Вы находитесь на странице: 1из 13

Review

A review of process simulations and the use of additives in spray drying


S. Wang
*
, T. Langrish
School of Chemical and Biomolecular Engineering, Building J01 The Chemical Engineering Building, The University of Sydney, NSW 2006, Australia
a r t i c l e i n f o
Article history:
Received 20 May 2008
Accepted 9 September 2008
Keywords:
Multicomponent simulation
Casein
Proteins
Spray-drying additive
Particles
Drying
a b s t r a c t
The driving forces for protein migration during drying have been reviewed and existing simulations on
drying process have been analysed. Current models of spray drying have not captured the diffusion pro-
cess of both the solvent and solutes, or the effects of the solubilities and surface activities of the compo-
nents. A new distributed-parameter, multicomponent model needs to be developed. Casein has been
examined and found to coat spray-dried particles at 1 w/w%, signicantly lower than other additives,
such as maltodextrin, which commonly make up between 10 w/w% and 20 w/w% of the total solids.
The efcacy of casein as a spray-drying additive thus needs to be studied.
Crown Copyright 2008 Published by Elsevier Ltd. All rights reserved.
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2. Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1. Evidence of component segregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2. Hypotheses to explain the segregation process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1. Crust formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.2. Effect of solubility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3. Effect of surface activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3. A review of existing models of drying processes for the purpose of predicting component segregation . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.1. Lumped-parameter models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2. Distributed-parameter approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4. A new distributed-parameter multi-component model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.1. The diffusion process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.2. Diffusion coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.3. Implementations of surface activity effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1. Introduction
Food is a major component of post-harvest agricultural prod-
ucts. Since the majority of fresh food coming from animal (such
as milk) and plant sources is highly perishable, value addition to
it is possible through post-harvest engineering. Highly functional-
ised powders, such as lactose, whey proteins, caseins and milk per-
meates, are produced in Australia by various large scale agro-dairy
businesses such as Murray Goulburn, Bonlac Foods (now part of
Fonterra), Warrnambool Cheese and Butter and Tatura Milk Indus-
tries through spray drying. In the 2001/2002 nancial year Austra-
lia exported 455,000 tonnes of milk powder, which is worth one
and a half billion dollars (Australian Bureau of Statistics).
Besides being used in the manufacturing of dried food, spray
drying is also used in other industries, such as in the production
of fertilisers and oxide ceramics and pharmaceuticals (Masters,
1996). Masters (1996) estimated that more than 15,000 spray dry-
ers of industrial size are in operation throughout the world and
0963-9969/$ - see front matter Crown Copyright 2008 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.foodres.2008.09.006
* Corresponding author. Tel.: +61 2 9351 5660.
E-mail addresses: swan3658@mail.usyd.edu.au (S. Wang), T.Langrish@usyd.
edu.au (T. Langrish).
Food Research International 42 (2009) 1325
Contents lists available at ScienceDirect
Food Research International
j our nal homepage: www. el sevi er. com/ l ocat e/ f oodr es
approximately double that number in pilot plants and laboratories.
In the pharmaceutical industry, spray drying is a common method
used to prepare dry powder aerosols (Kuo et al., 2007). These dry
powder aerosols have advantages compared with solution aerosols
due to the chemical stability associated with the powder form
(Chan, 2006).
Despite the signicant success in utilising spray dryers in the
dairy industry, the post-harvest processing of fruit and vegetables
has so far been limited to making fruit juices and canning (sugar
syrup and salt brine as main preservatives). The export of fresh
plant products has not been very successful because of their high
water content and the requirement to refrigerate them throughout
their life cycle. The conversion of fruits, vegetables and honey into
powder form has not been technically and commercially feasible
(Adhikari, Howes, Bjandari, & Troung, 2004). Converting fruit juice
and honey into dry powder form faces key challenges due to low
recovery rates from the dryers. This is because sugar rich and acid
rich foods are considered to be sticky materials. This sticky behav-
iour is commonly attributed to the presence of low molecular
weight sugars (Adhikari, Howes, & Troung, 2003). Spray drying at-
tempts for these food products have encountered problems,
including wall deposition of powders in dryers and the clumping
of powders on conveyors. The powders undergo caking during
packaging, storage and transportation. The powder that is stuck
on the dryer wall stays there for the whole drying cycle. As a con-
sequence it gets scorched or overcooked. This lowers the product
quality and increases the risk of re (Adhikari et al., 2003).
The yield of the spray-drying process is greatly affected by the
amount of wall depositions that occur. As commonly dened, yield
is the percentage of theoretical product that are actually collected.
Whenwall depositionoccurs, spray-driedparticles stick to the walls
of the spray dryer and thus decrease the amount of product col-
lected. Masters (1996) stated that wall deposition is a common
occurrenceinspraydryingandaccounts for the mainloss of product.
The stickiness of the spray-dried products signicantly affects
the degree of wall deposition. The stickiness in spray-drying sugar
and acid-rich foods has been attributed to the thermo-plasticity of
low molecular weight sugars and organic acids present in the
materials (Truong, Bhandari, & Tony, 2005). Truong et al. (2005) ci-
ted Bhandari, Datta, Crooks, Howes, and Rigby (1997), Hennings,
Kockel, and Langrish (2001) and Roos and Karel (1991) that amor-
phous materials change from a glassy state to a liquid-like rubber
state and become sticky when particle temperatures become high-
er than the sticky-point temperatures (T
sticky
). This rubbery state
causes material adhesion to dryer surfaces. Adhikari, Howes, Shres-
tha, and Bhandari (2007), citing Roos (1995) and Bhandari and
Howes (1999), further suggested that the sticky-point temperature
is 20 C higher than the glass-transition temperature and that a
particle surface is considered to be sticky if the surface tempera-
ture of the particle is higher than the sticky-point temperature.
This sticky-point temperature is related to the glass-transition
temperature, which is explained by Bhandari and Howes (1999)
to be the temperature where the transition from a glassy state to
a rubbery state occurs. The additives used as spray-drying aids of-
ten have a high glass-transition temperature, so by mixing this
with the materials having low glass-transition temperatures, a
higher overall glass-transition temperature (and thus the sticky-
point temperature) of the bulk material is achieved. By raising
the glass-transition temperature (and thus the sticky-point tem-
perature), the particle temperature is less likely to become higher
than the sticky-point temperature, thus reducing the stickiness of
the products. Adhikari et al. (2004) further suggested that the safe
(non-sticky) drying regime is when the glass-transition tempera-
ture of the surface layer is greater than 10 C higher than the par-
ticle surface temperature.
Many workers (Adhikari et al., 2004; Cano-Chauca, Stringheta,
Ramos, & Cal-Vidal, 2005; Chidavaenzi, Buckton, & Koosha, 2001;
Roustapour, Hosseinalipour, & Ghobadian, 2006) have used malto-
dextrin and other materials as additives in spray drying to decrease
the stickiness of spray dried products. Most of these materials have
high glass-transition temperatures and thus decrease the bulk
glass-transition temperatures of the mixtures. Production of pow-
Nomenclature
A
p
particle surface area (m
2
)
D diffusivity (m
2
s
1
)
DE
v
activation energy for drying (J mol
1
)
F Faradays constant (96,500 C mol
1
)
K
B
Boltzmanns constant (1.38 10
23
J K
1
)
J
m
molar ux vector in three dimensions (mol m
2
s
1
)
N drying rate (kg m
2
s
1
)
N
s
the specic drying rate (kg m
2
s
1
)
N
max
maximum drying rate (kg m
2
s
1
)
P pressure (Pa)
R the ideal gas constant (8.314 J mol
1
K
1
)
T temperature (K)
X moisture content (kg kg
1
)
X
e
equilibrium moisture content (kg kg
1
)
X
cr
Critical moisture content (kg kg
1
)
Nu Nusselt number
Re Reynolds number
Pr Prandtl number
C molar concentration (mol m
3
)
d
p
particle size (m)
h external mass transfer coefcient (m s
1
)
K thermal conductivity (W m
1
K
1
)
k
*
corrected mass transfer coefcient (m s
1
)
l distance (m)
p permeabilities of solid (m
2
)
r radius of layers within particles (m)
t time (s)
v velocity of component convection (m s
1
)
x mole fraction
y saturated vapour concentration (kg m
3
)
Greek symbols
b mass transfer coefcient (s
1
K
1
)
c molar chemical potential (J mol
1
)
q density (kg m
3
)
e volume fraction (m
3
m
3
)
l dynamic viscosities (kg m
1
s
1
)
r the Laplacian operator
ru electrostatic potential gradient (V m
1
)
Subscripts
av average
atm atmospheric
e equilibrium
b bulk
g gas
l liquid
v vapour
w water
14 S. Wang, T. Langrish/ Food Research International 42 (2009) 1325
dered honey, from its natural form, has remained a challenge both
to producers as well as researchers. Dry honey contains 23.5% of
water. The low water content means that it is lighter to transport.
Dried honey powder also ows more easily than natural honey and
thus is easier to store. Currently, carriers such as maltodextrins are
added to honey during spray drying to decrease stickiness (Adhik-
ari et al., 2004). Spray dried honey may contain as little as 50% or
less of natural honey, with the rest being made up by additives
(National Honey Board). For example, up to 63% maltodextrin is
added to make the production of honey powder commercially via-
ble. The excessive use of these carriers, which are not native to the
powders, alters the avour and texture of the products and hence,
is undesirable for consumers, manufacturers and other users, since
the product contains a signicant fraction of non-honey material.
Unlike the previous explained carriers, which were used to
change physical properties of the bulk solution, casein has been
used as a carrier during drying due to its surface activity (Faldt &
Bergenstahl, 1994). A comparison between the materials used as
spray-drying additives as well as the amount used is summarised
in Table 1.
As shown in Table 1, Faldt and Bergenstahl (1994) found that
products from concentrations of casein as low as 1% of the bulk so-
lid dominated the surface of the particles. This concentration of
casein was much lower than the rest of the additives, which may
be due to the surface activity of casein since casein absorbs at
air/water interfaces preferentially, and thus only a small amount
of it is required to cover the surface of the particle.
Hence, there is an interest in developing smart carriers based on
food proteins, non-protein surfactants and lm-forming biopoly-
mers. It is desirable that these smart carriers can improve yields
at low carrier concentrations, such as 2.5 w/w% of total solids. It
is suggested here that the surface-active nature (chemical struc-
ture) of the protein drives its movement towards the surface
(Kim, Chen, & Pearce, 2003). The carriers need to form a thin and
smooth lm or skin at the surface to be effective. The formation
of this thin lm during the spray-drying process requires segrega-
tion to occur, such that the surface-active component with a high
glass-transition temperature forms a protective layer around the
bulk materials with lower glass-transition temperatures.
To develop these ideas, the mechanism and magnitude of the
component segregation during spray drying both need to be under-
stood. The magnitude of the surface activity and the excess concen-
tration of the carriers on the surface are assessed here. For
example, if the surface activity of protein drives this segregation,
then this could be changed if strong surfactants that are already
being used in the food industry, such as polysorbate 80, are
introduced.
2. Literature review
A review of existing work will now be given for the segregation
of different components during drying and modelling the spray-
drying process to understand this segregation. The purpose of this
paper is to identify the key gaps in the literature and how these
gaps need to be addressed. First, the evidence for component seg-
regation is reviewed, before hypotheses to explain the segregation
process are discussed. Then the existing lumped-parameter and
distributed-parameter models for drying are analysed, and the crit-
ical gaps in these models for component segregation are identied.
2.1. Evidence of component segregation
Diffusion occurs within solutions with multiple components
when their concentrations are not uniform (Cussler, 1997). This
suggests that spray-dried particles of multi-component mixtures
could potentially possess different physical properties at the sur-
face, since segregation can occur during the drying process. Such
properties include ease of powder ow and stickiness, which affect
the amount of wall deposition in spray dryers. By controlling the
spray-drying process, products with different physical properties
can be manufactured, thus increasing the effectiveness of the
spray-drying process in terms of producing powders with better
storage or controlled-release properties, and/or simply giving a
higher powder yield from spray drying. Controlled release (selec-
tive release) of spray-dried products can be achieved if layered or
encapsulated particles can be produced where dissolution is al-
lowed to happen. This delayed or controlled release is desirable
for many pharmaceutical purposes with pH sensitive drugs (Tan,
Goh, & Tam, 2007).
Segregation has been observed during the spray drying of milk,
where the fat, lactose and protein segregate through the dried milk
particle (Adhikari et al., 2007; Bhandari & Howes, 1999; Cano-Cha-
uca et al., 2005; Chidavaenzi et al., 2001; Cussler, 1997; Faldt &
Bergenstahl, 1994; Honey Board, 2007; Kim et al., 2003; Nijdam
& Langrish, 2006; Roos, 1995; Roustapour et al., 2006; Tan et al.,
2007). Kim et al. (2003) have found that fat tends to accumulate
on the outermost layer of a spray-dried milk particle. Beneath this
surface layer, fat globules that are covered by proteins exist, fol-
lowed by lactose and salts. They suggest that the segregation of
materials through the particle is due to the difference in the diffu-
sivity of the components. Initially, the concentrations of the differ-
ent components are most probably uniform throughout a particle,
as assumed by Meerdink and Riet (1995). However, as the moisture
evaporates fromthe particle surface, the local concentrations of the
components increase there. This may cause some components to
diffuse inwards to the centre of the particle. Since different compo-
nents have different diffusivities (Kim et al., 2003), segregation
may occur through the entire particle.
The migration of components within milk droplets in spray
dryer has also been studied by Nijdam and Langrish (2006). They
conrmed that fat and protein accumulate near the surface, prefer-
entially to lactose. They also found that, at higher operating tem-
peratures, protein appears more signicantly than lactose near
the surface. This led to them postulating that high temperatures
promote the formation of a protein skin near the particle surface,
which hinders the migration of lactose to the surface.
Table 1
Summary of existing selected work on spray drying performed with carriers
Raw material Additives Amount of additives used Reasons for choices of additives Source of information
Lime juice Silicon dioxide 10% of solution High T
g
and sticky point Nijdam et al. (2000) and Roustapour et al. (2006)
Maltodextrin 20% of solution
Mango juice Maltodextrin 12% of solution N.A. Cano-Chaocua et al. (2005) and Ribeiro et al. (2006)
Arabic gum
Waxy starch
Cellulose 3%, 6% and 9% of solution Induce crystallinity
Lactose Casein 1% of total solid (=0.1% of solution) Surface activity Faldt and Bergenstahl (1994) and Roustapour et al. (2006)
S. Wang, T. Langrish/ Food Research International 42 (2009) 1325 15
Adhikari et al. (2007) studied the segregation effect in terms of
surface tension and viscosity. They measured the effect on the sur-
face tension of lactose solution droplets by adding whey protein
isolate (WPI). This was carried out using a probe tack testing
instrument. Theoretically, a state of no adhesion is achieved when
the surface tension or the cohesive forces of the droplets are higher
than the adhesive forces to other particles or surfaces. They have
found that, by adding WPI, which behaves as a surfactant, the sur-
face tension and the cohesive forces of the lactose solutions were
reduced. However, it was also realised that different results may
arise if the solutions of WPI and lactose were subjected to high
temperatures, because protein tends to form a skin layer as found
by Nijdam and Langrish (2006).
2.2. Hypotheses to explain the segregation process
The drying of milk powder has been described in Kim et al.
(2003) to occur via three different processes, which are not neces-
sarily mutually exclusive. Although suggested in the context of
milk drying, these hypotheses can also be applied to the segrega-
tion of components in other mixtures of soluble substances. The
rst mechanism suggests that an initial layer of crust forms on
the outermost layer of the droplet. As evaporation continues to oc-
cur, the crust increases in thickness towards the centre of the drop-
let. The second mechanism suggests that moisture and the
dissolved components migrate within the droplet as drying occurs.
The drying force for this diffusion is the difference in concentration
caused by precipitation of solids. Thus this mechanism takes the
solubility of the components into account. The third mechanism
considers the surface activity of the different components as the
driving force for the diffusion.
2.2.1. Crust formation
The rst mechanism has been described by Kim et al. (2003).
This mechanism assumes that an initial layer of saturated material
forms on the outside of a multi-component droplet, such as a milk
droplet. As more moisture is evaporated, the dissolved components
deposit as solids. The thickness of the crust increases with time as
moisture is evaporated. Kim et al. (2003) suggested that, for this
hypothesis, no segregation of components necessarily occurs dur-
ing the drying process, and the fat, protein and lactose may be dis-
tributed evenly through the milk particle. However, the formation
of a crust can continue to happen during the segregation and
migration of different components, since the processes are not
mutually exclusive.
2.2.2. Effect of solubility
Kim et al. (2003) also described the next hypothesis, where
the solvent migrates towards the surface of the droplet during
drying, while the solute migrates to the centre. At the surface
of the droplet where evaporation occurs, the concentration of
the dissolved solids increases. This change in solution concentra-
tion drives diffusion of the components towards the centre. The
diffusivities of the many components of milk differ from each
other. As a result, some components may travel to the centre
of the droplet at a faster rate than others. As the components mi-
grate, the substance with lower solubility precipitates rst, while
others continue to diffuse to the centre. Since diffusivity de-
creases with increasing molecular weight, substances with low
diffusivity often have low solubility as well. Thus the outermost
layer of the milk particle is dominated by large insoluble mole-
cules, such as fat. Meerdink and Riet (1995) support these views
about the competition between the transfer rates of the solutes
caused by their different diffusion rates, and these workers based
their model on this differential diffusion for predicting the extent
of segregation.
2.2.3. Effect of surface activity
The third hypothesis has been described by Kim et al. (2003),
where a surface-active protein or molecule, if present, moves pref-
erentially to the air/liquid surface. During spray drying, the solu-
tion passes through the nozzle, and surface-active proteins may
form a monolayer on the surface of the droplets quickly. This pro-
cess may cause proteins to precipitate out almost instantly on ini-
tial heating at the liquidair interface, causing a drop in the local
concentration of protein near the surface of the droplets. The dif-
ference in concentrations then drives the diffusion of protein from
the centre to the surface of the droplets.
Faldt and Bergenstahl (1994) performed experiments to com-
pare these hypotheses. In their work, they spray-dried casein and
lactose mixtures and studied the surface compositions of the prod-
ucts by using electron spectroscopy for chemical analysis (ESCA,
otherwise known as X-ray photoelectron spectroscopy (XPS)). They
found that, even at lowconcentrations with a 1:99 ratio of casein to
lactose, the surface of the products was dominated by casein. These
three hypotheses have been investigated by Kim et al. (2003) by
measuring the surface composition of spray-dried milk powder
using ESCA. They determined that the outermost layer of milk pow-
der was dominated by free fat. Under this layer, a layer of fat glob-
ules covered in protein and lactose was found. When the fat layers
were dissolved away using petroleum ether, protein was found to
be preferentially deposited in the next layer before lactose. Faldt
and Bergenstahl (1994) agreed with the view of Meerdink and Riet
(1995) and suggested that the experimental results support the
hypothesis that the difference in diffusivity of various solutes re-
sults in component segregation, since the outermost layers would
be covered by protein if surface activity was the dominating factor,
or no segregation would be found if the rst hypothesis of crust for-
mation alone was the true situation. They also estimated the diffu-
sivity ratio of the different components of milk and showed that the
order of diffusivity supports the segregation hypothesis. This esti-
mation will be discussed in more detail later.
The process of particle drying affects the segregation process,
because the drying process removes water, which then affects
the diffusivities of the solutes through the particles. Hence, drying
models need to be reviewed to assess their suitability for predict-
ing the particle formation process with segregation of different
components, and this review of drying models will be carried out
next.
2.3. A review of existing models of drying processes for the purpose of
predicting component segregation
There are two fundamental approaches to modelling drying
processes, namely lumped-parameter approaches and distrib-
uted-parameter approaches. Lumped-parameter approaches as-
sume that the physical properties and components of the drying
materials remain uniform throughout the thickness or depth in
the material. Examples of such models are the concept of a charac-
teristic drying curve (CDC) and the reaction engineering approach
(REA). In these models, there is no component migration assumed
to occur during drying. On the other hand, distributed-parameter
approaches allow for different regions of the drying material to
have different concentrations of components. Both these ap-
proaches are reviewed in light of how well they are likely to pre-
dict segregation of different components during drying.
The review of drying kinetics has been done to establish and as-
sess the extent to which existing drying theory is able to model the
particle drying process when segregation of different components
occurs. It is shown that lumpedparameter models cannot, in
themselves, account for component segregation, although the
speed of drying is given by such lumped-parameter models. The
reason why the speed of drying is important is that component dif-
16 S. Wang, T. Langrish/ Food Research International 42 (2009) 1325
fusion or migration is very likely to be strongly dependent on the
moisture content. Hence, if the particles dry very quickly, then
there is little time for the components to diffuse, so the segrega-
tion, and the effect of temperature on it, may be reduced.
2.3.1. Lumped-parameter models
2.3.1.1. Characteristic drying curve (CDC). The characteristic drying
curve approach assumes that the shape of the falling-rate drying
curve is a characteristic function of reduced, dimensionless, mois-
ture content for a particular material. This is a lumped-parameter
approach, since the mean value of the moisture content for the en-
tire medium is used. The approach uses a relative drying rate,
which is the specic drying rate normalized relative to the unhin-
dered drying rate (Keey, 1978). To normalize the actual drying rate
and moisture content, Eqs. (1) and (2) are used (Keey, 1978).
f
N
N
max
fnU 1
U
X X
e
X
cr
X
e
2
Here the symbols have the following meaning:
N is the actual drying rate (kg m
2
s
1
).
N
max
is the maximum drying rate (kg m
2
s
1
), which occurs
during the rst drying period when the process is unhindered
by the presence of solids and thus is limited only by the rate
of heat transfer to the particle surface.
U is the normalised moisture content.
X is the moisture content (kg kg
1
).
X
cr
is the critical moisture content (kg kg
1
). Above this critical
moisture content, unhindered drying occurs, and hindered dry-
ing occurs below it.
The CDC suggests that, for a particular material, the shape of the
normalised drying curve is independent of the actual drying condi-
tions. The concept of CDC, although powerful, is not always true.
Langrish and Kockel (2001) cited Keey and Suzuki (1975), who
found that a characteristic curve is found for thin materials with
high effective moisture diffusivity. They also cited Keey and Suzuki
(1975), who found that no characteristic drying curve was found
for drying of large particles of greater than 20 mm in diameter,
but this limitation is unlikely to be signicant for spray-dried par-
ticles, which are typically below 200 lm in size.
Langrish and Kockel (2001) used the specic drying rate de-
scribed by Eqs. (3) and (4) to assess some of the existing experi-
mental work on the drying of milk droplets. They found that for
both skim and whole milk droplets, a linear falling rate curve
was a very good approximation.
N X X
e
bT
b
T
wb
3
Here the symbols have the following meaning:
T
b
and T
wb
are the gas-phase dry-bulb and wet-bulb tempera-
tures, respectively (K).
A
p
is the surface area of the particle (m
2
).
b is the mass-transfer coefcient (s
1
K
1
), calculated using the
equation below (Langrish & Kockel, 2001).
b
a
ms
Ap
X
i
DH
v
4
X
i
is the initial moisture content (kg water kg dry solids
1
), and
a is the heat-transfer coefcient, calculated using the Ranz
Marshall correlation:
Nu
ad
p
k
g
2 0:6Re
0:5
Pr
1=3
5
where
Nu is the Nusselt number.
Re is the Reynolds number.
Pr is the Prandtl number.
d
p
is the particle size (m).
k
g
is the thermal conductivity of the gas (W m
1
K
1
).
2.3.1.2. Reaction engineering approach (REA). This concept considers
the drying process to be an activated-rate reaction. Patel and Chen
(2005) related the specic drying rate to the difference in the va-
pour concentrations between the surface of the droplet and the
bulk air. The equation that quanties the rate of drying follows (Pa-
tel & Chen, 2005):
N
s
h y
v;sat
exp
DE
v
RT
_ _
y
v;b
_ _
6
Here the symbols have the following meaning:
N
s
is the specic drying rate (kg m
2
s
1
).
y
v,sat
and y
v,b
are the saturated vapour concentration above the
particle surface and the bulk vapour concentration, respectively
(kg m
3
).
DE
v
is the activation energy for drying (J mol
1
).
h is the external mass-transfer coefcient (m s
1
). This term is
similar to the b term in Eq. (3), and the units can be converted
to units of kg m
2
s
1
by multiplying by the density of the air.
According to this concept, DE
v
represents the increase in dif-
culty of removing water from the particle as the moisture content
deceases. The value of DE
v
is small when free moisture covers the
surface and, as the amount of free moisture decreases, DE
v
in-
creases. Thus, DE
v
is dened as a function of the free moisture con-
tent of the particle and is tted to the equation below (Patel &
Chen, 2005).
DE
v
DE
v;e
a exp b X X
e

c
_
7
Here the symbols have the following meaning:
DE
v,e
is the activation energy at equilibrium, J mol
1
.
X and X
e
are the moisture content of the particle and the mois-
ture content at equilibrium respectively, kg kg
1
.
The coefcients a, b and c are tted parameters and were cited
to be 0.998, 1.405 and 0.930, respectively, for skim milk (only) in
Patel & Chen, 2005. This equation is likely to be different for other
materials.
2.3.1.3. Critical assessment of CDC and REA approaches. Langrish and
Kockel (2001) used computational uid dynamics modelling to as-
Table 2
Difference between the experimental and predicted results of using both CDC and
REA for milk powders in terms of temperatures and moisture contents (Roos and
Karel, 1991)
REA CDC
T (C) wt% T (C) wt%
Skim milk 1.6 1.1 3.6 3.8
Whole milk 2.0 1.6 3.4 2.9
S. Wang, T. Langrish/ Food Research International 42 (2009) 1325 17
sess the CDC approach for drying of milk, when considering the
data from the literature, nding that the linear falling-rate curve
was an acceptable approximation for most of the data. Patel and
Chen (2005) compared results from simulations using CDC and
REA models. They claimed that REA is a better approach than
CDC since it allows for initial (transient) adjustment periods, par-
ticularly in cases with high relative humidities and high feed con-
centrations. For other conditions, it was concluded that both
methods give similar results. The simulation results were not com-
pared with any experimental data. Both of the lumped-parameter
approach models assume that no segregation occurs within the so-
lid and that the concentrations of the components are the same
throughout the particles.
Chen and Lin (2005) compared the simulated results using both
CDC and REA and compared both simulated results with experi-
mental results for the drying of whole and skimmilk droplets. They
found that the predictions from the REA model are closer to the
experimental results from using CDC. This is shown in Table 2.
Although the predicted results of the REA model are closer to
the experimental results than the results predicted by the CDC
model, the difference is small between both predictions, since this
difference remained under 5% for both models. Chen and Lin
(2005) agreed that, although simple, the CDC provides a good
approximation to experimental results. Furthermore, it was also
recognised that the REA predicts an initial activation energy that
is higher than the actual one because, in the initial drying period,
the milk droplets behave similarly to water, where the activation
energy of drying is close to zero. The REA over predicts this value,
thus causing the initial drying rate to be underestimated, which
may cause the predicted temperature of the droplet to rise faster
than it really does. In their studies, Chen and Lin (2005) found that
this initial period of time, where over-prediction of the activation
energy occurs, is approximately the rst 10% of the total drying
time.
2.3.1.4. Short-cut approach of Thijssen & Coumans (1984). This ap-
proach, which is not widely used, is based on the similarity of
the shapes for the moisture content proles during the diffusional
drying process, assuming that drying occurs by diffusion. In the ini-
tial (penetration period) stages of drying, the moisture-content
proles penetrate, with a similar shape, to the centre of a material,
while during the so-called regular regime, the moisture-content
proles are also self-similar, but decreasing in magnitude through-
out the material, as shown in Fig. 1. More description of this ap-
proach and further details are given in the work of Keey (1992).
Not only is this short-cut approach based on the average mois-
ture content of the material, but it is also limited in assuming that
only diffusion occurs. Like the CDC and REA approaches, the
lumped-parameter nature of the approach means that it is not nor-
mally appropriate to the drying of multi-component mixtures.
2.3.2. Distributed-parameter approaches
Distributed-parameter approaches allow for different regions
within a drying material to contain different concentrations of
the components, with different physical properties. When applied
to the spray drying of milk particles, this means that the different
layers of the spray-dried milk particle may contain different con-
centrations of the components, such as fat and protein. This ap-
proach is therefore, consistent with the observed component-
migration phenomenon (Adhikari et al., 2007; Kim et al., 2003; Nij-
dam & Langrish, 2006). Modelling of the component-migration
process has been attempted by some researchers to predict the
product composition proles. The equations which they have used
are the MaxwellStefan and NernstPlanck equations. These works
are described in more detail next.
2.3.2.1. Diffusion of salt and water in cheese. Morison and Payne
(1999) have modelled the diffusion process of salt and water in
cheese. In this model, they assumed salt to be a single solute and
all other components of cheese to be solids. They modelled the dif-
fusion of salt through water and water through the solids using a
form of the MaxwellStefan equation, as shown in Eq. (8).

n
j1
ji
x
i
J
mj
x
j
J
mi
c
i
D
ij

x
i
RT
r
T;P
c 8
Here the symbols have the following meaning:
x
i
is the mole fraction of specie i.
J
mj
is the total molar ux vector in three dimensions
(mol m
2
s
1
).
D
ij
is the MaxwellStefan diffusivity between species i and j
(m
2
s
1
).
R is the ideal gas constant (8.314 J mol
1
K
1
).
T is the temperature (K).
c
i
is the molar chemical potential of species i (J mol
1
).
r is the Laplacian operator.
This equation has been rearranged to give the formshown in Eq.
(9).
J
n
cB
1
Crx 9
Fig. 1. Moisture-content proles during the diffusional drying of a material,
showing the penetration period and regular regime in the short-cut approach of
Thijssen and Coumans (1984).
Fig. 2. Predictions of salt and moisture concentrations as functions of time,
compared with experimental data for cheese (Morison and Payne, 1999).
18 S. Wang, T. Langrish/ Food Research International 42 (2009) 1325
Here the symbols have the following meaning:
J
n
is the molar diffusional ux vector in three dimensions of spe-
cies with respect to species n (mol m
2
s
1
).
c is the total molar concentration (mol m
3
).
Morison and Payne (1999) used estimated diffusion coefcients
of water and salt through solid cheese from the literature and com-
pared the modelled salt and moisture content with the experimen-
tal data. This comparison is shown in Fig. 2.
As observed in Fig. 2, the model predicted changes in moisture
and salt concentration similar to those observed in the experi-
ments. The small differences were attributed to the inaccuracy
of the literature diffusion coefcients (Morison & Payne, 1999).
This shows that the MaxwellStefan equation is suitable for the
modelling of component migration within cheese. Although this
work attempted the modelling of diffusion of both water and salt
through cheese, it assumed that all other components of the
cheese were in the solid form and thus cannot segregate. This
model describes a system where water and solutes migrates
through a solid matrix, and thus it is difcult to apply it directly
to the spray-drying process, which involves droplets of solution
drying to form particles. The solute in this application of the
model was primarily salt, not surface-active material like dis-
solved casein.
2.3.2.2. Diffusion in the production of supported coimpregnation
catalysts. In the production of supported coimpregnation cata-
lysts, drying conditions may be controlled to produce different
catalyst distributions (Lekhal, Glasser, & Khinast, 2001). During
impregnation, a support is rst soaked in the solution of the pre-
cursor of catalyst. The precursor diffuses into the support and is
adsorbed into the support; the solvent is then evaporated off.
Multi-component diffusion can be controlled to produce catalyst
proles, such as egg-shell, egg-white, egg-yolk and uniform distri-
butions. Egg-shell and egg-yolk refer to situations where the con-
centrations of the various components differ throughout the
particles. Lekhal et al. (2001) modelled the multi-component dif-
fusion and convective process using the NernstPlanck equation
as shown in Eq. (10).
N
l;i
D
l;i
rc
l;i
c
l;i
z
i
D
l;i
F
RT
r/ c
l;i
KK
l;eff
g
l
rP
l
10
Here the symbols have the following meaning:
N
l,i
is the ux of each component (kg m
2
s
1
).
D
l,i
is the diffusion coefcient of the ionic species in the solvent
(m
2
s
1
).
F is Faradays constant (96,500 C mol
1
).
ru is the electrostatic potential gradient (V m
1
).
K and K
l,eff
are the intrinsic and relative permeabilities of the
liquid phase, respectively (m
2
).
c
l,i
is the molar concentration of the specie i in liquid phase,
mol m
3
.
z
i
is the particle charge.
P
l
is the liquid pressure (Pa).
g
l
is the viscosity of the liquid phase (Pa s).
The NernstPlanck equation is used in this case, since the solu-
tion used in the production of coimpreganation catalyst contains
ionic solutes. This model allows the segregation of the solutes by
modelling the multi-component diffusion process. The model,
however, needs signicant adaptation to predict segregation in
the spray-drying process, since it models the diffusion process of
ionic solutes and not proteins and sugars in its current form. It also
does not currently include surface-active components.
2.3.2.3. Drop drying models for carbohydrate solutions. The drying of
drops of solutions has been modelled in many papers for carbo-
hydrates in order to investigate surface stickiness (Adhikari,
Howes, & Bhandari, 2006; Adhikari et al., 2003; Adhikari et al.,
2004). The materials that were investigated include sucrose, glu-
cose, fructose, citric acid and maltodextrin. The distributed-
parameter models used for these works were similar to each
other, in that they were all based on moisture diffusion through
solids. The model divides each droplet into layers, and evapora-
tion only occurs on the surface. The droplet is assumed to be a
non-hollow solid sphere (made of the solutes) and moisture dif-
fuses from the centre to the surface before evaporating. The dis-
solved solutes do not diffuse through the droplet, thus only the
concentration of moisture is different between different layers
of the droplet. The ux equation at the surface is shown below
for water (Adhikari et al., 2003).
N
l;i
k

g;i
A
p
M
w
R
P
atm
T
av;film
ln
1 P
v;db
=P
T
1 P
vs;Td
=P
T
_ _
11
Here the symbols have the following meaning:
k

g;i
is a corrected mass-transfer coefcient (m s
1
).
A
p
is the surface area of droplet (m
2
).
R is the gas constant (J mol
1
K
1
).
T
av,lm
is the lm temperature, the average temperature of the
uid lm in the boundary layer outside each particle (K).
P
atm
, P
v,db
and P
vs,Td
are atmospheric, vapour and saturated
vapour pressures, respectively (Pa).
Adhikari et al. (2003) used the model to estimate the moisture
content at the surface of the dried particle and thus predict the
glass-transition temperature T
g
, which was used as an indicator
of stickiness. They found that this distributed-parameter model
predicted the moisture content of the particles well, with an aver-
age error of 4.56%. Using the model they predicted that maltodex-
trin alters the surface stickiness of the products and thus is an
effective drying aid (Adhikari et al., 2003).
This distributed-parameter approach to modelling predicts dif-
ferences in moisture content in different regions of the particle,
thus predicting different surface stickiness than for the bulk of
the particle. However, the model does not currently allow for sol-
ute diffusion within the particle, since only moisture diffusion is
included. Thus this model cannot be used to predict segregation
of soluble components during spray drying.
2.3.2.4. Amodel for solid formation in spray drying. Seydel, Sengespe-
ick, Blomer, and Bertling (2004) proposed a mathematical model
Fig. 3. Solids formation processes in a spray-dried particle (Seydel et al., 2004).
S. Wang, T. Langrish/ Food Research International 42 (2009) 1325 19
for solid formation during spray drying. In this work, they based
the model for the drying process on the following scenario:
1. Evaporation occurs on the surface boundaries, causing forma-
tion of shell of solids. The formation of solids is based on nucle-
ation of crystals. The permeability of this shell depends on
whether it is made up of crystals or a lm.
2. A bubble can form in the drop, which consists of the drying gas.
This gas can escape the drop if the shell is permeable or remain
trapped if the shell is impermeable.
3. If the bubble remains trapped, the gas expands due to the
increase in temperature and causes ballooning of the particle.
If the shell is rigid, rupture of the shell occurs.
4. If rupture does not occur, the particle shrinks back to the origi-
nal size as the temperature drops.
The model is illustrated in Fig. 3.
Here the symbols have the following meaning:
r
i
is the inner radius of the droplet as well as the radius of the air
bubble (m).
r
old
and r
new
are the initial radius before expansion and nal
radius after expansion of the air bubble, respectively (m).
D is the diffusion coefcient of the liquid phase in the drop
(m
2
s
1
).
v is the velocity of component convection (m s
1
).
w
L
and w
G
are the weight percentages of the solvent and the
solute, respectively.
q
L
and q
G
are the densities of the solvent and the solute, respec-
tively (kg m
3
).
The mass ux equation for this model is shown below.
N
oqw
G
1e
ot

1
r
2
or
2
qw
G
v1e
or
..
convection

1
r
2
o
or
r
2
1eDq
ow
G
or
_ _
..
diffusion

oq
0
e
ot
..
phase transformation

oqw
G
1e
or
dr
i
dt

r r
i
t
r
a
t r
i
t
dr
a
dt

dr
i
dt
_ _ _ _
..
pseudo-convection
12
where
e is the volume fraction of the liquid phase (m
3
m
3
).
q is the density of the pure component (kg m
3
).
Seydel et al. (2004) used this model to predict the conditions
under which the formation of a bubble in a droplet is favoured.
They suggested that hollowed particles are formed preferentially
at high temperatures and with materials having low diffusivity.
Formation of full particles is favoured when lower temperatures
and high diffusivity materials are used. Seydel et al. (2004) ex-
plained that this is due to more even distribution of concentrations
in droplets when a material has a high diffusion velocity. They also
explained that lower temperatures allows more time for this distri-
bution to occur due to slower drying.
The model is a distributed-parameter approach, since it evalu-
ates component concentrations at different distances from the cen-
tre of the droplet. However, in their studies, Seydel et al. (2004)
emphasised the occurrence of hollowed particles and did not at-
tempt to predict the degree of component segregation. Similar to
the work done by Adhikari et al. (2003), this model includes diffu-
sion in a two-phase system (i.e. a liquid phase through the solid
shell) but does not account for the diffusion of solutes within the
solvent in the liquid phase. This means that only water is allowed
to migrate within the particles and that the solutes are assumed to
be xed in their positions. Hence, the model, in its present form,
does not allow for migration of solutes and cannot predict segrega-
tion. The model also addressed the solids-formation process using
crystal nucleation and growth approaches, but many of the prod-
ucts from spray drying are amorphous in nature (White & Cake-
bread, 1966). The extent of crystallization for these particles may
not necessarily be complete in spray drying.
2.3.2.5. Distributed-parameter model for the drying of timber. Nij-
dam, Langrish, and Keey (2000) modelled the drying process of
softwood board using a one-dimensional distributed-parameter
approach. In this model, both uxes of moisture in the vapour
phase and the liquid phase were considered. There are two types
of ux for liquid moisture, the convective uxes caused by capil-
lary pressure and the diffusive uxes caused by concentration dif-
ferences. Gaseous transport mechanisms include gas convection,
vapour and air diffusion. These ux equations are shown below
(Nijdam et al., 2000).
j
wv

q
v
e
k
g
l
g
oP
g
oz

P
g
RT
D
g
M
v
e
_ _
o
oz
P
v
P
g
_ _
13
j
a

q
a
e
p
g
l
g
oP
g
oz

P
g
RT
D
g
M
a
e
o
oz
P
a
P
g
_ _
14
j
wf
q
l
p
l
l
l
o
oz
P
g
P
c
15
Here the symbols have the following meaning:
j
wv
, j
a
and j
wf
are the uxes of the water vapour, the air and the
free (unbound) water, respectively (kg m
2
s
1
).
q
v
, q
a
and q
l
are the densities of the vapour, the air and the
liquid, respectively (kg m
3
).
p
g
and p
l
are the permeabilities of the moist solids to gas and
liquid ows, respectively (m
2
).
l
g
and l
l
are the dynamic viscosities of the gas and the liquid,
respectively (kg m
1
s
1
).
P
g
, P
c
, P
v
and P
a
are the gas, capillary, vapour and air pressures,
respectively (Pa).
D
g
is the gas diffusion coefcient (m
2
s
1
, or kg m
1
s
1
when
multiplied by the component density).
M
v
and M
a
are the molar masses of the vapour and the air,
respectively (kg mol
1
).
e is the void fraction (m
3
m
3
).
In this model, Eqs. (13) and (14) describe the ux of the airva-
pour movement through the wood bres and Eq. (15) describes the
movement of unbound water. The model simulates uxes of both
liquid and gases that move through the solid wood bres and thus
is signicantly different to the case of spray drying, where droplets
of solution are dried and form solids. The components that are in-
cluded in the simulations are those for the water vapour, liquid
water, wood and air.
2.3.2.6. Modelling of segregation using slab experiments. Meerdink
and Riet (1995) compared experimental results from drying slabs
of agar, containing a ternary system of water, sucrose and casein,
with the simulated result from a multi-component diffusion mod-
el. The equations in their model follow:
oC
w
ot

o
or
E
ww
oC
w
or

o
or
E
ws
oC
s
or
16
oC
s
ot

o
or
E
ww
oC
w
or

o
or
E
ss
oC
s
or
17
20 S. Wang, T. Langrish/ Food Research International 42 (2009) 1325
Here the symbols have the following meaning:
C
w
and C
s
are the molar concentrations of water and sucrose,
respectively (mol m
3
).
t is the drying time (s).
E is an element of matrix of transformed diffusion coefcients
(m
2
s
1
).
Since the model was written to simulate a slab-drying experi-
ment, water is assumed to evaporate froma front that progressively
withdraws into the slab as evaporation continues, leaving behind a
slab consisting of the dry solutes. The model does not include the
surface activity of casein and only considers the segregation due
to the differences in the diffusion rate between sucrose and casein
towards the centre. This model also does not explicitly account for
the effect of the differences in solubilities of the solutes. Instead, it
assumes that the diffusivities of solutes are dependent on the mois-
ture content. This means that, whenthe moisture content decreases,
the diffusivities of the solutes also change and, in the limit of no
water present, the diffusion coefcients approach zero. Although
this approach simulates the diffusion process as the concentrations
of solutes increase due to evaporation, it also implies that the
solutes that have precipitated (as solids within the spray-dried par-
ticles) continue to diffuse slowly as long as small amounts of mois-
ture are present. This is unlikely to be true when the solutes have
precipitated out of solution as solids (Cussler, 1997).
2.3.2.7. Critical analysis of existing models of drying for the purpose of
predicting component segregation. The existing models of drying re-
viewed above are summarised in Table 3.
As stated in Table 3, the lumped-parameter models do not allow
for the migration of dissolved components of milk. These lumped-
parameter approaches to modelling the spray-drying process can
still be used to predict the overall drying kinetics. They involve less
experimental effort to use than distributed-parameter models, in
that fewer experiments are required to quantify the tted param-
eters. However, they effectively assume that the concentrations of
various components within the particles are the same, which
means that these models cannot predict the occurrence of compo-
nent segregation.
Most of the distributed-parameter models fromthe literature do
not account for or allow for the phenomenon of the migration of
dissolved components of milk during drying, as observed. These
models generally assume that the dissolved components are in so-
lid formand that water diffuses through them(Adhikari et al., 2003;
Morison & Payne, 1999; Seydel et al., 2004). The models of Lekhal
et al. (2001) and Nijdam et al. (2000) both describe systems where
water and solutes may diffuse through solid matrices and allow
segregation to occur through multi-component diffusion. These
systems are different to the case of spray drying, in that droplets
are formed initially in spray drying, which then solidify as water
is evaporated. Thus these models cannot currently be used to pre-
dict the extent of component segregation in spray drying. Both
the distributed-parameter models set up by Seydel et al. (2004)
and Meerdink and Riet (1995), to simulate the spray-drying pro-
cess, allow for component migration through diffusion. However,
they do not yet account for surface-active components. Both mod-
els also do not account for the effects of different solubilities of the
solutes explicitly. To understand and predict this, a new distrib-
uted-parameter model needs to be set up to study the mechanisms
of diffusion for not only the solvent, but also the dissolved compo-
nents. The effects of different solubilities of solutes and surface-ac-
tive components also need to be incorporated into this new model.
2.4. A new distributed-parameter multi-component model
To understand the mechanisms of component segregation and
thus predict the surface properties of spray-dried products, a
new distributed-parameter model need to be developed. This mod-
el need to allow for the diffusion of all dissolved components,
while also taking the effects of their solubilities and surface-active
properties into account. These migration mechanisms are ex-
plained in detail next.
2.4.1. The diffusion process
Diffusion plays an important part in both hypotheses two and
three. In hypothesis two, the diffusivities of different components
affect the order in which they are segregated. In hypothesis three,
diffusion of protein to the surface may be the rate-limiting step. In
order to better understand the process for spray-drying liquid
products, the diffusion process needs to be studied in detail and
modelled.
All forms of the diffusion equation calculate the ux using a
driving-force term, such as the difference in concentration and a
diffusion coefcient. The diffusion coefcient for mass transfers
in a liquid can be estimated using the StokesEinstein equation
as shown in Eq. (18) Cussler, 1997.
D
K
B
T
6plR
0
18
Here the symbols have the following meaning:
R
0
is the component particle size (m).
D is the diffusion coefcient (m
2
s
1
).
K
B
is Boltzmanns constant (1.38 10
23
J K
1
).
T is the temperature (K).
Table 3
Existing drying models in the literature and the experimental work involved when using these models
Existing models Description Experimental determination of parameters
CDC (lumped-parameter) Shape of drying curve independent of drying conditions Need to measure drying-rate curves under different drying conditions
as well as different sample congurations to assess the shape of drying
curves
Lumped-parameter model does not predict segregation of
components
REA (lumped-parameter) Activated rate process with DE
v
representing the increase
in difculty of removing water from the particle as the
moisture content deceases
Relationship of DE
v
with moisture content needs to be determined. In
theory, only one experiment is needed, but in practice experiments are
required to check if DE
v
values are unique for different drying
conditions and congurations of samples Lumped-parameter model does not predict segregation of
components
Existing distributed-parameter
approach for spray drying
Diffusion of water from centre to surface. Moisture content
is allowed to be different through the particles
Experiments are required to estimate all parameters used in the model.
More experiments are needed than the number of parameters
Solubilities of dissolved components and surface activity
are not considered
S. Wang, T. Langrish/ Food Research International 42 (2009) 1325 21
l is the viscosity of the solvent (kg m
1
s
1
).
This equation suggests that diffusion coefcients increase with
temperature and decrease with solvent viscosity and the size of the
molecule. Diffusion is often accompanied by convection, and the
total mass ux is the sum of the ux due to diffusion and the ux
due to convection, as explained by Cussler (1997).
Some suitable equations, which include both convective and
diffusion components, are shown in Table 4.
In these equations, n is the total ux, j is the ux due to diffu-
sion, v is the average solute velocity, c is the local concentration,
V is the volume average velocity and y is the molar average velocity
(Cussler, 1997).
2.4.2. Diffusion coefcients
Many attempts have been made to determine the diffusion
coefcients for a large variety of solutes. Some of these solutes in-
clude components of milk, carbohydrates and amino acids. Some of
these coefcients may be estimated using the StokesEinstein
equation, while others have been experimentally quantied by
experimental methods, such as the Taylor dispersion method.
The source, species and descriptions of the diffusion coefcients
quantied in the literature for carbohydrates, protein and fat are
summarised in Table 5. Some of these works are further explained
next.
2.4.2.1. Milk components. As mentioned before, Kim et al. (2003)
have estimated the diffusivity ratios for the different components
of milk using Eqs. (18) and (19). Eq. (18) is the StokesEinstein
equation, as previously mentioned.
D
i
D
lactose

R
0;lactose
R
0
; i
_ _
n
19
Here i is the component and n = 1, 2 or 3.
The estimated diffusivity ratios are summarised in Kim et al.
(2003), who then explained that according to this equation, and
its empirical extensions (for higher concentrations), the diffusion
coefcient can be proportional to R
n
0
(n = 1, 2 or 3). This means
that the diffusion coefcient is inversely related to the size of the
solute particles (to some power) and thus the diffusivity ratios be-
tween lactose and other components can be estimated by Eq. (19).
This information is shown in Table 6.
Table 4
Examples of some diffusion equations (Roos, 1995)
Choice Total ux Diffusion equation Reference velocity Best used for
Mass n
1
j
m
1
c
1
v j
m
1
q
1
v
1
v Dqrx
1
v x
1
v
1
x
2
v
2
qv n
1
n
2
Constant density liquids; coupled mass and momentum
transport
Volume n
1
j
1
c
1
v
0
j
1
c
1
v
1
v

Drc
1
v
0
c
1
V
1
v
1
c
2
V
2
v
2
V
1
n
1
V
2
n
2
For constant density liquids and for ideal gases; may be used
either mass or mole concentration
NernstPlanck ry
1

y
1
y
2
D
v
2
v
1
none Suitable for ionic solutes in electrolyte
Table 5
Summary of sources, species and description of diffusion coefcients estimated in the existing literature
Source Species Description
Hennings et al. (2001) and Kim et al. (2003) Free fat Diffusion coefcients estimated based on size of particles using Eq. (2)
Fat globules
Casein micelles
Casein sub-units
Whey protein
Lactose
Salts
Chen and Lin (2005) and Ribeiro et al. (2006) Lactose Diffusion coefcients measured experimentally for different temperatures and concentrations
Sucrose
Thijssen and Coumans (1984) and Wu et al. (2001) L-Proline Diffusion coefcients measured experimentally for different concentrations
L-Threonine
L-Arginine
Table 6
Relative sizes of molecules and other components in milk and their diffusivity ratios
in an aqueous solution (Hennings et al., 2001)
Type of particles Size (diameter, m) Diffusivity ratio
R
1
0
R
2
0
R
3
0
Free fat >10
5
>10
4
>10
8
>10
12
Fat globules 10
5
10
6
10
4
10
8
10
12
Casein micelles 10
7
10
8
10
2
10
4
10
6
Casein sub-units 10
8
10
9
10
1
10
2
10
3
Whey protein 10
8
10
9
10
1
10
2
10
3
Lactose 10
9
10
10
1 1 1
Salts 10
9
10
10
1 1 1
Table 7
Mutual diffusion coefcient of aqueous lactose solution and the respective standard deviations, D S
D
, at different temperatures and concentrations (Chen and Lin, 2005)
s (mol/dm) Dc (mol/dm
3
) D S
D
(10
9
m
2
s
1
)
T = 298 K T = 303 K T = 308 K T = 312 K T = 318 K T = 328 K
0.001 0.568 0.035 0.643 0.010 0.723 0.010 0.789 0.030 0.862 0.057 1.060 0.099
0.005 0.005 0.565 0.040 0.639 0.011 0.720 0.085 0.785 0.025 0.855 0.082 1.058 0.074
0.010 0.01 0.541 0.037 0.631 0.012 0.708 0.035 0.774 0.040 0.850 0.031 1.044 0.009
0.100 0.06 0.541 0.012 0.602 0.005 0.677 0.027 0.740 0.033 0.785 0.023 1.018 0.009
22 S. Wang, T. Langrish/ Food Research International 42 (2009) 1325
As shown in Table 6, the diffusivity ratio of free fat to lactose is
the lowest of all the components, which suggests that its diffusion
rate is the slowest. This component is followed by fat globules,
which is the next in order of diffusivity ratio. The diffusivity ratios
are order of magnitude estimates based on approximate sizes of
the particles. Given the absolute values of the diffusion coefcient
of lactose, the diffusion coefcients of the individual components
can then be estimated.
2.4.2.2. Carbohydrates in aqueous solutions. Ribeiro et al. (2006)
have measured the diffusion coefcients of lactose, sucrose, glu-
cose and fructose in aqueous solutions using the Taylor dispersion
method. In this method, a small amount of solution is injected into
a carrier solution of a slightly different concentration, which is
owing in a laminar stream through a long capillary tube. As the
carrier solution ows along the tube, the change in the concentra-
tions caused by the diffusion of the solute is measured by a refrac-
tometer at the end of the tube. These results are then used to nd
the diffusion coefcients of the solute. The experiments were con-
ducted in concentrations between 0.001 to 0.1 mol/L, and temper-
atures between 298.15 K and 328.15 K. The experimental results
for the diffusion coefcients of lactose and sucrose in aqueous
solutions at different temperatures are reported in Tables 7 and 8.
As observed in both Tables 7 and 8, the diffusion coefcients of
both sugars increase with temperature and decrease with increas-
ing concentrations. The diffusion coefcient of lactose is slightly
higher than that of sucrose at the same conditions, for example,
at 0.001 M and 298.15 K, D
lactose
is 0.568 10
9
m
2
s
1
and D
sucrose
is 0.525 10
9
m
2
s
1
. The small difference between the values
probably reects the fact that the sugars have different structural
arrangements, despite having the same molecular formula.
Wu, Ma, Liu, & Li, 2001 have also measured diffusion coef-
cients of sucrose in more concentrated solutions at 298.15 K. The
results are shown in Table 9.
As observed in Table 9, the diffusion coefcient of sucrose de-
creased from 5.22 10
6
cm
2
s
1
to 2.88 10
9
cm
2
s
1
as the
concentration increased from innite dilution to 0.772 M. This
trend is similar to that observed by Ribeiro et al. (2006). For the dif-
fusion coefcient at 298.15 K and 0.1 M, the value reported by
Ribeiro et al. (2006) is 0.504 10
9
m
2
s
1
, while that reported
by Wu et al. (2001) is 0.490 10
9
m
2
s
1
. The values are close
and consistent with each other.
2.4.2.3. Amino acids. Cussler (1997) has explained that, since amino
acids contains side chains which end in amino and carboxylic acid
groups, their diffusion coefcients are likely to be pH dependent.
This is because, at low pH, the amino groups become positively
charged (NH

3
) and the carboxylic group becomes negatively
charged (COO

). The pH will affect the diffusion coefcient since


it changes the concentration of electrolytes. However at intermedi-
ate pH, the protein has no net charge and the numbers of positive
and negative charges are equivalent. Thus during measurements of
diffusion coefcient for proteins, pH needs to be carefully con-
trolled to ensure the accuracy of the results.
Table 8
Mutual diffusion coefcients of aqueous sucrose solutions and the respective standard deviations, D S
D
, at different temperatures and concentrations (Chen and Lin, 2005)
s (mol/dm) Dc (mol/dm
3
) D S
D
(10
9
m
2
s
1
)
T = 298 K T = 303 K T = 308 K T = 312 K T = 318 K T = 328 K
0.002 0.005 0.525 0.009 0.611 0.009 0.711 0.008 0.804 0.015 0.804 0.015 0.917 0.02
0.005 0.005 0.5621 0.002 0.607 0.004 0.706 0.011 0.799 0.010 0.799 0.010 0.911 0.02
0.010 0.005 0.520 0.005 0.605 0.012 0.704 0.017 0.793 0.015 0.793 0.015 0.905 0.01
0.100 0.005 0.504 0.004 0.594 0.017 0.692 0.007 0.775 0.011 0.775 0.011 0.888 0.012
Table 9
Diffusion coefcients of sucrose in solutions up to 0.772 M (Thijssen and Coumans,
1984)
C (mol L
1
) D
exp
(10
6
cm
2
s
1
)
0 5.22
0.1 4.90
0.2 4.58
0.3 4.27
0.4 3.96
0.5 3.66
0.6 3.37
0.7 3.08
0.003 5.21
0.152 4.73
0.409 3.94
0.772 2.88
Table 10
Diffusion coefcients of L-proline, L-threonine and L-arginine (Thijssen and Coumans,
1984)
C
M1
(mol L
1
) C
M2
(mol L
1
) C
M
(mol L
1
) D (10
6
cm
2
s
1
)
L-Proline
0.0058 0.004 0.031 8.76
0.096 0.007 0.052 8.72
0.183 0.013 0.098 8.66
0.234 0.016 0.125 8.64
0.451 0.315 0.383 8.34
0.864 0.735 0.799 7.79
1.298 1.066 1.182 7.44
1.837 1.634 1.736 6.74
2.392 2.189 2.291 6.29
2.858 2.826 2.842 5.75
3.298 3.176 3.237 5.55
3.931 3.742 3.836 5.16
4.338 4.223 4.280 4.73
4.706 4.588 4.647 4.32
L-Threonine
0.060 0.004 0.032 8.04
0.085 0.005 0.045 8.00
0.157 0.052 0.105 7.86
0.240 0.117 0.178 7.61
0.322 0.215 0.269 7.55
0.405 0.293 0.349 7.49
0.509 0.379 0.444 7.12
0.594 0.471 0.532 7.08
0.653 0.556 0.604 6.83
L-Arginine
0.054 0.003 0.029 7.44
0.085 0.005 0.045 7.34
0.109 0.038 0.073 7.16
0.160 0.054 0.107 7.05
0.215 0.078 0.146 6.93
0.271 0.138 0.204 6.76
0.333 0.198 0.266 6.65
0.392 0.241 0.316 6.57
0.442 0.347 0.394 6.42
0.558 0.463 0.511 6.28
0.676 0.536 0.606 6.15
0.732 0.583 0.635 6.11
0.787 0.630 0.707 6.01
S. Wang, T. Langrish/ Food Research International 42 (2009) 1325 23
Wu et al. (2001) have measured the diffusion coefcients of the
amino acids, namely L-proline, L-threonine and L-arginine, using a
diffusion cell. The diffusion cell consists of two compartments
which are separated by a micro porous diaphragm with a xed per-
meability. Solutions of different concentrations are placed within
the compartments at the start of the experiment, and diffusion is
allowed to happen for a xed period, in their case 1.5 h. At the
end of the experiment, the changes in concentrations of the solu-
tions in the compartments are measured using a polarimeter. To
measure the diffusion coefcients, the nal concentrations of the
solutions in the two compartments are measured as well as the
duration over which the experiment took place. The diffusion coef-
cients are then estimated using Eq. (20).
D
1
bt
ln
C
2f
C
1f
C
2i
C
1i
20
Here the symbols have the following meaning:
D is the diffusion coefcient (cm
2
s
1
).
C are the concentrations, the subscripts f symbolize the nal
concentrations, and i indicates the initial concentrations
(mol cm
3
).
b is a constant that is specic to the diaphragmcell and depends
on the different diaphragms used (cm
2
).
t is the duration over which diffusion is measured (s).
A solution with known diffusion coefcient is used to measure
b. Wu et al., 2001 chose to use potassium chloride solution at 25 C,
and b was measured for their cell to be 6.60 cm
2
. The average con-
centrations of the solutions in the two compartments, C
M1
and C
M2
,
are measured using initial and nal concentrations and C
M
, the
average concentration of the solution within the cell, is estimated.
The diffusion coefcients measured, and the different solutions
used, are tabulated in Table 10.
Wu et al. (2001) tted the results into Eq. (21), which is a semi
empirical model built upon Gordons model. Gordons model is
widely used for predicting the diffusion coefcient of dipolar
molecular and non-electrolyte solutions but can only be used for
dilute solutions. The tted form shown in Eq. (21) allows the diffu-
sion coefcient to be estimated for a wide range of concentrations
by tting empirical parameters to account for the effect of chang-
ing concentrations.
D A
1
D
0
g
A
2
r
1
C
C
0
_ _ _ _
A
3
21
Here the symbols have the following meaning:
A
1
, A
2
and A
3
are tted empirical parameters.
D
0
is the diffusion coefcient at a concentration of 1 g
(100 ml)
1
.
C
0
is the molarity of 1 M (g
A2
r
) is the viscous factor.
[1 C=C
0

A3
] is the thermodynamic factor.
In their work, Wu et al. (2001) found the value of A
1
, by tting
the empirical results, to be 0.4655371, A
2
to be 0.3065743 and
A
3
to be 0.05312948. These parameters could be potentially ex-
tended to other non-electrolytes. Although quantied, the values
of the parameters were not compared with any existing literature
values.
For the purpose of this model, the diffusion coefcients of case-
in have been estimated using the method shown in Kim et al.
(2003), since no experimental value for the diffusion coefcient
of casein is available. To ensure consistency, the diffusion coef-
cient of lactose has also been estimated using the same method.
This is justied because, even though these values might not be ex-
act, they are of a similar order of magnitude to the experimental
values. The experimental results found by Ribeiro et al. (2006)
range between 0.5 10
9
and 1.1 10
9
m
2
s
1
, in comparison
with the range of 10
9
10
10
m
2
s
1
, as estimated by Kim et al.
(2003). The ranges overlap, and thus it is suggested that the esti-
mated diffusion coefcients are initially sufcient for modelling
purposes.
2.4.3. Implementations of surface activity effects
Kim et al. (2003) have described the third hypothesis for com-
ponent migration as being caused by surface activity. In this
hypothesis, the proteins in the regions near the liquid/air interface
of the droplet absorb preferentially to the surface of the particles
and form a monolayer. This monolayer of protein then dries up
upon initial heating and thus deprives the solution near the surface
of any dissolved protein. This causes a concentration gradient
through the droplet and drives diffusion of protein from local re-
gions in the droplet towards the surface. It is possible that the
majority of the surface-active component may diffuse towards
the surface. The surface-active solute will be present in the solu-
tion/solid matrix within the droplet during drying. The surface
activity of the proteins may result in them being pulled to the sur-
face rapidly and precipitating, possibly causing the concentration
of the solute in the liquid to be low under the surface region. This
then simplies the effect of the surface tension to that of diffusion
of the surfactants from the centre to the surface of the droplet, as
assumed by Meerdink and Riet (1995).
To model the effect of the surface activities of the proteins, the
process of the initial formation of the monolayer must be consid-
ered. Graham and Phillips (1979) investigated the rate of migration
of casein to the surface from a bulk solution. In their experiment,
they introduced a homogenous solution of casein into a container
and monitored the changes in surface concentration casein at dif-
ferent times. The path length for the diffusion of casein was 1 cm,
and the experiment was performed at 20 C. These workers found
that casein reached 90% of its nal surface concentration in
approximately 4 h and that the rate of change of surface concentra-
tion in this region is diffusion controlled. This information can be
used to estimate the time needed for casein in the layer immedi-
ately below the surface to diffuse to the surface in the new distrib-
uted-parameter model, where the diffusion coefcient may be
assumed to be directly proportional to the absolute temperature,
according to the StokesEinstein equation. From the residence
time of particles in the spray dryer, the maximum path length
for the diffusion of the proteins can then be estimated. This path
length can then be compared with the thickness of the surface
layer in the distributed-parameter model to ensure that all of the
protein from this shell can migrate to the surface during spray
drying.
3. Conclusions
To improve the spray-drying process of sticky materials con-
taining low molecular weight sugars and organic acids, the segre-
gation process of the components in the particles has been
discussed. Existing evidence for component segregation in milk
powder has been reviewed, and the hypotheses for this process,
namely crust formation, solubility and surface activity, have also
been explored. To assess the conditions under which these pro-
cesses are dominant, simulations of particles drying in spray dryers
need to be carried out. Existing lumped-parameter approaches, as
well as distributed-parameter approaches, have been reviewed. A
new distributed-parameter model needs to be developed to allow
for the segregation of multiple components, as well as accounting
for their solubility and surface activities. This new model can then
24 S. Wang, T. Langrish/ Food Research International 42 (2009) 1325
be used to predict the surface as well as the bulk properties of the
products. To evaluate the effectiveness of using additives as drying
aids during spray drying, previous research has been reviewed,
suggesting that very low amounts of casein (1% overall) are re-
quired to dominate the surface composition of the particles, but
the effect of casein additives on the overall yield has not been mea-
sured yet. To further understand the effect of the casein on the
spray-dried products, there is a need to quantify the yields using
small amounts of casein as additives in spray drying and to assess
the effects of different operating conditions during spray drying on
the segregation. In addition, the most effective proteins, and the
reasons for their effectiveness, need to be assessed, since whey
proteins and casein have different properties.
References
Adhikari, B., Howes, T., & Bhandari, B. R. (2006). Use of solute xed coordinate
system and method of lines for prediction of drying kinetics and surface
stickiness of single droplet during convective drying. Chemical Engineering and
Processing, 46(5), 205419.
Adhikari, B., Howes, T., Bjandari, B. R., & Troung, V. (2004). Effect of addition of
maltodextrin on drying kinetics and stickiness of sugar and acid-rich foods
during convective drying experiments and modelling. Journal of Food
Engineering, 62(1), 5368.
Adhikari, B., Howes, T., Shrestha, A., & Bhandari, B. R. (2007). Effect of surface
tension and viscosity on the surface stickiness of carbohydrate and protein
solutions. Journal of Food Engineering, 79(4), 11361143.
Adhikari, B., Howes, T., & Troung, V. (2003). Surface stickiness of drops of
carbohydrates and organic acid solutions during convective drying:
Experiments and modelling. Drying Technology, 21(5), 839873.
Australian Bureau of Statistics. The Australian Dairy Industry. <http://
www.abs.gov.au/ausstats/abs@.nsf/Previousproducts/1301.0Feature%20Article-
182004?opendocument&tabname=Summary&prodno=1301.0&issue=2004&
num=&view=> Accessed 14.11.07.
Bhandari, B. R., Datta, N., Crooks, R., Howes, T., & Rigby, S. (1997). A semi-empirical
approach to optimise the quantity of drying aids required to spray dry sugar-
rich foods. Drying Technology, 15(10), 25092525.
Bhandari, B. R., & Howes, T. (1999). Implication of glass transition for the drying and
stability of foods. Journal of Food Engineering, 40(12), 7179.
Cano-Chauca, M., Stringheta, P. C., Ramos, A. M., & Cal-Vidal, J. (2005). Effect of the
carriers on the microstructure of mango powder obtained by spray drying and
its functional characterization. Innovative Food Science and Emerging
Technologies, 6(4), 420428.
Chan, H.-K. (2006). Dry powder aerosol drug deliveryOpportunities for colloid and
surface scientists. Colloids and Surfaces A: Physiochemical Engineering Aspects,
284285(A), 5055.
Chen, X. D., & Lin, S. X. Q. (2005). Air drying of milk droplets under constant and
time-dependent conditions. American Institute of Chemical Engineers Journal,
51(6), 17901799.
Chidavaenzi, O. C., Buckton, G., & Koosha, F. (2001). The effect of co-spray drying
with polyethylene glycol 4000 on the crystallinity and physical form of lactose.
International Journal of Pharmaceutics, 216, 4349.
Cussler, E. L. (1997). Diffusion mass transfer in uid systems (2nd ed.). Cambridge
University Press (pp. 114119).
Faldt, P., & Bergenstahl, B. (1994). The surface composition of spray-dried protein
lactose powder. Colloids and Surfaces A: Physicochemical and Engineering Aspects,
90(23), 183190.
Graham, D. E., & Phillips, M. C. (1979). Proteins at liquid interfaces: 1. Kinetics of
adsorption and surface denaturation. Journal of Colloid and Interface Science,
70(1), 403414.
Hennings, C., Kockel, T. K., & Langrish, T. A. G. (2001). New measurements of the
sticky behaviour of skim milk powder. Drying Technology, 19(34), 471484.
Keey, R. B. (1978). Introduction to industrial drying operations. Oxford: Pergamon (pp.
154156).
Keey, R. B. (1992). Drying of loose and particulate materials. New York: Hemisphere.
Keey, R. B., & Suzuki, M. (1975). On the characteristic drying curve. AIChE Symposium
Series, 73(163), 4756.
Kim, E. H.-J., Chen, X. D., & Pearce, D. (2003). On the mechanisms of surface
formation and the surface composition of industrial milk powder. Drying
Technology, 21(2), 265278.
Kuo, M., DSouza, A. J. M., Tep, V., Gordon, M. S., Schiavone, H., Charan, C., et al.
(2007). Preparation of stable and dispersible dry powder aerosol formulations by
spray drying. San Carlos, California/Lawrence, Kansas: Inhale Therapeutic
Systems, Inc./The University of Kansas (<http://www.nektar.com/pdf/
dried_powder.pdf> Accessed 14.11.07).
Langrish, T. A. G., & Kockel, T. K. (2001). The assessment of a characteristic drying
curve for milk powder for use in computational uid dynamics modelling.
Chemical Engineering Journal, 84(1), 6974.
Lekhal, A., Glasser, B. J., & Khinast, J. G. (2001). Impact of drying on the catalyst
prole in supported impregnation catalyst. Chemical Engineering Science, 56(15),
44734487.
Masters, K. (1996). Deposit-free spray drying: Dream or reality. In Proceedings of the
10th international drying symposium, IDS96, Krakow, Poland, 30 July2 August
1996 (Vol. A, pp. 5260).
Meerdink, G., & Riet, K. v. (1995). Modelling segregation of solute material during
drying of liquid foods. AIChE Journal, 41(3), 732736.
Morison, K. R., & Payne, M. R. (1999). A multi-component approach to salt and water
diffusion in cheese. International Dairy Journal, 9(12), 887894.
National Honey Board, Dried Honey Product. <http://www.honey.com/downloads/
driedhoney.pdf> Accessed 05.10.07.
Nijdam, J. J., & Langrish, T. A. G. (2006). The effect of surface composition on the
functional properties of milk powder. Journal of Food Engineering, 77(4),
919925.
Nijdam, J. J., Langrish, T. A. G., & Keey, R. B. (2000). A high-temperature drying model
for softwood timber. Chemical Engineering Science, 55(18), 35853598.
Patel, K. C., & Chen, X. D. (2005). Prediction of spray-dried product quality using two
simple drying kinetic models. Journal of Food Process Engineering, 28(6),
567594.
Ribeiro, A. C., Ortona, O., Simones, S. M. N., Santos, C. I. A. V., Prazeres, P. M. R. A.,
Valente, M. J. A., et al. (2006). Binary mutual diffusion coefcient of aqueous
solutions of sucrose, lactose, glucose and fructose in the temperature range
from (298.15 to 328.15) K. Journal of Chemical and Engineering Data, 51(5),
18361840.
Roos, Y. H. (1995). Characterization of food polymers using state diagrams. Journal
of Food Engineering, 24(3), 339360.
Roos, Y. H., & Karel, M. (1991). Plasticizing effect of water on thermal behaviour and
crystallization of amorphous food models. Journal of Food Science, 56, 3843.
Roustapour, O. R., Hosseinalipour, M., & Ghobadian, B. (2006). An experimental
investigation of lime juice drying in a pilot plant spray dryer. Drying Technology,
24(2), 181188.
Seydel, P., Sengespeick, A., Blomer, J., & Bertling, J. (2004). Experiment and
mathematical modeling of solid formation at spray drying. Chemical
Engineering and Technology, 27(5), 505510.
Tan, J. P. K., Goh, C. H., & Tam, K. C. (2007). Comparative drug release studies of two
cationic drugs from pH-responsive nanogels. European Journal of Pharmaceutical
Sciences, 32(45), 340348.
Thijssen, H. A. C., & Coumans, W. J. (1984). Short-cut calculation of non-isothermal
drying rates of shrinking and non-shrinking particles containing an expanding
gas phase. In Proceeding of the 4th IUFRO international drying symposium, IDS 84,
Kyoto, Japan (Vol. 1, pp. 2230).
Truong, V., Bhandari, B. R., & Tony, Howes (2005). Optimization of cocurrent spray
drying process for sugar-rich foods. Part IIOptimization of spray drying
process based on glass transition concept. Journal of Food Engineering, 71(1),
6672.
White, G. W., & Cakebread, S. H. (1966). The glassy state in certain sugar containing
food products. Journal of Food Technology, 1(1), 7382.
Wu, Y., Ma, P., Liu, Y., & Li, S. (2001). Diffusion coefcients of L-proline, L-threonine
and L-arginine in aqueous solutions at 25 C. Fluid Phase Equilibria, 185(12),
2738.
S. Wang, T. Langrish/ Food Research International 42 (2009) 1325 25

Вам также может понравиться