Subscriber access provided by LOUGHBOROUGH | UNIVERSITY LIBRARY
Crystal Growth and Design is published by the American Chemical Society. 1155
Sixteenth Street N.W., Washington, DC 20036
Article
The Impact of Direct Nucleation Control on Crystal Size
Distribution in Pharmaceutical Crystallization Processes
Mohd R. Abu Bakar, Zoltan K. Nagy, Ali N. Saleemi, and Chris D. Rielly
Cryst. Growth Des., 2009, 9 (3), 1378-1384• DOI: 10.1021/cg800595v • Publication Date (Web): 15 January 2009
Downloaded from http://pubs.acs.org on March 4, 2009
More About This Article
Additional resources and features associated with this article are available within the HTML version:
• Supporting Information
• Access to high resolution figures
• Links to articles and content related to this article
• Copyright permission to reproduce figures and/or text from this article
The Impact of Direct Nucleation Control on Crystal Size
Distribution in Pharmaceutical Crystallization Processes
Mohd R. Abu Bakar, Zoltan K. Nagy,* Ali N. Saleemi, and Chris D. Rielly
Department of Chemical Engineering, Loughborough UniVersity, Loughborough, Leicestershire,
LE11 3TU, United Kingdom
ReceiVed June 9, 2008; ReVised Manuscript ReceiVed December 5, 2008
ABSTRACT: The control of crystal size distribution (CSD) in pharmaceutical crystallization is of primary importance, as downstream
processes such as ﬁltration or drying are greatly affected by the properties of the CSD. It is recognized that the variability in the
ﬁnal CSD is mainly caused by the signiﬁcant uncertainties in the nucleation rates, and therefore, a good control of nucleation events
is necessary to achieve the desired CSD. In this paper, a new direct nucleation control (DNC) approach is introduced that directly
controls the apparent onset of nucleation deﬁned as the formation of new particles with detectable size using in situ instruments.
The approach uses information on nucleation and dissolution, provided by focused beam reﬂectance measurement (FBRM), in a
feedback control strategy that adapts the process variables, so that the desired quality of product is achieved, for example large
crystals with a narrow CSD. In addition, DNC provides in situ ﬁnes removal through the operating protocol, rather than having
additional equipment and external recycle loops. DNC does not require concentration measurement and has the advantage of being
a model-free approach, requiring no information on nucleation or growth kinetics in order to design an operating curve. The DNC
approach automatically and adaptively detects the boundary of the operating zone; hence it is more robust to the presence of impurities
or residual solvent than the supersaturation control approach. The approach has been applied for the crystallization of glycine and
experimental results demonstrate the beneﬁts of DNC of producing larger crystals with narrower CSD compared to classical operations.
1. Introduction
Crystallization is an important unit operation used in a variety
of industries. It is estimated that more than 80% of pharmaceutical
products involve at least one crystallization step in their
manufacturing process.1
The crystallization operation is often
critical because it determines product properties such as the
crystal size distribution (CSD), habit and polymorphic form.
The CSD is of primary importance since it inﬂuences subsequent
downstream operations such as ﬁltration and drying, and the
product performance such as dissolution and bioavailability.
Along with the United States Food and Drug Administration’s
(FDA) Process Analytical Technology (PAT) initiative, the
development of control approaches, which can improve the
manufacturing of products with desired properties, has become
of signiﬁcant interest.2-4
The majority of control approaches applied in crystallization
affect the CSD indirectly by implementing a temperature or
antisolvent proﬁle in time to follow a given supersaturation
proﬁle in the phase diagram.5,6
These proﬁles are obtained either
using simple trial-and-error experimentation or more complex
model-based6,7
or direct-design approaches.8-13
However, it is
recognized that the variability in the ﬁnal CSD is mainly caused
by the signiﬁcant uncertainties in the nucleation rate, and all of
the above-mentioned approaches in essence try to identify
operating protocols, which provide an acceptable compromise
between crystal growth and nucleation.3,14
Lasentec focused beam reﬂectance measurement (FBRM) is
one of the in situ instruments that has been used extensively in
the crystallization processes to provide qualitative as well as
quantitative information about nucleation and crystal growth.15-21
Although FBRM cannot give direct information about the
formation of critical clusters, the evolution of the counts of
the newly formed particles (after nucleation and growth to the
detectable size) and the chord length distribution provide
valuable information, which can be indirectly related to the
nucleation, growth, agglomeration or attrition phenomena during
the crystallization process.21
Despite the widespread use of
FBRM for monitoring crystallization processes, up until now
only a few attempts have been reported that use online FBRM
information to augment existing crystallization operations18,22-26
with almost all cases of FBRM being used in conjunction with
supersaturation control approaches. In this paper a novel
approach will be introduced that directly controls the onset of
nucleation in the case of (i) antisolvent and (ii) combined cooling
and antisolvent crystallization processes. A desired number of
FBRM counts, which indicate the number density of crystals,
is maintained during the entire duration of crystallization using
a feedback control strategy that adapts the antisolvent/solvent
addition, or cooling/heating rate, to directly control the nucleation
or dissolution rates. The experimental results from the
implementation of the DNC approach for the crystallization of
glycine are reported below.
2. Direct Nucleation Control Approach
Direct nucleation control (DNC) is a model-free approach in
which the number of counts measured by the FBRM is directly
controlled using a feedback control strategy. By deﬁnition, the
total number of counts/s (#/s) measured by FBRM is the sum
of the number of chord length measurements detected in all
the channels (1 through 38 in the FBRM model used) that
represents a chord length range from 0.8 to 1000 µm, and is
approximately proportional to the nucleation rate, which is the
predominant mechanism in changing the number of particles.
Figure 1 shows the schematic block diagram of the DNC
approach for antisolvent crystallization systems. A similar
structure can be used for cooling crystallization.
A typical operating proﬁle during DNC, represented in the
phase diagram, is illustrated in Figure 2. The DNC approach
automatically switches between heating and cooling, or solvent
* Corresponding author. Phone: + 44 (0)1509 222 516. Fax: + 44 (0)1509
223 953. E-mail: Z.K.Nagy@lboro.ac.uk.
CRYSTAL
GROWTH
& DESIGN
2009
VOL. 9, NO. 3
1378–1384
10.1021/cg800595v CCC: $40.75  2009 American Chemical Society
Published on Web 01/15/2009
and antisolvent addition, or a combination of the two, to generate
nucleation, or ﬁnes dissolution, to maintain the desired number
of counts/s. If the number of counts/s exceeds the desired value,
the excess particles are dissolved by solvent addition, or an
increase in temperature, which drives the process into the
undersaturated zone until the number of counts/s decreases. The
proposed DNC approach provides in situ ﬁnes removal through
the operating protocol, rather than having additional equipment
and external recycle loops. The method extends the ﬁnes
removal strategy proposed by Lewiner et al.27
in which the
online measurement signal is not only used for the monitoring
purposes but also as an input for a continuous feedback control
methodology.
An additional novel characteristic of the DNC is that it does
not use predetermined heating or cooling polices. The heating/
cooling rates (or solvent/antisolvent addition rates) are automatically
determined during the entire duration of the crystallization
in correlation with the extent of nucleation detected. The DNC
approach has the beneﬁt of automatically driving the crystallization
process toward the optimal operating curve by automatically
and adaptively detecting the metastable zone limits. It
provides a very robust methodology for crystallization scaleup,
through FBRM feedback control, which requires virtually
no a priori information about the model, kinetics or metastable
zone width of the process. Since the approach tends to operate
at high supersaturation, the problems related to solvent-inclusion,
impurities and agglomeration may be facilitated. Usually in these
Figure 1. A block diagram of the DNC approach. w(t) is the actual
antisolvent/solvent ﬂowrate into the crystallizer; wset(t) is the antisolvent/
solvent ﬂowrate set-point given by the DNC to the pump.
Figure 2. Typical DNC operating proﬁle.
Figure 3. A schematic representation of the experimental setup for
DNC experiments.
Figure 4. Solubility data and trajectory of the solution concentration
due to the dilution as a result of the addition of antisolvent.
Figure 5. Proﬁles of (a) number of counts/s and (b) SWCD for the
uncontrolled experiments with antisolvent addition rates of 8 g/min
and 10 g/min. The dashed lines are the target counts/s.
Direct Nucleation Control of Crystallization Crystal Growth & Design, Vol. 9, No. 3, 2009 1379
cases it is recommended to operate at lower supersaturation.
However, the repeated and controlled dissolutions have a
deagglomeration effect on the system with the potential additional
beneﬁt of minimizing solvent or impurity inclusion due
to agglomeration. The approach provides a simple and robust
inferential control method of the size distribution by maintaining
the number of counts/s constant, hence eliminating the difﬁculties
related to the conversion of chord length distribution (CLD)
to CSD, which would be required by more sophisticated
algorithms, such as nonlinear model predictive control.28
The DNC approach can also be extended to control polymorphic
transformations. This however requires the polymorphic
system to exhibit a distinguishable shape change upon transformation,
or show a separate nucleation event, which can be
correlated with FBRM signal. The approach is different from
the recently published robust model based approaches29
in that
it is model-free and does not require model development,
parameter estimation and optimization. The DNC approach
could be more favorable in an industrial environment as the
iterative way of modeling and experiment may consume
signiﬁcantly more resources and time.
3. Experimental Section
3.1. Materials. In all experiments, aqueous glycine (>99.5% purity,
Fisher BioReagents) solutions were prepared, corresponding to a
solubility of 22.5 g of glycine per 100 g of water at 20 °C.30
Glycine
was dissolved in water by heating to 35 °C at a rate of 5 °C/min. After
the solution equilibrated at 35 °C for 10 min, the temperature of the
solution was lowered to 30 °C with a cooling rate of 1 °C/min, after
which the temperature was further reduced to 25 °C at 0.3 °C/min.
The temperature of the solution was maintained at 25 °C prior to the
start of experiment. Ethanol (99.99% purity, analytical reagent grade,
Fisher Scientiﬁc) was used as an antisolvent.
3.2. Apparatus. The crystallization experiments were performed
in a jacketed 1000 mL glass vessel. The temperature in the vessel was
controlled with a stainless steel Pt100 thermocouple connected to a
thermoﬂuid (Dow Corning 200/20 cS) circulator bath (Huber Ministat
125). The temperature readings were recorded every 4 s on a computer
by a control interface written in LabVIEW (National Instruments). An
overhead stirrer with a stainless steel three-blade marine type impeller
was used to agitate the system at 400 rpm. This agitation speed was
chosen to be high enough to guarantee that particles are well suspended
throughout the process, but low enough to avoid attrition or generation
of bubbles due to vortex formation. A FBRM probe (model A100,
Lasentec) was inserted into the solution to measure chord length
distributions in the range 0.8 to 1000 µm (the lower limit therefore
requires nucleation plus some growth to occur before detection occurs
in the FBRM). The position and orientation of the probe were chosen
according the standard recommendations to avoid particles adhering
to the probe and provide suitable sampling. The distributions were
collected every 6 s and averaged during collection. They were monitored
using the control interface software (version 5.3). The solvent and
antisolvent were pumped into the vessel by a MasterFlex console driver
with EasyLoad II peristaltic pump, calibrated before the experiments.
Two-way solenoid valves were used for the alternate addition of solvent/
antisolvent, which was controlled by the computer. A schematic
representation of the experimental setup is shown in Figure 3.
3.3. Solubility of Glycine in Water-Ethanol Mixture. The method
to determine the solubility of glycine in a mixture of water-ethanol
was based on the one that used by Nozaki and Tanford.31
Glass tubes
containing glycine in various compositions of water-ethanol mixture
were shaken for 24 h in a water bath maintained at 25 °C. The solutions
were separated from the solids by ﬁltration through 0.45 µm syringe
ﬁlters, while the tubes were kept partially immersed in the bath. The
supernatants obtained were transferred into evaporating dishes, which
were weighted before the transfer and weighted again after the transfer.
The supernatants on the dishes were assayed by a gravimetric method
that involves evaporation overnight in a vacuum oven, approximately
Figure 6. Proﬁles of (a) number of counts/s and antisolvent/solvent
addition rate, and (b) number of ﬁnes and coarse for DNC experiment
at a desired number of counts/s of 2000 #/s.
Figure 7. Proﬁles of (a) number of counts/s and antisolvent/solvent
addition rate, and (b) number of ﬁnes and coarse for DNC experiment
at a desired counts/s of 3000 #/s.
1380 Crystal Growth & Design, Vol. 9, No. 3, 2009 Abu Bakar et al.
12 h at 35 °C initially, and then the temperature of the oven was
increased to 107 °C. The evaporating dishes were weighted until
constant weights were obtained.
3.4. Uncontrolled Antisolvent Addition. This experiment was
conducted to see the maximum total counts/s the system could achieve,
so that the value can be used as an upper limit, below which a suitable
desired value of counts/s could be selected for the subsequent controlled
experiments. The procedure involved the continuous addition of
antisolvent to an aqueous glycine solution at two ﬁxed rates of 8 g/min
and 10 g/min until the number of counts/s stabilized.
3.5. DNC by Antisolvent/Solvent Addition. In this experiment,
the desired number of counts/s was selected based on the maximum
counts/s achieved in the uncontrolled antisolvent addition experiment
described previously. Antisolvent was added to the aqueous glycine
solution until primary nucleation occurred and was detected by the
FBRM, after which the addition rate was reduced or stopped to avoid
an overshoot from the desired number of counts/s. Once the number
of counts/s exceeded the desired value, the addition of antisolvent was
stopped and, sequentially, solvent was added to dissolve the ﬁne
particles, thus correcting the number of counts/s. The antisolvent was
continuously added at a slow rate toward the end of the batch to ensure
that the supersaturation in the system is maintained without triggering
nucleation. The experiment was stopped once the suspension reached
the 1000 mL mark on the vessel. The crystals produced were ﬁltered
and dried for subsequent analysis. This procedure was repeated with
different desired numbers of counts/s. The solvent/antisolvent addition
rate was determined according to a proportional control algorithm based
on the error between the desired and measured number of counts/s.
According to this approach, as the number of counts/s starts to increase,
the antisolvent addition rate is decreased and switched to solvent
addition as soon as the number of counts/s exceeds the desired value.
Similarly, during the solvent addition, the addition rate was also in
correlation with the error between the desired and measured number
of counts/s. The decrease in the number of counts/s due to the dilution
effect during solvent/antisolvent addition was also considered (see
Figure 4).
3.6. DNC by Combined Antisolvent Addition and Heating-Cooling.
This experiment follows the same procedure as described in section
3.4, but instead of adding solvent, heating was used to correct the nuclei
count, should the generation of nuclei exceed the desired number of
counts/s. Once the desired number of counts/s was achieved, the
supersaturation was continuously generated through either cooling or
antisolvent addition.
4. Results and Discussion
4.1. Solubility of Glycine in Water-Ethanol Mixture. Although
the proposed DNC approach does not require any
information about the solubility of the crystallizing system, they
were determined in this work to provide useful guidelines for
the method of supersaturation generation as well as acting as
lower limits for the crystallization operating proﬁles. The
solubility values obtained in this study are compared with the
literature values from the work of Nozaki and Tanford,31
and
Dunn and Ross32
as shown in Figure 4. Based on the ﬁgure,
the experimental values appear to be in agreement with the
results obtained by Dunn and Ross, but slightly below the results
obtained by Nozaki and Tanford. A trajectory of solution
concentration due to dilution as a result of the antisolvent
addition, calculated based on mass balance, is also shown in
the ﬁgure.
4.2. Uncontrolled Antisolvent Addition. Figure 5(a) shows
the history of the number of counts/s measured by FBRM
from the start of the antisolvent addition (at t ) 0). Once
nucleated, the number of counts/s increased very rapidly to
almost 7000 #/s for the higher addition rate of antisolvent and
6000 #/s for the lower addition rate, after which they dropped
slightly and eventually stabilized at 6000 #/s and 5000 #/s,
respectively. This result is to be expected because the higher
the supersaturation generation rate, the greater the number of
nuclei that will form. The subsequent drop in counts/s may be
due to a combinational effect of Ostwald ripening and dilution.
Dilution is suggested because the sudden formation of nuclei
produced a rapid reduction in the solute concentration and
together with the continuous addition of antisolvent resulted in
the dissolution of some small crystals. Based on the result of
this experiment, three target numbers of counts/s for the
subsequent DNC experiments were selected, i.e. 2000, 3000 and
4000 #/s, as shown in Figure 5(a).
In order to get information about the size of the crystals, the
square-weighted mean chord length (SWMCL) statistic was
used, deﬁned as
SWMCL )
∑
i)1
k
niMi
3
∑
i)1
k
niMi
2
(1)
where Mi is the midpoint length of an individual channel i, k is
the number of measurement channels and ni is the number of
counts/s in an individual measurement channel. The SWMCL
statistic was found to resemble more closely the Sauter mean
diameter measured using laser diffraction instrument33,34
and
optical microscopy.17,34
The square-weighted chord distribution
(SWCD) is deﬁned as
SWCD )
niMi
2
∑
i)1
k
niMi
2
(2)
The FBRM data do not correspond quantitatively to laser
diffraction or optical microscopic data, since they are based on
different principles of measurement, however the trends could
be analyzed to give information about the dynamic progress of
the crystallization process.
As indicated by the proﬁle of the SWCD in Figure 5(b), due
to the more intense nucleation rate for the higher antisolvent
addition rate, the competition between growth and nucleation
for the solute molecules increases and as a result, the size of
the crystals produced is smaller.
4.3. DNC by Antisolvent/Solvent Addition. Figure 6(a)
shows the proﬁles of the number of counts/s and the antisolvent/
solvent addition rate for a DNC target of 2000 #/s. It can be
seen from the proﬁle that there were a couple of overshoots
and undershoots before the system ﬁnally reached the desired
Figure 8. Proﬁle of the number of counts/s for DNC experiments at
2000, 3000 and 4000 #/s.
Direct Nucleation Control of Crystallization Crystal Growth & Design, Vol. 9, No. 3, 2009 1381
number of counts/s. It was found that primary nucleation is
always difﬁcult to control as the increase in number of counts/s
after the onset of nucleation is very fast.
Figure 6(b) shows the evolution of ﬁne and coarse crystals
during DNC. Fines were deﬁned to be <22 µm, whereas coarse
particles were deﬁned to be >250 µm. The trend shows that
solvent addition decreases the amount of ﬁnes, while having
little effect on the number of coarse crystals. The continuous
increase in the number of coarse crystals indicates growth into
the range >250 µm, rather than nucleation events. This result,
which is in accordance to the mechanism of Ostwald ripening,
shows that DNC is able to provide in situ automatic ﬁne removal
without the need for additional equipment or design. The DNC
experiments at 3000 and 4000 #/s also produced similar results
and trends. Figures 7(a) and (b) show the proﬁles of number of
counts/s with the addition of antisolvent/solvent and the
evolution of ﬁne and coarse crystals, respectively, during the
DNC experiment at 3000 #/s. As before, after a couple of
overshoots and undershoots before the system ﬁnally reached
the desired number of counts/s.
Figure 8 shows the proﬁle of number of counts/s for all the
controlled experiments. It was found that the smallest desired
number of counts/s was more difﬁcult to control, and it gave
higher noise due to the smaller particle concentration. The results
show that different set-points consistently produced similar
solvent/antisolvent addition output proﬁles. This suggests that
an iterative learning control approach35
could be used to
eliminate the over- and undershoots during the ﬁrst part of the
batch and to enhance the overall control performance of the
DNC. According to this approach, a feedforward plus feedback
controller could be implemented, in which the feedforward law
given by the solvent/antisolvent addition proﬁle would be
iteratively improved in a batch-to-batch manner so that the
overall control error would decrease throughout consecutive
batches. The feedforward component of the iterative learning
control would provide batch-to-batch performance improvement,
whereas the feedback control component would minimize the
effects of within batch disturbances. The iterative learning
control approach would improve performance and would provide
increased robustness for industrial implementation. However,
the beneﬁts of DNC in producing higher quality crystals at the
end of the batch are evident already from the implementation
demonstrated in this paper, as discussed below.
Figure 9 compares the plots of SWMCLs for the uncontrolled
and DNC experiments at 2000 and 4000 #/s. The plot of
SWMCL for the DNC experiment at 3000 #/s is omitted to
prevent overcrowding, but it was found to stabilize between
those of 2000 and 4000 #/s. Based on Figure 9, the growth of
the crystals was a maximum in the experiment with the lowest
counts/s set-point, and a minimum in the uncontrolled experiment.
The calculated values of the SWMCL for the controlled
number of counts/s at 2000 #/s, 3000 #/s, 4000 #/s and
uncontrolled experiments are 281, 268, 263 and 232 µm
respectively. The SWMCL values shown in Figure 9 are more
scattered since the smaller the number of particles in the system,
the higher the noise in the statistics of the CSD, leading to an
increased difﬁculty in controlling the system at lower counts/s
set-points. This is exempliﬁed by the controlled experiment at
2000 #/s, which has the most scattered SWMCL values. The
smaller the number of particles, the larger the crystals will grow,
however the corresponding increased noise in the measurement
requires a suitable compromise when selecting the set-point
counts/s values for the DNC.
Figure 9. Proﬁle of the SWMCLs for the uncontrolled and DNC
experiments at 2000 and 4000 #/s.
Figure 10. Microscopic images of crystals obtained from (a) uncontrolled,
(b) DNC at 4000 #/s, and (c) DNC at 2000 #/s experiments.
1382 Crystal Growth & Design, Vol. 9, No. 3, 2009 Abu Bakar et al.
Figure 10 shows the microscopic images of the crystals at
the end of the experiments, which conﬁrmed the results obtained
from FBRM. Based on the images, the size of crystals obtained
from the controlled experiment at the lower counts/s, i.e. 2000
#/s, was generally found to be the largest, while the size of
crystals obtained from the uncontrolled experiments was found
to be the smallest.
The DNC by antisolvent/solvent addition requires a large
volume crystallizer since it involves addition of antisolvent and
solvent. This limitation of the approach may be critical as it
requires additional cost for a large volume vessel. Furthermore,
the large volume of suspension which results from the addition
of antisolvent and solvent may lead to mixing problems. Another
problem is the generation of large volumes of waste materials
that are costly to separate.
4.4. DNC by Combined Antisolvent Addition and Heating-Cooling.
In these experiments the hard constraint of
limited vessel capacity was relaxed by using heating/cooling
to control the level of supersaturation instead of solvent/
antisolvent additions. In this approach, the vessel’s volume
requirement is only to accommodate the initial antisolvent
addition for generating the nuclei.
Figure 11(a) shows the proﬁles of number of counts/s,
SWMCL, process temperature, and antisolvent addition rate for
the experiment using the combined approach of antisolvent
addition and heating-cooling, with a target number of counts/s
of 3000 #/s. Initially, when the number of counts/s went beyond
the desired value upon nucleation as a result of the antisolvent
addition, the addition was stopped and, subsequently, the
temperature was increased gradually, depending on the error
between the measured and desired number of counts/s, to
dissolve the excessive nuclei. The number of counts/s gradually
decreased as the temperature was increased. As soon as the
number of counts/s went below 4000 #/s, cooling was initiated.
The number of counts/s eventually settled at 3000 #/s. They
remained approximately at the same level throughout the
process, even when the system was cooled down to 15 °C at
the maximum rate, set to 0.3 °C/min. This continuous cooling
was required to generate the supersaturation for crystal growth.
The proﬁle of SWMCL of the system during the experiment
shows that the mean size of the crystals was not affected very
much by the heating, only reduced slightly at high temperature
(approximately at 38 °C). As conﬁrmed by the trends of the
evolution of ﬁnes and coarse crystals during DNC in Figure
11(b), the heating dissolves only ﬁne crystals, which do not
contribute a lot to the SWMCL. As the cooling continues, the
mean size of the crystals increased signiﬁcantly as shown by
the steady increase of the SWMCL (Figure 11(a)). It can also
be seen that, during cooling, there was a slight increase in the
number of counts/s due to secondary nucleation. However,
during the second part of the batch the process is dominated by
growth and the amount of ﬁnes generated was not signiﬁcant
enough to trigger the dissolution mechanism of the DNC. In
this experiment, the amount of ethanol consumed was very little
(53 g), as it was only used to initiate the nucleation.
As an alternative to cooling toward the end of the batch for
supersaturation generation, hence promoting growth, the application
of antisolvent addition was also tested. Figure 12 shows
the proﬁles of the number of counts/s, SWMCL, process
temperature and antisolvent addition rate during the last 100
min of the experiment.
The ﬁgure shows that the crystals started to grow in size with
cooling and then further growth was promoted with the addition
of antisolvent. Although antisolvent was continuously added, the
growth rate decreased signiﬁcantly after about 125 min. The
solubility of glycine has a lower sensitivity to the solvent/antisolvent
ratio as the amount of antisolvent increases in the system (see
Figure 4). The decrease in the growth is probably due to the fact
that, as the sensitivity in the solubility decreases, the dilution effect
becomes predominant in decreasing the supersaturation, thus
depleting the driving force for growth (see Figure 4).
In comparison, the calculated SWMCL for the crystals
obtained from antisolvent addition toward the end is 297 µm,
whereas that obtained from cooling alone toward the end is 258
µm. However, the former consumed a signiﬁcantly larger
amount of ethanol (730 g). The batch time of DNC by the
Figure 11. Proﬁles of (a) number of counts/s, SWMCL, process
temperature, and antisolvent addition rate, and (b) number of ﬁne and
coarse particles, and process temperature in the DNC experiment by a
combined approach.
Figure 12. Proﬁles of number of counts/s, SWMCL, process temperature,
and antisolvent addition rate in the DNC by an alternative
combined approach.
Direct Nucleation Control of Crystallization Crystal Growth & Design, Vol. 9, No. 3, 2009 1383
combined approach was longer than that of DNC by antisolvent/
solvent addition because heating/cooling gave a slower response.
Table 1 compares the outcomes of applying different DNC
strategies to reach a stabilized desired counts/s of 3000.
Suitable automatic control of the solvent/antisolvent addition
or heating and cooling is difﬁcult especially during the initial
part of the batch when the ﬁrst nucleation events are generated.
The initial overshoot can be minimized by automatically
stopping the antisolvent addition when a continuous increase
in the number of counts/s has been detected, but before the
desired set-point is achieved. The overshoots can be minimized
by using a control structure with antiwindup which allows an
immediate switch from antisolvent addition to solvent addition
(or from cooling to heating), and vice versa when the desired
set-point is achieved. An iterative learning control approach
could lead to batch-to-batch improvement in the performance
of the DNC. Although in the case of the model system used in
this work (glycine in water) attrition or agglomeration was not
observed, these are common phenomena in the case of industrial
crystallization systems, and are not accounted for either by the
supersaturation control or by the classical open-loop control
approaches. The DNC approach could have the additional
beneﬁts in the control of systems exhibiting these phenomena.
The ﬁnes generated by attrition would be detected as an increase
in the number of counts/s and hence would be eliminated by
the DNC, by controlled dissolution. Additionally, the particular
feature of the DNC approach of generating partial dissolution
by repeatedly driving the process into the undersaturated regime
has a beneﬁcial deagglomeration effect.
5. Conclusions
The direct nucleation control (DNC) approach implemented
in this work controls directly the amount of nuclei present using
information provided by FBRM, in a feedback control strategy
through (i) addition of antisolvent or reduction of the suspension
temperature to generate nuclei up to a desired number of counts/s
and (ii) addition of solvent or increment of the suspension
temperature to correct the nuclei count, should excess nucleation
occur. It has been found that the DNC approach implemented
in these experiments, using glycine in water-ethanol mixture
as a model system, was able to produce crystals with a larger
size than those obtained from uncontrolled experiments. Results
also suggest that the lower the desired number of counts/s, the
larger the size of the crystals produced. In the experiments which
used the application of the combined approach, the results
indicate that the crystals produced were even larger. In some
cases better control of nuclei generation and correction are still
needed, but nonetheless the approach has proved to provide a
robust crystallization control strategy.
Acknowledgment. Financial support provided by the Engineering
and Physical Sciences Research Council (EPSRC), U.K.
(Grant EP/E022294/1), is gratefully acknowledged. One of the
authors (M.R.A.B.) is grateful to the Malaysian Ministry of
Higher Education and the International Islamic University
Malaysia for a scholarship.
References
(1) Reutzel-Edens, S. M. Curr. Opin. Drug DiscoVery DeV. 2006, 9, 806–815.
(2) Yu, L. X.; Lionbergera, R. A.; Rawa, A. S.; D‘Costa, R.; Wub, H.;
Hussain, A. S. AdV. Drug DeliVery ReV. 2004, 56, 349–369.
(3) Barrett, P.; Smith, B.; Worlitschek, J.; Bracken, V.; O’Sullivan, B.;
O’Grady, D. Org. Process Res. DeV. 2005, 9, 348–355.
(4) Birch, M.; Fussell, S. J.; Higginson, P. D.; McDowall, N.; Marziano,
I. Org. Process Res. DeV. 2005, 9, 360–364.
(5) Braatz, R. D. Annu. ReV. Control 2002, 26, 87–99.
(6) Larsen, P. A.; Patience, D. B.; Rawlings, J. B. IEEE Control Syst.
Mag. 2006, 26, 70–80.
(7) Worlitschek, J.; Mazzotti, M. Cryst. Growth Des. 2004, 4, 891–903.
(8) Gron, H.; Borissova, A.; Roberts, K. J. Ind. Eng. Chem. Res. 2003,
42, 198–206.
(9) Nonoyama, N.; Hanaki, K.; Yabuki, Y. Org. Process Res. DeV. 2006,
10, 727–732.
(10) Yu, Z. Q.; Chow, P. S.; Tan, R. B. H. Ind. Eng. Chem. Res. 2006, 45,
438–444.
(11) Fujiwara, M.; Nagy, Z. K.; Chew, J. W.; Braatz, R. D. J. Process
Control 2005, 15, 493–504.
(12) Zhou, G. X.; Fujiwara, M.; Woo, X. Y.; Rusli, E.; Tung, H.; Starbuck, C.;
Davidson, O.; Ge, Z.; Braatz, R. D. Cryst. Growth Des. 2006, 6, 892–898.
(13) Nagy, Z. K.; Chew, J. W.; Fujiwara, M.; Braatz, R. D. J. Process
Control 2008, 18, 399–407.
(14) Ward, J. D.; Mellichamp, D. A.; Doherty, M. F. AIChE J. 2006, 52,
2046–2054.
(15) Barrett, P.; Glennon, B. Chem. Eng. Res. Des. 2002, 80, 799–805.
(16) Barthe, S.; Rousseau, R. W. Chem. Eng. Technol. 2006, 29, 206–211.
(17) Chew, J. W.; Black, S. N.; Chow, P. S.; Tan, R. B. H. Ind. Eng. Chem.
Res. 2007, 46, 830–838.
(18) Chew, J. W.; Chow, P. S.; Tan, R. B. H. Cryst. Growth Des. 2007, 7,
1416–1422.
(19) Liotta, V.; Sabesan, V. Org. Process Res. DeV. 2004, 8, 488–494.
(20) Fujiwara, M.; Chow, P. S.; Ma, D. L.; Braatz, R. D. Cryst.l Growth
Des. 2002, 2, 363–370.
(21) Nagy, Z. K.; Fujiwara, M.; Woo, X. Y.; Braatz, R. D. Ind. Eng. Chem.
Res. 2008, 47, 1245–1252.
(22) Doki, N.; Seki, H.; Takano, K.; Asatani, H.; Yokota, M.; Kubota, N.
Cryst. Growth Des. 2006, 4, 949–953.
(23) Tadayyon, A.; Rohani, S. Can. J. Chem. Eng. 2000, 78, 663–673.
(24) Sheikhzadeh, M.; Trifkovic, M.; Rohani, S. Chem. Eng. Sci. 2008,
63, 991–1002.
(25) Sheikhzadeh, M.; Trifkovic, M.; Rohani, S. Chem. Eng. Sci. 2008,
63, 1261–1272.
(26) Woo, X. Y.; Nagy, Z. K.; Tan, R. B. H.; Braatz, R. D. Cryst. Growth
Des. 2009, in press.
(27) Lewiner, F.; Fevotte, G.; Klein, J. P.; Puel, F. Ind. Eng. Chem. Res.
2002, 41, 1321–1328.
(28) Nagy, Z. K.; Braatz, R. D. AIChE J. 2003, 49 (7), 1776–1786.
(29) Hermanto, M. W.; Chiu, M.-S.; Woo, X. Y.; Braatz, R. D. AIChE J.
2007, 53, 2643–2650.
(30) Mullin, J. W. Crystallization; Butterworth-Heinemann: London, 2001.
(31) Nozaki, Y.; Tanford, C. J. Biol. Chem. 1971, 246, 2211–2217.
(32) Dunn, M. S.; Ross, F. J. J. Biol. Chem. 1937, 125, 309–332.
(33) Heath, A. R.; Fawell, P. D.; Bahri, P. A.; Swift, J. D. Part. Part. Syst.
Charact. 2002, 19, 84–95.
(34) Yu, W.; Erickson, K. Powder Technol. 2008, 185, 24–30.
(35) Lee, K. S.; Lee, J. H. J. Process Control 2003, 13, 607–621.
CG800595V
Table 1. Different DNC Strategies To Reach a Stabilized Desired Counts/s of 3000 and Their Resultant Batch Times, Amount of Antisolvent
Used and SWMCL of the Crystals Obtained
DNC strategy
batch time
(min)
amount of antisolvent
used (g)
SWMCL of crystals
obtained (µm)
addition of antisolvent to nucleate; control of counts/s by additions of
solvent and antisolvent; antisolvent addition toward the end to promote
growth
68 370 268
addition of antisolvent to nucleate; control of counts/s by heating and
cooling; cooling toward the end to promote growth
168 53 258
addition of antisolvent to nucleate; control of counts/s by heating and
cooling; antisolvent addition toward the end to promote growth
145 730 297
1384 Crystal Growth & Design, Vol. 9, No. 3, 2009 Abu Bakar et al.