Powder Technology 300 (2016) 2836

Contents lists available at ScienceDirect

Powder Technology

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

CFD-PBE simulation to predict particle growth in a uidized bed melt

granulation batch process
Philipp Lau, Matthias Kind
a
Karlsruhe Institute of Technology, Department of Thermal Process Engineering, Kaiserstrae 12, 76131 Karlsruhe, Germany

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

Article history: The uidized bed spray granulation process unites the steps of solid formation and product formulation in one
Received 2 September 2015 apparatus and is used to produce granulate products out of a liquid. According to different applications of
Received in revised form 3 February 2016 granules, a specic product particle size distribution is required. Thus, the ability to predict the development of
Accepted 20 February 2016
the particle size distribution is crucial for process design, optimization and scale-up. For this reason, population
Available online 23 February 2016
balance equations can be solved where the knowledge of particle size-dependent growth rates are the important
Keywords:
quantities, which are difcult to obtain by experiments.
Computational uid dynamics Using a computational uid dynamics multiphase model with one nozzle at its center, an advanced two-uid
Population balance equations model with ve uid phases is solved for short process times (~s) considering a high local and temporal resolu-
Fluidized bed spray granulation tion of the granulation process (including uid dynamics, drop deposition and energy equations) to evaluate
Industrial scale size-dependent growth rates, as a function of the current state of the process. By transferring particle growth
Particle size-dependent growth rate rates to population balance equations, the development of particle size distribution can be nally predicted for
long process times (~min).
The present work applies the predictive model to an industrial melt spray granulation process in batch mode.
Experimental data show that this model could be implemented successfully.
2016 Elsevier B.V. All rights reserved.

1. Introduction
Customizing of product properties is becoming increasingly im-
portant for the production of solid particles from liquids. Fertilizer, for
example, must be easy to dose and dust-free to ensure a uniform dis-
tribution on the eld and trouble-free lling without risk of dust
explosion.
The uidized bed spray granulation process is one possibility to pro-
duce and formulate granules out of a liquid. Thereby, a melt is atomized
in a two-component jet and sprayed into the uidized bed, which con-
sists of seed particles of the same material. Droplets deposit on the par-
ticles near the spray zone and form a lm which solidies in the cold
uidization air. Granulate products with dened particle sizes are
formed in a step-by-step manner.
Many predominant experimental research contributions on the u-
idized bed spray granulation process have been carried out regarding
uid dynamics, interactions between particles and gas [13], heat trans-
fer [4], moisture distribution [5] and growth and agglomeration kinetics
[69]. Uhlemann and Mrl [10] give a good overview of process-
relevant mechanisms.
The use and development of numerical simulation tools for design,
optimization and scale-up of uidized bed spray granulation systems
Corresponding author.
E-mail address: matthias.kind@kit.edu (M. Kind).
are becoming increasingly important due to the fast increase of comput-
er performance. The modeling of multiphase ows in uidized beds can
be divided into the EulerEuler model (two uid model: TFM) and the
discrete element method (DEM).
Using the TFM, which is proposed by Gidaspow [11], the particle
phases are treated as interpenetrating continua. Hereby, the volume
fraction of each phase is the characteristic transport quantity. Interac-
tions between particle phases can be modeled using the kinetic theory
of granular ow [1214]. The latter has its advantage in modeling a
high number of particles. Therefore, it can be used to model particulate
systems in industrial applications.
The DEM [15] calculates the trajectory of each particle using Newto-
nian equations of motion. The interactions between particles can be
modeled using spring and damper models. The gas phase is considered
as a continuous Euler phase. The DEM can consider a particle size distri-
bution (PSD), but is limited in the number of particles in the bed.
Coupling of computational uid dynamics (CFD) with DEM gives a
good possibility to model the ow behavior of particles in uidized
bed systems where each particle can be tracked individually. Fries
et al. [16] used a DEM-CFD model to describe the particle residence
time in the spray zone during uidized bed spray granulation for differ-
ent granulator applications in batch mode. Thereby, they worked with a
specic bed mass of approximately 37 kg/m2. Heine et al. [17] also used
a DEMCFD simulation to model the particle wetting in a uidized bed
agglomerator.

http://dx.doi.org/10.1016/j.powtec.2016.02.040
0032-5910/ 2016 Elsevier B.V. All rights reserved.
 P. Lau, M. Kind / Powder Technology 300 (2016) 2836
A lot of research contributions have been carried out regarding
predicting the development of the PSD during granulation using popu-
lation balance equations (PBE) for small specic bed masses and low
spray ow rates. In order to solve PBE, it is necessary to know the in-
ternal process variables, e.g. size-dependent growth rates or particle
residence times in the spray zone, as a function of the current state
of the process.
Mrl [18] and Mrl et al. [19] dened three different equations to
obtain a particle growth rate for a granulation process considering
drop deposition. Thereby, the liquid ow rate of the nozzle is divided
into different fractions depending on the diameter, surface or volume
of each particle phase. A lot of publications use the growth rate kinetics
of Mrl to model uidized bed sprays granulation processes [2023].
Hofmann et al. [24], Hampel et al. [25] and Meyer et al. [26] used a so-
called two-zone or multi-zone model to obtain the particle residence
times of different particle sizes in a drying zone together with the particle
size-independent growth rate of Mrl in the spray zone. Hampel et al. val-
idated their population balance model for particle growth with a Wurster
granulator for small bed heights and low solution spray ow rates. There-
fore, they used the residence times of Brner et al. [27], who investigated
the particle ow behavior in two-dimensional (2D) uidized bed granu-
lation chambers. Brner et al. showed that the knowledge of residence
times of particles in the spray and drying zone of the granulation chamber
is very important to predict the particle growth in granulation processes.
However, particle residence times depend on initial PSD, process param-
eters and particle properties, and cannot be obtained by experiments for
three-dimensional industrial scale granulation processes.
Li et al. [28] developed a model using CFD and PBE to predict the de-
velopment of PSDs in continuously drying uidized bed spray systems
with low solution spray ow rates (b5 kg/h) and small specic bed
masses (b 40 kg/m2). They validated their model with the experimental
results of Zank [9] and Grnewald [6].
The present work extends the model of Li et al. [28] to industrial
scale melt granulation processes with high spray ow rates (N 100 kg/h)
and high specic bed masses (N200 kg/m2) to predict the temporal
development of the product PSD and to optimize such type of processes
economically.
As will be shown in the following, with this model it is now possible
to predict the temporal development of PSD during industrial scale
granulation processes. For such predictions, it is necessary to evaluate
the internal process variables, e.g. size-dependent growth rates of the
Fig. 1. Procedural method of the developed CFDPBE model. Growth rate kinetics are obtained by a high local and temporal CFD multiphase simulation and are used to solve population
balance equations for long process times. Numerical results are compared with a single-nozzle granulation experiment in batch mode.
29
granules, as a function of the current state of the process. Particle
growth rates can be extracted from CFD for short process times (~ s).
The development of PSD can nally be predicted for long process
times (~min) by transferring particle growth rates to PBE. The extended
CFD-PBE model is applied to three different cases of a melt spray gran-
ulation process in batch mode. The numerical results are validated
with experimental data.
2. Methods and material
2.1. PBE model
In the following we consider a batch-operated industrial melt spray
granulation process with negligible dust formation. For further informa-
tion we refer to Section 2.3. For this case, the development of PSD can be
described with a simple one-dimensional PBE as follows:


 
nGran GL dp  nGran
0 1
t dp
In order to solve this equation for a given initial condition, the
knowledge of the size-dependent growth rate of the granules GL has
to be known as a function of the current state of the process. Following
the work of Li et al. [28], the particle size-dependent growth rate can be
obtained using a mass balance for each particle class i, considering the
temporal deposited mass ow rate of droplets on particle class i:

 dp;i 2 1 M_ m;i t
GL;i dp;i ; t   _ m;nozzle :
M 2
t 2 _
p Ni   dp;i Mm;nozzle
|{z}
K i t
Hereby, M _ m;i t is the mass ow rate of melt droplets depositing
on particle phase i. This unknown quantity can be obtained by a high
temporal and spatial resolution of the granulation process using an
advanced two-uid CFD multiphase model with ve uid phases. The
model is solved for short process times (~s) considering uid dynamics,
particle growth due to drop deposition and energy equations. Extracting
M_ m;i t out of CFD and, subsequently, obtaining the time-averaged infor-
mation about GL(dp), the PBE can be nally solved to predict the
30 P. Lau, M. Kind / Powder Technology 300 (2016) 2836

Table 1 interaction between particles and air for uidized beds can be described
Overview of computational multiphase models used for the ve-uid model. with the model of Gidaspow [11], who combined the models of Ergun
Euler phases 5 (air, 3 particle phases, melt) and Orning [31] and Wen and Yu [32]. In the case of the existence
Turbulence model k-Epsilon standard, dispersed of more than one particle phase, the interactions between particles
Timestep used 0.000010.0005 s are modeled with the symmetric model of Syamlal [33] and Syamlal
Solver Pressure-based, transient
et al. [34].
Gravity 9.81 m/s2
Drag particleair Gidaspow [11] The momentum exchange coefcients for granular ows include
Drag particleparticle Syamlal et al. [33] physical quantities which have to be modeled, for example, solids
Drag dropletair Schiller-Naumann [35] pressure, coefcient of restitution and radial distribution functions.
Granular viscosity Gidaspow [11]
Therefore, the so-called kinetic theory of granular ow is used.
Granular bulk viscosity Lun et al. [14]
Frictional viscosity Neglected
We refer to Gidaspow [11], Lun et al. [14], Syamlal et al. [33] and
Granular temperature Algebraic Syamlal et al. [34] for further information. An overview of general
Solids pressure Lun et al. [14] settings and multiphase model quantities used for the CFD simulation
Radial distribution Lun et al. [14] are specied in Table 1. These models were also used and validated
Packing limit 0.63; constant
with experiments by Li et al. [28].
Restitution coefcient 0.9; constant

2.2.2. Drop deposition model

development of the PSD for long process times (~ min). The PBE is
solved using MATLAB R2010b with the so-called high-resolution nite
volume method, which is a method of second order accuracy. The Van
Leer limiter was used for this. We refer to LeVeque [29] and Gunawan
et al. [30] for detailed information. The procedural method of the devel-
oped CFD-PBE model is shown in Fig. 1.
2.2. CFD model
2.2.1. Advanced two-uid model with ve uid phases
The industrial melt granulation process is simulated with CFD using
the EulerEuler two-uid model with ve Euler phases. Gidaspow [11]
showed that this is a possible method to consider a gas/solid suspension
with a large number of particles. In this method, the gas phase and
particle phase are treated as penetrated continua. The conservation
equations for mass, momentum and energy are solved for each phase con-
sidering the volume fraction i. The interactions between particles have to
be modeled because of the continuous treatment of particle phases.
There are different models to characterize the momentum exchange
between phases depending on the physical condition of the phases. The
Drop deposition due to inertia is the most signicant mechanism for
particle growth and is implemented with so-called user-dened func-
tions. Following Li et al. [28], the local drop deposition in each compu-
tational grid cell is calculated by using the model of Lfer [35] together
with the models of Bai et al. [36] and Panao and Moreria [37]. Lfer
describes the local deposition efciency in two parts, namely the im-
pingement efciency and the adhesion probability. The impingement
efciency denes the possibility of a droplet hitting the surface of a
particle. The adhesion probability denes the probability of a droplet
sticking to the particle surface. Both parts depend on the relative velocity
between the droplet and particle phase.
The overall deposition efciency in each cell can be calculated by
integrating Lfer's equation. Considering three particle phases and
the particle movement (in comparison to a packed bed), the equation
is extended by a factor which is the ratio of the relative velocity between
drops and particles to the drop velocity. Finally, we obtain Eq. (3) for
each cell, which is strongly dependent on the relative velocity between
particles and droplets, volume fractions of particles and air and particle
diameters. For N particle phases we obtain:
" #
M_ drop 1:5  h X N
p;i  vrel;drop;i
1 exp  i : 3
_ m;in
M g  vdrop dp;i
i1

The total mass of droplets deposited can be obtained by integrating

this equation over the whole chamber volume for each particle phase
by consideration of mass, momentum and energy equations. We refer
to Lfer [35], Bai et al. [36], Panao and Moreira [37] and Li [38] for
further information.

2.2.3. CFD set-up

The geometry for CFD simulation is based on a cone-shaped
single-nozzle uidized bed spray granulator in batch mode. A 2D
rotationally symmetric simulation is carried out here assuming
that a single-nozzle granulator (with a nozzle in the center) can be
approximated by rotational symmetric ow. The geometry and the
mesh are shown in Fig. 2.
The mesh is rened in the zone near the nozzle tip by the use of a
grid size function and has 13,062 grid cells and 26,381 nodes. The
mesh is adapted twice for the grid independence study. Table 2 gives
an overview of the different meshes used.

Table 2
Different types of meshes for the grid independency study.

Number of cells Number of nodes CPU time for 5 s process time

Mesh 1 13,062 26,381 ~13 h

Mesh 2 52,248 104,990 ~20 h
Fig. 2. The 2D mesh and boundary conditions of the cone-shaped single-nozzle uidized
Mesh 3 208,992 418,892 ~120 h
bed spray granulation model. The dimensions of the geometry are shown in Table 1.
 P. Lau, M. Kind / Powder Technology 300 (2016) 2836 31

Table 3
Details of the geometry and boundary conditions.

A [mm] 1530
B [mm] 225
C [mm] 400
Geometry 2D rotationally symmetric
Number of nozzles 1
Boundary conditions Perforated Plate (B)
Nozzle
Wall (A)
Pressure outlet (C)
Symmetry axis
Droplet to nozzle tip ratio ddroplet
0:005
dnozzle
Air to liquid nozzle ratio _
M
ALR M_
air;nozzle
0:625
m;nozzle
v
Fluidization number f 2mm vperfPlate 1:8
mf ;2mm

The structure of the perforated plate (holes) is not considered here,

but a homogeneous distribution of the uidization air is assumed. Fluid-
ization and atomization air enter the granulation chamber and leave it
observed through the pressure outlet boundary condition. The nozzle
is realized as a bottom spray conguration and is located in the center Fig. 3. Schematic diagram of the experimental setup of the melt granulation process in
of the granulation chamber. An additional high-resolution numerical batch mode.
simulation of the nozzle has been carried out in order to neglect its
complex inner structure. Hence, a simplied boundary condition for moments of all nodes j:
the nozzle could be developed. With this boundary condition the mass
ow rate of atomization air and the nozzle tip velocity prole is the Mi m3;i w  L3
N N i i : 4
same than in the high-resolution simulation. Consequently, it is possible X
N X X
to minimize the grid cells for the complete granulation model. Further- Mj m3; j w j  L3j
j1 j1 j1
more, the characteristic droplet diameter of the spray was estimated
using phase Doppler anemometry measurements. The results and as-
sumptions of the nozzle are not provided here. Table 3 gives an overview The numerical simulation is carried out with ANSYS Fluent 15.0.7 on
of the geometry and boundary conditions used. a NEC 128Rc-2 Server on 12 parallel processors (Intel Xeon X5690 6
Three cases of a single-nozzle batch granulation experiment are Core Westmere with 3.46 GHz). The rst 25 s is simulated to obtain
used for experimental validation. The initial bed mass for each experi- time-averaged quantities out of CFD.
mental case and CFD simulation is 40 kg. The PSD of the seed material
was obtained using a Camsizer 2006 (Retsch Technology). Using the 2.3. Experimental setup
EulerEuler model, each particle phase has a dened diameter, thus, a
high number of particle phases are needed to consider a PSD in the pro- Three cases of a single-nozzle batch granulation experiment are
cess. Due to high numerical costs using a high number of granular used for experimental validation. The dimensions of the granulation
phases, the rst six moments of the initial number density PSD q0(dp) chamber are the same as in Fig. 2.
can be reduced to three representative so-called nodes characterized Particles with a total mass of 40 kg and a dened PSD are lled into
by their weights wi and abscissas Li. Table 4 shows details of initial the granulation chamber. The uidization and atomization air are
particle size distributions for the three cases and the reduction to switched on and the hot melting is recirculated in a storage tank. The
three representative nodes, respectively. For further information, we valve to the nozzle is opened at the time t = 0 and the melt is atomized
refer to the product difference algorithm of Gordon [39]. The initial into the granulation chamber. Fig. 3 shows the experimental setup of
bed mass of each particle phase can be easily obtained from the ratio the batch granulation process.
of the third moment of the considered node i to the sum of the third Particles are sampled and analyzed with a Camsizer 2006 (Retsch
Technology) at 3 min intervals. The uidized bed is assumed to be ide-
ally mixed due to the high velocities around the nozzle. Taking this into
account, the PSD of the sampling is representative for the PSD in the
Table 4 granulation chamber. The loss of the particles sampled is not considered
Details of initial particle size distribution and the reduction to three representative nodes. in PBE, because the mass sampled can be neglected due to high bed
Case 1 Case 2 Case 3 mass and high melt ow rates. The experiment ends when the pressure
drop of the uidized bed is too high (N30 mbar). The bed mass in-
m0,0 [] 1 1 1
m0,1 [mm] 1.63257 1.197495 1.87563 creases very quickly due to the high spray ow rate. After 9 to
m0,2 [mm] 2.86929 1.60099 3.77009
m0,3 [mm] 5.34991 2.33674 8.02410 Table 5
m0,4 [mm4] 10.46570 3.66458 17.90340
Overview of K i for different cases; K i is extracted out of CFD simulation and used to obtain-
m0,5 [mm5] 21.30523 6.12153 41.57607
ed particle size-dependent growth rates for the rst 25 s.
L1 [mm] 0.9448 0.7112 1.2083
L2 [mm] 1.6667 1.4042 2.0486 Case 1 Case 2 Case 3
L3 [mm] 2.3685 2.1953 2.8395
2 7.43E03 5.85E03 9E03
w1 [] 0.2229 0.3629 0.3056 K 1 [1/m ]
w2 [] 0.5964 0.5805 0.5885 K 2 [1/m2] 9.74E03 8.12E03 11.69E03
w3 [] 0.1807 0.0566 0.1059 K 3 [1/m2] 11.1E03 9.64E03 13.66E03
M1 [kg] 1.405 2.235 2.687 GL ,1[mm/s] 9.957E04 6.419E04 1.203E03
M2 [kg] 20.645 27.514 25.225 GL ,2[mm/s] 1.325E03 8.910E04 1.563E03
M3 [kg] 17.95 10.251 12.088 GL ,3[mm/s] 1.535E03 1.058E03 1.826E03
32 P. Lau, M. Kind / Powder Technology 300 (2016) 2836

10 min, the rst particles are discharged and the experiment is n-

ished. The mass of dust in the lter was less than 0.3% of the total
product mass of the granules at the end of the process, although
the cleaning of the lter was turned off (no dust integration). This
implies that overspray and abrasion can be neglected (assuming
that these particle sizes are blown out). The number of particles is
constant during the process (no internal nucleation, breakage or ag-
glomeration exists). Consequently, only particle growth due to drop
deposition is considered for numerical simulation.
Due to experimental constraints, the mass ow rate of melt from the
nozzle had to be varied with time and had to be accounted for in the CFD
simulation. To do so, the normalized nozzle mass ow rate Y (the ratio
of the nozzle mass ow rate to a constant reference mass ow rate
M_ Ref ) is introduced:

_ m;nozzle t
M
Y t : 5
M_ Ref
Fig. 4. Measured and time-averaged normalized nozzle mass ow rate Y as a function of
time (Case 1).

Figs. 4, 5 and 6 show the ratio of mass ow rates of melt Y over time
for different cases. At the beginning, the melt mass ow rate of the noz-
zle is uctuating with very high amplitude due to the automatic regula-
tion of the nozzle. Due to the risk of crystallization of melt at the
beginning, the mass ow rate of the nozzle is low at the beginning to
avoid blocking and is increased stepwise until it reaches constant
mass ow rate. For the subsequent CFD-calculations these data were
time-averaged.

3. Results and discussion

3.1. CFD simulation

3.1.1. Grid independency study

Fig. 5. Measured and time-averaged normalized nozzle mass ow rate Y as a function of
time (Case 2).
A grid independency study is carried out here to ensure that the
numerical grid used has no inuence on the numerical results. Figs. 7
and 8 show the axial particle velocity and the particle volume fraction
in the cross-sectional area 100 mm and 300 mm above the perforated
plate as a function of distance to the nozzle (x-coordinate) for three differ-
ent meshes for a process time of 7 s. Considering the high gradient region
near the nozzle, it can be shown that numerical results are independent of
the number of grid cells. Therefore, the coarse grid (13,062 cells) is used
for further investigations.

Fig. 7. Axial particle velocity in the cross-sectional area 100 mm above the perforated plate
Fig. 6. Measured and time-averaged normalized nozzle mass ow rate Y as a function of as a function of distance to the nozzle (x-coordinate) for three different meshes for a
time (Case 3). process time of 7 s.
 P. Lau, M. Kind / Powder Technology 300 (2016) 2836 33

Fig. 8. Volume fraction of particle phase in the cross-sectional area 300 mm above the Fig. 10. Area-specic fraction of melt deposited Kias a function of time for three nodes
perforated plate as a function of distance to the nozzle (x-coordinate) for three different (Case 1).
meshes for a process time of 7 s.

3.1.2. Drop deposition model

The deposited mass ow rate on particle phase I is obtained in
each cell considering uid dynamics, drop deposition mechanism
and energy equations. With integration over the whole chamber
volume, the deposition efciency of each particle phase i is dened
as

M_ m;i t
X i t : 6
_ m;nozzle t
M

In the case that all melt droplets deposit on particle surfaces, the sum
of all Xi equals one:

X
N
X i X tot 1: 7
i1

Fig. 9 shows the temporal change of Xi for case 1 for 25 s, considering

the rst 5 s of the unsteady starting process of the nozzle ow. Due to
Fig. 11. Area-specic fraction of melt deposited Kias a function of time for three nodes
bubble formation in the uidized bed, the typical uctuating ow
(Case 2).
behavior of the particles is obtained, which directly inuences the
current drop deposition rate.

Fig. 12. Area-specic fraction of melt deposited Kias a function of time for three nodes
Fig. 9. Drop deposition efciency for each node for the rst 25 s of the process (Case 1). (Case 3).
34 P. Lau, M. Kind / Powder Technology 300 (2016) 2836

It can be shown that the highest mass of droplets is deposited on 3.2. PBE simulation and experimental validation
node 2, which characterizes the particles with a diameter of 1.66 mm.
Due to the high mass fraction of the initial bed of this particle phase, a The particle growth rate has to be actualized for dened time inter-
high surface for drop deposition in the spray zone can be provided. It vals due to the unsteady behavior of the nozzle mass ow rate. There-
can be shown that the time-averaged total deposition efciency Xtot fore, the particle growth rate is implemented as a temporal stepwise
equals one, thus, the total mass of droplets entering the chamber is function which varies depending on the nozzle ow. It is assumed
deposited on the bed surface. This leads to the conclusion that there is that the drop deposition efciency fraction of each particle phase Xi is
no overspray, due to the high bed height. Experimental results also constant over time. Solving PBE for dened process times, we obtain
show that the effect of overspray can be neglected. Figs. 14, 15 and 16, which show the experimental (dots) and numerical
Due to the different surfaces of the nodes used in the simulation, Ki is results (lines) of the temporal development of the PSD q3(dp) in the
introduced, which normalizes the deposition efciency of melt Xi to the melt spray granulation process considering drop deposition for three
total surface of particle phase i (related to the total particle surface at the cases.
beginning, respectively). It can be assumed that the total area of each The experimental results show that the width of all PSDs increases
particle phase is constant for the rst few seconds of the process. with ongoing time. Large particles grow faster because of their higher
inertia and longer residence time in the spray zone compared to small
X i t particles. This correlation can also be described by the particle size-
K i t 8 dependent growth rate obtained. The experimental results can be predict-
Ai
ed quite well with the high-resolution CFD multiphase model followed by
solving PBE.
The area-specic fraction of melt deposited Ki is given in Figs. 10, 11
and 12 for the three cases, neglecting the rst 5 s of the unsteady
starting process of the nozzle ow. Ki has to be on the same level for
each node for a growth rate which depends only on the particle surface
fractions of particle phases. Obviously, Ki still seems to be a function of
residence time in the spray zone. Larger particles have a higher inertia,
therefore, the relative velocity between droplets increases with the in-
crease of particle diameter. This implies that larger particles can collect
more droplets during ight, related to their surface.
Due to the unsteady behavior of the uid dynamics of particles, Ki
has to be averaged in time to obtain a mean particle growth rate. There-
fore, the evolution of Ki is averaged for 20 s (neglecting the rst 5 s of
unsteady starting process) with a linear function. It can be shown that
the gradient of this function is near zero, so K i (time averagedKi) is
the y axis intercept.

3.1.3. Particle size-dependent growth rate

With the knowledge of K i and the temporal evolution of mass
ow rate, Eq. (2) can be solved for each node and each case.
Table 5 gives an overview of the mean area-specic drop deposition
efciency and the particle size-dependent growth rates for the rst
25 s (considering the melt mass ow rate of Figs. 4, 5 and 6). The
particle size-dependent growth rate as a function of particle size is Fig. 14. Comparison of numerical (CFD + PBE) and experimental results of the description
given in Fig. 13. We obtain linear functions for growth rates over of the temporal development of PSD.
particle size with a rst order approximation.

Fig. 13. Particle size-dependent growth rate for three nodes (dots) for three different cases
(for the rst 25 s). A rst order approximation is carried out here (lines) to obtain a general Fig. 15. Comparison of numerical (CFD + PBE) and experimental results of the description
formulation of particle growth rate. of the temporal development of PSD (Case 2).
 P. Lau, M. Kind / Powder Technology 300 (2016) 2836 35

dnozzle Diameter of nozzle tip, [m]

dp,i Particle diameter of particle class i, [m].
f2mm Fluidization number, [].
GL,i Particle growth rate of particle class i, [].
h Height of a grid cell, [].
Ki Area-specic fraction of melt deposited on particle phase
i, [1/m2].
Ki Time-averaged, area-specic fraction of melt deposited,
[1/m2].
Li Abscissa of node i, [m].
M_ air;nozzle Mass ow rate of nozzle air, [kg/s].
M_ m;nozzle Mass ow rate of melt, [kg/s].
Mi Mass of particle phase i, [kg].
M _ drop Mass ow rate of droplets deposited, [kg/s].
M_ m;in Mass ow of droplets entering cell, [kg/s].
M_ m;i Deposited melt ow on phase i, [kg/s].
mk,i kth moment of distribution i, [mi].
N Number of particle phases, [].
Fig. 16. Comparison of numerical (CFD + PBE) and experimental results of the description Ni Number of particles in phase i, [].
of the temporal development of PSD (Case 3). nGran Particle number density, [1/m].
i Deposition efciency on phase i, [].
Smaller deviations occur when predicting particle sizes at longer p Density of particles, [kg/m3].
process times (9 min). There are two reasons for the deviations observed: q0(dp) Number distribution, [1/m].
On the one hand, the particle growth rate is approximated with a linear q3(dp) Volume distribution, [1/m].
function with three nodes. Depending on the trend of the growth rate t Time, [s].
function, the prediction for very small and very large particles is defective. vperfPlate Air velocity of perforated plate, [m/s].
On the other hand, the particle size-dependent growth rate is a function vmf,2mm Minimal uidization velocity of 2 mm particles, [m/s].
of time. Due to the change of PSD, the area-specic fraction of melt depos- vdrop Velocity of droplets, [m/s].
ited Ki also changes. In order to simulate longer process times, the particle vrel ,drop,i Relative velocity between droplets and particle phase i, [m/s].
size-dependent growth rate has to be actualized with CFD simulations wi Weight of node i, [].
more often for better agreement. This actualization has not yet been car- Xi Total droplet deposition efciency on particle phase i, [].
ried out. Nevertheless, the PSD can be predicted quite well. Xtot Total droplet deposition efciency on all particle phases, [].
Y Normalized nozzle mass ow rate, [].
4. Conclusions

The aim of this work was to extend the CFD-PBE model of Li et al.
[28] and to apply it to a batch operated industrial uidized bed melt
spray granulation processes with high spray ow rates (N 100 kg/h)
and a high bed mass (N 20 kg). Using this model, it is possible to predict
the development of the PSD in melt granulation processes and to opti-
mize such type of processes economically.
Therefore, a 2D rotationally symmetric CFD advanced two-uid
model with ve uid phases model with one nozzle at its center was de-
veloped. The ve uid phases were air, three particle phases and melt
(droplets). The model is solved for short process times (~s) considering
uid dynamics, drop deposition (implemented with user-dened func-
tions) and energy equations. The population balance was solved to pre-
dict the PSD for long process times (~min) by extracting size-dependent
and time- and space-averaged particle growth rates from CFD, which
are not experimentally available.
Comparison of numerical and experimental results shows that this
method can predict the particle growth for a specic melt granulation
case in batch mode quite well. Smaller deviations occur which can be
minimized using more nodes for CFD simulation and with actualizing
Ki more often. As a future perspective, the CFD model will be validated
with other case studies. This model can also be extended to a continuous
multi-stage granulation process for longer process times by updating
particle size-dependent growth rates in time.
Nomenclature
ALR Air to liquid nozzle ratio, [].
Ai Surface of particle phase i, [m2].
g Volume fraction of gas phase, [].
p,i Volume fraction of particle phase i, [].
i Volume fraction of Eulerian phase i, [].
ddroplet Diameter of droplets, [m].
References
[1] L.-S. Fan, C. Zhu, Principles of Gas Solid Flows, Cambridge Univ. Press, Cambridge
[u.a.], 1998.
[2] D. Geldart, Gas Fluidization Technology, Wiley, Chichester, New York, 1986.
[3] R. Jackson, The Dynamics of Fluidized Particles, Cambridge Univ. Press, Cambridge,
2000.
[4] E.-U. Schlnder, E. Tsotsas, Wrmebertragung in Festbetten, durchmischten
Schttgtern und Wirbelschichten, Thieme, Stuttgart u.a, 1988.
[5] S. Heinrich, L. Mrl, Fluidized bed spray granulationa new model for the descrip-
tion of particle wetting and of temperature and concentration distribution, Chem.
Eng. Process. 38 (1999) 635663.
[6] G. Grnewald, Staubeinbindung und Keimbildung bei der Wirbelschicht-
Sprhgranulation: Vom Prozessverstndnis zur Modellierung, Karlsruher Inst. fr
Technologie, Karlsruhe, 2010 (Diss.). (https://cuvillier.de/de/shop/publications/375).
[7] M. Peglow, J. Kumar, G. Warnecke, S. Heinrich, L. Mrl, A new technique to deter-
mine rate constants for growth and agglomeration with size- and time-dependent
nuclei formation, Chem. Eng. Sci. 61 (2006) 282292.
[8] S. Schaafsma, P. Vonk, N. Kossen, Fluid bed agglomeration with a narrow droplet
size distribution, Int. J. Pharm. 193 (2000) 175187.
[9] J. Zank, Tropfenabscheidung und Granulatwachstum bei der Wirbelschicht-
Sprhgranulation, Univ., Diss. Karlsruhe, Shaker, Aachen, 2003.
[10] H. Uhlemann, L. Mrl, Wirbelschicht-Sprhgranulation: Mit 28 Tabellen, Springer,
Berlin, 2000.
[11] D. Gidaspow, Multiphase Flow and Fluidization: Continuum and Kinetic Theory
Descriptions, Academic Press, Boston, 1994.
[12] J. Ding, D. Gidaspow, A bubbling uidization model using kinetic theory of granular
ow, AICHE J. 36 (1990) 523538.
[13] J.T. Jenkins, S.B. Savage, A theory for the rapid ow of identical, smooth, nearly
elastic, spherical particles, J. Fluid Mech. 130 (1983) 187.
[14] C.K.K. Lun, S.B. Savage, D.J. Jeffrey, N. Chepurniy, Kinetic theories for granular ow:
inelastic particles in Couette ow and slightly inelastic particles in a general
oweld, J. Fluid Mech. 140 (1984) 223.
[15] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies,
Gotechnique 29 (1979) 4765.
[16] L. Fries, S. Antonyuk, S. Heinrich, S. Palzer, DEMCFD modeling of a uidized bed
spray granulator, Chem. Eng. Sci. 66 (2011) 23402355.
[17] M. Heine, S. Antonyuk, L. Fries, G. Niederreiter, S. Heinrich, S. Palzer, Modeling of the
spray zone for particle wetting in a uidized bed, Chem. Ing. Tech. 85 (2013)
280289.
36 P. Lau, M. Kind / Powder Technology 300 (2016) 2836
[18] L. Mrl, Anwendungsmglichkeiten und Berechnung von
Wirbelschichtgranulationstrocknungsanlagen, Techn. Hochsch., Magdeburg,
1980 (Diss.).
[19] L. Mrl, S. Heinrich, M. Peglow, Fluidized bed spray granulation, Handbook of pow-
der technology, vol. 12, Elsevier, Amsterdam [u.a.] 2007, pp. 21188.
[20] S. Heinrich, M. Peglow, M. Ihlow, L. Mrl, Particle population modeling in uidized
bed-spray granulation analysis of the steady state and unsteady behavior, Powder
Technol. 130 (2003) 154161.
[21] J. Drechsler, M. Peglow, S. Heinrich, M. Ihlow, L. Mrl, Investigating the dynamic be-
haviour of uidized bed spray granulation processes applying numerical simulation
tools, Chem. Eng. Sci. 60 (2005) 38173833.
[22] M. Dosta, S. Heinrich, J. Werther, Fluidized bed spray granulation: analysis of the
system behaviour by means of dynamic owsheet simulation, Powder Technol.
204 (2010) 7182.
[23] A. Bck, S. Palis, E. Tsotsas, Model-based control of particle properties in uidised
bed spray granulation, Powder Technol. 270 (2015) 575583.
[24] T. Hoffmann, M. Peglow, E. Tsotsas, Prozessdynamik der Wirbelschichtgranulierung,
Chem. Ing. Tech. 83 (2011) 658664.
[25] N. Hampel, A. Bck, M. Peglow, E. Tsotsas, Continuous pellet coating in a wurster u-
idized bed process, Chem. Eng. Sci. 86 (2013) 8798.
[26] K. Meyer, A. Bck, E. Tsotsas, Dynamic multi-zone population balance model of par-
ticle formulation in uidized beds, Procedia Eng. 102 (2015) 14561465.
[27] M. Brner, M. Peglow, E. Tsotsas, Particle residence times in uidized bed granula-
tion equipments, Chem. Eng. Technol. 34 (2011) 11161122.
[28] Z. Li, M. Kind, G. Gruenewald, Modeling the growth kinetics of uidized-bed spray
granulation, Chem. Eng. Technol. 34 (2011) 10671075.
[29] R.J. LeVeque, Finite Volume Methods for Hyperbolic Problems, Cambridge University
Press, Cambridge, New York, 2002.
[30] R. Gunawan, I. Fusman, R.D. Braatz, High resolution algorithms for multidimensional
population balance equations, AICHE J. 50 (2004) 27382749.
[31] S. Ergun, A.A. Orning, Fluid ow through randomly packed columns and uidized
beds, Ind. Eng. Chem. 41 (1949) 11791184.
[32] C.Y. Wen, Y.H. Yu, Mechanics of uidization, Chem. Eng. Prog. Symp. Ser. (1966)
100111.
[33] M. Syamlal, W. Rogers, T.J. O'Brien, MFIX Documentation, Theory Guide, vol. 1, National
Technical Information Service, Springeld, VA, 1993.
[34] M. Syamlal, The Particleparticle Drag Term in a Multiparticle Model of Fluidization,
National Technical Information Service, Springeld, VA, 1987.
[35] F. Lfer, Staubabscheiden, Georg Thieme Verlag, Stuttgart, Germany, 1988.
[36] C.X. Bai, H. Rusche, A.D. Gosman, Modeling of gasoline spray impingement, Atomi-
zation Sprays 12 (2002) 128.
[37] M.R.O. Panao, A.L.N. Moreira, Experimental study of the ow regimes resulting from
the impact of an intermittent gasoline spray, Exp. Fluids 37 (2004) 834855.
[38] Z. Li, Fortgeschrittene Methoden zur Beschreibung der Wirbelschicht-
Sprhgranulation(KIT, Diss.) Karlsruher Institut fr Technologie, Karlsruhe, Baden,
2013 (KIT Scientic Publishing).
[39] R.G. Gordon, Error bounds in spectroscopy and nonequilibrium statistical mechanics,
J. Math. Phys. 9 (1968) 1087.