International Journal of Multiphase Flow 43 (2012) 72–84

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

Void fraction prediction in annular two-phase flow

Andrea Cioncolini 1, John R. Thome ⇑
Heat and Mass Transfer Laboratory, Swiss Federal Institute of Technology-EPFL, EPFL-STI-IGM-LTCM, Station 9, 1015 Lausanne, Switzerland

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

Article history: A new method to predict the void fraction in annular two-phase flow in macroscale and microscale chan-
Received 12 January 2012 nels is presented. The underlying experimental database contains 2673 data points collected from 29 dif-
Received in revised form 28 February 2012 ferent literature studies for 8 different gas–liquid and vapor–liquid combinations (water–steam, R410a,
Accepted 8 March 2012
water–air, water–argon, water–nitrogen, water plus alcohol–air, alcohol–air and kerosene–air), for tube
Available online 15 March 2012
diameters from 1.05 mm to 45.5 mm and for both circular and non-circular channels. The new prediction
method is strongly simplified with respect to most existing correlations, as it depends only on vapor qual-
Keywords:
ity and the gas to liquid density ratio and reproduces the available data better than existing prediction
Annular two-phase flow
Void fraction
methods. Importantly, this study shows that there appears to be no macro-to-microscale transition in
Microscale annular flows, at least down to diameters of about 1.0 mm.
Macroscale Ó 2012 Elsevier Ltd. All rights reserved.
Flow regime
Flow pattern

1. Introduction required in virtually any two-phase flow calculation, since it is used

as input for determining numerous other key flow parameters,
Annular two-phase flow is one of the most frequently observed including the two-phase flow density, the two-phase flow viscosity
flow patterns in gas–liquid and vapor–liquid two-phase flow sys- and the average velocities of the two phases. Besides, the void frac-
tems, such as steam generators, air conditioning and refrigeration tion plays a fundamental role in the modeling of two-phase flow
systems, chemical processing plants and nuclear reactors. In annu- pattern transitions, heat transfer and pressure drop. The knowledge
lar flow, a part of the liquid phase flows as a continuous film that of the void fraction is also crucial in many thermal–hydraulic simu-
streams along the channel wall, while the rest of the liquid phase is lations, such as coupled neutronics–thermal hydraulics calculations
dispersed as entrained droplets in the gas or vapor phase that flows and two-phase natural circulation loop flow rates and heat trans-
in the center of the channel. Due to its practical relevance, annular port rates predictions.
flow has been extensively studied in the last decades. Nonetheless, Due to its importance, numerous void fraction prediction meth-
annular flow is still actively investigated as more accurate and reli- ods have been proposed so far and several assessments of predic-
able prediction methods are required for several cutting-edge tion methods have been published, including the recent
applications, such as nuclear reactor fuel optimization and power contributions by Vijayan et al. (2000), Coddington and Macian
uprate, nuclear systems transient and safety analyses and micro- (2002), Woldesemayat and Ghajar (2007) and Godbole et al.
evaporators design for the thermal management of computer (2011). According to Vijayan et al. (2000), in particular, the avail-
chips, microelectronic components, laser diodes and high energy able void fraction prediction methods can be classified into four
physics particle detectors. groups. The first group is given by slip ratio models, which specify
One of the most important parameters used to characterize two- an empirical relationship for predicting the slip between the
phase flows is the cross sectional void fraction e (simply referred to phases. The second group is given by Keh models that predict the
as the void fraction in what follows) representing the fraction of the void fraction by multiplying the homogeneous model void fraction
channel cross sectional area occupied by the gas or vapor phase. As eh with an empirically derived correction factor K. Then, the third
such, the void fraction is a flow parameter bounded between 0, cor- group is given by drift-flux correlations, which are based on the
responding to single-phase liquid flow, and 1 corresponding to sin- Zuber and Findlay (1965) drift-flux model and specify two empiri-
gle-phase gas flow. The accurate prediction of the void fraction is cal relations to predict the distribution parameter and the drift
velocity. Finally, the fourth group is the so called miscellaneous cor-
⇑ Corresponding author. Tel.: +41 21 693 5981; fax: +41 21 693 5960.
relations, which are empirical relations that do not fit into any of
E-mail addresses: andrea.cioncolini@epfl.ch (A. Cioncolini), john.thome@epfl.ch
the other groups. By far, the majority of the void fraction prediction
(J.R. Thome). methods proposed to date are based on the drift-flux model. As a
1
Tel.: +41 21 693 5984; fax: +41 21 693 5960. matter of fact, the three most accurate correlations recommended

0301-9322/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijmultiphaseflow.2012.03.003
 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84 73

in the recent review by Woldesemayat and Ghajar (2007) are drift- water–nitrogen, water plus alcohol–air, alcohol–air and kerosene–
flux models. air) and 17 different values of the tube diameter in the range of
The purpose of the present study is to present a new void frac- 1.05–45.5 mm, thus spanning from ‘micro’ to ‘macroscale’. Most
tion prediction method specifically designed for annular two-phase of the test rigs in the database for adiabatic two-component flows
flow. This new prediction method covers both macroscale and have been designed with calming sections long enough to signifi-
microscale channels, adiabatic and evaporating flow conditions cantly damp out any dependence on inlet conditions (Wolf et al.,
and is strongly simplified with respect to most existing correlations, 2001), so that inlet effects can be neglected in the present study.
as it depends only on vapor quality and the gas to liquid density ra- As can be seen in Table 1, 89.3% of the collected data are for adi-
tio. As will be shown, the new prediction method reproduces the abatic upflow, while 4.9% are for evaporating upflow and 5.8% cov-
available data better than existing correlations and extrapolates er adiabatic horizontal flow conditions, so that the databank is
to non-circular channels. This new method is part of a unified annu- biased towards adiabatic upflow conditions. Besides, as can be seen
lar flow modeling suite that is currently being developed by the in Fig. 1, most of the data were taken at operating pressures below
authors that also includes methods to predict the axial frictional 2.0 MPa and at mass fluxes below 1500 kgm2 s1. As such,
and total pressure gradients, the annular liquid film thickness, the additional void fraction measurements are more than welcome,
liquid film and gas core velocity profiles, the convective boiling heat particularly at medium to high operating pressures, high mass
transfer coefficient and the entrained liquid fraction (Cioncolini fluxes and under evaporating flow conditions.
et al., 2009a, 2009b; Cioncolini and Thome, 2011, 2012). The pres- As noted by Levy (1999), the transition from intermittent to
ent method replaces the correlation of Woldesemayat and Ghajar annular flow typically corresponds to a void fraction between 0.7
(2007) that has been used previously in this modeling suite. and 0.8. As can be seen in Fig. 1, all the data collected in Table 1
In what follows, the experimental void fraction databank col- correspond to a local void fraction above 0.7 and the vast majority
lected for use here is presented in Section 2. The new void fraction of the data are actually above 0.8. As such, the contamination of the
prediction method is described in Section 3, followed by results data from intermittent flow can be expected to be minimal, and is
and discussion presented in Section 4. therefore neglected in the present study.
It is well known that when the hydrodynamic conditions are
2. Experimental database description appropriate the pull of gravity can delay the rise of the liquid phase
in vertical upflow conditions, thus affecting the void fraction. A
The main details regarding the experimental annular flow data- preliminary check of the influence that gravity may exert on the
bank for circular tubes are summarized in Table 1, while a selection flow can be obtained by extrapolating a criterion proposed by Wal-
of histograms that further describes the collected data is shown in lis (1961) for predicting flow reversal, the condition at which in an
Fig. 1. The database includes 2633 measurements of the void frac- initially cocurrent annular upflow some of the liquid in the film
tion collected from 24 different literature studies that cover 8 dif- starts flowing downward under the pull of gravity. This flow rever-
ferent gas–liquid and vapor–liquid combinations (both single- sal condition reads as follows:
qffiffiffiffi qffiffiffiffi
component saturated fluids such as water–steam and refrigerant
J l þ J g < 1 ) Flow reversal ð1Þ
R410a and two-component fluids, such as water–air, water–argon,

Table 1
Experimental annular flow data bank for circular tubes.

Reference Fluids d (mm) P (MPa) G (kgm2 s1) x e (1) (2) (3) (4) (5)
Anderson and Mantzouranis (1960) H2O–Air 10.8 0.11 22–1419 0.01–0.70 0.70–0.97 23.5 " a 52 QCV
Beggs (1972) H2O–Air 25.4; 38.1 0.55–0.68 64–1058 0.04–0.83 0.72–0.98 na " a 13 QCV
Celata and Frazzoli (1981) H2O–Steam 21.2 4.9–6.9 170–300 0.19–0.98 0.70–0.98 na " d 118 RA
Alia et al. (1965) H2O–Ar 15.0; 25.0 0.60–2.2 306–3000 0.11–0.81 0.70–0.99 60–157 " a 265 LD;RA
Alcohol–Ar
Dejesus and Kawagi (1990) H2O–Air 25.4 0.15 205–1320 0.01–0.12 0.70–0.89 66 " a 28 RA
Godbole et al. (2011) H2O–Air 12.7 0.12–0.26 93–880 0.01–0.32 0.70–0.90 50 " a 57 QCV
Kaji and Azzopardi (2010) H2O–Air 19.0 0.15 41–695 0.01–0.67 0.70–0.96 300 " a 59 CP
Leung et al. (2005) H2O–Steam 13.4 2.0 4500 0.03–0.11 0.73–0.95 na " d 12 RA
Mukherjee (1979) Kerosene–Air 38.1 0.29–0.61 49–2520 0.02–0.76 0.71–0.98 na " a 39 CP
Spedding and Nguyen (1976) H2O–Air 45.5 0.11–0.12 29–1057 0.01–0.93 0.71–0.99 na " a 101 QCV
Sujumnong (1998) H2O–Air 12.7 0.11–0.26 64–3260 0.01–0.77 0.72–0.99 na " a 41 QCV
Ueda (1967) H2O–Air 19.4 0.11 97–661 0.01–0.05 0.70–0.79 68 " a 25 QCV
Shedd (2010) R410a 1.05; 2.96 1.9–3.1 400–800 0.20–0.96 0.70–0.99 90–114 ? a 152 CP
Silvestri et al. (1963) H2O–Ar 25.0 0.60–2.1 280–2900 0.06–0.84 0.71–0.99 60 " a 436 FT
H2O + Alcohol–Ar
Adorni et al. (1963) H2O–Ar 15.1 0.60–2.1 312–3420 0.06–0.82 0.70–0.98 99 " a 121 FT
Casagrande et al. (1963) H2O–N2 25.0 0.29–2.4 255–2880 0.07–0.79 0.70–0.99 60 " a 109 FT
H2O + Alcohol–Ar
Cravarolo et al. (1964) H2O–Ar 15.1; 25.0 0.60–2.1 266–2880 0.04–0.79 0.70–0.98 60–232 " a 517 FT
Alia et al. (1966) H2O–Ar 15.1; 25.0 0.60–2.1 266–3290 0.04–0.90 0.70–0.99 140–233 " a 136 FT
Gill et al. (1964,1965) H2O–Air 31.8 0.11 24–555 0.09–0.94 0.91–0.99 64–171 " a 147 FT
Hall-Taylor et al. (1963) H2O–Air 31.8 0.11–0.16 34–76 0.41–0.78 0.95–0.98 184 " a 18 FT
Whalley et al. (1974) H2O–Air 31.8 0.12–0.35 78–789 0.10–0.90 0.91–0.99 590 " a 139 FT
Brown (1978) H2O–Air 31.8 0.17–0.31 158–316 0.33–0.66 0.94–0.98 420 " a 30 FT
Würtz (1978) H2O–Steam 20.0 7.0 500–2000 0.20–0.70 0.72–0.96 450 " a 18 FT

(1) – Dimensionless distance L/d of test section inlet from mixer (2 component fluids, adiabatic tests only).
(2) – Flow direction: " = vertical upflow; ? = horizontal flow.
(3) – Type of test: a = adiabatic; d = diabatic.
(4) – Number of data points.
(5) – Measuring technique: QCV = quick closing valves; RA = radiation attenuation; CP = capacitance probe; LD = liquid displacement; FT = annular film thickness.
74 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84

1400 1200

Number of Data Points

Number of Data Points
1200 1000
1000
800
800
600
600
400
400
200 200
0 0
0 5 10 15 20 25 30 35 40 45 50 0 1 2 3 4 5 6 7
Tube Diameter [mm] Pressure [MPa]

400 200

Number of Data Points

Number of Data Points

300 150

200 100

100 50

0 0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Mass Flux [kg/m2 s] Vapor Quality

Vertical Upflow Data from Table 1

200 250
Number of Data Points

Number of Data Points

150 200

150
100
100
50
50

0 0
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0 1 2 3 4 5 6 7 8
Void Fraction Eq.(1) Left-Hand

600 1200
Number of Data Points

Number of Data Points

500 1000
400 800
300 600
200 400
100 200
0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 -3 -2.5 -2 -1.5 -1 -0.5
Mach Number Log10 (Density Ratio)

2500
1200
Number of Data Points
Number of Data Points

1000 2000
800 1500
600
1000
400
200 500

0 0
-2.2 -2 -1.8 -1.6 -1.4 -1.2 -1 -0.8 -0.6 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Log 10 (Viscosity Ratio) Surface Tension [N/m]

Fig. 1. Selected histograms describing the experimental database in Table 1.

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where the non-dimensional average volumetric fluxes of liquid Jl ql qg
and vapor Jg are: J l ¼ J l ; J g ¼ J g ð2Þ
gdðql  qg Þ gdðql  qg Þ
 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84 75

where ql and qg are the liquid and vapor densities, g is the acceler- the measured parameter is the average annular liquid film thick-
ation of gravity, d is the tube diameter and the superficial velocities ness and not the void fraction. In order to use these data here, their
of liquid Jl and vapor Jg are: void fractions have been calculated to account for liquid entrain-
ment as follows. The analytical expression of the liquid film flow
ð1  xÞG xG
Jl ¼ ; Jg ¼ ð3Þ area yields the following relation:
ql qg
4tðd  tÞ
ð1  eÞð1  cÞ ¼ ð5Þ
where x is vapor quality and G the total mass flux. A histogram d
2

showing the left-hand side of Eq. (1) for the vertical upflow data
of Table 1 is included in Fig. 1. As can be seen, the influence of grav- where c is the liquid droplet hold-up, representing the fraction of
ity on the flow should be negligible for most of the data, so that the the liquid phase cross sectional area occupied by the entrained
collected vertical upflow data can be assumed to be essentially lim- droplets and t is the average liquid film thickness. A further relation
ited to shear-driven annular flows, where gravity plays a marginal involving the void fraction e and the droplet hold-up c can be de-
or negligible role. The effect of gravity on the horizontal microscale rived if the slip between the carrier gas phase and the entrained li-
data of Shedd (2010) is as well neglected, since the mass flux is quid droplets is neglected, so that gas and droplets are assumed to
judged to be high enough to prevent gravity from significantly travel at the same speed:
affecting the axial symmetry of the flow in the microtubes tested. 1e 1  x qg
To the authors’ knowledge, the measurements performed by Shedd c ¼e ð6Þ
e x ql
(2010) provide the first reliable (and verifiable) microscale void
fraction data in the annular flow regime that appear in the litera- where e is the entrained liquid fraction, representing the ratio of the
ture. Other previous microscale void fraction measurements, entrained liquid droplets mass flow rate to the total liquid mass
including Triplett et al. (1999), Serizawa et al. (2002), Kawahara flow rate. Eqs. (5) and (6) are a system of two nonlinear equations
et al. (2002) and Chung and Kawaji (2004) have been obtained with that can be solved for e and c provided that the right-hand side of
image processing techniques that do not appear to be reliable for both equations is known. In this respect, the only parameter that
annular flows at high vapor quality where the liquid films are very needs to be determined is the entrained liquid fraction e, which is
thin. predicted according to Cioncolini and Thome (2012) as follows:
In order to check the relevance of compressibility effects in the  2:209
databank, the following form of the Mach number M is used: e ¼ 1 þ 279:6We0:8395
c ; 101 < Wec < 105 ð7Þ

V tpf V g xG The core flow Weber number Wec is defined as:

M¼  ¼ ð4Þ
atpf ag qg eag
qc J2g d x þ eð1  xÞ
Wec ¼ ; qc ¼ x eð1xÞ ð8Þ
where Vtpf and atpf are the average velocity and the sonic velocity for r q þ q
g l
the two-phase flow. As can be seen, the average two-phase velocity
Vtpf is approximated with the gas velocity Vg calculated in a segre- where qc is the droplet laden gas core density and r the surface ten-
gated-phase approximation. Since entrained liquid droplets are ne- sion. The core flow density qc depends on the entrained liquid frac-
glected, Vg provides a slight upperbound to the actual two-phase tion e, so that in principle an iterative calculation would be required
flow velocity. The sonic two-phase velocity atpf, on the other hand, to resolve these expressions. As described by Cioncolini and Thome
is approximated in Eq. (4) with the gas sonic velocity ag, which (2012), however, the method was implemented with a two-step ex-
provides a lowerbound to the sonic velocity of a high void fraction plicit predictor–corrector scheme to simplify the procedure. Assum-
two-phase flow. Available measurements of sonic velocity in high ing an uncertainty of 34.1% for the entrained liquid fraction, as
pressure steam–water flows in the dispersed annular region give described in Cioncolini and Thome (2012), and using for the other
acoustic velocities slightly higher than those of single-phase gas parameters appearing in Eqs. (5) and (6) the uncertainties provided
flow (Tong and Weisman, 1996). As such, Mach number values pre- in the original studies, the propagated uncertainty for the void frac-
dicted with Eq. (4) can be expected to be slightly conservative, i.e. tion is typically of a few percent and never exceeds 5%, thus con-
the ‘true’ Mach numbers are slightly below those predicted. As firming that predicting the void fraction from the measured liquid
can be seen in Fig. 1, the vast majority of the data points in Table film thickness is a well-conditioned calculation. The inverse process
1 have a predicted local Mach number value below 0.1 and virtually of predicting the liquid film thickness from the void fraction, how-
all data points are below 0.3. Assuming that this same threshold ever, is well known to be highly inaccurate.
normally used in single-phase incompressible flow applications is The main details regarding the additional annular flow data for
applicable for annular flows as well, compressibility effects can then vertical upflow in non-circular channels are summarized in Table
be neglected. 2. In particular, Jones and Zuber (1975) measured the void fraction
As can also be seen in Fig. 1, the gas to liquid density ratio qg q1
l in a rectangular channel (4.98 mm side, 63.5 mm height and aspect
spans about two orders of magnitude from 0.00118 to 0.156. On ratio of 12.75) with water–air at low pressure. Hori et al. (1995)
the other hand, the variations of the gas to liquid viscosity ratio performed their measurements using water in evaporating flow
lg l1
l and the surface tension r are quite limited, being for most conditions in a channel designed to mimic an inner subchannel
of the data within 0.015–0.025 for the viscosity ratio and within of a pressurized water nuclear reactor fuel bundle. Takenaka and
0.070–0.075 Nm1 for the surface tension. The limited variation Asano (2005) tested adiabatic water–air at low pressure in a tight
of the viscosity ratio and surface tension is due to the fact that lattice rod bundle geometry, while Morooka et al. (1989) used low
these parameters depend mostly on the operating temperature, pressure evaporating water in a rod bundle designed to simulate a
which is limited to the range of 293–323 K for most of the data. boiling water nuclear reactor fuel assembly. Finally, Das et al.
The density ratio, on the other hand, depends also on the operating (2002) performed their measurements with adiabatic water–air
pressure, which is characterized by a more pronounced variation, at low pressure in a concentric annulus. It is worth highlighting
as can be seen in Fig. 1. that the new prediction method that will be described in Section
As can be seen in Table 1, the most frequently used measuring 3 is based on the circular tube data of Table 1. The additional data
techniques for the void fraction are quick closing valves, radiation collected in Table 2 will then be used to check the applicability of
attenuation and capacitance sensors. In some of the tests, however, the new prediction method to non-circular channels.
76 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84

Table 2
Additional annular flow data for non-circular channels.

Reference Fluids dhyda (mm) P (MPa) G (kgm2 s1) x e (1) (2) (3) (4) (5)
Jones and Zuber (1975) H2O–Air 9.23 0.11 171–3057 0.01–0.12 0.74–0.89 264 " a 5 RA
Hori et al. (1995) H2O–Steam 10.1 4.9–9.8 1389 0.11–0.18 0.70–0.72 na " d 4 RA
Takenaka and Asano (2005) H2O–Air 4.65 0.11 90–99 0.03–0.12 0.81–0.89 20 " a 5 RA
Morooka et al. (1989) H2O–Steam 12.1 0.49–0.98 833–1389 0.01–0.12 0.71–0.88 na " d 17 RA
Das et al. (2002) H2O–Air 25.4 0.11 141–329 0.01–0.03 0.70–0.79 na " a 9 QCV

(1) – Dimensionless distance L/d of test section inlet from mixer (2 component fluids, adiabatic tests only).
(2) – Flow direction: " = vertical upflow.
(3) – Type of test: a = adiabatic; d = diabatic.
(4) – Number of data points.
(5) – Measuring technique: QCV = quick closing valves; RA = radiation attenuation.
a
Hydraulic diameter (4 Aflow P 1wet , Aflow = flow area, Pwet = wetted perimeter).

0.95

0.9
Void Fraction

Eqs. (15) and (16)

0.85

0.8

Spedding&Nguyen (1976)
0.75 Alia et al. (1965)
Shedd (2010)
0.7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Vapor Quality

Fig. 2. Selected data from Table 1: o = H2O–Air, diameter 45.5 mm, pressure 0.11 MPa, qg q1 l ¼ 1:31  103 ; D = H2O–Ar, diameter 15.0 mm, pressure 2.1 MPa,
qg q1
l ¼ 3:58  102 ; h = R410a, diameter 1.05 mm, pressure 2.6 MPa, qg q1
l ¼ 1:26  101 .

14 1
12 0.9
10 0.8
8 0.7
h

6 0.6
4 0.5
2 0.4
0 -3 -2 -1 0 -3 -2 -1 0
10 10 10 10 10 10 10 10
Density Ratio Density Ratio

0.95

0.9
Void Fraction

0.85

0.8 Density Ratio 10 -3

Density Ratio 10-2
0.75 Density Ratio 10-1
Density Ratio 1
0.7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Vapor Quality

Fig. 3. Fitting functions h and n from Eq. (16) vs. density ratio qg q1
l (top) and void fraction predicted from Eq. (15) vs. vapor quality (bottom).
 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84 77

0.95

Void Fraction: Measured

0.9

0.85

0.8

0.75

0.7 + 5 %
-5%
0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted

0.95
Void Fraction: Measured

0.9

0.85

0.8

0.75

0.7 +5%
-5%
0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted

0.95
Void Fraction: Measured

0.9

0.85

0.8

0.75

0.7 + 5 %
-5%
0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted

Fig. 4. Void fraction: experimental data from Table 1 vs. predictions of the proposed correlation Eqs. (15) and (16) [top] and predictions of Woldesemayat and Ghajar (2007)
[middle] and Lockhart and Martinelli (1949) [bottom] correlations.

3. New prediction method trolled by buoyancy. Annular flows and most separated two-
phase flows, however, cannot be effectively handled by a drift-
As already discussed, the drift-flux model framework has been flux model since in these flows the relative motion is strongly
quite successful in developing general purpose void fraction pre- dependent on the pressure and velocity gradients existing in
diction methods, capable of handling two-phase flows regardless the two phases (Brennen, 2005). As such, the drift-flux model is
of the flow pattern. As noted by Beattie and Sugawara (1986) and not considered appropriate for modeling shear-driven annular
Brennen (2005), however, the drift-flux approach is not particu- flows of interest here, and therefore is not used in the present
larly appropriate for modeling annular flows. Drift-flux models, study.
in fact, are designed to handle distributed and unseparated From inspecting the 68 different void fraction prediction meth-
two-phase flows where one phase is continuous, the other phase ods described in a recent assessment by Woldesemayat and Ghajar
is dispersed and the dominant relative motion between the two (2007), it can be seen that many different parameters have been
phases is caused by an external force field such as gravity. As assumed to influence the void fraction, including the vapor quality,
such, drift-flux models are most useful in the analysis of sedi- the densities and viscosities of the two phases, the surface tension,
mentation, fluidized beds and more generally in the analysis of the operating pressure P, the mass flux, the tube diameter and the
two-phase flows where the relative motion is primarily con- acceleration of gravity, so that schematically:
78 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84

e ¼ eðx; qg ; ql ; lg ; ll ; r; P; G; d; gÞ ð9Þ quality becomes milder and the growth of the void fraction be-
comes more uniform as the vapor quality changes. In the limit of
As already noted, the collected data in Table 1 are essentially lim-
qg q1
l ! 1 the void fraction tends to the vapor quality e ? x+
ited to shear-driven annular flows where gravity plays a marginal or
and the rate of growth loses its dependence on vapor quality. In
negligible role and both the viscosity ratio and the surface tension
conclusion, therefore, the void fraction in Eq. (10) should have a
cover limited ranges. As such, inclusion of the acceleration of gravity
strictly positive first derivative and a non-positive second deriva-
g, viscosity ratio lg l1 and surface tension r into the new predic-
l tive with respect to vapor quality, i.e. oe/ox > 0 and o2e/ox2 6 0. Be-
tion method is not appropriate and these parameters are dropped
sides, as shown in Fig. 2, the void fraction profile versus vapor
out. It is worth highlighting that neglecting these parameters does
quality changes shape as the density ratio varies, so that factorizing
not mean that they have no influence on the void fraction. Rather,
the function in Eq. (10) into the product of two functions, one for
their influence on void fraction, whatever it is, cannot be investi-
each of the independent variables x and qg q1 l , is not appropriate.
gated within the present range of test conditions represented in Ta-
A good candidate to reproduce a growing and saturating trend is
ble 1. For what concerns the other influencing parameters collected
the Hill function, originally proposed in the modeling of biochemi-
in Eq. (9), the ones that are almost invariably included into existing
cal kinetics (Hill, 1910) and frequently used in mathematical phys-
prediction methods and therefore emerge as the most influential are
iology (Murray, 2002). In the present context, in particular, the Hill
the vapor quality x and the density ratio qg q1 l . The vapor quality function is a three parameter function that reads as follows:
represents the fraction of the total mass flow transported by the
n
gas phase. An increase of the vapor quality means that more mass hx
e¼ n; h; k > 0; 0<n<1 ð11Þ
is transported by the gas phase and this in turn normally requires 1 þ kx
a higher void fraction to accommodate the increased gas mass flow
The upperbound on the parameter n guarantees that the second
in the channel. The density ratio, on the other hand, captures pres-
derivative o2e/ox2 is non-positive, as required in the present con-
sure effects on two-phase flows. Thus, a very simple yet reasonable
text. Asymptotical consistency requires that as x ? 1 then
modeling approach for the void fraction is proposed here:
e ? 1, and this condition allows reducing the number of parame-
  ters in Eq. (11) from three to two as follows:
e ¼ e x; qg q1
l ð10Þ
n
hx
Assuming the void fraction to depend only on vapor quality and e¼ ; h > 0; 0<n<1 ð12Þ
1 þ ðh  1Þxn
density ratio can be regarded as a minimal model, which represents
the maximum simplification possible but still retains the essential where the parameters h and n are functions of the density ratio
physics of the problem. Minimal models are typically used as a start- qg q1
l as follows:
ing point in mathematical modeling and are normally upgraded on  a  b
an as-needed basis with the inclusion of additional influencing h ¼ a þ ð1  aÞ qg q1
l ; n ¼ b þ ð1  bÞ qg q1
l ð13Þ
parameters to improve their accuracy and predictive capability.
where a, b, a and b are empirical coefficients to be determined from
In order to get some clue regarding the relation that links the
the experimental data using a nonlinear regression technique. The
void fraction e with the independent variables x and qg q1 l , it is
functional form assumed for h and n in Eq. (13) guarantees that
useful to plot a selection of experimental data from Table 1 as void
both parameters converge to unity as the density ratio goes to
fraction versus vapor quality with the density ratio as parameter,
one, so that in the limit of qg q1 ! 1 then the void fraction
as done in Fig. 2. As can be seen, the void fraction depicts a growing +
l

and saturating trend with vapor quality, with the rate of growth,
e ? x , as required by asymptotical consistency. It can be seen from
Eq. (12) that as x ? 0+ then e ? 0+, so that Eq. (12) is also asymptot-
i.e. oe/ox, modulated by the density ratio. In particular, at low val-
ically consistent in the limit at small vapor quality. The void fraction
ues of the density ratio the rate of growth is a strong function of
prediction method that is being developed here, however, is specif-
vapor quality and the growth is very fast at low vapor qualities
ically designed for annular flows and is therefore limited to void
and very slow at high vapor qualities. As the density ratio in-
fraction values in the range of 0.7–1.0. As such, the asymptotical
creases, however, the dependence of the rate of growth on vapor
consistency of Eq. (12) for x ? 0+ is not actually relevant in the pres-
ent context.
Table 3 Large datasets such as the one collected here in Table 1 almost
Statistical comparison between experimental data of Table 1 and correlations. always contain a tail of outliers, anomalous records that deviate
(1) (2) (3) (4) (5) (6) from the behavior of the bulk of the data and are due to measuring
errors, measurements carried out beyond the range of optimal per-
Homogeneous model 7.4 9.4 7.4 13.1 43.6 75.7
Lockhart and Martinelli (1949) 2.9 4.1 1.5 46.4 85.8 96.0 formance of the experimental stand, sensor malfunctions or errors
Baroczy (1966) 8.4 10.6 7.8 19.9 35.1 63.2 in the recording, post processing or reporting of the data. Addi-
Zivi (1964) 5.7 9.3 4.5 39.7 65.4 82.0 tional sources of potential outliers include measurements taken
Smith (1969) 3.4 4.6 1.8 42.9 76.4 96.1 close to the intermittent to annular flow transition and measure-
Armand and Treschev (1947) 4.3 4.9 3.6 18.5 65.2 98.2
Woldesemayat and Ghajar (2007) 2.3 2.9 1.3 53.4 92.9 99.5
ments affected by a residual dependence on inlet conditions. Since
DIX (Chexal et al., 1986) 3.7 4.5 3.3 25.0 75.0 98.4 it is well known that even a few outliers can have a disproportion-
Rouhani and Axelsson (1970) 5.8 6.4 5.6 8.5 38.3 95.4 ate effect on the analytical results derived from the statistical anal-
Morooka et al. (1989) 5.9 6.9 4.8 12.8 38.1 91.3 ysis of the data, it is normally wise to properly handle these
Bestion (1990) 4.0 5.9 0.3 37.7 73.1 93.5
anomalous records during the modeling of the data. Here, in partic-
Ishii et al. (1976) 4.8 6.9 2.1 35.0 65.5 88.1
Present study Eqs. (15) and (16) 1.8 2.9 0.4 70.1 92.0 98.7 ular, a robust fitting technique and a simple model-based outlier
detection algorithm (Pearson, 2005) are used to estimate the
Pn jeexp ecal j
(1) – Mean absolute percentage error (%) 1
100
neexp . regression coefficients a, b, a and b in Eq. (13) from the data. First,
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P e e 2 a preliminary set of values for the regression coefficients a, b, a and
(2) – Root mean squared percentage error (%) 100 1n n1 expeexp cal .
(3) – 100
Mean percentage error (%) n
Pn eexp ecal b is derived using the entire databank in Table 1, outliers included,
1 eexp .
(4) – Percentage of experimental data captured within ±2%. employing the least-absolute-residuals robust regression tech-
(5) – Percentage of experimental data captured within ±5%. nique. By minimizing the sum of the absolute values of the resid-
(6) – Percentage of experimental data captured within ±10%. uals, instead of the sum of the squares of the residuals as
 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84 79

happens with least-squares methods, this robust technique is less ifjR  medianðRÞj > 2  IQD ) Current record is outlier ð14Þ
affected by extreme points in the databank and therefore better
captures the trend of the bulk of the data. The superior perfor- where the residual R is the difference between the measured void
mance of robust fitting techniques, such as the one used here, with
fraction value and the prediction of the preliminary fit generated
respect to least-squares when dealing with large and contaminated in the first step of the analysis. The median of the residuals provides
databanks is well documented in the statistics literature (among
an outlier-resistant estimate of the nominal residual value, while
others: Moler, 2004; Pearson, 2005; Aster et al., 2012). An example, the interquartile distance IQD provides an outlier-resistant estimate
showing a comparison between least-squares and least-absolute-
of the scale of the natural variation of the residuals about the nom-
residuals fitting with contaminated data, is included in the Elec- inal value. According to this decision rule, a data point is identified
tronic Annex, available in the online version of this article. Once
as an outlier if its residual deviates from the median of the residuals
a preliminary fitting equation is available, every data point in the by more than two times the interquartile distance of the residuals.
databank is tested to determine whether it is an outlier according
Once outliers have been identified, they are removed from the data-
to the following outlier detection rule, called standard symmetric bank and the preliminary set of values for the regression coeffi-
boxplot (Pearson, 2005):
cients a, b, a and b obtained in the first step is refined using the

Anderson & Mantzouranis (1960) Beggs (1972)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
+5%
0.75 0.75
0.7 0.7 + 5 %
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Celata & Frazzoli (1981) Alia et al. (1965)

1 1
Void Fraction: Measured

Void Fraction: Measured

Uniform Heat Flux

0.95 Peaked Heat Flux 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
+5% +5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Dejesus & Kawagi (1990) Godbole et al. (2011)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 +5% 0.75
+5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Kaji & Azzopardi (2010) Leung et al. (2005)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 +5%
0.75
+5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Fig. 5. Void fraction: measured data from Table 1 vs. predictions of proposed correlation Eqs. (15) and (16).
80 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84

clean data only. The final values of the regression coefficients are It is worth highlighting that outliers are removed from the data-
a = 2.129, b = 0.3487, a = 0.2186 and b = 0.5150, so that the void bank of Table 1 only to derive the regression coefficients, but in all
fraction correlating equation that is finally obtained is: the comparisons that will be presented below, the entire databank
of Table 1 is always used, outliers included. Since outliers are not
hx
n identified here based on solid physical or experimental arguments,
e¼ ; 0 < x < 1; 103 < qg q1
l < 1; it is considered inappropriate to exclude them from the databank.
1 þ ðh  1Þxn
It is worth remembering that Eqs. (15) and (16) are appropriate for
0:7 < e < 1 ð15Þ shear-driven annular flows only, where the effect of gravity on the
 0:2186 flow is negligible. The trend of the parameters h and n versus den-
h ¼ 2:129 þ 3:129 qg q1
l ; sity ratio qg q1
l in Eq. (16) is presented in Fig. 3 (top-left and top-
right respectively), while the trends of the void fraction e in Eq.
 0:5150
n ¼ 0:3487 þ 0:6513 qg q1 ð16Þ (15) versus vapor quality x and density ratio qg q1 l are shown in
l
Fig. 3 (bottom). As can be seen, the predicted void fraction profiles

Mukherjee (1979) Spedding&Nguyen (1976)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
+5% +5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Sujumnong (1998) Ueda (1967)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
+5% +5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Shedd (2010) Silvestri et al. (1963)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
+5% +5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Adorni et al. (1963) Casagrande et al. (1963)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
+5% +5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Fig. 6. Void fraction: measured data from Table 1 vs. predictions of proposed correlation Eqs. (15) and (16).
 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84 81

Cravarolo et al. (1964) Alia et al. (1966)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 +5% 0.75 +5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Gill et al. (1964,1965) Hall-Taylor et al. (1963)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
+5% +5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Whalley et al. (1974) Brown (1978)

1 1
Void Fraction: Measured

Void Fraction: Measured

0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
+5% +5%
0.7 0.7
-5% -5%
0.65 0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted Void Fraction: Predicted

Wurtz (1978)
1
Void Fraction: Measured

0.95
0.9
0.85
0.8
0.75
+5%
0.7
-5%
0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted

Fig. 7. Void fraction: measured data from Table 1 vs. predictions of proposed correlation Eqs. (15) and (16).

are growing and saturating with vapor quality at a rate of growth ularly small, indicating that in the present context the robust fit-
modulated by the density ratio and with the correct asymptotic ting technique used in the first step of the analysis is indeed
trends, as required. quite effective in capturing the trend of the bulk of the data. Be-
The procedure described above for automatic outlier detection sides its efficiency (Pearson, 2005), the outlier detection algorithm
identifies 245 data points as outliers, corresponding to a dataset used here was also selected because it can be easily coded into
contamination of 9.3% that appears reasonable, considering that most computing software tools using built-in features only. Be-
contamination levels of 1–10% and above are quite common in sides this technique, many other outlier detection algorithms are
large datasets (Pearson, 2005). It is worth highlighting that the re- also available in the robust statistics literature, as this topic is cur-
moval of the outliers from the databank in Table 1 yields only a rently experiencing a growing interest for data-mining applica-
minimal variation of the regression model coefficients. Robust tions, such as credit card fraud detection, network intrusion
techniques are not 100% immune from data contamination, they detection, clinical trials and voting irregularity analysis. Robust fit-
are just less vulnerable. Typically, the removal of the outliers yields ting techniques and outlier detection algorithms, such as those
a correction to the model coefficients that may be small but not used here, allow getting the most out of noisy and/or contaminated
negligible. In the present case, however, this correction was partic- large databanks that cannot be cleaned prior to processing with
82 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84

purely physical arguments due to incomplete knowledge of the although some of which are the same as used here. Nonetheless,
experiments that generated the data. The same fitting procedure the good correlating capability of Eqs. (15) and (16) suggests that
used here has already been used with success by the authors in the void fraction in annular flow is essentially dependent on vapor
processing a large entrained liquid fraction database (Cioncolini quality x and density ratio qg q1 l , so that all the other parameters
and Thome, 2012). As already noted, Eqs. (15) and (16) can be re- that can be relevant in other flow patterns seem to lose their
garded at this point as a minimal model for predicting the void importance when it comes to annular flows.
fraction in annular flows. It will be shown in the following Section As shown here by the good performance of the drift-flux corre-
4, however, that within the limits of the present study no further lation of Woldesemayat and Ghajar (2007), drift-flux models can
upgrade to Eqs. (15) and (16) seems necessary. handle annular flows quite effectively, even though their physics
is in principle not appropriate for separated flows. Nonetheless,
4. Results and discussion drift-flux models make use of many additional flow parameters
that appear to be of second order importance, as demonstrated
The comparison of the measured void fraction data from Table 1 by the new prediction method proposed here, which depends only
with the predictions of Eqs. (15) and (16) is presented in Fig. 4 on vapor quality and the density ratio, and by the good perfor-
(top). The statistical comparison between measured data from Ta- mance of the slip ratio correlation of Lockhart and Martinelli
ble 1 and predictions is reported in Table 3, which includes also the (1949), which depends on vapor quality and the density and vis-
results of a selection of other frequently used void fraction predic- cosity ratios.
tion methods, such as the homogeneous model and those of Lock- Further comparisons of the measured void fraction data in Table
hart and Martinelli (1949), Baroczy (1966), Zivi (1964), Smith 1 with the predictions of Eqs. (15) and (16) are presented in Figs.
(1969), Armand and Treschev (1947), Woldesemayat and Ghajar 5–7, where each data subset from Table 1 is separately presented.
(2007), DIX (Chexal et al., 1986), Morooka et al. (1989), Bestion Notwithstanding some scatter and some outliers in a few data sub-
(1990) and Ishii et al. (1976). Further details regarding these corre- sets, the agreement of the data with the predictions is for the most
lations and their implementation can be found in the Electronic part quite satisfactory. In Fig. 5 (second row, left) the data of Celata
Annex. As can be seen in Table 3, the new prediction method pro- and Frazzoli (1981) taken in evaporating flow conditions are sys-
posed here fits the available data better than the other methods tematically overpredicted except at very high void fraction values.
considered, with a mean absolute percentage error of 1.8%, a root According to them, their data are systematically low due to the cal-
mean squared percentage error of 2.9% and 7 points out of 10 cap- ibration of the neutron attenuation instrument used for the mea-
tured to within ±2% (in the present study, the correlations are surements, which was performed assuming ideal annular flow
ranked on the basis of their mean absolute percentage error, root without entrained liquid droplets in the vapor core. The measure-
mean squared percentage error, mean percentage error and ments carried out with a uniform heat flux profile and also with an
according to the percentage of data captured within ±2%, ±5% outlet peaked heat flux profile appear consistent, indicating that
and ±10%, considering as not significant any difference of less than the axial shape profile of the supplied power is apparently not
1% in the computed values). Since both the mean absolute percent- affecting the void fraction. The bias in the data, however, does
age error and the root mean squared percentage error of the new not allow any definite conclusion to be drawn on the potential
prediction method are comparable with the uncertainty of the void influence of the evaporating process on the void fraction. The only
fraction measurements, the inclusion into Eqs. (15) and (16) of fur- other data subset taken in evaporating flow conditions is provided
ther parameters to improve their accuracy is considered not neces- by Leung et al. (2005) and is shown in Fig. 5 (bottom row, right) but
sary. The best prediction by existing methods is given by the drift- is too limited and scattered to help resolve the issue. It is recom-
flux correlation of Woldesemayat and Ghajar (2007), followed by mended, therefore, to obtain additional data in evaporating flow
the slip ratio model of Lockhart and Martinelli (1949). The compar- conditions to duly assess the potential influence of the evaporating
isons of the measured data from Table 1 with the predictions of process on void fraction, although on the basis of the limited data
these two methods are presented in Fig. 4 (middle and bottom). available here it seems that the effect will be of second order
Analogous diagrams for the other correlations considered are in- importance.
cluded in the Electronic Annex. Direct comparison of the prediction In Fig. 6 (third row, left), the microscale data provided by Shedd
method proposed here with other correlations is somewhat unfair, (2010) are quite nicely reproduced by the proposed correlation
as the proposed method has been designed with the present data- Eqs. (15) and (16). Since the new prediction method does not de-
bank while the other correlations are based on different data sets, pend on the tube diameter and satisfactorily fits both macroscale

0.95

0.9
Void Fraction

Eqs. (15) and (16)

0.85

0.8 P=1.72 MPa

P=4.14 MPa
P=8.62 MPa
0.75 P=14.5 MPa
P=20.7 MPa
0.7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Vapor Quality

Fig. 8. Void fraction: independent measurements of Thom (1964) vs. vapor quality and predictions of the proposed correlation Eqs. (15) and (16).
 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84 83

1
Jones & Zuber (1975) +5%
0.95 Hori et al. (1995)

Void Fraction: Measured

Takenaka & Asano (2005) -5%
0.9 Morooka et al. (1989)
Das et al. (2002)
0.85

0.8

0.75

0.7

0.65
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Void Fraction: Predicted

Fig. 9. Void fraction: experimental data from Table 2 for non-circular channels vs. predictions of the proposed correlation Eqs. (15) and (16).

and microscale data, this appears to provide evidence that there is than existing correlations, works also for evaporating flows and
no macro-to-microscale transition when it comes to annular flow, extrapolates to non-circular channels. Importantly, there is no evi-
at least down to a diameter of about 1.0 mm. Similar findings were dence of a macro-to-microscale transition in the present study
reported by Cioncolini and Thome (2011) in the context of convec- down to a diameter of about 1.0 mm, although more microscale
tive boiling modeling for shear-driven annular flows. However, measurements are necessary to confirm this preliminary
more microscale measurements and smaller diameter data are conclusion.
necessary to properly support this preliminary conclusion. It is
worth highlighting that this preliminary conclusion is limited to
Acknowledgements
annular flow. As noted by Shedd (2010), with other flow patterns
such as bubbly flow and elongated bubble/slug flows, as the chan-
G.P. Celata (ENEA), A. Ghajar (Oklahoma State University) and T.
nel diameter decreases the void fraction behavior approaches that
Shedd (University of Wisconsin) are gratefully acknowledged for
of a homogeneous mixture, indicating that a reduction in channel
providing their void fraction data. A. Cioncolini is supported by
size prevents the phases from developing a slip.
the Swiss National Science Foundation (SNSF) under Contract No.
The predictions of Eqs. (15) and (16) are included in Fig. 2,
200020-129624/1.
where the selection of experimental data from Table 1 used in void
fraction modeling is plotted as void fraction versus vapor quality
with the density ratio as parameter. As can be seen, the new pre- Appendix A. Supplementary data
diction method nicely captures the trends in the data.
Independent measurements (water–steam, pressure of 1.72– Supplementary data associated with this article can be found, in
20.7 MPa, horizontal circular tube, diameter of 25.4 mm) provided the online version, at http://dx.doi.org/10.1016/j.ijmultiphaseflow.
by Thom (1964), but not included in the databank of Table 1, are 2012.03.003.
compared with the new prediction method in Fig. 8. As can be seen,
the proposed method fits these independent data quite satisfacto-
rily, even when extrapolating to the operating pressure of References
20.7 MPa, corresponding to a density ratio qg q1 l ¼ 0:406 that is
Adorni, N., Casagrande, I., Cravarolo, L., Hassid, A., Pedrocchi, E., Silvestri, M., 1963.
well outside the range covered by the databank of Table 1. Further investigations in adiabatic dispersed flow: pressure drop and film
Finally, the void fraction data for non-circular channels in Table thickness measurements with different channel geometries-analysis of the
2 are compared with the predictions of the proposed correlation in influence of geometrical and physical parameters. CISE Report R-53, Segrate,
Italy.
Fig. 9. Although the data are quite limited, the satisfactory compar- Alia, P., Cravarolo, L., Hassid, A., Pedrocchi, E., 1965. Liquid volume fraction in
ison suggests that not only the channel size but also its cross sec- adiabatic two-phase vertical upflow – round conduit. CISE Report R-105,
tional shape is of second order importance when it comes to the Segrate, Italy.
Alia, P., Cravarolo, L., Hassid, A., Pedrocchi, E., 1966. Phase and velocity distribution
void fraction in annular flows, the latter which requires more in two-phase adiabatic annular dispersed flow. CISE Report R-109, Segrate, Italy.
experimental investigation. Anderson, G.H., Mantzouranis, B.G., 1960. Two-phase (gas–liquid) flow phenomena
– I: Pressure drop and hold-up for two-phase flow in vertical tubes. Chem. Eng.
Sci. 12, 109–126.
5. Conclusions Armand, A.A., Treschev, G.G., 1947. Investigation of the resistance during the
movement of steam–water mixtures in heated boiler pipes at high pressure. Izv.
Ves. Teplotekh. Inst. 4, 1–5.
A new prediction method for the void fraction in annular two- Aster, R.C., Borchers, B., Thurber, C.H., 2012. Parameter Estimation and Inverse
phase flow has been proposed. The underlying experimental data- Problems, second ed. Academic Press.
Baroczy, C.J., 1966. A systematic correlation for two-phase pressure drop. Chem.
base contains 2633 data points for circular tubes covering both Eng. Progr. Sym. Ser. 62, 232–249.
macroscale and microscale flow conditions and 40 additional data Beattie, D.R.H., Sugawara, S., 1986. Steam–water void fraction for vertical upflow in
points for non-circular channels. The new prediction method is a 73.9 mm pipe. Int. J. Multiphase Flow 12, 641–653.
Beggs, H.D., 1972. An experimental study of two-phase flow in inclined pipes. Ph.D.
strongly simplified with respect to most existing correlations, as
Thesis, University of Tulsa, Oklahoma, USA.
it depends only on vapor quality and the gas to liquid density ratio, Bestion, D., 1990. The physical closure laws in the CATHARE code. Nucl. Eng. Des.
and covers vapor qualities from 0 to 1, density ratios from 0.001 to 124, 229–245.
1, channel diameters from 1.05 mm to 45.5 mm, mass fluxes from Brennen, C.E., 2005. Fundamentals of Multiphase Flow. Cambridge Iniversity Press,
New York, USA.
22 kgm2 s1 to 3420 kgm2 s1 and pressures from 0.11 MPa to Brown D.J., 1978. Disequilibrium annular flow. Ph.D. Thesis, University of Oxford,
20.7 MPa. The new method reproduces the available data better England.
84 A. Cioncolini, J.R. Thome / International Journal of Multiphase Flow 43 (2012) 72–84

Casagrande, I., Cravarolo, L., Hassid, A., Pedrocchi, E., 1963. Adiabatic dispersed two- Lockhart, R.W., Martinelli, R.C., 1949. Proposed correlation of data for isothermal
phase flow: further results on the influence of physical properties on pressure two-phase two-component flow in pipes. Chem. Eng. Progr. 45, 39–48.
drop and film thickness. CISE Report R-73, Segrate, Italy. Moler, C.B., 2004. Numerical Computing with MATLAB. SIAM, USA.
Celata, G.P., Frazzoli, F.V., 1981. High void fraction measurements in full-scale heat Morooka, S., Ishizuka, T., Iizuka, M., Yoshimura, K., 1989. Experimental study on
transfer tests. Nucl. Technol. 54, 422–425. void fraction in a simulated BWR fuel assembly. Nucl. Eng. Des. 114, 91–98.
Chexal, B., Horowitz, J., Lellouche, G., 1986. An assessment of eight void fraction Mukherjee, H., 1979. An experimental study of inclined two-phase flow. Ph.D.
models for vertical flows. EPRI Report NSAC-107, Palo Alto, USA. Thesis, University of Tulsa, Oklahoma, USA.
Chung, P.M.Y., Kawaji, K., 2004. The effect of channel diameter on adiabatic two- Murray, J.D., 2002. Mathematical Biology. Springer-Verlag, New York.
phase flow characteristics in microchannels. Int. J. Multiphase Flow 30, 735– Pearson, R.K., 2005. Mining Imperfect Data. SIAM, USA.
761. Rouhani, S.Z., Axelsson, E., 1970. Calculation of void volume fraction in the
Cioncolini, A., Thome, J.R., 2011. Algebraic turbulence modeling in adiabatic and subcooled and quality boiling regions. Int. J. Heat Mass Transfer 13, 383–393.
evaporating annular two-phase flow. Int. J. Heat Fluid Flow 32, 805–817. Serizawa, A., Feng, Z., Kawara, Z., 2002. Two-phase flow in microchannels. Exp.
Cioncolini, A., Thome, J.R., 2012. Entrained liquid fraction prediction in adiabatic Therm. Fluid Sci. 26, 703–714.
and evaporating annular two-phase flow. Nucl. Eng. Des. 243, 200–213. Shedd, T.A., 2010. Void fraction and pressure drop measurements for refrigerant
Cioncolini, A., Thome, J.R., Lombardi, C., 2009a. Algebraic turbulence modeling in R410a flows in small diameter tubes. Preliminary AHRTI Report No. 20110-01.
adiabatic gas–liquid annular two-phase flow. Int. J. Multiphase Flow 35, 580– Silvestri, M., Casagrande, I., Cravarolo, L., Hassid, A., Bertoletti, S., Lombardi, C.,
596. Peterlongo, G., Soldaini, G., Vella, G., Perona, G., Sesini, R., 1963. A research
Cioncolini, A., Thome, J.R., Lombardi, C., 2009b. Unified macro-to-microscale program in two-phase flow. CISE Report, Segrate, Italy.
method to predict two-phase frictional pressure drops of annular flows. Int. J. Smith, S.L., 1969. Void fractions in two-phase flow: a correlation based upon an
Multiphase Flow 35, 1138–1148. equal velocity head model. Proc. Inst. Mech. Eng., London 184, 647–657 (Part I).
Coddington, P., Macian, R., 2002. A study of the performance of void fraction Spedding, P.L., Nguyen, V.T., 1976. Data on hold-up, pressure loss and flow pattens
correlations used in the context of drift-flux two-phase flow models. Nucl. Eng. for two-phase air–water flow in an inclined pipe. Eng. Report 122, University of
Des. 215, 199–216. Auckland, New Zeland.
Cravarolo, L., Hassid, A., Pedrocchi, E., 1964. Further investigation on two-phase Sujumnong, M., 1998. Heat transfer, pressure drop and void fraction in two-phase,
adiabatic annular-dispersed flow: effect of length and some inlet conditions on two component flow in a vertical tube. Ph.D. Thesis, University of Manitoba,
flow parameters. CISE Report R-93, Segrate, Italy. Canada.
Das, G., Das, P.K., Purohit, N.K., Mitra, A.K., 2002. Liquid holdup in concentric annuli Takenaka, N., Asano, H., 2005. Quantitative CT-reconstruction of void fraction
during cocurrent gas–liquid upflow. Canadian J. Chem. Eng. 80, 153–157. distributions in two-phase flow by neutron radiography. Nucl. Instrum.
Dejesus, J.M., Kawagi, M., 1990. Investigation of interfacial area and void fraction in Methods Phys. Res. A 542, 387–391.
upward cocurrent gas–liquid flow. Canadian J. Chem. Eng. 68, 904–912. Thom, J.R.S., 1964. Prediction of pressure drop during forced circulation boiling of
Gill, L.E., Hewitt, G.F., Lacey, P.M.C., 1964. Sampling probe studies of the gas core in water. Int. J. Heat Mass Transfer 7, 709–724.
annular two-phase flow-II: Studies of the effect of phase flow rates on phase Tong, L.S., Weisman, J., 1996. Thermal Analysis of Pressurized Water Reactors.
and velocity distribution. Chem. Eng. Sci. 19, 665–682. American Nuclear Society, USA.
Gill, L.E., Hewitt, G.F., Lacey, P.M.C., 1965. Data on the upwards annular flow of air– Triplett, K.A., Ghiaasiaan, S.M., Abdel-Khalik, S.I., LeMouel, A., McCord, B.N., 1999.
water mixtures. Chem. Eng. Sci. 20, 71–88. Gas–liquid two-phase flow in microchannels. Part II: Void fraction and pressure
Godbole, P.V., Tang, C.C., Ghajar, A.J., 2011. Comparison of void fraction correlations drop. Int. J. Multiphase Flow 25, 395–410.
for different flow patterns in upward vertical two-phase flow. Heat Transfer Ueda, T., 1967. On upward flow of gas–liquid mixtures in vertical tubes. Bull. JSME
Eng. 32, 843–860. 10, 989–999.
Hall-Taylor, N., Hewitt, G.F., Lacey, P.M.C., 1963. The motion and frequency of large Vijayan, P.K., Patil, A.P., Pilkhwal, D.S., Saha, D., Venkat Raj, V., 2000. An assessment
disturbance waves in annular two-phase flow of air–water mixtures. Chem. of pressure drop and void fraction correlations with data from two-phase
Eng. Sci. 18, 537–552. natural circulation loops. Heat Mass Transfer 36, 541–548.
Hill, A.V., 1910. The possible effects of the aggregation of the molecules of Wallis, G.B., 1961. Flooding velocities for air and water in vertical tubes. United
hemoglobin on its dissociation curves. J. Physiol. 40, iv–vii. Kingdom Atomic Energy Authority, Report AEEW-R-123, England.
Hori, K., Akiyama, Y., Miyazaki, K., Kurosu, T., Sugiyama, S., 1995. Void fraction in a Whalley, P.B., Hewitt, G.F., Hutchinson, P., 1974. Experimental wave and
single channel simulating one subchannel of a PWR fuel assembly. In: Celata, entrainment measurements in vertical annular two-phase flow. Symp. Multi-
G.P., Shah, R.K. (Eds.), Two-Phase Flow Modeling and Experimentation. ETS, Phase Flow Systems, University of Strathclyde, Scotland paper A1.
Italy, pp. 1013–1027. Woldesemayat, M.A., Ghajar, A.J., 2007. Comparison of void fraction correlations for
Ishii, M., Chawla, T.C., Zuber, N., 1976. Constitutive equation for vapor drift velocity different flow patterns in horizontal and upward inclined pipes. Int. J.
in two-phase annular flow. AIChE J. 22, 283–289. Multiphase Flow 33, 347–370.
Jones, O.C., Zuber, N., 1975. The interrelation between void fraction fluctuations and Wolf, A., Jayanti, S., Hewitt, G.F., 2001. Flow development in vertical annular flow.
flow patterns in two-phase flow. Int. J. Multiphase Flow 2, 273–306. Chem. Eng. Sci. 56, 3221–3235.
Kaji, R., Azzopardi, B.J., 2010. The effect of pipe diameter on the structure of gas/ Würtz, J., 1978. An experimental and theoretical investigation of annular steam–
liquid flow in vertical pipes. Int. J. Multiphase Flow 36, 303–313. water flow in tubes and annuli at 30 and 90 bar. Risø National Laboratory,
Kawahara, A., Chung, P.M.Y., Kawaji, M., 2002. Investigation of two-phase flow Report No. 372, Denmark.
pattern, void fraction and pressure drop in a microchannel. Int. J. Multiphase Zivi, S.M., 1964. Estimation of steady-state steam void fraction by means of the
Flow 28, 1411–1435. principle of minimum entropy production. ASME J. Heat Transfer 86, 247–252.
Leung, L.K.H., Groeneveld, D.C., Teyssedou, A., Aubé, F., 2005. Pressure drop for Zuber, N., Findlay, J., 1965. Average volumetric concentration in two-phase systems.
steam and water flow in heated tubes. Nucl. Eng. Des. 235, 53–65. ASME J. Heat Transfer 87, 453–468.
Levy, S., 1999. Two-Phase Flow in Complex Systems. Wiley, New York.