Chemical Engineering Science 166 (2017) 1–10

Contents lists available at ScienceDirect

Chemical Engineering Science

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

Interaction of bubbles rising inline in quiescent liquid

M. Gumulya a,⇑, R.P. Utikar a, G.M. Evans b, J.B. Joshi c,d, V. Pareek a
a
Department of Chemical Engineering, Curtin University, U1987, Perth 6845, WA, Australia
b
Discipline of Chemical Engineering, The University of Newcastle, NSW 2308, Australia
c
Homi Bhabha National Institute, Anushaktinagar, Mumbai 400 094, India
d
Department of Chemical Engineering, Institute of Chemical Technology, Matunga, Mumbai 400019, India

h i g h l i g h t s

Numerical study on the interaction of two inline bubbles rising in stagnant liquid.

Trailing bubble interacts with wake of the leading bubble and accelerates beyond terminal velocity.

At low Reynolds (Re) numbers, inline configuration is stable and bubbles collide.

At higher Re values, trailing bubble deviates from rectilinear path due to vorticity development.

The onset of path deviation depends on the Reynolds and Eotvos numbers of the bubbles.

a r t i c l e i n f o a b s t r a c t

Article history: In the buoyant rise of two bubbles arranged in an inline configuration, the trailing bubble tends to accel-
Received 22 November 2016 erate beyond its terminal velocity, due to its interaction with the wake of the preceding bubble. It is
Received in revised form 2 March 2017 demonstrated that several different interactive behaviours could be obtained due to this acceleration.
Accepted 3 March 2017
Firstly, at low values of Reynolds numbers ðRe ¼ ql V t D=l 6 35Þ, the inline configuration was found to
Available online 6 March 2017
be stable, and the two bubbles would collide and coalesce due to the velocity difference between the
two. Secondly, at higher Re values (Re > 50), vorticity development around the trailing bubble causes it
Keywords:
to deviate away from the inline configuration, thus preventing the occurrence of a head-on collision
Bubble rise velocity
Bubble wake
between the two bubbles. The deviation of the bubble was found occur if a certain critical velocity is
Bubble-bubble interaction exceeded by the trailing bubble. The value of the critical velocity was found to decrease with increasing
Bubbly flow Re values. Further, an increase in Eotvos number ðEo ¼ ql gD2 =rÞ tends to increase the critical velocity,
Volume-of-fluid method indicating that enhanced bubble deformability actually improves the path stability of the trailing bubble.
The interaction of bubbles can therefore significantly influence the tendency of the bubbles to collide and
coalesce.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction behaviours could be developed over time, depending on the

The interaction of multiple bubbles rising in a liquid is a funda-
mental aspect in the understanding of bubble column hydrody-
namics (Joshi et al., 2017). Despite its apparent simplicity,
various aspects remain unknown, such that even the behaviour
of two bubbles rising inline with each other – a highly idealised
and simplified case in comparison to the complex hydrodynamics
of bubble columns – still demands further analysis. The DNS stud-
ies of Bunner and Tryggvason (2003), along with Esmaeeli and
Tryggvason (2005), on the behaviour of multiple bubbles in an ini-
tially quiescent liquid showed that a number of bubble distribution
⇑ Corresponding author.
E-mail address: m.gumulya@curtin.edu.au (M. Gumulya).
sphericity and deformability of the bubbles. Spherical bubbles ris-
ing at Eotvos number ðEo ¼ ql gD2 =rÞ of 1.0 and Reynolds number
ðRe ¼ ql V t D=lÞ of 35 tend to align horizontally, whereas ellipsoidal
(or ‘deformable’) bubbles (Eo = 5.0, Re = 26.0) tend to maintain a
vertical alignment and form vertical columns/streams. On the
other hand, Esmaeeli and Tryggvason (2005) showed that deform-
able bubbles rising at higher Re values (Eo = 3.0, Re = 77.6) tend to
remain uniformly distributed throughout the control volume. This
shows that various factors influence the interaction of bubbles, and
a further analysis on the contribution of these factors is required
for a greater understanding of bubble column dynamics.
The hydrodynamics of rising bubbles can be significantly influ-
enced by the wake characteristics of the neighbouring bubbles. In
the case of bubbles with inline configuration, this interaction

http://dx.doi.org/10.1016/j.ces.2017.03.013
0009-2509/Ó 2017 Elsevier Ltd. All rights reserved.
2 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10
causes a trailing bubble to attain a higher rise velocity in compar-
ison to the leading one. This was experimentally observed by Katz
and Meneveau (1996) for small air bubbles rising in distilled water
(D  475 lm, Re  35, Eo  0.027). It was observed that the veloc-
ity of the trailing bubble continuously increases as the distance
between the two decreases, leading to collision and coalescence
between the bubbles. Before colliding, the relative velocity
(DV ¼ V 2  V 1 , where V1 and V2 are the instantaneous velocities
of the leading and trailing bubbles, respectively) of larger bubbles
(349  D  475 lm) was found to increase considerably (up to 27%
of the terminal velocity), whereas smaller bubbles (D < 349 lm)
tend to attain a decrease in relative velocity. The difference in these
behaviours could be caused by the contribution of pressure gradi-
ents in the two systems, which become more dominant at smaller
distances.
The experimental observations of Katz and Meneveau (1996)
were found to be inconsistent with the numerical results of Yuan
and Prosperetti (1994), who conducted a study using a mixed
spectral/finite-difference scheme for the interaction of spherical
inline bubbles in an axisymmetric domain at Re  200. Yuan and
Prosperetti (1994) reported the presence of an equilibrium dis-
tance (Le) between two rising inline bubbles as a consequence of
the balance between the wake interaction effect and the pressure
gradients between the two bubbles. The magnitude of the equilib-
rium distance relative to the diameter of the bubbles was found to
scale with Re. The absence of the equilibrium distance phe-
nomenon in the experimental results of Katz and Meneveau
(1996) were attributed towards the non-deformability of the bub-
bles, which is one of the assumptions of the numerical study of
Yuan and Prosperetti (1994), as well as towards the presence of
impurities in the liquid. More recently, Watanabe and Sanada
(2006) showed that inline bubbles rising at Eo = 0.00878 and
30  Re  200 with an initial distance of 5R do tend to rise at a
stable equilibrium distance after some time through a series of
axisymmetric simulations. However, this behaviour was not repli-
cated in their experiments, conducted in silicone oils with bubbles
rising at 0.6  Eo  1.0 and 5  Re  40. It was noted that the inline
configuration of the bubbles as they approach equilibrium distance
is not stable, as some pairs of bubbles tend to escape from the ver-
tical line. Further, it was noted that the equilibrium distance
(defined as the bubble distance in the vertical direction just before
the bubbles escape from the vertical alignment in cases where it is
unstable) tends to be 3–8 times larger than the predictions of
Yuan and Prosperetti (1994). Finally, it was observed that the equi-
librium distance is dependent on the initial bubble distance.
Zhang and Fan (2003) have conducted a force balance analysis
on the rise of aligned spherical bubbles at Re  O(1 0 0), considering
effects of buoyancy, gravity, drag, added mass and Basset forces. A
model was proposed, based on the analytical solution for the far
wake region of the leading bubble, to determine the drag force
and acceleration of the trailing bubble. Several similar models
(Ramírez-Muñoz et al., 2013; Baz-Rodríguez et al., 2014) have
since been proposed with some adjustments in the axial velocity
profile of the wake region of the leading bubble. Ramírez-Muñoz
et al. (2013) suggested the incorporation of an ‘artificial origin’
parameter to describe the velocity profile. The magnitude of the
artificial origin was determined through a numerical analysis of
the wake region behind a rising spherical bubble, and was found
to be a linear function of Re. It was found that buoyancy, quasi-
steady drag and inertial forces are the main factors contributing
towards the motion of the trailing bubble in the wake region of
the leading one, i.e. effects of history and added-mass forces can
be neglected. This analysis was found to apply for cases where
the separation distance is greater than 4.0R. At smaller distances,
it was noted that the assumption of one-way interaction in the
hydrodynamic of the trailing bubble is no longer valid, and that
the balance of pressure gradient with the wake interaction effects
on the hydrodynamics of two inline bubbles, as shown by Yuan and
Prosperetti (1994), needs to be taken into account. Baz-Rodríguez
et al. (2014) incorporated a slightly different fit-equation for the
expression of artificial origin, and the updated force model was
found to be valid for spherical bubbles rising at 50  Re  300
and separation distances  5R. Furthermore, the hydrodynamic
force on the leading bubble was also analysed, incorporating the
upward inviscid repulsion due to potential flow generated by the
trailing bubble. As a result of this analysis, a prediction of the the-
oretical equilibrium separation distance, consistent with the find-
ings of Yuan and Prosperetti (1994) and Watanabe and Sanada
(2006), were made.
Based on the discussion above, it is clear that various uncertain-
ties still exist regarding the interactive behaviour of the bubbles,
particularly regarding the presence of equilibrium distance, the
stability of the alignment of the bubbles, and the rise velocity of
the bubbles as they get closer to each other, i.e. in the near-wake
region of the leading bubble. Further, as most air–water bubbly
flows exhibit bubbles of radius  1.5 mm, typical Re and Eo values
in practical industrial situations are 880 and 1.2, respectively
(Duineveld, 1998). On the other hand, the studies so far consider
the interaction of inline spherical bubbles at low values of Eo,
albeit at relatively high Re values (Re  300). In this study, there-
fore, the interaction of two inline bubbles at various Re values at
0.5  Eo  1.0 will be analysed through 3D numerical analysis.
The methodology used in this study will be discussed in Section 2,
whereas the parameters of the study will be outlined in Section 3,
followed by a discussion on the interactive behaviour of the
bubbles.
2. Numerical methodology
The Gerris flow solver (Popinet, 2003) was used for the compu-
tation of conservation equations for incompressible flows with
variable density and surface tension,
r  u ¼ 0; ð1Þ
qð@u=@t þ u  ruÞ ¼ rp þ r  ð2lDÞ þ rjds n: ð2Þ
u is the velocity vector, D is the deformation tensor, r the surface
tension coefficient, j the curvature of the interface, ds is the Dirac
function, and n is the normal vector to the interface. All other vari-
ables (pressure, density and viscosity) have been assigned with
their conventional notation.
The basis of the Volume of Fluid (VOF) methodology is the
assignment of a scalar variable, a, to indicate the volume fraction
of the primary phase in a computational cell. An a value of 1.0 indi-
cates that the cell is filled with the primary phase, and the opposite
is true for an a value of 0. The volume fraction field is advected
with the local velocity field,
@ a=@t þ r  ðauÞ ¼ 0: ð3Þ
The density and viscosity of the fluid are calculated at each compu-
tational cell based on the value of a, and the densities and viscosi-
ties of the individual phase,
q ¼ aq1 þ ð1  aÞq2 ; ð4Þ
l ¼ al1 þ ð1  aÞl2 : ð5Þ
The solver employs staggered temporal discretisation for the
solutions of Eqs. (1)(3), which results in a scheme that is second-
order accurate in time (Popinet, 2003). Furthermore, hierarchical
octree is used to spatially discretise the 3D domain, such that adap-
tive mesh refinement can be implemented with minimal impact to
accuracy. To minimise problems with parasitic currents in the
 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10 3

interface, the balanced-force discretisation scheme is implemented

for the surface tension component of Eq. (2). Detailed discussion on
the setup of the solver has been provided by Popinet (2003).

3. Results and discussion

Consider two bubbles, initially stationary and spherical, posi-

tioned in the centreline of a periodic rectangular domain
(18.6R 18.6R 93R), where R is the initial radius of the bubbles.
The bubbles are set to be 10R apart. At the start of the simulation,
the buoyancy of the bubbles causes them to move upwards and
accelerate towards their terminal velocity. The trailing bubble then
continues to accelerate due to the wake of the leading bubble.
Adaptive mesh refinement is used in the simulations, such that
the interface between the two phases is resolved with cells of size
R/110. Away from the interface, lower levels of refinements are
used, with the largest cell size being  R/3.44. The validity of the
current implementation has been assessed and verified by
Gumulya et al. (2016).
The parameters of the gas and liquid phases and the size of the
bubbles were selected such that 0.1  Eo  1.0 and Re  890 (see
Table 1). The indicated Re values are the Reynolds numbers of Fig. 1a. Rise velocities of the leading and trailing bubbles (bubble 1 and 2,
the bubbles if they were rising as isolated bubbles in boundless respectively) at Eo = 1 and Re = 35 (Case 9) and the vertical distance between the
stagnant fluids at their terminal velocity, Vt. In comparison to the two bubbles (normalised against the equivalent radius of the bubbles, R) as a
function of time.
recent work of Cano-Lozano et al. (2016), the parameters encom-
pass the rectilinear, chaotic, and planar zigzag regimes. Table 1 also
lists the aspect ratio (v = horizontal axis/vertical axis) of the bub- is accompanied by acceleration in the trailing bubble, which occurs
bles if they were rising as isolated bubbles in boundless stagnant at approximately the same rate. The decrease in the distance
fluids at Vt. between the two bubbles therefore occurs in an approximately lin-
Fig. 1a shows the velocity profile of two bubbles rising inline at ear manner. After the collision, the rise velocity of the merged bub-
Eo = 1.0 and Re = 35.0 (Case 9), as well as the variability of the dis- ble can be seen to fluctuate considerably, caused by some wobbling
tance between the bubbles, normalised against their initial radius, that occurs during coalescence.
x⁄ = x/R. The origin of t as displayed in Fig. 1a (and all subsequent As the two bubbles approach closer to each other, their aspect
figures), i.e. t = 0, is set to be the time where both bubbles attain ratios change as shown in Fig. 1b. It can be seen that the aspect
rise velocities that are equal to Vt. The leading bubble (bubble 1) ratios of the two bubbles are generally similar to each other at
thus rises at a constant velocity (Vt) for the most part of the simu- t < 0.2 s ðv  1:14Þ. At t > 0.2 s, pressure exerted by the acceleration
lation, whereas the trailing bubble (bubble 2) accelerates beyond of the trailing bubble causes the leading bubble to assume a flatter
the terminal velocity value. In the first instance, at configuration, as reflected by the increase in the v values of bubble
9.4 < x⁄  10.0, the velocity of the trailing bubble increases very 1. On the other hand, the aspect ratio of bubble 2 decreases
rapidly, reaching a velocity of 1.14Vt within 0.086 s. The accelera- towards 1.05, upon which the two bubbles collide and coalesce.
tion then decreases considerably, with the velocity of the trailing The acceleration of the trailing bubble in the current case is
bubble slowly reaching 1.16Vt in 0.204 s. Throughout this process, compared against the wake model provided by Zhang and Fan
the two bubbles stay inline, with the trailing bubble gradually (2003) for the interaction of two spherical inline bubbles. This
catching up to the leading one. Collision then occurs at t = 0.52 s, model is based on a force balance for the movement of the trailing
after which the two bubbles merge and coalesce. Prior to the colli- bubble,
sion, the leading bubble can be seen to accelerate (at x⁄  6.0). This
is caused by an effect of repulsion in the upwards direction due to p dV
qB D3 2 ¼ F g þ F D þ F p þ F A þ F B ; ð6Þ
the potential flow field above the trailing bubble as the two bub- 6 dt
bles get closer to each other. The acceleration in the leading bubble

Table 1
Parameters selected for the current study.

Case Eo Re Type of path if bubble was rising in Rectilinear path for trailing bubble? v
isolation (Cano-Lozano et al., 2016)
1 0.1 6 Rectilinear Yes 1.00
2 0.1 50 Rectilinear No 1.02
3 0.1 100 Rectilinear No 1.03
4 0.5 83 Rectilinear No 1.14
5 0.5 221 Rectilinear No 1.38
6 0.5 495 Chaotic No 1.78
7 0.5 890 Flattened spiral No 2.09
8 1.0 6 Rectilinear Yes 1.04
9 1.0 35 Rectilinear Yes 1.16
10 1.0 73 Rectilinear No 1.60
11 1.0 173 Rectilinear No 1.69
12 1.0 366 Planar zigzag No 2.09
4 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10

analytical solution of the velocity field in the far wake region of a

laminar, axisymmetric wake behind an axisymmetric body
(Zhang and Fan, 2002),
  
V w C D0 Re0
¼ 1  exp   : ð12Þ
Vt 2 8x

Zhang and Fan (2003) thus uses Eq. (12) in conjunction with Eq.
(11) to model the acceleration of the trailing bubble as a result of
the wake of the leading bubble.
The model of Zhang and Fan (2003) can be generalised for ellip-
soidal bubbles, by considering their aspect ratio. As such, the pro-
jected area of the bubble is,
p
AP ¼ v2=3 D2 ; ð13Þ
4
where D is the equivalent diameter of the bubble. The acceleration
of the trailing bubble can thus be written as,

dV 2 CD 2=3 FB
C AM ¼ g  0:75 ðV W  V 2 Þ2 v2 þ p 3 ð14Þ
dt D q6D
where v2 is the aspect ratio of the trailing bubble. This can be sim-
plified further through substitution of Eq. (10),
Fig. 1b. Aspect ratios of the two bubbles (Eo = 1.0, Re = 35.0, Case 9) vs time.
 2=3 !
dV 2 v2 VW  V2 FB
C AM ¼g 1þ þ : ð15Þ
where qB is the density of the bubble. The terms on the RHS of Eq. dt v1 Vt q p6 D3
(6) constitute the gravitational, drag, pressure gradient, added mass,
and history (Basset) forces, respectively. The pressure gradient term The coefficient of added mass can be calculated based on the aspect
is considered to be negligible, as the trailing bubble is assumed to ratio of the bubble, as per the model of Lamb (1932),
be located in the far wake region of the leading bubble. The added pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
E cos1 E  1  E2
mass and history terms are given as below, C AM ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð16Þ
E2 1  E2  E cos1 E
p dV 2
F A ¼ qC AM D3 ; ð7Þ
6 dt In the current study, the variability of the aspect ratio between the
Z t
pffiffiffiffiffiffiffiffiffiffi dV 2 =ds first and the second bubble is disregarded, and the ratio of v2 =v1 in
F B ¼ D2 pql pffiffiffiffiffiffiffiffiffiffiffi ds; ð8Þ
t0 ts Eq. (15) is assumed to be constant at 1. Furthermore, the coefficient
of added mass is calculated based on the aspect ratio of the bubble if
where q is the density of the liquid and C AM the added mass coeffi- it was rising in isolation in a boundless stagnant fluid, i.e. at termi-
cient (= 0.5, as the bubbles are considered spherical). The drag force nal velocity.
acting on the trailing bubble is affected by the local liquid velocity
field, which is altered by wake of the leading bubble,

ðV W  V 2 Þ2
F D ¼ C D AP q ð9Þ
2
where AP is the projected area of the bubble (¼ p4 D2 , as the bubble is
considered spherical). V W is the average axial velocity in the wake
of the leading bubble over the projected area of the trailing bubble.
Furthermore, the drag coefficient, C D is considered to be inversely
proportional to the ratio of Re of an isolated bubble (Re0) to that
of the trailing bubble,
CD Re0 V0  Vt
¼ ¼ : ð10Þ
C D0 Re VW  V2
C D0 is the drag coefficient of the bubble if it was rising in isolation in
a boundless stagnant fluid. As the current study considers the rise of
bubbles through stagnant liquid, V 0 ¼ 0. With these considerations
and the fact that qB
q, the acceleration of the trailing bubble can
be rewritten as follows,
 
dV 2 VW  V2 12F B
¼ 2g 1 þ þ ð11Þ
dt Vt qpD3
The drag coefficient is eliminated through the fact that for a bubble
rising in isolation in a boundless stagnant fluid, the drag force is bal-
anced exactly by its gravitational force. Fig. 2. Normalised rise velocity of the trailing bubble (Eo = 1, Re = 35, Case 9) as a
The wake velocity behind the leading bubble, averaged over function of normalised distance between the two bubbles, in comparison to the
the projected area of the trailing bubble, is obtained based on the models suggested by Zhang and Fan (2003).
 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10
The results of these calculations for the case of a bubble travel-
ling at Eo = 1 and Re = 35 can be seen in Fig. 2, where the dimen-
sionless velocity of bubble 2 ðV 2 =V t Þ has been presented as a
function of the dimensionless distance between the bubbles, x⁄.
The contribution of the added mass and Basset forces are firstly
analysed by considering the effect of wake velocity on the velocity
of the trailing bubble (‘‘model I”), i.e. the LHS of Eq. (15) and F B are
set to zero. Model II takes into account the added mass force in
addition to the wake velocity (F B ¼ 0), whereas model III considers
the effects of all three forces. Both models II and III are solved
numerically using the first-order Euler finite difference scheme.
As can be seen in Fig. 2, model I tends to over-predict the rise
velocity of the trailing bubble by 7%. The inclusion of the added
mass force tends to improve the agreement with the VOF result
slightly, as the over prediction in V2 decreases to 6%. Considerable
improvement of the model can be seen to occur with the inclusion
of the Basset force, as the discrepancy with the VOF results
5
The stable inline configuration was found to be missing in cases
of interactive bubbles rising at higher values of Re. Fig. 4a shows
the rise velocities of bubbles rising at Re = 173, Eo = 1 (Case 11).
As in the previous case, the trailing bubble initially accelerates
towards the leading one. This continues until x⁄ reaches 7.8
(t  0.05 s) and the velocity of the trailing bubbles exceeds the ter-
minal velocity by 14%, upon which the trailing bubble starts to
decelerate. At t = 0.05–0.07 s, V 2 can even be seen to decrease
below the rise velocity of the leading bubble, before increasing
again towards a constant value, i.e. V t . During this process, the
trailing bubble deviates away from the original inline configuration
of the two bubbles. This is evident in Fig. 4b, where the deviation of
the bubbles from the centreline of the column, normalised against
decreases down to 3%. Furthermore, the over prediction of V2
that results from the implementation of models I and II becomes
non-existent with the implementation of model III. This suggests
that both added mass and history (Basset) forces contribute greatly
towards the rise behaviour of a bubble rising in the wake of
another bubble.
The velocity profile of the two bubbles found in Case 9 was
found to be similar to those obtained for cases of low Reynolds
numbers (Cases 1 and 8, with Re = 6.0 and Eo = 0.1 and 1.0, respec-
tively). In all of these cases, the vertical alignment of the two bub-
bles is stable, and the trailing bubble accelerates towards the
leading one. As the distance between the two bubbles decreases,
potential flow at the front of the trailing bubble causes acceleration
in the leading bubble. This can be seen in Fig. 3, where the dimen-
sionless velocity of the first bubble (normalised against the termi-
nal velocity) as a function of x⁄ has been presented for Cases 1, 8
and 9. It is evident that the velocity of bubble 1 increases as x⁄
decreases below 4.3. Furthermore, within the parameters of the
current study (0.1  Eo  1), the increase in V1 appears to be rela-
tively independent from Re and Eo numbers and highly dependent Fig. 4a. Rise velocities of bubble 1 and bubble 2 (Eo = 1, Re = 173, Case 11) and the
normalised vertical distance between the bubbles vs time.
on the distance between the bubbles. Despite this increase in V1,
the velocity of the trailing bubble at this point is considerably
higher in comparison to the leading bubble, thus a collision
between the two occurs.

1.3
Eo = 0.1, Re = 6
Eo = 1, Re = 6
1.25
Eo = 1, Re = 35

1.2

1.15
V1/Vt (-)

1.1

1.05

0.95

0.9
1.5 2 2.5 3 3.5 4 4.5 5
x* (-)
Fig. 3. Velocity of bubble 1, normalised against Vt, vs the dimensionless distance Fig. 4b. Radial distance travelled by the bubbles (Eo = 1, Re = 173, Case 11) with
between the two bubbles: Eo = 0.1, Re = 6 (Case 1); Eo = 1, Re = 6 (Case 8), Eo = 1, respect to the centreline of the column, normalised against the radius of the
Re = 35 (Case 9). bubbles vs time.
6 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10
Fig. 4c. Aspect ratios of the bubbles (Eo = 1.0, Re = 173, Case 11) vs time.
Fig. 4d. Rise velocity of bubble 2 (Eo = 1, Re = 173, Case 11) vs normalised distance
between the bubbles, in comparison to the models suggested by Zhang and Fan
(2003).
the bubble radius, has been presented. While some deviation could
be seen to occur with bubble 1 (maximum deviation  0.304R), the
deviation experienced by the trailing bubble is considerably more
prominent, with a maximum deviation of 4.4R. Upon deviating
from the path of the leading bubble, the trailing bubble then rises
at Vt.
The aspect ratios of the bubbles (Eo = 1.0, Re = 173, Case 11) can
be seen in Fig. 4c. While bubble 1 maintains a relatively constant
value of aspect ratio ðv  1:6Þ, bubble 2 undergoes a few of stages
of transformation. During the wake-induced acceleration stage
(t < 0.05 s), the aspect ratio of bubble 1 is slightly lower than that
of bubble 1. The aspect ratio then drastically decreases, as the bub-
ble rotates about its centre of gravity and develops momentum in
the lateral direction. After some time (t > 0.05 s), the bubble grad-
ually returns to its upright orientation, with its major and minor
axes aligned in the horizontal and vertical directions, respectively.
The aspect ratio therefore gradually increases, eventually reaching
a similar constant value as the leading bubble ðv  1:6Þ. The con-
stant values of aspect ratio and rise velocity, both of which are rel-
atively similar to their respective values for the leading bubble,
indicates that the trailing bubble travels independently from the
wake of the leading bubble after its deviation from the inline
configuration.
The acceleration of the trailing bubble in comparison to the
wake model provided by Zhang and Fan (2003) can be seen in
Fig. 4d (‘‘model III”). In this Figure, the velocity of the trailing bub-
ble prior to its deviation from the centreline has been presented. It
can be seen that the model tends to under-predict the trailing bub-
ble velocity by 5%. This under prediction has been attributed
towards the discrepancies in the actual wake velocity behind the
leading bubble with the model outlined in Eq. (12). Another set
of simulations was thus performed, this time involving single bub-
bles rising in stagnant unbounded fluid, such that the decay of
axial velocity in the wake behind a rising bubble can be identified.
The results of this analysis have been shown in Fig. 5a and b, where
the averaged wake velocity for bubbles at various Re values over
the projected area of the trailing bubble (normalised against Vt,
at x distance away from the centre) have been presented. As can
be seen in Fig. 5a, at Re 6 35, a good agreement was obtained
between the model provided through Eq. (12) with the actual wake
velocity underneath the bubble. However, at higher Re values, the
wake velocity model (Eq. (12)) was found to under predict V w =V t
considerably (see Fig. 5b). The wake model as presented in Eq.
(15) therefore needs to be implemented with a more accurate rep-
resentation of the wake velocity behind the leading bubble.
The results of this analysis, i.e. the implementation of model III
with the actual bubble wake velocity as presented in Fig. 5b (i.e.
obtained through the single bubble simulations), have been shown
in Fig. 4d (‘‘model III VW adjusted”). As can be seen in this Figure, a
greater agreement is obtained between the model and the bubble
rise velocity, with the model tending to slightly over predict V2 by
2.5%. This suggests that the bubble wake model can be improved
through a better understanding on the velocity field in the wake
behind the bubbles.
Fig. 6 shows a compilation of the trailing bubble velocity as
obtained through the VOF simulations with those obtained
through Eq. (15). As previously discussed, the values of VW were
obtained through single bubble simulations, such that the acceler-
ation of the second bubble through the bubble wake can be accu-
rately captured. As can be seen in this Figure, while the model
tends to slightly over-predict the acceleration of the trailing bub-
ble, a generally good agreement is obtained with the VOF data.
The margin of error for this analysis was found to be 5.5%, indi-
cating that the current model can predict the trailing bubble veloc-
ity with a relatively good accuracy.
Fig. 7a shows the maximum streamwise vorticity values,
xz;max ¼ maxð@U=@y  @V=@xÞ in the wake of the trailing bubbles
for two cases (Re = 35 and 173 at Eo = 1.0). In both cases, the max-
imum value of the streamwise vorticity behind bubble 1, xz;max1 ,
stays relatively constant throughout the timeline of the simula-
tions. The maximum values of streamwise vorticity behind the
trailing bubble have therefore been normalised against xz;max1 ,
such that the variability of vorticity for the two cases can be com-
pared. For the case of Re = 35, xz;max2 =xz;max1 can be seen to stay
relatively constant at 1, despite the increase in the rise velocity of
bubble 2. On the other hand, the maximum vorticity around the
trailing bubble for the case of Re = 173 can be seen to increase con-
siderably at 10  t⁄  30, corresponding to t  0.076–0.13 s, before
 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10
Fig. 5. Averaged wake velocity over the projected area of bubble 2, normalised
against Vt, vs the dimensionless distance in the vertical direction from the centre of
the bubble (Eo = 1.0).
decreasing back to xz;max1 . While the occurrence of this increase
in xz;max2 is relatively late in comparison to the deviation of the
trailing bubble as shown in Figs. 4a and 4b, it nevertheless indi-
cates that the increase in the rise velocity of the trailing bubble
in this case has caused an increase in vorticity strength, which thus
causes an instability in the rise path of the bubble. This observation
is consistent with the findings of Cano-Lozano et al. (2016), who
found that path instability exhibited by a bubble without the pres-
ence of a standing eddy is caused by the generation of vorticity at
the surface of the bubble. Further, the iso-surfaces of the stream-
wise vorticity behind bubbles 1 and 2 at Eo = 1 and Re = 173 at
t⁄ = 1.4 (as well as after the trailing bubble has deviated) has been
7
Fig. 6. Comparison of the trailing bubble velocity as predicted through model III
(Eq. (15)) with the results of the VOF simulations (for cases 9, 11, 12, 5, and 7,
respectively).
Fig. 7a. Maximum streamwise vorticity around bubble 2, normalised against the
maximum streamwise vorticity around bubble 1, vs. dimensionless time, t ¼ tV t =R.
shown in Fig. 7b, where it is evident that there is an expansion in
the vorticity field around the trailing bubble in comparison that
around the leading one.
The deviation of the trailing bubble was studied further by set-
ting a different value of initial distance between the bubbles, x⁄0.
Fig. 8 presents the rise velocities of bubble 2 as a function of the
distance it has travelled for cases of x⁄0 of 10 and 15 at Eo = 1.0
and Re = 173. Additionally, the radial distance of the bubbles with
respect to the centreline was also presented in this figure. In both
cases, the deviation of bubble 2 can be seen to occur after it has
travelled over a distance of 16-17R. The separation distances
between the two bubbles at these stages are 8.3R and 14R for x⁄0
of 10 and 15, respectively, whereas the values of V2 at the start
of deviation were found to be 1.14Vt and 1.12Vt. The similarities
in V2 at the onset of path instability suggest that the deviation in
8 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10

Fig. 8. Normalised velocity of bubble 2 and radial distance from the centreline
(normalised against the bubble radius) vs. distance travelled by the bubble
(Eo = 1.0, Re = 173, Case 11).

Fig. 9. Critical rise velocity of the trailing bubble at which point path instability
occurs vs Re.

The value of rise velocity at the start of path instability has been
termed as the critical velocity, V2,crit. This parameter was deter-
mined as the rise velocity of the trailing bubble upon which its
deviation in the radial direction exceeds 50% of its equivalent
radius. The results of this analysis have been presented in Fig. 9
for bubbles of various Eo and Re values, such that the values of
V2,crit, normalised against Vt, can be compared. In this figure, it is
evident that V 2;crit =V t tends to decrease with increasing Re value.
This was as expected, due to the instability of the bubble wake at
Fig. 7b. Iso-surfaces of the streamwise vorticity ðxz R=V t  0:01Þ: (i) at t* = 1.4 for
the leading bubble, (ii) at t* = 1.4 for the trailing bubble, and (iii) at t* = 27.4 for the
trailing bubble.
the trailing bubbles is dependent on the acceleration and rise
velocity rather than the distance between the bubbles.
higher inertia and the decreased stabilising effect of the viscous
forces. Furthermore, V 2;crit =V t can be seen to decrease with decreas-
ing values of Eo. This indicates that increased contribution of sur-
face tension actually destabilises the rise path of the trailing
bubble. The decreased deformability of the trailing bubble at lower
 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10
values of Eo causes it to be more sensitive towards the distur-
bances generated in the wake of the leading bubble.
The rise velocity pattern of the trailing bubble as presented in
Fig. 4a for Case 11 was found to be similar to those of trailing bub-
bles of cases 2–7 and 10, despite the different classifications of flow
regime as proposed by Cano-Lozano et al. (2016, see Table 1). This
indicates firstly that bubbles that would normally rise in a rectilin-
ear path if they were rising in isolation would tend to deviate in
their paths if their rise velocity exceeds a certain critical velocity
due to interaction with the wake of another bubble. As a result,
the bubbles are prevented from undergoing head-on collisions
with each other. Furthermore, for most of the cases (Case 10 is
an exception – this will be discussed in the following paragraph),
the trailing bubbles would tend to rise in an approximately recti-
linear manner, as indicated by the constant rise velocity (Vt),
upon deviation (see Fig. 4a). This is supported by Fig. 7a, where
Fig. 10a. Rise velocities of bubble 1 and bubble 2 (Eo = 1, Re = 366, Case 10) and the
normalised vertical distance between the bubbles vs time.
Fig. 10b. Path of bubble 1 and bubble 2 (Case 10).
9
the maximum vorticity behind the trailing bubble was found to
decrease back down to xz;max2  xz;max1 when the rise velocity
reaches  Vt. The rectilinear path of the trailing bubble upon the
completion of the deviation was found even in cases that fall under
the ‘‘chaotic” and ‘‘flattened spiral” category according to the clas-
sifications of Cano-Lozano et al. (2016) (Cases 5 and 6). Based on
these classifications, it would have been expected that the trailing
bubbles would continue on a non-rectilinear path upon deviating.
This could be caused by the limitations of the current study, where
the flow path is considerably shorter in comparison to that
employed by Cano-Lozano et al. (2016). Any instability inherent
in the rise paths of bubbles rising in Cases 5 and 6 may therefore
have not developed sufficiently by the end of the flow path in
the current study, such that a noticeable change in the bubble rise
velocities could be observed. Nevertheless, the fact that the two
bubbles appear to rise independently from each other and at termi-
nal velocity indicate that it is unlikely that a collision would occur,
even if the flow path is lengthened considerably.
A slightly different rise velocity pattern was obtained in the
case of bubbles of Case 10 (see Fig. 10a), where the development
of a standing eddy has caused them to inhabit a planar zigzag
movement (see Fig. 10b). Upon deviating, the trailing bubble can
be seen to develop a greater zigzag pattern in comparison to the
leading bubble, which results in a considerable fluctuation in the
bubble rise velocity. Furthermore, as the rise velocity of bubble 2
is less than the terminal velocity, the vertical distance between
the two bubbles increases overtime, thus preventing the two from
colliding with each other.
The findings presented on the rise behaviour of inline bubbles
at moderate values of Re were found to be consistent with the
experimental observations of Watanabe and Sanada (2006), where
the vertical alignment of two bubbles that are originally inline was
found to be unstable (Eo  0.9 and Re = 40). Watanabe and Sanada
(2006) also noted that the onset of the deviation from the vertical
alignment appears to be dependent on the initial bubble distance.
This supports our conclusion that the path instability inhabited by
the trailing bubble is caused by its acceleration and development
of vorticity around the bubble. Further, within the parameters of
the current study, there doesn’t appear to be any case of equilib-
rium distance, wherein the balance between the attractive forces
caused by the wake of the leading bubble and the potential repel-
10 M. Gumulya et al. / Chemical Engineering Science 166 (2017) 1–10
ling force ahead of the trailing bubble causes the two bubbles to
travel at a stable vertical distance from each other. The presence
of this phenomenon in the studies of Yuan and Prosperetti
(1994) and Watanabe and Sanada (2006) could therefore be attrib-
uted towards the axisymmetric assumption taken in both studies.
The current findings were found to be comparable to the find-
ings of the studies of multiple bubbles conducted by Bunner and
Tryggvason (2003), along with Esmaeeli and Tryggvason (2005).
In their studies, spherical bubbles rising at Re = 35 were found to
have a tendency to align horizontally, whereas ellipsoidal bubbles
rising at Re = 26 tend to maintain a vertical alignment. It was noted
that the horizontal alignment of the spherical bubbles were caused
by the tendency of spherical bubbles to rotate about each other (or
‘tumble’) before they have a chance to collide with each other. This
observation agrees well with the current finding, wherein the crit-
ical velocity for the deviation of the bubbles was found to increase
with increasing values of Eotvos number, indicating an increase in
path stability with increased bubble deformation.
4. Conclusions
The interaction of two bubbles rising in an inline configuration
has been studied through 3D numerical simulations. The parame-
ters of the simulations were set such that the bubbles fall in the
spherical and ellipsoidal flow regimes (0.1  Eo  1, Re  890).
Due to interaction with the wake of a preceding bubble, the rise
velocity of a trailing bubble can increase considerably beyond its
terminal velocity. This acceleration was found to be dependent
on the velocity field in the wake behind the leading bubble, as well
as on the added mass and the history (Basset) force. The velocity of
the bubble was found to agree well with a model wherein the
effects of these phenomena, along with the deformation of the
bubble, were taken into account.
The inline configuration of the two bubbles was found to be
stable at low values of Re ( 35). At Re > 50, the development of
vorticity resulting from the acceleration of the bubble causes it
to deviate away from the rectilinear path. The onset of deviation
was found to be a function of the bubble velocity, i.e. it occurs if
a certain critical velocity is exceeded. The value of the critical
velocity was found to decrease with increasing Re values. On the
other hand, an increase in Eo was found to increase the critical
velocity of the bubbles, indicating that decreased contribution of
surface tension, and increased deformability of the bubble, actually
improves the path stability of the trailing bubble.
The interaction of bubbles therefore strongly determines the
likelihood of their collision and coalescence. For most of the cases
considered in the current study, the deviation in the trailing bubble
was found to be temporary, and an approximately rectilinear path
was obtained upon the completion of the deviation. The bubble
would then rise at a constant (terminal) velocity, and the two bub-
bles would then rise at a stable distance from each other. At the
highest Re value (Re = 890, Eo = 0.5), the deviation of the trailing
bubble causes it to attain a greater path deviation in comparison
to the leading bubble. As a result, the vertical distance between
the two increases and a collision becomes even more unlikely.
Acknowledgements
This work was funded through the ARC Discovery Projects
Scheme DP140104510. This work was supported by resources pro-
vided by The Pawsey Supercomputing Centre with funding from
the Australian Government and the Government of Western Aus-
tralia. The authors would also like to acknowledge the remarkable
contribution of the Gerris Flow Solver community (http://source-
forge.net/projects/gfs/).
References
Baz-Rodríguez, S.A., Ramírez-Muñoz, J., Soria, A., Sacramento-Rivero, J.C., 2014.
Hydrodynamic interaction of two spherical bubbles rising in-line: a semi-
analytical approach. Chem. Eng. Commun. 201 (5), 674–687.
Bunner, B., Tryggvason, G., 2003. Effect of bubble deformation on the properties of
bubbly flows. J. Fluid Mech. 495, 77–118.
Cano-Lozano, J.C., Martinez-Bazan, C., Magnaudet, J., Tchoufag, J., 2016. Paths and
wakes of deformable nearly spheroidal rising bubbles close to the transition to
path instability. Phys. Rev. Fluids 1 (5), 053604.
Duineveld, P.C., 1998. Bouncing and coalescence of bubble pairs rising at high
Reynolds number in pure water or aqueous surfactant solutions, in. In:
Fascination of Fluid Dynamics. Springer, Netherlands, pp. 409–439.
Esmaeeli, A., Tryggvason, G., 2005. A direct numerical simulation study of the
buoyant rise of bubbles at O (1 0 0) Reynolds number. Phys. Fluids 17 (9),
093303.
Gumulya, M., Joshi, J.B., Utikar, R.P., Evans, G.M., Pareek, V., 2016. Bubbles in viscous
liquids: time dependent behaviour and wake characteristics. Chem. Eng. Sci.
144, 298–309.
Joshi, J.B., Nandakumar, K., Evans, G.M., Pareek, V.K., Gumulya, M.M., Sathe, M.J.,
Khanwale, M.A., 2017. Bubble generated turbulence and direct numerical
simulations. Chem. Eng. Sci. 157, 26–75.
Katz, J., Meneveau, C., 1996. Wake-induced relative motion of bubbles rising in line.
Int. J. Multiph. Flow 22 (2), 239–258.
Lamb, H., 1932. Hydrodynamics. Cambridge University Press.
Popinet, S., 2003. Gerris: a tree-based adaptive solver for the incompressible Euler
equations in complex geometries. J. Comput. Phys. 190 (2), 572–600.
Ramírez-Muñoz, J., Baz-Rodríguez, S., Salinas-Rodríguez, E., Castellanos-Sahagún, E.,
Puebla, H., 2013. Forces on aligned rising spherical bubbles at low-to-moderate
Reynolds number. Phys. Fluids 25 (9), 093303.
Watanabe, M., Sanada, T., 2006. In-line motion of a pair of bubbles in a viscous
liquid. JSME International Journal Series B Fluids and Thermal Engineering 49
(2), 410–418.
Yuan, H., Prosperetti, A., 1994. On the in-line motion of two spherical bubbles in a
viscous fluid. J. Fluid Mech. 278, 325–349.
Zhang, J., Fan, L.S., 2002. A semianalytical expression for the drag force of an
interactive particle due to wake effect. Indust. Eng. Chem. Res. 41 (20), 5094–
5097.
Zhang, J., Fan, L.S., 2003. On the rise velocity of an interactive bubble in liquids.
Chem. Eng. J. 92 (1), 169–176.