chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623

Contents lists available at ScienceDirect

Chemical Engineering Research and Design

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

Kinetic study of adductive crystallization of bisphenol-A

A. Alamdari ∗ , S. Shariat Nasab, A. Jahanmiri

Chemical Eng. Dept., School of Chemical and Petroleum Eng., Shiraz University, 7134851154, Shiraz, Iran

a b s t r a c t

Kinetics of adductive crystallization of bisphenol-A (BPA) from a pure phenolic solution was investigated by model
fitting to experimental data. BPA is an organic compound with vast applications in manufacturing of epoxy resins
and polycarbonates; and adductive crystallization is used in industry to separate BPA from the reaction mixture of
BPA synthesis. Using kinetic equations and mass and population balances the adductive crystallization process of
BPA was modeled. Solid and liquid information was obtained from a set of crystallization experiments carried out
under a cooling program. A genetic algorithm was applied to optimize the kinetic parameters of nucleation and
growth rate equations. The nucleation rate equation for crystallization of BPA–phenol adduct from a pure synthetic
2.8
solution was B0 = 8.75 × 1017 exp(−10900/RT)(C) MT1.17 , where B0 is the nucleation rate, C is the supersaturation,
and MT is the slurry density. The growth rate equation suggested was G = 5.72 × 1013 exp(−10400/RT)(C)2.7 , where
G is the growth rate of adduct particles. Fitting model predictions to experimental data demanded no addition of
agglomeration terms to population balance equation which suggests an insignificant mechanism of agglomeration
in particle enlargement. The nucleation and growth rate equations obtained in this study may be used to enhance
the product quality of BPA plants.
© 2010 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.

Keywords: Adductive crystallization; Bisphenol-A; Modeling; Parameter estimation

1. Introduction The previous work on the subject of crystallization kinetics

of BPA-adduct is almost a rarity. An extensive literature review
Bisphenol-A (BPA) is an organic compound composed of two resulted in only one study of Moyers (1986) in which he related
phenolic rings attached to the center carbon of a propyl agroup the nucleation rate of BPA-adduct crystals to their growth as
((CH3 )2 C (C6 H4 OH)2 ) (Hawley, 1981). The industrial applica-
tions of BPA are in industries of resins, plastics, and flame B0 = kn G1.14 MT1.17 (1)
retardants. BPA is generally synthesized through the conden-
sation reaction of phenol and acetone under the influence of a where B0 is the nucleation rate and was defined as the number
cationic-exchange-resin catalyst. It is separated from the reac- of smallest detectable crystals generated in unit time in accor-
tion products through the adductive crystallization process. dance to a unit mass of free phenol in the solution, MT is the
BPA-adduct crystals are then dissolved in the pure phenol suspension density, kn is the nucleation rate coefficient, and G
and re-crystallized for a better purity. Filtration in both crys- is the linear growth rate of adduct crystals. Moyers (1986) took
tallization and recrystallization processes of adduct crystals samples from an industrial continuous MSMPR BPA-adduct
is carried out under thorough washing with pure phenol to crystallizer and found a relation between the nucleation and
remove traces of impurities from the surface of adduct crystals growth rates. However, he did not report the dependency
(Kosaka and Sinclair, 1982). of nucleation and growth mechanisms to supersaturation as
A main goal in industry of BPA is to produce high pure driving force.
adduct crystals with appropriate qualities in size and appear- Both crystallization mechanisms of growth and nucle-
ance suitable for downstream processing. The finer the ation need a solute concentration in crystallizing solution in
crystals the more impurities they carry on their surfaces and the range of metastability. However, a concentration at the
therefore, they have the lower purity and color quality. high extreme of metastability zone makes the particle size

∗
Corresponding author. Tel.: +98 9171184316; fax: +98 7116281548.
E-mail addresses: [email protected], [email protected] (A. Alamdari).
Received 30 October 2008; Received in revised form 27 March 2010; Accepted 9 April 2010
0263-8762/$ – see front matter © 2010 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.cherd.2010.04.006
1616 chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623

Nomenclature

b supersaturation order of nucleation rate

B0 nucleation rate [# s−1 (gfree phenol )−1 ]
B̃0 rate of generation of smallest detectable size
particles vmin in the bottle crystallizer [# s−1 ]
C BPA concentration in phenolic solution [g
(gfree phenol )−1 ]
C* saturation concentration of BPA in phenol [g
(gphenol )−1 ]
C̃ total BPA mass in liquid phase [g]
C̃0 initial total mass of BPA in liquid phase [g]
C supersaturation [g (gfree phenol )−1 ] Fig. 1 – Schematic diagram of the set up used in the
Eg activation energy of growth rate [J mol−1 ] preliminary experiments of BPA adductive crystallization.
En activation energy of nucleation rate [J mol−1 ]
g supersaturation order of growth rate present study to measure the metastable zone width for the
G linear crystal growth rate [␮m s−1 ] crystallization process of BPA–phenol adduct. The appropriate
Gv growth rate defined based on particle volume cooling rate was explored and the crystal breakage during the
as size [␮m3 s−1 ] course of experiments was investigated.
k1 a factor relating the solute mass in crystal to However, since little information was available in the lit-
crystal mass equal to 0.29 [gBPA g−1 crystal
] erature about the growth and nucleation mechanisms of
kg linear growth rate coefficient [␮m s−1 (g BPA–phenol adduct crystallization, this study was generally
(gfree phenol )−1 )−g ] aimed at obtaining the kinetics of adductive crystallization
kg0 inherent linear growth rate coefficient [␮m s−1 process of BPA from phenolic solution. Kinetic mechanisms
(g (gfree phenol )−1 )−g ] of growth and nucleation which are influential in size distri-
kn0 inherent nucleation rate coefficient bution and average size of product are essential in design and
−b
[(# g−1
free phenol
s −1 ) (g g −1
BPA free phenol ) · operations of industrial crystallizers.
liquid
−m
(gcrystal g−1 ) ]
free phenol solid 2. Experimental
kv shape factor of particles
u a dummy crystal size [␮m3 ]
Experiments were carried out in plastic bottles as crystal-
vmin smallest detectable size in the distribution
lizers in a rotating bath at appropriate cooling rates. Width
[␮m3 ]
of metastable zone for a phenol solution saturated with BPA
vmax largest size in the distribution [␮m3 ]
in the temperature range of 50–70 ◦ C was examined through
v particle size [␮m3 ]
cooling the saturated solution and recording the turbidity tem-
m slurry density order of nucleation rate
perature, and successively heating the solution and recording
MBPA molecular weight of BPA [g gmol−1 ]
the transparency temperature. Breakage of particles during
Mphenol molecular weight of phenol [g gmol−1 ]
agitation was examined by agitating a suspension of BPA-
MT slurry density [gcrystal g−1 free phenol
]
adduct particles in a simulated solution of sucrose with similar
M̃T total crystal mass in the bottle [g]
viscosity and density to the real BPA–phenol solution. Appro-
M̃T0 total mass of initial seed in the bottle [g]
priate cooling rate at which excessive nucleation is prevented
ñ population distribution in the bottle at time t [#
was found by performing a few cooling crystallization trials
␮m−3 ]
from 68.5 to 45 ◦ C through a third-order-temperature profile at
ñseed seed population distribution [# ␮m−3 ]
different duration times.
R universal gas constant [J mol−1 K−1 ]
Preliminary experiments: A few preliminary experiments
t time [s]
were carried out through a set up shown in Fig. 1. The set
T temperature [◦ C]
up consisted of a 2-l-jacketed glass reactor equipped with a
marine type agitator, a thermocouple, and a sampling tube.
The temperature of the crystallizing solution was controlled by
distribution of product unstable due to excessive primary a system including a cold water tank, a mixing tank equipped
nucleation. Therefore, operation in a narrow metastable zone with a heater and a thermometer, and a centrifugal pump
most likely will lead to appearance of extra nucleation. Width shown in the figure.
of metastable zone which may be considered as a charac- Seed preparation: A bulk of seed particles was split to five por-
teristic property of the crystallization system is a means to tions using a sample splitter. The split portions were expected
predict the product specifications (Kim and Mersmann, 2001). to have almost the same size distributions. Each portion was
The rate of cooling process, the mixing intensity, the presence introduced to a bottle as initial seed crystals.
of impurities, and the thermal history of solution are influen- Rotating bath: The schematic diagram of the rotating bath
tial in the location of the labile curve (Mullin, 2001). The less in which crystallization experiments for obtaining the kinetic
the width of the zone is, the more accurate control system data were performed is shown in Fig. 2. The rotating bath
should be employed to prevent primary nucleation and man- included a shaft with two disks. Several rods for holding the
age the product size distribution. Therefore, having an insight bottles were mounted on the disks. Rotation of the shaft by
into the metastable information is crucial in production of a a variable speed motor was immersing each bottle for half a
high quality crystalline product. An attempt was made in the turn in water. The bath was equipped with a heating element
 chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623 1617

tions. The structure of the adduct crystals was not expected

to be affected by the sucrose solution.
Cooling rate: Intending to find out suitable cooling rates for
crystallization operations, a set of experiments was performed
under a cooling program of third-order-temperature profile
from 68.5 to 45 ◦ C and an appropriate agitation intensity of
340 rpm in duration times of 1.5, 2.5 and 4 h. The cumulative
number distributions of seed and product samples from these
experiments were calculated using the results of sieve analy-
sis. The suitable duration time corresponding to the suitable
average cooling rate, was assigned to the case that the total
number of crystals of the final product compared to that of the
initial seed increased less. This set of experiments provided
the suitable average cooling rate at which the supersaturation
during the course of experiment was appropriately controlled
at a level below that which would result in primary nucle-
Fig. 2 – Schematic diagram of the rotating bath used for the ation. Based on the phase diagram available for BPA–phenol
kinetic study of BPA adductive crystallization. (Moyers, 1986) the adduct particles crystallize in the temper-
ature range of 68.5–45 ◦ C from the solution. At temperatures
and a cold water coil mounted on the bath roof. The com- above 98 ◦ C non-adduct crystals of BPA precipitate and below
bination of the heating element and the cooling coil which 37.8 ◦ C the eutectic point is approached. This range was also
were operated by a controller was used to adjust the water chosen in order to meet the temperature ranges of industrial
temperature in accordance with the cooling profile. It was fea- operations.
sible to use several bottles in the bath at the same time to Kinetics: A set of experiments aiming at finding the kinetic
perform the crystallization experiments. When the evolution data, was carried out in the rotating bath. An amount of 2600 g
of size distribution was under investigation, the bottles were of BPA–phenol solution, saturated at 70 ◦ C regarding to BPA
removed from the bath at different times and their contents was prepared and equally distributed in eight plastic screw-lid
were filtered and solid and liquid phases were analyzed for bottles. For thermal uniformity of solutions in the bottles with
size distribution and solute concentration, respectively. the bath, the bottles were rotated at 70 ◦ C for 1 h before seed
Metastability: An unsaturated clear solution of 9.5 wt% introduction. Each bottle containing 323 g solution was seeded
BPA in phenol (300 g) was provided in a 500 ml-glass-vessel with 3 g of already prepared seed crystals. Bath temperature
within a bath at 65 ◦ C. According to the phase diagram of the was reduced from 68 to 45 ◦ C in 8 h in accordance to the third-
phenol–BPA system, the solution was expected to be saturated order-temperature profile of
at 50 ◦ C (Moyers, 1986). The bath was allowed to be cooled nat-
 t 3
urally with an average cooling rate of 0.67 ◦ C/min while the T0 − T
= (2)
solution was moderately agitated. T0 − Tf 
The temperature at the first indication of turbidity was
recorded as the turbidity point at which the primary or where T0 and Tf are the solution temperatures at the begin-
spontaneous nucleation occurred. Then after, the bath was ning and at the end of the profile, respectively; and  is the
slowly heated up until the turbidity disappeared and the solu- duration time of the cooling process (Mullin, 2001). This kind
tion became clear and transparent. The temperature of the of profile provided a low cooling rate at the early stages of the
solution transparency was considered as the solubility or sat- cooling process where enough surface area of seed was not
uration point. Distinctness and indistinctness of the image available and provided higher cooling rates at the late stages
of a grid pattern behind the glass vessel helped to easily of cooling when enough area of crystal was available to reduce
distinguish the transparency and turbidity of the solution. the duration time of the whole process.
This policy was repeated till the temperature extremes were The precision of the temperatures employed to the water
found in which the turbidity appeared and disappeared. The bath was within 0.1 ◦ C from the predicted values by Eq. (2);
metastability range was also measured for solutions with con- however, the temperature of solutions in the bottles was not
centrations of 13 and 18.5 wt% BPA in phenol. continuously accessible for measurement. At the final stages
Breakage: BPA-adduct crystals suspended in a non-solvent of the run at the highest rate of cooling, a temperature dif-
liquid was used to study the breakage. A suspension of almost ference of maximum 2 ◦ C was noticed between the solution
15% BPA-adduct crystals in 900 g sucrose solution with a vis- in the bottles and the bath water due to a low thermal con-
cosity close to BPA–phenol solution (3.76 cP) was prepared and ductivity of the polypropylene bottles. During the course of
agitated in the vessel of Fig. 1 at 60 ◦ C and 340 rpm for about 3 h. experiment after each reduction of 3 ◦ C in bath temperature,
The agitation intensity was just enough to suspend the adduct a bottle was withdrawn from the bath and its contents were
crystals in the solution. Comparison of size distributions of filtered quickly using a Buchner funnel under vacuum con-
adduct crystals just after the particle addition to the suspen- ditions and the filter cake was washed with n-heptane. The
sion and after 3 h agitation indicated the extent of breakage last bottle was withdrawn at 45 ◦ C far enough the solution
due to particle impacts to each other and to the propeller approach to the eutectic temperature of 37.8 ◦ C (Moyers, 1986).
and the crystallizer wall. Use of simulated sucrose solution Solid and liquid phases were analyzed for size distribution
in place of the real saturated BPA–phenol solution eliminated and BPA concentration, respectively. Use of the rotating bath
the error introduced on the size distribution by the mecha- had the advantages of no particle breakage, elimination of
nisms of growth, dissolution, and nucleation of particles due sampling error during the course of experiment, and provi-
to supersaturation created by probable temperature fluctua- sion of enough solid samples for sieve analysis. However, it is
1618 chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623

essential that seed splitting in this method of crystallization where B0 is the nucleation rate defined as the number of small-
practice to be carried out precisely. It is expected that the error est detectable crystals generated in a unit time in accordance
of splitting seed crystals using a sample splitter is less than to a unit mass of free phenol in the solution, C is the super-
the error of taking a representative sample from a suspen- saturation defined as (C–C*), MT is the suspension density, kn0
sion under agitation. The error of sampling from solid phase is the nucleation rate coefficient, G is the linear growth rate of
will introduce the most inaccuracy when the size distribution adduct crystals, kg0 is the growth rate coefficient, En is the
measurement is involved in the experiments. The centrifu- nucleation rate activation energy, Eg is the growth rate acti-
gal and gravity forces tend to classify particles in an agitated vation energy, the parameters m and b are the nucleation rate
suspension and in the entrance of the sampling tube, and to order with respect to MT and C, respectively; and g is the
introduce error to the samples taken as representatives. growth rate order with respect to supersaturation.
It is notable that the rotating bath was used only to Mass of BPA deposited on solid phase is related to the con-
determine the kinetic data. However, in determination of centration of BPA in the solution by the mass balance equation
metastability range, extent of breakage and permissible cool- (Randolph and Larson, 1988):
ing rate where no sampling during the course of experiment
was necessary the rotating bath was not used. A rotating speed dC̃BPA dM̃T
+ k1 =0 (5)
of 24 rpm for bath was selected to ensure uniform suspension dt dt
of the crystals in the bottles.
Seed crystals were chosen between two successive sieves where C̃BPA is the mass of BPA solute and M̃T is the mass of
of 500 and 1000 ␮m as shown in Fig. 6. Since the metastability adduct crystals in the whole suspension in a bottle crystallizer.
of solution had already been determined precisely the dissolu- Since the adduct crystal is an equimolar compound of BPA and
tion of seed was expected to be insignificant due to adjustment phenol, the constant k1 is
of operation conditions. But of course, the secondary nucle-
MBPA
ation and growth were dominant mechanisms. Use of fully k1 = (6)
Mphenol + MBPA
sealed screw-lid bottles, high boiling points of phenol (181 ◦ C)
and BPA (220 ◦ C), and a low experienced temperature range of where MBPA and Mphenol are the molecular weight of BPA and
70–45 ◦ C ensured no evaporation of solution. phenol, respectively. Initial conditions for the mass balance
Washing liquid: In order to find a liquid suitable for crystal equation are
washing which would not alter the adduct structure, a small
sample of crystals was divided to several parts, and each part C̃BPA (t = 0) = C̃0 BPA (7)
was suspended for 15 min in one of the washing liquids of n-
hexane, n-heptane, n-pentane, water, and cyclohexane; and M̃T (t = 0) = M̃T0 (8)
then filtered and dried. The melting point of each solid sam-
ple was measured in order to realize which one had the same where C̃0 BPA is the initial mass of dissolved BPA in the whole
melting point as that of the BPA-adduct (98–100 ◦ C). A change liquid phase in the bottle and M̃T0 is the total mass of initial
in melting point gives an indication of damage to the crystal seed in the bottle. Similarly, a mass balance equation with
structure. initial conditions for phenol was written.
Analysis: Particle size distribution of product from each A macroscopic population balance for the adduct crystals
bottle crystallizer was determined using the sieve analysis in a batch crystallization process in the absence of agglomer-
method and the concentration of BPA in filtrate of bottles ation and breakage is (Randolph and Larson, 1988):
withdrawn from the bath at different sampling times was
measured using the gas chromatography method. The nuclei ∂ñ(v, t) ∂ñ(v, t)
+ Gv =0 (9)
size, defined as the smallest detectable size, was 44 ␮m, and ∂t ∂v
the largest size in the distribution was 1000 ␮m in the experi-
where Gv is the growth rate defined based on particle volume
ments of this study. In order to have an insight into the crystal
as sized, and ñ(v, t) is the whole population of crystals in the
morphology the method of Scanning Electron Microscopy
size range of v to v + dv in the bottle.
(SEM) was performed utilizing an Oxford 5526 (Link Pentafet)
The initial and boundary conditions for the population bal-
microscope operating at 20 kV. The volume shape factor of
ance equation, respectively, are
adduct crystals was calculated through counting and weighing
2105 crystals in a narrow size range in the distribution taken
ñ(v, t = 0) = ñseed (v) (10)
between two successive sieves in the analysis.

(B0 Mfree solvent ) B̃0

ñ(vmin , t) = = (11)
Gv Gv
3. Modeling
ñ(vmax , t) = 0 (12)
Nucleation and growth rates of crystallization process were
represented using semi-empirical equations of (Randolph and where B̃0 is the rate of generation of smallest detectable size
Larson, 1988): particles vmin in the bottle crystallizer, and vmax is the largest
size in the distribution. The rate of growth based on particle
 −E  volume is
n
B0 = kn0 exp Cb MTm (3)
RT 1/3
Gv = 3kg kv Cg v2/3 (13)

 −E 
g where kv is the volume shape factor of particles. The mass
G = kg0 exp Cg (4)
RT and population balance equations were solved simultane-
 chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623 1619

Fig. 4 – The metastable zone in crystallization of BPA adduct

from the pure Nourafkan et al., 2006; phenol solution.

analysis. It is notable that nt did not include the last bottle

Fig. 3 – Accuracy of the numerical solution with changes in
used for a test of model verification.
the value of time increment.
The objective function represents the difference between
the model predicted mass distribution of adduct particles
ously using the implicit method of Crank–Nicolson (Gerald
and the experimentally measured mass distribution. The dif-
and Wheatley, 1989). The kinetic parameters were estimated
ference between the model predicted concentrations and
by best fitting the model predictions to the experimental
the experimentally measured concentrations by GC was also
data.
included in the objective function. Thus, the information of
Since most of the particles in the distribution were in the
both liquid and solid phases was used to optimize the kinetic
small size ranges, logarithmic intervals in size were chosen
parameters used in the rate equations of nucleation and
to perform the calculations. The size range 44–1000 ␮m were
growth. The genetic algorithm procedure was employed in this
divided to 100 incremental intervals (log v = 0.0407).
optimization process.
Parameter m, the nucleation rate order with respect to MT ,
in Eq. (3) is generally close to 1 due to a linear relation between
the number of inter-crystal contacts leading to secondary 4. Results and discussion
nucleation and the magma density. Therefore, at the present
study the value of parameter m was fixed to 1.17 according to Metastability: The experimental results of measuring the width
research of Moyers (1986). of metastable zone of BPA–phenol solution are shown in Fig. 4.
Due to nonlinear nature of growth and nucleation, the The solubility and labile curves of BPA-adduct seem to be par-
analytical solutions for the population balance equations gen- allel and about 7 ◦ C apart from each other. A relatively wide
erally have not been reported. However, for some limited zone of metastability measured indicates that maintaining the
special cases of population balance equation, the analytical temperature inside the zone is manageable during the process
solutions are available (Gelbard and Seinfeld, 1978). The ana- of cooling, and a large value of supersaturation may be applied
lytical solutions for these cases were used in the present without risk of approach to the labile zone which leads to pri-
study to evaluate the numerical solutions. As the time inter- mary nucleation. The effect of cooling rate on metastability
val in numerical solutions decreases, the accuracy of solution was not investigated in this study.
increases; on the other hand, the CPU time consumed for cal- Breakage: The mass mean size of particles before and after
culation increases. To find an optimum value for time intervals 3-h-agitation in the glass vessel of breakage experiment was
an objective function based on the difference between the ana- 388.8 and 328.5 ␮m, respectively. The extent of reduction in
lytical and numerical solutions was defined. Fig. 3 shows the mass mean size (15%) indicated that the breakage due to the
variations of this objective function with the changes in time particle impact to the propeller and to other particles during
interval values. The reduction in time interval to less than 60 s the course of experiment was trivial. However, this extent of
did not increase the accuracy in the numerical solution. There- breakage in rotating bottle crystallizers in which no agitator
fore, an interval of 60 s was chosen to be used in numerical was used is expected to be entirely prevented. It is notable that
calculations of the present study. the hydrodynamic conditions in both vessels are not compa-
Estimation of parameters: In order to find the optimum values rable, but the probability of breakage for bottles in rotating
of kinetic parameters, some arbitrary values for parameters bath used in the crystallization experiments is substantially
were initially guessed and the product size distribution were less than that in vessel of Fig. 1.
calculated by the model using these assumed values of param- Cooling rate: Results of cooling rate experiments aiming at
eter. The objective function which its value was minimized in finding the rate at which the primary nucleation was min-
order to fit the model predictions to the experimental data was imum are shown in Fig. 5. As the duration time increased,
defined as: or consequently the average cooling rate decreased, the
total number of particles decreased due to lower levels of
   
  1   M̃exp (i, j) − M̃model (i, j)  1   C̃exp (i) − C̃model (i) 
nt nm nt

fobj kg0 , kn0 , ka , g, b, a, En , Eg =  ×   (14)

nt × nm  M̃exp (i, j)  nt  C̃exp (i) 
i=1 j=1 i=1

where nt is the number of bottles removed from the bath as supersaturation and lower nucleation rates. Results suggested
samples and nm is the number of sieves used in particle size that an average cooling rate of less than 4.6 ◦ C/h was suitable
1620 chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623

Fig. 5 – Oversize cumulative number of seed and product

crystals in adductive crystallization of BPA from pure
phenol in the glass crystallizer at the cooling rates of 12.3,
7.4, 4.6 ◦ C/h and an agitation rate of 340 rpm.

for succeeding runs. This value of cooling rate corresponded

to duration time of more than 4 h for a cooling range from 68.5
to 45 ◦ C.
Washing liquid: Results of washing experiments showed
that the adduct crystals washed with n-heptane, n-hexane,
and n-pentane had a melting point close to that of the equimo-
lar adduct, but the adduct crystal washed with water showed
Fig. 7 – BPA–phenol adduct crystals product of
a melting point close to that of BPA. However, the adduct
crystallization from pure phenol using the rotating bath
crystals washed with cyclohexane had a range of melting
under magnification.

points from that of the equimolar adduct (98–100 ◦ C) to that

of BPA (155–157 ◦ C). The variations in melting point were
most likely due to washing out of phenol molecules by cyclo-
hexane and water and destruction of the crystal structure
of adduct through the washing operations. Additionally, a
GC analysis of an adduct sample washed with n-heptane
resulted in an equimolar BPA–phenol compound. As a result
n-pentane, n-hexane, and n-heptane are recommended as
appropriate liquids for washing the adduct crystals. A suit-
able washing agent is expected to wash out the solution
on the crystal surface but not to damage the crystal struc-
ture.
SEM observations: The coarse adduct particles grown more
and less lengthy and cylindrical are shown in Fig. 7. The
appearance of particles shown and also more particles not
shown do not suggest polymorphism for the adduct crys-
tals of BPA. It seems unlikely that the primary particles
have agglomerated to make a bigger crystal, but new par-
ticles have nucleated on the faces of the mother particles
and grown to some extent. These photomicrographs illustrate
almost smooth surfaces of crystal faces. However, the irreg-
ular patches on the single crystals are likely to be solidified
phenol not removed from the adduct faces despite thorough
washing operations by n-heptane.
Kinetic parameters: The values of parameters in the rate
equations of nucleation and growth for BPA adductive crys-
Fig. 6 – Mass distribution in adductive crystallization of BPA tallization from the pure phenol solution, calculated by model
from pure phenol during the course of experiments in the fitting to the experimental data when all the information in
rotating bath. Top: cumulative undersize. Bottom: four sets of data were used simultaneously in the optimiza-
differential. Corresponding temperatures have been tion process, are illustrated in Eqs. (15) and (16). We used
remarked. the order of 1.14 in relation B0 = kn G1.14 MT1.17 of Moyers (1986)
 chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623 1621

and thereby reduced the number of kinetic parameters by

one.

G = 5.72 × 1013 exp(−10400/RT)C2.7 (15)

B0 = 8.75 × 1017 exp(−10900/RT)C2.8 MT1.17 (16)

The agglomeration mechanism in adductive crystallization

of BPA from the pure solution was evaluated as negligible. A
few kernel functions were employed in the model and the
optimization process was performed. The agglomeration rate
coefficient suggested by the model for all cases was trivial.
It is pertinent to note that a genetic algorithm of opti-
mization procedure was employed to estimate the kinetic
parameters by minimizing the difference between the model
predictions at different times and the corresponding experi- Fig. 8 – Comparison of mass distribution of product
mental data. predicted by the model beyond the time range at which the
Relatively high orders of growth and nucleation rates with model parameters were tuned with the corresponding
respect to supersaturation would suggest that the crystalliza- experimental data after 8 h.
tion mechanisms are highly sensitive to the driving force. The
higher order for nucleation indicates that operation at condi-
tions of a higher level of supersaturation would increase the while the product distributions included lots of particles
nucleation more than growth. This may lead to a reduction smaller than 400 ␮m. Generation of vast amount of fines would
in particle size average and creation of more fine particles suggest a prevailing secondary nucleation mechanism.
as Fig. 5 illustrates a higher total number of particles for a Accuracy: The transparency and turbidity points were
shorter total time of experiment corresponding to a higher checked a few times to detect the temperature extremes
rate of cooling and a higher level of supersaturation. in the metastability experiments. Additionally, the kinetic
The rate orders may be greatly influenced by the presence parameters were calculated due to four measurements of con-
of impurities even at an imperceptible concentration. How- centration and four measurements of particle size distribution
ever, in the synthetic liquors of the present study because of carried out on contents from four different bottles withdrawn
using pure feed materials the impurities are presumed to be from the bath at different times.
absent. In order to find the confidence intervals of the estimated
It is postulated here that the variation of growth rate for parameters, the solid and liquid phase information obtained
different crystal faces is influential in crystallization mech- from each bottle crystallizer was applied independently for
anisms; and therefore, the supersaturation order of growth parameter estimation; then four sets of kinetic parameters
rate perhaps be influenced by the form and habit of crystals. were calculated. The various values of each individual param-
Lengthy and cylindrical shape of the adduct crystals shown eter were used to calculate the average and confidence interval
in Fig. 7 suggests that an infinitesimal increase in crystal vol- (95% confidence level) for that individual parameter. Table 1
ume due to mass deposition may result in a marked increase shows these statistical properties for kinetic parameters of kg0 ,
in length of the adduct particles. On the other hand, the sieve kn0 , g, b, En , and Eg used in Eqs. (3), (4), (15) and (16).
analysis method used in the present study is more sensitive to Model verification: Model predictions were verified using
the length of particles than to their volume. This behavior and the product size distribution and liquor concentration expe-
its effects on size distribution will impose the model to detect rienced at the time beyond the range at which the model
a higher value for growth rate order. Furthermore, the order of was tuned. It is pertinent to note that during the course of
nucleation rate is affected by the surface area of crystals and experiment five samples were withdrawn from the bath. Size
therefore by the crystal habit. A lengthy crystal generates more distribution and concentration information collected from the
area for contact than a spherical crystal. These considerations first four samples were used in parameter estimation and
may clarify the relatively high values of orders calculated for information from the last sample was used for model veri-
the dependency of the rate equations to supersaturation. fication. A comparison between the model predictions of size
The cumulative and differential mass distributions of the distribution beyond the time range at which the model param-
adduct particles during the course of crystallization experi- eters were tuned, and the corresponding experimental data
ment are shown in Fig. 6. Size range of initial seed in the after 8 h is shown in Fig. 8. The relative error between the
rotating bath experiments for kinetic study was 500–1000 ␮m, model predictions of size distribution and the experimental

Table 1 – Statistical properties of kinetic parameters of BPA adductive crystallization from phenolic solution used in Eqs.
(3), (4), (15) and (16).
Parameter 365 min 397 min 430 min 458 min Standard deviation Confidence interval

kg0 8.97E+13 4.90E+13 8.18E+13 6.80E+13 1.89E+13 7.21E+13 ± 3.01E+13

kn0 6.84E+17 4.52E+17 8.33E+17 5.67E+17 1.63E+17 6.34E+17 ± 2.59E+17
g 2.6 2.8 2.7 2.6 0.112 2.675 ± 0.178
b 2.7 2.9 2.8 2.7 0.096 2.775 ± 0.152
Eg 11,060 10,303 10,443 10,609 328.88 10,603 ± 523
En 11,080 10,495 10,652 10,831 251.16 10,764 ± 400
1622 chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623

where nm is the number of sieves used in the size analy-

sis.
The model predictions of BPA concentration in the solu-
tions were collected at the corresponding times of sampling
in the experiments. Comparison of the model predictions and
the experimental values of concentrations is illustrated in
Fig. 9. The relative error between the concentration predicted
by the model for time 8 h and the corresponding experimental
concentration was calculated as 6% using the formula:

 
 C̃exp − C̃model 

Error =   (18)
C̃exp 

An acceptable error of 6% between the concentrations pre-

dicted by the model beyond the range at which the model was
tuned and the corresponding experimental concentration at
8 h revealed the model reliability.
Fig. 9 – Comparison of the model predictions of BPA
Sensitivity analysis: The sensitivity of the objective func-
concentrations in solution with BPA concentrations in
tion to the kinetic parameters was examined by changing
filtrates of bottles removed from the rotating bath at
each parameter around its optimum value when the other
different sampling times.
parameter remained fixed and observing the corresponding
data was about 7%. The following equation was used to calcu- values of objective function. Fig. 10 shows that the objective
late the relative error: function is sensitive to nucleation and growth parame-
ters.
 
1   M̃exp (j) − M̃model (j) 
nm

Error = (17)
nm  M̃exp (j)  5. Conclusions
j=1

The metastability zone in the process of BPA adductive crystal-

lization from the pure phenolic solution was experimentally
explored. The metastability zone width of about 7 ◦ C sug-
gested that maintaining the concentration inside the zone
would be practically feasible in the both industrial and exper-
imental practices. Crystal breakage through crystal–crystal
contacts and crystal-impeller impacts seems insignificant
since minor reduction in the mass mean size of adduct par-
ticles during the breakage examination was observed. Trials
of cooling intensity suggested that prevention of excessive
nucleation necessitates an average cooling rate of less than
4.6 ◦ C/h.
Washing experiments suggested that n-pentane, n-
hexane, and n-heptane are suitable liquids for washing the
BPA-adduct crystals. However, water and cyclohexane are
likely to be destructive to crystal structure of the adduct parti-
cles and will remove the phenol molecules from the structure
in the washing operations.
Generation of vast amount of fines in rotating bath exper-
iments would suggest a prevailing secondary nucleation
mechanism in adductive crystallization of BPA.
Using mass and population balance equations, adductive
crystallization of BPA from the phenolic solution was mech-
anistically modeled. Kinetic parameters of nucleation and
growth rate equations were estimated by fitting the model
predictions to the experimental data using the optimiza-
tion procedure of genetic algorithm. Optimization process
Fig. 10 – Sensitivity of the objective function with changes
of the parameters suggested the equation of B0 = 8.75 ×
in kinetic parameters around their optimum values for
1017 exp(−10900/RT)C2.8 MT1.17 for nucleation and the equa-
crystallization of BPA from the pure phenolic solution: (a)
tion of G = 5.72 × 1013 exp(−10400/RT)C2.7 for growth of
kg0 = 5.72 × 1013 , the nucleation parameter was changed; (b)
adductive crystallization of BPA from the pure phenolic solu-
kn0 = 8.75 × 1017 , the growth parameter was changed. Note
tion.
that the objective function is represented in Eq. (14) and
Close agreement of the model predictions beyond the time
units of the parameters are: [kg0 ] = ␮m s−1 (g
range for which the parameters were tuned with the exper-
(gfree phenol )−1 )−g and [kn0 ] =
−b −m
imental data indicated enough validity of the mechanistic
(# g−1
free phenol
s−1 ) (gBPA g−1
free phenol
) (gcrystal g−1
free phenol
) . model developed in this study.
liquid solid
 chemical engineering research and design 8 8 ( 2 0 1 0 ) 1615–1623 1623

Acknowledgements Kim, K.J. and Mersmann, A., 2001, Estimation of metastable zone
width in different nucleation processes. Chem. Eng. Sci., 56:
2315–2324.
Mr Iman Noshadi’s help in performing the experiments, Mr
Konrad, F.M., 1975, in Hancock, E.G. (ed). (John Wiley & Sons, New
Hojat Kordabadi’s help in removing the program error, and the York).
financial support from Iranian National Petrochemical Com- Kosaka, Y. and Sinclair, K.B., (1982). Bisphenol-A from Phenol and
pany (NPC-RT) are gratefully acknowledged. Acetone with an Ion Exchange Resin Catalyst-Union Carbide
Technology. (SRI International Menlo Park, California).
Moyers, C.G., Jr., & Union Carbide Corp., 1986, Industrial
References crystallization for ultrapure products. Chem. Eng. Prog., 82(5):
42–46.
Gelbard, F. and Seinfeld, J.H., 1978, Numerical solution of the Mullin, J.W., (2001). Crystallization. (Butterworth-Heinemann), p.
dynamic equation for particulate systems. J. Comput. Phys., 123
28: 357–375. Nourafkan, E., Alamdari, A. and Jahanmiri, A., 2006, Bisphenol-A
Gerald, C.F. and Wheatley, P.O., (1989). Applied Numerical Analysis adductive crystallization under the influence of 4-tert-butyl
(fourth ed.). (Addison-Wesley Publishing Company, phenol, In 11th Iranian National Conf. on Chem. Eng. 28–30
Sydney). November 2006, Tehran, Iran, (p. 05135).
Hawley, G.G., (1981). The Condensed Chemical Dictionary (10th ed.). Randolph, A.D. and Larson, M.A., (1988). Theory of Particulate
(Van Nostrand Reinhold Company Inc). Processes (2nd ed.). (Academic Press, London).