a study on batch cooling crystallization of...

A STUDY ON BATCH COOLING CRYSTALLIZATION OFSULPHATHIAZOLE

Process Monitoring Using ATR-FTIR and ProductCharacterization by Automated Image Analysis

K. POLLANEN, A. HAKKINEN�, S.-P. REINIKAINEN, M. LOUHI-KULTANEN and L. NYSTROM

Department of Chemical Technology, Lappeenranta University of Technology, Lappeenranta, Finland

Supersaturation as the driving force of crystallization processes is an essential parameterand its effects on the quality of the product should be carefully investigated. In thispaper, a systematic study on cooling rate and cooling mode effects on the supersatura-

tion level and, consequently, on the size and shape of the sulphathiazole crystals has been con-ducted. In situ concentration monitoring was carried out using ATR-FTIR spectroscopy andthe size and shape of the crystals produced were measured by automated image analysis. Thesimultaneous application of these two modern analytical methods was found to provide usefulinformation for the thorough study of the causes and consequences of different crystallizationconditions on the outcome of the process. The results obtained in this study showed a signifi-cant increase in the average crystal size with decreasing cooling rate when linear coolingprofiles where applied. The average size of the sulphathiazole crystals produced by usingthe programmed cooling was found to be slightly smaller than the size of the crystals obtainedby using linear or natural cooling profiles with the same batch time. The repeatability of theexperimental procedure was considered to be good since the relative standard deviationsbetween the results of the repeated batches were rather small in all cases. The variationsobserved in the properties of the crystals produced could be easily and comprehensivelyexplained by the differences observed in the concentration profiles.

Keywords: crystallization; supersaturation; ATR-FTIR spectroscopy; crystal characteriz-ation; image analysis.

INTRODUCTION

The driving force of all crystallization processes is super-saturation. In cooling crystallization, the supersaturation istypically expressed as the concentration of the crystallizedmaterial in excess of its solubility at a given temperature.It is well known that the level of supersaturation duringthe cooling crystallization process influences the proper-ties of the crystalline product obtained. This has beenindisputably confirmed by several experimental resultspresented in the literature (e.g., Myerson et al., 1986;Jagadeshi et al., 1996; Matthews and Rawlings, 1998;Togkalidou et al., 2001a; Srinivasakannan et al., 2002;Ulrich and Strege, 2002; Lewiner et al., 2002). Commonlyrequired properties of a crystalline material are certaincrystal size, narrow size distribution and desired crystal

shape. These properties are determined mainly by the pro-cesses of nucleation, growth, attrition, breakage andagglomeration. Supersaturation affects nucleation andcrystal growth considerably, both of which accelerate asthe level of supersaturation is increased. In order toachieve the previously mentioned crystal properties, it isnecessary to somehow minimize the effect of the nuclea-tion process and maximize the crystal growth so that mostof the material that is dissolved in the solution couldaccumulate on the surfaces of a small number of crystals.This can be done by controlling the level of super-saturation during the crystallization process. Nucleationtypically occurs at rather high supersaturation levelswhereas crystal growth can occur and proceed at consider-ably lower levels of supersaturation (Mersmann, 1999).Based on these considerations, the optimal crystal proper-ties should be achieved if the supersaturation could bekept at a relatively low level throughout the whole crystal-lization process. One method for influencing the level ofsupersaturation during cooling crystallization is to controlthe cooling rate applied.

�Correspondence to: Dr A. Hakkinen, Department of Chemical Technology,Lappeenranta University of Technology, P.O. Box 20, FIN-53851,Lappeenranta, Finland.E-mail: [email protected]

47

0263–8762/06/$30.00+0.00# 2006 Institution of Chemical Engineers

www.icheme.org/journals Trans IChemE, Part A, January 2006doi: 10.1205/cherd.05082 Chemical Engineering Research and Design, 84(A1): 47–59

Cooling Modes

The cooling methods traditionally employed in coolingcrystallization processes can be roughly classified intothree different types, which are natural, linear and pro-grammed cooling. Natural cooling refers to a methodwhere the crystallizer is cooled with a constant temperaturecoolant and this normally results in very high cooling ratesat the early parts of the batch. Consequently, the solution isusually supersaturated at a considerably faster rate than it ispossible to desupersaturate it by the growth on the existingcrystals (Jones and Mullin, 1974). This causes thesupersaturation to rapidly exceed the metastable limitthus allowing massive spontaneous nucleation to takeplace. Consequently, the product obtained with naturalcooling often consists of a large number of small,irregularly shaped crystals.Linear cooling profile refers to a constant cooling rate

throughout the whole crystallization process. If the appliedcooling rate is selected correctly, it is possible to reducethe high initial supersaturation peak associated with natu-ral cooling and therefore also to diminish the occurrenceof unfavourable spontaneous nucleation after the first pri-mary nuclei have been generated. This should lead to anincrease in the average crystal size compared to the crys-tals produced by natural cooling, as was reported by Jonesand Mullin (1974). If the cooling rate chosen is too high,the possibility of repeatedly exceeding the metastablelimit during the process becomes likely and this againcan lead to worsening quality of the product as a conse-quence of uncontrolled nucleation. Too low cooling rateson the other hand can result in uneconomically longbatch times.A lot of work has been carried out during the last few

decades on finding the optimal cooling profiles thatshould be followed in order to perform batch cooling crys-tallization processes in a controlled manner. The objectiveof such programmed cooling profiles is to ensure that thegeneration rate of supersaturation always matches the avail-able crystal surface area on which the crystal mass formedis to be transferred (Davey and Garside, 2000). This meansthat at the early stages of the batch, when the surface areaof the crystals existing in the solution is small, the rate ofsupersaturation production has to be slow and thereforealso the applied cooling rates need to be extremely low.As the crystallization process then proceeds, the surfacearea of the crystals increases and the cooling rate increasestowards the end of the batch.One of the first papers that dealt with programmed cool-

ing of batch crystallizers was published by Mullin andNyvlt (1971), who derived an exact theoretical equationthat represented the ideal cooling curve based on constantnucleation and growth rates in a seeded solution. Thisequation, however, was too complex for general use andit was therefore simplified by making some appropriateassumptions. For an unseeded batch cooling crystallizationprocess of a substance with its solubility depending linearlyon temperature, the simplified cooling curve that should befollowed in order to maintain approximately constantsupersaturation throughout the whole duration of thebatch became:

(T0 � T)=(T0 � Tf ) ¼ (t=t)4 (1)

where T0 is the initial solution temperature, T is the solutiontemperature at time t, Tf is the final solution temperature,t is the time and t is the batch time.

The obtained simplified profiles were utilized by Mullinand Nyvlt (1971) and by Jones and Mullin (1974) in seededcrystallization experiments performed with aqueous sol-utions of potassium sulphate and ammonium sulphate andit could be shown in both of these cases that the mediancrystal size was clearly increased when compared withthe size of the crystals obtained using natural cooling.The presented experimental results however also showedthat the programmed cooling profile was not very effectivein eliminating the occurrence of nucleation during thecrystallization since all of the obtained crystal productscontained a significant amount of fine crystals.

Even though a large number of different, even furtheroptimized and more specific theoretical cooling profilescan be found from the literature, it is not usually necessaryin practice to use the exact forms of these optimalequations. According to Davey and Garside (2000) it isoften only required that a reasonable approximation tothe appropriate time–temperature profile is achieved.

After the first crystals have appeared in the solution, sec-ondary nucleation is likely to take place and the number ofnuclei that are produced by this mechanism will increasewith the mass of crystals in the solution (Mersmann,1996; Davey and Garside, 2000). Secondary nucleationcan be divided into attrition nucleation and surface nuclea-tion. Attrition nucleation refers to the formation of newnuclei as a consequence of crystal breakage, which iscaused by collisions between the crystals and the walls ofthe crystallizer, a stirrer or other crystals in the suspension.Surface nucleation on the other hand refers to formation ofnew nuclei on the surfaces of the crystals existing in thesolution or in their immediate vicinity (Mersmann, 1996)and it depends on the level of supersaturation. The occur-rence of secondary nucleation might set some limits onthe size distribution that can be achieved with controlledcrystallization because it creates new crystal surfaces onwhich the available mass of the dissolved material candeposit. In addition, attrition nucleation also decreases thesize of the already formed crystals.

In situ Concentration Measurement with ATR-FTIR

To get real-time information on the supersaturation levelthroughout the crystallization process, a reliable solute con-centration monitoring technique is needed. The ATR-FTIRtechnique has proved to be a suitable technique for moni-toring crystallization processes due to its wide applicabilityto different systems, since most compounds absorb radi-ation in the mid-IR range. In addition, the measurementtakes place at the interface between the ATR element anda sample and therefore the existing crystals are assumednot to disturb the measurements. However, transformationof the spectral data to concentration information is a criticalissue in order to obtain reliable results. Traditionalregression methods can be applied by correlating theheights or areas of specific peaks, or alternatively a ratioof specific peaks to a concentration of the measured con-stituent can be used (Uusi-Penttila and Berglund, 1996;Dunuwila and Berglund, 1997; Fevotte, 2002; Lewineret al., 2001a, b, 2002). However, traditional regression

Trans IChemE, Part A, Chemical Engineering Research and Design, 2006, 84(A1): 47–13

48 POLLANEN et al.

methods should not be applied to correlated variables,which the spectral variables typically represent. In complexchemical systems, the bands in the IR spectrum from differ-ent constituents often overlap each other, the absorbenciesof specific compounds of interest can be low and, conse-quently, no single peak can be found to correlate reliablywith concentration. In addition, random variation isalways present in the spectra, and for that reason a singlepeak can provide inaccurate estimates of the concentrationlevel. Multivariate methods, e.g., partial least squares (PLS)regression and principal component regression (PCR) cali-bration models can be applied to overcome these problemsand they have previously been applied to solute concen-tration prediction from crystallization systems (Togkalidouet al., 2001b; Profir et al., 2002; Feng and Berglund, 2002;Liotta and Sabesan, 2004). PLS enables the linear model-ling of correlated variables. The collinearity between vari-ables is, in fact, a stabilizing advantage rather than aproblem (Wold et al., 1983). A large number of variables,i.e., spectral points, can be included in the model. PLSreduces the dimensions of the original data and, at thesame time, it may reduce the noise level. The PLS cali-bration can be improved by careful data and model vali-dation steps as well as using an appropriate datapreprocessing technique. A calibration routine that includesvalidation and preprocessing steps for solute concentrationprediction in cooling crystallization process (Pollanenet al., 2005) is used in this context.The physical and chemical purity of a product is a major

issue in crystallization. ATR-FTIR has been applied tomonitor the concentration level in crystallization of race-mic compounds and process conditions that make it pos-sible to separate an enantiomerically pure compound outof a racemic mixture using a cooling crystallization process(Profir et al., 2002). Concentration measurement using theATR-FTIR technique in crystallization of a certain poly-morphic form of glycine has been investigated (Dokiet al., 2004). In addition, the measurement of the level ofspecified impurity simultaneously with the solute concen-tration measurement using ATR-FTIR has been introduced,but the impact of concentration or impurity level onthe product properties was not considered in that context(Derdour et al., 2003).The information on the concentration of crystallizing

substance allows the possibility of investigating the effectsof the driving force changes on the obtained product mor-phology and crystal size distribution (CSD). Consequently,the process conditions can be optimized to meet the desiredproduct properties. The results from the studies inwhich ATR-FTIR was used for concentration measurementduring the crystallization experiments using differentconstant cooling rates with and without seed crystalsshowed that the metastable range increased with thecooling rate and also with the overall concentration level,as was expected theoretically. The overall concentrationlevel increased by seeding as well (Fevotte, 2002; Lewineret al., 2001a, b, 2002). The average size of the product wasobserved to be largest when seeding was used (Lewineret al., 2001a) and the number of crystals was found toincrease with the cooling rate (Lewiner et al., 2001b).The variations in crystal shape in terms of length-to-width ratio were found to be reduced when seeding wasapplied (Lewiner et al., 2001b). Current research has

been strongly concentrated on the application of closedloop feedback control based on ATR-FTIR concentrationmeasurements. Attempts to simultaneously obtain crystalsas large as possible, to avoid exceeding the metastablelimit during ongoing crystallization and to minimize thebatch time have been introduced. The results have shownthat by maintaining the concentration level under the meta-stable limit, larger and more uniform product crystals canbe obtained than when the metastable limit is exceeded(Fujiwara et al., 2002). Controlled crystallization processeswith constant supersaturation levels have been compared touncontrolled crystallizations in previous studies (Gronet al., 2003; Liotta and Sabesan, 2004). Crystallizationwith feedback control with constant supersaturation wasfound to provide larger crystal sizes than crystallizationwith constant cooling rate (Liotta and Sabesan, 2004).The size of the product crystals decreased as the level ofsupersaturation was increased (Gron et al., 2003; Liottaand Sabesan, 2004).

Automated Image Analysis

The level of supersaturation during the crystallizationprocess affects not only the size but also the shape of thecrystals produced. An ideal technique for characterizingcrystalline solids would therefore be such, that it couldsimultaneously provide information on the size and theshape of the analyzed crystals. Currently the only methodsthat are capable of separating different crystal dimensionsfrom each other and thus generating appropriate shape fac-tors are the ones that are based on imaging techniques. Theuse of different kinds of image analysis techniques for char-acterizing crystalline solids has been recently reported, forexample, by Lewiner et al. (2001a, b, 2002), Bernard-Michel et al. (2002), Pons et al. (2002), Alander et al.(2003), Faria et al. (2003), Hurley et al. (2004), CalderonDe Anda et al. (2005a, b) and Ferreira et al. (2005).Traditionally image analysis has been carried out usingmanual microscopy, which is such an arduous methodthat its utilization in practical applications is rather difficult.Manual microscopy measurements are very time-consuming and for that reason the number of particlesthat can be analysed in a reasonable period of time isoften not sufficient to ensure that the statistical significanceof the obtained results is high enough. The results obtainedfrom microscopy measurements also suffer from operatorto operator variability due to the fact that the human oper-ator needs to constantly make selections regarding theexamined particles and measured particle dimensions.Most of the difficulties related to manual microscopy can,however, be overcome by automated image analysistechniques.

In this work, the concentration levels were measuredwith ATR-FTIR from the crystallizations using differentconstant cooling rates as well as natural cooling and pro-grammed cooling modes, the principles of which weredescribed above. Characterization of the crystals producedwas performed with a modern automated image analyserdesigned especially for size and shape analysis of dry pow-ders. The cooling rate and mode had a remarkable effect onthe supersaturation level during the unseeded crystalliza-tion and consequently on the product properties. Variationsin crystal properties could be easily and comprehensively


BATCH COOLING CRYSTALLIZATION OF SULPHATHIAZOLE 49

explained by the differences observed in the crystallizationconditions that were detected using the in-line concentrationmeasurement data.

MATERIALS AND METHODS

The experimental work performed in this study consistedof unseeded laboratory scale batch cooling crystallizationexperiments carried out using various cooling rates andcooling profiles. The crystallized material in all experimentswas pharmaceutical grade (European Pharmacopoeia/United States Pharmacopoeia) sulphathiazole (IndustriasGMB S.A., Castellbisbal, Barcelona, Spain) and the solventused in the experiments was a 50/50-mixture (w/w) ofdeionized water and GC-grade n-propanol [purity min.99.7% (w/w); Aspokem Oy, Helsinki, Finland].

Solubility Measurements

In order to be able to properly control any cooling crys-tallization process, the solubility of the crystallized materialin the considered solvent has to be known. The solubility ofsulphathiazole in the chosen solvent mixture at tempera-tures ranging from 108C to 808C was determined experi-mentally using the gravimetrical method. Samples thatwere taken from saturated homogenous solutions at differ-ent temperatures were filtered through a 0.2 mm-pore-sizesyringe filter and evaporated to dryness in a vacuumoven. The solubility was calculated based on the measuredreduction in the mass of the sample during drying. Allmeasurements were made with three parallel samples inorder to ensure adequate accuracy. The results of thesesolubility measurements have already been presentedearlier (Hakkinen et al., 2003). The most relevant part ofthe results for the current study will also be given later inthis paper in Figure 2.

Crystallization Experiments

The equipment used in the crystallization experiments isillustrated in Figure 1. The main parts of the equipmentwere the jacketed 4.0 dm3 glass crystallizer (height

250 mm, diameter 160 mm), programmable LAUDARK 8 KP-thermostat unit and ABB BOMEM MB155Sspectrophotometer. The temperature in the crystallizerwas measured with a Pt-100 sensor and was registered ona PC. The same PC was also used to control the thermostatin such a way that the temperature in the crystallizeraccurately followed the predefined cooling profiles.The crystallizer was supplied with a three-bladed curvedblade impeller (diameter 100 mm) and four baffles, whichwere made of PTFE. The agitation rate in all experimentswas kept constant at 400 rpm, which corresponded to atip speed of 2.1 m s21. According to Mersmann and Loffel-mann (2000), tip speeds of this magnitude are typical forlaboratory scale crystallizers.

All crystallization experiments in this study were carriedout without seed crystals. The initial batches were preparedby introducing approximately 4.0 dm3 of solvent mixtureinto the crystallizer and adding a quantity of sulphathiazolesuch that it corresponded to the previously determined solu-bility at 808C. The solution was then heated from 258C to858C in 2 h and kept at that temperature for 1 h to makesure that all of the added sulphathiazole was dissolvedbefore the beginning of the cooling stage.

The experimental work can be divided into two partsaccording to the type of cooling profiles applied. The pur-pose of the first part of the crystallization experiments wasto examine the influence of the applied cooling rate whenthe cooling was carried out with a constant cooling ratethroughout the whole duration of the crystallization. Theexperiments were done with four different rates, 3.9, 5.5,9.2 and 27.58C h21, which resulted in cooling times of14.0, 10.0, 6.0 and 2.0 h for the temperature interval from808C to 258C. The temperature data collected from theexperiments with different cooling rates is presented laterin this paper in Figure 3.

The goal of the second part of the crystallizationexperiments was to investigate the influence of the appliedcooling profile. In addition to the linear cooling profiles, thecrystallization experiments were also carried out with natu-ral and programmed cooling. Selection of the batch timeused in these experiments was done based on the limit-ations caused by the available thermostat unit and theheat transfer characteristics of the equipment. Trial runswere made to find the minimum batch time, and it wasobserved from on those runs, that the crystallizer couldnot be cooled according to the temperature profile calcu-lated from equation (1) in an accurate manner in lessthan 6 h. The cooling time for the temperature intervalfrom 808C to 258C was therefore chosen to be 6 h inall experiments with different cooling modes.

Since all experiments in this study were done withoutseed crystals, it was decided that the starting temperatureof the programmed cooling profile should be well withinthe metastable zone. By using data collected during the pre-vious experiment with a linear (6 h) cooling profile, it waspossible to detect the temperature where the first nucleispontaneously appeared into the solution as the limit ofthe metastable zone was exceeded. The starting point ofprogrammed, and also natural, cooling profiles was decidedto be 1.08C above the detected nucleation temperature. Atthis temperature, which was found to be 758C, the solutionwas clearly supersaturated but no crystals were yet presentin the solution. The experiments with natural andFigure 1. Experimental set-up.


50 POLLANEN et al.

programmed cooling profiles were carried out by coolingthe solution first from 858C to 758C with the cooling rateof 9.28C h21 after which the chosen cooling profile wasstarted. The crystallizations with the programmed coolingprofile were performed using a PC that controlled thethermostat unit in such a way that the temperature in thecrystallizer accurately followed the profile calculatedfrom equation (1). The experiments with the natural cool-ing on the other hand were performed by rapidly changingthe temperature of the cooling medium to a constant temp-erature of 258C, which resulted in very high initial coolingrates and considerably lower cooling rates at the end of thebatch. In those experiments where natural cooling profilewas applied, the final temperature of the batch was reachedin approximately 1.5 h. It was, however, decided that inorder to keep the influence of the mixing conditionsequal, the batch time in all experiments with differentcooling modes should be the same. Therefore also theexperiments with the natural cooling profile were per-formed with a batch time of 6 h by maintaining the crystalsin the stirred solution at constant temperature of 258C forthe remaining 4.5 h. The temperature data collected fromthe experiments with different cooling profiles will beshown later in Figure 5.All the crystallization experiments performed in this

study were conducted at least twice to ensure adequaterepeatability. Comparison between the parallel batchessuggested that considerable batch-to-batch variations didnot exist for any of the monitored properties.

ATR-FTIR Measurements and Calibration Procedure

ATR-FTIR device and settingsATR-FTIR measurements were taken using an ABB

BOMEM MB155S spectrophotometer equipped with aDipper 210 ATR immersion probe with a conical ZnSeelement manufactured by Axiom Analytical Inc. Theshape of the element causes the light beam to undergotwo internal reflections in the interface of a sample andthe element before the attenuated reflection proceeds tothe detector. Grams32 software was used to collect thespectra. The mid-IR spectrum from 4000 to 750 cm21

was collected. As a compromise between the quality ofthe spectrum and the robustness of the measurements, thespectral resolution of 16 cm21 was used and each spectrumconsisted on average of 20 consecutive scans. Water wasused as a background spectrum.

Calibration measurementsA calibration routine is necessary in order to obtain

quantitative information from IR spectra. The calibrationroutine includes sets of measurements with known concen-trations. To obtain a stable calibration, the conditionspresent in true measurements should be covered. Theseconditions include: mixing conditions, reactor geometry,temperature range, solvent composition and, of course, con-centration range. However, in the case of crystallizationmeasurements the supersaturated stage is unstable, sincespontaneous nucleation can take place due to any mechan-ical or other distraction in the process, and, consequently,the calibration measurements with known concentrationsbecome precarious. For this reason the calibration

measurements were conducted in undersaturated conditionsbut concentrations as close as possible to the solubility con-centration were measured. This means that extrapolation isneeded when concentration is predicted. The calibrationmeasurements are done in a liquid phase, but in a truecrystallization process a solid–liquid suspension exists.Particularly in this type of situation careful data andmodel validation steps are required in order to obtain areliable model.

Calibration sets were measured using the previouslymentioned spectrophotometer settings. A sulphathiazoleconcentration from 0 to 30 g sulphathiazole/100 g solventand a solvent composition 0–100 w% n-propanol wascovered in the calibration measurements. The temperaturerange for the measurements was from 258C to 808C andthe number of measured points was 513.

ATR-FTIR solubility measurementsTo validate the model, the solubility of sulphathiazole

was measured by adding an excessive amount of sulpha-thiazole to the solvent and mixing this suspension atconstant temperature until the equilibrium concentrationwas reached. Equilibrium was detected when the consecu-tive spectra were constant. In practice, the time needed toreach equilibrium was approximately 2 h. This was donefrom 258C to 808C with increments of 58C. ATR-FTIRmeasured solubilities were then compared to the resultsobtained from gravimetric measurements. This particularvalidation step tested the applicability of the model to thesituation resembling true crystallization where there is asolid–liquid suspension present and its suspension densityis changing throughout the process.

Calibration modelling and validationThe spectral data was analyzed using Matlab 6.5 from

MathWorks, Inc. The spectral data together with tempera-ture and solvent compositions formed the X matrix. Thepredicted variable y was the solute concentration.

Measured calibration data was validated using multi-variate statistical process control charts (MSPC). TheMSPC charts are based on a principal component analysis(PCA) model and are applied to the X matrix. WithinMSPC, Hotelling’s T 2 charts from score vectors of thePCA analysis are formed and they reveal whether the vari-ation of variables in the plane of extracted PCs to the modelis greater than can be explained by random noise. TheQ-chart is calculated from the squared prediction error ofthe residuals, i.e., from the data left out from the modelof the observations. Variables can be evaluated within thestructure (T 2) and residual (Q) space using variable load-ings to detect the essential variables that cause a particularsample to represent a different structure than the othersamples. The most relevant wave number range to beused in the model was selected based on contributioncharts. The principle of MSPC is described in (MacGregorand Kourti, 1995).

From the evaluation of the quality of measuredcalibration data and previous knowledge of the variationof essential process variables that should be covered in apredictive model, a calibration set was selected. A cali-bration set (86 points) was used to build the stable cali-bration model and a separate test set was used to select



the best performing model. The data was pretreated usingan orthogonal signal correction (OSC) filter. The OSC fil-tering methods remove systematic variation in X uncorre-lated to Y by extracting the vector (OSC component)mathematically orthogonal to Y from the data matrix X(Wold et al., 1998). The number of PLS componentsneeded in the final predictive model should be decreasedby OSC preprocessing. This should also lead to more accu-rate predictions of a desired property using the model, if theappropriate validation procedures were applied.A PLS model was derived. PLS finds a score vector t in a

column space of X (t ¼ Xw) and a vector u in a columnspace of Y (u ¼ Yq) that gives the maximal squaredcovariance: max(u0t)2 ¼ max(q0Y0Xw)2, for jwj ¼ jqj,where w and q are loadings of the X and Y decompositionsrespectively (Wold et al., 1983). Several such derived com-ponents can be calculated and the number of componentsneeded in the model is defined using an appropriate vali-dation criterion. In this context, the selection of the bestperforming model was essentially based on the root meansquared error of validation (RMSEV) of the external testset. The selected model was further validated by comparingATR-FTIR solubility measurements to gravimetricallymeasured solubilities using concentrations higher than thecalibration measurement points and to the situation wherethere is a solid–liquid suspension present. A detaileddescription of the calibration procedure is presented inPollanen et al. (2005).

Crystal Characterization

The samples for crystal characterization were collectedimmediately after the crystallization experiments fromfour different locations in the crystallizer by using avacuum pipe. The four samples, which were combined toform a composite sample, were always taken from amixed suspension in order to reduce the undesired classifi-cation due to the sedimentation of the largest crystals. Thevolume of each composite sample was approximately500 ml that corresponded to roughly 1/8 of the totalvolume of the crystal suspensions. The crystals fromthese composite samples were gently separated from thesolvent with a Buchner funnel after which they weredried in a vacuum oven. The dry crystals were then usedfor crystal size and shape analyses that were performedusing an automated image analyser (PharmaVision 830,Malvern Instruments, Ltd). The completely automatedoperation of the analyser meant that the measurementswere always made routinely and very rapidly (approxi-mately 20 000 particles in 10 min). For this reason, it waspossible to analyse enough particles in a reasonableperiod of time to obtain reliable results. In addition, thehuman operator bias could be eliminated.The samples for the analyses were prepared by disper-

sing the dry crystals evenly onto a 100 mm � 100 mmsample plate that was placed on a sample tray underneatha video camera. The camera was then moved across thesample tray in a preprogrammed way by linear actuatorsand a large set of digitized video images was automaticallyacquired. The obtained raw video images were processedusing PharmaVision 830 software (version 4.2.1.15) thatseparated all the individual crystals from the images and

determined a set of various morphological parametersseparately for each crystal in the sample.

The optimal conditions for the image analysis can onlybe achieved when the crystals are dispersed onto thesample plate in a mono-layer, since otherwise the obtainedresults may be significantly distorted as a result of the over-lapping of the analysed crystals. The formation of a perfectmono-layer is never possible in practice and therefore theinfluence of the overlapping needs to be eliminated some-how. By comparing the crystal images provided by thePharmaVision and the morphological parameters of theindividual crystals, it was possible to detect the objectsthat contained the overlapping crystals and automaticallydischarge those from the original data. This techniquewas used here to ascertain that the results obtained trulydescribed the dimensions of the individual crystals. Twoor three sub-samples (approximately 0.3–0.4 g each)were taken from the dried composite samples and analysedseparately and the crystal data obtained from those analyseswere combined to give an average result. The number ofcrystals analysed from the different samples in this study,after the overlapping objects had been removed, variedbetween 69 368 and 148 716.

RESULTS AND DISCUSSION

Prediction Ability of the Model

The solubility of sulphathiazole was measured gravime-trically and using ATR-FTIR and is presented in Figure 2.

Figure 2 shows that there is a good agreement for thesetwo different measurements and therefore the calibrationprocedure can be expected to give accurate enough predic-tions for the studied crystallization process. The resultsclearly indicate that calibration in a different physicalstate (a liquid phase) from the solubility measurement (asolid–liquid suspension) did not decrease the predictiveability of the model.

The Influence of Cooling Rate on Supersaturation

The concentration profiles from crystallizations usingdifferent constant cooling rates (Figure 3) illustrate thatthe overall concentration level tends to increase when the

Figure 2. The solubility of sulphathiazole in a 50/50-mixture (w/w) ofdeionized water and n-propanol at different temperatures.


52 POLLANEN et al.

cooling rate is increased. This is an expected and logicalresult. When the cooling rate is very low, the mass transferof the solute molecules on the surface of existing crystals isfast enough to release most of the supersaturation and con-sequently the supersaturation level remains low. When thecooling rate increases, the solubility decrease with decreas-ing temperature becomes greater than the mass transfer rateonto crystals, which causes the supersaturation level toincrease and remain high. This effect was clearlydemonstrated with these results. When cooling rates of9.28C h21 and 5.58C h21 were used, there were fewsudden concentration decreases, ‘steps’, during the ongoing

crystallization process after the primary nucleation. Thesesteps may possibly be caused by an instantaneous exceed-ing of a metastable limit at these particular points. Thennew nuclei will be formed, which causes a sudden release ofsupersaturation and decreases the concentration rapidly.

The predicted concentration level in crystallization usinga 27.58C h21 cooling rate is probably too high to be true.There are several possible reasons. Firstly, the concen-tration level may be too high to be reliably predictedusing the calibration model derived and the model becomesunstable. Secondly, with high cooling rates, the concen-tration level tends to lie above the metastable limit continu-ously and new nuclei are generated throughout the wholecrystallization process. Therefore, the unstable systemmay cause distractions to ATR-FTIR measurements. Thewidth of the metastable zone increases with the coolingrate, which can be seen in Figure 3(b), but is also illustratedin Figure 4.

The Influence of Cooling Mode on Supersaturation

Temperature, concentration and supersaturation profilesfrom crystallizations using different cooling modes are pre-sented in Figure 5. It should be noted that the concentrationprofiles in Figure 5(b) are presented against temperature,but the supersaturation levels in Figure 5(c) are illustratedagainst time. By this way the authors on the one handwanted to visualize the concentration changes due to chan-ging temperature together with the solubility curve, but onthe other hand to illustrate the driving force changes againstthe cooling profile used.

The concentration profile from the crystallization processusing programmed cooling in Figure 5(c) illustrates thatthe concentration level after the first primary nucleationwas as low as the equilibrium concentration from 75 to72 which corresponds to almost two hours as illustratedin Figure 5(c). The reason is that the cooling rate wasvery low in the beginning of the crystallization processwith programmed cooling. The estimated supersaturationlevel [Figure 5(c)] for programmed cooling seems to dropbelow zero, which actually is not true, but the system ismost likely in the equilibrium state. All analytical measure-ments and calibration models have certain uncertainty

Figure 4. The width of metastable zone versus the cooling rate for asystem at high temperature where the primary nucleation occurs for thefirst time.

Figure 3. ATR-FTIR results for linear cooling profiles: (a) cooling ratesused, (b) concentration profiles and (c) supersaturation profiles.



level. The uncertainty in the concentration level predictionwith current calibration technique is approximated tobe +1.00 g solute/100 g solvent. Therefore the negativevalues obtained in the supersaturation prediction arewithin the calculated uncertainty level for this type of cali-bration, especially when this particular supersaturationvalue is calculated from two values measured with thistechnique: solubility and solute concentration, and both ofthese values have uncertainties. This subject could havebeen overcame by adjusting the restriction for negativevalues but the authors wanted to show actual predictionsfrom model which shows also the uncertainty due to

calibration. As can be seen in the Figure 5(c) the super-saturation level increased at the end of the crystallizationusing programmed cooling as the cooling rate wasincreased towards the end of the process. It can be con-cluded that the programmed cooling did not maintain aconstant supersaturation level. One reason was that the pro-cess conditions were not optimized using the crystallizationkinetics and thermodynamic properties when determiningthe programmed cooling profile but instead, the totalbatch time was fixed and the profile was fitted to that. Inaddition, the solubility of sulphathiazole increases expo-nentially with respect to temperature and not linearly aswas assumed in the approximations concerning the coolingprofile described by equation (1).

Concentration profiles from crystallization with naturalcooling show that the concentration level decreased rapidlyat the very moment the temperature of the cooling mediumwas set to a constant value of 258C. The suspension cooleddown rapidly and practically all the crystals were formed ina few minutes. The suspension reached the final tempera-ture at t ¼ 1.5 h [Figure 5(a)]. For the remaining 4.5 h,the crystals were kept in a mixed suspension at constanttemperature and the concentration was close to thesolubility concentration.

Crystal Characterization

Examples of the optical micrographs of the sulpha-thiazole crystals obtained from the different experimentsperformed in this study are presented in Figure 6.

The images presented show that the shape of the crystalsobtained with the fastest cooling rate seems to be highlyirregular which may be a result of uncontrolled primarynucleation. As the cooling rate is decreased to 9.28C h21,an obvious change in the crystal shape is detected and, inparticular, the larger crystals seem to commonly exist asrather long and thin plates. Further decrease in the coolingrate appears to make the largest crystals slightly morerounded, which is perhaps caused by the increased influ-ence of grinding and breakage of the crystals as a resultof collisions in the crystallizer. It can be observed thatbesides the large primary crystals, also a significant quantityof finer crystals exist in all samples and the shape of thosefine crystals is generally more rounded than the shape ofthe larger ones. These fine secondary crystals are mainlyassumed to be attrition fragments of large primary crystals,which have been formed as a result of grinding effectscaused by the mixing during crystallization.

The product crystals from the experiments with thenatural cooling profile are clearly smaller than the crystalsobtained with the other profiles and their shape appears tobe more irregular. When the crystallization is carried outusing the programmed cooling profile, major changescannot be detected in the size or shape of the largest crys-tals compared with the crystals obtained using the linearprofile with the same batch time, i.e. with a constant cool-ing rate of 9.28C h21. The size of the smaller crystalsobtained with programmed cooling, however, appears tobe slightly larger than with the other profiles and theirshape seems to be more needle-like. These crystals alsoappear to be more uniform in shape, which is probably aresult of a decrease in attrition since the majority of thecrystals has now been formed during the final parts of the

Figure 5. ATR-FTIR results for different cooling modes: (a) coolingmodes used, (b) concentration profiles and (c) supersaturation profiles.


54 POLLANEN et al.

crystallization process. The overall conclusion that can bemade from the microscope images is that sulphathiazolecrystals obtained from the 50/50-mixture of water andn-propanol tend to grow as elongated rods and that theshape of the crystals produced in this study seems to bestrongly influenced by the attrition occurred during thecrystallization process.To obtain numerical information concerning the size and

shape of the crystals produced in this study, the crystalsamples were analysed using an automated image analyser.It was decided to describe the size and shape of the crystalsby crystal length and roundness, respectively. The crystallength refers to the largest projection of the crystal, whereasroundness is a measure of the length to width ratio. For aperfect circle, the roundness equals 1, and for a needleshaped crystal, the value of roundness approaches 0.These parameters were readily provided by the Pharma-Vision-software. The crystal size distributions shown inFigure 7(a) and Figure 8(a) are given as differential distri-butions where the projected surface area of the crystalsis plotted against the crystal length. The crystal shapedistributions on the other hand are given as differentialroundness distributions by the number of crystals and areshown in Figure 7(b) and Figure 8(b). The data obtainedfrom the image analyses contained several morphologicalparameters that were measured separately for each individ-ual crystal. It was therefore possible to use this data toexamine the relationship between the size and shape ofthe crystals. For this purpose, the roundness of the crystalswas plotted against the corresponding crystal lengths asshown in Figure 7(c) and Figure 8(c).

The Influence of Cooling Rate on the Size andShape of the Product Crystals

Crystal size distributions of the samples obtained from theexperiments with different constant cooling rates [Figure 7(a)]show that the cooling rate clearly influenced the size of the

produced crystals. It can be observed that the widest sizedistribution was obtained when the cooling rate was the high-est. When the cooling rate is decreased, the resulting sizedistributions can be seen to become slightly narrower. Thistrend is especially obvious when the relative amounts of thesmallest crystals are compared whereas truly significantchanges in the size of the largest crystals cannot be observed.This kind of result can be for several different factors. First,since the crystal size distribution with studied cooling ratesis rather wide, it can be assumed that attrition nucleation incrystallization of sulphathiazole plays a considerable role.The largest size found in all samples was approximatelyequal and this could imply that the maximum attainablecrystal size was limited by the mixing conditions applied inthe experiments. A similar conclusion has also been madeby Mersmann and Loffelmann (2000), who suggested thatthe final product size can be strongly determined by attritionfor crystals above 100 mm. If this were not the case, onecould expect the largest crystals produced with the lowestcooling rates to be clearly larger than those produced withrapid cooling. It can also be further speculated that in factthe largest crystals are likely to suffer most from attritionbecause they have higher collisional probabilities due totheir larger area and mass (Matthews and Rawlings, 1998).

The observed differences in the size distributions canalso be explained by variations in the rate of spontaneousnucleation between the different cooling rates. It wasalready shown in Figure 3(c) that an increase in the coolingrate resulted in an increase in the supersaturation during thecrystallization. The supersaturation profile for the experi-ments with the highest cooling rate implied that the levelof supersaturation was extremely high and it was alreadysuggested earlier that one reason for this may be that spon-taneous nucleation has occurred throughout the wholeduration of the crystallization. The highest cooling rateresulted in a very wide crystal size distribution which canbe explained by massive uncontrolled nucleation. Spon-taneous nucleation has probably occurred repeatedly also

Figure 6. The shape of sulphathiazole crystals obtained by batch cooling crystallization experiments from a 50/50-mixture of deionized water andn-propanol using (a) constant cooling rate of 27.58C h21, (b) constant cooling rate of 9.28C h21, (c) constant cooling rate of 5.58C h21, (d) constant coolingrate of 3.98C h21, (e) natural cooling profile and (f) programmed cooling profile.



when cooling rates of 9.28C h21 and 5.58C h21 were usedsince obvious steps were observed in the supersaturationprofiles for those experiments. Such steps could not, how-ever, be detected in the supersaturation profile for theexperiments with the lowest cooling rate; experimentsthat resulted in the narrowest crystal size distribution.The crystal shape distributions in Figure 7(b) show that

although differences between the shapes of the crystals

produced with different cooling rates are rather small, thedirection of the observed changes is systematic. The round-ness of the crystals seems to increase with increasingbatch time. Also this result can be explained by the effectof attrition because it is reasonable to believe that due toattrition the sharp edges of the crystals are smoothed. Atthe same time, a population of small irregularly shapedfragments is formed. This conclusion can also be justified

Figure 7. Image analysis results for linear cooling profiles: (a) crystallength distributions, (b) roundness distributions and (c) crystal shapeversus crystal size of the samples.

Figure 8. Image analysis results for different cooling modes: (a) crystallength distributions, (b) roundness distributions and (c) crystal shapeversus crystal size.


56 POLLANEN et al.

with Figure 7(c) where the crystal shape is plotted againstthe corresponding crystal size. The shape of the smallestcrystals with all cooling rates clearly tends to approach asphere and significant decrease in the roundness valuescan be observed with increasing crystal size. A minimumvalue for roundness can be found in all cases, and it cantherefore be assumed that the crystal shape is affected byattrition especially with the largest crystals due to theirhigher collisional probabilities.As was already mentioned earlier, all crystallization

experiments in this study were repeated in order to ensureadequate repeatability. Table 1 shows the mean crystalsizes and shape factors, which were all calculated fromthe number distributions, for all batches conductedwith different constant cooling rates. The table alsoshows the relative standard deviations between the meansof the parallel batches that are presented here to quantifythe statistical accuracy of the sampling and analysisprocedure. The calculated relative standard deviations areless than +3.6% in all cases, which implies that consider-able variations between the repeated batches did not exist.

The Influence of Cooling Mode on the Size andShape of the Product Crystals

The crystal size distributions of the samples obtained withdifferent cooling profiles are shown in Figure 8(a). Obser-vation of the presented distributions reveals that only moder-ate differences can be observed between the differentsamples. The size and the relative amount of small crystals(,200 mm) in all samples are practically equal and signifi-cant differences can only be seen when the largest crystalsare compared. The narrowest distribution is obtained whenthe crystallization was carried out using natural coolingand the widest distribution results from the experimentswith the linear cooling profile. The size distribution of thecrystals obtained by the programmed cooling profile canbe seen to lie between the other two distributions. This canbe considered a somewhat unexpected result because itwas assumed that a programmed cooling profile wouldhave produced the largest crystals.Several reasons may explain these observations. The

supersaturation profiles presented in Figure 5(c) impliedthat programmed cooling could not maintain the super-saturation constant throughout the whole duration of thecrystallization. It can therefore be concluded that the sim-plifying assumptions made to obtain the equation for the

programmed cooling profile were not appropriate for thecase of sulphathiazole. One reason may be the highlyunlinear solubility curve of sulphathiazole (Figure 2) andanother possible explanation is once again the relativelyhigh rate of secondary nucleation. The results obtainedwith the different cooling rates implied that sulphathiazolecrystals were highly sensitive to mechanical attrition. If thisholds true, it can be concluded that the crystals formed atthe beginning of the batch have probably broken alreadyduring the initial slow cooling period. An obvious conse-quence of this is that the number of crystals has probablyincreased and therefore the maximum attainable crystalsize has become smaller. The increase in the supersatura-tion towards the end of the batch suggests that the coolingrate applied during the final stages was too high and it istherefore likely that spontaneous nucleation has occurredat the end of the batch.

When the crystallization was carried out using the natu-ral cooling profile, a rather narrow crystal size distributionwas obtained. It should, however, be mentioned here thatthe chosen batch time probably significantly influencedthe properties of the final product. The temperature profilepresented in Figure 5(a) showed that with natural coolingthe final temperature of the batch was reached very rapidly.Since the batch time was fixed to 6 h, the crystal suspensionwas mixed for approximately 4.5 h at a constant temperatureand the supersaturation during that period was practicallynegligible [Figure 5(c)]. It is assumed, however, thatdespite the constant low supersaturation, considerablechanges in the crystal properties during this mixingperiod have occurred. It is possible that the observed sizedistribution for crystals produced by natural cooling is con-siderably narrower than it would have been immediatelyafter the final temperature of the batch was reached.Presumably the size distribution right after the coolingperiod was even wider than was observed with the fastestcooling rate in Figure 7(a). The changes in the crystalproperties are assumed to be a consequence of two differentprocesses. The first is once again assumed to be attrition,which is expected to decrease the maximum size of thecrystals, thus narrowing the size distribution from theupper end. On the other hand, Ostwald ripening maycause the smallest crystals to dissolve and their dissolvedmass to be deposited on large crystals, which consequentlygrow. This phenomenon could therefore narrow the distri-bution from the lower end. The time required by Ostwaldripening to proceed, however, depends strongly on the

Table 1. Statistical values for the crystal samples obtained with different cooling rates.

Cooling rate

The number ofcrystals in the

sampleMean crystallength D[1,0]

Average crystallength+RSDa

Meanroundness

Averageroundness+RSDa

3.98C h21 Batch 1: 83 351 116.2 mm 114.3 mm+ 2.3% 0.478 0.482+ 1.0%Batch 2: 115 549 112.5 mm 0.485




aRSD refers to the calculated relative standard deviation between the means of two similar batches.



properties of the solute/solvent combination and it is there-fore difficult to estimate the actual effect of it in this case.The crystal shape distributions shown in Figure 8(b)

show that the differences between the samples crystallizedwith different cooling profiles are rather large. The crystalsproduced with the programmed cooling profile are clearlymore elongated than the crystals obtained by linear or natu-ral cooling. This can be explained by the previously madeassumption that a large fraction of the crystals producedwith programmed cooling was probably formed in thefinal parts of the batch. These crystals have most likely suf-fered less attrition than the ones formed earlier and theirshape is therefore closer to the preferred needle-likeshape of the sulphathiazole crystals grown in unmixed con-ditions. The largest roundness values can be found withcrystals produced by natural cooling and this implies thattheir external appearance has been significantly affectedby attrition, as was already assumed earlier.Figure 8(c) shows that the shape of the crystals depends

significantly on their size. A minimum value for roundnesscan again be found in all cases and it can be observed thatwith natural cooling this minimum value occurs at a clearlysmaller crystal size than with the other two profiles. Thereason for this is assumed to be the long constant tempera-ture mixing period at the end of the batch. All of the crys-tals had been formed before this period and therefore theeffect of crystal breakage on their shape is believed to bemore remarkable than with the other samples.Table 2 shows the mean crystal sizes and shape factors

for all the batches conducted with different constantcooling modes. The relative standard deviations betweenthe means of the parallel batches are less than +3.1%in all cases, which again can be considered as an indicatorof rather good repeatability regarding the performedexperiments and the sampling and analysis procedure.As an overall conclusion it seems that the size and shape

of sulphathiazole crystals is mostly affected by secondarynucleation. Although the cooling conditions applied inthis study varied considerably, the differences observed inthe properties of the produced crystals were rather small.It can therefore be concluded that most of the effectscaused by changes in cooling conditions could be almostcompletely overruled by the dominating influence ofattrition. Davey and Garside (2000) have stated that it isnowadays recognized that attrition nucleation is the mostsignificant nucleation mechanism in crystallizers formaterials with high or moderate solubility. This statementappears to be valid for sulphathiazole based on the resultsobtained in this study.

CONCLUSIONS

The effect of different cooling conditions on the super-saturation level and consequently on the outcome ofsulphathiazole crystallization was studied. The in situ con-centration measurements were carried out using an ATR-FTIR immersion probe and the concentration predictionswere calculated using a PLS calibration model. The productsize and shape characterization was done with an auto-mated image analysis technique which provided a methodto simultaneously evaluate the size and the shape of theproduct crystals in a plane of two dimensions.

The results presented in this paper demonstrated theaffect of cooling rate and mode on the supersaturationlevel during the unseeded crystallization process. As wasexpected, the supersaturation level increased with the cool-ing rate and a natural cooling mode resulted in a high initialsupersaturation peak. However, the concentration profilesobtained with the programmed cooling implied that thesupersaturation could not be maintained constant whichis probably due to the exponential solubility curve ofsulphathiazole. The product properties varied strongly interms of crystal size and shape and the observed variationscould be easily and comprehensively explained by thedifferences in the concentration profiles.

The results showed that by measuring an essential pro-cess parameter and elaborately characterizing the product,the conditions leading to certain product properties couldbe widely and extensively considered and also successivelyrelated to known theory of crystallization processes. Thiskind of information can be considered essential since itenables the optimization of the process conditions so thatthe product properties can be more efficiently controlled.

REFERENCES

Alander, E.M., Uusi-Penttila, M.S. and Rasmuson, A.C., 2003, Character-ization of paracetamol agglomerates by image analysis and strengthmeasurement, Powder Technol, 130(1–3): 298–306.

Bernard-Michel, B., Pons, M.N. and Vivier, H., 2002, Quantification, byimage analysis, of effect of operational conditions on size and shapeof precipitated barium sulphate, Chem Eng J, 87(2): 135–147.

Calderon De Anda, J., Wang, X.Z. and Roberts, K.J., 2005a, Multi-scalesegmentation image analysis for the in-process monitoring of particleshape with batch crystallizers, Chem Eng Sci, 60(4): 1053–1065.

Calderon De Anda, J., Wang, X.Z., Lai, X. and Roberts, K.J., 2005b,Classifying organic crystals via in-process image analysis and the useof monitoring charts to follow polymorphic and morphological changes,J Process Control, 15(7): 785–797.

Davey, R. and Garside, J., 2000, From molecules to crystallizers, 62–73(Oxford University Press Inc., New York, USA).

Table 2. Statistical values for the crystal samples obtained with different cooling modes.

Cooling mode

The number ofcrystals in the

sampleMean crystallength D[1,0]

Average crystallength+RSDa

Meanroundness

Averageroundness+RSDa

Linear Batch 1: 125 297 93.3 mm 95.4 mm+ 3.1% 0.460 0.456+ 1.4%Batch 2: 69 368 97.4 mm 0.451

Natural Batch 1: 113 512 91.9 mm 92.7 mm+ 1.3% 0.461 0.458+ 0.9%Batch 2: 99 795 93.6 mm 0.455

Programmed Batch 1: 124 545 89.0 mm 88.6 mm+ 0.7% 0.379 0.382+ 1.5%Batch 2: 147 742 88.2 mm 0.386

aRSD refers to the calculated relative standard deviation between the means of two similar batches.


58 POLLANEN et al.

Derdour, L., Fevotte, G., Puel, F. and Carvin, P., 2003, Real time evalu-ation of the concentration of impurities during organic solution crystal-lization, Powder Technol, 129(1–3): 1–7.

Doki, N., Seki, H., Takano, K., Asatani, H., Yokota, M. and Kubota, N.,2004, Process control of seeded batch cooling crystallization of themetastable a-form glycine using an in-situ ATR-FTIR spectrometerand an in-situ FBRM particle counter, Cryst Growth Des, 4(5):949–953.

Dunuwila, D.D. and Berglund, K.A., 1997, ATR FTIR spectroscopy forin situ measurement of supersaturation, J Cryst Growth, 179(1–2):185–193.

Faria, N., Pons, M.N., Feyo de Azevedo, S., Rocha, F.A. and Vivier, H.,2003, Quantification of the morphology of sucrose crystals by imageanalysis, Powder Technol, 133(1–3): 54–67.

Feng, L. and Berglund K.A., 2002, ATR-FTIR for determining optimalcooling curves for batch crystallization of succinic acid, Cryst GrowthDes, 2(5): 449–452.

Ferreira, A., Faria, N., Rocha, F., Feyo de Azevedo, S. and Lopes, A.,2005, Using image analysis to look into the effect of impurity concen-tration in NaCl crystallization, Chem Eng Res Des, 83(A4): 331–338.

Fevotte, G., 2002, New perspectives for the on-line monitoring of pharma-ceutical crystallization processes using in situ infrared spectroscopy, IntJ Pharm, 214(2): 263–278.

Fujiwara, M., Chow, P.S., Ma, D.L. and Braatz, R.D., 2002, Paracetamolcrystallization using laser backscattering and ATR-FTIR spectroscopy:metastability, agglomeration and control, Cryst Growth Des, 2(5):363–370.

Gron, H., Borissova, A. and Roberts, K.J., 2003, In-process ATR-FTIRspectroscopy for closed-loop supersaturation control of a batch crystal-lizer producing monosodium glutamate crystals of defined size, Ind EngChem Res, 42(1): 198–206.

Hakkinen, A., Pollanen, K., Louhi-Kultanen, M. and Nystrom, L., 2003,Pressure filtration of sulfathiazole crystallized by batch cooling crystal-lization: the influence of solvent composition, Chem Eng Trans,3: 931–936.

Hurley, L.A., Jones, A.G. and Hammond, R.B., 2004, Molecular packing,morphological modeling, and image analysis of cyanazine crystalsprecipitated from aqueous ethanol solutions, Cryst Growth Des,4(4): 711–715.

Jagadeshi, D., Kubota, N., Yokota, M., Sato, A. and Tavare, N.S., 1996,Large and mono-sized product crystals from natural cooling modebatch crystallizer, J Chem Eng Jpn, 29(5): 865–873.

Jones, A.G. and Mullin, J.W., 1974, Programmed cooling crystallization ofpotassium sulfate solutions, Chem Eng Sci, 29(1): 105–118.

Lewiner, F., Klein, J.P., Puel, F. and Fevotte, G., 2001a, On-line ATRFTIR measurement of supersaturation during solution crystallizationprocesses. Calibration and applications on three solute solvent systems,Chem Eng Sci, 56(6): 2069–2084.

Lewiner, F., Fevotte G., Klein, J.P. and Puel, F., 2001b, Improving batchcooling seeded crystallization of an organic weed-killer using on-lineATR FTIR measurement of supersaturation, J Cryst Growth,226(2–3): 348–362.

Lewiner, F., Fevotte, G., Klein, J.P. and Puel, F., 2002, An online strategyto increase the average crystal size during organic batch cooling crystal-lization, Ind Eng Chem Res, 41(5): 1321–1328.

Liotta, V. and Sabesan V., 2004, Monitoring and feedback control ofsupersaturation using ATR-FTIR to produce an active pharmaceuticalingredient of a desired crystal size, Org Process Res Dev, 8(3):488–494.

MacGregor J.F. and Kourti T., 1995, Statistical process control of multi-variate processes, Control of Engineering Practice, 3: 403–414.

Matthews, H.B. and Rawlings, B., 1998, Batch crystallization of aphotochemical: modeling, control and filtration, AIChE J, 44(5):1119–1127.

Mersmann, A., 1996, Supersaturation and nucleation, Chem Eng Res Des,74(A7): 812–820.

Mersmann, A., 1999, Crystallization and precipitation, Chem Eng Process,38(4–6): 345–353.

Mersmann, A. and Loffelmann, M., 2000, Crystallization and precipi-tation: the optimal supersaturation, Chem Eng Technol, 23(1):11–15.

Mullin, J.W. and Nyvlt, J., 1971, Programmed cooling of batch crystalli-zers, Chem Eng Sci, 26(3): 369–377.

Myerson, A.S., Decker, E. and Weiping, F., 1986, Solvent selectionand batch crystallization, Ind Eng Chem Process Des Dev, 25(4):925–929.

Pons,M.N., Vivier, H., Delcour, V., Authelin, J.-R. and Pailleres-Hubert, L.,2002, Morphological analysis of pharmaceutical powders, PowderTechnol, 128(2–3): 276–286.

Pollanen K., Hakkinen A., Reinikainen S.-P., Louhi-Kultanen, M. andNystrom, L., 2005, ATR-FTIR in monitoring of crystallization pro-cesses: Comparison of indirect and direct OSC methods, ChemometricsIntell Lab Syst, 76(1): 25–35.

Profir, V., Furusjo, E., Danielsson, L.-G. and Rasmusson A.C., 2002, Studyof the crystallization of mandelic acid in water using in-situ ATR-IRspectroscopy, Cryst Growth Des, 2(4): 273–279.

Srinivasakannan, C., Vasanthakumar, R., Iyappan, K. and Rao, P.G., 2002,A study on crystallization of oxalic acid in batch cooling crystallizer,Chem Biochem Eng Q, 16(3): 125–129.

Togkalidou, T., Braatz, R.D., Johnson, B.K., Davidson, O. and Andrews, A.,2001a, Experimental design and inferential modeling in pharmaceuticalcrystallization, AIChE J, 47(1): 160–168.

Togkalidou, T., Fujiwara, M., Patel, S. and Braatz, R.D., 2001b, Soluteconcentration prediction using chemometrics and ATR-FTIR spectro-scopy, J Cryst Growth, 231(4): 534–543.

Ulrich, J. and Strege, C., 2002, Some aspects of the importance of meta-stable zone width and nucleation in industrial crystallizers, J CrystGrowth, 237–239(Pt. 3): 2130–2135.

Uusi-Penttila, M. and Berglund, K.A., 1996, Spectroscopic monitoring ofenvironmentally benign anti-solvent crystallization, J Cryst Growth,166(1–4): 967–970.

Wold, S., Martens, H. and Wold H., 1983, The multivariate calibrationproblem in chemistry solved by the PLS method, Lecture Notes inMathematics, 973: 286–293.

Wold S., H. Antti, F. Lindgren, J. Ohman, 1998, Orthogonal signal correc-tion of near-infrared spectra, Chemometrics Intell Lab Syst, 44(1–2):175–185.

ACKNOWLEDGEMENTS

The National Technology Agency (TEKES, Finland), Orion Pharma(Finland), the Graduate School in Chemical Engineering (Finland) andthe Academy of Finland are acknowledged for their financial support.

The manuscript was received 13 April 2005 and accepted forpublication after revision 7 December 2005.



a study on batch cooling crystallization of...

Documents