Flow Measurement and Instrumentation 52 (2016) 190–207

Contents lists available at ScienceDirect

Flow Measurement and Instrumentation

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

Estimation of bubble size distributions and shapes in two-phase bubble

column using image analysis and optical probes
crossmark
⁎
Giorgio Besagni , Pietro Brazzale, Alberto Fiocca, Fabio Inzoli
Politecnico di Milano, Department of Energy, Via Lambruschini 4a, 20156 Milano, Italy

A R T I C L E I N F O A BS T RAC T

Keywords: A precise estimation of bubble size distributions and shapes is required to characterize the bubble column fluid
Bubble column dynamics at the “bubble-scale”, and to evaluate the heat and mass transfer rate in bubble column reactors.
Homogeneous flow regime Image analysis methods can be used to measure the bubble size distributions and shapes; unfortunately, these
Image analysis experimental techniques are limited to resolve bubble clusters and large void fractions, and can not be applied
Bubble chord
under relevant operating conditions (e.g., high temperature and pressure). On the other hand, needle probes
Bubble size distribution
Optical probe
(i.e, optical and conductive probes) can be used to measure bubble sizes in dense bubbly flows and under
relevant operating conditions; however, needle probes measure chord length distributions, which should be
converted into bubble size distributions by using statistical algorithms. These algorithms rely on correlations—
generally obtained for single droplets/bubbles—that predicts the bubble shapes, by relating the bubble
equivalent diameter to the bubble aspect ratio. In this paper, we contribute to the existing discussion through
an experimental study regarding the bubble sizes and aspect ratio in a large air-water bubble column. The
experimental investigation has consisted in gas holdup, image analysis and optical probe measurements. First,
the gas holdup measurements have been used to identify the flow regime transition between the homogeneous
flow regime and the transition flow regime. Secondly, the homogeneous flow regime has been described at the
“bubble-scale”: chord length distributions and bubble size distributions have been obtained by using an optical
probe and image analysis, respectively. Based on the experimental data from the image analysis, a correlation
between the bubble equivalent diameter and the bubble aspect ratio has been proposed and has been compared
with existing correlations. Finally, the chord length distributions have been converted into bubble size
distributions using a statistical method, supported by the aspect ratio obtained through image analysis. The
proposed approach has been able to estimate correctly the bubble size distributions at the center of the column
then near the wall. We have also demonstrated that the correlations used to predicts the bubble shapes are the
main point of improvement in the method.

1. Introduction
Two-phase bubble columns have found many applications in the
chemical industries thanks to their simplicity of construction, the lack
of any mechanically operated parts, the low energy input requirements,
the reasonable prices and high performances (i.e., a large contact area
between the liquid and gas phase and good mixing within the liquid
phase throughout the column). Despite the simple system arrangement
(typically, a vertical pipe with a gas sparger located at the bottom of the
column), bubble columns are characterized by extremely complex fluid
dynamics interactions between the phases. For this reason, their
correct design, operation and scale-up rely on the knowledge of the
fluid dynamics at different scales: mainly, the “bubble-scale” (i.e.,
bubble size distributions and shapes, single bubble dynamics, collective
bubble dynamics, …) and the “reactors-scale” (i.e., flow patterns, mean
residence time of the disperse phase, dynamics of mesoscale clusters,
…). The knowledge of the fluid dynamics at the different scales can be
quantified through the precise estimation of the local (i.e., the bubble
size distributions, BSD, and the bubble aspect ratio) and the global (i.e.,
the gas holdup, εG) fluid dynamic properties. In particular, the gas
holdup is a dimensionless parameter defined as the volume of the gas
phase divided by the total volume of the dispersed phase. It determines
the residence time and, in combination with the BSD, the interfacial
area for the rate of interfacial heat and mass transfer [1,2]. The global
and local fluid dynamic properties are strictly related to the prevailing
flow regime, which can be distinguished—considering a large-diameter
bubble column—in the homogeneous flow regime, the transition flow
regime and the heterogeneous flow regime. A complete description of

⁎
Corresponding author.
E-mail addresses: giorgio.besagni@polimi.it (G. Besagni), pietro.brazzale@mail.polimi.it (P. Brazzale), alberto.fiocca@mail.polimi.it (A. Fiocca), fabio.inzoli@polimi.it (F. Inzoli).

http://dx.doi.org/10.1016/j.flowmeasinst.2016.10.008
Received 1 July 2016; Received in revised form 16 October 2016; Accepted 19 October 2016
Available online 22 October 2016
0955-5986/ © 2016 Elsevier Ltd. All rights reserved.
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

Nomenclature conversion algorithm [−]

n exponent in Eq. (11) and (12) [−]
Non-dimensional numbers N number of classes used [−]
Pc (C ) probability density function of the chord length [−]
2
Eo=
g (ρL − ρG ) deq
eötvös number [−] Pp (a ) probability density function describing the distribution of
σ
U2
the bubbles detected by the probe [−]
Fr= gd froude number [−] Ps (a )(m, s) probability density function of the bubble major axis [−]
eq
g (ρL − ρG ) μL4 Si (i=1,2,3) parameters in the swarm velocity method (Eq. (6)) [−]
Mo= morton number [−]
ρL2 σ 3 T temperature [°C]
ρL Udeq
Re= μL
reynolds number [−] U superficial velocity [m/s]
deq U2ρL ub bubble velocity [m/s]
We= weber number [−]
μ
σ u∞ terminal velocity of an isolated bubble (Eq. (11) and (12))
Nμ= μL viscosity ratio [−] [m/s]
G
Ub parameter in the drift-flux method (Eq. (9)) [m/s]
Acronyms
ubb ubble velocity [m/s]
Δtsampling optical probe sampling time [s]
BSD bubble size distribution
ε holdup [−]
CLD chord length distribution
εG, Local local gas fraction [−]
ξ generic variable [−]
Symbols
ρ density [kg/m3]
σs surface tension [N/m]
a major axis of the bubble [m]
C chord length [m]
b minor axis of the bubble [m]
m, s parameters in the log-normal probability density function
ci (i=1, …, 5) coefficient in the ellipse equation (Eq. (13)) [−]
[−]
cL lift coefficient in Eq. (B.1) [−]
x horizontal axis [−]
D*H non-dimensional diameter [−]
xc horizontal coordinate of the bubble center [m]
D*H, Cr critical non-dimensional diameter [−]
x vertical axis [−]
d23 sauter mean diameter [mm]
xc vertical coordinate of the bubble center [m]
dbmax critical bubble diameter according in Eq. (26) [mm]
z, k , i, l coefficients in the aspect ratio correlation [−]
dcr critical diameter (accordingly to the lift force) [mm]
α volume fraction [−]
dc diameter of the column [m]
μ dynamic viscosity [Pa∙s]
deq bubble equivalent diameter [mm]
φ aspect ratio [−]
do gas sparger holes diameter [mm]
e eccentricity [m]
Subscripts
f bubble frequency [1/s]
g gravity acceleration [m / s 2]
exp experimental
h height along the column [m]
predicted computed from algorithm
H0 height of the free-surface before aeration [m]
G gas phase
HD height of the free-surface after aeration [m]
L liquid phase
Hc height of the column [m]
T, E subscripts in the drift-flux formulation (Eqs. (6)–(8))
HD height of the free-surface after aeration [m]
trans transition point
J drift-flux [m/s]
swarm swarm velocity
nalgorithm number of bubbles generated by the CLD to BSD

these flow regimes has been proposed by Besagni et al. [3] and is
summarized in the following. The homogeneous flow regime—generally
associated with small gas superficial velocities, UG—is referred as the
flow regime where only “non-coalescence-induced” bubbles exist (e.g.
as detected by the gas disengagement technique, ref. [4]). The
homogeneous flow regime can be further distinguished into “pure-
homogeneous” (or “mono-dispersed homogeneous”) flow regime and
“pseudo-homogeneous” (or “poly-dispersed homogeneous”) flow re-
gime, depending on the BSD and the change of sign of the lift force
coefficient (see, for example, ref. [3]). The transition flow regime has
been identified by the appearance of the “coalescence-induced” bubbles
and is characterized by large flow macro-structures with large eddies
and a widened bubble size distribution due to the onset of bubble
coalescence. At high gas superficial velocities, a fully heterogeneous
flow regime is reached [5]; it is associated with high coalescence and
breakage rates and a wide variety of bubble sizes. To design and
optimize bubble column processes (i.e., when chemical reactions have
to be considered in practical applications), it is essential to recognize
the prevailing flow regime and, subsequently, estimate the local and the
global fluid dynamic properties. In this respect, a complete knowledge
of the global fluid dynamics of the bubble column relies on the
complete knowledge of the “bubble-scale”. Indeed, the bubble motion
and bubble dynamics characterize and influence the medium-scale
circulation (i.e., eddies that transport the dispersed phase) and large-
scale circulation (i.e., the liquid phase flowing upward in the center of
the column and downward in the region near the wall). Moreover, the
size and shape of the interface of the dispersed phase characterize the
heat and mass exchange. In this respect, beside experimental investi-
gations to obtain analytical relations for the gas holdup curve (the
relation between the gas holdup and the gas superficial velocity),
reliable methods to estimate BSDs are needed, which is the main
subject of this paper. In the following, in order to discuss the role of the
present paper in the literature, a brief literature survey on the
experimental methods to estimate the BSDs and shapes is proposed.
In the last decades, different intrusive and non-intrusive experi-
mental techniques have been proposed to measure bubble sizes and
shapes [6]. Non-intrusive techniques (i.e., image analysis) are generally
preferred over intrusive methods, since the flow conditions are not
disturbed. However, these techniques are limited to resolve large
bubble clusters and large void fractions; in this respect, the reader
may refer to the literature survey proposed in our previous paper [7].
In addition, image analysis may not be used under relevant operating

191
G. Besagni et al.
conditions (i.e. high-pressure bubble column reactors, as reviewed in
refs. [8,9]). For these reasons, despite image analysis is a useful tool to
study the bubble size distributions and shape in small-scale bubble
column (e.g., in laboratories), it fails and/or may not be applied in
industrial-scale reactors. Contrarily, needle probes (e.g., optical or
conductive probes) may be used to obtain BSDs in dense bubbly flows
and under relevant operating conditions. Unfortunately, the needle
probes provide chord length distributions (CLDs), which should be
related to the corresponding BSDs by using statistical methods. A
process that deduces the BSD from a CLD is generally called “back-
ward transform”, whereas the deduction of a CLD generated from an
existing BSD is called “forward transform”. Many authors have
proposed different approaches, starting from the pioneering paper of
Clark and Turton [10], who have proposed a “forward transform”
considering different bubble shapes. Later, in 1995, Liu and Clark
[11]—assuming ellipsoidal and vertically rising bubbles— have pro-
posed a “backward transform” by using an analytical relationship
between probability density functions. Hu et. al. [12] have used a
“forward transform” to obtain the CLD from lognormal and uniform
BSDs. Al-Oufi [13] have proposed an algorithm based on a double
transformation (backward and forward transformations). Rudisuli
[14] have proposed a Monte Carlo method that simulates the bubble
sampling process. Recently, Hoang et al. [15] have decomposed the
CLD into sub-distributions, which are subsequently converted into
BSDs using the approach of Liu and Clark [11], and finally the system
BSD is computed. It is important to observe that, to relate a CLD to the
corresponding BSD, a correlation between the bubble shape and size is
required. Several attempts have been made in literature to correlate the
aspect ratio as a function of dimensionless parameters [16–18]: some
authors have used the Eötvös number, Eo [19,20], others have adopted
the Weber number, We [20–22], while Tadaki and Maeda [23] used the
Tadaki number [24]. Other authors have used more than one dimen-
sional number, i.e. Bozzano and Dente have used both the Mo and the
Eo numbers [25] and Aoyama et al. [26] have used the Re and the Eo
numbers. However, these correlations have been developed for single
bubbles/drops, and they may not be suitable for dense bubbly flows.
This issue has been discussed by Besagni and Inzoli [7] who have
analyzed bubble sizes and shapes in annular gap bubble column and
have proposed a correlation for the aspect ratio valid for dense bubbly
flows.
In this paper, we have presented an experimental study concerning
the bubble size distributions and shapes in a large-diameter and large-
scale air-water bubble column (the present bubble column has been
previously studied by the authors by a comprehensive experimental
investigation [27]). In particular, this paper studies aspect ratio
correlations for homogeneous flows and investigates the role of aspect
ratio correlations in bubble chord conversion algorithms. The paper is
structured as follows. In Section 2, the experimental setup and the
measurement techniques (gas holdup, image analysis and optical probe
measurements) have been described. In Section 3, the experimental
results have been presented. First, the gas holdup measurements have
been used to identify the flow regime transition between the homo-
geneous and the transition flow regime. Secondly, in the homogenous
flow regime, the chord lengths of bubbles have been measured by using
an optical probe and the bubble size distributions have been estimated
with image analysis. Thus, the chord length distributions have been
converted into bubble size distributions using a statistical method,
supported by the aspect ratio obtained through image analysis. Finally,
the conclusions have been given in Section 4.
2. Experimental setup and methods
2.1. The experimental setup
The experimental facility (Fig. 1a) is a non-pressurized vertical pipe
made of Plexiglas® with dc=0.24 m and Hc=5.3 m. A pressure reducer
192
Flow Measurement and Instrumentation 52 (2016) 190–207
controls the pressure upstream from the rotameters (1) and (2), used to
measure the gas flow rate (accuracy ± 2% f.s.v., E5-2600/h, manufac-
tured by ASA, Italy). The experimental facility has been also equipped
with boxes filled with water around the cylindrical channel for the flow
visualization (Fig. 1b, Section 2.3). The values of gas density (used to
compute the superficial gas velocity) have been based upon the
operating conditions existing at the column mid-point (computed by
using the ideal gas law) [28]. The gas distributor, is a spider-sparger
distributor (Fig. 1c, d) with hole diameters do=2–4 mm (Fig. 1e). The
spider sparger, shown in Fig. 1c and d, has six arms made of 0.12 m
diameter stainless steel tubes soldered to the center cylinder of the
sparger. The sparger has been installed with the six holes located on the
side of each arm facing upward. These holes have been distributed as
shown in Fig. 1e, with an increasing diameter moving toward the
column wall. The gas and liquid temperatures have been checked and
maintained constant at room temperature during all the experiments
(22 ± 1 °C). Filtered air from laboratory lines has been used as the
gaseous phase in all the experiments; the air-cleaning line consists in
filters (mechanical and activated carbon) and condensation drying unit,
in order to clean the gas phase properly and, thus, to avoid the presence
of contaminants in form of (i) solid particles and (ii) organic sub-
stances. Deionized water has been used as the liquid phase. During the
experimentation, great care has been taken to ensure that the bubble
column has been always clean to minimize any contamination that
might affect the results (the system has been previously flushed to
remove contaminants and to avoid the presence of additional surfac-
tants).
2.2. Gas holdup and flow regime transition
2.2.1. Gas holdup measurements
Measurements of the bed expansion allowed the evaluation of the
gas holdup, εG. The procedure involves measuring the location (height)
of the liquid free surface when air flows in the column. The gas holdup
is then obtained using the relation:
(HD − H0 )
εG =
HD (1)
where HD and H0 are the heights of the free-surface after and before
aeration, respectively (H0=3.0 m above the distributor, H0/dc=12.5).
2.2.2. Flow regime transition
Two main transitions exist in large-diameter bubble columns:
1. the transition between the homogeneous flow regime and the
transition flow regime (in the present bubble column, the “mono-
dispersed homogeneous” does not exist);
2. the transition between the transition flow regime and the hetero-
geneous flow regime.
Although the flow regimes transitions do not occur instantaneously,
the definition of approximate transition points is helpful to understand
and model the fluid dynamic behavior of bubble columns [29]. In the
literature, many authors consider only the first flow regime transition,
without any reference to the second one, except for a limited number of
studies [5,30,31]. In our previous paper we have shown how the swarm
velocity and the drift-flux/Wallis plot methods have been able to
identify the first flow regime transition point [27]; In particular, the
first flow regime transition point, in terms of the transition gas velocity
and transition gas holdup, is computed as the average of the values
obtained by the two methods:
UG, trans, swarm + UG, trans, Wallis
UG, trans =
2 (2)
εG, trans, swarm + εG, trans, Wallis
εG, trans =
2 (3)
G. Besagni et al.
Fig. 1. Experimental setup.
The details concerning the two methods have been provided below.
It is worth noting that the transition point between the transition
and the heterogeneous flow regime cannot be identified by means of
these two methods, which is a matter of future studies. Thus, in the
following, for the sake of clarity, we refer to “flow regime transition
point” considering the first flow regime transition.
2.2.2.1. Swarm velocity method. The swarm velocity method has
been developed by Zuber and Findlay [32] and is based on the
193
Flow Measurement and Instrumentation 52 (2016) 190–207
swarm velocity:
Uswarm = UG / εG (4)
The swarm velocity is plotted against the superficial gas velocity:
Uswarm is almost constant in the homogeneous flow regime (in some
cases, it can be slightly decreasing), but it starts to increase as the
system enters the heterogeneous flow regime at a transition superficial
velocity, Utrans. The appearance of the first large bubble is responsible
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

for this sudden increase in the swarm velocity and is an indication of course. even if the test section has been covered by the squared box
flow regime transition. In this study, the quantitative evaluation of filled with water, some refraction problems may still arise, as the
Utrans has been determined by the intersection between the trends of refraction index of pipe wall is different from that of water. Indeed, the
Uswarm in the two flow regimes. refractive index of the Plexiglas® is 1.48 and the one of water is 1.33.
Thus, Uswarm has been taken as constant in the homogeneous flow This effect is negligible, as demonstrated in the annular gap bubble
regime (Eq. (5)): column configuration [7]. The same has been verified in the present
Uswarm,homogeneous = cons configuration, by evaluating the conversion factor for a ruler placed in
(5)
the center of the column for different radial positions. The effect of the
where as, in the heterogeneous/transition flow regime it has been optical distortion has been found to be negligible: the maximum
determined by a least squares fitting of the following function: difference in the conversion factors is 0.5 pixel/mm, as discussed in
ref. [3]. Images have been acquired approximately, h=2.4 m from the
Uswarm,heterogeneous = S1 (UG )S2 + S3 (6)
sparger; which corresponds to the developed region of the two-phase
where S1, S2 and S3 are fitting parameters. flow [27].
The transitional velocity is then evaluated by solving the following
equation: 2.3.2. Image analysis methods: bubble shape and sizes
Uswarm,homogeneous = Uswarm,heterogeneous The study of bubble shape and size using image analysis is a matter
(7)
of intensive research. Ellingsen and Risso [37] have discussed how the
bubble shape is totally defined by knowledge of its major axis and its
minor axis: under these conditions, the projected contours are ellipses.
2.2.2.2. Drift-flux method. The drift-flux method has been proposed Taking into account their conclusion, in this paper, each bubble is
by Wallis [33]. This method is based on the drift-flux, which represents approximated and reconstructed ellipses. Moreover, in the present
the gas flux through a surface moving with the speed of the two-phase case, the images have been acquired in the developed region of the two-
mixture and is experimentally obtained as follows: phase flow, where the break-up and coalescence phenomena are not
JT = UG (1 − εG ) ± UL εG (8) relevant, all bubbles rose mostly rectilinearly (ref. [27]) and their
shapes were axisymmetric at most. In order to sample the ellipsoidal
bubbles, the same approach that has been applied to an annular gap
In the batch mode, UL=0 m/s. The detailed derivation of Eq. (8) bubble column [4,7] and to other systems [3,27] has been also used in
has been proposed by Besagni and Inzoli [34]. this work.
Theoretically, the drift-flux is written in terms of the bubble swarm In this approach, each bubble is approximated and reconstructed
velocity, whose dependence upon εG varies with the prevailing flow using following ellipse equation (Fig. 2a):
regime:
c1 x 2 + c2 xy + c3 y 2 + c4 y + c5 x + 1 = 0 (13)
JE = Ub (1 − εG ) (9)
The method is structured in four phases:
The idea in this method is to employ a model for Ub, that is valid for
the homogeneous flow regime, plot JE and JT in the same graph as a
function of εG. In the homogeneous flow regime, JE is equal to JT and,
then, the transition point is defined when:
JT ≠ JE (10)
The evaluation of Ub is a matter of discussion in the literature and
different models have been proposed and applied. In this study, the
approach of Krishna et al. [35] has been followed, which is based on the
empirical model of Richardson and Zaki [36]:
Ub = u∞ εG (1 − εG )n −1 (11)
where n is fluid-dependent and u∞ is the terminal velocity of an
isolated bubble. These values should be fitted with the aid of the
experimental data in the determination of the transition point.
Combining Eq. (9) and Eq. (11), the following equation is derived:
JE = u∞ εG (1 − εG )n (12)

2.3. Image analysis

The goal of the image analysis is to obtain bubble shape, bubble

orientation and bubble size distributions.

2.3.1. Instrumentation
The photos have been taken using a NIKON D5000 camera (f/3.5;
1/1600 s; ISO400; 4288×2848 pixels and spatial resolution approxi-
mately 11.8 pixel/mm). The back light method has been employed in
the experiments using a 500 W halogen lamp as light source. The
camera has been accurately aligned horizontally to the visualization
sections, which consist in squared boxes (filled with water) around the
vertical pipe designed to correct the distorted image (Fig. 1b). Of Fig. 2. Image processing: reference lengths.

194
G. Besagni et al.
1. Calibration. The reference conversion factor between pixels and
millimeters has been provided. In the present work, two references
have been used: a ruler along the external wall (Fig. 2b) and one
inside the column (Fig. 2c).
2. Edge detection. For each bubble, six points on the bubble have
been selected (by human vision). The analysis concerning the errors
made during the bubble shape estimation due to the handpicked
points has been proposed in ref. [27].
3. Bubble reconstruction. The ellipse equation parameters c1, c2,
c3, c4 and c5 have been computed using the least squares method. It
is worth noting that the number of sampling points requested has
been larger than the minimum number of points for defining an
ellipse: this has allowed sampling the bubbles also in overlapping
conditions. However, when increasing UG, the bubble overlapping
and the reduced back light has been such that there may be a lack of
information: it has been verified that the image analysis is suitable
for analyzing gas holdups up to εG=0.05 [38], regardless of the
operation mode and working fluids.
4. Bubble processing. The equivalent ellipse has been processed to
obtain the major axis 2a, the minor axis 2b, the center of the bubble
and the orientation angle. The bubble equivalent diameters, deq, and
the bubble aspect ratio have been obtained as:
deq = 2 3 a2b (14)
φ = b /a (15)
From Eq. (16) and Eq. (17), follows Eq. (18):
deq = 2a φ1/3
(16)
Of course, ellipsoidal bubbles are inclined not only in visualization
plane, but also in the perpendicular direction. This problem is well
known in the literature and, despite some proposals [39,40], a solution
is far from being reached and the use of 2D projected images is a
common way of processing bubble images by different authors from
different research groups, for examples refs. [37,41–46]. It is worth
noting that the bubble shape data obtained using this approach are
considered reliable: the bubble shape information allowed the extra-
polation of correlations for the aspect ratio [7] and flow regime
transition criteria [7]. The results of these analyses have been validated
with the experimental data and, therefore, the bubble shape informa-
tion is considered reliable. As for the approximation of bubbles with
ellipses, Lage and Espósito [44] have stated that the error in the
measurement of each axis of the ellipse is approximately 6%. Taking
into account the error introduced by the hypothesis of oblate spheroid
and the optical distortion, they have estimated that the experimental
error in the determination of deq is between 10% and 15%.
The number of bubbles to be sampled to achieve a reliable BSD is a
matter of discussion in the literature [47]. Various studies have
sampled different numbers of bubbles—between 50 and 100 [44],
200 [45], 250 [48], 300 [49], between 250 and 300 [50] and 100 [51].
In this study, for each case, at least 800 bubbles have been selected
using two (or more) photos, as discussed in ref. [7].
2.4. Optical probe
2.4.1. Instrumentations
A double fiber optical probe system (manufactured by RBI, France)
measures the local fluid dynamic properties. The optical probe is
inserted, via an access port, into the flow at h=1.9 m from the sparger.
Each probe is made of two 40 μm glass fibers whose tips are re-
enforced by two sharp sapphire pins: further information concerning
this equipment can be found in Boes and Hager [52]. Similar optical
probe systems have been used by different authors [4,27,34,53–58].
195
Flow Measurement and Instrumentation 52 (2016) 190–207
Optical probes distinguish the gas and liquid phases by measuring the
intensity of a laser light that is reflected and/or refracted at the probe
tip on the basis of the refractive indexes of the probe tip, gas and liquid
phases [59]. The probe signal is measured via an optoelectronic module
that emits the laser to the probe tip and converts the reflected optical
signal into a digital signal. From the digital signal, the bubble frequency
f (bubble number per unit time) and void fraction εG, Local (assuming
it equals the proportion of time when the tip is surrounded by gas) can
be obtained. By cross-correlating the signals from the two tips, bubble
traveling time from one tip to the other can be estimated and the
bubble velocity ub can be calculated. Finally, for every bubble, a chord
length is calculated by multiplying the residence time on the probe tip
for the bubble velocity. Considering the whole sampling time, a bubble
chord length distribution is obtained (which is converted into the
corresponding BSD by using the method of Section 2.5).
All the measurements have been obtained using a sampling period
equal to Δtsampling=1000 s, which is large enough to produce reliable
time-averaged values and is far above the typical values of 1–5 min for
similar optical probes [54,56,57,60].
2.4.2. Optical probe sources of errors
When considering the experimental error of the optical probes, two
aspects should be considered. The first is the statistical error associated
with the measuring time, which is proportional to the square root of the
measuring time: in our case, the absolute uncertainty in the gas
fraction is approximately 0.001. The second aspect is a bias error due
to the difficulties of piercing a bubble at the bubble edge [61]:
• improper dewetting at the probe tip (the blinding effect);
• alteration of bubble trajectory prior to or during the piercing process
(the drifting effect);
•
bubble deformation and/or deceleration at the probe tip (the
crawling effect).
Both the blinding effect and the crawling effect include deforma-
tion effects. The deformation of the blinding effect is mainly related
to a local deformation zone located in the bubble interface, which
may be produced by both the probe-induced liquid pressure over the
bubble and the direct impact itself. The crawling effect considers the
deformation of large parts of the bubble. The relative influence of
these effects on the final residence time estimates, as well as their
absolute magnitude, is not yet clear. A summary of the errors
reported by the optical probe in the literature is as follows:
• Void fraction. Generally, the optical probe is generally considered
rather accurate in terms of void fraction. Comparing optical probe
results with other measurement techniques (i.e., visual or pressure
sensors), the relative differences found in the literature are as
follows: between −0.8% and −16% [59], −2.2% [57], −3% [60],
−11% [56], between −6% and −14% [53], +12% [55]. Barrau et al.
observed the worst performance with no liquid flow and/or at low
gas fractions.
• Bubble rise velocity. The rise time errors are primarily affected
by the impact angle of the bubbles with the probe tip. The relative
difference reported in the literature, compared with other techni-
ques (i.e., image processing) are +45% [62], +5% [53], between +5
and +6% [60], between +10 and +30% [54], +15% [55], and +29%
[56], approximately +9% [57].
• Bubble diameter. The assumption of spherical bubbles leads to an
underestimation of the equivalent diameter by between +10% and
27% [54,56,57]. Applying a correction based on the aspect ratio
evaluation can reduce the error, e.g., from +26.6 to +7% [57]
2.5. CLD to BSD conversion algorithm
We have used an approach based on a double transformation (a
backward transformation and a forward transformation) to relate the
CLD (provided by the optical probe) to the BSD (representative of the
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

the algorithm. The assumption of two-dimensional bubbles is in

agreement with the discussion on bubble shape proposed by
Ellingsen and Risso [37], whereas, concerning the vertical rising
bubbles, the reader should refer to our previous paper, were was found
that most of the bubbles had a very small rising orientation [27]. It is
worth noting that, when the bubble size range becomes too wide, the
influence of the dynamic processes is more serious and bubbles are
deformed strongly and do not rise vertically: this is a matter of future
investigation.
The BSD (of the bubble column system) may be represented either
in terms of the equivalent diameters BSD(deq) or in terms of the
bubble major axis, BSD(a). In the present paper, we have represented
the BSD of the systems in terms of BSD(a). An assumption should be
made concerning the shape of the BSD and, in agreement with the
literature survey proposed in the introduction, we suppose that the
BSD is described by a log-normal probability density function, Ps(a)(m,
s). Hence, it follows:

a ⎧ [ ln(a ) − m]2 ⎫
BSD (a ) = Ps (a )(m, s) = exp ⎨− ⎬
s 2π ⎩ 2s 2 ⎭ (17)

Where a is the major (semi) axis of the bubble (Fig. 2a), m is the mean
and s is the standard deviation of the log-normal probability density
function.
Optical probes are well known to provide biased CLD because of
their intrusive nature; the bias is especially important for small or slow
bubbles, whereas the larger bubbles are more likely to be detected by
the optical probe. Therefore, the BSD of the systems should be related
to the BSD detected by the optical probe Pp(a). To this end, we have
used the same approach as proposed by Clark and Turton [10]: a
bubble has a probability to be caught proportional to the bubble
projected area, and, thus, proportional to a2. Following the proposal of
Clark and Turton [10], the probability density function describing the
distribution of the bubbles detected by the probe, Pp(a), reads as
follows:
⎧ [ ln(a) − m]2 ⎫
exp ⎨− ⎬
a
a2Ps (a ) s 2π ⎩ 2s 2 ⎭
Pp (a ) = =
∫
∞
a2Ps (a ) da +∞ ⎧ [ ln(a) − m]2 ⎫
exp ⎨− ⎬ da
a
0 ∫0 s 2π ⎩ 2s 2 ⎭ (18)

2.5.2. Description of the methods

In the following sections, the two steps of the CLD to BSD
conversion algorithm have been presented and discussed:

1. Phase A. Starting from a first-guess bubble size distribution BSD(a)

=Ps(a)(m, s), a “forward transform” has been applied to obtain a
predicted CLD.
2. Phase B. Once the log-normal parameters have been obtained, a
“backward transform” has been applied to compute BSD(deq) from
BSD(a)=Ps(a)(m, s).

2.5.2.1. Phase A: forward transform. Fig. 3b displays the flowchart of

Fig. 3. Optical probe piercing process.
bubble column system). In particular, we used as starting point the
approach proposed by Al-Oufi [13], which has been further elaborated,
improved and optimized. In the following, the method, its assumptions
and the mathematical formulation have been described.
2.5.1. The method and the assumptions
We have considered two-dimensional and vertically rising bubbles
(Fig. 3), in agreement with the considerations discussed in the image
analysis (Section 2.3.2); please notice that, under the assumptions of
two-dimensional bubbles, the transversal plane has been neglected in
the transformation methods, which has been described in the following
paragraphs.
First, the parameters (m and s) of the log-normal probability
density function Ps(a)(m, s) have been first-guessed and the corre-
sponding probabilities Ps(a)(m, s) (Eq. (17)) and Pp(a) (Eq. (18)) have
been computed along with the cumulative probability function:
a
CP (a ) = ∫0 Pp (a ) da
(19)
Secondly, we have randomly selected a bubble radius detected by
the optical probe (Eq. (18)). To this end we have applied the

196
G. Besagni et al.
Fig. 4. Gas holdup and flow regime transitions.
transformation method [63], which is generally used to generate
random deviates from probability distributions; the reader should
refer to the Appendix A for the presentation of the transformation
method.
Thirdly, the aspect ratio of the synthetic bubble has been computed
by using a correlation, which related the bubble shape and size. In our
perspective (as presented in the introduction and as verified in the
discussion of the results), the aspect ratio correlation works critically.
For example, the algorithm has been unable to reproduce the experi-
mental data using the correlation of Wellek et al. [20] (Section 3.3.2).
Therefore, a correlation based on the image analysis experimental data,
has been proposed and applied (Section 3.2.2). It is worth noting that
the correlation proposed relates the aspect ratio to the major semi-axis,
in order to reduce the computational cost of the algorithm. At this
point, the synthetic bubble has been pierced by a virtual probe at
197
Flow Measurement and Instrumentation 52 (2016) 190–207
randomly selected eccentricities, e, in the range 0 < e < a (a is the major
semi axis of the bubble) and the corresponding chord length has been
computed. It is worth noting that the assumption of vertically rising
bubbles has been applied in this point of the algorithm. The value e has
been randomly generated by a probability density function Pe(e|a).
Where (e|a) indicates a random e for a given a. A bubble having a
major semi axis a has been pierced between e and e+de with a
probability proportional to the projected area 2πede according to the
expressions (valid for e≤a):
2πede
Pe (e|a ) de =
πa2 (20)
2e
Pe (e|a ) =
a2 (21)
The cumulative function for e is given by equation, Eq. (22):
e 2e e2
Ce (e|a ) = ∫0 πa 2
de = 2
a (22)
In order to select a random value of e, under the probability density
function Pe(e|R) (Eqs. (18) and (19)), the transformation method has
been used. A random uniform variable ξ is generated (Appendix A) in
the range [0,1], therefore the corresponding value of e is defined by:
e=a x (23)
The corresponding value of chord length C has been computed as:
C = 2 a2 − e 2 (24)
The above–described processes have been repeated for a random
sample of the bubbles (nAlgorithm = 106). Summing all the chords, a
predicted CLD has been obtained, Pc(C)predicted; the predicted CLD has
been compared with the experimental CLD, Pc(C)exp, (optical probe)
and the error has been computed as follows:
N
error = ∑ [(Pc (Ci ))predicted − (Pc (Ci ))exp ]
i =1 (25)
where Pc(Ci) is the probability density evaluated of the chord length Ci.
Finally, the parameters (m, s) of the log-normal distribution
Ps(a)(m, s) have been varied for minimizing Eq. (25):
2.5.2.2. Phase B: backward transform. Once the log-normal
parameters have been obtained, a “backward transform” has been
applied to compute BSD(deq) from BSD(a)=Ps(a)(m, s). With the
transformation method a random sample of a has been obtained from
Ps(a). For each major (semi) axis of the bubble obtained, the aspect
ratio has been computed by applying the correlation (Section 3.2.2)
and, finally, the equivalent diameter has been obtained. It is worth
noting that the BSD(deq) derived from the log-normal distribution
Ps(a)(m, s) is a continuous function. Of course, in real bubble column
systems, the number of bubbles beyond a certain value is very small,
because these bubbles are not stable. This is taken into account by
verifying that the number of bubbles above the critical value, dbmax,
predicted by Ishii and Kojasoy [64] is below 0.4% (from experimental
data) and, then, the BSD is cut-off at dbmax.
The critical diameter according to Ishii and Kojasoy [64] has been
defined as follows:
σ
dbmax = 40
g (ρL − ρG ) (26)
dbmax=10 cm in air-water system at ambient condition.
G. Besagni et al.
Fig. 5. Bubbly flow conditions investigated.
3. Experimental results and discussion
3.1. Gas holdup and flow regime transition
Fig. 4a displays the gas holdup measurements. At low gas super-
ficial velocity, in the pseudo-homogeneous flow regime, the relation-
ship between the gas holdup and the UG is linear, followed by a change
in slope due to flow regime transition. The flow regime transition from
the homogeneous to the transition flow regime is evaluated by using
the swarm velocity (Fig. 4b) and the Wallis plot (Fig. 4c) methods; the
transition point is estimated as the mean of the two values and is UG,
trans=0.025 m/s. The reader should refer to our previous paper for a
complete discussion concerning the gas holdup curve, the flow regime
transition, and the comparison with the literature.
198
Flow Measurement and Instrumentation 52 (2016) 190–207
3.2. Image analysis
Five gas superficial velocities in the homogeneous flow regime—
between UG =0.0037 and UG=0.0188 m/s (Fig. 5)—have been ana-
lyzed by using the image analysis. In the pseud-homogeneous flow
regime some bubbles are spherical (Fig. 6a), others are ellipsoidal
(Fig. 6b), and others are irregular in shape (Fig. 6c). The bubbles have
been sampled both in the center and near the wall of the column at
UG=0.0037 m/s and UG=0.0074 m/s. For higher UG, the bubble
overlapping is such that sampling in the center of the column has
been impossible.
3.2.1. Size distributions and shapes
In this section we briefly summarize the data presented by Besagni
and Inzoli [27] that we will use, in this paper, to derive the correlation
for the aspect ratio. The BSDs near the wall are bimodal, whereas the
BSDs at the center of the column are unimodal (Fig. 7a and b). In
G. Besagni et al.
Fig. 6. Bubble shapes (UG=0.0037 m/s).
particular, near the wall, there is a higher number of small bubbles: this
may be because of the lift force, which pushed the small bubbles toward
the wall (Appendix B). Fig. 7c displays the BSDs for the different gas
superficial velocities. Up to UG=0.0149 m/s, the BSDs are bimodal,
whereas at UG=0.0188 m/s (closer to the flow regime transition), the
BSD changes from bimodal to unimodal and is shifted toward higher
bubble diameters with a peak between deq=3.5 mm and 4. Fig. 8
displays the relationship between the aspect ratio and the equivalent
diameter (φ–deq). Bubbles with deq < 1 mm have an aspect ratio
greater than 0.7 (spherical). Contrarily, bubbles with higher deq are
characterized by lower aspect ratios (between 0.4 and 0.7). At
UG=0.0037 m/s and UG=0.0074 m/s, the aspect distribution is com-
pared near the wall and at the center of the column (Fig. 8a, b and
Fig. 8c, d, respectively). Near the wall, there is a slightly higher number
of small bubbles (φ > 0.7, deq < 2); however, despite this small
difference, the data cover almost the same φ–deq space. This suggests
that the shape of the bubbles is not influenced by their position in the
column, as expected and as verified in the next section. All operating
conditions—until UG=0.0147 m/s—have a comparable φ–deq distribu-
tion (Fig. 8a, c, e and f); at UG=0.0188 m/s (Fig. 8g), there is a change
in the distribution with the appearance of larger and more distorted
bubbles.
It is worth noting that the aspect ratio data are scattered broadly.
First, the smaller the bubble, the larger the error of estimation: this
should change the results, causing the low deq bubbles to have a higher
aspect ratio. Secondly, the image analysis has been applied to the
pseudo-homogeneous regime in dense bubbly flow conditions and,
thus, a scattering was expected. The reader should refer to our previous
study for a complete discussion concerning the bubble sizes and shapes
199
Flow Measurement and Instrumentation 52 (2016) 190–207
and a comparison with the literature.
3.2.2. Aspect ratio correlation
The many relations between the bubble column fluid-dynamic
parameters and the variables characterizing the system make it difficult
to find general correlations for the precise estimation of the bubble
sizes and shapes, which is the analytical relation between deq (or
related non-dimensional groups) and φ. Numerous correlations are
available in literature, although due to the fluid dynamic complexity of
the problem none of them can be either considered of general validity
or applied to a wide range of geometrical parameters and operating
conditions. Therefore, the goal of this section is to assess a novel (and
simple) correlation between non-dimensional parameters and the
aspect ratio for bubbly flow and compare it to the results of previous
literature. The experimental data obtained have been used, and the
procedure applied by Wellek et al. [20] is considered as a reference, as
described by Besagni and Inzoli [7]. It is worth noting that the aspect
ratio data are scattered broadly; therefore, the data have been grouped
into classes of equivalent diameters: each class is represented by the
average aspect ratio of the bubbles belonging to that class.
The relationship between the bubble shapes, the size of the bubbles,
the speed of the bubbles and the physical properties of the system may
be written as function of the following parameters:
φ = f (ub , σ , deq, μL , μG , ρL , ΔρG − L , g) (27)
Applying the dimensional analysis, Eq. (28) has been derived from
Eq. (27):
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

Fig. 7. Bubble size distributions.

⎛ deq ub2ρ deq ub ρ gΔρdeq2 ub 2 μ ⎞ ⎛ 1⎞ 3

φ = f ⎜ σ L, μ L, σ , , L⎟
⎝ L gdeq μG ⎠
= f (We, Re, Eo , Fr , Nμr )
The use of multiple dimensionless groups did not provide a benefit
[7,20], and a single non-dimensional parameter can approximate the
system: in the present work, Eo has been selected. Please notice that we
have also performed a more detailed statistical description of the
experimental data (e.g. by using PDF of aspect ratios for every bubble
sizes) and we have observed that it does not provide considerable
benefits to the algorithm. Therefore, this approach and is not taken into
account in the present paper. In order to propose a new correlation,
firstly, a scheme of correlation must be selected for the function f, and
the following has been applied:
⎛ gΔρd 2 ⎞k1
φ = f (We , Re , Eo , Fr , Nμr ) → f (Eo) = k 0 Eo k1 = k 0 ⎜⎜ ⎟⎟ = k2 deq k3
eq
⎝ σ ⎠ is able to approximate the Bozzano-Dente [25] correlation) has been
Given the mathematical structure of the algorithm, a correlation
between the bubble aspect ratio and the bubble major semi-axis (a) has
been used to reduce the computational effort:
k
φ ====== ⇒ k2 ∙ ⎜2Rφ 3 ⎟ =k 4 ∙ak5
− 1 ⎝ ⎠
a =2d φ 3
eq (30)
(28)
At this point, two issues should be considered: (i) the smaller the
bubble, the higher the sampling error (Section 3.2.1 and ref. [7,27]);
(ii) there are not enough bubbles having high deq. In order to deal with
these issues, the following considerations have been accounted:
• Bubbles having a < 0.55 mm (deq < 1 mm). These data have
been not considered and, instead, the bubbles have been approxi-
mated as spherical bubbles, as described by the Bozzano-Dente [25]
correlation;
• Bubbles having a > 5.5 mm (deq > 8 mm). The Bozzano-Dente
[25] correlation has been applied.
Therefore, the coefficients k4 and k5 in Eq. (30) have been
computed by using a statistical regression of the experimental data
(Fig. 8) in the region 0.55 < a < 5.5 mm. A parabolic function (which
(29) used in the range a < 0.55 mm. An exponential function asymptotic
to the Bozzano-Dente [25] correlation has been used in the range a
> 5.5 mm. The resulting correlation reads as follows (and its
coefficients are listed in Table 1):

200
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

Fig. 8. Relationship between aspect ratio and equivalent diameters.

Table 1 ⎧ z1 a2 + z2 a + z3 a < 0.55

⎪
Aspect ratio correlation coefficients. φ = ⎨ k 4 ak5 0.55 < a < 5.5
⎪ l
z1 z2 z3 k4 k5 i l ⎩ ia a > 5.5 (31)

−0,6572 0,001 0,69 −0,251 0,75008 −0,3 −0,3

201
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

On the qualitative point of view, all the correlations have been able
to follow the two characteristic behaviors of the observed aspect ratios:
(i) for relatively small bubbles, the aspect ratios decrease and (ii) for
larger bubbles they stay constant. On the quantitative point of view, the
Wellek et al. [20] and the Bozzano-Dente [25] correlations over-
estimate the mean aspect ratio. Contrarily, the proposed correlation
fits the data fairly well. This suggests that the present experimental
dataset appears to be different from the ones used to develop and
validate the other correlations. Moreover, the present experimental
dataset has data in qualitative agreement with the ones in our previous
study concerning dense bubbly flows in annular gap bubble column [7].
This suggests that correlations obtained for single bubbles/drops
should be carefully evaluated before being applied to dense bubbly
flows.
Fig. 9. Aspect ratio correlation and comparison with the literature.
Eq. (31) has been obtained by considering the whole experimental
dataset obtained for different gas velocities (UG between 0.0037 m/s
and 0.0188 m/s) and at different locations (center and wall of the
column). Applying the same procedure described above to smaller
datasets, we observed that the relationship between φ and the bubble
size does not depend upon UG and the sampling locations (Fig. 10).
This result is interesting and let us conclude the followings:

1. a correlation obtained considering all the bubbles sampled at

different gas superficial velocities and in different sampling positions
can be used without loss of generality;
Fig. 10. Influence of UG and sampling positions on aspect ratio correlation.
2. the relationship between bubble sizes and shapes (in the homo-
geneous flow regime) mainly depends upon the system considered
and not on the operating conditions [7];
3. our correlation may be considered somehow general for dense
bubbly flows in large-scale bubble column; indeed, the diameter of
the column, its height and the sparger openings have been chosen
considering the well-known scale-up criteria for the results: dc >
0.15 m, Hc/dc > 5 and large sparger openings [8,65].

3.3. Optical probe

We have discussed the optical probe measurements in terms of

bubble rise velocity and void fraction profiles in our previous paper
[27] and, therefore, this section focuses on the chord length distribu-
tions.
For the sake of clarity, Fig. 11 displays the radial CLDs
(UG=0.0037 m/s) obtained by the probe. At the center of the column,
the CLDs are broader than the CLDs near the wall. We consider a
center-CLD and a wall-CLD: (i) the center-CLD has been obtained by
summing all the CLDs from deq=45–195 mm; (ii) the wall-CLD has
been obtained by summing the remaining CLDs. The center-CLD and
the wall-CLD have been used as inputs for the “CLD to BSD conversion
algorithm”, whereas the image analysis data have been used for
Fig. 11. Radial chord length distributions - h=1.9 m from the sparger (UG=0.0037 m/s).
validation. For the validation, we have considered UG=0.0037 m/s
and UG=0.0074 m/s, where we have applied the image analysis at both
Fig. 9 compares the experimental data (in terms of average aspect
the wall and at the center of the column.
ratio) against our correlation and two well-known correlations:
• the Bozzano-Dente [25] correlation, which has been obtained for
3.3.1. Algorithm validation
simple bubbles using a total energy minimization method:
Fig. 12 compares the BSDs obtained by the image analysis and the
⎡ ⎛ ⎤− 2
3 “CLD to BSD conversion algorithm” (using Eq. (12) as aspect ratio
1⎞
⎢ 10∙ ⎜1 + 1. 3Mo6 ⎟ +3. 1Eo ⎥ correlation). Qualitatively speaking, the algorithm predicts the BSDs at
⎢ ⎝ ⎠ ⎥
φ=⎢ the center of the column fairly well, but fails near the wall. Tables 2 and
⎛ 1⎞ ⎥
⎢ 10∙ ⎜1 + 1. 3Mo6 ⎟ + Eo ⎥ 3 summarize the performance of the algorithm by comparing the
⎣ ⎝ ⎠ ⎦ (32) Sauter mean diameter (d23) and the mean volume fraction (α) for all

• The Wellek et al. [20] correlation, which has been obtained for single
the bubbles and for the “small”/“large” bubble classes. The small/large
bubble classes have been defined based on change of sign of the lift
drops in liquid:
force at a critical diameter dcr (please refer to the Appendix B): (i)
“small” bubbles (d < dcr) have a positive lift coefficient and (ii) “large”
1 bubbles (d > dcr) have a negative lift coefficient. The comparison of the
φ=
1 + 0.163Eo0.757 (33) results considering “small” and “large” bubbles is interesting because

202
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

Fig. 12. BSDs: comparison between experimental data from image analysis and algorithm.

in the Euler Multi-fluid approach the dispersed phase may be fractions predicted by the algorithm are higher for large bubbles and
implemented as a bi-dispersed phase based on the “small” and “large” lower for small bubbles (the error increases with UG). This result was
bubble classes [4]. The sauter mean diameter and the volume fraction somehow expected because the BSD at the wall are bimodal, whereas,
have been computed as: the method assumes log-normal BSDs. Indeed, the optical probe has
been unable to sample the small bubbles at the wall and the algorithm
ΣiN ni di3
d23 = overestimates the contribution of the large bubbles and underestimates
ΣiN ni di2 (34) the bubbles having deq between 2.5 mm and 5.5 mm. It is worth noting
M that the experimental CLD at the wall is affected by higher uncertain-
∑i =1 ni di3
αi = ties, due to the liquid recirculation near the wall.
N
∑i =1 ni di3 (35) The role of the aspect ratio correlation in the “CLD to BSD
conversion algorithm” has been studied by comparing the results
where di and ni are the diameter and number of bubbles of size class i,
using the Bozzano-Dente [25] and the Wellek et al. [20] correlations
respectively, and N and M are the total number of classes and the
(UG=0.0037 m/s, center of the column). The Bozzano-Dente correla-
classes belonging to the group of interest, respectively.
tion has predicted a BSD shifter toward higher diameters (Fig. 13a),
The Sauter mean diameters have been well predicted both at wall
whereas the correlation of Wellek et al. [20] has predicted a large
and at center of the column. The volume fraction have been well
amount of small bubbles (Fig. 13b). Indeed, for a given major semi
predicted at the center of the column. Contrarily, the algorithm has
axis, a, the Wellek et al. [20] correlation predicted higher aspect ratio
been unable to predict the volume fractions at wall: the volume

Table 2
Comparison between experimental data and algorithm (column center).

UG [m/s] Bubble class d23 [mm] α [−]

Image Analysis Algorithm Error [%] Image Analysis Algorithm Error [%]

0.0037 All bubbles 5,75 5,68 −1,20 1,00 1,00 0,00

Small bubbles 3,91 3,92 0,25 0,40 0,36 −9,41
Large bubbles 8,31 7,58 −8,77 0,60 0,64 6,19

0.0074 All bubbles 5,67 5,32 −6,13 1,00 1,00 0,00

Small bubbles 4,04 3,72 −7,80 0,42 0,41 −3,98
Large bubbles 8,07 7,55 −6,46 0,58 0,59 2,91

203
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

Table 3
Comparison between experimental data and algorithm (wall).

UG [m/s] Bubble class d23 [mm] α [−]

Image Analysis Algorithm Error [%] Image Analysis Algorithm Error [%]

0.0037 All bubbles 4.86 4.56 −6.04 1.00 1.00 0.00

Small bubbles 3.79 3.30 −12.90 0.55 0.51 −8.64
Large bubbles 7.44 7.48 0.48 0.45 0.49 10.68

0.0074 All bubbles 4.41 4.82 9.22 1.00 1.00 0.00

Small bubbles 3.89 3.44 −11.69 0.74 0.47 −36.03
Large bubbles 7.01 7.49 6.88 0.26 0.53 100.04

compared with our correlation (Fig. 9): as consequence, deq and the
pierced chord is higher. Therefore, to obtain the same chord, smaller a
(smaller deq) has to be generated (Fig. 14). Contrarily, the Bozzano-
Dente [25] correlation, at high a, predicted lower aspect ratio
compared with our correlation (Fig. 9): the large bubbles are more
stretched and, therefore, it is easier to obtain a small chord piercing a Fig. 14. Aspect ratio correlation: influence on the algorithm.
large bubble (Fig. 13b and Fig. 14). These results further demonstrates
that the correlations based on single droplet/bubbles are not suitable
for describing the behavior of the homogeneous flow regime and
correlations based on experimental data should be used. It is worth
noting that Al-Oufi [13], in his dissertation, did not presented any
validation of the method and a comparison with our results is not
possible.

3.3.2. Algorithm application

The algorithm provides reliable results for the BSDs at the center of
the column and may be used to obtain BSDs for the gas velocities where
a bubble sampling at the column center has been not possible. Then,
the BSDs representative of the whole system has been obtained by
coupling the results of the “CLD to BSD conversion algorithm”
(information at the center of the column) with the image analysis
sampling (information at the wall of the column). The coupling of these
data has been performed by weighting the center and the wall bubble
frequencies for the cross-sectional area occupied by the two BSDs. The
resulting BSDs are displayed in Fig. 15. In Fig. 15 the BSDs at
UG=0.0037 m/s and UG=0.0074 m/s are the ones obtained weighting
the BSDs obtained by the image analysis.
4. Conclusions
We have presented an experimental study regarding the aspect ratio
and size of bubbles in two-phase bubble columns. In particular, this
Fig. 15. BSDs in the developed region: weighting wall and center column data.
paper studies aspect ratio correlations for bubbly flows and investigates
the role of aspect ratio correlations in bubble chord conversion
algorithms. The main results are summarized below:
• based on visualizations, a correlation between bubble size and aspect
ratio has been proposed and has been shown to be superior to
existing correlations;
• the proposed correlation does not depend on the gas velocities and
sampling position, suggesting that the relationship between bubble
sizes and shapes (in the homogeneous flow regime) mainly depends
upon the system considered and not on the operating conditions;

Fig. 13. Comparison between experimental data and algorithm BSDs: literature correlations.

204
G. Besagni et al. Flow Measurement and Instrumentation 52 (2016) 190–207

• the chord length distributions (CLD) obtained from the optical
probe are converted into bubble size distributions (BSD) through an
improved literature methods; by comparing the BSD from CLD from
probe measurements with the BSD from imaging, we observed that
the BSDs are predicted well at the center of the column then near the
wall within the homogeneous flow regime;
• the aspect ratio correlation derived from the image analysis is of
fundamental importance in the CFD to BSD conversion algorithm:
the proposed approach has been able to reproduce the experimental
BSDs when using the proposed correlation, whereas when using
aspect ratio correlation from the literature, the method performed
poorly.
The proposed methods (CLD conversion using an aspect ratio
correlation based on dense bubbly flows) is interesting to study BSDs
in bubble columns working under relevant operating conditions (i.e.
high pressure and temperature). In addition, this paper demonstrates
how multiple measurement techniques provide a comprehensive view
of the column fluid dynamics: image analysis is used for detailed
analysis on bubble shapes and sizes and the optical probe provides an
insight view in the two-phase flow of the bubble column. Future study
will be minly devoted to extend the present aspect ratio correlation to
other liquid phases.

Appendix A. The transformation method

We consider the random variable ξ and the variable Y of a generic probability density function Pp. The method it is based on the assumption that
a monotone function Y=y(ξ) of the variable ξ has the same cumulative probability function of ξ. This means that the probability of getting an Y value
in the range (Y, Y+dY) is the same as getting an ξ in the range (ξ, ξ+dξ):
dx
Pp (Y ) = P (ξ )
dY (A.1)
In the case considered in this paper, ξ is a uniform deviate density function:
⎧1 0 < ξ < 1
P (ξ ) = ⎨
⎩ 0 otherwise (A.2)
Therefore, from Eq. (A.1) and Eq. (A.2):
dξ
Pp (Y ) = ξ ∈ [0, 1]
dY (A.3)
The solution of Eq. (A.3) is:
Y (ξ )
x = Cp (Y (ξ )) = ∫0 Pp (Y ) dY
(A.4)
where Cp(Y(ξ)) is the cumulative probability of Y and its inverse function reads as:
Y (ξ ) = CY −1 (ξ ) (A.5)
Eq. (A.5) states that it is possible to pass from a random variable ξ (uniform deviate density function, valid within the boundaries [0,1]) and the
random variable Y(ξ), having a probability density function Pp(Y).

Appendix B. Lift force

The lift force acts perpendicularly to the bubble itself and arises from the net effect of pressure and stress acting on the bubble surface:
⎯→
⎯
FL, j = −CL αj ρk (→
uj − →
uk ) × (∇ × →
uk ) (B.1)
where CL is the lift coefficient: its sign is positive for “small bubble” and is negative for “large bubble”. For this reason, the direction of the lift force
depends upon the bubble size and shape. For “small bubbles”, the lift force acts in the direction of decreasing liquid velocity (i.e. in case of batch or
co-current mode in the direction toward the pipe wall), whilst for “large bubbles” it changes direction (a force that can be assimilated to the lift force
tends to push large and deformed bubbles towards the center of the column [46,66]). In counter-current mode, the behavior is the opposite: the
“small bubbles” migrate toward the center of the column. This is also confirmed by the DNS studies of Lu et al. [67] and Lu and Tryggvason [68,69].
The change of sign of the lift coefficient is well described using the model of Tomiyama et al. [46]:
⎧ min[0.288 tanh(0.121 Re), f (Eo⊥)] Eo⊥ ≤ 4
⎪
CL = ⎨ f (Eo⊥) 4 < Eo⊥ ≤ 10
⎪− 0.27
⎩ 10 < Eo⊥ (B.2)
where:
f (Eo⊥) = 0.00105Eo⊥3 − 0.0159Eo⊥2 − 0.0204Eo⊥ + 0.474 (B.3)
Eo⊥ is the Eötvös number considering the maximum horizontal dimension of the bubble d⊥, given by the empirical correlation for the aspect
ratio by Wellek et al. [20]:
d⊥ = deq (1 + 0.163Eo0.757)1/3 (B.4)
For the air-water system at ambient conditions, the bubble diameter at which the change in sign occurs is dcr=5.8 mm. Therefore, two bubble
groups can be considered: one of the bubble classes represents small and nearly spherical bubbles with a positive CL (“small bubbles”), while the

205
G. Besagni et al.
other one models large and deformed bubbles with a negative CL (“large bubbles”).
References
[1] A. Youssef Ahmed, H. Al-Dahhan Muthanna, P. Dudukovic Milorad, Bubble
columns with internals: a review, Int. J. Chem. React. Eng. (2013) 169.
[2] R. Rzehak, Modeling of mass-transfer in bubbly flows encompassing different
mechanisms, Chem. Eng. Sci. 151 (2016) 139–143.
[3] G. Besagni, F. Inzoli, G. De Guido, L.A. Pellegrini, Experimental investigation on
the influence of ethanol on bubble column hydrodynamics, Chem. Eng. Res. Des.
112 (2016) 1–15.
[4] G. Besagni, G.R. Guédon, F. Inzoli, Annular gap bubble column: experimental
investigation and computational fluid dynamics modeling, J. Fluids Eng. 138 (1)
(2016) (011302-011302).
[5] S. Sharaf, M. Zednikova, M.C. Ruzicka, B.J. Azzopardi, Global and local hydro-
dynamics of bubble columns – Effect of gas distributor, Chem. Eng. J. 288 (2015)
489–504.
[6] C. Boyer, A.-M. Duquenne, G. Wild, Measuring techniques in gas–liquid and gas–
liquid–solid reactors, Chem. Eng. Sci. 57 (16) (2002) 3185–3215.
[7] G. Besagni, F. Inzoli, Bubble size distributions and shapes in annular gap Bubble
column, Exp. Therm. Fluid Sci. 74 (2016) 27–48.
[8] C. Leonard, J.H. Ferrasse, O. Boutin, S. Lefevre, A. Viand, Bubble column reactors
for high pressures and high temperatures operation, Chem. Eng. Res. Des. 100
(2015) 391–421.
[9] P. Rollbusch, M. Bothe, M. Becker, M. Ludwig, M. Grünewald, M. Schlüter,
R. Franke, Bubble columns operated under industrially relevant conditions –
Current understanding of design parameters, Chem. Eng. Sci. 126 (2015) 660–678.
[10] N.N. Clark, R. Turton, Chord length distributions related to bubble size distribu-
tions in multiphase flows, Int. J. Multiph. Flow. 14 (4) (1988) 413–424.
[11] W. Liu, N.N. Clark, Relationships between distributions of chord lengths and
distributions of bubble sizes including their statistical parameters, Int. J. Multiph.
Flow. 21 (1995) 1073–1089.
[12] B. Hu, P. Angeli, O.K. Matar, C.J. Lawrence, G.F. Hewitt, Evaluation of drop size
distribution from chord length measurements, AIChE J. 52 (3) (2006) 931–939.
[13] F.M.Al-Oufi, An Investigation of Gas Void Fraction and Transition Conditions For
Two-phase Flow in an Annular Gap Bubble Column, (Ph.D. thesis), 2011.
[14] M. Rüdisüli, T.J. Schildhauer, S.M. a Biollaz, J.R.V. Ommen, Monte Carlo
simulation of the bubble size distribution in a fluidized bed with intrusive probes,
Int. J. Multiph. Flow. 44 (2012) 1–14.
[15] N.H. Hoang, D.J. Euh, B.J. Yun, C.H. Song, A new method of relating a chord
length distribution to a bubble size distribution for vertical bubbly flows, Int. J.
Multiph. Flow. 71 (2015) 23–31.
[16] G.P. Celata, M. Cumo, F. D’Annibale, P. Di Marco, A. Tomiyama, C. Zovini, Effect of
gas injection mode and purity of liquid on bubble rising in two-component systems,
Exp. Therm. Fluid Sci. 31 (1) (2006) 37–53.
[17] G.P. Celata, M. Cumo, F. D'Annibale, A. Tomiyama, The wake effect on bubble
rising velocity in one-component systems, Int. J. Multiph. Flow. 30 (7–8) (2004)
939–961.
[18] G.P. Celata, F. D’Annibale, P. Di Marco, G. Memoli, A. Tomiyama, Measurements of
rising velocity of a small bubble in a stagnant fluid in one- and two-component
systems, Exp. Therm. Fluid Sci. 31 (6) (2007) 609–623.
[19] T. Okawa, T. Tanaka, I. Kataoka, M. Mori, Temperature effect on single bubble rise
characteristics in stagnant distilled water, Int. J. Heat Mass Transf. 46 (5) (2003)
903–913.
[20] R.M. Wellek, A.K. Agrawal, A.H.P. Skelland, Shape of liquid drops moving in liquid
media, AIChE J. 12 (5) (1966) 854–862.
[21] D.W. Moore, The rise of a gas bubble in a viscous liquid, J. Fluid Mech. 6 (01)
(1959) 113–130.
[22] T.D. Taylor, A. Acrivos, On the deformation and drag of a falling viscous drop at low
Reynolds number, J. Fluid Mech. 18 (03) (1964) 466–476.
[23] T. Tadaki, S. Maeda, On the shape and velocity of single air bubbles rising in
various liquids, Kagaku Kogaku 25 (1961) 254–264.
[24] R.Clift, J.R.Grace, M.E.Weber, Bubbles, Drops, and Particles, New York, 1978.
[25] G. Bozzano, M. Dente, Shape and terminal velocity of single bubble motion: a novel
approach, Comput. Chem. Eng. 25 (4–6) (2001) 571–576.
[26] S. Aoyama, K. Hayashi, S. Hosokawa, A. Tomiyama, Shapes of ellipsoidal bubbles in
infinite stagnant liquids, Int. J. Multiph. Flow 79 (2016) 23–30.
[27] G. Besagni, F. Inzoli, Comprehensive experimental investigation of counter-current
bubble column hydrodynamics: holdup, flow regime transition, bubble size
distributions and local flow properties, Chem. Eng. Sci. 146 (2016) 259–290.
[28] I. Reilly, D. Scott, T. Debruijn, D. MacIntyre, The role of gas phase momentum in
determining gas holdup and hydrodynamic flow regimes in bubble column
operations, Can. J. Chem. Eng. 72 (1) (1994) 3–12.
[29] R. Krishna, P.M. Wilkinson, L.L. Van Dierendonck, A model for gas holdup in
bubble columns incorporating the influence of gas density on flow regime
transitions, Chem. Eng. Sci. 46 (10) (1991) 2491–2496.
[30] S. Nedeltchev, A. Shaikh, A new method for identification of the main transition
velocities in multiphase reactors based on information entropy theory, Chem. Eng.
Sci. 100 (0) (2013) 2–14.
[31] S. Nedeltchev, New methods for flow regime identification in bubble columns and
fluidized beds, Chem. Eng. Sci. 137 (2015) 436–446.
[32] N. Zuber, J.A. Findlay, Average volumetric concentration in two-phase flow
systems, J. Heat. Transf. 87 (4) (1965) 453–468.
206
Flow Measurement and Instrumentation 52 (2016) 190–207
[33] G.B.Wallis, One-dimensional Two-phase Flow, New York, 1969.
[34] G. Besagni, F. Inzoli, Influence of internals on counter-current bubble column
hydrodynamics: holdup, flow regime transition and local flow properties, Chem.
Eng. Sci. 145 (2016) 162–180.
[35] R. Krishna, M.I. Urseanu, A.J. Dreher, Gas hold-up in bubble columns: influence of
alcohol addition versus operation at elevated pressures, Chem. Eng. Process.
Process Intensif. 39 (4) (2000) 371–378.
[36] J.F. Richardson, W.N. Zaki, Sedimentation and fluidisation: Part I, Chem. Eng. Res.
Des. 75 (Suppl (0)) (1997) S82–S100.
[37] K. Ellingsen, F. Risso, On the rise of an ellipsoidal bubble in water: oscillatory paths
and liquid-induced velocity, J. Fluid Mech. 440 (2001) 235–268.
[38] G. Besagni, Bubble column fluid dynamics: experimental and numerical investi-
gations, Politec. di Milano (2016).
[39] Y. Bian, F. Dong, W. Zhang, H. Wang, C. Tan, Z. Zhang, 3D reconstruction of single
rising bubble in water using digital image processing and characteristic matrix,
Particuology 11 (2) (2013) 170–183.
[40] D.L. Sahagian, A.A. Proussevitch, 3D particle size distributions from 2D observa-
tions: stereology for natural applications, J. Volcanol. Geotherm. Res. 84 (3–4)
(1998) 173–196.
[41] Y.M. Lau, N.G. Deen, J.A.M. Kuipers, Development of an image measurement
technique for size distribution in dense bubbly flows, Chem. Eng. Sci. 94 (0) (2013)
20–29.
[42] Y.M. Lau, K.T. Sujatha, M. Gaeini, N.G. Deen, J.A.M. Kuipers, Experimental study
of the bubble size distribution in a pseudo-2D bubble column, Chem. Eng. Sci. 98
(2013) 203–211.
[43] A. Karn, C. Ellis, R. Arndt, J. Hong, An integrative image measurement technique
for dense bubbly flows with a wide size distribution, Chem. Eng. Sci. 122 (0) (2015)
240–249.
[44] P.L.C. Lage, R.O. Espósito, Experimental determination of bubble size distributions
in bubble columns: prediction of mean bubble diameter and gas hold up, Powder
Technol. 101 (2) (1999) 142–150.
[45] P. Wongsuchoto, T. Charinpanitkul, P. Pavasant, Bubble size distribution and gas–
liquid mass transfer in airlift contactors, Chem. Eng. J. 92 (1–3) (2003) 81–90.
[46] A. Tomiyama, H. Tamai, I. Zun, S. Hosokawa, Transverse migration of single
bubbles in simple shear flows, Chem. Eng. Sci. 57 (11) (2002) 1849–1858.
[47] M. Honkanen, P. Saarenrinne, T. Stoor, J. Niinimäki, Recognition of highly
overlapping ellipse-like bubble images, Meas. Sci. Technol. 16 (9) (2005) 1760.
[48] R. Rakoczy, S. Masiuk, Experimental study of bubble size distribution in a liquid
column exposed to a rotating magnetic field, Chem. Eng. Process. Process Intensif.
48 (7) (2009) 1229–1240.
[49] W. Hanselmann, E. Windhab, Flow characteristics and modelling of foam
generation in a continuous rotor/stator mixer, J. Food Eng. 38 (4) (1998) 393–405.
[50] F.M. Aloufi, An investigation of gas void fraction and transition conditions for two-
phase flow in an annular gap bubble column, Loughborough University, UK, 2011.
[51] A.D. Passos, V.P. Voulgaropoulos, S.V. Paras, A.A. Mouza, The effect of surfactant
addition on the performance of a bubble column containing a non-Newtonian
liquid, Chem. Eng. Res. Des. 95 (2015) 93–104.
[52] R.M. Boes, W.H. Hager, Fiber-optical experimentation in two-phase cascade flow,
in: K. Hansen (Ed.), Proceedings of the International RCC Dams Seminar, Denver,
USA, 1998.
[53] S.L. Kiambi, A.-M. Duquenne, J.B. Dupont, C. Colin, F. Risso, H. Delmas,
Measurements of bubble characteristics: comparison between double optical probe
and imaging, Can. J. Chem. Eng. 81 (3–4) (2003) 764–770.
[54] H. Chaumat, A.M. Billet-Duquenne, F. Augier, C. Mathieu, H. Delmas, Application
of the double optic probe technique to distorted tumbling bubbles in aqueous or
organic liquid, Chem. Eng. Sci. 60 (22) (2005) 6134–6145.
[55] M. Simonnet, C. Gentric, E. Olmos, N. Midoux, Experimental determination of the
drag coefficient in a swarm of bubbles, Chem. Eng. Sci. 62 (3) (2007) 858–866.
[56] I. Lima Neto, D. Zhu, N. Rajaratnam, Air Injection in Water with Different Nozzles,
J. Environ. Eng. 134 (4) (2008) 283–294.
[57] W. Zhang, D.Z. Zhu, Bubble characteristics of air–water bubbly jets in crossflow,
Int. J. Multiph. Flow 55 (2013) 156–171.
[58] G. Besagni, F. Inzoli, Influence of electrolyte concentration on holdup, flow regime
transition and local flow properties in a large scale bubble column, J. Phys. Conf.
Ser. 655 (1) (2015) 012039.
[59] E. Barrau, N. Rivière, C. Poupot, A. Cartellier, Single and double optical probes in
air-water two-phase flows: real time signal processing and sensor performance, Int.
J. Multiph. Flow 25 (2) (1999) 229–256.
[60] K.-A. Chang, H.-J. Lim, C.B. Su, Fiber optic reflectometer for velocity and fraction
ratio measurements in multiphase flows, Rev. Sci. Instrum. 74 (7) (2003)
3559–3565.
[61] J. Vejražka, M. Večeř, S. Orvalho, P. Sechet, M.C. Ruzicka, A. Cartellier,
Measurement accuracy of a mono-fiber optical probe in a bubbly flow, Int. J.
Multiph. Flow 36 (7) (2010) 533–548.
[62] A. Cartellier, E. Barrau, Monofiber optical probes for gas detection and gas velocity
measurements: conical probes, Int. J. Multiph. Flow 24 (8) (1998) 1265–1294.
[63] W.H. Press, B.P. Flannery, S.A. Teukolsky, W.T. Vetterling, Numerical Recipes in
FORTRAN: The Art of Scientific Computing, Cambridge University Press, 1992.
[64] M. Ishii, G. Kojasoy, Interfacial Area Transport Equation and Preliminary
Considerations For Closure Relations, Technical Report, PU-NE-93/6, Nuclear
Engineering Department, Purdue University West Lafayette, IN 1993.
[65] N. Kantarci, F. Borak, K.O. Ulgen, Bubble column reactors, Process Biochem. 40 (7)
G. Besagni et al.
(2005) 2263–2283.
[66] D. Lucas, H.M. Prasser, A. Manera, Influence of the lift force on the stability of a
bubble column, Chem. Eng. Sci. 60 (13) (2005) 3609–3619.
[67] J. Lu, S. Biswas, G. Tryggvason, A. DNS, Study of laminar bubbly flows in a vertical
channel, Int. J. Multiph. Flow 32 (6) (2006) 643–660.
207
Flow Measurement and Instrumentation 52 (2016) 190–207
[68] J. Lu, G. Tryggvason, Effect of bubble size in turbulent bubbly downflow in a
vertical channel, Chem. Eng. Sci. 62 (11) (2007) 3008–3018.
[69] J. Lu, G. Tryggvason, Dynamics of nearly spherical bubbles in a turbulent channel
upflow, J. Fluid Mech. 732 (2013) 166–189.