Computers and Chemical Engineering 33 (2009) 1014–1021
Contents lists available at ScienceDirect
Computers and Chemical Engineering
journa l homepage: www.e lsev ier .com/ locate /compchemeng
olvent design for crystallization of carboxylic acids
runprakash T. Karunanithia, Charles Acquahb,1, Luke E.K. Acheniec,∗,hanthakumar Sithambaramd, Steven L. Suibd
Department of Civil Engineering, University of Colorado, Denver, CO, USADepartment of Chemical and Biomolecular Engineering, University of Connecticut, Storrs, CT, USADepartment of Chemical Engineering, Virginia Polytechnic Institute and State University, Blacksburg, VA, USADepartment of Chemistry, University of Connecticut, Storrs, CT, USA
r t i c l e i n f o
rticle history:eceived 9 March 2008eceived in revised form8 September 2008ccepted 1 November 2008vailable online 21 November 2008
eywords:
a b s t r a c t
Critical to crystallization chemical product design is the choice of an appropriate solvent. Traditionalmethods have focused on bench scale experiments using classes of solvents (e.g. polarity) with the dif-ferent classes giving rise to different crystal morphologies. However, there are instances where somesolvents belonging to a particular class give completely different morphology from other solvents in thesame class. There has been some modeling effort aimed at predicting crystal morphology. A major draw-back with some of these morphology prediction models is that they tend to be limited in application. Itis clear that the solvent selection, with respect to crystal morphology cannot be carried out efficiently by
rystallizationroduct designolvents
just experimentation or modeling alone. This paper outlines a systematic methodology which combinestargeted bench scale crystallization experiments, an efficient computer-aided molecular design (CAMD)approach and a database search approach for the design and selection of solvents for crystallization of
orphologyarboxylic acids
carboxylic acids.
. Introduction
Crystallization is used extensively as a purification and sep-ration process in industry due to its ability to provide highurity separations. For efficient downstream processing and prod-ct effectiveness, controlling the crystal size and shape distribution
s very critical (Ma, Tafti, & Braatz, 2002). This is especially true forrystals produced in the pharmaceutical industry. Crystal size andorphology affect the ease of separating, washing, drying, packag-
ng, handling and storage. In the case of pharmaceutical products,rystal morphology affects dissolution characteristics, bioavailabil-ty and the ease with which the crystals are compressed into tablets.actors such as temperature, mixing intensity, solvent, supersatu-ation and addition of seed crystals have been routinely used toontrol the size and morphology of product crystals (Jones, Davey,
Cox, 2005). Solvents have a strong influence on the morphol-gy of product crystals and hence the morphology can be modified
y changing the crystallization solvent. In view of the impor-ance of product crystal morphology, solvent selection/design
ethodologies need to incorporate this critical aspect. This paperroposes an approach that integrates ‘bench scale crystallization
experiments–computer-aided molecular design–solvent databasesearch’ to solve the complex problem of solvent design/selection forcrystallization processes. We believe such an integrated approachwould be more effective and useful.
Succinic acid crystals grown from aqueous solution are plate-like while those grown from isopropanol solution are needle-like(Davey & Whiting, 1982). Crystals of 1,4-di-tert-butylbenzene(DTBB) grown from polar methanol and ethanol have a tabularhabit with a length to width ratio (aspect ratio) of 1 while crys-tals grown from less polar alcohol such as isopropyl alcohol causedthe DTBB to crystallize in tabular habit with aspect ratio of 2. TheDTBB crystals produced from acetone were acicular with an aspectratio of 6 while crystals with high aspect ratio of about 10 weredeposited from hexane (Garti, Leci, & Sarig, 1981). Alizarin (1,2-dihydroxy-9,10-anthraquinone) crystallizes as very long needlesfrom acetone, acetonitrile, hexane, toluene, acetic acid and in all ofthese cases the aspect ratio was at least 10. If dry ethanol, methanolor propanol were used as solvent flat crystals having an isosceles tri-angular shape was obtained (Algra, Graswinckel, Van Enckevort, &Vlieg, 2005). 6-Aminopenicillanic acid crystallizes as thin diamond-shaped plates from pure solution while it crystallizes as elongated
plates or needle-like crystals from phenoxyacetic acid (Mwangi& Garside, 1996). Theoretical and experimental studies (Cang,Huang, & Zhou, 1998) of solvent effects on crystal morphology ofmeta-nitroaniline show that non-polar toluene produces needleswhile more polar solvents like ethanol and acetone produce plates.
Nmax maximum number of positions in a molecule.Ni number of times the first order group ‘i’ is present.Nj number of times the second order group ‘j’ is presentR universal gas constant = 0.008314 kJ/gmol KT temperature (K)Tb boiling point (K)Tbi contribution of ‘ith’ first order group to the boiling
pointTbj contribution of ‘jth’ second order group to the boil-
ing pointTm melting point (K)Tmi contribution of ‘ith’ first order group to the melting
pointTmj contribution of ‘jth’ second order group to the melt-
ing pointTf flash point (K)Tfi contribution of ‘ith’ first order group to the flash
pointVm
298 molar volume (cm3/gmol)xi mole fraction of component ‘i’�ij binary variable indicating whether the ith position
in a molecule has structural group j.�j valence of group j.ı solubility parameter (MPA1/2)ıH Hydrogen bonding solubility parameter (MPA1/2)ıhi contribution of ith group to the hydrogen bonding
solubility parameter˛i contribution of ‘ith’ group to LC50� viscosity (cp)�i contribution of ‘ith’ first order group to viscosity�j contribution of ‘jth’ second order group to viscosity�fusH heat of fusion (kJ/gmol)
Rbt&dcNvaoTftad
prfrIal2fd
devise morphology-related property constraints that are specific tocompounds belonging to that solute class. Except for morphology-related property constraints all other solvent property constraintsremain same irrespective of the solute class.
Table 1Selected dicarboxylic acids.
Solute ı (MPA1/2) 2D molecular structure
Succinic acid (SA) 29.339
Glutaric acid (GA) 29.408
Adipic acid (AA) 25.372
� i liquid phase activity coefficient of component ‘i’
ecrystallization of naproxen from various solvents revealed a trendetween solvent properties and particle morphology, suggestinghe influence of solvent in the process of crystal growth (Tomasko
Timko, 1999). Naproxen recrystallized from acetone yielded nee-les having high aspect ratio but when crystallized from methanolomparatively low aspect ratio crystals were produced. However,aproxen from benzene yielded almost plate shaped crystals withery low aspect ratio. Srinivasan, Sankaranarayanan, Thangavelu,nd Ramasamy (2000) studied the influence of organic solventsn the habit of 4-nitro-4-methyl benzylidene aniline (NMBA).hey obtained needles from non-polar n-hexane, rhombic plateletsrom carbon tetrachloride, TCE and toluene, dendrites frometrahydrofuran and ethanol, triangular platelets from methanolnd octahedral morphology from ethylacetate, acetone andi-methylformamide.
The morphology of ibuprofen, an important carboxylic acid ofharmaceutical relevance, has been studied extensively. Isomet-ic shaped crystals were observed when ibuprofen was grownrom polar ethanol whereas growth from non-polar ethyl acetateesulted in elongated platelets (Cano, Gabas, & Canselier, 2001).buprofen crystallized from ethanol and methanol was grain-likend those crystallized from isopropanol and hexane were needle-
ike (Garekani, Sadeghi, Badiee, Mostafa, & Rajabi-Siahboomi,001). Rasenack and Muller (2002) obtained plate-like ibupro-en crystals from alcohols while needles were obtained fromiethyl ether. Gordon and Amin (1984) and Bunyan, Shankland,
ical Engineering 33 (2009) 1014–1021 1015
and Sheen (1991) show a relationship between morphology ofibuprofen crystals and hydrogen bonding solubility parameterof solvents from which they are grown. Modeling studies pre-dict that ibuprofen crystallized from polar solvents is plate-likewhile when crystallized from non-polar solvents it is needle-like(Winn & Doherty, 2000). Hydrogen bonding solubility parameterof solvents has been used as a key property to design solventswhich yield desired ibuprofen crystal morphology (Karunanithi,Achenie, & Gani, 2006) and further experimental verification stud-ies (Karunanithi et al., 2007) have been conducted to validate thesolvent design results.
From the above discussion, it is clear that for many organicsolutes the type of solvent used for crystallization strongly influ-ences the final crystal morphology and moreover polarity playsan important role. This is especially true in the case of car-boxylic acids such as succinic acid and ibuprofen. Highly polarsolvents tend to produce crystals with low aspect ratio (iso-metric) and non-polar solvents seem to produce crystals withhigh aspect ratio (elongated). It has been proposed that hydro-gen bonding plays a critical role and hydrogen bonding solubilityparameter of solvents can be correlated to crystal morphol-ogy (Bunyan et al., 1991; Gordon & Amin, 1984; Karunanithiet al., 2006). While there is ample literature on the above,there is not a systematic study and comprehensive expla-nation of the phenomenon for a particular class of solutes.Here we consider solvent selection for dicarboxylic acids crys-tals.
First, the effect of solvents on crystal morphology of dicarboxylicacids is studied through a series of bench scale crystallizationexperiments and valuable insights gained from the experimen-tal study are subsequently used to devise solvent selectioncriteria—specifically morphology related solvent property con-straints. Then, state of the art computer-aided molecular design(CAMD) approach and solvent database search approach are used topropose optimal solvent candidates. It is worth noting that morpho-logical constraints developed through the experiments are specificto the solute class of aliphatic dicarboxylic acids. However, if thereis the need to design solvents for crystallization of a compoundthat belongs to a different solute class, bench scale crystallizationexperiments would have to be carried out for a few compoundsthat belong to that particular solute class. After which we need to
Carboxylic acids constitute a very important class of chem-cal compounds. Some well-known pharmaceutical compoundsuch as ibuprofen and aspirin are carboxylic acids. Four aliphaticicarboxylic acids were selected for the present study. The total sol-
bility parameter, ı, estimated using a group contribution methodConstantinou & Gani, 1994), and chemical structures of these car-oxylic acids are shown in Table 1.
Six solvents having widely varying polarity were selected for thistudy. The total solubility parameter, ı, hydrogen bonding solubility
Figs. 1–6. Optical mic
ical Engineering 33 (2009) 1014–1021
parameter, ıH, and viscosity, �, of each of these solvents are shownin Table 2.
2.2. Crystallization experiments
A suitable amount of solute was dissolved in 5 mL of solventat 25 ◦C until saturation. A constant supersaturation of 1.10 waseffected and the temperature of supersaturated solution raised
to 65 ◦C. Precipitation was reached by cooling to 25 ◦C over a 2-h period. The precipitated solution was kept overnight to allowmaximum crystal recovery. Then crystals were collected by vac-uum filtration and subsequently dried in a vacuum dessicator.Experiments were repeated for each solute in order to verify
roscope images.
A.T. Karunanithi et al. / Computers and Chemical Engineering 33 (2009) 1014–1021 1017
* Values of ıH and ı taken from Hansen (2002).** Viscosity, �, values at 25 ◦C were taken from Lide (2001).
he repeatability and consistency of morphology of the crystalsbtained.
.3. Crystal characterization
.3.1. Optical micrographs (OM)Optical micrographs of all the samples were taken in order to
acroscopically study the crystal size and shape and to calculatespect ratio.
.3.2. Powder X-ray diffraction (PXRD)PXRD patterns of the original and recrystallized solutes were
aken and were used to rule out potential formation of polymorphs.oinciding peak positions is indicative of same internal crystaltructure. If the peak positions do not coincide, then the crystalsave different structures and are polymorphs.
. Results
OM images of recrystallized dicarboxylic acids from differentolvents are shown in Figs. 1–6. Images are arranged in orderf increasing solvent ıH, with ethyl acetate having the lowestıH = 7.2 MPa1/2) and water having the highest (ıH = 42.3 MPa1/2).
The shape, habit or morphology of a crystalline material isefined by a 3-dimensional array in space. The infinite variety ofombinations in crystal morphology is a primary cause of subjec-ivity in habit description (Gordon & Amin, 1984). A statement of
orphology should therefore provide a description of the length ofach axis. Since a 3-dimensional description cannot be obtainedicroscopically, the simplest system of morphology description
nvolves providing lengths of x and z axes. This length quantificationf the crystal morphology is generally referred to as the aspect ratioAR) defined as the crystal length to width ratio (Gordon & Amin,984). The length is interpreted as the longest crystallographic axisz) while width is the dimension obtained when a 90◦ angle to theength axis is imagined (x). Using OM images, the mean AR and cor-esponding standard deviation of 40 different crystals for a givenample were calculated. A plot of mean AR vs. solvent hydrogenonding solubility parameter is shown in Fig. 7.
Crystal circularity provides another way of quantifying crystalorphology:
ircularity = 4�A
P2(1)
Here ‘P’ is the perimeter and ‘A’ is the area of the crystal. UsingEM images, the crystal circularity was computed with the helpf the public domain software ImageJ (NIH). Plot of mean circu-arity vs. solvent hydrogen bonding solubility parameter is shownn Fig. 8.
A representative PXRD pattern of original and recrystallizeddipic acid is shown in Fig. 9. It can be observed that 2-theta valuespeak positions) for both original and recrystallized sample coin-ide. The superimposition of X-ray diffraction patterns shows thathe crystalline phase remains unchanged. The same observation
Fig. 9. PXRD pattern for adipic acid (original) and adipic acid recrystallized fromisopropanol.
was made for all dicarboxylic acids recrystallized from the differ-ent solvents. Therefore, the presence of different polymorphs canbe ruled out. This shows that only external crystal morphology isaltered and the internal crystal structure remains intact. Variationsin relative intensities of the peaks can be attributed to the differentcrystal morphologies.
4. Discussion
Fig. 7 shows that the AR of solutes decreases as the solvent ıHincreases till the third data point (corresponding to the solventmethanol). A significant increase in AR is then observed in the caseof propylene glycol. From propylene glycol through ethylene glycoland then to water, the AR decreases. Among all solvents, crystalsfrom, methanol has the lowest AR. The AR of the dicarboxylic acids
in the case of ethylene glycol and propylene glycol is always higherthan that observed for methanol, even though both ethylene glycoland propylene glycol have higher ıH values than methanol. A simi-lar trend is observed for crystal circularity, Fig. 8, except that valuesof circularity are somewhat inversely related to those of AR. This
1 Chem
sm
mdetogHmoaobT1dgbpispptituclh2vghbheol�vTif
5
5
ib2
TC
O
S
018 A.T. Karunanithi et al. / Computers and
hows that there is no direct relationship between ıH and soluteorphology (aspect ratio or circularity).To further enhance our understanding, the difference between
onohydric alcohols (e.g. methanol, ethanol and propanol) andihydric alcohols (e.g. ethylene glycol and propylene glycol) wasxplored. Monohydric alcohols differ from dihydric alcohols inhe way the solvent molecules are held together. One moleculef a monohydric alcohol has only one –OH group and for hydro-en bonding to occur there is a need for another –OH group.ence, hydrogen bonding in monohydric alcohols is primarily inter-olecular (i.e. hydrogen bonding between two molecules). On the
ther hand, one molecule of dihydric alcohol has two –OH groupsnd there is a tendency for intramolecular hydrogen bonding toccur. Hence, although these solvents are by themselves polar, theyehave in a ‘non-polar’ manner due to intramolecular association.his is sometimes referred to as the ‘chameleonic’ effect (Barton,985). Crystal morphology of carboxylic acids is determined by theegree of hydrogen bonding interaction between the solute –COOHroup and solvent. Solvents having high intramolecular hydrogenonding ability seem to form crystals of higher aspect ratio com-ared to solvents whose hydrogen bonding ability is intermolecular
n nature. Barton (1985) cautions the use of hydrogen bondingolubility parameter in situations where interactions within a com-onent are very different from those between components. Theroblem in some cases is that the solubility parameters are tryingo indicate the values of two or more quantities, i.e. hydrogen bond-ng within the pure compound and hydrogen bonding betweenwo compounds, with one number. Hence, hydrogen bonding sol-bility parameter of solvents alone is inadequate to characterizerystal morphology. Spectroscopic studies have shown that ethy-ene glycol and propylene glycol form a 3-dimensional network ofydrogen bonding (Rodnikova, Chumaevskii, Troitskii, & Kayumova,005). Crupi, Longo, Majolino, and Venturi (2006) also demonstrateia Raman spectroscopy that propylene glycol exhibits a hydro-en bonded network structure. This network of hydrogen bondsolding molecules of dihydric and polyhydric alcohols is responsi-le for their characteristically high viscosity. Thus, solvents havingigh intramolecular interactions tend to have high viscosity. Thisxplains why the aspect ratio increases significantly in the casef propylene glycol (� = 40.4 cP) and to a lesser extent in ethy-ene glycol (� = 16.0 cP). Hence, taken together the solvent ıH and
can be utilized to explain the relationship between the sol-ents and the observed crystal morphologies of carboxylic acids.hese observed relationships, identified through bench scale exper-ments, are more formally formulated into property requirementsor design/selection of solvents in the following section.
. Solvent selection/design strategy
.1. Database search
Process solvents are typically selected from databases by match-ng property requirements. Some well-known databases that cane used for this purpose are DIPPR and CAPEC databases (ICAS,003). In the database search approach, selection of solvents is
able 3ase study solute information.
rganic solute Chemical structure
ebacic acid
ical Engineering 33 (2009) 1014–1021
straight forward. The user defines the criteria for solvent selectionas property constraints. The database is then searched to retrieveall solvents which match the user-defined property constraints. Anyof those solvents from the list could then be selected or a rankingsystem for solvents such as the one proposed by Gani, Jimenez-Gonzalez, and Constable (2005) could be used to rank the solvents.The critical step in the database search approach is the formulationof the property constraints. Here the user needs to have extensiveknowledge about the processing issues. Environmental, health, andsafety issues should be given critical consideration. Then, the usershould be able to convert these requirements into suitable prop-erty constraints for the solvents. The drawback in the databasesearch approach is the availability of all compounds and their cor-responding experimental property values. Another limiting factorin the database search approach is consideration of certain qual-itative properties such as crystal morphology. The problem hereis the construction of quantitative solvent property constraint forqualitative properties like crystal morphology. More often, thereis no way to do that and hence these important considerations areignored during solvent selection stage and then addressed on a trialand error basis on a pilot plant scale. This is obviously an expen-sive and time-consuming endeavor. Development of relationshipsbetween solvent properties and crystal morphology would helptackle this issue in the selection stage itself. This is exactly whatour experimental study in the previous section addresses. Since forcarboxylic acids, a relationship between the two solvent properties,namely, solvent hydrogen bonding solubility parameter (ıH) andsolvent viscosity (�) to the solute crystal morphology was estab-lished, property constraints could be devised for morphology basedon ıH and �. In this section, a practical example is used to demon-strate how our experimental findings on carboxylic acids can beused for solvent selection through database search. The problem isto select a suitable solvent for crystallization of Sebacic acid. Sebacicacid is an aliphatic dicarboxylic acid whose structure and propertiesare shown in Table 3.
Karunanithi et al. (2006) have defined various property require-ments that need to be considered while selecting solvents forcrystallization processes. These properties are (1) potential recov-ery of solute, (2) solubility of solute, (3) crystal morphology, (4)solvent flash point, (5) solvent toxicity, (6) solvent boiling pointand (7) solvent melting point. Since potential recovery is an equi-librium property, it is excluded from the database search approachsince only pure component solvent properties can be searched. Thesolubility parameter of Sebacic acid is 22.5 MPa1/2 and a solventwith solubility parameter close to that of the solute is desirable sothat the solute solubility is high. With respect to crystal morphol-ogy, the experimental study in the previous section revealed thathigh solvent hydrogen bonding solubility parameter and low sol-vent viscosity are likely to yield low aspect ratio crystals. Therefore,constraints on these two properties are used to select appropriate
solvents. −log(LC50) is used as a toxicity constraint and flash pointis used as a safety constraint. The other constraints are on boil-ing and melting points of the solvent. Actual constraints for thissolvent selection problem are shown in Table 4. If this solvent selec-tion problem needs to be carried out for a different carboxylic acid,
he only constraint that needs to be modified is the total solubilityarameter, ı, where one would want to select solvents having sol-bility parameter close to that of the solute. On the other hand, ifhis solvent selection problem needs to be carried out for a solutehat does not belong to the carboxylic acid group, one needs toonduct bench scale preliminary crystallization experiments usingompounds that belong to the same class as that of the solute. Thesexperiments can be used to devise morphology related constraints.ence, for this case the morphology constraints and total solubilityarameter constraint would vary.
CAPEC database (ICAS, 2003) was used to screen solvents forhe properties solubility parameter, flash point, melting point andoiling point. The solvent property lists provided by Barton (1985),arcus (2001), and Martin and Young (2001) were used to screen
olvents with respect to viscosity, hydrogen bonding solubilityarameter and −log(LC50), respectively. Four solvents were foundo satisfy all of the property requirements. For some other solvents,xperimental values of certain properties were missing and if theseeed to be considered then property prediction models can be usedo estimate the missing values. Results of the database search arehown in Tables 5 and 6.
Table 5 provides the list of solvents that satisfied all the propertyonstraints along with the experimental values of the properties.able 5 provides a list of solvents for which certain experimentalalues were not available and were predicted using ProPred prop-rty prediction software (ICAS, 2003). From Tables 5 and 6, it isorth noting that all of the solvents selected are alcohols and to beore specific all of them are monohydric alcohols. Hence, mono-
ydric alcohols are the ideal solvents for carboxylic acids if crystalsf low aspect ratio are desired.
.2. Computer-aided molecular design (CAMD)
CAMD is a technique to design molecules having specificesired properties. CAMD has often been used to design solventsor separation processes such as liquid–liquid extraction, absorp-ion and crystallization. One of the commonly used methodology
(Karunanithi, Achenie, & Gani, 2005) is to formulate the CAMDproblem as a Mixed Integer Nonlinear Programming (MINLP) opti-mization model. In this technique the solvent properties are posedas constraints in an optimization problem where an objective func-tion such as potential recovery needs to be maximized. The solutionof the CAMD–MINLP optimization model results in an optimal sol-vent structure. As in the case of database search, in order to considerqualitative properties such as crystal morphology, constraints onsurrogate solvent properties hydrogen bonding solubility parame-ter and viscosity need to be developed. The experimental resultsfrom the previous section helped us devise these morphologi-cal constraints. In this section we demonstrate the application ofour experimental investigation in devising property constraintsfor crystallization solvent design of Sebacic acid. Karunanithi etal. (2006) developed a solvent design model for crystallizationsolvents. This model was slightly modified, in view of the newexperimental results, by considering viscosity of the solvent as amorphological constraint. The solvent design optimisation modelis shown below.
Max PR% = 1001 − X1
(1 − X1
X2
)(2)
∑i
∑j
uij(2 − �j) = 2 (3)
∑i
∑j
uij = Nmax (4)
∑j
uij = 1 (5)
Tm = 102.425∑
i
NiTmi +∑
j
MjTmj ≤ 270 K (6)
Tb = 204.359∑
i
NiTbi +∑
j
MjTbj ≥ 340 K (7)
Tf = 3.63∑∑
i
NiTfi + 0.409Tb + 8843 ≥ 323 K (8)
−log(LC50) ≤ 1.6 (9)
20 ≤ ı =[
1000HVap298 − RT
Vm298
]1/2
≥ 30 MPa1/2 (10)
ıH =∑∑
i
Niıhi/V298m ≥ 16 MPa1/2 (11)
� = Molecularweight × 1000
× exp
⎛⎝∑
i
Ni�i/T +∑
j
Mj�j ≤ 3
⎞⎠ cp (12)
ln xSati − �fusH
Tm
(1 − Tm
T
)+ ln �Sat
1 = 0 (13)
x1 + x2 = 1 (14)
260 ≤ T ≤ 320 (15)
The design variables are the structural variables uij. Eq. (2) rep-
resents the potential recovery that needs to be maximized. X1 andX2 represent the solubility at low and high temperatures, respec-tively. Eqs. (3)–(5) represent the structural constraints that needto be satisfied to form a structurally feasible solvent molecule. Thesolubilities (X1 and X2) present in the design objective (potential
ormal melting point 217.64 Kormal boiling point 406.63 Kiscosity 1.90 cP
ecovery) are a function of design variables (i.e. structural vari-bles) and temperature. Eqs. (6) and (7) represent the constraintsn melting and boiling points of the solvents to make sure thathe solvents are in liquid state at operating conditions. These con-traints are evaluated using the group contribution model proposedy Constantinou and Gani (1994). Eq. (8) represents the safety con-traint on flash point, which is also evaluated using the aboveroup contribution model. Eq. (9) represents constraint on toxic-ty, evaluated using group contribution model proposed by Martinnd Young (2001). Eq. (10) represents constraint on solubility. Eqs.11) and (12) represent constraints (hydrogen bonding solubilityarameter and viscosity) related to crystal morphology. Note thathe constraint on viscosity has been modified based on the newxperimental results (refer Section 4). Previously viscosity was justsed as a process requirement constraint, but in light of our newxperiments we have used viscosity as a morphological constraints well. Eqs. (13) and (14) are phase equilibrium constraints. Eq.15) represents constraint on operating temperature. More detailedescription about the above property constraints and group contri-ution models used to calculate the property values can be found
n Karunanithi et al. (2006). This CAMD-optimization model wasolved using a decomposition based solution approach proposedy Karunanithi et al. (2005). The design results of the optimal sol-ent are shown in Table 7. Here, the designed optimal solvent wasound to be a monohydric alcohol.
Note that the constraint values used in the database searchnd CAMD model are slightly different. This is because CAMD pro-ides only one optimal solvent and hence it makes sense to haveighter property constraints in order to simplify the solution of theptimization model (by reducing the feasible region in the searchpace). On the other hand, the database search method has thedvantage of selecting a list of potential candidate solvents that cane further tested. However, if constraints are made too tight thene might not find any solvent or find very few solvents from theatabase that match the search criteria. Hence, tighter constraintsere used for the CAMD approach and slightly relaxed constraintssed for database search approach. This was done to take advantagef the strengths of both methods of solvent selection.
. Conclusions
Solvent effect on crystal morphology of aliphatic dicarboxyliccids is investigated more systematically and comprehensivelyy using an integrated methodology which combines targetedench scale crystallization experiments, computer-aided molecularesign approach and database search approach. From the exper-
ments, it was observed that the hydrogen bonding interactionetween the carboxylic acid and the solvent plays a critical roleith respect to final crystal morphology. Stronger hydrogen bond-
ng interactions between solvent and the –COOH group of theolute resulted in crystals with lower aspect ratio (higher circu-
ical Engineering 33 (2009) 1014–1021
larity). However, for dihydric alcohols with predominantly higherintramolecular hydrogen bonding, crystals with high aspect ratiowere obtained. Thus, the proposition that the aspect ratio of car-boxylic acid crystals has an inverse relationship with the hydrogenbonding solubility parameter of the solvent being used is not alwaystrue. The solvent hydrogen bonding solubility parameter on itsown is therefore not enough to characterize crystal morphology.Solvents exhibiting high intramolecular hydrogen bonding inter-actions tend to be highly viscous in nature and hence we proposeusing solvent hydrogen bonding solubility parameter and solventviscosity for selection of appropriate crystallization solvents. Basedon the above arguments, property constraints were developed forthe selection/design of solvents for crystallization of carboxylicacids. An illustrative example involving Sebacic acid revealed thatmonohydric alcohols are best suited for carboxylic acid crystalliza-tion. Finally, the combined (unified) approach taken in this papermakes it a more powerful methodology for design and selection ofsolvents for crystallization of carboxylic acids than any one of themethods applied alone.
Acknowledgements
The authors are grateful to R. Gani (Technical University ofDenmark) for many useful discussions and for providing the ICASsoftware. SLS and SS acknowledge support of the US Departmentof Energy, Office of Basic Energy Sciences. Many thanks go to J.Romanow (EM Lab., University of Connecticut) for help with SEMoperation and E. Nyutu (Dept. of Chemistry, University of Connecti-cut) for some insightful suggestions.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.compchemeng.2008.11.003.
References
Algra, R. E., Graswinckel, W. S., Van Enckevort, W. J. P., & Vlieg, E. (2005). Alizarin crys-tals: An extreme case of solvent induced morphology change. Journal of CrystalGrowth, 285, 168–177.
Barton, A. F. M. (1985). Applications of solubility parameters and other cohesionparameters in polymer science and technology. Pure and Applied Chemistry, 57,905–912.
Bunyan, J. M. E., Shankland, N., & Sheen, D. B. (1991). Particle design via crystallization.New York: American Institute of Chemical Engineers., pp. 44–57
Cang, H. X., Huang, W. D., & Zhou, Y. H. (1998). Effects of organic solvents on the mor-phology of the meta-nitroaniline crystal. Journal of Crystal Growth, 192, 236–242.
Cano, H., Gabas, N., & Canselier, J. P. (2001). Experimental study on the ibuprofencrystal growth morphology in solution. Journal of Crystal Growth, 224, 335–341.
Constantinou, L., & Gani, R. (1994). A new group contribution method for estimatingproperties of pure compounds. AIChE Journal, 40(10), 697–1710.
Crupi, V., Longo, F., Majolino, D., & Venturi, V. (2006). The hydrogen-bond network inpropylene glycol studied by Raman spectroscopy. Journal of Molecular Structure,790(1–3), 141–146.
Davey, R. J., & Whiting, M. J. L. (1982). Habit modification of succinic acid crystalsgrown from different solvents. Journal of Crystal Growth, 58, 304–312.
Gani, R., Jimenez-Gonzalez, C., & Constable, D. J. C. (2005). Method for selection ofsolvents for promotion of organic reactions. Computers and Chemical Engineering,29, 1661–1676.
Garekani, H. A., Sadeghi, F., Badiee, A., Mostafa, S. A., & Rajabi-Siahboomi, A. R.(2001). Crystal habit modifications of ibuprofen and their physicomechanicalcharacteristics. Drug Development and Industrial Pharmacy, 27, 803–809.
Garti, N., Leci, C. L., & Sarig, S. (1981). The effect of solvents on crystal habit of 1,4-di-tert-butylbenzene (DTBB). Journal of Crystal Growth, 54(2), 227–231.
Gordon, R. E., & Amin, S. I. (1984). Crystallization of Ibuprofen. U.S. Patent Number4476248.
Hansen, C. M. (2002). Hansen solubility parameters: A user’s handbook (2nd ed.). NewYork: CRC Press.
ICAS Documentation. (2003). Internal Report. CAPEC, Department of Chemical Engi-neering, Technical University of Denmark.
ImageJ software. National Institute of Health, MD.Jones, H. P., Davey, R. J., & Cox, B. G. (2005). Crystallization of a salt of a weak organic
acid and base; solubility relations, supersaturation control and polymorphicbehavior. Journal of Physical Chemistry, B109, 5273–5278.
arunanithi, A. T., Achenie, L. E. K., & Gani, R. (2005). A new decomposition-basedcomputer-aided molecular/mixture design methodology for the design of opti-mal solvents and solvent mixtures. Industrial and Engineering Chemistry Research,44, 4785–4797.
arunanithi, T. A., Achenie, L. E. K., & Gani, R. (2006). A computer-aided molecu-lar design framework for crystallization solvent design. Chemical EngineeringScience, 61, 1247–1260.
arunanithi, T. A., Acquah, C., Achenie, L. E. K., Sithambaram, S., Suib, S. L., & Gani, R.(2007). An experimental verification of morphology of ibuprofen crystals fromCAMD designed solvent. Chemical Engineering Science, 62, 3276–3281.
ide, D. R. (2001). Handbook of chemistry and physics (82nd ed.). New York: CRC Press.
a, D. L., Tafti, D. K., & Braatz, R. D. (2002). Optimal control and simulation of multi-
dimensional crystallization processes. Computers and Chemical Engineering, 26,1103–1116.
arcus, Y. (2001). The properties of solvents. Chichester, New York: Wiley.artin, T. M., & Young, D. M. (2001). Prediction of acute toxicity (96-h LC50)
of organic compounds to the fathead minnow (pimephales promelas) using
ical Engineering 33 (2009) 1014–1021 1021
group contribution method. Chemical Research in Toxicology, 14(10), 1378–1385.
Mwangi, S. M., & Garside, J. (1996). Morphology of 6-aminopenicillanic crystals andthe effect of phenoxyacetic acid. Journal of Crystal Growth, 166, 1078–1083.
Rasenack, N., & Muller, B. W. (2002). Properties of ibuprofen crystallized under var-ious conditions: A comparative study. Drug Development & Industrial Pharmacy,28, 1077–1089.
Rodnikova, M. N., Chumaevskii, N. A., Troitskii, V. M., & Kayumova, D. B. (2005).Structure of liquid ethylene glycol. Russian Journal of Physical Chemistry, 80(5),826–830.
Srinivasan, K., Sankaranarayanan, K., Thangavelu, S., & Ramasamy, P. (2000). Influ-
ence of organic solvents on the habit of NMBA (4-nitro-4-methylbenzylideneaniline) crystals. Journal of Crystal Growth, 212, 246–254.
Tomasko, D. L., & Timko, M. T. (1999). Tailoring of specific interactions to modify themorphology of naproxen. Journal of Crystal Growth, 205, 233–243.
Winn, D., & Doherty, M. F. (2000). Modeling crystal shapes of organic materials grownfrom solution. AIChE Journal, 46(7), 1348–1367.
