The lifespan of clusters in confined bubbly liquids

B. Figueroa-Espinozaa , B. Menab , A. Aguilar-Coronac , R. Zenitd,∗

a
Laboratorio de Ingenierı́a y Procesos Costeros, Instituto de Ingenierı́a, Universidad
Nacional Autónoma de México, Sisal, Yucatán, 97835, México
b
Instituto de Ingenierı́a, Universidad Nacional Autónoma de México, Ciudad de México,
04510, México
c
Facultad de Ingenierı́a Mecánica, Universidad Michoacana de San Nicolás de Hidalgo,
Morelia, Michoacán, 58030, México
d
Instituto de Investigaciones en Materiales, Universidad Nacional Autónoma de México,
Ciudad de México, 04510, México

Abstract

A nearly monodispersed bubbly flow in a Hele-Shaw type channel was stud-

ied, for large but finite Reynolds numbers. Experiments were carried out
with millimetre-sized Nitrogen bubbles rising through water-glycerin mix-
tures.The velocities, geometry and other flow characteristics were measured
by means of a high speed camera and an image processing routine, in order to
validate this experimental set-up with previous results from the literature and
shed some light on the cluster formation, timespan and breakup mechanisms.
The timespan of clusters was estimated by tracing the clusters trajectories.
The results show that clustering occurs, but the bubble aggregates breakup
rapidly due to hydrodynamic effects. Scaling arguments were used to obtain
an estimate for the lifespan and compared to experiments. This estimate
does not take into account wake effects, so it can be interpreted as a lower
bound of clusters lifespan in an homogeneous bubbly flow.

∗
Corresponding author
Email address: zenit@unam.mx (R. Zenit)

Preprint submitted to International Journal of Multiphase Flow November 7, 2017

Keywords: clusters lifespan, clustering, bubbly flow, velocity fluctuations

1. Introduction

If anyone observes a single millimetric air bubble rising through quiescent

water, one notices that its trajectory is a straight line (for bubbles of diame-
ter smaller than one millimeter); however, if a pair of such bubbles rise close
to each other, their trajectories follow complicated paths, due to the hydro-
dynamic interaction between bubbles and the (initially still) liquid. For the
case of the dual limit of large Reynolds and small Weber numbers, potential
flow theory can be used to study hydrodynamic interactions; Biesheuvel and
Van Wijngaarden (1982), Kok (1989) Kumaran and Koch (1993a,b) analyzed
the motion of two bubbles ascending in still liquid and carried out the cal-
culation of the bubble trajectories, which were then compared with careful
experiments in the laboratory. The actual trajectories were fairly well pre-
dicted by the theory, based on velocities calculated from a potential and a set
of dynamical equations for the relative velocities and the center of mass of
the two bubbles array. These trajectories are complex, but in general terms,
the bubbles tend to come together and collide along the line perpendicular to
gravity if their initial orientation is not close to a vertical alignment between
bubbles centres. One would expect, based on the aforementioned argument,
that bubbles in an initially homogeneous bubbly flow would tend to form hor-
izontally oriented clusters. This clustering phenomenon was also predicted by
numerical simulations (Sangani and Didwania, 1993; Smereka, 1993), where
potential flow hydrodynamic interactions between (many) bubbles lead to
the formation of horizontal bubble clusters (in absence of coalescence).

2
 In the case of a cloud of bubbles, the motion of other bubbles generates
forces (that depend on the relative positions and velocities) on each test
bubble. Moreover, these forces cause deformations on the bubbles surfaces,
which are also coupled to the liquid. This interaction between the flow field,
forces and deformations is hard, if not impossible to describe mathematically
in a deterministic manner. This situation lead Van Wijngaarden (1993) to
treat the problem with stochastic tools, based on a set of simplified equations
of motion for a pair of bubbles that resulted from the work of Kok (1989).
These simplifications served the purpose of calculating an ensemble average
for the velocity hvi:

hvi = U∞ {1 − 1.56α + O(α2) + ...} (1)

where hvi is the mean rise velocity of the bubbles and U∞ is the terminal
rise velocity of a free bubble. Equation (1) states that at higher volume
fractions the mean bubble velocity decreases. This result agrees with the
actual qualitative behavior of bubble suspensions (Van Wijngaarden and
Kapteijn, 1990; Clift et al., 1978; Zenit et al., 2001).
Because of linearity, binary hydrodynamic interactions can also be used
to calculate properties and averaged equations for bubbly liquids, as was
done by Spelt and Sangani (1998), who determined average properties in
the limit of large Reynolds and small Weber numbers, as function of the
volume fraction, such as mean relative velocity and velocity variance of the
bubbles, using numerical simulations and pair interaction theory. Kok (1993),
also calculated the mean-square fluctuating velocity in a uniform suspension
and the hydrodynamic diffusivities in a nonuniform one, by performing an

3
ensemble average over pair interactions. This pair averaging procedure is
valid for dilute suspensions (α ≪ 18/Re). These statistical techniques are a
powerful tool for modeling complex systems like multiphase flows.
Experimental measurements of bubble clusters are difficult because of the
superposition of bubbles when observing a bubble cloud, as well as the cluster
breakup due to motion-induced agitation. Nevertheless, some observations
of clustering have been reported in 3D experiments by Zenit et al. (2001)
and Figueroa-Espinoza and Zenit (2005), the latter being able to obtain
statistical evidence of horizontal clustering in a thin channel. The problem
with the 2D viewpoint is that the wall effects are strong, and may themselves
cause clustering, since when a bubble rises near a wall its velocity decreases
and a transverse lift force tends to attract it towards the wall, causing it
to bounce (Figueroa-Espinoza et al., 2008), blocking the trailing bubbles
motion and (potentially) causing them to form clusters. A similar effect
has been observed near a wall in the presence of surfactants (Takagi et al.,
2009). Nevertheless, if one compares the radial distribution functions for the
experiments in Figueroa-Espinoza and Zenit (2005) with those coming from
3D numerical simulations (Bunner and Tryggvason, 2002a), the qualitative
agreement is reasonable.
Bubble clusters may cause the whole flow field to behave differently, with
respect to an uniformly distributed suspension of bubbles. Clusters could
influence the surrounding flow field (Salesse et al., 2002), introducing en-
hanced velocity fluctuations and hydrodynamic interactions that could affect
the overall flow structure. One possible mechanism of this phenomenon is
depicted in Fig. 1; to the left, a uniform bubbly flow is shown. The wakes

4
and velocity differences between these individual bubbles are not very signif-
icant. On the other hand, if the bubbles are clustered, these agglomerations
could act as single large objects, as in the rightmost image of the same figure,
and the wakes and velocity fluctuations could be very important (Van Wi-
jngaarden, 2005), leading to a different behaviour than the one predicted
by the theory, based on the uniformity assumption. The way clusters inter-
act with the overall flow is not yet well understood, but the importance of
certain flow parameters on their formation and dynamics has already been
identified: the effect of the volume fraction, α, on the added mass coefficient
was first theoretically calculated by Van Wijngaarden (1976) for uniformly
distributed bubbly flows. Spelt and Sangani (1998) noted that the added
mass coefficient for clustered bubbly liquids was much greater than the theo-
retical predictions for uniformly distributed bubbles. They also showed that
the tendency to form aggregates is diminished when the velocity fluctuations
in the bubbly liquids are increased. The formation of clusters was not ob-
served for sufficiently low values of the bubbly turbulent intensity (the ratio
between the magnitude of the mean relative velocity of the bubbles and their
root-mean-squared velocity). If the bubble aggregates contribution to the
velocity fluctuations is important, then they may contribute to their own
breakup, so it would be reasonable to expect their lifespan to be very short.
This was observed by Bunner and Tryggvason (2002a).
The reason for focusing our attention on velocity fluctuations are numer-
ous: whenever an attempt is made of obtaining averaged equations of motion
for a bubbly flow, just as in the case of turbulent flows (Van Wijngaarden,
1997), some terms appear, (equivalent to the Reynolds stresses in turbulence

5
models) that depend on velocity fluctuations. These terms are related to the
transport of momentum due to fluctuations in the velocity of bubbles, col-
lisions, and hydrodynamic interactions. Since these averaged equations are
usually developed from the assumption that the velocity field is the sum of a
deterministic part plus a stochastic part, the structure of the latter is of in-
terest in many situations. There is a previous work which is closely related to
our investigation: Zenit et al. (2001) measured the liquid and bubble veloc-
ities for a bubble monodispersed suspension using a dual impedance probe.
They also calculated a first approximation for the liquid velocity variance,
using the potential of a single bubble rising through a liquid at rest. By
calculating the squared velocity disturbance on a single bubble, multiplying
it by the number density and integrating over all the space, they obtained:

1
hu2x i = αu2b (2)
5
3
hu2y i = αu2b (3)
20

where ub is the bubble velocity relative to the liquid for a given α.

There are other attempts to evaluate the velocity fluctuations, in order to
obtain more accurate estimates for the quantities involved in calculations for
averaged equations. Van Wijngaarden (1997) also calculated expressions for
the excess turbulent energy caused by the presence of the bubbles, and also
found it to be dependent on αu2b . These quantities have units of energy, and
usually are the only quantities that can be obtained in turbulence related
experiments.
More recently, bubble-induced agitation motivated some experimental in-
vestigations (Martinez-Mercado et al., 2007; Mercado et al., 2013; Riboux

6
et al., 2009) on the properties, scaling and interaction between the gas vol-
ume fraction, velocity PDF, liquid velocity fluctuations and observed energy
spectrum, in order to shed some light on the origin of such enhanced fluc-
tuations; Riboux et al. (2009) carried out an experiment with homogeneous
bubbly flows with different bubble sizes and volume fractions, being able to
obtain bubble mean velocity, liquid velocity fluctuations and Energy Spec-
trum using optical probes and Laser Doppler anemometry behind homoge-
neous bubble swarms. The velocity fluctuations resulted to scale as α0.4 ,
where α is the gas volume fraction. The Energy Spectrum follows a power
law in terms of the wavelength close to 3, consistent with other studies re-
lated to bubble induced turbulence (Lance and Bataille, 1991; Mendez et al.,
2013). Interestingly, later numerical simulations (Riboux et al., 2008) (DNS),
where the bubbles were modelled by fixed momentum forces of finite size
randomly distributed in an uniform flow (for a volume fraction in the range
0.06 < α < 0.04), reproduced the velocity fluctuations and Energy Spectrum
scaling of the experiments, suggesting that the wakes play a very important
role in bubble-induced turbulence (Riboux et al., 2008). This set-up of fixed
objects (a random distribution of spheres in a channel) immersed in a uniform
flow was carried out experimentally by Risso et al. (2008); this configuration
allowed for both temporal and spacial averaging, which showed that Large-
Reynolds-number bubble-induced agitation was driven by (large scale) wake
interactions. Interest in explaining theoretically the aforementioned scaling
has motivated some new models based on potential flow (Eames et al., 2004;
Roig, 2007), as well as stochastic considerations (Risso, 2011, 2016).
The objective of this investigation is to contribute to the understanding

7
Figure 1: Sketch of clusters acting as single large objects, generating large velocity fluc-
tuations in the flow.

of bubbly flow hydrodynamics, by comparing theory and experiments car-

ried out using optical measurements and digital image processing techniques.
Some experiments from Figueroa-Espinoza and Zenit (2005) were revisited;
this particular experimental setup allows for the identification of individual
clusters, as well as the measurement of their characteristic parameters, such
as size, mean rise velocity, and the added mass effect (mean cluster veloc-
ity). Additionally, some statistical properties of the liquid phase such as the
mean and fluctuating velocity were obtained using Particle Image Velocime-
try (PIV). Particular attention was put on the clusters lifespan, which, by
means of a scaling argument based on an analogy to the mean free path of
statistical mechanics, was estimated and compared with the measurements
as discussed below.

8
2. Materials and Methods

In this investigation two experimental techniques were used: first, bubble

position and geometry were obtained by means of a digital high speed camera
and image processing, and secondly, the fluid mean and fluctuating velocity
were obtained using particle image velocimetry (PIV).
The experimental setup used in this investigation is shown in Fig.2. A
thin channel of 20cm width and 120cm height was fabricated with two glass
sheets, separated by two glass strips of 3mm thickness. Nitrogen millimetric
bubbles were produced using an array of 10 identical capillaries, connected to
a chamber where pressurized gas was fed through a valve at the channel base.
With this array a nearly mono-dispersed stream of bubbles was produced.
More details of the experiment can be found in Figueroa-Espinoza and Zenit
(2005).
The gas volume fraction α is calculated as follows:

Vg ∆H
α= = (4)
VT H

where VT and Vg are the total volume of the mixture and the gaseous phase
respectively, H is the level of liquid in the channel, and ∆H is the increase
in height. Since the channel is very narrow, a small volume difference causes
the level to increase considerably.
Mixtures of deionized water and glycerine were used in order to vary the
liquid properties. In order to reduce coalesce, a small amount of MgSO4 was
dissolved in the mixture following Zenit et al. (2001). Since contamination
may have an effect on the experiment outcome, the liquid resistivity was
monitored: it was always kept larger than 11MΩ/cm (table 1 ).

9
Figure 2: Experimental setup. The thin channel was built with two glass plates separated
by two glass strips. Nitrogen is injected through the capillary bank to produce the bubbles

Experiments were carried out with water, water-glycerine 15% (glyc15

in what follows) and water-glycerine 30% (glyc30). A few tests were also
conducted in a water-glycerine 50% mixture (glyc50).

2.1. Digital Image Analysis

A high speed video camera was used to capture images of the bubbly
c
flow at 150 fps. An image processing routine in matlab was implemented in
order to measure the bubbles geometry, as well as their velocity and position.
Digital treatment of the images allowed for the identification of clusters: a
distance criteria was applied and then neighboring bubbles that complied

10
 % Weight ρ(kg/m3 ) µ(×10−3 P as)
0% 9.98230 1.005
15% 1.03450 1.53
30% 1.07270 2.5
50% 1.12630 6.0

Table 1: Fluid properties for different Glycerine percentage in weight at 20o C.

with this condition were grouped to form individual clusters, as explained

in Figueroa-Espinoza and Zenit (2005). In the present study the threshold
distance was two pixels in order to measure clusters size more precisely.
Figure 3 shows a typical digitally treated image. To the left in (a) the digital
image processing routine was used to identify (and label) individual bubbles;
to the right (b) individual clusters were identified and labeled, for the same
image. The position of their centroids, their area and aspect ratio were
recorded, and velocities were calculated using the frame rate of acquisition
and calibrated scale images in the horizontal and vertical directions. Note
that a dark area without objects is present in the images (at the bottom).
This was a mask used to block a region where the back-light was not uniform,
in order to avoid false objects detection during the image processing routine.
A routine to identify individual objects in subsequent frames was devel-
oped in order to follow trajectories of bubbles and clusters. Clusters were
considered to ”die” or break when their area (and velocity) changed sub-
stantially during one time step. When bubbles separate from the cluster the
position of the centroid changes abruptly, and the same thing happens when
the cluster leaves the field of view (FOV).

11
Figure 3: Binarized image showing detected objects; (a) individual bubbles, (b) clusters.
Both views came from the same frame and experiment: water-glycerine 15%, α = 0.018

Figure 4 shows a set of trajectories, displaced to have the same origin,

so the dispersion can be observed as in a bubble plume: to the left in a)
individual bubbles and b) clusters. It is clear that lifespan of clusters is
shorter compared with single bubbles, which can be detected crossing the
entire field of view (of about 20 cm × 30 cm). Some individual bubbles have
shorter trajectories because some leave the FOV at the borders, and some
other are lost because they aggregate forming new clusters.

2.2. Liquid velocity fluctuations

Liquid velocity fluctuations were obtained using a PIV system. The laser
sheet was placed at the middle plane between the two glass sheets that form

12
 250 250

200 200

150 150

100 100

50 50

0 0
-40 0 40 -20 0 20 40
(a) (b)

Figure 4: Trajectories of detected objects through the camera field of view; (a) individual
bubbles, (b) clusters for water-glycerine 15%, α = 0.018. Axis labels in millimeters.

the channel walls. This is the plane where the dominant components of the
motion are located. We used fluorescent particles in order to avoid reflections
from the bubbles surface. After an adaptive correlation and filtering, the
velocity vectors were obtained, and the variances were calculated in a 60 mm
× 60 mm area of interest approximately, located at equal distances from the
channel edges. The averaging time was one minute, giving 960 points per
frame. The corresponding results will be discussed in section 3.1.

13
3. Results

The velocity probability distributions were obtained from a large set of

bubbles for each volume fraction as shown in Figueroa-Espinoza and Zenit
(2005). PDF’s of the bubble velocity are not shown here for brevity, but they
fit normal distributions well, with mean velocities decreasing with α. Figure
5 shows the mean velocities for all liquids, showing both individual bubbles
and clusters. As expected, velocity decreases with volume fraction, and the
cluster velocity is always smaller than single bubbles velocity, confirming an
added mass effect. The dashed line represents the theoretical prediction of
Van Wijngaarden (1976), given by equation (1). Theoretical considerations
lead to the correct trend for low bubble fractions, even if the flow is confined
between two walls. Note that it is not easy to produce the same bubble
size as the volume fraction increases; in our case the bubble size increased
slightly with α, so the curves may present some variations on the slope. The
experiments represented in Figure 5 presented a mean diameter variation of
approximately 15% for water, 7% for glyc15 and 18% for glyc30.
Bubbles and clusters velocities are also represented in a dimentionless
form as an equivalent Reynolds number in terms of the volume fraction α in
Fig. 6. This Reynolds was calculated with the equivalent diameter Deq and
the average bubble rise velocity ub , i.e. Reeq = ub Deq ρ/µ. In the Figure, filled
markers correspond to clusters while empty markers represent bubbles mov-
ing in different liquids: water (circles), glyc15 (diamonds), glyc30 (squares)
and glyc50 (stars).
The equivalent Reynolds number seems to decrease with α for low volume
fractions. This happens because of the aforementioned added mass effect. For

14
 0.2

0.19

0.18
terminal velocity (cm/s)

0.17

0.16

0.15

0.14

0.13

0.12

0.11

0.1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
α

Figure 5: Bubble and cluster velocity for different values of the gas volume fraction. The
markers represent: water bubbles (crosses), water clusters (asterisk), water-glycerine 15%
bubbles (empty squares), water-glycerine 15% clusters (filled squares), water-glycerine
30% bubbles (empty diamonds), and water-glycerine 30% clusters (filled diamonds). The
dashed line corresponds to equation (1)

low α the bubble size remains relatively constant; hydrodynamic interactions

cause the velocity to decrease. As the volume fraction further increases,
bubble diameter grows causing the terminal velocity to increase, and as a
result the equivalent Reynolds increases for larger values of α. It can be seen
that the clusters are always slower than individual bubbles in all cases.
As the volume fraction increases, the mean clusters area also grows. Fig-
ure 7 shows the clusters mean area Ac , normalized with respect to the mean
bubble area Ab for each volume fraction, in a PDF representation. In gen-
eral, most clusters are formed by two bubbles, and it is less likely to find

15
 700

600

500

400
Reeq

300

200

100

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
α

Figure 6: Equivalent Reynolds number for bubbles and clusters. Filled markers correspond
to clusters while empty markers represent bubbles moving in: water (circles), glyc15 (di-
amonds), glyc30 (squares) and glyc50 (stars).

clusters formed by many bubbles. As α grows, the mean number of bubbles

in each cluster increases slightly. Some peaks may appear in the vicinity of
Ac /Ab = 3 (or more, if α is further increased). The experimental conditions
did not allow us to test this tendency for volume fractions higher than 0.13.

16
Figure 7: PDF of clusters normalized area with respect to the mean bubble area Ab for
water

3.1. Velocity Fluctuations

The variance of the liquid velocity for approximately 960 points in the
frame was obtained as described in section 2.2. The spatial averages are
shown in Figure 8; the vertical velocity variance < u′2 2
y > /ub is shown in

(a), and the horizontal component < u′2 2

x > /ub in (b). The vertical axis

represents the velocity fluctuations for different values of the gas volume
fraction, normalized with the mean terminal velocity squared. The vertical
and horizontal velocity fluctuations are much greater than the theoretical
lower bound from equations (2) and (3). Also note that the vertical fluctua-
tions are bigger than the horizontal ones by a factor of two. The dotted line

17
in the figure represents the scaling law hu′2 2 0.8 ′
y i/ub ∝ α , where uy is the verti-

cal component of the liquid velocity fluctuation and ub is the average bubble
velocity for that gas volume fraction. This is consistent with the results in
the literature (Riboux et al., 2009, 2008; Risso et al., 2008).

(a)
0
10

−1
10
2
b
<u ’ >/u
2
y

−2
10

−3
10 −2 −1
10 10

−1
(b)
10
2
b
<u >/u

−2
10
2
x

−3
10 −2 −1
10 10
α

Figure 8: Normalized liquid velocity fluctuations as a function of gas volume fraction:

(a) vertical and (b) horizontal components. Crosses: water, squares: glyc15, diamonds:
glyc30, stars: glyc50. The dotted line is a power law proportional to α0.8 .

18
 The bubbles velocity fluctuations were calculated as well. Figure 9 shows
the bubble vertical (a) and horizontal (b) velocity fluctuations. Markers
represent different liquids. The dotted line is a power law proportional to α0.4 .
These fluctuations behave differently than those of the liquid. The observed
exponent for the bubbles case is not in agreement with the one observed
in 3D experiments (Riboux et al., 2009). Bubble velocity fluctuations are
approximately two times larger than the liquid phase fluctuations. Sangani
et al. (1991) studied the problem of a bubble mixture subjected to small-
amplitude oscillatory motion. They calculated the normalized variance of
the bubbles velocity to leading order in α as:

hu′2 2
y i/ub = 0.275α (5)

where ub is the bubble speed and uy is the bubble vertical velocity. The
fluctuations obtained experimentally are larger than expected from equation
(5), and also larger than those reported by Bunner and Tryggvason (2002b)
from 3D simulations. It seems that confinement has a strong effect on the
bubbles velocity fluctuations. This could be the result of the instability of
the trajectory influenced by the walls (Figueroa-Espinoza et al., 2008).

19
 (a)
0
10
2
b
<u ’ >/u

−1
10
2
y

−2
10 −2 −1
10 10

0
(b)
10
2
b
<u >/u

−1
10
2
x

−2
10 −2 −1
10 10
α

Figure 9: Normalized liquid velocity fluctuations as a function of gas volume fraction:

(a) vertical and (b) horizontal components. Crosses: water, squares: glyc15, diamonds:
glyc30, stars: glyc50. The dotted line is a power law proportional to α0.4 .

3.2. Clusters lifespan

The lifespan of a cluster was considered to be the time lapse between

the cluster detection (based on an area and shape criteria) and its breakup
(see section 2.1). Probability density functions were obtained from frequency

20
histograms of the lifespan ls for each volume fraction α. These lifespan PDFs
can be well approximated by log-logistic distributions, commonly used in
survival analysis (Kleinbaum and Klein, 2012). Within this context, lifespan
(or survival) can be intuitively represented by the Survival Function, which is
the complementary cumulative distribution function SF = 1 − CDF , where
CDF stands for the cumulative distribution function, in terms of lifespan.
Clusters lifespan is also shown in nondimensional form, as

ts = ls ub /Deq (6)

One point in the SF represents the probability of (a cluster) surviving at

least ts . Figure 10 (a) shows an example for α = 0.062 as a function of the
normalized time ts . The dotted line is the best fit (log-logistic) SF, and the
continuous (stepped) line represents the measured SF from the experimental
histogram. To the right, (b) shows the SF functions for three different volume
fractions α: solid line: α = 0.077, dashed line: α = 0.062 and dotted line:
α = 0.054. As the volume fraction increases, the lifespan decreases, however
for very low volume fractions this trend was not observed, as will be discussed
next.

21
Figure 10: Survival Distribution Function in terms of the normalized lifespan. (a) Ex-
ample for α = 0.062. The continuous (stepped) line is a normailzed histogram of the
measurements, and the dotted line is the best fit (log-logistic) SF. (b) Shows SF functions
for three different volume fractions: solid line: α = 0.077, dashed line: α = 0.062 and
dotted line: α = 0.054.

In Fig. 11 a plot of timespan as a function of gas volume fraction is pre-

sented, for different test liquids, represented with different markers: circles:
water, triangles: glyc15, squares: glyc30. Gray and black markers represent
different experiments. From this figure it is clear that timespan decreases
with volume fraction; however, this trend does not hold for α < 0.04. For
very low volume fractions clustering is incipient; most clusters are formed
by two bubbles and there are only a few of them. Therefore, in this case
the timespan remains unaffected by the change in volume fraction. The un-
certainty of the normalized timespan can be estimated using its standard
deviation, which decreased with volume fraction; the largest one was about
29% for α = 0.017, decreasing to about 16% for α = 0.12. The trend (dotted)

22
line represents a power law that scales as α−1.4 , which will be discussed in
the next section.

Figure 11: Clusters normalized lifespan ts for different volume fractions α, represented
with different markers: circles: water, triangles: glyc15, squares: glyc30. Gray and black
markers represent different experiments. The trend (dotted) line represents a power law
that scales as α−1.4 .

4. Discussion and Conclusions

A nearly monodispersed bubbly flow was studied in a thin channel. Dig-

ital image processing routines were used to measure individual bubbles and
clusters position, velocity, trajectories, shape and sizes. The added mass
effect predicted by potential theory was observed in our thin channel, as a
decrease in mean bubble velocity with gas volume fraction α. The trajecto-
ries of individual bubbles and clusters were plotted to show the remarkable

23
difference in lifespan between individual bubbles and clusters.
Bubble, clusters and liquid velocity fluctuations were calculated. The
latter were obtained using PIV. The normalized liquid (vertical and horizon-
tal) velocity fluctuations for the thin channel are larger than the theoretical
prediction (equation (2)), and scale as hu′2 2 0.4
y i/Ub ∝ α .

The lifetime of clusters was estimated with the individual clusters trajec-
tories through the camera field of view. This lifetime is dependent on the gas
volume fraction α. The experiments show that for very low volume fraction
α < 0.04 this dependency is very weak, which may result from weak hydro-
dynamic interactions between the few clusters in the liquid (most of them
formed by only two bubbles), at a relative large distance from each other for
low α. For larger values of α, the cluster lifespan is greatly reduced.
The mechanism that causes clusters to break can be related to two im-
portant factors: bubbles wakes and velocity fluctuations caused by hydrody-
namic interactions. Let us consider the latter mechanism in order to picture
an analogy to statistical mechanics: let us suppose that each cluster will
traverse a path, subjected to different hydrodynamic forces (some attractive,
that tend to keep the bubbles clustered, and others repulsive or disruptive
that tend to break it). Suppose that this trajectory will end, as the mean
free path (from statistical mechanics), at a length L, which is defined as:

1
L= √ (7)
2nπd2B
where dB is the bubble (or clusters) diameter, n = N/Vt is the bubble density
number. The number density and the gas volume fraction can be related by:

24
 VB π
α= = n d3B (8)
Vt 6
which leads to:

dB
L= √ (9)
6 2α
This length decreases with volume fraction, as expected. If one considers
that the clusters lifespan ls to be proportional to L and inversely proportional
to the variance of the velocity:

L
ls = p 2 (10)
hvB i

And considering that hvB2 i scales with the volume fraction as:

hvB2 i = kUB2 αn (11)

where k is a positive constant of order one, n is an exponent that relates

velocity fluctuations to volume fraction (n = 0.8 from experiments), and UB
is a mean rise velocity , then

dB 1
ls ∝ (12)
UB α 2+n
2

For the case that concerns us, this leads to the following scaling:

ts ∝ α−1.4 (13)

which is represented by the dotted line near the normalized lifespan ts in

Figure 11. This trend is close to the experimental behavior for α (α > 0.04).

25
This suggests that the clusters breakup mechanism is induced by the hy-
drodynamic interactions, whose exponent matches the trend in normalized
cluster lifespan. However, previous investigations (Bouche et al., 2012) have
reported that normalized liquid velocity fluctuations dominated by wake in-
teractions are independent of α, so in that case the resulting exponent would
be α−1 . Unfortunately it is not clear which is the case, since some of the
curves best fit are closer to α−1.4 and others to α−1 (above α = 0.04). A
more comprehensive experimental campaign could have to be conducted to
clarify this issue.
Finally, two regimes were identified in terms of the normalized lifespan:
one regime independent of α for very low gas volume fraction, and another
regime where normalized lifespan decreases as α−b , where according to a
scaling argument, the exponent b = −1 when interaction is dominated by
wakes, and b = −1.4 when hydrodynamic interaction is the main cause of
liquid velocity fluctuations.

References

Biesheuvel, A., Van Wijngaarden, L., 1982. The motion of pairs of gas bubble
in a perfect liquid. J. Eng. Math. 16, 349–365.

Bouche, E., Roig, V., Risso, F., Billet, A. M., 2012. Homogeneous swarm
of high-reynolds-number bubbles rising within a thin gap. part 1. bubble
dynamics. J. Fluid Mech. 704, 211–231.

Bunner, B., Tryggvason, G., 2002a. Dynamics of homogeneous bubbly flows.

26
 part 1. rise velocity and microstructure of the bubbles. J. Fluid Mech. 466,
17–52.

Bunner, B., Tryggvason, G., 2002b. Dynamics of homogeneous bubbly flows.

part 2. velocity fluctuations. J. Fluid Mech. 466, 53–84.

Clift, R., Grace, J., Weber, M., 1978. Bubbles, Drops, and particles. Aca-
demic Press.

Eames, I., Hunt, J. C. R., Belcher, S. E., 2004. Inviscid mean flow through
and around groups of bodies. J. Fluid Mech. 151, 371–389.

Figueroa-Espinoza, B., Zenit, R., 2005. Clustering in high re monodispersed

bubbly flows. Phys. Fluids 17, 091701.

Figueroa-Espinoza, B., Zenit, R., Legendre, D., 2008. The effect of confine-
ment on the motion of a single clean bubble. J. Fluid Mech. 616, 419–443.

Kleinbaum, D. G., Klein, M., 2012. Survival analysis: A self learning text.
Springer.

Kok, J., 1989. Dynamics of gas bubbles moving through liquid. Phd thesis,
University of Twente.

Kok, J., 1993. Dynamics of a pair of gas bubbles moving through liquid. part
i. theory. Eur. J. Mech. Fluids 12, 515–540.

Kumaran, V., Koch, D. L., 1993a. The effect of hydrodynamic interactions on

the average properties of a bidisperse suspension of high reynolds number,
low weber number bubbles. Phys. Fluids A 5, 1123–1134.

27
Kumaran, V., Koch, D. L., 1993b. The rate of coalescence in a suspension
of high reynolds number, low weber number bubbles. Phys. Fluids A 5,
1135–1140.

Lance, M., Bataille, J., 1991. Turbulence in the liquid phase of a uniform
bubbly air-water flow. J. Fluid Mech. 222, 95–118.

Martinez-Mercado, J., Palacios-Morales, C. A., Zenit, R., 2007. Measure-

ment of pseudo-turbulence intensity in mono-dispersed bubbly liquids for
10<Re<500. Phys. Fluids 19, 103302.

Mendez, S., Serrano, J. C., Zenit, R., 2013. Power spectral distributions of
pseudo-turbulent bubbly flows. Phys. Fluids 25, 043303.

Mercado, J. M., Prakash, V., Tagawa, Y., Sun, C., Lohse, D., 2013. On
bubble clustering and energy spectra in pseudo-turbulence. J. Fluid Mech.
650, 287–306.

Riboux, G., Legendre, D., Risso, F., 2008. A model of bubble-induced tur-
bulence based on large-scale wake interactions. 719 719, 362–387.

Riboux, G., Risso, F., Legendre, D., 2009. Experimental characterization of

the agitation generated by bubbles rising at high reynolds number. J. Fluid
Mech. 643, 509–539.

Risso, F., 2011. Theoretical model for k 3 spectra in dispersed multiphase

flows. Phys. Fluids 23, 011701.

Risso, F., 2016. Physical interpretation of probability density functions of

bubble-induced agitation. J. Fluid Mech. 809, 240–263.

28
Risso, F., Amoura, V., Riboux, G., Billet, A., 2008. Wake attenuation in large
reynolds number dispersed two-phase flows. Phil. Trans. R. Soc. Lond. A
366, 2177–2190.

Roig, V., 2007. Measurement of interstitial velocity of homogeneous bubbly

flows at low to moderate void fraction. J. Fluid Mech. 572, 87–110.

Salesse, A., Larue de Tournemine, A., Roig, V., 2002. Development of bubble
cluster detection and identification method. Exp. Therm Fluid Sci. 26,
163–171.

Sangani, A. S., Didwania, A. K., 1993. Dynamic simulations of flows of bub-

bly liquids at large reynolds numbers. J. Fluid Mech. 250, 304–337.

Sangani, A. S., Zhang, D. Z., Prosperetti, A., 1991. The added mass, basset,
and viscous drag coefficients in nondilute bubbly liquids undergoing small-
amplitude oscillatory motion. Phys. Fluids A 3, 2955–2970.

Smereka, P., 1993. On the motion of bubbles in a periodic box. J. Fluid Mech.
254, 79–112.

Spelt, P., Sangani, A., 1998. Properties and averaged equations for flows of
bubbly liquids. App. Sci. Res. 58, 337–386.

Takagi, S., Ogasawara, T., Fukuta, M., Matsumoto, Y., 2009. Surfactant ef-
fect on the bubble motions and bubbly flow structures in a vertical channel.
Fluid Dyn. Res. 41, 1–17.

Van Wijngaarden, L., 1976. Bubble interactions between bubbles in liquid.

J. Fluid Mech. 77, 27–44.

29
Van Wijngaarden, L., 1993. The mean rise velocity of pairwise-interacting
bubbles in liquid. J. Fluid Mech. 251, 55–78.

Van Wijngaarden, L., 1997. On pseudo turbulence. Theoret. Comput. Fluid

Dynamics 10, 449–458.

Van Wijngaarden, L., 2005. Bubble velocities induced by trailing vortices

behind neighbours. J. Fluid Mech. 541, 203–229.

Van Wijngaarden, L., Kapteijn, C., 1990. Concentration waves in dilute bub-
ble/liquid mixtures. J. Fluid Mech. 212, 111–137.

Zenit, R., Koch, D., Sangani, A., 2001. Measurements of the average proper-
ties of a suspension of bubbles rising in a vertical channel. J. Fluid Mech.
429, 307–342.

30