Sei sulla pagina 1di 7

Advanced Powder Technology 20 (2009) 220–226

Contents lists available at ScienceDirect

Advanced Powder Technology


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

Original Research Paper

A distributed parameter model for particles in the spray drying process


S. Wang *, T.A.G. Langrish
School of Chemical and Biomolecular Engineering, Building J01, The Chemical Engineering Building, The University of Sydney, Darlington, NSW 2006, Australia

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

Article history: A multicomponent particle drying model has been developed to describe the compositions and temper-
Received 8 January 2009 atures of spray-dried particles using distributed drying kinetics. This means that component migration is
Received in revised form 16 March 2009 expected to occur within the particles, thus temperature and moisture content gradients through the par-
Accepted 17 March 2009
ticles are present and might not be uniform. Particles are divided into a number of shells and may be
spray dried from a solution of several solutes dissolved in water, such as sugars and proteins. The model
simulates the drying kinetics within a concurrent spray dryer and also accounts for the effects of the sol-
Keywords:
ubilities of the solids, as well as surface activity of the components. The model has been successfully com-
Multicomponent model
Component migration
pared with an analytical solution for evaporation alone, to test the modelling of the external heat and
Spray drying mass-transfer processes. The model has also been compared with an analytical solution for internal dif-
Proteins fusion in a sphere, which has tested the modelling of the internal heat and mass-transfer processes.
Ó 2009 The Society of Powder Technology Japan. Published by Elsevier BV and The Society of Powder
Technology Japan. All rights reserved.

1. Introduction as the driving force for the diffusion. A model, which describes
these mechanisms, may be developed to predict the extent of com-
Segregation has been observed during the spray drying of milk, ponent segregation.
where the fat, lactose and protein segregate through the dried milk There are two fundamental approaches to modelling drying
particle [1,2]. Kim et al. [1] have done experimental work in this processes, namely lumped-parameter approaches and distrib-
area, but modelling of the problem has not yet been done. In their uted-parameter approaches. Lumped-parameter approaches,
work, Kim et al. [1] have found that fat tends to accumulate on the which assume that the physical properties and components of
outermost layer of a spray-dried milk particle. Nijdam and Lang- the drying materials remain uniform throughout the particle, in-
rish [2] confirmed that fat and protein accumulate near the surface, clude the concept of a characteristic drying curve and the reac-
preferentially to lactose. Segregation has advantages and disadvan- tion engineering approach. These models, by definition, do not
tages. The advantages of protein coating are that it may increase allow component segregation to occur [3–5]. Alternatively, using
the yield and improve the storage characteristics; and the potential a distributed-parameter approach, when applied to the spray
disadvantages include non-uniform dissolution and release of the drying of milk particles, means that the different layers of the
materials within the particles. Kim et al. [1] explained component spray-dried milk particle may contain different concentrations
segregation in spray dried milk by suggesting that the particle dry- of the components, such as fat and protein. This approach is
ing process occurs via three different processes. The first process therefore consistent with the component segregation phenome-
suggests that an initial layer of crust forms on the outermost layer non. Some researchers have attempted to model component seg-
of the droplet, which increases in thickness towards the centre of regation to predict the product composition profiles. For example,
the droplet as evaporation continues. The second mechanism sug- the drying of drops of solutions, based on moisture diffusion
gests that, as drying occurs, the difference in concentrations caused through solids, has been modelled for carbohydrates to investi-
by the precipitation of solids drives the diffusion of moisture and gate surface stickiness [6]. It divides each droplet into layers,
the dissolved components within the droplet. The third mecha- with evaporation only occurring on the surface and moisture dif-
nism considers the surface activity of the different components fusing outwards from the centre. The dissolved solutes are as-
sumed to be in solid form and cannot diffuse through the
droplet. Thus this model only allows the moisture content to dif-
fer within the droplet while component segregation is not possi-
ble. Seydel et al. [7] proposed a distributed-parameter model for
solid formation during spray drying which allows for the evalua-
* Corresponding author.
E-mail addresses: swan3658@mail.usyd.edu.au (S. Wang), T.Langrish@usyd.edu.au
tion of concentrations of components at different distances from
(T.A.G. Langrish). the centre of the droplet. In their studies, Seydel et al. [7] empha-

0921-8831/$ - see front matter Ó 2009 The Society of Powder Technology Japan. Published by Elsevier BV and The Society of Powder Technology Japan. All rights reserved.
doi:10.1016/j.apt.2009.03.004
S. Wang, T.A.G. Langrish / Advanced Powder Technology 20 (2009) 220–226 221

sised the occurrence of hollowed particles and did not attempt to 2.1. Assignment of state variables within particles and droplets
predict the degree of component segregation.
The lumped and the distributed-parameter models from the The assignment of state variables is shown schematically in
literature do not allow for the phenomenon of the migration of Fig. 1. Four layers are used here simply as an example, and any
milk components during drying as observed. To understand number of layers can be used.
and predict this, a new distributed-parameter model has been As shown in Fig. 1, the particle used to demonstrate the assign-
set up to study the implications of diffusion for both the solvent ment of the state variables within this particle is made up of four
and the dissolved components. The model also considers the ef- shells, with the outermost shell labelled as 1. Rpart is used to denote
fects of different solubilities of solutes. In this work, we compare the outer radius of each shell, i.e., Rpart1 denotes the distance from
the numerical simulation that solves this model with test cases, the centre of the particle to the outer surface of the shell. Dr is the
including some experimental data and some analytical solutions distance between the layers and can be found using Eq. (1):
for evaporation and for internal diffusion, to check the predic-
tions of this simulation for external and internal heat and Drj ¼ ðdpart;j  dpart;jþ1 Þ=2 ð1Þ
mass-transfer processes.
The positions, at which temperature and moisture contents of
the shells can be estimated, are symbolised by ‘‘d”. Dx is the dis-
tance between these positions and thus the pathway across which
2. Numerical methods – model development mass transfer occurs between the neighbouring layers. Dx is calcu-
lated using the equations below:
This model has been developed to predict the compositions and
temperatures of spray-dried particles using distributed drying At the surface : Dx1 ¼ Dr1 þ ðDr 2 =2Þ ð2Þ
kinetics. Component migration is expected to occur within the par- Between the surface and the centre : Dxj ¼ ðDr j þ Dr jþ1 Þ=2 ð3Þ
ticles, thus temperature and moisture gradients through the parti- In the centre of particleðnth shellÞ : Dxn1 ¼ Dr n þ ðDr n1 =2Þ ð4Þ
cles are present and might not be uniform. In this model particles
are assumed to be spherical. These particles are divided into j shells Tavx and Xavx are the temperature and moisture content esti-
and are formed from a solution containing several solutes such as mated at the interface between the shells. They are estimated
sugars and proteins. The model simulates the drying kinetics with- using weighted averages from T and X as stored at the designated
in a concurrent spray dryer. The following assumptions have been positions shown in Fig. 1. Tavx1 and Xavx1 are calculated in a similar
used for the spray dryer: way as shown in Eq. (5), here, SV stands for both of the state vari-
ables T and X:
 All particles are assumed to move downwards within the drying
SV av x 1 ¼ ðSV 1  Dr 1 þ SV 2  ðDr 2 =2ÞÞ=ðDr1 þ ðDr 2 =2ÞÞ ð5Þ
chamber.
 No agglomeration or break up of particles occurs within the
dryer. 2.2. The equations of the overall model
 No temperature and humidity gradients are present at any
cross-section for air within the drying chamber. 2.2.1. The trajectory of the particles
 There are no radial and tangential components of the inlet air The equations defining the velocity and trajectory of the parti-
velocity and the axial velocity is assumed to be uniform at every cles are adapted directly from Truong et al. [8] and will not be
cross-section of the drying chamber. shown here. These equations predict the axial (Upx), radial (Upr)
and tangential velocities (Upt) of particles as well as, UR,, the
The assumptions used in the particle drying model are as following: velocity of particles relative to that of the air. The radial distance
for each droplet or particle is also calculated using Upr and Upx. Re,
 Heat transfer only occurs through conduction inside the particle. the Reynolds number, is calculated using equations adapted from
Truong et al. [8]. Cd is found using the Schiller–Naumann equa-
These assumptions are only limitations of the model related to tion for estimating particle drag [9]. Droplets are assumed to
the treatment of the dryer and not related to the drying kinetics of shrink in size according to balloon shrinkage without crust form-
the particles. The approach to handling the drying kinetics can be ing [8].
implemented in a more advanced spray-dryer simulation such as
Computational Fluid Dynamics. At the moment, one assumption 2.3. Mass balances for particles
in the drying kinetics is that the shrinkage of droplets follows
the amount of water lost, thus inflation and deflation of the drop- 2.3.1. Mass flow rate by diffusion within the particles
lets due to puffing, expansion and contraction of trapped gases are The mass flux of each component by diffusion is calculated indi-
not included. The assignment of the state variables for the model vidually before being combined to estimate the total mass flux for
within each particle is described next. each layer. The overall mass balances as well as the component

Fig. 1. Schematic description of state variables within a four-layered particle.


222 S. Wang, T.A.G. Langrish / Advanced Powder Technology 20 (2009) 220–226

balances are calculated using these flux values. The mass fluxes of 2.3.4. Mass balance for the intermediate layers
all components between the 1st to the jth shell are estimated using Water balance of intermediate layers and component mass bal-
the general form shown in Eq. (6), which calculates the mass flow _ w and m
ances of dissolved solids are both shown in Table 1. m _ x are
rate of component x from layer j1 into j by diffusion. Rpart(i,j) is the mass flux of water and dissolved species x by diffusion, respec-
the radius of the jth shell of the ith particle: tively, passing through the intermediate layer interfaces. Cx is the
concentration of the dissolved species x.
_ x ði; jÞ ¼ 2pRpart ði; jÞ2 ððDx;T ði; jÞ
m
 ðC x ði; j  1Þ  C x ði; jÞÞÞ=Dxði; j  1ÞÞ ð6Þ 2.3.5. Mass balance for the central nth layer
_ x is the diffusion rate of species x into j layer from layer j1, Water balance of nth layer is simply the water mass flux by dif-
Here m
fusion, m_ w , passing through the nth layer interface into the layer by
and D is the diffusivity of the component. In the cases of dissolved
diffusion as shown in Table 1. The component balances for the dis-
solids, D takes the form of DxT which is the diffusivity of solid x
solved solids in the nth layer are the mass fluxes into the layer, as
through water in a dilute solution, and Cx is the concentration of
shown in Table 1. m _ x is the mass flux by diffusion passing through
the species for dissolved components. DxT follows the Stokes–
the nth layer interface of component x, and Cx is the concentration
Einstein equation and thus is directly related to temperature. The
of solid species x.
diffusivities may be estimated from the literature. Dw, the diffusion
coefficient of water through the dissolved solids is used to calculate
2.4. Solubility of the dissolved solids
the diffusion flux of water. The function used to estimate Dw is
adapted from Adhikari et al. [6]. Concentrations of the dissolved
As evaporation occurs on the surface of the particles, water dif-
components are calculated by Eq. (7):
, ! fuses out from the inner layers. The moisture contents within these
X layers decrease with time and the concentration of dissolved solids
C x ¼ mx =mtotal ði; jÞ ¼ mx mw ði; jÞ þ mx ði; jÞ ð7Þ changes due to diffusion. When the concentrations of these solutes
x
reach saturation levels, solids start to form within the layers. The
mw and mx are the mass of water and the dissolved components, mass balance of the solids that precipitate out in each layer is cal-
respectively. culated by Eq. (10):

2.3.2. Summary of heat and mass balances for different layers dmpxs
¼ ðC x ði; jÞ  C x max ði; jÞÞ  mpði; jÞ ð10Þ
The heat and mass balance for water and the dissolved compo- dh
nents of different parts of the particles are listed in Table 1. mpxs is the accumulated mass of the precipitated solid of species x
within each layer. Cxmax is the saturation concentration of the dis-
2.3.3. Mass balance for layer 1 solved solids, which are temperature dependent and have been ex-
Water, being volatile, evaporates from the particles at their sur- pressed as functions of temperature experimentally.
faces. The water balance of the first layer thus includes a term for
evaporation, as shown in Table 1. m _ w is the water mass flux be-
2.5. Heat flow rate equations
tween layers one and two, m _ w;v ap is the mass flux of water being
evaporated from the surface of the particles and is calculated using Like the mass balances, heat balances are also calculated by first
Eq. (8): calculating the flux of heat between the layers. It is assumed that
_ w;v ap ¼ p  ðK p ðp /  pbulk Þ  dpart ði; 1Þ2 Þ
m ð8Þ heat transfer within the particle or droplet only occurs by conduc-
tion. This flux is calculated using Eq. (11):
/ is the relative humidity of air, estimated using the modified Hen-
_ jÞ ¼ ðpdp ði; jÞ2 K mat ðTði; j  1Þ  Tði; jÞÞÞ=ðDxði; j  1ÞÞ
Hði; ð11Þ
derson model [8]. Kp is the partial pressure based external mass-
transfer coefficient (m2s/kg):
Kmat is the thermal conductivity of the material, which may vary
K p ¼ ðqa Dw=a Sh=dparti ÞM w =ðM a PÞ ð9Þ with composition and temperature, and H_ is the heat flow rate from
shell j1 into shell j.
Mw and Ma are the molecular weight of water and air, respectively
(g/mol), P is pressure (Pa), Sh is the Sherwood number and Dw/a is 2.5.1. Heat balance for layer 1
the diffusivity of water through air (m2/s), which can be found in lit- Besides warming the particles up, heat is used to evaporate
erature [8]. moisture at the surface of the particles. Energy is transferred
Component balances for layer one can be represented by a gen- to the first layer by the hot drying medium. The heat balance
eral equation, as shown in Table 1. m _ x is the mass flux of species x of the first shell is calculated by the equation shown in Table
from layer one into layer two and Cx is the concentration of solid x 1. Here Cpv, Cps and Cpw are the specific heats of vapour, solid
in layer one. The first term of the component balance for the shell and water, respectively. H_ a is the heat transferred from the dry-
one is due to the diffusion of dissolved solid species, and the sec- ing medium to heat up the first shell and H_ ev ap is the evaporative
ond term represents movement of this solid species while dis- heat load or duty. These terms are calculated in Eqs. (12) and
solved in water, which is also moving. (13):

Table 1
Mass and energy balance of the different components of particles.

Equations 1st shell Middle shells Centre


_ w;v ap ði;2Þm
m _ w ði;2Þ _ w ði;jÞm
ðm _ w ði;jþ1Þ _ w ði;nÞ
m
Water mass balance U px ðiÞ U px ðiÞ U px ðiÞ
_ x ði;2Þm
½m _ w ði;2ÞC x ði;1Þ _ x ði;jÞm
½m _ x ði;jþ1Þþm
_ w ði;jÞC x ði;j1Þm
_ w ði;jþ1ÞC x ði;jÞ _ x ði;nÞþm
½m _ w ði;nÞC x ði;n1Þ
Component mass balance U px U px ðiÞ U px ðiÞ
½H_ a þH_ ev ap Hði;2Þ
_ _
½ðHði;jÞ _
Hði;jþ1ÞÞ _
½ðHði;nÞÞ
Heat balance ½ms ðC ps þXC pw ÞU px  ms ðC ps þXC pw ÞU px ms ðC ps þXC pw ÞU px
S. Wang, T.A.G. Langrish / Advanced Powder Technology 20 (2009) 220–226 223

H_ a ¼ pdp ði; 1ÞK a NuðT B  Tði; 1ÞÞ ð12Þ cles of fraction i. The numerator adds up the total moisture mass,
dmp and the denominator calculates the total dry solid masses within
H_ ev ap ¼ Hfg U px ð13Þ all shells of the particles for all the size fractions.
dh
Nu is the Nusselt number [10], TB is the dry bulb temperature of the 2.8. Method of simulation
bulk air in the dryer, Hfg is the latent heat for water evaporation and
Ka is the thermal conductivity of air. The equations in the model have been solved using a fourth-or-
der Runge–Kutta procedure similar to that described in Truong
2.5.2. Heat balance for the intermediate and the central nth layer et al. [8]. For each step size, the heat and mass balances, as well
The heat balance for the intermediate shells and the central as the trajectory of all droplets, have been calculated first. The
layer are both shown in Table 1. H _ is the heat transfer rate through evaporated moisture from all fractions of particles has then been
the respective layer interface. added to the drying gas, increasing the humidity. A similar proce-
dure has been used to estimate the enthalpy and temperature of
2.6. Mass and heat balance for the drying medium (air) the drying air. The outputs from this step are used as the input val-
ues for the next iteration. The program stops when all particles exit
The mass balance for the drying medium is adapted directly the dryer. The distance from the atomizer to the exit is set to be
from Troung et al. [8]. A heat balance for the air, accounting for 0.48 m (based on the length of the Buchi small-scale 290 advanced
the heat required to evaporate moisture from the droplets or par- spray dryer). The program is written using Matlab.
ticles, convective heating or cooling of droplets or particles, and
heat lost from the dryer, is shown in Eq. (14): 3. Result and discussion – model testing
" #
dHh X h i
¼ _ _
ðHev ap ðiÞ þ Ha ðiÞÞ =U px  U dryer ðT a  T amb Þ=L =G 3.1. Grid sensitivity check
dh droplets

ð14Þ To carry out the grid sensitivity check and to test the external
G is the mass flow rate of the air, and Yb is the humidity of the bulk moisture evaporation rate predicted by the simulation, the simula-
air. Hh is the enthalpy of humid air. Ta and Tamb are the temperatures tion was run with low solid concentrations of 1% wt, no gravity and
of the drying medium (air) and ambient air, respectively. Udryer is no drag force. The effects of solubility of the dissolved solids and
the overall heat-transfer coefficient from the walls of the spray their diffusion mechanism were ignored by setting the solubilities
dryer to the ambient environment and L is the length of the to high values and setting the diffusion coefficients for the solutes
chamber. to zero values, while keeping the diffusion coefficient of water
none zero. By doing so, the model simulates the drying of a sus-
2.7. Overall average moisture content pended droplet of a very dilute solution under isothermal condi-
tions. A high air-to-liquid ratio was achieved to ensure there
The average moisture content of the ith weight fraction of the would be little change in the gas temperature or humidity, by
product can be calculated using the following equation: using a low liquid flow rate of 1.78  106 kg/s and a high gas
!, ! flow rate of 0.0108 m3/s. The initial conditions used are shown in
X X X X Table 2.
X ov ðiÞ ¼ FðiÞX i ms;i FðiÞms;i ð15Þ
droplets shells droplets shells
Using the parameters as specified above, simulations were car-
ried out for different number of layers, to find out the minimum re-
F(i) is the frequency of the number distribution for the particles of quired to give average moisture contents that were not sensitive to
the size fraction i. msi is the mass of solid in each shell of the parti- the use of further layers. The results are shown in Fig. 2.

Table 2
Parameters used in the analytical calculation of drying kinetics.

qa (kg/m3) la (105 kg/ms) Dp (107 m) P* (103 Pa) Ts (K) UR (m/s) P (105 Pa) Upx (m/s) Xs (kg/kg)
1.13 1.93 8.03 3.51 300 40.7 1.01 41.3 99

Fig. 2. Simulated Xov as a function of the number of layers and required simulation time.
224 S. Wang, T.A.G. Langrish / Advanced Powder Technology 20 (2009) 220–226

As observed in Fig. 2, the simulation results for the final average drying) were chosen for the analytical calculations because they al-
particle moisture content change from 0.5 kg/kg at three layers to low the solution droplets to dry more slowly compared with the
0.74 kg/kg at nine layers. Simulations using more than nine layers situation at gas temperatures such as 210 °C, which assist in com-
within the particle resulted in small fluctuations about the value of paring the model and analytical results. The analytical and the cor-
0.74 kg/kg. This result suggests that dividing the particles into nine responding results generated by the model are shown in Table 3.
control volumes is sufficient to give relatively insensitive results to Here, drying rate is in kg of water evaporated per unit length of
the numbers of control volumes, while further numbers of control dryer.
volumes mainly just increase the computation time. As observed in Table 3, the average rates of drying generated
from the simulations at inlet temperatures of 40 °C and 70 °C differ
3.2. Analytical calculations of the drying kinetics by 0.04% and 0.06% from the corresponding analytical results,
respectively. The simulated constants at the initial drying condi-
The rate of drying predicted by the model has then been com- tions are also similar to the analytical results. The negligible differ-
pared with the solution obtained by solving the equations analyt- ences are probably due to round off errors occurred during
ically. The parameters used for both the simulation and analytical analytical calculations. Overall, it can be concluded that the simu-
tests are as specified previously. The stepwise calculations for the lated external evaporation rate is practically identical to the ana-
analytical solution at 40 °C are shown below: lytical results.

Re ¼ ðqa U R dp Þ=l
3.3. Comparison with the literature: internal mass-transfer processes
¼ ð1:13  40:7  8:03  107 Þ=ð1:93  105 Þ ¼ 1:92 ð16Þ
Dw=a ¼ 1:17564  109  T 1:75
abs  P atm =P In this simulation, water evaporation is assumed to occur only
¼ 1:17564  10 9
 313 1:75 5
¼ 2:74  10 m s 2 1
ð17Þ at the surface. The mass transfer of water within the material lay-
ers occurs by diffusion. The rate of diffusion estimated by the sim-
Sc ¼ l=ðqa Dw;a Þ
ulation has been compared with the results of an analytical
¼ ð1:93  103 Þ=ð1:13  2:7  105 Þ ¼ 0:62 ð18Þ solution of the diffusion equation. In this analytical solution, time
pffiffiffiffiffiffi 1 is included in the Fourier number, which is more meaningful than
Sh ¼ 2:0 þ 0:6 Re  Sc3
pffiffiffiffiffiffiffiffiffiffi 1
the time itself, because the Fourier number includes the scale
¼ 2:0 þ 0:6 1:92  0:623 ¼ 2:71 ð19Þ (length or radius, r) and the diffusion coefficient, as will be de-
K m ¼ qa Dw=a Sh=dpart i  M w =M a P scribed next. This theoretical equation is shown below [12]:
¼ 1:13  2:7  105 =ð8:03  107 Þ v X
1
2 2 2
X
1
2 2
4 2 1 ¼1þ2 ð1Þn eðkn p tÞ=r ¼ 1 þ 2 ð1Þn eðF o n p Þ ð21Þ
 18:02=ð28:97  101325Þ ¼ 6:40  10 m s ð20Þ V n¼1 n¼1

The analytical results have been compared with the modelled where v/V is the fractional change in the moisture content at the cen-
results at inlet temperatures of 40 °C and 70 °C. These relatively tre of the spherical particles, defined by v =V ¼ XX i X
X c
e
 X c ; X i andX e are
i
low gas temperatures (compared to >100 °C typically used in spray the moisture contents of the centre, the initial particle and the

Table 3
Differences of calculated analytical results compared with simulated results.

T (°C) Dw,a (105 m2 s1) (%) Km (104 m1 s) (%) mw,vap (1012 kg s1) (%) Drying rate (1014 kg m1) (%)
40 0.03 0.03 0.03 0.04
70 0.00 0.06 0.06 0.06

Fig. 3. The rate at which the surface layer reaches moisture equilibrium, compared with the assumption made in the analytical solution.
S. Wang, T.A.G. Langrish / Advanced Powder Technology 20 (2009) 220–226 225

Fig. 4. The results from the analytical and numerical solutions of the diffusion equation for different numbers of layers in the numerical solution.

equilibrium moisture content, respectively. Fo is the Fourier number, about 0.6 s from the analytical solution. The small difference ob-
which is defined to be Dt/r2, and D is the diffusion coefficient (m2/s). served between the simulated results and the analytical one may
The analytical solution includes the assumptions that the parti- be due to the assumption discussed above about the change in
cle moisture content is initially uniform and that the surface of the the surface moisture content in the analytical solution, which is as-
particle comes to equilibrium instantly. In the numerical solution, sumed there to be instant. Physically, this is not the case, and this
the moisture content is uniform initially, but the surface does not physical behavior is reflected in the numerical solution as shown in
come to equilibrium instantly, like the real physical situation. The Fig. 3.
predicted delay in the surface coming to equilibrium, from the In Fig. 4, the results of simulations performed using different
numerical solution, is shown in Fig. 3 to be around 0.001 s. This de- layers of dividing the particles have also been compared with the
lay may cause some minor discrepancies between the analytical analytical solution. It is also observed from Fig. 1 that, as the num-
and numerical solutions, even when the numerical solutions have ber of layers is increased, the predicted shape of the initial drying
a large number of layers, because the boundary conditions are not curve from the numerical solution is more similar to that esti-
exactly identical. mated by the analytical solution, in terms of the initial delay in
The comparison of the results from the analytical solution with the drop of the centre moisture content (initial penetration period).
the results from the simulation is shown in Fig. 4 for different To explain this further, the schematic drawing in Fig. 5 is used and
numbers of layers in the numerical solution. the time taken by the initial penetration period is presented in
From Fig. 4, it is observed that the predicted rate of decrease in Table 4.
the moisture content from the numerical solution at the central As illustrated in Fig. 5, unlike the analytical solution where the
layer of the particle is similar to that from the analytical solution. ‘centre’ of the particle is taken to be the exact point in the middle
All of the simulated results came within 1% of the equilibrium of the particle, the ‘centre’ of the simulation is regarded as the nth
moisture content within a similar period of time to the result of (inner most) layer of the particle. However, this difference is likely

Fig. 5. Schematic diagram showing the diffusion pathways of the analytical solution and the numerical simulation.
226 S. Wang, T.A.G. Langrish / Advanced Powder Technology 20 (2009) 220–226

Table 4 wet-bulb temperature of 38.5 °C at a time that was between


Penetration periods (onset Fourier number for the regular regime) for the analytical 5–10 s after the start of the experiment. This result is longer com-
and numerical solutions and the maximum discrepancy between the numerical
solutions and the analytical solution where the maximum discrepancy is calculated
pared with the simulated results from both the solute fixed coordi-
by defined as (FoanalyticalFonumerical)/Foinitial  100(%). nate model (<5 s) as well as the new multicomponent distributed
parameter model (3–4 s). In their work, Adhikari et al. [6] con-
Fo Maximum discrepancy from
the analytical solution (%)
cluded that the experimental and predicted values of the temper-
ature history of the water droplet agreed well with each other,
Analytical solution 0.032 N.A
Numerical solution, 4 layers 0.007 11.3
with an absolute error of less than 5%. Comparing the predicted
Numerical solution 8 layers 0.015 9.21 temperature of the two models, the multicomponent distributed
Numerical solution 12 layers 0.018 6.52 parameter model predicted the droplet temperature rose to
Numerical solution 16 layers 0.022 5.62 37.4 °C, compared with 38.5 °C as predicted by Adhikari et al. [6].
Numerical solution 20 layers 0.026 5.62
The small difference is likely to be due to the different assumptions
made in predicting the flux of evaporation. Adhikari et al. [6] in-
cluded a correction factor for high heat fluxes. In the distributed
to be small when the gradient in moisture content (with respect to parameter model, this correction factor was not applied, but this
distance) is small, as is likely to be the case since the moisture is can be included in the future.
removed from the surface. The concept of an initial penetration
period has been discussed by Keey [3], where he quoted the work 4. Conclusions
of Thijssen and Coumans [11] who found that during Fickian diffu-
sion, there is a penetration period where ‘‘a moisture-deluded zone A distributed-parameter, multi-component model has been set
penetrate(s) the material”. During this period, the loss of moisture up to simulate the spray drying process. This model takes the dif-
front moves from the surface of the droplets progressively until it fusion of water, the diffusion of dissolved solids and the effects of
reaches the centre. After this happens, the regular regime occurs, the solubilities of the solids into consideration. Dividing the parti-
where the droplet starts to lose moisture through the entire mate- cles into nine layers allows the model to give a grid-insensitive
rial, including the centre. It is also observed in Table 4, where the solution. The drying kinetics of the particles have been examined
initial penetration period increased from a Fourier number of under isothermal conditions with low solids concentrations. The
0.01 to the Fourier number estimated by the analytical solution simulated results have been compared with the analytical results,
of 0.03, when the number of layers used in the grid division is in- and the agreement is very good. Both internal and external heat
creased from 4 to 20 layers. This situation also explains the fact and mass transfer processes have been checked against literature
that as the number of layers is increased, the discrepancy de- solutions, with the internal diffusion process simulated in the
creases, since the ‘centre’ becomes smaller, with finer grid divi- model matching the analytical solutions for diffusion very closely.
sions being used in the numerical simulation.
From Fig. 4, it is observed that as the number of layers increases, References
the shape of simulated curves gets more similar to that from the
[1] E.H.-J. Kim, X.D. Chen, D. Pearce, On the mechanisms of surface formation and
analytical solution in terms of the rate of diffusion. This is com-
the surface composition of industrial milk powder, Drying Technology 21 (2)
pared next in Table 4, by calculating the maximum percentage dis- (2003) 265–278.
crepancy between analytical and numerical solutions, where the [2] J.J. Nijdam, T.A.G. Langrish, The effect of surface composition on the
discrepancy is defined as (FoanalyticalFonumerical)/Foinitial  100(%). functional properties of milk powder, Journal of Food Engineering 77 (4)
(2006) 919–925.
This maximum discrepancy is observed from Table 4 to decrease [3] R.B. Keey, Introduction to Industrial Drying Operations, Pergamon, Oxford,
from 11.3% to 5.62% when the number of layer was increased from 1978. pp. 154–156.
4 to 16 layers. The discrepancy then became constant with a fur- [4] K.C. Patel, X.D. Chen, Prediction of spray-dried product quality using two
simple drying kinetic models, Journal of Food Process Engineering 28 (6)
ther increase of number of layers used in the grid division, possibly (2005) 567–594.
because of the slight difference in the boundary condition (Fig. 5). [5] T.A.G. Langrish, T.K. Kockel, The assessment of a characteristic drying curve for
A discrepancy of 5–6% seems tolerable, suggesting that 16 layers milk powder for use in computational fluid dynamics modelling, Chemical
Engineering Journal 84 (1) (2001) 69–74.
are sufficient for further simulation work using the numerical [6] B. Adhikari, T. Howes, B.R. Bhandari, Use of solute fixed coordinate system and
model. method of lines for prediction of drying kinetics and surface stickiness of single
droplet during convective drying, Chemical Engineering and Processing 46 (5)
(2006) 405–419.
3.4. Comparison with the literature: external heat and mass transfer
[7] P. Seydel, A. Sengespeick, J. Blomer, J. Bertling, Experiment and mathematical
processes modeling of solid formation at spray drying, Chemical Engineering and
Technology 27 (5) (2004) 505–510.
[8] V. Troung, B.R. Bhandari, T. Howes, Optimization of co-current spray drying
To check this new multicomponent distributed parameter mod-
process of sugar-rich foods. Part I – moisture and glass transition temperature
el in terms of external heat and mass transfer processes, the simu- profile during drying, Journal of Food Engineering 71 (1) (2005) 55–65.
lated results were compared against modelled as well as [9] M. Rhodes, Introduction to Particle Technology, Wiley, New York, 1998. pp. 2–
experimental results published by Adhikari et al. [6]. In their mod- 4.
[10] W.E. Ranz, W.R.J. Marshall, Evaporation from drops, Chemical Engineering
el, Adhikari et al. [6] studied the droplet and particle drying process Progress 148 (4) (1952) 141–146.
using experiments conducted within a cross-flow drying tunnel. In [11] H.A.C. Thijssen, W.J. Coumans, Short-cut calculation of non-isothermal drying
this study droplet is held on a holder and dried in a chamber with rates of shrinking and non-shrinking particles containing an expanding gas
phase, in: Proc. 4th IUFRO International Drying Symposium, IDS ‘84, Kyoto,
hot air blowing across it. The mass and temperature history of the Japan, vol. 1, 1984, pp. 22–30.
droplet is measured and recorded via the holder. During their [12] H.S. Carslaw, J.C. Jaeger, Conduction of Heat in Solids, second ed., Clarendon
experiment the temperature of the water droplet rose to the Press, Oxford, 1959. pp. 233–235.

Potrebbero piacerti anche