Handbook of Multiphase Flow Science and Technology (Guan Heng Yeoh (Eds.) )

Basic Theory and Conceptual Framework
of Multiphase Flows
Guan Heng Yeoh and Jiyuan Tu
Abstract
The fundamentals of computational multiphase fluid dynamics are presented
using discrete and continuum frameworks. Depending on the number, type, and
size of phases and their interaction between phases within the flow system, a
multiscale consideration of the multiphase flow physics allows the adoption of a
number of possible approaches. The Lagrangian formulation can be utilized to
track the motion of discrete constituents of identifiable portion of particular
phases occupying the flow system. This represents the most comprehensive
investigation that can be performed to analyze the multiphase flow physics.
Because of the complexity of the microscopic motions and thermal characteristics
of each discrete constituent which can be prohibitive at the (macro) device scale,
the Eulerian formulation which characterizes the flow of discrete constituents as a
fluid can be adopted for practical analysis of the flow system. This results in the
development of a multifluid approach which solves for the conservation equa-
tions of continuous and dispersed phases. In order to better resolve the micro-
physics of the discrete constituents at the mesoscale, population balance allows
the synthesization of the behavior and dynamic evolution of the population of the
discrete constituents occupying the flow system. Such an approach allows the
consideration of the spatial and temporal evolution of the geometrical structures
as a result of formation and destruction of agglomerates or clusters through
G.H. Yeoh (*)

School of Mechanical and Manufacturing Engineering, University of New South Wales, Sydney,
NSW, Australia
Australian Nuclear Science and Technology Organisation (ANSTO), Kirrawee DC, NSW, Australia
e-mail: g.yeoh@unsw.edu.au
J. Tu
School of Aerospace, Mechanical and Manufacturing Engineering, Platform Technologies
Research Institute (PTRI), RMIT University, Melbourne, VIC, Australia
e-mail: jiyuan.tu@rmit.edu.au
# Springer Nature Singapore Pte Ltd. 2016 1

G.H. Yeoh (ed.), Handbook of Multiphase Flow Science and Technology,
DOI 10.1007/978-981-4585-86-6_1-1
2 G.H. Yeoh and J. Tu
interactions between the discrete constituents and, more importantly, the colli-
sions with turbulent eddies.
Keywords
Multiphase flows • Lagrangian framework • Eulerian framework • Population
balance
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Gas-Solid Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Liquid-Solid Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Gas-Liquid Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Three-Phase Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Definitions of Basic Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Multi-Scale Consideration of Multiphase Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Lagrangian Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Models for Particulate-Particulate Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Classification of Particulate-Particulate Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Classification of Fluid-Particulate Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Eulerian Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Interpenetrating Media Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Derivation of Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Population Balance Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Introduction
Multiphase flows in the context of fluid mechanics can be perceived as a flow system
that consists of two or more distinct phases flowing in a fluid mixture where the level
of separation between the phases is at a scale well above the molecular level.
Principally, multiphase flows can be categorized by the number, type, and size of
phases and their respective interaction between each phase in the fluid mixture. A
multiphase flow system can thus be classified according to different types of flows
depending on the combination of different phases within the fluid mixture. The
physical understanding of multiphase flows especially when dealing with more than
one phase in the fluid mixture offers problems of complexity that are immeasurably
far greater than in single-phase flows. These phases generally do not uniformly mix,
and the prevalence of small-scale interactions occurring between the phases can have
profound effects on the macroscopic properties of the flow system. This clearly
reflects the ubiquitous challenges that persist when modeling the inherently complex
nature of the multiphase flow physics.
Basic Theory and Conceptual Framework of Multiphase Flows 3
Gas-Solid Flows
Gas-solid or gas-particle flows concern the motion of suspended solid particles in the
gas phase. When the particle number density is relatively small, the gas flow
influences the flow of solid particles. Such flows are governed predominantly by
the surface and body forces acting on the solid particles. These types of flows are
generally known as dilute gas-particle flows. When the particle number density is
very small, these solid particles are mere tracers in the gas phase. These types of
flows are commonly known as very dilute gas-particle flows. When the particle
number density is sufficiently large, particle-particle interactions now govern the
motion of solid particles. Collisions that exist between solid particles will signifi-
cantly alter the migration of these particles within the gas phase. These types of flow
are referred to as dense gas-particle or granular flows. For gas-particle flows in
conduits, the motion of solid particles following impact on the boundary walls is
affected by the surface characteristics and material properties, which is different
when compared to the free flight of solid particles in the gas phase. In gas-particle
flows, the solid particles constitute the dispersed or discrete phase and the gas is the
continuous phase.
Liquid-Solid Flows
Liquid-solid flows comprise the transport of solid particles in the liquid phase. Such
flows are driven largely by the presence of pressure gradients since the density ratio
between the two phases is generally low and the drag between the phases constitutes
the dominant effect in such flows. These types of flows are generally known as
liquid-particle flows or slurry flows. In liquid-particle flows, the solid particles
constitute the dispersed phase and the liquid represents the continuous phase. The
major concern in such flows is the characterization of the sedimentation behavior in
the liquid-particle mixture which is governed by the range of size of solid particles
traveling in the liquid phase.
Gas-Liquid Flows
Gas-liquid flows can exist in a number of different forms – the motion of gas bubbles
in the liquid phase or the motion of liquid droplets in the gas phase. For the former,
the liquid is taken as the continuous phase and the gas bubbles are considered as
discrete constituents of the gas phase or the dispersed phase. For the latter, the gas is
regarded as the continuous phase and the droplets are taken as the finite fluid
particles of the liquid phase or the dispersed phase. Since gas bubbles or liquid
droplets are allowed to deform within the continuous phase, several different
geometrical shapes are possible which include spherical, elliptical, distorted, toroi-
dal, cap, and other complex configurations. Gas-liquid flows undergo a spectrum of
flow transition regimes. Change of interfacial structures between the phases are due
to bubble-bubble interactions via coalescence and breakup of bubbles and any phase
change process from gas-to-liquid or liquid-to-gas. For the special case of separated
flows such as liquid film in gas phase or gas film in liquid phase and liquid core and
gas film or gas core and liquid film in conduits, such flows possess well-defined
interfaces, and they belong to the specific consideration of immiscible flows. Each
phase is treated as a continuous fluid co-flowing simultaneously with each other.
Three-Phase Flows
Three-phase gas-liquid-solid flows are encountered in a number of engineering

applications of technical relevance. Principally, this particular class of multiphase
flows considers the solid particles and gas bubbles as being the discrete constituents
of the dispersed phase co-flowing with the continuous liquid phase. The coexistence
of the three phases considerably complicates the computational modeling due to the
requirement of understanding the phenomena associated with particle-particle, bub-
ble-bubble, particle-bubble, particle-fluid, and bubble-fluid interactions modifying
the multiphase flow physics.
Definitions of Basic Terms
Some basic definitions that are fundamental to the description of multiphase flows
are introduced herein. For convenience, the notation that will be adopted corre-
sponds to the Cartesian tensor format. The lowercase subscripts (ijk) are employed in
the conventional manner to denote vector or tensor components. The single upper-
case subscript (n) refers to the property of a specific phase. In general, the generic
subscripts n = c (continuous liquid), n = d (dispersed or discrete phase), n = l
(liquid), n = g (gas), and n = s (solid) are employed for clarity in depicting the
different classes of multiphase flows.
Specific properties frequently encountered are as follows. The volume fraction of
the continuous phase can be defined as
δV c
αc ¼ lim (1)
δV!δV o
δV
where δVc is the volume of the continuous phase in the volume of δV. The volume
δVo represents the limiting volume whereby a stationary averaging is performed.
Equivalently, the volume fraction of the dispersed or discrete phase can be written as
δV d
αd ¼ lim (2)
δV!δV o
δV
where δVd is the volume of the dispersed or discrete phase in the volume of δV. This
volume fraction is also referred to as the void fraction which describes the portion of
the channel or pipe occupied by the dispersed gas phase at any instant in space and
time. In the chemical engineering terminology, it is known as holdup. For the case of
two-phase flow, it follows that αd = 1 αc. Hence, the sum of the volume fractions
P
of different constituents in a multiphase mixture must be equal to unity, i.e., n αn
¼ 1.
The mixture density can be evaluated in accordance with
X
ρ¼ α ρ
n n n
(3)
The bulk density of the dispersed phase is related to the material density by
ρd ¼ αd ρd (4)
of which the material density, in terms of a limit, is defined as
δMd
ρd ¼ limδV!δV o (5)
δV
with δMd being the mass of the dispersed phase. The bulk and material densities of
the continuous phase are analogously defined in accordance with the definition of the
bulk and material densities of the dispersed phase. Conversely, the specific enthalpy,
h, and specific entropy, s, in terms of per unit mass are weighted similar to the
mixture density to the following:
X
ρh ¼ α ρh
n n n n
(6)
X
ρs ¼ α ρs
n n n n
(7)
It should be noted that properties such as mixture viscosity or thermal conduc-

tivity may not be obtained through such simple weighted averaging. Other means of
evaluating these properties are required.
The true velocities (or actual local velocities) of the different phases are the
velocities the phases actually travel, which is the local instantaneous velocities of
the fluids. Defining uc and ud to be the local instantaneous or phase velocities of
continuous and dispersed phases, the superficial velocities and the phase velocities
are related by the volume fraction according to
U c ¼ αc uc (8)
Ud ¼ αd ud (9)
For the case of two-phase flow, the total superficial velocity is U = Uc + Ud.
Hence, the totalPsuperficial velocity for a multiphase mixture can be analogously
written as U ¼ n αn un .
Of more specific definitions, the quality of a mixture comprising liquid and vapor
is defined by
ρd
X¼ (10)
ρ
while the dispersed phase mass concentration, which is the ratio of the mass of the
dispersed phase to that of the continuous phase in a mixture, is given as
ρd
C¼ (11)
ρc
Multi-Scale Consideration of Multiphase Flows
Within the conceptual framework of computational multiphase fluid dynamics, the

Eulerian or Lagrangian formulation of multiphase flows requires the multiscale
consideration of the multiphase flow physics. The different physical characteristics
at different length scales are illustrated in Fig. 1. For the microscale physics, it is
paramount that interaction of the gas bubbles, liquid drops, and solid particles with
the continuum phase is properly understood through tracking the motion of the
individual discrete constituents in space and time. For the mesoscale physics,
significant interaction between the discrete constituents may result in local structural
changes due to agglomeration/coalescence and breakage/breakup processes of gas
bubbles, liquid drops, and solid particles. For the macroscale physics, the hydrody-
namic behavior of the background fluid on the clusters of gas bubbles, liquid drops,
and solid particles may yield large scale flow structures influencing the different
individual phases co-flowing with the continuous phase within the flow system.
Computational approaches can be utilized to reveal details of particular multi-
phase flow physics that otherwise could not be visualized through experiments or
clarify specific accentuating mechanisms that are consistently being manifested.
Techniques that are being adopted based on the utilization of advanced numerical
methods and models usually contain very detailed information, producing an accu-
rate realization of the fluid flow.
Lagrangian Formulation
The concept of Lagrangian tracking entails following the motion of individual

identities of the identifiable portion of a particular phase occupying the flow system.
Such consideration therefore includes molecular dynamics, Brownian dynamics, and
discrete element method, which is represented by an illustration of a plot depicting
the characteristic time scale versus length scale in Fig. 2. In general, these identities
being considered represent a wide range of discrete elements including atoms,
molecules, nuclei, cells, aerosol or colloidal particles, and granules.
Micro-Scale Meso-Scale Macro-Scale
Interactions of Formation of Hydrodynamic

Discrete Elements Clusters of Bubbles, Behaviour of Phases
with Continuum Fluid Drops and Particles at Device Scale
Motion of Local Structural Large Scale

Discrete Changes Flow Structures
Elements
Length Scale
Fig. 1 Multiscale consideration of multiphase flows (Yeoh et al. 2014)
Fig. 2 Illustration of time and length scales for Lagrangian simulations: molecular dynamics,
Brownian dynamics, and discrete element method (Yeoh et al. 2014)
Molecular dynamics, Brownian dynamics, and discrete element method share the
common characteristics whereby the discrete elements are allowed to interact for a
period of time under prescribed interaction laws, and the motion of each individual
element is resolved by solving the linear momentum and angular momentum
equations, subject to forces and torques arising both from particle interaction with
each other and those imposed on the particles by the surrounding fluid (Li et al.
2011). Consideration of which Lagrangian tracking approach should be adopted
stems primarily through the formulation of appropriate particle interaction laws as

well as by the imposition of random forcing to mimic collisions or interaction with
molecules of the surrounding fluid. By carefully identifying these particle interaction
laws, the physical behavior of the discrete element at a certain range of time and
length scales can be efficiently captured.
Considering the discrete elements which could be solid particles, liquid droplets,
or gas bubbles, the instantaneous velocity Ud and rotation rate Ωd for the discrete
particulate (particle, droplet, or bubble) through the time-driven discrete element
method can be obtained through solution of the linear and angular momentum
equations (Newton’s second law):
DUd X
md ¼ Fd (12)
Dt
DΩd X
Id ¼ Md (13)
Dt
where md is the mass of particulate, and Id is the moment of inertia of particulate. On
the right hand side, the time derivatives are essentially the material derivatives of the
particulate velocity and particulate rotation rate. On the left hand side, the source
terms represent the sum or cumulative of forces and moments acting on the
particulate.
For multiphase flow associated with heat exchange and phase change between
particulate and surrounding fluid, these heat and mass transfer processes are concur-
rently tracked along the discrete particulate trajectories and solved by the particulate
conservation equations of mass and energy:
Dmd
¼ Smd (14)
Dt
DT d
¼ ST d (15)
Dt
where Td is the temperature of the particulate. The source term Smd represents the
interphase mass transfer between the particulate and surrounding fluid, while the
source term ST d is governed primarily by the interphase convection heat transfer,
latent heat transfer associated with mass transfer, net radiative power absorbed by the
particulate, and particulate-particulate interaction due to conduction heat transfer.
Note that the product of the particulate mass, specific heat of constant pressure, and
material derivative of the particulate temperature denote the sensible heating term of
the particulate energy equation.
Models for Particulate-Particulate Interaction
Through specific consideration of using discrete element method for large particulate
assemblies, the hard-sphere or soft-sphere model can be effectively applied. Soft
contact forces are subsequently derived from a point on the bodies for the hard-
sphere model or the overlap of bodies for soft-sphere model.
Hard-Sphere Model
Main assumptions that are concerned with the particulate shape, deformation history
during collision, and nature of collisions are:
• Particulates are generally taken to be spherical and quasirigid

• Shape of particulates is retained after impact
• Dynamics of idealized binary collision
• Collisions between particulates are instantaneous
• Contact of particulates during collision occurs at a point
• Interaction forces are taken to be impulsive and all other finite forces are negli-
gible during collisions
These assumptions are believed to be sufficiently realistic for collisions of

relatively coarse particulates (>100 μm). One characteristic feature of this model
is the ability to process a sequence of collisions one at a time. Also, Lagrangian
tracking of particulates can be readily performed with realistic values of the restitu-
tion and friction coefficients. During the impact of two particles such as described in
Fig. 3, the motion is governed by the linear and angular impulse momentum laws for
a binary collision of two spheres:

md,k Ud,k U0d,k ¼ J (16)

md,l Ud,l U0d,l ¼ J (17)
I d,k
Ωd,k Ω0d,k ¼ J n (18)
rk
I d,l
Ωd,l Ω0d,l ¼ J n (19)
rl
where superscript 0 denotes conditions just before collision, md,k and md,l are the
mass, rk and rl are the radii, Ωd,k and Ωd,l are the angular velocities, Id,k and Id,l are
the moment of inertia of particulate k and particulate l. Velocities prior to collisions
are the velocities determined at the last time step just before collision. Note that the
corresponding time difference should not be larger than 104 s.
In Fig. 3, the normal and tangential unit vectors that define the collision coordi-
nate system are:
y rl
rk Ud,l
n
Ud,k Wd,l
l
k
Gd,kl Rl
Rk
Wd,k l
x
k t J
z
Fig. 3 Contact between two particles for hard-sphere model
xk xl
n¼ (20)
jxk xl j

Ud,kl G0d,kl ∙n n
t ¼
(21)
Ud,kl G0d,kl ∙n n
By definition, I ¼ mr 2gyration where rgyration is the radius of gyration of the

particulate and J is the impulse vector. Adopting (J n) n = J (J ∙ n)n, the
relative velocity at the contact point between two particulates with velocities Ud,k
and Ud,l is derived as

Ud,kl U0d,kl ¼ Bd,k J Bd,k Bd,l ðJ∙nÞn (22)
where

Ud,kl ¼ Gd,kl r k Ωd,k r l Ωd,l n (23)
The relative velocity of particulate centroids Gd , kl is given by
Gd,kl ¼ Ud,k Ud,l (24)
Some parameters are established to relate the velocities before and after colli-
sions. The first collision parameter is the coefficient of normal restitution, en:

Ud,kl ∙n ¼ en U0d,kl ∙n (25)
The normal component of the impulse vector can thus be written as:

U0d,kl ∙n
J n ¼ ð1þen Þ (26)
B1
where B1 = (md,k + md,l)/md,kmd,l. The second and third collision parameters consist
of the coefficient of tangential restitution, et, and the coefficient of friction, μf. These
two parameters concern the two kinds of collisions – particulate sticking and sliding
in the tangential impact process. For the case where the tangential component of the
impact velocity is sufficiently high or the friction coefficient is small by comparison,

ð1þet Þ Ud,kl ∙t
0
μf < (27)
Jn B2
where
1 1 r2 r2
B2 ¼ þ þ k þ l
md , k md , l I d , k I d , l
Here, sliding occurs throughout the whole duration of the contact. Applying
Coulomb’s law, the tangential component of the impulse is then given by
J t, sliding ¼ μf J n (28)
On the other hand, if the friction coefficient is sufficiently high,

ð1þet Þ Ud, kl ∙t
0
μf (29)
Jn B2
in which sticking collisions occur after an initial sliding phase – the relative
tangential velocity between two colliding particles becomes zero – the tangential
impulse for this case is:

U0d,kl ∙t
J t,sliding ¼ ð1þet Þ (30)
B2
where the coefficient of tangential restitution, et, is defined as:

Ud,kl ∙t ¼ et U0d,kl ∙t (31)
Once all the impulse vectors are known, the postcollision velocities can be
appropriately determined. More details on a two-step approach in determining the
hard-sphere particle dynamics can be found in Hoomans et al. (1996).
y x
rj Ud,j
n
ri Ud,i Wd,j
j
i Rj
Ri
Wd,i j
x
i t
z
Fig. 4 Contact between two particles for soft-sphere model
Soft-Sphere Model
This model requires that particulate collisions are of finite durations. The duration of
contact is related to the nonfinite particulate stiffness which is specified as the
particulate property. Thus, the force at contact is continuously changing as the
particulates begin to deform, which can be represented by the assumption of a
small overlap distance. Forces at all contacts are determined at one instant and
Newton’s equations of motion are then solved to obtain new particulate locations
and velocities. For dense flows, this model is considered to be much more efficient
than the hard-sphere model. More importantly, it can be applied to any configura-
tions, including static and dynamic situations.
As illustrated in Fig. 4, at any instant during the collision of two particulates, the
resultant force acting between the two particulates, assuming spheres for the purpose
of illustration, can be expressed by their respective normal and tangential compo-
nents. Note that the formation of normal and tangential stresses during impact is
described as a decoupled problem. The normal and tangential unit vectors can be
expressed as
xi xj
n ¼ (32)
xi xj
Utd
t ¼ (33)
Utd
From above, Utd is the slip velocity of the point of contact or tangential velocity
which can be obtained from

Utd ¼ Urd Urd ∙n n (34)
with the relative velocity of each particulate given by


Urd ¼ Ud,i Ud,j r i Ωd,i r j Ωd,j n (35)
The particulate deformation during contact can be characterized by the presence

of a normal overlap or displacement ξ of the two particulates, viz.,

ξ ¼ Ri þ Rj þ xi xj ∙n (36)
while the normal displacement rate ξ_ can be evaluated based upon the relative
translational velocities projected in the direction of the normal unit vector:

ξ_ ¼ Ud, i Ud, j ∙n (37)
The total collision force and torque fields on particle i can be written as
FCd ¼ Fnd þ Ftd (38)
d ¼ ri n F
MC t
(39)
where Fnd and Ftd are the normal and tangential contact forces.
Classification of Particulate-Particulate Forces
For the soft sphere model, the fundamentals of particulate-particulate interactions

can be represented using a combination of particulate and continuum mechanics.
Figure 5 illustrates the different levels of classification for the dynamics of random
packing in particulate beds, clusters, or agglomerates: continuum mechanics, micro-
mechanics, and molecular dynamics. Using this model with stiff particulates and soft
contacts, the influence of elastic-plastic repulsion in a representative particulate
contact can be demonstrated via contact force equilibrium. Also, a sphere-sphere
adhesion model without any contact deformation can be combined with a plate-plate
adhesion model for nanocontact flattening. Through this model, various contact
deformation paths can be realized. With the increasing flattening, normal and
tangential contact stiffness, rolling and twisting resistance, energy absorption, and
friction work increase.
The level of continuum mechanics constitutes the formulation of three-
dimensional continuum models as described by tensor equations of representative
elemental finite volumes. Balances of mass, momentum, moment, and energy are
considered for three translational and three rotational degrees of freedom.
The level of micromechanics constitutes the microtransition to a geometrical
equivalent of two-dimensional plane with a finite number of discrete subelements,
i.e., particulates. At this level, all particulate-particulate interactions within the
random packing structure are described by contact forces in the normal and tangen-
tial directions in conjunction with fundamental laws for elastic force-displacement,
inelastic deformations or plastic dislocations, solid friction, and viscous damping.
Fig. 5 Classification of different levels of particulate dynamics: continuum mechanics, micro-

mechanics, and molecular dynamics
These microscopic simulations of a small sample can be utilized to derive macro-

scopic constitutive relationships as required to describe the behavior within the
framework of the macroscopic continuum theory.
The level of molecular dynamics constitutes the dominant molecular interactions
at particulate contact surfaces. In general, fundamental material properties such as
elasticity; bonding strength; viscoelasticity; electromagnetic, thermal, and wave
propagation characteristics; or phase conversion enthalpies can be physically
explained by the molecular interaction energies and potential forces.
Contact Forces at the Level of Micromechanics

Four deformation effects can be considered in accordance with particulate-surface
contacts and their force-response (stress-strain) behaviors:
• Elastic contact deformation which is reversible, independent of deformation rate

and consolidation time effects and such deformation is valid for all particulate
solids
Table 1 Literatures on elastic, plastic, viscoelastic, and viscoplastic contact deformation

Viscoelastic
Elastic contact Plastic contact contact Viscoplastic contact
Hertz (1882) Derjaguin (1934) Pao (1955) Rumpf et al. (1976)
Huber (1904) Krupp and Sperling Lee and Radok Kuhn and
(1965) (1960) McMeeking (1992)
Derjaguin (1934) Greenwood and Hunter (1960) Bouvard and
Williamson (1966) McMeeking (1996)
Bradley (1936) Schubert et al. Yang (1966) Stroakers et al.
(1976) (1997)
Fromm (1927) Molerus (1975, Krupp (1967) Stroakers et al.
1978) (1999)
Cattaneo (1938) Maguis and Pollock Rumpf et al. Parhami and
(1984) (1976) McMeeking (1998)
Foppl (1947) Walton and Braun Walton (1993) Parhami et al.
(1986) (1999)
Mindlin (1949) Fleck et al. (1992) Sadd et al. (1993) Redanz and Fleck
(2001)
Deresiewicz (1954) Greenwood (1997) Leszczynski Heylinger and
(2003) McMeeking (2001)
Lurje (1963) Thornton (1997)
Krupp (1967) Thornton and Ning Brilliantov and Tomas (2004a, b)
(1998) Poschel (2005)
Greenwood and Tomas (2000, 2001) Luding et al. (2005)
Williamson (1966)
Johnson et al. (1971) Vu-Quoc and Zhang
(1999)
Dahneke (1972) Vu-Quoc et al.
(2000)
Maw et al. (1976) Mesarovic and
Johnson (2000)
Cundall and Strack Luding and
(1979) Herrmann (2001)
Tsai et al (1991) Luding et al. (2001)
Thornton and Yin (1991), Delenne et al. (2004)
Thornton (1991)
Sadd et al. (1993)
Tavares and King (2002)
Vu-Quoc and Zhang
(1999)
Di Renzo and Di Maio
(2004)
• Plastic contact deformation which is irreversible, deformation rate and consoli-

dation time being independent
• Viscoelastic contact deformation which is reversible and dependent on deforma-
tion rate and consolidation time
Normal Impact Shearing Twisting Rolling
Fig. 6 Different modes of particulate-particulate interactions
• Viscoplastic contact deformation which is irreversible and dependent on defor-

mation rate and consolidation time
More details on the respective models accounting for the above deformation
effects can be found in the literatures listed in Table 1.
In Fig. 6, the forces and torques acting on the particulates can be described
according to the contact forces, moments, and degrees of freedom:
• Along the line normal to the particulate centers

• Resistance from shearing or sliding, twisting, and rolling of one particulate over
another
For the simulation particulate dynamics by the discrete element method, the linear
spring-dashpot model (Cundall and Strack 1979) is normally employed. This model
can be combined to include viscous damping (Saluena et al. 1999), plasticity
(Luding and Herrmann 2001; Luding et al. 2001), and liquid bridge bonds (Lian
et al. 1993). The basic nonlinear elastic behavior derived by Hertz (1882), also
known as the nonlinear spring-dashpot model, can also be combined to include

viscoelasticity (Pao 1955, Lee and Radok 1960; Hunter 1960; Yang 1966) and
constant adhesion (Johnson et al. 1971).
Linear plastic behavior has been described by Walton and Braun (1986). The
increase of adhesion due to plastic contact information has nonetheless been intro-
duced by Molerus (1975) and Schubert et al. (1976). Nonlinear plastic,
displacement-driven contact has been investigated by Vu-Quoc and Zhang (1999)
while contact softening was considered by Tomas (2001).
Considering the above theories and constitutive models, a general contact model
for load, time, and rate dependent elastic, viscoelastic, plastic, viscoplastic, adhesion,
and dissipative behaviors can be developed to characterize the particulate-particulate
interactions for the specific problem of interest.
Contact Forces at the Level of Molecular Dynamics

The phenomenon of adhesion of particulates for fine (<100 μm), ultrafine (<10 μm),
and nanosized particulates (<0.1 μm) is important especially in particulate technol-
ogy and processing. Within the context of the soft sphere model, the adhesive forces
are assumed to act on length scales much smaller than the particulate size which
means that these forces have no effect until two or more particulate collide.
For the formation of particulate clusters or agglomerates, various forces between
particles act on length scale much smaller than the particle size give rise to partic-
ulate adhesion and microprocesses of particulate bond effects while in contact
including:
Surface and field forces at direct contact
• van der Waals forces

• electrostatic forces (electric conductor and nonconductor)
• magnetic forces (external and dipole-dipole attraction)
• interfacial attractive forces
Material bridges between particulate surfaces
• Organic macromolecules such as flocculants in suspensions

• Liquid bridging (low viscous wetting liquids by capillary pressure and surface
tension and high viscous bond agents such as resins)
• Solid bridging (freezing of liquid bridge bonds, recrystallization of liquid bridges
which contains solvents, contact fusion by sintering, chemical reactions with
adsorbed surface layers, solidification of high viscous bond agents, chemical
bonds by solid-solid reactions, and solidification of swelled ultrafine solid
particulates)
Interlocking by macromolecular and particulate shape effects
• Interlocking of chain branches at macromolecules such as proteins

+- NS
Dipole
Non- Magnetic
Molecule Conductor
Conductor Dipole
-+ + + +
SN
+ + Fluid 1
+ +
+ +
+- + +++++ + + + + NS
Interface
---
-+ - - -
S N
- - - -
- -
- - Fluid 2
- -
+- - - - NS
Van der Walls Electrostatic Magnetic Interfacial

Forces Forces Force Attractive Forces
Surface and Field Forces at Direct Contact
Interlocking by Hook-Like Bonds
Low High Recrystallization of

Viscosity Viscosity Liquid Bridges
Organic Liquid Bridge Solid Bridge

Macromolecules Bonds Bonds
Material Bridge between Contacts
Fig. 7 Particulate adhesion via different bonding effects
• Interlocking of contacts by overlaps of surface roughness

• Interlocking by hook-like bonds
All the above adhesive particulate forces that are schematically illustrated in
Fig. 7 are by no means exhaustive, but they are sufficient representatives to provide
a modeling framework to tackle the many challenging engineering problems of
interest.
Classification of Fluid-Particulate Forces
The Stokes number can be defined as
tp
St ¼ (40)
tf
where tp is the relaxation time and tf is the fluid integral scale of the particulate. For
St << 1, the particulates can be considered to follow the fluid streamline with a
small drift velocity relative to the fluid velocity. At St = O(1), large particle
dispersion relative to the fluid velocity becomes more significant. For St >> 1, the
particulates are considered to be no longer in equilibrium with the surrounding fluid
phase, and they divert rather substantially from the fluid stream path leading to
significant momentum transfer from the particulate to the fluid. Here, the inertia
effect of particulates becomes more prevalent and will therefore exert a significant
influence on the surrounding fluid.
According to Clift et al. (1978), Shirolka et al. (1996), Crowe et al. (1998), and
Gouesbet and Berlemont (1999), the fluid-particulate forces include the drag, virtual
or added mass, Basset history, Saffman lift, Magnus, pressure gradient, and reduced
gravity.
For the drag force, it can be expressed in terms of the drag coefficient CDas
π
¼ ρf d 2p CD Uf Ud Uf Ud
Drag
Fd (41)
8
The drag coefficient is normally formulated as a function of the particle Reynolds
number: Red = ρf|Uf Ud|dp/μf. Suitable relationships for the drag coefficient can
be specified depending on the different types of flows and configurations being
considered.
For the virtual or added mass force, it can be expressed in terms of the added mass
coefficient KA as

dUf dUd
FVirtual
d ¼ K A md (42)
dt dt
This force originates because of the difference in acceleration and becomes

dominant when significant difference in the density is present between the fluid
and particulate.
For the Basset history force, it can be expressed in terms of the Basset history
coefficient KB as
ðt
pffiffiffiffiffiffiffiffiffiffiffi dUf dUd ds
FBasset ¼ K B d2p πρf μf pffiffiffiffiffiffiffiffiffiffi (43)
d
dt dt ts
t0
The nature of this force becomes significant due to transitory nature of the
particulate’s boundary layer especially in an oscillatory flow field.
For the Saffman lift force (Saffman 1965), it can be expressed in terms of the lift
coefficient KL as

ρf Uf Ud Ωf
FSaf fman
¼ K L md pffiffiffiffiffiffiffiffiffiffiffiffi (44)
d
ρd Red αL
where αL = |Ωf|dp/(2|Uf Ud|) and Ωf = ∇ Uf. This force becomes prevalent

especially at very low Reynolds number where it has been demonstrated that a small
rotating particle moving in a uniform shear flow experiences a lift force both due to
pressure difference between the top and bottom of the particle when it rotates with
the fluid and local gradients of transitional fluid velocities. For flows of finite
Reynolds number, particle rotation appears to have little influence on this force.
Further increasing the Reynolds number has shown that the lift force decreases
significantly for both rotating and nonrotating conditions. According to Bagchi and
Balachandar (2002), the lift on nonrotating sphere decreases more rapidly than that
on the rotating sphere.
For the Magnus force, it can be expressed as

Magus 3 ρf Ωf
Fd ¼ md Uf Ud Ωd (45)
4 ρd 2
At very high Reynolds number, this force provides an added lift to the particulate
rotating at the same rate as the fluid flow or to obtain the total lift force acting on a
particle traveling through the fluid with arbitrary rotation rate. The corresponding
viscous torque acting on the particulate due to the differential fluid rotation rate can
be determined by

Magus Ωf
Mf ¼ πμf d 3p Ωd (46)
2
For the pressure gradient force, it can be expressed as
ρf DUf
FPressure ¼ md (47)
d
ρd Dt
This force is required to be considered for the case where it could accelerate the
fluid which would occupy the particulate volume if the particulate is absent.
For the reduced gravity force which includes both the gravitational force and the
corresponding fluid buoyancy force acting on the particulate, it can be expressed as

ρf
FPressure ¼ 1 md g (48)
d
ρd
For submicron particulates, a random force as a result from Brownian motion

induced by individual molecular collision with the particulate can be expressed as
rffiffiffiffiffiffiffi
πSo
FBrownian ¼ Gi (49)
d
Δt
According to Li and Ahmadi (1992), Gi represents the zero-mean, unit-variance,
independent Gaussian random numbers, and So is given by
216σT f μf
So ¼ (50)
πρ2d d5p CC
where σ is the Stefan-Boltzmann constant and Tf is the fluid temperature. The

Cunningham correction to Stokes’ drag law CC can be calculated by the relationship:
2λ
CC ¼ 1 þ 1:257 þ 0:4eð1:1dp =2λÞ (51)
dp
where λ represents the molecular mean free path. In turbulent flows, a random force
is often employed to model the effects of subgrid scale turbulence on the dispersion
of particulates.
Eulerian Formulation
With ever increasing computer power, the Lagrangian formulation allows the pos-
sibility of resolving the motion of all the fluid around every gas bubble, liquid drop,
and solid particle and the position of these individual discrete constituents. None-
theless, it is in the examination of very complex, large-scale systems, where such
comprehensive treatment remains restrictive to only turbulent flows of low Reynolds
number and dynamics of a limited amount of these individual discrete constituents.
There is still a great need to adopt a macroscopic consideration based on the
multiphase flow continuum formulation. The averaging of the transport equations
provides the means of eliminating the interfacial discontinuities and allowing prac-
tical computations for large-scale multiphase flows of high Reynolds number.
In the Eulerian formulation, discrete constituents being the identifiable portion of
a particular phase that occupy flow systems are characterized as a fluid in the
dispersed phase. The continuous phase which occupies a connected region of
space is usually described as the fluid that carries these discrete constituents. Within
a multiphase flow system, multiple phases that coexist simultaneously in the flow
often exhibit relative motion among the phases as well as heat transfer across the
phase boundary. Because of the complexity of the microscopic motions and thermal
characteristics of the individual discrete constituents, solutions to the microlevel
evolutionary equations can be prohibitive due to the uncertainty of the particular

constituents at any instant of time and space.
For most practical purposes, exact prediction on the evolution of details within
the multiphase flow system is not deemed desirable. On the other hand, the gross
features of the fluid flow and heat transfer are of more significant importance. Owing
to the complexities of interfaces and resultant discontinuities in fluid properties as
well as from physical scaling issues, it is thereby customary to apply some sort of
averaging process to the conservation equations, which leads to derivation of the
conservation equations in the interpenetrating media framework.
Interpenetrating Media Framework
In the interpenetrating media framework, the averaging process allows computations

of the multiphase fluid dynamics through suitable numerical techniques in the
Eulerian formulation. This framework also permits the means of comparing the
computational results with available experimental data. Numerous averaging
approaches have been proposed. The adoption of an appropriate choice of averaging
in a multitude of multiphase flow investigations may be found in Vernier and
Delhaye (1968), Yadigaroglu and Lahey (1976), Delhaye and Achard (1976), Panton
(1968), Agee et al. (1978), Banerjee and Chan (1980), Drew (1983), Lahey and
Drew (2001a), Besnard and Harlow (1988), Joseph et al. (1990), Drew and Passman
(1999), Kolev (2005), and Ishii and Hibiki (2006).
In principal, averaging may be performed in time, space, over an ensemble, or in
some combination of these. Defining the instantaneous field to be ϕ(x, y, z, t), the
commonly used averaging approaches in computational multiphase flow can be
mathematically described as:
Time averaging:
ð
1
ϕðx, y, z, tÞ ¼ lim ϕðx, y, z, tÞdt (52)
T!1 T
Space averaging:
ððð
1
hϕiV ðtÞ ¼ lim ϕðx, y, z, tÞdV (53)
V!1 V
Ensemble averaging:
1 XN
hϕiE ðx, y, z, tÞ ¼ lim ϕ ðx, y, z, tÞ (54)
N!1 N n¼1 n
where T is an averaging time scale, V is the volume based on an averaging length

scale, and N is the total number of realizations. While averaging allows the mathe-
matical solution of the problem to be more tractable, the lost information regarding
the local gradients between each phase is recovered which have to be obtained
through the use of constitutive relationships. These relationships are generally in the
form of semiempirical closure expressions for various interphase interaction prop-
erties including the interfacial mass, momentum, and energy exchanges in the
conservation equations of mass, momentum, and energy.
On the basis of the continuum assumption, all phases behave like fluids
intermingling and interacting as continuous phases. The formulation of the local
instantaneous conservation equations of mass, momentum, and energy can therefore
be achieved through classical consideration, which is similar to the derivation of
conservation equations governing the single-phase fluid flow. To allow a clear
physical description of the jump or interphase interaction conditions which express
the conservation of mass, momentum, and energy at the interface, a phase indicator
function is introduced. Through the use of this phase indicator equation and applying
appropriate averaging, conservation equations of mass, momentum, and energy to
solve highly turbulent multiphase flows at the device scale can be derived. It should
be noted that either the space (volume) averaging or ensemble averaging could be
employed to formulate the effective conservation equations since both of them result
essentially in the same form of equations. For succinctness, we denote 〈〉 as a
generic averaging process of representing either space (volume) averaging 〈〉V or
ensemble averaging 〈〉E so that 〈ϕ〉 corresponds effectively to a generic averaged
field.
Derivation of Governing Equations
Mass Conservation
In the Eulerain formulation, the instantaneous conservation equation of mass may be
derived from the consideration of the flow for a continuum fluid. The equation is
deemed to be valid for the continuous flow in each phase. Similar to the consider-
ation of continuum mechanics such as been illustrated in Fig. 5, the formulation of
three-dimensional continuum equation can be described by tensor equations of
representative elemental finite volumes.
Defining an elemental finite volume of dV containing kth (= 1, 2, 3,. . .,N ) phase,
the fundamental physical principles can be applied to the infinitesimal small fluid
element. Considering the fluid element to be fixed in space and the fluid is permitted
to flow through it, this approach leads directly to the fundamental equation in partial
differential form. Note that N phase herein includes one of the phases taken to be the
continuous fluid while the dispersed phase can be represented by a multitude of
continuum fluids dependent on the number, type, and size of the particulates being
present in the multiphase flow. Such a characterization is commonly known as the
multifluid approach in the Eulerian formulation.
The fundamental physical principle governing the conservation of mass is:
Rate of increase of mass within the elemental volume

¼ Net rate at which mass enters the elemental volume (55)
On the left hand side of Eq. 52, since the mass of any fluid mk is the product of the
k phase density and the elemental volume, ρkdV, the rate of increase of mass within
th
the elemental volume is thus given by
@mk @ρk
¼ dV (56)
@t @t
On the right hand side of Eq. 52, in order to account for the mass flux across each
of the faces of the elemental volume, the net rate at which mass enters the elemental
volume is simply the divergence of the mass flux which can be expressed as

∇∙ ρk Uk dV (57)
Combining Eqs. 56 and 57 and dividing by the volume dV yields:
@ρk
þ ∇∙ ρk Uk ¼ 0 (58)
@t
which represents the local instantaneous equation for the mass conservation of the kth
phase.
The phase indicator function can be defined as
8
>
< 1, if ðx, y, zÞ is in kth phase at time t
φk ðx, y, z, tÞ ¼ 0, otherwise (59)
>
:
Drew and Passman (1999) have demonstrated that the topological equation
reflecting the material derivatives of φk for each kth phase following the interface
velocity Uint vanish as
@φk
þ Uint ∙∇ φk ¼ 0 (60)
@t
Based on Eq. 60, it can be demonstrated that both partial derivatives of φk in
space and time vanish away from the interface. The phase indicator function φk on
the interface can be regarded as the jump condition which remains constant so that
the material derivatives following the interface vanish. If mass transfer exists across
the interface from one fluid to the other, the interface moves not only by convection
but also by the amount of mass being transferred between the fields. It should thus be
noted that the interface velocity is not equivalent to the neighboring velocities of
each phase.
Using Eq. 60 and multiplying φk to both sides of Eq. 58, the following equation
can be derived:

@ ρk φ k
þ ∇∙ ρk Uk φk ¼ ρk Uk Uint ∙∇φk (61)
@t
Applying averaging to Eq. 61 and using the Reynolds, Leibnitz, and Gauss rules,
the instantaneous averaged equation for the mass conservation is given by
@ρk φk
þ ∇∙ρk Uk φk ¼ ρk Uk Uint ∙∇φk (62)
@t
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Γk
where Γk denotes the interfacial mass transfer.
Momentum Conservation
The instantaneous equation for the kth phase momentum conservation can be derived
through the consideration of Newton’s second law of motion which states that:
Rate increase of momentum of the elemental volume

¼ Sum of forces acting on the elemental volume (63)
On the left hand side of Eq. 63, given that the mass of any fluid mk is ρkdV, the rate
of increase of momentum within the elemental volume is given by
mk ak ¼ ρk ak dV (64)
Since the acceleration ak in Eq. 64 is simply the material derivative of Uk, Eq. 64
can be rewritten as
DUk
ρk dV (65)
Dt
On the right hand side of Eq. 63, the sum of forces acting on the elemental volume
comprises of two sources: surface forces and body forces. Defining Tk to be the
Cauchy stress tensor representing the respective stresses along the directions of x, y,
and z such as depicted in Fig. 5: σ kxx , σ kyy , σ kzz , τkxy , τkxz , τkyz and the body forces to be
Fk,bodyforces, these forces are normally incorporated as additional source or sink
terms in Eq. 63 as
X
∇ ∙ Tk dV þ F k,bodyforces dV (66)
Combining Eqs. 65 and 66 and dividing by the volume dV yields:

k X
@U
ρk
þ U ∙ ∇U ¼ ∇ ∙ Tk þ
k k
Fk,bodyforces (67)
@t
It is customary to represent Tk in terms of pressure pk and extra stress tensor τk.

This entails expressing the normal stresses σ kxx , σ kyy , σ kzz as the combination of pressure
and normal viscous stresses: σ kxx ¼ pk þ τkxx , σ kyy ¼ pk þ τkyy , σ kzz ¼ pk þ τkzz :
Therefore, Eq. 67 can be alternatively expressed as
X
@Uk
ρk þ Uk ∙ ∇Uk ¼ ∇pk þ ∇ ∙ τk þ Fk, bodyforces (68)
@t
Similar to the derivation of the averaged equation governing the mass conserva-
tion, the averaged equation governing the momentum conservation can be derived
by multiplying Eq. 68 with the phase indicator function. After some rearrangement,

@ ρk φk Uk
þ ∇ ∙ ρk φk Uk Uk ¼ ∇ φk pk þ ∇ ∙ φk τk þ pk ∇φk
@t

þ ρk Uk Uk Uint ∙∇φk τk ∙∇φk
X
þ φk Fk, bodyforces (69)
Applying the Reynolds, Leibnitz, and Gauss rules, the averaged equation
governing the momentum conservation is given by
@〈ρk φk Uk 〉 X
þ ∇∙〈ρk φk Uk Uk 〉 ¼ ∇〈φk pk 〉 þ ∇∙〈φk τk 〉 þ 〈φk 〉〈 Fk, bodyforces 〉
@t
þ〈ρk Uk Uk Uint ∙∇φk 〉 þ 〈pk 〉〈∇φk 〉〈τk ∙∇φk 〉
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Ωk
(70)
where Ω denotes the interfacial momentum transfer.
k
Energy Conservation
Based on the first law of thermodynamics, the instantaneous equation for the kth
phase energy conservation can be derived assuming no additional or removal of heat
due to external heat sources as
Rate increase of energy of the elemental volume

¼ Net rate of heat added to the elemental volume
þ Net rate of work done on the elemental volume (71)
On the left hand side of Eq. 71, analogous to the consideration of the conservation
of momentum, the time rate of change of energy for the elemental volume of the kth
phase is simply the product of the density ρk and material derivative of energy Ek.
The rate of increase of energy is thus given by
DEk
ρk dV (72)
Dt
On the right hand side of Eq. 71, with reference to Fig. 5, the rate of work done on
the elemental volume in the x direction is the product between the surface forces
(caused by the normal stress σ kxx and tangential stresses τkyx and τkzx ) and the velocity
component in the x direction. Work done due to surface stress components along the
y direction and z direction is derived accordingly. In addition to the work done by
surface forces on the fluid element, possible work done due to body forces is also
considered. Hence, the net rate work done can be written as
X X
W_ ¼ ∇∙ Uk ∙Tk dV þ Uk ∙ Fk, bodyforces dV (73)
Also, the net rate of heat transfer to the fluid due to the heat flow is given by the
difference between the heat entering and leaving the elemental volume. The total rate
of heat added is expressed by
X
Q_ ¼ ∇∙qk dV (74)
Combining Eqs. 72, 73, and 74 and dividing by the volume dV yields:
k X
@E
ρ k
þ U ∙∇E ¼ ∇∙qk þ ∇∙ Uk ∙Tk þ Uk ∙
k k
Fk, bodyforces (75)
@t
Equation 75 can also be alternatively expressed in terms of combination of

pressure, normal viscous stresses, and extra stress tensor as

@Ek
ρk þ Uk ∙∇Ek ¼ ∇∙qk ∇∙ Uk pk þ ∇∙ Uk ∙τk
@t
X
þ Uk ∙ Fk, bodyforces (76)
Similar to the derivation of the averaged equation governing the mass and
momentum conservation, the averaged equation governing the energy conservation
can be derived by multiplying Eq. 76 with the phase indicator function. After some
manipulation,

@ ρ k E k φk
þ ∇∙ ρk Uk Ek φk ¼ ∇∙ φk qk ∇∙ φk Uk pk þ qk ∙∇φk
@t

þ Uk pk ∙∇φk þ ρk Ek Uk Uk Uint ∙∇φk

þ ∇∙ φk Uk ∙τk Uk ∙τk ∙∇φk
X
þ φk Uk ∙ Fk, bodyforces (77)
Applying the Reynolds, Leibnitz, and Gauss rules, the averaged equation
governing the energy conservation is given by
@〈ρk Ek φk 〉
þ ∇∙〈ρk Uk Ek φk 〉 ¼ ∇∙〈φk qk 〉 ∇∙〈φk Uk pk 〉 þ ∇∙〈φk Uk ∙τk 〉þ
@t P
〈φk〉〈Uk ∙ Fk, bodyforces 〉þ
(78)
〈ρk Uk Uk Uint ∙∇φk 〉 þ 〈qk ∙∇φk 〉 þ 〈Uk pk 〉∙〈∇φk 〉〈Uk ∙τk ∙∇φk 〉
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Πk
where Πk denotes the interfacial energy transfer.

It should be noted that the specific energy Ek of a fluid can often be defined as the
sum of the specific internal energy and kinetic energy of the kth phase:
Ek ¼ ek 1 k k
þ U ∙U (79)
2
|{z} |fflfflfflfflfflffl{zfflfflfflfflfflffl}
specific internal
kinetic energy
energy for the kth phase
for the kth phase
Conservation equations for specific internal energy and kinetic energy may be
derived by substituting Eq. 79 into Eq. 78. Based on the definitions of the sensible
enthalpy hks and the total enthalpy Hk of a fluid given as
pk 1 k k
hks ¼ ek þ H k ¼ hks þ U U (80)
ρk 2
and combining these two definitions yields
pk 1 k k pk
H k ¼ ek þ þ U ∙U ¼ Ek þ k (81)
ρ k 2 ρ
Using Eq. 81, the equation for the total enthalpy can be formulated which may be
written in the averaged form as:
@〈ρk H k φk 〉 @〈pk φk 〉
þ ∇∙〈ρk Uk H k φk 〉 ¼ ∇∙〈φk qk 〉 þ ∇∙〈φk Uk ∙τk 〉þ
@t P @t
〈φk 〉〈Uk ∙ Fk, bodyforces 〉þ
@φk (82)
〈ρk Uk Uk Uint ∙∇φk 〉 þ 〈qk ∙∇φk 〉 þ 〈pk 〉〈Uk ∙τk ∙∇φk 〉
@t
Πk
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
The term ∇ ∙ 〈φkqk〉 at the right hand side in Eq. 82 can normally be formulated
by applying the Fourier’s law of heat conduction that relates the heat flux to the local
temperature gradient. In other words, qk = ∇(λkTk) where λk is the thermal
conductivity for the kth phase. It is also worth noting that term @φk/@t in Eq. 82
represents the local acceleration term of the material derivative of the phase indicator
function as stipulated by Eq. 60. It may be rewritten as @φk/@t = Uint ∙ ∇φk
Conservation Equations for Turbulent Multiphase Flow

In most practical flows of interest, turbulence is associated with the existence of
random fluctuations in the fluid. The presence of the random nature of the fluid flow
generally precludes computations based on the equations that describe the fluid
motion to be carried out to the desired accuracy. It is therefore preferable that
some means of practically resolving the random transient distribution of the instan-
taneous flow field with time or space is realized for practical computations.
Reynolds-Averaged Closure
One feasible approach to handle turbulent multiphase flow in the context of
interpenetrating media framework is to consider any instantaneous field ϕ to be
decomposed into a steady mean motion ϕ and its fluctuating component ϕ0 :
ϕ ¼ ϕ þ ϕ0 (83)
This particular flow decomposition presents an attractive way of characterizing a

turbulent flow by the mean values of flow properties with its corresponding statis-
tical fluctuating property. The time averaged of the fluctuating component ϕ0 is zero.
In other words,
ð
0 1
ϕ ¼ limT!1 ϕ0 ¼ 0 (84)
T
By applying volume-averaging or ensemble-averaging, the instantaneous aver-
aged field can be written as
〈ϕ〉 ¼ 〈ϕ〉 þ ϕ00 (85)

0 0
In accordance with Eq. 84, the time averaged of the fluctuating component ϕ is
also, by definition, zero.
In multiphase flow analysis, preference is normally given to the Favre-averaging

approach in order to alleviate the complication of modeling additional correlation
terms containing averages of fluctuating quantities. Two types of averaged variables
are employed. The phase-weighted average for the field ζ can be defined by
〈ζφk 〉
〈ζ〉 ¼ (86)
〈φk 〉
and the mass-weighted average of the field ψ can also be defined in accordance with
〈ψρk 〉
〈ψ〉 ¼ (87)
〈ρk 〉
It follows that ζ 00 〈φk 〉 and ψ 00 〈ρk 〉 are zeros.

The local volume fraction (volumetric concentration or relative residence time)
represents an important parameter in multiphase flow investigations. It can usually
be defined as the fraction of time in which the continuous or dispersed phase
occupies a particular given point in space. In principal, the local volume fraction
αk can be regarded as the ratio of the fractional volume ϑk of the kth phase in an
arbitrary small region over the total volume V of the region within the multiphase
flow. It also corresponds to the volume-averaged of the phase indictor function, i.e.,
αk= ϑk /ϑ =〈φk〉. In the event where the transport equations governing the conser-
vation of mass, momentum, and energy are volume-averaged and subsequently time-
averaged, suitable forms of these equations via the phase-weighted and mass-
weighted averages for the multifluid model can be formulated.
Dropping the bars and parentheses which they denote be default the Favre-
averaging and volume-averaging processes being performed on the transport equa-
tions, the so-called Reynolds-averaged form of the governing equations written in
terms of the local volume fraction and products of averages can be derived as
Mass conservation:
@ k k
ρ α þ ∇∙ ρk Uk αk ¼ Γ0k (88)
@t
Momentum conservation:
@ k k k
ρ α U þ ∇∙ ρk αk Uk Uk ¼ ∇ pk αk þ ∇∙ αk τk ∇∙ αk τk00
@t
X
ak Fk, bodyf orces þ Ω0k (89)
Energy conservation:
@ k k k @ k k 00

ρ α H þ ∇∙ ρk αk Uk Hk ¼ p α ∇∙ αk λk T k ∇∙ αk qk
@t @t
00

þ ∇∙ αk Uk ∙τk ∇∙ αk Uk ∙τk
X
þ αk Uk ∙ Fk, bodyforces þ Π0k (90)
It is worth noting that if ensemble averaging is performed on the governing

equations, and fluctuating quantities are subsequently introduced into the equations,
the final forms of the governing equations are no different from those of the volume-
averaged and time-averaged conservation equations.
In Eqs. 88, 89, and 90, the interfacial terms Γ0 k, Ω0 k, and Π0k represent the Favre-
averaging that is subsequently performed on top of the volume-averaged terms: Γk,
Ωk, and Πk. These interfacial exchange terms can be modeled in accordance with
XN
Γ0k ¼ l¼1
ðm_ lk m_ kl Þ (91)
XN k, drag k, nondrag
Ω0k ¼ l¼1
m_ lk Ul m_ kl Uk þ pkint ∇αk þ FD þ FD (92)
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
FkD
X
N
Π0k ¼ m_ lk H l m_ kl H k Þ þ Qint (93)
l¼1
where m_ lk and m_ kl characterizes the mass transfer from the lth phase to kth phase and
from the kth phase to lth phase, respectively. From mass conservation,
m_ kk ¼ m_ ll ¼ 0: The interfacial force FkD in Eq. 92 is usually decomposed in terms
k, drag k, nondrag
of the drag force FD and nondrag forces in FD . The interfacial drag force
k, drag
FD and interfacial heat source Qint can usually be expressed as
k, drag
X
N
FD Bkl Ul Uk (94)
l¼1
X
N
Qint Ckl T l T k (95)
l¼1
where Bkl and Ckl are the interphase drag and heat transfer terms. Through appro-
priate modeling considerations, closure to the interfacial exchange terms can be
attained through prescribed algebraic functions of the governing flow parameters.
In most multiphase flow problems, the fluid is normally assumed to be Newto-

nian. Therefore, the normal and shear stress τk can be taken to be proportional to the
time rate of strain, which is the velocity gradient. The normal and shear viscous
stress components for the kth phase according to the Newton’s law of viscosity are
h T i 2 k
τk ¼ μk ∇Uk þ ∇Uk μ ∇∙Uk I (96)
3
where μk is the dynamic viscosity for the kth phase. In Eqs. 89 and 90, τk 00 and qk00
are turbulent fluxes which can be estimated from the eddy viscosity and eddy
diffusivity hypotheses. According to Lopez de Bertodano et al. (1994a, b), this
concept suggests that the Reynolds stress τk 00 can be correlated with the mean
rates of deformation for different phases in analogous to the Newton’s law of
viscosity. The Reynolds stress τk 00 can thus be expressed as
h T i 2 k 2
τk ¼ μkt ∇Uk þ ∇Uk μt ∇∙Uk I ρk kk I (97)
3 3
where kk is the turbulent kinetic energy and μkt is the so-called turbulent or eddy
viscosity which is taken to be a function of the flow rather than of the fluid and it is
required to be prescribed. Similar to the eddy viscosity hypothesis, the eddy diffu-
sivity hypothesis for the Reynolds flux qk00 can be taken to be proportional to the
gradient of the transported quantity. For the total enthalpy, the Reynolds flux term
can be modeled as
00
qk ¼ Γkt ∇H k (98)
The term Γkt in the above expression is the eddy diffusivity for total enthalpy of the
fluid. Since the turbulent transport of momentum and heat can be attributed through
the same mechanisms – eddy mixing – the value of the eddy diffusivity in Eq. 98 can
be taken to be close to that of the eddy viscosity μkt . By definition, the turbulent
Prandtl number, which is the ratio between momentum diffusivity (viscosity) and
thermal diffusivity, can thus be written as
μkt
Pr kt ¼ (99)
Γkt
To satisfy dimensional requirements, at least two scaling parameters are required

to relate the Reynolds stress to the rate of deformation. One feasible choice is the use
of the turbulent kinetic energy kk and the rate of dissipation of turbulent energy ek.
The local turbulent viscosity in Eq. 99 can be obtained either from dimensional
analysis or from analogy to the laminar viscosity as μkt / ρk vt l . Based on the
pffiffiffiffiffi
characteristic velocity vt defined as kk and the characteristic length l as (kk)3/2/ek, the
turbulent viscosity can be calculated according to
k 2
k
μkt ¼ Cμ ρ k
(100)
ek
where Cμ is an empirical constant and local values of kk and ek are obtained in time
and space through solutions of appropriate two-equation turbulence models.
Following proposals by Lopez de Bertodano et al. (1994a) and Lahey and Drew
(2001b), the system of turbulent scalar equations at high Reynolds numbers for the
continuum equations in the interpenetrating media framework is straightforward
generalizations of the single-phase counterpart of the two-equation k e model
developed by Tennekes and Lumley (1976) and Versteeg and Malalasekera (1995).
Other turbulence models such as the Reynolds stress and shear stress transport (SST)
could also be applied in place of the two-equation k e model. The Reynolds stress
model has a greater potential to account for flows having strongly anisotropic
turbulence. SST which combines the two-equation k e model and two-equation
k ω model where ω is the turbulent frequency can be employed to better handle
nonequilibrium boundary layer regions.
Spatial Filtering
In contrast to Reynolds-averaged closure, another approach to handling multiphase
flows within the interpenetrating media framework is through the consideration of
large eddy simulation (LES). The basic idea behind LES is that the large scale
motions are solved directly and the small scale motions are represented in terms of
subgrid scale models. Significant advances in computational resources are looking
towards LES as the preferred methodology to be adopted for many turbulence
investigations of fundamental fluid dynamics problems. Since all real-world flows
are inherently unsteady, LES provides the means of obtaining such solutions and is
gradually replacing the use of two-equation turbulence models.
The Favre-averaging approach, as utilized in Reynolds-averaged closure, is
adopted in LES in order to alleviate the complication of modeling additional
correlation terms containing fluctuating quantities. Here, the phase-weighted aver-
age for the field ζ can also be defined as
Ð 0
ζ ðx0 , tÞφk ðx0 , tÞ Δ ζ ðx , tÞφk ðx0 , tÞGðx x0 Þdx0
ζ ðg
x0 , tÞ ¼ ¼ Ð (101)
Δ φ ðx , tÞcdx
k 0 0
φk ðx0 , tÞ
and the mass-weighted average of the field ψ can also be defined in accordance with
Ð 0
ψ ðx0 , tÞρk ðx0 , tÞ Δ ψÐðx , tÞρk ðx0 , tÞGðx x0 Þdx0
ψ ðg
x0 , t Þ ¼ ¼ k 0 0 0
(102)
ρk ð x0 , t Þ Δ ρ ðx , tÞGðx x Þdx
The instantaneous variables of ζ and ψ are:

ζ ðx0 , tÞ ¼ ζ ðg
x0 , tÞ þ ζ 00 ðx0 , tÞ (103)
ψ ðx0 , tÞ ¼ ψ ðg
x0 , tÞ þ ψ 00 ðx0 , tÞ (104)
x0 , tÞ and ψ ðg
where ζ ðg x0 , tÞ represent the filtered or resolvable components (essentially
local averages of the complete field) and ζ 00 (x0 , t) and ψ 00 (x0 , t) are the subgrid
scale components that account for the unresolved spatial variations at a length
smaller than the filter width Δ.
In Eqs. 101 and 102, G(x x0 ) represents an appropriate spatial filter function to
be applied for the problem in question. The most common localized filter functions
can be represented by
Top hat:
8
< 1 for jx x0 j < Δ
0
Gðx x Þ ¼ Δ 2 (105)
:
0 otherwise
Gaussian:
rffiffiffiffiffiffiffiffi !
0 2
6 6 ð x x Þ
G ð x x0 Þ ¼ exp (106)
πΔ2 Δ2
Spectral cutoff:
sin ðke ðx x0 ÞÞ π
G ð x x0 Þ ¼ , ke ¼ (107)
π ðx x0 Þ Δ
Within the interpenetrating media framework, it is also customary in the context

of LES that the volume fraction can be expressed as
ð
αk ðx0 , tÞ ¼ ζ ðx0 , tÞ ¼ φk ðx0 , tÞGðx x0 Þdx0 (108)
Δ
Dropping the bars which by default denote Favre-averaging, the filtered conser-
vation equations are:
Mass conservation:
@ k k 00
ρ α þ ∇∙ ρk Uk αk ¼ Γ k (109)
@t
@ k k k
ρ α U þ ∇∙ ρk αk Uk ⨂Uk ¼ ∇ pk αk þ ∇∙ αk τk
@t
X k, bodyf orces
∇∙ αk τk00 αk F þ Ω00k (110)
Energy conservation:
@ k k k
ρ α U þ ∇∙ ρk αk Uk ⨂Uk ¼ ∇ pk αk þ ∇∙ αk τk ∇∙ αk τk00 (111)
@t
where Γ 0 0 k, Ω0 0k, and Π0 0 k are the filtered interfacial mass, momentum, and energy
balance source terms. Note that τk has been defined in Eq. 96 for Newtonian fluid.
The unresolved subgrid stress tensor τk 0 0 in Eqs. 89 and 90 can be modeled in
analogous with the Reynolds-averaged closure via the Boussinesq hypothesis:
τk ¼ ρk Ukg fk ¼ 2μk S~ k þ 1 τk 00 I
fk ⨂U
00
⨂Uk ρk U T SGS
1 h k k T i 1
3 (112)
~ k
S ¼ ∇U þ ∇U ∇∙Uk I
2 3
where S~ is the strain rate of the large scale or resolved field and μkT SGS is the subgrid
scale eddy viscosity for the kth phase are determined through solutions of appropriate
subgrid scale models. Also, the unresolved subgrid enthalpy flux qk00in Eq. 111 is
modeled in a manner similar to the subgrid turbulence stresses by the standard
gradient diffusion hypothesis as
00 g fk H μk
qk ¼ ρk U k
H ρk U ~ ¼ T SGS ∇H (113)
Pr kT SGS
where Pr kT SGS is the subgrid turbulent Prandtl number for the kth phase.
Since the smallest turbulence eddies are almost isotropic, the Boussinesq hypoth-
esis provides a good description of the unresolved eddies (Smagorinsky 1963).
Taking the length scale to be the filter width, the velocity scale can be expressed
as the product of the length scale and the average strain rate of the resolved flow. This
thus brings about the formulation of the Smagorinsky-Lilly model which assumes
that the subgrid scale eddy viscosity can be described in terms of a length and a
velocity scale. Nevertheless, the Smagorinsky-Lilly model has been designed for
flow which is highly turbulent, fully developed, and isotropic but does not accom-
modate any eventual departure of the flow from these assumptions. In order to attain
an automatic adaptation of the model for inhomogeneous flows, simulations of
multiphase flows have been performed on the dynamic formulation of the model
(Germano et al. 1991).
Population Balance Approach
Presence of particulates regardless whether they are inherently present within or

deliberately introduced into the flow system significantly affects the behavior of the
multiphase flow. Because of mounting interests in determining the influence of
particulates within the flow system, population balance has the capacity of resolving
the microphysics which occurs at the mesoscale level. This in turn allows the
behavior and dynamic evolution of the population of discrete particulates to be
better synthesized. Much consideration has been concentrated towards describing
the spatial and temporal evolution of the geometrical structures as a result of the
formation and destruction of agglomerates or clusters through interactions among
and between discrete particulates and collisions with turbulent eddies. The primary
issue concerning the use of macroscopic formulation based on averaging of the
transport equations is the determination of the interfacial rates of the interphase
interaction terms. These terms provide the appropriate closure to the equations which
can be realized through the consideration of suitable mechanistic models accounting
for the physical interaction between the different phases. One such approach to
determine the local size distribution in space and time is population balance
modeling.
In essence, the population balance of any system is a record for the number of
particulates whose presence or occurrence governs the overall behavior of the flow
system under consideration. Record of these discreet elements is dynamically
dependent on the birth and death processes that create new discrete and terminate
existing particulates within a finite or defined space. Since a multiphase flow
generally contains millions or billions of discrete particulates that are simultaneously
varying in space and time, the feasibility of direct numerical simulation in resolving
such flow is still far beyond the capacity of existing computational resources.
Population balance, which records the number of these particulates as an averaged
function through the population balance equation, has shown to be extremely
promising in handling the flow complexity because of its comparatively lower
computational requirements.
Definition of Density Function and Continuous Phase Vector

In this framework, the population of particulates can be treated of not only being
distributed in the physical space but also in an abstract property space. Dependent
variables of these entities can be taken to exist in two different coordinates: internal
and external coordinates. Mathematically, the internal coordinates are the property
coordinates while the external coordinates are the spatial coordinates. Figure 7
illustrates the internal and external coordinates involved in the population balance
for a three-phase flow. The joint space comprising the internal and external coordi-
nates is referred to as the space of particulate phase. In this space, the quantity of
basic interest which characterizes the distinct particulates is the consideration of the
density function.
Liquid flow
Internal Coordinates External Coordinates
Number
Fractional volume
occupy by gas
particles at (r2,t2)
Size
Gas
Particles
Changes of
internal properties Solid
caused by the particles
“Birth” or “Death” Changes of
processes external variables
resulted from the
Number
Fluid Motions
Fractional volume
occupy by gas
particles at (r1, t1)
Size
Fig. 8 Population balance for gas-liquid-particle flow (Yeoh et al. 2014)
Based on Ramkrishna (2000), the density function f1(x, r, t) can be defined as the
number of particles per unit volume of the particle phase at time t at the internal
coordinates x (x1, x2, . . . , xn) where n represents the number of different quan-
tities associated with the particle and at the external coordinates r (r1, r2, r3)
which may be employed to indicate the position vector of the particulate. This
number density f1(x, r, t)is usually taken to be smooth so that it can be differentiated
with respect to any of its arguments. The particle state vector accounts for both
internal and external coordinates where the domain of internal coordinate shall be
taken to be Vx while the domain of external coordinates to be Vr that represents a set
of points in the physical space in which particulates are present.
The behavior of each particulate being affected by the continuous phase may also
be collated into a finite dimensional vector field. The continuous phase vector which
can be defined by Y = (r, t) is a function of only the external coordinates r and time
t, and it is calculated from the governing conservation equations associated with the
particular problem. It should be noted that a continuous phase may not be necessarily
considered for some applications where interaction between the population of
particulates and the continuous phase does not result in a substantial change in the
continuous phase. Analysis on the population thereby reduces to only the consider-
ation of population balance (Fig. 8).
Population Balance Equation

The development of the population balance equation can be traced back to as early as
the end of eighteenth century via the Boltzmann-type equation, proposed by Ludwig
Boltzmann. Such an equation could be regarded as the first population balance

equation which could be expressed in terms of a statistical distribution of molecules
or particles in a state space. The fundamental variable is the particle distribution
function along with an appropriate choice of internal coordinates pertaining to a
particular problem being solved. Defining p(x, r, c, t)dxdrdc to be the particle
number density distribution function which is assumed to be continuous and spec-
ifies the probable number density of particles with internal coordinates about x in the
range of dx, about position r in the range of dr, and about velocity c in the range dc,
at about time t, an equation for the particle number density distribution function can
be written as
pðx þ dx, r þ dr, c þ dc, t þ dtÞdxdrdc pðx, r, c, tÞdxdrdc

¼ Sp dxdrdcdt (114)
Where dr = cdt, dc = Fdt in which F is the force per unit mass acting on a
particle and Sp consists of the rates of change of p(x, r, c, t) due to death and birth of
particulates as well as other sources or sinks due to particulate interactions (for
example, rebounding of particulates). By assuming that the change of particulate
velocity within the time interval t to t + dt is negligible, Eq. 113 reduces to
@f
þ ∇r ∙ðvr f Þ þ ∇x ∙ðvx f Þ ¼ Sf (115)
@t
The above equation is analogous to the Boltzmann-type equation in describing
the temporal and spatial rate of change of the distribution function:
ð þ1
f ðx, r, tÞ ¼ pdc (116)
1
which simply denotes the probable number density of particulates with internal
coordinates about x in the range of dx, about position r in the range of dr, and at
about time t. Incidentally, that this function is by definition the same as the density
function f1(x, r, t). The source/sink term at the left hand side of Eq. 116 is
ð þ1
Sf ¼ Sp dc (117)
1
In Eq. 115, integrating the force F over the whole velocity space results in no net
contribution due to this force and thus it does not explicitly appear in the population
balance equation since the distribution vanishes as the velocity approaches to infinity
(1). Also, the source/sink term consists of only the net generation rate of particles
due to death and birth processes while other sources or sinks due to particle
interactions vanish as the number of particulates is conserved especially for the
case during the rebounding processes in the system.
Integrated Forms of Population Balance Equation

For a nonreactive and isothermal multiphase flow system, it is customary to assume
that all relevant internal variables can be calculated from the consideration of the
particulate volume or diameter. If the internal coordinate is taken to be the particulate
volume Vp or diameter dp in describing incompressible particulate dispersions. For a
multiphase flow system where compressibility effect becomes important, the use of
particulate mass mp as internal coordinate may prove to be more advantageous. This
requirement is particularly important in the gas phase because this quantity is
required to be conserved under pressure changes. The population balance equation
can now be written in accordance with the particulate volume, diameter, and mass as
@ @
f 1 V p , r, t þ ∇r ∙ vr V p , r, Y, t f 1 V p , r, t þ ̇ V p f 1 V p , r, t
@t @V p

¼ Sf 1 V p , r, Y, t (118)
@ @
f dp , r, t þ ∇r ∙ vr dp , r, Y, t f V p , r, t þ ̇ dp f 1 d p , r, t
@t @V p

¼ Sf 1 dp , r, Y, t (119)
@ @
f 1 mp , r, t þ ∇r ∙ vr mp , r, Y, t f 1 mp , r, t þ ̇ m p f 1 V p , r, t
@t @V p

¼ Sf 1 mp , r, Y, t (120)
where ̇ V p denotes the time rate of change of particle volume, d p_ represents the time
rate of change of particle diameter, and ̇ m p represents the time rate of change of
particle mass. To close the population balance problem, closure is required for the
growth of particulates as well as the death and birth kernels through appropriate
models in the source/sink term Sf 1 . Note that these kernels are required to be
consistent with the internal coordinate used.
In practice, the population balance of particulates is solved through the consid-
eration of moment transformation. If the particle volume Vp is adopted as the internal
coordinate, the mth moments of the number density function in terms of volume can
be defined as
ð
1

mk ðr, tÞ ¼ f 1 V p , r, t V kp dV p (121)
0
th
Various m moments in equation (3–18) contribute to physically important
moments of the number density function. In retrospective, particle number density,
particle mass density, interfacial area concentration, and local volume fraction
essentially correspond to zero, first, second, and third order moments. If the volume
Vp is treated as the independent internal coordinate, the particle number density N,
average particle volume V , interfacial area concentration Ai, and local volume
fraction αp are:
ð
1

N ðr, tÞ ¼ m0 ðr, tÞ ¼ f 1 V p , r, t dV p (122)
0
01 1
ð

V ðr, tÞ ¼ m1 ðr, tÞ ¼ @ f 1 V p , r, t V p dV p A=N ðr, tÞ (123)
0
ð
1

Ai ðr, tÞ ¼ m2 ðr, tÞ ¼ f 1 V p , r, t πD2e dV p (124)
0
ð
1
π
αp ðr, tÞ ¼ m3 ðr, tÞ ¼ f 1 V p , r, t D3e dV p (125)
6
0
In Eqs. 124 and 125, it is noted that the particulate surface area and volume are
expressed in terms of equivalent diameters.
Source/Sink Term of Population Balance Equation

The death and birth processes in the source/sink term for the population balance
equation are taken to occur simultaneously. Thus, Sf 1 ðx, r, Y, tÞ can be obtained via
ð
Sf 1 ðx, r, Y, tÞ ¼ νðx0 , r, Y, tÞbðx0 , r, Y, tÞPðx, rj x0 , r, Y, tÞf 1 ðx, r, tÞdV x
Vx
ð
1 @ ðx~, rÞ
bðx, r, Y, tÞf 1 ðx, r, tÞ þ r ; x0 , r, Y, tÞf 2 ðx~,~r ; x0 , r, tÞ
aðx~,~ dV x
δ @ ðx, rÞ
Vx
ð
aðx0 , r0 ; x, r, Y, tÞf 2 ðx0 , r0 ; x, r, tÞdV x (126)
Vx
The terms ν(x0 , r, Y, t) and P(x, r| x 0 , r, Y, t) denote the average number of

particulates, and probability density function for particulates from the breakage of a
single particle of state (x0 , r0 ) in an environment of Y at time t. The probability
density function P(x, r| x 0 , r, Y, t) is generally taken as a continuously distributed
fraction over particle state space and is commonly associated with the daughter
particle size distribution function denoting the size distribution of daughter particles
produced upon the breakage of a parent particulate. This quantity needs to be
determined through either detailed modeling of the breakage process or from
experimental observation. The term b(x, r, Y, t) represents the specific breakage
rate of particulates or more commonly known as the breakage frequency.
In Eq. 126, δ represents the number of times identical pairs have been considered
in the interval of integration where 1δ is introduced to correct for the redundancy. The
aggregation frequency aðx~, r; x0 , r, Y, tÞ represents the fraction of pairs of particles of
states ðx~,~ r Þ and (x0, r0) that will aggregate. A coarse approximation of the pair
density function is assumed for f 2 ðx~,~ r ; x0 , r, tÞ f~1 ðx~,~ r , tÞf 01 ðx0 , r0 , tÞ, which implies
the absence of any statistical correlation between particles of state spaces (x, r) and
~
(x0 , r0 ) at any instant of time t. The term @@ ððxx,, rrÞÞ represents the Jacobian determinant.
In practice, the transport of particle number density is solved within the frame-
work of computational multiphase fluid dynamics. By taking the volume Vp as the
independent internal coordinate. Eq. 126 becomes
ð
1

Sf 1 ðx, r, Y, tÞ ¼ ν V 0p , r, Y b V 0p , r, Y P V p , rj V 0p , r, Y, t f 1 V 0p , r, t dV 0p b V p , r, Y
Vp
ðp
V
1 ~
f 1 V p , r, t þ r ; V 0p , r0 , Y f~1 V~ p ,~
a V p ,~ r , t f 01 V 0p , r0 , t dV 0p
2
0
ð
1

a V 0p , r0 ; V p , r, Y f 2 V 0p , r0 ; V p , r, t dV 0p (127)
0
Owing to the complexity associated with both the particle growth term as well as
birth and death processes in the source/sink term, appropriate constitutive relation-
ships are inevitably required to close the population balance equation. In most cases,
the detailed functionality of closure relationships and even the physical insights of
the birth and death processes are generally unknown or unresolved. Also, formula-
tion of the growth terms that may be dependent on the choice of internal coordinates
or particle properties leads to additional complexity of suitably characterizing the
particle phase in the population balance framework. Therefore, this parameterization
process represents the weakest component which still demands the greatest attention.
Practical Considerations of Population Balance Equation

For practical solutions, various numerical approaches have been developed to solve
the population balance equation. The three common methods are the Monte Carlo
methods, method of moments, and class methods.
Monte Carlo methods solve the population balance equation based on the statis-
tical ensemble approach (Domilovskii et al. 1979; Liffman 1992; Maisels et al.
2004). The main advantage is the flexibility and accuracy to track particulate
changes in a multidimensional system. Nevertheless, the accuracy of Monte Carlo
methods is greatly dependent on the number of simulation particles; extensive
computational time is thus required to track a multitude of particulates. Monte
Carlo methods are generally not easily coupled with the conceptual framework of
computational multiphase fluid dynamics.
The basic principle behind method of moments centers in the transformation of
the problem into lower-order of moments of the particulate size distribution. The
primary advantage is its numerical economy that condenses the problem substan-
tially by only tracking the evolution of a small number of moments (Frenklach
2002). More importantly, method of moments do not suffer from truncation errors in
the approximation of the particulate size distribution. Mathematically, the transfor-
mation from the space of the particulate size distribution to the space of moments is
extremely rigorous. Fraction moments, representing mean diameter or surface area,
pose serious closure problem. To overcome this, Frenklach and Wang (1991, 1994)
have proposed an interpolative scheme to aptly determine the fraction moment from
integer moments, namely, the method of moments with interpolative closure.
Another different approach to computing the moment is through the use of numerical
quadrature scheme such as the quadrature method of moments as suggested by
McGraw (1997). With the aim to solve multidimensional problems, Marchisio and
Fox (2005) extended the method by developing the direct quadrature method of
moments where the quadrature abscissas and weights are subsequently formulated as
transport equations. The main idea of direct quadrature method of moments is to
track the primitive variables appearing in the quadrature approximation, instead of
moments of the. In general, the method of moments represent a sound mathematical
approach and an elegant tool for solving the population balance equation with
limited computational burden.
Instead of inferring the particulate size distribution to derivative variables such as
moments, class methods directly simulate the main characteristics using primitive
variables. One approach is the adoption of an averaged quantity to represent the
overall changes of the particle population. Ishii and coworkers have formulated the
interfacial area concentration transport equation to simulate different flow regimes
and conditions (Hibiki and Ishii 2002; Sun et al. 2004a, b). In order to be consistent
with the form of conservation equations in the multifluid approach, the transport
equation of the averaged bubble number density can be adopted (Cheung et al.
2007a, b). A more sophisticated model namely the homogeneous multiple-size-
group model that was developed by Lo (1996) can provide the feasibility of
accounting different shapes and velocities. The inhomogeneous multiple-size-
group model developed by Krepper et al. (2005) consisted of further subdividing
the dispersed phase into N number of velocity fields.
Summary
Owing to the increasing reliance on computational investigations of multiphase

flows of natural and technological significance, the purpose of this chapter is to
present the appropriate conceptual frameworks that can be applied to aptly resolve
and attain a fundamental understanding of many different classifications of multi-
phase flows. For small scale flow system, it is becoming ever more possible to solve
directly the transport equations governing the conservation of mass, momentum, and
energy for each phase and compute every detail of the multiphase flow, the motion of
all the fluid around every particulate, and the position of every interface via the
Lagrangian formulation. For large scale flow system, such comprehensive treatment
remains prohibitive which is only restricted to turbulent flows of low Reynolds
number and the dynamics of a limited amount of individual particulates. The use of
multiphase flow continuum via the Eulerian formulation within the interpenetrating
media framework provides the effective way of predicting the gross features of the
multiphase flows. Because of mounting interests in determining the influence of
particulates affecting the flow system, the need of population balance to resolve the
microphysics which occurs at mesoscale is required in order to better synthesize the
behavior and dynamic evolution of particulates occupying the flow system. The
overarching issue especially using the macroscale formulation based on averaging
the transport equations governing the conservation of mas, momentum, and energy
is the determination of the interfacial transfer terms in providing the realistic physical
interaction between the different phases. One such approach to determine the local
size distribution of the particulates in space and time is through the population
balance approach. The population balance of any flow system is a record for the
number of particulates whose presence or occurrence governs the overall behavior of
the flow system under consideration. Record of these particulates is dynamically
dependent on the birth and death processes that terminate existing particulates and
create new particulates within defined space and time.
References
L.J. Agee, S. Banerjee, R.B. Duffey, E.D. Hughes, Some aspects of two-fluid models and their
numerical solutions, Second OECD/NEA Specialists Meeting on Transient Two-Phase Flow,
CEA, France, 1978
P. Bagchi, S. Balachandar, Effect on free rotation on the motion of a solid sphere in linear shear flow
at moderate Re. J. Fluid Mech. 473, 379–388 (2002)
S. Banerjee, A.M.C. Chan, Separated flow model I. Analysis of the averaged and local instanta-
neous formulations. Int. J. Multiphase Flow 6, 1–24 (1980)
D.C. Besnard, F.H. Harlow, Turbulence in multiphase flow. Int. J. Multiphase Flow 14, 679–699
(1988)
D. Bouvard, R.M. McMeeking, Deformation of interparticle necks by diffusion-controlled creep.
J. Am. Ceram. Soc. 79, 666–672 (1996)
R.S. Bradley, The cohesion between smoke particle. Trans. Faraday Soc. 32, 1080–1090 (1936)
N. V. Brilliantov, T. Pöschel, Adhesive interactions of viscoelastic spheres, Powder and Grains

2005, Ed. by R. Garcia-Rojo, H. J. Herrmann, S. McNamara, vols. 1 & 2 (Balkema Publisher,
Leiden, 2005), pp. 505–508
C. Catttaneo (1938). Sul Contatto di due Corpi Elastici: Distribuzione Locale Degli Sforzi,
Academia Nationale Lincei Rendiconti, Ser. 6, Vol. 27, pp. 342–348, 434–436, 474–478
S.C.P. Cheung, G.H. Yeoh, J.Y. Tu, On the modeling of population balance in isothermal vertical
bubbly flows – average bubble number density approach. Chem. Eng. Process. 46, 742–756
(2007a)
S.C.P. Cheung, G.H. Yeoh, J.Y. Tu, On the numerical study of isothermal bubbly flow using two
population balance approaches. Chem. Eng. Sci. 62, 4659–4674 (2007b)
R. Clift, J.R. Grace, M.E. Weber, Bubbles, Drops and Particles (Dover Publications, New York,
1978)
C.T. Crowe, M.P. Sharma, D.E. Stock, Particle-source-in cell (PSI-cell) model for gas-droplet flows.
J. Fluids Eng. 99, 325–332 (1998)
P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies. Geotechnique 29,
47–65 (1979)
B. Dahneke, The influence of flattening on the adhesion of particles. J. Colloid Interface Sci. 40,
1–13 (1972)
J.Y. Delenne, M.S. Youssoufi, F. Cherblanc, J.C. Benet, Mechanical behaviour and failure of
cohesive granular materials. Int. J. Numer. Anal. Methods Geomech. 28, 1577–1594 (2004)
J.M. Delhaye, J.L. Achard, On the averaging operators introduced in two-phase flow modeling,
Proceedings of CSNI Specialists Meeting on Transient Two-Phase Flow, Toronto, Canada, 1976
H. Deresiewicz, Contact of elastic spheres under an oscillating torsional couple. Trans. ASME
J. Appl. Mech. 21, 52–56 (1954)
B.V. Derjaguin, Untersuchungen über die Reibung und Adhäsion, IV. Kolloid Zeitschr. 69,
155–164 (1934)
A. Di Renzo, F.P. Di Maio, Comparison of contact-force models for the simulation of collisions in
DEM-based granular flow codes. Chem. Eng. Sci. 59, 525–541 (2004)
E.R. Domilovskii, A.A. Lushnikov, V.N. Piskunov, A Monte Carlo simulation of coagulation
processes. Izvestkya Akademi Nauk SSSR, Fizika Atmosfery I Okeana 15, 194–201 (1979)
D.A. Drew, Mathematical modeling of two-phase flow. Annu. Rev. Fluid Mech. 15, 261–291
(1983)
D.A. Drew, S.L. Passman, Theory of Multicomponent Fluids (Springer-Verlag, Berlin, 1999)
N.A. Fleck, L.T. Kuhn, R.M. McMeeking, Yielding of metal powder bonded by isolated contacts.
J. Mech. Phys. Solids 40, 1139–1162 (1992)
L. Fӧppl, Die Strenge Lӧsung für die Rollende Reibung (Lebnitz Verlag, Munchen, 1947)
M. Frenklach, Method of moments with interpolative closure. Chem. Eng. Sci. 57, 2229–2239
(2002)
M. Frenklach, H. Wang, Detailed modeling of soot particle nucleation and growth, Proceedings of
the Twenty Third Symposium on Combustion, Combustion Institute, University of Orleans,
France, 1991
M. Frenklach, H. Wang, in Soot Formation in Combustion: Mechanisms and Models, ed. by
H. Bockhorn. Detailed mechanism and modeling of soot formation (Springer-Verlag, Berlin,
1994)
G. Fromm, Berechnung des Schlupfes beim Rollen deformierbarer Scheiben, Zeitschr. Angew.
Math. Mech 7, 27–58 (1927)
M.U. Germano, U. Piomelli, P. Moin, W.H. Cabot, A dynamic subgrid-scale Eddy viscosity model.
Phys. Fluids A 3, 1760–1765 (1991)
G. Gouesbet, A. Berlemont, Eulerian and Lagrangian approaches for predicting the behaviour of
discrete particles in turbulent flows. Prog. Energy Combust. Sci. 25, 133–159 (1999)
J.A. Greenwood, J.B.P. Williamson, Contact of nominally flat surfaces. Proc. Roy. Soc. Lond. A
295, 300–319 (1966)
J.A. Greenwood, Adhesion of Elastic Spheres, Proc. R. Soc. Lond. A 453, 1277–1297 (1997)
H. Hertz, Über die Berührung fester elastischer Körper (on the contact of elastic solids). J. Reine.
u. agnew. Math. 92, 156–171 (1882)
P.R. Heyliger, R.M. McMeeking, Cold plastic compaction of powders by a network model. J. Mech.
Phys. Solids 49, 2031–3054 (2001)
T. Hibiki, M. Ishii, Development of one-group interfacial area transport equation in bubbly flow
systems. Int. J. Heat Mass Transf. 45, 2351–2372 (2002)
B.P.B. Hoomans, J.A.M. Kuipers, W.J. Briels, W.P. Swaaij, Discrete particle simulation of bubble
and slug formation in a two-dimensional gas-fluidized bed: a hard-sphere approach. Chem. Eng.
Sci. 51, 99–118 (1996)
M.T. Huber, Zur Theorie der Berührung fester elastischer Körper. Ann. Phys. 14, 153–163 (1904)
S.C. Hunter, The Hertz problem for a rigid spherical indenter and a viscoelastic half-space. J. Mech.
Phys. Solids 8, 219–234 (1960).
M. Ishii, T. Hibiki, Thermo-Fluid Dynamics of Two-Phase Flow (Springer-Verlag, Berlin, 2006)
K.L. Johnson, K. Kendall, A.D. Roberts, Surface energy and the contact of elastic solids. Proc. Roy.
Soc. Lond. A 324, 301–313 (1971)
D.D. Joseph, T.S. Lundgren, R. Jackson, D.A. Saville, Ensemble averaged and mixture theory
equations for incompressible fluid-particle suspensions. Int. J. Multiphase Flow 16, 35–42
(1990)
N.I. Kolev, Multiphase Flow Dynamics 1: Fundamentals, 2nd edn. (Springer-Verlag, Berlin, 2005)
E. Krepper, D. Lucas, H. Prasser, On the modeling of bubbly flow in vertical pipes. Nuc. Eng. Des.
235, 597–611 (2005)
H. Krupp, Particle adhesion – theory and experiment. Adv. Colloid Interf. Sci. 1, 111–239 (1967)
H. Krupp, G. Sperling, Z. Phys. 19, 259–265 (1965)
L.T. Kuhn, R.M. McMeeking, Power-law creep of powder bonded by isolated contacts. Int.
J. Mech. Sci. 34, 563–573 (1992)
R.T. Lahey Jr., D.A. Drew. The three-dimensional time and volume averaged conservative equa-
tions of two-phase flow, in Advances in Nuclear Science and Technology, vol. 20, ed. by
J. Lewins, M. Becker (Plenum, 2001a)
R.T. Lahey Jr., D.A. Drew, The analysis of two-phase flow and heat transfer using multi-
dimensional, four field, two-fluid model. Nuc. Eng. Des. 204, 29–44 (2001b)
E.H. Lee, J.R.M. Radok, The contact problems for viscoelastic bodies. Trans. ASME J. Appl.
Mech. 27, 438–444 (1960)
J.S. Leszczynski, A discrete model of a two-particle contact applied to cohesive granular materials.
Granul. Matter 3, 91–98 (2003)
A. Li, G. Ahmadi, Dispersion and deposition of spherical particles from point sources in a turbulent
channel flow. Aerosol Sci. Technol. 16, 209–226 (1992)
S.Q. Li, J.S. Marshall, G.Q. Liu, Q. Yao, Adhesive particulate flow: the discrete-element method
and its application in energy and environmental engineering. Prog. Energy Combust. Sci. 37,
633–668 (2011)
G. Lian, C. Thornton, M.J. Adams, A theoretical study of the liquid bridge force between rigid
spherical bodies. J. Colloid Interface Sci. 161, 138–147 (1993)
K. Liffman, A direct simulation Monte Carlo method for cluster coagulation. J. Comput. Phys. 100,
116–127 (1992)
S.M. Lo, Application of population balance to CFD modeling of bubbly flow via the MUSIG
Model, AEAT-1096, AEA Technology (1996)
M. Lopez de Bertodano, R.T. Lahey Jr., O.C. Jones, Development of a k-e model for bubbly
two-phase flow. J. Fluids Eng. 116, 128–134 (1994a)
M. Lopez de Bertodano, R.T. Lahey Jr., O.C. Jones, Phase distribution in bubbly two-phase flow in
vertical ducts. Int. J. Multiphase Flow 20, 805–818 (1994b)
S. Luding, H.J. Herrmann, in Institut für Mechanik, ed. by S. Diebels. Zur Beschreibung komplexen
Materialverhaltens (2001), Stuttgart, pp. 121–134
S. Luding, M. Lätzel, W. Volk, S. Diebels, H.J. Herrmann, From discrete element simulations to a
continuum model. Comput. Methods Appl. Mech. Eng. 191, 21–28 (2001) HYPERLINK
"https://pure.tudelft.nl/portal/en/persons/s-luding(b29c6ff7-35b1-4123-8724-e2ee249b6bb3).
https://pure.tudelft.nl/portal/en/publications/micromacro-transition-for-cohesive-granular-
media(b254117d-f6f6-47e4-8e2e-2eb53e1017b6).html"Micro-Macro Transition for Cohesive
Granular Media. in S Diebels (ed.), Zur Beschreibung komplexen Materalverhaltens, Institut
für Mechanik.. pp. 121-134.
S. Luding, K. Manetsberger, J. Müller, A discrete model for long time sintering. J. Mech. Phys.
Solids 53, 455–491 (2005)
A.J. Lurje, Räumliche Probleme der Elastizitätstheorie (Akademie-Verlag, Berlin, 1963)
A. Maisels, F.E. Kruis, H. Fissan, Direct simulation Monte Carlo for simulation nucleation,
coagulation and surface growth in dispersed systems. Chem. Eng. Sci. 59, 2231–2239 (2004)
D.L. Marchisio, R.O. Fox, Solution of population balance equations using the direct quadrature
method of moments. J. Aerosol Sci. 36, 43–73 (2005)
D. Maugis, H.M. Pollock, Surface forces, deformation and adherence at metal microcontacts. Acta
Metall. 32, 1323–1334 (1984)
N. Maw, J.R. Barber, J.N. Fawcett, The oblique impact of elastic spheres. Wear 38, 101–114 (1976)
R. McGraw, Description of aerosol dynamics by the quadrature method of moments. Aerosol Sci.
Technol. 27, 255–265 (1997)
S.D. Mesarovic, K.L. Johnson, Adhesive contact of elastic-plastic spheres. J. Mech. Phys. Solids
48, 2009–2033 (2000)
R.D. Mindlin, Compliance of elastic bodies in contact. Trans. ASME J. Appl. Mech. 16, 259–267
(1949)
O. Molerus, Theory of yield of cohesive powders. Powder Technol. 12, 259–275 (1975)
O. Molerus, Effect of interparticle cohesive forces on the flow behaviour of powders. Powder
Technol. 20, 161–175 (1978)
R.J. Panton, Flow properties for the continuum viewpoint of a non-equilibrium gas particle mixture.
J. Fluid Mech. 31, 273–304 (1968)
Y.H. Pao, Extension of the hertz theory of impact to the viscoelastic case. J. Appl. Phys. 26,
1083–1088 (1955)
F. Parhami, R.M. McMeeking, A network model for initial stage sintering. Mech. Mater. 27,
111–124 (1998)
F. Parhami, R.M. McMeeking, A.C.F. Cocks, Z. Suo, A model for the sintering and coarsening of
rows of spherical particles. Mech. Mater. 31, 43–61 (1999)
D. Ramkrishna, Population Balances. Theory and Applications to Particulate Systems in Engi-
neering (Academic Press, San Diego, 2000)
P. Redanz, N.A. Fleck, The compaction of a random distribution of metal cylinders by the discrete
element method. Acta Mater. 49, 4325–4335 (2001)
H. Rumpf, K. Sommer, K. Steier, Mechanismen der Haftkraftverstärkung bei der Partikelhaftung
durch plastisches Verformen, Sintern und viskoelastisches Fließen. Chem. Ing. Tech. 48,
300–307 (1976)
M.H. Sadd, Q. Tai, A. Shukla, Contact law effects on wave propagation in particulate materials
using distinct element modelling. Int. J. Non-Linear Mechanics 28, 251–265 (1993)
P.G. Saffman, The lift on small sphere in slow sphere flow. J. Fluid Mech. 22, 385–400 (1965)
C. Saluena, T. Pöschel, S.E. Esipov, Dissipative properties of vibrated granular materials. Phys.
Rev. E 59, 4422–4425 (1999)
H. Schubert, K. Sommer, H. Rumpf, Plastisches Verformen des Kontaktbereiches bei der
Partikelhaftung. Chem. Ing. Tech. 48, 716 (1976)
J.S. Shirolkar, C.F.M. Coimbra, M.Q. McQuay, Fundamental Aspects of Modeling Turbulent
Particle Dispersion in Dilute Flows. Prog. Energy Combust. Sci. 22, 363–399 (1996)
J. Smagorinsky, General circulation experiment with the primitive equations: part I. The basic
experiment. Mon. Weather Rev. 91, 99–164 (1963)
B. Storakers, S. Biwa, P.L. Larsson, Similarity analysis of inelastic contact. Int. J. Solids Struct. 34,
3061–3083 (1997)
B. Storakers, N.A. Fleck, R.M. McMeeking, The viscoplastic compaction of composite powders.
J. Mech. Phys. Solids 47, 785–815 (1999)
X. Sun, S. Kim, M. Ishii, S.G. Beus, Modeling of bubble coalescence and disintegration in confined
upward two-phase flow. Nucl. Eng. Des. 230, 3–26 (2004a)
X. Sun, S. Kim, M. Ishii, S.G. Beus, Model evaluation of two-group interfacial area transport
equation for confined upward flow. Nuc. Eng. Des. 230, 27–47 (2004b)
L.M. Tavares, R.P. King, Modeling of particle fracture by repeated impacts using continuum
damage mechanics. Powder Technol. 123, 138–146 (2002)
H. Tennekes, J.L. Lumley, A First Course in Turbulence (MIT Press, Cambridge, MA, 1976)
C. Thornton, Interparticle sliding in the presence of adhesion. J. Phys. D. Appl. Phys. 24,
1942–1946 (1991)
C. Thornton, Coefficient of restitution for collinear collisions of elastic–perfectly plastic spheres.
Trans. ASME J. Appl. Mech. 64, 383–386 (1997)
C. Thornton, Z. Ning, A theoretical model for stick/bounce behaviour of adhesive elastic-plastic
spheres. Powder Technol. 99, 154–162 (1998)
C. Thornton, K.K. Yin, Impact of elastic spheres with and without adhesion. Powder Technol. 65,
153–166 (1991)
J. Tomas, Particle adhesion fundamentals and bulk powder consolidation. KONA Powder Part 18,
157–169 (2000)
J. Tomas, Assessment of mechanical properties of cohesive particulate solids – part 1: particle
contact constitutive model. Part. Sci. Technol. 19, 95–110 (2001)
J. Tomas, Fundamentals of cohesive powder consolidation and flow. Granul. Matter 6, 75–86
(2004a)
J. Tomas, Product design of cohesive powders – mechanical properties, compression and flow
behavior. Chem. Eng. Technol. 27, 605–618 (2004b)
C. Tsai, D. Pui, B. Liu, Elastic flattening and particle adhesion. Aerosol Sci. Technol. 13, 239–255
(1991)
P. Vernier, J.M. Delhaye, General two-phase flow equation applied to the thermohydrodynamics of
boiling nuclear reactors. Acta Tech. Belg. Energie Primaire 4, 3–43 (1968)
H.K. Versteeg, W. Malalasekera, An Introduction to Computational Fluid Dynamics – The Finite
Volume Method (Prentice Hall, Pearson Education Ltd., England, 1995)
L. Vu-Quoc, X. Zhang, An accurate and efficient tangential force–displacement model for elastic
frictional contact in particle-flow simulations. Mech. Mater. 31, 235–269 (1999)
L. Vu-Quoc, X. Zhang, O.R. Walton, A 3-D discrete-element method for dry granular flows of
ellipsoidal particles. Comput. Methods Appl. Mech. Eng. 187, 483–528 (2000)
O.R. Walton, R.L. Braun, Viscosity, Granular Temperature and Stress Calculations for Shearing
Assemblies of Inelastic, Frictional Disks. J. Rheology. 30, 949–980 (1986)
O.R. Walton, Numerical Simulation of Inelastic, Frictional Particle–Particle Interactions, Particulate
Two-Phase Flow (Ed. M. C. Roco), Butterworth–Heinemann, chap. 25, pp. 884–911 (1993)
G. Yadigaroglu, R.T. Lahey Jr., On the various forms of the conservation equations in two-phase
flow. Int. J. Multiphase Flow 2, 477–494 (1976)
W.H. Yang, The contact problem for viscoelastic bodies. Trans. ASME J. Appl. Mech. 33, 395–401
(1966)
G.H. Yeoh, C.P. Cheung, J.Y. Tu, Multiphase Flow Analysis Using Population Balance Modelling
(Butterworth-Heinemann, Elsevier, 2014)
Recent Advances in Modeling Gas-Liquid
Flows
Sherman C. P. Cheung, Lilunnahar Deju, and Sara Vahaji
Abstract
Gas-liquid flows are commonly encountered in industrial flow systems. In order
to adequately capture the distribution and its effect on the local hydrodynamics in
vertical gas-liquid flow, this chapter presents a numerical assessment on seven
combinations of six widely adopted bubble coalescence and bubble breakage
kernels. Three different coalescence kernels by Coulaloglou and Tavlarides
(Chem Eng Sci 32:1289–1297, 1977), Prince and Blanch (AIChE J
36:1485–1499, 1990), and Lehr et al. (AIChE J 48:2426–2443, 2002) have
been selected and combined with three different breakage kernels where each
kernel considers a different shape of the daughter size distribution of the bubbles
such as the U-shape proposed by Luo and Svendsen (AIChE J 42:1225–1233,
1996), the bell-shape proposed by Maritnez-Bazan (J Fluid Mech 401:157–182,
1999a; J Fluid Mech 401:157–182, 1999b), and the M-shape proposed by Wang
et al. (Chem Eng Sci 58:4629–4637, 2003). Numerical predictions of the void
fraction, bubble size distribution, and interfacial area concentration are compared
against the TOPFLOW experimental data (Lucas et al. 2010). Numerical results
reveal that the predicted two-phase flow structure is very sensitive to the choice of
coalescence and breakage kernels. Bases on the results, the model of Wang et al.
(Chem Eng Sci 58:4629–4637, 2003) is found to have a tendency to predict
higher breakage rate than the other two kernels. Moreover, although similar order
of magnitude of breakage rates are given by Luo and Svendsen (AIChE J
42:1225–1233, 1996) and Martinez-Bazan (J Fluid Mech 401:157–182, 1999a;
J Fluid Mech 401:157–182, 1999b), the bell-shape daughter size distribution by
Martinez-Bazan (J Fluid Mech 401:157–182, 1999a; J Fluid Mech 401:157–182,
1999b) is found to be favorable for equal breakage of bubbles, leading to
overprediction of larger bubbles.
S.C.P. Cheung (*) • L. Deju • S. Vahaji

School of Engineering, RMIT University, Melbourne, VIC, Australia
e-mail: chipok.cheung@rmit.edu.au

DOI 10.1007/978-981-4585-86-6_3-1
2 S.C.P. Cheung et al.
Keywords
Gas-liquid flows • Bubble coalescence • Bubble breakage
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Two-Fluid Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Population Balance Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Coalescence Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Breakage Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Experimental Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Numerical Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Impact of the Kernels on the Void Fraction Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Impact on the Bubble Size Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Validation of Interfacial Area Concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Consideration of Collision Frequency and Coalescence Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . 19
Consideration of Breakage Rate and Daughter Size Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Greek Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Super/Subscripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Introduction
Two-phase gas-liquid flows appear in many industrial applications, including chem-

ical, civil, nuclear, mineral, energy, food, pharmaceutical, and metallurgy.
Depending on the flow conditions, the two-phase flow pattern could evolve dynam-
ically and transit to different flow regimes. Such transition of flow regimes unfortu-
nately poses significant impact on the system performance or even incurs safety
issues in the operation. Aiming to improve the performance and assess the safety of a
particular system, it is therefore essential to grasp the phenomenological understand-
ing of bubble size or interfacial area and its dispersion behavior in the complex
two-phase flow structures. Subject to local flow conditions, previous experimental
studies have shown that bubbles within the bulk liquid flow could undergo signif-
icant coalescence and break-up processes leading to a wide spectrum of bubble size
distribution. Especially in the transition from the bubbly to slug flow regime,
rigorous bubble coalescence and breakup processes gradually transform the bubble
size distribution from a single-peaked to a bimodal profile. The bimodel profile
signifies the coexistence of the two types of bubbles within the system: spherical and
cap/distorted bubbles. Obviously, with different geometrical shape, interfacial area
concentration and its corresponding heat and mass transfer processes exhibit sub-
stantial difference between the two types of bubbles. More importantly, the shape of
bubbles also affects its transversal motion within the system. Based on the previous
studies, it has been widely accepted that small bubbles are subject to positive lift
Recent Advances in Modeling Gas-Liquid Flows 3
forces and have tendency traveling toward the wall region (Bothe et al. 2006;
Tomiyama 1998; Lucas et al. 2007). Cap and distorted bubbles generally travel in
opposite direction migrating towards the pipe center, which could eventually turn
Taylor bubbles via additional coalescence. Accurate knowledge of these hydrody-
namic variables throughout the entire system is thus paramount to the successful
design and operation of the system. Developing a robust and reliable mathematical
framework for modeling the aforementioned complex flow structure remains as a
great challenge at the moment.
Theoretically speaking, the evolution of bubble size distribution can be modeled
by the population balance equation (PBE) which is expressed in integro-differential
form with corresponding coalescence and breakup kernels. The development of
population balance model stems from the Boltzmann equation back in the eighteenth
century which governs the statistical distribution of particles in a state space.
Nevertheless, the generic population balance concept was first introduced in the
middle of the nineteenth century. Adopting the statistical mechanics framework,
Hulburt and Katz (1964) presented the population balance concept to solve particle
size variation due to nucleation, growth, and agglomeration processes. A series of
research developments were thereafter presented by Fredrickson et al. (1967),
Ramkrishna and Borwanker (1973), and Ramkrishna (1979, 1985) where the treat-
ment of population balance equations were successfully generalized with various
internal coordinates. The detailed descriptions of the mathematical and the generic
issues of population balance have been documented in the textbook by Ramakrishna
(2000). Although the concept of population balance has been formulated over many
decades, implementation of population balance modeling was only realized until
very recent years. The breakthrough was facilitated by the rapid development of
computational fluid dynamics (CFD) and in situ experimental measuring techniques.
The flourish of commercial CFD packages in the past decades has made a reliable
framework for solving the PBE. The field information obtained from the CFD
framework enabled solution algorithms to be developed within the internal coordi-
nates. The advancement in measuring bubble size or population balance variables
from experiment has also provided vital information for model calibrations and
validations. Among all existing methods, the class method (CM) is widely adopted,
in which the internal coordinate (e.g., particle length or volume) is discretized into a
finite series of bins. Encouraging results using CM in the form of MUSIG model for
bubbly flow simulations can be found in literatures (Frank et al. 2004; Olmos et al.
2001; Pohorecki et al. 2001; Bordel et al. 2006; Krepper et al. 2005, 2007;
Cheung et al. 2007, 2013). Although encouraging results have been obtained, all
of the aforementioned studies at best have demonstrated the feasibility and perfor-
mance of various numerical approaches in solving the population balance of bubbles
within the two-phase flow domain. Limited emphasis has been devoted towards the
understanding and modeling of the mechanisms of bubble coalescence and bubble
breakage under different flow conditions.
Chen et al. (2005) presented a numerical assessment on the performance of several
bubble coalescence and breakage kernels. Predictions by the coalescence models of
Chesters and Hoffman (1982), Prince and Blanch (1990), as well as breakage kernels
of Luo and Svendsen (1996) and Martinez-Bazan et al. (1999a, b) were validated
against three different sets of experimental data. Numerical results show that the choice
of bubble coalescence and bubble breakage kernels did not pose significant impact on
the final predictions. Nonetheless, model validations were mainly focused on the
comparison of velocity and turbulent kinetic energy. Limited comparison of the gas
phase distribution was presented. Sensitivity of the coalescence and breakage kernels
on the bubble size distribution and phase distribution was rather inconclusive.
In view of current developments of the state of the art, this chapter aims to further
exploit the recent advancements of population balance modeling for multiphase
flows from two aspects. First, it attempts to implement some of the widely adopted
bubble coalescence and breakage kernels and assess their applicability in predicting
the local hydrodynamic variables. Second, it seeks to investigate and understand the
physical mechanisms of each kernel and the effect of the interfacial forces affecting
the two-phase flow structure with the two-fluid model. A total of six coalescence and
breakage kernels are considered. The widely adopted breakage kernels by Luo and
Svendsen (1996), Martinez-Bazan et al. (1999a, b), and Wang et al. (2003) are
selected. For the coalescence kernels, the models by Coulaloglou and Tavlarides
(1977), Prince and Blanch (1990), and Lehr and Mewes (2001) are chosen. Numer-
ical results are assessed against the experimental data by Lucas et al. (2010) for
air-water flow in a tall vertical pipe with an inner diameter of 195.3 mm.
Mathematical Models
With reference to the formulation of the PBE, one should notice that the left-hand
side of the equation denotes the time and spatial variations of the PSD, which
depends on the external variables. By incorporating the PBE within CFD solver,
external variables can be obtained. In this section, governing equations of the two
fluid model and its associated model for handling interfacial momentum and mass
transfer are introduced.
Two-Fluid Model
The three-dimensional two-fluid model solves the ensemble-averaged of mass,

momentum, and energy transport equations governing each phase. Denoting the
liquid as the continuum phase (αl) and the vapor (i.e., bubbles) as disperse phase
(αg), these equations can be written as:
Continuity equation of liquid phase

@ρl αl ⇀
þ ∇ ρl αl u l ¼ Γlg (1)
@t
Continuity equation of vapor phase
@ρg αg f i
⇀
þ ∇ ρg αg f i u g ¼ Si f i Γlg (2)
@t
Momentum equation of liquid phase
⇀ h i
@ρl αl u l þ ∇ ρl αl u⇀l u⇀l ¼ αl ∇P þ αl ρl g⇀ þ∇ αl μe ∇u⇀l þ ∇u⇀l T þ Γlg u⇀g Γgl u⇀l þ Flg
l
@t
(3)
Momentum equation of vapor phase
⇀ h i
@ρg αg u g þ ∇ ρ αg u⇀g u⇀g ¼ αg ∇P þ αg ρ g⇀ þ∇ αg μe ∇u⇀g þ ∇u⇀g T þ Γgl u⇀l Γlg u⇀g þ Fgl
g g g
@t
(4)
Energy equation of liquid phase
@ρl αl H l ⇀
þ ∇ ρl αl u l Hl ¼ ∇ αl λel ð∇T l Þ þ Γgl H l Γlg H g (5)
@t
Energy equation of vapor phase
@ρg αg Hg h i
⇀
þ ∇ ρg αg u g H g ¼ ∇ αg λeg ∇T g þ Γgl H l Γlg H g (6)
@t
On the right-hand side of Eq. 2, Si represents the additional source terms due to
coalescence and breakage. For isothermal bubbly turbulent pipe flows, it should be
noted that the mass transfer rate Γlg and Γgl are essentially zero. The total interfacial
force Flg appearing in Eq. 3 is formulated according to appropriate consideration of
different subforces affecting the interface between each phase. For the liquid phase,
the total interfacial force is given by:
Flg ¼ Fdrag
lg þ Flg þ Flg
lift lubrication
þ Fdispersion
lg (7)
The subforces appearing on the right hand side of Eq. 7 are: drag force, lift force,
wall lubrication force, and turbulent dispersion force. More detailed descriptions of
these subforces can be found in Anglart and Nylund (1996). Note that for the gas
phase, Fgl = Flg.
The lift coefficient CL has been correlated as a function of the Eotvos number, Eo (59);
it allows the lift coefficient to be positive or negative depending on the bubble size. CL
can be expressed as:
8
< min½0:288tanhð0:121 Reb Þ, f ðEod Þ Eo < 0:4
CL ¼ f ðEod Þ ¼ 0:00105Eo3d 0:0159Eo2d 0:0204Eod þ 0:474 4 Eo 10
:
0:29 Eo > 10
(8)
gðρl ρg ÞD2H
where the modified Eotvos number, Eod ¼ ; DH corresponds to maximum
σ
1=3
horizontal dimension given by Wellek et al. (1966) as DH ¼ Ds 1 þ 0:163 Eo0:757 .
Turbulence Modeling for Two-Fluid Model

In handling bubble induced turbulent flow, unlike single phase fluid flow problem,
no standard turbulence model is tailored for multiphase flow. For simplicity, the
standard k-e model has been employed with encouraging results in early studies
(Davidson 1990; Schwarz and Turner 1988). Nonetheless, based on our previous
study (Cheung et al. 2012), the Menter’s (1994) k-ω based Shear Stress Transport
(SST) model were found superior to the standard k-e model. Similar observations
have been also reported by Frank et al. (2004). Based on their bubbly flow validation
study, they discovered that standard k-e model predicted an unrealistically high gas
void fraction peak close to wall. Interestingly, they also found that the two turbulence
models behaved very similar by reducing the inlet gas void fraction to a negligible
value. This could be attributed to a more realistic prediction of turbulent dissipation
close to wall provided by the k-ω formulation. It revealed that further development
should be focused on multiphase flow turbulence modeling in order to better
understand or improve the existing models.
The SST model is a hybrid version of the k-e and k-ω models with a specific
blending function. Instead of using empirical wall function to bridge the wall and the
far-away turbulent flow, it solves the two turbulence scalars (i.e., k and ω) explicitly
down to the wall boundary. The ensemble-averaged transport equations of the SST
model are given as:

@ρl αl kl ⇀ μ t, l
þ ∇ ρl αl u l kl ¼ ∇ αl μl þ ∇kl þ αl Pk, l ρl β0 kl ωl
@t σ k3
@ρl αl ωl ⇀ μ t, l
þ ∇ ρl αl u l ωl ¼ ∇ αl μl þ ∇ωl
@t σ ω3
1 @kl @ωl ωl
2ρ1 αl ð1 F1 Þ þ α l γ 3 P k , l ρ l β 3 ωl 2
σ ω2 ωl @xj @xj kl
(9)
where σ k3, σ ω3, γ 3, and β3 are the model constants which are evaluated based on
the blending function F1. The shear induced turbulent viscosity μts , l is given by:
ρa1 kl pffiffiffiffiffiffiffiffiffiffiffiffi
μts, l ¼ , S ¼ 2Sij Sij , (10)
maxða1 ωl , SF2 Þ
The success of SST model hinges on the use of blending functions of F1 and F2
which govern the crossover point between the k-ω and k-e models. The blending
functions are given by:
" pffiffiffiffi ! #
4 kl 500μl 4ρl kl
F1 ¼ tanh Φ1 , Φ1 ¼ min max , ,
0:09ωl dn ρl ωl d2n Dþω σ ω2 d n
2
pffiffiffiffi ! (11)
kl 500μl
F2 ¼ tanh Φ22 , Φ2 ¼ max , ,
0:09ωl dn ρl ωl d2n
Here, default values of model constants were adopted. More detailed descriptions
of these model constants can be found in Menter (1994). In addition, to account the
effect of bubbles on liquid turbulence, the Sato’s bubble-induced turbulent viscosity
model was also employed (Sato et al. 1981). The turbulent viscosity of liquid phase
is therefore given by:
μt, l ¼ μts, l þ μtd, l (12)
and the particle induced turbulence can be expressed as:
⇀ ⇀
μtd, l ¼ Cμp ρl αg DS U g U l (13)
For the gas phase, dispersed phase zero equation model was adopted, and the
turbulent viscosity of gas phase can be obtained as:
ρg μt, l
μt, g ¼ (14)
ρl σ g
where σ g is the turbulent Prandtl number of the gas phase.
Population Balance Equation
The foundation development of the PBE stems from the consideration of the
Boltzman equation. Such equation is generally expressed in an integrodifferential
form describing the particle size distribution (PSD) as follow:
@f ðξ, r, tÞ
þ ∇ uðξ, r, tÞ f ξ, r, t
@tð ð1
1 ξ 0 0 0 0 0 (15)
¼ aðξ ξ , ξ Þf ðξ ξ , tÞf ðξ , tÞdξ f ðξ, tÞ aðξ ξ0 , ξ0 Þf ðξ0 , tÞdξ0
Ð21 0 0 0
þ ξ γ ðξ Þbðξ0 Þpðξ=ξ0 Þf ðξ0 , tÞds bðξÞf ðξ, tÞ
where f(ξ, r, t) is the particle size distribution dependent on the internal space
vector ξ, whose components could be characteristics dimensions, surface area,
volume, and so on. r and t are the external variables representing the spatial position
vector and physical time in external coordinate, respectively. u(ξ, r, t) is velocity
vector in external space. On the RHS, the first and second terms denote birth and
death rate of particle of space vector ξ due to merging processes, such as: coales-
cence processes; the third and fourth terms account for the birth and death rate
caused by the breakage processes, respectively. a(ξ, ξ0) is the coalescence rate
between bubbles of size ξ and ξ0. Conversely, b(ξ) is the breakage rate of bubbles
of size ξ. γ(ξ0) is the number of fragments/daughter bubbles generated from the
breakage. Two-phase gas-liquid flows of a bubble of size ξ0 and p(ξ/ξ0) represents the
probability density function for a bubble of size ξ to be generated by breakage of a
bubble of size ξ0.
Driven by practical interest, numerical approaches have been developed to solve
the PBEs. The most common methods are Monte Carlo methods, method of
moments, and class methods. Because of their relevance in CFD applications, the
widely adopted MUltiple-Size-Group (MUSIG) model (Lo 1996) has been used as
the numerical technique for solving PBE in the present study.
Coalescence Kernels
For coalescence between two colliding bubbles of i and j group, the coalescence
efficiency a(Mi, Mj) could be calculated as a product of collision frequency,
h(Mi, Mj), and coalescence efficiency λ(Mi, Mj).

Mi , Mj ¼ h Mi , Mj λ Mi , Mj (16)
In Liao and Lucas (2010) paper, they have listed a variety of mechanisms for the
collision frequency among the bubbles in the turbulent flow:
• Random motion induced collision due to fluctuating turbulent eddies.

• Collision due to velocity gradient.
• Collision due to capture in turbulent eddies.
• Collision due to buoyancy.
• Collision due to wake interaction.
Coulaloglou and Tavlarides Coalescence Kernel

Coulaloglou and Tavlarides (1997) developed their model based on the consider-
ation of turbulent random motion induced collisions as primary source of bubble
coalescence and, the model is formulated according to the film drainage model for
deformable particle with immobile surface. The total coalescence rate is given by:
" #
2 2=3 2=
1=2 1
=3 μρϵ di dj 4
a Mi , Mj ¼ C2 d i þ dj di þ dj 3
e exp CC&T l 2l
σ di þ dj
(17)
For the current set of experimental data, the coalescence efficiency parameter
(CC & T) was selected as 0:183 1010 cm2 .
Prince and Blanch Coalescence Kernel

Turbulent random collision is considered for the bubble coalescence by Prince and
Blanch (1990). Based on their model, coalescence process in turbulent flows has
been described in three steps. Firstly, the bubbles trap small amount of liquid
between them. Then the liquid drains out until the liquid film thickness equals to
the critical thickness. The two bubbles are then finally ruptured and merged into one
bigger bubble. Similar to particle kinetic theory, the coalescence rate of bubbles can
be related to the collision frequency of bubbles and the probability of successful
coalescence. The derivation of the kernel can be found out from the paper by Prince
and Blanch (1990). The total coalescence rate by Prince and Blanch is calculated as
following:

2 2= 2=
1=2 1
= tij
a Mi , Mj ¼ C3 di þ d j di 3 þ dj 3 e 3 exp (18)
τij
Here τij is the contact time for two bubbles, and tij is the time required for two
bubbles to coalesce having diameters di and dj.
Lehr et al. Coalescence Kernel

Lehr et al. (2002) proposed the coalescence frequency based on the critical approach
velocity model. An experimental investigation has been conducted to determine the
criterion of collision between two bubbles resulting in coalescence or bouncing.
They have found that the colliding bubbles might result in coalescence or bounce
back depending on the relative approach velocity perpendicular to the surface of
contact. They found that the critical approach velocity (ucritical) for distilled water and
air is of 0.08 m/s. They have also defined the critical velocity as the maximum
velocity of bubbles resulting in coalescence, which has no dependency on the size of
the bubbles. Coalescence will only occur when the relative approach velocity of
bubbles perpendicular to contact surface is lower than the critical approach velocity.
The total coalescence rate is given as following:
2 !2 3
2 2=3 1=2 1 α
1=
5min ucritical , 1
3
a Mi , Mj ¼ C4 di þ d j di þ dj 3 e 3 exp4 max
2= =
1
α1=3 u0
(19)
Breakage Kernels
For breakage of bubble from j group to i group, the partial breakage frequency
γ(Mi, Mj) is a function of total breakage frequency, γ(Mi), and the daughter size
distribution, β(Mi, Mj).

γ Mi , Mj
β Mi , Mj ¼ (20)
γ ðM i Þ
According to Liao and Lucas (2009) paper, the variety of mechanisms for the
breakage of particles in the turbulent flow could be categorized as:
• Turbulent fluctuation and collision

• Viscous shear force
• Interfacial instability
• Shearing off process
Luo and Svendsen Breakage Kernel

The bubble breakup rate by Luo and Svendsen (1996) is developed with the
assumption of binary breakup under isotropic turbulence influence. Breakage
event is determined by the energy level of arriving eddy with smaller or equal length
scale compared to the bubble diameter to induce the oscillation. The daughter size
distribution is accounted using a stochastic breakup volume fraction fBV. Denoting
2=3
the increase coefficient of surface area as cf = [f BV +(1fBV)2/3–1], the breakage rate
in terms of mass can be obtained as:
1= ð1 !
e 3
ð 1 þ ξ Þ2 12cf σ
γ Mi , Mj ¼ 0:923 1 αg n exp dξ (21)
dj ξmin ξ11=3 βρe2=3 d j 5=3 ξ11=3
where ξ = λ/dj is the size ratio between an eddy and a bubble in the inertial
subrange and consequently ξmin = λmin/dj and β = 2.0 are defined based on the
consideration of bubbles breakup in turbulent dispersion systems. For binary break-
age, the value of the dimensionless variable describing breakage volume fraction
should be between 0 and 1 (0<fBV<1). f BV ¼ 0:5 refers to equal breakage and f BV
¼ 0 or 1 would refer to no breakage.
The γ(Mi, Mj) represents the breakage rate of bubble with mass of Mi into fraction
of fBV and fBV + dfBV for a continuous fBV function. The total breakage rate of bubbles
can be obtained by integrating the partial breakage, γ(Mi, Mj), over the whole interval
of 0 to 1. Total breakage rate can be expressed as:
ð1
1
γ ðM i Þ ¼ r Mi , Mj df BV (22)
2 0
The advantage of this model is that it provides the partial breakage rate, γ(Mi, Mj),
directly. Then the daughter bubble size distribution for mother bubbles with size
fraction of fBV can be derived by normalizing the partial breakage rate, γ(Mi, Mj), by
the total breakage rate, γ(Mi)
!
Ð 1 ð1 þ ξÞ2 12Cf σ
2 ξmin 11=3 exp 5= 1=
dξ
r Mi , Mj ξ βρf e2=3 d i 3 ξ 3
βðf BV , 1Þ ¼ ¼ ! (23)
r ðM i Þ Ð 1 Ð 1 ð1 þ ξÞ2 12Cf σ
0 ξmin exp 5= 1=
dξdf BV
ξ11=3 βρf e2=3 di 3 ξ 3
Wang et al. Breakage Kernel

While Luo and Svendsen (1996) only considered the energy constraint, Wang et al.
(2003) extended the model by adding the capillary constraint to calculate the
breakage. According to this model, the capillary pressure must be overcome by the
dynamic pressure of the turbulent eddy leading towards a minimum breakage
fraction. On the other hand, for maximum breakage, the eddy kinetic energy should
be higher than the increase of the surface energy. The main advantage of this model
is having no adjustable parameter, and the daughter size distribution can be easily
obtained through the ratio between the partial breakage frequency and the total
frequency:
ð di ð 1
1=
γ Mi , Mj ¼ 0:923ð1 αd Þnϵ 3
λmin 0
1 ðλ þ d Þ2
Pe ð e ð λ Þ Þ deðλÞdλ (24)
f BV , max f BV , min 11=
λ 3
where Pe(e(λ)) is the energy distribution of eddies with size λ has been calculated
similarly as Luo and Svendsen (1996). The daughter bubble size distribution is
expressed as:
ð1
Ð di ðλ þ d Þ2 1 1 eðλÞ
λmin exp deðλÞdλ
λ
11=
3 f
0 BV , max f BV , min e ðλ Þ e ðλÞ
βðf BV ,1Þ ¼ ð (25)
Ð 1 Ð di ðλ þ d Þ2 1 1 1 eðλÞ
0 λmin exp deðλÞdλ df BV
0 f BV , max f BV , min e ðλÞ e ðλÞ
11=
λ 3
As the breakage of bubbles will occur only when the dynamic pressure of eddy
exceeds the capillary pressure of bubbles, the daughter size distribution function
tends to zero as the breakage fraction, fBV, approaches zero. This is due to the fact that
a smaller bubble with radius of curvature tends to zero characterized by higher
capillary pressure, and therefore it is unlikely to break a very small bubble. Wang
et al. (2003) have also concluded that a local minimum probability occurs for the
equal breakage as it requires more energy than a binary unequal breakage. The
daughter size distribution is influenced by mother bubble size and the energy
dissipation rate. This model has been formulated based on the premise that the
distribution should not have any singularity point or any uncertain parameter.
Martinez-Bazan et al. Breakage Kernel

Martinez-Bazan et al. (1999a, b) developed their model based on the assumption that
a bubble will break if the turbulent kinetic energy in the continuous phase is larger
than a critical value. They postulated that the rate of breakage processes is inversely
proportional to the difference between the deformation force, τt(di), per unit surface
produced by the turbulent stress and the surface restoring pressure of the bubble,
τs(di). When τt(di) > τs(di) the bubble will deform and eventually break. The bubble
breakage frequency is a function of bubble sizes and the turbulent dissipation rate
and can be given by:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
σ
CMB ðedi Þ2=3 12
ρf d i
γ ðM i Þ ¼ K g (26)
di
Here the value of Kg is equal to 0.25 and CMB = 8.2. The daughter bubble size
distribution is expressed as:
1 2=3 σ 1 1=3 2=3 σ

ρCMB ed j 6 ρCMB e d i 3 d j 3 6
2 di 2 di
βðf BV , 1Þ ¼ (27)
Ð1 1 2=3 σ 1 1=3 2=3 σ
0 ρCMB ed j 6 ρCMB e d i 3 d j 3 6 df BV
2 di 2 di
Experimental Details
The TOPFLOW experimental dataset is selected for model validation in this study.
The experiment data were reported by Lucas et al. (2010) and the experimental
arrangement are briefly described below. Figure 1 illustrates the schematic
representations of the TOPFLOW experiment. In this test facility, a large size vertical
cylindrical pipe with height 9000 mm and inner diameter of 195.3 mm was adopted.
Water with a constant temperature of 30 C, maintained by a heat exchange installed
in the water reservoir, was circulated from the bottom to the top. The maximum
superficial velocities for gas and liquid phase were jg = 14 m/s and jf = 4 m/s,
respectively. A variable gas injection system was constructed by equipping with gas
injection units at 18 different axial positions from Z/D = 1.1–9.9. Three levels of air
chambers were installed at each injection unit. The upper and the lower chambers
have 72 annular distributed orifices of 1 mm diameter for small bubble injection,
while the central chamber has 32 annularly distributed orifices of 4 mm diameter for
large bubble injection. A fixed wire-mesh sensor was implemented at the top of the
pipe where instantaneous information of gas void fraction as well as bubble size
distribution was measured. To balance the bubble coalescence and breakage, dimen-
sionless multiplying factors are selected for the mass transfer rates of breakage and
coalescence, as FB = 0.25 and FC = 0.05, respectively (Following the study carried
out by Krepper et al. (2008)). It is important to emphasize that such dimensionless
factors were calibrated for the Prince and Blanch coalescence and Luo and Svendsen
breakage kernels. One could also obtained a more “accurate” predictions compared
with the experimental data by adjusting these factor with other coalescence and
breakage kernels. Nonetheless, it will deficit the predictive nature of the numerical
model.
Numerical Details
Numerical simulations were carried out using the commercial computational fluid
dynamics package ANSYS-CFX12.1. Corresponding source and sink terms in
relation to the bubble coalescence and breakage rate were incorporated using user
Fortran subroutine. To reduce the computational time and resource, the flow was
considered as axisymmetric and the computational domain was constructed in a 60
radial pipe sector. Both vertical sides of the computational domain were specified as
symmetry boundary.
Two flow conditions were selected, namely, T107 and T118. The liquid superfi-
cial velocity ( jl) was 1.017 m/s for both cases and gas velocity ( jg) was 0.140 and
0.219 4 m/s for 107 and 118, respectively. Details of the boundary conditions are
summarized in Table 1. The mesh sensitivity study has revealed that even finer
computational meshes (96,000 elements) did not provide appreciably different
results (only 3% of difference) compared to meshes consisting of 48,000 elements.
The computational meshes of 48,000 elements were therefore adopted in this study
and convergence criterion of 1.0 104 RMS residual value was applied to
terminate the numerical simulation. The coalescence and breakage kernels are
applied in different combinations and denoted as Kernels 1, Kernel 2,
Kernels 3, Kernels 4, Kernels 5, Kernels 6, and Kernels 7 as tabulated in Table 2.
Fixed wire-mesh
D = 195.3 mm measuring sensor
Gas injection unit
locations
Small and narrow
range bubble size
Z/D = 1.7
Breakup dominant
9000 mm
Z/D = 22.6
Large and wide

Z/D = 39.9 range bubble size
Water
Fig. 1 Schematic of the TOPFLOW experimental arrangement

Table 1 Boundary TOPFLOW Experiment

conditions of the selected
Case T107 Case T118
flow conditions
[hjli|Z/D = 0] (m/s) 1.017 1.017
[hjgi|Z/D = 0] (m/s) 0.140 0.2194
[αg|Z/D = 0] (%) [12.1] [17.72]
[DS|z/D = 0.0] (mm) [20.18] [23.28]
Table 2 List of different kernel combinations adopted in this study

Kernels 1 C: Prince and Blanch (1990) B: Luo and Svendsen (1996)
Kernels 2 C: Coulaloglou and Tavlarides (1977) B: Luo and Svendsen (1996)
Kernels 3 C: Lehr et al. (2002) B: Luo and Svendsen (1996)
Kernels 4 C: Prince and Blanch (1990) B: Wang et al. (2003)
Kernels 5 C: Coulaloglou and Tavlarides (1977) B: Wang et al. (2003)
Kernels 6 C: Lehr et al. (2002) B: Wang et al. (2003)
Kernels 7 C: Prince and Blanch (1990) B: Martinez-Bazan (1999a, 1999b)
Results and Discussions
Impact of the Kernels on the Void Fraction Distribution
Figure 2 presents the comparison of the measured and predicted radial gas void
fraction distribution for the four selected Kernels (Lucas et al. 2007; Olmos et al.
2001; Wang et al. 2003; Krepper et al. 2008) under two different flow conditions. It
can be observed from the figure that consistent predictions were obtained from the
four Kernels at the upstream location near the gas injection locations (i.e.,
L/D = 1.7). The agreement among Kernels could be because of the considerably
short distance after the gas injection for the two-phase flow structure to become fully
developed and the insufficient time for the bubble coalescence and breakage pro-
cesses. At the downstream location (i.e., L/D = 39.9), the predicted radial void
fraction profiles by the four kernels exhibit substantial difference
In general, Kernel 1 and 7 have successfully captured the core peak of the gas
void fraction profiles. In contrast, for the Kernels 4 and 6 where breakage kernel
proposed by Wang et al. was adopted, the near wall gas void fractions were
overpredicted thereby showing a wall peaking characteristic. By comparing the
predictions from the Kernels 1 and 4, it clearly shows that the breakage kernel by
Wang et al. tends to predict a much higher breakage rate than the kernel by Luo and
Svendsen. With the higher breakage rate, more spherical small bubbles were pre-
dicted by the kernel causing a positive lift coefficient and leading to a wall peak void
fraction profile. Meanwhile, it is noted that the difference of predictions between
Kernels 4 and 6 is insignificant. This exemplifies that coalescence kernels by Prince
and Blanch and Lehr et al. yield reasonably similar predictions. The radial void
a 0.4 b 0.4
kernels 1 T107 kernels 1 T107
kernels 4 L/D =1.7 kernels 4 L/D =39.9
kernels 6 kernels 6
0.3 0.3
kernels 7 kernels 7
Experiment Experiment
Void Fraction [ ]
Void Fraction [ ]
0.2 0.2
0.1 0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Radial Distance [ ] Radial Distance [ ]
c 0.4 d 0.4
kernels 4 L/D =1.7 kernels 4 L/D =39.9
kernels 6 kernels 6
0.3 0.3
kernels 7 kernels 7
Void Fraction [ ]
Void Fraction [ ]
0.2 0.2
0.1 0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Fig. 2 Measured and predicted gas volume fraction distribution for the selected kernel
combinations
fraction profiles are captured marginally better by Kernels 7 such as evidenced by the
steep decrease of the void fraction near the pipe wall.
By comparing the predictions of Kernels 1, 2, and 3 at the downstream location
(i.e., L/D = 39.9), a closer examination of the influence of the three coalescence
kernels is shown in Fig. 3. With the same breakage kernel by Luo and Svensden, it
can be observed that the coalescence kernel by Lehr et al. tends to marginally
overpredict the gas void fraction near the wall region. Consequently, void fraction
at the core of the pipe is slightly underestimated. Similar predictions were given by
the Kernels 1 and 2 while Kernel 2 shows considerably better agreement with the
measurements at the core. Overall, both of the coalescence kernels by Coulaloglou
and Tavlarides and Prince and Blanch tend to predict similar coalescence rate. On the
other hand, void fraction at the wall region is overpredicted by Kernel 5, which is
due to the fact that the breakage kernel by Wang et al. tend to predict higher breakage
rate compared to other breakage kernels.
a 0.4 b 0.4
kernels 2 L/D =1.7 kernels 2 L/D= 39.9
kernels 3 kernels 3
0.3 0.3
kernels 5 kernels 5
Void Fraction [ ]
Void Fraction [ ]
0.2 0.2
0.1 0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
0.4
c d 0.4
kernels 2 L/D=1.7 kernels 2 L/D= 39.9
kernels 3 kernels 3
0.3 0.3
kernels 5 kernels 5
Void Fraction [ ]
Void Fraction [ ]
0.2 0.2
0.1 0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Fig. 3 Measured and predicted gas volume fraction distribution for coalesence kernel
combinations
Impact on the Bubble Size Distribution
Since most of the interfacial heat and mass transfer processes are governed by the
interfacial area concentration, it is therefore important to assess the performance of
kernels in capturing the evolution of bubble size distribution. Figure 4 depicts the
measured and predicted cross-sectional averaged bubble size distribution of the
selected Kernels at the two axial locations. As shown in the figure, due to the
injection method, the bubble size distribution appears as a bimodal distribution at
the upstream location (i.e., L/D = 1.7). The bimodal characteristic has been reason-
ably captured by all Kernels; while the Kernel 7 appears to give the best agreement
with the measurements. Surprisingly, the predicted upstream bubble size distribu-
tions (i.e., L/D = 1.7) by the Kernels 4 and 6 are practically identical even though
different coalescence kernels were adopted. The maximum difference between pre-
dictions of the two kernels is only 4%. This clearly demonstrates that the predictions
by the two kernels are dominated by the breakage rate. As a result, predictions from
the Kernel 1 and 4 are considerably different from the one from Kernel 7. At the
downstream location (i.e., L/D = 39.9), one could clearly observe from the mea-
surements that the bubble size distribution transits from bimodal to single peak
a 1.0
b4
Bubble Size Distribution [%/mm]

0.8 kernels 6 kernels 6
3
kernels 7 kernels 7
0.6
0.4
1
0.2
0.0 0
0 10 20 30 40 50 0 10 20 30 40 50
Bubble Diameter [mm] Bubble Diameter [mm]
c 1.2
d 4
kernels 1 T118 kernels 1
kernels 4 T118
L/D= 1.7 kernels 4

1.0
kernels 6 L/D= 39.9 kernels 6
kernels 7 3
kernels 7
0.8 Experiment Experiment
0.6 2
0.4
1
0.2
0.0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Fig. 4 Measured and predicted bubble size distribution for the selected kernel combinations
profile due to bubble breakage processes. Such transition characteristic has been
reasonably captured by most of the Kernels.
It should be also noticed that the bubble size distribution is substantially over-
predicted by the Kernels 7. This could be due to the bell-shaped daughter size
distribution assumption adopted by the Martinez-Bazan breakage kernel. One should
also notice that the Kernels 4 and 6 marginally overpredict the amount of small
bubbles (i.e., 0–3 mm bubbles) in comparison to Kernels 1. This further ascertain
that breakage kernel by Wang et al. tends to predict higher breakage rate in
comparison to the kernel by Luo and Svendsen. Figure 5 shows the predicted bubble
size distribution based on different coalescence kernels. As discussed above, the
figure shows that the predictions are dominant by the breakage kernel. Although
different coalescence kernels have been adopted, insignificant differences were
found between Kernel 1, 2, and 3.
Validation of Interfacial Area Concentration
Figure 6 shows the comparison of the measured and predicted radial interfacial area
concentration profiles at the two axial locations. Agreed with the findings from the
a 1.0
b 4
kernels 2 L/D=1.7 kernels 2 L/D=39.9

0.8 kernels 3 kernels 3
kernels 5 3
kernels 5
0.6
2
0.4
1
0.2
0.0 0
0 10 20 30 40 50 0 10 20 30 40 50
1.2
c kernels 1 T108
d 4
T118 kernels 1
kernels 2
L/D=1.7

1.0 kernels 2
kernels 3 L/D=39.9
3 kernels 3
kernels 5 kernels 5
0.8 Experiment Experiment
0.6 2
0.4
1
0.2
0.0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Fig. 5 Measured and predicted bubble size distribution for the selected coalescence kernel
combinations
void fraction profiles, the breakage kernel by Wang et al. (i.e., Kernels 4, 5, and 6)
tends to overpredict the interfacial area concentration at both pipe center and near
wall region. On the other hand, predictions provided by the Luo and Svensden and
Martinez-Bazan breakage kernels (i.e., Kernels 1, 2, 3, and 7) are in satisfactory
agreement with the experimental data.
Consideration of Collision Frequency and Coalescence Efficiency
Theoretically speaking, the bubble coalescence rate is the product of the bubble
collision frequency and bubble coalescence efficiency. For all selected coalescence
kernels in this study, the bubble collision frequency is derived based on the turbulent
fluctuation and random collision assumption. All the kernels assume that bubble
collision frequency is proportional to the collision cross-sectional area (i.e.,
π
2
4 di þ dj ). In other words, the bubble collision frequency increases with the
bubble size. Figure 7 shows the bubble collision frequency for the collision of two
identical size bubbles at the pipe center and near the pipe wall evaluated according to
the three different coalescence kernels. The eddy dissipation rates adopted in the
a 200 b 300
Interfacial area concentration [1/m3]

kernels 2 L/D= 1.7 250 kernels 2 L/D =39.9
kernels 3 kernels 3
150
kernels 4 kernels 4
200 kernels 5
kernels 5
kernels 6 kernels 6
kernels 7 kernels 7
100 150
100
50
50
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
c 300
d 350
T108 kernels 1 T108

kernels 1

300 kernels 2
250 kernels 2 L/D= 1.7 L/D=39.9
kernels 3
kernels 3
250 kernels 4
kernels 4 kernels 5
200
kernels 5
kernels 6
kernels 6 200
kernels 7
150 kernels 7
Experiment
Experiment 150
100
100
50
50
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Fig. 6 Measured and predicted interfacial area concentration along the radial direction at the axial
location L/D = 1.7 and 39.9
figure are extracted directly from the liquid phase turbulence model at the pipe center
and near the pipe wall (i.e., first grid cell away from the wall). [Note that,
ε = 0.006829 m2/s3 at the pipe center and ε = 0.186918 m2/s3 near the wall region].
With the lower eddy dissipation rate at the pipe center, all coalescence kernels
predict higher collision frequency at the near wall region. For all the selected
coalescence kernels, the bubble collision frequency is evaluated from the approach
velocity based on the eddy with the same size of bubble, which is proportional to the
eddy dissipation rate. Therefore, the bubble collision frequency increases with
higher turbulence level near the pipe wall region.
Meanwhile, the figure also exemplifies the reason why the predictions of void
fraction and bubble size distributions are surprisingly insensitive to the adopted
coalescence kernels for the selected flow conditions. As depicted, the bubble colli-
sion frequency is evaluated based on the two main factors: bubble size and eddy
dissipation rate. As the eddy dissipation rate becomes insignificant at the pipe center,
the corresponding bubble collision frequency starts to diminish. On the other hand,
with the higher eddy dissipation rate near the wall region, bubble sizes are mostly
small driven by the lateral lift force for small spherical bubbles. As a result, pre-
dictions of all coalescence kernel at this region are very similar. More importantly, it
must be emphasized that the collision frequency only considers collision of spherical
0.0035
a kernels 1 Center of the pipe
0.0035
b kernels 1 Near the wall
0.0030 kernels 2 T107 kernels 2 T107
0.0030
Collision frequency [m3/s]
kernels 3 L/D = 1.7 kernels 3 L/D=39.9
Collision frequency [m3/s]

0.0025 0.0025
0.0020 0.0020
0.0015 0.0015
0.0010 0.0010
0.0005 0.0005
0.0000 0.0000
0 15 30 45 0 15 30 45
Bubble diameter [mm] Bubble diameter [mm]
Fig. 7 Predicted frequency of collision for equal size bubbles at the center of the pipe
(ε = 0.006829 m2/s3) and near the wall region (ε = 0.186918 m2/s3)
bubbles where other bubble shaper (i.e., cap or Taylor bubble) could become
dominant, leading to substantial error in the simulation.
Figure 8 shows the corresponding bubble coalescence efficiency for the collision
of two identical size bubbles as shown in Fig. 7. Overall, all kernels predict higher
bubble coalescence efficiencies for small bubbles. The predictions agree with exper-
imental findings of Doubliez (1991) where they concluded that coalescence mech-
anism favors more gentle collisions between small bubbles. Moreover, all
coalescence kernels predict higher coalescence efficiency at the pipe center than
the near pipe wall region. This is related to the assumption taken by the kernel in
relation to the eddy dissipation rate.
Figure 9 shows the influence of eddy dissipation rate on the bubble contact time
and relative characteristic velocity of eddy at the two selected near wall locations.
For the coalescence kernels by Coulaloglou and Tavlarides and Prince and Blanch,
as discussed, the bubble coalescence efficiency is evaluated based on the function of
the required film thinning time and bubble contact time. For a successful coales-
cence, the bubble contact time needs to exceed the required coalescence time.
Unfortunately, increasing the eddy dissipation rate reduces the bubble contact time
which thereby results in lower bubble coalescence efficiency. For the coalescence
kernel by Lehr et al., the coalescence efficiency is assumed as a function of the
characteristic velocity of eddies and the critical velocity. The higher eddy dissipation
rate incurs higher characteristic velocity which sequentially lower bubble coales-
cence efficiency.
Consideration of Breakage Rate and Daughter Size Distribution
Figure 10 depicts the predicted birth rate of different size bubbles due to breakage of
mother bubble at the pipe center and near the pipe wall being evaluated by the three
selected breakage kernels. As all of the breakage kernels that assume the successful
breakage requires turbulent kinetic energy of the colliding eddy, where turbulent
2.0 2.0
1.8
a kernels 1 Center of the pipe b kernels 1 Near the wall
1.8
Coalescence efficiency [m3/s]
Coalescence efficiency [m3/s]

1.6 kernels 3 1.6
L/D= 39.9 kernels 3 L/D=39.9
1.4 1.4
1.2 1.2
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
0 10 20 30 40 50 0 10 20 30 40 50
Fig. 8 Coalescence efficiency of equal size of bubbles for selected kernel combination at the center
of the pipe (ε = 0.006829 m2/s3) and near the wall region (ε = 0.186918 m2/s3)
a 0.10
L/D=1.7, Near the wall (ε=0.13)
0.08
Bubble contact time [s]
T107
L/D=39.9 Kernels 1
0.06
ε = 0.18 m2/s3
0.04
0.02
0.00
0 15 30 45
Bubble diameter [mm]
b 1.0
Relative characteristic velocity

0.8
T107
Kernels 3
of eddy [m/s]
0.6
0.4
L/D=1.7
0.2
ε = 0.13 m2/s3
0.0
0 15 30 45
Bubble diameter [mm]
Fig. 9 Effect of bubble diameter and turbulent eddy dissipation (ε) on bubble contact time and
relative characteristic velocity of eddy at near wall region for L/D = 1.7 and 39.9
a 2.0 b 20
kernels 1 Center of the pipe 18 kernels 1 Near the wall
1.6
16
kernels 7 L/D=39.9 kernels 7 L/D =39.9
Breakage Frequency [1/s]
Breakage Frequency [1/s]

14
1.2 12
10
0.8
8
6
0.4
4
2
0.0
0
0 10 20 30 40 50 0 10 20 30 40 50
Fig. 10 Birth rate of daughter bubble due to breakage mother bubble for selected kernel combi-
nation at the center of the pipe (ε = 0.006829 m2/s3) and near the wall region (ε = 0.186918 m2/s3)
dissipation rate is at lowest value, the breakage rates at the pipe center vanish almost
to zero. Due to the high void fraction of larger bubbles at the location (see also in
Fig. 2), the breakage kernel by Martinez-Bazan predicts higher breakage rate for
larger bubbles. Nevertheless, the breakage rate is still insignificant. Moreover, albeit
higher turbulent dissipation rate at the near wall region, the turbulent kinetic energy
remains insufficient to support the breakage process. Therefore, the breakage rates
for the kernels by Luo and Svendsen and Martinez-Bazan are still close to zero
In contrast, the breakage kernel by Wang et al. predicts higher breakage rates as
compared to the other two breakage kernels. From the formulation of the kernel, the
main difference between Wang et al. and Luo and Svendsen is the introduction of
bubble breakage efficiency expressed in terms of the maximum and minimum
breakage fraction (i.e., fBV , max and fBV , min) which is determined by the energy
and force constraints, respectively. The breakage efficiency could be considered as a
multiplier in the model which magnifies the breakage rate 5 to 10 times higher than
the one by Luo and Svendsen. For some given situations (i.e., fBV , max fBV , min), the
breakage rate can be scaled up to 100 times. This also provides explanation for the
high breakage rate predictions which eventually leads to more small bubbles and a
wall peaking void fraction profiles in the present study.
Figure 11 presents the daughter bubble size distribution at the pipe center and
near the pipe wall predicted by the three selected breakage kernels. As shown in the
figure, the breakage kernel by Luo and Svendsen presents an U-shape characteristic
for the daughter bubble size distribution. The characteristic shape has a minimum
probability located at equal size breakage and the maximum values if the daughter
bubble size approach to 0 or 1. It is worthwhile to note that the distribution becomes
flatter near the pipe wall as the higher turbulent dissipation rate is available to
support breakage processes. On the other hand, the breakage kernel by Wang et al.
exhibit an M-shape daughter size distribution at both locations. The main difference
between two kernels is the breakage probability which vanishes to 0 for small
bubbles considering the capillary pressure becomes extremely high when the radius
of curvature tends to zero. To certain extent, one could conclude that the U-shape
a b
10 10
Dimensionless Daughter Size Distribution
Dimensionless Daughter Size Distribution

kernels 1 Center of the pipe kernels 1 Near the wall
8 8
6 6
4 4
2 2
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Breakup volume fraction [ ] Breakup volume fraction [ ]
Fig. 11 Daughter size distribution for selected kernel combination at the center of the pipe
(ε = 0.006829 m2/s3) and near the wall region (ε = 0.186918 m2/s3)
distribution with maximum probability for low breakage fraction is unrealistic

because the capillary pressure becomes very high for small bubbles which makes
the probability of breakage to be extremely low. Following the same argument, the
M-shape distribution appears to be more practical. From the numerical viewpoint,
both the U-shape and M-shape distributions exhibit very similar characteristic
throughout the simulations. Therefore, the higher breakage rate predicted by the
breakage kernel by Wang et al. is mainly caused by the contribution of the bubble
breakage efficiency.
Nevertheless, the breakage kernel by Martinez-Bazan shows a bell-shaped dis-
tribution where maximum probability located at the equal size breakage and the
minimum when daughter bubble size approaches to 0 or 1. This is also the main
reason by the kernel tends to overpredicted the bubble size distribution for large
bubble. With the bell-shaped distribution, most bubbles undergo even size breakage
leading to larger daughter bubble than the other breakage kernels. As a result, the
predicted bubble sizes are larger, even though the predicted breakage rates are of
similar order to the breakage kernel by Luo and Svendsen.
Conclusions
A numerical assessment on seven different combination of widely adopted bubble

coalescence and breakage kernels in capturing the evolution of bubble size distribu-
tion and its effect on the phase distribution and interfacial area concentration flow is
presented in this chapter. Model predictions have been validated against the exper-
imental data measured in TOPFLOW facility at the Helmholtz-Zentrum Dresden-
Rossendrof (HZDR) (Lucas et al. 2010). Overall, most of the kernels have reason-
ably captured the transition of phase distribution from wall peaking to core peaking
within the system. The evolution of bubble size distribution (i.e., from bimodal to
single peak) has been also captured by models.
A closer examination of the predicted results has also revealed that the bubble
collision frequency is dominant by the collided bubble size and local turbulence
level. For the given flow conditions, due to the low eddy dissipation rate at the pipe
center, insignificant coalescence rates are predicted by all coalescence kernels.
Furthermore, at the pipe wall region, the predicted collision frequencies are of the
same order magnitude for all coalescence kernels. While all models predict similar
void fraction profiles, the coalescence kernel by Coulaloglou and Tavlarides gives
slightly better predictions in the void fraction profile and bubble size distribution.
More profound findings were also observed for the selected breakage kernels.
The U-shape daughter size distribution by Luo and Svendsen could be unrealistic as
the capillary pressure is very high for small bubbles. The M-shaped distribution
appears to be more realistic with the same argument. Interestingly, both the U-shaped
and M-shaped distributions produce similar predictions in the present study. The
bell-shaped daughter bubble size distribution by Martinez-Bazan is found to be
favorable for the equal size breakage, leading to the overpredictions of the bubble
size distribution.
Acknowledgments The financial support provided by the Australian Research Council (ARC
project ID DP130100819) is gratefully acknowledged.
Notation
a Coalescence rate
a(Mi, Mj) Coalescence rate in terms of mass
aif Interfacial area concentration
BB,BC Mass birth rate due to breakup and coalescence
C1, C2, CC&T Coalescence model constant
C3, C4, CMB, Kg Breakage model constant
CD Drag coefficient
Cf Coefficient of surface area
CL Lift coefficient
Cw1, Cw2 Wall lubrication coefficients
CTD Dispersion coefficient
DH Maximum bubble horizontal dimension
dij Equivalent diameter
Ds Sauter mean bubble diameter
DB,DC Mass birth rate due to breakup and coalescence
e(λ) Kinetic energy of eddy with size λ
Eo Eötvos number
Eod Modified Eötvos number
f Size fraction
fBV Breakup volume fraction
Flg Total interfacial force
FB Breakup calibration factor
FC Coalescence calibration factor

drag
Flg Drag force
Flift
lg Lift force
Fwall
lg
lubrication
Wall lubrication force
Fturbulent
lg
dispersion
Turbulent dispersion force
ho Initial film thickness
hf Critical film thickness
h (Mi, Mj) Collision frequency in terms of mass
j Superficial velocity
k Turbulent kinetic energy
M Mass scale of gas phase (bubble)
nw Outward vector normal to the wall
n Average bubble number density or weight
P Pressure
Pb Breakage probability
Pe(e(λ)) Energy distribution function
r Breakage rate
r (Mi, Mj) Partial breakage rate in terms of mass
r (Mi) Total breakage rate in terms of mass
Reb Bubble Reynolds number
Si Mass transfer rate due to coalescence and breakup
Scb Turbulent bubble Schmidt number
t Physical time
tij Time for two bubbles to coalesce
u Velocity vector
ut Turbulent velocity
v Volume of bubble
We Webber number
Wecr Critical Webber number
yw Distance from the wall boundary
Greek Symbols
α Void fraction
αmax Maximum allowable void fraction
β(fBV, 1) Daughter bubble size distribution
e Dissipation of turbulent kinetic energy
ηkli Coalescence mass matrix
λ Size of eddy in inertia subrange
λ (Mi, Mj) Coalescence efficiency in terms of mass
λmin Minimum size of eddy in inertia subrange defined as 11.4(v3/e)1/4
μ Viscosity
ρ Density
σ Surface tension
τij Contact time for two bubbles
ξ Internal space vector of the PBE or size ratio between an eddy and a
particle
Super/Subscripts
e Effective
i, j, k Index of gas bubble class
t Turbulent
g Gas phase
l Liquid phase
References
H. Anglart, O. Nylund, CFD application to prediction of void distribution in two-phase bubbly
flows in rod bundles. Nucl. Sci. Eng. 163, 81–98 (1996)
S. Bordel, R. Mato, S. Villaverade, Modeling of the evolution with length of bubble size distribu-
tions in bubble columns. Chem. Eng. Sci. 61, 2663–2673 (2006)
D. Bothe, M. Schmidtke, H.J. Warnecke. VOF-simulation of the lift force for single bubbles in a
simple shear flow. Chemical Engineering & Technology, 29, 1048 (2006)
P. Chen, J. Sanyal, M.P. Duduković, Numerical simulation of bubble columns flows: effect of
different breakup and coalescence closures. Chem. Eng. Sci. 60, 1085–1101 (2005)
A.K. Chesters, G. Hoffman, Bubble coalescence in pure liquids. Appl. Sci. Res. 38, 353–361 (1982)
S.C.P. Cheung, G.H. Yeoh, J.Y. Tu, On the numerical study of isothermal vertical bubbly flow using
two population balance approaches. Chem. Eng. Sci. 62, 4659–4674 (2007)
S.C.P. Cheung, G.H. Yeoh, F. Qi, J.Y. Tu, Classification of bubbles in vertical gas–liquid flow: part
2–a model evaluation. Int. J. Multiphase Flow 39, 135–147 (2012)
S.C.P. Cheung, L. Deju, G.H. Yeoh, J.Y. Tu, Modeling of bubble size distribution in isothermal
gas–liquid flows: numerical assessment of population balance approaches. Nucl. Eng. Des. 265,
120–136 (2013)
C.A. Coulaloglou, L.L. Tavlarides, Description of interaction processes in agitated liquid-liquid
dispersions. Chem. Eng. Sci. 32, 1289 (1977)
M.R. Davidson, Numerical calculations of two-phase flow in a liquid bath with bottom gas
injection: the central plume. Appl. Math. Model. 14, 67–76 (1990)
L. Doubliez, The drainage and rupture of a non-foaming liquid film formed upon bubble impact
with a free surface. Int. J. Multiphase Flow 17, 783–803 (1991)
T. Frank, J. Shi, A.D. Burns. Validation of Eulerian multiphase flow models for nuclear safety
application, in Proceeding of the Third International Symposium on Two-Phase Modelling and
Experimentation, Pisa, 2004
A.G. Fredrickson, D. Ramkrishna, H.M. Tsuchiya, Statistics and dynamics of procaryotic cell
population. Math. Biosci. 1, 327–374 (1967)
H.M. Hulburt, S. Katz, Some problems in particle technology: a statistical mechanical formulation.
Chem. Eng. Sci. 19, 55–574 (1964)
E. Krepper, D. Lucas, H. Prasser, On the modelling of bubbly flow in vertical pipes. Nucl. Eng. Des.
235, 597–611 (2005)
E. Krepper, T. Frank, D. Lucas, H. Prasser, P.J. Zwart. Inhomogeneous MUSIG model – a

Population Balance approach for polydispersed bubbly flows, in Proceeding of the 6th Inter-
national Conference on Multiphase Flow, Leipzig, 2007
E. Krepper, D. Lucas, T. Frank, H. Prasser, The inhomogeneous MUSIG model for the simulation of
polydispersed flows. Nucl. Eng. Des. 238, 1690–1702 (2008)
F. Lehr, D. Mewes, A transport equation for the interfacial area density applied to bubble column.
Chem. Eng. Sci. 56, 1159–1116 (2001)
F. Lehr, M. Millies, D. Mewes, Bubble-size distributions and flow fields in bubble columns. AICHE
J. 48, 2426–2443 (2002)
Y. Liao, D. Lucas, A literature review of theoretical models for drop and bubble breakup in turbulent
dispersions. Chem. Eng. Sci. 64, 3389–3406 (2009)
Y. Liao, D. Lucas, A literature review of theoretical models for drop and bubble coalescence process
of fluid particles. Chem. Eng. Sci. 65, 2851–2864 (2010)
S. Lo, Application of the MUSIG model to bubbly flows. AEAT-1096, AEA Technology, 1996
D. Lucas, E. Krepper, H.M. Prasser, Use of models for lift, wall and trubulent dispersion force
acting on bubbles for poly-disperse flows. Chem. Eng. Sci. 62, 4146–4157 (2007)
D. Lucas, M. Beyer, L. Szalinski, P. Schutz, A new database on the evolution of air-water flows
along a large vertical pipe, Int. J. Ther. Sci. 49, 664–674 (2010)
H. Luo, H. Svendsen, Theoretical model for drop and bubble break-up in turbulent dispersions.
AICHE J. 42, 1225–1233 (1996)
C. Martinez-Bazan, J.L. Montanes, J.C. Lasheras, On the breakup of an air bubble injected into a
fully developed turbulent flow. Part 1. Breakup frequency. J. Fluid Mech. 401, 157–182 (1999a)
C. Martinez-Bazan, J.L. Montanes, J.C. Lasheras, 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, 157–182 (1999b)
F.R. Menter, Two-equation eddy viscosity turbulence models for engineering applications. AIAA
J. 32, 1598–1605 (1994)
E. Olmos, C. Gentric, Ch. Vial, G. Wild, N. Midoux, Numerical simulation of multiphase flow in
bubble column influence of bubble coalescence and break-up. Chem. Eng. Sci. 56, 6359–6365
(2001)
R. Pohorecki, W. Moniuk, P. Bielski, A. Zdrojkwoski, Modelling of the coalescence/redisperison
processes in bubble columns. Chem. Eng. Sci. 56, 6157–6164 (2001)
M.J. Prince, H.W. Blanch, Bubble coalescence and break-up in air sparged bubble columns. AICHE
J. 36, 1485–1499 (1990)
D. Ramkrishna, in Advances in Biochemical Engineering, ed. by T. K. Ghose, A. Fiechter,
N. Blakebrough. Statistical models of cell populations, vol 11 (Springer, Berlin, 1979)
D. Ramkrishna, The status of population balances. Rev. Chem. Eng. 3, 49–95 (1985)
D. Ramkrishna, in Theory and Applications to Particulate Systems in Engineering. Population
balances (Academic, New York, 2000)
D. Ramkrishna, J.D. Borwanker, A puristic analysis of population balance. Chem. Eng. Sci. 28,
1423–1435 (1973)
Y. Sato, M. Sadatomi, K. Sekoguchi, Momentum and heat transfer in two-phase bubbly flow – I. Int.
J. Multiphase Flow 7, 167–178 (1981)
M.P. Schwarz, W.J. Turner, Applicability of the standard k-e model to gas-stirred baths. Appl. Math.
Model. 12, 273–279 (1988)
A. Tomiyama, Struggle with computational bubble dynamics, in Proceeding of the Third Interna-
tional Conference on Multiphase Flow, Lyon, 1998
R.M. Wellek, A.K. Agrawal, A.H.P. Skelland, Shapes of liquid drops moving in liquid media.
AICHE J. 12, 854 (1966)
T. Wang, J. Wang, Y. Jin, A novel theoretical breakup kernel function for bubbles/droplets in a
turbulent flow. Chem. Eng. Sci. 58, 4629–4637 (2003)
Euler-Euler Modeling of Poly-Dispersed
Bubbly Flows
Roland Rzehak
Abstract
CFD simulations of bubbly flow on the scale of technical equipment become
feasible within the Eulerian two-fluid framework of interpenetrating continua. For
practical applications, suitable closure relations are required which describe the
interfacial exchange processes. To facilitate predictive simulations a reference set
of closure relations has been defined for bubble forces and bubble-induced
turbulence including numerical values for all parameters. Models for bubble
coalescence and breakup are not yet included at present due to their lacking
maturity. This means that the bubble size distribution must still be estimated by
other means or treated as a parameter in a sensitivity study. Using measured
bubble sizes, the validity of model predictions for gas fraction, mean velocities of
gas and liquid, and liquid turbulent kinetic energy have been validated over a
range of conditions. These include both pipe flows and bubble columns and
circular and rectangular cross sections. The bubble size is limited to the range
of 1 to 10 mm and the gas fraction to at most 10–20%. The present report gives a
brief but complete description of the model, a few validation examples, and
references for further study.
Keywords
Fluid Dynamics • Dispersed Gas Liquid Multiphase Flow • Euler-Euler
Two-Fluid Model • Closure Relations • CFD Simulation • Model Validation
R. Rzehak (*)
Institute of Fluid Dynamics, Helmholtz-Zentrum Dresden - Rossendorf, Dresden, Germany
e-mail: r.rzehak@hzdr.de

DOI 10.1007/978-981-4585-86-6_4-1
2 R. Rzehak
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Euler-Euler Framework of Interpenetrating Continua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Two-Fluid Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
MUSIG Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Initial and Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Baseline Closure Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Bubble Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Turbulence Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Other Model Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Tests of Liu (1998) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
MTLoop Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
TOPFLOW Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Tests of Mudde et al. (2009) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Tests of bin Mohd Akbar et al. (2012) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Introduction
Bubbly flows are common in many technical applications ranging from chemical
engineering and biotechnology to energy and transportation. Typical for such flows
is the existence of widely disparate length scales, most prominently the size of the
individual bubbles at the small end and the dimension of the domain occupied by the
fluid at the large end. A computational treatment becomes feasible within the
Eulerian two-fluid framework of interpenetrating continua in which the small scales
are eliminated by an averaging procedure and only the large scales are resolved.
However, to obtain a closed system of equations the physics on the scale of
individual bubbles or groups thereof has to be modeled by so-called closure
relations.
A large number of works exist, in each of which largely a different set of closure
relations is compared to a different set of experimental data. For the limited range of
conditions to which each model variant is applied, reasonable agreement with the
data is mostly obtained, but due to a lack of comparability between the individual
works no complete, reliable, and robust formulation has emerged so far. Moreover,
the models usually contain a number of empirical parameters that have been adjusted
to match the particular data that were used in the comparison. Predictive simulation,
however, requires a model that works without any adjustments within the targeted
domain of applicability.
To make a first step towards such a predictive model, we here consider adiabatic
flows in which only momentum is exchanged between the phases. Apart from
interest in its own right, results obtained for this restricted problem also provide a
good starting point for the investigation of more complex situations including heat
and mass transport and possibly phase change or chemical reactions.
Euler-Euler Modeling of Poly-Dispersed Bubbly Flows 3
Aspects requiring closure for the case under consideration are: (i) the exchange of
momentum between liquid and gas phases, (ii) the effects of the dispersed bubbles
on the turbulence of the liquid carrier phase, and (iii) processes of bubble-
coalescence and breakup that determine the distribution of bubble sizes. All of
these aspects are coupled and therefore in principle have to be considered as a
whole. However, to reduce the complexity of the problem, we focus in a first step
on cases where a fixed bubble size distribution can be imposed. We refer to such a
situation as “fixed polydispersity.” The monodisperse approximation, where all
bubbles have the same size, is a special case thereof.
Suitable closure relations for aspects (i) and (ii) relevant for adiabatic bubbly
flows with fixed polydispersity have been collected into a baseline model that serves
as a starting point for the simulation of such systems. Inasmuch as systems that differ
on the large scales are still governed by the same physics on the small length scales, a
unified description by the same closure relations should be possible. This assertion is
verified by applying the baseline model to a number of different configurations
commonly encountered in engineering applications. At the same time a range of
validity is established within which predictions to a certain degree of accuracy are
possible.
Expanding the range of validity as well as the achieved accuracy is a continuously
ongoing development effort. From a detailed analysis of the remaining deviations
between simulation and experiment, directions for improvement of the closure
models can be identified.
Section “Euler-Euler Framework of Interpenetrating Continua” for reference
briefly summarizes the Euler-Euler framework and the inhomogeneous MUSIG
model, which is needed to include effects of polydispersity. In addition, boundary
conditions needed to complete the problem specification are discussed. In section
“Baseline Closure Relations”, a detailed description of the baseline closure models is
given. Section “Model Validation” finally provides a number of illustrative valida-
tion examples. Different numerical setups are used as suitable for the various
problems. Details of these are given as well.
Euler-Euler Framework of Interpenetrating Continua
In this section the Euler-Euler framework for bubbly flows is covered, i.e., flows in
which both phases, dispersed gas and continuous liquid, are treated as
interpenetrating continua as described in the section “Introduction”. The balance
laws for mass and momentum which govern the fluid dynamics of the system are
stated and the terms requiring closure to be discussed in section “Baseline Closure
Relations” are identified. Depending on the treatment of bubble size, several
approaches of increasing complexity may be distinguished.
The so-called two-fluid model applies to a situation where all bubbles have the
same size, i.e., the monodisperse approximation. It considers two separate continuity
and momentum equations for the two phases, gas and liquid.
4 R. Rzehak
A frequently used treatment for polydisperse systems is the so-called MUSIG

(MUltiple SIze Group) approach, which is sometimes also called the class method or
sectional method. In the homogenous MUSIG model, a separate continuity equation
is solved for each size group, but only a single momentum equation for all size
groups together. The inhomogeneous MUSIG model also considers several velocity
groups and solves a separate momentum equation for each of these. The number of
size and velocity groups can be chosen independently to adjust a compromise
between accuracy and computational effort, where the latter increases particularly
with the number of velocity groups. The inhomogeneous MUSIG model thus applies
to a situation in which the bubble size may vary strongly, e.g., due to bubble
coalescence and breakup processes.
For situations in which the polydispersity may be taken as fixed, the full gener-
ality of the inhomogeneous MUSIG models is not needed. In particular, no mass
transfer between the groups occurs. Moreover, the consideration of more than one
size group for each velocity group is possible, but provides no greater accuracy. The
velocity groups are chosen to match marked changes in the bubble forces that cause
different motion of bubbles with different size. An important example is the change
of sign of the lift coefficient (see section “Lift Force”) which requires two velocity
groups.
The subject of the two-fluid model has been amply covered in several mono-
graphs (e.g., Drew and Passman 1998, Yeoh and Tu 2010, Ishii and Hibiki 2011) so
that the presentation in section “Two-Fluid Model” can be kept rather concise. The
emphasis is to provide a coherent background for the later developments including
all auxiliary relations. In addition, some issues that are frequently missed out in other
treatments will be clarified. The extension to the inhomogeneous MUSIG model has
been presented mostly in research papers (e.g., Frank et al. 2008; Krepper et al.
2008). The case of fixed polydispersity considered herein constitutes only a special
case thereof. A summary relevant to this particular application will be given in
section “MUSIG Model”. Finally in section “Initial and Boundary Conditions”,
initial and boundary conditions are discussed, which are necessary for a complete
problem specification. Since the mathematical theory for the multiphase fluid
dynamics of interpenetrating continua is much less developed than the still incom-
plete corresponding theory for single phase fluids, the presentation is largely based
on physical intuition rather than mathematical rigor.
Two-Fluid Model
The balance laws of the two-fluid model relevant to fluid dynamic questions, where
only momentum is exchanged between the phases, may be summarized as follows.
Mass conservation:
@
ðαG ρG Þ þ ∇ ðαG ρG uG Þ ¼ 0 (1)
@t
@
ðαL ρL Þ þ ∇ ðαL ρL uL Þ ¼ 0 (2)
@t
@
ðαG ρG uG Þ þ ∇ ðαG ρG uG uG Þ ¼ αG ∇pG þ ∇ ðαG T G Þ þ Fbody þ Finter
@t G G
(3)
@
ðαL ρL uL Þ þ ∇ ðαL ρL uL uL Þ ¼ αL ∇pL þ ∇ ðαL T L Þ þ Fbody þ Finter (4)
@t L L
Here the index “G” refers a quantity to the gas phase while the index “L” refers it
to the liquid phase. If the index is omitted in the following, it will be understood that
both phases are referred to.
Equations 1, 2, 3, and 4 are 8 equations for the 12 unknowns phase fraction α,
density ρ, velocity u, and pressure p for gas and liquid. Additional relations between
the unknowns are provided by a thermal equation of state for each phase relating
pressure, temperature, and density as
ρ ¼ ρðp, T Þ (5)
and the constraint of phase conservation to be satisfied by the phase fractions, i.e.,
αL þ αG ¼ 1: (6)
But even with these relations, the system is still underdetermined. Therefore,
commonly the additional so-called equilibrium pressure condition
pG ¼ pL p (7)
is assumed.
One may wonder why in contrast to all other terms the phase fraction α appears
outside the gradient in the pressure term in Eqs. 3 and 4. This is the result of a formal
manipulation ∇(α p) = α ∇p + p∇α and treating the second term on the right as part
of the interfacial force density Finter, which has to be modeled anyways. A discussion
of this procedure has been given by Prosperetti and Jones (1984).
The interfacial force densities Finter represent an exchange of momentum between
the phases. Conservation of total momentum may be expressed as
6 R. Rzehak
Finter
L ¼ Finter
G : (8)
Specific expressions for FGinter will be provided in section “Bubble Forces”.

A constitutive relation for the viscous stress tensor T is

2
T ¼ μ ∇u þ ð∇uÞT ð∇ uÞ1 , (9)
3
which applies for both the liquid and the gas phase, but the Newtonian dynamic
viscosity μ may contain turbulent contributions as described in section “Turbulence
Modeling”. As usual, by invoking Stokes’ hypothesis the coefficient of second or
bulk viscosity is eliminated (White 1991, ch 2.4.3). Generalizations to complex
rheological fluids are in principle straight forward.
For the body force, we have mainly gravity in mind, i.e.,
Fbody ¼ Fgrav ¼ αρg, (10)
but others may be included as well.

Realistic equations of state and material properties for liquid and gas may be
taken, e.g., from the NIST Standard Reference Database on Thermophysical Prop-
erties of Fluid Systems [http://webbook.nist.gov/chemistry/fluid]. Since heat transfer
is not considered, the temperature T is simply a constant, but pressure p remains a
variable in these data. Frequently, however, approximations are used instead. If
phases are taken as incompressible, ρ const. We note that strictly speaking
incompressibility means only that density does not depend on pressure. It could
still depend on other variables like temperature or composition, etc., which, how-
ever, under the restrictions laid out in the introduction are constants. If pressure
variations are significant, expansion of the gas
bubbles may bemoltaken into account
according to the ideal gas law ρG ¼ MG =R ðp=T Þ, where MG is the molecular
mol
mass of the gas and R is the universal gas constant. Other material properties are
often assumed as constant.
The notion of incompressibility in the present context has to be considered
carefully. Above we have referred to thermodynamic incompressibility or incom-
pressible materials. In fluid mechanics, incompressibility is considered as a property
of the flow rather than the material, and the requirement for incompressible flow is
that velocities remain small compared with the speed of sound, or in other words that
pressure variations propagate much faster than the flow. In the following, incom-
pressible flow of both phases will be assumed. As laid out this does not preclude
considering expansion of the bubbles due to pressure variations.
In single phase flows, one has gotten accustomed to identifying incompressibility
with a divergence-free velocity field. For multiphase flows, however, even if both
phases by themselves are incompressible, the divergence of the individual velocities
u in Eqs. 1 and 2 needs not vanish. For constant densities, these equations turn into
evolution equations for the phase fractions, i.e.,
@
ðαG Þ þ ∇ ðαG uG Þ ¼ 0
@t
@
ðαL Þ þ ∇ ðαL uL Þ ¼ 0:
@t
These may be added and simplified by using phase conservation to obtain the
constraint
∇ ðαG uG þ αL uL Þ ¼ 0, (11)
which has to be satisfied by an appropriate pressure field. This is seen to be a

vanishing divergence of the so-called mixture velocity umix = αGuG + αLuL.
The general form of the equations is commonly derived by averaging a more
detailed model in which the interface between the phases is resolved. Different
averaging procedures have been used, namely, time, volume, and ensemble averages
(Delhaye and Archard 1977; Drew and Passman 1998, ch 8–12; Prosperetti 2003,
Sect. 2). Correspondingly, the phase fraction α is interpreted as residence time
fraction, volume fraction or ensemble expectation value. Considering the time
evolution of the multiphase flow as a random process, the ensemble average is
taken as an average over different realizations of this process, whereas time and
volume averages are taken over a time interval or control volume in a single
realization. Time and volume averages relate directly to typical experimental data,
which are taken in a single run. An ensemble average can in principle be observed if
the same experiment is conducted multiply.
A distinguishing feature of the ensemble average is that it yields truly instanta-
neously and locally defined values, whereas time and volume averages have a limited
resolution. In fact, for integrals over a finite time or volume to become true averaging
operations rather than filtering, a separation of scales is needed (Delhaye and Archard
1977). On the other hand, the link of fluctuating quantities to small temporal or spatial
scales provides relations that are otherwise unavailable. Some theorems under which
condition the different kinds of averages are equivalent are provided by ergodic theory
(Cushman 1983), which however apply to infinite averaging times or volumes.
Awaiting more practically applicable results, equivalence may be assumed if the
variations to be investigated take place on sufficiently large temporal / spatial scales
compared to scale over which the averaging takes place.
Another aspect to keep in mind is that the variables appearing in Eqs. 1, 2, 3, and
4 represent averages with different weighting (e.g., Drew and Passman 1998, ch 11).
Some are weighted with respect to phase fraction, others with respect to phase
density. The reason for introducing these weights is to eliminate as many correlation
terms as possible from the equations. However, as discussed in Delhaye and Archard
(1977), this makes the average fluctuations of the weighted variables non-zero.
Finally, in relation to turbulence one further has to distinguish single and double
averaging depending on whether phase averaging and Reynolds decomposition are
applied at once or in two separate steps (Hill 1998, ch 1.6; Burns et al. 2004). Since
both phase and Reynolds averaging may likewise be considered as time, space, or
8 R. Rzehak
ensemble average, it is not quite clear wherein the distinction lies. A careful
examination of works that have employed the double averaging (e.g., Elgobashi
and Abou-Arab 1983; Besnard and Harlow 1988; Burns et al. 2004) revealed that the
second average has always been performed after some further modeling of the
exchange terms was introduced. Thus it appears more adequate to consider this
second average as a heuristic extension of closures that were originally conceived for
laminar flows to turbulent ones.
MUSIG Model
The MUSIG (MUltiple SIze Group) approach which is sometimes also called the
class method or sectional method is but one possible approach to treat polydisperse
bubbly flows. The main reasons to choose this approach are that it represents a rather
straight forward generalization of the two-fluid model to multiple gas “phases” and
that it is easily adapted to the dependences on bubble size found in the closure
relations. This contrasts with other frequently used approaches such as the method of
moments (MOM) (e.g., Marchisio and Fox 2013, ch 3) or the interfacial area
transport equation (IATE) (e.g., Ishii and Hibiki 2011, ch 10–11) which are mathe-
matically much more involved and less flexible to accommodate the size dependence
of closure relations. On the downside it must be admitted that there are indications
that the MUSIG approach may be computationally less efficient than these (e.g.,
Sanyal et al. 2005; Selma et al. 2010).
The first step to a model for polydisperse bubbly flows is the choice of a suitable
internal coordinate, for which the bubble size dB is an obvious possibility but not the
only one. Equally possible is the use of the bubble mass mB. Both are of course
closely related. Taking the spherical equivalent diameter as a measure of bubble size
this relation is expressed as
1=3
6 mB
dB ¼ : (12)
π ρG
When the gas density changes, for example, due to changes in hydrostatic
pressure, a different value of bubble size is obtained depending on the local gas
density, while the bubble mass is of course the same everywhere. Therefore, when
using the bubble size as the internal coordinate an extra term is needed to describe
the expansion or contraction of bubbles due to pressure change, which is not needed
when using the bubble mass. In addition, devising numerical methods that ensure
strict mass conservation becomes straight forward in the latter case. The advantage
of using the bubble size on the other hand is that it relates directly to closure
correlations. This allows to place group boundaries directly where significant
changes in the bubble dynamics occur. Most implementations to date use bubble
mass as the internal coordinate and we follow that approach here.
Consideration of polydispersity customarily starts from the population balance
equation (Ramkrishna 2000; Marchisio and Fox 2013, ch 4), from which a set of
continuity equations is derived for each size group. For the inhomogeneous MUSIG
model these are augmented by a set of momentum equations for the velocity groups
(Frank et al. 2008, Krepper et al. 2008). For the present situation assuming fixed
polydispersity, the number of size and velocity groups is taken the same or in other
words there is only a single size group for each velocity group. The balance laws
correspond to a multi-fluid model with Eqs. 2 and 4 for the continuous liquid and
multiple copies of Eqs. 1 and 3 for the bubbles belonging to each size group.
Numbering these by the index j = 1 . . . N, the system of equations is written as
@
αG, j ρG þ ∇ αG, j ρG uG, j ¼ 0 (13)
@t
@
αG, j ρG uG, j þ ∇ αG, j ρG uG, j uG, j
@t

¼ αG, j ∇p þ ∇ αG, j TG, j þ αG, j ρG g þ Finter
G, j (14)
@
ðαL ρL Þ þ ∇ ðαL ρL uL Þ ¼ 0 (15)
@t
@
ðαL ρL uL Þ þ ∇ ðαL ρL uL uL Þ ¼ αL ∇p þ ∇ ðαL TL Þ þ αL ρL g þ Finter
L : (16)
@t
With
X
N
αG ¼ αG, j (17)
j¼1
for the total gas fraction and
X
N
Finter
G ¼ Finter
G, j (18)
j¼1
for the total interfacial force density acting on the gas, the definitions of section
“Two-Fluid Model” carry over.
For comparison with experimental data we define furthermore a total gas velocity as
P
N
αG, j uG, j
j¼1
uG ¼ : (19)
P
N
αG, j
j¼1
To complete the setup, values for the bubble size and inlet gas fraction must be
determined from a measured or otherwise estimated bubble size distribution. We
take this distribution to be given in the form of a differential distribution of the gas
fraction contained in bubbles of size d, i.e., as @αG(d ) / @d.
10 R. Rzehak
The bubble size is discretized into a number of finite intervals j = 1. . .N, each
defining a size group as shown in Fig. 1. For the present purpose, typically N is a
small number. The upper and lower boundaries of the j-th interval are denoted as
dj1/2 with d1/2 = 0 and dM+1/2 defining the maximum bubble size that occurs. The
other values of dj1/2 here are chosen to correspond with sizes where the behavior of
the bubble changes. The length of the intervals typically varies and the representative
size dj for each group is typically different from the central value.
Sizes are d converted to / from masses m by means of Eq. 12 as part of the pre-/
post-processing. If pressure varies throughout the domain, so will the group bound-
aries dj1/2. This introduces deviations from the intended behavior that have to be
accepted in a formulation based on bubble mass.
The gas fraction for each group is calculated straightforwardly as
ð
djþ1=2
@αG ðdÞ
αG, j ¼ dd: (20)
@d
d j1=2
For the representative size of each group we take the Sauter diameter, i.e., the
diameter of a sphere that has the same ratio of volume to surface area as the
cumulative distribution of bubbles within the group. Thereby the individual bubbles
are assumed to be of spherical shape as well. The corresponding expression is
αG, j
dj ¼ : (21)
Ð
djþ1=2
1 @αG ðdÞ
dd
dj1=2 d @d
Fig. 1 Discretization of the internal coordinate

Initial and Boundary Conditions
Different types of boundary conditions have to be distinguished which represent

different levels of modeling, namely, wall, inlet, outlet, and symmetry conditions.
Wall boundary conditions describe the physical interaction of the fluid with another
continuum material, most often a solid, which is not spatially resolved in the model.
Inlet and outlet boundary conditions serve to truncate the computational domain and
thus ought to represent the effects of the neglected part of the real world in the
specific situation considered. Symmetry conditions similarly serve to truncate the
computational domain, but the assumption of symmetry implies a rather general
form of the relations to be imposed. Of course, when using this type of boundary
condition, one has to be aware that in reality symmetry of the flow may be broken
dynamically even in geometrically symmetric configurations. All of these different
types will be discussed in the following in general terms. Specific models for in- and
outlet conditions are described in section “Inlet and Outlet Conditions”.
Concerning wall conditions, we consider only solid walls on which the pertinent
boundary conditions are the following. As the walls are impenetrable to both liquid
and gas, the components of both liquid and gas velocities in direction normal to the
wall have to vanish. Conditions on tangential components of liquid and gas velocity
or shear stress are motivated by considering that the bubbles have no contact with the
wall, except for occasional bounces that do not matter here. This means that the
tangential components of the liquid velocity and the gas shear stress have to vanish.
Eqs. 1 and 13 suggest that no condition is required for the MUSIG gas fractions on
solid walls since these quantities are just transported along the path lines of the flow
field which cannot cross the walls. As will be seen in section “Turbulent Dispersion
Force” in connection with turbulent dispersion, a vanishing gradient should none-
theless be imposed since also no diffusive transport through the wall is possible.
At an outlet one typically specifies a constant pressure and requires vanishing
tangential velocity components both liquid and gas. No condition is needed for the
gas fractions. At an inlet it is common to specify profiles of liquid and gas velocities
as well as gas fractions and density. The liquid fraction is then determined by volume
conservation Eq. 6 and the remaining two thermodynamic variables – the other
density and pressure or both densities – are determined by the thermal equation of
state Eq. 5 of section “Two-Fluid Model”.
Symmetry conditions are no different from single phase flow, i.e., the normal
component of velocity and derivatives in normal direction of all other variables have
to vanish.
For transient calculations, of course the initial state of all fields at all locations of
the domain has to be given corresponding to the physical problem under investiga-
tion. For stationary calculations, the initial conditions do not matter and it is in
principle possible to start the calculation from zero field values. However, problems
with slow or no convergence at all may result, which can be avoided by using the
best available approximation to the true solution as initial condition.
12 R. Rzehak
Baseline Closure Relations
In this section closure relations for bubbly flows are described. Within the present
scope this comprises (i) the exchange of momentum between liquid and gas phases,
i.e., bubble forces and (ii) the effects of the dispersed bubbles on the turbulence of
the liquid carrier phase, the so-called bubble-induced turbulence. These aspects are
coupled and therefore in principle have to be considered as a whole. The purpose of
this contribution therefore is to present a selection of closures that works well
together in practical applications rather than to give the broadest possible coverage
of each topic.
Bubble Forces
Concerning momentum exchange between liquid and gas phase, we consider drag,
lift, wall, turbulent dispersion, and virtual mass forces, i.e., for FGinter in section
“Euler-Euler Framework of Interpenetrating Continua” we get
Finter
G ¼ Fdrag þ Flift þ Fwall þ Fdisp : (22)
All terms depend parametrically on the bubble size, so the bubble size of the j-th
MUSIG group has to be inserted for dB. The correlations are expressed in terms of
dimensionless numbers, namely, the Reynolds-, Eötvös-, and Morton numbers
defined as
jU rel j dB
Re ¼ , (23)
νL
Δρg d2B
Eo ¼ , (24)
σ
Δρρ2L g ν4L
Mo ¼ : (25)
σ3
Drag Force
The drag force reflects the resistance opposing bubble motion relative to the sur-
rounding liquid. The corresponding gas-phase momentum source is given by
3
Fdrag ¼ CD ρL αG juG uL jðuG uL Þ: (26)
4d B
Dimensional analysis (Grace 1973) indicates that for deformable bubbles rising in
a quiescent liquid the drag coefficient CD depends on two of the dimensionless
numbers Re, Eo, and Mo, the former two being the common choice. A correlation
distinguishing different shape regimes has been suggested by Ishii and Zuber (1979),
namely

CD ¼ max CD, sphere , min CD, ellipse , CD, cap , (27)
where
24
CD, sphere ¼ 1 þ 0:1Re0:75
Re
2 pffiffiffiffiffiffi
CD, ellipse ¼ Eo (28)
3
8
CD, cap ¼ :
3
This correlation was compared with an extensive data set on the terminal velocity
of bubbles rising in quiescent liquids covering several orders of magnitude for each
of Re, Eo, and Mo in Tomiyama et al. (1998) with good agreement except at high
values of Eo.
Lift Force
A bubble moving in an unbounded shear flow experiences a force perpendicular to
the direction of its motion. The momentum source corresponding to this shear lift
force, often simply referred to as lift force, can be calculated as (Zun 1980)
Flift ¼ CL ρL αG ðuG uL Þ rotðuL Þ: (29)
For a spherical bubble the shear lift coefficient CL is positive so that the lift force
acts in the direction of decreasing liquid velocity, i.e., in case of cocurrent pipe flow
in the direction towards the pipe wall. Experimental (Tomiyama et al. 2002) and
numerical (Bothe et al. 2006) investigations showed that the direction of the lift force
changes its sign if a substantial deformation of the bubble occurs. From the obser-
vation of the trajectories of single air bubbles rising in simple shear flow of a glycerol
water solution, the following correlation for the lift coefficient was derived:
8
> min½0:288tanhð0:121ReÞ, f ðEo⊥ Þ Eo⊥ < 4
<
CL ¼ f ðEo⊥ Þ for 4 < Eo⊥ < 10
>
: (30)
0:27 10 < Eo⊥
with f ðEo⊥ Þ ¼ 0:00105Eo3⊥ 0:0159Eo2⊥ 0:0204Eo⊥ þ 0:474:
This coefficient depends on the modified Eötvös number given by
gðρL ρG Þd2⊥
Eo⊥ ¼ , (31)
σ
14 R. Rzehak
where d⊥ is the maximum horizontal dimension of the bubble. It is calculated using

an empirical correlation for the aspect ratio by Wellek et al. (1966) with the
following equation:
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d⊥ ¼ dB 1 þ 0:163Eo0:757 ,
3
(32)
where Eo is the usual Eötvös number.

The experimental conditions on which Eq. 30 is based were limited to the
range 5.5 log10 Mo 2.8, 1.39 Eo 5.74, and values of the dimensionless
shear rate 0 Sr ¼ d B ̇ γ=uterm 2. The water-air system at normal conditions has a
Morton number Mo = 2.63∙1011 which is quite different, but good results have
nevertheless been reported for this case (Lucas and Tomiyama 2011).
Wall Force
A bubble translating next to a wall in an otherwise quiescent liquid also experiences a lift
force. This wall lift force, often simply referred to as wall force, has the general form
2
Fwall ¼ CW ρL αG juG uL j2 ^y , (33)
dB
where ^y is the unit normal perpendicular to the wall pointing into the fluid. The
dimensionless wall force coefficient CW depends on the distance to the wall y and is
expected to be positive so the bubble is driven away from the wall.
Based on the observation of single bubble trajectories in simple shear flow of a
glycerol water solution, Tomiyama et al. (1995) and later Hosokawa et al. (2002)
concluded a functional dependence
2
dB
CW ðyÞ ¼ f ðEoÞ : (34)
2y
In the limit of small Morton number, the correlation
f ðEoÞ ¼ 0:0217Eo (35)
can be derived from the data of Hosokawa et al. (2002). The experimental conditions
on which Eq. 35 is based are 2.2 Eo 22 and 6.0 log10 Mo 2.5 which is
still different from the water-air system with Mo = 2.63∙1011, but good predictions
have been obtained also for air bubbles in water (Rzehak et al. 2012).
Turbulent Dispersion Force

The turbulent dispersion force describes the effect of the turbulent fluctuations of
liquid velocity on the bubbles. Burns et al. (2004) derived an explicit expression by
Favre averaging the drag force as

3 αG μturb 1 1
F disp
¼ CD juG uL j L
þ gradαG : (36)
4 dB σ TD αL αG
In analogy to molecular diffusion, σ TD is referred to as a Schmidt number. In

principle it should be possible to obtain its value from single bubble experiments
also for this force by evaluating the statistics of bubble trajectories in well-
characterized turbulent flows, but to our knowledge this has not been done yet.
A value of σ TD = 0.9 is typically used.
Virtual Mass Force

When a bubble is accelerated, a certain amount of liquid has to be set into motion as
well. This may be expressed as a force acting on the bubble as

DG uG DL uL
FVM ¼ CVM ρL αG , (37)
Dt Dt
where DG / Dt and DL / Dt denote material derivatives with respect to the velocity of

the indicated phase. For the virtual mass coefficient, a value of CVM = 0.5 has been
derived for isolated spherical bubbles in inviscid and creeping flows by Auton et al.
(1988) and Maxey and Riley (1983), respectively. Results of direct simulations of a
single bubble by Magnaudet et al. (1995) suggest that this value also holds for
intermediate values of Re.
Turbulence Modeling
Due to the low density and small spatial scales of the dispersed gas, it suffices for
bubbly flows to consider turbulence in the continuous liquid phase. Two contribu-
tions to the turbulent fluctuations have to be taken into account, which are referred to
as shear-induced and bubble-induced turbulence. Despite some discussion in the
literature (e.g., Rensen et al. 2005, Riboux et al. 2010), we here treat both contribu-
tions as indistinguishable and describe them by means of a single total turbulent
kinetic energy. To this end a standard two-equation model is used, which is known to
work well for single phase flows where only shear-induced turbulence exists. Details
are given in section “Basic Turbulence Model”. To account for the bubble-induced
contribution, this model is augmented with suitable source terms which are
described in section “Source Terms for Bubble-Induced Turbulence”. To avoid the
need to resolve the viscous sublayer near solid walls, a turbulent wall function is
applied. Lacking definite results for two-phase flows this is presently taken the same
as for single phase flow. Details are given in section “Turbulent Wall Function”.
16 R. Rzehak
Basic Turbulence Model

Turbulence in bubbly flows is here described by an SST model (Menter et al. 2003,
Menter 20091) with additional source terms for the bubble-induced contribution. The
SST model is a wall-distance dependent blend of k-ε and k-ω models that combines
the respective advantages of both. Usually k and ω are employed as independent
variables by noting that the ε-equation may be transformed to an equivalent equation
for ω that contains a cross-diffusion term, which is not present in the usual
ω-equation. Hence the equations for the turbulent kinetic energy kL and the turbulent
frequency ωL to be solved are
@ 1 turb

ðαL ρL kL Þ þ ∇ ðαL ρL uL kL Þ ¼ ∇ αL μmol
L þ σ k μL ∇kL
@t (38)

þαL Pk Cμ ρL ωL kL þ SkL
@ 1 turb

ðαL ρL ωL Þ þ ∇ ðαL ρL uL ωL Þ ¼ ∇ αL μmol
L þ σ ω μL ∇ω
@t (39)
ρ Pk ∇kL ∇ωL
þαL CωP Lturb CωD ρL ω2L þ αL 2 σ 1 ω2 ρL ð1 F1 Þ þ SωL :
μL ωL
Here F1 denotes the blending function, which assumes a value of one for the k-ω
model and zero for the k-ε model. It is defined as
20 0 pffiffiffiffiffi 114 3
kL 500μmol
6B L
B max Cμ ωL y , ρ ωL y2 , CC 7
6B B CC 7
6B B
L
CC 7
F1 ¼ tanh6BminB 1
4σ ω2 ρL kL CC 7:
6@ @ AA 7
4 5
2
y2 max σω2 ρL ∇kLω∇ω
L
L
, 1:0 10 10
(40)
The model constants Cμ, CωP, CωD, σ 1 1

k , and σ ω are also interpolated between the
corresponding values of the k-ω model (index “1”) and the k-ε model (index “2”)
using the blending function F1 as
χ ¼ F1 χ 1 þ ð1 F1 Þχ 2 : (41)
All turbulence model parameters take their usual single phase values as summa-
rized in Table 1. Note that these values deviate slightly from those commonly used
for the k-ω and k-ε models alone (NASA 2014).
In terms of the strain rate tensor
1
Note that the ANSYS CFX User Guide (ANSYS 2012) quotes (Menter 1994) but describes the
later modifications as implemented in the code.
Table 1 Parameter values for k-o and k-E models

Cμ CωP CωD σ1
k σ1
ω
k-ω model (index “1”) 0.09 0.5532 0.075 0.85034 2.0
k-ε model (index “2”) 0.09 0.4463 0.0828 1.0 0.85616

S ¼ 1=2 ∇uL þ ð∇uL ÞT , (42)
the production term is Pek ¼ 2μturb

L S : uL but a limiter is introduced to prevent the
build-up of turbulent kinetic energy in stagnation zones so that

Pk ¼ min Pek , 10Cμ ρL ωL kL : (43)
Since bubble-induced effects are included in k and ω due to the respective source
terms, the turbulent viscosity is evaluated from the standard relation of the SST
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
model which includes a limiter based on the generalized shear rate γ ¼ 2 S : S
ρ k
μturb ¼ L L , (44)
L
max ωL , Cγ F2 γ
where F2 is a second blending function defined as

" pffiffiffiffiffi 2 #
2 kL 500μmol
F2 ¼ tanh max , L
(45)
Cμ ωL y ρL ωL y2
and Cγ = 1 / 0.31 is a further model constant.

The effective viscosity is simply μLeff = μLmol þ μLtrub.
Boundary conditions on k and ω are taken the same as for the single phase case,
which is consistent with the view that the full wall shear stress is exerted by the
liquid phase, as it contacts the complete wall area.
Source Terms for Bubble-Induced Turbulence

Concerning the source term SkL describing bubble effects in the k-equation, there is
large agreement in the literature (e.g., Kataoka et al. 1992, Troshko and Hassan
2001). A plausible approximation is provided by the assumption that all energy lost
by the bubble due to drag is converted to turbulent kinetic energy in the wake of the
bubbles. Hence, the k-source becomes
SkL ¼ Fdrag
L ðuG uL Þ: (46)
The source term SLe that describes the effect of bubble-induced turbulence in the
ε-equation of k-ε models is derived using similar heuristics as for the single phase
case, namely, the k-source is divided by some time scale τ so that
18 R. Rzehak
SkL
SeL ¼ CeB : (47)
τ
Further modeling then focuses on the time scale τ proceeding largely based on
dimensional analysis. This follows the same line as the standard modeling of shear-
induced turbulence in single phase flows (Wilcox 1993), where production terms in the
ε-equation are obtained by multiplying corresponding terms in the k-equation by an
appropriate time scale that represents the lifetime of a turbulent eddy before it breaks up
into smaller structures. In single phase turbulence the relevant variables are obviously k
and ε from which only a single time scale τ = kL/eL can be formed. For the bubble-induced
turbulence in two-phase flows, the situation is more complex and several plausible
expressions for the time scale are conceivable. In the absence of theoretical arguments to
decide which of these is the most relevant one, a comparison of all four alternatives has
shown the best performance for the choice τ = dB /√ kL (Rzehak and Krepper 2013a, b). For
the coefficient CeB a value of 1.0 was found to give reasonable results.
For use with the SST model, the ε-source is transformed to an equivalent
ω-source which gives
1 e ωL k
SωL ¼ S S : (48)
Cμ k L L k L L
This ω-source is used independently of the blending functions in the SST model
since it should be effective throughout the fluid domain.
Turbulent Wall Function

To avoid the need to resolve the viscous sublayer, a single phase turbulent wall
function assuming a smooth wall is applied, which consists of a blend between
inertial and viscous sublayers (ANSYS 20122). This treatment is facilitated by the
availability of analytical solutions for both sublayers in the k-ω model (Wilcox 1993,
Sect. 4.6.3) that cover the near wall region in the SST model. It allows mesh
refinement near the wall to a degree that the viscous sublayer becomes resolved.
Since turbulence is considered in the liquid phase only, all variables in the following
presentation refer to this phase, but the index “L” has been dropped for notational
convenience.
In terms of variables nondimensionalized with the friction velocity and
corresponding viscous length, denoted by subscript “+”, the analytical solutions
for the viscous and inertial sublayers are
uviscous
þ ¼ yþ , (49)
2
This approach is termed automatic near-wall treatment in the ANSYS CFX User Guide. No
reference is quoted there and only partial accounts could be found in the literature (e.g., Vieser
et al. 2002, Esch et al. 2003).
1
uinertial
þ ¼ ln Eyþ , (50)
κ
where κ = 0.41 and E = 9.8 for smooth walls.
A simple interpolation between both regimes is furnished by
h 4 4 i1=4
ucompound
þ ¼ uviscous
þ þ uinertial
þ : (51)
If u+ is known, the friction velocity can be obtained by straight forward inversion

of Eq. 51 or equivalently from
h 4 4 i1=4
ucompound
τ ¼ uviscous
τ þ uinertial
τ , (52)
where uτviscous and uτinertial are found from inverting Eqs. 49 and 50, respectively.
In case u+ vanishes, the friction velocity and the viscous length scale become
ill-defined. This may happen at points where the boundary layer separates. To
alleviate this issue, following Launder and Spalding (1974) the alternative velocity
scale
uinertial
k ¼ C1=4
μ k
1=2
(53)
is introduced for the inertial sublayer. Variables nondimensionalized with this

alternative velocity and corresponding viscous length will be denoted by subscript
“*”. With this modification, explicit expressions for the two terms in Eq. 52 are
sffiffiffiffiffiffiffiffiffi
μ j uj
uviscous
τ ¼ (54)
ρy
κ juj
uinertial
τ ¼ : (55)
lnðE y Þ
In addition, the overall alternative velocity scale is defined by

h 4 4 i1=4
ucompound
k ¼ uviscous
τ þ uinertial
k : (56)
Now for u and k the known values at the wall adjacent grid cell are taken, uτ and
uk are calculated, and from these the wall shear stress
τW ¼ ρuτ uk , (57)
which provides the flux of momentum into this cell through the wall.
Similarly to the above, the turbulent frequency ω is computed from the solutions
in the viscous and inertial sublayer
20 R. Rzehak
6μ
ωviscous ¼ , (58)
CωD1 ρ y2
1 uk
ωinertial ¼ pffiffiffiffiffiffi , (59)
Cμ κ y
however, with a different blending, namely

h 2 2 i1=2
ωcompound ¼ ωviscous þ ωinertial : (60)
The value from the analytical solution is specified in the wall adjacent grid cell for ω.
The boundary condition for k is a vanishing normal derivative at the wall.
Other Model Aspects
This section collects some aspects which do not strictly belong to closure modeling
in the sense that they would describe bubble-scale phenomena that have been
averaged out in the two-fluid model. However, they are necessary ingredients for
any computation and will certainly have an impact on the results obtained. Hence,
validation of the closures as presented in the previous sections also depends on these
aspects of the overall model. These miscellanies comprise the calculation of the wall
distance (section “Wall Distance Calculation”) and inlet and outlet conditions
(section “Inlet and Outlet Conditions”).
Wall Distance Calculation

An often overlooked and somewhat fuzzy aspect of bubbly flow modeling in
complex geometries as they frequently occur in engineering applications is the
calculation of the normal wall distance, which is required for both the blending
functions for the SST turbulence model (see section “Basic Turbulence Model”) and
the wall force (see section “Wall Force”). For a simple geometry, one might simply
take the closest distance to the nearest wall, but for complex geometries involving
curved boundaries or sharp corners, a more global measure is needed. Such a global
measure is obtained by solving a partial differential equation for a distance function ϕ
and computing the normal wall distance y from ϕ (see e.g., Tucker 2003). An
admittedly crude but widely used approach attributed by Liu et al. (2010) to a
contribution by Spalding (1994), which however apparently never appeared in
print, is as follows. The distance function is obtained as solution to the Poisson
equation
∇2 φ ¼ 1: (61)
On walls a Dirichlet boundary condition φ = 0 is specified and on other types of

boundaries a homogeneous Neumann boundary condition @@ φn ¼ 0 is used. The wall
distance y is then computed from
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
y ¼ j∇φj þ j∇φj2 þ 2φ, (62)
which gives the exact result for the one-dimensional case.
Inlet and Outlet Conditions

While conditions at solid walls or symmetry planes are determined from fundamen-
tal physical principles as discussed in section “Initial and Boundary Conditions”,
conditions at inlets and outlets are specific to each case. For model validation these
conditions are best taken from measurements to eliminate any error. For predictive
simulations, however, this is not possible and additional modeling is necessary.
Commonly used prescriptions are described in the following.
Typically volumetric fluxes JL and JG of liquid and gas through a reference area
Aref are given as part of the problem specification. Care must be taken correctly to
consider the reference area.
Inlet conditions for the liquid velocity and turbulence are frequently set to fully
developed single phase flow profiles corresponding to the required liquid volume
flux. In general these profiles have to be calculated numerically. This calculation may
be combined with the simulation at hand by introducing a sufficiently long flow
development zone.
For a round pipe with radius R, a useful approximation of the normal velocity u is
given by the empirical power law profile (Schlichting 1979, p. 599)
u r 1=7
¼ 1 , (63)
umax R
where the peak value umax is calculated such that the desired volume flux JL results.
The turbulent kinetic energy k is often computed from the liquid velocity and a
specified value for the turbulence intensity Iturb as
3
k ¼ ðI turb juL jÞ2 : (64)
2
A common value used for the turbulence intensity is Iturb = 5%. The turbulent
frequency ω is computed from the turbulent kinetic energy k and the integral
turbulent length scale Lturb as
k1=2
ω ¼ C1
μ : (65)
Lturb
For pipe flows, the integral turbulent length scale is often estimated as
Lturb = 0.2 R.
22 R. Rzehak
Concerning the gas two options are commonly used. For the first, the gas is
assumed to be distributed over the full cross section at the inlet. In this case the inlet
areas for gas and liquid coincide and a value for the gas fraction must be specified.
The gas velocity may be set to a constant value or a profile that corresponds to the
liquid velocity shifted by the relative velocity of the bubbles. The constant value or
the relative velocity, respectively, is determined to match the desired gas flux JG. The
gas is taken to be uniformly distributed over the inlet so that the gas fraction is
constant. Assuming that the bubbles enter with zero relative velocity to the liquid, its
value is obtained in terms of the given volume fluxes JL and JG as
JG
αG ¼ , (66)
JG þ JL
whereas for a nonvanishing relative velocity
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

1 JG þ JL 1 JG þ JL 2 JG
α¼ 1þ 1þ : (67)
2 urel 4 urel urel
For the second option, the true area through which gas enters is taken as the gas
inlet. Since only gas enters, the gas fraction is specified as 1. The liquid inlet then
must exclude this area and a gas fraction of 0 is specified there. The gas velocity is
typically set to a constant value over the gas inlet. If the gas injector works in the
bubbling regime, a suitable value is obtained as
4Aref
uG ¼ J G : (68)
πd2B
Precise conditions at the inlet do not matter as long as the axial distance to the
measurement location is large enough for fully developed conditions to be attained.
That this is indeed the case may be checked by looking at the axially resolved fields.
At an outlet, typically a constant pressure is prescribed while the normal deriv-
atives of the tangential velocity components of gas and liquid are set to zero. This
still allows the normal derivative of the normal components as well as the values of
all velocity components of both liquid and gas velocities to adjust in the simulation.
For other variables, a vanishing normal derivative is prescribed. For flow in pipes or
ducts, this corresponds with the assumption of fully developed flow.
For cases like bubble columns, where the liquid cannot leave the domain, a
common alternative is the so-called degassing condition. There the liquid normal
velocity is set to zero which means that now the pressure becomes part of the
solution. Its distribution over the outlet area can be considered to describe the effect
that while the fluid surface is taken fixed in the simulation, it may deform in reality.
If the outlet is sufficiently far away from the measurement location, the outlet
condition should have no influence on the observed results. If necessary an artificial
flow abatement zone may be added to the physical domain to ensure this. A length of
~10% of the main flow section is a reasonable practical estimate for the required
length.
Model Validation
The closure models defined in the previous sections have been implemented in
ANSYS-CFX and also in OpenFOAM (Rzehak and Kriebitzsch 2015). The basic
equations which can be augmented by user-defined closures are available also in
other closed and open software packages.
Several applications to bubbly pipe flow and flow in bubble columns will be
presented in the following to illustrate the state of the art. An overview of the
pertinent experimental conditions is given in Table 2, where D is the pipe or column
diameter or width, JL and JG are liquid and gas volume fluxes, 〈dB〉 is the average
bubble size, and 〈αG〉 the average gas fraction. The results are taken from previous
publications (Rzehak and Krepper 2013a, b; Rzehak et al. 2014, 2015; Rzehak and
Krepper 2015; Ziegenhein et al. 2015).
All of these applications are limited to upward vertical flows in simple geometries
and air bubbles in water (see Zidouni et al. 2015; Liao et al. 2016; Rzehak et al.
2017a, 2017b, for more complex applications). The bubble size or its distribution is
taken from the experiment rather than being modeled reflecting the state of model
development. While it may be reasonably assumed that other materials are also well
described due the use of nondimensional numbers in the correlations, other geom-
etries and boundary conditions as well as largely different parameter values require
additional validation. Many works on such cases using different models may be
found in the literature as well, a review of which, however, is beyond the scope of
this chapter.
Depending on the test under investigation, different setups were used. The
calculations were made either in stationary mode imposing plane/axisymmetric
conditions by considering only a thin slice/sector of the domain together with
symmetry conditions or in transient mode with subsequent averaging of the results
and fully 3D on the same domain as the experiments. The reason to choose the
stationary or quasi-2D approximation whenever applicable is that it drastically
reduces the computation time. For the transient simulations, the reported quantities
are averages over the statistically steady state.
At the inlet a uniform distribution of gas throughout the cross section was
assumed or the injection nozzles or needles were modeled as individual surface
patches. For the liquid, fully developed single phase velocity and turbulence profiles
were assumed in the pipe flow cases.
At the top, a pressure boundary condition was set for the pipe flow cases while the
degassing condition was employed for the bubble column cases.
24 R. Rzehak
Table 2 Main Name D JL JG 〈dB〉〈αG〉

experimental conditions for
mm m s1 m s1 mm %
the selected test cases
Liu (1998): round pipe
L21B 57.2 1.0 0.14 3.03 10.6
L21C 57.2 1.0 0.13 4.22 9.6
L22A 57.2 1.0 0.22 3.89 15.7
L11A 57.2 0.5 0.12 2.94 15.2
MTLoop: round pipe
MT039 51.2 0.4050 0.0111 4.50 1.89
MT041 51.2 1.0167 0.0115 4.50 1.00
MT061 51.2 0.4050 0.0309 4.50 5.03
MT063 51.2 1.0167 0.0316 4.50 2.64
Mudde et al. (2009): round bubble column
M1 150 – 0.015 4.02 6.1
M2 150 – 0.017 4.06 7.6
M3 150 – 0.025 4.25 11
M4 150 – 0.032 4.47 16
M5 150 – 0.039 4.53 20
M6 150 – 0.049 4.44 25
TOPFLOW: round pipe
TL12–041 195.3 1.017 0.0096 4.99 1.1
bin Mohd Akbar et al. (2012): flat bubble column
A1 240 – 0.003 4.3 1.4
A2 240 – 0.013 5.5 6.2
Concerning bubble size, a monodisperse approximation was used whenever the

bubbles are smaller than the diameter of ~6 mm where the lift force changes its sign
for the water-air system (cf. section “Lift Force”). In the other cases, two size and
velocity groups corresponding to bubbles smaller and larger than 6 mm were used.
If there is a significant variation of pressure within the domain, the gas density
will change according to the ideal gas law and consequently the bubble size changes,
since mass is conserved. Yet the flow of both gas and liquid remains incompressible
to a good approximation. It is possible to keep the computational advantage of
treating both gas and liquid as incompressible fluids with constant material proper-
ties if a fully developed flow is considered by adjusting the gas flux at the inlet to the
value obtained by evaluating
ð
D=2
J G ¼ 8=D 2
αG ðr ÞuG ðr Þrdr (69)
0
using the data at the measurement location (Rzehak et al. 2012). In cases in which
only uL but not uG has been measured, an estimate of the latter may be obtained from
the former and αG based on the assumption of fully developed stationary flow.
Where this procedure has been applied, the adjusted values are given in Table 2.
Turbulence data frequently give the axial intensity of turbulent fluctuations while
in the simulations based on two-equation models only the turbulent kinetic energy is
available. For a comparison, it has to be considered that wall-bounded turbulence is
anisotropic with the axial component of fluctuating velocity being larger than those
in radial and azimuthal directions. Data on axial and radial components for typical
turbulent two-phase flow conditions are given, e.g., in Michyoshi and Serizawa
(1986), Wang et al. (1987), or Shawkat et al. (2008), while data on the azimuthal
component could not be found in the literature. Concerning turbulent kinetic energy
k, an estimate of the ratio √ k / u’ may be obtained from the quoted works by
assuming that azimuthal and radial components are equal as √ k / u’ 0.8 . . .1.2.
The lower bound is a bit larger than the value √(1/2) 0.71 corresponding to the
unidirectional limiting case while the upper bound is almost the same as
√(3/2) 1.22 obtained for isotropic turbulence. Taking u’ as an estimate for √ k
thus provides an estimate that is accurate to within ~20%.
Tests of Liu (1998)
The system studied by Liu (1998) is vertical upflow of water and air in a round pipe
with inner diameter D = 57.2 mm at a temperature of 26 C and presumably
atmospheric pressure. The total length of the flow section was H = 3.43 m. Gas was
delivered through a tube of 9.7 mm inner diameter located at the center of the main
pipe. A special gas injector was used that allowed to adjust the bubble size inde-
pendently of liquid and gas superficial velocities. A variety of combinations of these
three parameters are available. Radial profiles of gas fraction, mean bubble-size,
axial liquid velocity, and axial liquid turbulence intensity were measured at an axial
position H / D = 60 corresponding to fully developed conditions.
In Rzehak and Krepper (2013a, b) stationary axisymmetric simulations were done
assuming a uniform distribution of gas throughout the pipe cross-section at the inlet.
Since the pressure effect is significant for the 3.43 m long pipe, gas volume fluxes
were adjusted to allow treating the gas as incompressible. A selection of test cases
was made for which a monodisperse bubble size distribution is sensible. The bubble
size was set equal to the average of the measured profiles.
The comparison of calculated and measured profiles in Fig. 2 shows reasonable
agreement for the gas fraction αG and the axial liquid velocity uL. Notable deviations
occur in the region close to the wall where the simulations predict the peak in the gas
fraction too high and the gradient of the liquid velocity too steep. For the turbulent
kinetic energy kL, the agreement between simulation and measurement is not as
good, but possibly to some extent due to the isotropic approximation.
26 R. Rzehak
Fig. 2 Fully developed gas fraction αG, axial liquid velocity uL, and square root of turbulent kinetic
energy √ kL for the tests of Liu (1998). Lines: simulation, symbols: experiment
MTLoop Tests
The MTLoop facility described in detail in Prasser et al. (2003) and Lucas et al.
(2005) was used to study upward vertical flow of air and water at a slightly elevated
temperature of 30 C and atmospheric pressure. The test section consists of a circular
pipe with inner diameter D = 51.2 mm. Gas was injected through up to 19 needles
depending on the required gas volume flux. The distance between the gas injector
and the measurement location was varied between H = 0.03 m and H = 3.03 m
which corresponds to a ratio H / D 60. Thus at the highest level fully developed
flow may be expected. Radial profiles of gas fraction and gas velocity as well as
distributions of bubble size were measured by a wire-mesh sensor. A large number of
combinations of liquid and gas volumetric fluxes JG and JL were investigated.
Simulations of four selected tests in the bubbly flow regime have been performed
with a setup that aims at calculating the fully developed flow at the end of the test
section (Rzehak et al. 2015). To this end, the gas volumetric flux was adjusted to
capture the gas expansion due to the drop of hydrostatic pressure over the height of
the pipe and the bubble size distribution measured at this location was imposed. To
capture the different direction of motion for large and small bubbles according to the
lift-force, two velocity groups were used for bubbles larger and smaller than 6 mm in
size. Values of bubble size dj and relative amount fj of both classes for all tests are
given in Table 3. At the inlet, a uniform distribution of gas was imposed and
axisymmetric flow was assumed so that only a narrow sector of the pipe needs to
be considered as the computational domain.
Measured and calculated radial profiles of gas fraction αG and gas velocity uG are
compared in Fig. 3. For the gas fraction the overall agreement with the measured
values is quite reasonable. Here the peak near the wall is predicted quite well, but
deviations are seen in the pipe where the predicted gas fraction is sometimes too high
and sometimes too low. For the gas velocity the agreement between simulation and
experiment is very good.
Table 3 MUSIG group Case d1 [mm] d2 [mm] f1 f2

sizes dj and relative
MT039 4.85 6.52 0.9364 0.0636
amounts fj for MTLoop
simulations MT041 4.79 6.51 0.9621 0.0379
MT061 4.89 6.69 0.7964 0.2036
MT063 5.00 6.54 0.8783 0.1217
Fig. 3 Fully developed gas fraction αG and gas velocity uG, for the MTLoop tests. Lines:
simulation, symbols: experiment
28 R. Rzehak
Table 4 Length of test section for different levels of gas injection for the TOPFLOW tests
Level A C F I L O R
H [mm] 221 335 608 1552 2595 4531 7802
H/D [-] 1.1 1.7 3.1 7.9 13.3 23.2 39.9
TOPFLOW Tests
The TOPFLOW facility has been specifically designed to obtain high quality data for
the validation of CFD models. The tests used here (Lucas et al. 2010) have been run
for cocurrent vertical upward flow of air and water in a round pipe with an inner
diameter of D = 195.3 mm. Measurements were made by a wire mesh sensor at the
top end of the pipe while gas injection occurs at different levels below. The operating
conditions were set to a temperature of 30 C and a pressure of 0.25 MPa at the
location of the active gas injection. In this way the flow development can be studied
as it would be observed for gas injection at a fixed position and measurements taken
at different levels above. Distances H between the injection devices and the sensor
are given in Table 4 for the different levels. The values of mean bubble size and
average gas fraction given in Table 2 correspond to the highest measurement level R.
Instrumentation with a wire mesh sensor allows collection of data on radial
profiles of gas - fraction and gas - velocity as well as distributions of bubble size.
A large range of liquid and gas superficial velocities was investigated in which all
flow regimes from bubbly to annular occur. In the detailed report (Beyer et al. 2008)
it has been noted that for bubbly flows the gas volume fluxes calculated from the
measured profiles by integrating the product of gas fraction and velocity were
systematically larger than those measured by the flow meter controlling the inlet.
This deviation is likely to be caused by the distance between the sending and
receiving wire planes, which leads to an increased value of gas fraction, but a
detailed explanation is not available yet. The ratio of the values calculated from
the profiles (cf. Eq. 37) to the values measured directly at the inlet has an approx-
imately constant value of 1.2 over the bubbly flow regime (Beyer et al. 2008,
Fig. 1–19). In the simulations the values measured by the flow meter will be used
to set the inlet boundary condition. To get the same integral value of this conserved
quantity for each cross-sectional plane, all measured void gas fractions are divided
by 1.2 throughout this section.
For this data set, the investigation of developing flow is possible. To this end tests
have been selected by (Rzehak and Krepper 2015) in which a significant polydis-
persity is present but the distribution of bubble masses does not change appreciably
and a treatment by fixed polydispersity is possible. The change in bubble size due to
the decrease in hydrostatic pressure with height is included in the simulations.
Stationary axisymmetric simulations were done with the gas inlet modeled as
individual nozzles. Two MUSIG groups were used with sizes of 4.4 mm and
6.6 mm at the inlet and relative amounts of 94.4% and 5.6%. As already mentioned,
the sizes increase with height due to the calculation based on mass discretization and
the pressure effect on gas density.
Fig. 4 Developing gas fraction αG and gas velocity uG, for the TOPFLOW test case TL12–041.
Top: experimental data corrected as described in the text, bottom: simulation results
Results for the development of gas fraction αG and gas velocity uG for one of the
cases from Rzehak and Krepper (2015) are shown in Fig. 4. It can be seen that near
the inlet the wall peak in the gas fraction is underestimated by the simulation but at
the higher levels it is overpredicted. Likewise the initial width of the peak comes out
too broad in the simulations but the shoulder that develops subsequently has a
narrower range than in the experiments. In particular at the highest level, R, the
gas has reached the pipe center in the experiment but not yet in the simulation. For
the gas velocity, experimental values exist only where a certain minimum amount of
gas is present. The simulated profiles are in reasonable accord with the data up to
level O. A large deviation is seen at the highest level, R, which probably corresponds
with the gas not being distributed over the full pipe radius.
30 R. Rzehak
Tests of Mudde et al. (2009)
The setup of Mudde et al. (2009) consists of a round bubble column with diameter
D = 150 mm again operated with air and water at ambient conditions. The ungassed
fill height was H = 1.3 m. Measurements of gas fraction and axial liquid velocity
profiles were taken at different levels of which the one at 0.6 m above the inlet has
been chosen for the comparison here. The sparger was designed specifically to
provide highly uniform inlet conditions. Several values of the gas superficial veloc-
ity are available reaching rather large values of gas fraction. The mean bubble size
and variation around it have been measured at two locations close to the inlet and
close to the top water level. Since a slight increase is observed, the average value of
both measurements corresponding to the middle level has been used in the
simulations.
Transient 3D simulations of these tests were performed by Rzehak et al. (2014)
assuming a uniform distribution of gas at the inlet. The height of the computational
domain was obtained by adding the average gas volume from the measurements to
the volume of liquid. A monodisperse bubble size distribution corresponding to the
measured values was used. For the column height of 1.3 m, the pressure effect is still
small enough to be neglected.
Measured and calculated values are compared in Fig. 5. Clearly the gas fractions
αG are predicted within the experimental errors. The calculated liquid velocity
profiles uL do not depend on the total gas hold up. Since the measured profiles do
not show any systematic trend as a function of this variable, their variation is most
likely an indication of the measurement errors.
Fig. 5 Gas fraction αG, axial liquid velocity uL, for the tests of Mudde et al. (2009). Lines:
simulation results, symbols: measured values. Only half of the column is shown
Tests of bin Mohd Akbar et al. (2012)
The experiments of bin Mohd Akbar et al. (2012) were conducted in a flat bubble
column of width D = 240 mm using air bubbles in water at ambient conditions.
Without gas supply the water level was at H = 0.7 m. Profiles of gas fraction, axial
liquid velocity, and axial turbulence intensity as well as the bubble size distribution
were measured at a plane 0.5 m above the inlet. The bubble size distribution in
addition was measured also near the inlet. Two values of superficial gas velocities JG
are available.
From the measured bubble size distributions, it can be seen that no significant
change occurs over the column height. Hence a treatment by fixed polydispersity is
possible. Due to the small column height the pressure effect on material properties is
negligible. In particular the gas density remains constant to a good approximation so
that there is no expansion of the bubbles. The simulations were run in transient
(URANS) mode on the full 3D domain (Ziegenhein et al. 2015). The resulting data
were averaged over a sufficiently long period to obtain stationary averages. For the
evaluation of turbulent kinetic energy, the axial component of the resolved transient
fluctuations has been added to the unresolved part obtained from the turbulence
model. The change in the water level due to the gas is small enough not to affect the
flow at the measurement level and was hence neglected. Individual needles of the
sparger were represented by area elements roughly corresponding to the size of the
bubbles. For the lower value of gas volume flux JG, a monodisperse bubble size
distribution was imposed. For the higher value, two MUSIG size and velocity groups
were used with diameters of 5.3 mm and 6.3 mm and relative amounts of 63% and
37%, respectively. This treatment allows to capture the effect of the sign change of
the lift coefficient (see section “Lift Force”), which for air bubbles in water takes
place at a bubble size of dB 6 mm.
A comparison between simulation results and measured data is shown in Fig. 6.
As may be seen, the agreement between both is quite good for gas fraction αG and
axial liquid velocity uL. Slight differences are that the predicted gas fraction profile is
a little bit too peaked near the wall and there is a small dip in the predicted liquid
velocity in the center of the column. The turbulent kinetic energy in the column
center is somewhat underpredicted by the simulations, and the peak in kL near the
wall is not reproduced by the simulations.
Discussion and Conclusions
As shown by the examples discussed in section “Model Validation,” in its present

state the Euler-Euler simulation of polydispersed bubbly flows may be used for case
studies, the design of better experiments, to enhance understanding in combination
with experiments, and within a limited range for prediction. In the future, full
predictive capability and use for optimization are envisaged, but further model
development and validation are needed to reach this goal.
32 R. Rzehak
Fig. 6 Gas fraction αG, axial liquid velocity uL, and square root of turbulent kinetic energy √kL for
the tests of bin Mohd Akbar et al. (2012). Lines: simulation results, symbols: measured values. Only
half of the column is shown
Concerning the basic framework presented in section “Euler-Euler Framework of

Interpenetrating Continua,” the most important desideratum would be a stronger link
between terms derived by averaging and closure models. At present this replacement
is treated heuristically. Possible approaches that allow a more fundamental corre-
spondence to be exploited are discussed, e.g., in Prosperetti (1998) and Buffo and
Marchisio (2014). In this respect it would also be desirable to relax the point particle
approximation. A possible way to achieve this is discussed in Tomiyama et al. (2006,
Sect. 4.2ff) together with the expected benefits.
For the closure models discussed in section “Baseline Closure Relations”
concerning bubble forces, the biggest issue in view of applications to industrial
problems is probably an extension of the range of applicability towards higher gas
fractions by the inclusion of so-called swarm effects. Suggestions in this direction
have been made, e.g., in Roghair et al. (2011) showing that direct numerical
simulations provide a viable tool to improve closures on bubble forces. This could
also be used to extend the range of applicability towards larger bubbles. Another
aspect requiring further developments is a general lift force combining shear-, wall-,
and other effects.
Concerning turbulence including the turbulent dispersion force, there is obvi-
ously a great need for model improvement until a similar state as in single phase
flows is reached. This comprises better source term models for the k-ε/ω models as
well as two-phase wall functions and also the nature of the turbulent dispersion
process. Extension of two-equation to full Reynolds stress models with source terms
is needed for more complex geometries. Direct numerical simulations hold a great
promise to make progress particularly in this area.
The modeling of coalescence and breakup processes is a cornerstone in a full
model of polydisperse bubbly flows since all other submodels strongly depend on
the bubble size. An attempt to include these processes has been made, e.g., in Liao
et al. (2015). However, the nature of these processes is not yet sufficiently under-
stood especially for turbulent flows. In addition, physicochemical properties of the
involved materials have significant influence on the rates of these processes.
For the validation of models, more comprehensive CFD-grade data sets are
needed. This means in particular that local values are needed, e.g., cross-sectional
profiles at different locations along the main flow direction. Full sets of observables
including in addition to phase fractions and phasic velocities also the bubble size
distribution and turbulent fluctuations are highly desirable. This serves both to cover
all aspects of the various submodels as well to facilitate the analysis of possible
deviations between the model predictions and the validation data. Parametric vari-
ations of all independent variables – gas and liquid fluxes, bubble size, pipe/column
diameter and height, material properties, etc. – are necessary to obtain a range of
applicability that is large enough to make the models useful for practical applica-
tions. Finally, accurate measurements in particular with small systematic errors are
needed. Ideally this criterion would be fulfilled by any good measurement technique,
but multiphase flows pose significant challenges in this respect.
Concerning all of these aspects, there is a dynamic ongoing development which
may be expected to lead to fully predictive models in the future.
Nomenclature
Notation Unit Denomination

CD – Drag coefficient
CL – Lift coefficient
CTD – Turbulent dispersion coefficient
CVM – Virtual mass force coefficient
CW – Wall force coefficient
Cμ – Shear-induced turbulence coefficient (k-ε model)
dB m Bulk bubble diameter
d⊥ m Bubble diameter perpendicular to main motion
D m Pipe / column diameter or width
(continued)
34 R. Rzehak

Eo – Eötvös Number
F N m3 Force
g m s2 Acceleration of gravity
H m Pipe length / column height
Iturb – Turbulence intensity
J m s1 Superficial velocity = volumetric flux
k m2 s2 Turbulent kinetic energy
Lturb m Characteristic eddy size Integral turbulent length scale
Mo – Morton Number
p Pa Pressure
r m Radial coordinate
R m Pipe / column radius or half-width
Re – Reynolds number
t s Time
T N m2 Stress tensor
u m s1 Velocity
uτ m s1 Friction velocity
U m s1 Velocity scale
u’ m s1 Fluctuation velocity
x m Axial coordinate
y m Wall normal coordinate
z m Spanwise coordinate
α – Phase fraction
δ m Viscous length scale
e m2 s3 Turbulent dissipation rate
μ kg m1 s1 Dynamic viscosity
ν m2 s1 Kinematic viscosity
ρ kg m3 Density
σ N m1 Surface tension
τW N m2 Wall shear stress
ω s1 Turbulent frequency
References
ANSYS, ANSYS CFX-Solver Theory Guide Release 14.5 (ANSYS Inc., Canonsburg, 2012)
T. Auton, J. Hunt, M. Prud'Homme, The force exerted on a body in inviscid unsteady non-uniform
rotational flow. J. Fluid Mech. 197, 241 (1988)
D. Besnard, F. Harlow, Turbulence in multiphase flow. Int. J. Multiphase Flow 14, 679 (1988)
M. Beyer, D. Lucas, J. Kusin, P. Schütz, Air-water experiments in a vertical DN200-pipe. Technical
report, Helmholtz-Zentrum Dresden - Rossendorf, FZD-505, 2008 (in German)
M. Bin Mohd-Akbar, K. Hayashi, S. Hosokawa, A. Tomiyama, Bubble tracking simulation of
bubble-induced pseudoturbulence. Multiph. Sci. Technol. 24, 197 (2012)
D. Bothe, M. Schmidtke, H.-J. Warnecke, VOF-simulation of the lift force for single bubbles in a
simple shear flow. Chem. Eng. Technol. 29, 1048 (2006)
A. Buffo, D.L. Marchisio, Modeling and simulation of turbulent polydi sperse gas-liquid systems
via the generalized population balance equation. Rev. Chem. Eng. 30, 73–126 (2014)
A.D. Burns, T. Frank, I. Hamill, J.-M. Shi, The Favre averaged drag model for turbulence dispersion
in Eulerian multi-phase flows, in Procceedings of 5th International Conference on Multiphase
Flow, ICMF2004, Yokohama, 2004
J.H. Cushman, Volume averaging, probabilistic averaging, and ergodicity. Adv. Water Resour. 6,
182 (1983)
J.M. Delhaye, J.L. Achard, On the use of averaging operators in two-phase. Flow Model. Metallurg.
Mater. Technologist 1, 289 (1977)
D.A. Drew, S.L. Passman, Theory of Multicomponent Fluids (Springer, New York, 1998)
S.E. Elghobashi, T.W. Abou-Arab, A two-equation turbulence model for two-phase flows. Phys.
Fluids 26, 931 (1983)
T. Esch, F. Menter, W. Vieser, Heat transfer predictions based on two-equation turbulence models,
in Proceedings of the 6th ASME-JSME Thermal Engineering Joint Conference, paper
TED-AJ03–542, 2003
T. Frank, P. Zwart, E. Krepper, H.-M. Prasser, D. Lucas, Validation of CFD models for mono- and
polydisperse air–water two-phase flows in pipes. Nucl. Eng. Des. 238, 647 (2008)
J.R. Grace, Shapes and velocities of bubbles rising in infinite liquids. Trans. Inst. Chem. Eng. 51,
116 (1973)
D.P. Hill, The computer simulation of dispersed two-phase flows, dissertation, Imperial College of
Science, London, 1998
S. Hosokawa, A. Tomiyama, S. Misaki, T. Hamada, Lateral migration of single bubbles due to the
presence of wall, in Proceedings of ASME Joint U.S.-European Fluids Engineering Division
Conference, FEDSM2002, Montreal, 2002
M. Ishii, T. Hibiki, Thermo-fluid Dynamics of Two-Phase Flow, 2nd edn. (Springer, New York,
2011)
M. Ishii, N. Zuber, Drag coefficient and relative velocity in bubbly, droplet or particulate flows.
AICHE J. 25, 843–855 (1979)
I. Kataoka, D.C. Besnard, A. Serizawa, Basic equation of turbulence and modeling of interfacial
transfer terms in gas-liquid two-phase flow. Chem. Eng. Commun. 118, 221 (1992)
E. Krepper, D. Lucas, T. Frank, H.-M. Prasser, P. Zwart, The inhomogeneous MUSIG model for the
simulation of polydispersed flows. Nucl. Eng. Des. 238, 1690 (2008)
B.E. Launder, D.B. Spalding, The numerical computation of turbulent flows. Comput. Methods
Appl. Mech. Eng. 3, 269–289 (1974)
Y. Liao, R. Rzehak, D. Lucas, E. Krepper, Baseline closure model for dispersed bubbly flow:
Bubble-coalescence and breakup. Chem. Eng. Sci. 122, 336 (2015)
J. Liao, T. Ziegenhein, T.R. Rzehak, Bubbly flow in an airlift column: a CFD study. J. Chem.
Technol. Biotechnol. 91, 2904–2915 (2016)
T. J. Liu, The role of bubble size on liquid phase turbulent structure in two-phase bubbly flow, in
Proceedings of 3rd International Conference on Multiphase Flow, ICMF1998, Lyon, 1998
C.-B. Liu, P. Nithiarasu, P. Tucker, Wall distance calculation using the Eikonal/Hamilton-Jacobi
equations on unstructured meshes. Int. J. Comput-Aided Eng Softw. 27, 645–657 (2010)
D. Lucas, A. Tomiyama, On the role of the lateral lift force in poly-dispersed bubbly flows. Int.
D. Lucas, E. Krepper, H.-M. Prasser, Development of co-current air–water flow in a vertical pipe.
Int. J. Multiphase Flow 31, 1304–1134 (2005)
D. Lucas, M. Beyer, J. Kussin, P. Schütz, Benchmark database on the evolution of two-phase flows
in a vertical pipe. Nucl. Eng. Des. 240, 2338 (2010)
J. Magnaudet, M. Rivero, J. Fabre, Accelerated flows past a rigid sphere or a spherical bubble part 1:
steady straining flow. J. Fluid Mech. 284, 97 (1995)
D.L. Marchisio, R.O. Fox, Computational Models for Polydisperse Particulate and Multiphase
Systems (Cambridge University Press, Cambridge/New York, 2013)
36 R. Rzehak
M.R. Maxey, J.J. Riley, Equation of motion for a small rigid sphere in a nonuniform flow. Phys.
Fluids 26, 883 (1983)
F. Menter, Two-equation eddy-viscosity turbulence models for engineering applications. AIAA
J. 32, 1598–1605 (1994)
F.R. Menter, Review of the shear-stress transport turbulence model experience from an industrial
perspective. Int. J. Comput. Fluid Dyn. 23, 305–316 (2009)
F. R. Menter, M. Kuntz, R. Langtry, Ten years of industrial experience with the SST turbulence
model, in Turbulence, Heat and Mass Transfer 4, ed. by K. Hanjalic, Y. Nagano, Tummers, M.
(Begell House, 2003)
I. Michiyoshi, A. Serizawa, Turbulence in two-phase bubbly flow. Nucl. Eng. Des. 95, 253–267
(1986)
R.F. Mudde, W.K. Harteveld, H.E.A. van den Akker, Uniform flow in bubble columns. Ind. Eng.
Chem. Res. 48, 148 (2009)
NASA, in Turbulence Modeling Resource (NASA Langley Research Center, 2014) http://
turbmodels.larc.nasa.gov/index.html
H.-M. Prasser, D. Lucas, E. Krepper, D. Baldauf, A. Böttger, U. Rohde, P. Schütz, F.-P.W. Zippe,
W. Zippe, J. Zschau, Flow maps and models for transient two-phase flows. Technical report,
Forschungszentrum Dresden Rossendorf, FZR-379 (2003) (in German)
A. Prosperetti, Ensemble averaging techniques for disperse flows, in Particulate Flows: Processing
and Rheology, eds. by D. Drew, D. Joseph, S. L. Passman (Springer, New York 1998)
A. Prosperetti, Two-fluid modelling and averaged equations. Multiph. Sci. Technol. 15, 181 (2003)
A. Prosperetti, A. Jones, Pressure forces in disperse two-phase flow. Int. J. Multiphase Flow 10,
425 (1984)
D. Ramkrishna, Population Balances – Theory and Applications to Particulate Systems in Engi-
neering (Academic, San Diego, 2000)
J. Rensen, S. Luther, D. Lohse, The effect of bubbles on developed turbulence. J. Fluid Mech. 538,
153 (2005)
G. Riboux, F. Risso, D. Legendre, Experimental characterization of the agitation generated by
bubbles rising at high Reynolds number. J. Fluid Mech. 643, 509 (2010)
I. Roghair, Y. Lau, N. Deen, H. Slagter, M. Baltussen, M. Van Sint Annaland, J. Kuipers, On the
drag force of bubbles in bubble swarms at intermediate and high Reynolds numbers. Chem.
Eng. Sci. 66, 3204–3211 (2011)
R. Rzehak, E. Krepper, Closure models for turbulent bubbly flows: a CFD study. Nucl. Eng. Des.
265, 701–711 (2013a)
R. Rzehak, E. Krepper, CFD modeling of bubble-induced turbulence. Int. J. Multiphase Flow 55,
138–155 (2013b)
R. Rzehak, E. Krepper, Bubbly flows with fixed polydispersity: validation of a baseline closure
model. Nucl. Eng. Des. 287, 108 (2015)
R. Rzehak, S. Kriebitzsch, Multiphase CFD-simulation of bubbly pipe flow: a code comparison. Int.
R. Rzehak, E. Krepper, C. Lifante, Comparative study of wall-force models for the simulation of
bubbly flows. Nucl. Eng. Des. 253, 41–49 (2012)
R. Rzehak, E. Krepper, T. Ziegenhein, D. Lucas, A baseline model for monodisperse bubbly flows,
in 10th International Conference on CFD in Oil & Gas, Metallurgical and Process Industries
(CFD2014), Trondheim, 2014
R. Rzehak, E. Krepper, Y. Liao, T. Ziegenhein, S. Kriebitzsch, D. Lucas, Baseline Model for the
Simulation of Bubbly Flows Chem. Eng. Technol. 38, 1972–1978 (2015)
R. Rzehak, T. Ziegenhein, S. Kriebitzsch, E. Krepper, D. Lucas, Unified modeling of bubbly flows
in pipes, bubble columns, and airlift columns. Chem. Eng. Sci. 157, 147–158 (2017a)
R. Rzehak, M. Krauß, P. Kovats, K. Zähringer, Fluid Dynamics in a Bubble Column: New
Experiments and Simulations. Int. J. Multiphase Flow 89, 299–312 (2017b)
J. Sanyal, D.L. Marchisio, R.O. Fox, K. Dhanasekharan, On the comparison between population
balance models for CFD simulation of bubble columns. Ind. Eng. Chem. Res. 44, 5063–5072 (2005)
H. Schlichting, in Boundary Layer Theory, 7th ed (McGraw-Hill, New York, 1979)

B. Selma, R. Bannari, P. Proulx, Simulation of bubbly flows: Comparison between direct quadrature
method of moments (DQMOM) and method of classes (CM). Chem. Eng. Sci. 65, 1925–1941
(2010)
M. Shawkat, C. Ching, M. Shoukri, Bubble and liquid turbulence characteristics of bubbly flow in a
large diameter vertical pipe. Int. J. Multiphase Flow 34, 767–785 (2008)
D. Spalding, Calculation of turbulent heat transfer in cluttered spaces, in Proceedings of 10th
International Heat Transfer Conference, Brighton, 1994
A. Tomiyama, A. Sou, I. Zun, N. Kanami, T. Sakaguchi, Effects of Eötvös number and dimension-
less liquid volumetric flux on lateral motion of a bubble in a laminar duct flow, in Proceedings of
2nd International Conference on Multiphase Flow, Kyoto, Japan, 3, 1995
A. Tomiyama, I. Kataoka, I. Zun, T. Sakaguchi, Drag coefficients of single bubbles under normal
and micro gravity conditions. JSME Int. J. B 41, 472–479 (1998)
A. Tomiyama, H. Tamai, I. Zun, S. Hosokawa, Transverse migration of single bubbles in simple
shear flows. Chem. Eng. Sci. 57, 1849–1858 (2002)
A. Tomiyama, K. Sakoda, K. Hayashi, A. Sou, N. Shimada, S. Hosokawa, Modeling and hybrid
simulation of bubbly flow. Multiph. Sci. Technol. 18, 73 (2006)
A.A. Troshko, Y.A. Hassan, A two-equation turbulence model of turbulent bubbly flows. Int.
J. Multiphase Flow 27, 1965 (2001)
P. Tucker, Differential equation-based wall distance computation for DES and RANS. J. Comput.
Phys. 190, 229–248 (2003)
W. Vieser, T. Esch, F. Menter, Heat transfer predictions using advanced two-equation turbulence
models, Technical Report, ANSYS Inc, 2002
S.K. Wang, S.J. Lee, O.C. Jones Jr., R.T. Lahey Jr., 3-D turbulence structure and phase distribution
measurements in bubbly two-phase flows. Int. J. Multiphase Flow 13, 327–343 (1987)
R.M. Wellek, A.K. Agrawal, A.H.P. Skelland, Shapes of liquid drops moving in liquid media.
AICHE J. 12, 854–862 (1966)
F.M. White, Viscous Fluid Flow (McGraw-Hill, New York, 1991)
D. C. Wilcox, Turbulence Modeling for CFD (DCW-Industries, La Canada, 1993)
G.H. Yeoh, J.Y. Tu, Computational Techniques for Multiphase Flows – Basics and Applications
(Butterworth-Heinemann, Burlington, 2010)
F. Zidouni, E. Krepper, R. Rzehak, S. Rabha, M. Schubert, M.U. Hampel. Simulation of gas-liquid
flow in a helical static mixer. Chem. Eng. Sci. 137, 476–486 (2015)
T. Ziegenhein, R. Rzehak, D. Lucas, Transient simulation for large scale flow in bubble columns.
Chem. Eng. Sci. 122, 1 (2015)
I. Zun, The transverse migration of bubbles influenced by walls in vertical bubbly flow. Int.
Euler-Euler Modeling of Segregated Flows
and Flows with Transitions Between
Different Flow Morphologies
Thomas Höhne
Abstract
Stratified two-phase flows are relevant in many industrial applications, e.g.,
pipelines, horizontal heat exchangers, and storage tanks. Special flow character-
istics as flow rate, pressure drop, and flow regimes have always been of engi-
neering interest. The numerical simulation of free surface flows can be performed
using phase-averaged multi-fluid models, like the homogeneous and the two-fluid
approaches, or non-phase-averaged variants. The approach shown in this chapter
within the two-fluid framework is the algebraic interfacial area density (AIAD)
model. It allows the macroscopic blending between different models for the
calculation of the interfacial area density and improved models for momentum
transfer in dependence on local morphology. An approach for the drag force at the
free surface was introduced. The model improves the physics of the existing
two-fluid approaches and is already applicable for a wide range of industrial two
phase flows. A further step of improvement of modeling the turbulence was the
consideration of sub-grid wave turbulence (SWT) that means waves created by
Kelvin-Helmholtz instabilities that are smaller than the grid size. The new
approach was verified and validated against horizontal two-phase slug flow data
from the HAWAC channel and smooth stratified flow experiments of a different
rectangular channel. The results approve the ability of the AIAD model to predict
key flow features like liquid holdup and free surface waviness. Furthermore, an
evaluation of the velocity and turbulence fields predicted by the AIAD model
against experimental data was done. The results are promising and show potential
for further model improvement.
Keywords
CFD • Horizontal flow • AIAD • Two-phase flow • HAWAC • HZDR
T. Höhne (*)
Helmholtz-Zentrum Dresden-Rossendorf (HZDR), Dresden, Germany
e-mail: t.hoehne@hzdr.de

DOI 10.1007/978-981-4585-86-6_5-1
2 T. Höhne
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Modeling Free Surface Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
The CFD Approaches Applicable to Free Surface Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Algebraic Interfacial Area Density Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Modeling the Free Surface Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Sub-grid Wave Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Turbulence Damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Verification and Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Horizontal Wavy and Stratified Flow: Fabre Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
CFD Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Horizontal Slug Flow: HAWAC Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Introduction
In the last decade, applications of computational fluid dynamic (CFD) methods for
industrial applications received more and more attention, as they proved to be a
valuable complementary tool for design and optimization. The main interest toward
CFD consists in fact in the possibility of obtaining detailed 3D complete flow-field
information on relevant physical phenomena at lower cost than experiments.
Stratified two-phase flows are relevant in many industrial applications, e.g.,
pipelines, horizontal heat exchangers, and storage tanks. Special flow characteristics
as flow rate, pressure drop, and flow regimes have always been of engineering
interest. Wallis and Dobson (1973) analyzed the onset of slugging in horizontal
and near-horizontal gas-liquid flows. Flow maps which predict transitions between
horizontal flow regimes in pipes were introduced, e.g., by Taitel and Dukler (1976)
and Mandhane et al. (1974). The most important flow regimes are smooth stratified
flow, wavy flow, slug flow, and elongated bubble flow. Taitel and Dukler (1976)
explained the formation of slug flow by the Kelvin-Helmholtz instability. They also
proposed a model for the frequency of slug initiation (Taitel and Dukler 1977). The
viscous Kelvin-Helmholtz analysis proposed by Lin and Hanratty (1986) generally
gives better predictions for the onset of slug flow.
Typically, free surfaces manifest as stratified, wavy, or slug flows in horizontal
flow domain where gas and liquid are separated by gravity. The simulation of slug
formation is a sensitive test case for the model setup regarding the quality of the
models for interfacial friction, respectively, momentum transfer. A general overview
on the phenomenological modeling of slug flow was given by Hewitt (2003) and
Valluri et al. (2008). Various multidimensional numerical models were developed to
simulate stratified flows: marker and cell (Harlow and Welch 1965), Lagrangian grid
methods (Hirt et al. 1974), volume of fluid (VOF) method (Hirt and Nichols 1981),
and level set method (Osher and Sethian 1988). These methods can in principle
capture accurately most of the physics of the stratified flows. However, they cannot
capture all the morphological formations such as small bubbles and droplets, if the
Euler-Euler Modeling of Segregated Flows and Flows with Transitions Between. . . 3
grid is not sufficiently small. One of the first attempts to simulate mixed flows was
^
presented by Cerne et al. (2001) who coupled the VOF method with a two-fluid
model in order to bring together the advantages of both formulations.
Mouza et al. (2001) were numerically investigating the characteristics of hori-
zontal wavy stratified flow in circular pipes and rectangular channels. They used the
CFD code CFX for a simulation of the gas and liquid flow in separate domains,
setting the time-averaged values of interfacial velocity and shear as boundary
condition at the free surface. They used the data set by Fabre et al. (1987) as test
case for rectangular channel flows. In a validation study for a preliminary version of
the NEPTUNE_CFD code, Yao et al. (2003) conducted 2D simulations of the
experiments by Fabre et al. (1987) as one of three test cases. They report a
qualitatively good agreement of the calculated profiles of velocity, turbulence kinetic
energy, and turbulent shear stresses for the cases with zero and medium gas velocity.
But some quantitative deviations occur. In the case with high gas velocity (case 400),
the code fails in predicting the turbulence parameters, which they account to the
inability of the 2D model to capture the transverse flow reported for that experiment.
Terzuoli et al. (2008) used the data set as test case for a cross-code comparison of
three different CFD codes and to validate the free surface flow models of the
respective codes. They compared the scientific code NEPTUNE_CFD and the
commercial codes ANSYS CFX and FLUENT. By comparing 2D and 3D simula-
tions with the experiments, they found that three-dimensional effects should not be
neglected. Furthermore, they pointed out the fundamental role of the drag modeling
at the free surface. A series of five different experiments were used by Coste et al.
(2012) for the validation of the NEPTUNE_CFD code, the experimental cases
250 and 400 from Fabre et al. (1987) being among them. They were able to achieve
a good agreement of their numerical data with the experimental data for velocity and
turbulence in case 250. For case 400, they found a significant deviation between their
simulations and the experiment, which they account to the inability of the
NEPTUNE_CFD code to predict the transverse flows occurring in case 400.
In general, CFD simulations for free surface flows require the modeling of the
non-resolved scales. For modeling of interfacial transfers, it is necessary to select the
adequate interfacial transfer models and to determine the interfacial area. The
numerical solution can resolve the statistically averaged motion of the free surface
(including waves) which may not be too small relatively to the channel height and to
the characteristic length of the spatial discretization. However, the detailed structure
of interacting boundary layers of the separated continuous phases and surface ripples
cannot be resolved. Instead, its influence on the average flow must be modeled.
Non-resolved small-scale structures of the interface have influence on mass,
momentum, and heat transfer between the phases. The type of required models
depends on the general modeling approach used. To model the momentum transfer,
e.g., in the frame of the two-fluid model, the correlations for the interfacial drag are
used. In the past, due to the lack of appropriate models, often drag correlations valid
for bubbly flows or correlations developed for 1D codes were used to simulate the
4 T. Höhne
interfacial momentum transfer at the free surface. Such approaches do not properly
reflect the physics of the phenomena.
From this point of view, in the framework of the two-field model, it is interesting
to consider, close to the interface, an anisotropic momentum exchange between
liquid and gas. This is done for the algebraic interfacial area density (AIAD) model
(Höhne and Vallée 2010; Höhne et al. 2011) which allows using different models to
calculate the drag force coefficient and the interfacial area density for the free surface
and for bubbles or droplets.
A further step of improvement of modeling the turbulence is the consideration of
non-predicted free surface waves or so-called sub-grid waves that means waves
created by Kelvin-Helmholtz instabilities that are smaller than the grid size. So far in
the present code versions, they are neglected. However, the influence on the turbu-
lence kinetic energy of the liquid side can be significantly large. A region of marginal
breaking is defined according to Brocchini and Peregrine (2001). In addition,
turbulence damping functions should cover all the free surface flow regimes, from
weak to strong turbulence.
Modeling Free Surface Flows
The CFD Approaches Applicable to Free Surface Flow
The three main types of two-phase CFD, namely, the RANS approach, the space-
filtered approaches (such as LES methods), and the pseudo-DNS approaches, are in
principle applicable to free surface flow (see Bestion 2010a, b). Table 1 shows the
main characteristics of these methods. If only two continuous fields (continuous
liquid and continuous gas) exist in the flow without any bubble below the free
surface and without any droplet in the gas flow, a one-fluid approach (homogeneous
model) is applicable together with an interface tracking method (ITM) to predict the
free surface.
Since there may be some bubble entrainment below the free surface, a two-fluid
approach was also used to be able to deal with various types of interface configu-
rations including both large interfaces (free surface) and interface of dispersed fields
(bubbles, droplets). Detailed derivation of the two-fluid model can be found in the
book of Ishii and Hibiki (2006).
On both sides of the free surface, shear layers are expected which require a
specific attention since complex phenomena with turbulent transfers coupled to
possible interfacial waves take place. It was found necessary to be able to track the
interface position in order to treat this zone in a similar way as a wall boundary layer
using wall functions. When trying to use a two-fluid approach, the development of
an interface recognition method was found necessary.
The AIAD method belongs to the third time and space filtering type. Because the
model can be directly applied for industrial cases, it is classified as a macroscale
model (Fox 2013). A different approach in this group, for instance, is done in the
NEPTUNE code (see Coste et al. 2007; Coste and Laviéville 2009).
Table 1 Flow rates and Case 250

resulting mean liquid level
and superficial velocities of Liquid flow rate V_ L in l/s 3.0
the experimental case Gas flow rate V_ G in l/s 45.4
250 (Fabre et al. 1987) Mean liquid level hmean in mm 38
Liquid superficial velocity Ws,L in m/s 0.395
Gas superficial velocity Ws,G in m/s 3.66
Fig. 1 Different morphologies under slug flow conditions
Algebraic Interfacial Area Density Model
Figure 1 shows different morphologies that occur under slug flow conditions.
Separate models are necessary for droplets or bubbles and separated continuous
phases (interfacial drag, etc.).
The basic idea of the AIAD model is:
• The interfacial area density allows the detection of the morphological form and
the corresponding switching of each correlation from one object pair to another.
• It provides a law for the interfacial area density and the drag coefficient for a full
range of phase volume fractions from no liquid to no gas (Fig. 2).
• The model improves the physical modeling in the asymptotic limits of bubbly and
droplet flows.
• The interfacial area density in the intermediate range is set to the interfacial area
density for free surface (Fig. 2).
The approach used in the AIAD model is to define blending functions depending
on the volume fraction that enable switching between the morphologies of dispersed
droplets, dispersed bubbles, and the free surface. Based on these blending functions,
different equations for the interfacial area density and the drag coefficient can be
applied according to the local morphology. The blending functions are defined as
Eqs. 1 and 2 for droplet and bubble region, respectively, and Eq. 3 for the free
surface:
h i1
f D 1 þ eaD ðαL αD, limit Þ (1)
6 T. Höhne
Fig. 2 Air volume fraction

and corresponding
morphologies/models
h i1
f B ¼ 1 þ eaB ðαG αB, limit Þ (2)
f FS ¼ 1 f D f B (3)
with aD and aB being the blending coefficients for droplets and bubbles, respec-
tively, and αD,limit and αB,limit the volume fraction limiters. In the simulations
presented here, values of aD = aB = 70 and αD,limit = αB,limit = 0.3 were used. For
all model coefficients, same values were used as in previous studies (Höhne and
Vallée 2010; Höhne 2013). They were chosen independent of the actual geometry
and flow regime and no tuning of the AIAD model was done for the work presented
here. The threshold value αB,limit = 0.3 is a critical volume fraction before the
coalescence rate increases sharply and is verified by experiments in both vertical
and horizontal flows. Published data agree that bubbly flow rarely exceeds a gaseous
volume fraction of about 0.25–0.35 when the transition to resolved structures occurs
(Griffith and Wallis 1961; Taitel et al. 1980; Murzyn and Chanson 2009). Parameter
studies also indicated that the model is not very sensitive toward a change of the
blending function parameters. In Fig. 3, the blending function for bubbles is plotted
against the gas volume fraction for different values of aB and αB,limit.
For simplicity, bubbles and droplets are for now assumed to be of spherical shape,
with a constant diameter of dB and dD, respectively. Non-drag forces in the regions of
dispersed flow are neglected. The resulting formulation for the interfacial area
density for droplets, AD, is given by
6αL
AD ¼ (4)
dD
in which αL is the volume fraction of the liquid phase. The IAD for bubbles, AB, is
formulated analogous. The IAD of the free surface, AFS, is defined as the magnitude
of the gradient of the liquid volume fraction αL, as given in Eq. 5, with n being the
normal vector of the free surface:
Fig. 3 Blending function fB plotted for different values of aB and αB,limit
@αL
AFS ¼ j∇αL j ¼ (5)
@n
The local interfacial area density A is then calculated as the sum of Aj, weighted
by the blending functions fj:
X
A¼ A,
j fj j
j ¼ FS, B, D (6)
Modeling the Free Surface Drag
In the general case of a two-phase flow, there is a velocity difference between the
fluids, which is commonly called slip velocity. In contradiction to most volume of
fluid (VOF) methods where only one velocity field is present, in the multi-fluid
framework, each phase has its own velocity and turbulence model. Thereby, a drag
force is induced at the phase boundary that is acting on both phases. The drag force
can be correlated with the slip velocity U, the fluid density ρ, the surface area a, and
the dimensionless drag coefficient CD. For geometry-independent modeling, the
drag force is expressed as the volumetric force density FD and a is then replaced
by the area density. The magnitude of the drag force density is given by Eq. 7:
1
jFD j ¼ CD A ρjUj2 (7)
2
For a dispersed flow, the density of the continuous phase is used in Eq. 7. In case
of a free surface, the phase-averaged density is used, i.e.,
ρ ¼ α G ρG þ α L ρL (8)
with ρG and ρL being the density of the gas and the liquid, respectively.
In simulations of free surface flows, Eq. 7 does not represent a realistic physical
model. It is reasonable to expect that the velocities of both fluids in the vicinity of the
interface are rather similar. In Höhne et al. (2011), it is assumed that the shear stress
near the surface behaves like a wall shear stress on both sides of the interface in order
to reduce the velocity differences of both phases. It is supposed that the morphology
region “free surface” is acting like a wall, and a wall-like shear stress is introduced at
8 T. Höhne
the free surface which influences the loss of gas velocity. The components of the
!
normal vector n at the free surface are taken from the gradients of the void fraction.
To use these directions of the normal vectors, the gradients of gas/liquid veloc-
ities, which are used to calculate the wall shear stress onto the free surface, are
weighted with the components of the normal vector. From theory, shear stress is a
!
symmetric tensor τ; if we have a surface normal vector n , then the wall-like free
surface shear stress vector τW is the product of the viscous stress tensor and the
surface normal vector:
2 3 2 3
τxx τxy τxz nx
τW ¼ 4 τyx τyy τyz 5 4 ny 5 (9)
τzx τzy τzz nz
This results in the following equations:
τW, x ¼ τxx nx þ τxy ny þ τxz nz

τW, y ¼ τyx nx þ τyy ny þ τyz nz
(10)
τ W, z p
¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
τzx nx þ τzy ny þ τzz nz
τW ¼ τW, x 2 þ τW, y 2 þ τW, z 2
with nx , ny , nz being the components of the normal vector, u,v,w the compo-
nents of the velocity vector, and μ the dynamic viscosity of the fluid; it can also be
written:

@u @u @v @u @w
τW, x ¼ nx μ 2 þ ny μ þ þ nz μ þ
@x @y @x @z @x

@u @v @v @v @w
τW, y ¼ nx μ þ þ ny μ 2 þ nz μ þ
@y @x @y @z @y

@w @u @w @v @w
τW, z ¼ nx μ þ þ ny μ þ þ nz μ 2
@x @z @y @z @z
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
τW ¼ τW, x 2 þ τW, y 2 þ τW, z 2
(11)
It is assumed that the drag force Eq. 7 is equal to the wall shear stress force acting
at the free surface in the vicinity of the free surface:
FW ¼ τ i A ¼ FD (12)
As a result, the modified drag coefficient is dependent on the viscosities of both

phases, the wall-like shear stresses (local gradients of gas/liquid velocities normal to
the free surface), the mixture density, and the slip velocity between the phases:

2 αL τW , L þ αG τW , G
CD, FS ¼ (13)
ρjUj2
The AIAD model uses the following three different drag coefficients: CD,B = 0.44
(Newton) for bubbles, CD,D = 0.44 for the droplets, and CD,FS (Höhne and Vallée
2010) according to Eq. 13 for the free surface. The advantage of this definition is that
the resulting drag force is not depending on U anymore, since |U|2 is being elimi-
nated, but instead the local velocity gradients and the viscosities of both phases are
included in the calculation. This is a more physical definition for the interfacial drag
force in a shear-driven flow. The total drag coefficient for a unit volume is calculated
analogous to the interfacial area density as the weighted sum of the drag coefficients
for all morphologies:
X
CD ¼ j
f j CD, j , j ¼ FS, B, D (14)
Sub-grid Wave Turbulence
Small waves (Fig. 4) created by Kelvin-Helmholtz instabilities that are smaller than
the grid size are neglected in traditional two-phase flow CFD simulations, but the
influence on the turbulence kinetic energy of the liquid side can be significantly
large. Brocchini and Peregrine (2001) try to quantify this in the L-q diagram (Fig. 5)
pffiffiffiffiffiffi
which predicts the free surface shape as a function of the liquid turbulence q ¼ 2 k
and a length scale L. They supposed that both gravity and surface tension act at a
liquid surface so the surface behavior depends on both the turbulent Froude number
1=
Fr ¼ q=ð2gLÞ 2 and Weber number We = q2Lρ/2σ (σ is the interfacial tension
coefficient). Thus, the value of both parameters must be considered. Their effect is
discussed by seeking to delineate a critical region of parameter space between
quiescent surfaces and surfaces that break up completely.
The shaded area in Fig. 5 represents the region of marginal breaking and has been
obtained by using the two estimated values for both the critical Weber number and
the critical Froude number. So far we assume that the local length scale L can be
obtained by local surface morphology created by the larger interface structures,
which are resolved. The shaded area in the diagram also indicates the range of
variations between surface that is no longer smooth or possibly broken because of
turbulent flow.
Depending on the values of the Froude and Weber numbers, the following four
regimes can be classified:
– Weak turbulence Fr 1, We 1: The turbulence is not strong enough to cause

significant surface disturbances.
10 T. Höhne
Fig. 4 Surface instabilities at the free surface (HAWAC)
Fig. 5 Diagram of the (L,q)-plane for water (Brocchini and Peregrine 2001)
– “Knobbly” flow Fr 1, We 1: The turbulence is strong enough to deform the

surface against gravity, but its turbulent length scale is small. Surface tension
causes the surface shape to be very smooth and rounded.
– Turbulence is dominated by gravity Fr 1, We 1: Surface distortions are
primarily counters by gravity, resulting in a nearly flat free surface. The turbulent
energy is sufficient to disturb the surface at relatively small scales, leading to
small regions of waves, vortex dimples, and scars.
– Strong turbulence Fr 1, We 1: The turbulence is strong enough to counter
gravity and surface tension which is no longer sufficient to prevent the surface
from breaking up into droplet and bubbles.
Since turbulence can have different length scales at the same time, many of these
regions can occur close to each other.
For an upper critical value, Brocchini and Peregrine (2001) compared the turbu-
lent kinetic energy density per unit volume of a blob that can disturb the surface with
the energy of a surface disturbance per unit surface area (Fig. 6). It is assumed that a
blob is just able to generate a spherical drop that touches the surface when it has lost
Fig. 6 Example for the upper

and lower bound of the
transition zone
any overall motion. This leads to an estimate for the upper bound of the transition
zone of
π πσ
qu 2 g Lþ (15)
24 n 2LρL
! !
with gn ¼ g n FS . The gravity vector normally stabilizes the free surface in
horizontal flows. In cases of slug flows or vertical flows, the vector can even
destabilize the flow. Therefore, the normal vector of the free surface is also taken
into account. If gn turns into negative values, the term is destabilizing the surface.
The interface normal vector is formulated using the volume fraction gradient of the
liquid phase:
! ∇αl
n¼ (16)
j∇αl j
With this formulation, the interface normal vector is always pointing into the
liquid. The product becomes zero in the case of an interface located parallel to the
! !
gravity vector n perpendicular to g n . Additionally, it becomes negative if the
inward-directed interface normal vector is pointing into the opposite direction
!
compared to g , which is the case occurring with Rayleigh-Taylor type of instabilities.
The model surface for the lower bound is derived at the molecular level from the
continuum surface model from Shikhmurzaev (1997). Surface diffusion studies
show that the first indication of turbulence “breaking” the surface can be represented
by the creation of fresh surface by the breaking of the surface skin (Fig. 6). It is
considered as a linear down welling feature bounded by two, convex-upward
quarter-circles. The result of the lower bound of the transition region is then
12 T. Höhne
Fig. 7 Diagram of the (L,k)-

plane for water

5 π gn L ðπ 2Þσ
ql
2
þ (17)
3 2 125 5LρL
We assume that we must add the potion of turbulent kinetic energy created by the
small waves at the liquid side which is so far missing in the simulation to the overall
turbulent kinetic energy (see Fig. 5 shaded area):

kSWT ¼ 0:5 q2u q2l (18)
The result is depended on the local length scale L shown in Fig. 7.

The consequence of the specific turbulent kinetic energy k is prescribed in the
following source term:
2 @Ui
Pk, SWT ¼ f FS ρ k (19)
3 @xi L
where the gradients of the local velocities and the liquid density are present and
which are added to the total turbulent kinetic energy (k-ω Model, Wilcox 1994):

@ ðρkÞ μt
þ ∇ ðρUkÞ ¼ ∇ μþ ∇k þ Pk β0 ρkω þ Pk, SWT (20)
@t σk
The term Pk, SWT in Eq. 20 is blended only in the vicinity of the free surface.
Turbulence Damping
Damping of turbulence as the interface is approached – on both sides – was found

vital for the modeling of interface deformations under strong gas-side shear (Reboux
et al. 2006). The phenomenon is similar to wall-turbulence decay, where it is known

and rigorously derived that the eddy viscosity scales with the cube of the distance to
the wall. The same has been found via DNS for interfacial flows, on the gas side
which obviously perceives the interface as a solid wall (Fulgosi et al. 2003); it was
also found that the liquid-side eddy viscosity scales with the square of the interface
distance. In fact, it was found that when no damping is introduced, a spurious
amount of eddy viscosity is generated at the interface, which tends to smear high-
frequency surface instabilities, like wrinkling and fingering, and introduces strong
errors in estimating the interfacial shear, which is the most important ingredient for
mass transfer modeling, as explained in the corresponding section. Furthermore,
without any special treatment of the free surface, the high velocity gradients at the
free surface, especially in the gaseous phase, generate levels of turbulence that are
too high throughout the two-phase flow when using differential eddy viscosity
models like the k-ε or the k-ω model. Therefore, a certain amount of damping of
turbulence is necessary in the region of the interface, because the mesh is too coarse
to resolve the velocity gradient in the gas phase at the interface.
A few empirical models have been suggested, which address the turbulence
anisotropy at the free surface; see among others Celik and Rodi (1984). However,
no model is applicable for a wide range of flow conditions, and all of them are
nonlocal: they require, for example, explicit specification of the liquid layer thick-
ness, of the amplitude and period of surface waves, etc. Direct and large eddy
simulations of turbulent multi-material flow have been applied to model surface
waves. Specifically, Reboux et al. (2006) used DNS to quantify the damping of
turbulence approaching the interface and incorporated this knowledge into the
damping of LES turbulence models (Liovic and Lakehal 2007a, b). Nourgaliev
et al. (2008), Boeck and Zaleski (2005), and Coward et al. (1997) reported repre-
sentations of surface instabilities that were obtained by DNS.
For the two-fluid formulation, Egorov (2004) proposed a symmetric damping
procedure. This procedure provides a solid wall-like damping of turbulence in both
gas and liquid phases. It is based on the standard ω-equation, formulated by Wilcox
(1994) as follows:
@ α i ρ i ωi
ðαi ρi ωi Þ þ ∇ ðαi ρi Ui ωi Þ ¼ α τt, i S_ i αi β ρi ω2i
@t ki

þ ∇ αi μi þ σ ω μt, i ∇ωi þ SD, i
i ¼ g, l
(21)
where α = 0.52 and β = 0.075 are the k-ω model closure coefficients of the
generation and the destruction terms in the ω-equation, σω = 0.5 is the inverse of the
turbulent Prandtl number for ω, τt is the Reynolds stress tensor, and S_ is the strain-rate
tensor. Asymptotic analysis in the viscous sublayer near the wall shows that the k-ω
model properly describes the turbulence damping in the internal part of the boundary
layer, if the following Dirichlet boundary condition is specified for ω on the wall
(subscript W):
14 T. Höhne
6μi
ωw , i ¼ B (22)
βρi Δn2
Here Δn is the near-wall grid cell height, and B is a coefficient. In order to mimic
the effect of this boundary condition near the free surface, the following source terms
have been introduced in the right-hand side of the gas and liquid phase ω-equations:
2
6μi
SD, i ¼ a Δy β ρi B , i ¼ g, l (23)
βρi Δn2
Here A is the interface area density, Δy the local characteristic lengths scale, Δn is
the typical grid cell size across the interface, and ρi and μi are density and viscosity of
the corresponding phase i. The factor A activates this source term only at the free
surface, where
it cancels the standard ω-destruction term of the ω-equation
αi βρi ω2i and enforces the required high value of ωi and thus the turbulence
damping.
As a consequence, there is a sink term and a source term in the k equation and
probably the effect might cancel it out. Nevertheless, the physical effects of sub-grid
turbulence and turbulence damping due to high velocity gradients are not the same
and both should be considered.
Verification and Validation
Horizontal Wavy and Stratified Flow: Fabre Channel
The first validation case of the AIAD model is an experimental data set from Fabre
et al. (1987). In a quasi-horizontal channel with rectangular cross section, experi-
ments with air and water were conducted. The flow regime investigated was smooth
cocurrent stratified flow. The channel consists of six segments made of Plexiglas,
with a total length of lc = 12.92 m and an inner cross section of height and width of
hc = 0.1 m and wc = 0.2 m, respectively. Water flows into the channel out of a
tranquilization tank. Air and water inlet are separated by a floating Plexiglas sheet.
At the channel outlet, the water is discharged into a tank. Air and water are
recirculated in separate loops.
In Fabre et al. (1987), measurement data along the vertical axis at z = 9.1 m from
the inlet are reported. The instantaneous interface height was measured with verti-
cally mounted capacitance wires. Velocity in gas was measured using hot-wire
anemometry and velocity in liquid was measured with laser Doppler anemometry.
A more detailed description of the experimental setup and measurement techniques
can be found in Fabre et al. (1987).
The experiment which has been investigated in this work is case 250. The liquid
flow rate was V_ L ¼ 3:0 l=s. The gas flow rate was adjusted at V_ L ¼ 45:4 l=s for case
250. The resulting mean liquid levels and corresponding superficial velocities are
Table 2 Number of nodes and cell width in each spatial direction of the meshes used in the grid
study
Mesh No. of nodes Δx [mm] Δy [mm] Δz [mm]
Mesh 1 237,446 4.76 4.77 10.00
Mesh 2 473,946 4.76 4.77 5.00
Mesh 3 1,174,845 3.03 2.94 5.00
summarized in Table 2. The specific case was chosen, since case 250 was reported to
be smooth stratified flow.
CFD Setup
An unstructured mesh with hexahedral cells was used. The AIAD model was
implemented into the commercial CFD code ANSYS CFX 14.5 by CFX Command
Language (CCL). The data for the comparison with the experiments were taken at a
vertical center line 2.0 m from the inlet. To generate appropriate inlet boundary
conditions, velocity profiles were taken from preliminary 2D simulations of the full-
length channel at 7.1 m from the inlet. The water level measured in the experiments
was specified at the inlet and the boundary velocity profiles of air and water were
scaled accordingly. The same water level was specified as initial condition for the
whole domain, together with the corresponding gas and liquid superficial velocities
and a hydrostatic pressure. As outlet boundary condition, a constant pressure outlet
was defined, with the volume fraction function and the hydrostatic pressure also used
for the initial conditions. Figure 8 shows an illustration of the domain at the initial
state.
The channel inclination was neglected in this study, since it is very low. All
simulations were run in transient mode. In a grid study, three meshes were compared.
The number of nodes and the cell widths in all three spatial directions is given in
Table 3. For the coarsest mesh 1, convergence was not achieved and no results are
presented. For mesh 2 and mesh 3, the profiles of the velocity w in main flow
direction for air and water are shown in Figs. 9 and 10, respectively. There is no
qualitative difference between the profiles and the quantitative deviation is suffi-
ciently small to stop the grid study. For further calculations, the finest mesh 3 was
used, since it showed better convergence behavior and allowed wider time steps.
Results and Discussion

The focus of the analysis is put on the profiles of the velocity w in flow direction and
the profiles of the turbulence kinetic energy k, because in previous validation studies
of the AIAD model, these parameters were not available. The time course of the
water level was also evaluated. Calculations with different turbulence models were
done, which confirm the necessity of turbulence damping in free surface flows.
16 T. Höhne
Fig. 8 Illustration of the initial volume fraction and the inlet velocity profile in case 400 (detail
view)
Fig. 9 Vertical profiles of the streamwise velocity w in air
Smooth Stratified Flow

The importance of turbulence modeling at the free surface was shown, e.g., by
Fulgosi et al. (2003), so it is interesting and important to know how the use of
different turbulence models influences the solution and how the AIAD turbulence
modeling performs in comparison to standard turbulence models. To get an insight
into this, an assessment of turbulence models was done. The AIAD k-ω model with
turbulence damping and small wave turbulence is compared to three reference
models. The models used as reference were k-ε (Launder and Sharma 1974),
unmodified k-ω (Wilcox 1998), and shear stress transport (SST) (Menter 1994), all
of which can be regarded as established standard models. All other parameters of the
setup and the AIAD model were left unchanged.
The time-dependent water levels at z = 2 m are shown in Fig. 11. Since no flow
history data is reported in Fabre et al. (1987), the simulation results are compared to
the time-averaged water level reported for the experiment. For all setups, an initial
Fig. 10 Vertical profiles of

the streamwise velocity w in
water
Fig. 11 Development of the

water level from the
simulations of case 250 for
different turbulence models
compared to the experimental
mean water level from Fabre
et al. (1987)
decrease of the water level occurs before a quasi-steady state is reached. There is
hardly any deviation in the course of the water level between k-ε and k-ω, which
both show the poorest performance among the models used. Also the SST model
predicts water levels that are much lower than in the experiment. For the k-ω model
with turbulence damping, as used in the AIAD model, good agreement with the
experiment is achieved in the quasi-steady state, but minor deviations leave room for
18 T. Höhne
Fig. 12 Scaled vertical

profiles of the streamwise
velocity w in air
further improvement. The results clearly show the necessity of turbulence damping
at the free surface in order to correctly predict the liquid holdup.
The ability of the AIAD model to qualitatively predict the key features of free
surface flows as well as liquid holdup was already demonstrated in the previous
work. But in those studies, measurement data on the kinematic structure of the flows
was not available from the experiments. Therefore, the analysis of the velocity and
turbulence data was the most important part of the work presented here. Figures 12
and 13 show the vertical profiles of the streamwise velocity w at the measurement
location for the different turbulence models in comparison to the experimental data.
To restore comparability of the velocity profiles despite the different water levels, the
velocity values are scaled with the superficial velocityws, i ¼ V_ i =Aave, i of the phase i,
where Aave,i is the average cross-sectional area occupied by phase i. The height is
scaled with the average water level hmean in a way that for water hL ¼ hL =hmean and
for air hG ¼ ðhG hmean Þ=ðhc hmean Þ.
In Fig. 12, especially the non-damped models expose a flattened profile in
comparison to the experimental data, with the velocity maximum shifted to the
upper part of the channel close to the wall. This trend can also be observed for the
AIAD k-ω model in a weaker form, but in that case, the qualitative agreement with
the experiment is better in the upper part of the channel. Nonetheless, the trend in the
free flow region is similar to the other turbulence models, but a steeper gradient is
observed close to the free surface.
In water, a satisfying agreement of the scaled velocity profiles is achieved
between the SST model and the experiment, as can be seen in Fig. 13. The
non-damped k-ω and k-ε models also deliver good results in the free flow region,
Fig. 13 Scaled vertical

profiles of the streamwise
velocity w in water
but predict too low velocities in the vicinity of the free surface. When using the
AIAD turbulence modeling, the velocity close to the free surface is calculated too
high. This inevitably distorts the profile also in the lower part of the channel. In order
to further investigate this problem, the turbulence parameters are analyzed in the
following.
The numerical coupling between the turbulence model and the conservation
equations in RANS-based eddy viscosity turbulence models is done via the turbu-
lence viscosity or eddy viscosity μt, which is a function of turbulence kinetic energy
and turbulence dissipation rate. For the k-ω model and the k-ε model, the eddy
viscosity is expressed by Eqs. 24 and 25, respectively, with Cμ = 0.09 being a
constant:
k
μt, kω ¼ ρ (24)
ω
k2
μt, ke ¼ Cμ ρ (25)
e
In the SST model, the eddy viscosity is expressed by Eq. 26:
a1 k
μt, SST ¼ ρ (26)
maxða1 ω, ΩF2 Þ
with a1 = 5/9 being a SST model coefficient, Ω the absolute value of the vorticity,
and F2 a blending function of the SST model:
20 T. Höhne

F2 ¼ tan h arg22 (27)
pffiffiffi
2 k 500v
arg2 ¼ max 0 , 2 (28)
β ωy y ω
with y being the wall distance, v the kinematic viscosity of the fluid, and
β0 = 0.09 an SST model coefficient. Eddy viscosity and molecular viscosity of the
fluid μ are added up to the effective viscosity μeff = μ + μt.
Since the eddy viscosity is a numerical construct, no experimental data is
available to compare with the simulation results. But an analysis of the eddy
viscosity can give an insight on the influence of the turbulence modeling on the
velocity field and the free surface position. The effective viscosity acts as propor-
tionality factor between the viscous force Fμ and the velocity gradient normal to the
flow direction @w/@h, as expressed in the standard shear stress in Eq. 29, with
τeff = Fμ/A:
@w
τeff ¼ μeff (29)
@h
Therefore, if the flow is incompressible and driven by a constant force, an
increase of the effective viscosity has the effect of a decrease in the velocity gradient
and vice versa. Since the molecular viscosity is assumed to be constant and is orders
of magnitude smaller than the turbulent viscosity in the case examined here, the latter
one is the governing parameter of the effective viscosity.
Figures 14 and 15 show the vertical profiles of the calculated eddy viscosity in air
and water, respectively. For better comparability, μt is also correlated with h*. At the
walls, all models are more or less in agreement, but the deviations between the
models grow drastically toward the free surface. There the unmodified k-ω and k-ε
model reach the maximum μt, while with SST and AIAD k-ω the maximum of μt is
located in the free flow region and toward the free surface μt decreases again. These
trends are observed in both phases. From these observations, it can be explained why
there are such strong deviations between the velocity profiles and free surface
positions predicted by the different turbulence models. According to Eq. 29, an
increase in turbulence viscosity has to result in a decrease of the velocity gradient.
The effect of this is a smoothing of the velocity profile, as can be observed in Figs. 12
and 13.
The velocity gradients also strongly affect the calculation of the free surface drag,
since the free surface drag force resulting from the combination of Eqs. 7 and 13 is
only depending on the area density and the free surface shear stresses. The latter ones
again are depending on the velocity gradients at the free surface according to Eq. 11.
The result of the problems discussed above is therefore a reduction in the calculated
drag force, which forces a reduction in the velocity difference between the phases.
The only remaining degree of freedom is the interface position, with the result of a
shifting of the free surface toward the slower phase, which is the water in this case.

the turbulent viscosity ηt in air

the turbulent viscosity ηt in
water
The above considerations in combination with the analysis of the eddy viscosity
explain why the prediction of the holdup is poor when using the non-damped
turbulence models. With the turbulence damping at the free surface used in the
22 T. Höhne
Fig. 16 Profiles of the turbulence kinetic energy k in air
AIAD model, steeper velocity gradients at the free surface are allowed, and this
effect is weakened. This explains the better performance of the AIAD model in
predicting the liquid holdup.
To get a deeper insight into the turbulence structure of the flow, also the turbu-
lence kinetic energy is analyzed, for which measurement data is available. The
vertical profiles of k in the gas phase are plotted in Fig. 16. For comparability, k is
scaled with ws2 and correlated with h*. It is obvious that with all non-damped
turbulence models, k is strongly overestimated, up to one order of magnitude close
to the free surface in case of the unmodified k-ω. According to Egorov (2004), this is
a well-known problem and was the initial motivation for the development of
turbulence damping functions. The k-ω model with damping term predicts a turbu-
lence kinetic energy profile much closer to the experimental one. But still there are
deviations, especially in the region below h* = 0.5.
For the liquid phase, the vertical profiles of the turbulence kinetic energy are
shown in Fig. 17. The situation is comparable to the gas phase; the standard
turbulence models fail in predicting k, while the k-ω with damping is much closer.
Noteworthy is the strong increase of k close to the gas-liquid interface when using
damped k-ω and SST. Unfortunately, there is no measurement data available above
h* = 0.9, so it cannot be determined whether this is an artifact of the turbulence
modeling or a realistic effect. From the examination of the turbulence parameters, it
can be deduced that the free surface turbulence damping has an effect not only in the
small region where the damping functions are blended in but also influences a major
part of the core flow. This effect is beneficial, because it leads to a more realistic
calculation of the turbulence kinetic energy. But the problem remains that the
performance of the k-ω model away from the walls is not optimal, which is a well-
known drawback (Menter 1994).
Future development of the AIAD model should therefore focus on the turbulence
modeling. It has to be determined more in detail, how the free surface turbulence
modeling takes effect on the core flow and on the free surface drag. An asymmetric
turbulence damping should be considered to overcome the weaker performance of
the model in the liquid phase, which was up to now not done for reasons of
numerical stability. Also a detailed assessment of the influence of the small wave
Fig. 17 Profiles of the

turbulence kinetic energy k in
water
turbulence is necessary. One prerequisite for these tasks are measurement data on the
velocity and turbulence fields close to the free surface, which are not available in
Fabre et al. (1987), as in most experimental studies on stratified flow, due to
limitations in the measurement techniques. A solution might be the analysis of
DNS data as complement to experimental studies. As an alternative approach, the
damping functions should be implemented with the SST model, since the k-ω model
is known to create problems in the free flow region (Menter 1994).
Horizontal Slug Flow: HAWAC Channel
The horizontal air/water channel (HAWAC) (Fig. 18) is devoted to cocurrent flow
experiments. A special inlet device provides defined inlet boundary conditions by
separate injection of water and air into the test section. A blade separating the phases
can be moved up and down to control the free inlet cross section for each phase,
which influences the evolution of the two-phase flow regime. The cross section of
this channel is smaller than the channel used in an earlier study described by Vallée
et al. (2008): its dimensions are 100 30 mm2 (height width). The test section is
about 8 m long, and therefore the length-to-height ratio L/h is 80. In terms of the
hydraulic diameter, the dimensionless length of the channel is L/Dh = 173.
The inlet device (Fig. 19) is designed for separate injection of water and air into
the channel. The air flows through the upper part and the water through the lower
part of this device. As the inlet geometry introduces many perturbations into the flow
(bends, transition from pipes to rectangular cross section), four wire mesh filters are
mounted in each part of the inlet device. The filters are made of stainless steel wires
24 T. Höhne
Fig. 18 Schematic view of the horizontal channel with inlet device for a separate injection of water
and air into the test section
Fig. 19 The HAWAC inlet device
with a diameter of 0.63 mm and have a mesh size of 1.06 mm. The wire mesh filters
are used to provide homogenous velocity profiles at the test section inlet. Moreover,
the filters produce a pressure drop that attenuates the effect of the pressure surge
created by slug flow on the fluid supply systems.
Air and water come in contact at the final edge of a 500 mm long blade that
divides both phases downstream of the filter segment. The free inlet cross section for
each phase can be controlled by inclining this blade up and down. In this way, the
perturbation caused by the first contact between gas and liquid can be either
minimized or, if required, a perturbation can be introduced (e.g., hydraulic jump).
Both, filters and the inclinable blade, provide well-defined inlet boundary conditions
for the CFD model and therefore offer very good validation possibilities. Optical
measurements were performed with a high-speed video camera.
CFD Setup
The new approach was implemented via the command language CCL into ANSYS
CFX-14 (ANSYS 2012). An Euler-Euler multiphase model using fluid-dependent
RANS k-ω turbulence models was applied. The high-resolution discretization
scheme was used for convection terms in the equations. For time integration, the
fully implicit second-order backward Euler method was applied with a constant time
step of dt = 0.001 s and a maximum of 30 coefficient loops per time step.
Convergence was defined in terms of the RMS values of the residuals, which was
less than 104 most of the time. An initial water level of half of the channel was
assumed for the entire model length. The inlet conditions are given by both super-
ficial velocities; a pressure distribution is set at the channel outlet.
The grid consists of 1.2 106 hexahedral elements. A slug flow experiment at a
water velocity of 2.0 m/s and an air velocity of 10.0 m/s was chosen for the CFD
calculations. These velocities correspond to the following correlation proposed by
Mishima and Ishii (1980) for the evaluation of the onset of slugging in a horizontal
pipe:
rffiffiffiffiffiffiffiffiffiffiffi
gy0 ρL
cG cL ¼ 0:487 (30)
ρG
where y0 = 50 mm is the height of the gas flow part of the channel (the distance of
the interface to the top) and ci the critical velocity of phase i.
In the experiment, the inlet blade was in the horizontal position. Accordingly, the
model inlet was divided into two parts: in the lower half of the inlet cross section,
water was injected with air injected in the upper half. To allow for upstream flow
development, the inlet blade was modeled as shown in Fig. 20a. An initial water
level of y0 = 50 mm was assumed (Fig. 20b). In the simulation, both phases have
been treated as isothermal and incompressible, at 25 C and at a reference pressure of
1 bar. A hydrostatic pressure was assumed for the liquid phase. At the inlet, the
turbulence properties were set using the “medium intensity and eddy viscosity ratio”
option of the flow solver. This is equivalent to a turbulence intensity of 5% in both
phases. The inner surface of the channel walls are defined as hydraulically smooth
with a nonslip boundary condition applied to both gaseous and liquid phases. The
channel outlet was modeled with a pressure-controlled outlet boundary condition.
The parallel transient calculation of 15.0 s of simulation time on eight processors
took 21 CPU days.
Verification
To first verify the damping procedures, this slug flow experiment of the HAWAC
channel was used. The horizontal two-phase flow was modeled with and without the
damping procedures. In addition, a single-phase flow of the gaseous phase was
modeled using the upper part (50%) of the channel, and a moving wall with constant
velocity of the liquid phase of 2 m/s at the lower wall boundary was utilized to mimic
the free surface. Figure 21 shows the verification of the k-ω turbulence model
damping procedures using a HAWAC experiment. On the left side, a case is
shown with gas velocity 10 m/s and liquid velocity 2 m/s with and without damping
functions. On the right side, a single-phase case with only air flowing over a moving
wall with 2 m/s is displayed. Figure 22 shows the results of the gas velocity field over
the channel height. The simulations used the k-ω turbulence model with and without
damping procedures inside the AIAD model framework in comparison to single
gas-phase flow. The damping functions have a strong effect on the gas velocity field.
26 T. Höhne
Fig. 20 Model and initial conditions of the volume fractions
Moving Wall 2m/s

Case with gas velocity 10 m/s and liquid Single phase case with only air
velocity 2 m/s with and without damping flowing over moving wall
functions
Fig. 21 Verification of the k-ω turbulence model damping procedures
16.00
14.00
12.00
K omega no Damping
Air Velocity [m/s]
Single Phase
10.00
k omega Damping
8.00
6.00
4.00
2.00
0.00
0 0.02 0.04 0.06 0.08 0.1
Channel height [m]
Fig. 22 Results of the gas velocities over the channel height, k-ω turbulence model with and
without damping procedures using the AIAD model in comparison with single gas-phase flow
Fig. 23 Measured picture sequence at JL = 1.0 m/s and JG = 5.0 m/s with Δt = 50 ms (depicted
part of the channel: 0–3.2 m after the inlet)
AreaDensVar [m^-1]
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 1.0
00 00 00 00 00 00 00 00 00 00 0
0e 0e 0e 0e 0e 0e 0e 0e 0e 0e 0e
+0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0
00 01 01 01 01 01 01 01 01 01 02
Fig. 24 Calculated picture sequence at JL = 1.0 m/s and JG = 5.0 m/s with Δt = 50 ms (depicted
part of the channel: 0.4–3.6 m after the inlet)
It is obvious that without modification of the turbulence at the free surface, the
velocity fields of a horizontal two-phase flow are not predicted correctly.
Results and Discussion

Picture sequences in Figs. 23 and 24 compare predictions of the phase distribution
from the CFD calculation and comparable camera frames of slugging behavior
observed in the experiment. In both cases, a slug is generated. The sequences
show that the qualitative behavior of the creation and propagation of the slug is
similar in both the experiment and the CFD calculation.
28 T. Höhne
Air at 25 C.Volume Fraction Air at 25 C.Volume Fraction

1.000e+000 1.000e+000
7.500e-001 7.500e-001
5.000e-001 5.000e-001
2.500e-001 2.500e-001
1.000e-015 1.000e-015
Y Y
X X
time = 8.97 [s] time = 9.21 [s]
8.97 s after start of simulation 9.21 s after start of simulation

Air at 25 C.Volume Fraction Air at 25 C.Volume Fraction
uboot Default uboot Default
1.000e+000 1.000e+000
7.500e-001 7.500e-001
5.000e-001 5.000e-001
2.500e-001 2.500e-001
1.000e-015 1.000e-015
Y Y
X X
time = 9.52 [s] time = 9.91 [s]
9.52 s after start of simulation 9.91 s after start of simulation
Fig. 25 Calculated picture sequences at JL = 1.0 m/s and JG = 5.0 m/s with isosurface at αL = 0.5,
development of waves and slug formations in the channel.
In addition, Fig. 25 shows the development of the slug. These slugs are induced
only by instabilities generated in the simulation, and the single effects leading to slug
formation that can be simulated in this model are:
• Instabilities and small waves randomly generated by the interfacial momentum

transfer. As a result, bigger waves are generated.
• The waves can have different velocities and can merge together.
• Bigger waves roll over smaller waves and can close the channel cross section.
Fig. 26 Sub-grid wave turbulence [W m3], development of a slug formation in the channel
The slug formation can also be induced by a perturbation appearing downstream

at the inlet blade which separates the water and the air flow. In the simulation as well
as in the experiment, this perturbation is induced from the blade lip. Figure 26 shows
the influence of the sub-grid wave turbulence during a slug development in the
channel. In order to extract quantitative information, a local position was chosen
1.5 m after the blade position in the middle of the channel over the channel height.
The sub-grid wave turbulence exists only in the area of the free surface and follows
the slug formations. At the wavy front and back of the slugs, the value of the sub-grid
wave turbulence is the highest in the channel.
The turbulence intensity with and without sub-grid wave turbulence is displayed
in Fig. 27. The turbulence intensity Tu is calculated according to Eq. 31:
pffiffiffiffiffiffiffiffiffiffiffi
2=3ki
Tu ¼ with i ¼ g, l (31)
ui
with the additional source term of the sub-grid wave turbulence in the water
turbulence kinetic energy the value is slightly higher in the vicinity of the interface.
The slug frequency analysis was done using fast Fourier transform (FFT). The
power spectral density (PSD), which describes how the power of a signal or time
series is distributed over different frequencies, is used. The position of 2.5 m away
from inlet blade in the middle of the channel (0.015 m, 0.05 m), where waves are
generated, was utilized to describe the change of water level in the channel. A
characteristic slug frequency of around 2.0 Hz is seen Fig. 28, which corresponds
roughly to the experimental value of approximately 2.4 Hz.
Summary and Conclusions
Stratified two-phase flows are relevant in many industrial applications, e.g., pipe-
lines, horizontal heat exchangers, and storage tanks. Special flow characteristics as
flow rate, pressure drop, and flow regimes have always been of engineering interest.
The numerical simulation of free surface flows can be performed using phase-
averaged multi-fluid models, like the homogeneous and the two-fluid approaches,
30 T. Höhne
0.10
0.09 Air SWT
0.08 Water SWT
0.07 Air
channel hight [m]
Water
0.06
0.05
0.04
0.03
0.02
0.01
0.00
1.0E-03 1.0E-02 1.0E-01 1.0E+00
turbulence intensity [-]
Fig. 27 Comparison of air/water turbulence intensity [] with and without sub-grid wave
turbulence
0,01
0,0001
Power Spectral Density [(ˆ2)/Hz]
1e-06
1e-08
1e-10
1e-12
1e-14
0,1 1 10 100
Frequency [Hz]
Fig. 28 Power spectral density (PSD) of water level at the middle of the channel, CFD simulation
or non-phase-averaged variants. The approach shown in this paper within the

two-fluid framework is the algebraic interfacial area density (AIAD) model. It allows
the macroscopic blending between different models for the calculation of the
interfacial area density and improved models for momentum transfer in dependence
on local morphology. An approach for the drag force at the free surface was
introduced. The model improves the physics of the existing two-fluid approaches
and is already applicable for a wide range of industrial two-phase flows. A further
step of improvement of modeling the turbulence was the consideration of sub-grid
wave turbulence that means waves created by Kelvin-Helmholtz instabilities that are
smaller than the grid size. A first CFD validation of the approach was done for an
adiabatic case of the HAWAC channel. In addition, an experimental test case of the
Fabre channel was simulated by using the AIAD model as closure for stratified
gas-liquid flows. For the first time, the AIAD model was validated with a detailed set
of velocity and turbulence data, and its performance was compared with three
standard turbulence models. It was confirmed that with the AIAD model, it is
possible to predict key flow features of stratified flow like waviness of the free
surface and liquid holdup with good accuracy, where the established turbulence
models fail. In predicting the streamwise velocity and turbulence kinetic energy in
the gas phase, the AIAD model was also superior to the standard turbulence models,
but there are still existing deviations to the experimental data. In calculating the
liquid velocity, the SST turbulence model is showing the best performance, although
it again fails in predicting the turbulence kinetic energy, as well as the other standard
turbulence models. The results show that a sophisticated modeling of drag and
turbulence at the free surface is necessary in order to correctly model stratified
flow. The AIAD model is fulfilling these requirements from a qualitative point of
view. But further research is necessary on the turbulence damping and its effect on
the free surface drag. More verification and validation of the approach is necessary –
more CFD-grade experimental data are required for the validation.
References
ANSYS CFX, User Manual. (Ansys, 2012)
D. Bestion, Extension of CFD code application to two-phase flow safety problems. Nucl. Eng.
Technol 42, 365–376 (2010a)
D. Bestion, Applicability of Two-Phase CFD to Nuclear Reactor Thermal Hydraulics and Elabo-
ration of Best Practice Guidelines (CFD4NRS-3, Washington, DC, 2010b) Sept 2010, to be
published in a special issue of Nuc. Eng. Des.
T. Boeck, S. Zaleski, Viscous versus inviscid instability of two-phase mixing layers with continuous
velocity profile. Phys. Fluids 17, 032106-1–032106-11 (2005)
M. Brocchini, D.H. Peregrine, The dynamics of strong turbulence at free surfaces. Part1. Descrip-
tion, J. Fluid Mech. 449, 225–254 (2001)
I. Celik, W. Rodi, A Deposition-Entrainment Model for Suspended Sediment Transport. Report SFB
210/T/6, Strömungstechnische Bemessungsgrundlagen für Bauwerke. University of Karlsruhe,
Karlsruhe (1984)
G. Cerne, S. Petelin, I. Tiselj, Coupling of the interface tracking and the two-fluid models for the
simulation of incompressible two-phase flow. J. Comput. Phys. 171(2), 776–804 (2001)
32 T. Höhne
P. Coste, J. Laviéville, A Wall Function-Like Approach for-Two-Phase CFD Condensation Model-

ing of the Pressurized Thermal Shock. Proceedings of NURETH-13, Kanazawa (2009)
P. Coste, J. Pouvreau, C. Morel, J. Laviéville, M. Boucker, A. Martin, Modeling Turbulence and
Friction Around a Large Interface in a Three-Dimension Two-Velocity Eulerian Code. Pro-
ceedings of International Conference on NURETH 12, Pittsburgh (2007)
P. Coste, J. Laviéville, J. Pouvreau, C. Baudry, M. Guingo, A. Douce, Validation of the large
interface method of NEPTUNE_CFD 1.0.8 for Pressurized Thermal Shock (PTS) applications.
Nucl. Eng. Des. 253(0), 296–310 (2012)
A.V. Coward, Y. Renardy, M. Renardy, J. Richards, Temporal evolution of periodic disturbances in
two-layer Couette flow. J. Comput. Phys. 132, 346–361 (1997)
Y. Egorov, Contact Condensation in Stratified Steam-Water Flow. EVOL-ECORA –D 07 (2004).
http://domino.grs.de/ecora/ecora.nsf/
J. Fabre, L. Masbernat, C. Suzanne, Experimental data set no. 7: Stratified flow, part i: Local
structure. Multiph. Sci. Technol. 3(1–4), 285–301 (1987)
R. Fox, Kinetic Theory Based CFD Models for Polydisperse Multiphase Flow. Proceedings ASME
Fluids Engineering Division Summer Meeting, Lake Tahoe (2013) http://www.asmeconferences.
org/FEDSM2013/Plenary.cfm. Last accessed 31 July 2013
T. Frank, Numerical Simulations of Multiphase Flows Using CFX-5. CFX Users conference,
Garmisch-Partenkirchen (2003)
M. Fulgosi, D. Lakehal, S. Banerjee, V. De Angelis, Direct numerical simulation of turbulence in a
sheared air-water flow with a deformable interface. J. Fluid Mech. 482, 319–345 (2003)
P. Griffith, G.B. Wallis, Two-phase slug flow. J. Heat Transf. (US) 83, 307 (1961)
F.H. Harlow, J.E. Welch, Numerical calculation of time-dependent viscous incompressible flow of
fluid with free surface. Phys. Fluids 8(12), 2182–2189 (1965)
G.F. Hewitt, Phenomenological Modelling of Slug Flow. Short Course Modelling and Computation
of Multiphase Flows (ETH Zurich, Zurich, 2003)
C.W. Hirt, B.D. Nichols, Volume of fluid (VOF) method for the dynamics of free boundaries.
J. Comput. Phys. 39(1), 201–225 (1981)
C.W. Hirt, A.A. Amsden, J.L. Cook, An arbitrary lagrangian-eulerian computing method for all
flow speeds. J. Comput. Phys. 14, 227–253 (1974)
T. Höhne, Modelling and Validation of Turbulence Parameters at the Interface of Horizontal
Multiphase Flows. Proceedings of 8th International Conference on Multiphase Flow,
ICMF2013–883 (2013)
T. Höhne, C. Vallée, Experiments and numerical simulations of horizontal two phase flow regimes
using an interfacial area density model. J. Comput. Multiphase Flows 2, 131–143 (2010)
T. Höhne, H.T. Deendarlianto, D. Lucas, Numerical simulations of counter-current two-phase flow
experiments in a PWR hot leg model using an interfacial area density model. Int. J. Heat Fluid
Flow 32, 1047–1056 (2011)
M. Ishii, T. Hibiki, Thermo-Fluid Dynamics of Two-Phase Flow (Springer, New York, 2006)
B. Launder, B. Sharma, Application of the energy-dissipation model of turbulence to the calculation
of flow near a spinning disc. Lett Heat Mass Transf. 1(2), 131–137 (1974)
P.Y. Lin, T.J. Hanratty, Prediction of the initiation of slugs with linear stability theory. Int.
P. Liovic, D. Lakehal, Interface-turbulence interactions in large-scale bubbling processes. Int.
J. Heat Fluid Flow 28, 127–144 (2007a)
P. Liovic, D. Lakehal, Multi-physics treatment in the vicinity of arbitrarily deformable gas-liquid
interfaces. J. Comput. Phys. 222, 504–535 (2007b)
J.M. Mandhane, G.A. Gregory, K. Aziz, A flow pattern map for gas-liquid flow in horizontal pipes:
predictive models. Int. J. Multiphase Flow 1, 537–553 (1974)
F.R. Menter, Two-equation Eddy-viscosity turbulence models for engineering applications. AIAA
J. 32, 1598–1605 (1994)
K. Mishima, M. Ishii, Theoretical prediction of onset of horizontal slug flow. ASME J. Fluids Eng.
102, 441–445 (1980)
A. Mouza, S. Paras, A. Karabelas, CFD code application to wavy stratified gas-liquid flow. Trans.
IChemE 79(Part A), 561–568 (2001)
F. Murzyn, H. Chanson, Experimental investigation of bubbly flow and turbulence in hydraulic
jumps. Environ. Fluid Mech. 9, 143–159 (2009)
R. Nourgaliev, M.-S. Liou, T.G. Theofanous, Numerical prediction of interfacial instability: Sharp
Interface Method (SIM). J. Comput. Phys. 227, 3940–3970 (2008)
S. Osher, J.A. Sethian, Fronts propagating with curvature dependent speed: algorithms based on
Hamilton-Jacobi formulations. J. Comput. Phys. 79(1), 12–49 (1988)
S. Reboux, P. Sagaut, D. Lakehal, Large-eddy simulation of sheared interfacial flow, Phys. Fluids
18, 105 (2006)
Y.D. Shikhmurzaev, Moving contact lines in liquid/liquid/solid systems. J. Fluid Mech. 334,
211–249 (1997)
Y. Taitel, A.E. Dukler, A model for predicting flow regime transitions in horizontal and near
horizontal gas-liquid flow. AICHE J. 22, 47–55 (1976)
Y. Taitel, A.E. Dukler, A model for slug frequency during gas-liquid flow in horizontal and near
horizontal pipes. Int. J. Multiphase Flow 3, 585 (1977)
Y. Taitel, D. Bornea, A.E. Dukler, Modelling flow pattern transitions for steady upward gas-liquid
flow in vertical tubes. AlChE J. 26, 345–354 (1980)
F. Terzuoli, M. Galassi, D. Mazzini, F. D’Auria, CFD code validation against stratified air-water
flow experimental data. Sci. Technol. Nucl. Installations 2008, 596 (2008)
C. Vallée, T. Höhne, H.M. Prasser, T. Sühnel, Experimental investigation and CFD simulation of
horizontal stratified two-phase flow phenomena. Nuc. Eng. Des. 238(3), 637–646 (2008)
P. Valluri, P.D.M. Spelt, C.J. Lawrence, G.F. Hewitt, Numerical simulation of the onset of slug
initiation in laminar horizontal channel flow. Int. J. Multiphase Flow 34, 206–225 (2008)
G.D. Wallis, J.E. Dobson, Onset of slugging in horizontal stratified air-water flow. Int. J. Multiphase
Flow 1, 173–193 (1973)
D.C. Wilcox, Turbulence Modelling for CFD (DCW Industries Inc., La Cañada, 1994)
W. Yao, P. Coste, D. Bestion, M. Boucker, Two-Phase Pressurized Thermal Shock Investigations Using
a 3D Two-Fluid Modeling of Stratified Flow with Condensation. Proceedings of 10th International
Topical Meeting on Nuclear Reactor Thermal Hydraulics (NURETH-10), pp. 5–9 (2003)
Recent Advances in Modeling Gas-Particle
Flows
Luca Mazzei
Abstract
This chapter is concerned with the mathematical modeling of dense fluidized
suspensions and focuses on the so-called Eulerian or multifluid approach. It
introduces newcomers to some of the techniques adopted to model fluidized
beds and to the challenges and long-standing problems that these techniques
present. After introducing the principal approaches for modeling fluid-solid
systems, we focus on the multifluid, overviewing the main averaging techniques
that consent to describe granular media as continua. We then derive the Eulerian
equations of motion for fluidized powders of a finite number of monodisperse
particle classes, employing volume averages. We present the closure problem and
overview constitutive relations for modeling the granular stress and the interac-
tion forces between the phases. To conclude, we introduce the population balance
modeling approach, which permits handling suspensions of particles continu-
ously distributed over the size and any other property of interest.
Keywords
Multiphase flows • Fluidization • Fluidized bed modeling • Averaging methods •
Multifluid modeling • Computational fluid dynamics • Closure problem • Popu-
lation balance modeling
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
An Overview of Fluidized Bed Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
An Overview of Averaging Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Statistical Averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Volume Averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
L. Mazzei (*)
Department of Chemical Engineering, University College London, London, UK
e-mail: l.mazzei@ucl.ac.uk

DOI 10.1007/978-981-4585-86-6_8-1
2 L. Mazzei
Time Averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
A Final Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Averaged Equations of Motion for Fluid-Particle Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Fluid Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Solid Phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
The Problem of Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Effective Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Fluid-Particle Interaction Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Particle-Particle Interaction Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Population Balance Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Fluid-Phase Volume Average of Point Variable Spatial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . 39
Fluid-Phase Volume Average of Point Variable Time Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Particle-Phase Volume Average of Point Variable Time Derivatives . . . . . . . . . . . . . . . . . . . . . . . . 41
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Introduction
Fluidization is a well-established technology used in several industrial processes

such as coal combustion, biomass gasification, waste disposal, and food processing.
To design fluidized beds, engineers have resorted for many years to experimental
correlations and pilot plants. This practical approach to fluidization is well reflected
in the first textbooks on the subject (see, for instance, Leva 1959). These correla-
tions, however, lack general validity and can help design only standard units: they
cannot tell us how changes in vessel geometry, introduction of internals (like heat
exchanger tubes), or feed repartition over various entry points affect fluid dynamics
and performance. To answer these and similar questions, and improve the design of
conventional units, one needs a theory for predicting how dense fluidized powders
behave; pilot plants are not a convenient alternative, being expensive, time-
consuming, and not always leading to adequate scale up.
When fluidized beds were first employed in the 1920s–1940s, engineers did not
appreciate this problem, probably because at the time, the required plant perfor-
mance was either not critical (like in FCC plants) or easily achievable (like in
roasting and drying). Nevertheless, when later the problem revealed itself in other
and more demanding applications, with some plants falling far short of the expected
conversions previously achieved in pilot units, it became clear that this matter had to
be addressed thoroughly. Researchers hence endeavored to find more reliable
methods to predict the dynamics of fluidized suspensions.
In the 1960s, scientists began to adopt the conservation laws of mass, momentum,
and energy to analyze nearly any physical and chemical problem. This innovative
approach, most probably fostered by the release of the influential textbook Transport
Phenomena (Bird et al. 1960), led to significant theoretical headway, bolstered the
hope to explain theoretically the behavior of fluidized powders and prompted the
first trials to develop fluid dynamic models based on transport equations.
Recent Advances in Modeling Gas-Particle Flows 3
Anderson and Jackson (1967) were among the first to model fluidized systems;
starting from the continuity and dynamical equations for single-phase, incompress-
ible fluids and the Newtonian equations for rigid-body motion, they derived aver-
aged equations of conservation for the fluid and solid phases by applying a formal
mathematical process of volume averaging. Afterward, several researchers did the
same, refer, for instance, to Whitaker (1969), Drew (1971), and Drew and Segel
(1971). Initially, they used these models to understand better the complex behavior
of multiphase systems, but without regarding them as a viable way to design real
industrial units. Nonetheless, when faster computer processors and advanced numer-
ical methods to integrate partial differential equations became available, they real-
ized that a mathematical theory of multiphase flows could indeed become a useful
design tool.
With the further development of new and more rigorous formulations of multi-
phase equations of motion (Buyevich 1971; Hinch 1977; Nigmatulin 1979; Drew
1983; Jenkins and Savage 1983), the late 1970s and early 1980s witnessed the first
endeavors to simulate numerically granular flowing systems (Pritchett et al. 1978;
Gidaspow and Ettehadieh 1983; Gidaspow et al. 1986). The promising results of
these few pioneering studies generated an increasing interest in computational fluid
dynamics (CFD) and multiphase flows, which rapidly started to turn into research
areas in their own right.
Nowadays, CFD has become an almost indispensable tool to solve problems of
academic and industrial interest. In the field of fluidization, CFD has assisted to
understand fluid-solid interactions and has enabled to predict numerous macroscopic
phenomena encountered in particulate systems. Similarly, multiphase flows and
fluidization dynamics have become topics of interest not only for the scientific
community but also for the industrial world. Over the years, many researchers
have developed mathematical models to predict the dynamics of dense fluidized
suspensions, proposing several approaches and mathematical schemes; we now
briefly overview some of them, highlighting their advantages and limitations.
An Overview of Fluidized Bed Modeling
Fluidized beds can be modeled at various levels of detail. At the most fundamental,
the motion of the system is determined by the Newtonian equations for translation
and rotation of each particle and the Navier-Stokes transport equations to be satisfied
at every point occupied by the interstitial fluid. In this approach, referred to as
Eulerian-Lagrangian, the flow field of the fluid is modeled at a length scale far
smaller than the particle size; one, therefore, is able to determine the pressure and
velocity gradient over the surface of each particle and from there the interaction force
between the fluid and the particles (see, for instance, Pan et al. 2002). No closure
problem hence arises for this force. Furthermore, since the particles are considered
individually, the solid phase is not modeled as a continuum, retaining its granular
nature; the equations describing its motion, consequently, feature no granular stress.
4 L. Mazzei
Also the closure problem for this term, which arises in macroscopic models, is
therefore absent.
This modeling strategy is appealing: it is conceptually quite simple – being
probably the most natural for describing the dynamics of particulate systems – and
it is the least affected by closure problems. It presents, nonetheless, a few disadvan-
tages. First of all, it is extremely demanding computationally: simulations of this
kind have been performed only for systems containing a relatively small number of
particles; extending these calculations to dense suspensions in large domains, like
those found in industrial fluidized beds, is presently inconceivable. Moreover, even
if this were feasible, the information provided by the solution would be much too
detailed, and one would have to filter the results to render them useful. Note that such
results are of little direct interest to most end users: the simulations yield the position
and velocity of each particle at any given time, but what are the spatial distributions
of the observables, such as the granular temperature and pressure, which are of real
interest in applications? With the output of Eulerian-Lagrangian simulations, one can
obtain these distributions (if one knows how the observables of interest are related to
the fluid and particle dynamics at the microscopic length scale, a knowledge that is
acquired when one derives the averaged equations which characterize continuum
models), but a complex calculation is required.
These observations suggest that it might be convenient to formulate equations of
change governing the evolution of these observables directly. In this approach, we
renounce to capture the details described above, satisfying ourselves with a far
reduced description of the flow. Although there exists no guarantee that these
simplified equations can be really obtained – in closed form, that is – several studies
have been conducted in such a pursuit (Anderson and Jackson 1967; Whitaker 1969;
Drew 1971, 1983; Drew and Segel 1971; Drew and Lahey 1993; Gidaspow 1994;
Zhang and Prosperetti 1994; Enwald et al. 1996; Jackson 1997, 1998, 2000). Owing
to the complexity of the problem, one does not aim to derive the general exact
averaged multiphase equations of motion; the intent is merely to formulate models
which may describe satisfactorily phenomena of interest for industrial applications.
Various mathematical techniques yield such equations, and several claims have
been advanced as to the superiority of each form of averaging versus the others.
However, the resulting transport equations are very similar and present many
common features. Two are the most significant. First, they are all written in terms
of mean variables defined over the entire physical domain, so, they resemble those
that one would write for a set of imaginary fluids which interpenetrate one another
and occupy simultaneously the same physical volume. The model, known as
Eulerian-Eulerian or multifluid, thus takes the form of coupled differential equations
subjected to initial and/or boundary conditions assigned only on the boundaries of
the domain containing the mixture, and no longer on the surface of the particles, as in
Eulerian-Lagrangian models. Second, the process of averaging generates a number
of indeterminate terms unrelated to the averaged variables but associated with details
of the motion at the particle (that is, microscopic) length scale. These are key terms,
represented by the fluid and solid stress tensors and by the interaction forces
exchanged by the phases. A closure problem hence arises, which one cannot usually
solve analytically; in fact, there is no guarantee that a solution exists. So, one has to
resort to empirical relations, this being the main shortcoming of the method.
Besides these two approaches, there is a third that can be regarded as a hybrid
between them. Averaged equations of motion are used for the fluid phase, but rigid-
body Newtonian equations are solved for every particle of the system. These do not
interact with the fluid through its microscopic velocity field – as is the case in
Eulerian-Lagrangian models – but with the averaged value of the latter. For instance,
the overall force exerted by the fluid on each particle is not computed by integrating
over the particle surface the local traction arising from the fluid velocity gradients:
the force is instead calculated in terms of slip velocity between the average fluid
velocity and the velocity of the particle center of mass and by resorting to empirical
relations. This strategy, called discrete particle modeling (DPM), is significantly less
demanding computationally than the Eulerian-Lagrangian and has met with resound-
ing success (Tsuji et al. 1993; Hoomans et al. 1996; Xu and Yu 1997; Ouyang and Li
1999; Kafui et al. 2002; Lu et al. 2005; Zhu et al. 2008; Di Renzo et al. 2011; He
et al. 2012; Wang et al. 2013; Deen et al. 2014).
To describe particle collisions, modelers use two approaches: hard and soft
sphere. In the first, particles interact via binary, instantaneous, pointwise collisions.
Their velocities after an encounter are computed by requiring that linear and angular
momentum are conserved in the collision. This approach was pioneered by Allen
and Tildesley (1990). Since their publication many authors have found it useful to
model the collision dynamics in granular systems. Hoomans et al. (1996) used it in
their model for gas-fluidized beds; it was the first time that the technique had been
applied to a dense system. Many authors have since published papers with this
strategy (see, for instance, Ouyang and Li 1999). The soft-sphere model for fluidized
beds was instead pioneered by Tsuji et al. (1993), who developed their approach on
the basis of earlier work done by Cundall and Strack (1979). Here, during an
encounter, particles overlap slightly, and the contact forces are calculated from the
deformation history of the solids using a linear spring-dashpot model. This has been
employed by Xu and Yu (1997), Pandit et al. (2005), Ye et al. (2005), and several
other researchers.
Among the three modeling approaches discussed, the second is often preferred
for its valuable feature of being computationally less demanding. Due to the number
of particles present in industrial plants, Eulerian (continuum) modeling is unlikely to
be replaced by its Lagrangian (discrete) counterparts in the near future; furthermore,
Eulerian models appeal more to end users, because they provide information of
direct interest. The role of discrete modeling is yet paramount. The method, to be
considered more as an effective research tool than a practical design instrument, by
yielding information about the dynamics of multiphase systems at the microscopic
length scale, may significantly help to develop and improve continuous models
through the derivation of accurate closure relations. Eulerian-Lagrangian and DPM
simulations are to multiphase flows what direct numerical simulations are to turbu-
lent flows. This multiscale modeling strategy is represented in Fig. 1. The goal of the
strategy is clear, but how to link the models and extract from each the information
needed by those higher up in the hierarchy is an open challenge.
6 L. Mazzei
Fig. 1 Multiscale modeling strategy
An Overview of Averaging Theory
In the present section, we focus on three techniques that one can adopt to derive
averages of point variables: statistical, volume, and time averaging.
Statistical Averaging
As previously mentioned, predicting the dynamics of fluidized suspensions may

appear conceptually simple: one has to solve the Newtonian equations of motion for
each particle and the Navier-Stokes equations for the fluid. In practice, doing so is
extremely demanding, because the number of particles is quite large; however, one
could argue that this is only a practical and temporary issue, which future generations
of computers will certainly overcome. There is, nevertheless, a more fundamental
problem: to integrate the equations, one has to know the initial positions and
velocities of all the particles. For large particle numbers, this information is impos-
sible to obtain. The problem, therefore, cannot be addressed deterministically: a
statistical approach is necessary. To clarify this concept, let us be more definite.
Consider a fluidized suspension of ν identical, spherical, smooth particles, and let
xs(t) and vs(t) be the position vector and linear velocity of the sth particle center,
respectively. Initially, the sth particle is located in the point xs with velocity vs. If Fs(t)
denotes the unit mass force acting on the particle, as time advances the latter moves
obeying the equations:
x_ s ðtÞ ¼ vs ðtÞ; v_s ðtÞ ¼ Fs ðtÞ; xs ð0Þ ¼ xs ; vs ð0Þ ¼ vs (1)
Consequently, if we know the initial conditions for each particle and the functional
expression of the overall force acting on each particle, by integrating the above
differential equations, we may predict, with certainty, the particle positions and
velocities at any future time. For a realistic number of particles, knowing the initial
positions and velocities of the particles appears to be impossible even in principle.
For systems comprising a great number of particles, therefore, we cannot know
the initial state of every particle; in consequence, we cannot assign the initial
conditions deterministically, as we did in Eq. 1. What we usually know are solely
macroscopic – and therefore measurable – properties of the system, such as its local
density, temperature, or mean velocity. But there are infinite system configurations
yielding the same macroscopic properties, each with a certain probability of occur-
rence. So, we must replace the deterministic initial conditions above with probabi-
listic initial conditions. To this end, we introduce a probability density function
(PDF) defined so that:
π ν ðx1 , . . . , xν , v1 , . . . , vν ; tÞdx1 . . . dxν dv1 . . . dvν (2)
gives the joint probability that at time t the first particle has position and velocity in
the ranges dx1 and dv1 about the real-space point x1 and the velocity-space point v1,
the second particle has position and velocity in the ranges dx2 and dv2 about the real-
space point x2 and the velocity-space point v2, and so on up to the last particle
forming the particulate system. If we let:
r ð x1 , . . . , x ν , v 1 , . . . , v ν Þ (3)
we can regard r as the position point identifying the state of the entire particulate
system in an abstract phase space of 6ν dimensions. Then π ν(r; t)dr is the probability
that at time t the configuration of the particulate system lies in the range dr about the
phase-space point r. We refer to the function π ν(r; t) as the ν-particle joint PDF or
master joint PDF. At any given time, we do not know the exact configuration of the
system, but the master joint PDF states how probable each configuration is. π ν(r; 0)
dr, in particular, is the probability that the initial configuration lies in the range dr
about r. No determinism is present: the system can be in any configuration, but for
each one, the PDF tells us the probability of occurrence. Knowing the master joint
PDF means having complete statistical knowledge of the population of particles.
Another useful function is the one-particle marginal PDF, which gives the
probability of finding a single particle in a differential neighborhood of a given
state – independently of the states of all the other particles. For a system in which the
particles are identical, and consequently indistinguishable, the ν-particle joint PDF is
symmetrical with respect to the particle state variables, and the one-particle marginal
PDF is equal for all the particles. Let π 1(x1, v1; t) denote the latter (for the real-space
and velocity-space variables, the subscript is unimportant; the one reported has been
selected only for convenience). By definition, π 1(x1, v1; t)dx1dv1 is the probability of
finding at time t a particle – any particle of the population, not just the first particle –
in the ranges dx1 and dv1 about the points x1 and v1. The one-particle marginal PDF
contains significantly less information about the particulate system than the master
8 L. Mazzei
joint PDF; however, as we shall see, in most cases of real interest, this is all the
information that is truly required.
We reduce the ν-particle joint PDF to the one-particle marginal PDF by integrat-
ing out the state variables of every particle but the first one:
ð ð ð ð ν
π 1 ðx1 , v1 ; tÞ ... ... π ν ðr; tÞ ∏ dxs dvs (4)
Ωx Ωx Ωv Ωv s¼2
where Ωx and Ωv represent the ranges of variation of the particle positions and
velocities, respectively; the former coincides with the region of physical space
enclosed by the vessel containing the suspension, whereas the latter is unbounded
and coincides with ℝ3.
Knowing the master joint PDF allows calculating any average associated with the
population of particles. Take a function b(r) that associates a scalar value with the
state of the particulate system. This is referred to as dynamical function (Balescu
1975). If the system changes configuration (that is, the value of r changes), the value
of the function changes. The observable 〈b〉s associated with the function b is the
average value of the latter over all the system configurations; therefore, it is:
ð
hbis ðtÞ bðrÞπ ν ðr; tÞdr (5)
Ωr
where Ωr represents the range of variation of r. In the integral above, we are

summing all the values that the dynamical function can take, each one weighted
by the probability of occurrence of the system configuration to which that value
refers. Being macroscopic variables, observables do not depend on the microscopic
state of the system but can be functions of the time, real-space, and velocity-space
coordinates; in the expression above, 〈b〉s is only a function of time, for b does not
depend on real-space and velocity-space coordinates; in general, however, this
dependence will be present. Equation 5 suggests that to calculate any kind of
macroscopic property of the system, one needs to know the master joint PDF (that
is, one has to have complete knowledge of the system). In general this is true, but
fortunately this is not always the case.
A class of dynamical functions of particular theoretical importance is given by
functions which take the following mathematical form:
X
ν
bðrÞ ¼ b1 ðxs , vs Þ (6)
s¼1
where b1 is an arbitrary function of the phase-space state of one particle. This

dynamical function depends on the state of the entire system, but the state of each
particle is taken one at a time. An example is given by the total kinetic energy of the
system, in which case b1 m(vs vs)/2, where m is the particle mass. For this class
of dynamical functions, we can write:
ν ð
X
h bi s ð t Þ ¼ b1 ðxs , vs Þπ ν ðr; tÞdr
s¼1 Ωr
ð ð ð ð
¼v b1 ðx1 , v1 Þπ 1 ðx1 , v1 ; tÞdx1 dv1 ¼ b1 ðx1 , v1 Þf 1 ðx1 , v1 , tÞdx1 dv1
Ωx Ωv Ωx Ωv
(7)
where it is:
f 1 ðx1 , v1 , tÞ νπ 1 ðx1 , v1 ; tÞ (8)
Here we have exploited the symmetry properties of the ν-particle PDF, which hold
insofar as all the particles are identical. Thus, for this class of dynamical functions, to
calculate the observables, one needs to know only the one-particle marginal PDF or
equivalently the scalar function f1(x1, v1, t).
Known as number density function (NDF), the latter arises naturally from the
passages shown above but has as well an important physical interpretation: f1(x, v, t)
dxdv represents the average number of particles present at time t in the range
(or infinitesimal volume) dx around the real-space point x with velocity in the
range dv around the velocity-space point v (we have removed the subscript from
the arguments of the NDF for convenience). f1(x, v, t), in other words, is an
observable representing the mean particle number density in the six-dimensional
phase space formed by the union of the real space Ωx and velocity space Ωv. To prove
this, we must show that the NDF is the mean value of the number density of particles
present in the real-space point x with velocity v. This density has this expression:
X
ν
φ1 ðr; x, vÞ δðxs x, vs vÞ (9)
s¼1
This is because if no particle is located in x with velocity v the density is zero, while
if a particle is therein located, the density diverges (assuming that the volume of the
particles is negligibly small compared with the macroscopic volumes of interest).
Notice that φ1(r; x, v) belongs to the special class of dynamical functions defined by
Eq. 6. Thus, we have:
ð ð
hφ1 is ðx, v, tÞ ¼ ν δðx1 x, v1 vÞπ 1 ðx1 , v1 ; tÞdx1 dv1 ¼ νπ 1 ðx, v; tÞ (10)
Ωx Ωv
This differs from the NDF defined in Eq. 8 merely in notation. Because of the
important physical meaning that the NDF possesses, one usually favors the latter
over the one-particle marginal PDF; knowledge of either function, however, permits
calculating observables associated with dynamical functions of the class defined by
Eq. 6. Of course, to determine observables of this kind, one needs to know how the
10 L. Mazzei
NDF evolves in each phase-space point; this, as we shall see at the end of this
chapter, is extremely challenging.
To conclude this section, we present three examples of observables that are
particularly significant: mass, linear momentum, and energy density. The dynamical
functions with which these macroscopic quantities are associated take the following
expressions:
X
ν X
ν
bM ðr; xÞ m δðxs xÞ; bL ðr; xÞ m vs δðxs xÞ
s¼1 s¼1
X
ν (11)
bE ðr; xÞ ðm=2Þ ðvs vs Þδðxs xÞ
s¼1
Eq. 7 then gives:
hbM is ðx, tÞ ¼ mnðx, tÞ; hbL is ðx, tÞ ¼ mnðx, tÞhvis ðx, tÞ

(12)
hbE is ðx, tÞ ¼ ðm=2Þnðx, tÞhvvis ðx, tÞ
where n represents the expected number of particles per unit real-space volume, or
equivalently the expected number density of particles in real space, 〈v〉s the expected
particle velocity, and 〈v v〉s twice the expected kinetic energy per particle unit mass.
Mathematically, their expressions are:
ð
Ð 1
nðx, tÞ Ωv f 1 ðx,
v, tÞdv; hvis ðx, tÞ v f 1 ðx, v, tÞdv
ð nðx, tÞ Ωv
1 (13)
hvvis ðx, tÞ ðvvÞf 1 ðx, v, tÞdv
nðx, tÞ Ωv
In this section, to simplify the treatment, we have assumed that the state of each
particle is identified only by position in real space and velocity. Additional coordi-
nates can be introduced, such as the particle size, but the concepts presented do not
change. For a more general treatment of this subject, we refer to the literature, in
particular to the textbook by Marchisio and Fox (2013).
Volume Averaging
Another method of deriving observables relies on volume averages; these are

computed over spatial domains that are large enough to contain a statistically
significant number of particles, but which are small compared with the length
scale of variation of the observables.
There are two kinds of volume averages: hard and soft. In the former, a volume Vx
bounded by a surface Sx is attached to every spatial point x; within this volume, one
averages the property of interest by using the mean value theorem of integral
calculus. The values of the property within Vx are accounted for and ascribed the
same weight in the average, while those outside Vx are ignored. Soft averages are
based on an alternative technique, more elegant and convenient from a mathematical
viewpoint, that uses radial weighting functions. These are continuous, monotone,
decreasing functions of the radial distance from the spatial point in which the
average is evaluated. This mathematical device ascribes a weight to the property
values within the whole physical domain; however, the length scale over which the
weighting function decays significantly (referred to as weighting function radius)
identifies a spherical volume around the point of average x outside which the
property values affect the average negligibly. The two averaging schemes, accord-
ingly, are not as different as they may appear. For mathematical convenience, in what
follows we favor soft averages.
Volume-averaged variables might appear to depend on the specific choice of
volume Vx or of weighting function (in particular on its radius). The larger the ratio
between the smallest length scale of variation of the observables and the particle size,
the more such a dependence dwindles provided that the weighting function radius is
properly chosen. If this radius is denoted by r2, the particle radius by r1, and the
macroscopic length scale by r3, the local average is expected not to depend on the
particular form of weighting function provided that the condition r1 r2 r3 is
satisfied. In such a case, there is said to exist separation of scales between the
macroscopic fluid dynamic problem and the detailed motion at the scale of a single
particle. Only in this instance the volume-averaged variables have an unambiguous
physical meaning.
In multiphase systems, made up of one continuous phase (the fluid) and one or
more discrete phases (the particles), one can employ volume averages to obtain mean
properties for each phase. We now first introduce formally the weighting functions
and then report how such averages are defined.
Weighting Functions
Weighting functions are characterized by the following mathematical properties:
1. The weighting function ψ is a scalar function of r defined for r > 0, where

r denotes the distance of a point z from a point x in Euclidean space:
ψ ¼ ψ ðr Þ, r jx zj (14)
2. ψ(r) is positive for any value of r, decreases monotonically with r and possesses
continuous derivatives of any order. In other words, it is a function of class C1.
3. ψ(r) is normalized, so that, if Ωx denotes the spatial domain occupied by the
system of interest (assumed here to stretch out to infinity), it is:
ð ð1
ψ ðjx zjÞdz ¼ 4π ψ ðr Þr 2 dr ¼ 1 (15)
Ωx 0
12 L. Mazzei
In the integral on the left-hand side, z is the spatial variable of integration, while x is
the spatial position in which the volume average is computed. The radius of the
weighting function is defined as the scalar r2 that satisfies the following equation:
ð r2 ð1
1
4π ψ ðr Þr 2 dr ¼ 4π ψ ðr Þr 2 dr ¼ (16)
0 r2 2
The weighting function radius is thus a measure of the linear size of the spherical
neighborhood of x in which the spatial points have appreciable weight in the
averaging process.
Fluid-Phase Volume Averages

The void fraction, or fraction of space occupied by the fluid, and the fluid-phase
volume average of a generic point variable ζ(x, t) calculated in x at time t are so
defined:
ð ð
1
eðx, tÞ ψ ðjx zjÞ dz; hζ ie ðx, tÞ ζ ðz, tÞψ ðjx zjÞdz (17)
Λe eðx, tÞ Λe
In the equations above, Λe represents the domain occupied by the fluid phase at time
t (we have left out the explicit dependence on t to simplify the notation).
Solid-Phase Volume Averages

In a system with ν solid phases, the volume fraction of the rth phase Sr and the solid-
phase volume average of a generic point variable ζ(x, t) calculated in x at time t are
so defined:
Xð 1 Xð
ϕr ðx, tÞ ψ ðjx zjÞ dz; hζ irs ðx, tÞ ζ ðz, tÞψ ðjx zjÞ dz
Sr Λr ϕr ðx, tÞ S Λr
r
(18)
where Λr is the region of physical space occupied by a generic particle of phase S r
at time t. The summation is over all the particles of phase Sr. In hζ irs the subscript
s indicates that this is a solid-phase volume average, while the superscript r indicates
that the average refers to solid phase Sr.
This average, used by several researchers (Enwald et al. 1996; Drew and Passman
1998), operates on the microscopic properties of the particle material, considering
point fields ζ(x, t) that vary within the particles. It is an average which exactly
parallels the one given for the fluid. Another approach, advanced by Anderson and
Jackson (1967), is based on properties ζ r(t) of each particle as a whole.
Particle-Phase Volume Averages

Since the particles are rigid, their motion is determined by the translation of their
centers of mass and by the rotation of their bodies about instantaneous axes of
rotation. Thus, the resultant forces and torques acting on the particles suffice to
establish their motion. We can then introduce a different kind of volume average that
depends only on properties of the particles as a whole. We define the number density
of particles of class Sr calculated in x at time t as follows:
X
nr ðx, tÞ ψ ðjx zr ðtÞjÞ (19)
Sr
zr(t) being the position occupied at time t by the center of mass of a generic particle
of solid phase Sr. The volume fraction ϕr(x, t) is related to the number density nr(x, t)
as follows:
ϕr ðx, tÞ nr ðx, tÞV r (20)
in which Vr is the volume of a particle of solid phase Sr. As indicated, this equation is
approximate, but it is accurate if the separation-of-scale requirement is met and the
weighting function radius is selected correctly; this, as said, must be far larger than
the particle radius.
Generalizing the averaging scheme of Jackson (1997), we define the particle-
phase volume average for a particle property ζ r(t) of solid phase Sr calculated in x at
time t as:
1 X
hζ irp ðx, tÞ ½ζ r ðtÞψ ðjx zr ðtÞjÞ (21)
nr ðx, tÞ S
r
In hζ irp the subscript p indicates that this is a particle-phase volume average, while the
superscript r indicates that the average refers to solid phase Sr.
Time Averaging
The third averaging method available is time averaging. Let us consider a field ζ(x, t);
for any fixed spatial position x, ζ(x, t) is a function of time that fluctuates irregularly.
We denote the time scale that represents these fluctuations as τ1. In x, we can obtain a
mean value of ζ(x, t) by time averaging over a large number of fluctuations,
considering a time interval τ2 much larger than the time scale of the fluctuations.
Again, we resort to the mean value theorem, this time writing:
ð tþα
1
hζ it ðx, tÞ ζ ðx, τÞdτ, α τ2 =2 (22)
τ2 tα
14 L. Mazzei
where 〈ζ〉t(x, t) denotes the time average and τ is a dummy integration variable. Also
now, the mean value is expected to be insensitive to the averaging time scale
provided that τ1 τ2 τ3, where τ3 represents the time scale of the mean flow
variations. That is, there has to be separation of scale (now in the time domain)
between the macroscopic motion of the fluid-solid mixture and the microscopic
motion of the particles; only in this case the time-averaged variables have an
unambiguous physical meaning.
A Final Remark
Before concluding this section on averaging, we would like to point out that different
averaging schemes can lead to different average values. If the values are equal, the
system is said to be ergodic, but not all systems present this feature. For details, refer,
for instance, to Jackson (2000).
Averaged Equations of Motion for Fluid-Particle Systems
We now derive the averaged equations of motion for a generic fluid-particle system
of ν solid phases by using volume averages. Similar equations can be obtained with
statistical and time averages; refer, for instance, to Gidaspow (1994), Drew and
Passman (1998), and Brilliantov and Poschel (2004). Our treatment is an extension
of the work of Jackson (2000) and Owoyemi et al. (2007). Below, we adopt
Einstein’s convention: repeated indices are summed over the values one to three,
with the exception of r and s, used as phase indices, and of e and p, used to specify
the volume average type.
Fluid Phase
Let us first derive the volume-averaged continuity equation. The starting point is the
microscopic continuity equation for the fluid. If we assume that the latter is incom-
pressible, this reads:
@ a ua ¼ 0 (23)
where @ a @/@xa and ua is the ath component of the fluid velocity vector u(x, t) with
respect to a generic orthonormal vector basis. Let us multiply both sides by ψ(|x z|)
and integrate over Λe with respect to z; doing so yields this averaged equation:
ð
eh@a ua ie ¼ ½@a ua ðz, tÞψ ðjx zjÞdz ¼ 0 (24)
Λe
In this form the equation is not useful, because it is written in terms of averaged
derivatives of point variables instead of derivatives of averaged point variables. We
may, however, manipulate the equation by using these mathematical relations,
whose proof is given in the appendix:
ν Xð
X
eh@a ζ ie ¼ @a ehζ ie ζ ðz, tÞka ðz, tÞψ ðjx zjÞ dσ z (25)
r¼1 Sr @Λr
Xν Xð
eh@t ζ ie ¼ @t ehζ ie þ ζ ðz, tÞuðz, tÞkðz, tÞψ ðjx zjÞ dσ z (26)
r¼1 Sr @Λr
Here @ t @/@t, k(x, t) is the outward unit normal to the surface @Λr bounding Λr,
while ka(x, t) is the ath component of the unit vector k(x, t). Setting ζ ua and ζ 1
in Eqs. 25 and 26, respectively, and adding the results yields the averaged continuity
equation in the form that we sought:

@t e þ @a ehua ie ¼ 0 (27)
In this equation, as we should have expected, the fluid volume fraction takes on the
role that the fluid density has for single-phase compressible fluids.
Let us go on to derive the volume-averaged linear momentum balance equation
for the fluid. The starting point is the corresponding microscopic balance equation:
ρe ½@t ua þ @b ðua ub Þ ¼ @b T ab þ ρe ga (28)
where ρe is the (constant) fluid density and Tab(x, t) is the abth component of the
point fluid stress tensor, while ga is the ath component of the gravitational field.
Multiply both sides by ψ(|x z|) and integrate over Λe with respect to z. To treat the
left-hand side of the averaged equation obtained, write Eqs. 25 and 26 with ζ ua ub
and ζ ua, respectively, while to treat the right-hand side, use Eq. 25 with ζ Tab.
With these relations, the averaged equation becomes:

ρe @t ehua ie þ @b ehua ub ie
ν X
X ð
(29)
¼ @b ehT ab ie þ eρe ga T ab ðz, tÞkb ðz, tÞψ ðjx zjÞdσ z
r¼1 S r @Λr
The last term on the right-hand side is the sum over all particle classes of the mean
resultant traction forces exerted by the fluid on the particles of each class. The force:
Xð
T ab ðz, tÞkb ðz, tÞψ ðjx zjÞdσ z (30)
Sr @Λr
is the sum of the average resultant forces exerted by the fluid on the rth phase
particles. To compute this force for each particle, we first weight the differential
16 L. Mazzei
traction forces acting on each infinitesimal region dσ z of the particle surface using
the value of ψ(|x z|) corresponding to each region, and then we sum the (infinite
number of) contributions. The fluid-solid interaction force, defined by Eq. 30,
couples the linear momentum balance equation of the fluid to that of each particle
class.
For reasons that will be clear later (when we deal with the solid phases), it is
convenient to express Eq. 30 differently. To do so, we expand the weighting function
in a Taylor series about the center zr(t) of a generic particle of phase Sr, writing:
8z @Λr : ψ ðjx zjÞ ψ ðjx zr jÞ

2 (31)
½@b ψ ðjx zr jÞr r kb ðzÞ þ ð1=2Þ @bc ψ ðjx zr jÞ r 2r kb ðzÞkc ðzÞ
where rr is the radius of the particles of phase Sr. As the particle radius is far smaller
than the radius of the weighting function, we may truncate the Taylor series at the
second-order term with acceptably small error. Using this relation, we approximate
the force in Eq. 30 as:

nr hf a irp @b nr hAab irp þ ð1=2Þ@bc
2
nr hBabc irp (32)
where it is:
X ð
nr ðx, tÞhf a irp ðx, tÞ ψ ð j x zr j Þ T ad ðz, tÞkd ðz, tÞdσ z (33)
Sr @Λr
X ð
nr ðx, tÞhAab irp ðx, tÞ ψ ðjx zr jÞr r T ad ðz, tÞkd ðz, tÞkb ðz, tÞdσ z (34)
Sr @Λr
X ð
nr ðx, tÞhBabc irp ðx, tÞ ψ ðj x zr jÞr 2r T ad ðz, tÞkd ðz, tÞkb ðz, tÞkc ðz, tÞdσ z
Sr @Λr
(35)
The quantities defined above are the components of a vector, a second-order tensor,
and a third-order tensor, respectively. The force in Eq. 30 is obtained by first
weighting the differential traction forces exerted on the infinitesimal surface ele-
ments of the fluid-particle interface, using the values of the weighting function at the
locations of the elements, and then by summing such contributions. The force in
Eq. 33, on the other hand, is obtained by first calculating the forces acting on the
entire surface of each particle, then by weighting them using the values of the
weighting function at the particle centers, and finally by summing such contribu-
tions. This second average interprets better the fluid-particle interaction force and
fulfills the principle of action and reaction, as we will see in section “Solid Phases”;
this is why we prefer to operate in terms of this average force and of the additional
contributions appearing in Eq. 32.
The convective term in Eq. 29 features the average of the product of point
velocity components. We find it convenient to decompose it into the sum of a
product of average velocity components and of an average of velocity fluctuations;
thus, we write:
hua ub ie hua ie hub ie þ hû a û b ie (36)
Here hatted variables denote the deviations of point variables from their respective
mean values. The relation above is not exact, holding only when there is separation
of scale between the microscopic and macroscopic descriptions of the flow. Intro-
ducing Eqs. 32 and 36 into Eq. 29 yields:
ν
X
ρe @t ehua ie þ @b ehua ie hub ie ¼ @b hSab ie nr hf a irp þ eρe ga (37)
r¼1
where it is:
ν h
X i
hSab ie ehT ab ie þ nr hAab irp ð1=2Þ@c nr hBabc irp eρe hû a û b ie (38)
r¼1
This term is the fluid-phase effective stress tensor. Finding an analytical closure for it
is extremely complex, but Jackson (1997) did so for the limiting case of diluted,
Stokesian, monodisperse suspensions fluidized by Newtonian fluids. We will
address the problem of closure, for all the terms featuring on the right-hand side of
Eq. 37 and of the ν solid-phase-averaged dynamical equations, later on in section
“The Problem of Closure.”
Solid Phases
The volume-averaged continuity equation for the generic solid phase S r can be
derived quite easily by using this mathematical relation, whose proof is given in the
appendix:

nr ζ_
r
p
¼ @t nr hζ irp þ @a nr hζva irp (39)
where the dot denotes a total time derivative and hζva irp ðx, tÞ is the average of the
product of ζ r(t) and of the ath component of the velocity vr(t) of the particle center.
Setting ζ r 1 gives:

@t nr þ @a nr hva irp ¼ 0 (40)
which is the equation sought. Here the particle number density, or equivalently the
volume fraction, takes on the (compressible) fluid density role.
18 L. Mazzei
To derive the volume-averaged equation of motion for the generic solid phase S r,
we adopt the equation governing the motion of the generic particle of such phase:
ð ν X
X
ρr V r v_r, a ðtÞ ¼ T ab ðz, tÞkb ðz, tÞdσ z þ f rs, a ðtÞ þ ρr V r ga (41)
@Λr s¼1 Ss
where ρr denotes the density of the particles of phase S r, v_r, a ðtÞ the ath component
of the acceleration of the particle center of mass while frs,a(t) the ath component of
the force exerted on the r particle by the generic s particle of phase S r when a
collision takes place. This force does not vanish only if particles r and s are in direct
contact (it is zero for most s particles). Notice also that frs,a(t) vanishes when r and
s refer to the same particle, because particles r and s need, of course, to be different.
The surface integral on the right-hand side of the equation is the overall force exerted
by the fluid on the particle.
To average Eq. 41, we multiply both sides by ψ(|x zr|) and sum over all the
particles belonging to phase S r. Doing so gives:
X ð
X
ρr V r ψ ðjx zr ðtÞjÞv_r, a ðtÞ ¼ ψ ðjx zr ðtÞjÞ T ab ðz, tÞkb ðz, tÞdσ z
Sr Sr @Λr
" # (42)
X ν X
X X
þ ψ ðjx zr ðtÞjÞ f rs, a ðtÞ þ ρr V r ga ψ ðjx zr ðtÞjÞ
Sr s¼1 S s Sr
We now employ Eqs. 19, 21, 33, and 39, with ζ r v_r, a in the second and ζ r vr, a in
the fourth, and the following relation, whose proof is left to the reader:
" # " #
X ν X
X ν X
X X
ψ ðjx zr ðtÞjÞ f rs, a ðtÞ ¼ ψ ðjx zr ðtÞjÞ f rs, a ðtÞ (43)
Sr s¼1 Ss s¼1 S r Ss
to obtain:
h i
ρr V r @t nr hva irp þ @b nr hva vb irp
" # (44)
r
ν X
X X
¼ nr hf a ip þ nr ρr V r ga þ ψ ðjx zr ðtÞjÞ f rs, a ðtÞ
s¼1 S r Ss
The first term on the right-hand side is the fluid-particle interaction force – which
also features, with opposite sign, in Eq. 37. This force satisfies the action-and-
reaction principle, as it should. The final term combines the resultant forces arising
from the particle-particle contacts among particles that belong to the same phase
(s = r) and to different phases (s 6¼ r). These contributions are conceptually different,
insofar as the former is a self-interaction term that represents the stress internal to the
phase under examination, while the latter is a contact force acting between the
Eulerian solid phases. To let the collisional solid stress tensor associated with phase
S r appear explicitly in Eq. 44, we need to manipulate the equation further. We first
consider the following double sum over the particles r and s of the rth phase:
X X
ψ ðjx zrs ðtÞjÞf rs, a ðtÞ (45)
Sr Sr
in which zrs(t) denotes the position vector of the point of mutual contact between the
rigid particles r and s. This double sum vanishes, inasmuch as zrs = zsr and, for the
action-and-reaction principle, frs,a = fsr,a. If we then expand ψ(|x zrs|) in a Taylor
series around zr, letting krs denote the unit vector of the vector zrs zr, we obtain
from the equation above that:
" #
X X h i
ψ ðjx zr ðtÞjÞ f rs, a ðtÞ @b nr hMab irp ð1=2Þ@c nr hN abc irp (46)
Sr Sr
where it is:
( )
X X
nr ðx, tÞhMab irp ðx, tÞ ψ ðjx zr ðtÞjÞr r f rs, a ðtÞkrs, b ðtÞ (47)
Sr Sr
( )
X X
nr ðx, tÞhN abc irp ðx, tÞ ψ ðjx zr ðtÞjÞr 2r f rs, a ðtÞkrs, b ðtÞkrs, c ðtÞ (48)
Sr Sr
The second-order tensor so defined:

hCab irp nr hMab irp ð1=2Þ@c nr hN abc irp (49)
is the collisional stress tensor of the rth particle phase which accounts for the transfer
of linear momentum at collisions between alike particles over the distance 2rr
separating their centers. This physical phenomenon is important in dense fluidized
suspensions, where the total volume occupied by the particles is not negligible
compared with the volume of the vessel containing them. For rarefied granular
gases, which one can model adopting the Boltzmann-Grad limit, defined as:
r r ! 0; νr ! 1; νr r 2r bounded (50)
in which νr represents the overall number of particles belonging to solid phase S r,

because zr ! zrs ! zs, the collisional stress vanishes (this result is well known in
kinetic theory of gases; see, for instance, Chapman and Cowling 1970 or Gidaspow
1994). This is consequence of the principle of action and reaction, insofar as in the
Boltzmann-Grad limit it is:
20 L. Mazzei
" #
X X X X
ψ ðjx zr ðtÞjÞ f rs, a ðtÞ ! ψ ðjx zrs ðtÞjÞf rs, a ðtÞ ¼ 0 (51)
Sr Sr Sr Sr
Consider now the other contribution to the overall particle-particle contact force
appearing on the right-hand side of Eq. 44. This term, which represents the contact
forces acting between r particles of phase S r and s particles of phase S s, can be
expressed as:
" #
X X
ψ ðjx zr ðtÞjÞ f rs, a ðtÞ (52)
Sr Ss
where s 6¼ r and with particles r and s belonging to phases S r and S s, respectively.

Given its definition, this force should fulfill the principle of action and reaction, so
that:
" # " #
X X X X
ψ ðjx zr ðtÞjÞ f rs, a ðtÞ ¼ ψ ðjx zs ðtÞjÞ f sr, a ðtÞ (53)
Sr Ss Ss Sr
Clearly, this condition is not satisfied, since even if frs,a = fsr,a, it is ψ(|x zr|) 6¼
ψ(|x zs|). Only when rr ! 0 and rs ! 0, this equation holds. We conclude that the
force in Eq. 52 cannot be regarded as the interaction force between phases S r and
S s but must include an additional contribution which does not satisfy the action-
and-reaction principle. To find this force, we expand ψ(|x zrs|) in a Taylor series
about the point zr. Doing so gives:
" #
X X
ψ ðjx zr ðtÞjÞ f rs, a ðtÞ nr hf a irs
p
Sr Ss
h i
þ @b nr hPab irs rs
p ð1=2Þ@c nr hQabc ip (54)
where it is:
( )
X X
nr ðx, tÞhf a irs
p ðx, tÞ ψ ðjx zrs ðtÞjÞ f rs, a ðtÞ (55)
Sr Ss
( )
X X
nr ðx, tÞhPab irs
p ðx, tÞ ψ ðjx zr ðtÞjÞr r f rs, a ðtÞkrs, b ðtÞ (56)
Sr Ss
( )
X X
nr ðx, tÞhQabc irs
p ðx, tÞ ψ ðjx zr ðtÞjÞr 2r f rs, a ðtÞkrs, b ðtÞkrs, c ðtÞ (57)
Sr Ss
The second-order tensor so defined:

hDab irs rs rs
p nr hPab ip ð1=2Þ@c nr hQabc ip (58)
is the collisional stress tensor related to the momentum transferred at collisions

between phases S r and S s. We find it natural now to introduce the following tensor:
X
ν
hSab irp hCab irp þ p nr ρr V r h^
hDab irs v a^v b irp (59)
s6¼r¼1
This is the effective stress tensor of phase S r. The first two contributions, taken
together, represent the (total) collisional stress tensor, while the last, which arises
from the Reynolds decomposition of the convection term in the averaged dynamical
equation, represents the kinetic stress tensor.
In light of the results obtained above, we can express the averaged linear
momentum balance equation for phase S r as follows:
h i
ρr V r @t nr hva irp þ @b nr hva irp hvb irp
X
ν
¼ @b hSab irp þ nr hf a irp þ nr hf a irs
p þ nr ρr V r ga (60)
s6¼r¼1
In this equation, the interaction forces between the phases (represented by the second
and third terms on the right-hand side) satisfy the action-and-reaction principle.
Table 1 reports, in absolute notation, the multifluid equations of motion just derived.
The Problem of Closure
The averaged equations of motion for the fluid and solid phases just derived are
mathematically unclosed, for they feature terms related to point (i.e., microscopic)
variables. In their current form, therefore, the equations cannot be solved. An
example of such terms is given by the interaction force between the fluid phase
and the generic solid phase S r. This force, as seen, is equal to:
1 X ð
h f irp ðx, tÞ ψ ð j x zr j Þ Tðz, tÞkðz, tÞdσ z (61)
nr ðx, tÞ S @Λr
r
To calculate it, one needs to know the point fluid stress distribution T(x, t) over the
surface of each particle as well as the position of each particle. This distribution is
related to the point velocity field of the fluid, not to its volume average; because in a
macroscopic description of the flow this field and the particle positions are unknown,
Eq. 61 has no practical use (in a macroscopic modeling context).
22 L. Mazzei
Table 1 Eulerian-Eulerian averaged equations of motion for a system of n-particle classes

Continuity equation – fluid phase
@ te + @ x (ehuie) = 0
Continuity equation – particle phase r

@ t ϕr þ @ x ϕr hvirp ¼ 0
Dynamical equation – fluid phase
P
ν
ρe @ t ehuie þ @ x ehuie huie ¼ @ x hSie nr hfirp þ eρe g
r¼1
Dynamical equation – particle phase r
h i P
ν
ρr @ t ϕr hvirp þ @ x ϕr hvirp hvirp ¼ @ x hSirp þ nr hfirp þ nr hfirs
p þ ϕr ρr g
s6¼r¼1
We manipulated the averaged equations of motion in such a way that the closure
problem is confined to a small number of well-defined terms. These are the effective
stress tensor for each phase and the interaction forces between the fluid and each
solid phase and between each pair of solid phases. Overcoming the closure problem
means deriving expressions for them in terms of averaged variables only. Analytical
closures based on purely theoretical arguments are prohibitively difficult to obtain;
there is no guarantee that such equations even exist. Usually, the goal is far less
ambitious and is finding equations that consent to analyze the systems of interest
with the desired accuracy; such equations should be the simplest able to capture
enough physics to describe the fluid dynamics of the suspension satisfactorily.
In what follows, we first present some strategies for modeling the effective fluid
and solid stress tensors; we then analyze the mean fluid-particle interaction force,
laying emphasis on the buoyancy and drag forces, and the mean particle-particle
interaction force.
Effective Stress
Owing to the many contributions, yielded by the averaging process, that make up the
effective stress tensors, these are complex to model. Closing these quantities is
further complicated by the absence of experimental measurements having a direct
bearing on them. Notwithstanding, researchers usually suppose that both fluid and
solid phases behave as Newtonian fluids, writing:

2
Se ¼ pe κe μe trDe I þ 2μe De ;
3
(62)
2
Sr ¼ pr κr μr trDr I þ 2μr Dr
3
where pe, pr, κe, κr, μe, and μr are the averaged pressures, dilatational viscosities and
shear viscosities of the fluid, and rth solid phase, respectively; furthermore, I is the
identity tensor, while De and Dr. are the rate of deformation (or strain) tensors,
defined as:
1 1
De @x ue þ @x uTe ; Dr @x vr þ @x vTr (63)
2 2
From now on, as done in these last expressions, we simplify the notation by leaving
out the angular brackets that imply averaging. Experimental evidence has shown that
in several fluidization regimes, the assumption of Newtonian behavior is satisfactory,
especially for powders far from maximum packing.
If Eq. 62 holds, the closure problem reduces to finding suitable constitutive
expressions for the pressure, dilatational viscosity, and shear viscosity of each
phase. As done in section “Averaged Equations of Motion for Fluid-Particle Sys-
tems,” often one assumes that the fluid is incompressible, so that no constitutive
equation is required for pe, that κe is zero and that μe is proportional to e, the
proportionality constant being the shear viscosity of the pure fluid.
For the solid phases, constitutive expressions for these quantities have been
derived from granular kinetic theory (Gidaspow 1994; Brilliantov and Poschel
2004), a generalization of the mathematical theory of dense nonuniform gases
(Chapman and Cowling 1970). The idea is that, because dense granular gases
resemble in many ways dense molecular gases, the constitutive equations that
govern the two should be derivable, at least in part, from the same theoretical
framework. Similarly to a molecular gas, particle pressure and viscosities are
functions of a granular temperature, which is governed by a balance equation for a
pseudointernal energy related to the particle peculiar velocity. For solid phase S r,
the balance equation is:
ρr ½@t ðϕr U r Þ þ @x ðϕr U r vr Þ ¼ @x qr þ Sr : @x vr þ Gd, r Sv, r Sc, r (64)
where Ur 3/2Θr is the pseudointernal energy per unit mass, Θr being the granular
temperature, and qr. is the pseudothermal heat flux. The equation differs from the
classical internal energy balance equation (Bird et al. 1960) because of a sink term
Sc,r(x, t) representing energy degradation caused by inelastic collisions, a source
term Gd,r(x, t) representing generation of particle velocity fluctuations by fluctuating
fluid-particle forces, and a sink term Sv,r(x, t) representing their dampening by
viscous resistance to particle motion. qr. is usually modeled using Fourier’s law,
writing:
qr ¼ kr @x Θr (65)
where kr is the granular thermal conductivity of the rth solid phase. Different
closures have been developed for this parameter; Gidaspow et al. (1992), for
instance, proposed:
24 L. Mazzei
150ρr sr ðπΘr Þ1=2

kr ¼ ½1 þ ð6=5Þϕr αr ð1 þ er Þ2
384αr ð1 þ er Þ
þ 2ϕ2r ρr sr αr ð1 þ er ÞðΘr =π Þ1=2 (66)
where sr denotes the particle diameter, er the coefficient of restitution for particle
collisions, and αr the radial distribution function for the rth solid phase. Various
expressions are available for this function; for instance, that advanced by Iddir and
Arastoopour (2005) reads:
X
ν
αr ¼ ½1 ðϕ=ϕmax Þ1 þ ð3sr =2Þ ðϕs =ss Þ (67)
s¼1
where ϕ is the overall solid volume fraction and ϕmax is the maximum solid
compaction (i.e., the maximum value which ϕ can take). Here αr diverges positively
when ϕ approaches ϕmax. An expression in which αr is bounded is that of Lebowitz
(1964), where ϕmax does not feature:
" #
X
ν
αr ¼ ð1=eÞ 1 þ ð3sr =2eÞ ðϕs =ss Þ (68)
s¼1
To close Eq. 64, one needs constitutive equations also for the terms Gd,r, Sv,r, and Sc,r.
For briefness, we do not report them; the interested reader may refer, for instance, to
Gidaspow (1994), Syamlal et al. (1993), Fan and Zhu (1998) and Jackson (2000).
Various closures for the solid pressure are available in the literature, all derived
from the granular kinetic theory. As an example, we report the expression advanced
by Lun et al. (1984), suitably extended to cater for polydisperse suspensions:
" #
X
ν
3
pr ¼ 1 þ 2 ðsrs =sr Þ ϕs αrs ð1 þ ers Þ ϕr ρr Θr (69)
s¼1
where sr and ss are the particle diameters for phases r and s, respectively, ers is the
coefficient of restitution for collisions between particles of phases r and s, while:
srs ðsr þ ss Þ=2; αrs ðsr αs þ ss αr Þ=ðsr þ ss Þ (70)
As an example of constitutive equations for the solid-phase dilatational and shear

viscosities, we report those given in Gidaspow (1994), even if several are available
in the literature:
κr ¼ ð4=3Þϕ2r ρr sr αr ð1 þ er ÞðΘr =π Þ1=2 (71)

10ρr sr ðπΘr Þ1=2

μr ¼ ½1 þ ð4=5Þϕr αr ð1 þ er Þ2
96αr ð1 þ er Þ
þ ð4=5Þϕ2r ρr sr αr ð1 þ er ÞðΘr =π Þ1=2 (72)
These expressions are those originally developed for monodisperse suspensions and
do not directly account for the presence of the other solid phases.
The expressions given above, as said, are based on the kinetic theory of granular
flows. This assumes that particles are smooth and spherical, that collisions are binary
and instantaneous, and that the suspension is far from the frictional packing limit,
which marks the transition from the viscous to the frictional flow regime. In the first
regime, particles undergo transient contacts, momentum transfer is translational and
collisional, and the granular kinetic theory holds; in the second, particles undergo
enduring contacts, and momentum transfer is mainly frictional. Granular kinetic
theory does not account for these interactions, and thus in the frictional flow regime,
the closures reported above are inadequate.
In regions of high solid volume fraction, particles interact with multiple neighbors
and the mechanism for stress generation is not merely due to kinetic and (particu-
larly) collisional contributions but also to sustained contacts among particles. Such
contacts make particles dissipate considerable energy, letting them form very dense
regions in the bed. This increases the ability of the granular assembly to resist
shearing, for tangential frictional forces at contact points are now present. Hence,
the solid viscosity is larger than that predicted by the granular kinetic theory.
To describe the frictional stress, other models, empirical, phenomenological, or
based on the theory of soil mechanics, are needed. Usually, one assumes that it is:
S⋆ ⋆ ⋆
r ¼ pr I þ 2μr Dr (73)
where the star indicates that the quantity refers to the frictional flow regime. Syamlal
et al. (1993) proposed this equation for the frictional pressure:
B
p⋆ ⋆
r ¼ ϕr p , p⋆ 10A ϕ ϕf (74)
where ϕf denotes the frictional solid packing (the solid volume fraction threshold
value at which the powder enters the frictional flow regime). The coefficients A and
B are very high, with typical values of 25 and 10, respectively. For other constitutive
equations, the reader is referred to the literature. An expression often used for the
frictional shear viscosity is that of Schaeffer (1987), which reads:
!1
p⋆
r sin ϑr
X
ν
1h i
μ⋆
r ¼ p ffiffiffiffiffiffiffiffiffiffiffiffiffi ϕs , I 2 ðD r Þ ðtrDr Þ2 trD2r (75)
2 I 2 ðDr Þ s¼1 2
where ϑr is the angle of internal friction of the rth granular material, while I2(Dr) is
the second invariant of the rate of deformation tensor. Other expressions are
26 L. Mazzei
available in the literature. In the frictional flow regime, one usually accounts also for
the kinetic and collisional contributions to the solid stress; the easiest way to do this
is adding the viscous stress tensor to the frictional one.
Fluid-Particle Interaction Force
There are five main contributors to the fluid-particle interaction force. The first is the
buoyancy force, whose definition in the context of multiphase flows is not unique
and needs to be discussed. The second acts in the direction of the fluid-particle slip
velocity – that is, the fluid velocity relative to an observer moving with the same
local mean velocity as the particles. The third is normal to the slip velocity, the fourth
is parallel to the relative acceleration between the phases, and the fifth is proportional
to the local mean acceleration of the fluid. The last four terms are commonly referred
to as drag force, lift force, virtual mass force, and local fluid acceleration force,
respectively. As we shall see, the local fluid acceleration force is not always present
but features only when one definition of buoyancy force is used – in particular, the
classical definition presented later on. Among these five terms, often the buoyancy
and drag forces are dominant.
Buoyancy Force
A first definition sets this force equal to the weight of the fluid displaced by the solid;
accordingly, if we refer the force to the unit volume of suspension, it is:
nr fB⋆, r ϕr ρe g (76)
Since it is consistent with the Archimedes’ principle original formulation, we call

this classical definition. For a given value of ϕr, this force is constant, being
unrelated to the flow.
The second definition relates the force to the effective fluid stress tensor, as
reported by Jackson (2000); per unit volume of suspension, it is:
nr fB•, r ϕr @x Se (77)
Another definition often encountered in the literature considers solely the isotropic
part of the effective stress tensor of the fluid; the closure therefore takes the form:
nr fB∘, r ϕr @x pe (78)
These definitions lead to different values of the buoyancy force. There is nothing
wrong with this, for we are free to define this force as we like: what is crucial is that
the total fluid-particle interaction force nr fr, which has an objective physical
meaning, be correctly calculated. Thus, modelers who adopt different definitions
of buoyancy force will also need to employ different expressions for the comple-
mentary force that makes up the total fluid-particle interaction force. The value of the
latter must be the same in all models. So, for instance, if one opts to use Eq. 78, the
contribution of the deviatoric part of Se must be included in the complementary
force: it will still be present, the only difference being that it is not regarded as part of
the buoyancy force, but as part of the complementary force.
To better understand the meaning of these definitions, consider a monodisperse
suspension of motionless particles equally distributed in space (i.e., an ideal homo-
geneous bed). The second and third definitions here coincide, because De vanishes
and Se is therefore isotropic:
Se ¼ pe I ) nfB• ¼ nfB∘ (79)
where n is the particle number density. We can derive an expression for this force
using the mean dynamical equations reported in Table 1; these reduce to:
nf ¼ @x pe þ eρe g; nf ¼ ϕρs g (80)
where ϕ denotes the solid volume fraction, ρs the particle density, and n f the fluid-
particle interaction force. Subtracting the two equations and using Eq. 78 gives:
@x pe ¼ ðeρe þ ϕρs Þg; nfB∘ ¼ nfB∘ ¼ ϕðeρe þ ϕρs Þg (81)
Thus, for ideal uniform fluidized beds, the difference between the first and the other
two definitions reduces to the density choice in the force expression: the first requires
the fluid density while the second the suspension bulk density. For further details
about this topic, we refer to Jackson (2000).
Local Fluid Acceleration Force

If the classical definition of buoyancy force is employed, the complementary force to
the total fluid-particle interaction force must include a term known as local fluid
acceleration force (this term is absent otherwise). Per unit volume of suspension, this
force is given by:
nr fA⋆, r ϕr ρe Det ue , Det ue @t ue þ ue @x ue (82)
where the derivative on the right-hand side is a material derivative relative to a

Lagrangian observer moving with the locally averaged velocity of the fluid.
If the fluid acceleration is far less than the gravitational acceleration, the local
fluid acceleration force is far less than the buoyancy force, and so its contribution is
negligible. This force, nonetheless, is conceptually important, as the following
thought experiment reveals. Consider a uniform assembly of particles at rest in a
body of fluid. The fluid is also at rest in a vertical container placed on a horizontal
plane. The system resides in a uniform gravitational field. If the plane supporting the
container and the constraints keeping the particles at rest are suddenly removed, the
entire system falls freely with an acceleration equal to g. Since the mean velocity
fields of both phases are uniform and no pressure gradients are present, the effective
28 L. Mazzei
stress tensors of both phases vanish, and the dynamical equations in Table 1 reduce
to:
eρe Det ue ¼ nf þ eρe g; ϕρs Dst vs ¼ nf þ ϕρs g (83)
For convenience, we have used the nonconservative formulation of the equations; to

obtain them, one must combine the dynamical and continuity equations (see, for
instance, Bird et al. 1960). The material derivative for the solid phase is defined
similarly to that for the fluid phase. In the case at hand, both material derivatives are
equal to the gravitational acceleration, and Eq. 83 leads to the same result: the fluid-
particle interaction force must vanish. This condition can be met only if the local
fluid acceleration force is accounted for. As the two phases move identically, no slip
velocity and acceleration are present between them; consequently, the drag, virtual
mass, and lift forces are all zero (see the sections below dedicated to these forces).
Conversely, if its classical definition is adopted, the buoyancy force is nonzero. So,
the total fluid-particle interaction force can vanish only if the local fluid acceleration
force is considered:
nf ¼ nfB⋆ þ nfA⋆ ¼ ϕρe g þ ϕρe g ¼ 0 (84)
If the other definitions of buoyancy force are adopted, the local fluid acceleration
force must not be included, because in both cases, being the effective fluid stress
tensor zero, the buoyancy force vanishes. Note that this is not only true for solid
suspensions but also for single bodies moving in pure fluids.
A Consideration on the Complementary Force

As stated, all models need to agree on the value ascribed to the total fluid-particle
interaction force, but they can use different repartitions for such force. A model may
use the classical definition for the buoyancy force and include the local fluid
acceleration force, while another may adopt one of the buoyancy force definitions
given in Eqs. 77 and 78 without including the local fluid acceleration force. Both
choices are acceptable, but the models will have to adopt different expressions for
the complementary force to the total fluid-particle interaction force. Let us write:

nr fr ¼ ϕr ρe Det ue g þ nr fr⋆ ; nr fr ¼ ϕr @x Se þ nr fr• (85)
We wonder how the complementary forces nr fr⋆ and nr fr• are related. To answer this
question, we consider the following relation, obtained by combining the equations
above:

nr fr• ¼ nr fr⋆ þ ϕr ρe Det ue ρe g @x Se (86)
Then, using the dynamical equation for the fluid phase reported in Table 1, with a
few mathematical passages not reported for briefness, one can prove that:
X
ν
ϕr ρe Det ue ρe g @x Se ¼ ðϕr =eÞ ns fs• (87)
s¼1
We now use this relation in Eq. 86 and then sum both sides of the resulting equation
over the phase index r. Doing so gives:
X
ν
nr fr⋆ ð1=eÞnr fr• ¼ 0 (88)
r¼1
This is the condition that needs to be satisfied to render the two models consistent.
This condition can be met by imposing the following restriction:

nr fr• ¼ e nr fr⋆ (89)
If we denote as τ e the deviatoric part of the effective stress tensor of the fluid, then,
using the equation above, we immediately obtain:

nr fr∘ ¼ e nr fr⋆ þ ϕr @x τ e (90)
where nr fr∘ represents the complementary force to the total fluid-particle interaction
force that arises when Eq. 78 is employed for defining the buoyancy force.
The main constituents of the complementary forces defined above are the drag
force, the lift force, and the virtual mass force. We will now discuss how these forces
are expressed constitutively.
Drag Force
By definition, the drag force is parallel to the fluid-particle slip velocity (a vector that
fulfills the principle of material frame-indifference); hence, it is:
nr f D , r β r ð ue v r Þ (91)
where βr denotes the drag coefficient for the rth particle phase. Finding a closure for
the drag force amounts to finding a constitutive expression for βr. We now report
some of these expressions, written in a way that is consistent with the classical
definition of buoyancy force.
Ergun and Orning (1949) developed an empirical correlation for assessing the
unrecoverable pressure drop through packed beds. Extending its range of validity to
homogeneous fluidized suspensions, one obtains the following constitutive
equation:
μ e ϕr ð 1 e Þ ρe ϕr j ue v r j
βr ¼ 150 þ 1:75 (92)
ðesr Þ 2 esr
Gidaspow (1994) recommends using this closure for values of the void fraction up to
0.80, even if the Ergun equation was developed (and has been extensively verified)
30 L. Mazzei
for fixed beds in which the void fraction is small, with values close to 0.40. For void
fraction values larger than 0.80, Gidaspow (1994) recommends using the expression
of Wen and Yu (1966), which is one of the most popular correlations for the
calculation of the drag coefficient in dense fluidized suspensions:
3 ρ ϕ jue vr j 2:70
βr ¼ CD ðRer Þ e r e (93)
4 sr
where:

ρ ejue vr jsr ð24=Rer Þ 1 þ 0:15Re0:687 for Rer < 1000
Rer e ; CD ðRer Þ ¼ r
μe 0:44 for Rer 1000
(94)
The expression above has been proposed by Schiller and Naumann (1935). In
Eq. 93, as we see, the exponent in the voidage function eα is constant and equal
to 2.70.
Di Felice (1994) suggested that the exponent α should be a function of the particle
Reynolds number; the expression that he proposed is:
h i
αðRer Þ ¼ 2:70 0:65exp ð1=2Þð1:50 log10 Rer Þ2 (95)
The value of the exponent reduces to that employed in the expression of Wen and Yu
(1966) for very small and very large values of the Reynolds number. In the
intermediate region, nevertheless, the deviation from 2.70 is significant, the expo-
nent reaching a minimum value of 2.05 when Rer 32.
The closures reported above are extensively used; nevertheless, they are not
consistent with the empirical equation developed by Richardson and Zaki (1954)
to describe the expansion of homogeneous fluidized beds of non-cohesive particles.
Since this equation is very accurate, the inconsistency is a shortcoming of the drag
force closures. To overcome this limitation, Mazzei and Lettieri (2007) derived an
expression that is consistent with the Richardson and Zaki correlation over the entire
range of fluid dynamic regimes and for any value of the suspension void fraction. It
has the following formulation:
8h i2 9
>
< 0:63 þ 4:80ðRer =eγ Þ1=2 >
=
2ð1γ Þ
αðe, Rer Þ ¼ ð1=lneÞln e (96)
>
: 0:63 þ 4:80Re1=2
2 >
;
r
where:
4:80 þ 2:40 0:175ðRer =eγ Þ3=4

γ ðe, Rer Þ ¼ (97)
1 þ 0:175ðRer =eγ Þ3=4
To calculate γ one needs to solve a nonlinear equation. Since γ has a very narrow
range of variation, finding the solution requires few iterations. For a detailed
discussion on how this closure was derived and on how it compares with the other
expressions reported above, we refer to Mazzei and Lettieri (2007).
Virtual Mass and Lift Forces

If a body immersed in a fluid accelerates, some of the surrounding medium must also
accelerate; this results in a force, named virtual mass force, equal to:

nr fV , r ϕr ρe CV ðϕr Þ Det ue Drt vr (98)
where Det ðÞ and Drt ðÞ are the material derivatives associated with the fluid and rth
solid phase, respectively. The virtual mass coefficient, denoted as CV(ϕr), depends
on the particle shape and on the volume fraction of the solid phase considered. For
very dilute mixtures of spherical particles, CV(ϕr) is taken to be 1/2, since this is the
calculated value for a single sphere in an infinite fluid (Maxey and Riley 1983). The
same result was found by Zhang and Prosperetti (1994) for an inviscid fluid and low
particle concentration. For larger values of the solid volume fraction, the coefficient
is expected to increase. Using lattice-Boltzmann simulations (but for bubbly sus-
pensions), Sankaranarayanan et al. (2002) showed that CV(ϕr) is nearly linear; at
moderate values of the solid volume fraction, Zuber (1964) suggested that:
CV ðϕr Þ ¼ ð1 þ 3ϕr Þ=2 (99)
The virtual mass force is important when the density of the fluid is higher than that of
the disperse phase; so, in fluidized beds, especially when the fluidization medium is a
gas, this force usually plays a secondary role. In bubble columns, conversely, it
strongly affects the dynamics of the system.
If an object moves in a fluid which is in shearing flow, it experiences a force
transverse to the direction of relative motion. This lift force is equal to:
nr fL, r ϕr ρe CL ðϕr Þð@x ue Þ ð ue v r Þ (100)
The lift coefficient, denoted as CL(ϕr), depends on the particle shape and on the
volume fraction of the solid phase considered. For very dilute mixtures of spherical
particles, CL(ϕr) is also taken to be 1/2. One reason for this is that Eqs. 98 and 100
are not frame independent when taken separately, but their sum satisfies the principle
of material objectivity if the coefficients of the two forces are equal. Hence, to satisfy
this principle, one should set CL(ϕr) = CV(ϕr). In fluid-solid systems, the lift force is
often (slightly) more important than the virtual mass force, but both forces are
outweighed by the drag force. For more details about these forces, we refer to
Marchisio and Fox (2013) and the references therein provided.
32 L. Mazzei
Other Forces
Other contributions to the fluid-particle force could be considered. A comprehensive
overview can be found in Drew and Passman (1998). Here we cite only the Faxen
force and a history-dependent term analogous to the Basset force for the motion of
isolated particles (Basset 1888). For the latter, we can reasonably believe that, for
fluidized suspensions, the averaging of history-dependent forces should result in a
vanishing contribution, since averaging would most probably erase any historical
effects of the motion of the particles on the fluid in their immediate neighborhood.
We thus expect this force to be negligible.
Particle-Particle Interaction Force
In fluidized mixtures of several monodisperse particle classes, each class exchanges

linear momentum with all the others; this momentum transfer arises from particle
collisions and results into a particle-particle drag force. Soo (1967) was among the
first to quantify it, deriving a theoretical expression for the force acting on a single
particle of species r in a cloud of particles of species s. Nakamura and Capes (1976)
and Arastoopour et al. (1982) made similar efforts. Many authors have since then put
forward other correlations, most of them being variations of earlier works. The force
is expressed as the product of a drag coefficient by the velocity of slip between the
particle classes:
nr frs ζ rs ðvs vr Þ (101)
where nrfrs is the force exerted by phase s on phase r per unit volume of suspension
(see Table 1) and ζ rs is the particle-particle drag coefficient for the two-particle
classes involved. The closure problem reduces to finding a constitutive expression
for ζ rs. Gidaspow et al. (1985) advanced the relation:
" #
ϕr ϕs ρr ρs ðsr þ ss Þ2
ζ rs ¼ Crs ð1 þ ers Þ j vs vr j (102)
ρr s3r þ ρs s3s
where ρr, ρs, sr, and ss are the densities and diameters of the particles of classes r and
s, respectively, ers is their coefficient of restitution, and Crs is given by:
3Φ1=3 þ ðϕr þ ϕs Þ1=3

Crs h rs i (103)
1=3
rs ðϕr þ ϕs Þ
4 Φ1=3
where:
Φr
Φrs ð1 srs Þ½Φr þ ð1 Φr ÞΦs ð1 Xrs Þ þ Φr for Xrs
Φr þ ð1 Φr ÞΦs
Xrs
Φrs ½ðΦr Φs Þ þ ð1 srs Þð1 Φr ÞΦs ½Φr þ ð1 Φr ÞΦs þ Φs otherwise
Φr
(104)
In the relations above, Φr and Φs are the particle volume fractions at maximum
packing for phases r and s, respectively; moreover, it is:
1=2 1=2
ϕr ss sr
Xrs ; srs if sr ss and srs otherwise (105)
ϕ r þ ϕs sr ss
Another popular closure is that of Syamlal (1987), which reads:

" #
3 π ϕr ϕs ρr ρs grs ðsr þ ss Þ2
ζ rs ¼ ð1 þ ers Þ 1 þ Frs j vs vr j (106)
4 4 ρr s3r þ ρs s3s
in which Frs denotes a coefficient of friction for phases r and s while grs the radial
distribution function of Lebowitz (1964), given by Eq. 68. Gera et al. (2004)
suggested that the equations above should include an additional term that is neces-
sary to prevent the particle phases from segregating when they are fully packed.
Without it, Eqs. 102 and 106 permit packed particles of different size to segregate, a
phenomenon which is not observed experimentally. To prevent this, they
recommended adding to the coefficient ζ rs the term Ψp⋆, where p⋆ is given by
Eq. 74 and Ψ is a coefficient that must be adjusted to match the actual segregation
rate of the powder considered. The value that Gera et al. (2004) used was 0.30, but
they stressed that this is not of general validity. Ψp⋆ is added so that when the
powder approaches maximum packing, the particle-particle drag increases suffi-
ciently to make the solid phases r and s move together as if they were one phase,
thereby hindering segregation. This additional term is included only for ϕ > ϕf.
Population Balance Modeling
In the previous sections, we have presented the Eulerian equations of motion for
dense fluidized suspensions constituted of ν-particle classes, the rth class being
characterized by a density ρr and a diameter sr. A serious limitation of this modeling
approach is that changes in particle size are not permitted: particles can segregate or
mix, and so the particle size distribution (PSD) in every real-space point can change
in time, but the size of the particles for each class is fixed. In general, nevertheless,
particles can grow, shrink, aggregate, and break, and new particles, of vanishing
small size, may nucleate; these size changes reflect the physical and chemical
processes taking place in the system and strongly affect the evolution of the PSD.
34 L. Mazzei
Predicting this evolution is essential for a realistic description of the system behavior.
We now introduce a modeling approach, referred to as population balance modeling,
which has this capability.
Population balance modeling is statistical in nature; it can be regarded as a
generalization of the statistical modeling approach presented in section “Statistical
Averaging.” There, the state of a particle was identified by two coordinates: position
in real space x and velocity v; this number can be increased, if additional properties
are required to fully characterize the particle state. Here we will add only the particle
size s. The complete description (in a statistical sense) of the system is given by the
master joint PDF, which we introduced in section “Statistical Averaging”; now, this
function depends on 7ν internal coordinates plus the time coordinate (the phase
space of the entire particulate system has 7ν dimensions). To calculate many of the
macroscopic properties of practical interest, however, the one-particle marginal PDF,
or equivalently the number density function, suffices; this is because many of the
microscopic functions of interest take the following form:
X
ν X
ν
bð r Þ ¼ b1 ðxs , vs , ss Þ ¼ b1 ð r s Þ (107)
s¼1 s¼1
where b1 is an arbitrary function of the phase-space state of one particle, and where r
and rs are the position points of the entire particulate system and of particle s in their
phase spaces, respectively. For a function of this kind, one can prove that:
ð ð ð
hbis ðtÞ ¼ b1 ðx1 , v1 , s1 Þf 1 ðx1 , v1 , s1 , tÞ dx1 dv1 ds1 (108)
Ωx Ωv Ωs
where Ωs is the range of variation of s, while f1 is the NDF. By definition, f1(x, v, s, t)

dxdvds represents the expected number of particles located at time t in the volume dx
around the point x with velocity in the range dv around the velocity v with size in the
range ds around the size s. f1(x, v, s, t)dxdvds, therefore, is an observable
representing the mean particle number density in the seven-dimensional phase
space made up by the union of the real space Ωx, velocity space Ωv, and size space
Ωs. Knowing the NDF is equivalent to knowing the particle size and velocity
distributions in any real-space point at any time. In Eq. 108, 〈b〉s is only a function
of time, since b does not depend on real-space, velocity-space, and size-space
coordinates.
So, knowing the NDF permits calculating observables associated with micro-
scopic functions of the class defined by Eq. 107. To calculate observables of this
kind, one has to know how the NDF evolves in the phase space of one particle. An
evolution equation for it is hence necessary. This is called (generalized) population
balance equation (PBE). One may derive this equation rigorously starting from the
microscopic description of the particulate system, given by the transport equation of
the master joint PDF, which is a generalization of the Liouville equation (Marchisio
and Fox 2013). Here we follow a less rigorous, and therefore easier, method that
regards the population balance equation as a simple continuity statement written in

terms of the number density function in the phase space of one particle (this space in
our case has seven dimensions: three in real space, three in velocity space, and one in
size space). This is the most popular derivation method used in the literature on
polydisperse fluid-particle systems.
Consider an arbitrary, fixed control volume Λr Λx[Λv[Λs in the phase space of
a single particle. The number of particles that accumulate in it per unit time is:
ð ð
ACC ¼ @t f 1 dx ¼ @t f 1 dr (109)
Λr Λr
where, to simplify the notation, we have denoted as r the position point of the
particle in its phase space. The operations of time differentiation and space integra-
tion can be interchanged insofar as the control volume is not time dependent. The net
number of particles entering Λr per unit time is:
ð ð
IN OUT ¼ _ r dσ r ¼
f 1 rn _
@r f 1 rdr (110)
@Λr Λr
where r_ and @ r are the particle velocity and the nabla operator in phase space,
respectively, and where nr is the unit vector normal to the hypersurface bounding Λr
directed outward. To turn the surface integral into a volume integral, we have used
the Gauss theorem.
The difference between the two terms above has to balance the net number of
particles generated per unit time within Λr. Particle generation is caused by colli-
sions, breakage, aggregation, and similar instantaneous processes (no process, of
course, is instantaneous; however, these processes have characteristic times that are
so smaller than those characterizing the evolution of the NDF that we can regard
them as instantaneous). For instance, if two particles located outside Λv collide, their
velocities vary abruptly, and after the collision, one particle (or even both) might be
located within Λv, having thus entered Λr without crossing its boundaries. If we
denote as Gr the net number of particles generated per unit volume of phase space
and unit time owing to instantaneous phenomena, it is:
ð
GEN ¼ Gr dr (111)
Λr
If we equate the accumulation term to the sum of the convection and generation
terms, we obtain, after a few minor rearrangements, the following integral equation:
ð
ð@t f 1 þ @r f 1 r_ Gr Þdr ¼ 0 (112)
Λr
Because the integration volume Λr is arbitrary, and the integrand is (assumed to be)
continuous, we conclude that the integrand must vanish:
36 L. Mazzei
@t f 1 ¼ @r f 1 r_ þ Gr (113)
This is the PBE. We find it convenient to rewrite it in terms of the velocities which
the particles possess in the real, velocity, and size spaces. Letting @ v and @ s represent
the nabla operators in the velocity and size spaces, respectively, and v_ and ṡ be the
particle velocities in the velocity and size spaces, respectively, ( v_ represents the
particle acceleration in real space and s_ the particle growth rate), we can write:
@t f 1 ¼ @x f 1 v @v f 1 v_ @s ðf 1 sÞ
_ þ Gr (114)
or equivalently:
@t f 1 ¼ v@x f 1 @v f 1 v_ @s ðf 1 sÞ
_ þ Gr (115)
The two expressions are equivalent insofar as the real-space particle velocity v is an
independent coordinate and not a function of x; the same is not true for v_ and ṡ, which
in general may depend on the coordinates v and s, respectively. For instance, since v_
is the real-space particle acceleration, and since this is equal to the total force per unit
mass acting on the particle, if the latter depends on the particle velocity, also v_ will.
This is surely the case in fluidized beds, where a component of the force is the drag.
The PBE, as said, is the transport equation of the NDF. Solving it allows
determining the NDF evolution. The equation, nevertheless, can be solved only if
it is closed. Here by closed we mean that all the terms in the equation can be
computed from knowledge of the number density function (of course, these func-
tionals need to be known). This is not the case for the generation term, because, as
we know from statistical mechanics, it involves correlations between two particles
(involving therefore the two-particle marginal PDF). A closure, consequently, will
have to be introduced to express Gr in terms of the NDF. This is a significant
challenge, because these closures are in general complex to derive (Balescu 1975).
Once this has been done, the PBE is closed, but its solution will be extremely
difficult to obtain. This is because in general, the PBE results to be a nonlinear,
integral, partial differential, functional equation in a seven-dimensional space. As a
consequence, one does not usually attempt to solve it, using the equation to extract
solely the information about the system behavior that is of interest in the application
at hand.
The topics of how to close the PBE and how to solve it are vast. We thus refer to
the specialized literature (we strongly recommend Chapman and Cowling 1970;
Gidaspow 1994; Ramkrishna 2000 and Marchisio and Fox 2013). Here we just
briefly mention a powerful solution method that allows tuning the PBE into a set of
four-dimensional equations that can be solved with normal computational fluid
dynamics numerical codes. This is called quadrature-based moment method (several
variants exist, but all of them are based on the same idea, which we will now
present).
Often engineers are only interested in few integral properties of the NDF. Called
moments, these may be important because they control the product quality or
because they are simple to measure and monitor. The idea behind the method of
moments is to derive transport equations for the moments of interest by integrating
out the coordinates v and s from the PBE. For any given function φ(v, s), we can
write:
ð ð ð ð ð ð
@t φðv, sÞf 1 dvds ¼ _
φðv, sÞ@r f 1 rdvds þ φðv, sÞGr dvds (116)
Ωv Ωs Ωv Ωs Ωv Ω s
The integral on the left-hand side is the moment of f1 associated with the function
φ(v, s) and depends only on x and t. As a consequence, the equation above, which
governs the evolution in time and real space of the moment of the NDF associated
with φ(v, s), can be solved by any CFD numerical code. We have therefore overcome
the dimensionality issue. The problem with the equation above is that it is usually
unclosed, since for any set of moments which the modeler wishes to track, obtained
with a finite set of functions φ1, . . ., φn, the equations involve also moments external
to the set.
To overcome the closure problem, we can operate as follows. As mentioned, the
moment method aims to solve the dimensionality issue by turning a problem
involving one higher-dimensional differential equation into a problem involving a
set of four-dimensional differential equations solvable by a CFD code. To capture all
the information contained in the PBE, one would have to consider an infinite set of
equations. But since we neither want nor can solve an infinite number of equations,
the idea behind the method of moments is to satisfy only a finite number of them.
This leaves the NDF largely undetermined, because only the infinite set would yield
the correct NDF. This means that we can choose – to a certain extent – the NDF
arbitrarily and then let the moment equations determine the details which we have
left unspecified. Moment methods differ in the choice of the function φ and in the
arbitrary input for the NDF. Their common feature is to choose the latter so that f1 is a
given function of v and s containing 3α undetermined parameters (two scalars and
one vector) depending on x and t. So, if we take 5α scalar moment transport
equations, we obtain 5α differential equations for the unknown parameters. One
hopes that, for α sufficiently large, the result is accurate enough and independent of
the form chosen for the NDF.
The quadrature methods of moments are examples of this approach; they over-
come the closure problem by assuming that the NDF has the following functional
expression:
X
α
f 1 ðx, v, s, tÞ ¼ nr ðx, tÞδ½v vr ðx, tÞδ½s sr ðx, tÞ (117)
r¼1
This is a quadrature formula, in which α is the number of nodes, vr(x, t) and sr(x, t)
are the rth quadrature nodes, and nr(x, t) is the rth quadrature weight. This formula
represents the particle population by means of α solid phases, the rth having number
density nr(x, t) and being made up of particles with velocity vr(x, t) and size sr(x, t).
38 L. Mazzei
The difference between this representation and that used in section “Averaged
Equations of Motion for Fluid-Particle Systems” is that here the size of each particle
class is not fixed but evolves in time and space. Here the 3α parameters which one
must obtain via the moment transport equations are nr(x, t), vr(x, t), and sr(x, t). For
details about how this is done, we refer the reader to the literature previously cited.
Conclusions
We presented three strategies for modeling fluidized beds: Eulerian-Lagrangian

modeling, discrete particle modeling, and Eulerian-Eulerian modeling. Tracking
the motion of each particle, the first two give a detailed description of the system
dynamics; these methods, however, are too expensive computationally to be of any
use for describing systems of industrial interest. We thus focused on Eulerian-
Eulerian modeling, describing the averaging techniques that turn granular systems
into continuous media and deriving the volume-averaged equations of mass and
linear momentum balance for fluidized suspensions made up of ν-particle classes.
We then addressed the closure problem, describing the main constitutive equations
used by modelers to express the fluid-particle and particle-particle interaction forces
and the effective fluid dynamic stress. We concluded the chapter by introducing the
population balance modeling, which permits describing systems in which the parti-
cles are continuously distributed over the size and in which the size is free to vary
owing to continuous and discontinuous processes, such as chemical reaction,
growth, aggregation, and breakage.
Appendix
Fluid-Phase Volume Average of Point Variable Spatial Derivatives
We intend to derive an expression for the fluid-phase volume average of point

variable spatial derivatives; to this end, we start by considering the derivative:

@a eðx, tÞhζ ie ðx, tÞ (118)
Now, using the definition of fluid-phase volume average given in Eq. 17 and the
derivation chain rule, we write the quantity above as:
Ð Ð Ð
@xa Λe ζ ðz,tÞψ ðjx zjÞdz ¼ Λe ζ ðz,tÞ@xa ψ ðjx zjÞdz ¼ Λe ζ ðz,tÞ@za ψ ðjx zjÞdz
Ð Ð (119)
¼ Λe ½@za ζ ðz,tÞψ ðjx zjÞdz Λe @za ½ζ ðz,tÞψ ðjx zjÞdz
For the first integral, we can write:

ð
½@za ζ ðz, tÞ ψ ðjx zjÞdz ¼ eðx, tÞh@a ζ ie ðx, tÞ (120)
Λe
For the second, the Gauss theorem allows writing:

Ð
Λe @za ½ζ ðz, tÞψ ðjx zjÞdz
Ð ν Xð
X
¼ @Λx ζ ðz, tÞna ðz, tÞψ ðjx zjÞdσ z ζ ðz, tÞka ðz, tÞψ ðjx zjÞdσ z
r¼1 S r @Λr
(121)
where @Λx is the surface bounding the domain containing the mixture and na(x, t) is
the ath component of the unit vector normal to @Λx pointing away from the mixture.
If the shortest distance from the generic point x @Λx is considerably larger than
the weighting function radius, the first term of the right-hand side of the equation
above is much smaller than the second. Neglecting it, we obtain Eq. 25.
Fluid-Phase Volume Average of Point Variable Time Derivatives
Similarly, to derive an expression for the fluid-phase volume average of point

variable time derivatives, we start by considering the derivative:

@t eðx, tÞhξie ðx, tÞ (122)
Using the definition of fluid-phase volume average given in Eq. 17 and then applying
the Leibnitz theorem allows writing this as:
Ð Ð
@t Λe ζ ðz, tÞψ ðjx zjÞdz ¼ Λe ½@t ζ ðz, tÞψ ðjx zjÞdz
ν Xð
X
ζ ðz, tÞuðz, tÞkðz, tÞψ ðjx zjÞdσ z (123)
r¼1 S r @Λr
Ð
þ @Λx ζ ðz, tÞuðz, tÞnðz, tÞψ ðjx zjÞdσ z
The integral on @Λx can be neglected for the same reasons given in Appendix “Fluid-
Phase Volume Average of Point Variable Spatial Derivatives.” Now, using the
definition of fluid-phase volume average given in Eq. 17, we have:
ð
½@t ζ ðz, tÞψ ðjx zjÞdz ¼ eðx, tÞh@t ζ ie ðx, tÞ (124)
Λe
Obtaining Eq. 26 is then immediate. Note that if Λx is time independent, u(z, t) = 0

on @Λx, and therefore the last integral on the right-hand size of Eq. 123 rigorously
vanishes.
40 L. Mazzei
Particle-Phase Volume Average of Point Variable Time Derivatives
We intend to derive an expression for the particle-phase volume average of point

variable time derivatives; in this case, we consider the derivative:
h i
@t nr ðx, tÞhζ irp ðx, tÞ (125)
Now, employing the definition of particle-phase volume average given in Eq. 21, we
can express the partial derivative above as:
X X
@t ½ζ r ðtÞψ ðjx zr ðtÞjÞ ¼ ζ_r ðtÞψ ðjx zr ðtÞjÞ
Sr Sr
X
þ ½ζ r ðtÞ@t ψ ðjx zr ðtÞjÞ (126)
Sr
From the definition of particle-phase volume average, it is:

X
ζ_r ðtÞψ ðjx zr ðtÞjÞ ¼ nr ðx, tÞ ζ_ p ðx, tÞ
r
(127)
Sr
Applying the derivation chain rule yields:

X X
½ζ r ðtÞ@t ψ ðjx zr ðtÞjÞ ¼ ζ r ðtÞvr, a ðtÞ@xa ψ ðjx zr ðtÞjÞ
Sr Sr h i
X (128)
¼ @xa ζ r ðtÞvr, a ðtÞψ ðjx zr ðtÞjÞ ¼ @xa nr ðx, tÞhζva irp ðx, tÞ
Sr
having used again the partial derivatives commutative property and the definition of
particle-phase average. Replacing these last two results in Eq. 126 yields Eq. 39.
References
M.P. Allen, D.J. Tildesley, Computer Simulations of Liquids (Oxford Science Publications,
New York, 1990)
T.B. Anderson, R. Jackson, A fluid mechanical description of fluidized beds. Equations of motion.
Ind. Eng. Chem. Fundam. 6, 527–539 (1967)
H. Arastoopour, C.H. Wang, S.A. Weil, Particle-particle interaction force in a dilute gas-solid
system. Chem. Eng. Sci. 37, 1379–1386 (1982)
R. Balescu, Equilibrium and Nonequilibrium Statistical Mechanics (Wiley, New York, 1975)
A.B. Basset, Treatise on Hydrodynamics (Deighton Bell, London, 1888)
R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena (Wiley, New York, 1960)
N.V. Brilliantov, T. Poschel, Kinetic Theory of Granular Gases (Oxford University Press, Oxford,
2004)
Y.A. Buyevich, Statistical hydrodynamics of disperse systems. Part 1. Physical background and
general equations. J. Fluid Mech. 49, 489–507 (1971)
S. Chapman, T.G. Cowling, The Mathematical Theory of Non-Uniform Gases (Cambridge Univer-
sity Press, Cambridge, 1970)
P.A. Cundall, O.D. Strack, A discrete numerical model for granular assemblies. Géotechnique 29,
47–65 (1979)
N.G. Deen, E.A.J.F. Peters, J.T. Padding, J.A.M. Kuipers, Review of direct numerical simulation of
fluid-particle mass, momentum and heat transfer in dense gas-solid flows. Chem. Eng. Sci. 116,
710–724 (2014)
R. Di Felice, The voidage function for fluid-particle interaction systems. Int. J. Multiphase Flow 20,
153–159 (1994)
A. Di Renzo, F. Cello, F.P. Di Maio, Simulation of the layer inversion phenomenon in binary liquid-
fluidized beds by DEM-CFD with a drag law for polydisperse systems. Chem. Eng. Sci. 66,
2945–2958 (2011)
D.A. Drew, Averaged field equations for two-phase media. Stud. Appl. Math. 50, 133–166 (1971)
D.A. Drew, Mathematical modelling of two-phase flow. Annu. Rev. Fluid Mech. 15, 261–291 (1983)
D.A. Drew, R.T. Lahey, Analytical modelling of multiphase flow, in Particulate Two-Phase Flow
(Butterworth-Heinemann, Boston, 1993)
D.A. Drew, S.L. Passman, Theory of Multicomponent Fluids, Applied Mathematical Sciences
(Springer, New York, 1998)
D.A. Drew, L.A. Segel, Averaged equations for two-phase flows. Stud. Appl. Math. 50, 205–231 (1971)
H. Enwald, E. Peirano, A.E. Almstedt, Eulerian two-phase flow theory applied to fluidization. Int.
S. Ergun, A.A. Orning, Fluid flow through randomly packed columns and fluidized beds. Ind. Eng.
Chem. 41, 1179–1184 (1949)
L.S. Fan, C. Zhu, Principles of Gas-Solid Flows (Cambridge University Press, Cambridge, 1998)
D. Gera, M. Syamlal, T.J. O’Brien, Hydrodynamics of particle segregation in fluidized beds. Int.
D. Gidaspow, Multiphase Flow and Fluidization (Academic Press, London, 1994)
D. Gidaspow, B. Ettehadieh, Fluidization in two-dimensional beds with a jet. Part II: Hydrodynamic
modeling. Ind. Eng. Chem. Fundam. 22, 193–201 (1983)
D. Gidaspow, M. Syamlal, Y.C. Seo, Hydrodynamics of fluidization of single and binary particles:
Supercomputer modeling, in Proceedings of the 5th International Conference on Fluidization,
Elsinore, 1985
D. Gidaspow, M. Syamlal, Y.C. Seo, Hydrodynamics of fluidization: Supercomputer generated
vs. experimental bubbles. J. Powder Bulk Solids Technol. 10, 19–23 (1986)
D. Gidaspow, R. Bezburuah, J. Ding, Hydrodynamics of circulating fluidized beds. Kinetic theory
approach. In Proceedings of the 7th International Conference on Fluidization, Brisbane, 1992
Y. He, T. Wang, N. Deen, A.M. van Sint, J.A.M. Kuipers, D. Wen, Discrete particle modeling of
granular temperature distribution in a bubbling fluidized bed. Particuology 10, 428–437 (2012)
E.J. Hinch, An averaged equation approach to particle interactions in a fluid suspension. J. Fluid
Mech. 83, 695–720 (1977)
B.P.B. Hoomans, J.A.M. Kuipers, W.J. Briels, W.P.M. van Swaaij, Discrete particle simulation of a
two-dimensional gas-fluidized bed: A hard sphere approach. Chem. Eng. Sci. 51, 99–118 (1996)
H. Iddir, H. Arastoopour, Modeling of multitype particle flow using the kinetic theory approach.
AICHE J. 51, 1620–1632 (2005)
R. Jackson, Locally averaged equations of motion for a mixture of identical spherical particles and a
Newtonian fluid. Chem. Eng. Sci. 52, 2457–2469 (1997)
R. Jackson, Erratum. Chem. Eng. Sci. 53, 1955 (1998)
R. Jackson, The Dynamics of Fluidized Particles, Cambridge Monographs on Mechanics (Cam-
bridge University Press, Cambridge, 2000)
J.T. Jenkins, S.B. Savage, A theory for the rapid flow of identical, smooth, nearly elastic, spherical
particles. J. Fluid Mech. 130, 187–202 (1983)
K.D. Kafui, C. Thornton, M.J. Adams, Discrete particle-continuum fluid modelling of gas-solid
fluidised beds. Chem. Eng. Sci. 57, 2395–2410 (2002)
42 L. Mazzei
J.L. Lebowitz, Exact solution of generalised Percus-Yevick equation for a mixture of hard spheres.
Phys. Rev. 133, 895–899 (1964)
M. Leva, Fluidization (McGraw-Hill, New York, 1959)
H. Lu, S. Wang, Y. Zhao, L. Yang, D. Gidaspow, J. Ding, Prediction of particle motion in a
two-dimensional bubbling fluidized bed using hard-sphere model. Chem. Eng. Sci. 60,
3217–3231 (2005)
C.K.K. Lun, S.B. Savage, D.J. Jeffrey, N. Chepurniy, Kinetic theories for granular flow: Inelastic
particles in Couette flow and slightly inelastic particles in a general flow field. J. Fluid Mech.
140, 223–256 (1984)
D.L. Marchisio, R.O. Fox, Computational Models for Polydisperse Particulate and Multiphase
Systems (Cambridge University Press, Cambridge, 2013)
M.R. Maxey, J.J. Riley, Equation of motion for a small rigid sphere in a nonuniform flow. Phys.
Fluids 26, 883–889 (1983)
L. Mazzei, P. Lettieri, A drag force closure for uniformly-dispersed fluidized suspensions. Chem.
Eng. Sci. 62, 6129–6142 (2007)
K. Nakamura, C.E. Capes. Vertical pneumatic conveying of binary particle mixtures, in Fluidization
Technology, ed. by D.L. Keairns (Hemisphere Publishing Corporation, Washington, DC, 1976),
pp. 159–184
R.I. Nigmatulin, Spatial averaging in the mechanics of heterogeneous and dispersed systems. Int.
J. Ouyang, J. Li, Particle-motion-resolved discrete model for simulating gas-solid fluidization.
Chem. Eng. Sci. 54, 2077–2083 (1999)
O. Owoyemi, L. Mazzei, P. Lettieri, CFD modeling of binary-fluidized suspensions and investiga-
tion of role of particle-particle drag on mixing and segregation. AICHE J. 53, 1924–1940 (2007)
T.W. Pan, D.D. Joseph, R. Bai, R. Glowinski, V. Sarin, Fluidization of 1204 spheres: Simulation
and experiments. J. Fluid Mech. 451, 169–191 (2002)
J.K. Pandit, X.S. Wang, M.J. Rhodes, Study of Geldart’s Group A behaviour using the discrete
element method simulation. Powder Technol. 160, 7–14 (2005)
J.W. Pritchett, T.R. Blake, S.K. Garg, A numerical model of gas fluidized beds. AIChe. Symp. Ser.
176, 134–148 (1978)
D. Ramkrishna, Population Balances (Academic Press, London, 2000)
J.F. Richardson, W.N. Zaki, Sedimentation and fluidization: Part I. Trans. Inst. Chem. Eng. 32,
35–53 (1954)
K. Sankaranarayanan, X. Shan, I.G. Kevrekidis, S. Sundaresan, Analysis of drag and virtual mass
forces in bubbly suspensions using an implicit formulation of the lattice Boltzmann method.
J. Fluid Mech. 452, 61–96 (2002)
D.G. Schaeffer, Instability in the evolution equations describing incompressible granular flow.
J. Diff. Eq. 66, 19–50 (1987)
L. Schiller, Z. Naumann, A drag coefficient correlation. Z. Ver. Deutsch. Ing. 77, 318–320 (1935)
S.L. Soo, Fluid Dynamics of Multiphase Systems (Blaisdell Publishing Company, Waltham, 1967)
M. Syamlal, The Particle-Particle Drag Term in a Multiparticle Model of Fluidization (National
Technical Information Service, DOE/MC/21353–2373, NTIS/DE87006500, Springfield, 1987)
M. Syamlal, W.A. Rogers, T.J. O’Brien, MFIX Documentation and Theory Guide (DOE/METC94/
1004, NTIS/DE94000087, 1993). Electronically available from: http://www.mfix.org
Y. Tsuji, T. Kawaguchi, T. Tanaka, Discrete particle simulation of two-dimensional fluidized bed.
Powder Technol. 77, 79–87 (1993)
J. Wang, M.A. van der Hoef, J.A.M. Kuipers, Particle granular temperature of Geldart A, A/B and B
particles in dense gas-fluidized beds. Chem. Eng. Sci. 97, 264–271 (2013)
C.Y. Wen, Y.H. Yu, Mechanics of fluidization. Chem. Eng. Prog. Symp. Ser. 62, 100–111 (1966)
S. Whitaker, Advances in the theory of fluid motion in porous media. Ind. Eng. Chem. 61, 14–28 (1969)
B.H. Xu, A.B. Yu, Numerical simulation of the gas-solid flow in a fluidized bed by combining
discrete and particle method with computational fluid dynamics. Chem. Eng. Sci. 52,
2785–2809 (1997)
M. Ye, M.A. van der Hoef, J.A.M. Kuipers, The effects of particle and gas properties on the
fluidization of Geldart A particles. Chem. Eng. Sci. 60, 4567–4580 (2005)
D.Z. Zhang, A. Prosperetti, Averaged equations for inviscid disperse two-phase flow. J. Fluid Mech.
267, 185–219 (1994)
H.P. Zhu, Z.Y. Zhou, R.Y. Yang, A.B. Yu, Discrete particle simulation of particulate systems: A
review of major applications and findings. Chem. Eng. Sci. 63, 5728–5770 (2008)
N. Zuber, On dispersed two-phase flow in the laminar flow regime. Chem. Eng. Sci. 19, 897–903 (1964)
Numerical Modelling of Pulverised Coal
Combustion
Zhao F. Tian, Peter J. Witt, Mark P. Schwarz, and William Yang
Abstract
Many thermal power generation plants rely on combustion of pulverised coal
carried out in large furnaces. Design and improvement of these furnaces can be
effectively assisted by using numerical modelling with Computational Fluid
Dynamics (CFD) techniques to develop a detailed picture of the conditions within
the furnace, and the effect of operating conditions, coal type, and furnace design
on those conditions. The equations governing CFD models of pulverised coal
combustion are described, with a focus on sub-models needed for
devolatilisation, combustion and heat transfer. The use of the models is discussed
with reference to examples of CFD modelling of brown coal fired furnaces in the
Latrobe Valley in Australia and black coal fired furnaces described in the litera-
ture. Extensions to the CFD models that are required to tackle specific industrial
and environmental issues are also described. These issues include control of NOx
and SOx emissions and the effect of slagging and fouling on furnace and boiler
operation.
Keywords
CFD • Coal combustion • Tangentially fired • Drying model • Devolatilisation
model • Char Combustion model • NOx • Soot model • Turbulence model
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Gas Phase Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Gas Phase Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Z.F. Tian (*)

School of Mechanical Engineering, The University of Adelaide, Adelaide, SA, Australia
e-mail: zhao.tian@adelaide.edu.au
P.J. Witt • M.P. Schwarz • W. Yang
CSIRO, Mineral Resources, Clayton, VIC, Australia

DOI 10.1007/978-981-4585-86-6_9-1
2 Z.F. Tian et al.
Turbulence Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Gas Phase Combustion: Gas Reaction Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Models for Particle Phase Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Eulerian-Eulerian Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Eulerian-Lagrangian Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Radiation Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Emissions Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Introduction
Coal-fired electricity generation is still dominant in the power industries of many

countries, despite the rapid increase of renewable electricity generation in recent
years. In 2010, coal-fired electricity generation accounted for 40% of the electricity
generation worldwide (EIA 2013). In Australia, coal including black coal and brown
coal generated about 64% of electricity in 2012–2013 (BREE 2014). It is particularly
noteworthy that in the state of Victoria, brown coal from the Latrobe Valley region
produces over 85% of the state’s electricity supply (Allardice 2000). Pulverised coal
(PC) combustion is one of the major technologies for the conversion of chemical
energy in coal into electricity. In the brown coal fired power plants in the Latrobe
Valley, all existing boilers use PC combustion technologies.
With advances in computing power and modelling techniques, computational
fluid dynamics (CFD) has evolved into a feasible tool for scientists and engineers
who can apply it to better understand PC combustion in furnaces (Tian et al. 2009)
and therefore optimise the design and operation of PC boilers. Nevertheless, PC
combustion in furnaces is one of the most difficult processes to model mathemati-
cally, since it generally involves the simultaneous coupled processes of three-
dimensional gas-particle fluid dynamics, turbulent mixing, heat transfer, and com-
plex homogeneous and heterogeneous chemical reaction kinetics (Viskanta and
Mengüç 1987). In modelling real industrial installations, additional complications
arise from slagging and fouling of heat transfer surfaces, variability in feed charac-
teristics and inevitable uncertainties in actual structural geometry due, for example,
to occasional damage or maintenance issues.
Figure 1 shows the major physical and chemical processes that occur during the
burning of pulverised coal particles in a tangentially-fired PC furnace. Fine coal
particles, pulverised in mills, are blown into the furnace through a number of
burners. Once the particles enter the furnace, they are heated by hot furnace gases
and radiation from the flame, they start to dry when their temperature reaches about
100–110 C (Wu 2005). When the particles are heated further to a certain critical
temperature (depending on the coal type and size), devolatilisation starts and vola-
tiles are released from the particles. The products of devolatilisation include
non-condensable volatiles (light gases), condensable volatiles (tars) and remaining
solid particles that normally comprise char and mineral matter. The volatiles react
with oxygen from the combustion air and other oxidants in the furnace. Finally char
Numerical Modelling of Pulverised Coal Combustion 3
Fig. 1 A schematic drawing Ash, flue gas, and

of PC coal fired furnace and emissions (including
some typical CFD sub-models NOx, SOx, etc. )
required to model these Convective parts
processes. Submodels:
(Superheaters,
(1) particle phase model,
Reheaters,
(2) evaporation/drying model,
(3) devolatilisation model, Economizer)
(4) char combustion model,
(5) turbulence model,
(6) turbulence-reaction
interaction model,
(7) radiation model, (8) soot
model, (9) NOx model, Furnace gas PC particles
(1,2,3,4)
(10) SOx model, (11) slagging
model Turbulent
Slag on flame
furnace zone (5-10) Raw Coal
wall(11)
Mill
Burners
Combustion air
particles react with gases in the furnace, leaving mineral matter and probably a small
fraction of unburnt char in the solid particles. These particles (ash) and the furnace
gas flow through convective heat transfer sections such as superheaters, reheaters
and economisers, exchanging heat with the working fluid (water/steam) in the
convective devices. Typical exhaust gases comprise CO2, N2, H2O, O2, and small
amounts of NOx, SOx, CO and particulate matters (PM). After leaving the convec-
tive passes exhaust gases will normally go through various air pollution control
equipment before being discharged through the stack.
In the CFD approach, mathematical descriptions of each of these processes in the
furnaces are called “sub-models” because they can be developed and updated in the
same way that modules in circuit boards can be replaced (Niksa 1996). As shown in
Fig. 1, a CFD code for modelling coal combustion probably needs the following
sub-models: (1) model for particle phase motion, (2) evaporation/drying model,
(3) devolatilisation model, (4) char combustion model, (5) turbulence model, (6) tur-
bulence-reaction interaction model (gas phase reaction models), (7) radiation model,
(8) soot model, (9) NOx model, (10) SOx model, (11) slagging model. Figure 2,
adapted from Tian et al. (2010a) shows some sub-models available in the commer-
cial CFD code ANSYS/CFX 14.
This book chapter briefly reviews and describes the mathematical equations of
some of these sub-models. The authors have implemented some of these sub-models
into a CFD model of a tangentially-fired PC furnace at the TRUenergy Yallourn
power plant, Latrobe Valley, Australia (Tian et al. 2010a). This CFD model was
developed based on the commercial CFD code, ANSYS/CFX. The model has been
4 Z.F. Tian et al.
CFX Coal combustion solver
Turbulence Particle phase Gas combustion Radiation Emissions

models models models models models
Standard Lagrangian EDM P-1 model Soot

k-ε model model model
RSM FRC DT model NOx

model
Drying model
k-ω model Monte Carlo
Combined model
Devolatilisatio model
n model
SST model
Char oxidation
model
Fig. 2 The structure of CFD coal combustion solver (Adapted from Tian et al. 2010a). RSM:
Reynolds Stress model; SST: Shear stress transport model; EDM: Eddy dissipation model; FRC:
Finite rate chemistry model; Combined model: Combined EDM/FRC model; DT: Discrete transfer
model
validated against plant measurements and applied to investigate the effects of several
operating conditions at full load, such as different out-of-service firing groups and
different combustion air distributions on the coal flames (Tian et al. 2010b). The
CFD furnace model was then used to assess the combustion of pre-dried brown coal
in the furnace that was designed for raw or non-pre-dried brown coal (Tian et al.
2012). In this book chapter, additional results of the CFD furnace model are reported
as examples of the sub-model applications.
Gas Phase Model
Gas Phase Governing Equations
In CFD models of pulverized coal combustion, gases in the furnace are normally
considered to be a mixture consisting of the gaseous components, O2, H2O, CO2,
CO, N2, NO, and volatiles (Tian et al. 2010a). Volatiles can be taken as one single
gas component, a mixture of light gas and tar, or a mixture of individual gases such
as CH4, C2H2, etc. To reduce the computing time, these components are normally
assumed to be mixed at the molecular level, hence having the same mean velocity,
pressure, temperature and turbulence fields (Tian et al. 2010a). The Navier-Stokes
equations are used to solve the continuity and momentum equations of the gas
mixture. The gas phase equations solved in ANSYS/CFX are given below as an
example of the steady state governing equations for CFD models. Equations 1 and 2
are the continuity equation and momentum equation of the mean steady state after
Reynolds averaging:

∇ ρg U g ¼ 0 (1)
n h T i o
∇ ρg U g U g ¼ ∇pg þ ∇ μg ∇U g þ ∇Ug ρg U g U g þ SM , (2)
here Ug is the gas phase mean velocity vector; Pg is the gas phase mean pressure;
SM is the external momentum source such as gravity and forces from the coal particle
phase; and ρg is the gas mixture density defined as:
X
Nc
ρg ¼ Y I ρI , (3)
I¼1
where ρI is the mass density of the component I. Nc is the number of modelled

species in the gas mixture, and YI is the mass fraction of the species I, solved by the
following equation:

∇ ρg U g Y I ¼ ∇ ðΓI:eff ∇Y I Þ þ SI (4)
In this equation, SI is the source term of the species related to generation or

destruction of the species by reaction. Other properties of the gas mixture, such as
the gas mixture molecular viscosity μg and the gas mixture specific heat capacity at
constant pressure Cp,g, are calculated in the same manner as Eq. (3):
X
Nc
αg ¼ Y I αI , (5)
I¼1
where αg in the gas mixture property being considered.

The effective diffusion coefficient of species I, ΓI.eff, in Eq. (4) is defined as:
μt
ΓI:eff ¼ ΓI þ , (6)
Sct
Where ΓI is the molecular diffusion coefficient of species, ΓI = ρIDI, here DI is
the kinematic diffusivity of the species I. Sct is the turbulent Schmidt number and μt
is turbulent viscosity.
The source term SI in Eq. (4) is due to chemical reaction involving the species.
There is one transport equation of each gas component except the constraint gas N2.
The mass fraction of N2 is calculated by using the following equation:
X
Nc
YI ¼ 1 (7)
I¼1
6 Z.F. Tian et al.
The Reynolds averaged energy equation for the gas mixture can be:

μt
∇ ρg U g hg ¼ ∇ λg ∇T g þ ∇hg þ SE , (8)
Prt
where hg is the gas mixture enthalpy. Prt is the turbulent Prandtl number. The
energy source term SE includes thermal energy from chemical reactions and thermal
radiative heat transfer.
In most CFD coal combustion models, the Reynolds averaged Navier Stokes
(RANS) modelling approach is used to handle turbulence. In RANS models, the
Reynolds stresses terms ρg Ug Ug in the momentum Eq. 2 are modeled based on the
Boussinesq hypothesis:
h T i 2
ρg U g Ug ¼ μt ∇U g þ ∇U g δij ρg kg þ μt ∇ Ug , (9)
3
where μt is the turbulence viscosity that can be calculated by applying turbulence
models that will be discussed later. Δij is the Kronecker delta that is 1 when i = j and
0 when i 6¼ j.
Turbulence Models
Turbulent mixing is one of the major factors controlling the local proportions of fuel
and oxygen throughout the primary flame zones and thereby exerting a predominant
effect on heat release rates, heat fluxes onto steam tubes, carbon burnout times and
pollutant formation rates (Niksa 1996). In CFD techniques, turbulence models are
used to close the equations for the fluctuating quantities and thereby include the
effects of eddies on the time averaged flow.
The Reynolds stresses in Eq. (9) can be also directly calculated from six transport
equations and this is called the Reynolds Stress model (RSM). The RSM is a second
order RANS model but has been used to predict coal combustion by only a few
researchers such as Weber et al. (1995), Zhang and Nieh (1997). The RSM has been
shown to perform better than k-e models in predicting of isothermal swirl jets
(German and Mahmud 2005; Weber et al. 1990) due to two equation models not
being able to reliably resolve flows with strong streamline curvatures. This limitation
of two equations models can be partly overcome by a curvature correction term to
the turbulence production term. However work at International Flame Research
Foundation (IFRF) (Weber et al. 1995) demonstrated that this advantage of the
RSM over k-e models was not found in the burning jet applications (Niksa 1996).
Backreedy et al. (2006) found that the performance of RSM in modelling a swirl coal
flame in a pilot-scale furnace is not better than that of k-e models. This is confirmed
by a recent modelling project that compared the performance of several RANS
models in modelling swirling flow in a vortex flow reactor (Tian et al. 2015). The
BSL RSM can predict the anisotropic Reynolds stresses that the SST model and the
standard k-e model cannot predict, but this does not make it more accurate in
predicting the mean flow-field than the other two models. Application of the RSM
requires more computational resources than is required by the standard k-e and
similar first order models, partially due to the need to solve additional transport
equations of the Reynolds Stresses and probably due to the poor convergence
characteristics and stability of RSM.
The standard k-e model is commonly applied in studies of coal combustion in
furnaces. For the standard k-e model the turbulent viscosity, μt, in Eq. (8) is
computed from:
μt ¼ Cμ ρg k2g =eg (10)
where kg is the turbulence kinetic energy of the gas mixture and eg is the kinetic
energy dissipation rate of the gas mixture. In the standard k-e model, the turbulence
kinetic energy and the kinetic energy dissipation rate of the gas mixture are calcu-
lated by solving two transport equations.
The turbulence kinetic energy equation for the standard k-e model is:

μ
∇ ρg U g kg ¼ ∇ μg þ t ∇kg þ Pk ρg eg (11)
σk
Here, the rate of production of turbulence kinetic energy Pk is modeled by:

T 2
Pk ¼ μt ∇U g ∇Ug þ ∇U g ∇ U g μt ∇ U g þ ρg kg (12)
3
The kinetic energy dissipation rate equation is:

μ eg
∇ ρg U g eg ¼ ∇ μg þ t ∇eg þ Ce1 Pk Ce2 ρg eg (13)
σe kg
The values of the constants are Cμ = 0.09, σ k = 1.0, σ e = 1.3, C1e = 1.44,
C2e = 1.92 (Launder and Spalding 1974).
Compared to the RSM, the standard k-e model is less computationally intensive
while providing a similar level of predictive accuracy to the RSM for most coal
combustion applications not involving strong swirling flow fields. The standard k-e
model has been used to simulate coal flames in pilot-scale furnaces (Tian et al.
2010a), e.g., Lockwood and Salooja (1983), Truelove and Holcombe (1991), Zhou
et al. (2002) and many others, and full scale furnaces by workers such as Belosevic
et al. (2006), Xu et al. (2001).
One of the major shortcomings of the standard k-e model is that it cannot predict
the adverse pressure gradient properly; the standard k-e model significantly over-
predicts shear stress levels and thereby delays separation (Menter 1992). Another
shortcoming relates to the numerical stiffness of the equations when integrated
through the viscous sublayer (Menter 1992). Many modified k-e models have been
8 Z.F. Tian et al.
derived from the standard k-e model in order to overcome these shortcomings, one of
which is the Re-Normalization Group (RNG) k-e model that has been used in coal
combustion modelling. The transport equations for gas phase kg and eg in the RNG
k-e model are given as follows:

μt
∇ ρg Ug kg ¼ ∇ μg þ ∇kg þ Pk ρg eg (14)
σ k, RNG

μt eg
∇ ρg U g e g ¼ ∇ μ g þ ∇eg þ Ce1, RNG Pk Ce2, RNG ρg eg (15)
σ e, RNG kg
Ce1,RNG is calculated as:
Ce1, RNG ¼ 1:42 f η , (16)
where
η
η 1
fη ¼ 4:38 (17)
ð1 þ βRNG η3 Þ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pk
η¼ (18)
ρg Cμ, RNG eg
and where βRNG is 0.012.

The values of the other constants are Cμ,RNG = 0.0845, σ k,RNG = 0.7179,
σ e,RNG = 0.7179, Ce2,RNG = 1.68 (Versteeg and Malalasekera 2007).
The Re-Normalisation Group (RNG) k-e model has been used in some coal
combustion modelling projects. Fan et al. (2001) compared modeling results from
the RNG k-e model and the standard k-e model in a tangentially fired furnace against
experimental data, and found the RNG k-e model gave better results for swirling flow
and sharper flow gradients within calculated regions than the standard k-e model.
Backreedy et al. (2006) used the RNG k-e model in a coal test furnace model, as the
RNG k-e model is believed to have advantages over the standard k-e model in
swirling flows. Nevertheless, these advantages of the RNG k-e model over the
standard k-e model in swirling flows have been a matter of some controversy (Saqr
2011).
The Wilcox k-ω model (Wilcox 1988), Menter k-ω model that also called
Baseline (BSL) model, and Shear-stress transport (SST) model (Menter 1994) are
another class of two-equation RANS models. The SST model is a hybrid approach
between the standard k-e model and the k-ω model. In the SST model, in the region
near walls, the k-ω model is used as it performs well for near wall flows and can
avoid the use of wall functions also it allows the accurate specification of ω values on
the wall surface hence avoiding issues of defining e near wall for fine grids. In the
region far away from wall, the standard k-e model is used as it is robust in the free
stream while k-ω model is sensitive to the free stream value of ω (Versteeg and
Malalasekera 2007).
The transport equations of kg and ωg in the SST model are shown below, with
ωg = eg/kg. The transport equations of kg and ωg of the SST model are:

μ
∇ ρg U g kg ¼ ∇ μg þ t ∇kg þ Pk β0 ρg kg ωg (19)
σ k3

μt ωg
∇ ρg Ug ωg ¼ ∇ μg þ ∇ωg þ α3 Pk
σ ω3 kg
1
þ ð1 F1 Þ2ρg ∇ωg ∇kg β3 ρg ω2g (20)
σ ω , 2 ωg
The turbulent viscosity, μt, is calculated as:
ρ kg α1
μt ¼ g , (21)
max α1 ωg , SF2
pffiffiffiffiffiffiffiffiffiffiffiffi
where S ¼ 2Sij Sij .
The blending function F1 in Eq.
20is p
calculated
ffiffiffiffi as:

4 kg 500μg 4ρg kg
F1 ¼ tanh arg1 , arg1 ¼ min max 0:09ωg y , ρ y2 ωg , σ ω, 2 Dþ 2
g ωy

1 1
Dþ
ω ¼ max 2ρg ∇kg ∇ωg , 1010 (22)
σ ω, 2 ωg
where y is the distane to the nearest wall. The blending function F2 in Eq. 21 is
given as:
pffiffiffiffiffi !
2 kg 500μg
F2 ¼ tanh arg22 , arg2 ¼ max , : (23)
0:09ωg y ρg y2 ωg
The values of the constants employed in the SST model are β0=0.09, α1 = 5/9,
α3 = 0.44, β3 = 0.0828, σ ω,2 = 1/0.856 (Versteeg and Malalasekera 2007).
Only a few studies of coal combustion using the k-ω or the SST model can be
found in the literature. In a CFD modelling study (Tian et al. 2009), six first order
RANS models , namely, the standard k-e model, a modified k-e model, RNG k-e
model, Wilcox k-ω model, BSL k-ω model and SST models were used to simulate a
non-swirling coal flame in a pilot-scale furnace of IFRF. The standard k-e model,
RNG k-e model, BSL and SST models were found to be generally in good agreement
with the experimental data. Predictions using the SST model and BSL k-ω model
were almost identical, and results of the standard k-e model and the RNG k-e model
were similar (Tian et al. 2010b). The SST model and the standard k-e model were
further tested in modelling an isothermal gas-particle flow in three inclined
10 Z.F. Tian et al.
rectangular jets in crossflow (Tian et al. 2011). The flow configuration and flow
conditions were scaled based on typical flow conditions experienced in the Victorian
brown coal furnace burners (Tian et al. 2010a). Gas and particle flows predicted by
both models were found to be in reasonable agreement with the detailed experimen-
tal data, although the SST model showed a slightly better agreement with the
measurements than the standard k-e model (Tian et al. 2010a). The SST and the
standard k-e model were then employed in a CFD model of a 375 MW tangentially
fired furnace (Tian et al. 2010a) burning high-moisture brown coal. Both turbulence
models provide similar predictions that were in good agreement with the plant data
(Tian et al. 2010a).
The standard k-e model has been found to perform particularly well in confined
flows where Reynolds shear stresses are most important (Versteeg and Malalasekera
2007). In tangentially fired furnaces, the strong vortex at the center formed by the
impinging jets from corners or walls greatly increases the turbulence transport in the
furnaces (Basu et al. 1999), therefore the flows in the furnaces can be taken as
turbulence transport dominated flows. The major advantage of the SST model over
the standard k-e model is the inclusion of the shear rate magnitude, S, in Eq. 21,
which ensures the ratio of turbulence production to turbulence dissipation larger than
one in adverse pressure gradient flows (Menter 1992). However, this supposed
advantage does not appear to result in a clearly better prediction forthe tangentially
fired flames, though the separation of flows are indeed found in the furnace as shown
in Fig. 3. Figure 3a-c show the predicted flow vectors on a horizontal plane at the exit
of the upper main burner when firing units 3&6, 5&6 and 2&6 are out of service,
respectively. Flow separations can be found in Fig. 3 between jets. For example, as
shown in Fig. 3a, primary gas flows in the furnace as jets and these high speed jets
entrain furnace gas and secondary air. This entrainment helps to form recirculations
and flow separations. The details of the CFD model and boundary conditions for the
cases shown in Fig. 3 can be found in previous papers (Tian et al. 2010a; Tian et al.
2010b).
Gas Phase Combustion: Gas Reaction Kinetics
The gas phase reactions in coal fired furnaces are very complex. Not all the species
and reaction chemistry can be included in the CFD models, partially due to the large
computing time required to transport all the species and the stiffness of the large
number of intermediate reactions. This problem is further complicated by the
heterogeneous nature of coal and the devolatilisation process making knowing the
detailed chemical composition of volatiles and subsequent reactions extraordinarily
difficult. As discussed in Yeoh and Yuen (2009), CFD techniques for partial differ-
ential equations require computing time roughly proportional to Ns2 (Ns is the
number of species). If all the reaction species found in the PC furnace are included
in the CFD model, the computing time will be excessive. Furthermore, except for
some simpler alkane hydrocarbon fuels such as CH4 and C2H2, comprehensive
reaction chemistry for complex fuel is still not well determined (Yeoh and Yuen
Fig. 3 Predicted flow patterns for cases (a) Firing units (FU) 3&6 out-of-service (b) Firing units
(FU) 5&6 out-of-service, (c) Firing units (FU) 6&7 out-of-service
2009). Therefore global reaction schemes are normally used in CFD modelling of the
gas phase volatile combustion of coal.
As noted earlier and described in detail below coal combustion consists of a
number of stages with a critical stage being the devolatilisation of the solid particle to
produce gas phase volatiles:
Coal ! Volatiles þ CðcharÞ (24)
Volatile combustion can be modeled by the global single step reaction:
VolatilesðHCÞ þ O2 ! CO2 þ H2 O (25)
Tian et al. (2010b) notes that the concentration of CO cannot be calculated by the
above single step reaction. If understanding the CO concentration is important for
12 Z.F. Tian et al.
a b
CO.Mass Fraction
CO concentration
0.030
0.025
0.020
0.015
0.010
0.005
0.000
Fig. 4 Predicted CO concentration for cases (a) Firing units 2&6 out-of-service (b) Firing units
5&6 out-of-service. CFD model details and boundary conditions can be found in Tian et al.
(2010a, b)
the coal combustion modeling work being undertaken, the following reaction
scheme was proposed in Tian et al. (2010b):
Coal ! Volatiles þ CðcharÞ (26)
VolatilesðHCÞ þ O2 ! CO þ H2 O (27)
1
CðcharÞ þ O2 ! CO (28)
2
1
CO þ O2 ¼ CO2 (29)
2
Figure 4 shows the predicted CO concentration in the furnace of Tian et al.
(2010b) based on the combustion scheme shown above. More species and reactions
can be added in the CFD model, however, as mentioned above, the computing time
will increase and details of the chemistry are required. When predition of NOx and/or
SOx emissions is required, the NOx species and SOx species can be added to the gas
mixture. NOx and SOx models are briefly reviewed later.
Several approaches can be used to calculate behaviours of gas species specified
by the chemistry. The most straight forward one is a species transport approach. In
this approach, transport equations for each species (or each species except a
constraint species) are solved. To close the transport equations (such as Eq. (4)), the
source term, SI, needs to be calculated. Usually the gas phase reactions in coal fired
boilers can be taken as a fast reaction system in respect to their modelling. A fast
reaction system means the chemical reaction rates are much faster than the mixing
processes in the system, in other words, reaction rates in the system are controlled by
the mixing process. Another characteristic of flames in coal-fired boilers is that they
generally can be classified as non-premixed combustion.
The source term SI can be computed as the sum of the reaction sources of
reactions involving species I,
XNKI
SI ¼ W I k¼1
v00kI v0kI RkI , (30)
where WI is the molar mass of species I. RkI is the reaction rate of species I in the
reaction k, which can be calculated by using either a finite rate approach or the eddy
dissipation model. v0kI is the stoichiometric coefficient of species I in the reaction k as
a reactant and v00kI is the stoichiometric coefficient of species I in the reaction k as a
product. NkI is the number of reactions that component I involves in.
For finite rate chemistry model, the reaction rate of reaction k, Rk, is calculated as:
0 00
rkI r kI
Rk ¼ Fk ∏Nc
I¼A, B, :::: ½I Bk ∏I¼A, B, :::: ½I
Nc
(31)
here [I] is the molar concentration of species I. Nc is the number of species in the
reaction k. The forward rate constant Fk can be calculated by the Arrhenius rate:

Ek
Fk ¼ Ak T βk exp (32)
RT
where Ak is the pre-exponential factor; T is temperature; βk is the temperature

exponent; Ek is the activation energy; and R is the universal gas constant, 8.314 J/
molK.
If applicable, the backward rate constant Bk can be calculated as:

βbk Ebk
Bk ¼ Ab T exp (33)
RT
The finite rate chemistry model is applicable to laminar flames as the effects of
turbulence on the reactions are not included. This approach can be used for com-
bustion with relatively slow chemistry and small turbulent fluctuations (Yeoh and
Yuen 2009).
In coal-fired flames, the eddy dissipation model (Magnussen and Hjertager 1977)
can be used to model the turbulence-chemistry interaction. In the eddy dissipation
model for pre-mixed flames, Rk is directly related to the time required to mix
reactants at the molecular level, i.e., a mixing time defined by the turbulent kinetic
energy of gas mixture, kg, and dissipation rate, eg:
14 Z.F. Tian et al.

eg ½I
RkI ¼ A min 0 (34)
kg vkIR
where v0kIR is the stoichiometric coefficient for reactant I in reaction k; and A is a

constant with a value of 4.
The advantage of the eddy dissipation model is that it is simple and takes accounts
effects of turbulence on chemistry. However, the eddy dissipation model, as shown
in Eq. (34), does not account for the effects of temperature on reaction rates and it
can only be used for one-step or two-step reactions without giving detailed chemistry
effects. When more detailed reaction kinetics are required in the CFD model, the
generalized eddy dissipation concept model can be used.
In the generalised eddy dissipation concept model, the mean reaction rate of
component I, RI, is assumed to occur in small turbulence structures over a mean
residence time τ(Magnussen and Hjertager 1981). The fine turbulence structures in
a computational cell are characterised a mean length fraction, ξ. The mean reaction
rate of I, RI, is calculated as:
W ρðξ Þ2
RI ¼ h I i YY YI (35)

τ 1 ðξ Þ 3
Here Y Y is the species mass faction in the fine structures after reacting over the
time scaleτ; Y Y can be determined through a laminar finite rate model (Yeoh and
Yuen 2009). The mean residence time τis calculated as:
1=2
vg
τ ¼ Cτ , (36)
eg
here Cτ is a constant with a default value of 0.4082.

The mean length fraction, ξ is calculated as:
!1=4
v g eg
ξ ¼ Cξ , (37)
kg 2
here Cξ is a constant with a default value of 2.1377.

More advanced turbulence-chemical interaction models such as the joint proba-
bility density function (PDF) and laminar flamelet models can be applied to coal
fired combustion. Their advantages over the EDC model and eddy dissipation
models have been found to be less pronounced in coal flames than the gas flames,
partially due to the fact that gaseous phase reactions are just a part of the reaction
sequence involved in coal combustion (Vascellari and Cau 2012): the heterogeneous
reaction of char is also important in the coal flame.
Models for Particle Phase Motion
In the most CFD studies of PC combustion in furnaces, two categories of approaches

are typically used for prediction of the coal particle phase motion: the Eulerian-
Eulerian model or the Eulerian-Lagrangian model.
Eulerian-Eulerian Model
The Eulerian-Eulerian model calculates the particle flow using Eulerian or fluid-like
equations, e.g. modified Navier-Stokes Eqs. (1) and (2). These equations can be
implemented efficiently in the existing solvers resulting in relatively less computa-
tional time being required to calculate mean parameters, such as velocity and volume
fraction for the particle flow. Some simplified Eulerian-Eulerian models, which
assumed a mechanical and thermal equilibrium between the two phases, have been
developed and used to model coal combustion, e.g., Benim et al. (2005), Fiveland
and Wessel (1988), Zhou et al. (2002) developed a two-fluid-trajectory model and
simulated coal combustion in a tangentially fired boiler. This two-fluid-trajectory
model uses Eulerian gas-phase equations, Eulerian particle-phase continuity and
momentum equations, two-phase turbulence models, and Lagrangian ordinary dif-
ferential equations of particle temperature and mass change to take into consider-
ation of the history effects (Zhou et al. 2002). The Eulerian-Eulerian coal
combustion model has been incorporated in the commercial CFD code PHEONICS.
Nevertheless, there are some inherent problems in the use of Eulerian models for
gas-particle flows as reviewed by Tian et al. (2005), propably making this approach
less attractive in modeling PC combustion.
Eulerian-Lagrangian Model
Most commercial and research CFD codes make use of the Eulerian-Lagrangian
approach to model pulverised coal particle combustion, for examples, ANSYS/CFX
and FLUENT. The Lagrangian model tracks the individual particle motion and
therefore overcomes some difficulties associated with the Eulerian model for the
particles (Tian et al. 2005).
The equations of particle motion are:
dxp
¼ up (38)
dt
dup
mp ¼ FD þ Fg þ Fother , (39)
dt
16 Z.F. Tian et al.
here FD is the drag force, Fg is the gravity force and Fother includes other forces
such as buoyancy force, virtual mass force, pressure gradient force, etc. The drag
force FD is calculated from (Tian et al. 2010a):
ug, instant up
FD ¼ m p , (40)
τr
here ug , instant is the instantaneous gas velocity. The discrete random walk (DRW)
model is widely used to model the effects of turbulence on particle trajectories. In the
DRW model, ug, instant = ug + u0 when particle dispersion is taken into consideration
(u0 is an approximation to the eddy fluctuation velocity determined using a random
walk approach); ug, instant = ug when the particle dispersion is off. The particle
relaxation time τr is given by:
ρp d2p
τr ¼ (41)
18μg f D
and the drag coefficient fD for a sphere can be calculated based on empirical
equations, e.g., (Tian et al. 2010a):
1 þ 0:15Re0:687 , Rep 1000

fD ¼ p (42)
0:01833Rep , Rep > 1000
24
CD ¼ f (43)
Rep D
One major concern of the Eulerian-Lagrangian approach is the expensive com-

puting time that may be experienced when tracking a substantial number of the
particles to obtain good statistical information of the particle phase (Tian et al. 2005).
With the progress in computer speeds, multi-core processors and parallelisation
techniques, the time expense of Eulerian-Largrangian models has been significantly
reduced and it has become a popular tool for coal combustion simulations. The
Eulerian-Lagrangian model is extended in coal combustion models to take into
account the particle combustion processes occurring in the furnace. The most widely
used coal drying models, devolatilisation models and char oxidation models that
have been implemented the Eulerian-Largrangian models are reviewed in the next
section.
Coal Devolatilisation and Char Oxidation Models

When pulverised coal particles enter the furnace through the burners, they rapidly
mix with hot intermediates and combustion products. The particles undergo the
following four well-defined steps during combustion in the furnace (Wu 2005)
shown in Fig. 5:
Volatile Combustion
1. 2.
Original
Coal Heating & 4.
Devolatilisation
Particle Drying
Process Char Combustion
Fig. 5 Coal particle combustion processes (After Wu 2005)
• Coal particle heating and drying;

• Devolatilisation of the coal particle to produce non-condensable volatiles (light
gases), condensable volatiles (tars), and a carbonaceous char;
• Gas phase volatile combustion;
• Char combustion.
To model these processes in the Lagrangian particle tracking model, the coal
particles are normally treated as spheres that do not interact with other particles. Each
particle is able to undergo internal reactions as well as being fully coupled through
the transfer of mass, momentum and energy with the gas phase, which enables heat
transfer and chemical reactions to occur between the particle and gas phase. Several
models are required to model the combustion of the coal particles shown in Fig. 5,
namely, a drying model for the raw coal particles, a devolatilisation model, and a
char combustion model.
Coal normally contains moisture that can be divided into surface moisture and
inherent (or bound) moisture. Surface moisture is moisture on the coal surface
including inter particle voids and contact points of particles; while inherent moisture
exists in the coal internal pore structure (Wu 2005). Old coal such as bituminous
coals has 1–12.2% moisture as received and subbituminous coals have moisture as
received in the range of 14.1–31% (Tillman 1991; Wu 2005). Young coals such as
brown coal and lignite normally contain higher moisture content, e.g., Victorian
brown coal typically has 66–70% moisture by weight (Tian et al. 2010b). In
pulverised coal furnaces, a fraction of water in coal particles is released from coal
during the pulverising process and also in the pre-drying process if there is any. The
content of the water in particles entering the furnace depends on the coal type and the
boiler type.
18 Z.F. Tian et al.
Sometimes the modelled drying process of coal particles in PC furnaces can be

incorporated in the devolatilisation process, or it can be modelled in a separate
drying model. When the water content of the particles is small, e.g., coal particles
after a pre-drying process, evaporation of water in the pre-dried particle can be
modelled as a species of volatile gas. This will slightly reduce the complexity of the
coal modelling process by eliminating the need for a separate drying model. How-
ever, it is more accurate to model the evaporation of water separately from the
devolatilisation process, because in real furnace combustion, most water in the
particles evaporates before the start of devolatilisation process.
Typical equations used to model the water evaporation in coal particles assuming
mass transfer control are given in Bhambare et al. (2010). The change of mass of coal
particles during the drying process can then be calculated as below:

dmp
¼ kc Cvapour, s Cvapour, 1 Ap Mvapour (44)
dt drying
here Ap is particle surface area. Mvapour is the molar mass of water vapour. kc is
mass transfer coefficient. Cvapour,s is vapour concentration at the coal particle surface:
psat ðTP Þ
Cvapour, s ¼ , (45)
RT p
here psat(TP) is the saturated vapour pressure at the particle temperature, Tp.
Cvapour,1 in Eq. (44) is the vapour concentration in the bulk gas:
pop
Cvapour, 1 ¼ ½H2 O , (46)
RT 1
here, [H2O] is the mole fraction of H2O vapour and pop is the operating pressure.
The mass transfer coefficient, kc, in Eq. (44) is calculated using the Nusselt
number:
kc d P 1=2
Nu ¼ ¼ 2:0 þ 0:6Red Sc1=3 (47)
Dvapour, m
where Dvapour,m is the diffusion coefficient of vapour, Sc is the Schmidt number.

In the CFD model, the reduction of particle mass can be accounted for either by
reducing the particle density without changing of the particle diameter or reducing
the particle diameter with a constant density.
Devolatilisation Models
After drying in the burner exit region, coal particles are heated to higher tempera-
tures rapidly, and start to decompose to produce non-condensible volatiles and tars.
Non-condensible volatiles are non-condensable gases that consist mainly of a
mixture of CO2, H2O vapour and combustible gases including CO, H2 and hydro-
carbons such as CH4, C2H4, C2H6, etc. (Field et al. 1967). The tar is a heavy
hydrocarbon-like substance that is condensable, with an atomic ratio of H/C > 1.0
(Tillman 1991; Wu 2005). Again, the exact products of the devolatilisation process
are determined by the coal types and decomposition condition that can be either
rapid or slow. Pulverised coal combustion always involves a high rate of heating
(104 K/s or greater) that is classified as rapid decomposition (Field et al. 1967).
Evolution of volatile matter under the influence of heat and the subsequent combus-
tion of the vapours evolved is an integral part of the combustion of coal, including
brown coal (Mulcahy et al. 1991). In fact, about 65% of the heat released by
combustion of Yallourn coal, one kind of Latrobe Valley coal, is derived from
combustion of the volatiles (Jones and Stacy 1986).
Two groups of devolatilisation models have been developed and used for PC
combuston models: simple global kinetic models and more comprehensive
computer-based network models. In CFD models, the products of devolatilisation
process are assumed to be the gas(es) and that remaining coal particles that comprise
char and ash only. Volatile gas(es) in the model can be a single species or several
major volatile components such as CH4, C2H4, C2H6, etc. Two simple global kinetic
models are widely used in PC coal combustion modelling. These are the single first-
order reaction (SFOR) model and the competing reaction model.
In SFOR model the devolatilisation of coal particles is assumed to be independent
of the particle size, porosity, specific surface area and surface/mass ratio, and other
coal characteristics (Tillman 1991). The rate of devolatilisation is assumed to be first-
order dependent on the amount of volatiles remaining in the particle:

dmp
¼ kv mp 1 f v, 0 mp, 0 (48)
dt devo
where mp is the instantaneous particle mass, mp,0 is the initial particle mass after
drying process if there is a separate drying model. fv,0 is the initial mass fraction of
volatiles in the particle before devolotiliation and kv is the kinetic rate:

Tv
kv ¼ Av exp (49)
Tp
The pre-exponential factor Av and the activation temperature Tv are constants

determined experimentally for the particular coal.
Some experiments have found that the yield of volatiles from PC particles can be
greater by as much as a factor of two than the proximate value in PC furnaces
(ANSYS/CFX 2015). The competing reaction model takes this into consideration by
assuming that two devolatilisation process undergo simultaneously, one reaction
dominants at low temperatures and the other at high temperatures (ANSYS/CFX
2015). Therefore, Eq. (48) can be written as:
20 Z.F. Tian et al.

dmp
¼ kv, 1 α1 þ kv, 2 α2 mp 1 f v, 0 mp, 0 , (50)
dt devo
where α1is near the proximate volatile fraction where α2 is higher, close to unity,
reflecting the characteristics of devolatilisation at high temperature (Wu 2005). kv , 1
and kv , 2 are the kinetic rate at low and high temperature, respectively.
A chemical percolation model for devolatilisation (CPD) model has been devel-
oped in Grant et al.(1989) by applying the lattice statistics. In contrast to the above
devolatilisation models based on empirical rate relations, the CPD model
characterises the devolatilisation behaviour of rapidly heated coal based on the
physical and chemical transformations of the coal structure (ANSYS/FLUENT
2015). The CPD has been implemented into several CFD codes such as ANSYS/
FLUENT and has been used for some coal combustion modelling, e.g., in Jovanovic
et al.(2012). In this chapter, the CPD model is not discussed in details due to the
space limitation. Interested readers can read more details in Grant et al.(1989), Wu
(2005) .
Devolatilisation of Latrobe Valley brown coal under fast heating rates (normally
larger than 104 K/s), which is experienced in pulverised brown coal combustion, has
been investigated in several studies using different methods such as a vertical laminar-
flow furnace (Roberts and Loveridge 1969), a plug-flow reactor (Duong 1985), and a
pressurised drop-tube furnace (Yeasmin et al. 1999), and corresponding kinetic
parameters for the SFOR model have been calculated based on the experimental
measurements. Duong (1987) conducted measurements of pulverised brown coal
combustion in a plug-flow reactor under different inlet conditions. It is found that
the fuel/air mass ratio is seen to be the only factor affecting both the rates and
mechanism of the volatile release. However, this cannot explain the large difference
between parameters developed in Roberts and Loveridge (1969), Yeasmin et al.
(1999), given that both experiments were carried out in an inert atmosphere of
nitrogen. The kinetics parameters from Yeasmin et al. (1999), Duong (1987) have
been tested in a drop tube furnace (Ouyang et al. 1998) and it was found the parameters
from run 3 and run 5 of Duong (1987) give better agreement than other parameters for
the measured particle mass loss along the axis line in the drop tube furnace.
It has been found that some bituminous coals swell considerably during heating.
A swelling coefficient can be used in CFD codes to take into account such swelling
effects during devolatilisation. The value of the swelling coefficient is determined by
coal types and combustion conditions. Experiments have shown that Latrobe Valley
coals do not undergo swelling but develop an internal bubble structure when
devolatilised in nitrogen (Sainsbury and Hawksley 1969). The non-swelling char-
acteristic is confirmed by the observations of several combustion experiments such
as in Street (1979). Therefore, the particle swelling can be neglected when simulat-
ing the Latrobe Valley coal combustion using CFD.
Char Combustion Models

Char remaining in the coal particle after devolatilisation contains fixed carbon and
subsequently undergoes a heterogeneous reaction with gaseous species at elevated
Fig. 6 Steps in
heterogeneous reactions
(After Williams et al. 2000)
Oxidising
reactants
5 1
2
4 3
temperatures (Wu 2005). Combustion of the residual char is relatively slow due to
the small reaction surface. The heterogeneous reactions in a coal particle include five
steps shown in Fig. 6 (Williams et al. 2000; Wu 2005):
Step 1. Diffusion of oxidants through the gas boundary layer surrounding the particle
(external diffusion) and through the pores of the particle (internal diffusion) to the
particle surface
Step 2 adsorption of reactants onto the particle surface
Step 3 surface reactions to form solid products
Step 4 desorption of the solid products into the gas phase
Step 5 diffusion of gas phase oxidisation products through the pores of the particle
and through the ambient gas phase to the gas stream (Williams et al. 2000;
Wu 2005).
In pulverised coal combustion, the main heterogeneous reactions include:
C þ O2 ! CO2 (51)
1
C þ O2 ! CO (52)
2
Reaction 51 dominates at lower temperature and reaction 52 is dominant with
increasing temperature. Furthermore, residual char may also react as follows:
C þ CO2 ! 2CO (53)
C þ H2 O ! CO þ H2 (54)
CO and H2 produced in the above heterogeneous reactions diffuse away from the
char particle into the ambient gas stream and react as follows:
22 Z.F. Tian et al.
CO þ 1=2 O2 ! CO2 (55)
H2 þ 1=2 O2 ! H2 O (56)
CO þ H2 O ! CO2 þ H2 (57)
Combustion of a char particle is controlled by the rate of oxygen diffusion to the

particle or the chemical reaction rate, or a combination of the two factors
(Wu 2005). In the low temperature zone (<600 C), the chemical reaction is
relatively slow and the available oxygen at the particle surface can readily diffuse
into the pores of the particle and react with carbon (Tillman 1991). The char
oxidation is determined by the chemical reaction rate rather than the rate of oxygen
diffusion. In the moderate temperature zone (about 600–800 C), the chemical
reaction rates become higher and consume oxygen faster. Both the chemical
reaction and oxygen diffusion determine the char oxidation rate (Wu 2005).
When the temperature increases further, the chemical reaction becomes so fast
that the oxygen diffusion cannot follow. The oxygen concentration at the particle
external surface diminishes to zero and no oxygen is left to diffuse into the pores
(Wu 2005). Therefore, char oxidisation rates are determined by the oxygen diffu-
sion rate to the particle surface and diffusion rate of oxygen through the porous ash
layer to the active char layer.
Several char oxidation models have been developed, mainly depending on the
reaction mechanism for the combustion. Some of these models are global reaction
models such as the diffusion-limited surface reaction model, the kinetic/diffusion
surface reaction rate model, and other models such as the Gibb model. The kinetic/
diffusion reaction rate model, both diffusion of oxidising reactors and surface
reaction rate control the total reaction rate of the char. The kinetic/diffusion surface
reaction rate model is given as follows:

dmp Pg
¼ πd2p ρg RT ½O2 (58)
dt char PA
where RT is overall reaction rate:

1
RT ¼ R1 1
diff þ RC (59)
Rdiff is the diffusion rate coefficient:

Dref T p þ T g α PA
Rdiff ¼ (60)
dp 2T ref Pg
where Dref is the dynamic diffusivity; Tref is the reference temperature normally
293 K.
Rc is the chemical rate coefficient:

TC
RC ¼ AC exp (61)
TP
where Ac and Tc are determined by the type of coal and specified as input
parameters.
For the diffusion-limited model, Rc is considered large enough, so based on
Eq. (59), RT is controlled by Rdiff that is the diffusion process of oxidant to the
reaction surface. Therefore, RT = Rdiff.
In the Gibb’s model (Gibb 1985) as cited in (ANSYS/CFX 2015), the oxidation
mechanism of carbon can be characterised by the parameter ϕ:
ϕC þ O2 ! 2ðϕ 1ÞCO þ ð2 ϕÞCO2 (62)
Here ϕ is the molar ratio of carbon atom to ϕ oxygen molecules, determined by

the particle temperature:

2ð ϕ 1Þ Ts
¼ As exp (63)
2ϕ Tp
As has a value of 2500 and Ts is 6240 K (ANSYS/CFX 2015). After solving the
above analytic equation, the rate of particle mass loss is calculated:

dmp 3ϕ Mc ρ1 1 1
¼ k1 þ ðk2 þ k3 Þ1 mp (64)
dt char 1 e c M O 2 ρc
where ec is char particle void faction; k1 is the rate of external diffusion:
D
k1 ¼ (65)
r 2p
Here D is the oxygen diffusion coefficient in the ambient gas. k2 is the surface
reaction rate:
kc
k2 ¼ ð1 ec Þ (66)
R
kc is the carbon oxidation rate:

Tc
kc ¼ Ac T p exp (67)
Tp
Ac is 14 m/sK and Tc is 21,580 K (ANSYS/CFX 2015). k3 in Eq. (64) is the rate of

internal diffusion and internal surface reaction:

k3 ¼ kc ðβc cothβc 1Þ= βc 2 aC (68)
24 Z.F. Tian et al.
0:5
where βc ¼ R kc
D p ec aC ; ac is 0.75. Dp is computed from external diffusivity D,
according to Dp = effic D.
There are other surface reaction rate models such as intrinsic model. However,
these models are not given due to the space limit. Interested readers are referred to
publications by Edge et al. (2011), Mitchell (2000) for further details.
Radiation Models
In pulverised coal boilers, thermal radiation is the dominant mode of heat

transfer. In a combustor, radiant heat transfer from the flame and combustion
products to the surrounding walls can be predicted if the radiative properties and
temperature distributions in the medium and on the walls are available (Viskanta
and Mengüç 1987). In CFD models of pulverised coal furnaces, a radiative
transfer equation (RTE) is normally employed to model radiation in the furnace
(Tian et al. 2010a):

! !
dI υ r , s K sυ
! !
¼ ðK aυ þ K sυ ÞI υ r , s þ K aυ I b υ, T g þ
ds 4π
ð 4π
! ! ! !
dI υ r , s Φ s s 0 dΩ0 (69)
0
Here, Iv is the spectral radiant intensity that depends on position r and direction
! ! !
s. r , s , s 0 are position vector, direction vector, and scattering direction vector,
respectively. Kaυ is the spectral absorption coefficient and Ksυis the spectral scatter-
ing coefficient. Ib is blackbody emission intensity; Ω0 is the solid angle.
The RTE is very difficult to solve directly due to the differential and integral
processes in the equation. Therefore, several radiation models have been developed
such as Monte Carlo model, discrete transfer (DT) model and P-1 model; these have
been used in modelling coal flames in pulverised coal furnaces.
The Monte Carlo model simulates the rays of radiation as photon trajectories and
has been recognised as the best method for modelling radiation (IIBD 2002).
However for coal-fired boilers, the computational time is extensive, making this
model less attractive.
For DT model, the scattering in the furnace is assumed to be isotropic and the
system is assumed to be reasonably homogeneous; the intensity Iv along rays leaving
from the boundaries is solved using the equation of transfer (ANSYS/CFX, 2009}:
I v ðr, sÞ ¼ I v0 expððK av þ K sv ÞsÞ þ I bv ð1 expðK a sÞÞ þ K sv I v (70)
where Iv0 is the radiation intensity leaving the boundary. The radiation intensity is
then integrated over a defined solid angle at discrete points to get the spectral
incident radiation, and the radiative heat flux; based on the homogeneity assumption
the solution is extended to the entire domain (ANSYS/CFX 2015).
In the P-1 model Iv is represented by an orthogonal series of spherical harmonics
(Sazhin et al. 1996). In the PC furnace, P-1 model assumes that radiation intensity is
isotropic anywhere in the CFD domain. In the P-1 model the following equation is
solved:

1
∇ ∇Gv ¼ K av ðEbv Gv Þ (71)
3ðK av K sv Þ AK sv
here Gv is spectral incident radiation

ð
Gv ¼ I v dΩs (72)
and A is the linear anisotropy coefficient. The thermal heat transfer is linked to the
energy equation by the term ∇qrv(Sazhin et al. 1996):
1
∇qrv ¼ ∇ ∇Gv (73)
3ðK av K sv Þ AK sv
The P-1 model is one of a number of flux methods that separate the dependency of
radiation intensity from the spatial dependency by discretization of the intensity into
vectors representing intervals of the solid angle. P-1 model has been proved adequate
for the study of pulverised fuel flames in regions away from the immediate vicinity
of the flame. This is because P-1 model is particularly useful for accounting for the
radiative exchange between gas and particles (Sazhin et al. 1996). The predicted wall
incident heat flux for a brown coal furnace based on DT and P-1 radiation model are
compared against power plant measurement and it is found that P-1 model under-
predicts the incident heat flux considerably (Tian et al. 2010a). This demonstrates
that these radiation models are not universal but case sensitive.
Another important aspect in radiation is spectral modelling. Several models can
be used for spectral integration in CFD modelling of radiative heat transfer, e.g., gray
model, multiband model and weighted sum of gray gases model (WSGGM). For
gray model, the furnace gas properties are assumed not to be a function of wave-
length, therefore saving computing time for integration of radiation intensity over
spectrum. However, this simple approach is not accurate for coal flames as gases in
the coal-fired furnaces such as CO2 and water vapour have significant different
spectral properties. In multiband models, gas properties and radiation sources are
represented by a stepwise function. The radiative flux can be calculated by the
integration through the stepwise function. In coal combustion CFD practice, the
WSGG model is also widely used. In WSGG, the emissivity and absorptivity are
approximated by the summation of gray media solutions, normally CO2, water
vapour and hydrocarbon fuels.
26 Z.F. Tian et al.
Emissions Modeling
Emissions from the tangentially fired boilers have been intensively studied using
CFD techniques. The major emission of concern is NOx, e.g., Backreedy et al.
(2005), Díez et al. (2008), Le Bris et al. (2007), because of the established restrictive
legislations that limit NOx emissions to the atmosphere (Díez et al. 2008).
The level of NOx emissions is dependent on the nitrogen content in the coal and
the combustion conditions (Hodges and Holden 2003). Nitrogen levels in Latrobe
Valley brown coals are very low by world standard and the separation firing system
of the Latrobe Valley boilers achieves a staged combustion by introducing about
20% of the fuel above the main burners, resulting in comparatively low NOx
emission.
In CFD models, the concentration of nitric oxide (NO) is normally modelled
because NO is the major NO species in emissions from pulverised coal power plant
(Visona and Stanmore 1998). As the concentration of NO is low in pulverised coal
furnaces, its effects on combustion in the furnace is negligible, so it can be modelled
after the flow fields, temperature fields and major species field are solved. Nitrogen
chemistry is extremely complex and modelling of details of chemical reactions of
NO is very difficult, therefore simple global reactions are usually used to model NO
formation and destruction in CFD models (Jones et al. 1998).
The major formation mechanisms of NO emissions in PC power plant are known
as thermal NO, prompt NO and fuel NO. The formation of thermal NO is dependent
on the temperature of the PC furnace. When the temperature in PC furnace is high
(typically greater than 1500 C), free radicals such as O and N from atmospheric
oxygen and nitrogen are abundant and start forming NO (Li et al. 2004). Major
reactions forming thermal NO are given as below (Li et al. 2004):
kN1
N2 þ O $ NO þ N (74)
kN1
kN2
N þ O2 $ NO þ O (75)
kN2
kN3
N þ OH $ NO þ H (76)
kN3
The Arrhenius form can be used to calculate reaction rate, kNi or k-Ni for each
reaction. The rate of thermal NO formation is therefore calculated by the following
equation (Li et al. 2004), under the assumption of quasi steady state:
d ½NOt 1 ½NO2 =kN ½N2 ½O2

¼ 2kN1 ½N2 ½O2 (77)
dt 1 þ kN1 ½NO=ðkN2 ½O þ kN3 ½OHÞ
Here kN1, kN2, kN3 are reaction rates of reaction (74–76), respectively.
kN = (kN1/k-N1)(kN2/k-N2).
The complete mechanisms for prompt NO formation are complex and usually
prompt NO is higher in rich flames than in lean flames. A single global model is
usually used to calculate prompt NO formation as below (Li et al. 2004):
d ½NOprompt
¼ f N T βNO Apr ½O2 αNO ½N2 ½FuelbNO exp Ea =RT g (78)
dt
where fN is a correction factor. Apr is the pre-exponential factor; aNO and bNO are
reaction orders.
Fuel NO accounts for 70–90% of NO emission in fossil fuel combustion and is
the major part of NO emissions in fossil fuel combustion (Li et al. 2004). Fuel NO is
mainly formed from the nitrogen compounds in the coal and HCN has been found to
be the major precursor of fuel NO. Major reactions for formation of fuel NO in many
CFD models are as below:
kN4
HCN þ O2 $ HCO þ NO (79)
kN4
Again an Arrhenius form of reaction rates can be used to calculate kN4 and k-N4 of
the above reactions. When reactions 76 and 79 are included in the CFD coal model,
the species transport equations of HCN, HCO and OH will be included as well.
Another concern of emissions is the emission of SOx that contributes to acid rain
and the corrosion of power plant equipment. More than 50% of SOx emission in the
world is from coal fired power plants (Perera and Faltsi-Saravelou 2007). Coals have
sulphur concentration that varies from 0.2% to 11% by weight depending on coal
type and the environmental conditions (Boardman and Smoot 1993; Wu 2005).
Sulphur exists in coals in two different forms, pyritic and organic sulphur and after
combustion, most sulphur converts to SO2 (about 90%) and a small fraction of
sulphur is captured in the ash (Wu 2005). The mechanisms of sulphur oxidation are
complex and can be modelled by either by a global model or reduced mechanism
(Perera and Faltsi-Saravelou 2007).
The levels of sulphur in Latrobe Valley coals are very low by world standards
(typically 0.2% dry-based) and a significant portion (10 ~ 30%) of the sulphur is
retained in the ash by sulphation of basic inorganic oxides (Kiss et al. 1984).
Therefore, the sulphur emission from Latrobe Valley power stations are below
licence limits without the need for flue-gas desulphurisation (Hodges and Holden
2003).
Soot is also an emission problem from PC power plants. Furthermore, soot plays
an important role in radiative heat transfer in the furnace. Therefore a soot model is
required to accurately calculate heat transfer and understand emission problems from
PC furnaces. Many methods have been developed to model soot formation in
gaseous flames. However these methods may not be appropriate to use in coal
flames, as tar is the principle precursor to soot in a coal flame, while acetylene is
the major precursor to soot in gaseous hydrocarbon flames (Brown and Fletcher
28 Z.F. Tian et al.
1998). Therefore, it is critical to include a tar transport equation in the CFD coal soot
model.
Several coal soot models have been developed in the literature and one advanced
model is from Brown and Fletcher (1998). In this model, the equations of mass
fraction of soot (Yc) and tar (YT) are given as below (Brown and Fletcher 1998):
μ
∇ ρg U g Y C ¼ ∇ ∇Y C þ SC (80)
σ
μ
∇ ρg U g Y T ¼ ∇ ∇Y T þ ST (81)
σ
A transport equation of soot number (Nc) is also included (Brown and Fletcher
1998):
μ
∇ ρg U g N C ¼ ∇ ∇N C þ SN (82)
σ
Source terms of the above equations are (Brown and Fletcher 1998):
SY C ¼ ρg ðr_FC r_OC Þ (83)
SY T ¼ ρg ðr_FT r_FC r_GT r_OT Þ (84)
SNC ¼ ρg ððN a =MC CminÞ r_AN Þ (85)
where
r_FT ¼ SPtar (86)
r_OT ¼ ρg ½cT ½cO2 AOT eEOT =RT (87)
r_GT ¼ ½cT AGT eEGT =RT (88)
r_FC ¼ ½cT AFC eEFC =RT (89)

PO2
r_OC ¼ SAv:C AOC eEOC =RT (90)
T 1=2
1=3
2=3 2=3 2=3
SAv:C ¼ 6 π
2=3 1=3
ρg N C Y C ρg =ρC (91)
1=6
6MC 6kT 1=2 ρg Y C 1=6 11=6
r_AN ¼ 2Ca ρg N C (92)
πρC ρC MC
The Arrhenius constants in the above equations can be found in the literature, for
example (Brown and Fletcher 1998).
Fig. 7 Slagging on the

superheater tubes in Loy Yang
power station boilers
Table 1 Predicted flue gas exit temperature for cases with clean condition (no slagging layer) and
with slagging layer thickness = 25 mm
Slagging layer thickness (mm) Flue gas exit temperature (C )
0 1091
25 1181
The model has been validated against limited data available in a flat flame and
reasonable agreement was achieved (Brown and Fletcher 1998). However, more
detailed and reliable measurements of soot and tar in coal flames are not yet available
and are required to further develop, tune and validate this model. Furthermore, the
effect of turbulence on the soot transport and formation is not clear in the model.
Ash particle dispersion, fouling and slagging present another issue where CFD
modelling can provide benefits. Slagging is the deposition of particles (ash, coal) on
the furnace parts where radiative heat transfer is dominant, e.g., the water wall tubes
and superheater tubes. Fouling is the deposition of ash on the convective parts of the
boiler. The ash yields of Latrobe Valley coals are lower than other overseas low rank
coals and generally between 1 ~ 3% dry-based (Hodges and Holden 2003). How-
ever, fouling and slagging remains a major concern for boiler operators. Figure 7
shows a picture of the slagging in the superheater of the tangentially fired brown coal
boiler in Loy Yang power station in the Latrobe Valley.
Table 1 shows the predicted flue gas exit temperature (FGET) of the unit No. 3 at
TRUenergy’s Yallourn power plant without considering slagging and when a slag-
ging layer thickness of 25 mm is accounted for. It can be seen that the FGET
increases when the slagging layer thickness increases. For moderate fouling brown
coals such as Yallourn and Loy Yang brown coal, the designed FGET is about 1050
C . The FGET is much higher than 1050 C when there is a slag layer of 25 mm
(about 130 C higher). The increase of FGET is quite likely to have a negative effect
on the fouling behaviour of ash downstream the convective parts of the furnace.
30 Z.F. Tian et al.
Fig. 8 Predicted gas temperature at the mid-plane of the furnace for cases, (a) clean condition
(no slagging layer), (b) slagging layer thickness = 25 mm
Figure 8 shows the predicted gas temperature at the mid-plane of the furnace with
and without a slagging layer. The temperature of the case with 25 mm slagging layer
(b) is higher than the clean condition case (a) in the center zone of furnace at the
primary burner level. The high temperature zone is further increased when the
slagging layer thickness is further increased. This increased temperature is due to
the reduction of heat transfer to the water walls by the presence of the slagging layer.
The increased temperature in the center zone of the furnace leads to the increased
FGET.
The above calculations given in Table 1 and Fig. 8 are based on the boundary
condition that assumes the thickness of slagging layer in the furnace to be uniform.
When the exact distribution of slagging layer thickness is required, a slagging model
is necessary to calculate the distribution of the slagging layer thickness and to
provide more accurate boundary conditions to the gas phase predictions.
Several slagging models have been developed for CFD modelling of coal gasi-
fication. Seggiani (1998) developed a one dimensional slag model that is coupled
with a 3 dimensional (3D) CFD coal gasifier code. The 3D CFD code provides the
particle deposition mass flow rate, particle temperature, and gas temperature to the
slag model. The slag model calculates the slag mass flow rate, liquid slag layer
thickness, solid slag layer thickness, slag layer outside surface temperature,
refractory temperature, etc., based on the slag mass conservation equation, energy
conservation equation for the refractory wall and momentum conservation equation
(Seggiani 1998). The slag layer outside surface temperature is sent back to the 3D
CFD code as the wall boundary condition. This slag model is 1D since it only
calculates the slag distribution in an axial direction. Wang et al. (2007) included a
wall burning model into the 1D slag model. The wall burning model is able to
calculate combustion of particles captured by the slagging layer.
Recently, Chen et al. (2013) developed a 3D slag model for coal combustion and
gasification. In this model, a volume-of-fluid (VOF) model is used to calculate the
slag phase which interacts with a Langrangian particle tracking model. This model is
able to calculate slag distributions and flows in a 3D manner. Nevertheless, the 3D
slag model is still not able to predict the solid slag and further work is needed to
capture the solid slag (Chen et al. 2013).
Conclusions
In 2010, coal-fired electricity generation accounted for 40% of the electricity gen-
eration worldwide. These plants rely on pulverised coal combustion carried out in
large furnaces with multiple interacting burners. Flow and thermal conditions in the
furnaces are not easily understood because they involve complex turbulent flow
fields interacting with complicated combustion processes and heat transfer. In real
industrial installations, additional complications arise from slagging and fouling of
heat transfer surfaces, variability in feed characteristics and inevitable uncertainties
in actual structural geometry due, for example, to occasional damage or maintenance
issues. Design and improvement of these furnaces can be effectively assisted by
using numerical modeling with CFD techniques to develop a detailed picture of the
conditions within the furnace, and the effect of operating conditions, coal type, and
furnace design on those conditions.
The equations governing CFD models of pulverised coal combustion are
described, with a focus on sub-models needed for devolatilisation, combustion and
heat transfer. The use of the models is discussed with reference to examples of CFD
modelling of brown coal fired furnaces in the Latrobe Valley in Australia and black
coal fired furnaces described in the literature. Extensions of the CFD models that are
required to tackle specific industrial and environmental issues are also described.
These issues include control of NOx, SOx, and soot emissions and the effect of
slagging and fouling on furnace and boiler operation.
In a PC furnace, fine coal particles pulverised in mills are blown into the furnace
through burners. Once the particles enter the furnace, they are heated by the hot
furnace gas and radiation from the flame, causing drying and devolatilisation.
Combustion of the volatiles occurs, followed by char combustion. The furnace
gases (together with remaining ash) flow through to the convective heat transfer
sections.
As a result, a CFD model of the process must include, in addition to the basic gas
flow, sub-models to account for the complex physics and chemistry involved. These
32 Z.F. Tian et al.
include: (1) particle phase model for particle phase motion, (2) evaporation/drying
model, (3) devolatilisation model, (4) char combustion model, (5) turbulence model,
(6) turbulence-reaction interaction model (gas phase reaction models), (7) radiation
model, (8) soot model, (9) NOx model, (10) SOx model, (11) slagging model. (Tian
et al. 2010a).
Because of the crucial importance of mixing on combustion and heat transfer, the
turbulence model is critical for achieving a reliable basis for a furnace model. Both
the standard k-e model and the SST model appear to perform satisfactorily, although
various enhancements have been found to improve the prediction of specific aspects
of the flow. In the case of swirl burners, higher order turbulence models such as the
RSM are required.
As with turbulence, the sub-model for radiative transfer is of great importance in
generating a reliable basis for the over-all model, because so many of the
sub-processes involved in coal combustion are temperature dependent. The Monte
Carlo method is the best technique for modelling radiation, but is computationally
very expensive. The Discrete Transfer (DT) method and the P-1 flux method have
been found to be adequate in most PC furnace cases.
Coal devolatilisation and combustion are modelled by means of equations that
depend on local gas and particle temperatures, and local concentrations, as predicted
by the CFD model. Various sets of equations have been developed, with the
complexity of the reaction schemes ranging from a small number of effective
reactions to complex schemes involving many species. Because of the variability
in coal properties, not only from plant to plant, but also from hour to hour, simpler
schemes are often found to be adequate. These also have the advantage of limiting
computational run times. For volatile combustion, the generalised eddy dissipation
concept model is often found to be satisfactory, since it takes into account the critical
aspects of turbulent mixing, micro-mixing and chemical kinetics. More sophisticated
flamlet models should undoubtedly be more accurate, but the advantages can be
outweighed by uncertainties in coal characterisation.
It has been found that char combustion is adequately represented by a diffusion-
limited surface reaction model, or by a mixed kinetics/diffusion surface reaction rate
model. As with devolatilisation, difficulties in characterisation and variability in
properties limits the extent to which additional model complexity achieves improved
predictive capability in real industrial applications.
SOx , NOx, and soot emissions are of major environmental importance, and the
ability to predict the amounts of these emissions can lead to furnace improvements
that give better environmental performance. It is essential therefore to use
sub-models for these species that capture all the critical sub-processes. Such
sub-models have been published in the literature and used with some success (e.g.,
Li et al. 2004; Brown and Fletcher, 1998), but on-going experimental work is
required to further refine the equations.
Slagging and fouling can have a major impact on furnace and boiler performance.
A comprehensive model of such complicated phenomena has not yet been devel-
oped, but CFD modelling can nonetheless provide extremely valuable insights into
slagging and fouling – their effects and how to minimise them. For example, the
authors have investigated the sensitivity of the flue gas exit temperature (FGET) of
the unit No. 3 at TRUenergy’s Yallourn power plant to the thickness of a slagging
layer.
In summary, CFD models for PC combustion have now been developed that
provide adequately accurate prediction of furnace performance to be useful for the
purposes of design and optimisation. On-going refinement of various sub-models
will be necessary to improve predictive capability. This endeavour will require
further experimental investigation of coal combustion sub-processes and improved
characterisation additionally improved measurement techniques for collecting
detailed data within operating furnaces will be needed. Further programs of the
sort carried out by Tian et al. (2012) involving modelling of industrial furnaces
together with industrial measurements for model validation are also necessary.
Acknowledgment The first author gratefully acknowledges the support of the Australian Research
Council through Grant DP150102230.
References
D. Allardice, The utilisation of low rank coals. Aust. Coal Rev. (10), 40–46 (2000)
ANSYS/CFX, ANSYS/CFX 16.0 Theory Guide (ANSYS, Inc, Canonsburg, 2015)
ANSYS/FLUENT, ANSYS FLUENT 16.0 User’s Guide (ANSYS, Inc, Canonsburg, 2015)
R. Backreedy, L. Fletcher, L. Ma, M. Pourkashanian, A. Williams, Modelling pulverised coal
combustion using a detailed coal combustion model. Combust. Sci. Technol. 178, 763–787
(2006)
R. Backreedy et al., Prediction of unburned carbon and NOx in a tangentially fired power station
using single coals and blends. Fuel 84, 2196–2203 (2005)
P. Basu, C. Kefa, L. Jestin, Boilers and Burners: Design and Theory (Springer, New York, 1999)
S. Belosevic, M. Sijercic, S. Oka, D. Tucakovic, Three-dimensional modeling of utility boiler
pulverized coal tangentially fired furnace. Int. J. Heat Mass Transf. 49, 3371–3378 (2006)
A. Benim, B. Epple, B. Krohmer, Modelling of pulverised coal combustion by a Eulerian-Eulerian
two-phase flow formulation. Progr. Comput. Fluid Dyn. Int. J. 5, 345–361 (2005)
K.S. Bhambare, Z. Ma, P. Lu, CFD modeling of MPS coal mill with moisture evaporation. Fuel
Process. Technol. 91, 566–571 (2010)
R.D. Boardman, L.D. Smoot, in Fundamentals of Coal Combustion: For Clean and Efficient Use,
ed by L. D. Smoot. Pollutant formation and control, Chapter 6 (Elsevier Science Publishers,
Amsterdam, 1993)
BREE, 2014 Australian Energy Update (Bureau of Resources and Energy Economics, Canberra,
2014)
A.L. Brown, T.H. Fletcher, Modeling soot derived from pulverized coal. Energy Fuel 12, 745–757
(1998)
L. Chen, S.Z. Yong, A.F. Ghoniem, Modeling the slag behavior in three dimensional CFD
simulation of a vertically-oriented oxy-coal combustor. Fuel Process. Technol. 112, 106–117
(2013)
L.I. Díez, C. Cortés, J. Pallarés, Numerical investigation of NOx emissions from a tangentially-fired
utility boiler under conventional and overfire air operation. Fuel 87, 1259–1269 (2008)
T.H. Duong, Mathematical modelling of pulverised Victorian brown coal combustion and heat
transfer in a plug-flow reactor, in Proceedings of Third Australian Conference on Heat and Mass
Transfer (1985)
34 Z.F. Tian et al.
T.H. Duong, Modelling of brown coal combustion in one-dimension, NERDDP Project 931, End of
grant Report, Report No ND/87/040, The State Electricity Commission of Victoria, 1987
P. Edge et al., Combustion modelling opportunities and challenges for oxy-coal carbon capture
technology. Chem. Eng. Res. Des. 89, 1470–1493 (2011)
EIA, International Energy Outlook 2013 (U.S. Energy Information Administration, Washington,
DC, 2013)
J. Fan, L. Qian, Y. Ma, P. Sun, K. Cen, Computational modeling of pulverized coal combustion
processes in tangentially fired furnaces. Chem. Eng. J. 81, 261–269 (2001)
M.A. Field, D.W. Gill, B.B. Morgan, P.G.W. Hawksley, Combustion of Pulverized Coal (BCURA,
Leatherhead, 1967), pp. 189–192
W. Fiveland, R. Wessel, Numerical model for predicting performance of three-dimensional pulver-
ized-fuel fired furnaces. J. Eng. Gas Turbines Power 110, 117–126 (1988)
A. German, T. Mahmud, Modelling of non-premixed swirl burner flows using a Reynolds-stress
turbulence closure. Fuel 84, 583–594 (2005)
J. Gibb, in Cited in ANSYS/CFX 16.0 Theory Guide. Combustion of residual char remaining after
devolatilization. Lecture at course of pulverized coal combustion imperial college (ANSYS,
Inc., Canonsburg, 1985)
D.M. Grant, R.J. Pugmire, T.H. Fletcher, A.R. Kerstein, Chemical model of coal devolatilization
using percolation lattice statistics. Energy Fuel 3, 175–186 (1989)
S. Hodges, J. Holden, in A Continuing Education Course on the Science and Technology of Lignite
Utilisation. Impact of coal quality on power station performance (CRC for Clean Power from
Lignite, Melbourne, 2003)
IIBD, Improving industrial burner design with computational fluid dynamics tools: progress, needs,
and R&D priorities. Workshop Report (2002)
J. Jones, M. Pourkashanian, A. Williams, R. Chakraborty, J. Sykes, Modeling of Coal Combustion
Processes – A Review of Present Status and Future Needs (Pittsburgh Coal Conference,
Pittsburgh, 1998)
J.C. Jones, W.O. Stacy Devolatilization of Victorian brown coal. Part 2. Oxidizing conditions, R&
D ReportND/86/025, State Electricity Commission of Victoria, (1986)
R. Jovanovic, A. Milewska, B. Swiatkowski, A. Goanta, H. Spliethoff, Sensitivity analysis of
different devolatilisation models on predicting ignition point position during pulverized coal
combustion in O2/N2 and O2/CO2 atmospheres. Fuel 101, 23–37 (2012)
L.T. Kiss, D.J. Brockway, A.M. George, W.O. Stacy Properties of brown coals from the Rosedale,
Stradbroke and Gormandale fields. SECV, Research and Development Department Report No
SC/84/85 (1984)
B.E. Launder, D. Spalding, The numerical computation of turbulent flows. Comput. Method Appl.
M 3, 269–289 (1974)
T. Le Bris, F. Cadavid, S. Caillat, S. Pietrzyk, J. Blondin, B. Baudoin, Coal combustion modelling
of large power plant, for NOx abatement. Fuel 86, 2213–2220 (2007)
K. Li, S. Thompson, J. Peng, Modelling and prediction of NOx emission in a coal-fired power
generation plant. Control. Eng. Pract. 12, 707–723 (2004)
F. Lockwood, A. Salooja, The prediction of some pulverized bituminous coal flames in a furnace.
Combust. Flame 54, 23–32 (1983)
B.F. Magnussen, B. Hjertager, On the structure of turbulence and a generalized eddy dissipation
concept for chemical reaction in turbulent flow, in 19th AIAA Aerospace Meeting, St. Louis,
1981.
B.F. Magnussen, B.H. Hjertager, in Symposium (International) on Combustion. On mathematical
modeling of turbulent combustion with special emphasis on soot formation and combustion, vol
1 (Elsevier, Amesterdam, 1977), pp. 719–729
F.R. Menter, Improved two-equation k-omega turbulence models for aerodynamic flows. NASA
STI/Recon Technical Report N 93: 22809 (1992)
F.R. Menter, Two-equation eddy-viscosity turbulence models for engineering applications. AIAA
J. 32, 1598–1605 (1994)
R.E. Mitchell, An intrinsic kinetics-based, particle-population balance model for char oxidation
during pulverized coal combustion. P Combust Inst 28, 2261–2270 (2000)
M. Mulcahy, W. Morley, I. Smith Combustion, gasification and oxidation. The Science of Victorian
Brown Coal: structure, Properties, and Consequences for Utilization (1991), p. 359
S. Niksa, Coal combustion modelling, vol 31 (IEA Coal Research, London, 1996)
S. Ouyang, H. Yeasmin, J. Mathews, A pressurized drop-tube furnace for coal reactivity studies.
Rev. Sci. Instrum. 69, 3036–3041 (1998)
S. Perera, O. Faltsi-Saravelou, SOx formation model for turbulent combustion applications, in Third
European Combustion Meeting ECM (2007)
R.A. Roberts, D.J. Loveridge, Devolatilisation and combustion rate measurements on pulverised
fuel particles of Morwell, Morwell Woody and Loy Yang coal. Document 4C/47, British Coal
Utilisation Research Association (1969)
R.B. Sainsbury, P.G.W. Hawksley, Devolatilisation and combustion rate measurements on
pulverised fuel particles of Yallourn Open Cut Coal. Combustion Note 825. British Coal
Utilisation Research Association (1969)
K.M. Saqr, Comments on:“CFD analysis on the influence of helical carving in a vortex flow solar
reactor” by N. Ozalp and D. JayaKrishna (Int. J. Hydrogen Energy 2010: 35, 6248–6260). Int.
J. Hydrogen. Energy 36, 2320–2322 (2011)
S. Sazhin, E. Sazhina, O. Faltsi-Saravelou, P. Wild, The P-1 model for thermal radiation transfer:
advantages and limitations. Fuel 75, 289–294 (1996)
M. Seggiani, Modelling and simulation of time varying slag flow in a Prenflo entrained-flow
gasifier. Fuel 77, 1611–1621 (1998)
P.J. Street Single particle studies of brown coal combustion. CEGB Research Division Memoran-
dum, MM/COMB, TH94 (1979)
Z.F. Tian, G.J. Nathan, Y. Cao, Numerical modelling of flows in a solar-enhanced vortex gasifier:
Part 1, comparison of turbulence models. Progr. Comput. Fluid Dyn. Int. J. 15, 114–122 (2015)
Z.F. Tian, J.Y. Tu, G.H. Yeoh, Numerical simulation and validation of dilute gas-particle flow over a
backward-facing step. Aerosol Sci. Technol. 39, 319–332 (2005)
Z.F. Tian, P.J. Witt, M.P. Schwarz, W. Yang, Comparison of two-equation turbulence models in
simulation of a non-swirl coal flame in a pilot-scale furnace. Combust. Sci. Technol. 181,
954–983 (2009)
Z.F. Tian, P.J. Witt, M.P. Schwarz, W. Yang, Modeling issues in CFD simulation of brown coal
combustion in a utility furnace. J. Compt. Multiph. Flow 2, 73–88 (2010a)
Z.F. Tian, P.J. Witt, M.P. Schwarz, W. Yang, Numerical modeling of Victorian brown coal
combustion in a tangentially fired furnace. Energy Fuel 24, 4971–4979 (2010b)
Z.F. Tian, P.J. Witt, M.P. Schwarz, W. Yang, Combustion of predried brown coal in a tangentially
fired furnace under different operating conditions. Energy Fuel 26, 1044–1053 (2012)
Z.F. Tian, P.J. Witt, W. Yang, M.P. Schwarz, Numerical simulation and validation of gas-particle
rectangular jets in crossflow. Comput. Chem. Eng. 35, 595–605 (2011)
D. Tillman, The Combustion of Solid Fuels and Wastes (Academic Press, San Diego, 1991)
J. Truelove, D. Holcombe, in Symposium (International) on Combustion. Measurement and
modelling of coal flame stability in a pilot-scale combustor, vol 1 (Elsevier, Amesterdam,
1991), pp. 963–971
M. Vascellari, G. Cau, Influence of turbulence–chemical interaction on CFD pulverized coal MILD
combustion modeling. Fuel 101, 90–101 (2012)
H.K. Versteeg, W. Malalasekera, An introduction to computational fluid dynamics: the finite volume
method (Pearson Education, Harlow, 2007)
R. Viskanta, M. Mengüç, Radiation heat transfer in combustion systems. Prog. Energy Combust.
Sci. 13, 97–160 (1987)
S. Visona, B. Stanmore, Modelling NO formation in a swirling pulverized coal flame. Chem. Eng.
Sci. 53, 2013–2027 (1998)
X. Wang, D. Zhao, L. He, L. Jiang, Q. He, Y. Chen, Modeling of a coal-fired slagging combustor:
development of a slag submodel. Combust. Flame 149, 249–260 (2007)
36 Z.F. Tian et al.
R. Weber, A. Peters, P. Breithaupt, B. Visser, Mathematical modeling of swirling flames of

pulverized coal: what can combustion engineers expect from modeling? J. Fluids Eng. 117,
289–297 (1995)
R. Weber, B. Visser, F. Boysan, Assessment of turbulence modeling for engineering prediction of
swirling vortices in the near burner zone. Int. J. Heat Fluid Flow 11, 225–235 (1990)
D.C. Wilcox, Reassessment of the scale-determining equation for advanced turbulence models.
AIAA J. 26, 1299–1310 (1988)
A. Williams, M. Pourkshanian, J.M. Jones, N. Skorupska, Combustion and Gasification of Coal
(Taylor & Francis, New York, 2000)
Z. Wu, Fundamentals of Pulverised Coal Combustion (IEA Clean Coal Centre Reports, London,
2005)
M. Xu, J. Azevedo, M. Carvalho, Modeling of a front wall fired utility boiler for different operating
conditions. Comput. Method Appl. M 190, 3581–3590 (2001)
H. Yeasmin, J. Mathews, S. Ouyang, Rapid devolatilisation of Yallourn brown coal at high
pressures and temperatures. Fuel 78, 11–24 (1999)
G.H. Yeoh, K.K. Yuen, Computational Fluid Dynamics in Fire Engineering: Theory, Modelling
and Practice (Butterworth-Heinemann, Amsterdam, 2009)
J. Zhang, S. Nieh, Comprehensive modelling of pulverized coal combustion in a vortex combustor.
Fuel 76, 123–131 (1997)
L. Zhou, L. Li, R. Li, J. Zhang, Simulation of 3-D gas-particle flows and coal combustion in a
tangentially fired furnace using a two-fluid-trajectory model. Powder Technol. 125, 226–233
(2002)
Cavitation Flow of Cryogenic Fluids
Xiaobin Zhang and Zhu Jiakai
Abstract
Cryogenic cavitation is important for the operation of rocket propulsion system and
is complex because of its strong thermal effects. In this chapter, a brief introduction
of this phenomenon is given out. Theoretical models historically developed to
estimate thermal effects are classified by their physical hypothesis and deduced
according to their proposers. The summary of the past experimental studies helps
to provide the appearance, thermodynamic state and features of cryogenic cavitation
over different geometries. Besides, a robust numerical framework for cryogenic
cavitation modeling is built in details. Vorticity transport analysis further reveals
the mechanism for unique partially shedding mode appeared in cryogenic cavitation.
Keywords
Cavitation • Cryogenic • Thermal effects • B-factor
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Parameters Developed to Estimate Thermal Effects on Cavitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
B Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Thermal Aspects From Bubble Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Experimental Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Appearance of Cavitation with Thermal Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Thermodynamic State of Sheet Cavitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Thermal Effects on Dynamics of Cavitation over Simple Geometries . . . . . . . . . . . . . . . . . . . . . . 15
Thermal Effects on Cavitation in Rocket Inducers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Numerical Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Establish an Effective Numerical Framework Based on the Mixture Model . . . . . . . . . . . . . . . . 20
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
X. Zhang (*) • Z. Jiakai

Institute of Refrigeration and Cryogenics, Zhejiang University, Hangzhou, Zhejiang, China
e-mail: zhangxbin@zju.edu.cn

DOI 10.1007/978-981-4585-86-6_13-1
2 X. Zhang and Z. Jiakai
Introduction
The phenomenon by which a liquid forms gas-filled or vapor-filled cavities under the
effect of tensile stress produced by a pressure below its vapor pressure is termed
cavitation (Batchelor 1967). On the one hand, the cavitation is unavoidable and
undesirable, which will reduce the flow rates of venture and the efficiency of pumps;
make the vane load asymmetry, vibration, and noise; cause the serious erosion
damage; and consequently, reduce the lifetime. On the other hand, unique phenom-
ena (such as strong pressure and temperature fluctuations) accompanied in the
cavitation process are virtual for numerous applications, for example, drug delivery
(Ohl et al. 2003), shock wave lithotripsy (Tanguay and Colonius 2003), and
manufacturing and material processing applications (Soyama and Macodiyo 2003).
Cryogenic liquids are the fluid whose temperature is below 123 K (Thipse 2013).
Typical cryogenic liquids, such as liquid hydrogen, liquid nitrogen, liquid natural
gas, and liquid oxygen, usually serve as the fuels in the rocket propulsion system,
and other low temperature applications, and are generally operated close to their
critical points. Cavitation in cryogenic fluids presents richer physics than in ordinary
fluids like water. This results from the unique properties of cryogenic fluids as shown
in Table 1. First and foremost, the liquid-vapor density ratio of liquid nitrogen is
lower than that of cold water by an order of magnitude, and more liquid mass has to
vaporize to sustain a vapor cavity with the equal size. Therefore, the liquid sur-
rounding the vapor cavity undergoes substantial evaporative cooling in case of liquid
nitrogen, and the temperature near the liquid-vapor interface drops below the free-
stream temperature. Besides, the thermal conductivity of cryogenic liquid is much
lower than water; thus, with the same heat flux, a larger gradient of temperature will
be generated within the cryogenic fluid. In addition, thermodynamic properties
(especially the vapor pressure) are very sensitive to temperature variations. With
the temperature depressions of only 1–2 K, the value of vapor pressure could vary a
factor of 2. Consequently, the constant temperature hypothesis used in cavitating
process in water is no longer valid when predicting cavitation in cryogenic fluids for
its thermosensitive properties. This phenomenon is also regarded as the thermal
effect or thermal delay (Brennen 1973). This chapter will strive to review the past
efforts made on the cryogenic cavitation research.
Table 1 Properties of some cryogens and water (Lemmon et al. 2007)

Specific Liquid/ Liquid Thermal
heat vapor density Vaporization conductivity
Substance (J/kg-K) density (kg/m3) heat (KJ/Kg) (W/mK)
H2 9,816 53 71 446 100
N2 2,046 175 809 199 135
Water 4,200 1,603 958 2,257 681
Cavitation Flow of Cryogenic Fluids 3
Parameters Developed to Estimate Thermal Effects on Cavitation
B Factor
Historically, thermal effect of cavitation draws the interest of researchers because of

the need for prediction of cavitating performance for a specific fluid-pump combi-
nation using test data from a different fluid-pump combination.
Cavitation characteristics are of great importance in the operation of centrifugal
pumps (Hirschi et al. 1998; Coutier-Delgosha et al. 2003; Medvitz et al. 2002).
To describe the dynamic similarity of different cavitating flow, the conventional
cavitation parameter σ is usually utilized, which is defined as (Brennen 1973)
h0 hv
σ¼ (1)
V 20
2g
where h0 is the free-stream static pressure head and V0 is the free-stream

velocity, hv is the vapor pressure head corresponding to free-stream temperature,
and g is the acceleration of gravity. Equation 1 makes sense only when the
cavity surface is at a constant pressure equal to the free-stream vapor pressure.
However, as pointed out in the introduction, in some fluids such as liquid
hydrogen, because of the thermodynamic effects, cavity pressure can be signif-
icantly less than free-stream vapor pressure. In such situation, hv should be
replaced by the actual cavity pressure hc. As will be pointed out in the next
sections, at the presence of thermal effects, the pressure isn’t a constant value
within the cavity. Thus the minimum cavity pressure head hc, min, which occurs
near the cavity leading edge, is usually used to define a modified cavitation
parameter for developed cavities,
h0 hc, min h0 hv þ hv hc, min Δhv

σ c, min ¼ 2
¼ 2
¼σþ 2 (2)
V0 V0 V0
2g 2g 2g
where Δhv is the vapor-pressure depression due to temperature depression.

NPSH (net positive suction head) is the most common term used to describe
pump cavitation performance. The NPSH is defined as the margin of fluid total head
above fluid vapor pressure head at the pump inlet or in terms of inlet static-pressure
head and velocity head (Rzlggeri and Moore 1969),
V 21
NPSH ¼ h0 þ hV (3)
2g
Combine Eqs. 2 and 3, then

NPSH þ ΔhV
σ c, min ¼ 1:0 (4)
V 21
2g
For geometrically similar cavities in a given flow device, the value of σ c,min is
assumed to a constant (Gelder et al. 1966); thus, we can obtain (Rzlggeri and Moore
1969)

NPSHref þ ðΔhV Þref ðV l Þref 2
¼ (5)
NPSH þ ΔhV Vl
where the subscript ref denotes a reference value that must be established by
experiment. The fluid velocity at the pump inlet can be expressed as the product of
pump rotative speed N and diameter D, and thus Eq. 5 may be expressed as

NPSHref þ ðΔhV Þref N ref Dref 2
¼ (6)
NPSH þ ΔhV ND
If (ΔhV)ref and ΔhV are known, then above relationship can be employed in
predicting prototype performance from model tests. These two terms also represent
the thermodynamic effects on cavitation, and estimation of their values is of critical
importance.
Stahl and Stepanoff (1956) were the first to analyze the thermodynamic
effect for developed cavitation. They proposed the B factor, which is defined
as the ratio of vapor volume to liquid volume involved in the vaporization
process. Fisher (1945) and Jacobs (1961) also considered the B-factor method,
and their formations are essentially equivalent with that of Stahl and Stepanoff
as indicated by Acosta and Hollander (1959). Spraker also utilized this
method (1965).
A simple heat balance between liquid and vapor is written:
ρv ϑv L ¼ ρl ϑl CPl ΔT (7)
So that
ϑv ΔT
B¼ ¼ (8)
ϑl ΔT
where ϑl is the volume of liquid cooled by the production of vapor volume ϑv, ΔT is
the actual temperature drop between the liquid bulk temperature and the temperature
inside the cavity, and ΔT is the characteristic temperature drop calculated by ΔT

¼ ρρvcLPl . B is also called thermal cavitation criterion.
l
Equation 8 can also be further written in a more usable form:

Fig. 1 Basic model for B-factor theories

ϑv ΔT ρl cPl dT
B¼ ¼ ¼ δhv (9)
ϑl ΔT ρv L dhv
where δhv is the reduction in local static pressure.

If we define

1 dT ρl cPl dT B
β¼ ¼ , then δH ¼ (10)
ΔT dhv ρv L dhv β
Combine Eqs. 8 and 9,

Bref N ref Dref 2 B
NPSH ¼ NPSHref þ (11)
βref ND β
Equation 11 is utilized to explain thermodynamic scale effects on NPSH require-

ment in pumps. The main practical problem before using Eq. 11 is the calculation
of B.
There are several approaches to estimate the value of B which will be described in
details in the following sections. As shown in Fig. 1, there are common assumptions
for the basic B-factor models: there exist a thermal boundary layer and a vapor cavity
with thickness of δl and δv, respectively; the vapor cavity is stable and attaches at the
solid wall; vapor evaporates on the liquid vapor interface and extract heat from the
thermal boundary layer within which there exists a temperature gradient between
reference liquid temperature and cavity temperature; and the vapor and liquid on the
liquid-vapor interface are in thermodynamic equilibrium.
Gelder et al. and Moore and Ruggeri Approach

Gelder et al. and Moore and Ruggeri developed the B-factor theory in a semiempir-
ical manner (Gelder et al. 1966; Moore and Ruggeri 1962). They implied a reference
value of B from measured cavity pressure depressions for one fluid, temperature, and
velocity and then estimating relative values of B for other conditions and liquids.
Consider a developed cavity shown in Fig. 1, the shape of cavity is stable, and
the vaporization of vapor extracts heat from the liquid layer. Liquid volume Vl is
assumed proportional to the product of thickness of the liquid layer δl and cavity
length Δx. Vapor volume Vv is assumed to be proportional to the product of
cavity thickness δv and cavity length Δx. δl is assumed to be proportional to the
square root of the product of thermal diffusivity α and vaporization time, as
indicated in Eisenberg and Pond (1948). The vaporization time is proportional to
Δx/V0, where V0 is the reference velocity. Thus, δl is proportional to (αΔx/V0)0.5.
Consequently, the vapor to liquid volume ratio is proportional to δv(αΔx/V0)0.5.
Then, the B factor can be predicted relative to a reference value obtained from
tests as follows:
n1 h
V0 αref in2 Δxref n3 δv n4
B ¼ Bref (12)
V 0, ref α Δx δv, ref
Based on the experiment for one venturi scale, Gelder, et al. and Moore and
Ruggeri calibrated the exponents from the experimental data of venturi cavitation in
Freon 114. It is noted that since it was difficult to determine the cavity thickness, δv
was assumed to be equal to δv,ref. A semiempirical form is expressed as
0:85 h i
V0 αref 0:5 Δxref 0:16
B ¼ Bref (13)
V 0, ref α Δx
This relation is able to predict the B factor based on a reference value of B for
changes in liquid, temperature, velocity, or cavity length.
In some situations, the scale of prototype is different from that of reference model,
then Eq. 13 isn’t suitable. Moore and Ruggeri further studied the geometric scale
effects on the value of B based on experiments of different scale venturis.
For different venturi scales, for a constant scaled cavity length (Δx/D = con-
stant), the flow conditions are assumed to be geometrically similar, and therefore
cavity thickness and cavity length will vary with scale factor,
δVS DS ΔxS DS
¼ and ¼ (14)
δVf Df Δxf Df
Thus, combining Eqs. 14 and 13 and based on the experimental Freon-114 data
from the 0.7- to 1.0 scale venturis, the following form is obtained:
0:85 0:5 0:15
V0 αf DS
B s ¼ Bf (15)
V 0, f αs Df
Finally, a more general semiempirical B-factor model, which is relative to a

reference value and can predict new value of B with changes in liquid, liquid
temperature, flow velocity, cavity length, and venturi scale, is derived by combining
Eqs. 13 and 15,
n h
V0 αref im D 1n Δx=D p
Bpred ¼ Bref (16)
V 0, ref α Dref ðΔx=DÞref
The exponents are recalibrated based on liquid hydrogen and Freon 114 data, and
resulting values are m = 1.0, n = 0.8, and p = 0.3.
For pumps, under conditions of flow similarity, the inlet velocity of pump is
proportional to pump rotative speed N. Moore and Ruggeri further assumed that Δx
remains constant for various liquids, liquid temperatures, and pump rotative speeds
when the flow coefficient φ and ψ/ψ NC don’t change. Then Eq. 16 becomes
(Rzlggeri and Moore 1969)
hα im N n
ref
Bpred ¼ Bref (17)
α N ref
MTWO
Although the method used by Gelder et al. has been successful in predicting the
B factor, it is physically inadequate because it neglected the convective heat transfer
involved in the cavitation process. Hord (1973a) developed a more general convec-
tion model and showed that the simple thermal conduction model used by Gelder is a
special case of it (Hord 1973a).
The convective heat transfer process can be described by the following equation,
Nux ¼ Co Rem1 Prn1 (18)
And the characteristic thermal boundary layer thickness was assumed as:
δl ¼ C1 Rem2 Prn2 (19)
From the cavity shape data in the study of Hord, the cavities’ shape can be
described by the form δv = C2xp.
Then the B factor can be written as
2 3 2 3
ðx ðx ðx ðx
B ¼ V V =V l ¼ δv dx= δl dx ¼ 4C2 xp dx5=4C1 Prn2 xRex m2 dx5
0 0 0 0 (20)
¼ C3 xp1 Rem2
x Pr
n2
¼ C3 αn2 V m2
0 x
p1þm2
ðρ=μÞm2n2
It can also be expressed based on a reference value as

α E1 V E2 x E3 v E4

ref 0 0 ref
B ¼ Bref (21)
α V 0, ref xref v0
Hord pointed out that Eq. 19 only contains the most common correlating param-
eters, and additional physical variables can be involved based on different Nusselt
relationships or cavity shape formations. After applying dimensional analysis of the
cavitation problem, Hord included size, surface tension, and two-phase sonic veloc-
ity into the B-factor expression and derived a new formation,
α E1 MTWO E2 x E3 v E4 σ E5 D E6

ref 0 ref ref ref
B ¼ Bref (22)
α MTWOref xref v0 σ D
The MTWO parameter is the ratio of V0/Vl where Vl is proportional to the two-phase
liquid-vapor sonic velocity across the cavity interface,
( )1=2
V1 1 þ BðρL =ρv Þðal =av Þ2
MTWO ¼ (23)
al 1 þ Bðρv =ρl Þ
where al and av are the speed sound of liquid and vapor, respectively.
The Entrainment Method

In order to improve the correlations, Holl et al. proposed another scale law called
“entrainment method.” (Holl et al. 1975) The cavity is assumed stable. The vapor is
continuously supplied from the cavity interface and must be balanced by the rate of
entrainment of volume of vapor away from the cavity in the wake. Vaporization of
vapor extracts heat from the adjacent liquid and a thermal boundary develops.
Therefore there exists a heat balance expressed as
m_ v hfg ¼ hAW ðT 1 T c Þ (24)
where Tc is the cavity temperature, T1 is the liquid temperature outside the thermal
boundary, h is the heat transfer coefficient, AW is the area of the cavity interface, and
hfg is the latent heat.
m_ v is the mass flow rate of vapor in the cavity expressed as
m_ v ¼ ρv D2 V 1 CQ (25)
where CQ is the flow coefficient and D is the diameter.

Substitute Eq. 25 to Eq. 24, then,
ρv D2 V 1 CQ hfg
ΔT ¼ ðT 1 T c Þ ¼ (26)
hAW
In terms of dimensionless coefficients, we obtain
CQ Pe ρv hfg
ΔT ¼ (27)
CA Nu ρl cP
where the flow coefficient CQ, area coefficient CA, and Nusselt number Nu were
determined from the experimental data, Pe is the Peclet number, Pr is the Prandtl
number.
Then the corresponding B factor proposed above can be expressed as

CQ Pe ρv λ
1
B ¼ ΔT=ΔT ¼ (28)
CA Nu ρl cP ΔT
Determination of CQ
From Eq. 25 the flow coefficient is defined as CQ ¼ m_ v =ρv D2 V 1 . The entrainment
theory is based on the concept that for a given set of flow conditions, the volume
flow rate of vapor required to sustain a vaporous cavity is equal to the volume flow
rate of the gas required to maintain a ventilated cavity during which
non-condensable gas was injected through a slit or hole situated slightly downstream
the cavity. Thus, the flow coefficient is usually determined and correlated from
ventilated cavitation experimental data.
c
L
CQ ¼ C1 Re Fr a b
(29)
D
Determination of CA
The area coefficient was determined from experimental photographs of both natural
and ventilated cavities, and an empirical correlation equation is obtained as below
d
L
CA ¼ C2 (30)
D
Determination of Nu
The Nusselt number was determined from Eq. 27 by using the measured values of
ΔT and the empirical relation for CQ and CA. The Nusselt number was assumed to
have the form
h
L
Nu ¼ C3 Re Pr Fr
l f
(31)
D
Fruman et al. Approach

Fruman derived a model to estimate the temperature difference between the cavity
and reference fluid from considering that the liquid flow on the cavity is analogous to
that turbulent flow on a plate (Fruman et al. 1991). On a smooth plate, the equation
below describes the relationship between the wall temperature and the heat flux
Q transferred by the fluid with Prandtl number Pr 1 (Brun et al. 1970).

Q ub
T p T ref s
¼ 1 ð1 PrÞ (32a)
1=2ρcp U ref Cf s
U ref
where Cf is the local friction coefficient,

. and ub is the velocity at the edge of the
0:1
2:1 Re
viscous boundary layer and ub =Uref ¼ x (Brun et al. 1970).
The heat flux Q can be expressed in terms of flow coefficient CQ as
Q ¼ ρv LVg ¼ ρv LCQ U ref (32b)
The temperature difference for a rough plate (Eckert and Drake 1972) is calcu-
lated from the one for a smooth plate by

Cf
T p T ref r
¼ η s
T p T ref s
(33)
Cf r
where η is a multiplying factor less or equal to 1 for the Prandtl’s number and the
roughness. Fruman assumed η = 1 in his model.
Thus

ρv LCQ 2:1
T p T ref r
¼ 1 0:1 ð1 PrÞ (34)
1=2ρcp Cf r
Rex
At very high Reynolds numbers, (Cf)r is expressed by
1 x1=7
Cf ¼ 0:00695 (35)
2 r e
where e is the equivalent roughness, which is calibrated from Hord’s data assuming
CQ = 5.2 1013 and expressed as e ¼ 2:2Re0:5x .
Thermal Aspects From Bubble Dynamics
Brennen (1995, 1973) illustrated the nature of thermal effect problem from the single
spherical bubble dynamics. The Rayleigh-Plesset equation can be rewritten in the
following general form (neglecting the effect of non-condensable gas, viscous, and
surface tension):

pv ðT 1 Þ p1 ðtÞ pv ðT B Þ pv ðT 1 Þ d2 R 3 dR 2
þ ¼R 2 þ
ρl ρl dt 2 dt (36)
ð 1Þ ð 2Þ ð 3Þ ð4Þ
The driving force for the bubble growth in the above equation is divided into two
parts: the instantaneous tension term (1) determined by the surrounding conditions
far from the bubble interface and the thermal term (2). It’s reasonable to derive the
thermal term into an approximation form:
pv ðT B Þ pv ðT 1 Þ 1 dpv
¼ ðT B T 1 Þ (37)
ρl ρl dT
Thus the magnitude of the difference between TB (temperature within the bubble)
and T1 determines whether the thermal term can have a major effect on the bubble
dynamics in Eq. 1.
There are two approaches to estimate the temperature difference depending on
two different heat transfer mechanisms between liquid and vapor, respectively.
Conductive Approach
One approach assumes that the heat transfer at the bubble wall is supposed to be
conductive dominated. The determination of (TB T1) need the solution of the heat
diffusion equation in spherical coordinates and an energy balance equation for the
bubble. However, the resulted equations are complex and difficult for application.
Plesset and Zwick (Plesset and Prosperetti 1977) assumed that there is a small
thermal boundary layer thickness around the bubble; thus, it becomes convenient to
pffiffiffiffiffiffiffi
use the magnitude of this thickness like αL t to estimate the conductive heat flux to
the interface per unit surface area by means of FOURIER’s law, thus the energy
equation is simplified to

T1 TB d 4 3
4πR kl pffiffiffiffiffiffiffi ¼
2
πR ρv L (38)
αL t dt 3
Then the temperature difference is obtained:

pffi
t ρ L dR
T 1 T B ¼ pffiffiffiffi v (39)
αl ρl cpl dt
Introducing this temperature difference into Eq. 38, the Rayleigh-Plesset Eq. 36
becomes

pv ðT 1 Þ p1 ðtÞ X 1 dR d2 R 3 dR 2
þ pffi ¼R 2 þ (40)
ρL t dt dt 2 dt
A close observation of the above equation yields the fact that the impact of
thermal effect on bubble dynamics depends on Σ which is introduced by Brennen
(1973) and defined as
X ρv L dp
¼ pffiffiffiffi v (41)
ρl 2 cpl αl dT
The term dpv/dT is approximated by using the Clausius-Clapeyron relation in

Brennen’s expression and then Σ is expressed as
X 2
ρv ðLÞ2
¼ pffiffiffiffi (42a)
ρl cpl T 1 αl
This parameter’s unit is m/s3/2, and it is crucially important to determine whether

the cavitation process is thermally controlled or not.
Comparison between the definition of B factor and Σ and the relation between
them is obtained
!
X ðT 1 T B Þ2 1
¼ 2 pffiffiffiffi
(42b)
2
T1 B αl
or
!
ðT 1 T B Þ 1
B¼ P1=2 1=4 (43)
T1 αl
in order to analyze the similarity between different cavitation processes. Franc et al.
(2003) recast Eq. 40 into a nondimensional form as following:
sffiffiffiffiffiffiffiffi
X C _ pffi Cp þ σ
€ þ 3 R_ 2 þ
RR R t¼ (44)
2 3 2
U1
All the derivatives are with respect to x ¼ x=C, x is the distance traversed by a
bubble in the flow field in time t with the formation of U1t, and C is the character-
istic length.
Kato (1984) also proposed a new nondimensional thermodynamic parameter
pffiffiffiffiffiffiffiffiffiffiffi
ρl =ρv Σ for attached cavity by rearranging the Hord’s experimental results and
P pffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffi
establish a good correlation between ρl =ρv C=U 3 and the nondimensional
pressure drop inside the cavity.
These nondimensional quantities introduced above can be utilized to estimate
whether two cavitation processes are thermally similar.
Convective Approach
The other approach assumes the heat transfer process at the liquid vapor interface to
be of convective domination (Franc and Pellone 2007; Franc et al. 2010a). The heat
balance for the bubble is expressed as

d 4 3
πR ρv L ¼ 4πR2 hðT 1 T B Þ (45)
dt 3
Then
ρv L _
T1 TB ¼ R (46)
h
By introducing Eq. 46 into Eq. 45 as done in “conductive approach,” Eq. 36 can
be expressed in a nondimensional form as
€ þ 3 R_ 2 þ τ R_ ¼ Cp þ σ
RR (47)
2 τT 2
where
pffiffiffiffi
C Nu αl
τ¼ , τT ¼
U Σ
C is the reference length and U is reference velocity, and τ is an approximation of
transit time of a bubble passing through the cavitation region.
The Nusselt number can be further correlated with experimental data in terms of
the Reynolds and Prandtl numbers as done in the B-factor theory introduced above.
The degree of thermodynamic effects on cavitation is obtained by calculating the
nondimensional ratio of the characteristic transit time τ and thermal time τT.
Experimental Studies
Appearance of Cavitation with Thermal Effects
Sarosdy and Acosta (1961) compared the cavitation phenomenon in water and in
Freon 113 using a cavitating disk. They found that the cavity in Freon was “frothy”
as compared to be “glassy” in cold water. The “frothy” nature was later reported in
literatures in liquid hydrogen, liquid nitrogen (Hord 1973a), and hot water (Cervone
et al. 2006). The tested geometries included ventures, ogives, and hydrofoils. A
typical comparison between the appearance of cavitation in cold water and in liquid
nitrogen is shown in Fig. 2 (Watanabe et al. 2010). Even for the same fluid, with the
increase of temperature to the critical point, cavitation will show strong thermal
effects. Appearance of cavitation in cold and hot water in Cervone’s experiment are
Fig. 2 Visualization of a cavitating inducer in (a) water (306 K) and (b) liquid nitrogen (79 K) for
the same cavitation number, the same rotational speed, and the same flow rate (From Yoshiki
Yoshida et al. Watanabe et al. 2010)
Fig. 3 Cavitation appearance in (a) cold water T = 25 C, α = 4 , σ = 1.25 and (b) hot water
T = 70 C, α = 8 , σ = 2.0, (From Cervone 2006)
shown in Fig. 3. In case of hot water (case a), cavity bubbles are smaller than those of
cold water (case b) and tend to coalesce more easily. Except for the common
appearance, unique characters of cavitation with thermal effects vary for fluid to
fluid. For example, in Gelder et al. (1966), the cavitating flow in nitrogen (196 C
~193 C) along the venturi wall was described as less violent and noisier than that
in Freon 114 (18 C ~27 C) and has a less stable leading edge and is less
uniformly developed circumferentially, although cavitation in Freon 114 also
shows thermal effects at temperatures like 15 C.
Thermodynamic State of Sheet Cavitation
Hord et al. (Hord 1972, 1973a, b) measured series of temperature and pressure data
under different cavitation number of cryogenic flows over venture nozzle, hydrofoil,
and ogive bodies. The typical type of cavity was sheet cavitation. A temperature
depression about 2–3 K was detected, and the pressure within the cavity wasn’t a
constant value. The state of thermodynamic equilibrium within the developed
cavities was determined from the direct measurements of pressures and temperatures
within the vaporous cavities. It was found that cavities developed on the hydrofoil
and ogives were in stable thermodynamic equilibrium while metastable vapor
existed in the central and aft regions of hydrogen cavities of hydrogen cavities in
the venturi. The stable thermodynamic equilibrium prevailed near the leading edge
of the cavities for all of the three geometries, which were also concluded in the work
of Gelder et al. (1966) and Moore and Ruggeri (1962). The thermodynamic equi-
librium state in sheet cavitation set the basis of the B-factor theories discussed in
section “Parameters Developed to Estimate Thermal Effects on Cavitation.”
Thermal Effects on Dynamics of Cavitation over Simple Geometries
Yutaka et al. (2009a, b) observed the cavitation patterns on a plano-convex hydrofoil

in a high-speed cryogenic cavitation tunnel using liquid nitrogen as the working
fluid. The frequency of periodical shedding of cloud cavitation was much smaller
than that estimated by an empirical equation for the vortex shedding from a bluff
body. The shedding of cloud cavitation even occurs when the cavity fully covers the
hydrofoil surface while in conventional cloud cavitation such as in water, the length
of cavity is shorter than hydrofoil chord length.
Cervone et al. (2006) conducted thermal cavitation experiments on a NACA 0015
hydrofoil at various incidence angles, cavitation numbers, and free-stream temper-
atures. For constant incident angle and free-stream cavitation number, the pressure
recovery at the cavity closure happens more upstream in hot water than in cold water.
The slight recovery of pressure along the hydrofoil wall within the cavity could also
be recognized from Hord’s experimental data (Hord 1973a). Cervone divided the
cavitation regime into super cavitation, bubble+cloud cavitation, and bubble cavita-
tion according to the variations of cavity length. For higher free-stream temperatures,
the bubble+cloud cavitation zone tends to spread over a wider range of cavitation
numbers and to begin at higher values of cavitation number.
Gustavsson et al. (2008) and Kelly et al. (2011) carried out lots of experiments on
a NACA 0015 hydrofoil with a perfluorinated ketone that exhibits a strong thermo-
dynamic effect at ambient conditions. The effects of velocity on cavitation number
were studied, and the pressure and temperature were measured along with high-
speed imaging. The perfluorinated ketone was chosen because the value is identical
to liquid hydrogen when heated to 70 C; thus it will avoid the cryogenic testing
difficulties and facilitates the investigation of thermal effects on cavitation at ambient
conditions. Raising the temperature with the flow speed and cavitation number kept
Fig. 4 Unsteady cloud cavitation forms of liquid nitrogen
constant, the cavity becomes shorter with higher pressure and consists of finner
bubble structure. The shedding of the vapor clouds becomes more frequent. A
downstream shift of peak Cp is present due to compressibility effects. Actually, in
1979, Hord (1973a) also reported the shock wave phenomenon due to the compress-
ibility effects in the experiment of cavitating hydrogen hydrofoil.
Niiyama et al. (2012) conducted experiments on the NACA16-012 hydrofoil with
an angle of attack 8 . The observed cavity consisted of numerous bubbles much
smaller than those in water observed by Franc et al. (1985). On the other hand, with
the reduction of cavitation number, the length of cavity became longer but it didn’t
strongly oscillate, while in Franc’s experiment, the cavity oscillated strongly when it
spread over 50–70% of the chord length of hydrofoil.
Giorgi et al. (2010) studied the cavitation phenomenon in internal flow in the
presence of thermal effects. For LN2, the frequency spectrum of the downstream
pressure signals is different from that of water with higher frequency content in the
range of low frequencies.
Fig. 5 Swing phenomenon of the cavitation zone (Zhao et al.)
Dular (Petkovšek and Dular 2013) utilized the infrared thermography in studying
the cavitation thermal effects in water; a warming effect due to the condensation was
detected.
In the cryogenic tunnel in Institute of Refrigeration and Cryogenics in Zhejiang
University (Zhao et al.), the visualization experiment of cavitating flow of cryogenic
fluid was conducted in venturi tube. The unsteady characteristics of cloud cavitation were
captured using a high-speed camera as shown in Fig. 4. The cavity appeared to be
“foggy” and oscillated periodically. At t = 0 ms, the cavity was short. At t = 1.0 ms, the
cavity length became longer and the phase change became more evident. As the length of
cavity increased to a critical value at t = 3 ms, the tail of cavity started to condensate, and
it broke into small cavitation clouds and finally vanished. Then the cavity was shortened.
The front part of the cavity seemed to be more stable than the tail. In such a cycle, there
existed two reentrant jet flows to make the cavity vary up and down as shown in Fig. 5.
Thermal Effects on Cavitation in Rocket Inducers
Different from the flow over a simple geometry (hydrofoil, nozzle, etc.), the flow
over a rocket inducer is more complicated. For example, velocities are not the same
along the blade in the radial direction which can be approximatively described as Vr
= ωr, where ω is the angular velocity and r is the radius of the corresponding
location on the blade. Therefore the velocity on the blade increases proportionally to
radius r. The thermodynamic effects at different locations can then be estimated from
P qffiffiffiffiffiffiffiffiffiffiffi3
the parameter D=V r . If the characteristic length D is chosen as the local blade
chord length at the same radial coordinate, then it will also increase proportionally to
P qffiffiffiffiffiffiffiffiffiffiffi3
radius r. Thus, the value of D=V r will be proportional to 1/Vr. The intensity of
thermal effects on cavitation is different even along the same blade.
Franc et al. (2010a, b; Franc and Pellone 2007) conducted experiments of a four-
bladed inducer in R114 refrigerant at different temperature and in cold water. A
challenging experiment of direct measurement of temperature on a rotating inducer
was conducted in (Franc et al. 2010b). The onset of rotating cavitation instabilities
had poor relationships with the leading edge cavitation. The temperature depression
within leading edge cavities increased nearly linearly with cavity length. The
alternate blade cavitation and supersynchronous rotating cavitation are delayed by
the thermodynamic effect.
Yoshida et al. (2009, 2011) studied the influence of thermodynamic effect on
rotating cavitation. The focus was on the tip cavitation. The experiments were
conducted on a three-bladed inducer at different temperature (74 K, 78 K, and
83 K). As the cavitation number decreases, supersynchronous rotating cavitation,
synchronous rotating cavitation, and subsynchronous rotating cavitation appeared
one after another. In supersynchronous rotating cavitation region, the propagating
speed ratio wasn’t affected by temperature. Synchronous rotating cavitation occurred
at a critical cavity length of the order of 80% of the blade spacing but at a lower
cavitation number with the increase of temperature due to the lag of cavity growth by
the thermodynamic effects. And the unevenness of cavity length was greatly affected
by the thermal effects. In subsynchronous rotating cavitation region, thermal effects
suppressed the cavity fluctuations; however, it didn’t affect the propagating speed.
Kikuta et al. (2009) investigated the influence of rotational speed on thermody-
namic effect in liquid nitrogen cavitating flow. At the same cavitation number, with
increased rotational speed, both of the cavity length and the temperature depression
became larger.
Numerical Studies
Reduction in design cycles and the desire to reduce testing have driven the demands
for numerical analysis. Various computational methods have been developed to
investigate the cavitating flow and can be generally divided into two categories:
single-phase modeling with cavitation interface tracking and multiphase modeling
with an embedded cavitation interface. In the former method, only the liquid phase is
modeled with the assumption that the vapor pressure in the cavity region is constant,
corresponding to the local temperature. Although it requires a considerable amount
of preliminary knowledge, many successful applications have been reported
(Deshpande et al. 1997; Stutz and Reboud 1997a). The latter method models both
phases via different approaches according to the properties and thermodynamic
assumptions. For example, the full nonequilibrium two-fluid models consider sep-
arate conservation equations for each phase (Saurel and Lemetayer 2001) and can
describe the rich physical details involved in the cavitating process, however require
parameters which are hard to be obtained. These models have been tested on
cavitation in metastable liquids, liquid helium, and other high-speed applications
(Zein et al. 2010; Ishimoto and Kamijo 2004; Petitpas et al. 2009). By further
assuming the pressure and thermal equilibrium between phases, a mixture model
is proposed, which regards the two-phase mixture as a single fluid. In the mixture
model, since cavitation primarily occurs in the low-pressure regions, where the
velocity is relatively high, the slip velocity between the phases is commonly not
considered in most numerical simulations, resulting in the so-called the
homogeneous mixture condition (Ahuja et al. 2001; Ji et al. 2013, 2014). In contrast,
the nonhomogeneous mixture model (also called the drift-flow model) considers the
slip velocity. Examples of this approach are provided in the works by Rhee et al.
(2005), which uses an algebraic relation to calculate the slip velocity in the modeling
of propeller cavitation.
The key for implementation of homogenous mixture model is to calculate the
variable density field in the cavitation process. One approach uses an equation of
state (EOS) to relate the pressure with the thermodynamic properties; various
relations have been proposed (Saurel et al. 1999; Barre et al. 2009; Goncalvès and
Patella 2010; Clerc 2000; Delgosha et al. 2003; Goncalvès et al. 2010; Sinibaldi et al.
2006). Another popular method introduces a gas volume/mass fraction transport
equation, whose source terms are the condensation and evaporation rate in the liquid-
vapor conversion (Singhal et al. 2002; Merkle et al. 1998; Kunz et al. 2000; Sauer
and Schnerr 2000a; Senocak and Shyy 2004a; Zwart et al. 2004). One apparent
advantage of this model comes from the potential for modeling the impact of inertial
forces on cavities like elongation and the drift of cavity bubbles, such as is demon-
strated in the works by Singhal et al. (2002), Merkle et al. (1998), and Kunz et al.
(2000). The main challenge in using this method is the formation of the source term
and the tunable parameters involved for vaporization and condensation processes.
Recently, Hosangadi and Ahuja (2005) employed the Merkle transport-based model
to simulate cryogenic cavitation; the optimum empirical parameters of the cavitation
model were found to be sufficiently lower than those for non-cryogenic fluids
(Ji et al. 2014). They showed that the cavity is sustained by mass directly being
convected into the cavity with the liquid vaporization as it travels across the vapor-
liquid interface and contains far less vapor than cavity in water. Later, they analyzed
the thermal effects in cavitating liquid hydrogen inducers (Hosangadi et al. 2007). A
more gradual breakdown of the head in liquid hydrogen and improved overall
performance were found if compared with water. Utturkar (Utturkar et al. 2005)
utilized the mushy interfacial dynamic-based cavitation model to study the charac-
teristics of cryogenic cavity over ogive and hydrofoil. Zhang et al. (2008a, b; Cao
et al. 2009) validated the full cavitation model in cryogenic cases with reasonable
predictions of temperature and pressure distribution over different test sections. And
later they conducted better predictions by recalculating the empirical bubble radium
in the full cavitation model taking the effect into account of local pressure (Zhang
et al. 2013). Huang et al. (2014a) involved a thermal term into the Kubota model and
recalibrated the empirical parameters to obtain a well prediction of temperature and
pressure depressions in the experiments by Hord. By further assuming that the mass
transfer is proportional to the velocity divergence and that the liquid is at its
saturation state, a thermodynamically well-posed four-equation model including a
void transport equation whose source term is derived from a modified barotropic
model has been tested in thermosensitive Freon R-114 (Goncalvès 2014).
It is recognized that traditional transport equation-based cavitation models are
still able to predict cavitation dynamics in cryogenic fluids after a robust calibration
or modification.
Establish an Effective Numerical Framework Based on the Mixture

Model
The homogeneous mixture approach is used to model the two-phase cavitating flow.
The phases are assumed to be well mixed moving at the same velocity and to be in
local thermodynamic equilibrium. They share the same temperature and the same
pressure. The set of governing equations for cryogenic cavitation based on the
homogenous fluid modeling comprise the conservative form of the unsteady
Reynolds-averaged Navier-Stokes equations, the enthalpy equation, the turbulence
closure model, and a transport equation for the vapor (or liquid) volume fraction.
Conservation of Mass, Momentum, and Energy

The mass continuity, momentum, and energy equations are given below (Theory
Guide 2012):

@ρm @ ρm uj
þ ¼0 (48)
@t @xj

@ ðρm ui Þ @ ρm ui uj @p @ @ui @uj 2 @uk
þ ¼ þ ðμm þ μt Þ þ δij (49)
@t @xj @xi @xj @xj @xi 3 @xk

@ @ @ @T
½ρm hm þ ρm hm uj ¼ ðk þ k t Þ hfg mr (50)
@t @xj @xj @xj
ρm ¼ ρv αv þ ρl αl , μm ¼ μv αv þ μl αl , ρm hm ¼ ρv αv hv þ ρl αl hl
In the above equations, x and the subscripts i, j, and k denote the coordinate axes.
ρ is the density, α represents the volume fraction, u represents the velocity vector,
P is the pressure, T is the temperature, μ is the viscosity, h is the sensible enthalpy
expressed as h = cp T, cp is the specific heat, k is thermal conductivity, m_ ris the net
mass source term defined in the subsection “Cavitation Model,” and hfg is the latent
heat. The subscripts v and l denote the vapor phase and liquid phase, respectively, the
subscript t denotes turbulent flow, and the subscript m denotes the mixture.
Cavitation Model
For cavitation modeling, a vapor mass transport equation is solved (Cavitation
models 2012):
@
!
ðρv αv Þ þ ∇ ρv αv u ¼ m_ r (51)
@t
or
@
!
ðf v ρm Þ þ ∇ f v ρm u ¼ m_ r (52)
@t
Models developed to determine the net mass source term m_ r are called cavitation
models.
Schnerr-Sauer Model (SSM)

The net mass source term m_ r can be expressed in the general form (Cavitation models
2012):
ρv ρl dαv
m_ r ¼ (53)
ρm dt
Schnerr and Sauer (2000b) used the following expression to relate the vapor
volume fraction to the number of bubbles per unit volume of liquid:
4
n πℜ3B
αv ¼ 3 (54)
4
1 þ n πℜ3B
3
where ℜB is the bubble radius and n is the bubble number per unit volume of liquid.
Combining Eqs. 53 and 54, m_ r is expressed as
3αv ð1 αv Þ ρv ρl dℜB
m_ r ¼ (55)
ℜB ρm dt
Assuming that the bubble is spherical and the bubble growth is an inertial-
controlled process, the simplified Rayleigh-Plesset equation is used to account for
time evolution rate of the bubble radius as follows (Plesset and Prosperetti 1977):
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dℜB 2 absfPv ðT Þ Pg
¼ signðPv PÞ (56)
dt 3 ρl
where T is the local temperature.

Substituting Eq. 56 into Eq. 55, the net mass source term has the following form:

3αv ð1 αv Þ ρv ρl 2 Pv ðT Þ P 1=2
P < Pv ð T Þ m_ re ¼ (57)
ℜB ρm 3 ρl

3αv ð1 αv Þ ρv ρl 2 P Pv ðT Þ 1=2
P P v ðT Þ m_ rc ¼ (58)
ℜB ρm 3 ρl
where Pv(T) is the local saturation vapor pressure.

The bubble radius ℜB is given by
13
αv 3 1
ℜB ¼ (59)
1 αv 4π n
The mass transfer rate in Eqs. 57 and 58 is proportional to αv(1αv) and

approaches zero when αv = 0 and αv = 1. The only parameter which must be
determined in this model is the number of vapor bubbles per volume of liquid (n).
Full Cavitation Model (FCM)

Singhal et al. (2002) have used the Rayleigh-Plesset (RP) equation (Franc and
Michel 2005) for depicting bubble dynamics in order to obtain the following
_
expression for R:

ρl ρv 2 Pb P 1=2
m_ r ¼ ðn4π Þ1=3 ð3αv Þ2=3 (60)
ρ 3 ρl
where n is bubble number density in the cavitation region. Pb is the pressure inside
the bubble, and it is approximated to be the saturation vapor pressure Psat in Singhal
et al. (2002).
There is only one unknown parameter, n, on the right hand side of the equation
above, which can be transformed to the unknown single bubble radius ℜb using
equation:
4
αv ¼ n πℜ3b (61)
3
To determine ℜb, the FCM regards it as the maximum possible size as deter-
mined from the balance between aerodynamic drag and surface tension and obtains
ℜb ¼ 0:061Weσ
2ρ u2
. Here, We is the Weber number defined as W e ¼ ρl u2rel Lch =σ, σ is the
l rel
surface tension, and urel is the relative velocity between the two phases. Finally, the
net mass source term has the following form:
pffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k 2ð pv pÞ
m_ re ¼ Ce ρl ρv 1 fv fg p pv (62)
σ 3ρl
pffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k 2ð pv pÞ
m_ rc ¼ Cc ρρ fv p > pv (63)
σ l l 3ρl
where Cc and Ce are condensate and evaporate parameters and need to be determined
empirically.
Dynamic Cavitation Model (DCM)

It may be noticed that the expression of ℜb in FCM is independent of the local
pressure, which is not a good assumption for unsteady cavitation. Therefore, there is
still room to improve the FCM, regardless of its success in modeling the quasi-steady
sheet cavitation (Singhal et al. 2002; Zhang et al. 2008a), by considering the effects
of pressure on the bubble radius.
In general, the liquid and vapor may not be in equilibrium locally, and the bubble
will essentially experience a nonequilibrium thermodynamic path and metastable
state during the phase-change process (Zein et al. 2010; Gavrilyuk and Saurel 2002).
However, as indicated in section “Thermodynamic State of Sheet Cavitation,” the
local thermodynamic equilibrium during the phase-change process is a valid
assumption. At thermodynamic equilibrium, the temperature and chemical potentials
between the liquid phase and vapor phase are equal. Integrating the Gibbs-Duhem
equation (Carey 1992) from the initial pressure P = Psat to an arbitrary pressure P for
the liquid and vapor phases, respectively, and using the ideal gas equation, a general
relation between the pressure inside and outside of the bubble is obtained. These two
pressures should satisfy the Young-Laplace equation. Combining the two equations
finally leads to the following expressions relating the local pressure and the bubble
radius. Details regarding the derivation can be found elsewhere (Zhang et al. 2013;
Carey 1992).
For the evaporation process:
2σ
ℜb , e ¼ (64)
Psat ðT l Þexp ½Pl Psat ðT l Þ=ðρl RTÞl Pl
and for the condensation process:
2σ
ℜb , e ¼ (65)
ρl RTv ln½Pv =Psat ðT v Þ Pv þ Psat ðT v Þ
where the subscript sat denotes the saturation state. Pl and Pv in above equations
represent the local pressure P during the evaporation or condensation process,
while Tl and Tv represent the local temperature Tref. Substituting Eqs. 60, 64, and
65 into 60 and considering the fact that the per unit volume phase-change rates
should be proportional to the volume fractions of the donor phase, leads to the
expressions:
When Psat > P (evaporation process)

Psat ðT ref Þexpf½P Psat ðT ref Þ=ðρl RTref Þg P 2 Psat P 1=2
m_ re ¼ Ce ρv 1 f v f gas
σ 3 ρl
(67)
When Psat P (condensation process)

ρl RTref ln½P=Psat ðT ref Þ þ Psat ðT ref Þ P 2 P Psat 1=2
m_ rc ¼ Cc ρl fv (68)
σ 3 ρl
In the above equations, the relation αv = fρ/ρv is used to calculate the vapor
volume fraction. Ce and Cc are empirical constants, while fgas is the non-condensable
gas mass fraction that is set to 108 simulations to avoid its influence. No assumption
has been made that the bubble pressure is equal to the saturation pressure in the
derivation of the bubble radius expression. However, those two pressures are
practically very close from the thermodynamic viewpoint, which means that
Eqs. 67 and 68 can be further simplified. Notice also that during the phase change
the bubble radius is a function of the local pressure and the saturation pressure. For
isothermal cavitation in water, both the saturation pressure and the surface tension
are constant; thus the bubble radius only varies with the local pressure. For the
evaporation process, the evaporation rate is larger if the local pressure P is lower; for
the condensation process, the condensation rate is larger if the local pressure P is
higher.
The effects of turbulence on cavitation are considered in the same fashion as in
the FCM (Singhal et al. 2002) by simply raising the phase-change threshold pressure
value as:
Pcav ¼ ½Psat ðT ref Þ þ Pt =2 (69)
where Pt = 0.39ρκ is the turbulent pressure. Thus, Eqs. 67 and 68 can be rewritten as
When Pcav > P (evaporation process)

Pcav ðT ref Þexpf½P Pcav =ðρl RTref Þg P 2 Pcav P 1=2
m_ re ¼ Ce ρv 1 f v f gas
σ 3 ρl
(70)
When Pcav P (condensation process)

ρl RTref lnðP=Pcav Þ þ Pcav P 2 P Pcav 1=2
m_ rc ¼ Cc ρl fv (71)
σ 3 ρl
Turbulence Model
For Quasi-Steady Cavitating Flow

For quasi-steady cavitating flow, the realizable turbulence model is utilized (Theory
Guide 2012):

@ ð ρm κ Þ @ ρm uj κ @ μt @κ
þ ¼ μþ þ G k ρm e (72)
@t @xj @xj σ κ @xj

@ ð ρm e Þ @ ρm uj e @ μt @e e2
þ ¼ μþ þ ρm C1 Se C2 ρm pffiffiffiffiffi (73)
@t @xj @xj σ e @xj κ þ νe

η κ pffiffiffiffiffiffiffiffiffiffiffiffi @uj
C1 ¼ max 0:43, , η ¼ S , S ¼ 2Sij Sij , Gκ ¼ ρu0i u0j
ηþ5 e @ui
In these equations, Gκ represents the generation of turbulence kinetic energy due to

the mean velocity gradients. σ κ and σ e are the turbulent Prandtl numbers for k and e.
k is the turbulent kinetic energy and e is turbulent dissipation rate. The eddy viscosity is
computed from
κ2
μt ¼ ρm Cμ (74)
e
The default constants for this turbulence model are given as follows:
C2 ¼ 1:9, σ κ ¼ 1:0, σ e ¼ 1:2, Prt ¼ 0:85 (75)
For Unsteady Cavitating Flow

For unsteady cavitating flow, the large eddy simulation model is used.
The LES equations are given below (Theory Guide 2012):

@ ðρm Þ @ ρm uj
þ ¼0 (76)
@t @xj

@ ð ρm ui Þ @ ρm ui uj @p @ @τij
þ ¼ þ σ ij (77)
@t @xj @xi @xj @xj

@ui @uj 2 @uk
σ ij ¼ μm þ δij
@xj @xi 3 @xk
where the over-bar denotes the filtered quantities. τij is the sub-grid scale (SGS)
stress and defined as:

τij ¼ ρm ui uj ui uj (78)
It is assumed that the SGS stress is proportional to the modulus of the strain rate
tensor, Sij , of the filtered large-scale flow:
1
τij τkk δij ¼ 2μt Sij (79)
3
where μt is the sub-grid scale turbulent viscosity. In the LES Wall-Adapting Local
Eddy-Viscosity (WALE) model (Nicoud and Ducros 1999), μt and Sij are modeled by
3=2
Sdij Sdij
μt ¼ ρL2s 5=4 (80)
5=2
Sij Sij þ Sdij Sdij

1 @ui @uj
Sij ¼ þ (81)
2 @xj @xi
1 2 1 @ui
Sdij ¼ gij þ g2ji δij g2kk , gij ¼ , Ls ¼ minðκd, Cw ΔÞ (82)
2 3 @xj
where Ls is the mixing length for the SGS, κ is von Karman’s constant, d is the
distance to the closest wall, Δ represents the local grid scale based on the cell
volume, and the WALE constant Cw is 0.325 (Theory Guide 2012).
Pressure Correlation Equation and Update of Thermodynamic

Properties
Thermodynamic properties (such as saturation vapor pressure, densities, specific
heat, thermal conductivity, viscosity, etc.) are specified as the function of tempera-
ture from the data generated from Prop v7.0 (Lemmon et al. 2007).
The key step in the implicit pressure-based algorithm is to establish the pressure
correlation equation for the cavitating flows. To satisfy the continuity constraint and
ensure numerical stability, the pressure correlation equation is built based on the total
volume continuity (Li and Vasquez 2012):

1 @ρl αl 1 @ρv αv
_ þ
þ ∇ ðρl αl uÞ mg þ ∇ ðρv αv uÞ ðm_ Þ ¼ 0 (83)
ρl @t ρv @t
where m_ ¼ R_ e þ R_ c and αl + αv = 1.
Steady Flow
It is noted that for steady flow computations, the pressure-density coupling scheme
only affects the convergence path, and the final solution is independent of the choice
because of the nature of pressure correlation (Senocak and Shyy 2004a, b; Li and
Vasquez 2012). The conclusions are verified that the calculations for quasi-steady
cavitation based on the incompressible fluids for both liquid and gas phases obtained
the quite accordant results with the experimental data (Senocak and Shyy 2004a;
Zhang and Khoo 2013) and the compressible calculations (Zhang and Khoo 2014).
Therefore, compressibility of either liquid or gas is neglected. Equation 83 is easily
converted to

1 1
∇u¼ m_ (84)
ρl ρv
The set of the pressure-based continuity equation and momentum equation are
simultaneously solved in the following sequences (Theory Guide 2012):
1. The momentum equations are discretized using the finite volume method to
obtain the linear algebraic equations, Au = t ∇p, which is then used to obtain
the relation, u = AH/AD (1/AD) ∇p and u0 = (1/AD) ∇p0, where A is the
coefficient matrix of the algebraic equations for velocity vectors (u); AD = diag(A),
AH = t ANu, and AN = A diag(A); u* and p* are the estimation velocities and
pressure; u0 and p0 are the correction velocities and pressure, respectively.
2. Substituting p = p + p0 and u = u + u0, together with the relation u = AH/
AD (1/AD) ∇p, into Eq. 17, yields the implicit governing equation for the
correction pressure p0 , which can be easily solved. It is emphasized that the mass
transfer term m_ also contains p* and p0 .
3. Pressure and velocities are updated by p = p + p0 and u = u + u0.
Scalars (T, k, e, αv) are solved by Eqs. 50, 76, 77, and 51, respectively, and the
fluid properties including the mixture density are then updated.
Unsteady Flow
For unsteady flow computations, the speed of sound of vapor and liquid will
influence the final results. Thus, it is important to establish a pressure correlation
equation to account for the compressibility.
Integrating Eq. 83 over a control volume and assuming that
@ρl 0 0 @ρ
ρl ¼ ρl þ ρ0l ; ρv ¼ ρv þ ρ0v ; αl ¼ αl þ ρ0l ; αv ¼ αv þ ρ0v ; ρ0l ¼ p ; ρv ¼ v p0 ;
@p @p
en, , 0 ; V en ¼ V en, þ V en, 0 ; α0 0; α0 0;
l ¼ Vl
V en þ V en
l v v v l v
The discretized pressure correction equation can be obtained (Li and Vasquez
2012):
( )
1 X αl Vol @ρl 0 X 1 @ρl e, 0
αeq, ρl Aen V en,0 0
m Vol þ p þ F p
ρl e
q
Δt @p e
ρle, @p e l
( )
1 X e, X 1 @ρ
þ αq ρv Aen V en , 0 þ m0 Vol þ αv Vol @ρv p0 þ v
F e, 0
p
ρv e
q
Δt @p e
ρev, @p e v
( ) ( )
1 αl ρl Vol α0l ρ0l Vol0 X e, 1 αv ρv Vol α0v ρ0v Vol0 X e,
¼ þ Fl m0 Vol þ Fv þ m0 Vol
ρl Δt e
ρv Δt e
(85)
Here, the superscript * and 0 , respectively, represent old values and corrections.
Vol is the cell volume, Δt is the time step, A is the area at face “e,” and F is the phase
mass flux. It should be noted that the extra terms related to @ρ/@p are the com-
pressible effect in the pressure correction equation. @ ρl = @ p and @ ρv = @ p are the
speed of sound in liquid and vapor, respectively, which are given as the function of
saturation temperature and coupled into the codes of fluent before simulations.
In the coupled algorithm (Theory Guide 2012), it solves the pressure correction
equation and momentum equation simultaneously to get the updated pressure and
velocities: p = p + p0, u = u + u0. Then scalars T and αv are solved by Eqs. 50 and
51, respectively, with the updated fluid properties.
Validation and Recalibration of the Cavitation Model

The Hord group from NASA has performed subscale tests of cavitation in cryogenic
fluids (liquid nitrogen and liquid hydrogen) in a transparent plastic blowdown
tunnel. The hydrofoil and ogive were placed in the center of the tunnels, respectively.
And several pressure transducers and thermal couples were mounted along the wall
of them. The accuracy for pressure and temperature was 6,900 Pa and 0.2 K,
respectively. Details of the experiments can be found in Hord (1972, 1973a, b).
The experimental data are usually used as bench mark to examine the cavitation
modeling framework.
Full Cavitation Model

The full cavitation model (FCM) was used for quasi-steady cryogenic cavitation
modeling by Zhang et al. (2008a, b). The realizable turbulence model is used to
investigate the effects of turbulent mixing. In the near wall treatment, the standard
wall function and the enhanced wall treatment are compared. Both the standard wall
function and the enhanced wall treatment are applicable for modeling cavitating
turbulent flows in liquid hydrogen, and comparable results with Hord’s experimental
data are obtained. However, the enhanced wall treatment provides a better temper-
ature and pressure distribution than the standard wall function, because it assumes a
local equilibrium that exists between the production of kinetic energy and its
dissipation rate at the wall-adjacent cells. As illustrated in Fig. 6 in liquid hydrogen
cavitating flow over the hydrofoil 229C and ogive 390B, the variation of the
vaporization and condensation rate parameters has a great effect on the simulation
results. As these parameters become larger, the temperature and pressure depressions
also increase. The reason for this behavior is that although the evaporation and
condensation rate change proportionally, the change in the absolute magnitude of the
evaporation rate is greater than that of condensation rate. Also, the locations of the
lowest temperature and pressure move away from the leading edge when the
parameters decrease. Apparently, the simulated results with the default model
parameters (Ce = 0.02, Cc = 0.01) correlate better with the experimental data than
those with significantly modified model parameters.
As introduced in section “Parameters Developed to Estimate Thermal Effects on
Cavitation”, the basic assumptions in the B-factor theory is that liquid flows over the
cavity as if it were a solid body and no convection of liquid mass flows across the
vapor-liquid surface. Therefore, the primary factor governing the vaporization is the
viscous diffusion of energy across vapor-liquid interface, which sustains the contin-
uous vaporization of the liquid. In the simulated liquid nitrogen cavitating flow
(Zhang et al. 2008b), the particle traces of the fluid streamlines are plotted near the
20.8 4x105
Temperature (K)
Ce=0.01,Cc=0.005
20.4
Pressure (Pa)
3x105 Ce=0.02,Cc=0.01
Ce=0.04,Cc=0.02
20.0
Ce=0.01, Cc=0.005 Hord,1973
19.6 Ce=0.02, Cc=0.01 2x105
Ce=0.04, Cc=0.02
19.2 Hord, 1973
1x105
18.8
0.13 0.14 0.15 0.16 0.17 0.13 0.14 0.15 0.16 0.17
x (m) x (m)
Hydrofoil - 229C
22.0
4x105 Ce=0.01,Cc=0.005
Temperature (K)
21.6
Ce=0.02,Cc=0.01
Pressure (Pa)
21.2 Ce=0.04,Cc=0.02
3x105
20.8 Hord,1973
Ce=0.01,Cc=0.005
20.4 Ce=0.02,Cc=0.01 2x105
Ce=0.04,Cc=0.02
20.0
Hord,1973
1x105
19.6
0.05 0.06 0.07 0.08 0.09 0.10 0.05 0.06 0.07 0.08 0.09 0.10
x (m) x (m)
Ogive - 390B
Fig. 6 Wall temperature and pressure distribution with different cavitating model parameters
(Zhang et al. 2008a)
Fig. 7 Closeup particle traces

of the fluid streamlines near
the hydrofoil wall of NASA
hydrofoil 283C (Zhang et al.
2008b)
hydrofoil wall in Fig. 7. It is noticed that only the outskirt of the cavity, where the
vapor fraction is relatively small, has the liquid flow through it; however, the cavity
core near the wall, where it has the largest vapor fraction, has no liquid flow through
it. Then, it is deduced that the sustaining mechanism of the cryogenic cavity maybe
contains two parts: for the outskirt of the cavity, it is mainly sustained by the
convection of liquid flowing across the vapor-liquid surface, and for the cavity
core, it is still mainly controlled by the viscous diffusion of energy from the
surrounding mixture, other than from the vapor-liquid surface contrasted to the
B-factor theory.
The distribution of pressure and temperature along the wall of hydrofoil 293A are
plotted together in Fig. 8 to present a much clearer understanding of the process of
cavitation in liquid nitrogen. First, as the cross area of passage shrinks, the static
pressure decreases. When it becomes 104,700 Pa (the saturation pressure
Fig. 8 Pressure and

temperature distribution in temperature depression
78.0 starting point
dp 6x105
liquid nitrogen cavitation over
the wall of hydrofoil 293A 77.7
cavitation 5x105
(Cao 2011)
region
77.4
Temperature (k)
Pressure (pa)
4x105
77.1
3x105
76.8 pressure
computed T
recovery
computed P
76.5 saturation P 2x105
of computed T
76.2
1x105
0.13 0.14 0.15 0.16
X (m) –0
corresponding to the temperature), theoretically cavitation will happen. However,

the turbulence increases the cavitation pressure so that cavitation occurs in advance.
The evaporation of liquid dominates at the front of the cavity with a temperature
depression. At the same time, the static pressure varies together with saturation
pressure corresponding to the local temperature. When the temperature decreases to
the lowest value, the intensity of cavitation is the strongest. At that time, the pressure
recovers as the cross area becomes bigger, which suppresses the development of
cavity. And the condensation dominates this stage and the released heat heated the
surrounding liquid. Due to the low conductivity of liquid nitrogen, the recovery of
temperature experiences a long distance along the hydrofoil wall. The cavity closes
earlier than the recovery of temperature.
Dynamic Cavitation Model

The dynamic cavitation model (DCM) was used for quasi-steady cryogenic cavita-
tion modeling by Zhang et al. (2013). As shown in Fig. 9, the results are in good
agreement with the full cavitation model of the experimental data. The dynamic
cavitation model can perform as precisely as the full cavitation model in computing
temperature and pressure fields of quasi-steady cryogenic cavitation.
This ability of dynamic cavitation model in modeling steady and unsteady
cavitation in water is also conducted in Zhang et al. (2014, 2015).
For quasi-steady cavitating flow, the CFD code with the dynamic cavitation
model and the full cavitation model is applied to a number of validation and
demonstration problems, which include cavitating flow over submerged cylindrical
bodies with different forehead geometries and the NACA66 (MOD) hydrofoil at
different boundary conditions. The predictions of the pressure distribution from the
dynamic cavitation model are found to agree well with the experimental data and the
results from the full cavitation model. The computed cavitation region using the
dynamic cavitation model spreads out over a smaller space than the results from the
Fig. 9 Temperature and a 84.0

pressure distributions along
the ogive wall for liquid 83.8 Ogive-419A, LN2
nitrogen cavitating flow
Temperature (K)
(Zhang et al. 2013) 83.6
83.4
83.2 FCM
DCM
83.0 Experiment
82.8
0.05 0.06 0.07 0.08 0.09
Distance (m)
5
b 3.4x10 FCM
3.2x105
DCM
3.0x105 Experiment
Pressure (Pa)
2.8x105
Ogive-419A, LN2
2.6x105
2.4x105
2.2x105
2.0x105
1.8x105
0.05 0.06 0.07 0.08 0.09
Distance (m)
full cavitation model, because the former model incorporates a higher sensitivity to
pressure than the latter.
Unsteady cavitating flows through a NACA66 hydrofoil were modeled. Figure 10
gives the contours of the volume fraction α at different times, which shows the
transient evolutions of the cavity’s developing, shedding, and collapsing. The
corresponding experimental observations (Leroux et al. 2004) at the same process
are also given for comparisons. Qualitatively, they are well accordant. A long and
narrow cavity is formed in the leading edge at t = 1.26 s and then it grows larger
along the wall of the hydrofoil. When the length is more than about 0.5c, the trailing
of the cavity becomes so unstable that the cavity breaks off and sheds from the wall.
At the same time, the length of the cavity that adheres to the wall becomes smaller in
Fig. 10f–g.
Sauer-Schnerr Model
Zhu et al. extended the Sauer-Schnerr model for quasi-steady cryogenic cavitation
(2015). The realizable turbulence model is used based on the recommendation from
FCM modeling results stated above. The bubble number density per unit liquid (n) is
DCM
9.89e-01
Experiments
(a) t=1.26 s
9.00e-01
(b) t=1.30 s
8.11e-01
7.22e-01
(c) t=1.34 s
6.33e-01
(d) t=1.38 s
5.44e-01
4.55e-01
(e) t=1.42s
3.67e-01
(f) t=1.46 s
2.78e-01
1.89e-01
(g) t=1.50s
1.00e-01
Fig. 10 Contours of the water volume fraction of DCM at different times. (Fluids flow from right
to left) (Zhang et al. 2014)
calibrated, as shown in Fig. 11; when n exceeds 1010, the simulation results become
unstable and can’t get to convergence. When n equals to 1010, the pressure distri-
bution is obviously away from the experiment. When n is fixed at 105, 108, and 109,
the pressure and temperature distributions along the hydrofoil wall are all consistent
well with the experimental data. Thus, 108 is finally chosen.
5
9x10 84.0
5 Hord,1973;P
8x10 290C 83.5 290C
5
n=1x105
7x10 83.0
n=1x108
5
6x10 n=1x109 82.5
P / Pa
T/K
5
Hord,1973;P
5x10 n=1x1010 82.0
5
n=1x105
4x10 81.5 n=1x108
5
3x10 81.0 n=1x109
2x10
5
80.5 n=1x1010
5
1x10 80.0
0.12 0.14 0.16 0.18 0.20 0.22 0.12 0.14 0.16 0.18 0.20 0.22
x/m x/m
Fig. 11 Computed temperature and pressure depression in liquid nitrogen compared with exper-
imental data of hydrofoil 290C with variable bubble number density (Zhu et al. 2015)
Together with the large eddy simulation model (described in section “Turbulence
Model”-b), the calibrated Sauer-Schnerr model is further utilized to model unsteady
cryogenic cavitation (Zhu et al. 2016). The pressure correction equation described in
section “Pressure correlation equation and update of thermodynamic properties”-b is
adopted to take account the compressibility of liquid and vapor. The modeling
results are analyzed to reveal the interactions of vortices, thermal effects, and
cavitation in liquid hydrogen cavitating flows.
General Observations of Unsteady Cavitation in LH2

To illustrate the dynamic evolution of cavitation, 15 numerical snapshots with an
interval of 0.3 ms are shown in Fig. 12 (Zhu et al. 2016). The whole cavity is
vaporous with less than 60% vapor in most regions and periodically detaches from
the wall with a frequency of 275 Hz. Two completely different cavitation shedding
phenomena can be observed in a complete period from the figures. From (a) to (f),
the shape of the primary cavity almost remains unchanged, while the size shrinks
during this time. Simultaneously, the small cavitation clouds are found to flow out of
the primary cavity along the ogive surface. This stage is defined as the partially
shedding mode (PSM). From (j) to (m), different from the PSM, the primary cavity
becomes unstable and fully shed off from the lead edge of the ogive, which stage is
defined as the fully shedding mode (FSM). The other two stages represent the
transition between the PSM and FSM.
Detailed Observations of PSM

In the PSM, the primary cavity is quasi-steady, while, it is interesting to see that there
are small cavitation clouds intermittently flowing out of the rear of the primary cavity
and collapsing downstream (Zhu et al. 2016) . In order to better understand the
occurrence mechanism of the small clouds, the vorticity transport equation in a
variable density flow is employed as follows:
34
a b c
0.015
0.015 0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055
X (m) X (m) X (m)
d e f
0.015
0.015 0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055
X (m) X (m) X (m)
g h i
0.015 0.015
0.015
0.01 0.01
0.01
y (m)
y (m)
y (m)
0.005 0.005
0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055
X (m) X (m) X (m)
X. Zhang and Z. Jiakai
Fig. 12 (continued)
j k l
0.015 0.015 0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.05 0.15 0.25 0.35 0.45 0.550.65 0.75 0.85 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055
X (m) X (m) X (m)

m n o
0.015 0.015
0.015
0.01 0.01
0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.05 0.150.25 0.35 0.45 0.55 0.65 0.750.85
0.05 0.15 0.25 0.35 0.45 0.550.65 0.750.85
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.025 0.03 0.035 0.04 0.045 0.05
X (m) X (m) X (m)
Fig. 12 Instantaneous vapor volume fraction contours for unsteady LES calculations shown at intervals of 0.3 ms (Zhu et al. 2016)
35
D ω ! ! ! ! ∇ρm ∇p
!
!
¼ ω ∇ V ω ∇ V þ þ ðνm þ νt Þ∇2 ω (86)
Dt ρ2m
!
In this equation, the term Dω =Dt is the material derivative of the vorticity vector
!
ω , which describes the rate of change of vorticity of the fluid particle. The first term
on the right-hand side represents the vortex stretching term, which is zero for a 2D
case. The second term is the vortex dilatation term due to volumetric expansion/
contraction, describing the effects of fluid compressibility. If the vapor and liquid
!
density are regarded as constants (Ji et al. 2013; Bensow and Bark 2010), ∇ V ¼
!
ð1=ρv 1=ρl Þm, _ then the vortex dilatation term becomes ω ð1=ρv 1=ρl Þm, _ indi-
cating the link between the mass transfer rate and the vorticity generation. The third
term means the baroclinic torque resulting from the misaligned pressure and density
gradients. The last term is viscous diffusion of vorticity and can be ignored because
of the much smaller effect on the vorticity transport in high Reynolds number flow
(Ji et al. 2014; Huang et al. 2014b).
Figure 13 presents the contours of the calculated vapor volume fraction, pressure,
!
temperature, streamline, and distribution of Dω=Dt near the ogive wall in the PSM at
decreased intervals of 0.1 ms. In Fig. 13a, small cavitation clouds are found to shed
at an interval of about 0.4 ms, which corresponds to a frequency of 2,500 Hz. The
small clouds are generated near the leading edge inside the cavity. Its center has the
elevated vapor fraction and decreased pressure and temperature values compared
with the rest of the primary cavity (Fig. 13b, c). As the small cloud travels beyond
the cavity, the vapor in the cloud begins to condensate because of the high surround-
ing pressure field, and even a higher temperature than that of inlet liquid occurs
around the cloud, indicating the warming effect. In Fig. 13d for streamline, it is
found that the liquid can flow into the cavity, and the streamlines are nearly straight
apart from near the ogive wall, where several small discrete vortexes appear with the
clockwise rotation direction. The positions of the vortexes are the same as the small
cavitation clouds, implying that the formation of the small cloud within the primary
one is primarily due to the vorticity structures. Correspondingly, because the vortex
interior has the lowest pressure, there triggers the most violent cavitation phase
change, resulting in the highest vapor phase fraction and the lowest temperature. The
!
vorticity structure is further verified quantitatively by the contours of the term Dω=Dt
in Fig. 13e. Since the vortex has the largest mass transfer rate due to cavitation and
baroclinic torque term, respectively, corresponding to the calculations of the right
!
second term and third one of Eq. 86, it is not surprising that the Dω=Dt is intensive
near the ogive surface, but negligible in the part of cavity away from the wall. These
results demonstrate that there are strong vortex-cavitation interactions near the ogive
wall and in the shedding small vapor cloud.
The pressure fluctuations and the corresponding power spectrum density (PSD) at
position x = 0.04 m on the ogive surface are plotted in Fig. 14. The primary
frequency of pressure fluctuations is 275 Hz, being equivalent to the primary
cavitation shedding events. It is also noted that pressure fluctuation at the higher
a 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.015 0.015 0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
0.02
0.02 0.02
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.015 0.015
0.015
y (m)
y (m)
y (m)
0.01 0.01 0.01
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
0.02 0.02
b 0.02
Pressure (Pa) Pressure (Pa) Pressure (Pa)
100000 160000 220000 280000 340000 400000 460000 100000 160000 220000 280000 340000 400000 460000
100000 160000 220000 280000 340000 400000 460000
0.015 0.015 0.015
y (m)
y (m)
y (m)
0.01 0.01 0.01
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
37
Fig. 13 (continued)
0.02 0.02 0.02
38
Pressure (Pa) Pressure (Pa) Pressure (Pa)

100000 160000 220000 280000 340000 400000 460000 100000 160000 220000 280000 340000 400000 460000 100000 160000 220000 280000 340000 400000 460000
0.015 0.015 0.015
y (m)
y (m)
y (m)
0.01 0.01 0.01
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
0.02 0.02 0.02

c Temperature (K) 20 20.2 20.4 20.6 20.8 21 21.2 21.4 21.6
Temperature (K) Temperature (K) 20 20.2 20.4 20.6 20.8 21 21.2 21.4 21.6
20 20.2 20.4 20.6 20.8 21 21.2 21.4 21.6
0.015 0.015 0.015
y (m)
y (m)
y (m)
0.01 0.01 0.01
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
0.02 0.02 0.02

Temperature (K)
Temperature (K) Temperature (K) 20 20.2 20.4 20.6 20.8 21 21.2 21.4 21.6 20 20.2 20.4 20.6 20.8 21 21.2 21.4 21.6
20 20.2 20.4 20.6 20.8 21 21.2 21.4 21.6
0.015 0.015 0.015
y (m)
y (m)
y (m)
0.01 0.01 0.01
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
Fig. 13 (continued)
d 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.015 0.015
0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
0.02 0.02 0.02
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
0.015 0.015 0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
e 0.02 → 0.02 0.02 →

→
Dw Dw Dw
: S–2 –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09 –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09 : S–2 –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09
: S–2
Dt Dt Dt
0.015 0.015 0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
39
Fig. 13 (continued)
40
0.02
0.02 0.02 →
→ → Dw
Dw –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09
: S–2 –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09 Dw –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09
: S–2
: S–2 Dt
Dt Dt 0.015
0.015 0.015
0.01 0.01
0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
!
Fig. 13 The predicted vapor volume fraction, pressure, temperature, streamline, and distribution of Dω =Dt in the partially shedding mode are shown at intervals
!
of 0.1 ms (Zhu et al. 2016). (a) Vapor content, (b) pressure, (c) temperature, (d) streamlines, (e) distribution of Dω =Dt
3.5x105 3.5x107
x=0.04 m x=0.04 m
3.0x105 3.0x107
f =275 Hz
Pressure (Pa)
2.5x105 2.5x107
Power
2.0x107
2.0x105
1.5x107
1.5x105
1.0x107
1.0x105 5.0x106
5.0x104
0.00 0.01 0.02 0.03 0.04 100 1000
Time (s) Frequency (Hz)
Fig. 14 Pressure fluctuations and power spectrum density (PSD) at the position x = 0.04 m (Zhu
et al. 2016)
frequency at 600 Hz and beyond is not ignorable, especially a signal peak exists
around 2,000 Hz. These higher frequencies are closely related to the faster shedding
process of the vortex-induced small cavitation clouds.
Thermal Effects on Vortex Formation in PSM

To explore the thermal effects on the cavity developments, the simulations of
cavitation in LH2 in isothermal condition are also carried out, of which the energy
equation (Eq. 50) is not solved (Zhu et al. 2016). The results are compared in Fig. 15
together. It is found that most regions in the isothermal cavity contain more than 85%
vapor, much larger than the value in the nonthermal cavity. The shedding phenom-
ena of the small cavitation cloud in the non-isothermal cases are not observed; in
contrast, an obvious reentrant jet exists in the rear of the cavity, which is considered
to be the reason of the shedding of the attached cavity (Stutz and Reboud 1997a, b;
Le et al. 1993). The incoming liquid cannot enter the interior of the cavity and has to
flow around it, which indicates that the pressure in cavitation zone is maintained
mainly through the evaporation of liquid at the gas-liquid interface. Therefore, the
dilatation term, due to the close relationship with the mass transfer, is found to be
dominant along the liquid-vapor interface but negligible inside the cavity region. In
addition, due to the larger vapor fraction in the cavity zone compared to the
non-isothermal cases, there is a larger density gradient near the cavity interface for
the isothermal cavity. As a result, the highest levels of the vortex baroclinic torque
term exist near the closure of the cavity. The combination of these two terms in
!
Eq. 86 predominates the vorticity transport; thus the highest levels of Dω =Dt appear
at the interface and closure of the cavity.
However, for the non-isothermal cavity, both the mixture density and tempera-
ture, as well as the saturated pressure, experience a gradual variation across the
cavity except near the ogive surface, as shown in Fig. 13. These smooth transitions at
the cavity interface lead to the negligible dilatation term and baroclinic torque term,
as shown in Fig. 15b. While the gradients of vapor content and temperature, so as the
saturated pressure, are the largest immediately near the ogive surface inside the
42
a 0.02 0.02 0.02

0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 Pressure (Pa) 100000 160000 220000 280000 340000 400000 460000
0.015 0.015 0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005
0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06
x (m) x (m) x (m)
0.02
0.02 0.02 → → ∇rm x∇r
→ w (∇.V ) :[s–2] :[s–2] –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09
Dw :As–2E –1E+09 –SE+07 –2E+07 –2E+08 SE+08 1E+09 –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09 r2m
0.015
0.015
Dt 0.015
0.01
0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005
0.005
0 0
0
0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06
x (m) x (m) x (m)
b 0.02 → 0.02 → → 0.02

w (∇.V ) :[s–2] ∇rm x∇r
–1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09 :[s–2] –1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09
Dw :As–2E
–1E+09 –5E+08 –2E+08 2E+08 5E+08 1E+09
Dt r2m
0.015 0.015 0.015
0.01 0.01 0.01
y (m)
y (m)
y (m)
0.005 0.005 0.005
0 0 0
0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05 0.025 0.03 0.035 0.04 0.045 0.05
x (m) x (m) x (m)
Fig. 15 Comparison of the features in isothermal and non-isothermal cavity in FSM (Zhu et al. 2016). (a) Isothermal, (b) non-isothermal (PSM)
Fig. 16 Mechanisms of the a

LH2 unsteady cavities with Liquid
strong thermal effects (a) and
Vapor
without thermal effects (b) S tre a m line
(Zhu et al. 2016)
Ogive
b Liquid
Vapor
Ogive
cavity. Therefore, there are high levels of both the dilatation term and baroclinic
torque term.
The different vortex-occurring characteristics of the isothermal and non- isother-
mal fluids mean the different cavitation dynamics. The unique phenomenon is
illustrated by the interaction of vortex and the cavity shown in Fig. 16. For the
non-isothermal cavitation, the vortex is small compared with the primary cavity, and
the interactions between the small vortex and cavitation only happen in the boundary
layer along the wall, resulting in the phenomenon of the coexistence of the quasi-
steady primary cavity and the unsteady shedding of small cavitation cloud. However,
for the isothermal cavitation, the rear vortex travels upstream to the leading edge and
can develop to the comparable size of the primary cavity, leading to the shedding off
of the whole cavity. The experimental observations seem accordant with the phe-
nomena. Niiyama et al. (2012) conducted experiments on the NACA16-012 hydro-
foil using LN2 with an angle of attack 8 . As the cavitation number decreases, the
primary cavity just gradually expands to the trailing edge. In this process, no large-
scale cavitation clouds sheds off and the observed cavity consists of lots of bubbles
smaller than those in water. Similar phenomenon was also found in other thermo-
sensitive fluids (Cervone et al. 2006; Gustavsson et al. 2008). However, around the
NACA16-012 hydrofoil in water, periodically oscillating cavity caused by the
reentrant jet was observed when the cavity length developed over chords of
0.5–0.7 C hydrofoil (Franc and Michel 1985).
Conclusion
In this chapter, historical theoretical models about thermal effects estimation in

cryogenic cavitation are classified by their hypothesis. Those models are limited to
quasi-steady cavitation. Appearances, thermodynamic state and features of cryo-
genic cavitation are summarized from past research reports. The thermal effects
inhibit the development of cryogenic cavitation bubbles and make the cavity porous
and cloudy. The numerical simulation of cryogenic cavitation shows that the liquid
can penetrate the cavity to evaporate instead of along the vapor liquid interface in
isothermal cavitation situations. What’s more, the vorticity mainly transfers along
the surfaces of objects within the cavity resulting in a unique partially shedding
mode for cryogenic cavitation. The details of a robust numerical framework are
presented and three cryogenic cavitation models are validated again the experimental
data.
References
A.J. Acosta, A. Hollander, Remarks on Cavitation in Turbomachines, Hydromechanics Laboratory,
California Institute of Technology, Report 79.3, Oct 1959, 1–48
V. Ahuja, A. Hosangadi, S. Arunajatesan, Simulations of cavitating flows using hybrid unstructured
meshes. J. Fluids Eng. 123, 331–340 (2001)
S. Barre, J. Rolland, G. Boitel, E. Goncalves, R.F. Patella, Experiments and modeling of cavitating
flows in venturi: attached sheet cavitation. Eur. J. Mech. B Fluids 28, 444–464 (2009)
G.K. Batchelor, An Introduction to Fluid Dynamics (Cambridge University Press, New York, 1967)
R.E. Bensow, G. Bark, Implicit LES predictions of the cavitating flow on a propeller. J. Fluids Eng.
132, 041302 (2010)
C.E. Brennen, The dynamic behavior and compliance of a stream of cavitating bubbles. J. Fluids
Eng. 95, 533–541 (1973)
C.E. Brennen, Cavitation and bubble dynamics. New York : Oxford University Press; 1995.
E.A. Brun, L.A. Martinot, J. Mathieu, MeÂcanique des Fluides 3 (Dunod, Paris, 1970)
Cao Z.L., CFD modeling and experimental study of cavitation in cryogenic liquids (In Chinese),
(2011)
X. Cao, X. Zhang, L. Qiu, Z. Gan, Validation of full cavitation model in cryogenic fluids. Chin. Sci.
Bull. 54, 1633–1640 (2009)
V.P. Carey, Liquid-Vapor Phase Change Phenomena: An Introduction to the Thermophysics of
Vaporization and Condensation Processes in Heat Transfer Equipment (Hemisphere Publishing
Corporation, Washington, DC, 1992)
Cavitation models, Theory guide, ANSYS FLUENT 14.5 Documentation, 2012
A. Cervone, C. Bramanti, E. Rapposelli, et al., Thermal cavitation experiments on a NACA 0015
hydrofoil. ASME J. Fluids Eng. 128, 326–331 (2006)
S. Clerc, Numerical simulation of the homogeneous equilibrium model for two-phase flows.
J. Comput. Phys. 161, 354–375 (2000)
O. Coutier-Delgosha, R. Fortes-Patella, J.L. Reboud, et al., Experimental and numerical studies in a
centrifugal pump with two-dimensional curved blades in cavitating condition. J. Fluids Eng.
125, 970–978 (2003)
O.C. Delgosha, J.L. Reboud, Y. Delannoy, Numerical simulation of the unsteady behaviour of
cavitating flows. Int. J. Numer. Methods Fluids 42, 527–548 (2003)
M. Deshpande, J. Feng, C.L. Merkle, Numerical modeling of the thermodynamic effects of
cavitation. J. Fluids Eng. 119, 420–427 (1997)
E.R.G. Eckert, R.M. Drake, Analysis of Heat and Mass Transfer (McGraw-Hill Book Co,
New York, 1972)
P. Eisenberg, H.L. Pond, Water Tunnel Investigations of Steady State Cavities, Rept. No. 668, David
W. Taylor Model Basin, Oct 1948
R.C. Fisher, Discussion of “A survey of modern centrifugal pump practice for oilfield and oil
refining services”, by N. Tetlou. Proc. Inst. Mech. Eng. 152, 305–306 (1945)
J.P. Franc, M. Michel, Attached cavitation and the boundary layer: experimental investigation and
numerical treatment. J. Fluid Mech. 154, 63–90 (1985)
J.P. Franc, J.M. Michel, Fundamentals of Cavitation (Kluwer, Dordrecht, 2005)

J.P. Franc, C. Pellone, Analysis of thermal effects in a cavitating inducer using rayleigh equation.
ASME J. Fluids Eng. 129, 974–983 (2007)
J.P. Franc, C. Rebattet, A. Coulon, An experimental investigation of thermal effects in a cavitating
inducer. Fifth International Symposium on Cavitation, Osaka, 2003
J.P. Franc, G. Boitel, M. Riondet, et al., Thermodynamic effect on a cavitating inducer-part I:
geometrical similarity of leading edge cavities and cavitation instabilities. ASME J. Fluids Eng.
132, 021303-1–021303-8 (2010a)
J.P. Franc, G. Boitel, M. Riondet, et al., Thermodynamic effect on a cavitating inducer-part II:
on-board measurements of temperature depression within leading edge cavities. ASME J. Fluids
Eng. 132, 021304-1–021304-9 (2010b)
D.H. Fruman, I. Benmansour, R. Sery, Estimation of the thermal effects on cavitation of cryogenic
liquids. Cavitation Multiphase Flow Forum ASME FED 109, 93–96 (1991)
S. Gavrilyuk, R. Saurel, Mathematical and numerical modeling of two-phase compressible flows.
J. Comput. Phys. 175, 326–360 (2002)
T.F. Gelder, R.S. Ruggeri, R.D. Moore, Cavitation similarity considerations based on measured
pressure and temperature depressions in cavitated regions of freon 114, NASATN D-3509, 1966
M.G.D. Giorgi, D. Bello, A. Ficarella, Analysis of thermal effects in a cavitating orifice using
Rayleigh equation and experiments. J. Eng. Gas Turbines Power 132, 092901 (2010)
E. Goncalvès, Modeling for non isothermal cavitation using 4-equation models. Int. J. Heat Mass
Transf. 76, 247–262 (2014)
E. Goncalvès, R.F. Patella, Numerical study of cavitating flows with thermodynamic effect.
Comput. Fluids 39, 99–113 (2010)
E. Goncalvès, R.F. Patella, J. Rolland, B. Pouffary, G. Challier, Thermodynamic effect on a
cavitating inducer in liquid hydrogen. J. Fluids Eng. 132, 111305 (2010)
J.P.G. Gustavsson, K.C. Denning, C. Segal, Hydrofoil cavitation under strong thermodynamic
effect. J. Fluids Eng. 130, 091303(1)–091303(5) (2008)
R. Hirschi, P. Dupont, F. Avellan, et al., Centrifugal pump performance drop due to leading edge
cavitation: numerical predictions compared with model tests. J. Fluids Eng. 120, 705–711
(1998)
J.W. Holl, M.L. Billet, D.S. Weir, Thermodynamic effects on developed cavitation. J. Fluids Eng.
97, 507–513 (1975)
J. Hord, Cavitation in liquid cryogens, I-Venturi. NASA Contractor Reports, CR-2054, 1972
J. Hord, Cavitation in liquid cryogens, II-hydrofoil. NASA Contractor Reports, CR-2156; 1973a
J. Hord, Cavitation in liquid cryogens, III-ogive. NASA Contractor Reports, NASA CR-2242,
1973b
A. Hosangadi, V. Ahuja, Numerical study of cavitation in cryogenic fluids. J. Fluids Eng. 127,
267–281 (2005)
A. Hosangadi, V. Ahuja, R. Ungewitter, Analysis of thermal effects in cavitating liquid hydrogen
inducers. J. Propuls. Power 23, 1225–1234 (2007)
B. Huang, Q. Wu, G. Wang, Numerical investigation of cavitating flow in liquid hydrogen. Int.
J. Hydrog. Energy 39, 1698–1709 (2014a)
B. Huang, Y. Zhao, G.Y. Wang, Large eddy simulation of turbulent vortex cavitation interactions in
transient sheet/cloud cavitating flows. Comput. Fluids 92, 113–124 (2014b)
J. Ishimoto, K. Kamijo, Numerical study of cavitating flow characteristics of liquid helium in a pipe.
Int. J. Heat Mass Transf. 47, 149–163 (2004)
R.B. Jacobs, Prediction of symptoms of cavitation. J. Res. NBS 65C, 147–156 (1961)
B. Ji, X. Luo, Y. Wu, X. Peng, Y. Duan, Numerical analysis of unsteady cavitating turbulent flow
and shedding horse-shoe vortex structure around a twisted hydrofoil. Int. J. Multiphase Flow 51,
33–43 (2013)
B. Ji, X. Luo, R.E. Arndt, Y. Wu, Numerical simulation of three dimensional cavitation shedding
dynamics with special emphasis on cavitation–vortex interaction. Ocean Eng. 87, 64–77 (2014)
H. Kato, Thermodynamic effect on incipient and developed sheet cavitation. International Sympo-
sium on Cavitation Inception, New Orleans, Dec 1984
S. Kelly, C. Segal, J. Peugeot, Simulation of cryogenics cavitation. AIAA J. 49, 2502–2510 (2011)
K. Kikuta, Y. Yoshida, T. Hashimoto, et al., Influence of rotational speed on thermodynamic effect
in a cavitating inducer. Proceedings of the ASME 2009 Fluids Engineering Division Summer
Meeting, Colorado, 2–6 Aug 2009
R.F. Kunz, D.A. Boger, D.R. Stinebring, et al., A preconditioned Navier–Stokes method for
two-phase flows with application to cavitation prediction. Comput. Fluids 29, 849–875 (2000)
Q. Le, J.P. Franc, J.M. Michel, Partial cavities-global behavior and mean pressure distribution.
J. Fluids Eng. 115, 243–248 (1993)
E.W. Lemmon, M.O. McLinden, M.L. Huber, REFPROP: Reference fluid thermodynamic and
transport properties, NIST standard reference database 23, 2007
J.B. Leroux, J.A. Astolfi, J.Y. Billard, An experimental study of unsteady partial cavitation. J. Fluids
Eng. 126, 94–101 (2004)
H.Y. Li, S.A. Vasquez, Numerical simulation of steady and unsteady compressible multiphase
flows. International Mechanical Engineering Congress & Exposition, Houston, 2012
R.B. Medvitz, R.F. Kunz, D.A. Boger, et al., Performance analysis of cavitating flow in centrifugal
pumps using multiphase CFD. J. Fluids Eng. 124, 377–383 (2002)
C.L. Merkle, J. Feng P.E.O. Buelow, Computational modeling of sheet cavitation. Proceedings of
3rd International Symposium on Cavitation, Grenoble, 1998
R.D. Moore, R.S. Ruggeri, Prediction of thermodynamic effects of developed cavitation based on
liquid-nitrogen and freon 114 data in scaled venturis, NASA TN D-4899, 1962
F. Nicoud, F. Ducros, Subgrid-scale stress modelling based on the square of the velocity gradient
tensor. Flow Turbul. Combust. 62, 183–200 (1999)
K. Niiyama, Y. Yoshida, S. Hasegawa, et al., Experimental investigation of thermodynamic effect
on cavitation in liquid nitrogen. Proceedings of the 8th International Symposium on Cavitation,
Singapore, 2012
C.D. Ohl, M. Arora, R. Ikink, M. Delius, B. Wolfrum, Drug delivery following shock wave induced
cavitation. Presented at Fifth International Symposium on Cavitation, Osaka, 2003
F. Petitpas, J. Massoni, R. Saurel, et al., Diffuse interface model for high speed cavitating
underwater systems. Int. J. Multiphase Flow 35, 747–759 (2009)
M. Petkovšek, M. Dular, IR measurements of the thermodynamic effects in cavitating flow. Int.
J. Heat Fluid Fl. 44, 756–763 (2013)
M.S. Plesset, A. Prosperetti, Bubble dynamics and cavitation. Annu. Rev. Fluid Mech. 9, 145–185
(1977)
S.H. Rhee, T. Kawamura, H.Y. Li, Propeller cavitation study using an unstructured grid based
Navier-Stoker solver. J. Fluids Eng. 127, 986–994 (2005)
R.S. Rzlggeri, R.D. Moore, Method for prediction of pump cavitation performance for various
liquids, liquid temperatures, and rotative speeds, NASA TN D-5292, 1969
L.R. Sarosdy, A.J. Acosta, Note on observations of cavitation in different fluids. J. Basic Eng. 83,
399–400 (1961)
J. Sauer, G.H. Schnerr, Unsteady cavitating flow-A new cavitation model based on a modified front
capturing method and bubble dynamics. Proceedings of 2000 ASME Fluid Engineering Sum-
mer Conference, Boston, 2000a
J. Sauer, G.H. Schnerr, Unsteady cavitating flow-A new cavitation model based on a modified front
capturing method and bubble dynamics. Proceedings of 2000 ASME Fluid Engineering Sum-
mer Conference, Boston, 2000b
R. Saurel, O. Lemetayer, A multiphase model for compressible flows with interfaces, shocks,
detonation waves and cavitation. J. Fluid Mech. 431, 239–271 (2001)
R. Saurel, P. Cocchi, P.B. Butler, Numerical study of cavitation in the wake of a hypervelocity
underwater projectile. J. Propuls. Power 15, 513–522 (1999)
I. Senocak, W. Shyy, Interfacial dynamics-based modelling of turbulent cavitating flows, part-2:
time-dependent computations. Int. J. Numer. Meth. Fl. 44, 997–1016 (2004a)
I. Senocak, W. Shyy, Interfacial dynamics-based modeling of turbulent cavitating flows, part-1:

model development and steady-state computations. Int. J. Numer. Meth. Fl. 44, 975–995
(2004b)
A.K. Singhal, M.M. Athavale, H.Y. Li, et al., Mathematical basis and validation of the full
cavitation model. J. Fluids Eng. 124(3), 617–624 (2002)
E. Sinibaldi, F. Beux, M.V. Salvetti, A numerical method for 3D barotropic flows in turbomachin-
ery. Flow Turbul. Combust. 76, 371–381 (2006)
H. Soyama, D. Macodiyo, Improvement of fatigue strength on stainless steel by cavitating jet in air.
Fifth International Symposium on Cavitation, Osaka, 2003
W.A. Spraker, The effects of fluid properties on cavitation in centrifugal pumps. J. Eng. Power 87,
309–318 (1965)
H.A. Stahl, A.J. Stephanoff, Thermodynamic aspects of cavitation in centrifugal pumps. ASME
J. Basic Eng. 78, 1691–1693 (1956)
B. Stutz, J. Reboud, Two-phase flow structure of sheet cavitation. Phys. Fluids 9, 3678–3686
(1997a)
B. Stutz, J.L. Reboud, Experiments on unsteady cavitation. Exp. Fluids 22, 191–198 (1997b)
M. Tanguay, T. Colonius, Progress in modeling and simulation of shock wave lithotripsy (SWL).
Fifth International Symposium on Cavitation, Osaka, 2003
Theory Guide, ANSYS, FLUENT 14.5 Documentation, 2012
S.S. Thipse, Cryogenics (Alpha Science International, Oxford, UK, 2013)
Y. Utturkar, J. Wu, G. Wang, W. Shyy, Recent progress in modeling of cryogenic cavitation for
liquid rocket propulsion. Prog. Aerosp. Sci. 41, 558–608 (2005)
M. Watanabe, L. Nagaura, S. Hasegawa, et al., Direct visualization for cavitating inducer in
cryogenic flow (The 3rd report: visual observations of cavitation in liquid nitrogen),
(in Japanese) JAXA Research and Development Memorandum, JAXA-RM-09-010, 2010
Y. Yoshida, Y. Sasao, M. Watanabe, et al., Thermodynamic effect on rotating cavitation in an
inducer. ASME J. Fluids Eng. 131, 091302-1–091302-7 (2009)
Y. Yoshida, H. Nanri, K. Kikuta, et al., Thermodynamic effect on subsynchronous rotating
cavitation and surge mode oscillation in a space inducer. ASME J. Fluids Eng. 133, 061301-
1–061301-7 (2011)
I. Yutaka, N. Tsukasa, N. Takao, Periodical shedding of cloud cavitation from a single hydrofoil in
high-speed cryogenic channel flow. J. Therm. Sci. 18, 58–64 (2009a)
I. Yutaka, N. Tsukasa, N. Takao, Cavitation patterns on a plano-convex hydrofoil in a high-speed
cryogenic cavitation tunnel. Proceedings of the 7th International Symposium on Cavitation,
Ann Arbor, 17–22 Aug 2009b
A. Zein, M. Hantke, G. Warnecke, Modeling phase transition for compressible two-phase flows
applied to metastable liquids. J. Comput. Phys. 229, 2964–2998 (2010)
L.X. Zhang, B.C. Khoo, Computations of partial and super cavitating flows using implicit pressure-
based algorithm (IPA). Comput. Fluids 73, 1–9 (2013)
L.X. Zhang, B.C. Khoo, Dynamics of unsteady cavitating flow in compressible two-phase fluid.
Ocean Eng. 87, 174–184 (2014)
X.B. Zhang, L.M. Qiu, H. Qi, X.J. Zhang, Z.H. Gan, Modeling liquid hydrogen cavitating flow with
the full cavitation model. Int. J. Hydrog. Energy 33, 7197–7206 (2008a)
X.B. Zhang, L.M. Qiu, Y. Gao, X.J. Zhang, Computational fluid dynamic study on cavitation in
liquid nitrogen. Cryogenics 48, 432–438 (2008b)
X. Zhang, Z. Wu, S. Xiang, L. Qiu, Modeling cavitation flow of cryogenic fluids with thermody-
namic phase-change theory. Chin. Sci. Bull. 58, 567–574 (2013)
X.B. Zhang, W. Zhang, J.Y. Chen, et al., Validation of dynamic cavitation model for unsteady
cavitating flow on NACA66. Sci. China Technol. Sci. 57, 819–827 (2014)
X.B. Zhang, J.K. Zhu, L.M. Qiu, et al., Calculation and verification of dynamical cavitation model
for quasi-steady cavitating flow. Int. J. Heat Mass Transf. 86, 294–301 (2015)
D.F. Zhao, J.K. Zhu, L. Xu, et al., Visualization experiment of cavitating flow of cryogenic fluid in
venturi tube. (In chinese). Cryog. Eng. (submitted)
J.K. Zhu, Y. Chen, D.F. Zhao, X.B. Zhang, Extension of the Schnerr–Sauer model for cryogenic
cavitation. Eur. J. Mech. B-Fluid 52, 1–10 (2015)
J.K. Zhu, D.F. Zhao, L. Xu, Interactions of vortices, thermal effects and cavitation in liquid
hydrogen cavitating flows. Int. J. Hydrog. Energy 41, 614–631 (2016)
P.J. Zwart, A.G. Gerber, T. Belamri, A two-phase flow model for predicting cavitation dynamics.
Fifth International Conference on Multiphase Flow, Yokohama, 2004
Experiments on Gas-Liquid Flow in Vertical
Pipes
D. Lucas, M. Beyer, and L. Szalinski
Abstract
A comprehensive database on upward two-phase flows in vertical pipes was
obtained using the wire-mesh sensor technologies for gas-liquid flows in vertical
pipes. The investigations were done for different pipe diameter as well as for
flows with and without phase transfer. Wire-mesh sensors provide detailed
information on the structure of the gas-liquid interphase. Basic characteristics
of gas-liquid flows can be observed in such experiments and are discussed in this
chapter. The quantitative results obtained in the measurements as radial volume
fraction profiles, radial gas velocity profiles, bubble size distributions, distributions
of interfacial area density, and others are valuable data for the development and
validation of Computational fluid dynamics (CFD) codes for multiphase flows.
Keywords
Gas-liquid flow • Pipe flow • Bubble size distribution • Phase transfer •
Experiment • Database
Nomenclature
CFD Computational fluid dynamics
DN50 Test section with the 52.3 mm (ID) pipe
Dbub Bubble diameter
Dorifice Diameter of the gas injection orifices for the variable gas injection
Dpipe Pipe diameter
HZDR Helmholtz-Zentrum Dresden – Rossendorf
IA Interfacial area
IAD Interfacial area density
ID Inner diameter
D. Lucas (*) • M. Beyer • L. Szalinski

Institute of Fluid Dynamics, Helmholtz-Zentrum Dresden-Rossendorf, Dresden, Germany
e-mail: D.Lucas@hzdr.de

DOI 10.1007/978-981-4585-86-6_15-1
2 D. Lucas et al.
K16 Experimental test series for condensing steam-water flow

L/D Length-to-diameter ratio
L12 Experimental test series for air-water flow in 195.3 mm (ID) pipe
L20 Experimental test series for air-water flow in 52.3 mm (ID) pipe
PR17 Experimental test series for evaporating flow in case of pressure
relief
r Radius [m]
TOPFLOW Transient two-phase FLOW test facility
UG Gas velocity [m/s]
ai Interfacial area density [1/m]
db Bubble diameter [m]
JG Gas superficial velocity [m/s]
JL Liquid superficial velocity [m/s]
ε Gas volume fraction [%]
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Experimental Setup and Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
The TOPFLOW Facility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Wire-Mesh Sensors and Data Evaluation Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Test Sections, Measuring Procedures, and Test Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Flow Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Separation of Small and Large Bubbles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Evolution of the Air-Water Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Interfacial Area Density (IAD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Steam Bubble Condensation in Sub-cooled Liquid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Transient Data from Pressure Relief Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Introduction
Gas-liquid flows in vertical pipes are well suited to study basic phenomena and
general characteristics of two-phase flows with deformable interfaces. Here, the flow
develops under clear boundary conditions and can be investigated for a relatively
long distance. Different flow pattern such as bubbly flow, slug flow, churn-turbulent
flow, and annular flow as well as transitions between them may occur depending on
the flow rates and size of the pipe.
For many practical applications as pipelines, steam generator tubes, and many
others, pipe flow is of interest on its own, but more generally the findings obtained
for two-phase flows in pipes can be transferred to more complex flow situations using
computational fluid dynamics (CFD). For medium- and large-scale applications, the
two- or multi-fluid approach is frequently applied. Here closure models are essential to
reflect the non-resolved phenomena on local scale. The establishment of CFD setups
with sets of closures which can be applied to a wide range of flow conditions without
Experiments on Gas-Liquid Flow in Vertical Pipes 3
Table 1 Test series

Pipe Range of gas
diameter Pressure Range of liquid flow rates
Test [mm] Fluids [bar(a)] Flow type flow rates [m/s] [m/s]
L20 52.3 Air- 1.1–2.1 Adiabatic 0.04–4 0.0025–19
water
L12 195.3 Air- 2.5 Adiabatic 0.04–1.6 0.0025–3.2
water
K16 195.3 Steam- 10, Condensing 0.4–1.0 0.09–0.53
water 20, 40, steam
65
PR17 195.3 Steam- 10, Evaporation 0 or 1.0 –
water 20, 40, by pressure
65 relive
any tuning is an important requirement as discussed in chapter “▶ Euler-Euler-Model-

ling of Poly-dispersed Bubbly Flows”. High-quality data on vertical pipe flow are one
important basis to develop, adapt, and validate such closure models. Afterward the
validated model setup can also be used for more complex flow situations.
The aim of the experiments discussed in this chapter is to achieve a better under-
standing of two-phase flows and to provide a comprehensive high-quality database
suitable for the CFD model development and validation. The latter issue requires data
with high resolution in space and time as they are provided by innovative measuring
techniques as presented in chapter “▶ Imaging Measuring Techniques”. The database
includes data for adiabatic air-water flows and steam-water flows with phase transfer for
different pressure levels up to 6.5 MPa. An overview is given in Table 1. For all these
cases flow rates and other parameter were varied over a wide range.
Experimental Setup and Boundary Conditions
The TOPFLOW Facility
TOPFLOW is an acronym standing for Transient twO-Phase FLOW test facility. It

is designed for the generic and applied study of transient two-phase flow phenomena
in the power and process industries. By applying innovative measuring techniques,
TOPFLOW provides data suitable for CFD code development and qualification.
TOPFLOW allows to perform steam-water or air-water experiments. The facility is
described in detail by Schaffrath et al. (2001) and Prasser et al. (2006).
The TOPFLOW facility can be operated at pressures up to 7 MPa and the
corresponding saturation temperature of 286 C. The maximum steam mass flow
is about 1.4 kg/s, produced by a 4 MW electrical heater. The maximum saturated
water mass flow rate through the vertical test section is 50 kg/s. Different test
sections can be operated between the heat source (steam generator) and the heat
sink (cooling systems).
4 D. Lucas et al.
Fig. 1 The integration of the Variable Gas Injection test section in the TOPFLOW facility in case
of the K16 experiments
Figure 1 shows a scheme of the integration of the vertical pipe test section loop in the
TOPFLOW facility for the example of the K16 experiments. Water is circulated in loop
entering the vertical pipe test sections from below. Air or steam is supplied from gas
injection systems which are different for the different tests and which are described
below. The TOPFLOW steam drum is used as separator for all experiments.
Wire-Mesh Sensors and Data Evaluation Procedures
The challenges to measure the gas-liquid phase distribution are discussed in detail in
chapter “▶ Imaging Measuring Techniques”. The aim of the experiments discussed
here was to get data on the evolution of gas-liquid flows in a pipe for a wide range of
flow rates with high resolution in space and time. Such data can be provided using
the so-called wire-mesh sensor techniques, originally developed by Prasser et al.
(1998). Improvements of this technology are presented by Prasser et al. (2001) and
Pietruske and Prasser (2007).
A wire-mesh sensor consists of two grids of parallel wires, which span over the
measurement cross section (see Fig. 1 in chapter “▶ Imaging Measuring Techniques”).
The wires of both planes cross under an angle of 90 but do not touch. Instead there
is a vertical distance between the wires at the crossing points. At these points, the
conductivity is measured. According to the different conductivity of gas and water,
the phase present in the moment of the measurement at the crossing point can be
determined. Many different types of wire-mesh sensors were built and successfully
used by an international community during the last 15 years.
Sensors with 16 16 wires (52.3 mm pipe) and 64 64 wires (195.3 mm pipe)
were used in the presented experiments. For both sensor sizes, measurements were
done with a frequency of 2,500 Hz, i.e., each crossing point was measured 2,500
time per second. This allows to measure all gas structures larger than about 3 mm
sphere equivalent diameter.
The measuring time was 10 s, i.e., the result of one single measurement is a three-
dimensional matrix of 16*16*25.000 or 64*64*25.000 values of the instantaneous
local conductivity, respectively. By a calibration procedure, a matrix of the instan-
taneous local volume void fraction with the same dimensions is calculated.
The void fraction values can be visualized to get detailed insights on the flow
characteristics. However, more important is the derivation of quantitative data by
using averaging procedures (Prasser et al. 2005b). Most important is the time
averaging, which, e.g., leads to time-averaged two-dimensional gas volume fraction
distributions in the pipe cross section. Due to the radial symmetry of the data, the
statistical error can be further lowered by an azimuthally averaging. To do this, the
cross section is subdivided into ring-shaped domains with equal radial width. The
contribution of each mesh cell is calculated by weight coefficients obtained from a
geometrical assignment of the fractions of a mesh belonging to these rings. In the
result, radial gas volume fraction profiles are obtained.
For the measurements, two sensors were used which measurement planes have a
distance in the range of few centimeters. This allows to cross-correlate the gas
volume fraction values of the two planes for all mesh points which are located
above each other. From the maxima of the cross-correlation functions, the typical
time-shift of the local void fraction fluctuations can be determined. Since the
distance between the measuring planes is known, the local time-averaged gas
velocity can be calculated. The point-to-point two-dimensional gas velocity distri-
butions in the pipe cross section are obtained in the results of this procedure. Again
an azimuthally averaging is applied to obtain the radial profiles of the gas velocity.
This procedure has clear limits for the transition region to annular flow with a gas
core in the pipe center. Here the gas velocity can be much larger than the velocity of
the interfaces determined here. Please consider that the second sensor is only used
for the determination of the gas velocities. Due to the perturbing effect of the first
sensor, other data, especially bubble size distributions obtained from the second
sensor, would be distorted.
The next step of the data evaluation procedure is the identification of single
bubbles. A bubble is defined as a region of connected gas-containing elements in
void fraction matrix which is completely surrounded by elements containing the
6 D. Lucas et al.
liquid phase. A complex procedure, introduced by Prasser et al. (2001), applies a

filling algorithm combined with sophisticated stop criteria to avoid artificial coales-
cence as well as artificial fragmentation of bubbles. In the result, the same identifi-
cation number is assigned to all volume elements which belong to the same bubble.
Different bubbles receive different identification numbers. These numbers are stored
in the elements of a second array. This array has the same dimension as the void
fraction array. Combining the information from the void faction and bubble number,
arrays together with the radial profiles of the gas velocity characteristic data of the
single bubbles as bubble volume, sphere equivalent bubble diameter, maximum
circle equivalent bubble diameter in the horizontal plane, coordinates of the bubble
center of mass, moments characterizing asymmetries, and others are obtained. Based
on these data, cross section and time-averaged bubble size distributions and radial
gas volume fraction profiles decomposed according to the bubble size are calculated.
The bubble size distributions are defined volume fraction related, i.e., they present
the volume fraction per width of a bubble diameter class (equivalent diameter of a
sphere with the measured bubble volume Vb is considered).
Besides these evaluation procedures, the wire-mesh sensor data comprise also
information on the interfacial area density (IAD). To obtain it, first the interfacial
area (IA) has to be determined. This is done by the evaluation of the surface area of
each single bubble measured. Due to the spatial resolution of the wire-mesh sensor
of usually 3 3 mm2 in the plane of the pipe cross section, the IA cannot be
determined directly from the measurement data. If a bubble moves through the wire
mesh, its surface area information is stored as averaged void fraction values. As
originally proposed by Prasser 2007, an iterative approach is used for the determi-
nation of the bubble surface. Based on an initial threshold for void fraction which is
characteristic for each single bubble, the surface area and the enclosed volume are
calculated. Since the bubble volume is already known from the bubble identification
algorithm, the void fraction threshold is modified iteratively until both volumes are
equalized.
Thereby the whole bubble surface area is assembled from single area parts from
each volume element of the wire-mesh sensor data related to the current bubble
surface. These volume elements are cuboids of 3*3* gas velocity/measurement
frequency mm3, on whose corner points void fraction data are available. Resulting
from the interpolation between these void fraction values, the algorithm calculates
geometrical points which are elements of the bubble surface area under consider-
ation of the void fraction threshold at the bubble surface. Using these points, a
reconstruction of the bubble surface is possible. Thereby the IA algorithm considers
different geometrical arrangements of the area parts inside the volume elements,
whereas these area parts may again consist of some more little subarea parts.
Based on the single area assembling, the algorithm slightly overestimates the IA
in dependence on the bubble size. As smaller the bubble, as more this effect
influences the results. For this reason, the calculated surface area of small bubbles
is corrected using the known bubble volume and the maximal cross section area to
obtain a surface area of an equipotential ellipsoid. Depending on the bubble size and
related to the wire pitch of the mesh sensor, a weighting function determines the final
bubble surface.
Due to the determination of the surface area of each single bubble, the algorithm
also provides detailed distributions. Thus, time-dependent cross section averaged
time sequences of the IA and IAD as well as azimuthal and time-averaged radial
distributions are calculated. Furthermore the algorithm provides distributions of both
parameters related to bubble size classes including quotients of the calculated IA and
the surface area of a volume equivalent sphere. Additionally for each measurement, a
matrix of 64 64 time-averaged IAD values is calculated that can be used for a
quick view over the flow data. For further investigations of the flow structure,
extensive files are available that contain detailed parameter for each detected single
bubble of the measured flow.
Test Sections, Measuring Procedures, and Test Matrices
As discussed in the previous section, wire-mesh sensors can provide quite detailed
information on the gas-liquid phase distribution in pipe flows. In addition it can also
be applied for oil-water flows as shown, e.g., by Szalinski et al. (2010) who presents
measurements on air-water flows in comparison with experiments done for oil-water
flows. However, the wire-mesh sensor is an intrusive measuring technique, which is
a clear disadvantage of this technology. Several investigations on this effect
were done; see, e.g., Prasser et al. (2001, 2005a), Sharaf et al. (2011), and Nuryadin
et al. (2015). Summarizing these investigations, it can be said that the sensor signal
well reproduces the bubble sizes and shapes as long as there is a minimum liquid
flow rate equivalent to a superficial velocity JL (volume flow rate divided by the
cross section area of the pipe) of about 0.2 m/s. For lower flow rates, a deceleration
of the bubbles may occur. Unfortunately the flow behind the sensor is clearly
disturbed. Most important is here the effect of a bubble fragmentation by the sensor.
As mentioned before, the aim of the experiments is to investigate the evolution of the
flow along the pipe. Because of the disturbed flow behind the sensor, it is however
not meaningful to place several sensors in the pipe to get the information of the gas
distribution at different inlet lengths. Instead two different methods were used. For
the 52.3 mm pipe (test series L20), pipe sections of different lengths were used to
vary the distance between the fixed position of the gas injection and the measuring
plane. For the 195.3 mm pipe, such a procedure, which requires to rebuild the facility
for measurements at different height positions, would be not feasible. For this
reason, the so-called Variable Gas Injection test section was constructed. Here the
measuring plane is located at the top of the pipe for all measurements. To vary the
inlet length, gas injection devices are placed on several distances below the measur-
ing plane. They allow a gas injection through holes in the pipe wall to minimize the
disturbance of the flow by the non-active gas injections.
8 D. Lucas et al.
Fig. 2 Vertical test section

DN50 (a), pipe section with
gas injection module (b), and
gas injection device (c)
Vertical Test Section DN50

The TOPFLOW test facility is equipped with two vertical pipes for the investigation
of vertical two-phase flows. Both pipes are connected with a circulation pump and a
separation tank (steam drum) to a test section circuitry (Fig. 1). Both test sections can
operate separately using isolating valves. The DN50 test section has an inner
diameter of 52.3 mm and a length of about 9 m. It is assembled from several sections
with various lengths, to allow different inlet length between the gas injection and the
measurement plane. The water from the steam drum flows through the circulation
pump and is injected into the vertical test section from the bottom. The gas was
mixed into the water flow also from the bottom, using a central injection device. It is
equipped with eight orifices of 4 mm diameter (Fig. 2). Leaving the vertical pipe, the
two-phase mixture flows via a horizontal pipe section into the steam drum, where it
separates. The gas is released into the blow-off tank and finally to the atmosphere.
The water is pumped again in the test section.
Fig. 3 General test matrix of HZDR for vertical pipe flow with the measured points for the DN50
air-water upward flow series L20 marked in gray
Applying this test section, comprehensive experiments were done including a

wide range of combinations of flow rates (represented by the superficial velocities JL
and JG). Four different inlet lengths were investigated, 100 (A), 1600 (C), 3100 (D),
and 7910 (F/F2) mm, so that the evolution of the flow along the vertical pipe can be
analyzed. Thereby the gas injection always was installed at the bottom of the pipe,
while one wire-mesh sensor is shifted to the different positions. The corresponding
length to diameter ratios L/D are 1.9 (A), 30.6 (C), 59.3 (D), and 151.2 (F/F2),
respectively. Figure 3 shows the test matrix for upward air-water flows. For the
matrix points with red numbers, the measurements for the largest inlet length were
measured twice to allow reproducibility checks. The first measurement is denoted
with F, the second with F2. After detailed consistency checks, measurements for
some height positions were omitted to include only reliable data into the database.
Due to the different position of the wire-mesh sensor, the pressure at sensor position
is slightly different. It varies between 0.11 and 0.21 MPa (absolute pressure). During
the measurements, the temperature changed in a range from 16 C to 45 C.
The Variable Gas Injection Test Section

This test section (Fig. 4) consists of a vertical steel pipe with an inner diameter of
195.3 mm and a length of about 8 m. The measuring plane is located at the upper end
of the test section. The device is equipped with six (L12 test series) or seven (K16
test series) gas injection units which allow to inject air or steam via orifices in the
pipe wall.
The gas injection via wall orifices offers the advantage that the two-phase flow
can rise smoothly to the measurement plane, without being influenced by the feeder
within the tube at other height positions. The injection devices are arranged almost
logarithmically over the pipe length since the flow structure varies quite fast close to
the gas injection mainly caused by the radial redistribution of the gas. Six of the gas
injection modules consist of three injection chambers each. Two of the three
chambers (the uppermost and the lowest) have 72 1 mm orifices. The middle
chamber has 32 4 mm orifices, which is used to vary the initial bubble size
10 D. Lucas et al.
Fig. 4 The test section “Variable Gas Injection” for the L12 experiments and a gas injection
module
distribution. For rotation-symmetric gas injection, all orifices per chambers are
equally distributed over the circumference of the pipe. For the K16 experiments,
an additional injection chamber with 1 mm orifices was installed as close to the
measuring plane as possible (40 mm between gas injection and measurement plane
of the first wire-mesh sensor in flow direction; L/D ~ 0.2). This was done to provide
more detailed information on the injected steam bubbles. Only one injection cham-
ber is activated for a single measurement. L/D is increased by using the gas injection
chambers from “A” (L12) or “@” (K16) through “R” and “B” through “Q” for 1 and
4 mm injection, respectively. Caused by the implementation of the additional
injection chamber, the injection lengths are 83 mm larger in the K16 experiments
compared to the data given in Fig. 4 for the L12 experiments.
Procedure and Test Matrix for the L12 Test Series

During each experiment, water is fed from the steam drum via the test section pump
through the vertical test section (see Fig. 1). For the experiments, water mass flow
rates between 1.2 and 48 kg/s were applied. The water temperature was kept constant
at 30 C 1 K during the experiments by a special procedure explained in detail by
Beyer et al. (2008). More information on the L12 experiments and their results can
be found in Lucas et al. (2010a, b).
Because of the hydrostatic pressure and friction pressure loss, the pressure
changes along the pipe. To reflect the same situation as having a fixed position for
the gas injection and shifting the measuring plane, the system pressure was adjusted
to guarantee the nominal value of 0.25 MPa exactly at the position of the respec-
tively activated injection device. In contrast to the K16 experiments discussed below,
no measurement for the pressure difference between the top of the pipe and the single
injection devices was available during the L12 test series. For this reason, the
pressure differences were estimated by a procedure described by Beyer et al. (2008).
The measured combinations of air and water superficial velocities are shown in
Fig. 5. Two test series with constant liquid superficial liquid velocities JL of about 0.4
and 1.0 m/s and two with constant gas superficial velocities JG about 0.01 m/s and
0.2 m/s were done. To investigate the evolution of the flow along the pipe, all levels
(A–R) were measured for any points smaller than 149. The maximum possible gas
Fig. 5 General experimental test matrix for vertical pipe flows with marked points measured for
the L12 test series
12 D. Lucas et al.
flow rate, which can be injected through the injection chambers with a diameter of
1 mm, is limited. For this reason, for the points 149, 151, 160, and 162, both
injection chambers with 1 mm orifices were operated parallel. For the measurement
points 171, 173, 182, and 184, only the 4 mm injections were used. The investigated
combinations of flow rates lead to different flow pattern starting from bubbly flow,
via churn-turbulent flow up to wispy-annular flow.
Procedure in the K16 Test Series

In this experimental series, condensing pipe flows were investigated. To do that,
some extensions of the test section “Variable Gas Injection” were done compared to
the L12 experiments. The experiments base of experiences obtained in a previous
test presented by Lucas and Prasser (2007). The test section pump (see Fig. 1)
circulates the saturated water from the steam drum to the lower end of the variable
gas injection. In addition cold water is injected through a mixing device at the lower
end of the test section. This allows to obtain a sub-cooling of the water of several
Kelvin depending on the flow rates. Several temperature measurements were added
to check and control the temperature. As in the L12 experiments, the nominal
pressure is set at the position of the respectively activated injection chamber. Thus
switching between different positions of the injection provides the same conditions
like in case of a fixed location of the injection and shifting the measuring plane. This
is especially important for the condensation experiments since the saturation tem-
perature and by that also the sub-cooling depends on pressure. To adjust the pressure,
the absolute value is measured at the upper end of the test section. In addition the
differential pressure between this measurement position and the position of the
single gas injection is determined using a newly installed measuring system. After
adjusting the pressure for the selected flow rates and for the position of the activated
steam injection, the aspired sub-cooling has to be set up. The water temperature is
measured by thermocouples, mounted in the saturated and the cold water pipes as
well as in the Variable Gas Injection pipe below the injection levels R and O for the
mixing temperature. The total water mass flow rate (saturated water from the loop
and the additional cold water injection) and the water temperature are adjusted
together finally to reach the aspired values.
Measurements were done for four different pressure levels in the range from 1 up
to 6.5 MPa. Furthermore steam and water flow rates as well as the sub-cooling of the
water were varied as indicated in Table 2.
PR17 Test Series

The “Variable Gas Injection” test section was also used for experiments on evapo-
rating pipe flow. The evaporation was induced by pressure relief, i.e., the gas
injection devices itself were not used in these experiments. Two different experi-
mental procedures were applied as illustrated in Fig. 6.
In case of the first procedure (left-hand side of Fig. 6), water was circulated with a
superficial velocity of about 1 m/s and flows upward through the test section. Before
starting the pressure release saturation, conditions are obtained in the steam drum.
Since the circulation pump is located at much lower elevation, the pressure exceeds
Table 2 Test matrix for Point JL [m/s] JG [m/s] ΔT [K]

the K16 experiments
1 MPa
118 1.017 0.219 3.9, 5.0
138 0.405 0.534 4.7, 5.3, 6.3, 6.6, 7.2
140 1.017 0.534 3.7, 4.8, 5.0, 6.0
2 MPa
118 1.017 0.219 3.7, 4.9, 6.0
138 0.405 0.534 4.8, 6.6, 8.7
140 1.017 0.534 3.2, 5.0, 6.8
4 MPa
118 1.017 0.219 2.7, 5.0, 7.2
138 0.405 0.534 2.6, 6.6, 12.6
140 1.017 0.534 2.6, 5.0, 7.6
6.5 MPa
096 1.017 0.0898 2.4, 5.0, 7.6
116 0.405 0.219 2.9, 6.0, 9.0
118 1.017 0.219 2.5, 5.0, 7.4
138 0.405 0.534 2.5, 9.8, 17.2
140 1.017 0.534 2.3, 5.0, 8.7
Fig. 6 Schemas of the two experimental procedures for the PR14 experiments (left procedure
1, right procedure 2) (From Mikuz et al. (2015))
14 D. Lucas et al.
Test 1 Test 2
Opening level of Pressure R t1 t2 t3 R t1 t2 t3
the blow-off valve
MPa % s s s % s s s
R
1 60 21 30 30 50 18 34 30
time 2 50 18 34 30 40 14 42 30
t1 t2 t1 t3 4 30 11 48 30 25 9 52 30
6,5 25 9 52 30 20 7 56 30
Fig. 7 Test matrix for procedure 1 in the PR17 test series, R is the relative degree of opening of the
blow-off valve (From Liao et al. (2013))
the saturation pressure at this position. It is important to maintain sub-cooled

conditions at the position of the pump also during the pressure relief to avoid
cavitation. This condition limits the speed of the depressurization which can be
used in the experiments. Therefore, the blow-off valve which is located at the steam
drum was only partially opened. According to the rather small pressure gradients,
also the maximum void fraction generated in test section by evaporation is limited.
The blow-off valve used in the procedure has relative long opening and closing
times. For this reason, the valve was opened and closed according to the ramp shape
shown at the left-hand side of Fig. 6. The relative degree of opening at the plateau
and the corresponding durations are shown in the test matrix, Fig. 7. In total there are
eight tests. Each test was repeated using the same conditions to check the
reproducibility.
For the second procedure, the facility was equipped with an additional blow-off
line which was mounted at the upper end of the test section (right-hand side of
Fig. 6). After heating up and before the initiation of the depressurization, the test
section was separated from the loop by valves, i.e., the experiments run from
stagnant liquid. The new blow-off line is equipped with a fast opening valve
allowing an opening ramp as shown at the left-hand side of Fig. 8. An orifice with
a diameter of 20 mm was implemented in the blow-off line to limit the speed of
depressurization. Much steeper pressure gradients resulting in much larger void
fraction are obtained by this procedure. The test matrix is presented in Fig. 8. As
in case of procedure 1, the pressure relief was start from 4 different pressure values.
Opening times of 10 and 20 s were used. Again each of the eight runs was repeated
once again.
Both procedures have advantages and disadvantages. The advantage of the first
procedure is the relatively large velocity of the fluid at the measuring plane. Previous
investigations on the intrusive effect of the wire-mesh sensor have shown that the
uncertainties of the measurements increase for small water velocities. Bubbles may
be considerably decelerated due to the interaction with the wires for water superficial
velocities below 0.2 m/s. For a water superficial velocity of 1 m/s as applied in this
procedure, this undesired effect is rather negligible, and reliable data are obtained.
For the second procedure, it is expected that the bubbles are also pushed through the
Test 1 Test 2
Pressure t1 t2 t3 t1 t2 t3
Opening level of the
fast-acting blow-off valve MPa s s s s s s
open 1 5 10 35 5 20 55
2 5 10 85 5 20 75
time
closed
t1 t2 t3 4 5 10 85 5 20 75
6,5 5 10 85 5 20 75
Fig. 8 Test matrix for procedure 2 in the PR17 test series (From Mikuz et al. (2015))
sensor due to the boiling up, but it is rather difficult to quantify a possible interaction
between the sensor wires and the bubbles. Another disadvantage of the second
procedure is caused by the fact the valves which separate the test section from the
loop (Fig. 6, right) have relatively long closing times. Due to heat losses during the
waiting time before the pressure relief is started, a slight sub-cooling will be obtained
in the test section. Nevertheless the second procedure has the advantage that stepper
pressure gradients can be realized leading to higher void fractions.
Results
For all the experiments described above, well-documented data on boundary condi-
tions and on information obtained from the wire-mesh sensor measurements are
available. It is a comprehensive database containing detailed information on the
phase distributions in vertical pipe flows under various conditions. Examples for
derived quantitative data are radial gas volume fraction, interfacial area concentra-
tion, and gas velocity profiles as well as bubble size distributions. This database is
frequently used for CFD model development and validation (e.g., Kaji et al. 2009;
Duan et al. 2011; Qi et al. 2012). The single documentations on the experiments also
discuss uncertainties and include plausibility checks of the measured data. Some
considerations on uncertainties specifically for the L12 experiments are discussed by
Beyer et al. (2010). In frame this handbook these findings will not be repeated.
Instead some more basic observations from the experiments are discussed in this
section.
16 D. Lucas et al.
Fig. 9 Examples for different

flow pattern observed in the
L20 experiments at
L/D = 151 (fully developed
flow)
121 040 085 118 129 140 215
Table 3 Assignment of flow pattern in Fig. 9

Test JL [m/s] JG [m/s] Flow pattern
121 4.047 0.2190 Finely dispersed
040 0.641 0.0096 Bubbly flow with wall peak
085 1.017 0.0574 Bubbly flow with wall and center peak
118 1.017 0.2190 Bubbly flow with center peak
129 1.017 0.3420 Bubbly to slug flow transition
140 1.017 0.5340 Slug flow
215 0.405 12.14 Annular flow
Flow Structure
Flow pattern maps for vertical pipe flow were introduced long time ago, e.g., by
Taitel et al. (1980). Other authors published similar maps using different criteria or
introducing sub-pattern. The wire-mesh sensor data can be visualized, e.g., by
plotting the gas distribution in a plane along the pipe diameter or by a virtual side
view generated by a special ray tracing algorithm developed by Manera et al. (2005).
This allows to distinguish flow pattern subjectively. Figure 9 shows an example of
such views, and Table 3 presents the assignment to different flow pattern for the
52.3 mm pipe.
Figure 9 demonstrates the dependency of the flow pattern on the flow rates, but in
general there is also a dependency on the pipe size and on the inlet length. Figure 10
compares flow structures for the 52.3 mm pipe and the 195.3 mm pipe for the same
flow rates.
0.037 0.057 0.090 0.14 0.22 0.34 0.53 0.84 1.3 0.037 0.057 0.090 0.14 0.22 0.34 0.53 0.83 1.3
J air [m/s] J air [m/s]
Dpipe = 52.3 mm Dpipe = 195.3 mm
Fig. 10 Dependency of air-water flow structures for JL = 1.017 m/s and increasing air flow rates on
the diameter of the pipe
The main difference is the shape of the large gas structures. While in the 52.3 mm
pipe slightly distorted Taylor bubbles are observed, rather irregular large gas struc-
tures can be found in the 195.3 mm pipe. This is in accordance with literature.
Because of the decreasing confining effect of the pipe with increasing pipe diameter,
Taylor bubbles become more and more unregularly finally forming highly distorted
gas structures which characterize the churn-turbulent flow regime. However for both
pipe sizes, bimodal bubble size distributions can be found for the same combination
of flow rates as shown in Fig. 11. The peak representing the large bubbles is shifted
to larger bubble sizes and is also broader for the larger pipe. This effect was
discussed in detail by Prasser et al. (2005b, 2007). The wire-mesh sensor technology
allows to extract single large gas structures from a churn-turbulent flow as shown in
the same paper. Such extracted gas structures can be understood as a strongly
distorted Taylor bubble. For this reason, there is some similarity between slug and
churn-turbulent flows with the most important difference of more regular shapes in
smaller pipes and more distorted shapes in larger ones. Also, it can be concluded
from these experiments and measurements done for other pipe diameters that there is
a smooth transition between slug flow and churn-turbulent flow in dependence on
the pipe diameter.
18 D. Lucas et al.
Fig. 11 Bubble size distributions for the same flow rates (JL = 1.017 m/s and JG = 0.53 m/s) but
different pipe diameter
The quantitative data obtained by the wire-mesh sensors allow to define more
objective criteria to define flow pattern as introduced, e.g., by Lucas et al. (2005) for
experiments done in a 51.6 mm pipe. Here the transition between bubbly flow and
slug flow was defined basing on the sphere equivalent diameter of the largest
bubbles. If it exceeds the pipe diameter, the flow was assigned to slug flow, otherwise
to bubbly flow. This criterion may hold for pipes with a diameter in the range up to
50 mm. In case of larger pipes, the flow characteristics are typical closer to churn
turbulent if bubbles larger 50 mm sphere equivalent diameter occur. For this reason,
a more generalized criterion for the transition between bubbly and slug or churn-
turbulent flow can be formulated. The flow can be assigned to bubbly flow
if db < min (Dpipe, 50 mm).
For annular flow, only small liquid drops – which cannot be registered by the
wire-mesh sensor – should occur in the pipe center. For the L12 experiments, some
water bridges are still observed for the largest gas superficial velocity of 3.185 m/s.
These measuring points still lie in the transition between churn-turbulent and annular
flow. Sometimes this region is also called wispy-annular flow. In this case, some
liquid wisps are overserved to detach from the liquid film as shown in Fig. 12. Here
also some larger drops are visible in the core region of the pipe.
For the bubbly flow region, in addition a distinction between flows with wall and
center peaks of the radial gas volume fraction profiles can be used to characterize the
flow structure. The nature of these profiles is discussed in detail in section “Separa-
tion of Small and Large Bibbles.” Also for some flow rates, bubbly flows in the
transition between these two situations occur, showing a wall peak as well as a center
peak of these profiles.
Fig. 12 Instantaneous liquid

distribution in the cross
section of the 195.3 mm pipe,
JL = 0.405, JG = 3.185,
L/D = 39.9
Using the criteria introduced above, flow pattern maps can be established for the
L20 and L12 experiments. Such maps are shown in Fig. 13. While the general
tendencies are rather similar for both pipe configurations, some differences can be
seen for the positions of the transition lines. Also, for some conditions, transitions
from bubbly flow to churn-turbulent flow caused by bubble coalescence can be
found for the L12 experiments.
Another interesting aspect is the influence of the initial bubble size distribution on
the flow characteristics. As explained before, the L12 measurements were done for
the same combinations of flow rates but injecting the gas through holes with different
diameter. This results in different initial bubble size distributions. The influence on
bubble size distributions and correspondingly to other characteristics like radial gas
volume fraction and gas velocity profiles is still visible for the largest L/D measured
in case of low gas volume rates. Starting from point 072 (JG = 0.0368 m/s) for
JL = 0.405 m/s and point 107 (JG = 0.14 m/s) for JL = 1.017 m/s, the distributions
and profiles are the same under consideration of the experimental uncertainties.
Separation of Small and Large Bubbles
The distribution of the bubbles over the pipe cross section can be characterized by
radial gas volume fraction profiles because of the symmetry of the flow, which was
checked for all experiments. For bubbly flows, the maximum of this distribution
can be located in the near wall region or in the center, while it is always in the pipe
center for all other flow regimes (Fig. 14). These profiles are established by the
20 D. Lucas et al.
Fig. 13 Flow pattern maps for the L20 experiments (upper figure) and L12 experiments (lower figure)
action of the so-called non-drag forces (Lucas et al. 2001, 2007, chapter “▶ Euler-
Euler-Modelling of Poly-dispersed Bubbly Flows”), which act perpendicular to the
main flow direction.
The lateral lift force acts in an upward vertical pipe flow to push the small bubbles
toward the pipe wall. However it strongly depends on the bubble size as found
experimentally by Tomiyama et al. (2002) and in Direct Numerical Simulation
(DNS), e.g., by Ervin and Tryggvason (1997) and Bothe et al. (2007). In case of
deformed bubbles, the lateral lift force even changes its sign. According to the
correlation obtained by Tomiyama et al. (2002), this change of sign of the lift
force occurs for air-water flows at ambient conditions for a sphere equivalent bubble
diameter of about 5.8 mm. Since the bubble deformation depends on the relation
between buoyancy and surface tension reflected by the Eötvös number, it also
depends on the material properties and decreases for steam-water flows at
Fig. 14 Examples for radial gas volume fraction distributions for developed flow; left, L20 (inner
diameter 52.3 mm); right, L12 (inner diameter 195.3); (a) bubbly flow with wall peak, (b) bubbly
flow with transition between wall peak and center peak, (c) slug or churn-turbulent flow, (d)
transition to annular flow
22 D. Lucas et al.
6.5–3.5 mm sphere equivalent diameter. Due to the change of sign, larger bubbles
can be found in upward flows preferred in the pipe center, i.e., in general there is a
clear separation of small and large bubbles in pipe flow. This separation can be
clearly shown by the wire-mesh sensor data as discussed by Prasser et al. (2002).
Since the bubble sizes are known, the radial volume fraction profiles can be
decomposed according to the bubble size. Figure 15 shows examples for such
decomposition. The separation of small and large bubbles can clearly be seen.
The correlation from Tomiyama et al. (2002) was obtained for single bubbles in a
highly viscous linear laminar shear flow. Surprisingly the correlation fits at least
regarding the change of the sign of the lift force very well with the observed
separation of small and large bubbles in the TOPFLOW experiments as discussed
in detail by Lucas and Tomiyama (2011). Here the flow is clearly turbulent, there are
cases with high void fraction, and the material parameters are clearly different. The
agreements hold not only for air-water flows in the L20 and L12 experiments but
also in the experiments for steam-water flows K16 under different pressure levels as
shown in Fig. 15. It also holds for the refrigerant dichlorodifluoromethane (R12) as
shown by Krepper et al. (2013).
For the CFD modeling of polydisperse bubbly flows, it is essential to reflect this
separation of small and large bubbles as soon as the bubble sizes are in the range of
the change of the sign of the lift force since the local distribution may have a strong
influence on the evolution of the flow as discussed by Krepper et al. (2005, 2008),
Frank et al. (2008), and Lucas et al. (2011). Accordingly the knowledge of bubble
size distributions is important. The wire-mesh sensor technology allows the mea-
surement of such bubble size distributions as discussed above. For fully developed
air-water flows, a mono-modal distribution is observed for flow with a maximum gas
volume fraction of about 10%. For larger gas volume fractions, the bubble size
distribution is usually bi-model with one peak in the range of 6–10 mm volume
equivalent bubble diameter which sharply decreases (especially for the smaller
diameter pipe) and a second peak which represents bubbles stabilized by the
stabilizing effect of the pipe as shown in Fig. 11. More examples for bubble size
distributions will be given in the next sections.
Evolution of the Air-Water Flows
Beside the fully developed flow, the experiments discussed here provide valuable
data on the evolution of the flow. In the L12 experiments, gas is injected via orifices
in the pipe wall. For this reason close to the injection (i.e., at small L/D), the radial
gas volume fraction profile always shows a wall peak. With increasing distance from
the injection, there is a redistribution of the bubbles depending on flow rates and
bubble sizes as shown in Fig. 16. The letters in the legend refer to the different
distances between gas injection and measuring plane according to Fig. 4. In both
cases presented in this figure, the bubbles start to distribute over the cross section of
the pipe. For the case with the lower gas flow, a wall peak is maintained for all
Experiments on Gas-Liquid Flow in Vertical Pipes
Fig. 15 Decomposed radial volume fraction profiles: upper line left, L20 experiments, JL = 1.611 m/s; JG = 0.219 m/s, L/D = 151.2; right, L12, JL = 1.017 m/s;
23
JG = 0.219 m/s, L/D = 39.9, injection via 1 mm orifices; lower line left, K16, 2 MPa, JL = 1.017 m/s; JG = 0.219 m/s, L/D = 39.9, injection via 1 mm orifices;
right, K16, 6.5 MPa, JL = 1.017 m/s; JG = 0.0898 m/s, L/D = 39.7, injection via 4 mm orifices
24 D. Lucas et al.
Fig. 16 Evolution of the radial gas volume fraction profile for the L12 experiments DOrifice = 1 mm;
top, JL = 1.017 m/s; JG = 0.0062 m/s; bottom, JL = 1.017 m/s; JG = 0.219 m/s
distances, while a core peak of the gas volume fraction is established for the larger
gas flow rate.
This again is caused by the action of the lateral lift force, and the effects fit well
with the evolution of the corresponding bubble size distributions shown in Fig. 17.
For the low gas flow rate, there is a slight increase of the bubble sizes along the pipe
which is mainly caused by the expansion caused by the decreasing pressure with
height position. However, most bubbles remain below 5.8 mm sphere equivalent
diameter for all positions resulting in the wall peak. Since the fraction of larger
Fig. 17 Evolution of the bubble size distributions for the L12 experiments DOrifice = 1 mm; top,
JL = 1.017 m/s; JG = 0.0062 m/s; bottom, JL = 1.017 m/s; JG = 0.219 m/s
bubbles increases, the wall peak becomes smaller, and the gas volume fraction in the
core region increases with the distance from the gas injection.
For the high gas flow rate, the bigger part of the gas is transported by bubbles
larger than 5.8 mm, and thus most of these bubbles migrate toward the pipe center.
The evolution of the bubble size distributions shows an increase of the small bubble
peak caused by breakup processes, while the large bubbles further increase by
coalescence. The data are used for the development and validation of improved
models for bubble coalescence and breakup; see, e.g., Liao et al. (2011, 2015).
26 D. Lucas et al.
Fig. 18 Evolution of the radial gas velocity profile for the L12 experiments DOrifice = 1 mm; top,
JL = 1.017 m/s; JG = 0.0062 m/s; bottom, JL = 1.017 m/s; JG = 0.219 m/s
The profiles of the gas velocity shown in Fig. 18 are mainly influenced by the gas
volume fraction profiles. The wall peak leads to a flattening of the liquid velocity
profile and in the result also of the gas velocity profile. Accordingly with the
decreasing height of the wall peak, the velocity profiles become more and more
pronounced along the pipe for the case with lower gas flow rate. Because of the high
gas volume fraction in the case with higher gas flow rate, the characteristics of the
gas velocity profile follow the gas volume fraction profile. In the region above the
injection, the gas is still close to the wall that also leads to a maximum gas (and
liquid) velocity in the wall region. Along the pipe, there is a transition to a core
peaked velocity profile in accordance with the gas volume fraction profiles.
It is worth to mention that all the measured data are highly consistent with each
other. Even for the small distances between the injection chambers with 1 mm
orifices from one injection device (see Fig. 4), the evolution of the flow always
show a clear trend in all the measurements (solid and dashed curves in the figures
above).
Interfacial Area Density (IAD)
The IAD is an important parameter in two-phase flows with mass transfer. Conden-
sation and evaporation rates in dynamic flows or heterogeneous chemical reactions
are proportional to this parameter. As mentioned before, the interfacial area is
determined by a special algorithm for each single bubble. In the following, only
some general trends will be discussed for the example of the L12 experiments.
To get an overview on the dependency of the IAD ai on flow rates, data for
developed flow, i.e., for largest length-to-diameter ratio (L/D) which is about 40, are
presented first in Fig. 19. The left-hand side shows the results for the two lines of the
experimental matrix (see Fig. 5) with constant liquid superficial velocities JL of
about 0.4 m/s and 1.0 m/s and increasing gas superficial velocity JG, while the right-
hand side shows the two lines with constant JG of about 0.01 m/s and 0.22 m/s and
increasing JL. It can be seen at the upper left picture that the IAD first increases with
JG (in the regions of bubbly flow and the transition to churn-turbulent flow) then
reaches a maximum in the churn-turbulent flow regime and starts to decrease as soon
as the transition to wispy-annular and annular flow regimes plays a role. For
JG = 0.01 m/s, the IAD deceases with increasing JL but remains almost constant
for JG = 0.22 m/s which is mainly in the churn-turbulent flow regime. This general
behavior corresponds to the expectations since in the bubbly flow regime, more and
more bubbles occur with increasing JG, while for churn-turbulent flow, more and
more large gas structures are formed which contribute only little to the gas-liquid
interfacial area. Finally in an annular flow, the gas-liquid interface is small.
For bubbly flow with monodispersed spherical bubbles, the IAD is related to the
gas volume fraction ε and the bubble diameter db according to
6e
ai ð d b Þ ¼ , (1)
db
i.e., for such a flow, the IAD should be proportional to the void fraction. In case of
monodisperse spherical bubbles with 5 mm diameter, the IAD to a ratio would be
12 (%*m)1. In the middle row of Fig. 19, the IAD related to the void fraction is
presented. The graphs clearly shows two different parts – an almost constant part
with values between 11 and 14 (%*m)1 which is characteristic for bubbly flows and
a decreasing part for churn-turbulent flows. For the churn-turbulent flows, the
increase of IAD is less than that of ε.
28 D. Lucas et al.
Fig. 19 IAD (top), IAD related to void fraction (middle), and relation between real IAD and the
values obtained from the bubble size distribution assuming spherical bubbles (bottom). Left,
experiments with constant JL; right, experiments with constant JG. Dashed lines (middle pictures)
present the IAD related to the void fraction under the assumption of spherical bubbles
The graphs for the two different injection orifices collapse to one graph starting
from about JG = 0.04 m/s for JL = 1 m/s and from about JG = 0.01 m/s for
JL = 0.4 m/s. Below these gas flow rates, there is still a dependency of the
IAD-to-ε ratio on the injection. The smaller bubbles in case of the 1 mm injection
lead to higher values of this ratio compared to the 4 mm injection. For constant JG
shown at the right picture in the middle row of Fig. 19, the trends that the ratio is in
the range between 11 and 14 for the bubbly flow region but clearly smaller for the
churn-turbulent flow regime are confirmed.
Fig. 20 IAD (top), IAD related to void fraction (middle), and relation between real IAD and the
values obtained from the bubble size distribution assuming spherical bubbles (bottom). Selected
matrix points. Left, JL = 0.4 m/s; right, JL = 1.0 m/s
The pictures in the middle row also show the IAD-to-ε ratio obtained from the
bubble size distributions assuming spherical bubbles (dashed lines). They show
quite similar trends but are smaller for all cases. This has to be expected since
spheres represent the smallest possible surface. To quantify the deviation from the
assumption of spherical bubbles, the ratio between IAD obtained with the new
algorithm and the one obtained from the bubble size distribution assuming spherical
bubbles is shown at the bottom of Fig. 19. For bubbly flow, this ratio lies between
30 D. Lucas et al.
@(40 mm) A(304 mm) D (577 mm) G (1521 mm) J (2564 mm) M (4500mm) B (361 mm) E (634 mm) H (1578 mm) K (2621 mm) N (4557 mm)
Fig. 21 Virtual side projects (left columns) and central cuts (right columns) for different distances
from the steam injection. JL = 1.017 m/s, JG = 0.534 m/s, 1 MPa, sub-cooling 3.7 K; left, injection
trough 1 mm orifices; right, injection trough 4 mm orifices
1 and 1.5, while it becomes much larger for churn-turbulent and (wispy-)annular
flow.
The evolution of the IAD along the pipe is presented in Fig. 20 for the same
parameter as discussed above for the matrix points 019, 074, 107, 129, and 184 (left
column) for JL = 1 m/s and the corresponding points (017, 072, 105, 127, and 182)
for the JL = 0.4 m/s line. Only for the case with the lowest JG shown in these figures,
a decrease of the IAD-to-ε ratio is observed in the injection region. This corresponds
to the measured bubble size distribution which only shows for this case a dominance
of bubble coalescence. For the other cases, because of the injection of relatively large
bubbles, the flow behind the gas injection is dominated by bubble breakup related to
an increase of the IAD-to-ε ratio. As the lower pictures in Fig. 20 demonstrate, the
bubbles are highly deformed in the vicinity of the injection. This deformation
decreases along the pipe due to bubble breakup.
Also the radial profiles of the IAD show very similar shapes like the gas volume
fraction profiles. The IAD-to-ε ratio always shows a wall peak for all cases of
developed flow (L/D = 40). This is due to the decreasing averaged bubble size
from the pipe center toward the wall.
Steam Bubble Condensation in Sub-cooled Liquid
A detailed discussion on the results of the K16 experiments is given by Lucas et al.
(2010c, 2013).
Beside the variations listed in the test matrix, Table 2, the 1 mm and 4 mm orifices
were used for the injection. Here again the 4 mm orifices lead to larger bubbles, i.e.,
to a lower interfacial area density comparing both injections for the same flow rates,
sub-cooling, and pressure. As an example, Fig. 21 compares the evolution of the
Fig. 22 Evolution of the cross section averaged bubble size distributions. Case 118 (JL = 1,017 m/s,
JG = 0,219 m/s), 2 MPa, sub-cooling 3.7 K, top 1 mm orifices, bottom 4 mm orifices
flow along the pipe for case 140 at 1 MPa pressure and a sub-cooling of 3.7 K
qualitatively.
32 D. Lucas et al.
Fig. 23 Evolution of the time

and cross section averaged gas
volume fraction along the pipe.
Case 118 (JL = 1,017 m/s,
JG = 0,219 m/s), 4 MPa
For both cases, the experimental conditions are exactly the same, but the size and
number of injection orifices are different. This is also true for the quantitative results
shown in Fig. 22. Clearly the bubbles are much larger in case of the 4 mm orifices.
Figure 23 shows the evolution of the time and cross section averaged gas volume
fraction for three different values of the initial sub-cooling. Of course the conden-
sation rate increases with the sub-cooling leading to a faster decrease of the steam
volume fraction. In addition, the abovementioned effect of the lower IAD in case of
the 4 mm injection clearly leads to lower condensation rates resulting in a slower
decrease compared to the 1 mm injection for the same boundary conditions. For the
case with the lowest sub-cooling saturation, conditions are achieved at about one half
of the pipe length. For this reason, condensation stops, and a slight reevaporation can
be observed.
This effect is more pronounced in the case shown in Fig. 24. On the one hand, the
condensation of the injected steam heats up the water, and on the other hand, the
pressure and accordingly also the saturation temperature decrease along the pipe.
Again the condensation rate is higher in case of the 1 mm injection for low L/D. In
both cases, a minimum gas volume fraction of about 15% is reached. After that there
is an increase of the gas volume fraction caused by reevaporation. The temperature
measurement obtained by one lance of thermocouples which was mounted directly
above the wire-mesh sensor and which spans over the whole pipe diameter confirms
these phenomena, in principle. The saturation temperature was calculated from the
pressure measured at the sensor position. First the water temperature increases, but
starting from about L/D = 12, the temperature again decreases in accordance to the
saturation temperature. The interfacial area has an important influence as long a clear
sub-cooling occurs. It is important to mention that the water temperature should be
slightly larger than the saturation temperature. That fact that this is not true for the
example shown in Fig. 24 can be attributed to the uncertainties of the temperature
measurement. As soon as the saturation temperature is reached, the fluid is in thermal
Fig. 24 Evolution of the time and cross section averaged void fraction (top) and averaged water
temperature (bottom) along the pipe. Case 140 (JL = 1.017 m/s, JG = 0.534 m/s), 2 MPa and 3.2 K
sub-cooling
equilibrium, and the interfacial area density should only determine the amount of
small overheat mentioned before.
Figure 25 demonstrates the influence of pressure on the condensation rate.
Because of the increase of the steam density with pressure, condensation along all
the pipe only occurs at 1.0 MPa, while reevaporation occurs for higher pressure.
An example for the evolution of the radial gas volume fraction profiles and gas
velocity profiles is shown in Fig. 26. For small L/D, the gas velocity (and
34 D. Lucas et al.
Fig. 25 Evolution of the time and cross section averaged void fraction (left) and averaged water
temperature (right) along the pipe. Case 140 (JL = 1.017 m/s, JG = 0.534 m/s), 5.0 K sub-cooling
accordingly the liquid velocity) is larger in the near wall region compared to the pipe
center. As already discussed for the L12 experiments, this is caused by the increase
of the vertical component of the liquid velocity caused by the injected gas. With
increasing L/D, the bubbles start to migrate toward the pipe center but are condens-
ing at the same time. The velocity profiles develop a center-peaked profile.
The data were used for the validation of CFD models for condensing polydisperse
bubble flows; see, e.g., Krepper et al. (2011) and Liao et al. (2014).
Transient Data from Pressure Relief Experiments
The time-plots of the pressure clearly differ for the two procedures used in the
pressure relief experiments (see section “PR17 Test Series”). Due to the slow
opening of the blow-off valve in case of procedure 1, the pressure transients are
rather smooth (Fig. 27, left). For procedure 2 (Fig. 27, right), a sharp decrease of
pressure occurs immediately after the opening of the valve. This leads to a consid-
erable sub-cooling of the liquid followed by a strong evaporation process. During a
short period, the volume of steam generated by evaporation exceeds to discharge
volume leading to an increase of pressure. After closing the valve, an increase of
pressure is observed which is more pronounced at the 1 MPa experiment compared
to the 6.5 MPa experiment. It is caused by the heat input from the walls.
In the following, the experiment for procedure 1 at 6.5 MPa and 25% opening of
the blow-off valve is discussed more in detail. The opening of the blow-off valve
starts at 2 s, i.e., according to the numbers given in Fig. 7, the valve is opened to the
Fig. 26 Evolution of the radial gas volume fraction profile (top) and the radial gas velocity profile
(bottom). Case 140 (JL = 1.017 m/s, JG = 0.534 m/s), 4 MPa, 7.6 K sub-cooling, injection via 4 mm
orifices
desired value (25%) at 11 s and closing starts at 63 s. The valve is closed completely
at 72 s.
Figure 28 shows the cross section averaged void fraction obtained from the wire-
mesh sensor measurement as function of time. The delay of the increase and decrease
of the void fraction compared to the opening ramp of the valve results from the delay
of the evaporation process but mainly from the fact that the measuring plane is
36
Fig. 27 Time-dependent pressure for the two procedures; left, procedure 1; right, procedure 2; top, 1 MPa; bottom, 6.5 MPa. The numbers on brackets indicate
the first and second realization of the test
D. Lucas et al.
Fig. 28 Cross section averaged void fraction for procedure 1, 6.5 MPa and 25% opening of the
blow-off valve
located at the upper end of the pipe. The steam which is produced along the pipe
needs some time to travel to the sensor.
This fact is also reflected in the evolution of the bubble size distribution with time
as shown in Fig. 29. With the increase of the averaged void fraction, also the bubble
sizes increase (Fig. 29, top, left). During the period in which a plateau of the
averaged void fraction is observed, also the bubble size distributions remain almost
unchanged (Fig. 29, top, right). The bubble sizes decrease with the averaged void
fraction after closing the blow-off valve (Fig. 29, bottom).
The boiling up during the pressure relief process is also reflected in the radial
profiles of the gas velocity which are shown in Fig. 30. Because the first bubbles are
generated at the pipe wall (see Fig. 31, top, left) in the first seconds after the start of
the blowdown, the velocity increases first only in the near wall region (up to 18 s in
Fig. 30, left). Later on the maximum of the gas volume fraction shifts away from the
pipe wall and forms intermediate peaks (Fig. 31, top line – 23–69 s) followed by a
center peak from 71 to 83 s. The velocity profiles have their maxima in the pipe
center for t > 29 s. They are again flattened with the decrease of the boiling process
after closing of the blow-off valve. Starting from about 90 s, bubbles are observed
only in the near wall region (Fig. 31, bottom, right). Compared to steady-state
experiments for air-water flows and condensing steam-water flow as discussed in
the previous sections in which the radial profiles and bubble size distributions were
obtained from an averaging over the whole measuring time (10 s), here the statistics
of the data are not so good. The averaging over period of only 2 s leads to some
fluctuations of the profiles as it can be seen in Fig. 31. The fluctuations are more
pronounced in the core region of the pipe compared to the wall region due to the
38
Fig. 29 Bubble size distributions during the transient, procedure 1, 6.5 MPa, 25% opening of the blow-off valve
D. Lucas et al.
Fig. 30 Radial profiles of the gas velocity during the transient, procedure 1, 6.5 MPa, 25% opening
of the blow-off valve
occurrence of few large bubbles and the lower statistics in central positions com-
pared to larger radii caused by the equal width of the radial rings (see section “Wire-
Mesh Sensors and Data Evaluation Procedures”).
During the period of the plateau of the averaged void fraction in Fig. 28, almost
stationary conditions are observed in respect to bubble size distributions (Fig. 29,
top, right), radial profiles of gas velocity (Fig. 30, right), and radial gas volume
fraction profiles (Fig. 31, top, right). For this reason, it should be justified to do a
40
Fig. 31 Radial profiles of the gas volume fraction during the transient, procedure 1, 6.5 MPa, 25% opening of the blow-off valve
D. Lucas et al.
Fig. 32 Radial volume fraction profiles decomposed according to the bubble size and averaged
over the gas volume fraction plateau (31–61 s) , procedure 1, 6.5 MPa, 25% opening of the blow-off
valve
Fig. 33 Volume averaged void fraction for different pipe sections for procedure 1, 6.5 MPa and
25% opening of the blow-off valve. The sections are characterized by length to diameter ratio (L/D)
measured from the axial position of the wire-mesh sensor
time averaging over this period in order to improve the statistics. This is especially
important for the radial volume fraction profiles decomposed according to the
bubbles size. Such profiles are presented in Fig. 32. Obviously the peak at half of
the pipe radius in the total gas volume fraction profiles shown in Fig. 31, top right, is
42 D. Lucas et al.
caused mainly by bubbles larger than 10 mm sphere equivalent bubble diameter.

Smaller bubbles are more or less equally distributed over the pipe cross section.
The measured data for the pressure difference recorded for several height posi-
tions were used to obtain some information on the axial void distribution along the
pipe. Results are shown in Fig. 33. Considerable uncertainties arise for these
measurements from the fluctuations in signals of the measured pressure differences
and from the correlations used for two-phase pressure drop due to friction and
acceleration. Compared to the averaged void fraction measured by the wire-mesh
sensor (see Fig. 28), some lower values are obtained for the topmost section. In
contrast to experiments done at 1 MPa in case of the 6.5 MPa experiment, steam is
also observed at the lowest section between the measurement positions for the
pressure difference, as shown in Fig. 33.
This observation agrees with the temperature measurement at the lower end of the
test section. Before the opening of the valve, the water temperature is slightly below
the saturation temperature which corresponds to the pressure measured at this
position. After the start of the blow-off, the measured temperature and saturation
temperature agree quite well.
The data that were from this experimental series were used for the validation of
one-dimensional system codes (Mikuz et al. 2015) as well as for the validation of
CFD codes (Liao et al. 2013).
Conclusion
This chapter presents experimental setups on gas-liquid vertical pipe flow, explains
which quantitative data can be obtained using the wire-mesh sensor technology, and
discusses observations regarding two-phase flow characteristics obtained in these
experiments. Flows without phase transfer and in condensing and evaporating flows
are considered. Results from two different pipe sizes are discussed, and the exper-
imental boundary conditions were varied over a wide range to obtain different flow
pattern. The latter is not only dependent on the gas and liquid volume flow rates but
also on pipe diameter and often initial bubble size distribution which depends on the
gas injection device. The reason lies in the fact that the evolution of bubbly flows is
strongly influenced by the bubble size distribution. It is essential to consider this fact
in a proper simulation of such flows. There is a complex interaction between local
gas-liquid phase fractions and the local bubble size distribution which changes
caused by bubble coalescence and break.
The situation becomes even more complex for flows with phase transfer. The
interfacial area density was shown to have an important influence on the dynamics of
the condensation of steam bubbles injected into sub-cooled water. The evolution of
such flows along the pipe also depends on flow rates, pressure, and sub-cooling.
Evaporation was initiated in the presented experiments by pressure relief. Transient
boiling processes were investigated by limited valve opening times. In case of flows
with phase transfer, the bubble size distribution changes because of phase transfer in
addition to the complex coalescence and breakup processes.
Beside qualitative insights in gas-liquid flows, a comprehensive database for CFD

code development and validation is obtained.
In this book chapter, wire-mesh sensor measurements were considered. It was
shown that it is an innovative technology which is able to provide valuable insights
into two-phase flows. However it is an intrusive measuring technique, i.e., it
influences the flow. For this reason, the range of applicability is limited. For the
cases discussed here, always a basic liquid flow with a superficial velocity of at least
0.2 m/s was present. This limits the influence of the interaction between the wires
and bubbles on the results of the measurements because the bubbles are pushed
through the sensor by some momentum. Detailed discussion on accuracy for such
flows can be found in the cited references giving more details on the experiments.
For noninvasive measurements, radiation-based methods can be applied.
Recently the ultrafast electron-beam X-ray tomography was developed at HZDR
(see chapter “▶ Imaging Measuring Techniques”). Presently is used to obtain similar
data as discussed in this chapter for a wider range of flow conditions including
counter-current air-water vertical pipe flow and cocurrent downward air-water and
steam-water pipe flows (Banowski et al. 2015). Since there is no interaction between
the measurement and the flow, bubbles measured in the upstream measuring plane
can be found with a high probability also in the downstream measuring plane. This
allows to determine velocities of single bubbles. There is a high potential for getting
new insights into the flow characteristics and getting better quantitative data from
these experiments.
Acknowledgments This work is carried out in the frame of a research project funded by the
German Federal Ministry of Economic Affairs and Energy, project numbers 150 1329 and
150 1411. The authors like to thank especially Prof. Dr. Horst Michael Prasser who planned the
experiments discussed here and developed the wire-mesh sensor technology as well as all members
of the TOPFLOW team who contributed to the successful performance of these experiments.
References
M. Banowski, D. Lucas, L. Szalinski, A new algorithm for segmentation of ultrafast X-ray
tomographed gas-liquid flows. Int. J. Therm. Sci. 90, 311–322 (2015)
M. Beyer, D. Lucas, J. Kussin, P. Schütz, Air-water experiments in a vertical DN200-pipe. Report
Forschungszentrum Dresden-Rossendorf (2008), http://www.hzdr.de/publications/PublDoc-
5374.pdf
M. Beyer, D. Lucas, J. Kussin, Quality check of wire-mesh sensor measurements in a vertical
air/water flow. Flow Meas. Instrum. 21, 511–520 (2010)
D. Bothe, M. Schmidtke, H.-J. Warnecke, Direct numerical computation of the lift force acting on
single bubbles. 6th International Conference on Multiphase Flow, ICMF 2007, 09.-13.07.2007,
Leipzig, Germany (2007)
X.Y. Duan, S.C.P. Cheung, G.H. Yeoh, J.Y. Tu, E. Krepper, D. Lucas, Gas–liquid flows in medium
and large vertical pipes. Chem. Eng. Sci. 66, 872–883 (2011)
E.A. Ervin, G. Tryggvason, The rise of bubbles in a vertical shear flow. J. Fluids Eng. 119, 443–449
(1997)
polydisperse air-water two-phase flows in pipes. Nucl. Eng. Design 238, 647–659 (2008)
44 D. Lucas et al.
R. Kaji, B.J. Azzopardi, D. Lucas, Investigation of flow development of co-current gas-liquid

vertical slug flow. Int. J. Multiphase Flow 35, 335–348 (2009)
E. Krepper, D. Lucas, H.-M. Prasser, On the modelling of bubbly flow in vertical pipes. Nucl. Eng.
Design 235, 597–611 (2005)
simulation of polydispersed flows. Nucl. Eng. Design 238, 1690–1702 (2008)
E. Krepper, M. Beyer, D. Lucas, M. Schmidtke, A population balance approach considering heat
and mass transfer – experiments and CFD simulations. Nucl. Eng. Design 241, 2889–2897 (2011)
E. Krepper, R. Rzehak, C. Lifante, T. Frank, CFD for subcooled flow boiling: coupling wall boiling
and population balance models. Nucl. Eng. Design 255, 330–346 (2013)
Y. Liao, D. Lucas, E. Krepper, M. Schmidtke, Development of a generalized coalescence and
breakup closure for the inhomogeneous MUSIG model. Nucl. Eng. Design 241, 1024–1033 (2011)
Y. Liao, D. Lucas, E. Krepper, R. Rzehak, Flashing evaporation under different pressure levels.
Nucl. Eng. Design 265, 801–813 (2013)
Y. Liao, D. Lucas, E. Krepper, Application of new closure models for bubble coalescence and
breakup to steam-water vertical pipe flow. Nucl. Eng. Design 279, 126–136 (2014)
Y. Liao, R. Rzehak, D. Lucas, E. Krepper, Baseline closure model for dispersed bubbly flow: bubble
coalescence and breakup. Chem. Eng. Sci. 122, 336–349 (2015)
D. Lucas, H.-M. Prasser, Steam bubble condensation in sub-cooled water in case of co-current
vertical pipe flow. Nucl. Eng. Design 237, 497–508 (2007)
D. Lucas, A. Tomiyama, On the role of the lateral lift force in poly-dispersed bubbly flows. Int.
D. Lucas, E. Krepper, H.-M. Prasser, Prediction of radial gas profiles in vertical pipe flow on basis
of the bubble size distribution. Int. J. Therm. Sci. 40, 217–225 (2001)
D. Lucas, E. Krepper, H.-M. Prasser, Development of co-current air–water flow in a vertical pipe.
Int. J. Multiphase Flow 31, 1304–1328 (2005)
D. Lucas, E. Krepper, H.-M. Prasser, Use of models for lift, wall and turbulent dispersion forces
D. Lucas, M. Beyer, L. Szalinski, Experimental investigations on the condensation of steam bubbles
injected into sub-cooled water at 1 MPa, Multiphase. Sci. Technometrics 22, 33–55 (2010a)
D. Lucas, M. Beyer, J. Kussin, P. Schütz, Benchmark database on the evolution of two-phase flows
in a vertical pipe. Nucl. Eng. Design 240, 2338–2346 (2010b)
D. Lucas, M. Beyer, L. Szalinski, P. Schütz, A new database on the evolution of two-phase flows in
a large vertical pipe. Int. J. Therm. Sci. 49, 664–674 (2010c)
D. Lucas, T. Frank, C. Lifante, P. Zwart, A. Burns, Extension of the inhomogeneous MUSIG model
for bubble condensation. Nucl. Eng. Design 241, 4359–4367 (2011)
D. Lucas, M. Beyer, L. Szalinski, Experimental database on steam-water flow with phase transfer in
a vertical pipe. Nucl. Eng. Design 265, 1113–1123 (2013)
A. Manera, H.-M. Prasser, T.H.J.J. van der Hagen, Suitability of drift-flux models, void-fraction
evolution and 3D flow pattern visualization during stationary and transient flashing flow in a
vertical pipe. Nucl. Technol. 152, 38–53 (2005)
B. Mikuz, I. Tiselj, M. Beyer, D. Lucas, Simulations of flashing experiments in TOPFLOW facility
with TRACE code. Nucl. Eng. Design 283, 60–70 (2015)
S. Nuryadin, M. Ignaczak, D. Lucas, D. Deendarlianto, On the accuracy of wire-mesh sensors in
dependence of bubble sizes and liquid flow rates. Exp.Thermal Fluid Sci. 65, 73–81 (2015)
H. Pietruske, H.-M. Prasser, Wire-mesh sensor for high-resolving two-phase flow studies at high
pressures and temperatures. Flow Meas. Instrum. 18, 87–94 (2007)
H.-M. Prasser, Evolution of interfacial area concentration in a vertical air-water flow measured by
wire-mesh sensors. Nucl. Eng. Design 237, 1608–1617 (2007)
H.-M. Prasser, A. Böttger, J. Zschau, A new electrode-mesh tomograph for gas/liquid flows. Flow
Meas. Instrum. 9, 111–119 (1998)
H.-M. Prasser, D. Scholz, C. Zippe, Bubble size measurement using wire-mesh sensors. Flow Meas.
Instrum. 12, 299–312 (2001)
H.-M. Prasser, E. Krepper, D. Lucas, Evolution of the two-phase flow in a vertical tube –
decomposition of gas fraction profiles according to bubble size classes using wire-mesh sensors.
Int. J. Therm. Sci. 41, 17–28 (2002)
H.-M. Prasser, M. Misawa, I. Tiseanu, Comparison between wire-mesh sensor and ultra-fast X-ray
tomograph for an air/water flow in a vertical pipe. Flow Meas. Instrum. 16, 73–83 (2005a)
H.-M. Prasser, M. Beyer, A. Böttger, H. Carl, D. Lucas, A. Schaffrath, P. Schütz, F.-P. Weiß,
J. Zschau, Influence of the pipe diameter on the structure of the gas-liquid interface in a vertical
two-phase pipe flow. Nucl. Technol. 152, 3–22 (2005b)
H.-M. Prasser, M. Beyer, H. Carl, A. Manera, H. Pietruske, P. Schütz, F.-P. Weiß, The multipurpose
thermal-hydraulic test facility TOPFLOW: an overview on experimental capabilities, instru-
mentation and results. Kerntechnik 71, 163–173 (2006)
H.-M. Prasser, M. Beyer, H. Carl, S. Gregor, D. Lucas, H. Pietruske, P. Schütz, F.-P. Weiss,
Evolution of the structure of a gas–liquid two-phase flow in a large vertical pipe. Nucl. Eng.
Design 237, 1848–1861 (2007)
F.S. Qi, G.H. Yeoh, S.C.P. Cheung, J.Y. Tu, E. Krepper, D. Lucas, Classification of bubbles in
vertical gas–liquid flow: part 1 – an analysis of experimental data. Int. J. Multiphase Flow 39,
121–134 (2012)
A. Schaffrath, A.-K. Krüssenberg, F.-P. Weiß, E.-F. Hicken, M. Beyer, H. Carl, H.-M. Prasser,
J. Schuster, P. Schütz, M. Tamme, TOPFLOW – a new multipurpose thermalhydraulic test
facility for the investigation of steady state and transient two-phase flow phenomena.
Kerntechnik 66, 209–212 (2001)
S. Sharaf, B. Azzopardi, U. Hampel, C. Zippe, M. Beyer, M.J. Da Silva, Comparison between wire
mesh sensor technology and gamma densitometry. Meas. Sci. Technol. 22, 104019 (2011)
L. Szalinski, M.J. Da Silva, S. Thiele, M. Beyer, D. Lucas, U. Hampel, V. Hernandez Perez,
L.A. Abdulkareem, B.J. Azzopardi, Comparative study of gas-oil and gas-water two-phase flow
in a vertical pipe. Chem. Eng. Sci. 65, 3836–3848 (2010)
Y. Taitel, D. Bornea, A.E. Duckler, Modelling flow pattern transitions for steady upward gas-liquid
flow in vertical tubes. AICHE J. 26, 345–354 (1980)
A. Tomiyama, H. Tamai, I. Zun, S. Hosokawa, Transverse migration of single bubbles in simple
shear flows. Chem. Eng. Sci. 57, 1849–1858 (2002)
Multiphase Flows in Biomedical
Applications
Jingliang Dong, Kiao Inthavong, and Jiyuan Tu
Abstract
A sound understanding of the physics of multiphase flow is important for
studying biofluid dynamics in human, and the air flow in the respiratory system
and blood flow in the cardiovascular system remain two of the most important
fields. This chapter presents case studies covering 3D model reconstruction, gas-
particle modelling in the human respiratory system and liquid-particle flow
modelling in the human arterial capillary by reviewing the current multiphase
modelling techniques and its challenges. The potential benefits of using compu-
tational fluid dynamics (CFD) in human biofluids modelling are demonstrated.
Keywords
Modelling • Gas particle flow • Blood flow • Nasal cavity • Capillary • Biofluid •
Inhalation exposure • Drug delivery • Red blood cells
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Multiphase Flow in Human Respiratory System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Multiphase Flow in Human Cardiovascular System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Model Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Case Study: Gas-Particle Flow in Human Nasal Cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Nasal Cavity Reconstruction and Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Inhalation and Particle Deposition Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Inhalation Flow Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Particle Deposition in the Human Nasal Cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Case Study: Nasal Spray Efficiency Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Numerical Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Inhalation Airflow Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
J. Dong (*) • K. Inthavong • J. Tu

School of Engineering, RMIT University, Bundoora, VIC, Australia
e-mail: jingliang.dong@rmit.edu.au; kiao.inthavong@rmit.edu.au; jiyuan.tu@rmit.edu.au

DOI 10.1007/978-981-4585-86-6_16-1
2 J. Dong et al.
Spray Trajectory Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Particle Deposition Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Case Study: Modelling of Blood Flow in Capillary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
CFD Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Flow Field and Mean Velocity Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Introduction
Airflow in the respiratory system and blood flow in the cardiovascular system remain
two of the most important fields within biomedical engineering. As these flows
constitute biofluids either in liquid or gaseous phase, fluid dynamics principles can
solve the biofluid mechanics to support the diagnosis and decision-making for
treatment of clinical pathologies. Exploring the relationship between flow phenom-
ena and pathophysiological observations is enhanced by continual advances in
imaging modalities, measurement techniques, and computational modelling capa-
bilities (Siebes and Ventikos 2010). In particular, computational fluid dynamics
(CFD) is a compelling, nonintrusive, virtual modelling technique with powerful
visualization capabilities that enables engineers to solve complex airflow in realistic
airways (Inthavong et al. 2006, 2008; Wen et al. 2008; Shang et al. 2015) or blood
flow in arterial vessels (Antiga et al. 2008; Dong et al. 2013, 2015).
This chapter presents case studies covering gas-particle modelling in the human
respiratory system and liquid-particle flow modelling in the human arterial capillary
by reviewing the current multiphase modelling techniques and its challenges. This
chapter can serve as an illustration of the necessary modelling framework in numer-
ical applications. Combined with experimental measurements and clinical observa-
tions, multiphase CFD modelling can address unmet clinical needs, predominantly
in the direction of enhanced diagnosis, as well as assessment and prediction of
treatment outcomes.
Multiphase Flow in Human Respiratory System
Based on function, the respiratory system can be divided into the conducting zone
(nose to bronchioles) and respiratory zone (alveolar duct to alveoli) shown in Fig. 1.
Alternatively, it can be divided based on anatomy leading to the upper (all structures
before larynx) and lower respiratory tract. As the first passageway for air entering the
body, the nasal cavity acts a critical role in the respiratory system, such as air
conditioning, hazardous airborne particles filtration, and the sense of olfaction.
The multiphase flow in this field is classified as gas-particle flows occurring
during inhalation. Dust particles, pollen, and spray droplets that exist in the micron
size range can be treated as discrete particles, which are generally visible and tend to
deposit on respiratory surfaces by inertial impaction (Kelly et al. 2005; Frank et al.
Multiphase Flows in Biomedical Applications 3
Fig. 1 Schematic of the respiratory system divided by anatomy into the upper and lower respira-
tory tract
2012). Finer particles such as fumes, gases, and smoke that exist in the submicron
range are less visible exhibiting low Stokes numbers, and its deposition is considered
to diffusion dominant (Cheng et al. 1990; Kreyling et al. 2006).
Occupational hazards associated with exposure to micron and submicron parti-
cles are widely recognized in work environments (Yokel and MacPhail 2011). For
example, workers exposed to manganese-containing welding fumes may face haz-
ardous risks associated with translocation of ultrafine manganese oxide particles
(<100 nm) to the central nervous system (Elder et al. 2006). Healthcare workers may
be exposed to patients with airborne infectious agents in the form of fine suspended
droplets. Other pathogens include fungi which can be in the size range of 2–200 μm,
while viruses groups can be as small as 0.02–0.3 μm (Weber et al. 1996; Johnson
et al. 2011). Studies have developed a measure of the likelihood of micron particles
being inhaled through the nose and mouth from the ambient air, and it is termed
particle inhalability (Menache et al. 1995). The inhalability decreased from a
maximum for 1 μm particle size to a minimum for 140 μm particles. The high
inhalability for smaller particles within the range of 1–5 μm subsequently leads to
deeper particle penetration into the airways, while even small particles in the
4 J. Dong et al.
Fig. 2 The human body

vasculature network
submicron range travel down further to the lower lungs including the alveoli
(Darquenne and Prisk 2004; Tsuda et al. 2013).
While the filtration function of the nose protects the lower respiratory tract from
harmful pollutants, there are cases where it is desirable to purposely introduce
particles into the nose. Nasal administration through sprayed or atomized intranasal
medication delivery (majority of the particles are in micron size range) is an alternate
drug delivery route that provides advantages over oral, dermal, or intravenous
routes. The drug molecule can be transferred quickly across the single epithelial
cell layer to the systemic circulation without first-pass hepatic and intestinal metab-
olism since the nasal cavity is covered by a thin mucosa (which is highly
vascularized). The nasal route provides rapid onset of therapeutic effects, no lung
toxicity, is noninvasive, and is essentially painless (Wolfe and Braude 2010).
Therefore, knowledge regarding particle deposition processes in the nasal cavity
is important in both inhalation toxicology and aerosol therapy fields (Wang et al.
2009). To demonstrate potential benefits of using computational fluid dynamics
(CFD) approach in the occupational health assessment and the drug delivery effi-
ciency analysis, multiphase (gas-particle) flow modelling studies are presented in
this chapter.
Multiphase Flow in Human Cardiovascular System
The cardiovascular system is an interconnection of arteries and veins that branch at

multiple levels to reach all parts of the human body via an intricate network of
vessels (Fig. 2). This interconnected and closed loop system transports oxygenated
red blood cells to tissues and organs of the human body, while deoxygenated red
blood cells are returned to the respiratory organs.
In recent years, CFD simulations based on single-phase assumptions, combined
with noninvasive medical imaging, have been used increasingly by researchers
seeking to better understand hemodynamic factors (Berthier et al. 2002; Dai et al.
2004; Gijsen et al. 1999; Hoi et al. 2004; Myers et al. 2001; Taylor and Draney
2004). Various investigators have reported the coexistence of regions of low WSS
and high tensile stress (Zhao et al. 2002; Dong et al. 2015) and found an inverse
relationship between WSS and arterial wall thickness (Krams et al. 1997; Ku 1997).
Single-phase modelling may be applicable in large arteries such as the aorta and
some larger coronary arteries in the heart. However it is not capable of explaining the
blood flow behavior when blood vessel diameters are small and in the same order as
the red blood cells (RBCs) (d ≲10 μm). Blood at physiologic conditions is a dense
suspension of cells and platelets and predominantly RBCs. The motions of RBCs in
concentrated suspensions were studied by tracking the particles in flow-through
tubes (Goldsmith and Marlow 1979). Non-Newtonian shear thinning with plug
flow in the core of the tubes was observed. The RBC viscosity was governed by
the particle deformation. Based on particle interactions in the concentrated suspen-
sions, Phillips et al. (1992) derived a constitutive equation that consisted of a stress
tensor dependent on the particle volume fraction and a diffusion equation accounting
for shear-induced particle migration. Based on in vitro experimental results, Munn
et al. (1996) described the important role of RBCs in leukocyte rolling, namely, that
the RBCs greatly enhance leukocyte adhesion to activated endothelium.
Therefore, the application of multiphase flow modelling to simulate blood flow as
a suspension of red cells (particles, also called erythrocytes) in plasma (fluid) has
potential to predict the blood rheology more precisely (e.g., as a Newtonian fluid),
especially for blood flow in capillaries.
Model Reconstruction
Prior to the CFD modelling, construction of a computational model of the human

airways or arteries is essential. It can be divided into three stages: medical image
acquisition, segmentation, and surface/volume reconstruction. Image acquisition
involves medical images that can be obtained from various modalities such as
computed tomography (CT) and magnetic resonance imaging (MRI). A 3D matrix
of a series of 2D cross sections, separated by an interval distance, contains informa-
tion about tissues and structures which are distinguished from one another by
differences in brightness or gray scale. When reconstructed into 3D, these images
provide medical practitioners with realistic computational models that can assist
6 J. Dong et al.
Fig. 3 Thresholding of
interested abdominal aorta on
one selected slice using the
Hounsfield scale within the
range of 168 and 596
clinical diagnosis and medical treatment and planning. Reconstruction of the mor-
phological structures from the medical images involves segmentation of the desired
region of interest of organs or structures. The premise of segmentation can be
summarized as the partitioning of an image into a number of homogeneous seg-
ments, such that any two neighboring segments yield a heterogeneous segment.
The simplest segmentation algorithm is based on the gray scale levels of pixels
which is referred to as thresholding. It applies a cutoff range for image pixels by
defining a minimum and maximum pixel gray scale value based on the Hounsfield
scale (HU) to extract the region of interest. The Hounsfield unit is defined as
μP μW
HU ¼ 1000
μW
where μp is the mean X-ray attenuation coefficient of the pixel and μw is the mean
X-ray attenuation coefficient of water. Bone structures attenuate the beam the
greatest, while air has minimal attenuation properties. This segmentation method
uses either global or local information to select only those pixels within an interested
gray scale range, and a binary function is applied as

1 if minx ðf ðxÞÞ θ maxx ðf ðxÞÞ
gðxÞ ¼
0
where θ is the selected HU value and f(x) is the HU values of the images. The output
is a binary image where each pixel falls within the desired HU value, depending on a
pixel’s label of 1 or 0. A CT image illustrating the segmentation of aorta artery is
shown in Fig. 3 where the selected pixels are colored in green. It is most effective
fore images containing uniform regions with clear differentiation from a background
or other objects in terms of pixel intensity values. However, additional side artery
branches and tissues that exhibit pixels within the same HU range are also included.
Fig. 4 Reconstructed
geometry of the
(a) preliminary model with the
presence of noise and
(b) refined model removing
unwanted voxels
In such a circumstance, manual operation is needed to remove these additional pixels

that are not of interest.
A segmented volume is then extracted from the threshold segmentation. The
preliminary 3D surface geometry is created into an .stl format (stereolithography file
format). During the CAD conversion, a large amount of noise is present and the
preliminary 3D geometry, which requires further refinements. A preliminary model
and its refined version are shown in Fig. 4.
Case Study: Gas-Particle Flow in Human Nasal Cavity
The accurate predictions of airflow structures and related particle depositions in

realistic models of the human respiratory system are of fundamental importance in
inhalation toxicology studies. As a major route of entry into the body for airborne
particles, the human nasal cavity, with an effective length of approximately 10 cm,
features a wide array of basic flow phenomena such as boundary-layer flows, flow
separation with recirculation zones, and laminar/turbulent transitional flow. Based on
a realistic human nasal cavity model reconstructed from CT images, this section
presents a case study of the micron particle deposition under steady laminar flow
rates.
8 J. Dong et al.
Fig. 5 Nasal airway passage NC04 model (a) and its inhalation exposure analysis (b)
Fig. 6 Unstructured mesh results with refined near wall mesh having 8-prism layers
Nasal Cavity Reconstruction and Model
A computational model of a human (48-year-old healthy male) nasal passage was

developed based on CT images. The model was truncated at the anterior trachea to
focus on particle deposition in the upper respiratory tract. Figure 5 shows the airway
reconstructed from CT images. The models include both left and right nasal passages
and the nasopharyngeal duct. To better represent the flow conditions at and around
the nostril inlets, the external nares, facial features, and the surrounding environment
a
1 Vestibule
2 Upper passage
3 Middle passage
4 Lower passage
5 Olfactory
6 Septum
7 Pharynx
I II III
lateral 4
side Right
1 3
2 chamber
7
5
septal
side 6
septal
side
Left
chamber
lateral
side
I II III
Fig. 7 3D and 2D views of the human nasal cavity showing regions used for particle deposition
analysis
near the face were included (Doorly et al. 2008; Ge et al. 2013;Inthavong et al. 2012,
2013; King Se et al. 2010).
The computational model was meshed using ICEM-CFD (ANSYS ®, USA) with
unstructured tetrahedral elements. Prism layers were applied in near-wall regions to
provide accurate near-wall particle behavior. The number of independent mesh
elements was 3.8 million (Fig. 6).
Figure 7 separates the nasal cavity into three main nasal areas: area I (includes the
nasal vestibule), area II (includes the main nasal passage), and area III (includes the
nasopharynx). To enable regional particle deposition analysis, the nasal surface was
divided into seven regions based on a combination of anatomical features and
epithelial tissue types (Gross et al. 1982; Schroeter et al. 2012, 2015). These regions
10 J. Dong et al.
Fig. 8 Representative flow streamlines in the human nasal cavity showing airflow patterns. A
cross-sectional slice taken at a-a’ in the human model shows velocity magnitude contours
are the (1) vestibule (squamous epithelium), (2) upper passage (mainly transitional
epithelium), (3) middle and (4) lower passages (mainly respiratory epithelium),
(5) olfactory (olfactory epithelium), (6) septum, and (7) pharynx. The surface-
mapping technique originally developed by Inthavong et al. (2014b) was adopted
to unwrap the nasal cavity morphology from its 3D domain onto a planar 2D
domain, and the 2D surface was normalized into an orthogonal shape allowing
direct comparisons between nasal chambers.
Inhalation and Particle Deposition Modelling
Inhalation airflow rate of 15 L/min was used to reflect resting breath condition (Kelly
et al. 2004). The flow was treated as laminar and incompressible using the commer-
cial CFD package, ANSYS Fluent v14.5 (ANSYS ®, USA). To ensure similarity
with particle exposure experiments, micron-sized particles were released in front of
breathing zone (Fig. 5b). It was reported that particle deposition fraction approaches
Fig. 9 Regional particle deposition fraction comparison for the human NC04 model using
semilog axes
100% in human nasal cavities when particle diameters larger than 20 μm (Inthavong
et al. 2006). Therefore, this study focused on particle diameters of 2.5 μm, 9 μm, and
20 μm, which represent low, medium, and high inertial particles.
Inhalation Flow Patterns
Tracing flow streamlines provide insight to the likely path trajectory an airborne
particle may follow. Flow streamlines were traced from the outside air as it enters the
nostril inlet. By including the facial features, natural flow paths are produced (Doorly
et al. 2008; Corley et al. 2012) which differ to cast replica model studies or
computational models that omit the face and outside air region.
Figure 8a shows the streamlines predominantly distributed through the center
rather than the ceiling or the floor of the airway. Swirling flow is found near the
olfactory region, and this can be attributed to the lower flow velocities found in the
upper airway region to allow gases to be taken up by olfactory nerves but prevent
larger micron-sized particles from reaching the nerves and damaging them. The
single streamline (Fig. 8b) shows flow acceleration immediately after the nostril inlet
and high acceleration at the nasopharynx caused by the airway passage decreasing to
a smaller cross-sectional area. These regions of acceleration propel the airborne
particles forward, increasing the particle inertia. This increases the likelihood of
particle impaction when the flow streamline changes path and the particle inertia is
too great. This flow feature makes the anterior nasal half and the nasopharynx region
acting as a gravimetric sampling filter. The cross section (Fig. 8c) shows a flow
preference to the left nasal chamber which is caused by the larger volume found from
the scanned data.
12 J. Dong et al.
Fig. 10 Particle deposition patterns for the human nasal model in the 3D domain (a, b), 2D planar
domain (c), and deposition penetration distribution (d). Particle sizes are colored based on the low
(2.5 μm, blue), medium (9 μm, green), and high inertial particles (20 μm, red) according to their
deposition fractions reported. The x coordinate in (d) is normalized by the maximum penetration
depth from nostril to nasopharynx
Particle Deposition in the Human Nasal Cavity
Major regional particle deposition results are summarized in Fig. 9. In general, the
micron particle deposition increases from low deposition to almost full deposition in
the nasal cavity from 2.5 to 20 μm particles. The majority of particles are deposited
in the main nasal passage (II), where 2.12%, 35.8%, and 77.4% of particles are
trapped for the low, medium, and high inertial micron particles, respectively.
Although the vestibule is the second preferred deposition site, the deposition fraction
remains small for low (0.95% deposition) and medium (2.85% deposition) inertial
particles. This increases significantly to 21% for high inertial particles. The naso-
pharynx (III) region has the lowest deposition for all particle sizes due to its posterior
location.
To enable complete data access for the complex 3D nasal structure, the 3D
particle spatial deposition results, particle deposition results on the unwrapped 2D
domain, and particle deposition penetration depth along the nasal axial distance are
presented. Thus, Fig. 10 examines the particle spatial deposition in the human nasal
model from three perspectives, 3D view (a, b), 2D view (c), and 1D view (d).
Despite the lowest total deposition fraction of 3.5% (left chamber: 1.1%, right
chamber: 2.4%) occurring for 2.5 μm particles, its deposition exhibits a widely
dispersed pattern over the whole nasal cavity. This is mainly due to that the low
inertial particles are more likely to adapt to the changing inhaled flow paths and
transported through the nasal cavity and penetrate the nasal cavity further. For 9 μm
particles, the total deposition fraction is 40.0% (left chamber: 17.7%, right chamber:
22.3%) as the particle inertia is increased. This leads to fewer particles able to
follow the airflow, and its deposition pattern region reduces. For 20 μm particles,
deposition fraction reaches 100% with a majority 70% depositing in the right
chamber. Particles are mainly concentrated in the main nasal passage (area II) by
direct impaction.
The corresponding 2D deposition (Fig. 10c) shows deposited particles concen-
trating near the entrance to the main nasal passage. Preferential deposition in this
region occurs because the airway opens up from the anterior nose allowing more
particles to access the main passage. In both cavities deposition occurs on the nasal
septal wall side. More particles deposited in the right chamber compared to the left
chamber for all three particle sizes tested. This was mainly due to the asymmetric
geometry of this human nasal model, where its right nasal chamber is slightly larger
than the left, allowing more airflow through with less flow resistance (which was
found in the cross-sectional contour in Fig. 8c). The mass flow attribution in each
chamber shows a 47:53% flow rate distribution in the left to right chambers.
The deposition penetration depth shown in Fig. 10d indicates that particles are
concentrated in four hot spots labelled as “A,” “B,” “C,” and “D” in the human nasal
cavity. Locations “A” and “D” are located at the top of vestibule of right cavity and
posterior septum of right cavity, respectively. These regions in the right chamber are
preferential deposition sites for all three particles. Further inspection of the geometry
shows that region is a consequence of the airflow entering the nostril inlet at higher
momentum, while location “D” exhibits a minor deviated septum causing flow
disturbance. Location “B” has high deposition of high inertial particles (20 μm)
only for both chambers. This location is located at the “nasal valve” which is close to
the common boundary of vestibule and septum of both cavities. Location “C”
exhibits high deposition for medium (9 μm) and high (20 μm) inertial particles.
The deposition occurs at the anterior septum for both chambers and at the common
boundary of septum and upper passage of left cavity.
Summary
The objective of this case study is to numerically present micron particles deposition
patterns in a human nasal cavity in terms of the anatomy, airflow, and particle
deposition. The results demonstrate that majority of the finest particles (2.5 μm)
can pass through the nasal cavity with extremely low deposition fraction. In contrast,
the micron particles with largest inertia (20 μm) are almost captured by the nasal
14 J. Dong et al.
0 0.100 0.200 (n)

0.050 0.150
Fig. 11 Computational model including the nasal spray device
epithelium. In terms of particle deposition location, the main passage receives the
largest portion of the particle exposure dose compared with other major locations,
regardless of inhaled micron particle size differences. Additionally, the unwrapped
nasal deposition results enable a complete data access to further evaluate the particle
exposure risks.
Case Study: Nasal Spray Efficiency Analysis
Nasal drug delivery provides an alternative approach to traditional delivery methods

such as oral drug routes that can fail in the systemic delivery of certain compounds.
The nasal airway is dominated by the nasal turbinates that are lined with highly
vascularized mucosa that contain openings to the paranasal sinuses. Because of these
characteristics, it is hypothesized that drug delivery to combat health problems, such
as lung diseases, cancers, diabetes, sinus infections, etc., may be viable if the drug
formulation can be deposited in the turbinate region (Kimbell et al. 2004). However
this requirement tends to be poorly implemented where a large proportion of the drug
particles is known to deposit in the anterior regions of the nasal vestibule, attributed
to the sprayed particles existing in a high inertial regime (Inthavong et al. 2006;
Kelly et al. 2004; Zwartz and Guilmette 2001). Therefore studies into local particle
deposition become of great significance in the delivery of drugs via the nasal airway.
In this case study, drug delivery efficiency assessment of a nasal spray device is
presented. Knowledge of the atomized drug particles’ initial conditions as they are
introduced into nasal cavity is required in order to simulate and optimize delivery
devices (nasal sprays) and correct drug formulations. Therefore, experimental work
to capture the spray atomization images (Fung et al. 2013; Inthavong et al. 2014a)
was used to extract basic measured parameters from a spray device. Information
such as nozzle diameter, breakup length, diameter at breakup length, spray cone
angle, and initial particle velocity was adopted for numerical simulation input.
Fig. 12 Computational mesh results
Table 1 Measured Measured parameters Value

parameters from
Nozzle diameter 0.5 mm
experimental measurements
of a nasal spray device Diameter at breakup length 4 mm
(Fung et al. 2013; Spray cone angle 30
Inthavong et al. 2014a) Initial particle velocity 15 m/s
Breakup length 3.5 mm
Numerical Model Setup
In this case study, the same human nasal model reconstructed from CT images with
previous case study model was used. In addition, a commercial nasal spray device
was included and inserted into the nostril of the left nasal chamber (Fig. 11).
The computational model was meshed using ICEM-CFD (ANSYS ®, USA) with
unstructured tetrahedral elements. Prism layers were applied in near-wall regions to
provide accurate near-wall particle behavior. The number of independent mesh
elements was 4.2 million (Fig. 12).
The airflow rate of 15 L/min representing a rest breathing condition was chosen.
To capture the flow separation and turbulence near and after the spray device
regions, the realizable k-e model was used, although variants of the model, such as
the k-ω models and the Reynolds stress model (RSM), have been shown to produce
good results. When a turbulence model is used, the near-wall regions must have a
16 J. Dong et al.
Fig. 13 Velocity streamlines passing through the airway
very fine mesh to resolve the turbulent boundary layer, or a wall function is used
which acts as a black box to model the rapid gradients in the turbulent boundary
layer. Here, an enhanced wall function is used to resolve the boundary layer.
The particles adopted the properties of spherical water droplets, as most drug
formulations are diluted with water. Particles were individually tracked under the
Lagrangian approach by integrating the force balance equations on the particle. The
internal walls of the nasal cavity were set to a “trap” condition, meaning that particles
touching a wall deposit at that location. The turbulent dispersion of the particles is
handled by the eddy interaction model (EIM). Spray boundary conditions extracted
from the measurement were listed in Table 1.
Inhalation Airflow Patterns
Path streamlines that track the airflow motion from the external breathing zone are
shown in Fig. 13. A significant velocity increase is observed in the confined space of
the nostril periphery and the vestibule region due to the presence of the spray device
which differs from the streamline results of the plain nasal cavity model (Fig. 8a).
The streamlines then separate and swirl through the nasal valve region. The bulk
flow passes into the oropharynx region with high velocity caused by the airway
passage decreasing to a smaller cross-sectional area.
Fig. 14 Particle trajectory comparison
Spray Trajectory Comparison
Numerical predicted particle spray trajectories of different particle size are presented
in Fig. 14. Four particle sizes, 20 μm, 15 μm, 10 μm, and 5 μm, were analyzed and
compared. It is found that the nasal spray droplets are deposited almost immediately
and the penetration depth remains very short when particle size is above 10 μm. This
is mainly due to the particle inertia. For relatively large micron particles (>5 μm),
particle inertia dominates its motion. As a result, particles hit the nasal vestibule wall
immediately after being released from the spray device. In contrast, when particle
size reduces to 5 μm, the influence of particle inertia turns to be minor, and spray
particles can follow the airflow streamlines well and travel further downstream of the
human up airway.
18 J. Dong et al.
Fig. 15 Particle deposition comparison
Particle Deposition Comparison
Sprayed particle deposition between 15 and 5 μm is compared in Fig. 15. Despite a

greater total deposition fraction for 15 μm particles, the 5 μm particle deposition is
more dispersed. For 15 μm particles, although the particle deposition achieves 11%
deposition in the middle region of the nasal cavity, the majority of them are deposited
in the vestibule region of the nasal cavity with a deposition of 75%. In the middle
region, there is comparable deposition between the two particle sizes. However, due
to the reduced particle inertia, a larger number of 5 μm particles escaped through the
pharynx. This will lead to deposition later downstream in the respiratory tract and
may reach the lungs, which can cause adverse health responses.
Summary
Nasal drug delivery offers an attractive alternative to invasive drug delivery for small
and large molecular weight drugs. The major advantages are the straightforward and
needle-free application mode and the permeable application site in the nasal cavity
that allow a rapid onset of local and systemic drug actions. However, the limited area
in the nasal cavity with an optimal drug absorption capability requires a precise drug
delivery to the targeted area during the nasal spray process. This case presents an
illustration of how to numerically assess the drug delivery efficiency of commercial
nasal spray devices in detail. Due to high inertia effects, large-sized particles are
immediately impacted on the anterior vestibule region, indicating an inefficient drug
delivery performance. While for fine particles (5 μm), the sprayed drug particles can
Table 2 RBCs case studies for different volume fractions

Pressure drop (Pa) VF % Tube diameter (D, μm) Tube length No. of RBCs
48 5 150 2D 3,400
48 15 150 2D 10,200
24 30 150 1D 10,200
travel further downstream the nasal cavity, with an increased likelihood to reach the
targeted epithelium area.
Case Study: Modelling of Blood Flow in Capillary
As a physiological fluid, blood is a complex suspension of polydisperse, flexible,

chemically and electrostatically active cells, which are suspended in an electrolytic
fluid consisting of numerous active proteins and organic substances (Baskurt and
Meiselman 2003). Macroscopic vessels represent only a small fraction of circulatory
system, although the largest veins contain 50% of blood. The vascular tissue is made
of microscopic capillary channels. There are about 1010 blood vessels whose diam-
eters are comparable with the dimensions of the red blood cells (RBCs), i.e.,
5–10 μm (Eggleton and Popel 1998). Therefore, the majority of defects in circulatory
blood system occur in capillary vessels where blood flows less vigorously than in
larger macroscopic vessels.
When analyzing disease development in arteries, it is important to understand the
local variations in blood rheology. Blood flow in large arteries is often assumed to
behave as a homogeneous fluid, an assumption that is not entirely correct (Chien
et al. 1970; Gijsen et al. 1999). The local viscosity changes with the local concen-
tration, and the rate of shear strongly influences the wall shear stress (WSS) and its
gradients, physiological parameters important in the study of atherosclerosis and
aneurysm (Evju et al. 2013; Xiang et al. 2012). Moreover, the flow behavior of
RBCs is influenced by the geometric structure of the flow environment, while
rheological properties across a tube cross section are difficult to measure if nonin-
vasive techniques are to be used (Fullana et al. 2007; Boynard et al. 2007).
In this section, a numerical study employing the Lattice Boltzmann method to
model the blood as a particle suspension of RBCs was reviewed (van Wyk et al.
2014), in which a tube flow with the RBC volume fractions (VF) of 5%, 15%, and
30% was studied.
CFD Models
The geometry is a simplified cylindrical arterial vessel and meshed into O-grid cells
to ensure near orthogonality close to the walls of the tubes. Cells are progressively
finer toward the walls to ensure resolution of the boundary layer. Common boundary
conditions are applied, such as the no-slip wall condition and pressure outlets. The
20 J. Dong et al.
Fig. 16 Instantaneous RBC distributions of the tube flow cases described in Table 2. (a) VF 5%,
(b) VF 15%, (c) VF 30% (Figure adapted with permission from van Wyk et al. (2013))
zero gradient condition is set at the walls for the scalar transport equation since the
RBCs never diffuse across arterial walls. Other boundary conditions are listed in
Table 2.
Fig. 17 Mean velocity 0.8

distributions for (a) VF 5%, VF 5%
(b) VF 15%, and (c) VF 30% 0.7 VF 15%
(Figure adapted with VF 30%
permission from van Wyk 0.6
et al. (2013))
0.5
UX / UN
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
r/R
Flow Field and Mean Velocity Distribution
An isometric view of the RBC distributions at an instant in time for each of the cases
is shown in Fig. 16. The image on the right for each flow case plots all particles in
semitransparent form to compare the particle density. The results demonstrate that
RBCs get forced closer to the walls as the bulk VF is increased. This is partially due
to the decrease in available volume in the center and therefore an increase in the
viscosity.
The mean radial velocity distributions of the studied tube flow cases are displayed
in Fig. 17. The mean radial velocity profile of the 5% case is still in a parabolic
shape, whereas the 15% and 30% cases develop a more pluglike flow shape in the
center of the tube. The error bars show that the velocity profiles of the 5% and 15%
cases are within an acceptable range of approximately 2%. The lack of data for the
30% case is partly responsible for the larger variation near the center of the tube.
However the concentration of RBCs near the center is greater leading to larger
variations in the flow field.
Summary
The application of multiphase flow modelling on Poiseuille flow of blood suspen-

sions in small capillaries was reviewed. The modelled suspension of monodispersed
RBCs exhibits distinct flow behavior compared with homogeneous plane-shear flow,
and the variation of the RBC volume fraction significantly alters the flow field. In
conjunction with physiological realistic arterial vessel and flow conditions, this
modelling approach can offer more fluid dynamic insights to the study of arterial
disease, such as atherosclerosis.
22 J. Dong et al.
Conclusion
This chapter outlines the main procedure of solving biofluid dynamics problems, and
provides practical modelling solutions through case study demonstrations. Multi-
phase modelling of inhalation exposure, nasal drug delivery and red blood cells in
capillary were presented in detail. This chapter can serve as an illustration of the
necessary modelling framework in numerical applications. Combined with experi-
mental measurements and clinical observations, multiphase CFD modelling can
address unmet clinical needs, predominantly in the direction of enhanced diagnosis,
as well as assessment and prediction of treatment outcomes.
References
L. Antiga, M. Piccinelli, L. Botti, B. Ene-iordache, A. Remuzzi, D. Steinman, An image-based
modeling framework for patient-specific computational hemodynamics. Med. Biol. Eng.
Comput. 46, 1097–1112 (2008)
O.K. Baskurt, H.J. Meiselman, Blood rheology and hemodynamics. Semin. Thromb. Hemost. 29,
435–450 (2003)
B. Berthier, R. Bouzerar, C. Legallais, Blood flow patterns in an anatomically realistic coronary
vessel: influence of three different reconstruction methods. J. Biomech. 35, 1347–1356 (2002)
M. Boynard, L. Haider, H. Lardoux, P. Snabre, Rheological and flow properties of blood investi-
gated by ultrasound. Indian J. Exp. Biol. 45, 18–24 (2007)
Y.S. Cheng, G.K. Hansen, Y.F. Su, H.C. Yeh, K.T. Morgan, Deposition of ultrafine aerosols in rat
nasal molds. Toxicol. Appl. Pharmacol. 106, 222–233 (1990)
S. Chien, S. Usami, R.J. Dellenba, M.I. Gregerse, Shear-dependent deformation of erythrocytes in
rheology of human blood. Am. J. Physiol. 219, 136 (1970)
R.A. Corley, S. Kabilan, A.P. Kuprat, J.P. Carson, K.R. Minard, R.E. Jacob, C. Timchalk,
R. Glenny, S. Pipavath, T. Cox, C.D. Wallis, R.F. Larson, M.V. Fanucchi, E.M. Postlethwait,
D.R. Einstein, Comparative computational modeling of airflows and vapor dosimetry in the
respiratory tracts of rat, monkey, and human. Toxicol. Sci. 128, 500–516 (2012)
G.H. Dai, M.R. Kaazempur-Mofrad, S. Natarajan, Y.Z. Zhang, S. Vaughn, B.R. Blackman,
R.D. Kamm, G. Garcia-Cardena, M.A. Gimbrone, Distinct endothelial phenotypes evoked by
arterial waveforms derived from atherosclerosis-susceptible and -resistant regions of human
vasculature. Proc. Natl. Acad. Sci. U. S. A. 101, 14871–14876 (2004)
C. Darquenne, G.K. Prisk, Aerosol deposition in the human respiratory tract breathing air and 80:
20 heliox. J. Aerosol Med.-Depos. Clearance Eff. Lung 17, 278–285 (2004)
J.L. Dong, K. Inthavong, J.Y. Tu, Image-based computational hemodynamics evaluation of athero-
sclerotic carotid bifurcation models. Comput. Biol. Med. 43, 1353–1362 (2013)
J.L. Dong, Z.H. Sun, K. Inthavong, J.Y. Tu, Fluid-structure interaction analysis of the left coronary
artery with variable angulation. Comput. Methods Biomech. Biomed. Eng. 18, 1500–1508
(2015)
D.J. Doorly, D.J. Taylor, R.C. Schroter, Mechanics of airflow in the human nasal airways. Respir.
Physiol. Neurobiol. 163, 100–110 (2008)
C.D. Eggleton, A.S. Popel, Large deformation of red blood cell ghosts in a simple shear flow. Phys.
Fluids 10, 1834–1845 (1998)
A. Elder, R. Gelein, V. Silva, T. Feikert, L. Opanashuk, J. Carter, R. Potter, A. Maynard,
J. Finkelstein, G. Oberdorster, Translocation of inhaled ultrafine manganese oxide particles to
the central nervous system. Environ. Health Perspect. 114, 1172–1178 (2006)
O. Evju, K. Valen-sendstad, K.A. Mardal, A study of wall shear stress in 12 aneurysms with respect
to different viscosity models and flow conditions. J. Biomech. 46, 2802–2808 (2013)
D.O. Frank, J.S. Kimbell, S. Pawar, J.S. Rhee, Effects of anatomy and particle size on nasal sprays
and nebulizers. Otolaryngol. Head Neck Surg. 146, 313–319 (2012)
J.M. Fullana, N. Dispot, P. Flaud, M. Rossi, An inverse method for non-invasive viscosity
measurements. Eur. Phys. J.-Appl. Phys. 38, 79–92 (2007)
M.C. Fung, K. Inthavong, W. Yang, P. Lappas, J. Tu, External characteristics of unsteady spray
atomization from a nasal spray device. J. Pharm. Sci. 102, 1024–1035 (2013)
Q. Ge, X. Li, K. Inthavong, J. Tu, Numerical study of the effects of human body heat on particle
transport and inhalation in indoor environment. Build. Environ. 59, 1–9 (2013)
F.J.H. Gijsen, F.N. Van de vosse, J.D. Janssen, The influence of the non-Newtonian properties of
blood on the flow in large arteries: steady flow in a carotid bifurcation model. J. Biomech. 32,
601–608 (1999)
H.L. Goldsmith, J.C. Marlow, Flow behavior of erythrocytes. 2. Particle motions in concentrated
suspensions of ghost cells. J. Colloid Interface Sci. 71, 383–407 (1979)
E.A. Gross, J.A. Swenberg, S. Fields, J.A. Popp, Comparative morphometry of the nasal cavity in
rats and mice. J. Anat. 135, 83–88 (1982)
Y. Hoi, H. Meng, S.H. Woodward, B.R. Bendok, R.A. Hanel, L.R. Guterman, L.N. Hopkins,
Effects of arterial geometry on aneurysm growth: three-dimensional computational fluid dynam-
ics study. J. Neurosurg. 101, 676–681 (2004)
K. Inthavong, Z.F. Tian, H.F. Li, J.Y. Tu, W. Yang, C.L. Xue, C.G. Li, A numerical study of spray
particle deposition in a human nasal cavity. Aerosol Sci. Technol. 40, 1034–10U3 (2006)
K. Inthavong, Z.F. Tian, J.Y. Tu, W. Yang, C. Xue, Optimising nasal spray parameters for efficient
drug delivery using computational fluid dynamics. Comput. Biol. Med. 38, 713–726 (2008)
K. Inthavong, Q.J. Ge, X.D. Li, J.Y. Tu, Detailed predictions of particle aspiration affected by
respiratory inhalation and airflow. Atmos. Environ. 62, 107–117 (2012)
K. Inthavong, Q.J. Ge, A. Li, J.Y. Tu, Source and trajectories of inhaled particles from a surrounding
environment and its deposition in the respiratory airway. Inhal. Toxicol. 25, 280–291 (2013)
K. Inthavong, M.C. Fung, X.W. Tong, W. Yang, J.Y. Tu, High resolution visualization and analysis
of nasal spray drug delivery. Pharm. Res. 31, 1930–1937 (2014a)
K. Inthavong, Y.D. Shang, J.Y. Tu, Surface mapping for visualization of wall stresses during
inhalation in a human nasal cavity. Respir. Physiol. Neurobiol. 190, 54–61 (2014b)
J.A. Johnson, K.C. Bloch, B.N. Dang, Varicella reinfection in a seropositive physician following
occupational exposure to localized zoster. Clin. Infect. Dis. 52, 907–909 (2011)
J.T. Kelly, B. Asgharian, J.S. Kimbell, B.A. Wong, Particle deposition in human nasal airway
replicas manufactured by different methods. Part I: inertial regime particles. Aerosol Sci.
Technol. 38, 1063–1071 (2004)
J.T. Kelly, B. Asgharian, B.A. Wong, Inertial particle deposition in a monkey nasal mold compared
with that in human nasal replicas. Inhal. Toxicol. 17, 823–830 (2005)
J.S. Kimbell, J.D. Schroeter, B. Asgharian, B.A. Wong, R.A. Segal, C.J. Dickens, J.P. Southall,
F.J. Miller, Optimization of nasal delivery devices using computational models. Respir. Drug
Deliv. IX 1, 233–238 (2004)
C.M. King SE, K. Inthavong, J. Tu, Inhalability of micron particles through the nose and mouth.
Inhal. Toxicol. 22, 287–300 (2010)
R. Krams, J.J. Wentzel, J.A.F. Oomen, R. Vinke, J.C.H. Schuurbiers, P.J. Defeyter, P.W. Serruys,
C.J. Slager, Evaluation of endothelial shear stress and 3D geometry as factors determining the
development of atherosclerosis and remodeling in human coronary arteries in vivo – combining
3D reconstruction from angiography and IVUS (ANGUS) with computational fluid dynamics.
Arterioscler. Thromb. Vasc. Biol. 17, 2061–2065 (1997)
W.G. Kreyling, M. Semmler-behnke, W. Moller, Ultrafine particle-lung interactions: Does size
matter? J. Aerosol Med.-Depos. Clearance Eff. Lung 19, 74–83 (2006)
D.N. Ku, Blood flow in arteries. Annu. Rev. Fluid Mech. 29, 399–434 (1997)
M.G. Menache, F.J. Miller, O.G. Raabe, Particle inhalability curves for humans and small
laboratory-animals. Ann. Occup. Hyg. 39, 317–328 (1995)
24 J. Dong et al.
L.L. Munn, R.J. Melder, R.K. Jain, Role of erythrocytes in leukocyte-endothelial interactions:
mathematical model and experimental validation. Biophys. J. 71, 466–478 (1996)
J.G. Myers, J.A. Moore, M. Ojha, K.W. Johnston, C.R. Ethier, Factors influencing blood flow
patterns in the human right coronary artery. Ann. Biomed. Eng. 29, 109–120 (2001)
R.J. Phillips, R.C. Armstrong, R.A. Brown, A.L. Graham, J.R. Abbott, A constitutive equation for
concentrated suspensions that accounts for shear-induced particle migration. Phys. Fluids Fluid
Dyn. 4, 30–40 (1992)
J.D. Schroeter, J.S. Kimbell, B. Asgharian, E.W. Tewksbury, M. Singal, Computational fluid
dynamics simulations of submicrometer and micrometer particle deposition in the nasal pas-
sages of a Sprague-Dawley rat. J. Aerosol Sci. 43, 31–44 (2012)
J.D. Schroeter, E.W. Tewksbury, B.A. Wong, J.S. Kimbell, Experimental measurements and
computational predictions of regional particle deposition in a sectional nasal model.
J. Aerosol Med. Pulm. Drug Deliv. 28, 20–29 (2015)
Y.D. Shang, K. Inthavong, J.Y. Tu, Detailed micro-particle deposition patterns in the human nasal
cavity influenced by the breathing zone. Comput. Fluids 114, 141–150 (2015)
M. Siebes, Y. Ventikos, The role of biofluid mechanics in the assessment of clinical and pathological
observations. Ann. Biomed. Eng. 38, 1216–1224 (2010)
C.A. Taylor, M.T. Draney, Experimental and computational methods in cardiovascular fluid
mechanics. Annu. Rev. Fluid Mech. 36, 197–231 (2004)
A. Tsuda, F.S. Henry, J.P. Butler, Particle transport and deposition: basic physics of particle kinetics.
Compr. Physiol. 3, 1437–1471 (2013)
S. Van wyk, L.P. Wittberg, L. Fuchs, Wall shear stress variations and unsteadiness of pulsatile
blood-like flows in 90-degree bifurcations. Comput. Biol. Med. 43, 1025–1036 (2013)
S. Van wyk, L.P. Wittberg, L. Fuchs, Atherosclerotic indicators for blood-like fluids in 90-degree
arterial-like bifurcations. Comput. Biol. Med. 50, 56–69 (2014)
S.M. Wang, K. Inthavong, J. Wen, J.Y. Tu, C.L. Xue, Comparison of micron- and nanoparticle
deposition patterns in a realistic human nasal cavity. Respir. Physiol. Neurobiol. 166, 142–151
(2009)
D.J. Weber, W.A. Rutala, H. Hamilton, Prevention and control of varicella-zoster infections in
healthcare facilities. Infect. Control Hosp. Epidemiol. 17, 694–705 (1996)
J. Wen, K. Inthavong, J. Tu, S.M. Wang, Numerical simulations for detailed airflow dynamics in a
human nasal cavity. Respir. Physiol. Neurobiol. 161, 125–135 (2008)
T.R. Wolfe, D.A. Braude, Intranasal medication delivery for children: a brief review and update.
Pediatrics 126, 532–537 (2010)
J.P. Xiang, M. Tremmel, J. Kolega, E.I. Levy, S.K. Natarajan, H. Meng, Newtonian viscosity model
could overestimate wall shear stress in intracranial aneurysm domes and underestimate rupture
risk. J Neurointerventional Surg. 4, 351 (2012)
R.A. Yokel, R.C. Macphail, Engineered nanomaterials: exposures, hazards, and risk prevention.
J. Occup. Med. Toxicol. 6, 7 (2011)
S.Z. Zhao, B. Ariff, Q. Long, A.D. Hughes, S.A. Thom, A.V. Stanton, X.Y. Xu, Inter-individual
variations in wall shear stress and mechanical stress distributions at the carotid artery bifurcation
of healthy humans. J. Biomech. 35, 1367–1377 (2002)
G.J. Zwartz, R.A. Guilmette, Effect of flow rate on particle deposition in a replica of a human nasal
airway. Inhal. Toxicol. 13, 109–127 (2001)
Flow Boiling Enhancement via
Cross-Sectional Expansion
Patrick Phelan and Mark Miner
Abstract
Heat transfer enhancements available from expanding the cross section of a
microchannel or microchannel system in flow boiling are presented, including
recommendations appropriate for design and selection of expanding channel heat
sinks. The principal relevant operating parameters of a boiling-channel heat sink
are the attainable critical heat flux (CHF), which limits the practical heat flux
permissible and the pressure drop across the channel, which may impose substan-
tial pumping costs on the loop and is coupled to stability of flow in the channels.
Keywords
Microchannel • Liquid cooling • High heat flux • Flow boiling • Pressure drop •
Two-phase
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Microchannels: An Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Evaluating Predictive Criteria for Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Basic Phenomenology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Observed Improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
P. Phelan (*)
School for Engineering of Matter, Transport and Energy, Arizona State University, Tempe, AZ,
USA
e-mail: phelan@asu.edu
M. Miner
School of Earth and Space Exploration, Arizona State University, Tempe, AZ, USA
e-mail: mark.miner@asu.edu

DOI 10.1007/978-981-4585-86-6_17-1
2 P. Phelan and M. Miner
A perturbation of the channel diameter may be employed to examine CHF and

pressure drop relationships from the literature with the aim of identifying those
adequately general and suitable for use in a scenario with an expanding channel.
Optimum rates of expansion which maximize the critical heat flux may be extracted
from some criteria, and optima in expansion have been observed by investigators.
The boiling number is considered, and it is seen that expansion typically produces an
increase in the boiling number in the region explored, though no optima are
observed.
Pressure drop relationships admit improvement with expansion according to the
same perturbation analysis, and no optimum appears. The relevant phenomena
surrounding flow boiling pressure drop are considered, along with a handful of
dimensionless numbers as qualitative selection aids. Decrease in the pressure drop
across the evaporator is observed with expanding channels, though low-frequency
oscillations are not necessarily damped out. Expansion is seen to improve stability of
the flow by reducing the dependence of the pressure drop on heat flux.
Introduction
The thermal management engineer may be justly likened to the undertaker; it is best
not to need, but the need is inevitable. Heat exchange apparatuses are required in
every thermodynamic cycle, no matter how simple. If the engineer wishes to
accomplish most any task consuming or generating useful work, heat exchange
will be required. How a cycle is accomplished directly affects its utility; wasted
work becomes entropy, and the exergy of the system is reduced. Heat transfer across
a finite temperature difference generates entropy; therefore, devices with low resis-
tance to heat flow allow any given process to capture more of the desirable output
with lower overhead.
Looking beyond efficiency, device effectiveness remains stubbornly tied to heat
exchange. Low-power design and thoughtful component layout can mitigate the
need to pump away unwanted heat, but performance demands inevitably outstrip
efficient architecture, and the waste heat generated by ever-smaller, higher-power,
and higher-frequency devices must be dissipated. The ability to employ processes
which rapidly convert energy into beneficial formats is increasingly limited by the
ability to reject substantial amounts of heat energy in space- or weight-limited
devices. Many engineering examples offer themselves, such as turbine blades,
high-concentration photovoltaics, portable catalytic reforming, and advanced batte-
ries, forcing the conclusion that thermal limitations are a ubiquitous challenge across
engineering disciplines.
Changing the thermodynamic phase of a substance is not strictly a mode of heat
transfer, as the substance undergoing the change must accept or reject heat via one of
the usual modes, but phase change does represent an opportunity for significant
energy arbitrage. Since the temperature of a liquid-vapor mixture is coupled to the
pressure of a system, the boiler or evaporator designer has a measure of predictabil-
ity. The coupling of convection with phase change to obtain improved cooling
Flow Boiling Enhancement via Cross-Sectional Expansion 3
performance presents potent possibilities, where the cooling device scale enforces
proximity of coolant molecules to the heated walls. A useful tool in this application
is the microchannel.
This chapter presents an advance in flow boiling heat-exchanger design, analyt-
ically and experimentally treating the possibility of optimizing the streamwise shape
of a channel or array of channels to enhance heat rejection and reduce associated
system pumping cost. Reducing the required work input directly increases first-law
efficiency metrics of a given cycle, and reducing the resistance to heat transfer of
either the accepting or rejecting heat exchanger improves second-law efficiency,
reducing entropy generation.
Microchannels: An Overview
Microchannels promise a compact, versatile, tunable solution to a substantial set of

thermal management problems. Enhancing convection by shrinking the scale of the
channel is an extremely attractive option at small device scales. Conventional
convective relationships for predicting the heat transfer coefficient look like
Nu k
h¼ (1)
d
where h is the convective heat transfer coefficient (in mW2 K ); Nu the dimensionless
Nusselt number which represents the ratio of convection to conduction in the fluid
and is typically
W obtained via experimental correlation, k, the thermal conductivity of
the fluid mK ; and d the appropriate length scale of the flow, which for internal flow
is the hydraulic diameter of the tube carrying the fluid. The Nusselt number is an
experimentally derived value, with a variety of correlations tailored to specific flow
regimes. In laminar flow, which is typical for single-phase microchannels, the
Nusselt number carries a subunity dependence on the length scale, meaning that
the limit of Eq. 1 tends to infinity as d goes to zero. This behavior clearly extrapolates
from an experimental-physical correlation into nonphysical values, but for some
region, as the diameter of a tube decreases, the resistance of the flow to convective
heat transfer also decreases. Single-phase cooling in microchannels is widely
employed and presents many interesting challenges, but it is always limited by
Poiseuille’s law and often hampered by the necessarily nonuniform fluid tempera-
tures obtaining in sensible heating.
The latent heat of vaporization of a liquid coolant holds out promise for moving
beyond some of the limitations of single-phase flow. The principal drawback, on the
surface, is the exchange of a quasi-static system that operates at a predictable and
repeatable steady state for the inherently dynamic boiling process, where flow is
dramatically less predictable in time and in space. For all that, flow boiling remains a
tantalizing tool to employ in the pursuit of higher heat-density cooling applications,
and the prospects of direct boiling for steam generation could be attractive due to
reduced capital requirements. Phase change promises obvious advantages in large
latent heats, reduced coolant inventories, and stable, pressure-selectable sink tem-
peratures. Engineering challenges, though, are to be found right the way down.
An excellent survey of the field is available (Kandlikar 2012), which, with some
overlap into single-phase flow, weighs well the state of the art and the challenges
impeding the widespread application of microscale flow boiling. The common
themes of pressure drop oscillation and critical heat flux (CHF) are sounded time
and again, and the reader is directed to Kandlikar’s extensive bibliography for further
exploration. The challenge of harnessing microscale flow boiling is not the theory,
but difficulties which appear in the practical application of the phenomenon. Many
investigators (Zhang et al. 2009; Chang and Pan 2007; Qu and Mudawar 2003;
Kandlikar 2004) have observed and investigated the unstable nature of multiphase
microchannel flow, where local stochastic processes like bubble nucleation cascade
up into global unpredictability of the entire heat sink. Much research has subse-
quently targeted the reduction or elimination of these thermohydraulic instabilities,
which tend to limit the achievable heat flux.
Thermal instabilities become pressure instabilities, as local temperature irregu-
larities cause the flow to boil in a more inhomogeneous fashion than usual, and the
subsequent changes in the flow field feed back into thermal behavior because of local
dryout or flow reversal. Because thermal loads will often be somewhat variable, due
to the locality of heating and time-varying loads in circuitry or other applications, the
principal emphasis of instability suppression has been to break oscillatory cycles by
manipulating the flow field through modification of the pressure profile.
Elimination of flow reversal in microchannels can be readily accomplished by
imposition of a steep pressure gradient at or near the inlet of the channels. This
clearly approaches the limiting case of an expansion valve in a vapor-compression
refrigerant loop, which is a known stable configuration. By using less aggressive
chokes in the flow path, called inlet restrictors, many investigators (Odom et al.
2012; Revellin and Thome 2007a; Koşar et al. 2006; Kandlikar et al. 2006; Basu
et al. 2011) have successfully reduced oscillatory instabilities, but at the cost of a
significant pressure penalty.
While a single restriction is beneficial, some investigators (Mukherjee and
Kandlikar 2009) further posited a greater beneficial effect from a series of finite
constrictions. An expanding vapor bubble in such a channel was modeled and the
output was promising. Channel expansion allowed the bubble growth to proceed
predominately downstream, somewhat ameliorating the flow reversal that can be
introduced by bubble expansion in a straight channel. As a series of finite constric-
tions could be a challenge to manufacture, this work suggested that continuous
expansion would serve better than none.
A separate approach of reducing instability is to attempt to homogenize the flow
field by distributing nucleation sites in desired areas, and some investigators (Kuo
and Peles 2008) fabricated reentrant cavities and laser-etched nucleation sites into
microchannel walls. This reduces the required wall superheat for bubble nucleation,
but may not directly affect pressure behavior. However, Pan’s group continued these
researches and obtained significant improvement in heat sink behavior by combining
artificial nucleation sites with the idea of an expanding channel cross section (Lu and
Pan 2009). This concept was first introduced to the thermal literature in an exami-
nation of the pressure effects of varying cross-sectional channels on ethanol-CO2
flows, and benefits were noted from expansion (Hwang et al. 2005).
Pan and colleagues continued to pursue a number of related experiments, using
continuously expanding etched silicon microchannels; sometimes as a single chan-
nel (Lee and Pan 2007), but also in parallel arrays (Lu and Pan 2008, 2009, 2011).
More recently, investigation of these arrays of expanding microchannels extended to
other flow patterns with a mind toward engineering applications (Liu et al. 2012).
Balasubramanian et al. have also inquired into expanding microchannels, using
copper with wire-EDM-cut channels; discrete cross-sectional area changes are made
by chopping a channel wall at a given point along the flow direction, thus merging
two neighboring channels and adding the wall space in between (Balasubramanian
et al. 2011). Results from all of these groups have been uniformly positive:
expanding the cross section of a microchannel reduces instabilities and improves
both CHF and pressure drop characteristics.
Miner et al. numerically, experimentally, and analytically pursued the optimiza-
tion of expanding microchannel geometry (Miner et al. 2013a, b, 2014), seeking to
maximize the critical heat flux (CHF) attainable within a fixed footprint. These
investigations pointed to a maximum CHF at a relatively small expansion angle
and realized this in laboratory experiments. These same experiments demonstrated
that expanding microchannels offered a substantial reduction in pumping costs and a
partial decoupling of the thermal and pressure effects (Miner and Phelan 2013).
Another avenue of research has examined microchannels of constant cross
section with an open manifold above, a configuration that allows vapor to leave
the channel orthogonal to the flow direction of refrigerant (Kandlikar et al. 2013;
Kalani and Kandlikar 2014, 2015a, b). These studies considered both a tapered and a
constant cross-sectional manifold and noted improvement with the taper. This
method facilitates rewetting of the heated surface, which is a substantial benefit,
and effectively acts as a modified form of pool boiling, where the superficial velocity
of the coolant is not a driver of performance. The approach eliminates confinement
effects at the scale of the channel dimension and may allow for reduced manufactur-
ing costs, as the tapered manifold may be simpler to produce than a tapered channel
array. Excellent visualization of the flow is presented by these investigators, and flow
pattern mapping is pursued.
Finally, a beautiful set of investigations with high-speed photography (Tamanna
and Lee 2015a, b) enabled further flow pattern mapping in an expanding microgap
(effectively a single, wide channel), and the fine-grained apparatus employed also
demonstrated conclusively that expansion has a substantial stabilizing effect on wall
temperatures, which is a major motivation for electronics-oriented researchers.
These investigators also reported an optimum expansion rate, also at a relatively
small expansion angle. While reporting reduced pressure drop, they also observed
large-scale slow pressure oscillations in the system, which seemed to intensify as the
expansion rate increased.
The research group of Odom et al. observed similar large-scale pressure oscilla-
tion behavior early in their investigations, but determined it to be caused by periodic
cycling in the chiller. The chiller was taken out of service, and an ice bath was
employed to stabilize the heat sink temperature. Higher heat rejection with expan-
sion increases chiller loading, increasing the amplitude of these cycles, which are the
result of systematic pressure reduction in the liquid-vapor system as heat is removed
by the chiller. It was also noted that chillers advertising PID control fail to note that
compressors are binary and effectively pursue a pulse-width modulation cooling
scheme, adding an unwanted oscillation to the loop. Researchers employing chillers
may wish to take additional care to reduce these effects.
Pressure drop in small channels has been addressed by a number of researchers,
beginning with Poiseuille, but modern consideration of boiling flow was dominated
by macroscale applications, such as boilers for steam generation and, later, evapo-
rators in refrigeration applications. The results of these researchers were carried
downward in channel size for some time, and departure from the venerable Lockhart
and Martinelli correlation in mini- and microchannels began to be contemplated by
Lazarek and Black (1982), who modified it for their frictional pressure drop in short
tubes with fluorocarbon refrigerants. Kandlikar provided an excellent review of the
literature up to the year 2001, and it is notable that only 10 of the 28 investigations he
surveyed took up pressure drop directly. The reader is directed to the extensive
bibliography in Kandlikar’s paper (Kandlikar 2002) for further investigation. Since
that survey, much more attention has been paid to understanding the mechanisms of
the boiling pressure effects, especially in light of their tight coupling to heat transfer
phenomena.
Subsequent investigators took up the questions of classification and prediction of
pressure drop through microchannels, and as usual, new correlations work well for
the data sets from which they are derived, but a lack of generality persists. Mudawar
surveyed the challenges encountered in his career investigating boiling flow and
noted in particular the multiplicity of factors contributing to pressure drop across
microchannel evaporators and the ongoing disagreement over their significance and
prediction (Mudawar 2011). However, his extensive paper is also a paean to a variety
of employments found for microchannels in emerging and established technology
and leaves the researcher encouraged that the applications engineers may be catching
on and catching up to the state of the art.
Evaluating Predictive Criteria for Expansion
The body of data considering microchannels is growing rapidly, and many investi-
gators have published correlations predicting both critical heat flux events and
pressure drop across a microchannel array. These correlations are generally derived
from limited data sets and invariably produce acceptable outcomes for the
researchers deriving them. The correlations typically extend only to very similar
circumstances and configurations, and extrapolation of any correlation is a danger. In
the interest of clarifying which relationships may be of use to the microchannel
engineer who wishes to improve flow boiling, a variety of correlations available
from the literature are examined to obtain a verdict on the effect of channel cross-
sectional expansion.
A simple test may be posed to some relationship predicting a flow parameter, such
as critical heat flux correlation q_ CHF ¼ F , or pressure drop dp dz ¼ G , purporting
generality: if the diameter terms are construed as d = di + ez, where di is the inlet
diameter, e the rate of channel expansion, and z the station downstream from the
inlet, what is the sign on the expression resulting from the partial with respect to e,
e.g., @F/@e? The results will plainly be best for small e, but if the effect on the
prediction corresponds to observed physical reality, the correlation may contain
adequate information and prove generally useful. If, in a reasonable parameter
space, @F/@e > 0, the prediction behaves realistically and predicts enhanced critical
heat flux from expansion. This is known to be the case from experiment, and the
predictive criterion may be useful. Similarly, if @G/@e<0, the correlation predicts
reduced pumping cost, as has been experimentally observed, and may be a useful
relation.
This procedure may be employed by any investigator wishing to evaluate a given
criterion subject to expansion. It is useful in downselecting from a plethora of criteria
and may be applied to any geometry, though with varying accuracy, by substitution
of hydraulic diameter in place of circular diameter. Implicit in this approach is the
assumption that small alterations should yield small effects, but those effects ought
to trend in the correct direction. Understanding that employing a perturbation on
derived equations rather than on governing equations is sometimes risky, analysis
proceeds by arguing that the original equation is always recovered if e tends to zero,
and that the results of such analysis will be tentative until compared with
experimental data.
A variety of predictive correlations for critical heat flux (CHF) have been
examined for improvement via cross-sectional expansion by considering the effect
of a small perturbation of the diameter of a circular channel on the predictions made
by the selected criteria (Miner and Phelan 2013). By examining the partial derivative
of the critical parameter with respect to expansion, extrema could occasionally be
located. For several criteria, nonzero diameter expansion necessarily increases
critical heat flux, and an optimum expansion rate exists for many of these criteria.
Additionally, it was demonstrated that CHF relations follow only a few distinct
types, relations which contemplate more involved physical effects, such as pressure
and phase velocity, and perform much better against the expansion examination than
those which only considered upstream variables. The upstream-only correlations are
manifestly more attractive to the system designer, who is beginning without an
extant system in hand. The best-performing criteria according to this test are
summarized in Table 1.
The calculations involved are straightforward, but often produce long expressions
with large groups of positive terms. For the purpose of examining the criterion, only
@
the sign of the @e expression is significant. If the sign is fixed by constant parameters
alone, the calculations become quite simple, and a straightforward conclusion may
be reached. If the sign is fixed by geometry or flow variables (like We or Lh/d ) which
Table 1 Correlations with favorable effect

Author(s) Condition
Katto and Ohno (1984)
(3) Lh
d > 645
(4) Lh
d > 376
(5) d
0:280Lh WeF
> :003
(13) Lh
d > 376
Qi et al. (2007) Lh
d > 83:33Coþ50:97
0:765Co
h i
Koşar and Peles (2007) 1 @P 17:281P
Lh
> 23:12 @e 2
d :0934P :34P 1:3104

Revellin and Thome (2007b) 17 37 @ V @V L

ðρL ρV Þgd2 1
V
Lh
d > :075 VVVL 4σ L 7
Lh
d 37 @e
VV @e
VL
cannot be negative, the result may be presented as an inequality. If there is a

competition between effects permitting the sign to change under certain conditions,
the expression has captured competing physics, which is encouraging. Additionally,
2
the optimum situation where @@e2 ¼ 0 may be explored analytically if the calculation
is tractable, or it may be considered numerically across desired parameter spaces.
Revellin and Thome’s criterion is displayed as in their work (Revellin and Thome
2007b) and not in the guise of a Bond or confinement number, since the authors
wished to call attention to the most dangerous wavelength in the Kelvin-Helmholtz
instability.
Investigators who desire to employ the CHF enhancements available from
manipulating channel geometry must choose wisely among predictive criteria,
some of which may not function well for even simple enhancements such as an
expanding cross section. Relationships which contemplate the flow variables
directly influencing CHF respond better to this interrogation by partials, but require
measurement, approximation, or modeling of the flow variables required. Some of
these criteria depend on geometry and flow variables easily obtainable early in the
design phase of a system, rendering them nearly as useful as the purely upstream-
dependent relations. In general, the more physical phenomena which are reflected in
the criterion, the more accurate it will be (Table 1).
A similar study was performed across a variety of pressure drop predictions in
microchannels (Miner et al. 2014) as an adjunct to an experimental study. In all
cases, expansion was predicted to reduce pumping cost, and it seems that pressure
relationships may be more generally physically grounded than the CHF relationships
noted above. Thus, the discussion of pressure drop correlations is somewhat more
limited in scope than that of critical heat flux prediction. While many investigators
have discussed the issue, their concern is typically with refining certain parameters
within an unanimous framework or characterizing and including pressure drops
associated with other elements of the apparatus, such as inlet contraction and outlet
expansion. Consistent across all investigators considered is the distinction between
frictional and acceleration pressure drop, the former due to bubble-induced tortuosity
and surface effects and the latter due to the thrust developed from evaporating liquid
turning to faster-moving vapor. The physical forces affecting these terms will be
taken up below.
Basic Phenomenology
The physical scientist will wish to know the nature of the forces acting inside a
boiling microchannel. The engineer will wish to know which of these forces matter
and what may be done to obtain a given objective. Kandlikar provides an excellent
discussion of these effects and settles on the inertia, surface tension, shear, buoyancy,
and evaporation momentum forces as the candidates for significance (Kandlikar
2010). The formulations here employed are those which consider force over a unit
area and are cast into variables more appropriate to a system designer.
The inertial force scale Fi considers momentum flux through a given area as:
_
mV m_ 2 G2
Fi 4 (2)
A ρd ρ
where m_ is the mass flow rate, V is the average velocity, A is the local flow cross-
sectional area, ρ is the fluid density (as appropriate to the quality at a given station),
and G is the mass flux. The final formulation, involving the constant mass flux and
quality-dependent density, will substantially increase in magnitude as the density of
the flow drops during boiling.
The surface tension force scale Fσ follows from the Young-Laplace equation
applied to the bulk channel dimension as for a circular channel cross section:
σ
Fσ (3)
d
where σ is the surface tension of the liquid phase of the fluid and d is the relevant
channel dimension; either radius or diameter may suffice, as the order of magnitude
is unchanged by the factor of 2. Plainly, this term is significant only in proportion to
the liquid fraction in the channel and will disappear for a pure vapor flow.
The shear force scale Fτ will be treated as Newtonian, with shear stress propor-
tional to the velocity gradient at the wall. When the channel is full of liquid, flow is
likely to be laminar and have a velocity gradient proportional to the velocity. As
vapor is generated in what is here assumed to be an annular core, the liquid film
against the walls is subjected to a zero-velocity no-slip condition at the wall and a
matching-velocity no-slip condition at the vapor. When the vapor density is signif-
icantly less than the liquid density, as is common in refrigerants, the vapor velocity
will be large. Regardless of the particular shape of the velocity distribution in the
liquid, the velocity gradient in the increasingly thin liquid film may be reasonably
assumed to vary linearly with the vapor core velocity, as this situation approaches a
type of Couette flow. Thus, the force scale may be simply represented as:
μV m_ vG
Fτ μ 3 (4)
d ρd d
where μ is the liquid viscosity, V is the velocity of whatever phase is appropriate, ρ is

again the quality-weighted bulk fluid density, and ν is the momentum diffusivity,
which is generally higher in the liquid phase than in the vapor phase, making this
term reduce in magnitude during boiling.
Gravity may affect the flow, and the characteristic buoyancy force Fg is a function
of the density difference as:
Fg ðρL ρV Þgd sinðθÞ (5)
where g is the acceleration due to gravity and θ the deviation of the channel from
horizontal.
The fluid boiling in the channel advances in the vapor phase, with a higher
velocity than the liquid phase. Thus, as fluid is boiled, the conservation of mass
requires that it is accelerated forward, developing a thrust. This evaporation momen-
tum force FM scales as:

_ gen ðV V V L Þ
mV _ m_
qL 1 1
FM (6)
d2 hLV d 3 ρV ρL
where mV_ gen is the rate of vapor mass generation (intimately related to the heat input)
and L is the channel length. The heat flux may be considered locally, as when writing
a spatially resolved solver, or globally, as when designing a channel ab initio.
These are the most significant fundamental forces at work within a channel, and it
will be instructive to consider how they may react to alterations in the channel
configuration.
Until flow reversal occurs, the inertial forces will tend downstream. The liquid-
vapor interface in annular separated flow is distended from the minimum area shape
sought by the surface tension, which will tend to pull liquid upstream; this follows
from consideration of a channel as a capillary tube. Shear forces will resist motion,
so until flow reversal occurs, these forces will be directed upstream. Gravity will be
directed according to the channel orientation. Evaporation momentum thrust will
always tend to drive the interface upstream. The system driving pressure, typically
from a pump or compressor, will define the downstream direction. Notably, only
driving pressure, evaporation momentum, and capillary forces have orientations
independent of the instantaneous flow behavior. This physically grounds the obser-
vation that rewetting dry channels or overcoming flow reversals is a difficult task: the
inertia force has reversed direction and must now be overcome. Additionally, the
evaporation momentum thrust tends to increase during dryout events, as thin liquid
Table 2 Sensitivity of forces to channel expansion

Force Inertia Surf. tens. Shear Buoyancy Evaporation thrust
σ

Expression m_ 2 μm_ (ρL ρV)gd _ m_
qL 1
ρ1
ρd4 d ρd3 hLV d 3 ρV L
σz _

@F
@e
4m_ 2 z
d2
3μmz (ρL ρV)gz 3qL_ m_ 1
ρ1
ρd 5 ρd 4 h d4 ρ
LV V L
Fig. 1 Representative force
Force Magnitude
scales in 10 mm long
microchannels with 3 gs R-134a
at 350 kPa, x = 0.5,
q_ ¼ 500 cm
W
2
100
–2
10–6 10–5 10–4 10–3 10
Diameter, m
Inertia
Evap. Thrust
dF/dε Magnitude
Shear
Surf. Tens
Buoyancy
100
10–6 10–5 10–4 10–3 10–2
Diameter, m
films evaporate extremely well, temporarily increasing the vapor mass

generation rate.
The sensitivity of each of these force scales to the channel expansion is a quantity
of interest. Table 2 presents the results of the sensitivity analysis, in which the
diameter is defined as d = d0 + ez and the resulting expression is differentiated
with respect to the small parameter e. A visualization of these force scales, following
in a familiar format (Kandlikar 2010), is presented in Fig. 1, assuming a represen-
tative 3 gs flow of R-134a at 350 kPa in a 10 mm long channel with an average quality
of 0.5 and a heat flux of 500 500 cm W
2.
All of the forces diminish with channel expansion, with the exception of buoy-
ancy, and all force scale effects increase linearly with the length of the channel. The
more significant quantity may be the strength of the relationship between a force and
the channel diameter d and by extension the expansion parameter e. The inertia force
is the most sensitive, dropping off as d5. This force scale tends to have the largest
magnitude in small channels and may be directed downstream in typical flow
patterns or upstream during flow reversal. Reduction of the inertial force scale allows
the conclusion that an expanding channel will be easier to rewet after a flow reversal.
Shear and evaporation momentum forces fall off as d4, a steep reduction, but not
so steep as inertia. The evaporation thrust force is always working against the desired
flow direction, so its reduction with expansion is seen as a significant benefit. The
reduction in shear force comes from the easing of the velocity gradient with an
expanding channel; this is seen as a benefit, as it contributes to the diode-like
properties of the channel, easing flow in the direction of expansion.
The surface tension force diminishes as well, but not nearly so rapidly. The d2
dependence means that this force will increase in magnitude with respect to inertia,
shear, and evaporation thrust; the surface tension is helpful in rewetting walls, which
are drier in the downstream direction, though it continues to pull liquid toward the
smaller end of the channel.
Buoyancy alone becomes more significant as the channel expands. This effect
would be most pronounced in vertical orientations of the evaporator, but since
Lazarek and Black (1982) did not report any significant directionality to their results
for upflow and downflow, directionality is not expected to be a first-order influence
on the flow boiling behavior.
If the critical parameter for predicting pressure drop in a microchannel array is
driven solely by evaporation thrust and inertia, we may require that FM < Fi:

_ m_
qL 1 1 m_ 2
3 ρ
< 4 (7)
hLV d V ρL ρd
which may be rearranged to yield the dimensionless term that predicts this behavior
as:

_
qLdρ 1 1
<1 (8)
_ LV ρV ρL
mh
and since the momentum at the exit of the channel will be most significant in
predicting flow reversal, the appropriate ρ value is likely to be that of vapor,
which is typically much less than the density of liquid refrigerant, resulting in a
familiar group of terms:
_ 2 L
qd L
¼ Bl (9)
_ LV d
mh d
Equation 9 indicates that the critical parameter in flow reversal, which may
precipitate critical heat flux, is linked to the boiling number and the confinement
of the channel. This parameter appears in great many CHF predictions, but it is not
necessarily a reliable group to employ for expanding channel diameters (Miner and
Phelan 2013).
If more forces are allowed to participate in the equation, perhaps even all five of
the aforementioned groups and the system driving pressure ΔP, it may be expected
that flow reversal in a channel inclined at angle θ to the horizontal is avoided if
FM + Fτ + Fg sin θ < ΔP + Fi + Fσ. Employing the force scales discussed

above yields:

_ m_
qL 1 1 μm_ m_ 2 σ
3 ρ
þ 3 þ ðρL ρV Þgd sin ðθÞ . . . < ΔP þ 4 þ (10)
hLV d V ρL ρd ρd d
and by invoking the density arguments previously made and by normalizing the
inertial force scale, which results in the pressure drop being presented as an Euler
number Eu (representing the ratio of imposed pressure forces to inertia forces),
Eq. 10 becomes:
L μd ðρL ρV ÞρV gd5 sin ðθÞ ρV σd3

Bl þ þ 1 < Eu (11)
d m_ m_ 2 m_ 2
The first term remains as before, but several other pertinent terms are built into
Eq. 11. The second term is observed to be the reciprocal of the Reynolds number Re,
the third term contains the Eötvös (Bond) number Eo combined with the Weber
number We (conceptually equivalent to the Richardson number Ri) and inclination,
and the final term on the left is noted to be the reciprocal of the Weber number, and a
further rearrangement yields:
L 1 1 Eo sin ðθÞ
Bl þ þ 1 < Eu (12)
d Re We We
The sum of effects on the left-hand side of Eq. 12 must not exceed the Euler
number. This expression captures the salient effects of geometry, refrigerant param-
eters, and experimental outcomes (in Bl and Eu) and represents the link between the
physical performance of the system and the constituent elements. Equation 12 may
have predictive or at least qualitative-comparative value, and it may provide a
starting point for predictive evaluations of flow boiling pressure drop and critical
heat flux in microchannel arrays.
At this point it may be objected that no account has been given of the driving
pressure drop. The inertial force employed in the above equations has no other
source than the externally applied pressure difference across the channel array, as is
plainly implied by the Hagen-Poiseuille equation, which is reasonably employed for
the liquid inlet conditions:
m_ Fi
ΔP μL 4
(13)
ρD Re
Equation 12 provides a conceptual framework for the manipulation of the rele-

vant channel parameters affecting pressure behavior in the channel array. Large
boiling numbers are deleterious, as are large L/d ratios and small Reynolds numbers.
Small Weber and Eötvös numbers are helpful, along with small inclinations of the
channel. Large Euler numbers provide room for the other parameters. The relative
Table 3 Microchannel expansion-related performance improvements

Improvement metric
Investigator Comparison Configuration and coolant and magnitude
Lee and Pan Expanding versus Water. 25 mm lg., 0.183 25% increased heat
(2007) uniform single exp., 21/μm dp. flux
channel 34 21 μm chan. Dhyd Reduced wall
dp. superheat
Increased temperature
stability
Lu and Pan Expanding versus Water. 26 mm lg., 0.5 4 reduced Δp
(2008) uniform channel exp., 76 μm dp., 100 μm oscillation
array (uniform wd. in 560 μm wd out Larger stable flow
internally cited) region
Increased slope of
stability boundary
Balasubramanian Expanding versus Water. 25 mm lg., 30% reduced Δp
et al. (2011) uniform channel 1.2 mm dp., 310 μm Reduced slope of Δp
array (chopped wd. In Aout/Ain = 1.65 versus q curve
wall) Increased heat trans,
coeff. Stable wall
temp to 0.2 C
Miner et al. Expanding versus R-134a, 13 mm 20–50% improvement
(2013b, 2014) uniform channel lg. 0 ,0.5 ,1 ,2 exp. in qCBF
array 150 μm. wd., 600 μm Increase in boiling
dp. in number
2 10 Δp
reduction
4 reduction in
slope of Δp versus
q curve substantial
stability improvement
Kandlikar et al. Tapered versus Water. 40–333 ml/min Uniform p fluctuation
(2013) uniform manifold 10 10 mm array 4–8 tapered
217 162 μm chan. 3.3 C wall superheat
wd. dp. 0–180/ μm taper reduction
2% heat flux
improvement
(no CHFj
Kalani and Tapered versus Water. 80 ml/min 2–5 C wall superheat
Kandlikar (2014) uniform manifold 10 10 mm array reduction
(here neglecting 127–727 μm mfid. Depth Δp uniform 10
plain chip data) 181 450 μm chan. Δp tapered
wd. dp.
Kalani and Tapered versus Water. 10 C subcoo1mg Taper reduces slope of
Kandlikar uniform manifold 10 10 mm array Δp versus q
(2015a) (here neglecting 127–727 μm mfid. Depth Δp uniform 4–8
plain chip data) 181 450 μm chan. dp tapered
wd. dp.
(continued)
Table 3 (continued)
Improvement metric
Investigator Comparison Configuration and coolant and magnitude
Tamanna and Lee Expanding versus Water. 1.27 cm lg., 200 μm 15–30% dp reduction
(2015a, b) uniform microgap dp. in, 200–460 μm dp. out at higher q
More stable thin liquid
film boiling
More stable wall
temperatures
Optimum geometry
had 15% higher
q for same Twall at
higher q
magnitude of these effects will become apparent after some thought, as common
channel conditions involve Re >> 1, L/d >> 1, We >> 1, and Eo 1, leaving the
boiling number and Euler number as the variables most readily manipulated. Given a
certain refrigerant and a target heat flux, minimization of the sum on the left-hand
side of Eq. 12 would illuminate the desirable parameter space in D and L, which are
likely to be the design variables in evaporator or boiler design. The insertion of the
expanding channel diameter d = d0 + ez would add e to the parameter space and
allow the applications engineer to select a desirable geometry and estimate the
attendant pressure drop penalty.
Observed Improvements
Observed improvements from a variety of investigators are presented in Table 3 to

allow an overview of the gains obtainable from expanding microchannels. Where the
investigators observed an optimum, it is noted.
Table 3 demonstrates an unanimity in the results obtained for the various
expanding flow passage configurations: stability is enhanced, wall temperatures
are moderated, and pressure oscillations lessen. Not all of the investigators pursued
experiments up to critical heat flux, but those that did observed increased CHF
available from expansion. The results for overall pressure drop are mixed, with the
very small channels tending toward higher pressure drops, and the larger channels,
manifolds, and gaps observing reduced pumping costs associated with expansion.
The dominant forces at very small (verging on nanochannel) scales are not the same
as those at the microchannel and nearly minichannel scales, and this behavior is not
surprising.
Example data were provided by Miner et al., and some of the key features of
expansion are illustrated in the figures that follow.
The typical increase of critical heat flux with mass flux is observed in Fig. 2
(Miner et al. 2013b). These data trend monotonically upward with mass flux, and
expanding channel cross-sectional shifts the curves toward higher heat fluxes for a
Fig. 2 CHF versus mass 500

flow rate
450
Critical Heat Flux (W/cm2)

400
350
300
0°
250 0.5°
200 1°
2°
150
400 600 800 1000 1200 1400 1600
Mass Flux (kg / (m2∗s))
given mass flux. Because the mass flux varies through the channel, what is presented
here is taken at the channel average cross-sectional area.
A very surprising feature was noted in a straight microchannel apparatus after
experimentation was concluded: damage appeared in the copper, looking very much
like early-stage cavitation scarring. Figure 3 shows the outlet face of the straight
channel turret, and bubble-like damage patterns are circled on the end faces of some
of the fins. This damage did not appear on the other channel arrays in that investi-
gation. The pressure drop occurring when liquid refrigerant exits the channels to the
outlet plenum can result in flashing, and the lowest pressure regions would be the
outlet faces of the fins. These sites may therefore be prone to cavitation-type
nucleation-induced damage. A reduction in the outlet pressure accompanies expan-
sion of the channel cross section, and this may ameliorate the likelihood of damage
and account for the lack of observed damage in the other turrets. The largest
observed damage site was approximately 60 μm diameter, and sites are observed
down to 15 μm in diameter and below, though it becomes increasingly difficult to
distinguish between possible damage sites and surface roughness features. A brief
examination of the critical bubble radius indicates that these site sizes lie within the
bounds of the minimum and maximum critical bubble radii. This damage mecha-
nism in an evaporator may be negligible at larger scales, but could pose a threat to the
longevity of a microchannel evaporator. Suitable engineering of the plenum could
ameliorate this; however, the open-manifold arrays employed by Kalani et al. and the
gap configuration of Tamanna and Lee would be much less susceptible to this
phenomenon.
Extensive data on the pressure drop and CHF coupling are presented in Fig. 4
(Miner et al. 2014). The figure shows results averaged over periods of constant
heater input power for every experiment performed in the study. Heat flux into the
channel array lies on the abscissa, and the ordinate plots the pressure drop over mass
flux. The trend shown in Fig. 4 is clear: cross-sectional expansion of the micro-
channel reduces the deleterious effect of heat input on the pressure drop across the
Fig. 3 Damage at straight

channel outlet
Fig. 4 Heat flux effect on 0.8

pressure drop 0°
0.5°
ΔP/G (kPa∗s∗mm2 / g)
0.6
1°
2°
0.4
0.2
0
0 1 2 3 4 5
Heat Flux (W / mm2)
array. This augurs well for the employment of expanding microchannels, as the
feedback effect from higher heat rates increasing the required pressure drop is
ameliorated.
It bears repeating that pressure drop and heat input remain strongly coupled in the
expanding channels; the degree of this coupling decreases for larger channel expan-
sions. Additionally, the decoupling seems to stop at the optimum 1 degree expansion
observed in that investigation, with the larger 2 degree angle exhibiting effectively
identical coupling between thermal and pressure effects.
Improved stability is also evident from an examination of the outlet qualities
attained by the expanding channels. A more stable channel will tolerate a higher
outlet quality and continue to function. The general trend is for expanding channels
Fig. 5 Outlet vapor quality 0.9

0°
0.5°
0.8
Outlet Quality (dim)

1°
2°
0.7
0.6
0.5
1.5 2 2.5 3 3.5 4 4.5
Mass Flow Rate (g/s)
to allow increased outlet quality, indicating a stability improvement. The enhanced

outlet quality is not adequately explained by greater inlet subcooling of the
expanding channels, as the additional heat input required to raise the quality is
about four times greater than the heat input required to make up for the largest
subcooling shown in Fig. 5, and the most dramatic outlet quality difference occurs
for lower mass flow rates, where there was no appreciable difference in the sub-
cooling between the expansions. This leaves expansion as the most logical expla-
nation for the improvement in boiling efficiency.
Applications
Expanding microchannels in all flavors allow enhanced cooling in high heat flux
applications. Chip cooling is especially amenable to these devices, as the typical chip
geometry requires a constrained footprint operating at high-power dissipations.
Expanding channels and expanding manifolds have both shown promise, using a
wide variety of coolants, including air. Chip cooling is expected to remain a driver of
research in this field, and the application is not limited to processors. A key
advantage of the expanding microchannel array is the improved uniformity in wall
temperature. This ameliorates hotspots on chips, which are almost invariably the
precursors to failure.
Steam generation in small formats could profitably employ the expansion advan-
tage, and steam generation in large format boilers could be substantially aided by the
concomitant pressure drop reduction, though it is not as clear what the CHF effects
would be in larger channels where different boiling modes (and traditional models)
hold sway. However, the conventional scale models are ripe for examination by the
methods discussed herein.
Reforming of fuels in fuel-cell applications demands high heat transfers and the
accommodation of multiphase and reacting flows, all of which are temperature and
pressure sensitive to some degree. The enhanced control offered by expanding

microchannels or tapered manifolds can make a beneficial difference in the optimal
operation of these reactors, with wall temperature stability being directly linked to
catalytic performance.
The heat buildup in modern lithium-ion batteries may be especially amenable to
expanding channel cooling, as the only possible coolants for use in batteries must be
dielectric fluids, a category in which air and common refrigerants readily fall, and
fluids for which expanding microchannel cooling has proven effective.
Developers of solar receivers could consider expanding microchannels as either
steam generation units for a true solar-thermal plant or for photovoltaic module
cooling under highly concentrated light.
The amazing advances in manufacturing technology developed in the last gener-
ation enable the fabrication of microchannels with greater ease than ever before.
Micromachining, electron discharge techniques, and selective etching allow almost
any desired channel profile to be placed on a suitable substrate. The demonstrable
improvements available from enhanced microchannels are well within reach of the
system designer.
Conclusions
This chapter presents geometric expansion as an eminently usable enhancement to

microchannel evaporators and boilers. The effects of channel cross-sectional expan-
sion or manifold expansion have been demonstrated to increase the critical heat flux
attainable by an evaporator while reducing the pump work required and improving
the stability of the cooling system. Experimental and analytical investigations
provide much support for the idea of optimizable expansion for critical heat flux
and significant reduction of pressure drop across microchannel systems.
Experiment bears out the optimization of expansion of the cross section of a
microchannel in the flow direction. The location of the optimum tends to higher
expansions when mass flow rates in the channel are higher. Boiling numbers are also
increased by cross-sectional expansion, indicating improved consumption of latent
heat in expanding channel evaporators and reflecting the improved stability of these
devices. Expansion of the microchannel cross section in the flow direction is readily
done with conventional fabrication techniques, and the optimum improvement does
not require dramatic expansion of the channel, leaving conduction heat paths
minimally affected.
The effects of microchannel cross-sectional expansion appear to be uniformly
beneficial: reduced associated pumping cost in the loop, reduced or eliminated
low-frequency oscillatory instabilities in the system, and notable decoupling of the
thermal and mechanical behaviors of the system. Applying this method to conven-
tional channels requires a greater understanding of the scaling of the relevant
phenomena, but besides fabrication costs, there are few perceived disadvantages.
Cross-sectional expansion may prove to be an expedient method of making multi-
phase flow more tractable to the applications engineer by enhancing the
predictability of evaporators or boilers subject to large heat fluxes. The reduction in

cyclic stresses from oscillatory instabilities may allow longer and more reliable
service life to a boiling system.
Examination of force scales which are likely to be relevant to the flow allows
insight to the effects of cross-sectional expansion on the fundamental forces in a
boiling channel. These considerations are phenomenologically specific with respect
to major fluid properties and apparatus geometry and are expected to yield qualita-
tive guidance to the engineer wishing to design and employ channel expansion as a
design tool. Care must be taken when selecting and balancing dominant mechanisms
in the optimization, as these will scale with the channel dimensions.
Channel cross-sectional expansion has a similar principle of effect as a flow
restriction, namely, the insertion of a diode into the flow circuit. Making the heat
source itself act as the enforcer of unidirectional flow reduces the part count of the
system and means that the entire system will behave better. The unpleasant side
effects of flow boiling are not merely contained, but are abated; the heat fluxes
available to the apparatus are increased; the system becomes more stable and more
capable. Well-tuned cross-sectional expansion in a boiler will provide the applica-
tions engineer with a system that is more capable, durable, and predictable at high
heat fluxes. As exergetic efficiency becomes a stronger driver of design in the face of
resource and lifetime cost constraints, it is hoped that this method will find increasing
use in the engineering community at large.
Acknowledgments The authors gratefully acknowledge the Office of Naval Research for funding
and support. This work was partially supported by the Office of Naval Research as a MURI award
(prime award number N00014-07-1-0723).
References
K. Balasubramanian, P.C. Lee, L. Jin, S. Chou, C. Teo, S. Gao, Experimental investigations of flow
boiling heat transfer and pressure drop in straight and expanding microchannels. A comparative
study. Int. J. Therm. Sci. 50(12), 2413–2421 (2011)
S. Basu, S. Ndao, G.J. Michna, Y. Peles, M.K. Jensen, Flow boiling of R134a in circular microtubes
part II: study of critical heat flux condition. ASME J. Heat Transf. 133(5), 051503 (2011)
K. Chang, C. Pan, Two-phase flow instability for boiling in a microchannel heat sink. Int. J. Heat
Mass Transf. 50(11), 2078–2088 (2007)
J. Hwang, F. Tseng, C. Pan, EthanolCO2 two-phase flow in diverging and converging micro-
channels. Int. J. Multiphase Flow 31(5), 548–570 (2005)
A. Kalani, S.G. Kandlikar, Evaluation of pressure drop performance during enhanced flow boiling
in open microchannels with tapered manifolds. ASME J. Heat Transf. 136(5), 051502 (2014)
A. Kalani, S.G. Kandlikar, Effect of taper on pressure recovery during flow boiling in open
microchannels with manifold using homogeneous flow model. Int. J. Heat Mass Transf. 83,
109–117 (2015a)
A. Kalani, S.G. Kandlikar, Flow patterns and heat transfer mechanisms during flow boiling over
open microchannels in tapered manifold (OMM). Int. J. Heat Mass Transf. 89, 494–504 (2015b)
S.G. Kandlikar, Fundamental issues related to flow boiling in minichannels and microchannels.
Exp. Thermal Fluid Sci. 26(2), 389–407 (2002)
S. Kandlikar, Heat transfer mechanisms during flow boiling in microchannels. ASME J. Heat
Transf. 126(1), 8–16 (2004)
S.G. Kandlikar, Scale effects on flow boiling heat transfer in microchannels: a fundamental
perspective. Int. J. Therm. Sci. 49(7), 1073–1085 (2010)
S.G. Kandlikar, History, advances, and challenges in liquid flow and flow boiling heat transfer in
microchannels: a critical review. ASME J. Heat Transf. 134(3), 034001–1–15 (2012)
S.G. Kandlikar, W.K. Kuan, D.A. Willistein, J. Borrelli, Stabilization of flow boiling in micro-
channels using pressure drop elements and fabricated nucleation sites. ASME J. Heat Transf.
128(4), 389–397 (2006)
S.G. Kandlikar, T. Widger, A. Kalani, V. Mejia, Enhanced flow boiling over open microchannels
with uniform and tapered gap manifolds. ASME J. Heat Transf. 135(6), 061401 (2013)
Y. Katto, H. Ohno, An improved version of the generalized correlation of critical heat flux for the
forced convective boiling in uniformly heated vertical tubes. Int. J. Heat Mass Transf. 27(9),
1641–1648 (1984)
A. Koşar, Y. Peles, Critical heat flux of R-123 in silicon-based microchannels. ASME J. Heat
Transf. 129(7), 844–851 (2007)
A. Koşar, C.-J. Kuo, Y. Peles, Suppression of boiling flow oscillations in parallel microchannels by
inlet restrictors. ASME J. Heat Transf. 128(3), 251–260 (2006)
C.-J. Kuo, Y. Peles, Flow boiling instabilities in microchannels and means for mitigation by
reentrant cavities. ASME J. Heat Transf. 130(7), 72402 (2008)
G.M. Lazarek, S.H. Black, Evaporative heat transfer, pressure drop and critical heat flux in a small
vertical tube with R-113. Int. J. Heat Mass Transf. 25(7), 945–960 (1982)
P.C. Lee, C. Pan, Boiling heat transfer and two-phase flow of water in a single shallow micro-
channel with a uniform or diverging cross section. J. Micromech. Microeng. 18(2),
025005–1–13 (2007)
T.-L. Liu, B.-R. Fu, C. Pan, Boiling two-phase flow and efficiency of co and counter-current
microchannel heat exchangers with gas heating. Int. J. Heat Mass Transf. 55(21–22),
6130–6141 (2012)
C.T. Lu, C. Pan, Stabilization of flow boiling in microchannel heat sinks with a diverging cross-
section design. J. Micromech. Microeng. 18(7), 075035–1–13 (2008)
C.T. Lu, C. Pan, A highly stable microchannel heat sink for convective boiling. J. Micromech.
Microeng. 19(5), 055013–1–13 (2009)
C.T. Lu, C. Pan, Convective boiling in a parallel microchannel heat sink with a diverging cross
section and artificial nucleation sites. Exp. Thermal Fluid Sci. 35(5), 810–815 (2011)
M.J. Miner, P.E. Phelan, Effect of cross-sectional perturbation on critical heat flux criteria in
microchannels. ASME J. Heat Transf. 135(10), 101009 (2013)
M.J. Miner, B.A. Odom, C.A. Ortiz, J. Sherbeck, R. Prasher, P.E. Phelan, Optimized expanding
microchannel geometry for flow boiling. ASME J. Heat Transf. 135(4), 042901 (2013a)
M.J. Miner, P.E. Phelan, C.A. Ortiz, B.A. Odom, Experimental measurements of critical heat flux in
expanding microchannel arrays. ASME J. Heat Transf. 135(10), 101501 (2013b)
M.J. Miner, P.E. Phelan, C.A. Ortiz, B.A. Odom, Experimental measurements of pressure drop in
expanding microchannel arrays. ASME J. Heat Transf. 136(3), 031502 (2014)
I. Mudawar, Two-phase microchannel heat sinks: theory, applications, and limitations. ASME
J. Electron. Packag. 133(4), 041002 (2011)
A. Mukherjee, S.G. Kandlikar, The effect of inlet constriction on bubble growth during flow boiling
in microchannels. Int. J. Heat Mass Transf. 52(21), 5204–5212 (2009)
B.A. Odom, M.J. Miner, C.A. Ortiz, J. Sherbeck, R. Prasher, P.E. Phelan, Microchannel two-phase
flow oscillation control with an adjustable inlet orifice. ASME J. Heat Transf. 134(12), 122901
(2012)
S.L. Qi, P. Zhang, R.Z. Wang, L.X. Xu, Flow boiling of liquid nitrogen in micro-tubes: part II heat
transfer characteristics and critical heat flux. Int. J. Heat Mass Transf. 50(25), 5017–5030 (2007)
W. Qu, I. Mudawar, Measurement and correlation of critical heat flux in two-phase micro-channel
heat sinks. Int. J. Heat Mass Transf. 47(10), 2045–2059 (2003)
R. Revellin, J.R. Thome, Adiabatic two-phase frictional pressure drops in microchannels. Exp.
Thermal Fluid Sci. 31(7), 673–685 (2007a)
R. Revellin, J.R. Thome, A theoretical model for the prediction of the critical heat flux in heated
microchannels. Int. J. Heat Mass Transf. 51(5), 1216–1225 (2007b)
A. Tamanna, P.S. Lee, Flow boiling heat transfer and pressure drop characteristics in expanding
silicon microgap heat sink. Int. J. Heat Mass Transf. 82, 1–15 (2015a)
A. Tamanna, P.S. Lee, Flow boiling instability characteristics in expanding silicon microgap heat
sink. Int. J. Heat Mass Transf. 89, 390–405 (2015b)
T. Zhang, T. Tong, J.-Y. Chang, Y. Peles, R. Prasher, M.K. Jensen, J.T. Wen, P.E. Phelan, Ledinegg
instability in microchannels. Int. J. Heat Mass Transf. 52(25), 5661–5674 (2009)
Nanoparticle-Laden Flow for Solar
Absorption
Vikrant Khullar, Sanjeev Soni, and Himanshu Tyagi
Abstract
Nanoparticle-laden fluid (or more popularly “nanofluid”) could be engineered to
efficiently absorb as well as transport solar energy. This flow involves suspension of
nano-sized particles (particle size <100 nm) within a fluid. The fluid may be in
gaseous or liquid state depending on the application or a naturally occurring phe-
nomenon. Nanoparticles of various materials, shapes, and sizes can be suspended
within a fluid, without agglomeration, through certain chemicals known as surfactants
or capping agents. Such nanoparticle-laden fluids can be prepared in situ through
chemical methods, or these can be obtained by mixing the solid nanoparticles to a
fluid. There are advantages of using nano-sized particles as compared to micro-sized
particles. Nanoparticle-laden flows do not lead to clogging of the fluid passage or
other components like pumps and valves. Moreover, the fact that the optical proper-
ties of metallic and semiconductor nanoparticles are dependent on their shape, size,
and the surrounding dielectric media makes them easily usable for engineering highly
solar selective nanoparticle dispersions at very low volume fractions.
This chapter describes in detail the mechanism of absorption of solar energy
by volumetric absorption solar thermal systems employing nanoparticle-laden
fluid flows. Thermophysical, rheological, and optical properties of the
V. Khullar (*)
Mechanical Engineering Department, Thapar University, Patiala, Punjab, India
e-mail: vikrant.khullar@thapar.edu; vikrantk@iitrpr.ac.in
S. Soni
Biomedical Instrumentation Division, CSIR-Central Scientific Instruments Organisation,
Chandigarh, India
e-mail: ssoni@csio.res.in
H. Tyagi
Department of Mechanical Engineering, Indian Institute of Technology Ropar, Rupnagar, Punjab,
India
e-mail: himanshu.tyagi@iitrpr.ac.in

DOI 10.1007/978-981-4585-86-6_19-1
2 V. Khullar et al.
nanoparticle-laden fluids have been explored. Critical analysis of the previous

studies relevant to nanofluid-based solar thermal systems reveals that these novel
solar thermal systems hold huge potential and warrant further exploration for
practical realization of such novel solar thermal systems.
Keywords
Nanoparticle • Nanofluid flow • Solar energy • Volumetric absorption • Overall
efficiency • Heat transfer • Absorption • Scattering
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Stability of Nanoparticle-Laden Fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Synthesis of Nanoparticle-Laden Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Possible Ways in Which Agglomeration (and Subsequent Settling Out) Can
Be Avoided . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Thermophysical, Rheological, and Optical Properties of Nanofluids . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Relevance of Nanoparticle-Laden Fluid Flow in Solar Thermal Systems . . . . . . . . . . . . . . . . . . . . . . 5
Limitations of Surface Absorption-Based Solar Thermal Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Emergence of Nanoparticle-Laden Fluid Flows for Solar Thermal Applications . . . . . . . . . . . 6
Can Nanoparticle-Laden Fluid Flow Be Engineered to Compete with Solar
Selective Surfaces? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Conditions in which Nanoparticle Dispersion Can Have Low Radiative Losses as
Compared to Solar Selective Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Thermal Modeling of Nanofluid-Based Volumetric Absorption Solar Thermal Systems . . . . . . 9
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Basic Modeling Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Fundamental Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Modeling Optical and Thermophysical Properties of Nanoparticle-Laden Fluid . . . . . . . . . . . 12
Performance Evaluation of Nanoparticle-Laden Fluid-based Solar Thermal Systems . . . . . . 14
Nanofluid-Based Solar Thermal Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Potential Basefluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Nanofluids Relevant to Solar Thermal Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Solar Thermal Systems Employing Nanofluids as the Working Fluid . . . . . . . . . . . . . . . . . . . . . . 20
Conclusions and Future Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Introduction
Nanoparticle-laden fluid (or more popularly “nanofluid”) is a stable dispersion of

nano-sized particles (particle size <100 nm) in a fluid. Nanoparticle-laden fluid flow
is the flow of nanofluid through a conduit or an open channel. This type of flow is
essentially a multiphase flow in the sense that in addition to the bulk flow of the fluid
as a whole, there is random motion (Brownian motion), agglomeration, and disper-
sion across the flow cross section and in the flow direction.
Historically, these fluid flows have found applicability in heat transfer applications
owing to their enhanced thermophysical properties. More recently, the optical tun-
ability has furthered the application areas where these fluid flows could be put to use.
Nanoparticle-Laden Flow for Solar Absorption 3
Stability of Nanoparticle-Laden Fluid
In order to employ nanoparticle-laden fluid for any heat transfer flow application, it
is imperative to ensure the stability (i.e., the very existence) of these fluids. The
stability of nanofluids depends primarily on the following:
• Synthesis route of these nanofluids

• Operating conditions (such as range of temperatures and pumping and plumbing
arrangements)
The following sections detail the basic synthesis routes of nanofluid preparation.
Synthesis of Nanoparticle-Laden Fluids
In order to ascertain the stability of nanofluids, it is important to understand how

these nanofluids are prepared. Broadly, nanofluids can be manufactured by the
following two methods (Ghadimi et al. 2011):
• One-step process: Involves in situ production of the nanoparticles in the basefluid

• Two-step process: Involves production of nanoparticles and subsequently dis-
persing these nanoparticles into the basefluid
Nanofluids produced by the one-step process have the advantage that they are
stable and generally do not require any additional stabilization process. However,
currently it is difficult to produce nanofluids (of desired volume fraction) on a very
large scale by the one-step process. On the other hand, the two-step process offers the
flexibility of precisely controlling the volume fraction of the nanoparticles in the
nanofluid. Furthermore, numerous number of nanoparticle-basefluid combinations
can be engineered in accordance with the specific application. However, the nano-
fluids produced by the two-step process are relatively less stable (compared to those
produced by one-step process), and the nanoparticles may tend to agglomerate and
settle out of the solution Otanicar et al. (2013).
Van der Waals force of attraction between the nanoparticles tends to combine the
nanoparticles to sub-/micron-sized particles (aggregation instability). These rela-
tively heavier particles subsequently settle out of the solution owing to the force of
gravity referred to as kinetic instability (Taylor et al. 2013a).
Possible Ways in Which Agglomeration (and Subsequent Settling

Out) Can Be Avoided
There are mainly two types of methods by which agglomeration can be prevented –
physical and chemical methods. Soft agglomerations can be broken down through
the application of physical methods such as ultrasonication and magnetic (and high
4 V. Khullar et al.
shear) stirrer. However, these physical methods may alter the length (and hence the
thermophysical/optical properties) of anisotropic nanoparticle geometries (for
instance, it can reduce the length of carbon nanotubes). Ultrasonication has proven
to be the most popular among the physical methods (Taylor et al. 2013a).
There are several types of chemical methods which can be used to prevent
agglomeration.
Addition of dispersing agents/surfactant: Surfactant molecules surround the nano-

particles and form an envelope/cap around the nanoparticle. The electrostatic
force of repulsion among these capped nanoparticles inhibits the possible
agglomeration of nanoparticles – hence imparting stability to the nanoparticle
dispersion. However, most of the common surfactants undergo thermal degrada-
tion (and subsequently detach from the nanoparticle surface) at high tempera-
tures; therefore this mode of stabilizing nanoparticle dispersions is not suitable for
solar thermal applications (Ghadimi et al. 2011).
Adjusting the pH: The pH of the nanoparticle dispersion should be such that it is far
away from the isoelectric point (IEP). At IEP the zeta potential of the fluid is zero.
For stable nanoparticle dispersions, the pH should be such that the absolute value
of the zeta potential is sufficiently high (Ghadimi et al. 2011).
Surface modification of the nanoparticles: This essentially means coating/capping
the nanoparticles with some functional group which imparts hydrophilic charac-
ters to the otherwise hydrophobic nanoparticles or vice versa. Plasma deposition
process has been found to be very effective in surface modification of carbon
nanotubes (CNTs), imparting controllable hydrophilicity to the CNTs and hence
improving the stability of CNT-based dispersions (Kim et al. 2010; Lamas et al.
2012; Hordy et al. 2014). Researchers have very recently reported that nanopar-
ticle dispersions employing plasma-treated multiwalled carbon nanotubes
(MWCNTs) have shown excellent stability over extended periods of time (up to
8 months) and high temperatures of about 170 C (Hordy et al. 2014). Overall, the
plasma functionalization of MWCNTs (proposed recently by (Hordy et al. 2014))
promises enhanced stability at high temperatures and hence presents a practical
solution to the agglomeration issue.
Thermophysical, Rheological, and Optical Properties

of Nanofluids
Among the thermophysical properties, thermal conductivity and specific heat capac-
ity have been widely researched. These two properties play a pivotal role in
determining the effectiveness of nanofluids as heat transfer liquid. Firstly, let us try
to understand the effect of adding nanoparticles on the thermal conductivity of the
basefluid. Choi and Choi and Eastman (1995) have predicted an enhancement in
thermal conductivity of the basefluid through addition of metallic nanoparticles.
Since these initial studies, nanofluids have received enormous interest. However, this
has also led to numerous conflicting claims by various researchers, and in the year
2009, an International Nanofluid Property Benchmark Exercise (INPBE) was

reported in order to reach a consensus. In this exercise thermal conductivity mea-
surements of identical nanofluid samples were carried out by over 30 organizations
worldwide. Various experimental techniques such as transient hot-wire method,
steady-state methods, and optical methods were employed to accomplish the thermal
conductivity measurements. The thermal conductivity of nanofluids was found to
increase with particle volume fraction and aspect ratio. Furthermore, the generalized
theory by Nan et al. (1997) was found to be in good agreement with the experimental
data, suggesting that no anomalous enhancement of thermal conductivity was
achieved.
As for specific heat capacity is concerned, it has been found that effective specific
heat of nanofluids can be approximated by parallel mixture rule. Specific heat of
solids being generally low compared to that of liquids, nanofluids have slightly low
specific heat capacity compared to basefluids. In essence, addition of nanoparticles
to the basefluid does alter its thermophysical properties. However, the observed
enhancement is not enough (at low volume fractions) to bring about a significant
change in effectiveness of the basefluid as heat transfer liquid.
In order to ascertain the effect of nanoparticles on the pumping power require-
ment of heat transfer liquids, it is imperative to understand the rheological properties
of nanofluids. More viscous the heat transfer liquid, more is the pumping power
requirement. Numerous studies pertinent to viscosity measurements of nanofluids
are well documented in the literature Murshed et al. (2008). A general consensus is
that viscosity increases with increase in volume fraction of the nanoparticles. Hence
addition of nanoparticles to the basefluid tends to increase the pumping power
requirement.
One of the properties which is based on a more sound scientific rationale and
better understood among the scientific community is the optical tunability of these
nanofluids. Addition of trace amounts of nanoparticles to the basefluid has shown to
enormously alter the optical properties of the basefluid. Optical properties could be
easily tailored as per specific application through careful control of nanoparticle size,
shape, material, and volume fraction.
Relevance of Nanoparticle-Laden Fluid Flow in Solar Thermal

Systems
Limitations of Surface Absorption-Based Solar Thermal Systems
Most of today’s solar thermal technologies that are being used at commercial scales
are constructed with surface-based absorbers, wherein solar irradiance is first
absorbed by a metal surface (usually employing selective coatings) and then trans-
ferred to a working fluid through conduction and convection modes of heat transfer
(see Fig. 1a). Selective coatings have high absorptivity across the spectrum of
incident solar energy (0.3–2.5 μm) while simultaneously having low emissivity in
the mid-infrared region (2.5–10 μm) of the spectra (Ritchie and Window 1977).
6 V. Khullar et al.
a b
SUN
SUN
Conduction Solar selective Transparent

absorber cover
Volumetric absorption
Convection
Flow of fluid Working f luid Flow of fluid
Fig. 1 Schematic showing the heat transfer mechanism in (a) surface absorption-based solar
thermal technologies and (b) volumetric absorption-based solar thermal technologies
These characteristics make selective coatings quite efficient for harnessing incident
solar irradiance. However, within the surface-based absorption systems, an inherent
thermal barrier is present in the form of conduction and convection resistances
between the absorbing surface and the working fluid. As a result, conversion of
solar energy into thermal energy of the working fluid is not a very efficient process
in such systems (Lenert 2010). This issue becomes even more pronounced in the case
of concentrating solar collectors as the flux that must be transferred across this thermal
barrier increases manifolds. In these cases, excessively high surface temperatures can
develop relative to fluid temperature – e.g., reported temperature differences (between
the surface and fluid temperatures) have been as high as 500 C (Abdelrahman et al.
1979). Since radiative losses are proportional to the fourth power of the surface
temperature, high surface temperatures lead to excessively high radiative heat losses.
One of the ways this problem can be addressed is by allowing the incident solar
radiation to interact directly with the working fluid without heating any other struc-
tures within the receiver (see Fig. 1b). Here, instead of surface absorption, there would
be volumetric absorption of the incident solar irradiation (i.e., absorption occurring
throughout the medium and not just confined at the surface).
Emergence of Nanoparticle-Laden Fluid Flows for Solar Thermal

Applications
Many of the commonly used heat transfer fluids (such as ethylene glycol, Therminol
VP-1, molten salts) are not very good absorbers of solar radiant energy because they
are transparent within a large range of the solar spectrum (Drotning 1978; Otanicar
et al. 2009). Therefore, additives are needed in order to create working fluids which
may directly absorb solar energy. Seeding the working fluid with micron-sized
particles, such as carborundum and silicon dioxide, can significantly enhance the
fluid’s solar energy absorption capability (Arai et al. 1984). However, micron-
particle suspensions are not very stable, and the particles tend to settle down.
Practical difficulties, such as clogging of pumps and valves, have therefore limited
the use of micron-particle suspensions as volumetric absorbers (Otanicar
et al. 2010a).
On the other hand, recent studies have shown that dispersing trace amounts of
nanoparticles can significantly enhance the solar irradiance absorption capability of
the basefluid– by up to seven orders of magnitude change in the extinction coeffi-
cient for 1% volume fraction (Taylor et al. 2011). Excellent stability (relative to
micron-particle suspensions) coupled with high solar energy absorption capability
makes nanofluids potential candidates for solar thermal applications (Tyagi et al.
2009; Lenert and Wang 2012; Veeraragavan et al. 2012; Khullar et al. 2013, 2014;
Khullar 2015). Moreover, the fact that the optical properties of metallic and semi-
conductor nanoparticles are dependent on their shape, size, and the surrounding
dielectric media (Eustis and El-Sayed 2006; Tan et al. 2012; Taylor et al. 2012,
2013a; Neumann et al. 2013) makes them easily tunable for engineering highly solar
selective nanoparticle-laden fluids at very low volume fractions.
Within the surface absorption-based solar thermal systems, the solar radiant
energy is first absorbed by the selectively coated surface and then transferred to
the working fluid via conduction and convection modes, i.e., the working fluid itself
does not directly interact with the incident solar irradiance. However, in the case of a
volumetric absorption nanoparticle-laden fluid-based solar thermal system, the
working fluid is allowed to directly interact with the incident solar irradiance, and
as a result volumetric energy absorption takes place via absorption and scattering
mechanisms (as shown in Fig. 2). Furthermore, due to the chaotic movement of the
nanoparticles, the absorbed energy is redistributed within the fluid therefore resulting
in more uniform temperature distribution relative to their surface absorption-based
counterparts.
Can Nanoparticle-Laden Fluid Flow Be Engineered to Compete

with Solar Selective Surfaces?
Conditions in which Nanoparticle Dispersion Can Have Low

Radiative Losses as Compared to Solar Selective Surfaces
As such, nanoparticle-laden fluid can be engineered to absorb almost the entire solar
irradiance incident on it. However, inherently high emissivity of the basefluid (in the
mid-infrared region) results in nanofluid which has emissivity values very close to
1. In the backdrop of the aforementioned facts, nanoparticle dispersion can be
thought to approximate a perfect black body.
Figure 3 shows radiative losses as a function of temperature for a perfect black
body, solar selective surface with cutoff λ of 1.5 μm, and solar selective surface with
cutoff λ of 2.0 μm. It is clear from the plots that for a given temperature, radiative
8 V. Khullar et al.
Spectral electromagnetic radiations (solar irradiance)
Scattering
and
absorption
Absorption
Basefluid
Nanoparticles
Fig. 2 Attenuation of solar irradiance as it passes through the nanofluid
Fig. 3 Radiative losses as a function of temperature in the case perfect black body, solar selective
surface (SSS) with cutoff λ of 1.5 μm, and solar selective surface (SSS) with cutoff λ of 2.0 μm
losses are bound to be higher for a perfect black body (and similarly for a nanofluid-
based solar thermal system too).
However, the following are the conditions under which nanoparticle dispersion
can have low radiative losses as compared to solar selective surfaces:
• Sufficiently high “temperature overheat” develops, resulting in high surface

temperature relative to the fluid temperature in the case of surface absorption-
based absorbers.
• Nanoparticle distribution is so engineered that highest temperatures occur away
from the free liquid surface, resulting in average fluid temperatures sufficiently
higher than that of the free liquid surface.
Let us quantitatively analyze the aforementioned cases by using some hypothet-

ical temperatures. Consider a solar selective surface at 1400 C, and due to ineffec-
tive convective heat transfer, let the average fluid temperature be 1100 C
(a temperature overheat of 300 C). Now, let us consider the case of a nanofluid-
based direct absorption collector, average temperature of the nanoparticle dispersion
being 1100 C and temperature at the free liquid surface being 800 C (achieved
through careful design of nanoparticle distribution). Under such conditions (see
Fig. 3) nanoparticle dispersions can have lower radiative losses as compared to the
solar selective surfaces (depending on the cutoff wavelength value, for instance in
the case of cutoff wavelength of 1.7 μm); however, such conditions are yet to be
practically realized.
As mentioned earlier, it is practically difficult to ensure low radiative losses from
nanoparticle-laden fluid flow; however, recently, a few researchers have pointed out
that if this flow is enveloped by transparent heat mirrors (transparent heat mirrors are
optical elements which have high transmissivity in the solar irradiance wavelength
band and high reflectivity in the mid infrared region), then it is possible to curb
radiative losses (Hewakuruppu et al. 2015; Khullar 2015; Khullar et al. 2016).
Again, comprehensive experimental investigation into practical realization of such
systems is yet to be verified.
Thermal Modeling of Nanofluid-Based Volumetric Absorption

Solar Thermal Systems
Overview
A nanofluid-based volumetric absorption solar thermal system is similar to its

surface absorption-based counterpart except for the receiver part. Both flat and
cylindrical receiver geometries have been proposed, and heat transfer models have
been formulated for these geometries. Figure 4 compares the nanofluid-based vol-
umetric receiver designs with their conventional surface absorption-based counter-
parts. In case of flat geometry, the selectively coated absorber surface has been
replaced by a transparent glass envelope, and the conventional HTL has been
10 V. Khullar et al.
a Selectively coated b Low iron

metal surface glass envelope
HTF flow Nanoparticle-laded

f luid f low
c Glass Envelope d
Nanoparticles
Heat Transfer Heat Transfer

Liquid Liquid
Air/Vacuum Air/Vacuum
Solar Selective Metal Low Iron Glass Tube
Surface
Fig. 4 Schematic of (a) conventional flat receiver, (b) nanofluid-based flat receiver, (c) conven-
tional cylindrical receiver, and (d) nanofluid-based cylindrical receiver
replaced with nanoparticle-laden fluid. In cylindrical geometries the absorber tube is

replaced by a glass tube which houses the nanofluid so that direct interaction of the
working fluid with the solar irradiance is possible.
In the conventional receiver designs, the solar radiant energy is first absorbed by
the selectively coated absorber surface and then transferred to the working fluid via
conduction and convection modes, i.e., the working fluid itself does not directly
interact with the incident solar irradiance. However, in the case of a nanofluid-based
volumetric absorption receiver, the working fluid is allowed to directly interact with
the incident solar irradiance, and as a result volumetric energy absorption takes place
via absorption and scattering mechanisms.
Basic Modeling Framework
Nanoparticle-laden fluid flow could be modeled as a participating media flow which

scatters, absorbs, as well as emits the electromagnetic radiations in addition to the
convective and conduction modes of heat transfer. In other words, nanoparticle-
laden fluid directly interacts with the incident solar irradiance, and the absorbed
energy is redistributed through convection, radiation, and conduction modes of heat
transfer.
In essence, the basic formulation draws its inspiration from the problem of
solving combined heat transfer modes for a participating media. Analytical solutions
to such problems are available in the literature for gray media enclosed in simple
geometries (Modest 2003)
However, heat transfer formulation relevant to nanoparticle-laden fluid flows
typically involves combined modes of heat transfer in non-gray media
(as nanoparticle-laden fluid typically has spectrally dependent optical properties).
As a result of which an energy balance applied to such systems fetch coupled
nonlinear integrodifferential equations. Closed-form analytical solution to such
equations is difficult to achieve, and one needs to resort to numerical techniques
for solving such equations. Numerical techniques such as finite difference method
(FDM), finite volume method (FVM), and finite element method (FEM) transform
the integrodifferential equations into a set of more solvable algebraic equations.
Many researchers have used the aforementioned numerical techniques and suc-
cessfully implemented their algorithms in MATLAB®. These attempts have been
successful in understanding and identifying the dominant heat transfer mechanisms
involved in nanoparticle-laden fluid flows. However, these are limited to simple
geometries only and are very difficult to be extended to complex geometries.
This has paved the way for using commercial software packages such as
COMSOL® for modeling such systems (Chang et al. 2011; Kluczyk and Jacak
2016). Specifically, the wave optics module of COMSOL provides solution for
optical scattering of nanoparticles like nanospheres as well as periodic nanorods.
There is also a Brewster interface model through which the interfaces/boundaries for
electromagnetic wave propagation can be accounted. Coupling of above modules
with heat transfer module may estimate the resulting temperatures. Another versatile
modeling tool, Monte Carlo coupled to FEM, can be deployed to model a nanofluid-
based direct solar thermal absorber (Lee et al. 2012).
Fundamental Governing Equations
We need to solve the basic energy balance equations in order to quantitatively

analyze the heat transfer mechanisms and predict the thermal performance of the
aforementioned nanoparticle-laden fluid-based volumetric absorption solar thermal
systems. Essentially, overall energy balance and optical energy balance equations
form the two equations which need to be solved.
Overall Energy Balance

The heat transfer within the nanoparticle-laden fluid flow could be approximated as
two-dimensional steady state and is represented by Eq. 1:

1 @ @T @ ðrqrad Þ @T
kr ¼ ρcp U , (1)
r @r @r r@r @x
ÐÐ
where the divergence of radiative flux (qrad = Iλ,ψ dψdλ) in the left hand side
of Eq. 1 represents the energy generation term of the generalized energy balance
equation. Here, T, k, U, and cp are the temperature, thermal conductivity, velocity,
and specific heat of the fluid.
Optical Energy Balance: Radiative Transfer Equation (RTE)

Radiative transfer equation represents the optical energy balance in participating
media. Absorption, emission, and scattering (out-scattering and in-scattering) are the
three processes responsible for influencing the optical energy balance. The
one-dimensional RTE in Cartesian and cylindrical coordinates (Modest 2003; Ben
Salah et al. 2004) can be mathematically represented by Eqs. 2 and 3, respectively:
dI λ
¼ K eλ I λ þ K absλ I bλ , (2)
dy
1 @ ðrI λ Þ
¼ K eλ I λ þ K absλ I bλ , (3)
r @r
where Keλ and Kabsλ are the spectral extinction coefficient and absorption coef-
ficients of the nanoparticle-laden fluid and Iλ is the spectral radiative intensity. The
two terms on the right hand side of Eqs. 2 and 3 represent attenuation of the incident
solar irradiance (absorption and out-scattering) and augmentation (thermal
reemission), respectively. The effect of in-scattering has not been accounted in
most of the proposed models owing to its negligible magnitude.
Now that we have got the governing equations relevant to nanoparticle-laden fluid-
based solar thermal systems, next we need to convert these nonlinear
integrodifferential equations into more solvable algebraic equations. As mentioned
earlier, this could be done using numerical techniques such as FVM, FDM, and FEM.
The generalized energy balance equation has been transformed from its nonlinear
integrodifferential form to a set of simplified algebraic equations. However, as the RTE
and the generalized energy balance equations are coupled through temperature distri-
bution, therefore an algorithm needs to be devised to solve these coupled equations.
Modeling Optical and Thermophysical Properties of Nanoparticle-

Laden Fluid
Modeling Optical Properties

Obtaining a suitable nanoparticle-laden fluid requires a fundamental understanding of
how light interacts with the nanoparticles. Once optical constants are known for the bulk
material, these values can then be employed for finding the corresponding values for
nanoparticles. Subsequently, spectral absorption coefficients could be computed for
various nanoparticle geometries and materials. Finally, solar-weighted absorptivity is
calculated for these geometries, and optimum nanoparticle geometry and material are
identified.
Nanoparticles typically have dimensions on the order of 0.2–100 nm. At this scale
the optical properties are drastically different from those at bulk and atomic/molec-
ular levels (Biju et al. 2008). Moreover, optical properties are highly sensitive to the
size, the shape of the nanoparticle material, and the surrounding dielectric media
(i.e., the basefluid in which nanoparticles are dispersed).
Nanoparticles scatter as well as absorb the solar irradiance. Relative to the
nanoparticles, scattering by the basefluid is negligible and therefore has not been
accounted for in most of the proposed models. Furthermore, scattering by the
nanoparticles has been approximated as independent scattering (the volume fraction
of the nanoparticles being very small) following Rayleigh’s regime (Bohren and
Huffman 1983). This can be justified as the size of the nanoparticles is small as
compared to the wavelength of the interacting solar irradiance, i.e., α ¼ πD λ 1,
where α is the size parameter.
Nanoparticle interaction with the spectral solar irradiance could be approximated
as dipoles under electromagnetic radiations provided that the particles size is small
relative to the wavelength of the incident EM radiation. Taking advantage of the
aforesaid approximation, researchers (Draine and Flatau 1994) have devised a very
power tool known as discrete ordinate approximation (DDA) in order to calculate
scattering and absorption of complicated nanoparticle geometries. Easily
implementable codes are available as freeware on the web which could be run on
a variety of operating systems (LINUX, WINDOWS, and UNIX).
Modeling Thermophysical Properties

Parallel mixture rule (Zhang et al. 2006; Das et al. 2008) gives sufficiently accurate
approximation of the heat capacity of the nanoparticle-laden liquid:
ρnf cp,nf ¼ f v ρnp cp, np þ ð1 f v Þρbf cp,bf , (4)
where fv is the volume fraction of the nanoparticles; ρnf , ρnp , ρbf are the densities;
and cp,nf, cp,np, and cp,bf are the specific heats of the nanofluid, nanoparticles, and the
basefluid, respectively.
Thermal conductivity of the nanoparticle-laden fluids has been predicted by many
researchers (Maxwell 1881; Hamilton and Crosser 1962; Nan et al. 1997; Wang et al.
2003; Prasher et al. 2005; Zhang et al. 2006). Among the proposed models, the
Maxwell effective medium model (Maxwell 1881), the predictions made by Ham-
ilton and Crosser (1962), and more recently the generalized model by Nan et al.
(1997) have been found to be consistent with the experimental findings. For spher-
ical nanoparticles and negligible interfacial thermal resistance (Buongiorno et al.
2009), the last two models essentially result in the same equation as proposed by
Maxwell (see Eq. 5):

knf knp þ 2kbf þ 2fv knp kbf
¼ , (5)
kbf knp þ 2kbf fv knp kbf
where knf , kbf, and knp are the thermal conductivities of the nanofluid, basefluid,
and nanoparticles, respectively, and fv is the volume fraction of the nanoparticles. At
low volume fractions of the nanoparticles, the thermophysical properties (heat
capacity and the thermal conductivity) of the nanoparticle-laden fluid are similar to
those of the parent basefluid. This has been recently verified by the experiments
conducted by Lenert and Wang (2012).
Performance Evaluation of Nanoparticle-Laden Fluid-based Solar

Thermal Systems
Performance evaluation of nanoparticle-laden fluid-based solar thermal systems

could be done on the same fundamental equations as employed for assessment of
their surface absorption-based counterparts. For this purpose optical and thermal
efficiencies need be evaluated:
Gnf
ηopt ¼ , (6)
Cratio Gaper
ρ cp ðT out T in Þ
ηth ¼ , (7)
Cratio Gaper
where Gnf is the solar flux that is able to reach the nanofluid, Cratio is the geometric
concentration
ratio defined as the ratio of the aperture area to the receiver area
Aaper
Cratio ¼ Arec , Gaper is the direct normal incident solar irradiance at the aperture,
ρ and cp are the density and specific heat of the nanofluid, and Tout and Tin are the
mean outlet and inlet temperatures, respectively.
As a representative example, a typical nanoparticle-laden fluid-based volumetric
receiver design has been analyzed for its performance characteristics. Furthermore,
the aforementioned receiver design has been compared with the surface absorption-
based receivers (cermet-based receiver); the geometric dimensions of the
nanoparticle-laden fluid-based volumetric receivers have been kept similar to that
of the cermet-based receiver (Dudley et al. 1995).
From practical standpoint the volumetric absorption receivers employing nano-
fluids as the working fluid are relevant. However, in order to clearly ascertain the
effect of adding nanoparticles to the basefluid, the cases of receivers which are
exactly similar to the nanofluid-based receivers but use conventional heat transfer
fluids such as Therminol VP-1 have been considered.
Figure 5 outlines the thermal efficiency and typical losses (per unit aperture area)
in the three types of receivers. The essential revelation from Fig. 5a is that there is a
definite improvement in terms of thermal efficiency in the case of the nanofluid-
based volumetric receiver. Basefluid-based volumetric receivers due to their low
a 100 b 120
100
80
Thermal Losses (Wm )

Thermal Efficiency (%)
−2
80
60
Nanofluid-based volumetric absorption receiver
60
Basefluid-based volumetric absorption receiver
40 Cermet based surface absorption receiver
40
20
20
0 0
0 50 100 150 200 250 0 50 100 150 200 250
Average Fluid Temperature above Ambient (°C) Average Fluid Temperature above Ambient (°C)
c 800
700
Optical Losses (Wm )
−2
600
500
400
300
200
100
0
0 50 100 150 200 250
Average Fluid Temperature above Ambient (°C)
Fig. 5 (a) Thermal efficiency as a function of average fluid temperature above ambient, (b) thermal
losses as a function of average fluid temperature above ambient, and (c) optical losses as a function
of average fluid temperature above ambient
thermal efficiencies do not warrant further consideration, and therefore from this
point onward, the analysis is restricted to nanofluid-based volumetric receivers only.
In the case of nanofluid-based volumetric receivers (see Fig. 5a), the maximum
improvement in thermal efficiency is for low values of fluid temperature, and the gap
between the two narrows down with increase in operating fluid temperatures. This
observation can be attributed to the fact that at higher average fluid temperatures, the
total losses for the two collectors tend to converge as the effect of thermal losses
becomes predominant. It is apparent from the plots that optical losses (see Fig. 5c)
are very high in the case of basefluid-based volumetric receivers as they allow most
of the solar radiant energy to pass through.
These results indicate that under carefully chosen operating conditions, the
nanofluid-based volumetric absorption solar thermal systems outperform their sur-
face absorption-based counterparts. It has been reported that thermal efficiency
enhancement of the tune of 5–15% is achievable (Khullar et al. 2013; Khullar
2015). Furthermore, these observations have recently been validated by proof-of-
concept experiments employing nanofluid as the working fluid (Khullar et al. 2014).
16
Table 1 Thermophysical properties of various heat transfer liquids

Boiling point/
upper Freezing Thermal Specific Range in which optical
Name of Visual temperature point conductivity heat constants (n and κ values)
basefluid Formula/composition appearance [ C] [ C] [Wm1 K1] [Jkg1 K1] are known
Water H2O Colorless 100 0 0.598 4180 0.2 μm200 μm
Therminol Eutectic mixture of 73.5% Clear 257/400 – 0.136 1546 0.2 μm–2.5 μm
VP-1 diphenyl oxide and 26.5% liquid
biphenyl
Therminol 55 Synthetic hydrocarbon Clear 351 – 0.127 1930 –
mixture yellow
liquid
Syltherm 800 Silicone oil Colorless >315 46 0.14 1600 –
(Si(CH3)2O-)n
Ethylene C2H4OH Colorless 195–198 – – – 0.2 μm–1.5 μm
glycol
Paraffin oil – Colorless 311 9.5 0.13 2100 –
Polyethylene – Colorless – – – – 0.2 μm–1.5 μm
glycol
Binary nitrate NaNO3:KNO3 (60:40 by – 600 220 – 1495 –
(solar salt) weight %)
Hitec NaNO3:KNO3: NaNO2 – 535 142 – 1560 –
(7:53:40 by weight %)
Hitec XL NaNO3:KNO3: Ca(NO3)2 – 120 500 – 1447 –
(calcium (7:45:48 by weight %)
nitrate salt)
V. Khullar et al.
Table 2 List of nanoparticles and nanoparticle dispersions previously studied
Nanoparticles Nanoparticle dispersion
Preparation Surfactant/
process/ Preparation sonication Known optical
Material Morphology procured Size Pretreatment Basefluid process time properties Reference
Carbon- Spheres Nano Amor D = 28 nm – Therminol Two-step –/30 min Absorption (Lenert and
coated Inc. VP-1 process coefficient Wang
cobalt (0.4–1.6 μm) 2012)
Carbon Nanohorns Carbonium D = 2–5 nm L = – Ethylene Two-step – Extinction (Sani et al.
Srl-Italy 30–50 nm cone glycol process coefficient 2011)
angle =20 (0.2–1.4 μm)
MWCNTs Thermal- D = 30 nm Plasma Ethylene Two-step –/15 min Transmittance (Hordy
chemical L = 4 μm functionalization glycol, process (0.2–1.5 μm) et al. 2014)
vapor polyethylene
deposition glycol,
Nanoparticle-Laden Flow for Solar Absorption
(T-CVD) denatured
alcohol
Shenzhen – Oxidation Ethylene Two-step – Transmittance (Meng et al.
Nanoport treatment with glycol process (0.2–2.5 μm) 2012)
Company HNO3
CNT Co. Ltd., D = 30 nm L = – Deionized Two-step CTAB/ – (Lee and
Korea, T-CVD 0.5 μm water process 60 min Jang 2013)
Research D = 10 nm Chemical base Water Two-step –/30 min Transmittance (Karami
Institute of L = 5 μm–10 μm treatment process (0.2–1.5 μm) et al. 2014)
Petroleum
Industry of
Iran
Nanoballs Research D <100 nm – Ethylene Two-step Sodium Transmittance (Karami
Institute of glycol, Water process dodecyl (0.2–1.5 μm) et al. 2013)
Petroleum sulfate/
30 min
17
(continued)
18
Table 2 (continued)
Nanoparticles Nanoparticle dispersion
Preparation Surfactant/
process/ Preparation sonication Known optical
Material Morphology procured Size Pretreatment Basefluid process time properties Reference
Industry, spray
pyrolysis
Graphite Advance D = 30 nm – Water Two-step Sodium Extinction (Taylor
Materials, process dodecyl coefficient et al. 2011)
Sigma sulfate/ (0.2–2.5 μm)
Aldrich, 15–30 min
NanoAmor
Aluminum – Advance – – Water Two-step Sodium Extinction (Taylor
NanoAmor
Nickel Spheres Vacuum D = 4.9 nm – Alkyl One-step – Absorption (Hanamura
evaporation naphthalene process coefficient and
onto a running (0.3–10 μm) Kameya
oil substrate 2011)
technique
Gold Spheres Citrate D= – Deionized One-step –/30 min Transmittance (Zhang
reduction 10 nm–30 nm water process (0.3–0.7 μm) et al. 2014)
method
V. Khullar et al.
Silver – Advance D = 20 nm – Water Two-step Sodium Extinction (Taylor
Aldrich, Nano 15–30 min
Amor
Copper – Advance – – Water Two-step Sodium Extinction (Taylor
NanoAmor
ZnO – – D = 10 nm – Deionized Two-step E80/60 min Transmittance (Zhu et al.
water process (0.2–2.0 μm) 2012)
AlN – – D = 40 nm – Deionized Two-step Arabic Transmittance (Zhu et al.
water process gum/60 min (0.2–2.0 μm) 2012)
ZrC – – D = 40 nm – Deionized Two-step Arabic Transmittance (Zhu et al.

TiN – – D = 20 nm – Deionized Two-step Arabic Transmittance (Zhu et al.
TiO2 – Advance – – Water Two-step Sodium Extinction (Taylor
NanoAmor
19
Nanofluid-Based Solar Thermal Systems
Potential Basefluids
Ideally, heat transfer liquids should be optically nonparticipating, in order to be

employed as basefluids for nanofluid preparation. This ensures that simply through
addition of nanoparticles, the nanofluid of desired optical properties can be
engineered. Furthermore, high boiling point, high specific heat, and chemical and
thermal stability are some of the other properties which lend a basefluid to be suitable
for nanofluid preparation. Table 1 lists various heat transfer liquids along with their
thermophysical properties and the state of the known optical properties.
Nanofluids Relevant to Solar Thermal Systems
In the last 5 years, researchers have explored the candidature of various nanofluids
for solar thermal systems. Primarily, these include carbon-based, metal nanoparticle-
based, and metal oxide nanoparticle-based nanofluids. Table 2 details a selected list
of the nanofluids investigated, along with the details of preparation method
employed.
Solar Thermal Systems Employing Nanofluids as the Working Fluid
Various solar thermal systems employing nanofluids have been theoretically as well
as experimentally analyzed. Most of the reported studies focus on flat receiver
geometries. Furthermore, most of the studies employ monodispersion nanofluids
as the working fluid. Table 3 details various characteristics and typical performance
trends relevant to nanofluid-based solar thermal systems.
The common conclusions that can be drawn from all these reported studies are
that:
• There is a definite improvement in solar irradiance absorption capability through

addition of nanoparticles.
• If operating parameters are carefully chosen, higher thermal efficiencies can be
achieved in the case of nanofluid-based direct absorption solar thermal systems.
In the backdrop of the aforementioned facts, it is worthwhile to explore these

nanofluid-based direct absorption solar thermal systems in greater detail. Thus both
theoretical as well as experimental studies relevant to these systems are warranted.
Table 3 Selected list of solar thermal systems employing nanoparticle-laden fluid as the working fluid
Solar thermal system Major findings
Concentrating E: Experimental
type (CT)/non-
concentrating Nanoparticle Operating
type (NCT) Receiver geometry dispersion temperatures
Shape Dimensions Irradiance source employed ( C) M: Modeling References
CT, Potential D’ = 0.0635 m 1.6 kW solar C-coated E: 20–50 C M: 1. Receiver efficiency (Lenert and
application in simulator (fully cobalt/ increases with increasing Wang 2012)
central Cylinder reflective, Therminol nanofluid height and
receivers Sciencetech Inc.) VP-1 incident solar flux
H = 0.06 m M: 2. When coupled to a power
127–1127 C cycle, combined optimum
efficiencies exceeding 35%
can be achieved for Cratio

>100 and H >5 cm
E: Experimental results are
in good agreement with the
model predictions
NCT L’ = 1.5 m Sun Hypothetical M: 20–36 C M: 1. Tailoring extinction (Otanicar
W = 1.5 m profile as a function of depth et al. 2011)
Cuboid improves thermal efficiency
H = 0.01 m 2. Efficiency improvement
of 6% as compared to
conventional surface
absorption-based systems
NCT D’ = 0.026 m Sun Carbon black/ E: 24–40 C E: 1. The shear viscosity (Han et al.
distilled increases with increasing 2011)
water volume fraction and
Cylinder
decreases with increase in
21
(continued)
Table 3 (continued)
22

type (CT)/non-
temperature at the same
shear rate
2. Thermal conductivity of
carbon black nanofluids
increases with increase in
volume fraction and
temperature
H = 0.150 m 3. Carbon black nanofluids
have good solar absorption
efficiency
NCT D’ = 0.026 m Sun CNTs/ E: 24–40 C E: 1. CNTs with oxidation (Meng et al.
ethylene treatment exhibit good 2012)
glycol dispersing performance
Cylinder
H = 0.150 m 2. At room temperature, 18%
enhancement was found in
the photothermal conversion
efficiency of the 0.5% mass
fraction CNTs glycol
nanofluids in comparison to
the basefluids
NCT D’ = 0.035 m Solar simulator Gold E: 20–34 C E: 1. GNPs have good (Zhang et al.
(Newport Co. Oriel nanoparticles photo-thermal conversion 2014)
Cylinder Xenon Arc lamp) (GNP)/water efficiency, at particle
concentration of 0.15 ppm,
V. Khullar et al.
GNP increases the
photothermal conversion
efficiency of basefluid by
20% and reaches a specific
absorption rate of 10 kWg1
H = 0.003 m 2. Photothermal conversion
efficiency increases with
increasing volume fraction
NCT, flat plate A’ = 1.51 m2 Sun MWCNT/ E: 30–40 C E: Efficiency of solar (Yousefi et al.
collector water collector increases as the pH 2012a)
is increased or decreased
Cuboid with respect to the isoelectric
point
NCT, flat plate A’ = 1.51 m2 Sun Al2O3/Water E: 30–40 C E: Using 0.2 Wt% Al2O3 (Yousefi et al.
collector nanofluid as absorbing 2012b)
medium in a flat plate solar

Cuboid water heater increases the
efficiency by 28.3%
CT, parabolic L’ = 0.02 m Sun Graphite/ E: 270 C M: 1. In case of nanofluid- (Taylor et al.
dish collector W = 0.02 m Therminol based receivers, efficiency 2011)
Cuboid VP-1 enhancement of 10% can be
achieved relative to surface
absorption-based receivers
when concentration ratios
are in the range of 100–1000
H = 0.001 m 2. Graphite nanofluids with
volume fraction on the order
of 0.001% are suitable for
10–100 MWe power plants
E: Experiments on
laboratory scale nanofluid-
23
(continued)
Table 3 (continued)
24

type (CT)/non-
based dish receiver suggest
that 10% increase in
efficiency can be achieved
relative to surface absorption
collector if operating
conditions are carefully
chosen
CT H= Sun Graphite/ – M: 1. An analytical model Veeraragavan
0.001 m–0.01 m Therminol was formulated to et al. 2012
Cuboid VP-1 investigate the effect of heat
loss, particle loading, solar
concentration, and channel
height on the receiver
efficiency
2. Model predicts an
optimum total efficiency of
0.35 for a volumetric
receiver (dimensionless
receiver length of 0.86)
employing graphite
nanoparticles dispersed in
Therminol VP-1
V. Khullar et al.
NCT H = 0.0012 m Sun Aluminum/ M: 34–35 C M: 1. The presence of (Tyagi et al.
Water nanoparticles increases the 2009)
Cuboid absorption of incident
radiation by more than nine
times over that of pure water
2. Under similar operating
conditions, the efficiency of
a direct absorption collector
using nanofluid as the
working fluid is found to be
up to 10% higher than of
conventional flat plate
collector
CT A’ = 0.0015 m2 SuperPAR64 lamp Graphite, E: 25–45 C E: 1. Efficiency (Otanicar
(color temperature CNT, silver/ improvements of up to 5% in et al. 2010a)
Cuboid of 3158 K) water solar thermal collectors by

utilizing nanofluids as the
absorption mechanism
H= 2. Experimental results
0.000150 m demonstrate an initial rapid
increase in efficiency with
volume fraction, followed by
leveling off in efficiency as
volume fraction continues to
increase
CT, potential D’ = 0.008 m, Artificial light Amorphous E: 95 C E: Under similar operating (Khullar et al.
application in 0.035 m source with optical carbon/ conditions, higher average 2014)
central Cylinder H = 0.008 m, fiber, color ethylene stagnation temperatures are
receivers 0.010 m temperature glycol achievable in the case of
3400 K MWCNT/ nanofluid-based volumetric
distilled absorption systems as
water compared to solar selective
25
surface absorption-based
system
(continued)
Table 3 (continued)
26

type (CT)/non-
CT, Fresnel lens L’ = 0.1 v m Sun Graphite/ E: 70 C M: Increase in concentration (Kaluri et al.
W = 0.05 m water ratio of solar radiation 2015)
Cuboid significantly enhances the
system efficiency
H = 0.002 m CuSO4 M: 70 C E: Compared to the system
solution employing basefluid the
absorbing liquid based
system has 28% higher
thermal efficiency
V. Khullar et al.
Conclusions and Future Scope
Extensive research has being going on in recent years into the theoretical as well as
experimental aspects of applications of nanoparticle-laden flows in direct absorption
solar volumetric collectors. In principle, it can be concluded from these investiga-
tions that increase in efficiency can definitely be achieved relative to surface
absorption-based solar collector if operating conditions are carefully chosen. Con-
tinued research into these systems is warranted in order to realize such nanofluid-
based collectors on a commercial scale. As for as the theoretical modeling pertinent
to nanofluid-based volumetric collectors is concerned, investigators have developed
various numerical (Tyagi et al. 2009; Otanicar et al. 2010a; Lenert and Wang 2012)
and analytical (Veeraragavan et al. 2012) heat transfer models for predicting the
efficiency of the collectors employing nanoparticle dispersions. These attempts have
been quite successful in understanding the radiative and conductive heat transfer
mechanisms in direct absorption nanofluid-based receivers. On the experimental
front, the investigations have been carried out on both miniature collectors having
stagnant nanofluid layers as well as nanofluid-based micro-channel flow collectors
under laboratory conditions. Majority of these studies are limited to low operating
temperatures (maximum operating temperatures being approx. 270 C). However,
the real benefit is expected to be demonstrated at higher operating temperatures
(typically 400–500 C) and under high solar concentration ratios owing to the fact
these systems shall offer better conversion efficiency when large amount of solar
irradiance needs to be transferred to the working fluid. Solar power towers seem to
be the most promising solar thermal systems where these nanofluids could be used as
the working fluid to volumetrically absorb the solar thermal energy. Furthermore,
heat mirror-based covers could be engineered into these systems to further improve
upon the efficiency of nanoparticle-laden fluid-based solar thermal systems. In terms
of the choice of nanoparticle material is concerned, recent improvements in the
stability of MWCNT-based nanofluids and the fact that they have broad absorption
spectra lend carbon-based nanoparticle to be the most appropriate and inexpensive
proposition for volumetric absorption of the solar thermal energy. Currently, there is
real impetus to fabricate and test nanofluid-based solar volumetric collectors under
outdoor conditions. Furthermore, this exercise shall play a seminal role in identifi-
cation of the key parameters governing the overall performance of such volumetric
solar collectors.
References
M. Abdelrahman, P. Fumeaux, P. Suter, Study of solid-gas-suspensions used for direct absorption of
concentrated solar radiation. Sol. Energy 22(1), 45–48 (1979)
N. Arai, Y. Itaya, M. Hasatani, Development of a “volume heat-trap” type solar collector using a
fine-particle semitransparent liquid suspension (FPSS) as a heat vehicle and heat storage
medium Unsteady, one-dimensional heat transfer in a horizontal FPSS layer heated by thermal
radiation. Sol. Energy 32(1), 49–56 (1984).
M. Ben Salah, F. Askri, K. Slimi, S. Ben Nasrallah, Numerical resolution of the radiative transfer
equation in a cylindrical enclosure with the finite-volume method. Int. J. Heat Mass Transf. 47
(10–11), 2501–2509 (2004)
V. Biju, T. Itoh, A. Anas, A. Sujith, M. Ishikawa, Semiconductor quantum dots and metal
nanoparticles: syntheses, optical properties, and biological applications. Anal. Bioanal. Chem.
391(7), 2469–2495 (2008)
C.F. Bohren, D. Huffman, Absorbing and Scattering of Light by Small Particles (Wiley, New York,
1983)
J. Buongiorno, D.C. Venerus, N. Prabhat, T. McKrell, J. Townsend, R. Christianson, . . . S.Q. Zhou,
A benchmark study on the thermal conductivity of nanofluids. J. Appl. Phys. 106(9), 094312
(2009)
W. Chang, B.A. Willingham, L.S. Slaughter, B.P. Khanal, L. Vigderman, E.R. Zubarev, S. Link,
Low absorption losses of strongly coupled surface plasmons in nanoparticle assemblies. PNAS
108(50), 19879–19884 (2011)
S.U.S. Choi, J. Eastman, Enhancing thermal conductivity of fluids with nanoparticles, in ASME
International Mechanical Engineering Congress and Exposition (ASME, San Francisco, 1995)
S.K. Das, S.U.S. Choi, W. Yu, T. Pradeep, Nanofluids: Science and Technology (Wiley, Hoboken,
2008)
B.T. Draine, P.J. Flatau, Discrete-dipole approximation for scattering calculations. J. Opt. Soc.
Am. A 11(4), 1491–1499 (1994)
W.D. Drotning, Optical properties of solar-absorbing oxide particles suspended in a molten salt heat
transfer fluid. Sol. Energy 20(4), 313–319 (1978)
V.E. Dudley, G. Msi, L.R. Evans, C.W. Matthews, Test Results, Industrial Solar Technology
Parabolic Trough Solar Collector (Sandia National Labs, Albuquerque, 1995)
S. Eustis, M.A. El-Sayed, Why gold nanoparticles are more precious than pretty gold: noble metal
surface plasmon resonance and its enhancement of the radiative and nonradiative properties of
nanocrystals of different shapes. Chem. Soc. Rev. 35(3), 209–217 (2006)
A. Ghadimi, R. Saidur, H.S.C. Metselaar, A review of nanofluid stability properties and character-
ization in stationary conditions. Int. J. Heat Mass Transf. 54(17–18), 4051–4068 (2011)
D. Han, Z. Meng, D. Wu, C. Zhang, H. Zhu, Thermal properties of carbon black aqueous nanofluids
for solar absorption. Nanoscale Res. Lett. 6, 457 (2011).
R.L. Hamilton, O.K. Crosser, Thermal conductivity of heterogeneous two-component systems. Ind.
Eng. Chem. Fundam. 1(3), 187–191 (1962)
K. Hanamura, Y. Kameya, Enhancement of solar radiation absorption using nanoparticle suspen-
sion. Sol. Energy 85(2), 299–307 (2011)
Y.L. Hewakuruppu, R.A. Taylor, H. Tyagi, V. Khullar, T. Otanicar, S. Coulombe, N. Hordy, Limits
of selectivity of direct volumetric solar absorption. Sol. Energy 114(2015), 206–216 (2015)
N. Hordy, D. Rabilloud, J.L. Meunier, S. Coulombe, High temperature and long-term stability of
carbon nanotube nanofluids for direct absorption solar thermal collectors. Sol. Energy 105,
82–90 (2014)
R. Kaluri, S. Vijayaraghavan, S. Ganapathisubbu, Model Development and performance studies of
a concentrating direct absorption solar collector. J. Sol. Energy Eng. 137(2), 021005 (2015)
M. Karami, M. Raisee, S. Delfani, M.A. Akhavan Bahabadi, A.M. Rashidi, Sunlight absorbing
potential of carbon nanoball water and ethylene glycol-based nanofluids. Opt. Spectrosc. 115(3),
400–405 (2013)
M. Karami, M.A. Akhavan Bahabadi, S. Delfani, A. Ghozatloo, A new application of carbon
nanotubes nanofluid as working fluid of low-temperature direct absorption solar collector. Sol.
Energy Mater. Sol. Cells 121, 114–118 (2014)
V. Khullar, Heat transfer analysis and optical characterization of nanoparticle dispersion based solar
thermal systems, Doctoral dissertation, Indian Institute of Technology Ropar, 2015
V. Khullar, H. Tyagi, P.E. Patrick, T.P. Otanicar, H. Singh, R.A. Taylor, Solar energy harvesting
using nanofluids-based concentrating solar collector. J. Nanotechnol. Eng. Med. 3(3), 031003
(2013)
V. Khullar, H. Tyagi, N. Hordy, T.P. Otanicar, Y. Hewakuruppu, P. Modi, R.A. Taylor, Harvesting
solar thermal energy through nanofluid-based volumetric absorption systems. Int. J. Heat Mass
Transf. 77, 377–384 (2014)
V. Khullar, H. Tyagi, T.P. Otanicar, Y.L. Hewakuruppu, R.A. Taylor, Solar selective volumetric
receivers for harnessing solar thermal energy, in Proceedings of the ASME 2016 International
Mechanical Engineering Congress & Exposition IMECE 2016, 11–17 Nov 2016, Phoenix,
Paper Number: IMECE2016–66599. 2016
Y.J. Kim, H. Ma, Q. Yu, Plasma nanocoated carbon nanotubes for heat transfer nanofluids.
Nanotechnology 21(29), 295703 (2010)
K. Kluczyk, W.A. Jacak, Size effect in plasmon resonance of metallic nanoparticles: RPA versus
COMSOL. in Proceedings the 44th International School and Conference on the Physics of
Semiconductors “Jaszowiec 2015”, vol 129, Acta Physica Polonica A No. 1-A. 2016
B. Lamas, B. Abreu, A. Fonseca, N. Martins, M. Oliveira, Assessing colloidal stability of long term
MWCNT based nanofluids. J. Colloid Interface Sci. 381(1), 17–23 (2012)
S.H. Lee, S.P. Jang, Extinction coefficient of aqueous nanofluids containing multi-walled carbon
nanotubes. Int. J. Heat Mass Transf. 67, 930–935 (2013)
B.J. Lee, K. Park, T. Walsh, L. Xu, Radiative heat transfer analysis in plasmonic nanofluids for
direct solar thermal absorption. J. Solar Energy Eng. 134(2.) 134, 021009–021001 (2012)
A. Lenert, Nanofluid-based receivers for high-temperature, high-flux direct solar collectors, Doc-
toral dissertation, Massachusetts Institute of Technology, 2010. Retrieved 12 Apr 2012, from
http://dspace.mit.edu/bitstream/handle/1721.1/61881/706146087.pdf
A. Lenert, E.N. Wang, Optimization of nanofluid volumetric receivers for solar thermal energy
conversion. Sol. Energy 86(1), 253–265 (2012)
J.C. Maxwell, A Treatise on Electricity and Magnetism (Clarendon Press, Oxford, 1881)
Z. Meng, D. Wu, L. Wang, H. Zhu, Q. Li, Carbon nanotube glycol nanofluids: Photo-thermal
properties, thermal conductivities and rheological behavior. Particuology 10(5), 614–618 (2012)
M.F. Modest, Radiative Heat Transfer (Academic, Amsterdam, 2003)
S.M.S. Murshed, C. Yang, K.C. Leong, Investigations of thermal conductivity and viscosity of
nanofluids. Int. J. Therm. Sci. 47(5), 560–568 (2008)
C.-W. Nan, R. Birringer, D.R. Clarke, H. Gleiter, Effective thermal conductivity of particulate
composites with interfacial thermal resistance. J. Appl. Phys. 81(10), 6692–6699 (1997)
O. Neumann, A.S. Urban, J. Day, S. Lal, P. Nordlander, N.J. Halas, Solar vapor generation enabled
by nanoparticles. ACS Nano 7(1), 42–49 (2013)
T.P. Otanicar, J.S. Golden, P.E. Phelan, Optical properties of liquids for direct absorption solar
thermal energy systems. Sol. Energy 83(7), 969–977 (2009)
T.P. Otanicar, P.E. Phelan, R.S. Prasher, G. Rosengarten, R.A. Taylor, Nanofluid-based direct
absorption solar collector. J. Renew. Sustain. Energy 2(3), 033102 (2010a)
T. Otanicar, I. Chowdhury, P.E. Phelan, R. Prasher, Parametric analysis of a coupled photovoltaic/
thermal concentrating solar collector for electricity generation. J. Appl. Phys. 108(11), 114907
(2010b)
T.P. Otanicar, I. Chowdhury, R. Prasher, P.E. Phelan, Band-gap tuned direct absorption for a hybrid
concentrating solar photovoltaic/thermal system. J. Solar Energy Eng. 133(4), 41014 (2011)
T. Otanicar, J. Hoyt, M. Fahar, X. Jiang, R.A. Taylor, Experimental and numerical study on the
optical properties and agglomeration of nanoparticle suspensions. J. Nanopart. Res. 15, 2039
(2013)
R. Prasher, P. Bhattacharya, P. Phelan, Thermal conductivity of nanoscale colloidal solutions
(nanofluids). Phys. Rev. Lett. 94(2), 025901 (2005)
I.T. Ritchie, B. Window, Applications of thin graded-index films to solar absorbers. Appl. Opt. 16
(5), 1438–1443 (1977)
E. Sani, L. Mercatelli, S. Barison, C. Pagura, F. Agresti, L. Colla, P. Sansoni, Potential of carbon
nanohorn-based suspensions for solar thermal collectors. Sol. Energy Mater. Sol. Cells 95(11),
2994–3000 (2011)
H. Tan, R. Santbergen, A.H.M. Smets, M. Zeman, Plasmonic light trapping in thin-film silicon solar
cells with improved self-assembled silver nanoparticles. Nano Lett. 12(8), 4070–4076 (2012)
R.A. Taylor, P.E. Phelan, T.P. Otanicar, R. Adrian, R. Prasher, Nanofluid optical property charac-
terization: towards efficient direct absorption solar collectors. Nanoscale Res. Lett. 6(1),
225 (2011)
R.A. Taylor, T. Otanicar, G. Rosengarten, Nanofluid-based optical filter optimization for PV/T
systems. Light: Sci. Appl. 1(10), e34 (2012)
R. Taylor, S. Coulombe, T. Otanicar, P. Phelan, A. Gunawan, W. Lv, . . . H. Tyagi, Small particles,
big impacts: a review of the diverse applications of nanofluids. J. Appl. Phys. 113(1), 011301
(2013a)
R.A. Taylor, T.P. Otanicar, Y. Herukerrupu, F. Bremond, G. Rosengarten, E.R. Hawkes, . . .
S. Coulombe, Feasibility of nanofluid-based optical filters. Appl. Opt. 52(7), 1413–1422
(2013b)
H. Tyagi, P. Phelan, R. Prasher, Predicted efficiency of a low-temperature nanofluid-based direct
absorption solar collector. J. Solar Energy Eng. 131(4), 041004 (2009)
A. Veeraragavan, A. Lenert, B. Yilbas, S. Al-Dini, E.N. Wang, Analytical model for the design of
volumetric solar flow receivers. Int. J. Heat Mass Transf. 55(4), 556–564 (2012)
B.-X. Wang, L.-P. Zhou, X.-F. Peng, A fractal model for predicting the effective thermal conduc-
tivity of liquid with suspension of nanoparticles. Int. J. Heat Mass Transf. 46(14), 2665–2672
(2003)
T. Yousefi, F. Veysi, E. Shojaeizadeh, S. Zinadini, An experimental investigation on the effect of
MWCNT-H2O nanofluid on the efficiency of flat-plate solar collectors. Renew. Energy 39,
207–212 (2012a)
T. Yousefi, F. Veysi, E. Shojaeizadeh, S. Zinadini, An experimental investigation on the effect of
Al2O3-H2O nanofluid on the efficiency of flat-plate solar collectors. Renew. Energy 39(1),
293–298 (2012b)
X. Zhang, H. Gu, M. Fujii, Effective thermal conductivity and thermal diffusivity of nanofluids
containing spherical and cylindrical nanoparticles. J. Appl. Phys. 100(4), 044325 (2006)
H. Zhang, H.-J. Chen, X. Du, D. Wen, Photothermal conversion characteristics of gold nanoparticle
dispersions. Sol. Energy 100, 141–147 (2014)
Q. Zhu, Y. Cui, L. Mu, L. Tang, Characterization of thermal radiative properties of nanofluids for
Selective absorption of solar radiation. Int. J. Thermophys. 34(12), 2307–2321 (2012)
Review of Subcooled Boiling Flow Models
Eckhard Krepper and Wei Ding
Abstract
In this chapter, the present capabilities of CFD modelling for wall boiling in
industrial applications are described. The basis is the Eulerian two-fluid frame-
work of interpenetrating continua. From the first attempts, heat flux partitioning
algorithms were used to describe boiling at a heated wall. Based on a mechanistic
model representation of the microscopic processes, the framework is described by
empirical correlations. The developments of the main correlations for the bubble
size at detachment and for the nucleation site density are described. Different
approaches for the bubble size in the bulk are presented. Further the extension of
the conventional heat partitioning model toward the high heat flux will be also
stated. Finally an outlook on further model improvement is given.
Keywords
Boiling • Bubble detachment frequency • Bubble diameter at detachment • CFD •
Euler-Eulerian approach • Nucleation site density • Wall boiling
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Critical Heat Flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Wall Boiling in the Framework of Euler/Euler Two-Phase Flow CFD Modelling . . . . . . . . . . . . . 4
Modelling of Boiling at a Heated Wall by Heat Flux Partitioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
General Model Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
E. Krepper (*)
Institute of Fluid-Dynamics, Helmholtz-Zentrum Dresden – Rossendorf, Dresden, Germany
Computational Fluid Dynamics, Helmholtz-Zentrum Dresden-Rossendorf, Dresden, Germany
e-mail: E.Krepper@hzdr.de
W. Ding
Computational Fluid Dynamics, Helmholtz-Zentrum Dresden-Rossendorf, Dresden, Germany
e-mail: W.Ding@hzdr.de

DOI 10.1007/978-981-4585-86-6_20-1
2 E. Krepper and W. Ding
Bubble Detachment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Nucleation Site Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Other Influence Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Evaporation and Condensation in the Bulk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Bubble Size in the Bulk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Model Extensions Toward Higher Heat Fluxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Outlook of Further Model Improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Individual Subprocess Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Introduction
Subcooled flow boiling becomes an essential phenomenon in many industrial

applications when large heat transfer coefficients are required. For lots of flow
conditions, heat transfer from a wall into a fluid can be much more effective than
for single-phase convection. The design or the safety assessment of such applica-
tions can be supported by simulation of these processes. The computational fluid
dynamic (CFD) modelling describes the phenomena dependent only on local quan-
tities. Therefore, it is suited especially for design and optimization of the flow
geometry.
For engineering calculations, currently the most widely used CFD approach to
model two-phase flows with significant volume fractions of both phases is the
Eulerian two-fluid framework of interpenetrating continua (see, e.g., Ishii 1975;
Drew and Passman 1998; Yeoh and Tu 2010). In this approach, balance equations
for mass, momentum, and energy are written for each phase, i.e., gas and liquid,
separately and weighted by the so-called volume fraction which represents the
ensemble averaged probability of occurrence for each phase at a certain point in
time and space. Microscopic phenomena are averaged and considered by empirical
correlations. Exchange terms between the phases appear as source/sink terms in the
balance equations. These exchange terms express the interfacial forces, as well as
heat and mass fluxes, as functions of the average flow parameters. Since most of
these correlations are highly problem specific, their range of validity has to be
carefully considered.
Up to medium values of the gas volume fraction, the two-phase flow is in the
bubbly flow regime. In this instance, the mass and momentum exchange between the
phases is conveniently parametrized by the bubble size. To account for deformations
the equivalent spherical diameter is used as a measure of bubble size. At high
Reynolds numbers common to engineering flows, the bubbles strongly affect the
turbulence intensity and structure in the bulk and near walls. These questions have
been investigated extensively for adiabatic flows (see, e.g., Lucas et al. 2007), but
not all issues have been resolved satisfactorily yet.
Review of Subcooled Boiling Flow Models 3
For the case of boiling flows, where heat is transferred into the fluid from a heated
wall at such high rates that vapour is generated, additional source terms describing
the physics of these processes at the heated wall have to be included. In the past,
several CFD wall boiling model approaches following the lines of Kurul and
Podowski (1990, 1991) were calibrated and validated by several authors against
different experimental results. In this approach according to a microscopic model
concept, a given overall heat flux is divided into various components.
An application of continuing high interest is the thermal hydraulic flow in the core
of a nuclear reactor. Accordingly in most of these tests, subcooled flow boiling of
water at high pressure flowing upward in a vertical pipe heated from the outside was
investigated, and measurements of the axial development of gas volume fraction,
wall temperature, and cross-sectionally averaged liquid temperature were provided.
However, typical flow conditions encountered in this application do not particu-
larly lend themselves to experimental investigation. High pressure, high tempera-
ture, narrow channels, and small expected sizes of steam bubbles represent
significant challenges for measurements. The use of refrigerants can greatly relieve
this burden. Advantages are that this allows a choice of test parameters that is more
convenient for the measurement compared to the water/steam system at high pres-
sure. The same vapour/liquid density ratio can be achieved at a much lower system
pressure, and the same Reynolds number can be achieved at a larger diameter of the
heated pipe. This enables a measurement of radial profiles for gas volume fraction,
temperature, liquid, and gas velocities and of bubble sizes which allows a stringent
validation of CFD models.
Critical Heat Flux
For lots of flow conditions, heat transfer from a wall into a fluid can be much more
effective than for single-phase convection. However, the efficient heat transfer
mechanism provided by vapour generation is limited at a point where liquid is
expelled from the surface over a significant area. This occurs at the critical heat
flux where the heat transfer coefficient begins to decrease with increasing tempera-
ture leading to an unstable situation. In this event, a rapid heater temperature
excursion occurs which potentially leads to heater melting and destruction. For a
given working fluid, the critical heat flux depends on the flow parameters as well as
the geometry of the flow domain. The verification of design improvements and their
influence on the critical heat flux requires expensive experiments. Therefore, the
supplementation or even the replacement of experiments by numerical analyses is of
high interest in industrial applications.
In the past, many different empirical correlations for critical heat flux were
developed and fitted to data obtained from experimental tests. These have been
implemented mainly in purpose-specific 1D codes and applied for engineering
design calculations. However, these correlations are valid only in the limited region
of fluid properties, working conditions, and geometry corresponding to the tests to
which they were fitted. Using large lookup tables based on a great number of
experiments, a significant range of fluid properties and working conditions can be

covered. But this method is still limited to only that specific geometry for which they
were developed. Independence of the geometry can only be achieved by the appli-
cation of computational fluid dynamic (CFD) methods. Existing CFD models,
however, are not yet able to describe critical heat flux reliably. A precondition
would be the complete understanding and simulation of boiling as a preliminary
state toward critical heat flux.
Wall Boiling in the Framework of Euler/Euler Two-Phase Flow CFD

Modelling
The general equations for diabatic two-phase flow in the Euler/Euler framework of
interpenetrating continua have been reviewed in many places before (e.g., Ishii
1975; Drew and Passman 1998; Yeoh and Tu 2009 and the corresponding chapter
of the actual handbook).
Therefore, here the focus is directed only on those issues that are particularly
relevant for the simulation of boiling. The first major block is the wall boiling model
describing vapour generation at the wall and transfer of sensible heat to the liquid.
Here, most of the CFD model implementations closely follow the heat flux
partitioning approach.
A second issue are the phase change phenomena occurring in the bulk fluid. Here,
the relative amount of liquid and vapour is allowed to change by condensation/
evaporation depending on the net transport of heat to the liquid-vapour interface. To
simplify matters, the vapour bubbles are assumed to be at saturation temperature
everywhere which is a rather good approximation except close to the critical heat
flux. In the first model approaches, a monodispersed size distribution was assumed
with a size dependent on the liquid temperature. More developed models consider
the bubble size distribution by momentum methods respective by population balance
models.
Turbulent fluctuations are modelled by a shear stress turbulence (SST) model
according to Menter (1994, 2009) applied to the liquid phase. This corresponds to a
k-ω model near the walls and a k-ε model far from walls. The enhancement of liquid
turbulence caused by the bubbles is considered following Sato et al. (1981) by an
additive contribution to the effective viscosity. In addition, a wall function for
boiling flows suggested by Ramstorfer et al. (2005) and subsequently verified by
Koncar and Krepper (2008) and Koncar and Mavkov (2010) was tested.
For momentum exchange between the phases, finally, lift and turbulent dispersion
wall forces are included in the model in addition to the ubiquitous drag force. The
consideration of these forces has been found in good agreement with data for
adiabatic air water bubbly flow (e.g., Lucas et al. 2007).
Modelling of Boiling at a Heated Wall by Heat Flux Partitioning
General Model Structure
Modelling of wall boiling is based on the microscopic phenomena observed in the

near wall region and transformed either by direct simulation or by empirical corre-
lations to the Euler/Euler framework.
Real wall surfaces have irregularities such as pits or scratches. The initial
nucleation of a vapour bubble typical occurs at such imperfections of the surface.
The bubble grows during a time tg and finally leaves the wall. During this time, heat
QE is consumed by the vapour generation. After the release of the initial bubble,
cooler bulk liquid comes in contact to the surface where the bubble was previously
attached. This new liquid adjacent to the wall leads to cooling QQ. This mechanism,
which is obviously not present in single-phase flows, is termed quenching. After a
waiting time tw, the thermal layer near the wall is reformed, and a new bubble occurs
at the same place. This so-called ebullition cycle extends over a time τ = tg + tw .
Consequently, the departure frequency f is calculated as f = 1/τ. On parts of the wall,
where no bubbles reside, heat QC flows directly to the subcooled liquid in the same
way as in single-phase flow. The complete cycle is presented in Fig. 1.
Judd and Hwang (1976) first proposed the heat flux partitioning concept which
was further developed by Kurul and Podowski (1990, 1991). Detailed informations
on heat flux partitioning are also given by Yeoh et al. (2008). A given overall het flux
Qtot is divided into various components according to a microscopic model concept
(see Fig. 1).
Accordingly, the given external heat flux Qtot, applied to the heated wall, is
written as a sum of three parts:
Qtot ¼ QC þ QQ þ QE (1)
Fig. 1 Microscopic concept

of the heat flux partitioning
model
Fig. 2 Principal algorithm of TL and U from flow solution

the heat flux partition model
approach guess for TW (nested intervals)
calculation of
π 3
evaporation heat flux QE = d BW fNh1g
6
convective heat flux QC = ( 1 − AW )hC (TW − TL )
quenching heat flux QQ = AW hQ (TW − TL )
overall heat balance already fulfilled?
No ?
QW = QE + QF + QQ
Yes
where QC, QQ, and QE denote the heat flux components due to single-phase turbulent
convection, quenching, and evaporation, respectively. The individual components in
this heat flux partitioning are then modelled as functions of the wall temperature and
other local flow parameters. Once this is accomplished, Eq. 1 can be solved
iteratively for the local wall temperature TW, which satisfies the wall heat flux
balance (see Fig. 2). Denoting the fraction of area influenced by the bubbles as AW,
the heat flux components are expressed as discussed in the following.
The turbulent convection heat flux is calculated in the CFX model version in
much the same way as for a pure liquid flow without boiling, but multiplied by the
fraction of area unaffected by the bubbles, i.e.,
QC ¼ AconvL hCL ðT W T L Þ (2)
Here AconvL = 1 AW, with AW as the area fraction occupied by bubbles. In

some implementations, TL is taken from the first grid cell adjacent the heated wall.
To avoid a grid size dependency, Wintterle (2004) proposed to consider here the
turbulent wall function. Here hCL is the heat transfer coefficient which is written
using the temperature wall function T+(y+) known from Kader (1981) as
ρ C P uτ
hCL ¼ (3)
Tþ
where nondimensional variables (indicated by superscript “+”) and the friction
velocity uτ are defined as usual. QQ is represented in terms of the quenching heat
transfer coefficient hQ:
QQ ¼ AW hQ ðT W T L Þ (4)
A grid independent solution for QQ is obtained by evaluating the nondimensional

temperature profile at a fixed value of y+. The evaporation heat flux QE is obtained
via the evaporation mass flux at the wall:
QE ¼ m_ W H LG (5)
where the generated vapour mass m_ W is expressed in terms of the bubble diameter at
detachment dW, bubble generation frequency f, and nucleation site density N as
π
m_ W ¼ ρG d3W f N (6)
6
Correlations for the yet undetermined quantities used in the heat flux partitioning
wall boiling model are discussed in the following.
Bubble Detachment
The bubble size at detachment depends on the liquid subcooling. Also the liquid
properties, the system pressure, wall material and surface properties, and the heat
flux have an influence. Finally the mechanical attraction of the surrounding flow, as
indicated by the fluid velocity, determines the detachment of a growing bubble. For
CFD modelling of boiling, different correlations for bubble detachment were devel-
oped derived from measurements or analytical considerations.
Tolubinsky and Kostanchuk

An investigation of the bubble size at detachment was performed by Tolubinsky and
Kostanchuk (1970) for water at different pressures and subcoolings. The observed
dependence on the liquid subcooling at atmospheric pressure can be fitted to a
correlation
T T L
ΔT
sat
d W ¼ dref e refd (7)
To match the tests of Bartholomej and Chanturiya (1967) which were conducted
at much higher pressures relevant under typical nuclear energy applications, the
values of dref and DTrefdwere adjusted to dref = 0.6 mmand ΔTrefd = 45K(e.g.,
Krepper et al. 2007).
Fritz
An attempt to determine the bubble diameter at detachment at atmospheric pressure
was performed by Fritz (1935). He integrated the surface of the arising bubble and
analyzed a force balance.

σ 0:5
dw ¼ CΘ (8)
gΔρ
Here Θ is the wall contact angle and was estimated for water/steel to π/4. σ is the
surface tension and Δρ the density difference between liquid and gas.
Kocamustafaogullari
To extend the range of validity, the correlation of Eq. 8 was extended by Kocamusta-
faogullari (1983):
0:5 !0:9
5 σ Δρ
dw ¼ 2:64 10 (9)
gΔρ ρg
The correlation was supported by experiments with water in the pressure range of
0.01–14 MPa. For atmospheric pressure, Eq. 10 reduces to Eq. 9.
Unal
A very detailed correlation for the bubble diameter at detachment was developed by
Unal (1976). He tried to consider both the thermal wall material properties of the
heater and the fluid field of the surrounding flow:
2:42 105 P0:709 a

dw ¼
ðr Þ0:5 ffi
bΦffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ΔT sup ρS CPS λS
a¼
2ρg H lv π
ΔT sub (10)
b¼ !
ρl
2 1
ρ
g
v
Φ ¼ max , 1½m=s
v0
Bubble Detachment Based on Force Balance

Klausner et al. (1993) proposed a mechanistic force balance model for the prediction
of both bubble departure and lift-off sizes in the nucleate boiling condition. They
applied the model successfully in the various flow conditions in both horizontal and
vertical channels under pool and flow boiling. Later, many investigators tried to
improve the model to achieve a general applicability for flow direction and fluids.
Zeng et al. (1993a, b) applied the model for both horizontal and vertical channels
under pool and flow boiling, whereas Situ et al. (2005) and Yeoh and Tu (2005)
extended its application to steam-water boiling flow condition.
The concept of Klausner et al. (1993) is based on a force balance at the arising
bubble (see Fig. 3). Based on the conservation of momentum, in the direction
parallel (tangential t – direction) and perpendicular (vertical n- direction) to the
heating wall, the forces acting on the bubble can be obtained as
Fig. 3 Schematic diagram of

a growing bubble at a heated
surface
Fn ¼ Fgroth, n þ Fdrag, n þ Fcpress, n þ Fsl, n þ Fb, n þ Fsurf , n (11)
Ft ¼ Fgrowth, t þ Fdrag, t þ Fb, t þ Fsurf , t (12)
Fgrowth is the bubble growth force, Fdrag the quasi-steady drag force due to the
viscous fluid flowing around the bubble, Fcpress a contact pressure due to the effect of
the wall, Fsl a force resulted from the flow distribution, Fb the buoyancy force, and
Fsurf the surface tension.
When the bubble is fixed at the heating surface and growing out of an nucleation
site, the sum of all forces has to be zero:Fn = Ft = 0. In horizontal pool boiling, the
detachment or lift off occurs when the Fn becomes positive. In flow boiling or
non-horizontal pool boiling, detachment occurs when Fn or Ft becomes larger than
0. If Ft becomes greater than 0, firstly the bubble departs from the nucleation site and
starts sliding along the wall until it lift off when the force in perpendicular direction
becomes positive. Conversely, the bubble directly lifts off from the nucleation site
without sliding when Fn becomes positive firstly. The phenomenon of sliding
bubbles was observed also in experiments (e.g., Prodanovic et al. 2002).
This model concept was further developed by different authors formulating more
precise the single components of the forces in Eqs. 11 and 12 (e.g., Chen et al. 2012).
Nucleation Site Density
If the nucleation temperature of the liquid is reached, a vapor bubble is formed. The
nucleation temperature slightly exceeds the saturation temperature of the fluid. The
initial nucleation of vapor typically occurs at a cavity or crevice on the heated

surface.
Lemmert and Chawla

Lemmert and Chawla (1977) proposed a quite simple correlation for the calculation
of the nucleation site density dependent on the wall superheating temperature:
p
TW TL
N ¼ N ref (13)
ΔT refN
More Detailed Models

Kocamustafaogullari and Ishii (1983) developed a nucleation site density correlation
based on measurements. They considered both surface conditions and fluid
properties.
2:157 107 ρ3:2 ð1 þ 0:0049ρ Þ4:13 RC 4:4 Δρ

N¼ ,ρ ¼ (14)
dw 2 ρg
with the critical cavity size RC. Additional in this analysis, they derived a correla-
tion for the bubble size at detachment (see section “Kocamustafaogullari”, Eq. 9).
Yang and Kim (1988) considered the probability distribution of the cavity size.
They investigated the activation capability of a cone dependent on the cone angle.
Wang and Dhir (1993) proposed an empirical correlation at pool boiling based on
measurements including the effect of contact angle. During the tests, they influenced
the wettability of the heated surface and measured the effect. For pool boiling, they
derived
N ¼ 7:81 1029 ð1 cos θÞRC 6:0 (15)
with RC as the critical cavity radius.

Benjamin and Balakrishnan (1997) investigated the effect of the heat flux on the
nucleation site density. They considered different heater materials (stainless steel and
aluminum) with different surface treatment by polishing. Furthermore, different
fluids were investigated. Their correlation for pool boiling is

1 0:4
N ¼ 218:8Pr 1:63
Ω ΔT W 3 (16)
γ
with the Prandtl number Pr, a surface-liquid interaction parameter γ, and a dimen-
sionless surface roughness parameter Ω.
Also Basu et al. (2002) analyzed experiments and tried to consider the contact
angle. During the tests, they controlled the surface wettability following a well-
defined surface treatment.
Hibiki and Ishii (2003) performed a very detailed investigation of the different
dependencies of the nucleation site density. They tried to collect the experiences of
the previous described correlations. The dependency on surface conditions and fluid
properties was expressed as dependencies on pressure and on the minimum cavity size.
0

θ2 þ λ
N ¼ N ref 1 exp 2 exp f ðρ Þ 1 (17)
8μ Rc
with the parameters m = 0.722, L0 = 2.5.106 m, the contact angle Θ, the critical
cavity radius Rc
n o
2σ 1 þ ρg =ρf =Pf
Rc ¼ (18)
exp ifg T g T sat = RTg T sat 1
and the function
f ðρþ Þ ¼ 0:01064 þ 0:48246ρþ 0:22712ρþ2 þ 0:05468ρþ3 , ρþ ¼ logðρ Þ

(19)
Later, it was proposed to replace the function f(ρ+) by measurements. Kandlikar
and Steinke (2002) performed corresponding experiments.
General Considerations on Nucleation Site Density

Most of the theoretical and experimental investigations show a clear dependency of
the nucleation site density on the wall superheating temperature. This is expressed in
the correlation of Lemmert and Chawla (1977); see Eq. 7. However, a recent
compilation (Kolev 2006) shows that vastly different parameter values are required
to match different data sets. A likely reason for this fact is that nucleation site density
is highly dependent on the microscale topography of the boiling surface, which in
turn depends strongly on the processes that were used to finish the surface. These
processes are very diverse, in most boiling experiments not specifically controlled,
and in most cases an unknown boundary condition for the simulation. This is the
reason for the doubts of the authors, to find any reasonable predictive model for
nucleation site density.
Krepper et al. (2013) proposed the following solution strategy: Dealing with the
model framework shows that the nucleation site density has almost no influence on
the liquid temperature and a small influence on the gas volume fraction, but a strong
influence on the wall superheatTW Tsat. In the most common case of missing
information on the nucleation site density, this can be compensated by the measured
wall temperature which is in most boiling experiments easily accessible. It might be
reasonable to apply a quite simple relation between nucleation site density and wall
temperature as in Eq. 13 and determine Nref in a way that the calculated wall
temperature agrees to the measured value.
Other Influence Factors
Bubble Influence Area/Factor

In terms of bubble detachment diameter dW and nucleation site density N, the wall
area fraction AW, influenced by vapor bubbles, is given by
2
dW
AW ¼ π a N (20)
2
Here, a is the so-called bubble influence factor, for which a value of 2 is

commonly used (Kurul and Podowski 1990, 1991). Direct experimental evidence
concerning this quantity is rather scarce. Probably the most relevant source is Han
and Griffith (1965) who in some “rough” experiments determined the hydrodynamic
disturbance caused by lifting a spherical particle from a horizontal surface and found
that it has a range of twice the size of the particle. A similar size has been claimed by
Cieslinski (2005) from PIV measurement of the flow field around departing bubbles,
although the quality of the images presented is rather poor.
AW = 1 corresponds to the case where the whole surface is under the influence of
bubbles. The parameters dW, a, and N should be such that the calculated therefrom
AW by Eq. 20 results in smaller values. Moreover, it should be kept in mind that
already as AW approaches 1, the assumptions of the model are not really satisfied
anymore.
Bubble Detachment Frequency

The bubble detachment frequency f is given according to Cole (1960) as a function
of the detachment size dW.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
4gðρL ρG Þ
f ¼ (21)
3CD dW ρL
Relations of this type have been reviewed critically by Ivey (1967) and Ceumern-
Lindenstjerna (1977).
Quenching Heat Transfer Coefficient

The quenching heat transfer coefficient is calculated using the analytical solution for
one-dimensional transient conduction, as suggested by Mikic and Rohsenow (1969);
Mikic et al. (1970):
2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
hQ ¼ pffiffiffi f twait kL ρL CPL (22)
π
where twait is waiting time between the bubble departure and the appearance of a next
bubble at the same nucleation site. A simple assumption by Kurul and Podowski
(1990, 1991) is adopted also here, where the waiting time takes a fixed fraction of the
bubble departure period:
0:8
twait ¼ (23)
f
Data supporting this simple estimate come from the work of del Valle and
Kenning (1985) which however is limited to the regime where the heat flux is
75% of the critical heat flux and larger.
Turbulent Wall Functions for Boiling Flow

The first wall boiling model originally was developed for the application in a 1D
system code. Then the liquid temperature can be taken from the cell center. In the
first CFD implementations of the model, the temperature was taken from the first cell
adjacent to the heated wall, resulting in a grid dependency.
Ramstorfer et al. (2005) proposed and subsequently Koncar and Krepper (2008)
and Koncar and Mavkov (2010) verified a turbulent wall function for boiling flow
where bubbles grow on the heated wall by analogy with flow over a rough wall. The
presence of the bubbles forces the liquid into a similar flow pattern as that observed
in single-phase turbulent flow with wall roughness. The latter is described by a
modified law of the wall (White 1991; Pope 2000).
u 1 y 1 y
¼ ln þ B ¼ ln þ B (24)
uτ κ δ κ s
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where uτ ¼ ðτW =ρÞis the friction velocity, δ = μ/(ρuτ) is the viscous length scale,
s is the hydrodynamic roughness length scale, and k 0.41 is the von Karman
constant. For flow over smooth walls, B const 5.5, while for flow over rough
walls, B* is a function of s/δ which in the limit of large s, i.e., for the so-called fully
rough walls, tends to a constant value of 8.5. An essentially new concept used here is
to directly relate the hydrodynamic roughness s directly to the bubble size and
nucleation site density as
s / Nd 3W (25)
The constant of proportionality is not known from theoretical considerations at

present, so its value is determined by matching the data. As will be shown by the
comparison in section 7.1, the representation of radial velocity profiles is greatly
enhanced by employing this two-phase boiling wall function over the often used
single-phase wall function. Moreover, a single value for the constant applies to a
range of flow situations.
Evaporation and Condensation in the Bulk
Vapor is assumed to be at saturation condition. Where the liquid is subcooled, i.e.,

TL < Tsatvapor is condensing with the mass transfer rate per unit volume:

hLG ðT sat T L ÞAI
m_ ¼ max ,0 (26)
H LG
while in superheated liquid, fluid is evaporating at the rate

hLG ðT L T sat ÞAI
m_ ¼ max ,0 (27)
H LG
Here, hLG is the interfacial heat transfer coefficient, calculated according to Ranz
and Marshall (1952) as
kL kL
hLG ¼ Nu ¼ 2 þ 0:6Re1=2 Pr1=3 (28)
dB dB
and AI is the interfacial area, which is expressed in terms of gas volume fraction and
equivalent spherical bubble diameter as
6
AI ¼ αG (29)
dB
Bubble Size in the Bulk
RPI Approach
To close the phase transition model in the bulk bubble flow with a mean bubble
diameter dB, Kurul and Podowski (1990, 1991) and also Anglart et al. (1997)
proposed to calculate the bubble diameter dB locally as a linear function of liquid
subcooling Tsub:
dB, 1 T sub T sub, 2 þ dB, 2 T sub T sub, 1

dB ¼ (30)
T sub, 2 T sub, 1
Reference subcooling conditions for typical nuclear energy applications have

been given as dB,1 = 0.1 mm at Tsub , 1 = 13.5 Kand dB,2 = 2 mm at Tsub,2 = 5 K
in Anglart et al. (1997).
Near a heated wall with increasing wall distance, the subcooling temperature
decreases. Calculating dB according Eq. 30 also dB would decrease with increasing
wall distance. In boiling tests however in many cases, an increase in dB with increase
in wall distance is observed, which might be caused by bubble coalescence. This

effect cannot be described by Eq. 30.
Zeitoun and Shoukri

Zeitoun and Shoukri (1996) performed experiments in a vertical channel under low
pressure and mass flux conditions. High-speed video techniques combined with
digital image processing were applied to measure the bubble sizes of the generated
vapor. For the bubble size in the bulk db, the following correlation was derived:
1:326
db 0:0683 ρl =ρg
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ 0 1 (31)
σ=gΔρ 149:2 ρl =ρg
1:326
B C
Re0:324 @Ja þ A
Bo0:487 Re1:6
iMUSIG Approach
A more detailed calculation of the interfacial area density is possible by an additional
transport equation, by a momentum method, or most comprehensively by a bubble
class approach. Here the latter method is described short for the application to
boiling processes. A detailed description can be found on another place of the
current handbook.
To describe polydispersed flows within a purely Eulerian approach, a number of
different (MUltiple) bubble size groups i = 1 . . . M are considered, each representing
bubbles of typical size di. The fraction of gas volume contained in each bubble size
group is denoted as αi so that the total gas volume fraction is given by
X
M
αG ¼ αi (32)
i¼1
From a theoretical point of view, there would be no problem to define each size
group i as additional phase with the complete setup of Navier-Stokes equations.
Practical reasons of computational effort and numerical stability are limiting factors
of such a procedure. Instead, the different size groups are considered only in the
continuous equation. Between the size groups, the exchange both by bubble coales-
cence and fragmentation and by condensation and evaporation is organized. The
advantage is that a large number of bubble size groups can be considered while
keeping the computational effort within reasonable bounds. On the other hand,
profound effects of bubble size are missed entirely like, for example, the change in
the sign of the lift force as discussed in other places of this handbook. To capture
such phenomena, provision has to be made that bubbles of different size may move
with different velocities. To overcome these limitations, the inhomogeneous MUSIG
model (Frank et al. 2008; Krepper et al. 2008) was developed and applied to boiling
Inhomgeneous MUSIG model
Velocity groups V1 V2 ... VN

J=1..N
conden- evapo-
sation ration
Size fractions
K=1..∑MJ d1 dM1 dM1+1 dM1+M2 d∑MJ
bubble break-up bubble coalescence
Fig. 4 Principle schema of the inhomogeneous MUSIG model including phase transfer
phenomena in Krepper et al. (2013) and Krepper and Rzehak (2014). The schema is
shown in Fig. 4.
It is then useful to define occupation numbersfi = ai/aG giving the contribution of
each size group to the total gas volume fraction. Obviously we then have Σi fi = 1.
For each size group, the equation of mass conservation assumes the form
@
ðαi ρG Þ þ ∇ðαi ρG uG Þ ¼ Γ topo þ Γ phase (33)
@t i i
where the right-hand side gives the net source of mass for group i which results from
topological changes due to coalescence and breakup as well as phase change due to
condensation and evaporation.
For the homogeneous MUSIG model, only one momentum and energy equation
for the total amount of vapor is considered as well as the conservation equations of
the liquid of course. In these equations, the total gas volume fraction αG is calculated
according to Eq. 32. In addition, also the bubble size dB appears which is taken in the
Sauter sense representing the interfacial area AI = 6αG/dB. In order to preserve this
interpretation, dB is calculated from the occupation number and bubble size for each
group as
!1
XM
fi
dB ¼ (34)
d
i¼1 i
Bubble Coalescence and Breakup

The net mass source for size group i due to bubble coalescence and breakup can be
expressed as the sum of bubble birth rates due to the breakup of larger bubbles from
groups j > i to group i and coalescence of smaller bubbles from size groups j < i to
group i as well as bubble death rates due to breakup of bubbles from size group i to
smaller bubbles in groups j < i and the coalescence of bubbles from size group i with
bubbles from any other group to even larger ones which belong to groups j > i. That is,
Γtopo
i ¼ Bbreak
i Dbreak
i þ Bcoal
i Dcoal
i (35)
The birth and death rates in turn are commonly expressed in terms of the
coalescence and breakup kernels such that
X mi
Bi break ¼ ρG αG b mj , mi f j Δmi
j>i
mj
1 X
Di break
¼ ρG α G b mi , mj f i Δmj
2 j<i
1 XX mj þ mk (36)
Bi coal ¼ ρ2G α2G c mj , mk Xjki f j f k
2 ji ki
mj mk
X 1
Di coal ¼ ρ2G α2G c mi , mj f i f j
j
mj
Here Xjki is an approximation to the delta-functiond(mj + mk mi). Unfortu-

nately according to the actual state of development, the breakup and coalescence
kernel functions b and c adapted to yield good results for air/water at adiabatic
conditions cannot directly be used for other fluid systems as steam/water of refrig-
erants. The parameters have to be newly adjusted to match the measured bubble
sizes.
Condensation and Evaporation Including Boiling at the Wall

When condensation or evaporation occurs, the volume fraction in size group
i changes for two reasons: (i) mass is transferred directly between the bubbles and
the liquid and (ii) since due to this direct mass transfer the bubbles are shrinking or
growing, they may subsequently belong to a different size group.
Written as a source term for size group i, the direct mass transfer to the liquid is
given by
e i ¼ AI, i hL, i ðT L T sat Þ

Γ (37)
H LG
for the assumption that the gas is at saturation temperature. The total source terms for
size class i including also the ensuing change of bubble size, i.e., Γ phase
i in Eq. 37,
have been derived by Lucas et al. (2011) as

mi ei mi
Γphase ¼ Γ e iþ1 for Γ
Γ e i < 0, i:e:
i mi mi1 miþ1 mi
mi e i1 mi e i for Γ
e i > 0, i:e: evaporation
condensation Γ Γ
mi mi1 miþ1 mi
(38)
where mi = r pd3/6 is the mass of each bubble in size group i. Basing the calculation
on bubble mass rather than size for compressible flows has the advantage that since
mass is conserved, no extra terms arise in the equations. Conversion to the

corresponding bubble size which depends on the local density can be done straight
forwardly as needed. For incompressible flows, no differences between mass- and
size-based groups arise.
The liquid side heat transfer coefficient finally is calculated according to Ranz and
Marshall (1952) as in Eq. 28.
In addition to the source terms for the continuity equations for the bubble size
groups, there is also a mass source for the liquid phase continuity equation which is
given by
X
ΓL ¼ ei
Γ (39)
i
Moreover, corresponding secondary sources appear in the momentum and energy

equations.
To include the generation of vapor bubbles at the heated wall, an additional
source term, Srpi, is included in Eq. 40 (Lifante et al. 2011). This source term applies
only to the equation corresponding to the size group whose diameter is the closest to
the bubble detachment diameter dW. It is given by the evaporation mass flux
computed in the wall heat partitioning distributed evenly throughout the grid cells
adjacent to the heated wall, i.e.,
S
Srpi ¼ m_ W (40)
V
where m_ W is given by Eq. 6 and S and V are wall surface area and volume of the
corresponding grid cell, respectively.
Model Extensions Toward Higher Heat Fluxes
Newer implementations of the heat flux partitioning model (e.g., Lifante 2013;
Mimouni et al. 2016) include model developments toward the simulation of higher
heat fluxes. The original heat flux partitioning model considers steam only at
saturation conditions. For lower vapor void fractions the energy equation has to be
solved only for the liquid phase. Gas will occur only at saturated conditions. At
higher heat fluxes, larger vapor void fractions will occur and this simplification has
to be extended by solving also the energy equation for gas and superheated steam can
be simulated. The heat flux partitioning algorithm has to be extended by an addi-
tional component, the convective heat flux into gas QG. Equation 1 reads now as
Qtot ¼ QC þ QQ þ QE þ QG (41)
The additional degree of freedom has to be closed by a function describing the

relation of heat transferred to liquid and to gas. For the CFX implementation, f(a)is
proposed (Lifante et al. 2013):
8 9
>
>
1 ½20ðααcrit Þ >
< 1 e α α crit >
=
2 20αcrit
f ðαÞ ¼ 1 α (42)
>
> >
: α αcrit > ;
2 αcrit
Equation 2 reads now as
QC ¼ ð1 AW Þf ðαL ÞhCL ðT W T L Þ
and similarly
QG ¼ ð1 AW Þf ðαG ÞhCG ðT W T G Þ (43)
In the same way as Eq. 3,
ρG CPG uτ
hCG ¼ (44)
Tþ
G
Figure 5 shows the proposed distribution functions for aLcrit = 0.2 respective
αGcrit = 0.8. With good adjusted values, a typical behavior for film boiling can be
described (Lifante et al. 2013). The weakness of the above referred implementation
is the arbitrariness of the applied distribution function independent on the micro-
scopic phenomena. These can be achieved by a detailed consideration of the
microscopic processes in the near bubble region.
Outlook of Further Model Improvements
Overview
As is introduced, the Euler/Euler two-phase flow CFD-Model has been already well
developed to model the global boiling process. Nevertheless, this approach is far
away from a predictive tool due to the correlations described by bubble dynamics. A
critical review of the detailed correlations shows that some of the parameters are not
suited for a broad usage for different fluids or different pressure levels but have to be
carefully recalibrated for the intended applications.
In order to improve the CFD model, it is necessary to investigate the individual
subprocess models that are able to present the bubble dynamics on the wall correctly.
The difficulties here are due to the small scales of bubbles which have a
non-negligible impact on the heat transfer; the high dynamics of boiling; the
1.0 1.0
0.8 0.8
F(αG) [-]
0.6 0.6
F(αL) [-]
0.4 0.4
0.2 0.2
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
αL [-] αG [-]
Fig. 5 Distribution functions F(αL)and F(αG) for αLcrit = 0.2 resp. αGcrit = 0.8
complexity of the transient heat transfer between solid, liquid, and vapor; and large-
scale heated surface.
In the recent decades, several different CFD models were developed to investi-
gate the bubble dynamics based on single-bubble simulation. Lee and Nydahl (1989)
simulate the single-bubble growth in nucleation boiling based on moving mesh with
the generalized arbitrary Lagrangian-Eulerian (ALE) approach. However, in their
models, the assumption of a constant wall temperature and lack of suitable modeling
of the detachment lead to only partial accurate prediction of the bubble dynamics.
Later in order to avoid the complex mesh generation process, other interface tracking
methods (ITM) were applied in the CFD simulations, such as the level set method,
the volume of fluid method (VOF), the front tracking method (FTM), the color/
density function, the phase-field method (PFM), and the lattice Boltzmann approach
(LB). Son et al. (1999) succeeded in modeling the bubble dynamics using level set
method to track the liquid/vapor interface. The computed bubble dynamics were in
good agreement with experimental measurement. However, the constant temperature
assumption for the heat transfer surface makes the model be unable to predict the
bubble waiting time and frequency which is very important in the global boiling
process. Further, Stephan and Busse (1992) tried to simulate the single bubble
dynamics based on the volume of fluid method, In this model, the temperature on
the wall is allowed to vary. Due to lack of validation, the accuracy of the model is
still uncertain. Recently, Sato and Niceno (2013) developed a new model based on
color/density function. In this model, they could simulate the dry spot underneath the
bubble; they could even show the bubble growth rate and the temperature distribu-
tion over the heat transfer surface which were in good agreement with experimental
data. However, because of the large computational effort, their model is only limited
in few nucleation sites and only available for pool nucleation boiling. Another big
disadvantage is that the simulation domain in the models above is strictly limited in
the millimeter to centimeter size.
Though computational science was well developed, it is still impossible to
directly simulate both small scale of bubbles (~mm, ~cm) and large scale (~meter)
of the global boiling process in one model. Nevertheless, the direct numerical
simulations could yield essential information for the derivation of correlation valid
in a wide range of parameters. For industrial applications, the simplification of the
subprocess models based on microphysics is still required in the numerical
simulation.
Individual Subprocess Models
Microlayer Model
Cooper and LIoyd (1969) considered the shape of the bubble is hemispherical at the
first growth phase of a bubble due to the hindering by the liquid surrounding the
bubble. Then there should be a “microlayer” between the vapor hemisphere and
heated surface. This “microlayer” was proposed by Snyder (1956) and verified by
Judd and Hwang (1976). They accounted for up to 30% of the total heat transfer for
the pool boiling of dichloromethane experimentally.
In the last decades, based on the microlayer theory, Zhao et al. (2002a, b, c) have
established a dynamic microlayer model that focuses on individual bubbles and can
be used to explain pool boiling mechanism. Their model can predict heat flux in the
nucleate boiling region at high heat flux and in the transition boiling region. They
concluded that in the nucleate boiling and the low-super heat region of transition
boiling, the heat flux is mainly contributed by the evaporation of the microlayer. In
the high-super heat region of transition boiling, the heat flux is contributed by both
microlayer and macrolayer.
Bubble Growth
The bubble growth takes place as inertia controlled growth and thermal diffusion
controlled growth. The inertia controlled growth follows the rule of Rayleigh
solution established by Mikic et al. (1970). In this period, a microlayer is formed
under the bubble. Later the evaporation of the microlayer and the evaporation around
cap controlled by thermal diffusion contribute to the bubble growth. The thermal
diffusion controlled growth follows the rule of Labuntsov (1975) solution. During
the bubble growth, the new microlayer is formed due to the expansion of gas liquid
interface on the wall. The evaporation of microlayer causes the dry out on the
boiling wall.
Bubble Departure
As is introduced, Klausner et al. (1993) developed a model based on the balance of
the forces acting on the bubble to predict the departure and lift off of the bubble (see
also section “Bubble Detachment Based on Force Balance”). He obtained a satis-
factory predictive accuracy against own data of flow boiling with refrigerant R113.
In the model, the base diameter (contact diameter dw) of bubble was recommended to
be 0.09 mm and advancing/receding contact angle (βad, βre) of π/4 and π/5, respec-
tively. Later, modified versions of the Klausner model have been applied by many
different authors to predict their own experimental data such as Thorncroft et al.
(2001), Situ et al. (2005), Sugrue (2012), and Chen (2012) but with their own
calibration of dw, βαd, and βre. Klausner applied the Mikic et al. (1970) model to
simulate the bubble growth, while Situ and most of latter authors employed the
Zuber (1961) formulation. Zuber included in his formulation a parameter b to
account for bubble sphericity. This parameter has been used by Sugrue (2012)
with different value from 0.24 to 24 as calibration to fit the models with their
experiment data.
With above well-developed subprocess model, the single bubble life cycle could
be modelled without complex vapor/liquid interface tracking. The complexity of this
sub-model should be suitable to be implemented in the modelling of the global
boiling process.
Summary
For the CFD modelling of boiling at a heated wall by an Euler/Euler description of

two-phase flow in the past, a heat flux partitioning model describing the microscopic
phenomena at the wall by empirical correlations adapted to experimental data has
gained some success. Such an approach was used and adjusted to boiling experi-
ments with water at a pressure of several MPa and also to experiments applying
different refrigerants.
Nevertheless, this approach is far away from a predictive tool. A critical review of
the detailed correlations shows that some of the parameters are not suited for a broad
usage for different fluids or different pressure levels but have to be carefully
recalibrated for the intended applications. Then some deviations can be calculated
with good success.
An improvement of the models can only be achieved by a more detailed micro-
physics. This particularly is necessary when critical heat flux should be simulated.
Then besides the physical micro phenomena, the morphology of the fluid has to be
considered.
In many cases, the models can process more information than is available from
the experiment. One example is the nucleation site density; their importance on
boiling modelling was shown. A model of the nucleation site density on a heated
wall is very difficult, if not impossible. Even when the heat properties of the material
are known commonly, no information exists on surface treatment. As a compromise
the gap can be closed by matching the measured and calculated temperature of the
heated wall which has a sensitive dependence on the nucleation site distribution. A
suitable model must be starting from the question to be solved considering a good
weight between necessary information and calculated parameters.
Nomenclature

a – Bubble influence factor
AconvL – Area fraction for heat convection to liquid
AW – Area fraction occupied by bubbles (AW + AconvL = 1)
AI – Interfacial area density
CB – Bubble-induced turbulence coefficient (Sato et al. 1981) model)
CD – Drag coefficient
CL – Lift coefficient
CP J K1 kg3 Specific heat capacity at constant pressure
CTD – Turbulent dispersion coefficient
CVM – Virtual mass force coefficient
CW – Wall force coefficient
Cμ – Shear-induced turbulence coefficient (k-ε model)
dB m Bulk bubble diameter
d⊥ m Bubble diameter perpendicular to main motion
dW m Bubble detachment diameter
D m Pipe diameter
Eo – Eötvös Number
f Hz Bubble detachment frequency
FD N m3 Drag force
FL N m3 Lift force
FTD N m3 Turbulent dispersion force
FVM N m3 Virtual mass force
FW N m3 Wall force
g m s2 Acceleration of gravity
hCL W m2K1 Heat transfer coefficient for single-phase convection to liquid
hCG W m2K1 Heat transfer coefficient for single-phase convection to gas
hLG W m2K1 Heat transfer coefficient for bulk evaporation / condensation
hQ W m21 Heat transfer coefficient for quenching
H J kg3 Specific enthalpy
HLG J kg3 Specific evaporation enthalpy
Ja – Jakob number
kL W m1 K1 Thermal conductivity
k m2s2 Turbulent kinetic energy
L m Length scale
m_ W kg s1 Vapour mass flux
Mo – Morton number
N m3 Nucleation site density
p Pa Pressure
Pr – Prandtl number
Qtot W m2 Total wall heat flux
QC W m2 Heat flux due to single-phase convection
(continued)

QQ W m2 Heat flux due to quenching
QE W m2 Heat flux due to evaporation
r m Radial coordinate
Re – Reynolds number
s m Hydrodynamic wall roughness
t s Time
twait s Waiting time
T K Temperature
TL K Liquid temperature
Tsat K Saturation temperature
Tsub K Liquid subcooling
Tsup K Wall superheat
TW K Wall temperature
u m s1 Velocity
uτ m s1 Friction velocity
U m s1 Velocity scale
V m3 Volume
x m Axial coordinate
y m Distance to the wall
α – Volume fraction
δ m Viscous length scale
ε m2s3 Turbulent dissipation rate
ΔT K Temperature scale
μ kg m1s1 Dynamic viscosity
ν m2s1 Kinematic viscosity
ρ kg m3 Density
σ N m1 Surface tension
τW N m2 Wall shear stress
Θ grd Wall contact angle
References
H. Anglart, O. Nylund, N. Kurul, M.Z. Podowski, CFD prediction of flow and phase distribution in
fuel assemblies with spacers, NURETH-7, 1995 Saratoga Springs, New York. Nucl. Eng. Des.
177, 215–228 (1997)
G.G. Bartolomej, V.M. Chanturiya, Experimental study of true void fraction when boiling sub-
cooled water in vertical tubes. Therm. Eng. 14, 123–128 (1967.) translated from
Teploenergetika 14(2), 80–83
N. Basu, G.R. Warrier, V.K. Dhir, Onset of nucleate boiling and active nucleation site density during
subcooled flow boiling. ASME J. Heat Transf. 124, 717–728 (2002)
R.J. Benjamin, A.R. Balakrishnan, Nucleation site density in pool boiling of saturated pure liquids:
effect of surface microroughness and surface and liquid physical properties. Exp. Thermal Fluid
Sci. 15, 32–42 (1997)
W.C.B. Ceumern-Lindenstjerna, in Heat Transfer in Boiling, ed. by E. Hahne, U. Grigull. Bubble

departure diameter and release frequencies during nucleate pool boiling of water and aqueous
NaCl solutions (Academic Press and Hemisphere, 1977)
D. Chen, L. Pan, S. Ren, Prediction of bubble detachment diameter in flow boiling based on force
analysis. Nucl. Eng. Des. 243, 263–271 (2012)
J.T. Cieslinski, J. Polewski, J.A. Szymczyk, Flow field around growing and rising vapour bubble by
PIV measurement. J. Vis. 8, 209 (2005)
R. Cole, A photographic study of pool boiling in the region of the critical heat flux. AIChE J. 6,
533–542 (1960)
M.G. Cooper, A.J.P. Lloyd, The microlayer in nucleate pool boiling. Int. J. Heat Mass Transf. 12,
895–913 (1969)
M. del Valle, D. Kenning, Subcooled flow boiling at high heat flux. Int. J. Heat Mass Transf. 28,
1907 (1985)
D.A. Drew, S.L. Passman, Theory of Multicomponent Fluids (Springer, New York, 1998)
W. Fritz, Phys. Z. 36, 379 (1935)
polydisperse air-water two-phase flows in pipes. Nucl. Eng. Des. 238, 647–659 (2008)
J. Garnier, E. Manon, G. Cubizolles, Local measurements on flow boiling of refrigerant R12 in a
vertical tube. Multiph. Sci. Technol. 13, 1–111 (2001)
C.-Y. Han, P. Griffith, The mechanism of heat transfer in nucleate pool boiling. Int. J. Heat Mass
Transf. 8, 887 (1965)
T. Hibiki, M. Ishii, Active nucleation site density in boiling systems. Int. J. Heat Mass Transf. 46,
2587–2601 (2003)
Ishii, M., Thermo-Fluid Dynamic Theory of Two-Phase Flow (Paris, Eyrolles, 1975)
H.J. Ivey, Relationships between bubble frequency, departure diameter, and rise velocity in nucleate
boiling. Int. J. Heat Mass Transf. 10, 1023 (1967)
R.L. Judd, K.S. Hwang, A comprehensive model for nucleate pool boiling heat transfer including
microlayer evaporation. J. Heat Transf. 98, 623–629 (1976)
B.A. Kader, Temperature and concentration profiles in fully turbulent boundary layers. Int. J. Heat
Mass Transf. 24, 1541–1544 (1981)
S.G. Kandlikar, M.E. Steinke, Contact angles and interface behavior during rapid evaporation of
liquid on a heated surface. Int. J. Heat Mass Transf. 45, 3771–3780 (2002)
J. Klausner, R. Mei, D. Bernhard, L. Zeng, Vapor bubble departure in forced convection boiling. Int.
J. Heat Mass Transf. 36, 651–662 (1993)
G. Kocamustafaogullari, Pressure dependence of bubble diameter for water. Int. Commun. Heat
Mass Transf. 10, 501–509 (1983)
G. Kocamustafaogullari, M. Ishii, Interfacial area and nucleation site density in boiling systems. Int.
J. Heat Mass Transf. 26, 1377 (1983)
N.I. Kolev, Uniqueness of the elementary physics driving heterogeneous nucleate boiling and
flashing. Nucl. Eng. Technol. 38, 175 (2006)
B. Koncar, E. Krepper, CFD simulation of convective flow boiling of refrigerant in a vertical
annulus. Nucl. Eng. Des. 238, 693 (2008)
B. Koncar, B. Mavko, Simulation of boiling flow experiments close to CHF with the neptune CFD
code. Sci. Technol. Nucl. Installations 2008, 732158 (2008)
B. Koncar, B. Mavko, Wall function approach for boiling two-phase flows. Nucl. Eng. Des.. 2010
240, 3910 (2010)
E. Krepper, R. Rzehak, CFD for subcooled flow boiling: analysis of DEBORA tests. J. Comput.
Multiphase Flows 6, 329–359 (2014)
E. Krepper, B. Koncar, Y. Egorov, Modelling of subcooled boiling – concept, validation and
application to fuel assembly design. Nucl. Eng. Des. 237, 716–731 (2007)
simulation of polydispersed flows, Nucl.Eng. Des. 238, 1690–1702 (2008)
E. Krepper, R. Rzehak, C. Lifante, T. Frank, CFD for subcooled flow boiling: coupling wall boiling
and population balance models. Nucl. Eng. Des. 255, 330–346 (2013)
N. Kurul, M.Z. Podowski, Multidimensional effects in forced convection subcooled boiling, in
Proc. 9th Int. Heat Transfer Conf., Jerusalem, Israel (1990)
N. Kurul, M. Podowski, On the modeling of multidimensional effects in boiling channels, in ANS
Proceedings of 27th National Heat Transfer Conference, Minneapolis (1991)
D. Labuntsov, Current theories of nucleate boiling of liquids. Heat Transf. – Sov. Res. 7, 1–15
(1975)
R.C. Lee, J.E. Nydahl, Numerical calculation of bubble growth in nucleate boiling from inception
through departure. J. Heat Trans. 111, 474–479 (1989)
M. Lemmert, J.M. Chawla, in Heat Transfer in Boiling, ed. by E. Hahne, U. Grigull, Influence of
Flow Velocity on Surface Boiling Heat Transfer Coefficient (Academic Press and Hemisphere,
1977), ISBN: 0-12-314450-7, 237–247
C. Lifante, F. Reiterer, T. Frank, A. Burns, Coupling of wall boiling with discrete population
balance model, The 14th International Topical Meeting on Nuclear Reactor Thermalhydraulics,
NURETH-14, Toronto, Ontario, Canada, September 25–30, (2011), NURETH14-087
C. Lifante, T. Frank, A. Burns, RPI wall boiling extension towards CHF 2013, presentation (2013)
D. Lucas, E. Krepper, CFD Models for Polydispersed Bubbly Flows (Forschungszentrum FZD,
Dresden, 2007), p. 486
D. Lucas, E. Krepper, H.-M. Prasser, Use of models for lift, wall and turbulent dispersion forces
D.M. Lucas, T. Frank, C. Lifante, P. Zwart, A. Burns, Extension of the inhomogeneous MUSIG
model for bubble condensation. Nucl. Eng. Des. 241, 4359–4367 (2011)
F. Menter, Two-equation eddy-viscosity turbulence models for engineering applications. AIAA-J.
32(8), 1598–1605 (1994)
F.R. Menter, Review of the shear-stress transport turbulence model experience from an industrial
perspective. Int. J. Comput. Fluid Dyn. 23, 305–316 (2009)
B.B. Mikic, W.M. Rohsenow, A new correlation of pool-boiling data including the fact of heating
surface characteristics. ASME J. Heat Transf. 91, 245–250 (1969)
B. Mikic, W. Rohsenow, P. Griffith, On bubble growth rates. Int. J. Heat Mass Transf. 13,
657 (1970)
S. Mimouni, C. Baudry, M. Guingo, J. Lavieville, N. Merigoux, N. Mechitoua, Computational
multi-fluid dynamics predictions of critical heat flux in boiling flow. Nucl. Eng. Des. 299, 28–36
(2016)
S.B. Pope, Turbulent flow, ISBN 0521 59125 2 (2000)
V. Prodanovic, D. Fraser, M. Salcudean, Bubble behaviour in subcooled flow boiling of water at low
pressures and low flow rates. Int. J. Multiphase Flow 28, 1–19 (2002)
F. Ramstorfer, B. Breitschädel, H. Steiner, G. Brenn, Modelling of the near-wall liquid velocity field
in subcooled boiling flow, in Proceedings of HT2005, ASME Summer Heat Transfer Conference
2005, San Francisco (2005)
W.E. Ranz, W.R. Marshall, Evaporation from drops. Chem. Eng. Prog. 48(3), 141–146 (1952)
Y. Sato, B. Niceno, A sharp-interface phase change model for a mass-conservative interface
tracking method. J. Comput. Phys. 249, 127–161 (2013)
Y. Sato, M. Sadatomi, K. Sekoguchi, Momentum and heat transfer in two-phase bubble flow-I. Int.
R. Situ, T. Hibiki, M. Ishii, M. Mori, Bubble lift-off size in forced convective subcooled boiling
flow. Int. J. Heat Mass Transf. 48, 5536–5548 (2005)
N.W. Snyder, Summary of conference on bubble dynamics and boiling heat transfer held at the jet
propulsion laboratory, JPL Memo No. 20–137 (Jet Propulsion Laboratory, California Institute of
Technology, 1956)
G. Son, V.K. Dhir, N. Ramanujapu, Dynamics and heat transfer associated with a single bubble
during nucleate boiling on a horizontal surface. J. Heat Trans. 121, 623–631 (1999)
P.C. Stephan, C.A. Busse, Analysis of the heat transfer coefficient of grooved heat pipe evaporator
walls. Int. J. Heat Mass Transf. 35, 383–391 (1992)
R.M. Sugrue, The effects of orientation angle, subcooling, heat flux, mass flux, and pressure on
bubble growth and detachment in subcooled flow boiling, Master Thesis in Nuclear Science and
Engineering, Massachusetts Institute of Technology, Cambridge, MA (2012)
G.E. Thorncroft, J.F. Klausner, R. Mei, Bubble forces and detachment models. Multiph. Sci.
Technol. 13, 35–76 (2001)
V.I. Tolubinsky, D.M. Kostanchuk, Vapour bubbles growth rate and heat transfer intensity at
subcooled water boiling; Heat Transfer 1970, Preprints of papers presented at the 4th Interna-
tional Heat Transfer Conference, Paris, 5, Paper No. B-2.8 (1970)
H.C. Ünal, Maximum bubble diameter, maximum bubble growth time and bubble growth rate. Int.
J. Heat Mass Transf. 19, 643–649 (1976)
C.H. Wang, V.K. Dhir, Effect of surface wettability on active nucleation site density during pool
boiling of saturated water. J. Heat Transf. 115, 659–669 (1993)
F.M. White, Viscous Fluid Flow (McGraw-Hill, 1991), ISBN 0-07-069712-4
T. Wintterle, Development of a numerical boundary condition for the simulation of nucleate boiling
at heated walls, Diploma Thesis University Stuttgart, IKE - 8- D- 014 (2004)
S. Yang, R. Kim, A mathematical model of the pool boiling nucleation site density in terms of the
surface characteristics. Int. J. Heat Mass Transf. 31, 1127–1135 (1988)
G.H. Yeoh, J.Y. Tu, A unified model considering force balances for departing vapour bubbles and
population balance in subcooled boiling flow. Nucl. Eng. Des. 235, 1251–1265 (2005)
G.H. Yeoh, J.Y. Tu, Modelling Subcooled Boiling Flows, (Nova Science Publishers, Inc., 2009)
ISBN 978-1-60456-943-8
G.H. Yeoh, J.Y. Tu, Computational Techniques for Multiphase Flows – Basics and Applications
(Butterworth-Heinemann, Elsevier Science and Technology, 2010)
G.H. Yeoh, S.C.P. Cheung, J.Y. Tu, M.K.M. Ho, Fundamental consideration of wall heat partition
of vertical subcooled boiling flows. Int. J. Heat Mass Transf. 51, 3840–3853 (2008)
O. Zeitoun, M. Shoukri, Bubble behaviour and mean diameter in subcooled flow boiling. ASME
J. Heat Transf. 118, 110–116 (1996)
L. Zeng, J. Klausner, D. Bernhard, R. Mei, A unified model for the prediction of bubble detachment
diameters in boiling systems – I. Pool boiling. Int. J. Heat Mass Transf. 36, 2261–2270 (1993a)
L. Zeng, J.F. Klausner, D.M. Bernhard, R. Mei, A unified model for the prediction of bubble
detachment diameters in boiling systems – II Flow boiling. Int. J. Heat Mass Transf. 36,
2271–2279 (1993b)
Y.-H. Zhao, T. Masuoka, T. Tsuruta, Unified theoretical prediction of fully developed nucleate
boiling and critical heat flux based on a dynamic microlayer model. Int. J. Heat Mass Transf. 45,
3189–3197 (2002a)
Y.-H. Zhao, T. Tsuruta, T. Masuoka, Prediction of bubble-behavior in subcooled pool boiling based
on microlayer model. JSME Int. J. B: Fluids Therm. Eng. 2002(45), 346–354 (2002b)
Y.-H. Zhao, T. Tsuruta, T. Masuoka, Critical heat flux prediction of subcooled pool boiling based on
the microlayer model. JSME Int. J. B: Fluids Therm. Eng. 2002(45), 712–718 (2002c)
N. Zuber, The dynamics of vapor bubbles in nonuniform temperature fields. Int. J. Heat Mass
Transf. 2, 83–98 (1961)
Multiphase Flows in Pharmaceutical
Applications
Z.B. Tong, R.Y. Yang, H.K. Chan, and A.B. Yu
Abstract
Dry powder inhalers (DPI) have been widely used for drug delivery to the
respiratory tract to treat disease such as asthma and cystic fibrosis. It is therefore
important to improve their efficiency through better understanding of dispersion
mechanisms of powders in devices. In this aspect, numerical modeling has been
playing an increasingly important role. This chapter reviews the recent progress
using computational fluid dynamics (CFD) and the discrete element method
(DEM) to investigate the airflow and de-agglomeration in DPIs. In particular,
with the coupled CFD and DEM approach, the models are able to generate
detailed information at the particle scale, leading to improved and novel designs
of next generations of DPI systems.
Keywords
Aerosol drug delivery • Dry powder inhaler • Particle-scale modeling • Discrete
element method • Computational fluid dynamics • Powder dispersion
Z.B. Tong (*)

Key Laboratory of Energy Thermal Conversion and Control of Ministry of Education, School of
Energy and Environment, Southeast University, Nanjing, China
e-mail: z.tong@seu.edu.cn
R.Y. Yang
School of Materials Science and Engineering, University of New South Wales, Sydney, NSW,
Australia
e-mail: r.yang@unsw.edu.au
H.K. Chan
Faculty of Pharmacy, University of Sydney, Camperdown, NSW, Australia
e-mail: kim.chan@sydney.edu.au
A.B. Yu
Department of Chemical Engineering, Monash University, Clayton, VIC, Australia
e-mail: Aibing.Yu@monash.edu

DOI 10.1007/978-981-4585-86-6_22-1
2 Z.B. Tong et al.
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Developments of Numerical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Computational Fluid Dynamics (CFD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Discrete Element Method (DEM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
CFD-DEM Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Model Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Applications of CFD to DPIs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
DEM Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
CFD-DEM Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Conclusion and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Introduction
Dry powder inhalation is a popular aerosol drug delivery for airway disease. Since
developed 45 years ago (Bell et al. 1971), the technique has been used to treat
asthma, cystic fibrosis (CF), and chronic obstructive pulmonary disease (COPD)
(Gonda 1992; Frijlink and De Boer 2004, Zhou et al. 2014). Attempts have also been
made to use the inhalation route to deliver many drugs such as proteins and genes for
systemic effects (e.g., using the lung to deliver insulin to the bloodstream for
diabetics) (Patton 1998). Dry powder inhalation has several advantages comparing
to other aerosol drug delivery methods, such as portable, chemically stable, delivery
of drug using the patient’s respiration and obvious environmental benefits. Currently,
various commercial dry powder inhalers (DPIs) are already on the market, and more
are under development.
The current pharmaceutical DPIs, however, have rather poor aerosol performance
with less than 30% of the dose being able to reach the lungs. Particles with
aerodynamic size range of 1–5 μm are required for the delivery to the more distal
parts of the respiratory tract. Such fine powders are often very cohesive and difficult
to disperse. To improve dispersion performance, development has been focused on
reducing cohesion between particles, e.g., using porous particles, micronized carrier
powders, or corrugated particles (Chan 2006a). On the other hand, powder disper-
sion performance is more efficient by increasing the dispersion/de-agglomeration
forces in devices, which can be achieved by changing inhaler design, environmental
conditions, and inhalation flow rate. The dispersion process generally involves
multiple physical factors due to inhaler-particle interactions as well as particle-
particle and particle-fluid interactions, as shown in Fig. 1 (Finlay 2001). But how
they exactly occur and what is their relative importance are still not clear. The
consequent lack of basic understanding of the dispersion process is a major hin-
drance to the advance of pharmaceutical powder aerosol technology.
Detailed information of the fluid flow and particle behavior will help understand
dispersion mechanisms of DPIs. However, obtaining such information from exper-
iments is extremely difficult (Zhu et al. 2007), if not impossible because of the small
particle size (less than 5 μm) and extremely short duration of impacts. So far
Multiphase Flows in Pharmaceutical Applications 3
Drug Agglomerates 2
1 Particle – device impaction
Particle – air interaction
3
Eddies
Particle – particle impaction
Fig. 1. De-agglomeration mechanisms (Finlay 2001)
experiments have been largely limited to post-dispersion analysis. On the other hand,
numerical modeling has been playing an increasingly important role in this area.
Computation fluid dynamic (CFD) has been used to understand the airflow inside
DPIs and to investigate the effect of design on dispersion performance (Wong et al.
2012; Ruzycki et al. 2013). The studies have shown that aerosol performance
depends on the inhaler characteristics and prevailing flow conditions. They also
highlighted the important role of computational modeling in inhaler design for
pharmaceutical aerosol generation. While CFD is unable to model the dynamics
and interaction of discrete particles, such limitation can be overcome by the discrete
element method (DEM)-based models (Calvert et al. 2009). The DEM models
explicitly calculate the interparticle forces (Cundall and Strack 1979) and provide
dynamic information of individual particles. By combining CFD with DEM, the
approach can generate information in two important aspects: flow structure and force
structure at individual particle scale, thus allowing the dispersion of powders to be
accurately described. This chapter will review the development in this area and
discuss a perspective on the future work.
Developments of Numerical Models
Computational Fluid Dynamics (CFD)
CFD is a tool widely used to simulate and analyze fluid flow. It is the numerical
technique to solve the Navier-Stokes equations which describe the behavior of fluid
flow. While direct numerical simulation (DNS) and large eddy simulation (LES) are
more accurate to model turbulent flows, their applications in this area are rare due to
their extremely high computational cost. The most commonly used approach for
drug aerosol modeling is to solve the Reynolds averaged Navier-Stokes (RANS)
equations, given by (Anderson and Jackson 1967):
∇ ðuÞ ¼ 0 (1)
@ ðuÞ
ρ ρ∇ ðuuÞ ¼ ∇P þ ∇ ðτÞ þ ρg þ ρ∇ ρ u0 u0 (2)
@t
4 Z.B. Tong et al.
where ū, u0, ρ, P, and τ are fluid mean velocity, turbulent velocity fluctuation, fluid
density, pressure, and fluid viscous stress tensor, respectively. The Reynolds aver-
aging of the transport equations introduces additional Reynolds stress term ρ u0 u0
whose closure is typically via a two-equation eddy viscosity model (Walters and
Leylek 2004; Genç et al. 2009) or a transported turbulent shear stress model
(Launder et al. 1975; Wang et al. 2006).
Discrete Element Method (DEM)
The macroscopic behavior of particle flow is governed by the interactions between

individual particles as well as interactions of particles with surrounding fluid and
wall. The DEM was originally developed by Cundall and Strack (1979) to study
granular dynamics and since has been extensively used in the study of various
phenomena, such as particle packing and compaction, particle flow, and particle-
fluid flow (Zhu et al. 2008).
In the DEM approach, the governing equations for the translational and rotational
motion of particle i with mass
dvi
mi ¼ Fi (3)
dt
dωi
Ii ¼ Mi (4)
dt
where vi and ωi are the translational and angular velocities of particle i, respec-
tively, Fi and Mi are the total force and torque acting on particle i. The total force
may include mechanical contact forces, noncontact forces such as van der Waals
force, capillary force and electrostatic force, particle-fluid force (e.g., the drag force,
Saffman and Magnus lift forces) as well as gravity (Zhu et al. 2008). Various models
have been proposed to calculate these forces and torques (Tong et al. 2015). Once the
forces and torques are known, Eqs. (3) and (4) can be solved numerically. Thus, the
trajectories, velocities, and the transient forces of all particles in a system considered
can be determined.
CFD-DEM Model
The CFD-DEM approach was firstly proposed by Tsuji et al. (1992), and then
followed by many others (Xu and Yu 1997). As mentioned above, with this
approach, the motion of discrete particles is described by the DEM based on the
Newton’s second law of motion and the flow of continuum fluid by CFD based on
the local averaged Navier-Stokes equations. The CFD-DEM coupling is achieved by
exchanging the information obtained from the DEM and the CFD at each step, as
shown in Fig. 2. At every time step with a given fluid flow condition, the DEM
Discrete Model vi , (x, y, z )i ,ε i
kc
fpf, i
∑ fpf,i
Fpf =
i=1
ΔV
ui , (u ‒ v )i , Re p,i Continuum Model
Fig. 2. Coupling and information exchange between continuum (CFD) and discrete (DEM)
models (Xu et al. 2001)
determines particle-related information such as the positions, velocities, and forces

of individual particles. The porosity and volumetric particle-fluid interaction force in
the individual computational cells are then calculated and passed to the CFD which
uses these data to determine the air flow field and determines the interaction from the
fluid to individual particles. The resulting forces are then passed to the DEM to
determine the state of particles at the next time step, and the process continues (Chu
et al. 2009). In some situations particles have a minimum effect on fluid flow (e.g.,
dispersion of single agglomerate in a uniform flow), a one-way coupling method in
which only the interaction of fluid on particles is considered is preferred to reduce
simulation time.
Model Applications
Applications of CFD to DPIs
CFD simulations have been performed to investigate the flow field inside the DPIs
(Matida et al. 2003; Coates et al. 2004; Matida et al. 2004; Coates et al. 2005). The
studies by Coates et al. showed that the change to DPI designs, such as grid size
(Coates et al. 2004), mouthpiece geometry (Coates et al. 2007), inlet air dimension
(Coates et al. 2006), the presence of a capsule (Coates et al. 2005), and air flow rate,
can affect aerosol performance significantly. Figure 3 shows the effect of the
structure of the inhaler grid on flow velocity. However the modified mouthpiece
6 Z.B. Tong et al.
Fig. 3. Different designs of the grid and their effects on velocity profile (Coates et al. 2004)
designs of Aerolizer ® (Fig. 4) are able to reduce throat deposition but have no effect
on the overall dispersion performance characterized in terms of fine particle fraction
(FPF) (Coates et al. 2007). The change of the inhaler mouthpiece length was also
found to play a less significant role. On the other hand, air inlets control the levels of
turbulence and particle impaction velocities generated in the device, as well as the
flow development rate and device emptying times. Modification of their dimensions
(Coates et al. 2006) has a varying effect on the inhaler performance depending on
flow rates (Fig. 5).
De Boer et al. (2012) used CFD to investigate the performance of a commercial
high-dose disposable DPI named Twincer™ as shown in Fig. 6. Their results
indicated that the flow across the inhaler was independent of the pressure drop.
The dispersion efficiency or deposition was not affected by the flow rate. By
Fig. 4. Schematic of the different modified mouthpiece designs (Coates et al. 2007)
Fig. 5. Schematic of the device designs of the air inlet size on inhaler performance (Coates et al.
2006)
reducing the resistance of the classifier bypass channels, the classifier symmetry can
be improved.
Donovan et al. (2012) simulated two commercial DPIs, Handihaler ® and
Aerolizer ® (Fig. 7), to investigate the effects of carrier particle size and shape on
inhaler performance. The results indicated the number of particle-wall collisions
increased with carrier particle size in Aerolizer ® but was independent of carrier
particle size in Handihaler ®. Milenkovic (Milenkovic et al. 2013; Milenkovic and
Alexopoulos 2014), on the other hand, investigated the airflow and particle
8
1 μM; 4 kPa 3 μM; 4 kPa
0.000 0.002 0.004 0.006 0.008 0.010 0.012 0.014 0.016 0.018 0.020 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10 0.11 0.12 0.13 0.14 0.15
10 μM; 4 kPa 100 μM; 4 kPa
Fig. 6. Particle trajectories in Twincer™ colored by residence time for particles of different sizes (de Boer et al. 2012)
Z.B. Tong et al.
Fig. 7. Handihaler ® (left)

and Aerolizer ® (right) a
a contour of velocity
magnitude; b carrier particle
trajectory (Donovan et al.
2012)
Speed (m/s)
90
80
70
60
50
40
30
20
10
0
deposition in the Turbuhaler ®, as shown in Fig. 8. Their simulation results for

particle deposition and fine particle fraction agreed well with available experimental
data.
Following the studies proposed by Coates et al. (2006), the effect of device design
of the Aerolizer ® on the aerosolization of a carrier-based formulation has been
investigated by Zhou et al. (2013). The results showed the air inlet size and grid
10 Z.B. Tong et al.
Fig. 8. a Contour of velocity a 32

magnitude (Milenkovic et al. 30
2013); b deposited particles
for steady flow in the 28
Turbuhaler ® (Milenkovic and 26
Alexopoulos 2014)
24
22
20
18
16
14
12
10
8
6
4
2
0
b
10
9.5
9
8.5
8
7.5
7
6.5
6
5.5
5
4.5
4
3.5
3
2.5
2
1.5
1
Mouthpiece
Grid
Inhaler
chamber
Air inlet
Original 1/3 mouthpiece 1/3 inlet size Cross grid

Iength
Fig. 9. Diagram of the original and modified Aerolizer ® devices (Zhou et al. 2013)
Fig. 10. Agglomerate-based formulation (left) and Carrier-based formulation (right)
structure of the Aerolizer ® have significant effect on the aerosolization performance

(Fig. 9).
The work mentioned above improved our understanding of the performance of
DPIs which can be extended to other dry powder inhaler systems to provide critical
information for optimal inhaler design. On the other hand, the work also showed the
limitations of simulating the powder dispersion process (Iimura et al. 1998;
Higashitani et al. 2001; Iimura and Higashitani 2005). The limitation of CFD
model can be overcome by explicit calculation of the particle contact mechanics
using the DEM.
12 Z.B. Tong et al.
DEM Applications
The DEM has been applied to investigate the mechanical impaction between
agglomerates and devices. There are two kinds of drug powder formulations used
in DPIs, e.g., agglomerate-based and carrier-based formulation, as shown in Fig. 10.
The applications of DEM modeling in the dispersion of two formulations due to
impaction are discussed separately in the following sections.
(i) Dispersion of agglomerate-based formulation due to impaction
In the early DEM research, Thornton and his co-workers (Thornton et al. 1996,
1999; Ning et al. 1997) have reported impact simulations on lactose agglomerates
(Fig. 11) and observed a minimum velocity below which no significant damage
occurs, and the agglomerate behaves like a large single particle. Similar conclusions
were also obtained by Subero et al. (1999) who showed that the extent of breakage
increased with impact velocity but eventually reached a limit beyond which the
breakage approached an asymptotic value. As shown in Fig. 12, a modified Weber
2
number We0 ¼ ρðUU Γ
oÞ D
was linked to the damaged ratio.
Moreno et al. (2003, 2006) studied the effect of impact angle and surface energy
on the breakage of agglomerates and found that the normal component of impact
velocity was the dominant factor. Comparing the previous work, the wider range of
surface energies have been tested by Moreno et al., and a new dimensionless group,
Δ ¼ WeI2=3e , was able to provide a much better unification of data compared to the
modified Weber number We0 proposed by Subero et al. (1999), as shown in Fig. 13.
In DPIs, powders are loosely bonded in order to have easy dispersion when in use
(Chan 2006b). Comparing with hard, strong agglomerates, soft agglomerates behave
quite differently on impacts and often disintegrate into many small pieces instead of
breaking into several large chunks (Ning et al. 1997; Boerefijn et al. 1998). Yang
et al. (2008) investigated the agglomeration of fine particles down to 5 μm in size
under an assumed centripetal force. Their results showed that the strength of an
agglomerate can be predicted by the modified Rumpf model (Rumpf 1962).
Recently, Tong et al. (2009) simulated the loose and weak agglomerates impacted
with a wall at different velocities and angles. The simulation results showed that the
agglomerates experienced significant plastic deformations before breaking into
many small pieces/fragments. While the normal (90 ) impaction generated large
forces inside the agglomerate, the deformation area was much larger with the 45
impaction. This suggested that the shear force played a very important role in the
breakage of the agglomerates. This is different from the breakage of hard agglom-
erates of which the normal impact is the dominant breakage factor (Moreno et al.
2003). This was further confirmed by analyzing the size distribution of fragments
after impaction. As shown in Fig. 14a, at a given impact velocity, the 45 impaction
always generated the smallest fragment size characterized by the 80 % passing size
(P80). To quantify the breakage efficiency, the total impact energy (work) from the
wall to the agglomerates was analyzed. The results showed that the mechanical
breakage efficiency was governed by the total impact energy which includes energy
in both the normal and shear directions, as shown in Fig. 14b.
(ii) Dispersion of carrier-based formulation
The carrier-based formulation is more popular in the DPIs than the agglomerate-
based formulation. Recently, the DEM method (Yang et al. 2013) has been used to
investigate the dispersion process of the carrier-based agglomerate due to wall
impaction. The effects of impact velocity, impact angle, and work of adhesion on
the dispersion performance were analyzed. As shown in Fig. 15, the results showed
that the impact-induced dispersion performance for carrier-based DPI formulations
can be well approximated using a cumulative Weibull distribution function that is
governed by the ratio of overall impact energy and adhesion energy.
Fig. 11. Space lattice at a t = 9 μs and b t = 12 μs of an agglomerate upon impaction (Thornton

et al. 1999)
Fig. 12. Correlation between

the damage ratio and the 0.5 Jm-2
modified Weber number We0 0.8 2.0 Jm-2
for all the assemblies and 5.0 Jm-2
Damage Ratio (-)
impact velocities above the

lowest threshold (Subero et al. 0.6
1999)
0.4
0.2
0
0.1 1 10 100 1000 10000
We' (-)
14 Z.B. Tong et al.
a
1.0
0.35 J/m2
0.8
3.50 J/m2
Damage ratio
0.6 35.0 J/m2
0.4
0.2
0.0
1E–7 1E–6 1E–5 1E–4 1E–3 0.01 0.1 1

We⬘
b
1.0
0.35 J/m2
0.8 3.50 J/m2
35.0 J/m2
Damage ratio
0.6
0.4
0.2
0.0
0.01 0.1 1 10 100 1000 10000 100000

Δ = We Ie2/3
Fig. 13. Relationship between damage ratio and a modified Weber number, We0 ; b new dimen-
sionless group, Δ, for different values of surface energy (Moreno and Ghadiri 2006)
Yang et al. (2013; 2015) also used the DEM method to investigate the mixing
process for carrier-based dry powder inhaler formulations (Fig. 16) and found that
amplitude and frequency of the vibration velocity can be controlled to maximize the
mixing of small particles with the carrier. The results also showed that the electro-
static force can result in a different mixing behavior comparting to the van der Waals
force.
CFD-DEM Applications
This section discusses the main applications of CFD-DEM models on the dispersion
of agglomerate-based and carrier-based formulation with the airflow.
(i) Dispersion of agglomerate-based formulation with airflow
a 1.0
0.9
0.8
Normalized P80
0.7
0.6
0.5
0.4
4 m/s
0.3 6 m/s
8 m/s
0.2 10 m/s
0.1
0 10 20 30 40 50 60 70 80 90
Impact Angle (degree)
Fig. 14. a Postimpact fragment size (P80) as a function of impact angle; b total agglomerate-wall
impact energy at different impact velocities and angles (Tong et al. 2009)
16 Z.B. Tong et al.
Fig. 15. The variation of

dispersion ratio with the
energy ratio from the
perspective of the whole
agglomerate (Yang et al.
2013)
Fig. 16. Snapshots of drug and carrier particles mixing at different time instances
Fig. 17. Dispersion behavior of an agglomerate in a uniform flow field (Calvert et al. 2011)
a b
Particle velocity
15
Particle velocity (cm/s)
15
12
12
8 8
4
4
0.980167
0.787641
c d
Particle velocity
15
15
12
12
8 8
4
4
0.980167
0.787641
e f
Particle velocity
15
15
12
12
8
8
4
0.980167
0.787641
Fig. 18. The change in the agglomerate after impact (Nguyen et al. 2014)
The dispersion of agglomerates by airflow has been investigated by Iimura et al.

(2009a, b) and Calvert (Calvert et al. 2011; Calvert et al. 2013). As shown in Fig. 17,
the results showed beyond a threshold of particle-fluid velocity, dispersion occurs
quickly and approaches a completely dispersed state asymptotically. Agglomerate
dispersions due to impaction with a target particle or cylindrical obstacle with
airflow were simulated by Iimura et al. (2009a, b) and Nguyen et al. (2014). Figure 18
showed the fine fragments were attached to the target due to the restructuring
mechanism that occurred during impact (Nguyen et al. 2014).
Recently, the effect of multiple impactions on the agglomerate dispersion
(Fig. 19) has been investigated by Tong et al. (2011). The results indicated that the
generation of fine particles were mainly caused by the second impaction. After the
first impaction, the agglomerate broke into large-sized fragments which are
18 Z.B. Tong et al.
Fig. 19. De-agglomeration of powders in different throat designs (Tong et al. 2011)
subsequently broken into smaller pieces upon the second impaction. While increas-
ing flow rate and the number of impactions increased breakage, they also resulted in
larger powder deposition on the wall. To have an optimal dispersion, both throat
design and flow rate should be well considered.
While some studies (French et al. 1996; Li et al. 1996; Dunbar et al. 1998)
suggested that the drag force generated by turbulent flow is a major factor of
Fig. 20. Flow field about the particle cluster and resulting total forces on the fine particles for three
Reynolds numbers (Cui et al. 2014)
de-agglomeration, others (Finlay 2001; Voss and Finlay 2002) indicated that
mechanical impaction is more important for powder dispersion. The aerodynamic
dispersions of a loose aggregate (Calvert et al. 2011) and a carrier-based DPI
formulation (Yang et al. 2014) in a uniform fluid flow have been investigated. The
results showed that unless the relative velocity was above a threshold limit, the loose
aggregate or carrier-based DPI formulation would not deform and disperse but
accelerate as a single entity. As the flow velocity in DPIs is normally smaller than
the threshold limit, the dispersion directly induced by air flow only plays a minor
role in powder dispersion in DPIs. This was also confirmed by the study of a
cyclonic flow model (Tong et al. 2010).
20 Z.B. Tong et al.
Fig. 21. a Schematic view of a

the inhaler; b size distribution Air Inlet
of powders forming the
agglomerate. The inset shows
the formed agglomerate Grid
(colors represent particle
diameters)
Outlet
Air Inlet
Capsule
Barrel
b
Probability density function (mass)
0.3
0.2
0.1
0.0
0 1 2 3 4 5 6 7 8 9
Diameter (μm)
(ii) Dispersion of carrier-based formulation with airflow
Work has also been conducted to investigate the aerosolization of carrier-based

formulation caused by airflow (Cui et al. 2014; Yang et al. 2014). For the detachment
of the drug particles from the carrier through the fluid stresses (normal and tangential
force), the adhesion force (van der Waals) and the friction force have to be overcome.
Figure 20 shows the detachment probability by the three mechanisms as a function
of carrier particle Reynolds number. These results can be used for deriving the drug
powder detachment model.
(iii) Powder dispersion in Aerolizer ®
Powder dispersion in a commercial Aerolizer ® inhaler model was studied. Ini-

tially, the agglomerate-based formulation was simulated. Figure 21 shows the inhaler
Fig. 22. Impaction of single carrier particle with a wall in a uniform shear flow (Tong et al. 2015)
device and agglomerates used in the study. The inhaler consists of a chamber with
two inlets, a barrel, and a grid in between. A capsule is inside the chamber to store
agglomerated powders (Tong et al. 2013). The simulations showed that the dominant
mechanism was agglomerate-device impaction. The turbulent flow was not strong
enough to break the agglomerates, and the agglomerate-agglomerate impactions
occurred only at the very early stage when the agglomerates were spun out from
the capsule. The performance of the inhaler was very sensitive to the flow condition.
While better dispersion performance was observed with increasing airflow velocity,
larger flow rates also increased the amount of powder deposition and exceedingly
high flow rate actually reduced the performance of the inhaler. Thus the flow rate
should be controlled for optimum dispersion efficiency.
The carrier-based DPI formulations enable the fine drug particles being detached
from the carrier upon inhalation. However, simulating such a system has a very
computational cost due to the large number of particles involved but also the large
carrier-particle size ratio. Therefore, a multi-scale approach was proposed (Tong
et al. 2015). This approach was based on an assumption that the powder dispersion in
an inhaler depends on the detachment of fine drug powders from carriers and energy
distribution inside the inhaler. Figure 22 shows the detachment of fine powders from
the carrier impaction. The detachment behavior of particles depended on the char-
acteristics and material properties of carrier and drug particles. By linking such
information with the collision energy distribution inside the inhaler (Fig. 23), the
overall aerosolization performance of the inhaler was predicted. The results showed
22 Z.B. Tong et al.
Fig. 23. Spatial distribution of total particle-wall (left) and particle-particle (right) collision energy
(the colors represent magnitude (mJ) of collision energy) (Tong et al. 2015)
that the dispersion performance of carrier-based formulation decreased with the

carrier particle size and increased with the air flow velocity.
Conclusion and Perspectives
The numerical studies of DPI have demonstrated that the approach is an effective
tool in analyzing and predicting the performance of various inhalation devices and
formulations. The numerical results can generate information leading to a better
understanding of the internal flow structure of inhaler systems by analyzing the flow
patterns of particle phase at particle scale and fluid phase at computational cell scale
and particle-fluid, particle-particle, and particle-wall interaction forces. The results
can be used to improve the performance of existing DPIs or designing new inhalers
with desirable dispersion. It also has a wide applicability to a variety of inhalers and
operational conditions.
Future effort should be made in the following three aspects:
(i) Model development: Theoretical development is the foundation and guarantees

the rationality of the model. One major area is to develop a more comprehen-
sive model (e.g., LES) to study the complicated situation.
(ii) Computational performance: the accuracy, robustness, versatility, and speed
(e.g., GPU) (Stone et al. 2010), which are normally the key factors to judge
numerical techniques, can be improved in the future.
(iii) Application: The findings on the role and relative importance of particle-
device, particle-particle, and particle-air interactions can be applied to devel-
oping better design of inhalers for improved de-agglomeration.
The new designs will be evaluated by measuring the powder dispersion per-
formance and compared with the model prediction.
Acknowledgments Authors are grateful to the Australian Research Council (ARC) for the
financial support.
References
T.B. Anderson, R. Jackson, Ind. Eng. Chem. Fundam. 6, 4 (1967)
J.H. Bell, P.S. Hartley, J.S.G. Cox, J. Pharm. Sci. 10 (1971)
R. Boerefijn, Z. Ning, M. Ghadiri, Int. J. Pharm. 172, 199 (1998)
G. Calvert, M. Ghadiri, R. Tweedie, Adv. Powder Technol. 20, 1 (2009)
G. Calvert, A. Hassanpour, M. Ghadiri, Chem. Eng. Res. Des. 89, 5 (2011)
G. Calvert, A. Hassanpour, M. Ghadiri, Chem. Eng. Sci. 86, 146 (2013)
H.-K. Chan, Colloids Surf. A: Physicochem. Eng. Asp., 284–285 (2006a)
H.K. Chan, J. Aerosol Med. 19, 1 (2006b)
K.W. Chu, B. Wang, A.B. Yu, A. Vince, Powder Technol. 193, 3 (2009)
M.S. Coates, D.F. Fletcher, H.K. Chan, J.A. Raper, J. Pharm. Sci. 93, 11 (2004)
M. Coates, H.-K. Chan, D. Fletcher, J. Raper, Pharm. Res. 22, 6 (2005)
M.S. Coates, H.K. Chan, D.F. Fletcher, J.A. Raper, J. Pharm. Sci. 95, 6 (2006)
M.S. Coates, H.K. Chan, D.F. Fletcher, H. Chiou, Pharm. Res. 24, 8 (2007)
Y. Cui, S. Schmalfuß, S. Zellnitz, M. Sommerfeld, N. Urbanetz, Int. J. Pharm. 470, 1–2 (2014)
P.A.O.D. Cundall, L. Strack, Geotechnique 29, 47 (1979)
A.H. de Boer, P. Hagedoorn, R. Woolhouse, E. Wynn, J. Pharm. Pharmacol. 64, 9 (2012)
M.J. Donovan, S.H. Kim, V. Raman, H.D. Smyth, J. Pharm. Sci. 101, 3 (2012)
C.A. Dunbar, A.J. Hickey, P. Holder, Kona 16, 7 (1998)
W.H. Finlay, The Mechanics of Inhaled Pharmaceutical Aerosols, An Introduction (Academic,
London, 2001), p. 134
D.L. French, D.A. Edwards, R.W. Niven, J. Aerosol Sci. 27, 5 (1996)
H.W. Frijlink, A.H. De Boer, Expert Opin. Drug Deliv. 1, 1 (2004)
M.S. Genç, Ü. Kaynak, G.D. Lock, P I Mech. Eng. G-J Aer. 223, 3 (2009)
I. Gonda, Physicochemical Principles in Aerosol Delivery. Topics in Pharmaceutical Sciences 1991
(Medpharm Sci Publ, Stuttgart, 1992), p. 95
K. Higashitani, K. Iimura, H. Sanda, Chem. Eng. Sci. 56, 9 (2001)
K. Iimura, K. Higashitani, Adv. Powder Technol. 16, 1 (2005)
K. Iimura, H. Nakagawa, K. Higashitani, Adv. Powder Technol. 9, 4 (1998)
K. Iimura, M. Suzuki, M. Hirota, K. Higashitani, Adv. Powder Technol. 20, 2 (2009a)
24 Z.B. Tong et al.
K. Iimura, M. Yanagiuchi, M. Suzuki, M. Hirota, K. Higashitani, Adv. Powder Technol. 20,

6 (2009b)
B.E. Launder, G.J. Reece, W. Rodi, J. Fluid Mech. 68, 537 (1975)
W.-I. Li, M. Perzl, J. Heyder, R. Langer, J.D. Brain, K.H. Englmeier, R.W. Niven, D.A. Edwards,
J. Aerosol Sci. 27, 8 (1996)
E.A. Matida, W.H. DeHaan, W.H. Finlay, C.F. Lange, Aerosol Sci. Technol. 37, 11 (2003)
E.A. Matida, W.H. Finlay, A. Rimkus, B. Grgic, C.F. Lange, J. Aerosol Sci. 35, 7 (2004)
J. Milenkovic, A.H. Alexopoulos, C. Kiparissides. Int. J. Pharm. 461, 1–2 (2014)
J. Milenkovic, A.H. Alexopoulos, C. Kiparissides, Int. J. Pharm. 448, 1 (2013)
R. Moreno, M. Ghadiri, Chem. Eng. Sci. 61, 8 (2006)
R. Moreno, M. Ghadiri, S.J. Antony, Powder Technol. 130, 1–3 (2003)
D. Nguyen, A. Rasmuson, K. Thalberg, I. Niklasson Bjo¨rn, Chem. Eng. Sci. 116, 91 (2014)
Z. Ning, R. Boerefijn, M. Ghadiri, C. Thornton, Adv. Powder Technol. 8, 1 (1997)
J. Patton, Nat. Biotechnol. 16, 2 (1998)
H. Rumpf, in The Strength of Granules and Agglomerates. Agglomeration, ed. by W. A. Knepper.
(Interscience, New York, 1962), p. 379
C.A. Ruzycki, E. Javaheri, W.H. Finlay, Expert Opin Drug Del 10, 3 (2013)
J.E. Stone, D.J. Hardy, I.S. Ufimtsev, K. Schulten, J. Mol. Graph. Model. 29, 2 (2010)
J. Subero, Z. Ning, M. Ghadiri, C. Thornton, Powder Technol. 105, 1–3 (1999)
C. Thornton, K.K. Yin, M.J. Adams, J. Phys. D. Appl. Phys. 29, 2 (1996)
C. Thornton, M.T. Ciomocos, M.J. Adams, Powder Technol. 105, 1–3 (1999)
Z.B. Tong, R.Y. Yang, A.B. Yu, S. Adi, H.K. Chan, Powder Technol. 196, 2 (2009)
Z.B. Tong, R.Y. Yang, K.W. Chu, A.B. Yu, S. Adi, H.K. Chan, Chem. Eng. J. 164, 2–3 (2010)
Z.B. Tong, S. Adi, R.Y. Yang, H.K. Chan, A.B. Yu, J. Aerosol Sci. 42, 11 (2011)
Z.B. Tong, B. Zheng, R.Y. Yang, A.B. Yu, H.K. Chan, Powder Technol. 240, 19 (2013)
Z. Tong, H. Kamiya, A. Yu, H.-K. Chan, R. Yang, Pharm. Res. 32, 2086–2096 (2015)
Z. Tong, A. Yu, H.K. Chan, R. Yang, Curr. Pharm. Des. 21, 27 (2015)
Y. Tsuji, T. Tanaka, T. Ishida, Powder Technol. 71, 3 (1992)
A. Voss, W.H. Finlay, Int. J. Pharm. 248, 1–2 (2002)
D.K. Walters, J.H. Leylek, J. Turbomach. 126, 1 (2004)
B. Wang, D.L. Xu, K.W. Chu, A.B. Yu, Appl. Math. Model. 30, 11 (2006)
W. Wong, D.F. Fletcher, D. Traini, H.K. Chan, P.M. Young, Adv. Drug Deliv. Rev. 64, 4 (2012)
B.H. Xu, A.B. Yu, Chem. Eng. Sci. 52, 16 (1997)
B.H. Xu, Y.Q. Feng, A.B. Yu, S.J. Chew, P. Zulli, Powder Handl. Process. 13, 71 (2001)
R.Y. Yang, A.B. Yu, S.K. Choi, M.S. Coates, H.K. Chan, Powder Technol. 184, 1 (2008)
J. Yang, C.-Y. Wu, M. Adams, Granul. Matter 15, 4 (2013)
J. Yang, C.-Y. Wu, M. Adams, Acta Pharm. Sin. B 4, 1 (2014)
J. Yang, C.-Y. Wu, M. Adams, Particuology. 23, 25–30 (2015)
Q.T. Zhou, Z.B. Tong, P. Tang, M. Citterio, R.Y. Yang, H.K. Chan, AAPS J. 15, 2 (2013)
Q. Zhou, P. Tang, S.S.Y. Leung, J.G.Y. Chan, H.-K. Chan, Adv. Drug Deliv. Rev. 75, 141 (2014)
H.P. Zhu, Z.Y. Zhou, R.Y. Yang, A.B. Yu, Chem. Eng. Sci. 62, 13 (2007)
H.P. Zhu, Z.Y. Zhou, R.Y. Yang, A.B. Yu, Chem. Eng. Sci. 63, 23 (2008)

Handbook of Multiphase Flow Science and Technology (Guan Heng Yeoh (Eds.) )

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Handbook of Multiphase Flow Science and Technology (Guan Heng Yeoh (Eds.) )

Uploaded by

Copyright:

Available Formats

Basic Theory and Conceptual Framework

Guan Heng Yeoh and Jiyuan Tu

G.H. Yeoh (*)

# Springer Nature Singapore Pte Ltd. 2016 1

Three-phase gas-liquid-solid ﬂows are encountered in a number of engineering

Definitions of Basic Terms

of which the material density, in terms of a limit, is deﬁned as

It should be noted that properties such as mixture viscosity or thermal conduc-

Multi-Scale Consideration of Multiphase Flows

Within the conceptual framework of computational multiphase ﬂuid dynamics, the

The concept of Lagrangian tracking entails following the motion of individual

Micro-Scale Meso-Scale Macro-Scale

Interactions of Formation of Hydrodynamic

Motion of Local Structural Large Scale

Fig. 1 Multiscale consideration of multiphase ﬂows (Yeoh et al. 2014)

stems primarily through the formulation of appropriate particle interaction laws as

Models for Particulate-Particulate Interaction

• Particulates are generally taken to be spherical and quasirigid

These assumptions are believed to be sufﬁciently realistic for collisions of

By deﬁnition, I ¼ mr 2gyration where rgyration is the radius of gyration of the

The relative velocity of particulate centroids Gd , kl is given by

Gd,kl ¼ Ud,k Ud,l (24)

J t, sliding ¼ μf J n (28)

On the other hand, if the friction coefﬁcient is sufﬁciently high,

with the relative velocity of each particulate given by

The particulate deformation during contact can be characterized by the presence

FCd ¼ Fnd þ Ftd (38)

Classification of Particulate-Particulate Forces

For the soft sphere model, the fundamentals of particulate-particulate interactions

Fig. 5 Classiﬁcation of different levels of particulate dynamics: continuum mechanics, micro-

These microscopic simulations of a small sample can be utilized to derive macro-

Contact Forces at the Level of Micromechanics

• Elastic contact deformation which is reversible, independent of deformation rate

Table 1 Literatures on elastic, plastic, viscoelastic, and viscoplastic contact deformation

• Plastic contact deformation which is irreversible, deformation rate and consoli-

Normal Impact Shearing Twisting Rolling

Fig. 6 Different modes of particulate-particulate interactions

• Viscoplastic contact deformation which is irreversible and dependent on defor-

• Along the line normal to the particulate centers

known as the nonlinear spring-dashpot model, can also be combined to include

Contact Forces at the Level of Molecular Dynamics

• van der Waals forces

Material bridges between particulate surfaces

• Organic macromolecules such as ﬂocculants in suspensions

Interlocking by macromolecular and particulate shape effects

• Interlocking of chain branches at macromolecules such as proteins

Van der Walls Electrostatic Magnetic Interfacial

Interlocking by Hook-Like Bonds

Low High Recrystallization of

Organic Liquid Bridge Solid Bridge

Fig. 7 Particulate adhesion via different bonding effects

• Interlocking of contacts by overlaps of surface roughness

Classification of Fluid-Particulate Forces

The Stokes number can be deﬁned as

This force originates because of the difference in acceleration and becomes

where αL = |Ωf|dp/(2|Uf Ud|) and Ωf = ∇ Uf. This force becomes prevalent

For the pressure gradient force, it can be expressed as

For submicron particulates, a random force as a result from Brownian motion

where σ is the Stefan-Boltzmann constant and Tf is the ﬂuid temperature. The

evolutionary equations can be prohibitive due to the uncertainty of the particular

Interpenetrating Media Framework

In the interpenetrating media framework, the averaging process allows computations

where T is an averaging time scale, V is the volume based on an averaging length

1 2=3 σ 1 1=3 2=3 σ