Separation and Purification Technology 48 (2006) 159–166

Dynamic modelling optimisation and control of lactose crystallisations:

Comparison of process alternatives
T.T.L. Vu a,∗ , R.J. Durham a , J.A. Hourigan a , R.W. Sleigh b
a Food Research Group, CPAFS, University of Western Sydney, Locked Bag 1797, Penrith South DC, NSW 1797, Australia
b Food Science Australia, Sydney, NSW 2113, Australia

Abstract

A general technique is derived to formulate and solve the dynamic optimisation problems for crystallisation, in order to re-evaluate the
growth rate correlations and search for the optimal cooling and heating strategies (evaporative mode only). The dynamic model of a process
making ␣-lactose monohydrate is first revised for these operations using seed crystals: batch cooling, semi-batch cooling and heating, and
continuous cooling. Experiments are conducted in 2 and 20 L crystallisers under various seeding, cooling and heating strategies to evaluate
the constants in the power law function, which correlates the growth rate of the particles with supersaturation. Using a nonlinear least square
method, the best curve fitting of experimental data to the model yields G = 0.007T(100(x␣ /xw ) − sol␣ )2.7 . The value of n = 2.7, which is
applicable to all operations investigated in the study. The value of k however, must be estimated for selected seeding and cooling strategies,
and most importantly for the type of syrup fed to the crystalliser. Using k = 0.007 in simulations and experiments of jet-milled seeds at 1.5%
seeding ratio, the experimental results agreed well with predictions before the growth became slow. A semi-batch is slightly faster than a batch
process and evaporative semi-batch is much faster than cooling operation but is more difficult to control. The performance of a cooling and
seeding run in continuous mode is simulated. The system reached steady state after seven residence times but the predicted particle size could
only be stabilised after 100 h. The developed methodology can be applied to the crystallisation of many other substances but is not limited to
the food or dairy industries.
© 2005 Published by Elsevier B.V.

Keywords: Continuous performance; Cooling profile; Evaporative semi-batch; Growth rate; Cooling crystallisation; Seeding

1. Introduction
The possibility of growing pure crystals of a controlled
size distribution has made crystallisation one of the most
important purification and separation techniques in the food
industries, particularly in the sugar and dairy industries.
While either evaporative vacuum batch or evaporation in con-
tinuous mode is the only technique applied to extract sugar
from cane juice, both cooling and heating can be employed
to create supersaturation in crystallisation of lactose from
whey permeate, a yellow waste material, which contains min-
eral salts and nitrogenous compounds, remaining after cheese
manufacturing. Lactose is mainly manufactured by cooling
crystallisation of whey permeate in the dairy industry since
the lactose solubility–temperature curve is not flat and its sol-
∗ Corresponding author. Tel.: +61 2 4570 1426; fax: +61 2 4570 1579.
E-mail address: l.vu@uws.edu.au (T.T.L. Vu).
ubility at the ambient temperature is low. Last but not least,
the difference between the metastable concentration and sol-
ubility of lactose, referred to as the metastable zone, is quite
large, which allows controlled cooling in seeded batch oper-
ation to minimise unwanted nucleation.
In 1998 the Centre for Advanced Food Research, in con-
junction with Food Science Australia developed new technol-
ogy to generate pure lactose syrup from whey ultra-filtrate [1].
This opens major new opportunities for crystallisation, since
for the first time crystallisation can be optimised for particle
formation rather than purification as in the current indus-
trial practices of manufacturing ␣-lactose monohydrate. In
order to maximise the advantages of using high purity lactose
feed for crystallisation, different operating conditions must
be compared, e.g., batch versus continuous modes, cooling
versus evaporation. The best seeding and cooling or heating
strategies must be selected to ensure the highest yield and
quality of the product. To minimise the number of labora-

1383-5866/$ – see front matter © 2005 Published by Elsevier B.V.

doi:10.1016/j.seppur.2005.07.015
160 T.T.L. Vu et al. / Separation and Purification Technology 48 (2006) 159–166

[8]. A nonlinear least square method must be applied instead

Nomenclature of a single regression analysis in order to take into account
the effect of cooling temperatures.
C mass fraction (kg kg−1 ) In the following sections, the dynamic model of a cooling
F feed rate (kg h−1 ) crystalliser to produce ␣-lactose in a seeded batch is firstly
G growth rate (␮m h−1 ) defined and then modified for other operations: evaporative
M mutarotation rate (h−1 ) semi-batch and cooling in continuous mode. The method-
sol solubility (◦ Brix) ology of formulating and solving the dynamic optimisation
T temperature (◦ C) problems generated in the study is described. Cooling semi-
t time (h) batch operation is the focus of the work since it is a start-up
W total mass (kg) stage of both batch and continuous operations. Hence exper-
x mass (kg) iments are carried out in batch and semi-batch crystallisers
xd particle mean size (␮m) along with simulation results to validate the computer model,
ẋ rate change in mass (kg h−1 ) which can be used for industrial scale crystallisations. For
comparison, dynamic optimisations are also conducted for
Greek symbol an evaporative case study. Neither continuous nor evapo-
λ conversion ratio rative lactose crystallisation is commonly practiced in the
dairy industry or studied in the literature except some work
Subscripts
on continuous cooling done by Thurlby and Sitnai [9] and
c crystals of pure lactose
by Shi et al. [5]. Based on the results found for semi-batch
f feed
operation, experimental design and simulation in continuous
fin final
mode are also investigated here to compare with batch and
imp impurities
semi-batch.
lac lactose
The outcome anticipated from this study is a mathemati-
nv number of variables
cal model applicable to alternative processes, which produce
p product
uniform distribution of crystals in the highest yield.
w water
␣ dissolved ␣-lactose
␤ dissolved ␤-lactose
0 initial seed 2. Methodology

2.1. Process model

tory and pilot-plant experiments, preliminary results can be
obtained from simulations if mathematical models describing
the processes are well developed.
In the last quarter century many authors measured the
growth rate of ␣-lactose monohydrate in an isothermal seeded
batch or a continuous cooling crystalliser [2–5]. Recently
Butler [6] and Lifran et al. [7] have resumed lactose kinetic
studies. Based on experimental results at various tempera-
tures, a power law function was proposed to correlate the
growth rate of the particles with supersaturation. Previous
authors have presented the growth rate in different units but
used a similar form of power law function for lactose crystalli-
sation kinetics growth rate (G) = f(temperature) × (driving
force)n . In this format a nonlinear function k exp(−Eg /RT)
or a linear relationship k × T can be used to incorporate the
temperature. Values of the constants n, k and Eg were found
from a single regression analysis of the experimental data and
were reported differently due to their dependence on the ini-
tial weight, average mean size and particle size distribution
of the seed crystals used in the experiments. The growth rate
correlation, based on experimental results of an isothermal
static crystallisation cannot predict well the performance of
a dynamic cooling crystallisation; therefore, it must be re-
evaluated for different seeding weight and cooling strategies
Fig. 1 shows a schematic of a batch cooling process
implemented both in the laboratory and pilot plant, and a
bench-scale evaporative crystalliser. The dynamic model of
a batch cooling process is fully described by the mass and
population balances. In the case of ␣-lactose monohydrate
additional differential equations are required to incorporate
the mutarotation rate (M), calculating the dynamic transfor-
mation between ␣- and ␤-lactose in water. If a batch is initially
fed with seed crystals closely sized between two adjacent
sieve sizes, the dynamic of the particle size distribution can be
ignored. The state equations become the rate change (kg h−1 )
in mass of water (xw ) dissolved ␣-lactose (x␣ ), ␤-lactose (x␤ ),
crystals of pure lactose (xc ) and average size of lactose par-
ticles (xd ).
The state equations are also modified for the following
cases, in which the syrup feed is pumped into the vessel at
the rate F (kg h−1 ) but the differences are:
• cooling semi-batch: heat is removed;
• evaporative semi-batch: steam is supplied at the rate E
(kg h−1 ) to evaporate λE kilogram of water per hour
(1 > λ > 0);
• continuous flow: seed particles are pumped in with the feed
while product and heat are withdrawn.
 T.T.L. Vu et al. / Separation and Purification Technology 48 (2006) 159–166 161

Fig. 1. Schematics of batch cooling and evaporative crystallisers.

The state equations and their modification are shown sollac − K(0.0159 + 0.00023T 1.36 )
below. The subscript “f” denotes feed. Eqs. (1a) and (1b)
× (100((x␣ + x␤ )/xw ) − sollac )
define the rate change in mass of water in cooling and sol␣ = (9)
heating semi-batch, respectively. In these equations C with 1+K
the corresponding subscripts are the mass fractions of water, Under the conditions of no nucleation, agglomeration or
␣- and ␤-lactose in the feed. Some common symbols and breakage, in a batch or semi-batch operation, the number of
units are included in the nomenclature list.
Batch Semi-batch
dxw
Cooling ẋw = = −0.05R ẋw = −0.05R + FCwf (1a)
dt
dxw
Heating ẋw = = −0.05R ẋw = −0.05R + FCwf − λE (1b)
dt
Mx␤ Mx␤
ẋ␣ = −Mx␣ + − 0.95R ẋ␣ = −Mx␣ + − 0.95R + FC␣f (2)
K K
x␤ x␤
ẋ␤ = Mx␣ − M ẋ␤ = Mx␣ − M + FC␤f (3)
K K
3xd2 Gxc0
ẋc = R = (4)
xd30

x␣
n
ẋd = G = kT 100 − sol␣ (5)
xw

Eqs. (4) and (5) are applied in both batch and semi-batch
cases. In Eq. (4) xc0 and xd0 are mass and average size of crystals in the crystalliser remains constant. In a continuous
initial seed particles, respectively. Two constants k and n in run, it is assumed that the number of seed particles in the
the power law function (5) will be re-evaluated in this work to feed stream and the product stream are the same but their
compare with values found in the literature. The mutarotation average sizes and total masses are different. If this condition
rate M and K used in the state equations [6], solubility of is held, at steady state the rate changes in mass of water,
lactose and ␣-lactose [10] are correlated with temperature in dissolved ␣- and ␤-lactose approach zero in a continuous
the following equations. operation. The component balances, Eqs. (1a), (2) and (3)
  can be written in the same fashion as the following water
ximp balance.
M = 9.6 + 0.01 eT/12.4 (6)
xw
0 = −0.05R + F (Cwf − Cwp ) (10)
K = 1.64 − 0.0025T (7)
Since the feed stream is seeded an additional ordinary
sollac = e2.389+0.028T (8) differential equation describing the change in the number of
162 T.T.L. Vu et al. / Separation and Purification Technology 48 (2006) 159–166
seed crystals must be included. At steady state, the change
in mass of crystals and number of seeds in the vessel also
approach zero. This implies the numbers of seeds in the feed
and in the product stream are the same, but the mean size of
the feed seeds and their total mass are smaller. The difference
in solid content between the product (Ccp ) and feed (Ccf )
streams is shown in Eq. (11).
xd2 G
Ccp − Ccf = (11)
2F
2.2. Problem formulation and solving methods
2.2.1. Batch and semi-batch
2.2.1.1. Set-point profile. The aim of the first dynamic opti-
misation problem is to search for an optimal cooling tempera-
ture, or feed rate and evaporation rate profiles (in evaporative
mode) using values of n and k found in the literature or in
a single curve fitting [7]. The same technique is applied to
optimise performances of a cooling batch, cooling or evap-
orative semi-batch. To avoid solving the ordinary differen-
tial equations (ODEs) required for each recursion, Lagrange
interpolation polynomials are used to approximate the state
variables x (Eqs. (1)–(5)). These approximations can be dif-
ferentiated and back substituted into the ODEs, which are
converted to an algebraic set. The dynamic optimisation prob-
lem now becomes a normal nonlinear programming prob-
lem, which can be solved using any existing optimisation
technique.
The constraints commonly imposed on seeded batch
crystallisation include solubility and secondary nucleation
threshold (SNT) concentration. In particular, Eq. (8) above
and the following derived by Butler [6] are used for
lactose.
solsnt = e2.992+0.0196T (12)
Below the solubility, crystals dissolve instead of growing.
Beyond the SNT concentration, nucleation occurs, result-
ing in downstream processing inefficiencies. Working within
these limits justifies the assumption that the number of seed
crystals remains constant through an operation. Some addi-
tional constraints are required for an evaporative semi-batch
since simultaneous syrup feeding and water evaporation
affect the vessel inventory. The first additional constraint lim-
its the total mass (W) in a crystalliser within initial and final
values: Wi ≤ W ≤ Wfin . Secondly, the solid content (Cc ) in a
crystalliser has an upper limit, at which circulation will suffer
Cc = (xc /W) ≤ Ccfin .
The target of both heating and cooling problems is the
minimum operating time but the objective function set for
cooling mode can also be to maximise the growth at a spec-
ified time OFcool = max{xc (tfin )}. In contrast the objective
function of an evaporative semi-batch, referred to as a multi-
ple objective function, is much more complicated as shown in
Eq. (13).


i=n
OFevap = min wt t + ww [W(i) − Wfin ]2
i=1


i=n
+ wCc [Cc (i) − Ccfin ] 2
(13)
i=1
wt,w,Cc : weighing constants defined by users.
The solutions of the first optimal control problem are used
as set point profiles in the following dynamic optimisation
and experiments carried out in the laboratory and pilot plant
described in Section 2.3.
2.2.1.2. Nonlinear curve fitting. The aim of the second
dynamic optimisation problem is to re-evaluate the two con-
stants k and n for the cooling and heating strategies found
above. A nonlinear least square dynamic optimisation prob-
lem is formulated to fit the model equations to experimental
data and solved for the constants n and k in Eq. (5). The
same constraints described above are applied. The objective
function can be expressed as:
 

predicted
 j=fin
i=nv 2
OFfit = Min  xi (tj ) − ximeasured (tj ) 
i=1 j=0
(14)
In Eq. (14) xi represents the state variables, measurable on line
or off-line. In this work, solution concentration, solid content
and mean size of crystals are three measured variables.
2.2.2. Continuous operation
Simulations of start-up stages must be performed first
since the feed stream is seeded and the number of seed parti-
cles in the vessel increases with time. A model of a continuous
stirred tank reactor operating at a constant temperature is
applied to describe the performance of the crystalliser at
steady state.
2.3. Experiments
A 2 L stainless steel vessel, 4 L glass reactor and a 20 L
stainless steel jacketed crystalliser were set up in the labora-
tory and the pilot plant, in accord with the schematic shown
in Fig. 1. Lactose was crystallised from high purity syrup at
39 ◦ Brix, 60 ◦ C, generated in upstream units of the pilot plant.
Seeds having different median sizes taken from a jet mill,
under sieve 25 ␮m, between sieves 25–38 and 45–53 ␮m,
with two seeding ratios 1.0 and 1.5% (w/w) of crystal-free
solution were used in the experiments and simulations of
cooling or evaporative operations, batch, semi-batch or con-
tinuous mode. Cooling crystallisations were conducted in a 2
and 20 L stainless steel vessel, thermostated between 10 and
70 ◦ C and equipped with a stirrer. Different cooling and seed-
ing strategies were applied to evaluate the model. Sampling
was conducted every 30 min for the initial 3 h then hourly
 T.T.L. Vu et al. / Separation and Purification Technology 48 (2006) 159–166
for 5 h and for the last 3 h of each batch (25 h). The samples
were analysed for concentration (◦ Brix), solid content (mass
of lactose crystals/mass of content of the crystalliser), parti-
cle median size (␮m) and size distribution. For comparison
one evaporative run was conducted isothermally in a 4 L glass
vessel under vacuum, starting at the same initial conditions
applied for cooling. To mimic the real process the final prod-
ucts were filtered, washed and dried. The weights of dried
crystals and the filtrates were recorded to ensure a closed
mass balance and to calculate the yields.
3. Results and discussions
3.1. Batch cooling operation
3.1.1. Set point profiles
The cooling profiles, i.e. the solutions of the first dynamic
optimisation problem, are dependent on the mass and mean
size of initial seed crystals but not on the size of the crys-
talliser. A faster temperature drop is applied for finer mean
size and a larger weight of seeds. Fig. 2 shows the optimal
cooling profile (1a) of jet-milled seed particles. At the begin-
ning of a cooling batch, the temperature profiles are restricted
by the SNT concentration (Eq. (12)) then later dictated by the
mutarotation rate (Eq. (6)). For possible implementation the
step profile is approximated by three first order polynomi-
als, labelled as (1b) in the same figure. Curves 1, 2, 3 and
4 represent measured temperatures and corresponding solid
contents in the crystalliser when different cooling rates were
implemented. Cooling profiles 2 and 4 are the fastest and
Fig. 2. Cooling strategies and solid yields recorded in cooling batch (1: optimal, 2: fast, 3: natural and 4: slow strategies).
163
slowest temperature drops, which can be achieved manually
in the laboratory. For a short batch from 10 to 12 h, applying
the optimal strategy achieves the highest yield.
3.1.2. Nonlinear curve fitting
The optimal cooling trajectory was applied as a set point
profile in the experiments and in the nonlinear least square
dynamic optimisation problem. In this problem three mea-
sured variables: solid content, particles median size and
solution concentration were simultaneously fit to the model
equations, to re-evaluate two constants n and k used in the
growth rate Eq. (6). Experimental data were collected from
various scenarios: different batch sizes and seeding and cool-
ing strategies. Fig. 3 shows the best curve fitting of the initial
12 h, giving n = 2.7 for all the cases investigated here, com-
pared to the value found in the literature varied from 2 to 3.
The value of k varies with the mass and mean size of initial
seed crystals but is independent of the size of the crystalliser.
Implementing the optimal cooling strategy (curve 1) with jet-
milled seeds at 1.5% seeding ratio yields the average value,
k = 0.007 ± 0.003; meanwhile fast cooling (curve 2) gives a
higher value, slow cooling (curve 3) gives a lower k. To min-
imise error when scaling up the model to industrial batch
crystallisation, k must be re-estimated for selected seeding
and cooling strategies, and most importantly for the type of
syrup fed to the crystalliser, coming from the upstream stage.
This can be achieved inexpensively in the laboratory.
Particle size distributions of samples taken from a cooling
batch seeded with jet-milled crystals are plotted versus time
in Fig. 4 to study the spreading of initially narrow sized seed
population. A log–normal distribution curve is defined by the
164 T.T.L. Vu et al. / Separation and Purification Technology 48 (2006) 159–166

Fig. 3. Solid yields and particle median size in cooling batch (1: optimal, 2: fast and 3: slow strategies; data points represent experimental results).

following equation [11]. particle size distribution of the product after 3 h as shown in

Fig. 4.
CF (ln(d/xd ))2
Fd = √ exp −
σ 2π 2σ 2
In Eq. (15), Fd is the relative frequency at size d, σ is the
standard deviation of the distribution and CF is a dimension-
less constant related to the initial seed particles. If the seed
crystals possess a constant relative growth rate then σ and
CF remain constant regardless of the particle mean size. This
assumption does not hold for the first 3 h, while the shape,
standard deviation and skewness of the distribution change.
After 3 h, the size distribution shape remains constant; val-
ues of σ and CF reach 0.5 and 5, respectively. These values
and the median size can be used to predict the shape of the
Fig. 4. Particle size distributions of jet-milled seeds up to 11 h in semi-batch
cooling crystallisation (data points represent predicted results).
(15)
3.2. Alternative processes
In addition to cooling batch mode the model can be
extended and applied to semi-batch and continuous opera-
tions, which were ignored in the lactose crystallisation liter-
ature.
3.2.1. Semi-batch: cooling and heating
In industrial seeded cooling crystallisation, seed crystals
and lactose syrup are initially drawn into the crystalliser to
allow thorough mixing. Heat is then removed in accordance
with the cooling temperature profile while syrup is contin-
uously fed at the maximum rate up to the volume of the
vessel. This process, referred to as semi-batch mode, offers
a higher yield or shorter operating time than a batch perfor-
mance due to the following reasons. Semi-batch operation
starts with a smaller mass of solution but the same mass of
initial seed crystals (xc0 ), therefore the initial growth sur-
face area or the initial seeding ratio is much higher. As the
seeds grow, the supersaturation decreases. Continuous feed-
ing of fresh syrup (without seeds) maintains a high driving
force for crystallisation and constrains the temperature drops
within limits to avoid a number of practical complications
such as non-uniform mixing or fouling the cooling surface
with crystals.
Steps (3.1.1) and (3.1.2) are applied to the semi-batch
cooling case using the modified equations (from 1 to 3). The
same value of k and a cooling strategy similar to the batch
case can be used in semi-batch simulations and bench-scale
 T.T.L. Vu et al. / Separation and Purification Technology 48 (2006) 159–166
Fig. 5. Solid yields and particle median size in semi-batch cooling, heating and continuous cooling operations (1 and 2: semi-batch and 3: continuous mode;
data points represent predicted results).
runs. Experimental results and prediction on both solid yield
and particle size agree well for an 8-h period as shown in
curve (1), upper graphs of Fig. 5. After 8 h measured particle
size starts to deviate from prediction since it is more difficult
to avoid nucleation, breakage and agglomeration in a high
solid content solution.
Evaporative semi-batch crystallisation was conducted at
a constant temperature and pressure lower than atmospheric
pressure to reduce the boiling temperature of the lactose solu-
tion. Solving the dynamic optimisation of the evaporative
semi-batch case, in which syrup feed is pumped into the ves-
sel, while heat is supplied to evaporate water yields feed and
evaporation rate profiles. It was difficult to implement both
strategies in a bench scale experiment; therefore heat was
supplied at a fixed rate, while feed was pumped in to main-
tain a constant level in the vessel. The final products were
filtered, washed, dried and analysed to compare with predic-
tion. Curve (2) in Fig. 5 plot simulation results. Between 2
and 4 h, an evaporative run could achieve more than a cooling
batch, which operated for more than 20 h. In cooling, the solid
content cannot reach the limit of 35% while in heating the
solid can easily pass 50% due to the evaporation of water. That
means in the same size crystalliser, heating is more efficient
than cooling. However, operating and controlling an evapo-
rative batch while avoiding nucleation poses expensive and
difficult problems compared with a cooling batch, in which
temperatures are easily measured and controlled. In contrast,
evaporative crystallisation must be performed isothermally
under vacuum to reduce the boiling temperature. This imple-
165
mentation is more complicated and nucleation is difficult to
control because the gap between solubility and secondary
nucleation threshold concentration is very narrow (from 37 to
40 ◦ Brix, at 60 ◦ C). Modern techniques to measure solid con-
tent and concentration online as applied in the sugar industry
have not been extended to lactose manufacturing, therefore,
control of the lactose evaporative vacuum tank must rely on
a simple conductivity-based scheme. As a result, production
will be unreliable because the process cannot be exactly mon-
itored, predicted and controlled.
3.2.2. Continuous operation
Cooling in continuous mode is a promising alternative,
which includes two stages, to be investigated. During the
start-up stage, seeded syrup feed is pumped in while heat
is removed but no product is withdrawn. The first stage,
case (3) in Fig. 5 is similar to the cooling semi-batch case
except that seed particles are added in the feed stream. After
withdrawing the product, it takes about six or seven resi-
dence times for the solid content to reach steady state as
shown in the lower graphs of Fig. 5. The mean size, how-
ever is only stabilised after 100 h. This indicates that seed
particles are washed out of the tank during a long transition
period. Further more nucleation cannot be avoided if the feed
syrup at a high temperature (60 ◦ C) is directly introduced to
the tank kept at a low temperature (20 ◦ C). A continuous
stirred tank reactor operating at a constant temperature can-
not describe well the performance when nucleation occurs
in the process. A plug flow reactor installed before a contin-
166 T.T.L. Vu et al. / Separation and Purification Technology 48 (2006) 159–166
uous stirred tank reactor is being studied to overcome this
drawback.
4. Conclusions
In this study the dynamic model of a crystallisation pro-
cess producing ␣-lactose monohydrate was derived for these
performances: cooling batch, cooling and evaporative semi-
batch, and cooling in continuous mode. The methodology of
formulating and solving the dynamic problems to find the
optimal set point profiles and re-evaluate the growth rate cor-
relation was described. For many operations the value of n
equal to 2.7 can be applied, but k should be estimated for
selected seeding and cooling strategies, and most importantly
for the type of syrup fed to the crystalliser. Using the value of
k = 0.007 in simulations and experiments of jet-milled seeds
at 1.5% seeding ratio, experimental results agreed well with
predictions in a 10-h period for cooling batch and semi-batch.
In addition to lactose and sucrose, the developed methodol-
ogy can generally be applied to the crystallisation of many
other food substances such as salt, protein, glucose and tar-
taric acid, but is not limited to the food or dairy industries.
Implementing the optimal cooling strategy in semi-batch
mode achieves a higher yield than in batch operation. Yet
compared to heating, cooling is less efficient. Despite a low
production yield limited by lactose solubility, cooling batch
operation is still commonly practiced in lactose factories, as
the process requires low capital and maintenance costs; prod-
uct size and size distribution can be easily manipulated. In
contrast, these results suggest that evaporative batch opera-
tion is much faster but implementation is more complicated
and nucleation is difficult to control. Further investigation
will be focused on applying the control schemes developed
for sucrose to lactose evaporative crystallisation. Experimen-
tation and a more rigorous model including plug flow and
continuous stirred tank reactors are being studied to describe
the performance of a continuous cooling and seeding run with
minimal nucleation.
Acknowledgements
The authors would like to acknowledge Dairy Australia,
Food Science Australia and the Centre for Advanced Food
Research, University of Western Sydney for their support
of this project. Our special thanks are due to Professor E.T.
White and Dr. B. Butler, who gave us permission to use infor-
mation in reference [6].
References
[1] R.J. Durham, J.A. Hourigan, R.W. Sleigh, Australian Patent
84264/98 (2001), US Patent 09/463508 (2002).
[2] P. Jelen, S.T. Coulter, Effect of supersaturation and temperature
on the growth of lactose crystals, J. Food Sci. 38 (1973) 1182–
1185.
[3] J.A. Thurlby, Crystallisation kinetics of alpha lactose, J. Food Sci.
41 (1976) 38–42.
[4] R.C. Griffiths, G. Paramo, R.L. Merson, Preliminary investigation of
lactose crystallisation using the population balance technique, AIChE
Symp. Ser. 78 (1982) 118–128.
[5] Y. Shi, B. Liang, R.W. Hartel, Crystallisation kinetics of alpha-
lactose monohydrate in a continuous cooling crystalliser, J. Food
Sci. 55 (1990) 817–820.
[6] B. Butler, Modelling industrial lactose crystallisation, PhD Thesis,
The University of Queensland, St. Lucia, Australia, 1998.
[7] E.V. Lifran, T.T.L. Vu, R.J. Durham, J.A. Hourigan, R.W. Sleigh,
Crystallisation kinetics of lactose in the absence of lactose phos-
phate, in: Proceedings of the 32nd Australasian Chemical Engineer-
ing Conference on Sustainable Processes, Sydney, Australia, 26–29
September, 2004.
[8] T.T.L. Vu, R.J. Durham, J.A. Hourigan, R.W. Sleigh, Dynamic
modelling and optimisation of ␣-lactose monohydrate seeded batch
cooling crystallisation, in: Proceedings of the European Symposium
on Computer Aided Process Engineering-13, Lappeenranta, Finland,
1–4 June, Elsevier, Amsterdam, 2003, pp. 847–853.
[9] J.A. Thurlby, O. Sitnai, Lactose crystallisation: investigation of some
process alternatives, J. Food Sci. 41 (1976) 43–47.
[10] R.A. Visser, Supersaturation of ␣-lactose in aqueous solutions in
mutarotation equilibrium, Netherlands Milk Dairy J. 36 (1982)
89–101.
[11] T. Allen, Particle Size Measurement, vol. 1, Chapman & Hall, 1997
(Chapter 4).