turbulent dissipation, and r is the In this case the collision cross-sectional area is dominated by the
surface tension. Substituting Eqs. (11) and (12) into Eq. (10) and Group 2 bubble, but the turbulent fluctuating velocity u2t (V0 ) for
integrating results in the Group 2 bubble is given by
" !#  
ð1Þ a1=3 1=3
1;max a1 1=3 2 a1 2=3 16=9  2t ðV 0 Þ  2ðeDb2 Þ2=3  2:69e2=3 ðV  V 0 Þ2=9
u ð20Þ
SðVÞ  0:75kRC 1  exp C RC1 1=3 e f 1 V
ða1;max  a1=3
1 Þ
a1;max
where Db2 is the diameter of the Group 2 bubble. Accounting for the
ð13Þ random nature of bubble motion and assuming that the coalescence
When two Group 1 bubbles coalesce the resulting bubble may efficiency is driven mainly by the effect of turbulence on Group 1
not always be in Group 1, which results in the two different cases bubbles, the loss rate can be expressed as
(1) and (2). The total sink term in interfacial area concentration for Z V m1
0
Group 1 is determined by the difference in the loss rate of Group 1 SðVÞ ¼ kðV  V 0 ; V 0 ÞhðV  V 0 ; V 0 Þf ðV  V 0 Þf ðV 0 ÞdV
V min
bubbles which coalesce and the gain rate of Group 1 bubbles pro- " !#
1=3
duced by coalescence as well as by the average surface area of the ð12;2Þ a1;max a1=3
1
 3:15kRC 1  exp C RC1 1=3
lost and gained bubbles. Thus 1=3
ða1;max  a1 Þ
Z 2=3
 e1=3 f1 f2 ða1 a2 ÞV m1 V 7=9
ð1Þ
2V m1
ð1Þ ð1Þ ð1Þ e1=3 ai111=3 ð21Þ
RRC ¼ SRC ðVÞdV  0:0044C RC kRC
V min a1 a2=3
1;max Then the coalescence frequency can be expressed as
" 1=3
!#
a1;max a1=3
1
Z V m1 þV m2
 1  exp C RC1 1=3 ð14Þ ð12;2Þ
RRC ¼ SRC
ðVÞdV
ð12;2Þ
ða1;max  a11=3 Þ V c þV min
8   
>
< 1:77kð12;2Þ 1  exp C RC1 a1=3 a1=3
by assuming that f1 is independent of V and that Vmin is much smal- RC
1;max 1
1=3 1=3
ða1;max a1 Þ
e1=3 f1 f2 ða2=3
1 a2 Þ
ð1Þ
ler than Vm1, the bubble volume at the group boundary. Here C RC ac- ¼
>
:
counts for the distortions introduced by the simplifying V m1 ½ðV m2 þ V m1 Þ16=9  ðV c þ V min Þ16=9 
assumptions and the behavior of the bubble size distribution (Wu ð22Þ
et al., 1998). Additionally, the superscript (1) indicates that this
mechanism involves only Group 1 bubbles. The net loss in interfa- where the superscript (12,2) indicates that a Group 1 and Group 2
cial area concentration is then determined using bubble coalesce to form a Group 2 bubble. By combining Eq. (21)
Z 2V m1 with Eq. (15), the total loss rate for Group 1 bubbles is derived as
ð1Þ ð1Þ
/RC ¼ SRC ðVÞAi dVðVÞ ð15Þ
2V min
ð12;2Þ
/RC;1 ¼ 1:14C RC e a2=3 4=3 2=3
ð12;2Þ ð2Þ 1=3
1 a2 ai1 ai2
kRC
" 1=3 1=3
!#
Accounting for the loss rate of bubbles entering collisions and a1;max a1
 1  exp C RC1 1=3 ð23Þ
the Group 1 bubbles produced from those collisions results in a to- ða1;max  a1=3
1 Þ
tal loss rate of
where
ð1Þ ð1Þ ð1Þ e1=3 a1 ai15=3
/RC  0:17C RC kRC   5=6 1=2 1=3
!
1=3
a1;max a1=3
1;max  a1
1=3
ð2Þ
DSm2 qf e
" !# kRC ¼ exp C RC0 ð24Þ
r1=2
a1=3 1=3
1;max a1
 1  exp C RC1 1=3 : ð16Þ
ða1;max  a1=3
1 Þ
And the net gain rate for Group 2 becomes

The gain rate for ai2 due to collision of two Group 1 bubbles is
ð12;2Þ
/RC;2 ¼ 1:80C RC e a15=3 a1=3
kRC2 ai2
5=3
ð12;2Þ ð12;2Þ 1=3

derived in similar fashion. The gain rate of Group 2 bubbles pro- " !#
a1=3
1;max a1
1=3
duced from the collision of two Group 1 bubbles can be given by  1  exp C RC1 1=3 ð25Þ
Z "Z " !# ða1;max  a1=31 Þ
1=3 1=3
ð11;2Þ
2V m1 V=2
ð1Þ a a
1;max 1
RRC ¼ 1:84kRC 1  exp C RC1 1=3 1=3
ð12;2Þ
With C RC once again an adjustable constant to be determined
V min VV m1 ða
1;max  1 Þ a
# based on experimental data but which should have a similar value
 2=3 ð1Þ
a1 0 to C RC .
 e1=3 f12 ðV  V 0 ÞdV dV ð17Þ
a1;max The last case, (4), gives the random collision of two Group 2
bubbles. By accounting for the cross-sectional area of the bubbles
where the superscript (11,2) indicates that two Group 1 bubbles be- and the turbulent fluctuations in their velocity, the probability of
ð2Þ
come a single Group 2 bubble by coalescence. By combining this collision P RC can be found to be
with Eq. (15) and performing the resulting integrations, the source  2
rate of Group 2 bubbles can be found to be Db2 ðV  V 0 Þ þ Db2 ðV 0 Þ 2:44
¼ 2 ½ðV  V 0 Þ1=3 þ ðV 0 Þ1=3 2
ð2Þ
PRC ¼
" !# Dh Dh
ð11;2Þ ð1Þ ð1Þ e1=3 a1 a5=3
i1
a1=3 1=3
1;max a1 ð26Þ
/RC;2 ¼ 4:1C RC kRC 2=3
1  exp C RC1
a1;max ða1=3 1=3
1;max  a1 Þ
  where Dh is the hydraulic diameter and the superscript (2) indicates
2
 1  Dc1 ð18Þ that only Group 2 bubbles are involved. Also, the collision efficiency
3
can be determined by assuming that only turbulent eddies with
 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16 7

similar size to the Group 2 bubbles have a significant influence. This Using Eqs. (33) and (15) then allows the determination of the
results in the bubble number density sink of net interfacial area sink by accounting for the total destruction
Z and production of bubbles as (Wu et al., 1998)
V=2
ð2Þ 0
SRC ðVÞ ¼ kðV  V 0 ; V 0 ÞhðV  V 0 ; V 0 Þf ðV  V 0 Þf ðV 0 ÞdV ð1Þ 1=3
/WE ¼ 0:17C WE C D1
ð1Þ
ur1 a2i1 : ð34Þ
Vc
h  i
¼ 3:14kRC 1  exp C RC2 a1=2
ð2Þ
e1=3 In case (2), the resulting bubble belongs to Group 2. By ignoring
2
" the dependence of the relative velocity and the drag coefficient on
 22=9 #
1 V the bubble volume, then the bubble density distribution change
 2 f22 ðV  V c Þ22=9  ð27Þ becomes
Dh 2
SWE ðVÞ ¼ 1:38kWE C v f ðLw1 Þur1 C 1=3
ð1Þ ð1Þ 2 5=3
D1 f1 V ð35Þ
By combining Eq. (27) with Eq. (15) and accounting for the max-
imum stable bubble size in the resulting integration, the total The loss rate of Group 1 bubbles by this mechanism has already
interfacial area concentration sink for collisions of Group 2 bubbles been accounted for in Eq. (34). By combining this equation with Eq.
can be found to be (15) and using the correct limits of integration, then the production
rate of Group 2 interfacial area concentration is found to be
a7=3 1 h i 
 
1  expðC RC2 a1=2
ð2Þ ð2Þ ð2Þ
/RC ¼ 95:7C RC kRC e1=3 2
2 Þ 1  0:37D3
c2 2 
D2h a1=3
i2 /WE;2 ¼ 2:57C WE C 1=3
ð11;2Þ ð11;2Þ 2
D1 ur1 ai1 1  Dc1 ð36Þ
3
ð28Þ
ð2Þ In case (3), the preceding bubble is a Group 2 bubble. Account-
where once again C RC is an adjustable constant which should have
ð1Þ ing for the cross-section area of the bubbles and the local relative
similar value to C RC and
wake velocity uw2 and cap bubble relative velocity,
Dc  1=3
Dc2 ¼ : ð29Þ C D2 Ac
DSm2 uw2  ur2 and ð37Þ
z2
pffiffiffiffiffiffiffiffi
ur2  0:35 gDb ð38Þ
3.1.2. Coalescence due to wake entrainment
Entrainment of one or more bubbles in the wake of a preceding The average velocity of bubbles in the wake region can be de-
bubble is also a major coalescence mechanism. When bubbles en- rived as
ter the wake region of the preceding bubble they accelerate and Z Lw  1=3
collide with the preceding bubble. As with random collision there 1 C D2 Ac
 rw2 
u ur2 dz ¼ 0:94ur2 C 1=3
D2 ð39Þ
are four possible cases for coalescence by wake entrainment: (1) Lw2 0 z2
coalescence of two Group 1 bubbles to form a Group 1 bubble, If the dependence of the drag coefficient and relative velocity on
(2) coalescence of two Group 1 bubbles to form a Group 2 bubble, the bubble volume is ignored, then the collision frequency can be
(3) coalescence of a Group 1 and Group 2 bubble and (4) coales- found to be
cence of two Group 2 bubbles.
1=3
For case (1), a cylindrical wake region is assumed to form for a  rw2  1:91C D2 ur2 ðV  V 0 Þ2=3
hðV  V 0 ; V 0 Þ ¼ Ac ðV  V 0 ; V 0 Þu ð40Þ
distance Lw behind the preceding bubble. Then the cross-section
When the entrained bubble belongs to Group 1, then the local
for wake entrainment can be determined based on the bubble vol-
wake velocity of the bubble is given as
umes. The average wake velocity of the trailing bubble can be
approximated from the relative velocity between the liquid in u  rw2 þ ur1  ur2
 w12 ¼ u ð41Þ
the wake and the surrounding liquid as
Then the number density change is given by
 
Lw 1=3 Z V m1
 rw1  2:77f
u ur1 C D1 ð30Þ
Db =2 ð12;2Þ
SWE ðVÞ ¼
ð12;2Þ
1:91kWE u  w12 f1 f2 ðV  V 0 Þ2=3 dV 0
V min
where CD1 is the drag coefficient for the leading Group 1 bubble and ð12;2Þ
 1:91C WE u w12 f1 f2 V m1 V 2=3 ð42Þ
ur1 is the relative velocity of the preceding Group 1 bubble,
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where
ð12;2Þ
C WE
accounts for the proportionality constants and distor-
gDb Dq
ur1 ¼ ð31Þ tions due to the bubble volume distribution. By combining Eqs. (42)
3C D1 qf and (15) and accounting for the proper integration limits, the net
sink of interfacial area in Group 1 and net source of interfacial area
Based on previous studies the ratio of the wake length to bubble
in Group 2 can be derived as
radius ranges from 2 to 8 (Katz and Meneveau, 1966; Tsuchiya
et al., 1989) (Wu et al., 1998). ð12;2Þ ð12;2Þ
/WE;l1 ¼ 0:33C WE u w12 ai1 ai2 ð43Þ
Using Eqs. (30) and (31), the collision frequency can be ex-
ð12;2Þ ð12;2Þ a2i2
pressed as /WE;g2 ¼ 0:922C WE u  w12 a1 ð44Þ
 1=21 a2
11=84 1  a
hðV  V 0 ; V 0 Þ ¼ 1:81C v f Lw1 g V 6=7 ð32Þ When the entrained bubble belongs to Group 2, the bubble
vf number density change becomes
where Cv is a covariance term with a value of approximately 1, Lw1 is Z V=2
0
the non-dimensional wake length and vf is the liquid velocity. Based
ð2Þ
SWE ðVÞ ¼
ð2Þ
 rw2 f22 ðV  V0Þ2=3 dV
1:91C WE P WE u
ð2Þ

Vv
on this equation, the change in the bubble number density can be
ð2Þ ð2Þ
derived as  rw2 f22 ½V 5=3  2:43V c V 2=3 
¼ 0:785C WE PWE u ð45Þ
 1=21 where
ð1Þ ð1Þ 1a
SWE ðVÞ ¼ 0:71kWE f ðLw1 Þg 11=84 f12 V 13=7 ð33Þ
vf ð2Þ
PWE ¼ 1  expð0:7a2 Þ ð46Þ
8 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16

is the probability of wake entrainment. For each leading bubble, the Accounting for the fact that only eddies with certain sizes can
probability of wake entrainment depends on the number density of cause breakup, combining Eq. (50) with Eq. (15), and including
Group 2 bubbles in the wake region. In smaller pipes, this eventu- both the losses due to bubble destruction and gains due to bubble
ally leads to stable slug flow. Thus, the probability of entrainment creation gives the total interfacial area source for Group 1 as (Wu
decreases with axial distance. In the large pipe case there is no sta- et al., 1998)
ble slug flow and wake entrainment may continue indefinitely. !  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Therefore the distribution of Group 2 bubbles can be represented ð1Þ ð1Þ a5=3 Wecr1 Wecr1
/TI ¼ 0:12C TI 1=3 e i1
ð1  atot Þ exp  1 :
as a Poisson process. Using Eqs. (45) and (15) results in the net loss a12=3 We1 We1
rate of Group 2 interfacial area due to wake entrainment as
ð52Þ
ð2Þ ð2Þ a2i2   ð1Þ
/WE ¼ 1:02C WE ½1  expð0:7a2 Þu
 rw2 1 0:10D2
c2 ð47Þ where C TI
accounts for the proportionality constant Ck as well as
a2 the bubble volume distribution.
Breakup of Group 2 bubbles is complicated by the fact that
3.2. Bubble breakup mechanisms when a large bubble disintegrates, multiple daughter bubbles
may form which can be located in the Group 1 or Group 2 regions.
3.2.1. Disintegration due to turbulent impact For the current analysis, binary breakup is assumed with uniform
Turbulence is one of the most significant mechanisms causing bubble size distribution. This allows computation of the collision
bubble breakup. Eddies that contribute to bubble breakup are as- cross-sectional area between Group 2 bubbles and eddies. Thus,
sumed to be of approximately the same size as the bubble; larger in similar fashion to the Group 1 analysis the collision frequency
eddies simply carry along smaller bubbles without causing break- between eddies and Group 2 bubbles is
up while smaller eddies do not have sufficient energy to induce
hðV 0 ; Ded Þ  4:45e1=3 V 07=9 ð53Þ
disintegration. Three possible cases are analyzed here, (1) breakup
of a Group 1 bubble, (2) breakup of a Group 2 bubble to form Group And the bubble number density change is given by
1 bubbles and (3) breakup of a Group 2 bubble to form Group 2 Z Db2
ð2Þ ð2Þ 1=3
bubbles. STI ðV 0 Þ ¼ hðV 0 ;Ded ÞFf ðV 0 Þfed ðDed ÞdDed  4:025C TI e f2 ð1 atot Þ
To determine the bubble breakup rate, it is necessary to deter- 0:5Db2

mine the collision frequency between a bubble and an eddy with  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Wecr2 Wecr2 02=9
length scale near the bubble size. By assuming that turbulence is exp  1 V ð54Þ
We2 We2
isotropic and that the eddy sizes of interest are in the inertial sub-
range, the concentration of eddies as a function of wave number With the Weber numbers now defined in terms of Group 2
can be given by Azbel and Liapis (1983). Since the number of ed- bubble properties,
dies becomes infinitely large as the eddy size approaches zero,
50% of the bubble size is adopted as the minimum eddy size that
qf u 2t2 Dsm2
We2 ¼ ð55Þ
can cause disintegration as eddies smaller than this do not carry r
sufficient energy (Wu et al., 1998). If Group 1 bubbles are produced by the breakup, then the
Because not all bubble-eddy collisions results in breakup, a col- change rate of the Group 1 bubble number density can be written
lision efficiency must be associated with the process. This is done as
by balancing the cohesive and disruptive forces to yield a critical Z Vc
Weber number,
bðV 0 ; VÞnðV 0 ÞSTI ðV 0 ÞdV  8:05C 3k e1=3 f2 ð1  atot Þ
ð2;1Þ ð2Þ
STI ðV 0 Þ ¼
V min
qf u 2tc Db  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Wecr ¼ ð48Þ
r Wecr2 Wecr2 V c
 exp  1 ð56Þ
where u 2tc is the critical eddy velocity for breakup of a bubble with We2 We2 V 02=9
diameter Db. When the Weber number is smaller than this value,
which, with appropriate volume limits, leads to the source of Group
the bubble remain stable. Then the fraction of eddies with sufficient
1 interfacial area:
energy to cause bubble disintegration is given by Coulaloglou and
Z "Z #
Tavlarides (1977). V m2 Vc
ð2;1Þ ð2Þ 0
For case (1), with the considerations above, the collision fre- /TI;1 ¼ bðV 0 ; VÞnðV 0 ÞSTI ðV 0 ÞAi ðVÞdV dV
Vc V min
quency can be expressed in terms of the turbulent velocity fluctu- !  
ations and the collision cross-section of the bubble and eddy as ð2;1Þ a5=3 Wecr2
¼ 6:165C TI 1=3 ð1 e  atot Þ i2
2=3
exp 
7=3 a We2
hðV 0 ; Ded Þ  2pe1=3 Db ð49Þ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2

Then the change rate in the bubble number density, assuming Wecr2  13=3

 1 0:212Dc2  0:167D5
c2 ð57Þ
binary breakup, becomes We2
Z Db   ð2;1Þ
ð1Þ Wecr As above, C TI accounts for the proportionality constant as well
STI ðV 0 Þ ¼ hðV 0 ; Ded Þ exp  f ðV 0 Þfed ðDed ÞdDed as the effects of nonuniform bubble volume distribution.
0:5Db We
  For Group 2 bubbles, the gain and loss rates for bubble breakup
Wecr
 1:27C 3k e1=3 f1 exp  ð1  atot ÞV 02=9 ð50Þ must be balanced in order to obtain the net gain in interfacial area
We concentration. The bubble number density change is given by
where Ck is proportionality constant given by Azbel and Liapis Z V0
bðV 0 ; VÞnðV 0 ÞSTI ðV 0 ÞdV  8:05C 3k e1=3 f2 ð1  atot Þ
ð2;2Þ ð2Þ
(1983) and Wecr is the critical Weber number. The Weber number STI ðV 0 Þ ¼
Vc
defined in terms of average Group 1 properties:
 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
Wecr2 Wecr2 02=9
qf u 2t1 Dsm1 2qf e2=3 ðDsm1 Þ5=3  exp  1 V  V c V 02=9 : ð58Þ
We1 ¼ ¼ ð51Þ We2 We2
r r
 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16 9

Z V m1
Accounting for the loss due to bubble destruction for breakup ð2;12Þ 0
SSO ðVÞ ¼ bðV 0 ; VÞnðV 0 ÞgðV 0 Þf ðV 0 ÞdV
resulting in both Group 1 and Group 2 bubbles as well as the V min
production of interfacial area for breakup resulting in Group 2
bubbles results in the net source of interfacial area concentration
q9=5 13=5
f v g2 b
6=5
 38:88 f2 V 01=3 ð63Þ
for Group 2: qg r4=5 Dh We9=5
6=5
c
!  
5=3
ð2Þ ð2Þ 1=3 ai2 Wecr2 Then the total interfacial area source for Group 1 becomes
/TI;2 ¼ 0:378C TI e ð1  atot Þ exp  Z
a22=3 We2
ð2;12Þ
V m2
0
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi /SO;1 ¼ SSO ðVÞAi ðVÞdV
Wecr2  13=3
 V SO;c
1 1  0:212Dc2 ð59Þ  4 " #
We2 q3=5 1=5 2=5 2
f v g2 r ai2 Wec;SO
¼ 8:0C SO 2=5 3=5 a
1 ð64Þ
qg Dh Wec 2 Wem2
3.2.2. Shearing-off at the base of cap bubbles
where CSO absorbs the proportionality constant b⁄ and accounts for
The shearing-off of Group 1 bubbles at the base of large cap
the nonuniform bubble volume distribution. Based on the work of
bubbles may be the single most important mechanism for produc-
Evans et al. (1992), the value of the critical Weber number Wec,SO
ing Group 1 bubbles, but remains one of the least well understood.
is 1.2.
Previous analyses have focused on viscous shear as the main cause
The reduction in the Group 2 interfacial area concentration can
of this mechanism. In highly viscous flows cap bubbles can form
be computed from the volume loss rate of the Group 2 bubble,
thin skirts around the base of the bubble resulting in small bubbles
which is given by
being sheared off when the shear force on the skirt is strong en- !
0
ough to overcome the surface tension force at the bubble base. In dV r
modeling this mechanism, these forces can be identified as ¼ Q g:SO ¼ pDb v g2 d ¼ 30:86 V 01=3 ð65Þ
dt qg v g2
Surface tension force : F r  2rpDb ð60Þ
Then the loss rate for the bubble surface area can be computed
1 1 from the cap bubble geometry as
Interfacial shear force : F s  C fi qf v 2r As ¼ C fi qf v 2r pDb l ð61Þ
2 2 !
dAi ðVÞ r
where Cft is a friction factor specified by a constitutive relation and ¼ 118:45 ð66Þ
vr is the relative velocity of a large bubble. The skirt thickness is pre- dt qg v g2
dicted as (Clift et al., 1978) as Finally, the loss rate of interfacial area concentration can be
  found to be
6lg U T 1=2
d¼ ð62Þ Z
g Dq ð2;12Þ @ai2 V m2
dAi ðV 0 Þ 0
/SO;2 ¼ ¼ f ðV 0 Þ dV
@t SO V SO;c dt
where UT is the terminal rise velocity of a bubble and lg is the viscos- !   
ity of the gas phase. Under typical air–water two-phase flow condi- r a3i2 Wec;SO
tions, this skirt thickness is only about 50 lm. Given the significant ¼ 0:36C SO 1  ð67Þ
qg v g2 a22 Wem2
impact of this mechanism on the Group 1 interfacial area concentra-
tion, such a thin skirt cannot account for all of the bubble production
from this source and another mechanism must be determined. 3.2.3. Breakup due to surface instability
Another possible mechanism accounts for the gas velocity pro- When a coalescence event results in a bubble larger than the
file inside the cap bubble (Fu and Ishii, 2003a). In order to estimate maximum stable bubble limit, it becomes unstable and disinte-
the thickness of the gas layer that can be sheared off, the boundary grates. If this breakup is assumed to occur instantaneously, then
layer thickness is used. Then, shearing off is considered to be a the coalescence rate is equal to the disintegration rate and there
two-step process with shearing-off and secondary breakup due is no net bubble number change for Group 2 bubbles. The two pos-
to strong turbulence in the wake region. From experiments on ver- sible coalescence mechanisms that can produce such large bubbles
tical jets (Evans et al., 1992) the maximum bubble diameter in the are random collision and wake entrainment.
mixing zone (or wake region) can be determined with a critical The number density change for Group 2 bubbles due to random
Weber number for breakup being approximately 1.2. collision has already been given in Eq. (27). With appropriate limits
Images taken in pipes of various sizes as well as rectangular of integration the production rate of unstable bubbles can be com-
channels show that Group 1 bubbles are formed when interfacial puted as
waves on the surface of a Group 2 bubble are sheared off. The for- Z V m2 þV c
ð2Þ ð2Þ
mation of the sheared-off volume is related to Kelvin–Helmholtz RSI;RC ¼ SRC ðVÞdV
type instability. Based on jet instability studies (Spangler, 1994; V c;max
 
h  i
Reitz and Bracco, 1986) the sheared-off volume can be estimated, 1=3 1 2 1 Vc
 2:98C RC 1  exp C RC2 a1=2
ð2Þ
2 e a 2 ð68Þ
with the maximum stable bubble diameter defined as by Evans D2h V 5=9
c;max
V c;max
et al. (1992). Thus the formation rate of Group 1 bubbles can be
estimated as where Vc,max is the maximum stable bubble volume. Assuming that
! Vm2 is approximately equal to the maximum stable bubble size and
9=5 13=5
pDb v g2 d q v f g2 b
6=5
Db substituting then gives
nðV 0 ÞgðV 0 Þ  ¼ 24:87 ð62Þ
ðp=6Þds
3 qg r 4=5
Wec9=5 D6=5 ð2Þ ð2Þ
h RSI;RC ¼ 1:34  105 C RC e1=3
 
where n is the number of bubbles produced, ds is the skirt thickness 1 r 5=6 h 
1=2
i
and b is a proportionality constant. Because all of the daughter  2 a22 1  exp C RC2 a2 ð69Þ
Dh g Dq
bubbles are assumed to be of the same size, the daughter size dis-
tribution can be approximated by a delta function. This allows cal- Similarly, the number density change for Group 2 bubbles due
culation of the number density change of Group 1 bubbles as to wake entrainment has been given previously. Based on this
10 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16

formulation and with appropriate limits of integration the produc- In one-dimensional analysis codes turbulent dissipation is typically
tion rate of unstable bubbles by wake entrainment is estimated using similar, simplified models. Further, even more
Z V m2 þV c  2 complex and detailed models carry a significant amount of uncer-
ð2Þ ð2Þ ð2Þ ð2Þ r tainty and do not specify every aspect of the two-phase turbulence
RSI;WE ¼ SWE ðVÞdV  7:30  109 C WE PWE u
rw2 a22
Vc g Dq such as the eddy size distribution, and often do not sufficiently ac-
ð70Þ count for the effect of bubble-induced turbulence. Thus the use of a
single parameter to characterize turbulence – the method used in
The analysis of surface instability breakup is simplified by
nearly every computational approach – can be considered a very
assuming that any bubbles that are created above the stable size
primitive method in any case despite the more complex k–e or
limit undergo instantaneous breakup. This allows the breakup rate
k–x models used to predict it. For these reasons this relatively sim-
to be written as the sum of the two coalescence rates. The interfa-
ple model was chosen to reduce computational complexity while
cial area concentration source can then be written as
still providing an approximate value for the turbulent dissipation.
ð2Þ ð2Þ
/SI ¼ DAi RSI ð71Þ
3.4. Void transport source and sink terms
where DAi is the average change in bubble surface area when it
undergoes breakup. Unstable bubbles tend to break apart into two The inter-group transport of void fraction is modeled in similar
bubbles with slightly different volumes, resulting in fashion to interfacial area concentration from the bubble number
 2  2  density change, but using particle volume instead of particle sur-
D1 D2
DA i ¼ Dc;max
þ Dc;max
 1  Ai;max face area. This results in the following equations for transport of
ð72Þ void fraction between groups by random collision, wake entrain-
D1 ¼ 0:452Dc;max ; D2 ¼ 0:767Dc;max and therefore
ment, turbulent impact and shearing off:
DA i ¼ 0:488D2c;max
e1=3 a21 a2=3
8  1=6 h  i 9 gð11;2Þ ð1Þ ð1Þ
RC;2 ¼ 3:15C RC kRC
i1

>
< 2:616  104 C ð2Þ e1=3 1 2
a r 1  exp C a1=2
þ>
=
a2=3
1;max
RC Dh 2 g Dq
2 RC2 2 2 0 13
ð2Þ 1=3  
/SI ¼
>  1 > a1;max a1=3
1 2
: 1:425  107 C ð2Þ Pð2Þ u 2 r ;  41  exp @C RC1  A5 1  DC1 ð78Þ
WE WE rw2 a2 g Dq
 3
a1=3
1;max  a1=3
1
ð73Þ ð12;2Þ ð12;2Þ 1=3 5=3 4=3 2=3
gð12;2Þ
RC;2 ¼ 1:44C RC kRC e a1 a2 ai2
2 0 13
1=3
a1;max a1=3
1
3.3. Turbulence modeling  41  exp @C RC1  A5 ð79Þ
a1=3 1=3
1;max  a1
Many of the models above are dependent on the level of turbu-  
ð1Þ 1=3 2
lent dissipation in the flow field. In order to determine this, the gð11;2Þ 
WE;2 ¼ 3:85C WE C D1 ur1 a1 ai1 1  DC1 ð80Þ
3
shear-induced and bubble-induced components must be known. ð12;2Þ ð12;2Þ
This can be done by solving the k–e model for two-phase turbulence, g WE;2 ¼ 0:33C WE u a
 w12 1 ai2 ð81Þ
 
however for the current study a simple algebraic expression can suf- Wecr2
gð2;1Þ ¼ 11:65C TI e1=3 ð1  at Þa1=3 2=3
ð2;1Þ
TI;2 2 ai2 exp 
fice. The shear-induced turbulence can be expressed as (Kim, 1999) We2
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Wecr2  
hv m i3  1 16=3
 0:15DC2  0:117D6 ð82Þ
esi ¼ fTW ð74Þ We2 C2
2Dh
where gð2;1Þ ð2;1Þ
TI;1 ¼ gTI;2 ð83Þ
! "  4 #
 0:25 r a2i2 Wec;SO
0:316 1 qg jg þ qf jf gð2;12Þ
SO;2 ¼ 2:33C SO 1 ð84Þ
fTW ¼ and v m ðzÞ ¼ ð75Þ qg v g2 a2 Wem2
Re0:25
m
1a aqg þ ð1  aÞqf
gð2;12Þ ð2;12Þ
SO;1 ¼ gSO;2 ð85Þ
With Rem being the mixture Reynold’s number and j indicating
the superficial velocity or volumetric flux. While the bubble in-
duced turbulence can be given by the correlation of Lance and Bat- 4. Model validation
taille (1991) as
a 4.1. Validation scheme
ebi  C D u3r ð76Þ
Db
4.1.1. Computational scheme
where CD is the drag coefficient and ur the relative velocity of the In order to validate the models, the IATE is compared to the
bubble, or by Hibiki and Ishii (2002) as experimental data of Smith et al. (2011) by evaluating the perfor-
ebi  ghjg i ð77Þ mance of the one-dimensional IATE. For a steady-state, adiabatic
system this model is given through area-averaging as
It should be noted that as the pipe diameter increases, the  
d 2 hai1 i d
shear-induced turbulence predicted by Eq. (74) decreases to zero, v i1 iiÞ ¼  CD2
ðhai1 ihh~ c1 v g1 iiÞ
ðha1 ihh~
indicated that the collision frequencies will decrease to zero. This dz 3 ha1 i dz
X
is physically not the case, meaning that the bubble-induced turbu- þ h/j;1 i and ð86Þ
lence term is extremely important for flows where the channel size j

is large. d 2 hai2 i d
While this turbulence model may seem relatively primitive in
v i2 iiÞ ¼
ðhai2 ihh~ v g2 iiÞ
ðha2 ihh~
dz 3 ha2 i dz
comparison to the models previously detailed, there are several hai1 i d X
points to consider when discussing turbulence modeling. First, þ CD2
c1 ðha1 ihh~v g1 iiÞ þ h/j;2 i ð87Þ
ha1 i dz j
these IATE models are developed for a one-dimensional approach.
 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16 11

where the covariance terms that arise during area-averaging are culation from the two-fluid momentum equations. This is done by
close to unity and therefore neglected. Also, the single brackets taking the gas phase velocities at each measurement port as inputs
indicate simple area-averaged quantities, while the double brackets and using a quadratic curve fit to calculate the velocity at each
indicate void-weighted area-averaged quantities. This averaging point along the test section. With the velocities and inlet condi-
method is also applied to all of the coalescence and breakup models tions specified, the solution is iterated at each step dz along the test
given in Section 3. Generally, it is possible to assume that the inter- section until convergence is reached.
facial velocities are equal to the transport velocity of the bubble
group. 4.1.2. Determination of model constants
The IATE is also highly dependent on the void fraction. For this The models described in Section 3 result in a total of 18 coeffi-
reason, the two-group void transport equations are also area-aver- cients and critical values that must be determined. These values
aged as arise from our inability to predict every aspect of the two-phase
  flows. Many of these constants have analytical expressions; how-
d ðhag ihhv gz iiÞ dpðzÞ d
ðha1 ihhv gz1 iiÞ ¼   ðha2 ihhv gz2 iiÞ and ð88Þ ever no models exist to predict the parameters necessary to com-
dz pðzÞ dz dz pute their values. Four of these constants, those for bubbly flows,
  
d 1 ha2 ihhv gz2 ii dhpi have been determined through previous studies (Wu et al., 1998;
ðha2 ihhv gz2 iiÞ ¼   
dz 1þCD3 hpi dz Fu and Ishii, 2003a,b). The remaining models are newly derived
c1
# and apply specifically to flows in large diameter pipes. Based on
d X the hydrodynamics of the flow and previous experimental data
þ CD3c1 ðh a 2 ihhv gz2 iiÞþ g j;2 ð89Þ
dz j
many of these constants such as those for random collision can
be grouped together into a single constant that applies for several
and used to predict the evolution of the void fraction where p is the cases. This allows the number of independent constants to be re-
pressure and z the axial coordinate. duced to 13.
A rough outline of the method used for evaluating the IATE coa- Sensitivity analysis of the various mechanisms shows that each
lescence and breakup models is shown in Fig. 3. An explicit Euler of these mechanisms is dominant under different flow conditions.
method is applied to solve the ordinary differential equation using The regions for which each type of mechanism is dominant, and
a finite differencing approach with a cell size of 1 cm and imple- therefore the regions where each constant should be determined,
mented using a MATLAB script. The measured conditions at the are shown in Fig. 4. The green symbols denote constants that have
lowest measurement location are taken as inputs and used as the already been determined for bubbly flows based on the work of
initial condition in the simulation. Wu et al. (1998), while the blue symbols indicate mechanisms that
Ideally this system of equations would be best solved by inclu- are expected to have similar constants to those given by Wu et al.
sion in a one-dimensional system code that uses the two-fluid (1998). Using this study as a basis gives the values for the con-
ð1Þ ð12;2Þ ð2Þ
model to compute the velocity of the gas phase and the void frac- stants C RC , C RC , C RC , CRC0, CRC1 and CRC2 in the random collision
ð1Þ ð1Þ
tion. To simplify the calculation process and eliminate numerical models, C WE in the wake entrainment models, C TI , WEcr1 and WEcr2
issues that may arise the current approach decouples the IATE cal- in the turbulent impact models. Additionally, Wu et al. (1998) and
Fu and Ishii (2003a,b) give the value of a1,max as 0.62 and Fu and
Ishii (2003a,b) give the value of C as 1.0. Based on the work of
Evans et al. (1992), the critical Weber number for shearing off,
WeCR,SO, has a value of 1.2. It should be noted here that although
the inter-group and Group 2 models for random collision and tur-
bulent impact are newly derived as part of the current study and
quite different from the two-group models developed by Fu and
Ishii (2003a), analysis of the models indicates that the values
of the adjustable constants should be similar to those in the

Fig. 3. Diagram of IATE evaluation process. Fig. 4. Dominant regions for various coalescence and breakup mechanisms.
12 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16

Table 2
Coefficients and critical values for IATE Models.

C = 1.0 Coefficient in the transport equation for inter-group transport at the group boundary
ð1Þ
C RC ¼ 0:01;
ð12;2Þ
C RC
ð2Þ
¼ 0:01; C RC ¼ 0:01 Coefficients in bubble coalescence due to RC
ð1Þ
C WE ¼ 0:002;
ð12;2Þ ð2Þ
C WE ¼ 0:01; C WE ¼ 0:06 Coefficients in bubble coalescence due to WE
ð1Þ
C TI ¼ 0:05;
ð2;1Þ ð2Þ
C TI ¼ 0:04; C TI ¼ 0:01 Coefficients in bubble breakup due to TI
6
CSO = 2.5  10 Coefficient in bubble breakup due to SO
C¼
RC0 3:0; C¼
RC1 3:0; C¼
RC2 3:0 Coefficients in the Random Collision model accounting for effective range of influence of turbulent eddies
a1,max = 0.62 Dense packing limit for Group 1 bubbles
Wecr1 = 1.2, Wecr2 = 1.2 Critical Weber number for breakup due to TI
Wec,SO = 4000 Critical Weber number for shearing-off of small bubbles from large cap bubbles

one-group model of Wu et al. (1998). For this reason the values Group 2 wake entrainment and Group 2 turbulent impact in the re-
given in previous studies are used. Thus for the current study the gions where each is dominant as noted in Fig. 4. This provides the
ð2Þ ð2Þ
performance of the models using the previous values must be values of C WE and C TI . When this has been completed, only the
confirmed for the two-group case. Further, new constants must shearing-off mechanism remains. Comparison with the experi-
ð12;2Þ ð2Þ ð2;1Þ ð2Þ
be determined for C WE , C WE , C TI , C TI and CSO. mental data collected in the churn-turbulent region is then used
Determining the values of these constants is a multi-step ap- to evaluate these constants using the models developed and
proach. The methodology used to reduce error and subjectivity is benchmarked for all of the other mechanisms. This results in the
detailed in the work of Kim (1999), and the same method has been values for each constant as given in Table 2. The IATE with bench-
used in this case. First, the experimental data of Smith et al. (2011) marked source and sink models is then compared to the experi-
has been used to evaluate the Group 2 void transport equation and mental data, which is detailed in the following sections.
inter-group void transport mechanisms. This multi-step process
involves isolating each coefficient by comparison with data in the 4.2. Comparison with experimental data
region for which the specific inter-group mechanism dominates
the other inter-group mechanisms. This provides the values for 4.2.1. Model performance in transition to cap bubbly flow
ð12;2Þ ð2;1Þ
C WE and C TI . When these inter-group constants have been Figs. 5–7 show samples of the IATE prediction for the data of
determined, the Group 2 IATE is used to evaluate the effects of Smith et al. (2011). In the upper plots, the black lines represent

Fig. 5. Experimental data and model results for hjfi = 0.30 m/s and hjgi = 0.15 m/s.
 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16 13

Fig. 6. Experimental data and model results for hjfi = 0.30 m/s and hjgi = 0.30 m/s.

the prediction of total interfacial area concentration (upper left) Wake entrainment of Group 1 bubbles and production of Group
and void fraction (upper right). The blue lines represent the predic- 1 bubbles by turbulent impact in this instance represent compet-
tion for Group 2, while the red lines represent the prediction for ing mechanisms. Wake entrainment is slightly more dominant
Group 1. The appropriately colored symbols indicate the measured than turbulent impact, resulting in a net transfer of interfacial area
values from experiment. In the lower plots, the various colored to Group 2. At first, this causes a slight increase in the Group 2
lines indicate the different bubble breakup and coalescence interfacial area concentration, but the increase Group 2 bubbles re-
mechanisms. sults in increased wake entrainment of Group 2 bubbles leading to
Significant changes to the interfacial structure begin to occur as a dramatic decrease in the Group 2 interfacial area concentration.
the flow transitions from bubbly flow to cap-bubbly conditions. To As Group 2 bubbles continue to increase in size shearing off begins
highlight the performance of the model across the transition to have more significant effect, resulting in an increase in the
boundaries, several experiments in the 0.152 m diameter test facil- Group 1 interfacial area concentration. These behaviors are accu-
ity are discussed. At low superficial velocities (hjfi = 0.30 m/s and rately represented by the IATE as the figures show.
hjgi = 0.15 m/s), wake entrainment and random collision effects
dominate the transfer of Group 1 bubbles to Group 2. The physical 4.2.2. Model generality and scaling considerations
changes to the interfacial structure are reflected in the model pre- The interfacial area transport equation is derived without many
diction shown in Fig. 5, with error bars representing ±10% error. restrictive assumptions and is therefore applicable to a variety of
Wake entrainment of Group 2 bubbles also tends to dominate two-phase flows. The foundation of the transport equation is the
the sources of Group 2 bubbles along the test section, which acts Boltzmann transport equation, which uses the particle density dis-
to reduce the total interfacial area concentration. tribution function as a fundamental quantity. The actual form of
The data in Fig. 6, for hjfi = 0.30 m/s and hjgi = 0.30 m/s, further the particle density distribution function can, however, be very
illustrates the transition from bubbly to cap-bubbly flow. The complex and it is then necessary to make certain simplifications.
experimental data shows little change in the average Group 1 bub- Bubbles are divided into two groups based on their drag and trans-
ble size with a decrease in the Group 1 void fraction, while the port properties, and the distribution function is assumed to be lin-
Group 2 bubble size increases significantly along the axial direc- ear within each group. In large diameter channels the bubble
tion. The dominant mechanism for this change is wake entrain- distribution tends to be smooth across the pipe area and this
ment by Group 2 bubbles, of both Group 1 and other Group 2 assumption may be valid.
bubbles. In addition, the bubble shape was assumed for the purposes of
In Fig. 7, the data shown was collected at hjfi = 0.30 m/s and computing surface area. Group 1 bubbles are assumed to be
hjgi = 1.00 m/s. This flow condition is fully in cap-bubbly flow. spherical or nearly so, while Group 2 bubbles are assumed to be
14 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16

Fig. 7. Experimental data and model results for hjfi = 0.30 m/s and hjgi = 1.00 m/s.

hemispherical. In a large diameter channel the Group 2 bubbles are Shearing-off also has geometry dependence for small diameter
not confined by the pipe walls and so are free to assume any shape, channels. The dissipation rate in the mixing region depends on
but in small diameter channels the bubble shape and size are the ratio of the cross sectional area of the wake to the cross-
restricted by the channel walls. This may significantly change the sectional area of the pipe. This is much larger in small diameter
bubble density distribution function, especially for slug flows. pipes than in large diameter pipes, so this modification must be
For the most part, the bubble interaction mechanisms modeled considered when applying the model to smaller geometries.
here are independent of the channel geometry because for large Finally, for small diameter channels bubble breakup due to sur-
channels, Dh is significantly larger than Db and channel geometry face instability can be eliminated, as bubbles larger than the stabil-
effects are small. Thus the underlying assumptions in many of ity limit cannot form in small diameter channels.
the models are very general and should be applicable to other flow To illustrate these effects, the above considerations have been
geometries. A few of these interaction mechanisms however are used to predict the interfacial area concentration in pipes of diam-
geometry dependent and may not be applicable to more confined eter 0.0508 m (Fu and Ishii, 2003b), 0.102 m and 0.152 m (Smith
geometries. et al., 2011). The results are shown in Fig. 8 as the ratio of predicted
The random collision of Group 2 bubbles is dependent on the to experimental value of interfacial area concentration, with error
channel geometry, as the probability of collision decreases as the bars of ±15%. Thus a value of 1.0 means that the model predicted
channel size increases. In small pipes however there is very little the data exactly, a value of 1.15 means the model over-predicted
random collision of Group 2 bubbles because these bubbles have the data by 15%, and a value of 0.85 means that the model un-
little freedom to move, thus wake entrainment becomes the dom- der-predicted the data by 15%. For the 0.0508 m diameter case
inant coalescence mechanism. Thus for small pipes, random colli- the predictions are reasonable, however the model greatly over-
sion of Group 2 bubbles can be neglected. Wake entrainment of predicts the effect of shearing-off, turbulent impact, and wake
Group 2 bubbles in small channels also depends on the develop- entrainment. For the experiments shown in the figure these effects
ment length of the flow, L/D, as the probability of wake entrain- canceled each other resulting in an overall reasonable prediction,
ment depends on the wake length and number of Group 2 however these scaling effects mean that the model may not per-
bubbles in the wake region. For large pipes, the wake length has form well for all conditions in smaller diameter channels and a
been assumed to be twice the bubble diameter, however Taitel model developed with consideration of the geometry should be
et al. (1980) note that for small pipes with stable slug flow Lw/Dh used. The model predicts the data from the 0.102 m diameter chan-
is approximately 16. This indicates geometry dependence for the nel with reasonable accuracy for void fractions of up to 0.7, well
small pipe case. within the churn-turbulent flow region. The data for the 0.152 m
 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16 15

large-diameter cases where bubble-induced turbulence is domi-

nant. Also, the majority of the experiments used in this effort were
performed in a 0.102 m diameter channel, which is very close to
the maximum stable bubble size of just under 0.10 m. Additional
data should be collected in pipes with diameter of 0.152 m and lar-
ger to further validate the model performance under conditions
where the effect of the pipe wall is even less significant.

Acknowledgements

This work was performed at Purdue University under the aus-

pices of the U.S. Nuclear Regulatory Commission, Office of Nuclear
Regulatory Research, through the Institute of Thermal Hydraulics.

References

Azbel, D., Liapis, A.I., 1983. Mechanisms of liquid entrainment. In: Cheremisinoff,
N.P., Gupta, R. (Eds.), Handbook of Fluids in Motion. Ann Arbor Science, pp. 453–
482.
Batchelor, G.K., 1951. Pressure fluctuation in isotropic turbulence. Proc. Cambridge
Fig. 8. Model performance in pipes of various sizes. Philos. Soc. 47, 359–371.
Cheung, S., Yeoh, G.H., Tu, J.Y., 2007. On the numerical study of isothermal vertical
bubbly flow using two population balance approaches. Chem. Eng. Sci. 62,
4659–4674.
diameter facility is also well-predicted for void fractions of up to Clift, R., Grace, J.R., Weber, M.E., 1978. Bubbles, Drops, and Particles. Academic Press,
0.4. It should be noted here that the 0.102 m diameter channel is Inc.
only slightly larger than the maximum stable bubble size, and Coulaloglou, C.A., Tavlarides, L.L., 1977. Description of interaction processes in
agitated liquid–liquid dispersions. Chem. Eng. Sci. 32, 1289.
therefore some effects of the transition from small to large diame- Evans, G.M., Jameson, G.J., Atkinson, B.W., 1992. Prediction of the bubble size
ter channels may still be present. As these effects mainly affect the generated by a plunging liquid jet bubble column. Chem. Eng. Sci. 47, 3265–
behavior of Group 2 bubbles, the changes are not apparent in the 3272.
Fu, X.Y., Ishii, M., 2003a. Two-group interfacial area transport in vertical air–water
prediction of the data from the 0.152 m diameter facility because flow: I. Mechanistic model. Nucl. Eng. Des. 219, 143–168.
the Group 2 influence is not so large for void fractions smaller than Fu, X.Y., Ishii, M., 2003b. Two-group interfacial area transport in vertical air–water
0.4. While the model performs well for the conditions studied, flow: II. Model evaluation. Nucl. Eng. Des. 219, 169–190.
Hibiki, T., Ishii, M., 2002. Interfacial area concentration of bubbly flow systems.
additional data is needed for higher void fraction conditions – into Chem. Eng. Sci. 57, 3967–3977.
the churn-turbulent flow regime – in pipes with diameters larger Hibiki, T., Ishii, M., 2003. One-dimensional drift-flux model for two-phase flow in a
than 0.102 m in order to validate the performance of the model un- large diameter pipe. Int. J. Heat Mass Transfer 46, 1773–1790.
Huh, B.G., Euh, D.J., Yoon, H.Y., Yun, B.J., Song, C.H., Chung, C.H., 2006. Mechanistic
der these conditions. The average error in interfacial area concen-
study for the interfacial area transport phenomena in an air/water flow
tration for the conditions in this study was 10.2% for the 0.102 m condition by using fine-size bubble group model. Int. J. Heat Mass Transfer 49,
diameter cases and 6.5% for the 0.152 m diameter cases, indicating 4033–4042.
that the model is quite successful. Ishii, M., Zuber, N., 1979. Drag coefficient and relative velocity in bubbly, droplet or
particulate flows. AIChE J. 25, 843–855.
Ishii, M., Hibiki, T., 2010. Thermo-fluid Dynamics of Two-phase flow, second ed.
Springer, New York.
5. Conclusions Kataoka, I., Ishii, M., 1987. Drift-flux model for large diameter pipe and new
correlation for pool void fraction. Int. J. Heat Mass Transfer 30, 1927–
1939.
The two-group interfacial area transport equation is an impor- Katz, J., Meneveau, C., 1966. Wake-induced relative motion of bubbles rising in line.
tant tool for the prediction of interfacial area concentration – and Int. J. Multiphase Flow 22, 239–258.
Kim, S., 1999. Interfacial area transport equation and measurement of local
therefore interfacial transfer of mass, momentum and energy – in
interfacial characteristics. Ph.D. Thesis, Purdue University, West Lafayette, IN,
two-phase flows. This dynamic approach can improve the robust- USA.
ness and accuracy of the two-fluid model. Development of the IATE Kocamustafaogullari, G., Ishii, M., 1995. Foundations of the interfacial area transport
is ongoing, but various models have already been tested in some equation and its closure relations. Int. J. Heat Mass Transfer 38, 481–493.
Lance, M., Battaille, J., 1991. Turbulence in the liquid phase of a uniform bubbly air–
geometries. Here a model has been developed for two-phase flows water flow. J. Fluid Mech. 222, 95–118.
in channels with large hydraulic diameter, where wall effects on Liao, Y., Lucas, D., Krepper, E., Schmidtke, M., 2011. Development of a generalized
the flow field are small. Interfacial area concentration source and coalescence and breakup closure for the inhomogeneous MUSIG model. Nucl.
Eng. Des. 241, 1024–1033.
sink terms have been developed by mechanistically modeling Lucas, D., Beyer, M., Szalinski, L., Schutz, P., 2010. A new database on the
interactions between fluid particles including random collision, evolution of air–water flows along a large vertical pipe. Int. J. Therm. Sci. 49,
wake entrainment, turbulent impact, shearing-off, and surface 664–674.
Millies, M. et al., 1996. A first order relaxation model for the prediction of the local
instability. interfacial area density in two-phase flows. Int. J. Multiphase Flow 22, 1073–
To evaluate the IATE source and sink term models, the model 1104.
predictions for one-dimensional, steady-state cases have been Millies, M., Mewes, D., 1999. Interfacial area density in bubbly flow. Chem. Eng.
Process. 38, 307–319.
compared to a reliable database for pipes with diameters of Lehr, F., Mewes, D., 2001. A transport equation for the interfacial area density
0.0508 m, 0.102 m and 0.152 m in adiabatic, air–water flows. Sev- applied to bubble columns. Chem. Eng. Sci. 56, 1159–1166.
eral constants in the models were determined based on the exper- Ohnuki, A., Akimoto, H., 2000. Experimental study on transition of flow pattern and
phase distribution in upward air–water two-phase flow along a vertical pipe.
imental data for flow conditions which highlight each of the
Int. J. Multiphase Flow 26, 367–386.
mechanisms. Average errors in interfacial area concentration were Ohnuki, A., Akimoto, H., 2001. Model development for bubble turbulent diffusion
10.2% for the 0.102 m diameter cases and 6.5% for the 0.152 m and bubble diameter in large vertical pipes. J. Nucl. Sci. Technol. 38, 1074–
diameter tests. 1080.
Prasser, H.M., 2007. Evolution of interfacial area concentration in a vertical
Some future work remains to be done. More detailed modeling air–water flow measured by wire-mesh sensors. Nucl. Eng. Des. 237, 1608–
of two-phase turbulence should be performed, especially in 1617.
16 T.R. Smith et al. / International Journal of Multiphase Flow 47 (2012) 1–16

Prince, M.J., Blanch, H.W., 1990. Bubble coalescence and break-up in air-sparged Smith, T.R., Schlegel, J.P., Hibiki, T., Ishii, M., 2011. Two-phase flow structure in large
bubble columns. AIChE J. 36, 1485–1499. diameter pipes. Int. J. Heat Fluid Flow. http://dx.doi.org/10.1016/
Reitz, R.D., Bracco, F.V., 1986. Mechanisms of breakup of round liquid jets. j.ijheatfluidflow.2011.10.008.
Encyclopedia of Fluid Mechanics, vol. 3. Gulf Publishing, Houston, TX (Chapter Spangler, C., 1994. Nonlinear modeling of jet atomization in the wind-induced
10). regime. Masters Thesis, Purdue University.
Sari, S., Ergun, S., Barik, M., Kocar, C., Sokmen, C.N., 2009. Modeling of isothermal Sun, X., Smith, T.R., Kim, S., Ishii, M., Uhle, J., 2002. Interfacial area of bubbly flow in
bubbly flow with interfacial area transport equation and bubble number a relatively large diameter pipe. Exp. Therm. Fluid Sci. 27, 97–109.
density approach. Ann. Nucl. Energy 36, 222–232. Taitel, Y., Bornea, D., Duckler, A., 1980. Modeling flow pattern transitions for steady
Schlegel, J.P., Sawant, P., Paranjape, S., Ozar, B., Hibiki, T., Ishii, M., 2009. Void upward gas–liquid flow in vertical tubes. AIChE J. 26, 345–354.
fraction and flow regime in adiabatic upward two-phase flow in large diameter Tsuchiya, K., Miyahara, T., Fan, L.S., 1989. Visualization of bubble–wake interactions
pipes. Nucl. Eng. Des. 238, 2864–2874. for a stream of bubbles in a two-dimensional liquid–solid fluidized bed. Int. J.
Shen, X., Saito, Y., Mishima, K., Nakamura, H., 2006. A study on the characteristics of Multiphase Flow 15, 35–49.
upward air–water two-phase flow in a large diameter pipe. Exp. Therm. Fluid Wu, Q., Kim, S., Ishii, M., Beus, S.G., 1998. One-group interfacial area transport in
Sci. 31, 21–36. vertical bubbly flow. Int. J. Heat Mass Transfer 41, 1109–1112.
Shen, X., Matsui, R., Mishima, K., Nakamura, H., 2010. Distribution parameter and Yao, W., Morel, C., 2004. Volumetric interfacial area prediction in upward bubbly
drift velocity for two-phase flow in a large diameter pipe. Nucl. Eng. Des. 240, two-phase flow. Int. J. Heat Mass Transfer 47, 207–328.
3991–4000. Yoneda, K., Yasuo, A., Okawa, T., 2002. Flow structure and bubble characteristics of
Shoukri, M., Hassan, I., Gerges, I., 2003. Two-phase bubbly flow structure in large- steam-water two-phase flow in a large diameter pipe. Nucl. Eng. Des. 217, 267–
diameter vertical pipes. Can. J. Chem. Eng. 81, 205–211. 281.