You are on page 1of 50

A deterministic breakup model for Euler–Lagrange simulations of turbulent microbubble-laden flows

Journal Pre-proof

A deterministic breakup model for Euler–Lagrange simulations of


turbulent microbubble-laden flows

F. Hoppe, M. Breuer

PII: S0301-9322(19)30312-X
DOI: https://doi.org/10.1016/j.ijmultiphaseflow.2019.103119
Reference: IJMF 103119

To appear in: International Journal of Multiphase Flow

Received date: 29 May 2019


Revised date: 30 July 2019
Accepted date: 23 September 2019

Please cite this article as: F. Hoppe, M. Breuer, A deterministic breakup model for Euler–Lagrange
simulations of turbulent microbubble-laden flows, International Journal of Multiphase Flow (2019), doi:
https://doi.org/10.1016/j.ijmultiphaseflow.2019.103119

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2019 Published by Elsevier Ltd.


1

Highlights

• A deterministic and viable breakup model for microbubble-laden turbulent flows.

• Transfer from the Euler-Euler method to the Euler-Lagrange approach .

• Deterministic estimation of the size of arising daughter bubbles .

• Physically motivated model for the separation axis and separation velocity .

• Time lag between two successive breakup processes proposed.


A deterministic breakup model for Euler–Lagrange simulations of
turbulent microbubble-laden flows
F. Hoppe, M. Breuer∗
Professur für Strömungsmechanik, Helmut–Schmidt–Universität Hamburg, D–22043 Hamburg, Germany

Abstract

The present study is concerned with breakup models for microbubbles in turbulent flows. An-
alyzing the different physical mechanisms responsible for breakup based on a literature review,
breakage due to turbulent fluctuations in the inertial subrange is identified as the most im-
portant one. Widely used breakup models for this mechanism are discussed concerning their
advantages and drawbacks with special emphasis on thoughts how these models developed in
the Euler-Euler context can be transferred into the Euler-Lagrange approach favored in this
study. The most promising model is chosen as a basis and then implemented in an efficient
bubble tracking scheme relying on the large-eddy simulation technique. The size of the daugh-
ter bubbles is deterministically estimated based on the breakup mechanism. Furthermore, a
physically motivated model for the axis along which bubbles separate and for the separation
velocity of the daughter bubbles is developed. Lastly, an estimate of the time lag breakup
time between two successive breakup processes is provided. The simulation methodology is
validated against an experimental study by Martı́nez-Bazán, C., Montañes, J.L., Lasheras,
J.C., 1999 (On the breakup of an air bubble injected into a fully developed turbulent flow.
Part 1/2. J. Fluid Mech. 401, 157–182 and 183–207) investigating bubble breakup within a
turbulent jet flow. The predicted results are found to be in reasonable agreement with the
measurements. Furthermore, the effect of coalescence and other properties of the bubbles on
the breakup behavior is investigated.
Keywords: Turbulence-induced bubble breakup, Euler–Lagrange, daughter bubble size,
separation axis and velocity, breakup time, coalescence


Corresponding author
Email address: breuer@hsu-hh.de (M. Breuer)

Preprint submitted to Elsevier September 24, 2019


1 INTRODUCTION AND LITERATURE OVERVIEW 3

1. Introduction and Literature Overview

Turbulent bubble-laden flows occur in many natural processes and industrial applications,
e.g., the bubble entrainment by the breaking of waves in the oceans (Melville, 1996) or the
chemisorption process used to remove H2 S or CO2 from exhaust gases (Sujatha et al., 2017).
Since the interfacial area of the gas bubbles with the liquid carrier phase governs the mass
transfer rate between the two phases, this interface given by the number and the size of bubbles
dispersed in the flow, is of great importance in many two-phase flow problems. Besides the
capability of the bubbles to coalesce forming larger bubbles, the breakup into smaller fragments
due to different physical mechanisms strongly affects their size evolution.
Consequently, it is mandatory for numerical predictions of turbulent two-phase flows to capture
the breakup processes as accurately as possible. Large-eddy simulations in the framework
of Euler–Lagrange predictions are a promising approach, since both the fluid phase is cost-
efficiently and reliably predicted, while at the same time all relevant properties of the individual
bubbles are provided. However, due to the complex physics involved in the interaction of
bubbles with the surrounding fluid, the forecast of breakup is highly challenging. Typically,
the existing breakup models are formulated in the framework of the Euler–Euler approach
meaning that population balance equations are solved for the bubbles (see, e.g., Liao and
Lucas, 2009; Solsvik et al., 2013, for a literature review on breakup models). Since only the
number and size distribution of the bubbles are determined by the population balance, the
breakup process is described by breakup rates Ωbu for bubbles of a certain size. In order to
transfer these breakup models to the Euler–Lagrange framework, the breakup criteria can be
directly applied to individual bubbles as done by Lau et al. (2014) and Jain et al. (2014).
However, the prediction of the size of the daughter bubbles remains challenging. In the Euler–
Euler framework the size is usually modeled as a random variable following a bubble size
distribution, which in some models is estimated based on the breakup criterion. In the Euler–
Lagrange framework the size distributions can be used as well (see, e.g., Lau et al., 2014; Jain
et al., 2014) making the daughter sizes random variables following the distribution functions.
Furthermore, if the diameters of the generated bubbles appear in the criterion, it is possible
to obtain the daughter sizes directly from the breakup criterion. Note that the coalescence
of bubbles can be treated in an equivalent way, i.e., the coalescence criteria developed in
the Euler–Euler framework (see, e.g., Liao and Lucas, 2010) are directly applied to bubble
1 INTRODUCTION AND LITERATURE OVERVIEW 4

collisions, for example in the studies by Mattson and Mahesh (2012), Sungkorn et al. (2012)
or Hoppe and Breuer (2018).
According to the reviews by Liao and Lucas (2009) and Solsvik et al. (2013) the breakup
models are divided into different categories based on the physical mechanism responsible for
breakup, i.e., the breakup due to surface instabilities, due to viscous shear stresses and due to
turbulent fluctuations. Surface instabilities responsible for bubble breakup are the Rayleigh–
Taylor and the Kelvin–Helmholtz instabilities. According to Sharp (1984) the Rayleigh–Taylor
instability occurs when the less dense fluid pushes the denser one, e.g., gas bubbles accelerating
the surrounding fluid. This leads to spikes of denser fluid penetrating the less dense fluid,
ultimately causing the rupture of the lighter phase. On the other hand, the Kelvin–Helmholtz
instability is caused by a difference in the tangential velocity across the interface between two
fluids. Based on fully-resolved simulations of a bubble rising in a quiescent liquid, Tripathi
et al. (2015) demonstrated that this difference first leads to the deformation of the bubbles
into a toroidal shape and then to the complete breakup. A second form of breakup due to the
Kelvin–Helmholtz instability is the shearing off of small bubbles from the edge of a larger bubble
having the form of a spherical cap (Tripathi et al., 2015). Note that the breakup caused by
surface instabilities only occurs for large, non-spherical bubbles, since the high surface tension
of small bubbles resists the deformation of their surface.
The second prominent breakup mechanism is due to viscous shear stresses originating from a
velocity gradient of the mean flow around the bubbles leading to different fluid velocities at
opposing sides of the bubble. This causes the bubble to be transformed from a spherical to a
sigmoidal form with thin, pointed tips (Müller-Fischer et al., 2008). According to the exper-
iments of air bubbles dispersed in silicone oil by Müller-Fischer et al. (2008), small daughter
bubbles are ruptured from the tips of the parent bubble. In these experiments it was shown
that the breakup due to shear stresses occurs if the capillary number Ca exceeds a critical
value Cacrit , which lies between 30 and 45. Two-dimensional Lattice Boltzmann simulations
by Wei et al. (2012) also predicted breakup due to shear stresses at Cacrit ≈ 35. Yet, again
the breakup due to shear stresses occurs mainly for large bubbles, since the velocity gradients
required to deform small bubbles with a high surface tension have to be very large.
The last physical mechanism related to the breakup by turbulent fluctuations of the liquid
surrounding the bubbles makes up the largest group of breakup models. Note that this group
can be further classified into the breakup due to turbulence in the viscous and the inertial
1 INTRODUCTION AND LITERATURE OVERVIEW 5

subrange depending on the size of the bubbles. Bubbles of a size comparable to the viscous
subrange of turbulence are subjected to shear stresses of this range. However, Hinze (1955)
argued that, except for low Reynolds numbers, the length scales of the viscous subrange are
usually much smaller than the bubbles. Hence, it is a common assumption in the literature on
turbulence-induced breakup (see, e.g., Tsouris and Tavlarides, 1994; Luo and Svendsen, 1996;
Lehr et al., 2002) that bubbles are not affected by this breakup mechanism. Consequently, the
turbulent fluctuations in the inertial subrange are assumed to be dominant and responsible
for the breakup of small bubbles. However, in recent studies several authors (see, e.g.,
Solsvik and Jakobsen, 2016b,a; Karimi and Andersson, 2019; Castellano et al., 2019) stated
that the complete spectrum of turbulence should be considered for breakup, since fluid particles
(droplets or bubbles) can have largely varying sizes.
The physical mechanism responsible for breakup by turbulence in the inertial subrange are
turbulent pressure fluctuations associated with velocity fluctuations around the surface of the
bubbles leading to a deformation of the bubble (Hinze, 1955). If this deformation is sufficiently
strong, the bubbles are not able to restore their spherical form and break up into smaller
parts. However, the actual deformation processes are highly complex making the prediction of
breakup very challenging. The deformation due to turbulent fluctuations has been investigated
by several authors (see, e.g., Risso and Fabre, 1998; Andersson and Andersson, 2006; Qian et al.,
2006), who observed that the impact of a turbulent eddy on the surface of a spherical bubble
leads to the deformation of the bubble to an elongated, dumbbell-like shape, i.e., a shape
consisting of two, usually unequally-sized non-spherical volumes of gas connected by a small
neck. According to Andersson and Andersson (2006) three different outcomes are possible
after this initial deformation. Firstly, it is possible that the bubble does not break up and can
restore its original form. Due to the fact that the ends of the dumbbell-like bubble are usually
not of the same size, an internal pressure difference between the two ends exist caused by the
surface tension and the radii of the two ends. This pressure difference leads to a rapid flow
from the smaller-sized end through the neck back into the larger end restoring the original
shape of the bubble. The second possible outcome of the initial deformation is the pinch off
of one daughter bubble. Due to the reduced pressure in the neck caused by the backflow, the
neck is shrinking, while the gas is redistributed in the bubble. Hence, it is possible that the
neck collapses before the complete volume of the smaller end of the dumbbell-shaped bubble
has been flowing back. Thus, the parent bubble splits into two daughter bubbles with one
1 INTRODUCTION AND LITERATURE OVERVIEW 6

daughter typically consisting of only a small fraction of the original bubble volume. Lastly,
the dumbbell-shaped bubble can break up directly without the occurrence of backflow, which
requires the energy of the turbulent structure initially deforming the parent bubble to be large
(Andersson and Andersson, 2006). Note that in the experiments by Risso and Fabre (1998),
Andersson and Andersson (2006), Hesketh et al. (1991a,b), Walter and Blanch (1986), Stewart
(1995) and also Martı́nez-Bazán et al. (1999a) the turbulence-induced breakup was usually
binary, i.e., forming two daughter bubbles. However, in some cases Wilkinson et al. (1993),
Risso and Fabre (1998) and Solsvik and Jakobsen (2015) also observed the breakup into up to
ten fragments.
In addition to the previously described breakup mechanisms, where bubbles either break up
directly or restore their original spherical shape, Risso and Fabre (1998) demonstrated that
a resonance mechanism can also lead to bubble breakup. An eddy hitting the bubble with
insufficient turbulent kinetic energy to break the bubble up can excite an oscillation. If the
oscillating bubble is subsequently excited by such eddies, it is possible that the deformation
is amplified and the breakup occurs due to the resonance of turbulent eddies and bubble
oscillations. Finally, it is known from experiments of a bubble-laden flow through a Venturi-
shaped pipe by Wilkinson et al. (1993) that both bubble (gas density ρb , surface tension σ)
and fluid properties (fluid density ρf , fluid viscosity νf , structure of turbulence) play a role for
the breakup of bubbles.
In order to model the breakup process, a criterion for the occurrence of breakup reflecting the
physical mechanism is required. This might vary depending on further assumptions made by
the model in order to simplify the complex breakup process. Typically, the criterion can be
expressed by the dimensionless Weber number We, which either relates the kinetic energy of
the fluid to the surface energy of the bubbles or analogously by the ratio of inertial to surface
forces. Several assumptions are common to most of the prominent breakup models. Firstly,
it is usually assumed that only turbulent structures up to a length scale comparable to the
bubble diameter contribute to the turbulent breakup, since larger turbulent structures only
transport the bubbles (Coulaloglou and Tavlarides, 1977). Additionally, the assumption by
Hinze (1955) that turbulence is locally isotropic is also made by most authors. Furthermore,
it is assumed that the breakup of the parent bubble is binary neglecting the cases in which a
larger number of daughter bubbles is generated. Lastly, the resonance mechanism described
by Risso and Fabre (1998) is neglected accounting only for the direct breakup by sufficiently
1 INTRODUCTION AND LITERATURE OVERVIEW 7

strong turbulence.
One of the first models considering the breakup of gas bubbles was the model by Prince
and Blanch (1990). They assumed that the breakage occurs if We exceeds a critical value of
Wecrit = 2.3 meaning that the turbulent kinetic energy of the fluid fluctuations of the size
equal to the bubble diameter must be larger than a certain fraction of the surface energy of the
bubbles. The estimation of this fraction was based on the maximum stable bubble size in a gas-
sparged turbulent bubble column studied by Bhavaraju et al. (1978) applying a parameter fit to
experimental data. Hence, the criterion by Prince and Blanch (1990) is probably not generally
valid for arbitrary turbulent flows. Additionally, the breakup model requires the modeling of
the daughter bubble sizes based on an arbitrary size distribution, which is independent of the
breakup criterion.
Luo and Svendsen (1996) advanced the breakup criterion by providing an estimate of the critical
Weber number Wecrit , which did not depend on experimental observations. They based their
model on the idea by Tsouris and Tavlarides (1994), who proposed that the energy required
for breakup is associated with the increase ∆ES of the surface energy due to breakup. The
total surface of the two daughter bubbles is always larger than the surface of the parent bubble
leading to an increase of the total surface energy. According to Luo and Svendsen (1996)
the turbulent fluctuations have to provide this energy leading to breakup if the turbulent
kinetic energy of the impacting eddy is equal or larger than the difference in surface energy.
Another advantage of the model by Luo and Svendsen (1996) is that the breakup criterion
also provides the size of the daughter bubbles by relating the turbulent kinetic energy of the
fluctuations to the diameters of the daughter bubbles. For this purpose it is assumed by Luo
and Svendsen (1996) that the size provided by the breakup criterion is just the upper boundary
of the size of the smaller daughter bubble, i.e., the diameter of the smaller bubble can have
any random value up to this boundary according to a bubble size distribution estimated from
the breakup rate ΩLS
bu of the model. However, an obvious disadvantage is that regardless of

We one can find two daughter bubbles that fulfill the criterion by Luo and Svendsen (1996),
since the increase in surface energy tends to zero if the size of the smaller daughter bubble
tends to zero. Consequently, in the framework of Euler–Lagrange predictions this would lead
to bubble breakup at every instance of the simulation doubling the number of individually
tracked bubbles. This problem was also critically annotated by Hagesaether et al. (2002) who
mentioned that the rate of breakup into a very small daughter bubble and one which is of a
1 INTRODUCTION AND LITERATURE OVERVIEW 8

size nearly equal to the parent bubble tends to infinity in the model by Luo and Svendsen
(1996).
A different approach was chosen by Martı́nez-Bazán et al. (1999a,b) who proposed that the
parent bubble breaks into two daughter bubbles, if the turbulent stresses of the fluid fluctu-
ations exceed the surface restoring pressure. The turbulent stresses and the surface restoring
pressure are estimated based on the turbulent kinetic energy of the fluid fluctuations and the
surface energy of the parent bubble, respectively. The daughter size is again a random variable
with a distribution estimated from the breakup rate ΩMB
bu of the model, since it is not possible

to derive the daughter size directly. The reason is that solely the diameter of the parent bubble
appears in the breakup criterion. The authors performed experiments of bubble breakup in a
turbulent jet flow and found a good agreement of the model with their experimental results.
Note that the model by Martı́nez-Bazán et al. (1999a,b) was applied in Euler–Lagrange predic-
tions by Lau et al. (2014) and Jain et al. (2014). However, a problem concerning this model is
that breakup is predicted even if the turbulent kinetic energy only slightly exceeds the surface
stresses required to deform but not necessarily break up the bubble. Hence, in such a case the
associated deformation is still rather small and the breakage of the bubble is rather unlikely.
It was shown by Risso and Fabre (1998) that the surface of the parent bubble has to increase
by at least 50% compared to the surface area of the spherical shape for breakup to occur. It
has to be mentioned that this issue has been addressed by Martı́nez-Bazán et al. (2010). An
additional improvement of the model by Martı́nez-Bazán et al. (1999a,b, 2010) was recently
proposed by Solsvik et al. (2017).
Lehr et al. (2002) proposed a similar model, where the turbulent stresses deform the parent
bubble into a cylindrical shape. At some point of the bubble tube a necking occurs which
contracts resulting in the breakup of the bubble. Lehr et al. (2002) assume that the surface
restoring pressure trying to prevent the necking is equal to the pressure increase inside the
cylindrical bubble given by the Young–Laplace equation. Hence, bubble breakup is predicted
if the turbulent stresses exceed this surface restoring pressure. Similar to the model by Luo and
Svendsen (1996) it is possible to derive the sizes of the daughter bubbles as random variables
depending on the breakup rate. One major point of critique regarding the model by Lehr et al.
(2002) is that the assumption that bubbles deform into a cylindrical shape does not reflect the
real bubble deformation as shown by experiments (see, e.g., Risso and Fabre, 1998).
Lastly, another criterion was derived by Hagesaether et al. (2002), who proposed that the
1 INTRODUCTION AND LITERATURE OVERVIEW 9

energy density of the impacting eddy must be higher or equal to the surface energy density of
the daughter bubbles generated by breakup. Since the energy densities are given by dividing
the respective energies by the corresponding bubble volumes, this criterion translates to the
fact that the surface deforming turbulent stresses must be larger than the surface restoring
pressures of the daughter bubbles. Thus, it is similar to the approach by Martı́nez-Bazán
et al. (1999a,b) with the difference that the restoring pressure of the daughter bubbles is used
instead of the surface restoring pressure of the parent bubble. From a physical point of view
this assumption by Hagesaether et al. (2002) makes sense, since the restoring pressure of the
daughter bubbles would act against the contraction of the neck and, thus, against the breakup
of the parent bubble. It is again possible to estimate the size of the daughter bubbles as
random variables following a distribution function derived from the breakup criterion.
In addition to the above criterion Hagesaether et al. (2002) proposed that the criterion by Luo
and Svendsen (1996) has to hold as well, i.e., two breakup criteria are combined into one model.
Furthermore, it is assumed by Hagesaether et al. (2002) that all turbulent fluctuations up to the
size of the parent bubble play a role for breakup, which are accounted for by integrating over
all turbulent scales up to the size of the parent bubble. For the case of small-scaled turbulent
fluctuations the criterion by Hagesaether et al. (2002) yields a lower boundary for the possible
sizes of the daughter bubbles, while the criterion by Luo and Svendsen (1996) provides an upper
boundary. Then, the size of the daughter bubbles is randomly chosen between the lower and the
upper boundary values following a size distribution which is connected to the breakup criterion.
Note that the assumption that all turbulent scales up to the size of the parent bubble play
a role for the breakup contradicts experimental results by Andersson and Andersson (2006).
However, if only turbulent fluctuations with a size equal to the diameter of the parent bubble
are taken into account, a We number exceeding the criterion by Hagesaether et al. (2002)
automatically fulfills the criterion by Luo and Svendsen (1996). Hence, all diameters of the
daughter bubbles larger than the lower boundary given by the criterion by Hagesaether et al.
(2002) up to equally-sized breakup are possible. Consequently, in this case the criterion by
Luo and Svendsen (1996) does not play a role for breakup. The same reasoning applies for
the model by Wang et al. (2003), who combined the breakup criteria by Luo and Svendsen
(1996) and Lehr et al. (2002). Note that only some of the breakup models available in the
literature are mentioned here. For a more detailed summary, the interested reader is referred
to the extensive review by Solsvik et al. (2013).
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 10

Lastly, it has to be noted that the post-breakup treatment of the bubbles is also a crucial point
for the prediction of turbulent bubble-laden flows in the Euler–Lagrange framework. Since all
bubbles are tracked individually, an axis along which the daughter bubbles separate from each
other and a velocity at which they separate is required. To the best of the authors’ knowledge,
no model exists for both quantities. In the mostly used framework of Euler–Euler predictions it
is not necessary to estimate the separation axis and the separation velocity, since only number
distributions are considered. In the Euler–Lagrange predictions by Lau et al. (2014) and Jain
et al. (2014) the daughter bubbles were simply separated along an arbitrary direction with
a separation velocity of zero. Consequently, the objective of this study is two-fold: First,
the most promising breakup model derived in the Euler–Euler context is identified based on
the arguments discussed above. This model is transferred to the Euler–Lagrange framework
including the estimation of the size of the daughter bubbles. Second, a physically motivated
model for the separation axis and the separation velocity has to be incorporated in order to
make the breakup model applicable in the Euler–Lagrange context.

2. Breakup model and post-breakup treatment

Table 1 summarizes the literature review concerning the turbulence-induced breakup models
and their main advantages and drawbacks. Based on these considerations the breakup model
proposed by Hagesaether et al. (2002) is chosen in the present study as the starting point
for further developments. It is based on a physically reasonable assumption and provides
an estimate for the size of the daughter bubbles, which can be readily applied in the Euler–
Lagrange framework.

2.1. Breakup model of Hagesaether et al. (2002)


The starting point for the breakup modeling is that the parent bubble of diameter db,p is
exposed to turbulent velocity fluctuations hu0i u0i idb,p of the same size. The associated pressure
fluctuations deform the surface of the bubble. If the deformation is sufficiently strong, the
bubble breaks into two daughter bubbles with diameters db,s and db,` , where the subscripts ’s’
and ’`’ denote the smaller and the larger daughter bubble, respectively. In order to model the
breakup process, Hagesaether et al. (2002) proposed that the energy density wt of the turbulent
eddy must be higher or equal to the surface energy density wb,s of the smaller daughter bubble:

wt ≥ wb,s . (1)
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 11

Table 1: Overview of well-known breakup models and their critical Weber numbers Wecrit , which
have to be exceeded for breakup to occur. Additionally, comments about their advantages
and disadvantages are made.

Model Wecrit Comments


Prince and Blanch (1990) 2.3 Based on specific experiment, hence,
not generally valid
No daughter size estimation possible
h d2 2 i
b,s +db,`
Luo and Svendsen (1996) 12 d2b,p
−1 Daughter size obtainable
Breakup estimated for arbitrary
small We (critical for E–L predic-
tions)
Martı́nez-Bazán et al. (1999a,b) 12 Physical mechanism reasonable, al-
though breakup is estimated for
slightly deformed bubbles
No daughter size estimation possible
db,p
Lehr et al. (2002) 4 db,s
Daughter size obtainable
Assumption of cylindrical shape not
realistic
db,p
Hagesaether et al. (2002) 12 db,s
Daughter size obtainable
Physical mechanism reasonable

According to Hagesaether et al. (2002) the energy density of the turbulent eddy is given by the
turbulent kinetic energy Et of the fluid fluctuations divided by the volume of the corresponding
eddy. Since the deforming fluctuations are assumed to be of the same size as the diameter
db,p of the parent bubble (Hagesaether et al., 2002), the volume of the eddy is equal to the
volume Vb,p of the parent bubble. Hence, the energy density of the turbulent eddy deforming
the parent bubble is given by:

Et 1
wt = = ρf hu0i u0i idb,p . (2)
Vb,p 2

Here, ρf is the density of the fluid. On the other hand, the surface energy density of the smaller
daughter bubble is estimated by the ratio of the surface energy Es,s of the smaller daughter
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 12

bubble to its volume Vb,s :


Es,s σ
wb,s = =6 , (3)
Vb,s db,s
where σ denotes the surface tension of the bubble. Since the relations for the surface and the
volume of a sphere are used to obtain Eq. (3), this implies that the generated daughter bubble
is perfectly spherical, which may not be true in reality (see, e.g., the experiments by Andersson
and Andersson, 2006; Risso and Fabre, 1998). Hence, relation (1) is an inequality, which reflects
the fact that not all of the energy density of the turbulent eddy will be transformed into the
surface energy density of a spherical daughter bubble. A part of wt will be contained in the
energy density of the deformation, which is not accounted for by the model.
Note that the energy density of the turbulent eddy described by Eq. (2) corresponds to the
turbulent pressure fluctuations associated with the velocity fluctuations, while the surface
energy density defined by Eq. (3) corresponds to the surface restoring pressure of the smaller
daughter bubble. Consequently, from a physical point of view the breakup criterion (1) means
that the turbulent pressure fluctuations of the fluid have to be larger than the restoring pressure
of the smaller daughter bubble trying to prevent the contraction of the neck formed between
the two daughter bubbles. Thus, the model by Hagesaether et al. (2002) captures the direct
breakup of the parent bubble. However, the case where a small daughter bubble is pinched off
during the backflow into the deformed bubble described above can not be accounted for.
Inserting Eq. (2) and (3) into Eq. (1) allows to formulate the breakup criterion by Hagesaether
et al. (2002) in form of a dimensionless Weber number:

db,p −1/3
We ≥ 12 = 12 fV , (4)
db,s

In Eq. (4) the volume fraction fV denotes the ratio of the volume of the smaller daughter
bubble to the volume of the parent bubble. The dimensionless Weber is defined by:

ρf db,p hu0i u0i idb,p


We = . (5)
σ
−1/3
The ratio fV reaches its smallest value of 0.5−1/3 ≈ 1.26 for breakup into two equally-sized
daughter bubbles. Consequently, in order to break up a bubble, the Weber number has to be
at least Wemin = 12 × 0.5−1/3 ≈ 15.12, which leads to two daughter bubbles of the same size.
Breakup resulting in two unequally-sized bubbles requires a larger Weber number, since the
associated restoring pressure resisting the breakup is larger.
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 13

In order to determine the Weber number by Eq. (5), the turbulent fluctuations hu0i u0i idb,p over
the size of the parent bubble have to be estimated. In most of the existing breakup models (see,
e.g., Luo and Svendsen, 1996; Martı́nez-Bazán et al., 1999a; Lehr et al., 2002) the fluctuations
are assumed to be in the inertial subrange of turbulence. Hence, the fluctuations hu0i u0i idb,p
over a distance db,p can be estimated by the relation of Batchelor (1953):

2
hu0i u0i idb,p = β ( db,p ) 3 , (6)

with  denoting the energy dissipation rate of the turbulent flow at the position of the bubble
and β represents a proportionality constant. Following Hinze (1975), the dissipation rate is
defined by:
 = 2 νf Sij Sij , (7)

where νf is the kinematic viscosity of the fluid and Sij denotes the strain rate tensor, which
can be calculated based on the derivatives of the instantaneous fluid velocities:
 
1 ∂ui ∂uj
Sij = + . (8)
2 ∂xj ∂xi

According to Batchelor (1953) the proportionality constant β is related to the Kolmogorov


constant cK :  
9 1
β= Γ cK , (9)
5 3
1

where Γ is the Gamma-function, which has a value of Γ 3
≈ 2.68. The value of the Kol-
mogorov constant varies throughout the literature. For example, Batchelor (1953) gave a value
of cK = 1.7, while Tennekes and Lumley (1972) set the constant to cK = 1.5. In the present
work the Kolmogorov constant is set to cK = 1.62 based on the review by Sreenivasan (1995)
who evaluated a large number of experimental studies on the value of the Kolmogorov constant.
Consequently, the proportionality constant is given by β ≈ 7.81.
As mentioned in Sect. 1, in recent studies some authors (Solsvik and Jakobsen, 2016b,a; Karimi
and Andersson, 2019; Castellano et al., 2019) critically annotated the assumption of bubbles
being only subjected to turbulent structures in the inertial subrange, i.e., the whole spectrum
of turbulence should be considered. This is achieved by replacing Eq. (6) for the turbulent
fluctuations by other relations, which are based on model energy spectra like the one by Pope
(2000). Based on this energy spectrum, Solsvik and Jakobsen (2016a) proposed a highly com-
plex relation for the fluctuations in the inertial and energy-containing subranges of turbulence,
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 14

which does not account for the dissipative subrange. While such relations would allow for a
more physical description of the breakup-inducing turbulent fluctuations, Eq. (6) is neverthe-
less used in the present work for the sake of simplicity. This choice is motivated by the fact
that the size of the majority of the bubbles (c.f., Sect. 4) in the presently considered test case
by Martı́nez-Bazán et al. (1999a,b) are within the inertial subrange of turbulence. This can
be seen from the well-known relation lID ≈ 60 η by Pope (2000) for the boundary between the
inertial and the dissipative subranges of turbulence. Based on the value of η/dj ≈ 2 × 10−3
measured by Sadeghi et al. (2018) for a jet flow at comparable Reynolds numbers (Re = 30, 000
and 50, 000), one obtains a boundary of lID /dj ≈ 0.12, which is exceeded by a majority of the
bubbles. For an overview on available model energy spectra and the resulting relations for the
fluctuations, the reader is referred to the review by Solsvik and Jakobsen (2016b).

2.2. Post-breakup treatment


After the occurrence of breakup is determined by evaluating if We exceeds the required mini-
mum value Wemin , it remains to estimate the post-breakup conditions of the daughter bubbles.
This includes the diameters db,s and db,` of the daughters as well as the axis x sep along which the
bubbles separate and the corresponding velocities u b,s and u b,` of the daughter bubbles. Ad-
ditionally, in the framework of Euler–Lagrange predictions a breakup time ∆tbreak is required
to prevent repeated breakups at successive time steps.

2.2.1. Size of daughter bubbles


Before the daughter bubble size of each individual breakup event can be estimated, the number
of generated bubbles has to be specified. As mentioned in the literature overview (Sect. 1),
binary breakup is the most commonly observed type of breakup for bubbles (see, e.g., Risso
and Fabre, 1998; Andersson and Andersson, 2006; Hesketh et al., 1991a,b; Walter and Blanch,
1986; Stewart, 1995; Martı́nez-Bazán et al., 1999a; Wilkinson et al., 1993). The observation of
non-binary breakup (more than two daughter bubbles) by Risso and Fabre (1998) and Solsvik
and Jakobsen (2015) may be caused by the relatively large diameters (> 2 mm) of the parent
bubbles in these experiments (water). Since the present work is mostly considering small
bubbles (db . 1.5 mm) in water, the assumption of binary breakup is kept. Furthermore,
in the experiment by Martı́nez-Bazán et al. (1999a,b) forming the basis of the present test
case, binary breakup was mostly observed. It was stated that binary breakup events are
characterized by low to moderate Weber numbers. Only for rarely occurring large values of
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 15

We the probability of non-binary breakup increases. Since in the present study the considered
bubbles are small and the observed Weber numbers are only moderately larger than Wemin ,
the breakup is assumed to be binary. However, it has to be kept in mind, that non-binary
breakups of bubbles may occur, but are not taken into account by the present model.
The typical approach in the literature on bubble breakup based on Euler–Euler predictions is
to model the size of the daughter bubbles according to a size distribution. This concept was
also pursued in the Euler–Lagrange predictions of Lau et al. (2014) and Jain et al. (2014). In
order to circumvent the need to randomly choose the size of the daughters, it is assumed here
that breakup can only occur if criterion (4) is fulfilled exactly, i.e., the inequality is transformed
into an equality. Consequently, under the condition We ≥ Wemin it is possible to solve Eq. (4)
for the diameter db,s of the smaller daughter bubble:

12
db,s = db,p . (10)
We

Since the total bubble volume has to be conserved, the diameter of the larger daughter bubble
can be readily determined:
 13
db,` = d3b,p − d3b,s . (11)

Note that for We = Wemin Eqs. (10) and (11) yield two equally-sized daughter bubbles. With
increasing We the size of the smaller daughter bubble decreases, which is plausible, since the
stresses required to directly break off a small daughter bubble becomes larger as well.

2.2.2. Separation axis


Next, it is necessary to specify the axis of separation xsep along which the bubbles separate after
breakup. To the best of the authors’ knowledge the separation axis of bubbles has not been
the subject of numerical or experimental investigations. While Euler–Euler predictions do not
need to specify this axis, the existing Euler–Lagrange predictions considering breakup by Lau
et al. (2014) and Jain et al. (2014) simply choose this axis randomly. However, an experimental
study by Saha (2013) exists, who investigated the direction of separation of solid particles after
the breakup of agglomerates in a quasi-homogeneous and isotropic turbulent flow inside a tank.
Saha (2013) observed about 100 binary breakup events and measured the fluid stresses around
the agglomerates at the instance of breakage based on tracked tracer particles. He found that
the axis of separation x sep of the solid agglomerates was mostly aligned with the direction
in which the agglomerates experienced the largest fluid stresses. Breuer and Khalifa (2019)
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 16

successfully applied the idea of Saha (2013) in their improved model for the turbulence-induced
breakup of agglomerates within the framework of Euler–Lagrange simulations. Although Saha
(2013) considered solid particles, it is assumed that the breakup of gas bubbles is governed
by a similar mechanism. Since the parent bubble is deformed by the fluid stresses around its
surface, it is plausible to assume that the bubble is stretched in the direction of the largest
stresses. Hence, if the bubble breaks up, the generated daughter bubbles should separate in
the corresponding direction as well.
Saha (2013) analyzed the fluid stresses around the agglomerates by determining the eigenvalues
λ and the corresponding eigenvectors x eig of the strain rate tensor Sij (Eq. (8)), which are
defined by the characteristic equation:

Sij xeig,j = λ xeig,i . (12)

He found that Sij always has one largest, positive eigenvalue λ` and one smallest, negative
eigenvalue λs . The third eigenvalue λm is somewhere in between λ` and λs and can be either
positive or negative. According to Saha (2013) a positive eigenvalue means that a fluid element
is stretched by the fluid stresses in the direction of the corresponding eigenvector. Hence, x`eig
corresponding to the largest, positive eigenvalue λ` is the direction of the largest fluid stresses
and, thus, the direction in which the particles separate. Consequently, in the present case the
axis of separation of the daughter bubbles is given by:

x sep = x `eig . (13)

In order to determine the eigenvalues and eigenvectors of the strain rate tensor Sij , a method
given by Press et al. (2007) is applied: First, Sij is decomposed into an orthogonal matrix Qij
and a real, symmetric, tridiagonal matrix Aij by a Householder transformation (Press et al.,
2007). Afterwards, a QL-algorithm with implicit shifts also given by Press et al. (2007) is used
to obtain a lower triangular matrix, where the eigenvalues λ of Sij appear on the diagonal.
Additionally, this method yields the corresponding eigenvectors x eig , which are sorted by a
sorting algorithm to obtain the largest eigenvalue λ` .

2.2.3. Separation velocity


Since the daughter bubbles do not overlap after the breakup, their centers of mass should
at least be separated by a distance equal to the sum of their radii. In the Euler–Lagrange
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 17

framework it is in principle possible to set the positions of the daughter bubbles along the axis
of separation x sep starting from the last position of the parent bubble. However, this poses the
difficulty that the daughter bubbles may overlap with other surrounding bubbles. Furthermore,
if a block-structured grid is used to predict the fluid phase on a parallel computer, it is also
possible that the daughter bubbles move out of the block executed by a certain processor
which requires time-consuming exchange processes between the processors. In order to avoid
the large computational effort required for the solution of these problems, the positions of
both daughter bubbles are set equal to the position of the parent bubble. To prevent the
immediate re-coalescence of the daughter bubbles, the velocities u rb,s and u rb,` with which the
daughter bubbles separate from each other are estimated. Here the superscript ’r’ denotes that
the velocities are estimated in the frame of reference of the parent bubble moving with the
velocity u b,p . Note that the relative velocities u rb,s and u rb,` point in opposite directions along
the separation axis x sep :

u rb,s = urb,s x sep , (14)

u rb,` = −urb,` x sep , (15)

where the choice which velocity is positive and which is negative is arbitrary. Note that the
vector x sep is normalized.
r r
The estimation is based on an energy balance between the kinetic energies Ekin,s and Ekin,` of
the daughter bubbles in the moving frame of reference, the turbulent kinetic energy Eturb,db of
the velocity fluctuations exerted on the parent bubble during breakup and the kinetic energy
required for the breakup process:

r r
Ekin,s + Ekin,` = Eturb,db − Ebreak . (16)

As described in Sect. 1 prior to the breakup the parent bubble is deformed to a dumbbell-like
shape. The largest size of this deformed shape can be estimated based on the daughter bubbles
as follows:
ddb = db,s + db,` (17)

This larger size leads to additional stresses acting on the deformed bubble. If the bubble
breaks, the additional stresses act on the daughter bubbles and drive them away from each
other. Consequently, it is assumed that the turbulent kinetic energy Eturb,db is related to
the turbulent velocity fluctuations hu0i u0i iddb over the size ddb . The fluctuations hu0i u0i iddb are
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 18

estimated based on Eq. (6) given by Batchelor (1953), but with db,p replaced by the larger size
ddb . Since the fluctuations hu0i u0i idb,p over the size of the parent bubble are responsible for the
deformation of the parent bubble and the subsequent breakup, the corresponding kinetic energy
is denoted Ebreak . It is assumed that this energy is consumed by the breakup process and thus
does not contribute to the relative separation velocity. Hence, it is proposed to estimate the
excess kinetic energy available for the separation process based on the difference between the
velocity fluctuations hu0i u0i iddb exerted during breakup and the velocity fluctuations hu0i u0i idb,p
responsible for the breakup according to:

1 
∆E = Eturb,db − Ebreak = mf hu0i u0i iddb − hu0i u0i idb,p . (18)
2

Since the fluid fluctuations are exerted on the volume Vb,p of the parent bubble, mf = ρf Vb,p
in Eq. (18) denotes the mass of the fluid displaced by the parent bubble.
Consequently, the energy balance describing the relative daughter bubble velocities is given
by:
2 2
ms urb,s + m` urb,` = 2 ∆E. (19)

Since the separating bubbles have to accelerate the fluid surrounding them, the masses ms and
m` in Eq. (19) are the masses mb,s and mb,` of the two daughter bubbles combined with the
corresponding added masses of the displaced fluid mAM,s and mAM,` :
 
0 ρf
ms = mb,s + mAM,s = ρb Vb,s 1 + Cm , (20)
ρb
 
0 ρf
m` = mb,` + mAM,` = ρb Vb,` 1 + Cm , (21)
ρb
0
where Cm denotes the added-mass coefficient adjusted for the case that two bubbles are in
close vicinity to each other. Furthermore, it was shown by Lamb (1932) that the added-mass
coefficient depends on the diameters of the bubbles. Based on this work, Kamp et al. (2001)
demonstrated that the added-mass coefficient for two bubbles of different size can be written
according to:  3
0 1 L N − M2 db,s + db,`
Cm = , (22)
8 L − 2M + N db,s db,`
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 19

where L, N and M are defined by:


" #
d3b,` d3b,`
L = d3b,s 1 + 3 +3 , (23)
(db,s + 2 db,` )3 (2 db,s + 3 db,` )3
" #
d3b,s d3b,s
N= d3b,` 1+3 +3 , (24)
(2 db,s + db,` )3 (3 db,s + 2 db,` )3
 3
251 db,s db,`
M= . (25)
72 db,s + db,`
0
Consequently, the adjusted added-mass coefficient varies from Cm = 0.8 for equally-sized bub-
0
bles (db,s /db,` = 1) to Cm = 0.2, if the ratio of the diameters tends to zero (db,s /db,` → 0).
The actual velocities urb,s and urb,` of the daughter bubbles after breakup are obtained by the
additional application of the conservation of momentum. In the frame of reference moving
with the velocity u b,p of the parent bubble, the momentum balance in the separation direction
requires the total momentum of the daughter bubbles to be:

ms urb,s − m` urb,` = 0. (26)

Solving Eq. (26) for the relative velocity of the larger daughter bubble and inserting this into
Eq. (19) yields a relation containing only the velocity of the smaller daughter bubble as an
unknown, which can be readily solved. Using Eq. (26) again provides the velocity of the larger
daughter bubble. Consequently, the relative velocities of the daughter bubbles are given by:
  12
r 1 m`
ub,s = 2 ∆E , (27)
ms + m` ms
  12
r 1 ms
ub,` = 2 ∆E . (28)
ms + m` m`
Consequently, applying the definitions of mf , ms and m` to Eqs. (27) and (28) and inserting
into relations (14) and (15) provides the velocities of the daughter bubbles after the breakup
in the Eulerian frame of reference:
" # 12
1 d3b,` 
u b,s = u b,p + ρb 0 d3
hu0i u0i iddb − hu0i u0i idb,p x sep (29)
ρf
+ Cm b,s
" # 12
1 d3b,s 
u b,` = u b,p − ρb 0 d3
hu0i u0i iddb − hu0i u0i idb,p x sep (30)
ρf
+ Cm b,`

2.2.4. Breakup time


An additional difficulty arises in the framework of Euler–Lagrange predictions of breakup phe-
nomena in form of repeated breakups of the same bubble at several successive time steps. This
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 20

happens when a large bubble breaks up due to strong fluid fluctuations leading to a small and
a large daughter bubble. Hence, the larger daughter bubble may fulfill the breakup criterion in
the subsequent time step again causing the bubble to further split into two daughter bubbles,
where the larger daughter bubble is slightly smaller than the previous large daughter bubble.
This process is repeated until the breakup criterion is not fulfilled anymore. From a numerical
point of view, this issue is related to the modeling of the breakup (deformation of the parent
bubble and formation of the daughters) in the Euler–Lagrange framework as an instantaneous
sudden process taking place within one computational time step ∆t. Consequently, if no coun-
termeasures are taken (e.g., in form of an idle time), the original parent bubble is unphysically
eroded into many small daughter bubbles, with a breakup rate that is directly depending on the
numerical time-step size and not on physical arguments. Since for LES the chosen time-step
sizes are typically quite small, such an artificial erosion process may take place very fast.
In order to prevent such unphysical behavior, it is proposed that the daughter bubbles gener-
ated by the breakup of a parent bubble can not break up during an idle time ∆tidle . Since it
takes a certain time period to deform and split the parent bubble, the breakup time ∆tbreak
is chosen for the idle time ∆tidle = ∆tbreak . Because the breakup is caused by turbulent fluc-
tuations, the breakup time should be connected to a turbulent time scale ∆tturb . Different
possibilities exist for the estimation of ∆tturb .
Andersson and Andersson (2006) observed that the breakup processes of air bubbles with a
diameter of 1 mm in turbulent water flows (Re = 66, 000) take approximately 2 to 3.4 ms,
which is roughly 1/2 to 2/3 of the turbulent time scale defined by the ratio of turbulent kinetic
energy k to the dissipation rate :
k
∆tturb,AA = . (31)

The turbulent kinetic energy k can be estimated by hu0i u0i idb,p /2 with the fluctuations given by
Eq. (6). Hence, the turbulent time scale given by Andersson and Andersson (2006) is:
2

k β db,p
3

∆tturb,AA = = . (32)
 2  31

Consequently, using the experimentally observed factor of CAA = 1/2 − 2/3 between the
breakup and the turbulent time scale, the breakup time according to Andersson and Andersson
(2006) is:
2
db,p
3
β
∆tbreak,AA = Cbreak,AA 1 , with Cbreak,AA = CAA . (33)
 3 2
2 BREAKUP MODEL AND POST-BREAKUP TREATMENT 21

Alternatively, a turbulent time scale can be defined by the eddy turnover time, which is ob-
tained by dividing a characteristic length scale by a characteristic velocity:

lc
∆tturb,et = . (34)
vc

Here, the diameter of the parent bubble is used as the characteristic length scale, since this
is the length scale of the turbulent fluctuations. The characteristic velocity can be defined in
different ways. One meaningful possibility is to use the square root of the turbulent kinetic
energy k 1/2 . Consequently, one obtains for the eddy turnover time:
  12 32
db,p 2 db,p
∆tturb,etk = 1 = 1 , (35)
k2 β 3
Hence, the breakup time is given by:
2
  12
db,p
3
2
∆tbreak,etk = Cbreak,etk 1 , with Cbreak,etk = . (36)
3 β

Another alternative is to define the characteristic velocity in Eq. (34) by the square root of the
1/2
turbulent fluctuations hu0i u0i idb,p . This leads to a slightly different breakup time of:
2
  12
db,p
3
1
∆tbreak,etf = Cbreak,etf 1 , with Cbreak,etf = , (37)
3 β

which differs from Eq. (36) by a factor of 2. Obviously, all formulations yield the same
functional relationship including the diameter of the parent bubble and the dissipation rate.
Solely the proportionality constants differ.
Note that Coulaloglou and Tavlarides (1977) also define a breakup time based on the same func-
tional relationship without further specification of the model constant. However, Coulaloglou
and Tavlarides (1977) and later other authors (e.g., Solsvik and Jakobsen, 2015) used this
breakup time to estimate a breakup frequency of the bubbles, i.e., it is assumed that a certain
fraction of the bubbles always break up after this time. It has to be clearly noted that this
is not the case in the present work. Here, ∆tbreak specifies the minimum period of time the
daughter bubbles cannot break up after their formation. After this idle time ∆tbreak breakup
is possible for the daughter bubbles depending on the local properties of the fluid and the
bubbles, i.e., whether the breakup condition is fulfilled at this new instant in time. Hence,
the phenomenon of a breakup cascade, where repeated breakups of bubbles formed from one
mother bubble are observed (Solsvik et al., 2016) can be captured.
3 SIMULATION METHODOLOGY 22

In order to justify the above definitions and to choose the most appropriate formulation,
the breakup times given by Eqs. (33), (36) and (37) are compared with an experimental
observation by Andersson and Andersson (2006) of the breakup of an air bubble in a turbulent
water flow (Re = 66, 000). The breakup time of the bubble with a diameter of db,p = 1 mm
was ∆tbreak = 2.25 ms. According to Andersson and Andersson (2006) the dissipation rate
in the fluid was  = 16 m2 /s3 . Consequently, inserting those parameters into Eq. (33) gives
∆tbreak,AA = 7.74 ms for CAA = 1/2 and ∆tbreak,AA = 10.3 ms for CAA = 2/3. Using definition
(36) yields a breakup time of ∆tbreak,etk = 2.01 ms, while Eq. (37) gives ∆tbreak,etf = 1.42 ms.
Consequently, in the present work it is assumed that the breakup time is equal to the turbulent
time scale based on the eddy turnover time. Comparing the results of the present simulations
to the experimental data by Martı́nez-Bazán et al. (1999a) showed that the estimation of the
breakup time by Eq. (37) yields the best agreement with the experiment (see Sect. 6.3). The
ratio of the breakup time and the time-step size of the simulation defines the number of time
steps, in which a bubble can not experience a further breakup process. This value is stored as
a property of the bubble and transported through the flow field.

3. Simulation methodology

An Euler–Lagrange approach combined with the large-eddy simulation technique is applied in


the present work to predict the evolution of the turbulent fluid flow and the dispersed phase
both reliably and cost-efficiently. Hence, the continuous phase is treated in an Eulerian frame
of reference by solving the conservation equations of mass and momentum. The dispersed
bubbles are tracked individually through the flow field in the Lagrangian frame of reference.
It was shown in previous studies that this approach can successfully predict phenomena oc-
curring in turbulent multiphase flows including the coupling between the phases (particle-fluid
and particle-particle), agglomeration of solid particles, coalescence of bubbles and droplets, de-
position of particles at walls and turbulent breakup of solid agglomerates (see, e.g., Alletto and
Breuer, 2012, 2013; Breuer and Alletto, 2012; Breuer and Almohammed, 2015; Almohammed
and Breuer, 2016a,b; Hoppe and Breuer, 2018; Almohammed and Breuer, 2019; Breuer and
Khalifa, 2019). Since all details of the simulation methodology were previously described in
Breuer and Hoppe (2017), only a brief summary is given here.
3 SIMULATION METHODOLOGY 23

3.1. Continuous phase

The governing equations of the fluid flow are the filtered, three-dimensional, time-dependent
Navier–Stokes equations. The additional subgrid-scale stress tensor caused by the filtering
process mimics the influence of the non-resolved small-scale structures on the resolved large
eddies. In the present work the subgrid-scale model of Smagorinsky (1963) is used where the
Smagorinsky ”constant” CS is determined dynamically by the procedure proposed by Germano
et al. (1991), which was later improved by Lilly (1992). The feedback effect of the dispersed
phase on the fluid due to two-way coupling effects is accounted for by an additional source
term in the momentum equation, which is described by the particle-source-in-cell method of
Crowe et al. (1977). The force on the fluid is determined indirectly based on the acceleration
of the bubbles, thereby considering all fluid-related forces in the source term.
The governing equations of the continuous phase are solved by the in-house code LESOCC based
on a finite-volume method for arbitrary non-orthogonal and block-structured grids (Breuer,
1998, 2000, 2002; Breuer et al., 2006). All fluxes are spatially discretized by the midpoint rule
with a linear interpolation of the variables to the cell faces leading to an accuracy of second
order. A low-storage Runge–Kutta method within the predictor-corrector scheme is applied to
achieve second-order accurate time-marching. The corrector step ensures conservation of mass
by solving the pressure correction equation. Lastly, the momentum interpolation technique of
Rhie and Chow (1983) is applied to ensure the appropriate coupling of the pressure and the
velocity fields on non-staggered grids.

3.2. Dispersed phase

The motion of the bubbles is predicted individually based on Newton’s second law, where the
fluid forces acting on the bubbles are derived from the displacement of a small sphere in a non-
uniform flow (Gatignol, 1983; Maxey and Riley, 1983). Owing to the point-particle approach
finite-size effects cannot be taken into account. Accordingly, the simulation methodology is
limited to small bubbles denoted as microbubbles as done in previous studies (see, e.g., Hoppe
and Breuer, 2018, and the citations therein).
The forces taken into account are the drag force, the gravity force, the buoyancy force, the
added-mass force, the pressure-gradient force and the lift force. The Basset history force is
neglected for the sake of simplicity and computational effort.
Different drag coefficients have to be applied for the drag force on clean and contaminated
3 SIMULATION METHODOLOGY 24

bubbles. For clean bubbles with slip condition on the surface the drag coefficient by Mei and
Klausner (1992) and Mei et al. (1994) is applied. For bubbles covered with contaminants
the classical formulation of the drag coefficient for a Stokes flow around a solid sphere is
used (Tomiyama et al., 1998) with the correction factor by Schiller and Naumann (1933)
for Reb ≤ 800. The effect of non-spherical bubbles is taken into account by the model of
Dijkhuizen et al. (2010), who describe the influence of the bubble shape by a separate drag
coefficient depending on the Eötvös number. Both drag coefficients are then combined.
The gravity and buoyancy forces are calculated directly. The added-mass force is estimated by
the difference of the substantial derivative of the fluid velocity at the bubble position and the
total derivative of the bubble velocity itself. Here the added-mass coefficient is set to CM = 1/2
for spherical bubbles (Brennen, 1982) assuming that in the present case the deviations from
sphericity are small. The pressure-gradient force is described by the formulation of Maxey and
Riley (1983).
The lift force is determined according to Auton (1987) and Auton et al. (1988). The corre-
sponding lift coefficient is given by Legendre and Magnaudet (1998) distinguishing between
the cases of high (5 < Reb ≤ 500) and small (Reb ≤ 1) bubble Reynolds numbers. In order
to cover the intermediate range of bubble Reynolds numbers, the lift coefficients for large and
small Reb are combined.
Wall collisions are accounted for by a fully elastic hard-sphere collision model without friction
(Crowe et al., 1998), i.e., in case of a wall impact the sign of the wall-normal velocity of the
bubble is inverted, while the tangential components remain unchanged. For more details see
Breuer et al. (2012) or Almohammed and Breuer (2016b).
The inter-bubble collisions are deterministically predicted using a recently developed collision
algorithm (Alletto and Breuer, 2012; Breuer and Alletto, 2012). In the first stage the bubble is
moved based on the equation of motion. In the second stage the occurrence of bubble-bubble
collisions during the first stage is examined for all bubbles by the concept of virtual cells. If
a collision takes place, the velocities of the colliding bubbles are predicted based on a hard-
sphere model. In contrast to the solid particle case (Alletto and Breuer, 2012, 2013; Breuer
and Alletto, 2012; Breuer and Almohammed, 2015) friction between the colliding bubbles is
presently not considered and a fully elastic collision is assumed. Inter-bubble collisions are a
prerequisite for possible coalescence processes of bubbles. Recently, an enhanced film drainage
model for the prediction of bubble coalescence was developed by Hoppe and Breuer (2018).
4 DESCRIPTION OF THE TEST CASE 25

This model can be switched on to investigate the effect of coalescence. In the present study
all simulations have been performed with coalescence considered except for one case where the
effect of coalescence was investigated by deactivating the model.
As shown by several authors (see, e.g., Armenio et al., 1999; Kuerten and Vreman, 2005;
Marchioli et al., 2008; Bianco et al., 2012; Innocenti et al., 2016) the smallest, unresolved
scales in LES have an non-negligible influence on the dispersed phase. To take the impact
of these scales into account, the subgrid-scale velocity fluctuations at the bubble position are
estimated by an extended version of the Langevin model of Pozorski and Apte (2009) described
in Breuer and Hoppe (2017).
The governing equations for the bubbles are solved in a two-step procedure. First, Newton’s
second law is solved in the physical space by a fourth-order Runge-Kutta scheme providing the
velocity of the bubble. The second integration determining the position of the bubble on the
block-structured curvilinear grid is done in the computational space avoiding time-consuming
search algorithms in the physical space (Breuer et al., 2006, 2007). Consequently, a highly
efficient tracking scheme capable of predicting the paths of millions of bubbles is obtained
(see, e.g., Alletto and Breuer, 2012, 2013; Breuer and Alletto, 2012; Breuer and Almohammed,
2015).

4. Description of the test case

The bubble breakup is investigated based on the experiments by Martı́nez-Bazán et al. (1999a,b)
who observed the breakup of bubbles injected into a fully-developed, turbulent jet flow. In
the experiments a water jet is discharged from the bottom vertically upwards into a hexagonal
tank. The height of the tank is ht = 1.8 m, while the minimal diameter (separation distance
of the parallel sides) is dt,m = 0.597 m. The recirculation caused by the high-momentum flow
is minimized by allowing the water to overflow at the top of the tank. The diameter of the jet
nozzle is dj = 3×10−3 m and the exit velocity of the jet is uE0 = 17 m/s leading to a jet Reynolds
number of ReEj = uE0 dj /νf = 51, 000 with νf the kinematic viscosity of water. However, since the
air bubbles are coaxially injected through a tube with a diameter of dt = 1.5 × 10−3 m, which
passes through the center of the jet nozzle, the open section of the jet is reduced, i.e., the open
section diameter defined by Martı́nez-Bazán (2018) is d∗j = (d2j − d2t )/(dj + dt ) = 1.5 × 10−3 m.
Consequently, the jet Reynolds number based on the exit velocity uE0 of the fluid, the open

section diameter d∗j and the kinematic viscosity νf is ReEj = uE0 d∗j /νf = 25, 500 (Martı́nez-
4 DESCRIPTION OF THE TEST CASE 26

Bazán et al., 1999a,b). Note that a jet flow out of a nozzle with a diameter of dj but without
taking the tube through the center of the nozzle into account (see Section 5) requires a lower
flow velocity u0 in order to ensure an equivalent mass flow rate as in the configuration used
by Martı́nez-Bazán et al. (1999a,b). Based on the ratio of the area of the open section of the
jet in the experiment and the area of the complete nozzle, the required velocity can be deter-
mined to be u0 = uE0 (d2j − d2t )/d2j = 12.75 m/s. Consequently, the resulting Reynolds number
is lower as well, i.e., Rej = u0 dj /νf = 38, 250. The air bubbles are injected through a needle
mounted at the end of the tube at an axial position of x/dj = 15. The diameter of the needle
is dn = 0.394 × 10−3 m. The gas flow rate is set to Qg = 1.2 × 10−6 m3 /s, which corresponds to
a superficial gas velocity of ug = 9.84 m/s.

4
Bubble size distr.
3
hd∗b,rel i = 0.514
PDF

0
0 0.2 0.4 0.6 0.8 1
db /dj

Figure 1: Volumetric bubble size distribution of the bubbles in the experiment by Martı́nez-Bazán
et al. (1999a,b) shortly after their release from the air injection needle at an axial position
of x/dj = 16.1.

The size distribution is given by Martı́nez-Bazán et al. (1999a,b) a short distance downstream
of the tip of the needle (x/dj = 16.1). Figure 1 displays the volumetric probability density
function of the bubble diameters together with the dimensionless mean diameter of the released
bubbles hd∗b,rel i = hdb,rel i/dj = 0.514, which corresponds to a dimensional mean bubble size of
hdb,rel i = 1.542 × 10−3 m. According to Martı́nez-Bazán et al. (1999a,b) the void fraction of the
bubbles is always less than Φ = 10−5 making coalescence negligible. After the release of the
bubbles, their evolution is tracked along the major axis of the jet by capturing 1000 images
of the flow at different axial positions with a camera (Martı́nez-Bazán et al., 1999a,b). The
flow is illuminated by a diffuse white light source positioned opposite to the camera allowing
to identify individual bubbles and their respective diameters. According to Martı́nez-Bazán
5 SIMULATION SETUP 27

et al. (1999a,b) overlapping bubbles are not a problem for the image processing technique, due
to the low volume fraction of the bubbles.

5. Simulation setup

The present bubble breakup model is investigated by tracking the evolution of bubbles inserted
into the unsteady spatially-evolving jet based on the experiments by Martı́nez-Bazán et al.
(1999a,b) described in Sect. 4. However, the tube going through the center of the jet in
the experiment is neglected here for the sake of simplicity and since Martı́nez-Bazán et al.
(1999a,b) also treated the jet flow as unaffected by the tube during the analysis of their results.
Consequently, the jet flow with a Reynolds number of 38, 250 out of a circular nozzle with the
same diameter dj as in the experiment and a jet velocity of ume is considered ensuring a mass
flow that is equivalent to the experiment. The jet is placed in a weak co-flow with a velocity
of u∗co = uco /u0,s = 0.025 in order to prevent numerical problems due to entrainment at the
outer boundaries. Hence, to guarantee a Reynolds number of Rej = ∆u dj /νf = 38, 250 based
on the jet diameter dj and the velocity difference ∆u = ujc − uco between the center of the jet
flow and the co-flow, the velocity in the center of the jet is u∗jc = ujc /u0 = 1.025.
The computational domain containing the jet flow is of cylindrical form. The computational
domain has a dimensionless radius of R/dj = 10 preventing an influence of the boundary on the
jet flow. The length of the computational domain in the downstream direction is xd /dj = 40,
which is sufficient to compare the simulation results with the experimental data of Martı́nez-
Bazán et al. (1999a,b). The flow is resolved by a block-structured grid consisting of 5 geometric
blocks and a total of about 10.65×106 control volumes. One block is placed in the middle and 4
geometric blocks are arranged around this central block building an O-grid arrangement. Since
large velocity gradients occur in the shear layer of the jet flow, a finer grid resolution is required
in this region, especially near the inflow plane. Consequently, in the streamwise direction 512
control volumes are used, which increase in size following a geometric stretching function. The
size of the first cell in streamwise direction is set to ∆x1 /dj = 2.5 × 10−2 and the stretching
factor is q ≈ 1.0039. Hence, the last cell has a size of ∆x512 /dj = 0.1784. The control volumes
of the central block have a dimensionless size of 1.2×10−2 in both the y- and the z-direction. In
the surrounding O–grid blocks Nr = 120 and Nθ = 160 control volumes are used in the radial
and circumferential direction, respectively. The control volumes are stretched in the radial
p
direction with a geometric stretching factor of 1.0044 up to a radius r/dj = y 2 + z 2 /dj ≤ 0.8.
5 SIMULATION SETUP 28

Consequently, a radial cell size of ∆r/dj = 1.35 × 10−2 is obtained at r/dj = 0.5, where the
largest velocity gradients appear in the shear layer close to the inflow plane. For r/dj > 0.8
the stretching factor is increased to 1.0444. In the circumferential direction the equally-spaced
cell size is set to ∆θ = 2 π/Nθ = 3.93 × 10−2 .
The inflow boundary condition for the continuous phase applied at the base of the cylinder
is that of a jet a short distance downstream of a typical nozzle exit. Hence, the streamwise
velocity profile is approximated by (Morris, 1976; Michalke, 1984):
   
∗ ∗
u∗j r rj
u (r) = uco + 1 − tanh b − , (38)
2 rj r

where u∗co and u∗j = 1 are the dimensionless velocities of the co-flow and the jet and rj = dj /2
is the radius of the nozzle. According to Michalke (1984) the factor b = 0.25 rj /δ2 is related to
the momentum thickness δ2 of the shear layer at the inlet. The value of δ2 /rj varies throughout
the literature. Here, the commonly used (see, e.g., Bogey and Bailly, 2006; Gohil et al., 2011,
2014) value of δ2 /rj = 1/20 is applied. At the lateral surface limiting the domain in the radial
direction, a no-slip boundary condition is set with the velocity of the co-flow. Finally, at the
outlet at the top surface of the domain a convective boundary condition with a fixed convection
velocity of u∗conv = 0.21 is used at this outlet, which is slightly larger than the average centerline
velocity of a jet flow at a distance of x/dj = 40 downstream of the nozzle, i.e., uc /uj = 0.161
(Hussein et al., 1994). The dimensionless time-step size is ∆t∗ = ∆t ume /dj = 6 × 10−3 . The
jet flow is allowed to evolve for a dimensionless time period of ∆T ∗ = 250 ensuring a fully
developed flow field. Then the bubbles are released continuously into the flow field. After a
dimensionless time period of ∆T ∗ = 300 a fully developed bubble distribution is obtained in
the jet. Afterwards, the statistics of the breakup parameters are averaged over a dimensionless
time interval of ∆T ∗ = 6000.
Since Martı́nez-Bazán et al. (1999a,b) provide the bubble size distribution at an axial position
of x/dj = 16.1, the bubbles are inserted at this position in the simulation as well. In order
to account for the radial migration of the bubbles after their detachment from the needle,
the bubbles are inserted randomly within a radius of rin /dj = 0.2 from the center of the jet.
The size of the released bubbles is randomly chosen by using an inverse transform sampling
method (Devroye, 1986) based on the probability density distribution shown in Fig. 1. The

bubbles are released individually with a dimensionless time lag of ∆Tint = 6.798 between the
successive releases, i.e., a single bubble is released every 1133 time steps. Since the average
6 RESULTS 29

size of the released bubbles based on the volumetric probability density is hd∗b,rel i = 0.514,
the corresponding dimensionless volumetric flow rate Q∗b = πhd∗b,rel i3 /(6∆Tint

) of the bubbles
is Q∗b = 1.046 × 10−2 , which is equal to the flow rate in the experiments (Martı́nez-Bazán

et al., 1999a,b). Note that this dimensionless time lag ∆Tint means that a total of 883 bubbles
are released during the averaging time period, while on average 100 bubbles are present in
the computational domain simultaneously. According to Martı́nez-Bazán et al. (1999a,b) the
mean velocity of the bubbles is equal to the mean fluid velocity shortly after the bubbles detach
from the needle. Therefore, the initial velocity of the bubbles is set to the fluid velocity at the
corresponding position. Since Martı́nez-Bazán et al. (1999a,b) used regular tap water in their
jet flow experiments, the surface condition of the bubbles is set to that of contaminated ones.
The dimensionless gravitational acceleration is g ∗ = g dj /u2me = 1.8103 × 10−4 . The surface
tension of the bubbles is given by Pallas and Harrison (1990) as σ = 72.86 × 103 N/m, which
corresponds to a dimensionless value of σ ∗ = σ/(ρf u2me dj ) = 1.49 × 10−4 . Lastly, the Hamaker
constant AH = 3.7 × 10−20 J of water molecules interacting in a vacuum (Israelachvili, 2011) is
applied, which is A∗H = AH /(ρf u2me d3j ) = 8.43 × 10−18 in dimensionless form.

6. Results

6.1. Continuous flow field


Before the breakup of the bubbles is investigated, the properties of the fluid flow are depicted
in Figs. 2, 3 and 4. In Fig. 2 the mean streamwise velocity of the fluid is compared with the
experimental results by Zaman and Hussain (1980) and Hussein et al. (1994) who investigated
jets at Reynolds numbers of 3.2 × 104 and 9.55 × 104 based on the jet diameter and the exit
velocity, respectively. It can be seen in Fig. 2(a) that the predicted mean streamwise velocity
along the centerline of the jet normalized by the inflow velocity ujc of the jet agrees well with
the result by Zaman and Hussain (1980). After a short distance downstream of the inlet
(x/dj ≈ 10) the centerline velocity follows a typical inversely proportional decrease with the
distance x/dj . A common (see, e.g., Panchapakesan and Lumley, 1993; Hussein et al., 1994)
expression of the centerline velocity is given by:
 −1
huf,x,c i x x0
= Bu − , (39)
ujc dj dj
where Bu and x0 are constants, which have to be determined by fitting Eq. (39) to the fluid
velocity along the centerline. Fitting Eq. (39) to the results of the present simulations yields
6 RESULTS 30

about Bu = 4.8 and x0 = 1.1. Hussein et al. (1994) obtained values of Bu = 5.8 and x0 = 4.0.
In the experiments by Wygnanski and Fiedler (1969) coefficients of Bu = 5.7 and x0 = 3.0
were estimated, while Panchapakesan and Lumley (1993) found Bu = 6.06 and x0 = 0. Hence,
considering the large scatter in the experimental data, the present results are in reasonable
agreement with the available reference literature.
In Fig. 2(b) the radial distributions of the mean streamwise velocity huf,x i normalized by the
corresponding centerline velocities huf,x,c i are depicted as a function of r/(x − x0 ) at four
downstream positions (x/dj = 20, 25, 30 and 35). The radial profiles of the mean streamwise
velocity in the self-similar part of the jet (x/dj > 10) ideally have to overlap at all downstream
positions, if plotted in this way. As visible in Fig. 2(b) this is the case for the present simulation.
Furthermore, at all downstream positions the results agree well with the experiment by Hussein
et al. (1994).

1.2
Zaman & Hussain (1980) 1.2 Hussein et al. (1994)
1 Present Sim. Present Sim., x/dj = 20
1
x/dj = 25
huf,x i/huf,x,c i

0.8
huf,x,c i/ujc

0.8 x/dj = 30
0.6 x/dj = 35
0.6
0.4 0.4
0.2 0.2
0 0
0 4 8 12 16 20 0 0.05 0.1 0.15 0.2 0.25
x/dj r/(x − x0 )
(a) (b)

Figure 2: Temporally and circumferentially averaged streamwise fluid velocity (a) along the center-
line of the jet normalized by the jet inlet velocity ujc and (b) in radial direction at four
different downstream positions normalized by the corresponding centerline velocity.

Figure 3 depicts the radial distributions of the velocity fluctuations normalized by the square
of the averaged centerline velocity in the streamwise (Fig. 3(a)), the radial (Fig. 3(b)) and the
circumferential (Fig. 3(c)) direction as well as the relevant Reynolds shear stress (Fig. 3(d)).
Again the curves should ideally overlap. The small deviations among each other and to the
results by Hussein et al. (1994) may be the result of the weak co-flow applied in the simulations.
In Fig. 4 two-dimensional contour plots of the instantaneous streamwise fluid velocity uf,x
(Fig. 4(a)), the mean streamwise fluid velocity huf,x i (Fig. 4(b)), the turbulent kinetic energy
6 RESULTS 31

0.12 0.12
Hussein et al. (1994) Hussein et al. (1994)
0.1 Present Sim., x = 20 0.1 Present Sim., x = 20
hu0f,x u0f,x i/huf,c i2

hu0f,r u0f,r i/huf,c i2


x = 25 x = 25
0.08 0.08
x = 30 x = 30
0.06 x = 35 0.06 x = 35

0.04 0.04
0.02 0.02
0 0
0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25
r/(x − x0 ) r/(x − x0 )
(a) (b)

0.12 0.04
Hussein et al. (1994) Hussein et al. (1994)
0.1 Present sim, x=20 Present Sim., x = 20
0.03
hu0f,θ u0f,θ i/huf,c i2

hu0f,x u0f,r i/huf,c i2

x=25 x = 25
0.08
x=30 x = 30
0.06 x=35 0.02 x = 35

0.04
0.01
0.02
0 0
0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25
r/(x − x0 ) r/(x − x0 )
(c) (d)

Figure 3: Time-averaged radial distributions of all normal Reynolds stress components and the
relevant Reynolds shear stress component. All quantities are normalized by the square of
the averaged centerline velocity.

hkturb i = 0.5 hu0f,i u0f,i i (Fig. 4(c)) and the Reynolds shear stress component hu0f,x u0f,r i (Fig. 4(d))
are displayed. The meanAll results are averaged in time as well as in circumferential direction.
In Fig. 4(a) a snapshot of the turbulent structures of the actual flow field at an arbitrarily
chosen instant in time is depicted. Figure 4(b) is supposed to illustrate the spreading of the jet
and the decrease of the centerline velocity along the axis. Figures 4(c) and (d) show as expected
that the highest turbulent fluctuations are found in the shear layer of the jet. Obviously, at
the release position of the bubbles (x/dj = 16.1) the streamwise velocity has already dropped
to about half of the exit velocity of the jet. Furthermore, the normal and shear stresses are
significantly attenuated at this position.
6 RESULTS 32

40 40

35 uf,x /u0 35 huf,x i/u0


30 30

25 25
x/dj

x/dj
20 20

15 15

10 10

5 5

0 0
-4 -2 0 2 4 -4 -2 0 2 4
y/dj y/dj

(a) huf,x i/u0 (b) huf,x i/u0

40 40

35 hkturb i/u20 35 hu′f,x u′f,r i/u20


30 30

25 25
x/dj

x/dj

20 20

15 15

10 10

5 5

0 0
-4 -2 0 2 4 -4 -2 0 2 4
y/dj y/dj

(c) hkturb i/u20 (d) hu0f,x u0f,r i/u20

Figure 4: (a) Contour plot of the instantaneous streamwise velocity. Contour plots of temporally
and circumferentially averaged fluid results: (b) mean streamwise velocity, (c) turbulent
kinetic energy and (d) Reynolds shear stress.

6.2. Bubble statistics

In order to illustrate the distribution of the bubbles inside the computational domain, a snap-
shot of the bubbles taken at a dimensionless instant in time of ∆T ∗ = 6000 after the release
of the bubbles is shown Fig. 5. The color of the bubbles is chosen according to their size. At
6 RESULTS 33

this instance a total of 100 bubbles are present in the computational domain. It can be seen
that the bubbles are carried downstream by the flow with a slight migration in the radial di-
rection. Due to turbulent velocity fluctuations the bubbles can break up on their way resulting
in bubbles of different size.

40

35 db /dj
0.6
0.56
0.52
30
0.48
x/dj

0.44
0.4
25 0.36
0.32
0.28
0.24
20
0.2

injection
15
-5 0 5
y/dj

Figure 5: Snapshot of the bubbles at a dimensionless instance of time of ∆T ∗ = 6000 after the
release of the bubbles.

The size distributions of the bubbles are shown in form of volumetric probability density func-
tions in Fig. 6. The results of the present breakup model are compared with the experimental
result by Martı́nez-Bazán et al. (1999a). In order to demonstrate the evolution of the bub-
ble sizes, the size distributions are depicted at three downstream positions at x/dj = 19.5
(Fig. 6(a)), x/dj = 25.0 (Fig. 6(b)) and x/dj = 34.1 (Fig. 6(c)). Note again that the bubbles
were inserted at a distance of x/dj = 16.1 from the nozzle exit with bubble diameters based
on the volumetric size distribution by Martı́nez-Bazán et al. (1999a) shown in Fig. 1. Conse-
quently, at this position the probability density functions of the bubble diameter of the present
simulation and by Martı́nez-Bazán et al. (1999a) coincide.
In Fig. 6(a) it can be seen that in the present results the bubble size distribution is slightly
shifted towards smaller bubble diameters compared with the experiment. Hence, the peak of
the bubble size distribution of the simulation is found at about db /dj = 0.3, while the peak is at
db /dj = 0.4 in the experimental results by Martı́nez-Bazán et al. (1999a). This means that too
6 RESULTS 34

6 6
x/dj = 19.5, Exp. x/dj = 25.0, Exp.
Sim. Sim.
4 4
PDF

PDF
2 2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
db /dj db /dj
(a) x/dj = 19.5 (b) x/dj = 25.0

6
x/dj = 34.1, Exp.
Sim.
4
PDF

0
0 0.2 0.4 0.6 0.8 1
db /dj
(c) x/dj = 34.1

Figure 6: Comparison of the volumetric probability density function of the bubble diameter of the
present simulation with the experiments by Martı́nez-Bazán et al. (1999a) at three down-
stream positions.

many bubbles with a diameter of less than db /dj = 0.4 are present in the simulation compared
with the experiment. Consequently, the breakup into small bubbles is somewhat overestimated
by the present model in the region shortly downstream of the release position of the bubbles.
Further downstream at x/dj = 25.0 the probability density functions shown in Fig. 6(b) are
nearly identical. The most noticeable difference between simulation and experiment is that
the PDF of the present simulation exhibits a more pronounced peak at about db /dj = 0.3.
Lastly, at x/dj = 34.1 the location of the peak of the bubble size distribution is shifted to a
slightly smaller diameter (db /dj ≈ 0.22) in the experiments by Martı́nez-Bazán et al. (1999a),
while the peak in the simulation remains at db /dj = 0.3. This indicates that breakup is
underestimated downstream of x/dj ≈ 20 leading to the relatively small differences between
6 RESULTS 35

the size distributions at x/dj = 25.0 and 34.1 in the simulation.


In Fig. 7 the bubble number flux of the present simulation is compared with the experiment. In
the simulation the bubble number flux is determined by first counting the number of bubbles
passing through a certain downstream plane during the averaging time period (∆T ∗ = 6000).
Afterwards, the resulting bubble numbers at each downstream position are normalized by the
sim
number of totally released bubbles (Nb,tot = 883) during this time period. In Martı́nez-Bazán
et al. (1999a) the bubble number flux was obtained by counting the total number of bubbles at
several downstream positions during a certain time interval. However, the exact time interval
is not specified in Martı́nez-Bazán et al. (1999a). To circumvent this problem, the bubble
number flux given by Martı́nez-Bazán et al. (1999a) is normalized in the present study by the
exp
total number (Nb,tot = 5785) of bubbles counted at the first downstream position (x/dj = 16.1)
resulting in the quantity depicted in Fig. 7.
The bubble number flux reflects the behavior of the present breakup model described above.
Shortly after the insertion of the bubbles the breakup rate is overestimated leading to a too
large number of small bubbles. Consequently, the normalized bubble number flux is slightly
overestimated by the present simulation. Downstream of the position x/dj = 23 the normalized
bubble number flux does not increase any more due to the absence of further breakup processes
observed in the simulations. In the experiment by Martı́nez-Bazán et al. (1999a) this plateau
is reached further downstream at x/dj ≈ 26. In other words, the breakup criterion (4) is only
rarely fulfilled downstream of this position. The maximum of the normalized bubble number
flux differs between the prediction and the experiment. In the experiment a maximum of
Nb /Nb,rel. ≈ 4.7 is reached, while the simulation only reaches a maximum of Nb /Nb,rel. ≈ 4.1.
This is a direct consequence of the overestimated number of breakup processes into bubbles
with a diameter of db /dj = 0.3, which is too large compared with the experiment (c.f., Fig. 6(c)).
In other words, a larger number of smaller bubbles exist in the experiment, while fewer bubbles
with a larger diameter are present in the simulation.

6.3. Influence of the time lag ∆tidle

In Sect. 2.2.4 three possibilities to estimate the time lag based on a turbulent breakup time are
presented: ∆tbreak,AA , ∆tbreak,etk or ∆tbreak,etf defined by Eqs. (33), (36) and (37), respectively.
All formulations have the same functional dependency on the parent bubble diameter and the
dissipation rate, but the proportionality constant differs. Hence, for the same parameter values
6 RESULTS 36

Nb /Nb,rel.
3

2
Exp.
Sim.
1
15 20 25 30 35 40
x/dj

Figure 7: Number of bubbles passing through a certain downstream plane within ∆T ∗ = 6000
normalized by the number of released bubbles: Comparison of the present simulations
with the experiments by Martı́nez-Bazán et al. (1999a).


the idle time ∆tbreak,etk estimated by Eq. (36) is by a factor of 2 ≈ 1.41 larger than the time
lag obtained by Eq. (37). Using Eq. (33) proposed by Andersson and Andersson (2006) yields
a time lag ∆tbreak,AA which is by a factor of CAA β 3/2 /2 ≈ 5.46 larger than the value obtained
by Eq. (37), if CAA = 1/2 and β ≈ 7.81 is inserted. Remember that ∆tbreak,etf was chosen as
the standard idle time in the present work, i.e., all other results presented here are obtained
using Eq. (37). The influence of the different estimations is investigated next.
In Fig. 8 the results of simulations carried out with different approaches for the time lag
are shown in form of volumetric probability density functions of the bubble diameter at three
downstream positions. The experimental results by Martı́nez-Bazán et al. (1999a) are included
as a reference. It is obvious from Fig. 8 that the results obtained with Eq. (36) are very
similar to the results for Eq. (37), which can be readily explained by the relatively small
difference between ∆tbreak,etk and ∆tbreak,etf . Furthermore, one can see that the agreement
with the experiment is slightly worse in case of ∆tbreak,etk , e.g., at a downstream position of
x/dj = 25.0 (Fig. 8(c)) the peak of the volumetric probability density function is found at a
somewhat too large bubble diameter of db /dj = 0.35. The explanation for this observation is
the larger idle time of the daughter bubbles after a breakup event in the case of ∆tbreak,etk . This
prevents the breakup of the daughter bubbles for a longer period of time resulting in larger
bubbles. The same reasoning applies to the third case considered, where the time lag is given
by ∆tbreak,AA . Due to the even larger idle time compared to ∆tbreak,etf , the shift towards larger
bubble diameters is more pronounced. Lastly, it can be concluded from Fig. 8 that the results
6 RESULTS 37

6 6
∆tbreak,AA ∆tbreak,AA
∆tbreak,etk ∆tbreak,etk
∆tbreak,etf ∆tbreak,etf
4 4
Exp. Exp.
PDF

PDF
2 2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
db /dj db /dj
(a) x/dj = 19.5 (b) x/dj = 25.0

6
∆tbreak,AA
∆tbreak,etk
∆tbreak,etf
4
Exp.
PDF

0
0 0.2 0.4 0.6 0.8 1
db /dj
(c) x/dj = 34.1

Figure 8: Comparison of the volumetric probability density functions of the bubble diameter of the
present simulations with different estimations of the idle time ∆tidle = ∆tbreak at three
downstream positions. The experimental results by Martı́nez-Bazán et al. (1999a) are
included as a reference.

of the simulations agree best with the experimental data by Martı́nez-Bazán et al. (1999a), if
the time lag is estimated by ∆tbreak,etf motivating the choice in the present work.

6.4. Influence of bubble coalescence

The influence of bubble coalescence is considered next. This is achieved by deactivating the
coalescence model in the present simulation, while the bubble properties are unchanged.
In Fig. 9 the results of the simulation without coalescence are compared with the simulation
considering bubble coalescence and the experimental results by Martı́nez-Bazán et al. (1999a).
It is visible in Fig. 9(a) that in the region shortly downstream of the location of the bubble
insertion, coalescence plays only a marginal role. Further downstream (x/dj = 25.0) shown
6 RESULTS 38

6 6
x/dj = 19.5, Exp. x/dj = 25.0, Exp.
Sim. with coal. Sim. with coal.
Sim. without coal. Sim. without coal.
4 4
PDF

PDF
2 2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
db /dj db /dj
(a) x/dj = 19.5 (b) x/dj = 25.0

6
x/dj = 34.1, Exp.
Sim. with coal.
Sim. without coal.
4
PDF

0
0 0.2 0.4 0.6 0.8 1
db /dj
(c) x/dj = 34.1

Figure 9: Comparison of the volumetric probability density function of the bubble diameter of the
present simulations with and without coalescence with the experiments by Martı́nez-Bazán
et al. (1999a) at three downstream positions.

in Fig.9(b) neglecting coalescence leads to a larger peak value of the bubble size distribution
at about db /dj = 0.25 compared to the case with coalescence included. For bubbles with a
diameter larger than db /dj = 0.45 the probability densities without considering coalescence
are too small. Consequently, in the simulations taking coalescence into account the presence
of bubbles in this larger diameter range is slightly higher due to coalescence of bubbles that
previously broke up. Lastly, Fig.9(c) depicts the bubble size distribution at x/dj = 34.0. The
most noticeable deviation to the case considering coalescence is again the position of the peak of
the size distribution, which agrees slightly better with the experiment. It is obvious from Fig. 9
that the overall influence of the coalescence of bubbles is relatively small. This observation is
in accordance with the results by Martı́nez-Bazán et al. (1999a).
6 RESULTS 39

6.5. Influence of bubble properties

Lastly, the surface tension of the bubbles is varied in order to investigate the influence of
the bubble properties on breakup. On the one hand, a 30 percent smaller surface tension of
σ = 51.0 × 10−3 N/m is applied (standard case: σ = 72.86 × 10−3 N/m). On the other hand,
a 70 percent larger value of σ = 123.86 × 10−3 N/m is used. The corresponding dimensionless
surface tensions are σ ∗ = 1.05 × 10−4 and 2.54 × 10−4 , respectively. Remember that in the
preceding investigations the usual surface tension of an air-water interface of was used, which
corresponds to σ ∗ = 1.49 × 10−4 .
Figure 10 compares the volumetric probability density functions of the bubble diameter for
the three different values of σ ∗ at two downstream positions of x/dj = 19.5 and 34.0. The
experimental results by Martı́nez-Bazán et al. (1999a) are included as a reference. The size
distribution at x/dj = 25.0 is omitted here for the sake of brevity. It is apparent from Figs. 10(a)
and (b) that decreasing the surface tension leads to a shift in the bubble size distribution
towards smaller bubble diameters. For example, at the downstream position of x/dj = 19.5
the peak value of the size distribution for the case of σ ∗ = 1.05 × 10−4 is found at about
db /dj = 0.18, while it is at about db /dj = 0.3 for σ ∗ = 1.49 × 10−4 . On the other hand,
increasing the surface tension to σ ∗ = 2.54 × 10−4 causes the movement of the peak of the size
distribution towards larger bubble diameters, i.e., the peak is found at roughly db /dj = 0.4 for
both downstream positions shown in Fig. 10.
These results can be readily explained by considering the definition of the dimensionless Weber
number given by Eq. (5) and the breakup criterion (4), which requires a minimum Weber
number of Wemin = 15.12 for breakup to occur (see Sect. 2.1). Since the surface tension
appears in the denominator of Eq. (5), decreasing σ ∗ makes the fulfillment of criterion (4)
easier, i.e., less intense turbulent fluctuations are required for breakup. Hence, the breakup
into smaller daughter bubbles is augmented compared to the case of σ ∗ = 1.49×10−4 leading to
the shift of the peak position of the bubble size distribution towards smaller bubble diameters
in Fig. 10. Increasing the surface tension has the opposite effect, i.e., the fulfillment of the
breakup criterion requires stronger turbulent fluctuations. From a physical point of view both
findings make sense. A smaller surface tension means that the bubble deforms more readily
increasing the probability of breakup into smaller bubbles, while an increased surface tension
results in a bubble that is harder to deform and, hence, more difficult to break up.
7 CONCLUSIONS 40

6 6
σ ∗ = 1.05 × 10−4 σ ∗ = 1.05 × 10−4
σ ∗ = 1.49 × 10−4 σ ∗ = 1.49 × 10−4
σ ∗ = 2.54 × 10−4 σ ∗ = 2.54 × 10−4
4 4
Exp. Exp.
PDF

PDF
2 2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
db /dj db /dj
(a) x/dj = 19.5 (b) x/dj = 34.1

Figure 10: Comparison of the volumetric probability density functions of the bubble diameter for
three surface tension values at two downstream positions. The experimental results by
Martı́nez-Bazán et al. (1999a) are included as a reference.

7. Conclusions

The objective of the present investigation was an improved modeling concept for microbubble
breakup in the framework of the Euler-Lagrange approach relying on the large-eddy simulation
technique. A detailed evaluation of existing breakup models revealed that the model introduced
by Hagesaether et al. (2002) is the most promising one for the Euler–Lagrange approach since
the underlying physical mechanism is most reasonable. Beside other advantages it allows to
predict the size of the resulting daughter bubbles based on a deterministic scheme. The transfer
from the Euler-Euler to the Euler-Lagrange context requires to define the separation axis, the
separation velocity of the daughter bubbles and finally a breakup time. All these quantities
were derived purely based on physical arguments. The latter characterizes a time lag between
two successive breakup processes of the same bubble avoiding a dependency of the breakup
frequency from the numerical time-step size.
In order to assess the modeling assumptions and the numerical methodology, bubble breakup
within a fully developed turbulent jet flow is investigated. The test case is based on the
experimental study by Martı́nez-Bazán et al. (1999a,b) who discharged a water jet vertically
upwards into a tank and released air bubbles from a tiny needle inside the main self-similar
region. The turbulent jet in the tank is simulated based on LES and bubbles are tracked
through the flow field while analyzing the occurrence of breakup.
REFERENCES 41

The following conclusions can be drawn from the results:

• The predicted jet flow is in good agreement with reference data showing all characteristic
features in the self-similar part of the jet.

• The overall agreement of the predicted bubble size distributions with the experimental
results is good.

• The present breakup model slightly overpredicts breakup in the region shortly down-
stream of the bubble inlet, while it underestimates the breakup further downstream.

• Estimating the time lag ∆tidle based on Eq. (37) yields the best agreement with the
experiment. Generally, prolonging the time lag leads to larger bubbles at positions further
downstream of the injection point.

• The effect of bubble coalescence is relatively small, which agrees with the experiment
(Martı́nez-Bazán et al., 1999a).

• A variation of the surface tension shows that decreasing the surface tension leads to an
augmented breakup behavior into smaller bubbles, while increasing the surface tension
has the opposite effect as expected based on physical arguments.

In summary, the present methodology successfully transfers the breakup criterion by Hage-
saether et al. (2002) to the Euler-Lagrange framework allowing for a fully deterministic de-
scription of the post-breakup properties of the daughter bubbles. To overcome the restrictions
of the present model mentioned above, the inclusion of the entire spectrum of turbulence and
non-binary breakup events are reasonable next steps of the development of bubble breakup
models in the Euler–Lagrange context.

References

Alletto, M., Breuer, M., 2012. One–way, two–way and four–way coupled LES predictions of a
particle–laden turbulent flow at high mass loading downstream of a confined bluff body. Int.
J. Multiphase Flow 45, 70–90.

Alletto, M., Breuer, M., 2013. Prediction of turbulent particle–laden flow in horizontal smooth
and rough pipes inducing secondary flow. Int. J. Multiphase Flow 55, 80–98.
REFERENCES 42

Almohammed, N., Breuer, M., 2016a. Modeling and simulation of agglomeration in turbulent
particle-laden flows: A comparison between energy-based and momentum-based agglomera-
tion models. Powder Technology 294, 373–402.

Almohammed, N., Breuer, M., 2016b. Modeling and simulation of particle–wall adhesion of
aerosol particles in particle-laden turbulent flows. Int. J. Multiphase Flow 85, 142–156.

Almohammed, N., Breuer, M., 2019. Towards a deterministic composite collision outcome
model for surface–tension dominated droplets. Int. J. Multiphase Flow 110, 1–17.

Andersson, R., Andersson, B., 2006. On the breakup of fluid particles in turbulent flows. AIChE
J. 52 (6), 2020–2030.

Armenio, V., Piomelli, U., Fiorotto, V., 1999. Effect of the subgrid scales on particle motion.
Phys. Fluids 11, 3030–3042.

Auton, T. R., 1987. The lift force on a spherical body in a rotational flow. J. Fluid Mech. 183,
199–218.

Auton, T. R., Hunt, J. C. R., Prud’Homme, M., 1988. The force exerted on a body in inviscid
unsteady non-uniform rotational flow. J. Fluid Mech. 197, 241–257.

Batchelor, G. K., 1953. The Theory of Homogeneous Turbulence. Cambridge University Press.

Bhavaraju, S. M., Russell, T. W. F., Blanch, H. W., 1978. The design of gas sparged devices
for viscous liquid systems. AIChE J. 24 (3), 454–466.

Bianco, F., Chibbaro, S., Marchioli, C., Salvetti, M. V., Soldati, A., 2012. Intrinsic filtering
errors of Lagrangian particle tracking in LES flow fields. Phys. Fluids 24, 045103–045103–22.

Bogey, C., Bailly, C., 2006. Large eddy simulations of transitional round jets: Influence of the
Reynolds number on flow development and energy dissipation. Phys. Fluids 18 (6), 065101–
065101–14.

Brennen, C. E., 1982. A review of added mass and fluid inertial forces. Tech. rep., Department
of the Navy, Naval Civil Engineering Laboratory, Port Hueneme, CA, USA.

Breuer, M., 1998. Large–eddy simulation of the sub–critical flow past a circular cylinder:
Numerical and modeling aspects. Int. J. Numer. Meth. Fluids 28 (9), 1281–1302.
REFERENCES 43

Breuer, M., 2000. A challenging test case for large–eddy simulation: High Reynolds number
circular cylinder flow. Int. J. Heat Fluid Flow 21 (5), 648–654.

Breuer, M., 2002. Direkte Numerische Simulation und Large–Eddy Simulation turbulenter
Strömungen auf Hochleistungsrechnern. Habilitationsschrift, Universität Erlangen–Nürn-
berg, Berichte aus der Strömungstechnik. Shaker Verlag, Aachen.

Breuer, M., Alletto, M., 2012. Efficient simulation of particle–laden turbulent flows with high
mass loadings using LES. Int. J. Heat Fluid Flow 35, 2–12.

Breuer, M., Alletto, M., Langfeldt, F., 2012. Sandgrain roughness model for rough walls within
Eulerian–Lagrangian predictions of turbulent flows. Int. J. Multiphase Flow 43, 157–175.

Breuer, M., Almohammed, N., 2015. Modeling and simulation of particle agglomeration in tur-
bulent flows using a hard-sphere model with deterministic collision detection and enhanced
structure models. Int. J. Multiphase Flow 73, 171–206.

Breuer, M., Baytekin, H. T., Matida, E. A., 2006. Prediction of aerosol deposition in 90 degrees
bends using LES and an efficient Lagrangian tracking method. J. Aerosol Sci. 37 (11), 1407–
1428.

Breuer, M., Hoppe, F., 2017. Influence of a cost–efficient Langevin subgrid–scale model on
the dispersed phase of a large–eddy simulation of turbulent bubble–laden and particle–laden
flows. Int. J. Multiphase Flow 89, 23–44.

Breuer, M., Khalifa, A., 2019. Revisiting and improving models for the breakup of compact
dry powder agglomerates in turbulent flows within Eulerian–Lagrangian simulations. Powder
Technology 348, 105–125.

Breuer, M., Matida, E. A., Delgado, A., 2007. Prediction of aerosol drug deposition using an
Eulerian–Lagrangian method based on LES. In: Int. Conf. on Multiphase Flow, July 9–13,
2007. Leipzig, Germany.

Castellano, S., Carrillo, L., Sheibat-Othman, N., Marchisio, D., Buffo, A., Charton, S., 2019.
Using the full turbulence spectrum for describing droplet coalescence and breakage in indus-
trial liquid-liquid systems: Experiments and modeling. Chem. Eng. J. 374, 1420–1432.
REFERENCES 44

Coulaloglou, C. A., Tavlarides, L. L., 1977. Description of interaction processes in agitated


liquid-liquid dispersions. Chem. Eng. Sci. 32 (11), 1289–1297.

Crowe, C. T., Sharma, M. P., Stock, D. E., 1977. The Particle-Source-In-Cell (PSI-CELL)
model for gas–droplet flows. Trans. ASME J. Fluids Eng. 99, 325–332.

Crowe, C. T., Sommerfeld, M., Tsuji, Y., 1998. Multiphase Flows with Droplets and Particles.
CRC Press.

Devroye, L., 1986. Non-Uniform Random Variate Generation. Springer-Verlag New York.

Dijkhuizen, W., Roghair, I., Van Sint Annaland, M., Kuipers, J. A. M., 2010. DNS of gas
bubbles behaviour using an improved 3D front tracking model – Drag force on isolated
bubbles and comparison with experiments. Chem. Eng. Sci. 65 (4), 1415–1426.

Gatignol, R., 1983. The Faxén formulae for a rigid particle in an unsteady non-uniform Stokes
flow. J. Mec. Theor. Appl. 1, 143–160.

Germano, M., Piomelli, U., Moin, P., Cabot, W. H., 1991. A dynamic subgrid-scale eddy
viscosity model. Phys. Fluids A 3 (7), 1760–1765.

Gohil, T. B., Saha, A. K., Muralidhar, K., 2011. Direct numerical simulation of naturally
evolving free circular jet. J. Fluids Eng. 133 (11), 111203–111203–11.

Gohil, T. B., Saha, A. K., Muralidhar, K., 2014. Large eddy simulation of a free circular jet.
J. Fluids Eng. 136 (5), 051205–051205–14.

Hagesaether, L., Jakobsen, H. A., Svendsen, H. F., 2002. A model for turbulent binary breakup
of dispersed fluid particles. Chem. Eng. Sci. 57, 3251–3267.

Hesketh, R. P., Etchells, A. W., Russell, T. W. F., 1991a. Bubble breakage in pipeline flow.
Chem. Eng. Sci. 46 (1), 1–9.

Hesketh, R. P., Etchells, A. W., Russell, T. W. F., 1991b. Experimental observations of bubble
breakage in turbulent flow. Indust. Eng. Chem. Res. 30 (5), 835–841.

Hinze, J. O., 1955. Fundamentals of the hydrodynamic mechanism of splitting in dispersion


processes. AIChE J. 1 (3), 289–295.
REFERENCES 45

Hinze, J. O., 1975. Turbulence, 2nd Edition. McGraw-Hill.

Hoppe, F., Breuer, M., 2018. A deterministic and viable coalescence model for Euler–Lagrange
simulations of turbulent microbubble-laden flows. Int. J. Multiphase Flow 99, 213–230.

Hussein, H. J., Capp, S. P., George, W. K., 1994. Velocity measurements in a high-Reynolds-
number, momentum-conserving, axisymmetric, turbulent jet. J. Fluid Mech. 258, 31–75.

Innocenti, A., Marchioli, C., Chibbaro, S., 2016. Lagrangian filtered density function for LES-
based stochastic modelling of turbulent particle-laden flows. Phys. Fluids 28 (11), 115106.

Israelachvili, J. N., 2011. Intermolecular and Surface Forces, Third Edition. Academic Press,
San Diego.

Jain, D., Kuipers, J. A. M., Deen, N. G., 2014. Numerical study of coalescence and breakup in
a bubble column using a hybrid volume of fluid and discrete bubble model approach. Chem.
Eng. Sci. 119, 134–146.

Kamp, A. M., Chesters, A. K., Colin, C., Fabre, J., 2001. Bubble coalescence in turbulent flows:
A mechanistic model for turbulence-induced coalescence applied to microgravity bubbly pipe
flow. Int. J. Multiphase Flow 27 (8), 1363–1396.

Karimi, M., Andersson, R., 2019. Dual mechanism model for fluid particle breakup in the
entire turbulent spectrum. AIChE J. 65 (8), 1–15.

Kuerten, J. G. M., Vreman, A. W., 2005. Can turbophoresis be predicted by large–eddy


simulation? Phys. Fluids 17, 011701–011701–4.

Lamb, H., 1932. Hydrodynamics, 6th Edition. Cambridge University Press.

Lau, Y. M., Bai, W., Deen, N. G., Kuipers, J. A. M., 2014. Numerical study of bubble break-up
in bubbly flows using a deterministic Euler–Lagrange framework. Chem. Eng. Sci. 108, 9–22.

Legendre, D., Magnaudet, J., 1998. The lift force on a spherical bubble in a viscous linear
shear flow. J. Fluid Mech. 368, 81–126.

Lehr, F., Millies, M., Mewes, D., 2002. Bubble-size distributions and flow fields in bubble
columns. AIChE J. 48 (11), 2426–2443.
REFERENCES 46

Liao, Y., Lucas, D., 2009. A literature review of theoretical models for drop and bubble breakup
in turbulent dispersions. Chem. Eng. Sci. 64 (15), 3389–3406.

Liao, Y., Lucas, D., 2010. A literature review on mechanisms and models for the coalescence
process of fluid particles. Chem. Eng. Sci. 65, 2851–2864.

Lilly, D. K., 1992. A proposed modification of the Germano subgrid-scale closure method.
Phys. Fluids A 4 (3), 633–635.

Luo, H., Svendsen, H. F., 1996. Theoretical model for drop and bubble breakup in turbulent
dispersions. AIChE J. 42 (5), 1225–1233.

Marchioli, C., Salvetti, M. V., Soldati, A., 2008. Some issues concerning large–eddy simulation
of inertial particle dispersion in turbulent bounded flows. Phys. Fluids 20, 040603–040603–11.

Martı́nez-Bazán, C., 2018. Private Communication.

Martı́nez-Bazán, C., Montañes, J. L., Lasheras, J. C., 1999a. On the breakup of an air bubble
injected into a fully developed turbulent flow. Part 1. Breakup frequency. J. Fluid Mech.
401, 157–182.

Martı́nez-Bazán, C., Montañes, J. L., Lasheras, J. C., 1999b. On the breakup of an air bubble
injected into a fully developed turbulent flow. Part 2. Size PDF of the resulting daughter
bubbles. J. Fluid Mech. 401, 183–207.

Martı́nez-Bazán, C., Rodrı́guez-Rodrı́guez, J., Deane, G. B., Montañes, J. L., Lasheras, J. C.,
2010. Considerations on bubble fragmentation models. J. Fluid Mech. 661, 159–177.

Mattson, M. D., Mahesh, K., 2012. A one-way coupled, Euler–Lagrangian simulation of bubble
coalescence in a turbulent pipe flow. Int. J. Multiphase Flow 40, 68–82.

Maxey, M. R., Riley, J. J., 1983. Equation of motion for a small rigid sphere in a non–uniform
flow. Phys. Fluids 26, 883–889.

Mei, R., Klausner, J. F., 1992. Unsteady force on a spherical bubble at finite Reynolds number
with small fluctuations in the free-stream velocity. Phys. Fluids A-Fluid 4, 63–70.

Mei, R., Klausner, J. F., Lawrence, C. J., 1994. A note on the history force on a spherical
bubble at finite Reynolds number. Phys. Fluids 6, 418–420.
REFERENCES 47

Melville, W. K., 1996. The role of surface-wave breaking in air-sea interaction. Annu. Rev.
Fluid Mech. 28 (1), 279–321.

Michalke, A., 1984. Survey on jet instability theory. Prog. Aerosp. Sci. 21, 159–199.

Morris, P. J., 1976. The spatial viscous instability of axisymmetric jets. J. Fluid Mech. 77 (3),
511–529.

Müller-Fischer, N., Tobler, P., Dressler, M., Fischer, P., Windhab, E. J., 2008. Single bubble
deformation and breakup in simple shear flow. Exp. Fluids 45 (5), 917–926.

Pallas, N. R., Harrison, Y., 1990. An automated drop shape apparatus and the surface tension
of pure water. Colloid Surface 43, 169–194.

Panchapakesan, N. R., Lumley, J. L., 1993. Turbulence measurements in axisymmetric jets of


air and helium. Part 1. Air jet. J. Fluid Mech. 246, 197–223.

Pope, S. B., 2000. Turbulent Flows. Cambridge University Press.

Pozorski, J., Apte, S. V., 2009. Filtered particle tracking in isotropic turbulence and stochastic
modeling of subgrid–scale dispersion. Int. J. Multiphase Flow 35, 118–128.

Press, W. H., Teukolsky, S. A., Vetterling, W. T., Flannery, B. P., 2007. Numerical Recipes:
The Art of Scientific Computing, 3rd Edition. Cambridge University Press, New York, NY,
USA.

Prince, M. J., Blanch, H. W., 1990. Bubble coalescence and break-up in air-sparged bubble
columns. AIChE J. 36 (10), 1485–1499.

Qian, D., McLaughlin, J. B., Sankaranarayanan, K., Sundaresan, S., Kontomaris, K., 2006.
Simulation of bubble breakup dynamics in homogeneous turbulence. Chem. Eng. Commun.
193 (8), 1038–1063.

Rhie, C. M., Chow, W. L., 1983. A numerical study of the turbulent flow past an isolated
airfoil with trailing-edge separation. AIAA J. 21, 1525–1532.

Risso, F., Fabre, J., 1998. Oscillations and breakup of a bubble immersed in a turbulent field.
J. Fluid Mech. 372, 323–355.
REFERENCES 48

Sadeghi, H., Lavoie, P., Pollard, A., 2018. Effects of finite hot-wire spatial resolution on tur-
bulence statistics and velocity spectra in a round turbulent free jet. Exp. Fluids 59 (40),
1–10.

Saha, D., 2013. Experimental analysis of aggregate breakup in flows observed by three dimen-
sional particle tracking velocimetry. Ph.D. thesis, ETH Zürich.

Schiller, L., Naumann, A., 1933. A drag coefficient correlation. VDI Zeitschrift 77, 318–320.

Sharp, D. H., 1984. An overview of Rayleigh–Taylor instability. Physica D 12 (1–3), 3–10.

Smagorinsky, J., 1963. General circulation experiments with the primitive equations, I, The
basic experiment. Month. Weath. Rev. 91, 99–165.

Solsvik, J., Jakobsen, H. A., 2015. Single air bubble breakup experiments in stirred water tank.
Int. J. Chem. React. Eng. 13 (4), 477–491.

Solsvik, J., Jakobsen, H. A., 2016a. Development of fluid particle breakup and coalescence
closure models for the complete energy spectrum of isotropic turbulence. Ind. & Eng. Chem.
Res. 55 (5), 1449–1460.

Solsvik, J., Jakobsen, H. A., 2016b. A review of the statistical turbulence theory required
extending the population balance closure models to the entire spectrum of turbulence. AIChE
J. 62 (5), 1795–1820.

Solsvik, J., Maa, S., Jakobsen, H. A., 2016. Definition of the single drop breakup event. Ind.
& Eng. Chem. Res. 55 (10), 2872–2882.

Solsvik, J., Skjervold, V. T., Jakobsen, H. A., 2017. A bubble breakage model for finite Reynolds
number flows. J. Disper. Sci. Tech. 38 (7), 973–978.

Solsvik, J., Tangen, S., Jakobsen, H. A., 2013. On the constitutive equations for fluid particle
breakage. Rev. Chem. Eng. 29 (5), 241–356.

Sreenivasan, K. R., 1995. On the universality of the Kolmogorov constant. Phys. Fluids 7 (11),
2778–2784.

Stewart, C. W., 1995. Bubble interaction in low-viscosity liquids. Int. J. Multiphase Flow
21 (6), 1037–1046.
REFERENCES 49

Sujatha, K. T., Jain, D., Kamath, S., Kuipers, J. A. M., Deen, N. G., 2017. Experimental and
numerical investigation of a micro-structured bubble column with chemisorption. Chem.
Eng. Sci. 169, 225–234.

Sungkorn, R., Derksen, J. J., Khinast, J. G., 2012. Euler–Lagrange modeling of a gas–liquid
stirred reactor with consideration of bubble breakage and coalescence. AIChE J. 58 (5),
1356–1370.

Tennekes, H., Lumley, J. L., 1972. A First Course in Turbulence. The MIT Press.

Tomiyama, A., Zun, I., Sakaguchi, T., 1998. Drag coefficients of single bubbles under normal
and micro gravity conditions. Jap. Soc. of Mech. Eng., Int. J. Series B - Fluid and Thermal
Eng. 41, 472–479.

Tripathi, M. K., Sahu, K. C., Govindarajan, R., 2015. Dynamics of an initially spherical bubble
rising in quiescent liquid. Nat. Commun. 6, 1–9.

Tsouris, C., Tavlarides, L. L., 1994. Breakage and coalescence models for drops in turbulent
dispersions. AIChE J. 40 (3), 395–406.

Walter, J. F., Blanch, H. W., 1986. Bubble break-up in gas-liquid bioreactors: Break-up in
turbulent flows. The Chem. Eng. J. 32 (1), B7–B17.

Wang, T., Wang, J., Jin, Y., 2003. A novel theoretical breakup kernel function for bub-
bles/droplets in a turbulent flow. Chem. Eng. Sci. 58 (20), 4629–4637.

Wei, Y. K., Qian, Y., Xu, H., 2012. Lattice Boltzmann simulations of single bubble deformation
and breakup in a shear flow. J. Comp. Multiphase Flows 4 (1), 111–117.

Wilkinson, P. M., Van Schayk, A., Spronken, J. P. M., Van Dierendonck, L., 1993. The influence
of gas density and liquid properties on bubble breakup. Chem. Eng. Sci. 48 (7), 1213–1226.

Wygnanski, I., Fiedler, H., 1969. Some measurements in the self-preserving jet. J. Fluid Mech.
38 (3), 577–612.

Zaman, K. B. M. Q., Hussain, A. K. M. F., 1980. Vortex pairing in a circular jet under
controlled excitation. Part 1. General jet response. J. Fluid Mech. 101 (3), 449–491.

You might also like