Вы находитесь на странице: 1из 10

International Journal of Heat and Fluid Flow 69 (2018) 33–42

Contents lists available at ScienceDirect

International Journal of Heat and Fluid Flow


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

Drift-flux correlation for gas-liquid two-phase flow in a horizontal pipe T


a, b
Somboon Rassame *, Takashi Hibiki
a
Department of Nuclear Engineering, Faculty of Engineering, Chulalongkorn University, Wangmai, Patumwan, Bangkok 10330, Thailand
b
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

Keywords: A drift-flux correlation has been often used to predict void fraction of gas-liquid two-phase flow in a horizontal
Drift-flux model channel due to its simplicity and practicality. The drift-flux correlation includes two important drift-flux para-
Void fraction meters, namely, the distribution parameter and void-fraction-weighted-mean drift velocity. In this study, an
Gas-liquid flow extensive literature survey for horizontal two-phase flow is conducted to establish void fraction database and to
Horizontal flows
acquire existing drift-flux correlations. A total of 566 data is collected from 12 data sources and 4 flow-regime-
dependent and 1 flow-regime-independent drift-flux correlations are identified. The predictive capability of the
existing drift-flux correlations is assessed using the collected data. It is pointed out that the drift velocity de-
termined by a regression analysis may include a significant error due to a compensation error between dis-
tribution parameter and drift velocity. In this study, a simple flow-regime-independent drift-flux correlation is
developed. In the modeling approach, the void-fraction-weighted mean drift velocity is approximated to be 0 m/
s, whereas the distribution parameter is given as a simple function of the ratio of non-dimensional superficial gas
velocity to non-dimensional mixture volumetric flux. The newly developed correlation shows an excellent
predictive capability of void fraction for horizontal two-phase flow. Mean absolute error (or bias), standard
deviation (random error), mean relative deviation and mean absolute relative deviation of the correlation are
0.0487, 0.0985, 0.0758 and 0.206, respectively. The prediction accuracy of the correlation is similar to the
correlation of Chexal et al. (1991), which was formulated based on the drift-flux parameters by means of many
cascading constitutive relationships with numerous empirical parameters.

1. Introduction wavy-stratified and annular flow regimes. Lamari (2001) developed a


similar drift-flux correlation for those four flow regimes with different
The gas-liquid two-phase flow in a horizontal channel is often en- sets of the drift-flux parameters. Silva et al. (2011) proposed a drift-flux
countered in apparatuses, plants and pipelines in various industries correlation for intermittent flow regime, while Kong and Kim (2017)
such as chemical, petroleum, and nuclear industries. The void fraction developed a drift-flux correlation for dispersed bubble flow regime.
is one of most important parameters to characterize horizontal gas-li- Although the drift-flux correlations mentioned above are flow-regime-
quid two-phase flows and is a critical parameter for the design of in- dependent correlations, they provide the insight on how the flow re-
dustrial systems. A correlation based on the drift-flux model proposed gime affects the drift-flux parameters. However, most of the existing
by Zuber and Findlay (1965) has been used to predict void fraction in empirical drift-flux correlations were developed based on the re-
horizontal gas-liquid two-phase flows due to its simplicity and practi- searchers’ own data or a limited number of collected data, and the
cality. The drift-flux correlation includes two-important drift-flux applicability of the correlations to a wide range of flow conditions has
parameters such as the distribution parameter characterizing the phase not been well-examined.
distribution and the void-fraction-weighted mean drift velocity char- For the purpose of implementing a drift-flux correlation into one-
acterizing the relative velocity between phases. These drift-flux para- dimensional thermal-hydraulic system analysis code such as RELAP5
meters should be given by either a phenomenological model or an (Information Systems Laboratories, inc., 2001), Chexal et al. (1991)
empirical correlation. developed a flow-regime-independent drift-flux correlation applicable
Franca and Lahey (1992) proposed a drift-flux correlation for hor- for a whole range of void fraction. Chexal et al. (1991) adopted an
izontal gas-liquid two-phase flows. The distribution parameter is given approach to formulate the drift-flux parameters by means of many
by a constant value depending on the flow regime such as plug, slug, cascading constitutive relationships with numerous empirical


Corresponding author.
E-mail address: somboon.ra@chula.ac.th (S. Rassame).

https://doi.org/10.1016/j.ijheatfluidflow.2017.11.002
Received 18 August 2017; Received in revised form 31 October 2017; Accepted 19 November 2017
Available online 01 December 2017
0142-727X/ © 2017 Elsevier Inc. All rights reserved.
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

Nomenclature Δρ Density difference between phases [kg/m3]


μ Viscosity [Pa•s]
A Flow area [m2] ρ Density [kg/m3]
C0 Distribution parameter [-] σ Surface tension [N/m]
C∞ Asymptotic value of distribution parameter [-]
D Diameter [m] Subscripts
DH Hydraulic equivalent diameter [m]
g Gravitational acceleration [m/s2] cal. Calculated value
j Superficial velocity [m/s] exp. Experimental value
md Mean absolute error [-] f Liquid phase
mrel Mean relative deviation [-] g Gas phase
mrel, ab Mean absolute relative deviation [-]
N Number of samples [-] Superscripts
Re Reynolds number [-]
sd Standard deviation [-] + Non-dimensional quantity
vgj Drift velocity [m/s]
Mathematical symbols
Greek symbols
<> Area averaged properties
α Void fraction [-] <<>> Void-weight area-averaged properties

parameters, and sacrificed a physics principle characterizing the drift- αvgj


vgj =
flux parameters in exchange for developing the flow-regime-in- α (4)
dependent drift-flux correlation. The drift-flux correlation proposed by
Chexal et al. (1991) may be considered a kind of regression curve to αj
C0 =
reproduce the void fraction data. α j (5)
As reviewed above, most of the flow-regime-dependent drift-flux
where jg and α are the superficial gas velocity and void fraction, re-
correlations adopt a flow-regime-dependent constant value for the
spectively.
distribution parameter, which implies that the recommended distribu-
The distribution parameter and drift velocity can be, respectively,
tion parameter may not include the effects of physical properties and
obtained by Eqs. (5) and (4) if the distributions of local void fraction
flow kinematics on the distribution parameter. The flow-regime-in-
and gas and liquid velocities are measured (Hibiki and Ishii, 2002).
dependent correlation may not provide a clear understanding of de-
However, these distributions are often not available. Zuber and
pendent parameters on the drift-flux parameters. In view of these, this
Findlay (1965) proposed an alternative method to determine the dis-
study revisits the development of a drift-flux correlation for horizontal
tribution parameter and drift velocity. If a linear relationship between
gas-liquid two-phase flows with extensive data sets collected in a wide
〈j〉 and 〈〈vg〉〉 exists in a test condition, the distribution parameter and
range of flow conditions. The predictive capability of the newly de-
drift velocity can be obtained as the slope and intercept in the drift-flux
veloped drift-flux correlation is compared with that of the correlation
plot of 〈j〉 and 〈〈vg〉〉, respectively. However, a “compensation error” is
developed by Chexal et al. (1991).
common in this approach when the distribution parameter and drift
velocity are determined. The uncertainty of one parameter (for ex-
2. One-dimensional drift-flux model formulation and existing ample, distribution parameter) directly affects the value of another
drift-flux correlations for horizontal two-phase flow parameter (for example, drift velocity). If data are taken at high mixture
volumetric flux conditions, it is difficult to determine the drift velocity
2.1. One-dimensional formulation of drift-flux model accurately with a drift-flux plot. Therefore, some efforts to model the
drift velocity based on drag law have been made to develop a con-
Zuber and Findley (1965) introduced a concept of “drift velocity” to stitutive equation for predicting the drift velocity (Ishii, 1977; Hibiki
take into account the velocity difference between two phases. Local and Ishii, 2003).
drift velocity, vgj, is defined as the difference between gas velocity, vg, The approximation of the linear relationship between 〈j〉 and 〈〈vg〉〉
and mixture volumetric flux, j, as: is found to be reasonable in some test conditions such as vertical slug
and churn flows in confined channels (Mishima and Hibiki, 1996; Ishii,
vgj = vg − j (1) 1977; Ozaki and Hibiki, 2015). However, it is pointed out that the
approximation may not work for some test conditions such as two-
Averaging Eq. (1) over a flow channel yields one-dimensional drift-
phase flows in large size channels at low flow conditions (Hibiki and
flux model as:
Ishii, 2003; Clark et al., 2014), subcooled boiling flows (Ishii, 1977;
vg = C0 j + vgj (2) Hibiki et al., 2003; Brooks et al., 2012) and bubbly flows (Hibiki and
Ishii, 2002). Ishii and Hibiki (2010) provided a brief review of existing
where 〈〈〉〉 and 〈〉 are the void-fraction-weighted mean quantity and constitutive equations for distribution parameter and drift velocity in
cross-sectional area-averaged quantity over the flow channel, respec- various flow channels.
tively. The void-fraction-weighted mean gas velocity (hereafter, gas
velocity for simplicity), 〈〈vg〉〉, distribution parameter and void-frac- 2.2. Existing drift-flux correlations for horizontal gas-liquid two-phase flows
tion-weighted mean drift velocity (hereafter, drift velocity for simpli-
city), 〈〈vgj〉〉, are defined as Eqs. (3)–(5), respectively. In this section, some existing drift-flux correlations developed for
jg horizontal gas-liquid two-phase flows will be reviewed briefly. The
vg = summary of the reviewed correlations is tabulated in Table 1.
α (3)
Chexal et al. (1991) developed a flow-regime-independent drift-flux

34
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

Table 1
Existing drift-flux correlations developed for the horizontal gas-liquid two-phase flow.

Researchers Distribution parameter, C0 [-] Drift Velocity, 〈〈vgj〉〉 [m/s] Flow Applicable Range [m/s]

Chexal et al. (1991) C0 = [1 + α 0.05 (1 − α )2] C0V vgj = V gj0 Cg Not specified
C0V = L/[K 0 + (1 − K 0) αr ] Cg = (1 − α ) B1
L = min[1.25α 0.6, 1.0] 0.25
K 0 = B1 + (1 − B1)(ρg / ρf )0.25 ⎡ ⎛⎜ρf − ρg ⎞⎟ σggc ⎤
⎢ ⎥
V gj0 = 1.41 ⎢ ⎝ ⎠
⎥ C2 C3 C4
r = (1.0 + 1.57(ρg / ρf ))/(1 − B1) ρ2f
⎢ ⎥
B1 = min(0.8, A1 ) ⎣ ⎦
C2 = 1ifC5 ≥ 1
A1 = 1/[1 + e (−Re/60000) ]
= 1/{1 − exp[−C5/(1 − C5)]} ifC5 < 1
Re = Reg if Reg > Ref
= Ref if Reg < Ref ⎡ ⎡−Re /60000⎤ ⎤
⎢ f ⎥⎥
C3 = max ⎢0.5, 2e⎣ ⎦
⎢ ⎥
ρg jg DH ρf jf DH
⎣ ⎦
Reg = , Ref = C4 = 1ifC7 ≥ 1
μg μf
= 1/[1 − exp(−C8)] ifC7 < 1
C5 = 150/(ρf / ρg )
C7 = (0.09144/ DH )0.6
C7
C8 =
1 − C7
Franca and Lahey (1992) ≅1.0 (Wavy-stratified and Annular flow) 0.2 for jf = 0.005 m/s (Wavy stratified) 0.127 ≤ 〈jg〉 ≤ 4.01
≅1.0 (Plug flow) 2.7 for jf = 0.27 m/s (Annular flow) 0.198 ≤ 〈jf〉 ≤ 1.49
1.2 (Slug flow)
0.16 (Plug flow)
−0.20 (Slug flow)
Lamari (2001) 1.04 (Stratified wavy flow) 0.466 (Stratified wavy flow) 0.0950 ≤ 〈jg〉 ≤ 23.3
0.98 (Plug flow) 0.068 (Plug flow) 0.002 ≤ 〈jf〉 ≤ 1.08
1.06 (Slug flow) 0.991 (Slug flow)
0.92 (Annular flow) 3.688 (Annular flow)
Silva et al. (2011) 1.18 (Slug flow) 0.34 (Slug flow) 0.303 ≤ 〈jg〉 ≤ 2.06
0.280 ≤ 〈jf〉 ≤ 0.748
Kong and Kim (2017) 1.05 (bubbly flow) −1.21(bubbly flow) 3.6 ≤ 〈jg〉 ≤ 6.00
0.04 ≤ 〈jf〉 ≤ 1.00

correlation applicable for a whole range of void fraction. annular, and plug flow regimes, respectively. The superficial gas and
Chexal et al. (1991) formulated the drift-flux parameters by means of liquid velocities in their experiments were in the ranges of
many cascading constitutive relationships with numerous empirical 0.0950–23.3 m/s and 0.002–1.08 m/s, respectively.
parameters. The drift-flux parameters were given as a function of void Silva et al. (2011) performed air-water slug flow experiments in a
fraction, fluid properties and channel dimension. Chexel et al. (1991) horizontal pipe with the inner diameter of 0.026 m using a wire-mesh
sacrificed a physics principle characterizing the drift-flux parameters in sensor. They utilized the drift-flux model to correlate their experimental
exchange for developing the flow-regime-independent drift-flux corre- data. The distribution parameter of 1.13 and drift velocity of 0.23 m/s
lation. The drift-flux correlation proposed by Chexal et al. (1991) may were obtained experimentally based on their drift-flux plot. The su-
be considered a kind of regression curve to reproduce the void fraction perficial gas and liquid velocities in their experiments were in the
data. ranges of 0.303–2.06 m/s and 0.280–0.748 m/s, respectively.
Franca and Lahey (1992) developed a drift-flux correlation for Kong and Kim (2017) conducted air-water two-phase flow experi-
horizontal two-phase flows based on 200 air-water flow data taken in a ments in a horizontal pipe with the inner diameter of 0.0381 m using a
horizontal pipe with the inner diameter of 0.019 m. The data were local multi-sensor conductivity probe. Observed flow patterns in their
measured by a quick-closing valve method in four flow regimes such as experiments were bubbly, plug, slug, stratified, and annular flow re-
plug, slug, wavy- stratified and annular flow regimes. The range of gimes. The distribution parameter and drift velocity only for bubbly
superficial gas and liquid velocities set in their experiments were flow regime were determined to be 1.05 and −1.21 m/s, respectively.
0.127–4.01 m/s and 0.198–1.49 m/s, respectively. The distribution The superficial gas and liquid velocities in their experiments were in the
parameter and drift velocity were determined by a regression analysis ranges of 3.6–6.00 m/s and 0.04–1.00 m/s, respectively.
for the drift-flux plot in each flow regime. The distribution parameter The above brief review of the flow-regime-dependent drift-flux
obtained by the regression analysis was approximately unity for plug parameters indicates that the distribution parameter is approximately
and wavy- stratified and annular flow regimes, whereas the distribution unity except for slug and plug flow regimes and the drift velocity except
parameter was 1.2 for slug flow regime. The drift velocities in plug and for annular flow regime is fairly small in comparison with highest
slug flow regimes obtained by the regression analysis were 0.16 m/s mixture volumetric flux in each test condition.
and −0.20 m/s, respectively, while the drift velocities in wavy-strati-
fied and annular flow regimes ranged between the 0.2 m/s and 2.7 m/s 3. Existing database for horizontal flow
for the conditions of superficial liquid velocity between 0.005 m/s and
0.27 m/s. As discussed in the Section 2.1, a “compensation error” is Fig. 1 shows a schematic diagram of typical gas-liquid two-phase
common in determining the drift velocity, and thus the values of the flow regimes observed in a horizontal channel (Taitel et al., 1978). The
drift velocity may include a significant uncertainty. horizontal gas-liquid two-phase flow regimes are classified into five
Lamari (2001) developed a drift-flux correlation for two-phase flows basic patterns such as stratified smooth flow (SS), stratified wavy flow
in a horizontal pipe with the inner diameter of 0.0254 m. The mea- (SW), intermittent flow (I), annular flow with dispersed liquid droplets
surement was performed using an optical probe. The distribution (AD), and dispersed bubble (DB) flow. It should be noted that plug or
parameters of 1.04, 1.06, 0.92, and 0.98 and drift velocities of 0.466, elongated bubble and slug flow regimes can be classified as the inter-
0.991, 3.688, and 0.068 m/s were obtained for stratified-wavy, slug, mittent flow regime (Taitel et al., 1978). In this study, 566 data are

35
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

Fig. 1. Schematic diagram of flow regimes for gas-


liquid two-phase flow in a horizontal channel
(Taitel et al., 1978).

acquired from 12 data sources developed for gas-liquid two-phase flows fraction over a flow channel. The measurements at same flow condi-
in horizontal channels. Table 2 summarizes the test conditions and tions should be performed at many local points to characterize the void
measurement methods in the 12 data sources. The fluid systems in the fraction distribution in the flow channel. Parallel plate capacitance,
databases include air-water and air-kerosene systems, and the pipe gamma ray absorption and wire-mesh sensor methods were used in 3
diameter ranges from 19 mm to 77.9 mm. The superficial gas and liquid experiments.
velocities are in the range of 0.0253– 47.5 m/s and 0.000057 m/s–
5.97 m/s, respectively. Fig. 2 shows the flow conditions of the databases
4. Results and discussion
in a superficial gas velocity vs. superficial liquid velocity plane with
flow regime transition boundaries proposed by Taitel and
4.1. Comparison of existing drift-flux correlations with collected data
Dukler (1976). The figure indicates that the collected data are well
distributed in all five gas-liquid two-phase flow regimes.
In this section, the existing flow-regime-dependent correlations re-
As indicated in Table 2, several measurement techniques were uti-
viewed in the Section 2.2 are compared with the data collected in the
lized in the experiments and measurement accuracy information was
Section 3.
not available in 9 databases. A quick-closing valve technique was used
Fig. 3(a) shows a drift-flux plot in SS flow regime. No correlation
in 6 experiments to measure time- and volume-averaged void fraction.
has been developed for the SS flow regime. Since the flow character-
Instantaneous volume-averaged void fraction is obtained by the quick-
istics of the SS flow regime may be close to those of SW flow regime, the
closing valve technique. Experiments at same flow conditions should be
correlation of Lamari (2001) developed for the SW flow regime is
repeated 30 times or more to obtain accurate time- and volume-aver-
tentatively compared with the experimental data taken in the SS flow
aged void fraction. An appropriate number of repeated tests at same
regime. The data in the SS flow regime are directly aligned with some
flow conditions is particularly important for intermittent flow, because
scatter in the drift-flux plot. The correlation of Lamari (2001) indicated
the flow characteristics are irregular relative to other flow regimes.
by a red broken line tends to overestimate the gas velocity.
Conductivity, resistivity, and optical probe techniques were used in 3
Fig. 3(b) shows a drift-flux plot in SW flow regime. A broken red line
experiments to measure time-averaged local void fraction. Time- and
indicates gas velocities for the SW flow regime calculated by the cor-
area-averaged void fraction is obtained by integrating local void
relation of Lamari (2001). The data in the SW flow regime are directly

Table 2
Collected database of void fraction taken in horizontal gas-liquid two-phase flow.

Researchers Fluids Sys. <jg > [m/s] <jf > [m/s] Pipe Dia. [mm] Type of α Instrument Accuracy Information No. of Data

Beggs (1972) Air-water 0.0400–23.0 0.000400–1.83 25.4/38.1 Quick-closing valves NA 58


Spedding and Nguyen (1976) Air-water 0.0940–47.5 0.224–1.04 45.5 Quick-closing valve NA 91
Mukherjee (1979) Air-kerosene 0.0253–23.1 0.000305–3.59 38.1 Parallel plate capacitance Low at α < 10% or > 10% 74
Simpson et al. (1981) Air-water 2.967–11.5 3.56–4.69 26 Gamma ray technique NA 13
Minami and Brill (1987) Air-water 0.475–14.9 0.00518–0.901 77.9 Quick-closing valves NA 54
Minami and Brill (1987) Air- kerosene 0.542–16.6 0.00579–0.950 77.9 Quick-closing valves NA 57
Franca and Lahey (1992) Air-water 0.127–23.76 0.00560–1.49 19 Quick-closing valves ± 5% 99
Abdul-Majeed (1996) Air-kerosene 0.125–46.9 0.000057–0.793 50.8 Quick-closing valves NA 86
Lamari (2001) Air-water 0.0950–23.3 0.00200–1.08 25.4 Optical probes ± 10%. 61
Silva et al. (2011) Air-water 0.304–2.066 0.265–0.748 26 Wire-Mesh Sensor NA 12
Kong and Kim (2017) Air-water 0.0877–1.13 3.97–5.97 38.1 Four-sensor conductivity probe NA 10
Kong et al. (2017) Air-water 0.190–3.00 2.00–4.00 38.1 Four-sensor conductivity probe NA 9

36
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

the distribution parameter with the approximation of 〈〈vgj〉〉 = 0 m/s


are 1.17, 1.05, 1.20, 1.09 and 0.820 for the SS, SW, I, AD, and DB flow
regimes, respectively. Solid lines in Fig. 3 indicate the gas velocity
calculated by the newly determined distribution parameter and the
approximation of 〈〈vgj〉〉 = 0 m/s. The solid lines agree with the data.

4.2. Development of new drift-flux correlation for horizontal flow

The use of a flow-regime-dependent drift-flux correlation requires


accurate flow regime transition criteria to identify the flow regime
corresponding to superficial gas and liquid velocities. However, due to
complex horizontal two-phase flow characteristics, it may be difficult to
determine the flow regime with sufficient accuracy. In view of this,
Chexal et al. (1991) developed a flow-regime-independent drift-flux
correlation applicable for a whole range of void fraction.
Chexal et al. (1991) adopted an approach to formulate the drift-flux
parameters by means of many cascading constitutive relationships with
numerous empirical parameters, and sacrificed a physics principle
characterizing the drift-flux parameters in exchange for developing the
flow-regime-independent drift-flux correlation. The drift-flux correla-
tion proposed by Chexal et al. (1991) may be considered a kind of re-
gression curve to reproduce the void fraction data. In what follows, a
simple flow-regime-independent drift-flux correlation for gas-liquid
two-phase flow in a horizontal channel will be developed. The simple
Fig. 2. Flow conditions of collected data in superficial gas and liquid velocity plane. correlation is expected to provide some insight of the dependence of the
drift-flux correlation on flow parameters. It is also expected that the
aligned in the drift-flux plot. The correlation of Lamari (2001) agrees correlation can be applicable for all void fraction range from zero to
with the data taken in the SW flow regime. unity.
Fig. 3(c) represents a drift-flux plot in I flow regime. Blue solid and
broken lines are, respectively, gas velocities for plug and slug flow re- 4.2.1. Drift velocity in horizontal channel
gimes calculated by the correlation of Franca and Lahey (1992), As discussed in the Section 4.1, local drift velocity or relative ve-
whereas red solid and broken lines are, respectively, gas velocities for locity in the horizontal direction should be zero due to the absence of
plug and slug flow regimes calculated by the correlation of gravitational acceleration in the horizontal direction. No local drift
Lamari (2001). The data in the I flow regime are directly aligned with velocity yields 〈〈vgj〉〉 = 0 m/s. The validity of this approximation has
large scatter in the drift-flux plot. The correlations developed for the been experimentally verified in Fig. 3. In this study, 〈〈vgj〉〉 = 0 m/s is
plug flow regime tend to underestimate the gas velocity. The correla- assumed.
tions developed for the slug flow regime agree with the data at the
mixture volumetric flux higher than 5 m/s. No clear linear relationship
4.2.2. Distribution parameter in horizontal channel
between gas velocity and mixture volumetric flux is identified at the
Ishii (1977) considered the following functional form to take into
mixture volumetric flux lower than 5 m/s.
account the effect of inertia force on the distribution parameter.
Fig. 3(d) represents a drift-flux plot in AD flow regime. A broken red
line indicates gas velocities for the AD flow regime calculated by the ρg
C0 = C∞ − (C∞ − 1)
correlation of Lamari (2001). The data in the AD flow regime are di- ρf (6)
rectly aligned in the drift-flux plot. The correlation of Lamari (2001)
tends to overestimate or underestimate the gas velocity at low or high where C∞, ρg, and ρf are the asymptotic value of the distribution
mixture volumetric flux conditions, respectively. parameter, gas density and liquid density, respectively. The functional
Fig. 3(e) shows a drift-flux plot in DB flow regime. A broken red line form represented by Eq. (6) ensures that the distribution parameter
indicates gas velocity for the DB flow regime calculated by the corre- approaches unity as the density ratio approaches unity.
lation of Kong and Kim (2017). The data in the DB flow regime are The dependence of the distribution parameter on flow parameters is
directly aligned in the drift-flux plot. The correlation of Kong and examined as follows. Hibiki and Ishii (2003) performed an analysis of
Kim (2017) agrees with the data but predicts negative gas velocity at the flow parameter dependence of the distribution parameter for gas-
the mixture volumetric flux lower than 1 m/s, which is not physically liquid two-phase flow in a vertical large diameter pipe using the ratio of
correct.
non-dimensional superficial gas velocity, jg+ , to non-dimensional
As discussed above, a linear relationship between gas velocity and
mixture volumetric flux is found for most of flow regimes. The values of mixture volumetric flux, j+ , successfully. The non-dimensional su-
the distribution parameter and drift velocity in the existing drift-flux perficial gas velocity and mixture volumetric flux are defined as re-
correlations were determined by regression analyses. Since the second presented by Eqs. (7) and (8), respectively.
term in the right hand side of Eq. (2), 〈〈vgj〉〉, is much smaller than the
first term in the right hand side of Eq. (2), C0〈j〉, a significant un- jg
jg+ =
certainty is expected for the value of the drift velocity, 〈〈vgj〉〉. Local Δρgσ
1/4


⎞ ⎟
drift velocity or relative velocity in the horizontal direction should be ρf2 (7)
⎝ ⎠
zero due to the absence of gravitational acceleration in the horizontal
direction. No local drift velocity yields 〈〈vgj〉〉 = 0 m/s. The values of j
j+ =
the distribution parameter for each flow regime can be re-determined 1/4
⎛ Δρgσ ⎞
with the approximation of 〈〈vgj〉〉 = 0 m/s. The re-determined values of ⎜
ρf2

⎝ ⎠ (8)

37
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

Fig. 3. Comparison of existing flow-regime-


dependent correlations with collected data.

where Δρ, g and σ are the density difference between two phases, vg

ρg
j ρf
gravitational acceleration and surface tension, respectively. The de- C∞ = .
ρg
pendence of the distribution parameter on jg+ / j+ will be examined 1−
ρf (9)
first.
The asymptotic value of the distribution parameter for each data Fig. 4 shows the dependence of the asymptotic value of the dis-
point can be calculated by Eqs. (2) and (6) with the assumption of
〈〈vgj〉〉 = 0 m/s as: tribution parameter on the velocity ratio, jg+ / j+ . Open circle,

38
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

It should be noted here that the data taken in the I flow regime at
the mixture volumetric flux lower than 3 m/s are not used in developing
Eqs. (10) and (11). This is because it is difficult to determine the
asymptotic value of the distribution parameter by Eq. (9) accurately at
the mixture volumetric flux lower than 3 m/s. For the flow conditions of
the mixture volumetric flux lower than 3 m/s, data are significantly
scattered due probably to insufficient measurement accuracy as dis-
cussed in the Section 3. Eqs. (10) and (11) can reproduce the depen-
dence of the asymptotic value of the distribution parameter on the

velocity ratio, jg+ / j+ , for SS, SW, AD and DB flow regimes but they
tend to underestimate the asymptotic value of the distribution para-

meter for I flow regime at jg+ / j+ between 0.75 and 0.90. The final
form of the distribution parameter in one-dimensional drift-flux corre-
lation for gas-liquid two-phase flow in a horizontal channel is given by:
⎛ ⎞
For 0 ≤ ⎜ jg+ / j+ ⎟ < 0.9
⎝ ⎠ 1.50
⎧ ⎛ ⎞ ⎫
⎪ jg+ / j+ ⎪
⎪ ⎜ ⎟ ⎪
C0 = 0.800 exp 0.815 ⎜ ⎟
⎨ 0.900 ⎬
⎪ ⎜ ⎟ ⎪
⎪ ⎜ ⎟ ⎪
⎩ ⎝ ⎠ ⎭
⎛ ⎞ 1.50
Fig. 4. Relation between C∞ and ⎜ jg+ / j+ ⎟ for horizontal gas-liquid two-phase flow. ⎡ ⎤
⎧ ⎛ + ⎞ ⎫
⎝ ⎠ ⎢ ⎪ j / j+ ⎪ ⎥
⎢ ⎪ ⎜ g ⎟ ⎪ ⎥ ρg
− ⎢0.800 exp 0.815 ⎜ ⎟ − 1⎥
⎨ 0.900 ⎬ ρ
triangular, square, pentagon and hexagon indicate data for SS, SW, I, ⎢ ⎪ ⎜ ⎟ ⎪ ⎥ f
⎢ ⎪ ⎜ ⎟ ⎪ ⎥
AD and DB flow regimes, respectively. This figure indicates two distinct ⎣ ⎩ ⎝ ⎠ ⎭ ⎦ (12)
trends which were also found in gas-liquid two-phase flow in a large
diameter pipe (Hibiki and Ishii, 2003). The asymptotic value of the ⎛ ⎞
For 0.9 ≤ ⎜ jg+ / j+ ⎟ ≤ 1
distribution parameter gradually increases as the velocity ratio, ⎝ ⎠
jg+ / j+ , increases. The asymptotic value of the distribution para- ⎛ ⎞ ⎛ ⎞ ρg
C0 = ⎜−8.08 jg+ / j+ + 9.08⎟ − 8.08 ⎜ jg+ / j+ + 1⎟ ,
ρ (13)
⎝ ⎠ ⎝ ⎠ f
meter takes its maximum value and approaches unity as the velocity
The final form of the drift velocity in one-dimensional drift-flux
ratio, jg+ / j+ , approaches unity. The trend of C∞ → 1 as correlation for gas-liquid two-phase flow in a horizontal channel is
given by:
jg+ / j+ → 1 is physically sound because the asymptotic value of the
vgj = 0m/s, (14)
distribution parameter should be unity at the limiting condition of

jg+ / j+ = 1 or gas single-phase flow. Due to the data scatter, it is 4.3. Assessment of predictive capability of newly developed drift-flux
correlation for horizontal two-phase flow
difficult to set the value of jg+ / j+ at the maximum C∞ precisely. The
In this section, the predictive capability of void fraction using the
newly developed drift-flux correlation (hereafter, Rassame–Hibiki's
value of jg+ / j+ is tentatively set at 0.9, which is the same value for correlation) will be assessed by introducing four statistical parameters.
gas-liquid two-phase flow in a vertical large diameter pipe (Hibiki and They are mean absolute error (or bias), md, standard deviation (or
Ishii, 2003). The following correlation is proposed to reproduce the random error), sd, mean relative deviation, mrel, and mean absolute
relationship between the asymptotic value of the distribution parameter relative deviation, mrel, ab (Hibiki et al., 2017).
N
and the velocity ratio, jg+ / j+ . 1
md =
N
∑( α (i)cal . − α (i)exp . ),
i=1 (15)
⎛ ⎞
For 0 ≤ ⎜ jg+ / j+ ⎟ < 0.9 N
⎝ ⎠ 1
1.50 sd =
N−1
∑ {( α (i)cal . − α (i)exp . ) − md}2 ,
⎧ ⎛ ⎞ ⎫ i=1 (16)
⎪ jg+ / j+ ⎪
⎪ ⎜ ⎟ ⎪
C∞ = 0.800 exp 0.815 ⎜ N
⎟ 1 ( α (i)cal . − α (i)exp . )
⎨ 0.900 ⎬ mrel = ∑ ,
⎪ ⎜ ⎟ ⎪ N α (i)exp .
⎪ ⎜ ⎟ ⎪ i=1 (17)
⎩ ⎝ ⎠ ⎭ (10)
N
1 α (i)cal . − α (i)exp .
⎛ ⎞ mrel, ab = ∑ ,
For 0.9 ≤ ⎜ jg+ / j+ ⎟ ≤ 1 N i=1
α (i)exp . (18)
⎝ ⎠
where N, αexp. and αcal. are the number of samples, measured void
C∞ = −8.08 jg+ / j+ + 9.08,
fraction and calculated void fraction, respectively.
(11)
Fig. 5(a)– (f) show the ratio of calculated void fraction to measured

39
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

Fig. 5. Predictive capability of Rassame–Hibiki's


correlation and the correlation of
Chexal et al. (1991) for each flow regime.

void fraction against measured void fraction for SS, SW, I (all data), I deviation between calculated and measured void fractions is identified.
(data measured at 〈j〉 ≥ 3 m/s), AD and DB flow regimes, respectively. Rassame–Hibiki's correlation and the correlation of Chexal et al. (1991)
Fig. 6(a) and (b) show the ratio of calculated void fraction to predict void fraction in the SS flow regime with +0.0518 (bias) ±
measured void fraction against measured void fraction for total data 0.123 (random error) and −0.0141 (bias) ± 0.0837 (random error),
and total data except I flow regime data measured at 〈j〉 < 3 m/s, re- respectively. As indicated in Fig. 5(b), both correlations show similar
spectively. Open and solid circles in Figs. 5 and 6 indicate the ratio predictive capability of void fraction in SW flow regime and no sys-
calculated by Rassame–Hibiki's correlation and the correlation of tematic deviation between calculated and measured void fractions is
Chexal et al. (1991). Tables 3 and 4 summarize the predictive identified. Rassame-Hibiki's correlation and the correlation of
capability of Rassame–Hibiki's correlation and the correlation of Chexal et al. (1991) predict void fraction in the SW flow regime with
Chexal et al. (1991), respectively. +0.0419 (bias) ± 0.0629 (random error) and 0.0394 (bias) ±
As shown in Fig. 5(a), both correlations show similar predictive 0.0494 (random error), respectively. Based on the above brief discus-
capability of void fraction in SS flow regime and no systematic sion, both correlations can predict the void fraction for stratified

40
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

Fig. 6. Predictive capability of Rassame–Hibiki's


correlation and the correlation of
Chexal et al. (1991) for all data.

Table 3 predict void fraction in the AD flow regime with +0.00262 (bias) ±
Predictive capability of void fraction by Rassame–Hibiki's correlation. 0.0458 (random error) and 0.0392 (bias) ± 0.0495 (random error),
respectively. Based on the above brief discussion, both correlations can
Statistical parameters md [-] sd [-] mrel [-] mrel,ab [-]
predict the void fraction for separated horizontal two-phase flow re-
SS 0.0518 0.123 0.0674 0.138 gime or AD flow regime.
SW 0.0419 0.0629 0.0476 0.0534 As indicated in Fig. 5(f), both correlations show similar predictive
I −0.0321 0.116 0.125 0.378
capability of void fraction in DB flow regime and no systematic de-
I for data measured at <j> > −0.104 0.134 −0.154 0.166
3 m/s only
viation between calculated and measured void fractions is identified. It
AD 0.00262 0.0458 0.00194 0.0389 may be noted that the correlation of Chexal et al. (1991) may over-
DB −0.00271 0.00832 −0.0155 0.00739 estimate the void fraction systematically as void fraction decreases.
Total data except for data in I 0.00770 0.0947 0.00383 0.0929 Rassame-Hibiki's correlation and the correlation of Chexal et al. (1991)
flow regime measured at
predict void fraction in the DB flow regime with −0.00271 (bias) ±
<j> <3 m/s
Total data 0.00487 0.0985 0.0758 0.206 0.00832 (random error) and 0.0364 (bias) ± 0.0454 (random error),
respectively. Based on the above brief discussion, both correlations can
predict the void fraction for dispersed horizontal two-phase flow regime
Table 4 or DB flow regime.
Predictive capability of void fraction by the correlation of Chexal et al. (1991). The above assessment of Rassame–Hibiki's correlation and the cor-
relation of Chexal et al. (1991) has verified their predictive capability of
Statistical parameters md [-] sd [-] mrel [-] mrel,ab [-]
void fraction in horizontal gas-liquid two-phase flows with about ± 20
SS −0.0141 0.0837 −0.0203 0.0868 %. Although the predictive capability of void fraction by Rassame–Hi-
SW 0.0394 0.0494 0.0445 0.0453 biki's correlation is similar to that by the correlation of
I −0.0227 0.0858 0.0433 0.272
Chexal et al. (1991), Rassame–Hibiki's correlation is much simpler than
I for data measured at <j> > −0.0167 0.0811 −0.0398 0.125
3 m/s only the correlation of Chexal et al. (1991). This fact implies that Rassa-
AD 0.0392 0.0495 0.0452 0.0462 me–Hibiki's correlation may capture the dominant parameter affecting
DB 0.0364 0.0454 0.101 0.101 the distribution parameter successfully.
Total data except for data in I flow 0.0156 0.0673 0.0155 0.0799
regime measured at <j>
<3 m/s
5. Conclusions
Total data 0.00329 0.0740 0.0344 0.156
In this study, the newly developed drift-flux correlation (namely,
Rassame–Hibiki's correlation) has been developed for predicting void
horizontal two-phase flow regime including SS and SW flow regimes. fraction in horizontal gas-liquid two-phase flow. The summary of this
As shown in Fig. 5(c), both correlations fail to predict void fraction. study is given below.
Fig. 5(d) shows the predictive capability of the correlations based on
the data at 〈j〉 ≥ 3 m/s. Comparison between Fig. 5(c) and (d) con-
cludes that the discrepancy between the calculated and measured void
• An extensive literature survey has been conducted from 12 data
sources to establish comprehensive databases used for the assess-
fraction occurs at 〈j〉 ≤ 3 m/s. One of possible reasons for the dis- ment of predictive capability of existing drift-flux correlations and
crepancy between the calculated and measured void fraction observed for the development of a new drift-flux correlation. A total of 566
at 〈j〉 ≤ 3 m/s may be due largely to insufficient measurement accu- data points has been collected in a wide range of test conditions such
racy of the data taken at 〈j〉 ≤ 3 m/s. Rassame–Hibiki's correlation and as superficial gas velocity ranging from 0.0253 to 47.5 m/s, super-
the correlation of Chexal et al. (1991) predict void fraction in the I flow ficial liquid velocity ranging from 0.000057 m/s to 5.97 m/s, inner
regime with −0.0321 (bias) ± 0.116 (random error) and −0.0227 pipe diameter ranging from 0.019 m to 0.0779 m and void fraction
(bias) ± 0.0858 (random error), respectively. ranging from ∼0 to ∼1. Horizontal two-phase flow regimes covered
As indicated in Fig. 5(e), both correlations show similar predictive by the collected database are the stratified smooth flow (SS), stra-
capability of void fraction in AD flow regime and no systematic de- tified wavy (SW), intermittent (I), annular flow with dispersed li-
viation between calculated and measured void fractions is identified. quid droplets (AD) and dispersed bubbly (DB).
Rassame–Hibiki's correlation and the correlation of Chexal et al. (1991) • Existing drift-flux correlations developed for horizontal gas-liquid

41
S. Rassame, T. Hibiki International Journal of Heat and Fluid Flow 69 (2018) 33–42

two-phase flow have been reviewed. The existing drift-flux corre- 1–14.
lations can be classified into two types, namely, flow-regime-de- Franca, H., Lahey, R.T., 1992. The use of drift flux techniques for the analysis of phase
flows. Int. J. Multiphase Flow. 18 787-, 801.
pendent and flow-regime-independent drift-flux correlations. For Hibiki, T., Ishii, M., 2002. Distribution parameter and drift velocity of drift-flux model in
the flow-regime-dependent drift-flux correlation, constant values of bubbly flow. Int. J. Heat and Mass Trans. 45, 707–721.
the distribution parameters and drift velocities are given for each Hibiki, T., Ishii, M., 2003. One-dimensional drift-flux model for two-phase flow in a large
diameter pipe. Int. J. Heat and Mass Trans. 46, 1773–1790.
flow regime. It has been identified that the existing flow-regime- Hibiki, T., Situ, R., Mi, Y., Ishii, M., 2003. Modeling of bubble-layer thickness for for-
dependent drift-flux correlations do not incorporate the effect of mulation of one-dimensional interfacial area transport equation in subcooled boiling
flow conditions into the correlation. It has been pointed out that the two-phase flow. Int. J. Heat Mass Transfer 46, 1409–1423.
Hibiki, T., Mao, K., Ozaki, T., 2017. Development of void fraction-quality correlation for
value of the drift velocity in the flow-regime-dependent drift-flux two-phase flow in horizontal and vertical tube bundles. Prog. Nuc. Eng. 97, 38–52.
correlation may not be accurate due to a “compensation error” be- Information Systems Laboratories, Inc., 2001. RELAP5/MOD3.3 Beta Code Manual
tween the distribution parameter and drift velocity. Volume IV, Models and Correlations. NUREG/CR-5535/Rev 1-Vol IV.

• Assuming the void-fraction-weighted mean drift velocity being zero, Ishii, M., 1977. One-Dimensional Drift-Flux Model and Constitutive Equations For
Relative Motion Between Phases in Various Two-Phase Flow Regimes, ANL-77- 47.
a correlation for the distribution parameter has been developed as a Argonne National Lab., IL, USA.
function of the ratio of non-dimensional superficial gas velocity to Ishii, M., Hibiki, T., 2010. Thermo-Fluid Dynamics of Two-Phase Flow. Springer, NY.
non-dimensional mixture volumetric flux. The newly developed Kong, R., Kim, S., Bajorek, S., Tien, K., Hoxie, C., 2017. Experimental investigation of
horizontal air–water bubbly-to-plug and bubbly-to-slug transition flows in a 3.81 cm
drift-flux correlation (namely Rassame-Hibiki's correlation) is much ID pipe. Int. J. Multiphase Flow. 94 137-, 155.
simpler than the flow-regime-dependent drift-flux correlation such Kong, R., Kim, S., 2017. Characterization of horizontal air–water two-phase flow. Nucl.
as Chexel et al. (1991). Nevertheless, Rassame–Hibiki's correlation Eng. Des. 312, 266–276.
Lamari, M.L., 2001. An Experimental Investigation of Two-Phase (Air-Water) Flow
shows an excellent predictive capability of void fraction, which is Regimes in a Horizontal Tube At Near Atmospheric Conditions. Carleton University.
similar to that of the correlation of Chexel et al. (1991). The mean Ottawa, Ontari, Canada.
absolute error (or bias), standard deviation (random error), mean Mukherjee, H., 1979. An Experimental Study of Inclined Two-Phase Flow, The University
of Tulsa. Tulsa, USA.
relative deviation, and mean absolute relative deviation of Rassa- Minani, K., Brill, J.P., 1987. Liquid holdup in wet-gas pipelines. SPE Prod. Eng. 2, 36–44.
me–Hibiki's correlation are 0.0487, 0.0985, 0.0758 and 0.206, re- Mishima, K., Hibiki, T., 1996. Some characteristics of air-water two-phase flow in small
spectively. Rassame–Hibiki's correlation can predict void fraction diameter vertical tubes. Int. J. Multiphase Flow. 22, 703–712.
Ozaki, T., Hibiki, T., 2015. Drift-flux model for rod bundle geometry. Prog. Nucl. Eng. 83,
for horizontal gas-liquid two-phase flow with about ± 20% un- 229–247.
certainty. Silva, M.J., Hampel, U., Arruda, L.V.R., Amaral, C.E.F., Marales, R.E.M., 2011.
Experimental investigation of horizontal gas-liquid slug flow by means of wire-mesh
sensor. J. Braz. Soc. Mech. Sci. Eng. XXXII 237–242.
References
Simpson, H.C., Rooney, D.H., Grattan, E., Al-samarrae, F.A.A., 1981. Two-phase flow
structure in large diameter horizontal tubes. NEL Report No. 677.
Abdul-Majeed, G.H., 1996. Liquid holdup in horizontal two-phase gas-liquid flow, J. Spedding, L., Nguyen, V.T., 1976. Data on Holdup, Pressure Loss and Flow Patterns for
Petro. Sci. Eng. 15, 271–280. Two-Phase Air Water Flow in an Inclined Pipe. University of Auckland, Auckland
Beggs, H.D., 1972. An experimental study of two-phase flow in inclined pipes. New Zealand Report Eng. 122.
Dissertation, The University of Tulsa, Tulsa, USA. Taitel, Y., Dukler, E., 1976. A model for predicting flow regime transitions in horizontal
Brooks, C.S., Ozar, B., Hibiki, T., Ishii, M., 2012. Two-group drift-flux model in boiling and near horizontal gas-liquid flow. AIChe J. 22, 47–55.
flow. Int. J. Heat Mass Transfer. 55, 6121–6129. Taitel, Y., Lee, N., Dukler, E., 1978. Transient gas-liquid flow in horizontal pipes:
Chexal, B., Lellouche, G., Horowitz, J., Healzer, J., Oh, S., 1991. The Chexal-Lellouche Modeling the flow pattern transitions. AIChe J. 24, 920–934.
void fraction correlation for generalized applications. EPRI, NSAC-139. Zuber, N., Findlay, J.A, 1965. Average volumetric concentration in two-phase flow sys-
Clark, C., Griffiths, M., Chen, S.W., Hibiki, T., Ishii, M., Ozaki, T., Kinoshita, I., Yoshida, tems. J. Heat Transfer 87, 453–468.
Y., 2014. Drift-flux correlation for rod bundle geometries. Int. J. Heat Fluid Flow. 48,

42