This article was downloaded by: [University of Regina]

On: 25 August 2014, At: 03:46

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Drying Technology: An International Journal

Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/ldrt20

A Three-Dimensional Numerical Study of the Gas/

Particle Interactions in an Industrial-Scale Spray Dryer
for Milk Powder Production
a a
Yan Jin & Xiao Dong Chen
a
Biotechnology and Food Engineering Group, Department of Chemical Engineering , Monash
University , Clayton Campus , Victoria , Australia
Published online: 13 Oct 2009.

To cite this article: Yan Jin & Xiao Dong Chen (2009) A Three-Dimensional Numerical Study of the Gas/Particle Interactions
in an Industrial-Scale Spray Dryer for Milk Powder Production, Drying Technology: An International Journal, 27:10, 1018-1027,
DOI: 10.1080/07373930903203588

To link to this article: http://dx.doi.org/10.1080/07373930903203588

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 Drying Technology, 27: 1018–1027, 2009
Copyright # 2009 Taylor & Francis Group, LLC
ISSN: 0737-3937 print=1532-2300 online
DOI: 10.1080/07373930903203588
A Three-Dimensional Numerical Study of the Gas/Particle
Interactions in an Industrial-Scale Spray Dryer for Milk
Powder Production
Yan Jin and Xiao Dong Chen
Biotechnology and Food Engineering Group, Department of Chemical Engineering, Monash
University, Clayton Campus, Victoria, Australia
Gas/particle interaction plays an important role in modern spray
dryers and may have influences on wall deposition, agglomeration,
Downloaded by [University of Regina] at 03:46 25 August 2014
powder degradation, etc. In the present study, the three-dimensional
(3-D) transient multiphase flow in an industrial-scale spray dryer
has been investigated using the CFD package FLUENT. The
Eulerian–Lagrangian approach and two-way coupling method were
used in the simulations. The reaction engineering approach (REA
model) for milk particles has been implemented. Some new
characteristics of the gas flow pattern and the particle behavior
(e.g., temperature–time profiles) were identified from the numerical
results; for example, the milk particles flow in such a way that
makes the central jet oscillation more nonlinear. The discrete phase
enhances the turbulence near the air/droplet inlet but damps it down-
stream. The transient turbulent flow causes significant uncertainties
in the particle tracking, which presented some challenges in simula-
tions. The study has highlighted the importance in performing 3-D
transient simulations in order to understand the industrial-scale
dryers.
Keywords Computational fluid dynamics; Fluent; Gas=particle
interaction; Industrial-scale spray dryer; Multiphase
flow; REA drying kinetics model
INTRODUCTION
Spray dryers are widely used in the dairy industry to
produce dry powders from liquids. The capacity of spray
dryers in dairy industry is from one ton to some 20 tons
of dry powder produced per hour. Therefore, spray dryers
are in general extremely energy hungry. Design of
high-performance spray dryers that are more energy effi-
cient and yet can maintain high quality is difficult because
of the complex nature of turbulent flow, the strong air–
particle interactions, and the process–product quality
interactions in the dryers.
Computational fluid dynamics (CFD) packages provide
good tools to understand the detailed flow fields in the
dryers. Crowe[1] simulated the flow field in a spray
Correspondence: Xiao Dong Chen, Biotechnology and Food
Engineering Group, Department of Chemical Engineering,
Monash University, Clayton Campus, Victoria 3800, Australia;
E-mail: dong.chen@eng.monash.edu.au
1018
dryer with a particle source-in-cell model, which is one of
the early studies on this topic. Oakley et al.[2] studied
the scale-up of spray dryers with a CFD program
(FLOW3DTM) and found that the air flow patterns were
sensitive to the values of the turbulence parameters selected
at the annular air inlet. Oakley and Bahu[3] predicted the
low-frequency oscillations of the central jet, which were
also observed in the experiments. Guo et al.[4] investigated
the swinging and swirling oscillations of the central jet at
low frequency, which they found even without inlet swirl.
Stafford et al.[5] and Kieviet et al.[6] found the instabilities
in the slow recirculation zones near the walls.
The transient air flow and the particle dynamics in spray
dryers are very complicated, and the interactions of the air
and the particles make the problem even more difficult to
be solved computationally. Langrish and Fletcher[7] sum-
marized the work on applications of CFD in spray drying
before 2001 and stated that one of objectives of the numer-
ical simulations is to reduce wall depositions rates by mod-
ifying the air flow patterns in the chamber through small
changes in the air inlet geometry for existing dryers. Huang
et al.[8,9] and Huang and Mujumdar[10–12] performed a vari-
ety of 2-D=3-D simulations of spray dryers with the com-
mercial CFD package FLUENT.[13] The numerical
results showed the effect of drying air temperature, spray
rate, air flow rate, atomizers, etc., on the performance of
the spray dryers. It has been argued that the drying cham-
ber geometry is very important in the dryer design and the
dryer’s performance. The incorporation of a fluid bed can
significantly improve the overall drying performance.
Fletcher et al.[14] indicate the three-dimensional character-
istics of transient flow in spray dryers. Kota and
Langrish[15] predicted the deposition patterns by solving
unsteady Navier-Stokes equations. One-way droplet=air
coupling was employed in the simulation, which assumed
that the droplets had no effect on the continuous phase.
Mezhericher et al.[16,17] simulated droplet drying in spray
chambers using 2-D and 3-D CFD methods. The results
show the three-dimensionality and asymmetry of the flow
 GAS=PARTICLE INTERACTIONS IN MILK POWDER PRODUCTION 1019

field, which cannot be captured by 2-D axisymmetric

simulations.
Despite these previous works, the 3-D transient simula-
tions in spray dryers are still a relatively new topic in recent
years. Only a few papers[14,17,18] have studied the 3-D tran-
sient multiphase flow in dryers with the two-way coupling
method. The relationship of the transient particle behavior
and the transient air flowing pattern still has not been fully
understood. The objective of the present study is to per-
form the transient 3-D simulations in an industrial spray
dryer. The interactions of the particles and the air in this
large dryer are studied using the numerical results.

THEORETICAL MODELS AND EQUATIONS

A Eulerian–Lagrangian approach for dispersed multi-
phase flow has been employed in the simulation because
the volume fraction of the particles in the spray dryer, except
Downloaded by [University of Regina] at 03:46 25 August 2014

the narrow atomization zone, is smaller than 10%. The tran-

sient three-dimensional multiphase (and multi-species) flow
is simulated (the geometry of the dryer is shown in Fig. 1).
The gas phase is treated as a compressible ideal gas. The gov-
erning equations of the gas phase are compressible
Navier-Stokes equations. A k–e model was used to predict
the turbulent flow of the gas phase. Two-way coupling
method was used to consider the mass, momentum, and heat
transfer between the gas phase and the discrete phase. The
effect of the turbulent flow on the discrete phase was modeled
by a discrete random walk (DRW) stochastic tracking
model. The unsteady flow field is simulated with a dual-time-
step method. The details of the governing equations, the tur-
bulence model, and the numerical method can be found in
FLUENT 6.3 User’s Guide.[13]
The reaction engineering approach[19] was implemented
to the CFD package FLUENT to model the drying process
of milk particles. Woo et al.[20] implemented the reaction
engineering approach (REA) model and studied the 2-D
steady flow of a pilot-scale dryer. The advantage of the
REA model was implied by the comparison with the charac-
teristic drying rate curve (CDRC) model. The REA model is
described briefly here; hence, this is actually the first attempt
of applying the REA model to a transient 3-D simulation.
Basically, the rate of drying is firstly expressed as:

dmp
¼
hm Ap ðqn;s
qn;b Þ ð1Þ
dt

where mp is the particle mass, hm is the mass transfer coeffi-

cient, qn,s is the vapor concentration at the particle–gas inter-
face, and qn,b is the vapor concentration in the surrounding
gas (kg  m
3). The concentration of vapor in the surround-
ing is calculated as:
FIG. 1. Geometrical diagram of the compact disc dryer where
p H  15 m and D  8 m (a: spray dryer, b: mesh, c: flowing pattern in
qn;b ¼ Uwater ð2Þ
RTb the fluidized bed).
 1020 JIN AND CHEN
where Uwater is the local mole fraction of water, p is the local
absolute pressure, and Tb is the local bulk temperature in the
gas. hm in Eq. (1) is calculated from the Sherwood number
correlation of the Ranz and Marshall type:[21,22]
hm dp 1=2
ShAB ¼ ¼ 2:0 þ 0:6Red Sc1=3 ð3Þ
Dv;air
where dp is the particle diameter, Dv,air is the diffusivity of
qd ju
uj
vapor in air, Red ¼ p lp is the Reynolds number, and
l and the REA drying model was implemented in the
Sc ¼ qDv;air is the Schmidt number. l, q, and u are the dynamic
viscosity, density, and velocity of the gas phase, and up is the
particle velocity.
qv,s is an empirical parameter for solute=solids contain-
ing particles, which is time dependent during the drying
process as a result of the increasing solid content at the sur-
face. The REA approach[19,23,24] is used to calculate qv,s:
Downloaded by [University of Regina] at 03:46 25 August 2014
qn;s ¼ uqn;sat ðTs Þ ð4Þ
where u is a fractionality coefficient relative to the moisture
content of the dried product at the interface and Ts is the
interface temperature. In effect, u is the relative humidity
at the interface of the droplet=particle and the drying
air. The reaction engineering approach assumes that
evaporation is an ‘‘activation’’ process having to overcome
an ‘‘energy barrier,’’ whereas condensation or adsorption
is not. It uses a simple relationship to express u (approxi-
mately) as a function of water content and temperature
as follows:[23]


DEn
u ¼ exp
ð5Þ
RTp
where DEn may be viewed as a ‘‘correction factor’’ of the
apparent activation energy for drying because of the
increasing difficulty of removing water at low moisture
content levels. For 50 wt% skim milk as the feed, DEv
is computed by the formula based on a strategy
recommended by Chen:[24]
DEv
¼
0:1291ðX
Xb Þ3 þ 0:6667ðX
Xb Þ2
DEv;b

1:2803ðX
Xb Þ þ 0:9891 ð6Þ
where X is the particle moisture content and Xb is the
equilibrium moisture content. The equilibrium activation
energy DEv,b was obtained using the gas bulk temperature
Tb and gas relative humidity RHb:
! of the particles are reflected after colliding with the wall.
qv;b ðTb Þ
DEv;b ¼
RTb lnðRHb Þ ¼
RTb ln ð7Þ
qv;sat ðTb Þ
where qv,sat(Tb) is the saturated vapor concentration.
The interface temperature Ts in Eq. (4) is approximately
the particle temperature Tp because the Biot number is
small enough.[25] Then the energy equation of the particles
can be written as follows:
dTp dmp
mp cp ¼ hc ðT
Tp ÞAp þ hfg ð8Þ
dt dt
where cp, hfg, and hc are the heat capacity, the latent heat,
and the heat transfer coefficient of the milk particles.
The multiphase flow was solved with FLUENT 6.3.26
FLUENT package through the user-defined function
(UDF).
DESCRIPTION OF THE TEST CASE
The structure of the spray dryer simulated in this study
is shown in Fig. 1. The dryer, which is 15 m tall and 8 m
wide, can produce 9 tons of dried milk particles per hour.
This is a dryer in one commercial company. The geometri-
cal diagram of the dryer is shown in Fig. 1a. The inlet
parameters are given in Table 1. The initial droplets are
fed into the dryer from the rotating disc. The initial droplet
size is theoretically distributed according to the
Rosin-Rammler method. The mean droplet size is 150 mm
(estimated from the industrial result of mean particle size
[dried] of about 110 mm and with the shrinkage factor of
0.73 according to Lin and Chen[26]). The maximum and
the minimum droplet size are 250 and 30 mm. The initial
velocity of all the particles is in radial direction, whose
magnitude is 50 m  s
1. The Rosin-Rammler spread param-
eter is assumed to be 3.05. Twenty groups of particles
are employed to consider the diameter distribution of the
droplets. The particles are released every time step.
The wall is made of stainless steel. The ‘‘reflecting wall’’
boundary condition was employed at the dryer walls,
which assumes that all the particles are reflected elastically
by the wall. After reflected by the wall, the particle velocity
is computed by:
* *
V n;out ¼
V n;in
ð9Þ
* *
V t;out ¼ V t;in
* *
where V n;out and V t;out are the particles’ velocity compo-
nents, which are normal and tangential to the wall after
the reflection. In most normal operations, in dairy powder
production at least, the deposit layer nearly remains a
constant after hours of running, which means that most
Therefore, the ‘‘reflecting wall’’ is physically more accurate
than the ‘‘escaping wall’’ in the operation of the spray
dryers.
There are 58,700 quadrilateral cells in the computational
domain in total. Forty-five grid points are given in axial
 GAS=PARTICLE INTERACTIONS IN MILK POWDER PRODUCTION
TABLE 1
Dryer operating parameters
Droplet inlet Flow rate (ton=h)
Solid mass fraction (%)
Temperature ( C)
Air inlet Air flow rate (m3=min)
Air temperature ( C)
Absolute humidity kg H2O=(kg dry air)
Fluidized bed inlet Air flow rate (m3=min)
Air temperature ( C)
Absolute humidity kg H2O=(kg dry air)
Air outlet pipe Ambient pressure (Pa)
direction and 65 grid points are given in the circumferential
Downloaded by [University of Regina] at 03:46 25 August 2014
direction. The grid points in the radial direction are distrib-
uted differently at different cross sections. Seven grid
points in the radial direction are given at the air inlet,
and 20 grid points in the radial direction are given in the
spray chamber. The mesh resolution is the highest near
the air=droplet inlet, whose mesh spacing is roughly
0.04 m. The sensitivity of the mesh resolution was studied
by Jin and Chen[27] that indicates the independency of the
numerical results on the mesh resolution. The same mesh
resolution is used in the present study. It can also be seen
from the present numerical results that the mesh resolution
is high enough to capture the main vortical structures in the
chamber. The characteristics of the flow field, such as the
oscillation of the central jet, were also captured success-
fully. The global mesh and the mesh near the air inlet are
shown in Fig. 1b.
Although the present study focuses on the multiphase
flow in the spray dryer, the fluidized bed was also added
to the computational domain so that a more realistic
velocity profile can be given at the bottom of the spray
dryer. Figure 1c shows the streamlines of the gas phase
from the fluidized bed inlet, which are colored with the
streamline ID. The figure indicates the necessity of includ-
ing the fluidized bed into the computational domain. It can
be noticed that the streamlines that are released from the
inlet of the fluidized bed usually direct to the wall before
entering the spray dryer. Some small vortical structure
can be found at the corner. The flow field near the dryer
bottom is so complicated that it is almost impossible to
address the inlet profile without simulating the dryer and
the fluidized bed as one entity. Thus, the fluidized bed
and the spray dryer were simulated together in the present
study. The mass flow rate, temperature, and humidity of
the fluidizing air are given in Table 1. The air velocity is
normal to the fluidized bed inlet. The inlet area for fluid-
ized bed is 18 m2. The detailed porous structure of the fluid-
ized bed inlet was neglected in the present study.
1021
Case 1 Case 2 Case 3 Case 4
18 10.8 3.6 0
48%
78
1650
210
0.01
480
90
0.01

3210
The chamber wall is assumed to be adiabatic. The angle
between the velocity of the inlet hot air and the vertical axis
is 20 . The magnitude of the swirling velocity is 15% of that
of the axial velocity. k and e at the air inlet are 0.03 m2=s2
and 0.2 m2=s3. The numerical results show that the influ-
ence of the air inlet turbulence is quite small. The gravity
is 9.81 m2=s, which is in the vertical axial direction. The
droplet–droplet interaction is not considered in the present
study. The DRW stochastic tracking model is utilized to
consider the effect of the turbulent flow on the particles.
The time scale constant in the model is set to be 0.15, which
is recommended by the FLUENT 6.3 User’s Guide.[13]
The second-order up-wind scheme is used in space dis-
cretization. First-order implicit solver is used in time
discretization.
Four cases were simulated to understand the effect of
the milk droplet feeding rate. The gas phase is two-way
coupled with the discrete phase in the first three cases.
The pure gas phase flow was simulated in the fourth case.
The flow field is distributed uniformly before the beginning
of the simulation, where the initial temperature, pressure,
velocity, and absolute humidity are 20 C, 0 Pa, 0 m=s,
and 0.01 kg H2O=(kg dry air). The time step used in the
simulation is 0.01 s, which is verified to be small enough
according to the numerical results. The maximum number
of subiterations is 20. The subiterations are skipped when
the solution is converged. The convergence is decided if
the residues are smaller than 0.001.
RESULTS AND DISCUSSION
Some numerical results in Case 1 was compared with the
experimental data, which are reliable results taken from
plant trials by the research group led by Chen. The pre-
dicted exhaust air temperature is 73 C, which is close to
the plant data. Figure 2 shows the agreement of the aver-
aged wall temperature predicted by the simulation and
the experimental data by Chen et al.[28] The experimental
 1022 JIN AND CHEN

FIG. 2. Comparison of the predicted and the measured temperature

at the wall, Case 1.
Downloaded by [University of Regina] at 03:46 25 August 2014

data were measured at two different locations of the side

wall. Two experimental data differ from each other because
the wall temperature is strongly time-dependent and
unsymmetrical.
Figure 3 shows the flowing patterns of the gas phase at
190 s (Case 1). The 3-D streamlines are colored according
to the time since they leave the spray dryer air inlet
(Fig. 3a). Some of the streamlines leave the dryer directly
through the air exhaust, whereas others reenter the main
flow domain. After touching the ceiling wall, the stream-
lines swirl around the central jet and generate the recircula-
tion flow. The structure of the recirculation flow can be
seen from the 2-D streamlines in Figs. 3b and 3c, in which
the axial velocity contours were drawn together. A recircu-
lation center near the jet can be seen in Fig. 3b. The movie
made of the numerical results shows that the stream-wise
and span-wise recirculation flow are strongly time-
dependent. It can be seen that the stream-wise vortexes
oscillate with the movement of the central jet, while the
span-wise vortex rotates along the axis.
To understand the central jet flow, the iso-surfaces
where Vx ¼ 12 m  s
1 were generated from the numerical
results of Case 1 and Case 4 (Fig. 4). The iso-surfaces are FIG. 3. Simulated flowing patterns of the gas phase (a: 3-D streamlines
colored according to the air temperature. Rotation of the colored by the time; b: 2-D streamlines and axial velocity contours at the
iso-surfaces can be observed in both test cases. The air section [x ¼ 4 m]; c: 2-D streamlines and axial velocity contours at the
section [y ¼ 0 m]), Case 1, t ¼ 190 s.
velocity in Case 1 decays faster because of the damping
effect by the particles. Obviously, the particles also cause
the air temperature to drop faster. Figure 5 shows the where y and z are the coordinates in the section x ¼ 2.
history of the mass center of the section x ¼ 2 m (2 m from Many numerical results of the one-phase flow show that
the top). The coordinates of the mass center (ym, zm) are the central jet oscillates in low frequency.[3,29] This phe-
computed by nomenon is also captured by the present simulation
(Fig. 5a). Figure 5b shows that the central jet not only
R R rotates around the axis but also oscillates in a smaller
x¼2 y  qVx dA z  qVx dA
ym ¼ R ; zm ¼ x¼2
R ð10Þ length scale. Two frequencies can be found in the spectrum
x¼2 qVx dA x¼2 qVx dA of ym (solid line in Fig. 6). The lower frequency (about
 GAS=PARTICLE INTERACTIONS IN MILK POWDER PRODUCTION 1023
Downloaded by [University of Regina] at 03:46 25 August 2014

FIG. 4. Evolution of the iso-surface where Vx ¼ 12 m  s
1, (a)–(c)

one-phase flow (air only, Case 4), (e)–(f) two way coupling (Case 1).

0.03 Hz) represents the rotating central jet around the

vertical axis and the higher frequency (about 0.18 Hz) FIG. 5. Oscillation of the air-phase flow field. (a) Mass center history,
(b) pattern of the mass center (section x ¼ 2 m, air only, Case 4).
represents the smaller oscillation pattern in Fig. 5b.
Langrish et al.[29] also indicate the double-frequency oscil-
lation at high inlet swirl angle.
The oscillating history of the mass center (Fig. 7) shows
that the particles’ effect makes the oscillation more non-
linear. The low-frequency oscillation can still be observed,
whereas the frequency decreases to about 0.26. Filtering
the oscillation whose frequency is higher than 0.03, ym’s
low-pass oscillation in Case 4 reaches its first peak value
at about 60 s (see the doted line in Fig. 5a). The dotted line
in Fig. 7a shows that the first peak value is delayed to
about 100 s in Case 1. The high frequency of the oscillation
drops with the increasing of inlet droplet flow rate. The
amplitudes of the second peak frequency are about 0.15,
0.16, and 0.18 Hz in Case 2, Case 3, and Case 4, respec-
tively. In Case 1, the second peak frequency cannot be
observed clearly any longer. FIG. 6. Spectrum of ym (section x ¼ 2 m).
 1024
Downloaded by [University of Regina] at 03:46 25 August 2014
FIG. 7. Oscillation of the air-phase flow field. (a) Mass center history,
(b) pattern of the mass center positioning (section x ¼ 2 m, air=particle
two-way coupling, Case 4).
Figure 8 shows the turbulent kinetic k distribution at the
line x ¼ 2 m, y ¼ 0 m and x ¼ 10 m, y ¼ 0 m, which are
obtained at 190 s. Two peak values of the turbulent kinetic
energy k, as a result of the shear layer produced by the cen-
tral jet, can be found at the points r 
0.8 m and
r  1.1 m. The interesting discovery of the present study
is that there is another peak of k near the jet center. The
values of the central peak in Cases 1–4 are 17.0, 14.2,
1.6, and 4.0 (m2s
2), respectively. This means that the
turbulence near the jet center can be strengthened by the
particles if the inlet droplet flow rate is high, whereas it
can be damped if the inlet liquid feed is small. This
phenomenon indicates again the importance of simulating
industrial-size spray dryers directly, which have much
higher droplet feeding rate than in the small-scale dryers.
JIN AND CHEN
FIG. 8. Influence of the dispersed phase on the turbulent kinetic energy
(a: x ¼ 2 m; b: x ¼ 10 m, t ¼ 190 s).
If the section is far away from the air=droplet inlet
(Fig. 8b), the turbulent kinetic energy k is clearly decayed
due to the presence of the particles. There are two peaks
of k in Cases 3 and 4, whereas only one remains in
Cases 1 and 2. Moreover, the turbulence near the side wall
is enhanced by the particles in Case 1.
The dispersion of particles due to the turbulent flow is
considered using the stochastic tracking model. Figure 9
indicates the importance of this turbulence effect.
Figures 9a and 9c are from the numerical simulations with
the stochastic tracking model, whereas Figs. 9b and 9d are
without the stochastic tracking model. The particles were
colored with their residence time in the dryer. The red
particles have the longest residence time and the blue
particles have the shortest residence time. An ‘‘inverse
mushroom’’-like structure of the particles was generated
at 1 s (Fig. 9a). Without considering turbulent effect, fewer
particles are swirled back to the dryer. The particle move-
ment history is significantly influenced by the turbulent
 GAS=PARTICLE INTERACTIONS IN MILK POWDER PRODUCTION
Downloaded by [University of Regina] at 03:46 25 August 2014
FIG. 9. Effect of the turbulence on the particle behavior in the dryer,
Case 1.
flow, which is important in studying the particle residence
time, moisture content, and the insolubility index.[7]
Figure 10 displays all the particles having the same
initial particle diameter (215 mm) and the same initial veloc-
ity direction in the dryer at 140 s. The x coordinate is the
instantaneous particle residence time since the particles
enter the dryer. Due to the transient gas flow, especially
the turbulent effect, the particles travel in quite different
routes. The particle velocity can vary between 0 and
15 m  s
1, and the moisture content varies between 0 and
0.18 wt% on dry basis. The average residence time of the
particles that leave the dryer from the particle exit is about
51 s. Figure 10d shows the temperature of the particles that
have the same size (215 mm). Figures 10e and 10f indicate
the particle temperature from 0 to 1 s and from 0 to 0.1 s.
Results of two time steps (0.01 and 0.001 s) are compared,
1025
which shows that the results are independent of time step
if it is smaller than 0.01 s. The particles reach their peak
temperature at about 0.06 s. Differing from the previous
studies,[20] the particle temperature obtained from the
present study varies between 50 and 80 C after 2 s. The rea-
son for the difference lies in the fact that the present study
considers the turbulence effect on the particles that leads to
the uncertainty of the particle tracking.
The particle tracks are fixed when t < 0.6 s. The initial
drop of the temperature can be observed as the droplets
enter the dryer because of the evaporation under the influ-
ence of the high mass transfer coefficient (Fig. 10f). This
phenomenon is also observed by the experiment.[30] It is
known that most of the water in the liquid droplets is
removed soon after they enter the dryer,[27] but the varia-
tion of the particle histories after 0.6 s is still important
for its effect on the powder quality such as insolubility
index.[7]
It is worth noting that the 3-D simulation predicts much
stronger heat=mass transfer near the droplet inlet than the
1-D model.[31] The difference can be seen from the defini-
qd ju
uj
tion of the Reynolds number, Red ¼ p lp , which is used
to compute the heat and mass transfer coefficient. The
cross-section averaged air velocity, u ¼ m _ =ðqAÞ, is used in
the 1-D model, where m _ is the mass flow rate of the hot
air and A is the cross-section area. However, the local air
velocity in 3-D simulation can be much larger than the
average velocity. For example, the inlet air velocity at the
air inlet is about 27 m=s, which is about 10 times the aver-
age velocity. Thus, the 3-D simulation predicts much stron-
ger heat and mass transfer near the air=droplet inlet.
This shows, again, the importance of performing 3-D
transient simulation in order to understand the spatial
effect on particles. Even at 3-D, the steady-state simulation
can only give one sort of particle history, but the real par-
ticles move in a wide range of possibilities. The statistical
study of the transient simulation is significant to quantify
the particle variables in the dryer.
CONCLUSIONS
Great effort has been done in simulating 2-D=3-D
transient=steady multiphase flow fields in spray dryers in
recent years.[8–12,14,16,17,20] Compared with the previous
study, the present article focuses on the air=particle interac-
tion in an industrial-scale dryer, on which there is still very
few 3-D transient simulations. The REA model has been
used in 3-D simulations for the first time. Moreover, a
reflecting wall boundary condition is used in the simula-
tion, which is more accurate in the real plant operation
than the escaping wall boundary condition.
From the numerical results, some characteristics that
are indicated by previous researchers are also captured,
such as the low-frequency oscillation of the central jet
and the three-dimensionality of the flow field. Some new
 1026
Downloaded by [University of Regina] at 03:46 25 August 2014
FIG. 10. The influence of the turbulent gas flow on the particle variables history (a: time dependent velocity; b: time dependent moisture content;
c: particle mass, d, e, f: particle temperature), Case 1.
characteristics in the spray dryer are argued: It is found
that the particles make the central jet oscillate more nonli-
nearly. The second frequency of the oscillation drops with
the increasing droplet feeding rate. The turbulence kinetic
energy can be strengthened by the particles near the air=
liquid inlet for large feeding rate, whereas it is damped in
the main dryer chamber. The particle dispersion caused
by the turbulent flow has a rather important effect on the
particle behaviors. The particle behavior is not certain
JIN AND CHEN
in the dryer but can vary in a wide range due to the turbu-
lent effect. A statistical method is necessary in studying
particle behavior because of the uncertainty of the particle
tracking.
ACKNOWLEDGEMENT
The project is sponsored by the Gardiner Foundation
and Dairy Innovation Australia Ltd., Melbourne,
Australia. We also thank Dr. Kamlesh Patel in the same
 GAS=PARTICLE INTERACTIONS IN MILK POWDER PRODUCTION
research group for his help in discussing the properties of
the drying model and the properties of the milk particles.
REFERENCES
1. Crowe, C.T. Modelling spray–air contact in spray-drying systems.
In Advances in Drying; Mujumdar, A.S., Ed.; Hemisphere: New York,
1980; 63–99.
2. Oakley, D.E.; Bahu, R.E.; Reay, D. The aerodynamics of cocurrent
spray dryers. In Proceedings of the Sixth International Drying Sympo-
sium IDS; Roques, M., Ed.; Versailles, France, 1988; 373–378.
3. Oakley, D.E.; Bahu, R.E. Spray=Gas Mixing Behavior within Spray
Dryers; Mujumdar, A.S.; Filkova, I. (Eds.); Drying, Elsevier: Amster-
dam, 1991; 303–313.
4. Guo, B.; Langrish, T.A.G.; Fletcher, D.F. Time-dependent simulation
of turbulent flows in axisymmetric sudden expansions. In Proceeding
of the 13th Australasian Fluid Mechanics Conference; Thompson,
M.C.; Hourigan, K., Eds.; Melbourne, Australia, 1998; 283–286.
5. Stafford, R.A.; Fauroux, O.; Glass, D.H. Flow visualization and
instantaneous velocity measurements of spray dryer gas and spray
flows using particle image velocimetry. In Proceedings of the Tenth
Downloaded by [University of Regina] at 03:46 25 August 2014
International Drying Symposium (IDS ’96); Strumillo, C.; Mujumdar,
A.S., Eds.; Krakow, Poland, 1996; 555–562.
6. Kieviet, F.G.; Van Raaij, J.; De Moor, P.P.E.A.; Kerkhof, P.J.A.M.
Measurement and modeling of the air flow pattern in a pilot-plant
spray dryer. Transactions of the Institution of Chemical Engineers
1997, 75 (A), 321–328.
7. Langrish, T.A.G.; Fletcher, D.F. Spray drying of food ingredients and
applications of CFD in spray drying. Chemical Engineering and
Processing 2001, 40, 345–354.
8. Huang, L.X.; Kumar, K.; Mujumdar, A.S. Simulation of a spray
dryer fitted with a rotary disk atomizer using a three-dimensional
computational fluid dynamic model. Drying Technology 2004, 22
(6), 1489–1515.
9. Huang, L.X.; Passos, M.L.; Kumar, K.; Mujumdar, A.S.
A three-dimensional simulation of a spray dryer fitted with a rotary
atomizer. Drying Technology 2005, 23, 1859–1873.
10. Huang, L.X.; Mujumdar, A.S. Development of a new innovative
conceptual design for horizontal spray dryer via mathematical
modeling. Drying Technology 2005, 23, 1169–1187.
11. Huang, L.X.; Mujumdar, A.S. Numerical study of two-stage horizon-
tal spray dryers using computational fluid dynamics. Drying Technol-
ogy 2006, 24, 727–733.
12. Huang, L.X.; Mujumdar, A.S. Simulation of an industrial spray dryer
and prediction of off-design performance. Drying Technology 2007,
25, 703–714.
13. FLUENT 6.3 User’s Guide; Fluent Inc., 2006.
14. Fletcher, D.F.; Guo, B.; Harvie, D.; Langrish, T.A.G.; Nijdam, J.J.;
Williams, J. What is important in the simulation of spray dryer
performance and how do current CFD models perform? Applied
Mathematical Modeling 2006, 30 (11), 1281–1292.
15. Kota, K.; Langrish, T. Prediction of deposition patterns in a
pilot-scale spray dryer using computational fluid dynamics (CFD)
1027
simulations. Chemical Product and Process Modeling 2007, 2 (3),
1–18.
16. Mezhericher, M.; Levy, A.; Borde, I. Droplet-droplet interactions
in spray drying using 2D computational fluid dynamics. Drying
Technology 2008, 26 (3), 265–282.
17. Mezhericher, M.; Levy, A.; Borde, I. Modelling of droplet drying
in spray chambers using 2D and 3D computational fluid dynamics.
Drying Technology 2009, 27 (3), 359–370.
18. Verdurmen, R.E.M.; Van Houwelingen, G.; Gunsing, M.;
Verschueren, M.; Straatsma, J. Agglomeration in spray drying
installations (The EDECAD Project): Stickiness measurements and
simulation results. Drying Technology 2006, 24, 721–726.
19. Chen, X.D.; Lin, S.X.Q. Air drying of milk particle under constant
and time-dependent conditions. AIChE Journal 2005, 51 (6), 1789–
1799.
20. Woo, M.W.; Daud, W.R.W.; Mujumdar, A.S.; Wu, Z.H.; Talib,
M.Z.M.; Tasirin, S.M. CFD evaluation of droplet drying models in
a spray dryer fitted with a rotary atomizer. Drying Technology
2008, 26 (10), 1180–1198.
21. Ranz, W.E.; Marshall, W.R. Evaporation from drops, Part I.
Chemical Engineering Progress 1952, 48 (3), 141–146.
22. Ranz, W.E.; Marshall, W.R. Evaporation from drops, Part II.
Chemical Engineering Progress 1952, 48 (4), 173–180.
23. Chen, X.D.; Xie, G.Z. Fingerprints of the drying behavior of particu-
late or thin layer food materials established using a reaction engineer-
ing model. Transactions of the Institution of Chemical Engineers
1997, 75 (C), 213–222.
24. Chen, X.D. The basics of a reaction engineering approach to
modeling air-drying of small droplets or thin layer material. Drying
Technology 2008, 26 (6), 627–639.
25. Chen, X.D.; Peng, X.F. Modified Biot number in the context of air
drying of small moist porous objects. Drying Technology 2004, 23,
83–103.
26. Lin, S.X.Q.; Chen, X.D. Changes in milk droplet diameter during
drying under constant drying conditions investigated using the
glass-filament method. Transactions of the Institution of Chemical
Engineers 2004, 82 (C), 213–218.
27. Jin, Y.; Chen, X.D. Numerical study of the drying process of different
sized particles in an industrial-scale spray dryer. Drying Technology
2009, 27, 371–381.
28. Chen, X.D.; Lake, R.; Jebson, S. Study of milk powder deposition on
a large industrial dryer. Transactions of the Institution of Chemical
Engineers 1993, 71 (C), 180–186.
29. Langrish, T.A.G.; Williams, J.; Fletcher, D.F. Simulation of the
effects of inlet swirl on gas flow patterns in a pilot-scale spray dryer.
Transactions of the Institution of Chemical Engineers 2004, 82
(A7), 821–833.
30. Lin, S.X.Q.; Chen, X.D. Improving the glass-filament method
for accurate measurement of drying kinetics of liquid droplets.
Transactions of the Institution of Chemical Engineers 2002, 80
(A), 401–410.
31. Patel, K.C.; Chen, X.D. Drying of aqueous lactose solutions in a single
stream dryer. Food and Bioproducts Processing 2008, 86, 185–197.