You are on page 1of 16

Chemical Engineering Science 66 (2011) 23402355

Contents lists available at ScienceDirect

Chemical Engineering Science


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

DEMCFD modeling of a uidized bed spray granulator


L. Fries a,n, S. Antonyuk a, S. Heinrich a, S. Palzer b
a
Institute of Solids Process Engineering and Particle Technology, Hamburg University of Technology, Denickestrae 15, 21073 Hamburg, Germany
b
Nestle Research Center Vers-Chez-Les-Blanc, Route du Jorat 57, 1000 Lausanne 26, Switzerland

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

Article history: Coupled DEMCFD simulations have been performed to study the uid and particle dynamics in a
Received 31 August 2010 uidized bed spray granulator on the scale of individual particles. The aim of this study is to develop a
Received in revised form model of a uidized bed granulator by combining the gas and particle dynamics with a simple model of
3 February 2011
particle wetting. Based on material tests, the collision behavior of g-Al2O3 particles was characterized
Accepted 17 February 2011
and incorporated into the contact model. For two different granulator congurations, a bubbling
Available online 24 February 2011
uidized bed with top-spray injection and a Wurster-coater, wetting of the particle surface is estimated
Keywords: based on the residence time distribution inside a biconical spray zone. The effect of the geometry of the
Particle formation apparatus on the homogeneity of wetting is analyzed in order to understand the performance and
Fluidization
specicity of different granulator congurations. Compared to the top-spray granulator, the Wurster-
Granulation
coater has a narrow residence time distribution of the particles in the spray zone, which corresponds to
Discrete element modeling (DEM)
Wurster-coater more homogeneous particle wetting. For the Wurster-coater, the effect of process parameters like air
Computational uid dynamics (CFD) ow rate and geometry details like the position of the Wurster tube is studied using the model.
A stable circulating uidization regime is established by keeping the jet velocity above 100 m/s and the
gap distance between bottom plate and Wurster tube below 15 times the particle diameter
Based on a description of the physical material properties, an effective modeling tool for design and
scale-up of a uidized bed spray granulator is obtained. Modeling the interactions of the individual
particles, a step forward is taken towards a description of the micro-processes in granulation, which
also considers the material properties.
& 2011 Elsevier Ltd. All rights reserved.

1. Introduction apparatus is a key parameter affecting both particle size and


structure of the product (Palzer, 2009).
Fluidized bed spray granulation and agglomeration both play Fluidized bed granulators or agglomerators can be categorized
an important role in the manufacturing of powder granules in the by ve different congurations, which are schematically shown
food, ne chemicals and pharmaceutical industries as dust-free in Fig. 1:
and free-owing particles can be produced in a process with
favorable heat and mass transfer conditions and good solids (a) Top spray: conventional, no draft tube, the nozzle is usually
mixing (Heinrich et al., 2003). Further motives for this size- placed at the top outside the moving powder bed spraying
enlargement process are the improvement of appearance, taste downwards.
or odors of particles or protection from oxygen, humidity, light or (b) Bottom spray: the spray is introduced from a nozzle located at
incompatible active agents and to control the dissolution time the center and bottom of the bed, inside the powder bulk.
(Hemati et al., 2003). (c) Wurster-coater: bottom spray granulator where the bed contains
Liquid suspension or a solution or melt is injected into a a draft tube (Wurster insert) to create a circulating ow pattern.
uidized bed of particles in order to coat the particle surface or to (d) Rotating uidized bed: a rotor placed at the bottom of the bed
achieve particle growth by layering. The injected liquid may also is rotated and air is fed through a gap between the rotor and
work as a binder to induce agglomeration. A homogeneous the wall. Liquid is sprayed onto the roping solids ow from a
distribution of the spray liquid in the uidized bed is a prerequi- tangential injection nozzle.
site for uniform growth, whereas local over-wetting leads to the (e) Spouted bed: the uidization air is introduced from the bottom
formation of particle clusters. The moisture distribution in the via adjustable cylinders. The powder is entrained by a spout at
the center. Liquid is injected from a bottom spray nozzle.

n
Corresponding author. Tel.: + 49 40 42878 2143; fax: + 49 40 42878 2678. Top spray agglomerators (Fig. 1a) are used by the food
E-mail address: lennart.fries@tuhh.de (L. Fries). industry for size enlargement of culinary powders, seasonings,

0009-2509/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2011.02.038
L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355 2341

Fig. 1. Schemes of different uidized bed spray granulator congurations: (a) top spray granulator, (b) bottom spray granulator, (c) Wurster-coater, (d) rotor granulator
and (e) spouted bed granulator.

beverage powders and food ingredients. Bottom spray installa- adjustable, so that the gap distance between the Wurster tube
tions are preferably applied for agglomeration of dairy powders and the bottom plate can be varied. This gap distance is an
(Fig. 1b). important process parameter controlling the number of particles
The Wurster-coater (Fig. 1c) is a common device in the transported towards the center into the spout and the particle
pharmaceutical industry used to coat tablets and smaller particles concentration inside the wetting zone.
in the range 201000 mm (Wurster, 1966; Tang et al., 2008; Since so far operating conditions and the equipment geometry
Karlsson et al., 2009). It is described by Teunou and Poncelet have been developed empirically, the fundamental physical
(2002) as the most efcient batch uid bed coating equipment. mechanisms governing the process on the micro scale are not
A cylindrical draft tube is inserted vertically into the granulator. well understood (Werner et al., 2007; Turton, 2008). Experimen-
The segmented distributor plate has larger orices below the tal results are mostly specic to material and equipment selec-
Wurster tube and therefore the uidization gas enters at higher tion. Furthermore, the effectiveness of improvements on the
velocity at the center than in the annulus. Due to the central gas/ vessel shape or inserts like the Wurster tube is hard to assess,
liquid injection which is concurrent to the particle motion, a as it is difcult to measure the ow eld and particle motion
circulating movement of the particles is induced and the geome- inside the bed.
try can be divided into a wetting and a drying zone. In the central A lot of research work has been done recently to overcome
part inside the Wurster tube the particles are transported these limitations and to achieve a fundamental understanding of
upwards in a spout and collide with droplets from the atomizer. the granulation process on the micro-level scale (Iveson et al.,
The wetted particles decelerate in the expansion chamber above 2001; Litster, 2003; Tan et al., 2006; Peglow et al., 2007; Rajniak
the tube and fall down to the dense region of particles in the et al., 2009). However, due to its complexity, a comprehensive
annulus outside the tube, while they are dried by the heated physically based process model of uidized bed granulation is still
uidization air current. From the dense region, the particles are not available. The choice of an agglomerator/granulator type still
sucked into the Wurster tube, where they enter the next motion remains rather a question of philosophy than a decision based
cycle. The height of the draft tube inside the apparatus is on facts.
2342 L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355

2. Modeling the granulation process However, all Eulerian multiphase models developed so far
fundamentally lack a description of the particleparticle friction
In the literature, many attempts have been made to model and inter-particle cohesive forces (Patil et al., 2005). Energy
the process of particle formulation in a uidized bed in terms of dissipation through nonideal elastic collisions can only be incor-
population balances (Hounslow et al., 1988; Litster et al., 1995; porated into the model by using a damping or restitution
Heinrich et al., 2003; Kumar et al., 2006). coefcient to describe the material properties and their tempera-
KuShaari et al. (2006) developed a model to predict the coating ture and moisture dependency.
mass uniformity in a Wurster-coater using a Monte Carlo method. To describe the process in detail on the scale of individual
The movement of the particles is modeled as a random walk, particles, the Discrete-Element-Method (DEM) is suitable. As each
where steps in the horizontal and vertical direction are based on particle is tracked individually, the method allows a complete
the particle velocity distribution that was determined experimen- representation of the particleparticle and particlewall interac-
tally using high speed image analysis. The wetting intensity in the tions and their inuence on the process dynamics.
spray zone is higher in the vicinity of the nozzle in order to To be able to model a uidized bed, several authors have
consider the effect of particle sheltering. KuShaari et al. (2006) found coupled DEM and CFD to set up a Discrete Particle Model (DPM)
good agreement between simulations and experimental data. The (Hoomans et al., 1996; Tsuji et al., 1993; Deen et al., 2007; van
simulation results were highly dependent on the applied geometry Buijtenen et al., 2009).
of the spray zone, which was arbitrarily chosen due to the lack of Some attempts have also been made to model the uidized
experimental data on the droplet deposition. bed spray granulation process using DEM. Goldschmidt et al.
Link et al. (2007) found for a two phase injection nozzle using (2003) modeled in 2D the agglomeration of glass beads in a top
a discrete element simulation that 95% of the droplets were spray uidized bed where an aqueous solution of polyethylene
deposited on a particle within a limited spray zone. The geometry glycol (PEG) binder is injected. In their concept, 3 particle species
of this spray zone can be approximated by a biconical shape. are tracked: dry particles, wet particles and droplets. While the
Werner et al. (2007) proposed a list of 10 micro-level processes collision of dry particles is described using a traditional hard-
relevant for coating of food powders. Terrazas-Velarde et al. (2009) sphere contact model, collisions at a wet spot on the particle
presented a stochastic model for uidized bed granulation where surface always lead to agglomeration (which is in practice not
the micro-process kinetics of binder addition, particle collision, always the case). Upon collision with a droplet, the particle
particle wetting, agglomeration/rebound, drying of deposited dro- surface is wetted partially with a lm of constant thickness. The
plets and breakage are considered individually. The simulation box wetted area is shrinking at a constant rate to account for drying of
used in their work contained 2000 particles but uid dynamics the binder. In reality, the liquid penetrates into the solid matrix or
were not represented by the model. The results indicate that the dissolves. Drying of the binder is characterized by a continuously
mechanism of droplet-capture is the rate-determining step during increasing surface viscosity. Link et al. (2007) extended this
agglomeration. model to represent a spout uidized bed.
In the development of stochastic and population balance Gantt and Gatzke (2005) modeled the agglomeration of glass
models, the assumption of a spatially homogeneous (well-mixed) beads in a high shear granulator. They developed a comprehen-
system is usually employed. However, since powder character- sive implementation of the coalescence model by Liu and Litster
istics and essential hydrodynamic and kinetic parameters regard- (2000) including compaction and breakage mechanisms. Unfortu-
ing liquidsolid contact and granule agglomeration, particle nately, due to the high numerical effort required, the investigated
mixing and segregation are lumped into the kinetic rate con- system included only 192 particles.
stants, population balance models assuming homogeneity cannot Kafui and Thornton (2008) presented a 3D model of a uidized
be applied for a priori design and scale-up of the uid bed bed spray granulator, where the contact model for wet particles is
granulation process (Goldschmidt, 2001). based on the concept of surface energy (JKR Model, Johnson et al.,
In a uidized bed granulator, the wetting and drying rates are 1971). Particles are wetted, i.e. they pick up surface energy
highly dependent on the ow eld of gas and particles. Local according to their position and residence time inside a conical
temperatures in the equipment volume are important for the spray zone.
moisture equilibrium and inuence the drying rate (Hede et al., In the framework of a multi-scale modeling scheme it is
2009). But not only the drying kinetics are a function of moisture, theoretically and practically attractive to employ DEM for the
relative velocity and temperature. Also the agglomeration kinetics derivation of physically based expressions for aggregation and
that strongly depend on the mechanical material properties are breakage kernels of a population balance model representing the
governed by the local process conditions and the humidity granulation process on a macroscopic scale.
distribution. Especially for amorphous food powders, the mechan- In this contribution, a DEM model is coupled with a CFD
ical particle properties are highly dependent on the moisture simulation to consider the interactions between uid and
content (Palzer, 2009). While exceeding the glass transition particle phase using the EulerLagrange approach. The aim of
temperature Tg by 1030 1C, the viscosity decreases from 1010 this study is to develop a model of a uidized bed granulator by
to 1012 Pa s to approximately 108109 Pa s (Ferry, 1980). At such combining the gas and particle dynamics with a simple model of
viscosity, particles get adhesive. To predict agglomeration events, particle wetting. The effect of the geometry of the apparatus
a model of a uidized bed should include a ne resolution of the on the homogeneity of wetting is analyzed in order to under-
local temperature and humidity elds. stand the performance and specicity of different granulator
Following the concept of the kinetic theory of granular ow congurations.
(KTGF), multiphase computational uid dynamic (CFD) models have Two simulation studies are presented. The rst study describes
been established to describe the uid dynamics of spout uidized particle wetting based on the residence time distribution of the
bed systems (Gryczka et al., 2009). The method allows simulating a particles in a conical spray zone at the tip of the injection nozzle.
lab-scale uidized bed within a reasonable period of time. Two granulator congurations are compared: a bubbling uidized
Karlsson et al. (2009) developed a 3D multiphase CFD model of bed with a top-spray nozzle and a Wurster-coater. In the second
the particle and gas motion in the Wurster process including heat study the inuence of the velocity of the injected jet and the gap
and mass transfer between the particles and the gas phase and distance below the Wurster tube on the particle and uid
the continuous injection of liquid through a central spray nozzle. dynamics in the Wurster-coater is investigated.
L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355 2343

3. Mathematical model relatively coarse grid compared to the particle size. The side
length of an average Eulerian grid element has to be large
3.1. Modeling the motion of the discrete particles compared to the particle diameter (Deen et al., 2007), so that at
least 10 particles t into each uid grid cell.
The motion of each individual particle (index i) with mass mi Turbulence is included using the standard ke turbulence
and volume Vi in the system is calculated using Newtons second model. The gas injection by the nozzle is included as a free jet.
law: The nozzle tip is a circular area (diameter 2 mm) where the gas
velocity is prescribed by the boundary condition (velocity inlet).
dvi d2 r Vb
mi mi 2i Vi rp i ug vi mi g Fc,i FA,i , 1 The gas volume around the nozzle tip has a rened mesh
dt dt 1e
compared to the rest of the volume of the granulator. Still, this
where vi is the velocity and ri the position vector of the particle i. mesh is too coarse in order to achieve a precise description of the
The forces on the right hand side of Eq. (1) are, respectively, due ow phenomena inside the jet. The resulting gas ow prole is
to the pressure gradient, drag, gravity, contact forces (i.e. due to shown in Figs. 11, 16 and 20.
collisions) and adhesive particleparticle interactions (for instance
Van-der-Waals forces) (Deen et al., 2007).
The angular momentum of the particle i is calculated with the
Euler equation 3.3. Two-way coupling scheme for the motion of continuous and
discrete phase
doi
Ii Ti , 2
dt The continuous gas phase and the discrete particle phase
where Ti is the torque and Ii is the moment of inertia of particle i. interact with each other through momentum transfer and
The interphase momentum transfer coefcient b is frequently through the phase volume fractions. The integrated drag force
modeled by combining the Ergun (1952) equation for dense (Eqs. (3) and (4)) acting on all particles within a uid cell is
regimes (e o0.8) equivalent to the sink term added to the momentum balance
(Eq. (9)). In order to obtain a realistic momentum transfer, it is
1e2 Zg
bErgun 150 1:751eRep 3 necessary to calculate for each CFD mesh cell the volume fraction
e d2p
es occupied by the solid. For this purpose, the volume of each
and the correlation proposed by Wen and Yu (1966) for dilute particle is approximated by N cubic sample volumes. All sample
regimes (e Z0.8) cubes are checked to determine which mesh cell they lie within.
The solid volume fraction es of a particular mesh cell is the ratio of
3 1e Zg
bWen&Yu cD Rep 2:65 2 , 4 volume of all sample cubes that lie within that mesh cell and the
4 e dp
total volume of the cell, as given by
8  0:687  P
< 24 1 0:15 Rep if Rep o 1000, Vcube
cD
Rep
5 es 1e , 10
: 0:44 Vcell
if Rep Z 1000,
where Vcube is the volume of a sample cube and Vcell is the volume
where Rep erg9ug  vi9dp/Zg is the particle Reynolds number and
of the uid mesh cell. The number N is chosen in a range 1020 as
e is the gas volume fraction.
a compromise between accuracy of the porosity calculation and
numerical effort.
3.2. Modeling the motion of the continuous gas phase
The general algorithm of the coupled DEMCFD simulation is
shown in Fig. 2.
The gas phase is considered as continuum. The geometry of the First, the ow eld of the gas phase is resolved by the CFD
apparatus is discretized in mesh cells and the motion of the gas solver. When a stable solution is obtained, the ow eld is passed
phase is calculated using the continuity and volume-averaged to the coupling module, where the relative velocity between each
NavierStokes equations for each cell: particle and the surrounding gas is calculated in order to obtain
@ the drag force. The drag force acting on each particle is then
erg r  erg ug 0, 6
@t passed to the DEM solver which will update the particle positions
in a loop, until the end of the CFD time step is reached. The new
@
erg ug r  erg ug ug erpg r  esg Sp erg g: 7 particle positions are handed back to the coupling module, which
@t
will then update the uid cell porosities and calculate the
The gas-phase stress tensor sg is given by momentum sink term for each cell. Based on this input, the CFD
  solver iterates over the next time step until the ow eld again
sg  lg 23Zg r  ug IZg r  ug r  ug T : 8
converges to a stable solution.
The bulk viscosity lg can be set to zero for gases. Due to the
volume fraction they occupy, the particles act as obstacles and
inuence the velocity prole of the gas phase. This effect is
accounted for by adding a sink term Sp to the momentum balance
(Eq. (7)):
Z XNp
1 Vi b
Sp ug vi Drri dV: 9
Vcell Vcell i0
1e

The distribution function D distributes the reaction force


acting on the gas phase over the Eulerian grid. This force is the
equivalent reaction to the particle drag. To avoid local extrema Fig. 2. Steps in coupling the motion of discrete and continuous phase (based
of the solids concentration, the gas ow eld is resolved on a on Favier, 2009).
2344 L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355

3.4. Model describing particle contacts The effective modulus of elasticity En of the two contact
partners is given by Eq. (16), where n is the Poisson ratio. The
In case of a collision between two particles, the contact forces median radius Rn of the contact partners is a characteristic radius
in the DEM model are calculated according to a soft-sphere of the contact surface curvature.
contact model based on the theory developed by Hertz (1882) The elastic contribution of the impact energy absorbed during
for the normal impact. A no-slip approximation of the model the compression is released during the restitution phase of the
by Mindlin and Deresiewicz (1953) is used for the tangential impact and leads to the elastic force that separates the contact
component of the contact force, as proposed by Tsuji et al. (1992). partners. The absorption of kinetic energy during the impact
Consider two colliding spheres a and b with position vectors ra can be described by a restitution coefcient. The coefcient of
and rb and radii Ra and Rb (see Fig. 3). The relative velocity at the restitution is a ratio of the square root of the elastic strain energy
contact point vab is dened as follows: Wkin,R released during the restitution to the impact energy, i.e. the
initial kinetic energy Wkin (Antonyuk et al., 2010):
vab va vb Ra oa Rb ob  nab : 11 s
Here, nab is the unit vector normal to the contact plane. The Ekin,r 9vr 9
en : 18
normal component of the relative velocity vab,n is given by Ekin v
Eq. (12):
The restitution coefcient can be obtained through experi-
vab,n vab Unab nab : 12 ments, as described in the next section.
To account for visco-elastic material properties that cause
According to the Hertzian theory, the relation between the
energy dissipation, a damping factor Zn related to the normal
elastic force and the displacement is nonlinear, due to the
coefcient of restitution en is included into the DEM model, as
elliptical pressure distribution in the circular contact area
proposed by Tsuji et al. (1992):
3=2 p
Fc,n kn Udn Unab Zn vab,n : 13
Zn 2a m kn d1=4 n , 19
In the DEM model, the elastic part of the normal contact force 8
ln en
Fc,n is represented by a nonlinear spring, where the force is < p
2
if 0 oen r 1,
3=2 a p2 ln en 20
proportional to the stiffness kn and the displacement dn in the :1 if en 0:
contact, i.e. the overlap:
dn Ra Rb 9rb ra 9: 14 The effective mass mn of the contact partners is given by
Eq. (21):
The displacement-related elastic contact stiffness kn depends
1 1 1
on the geometry and the elastic properties of the colliding bodies: : 21
m ma mb
4 p
kn E R , 15 For the tangential component of the contact force a Coulomb-
3
type friction law is used:
1 1n2a 1n2b (
, 16 kt dt Zt vab,t if 9Fc,t 9 r m9Fc,n 9,
E Ea Eb Fc,t   22
m Fc,n  tab if 9Fc,t 9 4 m9Fc,n 9,
Ra Rb
R : 17 where kt, dt, Zt and m are the tangential spring stiffness, tangential
Ra Rb
displacement, tangential damping coefcient and friction coef-
cient, respectively. The tangential relative velocity vab,t is dened
by Eq. (23):
vab,t vab vab,n : 23
The tangential stiffness kt depends on the normal displace-
ment dn
p
kt 8G R dn , 24
where Gn is the equivalent shear modulus of the contact partners:
1 2va 2nb
: 25
G Ga Gb
The tangential damping coefcient Zt is dened according to
Eq. (26) (Tsuji et al., 1992):
p
Zt 2a m kt : 26

3.5. Material characterization

In order to mimic the mechanical behavior of the experimental


material (spherical g-Al2O3 granules, see Fig. 4) the input para-
meters for the contact model were determined experimentally
(Antonyuk et al., 2010). Table 1 gives an overview on the material
and contact parameters required by the DEM model.
The impact and rebound behavior of dry g-Al2O3 granules was
Fig. 3. The coordinate system used in the description of the collision model (based analyzed with the help of a free fall tester as shown in Fig. 5. The
on Deen et al., 2007). device consists of a stand (1) and a hardened steel target plate (2).
L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355 2345

Fig. 6. Segmented distributor plate of the Wurster-coater.

of restitution of 0.8 for g-Al2O3 granules was used in the simulations


Fig. 4. Picture of the g-Al2O3 granules. in this study.

Table 1
Input parameters for the DEM model. 4. Simulation conditions

Particle density, r kg/m3 1500 Simulations were performed using the commercial simulator
Shear modulus, G Pa 1  108
EDEM 2.2 (DEM Solutions Ltd., Edinburgh) coupled with Ansys
Poissons ratio, n dimensionless 0.25
Restitution coefcient, e dimensionless 0.8
Fluent 12 to describe the particle and uid dynamics in a uid bed
Static friction coefcient, m dimensionless 0.1 granulator.
Rolling friction coefcient, b dimensionless 0.01

4.1. Example 1: comparison between top-spray and


Wurster-coater conguration

For two different granulator congurations the residence time


distribution of the particles inside a biconical spray zone starting
at the nozzle tip was analyzed. The biconical shape of the spray
zone is an approximation based on the DEM analysis of droplet
deposition by Link et al. (2007).
A top-spray granulator (Fig. 1a) and a Wurster-coater (Fig. 1c)
were compared. In the Wurster-coater, air enters the equipment
through the bottom via a segmented distributor plate that
consists of three zones, as shown in Fig. 6. Below the Wurster
tube (zone 1) the uidization air velocity is higher than in the
annulus (zone 2) due to a larger porosity of the plate. In zone
3 near the wall of the apparatus a slightly higher porosity of the
plate and therefore higher gas velocity is applied to stagnant
areas. In addition, air is injected by a single phase bottom-spray
nozzle which is situated at the center inside the Wurster tube.
The top spray granulator has a distributor plate with constant
porosity over the whole diameter.
Simulations were performed with 45,000 monomodal spheri-
cal particles with a diameter of 3 mm, which corresponds to a
batch size of 0.95 kg at a particle density of 1500 kg/m3. The
geometry of the GF3-Wurster-insert (see Fig. 7a) was supplied by
Glatt Ingenieurtechnik GmbH (Weimar, Germany) and slightly
Fig. 5. Setup of the used free-fall tester.
simplied for the DEMCFD simulations (see Fig. 7b).
The outer dimensions of the top-spray granulator are identical
to those of the Wurster-coater. The uidization chamber has a
A vacuum nozzle (4) is used to hold the particle and to drop it conical shape as shown in Fig. 8, so that the gas velocity decreases
without initial rotation. During the test, a particle falls from a towards the outlet. In the Glatt GF3 insert, a two-uid nozzle
dened height h onto the horizontal target plate. The impact is (Schlick, model 970) is used. Here, the spraying angle can be
lmed at 8000 frames/s with the help of the high speed camera adjusted via the position of the nozzle cap. For the Wurster-
(3) (Imaging Solutions Motion Pro Y4 S2). By image analysis coater, the smallest spraying angle is recommended in order to
(using Matlab), the impact and rebound velocity as well as the avoid the deposition of spray liquid on the wall of the draft tube.
rebound angle and the rotational speed of the particle are 401 is a realistic value for this type of nozzle.
determined. This study focuses on the uid-dynamic behavior and neglects
According to experimental results by Antonyuk et al. (2010), who the injection of liquid binder. The gas injection by the nozzle is
found that the normal coefcient of restitution en is independent of included as a free jet. The resulting time-averaged gas ow prole
the impact velocity between 0.5 and 4.5 m/s, a constant coefcient is shown in Fig. 10.
2346 L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355

Fig. 7. Cross-sectional view of the modeled Wurster-coater: (a) Glatt GF3 geometry, (b) simplied geometry for DEM simulations and (c) mesh for CFD simulation with
12,000 tetrahedral cells.

Fig. 8. Instantaneous particle positions and velocity distributions inside the Wurster-coater (a) and the top spray granulator (b) at the simulation time t 1.4 s.
(For interpretation of the references to color in this gure, the reader is referred to the web version of this article.)

Table 2
Overview on simulation parameters for example 1.

Distributor cross section A m2 0.0254


Fluidization velocity bottom spray (zone 1) uF,W m/s 11 ( 10.2 umf)
Fluidization velocity bottom spray (zone 2) uF,ann m/s 4 ( 3.7 umf)
Fluidization velocity bottom spray (zone 3) uF,wall m/s 5.5 ( 5.1 umf)
Fluidization velocity top-spray uF,top m/s 5.5 ( 5.1 umf)
Gas injection velocity at the nozzle uspout m/s 160
Fluidization air ow rate uF m3/h 600
Atomizer air ow rate uspout m3/h 7
Gap distance below Wurster tube hgap mm 15 ( 5dp)
Total particle mass mp g 950
Normal coefcient of restitution en dimensionless 0.8
Static friction coefcient m dimensionless 0.1
Poissons ratio n dimensionless 0.25
Shear modulus G Pa 1  108
Time step for DEM simulation DtDEM s 2  10  6
Time step for CFD simulation DtCFD s 2  10  4
Total simulation time t s 5
Number of grid cells (CFD) N dimensionless 12,000

All simulation parameters are shown in Table 2. set for the gas, whereas the particlewall friction is dened by the
At the distributor plate, the gas inlet velocities were specied friction coefcient. At the outlet of the bed, atmospheric pressure
for each zone as boundary condition in the DEMCFD simulation was assumed, using a pressure outlet boundary condition. The
of the Wurster-coater (see Table 2). The boundary conditions DEM simulation time step was chosen according to the procedure
were set to match the gas ow speed through the porous plate of described in Antonyuk et al. (2011).
the Glatt GF3 granulator. Equations for the pressure drop were
used to estimate the distribution of the air ow over the plate. For 4.2. Example 2: inuence of process parameters on particle and uid
the circular bottom plate of top spray granulator, one homoge- dynamics in a Wurster-coater
nous inlet velocity uF,top was dened. The overall uidization air
ow rate is identical for both the Wurster-coater and the top To study the uid and particle dynamics of the Wurster-coater
spray granulator. No slip boundary conditions at the walls were in detail, 6 simulations with different process conditions were
L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355 2347

performed. Simulations were run with 150,000 monodisperse so that the central plane and all particles and geometry parts
spherical particles of a particle diameter dp 2 mm, which corre- behind the central plane are displayed. In the Wurster-coater the
sponds to a batch size of 0.94 kg at an average particle density of particles move at high velocity inside the draft tube, which is
1500 kg/m3. indicated by red color in Fig. 8a. The gray biconical shape above
In a series of case studies, the injection velocity uspout at the the nozzle indicates the spray zone. In the top-spray granulator
single phase nozzle (cases 13), and the gap distance hgap (see Fig. 8b) the particle motion is irregular. Particle eruptions
between the Wurster tube and the bottom plate (cases 35) were caused by collapsing bubbles can be observed at an average
varied. Material parameters were kept constant throughout the frequency of 0.3 s during the simulation.
simulations (see Table 2) as well as the height and diameter of the The number of particles which is entering the draft tube and
Wurster tube (hW 200 mm, dW 72 mm). The operating condi- the solid concentration in the spray zone can be controlled by the
tions in the case studies are shown in Table 4. adjustable gap distance between the distributor plate and the
lower border of the Wurster tube, which was set equal to ve
times the particle diameter (15 mm) for the performed simula-
tions in this example.
5. Simulation results
Due to the granulator geometry, the particles follow different
ow patterns in the Wurster-coater and the top-spray granulator.
5.1. Example 1: comparison between top-spray and
Fig. 9 shows streamlines of particles that where situated below
Wurster-coater conguration
the nozzle tip at the simulation time t 1.5 s. Their motion was
followed for 0.2 s leading to the streams drawn in Fig. 9. The
5.1.1. Particle and uid dynamics
streamlines illustrate the directed ow prole found in the
The particle and uid dynamics for both granulator congura-
Wurster-coater which is in contrast to the random motion of
tions were studied. Representative snapshots of particle positions
the particles in the top-spray granulator.
and the particle velocity distribution are shown in Fig. 8. For a
The time-averaged gas ow proles in the central vertical
better visualization the apparatus was cut vertically in the middle
plane of both granulator geometries are shown in Fig. 10. Jet
velocities larger than 14 m/s are cut off and displayed in white
Table 3
Overview on simulation parameters for simulation example 2.
color in the graphic. A relatively low injection depth h of the jet
is observed in the simulations. Approximately 3045 mm from
Fluidization velocity (zone 1) uF,W m/s 8 (10.1 umf) the nozzle tip the velocity magnitude of the gas has decreased to
Fluidization velocity (zone 2) uF,ann m/s 3 (3.8 umf) 10% of its initial value of uspout 160 m/s. In the top spray
Fluidization velocity (zone 3) uF,wall m/s 4 (5.1 umf)
granulator the direction of the injection is counter current to
Gas injection velocity at the nozzle uspout m/s 20160
Total particle mass mp g 940 the background uidization gas ow. Due to this, the jet injection
Fluidization air ow rate mF m3/h 440 depth h is smaller in the top spray granulator than in the
Atomizer air ow rate mspout m3/h 17 Wurster-coater.
Time step for DEM simulation DtDEM s 1  10  6 Due to the higher uidization velocity below the tube and the
Time step for CFD simulation DtCFD s 1  10  4
Total simulation time t s 3.0
concurrent gas injection, the air ow velocity inside the Wurster
tube is signicantly higher than in the annulus region. In the top-
spray granulator the average gas velocity is lower than in the
Table 4 Wurster conguration. A stagnant zone can be seen above the
Simulation scenarios performed describing the Wurster-coater. counter current injection nozzle.
Case uspout (m/s) hgap (mm)
5.1.2. Residence time distribution of particles in the spray zone
1 20 10 Particle wetting is a key mechanism in granulation and
2 100 10 agglomeration. To study the effect of particle and uid dynamics
3 160 10
4 160 20
on particle wetting, the residence time of the particles inside the
5 160 30 spray zone was monitored. The biconical spray zone was dened
in the simulations as shown in Fig. 11. A constant length L of

Fig. 9. Streamlines of particles leaving the spray zone for the Wurster-coater (a) and the top spray granulator (b).
2348 L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355

Fig. 10. Time-averaged uid velocity distribution in the central vertical plane of the Wurster-coater (a) and the top spray granulator (b).

inhomogeneous wetting. Due to the irregular motion in the bed,


some particles spent up to 10% of the simulation time inside the
spray zone, whereas 50% were not wetted at all.
In the top spray granulator the upper particles layers of the
uidized bed, where most particles are stagnant or fall backwards
into the dense region lie inside the spray zone. For this reason the
average particle velocity inside the spray zone in the top spray
granulator is low compared to the Wurster-coater.
For the given simulation conditions the average solids volume
fraction es inside the spray zone and therefore the average
residence time of the particles is very similar for both granulator
congurations, as illustrated by the data in Table 5. As mentioned
before, the fundamental difference between the two congura-
tions is the standard deviation of the distribution.
Fig. 11. Geometry of the spray zone.

45 mm was used, based on the jet injection depth found in the 5.2. Example 2: inuence of process parameters on particle and uid
DEM simulation of the Wurster-coater. The size of the spray zone dynamics in a Wurster-coater
is related to the injection depth of the jet according to Link et al.
(2007). The spraying angle j 401 was kept constant throughout 5.2.1. Inuence of the nozzle jet velocity on particle and uid motion
the simulations. The residence time t in the spray zone is a The simulation results show that the velocity of the air
property variable of each individual particle. At the beginning of injected via the nozzle has a strong inuence on the uid
the simulation, the residence time t is zero for all particles. The dynamics in the granulator. In Fig. 14, snapshots of the particle
particle positions are updated in the DEM simulation using small positions and their velocities are displayed. The color indicates
discrete time steps (Dt2  10-6 s). Whenever the center of a the particle velocity magnitude: blue particles move slowly
particle i is positioned inside the spray zone, the value of the (v o0.5 m/s) and red particles are fast moving (v41.5 m/s). Three
residence time variable ti of this particle is increased by Dt. simulation cases are depicted for the simulation time t 1.4 s. It
For the Wurster-coater conguration, the residence time of the can be seen that a higher spout velocity provokes higher particle
particles inside the spray zone is primarily determined by the velocities inside the Wurster tube. In the acceleration zone near
upow velocity of gas and particles in the draft tube. Once a the nozzle, the particles are concentrated to the center of the
stable circulating regime is established in the apparatus, the Wurster for high spout velocities (case 3) whereas they are more
particles are transported through the spray zone at constantly evenly distributed over the whole tube for low spout velocities
high velocity (driven by the spout). Consequently a narrow (case 1). According to the results of the simulation, the height of
residence time distribution and a homogeneous wetting of the the particle fountain increases with higher spout velocity from
particles are obtained. 430 mm above the nozzle tip in case 1 to 490 mm in case 3.
For the given simulation conditions the particles spend on Fig. 15 shows the time-averaged velocity eld of the uid in
average 22 ms/cycle inside the spray zone (see Fig. 12). The the central vertical plane for the simulation cases 1 to 3. It can be
residence time distribution obtained after 5 s of simulation shows seen that the nozzle jet has a low injection depth. Even at the
a clear peak at 22 m s and smaller peaks at multiples of this highest injection velocity of 160 m/s at the nozzle tip (case 3), the
period. These are related to particles which have passed through uid velocity decays to 14 m/s within a distance of less than
the spray zone more than once. If the simulation is continued, a 50 mm from the tip. The momentum introduced by the jet is
discrete Gaussian distribution can be expected which will con- immediately transferred to the particles in a relatively small zone
tinuously shift to the right with progressing simulation time. above the nozzle tip. In case 1, the penetration depth of the jet
For the top spray conguration a different picture is obtained injected at 20 m/s is almost invisible. In contrast to the distribu-
(see Fig. 13). Since there is no clear particle ow pattern as in the tion of the particle velocity, the gas ow eld inside the Wurster
Wurster-coater, the particle trajectories are randomly inuenced tube is hardly inuenced by the jet velocity.
by the high number of individual collisions in the uidized bed The small inuence of the atomizer velocity on the uid
(see Fig. 9). This leads to a broad residence time distribution motion in the granulator is due to the fact that even in case 3 less
of the particles in the spray zone, which corresponds to more than 2% of the total gas ow is injected via the nozzle. More than
L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355 2349

Fig. 12. Residence time distribution of particles in the spray zone (Wurster-coater).

Fig. 13. Residence time distribution of particles in the spray zone (Top spray granulator).

Table 5 particles towards the center of the tube. In the annulus the
Motion of particles in the spray zone. particles mainly slip downwards along the walls of the granulator
and the Wurster tube. At the center of the annulus a larger
Wurster Top-spray
porosity and stagnant or upwardly directed motion of the
Average solids volume fraction es % 8.9 9.7 particles can be observed. This is due to the bubbling regime of
in spray zone the uidized bed in the annulus. A comparison between cases
Average particle velocity vp,mean m/s 1.41 0.54 1 and 2 in Fig. 16 reveals that at identical ow conditions in the
annulus, the gas injection affects the bubble formation. At higher
spout velocity (case 2) larger bubbles are formed.
In slice 2, which is positioned at the upper end of the Wurster
98% of the gas enters the system via the distributor plate through tube, only few particles are present (see Fig. 17). Their movement
the bottom of the granulator. is directed upwards inside the tube and downwards in the
To assess the radial distribution of particles in the granulator, annulus. This indicates that the circulating regime is intact in a
virtual horizontal slices were cut through geometry. This was done wide uidization range. Along the surface of the Wurster tube, a
at two different heights, as indicated schematically in Fig. 7b. The deceleration of the particles can be observed (cases 1 and 2),
rst slice is situated in the lower part of the granulator, just below whereas at high spout velocity (case 3), all particles move faster
the tip of the injection nozzle. The second slice is placed at the than 1 m/s at this level. For low spout velocities as in case 1, some
upper border of the Wurster tube and allows visualizing the ow particles are moving downwards at the inside of the Wurster
conditions of particles entering the expansion zone. The thickness of tube, which indicates unwanted back-mixing.
both slices is 10 mm. With the help of a detailed analysis of the DEM data on the
Figs. 16 and 17 show instantaneous particle positions in the particle motion inside the Wurster tube, the visual observations
two slices, as seen from the top. The white bar in the annulus concerning the particle distribution in the granulator from Fig. 14 can
in Fig. 16 is the xation of the Wurster tube to the wall. Different be quantied (see Table 6). By increasing the jet velocity, the average
colors indicate the magnitude of the vertical component of the solids volume fraction inside the draft tube is reduced up to a certain
particle velocity. Red particles are transported upwards, blue limit. If the uid dynamic conditions create stable particle circulation,
particles fall downwards. A comparison can be drawn between the quantity of particles dragged into the tube is controlled by the
the case studies 13 to evaluate the inuence of the jet velocity gap distance below the Wurster tube (cases 2 and 3). If a stable
on the horizontal distribution of the particles and their vertical circulating regime is not established, the average particle residence
motion patterns. Looking at the particles inside the Wurster tube time in the Wurster tube increases (case 1). Particles that are
it can be observed that the high spout velocity in case 3 tears the stagnant or fall back downwards in the Wurster tube due to the
2350 L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355

Fig. 14. Inuence of the jet velocity on instantaneous particle positions and particle velocity distributions in the Wurster-coater at the simulation time t 1.4 s: (a) case 1,
(b) case 2 and (c) case 3.

Fig. 15. Inuence of the jet velocity on the time-averaged uid velocity distribution in the central vertical plane of the Wurster-coater: (a) case 1, (b) case 2 and (c) case 3.

Fig. 16. Inuence of the jet velocity on the instantaneous horizontal distribution of particles in slice 1 at the simulation time t 1.4 s: (a) case 1, (b) case 2 and (c) case 3.
(For interpretation of the references to color in this gure, the reader is referred to the web version of this article.)

Fig. 17. Inuence of the jet velocity on the instantaneous horizontal distribution of particles in slice 2 at the simulation time t 1.4 s: (a) case 1, (b) case 2 and (c) case 3.

insufcient gas velocity cause an increase of the solids volume volume fraction along the wall of the draft tube are formed less
fraction. often and that the particle ow direction is more uniform.
The jet injection velocity uspout shows a clear correlation with Collisions between wet particles inside the Wurster tube are
the average particle velocity vp,mean. The collision frequency fc particularly important to predict agglomeration. To build up
on the other hand decreases with increasing jet velocity. This agglomerates, the kinetic energy of the collision partners must be
observation can be related to the fact that regions of high solids dissipated by viscous and capillary forces. Increasing the relative
L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355 2351

collision velocity vc and decreasing the contact time tc will prevent green and red color. Contrary to that, particles can be found
the formation of agglomerates (Ennis et al., 1991; Antonyuk et al., spread over the whole diameter of the Wurster tube if the gap
2009). The jet velocity is a process parameter that directly affects the increases (cases 4 and 5). Near the wall of the tube they move at
average relative collision velocity and the average contact time. The low velocity, indicated by blue color. Particles are slowed down,
momentum of the jet injection can be used to control or to prevent but the circulating uidization regime is still intact. In case 6 the
agglomeration. In addition, the mechanical properties of the mate- particle concentration inside the draft tube is further increased.
rial and their distribution among the particle bulk and single The entire cross section of the Wurster tube is lled. In the center
particles have to be considered. the particles are accelerated by the jet and move at high velocity
(red color). However, along the walls particle strands slide down-
5.2.2. Inuence of the gap distance between Wurster tube and wards, which clearly indicates unwanted back-mixing. At a gap
bottom plate distance larger than 10 times the particle diameter (case 5), the
The gap distance between the distributor plate and the circulating regime is no longer intact.
Wurster tube is a parameter that strongly inuences the particle In Fig. 19, the time-averaged uid velocity eld is shown.
dynamics inside the granulator as it controls the recirculation Differences can be observed for the gas velocity inside the
of particles into the spout. Gaps of 10, 20 and 30 mm were Wurster tube, while the velocity in the rest of the granulator is
compared in the case studies 35. These distances are equal to 5, more or less equal for all cases.
10 and 15 times the particle diameter. The simulation parameters In case 3, a region of relatively low gas velocity can be seen
are given in Tables 2 and 3. below the nozzle tip. Due to the high solids volume fraction (see
By adjusting the gap distance below the Wurster tube, the Fig. 20) in this zone, the preferential path of gas is shifted towards
number of particles that are dragged into the draft tube can be the wall of the Wurster tube. In case 4 this zone of low gas
controlled. It can be seen in Fig. 18 that larger gap sizes increase velocity is much smaller and in case 5 it has disappeared. Here,
the particle concentration in the Wurster tube. Below the tube the the average gas velocity is high at the center below the nozzle and
uidization gas velocity is higher than in the annulus due to the low along the wall of the tube, which is in agreement with the
segmented distributor plate. This causes a radial ow directed elevated average solids volume fraction (Fig. 20) near the wall of
inwards (Venturi effect) through the gap between bottom plate the draft tube.
and Wurster tube. This ow is accelerated, if the gap distance is Above the nozzle a relatively homogeneous gas velocity of
small. For this reason, in case 3 the particles in the lower part of approximately 10 m/s is found inside the Wurster tube in case 3.
the Wurster tube are drawn towards the center. The increased solids volume fraction in cases 4 and 5 especially
For the economics of a coating process it is desirable to reduce along the wall of the Wurster tube (see Fig. 20) is slowing down
the total process time necessary to build up a dened coating the gas ow in this zone and tears the gas ow towards the
layer. This can be achieved by increasing the spray rate and center. Here, the average gas velocity increases from 9 m/s in case
simultaneously increasing the particle ux through the spray 3 to 12 m/s in case 5.
zone in order to avoid over-wetting and subsequent agglomera- Although the jet velocity uspout was kept constant, the injection
tion. Another option would be to increase both the spray rate and depth of the jet increases with higher Wurster gap distances
the temperature of the uidization air. ( +17% from cases 3 to 4, + 67% from cases 3 to 5, see Fig. 19). This
In case 3, particles are only present at the center of the tube. effect can be related to the higher background gas velocity in the
All particles rise at a velocity above 1 m/s which is indicated by center of the tube. The increased particle concentration creates a
preferred path of the gas ow in the center of the tube, which is
Table 6
supported by the jet injection.
Parameters characterizing the motion of particles inside the Wurster tube for In the annulus region no signicant effect of the gap distance
simulation cases 13. on particle and gas velocity and the solids volume fraction can be
observed.
Case 1 Case 2 Case 3
In the simulations the overall ow regime is less homogeneous
Average solids volume fraction es % 5.9 5.1 5.0 and tends to become unstable, if the gap distance between
Average particle velocity vp,mean m/s 0.71 0.88 0.94 bottom plate and Wurster tube is increased beyond 10 times
Collisions per particle fc s1 102 80 64 the particle diameter.
Average relative collision vc m/s 0.26 0.29 0.35 A quantitative analysis of the effect of an increased gap
velocity (particle-particle)
Average contact time tc ms 69 65 63
distance below the Wurster tube is shown in (Table 7). The
average solids volume fraction inside the draft tube increases

Fig. 18. Instantaneous particle positions and particle velocity distributions in the Wurster-coater for different gap distances at the simulation time t 1.4 s: (a) case 1,
(b) case 2 and (c) case 3.
2352 L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355

Fig. 19. Time-averaged uid velocity distribution in the central vertical plane of the Wurster-coater. Colors indicate the velocity magnitude: (a) case 1, (b) case 2 and
(c) case 3. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

Fig. 20. Time-averaged solids volume fraction es in the central vertical plane of the Wurster-coater: (a) case 1, (b) case 2 and (c) case 3.

Table 7 Roquette SA, France) in a uidized bed (Glatt GCPG 3.1). Both the top
Parameters characterizing the motion of particles in the Wurster tube for spray conguration and the Wurster-coater were tested. Pure water
simulation cases 35.
was injected for local wetting and plasticising of the particle
Case 3 Case 4 Case 5 surfaces.
In Fig. 21 the effect of different injection rates on the particle
Average solids volume fraction es % 5.1 6.6 8.8 size distribution (q3) of the agglomerates at otherwise identical
Average particle velocity vp,mean m/s 1.06 0.86 0.70 process conditions for a top spray granulator is shown. The
Collisions per particle fc s1 91 71 87
Average relative collision velocity vc m/s 0.34 0.32 0.28
injected total water mass is kept constant at 15% of the bed mass.
(particle-particle) It can be seen that the size distribution becomes narrower with
Average contact time tc ms 63 65 68 lower injection rates. High injection rates promote the formation
of large (oversize) agglomerates.
In Fig. 22 the moisture content of the agglomerates is shown as
a function of their diameter. At different rates, 15% of the bed
by 70% from cases 3 to 5. At the same time, the average particle mass of pure water were injected into the uidized bed of
velocity is reduced, which is again mainly related to the formation dextrose sirup. After the injection the uidization was stopped
of particle strands at the wall of the tube. In these regions of high and the bed material was sieved into size fractions in order to
solids volume fraction the inter-particle friction is increased, analyze the moisture content of each fraction individually by
leading to energy dissipation which is reected in a loss of kinetic differential weighing.
energy and lower average particle velocities. For the collision The results underline the issue that high spray rates favor the
frequency, no clear trend is observed. Along with the reduced generation of large, moist agglomerates while the smaller parti-
particle velocity, also the average relative collision velocity cles remain dry. The distribution of the liquid among the particle
decreases, as most collisions occur among slowly moving clus- size classes is uneven due to the insufcient mixing in the
tered particles. The higher average contact time in case 5 may granulator. At a lower injection rate (open symbols in Fig. 22), it
become relevant, if the particles tend to form agglomerates. is seen that also the small particles are wetted. Based on the
particle- and uid dynamics simulations this behavior can very
well be explained. Slower injection rates allow a better distribu-
6. Experimental results tion of the liquid among the particles. Instead of local wet spots
leading to large agglomerates, even wetting results in a homo-
Experimental investigations were performed to study the effect geneous growth rate and a narrow particle size distribution of the
of different injection rates and nozzle positions on the agglomera- agglomerates (see Fig. 21). Especially in a top spray granulation
tion behavior of spray dried dextrose sirup (DE 21, supplied by process a wet zone exists below the nozzle, which leads to the
L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355 2353

Fig. 23. Comparison of the agglomerate size distribution obtained in different two
Fig. 21. Inuence of the injection rate on the particle size distribution of granulator congurations.
agglomerated dextrose sirup.

particles in the spray zone indicating unfavorable wetting condi-


tions. This is reected by the broad size distribution of the
produced agglomerates. The Wurster-coater had a narrow resi-
dence time distribution, which matches well with the shape of
the particle size distribution of the agglomerates.

7. Conclusion

For two different granulator congurations the residence time


distribution of the particles inside a biconical spray zone at the tip
of the injection nozzle was investigated numerically on the scale
of individual particles with the help of coupled DEMCFD simula-
tions. The residence time distribution allows estimating the
homogeneity of the liquid distribution among the particles. The
results show that the Wurster-granulator is characterized by a
narrow residence time distribution, whereas the use of a top
spray granulator leads to a wide residence time distribution. Due
to the directed ow regime in the Wurster-coater, the wetting of
the particles in this type of apparatus is particularly homoge-
Fig. 22. Liquid distribution in particle size classes at different injection rates.
neous. In contrast to that, due to the unstable ow structure the
particles are inhomogeneously wetted in the top spray granulator.
formation of oversized agglomerates for a sensitive product like Based on the knowledge of the particle residence time in the
dextrose sirup. spray zone, it is a logical next step to adapt the collisional
At lower injection rate the process time to reach the xed total properties of the particles according to their residence time. This
injection mass is longer. Due to the intense heat and mass transfer is ongoing work, as well as the representation of variable particle
in the uidized bed, in this case more moisture is leaving the properties.
agglomerator with the exhaust air, which also helps to prevent Fluidized bed agglomeration experiments with amorphous
the formation of large agglomerates. water-soluble food particles underline the importance of a homo-
For a xed injection rate of 30 g/min and at identical process geneous liquid distribution among the particles and conrm the
conditions, two different granulator congurations (bottom spray ndings of the numerical simulations. If wet spots exist in the bed
with Wurster tube and top spray) were compared. The results are due to a high binder injection rate or insufcient mixing, wet
shown in Fig. 23. While a narrow particle size distribution of the oversize agglomerates are produced while large portions of the
agglomerates is obtained with the Wurster-coater, the size dis- particle bulk remain unagglomerated. The positive effect of a
tribution of the product from the top spray granulator is much Wurster tube on the distribution of the spray liquid seen in the
broader. In the top spray granulator more oversize agglomerates DEMCFD simulations was conrmed by the experiments.
are produced. Process parameters like the spout velocity and inserts like the
The experimental results of the particle size distribution Wurster tube and its distance to the bottom plate inuence the
(Fig. 23) obtained in the top spray granulator and the Wurster- uid- and particle dynamics of a Wurster-coater. This impact was
coater correspond very well with the residence time distributions analyzed through coupled DEMCFD-simulations. Based on
of the particles in the spray zone found in the numerical simula- experimental data characterizing the collision behavior of visco-
tions (Figs. 12 and 13). The simulations showed that the top spray plastic particles, the uidization regime and the functionality of a
conguration had an inhomogeneous residence time of the Wurster-coater can be predicted.
2354 L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355

Future experimental studies in a Glatt GF3 Wurster-coater n normal direction


using spherical g-Al2O3 granules will deliver results to validate p particle
results of the DEMCFD simulations. r rebound
s solid
spout spout, injection at the nozzle
Nomenclature t tangential direction
top top spray
A distributor cross section (m2) v vertical direction
cD drag coefcient (dimensionless) w Wurster
d diameter (m)
D distribution function (dimensionless)
e coefcient of restitution (dimensionless) Acknowledgments
Ekin kinetic energy (N m)
E modulus of elasticity We would like to thank Nestec S.A. for the nancial support
En effective modulus of elasticity (Pa) and Dr.-Ing. Michael Jacob from Glatt Ingenieurtechnik GmbH,
fc collision frequency (s  1) Weimar for the supply of geometry data.
F force (N)
g gravitational acceleration (m/s2) References
G shear modulus (Pa)
Gn effective shear modulus (Pa) Antonyuk, S., Heinrich, S., Deen, N., Kuipers, J.A.M., 2009. Inuence of liquid layers
h height (m) on energy absorption during particle impact. Particuology 7 (4), 245259.
Antonyuk, S., Heinrich, S., Tomas, J., Deen, N.G., van Buijtenen, M.S., Kuipers, J.A.M.,
I moment of inertia (N m)
2010. Energy absorption during compression and impact of dry elasticplastic
k displacement-related contact stiffness (N m  1.5) spherical granules. Granular Matter 12, 1547.
L length (m) Antonyuk, S., Palis, S., Heinrich, S., 2011. Breakage behaviour of agglomerates and
m mass (kg) crystals by static loading and impact. Powder Technology 206, 8898.
Deen, N.G., Van Sint Annaland, M., Van der Hoef, M.A., Kuipers, J.A.M., 2007.
nab normal unit vector (dimensionless) Review of discrete particle modelling of uidized beds. Chemical Engineering
N number specied by subscript (dimensionless) Science 62, 2844.
p pressure (Pa) Ennis, B.J., Tardos, G., Pfeffer, R., 1991. A microlevel-based characterization of
granulation phenomena. Powder Technology 65, 257272.
q3 mass density distribution (m  1) Ergun, S., 1952. Fluid ow through packed columns. Chemical Engineering
r particle position (m) Progress 48, 8994.
R radius (m) Favier, J., 2009. EDEMCFD Coupling Module for FLUENT User Guide, Revision 2A.
DEMSolutions Ltd.
Sp particle drag sink term (N m  3) Ferry, J.D., 1980. Visco-elastic Properties of Polymers 3rd edition J. Wiley & Sons,
t time (s) New York, USA.
T torque (N m) Gantt, J.A., Gatzke, E.P., 2005. High-shear granulation modelling using a discrete
element simulation approach. Powder Technology 156, 195212.
u gas velocity (m/s)
Goldschmidt, M.J.V., 2001. Hydrodynamic modelling of uidised bed spray
v particle velocity (m/s) granulation. Ph.D. Thesis, Twente University, Enschede, The Netherlands, p. 13.
vab relative velocity at the contact point (m/s) Goldschmidt, M.J.V., Kuipers, J.A.M., 2003. Discrete element modelling of uidised
bed spray granulation. Powder Technology 138, 3945.
V volume (m3)
Gryczka, O., Heinrich, S., Deen, N.G., Van Sint Annaland, M., Kuipers, J.A.M., Jacob,
Greek letters L., 2009. Characterization and CFD-modelling of the hydrodynamics
M., Morl,
of a prismatic spouted bed apparatus. Chemical Engineering Science 64 (14),
a energy dissipation factor (dimensionless) 33523375.
b interphase momentum transfer coefcient (kg m  3 s  1) L., 2003. Particle population modelling in
Heinrich, S., Peglow, M., Ihlow, M., Morl,
d displacement (m) uidized bed-spray granulationanalysis of the steady state and unsteady
behaviour. Powder Technology 130 (special issue), 154161.
e porosity (dimensionless) Hede, P.D., Bach, P., Jensen, A.D., 2009. Batch top-spray uid bed coating: scale-up
Zg dynamic viscosity of the gas phase (kg m  1 s  1) insight using dynamic heat- and mass-transfer modelling. Chemical Engineer-
Z damping factor (N s m  1) ing Science 64, 12931317.
Hemati, M., Cherif, F., Saleh, K., Pont, V., 2003. Fluidized bed coating and
lg gas phase bulk viscosity (kg m  1 s  1)
granulation: inuence of process related variables and physicochemical
m friction coefcient (dimensionless) properties on the growth kinetics. Powder Technology 130, 1834.
n Poissons ratio (dimensionless)
Hertz, H., 1882. Uber
die Beruhrung
fester elastischer Korper. die Reine
Journal fur
r density (kg/m3) und Angewandte Mathematik 92, 156171.
Hoomans, B.P.B., Kuipers, J.A.M., Briels, W.J., Van Swaaij, W.P.M., 1996. Discrete
t residence time (s) particle simulation of bubble and slug formation in a two-dimensional gas-
tg gas phase stress tensor (Pa) uidised bed: a hard-sphere approach. Chemical Engineering Science 51 (1),
j spraying angle (deg) 99118.
Hounslow, M.J., Ryall, R.L., Marshall, V.R., 1988. A discretized population balance
o angular velocity (s  1) for nucleation, growth and aggregation. A.I.Ch.E. Journal 41, 18211832.
Iveson, S.M., Litster, J.D., Hapgood, K., Ennis, B.J., 2001. Nucleation, growth and
Subscripts breakage phenomena in agitated wet granulation processes: a review. Powder
a particle a Technology 117, 339.
ann annulus Johnson, K.L., Kendall, K., Roberts, A.D., 1971. Surface energy and the contact of
elastic solids. Proceedings of the Royal Society of London A 324, 301313.
A adhesion Kafui, D.K., Thornton, C., 2008. Fully-3D DEM simulation of uidised bed spray
b particle b granulation using an exploratory surface-energy based spray zone concept.
c contact Powder Technology 184, 177188.
Karlsson, S., Rasmuson, A., van Wachem, B., Niklasson Bjorn, I., 2009. CFD
cell computational grid cell
modelling of the Wurster bed coater. A.I.Ch.E. Journal 55 (10), 25782590.
f uid Kumar, J., Peglow, M., Warnecke, G., Heinrich, S., Morl, L., 2006. Improved accuracy
F uidization and convergence of discretized population balance for aggregation: the cell
g gas average technique. Chemical Engineering Science 61, 33273342.
KuShaari, K., Pandey, P., Song, Y., Turton, R., 2006. Monte Carlo simulations to
gap gap between bottom plate and Wurster tube determine coating uniformity in a Wurster uidized bed coating process.
i particle index Powder Technology 166, 8190.
L. Fries et al. / Chemical Engineering Science 66 (2011) 23402355 2355

Link, J.M., Godlieb, W., Deen, N.G., Kuipers, J.A.M., 2007. Discrete element study of Tang, E.S.K., Wang, L., Liew, C.V., Chan, L.W., Heng, P.W.S., 2008. Drying efciency
granulation in a spout uidized bed. Chemical Engineering Science 62, 195207. and particle movement in coatingimpact on particle agglomeration and
Litster, J.D., Smit, D.J., Hounslow, M.J., 1995. Adjustable discretization balance for yield. International Journal of Pharmaceutics 350, 172180.
growth and aggregation. A.I.Ch.E. Journal 41, 591603. Terrazas-Velarde, K., Peglow, M., Tsotsas, E., 2009. Stochastic simulation of
Litster, J.D., 2003. Scaleup of wet granulation processes: science not art. Powder agglomerate formation in uidized bed spray drying: a micro-scale approach.
Technology 130, 3540. Chemical Engineering Science 64, 26312643.
Liu, L.X., Litster, J.D., 2000. Coalescence of deformable granules in wet granulation Teunou, E., Poncelet, D., 2002. Batch and continuous uid bed coatingreview and
processes. A.I.Ch.E. Journal 46 (3), 529539. state of the art. Journal of Food Engineering 53, 325340.
Mindlin, R.D., Deresiewicz, H., 1953. Elastic spheres in contact under varying Tsuji, Y., Tanaka, T., Ishida, T., 1992. Lagrangian numerical simulation of plug ow
oblique forces. Journal of Applied MechanicsTransactions of the ASME, of cohesionless particles in a horizontal pipe. Powder Technology 71, 239250.
Series E 20, 327344. Tsuji, Y., Kawaguchi, T., Tanaka, T., 1993. Discrete particle simulation of two-
Palzer, S., 2009. Inuence of material properties on the agglomeration of water- dimensional uidized bed. Powder Technology 77 (1), 7987.
soluble amorphous particles. Powder Technology 189, 318326. Turton, R., 2008. Challenges in the modelling and prediction of coating of
Patil, D.J., Van Sint Annaland, M., Kuipers, J.A.M., 2005. Critical comparison of pharmaceutical dosage forms. Powder Technology 181, 186194.
hydrodynamic models for gassolid uidized bedsPart I: bubbling gassolid van Buijtenen, M., Deen, N., Heinrich, S., Antonyuk, S., Kuipers, J.A.M., 2009.
uidized beds operated with a jet. Chemical Engineering Science 60, 5772. A discrete element study of wet particleparticle interaction during granula-
Peglow, M., Kumar, J., Hampel, R., Tsotsas, E., Heinrich, S., 2007. Towards a tion in a spout uidized bed. Canadian Journal of Chemical Engineering 87,
complete population balance model for uidized-bed spray agglomeration. 308317.
Drying Technology 25, 13211329. Wen, Y.C., Yu, Y.H., 1966. Mechanics of uidization. Chemical Engineering Progress
Rajniak, P., Stepanek, F., Dhanasekharan, K., Fan, R., Mancinelli, C., Chern, R.T., Symposium Series 62, 100111.
2009. A combined experimental and computational study of wet granulation Werner, S.R.L., Jones, J.R., Paterson, A.H.J., Archer, R.H., Pearce, D.L., 2007.
in a Wurster uid bed granulator. Powder Technology 189, 190201. Air suspension coating in the food industry: Part IImicro-level process
Tan, H.S., Salman, A.D., Hounslow, M.J., 2006. Kinetics of uidized bed melt approach. Powder Technology 171, 3445.
granulation I: the effect of process variables. Chemical Engineering Science Wurster, D.E., 1966. Particle coating process. US Patent No. 3,253,944, Wisconsin
61, 15851601. Alumni Research Foundation.

You might also like