ARTICLE IN PRESS

Chemical Engineering Science 65 (2010) 3191–3198

Contents lists available at ScienceDirect

Chemical Engineering Science

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

A method of characteristics for solving population balance equations (PBE)

describing the adsorption of impurities during crystallization processes
Franc- ois Févotte a, Gilles Févotte b,c,
a
99, Avenue de Verdun, 92130 Issy les Moulineaux, France
b
Ecole des Mines de Saint Etienne, Centre SPIN, 158, Cours Fauriel, 42000 Saint Etienne, France
c
Université Lyon 1, 43, Avenue A. Einstein, 69622 Villeurbanne Cedex, France

a r t i c l e in fo abstract

Article history: Possible hindering effects of impurities on the crystal growth were shown to take place because of the
Received 25 August 2009 adsorption of impurity species on the crystal surface. Transient features of this adsorption were
Received in revised form observed, such that the growth of a given crystal does not depend on supersaturation only, but also on
29 January 2010
the time a given particle spent in contact with impurities present in the mother liquor. Meanwhile, few
Accepted 2 February 2010
kinetic models describe the effect of impurities on the growth of crystals in solution, and published
Available online 6 February 2010
models are usually derived from data obtained, thanks to specific experiments based on the evaluation
Keywords: of the growth rate of single crystals. Such models are obviously questionable because, in the industrial
Crystallization practice, distributed properties of crystals are actually involved. Considering the ‘‘time of contamina-
Dynamic simulation
tion’’ of particles as a new internal variable is thus made necessary. This is the reason why a specific PBE
Nucleation
resolution algorithm is presented in this paper. The numerical scheme for the resolution of PBEs is
Population balance
Particulate processes based on the method of characteristics and shown to allow fast and accurate simulation of transient
Impurities features of the crystal size distribution in the particular case when the growth or nucleation rates are
assumed to exhibit unsteady-state dynamics. The algorithm is finally used to simulate the isothermal
desupersaturation crystallization of citric acid in water.
& 2010 Published by Elsevier Ltd.

1. Introduction From an industrial point of view, obvious unwanted con-

sequences of impurities in crystallizing media have to be dealt
Actually, modeling and control papers in the field of crystal- with:
lization deal essentially with pure solute/solvent systems. This is
obviously a very significant limitation because industrial crystal-
 Due to final intractable residual supersaturation, the yield of
lization processes cannot avoid undesirable impurities to be
the crystallization operation is reduced because the solubility
generated during the many chemical reactions preceding the
equilibrium can no more be reached.
crystallization steps. Even minute concentrations of impurities
 Undesirable growth rates reductions require increasing the
present in the initial solution can affect tremendously the
duration of the crystallization operation, and therefore reduce
crystallization process (Sangwal, 1996, 2007; Wood, 2001;
the process productivity.
Myerson, 2001; Chernov, 2004) and induce significant reductions
 As impurities can incorporate the crystal lattice, controlling
of the growth rate (Sangwal, 1996; Kubota et al., 2000; Kubota,
the final chemical purity of the particles is obviously a major
2001).
quality issue. In the field of pharmaceutical production for
As far as the advanced dynamic modeling of crystallization
example (see e.g. Chow et al., 1985; Johnson and Li, 2007;
systems is concerned, and to the best of our knowledge, such key-
Raghavan et al., 2001), it is clear that for the sake of public
features of ‘‘real’’ industrial crystallization processes (i.e. pro-
health, the concentration of harmful impurity species in final
cesses performed in the unavoidable presence of impurities) have
active ingredients is strictly controlled by public regulatory
never been described using PBEs so that one cannot evaluate the
agencies such as the US FDA (Federal Drug Administration).
distribution and the time-variations of the detrimental effects of
impurities during crystallization processes.
In order to investigate the effect of impurities, new dynamic
 Corresponding author at: Université Lyon 1, 43, Avenue A. Einstein, 69622 simulation approaches should be developed and it is the goal of
Villeurbanne Cedex, France. the present paper to present a numerical scheme allowing one to
E-mail address: fevotte@emse.fr (G. Févotte). solve the related PBEs.

0009-2509/$ - see front matter & 2010 Published by Elsevier Ltd.

doi:10.1016/j.ces.2010.02.009
 ARTICLE IN PRESS
3192 F. Févotte, G. Févotte / Chemical Engineering Science 65 (2010) 3191–3198
2. Modeling the crystal growth rate in the presence of
impurities
Many systems exist where impurities lead to supersaturation
thresholds below which the development of crystallization is fully
inhibited (see e.g. Sangwal, 2002). To the best of our knowledge
this feature of ‘‘real’’ industrial crystallization processes was never
described using PBEs. As shown below, unsteady-state growth
rate inhibition was proposed to explain the effect of impurities,
and such inhibition requires accounting for the time a given
particle spent in the presence of impurities. This is the reason why
it is important to account for the ‘‘birthdate’’ of crystals in the
governing crystallization PBEs. A numerical scheme, based on
MOC (Method Of Characteristics), is thus presented in the
following. Actually, the approach proposed by Kumar and
Ramkrishna (1996a, 1996b, 1997) in their series of three papers
is much more ‘‘advanced’’ than the present one, both in terms of
accuracy of the size integration technique used and of the ability
of the method to describe agglomeration and breakage phenom-
ena. Nevertheless, the present algorithm offers another way of
considering nucleation phenomena and, thanks to its great
simplicity, can be valuable for applications where fast computa-
tion is required (e.g. for in-line feedback control applications or
for kinetic parameter estimation).
The crystal growth rate G is usually assumed not to depend on
the particle size (McCabe’s hypothesis), but essentially on super-
saturation. Several expressions of the supersaturation may
actually be used, the following define the relative supersaturation
s(t) and the supersaturation ratio b(t), which both are adimen-
sional variables:
CðtÞC 
sðtÞ ¼
C
bðtÞ ¼ CðtÞ=C 
DCðtÞ ¼ CðtÞC 
*
where C is the equilibrium concentration (i.e. the solubility of the
solid compound which crystallizes) and C(t) is the actual solute
concentration.
The growth of crystals from solution is a complex process,
which can roughly be defined as the rate at which a specific or an
average characteristic size of the solid particle increases. Despite
its lack of physical meaning, the following simplified overall
kinetic law can be used:
dL
Go ðtÞ ¼ ¼ kg ðCðtÞC  ðtÞÞi ¼ kg DCðtÞi
dt
where index o will refer in the following to the growth rate in
pure solvent.
In Eq. (4), exponent i depends on the involved growth
mechanism(s). In practice, consistently with ‘‘standard’’ theore-
tical models, most published values of i are given between 1
and 2.
2.1. Effect of the adsorption of impurities on the crystal growth rate
When compared to the growth of crystals in pure solvent, the
time-averaged advancement velocity of a step in impure media
can be hindered by the adsorption of impurity species on the
growing crystal surface. Kink sites can be blocked by foreign
species that cannot easily incorporate in the crystal lattice. To
allow further crystal growth, the growth-step has to circumvent
the adsorbed impurity, which obviously reduces the overall
growth rate. Several models describing such ‘‘pinning mechan-
ism’’ were early described in the literature (Cabrera and
Vermilyea, 1958) and many convincing experimental observa-
tions of the pinning mechanism were reported in the literature,
notably thanks to the use of advanced imaging techniques such as
AFM (Atomic Force Microscopy, see e.g. Land et al., 1999; Jiang
et al., 2003; Thomas et al., 2004; Geng et al., 2005).
Kubota–Mullin’s model (Kubota and Mullin, 1995; Kubota
et al., 1997) was proposed to describe the pinning mechanism
through the following ratio G between the step velocities in pure
(u0) and impure (u) solvents:
u h ga i
G¼ ¼ 1 y ¼ 1ay ð5Þ
u0 kT sL
where g is the edge free energy, a is the size of the growth unit, T
is the absolute temperature, k is the Boltzmann constant and y* is
the steady-state coverage fraction of active growing crystal
surface by adsorbed impurities.
The overall parameter a is the effectiveness factor quantifying
the efficiency of the impurity specie in hindering the crystal
growth. Eq. (5) shows that a does not only depend on properties
of the involved solid, but also on supersaturation. Several
approaches can be used to compute the equilibrium coverage y*.
In Kubota–Mullin’s Model, the equilibrium coverage of the
growing surface is estimated thanks to Langmuir adsorption
theory:
y ¼ KC =ð1 þKC Þ
i i ð6Þ
where K is the Langmuir adsorption constant and Ci is the
concentration of impurity.
Even though the adsorption process is often regarded as
instantaneous (i.e., the steady-state coverage y* is reached
instantaneously), it was shown that the dynamics of adsorption
on the crystal surface cannot always be neglected. This is the
ð1Þ
reason why, as a first phenomenological approximation, the
transient behavior of the coverage process (i.e. the time variations
ð2Þ of y) was suggested by Kubota (2001) to obey the following first-
order dynamics where t is the time constant of the coverage
ð3Þ process:
y ¼ y ½1expðt=tÞ ð7Þ
2.2. Expression of the PBEs accounting for the adsorption of
impurities
Mono-dimensional PBEs (Population Balance Equations) re-
duce to the well-known following partial differential equation
allowing one to compute the time variations of the Crystal Size
ð4Þ Distribution (CSD). For the sake of simplicity, supersaturation s is
omitted in the list of arguments, even though it is the driving
force of almost all crystallization phenomena:
@cðL; tÞ @cðL; tÞ
þGðtÞ ¼0 ð8Þ
@t @L
cðL; 0Þ ¼ 0 or cðL; 0Þ ¼ cseed ð9Þ
RN ðtÞ
cð0; tÞ ffi cðL ; tÞ ¼ ð10Þ
GðtÞ
RN is the rate of nucleation expressed in # s  1 m  3, G= dL/dt is the
crystal growth rate, in m s  1.
The initial condition (9) accounts for the possibility of the
crystallization to start through primary nucleation (i.e. no solid
phase is initially present in the crystallizer) or through seeding.
Due to the supersaturated state of the solution, seed particles
initiate the crystallization process and are characterized by their
initial size distribution cseed.
Assuming that the growth rate is proportional to the step
velocity (i.e. G/GoG) it finally turns out that G depends on both
 ARTICLE IN PRESS
F. Févotte, G. Févotte / Chemical Engineering Science 65 (2010) 3191–3198 3193

time and supersaturation. Combining expressions (4)–(7) leads to

the following expression where n is the time at which the crystal
surface is set in contact with impurities. In the industrial practice
the impurities are present from the beginning of the crystal-
lization process so that one can assume that the nucleation events
(i.e. primary homogeneous or heterogeneous and further second-
ary nucleation) occur in the presence of impurities. The time n is
therefore assumed to coincide with the nucleation time. As an
example, the following growth rate expression is now considered:
   
KCi ðtnÞ
GðtÞ ¼ Go ðtÞ 1a 1exp 
1þ KCi t
   
i KCi ðtnÞ
¼ kg DCðtÞ 1a 1exp  ð11Þ
1 þ KCi t

Applying the impurity adsorption model to Eqs. (8)–(10) is not

straightforward as it increases the dimension of the problem: the
time (t  n) spent by the crystals in contact with impurities should
now be accounted for. Even though they are exposed to the same Fig. 1. Schematic representation of the coupling between the numbers of particles
supersaturation, two crystals taken at a given time t do not exhibit nucleated at time n and the overall distribution at time t.

the same growth rate G. This is why a population density function

f is now introduced, which depends on the ‘‘classical’’ variables, L provided that experimental data about the solubility curve C*(T)
and t, as well as n, the time of nucleation of a given particle: are available.
Moreover, during crystallization the suspension is kept in
@fðL; t; nÞ @fðL; t; nÞ
þ Gðt; nÞ ¼0 ð12Þ supersaturated conditions, one can therefore write that s 40 and
@t @L
the following condition is always fulfilled:
fðL; 0; nÞ ¼ 0 ð13Þ 8t A R þ ; 8n A ½0; t; GðtÞ 4 0 ð17Þ
Now, the following characteristic curve is considered to
RN ðtÞ
fð0; t; nÞ ¼ dðtnÞ ð14Þ describe the size population of particles nucleated at time n:
Gðt; nÞ
Z t
The standard definition of the crystal size distribution can still 8t A R þ ; 8n A ½0; t; L~ v ðtÞ ¼ Gðt 0 Þ dt 0 ð18Þ
be retrieved as v

Z 1 Z t As depicted in Fig. 1, the CSD along a given characteristic curve

cðL; tÞ ¼ fðL; t; nÞ dn ¼ fðL; t; nÞ dn ð15Þ is defined as follows:
0 o
c~ n ðtÞ ¼ cðL~ n ðtÞ; tÞ;
so that one can write:
3. A method of characteristics for solving population balance
~ ðtÞ
dc @cðL~ n ðtÞ; tÞ @L~ n ðt; nÞ @cðL~ n ðtÞ; tÞ
equations accounting for impurity effects n
¼ þ
dt @t @t @L
~
@cðL n ðtÞ; tÞ @cðL~ n ðtÞ; tÞ
A simple method of characteristics for the resolution of the PBE ¼ þ Gðt; nÞ ¼0 ð19Þ
without agglomeration and breakage is now presented in the @t @L
‘‘standard’’ case corresponding to mono-dimensional crystals
It therefore turns out that c ~ does not depend on t: this
grown in pure solvent; then, the case when unsteady-state v
impurity adsorption occurs is envisaged. implies that the solution of Eq. (8) is fully determined by the
boundary condition (10) and the resolution of (18) describing the
time evolutions of the characteristic curves:
3.1. A method of characteristics for monodimensional 8t A R þ ; 8n A ½0; t;
‘‘standard’’ PBEs
cðL~ n ðtÞ; tÞ ¼ c~ n ðtÞ ¼ c~ n ðnÞ ¼ cð0; nÞ ð20Þ
As already mentioned supersaturation is the driving force of
crystallization and computing s requires computing C(t) during RN ðnÞ
the crystallization process. Actually, the decrease of solute
cðL~ n ðtÞ; tÞ ¼ ð21Þ
GðnÞ
concentration C(t) is caused by the generation of crystals: the
Now, let us show that Eq. (21) allows determining the CSD for
molecules of solute are transferred through crystallization to the
every time and size, i.e. every point (L, t) in the phase space can be
dispersed solid phase. The total amount of solid is therefore given
represented thanks to Eq. (18). To this effect, the following
by the total volume of particles computed through the integration
application is considered:
of the whole CSD:
r s jp Z 1 rj Z 1 lt : ½0; t-R þ ; n/L~ n ðtÞ
Cs ðtÞ ¼ cðL; tÞL3 dL ffi s p cðL; tÞL3 dL ð16Þ þ
8t A R , lt is clearly continuous and
Ms L Ms 0
3
where rs (kg/m ) and Ms are the density and the molecular weight dlt
8t A R þ ; 8n A ½0; t; ðnÞ ¼ GðnÞ
of the solid compound, jp is a volumetric particle shape factor dn
(i.e. p/6 in the case of spherical crystals.) which, given Eq. (17), shows that lt is strictly decreasing and
An elementary mass balance of the solute allows computing at therefore invertible from [0, t] to ½0; L~ 0 ðtÞ. It follows that the
each time the evolutions of C(t) and consequently yields s(t), characteristic curves do not exhibit shock or rarefaction.
 ARTICLE IN PRESS
3194 F. Févotte, G. Févotte / Chemical Engineering Science 65 (2010) 3191–3198

A means of computing the distribution density function is ~ ðt; nÞ does not depend on t, and it can
It finally turns out that f m
therefore given by also be concluded that
1
RN ðlt ðLÞÞ 8t A R þ ; 8ðn; mÞ A ½0; t2
8t A R þ ; 8LA ½0; L~ 0 ðtÞ; cðL; tÞ ¼ 1
ð22Þ
Gðlt ðLÞÞ
fðL~ m ðt; nÞ; t; nÞ ¼ f~ m ðt; nÞ ¼ f~ m ðm; nÞ ¼ fð0; m; nÞ ð30Þ
Considering successive sampling times, the time variable n is
discretized as follows: RN ðmÞ
Z Li1 ðtÞ fðL~ m ðt; nÞ; t; nÞ ¼ dðmnÞ ð31Þ
Gðm; nÞ
8t A R þ ; 8vA ½0; t; Li ðtÞ ¼ L~ vi ðtÞ ci ðtÞ ¼ cðL; tÞ dL ð23Þ
Li ðtÞ The time variable m is discretized considering successive
From Eq. (21), and setting the following change of coordinates: sampling times m A fmi ; i A Ng and one can define the following
8 distribution function:
1
>
< n ¼ lt ðLÞ
ð24Þ 8t A R þ ; 8n A ½0; t
dL
: L ¼ L~ n ðtÞ;
> ¼ GðnÞ Li ðtÞ ¼ L~ mi ðt; nÞ
dn
Z Li1 ðt;nÞ
we obtain Fi ðt; nÞ ¼ fðL; t; nÞ dL
Li ðt;nÞ
Z Li1 ðtÞ Z ni Z ni Z t
Ci ðtÞ ¼ cðL; tÞ dL ¼ cðL~ n ðtÞ; tÞGðnÞ dn ¼ RN ðnÞ dn Ci ðtÞ ¼ Fi ðt; nÞ dn ð32Þ
Li ðtÞ ni1 ni1 0
ð25Þ
As illustrated in Fig. 1, it finally turns out that integrating c in Using Eq. (28) and setting
size between Li(t) and Li–1(t), at a given time t, amounts to 8 1
integrating c in n between ni–1 and ni, for a given size, and that the < m ¼ lt;n ðLÞ
>
result of this integration does not depend on time t. Consequently, dL ð33Þ
: L ¼ L~ m ðt; nÞ
> ðmÞ ¼ Gðm; nÞ
to solve the system (8)–(10), one has to solve the following two dm
systems coupled by the integral operator G. In order not to make Z Li1 ðt;nÞ
the notation heavier, the latter will still be referred to as G after
yields : Fi ðt; nÞ ¼ fðL; t; nÞ dL
discretization: Li ðt;nÞ
8 8 Z mi
>
< dLi ðtÞ ¼ GðtÞ; >< dCi ðtÞ ¼ 0
dt Fi ðt; nÞ ¼ fðL~ m ðt; nÞ; t; nÞGðm; nÞ dm ð34Þ
dt R ð26Þ mi1
>
: Li ðni Þ ¼ 0; >
: Ci ðni Þ ¼ nni RN ðnÞ dn
i1
Z mi
Fi ðt; nÞ ¼ dðmnÞRN ðmÞ dm ð35Þ
Any time-discretization scheme can be used to solve jointly mi1
the two systems (26). As an example, t might be discretized in the 
same way as n (i.e. the same sampling interval dt is used to scan t RN ðnÞ if n A ½mi1 ; mi 
Fi ðt; nÞ ¼ ð36Þ
and n), which leads to the following simple numerical scheme: 0 otherwise
k ¼ 0; It follows that:
while tk otend do Z t Z mi
 Ci ðtÞ ¼ Fi ðt; nÞ dn ¼ RN ðnÞ dn ð37Þ
 f or i ¼ 1; k1 do
 Rt 0 mi1

 jLi ðtk Þ ¼ Li ðtk1 Þ þ tkk1 GCðtÞ dt;
 The following two systems are thus obtained which are
 end
 coupled by the integral operator G:

  X

 L ðt Þ ¼ 0; GðtÞ ¼ f t; mi ; C  ; Cs pms p Ci ðmi ÞL3i ðtÞ ð38Þ
 k k
 R i
 Ck ðtÞ ¼ tk RN ðnÞ dn;
 tk1 8

 k ¼ k þ1; < dLi ðtÞ ¼ GðtÞ
dt ð39Þ
end : L ðm Þ ¼ 0
i i

8
3.2. Extension of the method for growth rate depending on n
< dCi ðtÞ ¼ 0
>
dt ð40Þ
As explained above, the growth rate G depends now on the > C ðm Þ ¼ R mi R ðmÞ dm
: i i m N
i1
time spent crystals in the presence of adsorbing impurities. Now,
let us consider characteristic curves defined as follows: The principle of the numerical scheme is the same as explained
Z t previously (see Fig. 1).
8t A R þ ; 8ðn; mÞ A ½0; t2 ; L~ m ðt; nÞ ¼ Gðt 0 ; nÞ dt 0 ð27Þ
m
4. Application to the simulation of the crystallization of citric
The distribution along a given characteristic curve is noted: acid in the presence of impurities
f~ m ðtÞ ¼ fðL~ m ðt; nÞ; t; nÞ and one can write:
4.1. Effects of impurities on the development of isothermal
~ ðt; nÞ
@f @fðL~ m ðt; nÞ; t; nÞ @L~ m ðt; nÞ @fðL~ m ðt; nÞ; t; nÞ crystallizations
m
¼ þ
@t @t @t @L
~
@fðL m ðt; nÞ; t; nÞ @fðL~ m ðt; nÞ; t; nÞ In order to illustrate the resolution method, the crystallization
¼ þ Gðt; nÞ ¼0 ð28Þ
@t @L of citric acid monohydrate is simulated using kinetic data
 ARTICLE IN PRESS
F. Févotte, G. Févotte / Chemical Engineering Science 65 (2010) 3191–3198 3195

previously published by Févotte et al. (2007). These data outline resolution method and to analyze possible effects of the
the major role played by secondary nucleation mechanisms in the adsorption of impurities on the development of the crystallization
development of the crystallization of citric acid. In the absence of process, than to investigate real solute/solvent/impurity systems.
reported experimental studies dealing with impure media, the Isothermal desupersaturation crystallization operations were
parameters of Kubota–Mullin’s model are set arbitrarily in order simulated at 15 1C. To initiate the crystallization in the super-
to compare the features of crystallization operations performed saturated zone (i.e. Cinit 4C* at 15 1C) a seed mass of 10 kg is
with and without impurities. The latter data were simply set to supposed to be introduced in a pilot-scale crystallizer fed with a
reproduce the kinetic features of systems for which, according to supersaturated citric acid solution containing 1 m3 of water. The
Eqs. (5)–(7), supersaturation thresholds can be observed at the initial dissolved citric acid concentration is Cinit = 1.825 kg anhy-
end of the batch process. It is important to notice that no effect of drous CA/kg water. The algorithm presented in Section 3 was
impurities on the rate of nucleation is assumed to occur in the applied to system (8)–(10) after setting the parameters presented
following. Such assumption is probably unrealistic but published in Tables 1 and 2. As one can see in Fig. 2, the computed CSD is
data about nucleation in the presence of impurities is really smooth and does not exhibit any oscillatory behavior, even when
lacking. The goal here is rather to show the usefulness of the coarse time intervals are used for the numerical simulation (see
Section 4.2).
After seeding, the number of particles increases due to
Table 1 secondary nucleation, and the initial supersaturation is consumed
Kinetic equations and parameters used for the simulation of the crystallization of through the growth of crystals, as displayed in Fig. 3d and a.
citric acid monohydrate.
Despite the adsorption of impurity species at the crystal surfaces,
dL the impurity concentration Ci is assumed to remain constant
¼ GðtÞ ¼ kg ðCðtÞCm  ðTÞÞj ¼ 7:18106 ðCðtÞCm  ðTÞÞ1:58
dt during the crystallization (i.e. the amount of adsorbed molecules
is assumed negligible with respect to the dissolved impurities.)
RN ðtÞ ¼ k2 CS ðtÞim ðCðtÞCm

ðTÞÞjm ¼ 1:72108 CS ðtÞ0:47 ðCðtÞCm

ðTÞÞ1:14
Table 2 shows the parameters used for comparative simula-
where: RN(t) is the secondary surface nucleation rate of
tions aiming at analyzing the effect of the concentration Ci on the
monohydrated citric acid [# s  1 m  3] development of isothermal batch crystallizations. Typical results
k2 is a ‘‘lumped’’ kinetic constant for the secondary are displayed in Fig. 3 to illustrate the main features of the effects
nucleation of monohydrated citric acid of impurities. As a reference, simulation no. 1 corresponds to the
Cm  is the solubility in water of monohydrated citric acid
crystallization of monohydrate citric acid in pure water, the
(1.35 kg/kg water at 15 1C)
i, im, jm are exponents proposed by Févotte et al. (2007) operating and modeling conditions are the same as previously
(see Table 1). The presence of impurities has a clear impact on the
Parameters of Kubota’s–Mullin’s adsorption model development of the CSD and yields, in the average, smaller
K =5 (m3/kg) particles than the ones obtained in pure solvent. In particular, as
t = 500 (s)
Ci = 0.01 (kg/m3)
shown in Fig. 3b, the particles obtained at the end of the
as = 1 crystallization process in impure water (Run 4) are about 25%
smaller than in pure water (see also Fig. 4).
As expected, the supersaturation reaches a constant asympto-
tic value at the end of batch processes with impurities (e.g.
Table 2 slim = 0.05 for simulation no. 4), while less than 1% supersaturation
Parameters of Kubota–Mullin’s model used for the batch simulations displayed in remains after 1500 s during Run 1. Such value of slim corresponds
Figs. 2 and 3.
to the final threshold supersaturation expected from Kubota–
Simulation no. K (kg m  3)  1 Ci (kg m  3) t (s) Mullin’s model for a 41.
The CSDs are significantly modified by the presence of
1 – 0 – impurities and the main effect of impurities appears to be the
2 5 0.005 500 significant generation of fines during the whole batch process,
3 5 0.01 500
4 5 0.01 0
especially at the end where almost no fines are generated in pure
solvent (see final curves in Fig. 3a) while more than 5  106

Fig. 2. Simulation of seeded isothermal crystallization of citric acid monohydrate in pure water at 15 1C: time variations of the CSD (# L  1 s  1). The kinetic parameters are
given in Tables 1 and 2: (a) crystallization performed in pure water and (b) crystallization performed with 0.01 kg m  3 impurities and unsteady-state adsorption (t =500 s).
 ARTICLE IN PRESS
3196 F. Févotte, G. Févotte / Chemical Engineering Science 65 (2010) 3191–3198

Fig. 3. Simulation of seeded isothermal crystallization of citric acid monohydrate in pure water (curve 1) and with the presence of impurities, at 15 1C:
(a) desupersaturation profile, (b) time evolution of the number average crystal size, (c) generation of total crystallized solid during time in kg/L, and (d) time variations
of the overall number of crystals in nb/L. Run 1 (continuous), 2 (dashed), 3 (dashed-dotted), 4 (dotted), Indices 1–4 refer to the parameters in Table 2.

Fig. 4. Number and weight particle size distribution histograms obtained at the end of crystallization operations performed without (Run 1) and with impurities (Run 3).

particles/L  1 s  1 are generated in the presence of impurities constant t soften the deterioration of the CSD, as well as the
during Run 3. With the same impurity concentration it is clear reduction of the solid production. However, it is important to
that the dynamic features represented by the overall time remind that the simulations are based on the assumption that the
 ARTICLE IN PRESS
F. Févotte, G. Févotte / Chemical Engineering Science 65 (2010) 3191–3198
nucleation process remains unaffected by the presence of
impurities; the availability of data describing the sensitivity of
the nucleation process to impurities might therefore significantly
change the main trends observed concerning the effects of
impurities on the CSD variations.
Concerning the crystallization advancement, Fig. 3c shows that
only 86% of the expected solid is obtained in the presence of
impurities (0.5 kg/L of crystals is expected from the values of Cinit
and C*, while only 0.43 kg/L is obtained).
The results presented in Fig. 2 might appear as rather
misleading because the displayed PSDs are not true population
density functions: according to the proposed resolution numerical
method every point of a given curve represents the number of
particles in the size interval [L, L+ G.dt], where dt is the constant
sampling time. It is clear that such size interval is large at the
beginning of the crystallization process; while it is small at the
end (i.e. G is much smaller at the end). Consequently, the
distribution of small particles is under-represented by Fig. 2. This
is why the results were recomputed in terms of normalized size
distribution histograms and presented in Fig. 4. The increase of
the amount of fines is now clearly shown by the final number
distribution histogram in Fig. 4a while the reduction of the size of
bigger particles is outlined by the weight distribution displayed in
Fig. 4b.
4.2. Some comments on the convergence of the numerical method
In order to evaluate the convergence of the method as a
function of the discretization time-step dt, the same simulation
was performed with different values of dt. Due to the lack of
analytical solution, no exact reference simulation is available
here, this is why the time-step dt= 0.005 s was arbitrarily set as a
reference allowing to compare the convergence of the solution for
increasing time-steps. The ‘‘simulation error’’ was then estimated
as follows:
Rn Rn
0 N ref ðtÞdt 0 N i ðtÞdt
eiðnÞ ¼ Rn for n A ½0; 3000 ð42Þ
0 N ref ðtÞ dt
where N i ðtÞ is the computed number of crystals generated
between time t and t +dt, index i refers to the simulation under
consideration (i.e. performed with a given value of the time step
dt) and index ‘‘ref ‘‘refers to the ‘‘reference’’ simulation.
The relative error on the computed CSD increases with the
time-step. For example, when the time-step increases from 0.02
to 0.05 s, the maximum discrepancy between the 2 cumulative
distributions is equal to 10  4 (at time t =300 s) which, however,
remains quite acceptable. A log–log plot of the numerical error in
the cumulative CSD, as a function of the time step shows a
convergence order of 1, which is probably mostly limited by the
order of the numerical integration formulas used in the scheme to
compute Eq. (16).
5. Conclusions
A simple method of characteristics for the resolution of
population balance equations applied to crystallization processes
without agglomeration and breakage was developed and eval-
uated through the dynamic simulation of the crystallization of
citric acid in water. The method is intended to allow representing
growth rate reductions observed during solution crystallizations
performed in the presence of industrial impurities. Indeed, the
effect of impurities was shown by Kubota and Mullin (1995) to
depend on the time spent by every crystal in the presence of
dissolved impurities and consequently, to depend on the ‘‘age’’ of
the crystals. Such a particular problem required accounting for an
3197
additional time variable in the expression of the PBEs and finding
a way of solving the resulting PBE system.
From a physical viewpoint, the PBE model turns out to be
consistent with industrial observations and among other results,
the PBE modeling allows representing the reduction of the yield of
industrial crystallizations which is generally poorly understood
but often attributed to the presence of impurities. The simulations
also demonstrate the impact of the unsteady-state behavior of the
adsorption of impurities on the performance of the crystallization
process in terms of yield and CSD of the solid product. Such
simulations could be applied, for example, to the design of model-
based control strategies aiming at improving the productivity of
batch crystallizers and/or the quality of crystals generated in
impure industrial mother liquors. To the best of our knowledge,
such control studies still remain to be undertaken.
Nomenclature
C solute concentration, kg solute/kg solvent
C* solubility concentration, kg solute/kg solvent
Ci impurity concentration, kg m  3
CS solid concentration, kg m  3
G growth rate, m s  1
i exponent of the supersaturation dependency of the
crystal growth rate, dimensionless
im exponent of the dependency of the nucleation rate on the
concentration of solid in suspension, dimensionless
jm exponent of the supersaturation dependency of the
nucleation rate, dimensionless
K Langmuir’s constant, m3 kg  1
kg growth rate constant, dimensionless
K2 kinetic nucleation parameter
L particle size, m
N(r, number of particles at time t in a given subset, # m  3
t)
RN nucleation rate # s  1 m  3
T time, s
Greek letters
c(L,t)population density function, # m  1 m  3
y fraction of coverage of growing crystal surface by
adsorbed impurity
yn fraction of coverage of growing crystal surface by
adsorbed impurity at the equilibrium, dimensionless
n nucleation time, s
t adsorption time constant, s
s supersaturation, dimensionless
Acknowledgements
We greatly acknowledge the French Research Agency, ANR for
the support granted to the Project ‘‘IPAPI’’ (Improving the
Properties of Active Pharmaceutical Ingredients), Ref. 07-BLAN-
0183.
References
Cabrera, N., Vermilyea, D.A., 1958. Growth and Perfection of Crystals, New York,
393 pp.
Chernov, A.A., 2004. Notes on interface growth kinetics 50 years after Burton,
Cabrera and Frank. J. Cryst. Growth 264 (4), 499–518.
Chow, A.H.-L., Chow, P.K.K., Zhongshan, W., Grant, D.J.W., 1985. Modification of
acetaminophen crystals: influence of growth in aqueous solutions containing
p-acetoxyacetanilide on crystal properties. Int. J. Pharm. 24, 239–258.
 ARTICLE IN PRESS
3198 F. Févotte, G. Févotte / Chemical Engineering Science 65 (2010) 3191–3198
Févotte, G., Caillet, A., Sheibat-Othman, N., 2007. A population balance model of
the solution-mediated phase transition of citric acid. AIChE J. 53 (10),
2578–2589.
Geng, Y.L., Xu, D., Wang, Y.L., Du, W., Liu, H.Y., Zhang, D.H., Wang, X.Q., 2005.
Atomic force microscopy study on surface morphologies of {1 1 0} faces of
MnHg(SCN)4 crystals. J. Cryst. Growth 277 (1–4), 555–559.
Jiang, X.N., Xu, D., Sun, D.L., Yuan, D.R., Zhang, Q.Y., 2003. Atomic force microscopy
studies on surface morphologies of CdHg(SCN)4 crystals grown in solutions
containing excessive amount of Cd(II) cations. Cryst. Res. Technol. 38 (2), 143–
149.
Johnson, Douglas S., Li, Jie Jack, 2007. The Art of Drug Synthesis. Wiley-Interscience
276 pp.
Kubota, N., Mullin, J.W., 1995. A kinetic model for crystal growth from aqueous
solution in the presence of impurity. J. Cryst. Growth 152 (3), 203–208.
Kubota, N., 2001. Effect of impurities on the growth kinetics of crystals. Cryst. Res.
Technol. 36 (8–10), 749–769.
Kubota, N., Yokota, M., Mullin, J.W., 1997. Supersaturation dependence of crystal
growth in solutions in the presence of impurity. J. Cryst. Growth 182 (1–2),
86–94.
Kubota, N., Yokota, M., Mullin, J.W., 2000. The combined influence of super-
saturation and impurity concentration on crystal growth. J. Cryst. Growth 212
(3–4), 480–488.
Kumar, S., Ramkrishna, D., 1996a. On the solution of population balance equations
by discretization—I. A fixed pivot technique. Chem. Eng. Sci. 51 (8),
1311–1332.
Kumar, S., Ramkrishna, D., 1996b. On the solution of population balance equations
by discretization—II. A moving pivot technique. Chem. Eng. Sci. 51 (8),
1333–1342.
Kumar, S., Ramkrishna, D., 1997. On the solution of population balance equations
by discretization—III. Nucleation, growth and aggregation of particles. Chem.
Eng. Sci. 52 (24), 4659–4679.
Land, T.A., Martin, T.L., Potapenko, S., Palmore, G.T., De Yoreo, J.J., 1999. Recovery
of surfaces from impurity poisoning during crystal growth. Nature 399,
442–445.
Myerson, A.S., 2001. Handbook of Industrial Crystallization, second ed Butter-
worth-Heinemann 304 pp.
Raghavan, S.L., Trividic, A., Davis, A.F., Hadgraft, J., 2001. Crystallization of
hydrocortisone acetate: influence of polymers. Int. J. Pharm. 212 (2), 213–221.
Sangwal, K., 1996. Effects of impurities on crystal growth processes. Prog. Cryst.
Growth Charact. Mater. 32 (1–3), 3–43.
Sangwal, K., 2002. On the nature of supersaturation barriers observed during the
growth of crystals from aqueous solutions containing impurities. J. Cryst.
Growth 242 (1–2), 215–228.
Sangwal, K., 2007. Additives and Crystallization Processes. From Fundamentals to
Applications. John Wiley & Sons 451 pp.
Thomas, T.N., Land, T.A., Martin, T., Casey, W.H., DeYoreo, J.J., 2004. AFM
investigation of step kinetics and hillock morphology of the {1 0 0} face of
KDP. J. Cryst. Growth 260 (3–4), 566–579.
Wood, W.M.L., 2001. A bad (crystal) habit—and how it was overcome. Powder
Technol. 121 (1), 53–59.