Академический Документы
Профессиональный Документы
Культура Документы
column reactors
Bai, W.
DOI:
10.6100/IR693280
Published: 01/01/2010
Document Version
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
• A submitted manuscript is the author's version of the article upon submission and before peer-review. There can be important differences
between the submitted version and the official published version of record. People interested in the research are advised to contact the
author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal ?
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
PROEFSCHRIFT
door
Wei Bai
Copromotor:
dr.ir. N.G. Deen
ISBN: 978-90-386-2405-1
Cover design by Jie Fan.
Printed by Ipskamp Drukkers B.V., Enschede, the Netherlands.
For my parents
Contents
Contents vii
1 Introduction 1
1.1 Bubble column reactors . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Experimental techniques . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Computational Fluid Dynamics . . . . . . . . . . . . . . . . . . . 8
1.4 Objectives & outline . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
vii
Contents
viii
Contents
Summary 187
Samenvatting 191
总结 195
Acknowledgement 199
ix
Chapter
1
Introduction
Bubble column reactors are commonly used in chemical, petrochemical, biochemical,
pharmaceutical, metallurgical industries and so on for a variety of processes, i.e.
hydrogenation, oxidation, chlorination, alkylation, chemical gas cleaning, various
bio-technological applications, etc.
Bubble column reactors have advantages of ease of operation, low operating and
maintenance costs as it requires no moving parts, and compactness. Also, they have
the characteristics of high catalyst durability and excellent heat and mass transfer
characteristics. Furthermore, bubble column reactors can be adapted to specific con-
figurations according to practical requirements.
In spite of the simple geometry of bubble column reactors, complex hydrodynamics
and its influence on transport characteristics make it difficult to achieve reliable design
and scale-up of bubble column reactors. Research on bubble column reactors covers
a wide range of activities, i.e. gas holdup, bubble properties, interfacial area, flow
regime, heat and mass transfer, back mixing, pressure drop, etc. Investigations on the
characteristics of bubble column reactors include both experimental and numerical
work. Some experimental techniques and numerical methods used in multiphase
flows are briefly introduced in this chapter. Finally, the objectives and outline of this
thesis will be described.
1
1. Introduction
2
1.1. Bubble column reactors
3
1. Introduction
(a) Cascade bubble column (b) Packed bubble column (c) Multishaft bubble col-
umn
(d) Bubble column with (e) Bubble column with in- (f) Bubble column with ex-
static mixers ternal loop ternal loop
perficial gas velocity, physical properties, solid particle properties and concen-
tration and so on, which influence the performance of this type of gas-liquid
contactors significantly. During the past decades, scientific interest in bubble
column reactors has increased considerably (Deckwer, 1992; Kantarci et al.,
2005). The research on the bubble columns covers a wide range including
gas holdup (i.e. Fair et al., 1962; Shah et al., 1982; Heijnen and Van’t Riet,
1984; Kawase and Moo-Young, 1987; Krishna et al., 1991; Saxena and Rao,
1991; Ruzicka et al., 2001), bubble characteristics (i.e. Abuaf et al., 1978; Lin
and Fan, 1999; Manera et al., 2009; Guet et al., 2003; Luther et al., 2004), in-
terfacial area (i.e. Kataoka et al., 1986; Tan and Ishii, 1990; Revankar and
Ishii, 1993; Delhaye and Bricard, 1994; Kiambi et al., 2001; Manera et al., 2009),
4
1.2. Experimental techniques
flow regime (i.e. Shah et al., 1982; Shaikh and Al-Dahhan, 2007), heat and
mass transfer coefficients (i.e. Deckwer et al., 1974; Wang and Fan, 1978; Shah
et al., 1982; Heijnen and Van’t Riet, 1984; Zehner, 1986; Verma, 1989; Saxena
and Rao, 1991; Avdeev et al., 1992; Lin and Fan, 1999; Merchuk et al., 1994;
Schlüter et al., 1995; Dudley, 1995), back mixing (i.e. Ohki and Inoue, 1970;
Deckwer et al., 1974; Hikita and Kikukawa, 1974; Joshi, 1980, 1982; Heijnen
and Van’t Riet, 1984; Kawase and Moo-Young, 1986; Zehner, 1986; Westerterp
et al., 1987; Wachi and Nojima, 1990; Majumder, 2008), and pressure drop
(i.e. Carleton et al., 1967; Gharat and Joshi, 1992; Molga and Westerterp, 1997;
Majumder et al., 2006).
5
1. Introduction
dynamic gas disengagement (DGD) measures the rate at which the liquid
level or the pressure at different levels in the reactor drops after the gas flow
is shut off. In such way, some characteristics, i.e. the gas holdup, bubble
size distribution, etc. can be obtained (Sriram and Mann, 1977; Daly et al.,
1992; Krishna and Ellenberger, 1996). In addition, photographic techniques,
tomographic techniques, radiographic techniques and so on have also been
used in study of multiphase flows. For a detailed review of the measurement
techniques we refer to Chaouki et al. (1997) and Boyer et al. (2002). Some
examples of experimental techniques used in multiphase flows are listed in
Table 1.1.
6
1.2. Experimental techniques
7
1. Introduction
8
1.3. Computational Fluid Dynamics
d
(mv) = ΣF (1.3)
dt
where m = ρV is the mass of the element. The term on the right hand side ΣF
denotes net force acting on the particle including surface and body forces, i.e.
gravity, pressure, drag, lift, virtual mass, wall force, etc.
The element trajectory is then calculated as:
dr
=v (1.4)
dt
The interfacial coupling between the phases is considered in different
ways. A simple way is only to consider the impact of the local characteristics
of the continuous phase on the individual element and neglect any effects
that the presence of the dispersed phase may have on the continuous phase.
This is usually referred to as one way coupling and is only valid in systems
with a very low fraction of the dispersed phase. When the fraction of the
dispersed phase is relatively high and the effects of the individual elements
on the continuous phase cannot be ignored, two way coupling is required.
For dense systems, four way coupling is necessary to take into account the
additional collision effects between elements.
The Eulerian-Lagrangian model has the advantage of considering the mi-
croscopic transport phenomena by taking into account the direct collision
and hydrodynamic interaction between neighboring elements. Moreover, in
bubble column reactors, the residence time of the gas phase in the form of
bubbles can be easily obtained with the Eulerian-Lagrangian model. Hence,
it is possible to study the back mixing of the gas phase with the theory of
residence time distribution (RTD) and then evaluate the performance of the
bubble column reactor.
A disadvantage of the Eulerian-Lagrangian approach is its relatively high
computational cost for the dispersed phase, particularly in very dense systems
which are common in industrial multiphase chemical reactors. Even though
efforts can be made to improve the model, i.e. improving the efficiency of the
9
1. Introduction
algorithm for dealing with collisions (Hoomans et al., 1996; Wu et al., 2010), it
may not be feasible for the Eulerian-Lagrangian model to keep track of a high
number of discrete elements of the dispersed phase.
The objective of this thesis is to study the bubble properties, i.e. bubble
velocity, bubble size and local void fraction, in heterogeneous flow regime
with a four-point optical fibre probe and to further develop and improve the
Eulerian-Lagrangian model in the study of fluid dynamics, back mixing of
both the gas and the liquid phase and performance of bubble column reactors.
In the experimental work, the accuracy of a four-point optical fibre probe is
investigated with a high speed camera in single bubble experiments. Bubble
properties in the heterogeneous flow regime are then measured with the four-
point optical probe. Effects of the superficial gas velocity and the column
height on the bubble properties are studied. In addition, effect of the gas
distributor on the hydrodynamics and back mixing of the gas phase in bubble
column reactors is investigated by using the Eulerian-Lagrangian model and
the theory of residence time distribution (RTD). The swarm effects on the drag
force in bubble column are studied by comparing simulations using various
drag coefficient correlations from literature and measurements using particle
image velocimetry (PIV). Breakup models in literature are implemented in
the Eulerian-Lagrangian model to consider the coalescence and breakup of
bubbles in turbulent flows. Finally, massless tracer particles which move
with the liquid phase are introduced into the model. Moreover, overall gas
holdup and back mixing of both the gas and the liquid phases are studied and
compared with correlations in literature.
The contributions to these topics are organized in chapters as follows:
Chapter 2 investigates the performance of a four-point optical fibre probe
for the determination of the bubble velocity. Single bubble experiments are
carried out and five liquids with different physical properties are used. The
results obtained from the four-point optical fibre probe are compared with
those obtained by image processing using a high speed camera. The accu-
10
Nomenclature
racy of the bubble velocity determination from the optical probe is discussed
according to the physical properties of the liquid phase.
Chapter 3 studies the bubble properties in the heterogeneous flow regime
in a square bubble column with the four-point optical fibre probe. The per-
formance of the four-point optical fibre probe is further discussed. Bubble
size and specific interfacial area are estimated according to the measurements
from the optical probe. Furthermore, the effects of the column height and the
gas flow rate on the bubble properties are investigated.
Chapter 4 discusses the effect of the gas distributor on the hydrodynamics
and back mixing of the gas phase in a square bubble column by using an
Eulerian-Lagrangian model. Theory of residence time distribution (RTD) is
adopted to study the back mixing of the gas phase.
Chapter 5 studies the drag coefficient correlations for bubble swarms in
literature with the aid of the Eulerian-Lagrangian model. The simulation
results are compared with the measurements of particle image velocimetry
(PIV).
Chapter 6 implements breakup models reported in literature in the
Eulerian-Lagrangian model. Furthermore, critical Weber numbers related
to bubble breakup in turbulent flows are also considered and combined with
a coalescence model. Simulation results of the Eulerian-Lagrangian model are
compared with PIV measurements. Moreover, the bubble size distributions
from the numerical simulations are compared with those obtained from the
four-point optical fibre probe.
Chapter 7 discusses the performance of a square bubble column. Massless
tracer particles are introduced into the Eulerian-Lagrangian model which
move with the liquid phase in the bubble column. Gas holdup and back
mixing of both the gas and liquid phases are investigated and compared with
correlations reported in literature.
Nomenclature
11
1. Introduction
Greek letters
References
N. Abuaf, O. C. Jones Jr., and G. A. Zimmer. Optical probe for local void frac-
tion and interface velocity measurements. Review of Scientific Instruments,
49(8):1090–1094, 1978.
K. Akita and F. Yoshida. Gas holdup and volumetric mass transfer coefficient
in bubble columns: Effects of liquid properties. Ind. Eng. Chem. Process Des.
Develop., 12(1):76–80, 1973.
12
References
W. -D. Deckwer. Bubble column reactors. Chichester [etc.] : Wiley and sons,
1992.
W. -D. Deckwer, R. Burckhart, and G. Zoll. Mixing and mass transfer in tall
bubble columns. Chemical Engineering Science, 29(11):2177–2188, 1974.
W. Dijkhuizen. Derivation closures for bubbly flows using direct numerical simu-
lations. PhD thesis, Enschede, April 2008.
13
1. Introduction
J. J. Heijnen and K. Van’t Riet. Mass transfer, mixing and heat transfer phe-
nomena in low viscosity bubble column reactors. The Chemical Engineering
Journal, 28(2):B21–B42, 1984.
A. Iskandrani and G. Kojasoy. Local void fraction and velocity field descrip-
tion in horizontal bubbly flow. Nuclear Engineering and Design, 204(1-3):
117–128, 2001.
14
References
R. Krishna, P.M. Wilkinson, and L.L. Van Dierendonck. A model for gas
holdup in bubble columns incorporating the influence of gas density on flow
regime transitions. Chemical Engineering Science, 46(10):2491–2496, 1991.
T. -J. Lin and L. -S. Fan. Heat transfer and bubble characteristics from a
nozzle in high-pressure bubble columns. Chemical Engineering Science, 54
(21):4853–4859, 1999.
15
1. Introduction
S. Luther, J. Rensen, and S. Guet. Bubble aspect ratio and velocity measure-
ment using a four-point fiber-optical probe. Experiments in Fluids, 36(2):
326–333, 2004.
16
References
17
1. Introduction
S. K. Wang, S. J. Lee, O. C. Jones Jr., and R. T. Lahey Jr. Local void frac-
tion measurement in techniques in two-phase bubbly flows using hot-film
anemometry. volume 31, pages 39–45, 1984.
P. Zehner. Momentum, mass and heat transfer in bubble columns. part 2. axial
blending and heat transfer. International chemical engineering, 26(1):29–35,
1986.
18
Chapter
2
Single bubble experiment
For the understanding of high void fraction flows in bubble columns, quantitative
information on bubble size, bubble velocity as well as gas void fraction is of crucial
importance. Optical probes have the advantage of low cost, simplicity of operation and
easy interpretation of the results. Moreover optical probes are well-suited to obtain
bubble properties in bubbly flows at high void fraction.
In this chapter, the performance of a four-point optical fibre probe was investigated
by comparing probe data with results obtained with photography. Five liquids with
different material properties were used in combination with air sparged from the dis-
tributor at the bottom a flat bubble column.
It is found that the liquid properties have a significant influence on the bubble ve-
locity obtained from the optical probe. In viscous liquids, the bubble deforms and is
decelerated by the optical probe during the approach process and thus the inaccuracy
of velocity determination results. In low viscosity liquids, the bubble’s wobbling
behavior also results in inaccuracy.
19
2. Single bubble experiment
2.1 Introduction
20
2.2. Experimental setup
piercing of the bubble by the tips of the probe. Meanwhile, signals generated
by the four-point optical fibre probe were recorded simultaneously. Images
taken by the camera and signals generated by the four-point optical fibre probe
were processed afterwards separately. Velocities from the image processing
were treated as reference velocities to evaluate the velocities obtained from the
four-point optical fibre probe. Five liquids with different material properties,
including, viscosity, surface tension, density were used during measurements.
The experimental setup mainly consists of a small flat bubble column (10 ×
110 × 500 mm), an Imager Pro HS CMOS camera with 12 bit, 1024 × 1280
pixel resolution, a four-point optical probe and a light source, which are
schematically shown in Figure 2.1.
During the experiments, an air bubble was released from a small hole in
the center of the bottom plate with a diameter of approximately 1 mm. The
range of produced bubble sizes ranged from 3 to 5 mm.
21
2. Single bubble experiment
The four-point optical probe was positioned at the top of the bubble col-
umn, a few centimeters below the liquid surface. The signals of the four-point
optical probe were read with a LabView program and stored onto a hard disk.
The high speed camera is mounted in front of the bubble column and
a light source illuminated the bubble column from behind, thus employing
a shadowgraphy technique. Shadow images of bubble and the four-point
optical probe were recorded by the camera and stored on the computer. The
image acquisition program is DaVis from LaVision. The applied field of view
of the camera is 700 × 1024 pixels.
The four-point optical probe has three tips of the same length that form
an equivalent angle. The fourth tip is positioned in the center between the
other three tips. The radial distance d = 0.5 mm and the vertical distance is
∆s = 1.4 mm. The geometrical configuration of the probe tips is shown in
Figure 2.2.
22
2.3. Data processing
tion of glycerol were used during the measurements. The physical properties
(at room temperature) of each of the selected liquids are listed in Table 2.1.
pi = ni /N, pi ≤ 0 (2.1)
X
L
pi = 1 (2.2)
i=1
23
2. Single bubble experiment
Suppose the pixels are dichotomized into two classes C0 and C1 (back-
ground and objects, or vice versa) by a threshold at level k. C0 denotes pixels
with levels [1, ..., k], and C1 denotes pixels with levels [k + 1, ..., L]. The prob-
abilities of class occurrence, ω and the class mean levels, µ, respectively are
given by
X
k
ω0 = P(C0 ) = pi = ω(k) (2.3)
i=1
X
L
ω1 = P(C1 ) = pi = 1 − ω(k) (2.4)
i=k+1
and
X
k X
k
µ(k)
µ0 = iP(i|C0 ) = ipi /ω0 = (2.5)
ω(k)
i=1 i=1
X
L X
L
µT − µ(k)
µ1 = iP(i|C0 ) = ipi /ω1 = (2.6)
1 − ω(k)
i=k+1 i=k+1
X
L
µT = µ(L) = ipi (2.7)
i=1
24
2.3. Data processing
zi+1 − zi
vi = (2.9)
∆t
where z is the vertical co-ordinate of the centre of mass of the bubble and ∆t
is the time difference between two subsequent images.
Three characteristic velocities were calculated separately. One is the aver-
age velocity before the bubble hits the probe, which will be termed the terminal
velocity. The second characteristic velocity is the velocity at the moment of
interaction between the bubble and the longest tip of the optical probe. The
third one is the average velocity during the bubble-probe contact, which will
be termed the piercing velocity. These three velocities will be used as a basis
for comparison with the velocity obtained from the four-point optical probe.
25
2. Single bubble experiment
threshold should be set to 10% of the bubble plateau level. However, it is hard
to remove pre-signals in this way if the signal-to-noise ratio of the probe is
low.
In the present study, the entry and exit moments were determined on basis
of the following equations:
and thus identical to the threshold used by Harteveld (2005). In some cases
where a pre-signal occurs, additional processing was employed. The entry
point was selected accordingly at the very beginning of the ascending ramp
after the primary peak.
Based on above steps, the moments of probe entry and exit can be deter-
mined for each bubble. With such information, the time difference between
the upper surface of a bubble hitting the longest tip and the other three tips
can be derived as show in Figure 2.3.
The velocity of bubble was calculated from the following equation:
∆s ∆s
u= = P , i = 1, 2, 3 (2.12)
∆t 1
3 ∆ti
26
2.3. Data processing
where ∆s is the vertical distance between the longest tip and the other three
tips, whereas ∆ti is the time difference between hitting of the longest tip and
short tip i. Due to the dynamics of a bubble during the interaction with the
probe, these three time intervals may vary from one another. This is amongst
others due to the bubble hitting the probe off-centre and irregular deformation
of the upper bubble surface during the interaction. To reduce the error in the
bubble velocity determination, the following selection criterion was used:
∆ti − ∆t
∆t < β, i = 1, 2, 3 (2.13)
Mudde and Saito (2001) and Fortunati et al. (2002) studied the influence of
this criterion on the accuracy of the bubble velocity both numerically and ex-
perimentally. In Mudde’s simulations, in case 8% < β < 20% the influence on
the accuracy is negligible in comparison to other error sources. In Fortunati’s
experiment, no significant effect was observed on the average rise velocity
and its standard deviation in case β < 25%. In the present single bubble
experiments, β = 30% was adopted in order to allow more bubbles to be con-
sidered. The choice of this selection criterion will be discussed in more detail
in chapter 3. Moreover, the chord length of bubble can then be determined
from the bubble velocity u and the time span for the bubble travelling through
the optical probe t, which will also be introduced in the following chapter.
2.3.3 Uncertainty
During image processing, the uncertainty in the instantaneous bubble veloc-
ity is related to the image acquisition rate, the determination of geometrical
properties (i.e. center of mass of the bubble) and applied magnification factor.
The magnification factor is obtained by taking the diameter of the probe sup-
port as a reference length. The DaVis software is used to acquire images and
the time interval between sequential images is very accurate. The accuracy
of geometrical properties of the bubble, such as, centre of mass of the bubble,
bubble diameter and so on, mainly depends on the performance of the image
segmentation method provided that the acquired image has good contrast
between bubble and its background.
27
2. Single bubble experiment
28
2.4. Results and discussion
29
2. Single bubble experiment
30
2.4. Results and discussion
0 0
Tip0 Tip0
Tip1 Tip1
Tip2 Tip2
Signal [Voltage]
Signal [Voltage]
Tip3 Tip3
-2 -2
-4 -4
-6 -6
-0.05 0 0.05 0.1 -0.02 -0.01 0 0.01 0.02 0.03 0.04
Relative time [s] Relative time [s]
0.2
Relative position [m]
0
Velocity [m/s]
0.15
-0.01
(a) Relative position vs. time (b) Vertical velocity vs. relative position
However, the optical probe has little retarding effect on bubbles rising
in low viscosity liquid (i.e. water). It is hard to distinguish changes of the
instantaneous bubble velocity due to the presence of the probe from those
due to the wobbling behavior of the bubble. It is also found that the bubble
velocity from the optical probe is within the range of fluctuation of the bubble
velocity.
31
2. Single bubble experiment
Velocity [m/s]
0.25
-0.01
0.2
(a) Relative position vs. time (b) Vertical velocity vs. relative position
The bubble velocities obtained from both the high speed camera and the
optical probe are compared with the use of parity plots. The terminal bubble
velocities obtained from image processing are used to compare with those
obtained from the optical probe (Figure 2.9). The terminal bubble velocity
is determined by averaging the instantaneous bubble velocities before the
bubble approached the optical probe. Since the field of view of the camera is
far away from the bubble injector, the bubble is believed to rise at its terminal
velocity in the region of interest. This can also be motivated from the plots of
the bubble displacement versus time (i.e. Figure 2.7(a) & 2.8(a)). However, the
presence of the optical probe in the path of the rising bubble seems to obstruct
the rising bubble. For instance, it is obvious that the bubble velocity reduces
slightly when approaching the probe in Figure 2.7(b). Therefore, in order
to evaluate the performance of the optical probe, the bubble velocity at the
moment of interaction with the longest tip of the optical probe is also used as a
basis for comparison with results obtained from the optical probe (Figure 2.10).
In addition, the velocity obtained from the four-point optical fibre probe is the
one which is neither the terminal velocity nor the instantaneous velocity of the
bubble at the moment that the roof of the bubble touches the longest tip of the
probe. This velocity is actually measured during the deceleration of the bubble
32
2.4. Results and discussion
during the contact with the probe. Hence, a comparison between the velocity
obtained from the probe and an average velocity during the interaction with
the probe obtained from the high speed camera is presented as well, as shown
in Figure 2.11.
Optical probe
0.2
0.2
0.15
0.15
0.1
0.05 0.1
0.05 0.1 0.15 0.2 0.25 0.1 0.15 0.2 0.25
HS camera HS camera
The straight lines in the parity plots are identity. In Figure 2.9, it can be seen
that the presence of the four-point optical probe has a considerable retarding
effect on the bubble velocity in viscous liquids. A large viscosity of the liquid
causes deceleration of the bubble movement along the tip. Accordingly, inac-
curacies arise during bubble velocity measurements using the optical probe.
For instance, the average deviation between the terminal bubble velocity and
velocity measured with the probe is about 45% in diethylene glycol. In the
glycerol solution, the deviation is around 33%. In the glycerol solution with
60% glycerol, the deviation is about 29%.
However, the terminal bubble velocities in low viscosity liquids are much
closer to those obtained with the camera. Meanwhile, one can find that
there is significant scatter of the bubble velocity measured by the optical
probe. The wobbling behavior of the air bubble in low viscosity liquids
induces some difficulties for measuring the bubble velocity. The shape of the
bubble is already changing continuously when it hits the tips. The velocity
measurement by the probe is strongly influenced by the dynamics of the
33
2. Single bubble experiment
bubble surface . On the contrary, the obstructing effect of the probe on the
rising bubble is not so significant. The average deviation between the terminal
bubble velocity and the one measured with the probe is about 18% in decane
and 20% in water.
Optical probe
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0.05 0.1 0.15 0.2 0.25 0.05 0.1 0.15 0.2 0.25
HS camera HS camera
In Figure 2.10(a), it can be seen that points in the plot are much closer
to the identity compared with the points in Figure 2.9(a). This can also be
seen from the average deviations. The average deviation between the instant
bubble velocity and that obtained from the optical probe in diethylene glycol
reduces from 45% to 40%. In the two glycerol solutions, the average deviation
also decreases. In glycerol solution (72%), it is about 25%, whereas in the
glycerol solution with 60% glycerol, the deviation is 25%. However, there is
no difference in the average deviation in the two low viscosity liquids.
According to the comparison shown in Figure 2.11(a), one can find that
the discrepancy between the velocity obtained from the optical probe and
the average velocity during the interaction determined from the high speed
camera gets smaller. For instance, the average deviation between the bubble
velocity obtained from the optical probe and the average velocity during the
interaction from image processing in diethylene glycol is 34%. In the glycerol
solutions, the average deviations for the two velocities are 18% in the glycerol
solution (60%) and 19% in the glycerol solution (72%) respectively.
34
2.4. Results and discussion
Optical probe
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0.05 0.1 0.15 0.2 0.25 0.05 0.1 0.15 0.2 0.25
HS camera HS camera
Velocities measured in diethylene glycol with the probe have the highest
deviation within these five liquids. Whereas, the deviation measured in
35
2. Single bubble experiment
decane is the lowest one. The order of magnitude of the Reynolds number
is about 103 in water and about 102 in decane. In these cases, the influence
of viscous forces is much smaller than the inertial force acting on the bubble.
Therefore, viscous forces do not dominate in the range of the investigated
Reynolds numbers. As a result the probe has little influence on the motion of
the bubble in these cases. The deviation in water is slightly larger than that
in decane. This could be due to the larger surface tension of water compared
to that of decane. A large surface tension tends to prevent the bubble from
being pierced.
In measurements with viscous liquids, the order of magnitude of the
Reynolds number is about 10. In this situation, viscous forces have a sig-
nificant influence on the motion of the bubble. Therefore, the deviations of
bubble velocity between high speed camera and the optical probe are larger.
2.5 Conclusions
36
Nomenclature
Nomenclature
37
2. Single bubble experiment
Greek letters
Indices
0, 1 class, [-]
B between-class, [-]
T total, [-]
entry probe entering bubble, [-]
exit probe exiting bubble, [-]
References
E. Barrau, N. Rivière, Ch. Poupot, and A. Cartellier. Single and double optical
probes in air-water two-phase flows: Real time signal processing and sensor
performance. International Journal of Multiphase Flow, 25(2):229–256, 1999.
38
References
S. Luther, J. Rensen, and S. Guet. Bubble aspect ratio and velocity measure-
ment using a four-point fiber-optical probe. Experiments in Fluids, 36(2):
326–333, 2004.
39
Chapter
3
Bubble properties of heterogeneous
bubbly flows in bubble column
This chapter focuses on the measurements of bubble properties in heterogeneous bubbly
flows in a square bubble column. A four-point optical fibre probe was used for this
purpose. The accuracy and intrusive effect of the optical probe was investigated first.
The results show that the optical probe underestimates bubble properties, such as,
bubble velocity and local void fraction. The presence of the probe in the bubble column
influences the local flow conditions. Particularly, when the probe is placed close to the
liquid surface, this influence is more pronounced. Furthermore, two methods for the
determination of the interfacial area were compared. The results from both methods
agree quite well at low superficial gas velocity, whereas significant discrepancies were
obtained at high superficial gas velocity. Finally, the effect of (initial) liquid height
on bubble properties was studied. No significant difference was found for the three
investigated (initial) liquid heights.
41
3. Bubble properties of heterogeneous bubbly flows
3.1 Introduction
42
3.2. Experimental setup
43
3. Bubble properties of heterogeneous bubbly flows
For each measurement a sampling time exceeding thirty minutes was taken.
dc = ut (3.2)
Due to the fact that many bubbles can be detected at each position within
the sampling period, many bubble velocities and chord lengths will be ob-
tained at that position for one measurement. For instance, in a small col-
umn (H/W = 3), the probability density functions of bubble velocity and
chord length at position x/W = 0.5, z/H = 0.75 for superficial gas velocities
u g = 0.005 m/s and u g = 0.04 m/s are respectively shown in Figure 3.2.
In order to obtain the average bubble properties at a certain position, the
probability density functions, such as f (u) and f (dc ), are used to determine
the corresponding mean quantities:
Z ∞
u= u f (u) du (3.3)
0
Z ∞
dc = dc f (dc ) ddc (3.4)
0
44
3.3. Data processing
0.05 0.02
0.04
0.015
0.03
f(dc)
f(u)
0.01
0.02
0.005
0.01
0 0
0 0.5 1 1.5 0 0.002 0.004 0.006 0.008 0.01 0.012
u [m/s] dc [m]
(a) Bubble velocity (u g = 0.005 m/s) (b) Chord length (u g = 0.005 m/s)
0.2 0.25
0.2
0.15
0.15
f(dc)
f(u)
0.1
0.1
0.05
0.05
0 0
0 1 2 3 4 5 0 0.05 0.1
u [m/s] dc [m]
(c) Bubble velocity (u g = 0.04 m/s) (d) Chord length (u g = 0.04 m/s)
45
3. Bubble properties of heterogeneous bubbly flows
mentioned above were used to calculate the average bubble velocity at each
location. Particularly, at high superficial gas velocity, more than 3000 bubbles
were selected, i.e. above 10 000 bubbles were used in case of superficial gas
velocity u g = 0.6 m/s.
3
de = dc (3.6)
2
46
3.3. Data processing
with a zero angle of attack, the mean horizontal diameter of a bubble with
aspect ratio E (Liu and Clark, 1995) can be obtained from:
3 −1
dH = E dc (3.7)
2
Therefore, the equivalent diameter of the bubble is:
3 −2/3
de = E dc (3.8)
2
Once the aspect ratio of the bubble E is known, one is able to determine the
equivalent diameter of the bubble. However, it may not be easy to determine
the aspect ratio in high void fraction flows. Due to the phenomena of coa-
lescence and breakup of bubbles, the bubble size varies and thus, the bubble
shape varies. Hence, assumptions with respect to the shape of the bubbles
may lead to (considerable) inaccuracies in bubble size.
The local specific interfacial area is related to the local void fraction and
the local equivalent bubble size:
6α
de = (3.9)
a
where α is the local void fraction.
This provides an alternative means of estimating the equivalent bubble
size if the specific interfacial area is known. In the next section, we will
introduce techniques for measuring the specific interfacial area using a four-
point optical fibre probe.
47
3. Bubble properties of heterogeneous bubbly flows
where N is the number of times over the sampling time T that an interface
passes through x0 where v0 and n represent the interface velocity and unit
normal vector of the interface respectively at x0 .
Let f (x, t) = 0 represent an interface. The jth interface passes through
x0 at time t = t0 j and can be represented as where f j (x0 , t0j ) is assumed to
sufficiently differentiable, f j (x0 , t0j ) = 0.
N
1 X O f j (x0 , t0j )
a= (3.11)
T ∂f
j=1
∂t (x0 , t0 j )
∂f
f j (xk , tk j ) = f j (x0 , t0 j ) + sk O f j (x0 , t0 j ) · ξk + ∆tk j (x0 , t0j ) k = 1, 2, 3 (3.12)
∂t
48
3.3. Data processing
∂ fj
(x0 , t0 ) A
cosα j = − ∂t
1j
(3.13)
O f j (x0 , t0 j ) A0
∂ fj
(x0 , t0 ) A
cosβ j = − ∂t
2j
(3.14)
O f j (x0 , t0 j ) A0
and
∂ fj
(x0 , t0 ) A
cosγ j = − ∂t
3j
(3.15)
O f j (x0 , t0j ) A0
where
ξ1x ξ1y ξ1z
A0 = ξ2x ξ2y ξ2z (3.16)
ξ3x ξ3y ξ3z
∆t1 j ξ1y ξ1z
s1
∆t2 j
A1 j = ξ2y ξ2z (3.17)
s2
∆t3 j
s ξ3y ξ3z
3
ξ ∆t1j
1x ξ1z
s1
∆t2j
A2j = ξ2x ξ2z (3.18)
s2
∆t3j
ξ3x ξ3z
s3
49
3. Bubble properties of heterogeneous bubbly flows
and
ξ ∆t1j
1x ξ1y
s1
∆t2j
A3j = ξ2x ξ2y (3.19)
s2
∆t3j
ξ3x ξ3y
s3
It follows that the identity cos2 α + cos2 β + cos2 γ = 1 and from equation 3.11
that a is given by:
1 Xq 2
N
a= A1j + A22j + A23 j (3.20)
T|A0 |
j=1
50
3.4. Results and discussion
Z
1
α= α(x, y) dA (3.22)
A A
The surface-averaged void fractions are compared with the gas holdups
obtained from the measurements using the differential pressure transducer.
The comparison for both gas flow rate and gas holdup is given in Table 3.1.
Table 3.1: Comparisons of both gas flow rate and gas holdup.
It can be seen that the gas flow rates obtained from the optical probe
are less than those from the flow meter. The gas flow rate from the optical
probe differs about 30% from that obtained with the flow meter. The surface-
averaged void fraction from the optical probe has a much smaller deviation
compared with results from the differential pressure transducer.
The possible reasons for the observed discrepancies may be various. Be-
sides the assumption of symmetry of flows inside column, the processing
method of the optical probe rules out many bubbles that are rising inside the
column at a large angle of attack and that do not hit the four tips of the probe.
Furthermore, the intrusive effect of the probe is also one of the reasons. In
spite of the discrepancies, the optical fibre probe can still be used in two-phase
bubbly flows to provide meaningful data.
51
3. Bubble properties of heterogeneous bubbly flows
The bubble size is calculated on basis of the two methods described in sec-
tion 3.3.3. For the purpose of simplicity, equation 3.6 is used to calculate
bubble size from the chord length distribution (method A), since the bubble
shape is difficult to determine. Furthermore, equation 3.9 is used to determine
the bubble size from the measurement of the specific interfacial area (equa-
tion 3.20) (method B). Comparisons are made in a tall column (H/W = 6) with
four different superficial gas velocities (u g = 0.005, 0.04, 0.06, 0.08and0.1 m/s).
The results are shown in Figure 3.3.
0.02
ug =0.005m/s, Eqn. 3.6 ug=0.005m/s, Eqn. 3.9
ug=0.04m/s, Eqn. 3.6 ug=0.04m/s, Eqn. 3.9
ug=0.06m/s, Eqn. 3.6 ug=0.06m/s, Eqn. 3.9
ug=0.08m/s, Eqn. 3.6 ug=0.08m/s, Eqn. 3.9
ug=0.1m/s, Eqn. 3.6 ug=0.1m/s, Eqn. 3.9
0.015
de [m]
0.01
It can be seen that the method A agrees well with the method B at low
superficial gas velocity (u g = 0.005 m/s). However, discrepancies arise. The
bubble sizes from method A are much larger than those from method B. That
is reasonable, since the shape of the bubble may no longer be spherical due to
strong coalescence of bubbles. Therefore, method B will be used for further
discussions.
52
3.4. Results and discussion
(a) Mean bubble velocity [m/s] (b) Local void fraction [-]
(c) Specific interfacial area [1/m] (d) Equivalent bubble size [m]
From Figure 3.4, one can find that there are only small differences in
the bubble properties among these three liquid heights. In Figure 3.5, bub-
ble properties at higher superficial gas velocity are compared for two liquid
heights (H/W = 3 and 6).
One can see that the bubble properties show little difference. Hence,
53
3. Bubble properties of heterogeneous bubbly flows
(a) Mean bubble velocity [m/s] (b) Local void fraction [-]
(c) Specific interfacial area [1/m] (d) Equivalent bubble size [m]
one may conclude that the initial liquid height barely influences the bubble
properties. However, it is more obvious that the presence of the optical probe
in column disturbs the local flow. The optical probe is inserted through the
wall at the right side of the column. The bubble properties at the right half part
of the distributions, such as the mean bubble velocity, the local void fraction
and the specific interfacial area, are smaller than those at the left half part of
the distributions. The closer to the side wall the measurement point is, the
larger the difference is.
In addition, it can be seen that the profile of the mean bubble velocity at
superficial gas velocity u g = 0.005 m/s is flat (see Figure 3.4). That is due to
meandering behavior of the bubble plume in the column. However, bubbles
have more probability to rise in the middle of the column than at the sides,
which can be seen from the profiles of the local void fraction. The local
54
3.4. Results and discussion
void fraction has a peak in the middle of the column. Furthermore, one can
also observe that the specific interfacial area has peak in the middle. The
equivalent bubble size is uniform throughout the horizontal direction in the
column.
As the superficial gas velocity increases, the meandering motion of the
bubble plume changes to a circulation pattern. Most of the bubbles rise in
the middle of the column and some move downwards at the wall region
due to liquid circulation. The profiles of the mean bubble velocity become
parabolic. Moreover the mean bubble velocity, the local void fraction and the
specific interfacial area increase accordingly. For instance, the mean bubble
velocity in the center increases from about 0.43 m/s at a superficial gas velocity
u g = 0.005 m/s to around 0.90 m/s at a superficial gas velocity u g = 0.1 m/s.
The local void fraction in the center increases about ten times (from 2.3% to
around 23%). The specific interfacial area increases from around 23 to 170.
However, the equivalent bubble diameter does not increase so much. The
equivalent bubble diameter ranges from 0.0064 m at a superficial gas velocity
u g = 0.005 m/s while the equivalent bubble diameter becomes more or less
constant (around 0.009 m) at higher superficial gas velocity.
In Figure 3.6, one can find that the bubble properties, such as the local void
fraction and the specific interfacial area, remain constant along the height at
superficial gas velocity u g = 0.005 m/s. The mean bubble velocity and the
equivalent bubble diameter possess a little variation along the vertical co-
ordinate. At higher the superficial gas velocity, the mean bubble velocity
decreases slightly along the column height which may be due to the spread-
ing of the bubble plume over the cross sectional area. The local void fraction
55
3. Bubble properties of heterogeneous bubbly flows
(a) Mean bubble velocity [m/s] (b) Local void fraction [-]
(c) Specific interfacial area [1/m] (d) Mean bubble size [m]
Figure 3.6: Axial (or vertical) bubble properties (H/W = 3).
and the specific interfacial area first decrease and then increase slightly. Fur-
thermore, the equivalent bubble size increases a little along the height up to
half the height of the column where the increase slows down in the upper
part of the column. One may notice that the equivalent bubble size increases
with the superficial gas velocity up to u g = 0.06 m/s and then, decreases with
the superficial gas velocity.
The axial evolution of the bubble properties at large initial liquid height
(H/W = 6) are plotted in Figure 3.7. It can be seen that the mean bubble
velocity decreases slightly along the height for all superficial gas velocities.
For u g = 0.005 m/s, the local void fraction remains constant and the specific
interfacial area increases slightly. As a result, the equivalent bubble diameter
shows a small decrease in the upper part of the column. Beyond that super-
56
3.5. Conclusions
(a) Mean bubble velocity [m/s] (b) Local void fraction [-]
(c) Specific interfacial area [1/m] (d) Mean bubble size [m]
Figure 3.7: Axial (or vertical) evolution bubble properties (H/W = 6).
ficial gas velocity, the local void fraction and the specific interfacial area first
decrease and then increase slightly. There are similar observations with those
found in the small column. The increase of the equivalent bubble diameter
slows down along the column height.
3.5 Conclusions
In the present work, bubble properties in bubble columns with different initial
liquid heights have been investigated experimentally. In addition the effect of
the superficial gas velocity has been investigated. A four-point optical fibre
probe was used to measure the relevant quantities, such as bubble velocity,
local void fraction, chord length and specific interfacial area. Bubble size
57
3. Bubble properties of heterogeneous bubbly flows
has been determined by using two different methods. One method is based
on an inverse transform of the chord length distribution using geometrical
probability theory. The equivalent bubble diameter is then related to the mean
chord length. Another method is to determine the equivalent bubble size from
the specific interfacial area which is deduced from the data processing of the
four-point optical probe. Comparison of the two methods reveals that results
from the two methods agree with each other quite well at low superficial gas
velocity. However, differences become apparent as the superficial gas velocity
increases. The former method assumes that bubbles have a spherical shape.
That may be true at low superficial gas velocity due to lack of coalescence of
bubbles. The shape of bubble deviates from spherical when the bubble size
increases and deforms. The assumption of spherical bubbles is not valid any
more. Therefore, the latter method was used for further analysis.
The accuracy of the four-point optical probe was also studied. Mean
bubble velocities and local void fractions at three different depths were used
to calculate area-averaged properties, such as gas flow rate and average void
fraction. The area-averaged gas flow rate was compared with that from a
flow meter whereas the area-averaged void fraction was compared with the
gas holdup obtained from differential pressure measurements. Discrepancies
were found for both methods. Possible reasons for the observed discrepancies
may be due to lack of symmetry of flow inside the column, the processing
method of the optical probe and the intrusive effect of the optical probe.
In addition, the effect of initial liquid height on the bubble properties was
investigated. Bubble properties were compared at three (initial) liquid heights
(H/W = 3, 6 and 9) at a superficial gas velocity u g = 0.005 m/s and for two
liquid heights (H/W = 3 and 6) at higher superficial gas velocities. Results
show that the initial liquid height barely influences the bubble properties at
the same height. However, the axial evolution of bubble properties differs
between the systems with different initial liquid heights. Particularly, the
mean bubble velocity decreases continuously along the column height at
small liquid height while it keeps nearly constant at higher superficial gas
velocities.
58
Nomenclature
Nomenclature
Greek letters
59
3. Bubble properties of heterogeneous bubbly flows
Indices
References
60
References
61
Chapter
4
Numerical analysis of the effect of
gas sparging on bubble column
hydrodynamics
A discrete bubble model (DBM) has been used to study the effect of gas sparger prop-
erties on the hydrodynamics in a bubble column. As a first step the performance of
the model was evaluated by comparison with experimental data. Subsequently, four
different perforated plates with different sparged areas were used as a gas sparger.
Distributions of liquid velocity, turbulent kinetic energy and void fraction in the cen-
tral plane were compared for the four different systems. Furthermore, the effect of the
sparger location was also investigated. It was found that the liquid phase circulation
becomes more pronounced as the sparged area location is more distant from the center
of the bottom plate.
Finally, gas phase Residence Time Distributions (RTD) were obtained from the sim-
ulations. By employing standard axial dispersion model, the gas phase mixing in the
bubble column was characterized. Results show that the extent of mixing increased
when the sparged area decreased. The axial dispersion coefficient increased as the
sparged area was shifted to the edge of the bottom plate.
63
4. Effect of gas sparging on bubble column hydrodynamics
4.1 Introduction
64
4.2. Discrete bubble model
Our discrete bubble model (DBM) was originally developed by Delnoij et al.
(1997) and Delnoij et al. (1999). The liquid phase hydrodynamics is repre-
sented by volume-averaged continuity and Navier-Stokes equations, while
the motion of each individual bubble is tracked in a Lagrangian way.
65
4. Effect of gas sparging on bubble column hydrodynamics
dv
ρg V = ΣF (4.1)
dt
dr
=v (4.2)
dt
The net force acting on each individual bubble is calculated by considering
all the relevant forces. It is assumed that the net force is composed of separate,
uncoupled contributions due to gravity, pressure, drag, lift, virtual mass and
wall force respectively:
ΣF = FG + FP + FD + FL + FVM + FW (4.3)
FG = ρ g Vg (4.4)
FP = −V∇P (4.5)
1
FD = − CD ρl πd2 |v − u|(v − u) (4.6)
8
where the drag coefficient is taken from Tomiyama et al. (1998):
16 48 8 Eo
CD = max min (1 + 0.15Re0.687 ), , (4.7)
Re Re 3 Eo + 4
where Re and Eo are the bubble Reynolds number and the Eötvös number
respectively.
66
4.2. Discrete bubble model
Zhang et al. (2006) studied the influence of the lift coefficient on the bubble
column dynamics using a two-fluid model. It was found that the results with
CL = 0.5 fit best with PIV experimental data. Therefore, CL = 0.5 is used in
the present study as well.
Accelerating bubbles experience a resistance, which is termed the virtual
mass force (Auton, 1987):
Dv Du
FVM = −CVM ρl V − (4.9)
Dt Dt
where the D/Dt operators denote the material derivatives pertaining to the
respective phase. In the present work, bubbles are assumed to have a spherical
shape and a virtual mass coefficient of CVM = 0.5 is used.
Bubbles in the vicinity of a solid wall experience a force referred to as the
wall force (Tomiyama et al., 1995):
" #
1 1 1
FW = − C W d 2 − ρl |(v − u)· nz |2 nW (4.10)
2 y (L − y)2
where nz and nW , respectively, are the normal unit vectors in the vertical
and wall normal direction, L is the dimension of the system in the normal
direction, and y is the distance to the wall in that direction. Finally, the wall
force coefficient CW is given by:
exp(−0.933Eo + 0.179) 1 ≤ Eo ≤ 5,
CW =
(4.11)
0.007Eo + 0.04 5 < Eo ≤ 33.
67
4. Effect of gas sparging on bubble column hydrodynamics
equations. The presence of the bubbles is reflected by the liquid phase volume
fraction αl and the interphase momentum transfer rate Φ:
∂
αl ρl + ∇ · αl ρl u = 0 (4.12)
∂t
∂
αl ρl u + ∇ · αl ρl uu = −αl ∇p − ∇ · (αl τl ) + αl ρl g + Φ (4.13)
∂t
The liquid phase is assumed to be Newtonian, thus the stress tensor τl can be
expressed as:
2
τl = −µeff,l (∇u) + (∇u)T − I(∇ · u) (4.14)
3
where µeff,l is the effective shear viscosity. In the present model, the effective
viscosity is composed of two contributions, the molecular viscosity and the
turbulent viscosity:
µeff,l = µL,l + µT,l (4.15)
68
4.2. Discrete bubble model
where Bβ = β11 β22 − β212 + β11 β33 − β213 + β22 β33 − β223 , βij = ∆2m αmi αm j and
αij = ∂u j /∂xi . ∆i is the filter width in the i direction.
where rlm = rl − rm and vlm = vl − vm . Note that if rlm · vlm > 0 the bubbles are
moving away from each other and will not collide. In case of a collision with
a wall the collision time follows simply from the distance to the wall and the
69
4. Effect of gas sparging on bubble column hydrodynamics
normal velocity component toward that wall, which leads for a vertical wall
(i.e. xwall = 0) to the following expression:
rl,x − (xwall + Rl )
tl,wall = (4.21)
vl,x
For each bubble, the minimum collision time is determined by scanning
all collision partners for a possible collision. In order to prevent calculating
the collision time for a certain bubble and its partner twice, only those bubbles
with index larger than the bubble’s index (i.e. l) and obstacles are considered
in the possible collision partners N(l).
Once the minimum collision time for each bubble is determined, a global
minimum collision time (i.e. tab ) and the corresponding collision pair (i.e. a
and b) can be found. First, all bubble positions are updated to the instant of
the collision using a simple explicit integration:
Following the movement of all bubbles, collision pairs a and b are touching
and the collision dynamics is applied to process the collision event. The post-
collision velocities of a and b can be calculated according to Hoomans et al.
(1996).
The bubble column studied here is shown in Figure 4.1. The cross-sectional
area of the column is 0.15 m × 0.15 m (W × D). The column is initially filled
with water to a height of 0.45 m (H). Air is used as the dispersed phase and
introduced into the column through a perforated plate at the bottom of the
column. The material properties of both phases are taken according to room
temperature. The bubble column is operating at atmospheric pressure.
The effects of different configurations of gas spargers on the column hy-
drodynamics are investigated. Perforated plates with 9 holes (3 × 3), 49 holes
(7 × 7), 225 holes (15 × 15) and 484 holes (22 × 22) as shown in Figure 4.2 are
used. All the holes have a diameter of 1 mm and are located in the central
region of the plate with a square pitch of 6.25 mm.
70
4.3. Simulation details
71
4. Effect of gas sparging on bubble column hydrodynamics
Furthermore, the effect of the position of the sparged area on the perfor-
mance of the bubble column is studied by using perforated plates with 49
holes. That is, besides the perforated plate with 49 holes used above, another
two perforated plates with the same number of holes are adopted for the pur-
pose, as shown in Figure 4.3. The centers of the sparger regions are located at
one-fourth, one-third and a half of the width of the column.
72
4.4. Results and discussion
In order to validate the discrete bubble model and the adopted closures, the
simulation results are compared with the experimental data reported by Deen
(2001). Information at three different heights (z/H = 0.35, 0.55, and0.75) in the
column was extracted from the simulation using a distributor with 49 holes.
The two eddy-viscosity models (Smagorinsky, 1963; Vreman, 2004) were used
for the simulations.
The profiles of the time-averaged vertical velocity of the liquid phase are
presented in the Figures 4.4(a)–4.4(c). It can be seen that the results obtained
with the eddy-viscosity model proposed by Vreman (2004) agree with the
experimental data better than those obtained with the eddy-viscosity model
of Smagorinsky (1963). This difference is much more obvious in the lower
part of the column(i.e. z/H < 0.5), whereas in the higher part of the column,
the simulation data from both two eddy-viscosity models are close to the
experimental data.
Furthermore, the profiles of turbulent kinetic energy (tke) from PIV mea-
surements and simulations are also compared (Figures 4.4(d)–4.4(f)). Note
that there are only two components of the liquid velocity available from PIV
measurements. Therefore, the reported turbulent kinetic energy obtained
from experimental data is calculated by assuming that the two horizontal
0 0
components of the liquid velocity are equal, i.e. k ≈ 12 (2ux2 + uz2 ). The turbu-
lent kinetic energy from the simulations, however, is obtained from all three
0 0 0
components of the liquid velocity, i.e. k = 21 (ux2 + u y2 + uz2 ).
It can be seen that the turbulent kinetic energy from the simulations is much
smaller in comparison with the experimental data at the lower height (z/H =
0.35). However, the difference again becomes smaller along the column height.
Meanwhile, it can clearly be seen that the simulations with the eddy-
viscosity model of Vreman (2004) are in better agreement with the experiments
than those with the eddy-viscosity model proposed by Smagorinsky (1963).
The above comparisons demonstrate that the discrete bubble model, with
supplemented proper force closures and turbulence eddy-viscosity model,
73
4. Effect of gas sparging on bubble column hydrodynamics
0.3
0.25
0.2
0.2
0.15
uz [m/s]
uz [m/s]
0.1
0.1
0.05
0.0 0
PIV data PIV data
Vreman (2004) -0.05 Vreman (2004)
Smagorinsky (1963) Smagorinsky (1963)
-0.1 -0.1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/W [-] x/W [-]
(a) Vertical liquid velocity (z/H = 0.35) (b) Vertical liquid velocity (z/H = 0.55)
0.2 0.03
0.15
0.1
0.02
tke [m2/s2]
uz [m/s]
0.05
0
0.01
-0.05
PIV data PIV data
-0.1 Vreman (2004) Vreman (2004)
Smagorinsky (1963) Smagorinsky (1963)
0.00
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/W [-] x/W [-]
(c) Vertical liquid velocity (z/H = 0.75) (d) Turbulent kinetic energy (z/H = 0.35)
0.03
0.02
0.02
tke [m2/s2]
tke [m2/s2]
0.01
0.01
(e) Turbulent kinetic energy (z/H = 0.55) (f) Turbulent kinetic energy (z/H = 0.75)
74
4.4. Results and discussion
75
4. Effect of gas sparging on bubble column hydrodynamics
9 holes 9 holes
0.3 49 holes 49 holes
225 holes 225 holes
484 holes 0.2 484 holes
0.2
uz [m/s]
uz [m/s]
0.1
0.1
0 0
-0.1 -0.1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/W [-] x/W [-]
(a) Vertical liquid velocity (z/H = 0.35) (b) Vertical liquid velocity (z/H = 0.55)
0.04 0.04
9 holes 9 holes
49 holes 49 holes
225 holes 225 holes
0.03 484 holes 0.03 484 holes
tke [m2/s2]
tke [m2/s2]
0.02 0.02
0.01 0.01
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/W [-] x/W [-]
(c) Turbulent kinetic energy (z/H = 0.35) (d) Turbulent kinetic energy (z/H = 0.55)
0.08 0.06
9 holes 9 holes
49 holes 49 holes
225 holes 225 holes
0.06 484 holes 484 holes
0.04
αg [-]
αg [-]
0.04
0.02
0.02
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/W [-] x/W [-]
(e) Void fraction (z/H = 0.35) (f) Void fraction (z/H = 0.55)
Figure 4.5: Comparisons of different gas spargers.
76
4.4. Results and discussion
to the sides more easily and thus, the local void fraction becomes smaller.
The effect of different location of the same sparged area on a perforated plate
has also been investigated. The locations of the sparged area on the per-
forated plates are shown in Figure 4.3. For each of the configurations, the
time-averaged vertical liquid velocity, turbulent kinetic energy and local void
fraction at two different heights are plotted in Figure 4.6(a)–4.6(f).
It is quite clear that the profiles of time-averaged vertical liquid velocity
are considerably different. Due to the deviation of the sparged area from the
center of the plate, the peak of the vertical liquid velocity distribution also
moves from the center towards the wall. Meanwhile, one can also observe
that the highest vertical liquid velocity for the plate with 9 holes is slightly
larger than that of the others. This originates from the relatively intensive
liquid circulation in the bubble column.
In Figure 4.6(c)–4.6(d), the distributions of turbulent kinetic energy for
the three different systems are plotted. On the right half of the column, the
turbulent kinetic energy decreases with increasing distance of the sparged
area from the center of the column. However, in the left half of the column,
the distributions of the turbulent kinetic energy in the columns are somewhat
more complicated. In the lower part of the column, the turbulent kinetic
energy is the highest for the W/3 case. whereas, at the intermediate height,
the three turbulent kinetic energy distributions are approaching each other.
The distributions of void fraction in the three cases are also quite different
(Figure 4.6(e)–4.6(f)). However, the trends follow those of the vertical liquid
velocity. That is, the peak of the void fraction distribution shifts from the
center to the edge with increasing sparger asymmetry. Furthermore, the void
fraction is highest when the sparger is located nearest to the wall.
It can be concluded that the location of the sparged area on the perforated
plate influences the bubble column hydrodynamics significantly. It also in-
fluences the gas phase mixing in the bubble column, which will be discussed
in the following section.
77
4. Effect of gas sparging on bubble column hydrodynamics
0.4
W/4 W/4
W/3 W/3
W/2 0.2 W/2
0.2
uz [m/s]
uz [m/s]
0.1
0
0
-0.1
-0.2 -0.2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/W [-] x/W [-]
(a) Vertical velocity (z/H = 0.35) (b) Vertical velocity (z/H = 0.55)
0.02 0.025
0.02
0.015
tke [m2/s2]
0.005
W/4 0.005 W/4
W/3 W/3
W/2 W/2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/W [-] x/W [-]
(c) Turbulent kinetic energy (z/H = 0.35) (d) Turbulent kinetic energy (z/H = 0.55)
0.1 0.05
W/4 W/4
W/3 W/3
0.08 W/2 0.04 W/2
Void fraction [-]
0.06 0.03
αg [-]
0.04 0.02
0.02 0.01
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/W [-] x/W [-]
(e) Void fraction (z/H = 0.35) (f) Void fraction (z/H = 0.55)
Figure 4.6: Comparisons of locations of gas spargers.
78
4.4. Results and discussion
The mean residence time tm and the variance σ2 of the residence time
distribution are calculated from:
Z ∞
tm = tE(t) dt (4.24)
0
Z ∞
σ2 = (t − tm )2 E(t) dt (4.25)
0
The axial dispersion model has been used extensively to characterize phase
mixing in chemical reactors. The residence time distribution can be used to
evaluate the dispersion coefficient Da in the axial dispersion model. According
to the axial dispersion model, the mass balance of a species in an unsteady-
state is given by (Levenspiel, 1999):
∂C ∂2 C ∂C
= Da 2 − U (4.26)
∂t ∂z ∂z
79
4. Effect of gas sparging on bubble column hydrodynamics
where the velocity U represents the mean velocity (for the gas phase U =
U g / g ).
In a so-called closed-closed system, a Peclet number Pe can be calculated
(Levenspiel, 1999) by using tm and σ2 determined from RTD data:
σ2 2 2
2
= − 2 (1 − e−Pe )
σ2θ = (4.27)
tm Pe Pe
where σθ is the dimensionless variance in the residence time. The dispersion
coefficient Da can now be determined from the Peclet number:
UH
Da = (4.28)
Pe
In the discrete bubble model, the determination of the residence time of the
gas phase is straightforward. Each bubble traveling in the reactor is tracked
exactly by the discrete bubble model. Therefore, the residence time of each
bubble can be determined from the DBM simulations directly and thus the
residence time distribution of the gas phase can be obtained.
The resulting residence time distributions of the columns with the four
different perforated plates are shown in Figure 4.7.
0.05
9 holes
49 holes
0.04 225 holes
484 holes
E(t) [1/s]
0.03
0.02
0.01
0
0 1 2 3 4 5 6 7
t [s]
One can find that the residence time distributions obtained from the DBM
simulations agree well with characteristics of typical residence time distribu-
80
4.4. Results and discussion
tion of the bubble phase reported by Molerus (1986). That is, (i) even the
fastest bubbles rise with a finite absolute velocity, i.e. the RTD shows a mini-
mum time; (ii) a steep slope of the RTD is observed for short residence times;
(iii) partial recirculation of small bubbles results in a long tail of the RTD.
The mean residence time, variance, Peclet number and dispersion coeffi-
cient in each column have been calculated from the residence time distribu-
tions shown in Figure 4.7 and are listed in Table 4.1.
According to the Table 4.1, the gas phase dispersion coefficient Da of
the column decreases with decreasing sparger area. This implies that the
degree of backmixing of the gas phase in the bubble column is increasing
with decreasing sparger area.
Table 4.1: Gas holdup and calculations from RTDs of the columns.
7×7
Perforated plate 3×3 15 × 15 22 × 22
W W W
4 3 2
[-] 0.016 0.014 0.015 0.016 0.020 0.024
tm [s] 1.37 1.23 1.31 1.35 1.61 1.92
σ2 [s2 ] 0.34 0.25 0.25 0.25 0.32 0.18
σ2θ [-] 0.18 0.17 0.15 0.14 0.12 0.05
Pe [-] 9.93 13.50 11.00 12.64 15.13 39.93
Da [m2 /s] 0.014 0.015 0.012 0.011 0.008 0.002
81
4. Effect of gas sparging on bubble column hydrodynamics
W/4
0.06 W/3
W/2
E(t) [1/s]
0.04
0.02
0
0 1 2 3 4 5 6 7
t [s]
becomes more intensive. That is quite reasonable because the flow field in the
column with a asymmetrical gas sparger is much more irregular than that in
a column with a symmetrical gas sparger.
4.5 Conclusions
A discrete bubble model (DBM) has been utilized to investigate the perfor-
mance of a laboratory scale bubble column with a square cross-section. Firstly,
the model including its closures has been validated by comparing the simula-
tion results with experimental data. Two eddy-viscosity models were adopted
for the comparison purpose. Computed time-averaged profiles of vertical liq-
uid velocity and turbulent kinetic energy of the liquid phase at three different
heights (z/H = 0.35, 0.55 and 0.75) were compared with PIV data. The results
show that the eddy-viscosity model proposed by Vreman (2004) performs
better than the model proposed by Smagorinsky (1963). Therefore, the for-
mer eddy-viscosity model was adopted for further investigations. Several
columns with different perforated plates were simulated to study the effect
of the gas sparger on the bubble column hydrodynamics. It was found that
the distributions of liquid velocity, turbulent kinetic energy and void fraction
82
Nomenclature
Nomenclature
B bubbles
C model coefficient, [-]; concentration of a species, [kg m−3 ]
83
4. Effect of gas sparging on bubble column hydrodynamics
Greek letters
84
References
Indices
References
85
4. Effect of gas sparging on bubble column hydrodynamics
O. Levenspiel. Chemical Reaction Engineering. John Wiley & Sons, 3rd edition,
1999.
86
References
87
4. Effect of gas sparging on bubble column hydrodynamics
88
Chapter
5
Discrete bubble modeling of
bubbly flows: Swarm effects
The performance of several drag correlations reported in literature for bubble swarms
has been investigated. A discrete bubble model (DBM) based on the Eulerian-
Lagrangian approach was adopted for this purpose. Numerical simulations for a
square bubble column were performed and the results were compared with PIV mea-
surements. The drag model reported by Lima Neto et al. (2008) predicts the vertical
liquid velocity and the relative velocity better than the other drag models at a superfi-
cial gas velocity u g = 0.0024 m/s. As the superficial gas velocity increases, however,
Lima Neto’s drag model and Wen & Yu’s model tend to overestimate the relative
velocity between the two phases. Among the other models, Rusche’s model gives a
better prediction of the vertical liquid velocity in the lower part of the bubble column.
Furthermore, most of the models can predict the liquid velocity in higher parts of the
column well. However, there are still some aspects that need to be considered and
improved to advance the accurate simulation of bubbly flows at high void fraction.
89
5. Discrete bubble modeling of bubbly flows: Swarm effects
5.1 Introduction
90
5.1. Introduction
91
5. Discrete bubble modeling of bubbly flows: Swarm effects
bubbles given by Clift et al. (1978) and proposed a new correlation for the
drag coefficient as function of the bubble Reynolds number.
In the present work, Eulerian-Lagrangian modelling of bubbly flows in a
square column is investigated with emphasis on the performance of different
closures for drag.
Our discrete bubble model (DBM) was originally developed by Delnoij et al.
(1997, 1999) and is based on volume-averaged continuity and Navier-Stokes
equations for the liquid phase, while the motion of each individual bubble is
tracked in a Lagrangian fashion taking into account bubble-bubble encoun-
ters.
dv
ρg V = ΣF (5.1)
dt
dr
=v (5.2)
dt
The net force acting on each individual bubble is calculated by considering
all the relevant forces. It is assumed that the net force is composed of separate,
uncoupled contributions due to gravity, pressure, drag, lift, virtual mass and
wall force respectively:
ΣF = FG + FP + FD + FL + FVM + FW (5.3)
FG = ρ g Vg (5.4)
92
5.2. Discrete bubble model
FP = −V∇P (5.5)
1
FD = − CD ρl πd2 |v − u|(v − u) (5.6)
8
where CD represents the drag coefficient (see section 5.3).
A bubble rising in a non-uniform liquid flow field experiences a lift force
due to vorticity or shear in the flow field. The shear induced lift force acting
on a bubble is usually written as (Auton, 1987):
where
f (EoH ) = 0.00105EoH 3 − 0.0159EoH 2 + 0.474 (5.9)
EoH is the Eötvös number defined by using the maximum horizontal dimen-
sion of a bubble:
(ρl − ρ g )gd2H
EoH = (5.10)
σ
The maximum horizontal diameter of the bubble is obtained from the
bubble aspect ratio E according to Wellek et al. (1966):
dV 1
E= = (5.11)
dH 1 + 0.163Eo0.757
where dV is the maximum vertical diameter of the bubble and Eo is the Eötvös
(ρl − ρ g )gd2
number, Eo = .
σ
93
5. Discrete bubble modeling of bubbly flows: Swarm effects
The relation between the above two diameters and the diameter of the
bubble d in the discrete bubble modeling is as follows:
94
5.2. Discrete bubble model
The liquid phase is assumed to be Newtonian, thus the stress tensor τl can be
expressed as:
2
τl = −µeff,l ((∇u) + (∇u)T − I(∇ · u) (5.18)
3
where µeff,l is the effective viscosity. In the present model, the effective viscos-
ity is composed of two contributions, the molecular viscosity and the turbulent
viscosity:
where Bβ = β11 β22 − β212 + β11 β33 − β213 + β22 β33 − β223 , βij = ∆2m αmi αm j and
αij = ∂u j /∂xi . ∆i is the filter width in the i direction.
In the present paper, a hard sphere collision model (Hoomans et al., 1996)
is adopted to describe the (possible) bouncing of bubbles. It consists of two
main parts. One part is processing the sequence of collisions and another part
is dealing with the collision dynamics. The former is described in detail by
Darmana et al. (2006). More details are given in Chapter 4 of this thesis.
95
5. Discrete bubble modeling of bubbly flows: Swarm effects
Although the correlation of Wen and Yu (1966) was proposed for gas-solid sys-
tems, we used this model here for the purpose of comparison. The associated
expression for the drag coefficient is give by:
24 −1.65
Re (1 + 0.15Res )(1 − α g ) Res < 1000
0.687
CD =
s (5.21)
0.44(1 − α g )−1.65 Res ≥ 1000
where the bubble Reynolds number Res is based on the superficial velocity:
(1 − α g )ρl |v − u|d
Res = (5.22)
µl
According to Ishii and Zuber (1979) and Ishii and Hibiki (2005), for the drag
correlation of bubbles four flow regimes should be distinguished: the viscous
regime, the distorted flow regime, the churn-turbulent flow regime and the
slug regime.
The drag coefficient correlations are given as follows:
24
(1 + 0.1Re0.75
s ) Viscous regime
Re
2 √ " 1 + 17.67(1 − α g )1.3 #2
s
Eo Distorted regime
CD =
3 18.67(1 − α g )1.5 (5.23)
8
3 (1 − α g )
2
Churn-turbulent regime
9.8(1 − α g )3 Slug regime
96
5.3. Drag coefficient correlations
f (α g ) = exp(3.64α g ) + α0.864
g (5.27)
97
5. Discrete bubble modeling of bubbly flows: Swarm effects
where CD∞ is the drag coefficient of single bubbles in infinite liquids, which
is deduced from Jamialahmadi et al. (1994)’s correlation:
V1 V2
V∞ = q (5.29)
V12 + V22
where
1 ∆ρ 2 3µl + 3µb
V1 = gd (5.30)
18 µl 2µl + 3µb
and s
2σ gd
V2 = + (5.31)
d(ρl + ρ g ) 2
The drag coefficient for a single bubble in infinite liquids CD∞ is then
determined as follows:
4 (ρl − ρ g )gd
CD∞ = (5.32)
3 V∞ 2
ρl
Due to the fact that Simonnet’s model only provides the drag coefficient
correlation in the range of the local void fraction between 0 and 0.3, we use
the drag coefficient at 0.3 when the local void fraction exceeds 0.3.
98
5.3. Drag coefficient correlations
0.2 m/s. And thus, the drag coefficient corresponding to each model can be
calculated as function of the local void fraction, which is shown in Figure 5.1.
Note that the drag coefficients from those models independent of the local
void fraction, such as Tomiyama’s model and Lima Neto’s model, are treated
as constants in the plot. Moreover, for those models without explicit specifi-
cation of the correlation for single bubble, i.e. Holland’s model and Rusche’s
model, the drag coefficient CD∞ is calculated according to the correlation of
Tomiyama et al. (1998).
10
Wen & Yu (1966)
Ishii & Zuber (1979)
8 Holland (1995)
Tomiyama (1998)
6 Rusche (2002)
CD [-]
Simonnet (2007)
Lima Neto (2008)
4
0
0 0.2 0.4 0.6 0.8
αg [-]
From Figure 5.1, one can see that the drag coefficient obtained from Lima
Neto’s model is very small compared to the others and about 8.5 times smaller
than that from Tomiyama’s single bubble model. Based on Wen & Yu’s model,
the drag coefficient increases monotonically and the slope becomes larger with
increasing local void fraction. Furthermore, the drag coefficient obtained
from Ishii & Zuber’s model first increases slightly up to α g = 0.3 and then
decreases. The drag coefficient increases considerably with α g fraction in
Holland’s and Rusche’s models. On the contrary, the drag coefficient obtained
from Simonnet’s model first slightly increases with α g . Subsequently the drag
coefficient decreases rapidly beyond the local void fraction of 0.15. When
99
5. Discrete bubble modeling of bubbly flows: Swarm effects
the local void fraction approaches 0.3, the drag coefficient obtained from
Simonnet’s model approaches the value from Lima Neto’s model.
The bubble column studied here is shown in Figure 5.2. The cross-sectional
area of the column is 0.15 m × 0.15 m (W × D). The column is initially filled
with water to a height of 0.45 m (H). Air is used as the dispersed phase and
introduced into the column through a perforated plate with 49 holes (7 × 7) at
the bottom of the column. The material properties of both phase are taken at
room temperature (20 °C) and atmospheric pressure.
A computational grid with (20×20×60) cells is adopted in the simulations.
Four superficial gas velocities, i.e. u g = 0.0024 m/s, 0.0049 m/s, 0.0073 m/s and
0.0097 m/s are used. The bubble diameter in the simulations is assumed to
be a constant, 0.005 m. Note that this assumption may not be correct since
the bubble size may vary in the applied range of superficial gas velocities due
to coalescence and breakup of bubbles. For the purpose of comparison of
different drag coefficient correlations for bubbles rising in a swarm, however,
a uniform bubble size is used here. Moreover, the time step in the simulation
for the liquid phase is set as 1.0 × 10−3 s and the collisions among bubbles and
the movements of the bubbles are processed multiple times within each time
step. The total simulation time is 150 s.
100
5.5. Results and discussion
are continuously injected into the bubble column and start rising in the col-
umn. At the moment when the peak maximum is reached, some bubbles
have already reached the level of the liquid surface and most of them escape
from the column. Hence, there is a rapid decrease in the curve right after the
peak. Once the circulation flow pattern has developed, some bubbles at the
top of the column may be trapped by the circulating liquid and remain in the
column. Some of these bubbles may leave the column soon, whereas others
may take a little more time to escape. These effects produce the fluctuation
of the gas holdup in the curve. In addition, the instantaneous vertical liquid
101
5. Discrete bubble modeling of bubbly flows: Swarm effects
0.01 0.8
0.6
0.008
0.4
uz [m/s]
0.006
ε [-]
0.2
0.004
0
0.002
-0.2
0 -0.4
0 50 100 150 0 50 100 150
t [s] t [s]
(a) Instantaneous gas holdup [-] (b) Instantaneous liquid velocity [m/s]
velocity at position (x/W = 0.5, y/D = 0.5, z/H = 0.96) is also plotted versus
time (Figure 5.3(b)). The obtained trend is similar to that obtained for the gas
holdup. However, unlike the gas holdup, the liquid velocity shows more fluc-
tuations at the beginning of the simulation. The averaging of hydrodynamic
quantities, i.e. velocities of both phases, is initiated 30 seconds after the start
of the simulation.
u g = 0.0024 m/s
Profiles of the average vertical liquid velocity and relative velocity between
the gas phase and the liquid phase at three heights are shown in Figure 5.4
and compared with the PIV measurements at corresponding heights. The
102
5.5. Results and discussion
results reveal that Lima Neto’s drag correlation predicts the vertical liquid
velocity better than all other correlations in the central region of the column.
However, the model underestimates the vertical liquid velocity in the region
between the center and the wall, particularly in the right part of the column
(x/W > 0.5). Among the other five drag coefficient correlations, Wen & Yu’s
model exhibits better performance for the prediction of the vertical liquid
velocity compared to the other models. Simonnet’s model has the largest
deviation in the central region in lower parts of the column (z/H ≤ 0.5). Ishii
& Zuber’s correlation and Tomiyama’s single bubble correlation show similar
performance regarding the prediction of the vertical liquid velocity in the
central region. Rusche’s model combined with Tomiyama’s single bubble
correlation shows moderate performance of all the six models. The similar
behavior is also found with Holland’s model combined with Tomiyama’s
single bubble correlation.
In addition, by comparing the relative velocity between the two phases
obtained from both the PIV measurements and the simulations, one can find
that the relative velocities between the gas phase and the liquid phase ob-
tained with Wen & Yu’s model and Lima Neto’s model for the drag coefficient
correlations are higher than those with the other four correlations. This is
due to the fact that the drag forces calculated from these two correlations are
smaller than the others. Furthermore, the other four correlations produce
similar profiles along the x direction at the three heights. In the lower parts
of the column, i.e. z/H = 0.3, Lima Neto’s model overestimates the relative
velocity in the central region. Along the height, however, this model can
predict the relative velocity quite well. This also holds for Wen & Yu’s model.
u g = 0.0049 m/s
In Figure 5.5, the simulation results and PIV measurements at u g = 0.0049 m/s
are compared with each other. It can be seen that all the models again overesti-
mate the vertical liquid velocity in the central region at low part of the column
(z/H ≤ 0.5). Ishii & Zuber’s model and Tomiyama’s model for single bubbles
exhibit similar trends and produce the largest deviation from the PIV data in
103
5. Discrete bubble modeling of bubbly flows: Swarm effects
(a) Vertical liquid velocity (z/H = 0.3) (b) Relative velocity (z/H = 0.3)
(c) Vertical liquid velocity (z/H = 0.5) (d) Relative velocity (z/H = 0.5)
(e) Vertical liquid velocity (z/H = 0.7) (f) Relative velocity (z/H = 0.7)
104
5.5. Results and discussion
that region. Moreover, the other four models yield better predictions with
respect to the vertical liquid velocity. At increased heights, i.e. z/H = 0.7,
one can find that Wen & Yu’s model and Simonnet’s model perform well.
Furthermore, Rusche’s model also predicts the vertical liquid velocity quite
well in the central region. However, the profile of the vertical liquid velocity
according to Rusche’s drag model seems to underestimate the velocity in the
left part of the column (x/W < 0.5) and slightly overestimates the velocity
in the right part (x/W > 0.5). On the contrary, Holland’s model seems can
predict the vertical liquid velocity in the left part of the column well, while
the model underestimates the liquid velocity in the right part. In addition, it
is worth to mention that Simonnet’s drag model can predict the vertical liquid
velocity away from the central region very well in the entire bubble column.
Comparisons for the relative velocity are presented in Figure 5.5(b), Fig-
ure 5.5(d) and Figure 5.5(f). It is quite clear that Lima Neto’s drag model
overestimates the relative velocity at all three heights. Wen & Yu’s model can
predict the relative velocity well in the central region of the lower part of the
column (z/H ≤ 0.5) compared with the other models. However, it overesti-
mates the relative velocity at the height z/H = 0.7. Furthermore, Simonnet’s
drag model, Ishii & Zuber’s model as well as Tomiyama’s model are able to
predict the relative velocity quite well. Finally Rusche’s drag model underes-
timates the relative velocity between the two phases slightly, while Holland’s
model has better estimation.
u g = 0.0073 m/s
When the superficial gas velocity increases up to 0.0073 m/s, it can be seen
that Tomiyama’s model for isolated bubbles produces large deviations for the
vertical liquid velocities in the central region of the column. However, the
model predicts the vertical liquid velocity well in the upper part of the column.
It is also clear that all the other models again overestimate the vertical liquid
velocity in the central region in the lower part of the column (z/H ≤ 0.5). The
models due to Ishii & Zuber, Holland, Rusche and Lima Neto yield better
prediction with respect to the vertical liquid velocities. In the upper part of
the column (z/H = 0.7), most of the models can predict the vertical liquid
105
5. Discrete bubble modeling of bubbly flows: Swarm effects
(a) Vertical liquid velocity (z/H = 0.3) (b) Relative velocity (z/H = 0.3)
(c) Vertical liquid velocity (z/H = 0.5) (d) Relative velocity (z/H = 0.5)
(e) Vertical liquid velocity (z/H = 0.7) (f) Relative velocity (z/H = 0.7)
106
5.6. Conclusions
velocity quite well except that the models of Wen & Yu, Holland, Simonnet
and Lima Neto underestimate the vertical liquid velocity in either the left or
the right part of the column.
By comparing the profiles of the relative velocity from numerical sim-
ulations with PIV measurements, it can been seen that Wen & Yu’s model
and Lima Neto’s model clearly increasingly overestimate the relative velocity
between gas phase and liquid phase along the column height. In addition,
the other models predict the relative velocity quite well in the central region
compared to the wall region. However, one can find that these models agree
with PIV measurements at height z/H = 0.7 very nicely.
u g = 0.0097 m/s
At a superficial gas velocity u g = 0.0097 m/s, one can find that the vertical
liquid velocity predicted with all the drag models is much larger than the
PIV measurements in the central region in the lower part of the column. Wen
& Yu’s model are much closer to the PIV data than the others. Moreover,
Rusche’s model and Lima Neto’s model show moderate performance. At the
height z/H = 0.7, however, Wen & Yu’s model apparently underestimates the
vertical liquid velocity in the central region of the column. In addition the
results from Holland’s model and Rusche’s model also show slightly different
profiles compared with the PIV data. The others agree with the data quite
well.
When looking at the comparisons of the relative velocity between the two
phases, one can see that Wen & Yu’s model and Lima Neto’s drag model
overestimate the relative velocity at all three heights in the bubble column.
The other models show similar trends for the relative velocity along the x
direction, particularly at the height z/H = 0.7. However, the profile of the
relative velocity in the upper part of the column is reproduced.
5.6 Conclusions
107
5. Discrete bubble modeling of bubbly flows: Swarm effects
(a) Vertical liquid velocity (z/H = 0.3) (b) Relative velocity (z/H = 0.3)
(c) Vertical liquid velocity (z/H = 0.5) (d) Relative velocity (z/H = 0.5)
(e) Vertical liquid velocity (z/H = 0.7) (f) Relative velocity (z/H = 0.7)
108
5.6. Conclusions
(a) Vertical liquid velocity (z/H = 0.3) (b) Relative velocity (z/H = 0.3)
(c) Vertical liquid velocity (z/H = 0.5) (d) Relative velocity (z/H = 0.5)
(e) Vertical liquid velocity (z/H = 0.7) (f) Relative velocity (z/H = 0.7)
109
5. Discrete bubble modeling of bubbly flows: Swarm effects
110
Nomenclature
in bubbly flows as well need to be considered, i.e. the effect of bubble swarms
on the lift force. Meanwhile, coalescence and breakup of bubbles should also
be taken into account in the modelling in order to simulate the bubbly flows
reasonably at high void fraction flows.
Nomenclature
Greek letters
Indices
b bubble
D drag
eff effective
g gas phase
G gravity
H horizontal direction
i i direction
j j direction
l liquid phase
L lift; molecular viscosity
m mixture
P pressure
s superficial velocity
T turbulence
V vertical direction
VM virtual mass
W wall
∞ infinite medium
112
References
Acknowledgement
The authors would like to gratefully acknowledge Dr. Lima Neto for sharing
his experimental results.
References
R. Clift, J. R. Grace, and M. E. Weber. Bubbles, Drops, and Particles. New York
[etc.] ; London : Academic Press, 1978.
113
5. Discrete bubble modeling of bubbly flows: Swarm effects
F. A. Holland and R. Bragg. Fluid Flow for Chemical Engineers. Elsevier Ltd.,
second edition, 1995.
M. Ishii and N. Zuber. Drag coefficient and relative velocity in bubbly, droplet
or particulate flows. AIChE Journal, 25(5):843–855, 1979.
114
References
115
5. Discrete bubble modeling of bubbly flows: Swarm effects
116
Chapter
6
Discrete bubble modeling of
bubbly flows: Implementation of
breakup models
Breakup models developed in literature were implemented in our model, which is
based on an Eulerian-Lagrangian approach. Moreover, the critical Weber number for
bubble breakup utilized by many authors in turbulent flows was also incorporated in
the model. Only binary breakage is considered in this work. For the models utilizing
the critical Weber number, two different daughter size distributions, namely bell shape
and a U shape were used.
First the implementation was verified by simulating bubbly flows in a square bubble
column. The simulated breakup frequency and daughter size distribution were com-
pared with those obtained from the models. Subsequently, the simulation results were
compared with detailed PIV measurements. Finally, the predicted bubble size distri-
butions were compared with chord length distributions obtained from measurements
with a four-point optical fibre probe.
117
6. Implementation of breakup models
6.1 Introduction
In two-phase gas-liquid flows, the properties of the dispersed phase are very
important (i.e. interface topology). The complex interface topology and its
dynamics (due to coalescence and breakup) poses a considerable difficulty in
the study of bubbly flows. Hence, mechanisms of coalescence and breakup
are of special significance for the investigation of two-phase gas-liquid flows.
The deformation and breakup of a bubble can be described with a sim-
ple static force balance, which introduces the ratio between the force that
causes the deformation and the surface tension that tends to counteract the
deformation (Hinze, 1955). Depending on the type of flow, the cause of the
deformation varies. No matter what the nature of the deformation is, how-
ever, the breakup occurs when the ratio exceeds a critical value. The ratio
is expressed as a critical Weber number. Sevik and Park (1973) studied the
critical Weber number for air bubbles in a high Reynolds number water jet
(Kolev, 2007). Walter and Blanch (1986) proposed an expression for the maxi-
mum stable bubble size in solutions and conducted a comprehensive study on
the effect of the presence of surfactants on bubble breakup. Risso and Fabre
(1998) analyzed the breakup mechanism of a bubble in turbulent flows under
microgravity conditions and estimated the critical Weber number. Qian et al.
(2006) studied the critical Weber number of bubbles in homogeneous turbu-
lence under zero gravity conditions using the lattice Boltzmann method. A
detailed review of the mechanisms of deformation and breakup of drops and
bubbles has been given by Risso (2000).
In recent years, CFD has emerged as a powerful tool to study multiphase
flow phenomena and assess their impact on the performance of multiphase
chemical reactors (Jakobsen, 2008). In particular population balance mod-
elling has received considerable attention in the past decade to account for
the size distribution of the dispersed phase. In order to close the population
balance equations, i.e. through expressions for the birth rate and the death
rate due to breakup of bubbles, many efforts on the mathematical formulation
of coalescence and breakup models in turbulent flows have been made since
then. Based on the turbulent nature of liquid-liquid dispersion, a phenomeno-
118
6.2. Discrete bubble model
Our discrete bubble model (DBM) was originally developed by Delnoij et al.
(1997, 1999). The liquid phase is described by volume-averaged continuity
119
6. Implementation of breakup models
dr
=v (6.3)
dt
The net force acting on each individual bubble is calculated by considering
all the relevant forces. It is assumed that the net force is composed of separate,
uncoupled contributions due to gravity, pressure, drag, lift, virtual mass and
wall force respectively:
ΣF = FG + FP + FD + FL + FVM + FW (6.4)
FG = ρ g Vg (6.5)
FP = −V∇P (6.6)
CD = CD∞ f (α g ) (6.8)
120
6.2. Discrete bubble model
where f (α g ) represents the effect arising from the presence of other bubbles:
f (α g ) = exp(3.64α g ) + α0.864
g (6.9)
where
f (EoH ) = 0.00105EoH 3 − 0.0159EoH 2 + 0.474 (6.13)
EoH is the Eötvös number defined by using the maximum horizontal dimen-
sion of a bubble:
(ρl − ρ g )gd2H
EoH = (6.14)
σ
The maximum horizontal diameter of the bubble is obtained from the
bubble aspect ratio E according to Wellek et al. (1966):
dV 1
E= = (6.15)
dH 1 + 0.163Eo0.757
where dV is the maximum vertical diameter of the bubble and Eo is the Eötvös
(ρl − ρ g )gd2
number, Eo = .
σ
121
6. Implementation of breakup models
The relation between the above two diameters and the diameter of the
bubble d in the discrete bubble modeling is as follows:
where nz and nW , respectively, are the normal unit vectors in the vertical and
wall normal direction, L is the dimension of the system in the wall normal
direction, and y is the distance to the wall in that direction. Finally, the wall
force coefficient CW is given by:
exp(−0.933Eo + 0.179) 1 ≤ Eo ≤ 5,
CW =
(6.19)
0.007Eo + 0.04 5 < Eo ≤ 33.
∂
αl ρl + ∇ · αl ρl u = 0 (6.20)
∂t
∂
αl ρl u + ∇ · αl ρl uu = −αl ∇p − ∇ · (αl τl ) + αl ρl g + Φ (6.21)
∂t
122
6.3. Coalescence model
The liquid phase is assumed to be Newtonian, thus the stress tensor τl can be
expressed as:
2
τl = −µeff,l ((∇u) + (∇u)T − I(∇ · u) (6.22)
3
where µeff,l is the effective viscosity. In the present model, the effective viscos-
ity is composed of two contributions, the molecular viscosity and the turbulent
viscosity:
µeff,l = µL,l + µT,l (6.23)
where µT,l is the turbulent viscosity (or eddy viscosity), which is determined
from turbulence modeling of the liquid phase.
123
6. Implementation of breakup models
The contact time is dependent on the bubble size and the turbulent intensity.
An estimate of the contact time in turbulent flows is given as:
R2/3
ij
τij = (6.26)
1/3
where is the turbulent energy dissipation rate.
Sommerfeld et al. (2003) provided an estimate of the contact time by as-
suming that it is proportional to a deformation distance divided by the normal
component of the relative velocity:
CC Rij
τij = (6.27)
|vni − vnj |
where CC is a model constant.
The properties of the new bubble in case of coalescence are obtained from
conservation of mass and momentum.
124
6.4. Breakup models & implementation
25 2e+04
ε=0.5 m2/s3 ε=0.5 m2/s3
ε=1.0 m2/s3 ε=1.0 m2/s3
20 ε=2.0 m2/s3 ε=2.0 m2/s3
1.5e+04
Ω(d) [1/s]
Ω(d) [1/s]
15
1e+04
10
5,000
5
0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
d [m] d [m]
√
d5/3 19/15 ρ7/5 2σ9/5
Ω(d) = 0.5 l
exp − (6.29)
σ7/5 d3 ρ9/5 6/5 l
125
6. Implementation of breakup models
By assuming that a bubble only breaks into two bubbles, Martı́nez-Bazán et al.
(1999b) postulated that the probability of the formation of a pair of bubbles
of sizes d1 and d2 from a mother bubble of size d, P(d∗ ), is weighted by the
product of two surplus stresses which are associated with the formation of the
two daughter bubbles. Hence, the daughter size distribution can be written
as:
P(d∗ )
β(d∗ ) = R 1 (6.30)
P(d∗ ) dd∗
0
and
!" #
∗ ∗2/3 12σ ∗3 2/9 12σ
P(d ) = d − (1 − d ) − (6.31)
8.2ρl d5/3 2/3 8.2ρl d5/3 2/3
where V ∗ = V1 /V.
As shown in Figure 6.2(b), unequal-size breakage is more likely according
to the above daughter size distribution. And the probability of unequal-sized
breakage increases rapidly with the mother bubble size.
126
6.4. Breakup models & implementation
3 25
d=0.005 m d=0.005 m
d=0.007 m d=0.007 m
d=0.01 m 20 d=0.01 m
2
15
β(V*)
β(V*)
10
1
5
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
*
V [-] V* [-]
127
6. Implementation of breakup models
7.8. They also obtained a value of the critical Weber number close to 4.5 by
analyzing how the mean deformation is related to the mean turbulence in-
tensity. Hence, they suggested that a minimal Weber number of about 5 is
necessary for breakup when the turbulence is the only cause of deformation.
Qian et al. (2006) performed lattice Boltzmann simulations for the deforma-
tion and breakup of bubbles in homogeneous turbulence under zero gravity
conditions. The minimum Weber number for bubble breakup was found to
be about 3.0.
The beta distribution can be used to describe the daughter bubble size
when a critical Weber number is involved to determine bubble breakup in the
turbulent flows. The beta distribution is formulated as follows:
Γ(a + b) ∗a−1
f (V ∗ ) = V (1 − V ∗ )b−1 (6.35)
Γ(a)Γ(b)
2.5 a=b=0.5
a=b=2.0
2
1.5
f(V*)
0.5
0
0 0.2 0.4 0.6 0.8 1
V* [-]
128
6.5. Simulation details
6.4.3 Implementation
The above breakup models provide quantitative information on dynamics
of bubble breakage in turbulent flows. For instance, the breakup frequency
is related to the time interval in which the bubble breakage occurs. Hence,
one may find a way to integrate the bubble breakage into the discrete bubble
model (DBM) by taking the breakup time into account.
In the discrete bubble model, each bubble is tracked individually in the
turbulent flow field. The bubble properties, such as bubble velocity, bubble
diameter and bubble position in the flow, and local flow properties, i.e. dis-
sipation rate of turbulence kinetic energy, are known. Hence, it is possible
to determine the breakup time for each bubble at its position in the flow. If
the breakup time of a bubble is smaller than the computational time step, the
bubble breaks immediately within that time step. On the other hand, if the
calculated breakup time of the bubble exceeds the computational time step,
the program starts counting the time since that instant. The computation
moves to the next time step and once the counted time exceeds the breakup
time, the bubble breaks. In this approach, the history of turbulent eddies
colliding with the bubble is considered.
The resulting daughter bubble sizes are determined according to the above
introduced daughter size distributions. The velocities of the two daughter
bubbles are assumed to be the same as that for the mother bubble. One of the
two daughter bubbles is located at the same position as the mother bubble
while the other daughter bubble is located around that daughter bubble ran-
domly. After breakage, each bubble in the turbulent flow field is then tracked
by considering the breakage criterion again.
The bubble column studied here is shown in Figure 6.4. The cross-sectional
area of the column is 0.15 m × 0.15 m (W × D). The column is initially filled
129
6. Implementation of breakup models
with water to a height of 0.45 m (H). Air is used as the dispersed phase and
introduced into the column through a perforated plate with 49 holes (7 × 7) at
the bottom of the column. The material properties of both phases are taken at
room temperature and atmospheric pressure.
130
6.6. Results and discussion
liquid phase is set at 1.0 × 10−3 s and the collisions among bubbles and the
movements of the bubbles are processed multiple times within each time
step. The total simulation time is 150 s.
The coalescence model and two breakup models described above are used
in the simulations. Moreover, two critical Weber numbers, i.e. We = 2.48 and
We = 5.0, are adopted for the purpose of comparison.
The sub-grid scale model is required to represent the turbulent flow in-
side the bubble column. This means that the continuity and Navier-Stokes
equations are resolved for the “large” eddies, while the ”small scales” of the
resolved filed are included through an eddy-viscosity subgrid-scale model.
The eddy-viscosity model proposed by Vreman (2004) was used to calcu-
late the eddy viscosity:
s
Bβ
µT,l = 2.5ρl C2S (6.36)
αij αij
where Bβ = β11 β22 − β212 + β11 β33 − β213 + β22 β33 − β223 , βij = ∆2m αmi αm j and
αij = ∂u j /∂xi . ∆i is the filter width in the i direction.
In the large eddy simulations, however, the dissipation rate of turbulent
kinetic energy cannot be obtained directly. Moreover, the dissipation rate only
from the resolved scales is negligible compared with that from the subgrid
scales. By assuming a local equilibrium between turbulent kinetic energy
production and dissipation, the dissipation rate can be estimated as (Jiménez
et al., 2001; Hartmann et al., 2004; Delafosse et al., 2009):
!2
(µL,l + µT,l ) ∂ui ∂u j
= + (6.37)
2ρl ∂x j ∂xi
131
6. Implementation of breakup models
40 600
Model (ε=0.5 m2/s3) Model (ε=0.5 m2/s3)
Model (ε=1.0 m2/s3) Model (ε=1.0 m2/s3)
Model (ε=2.0 m2/s3) Model (ε=2.0 m2/s3)
30
DBM (ε=0.5 m2/s3) DBM (ε=0.5 m2/s3)
400
Ω(d) [1/s]
Ω(d) [1/s]
DBM (ε=1.0 m2/s3) DBM (ε=1.0 m2/s3)
DBM (ε=2.0 m2/s3) DBM (ε=2.0 m2/s3)
20
200
10
0 0
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015
d [m] d [m]
132
6.6. Results and discussion
25
3 Model (d=0.005 m) DBM (d=0.005 m) Model (d=0.005 m)
Model (d=0.007 m) DBM (d=0.007 m) Model (d=0.007 m)
Model (d=0.01 m) DBM (d=0.01 m) 20 Model (d=0.01 m)
DBM (d=0.005 m)
2 DBM (d=0.007 m)
15
β(V*)
β(V*)
DBM (d=0.01 m)
10
1
5
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
*
V [-] V* [-]
0.06
0.1
β(V*)
β(V*)
0.04
0.02
0.05
0
0 0.2 0.4 0.6 0.8 1 0 1
V* [-] V* [-]
liquid velocity and relative velocity between the gas phase and the liquid
phase, will be considered. Comparisons are carried out at three heights in
the column (z/H = 0.3, 0.5 and 0.7). Moreover, simulations assuming uni-
form bubble size with and without considering swarm effects are included for
the purpose of reference. In the following figures, the symbols representing
different conditions of simulations are kept the same.
u g = 0.0049 m/s
The average vertical liquid velocity and the local void fraction obtained from
the simulations with superficial gas velocity u g = 0.0049 m/s are plotted in
133
6. Implementation of breakup models
Figure 6.7. It can be seen that the value of the critical Weber number influences
the vertical liquid velocity considerably, particularly in the central region of
the bubble column. Assuming the same daughter size distribution, i.e. the
bell-shaped distribution, the vertical liquid velocities predicted with We = 5.0
are higher than those predicted with We = 2.48. Furthermore, assuming the
same critical Weber number, i.e. We = 2.48, but with different daughter size
distributions, i.e. bell shape and U shape, the difference between the predicted
and experimental vertical liquid velocities in the central region is getting larger
along column height. Martı́nez-Bazán’s breakup model produces similar per-
formance regarding the prediction of the vertical liquid velocity assuming the
critical Weber number We = 2.48 and a bell-shaped daughter size distribu-
tion. Lehr’s breakup model produces a relatively large deviation. For the
case of uniform bubble size with a single bubble drag closure also a large
discrepancy with the experiments is obtained. However, it can be seen that at
the relatively low superficial gas velocity, i.e. u g = 0.0049 m/s, application of
a uniform bubble size considering the swarm effects on the drag force gives
better predictions of the vertical liquid velocity in the central region of the
column.
Comparison of local void fraction reveals that the local void fractions
predicted with the critical Weber number, We = 5.0, in the center is larger
than those predicted with other models. The predicted local void fractions
assuming an uniform bubble size considering the swarm effect and critical
Weber number, We = 2.48, with bell-shaped daughter size distribution are
close to each other.
u g = 0.0073 m/s
In Figure 6.8, profiles of the vertical liquid velocity and the local void fraction
at three heights in the bubble column are compared. The simulation with a
critical Weber number, We = 5.0, again shows the largest discrepancy with re-
spect to the prediction of the vertical liquid velocity in the center. In the central
region, Martı́nez-Bazán’s model and Lehr’s model have similar performance
compared to the case with the uniform size and single bubble drag closure in
the lower part of the column (z/H ≤ 0.5). Moreover, the models with a critical
134
6.6. Results and discussion
(a) Vertical liquid velocity (z/H = 0.3) (b) Local void fraction (z/H = 0.3)
(c) Vertical liquid velocity (z/H = 0.5) (d) Local void fraction (z/H = 0.5)
(e) Vertical liquid velocity (z/H = 0.7) (f) Local void fraction (z/H = 0.7)
135
6. Implementation of breakup models
Weber number, We = 2.48, predict better the liquid velocity in the lower part
of the column. However, they underestimate the liquid velocity in the higher
parts of the column. The model with uniform bubble size considering the
swarm effect predicts the liquid velocity quite well in the bubble column.
Profiles of the local void fraction show that the model with the critical
Weber number, We = 5.0, gives the highest void fraction in the central region
of the column. Furthermore, the local void fractions obtained from the models
with the critical Weber number, We = 2.48, are close to those predicted with
the model with uniform bubble size considering the swarm effect in the lower
part of the column (z/H ≤ 0.5). However, they start to deviate from each other
along the column height.
u g = 0.0097 m/s
At large superficial gas velocity, u g = 0.0097 m/s, profiles of the vertical liquid
velocity and the local void fraction obtained from the above models (Fig-
ure 6.9) show some differences compared with low superficial gas velocity.
It can be seen that the model with the critical Weber number, We = 5.0, can
predict the vertical liquid velocity quite well in the central region of the bubble
column. On the contrary, the model with critical Weber number, We = 2.48
and bell-shaped daughter size distribution and the model with uniform bub-
ble size considering swarm effects produce large deviations in the lower part
of the column. However, the former predicts the profile of the vertical liquid
velocity in the higher part of the column (i.e. z/H = 0.7) well. The model with
uniform bubble size and single bubble drag closure has the largest deviation
however in the higher part of the column this model performs quite well.
Martı́nez-Bazán’s model, Lehr’s model and the model with the critical Weber
number, We = 2.48 and U-shaped daughter size distribution show similar
performance in predicting the liquid velocity in the lower part of the bubble
column (z/H ≤ 0.5). With increasing column height, however, these models
show a worse performance.
In addition, one can see that the local void fractions obtained from the
model with uniform bubble size are slightly higher than those obtained from
the other models in the central region. In the lower part of the column, the
136
6.6. Results and discussion
(a) Vertical liquid velocity (z/H = 0.3) (b) Local void fraction (z/H = 0.3)
(c) Vertical liquid velocity (z/H = 0.5) (d) Local void fraction (z/H = 0.5)
(e) Vertical liquid velocity (z/H = 0.7) (f) Local void fraction (z/H = 0.7)
137
6. Implementation of breakup models
model with uniform bubble size and swarm effects predicts higher values of
the local void fraction than the others in the central region. The model that
uses the critical Weber number We = 2.48 and the U-shaped daughter size
distribution has similar performance as Martı́nez-Bazán’s model. Along the
column height, differences between the models are getting more pronounced.
(a) Vertical liquid velocity (z/H = 0.3) (b) Local void fraction (z/H = 0.3)
(c) Vertical liquid velocity (z/H = 0.5) (d) Local void fraction (z/H = 0.5)
(e) Vertical liquid velocity (z/H = 0.7) (f) Local void fraction (z/H = 0.7)
138
6.6. Results and discussion
139
6. Implementation of breakup models
0.5 0.4
Chord length Chord length
Bubble diameter Bubble diameter
0.4
0.3
Distribution [-]
Distribution [-]
0.3
0.2
0.2
0.1
0.1
0 0
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02
Dimension [m] Dimension [m]
(a) z/H = 0.55, x/W = 0.5 (b) z/H = 0.75, x/W = 0.5
Figure 6.10: Distributions of the chord length and the bubble size.
model. At the central part of the column, the bubble size distribution pre-
dicted by Martı́nez-Bazán’s model has higher probability than others at about
d = 0.007 m. At the higher part of the column, the distributions obtained
from the numerical simulations are slightly shifted to the larger bubble size
compared with that determined from the optical probe. That suggests that
the implemented coalescence and breakup models into the discrete bubble
model are able to predict the bubble size distribution reasonably well. Fur-
thermore, the model that uses a U-shaped daughter size distribution has a
high probability to obtain some very small bubbles, which suggests that the
implementation of the U-shaped daughter size distribution is successful.
6.7 Conclusions
In this work, coalescence and breakup models have been combined with a
detailed Euler-Lagrange (EL) model incorporating four-way coupling. The
implementation of the breakup models in the EL model is based on the con-
cept of breakup frequency (breakup time). By considering the breakup time
for bubbles in turbulent flows and the history of bubble and turbulent eddies
encounter, bubble breakup is incorporated into the discrete bubble model.
Only binary breakup is considered in the breakup models. The daughter size
140
6.7. Conclusions
0.5 0.5
We=2.48, bell shape Martinez-Bazan, 1999
We=2.48, U shape Lehr, 2002
Bubble size distribution
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
0.000 0.005 0.010 0.015 0.020 0.025 0.000 0.005 0.010 0.015 0.020 0.025
d [m] d [m]
(a) z/H = 0.55, x/W = 0.5 (b) z/H = 0.55, x/W = 0.5
0.4 0.4
We=2.48, bell shape Martinez-Bazan, 1999
We=2.48, U shape Lehr, 2002
Bubble size distribution
0.2 0.2
0.1 0.1
0.0 0.0
0.000 0.005 0.010 0.015 0.020 0.000 0.005 0.010 0.015 0.020
d [m] d [m]
(c) z/H = 0.75, x/W = 0.5 (d) z/H = 0.75, x/W = 0.5
Figure 6.11: Bubble size distribution.
141
6. Implementation of breakup models
Nomenclature
142
Nomenclature
Greek letters
Indices
i bubble index
j bubble index
l liquid phase
L lift; molecular viscosity
m mixture
P pressure
T turbulence
VM virtual mass
W wall
∞ infinite medium
References
144
References
145
6. Implementation of breakup models
F. Lehr and D. Mewes. A transport equation for the interfacial area density
applied to bubble columns. Chemical Engineering Science, 56(3):1159–1166,
2001.
Y. Liao and D. Lucas. A literature review of theoretical models for drop and
bubble breakup in turbulent dispersions. Chemical Engineering Science, 64
(15):3389–3406, 2009.
H. Luo and H. F. Svendsen. Theoretical model for drop and bubble breakup
in turbulent dispersions. AIChE Journal, 42(5):1225–1233, 1996.
146
References
147
6. Implementation of breakup models
T. Wang, J. Wang, and Y. Jin. A novel theoretical breakup kernel function for
bubbles/droplets in a turbulent flow. Chemical Engineering Science, 58(20):
4629–4637, 2003.
T. Wang, J. Wang, and Y. Jin. A CFD-PBM coupled model for gas-liquid flows.
AIChE Journal, 52(1):125–140, 2006.
148
Chapter
7
Numerical investigation of gas
holdup and phase mixing in
bubble column reactors
A discrete bubble model (DBM) has been used to study the overall gas holdup and the
phase mixing in bubble column reactors. The Eulerian-Lagrangian approach has the
advantage that it is possible to study the overall gas holdup and the gas phase mixing
in a direct way. Furthermore, by introducing tracer particles, also the liquid phase
mixing can be studied in a Lagrangian manner.
Comparisons suggest that the overall gas holdups obtained from the discrete bubble
model agree very well with the correlations of Kumar et al. (1976), Heijnen and
Van’t Riet (1984) and Ruzicka et al. (2001) within the applied range of the superficial
gas velocity. The gas phase dispersion coefficients from the simulation results agree
pretty well with Wachi and Nojima (1990)’s correlation, which is derived based on
the recirculation theory. In addition, the turbulent diffusion coefficient calculated
from the tracer particles velocities are very close to those from the literature within
the applied range of the superficial gas velocity.
149
7. Gas holdup and phase mixing in bubble column reactors
7.1 Introduction
150
7.2. Discrete bubble model
the reaction mixture is uniform and is typically used to describe either batch
reactors or continuously mixed tank reactors. On the other hand, mixing or
diffusion of the reacting mixture does not occur in the flow direction in plug
flow reactors. Real chemical reactors, however, may neither exhibit perfect
mixed nor plug-flow behavior. For this reason, axial dispersion models, have
been developed to describe the deviations from these simplified model reac-
tors. In addition, the concept of residence time distribution (RTD) has proven
a powerful tool to diagnose and characterize non-ideal reactors since Danck-
werts (1953). In bubble column reactors, investigations on the mixing of both
the gas and liquid phase have been reviewed by several authors, such as Joshi
(1980, 1982); Shah et al. (1982); Heijnen and Van’t Riet (1984) and Deckwer
and Schumpe (1993).
Computational Fluid Dynamics (CFD) has gained considerable interest in
recent years in the field of chemical engineering (Jakobsen, 2008). In addition
to experimental routes, it provides an alternative way to study and character-
ize chemical reactors in detail. On the other hand, chemical reactor modeling
is also crucial for scaleup and design of chemical reactors.
In this work we will focus on the prediction of gas holdup and mixing of
both the gas and liquid phase using the Eulerian-Lagrangian approach. The
simulation results will be compared with corresponding correlations obtained
from literature.
The discrete bubble model (DBM) used in this study was originally developed
by Delnoij et al. (1997) and Delnoij et al. (1999). In this model, the liquid
phase hydrodynamics is represented by the volume-averaged continuity and
Navier-Stokes equations while the motion of the each individual bubble is
tracked in a Lagrangian way.
151
7. Gas holdup and phase mixing in bubble column reactors
dv
ρg V = ΣF (7.1)
dt
dr
=v (7.2)
dt
The net force acting on each individual bubble is calculated by considering
all contributing forces. It is assumed that the net force is composed of separate,
uncoupled contributions due to gravity, pressure, drag, lift, virtual mass and
wall force respectively:
ΣF = FG + FP + FD + FL + FVM + FW (7.3)
FG = ρ g Vg (7.4)
FP = −V∇P (7.5)
1
FD = − CD ρl πd2 |v − u|(v − u) (7.6)
8
where the drag coefficient CD is determined according to Rusche (2002) which
takes swarm effects into account:
h i
CD = CD∞ exp(3.64α g ) + α0.864
g (7.7)
where the drag coefficient of a single bubble CD∞ is taken from Tomiyama
et al. (1998):
16 48 8 Eo
CD∞ = max min (1 + 0.15Re0.687 ), , (7.8)
Re Re 3 Eo + 4
152
7.2. Discrete bubble model
ρl |v − u|d
where Re is the bubble Reynolds number, Re = .
µl
A bubble rising in a non-uniform liquid flow field experiences a lift force
due to vorticity or shear in the flow field. The shear induced lift force acting
on a bubble is usually written as (Auton, 1987):
where
f (EoH ) = 0.00105Eo3H − 0.0159Eo2H − 0.0204EoH + 0.474 (7.11)
The modified Eötvös number, EoH is defined by using the maximum horizon-
tal dimension of a bubble as a characteristic length as follows:
(ρl − ρ g )gd2H
EoH = (7.12)
σ
The maximum horizontal diameter of the bubble is obtained from the
bubble aspect ratio E according to Wellek et al. (1966):
dV 1
E= = (7.13)
dH 1 + 0.163Eo0.757
where dV is the maximum vertical diameter of the bubble and Eo is the Eötvös
(ρl − ρ g )gd2
number, Eo = .
σ
The relation between the above two diameters and the diameter of the
bubble d in the discrete bubble modeling is as follows:
153
7. Gas holdup and phase mixing in bubble column reactors
where nz and nW , respectively, are the normal unit vectors in the vertical and
wall normal direction, L is the dimension of the system in the wall normal
direction, and y is the distance to the wall in that direction. Finally, the wall
force coefficient CW is given by:
exp(−0.933Eo + 0.179) 1 ≤ Eo ≤ 5,
CW =
(7.17)
0.007Eo + 0.04 5 < Eo ≤ 33.
∂
αl ρl + ∇ · αl ρl u = 0 (7.18)
∂t
∂
αl ρl u + ∇ · αl ρl uu = −αl ∇p − ∇ · (αl τl ) + αl ρl g + Φ (7.19)
∂t
The liquid phase is assumed to be Newtonian, thus the stress tensor τl can be
expressed as:
2
τl = −µeff,l ((∇u) + (∇u)T − I(∇ · u) (7.20)
3
where µeff,l is the effective viscosity. In the present model, the effective viscos-
ity is composed of two contributions, the molecular viscosity and the turbulent
viscosity:
µeff,l = µL,l + µT,l (7.21)
154
7.3. Correlations of the gas holdup
where αij = ∂u j /∂xi , βij = ∆2m αmi αmj and Bβ = β11 β22 −β212 +β11 β33 −β213 +β22 β33 −
β223 . ∆i is the filter width in the i direction.
155
7. Gas holdup and phase mixing in bubble column reactors
paring with results obtained in those columns with smaller diameter, they
found that the gas holdup varies directly with the superficial gas velocity and
is less in the column with a large diameter than in the column with a small
diameter. However, when the diameter is larger than 0.54 m, the gas holdup
is no longer influenced by the column diameter. The experimental data repro-
duced from Fair et al. (1962), a linear fitted curve and the gas holdup curve
from Shulman (1950) for a bubble column with a diameter of 0.12 m are shown
in Figure 7.1.
Φ=0.54 m
0.3 Φ=1.26 m
Fitted curve
Gas holdup [-]
Φ=0.12 m
0.2
0.1
0
0 0.02 0.04 0.06
ug [m/s]
Figure 7.1: Gas holdup curve (reproduced from) Fair et al. (1962).
Kato and Nishiwaki (1972) proposed the following correlation for the gas
holdup in an air-water system (Shah et al., 1982):
2.51u g
= γ
(7.23)
0.78 + βu0.8
g (1 − e )
where
β = 4.5 − 3.5 − 2.548φ1.8 (7.24)
and
γ = 717u1.8
g /β (7.25)
156
7.3. Correlations of the gas holdup
Kumar et al. (1976) correlated 382 gas holdup data points ( < 0.35andu g <
0.15 m/s) with the following equation:
According to the circulation theory and an assumption for the single bub-
ble rise velocity, Heijnen and Van’t Riet (1984) proposed the following expres-
sion for the homogeneous flow regime:
= 4u g (7.28)
157
7. Gas holdup and phase mixing in bubble column reactors
There are several factors that may induce longitudinal mixing in a bubble
column, such as turbulent eddies, a nonuniform velocity distribution over the
cross section of the reactor and molecular diffusion. In most of the practical
cases, the former two factors dominate the molecular diffusion (Westerterp
et al., 1987).
∂C ∂2 C ∂(UC)
=D 2 − (7.32)
∂t ∂z ∂z
where C represents the species concentration, D the axial dispersion coefficient
and U a characteristic velocity that depends on the phase. For instance for the
gas phase, the characteristic velocity is defined as U = u g /.
The axial dispersion coefficient D is directly related to the process of mixing
in reactors. For instance, a large D means rapid mixing while a small D implies
slow mixing, whereas D = 0 corresponds to plug flow conditions.
When the boundary conditions are known, i.e. open-open boundary con-
ditions, the above equation can be solved analytically. From a pulse tracer
experiment, the dispersion coefficient can then be obtained from the tracer
curve at the outlet of the reactor, that is, from the mean and variance of the
residence time distribution of the tracer.
158
7.4. Phase mixing
The residence time distribution E(t) normally has the property that the
area under the curve is unity:
Z ∞
E(t) dt = 1 (7.33)
0
In addition, the mean residence time tm and the variance σ2 of the residence
time distribution can be calculated as follows:
Z ∞
tm = tE(t) dt (7.34)
0
Z ∞
σ2 = (t − tm )2 E(t) dt (7.35)
0
Consequently the axial dispersion coefficient can be obtained with the
mean residence time tm and the variance σ2 . For instance, the Peclet number
Pe can be calculated in a closed-closed system (Levenspiel, 1999):
σ2 2 2
σ2θ = = − (1 − e−Pe ) (7.36)
t2m Pe Pe2
159
7. Gas holdup and phase mixing in bubble column reactors
where
v(t)v(t + τ)
RL (τ) = (7.42)
v02
ur φ
Bo = = 0.2 (7.43)
Dg
where Bo is Bodenstein number and ur is the relative velocity between the gas
and the liquid phase.
Men’shchikov and Aerov (1967) studied the gas phase axial mixing in a
bubble column (φ = 0.3 m) with the superficial gas velocity varying in the
range of 0.008-0.1 m/s and proposed the following correlation (Joshi, 1982):
D g = 1.47u0.72
g (7.44)
Towell and Ackermann (1972) correlated their own experimental data and
those from the literature with a correlation as follows (Joshi, 1982):
D g = 19.7φ2 u g (7.45)
The superficial gas velocity applied for the correlation ranges from 0.009 m/s
to 0.13 m/s, whereas the diameter of the bubble column ranges from 0.09 m to
1 m.
160
7.4. Phase mixing
Joshi (1982) reviewed six experimental investigations for the gas phase
dispersion in bubble columns. Based on this, he proposed the following
correlation which covers all the experimental data presented in the six exper-
imental studies:
u2g
D g = 110φ2 (7.46)
Heijnen and Van’t Riet (1984) suggested that the dispersion coefficient is
constant in the homogeneous flow regime according to experimental data in
the literature. They gave the following dispersion coefficient correlation in
the heterogeneous flow regime for the air-water system:
However, they also mentioned that the transition from the homogeneous
to heterogeneous flow regime is difficult to distinguish due to geometric
parameters, such as sparger location and uniformity of gas distribution.
On the basis of the recirculation theory for a bubble column in the turbulent
flow regime, Wachi and Nojima (1990) derived the following equation for the
axial dispersion coefficient of the gas phase:
D g = 20φ3/2 u g (7.48)
They also measured the gas phase dispersion coefficients in two bubble
columns (φ = 0.2 m and 0.5 m) using pulse experiments at superficial gas
velocities in the range of 0.029-0.456 m/s. Comparison between the theoreti-
cal correlation and the measurements revealed that the derived correlation is
able to reflect the dependencies of the axial dispersion of the gas phase on the
column diameter and the superficial gas velocity.
Dl = 0.30φ2 u1.2
g + 170δ (7.49)
161
7. Gas holdup and phase mixing in bubble column reactors
where δ is hole diameter of the perforated plate. Note that the units in the
equation are non-SI units. That is, the unit of dispersion coefficient is cm2 /s,
superficial gas velocity has unit of cm/s and the hole diameter is in cm.
Furthermore, the range of superficial gas velocity in his study ranges 0.02 m/s
and 0.25 m/s.
Towell and Ackermann (1972) developed a correlation for dispersion coef-
ficient as a function of the column diameter and the gas velocity (Majumder,
2008):
Dl = 1.23φ1.5 u0.5
g (7.50)
The diameters of the bubble columns used in this study are 0.04 m and 1.07 m,
while the superficial gas velocity ranged from 0.016 to 0.13 m/s and 0.009 to
0.034 m/s respectively (Joshi, 1982).
By measuring the dispersion coefficient in two tall bubble columns with
(φ = 0.15 m and 0.2 m), Deckwer et al. (1974) correlated his own experimental
data with those in literature and proposed an expression which is similar to
the correlation proposed by Towll and Ackermann (1972):
Dl = 0.678φ1.4 u0.3
g (7.51)
The superficial gas velocity used in the two bubble columns ranged from
0.004 to 0.13 m/s and 0.01 to 0.14 m/s respectively.
Hikita and Kikukawa (1974) studied the liquid dispersion coefficient in two
bubble columns and presented a correlation for liquid dispersion coefficient:
Dl = (0.15 + 0.69u0.77
g )φ
1.25
(7.52)
Baird and Rice (1975) adopted dimensional analysis to derive the disper-
sion coefficient based on isotropic turbulence model:
162
7.5. Numerical aspects of phase mixing study
Note that most of the above correlations that are either obtained from
experiments or derived theoretically in the literature show proportional de-
pendencies of the liquid phase axial dispersion coefficient on the diameter of
the bubble column and the superficial gas velocity. The only difference is that
the powers of these two parameters and the constant vary slightly from each
other.
163
7. Gas holdup and phase mixing in bubble column reactors
2002). In the present work, tracer particles are used to study the mixing of the
liquid phase.
For the liquid phase, the Navier-Stokes equations are solved on an Eulerian
grid and the liquid velocities are obtained on the faces of each computational
cell. The velocity of a tracer particle v at a certain moment t can then be
interpolated according to its position r(t) in the grid. Accordingly, after a time
∆t, the new position of the tracer particle can be calculated as:
Z t+∆t
r(t + ∆t) = r(t) + v(r(t), t) dt (7.57)
t
The bubble column studied here is shown in Figure 7.2. The cross-sectional
area of the column is 0.15 m × 0.15 m (W × D). The column is initially filled
with water to a height of 0.6 m (H). Air is used as the dispersed phase and
introduced into the column through a perforated plate at the bottom of the
column. The material properties of both phases are taken according to their
room temperature values. The bubble column is operating under atmospheric
pressure.
A perforated plate with 576 holes (24 × 24) is used as gas distributor. All
the holes have a diameter of 1 mm and are positioned in the central region of
164
7.6. Simulation details
165
7. Gas holdup and phase mixing in bubble column reactors
two sets of tracer particles are released uniformly over the cross-section of the
bubble column. The set of tracer particles released from the bottom of the
bubble column is denoted as “Tracer0” and the other set of tracer particles
released from the top of the bubble column is denoted as “Tracer1”. Moreover,
both sets of tracer particles are released simultaneously at the simulation time
t = 30 s at which the flow in the bubble column is fully developed.
The instantaneous gas holdup is recorded at every 0.04 s during the sim-
ulation. Examples of gas holdup histories at a superficial gas velocity
u g = 0.005 m/s and u g = 0.025 m/s are plotted in Figure 7.3.
0.03
0.15
0.02 0.1
ε [-]
ε [-]
0.01 0.05
0 0
0 50 100 150 0 50 100 150
t [s] t [s]
The plots show that the gas holdup increases monotonically at the begin-
ning of the simulation. After a short time, the gas holdup reaches a peak
and then starts to fluctuate during the simulation. Bubbles are injected into
the bubble column continuously and rise in the vertical direction, which is
reflected in the increasing initial part of the curves. Once the bubbles reach the
166
7.7. Results and discussion
top of the bubble column, most of them escape from the column directly. This
leads to the small decrease of the gas holdup after the first peak. As soon as a
liquid circulation pattern in the bubble column has developed, some bubbles
may be trapped in the circulating liquid and remain in the column for a longer
time. Hence, fluctuations in the gas holdup curves result.
In order to obtain the overall gas holdup in the bubble column for each
superficial gas velocity, the instantaneous gas holdups are averaged after 30
seconds, in a similar fashion as described in Chapter 5. The resulting overall
gas holdup for each superficial gas velocity is listed in Table 7.1. From this
table it can be seen that the standard deviation of the averaged gas holdup σ
is quite small compared to the overall gas holdup. Furthermore, within the
adopted range of the superficial gas velocity, the overall gas holdup is related
to the superficial gas velocity with a multiplication factor of 4 approximately.
In Figure 7.4, the overall gas holdups obtained from the simulations are com-
pared with the literature correlations introduced earlier.
It is seen that the simulation results agree with the correlations of Kumar
et al. (1976), Heijnen and Van’t Riet (1984) and Ruzicka et al. (2001) quite well.
These results indicate that the overall gas holdup increases nearly linearly
with the superficial gas velocity with a factor of 4 in the adopted range of
the superficial gas velocity. It can also be seen that the gas holdups obtained
from the correlation of Shulman (1950) are higher than those obtained from
others, which may be due to the diameter of the bubble column as reported
by Fair et al. (1962). In addition, the correlations of Fair et al. (1962) and Kato
and Nishiwaki (1972) are close to each other when the superficial gas velocity
167
7. Gas holdup and phase mixing in bubble column reactors
0.25
Shulman (1950)
Fair (1962)
0.2 Kato (1972)
Kumar (1976)
0.15 Kawase (1987)
ε [-] Heijnen (1984)
Ruzicka (2001)
0.1 DBM
0.05
0
0 0.005 0.01 0.015 0.02 0.025 0.03
ug [m/s]
Figure 7.4: Comparison of the computed gas holdup with literature correla-
tions.
u g < 0.025 m/s. As the superficial gas velocity increases, the gas holdup
obtained according to Kato and Nishiwaki (1972) increases faster than that
obtained from Fair et al. (1962). Moreover, Kawase and Moo-Young (1987)’s
correlation possesses a more flat trend.
The residence times of all individual bubbles in the bubble column are
recorded during the simulation and from this data set, the residence time
distribution is determined. In Figure 7.5, the residence time distribution for
each superficial gas velocity is shown. The plot shows a single peak in the
residence time distribution, which means that most of the bubbles in the bub-
ble column exit the column at the corresponding time. Moreover, the long tail
of the curve indicates that some bubbles remain in the column for longer res-
idence times due to reasons mentioned before. As the superficial gas velocity
increases, it is found that the peak of the residence time distribution shifts to
smaller residence time. That can be explained by the fact that at higher gas
168
7.7. Results and discussion
0.06
ug=0.005 m/s
ug=0.01 m/s
ug=0.015 m/s
ug=0.02 m/s
0.04 ug=0.025 m/s
E(t) [1/s]
0.02
0
0 2 4 6 8 10
t [s]
169
7. Gas holdup and phase mixing in bubble column reactors
Table 7.2: Mean and variance of the simulated residence time distributions.
u g [m/s] 0.005 0.01 0.015 0.02 0.025
tm [s] 2.88 2.94 2.90 2.99 3.01
σ2 [s2 ] 0.81 1.35 2.06 2.77 3.38
σ2θ [-] 0.10 0.16 0.25 0.31 0.37
Pe [-] 19.50 11.74 7.00 5.25 4.05
D g [m2 /s] 0.007 0.011 0.019 0.025 0.033
the earlier introduced correlations, the correlation of Joshi (1982) involves the
relation between the superficial gas velocity and the overall gas holdup in
the bubble column. Due to the fact that the overall gas holdups from the
simulations agree with the correlations in the literature quite well, we used
the overall gas holdup data obtained from the simulations to calculate the gas
phase dispersion coefficient from the correlation proposed by Joshi (1982). A
comparison is shown in Figure 7.6.
0.12
Diboun (1967)
0.1 Men'shchikov (1967)
Towell (1972)
Joshi (1982)
0.08
Dg [m2/s]
Heijnen (1984)
Wachi (1990)
0.06 DBM
0.04
0.02
0
0 0.005 0.01 0.015 0.02 0.025 0.03
ug [m/s]
Figure 7.6: Comparison of the simulated gas dispersion coefficient with liter-
ature correlations.
The comparison suggests that the simulation results agree very well with
the correlation of Wachi and Nojima (1990) which is derived based on the recir-
170
7.7. Results and discussion
culation theory. The gas phase dispersion coefficients from the simulations,
however, are far below the correlation of Men’shchikov and Aerov (1967).
Men’shchikov and Aerov (1967) proposed the correlation according to the ex-
perimental data from a bubble column with the diameter of 0.3 m and did not
consider the effect of the column diameter in the correlation. The dependence
of the gas phase dispersion on the column diameter, however, is used in most
of the correlations. Moreover, the effect of the column diameter on the gas
phase dispersion coefficient has been proven by Mangartz and Pilhofer (1981).
In addition, the simulation results are close to the correlations of Diboun and
Schügerl (1967), Towell and Ackermann (1972), Joshi (1982) and Heijnen and
Van’t Riet (1984) at low superficial gas velocities (i.e. u g ≤ 0.01 m/s), while
the difference becomes larger when the superficial gas velocity increases.
The two methods for interpolating the Eulerian velocity to the Lagrangian
velocity, tricubic interpolation and trilinear interpolation are compared with
each other. For instance, the mean-square displacement of the tracer particles
and the autocorrelation of vertical component of the tracer particle velocities
at superficial gas velocity u g = 0.005 m/s are shown in Figure 7.7.
It can be seen that there are some minor differences between the two
171
7. Gas holdup and phase mixing in bubble column reactors
methods with respect to both the mean-square displacement and the autocor-
relation, which suggests that the interpolation scheme only slightly influences
the resulting Lagrangian time series. In the following discussions, only results
obtained with the tricubic interpolation scheme are considered.
The turbulent diffusion coefficient of the liquid phase in the bubble column
is determined according to Equation 7.40. The Lagrangian time scale TL is
determined as the time for the autocorrelation to decrease to 1/e of its initial
value (Squires and Eaton, 1991). The results obtained from both sets of tracer
particles are listed in Table 7.3. The turbulent diffusion coefficient determined
from the tracer particles “Tracer0” is denoted as Dl0 and that determined from
the tracer particles “Tracer1” is represented as Dl1 . Moreover, for each set of
tracer particles, the turbulent diffusion coefficient is determined four times
and the average of these four samples is presented in Table 7.3 along with the
standard deviation which are denoted as σ0 and σ1 .
The table shows that the turbulent diffusion coefficients obtained from the
two set of tracer particles are quite close to each other. However, the standard
deviation of the turbulent diffusion coefficient is relatively large, particularly
at low superficial gas velocity (i.e. u g = 0.005 m/s). Moreover, it can be found
that the turbulent diffusion coefficient of the liquid phase nearly increases
with the superficial gas velocity.
172
7.7. Results and discussion
The turbulent diffusion coefficient of the liquid phase obtained from the sim-
ulations are compared with the correlations in the literature and shown in
Figure 7.8.
0.04
Ohki (1970) Towell (1972)
Deckwer (1974) Hikita (1974)
Zehner (1986) Heijnen (1984)
0.03
DBM (Tracer0) DBM (Tracer1)
Dl [m2/s]
0.02
0.01
0
0 0.005 0.01 0.015 0.02 0.025 0.03
ug [m/s]
173
7. Gas holdup and phase mixing in bubble column reactors
the dispersion coefficients calculated from Hikita and Kikukawa (1974)’s cor-
relation are higher than those obtained from the others at low superficial gas
velocity (i.e. u g < 0.02 m/s). Ohki and Inoue (1970)’s correlation increases
with the superficial gas velocity faster than the others and is larger than the
others when u g > 0.02 m/s. It is clearly found that the turbulent diffusion
coefficients obtained from the previous section are very close to the disper-
sion coefficients calculated from the correlations in the literature within the
applied range of the superficial gas velocity.
7.8 Conclusions
The present chapter utilized the discrete bubble model (DBM) to investigate
the overall gas holdup and the mixing of the both phases in bubble columns.
One of the advantages of the Eulerian-Lagrangian approach is that the dis-
persed phase is treated as individual elements, i.e. bubbles that are tracked
individually. The total number and the residence times of bubbles in the
bubble column reactors can be determined conveniently during simulations
and thus, the overall gas holdup and the residence time distribution of the
gas phase can be obtained accordingly. Hence, the dispersion coefficient de-
scribing the extent of the gas phase mixing can be calculated. Furthermore,
by introducing liquid phase tracer particles, it is possible to study the liquid
phase mixing as well.
In the first part of this chapter, the overall gas holdups from the dis-
crete bubble model are obtained within the range from u g = 0.005 m/s to
u g = 0.025 m/s. The results are then compared with correlations in the litera-
ture. The comparison reveals that the simulation results agree well with the
correlations of Kumar et al. (1976), Heijnen and Van’t Riet (1984) and Ruzicka
et al. (2001). The results suggest that the overall gas holdup increases nearly
linearly with the superficial gas velocity with a scaling constant of 4 in the
adopted range of the superficial gas velocity.
Secondly, the residence time distributions (RTD) of the gas phase are de-
termined in the residence times of bubbles from the discrete bubble model. A
sharp peak in the residence time distribution is obtained, which means that
174
Nomenclature
most of the bubbles in the bubble column exit the column at the corresponding
time. Moreover, the long tail of the curve suggests that some bubbles remain
in the column for longer residence times. As the superficial gas velocity in-
creases, one can find that the peak of the residence time distribution shifts to
smaller residence time, which is attributed to the stronger liquid circulation.
In addition, the distribution becomes wider when the superficial gas velocity
is higher, which is reflected by the increase of the variance of the residence
time distribution.
The comparison with the gas dispersion correlations in the literature shows
that the simulation results agree very well with Wachi and Nojima (1990)’s
correlation which is derived on basis of the recirculation theory.
Finally, the liquid phase mixing is studied through the introduction of
liquid phase tracer particles. Two methods for the interpolation scheme of
the Lagrangian velocity from the Eulerian velocity of the liquid phase are
compared with each other. Only small differences have been found between
the different methods. The turbulent diffusion coefficient is calculated from
the Lagrangian velocities and compared with correlations in literature. The
comparison shows that the obtained turbulent diffusion coefficients are very
close to those from the literature within the applied range of the superficial
gas velocity.
Nomenclature
175
7. Gas holdup and phase mixing in bubble column reactors
Greek letters
176
Nomenclature
Indices
b bubble
cell computational cell
D drag
eff effective
g gas phase
G gravity
i i direction
j j direction
l liquid
L lift; molecular viscosity
P pressure
S subgrid
T turbulent
VM virtual mass
W wall
177
7. Gas holdup and phase mixing in bubble column reactors
dr
= v(r(t), t) (A.I)
dt
k1 = hv (rk , tk )
1 1
k2 = hv rk + k1 , tk + h
4 4
3 9 3
k3 = hv rk + k1 + k2 , tk + h
32 32 8 (A.III)
1932 7200 7296 12
k4 = hv rk + k1 − k2 + k3 , tk + h
2197 2197 2197 13
439 3680 845
k5 = hv rk + k1 − 8k2 + k3 − k4 , tk + h
216 513 4104
8 3544 1859 11 1
k6 = hv rk − k1 + 2k2 − k3 + k4 − k5 , tk + h
27 2565 4104 10 2
178
7.B. Interpolation methods
179
7. Gas holdup and phase mixing in bubble column reactors
As shown in the Figure B.1, the function f at the position p in the regular
grid can be represented by:
X
N X
N X
N
f (dx, dy, dz) = ai jk dxi dy j dzk (B.II)
i=0 j=0 k=0
180
References
by defining:
f (pi ) if 1≤i≤8
∂
9 ≤ i ≤ 16
f (pi−8 ) if
∂x
∂
17 ≤ i ≤ 24
∂y
f (pi−16 ) if
∂
f (pi−24 ) if 25 ≤ i ≤ 32
∂z
bi =
(B.IV)
∂2
f (pi−32 ) if 33 ≤ i ≤ 40
∂x∂y
∂2
if 41 ≤ i ≤ 48
f (pi−40 )
∂x∂z
∂2
if 49 ≤ i ≤ 56
∂y∂z
f (pi−48 )
∂3
f (pi−56 ) if 57 ≤ i ≤ 64
∂x∂y∂z
Bα = b (B.V)
The matrix B−1 is the core of the tricubic interpolator and can be found in
Lekien and Marsden (2005).
References
181
7. Gas holdup and phase mixing in bubble column reactors
W. -D. Deckwer and A. Schumpe. Improved tools for bubble column reactor
design and scale-up. Chemical Engineering Science, 48(5):889–911, 1993.
W. -D. Deckwer, R. Burckhart, and G. Zoll. Mixing and mass transfer in tall
bubble columns. Chemical Engineering Science, 29(11):2177–2188, 1974.
J. J. Heijnen and K. Van’t Riet. Mass transfer, mixing and heat transfer phe-
nomena in low viscosity bubble column reactors. The Chemical Engineering
Journal, 28(2):B21–B42, 1984.
182
References
183
7. Gas holdup and phase mixing in bubble column reactors
R. Krishna, P.M. Wilkinson, and L.L. Van Dierendonck. A model for gas
holdup in bubble columns incorporating the influence of gas density on flow
regime transitions. Chemical Engineering Science, 46(10):2491–2496, 1991.
O. Levenspiel. Chemical Reaction Engineering. John Wiley & Sons, 3rd edition,
1999.
184
References
185
7. Gas holdup and phase mixing in bubble column reactors
P. Zehner. Momentum, mass and heat transfer in bubble columns. part 2. axial
blending and heat transfer. International chemical engineering, 26(1):29–35,
1986.
186
Summary
Experimental and Numerical Investigation of Bubble Column
Reactors
187
Summary
188
and Wen & Yu’s model have a better performance at low superficial gas
velocity and Rusche’s model can predict the hydrodynamics of the bubbly
flows better compared to the other models at high superficial gas velocity.
In Chapter 6, breakup models developed in literature are implemented
into the Eulerian-Lagrangian model. Moreover, the critical Weber number
for bubble breakup studied by many authors in turbulent flows is also in-
corporated in the model. The performance of different breakup models and
the critical Weber number for predicting hydrodynamics and the bubble size
distribution are compared with experimental data.
Finally, the Eulerian-Lagrangian model is further extended to study the
performance of bubble column reactors, i.e. predicting overall gas holdup
and phase mixing in Chapter 7. The residence time distribution of the gas
phase and tracer particles introduced in the liquid phase are used to study the
mixing of both the gas and liquid phase. It is found that the applied model
shows very good agreement with empirical correlations reported in literature.
189
Samenvatting
Experimentele en Numerieke Studie van Bellenkolom
Reactoren
191
Samenvatting
192
naburige bellen in een bellenzwerm resulteren in een afwijking op de wrijv-
ingskracht die geldt op individuele, geı̈soleerde bellen. In hoofdstuk 5 worden
de prestaties van verschillende wrijvingskrachtcorrelaties voor bellenzwer-
men uit de literatuur onderzocht met behulp van een discrete bellenmodel.
Aan de hand van een vergelijking met experimentele data is vastgesteld dat
bij lage gassnelheden de wrijvingskrachtcorrelaties van Lima Neto en Wen
& Yu de beste voorspellingen geven, terwijl bij hoge gassnelheden Rusche’s
correlatie de beste resultaten geeft.
In hoofdstuk 6 zijn literatuurmodellen voor het opbreken van bellen
geı̈mplementeerd in het Euler-Lagrange model. Bovendien zijn verschillende
literatuurmodellen voor het kritische Weber kental bij opbreken van bellen
in turbulente stromingen verdisconteerd in het model. De prestaties van de
verschillende modellen voor het opbreken van bellen en het kritische Weber
kental in termen van het voorspellen van hydrodynamica en de belgroot-
teverdeling zijn vergeleken met experimentele data.
Ten slotte is in hoofdstuk 7 het Euler-Lagrange model verder uitgebreid
om de prestaties van de bellenkolom reactoren te bestuderen, in het bijzonder
voor het voorspellen van de totale gasfractie en het mengen van beide fasen.
De verblijftijdverdeling van de gasfase en tracer deeltjes geı̈ntroduceerd in de
vloeistoffase worden gebruikt om het mengen van zowel de gas- als vloeistof-
fase te bestuderen. Het is gebleken dat de door het model voorspelde resul-
taten zeer goed overeenkomen met empirische correlaties gerapporteerd in
de literatuur.
193
总结
鼓泡塔反应器的实验与数值研究
由于具有结构简单、易操作、低运行与维护成本、高传热传质性质等优点,
鼓泡塔反应器被广泛地应用在化学、石油化学、生物化学、制药、冶金等工
业中的各种过程中,例如制氢、氧化、烷基化、化学气体清洗以及各种生物
技术应用等。然而,复杂的流体力学特征及其对传输性质(例如传热传质)
的影响,使得很难获得可靠设计和按比例放大的鼓泡塔反应器。很多因素,
极大地影响这类反应器的性能,例如反应器尺寸、反应器内部设计、气体分
布器设计、运行条件(如压力、温度)、表观气速以及各相的物理与化学性
质。在过去几十年里,人们做了大量关于鼓泡塔反应器的科学研究,运用了
实验与模拟技术。本研究利用实验与模拟技术详细地研究了一个具有方形截
面的鼓泡塔。
第一章介绍了鼓泡塔反应器及其根据不同实际需求的变体,以及此类反应
器的优缺点。并且介绍了与鼓泡塔反应器性能相关的关键参数。此外,第一
章中,文献综述简要地介绍了在过去几十年里,研究鼓泡塔反应器性能的实
验以及模拟技术。
在第二章里,研究了用于测量气泡性质的四针光纤探针的准确度。高速照
相技术所获得的气泡性质用于比较四针光纤探针的测量结果。比较发现,液
体物性显著地影响着探针所测量的气泡速度。最后,通过Morton数,描述了
利用探针测量气泡速度的不准确程度。
第三章通过一个方形鼓泡塔,进一步地研究了在较高的表观气速下,光
纤探针测量的准确度及其侵入性的影响。除了气泡速度,四针光纤探针还测
量了其他气泡性质,例如局部气含率、气泡弦长和比界面积。此外,通过不
同的方法,确定了气泡的尺寸。并且讨论了造成确定气泡尺寸差异的可能原
因。最后,研究了鼓泡塔内初始液体高度对气泡性质的影响。
第四章利用欧拉-拉格朗日模型研究了气体分布器对鼓泡塔内流动性质的影
响。首先,通过比较实验数据,验证了模型。随后,利用模型研究了不同的
195
总结
鼓泡区域对流体力学性质,如液体速度,湍动能及气含率的影响。此外,利
用模拟结果,获得了气相驻留时间分布。通过轴向分布模型,利用气相驻留
时间分布,描述了鼓泡塔内的气相混合。结果显示,混合程度随着鼓泡区域
的减小而变大。鼓泡区域越靠近侧壁,轴向分布系数越大。
泡状流的数值模拟需要可靠的封闭模型来描述界面动量传输率,例如作用
在气泡上的有效曳力。此外在大量气泡里,由于周围气泡的影响,作用在气
泡上的曳力可能会不同于作用在单个气泡上的曳力。第五章借助离散气泡模
型,研究了文献中的一些关于气泡群中的曳力关联式。通过比较实验数据发
现,在低表观气速下,Lima Neto以及Wen和Yu的模型优于其他模型。而在
高表观气速下,同其他模型相比,Rusche的模型能够较好地预测泡状流的流
动。
第六章将文献中发展的破碎模型应用到欧拉-拉格朗日模型。此外,湍流里
关于气泡破碎的临界Weber数也被应用到了此模型中。通过比较实验数据,研
究了这些不同破碎模型以及临界Weber数预测流动和气泡尺寸分布的性能。
最后,在第七章中,欧拉-拉格朗日模型被进一步扩展到用于研究鼓泡塔反
应器的性能,例如预测整体气含率与相混合。气相驻留时间分布和液相中的
示踪粒子被用于研究气相和液相的混合。结果表明,模型所预测的结果同文
献中的实验关联式吻合得很好。
196
List of publications
Publications
Conference papers
197
Acknowledgement
In this thesis the results of a four-year research study are presented, which
was conducted at the research group Fundamentals of Chemical Reaction
Engineering, Faculty of Science and Technology, University of Twente, the
Netherlands. I’d like to thank the Institute of Mechanics, Processes and Con-
trol - Twente (IMPACT) for the financial support to the project. Furthermore,
I would like to express my sincere thanks to those who have contributed their
help, support and care to both my work and my life in Enschede during the
past four years.
First of all, I would like to convey my thanks to my promotor, Prof. Hans
Kuipers. Your careful attitude on research and profound knowledge of process
engineering have made a strong impression on me. Your patient guidance
during monthly meetings and every group activity organized by your family
will be part of my beautiful memories for the rest of my life.
A special thank to my co-promotor, Dr. Niels Deen, who acted as a daily
supervisor. I couldn’t forget our fruitful weekly discussions, your quick-
witted mind and the encouragement you gave. Thank you for being there to
help me always. I would also like to thank you for all the grammar checking
of the thesis, the Dutch translation of the summary and all other support you
offered. I wish both Hans and you great success at Eindhoven University of
Technology.
I would like to thank the secretary of the group, Nicole Haitjema, for
preparing all the paperwork for my appointment as a PhD student and ar-
ranging housing for me before I came to Enschede. Everything you have done
for me during the past four years is highly appreciated.
Sincere thanks to Gerrit Schorfhaar. I’d like to thank you for constructing
the experimental setup for me. I also want to thank you for the fruitful
conversations and the sharing of your travelling experience and your life.
199
Acknowledgement
I’m impressed that you can still run 5 kilometers twice a week as a senior
person. I’d like to thank the other technicians in the group, Wim Leppink,
Johan Agterhorst and Erik Analbers for their help and kindness. I would also
like to thank Robert Meijer for his help on PC installing, email settings and
everything related to electronics.
I’d like to send my gratitude to the former group members for their sup-
port and kindness: Mao Ye, Christiaan Zeilstra, Dadan Darmana, Dongsheng
Zhang, Sabita Sarkar, Wouter Dijkhuizen, Tymen Tiemersma, Sander Noor-
man, Willem Godlieb, Jan Albert Laverman and Micheline Abbas. Many
thanks to warm-hearted Mao and Chris for their enthusiastic help when I first
came here. Special thanks to Dongsheng for his helping hand and inviting
me for dinners and outings with his family. I would also like to thank all the
current members of the group: Prof. Martin van Sint Annaland, Dr. Martin
van der Hoef, Maureen van Buijtenen, Nhi Dang, Jelle de Jong, Tom Kolk-
man, Sebastian Kriebitzsch, Yuk Man Lau, Ivo Roghair, Lianghui Tan, Martin
Tuinier, Olasaju Olaofe and Carles Mesado Melia for bringing a friendly and
creative atmosphere, nice X’mas parties, a memorable sailing event, borrels
and outings. Many thanks to Ivo and Tom for helping me on solving prob-
lems with clusters and other things. Thank Yuk Man for discussing the DBM
Code with me and offering other help. I’d also like to thank Martin (小马) for
talking to me in Chinese and offering his help.
I would like to extend my sincere thanks to Fausto Gallucci and his family.
The delicious Italian food and the hospitality of his family have been embed-
ded into my memory. Many thanks to Junwu Wang for the comments on my
work and the nice food at his place.
Many thanks to Wei Zhao & Wei Zhou(F), Yang Zhang & Rui Ge, Wei
Zhou(M), Xinhui Wang & Xuehui He, Xin Wang, Pengxiang Xu & Xia Bian, Yan
Song, Rui Guo and Hongmei Zhang for those happy times of dinner gathering,
table tennis and dizhu fighting. I would like to thank my former roommates
Xuexin Duan, Songyue Chen and Lixian Xu for the warm atmosphere at home,
nice conversation and pleasant dinner. I also thank Songyue’s husband Yifan
He and Lixian’s family, Hongping Luo and their son, Beibei for unforgettable
memories. I would like to thank Huaping Xu for enjoying chats and beer with
200
me. Thank Xiaoying Shao and Shu-Han Hsu for inviting me for dinners and
the trip to Greece. I also thank all the other friends: Rongmei Li, Yixuan Li,
Weihua Zhou, Xia Shang, Peng Zhang, Chunlin Song, Jinping Han, Yu Song,
Yu Zeng and Xinyan Wen.
Special thanks to Ying Zhang and her parents for taking care of me and
treating me as part of their family. I am grateful to Jie Fan for designing the
cover of my thesis.
最后,特别感谢我的家人多年来对我的关心和支持:父母业已年迈,尚未
回报丁点养育之恩,深感内疚,而父母仍理解并支持儿子的求学之情。感谢
我的哥哥、嫂子和姐姐、姐夫,他们给我了很大的关心,并赡养父母,使得
我不论在何处学习或工作,均能够安心于学业与工作。感谢我的小外甥和小
侄女,给我带来的开心和快乐。诚挚感谢周彦婷小姐所给予的关心以及在英
文写作上的帮助,使得我能够充满信心地完成论文。
柏巍 Wei Bai
September 2010, Ames
201
About the author
Wei Bai was born on March 14th 1977 in Hanzhong, Shaanxi, China. After
completing his secondary education in 1996 at Mian county’s 1st high school
in Shaanxi, China, he continued his study at Xi’an Jiaotong University, China
and obtained his bachelor’s degree in Heating, Ventilation and Air Condition-
ing Engineering on the subject of “Heating system design of high buildings”
in 2000. In 2001, he started his master study under the supervision of Prof.
Qiuwang Wang at Xi’an Jiaotong University. He received the master degree in
Engineering Thermophysics with a thesis titled ”Three-dimensional numeri-
cal simulation of periodical flow and heat transfer in primary surface channel”
in 2004. He then joined the Division of Building Science and Technology as a
research assistant at City University of Hong Kong, China. From 2006 to 2010,
he started his PhD research and was supervised by prof.dr.ir. J.A.M. Kuipers
and dr.ir. Niels Deen in the Fundamentals of Chemical Reaction Engineering
group at University of Twente, the Netherlands. The results of this research
are presented in this dissertation. In June 2010, he joined Prof. Rodney O.
Fox’s group as a post doc at Iowa State University in the U.S.A.
203