Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Computers and Chemical Engineering 112 (2018) 292–303

Contents lists available at ScienceDirect

Computers and Chemical Engineering


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

Multiobjective optimization and experimental validation for batch


cooling crystallization of citric acid anhydrate
K. Hemalatha a,b, P. Nagveni a,b, P. Naveen Kumar a, K. Yamuna Rani a,b,∗
a
Process Dynamics and Control group, Chemical Engineering Department, CSIR-Indian Institute of Chemical Technology, Hyderabad 500007, India
b
Academy of Scientific and Innovative Research (AcSIR), CSIR-Indian Institute of Chemical Technology, Hyderabad 500007, India

a r t i c l e i n f o a b s t r a c t

Article history: Multiobjective optimization (MOO) of crystallization systems is gaining importance due to its ability to
Received 21 December 2017 handle multiple conflicting objectives together for finding optimal operating policies. The present study
Revised 9 February 2018
focuses on batch cooling crystallization of citric acid. Among the two forms of citric acid, citric acid
Accepted 19 February 2018
anhydride (CAA) is chosen for experimentation as no such study is available. MOO is carried out to
Available online 22 February 2018
seek optimal cooling policy for unseeded cooling crystallization of CAA to maximize mean crystal size
Keywords: while minimizing variance in size. In this procedure, temperature is discretized using piecewise constant-
Multiobjective optimization control vector parameterization which is simple and convenient for practical implementation. The model
Batch cooling crystallization reported in literature is suitably modified for solubility parameters which are verified experimentally, and
Citric acid anhydrate employed for optimization. One of the optimal solutions from the Pareto solution set is implemented
Crystal size distribution through experimentation successfully and the measured product crystal properties are comparable to the
Optimal cooling policy
predicted results obtained through optimization.
Experimental implementation
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction ers can be achieved through supersaturation control by manipulat-


ing the temperature with the help of defined performance criteria
Crystallization processes have been widely used for production, throughout the batch. Thus, for batch cooling crystallization pro-
purification or recovery of solid materials in almost all process cesses, it is necessary to determine an optimal temperature trajec-
industries. In particular, batch crystallization is primarily used in tory satisfying performance objectives through which a desirable
pharmaceutical, food, specialty and fine chemical industries. The product quality is obtained.
solids obtained through crystallization are further processed to Determination of optimal model-based control policies has
yield powders, pastes, dispersions, tablets, etc. Thus the control of been widely investigated in the literature. The concept of pro-
crystal properties such as size, shape, morphology, and purity that grammed cooling of batch crystallizers has been introduced by
decides the quality and solid characteristics of the final product is Mullin and Nyvlt (1971) who proposed that a significant increase
of major concern in batch crystallization as it has significant ef- in the crystal size is obtained by implementing a cooling profile
fect on the ease of down-stream processing such as filtration and obtained for constant nucleation rate throughout the batch. This
drying. fact has been emphasized by several researchers who ascertained
Considerable amount of research has focused on optimization improvement in crystal size with controlled cooling compared to
and control of batch cooling crystallizers including online monitor- different temperature profiles including natural and linear cooling
ing that are discussed in recent review papers (Nagy and Braatz, curves (Mullin and Nyvlt, 1971; Jones and Mullin, 1974; Mayrhofer
2012; Nagy et al., 2013; Ulrich and Frohberg, 2013; Samad et al., and Nyvlt, 1988; Rohani and Bourne, 1990; Bohlin and Rasmuson,
2013; Simon et al., 2015). Majority of the studies have focused on 1992; Xie et al., 2001). Several studies have focused on optimal
control of crystal size distribution (CSD) or shape, as these crystal seeding in a batch crystallizer (Chung et al., 1999; Ma et al., 1999;
properties strongly influence the product quality, functionality and Chung et al., 20 0 0; Zhang and Rohani, 2003; Hu et al., 2005; Costa
processing (Doherty, 2007). The optimal control of batch crystalliz- and Filho, 2005; Shi et al., 2006; Choong and Smith, 2004; Patience
et al., 2004; Aamir et al., 2010). Seed properties such as seed size
and mass and seed loading (Kubota et al., 2002) are also proven

Corresponding author: Process Dynamics and Control group, Chemical Engineer- to affect final crystal size distribution in addition to tempera-
ing Department, CSIR-Indian Institute of Chemical Technology, Hyderabad 50 0 0 07,
ture and have also been optimized. Substantial research activity
India.
E-mail address: kyrani@iict.res.in (K.Y. Rani).
has been devoted to closed loop temperature/concentration control

https://doi.org/10.1016/j.compchemeng.2018.02.019
0098-1354/© 2018 Elsevier Ltd. All rights reserved.
K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303 293

studies including classical control approach and predictive control ter) addition profile has been sought with respect to maximum
approach for control of crystal size distribution (Rawlings et al., mean size and high yield through simulation and also validated
1993; Miller and Rawlings, 1994; Matthews and Rawlings, 1998; experimentally. They even compared the results of MOO with sin-
Zhang and Rohani, 2003; Fujiwara et al., 2002; Lewiner et al., gle objective optimization and suggested that better crystal growth
2002; Fujiwara et al., 2005; Worlitschek and Mazzotti, 2004; Shi and yield are achieved through MOO. Acevedo et al. (2015) have
et al., 2005; Shi et al., 2006; Christofides et al., 2007, 2008; Yu implemented a multiobjective optimization framework for size and
et al., 2007; Nagy et al., 2008; Nagy, 2009; Aamir et al., 2009; shape optimization of two systems, potassium dihydrogenphos-
Hermanto et al., 2009; Mesbah et al., 2011; Kwon et al., 2014; phate and paracetamol. The optimal cooling policies have been de-
Kwon et al., 2015; Sheikholeslamzadeh and Rohani, 2013). Few termined with mean size and target aspect ratio as objectives for
experimental studies have been reported for implementation of unseeded cooling crystallization. Hemalatha and Rani (2017) have
optimal temperature profiles for some case studies. Jones and studied MOO of batch cooling crystallization of unseeded parac-
Mullin (1974) compared different cooling curves experimentally for etamol crystallization and seeded potassium nitrate crystallization
potassium sulfate-water system and found that the programmed using maximization of mean size and minimization of CV as the
cooling with constant nucleation control and metastable control objectives in the unseeded case and, additionally, minimization of
showed improvement in product crystal size compared to natu- nucleated mass in the seeded case. Yet, research studies consider-
ral cooling. Similarly, Mohameed et al. (2002) have also imple- ing multiple objectives in the context of batch cooling crystalliza-
mented three different cooling curves and studied the effect of tion are less explored compared to traditional single objective op-
cooling rate on crystal size distribution for potassium chloride- timization studies. Further, though several simulation studies are
water system to choose an optimal trajectory. Worlitschek and reported on batch crystallization of citric acid anhydrate (CAA)
Mazzotti (2004) optimized the temperature trajectory for the batch (which has applications in food, beverage and pharmaceutical in-
cooling crystallization of paracetamol from ethanol with the fi- dustries), the process has not been studied in the context of MOO
nal particle size distribution defined as the control objective. and the experimental validation for such process has also not been
Aamir et al. (2012) proposed a novel targeted direct design ap- explored. Thus, crystallization of citric acid anhydrate is chosen as
proach to systematically design different cooling profiles and hence the topic of interest in the present work.
batch times, with the objective being the desired shape of prod- In the present study, unseeded batch cooling crystallization of
uct CSD and validated experimentally for potassium dichromate- citric acid anhydrate is considered for multiobjective optimization
water system. Significant research on batch cooling crystallizers fo- and experimental implementation. In this work, maximizing aver-
cused on finding the optimal cooling profiles in order to improve age crystal size and minimizing coefficient of variation are consid-
the process performance using a single objective pertaining to the ered as objectives. Average crystal size, expressed in terms of num-
product size distribution (Chung et al., 1999; Chung et al., 20 0 0; ber mean size, and coefficient of variation are considered based
Choong and Smith, 2004; Costa and Filho, 2005; Ma et al., 1999; on the suitability for measurement according to our recent work
Mohameed et al., 2003; Ward et al., 2006; Hounslow and Reynolds, (Hemalatha and Rani, 2017). In this recent work, MOO of unseeded
2006; Hsu and Ward, 2013; Tseng and Ward, 2017) while incor- and seeded batch cooling crystallization are studied through simu-
porating process model equations and temperature magnitude and lation only for Paracetamol and Potassium nitrate systems respec-
rate limits, and optionally minimum yield, as constraints. However, tively, whereas in the present study, first attempt is made to imple-
there has been a lot of interest in seeded cooling crystallization ment the optimal cooling policy that is determined through MOO
processes compared to unseeded crystallization processes. The op- through experimentation for citric acid anhydrate, and the results
timization or control objectives that are usually considered as mea- obtained are compared with simulation studies.
sures of product characteristics in the above cited literature are
large mean size, uniform distribution or minimum coefficient of 2. Experimental system
variation, desired shape of product size distribution, seed proper-
ties, shorter batch times or high yield. Some of these studies are This section discusses details regarding the system chosen i.e.
dedicated to choice and comparison of objective functions (Ward CAA including specific literature reported on its crystallization, sol-
et al., 2006; Hsu and Ward, 2013; Tseng and Ward, 2017) used in ubility characteristics of CAA, experimental setup, the procedure
batch cooling crystallization processes. used to carry out cooling crystallization in a batch crystallizer,
Majority of the literature cited above on cooling crystallization measurement and product characterization techniques used.
has concentrated on solving single objective optimization problems
for different crystallization systems. However, crystallization pro- 2.1. System Overview
cesses have multiple objectives that are conflicting in nature and
there exists a trade off between these objectives which makes it Citric acid is widely used in the food, beverage and pharmaceu-
necessary to consider multiobjective optimization to obtain op- tical industries to impact a clean, refreshing tartness. Its prime use
timal operating policies for better CSD. Thus, it is necessary to is as an acidulant, but it is also used as sequestrant of metal ions to
study multiobjective optimization (MOO) of batch cooling crys- give protection from the development of off-flavours and off-doors
tallization processes. Sarkar et al. (2006) have carried out MOO in certain food stuffs (Teixeira et al., 2012). Citric acid is also used
for the first time for simulation of seeded cooling crystallization in cosmetics and toiletries as buffer, and in a wide variety of in-
of potassium sulfate from water. Since then, the application of dustrial applications as a buffering and chelating agent. Citric acid
MOO to crystallization processes has received focus not only in is also a reactive intermediate in chemical synthesis. In addition,
cooling crystallization but also in reactive or antisolvent crystal- its carboxyl and hydroxyl groups permit the formation of a variety
lization processes. Sarkar et al. (2007) have carried out MOO for of complex molecules and reactive products of commercial interest
reactive crystallization processes through simulation to find op- (Soccol et al., 2006).
timal feeding policies for maximum weight mean size and min- Citric acid crystallizes in anhydrous and monohydrate forms.
imum coefficient of variation for barium sulfate-water system. The transition temperature of CAA to Citric acid monohydrate
Sheikhzadeh et al. (2008) and Trifkovic et al. (2009) have proposed (CAM) is first reported by Bennet and Yuill (1935) to be below
novel real-time dynamic optimal control methodology for seeded, 36.3 °C. Caillet et al. (2008) studied monitoring of solvent-mediated
anti-solvent crystallization of paracetamol from isopropanol and phase transition of citric acid anhydrate to monohydrate using in
water in an offline and online manner. The optimal antisovent (wa- situ Raman spectroscopy and reported the transition temperature
294 K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303

to be below 34 °C. According to several studies reporting the transi- and is diluted 10 times with distilled water. A sample of which
tion temperature of citric acid, it can be considered to be between is titrated against standard sodium hydroxide (1 N) solution stan-
34 °C and 36.5 °C (Nemdili et al., 2016). There has been an in- dardized with oxalic acid dihydrate (1 N), using phenolphthalein
creased interest in the study of crystallization of citric acid mono- (8-12 pH) as indicator in all the cases.
hydrate (CAM) compared to CAA through experimentation (Sikdar
and Randolph, 1976; Berglund and Larson, 1984; Bravi and Maz- 2.3. Cooling crystallization of citric acid anhydrate
zarotta, 1998a, b; Caillet et al., 2007a, b). However, several simu-
lation studies have been reported for crystallization of citric CAA The experimental unit consists of a jacketed glass vessel of ca-
where crystallization kinetics for CAM has been utilized to sim- pacity 250 ml that serves as a batch crystallizer. Stirring is pro-
ulate crystallization of CAA (Bohlin and Rasmuson, 1992; Choong vided using a pitched blade of 45 °angle agitator rotating at a con-
and Smith, 2004). Recent literature on crystallization of CAA in- stant speed of 300 rpm. The crystallizer is connected to a heating-
cludes experimental investigation of a seeded crystallization pro- cooling system or a bath circulator with temperature controller as
cess on a vibrated bed bath crystallizer by Teixeira et al. (2012) and shown in Fig. 1. Pt 100 resistance temperature detectors (RTD’s)
measurement of primary nucleation and growth parameters in a are used for online temperature measurement of the crystallizer
batch crystallizer by Nemdili et al. (2016). and the jacket. Also all the temperatures are recorded in a com-
puter connected through data acquisition interface using LABVIEW.
2.2. Solubility Studies The crystallizer is fed with water and solid CAA according to a cho-
sen initial concentration which is more than the saturation con-
Solubility of a solute in a specific solvent is a key step in the centration. The initial conditions are chosen after rigorous simu-
study of the crystallization processes. Solubility of CAA in water lation as explained in Section 5. The dissolution of CAA in water
and other solvents has been discussed in literature. However, in is carried out at a temperature higher than 70 °C and held con-
this study, the solubility of CAA in water has been verified ex- stant while stirring at a speed of 300 rpm for 60 min to ensure
perimentally first, to choose solubility data for crystallization ex- complete dissolution. The contents are quickly cooled to 62 °C and
periment in comparison with the reported literature. The solubil- held constant for 30 min. The crystallization experiment is carried
ity of citric acid anhydrate in aqueous solution is determined by out by implementing a cooling profile from 62 °C to 41 °C obtained
gravimetry and is also confirmed through titrimetry. Gravimetry through MOO. The total batch time is fixed at 2.5 h. As there is no
is the oldest and simplest analytical method used for determina- provision for automatic implementation of crystallizer cooling pro-
tion of solubility in any solvent. The basic steps involved in gravi- file in the experimental setup, the temperature profile is tracked
metric analysis are dissolution, filtration, drying and weighing. To by adjusting the jacket (coolant) temperature set point in accor-
prepare CAA solution, solid citric acid anhydrate is purchased from dance with the crystallizer temperature profile and the actual pro-
Alfa Aesar (purity ≥ 99.5), distilled water with PH of 6.5 serves as a cedure is explained in Section 5.3. At the end of the batch time,
solvent. Distilled water is filtered through vacuum filtration setup the slurry is filtered for product crystals that are later analyzed to
through 0.22 μm membrane filter. Saturated solutions are prepared determine crystal size distribution. The results obtained from this
in100ml beakers. Solubility is measured at temperatures ranging experiment are then compared with the predicted simulation data
from 40 °C to 60 °C in 10 °C increments. The temperature is con- obtained using multiobjective optimization.
trolled and maintained by thermostat baths, and the actual tem-
perature is validated by a digital calibration thermometer (Wika, 2.4. Measurement of crystal size distribution
Germany, uncertainty of ±0.05 °C). Syringes (25 mL) fitted with
0.44 μm membrane filters are used to sample 10 mL of solution The contents of the crystallizer are filtered using vacuum fil-
into pre-weighed crucibles or conical flasks. Syringes are preheated tration. The crystals are then washed with acetonitrile solution in
when necessary to prevent nucleation inside the syringes during which citric acid is poorly soluble and are dried overnight at 40 °C
sampling. The filters are also preheated to exceed the solution tem- in the oven. Following this, a solid sample is taken on a glass slide
perature. using a small spatula and the size is measured with an image anal-
Equilibrium is established by dissolution to saturation of solid ysis system, which consists of a table microscope, a digital camera
material. Three dissolution experiments at the same temperature and a computer equipped with Image analysis software. Images of
are conducted by adding an excess amount of commercial citric crystals are taken and properties of final product crystals are mea-
acid anhydrate in 20 ml water taken in 100 ml glass beakers. The sured using offline Image Pro Premier 9.1 software.
beakers are placed on a hot plate provided with magnetic stirring The parameter chosen for the crystal size characterization is the
for three hours. The temperature is monitored and kept constant diameter of the crystal along its major axis as observed by mi-
throughout the dissolution process. The samples are allowed to croscopy. The experimental results are presented in terms of num-
settle down by keeping them in a hot water bath maintained at ber average of crystal size and coefficient of variation as obtained
constant temperature for 24 h without disturbing them. For gravi- from the image analysis process. The number mean size L̄ is given
metric analysis, a sample of 5 ml each is withdrawn from the clear as:
supernatant solution into pre-weighed crucibles through a syringe 1 n
L̄ = Li (1)
provided with filter. The mass of the crucible with saturated so- N i=1

lution is recorded. Drying of samples is conducted by keeping the where Li is the individual crystal size and N is the number of crys-
crucible in a hot air oven at the constant solubility temperature for tals.
two days. The mass of the samples is recorded repeatedly through- Coefficient of variation (CV) is determined from standard devi-
out the drying process to establish the point at which no water ation (σ ) and mean size as
remains. Complete drying occurs when the mass of the sample re- σ
mains constant over time. The samples are weighed at this final CV = (2)

time when all the solvent is evaporated. The weight of the solid
is recorded for all the three samples. The average is taken as solu- 3. Simulation system
bility (kg. acid/kg. water) at that particular temperature. The re-
sults of this gravimetric analysis are confirmed through titrime- Determination of optimal cooling policy for unseeded batch
try also. A 5 ml sample of clear supernatant liquid is withdrawn cooling crystallization of citric acid anhydrate for implementation
K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303 295

Fig. 1. Schematic diagram of experimental setup for cooling crystallization of CAA.

to achieve better product CSD is of interest in this work. For this The concentration C is a function of the third moment μ3 , due
purpose, the dynamic model with crystallization kinetics reported to the crystal mass balance.
in the literature (Bohlin and Rasmuson, 1992) is chosen for sim-
dC d μ3
ulation. The model assumptions are: uniform mixing, size inde- = −ρc kv (10)
dt dt
pendent growth with no growth rate dispersion, no agglomera-
tion or breakage. The crystals formed initially in the process due where C is the solute concentration, ρ c is the density of the crystal,
to primary nucleation act as seeds for the remaining process and kv is the volumetric shape factor for a sphere crystal model and it
thus secondary nucleation also plays a role. The population bal- is assumed to be constant throughout the crystallization process.
ance equation (PBE) is solved using method of moments, which The solubility is a function of temperature which is expressed in
has been used historically by many researchers (Chung et al., 1999; the form of second order polynomial equation.
Choong and Smith, 2004; Shi et al., 2005; Shi et al., 2006), as it is C ∗ = C0 + C1 T + C2 T 2 + C3 T 3 (11)
simple and consumes less time during optimization, and is repre-
sented as follows: To obtain larger and fewer crystals it is desirable to suppress
nucleation and enhance growth. This can be achieved by control-
∂n ∂n
+ G(t ) =0 (3) ling the super saturation, which is the actual driving potential for
∂t ∂L both the nucleation and growth rates and is represented as a con-
This PBE is reduced to moment equations as centration difference C.
d μ0
=B (4) C = C − C ∗ (12)
dt
C∗ is solubility in g. of solute/g. of solvent and is a function of
dμ j temperature (T in K) as shown in Table 1.
= jGμ j−1 (5) Assuming constant volume, the amount of solute leaving the
dt
solution must be accounted for by crystal growth and nucleation
Boundary conditions: μ j = 0 at t = 0 j = 0, 1, 2, 3, 4, 5
and the mass balance of solute concentration is given as
where n represents the number density of crystals with length L
at time t. The total nucleation rate B is the sum of primary (BP ) dC
= −3ρc kv Gμ2 (13)
and secondary nucleation (BS ). Primary, secondary nucleation and dt
growth rate (G) are expressed in the form of power laws as follows. where ρ c is the density of the crystal and the solute concentration
C, kg acid/kg. solvent.
BP = kP (C − C ∗ )P (6) The crystallization kinetic parameters used by Bohlin and Ras-
muson (1992) are used for simulation of the CAA-water system.
BS = kS (C − C ∗ )S (7) The properties of CAA and parameters used for simulation of the
model are reported in Table 1. The temperature of the system is
decreased from 62 °C to 41 °C in a batch time of 2.5 h or 90 0 0 s fol-
B = BP + BS (8)
lowing any chosen cooling trajectory (to be discussed in Section 5).
The system of equations from Eqs. (1) to (13) is simulated where
G = kg (C − C ∗ )g (9) at first, the super saturation is calculated from Eqs. (12), and
296 K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303

Table 1
Values of Parameters/Variables used for simulation of the model.

Parameters/Variables Value Units

Total batch time tf 90 0 0 sec


Final temparature Tmin 41 °C
Initial temparature Tmax 62 °C
0.08412 + (0.09393)T
Solubility C∗ kg CAA/kg H2 O where T is in °C
−(0.00148)T 2 + (0.0 0 0 0110 0)T 3
Density of crystals ρ 1665 kg/m3
Volume shape factor kv 0.52 -
Primary nucleation rate constant kp 1.0 × 107 no/kg(kg/kg)−p
Primary nucleation rate exponent p 3.54 -
Growth rate exponent g 0.65 -
Secondary nucleation rate constant ks 0.88774 × exp(4781/T) no/kg(kg/kg)−p−1 (no/kg)−b where T is in K
Secondary nucleation rate exponent s 0.543 -
Empirical exponent b 0.84 -
Growth rate constant kg 0.02652 × exp(−3584/T) kg/kg, where T is in K
Heat of crystalization -H 206.09 kJ/kg
Heat capacity CP 1.2538 kJ/(kg.K)
Heat transfer coefficient UA 0.0216 kJ/(s °C)

Eqs. (6) to (9) are employed to calculate nucleation and growth 4.1. MOO problem formulation and solution methodology
rates. Subsequently, moment equations, Eqs. (4) and (5) are inte-
grated in order to find the final product crystal properties. The As the focus of the present problem is on the practical imple-
concentration for the next time instant is calculated through the mentation of the optimal cooling profile and to compare results
mass balance equation Eq. (13). During optimization (discussed in of experimentation and simulation, maximization of average crys-
Section 4), it is assumed that the crystallizer temperature is per- tal size and minimization of coefficient of variation are chosen as
fectly implemented (perfect control). For the purpose of control objective functions as these are popularly studied and can be mea-
calculations, Tj is used as manipulated input which is utilized for sured through experiment. Also, average crystal size is represented
implementation of optimal temperature trajectory and the energy in terms of number mean size and corresponding representation of
balance equation is coefficient of variation is considered.
The chosen objective functions are NMS and CV and they are
dT −UA   −H M
= T − TJ − 3ρ kv Gμ2 sol (14) represented by moments of CSD to simultaneously optimize with
dt MsluC p Cp Mslu respect to the temperature/cooling profile as a function of time.
The appropriate constraints are also incorporated. These individ-
where Msol is the mass of the solvent and Mslu is mass of the
ual objectives are optimized simultaneously (J1 = [J11, J12 ]) and the
slurry.
optimization problem is represented as follows with the objective
function J11, representing maximization of NMS and minimization
of CV is represented by the objective function J12 .
4. Multiobjective optimization
μ1
max J11 = (15)
In the face of the importance of the final CSD in the down- T (t ) μ0
stream processes and in product applications, the objectives of the  
optimization problems in crystallization are normally chosen ac- μ2 μ0
min J12 = −1 (16)
cording to specific features related to product quality and market T (t ) μ21
demands. The most common objective functions in crystallization
Subject to:
optimization problems are maximization of the mean crystal size
at the end of the batch, minimization of the standard deviation of c1 : Tmin ≤ T (t ) ≤ Tmax
the final CSD or minimization of its coefficient of variation (CV), dT
c2 : ≤0 (17)
and sometimes minimization of the batch time (Hemalatha and dt
Rani, 2016, 2017). A desirable change in one objective produces c3 : t ≤ t f
an undesirable change in another objective. This necessitates the where Tmax is the initial temperature and Tmin is the final temper-
application of MOO for crystallization processes. Many MOO stud- ature.
ies have dealt with multiple objectives by combining them into The temperature should always be a decreasing function of
one objective function composed of the weighted sum of indi- time for cooling crystallization throughout the batch. The inequal-
vidual objectives, or by transforming one objective into a single ity constraints ensure that the temperature profile can be imple-
response function while using others as constraints. Such an ap- mented. Temperature is the manipulated variable and it is usu-
proach requires prior knowledge of weights, which makes it dif- ally parameterized as a piecewise linear function with respect to
ficult to properly interrelate the properties and has the disadvan- time in earlier works related to MOO (Hemalatha and Rani, 2017;
tage of masking the physical significance of individual objectives. Acevedo et al., 2015; Sarkar et al., 2006). Experimental validation
Using proper MOO solution strategies, it is possible to consider of such kind of an optimal temperature profile sought through
multiple objectives all at a time and solve simultaneously (Sarkar MOO has been reported by Acevedo et al. (2015) where the tem-
et al., 2006; Hemalatha and Rani, 2017). No studies are reported on perature is controlled through automated environment. Such kind
MOO of cooling crystallization of CAA. In the present work, num- of automated environment is unavailable for experimental imple-
ber mean size (NMS) and number coefficient variation (CV) are the mentation for the present work. Hence, the temperature is cho-
two objectives considered for MOO of unseeded batch cooling crys- sen to be parameterized as a piecewise constant function which is
tallization of CAA and the optimization problem is formulated as simple and convenient for practical implementation in the present
follows. situation.
K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303 297

Fig. 2. Comparison of solubility data with literature.

The solution methodology using Non-dominated sorting ge- Table 2


Comparison of mean size and coefficient of variation of different cooling profiles.
netic algorithm (NSGA-II) as explained in an earlier work by
Hemalatha and Rani (2017) is chosen for MOO problem to obtain Profile NMS, μm CV, %
Pareto optimal set in the present study. In order to parameterize L 274.20 30.47
the input temperature trajectories T(t), the time interval (0, tf ) is P3 418.46 41.67
divided into P time stages of unequal length and piecewise con- P4 391.72 51.77
stant control policy is sought in each time interval (tk, tk +1 ). CP 364.45 48.02

T (t ) = T (k ) for t (k ) ≤ t < t (k + 1 ) (18)


5.2. Optimization of batch cooling crystallization of CAA

where T(k) is the temperature at the beginning of kth interval. Different polynomial cooling profiles reported in the literature
The optimal control problem then is to find T(k), k = 1,2…P. In that are commonly employed for simulations or implementation
the present study, P is considered as 11 between the temperatures are compared through simulations using the dynamic model for
Tmax and Tmin . The number of decision variables to the algorithm citric acid anhydrate detailed in Section 3. The cooling profile is
which are temperature variables is taken as 9, excluding Tmax and given by the following equation.
Tmin which are fixed initial and final temperatures. This temper- 
T (t ) = Tmax + (Tmin − Tmax )(t/t f )
a
ature profile is used as the input for MOO through NSGA-II. The (19)
time stages are considered according to time instants represented where exponent a = 1 results in a linear profile, a = 3 results in a
by t(k). The explanation for considering unequal time intervals for convex profile suggested for controlled cooling of seeded process,
optimization is given in Section 5.2. and a = 4 results in a convex profile suggested for controlled cool-
ing for unseeded process. The profiles for a = 3 and 4 are popularly
known as Mullin-Nyvlt profiles for constant rate of nucleation.
For unseeded cooling crystallization of citric acid anhydrate,
5. Results and Discussion Bohlin and Rasmuson (1992) reported that irrespective of cooling
profile followed, larger mean size is achieved at an initial super
5.1. Solubility study saturation of 0.1 kg/ kg. This has been confirmed in our simulations
and thus for the present work, initial super saturation is consid-
Solubility of CAA in water is determined at temperatures of ered to be 0.1 kg/ kg and initial concentration of CAA is chosen
40 °C, 50 °C and 60 °C respectively. The experimental solubility data accordingly. With C = 2.9403 kg/kg as the initial concentration of
is compared to those reported by several sources (Dalman, 1937; CAA, the simulations are carried out to study the effect of three
Bohlin and Rasmuson, 1992; Teixeira et al., 2012; Apelblat, 2014; different cooling profiles on final achievable mean size and coeffi-
Nemdili et al., 2016). Fig. 2 shows comparison of the experimental cient of variation. The results are shown in Table 2 and the corre-
results with the available literature. The solubility units reported sponding profiles are shown in Fig. 3. It is evident from Fig. 3 that
in the literature by different studies is converted to weight frac- the number mean size and CV obtained through Mullin-Nyvlt tra-
tion for the purpose of comparison. The experimental results ob- jectory with power 4 denoted by P4 is inferior to those obtained
tained are comparable to most of those reported in the literature. by other two profiles. The Mullin-Nyvlt profile with power 3, also
The present experimental solubility results are closer to the data known as a cubic cooling profile denoted by P3 gives better mean
reported by Teixeira et al. (2012). Hence, the solubility curve re- size (418.46 μm) compared to the other two profiles but inferior
ported in their work as shown in Table 1, is chosen for simulation value of CV (41.6%) compared to the linear profile denoted by ‘L’.
studies. This type of behaviour is concurrent with the results reported by
298 K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303

Fig. 3. Comparison of different cooling profiles.

Bohlin and Rasmuson (1992). For the linear profile, for an improve- found that this is comparatively inferior to the terminal mean size
ment of approximately 10% of CV, there is a decrease in mean and CV obtained for profile P3. Hence, an optimization is consid-
size by approximately 150 μm which is not desirable. Among the ered to be necessary with chosen objectives NMS and CV, to seek
three polynomial cooling profiles, for practical implementation, cu- an optimal cooling profile for practical implementation. The pop-
bic profile can be considered as the best operating profile with re- ular Non-dominated sorting genetic algorithm (NSGA-II) is used
spect to mean size and coefficient of variation. However, according for solving MOO problem. The solution methodology using NSGA-
to the optimization study reported by Choong and Smith (2004), II and genetic algorithm parameters reported by Hemalatha and
slight deviation from the cubic profile is obtained as the optimal Rani (2017) are used for the present MOO problem. 9 tempera-
profile through single objective optimization studies. They have ture variables (excluding initial and final temperature values) are
reported various single objective optimization studies with maxi- considered as decision variables for NSGA corresponding to the 11
mization of mean size, maximization of yield, minimization of co- time stages.
efficient of variation and minimization of batch time as different Unlike single objective optimization that results in a unique so-
objectives that yielded different optimal cooling curves for each lution, MOO results in a set of optimal solutions known as Pareto
case. It is concluded from their study that the objectives are con- optimal set or Pareto front. The trade off between the objectives
flicting in nature and a desirable change favouring one objective NMS and CV can be observed from the Pareto optimal set of MOO
results in an unfavourable change in the other objectives. Hence, as shown in Fig. 4.
multiobjective optimization is necessary for determination of an In order to explore an alternative method of arriving at the so-
optimal cooling profile with respect to specific objectives as it has lutions using weighted sum approach and to compare with the
the potential to consider trade off between multiple objectives. solutions of MOO, the single objective optimization (SOO) is con-
To carry out MOO of citric acid, mean size and coefficient of sidered with weighted functions of the two chosen objectives. The
variation are considered as objectives as these two objectives ex- following expression is used for SOO where the problem is to min-
hibited a trade off between each other in our simulation stud- imize the objective function J.
ies. A multiobjective optimization problem for unseeded cooling
J = (x × (−J11 ) ) + ( (1 − x ) × J12 ) (20)
crystallization is formulated with two objectives as discussed in
Section 4. Usually, the manipulated variable, temperature is param- By changing the values of the weighing factor ‘x’ (ranging from
eterized as a piecewise linear policy with respect to time. Due to 0.0 to 1.0), 25 optimization runs are carried out and the solutions
limitations in practical implementation of such a cooling profile in from these results are used to construct a Pareto set as shown in
terms of a variable set point at every sampling instant, piecewise the Pareto plot between the two objectives J11 (NMS) and J12 (CV)
constant policy is chosen where the total batch time is discretized in Fig. 5. Whereas a single run of MOO results in a Pareto solu-
into 11 unequal time stages corresponding to various temperatures tion set which consists of 100 optimal solutions (as the popula-
which can be read from the cubic profile: tion size in NSGA is chosen as 100) that show trade off between
t (k )={0, 1400, 2300, 3900, 5300, 6200, 6600, 7400, 8000, the two objectives. SOO solutions are obtained in the region be-
tween A and Din Fig. 5, whereas the MOO solutions are obtained
830 0, 870 0, 90 0 0}; between the region B and C. In the region of overlap (region BC),
A controlled cooling profile can be constructed with these time the solutions of SOO are slightly inferior to solutions of MOO. In
stages, without any optimization and compared with the other the region BA, for a small drop in CV value, NMS is drastically re-
polynomial profiles as shown in Fig. 3. When the unseeded pro- duced. On the other hand, in the region CD, for a small rise in
cess is simulated with the newly constructed profile CP, the ob- WMS, CV is increasing drastically. Therefore, these regions cannot
tained NMS and CV are 364.45 μ and 48.01% respectively and it is be considered to provide necessary trade off between the objec-
K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303 299

Fig. 4. Pareto solution set obtained through MOO.

Fig. 5. Comparison of optimization through weighted sum approach and MOO.

tives. Thus choosing MOO methodology over SOO using weighted Table 3
Comparison of objective function values for selective optimal solutions from Pareto
sum approach provides trade off solutions with minimal effort.
solution set.
From the Pareto set of MOO, three representative solutions of
Pareto set from Fig. 4, i.e. end point of the upper Pareto front (UP), Profile NMS, μm CV, %
end point of the lower Pareto front (LP) and the middle point (MP) UP 363.18 42.98
are chosen for comparison in terms of objectives in Table 3, the MP 341.02 37.21
corresponding optimal cooling profiles are shown in Fig. 6. It is LP 310.96 33.22

observed that the range of NMS pertaining to the Pareto set is 310-
363 μm and CV is 42-33%, and thus the constructed profile CP is
5.3. Tracking of optimal cooling profile for cooling crystallization of
not a part of Pareto solution set and is considered suboptimal. MP
CAA
is chosen to be the optimal Pareto point and the corresponding
optimal cooling profile is considered for practical implementation.
Unseeded cooling crystallization of citric acid anhydrate from
aqueous solution is carried out in a batch crystallizer set up ex-
300 K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303

Fig. 6. Comparison of cooling profiles for selected solutions from the Pareto set of MOO.

Fig. 7. Comparison of crystallizer temperature profiles through simulation and experiment.

plained in Section 2. The optimal cooling profile obtained through employed for simulations. The crystallizer temperature is consid-
MOO is considered as a set point trajectory for experimental val- ered as the set point and its optimal set point trajectory is tracked
idation. Here, the controlled variable is the crystallizer tempera- using the inlet jacket temperature (Tj ) as the manipulated input
ture. The contents of the crystallizer are maintained at an ini- using a PI controller. The controller parameters are chosen through
tial temperature of 62 °C. The heating-cooling system used for the trial and error method and are found to be: control gain Kc = 0.15
present work does not have a provision for direct control of the and time constant = 10 s. During the experiment, the temperature
crystallizer temperature (Tc ). The heating-cooling system has pro- data of the crystallizer, jacket inlet and outlet are recorded for ev-
vision to implement a given set point as the input jacket tem- ery 20 s throughout the batch. The bath temperature set point pro-
perature or bath temperature. Trial experiments indicated a differ- file is changed manually by giving step changes at specified inter-
ence of 1 °C between crystallizer temperature and bath tempera- vals. The crystallizer temperature obtained through experiment is
ture. Therefore, the set point temperature profile for jacket inlet plotted along with the optimal cooling profile generated through
bath temperature has been calculated from the determined opti- offline optimization as shown in Fig. 7. The actual bath tempera-
mal cooling profile. This trajectory so obtained has been used for ture is calculated as the average of jacket inlet and outlet tempera-
tracking. tures represented by T̄ j . The manipulated input temperature profile
For the purpose of comparison of experimental results with obtained through simulation and experiment are reasonably com-
simulations, Eq. (14) has also been included in the set of equations parable as shown in Fig. 8. The experiment is continued for 2.5 h
K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303 301

Fig. 8. Comparision of jacket temperature profiles through simulation and experiment.

Fig. 9. Image of Citric acid anhydriate crystals taken using optical microscope with 40X magnification.

Table 4 6. Conclusions
Comparison of results obtained through experiment and simulation.

Profile NMS, μm CV, % In the present study, it is attempted to carry out multiobjec-
Chosen optimal cooling policy (MP) simulation 341.02 37.21 tive optimization for cooling crystallization of citric acid anhydrate.
Experiment 334.99 35.01 Firstly, the solubility of citric acid in aqueous solution for differ-
ent temperatures is determined through experiments and is found
to be comparable with the data reported in the literature. Differ-
ent polynomial cooling temperature profiles have been evaluated
through simulation and it has been observed that cubic profile P3
and is terminated when the final temperature is reached at the end is found to be optimal with respect to larger mean size only. In
of the batch time. The final crystals are analyzed to measure mean order to find a cooling profile for experimental implementation,
size and variance. Fig. 9 shows an image of few of the experimen- the cubic temperature profile is approximated as piecewise con-
tally obtained final citric acid crystals taken using an optical mi- stant profile with unequal time intervals. Further, MOO optimiza-
croscope. The number mean size is measured to be 334.99 μm and tion problem is considered with NMS and CV as two objectives
coefficient of variation to be 35.01 %. From Table 4, it can be con- to find the piecewise constant crystallizer temperature profile. The
cluded that the experimentally obtained results are in good agree- obtained temperature profile is implemented in an experimental
ment with the predicted results.
302 K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303

crystallizer unit equipped with a heating cooling system. The sim- Hemalatha, K., Rani, K.Y., 2016. Sensitivity analysis of Pareto solution sets of mul-
ulation results indicate that NMS and CV obtained by implement- tiobjective optimization for a batch cooling crystallization process. Proceedings
of Indian Control Conference (ICC) IEEE 493–498.
ing cubic profile is found to be closer to the end point of lower Hemalatha, K., Rani, K.Y., 2017. Multiobjective optimization of unseeded and seeded
Pareto front, LP in MOO. Whereas, the middle Pareto point, MP is batch cooling crystallization processes. Ind. Eng. Chem. Res. 56, 6012–6021.
chosen for experimental implementation. The experimental results Hermanto, M.W., Chiu, M.S., Braatz, R.D., 2009. Nonlinear model predictive control
for the polymorphic transformation of L-glutamic acid crystals. AIChE J. 55 (10),
(334.99 μm, 35.01%) are found to be in good agreement with the 2631–2645.
optimal solution (341.02 μm, 37.21%) in terms of NMS and CV re- Hounslow, M.J., Reynolds, G.K., 2006. Product engineering for crystal size distribu-
spectively. This study illustrates the usefulness of MOO for cooling tion. AIChE J. 52 (7), 2507–2517.
Hsu, C.W., Ward, J.D., 2013. The best objective function for seeded batch crystalliza-
crystallization process in giving a trade off solution between two
tion. AIChE J. 59 (2), 390–398.
objectives. Hu, Q., Rohani, S., Jutan, A., 2005. Modelling and optimization of seeded batch crys-
tallizers. Comput. Chem. Eng. 29, 911–918.
Jones, A.G., Mullin, J.W., 1974. Programmed cooling crystallization of potassium sul-
phate solutions. Chem. Eng. Sci. 29, 105–118.
Kubota, N., Doki, N., Yokota, M., Jagadesh, D., 2002. Seeding effect on product crystal
Acknowledgement size in batch crystallization. J. Chem. Eng. Jpn. 35 (11), 1063–1071.
Kwon, J.S.I., Nayhouse, M., Christofides, P.D., Orkoulas, G., 2014. Protein crystal shape
and size control in batch crystallization: comparing model predictive control
The first author acknowledges the financial support from De- with conventional operating policies. Ind. Eng. Chem. Res. 53 (13), 5002–5014.
partment of Science and Technology (DST), India, under Women Kwon, J.S.I., Nayhouse, M., Orkoulas, G., Ni, D., Christofides, P.D., 2015. Run–
scientist’s program WOS-A, ref. no. SR/WOS-A/ET-40/2013. to-run-based model predictive control of protein crystal shape in batch crys-
tallization. Ind. Eng. Chem. Res., 54 (16), 4293–4302.
Lewiner, F., Févotte, G., Klein, J.P., Puel, F., 2002. An online strategy to increase the
average crystal size during organic batch cooling crystallization. Ind. Eng. Chem.
Res. 41, 1321–1328.
References Ma, D.L., Chung, S.H., Braatz, R.D., 1999. Worst-case performance analysis of optimal
batch control trajectories. AIChE J. 45, 1469–1476.
Aamir, E., Nagy, Z.K., Rielly, C.D., Kleinert, T., Judat, B., 2009. Combined quadra- Matthews, H.B., Rawlings, J.B., 1998. Batch crystallization of a photochemical: mod-
ture method of moments and method of characteristics approach for efficient eling, control, and filtration. AIChE J. 44 (5), 1119–1127.
solution of population balance models for dynamic modeling and crystal size Mayrhofer, B., Nývlt, J., 1988. Programmed cooling of batch crystallizers. Chem. Eng.
distribution control of crystallization processes. Ind. Eng. Chem. Res. 48 (18), Process. 24 (4), 217–220.
8575–8584. Mesbah, A., Huesman, A.E.M., Kramer, H.J.M., Van den Hof, P.M.J., 2011. A compari-
Aamir, E., Nagy, Z.K., Rielly., C.D., 2010. Evaluation of the effect of seed preparation son of nonlinear observers for output feedback model-based control of seeded
method on the product crystal size distribution for batch cooling crystallization batch crystallization processes. J. Process Control 21, 652–666.
processes. Cryst. Growth Des. 10 (11), 4728–4740. Miller, S.M., Rawlings, J.B., 1994. Model identification and control strategies for
Aamir, E., Rielly, C.D., Nagy, Z.K., 2012. Experimental evaluation of the targeted batch cooling crystallizers. AIChE J. 40, 1312–1327.
direct design of temperature trajectories for growth-dominated crystallization Mohameed, H.A., Abu-Jdayil, B., Al Khateeb, M., 2002. Effect of cooling rate on un-
processes using an analytical crystal size distribution estimator. Ind. Eng. Chem. seeded batch crystallization of KCl. Chem. Eng. Process 41, 297–302.
Res. 51, 16677–16687. Mohameed, H.A., Abdel-Jabbar, N., Takrouri, K., Nasr, A., 2003. Model-based opti-
Acevedo, D., Tandy, Y., Nagy, Z.K., 2015. Multiobjective optimization of an unseeded mal cooling strategy for batch crystallization processes. Chem. Eng. Res. Des.
batch cooling crystallizer for shape and size manipulation. Ind. Eng. Chem. Res. 81, 578–584.
54, 2156–2166. Mullin, J.W., Nývlt, J., 1971. Programmed cooling of batch crystallizers. Chem. Eng.
Apelblat, A., 2014. Properties of citric acid and its solutions. In: Citric Acid. Springer Sci. 26 (3), 369–377.
International Publisher, pp. 13–141. Nagy, Z.K., Chew, J.W., Fujiwara, M., Braatz, R.D., 2008. Comparative performance of
Bennett, G.M., Yuill, J.L., 1935. The crystal form of anhydrous citric acid. J. Chem. concentration and temperature controlled batch crystallizations. J. Proc. Control
Soci. (Resumed) 29, 130. 18, 399–407.
Berglund, K.A., Larson, M.A., 1984. Modeling of growth rate dispersion of citric acid Nagy, Z.K., 2009. Model based robust control approach for batch crystallization
monohydrate in continuous crystallizers. AIChE J. 30, 280–287. product design. Comput. Chem. Eng. 33 (10), 1685–1691.
Bohlin, M., Rasmuson, A.C., 1992. Application of controlled cooling and seeding in Nagy, Z.K., Braatz, R.D., 2012. Advances and new directions in crystallization control.
batch crystallization. Can. J. Chem. Eng. 70, 120–126. Annu. Rev. Chem. Biomol. Eng. 3, 55–75.
Bravi, M., Mazzarotta, B., 1998a. Size dependency of citric acid monohydrate growth Nagy, Z.K., Fevotte, G., Kramer, H., Simon, L.L., 2013. Recent advances in the moni-
kinetics. Chem. Eng. J. 70, 203–207. toring, modelling and control of crystallization systems. Front. Chem. Eng. Res.
Bravi, M., Mazzarotta, B., 1998b. Primary nucleation of citric acid monohydrate: in- Des. 91, 1903–1922.
fluence of selected impurities. Chem. Eng. J. 70, 197–202. Nemdili, L., Koutchoukali, O., Bouhelassa, M., Seidel, J., Mameri, F., Ulrich, J., 2016.
Caillet, A., Rivoire, A., Galvan, J.M., Puel, F., Fevotte, G., 2007a. Crystallization of Crystallization kinetics of citric acid anhydrate. J. Cryst. Growth. 451, 88–94.
monohydrate citric acid. 1. In situ monitoring through the joint use of Raman Patience, D.B., Dell’Orco, P.C., Rawlings, J.B., 2004. Optimal operation of a seeded
spectroscopy and image analysis. Cryst.Growth Des. 7, 2080–2087. pharmaceutical crystallization with growth-dependent dispersion. Org. Proc.
Caillet, A., Sheibat-Othman, N., Fevotte, G., 2007b. Crystallization of monohydrate Res. Dev. 8, 609–615.
citric acid. 2. Modeling through population balance equations. Cryst.Growth Des. Rawlings, J.B., Miller, S.M., Witkowski, W.R., 1993. Model identification and con-
7, 2088–2095. trol of solution crystallization processes: a review. Ind. Eng. Chem. Res. 32 (7),
Caillet, A., Puel, F., Fevotte, G., 2008. Quantitative in situ monitoring of citric acid 1275–1296.
phase transition in water using Raman spectroscopy. Chem. Eng. Process. 47, Rohani, S., Bourne, J.R., 1990. A simplified approach to the operation of a batch crys-
377–382. tallizer. Can. J. Chem. Eng. 68 (5), 799–806.
Choong, K.L., Smith, R., 2004. Optimization of batch cooling crystallization. Chem. Samad, N.A.F.A., Sin, G., Gernaey, K.V., Gani, R., 2013. A systematic framework for
Eng. Sci. 59, 313–327. design of process monitoring and control (PAT) systems for crystallization pro-
Christofides, P.D., Li, M., Mädler, L., 2007. Control of particulate processes: recent cesses. Comput. Chem. Eng. 54, 8–23.
results and future challenges. Powder Tech. 175 (1), 1–7. Sarkar, D., Rohani, S., Jutan, A., 2006. Multiobjective optimization of seededbatch
Christofides, P.D., El-Farra, N., Li, M., Mhaskar, P., 2008. Model-based control of par- crystallization processes. Chem. Eng. Sci. 61, 5282–5295.
ticulate processes. Chem. Eng. Sci., 63 (5), 1156–1172. Sarkar, D., Rohani, S., Jutan, A., 2007. Multiobjective optimization of semibatch reac-
Chung, S.H., Ma, D.L., Braatz, R.D., 1999. Optimal seeding in batch crystallization. tive crystallization processes. AIChE J. 53, 1164–1177.
Can. J. Chem. Eng. 77, 590–596. Sheikholeslamzadeh, E., Rohani, S., 2013. Modeling and optimal control of solution
Chung, S.H., Ma, D.L., Braatz, R.D., 20 0 0. Optimal model-based experimental design mediated polymorphic transformation of L-glutamic acid. Ind. Eng. Chem. Res.
in batch crystallization. Chemom. Intell. Lab. Syst. 50 (1), 83–90. 52, 2633–2641.
Costa, C.B.B., Filho, R.M., 2005. Evaluation of optimization techniques and control Sheikhzadeh, M., Trifkovic, M., Rohani, S., 2008. Real-time optimal control of an
variable formulations for a batch cooling crystallization process. Chem. Eng. Sci. anti-solvent isothermal semi-batch crystallization process. Chem. Eng. Sci. 63,
60, 5312–5322. 829–839.
Dalman, L.H., 1937. The solubility of citric and tartaric acids in water. J. Am. Chem. Shi, D., Mhaskar, P., El-Farra, N.H., Christofides, P.D., 2005. Predictive control of crys-
Soc. 59, 2547–2549. tal size distribution in protein crystallization. Nanotechnology 16 (7), S562.
Fujiwara, M., Chow, P.S., Ma, D.L., Braatz, R.D., 2002. Paracetamol crystallization us- Shi, D., El-Farra, N.H., Li, M., Mhaskar, P., Christofides, P.D., 2006. Predictive con-
ing laser backscattering and ATR-FTIR spectroscopy: metastability, agglomera- trol of particle size distribution in particulate processes. Chem. Eng. Sci. 61 (1),
tion, and control. Cryst. Growth Des., 2, 363–370. 268–281.
Fujiwara, M., Nagy, Z.K., Chew, J.W., Braatz, R.D., 2005. First-principles and direct de- Sikdar, S.K., Randolph, A.D., 1976. Secondary nucleation of two fast growth systems
sign approaches for the control of pharmaceutical crystallization. J. Proc. Control in a mixed suspension crystallizer: magnesium sulfate and citric acid water sys-
15 (5), 493–504. tems. AIChE J., 22, 110–117.
K. Hemalatha et al. / Computers and Chemical Engineering 112 (2018) 292–303 303

Simon, L.L., Pataki, H., Marosi, G., Meemken, F., Hungerbühler, K., Baiker, A., Tum- Tseng, Y.T., Ward, J.D., 2017. Comparison of objective functions for batch crystalliza-
mala, S., Glennon, B., Kuentz, M., Steele, G., Kramer, H.J.M., Rydzak, J.W., tion using a simple process model and Pontryagin’s minimum principle. Comp.
Chen, Z., Morris, J., Kjell, F., Singh, R., Gani, R., Gernaey, K.V., Louhi-Kulta- Chem. Eng. 99, 271–279.
nen, M., O’Reilly, J., Sandler, N., Antikainen, O., Yliruusi, J., Frohberg, P., Ulrich, J., Ulrich, J., Frohberg, P., 2013. Problems, potentials and future of industrial crystalliza-
Braatz, R.D., Leyssens, T., Stosch, M.V., Oliveira, R., Tan, R.B.H., Wu, H., Khan, M., tion. Front. Chem. Sci. Eng. 7, 1–8.
O’Grady, D., Pandey, A., Westra, R., Delle-Case, E., Pape, D., Angelosante, D., Ward, J.D., Mellichamp, D.A., Doherty, M.F., 2006. Choosing an operating policy for
Maret, Y., Steiger, O., Lenner, M., Abbou-Oucherif, K., Nagy, Z.K., Litster, J.D., Ka- seeded batch crystallization. AIChE J. 52, 2046–2054.
maraju, V.K., Chiu, M.S., 2015. Assessment of recent process analytical technol- Worlitschek, J., Mazzotti, M., 2004. Model-based optimization of particle size dis-
ogy (PAT) trends: a multi author review. Org. Proc. Res. Dev. 19, 3–62. tribution in batch-cooling crystallization of paracetamol. Cryst. Growth Des., 4,
Soccol, C.R., Vandenberghe, L.P., Rodrigues, C., Pandey, A., 2006. New perspectives 891–903.
for citric acid production and application. Food Tech. Biotech. 44 (2), 141–149. Xie, W., Rohani, S., Phoenlx, A., 2001. Dynamic modeling and operation of a seeded
Teixeira, G.A., Vieira, W.F., Finzer, J.R.D., Malagoni, R.A., 2012. Operational optimiza- batch cooling crystallizer. Chem. Eng. Comm. 187 (1), 229–249.
tion of anhydrous citric acid crystallization using large number of seed crystals. Yu, Z.Q., Chew, J.W., Chow, P.S., Tan, R.B.H., 2007. Recent advances in crystallization
Powder Tech. 217, 634–640. control: an industrial perspective. Chem. Eng. Res. and Des. 85 (7), 893–905.
Trifkovic, M., Sheikhzadeh, M., Rohani, S., 2009. Multivariable real-time optimal con- Zhang, G.P., Rohani, S., 2003. On-line optimal control of a seeded batch cooling crys-
trol of a cooling and antisolvent semibatch crystallization process. AIChE J. 55 tallizer. Chem. Eng. Sci. 58, 1887–1896.
(10), 2591–2602.

You might also like