Está en la página 1de 25

Applied Mathematical Modelling 22 (1998) 773±797

On the modelling of bubble plumes in a liquid pool


B.L. Smith *

Thermal-Hydraulics Laboratory, Paul Scherrer Institute, CH-5232 Villigen PSI, Switzerland


Received 7 February 1997; received in revised form 20 February 1998; accepted 3 March 1998

Abstract

Turbulent, bubble plumes are investigated numerically using the commercial, Computational Fluid Dynamics
(CFD) code CFX-F3D. A six-equation, two-¯uid model approach is adopted, in which interphase momentum ex-
change models include buoyancy, drag, added mass, lift and turbulent dispersion e€ects. Particular attention is paid to
turbulence modelling, in which generation and dissipation resulting from interaction between bubbles and liquid are
speci®cally taken into account within the context of an extended k )  turbulence model. Results from a number of
calculations are presented and compared against published, experimental bubble plume data. It is suggested that ex-
isting bubble/liquid interaction models for plumes may be grouped into three categories: those which produce lateral
bubble spreading, those which di€use the ambient liquid velocity ®eld, and those which couple the plume to the sur-
rounding liquid and thereby ultimately govern the pool mixing behaviour. Ó 1998 Elsevier Science Inc. All rights
reserved.

Keywords: Numerical simulation; Bubble plume; Plume spreading; Pool mixing; Two-phase ¯ow

1. Background

In the context of the Framework Programme IV of the European Atomic Energy Community,
and in collaboration with similar laboratories in the Netherlands and Spain, the Paul Scherrer
Institute (PSI) is participating in the Technology Enhancement of Passive Safety Systems
(TEPSS) project [1]. The project aims to make signi®cant additions to the technology base relating
to advanced, passive-type Boiling Water Reactors (BWRs), which o€er improved safety and
public acceptance through product design simplicity and improved safety margins [2].
As part of this activity, research is being undertaken to support development of improved
countermeasures against pool strati®cation. These e€orts, embodied in the sub-project Large-
Scale Investigation of Natural Circulation and Mixing (LINX), are directed towards a better
understanding of the most important physical processes taking place in deep-pool, thermal and
two-phase plumes, and the testing of concepts for avoiding or minimizing hot, strati®ed pool
regions. Small-scale experiments (LINX-1 and LINX-1.5), which addressed single-phase, thermal
plume mixing phenomena, have already been conducted [3,4]. The main e€ort of the current phase
of the programme centres on a medium-scale, highly instrumented test facility, LINX-2, which
will be used to study multi-phase aspects of natural circulation, mixing and condensation

*
Tel.: +41 056 2726; fax: +41 056 31021 99.

0307-904X/98/$ ± see front matter Ó 1998 Elsevier Science Inc. All rights reserved.
PII: S 0 3 0 7 - 9 0 4 X ( 9 8 ) 1 0 0 2 3 - 9
774 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

Fig. 1. Schematic of LINX-2 Facility.

phenomena in pressure suppression pools and containment volumes, under prototypic pressure
and temperature conditions, and in the presence of non-condensable gases [5].
The facility, Fig. 1, consists of a pressure vessel (rated at 10 bar and 250°C), steam and ni-
trogen (or air) supply lines, and a water conditioning loop. The vessel is well insulated, with heat
losses monitored using thermocouples both on and in the vessel wall, as well as within the in-
sulation material itself. Steam/nitrogen mixtures can be injected from either the top or bottom of
the vessel, as appropriate.
In parallel with the experimental programme, Computational Fluid Dynamics (CFD) models
are being assembled, aimed at developing reliable, multi-phase turbulence models for application
to mixing and natural circulation problems of the type which fall within the scope of the TEPSS
project. First analyses concern bubble-induced mixing in open pools. In lieu of relevant experi-
mental data from the LINX-2 test series, which, due to other commitments, are not expected for
the pool-mixing experiments until late 1998, the present investigation utilizes data from the open
literature to assess the performance of the numerical models.

2. Available experimental data

Bubble-induced mixing in open pools has been extensively researched because of its relevance to
a number of industrial processes, including bubble columns in chemical reactors, destrati®cation
B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 775

of lake water, and metallurgical processing. The majority of experiments feature air-bubble
plumes in water as simulant materials [6±13], although some important work on helium, argon
and nitrogen injection into water and liquid metal systems can also be cited [14±17]; see also
Mazumdar and Guthrie [18] for a recent review article. The data refer exclusively to the injection
of non-condensable gases into liquid pools. To the author's knowledge, no data are available
concerning the injection of a condensable/non-condensable mixture; the LINX-II tests aim to
provide ®rst-of-a-kind data of this type.
A typical experimental set-up is shown in Fig. 2(a) and consists of a cylindrical vessel, partially
®led with liquid, with bottom gas injection along the axis of symmetry, via a nozzle or porous
plug. The buoyant bubble plume so produced rises to the surface, spreading radially, and entrains
liquid from the pool, inducing a large-scale recirculation. Void fraction and bubble frequency
distributions are traditionally measured using electroresistivity probes, and recirculation velocities
in the pool with laser-Doppler anemometry. These two techniques have been successfully com-
bined to yield gas and liquid rise velocities, turbulence levels and Reynolds stresses in two-phase
plumes by Iguchi et al. [13] The principal experimental ®ndings are:
· With air/water systems, bubble spreading is approximately linear, but increased lateral plume
growth has been observed for helium/water and nitrogen/mercury systems [14].
· Radial distributions of void fraction, bubble frequency, and gas and liquid rise velocities in
the plume are essentially Gaussian, and exhibit good similarity along the length of the plume
[14]; this suggests that bubble dispersion is essentially a stochastic process. However, Cast-
illejos and Brimacombe [19] have reported that, for their large-ori®ce experiments, pro®les
are somewhat narrower than Gaussian. Perhaps, in their case, there remained some in¯uence
of the injection mechanism.
· The ¯ow in the plume is buoyancy-driven, except in the developing region near the injection
ori®ce(s), where the ¯ow is inertia-dominated, and the rise velocity of the bubbles with respect
to the liquid can be adequately represented by the terminal rise velocity of a single bubble (of
the same size) in a quiescent liquid [6,12].
· Turbulence is mostly isotropic, but with a bias in the vertical direction where the shearing
forces are stronger, with levels about twice those of single-phase plumes with the same liquid
rise velocity [12].
For gas injection through simple ori®ces, visual inspection of bubble plume dynamics has
enabled experimenters to identify three main ¯ow regimes, each governed by di€erent ¯ow
phenomena [18]. In the immediate vicinity of the ori®ce is the primary bubble zone, where bubble
detachment from the ori®ce(s) takes place. This region is inertia-dominated and the input gas
¯ow-rate and mode of injection have important in¯uences on bubble sizes, distribution and rise
velocities. In the free bubble zone above this, buoyancy forces predominate, there is little bubble
break-up and the overall plume development proceeds largely independently of the inlet condi-
tions. Above this comes the plume zone proper, where bubble break-up occurs, producing a
spectrum of bubble sizes. The primary reason for break-up has been observed to be the result of
the growth of instabilities on the surfaces of large bubbles, a process enhanced by the turbulent
and shearing motions in the liquid phase [18]. This general picture can change, of course, for
shallow-depth injection and/or high gas injection velocities where, for example, the inertia-
dominated region could prevail to the liquid surface.
Gas injection through porous plugs, at low ¯ow-rates, produces uniform, discrete bubble
swarms, with little evidence of bubble break-up or coalescence. At progressively higher injection
rates, incipient coalescence of the bubbles near the pore ori®ces takes place, bubble column
pulsation begins and, ®nally, bubble blanketing of the entire surface occurs. At this stage, large
bubble packets are produced, which break up almost immediately after detachment from the plug
776 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

Fig. 2. Simulation Model (2-D, Axisymmetric); (a) Schematic, (b) Mesh Layout.
B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 777

[9]. The stable free bubble zone seen with ori®ce injection appears not to exist with injection
through porous plugs.
Data from one of the porous-plug experiments of Anagbo and Brimacombe [9] are used in the
present work as a basis for the development of appropriate numerical models for bubble plumes.
For low gas-injection rate (Q_ ˆ 12 l/min), the bubbles were of uniform size (diameter 4 mm), were
suciently dispersed in the liquid to avoid complications associated with bubble coalescence and,
from the plug to the liquid surface, remained small enough to resist break-up forces. A well-
de®ned, dispersed, bubbly ¯ow regime therefore prevails throughout the plume in this experiment,
providing an attractively simple database for the development of two-phase numerical models.
Data are available in the form of void fraction and bubble velocity distributions at various heights
in the plume, and vertical liquid velocities in the surrounding pool.

3. Modelling

Historically, analytical studies of bubble plumes began with lumped-parameter models in


which the mixture conservation equations for mass and momentum were solved in integral form,
analogously to the treatment for single-phase jets [14]. Radial distributions of void fraction and
rise velocity were assumed (top-hat or Gaussian), and an entrainment coecient was imposed at
the plume boundary to provide a dynamic coupling to the surrounding pool. The model repre-
sents a cheap and ecient analysis tool, but obviously relies heavily on the availability of ap-
propriate empirical data.
More recently, CFD techniques have been employed to calculate directly the void and velocity
distributions from the underlying phasic conservation equations, together with suitable models
representing the phase interactions [12,20±26]. Two approaches are possible for computing the
bubble motions in dispersed ¯ow systems: Lagrangian and Eulerian. In the former, the ¯ow ®eld
for the continuous phase is calculated on an Eulerian grid for an assumed void fraction distri-
bution. From the ¯ow ®eld thus obtained, trajectories are computed in a Lagrangian manner for a
specimen number of bubbles, typically 1000. This provides a new estimate for the void fraction
distribution, which is then used to update the liquid ¯ow ®eld. In principle, the iteration can be
continued to self-consistency, though in practice this is seldom done. This technique has been used
recently by Sheng and Irons [25] in an analysis of their own experiments. They obtained good
correspondence with measured void fraction distributions, provided a bubble break-up model was
installed and the daughter bubbles were also tracked.
In the second approach, the bubbles and liquid are treated as two interpenetrating continua,
each satisfying its own conservation and constitutive equations (two-¯uid model), and coupled via
phase interaction relations. The computations are performed simultaneously for both phases, on
the same Eulerian grid. Two-¯uid models have been used extensively in nuclear reactor thermal
hydraulics over many years. In this context, applications are mostly one-dimensional and dom-
inated by heat transfer and phase change e€ects, for which reliable correlations are available. The
extension to multi-dimensional situations is much more recent and the development of suitable
models for the phase interaction constitutes a signi®cant research challenge.
The present study aims, through numerical simulation of one of the porous plug experiments of
Anagbo and Brimacombe [9], to make a critical assessment of the various models which have been
advanced to support multi-dimensional, CFD modelling of bubbly ¯ows. Transient, numerical
simulation of the experiment has been carried out to compute the plume growth, the generation of
large-scale motions in the pool, the accompanying deformations of the liquid surface, and the
¯ow ®eld in the gas space above it. All calculations have been carried out using the CFX-F3D
778 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

code [27], with appropriate modi®cations and code extensions implemented via the user Fortran
routines. The models considered are itemized in the following sections, under appropriate sub-
headings, including some comments regarding their origins and applicability. Results and com-
parisons with measured data follow in Section 4 and ®nal comments are given in Section 5.

3.1. Model set-up

The experimental set-up, shown schematically in Fig. 2(a), consists of a cylindrical perspex
tank, of diameter 50 cm, ®lled with de-ionized water to a depth of 40 cm. The porous plug, of
diameter 6 cm, is positioned at the centre of the base, ¯ush with the surface. For the high ¯ow-rate
experiments it was found necessary to install a ring ba‚e 3 mm above the initial pool surface to
prevent sloshing [9]. However, for the low ¯ow-rate test simulated here, the liquid level always
remained below the level of the ba‚e, which was therefore not modelled.
A 2-D, axisymmetric mesh was constructed (Fig. 2(b)) with 50 meshes in the radial direction ±
the porous plug covering the ®rst 10 next to the axis ± and 60 in the axial direction, of which the
®rst 50 were below the initial water level. The mesh sizes are in geometric progression with ratios
1.0232 and 1.035, respectively, in the radial and axial directions, giving sucient mesh concen-
tration near the axis to capture the void fraction and velocity pro®les in the plume region, par-
ticularly near the plug. The gas space above the liquid surface has been extended to eliminate the
in¯uence of the (pressure constant) boundary condition at the top of the model on the ¯ow near
the water surface.
Appropriate inlet conditions at the plug surface are dicult to estimate because of gas blan-
keting, either partial or total, and subsequent break-up to form the bubble swarm proper, details
of which cannot be followed with the current model. The experiments had been carried out for
three di€erent gas ¯ow-rates, Q_ ˆ 200, 600 and 1200 N cm2 /s 1 and, dividing by the total plug
area, correspond to average volumetric ¯uxes of jg ˆ 7.1, 21.2 and 42.4 cm3 /s cm2 , respectively.
However, the ¯uxes reported in the paper are 8.5, 25.3 and 50.5 cm3 /s cm2 and may have been
scaled with respect to the e€ective bubble discharge area of the plug. If so, the scaling factor
(e€ective/true area) is consistently in the range 0.83±0.84, for the three ¯ow rates. Best would be to
clear up the discrepancy with the authors of the paper but, sadly, the principal author has since
died, and attempts to contact the second author produced no response. Thus, in the absence of
further information, it is here assumed that the scaling factor of 0.84 is indeed a measure of the
e€ective to actual area of the plug in regard to the bubble generation. Consequently, the following
inlet conditions are adopted for the low ¯ow-rate case
rl ˆ 0:16 ul ˆ 0:0 m=s;
…1†
rg ˆ 0:84 ug ˆ 0:101 m=s;
in which rl , rg are the liquid and gas volume fractions, respectively, and u is the axial velocity
component; the total gas ¯ow rate of 200 N cm3 /s is then maintained. Given the uncertainty, it
was thought prudent to assess the sensitivity of the results to the assumed boundary condition (1);
results are discussed in Section 4.
The side wall and base are considered non-slip surfaces. Turbulent ¯ow conditions are assumed
for the liquid phase, and laminar ¯ow for the gas phase, the latter since Reynolds stresses in the
gas would be small [28]. Initially, the liquid surface is de®ned by assigning appropriate values for
the gas and liquid volume fractions: that is, rl . 0 and rg . 1 for the cells above the surface, and
the converse for those below, the mesh being adjusted appropriately to ensure that the initial

1
Flow-rate normalized to STP conditions.
B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 779

liquid level coincided with a mesh boundary. Subsequently, the volume fractions change as a
consequence of the phase motions. Thus, the level swelling commonly associated with bubble
plumes, and which has been shown to in¯uence the surface dynamics [29,30], can be simulated
with the present model, though the interface will not remain sharper than one cell width.

3.2. Numerical simulations

Equation set: For the two-¯uid model, separate mass and momentum equations are required
for each phase. In the present application, there is no mass or heat transfer between the phases,
but momentum transfer does take place because of interfacial drag. The full equation set used
here is as follows:
 
d o orl
…rl ql † ÿ j ql Dt j ˆ 0; …2†
dt ox ox
 
d o org
…rg qg † ÿ j qg Dt j ˆ 0; …3†
dt ox ox
dÿ  o   op
rl ql ujl ÿ j rl rijl ˆ ÿrl j ‡ Flj ; …4†
dt ox ox
d   o h i op
rg q0g ujg ÿ j rg rijg ˆ ÿrg j ‡ Fgj ; …5†
dt ox ox
 
d o rl lt okl
…rl kl † ÿ j ˆ Sk ; …6†
dt ox rk oxj
 
d o rl lt ol
…rl l † ÿ j ˆ S ; …7†
dt ox r oxj
in which
d o o d o o
ˆ ‡ ujl j or ˆ ‡ ujg j …8†
dt ot ox dt ot ox
as appropriate, and
 i 
ij
oug oujg
rg ˆ lg ‡ ; …9†
oxj oxi
 i 
ij oul oujl 2
rl ˆ …ll ‡ lt † ‡ ÿ ql kl dij : …10†
oxj oxi 3
Turbulent mass di€usion: The inclusion of the inter-phase mass di€usion terms in Eqs. (2) and
(3) is a matter of some controversy. The terms do not appear in the standard derivations of the
two-¯uid model [31,32], where inter-penetration of the phases occurs only as a result of the in-
terfacial forces and is calculated directly from the momentum equations [33]. However, Besnard
and Harlow [34], for example, argue that, at least for disperse ¯ow situations, Reynolds-averaging
of the (already averaged) two-¯uid equations can proceed provided the scale of the turbulent
¯uctuations is much larger than control volumes associated with the phasic averaging, but still
small compared with the scale of variation of the mean-¯ow ®eld. Providing this intermediate
scale exists, the void fraction can be decomposed into mean and ¯uctuating parts, and the
780 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

turbulent mass di€usion terms will appear naturally as a consequence of the Reynolds averaging.
The terms are retained here, for generality, with the phase di€usion coecient
Dt ˆ lt =rr ; …11†
where lt is the (liquid) eddy viscosity and rr is the turbulent Prandtl number for mass di€usion.
Virtual mass force: A gas bubble accelerating through a liquid experiences a resistance force
due to the liquid it displaces [35]. This induces an apparent increase in bubble inertia, here rep-
resented by an enhanced gas density
q0g ˆ qg ‡ Cvm ql …12†
in the acceleration term of the gas momentum Eq. (5). For a rigid sphere, the virtual mass co-
ecient Cvm ˆ 1/2, and this value is adopted for the present study (see also Sheng and Irons [25]
and Davidson [33]). Strictly, the expression (12) is only appropriate for a bubble rising in qui-
escent surroundings, whereas in a bubble plume the virtual mass force is proportional to the
relative acceleration between the phases [36]. Computationally, the virtual mass force is important
to stabilize the calculation during the early stages of plume development, by limiting bubble
accelerations. Later, once pseudo-steady conditions are established, the e€ect is negligible [26].
Buoyancy: The buoyancy forces are imposed, unconventionally, as extra source terms in the
momentum Eqs. (4) and (5), and separately for the liquid and gas spaces, i.e. below and above the
liquid surface, respectively, as follows
F Bg ˆ …qg ÿ ql †g; F Bl ˆ 0; below pool surface;
…13†
F Bl …ql ÿ qg †g; F Bg ˆ 0; above pool surface:
Thus, the gas phase is positively buoyant in the pool region below the surface and the liquid phase
is negatively buoyant in the gas region above the surface; otherwise, there is neutral buoyancy.
Apart from these terms, the acceleration due to gravity (g) appears nowhere else in the equation
set. This means that there is no hydrostatic pressure component to the pressure gradient terms
appearing in Eqs. (4) and (5). Though unorthodox, this formulation avoids the need to specify a
reference density, which would be inappropriate for one of the phases and would lead to nu-
merical inaccuracies in the treatment of the pressure drop for that phase.
Drag: The interphase drag force can be derived from a generalization of that for a single bubble
[33,37]:
3 CD
F Dl ˆ ÿF Dg ˆ rg ql jug ÿ ul j…ug ÿ ul †: …14†
4 d
Anticipating a relative velocity between the phases of jug ÿ ul j ˆ 0:25 m=s, corresponding to the
terminal rise velocity of a single 4 mm diameter bubble in a quiescent liquid [38], gives a bubble
Reynolds number Reb ˆ 1200. Form drag dominates in this regime and the drag coecient CD is
then independent of velocity. The standard value, CD ˆ 0.44, is adopted for the present study,
which is appropriate for small bubbles in pure water at the given Reynolds number [11].
Lift force: A bubble rising in a liquid in which there is a gradient in the mean velocity will be
subject to a lift force down the velocity gradient, and arises as a consequence of asymmetries in
the wake. The force is proportional to the relative velocity and the local liquid vorticity. For a
bubble swarm [39]
F Ll ˆ ÿF Lg ˆ CL ql rg …ug ÿ ul †  curl ul : …15†
It was already noted that the rise velocities of the bubbles in the plume have a Gaussian radial
distribution. The action of the lift force on the bubbles will therefore be away from the axis of the
plume, causing the plume to spread laterally. A lift coecient CL ˆ 0.1 is assumed here, in keeping
B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 781

with the experimental observations of Sheng and Irons [25] and the numerical analyses of Lopez
de Bertodano [40].
Turbulent dispersion force: This force has been proposed as an alternative to the inclusion of the
interphase mass di€usion terms in the phasic continuity Eqs. (2) and (3) to account for the dif-
fusion of bubbles due to the random in¯uence of the turbulent eddies in the liquid [33,40]. By
analogy with molecular Brownian di€usion [41], the force is proportional to the local bubble
concentration (or void fraction) gradient, with a di€usion coecient proportional to the eddy
viscosity. The form taken here for the turbulent dispersion force
3 CD
F TD ˆ ÿF TD
g ˆ l rl jug ÿ ul j $rg …16†
l
4 d t
is taken from the work of Davidson [33]. Because the radial distribution of void fraction in the
plume is Gaussian, the bubbles will di€use radially outwards from the axis of the plume under the
action of this force.
Turbulence modelling: Turbulence modelling for two-phase ¯ows requires considerable further
development ± there is no industrial standard model such as the k )  model for single-phase ¯ow.
There is a general belief, however, that the turbulence in the liquid phase has a strong in¯uence on
the void fraction distribution [42,43], and that phenomena such as bubble ¯attening, break-up
and wobble will have feed-back e€ects on the turbulent kinetic energy production [12].
Constant eddy-viscosity models, despite their simplicity, have nonetheless been used success-
fully in bubble-plume applications [21,33], though the current trend is to solve transport equa-
tions for the turbulence quantities, with the k )  model the most popular approach. Solving
mixture equations for the two-phase ¯ow ®eld, Mazumdar and Guthrie [20] found it necessary to
modify the coecients in the k and  equations in order to obtain satisfactory agreement with
measured ¯ow velocities. This work has been challenged by Schwarz and Turner [21] who
maintain that the k )  model with standard (single-phase) coecients can be used, provided a
two-¯uid model is adopted for the ¯ow calculations.
Others have added extra source terms to the k and  equations to account for the increased
generation of 10 turbulence in the liquid due to the momentum exchange between the phases
[12,22,25]. This gives, for the source terms in Eqs. (6) and (7)
Sk ˆ rl
…Pl ÿ l † ‡ C k1 rg rl Pl ‡ Ck2 Cf rg rl k;
C2 2l …17†
S ˆ rl kl ÿ kl ‡ C1 rg rl Pl kll ‡ C2 Cf rg rl ;
C1 1 Pl

in which Cf is the interface friction coecient, given by


3 CD
Cf ˆ jug ÿ ul j:
4 d
The ®rst two terms in each equation are standard for the single-phase k )  model, except for the
multiplication by rl . The third terms have been employed by Malin [44] and Malin and Spalding
[45] to obtain correct spreading rates for single-phase jets and wakes. As originally formulated,
the model made use of an analogy between turbulent and non-turbulent regions of the ¯ow and
the two phases in a multi-phase situation. The intermittency, which is essentially a probability
density function for turbulence, is likened to the volume fraction of two-phase ¯ow. Though the
intention was to use a two-¯uid approach to model single-phase turbulence, the model has been
subsequently used in the context of bubble plume modelling [12,22,25] and is therefore included in
the present work. The ®nal terms in Eq. (17) have been proposed by Simonin and Viollet [46] to
represent the migration of the gas bubbles through the liquid. These models have been used to
model two-phase plumes by Ilegusi and Szekely [22] and by Sheng and Irons [12,25], though the
782 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

Simonin and Viollet model appears to be dimensionally incorrect in the former paper, and the
values adopted for the various coecients were modi®ed in the latter two papers. The modi®ed
coecients are used in the present work
Ck1 ˆ 6:0; C1 ˆ 4:0; Ck2 ˆ 0:75; C2 ˆ 0:6:
Another approach is to assume that the shear-induced and bubble-induced turbulence e€ects are
essentially decoupled, so that separate models may be developed for each, and then linearly su-
perimposed [40,47,48]. There is experimental evidence in support of this partitioning, at least for
dilute bubbly ¯ows [43]. The stress-induced turbulence quantities are calculated from an appro-
priate single-phase turbulence model (the k )  model with standard coecients has been adopted
in more recent applications), while bubble-induced turbulence is derived by analogy to potential
¯ow around a sphere, with anisotropy e€ects included. A variant of this model is used here in
which the turbulent viscosity is calculated according to
kl2 1
lt ˆ Cl ql‡ Clb ql rg djug ÿ ul j with Cl ˆ 0:09; Clb ˆ 1:2: …18†
l 2
The second term in this relation is the asymptotic form of the bubble-induced viscosity, as used by
Sato et al. [47] and Anglart et al., [48] and obviates the necessity to solve further transport
equations.

4. Results

Transient calculations have been performed and continued through to steady-state. This is
usually established after about 5 s, but some calculations have been extended to 10 s for veri®-
cation purposes. Below the water surface, initially ¯at at h ˆ 40 cm, a seed gas volume fraction
rg ˆ 10ÿ5 is initially set to avoid singularities in the gas-phase ®eld equations. Likewise, in the gas
space above the surface, the liquid volume fraction is de®ned similarly. The gas and liquid phases
are assumed to be at rest everywhere at the start of the transient. The buoyancy force is imposed
on the gas phase below the liquid surface and negative buoyancy is applied to the liquid phase
above the surface, as described earlier. Otherwise, no special modelling is required and the surface
is free to deform with the ¯ow. One modelling de®ciency remains, however. The simple expedient
of increasing the gas density to account for the virtual mass of the submerged bubbles in Eq. (12)
is, in order to maintain overall mass conservation, applied also to the gas space above the liquid

Table 1
Bubble plume simulations

Case L TD T(MS) T(SV) TMD BIV Comments

1 ´ ´ ´ ´ ´ ´ No spreading of bubble plume or velocity ®eld


p p
2 ´ ´ ´ ´ Improved bubble plume spreading, velocity unchanged
p
3 ´ ´ ´ ´ ´ Correct bubble plume spreading, velocities unchanged
p p p
4 ´ ´ ´ Poor void fraction distribution
p p p
5 ´ ´ ´ Good void and velocity distributions
p p p p
6 ´ ´ General deterioration in results
p
7 ´ ´ ´ ´ ´ Better liquid velocities at edge of plume
p p p
8 ´ ´ ´ Improved bubble plume spreading, velocity slightly changed
p p p
9i±iv ´ ´ ´ Sensitivity to inlet boundary conditions

(All calculations include buoyancy, virtual mass and drag forces.)


L ± Lift force; TD ± Turbulent dispersion force; T(MS) ± Turbulence model (Malin and Spalding); T(SV) ± Turbulence model
(Simonin and Viollet); TMD ± Turbulent mass di€usion; BIV ± Bubble-induced viscosity.
B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 783

Fig. 3. Void fraction contours at three stages of bubble plume development.

surface which therefore has too much inertia. The drag and added-mass forces are included for all
simulations but other e€ects are added progressively, both singly and in combinations. Not all
calculations are reported here, though the ones exhibiting major trends are summarized in Ta-
ble 1, with further details appearing in the text following.
Case 1 is the base case calculation and contains the drag and virtual mass forces only. Void
fraction contours at three stages in the plume development are plotted in Fig. 3. At t . 1 s the
plume ®rst breaks through the surface of the liquid and thereafter level swelling is established.

Fig. 4. Case 1: Velocity pro®les at h ˆ 50 mm.


784 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

Fig. 5. Case 1: Void fraction pro®les at h ˆ 50 mm.

Steady-state conditions prevail after t . 5 s. Final bubble/liquid rise velocities, and void fraction
distributions, at a height h ˆ 50 mm above the plug surface, are compared with experimental data
in Figs. 4 and 5, respectively. At this elevation, the half-width of the bubble plume is only
marginally greater than the radius of the plug ( ˆ 3 cm) and there is no discernible (axial) motion
in the liquid pool beyond this. The gas and liquid velocities display the characteristic Gaussian
pro®les, and the velocity di€erence between them is about 0.3 m/s, close to the terminal velocity

Fig. 6. Case 1: Velocity pro®les at h ˆ 200 mm.


B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 785

Fig. 7. Case 1: Void fraction pro®les at h ˆ 200 mm.

for bubbles of diameter 4 mm rising in quiescent water [38], and is uniform across the plume. The
calculated maximum gas velocity is about 30% lower than measurement ± a persistent feature of
all calculations in the series. The void fraction distribution is poorly predicted (Fig. 5), with a
depression on the axis.
The velocity and void fraction pro®les at h ˆ 200 mm are shown in Figs. 6 and 7. The rise
velocities of the bubbles now exceed the measured values over the central part of the plume, and

Fig. 8. Case 2: Void fraction pro®les at h ˆ 50 mm.


786 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

Fig. 9. Case 2: Void fraction pro®les at h ˆ 200 mm.

the liquid velocities beyond the plume (r > 50 mm) are also low. The void fraction distribution is
also over-predicted near the centre of the plume, but with the depression on the axis persisting.
Fig. 7 con®rms that the calculation has not reproduced the measured plume spreading (plug
radius ˆ 3 cm). Clearly, mechanisms are required for spreading the bubble plume and enhancing
the viscous shear at the plume edge.

Fig. 10. Case 3: Void fraction pro®les at h ˆ 50 mm.


B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 787

Fig. 11. Case 3: Void fraction pro®les at h ˆ 200 mm.

The lift (15) and turbulent dispersion (16) e€ects provide radial forces on the bubbles pro-
portional to the velocity shear in the liquid and void fraction gradient, respectively. Since both the
velocity and void fraction pro®les are Gaussian, one would expect the in¯uence of these forces on
the bubble distribution to be similar (with magnitudes dependent on their respective empirical
coecients CL and CD ), and cumulative. Detailed calculations, in which the forces are included
separately or in combination, con®rm this expectation. Figs. 8 and 9 show void fraction pro®les

Fig. 12. Case 4: Velocity pro®les at h ˆ 200 mm.


788 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

Fig. 13. Case 4: Void fraction pro®les at h ˆ 200 mm.

at h ˆ 50 mm and h ˆ 200 mm for Case 2, in which both forces are included. The depression on the
axis seen for the base case calculation (Fig. 5) has disappeared and plume spreading has now
de®nitely been reproduced (cf. Fig. 7). Velocity pro®les (not shown) remained virtually un-
changed from those of the base case.
Turbulent mass di€usion has been advanced as an alternative to the lift and dispersion forces in
some bubble plume simulations. The model can be used here by setting the Prandtl±Schmidt

Fig. 14. Case 4: Void fraction pro®les at h ˆ 50 mm.


B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 789

Fig. 15. Case 5: Velocity pro®les at h ˆ 200 mm.

number rr in Eq. (11) to a ®nite (positive) value. The value rr ˆ 1 is adopted in other studies [21±
23] but, in the present application, this produced very little change compared with the base case
calculation. To exaggerate the e€ect therefore, the di€usion coecient is increased by a factor ten
by setting rr ˆ 0.1. This corresponds to Case 3 in Table 1. Void fraction pro®les at h ˆ 50 mm and
h ˆ 200 mm, shown in Figs. 10 and 11, con®rm that the model has a similar e€ect to the inclusion
of the lift and turbulent dispersion forces, presented as Case 2 above. However, for the parameters

Fig. 16. Case 5: Void fraction pro®les at h ˆ 200 mm.


790 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

Fig. 17. Case 7: Velocity pro®les at h ˆ 200 mm.

adopted, the e€ect is less pronounced and results in poorer agreement with experiment. Again,
there was little noticeable change in the velocity ®eld compared with the base case.
The e€ects of including extra source terms in the k and  equations, to represent bubble-in-
duced, turbulent energy production, are explored in the next three calculations in the series. In
Case 4, the terms suggested by Malin and Spalding [45] are included. These lead to an overall
deterioration in results. Figs. 12 and 13 show the velocity and void distributions at h ˆ 200 mm.

Fig. 18. Case 7: Void fraction pro®les at h ˆ 200 mm.


B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 791

The gas and liquid velocity pro®les are essentially the same as those of the base case (Fig. 6), with
the void fraction distribution too peaked. Even though the lift and turbulent dispersion forces are
included, the depression in the void fraction pro®le at h ˆ 50 mm returns (Fig. 14) and is more
pronounced than that for the base case (Fig. 5).
However, including the source terms suggested by Simonin and Viollet [46] resulted in the best
comparisons obtained so far. Figs. 15 and 16 show the velocity and void fraction pro®les at
h ˆ 200 mm for this case (Case 5). The correct rate of spreading of the plume has been predicted
and velocity pro®les are of the correct magnitude and shape. However, outside of the plume
(r > 50 mm), liquid velocities remain smaller than measured values indicating lack of momentum
transfer between the plume and the pool.
The inclusion of both sets of source terms (Case 6) led to a noticeable deterioration in the
predictions (not shown here) and, taken together with the poor results obtained in Case 4, it must
be concluded that the modi®cations to the k )  model suggested by Malin and Spalding are
inappropriate for the present application. The e€ects of enhancing the eddy viscosity by the
addition of a bubble-induced component according to Eq. (18) is examined in the next two
calculations in the series, singly in Case 7, and in combination with the lift and turbulent di€usion
forces in Case 8. Figs. 17 and 18 show the velocity and void fraction pro®les at h ˆ 200 mm for
Case 7. From the latter ®gure it is evident that the model has had little e€ect on the spreading of
the plume, though the pro®le is Gaussian, in contrast to that of the base case (Fig. 7). However,
the velocity pro®le, given by the dashed line in Fig. 17, is considerably broader than for the base
case (Fig. 6), with velocities approaching measured values at the edge of the plume. The main
e€ect of the bubble-induced component of the turbulent viscosity thus appears to be an enhanced
coupling between the plume and the surrounding ¯uid due to increased turbulent shear at the edge
of the plume. This coupling is stronger than that produced by including extra terms in the k and 
equations (Fig. 15).
The lack of adequate pool recirculation has also been noted by Schwarz [26] in a simulation of
one of the higher ¯ow-rate experiments by Anagbo and Brimacombe. He attributed the problem
to spurious vertical currents arising immediately under the pool surface, caused by inadequate
treatment of partially ®lled gas/liquid cells. Ironically, better predictions were obtained by
modelling only the liquid and imposing a ®xed (but deformed) upper free boundary, which forced
lateral ¯ow. In the simulations performed here, the ¯ow rate is much lower and the surface swell is
only 1±2 mm. No unusual, vertical liquid velocities were noticed near the free surface and the
reduced pool recirculation appears here to be due to lack of shear coupling between the plume
and its surroundings.
In Case 8, the bubble-induced viscosity model is combined with the mechanisms identi®ed
earlier for spreading the plume: the lift and turbulent dispersion forces. Figs. 19 and 20 show the
velocity and void fraction pro®les at h ˆ 200 mm for this case. For the parameters adopted, which
are unchanged from Case 2, it is evident comparing Figs. 9 and 20 that plume spreading is now
over-predicted and this has resulted in smaller bubble rise velocities (Fig. 19) compared with those
for Case 7. However, both the bubble and liquid velocity pro®les are now of the correct shape.
Again, no attempt has been made to tune parameters to obtain better comparisons.
Finally, the sensitivity of the results to uncertainties in the speci®cation of the inlet boundary
conditions is explored by performing four repeat calculations in which the e€ective void fraction
rg at the plug surface is varied and, in one case, with the discharge area reduced to the central plug
region. Otherwise, the modelling assumptions are identical to those for Case 5, described above.
In each case the gas inlet velocity ug is adjusted appropriately to maintain the total volumetric
¯ow rate of Q_ ˆ 200 N cm3 =s; details are given in Table 2.
792 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

Fig. 19. Case 8: Velocity pro®les at h ˆ 200 mm.

The base case, Case 9, corresponds to the inlet conditions de®ned earlier in Eq. (1) and is
included here for completeness. Cases 9(i)±(iii) are variants on this, the second case representing
perfect packing of uniformly sized spheres. Case 9(iv) is included as a result of the general ob-
servation made earlier that, for all simulations, bubble rise velocities at the centre of the plume at
h ˆ 50 mm were consistently low compared with measured values, though much better compar-
isons were obtained at higher elevations. It should also be noted (see for example Fig. 4) that the

Fig. 20. Case 8: Void fraction pro®les at h ˆ 200 mm.


B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 793

Table 2
Variants on Case 5 to assess the in¯uence of the inlet boundary conditions

Case no. Void fractions gas/liquid Axial velocities gas/liquid

rg rl ug (m/s) ul (m/s)

9 0.84 0.16 0.101 0.0


9(i) 0.62 0.38 0.137 0.0
9(ii) 0.50 0.50 0.170 0.0
9(iii) 1.0 0.0 0.085 0.0
9(iv) Case 9 with reduced inlet area

experimental data does not display the characteristic Gaussian pro®le at this level, but appears to
be more triangular. A non-uniform bubble distribution at the plug surface might explain the
discrepancy and, in the absence of more speci®c data, an extreme case is taken here in which the
entire gas injection is con®ned to the inner part the plug (.1/3 in area). The inlet boundary
conditions are then
)
rl ˆ 0:16 ul ˆ 0:0 m=s;
…19†
rg ˆ 0:84 ug ˆ 0:257 m=s;
in place of those given in Eq. (1).
Void fraction distributions at h ˆ 50 mm are displayed in Fig. 21 for Cases 9±(iii). At higher
elevations the di€erences are minor, and no discernible di€erences in the velocity pro®les were
apparent at any level. Thus, provided the assumption is made that gas discharge into the tank

Fig. 21. Cases 9±9iii: A comparison of void fraction pro®les at h ˆ 50 mm.


794 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

takes place over the entire plug surface, results are rather insensitive to the speci®c initial void
fraction adopted, except locally.
Velocity pro®les at h ˆ 50 mm are displayed in Fig. 22 for Case 9(iv). Maximum bubble rise
velocities are now much closer to measured values compared with those in Fig. 4. The pro®les are
too sharp and the width of the plume too narrow, re¯ecting the extreme case taken, but at least
indicate that a non-uniform bubble injection rate at the plug could explain the persistent velocity
under-predictions at this elevation. Further calculations were performed in which the e€ects of
adding extra terms to the momentum and turbulence equations, both singly and in combinations,
were explored. In all major respects the trends were similar to those noted already for the uniform
injection cases and no further discussion is given here.

5. Conclusions

CFD calculations have been performed in simulation of an air-water experiment of Anagbo


and Brimacombe [9] in which air bubbles of a well-determined size are generated in a water pool
by bottom injection through a porous plug. For the computations carried out here, a six-equa-
tion, two-¯uid model is utilized and several mechanistic closure laws are evaluated by comparing
code predictions with measured data. Extensions to the k )  turbulence model, which have
variously been advanced to account for the interaction between the bubbles and the liquid
[40,45,46], are also critically assessed. The principal conclusions reached are itemized below.
· Buoyancy, drag and added-mass e€ects must be included for realistic simulation of bubbly
¯ows. These combine to produce the correct initial bubble accelerations and terminal veloc-
ities, and to ensure that the relative slip between the phases is as observed experimentally.
· Lift and turbulent dispersion forces supply lateral forces to radially spread the bubbles, but
have little e€ect on the velocity ®elds. If these forces are not included in the (Reynolds-aver-

Fig. 22. Case 9iv: Velocity pro®les at h ˆ 50 mm.


B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 795

aged) two-phase ¯ow equations, no plume spreading occurs and the characteristic Gaussian
shape of the void fraction distribution in the plume is not reproduced.
· The inclusion of turbulent di€usion terms in the phasic mass continuity equations have a sim-
ilar e€ect to the inclusion of the mechanistic lift and turbulent dispersion forces but, in the
present application, the e€ect had to be magni®ed by a factor 10 to produce comparable
plume spreading.
· The extra source terms in the k and  transport equations, to account for turbulent interaction
between the phases, di€use the velocity ®eld in the plume, producing a broader Gaussian
curve, with little e€ect on the void fraction distribution. The model of Simonin and Viollet
[46] works well, but that of Malin and Spalding [45] appears to be inappropriate for the pres-
ent application.
· Enhancement of the turbulent viscosity by the addition of a bubble-induced contribution to
the Prandtl-Kolmogorov relation (18) has the e€ect of redistributing the liquid velocity ®eld
within the plume, similar to the Simonin and Viollet model, but in addition more strongly
couples the plume to the surrounding liquid. This latter e€ect is important for correctly pre-
dicting the long-term mixing of the pool.
· Calculations have been performed for di€erent assumed inlet conditions. Provided the total
gas ¯ow rate is maintained, and the discharge into the vessel is assumed uniform over the plug
surface, results are insensitive to the actual void fraction distribution adopted, though there
are minor di€erences in the void fraction distributions locally.
· Bubble-rise velocities at 50 mm above the plug surface are consistently under-predicted by the
computations, though correct values have been obtained at higher elevations. Possible causes
would be a non-uniform bubble distribution at the plug, or inappropriate values adopted for
the drag and virtual mass coecients in the developing region of the plume. No ®rm conclu-
sion can be reached on this issue due to lack of appropriate experimental data. However, the
forthcoming LINX-2 experiments will provide opportunities for a more careful scrutiny of
the bubble generation process and could help resolve this discrepancy.
All calculations reported here were carried out assuming 2-D axial symmetry. Some follow-up
calculations carried out in three dimensions [49] con®rmed the validity of this assumption.

Acknowledgements

The author is grateful to his colleagues Andreani, for his input regarding several technical
points relating to this paper, and Dury for his critical reading of the ®nal manuscript.

References

[1] P.M. Stoop, S. Spoelstra, M. Huggenberger, G. Varadi, G. Yadigaroglu, TEPSS ± Technology enhancement of
passive safety systems, Proceedings of the Fifth International Conference on Nuclear Engineering (ICONE 5),
Nice, France, 26±30 May 1997.
[2] J.J. Taylor, K.E. Stahlkopf, The US advanced light water reactor program ± A case for simple, passive safety
systems, Proceedings of the Int. Topl. Mtg. Safety of Next Generation Power Reactors, Seattle, Washington,
American Nuclear Society, 1±5 May 1988, p. 22.
[3] B.L. Smith, T.M. Dury, M.M. Huggenberger, H. N othiger, Analysis of single-phase mixing in open pools, ASME
Winter Annual Meeting, HTD-vol. 209, Anaheim, CA, USA, 1992, pp. 101±109.
[4] P. Coddington, M. Andreani, SBWR-PCCS vent phenomena and suppression pool mixing, Proceedings of the
Seventh International Topical Meeting on Nuclear Reactor Thermal Hydraulics (NURETH-7), vol. 2, Saratoga
Springs, New York, USA, 1995, pp. 1249±1271.
796 B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797

[5] F. De Cachard, S. Lomperski, G.R. Monauni, V. Cavicchia, The ®rst LINX-2 tests, Proceedings of the Fifth
International Conference on Nuclear Engineering (ICONE 5), Nice, France, 26±30 May 1997.
[6] F. Durst, B. Sch onung, K. Selanger, M. Winter, Bubble-driven liquid ¯ows, J. Fluid Mech. 170 (1986) 53±82.
[7] A.H. Castillejos, J.K. Brimacombe, Measurement of physical characteristics of bubbles in gas±liquid plumes: Part
I. An improved electroresistivity probe technique, Metallurgical Trans. B 18 (1987) 649±658.
[8] A.H. Castillejos, J.K. Brimacombe, Measurement of physical characteristics of bubbles in gas±liquid plumes: Part
II. Local properties of turbulent air±water plumes in vertically-injected jets, Metallurgical Trans. B 18 (1987) 659±
667.
[9] P.E. Anagbo, J.K. Brimacombe, Plume characteristics and liquid circulation in gas injection through a porous
plug, Metallurgical Trans. B 21 (1990) 637±647.
[10] Y.Y. Sheng, G.A. Irons, A combined laser doppler anemometry and electrical probe diagnostic for bubbly two-
phase ¯ow, Int. J. Multiphase Flow 17 (5) (1991) 585±598.
[11] Y.Y. Sheng, G.A. Irons, Measurement of the internal structure of gas±liquid plumes, Metallurgical Trans. B 23
(1992) 779±788.
[12] Y.Y. Sheng, G.A. Irons, Measurement and modeling of turbulence in the gas/liquid two-phase zone during gas
injection, Metallurgical Trans. B 24 (1993) 695±705.
[13] M. Iguchi, H. Ueda, T. Uemura, Bubble and liquid ¯ow characteristics in a vertical bubbling jet, Int. J. Multiphase
Flow 21 (5) (1995) 861±873.
[14] K.-H. Tacke, H.-G. Schubert, D.-J. Weber, K. Schwerdtfeger, Characteristics of round vertical gas bubble jets,
Metallurgical Trans. B 16 (1985) 263±275.
[15] M.J. McNallan, T.B. King, Fluid dynamics of vertical submerged gas jets in liquid metal processing systems,
Metall. Trans. B 13 (1982) 165±173.
[16] M. Kawakami et al., Dispersion of bubbles in molten iron and the nitrogen transfer in the bubble dispersion zone
at 1250°C trans., ISIJ 25 (1985) 394±402.
[17] M. Iguchi, Measurement of bubble characteristics in a molten iron bath at 1600°C using an electroresistivity
probe, Metallurgical and Materials Transactions B 26 (1995) 67±74.
[18] D. Mazumdar, R.I.L. Guthrie, The physical and mathematical modelling of gas stirred ladle systems, ISIJ
International 35 (1995) 1±20.
[19] A.H. Castillejos, J.K. Bimacombe, Measurement of physical characteristics of bubbles in gas±liquid plumes: Part
II. Local properties of turbulent air±water plumes in vertically injected jets, Metall. Trans. B 18 (1987) 659±671.
[20] D. Mazumdar, R.I.L. Guthrie, Numerical computation of ¯ow and mixing in ladle metallurgy steelmaking
operations (C.A.S. Method), Appl. Math. Modelling 10 (1986) 25±32.
[21] M.P. Schwarz, W.J. Turner, Applicability of the standard k )  model to gas-stirred baths, Appl. Math. Modelling
12 (1988) 273±279.
[22] O.J. Ilegbusi, J. Szekely, The modeling of gas±bubble driven circulation systems, ISIJ International 30 (9) (1990)
731±739.
[23] O.J. Ilegbusi, J. Szekely, M. Iguch, H. Takeuchi, Z. Morita, A comparison of experimentally measured and
theoretically calculated velocity ®elds in a water model of an argon stirred ladle, ISIJ International 33 (4) (1993)
474±478.
[24] Y. Matsumoto, Y. Murai, Flow structure in bubbly ¯ow: Bubble driven plume in aeration tank, in: V.E. Schrock,
T. Sakaguchi (Eds.), Proceedings of the Japan±US Seminar on Two-Phase Flow Dynamics, Berkeley, CA, USA,
5±11 July 1992, pp. 105±115.
[25] Y.Y. Sheng, G.A. Irons, The impact of bubble dynamics on the ¯ow in plumes of ladle water models,
Metallurgical and Materials Trans. B 26 (1995) 625±635.
[26] M.P. Schwarz, Simulation of gas injection into liquid melts, Appl. Math. Modelling 20 (1996) 41±51.
[27] CFX-F3D Version 4.1 User Manual, AEA Technology, Harwell, UK, 1995.
[28] D.A. Drew, R.T. Lahey Jr., Phase distribution mechanisms in turbulent two-phase ¯ow in channels of arbitrary
cross section, ASME Journal of Fluids Engineering 103 (1981) 583±589.
[29] J.H. Milgram, R.J. van Houton, Plumes from subsea well blowouts, Proceedings of the Third International
Conference on the Behaviour of O€-Shore Structures, vol. 1, Cambridge, MA, USA, 2±5 August 1982, pp. 659±
684.
[30] V. Sahajwalla, J.K. Brimacombe, M.E. Salcudean, SCANINJECT V, Part 2, Fifth International Symposium on
Ladle Metallurgy, Jernkontoret, Sweden, 1989, pp. 103±144.
[31] M. Ishii, Thermo-Fluid Dynamic Theory of Two-Phase Flow, Eyrolles, Paris, 1975.
[32] R.T. Lahey, D.A. Drew, The Current State-of-the-Art in the Modeling of Vapor/Liquid Two-Phase Flow, ASME
(Mech. Engineers), New York, Reprint, 90-WA/HT-13, 1990.
B.L. Smith / Appl. Math. Modelling 22 (1998) 773±797 797

[33] M.R. Davidson, Numerical calculations of two-phase ¯ow in a liquid bath with bottom gas injection: The central
plume, Appl. Math. Modelling 14 (1990) 67±76.
[34] D.C. Besnard, F.H. Harlow, Turbulence in multi-phase ¯ow, Int. J. Multiphase Flow 14 (6) (1988) 679±699.
[35] H. Lamb, Hydrodynamics, Cambridge University Press, Cambridge, 1932.
[36] M. Ishi, K. Mishima, Two-¯uid model and hydrodynamic constitutive relations, Nucl. Eng. Des. 82 (1984) 107±
126.
[37] T.L. Cook, F.H. Harlow, VORT: A computer code for bubbly two-phase ¯ow, Report No. LA-10021-MS, Los
Alamos National Laboratory, 1983.
[38] G.B. Wallis, One-Dimensional Two-Phase Flow, McGraw-Hill, New York, 1969.
[39] D.A. Drew, R.T. Lahey Jr., The virtual mass and lift force on a sphere in rotating and straining inviscid ¯ow, Int.
J. Multiphase Flow 13 (1) (1987) 113±121.
[40] M. Lopez de Bertodano, Turbulent bubbly two-phase ¯ow in a triangular duct, Ph.D. Thesis, Rensselaer
Polytechnic Institute, Troy, NY, 1992.
[41] L.D. Landau, E.M. Lifshitz, Fluid Mechanics, Pergamon Press, Oxford, 1959.
[42] D.A. Drew, R.T. Lahey Jr., Phase-distribution mechanisms in turbulent low-quality two-phase ¯ow in circular
pipes, J. Fluid Mech. 117 (1982) 91±106.
[43] M. Lopez de Bertodano, R.T. Lahey Jr., O.C. Jones, Phase distribution in bubbly two-phase ¯ow in vertical ducts,
Int. J. Multiphase Flow 20 (5) (1994) 805±818.
[44] M.R. Malin, Calculations of intermittency in self-preserving, free turbulent jets and wakes, Imperial College
CFDU Report CFD/83/10, University of London, 1983.
[45] M.R. Malin, D.B. Spalding, A two-¯uid model of turbulence and its application to heated plane jets and wakes,
PCH PhysicoChemical Hydrodynamics 5 (5/6) (1984) 339±362.
[46] O. Simonin, P.L. Viollet, On the Computation of turbulent two-phase ¯ows in the Eulerian formulation,
EUROMECH vol. 234, Toulouse, France, 1988.
[47] Y. Sato, M. Sadatomi, K. Sekoguchi, Momentum and heat transfer in two-phase bubble ¯ow ± I theory, Int. J.
Multiphase Flow 7 (1981) 167±177.
[48] H. Anglart, S. Andersson, M.Z. Podowski, N. Kurul, An analysis of multidimensional void distribution in two-
phase ¯ows, Proceedings of the Sixth International Topical Meeting on Nuclear Reactor Thermal Hydraulics
(NURETH-6), vol. 1, Grenoble, France 5±8 October 1993, pp. 139±153.
[49] B.L. Smith, M. Milelli, An investigation of con®ned bubble plumes, Proceedings of the Third International
Conference on Multiphase Flow 98 ± ICMF'98, Lyon, France, 8±12 June 1998.

También podría gustarte