Content uploaded by Zai-Qun Yu
Author content
All content in this area was uploaded by Zai-Qun Yu on Oct 30, 2017
Content may be subject to copyright.
REVIEW PAPER
RECENT ADVANCES IN CRYSTALLIZATION
CONTROL
An Industrial Perspective
Z. Q. Yu
1
, J. W. Chew
1
, P. S. Chow
1
and R. B. H. Tan
1,2,
1
Institute of Chemical and Engineering Sciences Ltd, Singapore.
2
Department of Chemical & Biomolecular Engineering, National University of Singapore, Singapore.
Abstract: Crystallization is the most important unit operation for the separation and purification
of chemicals in the pharmaceutical and fine chemical industries. Crystallization processes in
pharmaceutical active ingredient manufacturing have been traditionally a recipe-based
operations, offering little scope for dynamic process control and improvement. With the
change in regulatory climate from quality-by-testing (QbT) to quality-by-design (QbD) and with
the advent of the process analytical technology (PAT) initiative, it is timely to examine the
impact of such quality-based emphasis on crystallization control. In this paper, we review the
important recent developments in the control of crystallization process, and discuss their
feasibility and scope for implementation in industrial processes. The control methods to achieve
different aspects of crystal product quality, including particle size distribution (PSD), crystal habit
and polymorphic form, are discussed separately.
Keywords: crystallizaton; PAT; quality by design; PSD; crystal habit; polymorph control.
INTRODUCTION
Crystallization of the solid phase from solution
remains the predominant process for separ-
ation and purification of pharmaceuticals and
fine chemicals. The operating conditions of a
crystallization process determine the proper-
ties of the crystal product such as purity, size
and shape distributions and polymorphic
form. These properties in turn affect the down-
stream processing and handling, e.g., filtration
and drying, and eventually the therapeutic
properties and shelf-life of the final formulated
product. As such, the control and consistency
of solid phase properties through crystalliza-
tion has been the focus of considerable indus-
trial and academic research.
Crystallization control is essentially product
quality control, i.e, how to operate the process
to obtain the desired product quality in terms
of crystal solid attributes such as particle
size and shape, particle size distribution
(PSD) and polymorphic form. Pharmaceutical
manufacture has traditionally been a
recipe-based operation, in which the pro-
cesses are controlled such that the trajec-
tories follow the specifications submitted for
regulatory filings. There is also little scope
for crystallization process control apart from
low-level control such as temperature control
to ensure the specified temperature profile is
followed throughout the process. The quality
of the product is only determined by testing
at the end. Such a quality-by-testing (QbT)
approach often leads to failed batches and
loss of profit to the company. In view of
these shortcomings, the regulatory environ-
ment has been changed recently to shift
from regulating of the process to regulating
of the product, i.e., from a QbT to QbD (qual-
ity-by-design) approach.
The FDA (Food and Drug Administration)
now advocates the adoption of the QbD
approach with the aim that product quality
and performance are achieved and assured
by design of effective and efficient manufactur-
ing processes based on the mechanistic
understanding of the underlying science (Hus-
sain, 2006; McKenzie et al., 2006). Manufac-
turers would be given certain degree of
flexibility to make changes to process as long
as they can show that the changes made are
within the design space and that the quality
of the product is maintained. Therefore, there
is more scope for optimal and adaptive pro-
cesses as long as quality can be assured.
893 Vol 85 (A7) 893–905
Corresponding to:
Associate Professor
R.B.H. Tan, Institute of
Chemical and Engineering
Sciences Ltd, 1 Pesek Road,
Jurong Island,
Singapore 627833.
E-mail: reginald_tan@
ices.a-star.edu.sg
DOI: 10.1205/cherd06234
0263–8762/07/
$30.00 þ0.00
Chemical Engineering
Research and Design
Trans IChemE,
Part A, July 2007
#2007 Institution
of Chemical Engineers
Under QbD, real-time optimisation would be possible, with real
economic benefits, e.g., maximizing yield or productivity, and
reducing wastage due to failed batches compared with pre-
vious recipe-based operation.
Crystal quality is a result of the interplay among state vari-
ables in the course of crystallization as illustrated schemati-
cally in Figure 1. The interaction network is represented by
notched arrows. Various control strategies of crystallizers
differ in the use of open loop or closed loop, or which state vari-
ables are used as feedback signal in closed loop, or which
state variable is treated as the controlled variable (solid-state
attributes or supersaturation).
Solid-state attributes under frequent study include particle
size distribution (PSD), polymorphic specificity and crystal
habit. Most previous studies have been targeted at optimising
PSD (especially in cooling crystallization), probably because
the formation mechanism and manipulation approach of
PSD are better established than those of other solid-state
attributes. It has been reported that desirable PSDs are some-
times a compromise between conflicting requirements of
downstream processing (Kim et al., 2003; Liotta and Sabesan,
2004). Methods and algorithms developed for PSD control can
be adopted for the control of other solid-state attributes, and
tightening of crystallization control is an essential part of the
initiative to improve the performance of pharmaceutical
processes from two Sigma (4.6% defectives) to six Sigma
(2 ppb defectives) (Leuenberger and Lanz, 2005).
The emergence and development of the latest generation
of state-of-the-art PAT (process analytical technology) tools
and techniques have been very much in step with the
emphasis on process understanding of QbD and open up
tremendous new opportunities for effecting process control.
Yu et al. (2004) provided a review of the applications of
PAT to crystallization processes.
In this paper, we review the important recent developments
in crystallisation control, and discuss their feasibility and
scope for implementation in industrial processes. Control of
PSD, polymorph and crystal habit will be dealt with in the
next three sections respectively. Challenges and future
developments will be considered at the end of this review.
CONTROL OF CRYSTALLIZATION TO OBTAIN
DESIRED PSD
A Historical Viewpoint of PSD Engineering: From
Meta-Stable Zone to Optimal Control
The metastable zone, the region bound by the solubility
curve and the metastable limit in which spontaneous primary
nucleation does not occur, is a fundamental concept in
crystallisation control. It is considered the ideal region for
crystal growth once the nuclei have formed. Therefore the
temperature profiles used during cooling crystallization are
usually designed to progress within the meta-stable zone.
Different temperature profiles have been derived either
empirically or mathematically. In the pioneering work by
Mullin and Nyvlt (1971), they derived ‘programming cooling’
profiles by establishing mathematical models for cooling crys-
tallization of potassium sulphate and ammonium sulphate.
The temperature profiles were derived such that the supersa-
turation was kept constant throughout the crystallization. It
has been reasoned that the supersaturation does not necess-
arily have to be constant, and it can be allowed to fluctuate
within the meta-stable zone as long as secondary nucleation
is suppressed (Jones and Mullin, 1974). On the other hand,
keeping it constant simplifies the deduction of operating pro-
file to some degree. For instance, Wey and Karpinski (2002)
deduced the following temperature profile for seeded crystal-
lization under the assumptions of no spontaneous nucleation,
constant growth rate Gand a linear solubility-temperature
relation:
T¼TI3MS
aMsol
G3t3
3L3
S
þG2t2
L2
S
þGt
LS
! (1)
where Tstands for temperature, Mfor mass, tfor time, L
S
for
seed size, and T
I
for initial slurry temperature. Subscripts S
and sol represent seed and solvent and ais the slope in a
linearised solubility-temperature relation:
c¼aT þb(2)
Figure 1. Interplay among state variables of the crystallizer.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
894 YU et al.
Assuming further that the initial seed size is virtually zero
compared with their subsequent size, Mullin pointed out
that the temperature profile to maintain a constant
supersaturation should be cubic in time in seeded crystallisation
(Jones and Mullin, 1974; Mullin, 2001):
T¼TI(TITF)t
t
3
, (3)
where
t
is the overall batch time. T
F
denotes the final slurry
temperature.
For the case of semi-batch crystallizations, several studies
have shown that more uniform and larger crystals can be pro-
duced in reactive crystallizations when reagents are added
according to curved profiles rather than at constant rates.
Karpinski and Wey (2002) proposed a growth ramp equation
similar to equation (1) for the precipitation of silver bromide
and found very good agreement between measured and
calculated final size of crystals. Kim et al. (2005) adopted a
simple growth ramp analogical to equation (3) for agent
addition in reactive crystallization of an active pharmaceutical
ingredient (API) in Bristol-Myers Squibb. Results showed that
a cubic addition profile yielded larger and more well-defined
crystals than constant addition rate. The same principle
also applies to the addition rate of anti-solvent in anti-solvent
crystallization as described by Tavare (1995).
The concept of ‘optimal operation’ of crystallizers (Jones,
1974) takes a step further from the abovementioned
‘programmed cooling’ in that a specific control objective is
defined in terms of desired properties of final crystal products,
such as maximum average size, a narrow crystal size distri-
bution, minimum mass ratio of newly-formed crystals to
grown seeds, shortest batch time or combination of these.
The objective is targeted by manipulation of temperature tra-
jectory based on a priori knowledge of nucleation and growth
kinetics as a function of supersaturation. This is often referred
to as the first-principles approach (Rawlings et al., 1993;
Braatz, 2002), where optimisation procedures and process
modelling are needed to generate the optimal temperature
profiles. For a detailed overview of model development for
solution crystallization, the readers are referred to the
review article by Rawlings et al. (1993).
Renewed interest in optimal design of crystallizers has
been partly fuelled by a fundamental shift in the regulatory
climate of API manufacturing towards a QbD approach
(Ma et al., 2002; Zhang and Rohani, 2003; Choong and
Smith, 2004; Patience et al., 2004; Togkalidou et al., 2004;
Worlitschek and Mazzotti, 2004; Costa et al., 2005; Hu
et al., 2005). Advances in computer hardware and software
allow more factors to be accounted for in crystallization
models than possible previously. Hu et al. (2005) included
growth rate dispersion in their model for ammonium
sulphate–water system and Costa et al. (2005) included
agglomeration in their model for adipic acid –water system.
Ma et al. (2002) generated optimal temperature profiles for
potassium dihydrogen phosphate–water system where two-
dimensional growth was considered in crystallization
modelling. The optimal temperature profiles (and thus super-
saturation profiles) depend heavily on the expression of the
objective function (Chung et al., 1999; Ward et al., 2006).
For example, minimization of coefficient of variation of PSD
in unseeded crystallization leads to fast cooling at the
beginning of operation with the aim to generate nuclei in a
short time interval, while maximization of mass-based aver-
age crystal size necessitates convex temperature profiles.
Most of these optimization studies consider a single objective
even though crystallization typically has multiple performance
objectives, e.g., achieving CSD with narrow size distribution
and specified end size in a minimum time of operation. The
recent work by Sarkar et al. (2006) demonstrated the
potential for multi-objective optimization approach for a
batch cooling crystallization process.
The model-based optimization studies mentioned so far
only concentrate on batch cooling crystallization. There are
few reports on the modelling of anti-solvent or drowning out
crystallization processes because of the additional complex-
ity associated with mixing. The assumption of perfect mixing
used in most studies may not be applicable for practical
anti-solvent systems because supersaturation level—and
therefore nucleation and crystallization kinetics—is heavily
dependent on how the different components are mixed. A
recent paper by Woo et al. (2006) has addressed this issue
by incorporating a turbulent computational fluid dynamics
(CFD) code with a multi-environment probability density func-
tion (PDF) model, and the population balance equation
(PBE). This coupled CFD-PDF-PBE algorithm was able to
simulate the evolution of the crystal size distribution of a
semi-batch anti-solvent crystallization process taking into
account micromixing between the different components.
This modelling effort provides valuable understanding of the
effects of mixing on crystallization and would offer a more
scientific basis for the design and scale-up of crystallizers.
It can be used to test the robustness of the optimal control
profile to the ‘mixing perturbations’ for a given crystallizer
configuration and scale as illustrated in Figure 2. If the simu-
lation results show that the optimal profile does not produce
the desired crystal product quality when imperfect mixing
is considered, simulating the process using the CFD-
PDF-PBE model for a range of profiles can guide the process
Figure 2. Procedure for testing the robustness of the optimal control
profile to the ‘mixing perturbations’ for a given crystallizer configur-
ation and scale.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
RECENT ADVANCES IN CRYSTALLIZATION CONTROL 895
designer to systematically select the control profile that gives
the desired product quality. This can reduce the number of
experiments required to arrive at a robust design.
After a temperature or anti-solvent/reactant addition profile
as a function of time corresponding to constant supersatura-
tion or optimal operation has been generated, it is traditionally
tracked by a low-level temperature controller of PID type.
Crystallizers are largely operated in ‘open-loop’ mode, in
which solid-state attributes and supersaturation are not
used as feedback signals. Performance is subject to model
imprecision, uncertainties in parameter estimation and oper-
ating disturbances. Simplified models may fail to capture
the process dynamics satisfactorily, causing the computed
trajectory deduced from modelling to result in sub-optimal
operation. In addition, uncertainties in parameter estimation
are inherent and related to the random errors in real data.
Operating disturbances are inevitable since crystallizers are
located downstream to various synthesis trains. The actual
supersaturation therefore may be far from its theoretical
values in open-loop control strategies. Efforts have been
made to handle these issues via model improvement and
robust control theory (Ma et al., 1999; Matthews et al.,
1996; Nagy and Braatz, 2004).
Although the first-principles approach has the advantages
of enhancing process understanding and potential use in pro-
cess scale-up, the considerable expertise and time required
in process modelling and optimization has limited its research
and application mainly to small scale laboratory studies. The
application of first-principles approach to industrial scale has
been rarely reported. A simpler approach, termed the direct
design approach, is more amenable for large-scale
implementation in industry. In this approach, feedback control
is used to follow a setpoint based on a state measurement
which may include temperature, solution concentration, or
crystal size and shape distribution. Direct design approaches
based on measurements of solution concentration, crystal
size and shape distribution and polymorphic form will be dis-
cussed in the following sections.
Control Strategy for PSD Based on Concentration
Measurement
The predominant role of supersaturation in particle
formation as illustrated in Figure 1 has prompted its use as
the controlled variable in the feedback control of crystalliza-
tion. The direct design approach had been attempted
based on concentration measurement from online techniques
such as densitometry, conductivity and refractometry as early
as the 1980s and 1990s (Bordui et al., 1985; Gabas and
Laguerie, 1992; Gutwald and Mersmann, 1990), but the
implementation of this approach has not been widespread
because of the limitations associated with the online sensors
at that time. As attenuated total reflectance Fourier transform
infrared spectroscopy (ATR-FTIR) and other in situ tech-
niques such as ATR-UV and Raman for concentration
measurement became available, interest in the direct
design approach has been renewed. Since ATR-FTIR is the
most widely-used technique for concentration measurement
during crystallization, the following discussion will focus
only on this technique. The advantages of ATR-FTIR include
accurate liquid phase concentration measurement even in
the presence of solids, the capability to measure multiple
components simultaneously, and eliminating the need for
an external sampling stream. The measurement principles
of ATR-FTIR and its first use for in situ measurement of
supersaturation are described in Dunuwila et al. (1994).
The feasibility and usefulness of ATR-FTIR have been
demonstrated by several research groups for feedback con-
trol of cooling crystallisation (Feng and Berglund, 2002;
Gro
¨net al., 2003; Jones and Teodossiev, 1988; Lewiner
et al., 2001; Liotta and Sabesan, 2004) and anti-solvent crys-
tallization (Yu et al., 2006a; Zhou et al., 2006). This strategy
has a two-level structure (Liotta and Sabesan, 2004) and is
often termed as ‘supersaturation control’ or ‘concentration
control’. The transient value of supersaturation measured
by ATR-FTIR is compared with its setpoint and the difference
is translated to an updated cooling or anti-solvent addition
rate for a low-level controller. In other words, the setpoints
for lower level controllers are not time-dependent as in
open-loop control but state-dependent (Fujiwara et al.,
2005). Obviously supersaturation control possesses the
capability to absorb some disturbances in operating con-
ditions, such as changes in initial solute concentration,
seed loading and crystal growth rate, and so on. For
example, the anti-solvent crystallization of paracetamol
remained at the supersaturation setpoint of 0.05 even when
the seed loading was doubled (Figure 3) (Yu et al., 2006a).
As was obvious in this example, batch time is allowed to
vary in order to accommodate the operating disturbances or
changes.
With supersaturation control in place, the determination of
a setpoint supersaturation becomes the key consideration.
Choosing a constant supersaturation level within the meta-
stable zone would be the simplest option compared to any
supersaturation profile that varies with time. The study by
Worlitschek and Mazzotti (2004) showed that optimized cool-
ing did not produce crystals with CSD any more superior to a
simple constant supersaturation operation. Moreover, con-
stant supersaturation is claimed to be beneficial in developing
a well-defined crystal morphology (Karpinski and Wey, 2002).
The proper set point value of supersaturation and corre-
sponding batch time can be determined readily from a
small number of batches along with other operating par-
ameters aided by design of experiment (DOE).
Supersaturation control has been proven to be a beneficial
technique for obtaining product crystals of better quality, yet it
is not always necessary. Chew et al. (2007) have reported
that for a fast growth system, sophisticated methods of
Figure 3. Relative supersaturation was maintained at a constant level
by feedback control although seed loading (C
s
) was doubled
(Yu et al., 2006a).
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
896 YU et al.
control did not prove advantageous. Figure 4 shows that
open-loop temperature control following different temperature
trajectories and feedback concentration control did not differ
much in terms of chord length distribution (CLD), which is
related to PSD.
One of the advantages of the ATR-FTIR technique is its
ability to measure concentrations of multiple components
simultaneously after careful calibration (Togkalidou et al.,
2002), which is highly desirable for monitoring of industrial
processes wherein changes in concentrations of impurities
may pose problem to crystal quality. However, its calibration
in industrial context may be difficult because of the difficulty
in manipulating the concentration of impurities in calibration
solutions. The concentration range of both impurities and
desired solute in calibration solutions must cover the possible
fluctuations in commercial crystallizers to achieve a satisfac-
tory measurement precision. Additionally, from an industrial
standpoint, the vulnerability of the ATR element poses sev-
eral setbacks (Lewiner et al., 2001). Mechanical damage or
chemical deterioration of the ATR element immediately
affects the accuracy of the calibration; and encrustation of
the probe can easily occur. These concerns pose difficulties
in implementing ATR-FTIR in industrial crystallizers. Super-
saturation measurement is often seen as too troublesome
or complex for routine industrial crystallizations, but may
become a critical factor in the control of specific polymorphs
(Kee et al., 2006) as discussed later.
Feedback Control Based on PSD Measurements
If the ultimate control target is PSD, one may think that
using PSD measurements directly as feedback signal is the
most sensible way forward instead of using supersaturation.
This approach has, however, been hampered by the lack of
reliable in-situ sizing devices. Off-line sensors for the
measurement of PSD typically rely on good sampling which
is difficult to achieve. Changes to the crystals may occur
during the transfer of samples to the offline devices and
therefore dynamic changes of the process cannot be tracked
accurately. In recent years, probe-based Lasentec focused
beam reflectance measurement (FBRM) has gained popular-
ity for in situ characterization of high-concentration particulate
slurries. The FBRM probe utilizes laser light backscattering
technology to supply, in real time, a chord length distribution
(CLD) that is a function of the number, size, and shape of par-
ticles under investigation (Barrett and Glennon, 1999).
FBRM measures the backscattering properties of suspen-
sions, and FBRM data are related not only to PSD but also to
refractive index of solvents and particles, as well as particle
morphology. Research work has been reported to restore
PSD from FBRM data based on first principles (Barthe and
Rousseau, 2006; Bloemen and De Kroon, 2005; Hukkanen
and Braatz, 2003; Li and Wilkinson, 2005; Togkalidou et al.,
2004; Worlitschek and Mazzotti, 2002; Wynn, 2003). How-
ever this is an extremely difficult job and many assumptions
have to be made to accomplish the restoration. Although
these authors were able to verify their models with exper-
imental data, their algorithms are only applicable to well
defined systems with known shape and optical properties
and may not be extendable to systems in general. Much
work needs to be done in understanding the optics and phy-
sics of backscattering before PSD data can be restored
reliably from FBRM data. Another way to use FBRM data is
to empirically correlate the statistics of CLD with the results
of traditional sizing techniques. Heath et al. (2002) presented
a comprehensive review on correlation studies of FBRM data
where the similarity between PSD and square-weighted CLD
was confirmed experimentally in some particulate systems.
De Clercq et al. (2004) performed a comparative study of
FBRM data with results obtained via laser diffraction and
image analysis, and concluded that although the measure-
ment principles are completely different, PSDs with number-
weighted mean diameters above 150 mm were similar for
all measurement techniques. Nevertheless, most of these
correlations are system-specific and are difficult to
generalize.
More often, FBRM data are used qualitatively for monitor-
ing the process evolution with time, e.g., to identify the
onset of primary nucleation, detect attrition and agglomera-
tion. A widely used statistic of CLD is the counts of chord
lengths in the whole range or certain channels of chord
length. A jump in the total counts in the whole range can
be used to indicate the onset of nucleation, so it can be
used in measuring meta-stable limit (Barrett and Glennon,
2002; Fujiwara et al., 2002; Kline et al., 2006; Liotta and
Sabesan, 2004). A continuous increase in total counts
during crystallization was sometimes interpreted as the
result of secondary nucleation (Doki et al., 2004). The
counts in different channels have been frequently used to
isolate nucleation events and crystal growth. The counts in
channels of short chord length are related with fines and
thus with nucleation, while those in channels of long chord
length are associated with big crystals and thus with crystal
growth. The dividing line between ‘short’ and ‘long’ chord
length is not universally agreed upon, and is somewhat arbi-
trary depending on the specifics of particulate systems and
individual judgement. For example, in the study by Lafferre`re
et al. (2004), the change in counts from 1 to 50mm were
used to interpret nucleation and those from 50 to 160mm
Figure 4. Normalized square-weighted chord length distributions
(CLD) obtained from cooling crystallization of glycine using open-
loop temperature control following four different temperature profiles
and feedback concentration control at two different supersaturation
levels (Chew et al., 2007).
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
RECENT ADVANCES IN CRYSTALLIZATION CONTROL 897
were used for indicating crystal growth. Barrett and
Glennon (2002) adopted a different dividing line: 0–20 mm
and20–250 mm. Similarly, counts of chord lengths were
also used to explain the changes in PSD due to agglomera-
tion and breakage (Richmond et al., 1998). Moreover, FBRM
has also been used to monitor polymorphic forms (O’Sullivan
et al., 2003; Scho
¨ll et al., 2006).
Due to the complicated relationship between CLD and
PSD, there have been few reports on the use of FBRM
signal for feedback control despite the widespread use of
FBRM in process monitoring. Tadayyon and Rohani (2000)
constructed a feedback control system for a continuous cool-
ing crystallizer using FBRM signals (counts of chord lengths
smaller than 125 mm) in which the flow rate of fines dissol-
ution stream was adjusted in a real-time manner to increase
average crystal size.
Most recently, Chew (2006) demonstrated that a fully-
automated technique using FBRM measurements and
feedback control of temperature was successful in achieving
consistent PSD in repeated batch crystallizations of glycine
and paracetamol. Figure 5(a) shows the temperature –time
profile for a typical run. The saturated solution (Point A) is
cooled at a pre-set rate until nucleation is detected by the
FBRM (Point B). The system is allowed to stabilize at the
temperature of Point B until primary nucleation is completed.
Then, a heating ramp is implemented while using the FBRM
to monitor the PSD of the ‘seed’ crystals. The heating gradu-
ally redissolves the fines, thereby narrowing the PSD towards
the setpoint of coefficient of variation of CLD which is used as
the controlled variable. When the desired quality of these
internally generated ‘seeds’ is achieved (Point D), the
system is cooled at a constant rate to allow the crystals to
grow until the final yield is attained (Point E). Figure 5(b)
reveals that a desirable CLD (and, by inference, PSD)
could be consistently achieved using this technique. This
new development appears to be very promising for
implementation at industrial scale owing to its simplicity and
robustness.
Effectiveness of Seeding in PSD Engineering
Seeding is often employed to obviate the adverse impacts
of uncontrolled unseeded crystallizations, namely encrusta-
tion on internal fittings of crystallizers, and random variations
in the number of nuclei leading to inconsistent product and
possibly unfavourable PSD with a large amount of fines. It
is estimated that up to 30–50% of solutes crash out of sol-
ution rapidly through spontaneous nucleation for organic sys-
tems (Beckmann, 2000), leaving little that can be done about
PSD engineering through supersaturation control (Chew
et al., 2007). Seeding is also especially important for systems
which are difficult to nucleate or tend to induce liquid-phase
separation, because in this case, it reduces nucleation time
that may be otherwise too long from an economic perspec-
tive. Seeding is also known to be advantageous in ensuring
product consistency because the size range of the seeds,
whether the seeds are added dry or wet, the temperature at
which the seeds are added, and the amount of seeds are
all pre-determined.
In PSD engineering, the purpose of seeding a supersatu-
rated solution is to provide starting surface area for crystal
growth and avoid nucleation as much as possible. Seed
loading, seed size and timing of seeding are three critical
quantitative parameters in a seeding policy towards such a
goal.
Seed loading and size
Research on seeding techniques has been mainly targeted
towards how to determine seed size and seed loading.
Kubota et al. (2001) emphasized the importance of seed
loading to guarantee the unimodal distribution of final pro-
ducts. Their work showed that unimodal distribution of final
products could be achieved over a wide range of generation
rates of supersaturation when seed loading exceeded a
critical value which was determined empirically. One
consequence of accommodating high generation rates of
supersaturation is the high seed loading which inevitably
compromises productivity. Lung-Somarriba et al. (2004) put
forward a procedure to determine seed size and seed loading
which took into consideration the attrition of large crystals.
They proposed that the seed size should be smaller than
one quarter of the maximal product size over which attrition
will hinder crystal growth. Similar to the critical seed loading
Figure 5. Temperature –time profile for a typical run, (b) CLDs
recorded at the end of eight different batches of cooling crystallization
runs of glycine using the FBRM control technique by Chew (2006).
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
898 YU et al.
proposed by Kubota et al. (2001), they used critical surface
area obtained empirically to determine the size-mass
couple of seeds.
Gutwald and Mersmann (1994) incorporated generation
rate of supersaturation in the calculation of seed loading.
For a given cooling crystallization system with a cooling
rate of T
†
, the necessary seed mass M
S
was:
MS¼
w
v
w
a
dc
dT
Msol T
†
LS
Gmet
1þ1
2
GmetDTmet
T
†
LS
!
2
(4)
where
f
v
and
f
a
are the volume and surface shape factors of
crystals respectively, LSthe average seed size, dc/dT the
slope of the solubility curve, M
sol
the mass of solvent, G
met
the maximum allowable growth rate within metastable zone,
DT
met
the maximum allowable undercooling. It can be seen
that this seeding policy integrates important operating par-
ameters, two of which need to be measured separately,
namely G
met
and DT
met
.
Combination of seeding technique and manipulating the
profile of supersaturation-generating variables increases the
likelihood of removing inconsistency from final PSD. Mullin
(2001) proposed an expression to determine the seed
mass-size couple which has the following equivalent form:
LS
LP
3
¼MS
MP
(5)
where LPand M
P
stand for the final average product size and
mass (the product mass consists of two parts: seed mass and
the mass of solute that has deposited on seed surfaces)
respectively. This expression is derived from a mass balance
of the crystallization batch by assuming the number of product
crystals to be equal to that of seeds and that secondary
nucleation, agglomeration and attrition are negligible. It
should be noted that LSand LPare number-mean size,
defined as the first moment of population density function. In
conjunction with programmed cooling, it is assumed that
nucleation is subdued to such a degree that almost all solute
molecules deposit on the seed surfaces, and the product
size is then related to seed mass-size couple by equation
(5). Genck (2000) gave two examples of the application of
this methodology to cooling and evaporative crystallizations.
Yu et al. (2006b) examined the applicability of equation (5)
to the anti-solvent crystallization of paracetamol– acetone–
water system. Despite the tendency of agglomeration (Yu
et al., 2005), when coupled with supersaturation control,
equation (5) was found to be a good starting point for the
refinement of particle size distribution and adjustment of
batch time in different circumstances to obtain a target average
crystal size with narrow size distribution. Figure 6 shows the
PSDs of products using equation 5 as first estimate for seed
mass-size couple and when the seed loading was doubled.
By simply doubling the seed loading, the fines fraction has sig-
nificantly reduced to a narrower and unimodal distribution.
Seed loading and size was also studied through crystalliza-
tion modelling and optimisation. Chung et al. (1999) and
Choong and Smith (2004) included seed properties as the
optimised variables along with temperature profile in the for-
mulation of optimization problems. The relationship between
the properties of seeds and products is complicated by
secondary nucleation whose rate is usually a function of the
third moment and/or average size of the particles in the slurry.
‘Internal seeding’
Despite the apparent advantages of using seeds, seeding
may not be feasible all the time due to operational or safety
concerns, e.g., unavailability of ports for addition of seeds,
potential additional hazards associated with operator’s
exposure to API and solvent, dust explosion and so on.
In contrast to seeded systems in which the amount of
seeds added is specific, the initial nuclei formed by primary
nucleation in unseeded systems are random and irreproduci-
ble for different runs. Even with exactly the same initial con-
ditions and cooling rate in approaching nucleation, primary
nucleation gives different number of nuclei; hence product
consistency cannot be guaranteed for every run (Chew
et al., 2007). This therefore motivates a means to manipulate
the nuclei generated by primary nucleation in unseeded sys-
tems to achieve consistent nuclei from primary nucleation in
different runs, which thereby provides a viable alternative to
external seeding. The FBRM-based technique developed
by Chew (2006) can be viewed as a reliable means of
‘internal seeding’, as it involves the automatic detection
of nucleation followed by feedback monitoring and tuning of
the primary nuclei formed to achieve a desired distribution
of seed sizes. Results show that the batch-to-batch consist-
ency achieved using this new technique was comparable to
or even surpassed that with traditional external seeding.
Hence, ‘internal seeding’ with the aid of PAT may be a
viable industrial alternative in view of the added complexities
and uncertainties of external seeding.
POLYMORPH MONITORING AND CONTROL
Crystallization of a desired solid form from a solute-solvent
system exhibiting polymorphism is one of the most important
considerations in crystallization process design and control.
Polymorphism has profound influences on the properties of
the API and dosage form such as bioavailability and stability.
Recent high profile cases involving pharmaceutical products
in which the unexpected appearance of a second poly-
morphic form resulted in withdrawal of the products, have
highlighted the importance of better understanding of the
whole area of polymorph formation, prediction, transform-
ation and stability. Here we adopt a broad definition of
Figure 6. PSD of product crystals obtained from anti-solvent crystal-
lization seeded with seeds of size fraction 212 –250 mm(Yuet al.,
2006b).
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
RECENT ADVANCES IN CRYSTALLIZATION CONTROL 899
polymorph to include hydrate and solvates (also known as
pseudo-polymorphs) because their impacts on crystallization
design and control are the same.
Thermodynamic stability alone is not sufficient to ensure
that the stable polymorph will always be produced (Saranteas
et al., 2005). The unstable polymorph or psuedopolymorph
may crystallize out first during primary nucleation in an
unseeded system in accordance with the Oswald’s rule of
stages (Threlfall, 2003). Often it is the transformation kinetics
between the metastable and stable forms that govern the
final isolatable form. Solvent-mediated transformation has
been considered the main mechanism by which polymorph
transformation takes place during crystallization (Beckmann,
2000). Cardew and Davey (1985) and Davey et al. (1986)
expressed solvent-mediated transformation as a two-step
process in which the metastable form is first dissolved,
followed by the crystallization of the stable form. An under-
standing of the controlling step in such polymorphic trans-
formations is crucial for the design of the crystallization
process to obtain a single desired polymorph. This has motiv-
ated several recent studies on polymorphic transformation
(Brittain, 2004; Dharmayat et al., 2006; Gu et al., 2001; Hu
et al., 2005; Kitamura and Sugimoto, 2003; Mukuta et al.,
2005; Murphy et al., 2002; O’Brien et al., 2004; O’Sullivan
et al., 2003; O’Sullivan and Glennon, 2005; Ono et al.,
2004a, b; Qu et al., 2006; Roelands et al., 2006; Saranteas
et al., 2005; Scho
¨ll et al., 2006; Skrdla et al., 2001; Starbuck
et al., 2002; Tian et al., 2006; Wang et al., 2000).
In-Situ Monitoring of Polymorphs
Traditionally, offline analytical techniques such as powder
X-ray diffraction (PXRD), differential scanning calorimetry
(DSC), solid state NMR and infrared spectroscopy (IR)
have been used to characterize polymorphs. However, due
to the dynamic nature of polymorphic transformation and
the instability of certain polymorphs, real time monitoring of
polymorphs would be advantageous. During process devel-
opment, in-situ monitoring provides kinetic data of the pro-
cess that would enable a robust process to be designed to
produce consistently the right crystal form. During production,
the ability to identify and quantify undesirable polymorphs in
real time would enable the appropriate remedial action to be
taken, e.g., by increasing batch time to allow complete con-
version of metastable form to the stable form. Recently
Raman spectroscopy has been successfully applied to moni-
tor the polymorphic transformation in situ (Agarwal and
Berglund, 2003; Falcon and Berglund, 2003; Hu et al.,
2005; O’Brien et al., 2004; O’Sullivan et al., 2003; Ono
et al., 2004b; Qu et al., 2006; Scho
¨ll et al., 2006; Starbuck
et al., 2002; Wang et al., 2000). Falcon and Berglund
(2003) demonstrated the versatility of Raman spectroscopy
in simultaneous monitoring of solute concentration as well
as polymorphic form during the anti-solvent crystallization of
cortisone acetate. In this case, the on-line Raman measure-
ments were used quantitatively for solute concentration
measurement but only qualitatively for polymorph monitoring.
Despite the success of a few research papers in using
Raman spectroscopy for quantitative analysis of polymorphic
content, calibration of Raman signals for analysis of solid
form in a slurry is still a challenging task because the vari-
ation in spectra depends not only on the amount of poly-
morphic form present but also on the particle size. As
particle size distribution is constantly evolving during crystal-
lization, accurate calibration of Raman spectra with respect to
the relative polymorphic content is difficult to achieve. There-
fore, O’Sullivan et al. (2003) cautioned that the technique
may only be useful in a qualitative sense, for example to
detect the existence of different polymorphs in the crystallizer,
unless corrections for particle size effects were considered.
As polymorphic changes are often accompanied by a
change in morphology, alternative techniques such as
FBRM and online video microscopy have been used to moni-
tor polymorphic transformations. Relying on the distinct mor-
phological difference between the
d
and ß polymorphs,
O’Sullivan and Glennon (2005) successfully monitored the
polymorphic transformation of D-mannitol in aqueous solution
using FBRM and in situ ATR-FTIR. Their results identified the
transformation mechanism to be solvent-mediated transform-
ation instead of solid-state transition. Dharmayat et al. (2006)
and De Anda et al. (2005a, c) studied polymorphic transform-
ation of L-glutamic acid using in-process image analysis. The
volume fraction of each polymorph was derived from the real
time images and the crystal growth dynamics and related
polymorphic phase transitions in the batch cooling crystalliza-
tion process could be obtained.
Control of Process to Consistently Produce the
Desired Polymorph
Unlike the control of PSD described earlier where PSD
information acquired by FBRM can be used as input signal
for feedback control, direct real time feedback of polymorphic
information measured by the various inline techniques to
crystallization control has proved difficult to implement.
Instead, the general approach is to design the process
based on information on thermodynamics and transformation
kinetics (i.e., the first-principle method of control), usually
combined with seeding to ensure that the desired poly-
morphic form is obtained as the final product. There are
several reported studies on the determination of operating
conditions (solvent type, temperature range, cooling rate,
seeding strategy, additives) for the selective crystallization
of the desired polymorph. Beckmann (2000) has provided a
comprehensive guide into how polymorphism can be con-
trolled via seeding. Muller et al. (2006) and Saranteas et al.
(2005) presented examples of process design and scale-up
methodology to ensure the manufacture of the desired
polymorph. Once the operating conditions or constraints
(solubility and metastable limit of each polymorph) are estab-
lished, control method can be implemented to guide the
temperature or anti-solvent addition trajectory. A recent
paper by Kee et al. (2006) described the application of feed-
back concentration control using ATR-FTIR to selectively
crystallize the metastable
a
-form of L-glutamic acid. A temp-
erature range for seeding was carefully chosen to avoid the
nucleation and growth of
b
crystals and a suitable supersa-
turation value was selected as the control setpoint to
ensure that the solution concentration stayed within the meta-
stable limit and minimize secondary nucleation. Figure 7
shows that
a
-crystals of fairly uniform size were obtained
with minimal secondary nucleation or agglomeration.
A few attempts have been made recently to develop popu-
lation balance based process models that take into account
solvent-mediated transformation kinetics of polymorphs
(Ma et al., 2006; Ono et al., 2004a; Roelands et al., 2006;
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
900 YU et al.
Scho
¨ll et al., 2006). Ono et al. (2004a) combined PBE with
kinetic expressions for the polymorphic transformation that
include growth and secondary nucleation rates of
b
-form,
and dissolution rate of
a
-form of L-glutamic acid. Using the
process model developed, process simulation data (e.g.,
separate CSD information of the different polymorphs)
which is otherwise difficult or impossible to measure can be
obtained. Scho
¨ll et al. (2006) incorporated more complex kin-
etic expressions into their process model of crystallization of
L-glutamic acid, e.g., heterogeneous and surface nucleation
rates. Their simulation results showed that the overall trans-
formation rate depends mainly on the available surface of
a
crystals. Roelands et al. (2006) simulated the overall effect
of competitive nucleation and growth rates of polymorphs
during a batch anti-solvent crystallization of L-histidine. Ma
et al. (2006) developed a multi-dimensional population bal-
ance model for the crystallization of
b
-form of L-glutamic
acid that incorporates real-time crystal shape measurement
via an in-process imaging technique. Mazzotti et al. (2006)
presented a population balance model that accounts for the
nucleation, growth and agglomeration for a pH shift precipi-
tation of L-glutamic acid. Despite the computational complex-
ity and the difficulties in estimating rate parameters
accurately, such modelling efforts are still valuable as they
provide a deeper understanding of the transformation
process and the influences of different operating parameters,
and thereby could facilitate more robust control of the
process.
PARTICLE MORPHOLOGY MONITORING
AND CONTROL
Particle morphology or habit is an important property that
affects not only the downstream processing and handling
but also the end-use functional properties. Crystal habit is
closely associated with filterability, flowability and compaction
behaviour amongst other properties. A sudden change in
crystal habit may suggest the appearance of a new poly-
morph or the presence of trace impurities. Agglomeration
may lead to profound changes in final product quality, as
well as entrapment of unacceptable levels of occluded
solvent, causing difficulties in washing and drying.
Particle morphology has been traditionally analysed by
means of off-line microscopy. The availability of in-situ
video camera systems such as the Particle Vision and
Measurement system (PVM) developed by Lasentec and
the online high-speed imaging system developed by GlaxoS-
mithKline (de Anda et al., 2005–c; Dharmayat et al., 2006)
offer the opportunity for real-time monitoring as well as
control of shapes and sizes of crystals during crystallization
process. Until recently, information from in-line and on-line
imaging systems has only been used in a qualitative
manner to complement data from other sensors such as
FBRM and Raman spectroscopy. Quantitative analysis of
images acquired online from crystalliser has not been suc-
cessful using commercial image analysis software mainly
due to the inherent poor quality of the images, which often
are out-of-focus, poorly lit and contain overlapping particles.
Image Segmentation and Morphological
Quantification
The first step towards quantitative analysis of images is
image segmentation in which the particles are identified
and extracted from the image background. Various image
segmentation techniques have been reported recently, as
reviewed by De Anda et al. (2005b). As mentioned before,
online images taken in crystallizer containing slurries with
particles suspended in a solution are particularly challenging
to analyse because of out-of-focus and overlapping particles
and uneven background intensity. De Anda et al. (2005b)
presented a multi-scale Canny method to segment in situ
images of varied background pixel intensity resulting from
the light effect and temporal changes of hydrodynamics
within the crystallizer. Larsen et al. (2006) developed an
algorithm to deal with images of high-aspect-ratio particles
captured in moderately dense suspensions.
After segmentation, one straightforward way to quantify
particle morphology is to use simple descriptors such as
aspect ratio, parameter ratio, robustness, concavity index,
heterogeneity, fractal dimension, and so on (Belaroui et al.,
2002; BernardMichel et al., 1997; Hentschel and Page,
2003). However, any single simple descriptor describes
only one global feature of particle morphology and similar
values may be obtained for particles of visually quite different
shapes. The lack of mapping uniqueness has prompted the
simultaneous use of multiple simple descriptors to achieve
a more comprehensive and discriminating description of par-
ticle morphology. BernardMichel et al. (1997) used a set of
seven simple descriptors to classify general particle shape.
A
˚lander et al. (2004); Faria et al. (2003) quantified agglom-
eration degree of crystals on the basis of seven and six
simple descriptors respectively.
A more structured method for shape representation is to
use boundary Fourier descriptors which are a series of coef-
ficients obtained by applying one-dimensional Fourier trans-
form on the shape signature function of particle images.
Resolution of description can be adjusted according to
requirements: coarse shape features can be captured by
lower order coefficients and finer features by higher order
coefficients. Such a structure affords robustness, compact-
ness and computational efficiency. Its application in classifi-
cation of particle shape has been demonstrated in several
recent studies (BernardMichel et al., 1997; De Anda et al.,
2005b; Raj and Cannon, 1999). Yu et al. (2006c) derived
Figure 7. Microscopy image of
a
-form of L-glutamic acid crystals
obtained using feedback concentration control (Kee et al., 2006).
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
RECENT ADVANCES IN CRYSTALLIZATION CONTROL 901
another series of coefficients by applying two-dimensional
Fourier transform to particle images, which was proved to
be complementary to the boundary Fourier descriptors in
morphological representation.
Use of in Situ Images in Feedback Control and
Process Monitoring
Patience and Rawlings (2001) developed a feedback con-
trol scheme for cooling crystallization wherein quantitative
information of crystal habit was extracted from in situ particle
images. The concentration of an additive used to modify crys-
tal habit was adjusted in response to changes in crystal habit.
As knowledge of interactions between crystal faces and addi-
tive molecules is improved (Davey et al., 2002; Poornachary
et al., 2007; Scott and Black, 2005; Weissbuch et al., 2003),
‘spiking’ the crystallization medium with trace amounts of
approved impurities to engineer crystal habit may become
a common manipulated variable in industrial control systems.
De Anda et al. (2005a) employed a multi-scale Canny
method to segment in situ crystal images and boundary Four-
ier transform to characterize crystal habit. Their monitoring
system successfully followed the polymorphic transformation
process of L-glutamic acid where the polymorphs concerned
exhibit distinctive characteristic habits. The monitoring
system can also be used to extract two-dimensional crystal
growth kinetic parameters (Ma et al., 2006; Wang, 2006).
CHALLENGES AND FUTURE DEVELOPMENTS
The FDA’s PAT initiative and QbD approach have motiv-
ated the use of the latest generation of PAT to improve and
ensure quality and consistency of crystallization products.
Recent research has shown that several techniques are
very promising for the monitoring and control of crystallization
processes. These include Lasentec FBRM and PVM, ATR-
FTIR and Raman spectroscopy. In laboratory studies, feed-
back control strategies based on these sensors exhibit
more robustness than traditional open-loop strategies. The
challenge remains to implement these methods successfully
and ubiquitously in pharmaceutical and fine chemicals indus-
tries. Some of the hurdles to be overcome include calibration
of ATR-FTIR in the presence of multiple impurities, decou-
pling PSD effect from polymorphic content for quantitative
use of Raman spectra, correlation of FBRM data with process
performance of crystals and segmentation of low-quality on-
line images of particle. Resolution of these issues will
expedite the transfer of advanced control strategies from
academia to industry, advancing quality control towards
six-Sigma in pharmaceutical and fine chemical production.
Advanced on-line sensors help to improve the understand-
ing of crystallization process, thereby enabling more efficient
and cost-effective process design and control. More work
remains to be done to improve the understanding of the work-
ing principles of each PAT system in order to exploit the full
potential of each tool. If the relationship between CLD and
PSD can be better established, it may be possible to use
FBRM data more precisely for feedback control. If the particle
size effect on Raman signals can be decoupled from poly-
morphic purity, one can envision the use of Raman signals
quantitatively for optimal process design and eventually
real-time feedback control instead of simply using the tech-
nique to identify the end-point of crystallization processes.
The vast amount of data collected by PAT systems
implemented online has posed a practical data management
problem. Efficient data analysis and mining methods, such as
chemometrics, are required to identify the crucial information
from the data acquired.
Operation protocols developed in laboratory scale crystal-
lizers during process development are expected to reproduce
crystal quality (at least in a number of key solid-state attri-
butes) on production scale. However, crystal quality often suf-
fers in larger scales due to great inhomogeneities in space
and time. A new trend in the scale-up study relies on process
modelling which integrates computational fluid dynamics
(CFD) and crystallization simulation (Woo et al., 2006;
Zauner and Jones, 2000). Some of the parameters in the
crystallization model need to be calculated from detailed
information of flow field and turbulent energy dissipation
rate which can be provided by CFD. Crystallization kinetics
are then derived from smaller scale experimental data and
transferred directly to larger scale. Changes in mixing pro-
cesses and thus the scale-up effects on crystallization are
accounted for by the model parameters determined using
CFD. Simulation results offer valuable insights for determin-
ing optimal operating conditions and equipment parameters
on scale. Meanwhile, improvement in crystallization model-
ling will enhance the validity of these model-based
approaches. For example, crystal breakage resulting from
crystal–crystal, wall –crystal and crystal– impeller collisions
may be a serious concern in industrial-scale crystallizers.
Since the relative collision area of crystallizer wall and impel-
ler changes with scale, breakage due to crystal –wall and
crystal impeller will change along with breakage due to
crystal– crystal collisions. New and more complex modelling
will be required to address these factors at multiple size-scales.
NOMENCLATURE
a,bcoefficients in equation (2)
csaturation concentration of solute, g/g solvent
tany time during the process, min
Gmet maximum allowable growth rate within metastable zone,
ms
21
Lcrystal size, mm
Lnumber-average size, mm
Mmass, g
Ttemperature, 8C
T
†
cooling rate, 8Cs
21
DTmet the maximum allowable undercooling, 8C
w
v,
w
svolume and surface shape factors of crystals
t
the overall batch time, [s]
Subscripts
sol solvent
S, P seed and product respectively
I, F initial and final condition respectively
REFERENCES
Agarwal, P. and Berglund, K.A., 2003, In situ monitoring of calcium
carbonate polymorphs during batch crystallization in the presence
of polymeric additives using Raman spectroscopy, Crystal Growth &
Design, 3(6): 941–946.
A
˚lander, E.M., Uusi-Penttila, M.S. and Rasmuson, A.C., 2004,
Agglomeration of paracetamol during crystallization in pure and
mixed solvents, Ind Eng Chem Res, 43(2): 629–637.
Barrett, P. and Glennon, B., 1999, In-line FBRM monitoring of particle
size in dilute agitated suspensions, Particle & Particle Systems
Characterization, 16(5): 207– 211.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
902 YU et al.
Barrett, P. and Glennon, B., 2002, Characterizing the metastable
zone width and solubility curve using Lasentec FBRM and PVM,
Chem Eng Res Des, 80(A7): 799–805.
Barthe, S. and Rousseau, R.W., 2006, Utilization of focused
beam reflectance measurement in the control of crystal size distri-
bution in a batch cooled crystallizer, Chem Eng Technol, 29(2):
206– 211.
Beckmann, W., 2000, Seeding the desired polymorph: background,
possibilities, limitations, and case studies, Organic Process
Research & Development, 4(5): 372– 383.
BernardMichel, B., Rohani, S., Pons, M.N., Vivier, H. and Hundal,
H.S., 1997, Classification of crystal shape using Fourier descrip-
tors and mathematical morphology, Particle & Particle Systems
Characterization, 14(4): 193– 200.
Belaroui, K., Pons, M.N. and Vivier, H., 2002, Morphological charac-
terisation of gibbsite and alumina, Powder Technol, 127(3):
246– 256.
Bloemen, H.H.J. and De Kroon, M.G.M., 2005, Transformation of
chord length distributions into particle size distributions using
least squares techniques, Particulate Science and Technology,
23(4): 377– 386.
Bordui, P.F., Herko, S.P. and Kostecky, G., 1985, Computer-
controlled crystal growth of KDP from aqueous solution, Journal
of Crystal Growth, 72(3): 756.
Braatz, R.D., 2002, Advanced control of crystallization processes,
Annual Reviews in Control, 26(1): 87–99.
Brittain, H.G., 2004, Fluorescence studies of the transformation of
carbamazepine anhydrate form III to its dihydrate phase, Journal
of Pharmaceutical Sciences, 93(2): 375– 383.
Cardew, P.T. and Davey, R., 1985, The kinetics of solvent-mediated
phase-transformations, Proceedings of the Royal Society of
London, Series A, 398, 415–428.
Chew, J.W., 2006, Experimental investigation of the applicability of
FBRM in the control of batch cooling crystallization, MEng thesis,
Department of Chemical and Biomolecular Engineering, National
University of Singapore, Singapore.
Chew, J.W., Black, S.N., Chow, P.S. and Tan, R.B.H., 2007, A com-
parison between open-loop temperature-control and closed-loop
supersaturation-control for cooling crystallization of glycine, Ind
Eng Chem Res, 46(3): 830– 838.
Choong, K.L. and Smith, R., 2004, Novel strategies for optimisation
of batch, semi-batch and heating/cooling evaporative crystalliza-
tion, Chem Eng Sci, 59(2): 329– 343.
Chung, S.H., Ma, D.L. and Braatz, R.D., 1999, Optimal seeding in
batch crystallization, Canadian Journal of Chemical Engineering,
77(3): 590– 596.
Costa, C.B.B., da Costa, A.C. and Maciel, R., 2005, Mathematical
modelling and optimal control strategy development for an adipic
acid crystallization process, Chemical Engineering and Proces-
sing, 44(7): 737–753.
Davey, R.J., Cardew, P.T., Mcewan, D. and Sadler, D.E., 1986, Rate
controlling processes in solvent-mediated phase-transformations,
Journal of Crystal Growth, 79(1–3): 648 –653.
Davey, R.J., Allen, K., Blagden, N., Cross, W.I., Lieberman, H.F.,
Quayle, M.J., Righini, S., Seton, L. and Tiddy, G.J.T., 2002,
Crystal engineering-nucleation, the key step, CrystEngComm,
257– 264.
De Anda, J.C., Wang, X.Z., Lai, X. and Roberts, K.J., 2005a, Classi-
fying organic crystals via in-process image analysis and the use of
monitoring charts to follow polymorphic and morphological
changes, Journal of Process Control, 15(7): 785 –797.
De Anda, J.C., Wang, X.Z. and Roberts, K.J., 2005b, Multi-scale
segmentation image analysis for the in-process monitoring of
particle shape with batch crystallisers, Chem Eng Sci, 60(4):
1053–1065.
De Anda, J.C., Wang, X.Z., Lai, X., Roberts, K.J., Jennings, K.H.,
Wilkinson, M.J., Watson, D. and Roberts, D., 2005c, Real-time
product morphology monitoring in crystallization using imaging
technique, AIChE J, 51(5): 1406–1414.
De Clercq, B., Lant, P.A. and Vanrolleghem, P.A., 2004, Focused
beam reflectance technique for in situ particle sizing in wastewater
treatment settling tanks, Journal of Chemical Technology and
Biotechnology, 79(6): 610–618.
Dharmayat, S., De Anda, J.C., Hammond, R.B., Lai, X.J., Roberts,
K.J. and Wang, X.Z., 2006, Polymorphic transformation of L-gluta-
mic acid monitored using combined on-line video microscopy and
x-ray diffraction, Journal of Crystal Growth, 294(1): 35– 40.
Doki, N., Seki, H., Takano, K., Asatani, H., Yokota, M. and Kubota, N.,
2004, Process control of seeded batch cooling crystallization of the
metastable alpha-form glycine using an in-situ ATR-FTIR spec-
trometer and an in-situ FBRM particle counter, Crystal Growth &
Design, 4(5): 949–953.
Dunuwila, D.D., Carroll, L.B. and Berglund, K.A., 1994, An investi-
gation of the applicability of attenuated total-reflection infrared-
spectroscopy for measurement of solubility and supersaturation
of aqueous citric-acid solutions, Journal of Crystal Growth,
137(3–4): 561–568.
Falcon, J.A. and Berglund, K.A., 2003, Monitoring of antisolvent
addition crystallization with Raman spectroscopy, Crystal Growth
& Design, 3(6): 947–952.
Faria, N., Pons, M.N., de Azevedo, S.F., Rocha, F.A. and Vivier, H.,
2003, Quantification of the morphology of sucrose crystals by
image analysis, Powder Technol, 133(1–3): 54– 67.
Feng, L.L. and Berglund, K.A., 2002, ATR-FTIR for determining
optimal cooling curves for batch crystallization of succinic acid,
Crystal Growth & Design, 2(5): 449–452.
Fujiwara, M., Chow, P.S., Ma, D.L. and Braatz, R.D., 2002, Paraceta-
mol crystallization using laser backscattering and ATR-FTIR
spectroscopy: Metastability, agglomeration, and control, Crystal
Growth & Design, 2(5): 363–370.
Fujiwara, M., Nagy, Z.K., Chew, J.W. and Braatz, R.D., 2005,
First-principles and direct design approaches for the control of
pharmaceutical crystallization, Journal of Process Control, 15(5):
493– 504.
Gabas, N. and Laguerie, C., 1992, Batch crystallization of D-xylose
by programmed cooling or by programmed adding of ethanol,
Chem Eng Sci, 47(12): 3148– 3151.
Genck, W.J., 2000, Better growth in batch crystallisers, Chemical
Engineering, 107(8): 90–95.
Gro
¨n, H., Borissova, A. and Roberts, K.J., 2003, In-process ATR-
FTIR spectroscopy for closed-loop supersaturation control of a
batch crystalliser producing monosodium glutamate crystals of
defined size, Ind & Engi Chem Res, 42(1): 198 –206.
Gu, C.H., Young, V. and Grant, D.J.W., 2001, Polymorph screening:
Influence of solvents on the rate of solvent-mediated polymorphic
transformation, Journal of Pharmaceutical Sciences, 90(11):
1878–1890.
Gutwald, T. and Mersmann, A., 1990, Batch cooling crystallization at
constant supersaturation: Technique and experimental
results, Chemical Engineering & Technology-CET, 13(1):
229– 237.
Gutwald, T. and Mersmann, A., 1994, Evaluation of kinetic-par-
ameters of crystallization from batch and continuous experiments,
Separations Technology, 4(1): 2– 14.
Heath, A.R., Fawell, P.D., Bahri, P.A. and Swift, J.D., 2002, Estimat-
ing average particle size by focused beam reflectance measure-
ment (FBRM), Particle & Particle Systems Characterization,
19(2): 84–95.
Hentschel, M.L. and Page, N.W., 2003, Selection of descriptors for
particle shape characterization, Particle & Particle Systems
Characterization, 20(1): 25– 38.
Hu, Q., Rohani, S., Wang, D.X. and Jutan, A., 2005, Optimal control
of a batch cooling seeded crystalliser, Powder Technol, 156(2 –3):
170– 176.
Hukkanen, E.J. and Braatz, R.D., 2003, Measurement of particle
size distribution in suspension polymerization using in situ laser
backscattering, Sensors and Actuators B-Chemical, 96(1– 2):
451– 459.
Hussain, A.S., 2006, Quality by Design (QbD) – integration of prior
knowledge and pharmaceutical development into CMC submission
and review, AAPS workshop on Pharmaceutical Quality
Assessment – A Science and Risk-based CMC Approach in the
21st Century, North Bethesda, Maryland, USA.
Jones, A.G., 1974, Optimal operation of a batch cooling crystalliser,
Chem Eng Sci, 29(5): 1075–1087.
Jones, A.G. and Mullin, J.W., 1974, Programmed cooling
crystallization of potassium sulphate solutions, Chem Eng Sci,
29(1): 105.
Jones, A.G. and Teodossiev, N.M., 1988, Microcomputer program-
ming of dosage rate during batch precipitation, Crystal Research
and Technology, 23(8): 957– 966.
Karpinski, P.H. and Wey, J.S., 2002, Precipitation processes, in
Myerson, A.S. (ed.). Handbook of Industrial Crystallization, 151
(Butterworth Heinemann, Boston, USA).
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
RECENT ADVANCES IN CRYSTALLIZATION CONTROL 903
Kee, N.C.S., Tan, R.B.H. and Braatz, R.D., 2006, Selective
crystallization of metastable
a
-form of L-glutamic acid through
feedback concentration control, AIChE Annual Meeting, San
Francisco, USA.
Kim, S.J., Wei, C.K. and Kiang, S., 2003, Crystallisation process
development of an active pharmaceutical ingredient and particle
engineering via the use of ultrasonics and temperature cycling,
Organic Process Research & Development, 7(6): 997–1001.
Kim, S., Lotz, B., Lindrud, M., Girard, K., Moore, T., Nagarajan, K.,
Alvarez, M., Lee, T., Nikfar, F., Davidovich, M., Srivastava, S.
and Kiang, S., 2005, Control of the particle properties of a drug
substance by crystallization engineering and the effect on drug
product formulation, Organic Process Research & Development,
9(6): 894– 901.
Kitamura, M. and Sugimoto, M., 2003, Anti-solvent crystallization and
transformation of thiazole-derivative polymorphs—I: Effect of
addition rate and initial concentrations, Journal of Crystal Growth,
257(1–2): 177–184.
Kline, B.J., Saenz, J., Stankovic, N. and Mitchell, M.B., 2006,
Polymorph and particle size control of ppar compounds
pf00287586 and ag035029, Organic Process Research & Devel-
opment, 10(2): 203–211.
Kubota, N., Doki, N., Yokota, M. and Sato, A., 2001, Seeding policy
in batch cooling crystallization, Powder Technol, 121(1): 31–38.
Lafferre
`re, L., Hoff, C. and Veesler, S., 2004, In situ monitoring of the
impact of liquid-liquid phase separation on drug crystallization by
seeding, Crystal Growth & Design, 4(6): 1175– 1180.
Larsen, P.A., Rawlings, J.B. and Ferrier, N.J., 2006, An algorithm for
analyzing noisy, in situ images of high-aspect-ratio crystals to
monitor particle size distribution, Chem Eng Sci, 61(16):
5236–5248.
Leuenberger, H. and Lanz, M., 2005, Pharmaceutical powder tech-
nology—from art to science: The challenge of the FDA’s process
analytical technology initiative, Advanced Powder Technology,
16(1): 3–25.
Lewiner, F., Fevotte, G., Klein, J.P. and Puel, F., 2001, Improving
batch cooling seeded crystallization of an organic weed-killer
using on-line ATR FTIR measurement of supersaturation, Journal
of Crystal Growth, 226(2– 3): 348–362.
Li, M.Z. and Wilkinson, D., 2005, Determination of non-spherical
particle size distribution from chord length measurements. Part 1:
Theoretical analysis, Chem Eng Sci, 60(12): 3251–3265.
Liotta, V. and Sabesan, V., 2004, Monitoring and feedback control of
supersaturation using ATR-FTIR to produce an active pharma-
ceutical ingredient of a desired crystal size, Organic Process
Research & Development, 8(3): 488– 494.
Lung-Somarriba, B.L.M., Moscosa-Santillan, M., Porte, C. and Delacroix,
A., 2004, Effect of seeded surface area on crystal size distribution in
glycine batch cooling crystallization: A seeding methodology, Journal
of Crystal Growth, 270(3–4): 624 –632.
Ma, D.L., Chung, S.H. and Braatz, R.D., 1999, Worst-case perform-
ance analysis of optimal batch control trajectories, AIChE J, 45(7):
1469–1476.
Ma, D.L., Tafti, D.K. and Braatz, R.D., 2002, Optimal control and
simulation of multidimensional crystallization processes, Compu-
ters & Chemical Engineering, 26(7– 8): 1103– 1116.
Ma, C.Y., Wang, X.Z., de Anda, J.C. and Roberts, K.J., 2006,
Multi-dimensional population balance modelling incorporated with
real-time crystal shape measurement via in-process imaging,
Control of Particulate Processes VII, Harrison Hot Springs, BC,
Canada.
Matthews, H.B., Miller, S.M. and Rawlings, J.B., 1996, Model identi-
fication for crystallization: Theory and experimental verification,
Powder Technol, 88(3): 227 –235.
Mazzotti, M., Vicum, L., Lindenberg, C., Scho
¨ll, J. and Brozio, J.,
2006, The kinetics of L-glutamic acid precipitation: Characteris-
ation and modelling, AIChE Annual Meeting, San Francisco,
USA.
McKenzie, P., Kiang, S., Tom, J., Rubin, A.E. and Futran, M., 2006,
Can pharmaceutical process development become high tech?
AIChE J, 52(12): 3990–3994.
Mukuta, T., Lee, A.Y., Kawakami, T. and Myerson, A.S., 2005, Influ-
ence of impurities on the solution-mediated phase transformation
of an active pharmaceutical ingredient, Crystal Growth & Design,
5(4): 1429–1436.
Muller, M., Meier, U., Wieckhusen, D., Beck, R., Pfeffer-Hennig, S.
and Schneeberger, R., 2006, Process development strategy to
ascertain reproducible API polymorph manufacture, Crystal
Growth & Design, 6(4): 946–954.
Mullin, J.W., 2001, Crystallisation (Butterworth-Heinemann,
Oxford, UK).
Mullin, J.W. and Ny
ˇvlt, J., 1971, Programmed cooling of batch crystal-
lisers, Chem Eng Sci, 26(3): 369–377.
Murphy, D., Rodriguez-Cintron, F., Langevin, B., Kelly, R.C. and
Rodriguez-Hornedo, N., 2002, Solution-mediated phase transform-
ation of anhydrous to dihydrate carbamazepine and the effect of
lattice disorder, International Journal of Pharmaceutics, 246
(1–2): 121–134.
Nagy, Z.K. and Braatz, R.D., 2004, Open-loop and closed-loop
robust optimal control of batch processes using distributional
and worst-case analysis, Journal of Process Control, 14(4):
411–422.
O’Brien, L.E., Timmins, P., Williams, A.C. and York, P., 2004, Use of
in situ FT-Raman spectroscopy to study the kinetics of the trans-
formation of carbamazepine polymorphs, Journal of Pharma-
ceutical and Biomedical Analysis, 36(2): 335–340.
O’Sullivan, B., Barrett, P., Hsiao, G., Carr, A. and Glennon, B., 2003,
In situ monitoring of polymorphic transitions, Organic Process
Research & Development, 7(6): 977– 982.
O’Sullivan, B. and Glennon, B., 2005, Application of in situ FBRM and
ATR-FTIR to the monitoring of the polymorphic transformation of
D-mannitol, Organic Process Research & Development, 9(6):
884– 889.
Ono, T., Kramer, H.J.M., ter Horst, J.H. and Jansens, P.J., 2004a,
Process modelling of the polymorphic transformation of L-glutamic
acid, Crystal Growth & Design, 4(6): 1161–1167.
Ono, T., ter Horst, J.H. and Jansens, P.J., 2004b, Quantitative
measurement of the polymorphic transformation of L-glutamic
acid using in-situ Raman spectroscopy, Crystal Growth & Design,
4(3): 465– 469.
Patience, D.B. and Rawlings, J.B., 2001, Particle-shape
monitoring and control in crystallization processes, AIChE J,
47(9): 2125–2130.
Patience, D.B., Dell’Orco, P.C. and Rawlings, J.B., 2004, Optimal
operation of a seeded pharmaceutical crystallization with growth-
dependent dispersion, Organic Process Research & Development,
8(4): 609– 615.
Poornachary, S.K., Chow, P.S., Tan, R.B.H. and Davey, R., 2007,
Molecular speciation controlling stereoselectivity of additives:
Impact on the habit modification in alpha-glycine crystals, Crystal
Growth & Design, 7(2): 254–261.
Qu, H.Y., Louhi-Kultanen, M., Rantanen, J. and Kallas, J., 2006,
Solvent-mediated phase transformation kinetics of an anhydrate/
hydrate system, Crystal Growth & Design, 6(9): 2053–2060.
Raj, P.M. and Cannon, W.R., 1999, 2-D particle shape averaging and
comparison using Fourier descriptors, Powder Technol, 104(2):
180– 189.
Rawlings, J.B., Miller, S.M. and Witkowski, W.R., 1993, Model
identification and control of solution crystallization processes—a
review, Ind Eng Chem Res, 32(7): 1275– 1296.
Richmond, W.R., Jones, R.L. and Fawell, P.D., 1998, The relation-
ship between particle aggregation and rheology in mixed
silica-titania suspensions, Chemical Engineering Journal, 71(1):
67–75.
Roelands, C.P.M., Jiang, S.F., Kitamura, M., ter Horst, J.H., Kramer,
H.J.M. and Jansens, P.J., 2006, Antisolvent crystallization of the
polymorphs of L-histidine as a function of supersaturation ratio
and of solvent composition, Crystal Growth & Design, 6(4):
955– 963.
Saranteas, K., Bakale, R., Hong, Y.P., Luong, H., Foroughi, R. and
Wald, S., 2005, Process design and scale-up elements for
solvent mediated polymorphic controlled tecastemizole
crystallization, Organic Process Research & Development, 9(6):
911–922.
Sarkar, D., Rohani, S. and Jutan, A., 2006, Multi-objective optimis-
ation of seeded batch crystallization processes, Chem Eng Sci,
61(16): 5282– 5295.
Scho
¨ll, J., Bonalumi, D., Vicum, L., Mazzotti, M. and Muller, M.,
2006, In situ monitoring and modelling of the solvent-mediated
polymorphic transformation of L-glutamic acid, Crystal Growth &
Design, 6(4): 881–891.
Scott, C. and Black, S., 2005, In-line analysis of impurity effects on
crystallization, Organic Process Research & Development, 9(6):
890– 893.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
904 YU et al.
Skrdla, P.J., Antonucci, V., Crocker, L.S., Wenslow, R.M., Wright,
L. and Zhou, G., 2001, A simple quantitative FT-IR approach for
the study of a polymorphic transformation under crystallization
slurry conditions, Journal of Pharmaceutical and Biomedical
Analysis, 25(5– 6): 731–739.
Starbuck, C., Spartalis, A., Wai, L., Wang, J., Fernandez, P.,
Lindemann, C.M., Zhou, G.X. and Ge, Z.H., 2002, Process
optimisation of a complex pharmaceutical polymorphic system
via in situ Raman spectroscopy, Crystal Growth & Design, 2(6):
515– 522.
Tadayyon, A. and Rohani, S., 2000, Control of fines suspension
density in the fines loop of a continuous KCl crystalliser using
transmittance measurement and an FBRM (R) probe, Canadian
Journal of Chemical Engineering, 78(4): 663– 673.
Tavare, N.S., 1995, Industrial Crystallization, 112– 118 (Plenum
Press, New York, USA).
Threlfall, T., 2003, Structural and thermodynamic explanations of
Ostwald’s rule, Organic Process Research & Development, 7(6):
1017–1027.
Tian, F., Zeitler, J.A., Strachan, C.J., Saville, D.J., Gordon, K.C. and
Rades, T., 2006, Characterizing the conversion kinetics of carba-
mazepine polymorphs to the dihydrate in aqueous suspension
using Raman spectroscopy, Journal of Pharmaceutical and Biome-
dical Analysis, 40(2): 271–280.
Togkalidou, T., Tung, H.H., Sun, Y.K., Andrews, A. and Braatz, R.D.,
2002, Solution concentration prediction for pharmaceutical
crystallization processes using robust chemometrics and ATR
FTIR spectroscopy, Organic Process Research & Development,
6(3): 317– 322.
Togkalidou, T., Tung, H.H., Sun, Y., Andrews, A.T. and Braatz, R.D.,
2004, Parameter estimation and optimisation of a loosely bound
aggregating pharmaceutical crystallization using in situ infrared
and laser backscattering measurements, Ind Eng Chem Res,
43(19): 6168–6181.
Wang, X.Z., 2006, Shape—the final frontier, Control of Particulate
Processes VII, Harrison Hot Springs, BC, Canada.
Wang, F., Wachter, J.A., Antosz, F.J. and Berglund, K.A., 2000, An
investigation of solvent mediated polymorphic transformation of
progesterone using in situ Raman spectroscopy, Organic Process
Research & Development, 4(5): 391– 395.
Ward, J.D., Mellichamp, D.A. and Doherty, M.F., 2006, Choosing an
operating policy for seeded batch crystallization, AIChE J, 52(6):
2046–2054.
Weissbuch, I., Lahav, M. and Leiserowitz, L., 2003, Toward
stereochemical control, monitoring, and understanding of crystal
nucleation, Crystal Growth & Design, 3(2): 125–150.
Wey, J.S. and Karpinski, P.H., 2002, Batch crystallization, in
Myerson, A.S. (ed.). Handbook of Industrial Crystallization, 245
(Butterworth Heinemann, Boston, USA).
Woo, X.Y., Tan, R.B.H., Chow, P.S. and Braatz, R.D., 2006,
Simulation of mixing effects in antisolvent crystallization using a
coupled CFD-PDF-PBE approach, Crystal Growth & Design,
6(6): 1291–1303.
Worlitschek, J. and Mazzotti, M., 2002, On-line monitoring of batch
cooling crystallization, Chem Eng Trans, 1, 1317–1322.
Worlitschek, J. and Mazzotti, M., 2004, Model-based optimisation of
particle size distribution in batch-cooling crystallization of paraceta-
mol, Crystal Growth & Design, 4(5): 891– 903.
Wynn, E.J.W., 2003, Relationship between particle-size and chord-
length distributions in focused beam reflectance measurement:
Stability of direct inversion and weighting, Powder Technology,
133(1–3): 125–133.
Yu, L.X., Lionberger, R.A., Raw, A.S., D’Costa, R., Wu, H.Q. and
Hussain, A.S., 2004, Applications of process analytical technology
to crystallization processes, Advanced Drug Delivery Reviews,
56(3): 349–369.
Yu, Z.Q., Tan, R.B.H. and Chow, P.S., 2005, Effects of operating
conditions on agglomeration and habit of paracetamol crystals in
anti-solvent crystallization, Journal of Crystal Growth, 279(3– 4):
477– 488.
Yu, Z.Q., Chow, P.S. and Tan, R.B.H., 2006a, Application of
attenuated total reflectance-Fourier transform infrared (ATR-
FTIR) technique in the monitoring and control of anti-solvent
crystallization, Ind Eng Chem Res, 45(1): 438– 444.
Yu, Z.Q., Chow, P.S. and Tan, R.B.H., 2006b, Seeding and constant-
supersaturation control by ATR-FTIR in anti-solvent crystallization,
Organic Process Research & Development, 10(4): 717–722.
Yu, Z.Q., Chow, P.S. and Tan, R.B.H., 2006c, Quantification of
particle morphology by boundary Fourier transform and generic
Fourier transform, Chem Eng Sci, submitted.
Zauner, R. and Jones, A.G., 2000, Mixing effects on product particle
characteristics from semi-batch crystal precipitation, Chem Eng
Res Des, 78(A6): 894–902.
Zhang, G.P. and Rohani, S., 2003, On-line optimal control of a seeded
batch cooling crystallizer, Chem Eng Sci, 58(9): 1887– 1896.
Zhou, G.X., Fujiwara, M., Woo, X.Y., Rusli, E., Tung, H.H., Starbuck,
C., Davidson, O., Ge, Z.H. and Braatz, R.D., 2006, Direct design of
pharmaceutical antisolvent crystallization through concentration
control, Crystal Growth & Design, 6(4): 892–898.
The manuscript was received 30 November 2007 and accepted for
publication after revision 9 February 2007.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A7): 893– 905
RECENT ADVANCES IN CRYSTALLIZATION CONTROL 905
