Date post: | 14-Apr-2018 |
Category: |
Documents |
Upload: | dr-mohammed-azhar |
View: | 214 times |
Download: | 0 times |
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 1/12
Analysis of breakage and coalescence models for bubble columns
J.F. Mitre, R.S.M. Takahashi, C.P. Ribeiro Jr.1, P.L.C. Lage Ã
Programa de Engenharia Quımica — COPPE, Universidade Federal do Rio de Janeiro, PO Box 68502, Rio de Janeiro, RJ 21941-972, Brazil
a r t i c l e i n f o
Article history:
Received 30 January 2010
Received in revised form
11 July 2010
Accepted 20 August 2010Available online 27 August 2010
Keywords:
Bubble
Simulation
Population balance
Bubble columns
Coalescence
Breakage
a b s t r a c t
This work was aimed at evaluating the performance of bubble breakage and coalescence models in
bubble column simulations. A total of five different models were considered (two for breakage and
three for coalescence). These selected models had their parameters estimated using experimental data
of bubble size distributions for the air–water system in an isothermal bubble column. Bubble sizes were
measured with a photographic technique for two gas superficial velocities at three bubbling heights.
Model parameters were estimated using the maximum likelihood technique applied to a one-
dimensional population balance model, which was solved by the method of classes. The results allowed
to conclude, within the experimental uncertainty, that it was necessary to adjust the parameters of
some of the models according to the operational conditions. For other models, the same parameter
values could be used to represent different operational conditions.
& 2010 Elsevier Ltd. All rights reserved.
1. Introduction
Several unit operations in chemical engineering involve the
contact of two or more phases to promote mass and heat transferwith or without chemical reactions. Distillation towers, direct-
contact evaporators, bubble columns and multiphase reactors are
some examples (Ribeiro and Lage, 2004a; Chen et al., 2005; Joshi,
2001).
A multiphase flow is characterized by the existence of at least
two phases. The spatial distribution of these phases characterizes
the flow. A two-phase flow is called disperse when one of these
phases is distributed as small elements. These may be fluid or
solid particles, whereas the other phase must be a fluid.
When the particles in a two-phase dispersed flow can be
differentiated by their own properties (size, temperature, species
concentration, etc.), the flow is called polydisperse. These proper-
ties are called internal variables (Ramkrishna, 2000). One of the
most important internal variables is the particle size, which can beexpressed by mass, diameter, volume or other extensive particle
property.
Particles in a dispersed flow can interact among themselves.
They can aggregate to produce larger particles and they can also
break to generate smaller ones. These processes are important in
chemical engineering applications because they imply the
exchange of relevant properties, like energy or mass, among
mother and daughter particles.
The population balance is a modeling framework that
incorporates the interaction among particles in a dispersed flow(Ramkrishna, 2000). It is based on an equation that states the
conservation of particle number density which is called the
population balance equation (PBE). In order to include particle
interactions, the particle breakage and aggregation phenomena
must be modeled.
In turbulent flows, the main cause of fluid particle breakage is
the particle–turbulence interaction. Basically, the existing break-
age models can be divided into two groups. The first group
comprises statistical models where breakage occurs due to
collisions between particles and turbulent eddies that carry
enough energy to promote particle deformations sufficiently
intense to produce an immediate breakup. The second kind of
models assumes pure kinematic arguments based on cohesive and
disruptive tensions to establish the necessary conditions forparticle rupture (Lasheras et al., 2002).
Aggregation of fluid particles is called coalescence. Coalescence is
assumed to occur in three stages. First, there is a collision between
two particles that are forced against each other by the turbulent
dynamic pressure, thereby forming a thin film between their
interfaces. Then, the fluid within this film is drained until it reaches
a critical thickness at which molecular attractive forces dominate,
quickly promoting the film rupture and leading to coalescence
(Marrucci, 1969; Coulaloglou and Tavlarides, 1977; Prince and
Blanch, 1990; Chesters, 1991; Chaudhari and Hofmann, 1994).
There are several models in the literature for the breakage
and coalescence of fluid particles (Coulaloglou and Tavlarides,
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/ces
Chemical Engineering Science
0009-2509/$- see front matter& 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2010.08.023
à Corresponding author. Tel.: +55 21 25628346; fax: 55 21 25628300.
E-mail address: [email protected] (P.L.C. Lage).
URL: http://www.peq.coppe.ufrj.br/areas/tfd/ (P.L.C. Lage).1 Current address: Center for Energy and Environmental Resources, University
of Texas at Austin, 10100 Burnet Road, building 133, Austin, 78758 TX, USA.
Chemical Engineering Science 65 (2010) 6089–6100
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 2/12
1977; Prince and Blanch, 1990; Martınez-Bazan et al. 1999a,b;
Luo and Svendsen, 1996; Kamp et al., 2001; Wang et al., 2003,
2005b; Andersson and Andersson, 2006). Unfortunately, most of
these models were developed using complex flows, like aerated
stirred tanks or bubble columns, in which the turbulent field is
not well characterized. It is not only inhomogeneous, but also
anisotropic along several length scales (Martınez-Bazan et al.,
1999a).
Chesters (1991) and Chaudhari and Hofmann (1994) reviewedthe existing coalescence models and concluded that they are quite
similar to each other being also based on the same hypotheses.
There are a few models that have been published after these
reviews, but the main conclusion remains. Lasheras et al. (2002)
compared several breakage models and showed the existence of
inconsistencies in almost all of them.
There have been a few studies on the influence of several
bubble coalescence and breakage models on the evolution of the
particle size distribution function (Mitre et al., 2004; Wang et al.,
2005a; Bordel et al., 2006). Wang et al. (2005a) concluded that the
predicted bubble size distributions were quite different when
different bubble coalescence and breakup models were used. They
also concluded that it is necessary to take into account bubble
coalescence and breakup due to different mechanisms. However,
some of the models employed by Wang et al. (2005a) have some
theoretical inconsistencies, like the asymmetry of the coalescence
frequency function generalized from the Hibiki and Ishii (2000)
coalescence model (Mitre, 2006). This could have affected their
results.
In a previous work, Mitre et al. (2004) used bubble size
distribution data obtained in a direct-contact evaporator to test
the bubble breakage model of Luo and Svendsen (1996) and the
coalescence model of Prince and Blanch (1990). These authors
concluded that it was necessary to fit parameters in these models
to obtain a good agreement with the experimental data.
In the specific case of bubble columns, Bordel et al. (2006)
compared the performance of different bubble coalescence and
breakage models in the prediction of the evolution of bubble size
distributions under different experimental conditions. However,only 100 bubbles were analyzed at each experimental condition,
which does not guarantee a statistically representative sample. In
addition, the adopted gas superficial velocities were very low
(below 1.0 cm/s), for which appreciable rates of bubble breakage
and coalescence are not expected.
A somewhat recent trend in the literature is the coupling of the
PBE with CFD codes to model polydispersed multiphase systems,
the so-called PB-CFD simulation (Chen et al., 2005; Jakobsen and
Dorao, 2005; Podila et al., 2007; Frank et al., 2007; Bhole et al.,
2008). Some of these works focused on evaluating breakage and
coalescence models.
Chen et al. (2005) used two-dimensional axisymmetric
Eulerian models for simulating bubble column flows with several
breakage and coalescence closures using the FLUENT software. Inorder to simplify the model, it was assumed that all bubbles share
the same velocity field. For breakage, the models of Luo and
Svendsen (1996) and Martınez-Bazan et al. (1999a,b) were
chosen, whereas in the case of coalescence the models of Prince
and Blanch (1990), Luo (1993) and Chesters (1991) were tested.
The PBE was solved by the Kumar and Ramkrishna (1996) method
of classes. It was concluded that the solution including the
population balance is much better than the one using a single
bubble size even though bubble breakage had to be 10 times more
intense to match the experimental data. The choice of different
breakage and coalescence models did not impact the simulated
results for the flow velocity field, gas hold-up and volume-based
particle distribution function. However, no experimental data for
bubble size distribution were used to validate the simulations.
Furthermore, the simulation results might have been affected by
the use of Miyahara et al. (1983) model to predict the bubble
mean diameter at the gas sparger and by the turbulence closure
model employed.
Podila et al. (2007) analyzed the gas–liquid flow in tubular
reactors using axially symmetric two-dimensional simulations
coupled with population balance modeling of the bubble sizes in
FLUENT. It was shown that the employed turbulence closure
adequately predicts the turbulent intensity field. Similar to Chenet al. (2005), Podila et al. (2007) also assumed that all bubbles
share the same velocity field. The breakage models of Luo and
Svendsen (1996) and Lehr et al. (2002), as well as the coalescence
models of Prince and Blanch (1990), Luo (1993) and Lehr et al.
(2002), were employed. Experimental data on gas hold-up,
turbulent intensity and Sauter mean diameter were used for
comparison. The results were somewhat dependent on the
breakage and coalescence models. Reasonably good predictions
were obtained using the Lehr et al. (2002) breakage model, the
Prince and Blanch (1990) coalescence model, and the Lehr et al.
(2002) coalescence model with a different coalescence efficiency
constant.
Bhole et al. (2008) simulated the axisymmetric steady-state
gas–liquid flow in bubble columns including the PBE solution by
the Kumar and Ramkrishna (1996) method. In this case, a
different velocity field for each bubble class and an algebraic slip
model were considered. A modification of the Prince and Blanch
(1990) coalescence model to correct the fluctuating bubble
velocity was proposed, which led to a decrease in the coalescence
rate, in complete agreement with the well-known characteristic
of the original model to overestimate it. Bubble breakage was
represented by the model of Luo and Svendsen (1996), while the
model of Miyahara et al. (1983) was utilized to predict the initial
bubble sizes. Bhole et al. (2008) concluded that the assumption of
the same velocity field for all bubbles is inadequate. Comparison
of simulations with experimental data for the radial profiles of
mean bubble diameter, gas hold-up and axial liquid velocity
showed a good agreement.
It seems reasonable to assume that if the simulated flowcharacteristics in the multiphase system matched the experi-
mental results, then one would have an indirect evidence of the
adequacy of the breakage and coalescence models used in the
simulation. However, as breakage and coalescence are phenom-
ena with opposite trends, their models might be in error even if
the main simulated flow variables agree with their experimental
data. Therefore, although much has been made on advancing the
modeling of gas–liquid two-phase flow, no work has actually
validated the breakage and coalescence models used in PB-CFD
simulations because, as highlighted by Bhole et al. (2008), this
requires experimental data on local bubble size distributions.
Besides, other sources of errors do exist, such as the sparger
(Miyahara et al., 1983), turbulence and interface force modeling.
Therefore, the evaluation of breakage and coalescence modelsrequires simple models, with much less modeling uncertainties,
but considering different velocities for different bubble sizes
(Bhole et al., 2008).
In this work, we evaluated the application of some breakage
and coalescence models to bubble columns. Instead of a PB-CFD
simulation, a one-dimensional steady-state population balance
model with different bubble velocities was used. Such model is
more adequate to represent the experimental data on bubble size
distributions obtained at different heights by the image analysis
technique. Experiments were performed with the air–water
system in a cylindrical column operating in semi-batch mode at
two distinct gas superficial velocities. The model was solved by
the method of classes of Kumar and Ramkrishna (1996) using the
bubble size distribution data at the lowest height as initial
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089 –61006090
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 3/12
condition. The experimental data were interpreted using different
combinations of the selected breakage and coalescence models. In
order to verify model consistency, parameters were introduced to
correct the breakage and coalescence rates and the coalescence
efficiency. The values of such parameters were estimated by the
maximum likelihood technique. Experimental data uncertainty
was taken into account in evaluating the parameters values and
their standard deviations.
2. Material and methods
Experimental bubble size distributions in a bubble column
were obtained using the same photographic technique previously
employed by Silva and Lage (2000) and Ribeiro and Lage (2004a).
In the following, the apparatus and experimental conditions are
described.
2.1. Experimental set-up
The experimental set-up is shown in Fig. 1. It consisted of a
cylindrical glass column, 7.3cm in inner diameter and 2 m in
height, at the bottom of which the sparger was placed. The gasdistribution system was composed of a ceramic plate containing
89 orifices, 0.5mm in diameter, arranged in a quadrangular
pattern with a pitch of 6.3mm. Beneath this ceramic plate, a
stainless steel porous plate was used to reduce the gas velocity
fluctuations through the orifices and thereby keep bubble
formation in the constant flow rate regime (Gaddis and Vogelpohl,
1986). The distribution plates were fixed between the two flanged
Nylon sections that form the gas distribution system. The system
was sealed with Vitons O-Rings. Three parallel rotameters were
available for flow rate measurement, whose error is estimated to
be about 5%. Liquid temperature during operation was monitored
with the aid of a 1.8-m-long platinum resistance thermometer
(PT-100).
2.2. Experimental procedure
The column was operated in semi-batch, isothermal (251C)
mode using distilled water as the continuous phase and
compressed air as the disperse phase. A constant gauge pressure
of 4 bar was kept at the gas distribution chamber. Air was
considered to be an ideal gas. Due to its sensitivity to impurities
and its importance in the coalescence and breakage phenomena,
the surface tension of the employed distilled water was measured
to be 0.067 N/m in a KSV Sigma 70 tensiometer. Other properties
of air and water came from the literature.
A photographic technique was employed to measure statisti-
cally significant bubble size distributions under different operat-
ing conditions. All photos were taken with a SONY digital camera
(model MAVICAs MVC-FD91) adjusting the shutter speed to 1ms.
The camera flash light provided the required illumination.Aiming at determining the bubble size distribution, five
pictures were periodically taken at three different column heights.
The camera was focused at a ruler located outside the column but
at its middle plane. Therefore, the focus was approximately at the
column center. The focal depth was experimentally determined to
be around 1.5 cm. Thus, the photograph method captures bubbles
at the center of the column and near the column wall, as long as
they are within the focused slab. Due to axisymmetric behaviour
of the mean flow, this does not affect the results if a statistically
relevant number of bubbles are measured. The definition of the
height of a picture was made considering the center of the photo.
The field of view of each photo is approximately 6 cm in height.
The bubble size distributions were obtained for two values of gas
superficial velocity. These conditions are listed in Table 1.
A correction for the bubble volume during its ascension along
the column was computed considering isothermal operation. This
correction accounts for the gas expansion caused by the reduction
in the hydrostatic pressure along the column height. The area of
each bubble was evaluated using a graduate scale fixed to the side
of the column and located at its central plane, on which the
camera was focused for all experiments.
All photos were analyzed using the software Tnimages, which
can calculate the number of pixels associated with any delimited
area in a picture. Thus, in each photo, different regions with the
same area were initially selected in the image of the graduate
scale, so as to obtain a relationship between the areas in pixels
and in square millimeters. Next, the contours of the bubble on
focus were manually traced, the area of each bubble was
determined, and then the diameter of the circle with the samearea was computed. Based on a hypothesis of random positioning
of the bubble in the photographic plane, this diameter was
considered to be equivalent to the one related to the sphere with
the same volume. A minimum of 500 bubbles was analyzed for
each experimental condition in order to guarantee the statistical
significance of the determined size distributions (Ribeiro and
Lage, 2004a). The same photos were analyzed by two different
experimenters, but no significant difference in the obtained
bubble size distributions was noticed.
Due to the curved surface of the glass column and to the
considerable differences between the refractive indices of glass
and air, there is a systematic parallax error in the measurement of
the bubble areas by the photographic technique. In order to
quantify this error and correct the measurements, Ribeiro andLage (2004a) described a procedure that was used in the present
work. This procedure applies a single correction factor value of
0:6970:02 to all bubble diameter measurements.
Fig. 1. Schematic diagram of the experimental set-up: (1) flow meter (2) camera
and (3) bubble column. The arrows indicate the gas flow direction. The operation
was in semi-batch mode.
Table 1
Operating conditions associated with experimental data for bubble size
distributions.
uG (cm/s) Distance from sparger (cm) f d30 (mm)
Base
height
Middle
height
Top
height
2.0 70.2 11.0 64.6 142.0 0.09570.002 4.1 70.2
3.8 70.4 11.0 80.1 132.4 0.10670.002 4.9 70.2
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089–6100 6091
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 4/12
3. Mathematical formulation
3.1. Population balance model
The number of particles per unit volume of the state variable
space is the particle number distribution function, f . Considering
only one internal variable, the particle mass, f (t ,m,z) is called the
particle size distribution function (PSDF). In order to predict the
evolution of the PSDF, it is necessary to solve the populationbalance equation (PBE). Considering breakage and coalescence of
particles and neglecting nucleation and particle growth, the PBE is
given by Ramkrishna (2000) as
@ f
@t þr Á ðUf Þ ¼ H ð1Þ
where U ¼U (t ,m,z) is the velocity of the particle with mass m at
spatial position z and at instant t. H (t ,m,z, y ) is the particle net
production rate, which can be written as
H ¼ BBÀDB þBC ÀDC ð2Þwhere BB, DB, BC and DC represent, respectively, the particle
production (birth) and the destruction (death) by breakage and
the birth and the death by coalescence. For the monovariate
distribution on particle mass, these terms are given by Ram-krishna (2000) as
BBðt ,m,z, y Þ ¼Z 1
mu
zðmu, y ÞP ðmjmu, y Þbðmu, y Þ f ðt ,mu,zÞdmu ð3Þ
DBðt ,m,z, y Þ ¼ bðm, y Þ f ðt ,m,zÞ ð4Þ
BC ðt ,m,z, y Þ ¼ 1
2
Z m
0 f ðt ,mÀmu,zÞ f ðt ,m,zÞQ ðmÀmu,mu, y Þdmu ð5Þ
DC ðt ,m,z, y Þ ¼Z 1
0
f ðt ,m,zÞ f ðt ,mu,zÞQ ðm,mu, y Þdmu ð6Þ
In all these terms, it was assumed that they are affected by ( t ,z)
only through y . From now on, in order to simplify the notation, y
will be omitted in all the equations.
In the present case, the conditional probability P ðmujmÞ has
three important properties (Ramkrishna, 2000). First, none of the
daughter particles can be bigger than the mother particle. This leads
to the second property: the probability of a daughter particle being
smaller than the mother particle is equal to one. Finally, if there is
mass conservation, then the sum of the masses of all daughter
particles was equal to the mother particle mass. This last property is
represented by Eq. (7).
zðmÞZ m
0
P ðmujmÞdmu¼ m ð7Þ
4. A simple PBE model for a bubble column
Considering that the liquid phase is stagnant (no liquid
circulation), the column height is sufficiently small to neglect
the effect of gas expansion on the bubble ascension velocities, f
varies only along the axial direction of the column, and also
assuming steady-state conditions, Eq. (1) simplifies to
U z df
dz ¼ H ð8Þ
where U z is the bubble ascension velocity. In our model, U z was
assumed equal to the terminal velocity, which is a function of the
the bubble diameter. The value of U z was calculated from the
following balance between the drag and the buoyancy forces:
p
6d3 g
ðrÀr g
ÞÀF D
¼0
ð9Þ
where the drag force was given by
F D ¼ p
8d2rðU z Þ2C D
dh
d
2
ð10Þ
The ratio between the hydraulic bubble diameter, dh, and the
equivalent bubble diameter (diameter of the sphere with the
bubble volume), d, was estimated using the correlation recom-
mended by Clift et al. (1978). The drag coefficient was given byC D ¼ C D,isoð f 1 f 2ÞÀ2 ð11Þwhere C D,iso is the drag coefficient for an isolated bubble,
calculated by the drag correlation of Karamanev (1994), f 1 is the
correction factor for the population effect given by Behring
(1936), and f 2 is the wall correction factor given by Clift et al.
(1978). Further details can be found elsewhere (Ribeiro and Lage,
2004b,c; Ribeiro et al., 2005).
The breakage model consists of defining the functions zðmÞ,P ðmujmÞ and b(m). In some works, OðmujmÞ zðmÞP ðmujmÞbðmÞ is
modeled instead, being defined as the specific rate of production
of daughters of mass mu by the breakage of particles with mass m.
From the P ðmujmÞ properties, it is possible to write that
bðmÞ ¼ ½1=zðmÞ R m0 OðmujmÞdm
u. In order to model the coalescence,it is only necessary to define Q ðm,muÞ.
4.1. Breakage models
All the analyzed models assumed zðmÞ ¼ 2, although the
Martınez-Bazan et al. (1999a,b) model can be generalized to
zðmÞ42.
The breakage models were based on two theories. Breakage
occurs due to collisions between particles and turbulent eddies or
due to deformation induced by interaction with the fluid flow
(Lasheras et al., 2002).
The first theory encloses the breakup models of Coulaloglou
and Tavlarides (1977), Prince and Blanch (1990), Luo and
Svendsen (1996), Hagesaether (2002), Lehr et al. (2002), Wanget al. (2003), Andersson and Andersson (2006) and others.
Coulaloglou and Tavlarides (1977), Prince and Blanch (1990)
and Andersson and Andersson (2006) only modeled the breakage
frequency of particles. Coulaloglou and Tavlarides (1977) and
Prince and Blanch (1990) suggested statistical models forP ðmjmu, y Þ. Luo and Svendsen (1996), Hagesaether (2002), Lehr
et al. (2002) and Wang et al. (2003) models suffer from the
hypothesis associated with the eddy sizes which are considered to
promote particle rupture. However, the model of Luo and
Svendsen (1996) is still used in the literature (Podila et al.,
2007; Frank et al., 2007; Bhole et al., 2008) and is the only
breakage model implemented in CFX s 12 (ANSYS Inc., 2009).
For comparison with experimental data, only the models of
Luo and Svendsen (1996) and Martınez-Bazan et al. (1999a,b)were considered. The former was included due to historical
reasons and the latter was considered the most consistent model
available (Lasheras et al., 2002).
The model of Luo and Svendsen (1996) is given by
bðmÞ ¼ K 10:923ð1ÀfÞ e
d2
1=3Z 1
xmin
ð1þxÞ2
x11=3
Z 1=2
0expðÀwc Þ d f v dx
ð12Þand
P ðm1jmÞ ¼
1
2m
R 1xmin
ð1þxÞ2
x11=3expðÀwc Þ dx
R 1xmin
ð1þxÞ2
x11=3 R 1=2
0 exp
ðÀwc
Þd f v dx
ð13Þ
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089 –61006092
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 5/12
with
wc ¼12C f s
bre2=3d5=3x11=3
ð14Þ
C f ¼ f 2=3
v þð1À f vÞ2=3À1 ð15Þ
xmin ¼C x
l
d ð16Þwhere K 1 is the correction factor for the breakage rate, which has
to be estimated. In this model, K 1 represents a correction for the
bubble-eddy collision rate and it would be equal to 1 if this model
were perfect.
Martınez-Bazan et al. (1999a,b) defined the breakage fre-
quency as
bðmÞ ¼ K 1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffibðedÞ2=3À12
s
rd
r d
ð17Þ
where K 1 is the correction factor for the breakage rate, whose
value was experimentally determined by Martınez-Bazan et al.
(1999a) to be equal to 0:2570:03 in the case of air bubbles in
water, and
P ðm1jmÞ ¼
D1
3m1½D2=3
1 ÀD5=3c ½ð1ÀD3
1Þ2=9ÀD5=3c R Dmax
Dmin½D2=3ÀD
5=3c ½ð1ÀD3Þ2=9ÀD
5=3c dD
, D1A½Dmin,Dmax
0, D1=2½Dmin,Dmax
8>>>><>>>>:
ð18Þwhere D1 ¼ d1/d, Dmin ¼ dmin=d, Dmax ¼ dmax=d and Dc ¼ dc /d,
being d the diameter of the mother bubble, d1 the diameter of the
smallest daughter and dc is the critical diameter, which is given by
dc ¼ 12s
br
3=5
eÀ2=5 ð19Þ
The minimum diameter of a daughter bubble, dmin is given by
dmin ¼ ð12s=ðbrdÞÞ3=2eÀ1 ð20Þ
Therefore, the maximum diameter of a daughter bubble, dmax, is
dmax ¼ d 1À dmin
d
3" #1=3
ð21Þ
4.2. Coalescence models
The coalescence frequency of the particles with mass m and mu
is usually decomposed into two independent functions: thecollision frequency, y, and the coalescence efficiency (or coales-
cence probability), Z. Considering these, Q ðmu,mÞ could be
expressed by
Q ðmu,mÞ ¼ yðmu,mÞZðmu,mÞ ð22Þ
The coalescence efficiency is defined as the conditional
probability of occurring coalescence considering that a collision
has already occurred. Basically, it depends on the result of a two-
stage process: the draining of liquid film between the two
particles and the rupture of this film that occurs when its
thickness reaches a critical limit. If the collision time between the
particles is not long enough to allow the film thickness to reach
this limit, coalescence does not occur and the particles bounce.
4.2.1. Collision frequency
Basically, there are three different sources of relative bubble
motion: motion induced by turbulence, motion induced by mean-
velocity gradients and body-force-induced motion, arising from
different bubble slip velocities or wake interactions.
The motion induced by turbulence in the continuous phase is
always present in bubble columns as long as turbulent flow
prevails.
In contrast, the motion induced by mean-velocity gradients issignificant only in small-diameter bubble columns (Kamp et al.,
2001). The only coalescence model for this type of bubble collision
is the one given by Prince and Blanch (1990). However, according
to Prince and Blanch (1990), collisions due to laminar shear occur
if the bubble column operates in the heterogeneous regime, for
which liquid circulation does exist. Since the heterogeneous
regime is associated with gas superficial velocities larger than
4 cm/s for the air–water system, this mechanism was not
considered in the present work.
Finally, the motion induced by body forces has been con-
sidered in the past by two different reasonings. Prince and Blanch
(1990) and Lehr and Mewes (2001) considered that buoyancy is
responsible for the relative bubble motion. However, their models
are not correct because they consider the same pair of particles
twice disregarding the magnitude of the speed of each particle
(Mitre, 2006). Besides, it seems that the relative bubble motion
that generates more bubble coalescence is not caused by buoy-
ancy. In fact, bubble wake interactions are considered to be
responsible for it (Otake et al., 1997; Wu et al., 1998; Colella et al.,
1999; Hibiki and Ishii, 2000; Rafique and Dudukovic, 2004; Wang
et al., 2005b).
Wu et al. (1998) and Hibiki and Ishii (2000) developed models
for the collision frequency due to wake interactions. However,
theses models cannot be directly applied to population balance
models because they were developed for interfacial area models
with one or two bubble sizes. Wang et al. (2005b) generalized the
model of Wu et al. (1998), but the resulting model has a serious
inconsistency because it violates the principle of symmetry of the
coalescence frequency (Ramkrishna, 2000). Therefore, there is stillno appropriate model for coalescence due to collisions induced by
buoyancy. Moreover, this type of bubble interaction is only
relevant in the heterogeneous regime. Thus, it was not considered
in the present work either.
There are two main models of turbulence-induced collisions in
the literature: those given by Prince and Blanch (1990) and Kamp
et al. (2001).
Prince and Blanch (1990) stated the following expression for
the turbulent-induced collision rate:
yðmu,mÞ ¼ K 2S uðdsÞ ð23Þ
where S is the cross-section collision area, S ¼ ðp=4ÞðdþduÞ2, and
K
2
is a correction factor, theoretically equal to 1. Originally, Princeand Blanch (1990) mistakenly wrote S using the radius instead of
the diameters, which was properly corrected here. The fluctuation
velocity, u(ds) is given by
uðdsÞ ¼ ffiffiffi
2p
½ðedÞ2=3 þðeduÞ2=31=2 ð24Þ
For Kamp et al. (2001), u(ds) is a characteristic velocity
between two points separated by the distance ds, which is the
mean diameter, dm ¼ ðdþduÞ=2. Therefore, uðdsÞ ¼ ðedmÞ1=3.
In the model of Kamp et al. (2001) the collision frequency was
given by
y
ðmu,m
Þ ¼2
8
3p
1=2 C t ffiffiffiffiffiffiffiffiffiffi1:61p
S
ðedm
Þ1=3
ð25
Þ
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089–6100 6093
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 6/12
where the parameter C t considers the difference between the
fluctuation velocity of each phase. For the air–water system it was
given by
C 2t ¼ 9þ43:2n=ðe1=3d4=3m Þ
1þ43:2n=ðe1=3d4=3m Þ
ð26Þ
and K 2 is a correction factor, theoretically equal to 1.
4.2.2. Coalescence efficiency
In general, bubble sizes are assumed to be in the inertial
subrange of the turbulence spectrum and the film drainage model
assumes that bubbles are either rigid or deformable fluid particles
with fully mobile interfaces. Taking this into consideration, the
coalescence efficiency for the air–water system in the absence of
any solute can be generalized by the expression (Chesters, 1991):
Z¼ exp½ÀK E ðWeÞ0:5 ð27Þwhere We is a relevant Weber number. Three models of
coalescence efficiency were selected: Prince and Blanch (1990),
Kamp et al. (2001) and Luo (1993) models. All these models are
quite similar and could be expressed in the form of Eq. (27) with
different expressions for K E
and We.For rigid particles, Prince and Blanch (1990) stated
Z¼ exp ÀK 3
ðdeq=2Þ3eð2=3Þr16s
!1=2
lnðh0=h f Þ
ðdeq=2Þ2=3
2666664
3777775
¼ exp ÀK 3lnðh0=h f Þ
3:5636
We
2
0:5" #
ð28Þ
where h0 is the initial thickness of the film formed between the
bubbles, h0¼10À4 m, (Kirkpatrick and Lockett, 1974), h f is the
final thickness of the film formed between bubbles, h f ¼
10À8 m,
(Kim and Lee, 1987) and We is defined as We rðedeqÞ2=3deq=2s,where deq is the equivalent diameter. Luo (1993) defined the
coalescence efficiency as
Z¼ exp ÀK 3½0:75ð1þB2Þð1þB3Þ1=2
ðrd=rþ0:5Þ1=2ð1þB3ÞðWeÞ0:5
" #ð29Þ
where the Weber number is defined as above.
Kamp et al. (2001) stated that
Z¼ exp ÀK 3WeK
C vm
0:5" #
ð30Þ
where WeK is the Weber number based on the relative velocity of
the bubbles at the onset of deformation, V 0, WeK rV 20 deq=2s,
given by V 0 ¼ ðC t = ffiffiffiffiffiffiffiffiffiffi
1:61p Þ½eðdþduÞ=21=3
and C vm is the virtual masscoefficient given by
C vm ¼ LN ÀM 2
LÀ2M þN
1
d3eq
ð31Þ
where the coefficients L, M and N are geometrical parameters
given by convergent series of the diameters of the particles (Kamp
et al., 2001).
In all of these models, K 3 is the correction factor for the
coalescence efficiency, being an experimental parameter of unity
order.
It is important to emphasize that each model of coalescence
efficiency was used only with its respective model of collision
frequency, except the model of Luo (1993) that also employs the
Prince and Blanch (1990) collision frequency model.
5. Numerical procedure
Several numerical methods have been developed for the
solution of the population balance equation. Among them, two
types are commonly employed: the moment based methods
(McGraw, 1997; Marchisio and Fox, 2005) and the sectional
methods (Ramkrishna, 2000; Campos and Lage, 2003). This work
used the method of classes of Kumar and Ramkrishna (1996). By
this method, only two population properties can be exactlyconserved, which are usually taken as the zeroth and first order
moments.
The sectional moment N k includes all the particles with
properties between mk and mk+1 that are represented by the
pivot xk ðmko xkomk þ1Þ and is defined by
N kð z Þ ¼Z mk þ 1
mk
f ðm, z Þdm, k ¼ 1 . . . n ð32Þ
where n is the number of classes.
Following Kumar and Ramkrishna (1996), the method of
classes was applied to (8) leading to the following expression:
U z k
dN k
dz ¼ XiZ j
i, j xkÀ1 r ð xi þ x j Þr xk þ 1ð
1
À0:5dij
ÞCijkQ i, jN iN j
ÀN k X
n
i ¼ 1
N iQ i,k
þXn
i ¼ k
jk,ibiN iÀN kbk ð33Þ
where Ci, j,k and jk,i are defined as
Ci, j,k ¼
xk þ 1Àð x j þ xiÞ xk þ 1À xk
for xkrð x j þ xiÞr xk þ 1
ð x j þ xiÞÀ xkÀ1
xkÀ xkÀ1for xkÀ1rð x j þ xiÞr xk
8>>><>>>:
ð34Þ
jk,i ¼Z xk þ 1
xk
xk þ1Àm
xk þ 1À xkP ðmj xiÞ dmþ
Z xk
xkÀ1
mÀ xkÀ1
xkÀ xkÀ1P ðmj xiÞ dm ð35Þ
Further details about these equations can be found in Ramkrishna(2000) and Campos and Lage (2003).
The breakage models included one empirical parameter, K 1,
whereas the coalescence models have two empirical parameters,
K 2 and K 3. These parameters were estimated by means of the
maximum likelihood method (Bard, 1974; Anderson et al., 1978).
According to this method, the experimental error of all measured
variables is assumed to follow the normal distribution, and the
estimated parameters are those which minimize the objective
function Z given by
Z ¼XNE i ¼ 1
ð$mi À$iÞT V À1
i ð$mi À$iÞ ð36Þ
where V i is the matrix of covariance of the measured variables intest i, $mi and $i are the values calculated by the population
balance model. The minimization problem was solved by the
particle swarm method (Kennedy and Eberhart, 1995) followed by
a direct Newton method for refining the solution utilizing the
software ESTIMA (Pinto et al., 1990; Pinto, 1999; Schwaab et al.,
2008).
6. Results
6.1. Experimental results
After measuring the size of each particle and its respective
error, it was necessary to express these data in terms of the
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089 –61006094
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 7/12
relative frequency, G, and its corresponding error. This was done
following the procedure detailed by Ribeiro and Lage (2004a).
At each experimental condition, two data sets consisting
of around 500 bubble sizes with their measurement error
were generated. Since bubble contours were manually traced,
each data set was obtained with a different operator to properly
check for reproducibility. The bubble sizes were discretized into
five classes. When a larger number of classes was used, the errors
in relative frequency increased too much to allow modeldiscrimination.
The reproducibility of the replicate experiments is illustrated
in Fig. 2 by the individual data sets for uG ¼ 2.0 cm/s at the lowest
height. Similar agreement between the replicates was observed
for all the other experimental conditions. Considering that the
two data sets for each experimental condition actually represent
different estimates of the same quantity, the bubble size values
used in their determination were combined to obtain a single data
set based on a higher number of bubbles (above 1000 bubbles).
These results are shown in Fig. 3 and were used in the parameter
estimation procedure.
Since the experimental results were for relative frequency, it
was necessary to relate this frequency with the sectional moment.
This relation is given by
N i ¼ N T Gi ð37aÞ
N T ¼f
p
6d
3
30
ð37bÞ
d3
30 ¼Z 1
0d3GðdÞddffi
Xn
i ¼ 1
d3i Gi ð37cÞ
where N T is the total particle number density, f is the global gas
hold-up and d30 is the volumetric mean bubble diameter. At each
gas superficial velocity, the value of f was measured by the height
expansion method and it is given in Table 1. The referred table
also includes the values of d30 obtained from the bubble size
distributions at the lowest column height. It is clear that the mean
bubble size increased 20% for a 90%-increase in the gas superficial
velocity. If the Gaddis and Vogelpohl (1986) model for bubble
formation at submerged orifices is used to predict the mean
bubble size at the gas distributor, the estimated bubble diameters
are 3.6 and 4.3 mm for uG ¼ 2.0 and 3.8 cm/s, respectively, which
also implies a 20% increase with the change in gas superficial
velocity. These diameters are different from those listed in Table 1,
indicating that bubble coalescence takes place between the gas
distributor and the first measurement height.
6.2. Simulation results and parameter estimation
6.2.1. Additional considerations about the simulation process
The simulations were carried out by using the size distribution
at the smallest column height as the initial condition of the
problem.
The model consists of Eqs. (33)–(35) and the selected models
represented by
Eqs. (12) and (13) for the breakage model of Luo and Svendsen
(1996),
Eqs. (17) and (18) for the breakage model of Martınez-Bazan
et al. (1999a,b),
Fig. 2. Relative frequency of bubble diameters for uG ¼ 2.0 cm/s at the base height.
Different symbols refer to replicates.
Fig. 3. Relative frequency of bubble diameters with five classes: (a) with
uG¼2.0 cm/s and (b) with uG¼3.8 cm/s. More than 1000 bubbles were analyzed
for each height.
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089–6100 6095
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 8/12
Eqs. (22), (23) and (28) for the coalescence model of Prince and
Blanch (1990),
Eqs. (22), (25) and (30) for the coalescence model of Kamp
et al. (2001) and
Eqs. (22), (23) and (29) for the coalescence model of Luo
(1993).
The evolution of the bubble size distribution along the height
was simulated and compared with the experimental data at thetwo highest column positions.
In the simulations, the global variables f and e were considered
instead of the local variables required by the models. The specific
rate of dissipation of turbulent energy, e, was assumed to be the
input power to the small height bubble column, as given
approximately by e¼ uG g (Bhavaraju et al., 1978).
In the parameter estimation procedure, the errors in uG, f andN i were all considered.
6.2.2. Results of parameter estimation
The lower superficial velocity was chosen to be small enough
to render unlikely bubble breakage by turbulence. This was
confirmed by the Martınez-Bazan et al. (1999a,b) model that
predicts that all existing bubble sizes are stable under suchconditions. Therefore, only coalescence needed to be considered
for the data associated with uG ¼ 2.0 cm/s, and accordingly, only
the two parameters K 2 and K 3 were estimated for these data. The
experimental data for the two highest column heights were used
first alone and then together to fit these model parameters.
Parameter K 3 could not be obtained with statistical signifi-
cance for the present experimental data. In other words, K 3 was
always estimated with errors that were larger than the K 3 value
itself. This is due to the high coalescence efficiency (generally
higher than 0.9) under these experimental conditions. Since the
coalescence efficiency is close to its limiting value of one, any
value of K 3 in the order of magnitude of its standard value can
lead to a successful data regression when both K 2 and K 3 are
estimated. Therefore, K 3 was set to its standard value in each
model and kept constant for the other parameter estimation
attempts.Using the values estimated for K 2 of each model, all simulated
results are very similar to each other because all models are
capable of representing the experimental data with fitted para-
meters. For instance, Fig. 4 shows the results for the normalized
bubble distribution function obtained using the coalescence model
of Prince and Blanch (1990) with K 2 fitted considering the data for
the two highest column heights simultaneously, which gives the
worst agreement with the experimental data. This figure repro-
duces the trends observed for all comparisons between calculated
and experimental data in this work.
Since all analyzed models could represent the data with
appropriate parameter values, it was necessary to analyze the
results statistically to draw conclusions regarding the perfor-
mance of these models. Table 2 shows the fitted K 2 values. The
objective function, Z , was almost constant for the parameter
estimation of a given data set. This means that it was not possible
to judge the adequacy of each model simply by comparing these
simulated results with the experimental data. This is an evidence
of the good representation of the experimental results by all
models.
Table 2 shows that, contrary to the model of Kamp et al. (2001)
and Luo (1993), the Prince and Blanch (1990) model behaved
inconsistently, since the K 2 values fitted using the base-top height
and base-both heights data did not agree within the error ranges.
According to Prince and Blanch (1990), K 2 is close to one.
Nevertheless, the estimates were all in the 10À2–10À3 range.
Using the data for the higher gas superficial velocity, uG¼3.8
cm/s, parameter estimation was carried out for the parameters K 2
(in the coalescence models) and K 1 (in the breakage models). Thevalues obtained for K 2 and K 1 in this case are, respectively, listed
in Tables 3 and 4.
It is possible to observe from Table 3 that the Prince and
Blanch (1990) coalescence model has indeed low internal
consistency. For both breakage models, the K 2 value fitted
using base-middle height and base-top height data in Table 3
did not agree within the error range. This lack of agreement also
occurred for the other two coalescence models but only when
they were used with the Luo and Svendsen (1996) model.
When the Martınez-Bazan et al. (1999a,b) breakage model was
combined with either Kamp et al. (2001) or Luo (1993) models,
the six fitted K 2 values for the two values of uG using the
three combinations of the data sets (base-middle height, base-top
height, base-both heights) agreed within the error range.Thus, when using the model of Luo and Svendsen (1996), Table 3
shows that all coalescence models had estimates for K 2 using
the base-middle height and base-top height data that did not
Fig. 4. Comparison between experiment (uG ¼ 2.0 cm/s) and simulation using the
model of Prince and Blanch (1990) with parameters fitted considering the two
heights simultaneously. The error bars for both experimental and simulated data
are shown.
Table 2
Estimates for K 2 for the coalescence models using the data for uG¼2.0 cm/s.
Coalescence models Base-middle height Base-top height Base-both heights
ðK 27eÞ Z ðK 27eÞ Z ðK 27eÞ Z
Kamp et al. (2001) ð3:072:8Þ Â 10À3 48 ð7:075:8Þ Â 10À3 22 ð5:074:5Þ Â 10À3 28
Luo (1993) ð1:570:7Þ Â 10À3 49 ð3:072:1Þ Â 10À3 24 ð3:573:0Þ Â 10À3 29
Prince and Blanch (1990)
ð8:475:6
Þ Â10À3 48
ð2:071:2
Þ Â10À2 22
ð13:076:0
Þ Â10À3 28
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089 –61006096
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 9/12
agree within their error ranges, indicating that this breakage
model is inadequate.
From Table 4, it can be seen that similarly to the K 2 values, the
estimate for K 1 for the model of Luo and Svendsen (1996) was
much smaller than the expected value, being in the ð1À2Þ Â 10À3
range. Unfortunately, for uG¼3.8 cm/s data, the model of
Martınez-Bazan et al. (1999a,b) predicts that there would be
breakage only for the largest bubble class. Therefore, small errors
in the relative frequency of this bubble class can lead to large
errors in the parameter estimate. Despite this limitation, most K 1
estimates for this model agree within their error ranges with the
value of 0:2570:03 reported by Martınez-Bazan et al. (1999a,b).
Furthermore, when using the Martınez-Bazan et al. (1999a,b)
model together with either Kamp et al. (2001) or Luo (1993)
models, the K 1 values fitted for base-middle height and base-top
height data did not agree in their error ranges because of the
small intensity of the breakage phenomena.
Finally, the values of both parameters (K 1 and K 2) were
adjusted using the experimental data obtained with uG¼2.0 and
3.8 cm/s simultaneously. The fitted values of K 2 and K 1 are given
in Tables 5 and 6, respectively.
Tables 7 and 8 show some statistical functions related to the
estimates of K 1 and K 2 using only the data for uG
¼3.8cm/s and
the data for both superficial velocities, respectively.The correlation coefficient, a, is defined as the ratio of the
covariance between two parameters to the product of the
standard deviation of both parameters. The lower the value of
the correlation coefficient, the lower is the correlation between
the two parameters. There is a 95% probability of the parameters
be uncorrelated if the value of the correlation coefficient is below
0.1. Tables 7 and 8 show that there is no linear correlation
between the parameters and that the objective function has
always small values. This means that the parameters are
independent and relevant to the model.
The mean value of K 1 for the model of Luo and Svendsen
(1996) does not present significant variations in Tables 4 and 6,
but its errors in the latter are very large, being equal to or larger
than the parameter value itself.
Table 4
Estimates for K 1 using the data for uG¼3.8 cm/s.
Using these coalescence models Base-middle height Base-top height Base-both heights
K 17e K 17e K 17e
Considering the breakage model of Luo and Svendsen (1996)
Kamp et al. (2001)
ð1:670:6
Þ Â10À3
ð2:170:4
Þ Â10À3
ð1:670:4
Þ Â10À3
Luo (1993) ð1:170:4Þ Â 10À3 ð1:970:4Þ Â 10À3 ð1:870:4Þ Â 10À3
Prince and Blanch (1990) ð1:170:5Þ Â 10À3 ð1:870:4Þ Â 10À3 ð1:870:4Þ Â 10À3
Considering the breakage model of Martınez-Bazan et al. (1999a,b)
Kamp et al. (2001) 0.6070.30 0.1670.09 0.4270.30
Luo (1993) 0.5770.25 0.1670.11 0.4170.21
Prince and Blanch (1990) 0.5270.30 0.7670.40 0.5870.32
Table 5
Estimates for K 2 using the data for uG¼2.0 and 3.8 cm/s.
Coalescence models Base-middle height Base-top height Base-both heights
K 27e K 27e K 27e
Considering the breakage model of Luo and Svendsen (1996)
Kamp et al. (2001) ð3:370:2Þ Â 10À3 ð3:971:0Þ Â 10À3 ð4:171:1Þ Â 10À3
Luo (1993) ð1:570:4Þ Â 10À3 ð1:571:1Þ Â 10À3 ð1:572:0Þ Â 10À3
Prince and Blanch (1990) ð9:174:0Þ Â 10À3 ð1278Þ Â 10À3 ð1277Þ Â 10À3
Considering the breakage model of Martınez-Bazan et al. (1999a,b)
Kamp et al. (2001) ð3:670:2Þ Â 10À3 ð3:970:7Þ Â 10À3 ð4:071:0Þ Â 10À3
Luo (1993) ð2:170:4Þ Â 10À3 ð2:471:5Þ Â 10À3 ð372Þ Â 10À3
Prince and Blanch (1990) ð774Þ Â 10À3 ð1074Þ Â 10À3 ð874Þ Â 10À3
Table 3
Estimates for K 2 using the data for uG¼3.8 cm/s.
Coalescence models Base-middle height Base-top height Base-both heights
K 27e K 27e K 27e
Considering the breakage model of Luo and Svendsen (1996)
Kamp et al. (2001) ð3:770:2Þ Â 10À3 ð1:670:7Þ Â 10À3 ð2:870:9Þ Â 10À3
Luo (1993) ð1:570:4Þ Â 10À3 ð6:571:4Þ Â 10À3 ð4:071:8Þ Â 10À3
Prince and Blanch (1990)ð9:874
:0Þ Â
10À3
ð4:270
:8Þ Â
10À3
ð6:170
:7Þ Â
10À3
Considering the breakage model of Martınez-Bazan et al. (1999a,b)
Kamp et al. (2001) ð4:270:7Þ Â 10À3 ð2:970:8Þ Â 10À3 ð3:670:7Þ Â 10À3
Luo (1993) ð2:770:9Þ Â 10À3 ð1:870:5Þ Â 10À3 ð2:370:7Þ Â 10À3
Prince and Blanch (1990) ð3:670:8Þ Â 10À3 ð5:970:5Þ Â 10À3 ð4:070:5Þ Â 10À3
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089–6100 6097
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 10/12
The parameter of the model of Martınez-Bazan et al. (1999a,b)
does not present significant differences when its values shown in
Tables 4 and 6 are compared. This was expected because thismodel does not predict bubble breakage for uG ¼ 2.0 cm/s.
From Table 5, it can be seen that all six estimates of K 2 for all
coalescence models agree within their error ranges. However,
their relative errors are quite different. Luo (1993) model showsK 2 relative errors that range from 20% to 133%, while Prince and
Blanch (1990) model shows relative errors around 40–70%.
Finally, Kamp et al. (2001) model shows relative errors smaller
than 27%.
7. Conclusions
This work presents a critical analysis by parameter estimation
of some breakage and coalescence models commonly applied to
bubble columns. Experimental data were obtained in a bubble
column for two different gas superficial velocity at three height
positions. The largest superficial velocity is almost at the end of
the homogeneous bubbling regime for the air–water system.
The coalescence models of Prince and Blanch (1990), Luo
(1993) and Kamp et al. (2001) and the breakage models of Luo and
Svendsen (1996) and Martınez-Bazan et al. (1999a,b) were
analyzed.
Among the coalescence models, Kamp et al. (2001) model wasfound to be more consistent. Its parameter was shown to be
independent of the experimental conditions and its relative error
was the smallest among the parameter errors of all coalescence
models.
Although the evaluation of the breakage models was impaired
by the little intensity of breakage in the experimental conditions,
the Martınez-Bazan et al. (1999a,b) model behaved consistently
when all the available data were used in the parameter
estimation. Besides, its parameter value was in agreement with
the value supplied by these authors in very different experimental
conditions.
Although the experimental data were for a limited range of the
gas superficial velocity, the coalescence model of Kamp et al.
(2001) with K 2
¼ ð471Þ Â 10À3
and the breakage model of Martınez-Bazan et al. (1999a,b) with K 1 ¼ 0:470:3 are recom-
mended for bubble columns operating with air–water in the
homogeneous bubbling regime.
Nomenclature
b(m) breakage frequency of the particle with mass m, sÀ1
BB production of particles by breakage, mÀ3kgÀ1 sÀ1
BC production of particles by coalescence, mÀ3 kgÀ1 sÀ1
C x parameter between 11.4 and 31.4 (21.4 was used in
the present work), dimensionlessC D drag coefficient, dimensionless
C f coefficient associated with the increase in area duethe breakage process, dimensionless
C t parameter that considers the difference between the
fluctuation velocity of each phase, dimensionless
C vm virtual mass coefficient, dimensionlessd equivalent bubble diameter (diameter of the sphere
with the same volume of the bubble), m
d30volumetric mean diameter, m
dh hydraulic bubble diameter, md1 diameter of the smallest daughter particle, m
deq equivalent diameter, 1=deq ¼ 1=duþ1=d, m
D1 diameter of the smallest daughter, D1¼d1/d,
dimensionlessDB destruction of particles by breakage, mÀ3 kgÀ1 sÀ1
DC destruction of particles by coalescence, mÀ3
kgÀ1
sÀ1
Table 6
Estimates for K 1 using the data for uG¼2.0 and 3.8cm/s.
Using these coalescence models Base-middle height Base-top height Base-both heights
K 17e K 17e K 17e
Considering the breakage model of Luo and Svendsen (1996)
Kamp et al. (2001) ð1:070:6Þ Â 10À3 ð1:072:0Þ Â 10À3 ð1:072:0Þ Â 10À3
Luo (1993) ð1:070:9Þ Â 10À3 ð1:071:0Þ Â 10À3 ð1:072:0Þ Â 10À3
Prince and Blanch (1990)ð1:071
:0Þ Â
10À3
ð2:072
:0Þ Â
10À3
ð1:071
:0Þ Â
10À3
Considering the breakage model of Martınez-Bazan et al. (1999a,b)
Kamp et al. (2001) 0:670:3 0:370:1 0:470:3
Luo (1993) 0:570:2 0:370:1 0:470:2
Prince and Blanch (1990) 0:570:3 0:270:1 0:570:3
Table 7
Objective function, Z , and correlation coefficient, a, in the parameter adjustment
using the data for uG¼3.8 cm/s.
Coalescence models Base-middle
height
Base-top
height
Base-both
heights
a Z a Z a Z
Considering the breakage model of Luo and Svendsen (1996)
Kamp et al. (2001) 0.03 2 0.02 5 0.02 8
Luo (1993) 0.02 1 0.03 5 0.03 9
Prince and Blanch (1990) 0.02 1 0.03 6 0.03 9
Considering the breakage model of Martınez-Bazan et al. (1999a,b)
Kamp et al. (2001) 0.03 1 0.02 5 0.02 8
Luo (1993) 0.02 1 0.03 5 0.03 9
Prince and Blanch (1990) 0.02 1 0.03 5 0.02 8
Table 8
Objective function, Z , and correlation coefficient, a, in the parameter adjustment
using the data for uG¼2.0 and 3.8cm/s.
Coalescence models Base-middle
height
Base-top
height
Base-both
heights
a Z a Z a Z
Considering the breakage model of Luo and Svendsen (1996)
Kamp et al. (2001) 0.03 7 0.01 11 0.02 10
Luo (1993) 0.02 5 0.03 12 0.01 11
Prince and Blanch (1990) 0.03 6 0.03 10 0.02 10
Considering the breakage model of Martınez-Bazan et al. (1999a,b)
Kamp et al. (2001) 0.03 2 0.02 5 0.02 8
Luo (1993) 0.02 1 0.03 5 0.03 9
Prince and Blanch (1990) 0.02 1 0.03 5 0.02 8
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089 –61006098
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 11/12
Dc critical diameter, Dc ¼dc /d, dimensionlessdm mean diameter, dm ¼ ðdþduÞ=2, mDmax maximum size of a daughter, Dmax ¼ dmax=d,
dimensionlessDmin minimum size of a daughter, Dmin ¼ dmin=d,
dimensionless
e error for a 95% confidence interval of a given variable,
dimensionless
f particle number distribution function, mÀ3
kgÀ1
f 1 drag coefficient correction for the population effect,
dimensionless f 2 drag coefficient correction for the wall effect,
dimensionless f v volume fraction of the smallest daughter particle,
f v¼(d1/d)3, dimensionlessF D drag force, kgm sÀ2
g acceleration of gravity, m sÀ2
G relative frequency, dimensionlessH particle net production term in the population
balance equation, mÀ3 kgÀ1 sÀ1
h f final thickness of the film formed between bubbles, mh0 initial thickness of the film formed between bubbles, m
K
1
correction factor for the breakage rate, dimensionlessK 2 correction factor for the bubble collision rate,
dimensionlessK 3 correction parameter for coalescence efficiency
models, dimensionless
l eddy size, mm particle mass, kgm1 mass of the smallest daughter particle,
m1 ¼ r g ðp=6Þd31, kg
N sectional zeroth-order moment, mÀ3
n number of classes, dimensionless
Q ðm,muÞ aggregation frequency of the particles with mass m
and mu, m3 sÀ1
Re Reynolds number, dimensionless
S cross-section collision area into two particles,S ¼ ðp=4ÞðdþduÞ2, m2
t time, sU particle velocity, m sÀ1
u(ds) characteristic velocity between two points separated
by distance ds, m sÀ1
U z bubble ascension velocity, m sÀ1
uG gas superficial velocity, m sÀ1
We Weber number, dimensionless y vector of all continuous phase variables that
influence the particle processes
z spatial position vector, m
Z objective function, dimensionless
Greek letters
a correlation coefficient, dimensionless
b constant equal to 8.2, dimensionless
wc critical eddy energy density, dimensionless
e dissipation rate of energy per unit mass, m2 sÀ3
Z coalescence efficiency, dimensionless
r density of continuous phase, kg mÀ3
r g density of dispersed phase, kg mÀ3
s surface tension, kg sÀ2
B diameter ratio given by B¼ d=du, dimensionless
f gas hold-up, dimensionless
y collision frequency, sÀ1
x size of the eddy, dimensionless
xmin minimum size of the eddy, dimensionless
z mean number of particles daughters, dimensionless
Acknowledgments
The authors would like to acknowledge the financial support from
CNPq (Grant nos. 485817/2007-1, 301672/2008-3), CAPES, FAPERJ
(Grant no. E-26/152704/2006), FINEP (Grant no. 01.06.1004.00-Ref.
2524/06) and ESSS.
References
Anderson, T.F., Abrams, D.S., Grens, E.A., 1978. Evaluation of parameters fornonlinear thermodynamic models. A.I.Ch.E. Journal 24 (1), 20–29.
Andersson, R., Andersson, B., 2006. Modeling the breakup of fluid particles inturbulent flows. A.I.Ch.E. Journal 52 (6), 2031–2038.
ANSYS Inc., 2009. ANSYS CFX-12.0 User Manual.Bard, Y., 1974. Nonlinear Parameter Estimation. Academic Press, San Diego.Behring, H., 1936. The flow of liquid–gas mixtures in vertical tubes. Zeitschrift fur
die Gesamte Kalte-Industrie 43, 55–58.Bhavaraju, S.M., Russel, T.W.F., Blanch, H.W., 1978. The design of gas sparged
devices for viscous liquid systems. A.I.Ch.E. Journal 24 (3), 454–466.Bhole, M.R., Joshi, J.B., Ramkrishna, D., 2008. CFD simulation of bubble columns
incorporating population balance modeling. Chemical Engineering Science 63,2267–2282.
Bordel, S., Rafael, M., Villaverde, S., 2006. Modeling of the evolution with length of bubble size distributions in bubble columns. Chemical Engineering Science 61,3663–3673.
Campos, F.B., Lage, P.L.C., 2003. A numerical method for solving the transient
multidimensional population balance equation using an Euler–Lagrangeformulation. Chemical Engineering Science 58 (12), 2725–2744.
Chaudhari, R.V., Hofmann, H., 1994. Coalescence of gas bubbles in liquids. Reviewsin Chemical Engineering 10 (2), 131–190.
Chen, P., Sanyal, J., Dudukovic, M.P., 2005. Numerical simulation of bubblecolumns flows: effect of different breakup and coalescence closures. ChemicalEngineering Science 60, 1085–1101.
Chesters, A.K., 1991. The modelling of coalescence processes in fluid-liquid dispersions:a review of current understanding. Trans IChemE 69 (part A), 259–270.
Clift, R., Grace, J.R., Weber, M.E., 1978. Bubbles, Drops and Particles. AcademicPress, London.
Colella, D., Vinci, D., Bagantin, R., Masi, M., Bakr, E.A., 1999. A study on coalescencebreakage mechanisms in three different bubbles columns. Chemical Engineer-ing Science 54, 4767–4777.
Coulaloglou, C.A., Tavlarides, L.L., 1977. Description of interaction processesin agitated liquid–liquid dispersions. Chemical Engineering Science 32,1289–1297.
Frank, T., Zwart, P.J., Krepper, E., Prasser, H.-M., Lucas, D., 2007. Validation of CFDmodels for mono- and polydisperse air–water two-phase flows in pipes.
Nuclear Engineering and Design 238, 647–659.Gaddis, E.S., Vogelpohl, A., 1986. Bubble formation in quiescent liquids under
constant flow conditions. Chemical Engineering Science 41, 97–105.Hagesaether, L., 2002. Coalescence and break-up of drops and bubbles. Ph.D.
Thesis, Norwegian University of Science and Technology, Trondheim, Noruega.Hibiki, T., Ishii, M., 2000. Two-group interfacial area transport equations at bubbly-
to-slug flow transition. Nuclear Engineering and Design 202, 39–76. Jakobsen, H.H.L., Dorao, C.A., 2005. Modeling of bubble column reactors:
progress and limitations. Industrial & Engineering Chemistry Research 44,5107–5151.
Joshi, J.B., 2001. Computational flow modelling and design of bubble columnreactors. Chemical Engineering Science 56, 5893–5933.
Kamp, A.M., Chesters, A.K., Colin, C., Fabre, J., 2001. Bubble coalescence inturbulent flows: a mechanistic model for turbulence-induced coalescenceapplied to microgravity bubbly pipe flow. International Journal of MultiphaseFlow 27, 1363–1396.
Karamanev, D.G., 1994. Rise of gas bubbles in quiescent liquids. A.I.Ch.E. Journal40 (8), 1418–1421.
Kennedy, J., Eberhart, R.C., 1995. Particle swarm optimization. In: Proceedings of the 1995 IEEE International Conference on Neural Networks, vol. 4, Perth,Australia, pp. 1942–1948.
Kim, W.K., Lee, K.L., 1987. Coalescence behavior of two bubbles in stagnant liquids. Journal o f Chemical E ngineering of Japan 20 (5), 448–453.
Kirkpatrick, R.D., Lockett, M.J., 1974. The influence of approach velocity on bubblecoalescence. Chemical Engineering Science 29, 2363–2373.
Kumar, S., Ramkrishna, D., 1996. On the solution of population balance equationsby discretization—I. A fixed pivot technique. Chemical Engineering Science51 (8), 1311–1332.
Lasheras, J.C., Eastwood, C., Martınez-Bazan, C., Montanes, J.L., 2002. A reviewof statistical models for the break-up of an immiscible fluid immersed intoa fully developed turbulent flow. International Journal of Multiphase Flow 28,247–278.
Lehr, F., Mewes, D., 2001. A transport equation for the interfacial area densityapplied to bubble columns. Chemical Engineering Science 56, 1159–1166.
Lehr, F., Millies, M., Mewes, D., 2002. Bubble-size distribution and flow fields inbubble columns. A.I.Ch.E. Journal 48, 655–663.
Luo, H., 1993. Coalescence, breakup and liquid circulation in bubble column
reactors. Ph.D. Thesis, Norwegian Institute of Technology, Norwegian.
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089–6100 6099
7/27/2019 2010_LAGE_CES_v65_p6089-6100
http://slidepdf.com/reader/full/2010lagecesv65p6089-6100 12/12
Luo, H., Svendsen, H.F., 1996. Theoretical model for drop and bubble breakup inturbulent dispersions. A.I.Ch.E. Journal 42 (5), 1225–1233.
Marchisio, D.L., Fox, R.O., 2005. Solution of population balance equations using thedirect quadrature method of moments. Aerosol Science 36, 43–73.
Marrucci, G., 1969. A theory of coalescence. Chemical Engineering Science 24,975–985.
Martınez-Bazan, C., Montanes, J.L., Lasheras, J.C., 1999a. On the breakup of an airbubble injected into a fully developed turbulent flow. Part 1. Breakupfrequency. Journal of Fluid Mechanics 401, 157–182.
Martınez-Bazan, C., Montanes, J.L., Lasheras, J.C., 1999b. On the breakup of an airbubble injected into a fully developed turbulent flow. Part 2. Size PDF of the
resulting daughter bubbles. Journal of Fluid Mechanics 401, 183–207.McGraw, R., 1997. Description of the aerosol dynamics by the quadrature method
of moments. Aerosol Science and Technology 27, 255–265.Mitre, J.F., 2006. Study of break-up and coalescence models for disperseds flows.
Master’s Thesis, Universidade Federal do Rio de Janeiro, PEQ/COPPE, RJ, Brazil,(in Portuguese).
Mitre, J.F., Ribeiro, C.P., Lage, P.L.C., Pinto, J.C., 2004. Parameter estimation frombubble breakage and coalescence models applied to a direct-contactevaporator. In: ENEMP 2004. pp. 1–8, (in Portuguese).
Miyahara, T., Matsuba, Y., Takahashi, T., 1983. The size of bubbles generated fromperforated plates. International Chemical Engineering 23, 517.
Otake, T., Tone, S., Nakao, K., Mitsuhashi, Y., 1997. Coalescence and breakup of bubbles in liquids. Chemical Engineering Science 32, 377–383.
Pinto, J.C., 1999. ESTIMA—a software for parameter estimation and experimentaldesign, user’s guide. PEQ/COPPE, UFRJ, (in Portuguese).
Pinto, J.C., Lob~ao, M.W., Monteiro, J.L., 1990. Sequencial experimental design forparameter estimation: a different approach. Chemical Engineering Science 45,883–892.
Podila, K., Al Taweel, A.M., Koksal, M., Troshko, A., Gupta, Y.P., 2007. CFDsimulation of gas–liquid contacting in tubular reactors. Chemical EngineeringScience 62, 7151–7162.
Prince, M.J., Blanch, H.W., 1990. Bubble coalescence and breakup in air-spargedbubble columns. A.I.Ch.E. Journal 36, 1485–1499.
Rafique, M.P.C., Dudukovic, M.P., 2004. Computational modeling of gas–liquidflow in bubble columns. Reviews in Chemical Engineering 20 (3–4),225–373.
Ramkrishna, D., 2000. Population Balances—Theory and Applications to Particu-late Systems in Engineering. Academic Press, New York.
Ribeiro, C.P., Borges, C.P., Lage, P.L.D.C., 2005. Modelling of direct-contactevaporation using a simultaneous heat and multicomponent mass-transfermodel for superheated bubbles. Chemical Engineering Science 60, 1761–1772.
Ribeiro, C.P., Lage, P.L.C., 2004a. Experimental study on bubble size distributions ina direct-contact evaporator. Brazilian Journal of Chemical Engineering 21 (01),69–81.
Ribeiro, C.P., Lage, P.L.D.C., 2004b. Direct-contact evaporation in the homogeneousand heterogeneous bubbling regimes. Part ii: dynamic simulation. Interna-tional Journal of Heat and Mass Transfer 47, 3841–3854.
Ribeiro, C.P., Lage, P.L.D.C., 2004c. Population balance modeling of bubble sizedistributions in a direct-contact evaporator using a sparger model. ChemicalEngineering Science 59, 2363–2377.
Schwaab Jr., M.E.C.B., Monteiro, J.L., Pinto, J.C., 2008. Nonlinear parameterestimation through particle swarm optimization. Chemical EngineeringScience 63, 1542–1552.
Silva, L.F.L.R., Lage, P.L.C., 2000. Gas hold-up and bubble sizes in non-isothermalbubble columns. In: Proceedings of the XXVIII Brazilian Congress onParticulate Systems, pp. 207–214, (in Portuguese).
Wang, T, Wang, J., Jin, Y., 2003. A novel theoretical breakup kernel functionfor bubbles/droplets in a turbulent flow. Chemical Engineering Science 58,4629–4637.
Wang, T., Wang, J., Jin, Y., 2005a. Population balance model for gas–liquid flows:influence of bubble coalescence and breakup models. Industrial & EngineeringChemistry Research 44, 7540–7549.
Wang, T., Wang, J., Jin, Y., 2005b. Theoretical prediction of flow regime transition inbubble columns by the population balance model. Chemical EngineeringScience 60, 6199–6209.
Wu, Q., Kim, S., Ishii, M., 1998. One-group interfacial area transport in verticalbubbly flow. International Journal of Heat and Mass Transfer 41, 1103–1112.
J.F. Mitre et al. / Chemical Engineering Science 65 (2010) 6089 –61006100