Professional Documents
Culture Documents
Challenges and Issues On The CFD Modeling of Fluidized Beds A Review
Challenges and Issues On The CFD Modeling of Fluidized Beds A Review
Abstract
We first describe the main approaches used to model fluidized suspensions. Focusing
on the multifluid approach, we overview the principal averaging techniques that con-
sent to turn granular systems into continua; in particular, we discuss volume, ensem-
ble and time averages. We then use volume averages to derive the Eulerian equations
of motion for fluidized suspensions of a finite number of monodisperse particle class-
es. We introduce the closure problem, and overview some widely adopted closure equa-
tions used to express the granular stress and the interaction forces between the phases,
giving emphasis to the fluid-particle interaction force, in particular to the buoyancy and
drag contributions. We conclude the work by discussing some published CFD simula-
tions of mono and bidisperse fluidized beds, spanning different fluidization regimes
and commenting on the insight that these studies provide.
1. INTRODUCTION
Fluidization is a well-established technology used in several industrial processes such as coal com-
bustion, biomass gasification, waste disposal and food processing. To design fluidized beds, engi-
neers have resorted for many years to experimental correlations and pilot plants. Nonetheless, since
these correlations are valid only for the specific units investigated, they cannot help us to improve
design and performance; for instance, they can tell us nothing about the effect of changing geom-
etry, introducing internals or repartitioning the feeds over more entry points. Pilot plants, on the
other hand, are costly and time-consuming; moreover, they do not always lead to adequate scale up.
When fluidized beds were first employed in the 1920s -1940s, engineers did not appreciate this
problem, most probably because at the time the required plant performance was either not critical
(as in FCC plants) or easily achievable (as in roasting and drying). Nevertheless, when later on 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 mat-
ter had to be addressed thoroughly. Hence, researchers endeavored to find more reliable methods
to predict the dynamics of fluidized suspensions. In the 1960s, chemical engineers started to use
the conservation laws of mass, momentum and energy to analyze nearly any physical and chemi-
cal problem. This new approach, probably also fostered by the release of the first edition of the
influential textbook Transport Phenomena (Bird et al., 1960), led to significant theoretical head-
way, bolstered the hope to explain theoretically the behavior of fluidized beds and prompted the
first trials to develop fluid dynamic models based on conservation equations.
Anderson & Jackson (1967) were among the first to model fluidized systems. Starting from the
transport equations of single-phase fluid dynamics and the Newtonian equations of rigid-body
motion, they derived averaged equations of conservation for the fluid and solid phases by applying
a formal mathematical process of volume averaging. Afterwards, several other researchers did the
same, see for instance Whitaker (1969), Drew (1971) and Drew & Segel (1971). Initially, they used
these models to better understand the complex behavior of multiphase systems, never considering
them as a viable alternative to design real process units. But successively, when faster computer
processors and advanced numerical methods to integrate partial differential equations became
available, they realized that a mathematical theory of multiphase flows might provide a useful
design tool for industrial applications.
84 Challenges and Issues on the CFD Modeling of Fluidized Beds: a Review
With the further development of new and more rigorous formulations of multiphase equations
of motion (Buyevich, 1971; Hinch, 1977; Nigmatulin, 1979; Drew, 1983; Jenkins & Savage, 1983),
the late 1970s and early 1980s witnessed the first endeavors to describe qualitatively, by direct
numerical simulation, granular systems in flow (Pritchett et al., 1978; Gidaspow & Ettehadieh,
1983; Gidaspow et al., 1986). The promising results of these pioneering studies fueled an increas-
ing interest in computational fluid dynamics (CFD) and multiphase flows, which rapidly turned
into research areas in their own right. Nowadays, CFD has become an almost indispensable tool to
solve many complex academic and industrial problems. In fluidization, CFD has helped to under-
stand fluid-solid interactions and has enabled to predict various macroscopic phenomena encoun-
tered in particulate systems. Similarly, multiphase flows and, more specifically, fluidization dynam-
ics now interest not only the scientific community but also the industrial world. Over the years,
researchers have developed mathematical models to predict the dynamics of dense fluidized beds.
Several approaches and mathematical techniques have been employed; in the next sections, we
overview them and highlight their respective advantages and limitations.
Lagrangian simulations of a fluidized bed at the finite Reynolds numbers encountered in practical
applications. Numerical bed expansion profiles are processed to give straight lines in log-log plots
leading to power laws as in the well-known experimental correlation of Richardson & Zaki (1954);
furthermore, for the first time ever, the method allows to calculate directly the slip velocity and
other average quantities used in multifluid, continuous models.
Despite its potential, because of the severe computational effort that it requires, the Eulerian-
Lagrangian approach is seldom applied; also, from an industrial perspective, this strategy is not the
most effective. In the first place, the computational cost is way too demanding. Simulations so
detailed have been performed only for much diluted systems containing a truly small number of
particles (1204 in the work of Pan et al., 2002); extending these calculations to dense suspensions
like those characterizing industrial fluidized beds is presently inconceivable. Moreover, even if this
were feasible, the amount of information provided by the solution would be far too detailed, and in
any case a method of filtering or averaging would be required to elicit the useful results in which
we are really interested.
These observations suggest that it might be advantageous to formulate transport equations gov-
erning the evolution, in time and physical space, of these averages directly. In this alternative
approach, rather than aiming at the detailed solution described above, we are satisfied with a much
reduced description of the flow. Albeit there is no guarantee that these simplified equations can be
really found – in closed form, that is – several studies have been conducted in this pursuit
(Anderson & Jackson, 1967; Whitaker, 1969; Drew, 1971; Drew & Segel, 1971; Drew, 1983; Drew
& Lahey, 1993; Gidaspow, 1994; Zhang & Prosperetti, 1994; Enwald et al., 1996; Jackson, 1997;
Zhang & Prosperetti, 1997; Jackson, 1998). Owing to the complexity of the problem, we do not
expect, at least for the time being, that the exact averaged equations of motion that govern multi-
phase flows should be derived. The intent is far more practical and aspires to formulate relations
that should be good enough to describe satisfactorily the phenomena of interest.
Different mathematical techniques yield such equations, and several claims have been advanced
as to the superiority of each form of averaging versus the others. Regardless of the specific math-
ematical scheme adopted, 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 aver-
aged variables defined over the whole physical domain; thus, the resulting averaged equations
resemble those that one would write for n imaginary fluids capable of interpenetrating each other
while occupying simultaneously the same volume. The model, referred to as Eulerian-Eulerian or
also as multifluid, takes therefore the form of coupled partial differential equations subjected to ini-
tial and/or boundary conditions assigned only on the system boundaries. Second, the process of
averaging leaves behind a number of indeterminate terms not directly related to the averaged vari-
ables but still associated with details of the motion at the microscopic length scale. These are rep-
resented by the fluid and solid stress tensors, and by the interaction force exchanged by the phas-
es. A closure problem therefore arises, which usually cannot be solved analytically and has to be
overcome by means of empirical or semiempirical expressions. This is indeed the main shortcom-
ing of the method.
Many researchers have successfully simulated monodisperse fluidized beds. Ding & Gidaspow
(1990), for instance, investigated the bubble dynamics of Group B particles (Geldart, 1973), vali-
dating the results qualitatively. van Wachem et al. (1998) simulated bubbling gas-fluidized beds of
Group B particles, using columns of various diameters and different superficial velocities, and val-
idating their work both qualitatively and quantitatively. Peirano et al. (2001) studied the effect of
the dimensionality of the computational domain on the simulated fluidization dynamics, pointing
out that two-dimensional simulations should be used with caution and only for sensitivity analyses.
Pain et al. (2001) simulated bubbling and slugging gas-fluidized beds, analyzing the effect of an
obstruction on the fluidization quality. Gelderbloom et al. (2003) studied the bubbling and collaps-
ing of Group A, B and C powders, and validated the computed bubble sizes using the Davidson &
Harrison (1963) empirical relation. Lettieri et al. (2003, 2004) used different Eulerian models to
simulate the transition from bubbling to slugging and turbulent gas fluidization of Group B pow-
ders. More recently, Owoyemi et al. (2005) have simulated the fluid dynamics of industrial Group
B powders, whereas Mazzei & Lettieri (2008) of expanding and contracting homogeneous flu-
idized beds and their transition to the bubbling regime, validating in both studies the numerical
results with experimental data.
In addition to the modeling approaches just described, there is a third that can be regarded as a
hybrid between them. Averaged equations of motion are derived for the fluid phase, but Newtonian
equations for rigid-body motion are solved for each particle of the suspension. These do not inter-
act with the fluid via its microscopic velocity field, but with the averaged value of the latter. For
example, the overall force exerted by the fluid on each particle is not computed by integrating over
the particle surface the local traction that arises from the fluid velocity gradients; the force is
instead evaluated in terms of slip velocity between the average fluid velocity and the velocity of
the particle center of mass and by resorting to semiempirical correlations. The procedure, known
as discrete particle modeling (DPM), is considerably less demanding computationally than the pure
Eulerian-Lagrangian and has met with resounding success.
To describe particle collisions, modelers use two types of approaches: hard-sphere and soft-
sphere. In the first, particles interact via binary, quasi-instantaneous, pointwise collisions, which
preserve the overall linear and angular momentum of the system. This technique was pioneered by
Allen & Tildesley (1990); since their publication, many authors have adopted it to model the colli-
sion dynamics in granular systems (Tsuji et al., 1987; Hoomans et al., 1996; Ouyang & Li, 1999;
Lu et al., 2005). The soft-sphere model for fluidized beds was instead pioneered by Tsuji et al.
(1993), who based it on the earlier work of Cundall & Strack (1979). This approach allows the par-
ticles to slightly overlap during the collisions and computes the contact forces taking into account
the deformation history of the contact and using a linear spring/dashpot model (Xu & Yu, 1997;
Pandit et al., 2005; Ye et al., 2005).
Between the three modeling approaches discussed above – Eulerian-Lagrangian, Eulerian-
Eulerian and discrete particle modeling – the second is often favored because is computationally
less demanding. Owing to the vast number of particles involved in industrial plants, continuum
modeling is unlikely to be replaced by discrete modeling in the foreseeable future. Yet, the latter
plays an important role. To be considered more an effective research tool than a practical design
instrument, by providing information about the microscopic dynamics of multiphase systems, it can
significantly help to develop and improve continuous, average-based, macroscopic models. This
multiscale modeling strategy, clearly described by van der Hoef et al. (2005b), is represented
schematically in Figure 1. However, how to effectively link the different models and extract from
each the information needed by the others is still an open challenge.
Particle-Particle
Discrete Particle Modeling
Interaction Force Closure Laws
Volume averages are computed over spatial domains large enough to contain several particles
but small compared with the scale of point-to-point macroscopic variations in which the modelers
are interested and which they wish to capture. A volume V(x) bounded by a surface S(x) is attached
to each spatial point x; within such a volume, the property of interest is averaged out by using the
mean value theorem of integral calculus. When this averaging scheme is adopted, all the property
values within V(x) are accounted for and given the same weight in the average; conversely, the val-
ues outside V(x) are ignored, that is, are given zero weight. These averages are usually referred to
as hard volume averages to distinguish them from soft volume averages. The latter are based on an
alternative technique, more elegant and convenient from a mathematical standpoint, that uses a
radial weighting function: a continuous, monotone, decreasing function of the radial distance from
the point wherein the average is evaluated. This mathematical device ascribes a weight to the prop-
erty values within the whole physical domain. As a consequence, the average is no longer calcu-
lated by considering only the values attached to a well-defined averaging volume; on the contrary,
all are taken in consideration, each one suitably weighted. For a formal and more detailed discus-
sion concerning hard and soft volume averages, we refer to §5.
Volume-averaged variables thus defined appear to depend on the specific form chosen for the
weighting function and, in particular, on its radius (that is, the size of the region over which the
averaging effectively takes place; for a formal definition of weighting function radius we refer to
§5). Nonetheless, the larger the ratio between the smallest macroscopic length scale and the parti-
cle size, the more this dependence dwindles provided that the weighting function radius is proper-
ly chosen. If this radius is denoted by ra, the particle radius by rp and the aforementioned macro-
scopic length scale by rc, the local average is expected to be insensitive to the particular form of
the weighting function provided that rp << ra << rc. Such a choice of ra is, of course, possible only
if rc is far greater than rp; in such a case, there is said to be separation of scales between the macro-
scopic fluid dynamic problem and the detailed motion on the scale of a single particle. Only in this
instance the locally averaged variables do possess an unambiguous physical meaning and can be
employed in the development of averaged equations of motion.
Ensemble averages are based on a completely different idea related to a fundamental character-
istic of particulate systems: it is impossible to know the precise locations and properties (e.g., lin-
ear and angular velocities) of each individual constituent of the particle population at any given
time. In fact, this sort of details are not only impractical to measure but, most importantly, are usu-
ally entirely inconsequential. Often, more gross features of the motion are of interest. Let us con-
sider, for instance, two particulate systems subjected to the same boundary and initial conditions,
with the exception of a few undefined properties at the particle level such as the exact position ini-
tially occupied by each discrete element. Albeit the evolution of the two systems will be different
from a microscopic standpoint – reflecting the fact that the two flows originate from dissimilar con-
ditions at the microscopic level – they are macroscopically equivalent, since the evolution of all the
macroscopic variables characterizing the systems is identical. Note that if this were not the case,
repeatability of physical experiments would be impossible, for it is clearly impossible to set up
intentionally the same microscopic initial conditions twice.
The set of infinite systems sharing the same boundary and initial conditions in the sense speci-
fied above is said to constitute an ensemble, with each individual system representing a realization
of the latter. Each realization is unique, insomuch as it differs microscopically from all the others;
all realizations, however, are macroscopically equivalent, since they feature identical evolutions of
all the macroscopic measurable variables. Such ensembles are reasonable sets over which to per-
form averages because variations in the flow details are always assured, whereas variations in the
gross features of the flows cannot occur.
At any given time, the ensemble average of a generic property at a specified spatial location x
can be defined as the arithmetic mean of the instantaneous point property of the material located at
x at the time in question for each of the infinite number of realizations of the ensemble. Hence,
following for instance Kleinstreuer (2003), for a given field ξ (x,t) it is:
(4.1)
where 〈ξ〉e (x,t) is the ensemble average, N is the number of times the value of ξ (x,t) is sampled
and ξn (x,t) is the functional form of the field at the n-th realization of the flow. An alternative
technique for representing the ensemble average of the field ξ (x,t) is:
(4.2)
where µ denotes a generic realization of the ensemble, ε represents the ensemble of all possible
realizations, ξ (x,t;µ) expresses the dependence of the field on the realization and m(µ) quantifies,
loosely speaking, the probability of occurrence of each realization (Drew & Passman, 1998; Crowe,
2005).
To explain time averaging, we consider again the field ξ (x,t); for any spatial fixed position x*,
ξ (x*,t) is a pure function of time that fluctuates irregularly. The time scale τt that represents these
fluctuations is usually referred to as turbulent time scale. At each given location, we can obtain a
mean value of ξ (x,t) by time averaging over a large number of fluctuations, considering a time
interval τa much longer than the turbulent time scale. Again, we evaluate the average through the
mean value theorem:
(4.3)
where 〈ξ〉t (x,t) is the time average and τ is a dummy integration variable. Provided the averaging
time scale is sufficiently long, the turbulent fluctuations are entirely smoothed out. Also now, the
mean value is expected to be insensitive to the averaging time scale provided that τt << τa << τm,
where τm represents the time scale of the mean flow fluctuations (Delhaye & Achard, 1977, 1978).
For fluidized suspensions, volume, ensemble and time averages are not equivalent; this is
because the ergodicity principle in general is not satisfied. We remind that a system is said to be
ergodic if the values of the ensemble-averaged variables, at any point in space and time, are the
same as those of the respective volume and time-averaged variables obtained from a single real-
ization of the system randomly chosen from the ensemble. In our setting, this property is not always
met; to this end, one example provided by Jackson (2000) can be enlightening. Let us consider a
bubbling fluidized bed, and let us focus on one specific mean variable: the bulk density. This prop-
erty, if obtained by volume averaging, represents the mass to volume ratio for a blob of mixture
enclosed in an averaging volume sufficiently big to contain several particles but small compared
with the macroscopic length scale characterizing the flow. Consequently, the property is sometimes
representative of the mean density of the dense suspension and some other times of the mean den-
sity of a bubble, depending on which of the two, at the time in question, occupies the spatial point
in which the average is computed. Conversely, if obtained by ensemble averaging, the property
results to be equal at all times to a value intermediate between the bulk densities of a bubble and
of the dense suspension.
This ensues directly from the definition of ensemble average given above. The two averages are
obviously quite different, and the one more appropriate to investigate the dynamics of fluidized sys-
tems is clearly the former. In what follows, we adopt volume averages for these are closer to the
physical properties that are experimentally measurable and in which we are usually interested.
To define locally averaged variables, we use a weighting function; this is characterized by the
following mathematical properties:
1) The weighting function is a scalar function of r defined for r > 0, where r denotes the dis-
tance of a generic point y from a generic point x in a three-dimensional Euclidean space:
(5.1)
(5.2)
where dVy indicates that the spatial variable of integration is y and not x – the latter representing
the spatial position wherein the average is computed.
The averaging radius of ψ (r) is defined to be the scalar ra that fulfills the integral relation:
(5.3)
Thus, ra is a measure of the linear size of the spherical neighborhood of x wherein the spatial
points have appreciable weight in the averaging process.
(5.4)
To explain the physical meaning of this definition, we consider the function ψ (r) = 1/Ωa for r
ra and ψ (r) = 0 for r > ra, where Ωa is the volume of a sphere of radius ra. This function is not
smooth and does not decrease monotonically with r as required; we only use it to render the phys-
ical meaning of equation (5.4) more apparent. With this choice, equation (5.4) yields:
(5.5)
This is the mean value theorem written over the averaging volume Ωa. This expression defines hard
spatial averages; soft spatial averages, based on continuous weighting functions meeting the
requirements 1) to 3), are more convenient for they allow us to use all the theorems of analytical
calculus.
(5.6)
The fluid-phase average of a point field ξ (x,t) calculated in x at time t is then defined to be:
(5.7)
In equations (5.6) and (5.7), Ωf represents the volume occupied by the fluid phase at time t (we
have left out the explicit dependence on t to simplify the notation).
We can obtain the fluid-phase averages of spatial and time derivatives of point variables by gen-
eralizing the relationships originally derived by Anderson & Jackson (1967) for monodisperse sus-
pensions; they are given by (for further details, refer to Mazzei, 2008):
(5.8)
(5.9)
In these equations, the integrals are taken over the particle surfaces ∂Ωp, the vectors n (x,t) and u (x,t)
are the outward unit normal to ∂Ωp and the fluid velocity, respectively, and nk (x,t) is the k-th com-
ponent of n (x,t) with respect to a generic orthonormal vector basis. Moreover, i is the generic
particle class of the bidisperse fluidized suspension. In what follows, we denote by vi the physical
volume of the generic particle of class i and by Ωp the region of Euclidean space occupied by the
generic particle p, whichever the class, at the time of interest. Thus, vi is a property of the particles
of phase i and is time-independent (because the particles are rigid); conversely, Ωp is a time-
dependent domain of integration bounded by the oriented surface ∂Ωp.
(5.10)
where the summation is intended over all the particles of phase i and where the integral is taken
over the volume Ωp occupied at time t by a generic particle of the set.
We can also define solid-phase averages of point variables for each solid phase i of the sys-
tem; these are analogous to the fluid-phase averages, and take the form:
(5.11)
This average, employed by several researchers (Enwald et al., 1996; Crowe et al., 1997; Drew &
Passman, 1998; Fan & Zhu, 1998), operates on the microscopic properties of the particle material,
considering point fields ξ (x,t) that vary within the particles. It is an average that exactly parallels
the one just given for the fluid. This approach, however, is not always straightforward and presents
conceptual difficulties when the particles are rigid (as in the present treatment) or massless (for
instance, in bubble columns); in these cases, artifices are necessary to overcome these difficulties
(Batchelor, 1970; Hinch, 1977; Sangani & Didwania, 1993; Zhang & Prosperetti, 1994). Thus, to
derive the averaged transport equations for the particles, we do not use this average, but favor an
other based on properties ξp(t) of the particles as a whole, as opposed to the point properties ξ (x,t)
varying within the particles used in equation (5.11).
(5.12)
where xp(t) denotes the position occupied at time t by the center of mass of a generic particle of
the phase under consideration. The volume fraction φi (x,t) is related to the number density ni (x,t)
by the following relationship (Jackson, 1997; Mazzei, 2008):
(5.13)
where the last term on the right-hand side can be safely neglected if there is separation of scale
between the macroscopic motion of the suspension and the microscopic motion of the particles.
Generalizing the averaging scheme advanced by Jackson (1997), we define the particle-phase
average for the phase i of a particle property ξp(t) calculated in x at time t as:
(5.14)
The particle phase average of a particle property time derivative is given by:
(5.15)
where 〈ξvk〉ip(x,t) is the average of the product of the particle property ξp(t) and the k-th compo-
nent vkp (t) of the velocity of the center of a generic particle of the phase under consideration. Here
and in what follows, we adopt the convention that repeated lower indices are summed over the val-
ues one to three, with the exception of i, used as phase index, and of f and p, used to specify the
type of average; upper indices are not dummy indices and do not imply summation. Until now, we
have stated explicitly the dependence on time and position; from now on, to simplify the notation,
we will often leave it out.
Continuity equation
The equation is obtained by setting ξ = uk and ξ = 1in equations (5.8) and (5.9), respectively, and
then by adding the results. Since the fluid is incompressible, doing so yields:
(5.16)
Dynamical equation
At the microscopic level, the fluid satisfies the point Navier-Stokes differential equation:
(5.17)
where ρf is the fluid density, Tjk(x,t) is the jk-th component of the point fluid stress tensor and gj
is the j-th component of the gravitational field. To average this equation, we multiply both sides by
ψ(x – y) and integrate over Ωf with respect to y. To treat the left-hand side of equation (5.17),
we use equations (5.8) and (5.9) with ξ = ujuk and ξ = uj, respectively; to treat the right-hand side,
we use equation (5.6) and equation (5.8) with ξ = Tjk. The final result is:
(5.18)
The last term on the right-hand side is the sum over the two particle classes of the mean resultant
traction force exerted by the fluid on the particles of each class. If we expand ψ(x – y) in a Taylor
series about the center xp(t) of the generic particle of phase i, truncate the series at the third-order
term (the reason why it is necessary to retain also this term is clearly explained by Jackson, 1997)
and introduce the result in the integral on the right-hand side of equation (5.18), we obtain (for fur-
ther mathematical details we refer to Jackson, 1997, Owoyemi et al., 2007 and Mazzei, 2008):
(5.19)
(5.20)
(5.21)
(5.22)
where ri is the radius of the particle of phase i. The three quantities defined above are the com-
ponents of a vector, a second-order tensor and a third-order tensor, respectively. Using the Reynolds
decomposition for the average of the dyadic product on the left-hand side of equation (5.18), we
can write:
(5.23)
where û j (x,t) and û k (x,t) are the deviations of the velocity components uj (x,t) and uk (x,t) from
their respective means 〈uj〉f (x,t) and 〈uk〉f (x,t), respectively. Note that this relation is not exact,
but holds only when point variations can be decomposed into two contributions of scales much
smaller and much larger than the weighting function radius (Anderson & Jackson, 1967). If we
introduce equations (5.19) and (5.23) into equation (5.18), after few manipulations, we obtain:
(5.24)
(5.25)
This term is the fluid-phase effective stress tensor. The quantity 〈û j û k〉f (x,t) is analogous to the
Reynolds stresses of turbulent flows and arises from the fluctuations of the fluid velocity about its
mean value. Closing analytically 〈Sjk 〉f (x,t) is extremely complex, but Jackson (1997) derived a
closure on theoretical grounds in the limiting case of diluted, Stokesian monodisperse mixtures flu-
idized by Newtonian fluids. Appropriate closures for the terms on the right-hand side of equation
(5.24) are discussed in §6.
Continuity equation
We derive the equation by simply setting ξp = 1in equation (5.15); doing so yields:
(5.26)
Dynamical equation
To derive this equation, we start from the linear momentum balance equation written for a generic
particle of species i. The forces at play are the traction force exerted by the fluid on the particle,
the forces that result from the collisions between the particle in question and particles of the same
and of different species at their mutual points of contact and the effect of gravity. Thus, we write:
(5.27)
where ρi is the density of the particles of phase i, v˙pj (t) is the j-th component of the acceleration
of the particle center of mass and fjpq (t) is the j-th component of the force exerted by the generic
particle q of phase k on the particle p under consideration when a collision takes place. Note that
fjpq (t) ≠ 0 only for the particles that are in direct contact with the particle p.
To average equation (5.27), we multiply both sides by ψ(x – xp) and sum over all the parti-
cles that belong to the phase under consideration. Doing so gives:
(5.28)
If we now use equations (5.12), (5.14), (5.15) and (5.20), choosing in the second ξp = v˙pj and in
the third ξp = v pj , and employ the relation:
(5.29)
(5.30)
The first term on the right-hand side is the fluid-particle interaction force, which has its exact ana-
logue in equation (5.24), while the last term combines the resultant forces arising from the particle-
particle contacts between particles belonging to the same phase (k = i) and to different phases (k ≠i).
These contributions are conceptually different: the former is a self-interaction term that represents
the stress internal to the solid phase under examination, whereas the latter is a contact force acting
between the solid phases. For the solid stress tensor associated to phase i and for the particle-par-
ticle interaction force between phases i and k to appear explicitly in equation (5.30), further
mathematical manipulations are required. For these details we refer to Owoyemi et al. (2007) and
Mazzei (2008), who have shown that:
(5.31)
(5.32)
(5.33)
(5.34)
and:
(5.35)
(5.36)
(5.37)
Here xpq(t) is the position vector of the point of mutual contact between the generic rigid particles
p and q and npq(t) is the unit vector of xpq(t) – xp(t). Note that the last three quantities defined
above are not particle averages, because the weighting function is evaluated at the particle surface,
xpq ∈ ∂Ωp and not at its center. If we now introduce equations (5.31) and (5.32) into equation
(5.30), and we apply the Reynolds decomposition to the convective term, we obtain:
(5.38)
(5.39)
Here vˆj (x,t) and vˆm(x,t) are the deviations of the velocity components v pj (x,t) and v m p (x,t) from
The term ni〈fj 〉pik (x,t) is the interaction force between the granular phases i and k. Equation
(5.39) defines the effective solid stress tensor of the phase i ; the first four terms of the right-hand
side constitute the collisional stress, whereas the last contribution, related to the velocity fluctua-
tions, represents the kinetic stress (Chapman & Cowling, 1970; Brilliantov & Poschel, 2004).
Table 1 reports, in absolute notation, the Eulerian-Eulerian averaged equations of motion for
suspensions of n particle classes; these are an immediate generalization of the equations that we
have just derived for bidisperse suspensions. To write them, we have used equation (5.13), retain-
ing only the dominant term of the Taylor series. Notice that in the dynamical equation for the i-th
solid phase, the index k, which appears in the summation on the right-hand side of the equation,
must be different from the index i. This is because the force ni 〈f 〉pi k (x,t) models the interaction
between different particle classes.
ness and fulfillment of the entropy inequality. The first asserts that constitutive equations must be
objective, and must not depend on any specific physical observer (Astarita & Marrucci, 1974;
Truesdell, 1977; Lai et al., 1993; Massaudi, 2002; Gurtin, 2003). The second requires that the fluid
dynamics should be described by a set of equations with unique solution and dependent continu-
ously on their initial and boundary conditions (Drew & Passman, 1998). The third demands the ful-
fillment of the second law of thermodynamics: entropy generation must never be negative
(Astarita, 1989; Sandler, 1989). These principles, of course, cannot tell us what the closure equa-
tions are, but by imposing restrictions on the allowable form of these equations, they tell us what
the closures cannot possibly be.
The problem of closure consists in finding constitutive expressions for the effective stress ten-
sors of the fluid and particle phases and for the average interaction forces between the phases. As
already pointed out, finding analytical closures based on purely theoretical arguments is prohibi-
tively difficult; in fact, there is no guarantee that such equations even exist. The goal here is less
ambitious and aims to find equations that consent to analyze the systems of interest with the desired
accuracy; hence, the problem becomes finding closures that are as simple as possible, whilst still
being able to capture enough physics to describe the dynamics satisfactorily. This is the opposite
of the principle of equipresence, which states that in functional dependencies any possible variable
should be included, unless it can be shown that a particular dependence cannot occur. Although in
theory the idea is correct, the weakness of the method lies in the multitude of parameters present in
these equipresent equations and in the impossibility of measuring them experimentally. Thus, espe-
cially in the framework of multiphase flows, this principle is seldom observed.
In what follows, we will first present some strategies for modeling the effective fluid and solid
stress tensors; we will then analyze the average fluid-particle interaction force, laying emphasis on
the buoyancy and drag forces; finally, we will review some closures for the average interaction
force between particles of different phases. To simplify the analysis, while for the latter contribu-
tion we will consider polydisperse powders, for the others we will focus on monodisperse powders.
mental measurements having a direct bearing on them. Several attempts to investigate the viscous
behavior of particulate mixtures are found in the literature. In these studies, the suspensions are
usually treated as Newtonian pseudofluids, and quantitative measurements of their effective vis-
cosities are provided (Rutgers, 1962; Prudhoe & Whitmore, 1964; Grace, 1970; King et al., 1981;
Reiling, 1992; Poletto & Joseph, 1995; Colafigli et al., 2009). This problem has also been the sub-
ject of various theoretical (Einstein, 1906, 1911; Vand, 1948; Brinkman, 1952; Frankel & Acrivos,
1967; Batchelor & Green, 1972; Graham, 1981) and semitheoretical (Mooney, 1951; Thomas,
1965; Barnea & Mizrahi, 1973; Metzner, 1985; Gibilaro et al., 2007) investigations. However, as
mentioned earlier, all these analyses are concerned with, and provide insight into, the rheology of
the mixtures and not of their Eulerian constituent phases. Relating the two is difficult and research
is still needed.
Jackson (1997, 1998) and Zhang & Prosperetti (1997) derived complete theoretical closures for
dilute monodisperse fluidized suspensions characterized by small Reynolds and Stokes numbers
(for more details concerning the degree of dilution and the definition of Stokes number, we refer to
the cited articles). Jackson adopted volume averages, whereas Zhang & Prosperetti adopted ensem-
ble averages; the results, however, are similar, and therefore we report only Jackson’s. In the limit
considered, the effective solid stress tensor vanishes, whereas the effective fluid stress tensor takes
the form:
(6.1)
where δjk is the Dirac delta, 〈p〉f (x,t) is the average fluid pressure, µf is the fluid shear viscosity,
〈uj 〉(x,t) is the j-th component of the overall average suspension velocity defined as per equation
(5.4), εjrk is the Kronecker permutation symbol and 〈wr 〉p (x,t) is the r-th component of the aver-
age angular velocity of the particles. The first term in equation (6.1) is the jk-th component of the
isotropic effective fluid stress tensor; the second and third are viscous contributions, the third in
particular representing the Einstein (1906, 1911) correction for dilute suspensions; the forth and
fifth involve the average linear and rotational slip velocities between the fluid and particle phases.
The closure above has limited validity, but is insightful. It tells us that, even in the simple case
of low particle concentrations and small Reynolds and Stokes numbers, the deviatoric effective
fluid stress tensor is not simply related to the fluid average rate of deformation tensor (represented
by the third contribution shown in the equation), but depends also on terms related to the average
linear and angular velocity fields of the particle phase, since 〈uj〉 combines both fluid and particle
average velocities, and to the angular velocity associated with the average fluid velocity field. Thus,
at least in principle, to fully characterize the system dynamics we must also solve an averaged bal-
ance equation for the particle angular momentum; for more details about how to derive this equa-
tion, we refer to Jackson (1997).
Despite these considerations, researchers often use very simple closures for the effective stress
tensors, assuming that each phase is Newtonian; thus, for monodisperse systems they usually write:
(6.2)
(6.3)
where 〈p〉p (x,t), µp and κp are the average pressure and shear and dilatational viscosities of the
particle phase, respectively, and κf is the dilatational viscosity of the fluid phase. Moreover, I is the
identity tensor, while 〈D 〉f (x,t) and 〈D 〉p (x,t) are the rate of deformation tensors:
(6.4)
The problem of closure then reduces to finding appropriate constitutive expressions for the pres-
sure, shear viscosity and dilatational viscosity of each phase. The fluid is usually considered incom-
pressible; hence, its pressure does not need to be specified constitutively. Also, often µf is assumed
to be constant, and κf is neglected. For the solid phase, conversely, more elaborate expressions
have been advanced.
We can use various strategies to model the flow parameters featuring in equation (6.3). One is
resorting to empirical relations based on the particle properties and local void fraction. These are
easily implemented in numerical codes and relatively inexpensive computationally. For a detailed
review, refer to Massaudi et al. (1992) and Enwald et al. (1996). Another strategy is using the kinet-
ic theory approach, derived from the mathematical theory of dense non-uniform gases (Chapman
& Cowling, 1970). These models are more complex and numerically demanding, but are more
promising and suitable for a wider range of applications. The idea behind them is that, as fluidized
suspensions resemble in many ways a molecular gas, the transport equations that govern the two
should be similar and derivable, at least partially, from the same theoretical framework. Many
researchers have exploited this analogy for both granular (Haff, 1983; Jenkins & Savage, 1983; Lun
et al., 1984; Jenkins, 1987; Goldshtein & Shapiro, 1995) and fluidized (Koch, 1990; Ding &
Gidaspow, 1990; Buyevich, 1994; Gidaspow, 1994; Buyevich & Kapbasov, 1994; Balzer et al.,
1995, 1996; Buyevich, 1997; Buyevich & Kapbasov, 1999; Koch & Sangani, 1999) materials. The
first are relatively easier to treat since, being characterized by particles with no interstitial fluid,
they differ from molecular gases only because the particle collisions are inelastic; in fluidized sus-
pensions, the problem is complicated by the fluid-particle interactions. Similarly to a molecular
gas, for a granular material particle pressure and viscosity are functions of a granular temperature
Tp(x,t), defined so that Tp ≡ − 1 〈v̂ v̂ 〉 , that is governed by a balance equation for a pseudointer-
3 j j p
nal energy related to the particle velocity fluctuations. For granular materials this equation differs
from the classical internal energy balance equation because of a sink term Sc(x,t) representing
pseudointernal energy losses caused by inelastic collisions; for fluidized suspensions other two
terms appear: a source term Gd(x,t) representing the generation of particle velocity fluctuations by
fluctuating fluid-particle forces, and a sink term Sv(x,t) representing their dampening by the vis-
cous resistance to particle motion. Accordingly, the pseudointernal energy balance equation takes
the form:
(6.5)
where Up(x,t) is the pseudointernal energy per particle unit mass, defined to be Up ≡ − 1 〈v̂ v̂ 〉 ,
2 j j p
and q(x,t) is the pseudothermal heat flux. The closure problem then requires finding constitutive
expressions also for q, Sv, Sc and Gd. The most difficult contribution to close is the latter; for this
reason, this is often neglected (Ding & Gidaspow, 1990; Gidaspow, 1994; Balzer et al., 1995,
1996). Some authors (see, for instance, Koch, 1990; Buyevich, 1994; Buyevich & Kapbasov, 1994;
Buyevich, 1997; Buyevich & Kapbasov, 1999; Koch & Sangani, 1999) have derived theoretical
closures for the terms above in the limit of large Stokes numbers; their expressions, however, are
extremely complex and present some inconsistencies (Jackson, 2000). Finding a general solution
to the problem is still an open challenge, which becomes more complex when more than one par-
ticle phase is involved (Gidaspow, 1994).
We conclude by pointing out that the expression (6.3) and the granular kinetic theory model fail
when the particle concentration approaches the close packing limit. In dense fluidized suspensions,
particles can form extended networks able to transmit stress via sustained particle-particle contacts.
In this instance, the granular material no longer behaves as a fluid and concepts borrowed from soil
mechanics are required to provide appropriate closure.
(6.6)
This definition coincides with the original formulation of the Archimedes’s principle (force equal
to the weight of the fluid displaced by the particles); therefore, we refer to it as classical definition.
According to equation (6.6), the buoyancy force is constant, regardless of the local mean pressure
and velocity fields that surround the particles. The second definition relates the force to the effec-
tive fluid stress tensor, as reported by Jackson (2000). Per unit volume of suspension, it is:
(6.7)
The last definition, endorsed by Foscolo et al. (1983), Gibb (1991), Di Felice (1994) and Gibilaro
(2001), to mention just a few, deems the force proportional to the gradient of the average fluid pres-
sure. The constant of proportionality is again the solid volume fraction:
(6.8)
For homogeneous dispersions, the second and third definitions clearly coincide, since 〈D〉f (x,t)
vanishes. In uniform mixtures, the solid particles are motionless and equally distributed in space;
moreover, whereas the point fluid velocity is a non-homogeneous field due to the fluid flow around
the particles occurring at the microscopic length scale, the locally averaged velocity field is uni-
form (in the bulk of the suspension, sufficiently far from the system boundaries). Hence, the rate
of deformation tensors vanish and the stress tensors of both phases are isotropic. Furthermore, since
no collisions are present between the particles, the solid pressure is zero and, from equation (6.3),
no stress arises in the solid phase. Equation (6.2), taken in conjunction with the definitions (6.7)
and (6.8), instead gives:
(6.9)
A useful expression for n〈fs〉p•(x,t) can be derived from the locally averaged dynamical equations
reported in Table 1; in light of the previous considerations, these reduce to:
(6.10)
where ρp denotes the particle density. The equation on the right merely states that the particle
weight must be counterbalanced by the fluid-particle interaction force. By subtracting the two
equations, we obtain an explicit expression for the pressure gradient valid for homogeneous sus-
pensions:
(6.11)
(6.12)
We then conclude that for homogeneous fluidized suspensions the difference between the first and
the other two definitions of buoyancy force reduces to the mere choice of which density to employ
in the expression of the force: in the first instance the fluid density is required, whereas in the sec-
ond the suspension bulk density. As we shall see in §6.3.4, the buoyancy force definition is impor-
tant in the development, or simply the choice, of drag force closures.
In the authors’ opinion, the classical definition of buoyancy force, equation (6.6), should be
favored since it is the only definition that preserves the distinctive feature of such a force: being
constant and altogether unrelated to the specific characteristics of the flow (Bird et al., 2007). It
should be emphasized that in the present setting relating the buoyancy force to the suspension bulk
density, as the other two definitions implicitly do, is physically incorrect. In a fluidized suspension
of monodisperse solid particles, or more generally of particles whose dimensions do not differ by
orders of magnitude, each single particle moves through the fluid and displaces its own volume of
fluid as it flows, not of suspension. If it were not for the collisions that inevitably take place among
the particles, these would not feel the presence of the surrounding others. This is evident in dilut-
ed systems. In light of these considerations, the definition of buoyancy force that most naturally
suggests itself is the one that relates the force to the fluid density. This conclusion, as already point-
ed out, holds true as long as all the particles have similar dimensions. For large objects in suspen-
sions of small particles the situation is quite different: here the motion of such bodies takes place
through the suspension and the volume displaced comprises both fluid and particles. It makes there-
fore physical sense to relate the buoyancy force to the suspension bulk density. In this regard, we
mention the work of Poletto & Joseph (1995), where the motion of single test spheres in monodis-
perse mixtures was investigated. The dispersion was modeled as a pseudofluid with effective den-
sity and viscosity. The study aimed to determine the values of these properties that allowed pre-
dicting the settling velocity of test spheres in sedimenting and fluidized suspensions using appro-
priate modifications of the equation of Francis (1933) for the settling of single spheres in pure
Newtonian fluids. The article, in particular, proposed to establish the limits of applicability of the
notion that the effective density of a mixture, on which the expression of the buoyancy force is
based, should be the suspension bulk density. Experimental evidence showed that this approach is
applicable as long as the test spheres are rather larger than the suspended particles and provided
that the solid concentration in the suspension is sufficiently high (more quantitative data can be
found in the paper). For dilute suspensions and test particles of similar or smaller size than the sus-
pended particles the model failed. This corroborates the idea that the suspension bulk density
should be used only when large objects move through suspensions of much smaller particles.
This is true not only for solid suspensions but also for single bodies. The origin of this force is
clearly presented by Maxey & Riley (1983) in the study of the motion of small rigid spheres in non-
uniform flows. Per unit volume of suspension, the force is taken to be:
(6.13)
Here the derivative on the right-hand side is a material derivative relative to a Lagrangian observ-
er moving with the locally averaged velocity of the fluid. An analogous material derivative can also
be introduced for the particle phase; the two are formally defined as follows:
(6.14)
Often the fluid acceleration is far less than the gravitational acceleration; if so, the local fluid accel-
eration force is negligible compared with the buoyancy force. However, the force plays an impor-
tant conceptual role; to prove this point, we resort to a thought experiment reported by Jackson
(2000). A uniform set of solid particles is at rest in a body of fluid. The fluid is also at rest in a ver-
tical container placed on a horizontal plane. The whole system resides in a uniform gravitational
field. At a specified time, the plane that supports the container and the constraints that keep the par-
ticles at rest are removed. The whole system falls freely with an acceleration equal to the gravita-
tional field. As the mean velocity fields of both phases are uniform and no pressure gradients are
present, the effective stress tensors 〈S〉f (x,t) and 〈S〉p (x,t) vanish, and the averaged dynamical
equations reported in Table 1 reduce to:
(6.15)
For convenience, we have used the non-conservative formulation of the equations; to obtain
them, we must combine the dynamical and continuity equations (Bird et al., 2007). Both material
derivatives are equal to the gravitational acceleration; hence, the two equations of motion lead to
the same result: the fluid-particle interaction force must vanish. Clearly, this condition can be met
only if the contribution due to the local fluid acceleration force is accounted for. In fact, as the two
phases undergo the same motion, no slip velocity and acceleration are present between them; con-
sequently, the drag, the virtual mass and the lift forces are all zero. The buoyancy force, converse-
ly, is constant; therefore, the overall interaction force can go to zero only if the local fluid acceler-
ation force is considered:
(6.16)
Since the averaged fluid acceleration is often far less than the local gravitational field, this ideal
experiment does not reflect the normal conditions wherein fluidized systems usually operate; as
pointed out, the local fluid acceleration force can be often neglected (Gidaspow, 1994).
develop theoretical closures for monodisperse fluidized suspensions; we mention, for instance, the
work of Zhang & Prosperetti (1994, 1997) and Jackson (1997, 1998). The latter, however, are valid
for low particle concentrations and under very restricting assumptions, such as vanishing viscosity
(Zhang & Prosperetti, 1994) or small Stokes and Reynolds numbers (Jackson, 1997, 1998; Zhang
& Prosperetti, 1997). Their generalization to other fluid dynamic conditions is not simple and rais-
es conceptual issues (related, for example, to frame indifference and objectivity). For instance, the
lift force on an isolated spherical particle takes quite different functional forms in the inviscid and
low Reynolds number cases (Saffman, 1965; Auton et al., 1988; Jackson, 2000); finding a general
expression for this term still appears to be a daunting task.
A well established expression for the virtual mass force is:
(6.17)
Theoretical, and even empirical, expressions for the coefficient Cv (φ) are still to be found. For very
dilute mixtures of spherical particles, Cv (φ) is taken to be 1/2, since this is the calculated value for
a single sphere in an infinite fluid (Maxey & Riley, 1983; Nadim & Stone, 1991). A similar result
was also found analytically by Zhang & Prosperetti (1994), who derived an exact expression of the
force for inviscid fluids at low particle concentrations. The lift force is usually taken to be:
(6.18)
Equations (6.17) and (6.18) are not frame independent, if taken separately, yet their sum fulfills the
principle of material objectivity provided that Cv (φ) and Cl (φ) are equal (Drew & Passman, 1998).
For this reason, in dilute flows of spherical particles Cl (φ) is taken to be 1/2.
(6.19)
where β depends on voidage, slip velocity magnitude, density and viscosity of the fluid and parti-
cle diameter. Finding a closure for the drag force amounts to finding a constitutive expression for
β.
The definition of buoyancy force is important, insomuch as it directly affects the closure for the
drag. If we denote by n〈fd 〉p and n〈fd 〉p• the drag forces per unit volume of suspension consis-
tent with the buoyancy force definitions (6.6) and (6.12), respectively, it is:
(6.20)
To prove this, we note that for a uniform fluidized suspension, the local fluid acceleration force (if
present) and the virtual mass and lift forces vanish; thus, equation (6.10) yields:
(6.21)
(6.22)
whence equation (6.19) immediately ensues. This relation is necessary to render drag force closures
based on different buoyancy force definitions consistent with one another.
We report here some of the most well-known expressions used to model the drag force in flu-
idized systems. All the closures below are consistent with the classical definition of buoyancy
force, or are rendered so by resorting to equation (6.20).
The Ergun (1952) empirical correlation was originally developed to assess the unrecoverable
pressure drop through packed beds. If we extend its range of validity to homogeneous fluidized
suspensions, we can express the pressure gradient for such systems as:
(6.23)
Here dp is the particle diameter. This relation, taken along with the fluid averaged dynamical equa-
tion, can be used to derive the following drag constitutive expression:
(6.24)
Note that in the limit of very low fluid-particle slip velocities and solid concentrations, equation
(6.24) does not reduce to the Stokes closure law. The latter in this case yields:
(6.25)
Here the first bracketed term represents the number of particles per unit volume of suspension,
whereas the second the Stokes drag force per unit particle. This result, which holds as long as the
suspension is so diluted that the fluid flow field around each particle is not affected by the presence
of the surrounding others, differs from the limiting value taken by the Ergun viscous term for very
low solid concentrations:
(6.26)
This does not surprise, as the original Ergun closure, derived for fixed beds, was not intended to
account for the large voidage variations that fluidized suspensions usually experience. For this rea-
son, equation (6.24) should not be used for dense beds; all the same, in many CFD codes it is adopt-
ed for values of ε up to 0.80, a threshold suggested by Gidaspow (1994).
Other popular closures are those of Lewis et al. (1949), Wen & Yu (1966) and Kmiec (1982), the
first being the default correlation in most CFD codes when the suspension void fraction exceeds
0.80. All three relations can be put in the form:
(6.27)
where α = 2.65 for Lewis et al., α = 2.70 for Wen & Yu and α = 2.78 for Kmiec. Here the drag
coefficient is calculated using the expression of Schiller & Naumann (1935):
(6.28)
Di Felice (1994) modified equation (6.27) considering a functional dependence of the coefficient
α on the particle Reynolds number; he advanced the empirical expression:
(6.29)
Note that regardless of the value taken by the exponent α, equation (6.27) does reduce to the Stokes
closure law in the limit of very low fluid-particle slip velocities and solid concentrations, since for
vanishingly small particle Reynolds numbers and volume fractions of solid, it is:
(6.30)
Ishii & Zuber (1979) modified the drag force closure for single particles by introducing an effec-
tive viscosity for the suspension related to the bed void fraction. The idea is that the presence of
other particles in the bed hinders the motion of each single particle by raising the drag exerted on
it, as if the body were surrounded by a far more viscous fluid. Their closure takes the form:
(6.31)
(6.32)
Here φmax represents the maximum value of φ attainable by the particles in question (that is, the
maximum powder compaction) and is usually set equal to 0.62. The choice made by Ishii & Zuber
of replacing the fluid viscosity with a suspension effective viscosity raises similar objections to
those previously reported for the buoyancy force and the use of a suspension bulk density. Since
each particle moves through the fluid and not through the suspension, the need to modify the fluid
viscosity seems unwarranted. The effect that the solid concentration has on the drag force experi-
enced by a single particle is attributable, in our opinion, to the increase in the fluid velocity gradi-
ents at the particle surface rather than to an actual change in the viscosity of the medium. For com-
pleteness, we report the empirical correlation that Ishii & Zuber used to express the drag coeffi-
cient; this takes the form:
(6.33)
Note that equation (6.31) can be only used if the suspension does not approach its maximum com-
paction, since when this happens, the effective viscosity and in turn the drag force diverge. This is
clearly incorrect for a packed assembly of particles always retains a finite permeability to fluid
flow. A way to fix this problem is setting φmax to a value slightly higher than the real maximum
solid compaction.
We finally report the closure recently advanced by Mazzei & Lettieri (2007); this can be
arranged in a form similar to equation (6.27), where the coefficient α is a function of both particle
Reynolds number and suspension void fraction. In particular, it is:
(6.34)
Note that, since ne is a function of Re*, and Re* in turn depends on ne, to compute these quanti-
ties we need to solve a non-linear system. However, since ne has a very narrow range of variation
(being bounded between 2.40 and 4.80), finding the solution usually requires few iterations.
The accuracy of the closure laws reported above should be checked case by case against exper-
imental evidence. A simple way of doing this would be comparing theoretical with experimental
expansion profiles of non-cohesive homogeneous fluidized beds, since for these systems the theo-
retical profiles depend solely on the constitutive expressions used for the drag force. This is surely
the best methodology of testing any drag force closure, but is time consuming. Alternatively, we
can use the empirical correlation of Richardson & Zaki (1954), which permits to determine the
expansion of a non-cohesive fluidized suspension for any given value of the superficial fluid veloc-
ity. Of course, this correlation is not perfect, but being based on an extensive amount of experi-
mental data, is the most reliable we have. It takes the form:
(6.35)
where us and ut are the superficial fluid velocity and the particle unhindered terminal settling
velocity, respectively, and ne depends on the free fall particle Reynolds number:
(6.36)
Several correlations have been proposed to estimate ne, their main difference consisting in the
limiting values ascribed to the parameter. In some equations, in the limits of viscous and inertial
regimes, ne is equal to 4.65and 2.39 (Richardson & Zaki, 1954), in others takes the values of 4.70
and 2.35 (Rowe, 1987), whereas in still others tends to 4.80 and 2.40 (Khan & Richardson, 1989;
Gibilaro, 2001), respectively. Here, we use a modification of the empirical expression of Rowe,
where the limits for ne suggested by Khan & Richardson and Gibilaro are adopted:
(6.37)
Equation (6.35) tells us that once the physical properties of the fluid (density and viscosity) and of
the particles (density and diameter) have been assigned, the voidage of a non-cohesive fluidized
suspension, and in turn its expansion, is a simple function of us. Now, if we assume that this equa-
tion is correct (or at least sufficiently reliable for our purposes), we can say that for a generic non-
cohesive fluidized suspension, uniformly dispersed and in fluid dynamic equilibrium, a drag force
expression is accurate only if it yields similar expansion profiles to those predicted by equation
(6.35) for any fluid dynamic regimes and void fractions. This concept can be formalized as follows.
For the systems under scrutiny, equation (6.22) holds; thus, equation (6.19) yields:
(6.38)
For a homogeneous suspension and an observer at rest relatively to the particles, it is:
(6.39)
where us is the superficial fluid velocity. As us and g are parallel and opposite, introducing equa-
tion (6.39) into equation (6.38) and replacing φ with 1– ε gives:
(6.40)
Similar to equation (6.35), equation (6.40) is also a functional relationship between us and ε. This
equation is directly related to the closure used for β. Thus, a most convenient way of testing the
validity of the latter is to ascertain that the predictions based on equation (6.40) are fully consistent
with those obtained from its counterpart, equation (6.35), over the whole range of fluid dynamic
regimes and for any value of the fluid volume fraction. Our methodology is therefore the follow-
ing: for a generic Group A powder fluidized by liquid or gas, we assign the superficial velocity of
the fluid, and we calculate the corresponding value of the suspension void fraction yielded by equa-
tion (6.35); then, for the same value of us, we solve equation (6.40) using the closure for β in which
we are interested, and we evaluate the corresponding value of the voidage; finally, we compute the
ratio between the two results and quantify its deviation from unity. Naturally, the more the devia-
tion, the less the agreement between the two relations.
Figures 2 to 6 report the voidage ratio as a function of the particle Reynolds number parametric
in the void fraction for all the closures described above; for equation (6.27) we have selected the
closure of Wen & Yu (1966), choosing α equal to 2.70. The Ergun (1952) closure agrees with equa-
tion (6.35) in the viscous and intermediate fluid dynamic regimes; in the first the agreement
improves as the void fraction decreases, whereas in the second an opposite trend is observed. For
high particle Reynolds numbers, the equations agree only for fixed or nearly fixed beds. The Wen
& Yu (1966) closure yields the same results as equation (6.35) in both viscous and inertial limits
for any void fraction; in the intermediate region, however, a good agreement is found only for high
dilutions. The predictions are partially improved if we use the expression for o advanced by Di
Felice (1994), whereas the match is perfect if we use the closure of Mazzei & Lettieri (2007); this
is because the latter is based on the Richardson & Zaki (1954) correlation. Finally, the Ishii & Zuber
(1979) closure agrees with equation (6.35) in the viscous and intermediate fluid dynamic regions
and as long as the suspension is not too diluted; in the inertial regime, especially for high void frac-
tions, the closure significantly deviates from the Richardson & Zaki (1954) correlation.
1.50
0.4
0.5
1.40
0.6
Voidage Ratio: Ergun over Richardson & Zaki
1.20
1.10
1.00
Low Bed Voidage
0.90
0.80
1.0E-06 1.0E-04 1.0E-02 1.0E+00 1.0E+02 1.0E+04 1.0E+06
Re
Figure 2. Ratio between the equilibrium void fractions based on the Ergun (1952) clo-
sure and on the Richardson & Zaki (1954) empirical correlation. The curves are para-
metric in the Richardson and Zaki void fraction.
1.20
0.4
0.5
Voidage Ratio: Wen & Yu over Richardson & Zaki
0.6
1.15
0.7
0.8
0.9
1.10
1.05
1.00
High Bed Voidage
0.95
1.0E-06 1.0E-04 1.0E-02 1.0E+00 1.0E+02 1.0E+04 1.0E+06 1.0E+08
Re
Figure 3. Ratio between the equilibrium void fractions based on the Wen & Yu (1966)
closure and on the Richardson & Zaki (1954) empirical correlation. The curves are para-
metric in the Richardson and Zaki void fraction.
1.15
0.4 0.5
0.6 0.7
Voidage Ratio: Di Felice over Richardson & Zaki
0.8 0.9
1.10
Low Bed Voidage
1.05
1.00
0.95
1.0E-06 1.0E-04 1.0E-02 1.0E+00 1.0E+02 1.0E+04 1.0E+06 1.0E+08
Re
Figure 4. Ratio between the equilibrium void fractions based on the Di Felice (1994) clo-
sure and on the Richardson & Zaki (1954) empirical correlation. The curves are para-
metric in the Richardson and Zaki void fraction.
1.30
0.4
Voidage Ratio: Ishii & Zuber over Richardson & Zaki
1.20 0.5
0.6
1.10
Low Bed Voidage
1.00
0.90
0.80
0.60
1.0E-06 1.0E-04 1.0E-02 1.0E+00 1.0E+02 1.0E+04 1.0E+06 1.0E+08
Re
Figure 5. Ratio between the equilibrium void fractions based on the Ishii & Zuber (1979)
closure and on the Richardson & Zaki (1954) empirical correlation. The curves are para-
metric in the Richardson and Zaki void fraction.
1.05
1.03
1.00
0.98
0.95
1.0E-06 1.0E-04 1.0E-02 1.0E+00 1.0E+02 1.0E+04 1.0E+06 1.0E+08
Re
Figure 6. Ratio between the equilibrium void fractions based on the Mazzei & Lettieri
(2007) closure and on the Richardson & Zaki (1954) empirical correlation. The curves
are parametric in the Richardson and Zaki void fraction.
(6.41)
Single-sphere calculations indicate a value of 3/4 for Cf (φ); accordingly, this value is expected to
be valid in the limit of vanishing small particle concentrations (Jackson, 1997, 1998).
The elastic force is related to the suspension void fraction gradients; in uniform suspensions it
vanishes, but is important in the study of the stability of the uniform fluidization state. The transi-
tion from particulate to aggregative fluidization has been, for many decades, object of investiga-
tion. The phenomenon, deemed to result from fluid dynamic instabilities, is still nowadays incom-
pletely understood. Most of the theoretical studies regarding the stability of homogenous disper-
sions was carried out analytically, without the aid of computational tools. This is because decades
ago the computing power of personal computers was much less than it is today. Some researchers
ascribed the stability of uniform suspensions to the effect of interparticle forces (Massimilla et al.,
1972; Mutsers & Rietema, 1977; Rietema & Piepers, 1990), whereas others sought for a purely
fluid dynamical explanation (Jackson, 1963; Murray, 1965; Pigford & Baron, 1965; Verloop &
Heertjes, 1970). The first approach failed to yield quantitative results; this major shortcoming
might explain why the second approach has undergone much more development. The method
adopted in this instance was usually that of linear stability analysis. The equations of motion were
first written one-dimensionally and then linearized; finally, the stability of the system, when sub-
jected to small perturbations, was examined. The hope was to find formal stability for sufficiently
low values of the fluid flux and instability for values greater than a certain, well-defined threshold
limit. But this expectation was not met: the homogeneous state resulted to be intrinsically unstable,
this being true even if, in addition to the buoyancy and drag forces, the other contributions previ-
ously discussed – that is, virtual mass force, lift force and the like – are taken into account in the
interaction force between the phases.
Some researchers accepted this conclusion, putting forward the following explanation: homo-
geneous beds do not exist formally – that is, in a mathematical sense; however, what really counts
is the rate of growth of the disturbances within the bed. In liquid-fluidized systems the rate is
always very contained, so that disturbances develop so slowly that the status of visible bubbles is
never attained. The suspension, although formally unstable, appears to be uniform. In systems flu-
idized by gases, conversely, the rate of growth is such that disturbances swiftly reach detectable
size, even in shallow beds. Other researchers, most noticeably Foscolo & Gibilaro (1987), found
this explanation unconvincing. They maintained that several fluidized systems display an indis-
putable homogeneous behavior, which equations of motion of sufficient accuracy should be able to
predict. This led Foscolo & Gibilaro to introduce a new force, the elastic force, whose origin is
purely hydrodynamical and based on rather simple physical considerations (for a detailed deriva-
tion, refer to Gibilaro, 2001). The force takes the form:
(6.42)
where e is a unit vector parallel and opposite to the gravitational field and Ef (ε) is a positive scalar,
named elastic modulus, that results to be equal to:
(6.43)
If this additional contribution is included in the fluid-particle interaction force, applying linear sta-
bility analysis results in an easy analytical criterion for discriminating between stable and unstable
fluidization: the particulate regime is stable if the stability function below is negative.
(6.44)
Loss of stability was interpreted by Foscolo & Gibilaro as the onset of bubbling; in this view, the
minimum bubbling voidage coincides with the zero of the stability function. This does not have to
be necessarily so, since loss of formal stability merely indicates that small disturbances no longer
dampen out; they might, however, grow so slowly as to never really turn into visible bubbles. In
this case, the system would continue to appear perfectly stable. Mazzei et al. (2006) and Mazzei &
Lettieri (2008) have advanced a new closure for the elastic force based on the classical buoyancy
force definition, equation (6.6); Foscolo & Gibilaro (1987), conversely, had based theirs on the def-
inition (6.8). The force takes the form:
(6.45)
where nd (x,t) is the drag force versor. With this closure, which regards the elastic force as the com-
ponent of the drag related to void fraction gradients, the stability function is:
(6.46)
(6.47)
where ni〈f〉pik (x,t) is the force exerted by phase k on phase i per unit volume of suspension and ζik
is the particle-particle drag coefficient for the two particle classes involved. If we assume that equa-
tion (6.47) holds, then the closure problem reduces to finding a constitutive expression for ζik.
Gidaspow et al. (1985) have advanced the relation:
(6.48)
where ρi, ρk, di and dk are the densities and diameters of the particles of classes i and k, respec-
tively, eik is their coefficient of restitution and Cik is given by:
(6.49)
(6.50)
Otherwise, it is:
(6.51)
In the relations above, Φi and Φk are the particle volume fractions at maximum packing for the
phases i and k, respectively; moreover, it is:
(6.52)
The closure of Syamlal (1987) models the particle-particle drag coefficient as:
(6.53)
where Fik is a coefficient of friction and gik is a radial distribution function that takes the form:
(6.54)
Bell (2000) modified the closure of Gidaspow et al. (1985), defining Cik as:
(6.55)
Gera et al. (2004) suggested that the equations above should include an additional term that is nec-
essary to prevent the particle phases from segregating when they are packed. Without it, equations
(6.48) and (6.53) allow packed particles of different size to segregate, a phenomenon that is not
observed experimentally. To avoid this, they recommended adding to the coefficient ζik the term
KP, where K is a parameter that must be adjusted for different particle mixtures to match the
actual segregation rate, and P is given by:
(6.56)
Here ε is a threshold that should coincide with the packed bed void fraction. The term KP is
added so that when the powder nears maximum packing the particle-particle drag increases suffi-
ciently to make the two solid phases move together as if they were one phase (thus hindering seg-
regation).
a more sophisticated model for the simulation of polydisperse dense gas-fluidized beds, presented
in Mazzei et al. (2008).
(7.1)
The Richardson & Zaki correlation and the Mazzei & Lettieri (2007) closure agree perfectly. The
latter is always more accurate than the other two closures. The agreement between the three con-
stitutive expressions varies with the void fraction: for System 1 it improves as the latter rises, while
for System 2 we observe an opposite behavior. Furthermore, whereas the accuracy of the Ergun
(1952) and Wen & Yu (1966) equations is strongly dependent on the voidage, that of the Mazzei &
Lettieri (2007) equation is not.
Note that, in all the simulations a commercial CFD code would use the Ergun equation, since
the voidage is always less than 0.80 (refer to §6.3.4). This is unfortunate, because the Wen & Yu
closure is clearly more accurate and should be used in its place. This conclusions, however, cannot
be generalized to any system. Sometimes, equation (6.24) does yield better predictions; the ques-
tion, therefore, is when the CFD code should switch from one equation to the other. The threshold
of 0.80 is not always the best choice. Also in this respect, the closure of Mazzei & Lettieri (2007)
offers an improvement: holding for any fluid dynamic regime and void fraction, no switching is
required.
Figure 7 reveals that for System 1 the accuracy of equation (6.24) improves as the suspension
becomes more diluted. This surprises because the Ergun correlation was derived for packed beds
and was not meant to account for large variations in powder compaction. Note also that the best
prediction is found for a void fraction equal to 0.80, the threshold where the closure is usually no
longer adopted because deemed too inaccurate. To explain this apparent inconsistency Mazzei &
Lettieri (2008) resorted to Figure 2, previously discussed in §6.3.4, reporting in it the equilibrium
states wherein Systems 1 and 2 reside when full expansion is reached. Let us clarify how they rep-
resented these states on the diagram. For a given superficial fluid flux, the numerical equilibrium
void fractions were determined using the Ergun (1952) closure; from these, they calculated the
Reynolds number Re and the voidage ratio r between the computational voidage ε and that pre-
dicted by the Richardson & Zaki (1954) correlation. The equilibrium states were identified by the
points with coordinates (Re, r ). This is done in Figure 9. The points can be regarded as states in
which the systems operate when the transient regime comes to an end. System 1 operates in the
intermediate fluid dynamic region; here the Ergun closure is more accurate at high dilution, in
agreement with the trend found in Figure 7. System 2, conversely, operates in the inertial region of
the chart, where the closure rapidly loses accuracy as dilution increases. This behavior agrees with
the trend found in Figure 8.
12.0
Ergun (1952)
6.0
4.5
3.0
1.5
0.0
0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85
Experimental Voidage
Figure 7. Percent error of the computational equilibrium void fractions based on the
Ergun (1952), Wen & Yu (1966) and Mazzei & Lettieri (2007) closures and on the
Richardson & Zaki (1954) empirical correlation against the experimental void fractions
for System 1.
30.0
System 2
27.5 Richardson & Zaki (1954)
25.0
Mazzei & Lettieri (2007)
22.5
20.0
Wen & Yu (1966)
Percent Error
17.5
12.5
10.0
7.5
5.0
2.5
0.0
0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85
Experimental Voidage
Figure 8. Percent error of the computational equilibrium void fractions based on the
Ergun (1952), Wen & Yu (1966) and Mazzei & Lettieri (2007) closures and on the
Richardson & Zaki (1954) empirical correlation against the experimental void fractions
for System 2.
1.50
0.4
0.5
1.40
Voidage Ratio: Ergun over Richardson & Zaki
0.6
High Bed Voidage
0.7
1.30
CFD simulations - System 1
1.10
1.00
0.80
1.0E-06 1.0E-04 1.0E-02 1.0E+00 1.0E+02 1.0E+04 1.0E+06
Re
7.2. 2D simulations of the transition from bubbling to slugging and turbulent flu-
idization
This section shows the results of 2D simulations of the transition from bubbling to slugging and
turbulent fluidization of a sand-like Geldart Group B material having dp = 300µm and
ρp = 2500kg/m3. The results from the simulations are presented in Figure 10 and show the ability
of general multiphase Eulerian models to reproduce qualitatively some of the fundamental aspects
of different fluidization regimes. Full details can be found in Lettieri et al. (2003).
Simulations of the bubbling fluidized bed are shown in Figure 10a, where the voidage profile is
reported for a fluidizing gas velocity of 0.25m/s. The chosen velocity corresponds to about 3umf
(umf = 0.08m/s). The simulation shows formation of waves of voidage at the bottom of the bed,
which travel through the bed and subsequently break to form bubbles as the simulation progresses.
The computed bubbles show regions of voidage distribution at the bubble edge, as experimentally
observed by Yates et. al (1994).
Slugging fluidization occurs in beds of small diameter, such as laboratory-scale reactors and
pilot-scale units, if three conditions are met: 1) maximum bubble size is greater than 0.6 times the
diameter of the bed, 2) the superficial gas velocity is sufficiently high and 3) the bed is sufficient-
ly deep. Baeyens & Geldart (1974) combined conditions 2 and 3 into a criterion for slug flow:
(7.2)
where D is the vessel diameter and hmf is the bed height at minimum fluidization velocity. This
relation predicts the minimum superficial gas velocity ums at which slugging will occur for a given
gas-solid system. The criterion shows that ums depends on the velocity and bed height at minimum
fluidization. If we apply this criterion to the 300µm sand material, we see that slugging should
occur for gas velocities in excess of 0.26m/s. The voidage profiles shown in Figure 10b were
obtained for a fluidization velocity equal to 0.35m/s. Notably the bubbles formed at the distributor
grow by coalescence until their dimensions reach the size of the diameter of the vessel containing
the bed; this is a typical feature of slugging beds (Geldart, 1973) that appears to be correctly repro-
duced. The bed expands as the slugs rise through it, and its surface oscillates and drops sharply as
the slugs break through.
Figure 10. CFD simulations of the transition from bubbling (a) to slugging (b) and turbu-
lent (c) fluidization for a Group B gas-fluidized bed.
Using the two-phase theory of Davidson & Harrison (1963) and the expression of Clift et al.
(1978) for a slug rise velocity usl , we can predict the maximum height of a slugging bed; this is:
(7.3)
Here uex stands for the excess velocity, defined to be the difference between the superficial fluid
velocity us and the minimum fluidization velocity umf. The values for hmax predicted by equation
(7.3) were compared with those obtained from the simulation; the latter were slightly underesti-
mated by about 10%.
The effect of the operating conditions on the transition from bubbling to turbulent fluidization
of Geldart Group A and B materials has been extensively studied (Yates, 1996). In the literature
there are correlations that permit to estimate the superficial velocity at which the transition for a
given fluid-particle system should occur (Geldart, 1973). The transition is a gradual phenomenon
that spans a range of fluid velocities and depends on the gas and solid properties and also on the
equipment scale. In this work, the transition from slug to turbulent flow was investigated by simu-
lating an increase in the gas pressure from ambient up to 10bar. The correlation proposed by Cai et
al. (1989) was used to calculate the transition velocity uc. This is a function of the bed diameter
and of the fluid and particle properties:
(7.4)
where ρ¯f and µ̄f denote the fluid density and viscosity at ambient conditions, respectively.
According to this correlation, the transition from bubbling to turbulent fluidization for the 300µm
particles fluidized at 10bar should occur at velocities greater than 0.25m/s. The void fraction pro-
files presented in Figure 10c were obtained using the same value for the fluidizing gas velocity
employed to simulate the slugging regime (0.35m/s); this enabled to highlight the transition to tur-
bulent fluidization by means of simply increasing the gas density, which is equivalent to increas-
ing the gas pressure in the bed. It is well known in fluidization that increasing the fluid pressure
leads to an increased homogeneity in the fluid bed, as reported by Avidan & Yerushalmi (1982).
The high-pressure simulation clearly shows the principle elements of the mechanism of transition
from the bubbling to the turbulent regime. A considerable increase in the expansion of the dense
phase is observed, and the large discrete bubbles, which were simulated at ambient pressure, are on
the whole absent at high pressure. At 10bar, the gas percolates through small bubbles and channels,
making the bed progress towards a structure of greater homogeneity.
0.590
0.570
0.550
Average Bed Voidage
0.530
0.510
0.490
0.450
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
Time (s)
Average values of the bed height and voidage resulted slightly higher for the 2D case, in agree-
ment with the experimental evidence of Fitzgerald (1985), who reported an overestimation of bed
expansion in two dimensional fluid beds as compared to three dimensional ones. This points out
that 2D simulations may be used only in cases where the flow is by nature two dimensional, i.e., in
cases where the variations in space and time in a given direction of the physical space are negligi-
ble compared with those in the other directions, as emphasized by Peirano et al. (2001).
Regarding the difference in computational time between 2D and 3D simulations, it is worth
reporting that the computational time required for the 3D simulations is as much as 30 times the
computational time for the 2D simulations. For 3D simulations, one second of real time required
about 540 hours (equivalent to 23 days) on a 1.7 Ghz Dell 530 Precision workstation.
2D 3D
Bed height (m) 0.180 0.174
Bed voidage 0.541 0.527
Bubble hold-up 0.067 0.091
0.040
0.030
0.025
0.020
0.015
0.010
0.005
0.005 0.010 0.015 0.020 0.025 0.030 0.035 0.040
Predicted Bubble diameter (m)
Figure 12. Simulated average bubble diameters against bubble diameters predicted by
the Darton et al. (1977) equation for 2D and 3D simulations.
For such analysis, a definition of bubble is needed. In this study, this is assumed as the coherent
region within the fluidized bed where the solid volume fraction is lower than 0.15, after Kuipers et
al. (1993). There is no broad agreement on this definition in the literature. van Wachem et al. (1998)
and Gelderbloom et al. (2003) assumed that the upper limit for the solid volume fraction within a
bubble is 0.20. Yates et. al (1994) reported on a gas voidage distribution around a bubble, which
was measured experimentally by means of an x-rays visualization technique. They observed a void
fraction of unity for the bubble core and of 0.627 for the cloud region surrounding it.
The Darton model, based on the two-phase theory of Davidson & Harrison (1963), assumes that
bubbles coalesce along preferred paths, where the distance traveled by two neighboring bubbles
before coalescence is proportional to their horizontal distance. Their proposed equation, which
gives the bubble diameter as a function of the bed height and fluid excess velocity, is:
(7.5)
where db is the bubble diameter, h is the height above the distributor and A0 is the catchment area,
which characterizes the gas distributor. The constant 0.54 was obtained experimentally. Note that
the model does not account for any splitting, nor is applicable for fine powders like Geldart Group
C materials. It also does not apply to slug flow. In absence of available data on the distributor char-
acteristics, Darton et al. suggested a value for A0 equal to zero; this fairly corresponds to the inlet
boundary conditions set-up during this work, where no distributor was simulated.
The literature shows that the results obtained from the Darton model were thoroughly validated
against experimental data obtained by different authors, including Werther (1976), Rowe & Everett
(1972), Yasui & Johanson (1958), Geldart (1968) and Fryer & Potter (1976). The experimental data
reported by Darton et al. give measured bubble sizes for a number of bubbling fluid beds with dif-
ferent geometries, gas distributors, particle systems, excess gas velocities and bed heights. The
fluid beds simulated by Cammarata et al. (2003) fell within the range of operating conditions
employed in the experiments reported by Werther. In this case, Darton et al. found that the agree-
ment between experimental measures and predictions from their equation was good up to a ratio
between the excess gas velocity and the minimum fluidization velocity equal to 6.4 and a bed
height of 0.40m. Throughout all the simulations, Cammarata et al. set the ratio between excess and
minimum gas velocity to 2, giving a maximum bed height of 0.18m.
The bubble diameter analysis was performed at different heights in the bed, from above the dis-
tributor up to its surface, considering 13 planes across the bed. In 3D and 2D simulations, the bub-
ble diameters at the various bed heights were estimated considering the equivalent diameter of the
sphere and of the circle that have the same volume and area as the bubble, respectively. To com-
pare 2D with 3D results, however, Cammarata et al. modified the 2D bubble diameters to take into
account the actual 3D shape of a bubble. This is done as follows. Suppose that the 2D and 3D bub-
bles are given by half a circle and half a sphere, respectively, having the same diameter db. Then,
it is:
(7.6)
where db,a and db,v denote the equivalent diameters for the 2D and 3D bubbles, respectively, as
defined above. Thus, to account for the shape factor, we must modify the 2D bubble diameters as
follows:
(7.7)
As Figure 12 shows, the 2D bubble diameters are always smaller than the 3D ones, and are also
less than those predicted by the Darton et al. equation, the difference increasing with the bed height.
The data in Table 2 show similar values of the bed voidage for both 2D and 3D simulations, but a
higher bubble holdup for the 3D case. This indicates that while in both 2D and 3D the overall vol-
ume of gas in the fluid bed is the same, the bubbles produced in the 3D simulations are greater than
those in the 2D case, in agreement with the results shown in Figure 12.
The results obtained for bed expansion, bubble holdup and bubble size provide a coherent pic-
ture of the fluid dynamic behavior of the system investigated, in agreement with experimental
observations. Even if the Darton et al. model is semiempirical, the finding that the computed bub-
ble size increases steadily along the bed height and matches the prediction of the model represents
a significant test of the predictive capability of the granular kinetic models, which do not assume a
priori the existence of a bubble phase.
a binary fluidized mixture of natural rutile (flotsam) and slag (jetsam). These two powders were
characterized by the size distributions reported in Figure 13, had mean diameters of 186 µm and
305 µm, respectively, and particle density equal to 4200kg/m3. Owoyemi & Lettieri investigated
three mixture compositions corresponding to jetsam mass fractions averaged over the entire bed of
0.25, 0.50 and 0.75. For each case, they considered two excess gas velocities equal to 0.10m/s and
0.20m/s; this enabled them to test thoroughly the model for different degrees of mixing and segre-
gation. Each simulation ran for ten seconds real time; computational results obtained for mixing
and segregation patterns, bed height and bubble dynamics were validated against experimental
data.
The equations were solved in the CFD commercial code CFX 4.4. The Eulerian stress associat-
ed with the solid phases was neglected. This omission might give rise to localized overcompaction
in some regions of the simulated fluid bed; to counter this effect, the authors implemented the cor-
rective numerical algorithm for solid overcompaction developed by Owoyemi et al. (2007) for
binary systems.
35
(a)
35
30
30
25
25
%w t in sieve
20
20
%wt in sieve
15
15
10
10
5
5
0
212
300
425
500
600
0
106
125
150
180
212
250
300
90
Figure 13. Particle size distribution of (a) natural rutile and (b) slag.
7.4.1. Experimental
In all the experiments, the bed was initially completely segregated, the two powders being arranged
in two layers where the flotsam particles (natural rutile, smaller particles) occupied the bottom half
of the bed whilst the jetsam particles (slag, bigger particles) the top half. The mixture was then flu-
idized at a constant excess gas velocity, its value being determined from the expression of Wu &
Baeyens (1998) that relates the mixing index M representative of the bed mixedness to the visible
bubble flow rate. The mixture minimum fluidization velocity was calculated using the correlation
of Gossens et al. (1971); the bed freeze test was used to obtain experimentally the mixing and seg-
regation profiles in the bed.
(7.8)
where 〈wjet 〉t is the average mass fraction of the jetsam phase in the top region of the bed and
〈wjet 〉o is the average mass fraction of the jetsam phase evaluated over the entire bed. Following
van Wachem et al. (2001), the top region was assumed to be the top 25% of the bed. This definition
equates a state of perfect jetsam segregation at the bottom of the bed to a mixing index of M = 0 and
a state of perfect mixing to a mixing index of M = 1. Perfect jetsam segregation at the top of the
bed, which is usually imposed before the system is fluidized, is instead given by a mixing index of
M = (mj + mf )/mj, where mj and mf are the overall masses of jetsam and flotsam in the bed,
respectively.
Table 3 compares the experimental and numerical values of the mixing index obtained for the
different initial jetsam fractions. The numerical values were reached after ten seconds of simula-
tion and refer to an excess gas velocity of 0.20m/s; they are within 8% of the experimental results.
Whereas the experimental mixing index slightly decreases when the initial percentage of the jetsam
in the mixture increases, the numerical mixing index slightly increases, following therefore an
opposite trend.
Table 3. Experimental and theoretical values for the mixing index M obtained for different
jetsam fractions at an excess gas velocity of 0.20 m/s.
Mixing Index M
ω jet CFD Experimental Wu & Baeyens (1998)
0.75 1.00 0.92 0.90
0.50 0.98 1.00 0.90
0.25 0.95 1.00 0.90
Both experimental and numerical results were also compared with the predictions of the semi-
empirical correlation of Wu & Baeyens (1998) and were found to be within 10%. For all the cases
investigated, the correlation underpredicts both the experimental and computational mixing index-
es, giving a constant value for M. This is because the correlation does not take into account the dif-
ferent average composition of the mixtures, assuming that M depends only on the ratio between the
larger and smaller particle diameters and the excess gas velocity, both of which are the same in all
the three cases examined. Note, however, that a fluidized bed with a mixing index greater than 0.9
is usually considered to be in a well-mixed state. For all the cases investigated, both the simulations
and experiments showed that when the fluid bed becomes fully mixed, it remains well mixed from
then onwards.
Figure 14 shows the time evolution of the jetsam mass fraction profile through the bed for the
mixture with initial mean jetsam fraction equal to 75%. The figure shows that during the first two
seconds the jetsam phase spreads rapidly through the system, its mass fraction increasing at the bot-
tom of the bed until an even distribution is attained throughout the system after about four seconds
of simulation. Little variation in the jetsam mixing profile occurs afterwards. Figure 15 shows the
same profile for the mixture with initial mean jetsam fraction equal to 25%. Also in this case, the
total computational mixing time is around four seconds. This might be because the excess gas
velocity was the same for all the simulations, so that the same amount of gas was available to form
bubbles, which are the driving force that promotes mixing. Similar results were also obtained for
the 50% jetsam fraction case, as reported by Owoyemi & Lettieri (2008).
To validate the model, Owoyemi & Lettieri also compared quantitatively the predicted bulk
properties of the bed, such as bed height, voidage and bubble holdup, with those measured exper-
imentally. The results are reported in Table 4 and refer to an excess gas velocity of 0.20m/s. The
properties were time averaged after the first two seconds of simulation, since this reduces the start-
up influence. There are significant differences between numerical predictions and experimental
findings for the initial jetsam mass fractions investigated. Quantitatively, a percent error between
2.0% (25% jetsam fraction) and 4.0% (75% jetsam fraction) was found for the bed height. A greater
percent error ranging between 5.6% (25% jetsam fraction) and 7.0% (75% jetsam fraction) was
found for the bed voidage. The agreement between numerical and experimental results is better at
lower jetsam mass fractions. These deviations might be highlighting the shortcoming of the sim-
plification adopted in treating the powders as a perfect binary mixture, without accounting for their
realistic particle size distributions shown in Figure 13.
For the bubble holdup, the authors found an even greater disparity between numerical predic-
tions and experimental results, the percent error lying between 6.7% (25% jetsam fraction) and
97.8% (75% jetsam fraction). We remind that the bubble holdup is defined as the fraction of gas
that leaves the fluid bed through bubbles. The experimental data reported in Table 4 show that the
bubble holdup increases with increasing jetsam mass fraction, suggesting that the more the latter
increases, the more the excess gas leaves the bed preferentially via bubbles. Computationally, val-
ues obtained for bubble holdup are constant and are accurate only at low jetsam fractions, the error
being within 10%. The agreement becomes poorer with increasing jetsam concentration. A possi-
ble explanation for this might be that the computational bubbles were smaller than the real ones, as
we discuss in greater detail below.
0.35
0.0 seconds Limit of
0.30
1.0 seconds perfect mixing
2.0 seconds
4.0 seconds
0.25
Bed Height (m)
0.20
0.15
0.10
0.05
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Jetsam Mass Fraction
Figure 14. Computational evolution of the jetsam mass fraction profile with time for a
mixture with overall slag mass fraction equal to 75%.
0.35
Limit of
perfect mixing
0.30
0.25
Bed Height (m)
0.20
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Jetsam Mass Fraction
Figure 15. Computational evolution of the jetsam mass fraction profile with time for a
mixture with overall slag mass fraction equal to 25%.
Owoyemi & Lettieri analyzed the bubble dynamics within the bed by comparing numerical
results with experimental data and predictions of the Darton et al. (1977) bubble growth equation
previously presented in §7.3.2. Figure 16 plots the bubble diameter versus the bed height for all the
simulated bubbles computed at a constant excess gas velocity of 0.20m/s for an initial jetsam con-
centration of 75%. Over 5000 bubbles – the total number of bubbles that evolved during ten sec-
onds of real time simulation – are plotted on the graph. The data show an enormous spread in pre-
dicted bubble diameters due to coalescence, breakup and the interaction of bubbles with the wall
of the computational vessel. We see that the predicted bubble diameters are smaller in the higher
part of the fluidized bed, their values being within ±20% of the predictions given by the Darton et
al. equation. The prominence of small bubbles in the bed may be due to less pronounced bubble
coalescence in the higher regions of the bed.
0.07
Bubble diameter (m)
0.06
0.05
0.04
0.03
0.02
0.01
0.00
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
Bed height (m)
Figure 16. Comparison between computational bubble diameters and diameters predict-
ed by the Darton et al. (1977) equation at an excess gas velocity of 0.20 m/s for a mix-
ture with overall slag mass fraction equal to 75%.
Experimentally, a total of over 500 bubbles at different bed heights were analyzed and their
diameter obtained. Owoyemi & Lettieri found that these values were within ±30% of the predic-
tions of the Darton et al. equation. This discrepancy might be due to the assumption of Darton et
al. that bubbles line up and grow as close together as possible, following specific paths; this is
improbable in the chaotic flow patterns found in fluidized beds. Hence, differences between exper-
imental bubble sizes and sizes obtained from the Darton et al. correlation may be due to the differ-
ent bubble growth mechanisms which take place in real systems, i.e., the combination of the pres-
sure decrease towards the top of the fluidized bed and the horizontal and vertical coalescence of
trailing and neighboring bubbles. These might not necessarily line up and grow as close together
as possible as assumed in the Darton et al. correlation.
Figure 17 compares the simulated and experimental bubble diameters at different bed heights.
Here, the bubble sizes were averaged into ten classes by interpolation. The experimental bubble
diameters are larger than the simulated ones in most part of the bed. This agrees with the evidence
reported in §7.3, where 2D and 3D simulations were compared and it was found that bubbles
obtained from 2D simulations are always smaller that those obtained from 3D simulations.
However, the discrepancy between the CFD simulations and the results obtained from both exper-
iments and the semiempirical correlation may also be attributed to the omission of the distributor
plate in the simulations; this may be the cause of the CFD bubble sizes being smaller than the
experimental ones and those predicted by the Darton et al. correlation. A change in mixture com-
position, for jetsam mixture fractions of 0.50 and 0.25, had no effect on the numerical predictions
of bubble size in the computational simulations.
(7.9)
where 〈wjet 〉b and 〈wjet〉t are the jetsam mass fractions in the bottom and top halves of the bed. Cs
varies between -100 and +100, with the first denoting perfect jetsam segregation at the bed top, 0
representing perfect mixing, and +100 indicating perfect jetsam segregation at the bed bottom. Both
experimentally and numerically, Cs increases inversely with the jetsam mass fraction.
As done previously, we also compare the values of the mixing index with the predictions
obtained from the semiempirical correlation of Wu & Baeyens (1998); these deviate from the for-
mer by 18%, see Table 5. This might be because segregation is overwhelmed by the solids recir-
culation and mixing present in the experimental and computational beds. Cooper & Coronella
(2005) arrived at similar conclusions, when investigating the segregation dynamics of natural rutile
and coke particles.
Table 5. Experimental and theoretical values for the mixing index M obtained for different
jetsam fractions at an excess gas velocity of 0.10 m/s.
Figure 18 compares theoretical and experimental concentration profiles of the jetsam phase
through the bed for 75% initial jetsam fraction. The numerical values were computed after ten sec-
onds of simulation by splitting the bed into 28 horizontal layers of equal volume and by calculat-
ing the average jetsam mass fraction in each layer. The computational and the experimental jetsam
mass fraction profiles are found to almost overlap with the limit of perfect mixing profiles and are
consistent with the figures reported in Table 5, highlighting that mixing prevails over segregation.
0.07
0.05
Bubble Diameter (m)
0.04
0.03
0.02
0.01
0.35
Limit of
CFD
0.30 Experimental perfect mixing
0.25
Bed Height (m)
0.20
0.15
0.10
0.05
0.00
0.3 0.4 0.5 0.6 0.7 0.8 0.9
Jetsam Mass Fraction
Figure 18. Comparison between computational and experimental segregation patterns
for a mixture with overall slag mass fraction equal to 75%.
8. CONCLUSIONS
Fluidization is a well-established technology used in many industrial processes; however, accurate
modeling of the complex phenomena taking place in fluid bed reactors is still a great challenge for
both academics and industrialists. Whilst different modeling approaches are available to model the
dynamics of fluidized beds at different scales, CFD is recognized to be one of the most promising
techniques to aid an in-depth understanding of the physics underpinning fluidized bed behaviors,
while also enabling simulations at a scale of industrial interest. Within this context, this paper has
presented a review of the different modeling approaches used for fluid beds and guides the reader
through some of the principal averaging techniques that consent to turn granular systems into con-
tinua and to derive the Eulerian equations of motion for fluidized suspensions of a finite number of
monodisperse particle classes. The paper highlights the importance of the closure problem; in par-
ticular it critically provides an overview of some widely adopted closure equations for the fluid-
particle interaction force, demonstrating their suitability for different flow regimes and voidage in
the fluid bed. The discussion of some key results obtained from the CFD modeling of mono and
binary fluidized beds highlights some of the challenges ahead. In this regard, much work still
remains to be done towards developing correct mathematical expressions for the solid stress ten-
sors and the particle-particle interactions that will ultimately lead to model realistic multicompo-
nent systems and underpin the design of novel fluidized bed processes based on some fundamen-
tal science.
REFERENCES
Allen, M. P. & Tildesley, D. J. 1990. Computer Simulations of Liquids. Oxford Science Publications.
Anderson, T. B. & Jackson, R. 1967. A fluid mechanical description of fluidized beds. Equations of motion. Ind. Eng. Chem.
Fundam. 6, 527.
Arastoopour, H., Wang, C. H. & Weil, S. A. 1982. Particle-particle interaction force in a dilute gas-solid system. Chem. Eng.
Sci. 37, 1379.
Astarita, G. 1989. Thermodynamics: An Advanced Textbook for Chemical Engineers. Springer.
Astarita, G. & Marrucci, G. 1974. Principles of Non-Newtonian Fluid Mechanics. McGraw-Hill.
Auton, T. R., Hunt, J. C. R. & Prud’homme, M. 1988. The force exerted on a body in inviscid, unsteady, non-uniform, rota-
tional flow. J. Fluid Mech. 197, 241.
Avidan, A. A. & Yerushalmi, J. 1982. Bed expansion in high velocity fluidization. Powder Technol. 32, 223.
Baeyens, J. & Geldart, D. 1974. An investigation into slugging fluidized beds. Chem. Eng. Sci. 29, 255.
Balzer, G., Boelle, A. & Simonin, O. 1995. Eulerian gas-solid flow modelling of dense fluidized bed. In proc. of the 8th
International Conference on Fluidization. Tours, France.
Balzer, G., Simonin, O., Boelle, A. & Lavieville, J. 1996. A unifying modelling approach for the numerical prediction of
dilute and dense gas-solid two phase flow. In proc. of the 5th International Conference on Circulating Fluidized Bed.
Beijing, China.
Barnea, E. & Mizrahi, J. 1973. A generalized approach to the fluid dynamics of particulate systems. Chem. Eng. Sci. 5, 171.
Basset, A. B. 1888. Treatise on Hydrodynamics. Deighton Bell.
Batchelor, G. K. 1970. The stress system is a suspension of force-free particles. J. Fluid Mech. 41, 545.
Batchelor, G. K. & Green, J. T. 1972. The determination of the bulk stress in a suspension of spherical particles to order c2.
J. Fluid Mech. 56, 401.
Bell, R. 2000. Numerical modelling of multi-particle flows in bubbling gas-solid fluidized beds. Ph.D. Dissertation. School
of Mechanical and Manufacturing Engineering, Swinburne University of Technology.
Bird, R. B., Stewart, W. E. & Lightfoot, E. N. 1960. Transport Phenomena. Wiley.
Bird, R. B., Stewart, W. E. & Lightfoot, E. N. 2007. Transport Phenomena. Wiley.
Birkhoff, G. 1950. Hydrodynamics. Princeton University Press for University of Cincinnati.
Brenner, H. 1964. The Stokes resistence of an arbitrary particle -Part IV. Arbitrary fields of flow. Chem. Eng. Sci. 19, 703.
Brilliantov, N. V. & Poschel T. 2004. Kinetic Theory of Granular Gases. Oxford University Press.
Brinkman, H. C. 1952. The viscosity of concentrated suspensions and solutions. J. Chem. Phys. 20, 571.
Buyevich, Y. A. 1971. Statistical hydrodynamics of disperse systems. Part 1. Physical background and general equations. J.
Fluid Mech. 49, 489.
Buyevich, Y. A. 1994. Fluid dynamics of coarse dispersions. Chem. Eng. Sci. 49, 1217.
Buyevich, Y. A. 1997. Particulate pressure in monodisperse fluidized beds. Chem. Eng. Sci. 52, 123.
Buyevich, Y. A. & Kapbasov, S. K. 1994. Random fluctuations in a fluidized bed. Chem. Eng. Sci. 49, 1229.
Buyevich, Y. A. & Kapbasov, S. K. 1999. Particulate pressure in disperse flow. Int. J. Fluid Mech. Res. 26, 72.
Cai, P., Chen, S. P., Jin, Z. Q. & Wang, Z. W. 1989. Effect of operating temperature and pressure on the transition from bub-
bling to turbulent fluidization. AIChE Symp. Series 85, 37.
Cammarata, L., Lettieri, P., Michale, G. & Colman, D. 2003. 2D and 3D CFD simulations of bubbling fluidized beds using
Eulerian-Eulerian models. IJCRE. 1, 48.
Chapman, S. & Cowling, T. G. 1970. The Mathematical Theory of Non-Uniform Gases. Cambridge University Press.
Clift, R., Grace, J. R. & Weber, M. E. 1978. Bubbles, Drops and Particles. Academic Press.
Clift, R., Seville, J. P. K., Moore, S. C. & Chavarie, C. 1987. Comments on buoyancy on fluidized beds. Chem. Eng. Sci.
42, 191.
Colafigli, A., Mazzei, L. Lettieri, P. & Gibilaro, L. G. 2009. Apparent viscosity measurements in a homogeneous gas-flu-
idized bed. Chem. Eng. Sci. 64, 144.
Cooper, S. & Coronella, C. J. 2005. CFD simulations of particle mixing in a binary fluidized bed. Powder Technol. 151, 27.
Crowe, C. 2005. Multiphase Flow Handbook. CRC Press Inc.
Crowe, C. T., Sommerfeld, M. & Tsuji, Y. 1997. Multiphase Flows with Droplets and Particles. CRC Press.
Cundall, P. A. & Strack, O. D. 1979. A discrete numerical model for granular assemblies. Géotechnique 29, 47.
Darton, R. C., LaNauze, R. D., Davidson, J. F. & Harrison D. 1977. Bubble growth due to coalescence in fluidised beds.
Trans. Instn. Chem. Eng. 55, 274.
Davidson, J. F. & Harrison, D. 1963. Fluidised Particles. Cambridge University Press.
Delhaye, J. M. & Achard, J. L. 1977. On the use of averaging operators in two phase flow modeling. Thermal and Hydraulic
Aspects of Nuclear Reactor Safety. Part 1: Light Water Reactors. ASME Winter Meeting.
Delhaye, J. M. & Achard, J. L. 1978. On the averaging operators introduced in two-phase flow. In Transient Two-phase
Flow. In Proc. of CSNI Specialists Meeting, Toronto, Canada.
Di Felice, R. 1994. The voidage function for fluid-particle interaction systems. Int. J. Multiphase Flow 20, 153.
Ding, J. & Gidaspow, D. 1990. A bubbling fluidization model using kinetic theory of granular flow. AIChE J. 36, 523.
Drew, D. A. 1971. Averaged field equations for two-phase media. Stud. Appl. Math. 50, 133.
Drew, D. A. 1983. Mathematical modelling of two-phase flow. Annu. Rev. Fluid Mech. 15, 261.
Drew, D. A. & Lahey, R. T. 1993. Analytical modelling of multiphase flow. In Particulate Two-Phase Flow. Butterworth-
Heinemann.
Drew, D. A. & Passman, S. L. 1998. Theory of Multicomponent Fluids. Applied Mathematical Sciences. Springer.
Drew, D. A. & Segel, L. A. 1971. Averaged equations for two-phase flows. Stud. Appl. Math. 50, 205. Einstein,
A. 1906. Eine Neue Bestimmung der Molekuldimensionen. Ann. Phys. 19, 289.
Einstein, A. 1911. Berichtigung zu meiner Arbeit: Eine Neue Bestimmung der Molekuldimensionen. Ann. Phys. 34, 591.
Enwald, H., Peirano, E. & Almstedt, A. E. 1996. Eulerian two-phase flow theory applied to fluidization. Int. J. Multiphase
Flow 22, 21.
Epstein, N. 1984. Comments on a unified model for particulate expansion of fluidized beds and flow in fixed porous media.
Chem. Eng. Sci. 39, 1533.
Ergun, S. 1952. Fluid flow through packed columns. Chem. Eng. Progr. 48, 89.
Fan, L. S. & Zhu, C. 1998. Principles of Gas-Solid Flows. Cambridge University Press.
Fan, L. S., Han, L. S. & Brodkey R. S. 1987. Comments on the buoyancy force on a particle in a fluidized suspension. Chem.
Eng. Sci. 42, 1269.
Fitzgerald, T. J. 1985. Coarse particle systems. In Fluidization. Academic Press.
Foscolo, P. U. & Gibilaro, L. G. 1987. Fluid dynamic stability of fluidized suspensions: The particle bed model. Chem. Eng.
Sci. 42, 1489.
Foscolo, P. U., Gibilaro, L. G., Waldram, S. P. 1983. A unified model for particulate expansion of fluidized beds and flow
in fixed porous media. Chem. Eng. Sci. 38, 1251.
Francis, A. W. 1933. Wall effect in falling fall method for viscosity. Physics 4, 403.
Frankel, N. A. & Acrivos, A. 1967. On the viscosity of a concentrated suspension of solid spheres. Chem. Eng. Sci. 22, 847.
Fryer, C. & Potter, O. E. 1976. Experimental investigation of models for fluidized bed catalytic reactors. AIChE J. 22, 38.
Geldart, D. 1968. The expansion of bubbling fluidized beds. Powder Technol. 1, 355.
Geldart, D. 1973. Gas Fluidization Technology. Wiley.
Gelderbloom, S. J., Gidaspow, D. & Lyczkowski, R. W. 2003. CFD simulations of bubbling/collapsing fluidized beds for
three Geldart groups. AIChE J. 49, 844.
Gera, D., Syamlal, M. & O’Brien, T. J. 2004. Hydrodynamics of particle segregation in fluidized beds. Int. J. Multiphase
Flow. 30, 419.
Gibb, J. 1991. Pressure and viscous forces in an equilibrium fluidized suspension. Chem. Eng. Sci. 46, 379.
Gibilaro, L. G. 2001. Fluidization-Dynamics. Butterworth Heinemann.
Gibilaro, L. G., Gallucci, K., Di Felice, R. & Pagliai, P. 2007. On the apparent viscosity of a fluidized bed. Chem. Eng. Sci.
62, 294.
Gidaspow, D. 1994. Multiphase Flow and Fluidization. Academic Press.
Gidaspow, D. & Ettehadieh B. 1983. Fluidization in two-dimensional beds with a jet. Part II: hydrodynamic modeling. Ind.
Eng. Chem. Fundam. 22, 193.
Gidaspow, D., Syamlal, M. & Seo, Y. C. 1985. Hydrodynamics of fluidization of single and binary particles: supercomput-
er modeling. In proc. of the 5th International Conference on Fluidization. Elsinore, Denmark.
Gidaspow, D., Syamlal, M. & Seo, Y. C. 1986. Hydrodynamics of fluidization: supercomputer generated vs. experimental
bubbles. J. Powder & Bulk Solids Tech. 10, 19.
Goldshtein, A. & Shapiro, M. 1995. Mechanics of collisional motion of granular materials. Part I. General hydrodynamic
equations. J. Fluid Mech. 282, 75.
Gossens, W. R. A., Dumont, G. L. & Spaepen, G. L. 1971. Fluidization of binary mixtures in the laminar flow region. Chem.
Eng. Prog. Symp. Ser. 67, 38.
Grace, J. R. 1970. The viscosity of fluidized beds. CJChE 48, 30.
Graham, A. L. 1981. On the viscosity of suspensions of solid spheres. Appl. Sci. Res. 37, 275.
Gurtin, M. E. 2003. An Introduction to Continuum Mechanics. Academic Press.
Haff, P. K. 1983. Grain flow as a fluid mechanical phenomenon. J. Fluid Mech. 134, 401.
Hinch, E. J. 1977. An averaged equation approach to particle interactions in a fluid suspension. J. Fluid Mech. 83, 695.
Hoomans, B. P. B., Kuipers, J. A. M., Briels, W. J. & van Swaaij, W. P. M. 1996. Discrete particle simulation of a two-dimen-
sional gas-fluidized bed: a hard sphere approach. Chem. Eng. Sci. 51, 99.
Ishii, M. & Zuber, N. 1979. Drag coefficient and relative velocity in bubbly, droplet or particulate flows. AIChE J. 25, 843.
Jackson, R. 1963. The mechanics of fluidized beds: Part I. The stability of the state of uniform fluidization. Trans. Instn.
Chem. Eng. 41, 13.
Jackson, R. 1997. Locally averaged equations of motion for a mixture of identical spherical particles and a Newtonian fluid.
Chem. Eng. Sci. 52, 2457.
Jackson, R. 1998. Erratum. Chem. Eng. Sci. 53, 1955.
Jackson, R. 2000. The Dynamics of Fluidized Particles. Cambridge Monographs on Mechanics. Cambridge University
Press.
Jean, R. H. & Fan, L. S. 1992. On the model equations of Gibilaro and Foscolo with corrected buoyancy force. Powder
Technol. 72, 201.
Jenkins, J. T. 1987. Rapid flows of granular materials. In Non-Classical Continuum Mechanics. Cambridge University
Press.
Jenkins, J. T. & Savage, S. B. 1983. A theory for the rapid flow of identical, smooth, nearly elastic, spherical particles. J.
Fluid Mech. 130, 187.
Khan, A. R. & Richardson, J. F. 1989. Fluid-particle interactions and flow characteristics of fluidized beds and settling sus-
pensions of spherical particles. Chem. Eng. Comm. 78, 111.
King, D. F., Mitchell, F. R. G. & Harrison, D. 1981. Dense phase viscosities of fluidized beds at elevated pressures. Powder
Technol. 28, 55.
Kleinstreuer, C. 2003. Two-phase Flow. Routledge.
Kmiec, A. 1982. Equilibrium of forces in a fluidized bed -experimental verification. Chem. Eng. J. 23, 133.
Koch, D. L. 1990. Kinetic theory for a monodisperse gas-solid suspension. Phys. Fluids 2, 1711.
Koch, D. L. & Sangani, A. S. 1999. Particle pressure and marginal stability limits for a homogeneous monodisperse gas flu-
idized bed: kinetic theory and numerical simulation. J. Fluid Mech. 400, 229.
Kuipers, J., van Duin, K., van Beckum, F. & van Swaaij, W. 1993. Computer simulation of the hydrodynamics of a two-
dimensional gas-fluidized bed. Computers Chem. Eng. 17, 839.
Kunii, D. & Levenspiel, O. 1989. Fluidization Engineering. Butterworth Heinemann.
Ladd, A. J. C. & Verberg, R. 2001. Lattice-Boltzmann simulations of particle fluid suspensions. J. Stat. Phys. 104, 1191.
Lai, W. M., Rubin, D. & Krempl, E. 1993. Introduction to Continuum Mechanics. Butterworth Heinemann.
Lettieri, P., Saccone, G. & Cammarata, L. 2004. Predicting the transition from bubbling to slugging fluidization using com-
putational fluid dynamics. Chem. Eng. Res. Des. 82, 939.
Lettieri, P., Cammarata, L., Michale, G. & Yates, J. G. 2003. CFD simulations of gas-fluidized beds using alternative
Eulerian-Eulerian modelling approaches. IJCRE. 1, 1.
Lewis, W. K., Gilliland, W. C. & Bauer, W. C. 1949. Characteristics of fluidized particles. Ind. Eng. Chem. 41, 1104.
Lu, H., Wang, S., Zhao, Y., Yang, L., Gidaspow, D. & Ding, J. 2005. Prediction of particle motion in a two-dimensional bub-
bling fluidized bed using hard-sphere model. Chem. Eng. Sci. 60, 3217.
Lun, C. K. K., Savage, S. B., Jeffrey, D. J. & Chepurniy, N. 1984. Kinetic theories for granular flow: Inelastic particles in
Couette flow and slightly inelastic particles in a general flow field. J. Fluid Mech. 140, 223.
Massaudi, M. 2002. On the importance of material frame-indifference and lift forces in multiphase flows. Chem. Eng. Sci.
57, 3687.
Massoudi, M., Rajagopal, K. R., Ekmann, J. M. & Mathur, M. P. 1992. Remarks on the modelling of fluidized systems.
AIChE J. 38, 471.
Massimilla, L., Donsi’, G. & Zucchini, C. 1972. The structure of bubble-free gas fluidized beds of fine fluid cracking cat-
alyst particles. Chem. Eng. Sci. 27, 2005.
Maxey, M. R. & Riley, J. J. 1983. Equation of motion for a small rigid sphere in a nonuniform flow. Phys. Fluids 26, 883.
Mazzei, L. 2008. Eulerian modelling and computational fluid dynamics simulation of mono and polydisperse fluidized sus-
pensions. Ph.D. Dissertation. Department of Chemical Engineering, University College London.
Mazzei, L. & Lettieri, P. 2007. A drag force closure for uniformly-dispersed fluidized suspensions. Chem. Eng. Sci. 62,
6129.
Mazzei, L. & Lettieri, P. 2008. CFD simulations of expanding/contracting homogeneous fluidized beds and their transition
to bubbling. Chem. Eng. Sci. 63, 5831.
Mazzei, L., Lettieri, P. & Marchisio, D. L. 2008. Investigation into the dynamics of polydisperse fluidized beds by using the
direct quadrature method of moments. In proc. of the 11th International Conference on Multiphase Flow in Industrial
Plants. Palermo, Italy.
Mazzei, L., Lettieri, P., Elson, T. & Colman, D. 2006. A revised monodimensional particle bed model for fluidized beds.
Chem. Eng. Sci. 61, 1958.
Metzner, A. B. 1985. Rheology of suspensions in polymeric liquids. J. Rheol. 29, 739.
Mooney, J. 1951. The viscosity of a concentrated suspension of spherical particles. J. Colloid Interphase Sci. 91, 160.
Murray, J. D. 1965. On the mathematics of fluidization. J. Fluid Mech. 21, 465.
Mutsers, S. M. P. & Rietema, K. 1977. The effect of interparticle forces on the expansion of a homogeneous gas-fluidized
bed. Powder Technol. 18, 239.
Nadim, A. & Stone, H. A. 1991. The motion of small particles and droplets in quadratic flows. Stud. Appl. Math. 85, 53.
Nakamura, K. & Capes, C. E. 1976. Vertical pneumatic conveying of binary particle mixtures. In Fluidization Technology.
Hemisphere Publishing Corporation.
Nigmatulin, R. I. 1979. Spatial averaging in the mechanics of heterogeneous and dispersed systems. Int. J. Multiphase Flow.
5, 353.
Ouyang, J. & Li, J. 1999. Particle-motion-resolved discrete model for simulating gas-solid fluidization. Chem. Eng. Sci. 54,
2077.
Owoyemi, O. & Lettieri, P. 2008. CFD modelling and validation of bi-disperse fluidized industrial powders. Ind. Eng.
Chem. Res. 47, 6316.
Owoyemi, O., Lettieri, P. & Place, R. 2005. Experimental validation of Eulerian-Eulerian simulations of rutile industrial
powders. Ind. Eng. Chem. Res. 44, 9996.
Owoyemi, O., Mazzei, L. & Lettieri, P. 2007. CFD modeling of binary-fluidized suspensions and investigation of role of
van Wachem, B. G. M., Schouten, J. C., Krishna, R. & van den Bleek, C. M. 1998. Eulerian simulations of bubbling behav-
iour in gas-solid fluidised beds. Computers Chem. Eng. 22, 299.
van Wachem, B. G. M., Schouten, J. C., van den Bleek, C. M., Krishna, R. & Sinclair, J. L. 2001. Comparative analysis of
CFD models of dense gas-solid systems. AIChE J. 5, 1035.
Verloop, J. & Heertjes, P. M. 1970. Shock waves as a criterion for the transition from homogeneous to heterogeneous flu-
idization. Chem. Eng. Sci. 25, 825.
Wen, C. Y. & Yu, Y. H. 1966. Mechanics of Fluidization. Chem. Eng. Prog. Symp. Series 62, 100.
Werther, J. 1976. Bubble growth in large diameter fludized beds. In proc. of the 1st International Conference on Fluidization.
Asilomar, California.
Whitaker, S. 1969. Advances in the theory of fluid motion in porous media. Ind. Eng. Chem. 61, 14.
Wu, S. Y. & Baeyens, J. 1998. Segregation by size difference in gas fluidized beds. Powder Technol. 98, 139.
Xu, B. H. & Yu, A. B. 1997. Numerical simulation of the gas-solid flow in a fluidized bed by combining discrete and par-
ticle method with computational fluid dynamics. Chem. Eng. Sci. 52, 2785.
Yasui, G. & Johanson, L. N. 1958. Characteristics of gas pockets in fluidized beds. AIChE J. 4, 445.
Yates, J. G. 1996. Effects of temperature and pressure on gas-solid fluidization. Chem. Eng. Sci. 51, 167.
Yates, J. G., Cheesman, D. & Sergeev, Y. A. 1994. Experimental observations of voidage distribution around bubbles in a
fluidized bed. Chem. Eng. Sci. 49, 1885.
Ye, M., van der Hoef, M. A. & Kuipers, J. A. M. 2005. The effects of particle and gas properties on the fluidization of
Geldart A particles. Chem. Eng. Sci. 60, 4567.
Zhang, D. Z. & Prosperetti, A. 1994. Averaged equations for inviscid disperse two-phase flow. J. Fluid. Mech. 267, 185.
Zhang, D. Z. & Prosperetti, A. 1997. Momentum and energy equations for disperse two-phase flows and their closure for
dilute suspensions. Int. J. Multiphase Flow. 23, 425.