Sunteți pe pagina 1din 13

Journal of Food Engineering 111 (2012) 642654

Contents lists available at SciVerse ScienceDirect

Journal of Food Engineering


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

Designing a lactose crystallization process based on dynamic metastable limit


Shin Yee Wong a, Rajesh K. Bund b, Robin K. Connelly a,b,1, Richard W. Hartel a,b,
a
b

Department of Biological Systems Engineering, University of Wisconsin, Madison, 460 Henry Mall, Madison, WI 53706, United States
Department of Food Science, University of Wisconsin, Madison, 1605 Linden Drive, Madison, WI 53706, United States

a r t i c l e

i n f o

Article history:
Received 22 January 2012
Received in revised form 27 February 2012
Accepted 1 March 2012
Available online 20 March 2012
Keywords:
Lactose
Cooling crystallization
Metastable zone
Computational uid dynamics
Secondary nucleation
Encrustation

a b s t r a c t
In the dairy industry, lactose crystallization during rening typically generates a large number of nes
(<100 lm), which greatly reduces the efciency of downstream processes. To overcome this problem,
a strategy to minimize nes production was developed. On lab scale units, lactose crystals were produced
from three crystallizers (draft-tube bafed, anchor and paddle) operated with three cooling proles (at
different region inside metastable zone (MSZ)). Computational uid dynamics was used to simulate
the ow prole. Among all combinations investigated, anchor crystallizer (lowest shear) operated at slow
cooling rate (in upper MSZ) produces the largest crystals with minimal nes. Then, the design strategy
was applied in industrial scale crystallizer. The 13 h cooling prole created for operation in the medium
MSZ region successfully produced crystals with 28% less nes than the typical process. Therefore,
depending on the crystallizer design and operational region (in MSZ), production of nes can be
minimized.
2012 Elsevier Ltd. All rights reserved.

1. Introduction
Lactose is produced from whey permeates obtained after the
production of cheese and/or whey proteins. The three main steps
in lactose production are concentration, crystallization and separation. The concentration process involves the evaporation of water
in whey permeate to increase lactose concentration. Concentrated
whey permeate has about 6570% total solids, with about 80% of
the total solids as lactose. The mixture is then cooled during the
crystallization process where the lactose is separated as a-lactose
monohydrate crystal. Crude/food grade lactose is obtained after
two stages of centrifugation, and nal drying. Pharmaceutical
grade lactose is produced by a subsequent rening process, which
includes a series of washing, evaporation, drying and centrifugation steps to remove trace impurities.
Among all processing steps, crystallization is the most important separation step, but the crystallization process in the dairy
industry is far from optimized. The lling of the tank takes about
6 h, followed by gradual cooling and crystallization that lasts for
1418 h, which means the crystallization process lasts for 20
24 h (Shi et al., 2006). The process is usually carried out in a large
Abbreviations: MSZ, metastable zone; ML, metastable limit; MSZW, metastable
zone width; CFD, computational uid dynamics.
Corresponding author at: 1605 Linden Drive, Madison, WI 53706, United States.
Tel.: +1 608 263 1965; fax: +1 608 262 6872.
E-mail address: rwhartel@wisc.edu (R.W. Hartel).
1
Current address: Solae LLC, 4300 Duncan Avenue, St. Louis, MO 63110, United
States.
0260-8774/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jfoodeng.2012.03.003

stirred tank cooling crystallizer. As the solution is gradually cooled,


the supersaturation increases and lactose crystals are formed. The
growth of crystals is typically accompanied by secondary nucleation, so the nal product has a lot of small crystals, as shown in
Fig. 1a. The resulting crystal size distribution makes the downstream processes (e.g., ltration, washing, drying, etc.) difcult.
This results in low quality product, large product loss and low
efciency.
To overcome these production problems, the crystallization
process could be greatly improved by operating at conditions that
promote growth and minimize secondary nucleation, leading to
the production of a narrow distribution of larger crystals. The ideal
operating conditions can be selected by examining the lactose
supersolubility diagram (Wong et al., 2011) as reproduced in
Fig. 1b.
In Fig. 1b, the metastable zone width (MSZW) is dened as the
region between the solubility and supersolubility lines. The region
between forced crystallization and supersolubility lines was reported by Hunziker (1926) as the optimum region for mass crystallization in condensed milk. Recently, the metastable zone width
(MSZW) was rened by Wong et al. (2011), with additional reference lines:
1) TD is the temperature at the detection of the rst nuclei (via
microscopy test) at various lactose solution concentrations.
Therefore, seeded crystallization operations in the upper
metastable limit (ML) dened by the region between solubility and TD lines have minimum secondary nucleation.

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

(a)

nucleation through impact (agitation) (Wong et al., 2011), shear


damage to crystals, and impact on agglomerate formation/breakup
(Tung et al., 2009). For energy transport, uid mixing affects the
rate of heat transfer through the jacket wall. In terms of mass
transport, proper blending of the solution to the molecular level
is required to achieve uniform supersaturation or to promote reactive crystallization (Tung et al., 2009).
The ow prole in a crystallizer can be analysed using computational uid dynamics (CFD). CFD is a numerical method that predicts parameters such as velocity, temperature, and pressure
(among others) by solving the associated governing equations
describing uid ow, which are the mass, momentum and energy
conservation equations (Bird et al., 2007). CFD is widely used to
help describe a variety of uid motion in mixing (Paul et al.,
2004). For lactose crystallization, Wood-Kaczmar (2006) used
CFD to reveal the poor circulation regions below the impeller in a
crystallizer.
When designing a lactose crystallization process, the operating
regions and uid mixing are the two most important parameters.
Therefore, the objective of this study was to examine the correlation between uid mixing and the operating conditions along different regions in the lactose supersolubility diagram (Fig. 1b).
Then, based on this understanding, an optimized solution suitable
for industrial operation was proposed.

(b) 80

Temperature (oC)

70

UNDERSATURATED
ZONE

60
50
LABILE
ZONE

40
30

Extrapolated T
TD
D
0.5% Tr

20

2. Materials and methods

1% Tr
Industrial operation

10
0.1

643

0.2

0.3

0.4

0.5

0.6

Lactose solution concentration (g g-1)


Fig. 1. (a) Polarized light microscope image of the lactose crystals obtained after an
industrial cooling crystallization process. (b) Lactose supersolubility diagram (TD
temperature at the detection of the rst nuclei; 0.5, 1% DTr temperatures at 0.5%
and 1% change in the transmittance relative to the initially seeded lactose solution;
Lactose solution concentration (g g1) = mass of anhydrous lactose/mass of
solution).

This study was split into four parts. The rst three parts were
lab scale studies (Sections 2 and 3), while the last part involved
industrial verication and plant trials (Section 4). For the lab scale
studies, lactose crystallization experiments with three distinct
cooling proles were rst conducted using three crystallizer
types/impellers with different ow proles. Then, a model capable
of predicting the concentration prole according to the cooling
prole and initial lactose solution concentration was developed. Finally, the uid ow proles were analysed via CFD simulations.
2.1. Crystallizer design

2) The other two reference lines, 0.5% and 1% DTr, indicate the
temperatures at different levels of transmittance change
(DTr) (via spectroscopy test) when lactose solution was
cooled at 1 C/min. Higher changes in transmittance (relative to the seeded lactose solution) suggests the presence
of higher numbers of nuclei in the solution and, thus, higher
extent of secondary nucleation. The upper ML, dened by
the region between solubility and 0.5% DTr or 1% DTr lines,
have low or medium levels of secondary nucleation,
respectively.
Typical operating conditions for an industrial lactose rening
operation (Shi et al., 2006) are also shown in Fig. 1b. The bulk of
the operation operates in the lower ML between forced crystallization and the supersolubility lines, which is a region prone to secondary nucleation under agitated conditions. Therefore, commercial
operating conditions result in the production of lactose crystals
with wide range of size and numerous small crystals (Shi et al.,
2006), as shown in Fig. 1a. In short, for a seeded crystallization process, the ideal operating region would be the upper metastable region, which is characterized by the area between the solubility
and the TD, 0.5% DTr or 1% DTr lines, depending on the tolerance
to the extent of secondary nucleation.
In addition, uid mixing during the crystallization process is
also important in all aspects of transport properties, including
momentum, energy and mass. A few areas of concern for momentum transport include the homogeneity of crystal slurry, secondary

In this study, three crystallizers with different types of impellers, shown in Fig. 2, were used. The rst design (Fig. 2a) was a
draft tube bafe (DTB) crystallizer. It consisted of a tall form
250 ml beaker, a polycarbonate draft tube bafe set, and a three
blade propeller attached to a stainless steel shaft. The crystallizer
was placed in a water bath (Haake DC30) and the temperature of
the crystallization process was controlled by the bath temperature.
The draft tube bafe system was constructed so that both mother
liquor and suspension were well mixed. This system allowed axial
ow (of solution) inside the crystallizer: ow was down inside the
draft tube and up in the annular region. However, a minimum rotational speed of 440 rpm was required to suspend the crystals,
resulting in signicantly higher shear rate (around the impeller region) compared to the other two designs.
The second and third designs had close-clearance impellers,
with diameters nearly the same size as crystallizer tank diameter
to provide macro-scale blending of liquids at low shear (Paul
et al., 2004). The clearance between the impeller and the crystallizer (bottom and side) wall was ca. 3 mm. Due to the close clearance, the rotational motion of the impeller sweeps the wall and
prevents build-up of encrusted crystals on the wall. In addition,
this also improves heat transfer through the jacket.
The anchor crystallizer (Fig. 2b) consisted of a tall form 250 ml
jacketed beaker and an anchor stirrer. The temperature of the crystallization process was controlled by the circulating water bath
(Thermo Electron Neslab RTE7) connected to the jacketed beaker.

644

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

speed
95 rpm

speed
440 rpm

speed
110 rpm

Baffle

Draft
tube

Propeller
Total Volume
= 206 ml

Total Volume
= 206 ml

(a) Draft tube baffle (DTB)

(b)Anchor

Total Volume
= 300 ml

(c) Paddle

Fig. 2. Schematic diagram of all crystallizers.

The paddle assembly conguration (Fig. 2c) was similar to the anchor stirrer, with additional two-blade propellers at the top and
bottom of the assembly. This crystallizer consisted of a 350 ml beaker and the paddle assembly. Similar to the DTB crystallizer, the
paddle crystallizer was placed in the same water bath (Haake
DC30). The working volumes of the three crystallizers were 206,
206 and 300 ml, respectively. The three crystallizers were selected
to represent different mixing proles and impeller designs, which
can be further examined through CFD simulations.

average rate of 0.031 C/min, which is a long cooling time for


the industry. The medium cooling prole (MEDIUMC) had similar
cooling time (14 h) as an industrial cooling crystallizer (Shi et al.,
2006) with an average cooling rate of 0.055 C/min. As the solution gradually cooled, the supersaturation increased and lactose
crystals were formed. The concentration prole of the crystallization process changes as a function of the cooling prole and the
crystallization rates. These three different cooling proles allowed
the crystallization operation to be investigated in different operational regions as outlined in Fig. 1b.

2.2. Cooling prole


2.3. Experimental setup
Three cooling proles, representative of fast, medium and slow
cooling, were selected. The lactose solution was cooled from a high
temperature of 76 C to 25 or 30 C in a given period of time, as
shown in Fig. 3. The fast cooling prole (FASTC) had a cooling time
of 2.2 h with an average cooling rate of -0.43 C/min. The slow
cooling prole (SLOWC) had a total cooling time of 25 h with an

75
70

Temperature (o C)

65
60
55
50
45
40
35
30
25
0

10

12

14

16

18

20

22

24

Time (hr)
Fig. 3. Three cooling proles investigated (SLOWC, MEDIUMC and FASTC are
cooling proles with cooling rates of 0.031, 0.055 and 0.43 C/min,
respectively).

Supersaturated lactose solution (58%) was prepared by dissolving


pharmaceutical grade lactose (310, Foremost Farms USA) in deionized water. The solution was heated to 8085 C to make sure all
the lactose crystals were dissolved. At the start of the experiment,
seed crystals were added to achieve the following objectives reported in previous studies: (1) to create a number of initial nuclei
that are in fast growing state, so that secondary nuclei do not survive
when there is a balance between the growth rate of existing crystals
and the rate of change from clusters to crystals (in the supersaturated
solution) (Shi et al., 2006); and (2) to improve the yield (Vu et al.,
2003). In industrial operation, multiple crystal nuclei are often already present in the concentrated whey permeate during lling
and prior to the onset of cooling. Therefore, the solution was seeded.
The seed lactose crystals were prepared by sieving pharmaceutical
grade lactose through sieve sizes of 2509045 lm; crystals collected on the sieve size of 4590 lm (Vu et al., 2003) were used.
The crystallizer was rst lled with a xed volume of supersaturated lactose solution. At the start of the experiment (t = 0 min),
the supersaturated solution (at T = 76.5 C) was seeded with
25 mg of lactose seed crystals (Shi et al., 2006). The temperature
of the crystallization process was controlled by either the circulating water bath (Thermo Electron Neslab RTE7) connected to the
jacketed beaker (Anchor crystallizer) or the water bath (Haake
DL30-V26/B) (DTB and Paddle crystallizer). The crystallizer was
cooled to 25 or 30 C via the three different cooling proles, as
shown in Fig. 3. The cooling prole was implemented by controlling the set point of the circulating water bath with a modied Labview program.

645

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

Lactose crystals were produced as cooling progressed. The concentration of the lactose solution was measured following every
10 C drop in temperature. For this measurement, a small volume
(1 ml) of lactose slurry was rst ltered through a Whatman
GD/X syringe lter. The refractive index of the ltrate was measured (refractometer, Bausch and Lomb) and converted into lactose
concentration using pre-calibrated standard curve. At the end of
the cooling cycle, the crystallizer was emptied and the lactose crystals were separated by vacuum ltration (Whatman No. 42 lter
paper, 2.5 lm). The crystals were washed with 2-propanol (nonsolvent) and dried overnight in a 60 C oven. The crystal size distribution (CSD) was measured by laser diffraction via Malvern MasterSizer 2000 (Wong et al., 2010). In addition, the crystal shape
characteristic was examined by microscope (Nikon Labophot-2)
with 40 magnication.
The encrustation amount (g) and yield (g) were also measured.
Encrustation is dened as the coating/scaling of lactose crystals on
the surface of the crystallizer assembly (wall, bafes, and draft
tube). To measure the level of encrustation, the emptied crystallizer (after vacuum ltration of the slurry) was dried in a 60 C
oven overnight and the weight of encrusted lactose crystals measured. The yield of the crystallization process was calculated by
two approaches. The recovery percentage (Eq. (1)) denotes the ratio of lactose crystals produced and the amount of lactose initially
present in the solution. This is the approach commonly used in the
lactose rening industry.

Recovery %

Experimental yield g
initial mass of lactose in solution @t 0g

V I  1  C I V F  1  C E 0:05  M crystal

The value of Mcrystal was determined by solving both Eqs. (3)


and (4) simultaneously. The theoretical yield (%) can be calculated
via Eq. (5):

Theoretical yield %

Experimental yield g
 100
Mcrystal M seed g

To investigate the crystallization operation in different regions


along the lactose supersolubility diagram, crystallization experiments were conducted for all three crystallizer designs (Fig. 2) and
three cooling proles (Fig. 3). There were a total of 3  3 experiments, with all experiments repeated at least twice.
2.4. Modeling the concentration prole
For a batch cooling crystallization process, the mass conservation equation for all species (water, dissolved lactose and a-lactose
monohydrate crystals) can be written as:
Water concentration (x1) (kg water/m3 solution),

dx1
dx3
0:05
dt
dt

Lactose concentration (x2) (kg lactose/m3 solution),

dx2
dx3
0:95
dt
dt

Suspension density of a-lactose monohydrate crystals (x3) (kg


lactose crystal/m3 solution),

 100
1
Theoretically, the maximum yield (Mcrystal) is limited by the saturation concentration of lactose (Cs) at the nal cooling temperature (TF). The lactose crystallization process is dened as:

dx3 qpNGx24

dt
2

Average diameter of crystals (x4) (m),


22
dx4
G 3:2  109 eRT273 S  12:4 =a1
dt

Number concentration of crystals of a-lactose monohydrate, N


(#/m3 solution).
0.95 Lactose 0.05 H2O
(Dissolved in
solution)
VI  CI
VI  (1  CI)
VF  CS
VF  (1  CS)

Lactose
H2O

Initial
Mseed
Final
Mcrystal
(Equilibrium
+ Mseed
@ TF)
Used
VI  CI  VF  CS VI  (1  CI)  VF  (1  CS)

Here, VI and VF are the initial and nal volumes (ml) of the lactose solution. CI and CS are the initial and saturation (equilibrium at
TF) concentrations (g lactose/ml solution) of the lactose solution. Cs
can be calculated from (Eq. (2)) (Visser, 1982). Mseed is the mass of
seed added initially (=25 mg). Mcrystal is the maximum yield (g) of
crystals if the system achieves equilibrium at TF.
Cg

12:230:3375T 0:001236T 2 0:00007257T 3 5:188107 T 4


10012:230:3375T 0:001236T 2 0:00007257T 3 5:188107 T 4
2

The overall mass balance of each species in the reaction can be


written as:
Mass balance of lactose,

V I  C I V F  C E 0:95  Mcrystal
Mass balance of water,

11:4
dN
B 1:2  1018 eRT273 S  11:5 x3 =a2
dt

10

These equations were revised from the study of Vu et al. (2003)


assuming (1) constant density (q = 1150 kg/m3); and (2) nucleation and growth of lactose crystals were the dominant mass transfer mechanisms (minimal aggregation or encrustation). Here, the
growth (G) and nucleation (Bo) rates models were obtained from
Liang et al. (1991) with the addition of model parameters, a1 and
a2. Different from the study of Vu et al. (2003), Eqs. (6)(10) included the effect of secondary nucleation by incorporating the
nucleation rate model (Eq. (10)) developed by Liang et al. (1991).
Crystals of a-lactose monohydrate were produced (Eq. (8)) from
95% of dissolved lactose (Eq. (7)) and 5% of water (Eq. (6)). G is the
growth rate (m/min), as described by (Eq. (9)). At the start of the
batch process (t = 0 min), N (Eq. (10)) is the total number concentration of particles (number/ml solution) added as seed crystals.
Due to nucleation, N, changed during the crystallization process.
The rate of evolution of the particles can be described as the nucleation rate (Bo) (#/ml-min) (Eq. (10)). It is known that crystallization kinetics change with different types of crystallizers and
operating conditions. Therefore, the factors (a1, a2) in both Eqs.
(9) and (10) were added to determine the kinetic expression for
this study. These factors (a1, a2) were estimated using the experimental concentration prole of FASTC for anchor crystallizers.
For this purpose, an iteration loop that scanned all combinations
of both parameters (a1, a2) in the range of (1300) (interval
size = 1) was setup in Matlab R2008a. The parameter combination

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

with the smallest sum of error between eight experimental and


modeled concentration was determined as a1 = 39, a2 = 258. After
that, the performance of the model (Eqs. (6)(10)) was also validated with experimental concentration for SLOWC and MEDIUMC.
As cooling progressed, the temperature (T) decreased, giving an
increase in the supersaturation (S). S of the a-lactose solution was
calculated by a model developed by Visser (1982), as shown in Eqs.
(11)(13).

C
C s  FK m C  C s

F exp



2374:6
4:5683
Tk

K m 0:002286T 2:6371

11

75

Modeled
Experimental (Anchor)
Experimental (DTB)
Experimental (Paddle)

65

Temperature (oC)

646

55

45

35

12
25
0.2

13

0.3

0.4

0.5

0.6

Lactose solution concentration (g g-1)

Given the cooling proles (Fig. 3), all equations were solved
simultaneously by the ODE solver in Matlab R2008a to predict solution concentration and yield as function of time and temperature.

(a) SLOWC (25 hr)


75

2.5. CFD simulation approach

Modeled
Experimental (Anchor)
Experimental (DTB)
Experimental (Paddle)

ND2 q

3. Results and discussion


3.1. Experimental concentration prole
The concentration proles measured experimentally for the crystallization operation with all three cooling proles (SLOWC, MEDIUMC and FASTC) are shown in Fig. 4. From Fig. 4a, for anchor and
paddle crystallizer, crystallization in SLOWC occurred in the region
between 0.5% and 1% DTr initially. When the temperature dropped
below 65 C, the operation transitioned into the region between TD
and 0.5% DTr, and moved closer to TD after 46 C. The experimental
concentration prole of MEDIUMC (Fig. 4b) is similar to the prole
of SLOWC. However, the operation transitioned to the region between TD and 0.5% DTr only after 55 C, which is 10 C lower than
the operation in SLOWC, and therefore might experience higher degree of secondary nucleation. When the cooling rate was very high
(FASTC, Fig. 4c), the crystallization process was operating in the labile zone throughout, with high degree of secondary nucleation.

45

25
0.2

14

Here, N is the impeller speed (rev/s) and D is the impeller diameter (m). For all crystallizers, the modied Reynolds numbers (Re)
were greater than 5000, which indicated that all ows were in the
turbulent regime. Thus, the Reynolds stress model (RSM) (Deen,
1998) was used to model the relationship between the turbulent
uxes and the smoothed eld variables.
The cooling prole for FASTC (Fig. 3) was used for the CFD simulation. The prole was implemented as temperature boundary
conditions to the side and bottom wall of the crystallizers. The time
step size for the unsteady simulation was 60 s, and the total number of time steps was 130, which was equivalent to a total simulation time of 130 min (2.17 h).

55

35

0.3

0.4

0.5

0.6

Lactose solution concentration (g g-1)

(b) MEDIUMC (14hr)

75

Modeled
Experimental (Anchor)
Experimental (DTB)
Experimental (Paddle)

65

Temperature (oC)

Re

Temperature (oC)

65

To better correlate the differences between uid ow proles


and the crystallization process output, CFD simulations were conducted. Three 3D, pressure-based, implicit, unsteady state simulations were setup in Fluent 6.3.26 for each crystallizer (Fig. 2).
In the CFD model, the time-smoothed continuity, momentum
and energy transport equations (Bird et al., 2007) were solved for
lactose solution (q = 1150 kg/m3, l = 2cP, Cp = 3470 J/kg-K,
k = 0.6 W/m-K). For mixing, the modied denition of Reynolds
number (Paul et al., 2004) was used:

55

45

35

25
0.2

0.3

0.4

0.5

0.6

lactose solution concentration (g g-1)

(c) FASTC (2hr)


Fig. 4. Experimental and calculated concentration prole for crystallization operation in (a) SLOWC (25 h); (b) MEDIUMC (14 h); (c) FASTC (2 h). (TD temperature
at the detection of the rst nuclei; 0.5%, 1% DTr temperatures at 0.5%, 1% change in
the transmittance; Modeled concentration prole estimated by Eqs. (6)(13)).

From Fig. 4, the concentration proles predicted from the crystallization model were close to the experimental proles at all
cooling rates for the anchor and paddle crystallizers even though
crystallization kinetics were predicted based on the FASTC cooling
prole. This suggests that the kinetic models used here adequately
t lactose crystallization kinetics over a broad range of conditions.

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

However, in Fig. 4a and b, the experimental concentration


proles of DTB crystallizer deviated from both the modeled and
experimental (anchor and paddle) concentration proles. This is
most likely due to the high extent of encrustation (Fig. 6) observed
only in the DTB crystallizer, which will be discussed in the next
section.
For crystallization operation with minimal encrustation, all
modeled concentration proles for all cooling conditions are summarized in Fig. 5. As discussed in Section 1, the extent of secondary
nucleation increases as the operation moves further away from the
TD line towards the supersolubility line. Therefore, to limit the generation of small nuclei (extend of secondary nucleation), it is desirable to operate closer to TD. However, the cooling time increases
signicantly as the operation moves in the reverse direction (from
supersolubility line towards TD). For example, there is an 11 h difference between the cooling times based on MEDIUMC and
SLOWC. The model (Eqs. (6)(13)) provided a quick estimate of
process output corresponding to process input (initial lactose concentration) and operational parameters (cooling prole). With the
availability of the model, the lactose rener can customize their
cooling prole by balancing the available resources (operational
time) and the desired operational efciency (product quality,
recovery, etc.). This will be demonstrated in the next section.

SLOWC

75

MEDIUMC
FASTC

Temperature (oC)

65

55

Increased
cooling
time

45
Increased
extent of
secondary
nucleation

35

25
0.2

0.3

0.4

0.5

0.6

Lactose solution concentration (g g-1)


Fig. 5. Calculated temperature-concentration proles of the three operating
conditions investigated (for crystallization processes with minimal encrustation).

Fig. 6. Encrustation observed experimentally for crystallization operation at


SLOWC (25 h).

647

3.2. Crystallizer outputs


The yield and encrustation amount for all experiments are summarized in Table 1. Among the three crystallizers (at all operating
conditions), the DTB crystallizer had the lowest yield and highest
amount of encrustation. That means most of the lactose was lost
as encrusted crystals (as high as 67% for operation in SLOWC) while
only a limited quantity were recovered as the nal product crystals. Fig. 6 shows a snapshot comparing encrustation level in all
three crystallizers after operating in SLOWC. For the DTB crystallizer, severe encrustation was observed on the bafes, draft tube
assembly, and a portion of the crystallizer wall. However, the anchor and paddle crystallizers were almost free from encrustation
(Fig. 6), and the amounts of encrustation were less than 5% of the
total amount observed in DTB crystallizer (Table 1). Clearly, to obtain optimum yield, the DTB crystallizer design is not suitable for
lactose rening.
For both anchor and paddle crystallizers, where encrustation
levels were minimal, operation in SLOWC gave the highest yield
followed by MEDIUMC and FASTC. From Fig. 5, the horizontal difference between the nal concentration of operation in FASTC
and the solubility line (the residual supersaturation at the end of
the cooling cycle) is greater than the same distance for operation
in SLOWC and MEDIUMC; therefore, the lower yields observed
for FASTC are expected.
The crystal size distributions (CSD) of the lactose crystals obtained from the three crystallizers at end of the three operating
conditions are shown in Figs. 79. The corresponding microscopic
photos of all product crystals are shown in Fig. 10.
In Fig. 10, a distinct crystal shape characteristic can be observed for all crystals obtained from experiments operating in
FASTC. The lactose crystals primarily were found as large aggregates, accompanied by the presence of numerous small crystals.
The DTB crystallizer had the largest number of small crystals,
with 55.9% (of all crystals) smaller than 100 lm, as shown in
Fig. 7. In Fig. 9, although a higher proportion of larger crystals
were obtained for FASTC operation using the paddle crystallizer,
the aggregated crystals (Fig. 10) were not favorable as a nal
product (Shi et al., 2006). So, operation in this region should be
avoided.
In general, among all operating conditions, the crystals produced from operation in SLOWC had the highest average size (Figs.
79). However, there were no distinct differences between the CSD
proles for operation in SLOWC and MEDIUMC in all crystallizers.
Most of the crystals produced in SLOWC and MEDIUMC were
shaped as segregated tomahawk crystals with minimal aggregation
(Fig. 10). Among all 6 experiments (rst two columns in Fig. 10),
operation in SLOWC & MEDIUMC of DTB crystallizer produced
the largest lactose crystals, with ca. 3% of nes (<100 lm), whereas
the anchor and paddle crystallizers yielded ca. 5.4% and 18.1%
nes, respectively.
From Figs. 79, it is evident that the extent of secondary nucleation increased as the operating condition moved from the metastable regions dened by TD (Fig. 1b), from SLOWC ? MEDIUMC
? FASTC. In addition, if the rate of secondary nucleation was not
controlled/limited, the presence of high number of nuclei centers
promoted the formation of aggregates, as shown in column 3 of
Fig. 10. From these results, the production of nes can be minimized by operating within the upper ML dened between the
0.5% DTr and solubility lines.
Based on the experimental data, operation in SLOWC using
the anchor crystallizer produced the most optimized combination
with a reasonable level of nes (4.4%) and high yield (90%
Theoretical yield). The use of CFD will help clarify the ow
proles to better document the differences between crystallizer
designs.

648

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

Table 1
Experimental yield and encrustation amount from all crystallizers at different operating conditions.
Operating conditions

Recovery (%) (Eq. (1))

Theoretical yield (%) (Eq. (5))

Encrustation amt (g)

DTB crystallizer

SLOWC
MEDIUMC
FASTC (Tf = 27 C)

0.18
0.43
1.61

0.20
0.49
1.82

80.31
76.20
48.69

Anchor crystallizer

SLOWC
MEDIUMC
FASTC (Tf = 25 C)

78.80
74.29
54.33

90.34
85.17
60.68

1.50
2.51
2.42

Paddle crystallizer

SLOWC
MEDIUMC
FASTC (Tf = 27 C)

82.66
82.04
39.33

94.77
94.06
44.38

1.95
1.79
3.93

Fig. 7. Crystal size distribution of crystals obtained from the draft tube with bafes
(DTB) crystallizer at three operating conditions.

Fig. 8. Crystal size distribution of crystals obtained from the anchor crystallizer at
three operating conditions.

3.3. CFD ow prole analysis


The 3D ow patterns inside the crystallizers can be visualized
using path lines. Path lines follow the trajectories that would be

Fig. 9. Crystal size distribution of crystals obtained from the paddle crystallizer at
three operating conditions.

imposed by mass-less particles seeded at the plane y = 0 inside


the crystallizer domain. The path line plots of all crystallizers (colored by velocity (m/s)) are shown in Fig. 11. These path lines were
generated with 200 steps of (size) 0.005 m, which means each particle had advanced through a distance of 1 m inside the crystallizer.
From Fig. 11a, the bafes in the DTB crystallizer converted the
swirling motion (of the propeller) into top-down or axial uid motion that helped to lift and suspend the crystals (Paul et al., 2004).
It had upward z-directional ow in the annular region between the
draft tube and crystallizer wall, and downward z-directional ow
inside the draft tube (Fig. 12a). This arrangement allows the optimal mixing in the axial (z) direction.
In contrast, a radial ow pattern dominated in both anchor and
paddle crystallizers (Fig. 11b and c). As shown in Fig. 12b, the anchor impeller provided minimal axial ow and there was no distinct region of dominant upward (positive) or downward
(negative) axial movement. Therefore, the larger particles had a
tendency to settle at the bottom of the crystallizer; however, bottom encrustation was prevented by the close clearance between
the impeller and the crystallizer wall. The presence of the top
and bottom center propellers in the paddle impeller assembly
helped to enhance axial mixing. Similar to the axial ow pattern
inside the DTB crystallizer, particles moved upward in the region
between the paddle impeller and crystallizer wall and downward
inside the impeller in the paddle crystallizer (Fig. 12c). However,
the extent of axial mixing in the paddle crystallizer was less than
that of the DTB crystallizer.

649

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

SLOWC (25 hr)

MEDIUMC (14 hr)

FASTC (2.2 hr)

Draft tube
baffle
(DTB)
Crystallizer

Anchor
Crystallizer

Paddle
Crystallizer

Fig. 10. Microscopic images of the lactose crystals obtained from different crystallizers under different operating conditions.

Fig. 11. Path lines plots (colored by velocity (m/s)) of mass-less particles released from plane y = 0 for all crystallizers (Step size = 0.005 m, number of steps = 200, equivalent
to a tracked distance of 1 m). (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)

During the crystallization process, the driving force (supersaturation) required for the production of crystals is created by cooling.
The mixing proles will have an impact on the temperature and,
thus, the local supersaturation prole inside the crystallizers (Tung
et al., 2009). The simulated temperature proles at 1 h into the
cooling prole for FASTC operation are shown in Fig. 13. Due to
the presence of axial ow prole, the temperatures were mostly
uniform in DTB and paddle crystallizers. The highest average temperature was found inside the anchor crystallizer, where the heat
transfer from the center of the crystallizer to the wall is the slowest. Note that, the maximum temperature difference between the
wall and the center was less than 0.4 C for any of the crystallizer
congurations.

In the DTB crystallizer, the presence of low velocity (<0.01 m/s)


zones (Fig. 14) might be a signicant factor that leads to the high
encrustation level. The low velocity zones overlap with the locations of the encrusted crystals, as shown in Fig. 6. The encrusted
crystals were mostly present in the annular region between the
draft tube and the crystallizer wall. Inside the annular region, the
ow was mostly slow and unidirectional (vertical) (Fig. 11a), with
no turbulent eddies to lift any particles settled at the side walls/
bafes. Therefore, once in the low velocity zones, particles settle
and accumulate (at the same location) over time, leading to higher
level of encrustation as operational time increased (Table 1).
The inuence of shear on secondary nucleation has been reported in the literature (Koscher and Fulchiron, 2002; Paul et al.,

650

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

Fig. 12. Axial velocity (m/s) prole at the center plane of all crystallizers.

Fig. 13. Temperature distribution (1 h into the cooling cycle with FASTC operation) at the cross sectional plane (x = 0 or y = 0) of all crystallizers.

2004; Soos et al., 2008; Tung et al., 2009; Wong et al., 2010). In
general, the rate of secondary nucleation, which in turn determines
the number of nuclei formed, increases following shear treatment.
In a stirred tank, the high shear rates are generated in the immediate vicinity of the impeller (Paul et al., 2004). Fig. 15 shows the
contours of wall shear stress on the impeller assembly for each
crystallizer. The wall shear stress is dened as the force acting
tangential to the surface due to friction. It is proportional to the
velocity (gradient) parallel to the wall and, thus, the highest shear
stresses are found at the tips of the impellers. Among all crystallizers, the highest wall shear stress was observed in the propeller of
DTB crystallizer (Fig. 15a), since it had the highest rotational speed.
The lowest shear stress was found in the anchor crystallizer
(Fig. 15b). Although the paddle impeller had an intermediate level
of shear stress, it had the highest surface area of high shear stress
zones (non-blue region) (Fig. 15c). That means there will be higher
probability of (high shear) collisions between the lactose slurry
and the paddle impeller assembly.

The overall mixing process is described by the combination of


shear rate and the volume (Paul et al., 2004). The volume-averaged
shear rate of all crystallizers is summarized in Table 2. In Fig. 15,
although the propeller in DTB crystallizer had the highest wall
shear stress, the volume of this region is relatively small and,
therefore, any given uid volume will experience the high shear
for a small amount of time. This results in intermediate volumeaveraged shear rate. In contrast, the paddle assembly has nearly
the same size as the crystallizer; thus, it had the highest volumeaveraged shear rate.
It is logical that the magnitude of the volume-averaged shear
rate may be correlated to the amount of nes generated in each
crystallizer. In Table 2, the volume-averaged shear rate in the anchor crystallizer was the smallest, resulting in the production of less
than 6% nes (<100 lm). The largest number (24%) of nes was produced by the paddle crystallizer with the highest volume-averaged
shear rate. The impact of shear on CSD is twofold. First, it increases
the rate of secondary nucleation. Also, when large additional num-

651

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

Fig. 14. Velocity (m/s) contour plot in the lower velocity range (00.01 m/s) of the DTB crystallizer (All colored zones are prone to encrustation).

Fig. 15. Wall shear stress (Pa) on the impeller assembly of all crystallizers.

Table 2
Flow properties of all crystallizers (The cumulative volume fraction (%) of nes was calculated as the sum of the volume fraction (%) of all crystals of size less than 100 lm).
DTB crystallizer

Anchor crystallizer

Paddle crystallizer

Rotational speed (rpm)


Tip speed (m/s)
Volume-averaged shear rate (1/s)

440
0.58
15.71

95
0.25
4.83

110
0.38
18.77

Cumulative volume fraction (%) of nes (<100 lm)


for operation in
SLOWC
MEDIUMC
FASTC

3.53
2.31
50.39

4.41
6.31
21.07

17.80
18.36
18.59

652

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

Fig. 16. CFD analysis on Industrial scale crystallizer (a) Path lines plot of mass-less particles released from plane y = 0; (b) Wall shear stress (Pa) on the impeller assembly of
the crystallizers (c_ is the volume average shear rate).

Fig. 17. Proposed and existing industrial cooling and concentration proles (The proposed (13 h) prole is proposed by establishing a cushion of 1 C higher than the
concentration prole of 0.5% DTr).

bers of crystal units are introduced, the available supersaturation is


divided equally among crystals, which results in less growth for
individual crystals (Hartel, 2001). Despite the large volume-averaged shear rate, only 4.4% nes were produced in DTB crystallizer.
This is most likely due to the high level of encrustation, since the
secondary nuclei that were generated might be entrapped as encrusted crystals, resulting in lower recovery of smaller crystals.
Therefore, for DTB crystallizer, no conclusive remark on shear rate
and the extent of secondary nucleation can be made.

4. Industrial validations
From the lab scale (200300 ml) study, a good understanding
of the crystallizer designs, and the appropriate regions of operation
were established with model lactose in water solution systems.
However, in the dairy industry, lactose is produced in an industrial
scale crystallizer (2.6  107 ml) from whey permeate that
contains other components (e.g., residual whey protein, minerals,
riboavin, etc.). To verify the applicability of this study in industrial

653

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

(b)
(a)

Typical process
13 hr plant trial

Fig. 18. (a) Polarized light microscope images; (b) Cumulative crystal size distribution (CSD) of the lactose crystals obtained from 13 h industrial trial (The typical process is
representative of the current industrial operation with 14 h cooling prole).

operation, the same design strategy was applied to an industrial


scale crystallization process. The objective of this part of the work
was to apply the design strategy directly onto existing industrial
crystallizer without the need of high capital investment. Therefore,
a soft modication, which would only require modication in
processing parameters, was developed. As shown in the lab scale
study, the two most important criteria in optimizing/designing a
lactose crystallization process are (1) crystallizer design, and
(2) concentration prole (along MSZW). The soft processing
parameters corresponding to these two factors are the impellers
agitation rate and cooling prole, respectively.
4.1. Industrial scale crystallizer (7000 gal)
An industrial scale lactose crystallizer with a height of 5.5 m
and an internal diameter of 2.6 m was used. Similar to the anchor
crystallizer (Fig. 2b), the industrial scale crystallizer had a close
clearance impeller that swept the wall and prevented build-up of
encrusted crystals on the wall. Currently, the typical lactose crystallization process at this facility takes 12 h for lling and 14 h
for cooling. To examine the design and to determine the appropriate agitation rate for the industrial crystallizer, a computational
uid dynamics (CFD) simulation was created.
4.2. CFD analysis
A CFD analysis was conducted based on the agitation rate
(14.5 rpm) of the industrial crystallizer during cooling. The path
line plots of the industrial crystallizer are shown in Fig. 16a. Compared to the anchor crystallizer, the impeller in the industrial
crystallizer allows better axial mixing. The tip speed of the industrial crystallizer (1.84 m/s) is 7.4 times higher than that of the
lab-scale anchor crystallizer (0.253 m/s). However, even though
the tip speed of industrial crystallizer is signicantly higher, the
volume-averaged shear rate inside the industrial crystallizer is
only approximately 10% of the same shear rate of anchor
crystallizer, as shown in Fig. 16b. Based on the low volumeaveraged shear rate, agitation in the industrial crystallizer is satisfactory and should not by itself cause excessive secondary
nucleation.

4.3. Plant trial


Given the low volume-averaged shear rate estimated from the
CFD analysis, it is possible to design a crystallization operation
close to the 0.5% DTr line, as shown in Fig. 1b. However, it may
not be ideal to operate directly on the 0.5% DTr reference line.
Thus, a cushion of +12 C was included to provide additional protection against secondary nucleation. The cooling prole required
to achieve the desired concentration prole was calculated using
Eqs. (6)(13) via a Matlab R2008a routine. In this routine, the temperature was rst divided into 5 C grids. Then, the holding time
required at each temperature interval (to achieve a minimum difference between calculated and desired prole) was estimated. A
13 h cooling prole based on a concentrationtemperature prole
that avoided the secondary nucleation zone was estimated and applied to the industrial scale crystallization process.
The concentration and cooling prole of the 13 h plant trial is
shown in Fig. 17. A microscopic image of the lactose crystals obtained from the plant trial is shown in Fig. 18a. Compared to the lactose crystals obtained in a typical industrial process (Fig. 1a), the
crystals obtained from the 13 h plant trial are larger, with 28% less
nes than the typical process, as shown in Fig. 18b. The 13 h plant
trial sample had an L50 of 360 lm, compared to 250 lm obtained
from a typical industrial process. Thus, given a good crystallizer design, large crystals can be obtained from a commercial crystallizer
simply by using an appropriate cooling and agitation prole.
5. Conclusions
The results from this study showed that the characteristics of the
crystal products can be controlled by careful selection of operation
region in the lactose supersolubility diagram and crystallizer design. Among all the conditions tested, the operation in the anchor
crystallizer along the upper metastable limit (ML), closest to the rst
detection temperature (TD) produces the least number of nes
(<100 lm). In addition, the practicability of the approach established
using the lab scale crystallizer was tested successfully with an industrial scale crystallizer in whey permeate concentrate system.
To summarize, the two key factors in designing a cooling lactose
crystallization process are the cooling prole and the crystallizer

654

S.Y. Wong et al. / Journal of Food Engineering 111 (2012) 642654

design. To minimize secondary nucleation or the production of


nes, the cooling prole should be selected such that the concentration prole of the crystallization process lies within the upper
metastable limit as depicted in Fig. 1b. Once the cooling prole is
selected, the crystallization process should be conducted in a crystallizer with optimal mixing prole, heat transfer efciency (to
avoid encrustation) and minimal shear.
Acknowledgement
This project was supported by National Research Initiative
Grant #2007-55503-18448 from the USDA Cooperative State Research Education and Extension Service.
References
Bird, R.B., Stewart, W.E., Lightfoot, E.N., 2007. Transport phenomena, 2nd edn. J.
Wiley, New York.
Deen, W.M., 1998. Analysis of transport phenomena. Oxford University Press, New
York, pp. 597.
Hartel, R.W., 2001. Crystallization in foods. Aspen Publishers, Inc., Gaithersburg,
Maryland, pp. 192232.
Hunziker, O.F. 1926. Condensed Milk and Milk Powder, fourth ed. In: Otto F.
Hunziker, La Grange, Illinois.

Koscher, E., Fulchiron, R., 2002. Inuence of shear on polypropylene crystallization:


morphology development and kinetics. Polymer 43 (25), 69316942.
Liang, B., Shi, Y., Hartel, R.W., 1991. Growth-rate dispersion effects on lactose crystal
size distributions from a continuous cooling crystallizer. Journal of Food Science
56 (3), 848854.
Paul, E.L., Atiemo-Obeng, V.A., Kresta, S.M., 2004. Handbook of industrial mixing:
science and practice. Wiley-Interscience, Hoboken, NJ, pp. 1377.
Shi, Y., Liang, B., & Hartel, R.W. (2006). Crystal rening technologies by controlled
crystallization. (US 2006/0128953 A1).
Soos, M., Moussa, A.S., Ehrl, L., Sefcik, J., Wu, H., Morbidelli, M., 2008. Effect of
shear rate on aggregate size and morphology investigated under turbulent
conditions in stirred tank. Journal of colloid and interface science 319 (2),
577589.
Tung, H., Paul, E.L., Midler, M., McCauley, J.A., 2009. Crystallization of organic
compounds: an industrial perspective. Wiley InterScience, Hoboken, NJ.
Visser, R.A., 1982. Supersaturation of alpha-lactose in aqueous-solutions in
mutarotation equilibrium. Netherlands Milk and Dairy Journal 36 (2), 89101.
Vu, T.T.L., Hourigan, J.A., Sleigh, R.W., Ang, M.H., Tade, M.O., 2003. Metastable
control of cooling crystallization. European Symposium on Computer Aided
Process Engineering 13, 527532.
Wong, S.Y., Bund, R.K., Connelly, R.K., Hartel, R.W., 2011. Determination of the
dynamic metastable limit for a-lactose monohydrate crystallization.
International Dairy Journal 21 (11), 839847.
Wong, S.Y., Bund, R.K., Connelly, R.K., Hartel, R.W., 2010. Modeling the
crystallization kinetics of lactose via articial neural network. Crystal Growth
& Design 10 (6), 26202628.
Wood-Kaczmar, M. (2006). Process for crystallizing lactose particles for use in
pharmaceutical formulations. (WO 2006/086130 A2).

S-ar putea să vă placă și