DOI: 10.1021/cg1011988
Online Evaluation of Paracetamol Antisolvent Crystallization Growth Rate with Video Imaging in an Oscillatory Baffled Crystallizer
2011, Vol. 11 719–725
Cameron J. Brown and Xiongwei Ni* School of Engineering and Physical Science, Heriot-Watt University, Edinburgh EH14 4AS, U.K. Received September 9, 2010; Revised Manuscript Received January 6, 2011
ABSTRACT: We report a method for observing and quantifying antisolvent crystallization (acetone/water system) of paracetamol in an oscillatory baffled crystallizer (OBC). With the help of optical means, paracetamol crystals passing the light sheet are recorded by a CCD camera. This allows the mass of crystals to be tracked with time and in turn the instantaneous and overall averaged crystal mass growth rates to be evaluated. By the standard sampling method, the overall averaged crystal growth rate and the percentage recovery of crystals were also determined. This work shows that there is a high correlation in data between the optical and the sampling techniques, validating the accuracy and reliability of the system. Our data are also consistent with previous work in this field.
*Corresponding author. Tel: 0044 1314513781. Fax: 0044 1314513129. E-mail:
[email protected].
and fluctuations in the solvent composition, thus causing a disruption that will not be taken into account by the followed temperature or solvent composition profile, leading to out of specification crystals being generated. An alternative method would be to adjust the cooling or antisolvent addition rate online in response to the changes in the crystallization process. The first step in achieving this is the ability to accurately measure the crystal growth within the crystallizer, ideally noninvasively as not to affect the fluid mechanics within the device. Video imaging of crystallization processes has been utilized previously; for example, Scott et al.8 employed an in-line camera for qualitative monitoring of particle-particle interactions; De Anda et al.9 used an imaging technique to study crystal morphology; Wilkinson et al., 10 De Anda et al.,11 and Li et al.12 reported the use of online imaging for observing the shapes of crystals. Monitoring of crystal shape and size was further advanced by the work of Larsen et al.13 where they derived an algorithm to determine crystal size and shape based on characteristic lengths. However, their system only identified 50% of the crystals at low concentrations compared with that by manually analyzing the images using human eye; this figure decreased to 33% of the crystals at medium to high concentrations. The degree of false positives (the algorithm designated as crystals from a section of the image where there was none present) was also found to increase with concentration. Nevertheless the cumulative size distribution curve showed a high correlation to that of traditional measurements, highlighting the potential of video imaging. This potential was further exploited in a recent work by Larsen and Rawlings14 who showed high-resolution imaging could be used quite readily. The limitation is however in the algorithms utilized to analyze the images, in particular, relating to problems with image segmentation and particle overlap. The issue of overlap was identified by Larsen et al.15 when they implemented an algorithm to analyze real in situ images (as appose to simulated images in their other work 13,14 ) of high aspect ratio crystals. Again this illustrated that real time image analysis was possible for monitoring crystallization processes, although it is limited at high crystal concentrations. This issue would further be compounded when crystals of a low aspect ratio are analyzed where the crystal
r 2011 American Chemical Society
Published on Web 01/19/2011
1. Introduction Many fine/specialty chemicals and pharmaceuticals involve a crystallization step at some point in their manufacture, which typically involves organic solvents or a mixture of solvents. The solvent composition along with temperature and mixing affects the degree of the solubility of the crystals. As a result, this impacts the method of generating supersaturation. Supersaturation can be achieved through cooling, evaporation, reaction, or the addition of an antisolvent (or drowning out), with cooling and antisolvent being the most used processes in industry, an example of which is shown in Figure 1. Cooling crystallization occurs when a solution containing solute is cooled at a constant concentration of dissolved crystals. Once the solubility curve is crossed crystallization can occur (instantaneous if seeds are added to initiate nucleation or if the system is further cooled within the metastable limit). A similar process occurs for antisolvent crystallization; instead of cooling the system, antisolvent (e.g., water) is added to the solvent (e.g., acetone) at a constant temperature (Figure 1). This has the same effect as reducing the temperature in the cooling crystallization. In both cooling and antisolvent crystallization, solvent composition may also affect the nucleation rate as well as the relative growth rate of each crystal facet, resulting in influencing the crystal shape and size distribution.1 Solvent composition may also impinge on agglomeration of crystals2,3 as well as the degree of solvent incorporation resulting in effects on crystal purity.4 Because all these factors would contribute to the final specifications of the crystals produced, it is important to be able to measure and potentially control the rate at which supersaturation decreases during the process. Traditionally operating a pharmaceutical crystallization process is to follow a temperature or solvent composition profile determined previously in laboratory scale experiments.5,6 However, in practice kinetic and thermodynamic variables may vary considerably from batch to batch7 due to impurities
pubs.acs.org/crystal
720
Crystal Growth & Design, Vol. 11, No. 3, 2011
Brown and Ni
Figure 2. Experimental setup with a dual light sheet (gray shaded area represents one baffled cell and the image capture area).
Figure 1. Phase diagram for unseeded batch cooling and antisolvent crystallization.
shape is not obvious and distinct from the background image. They went on to suggest a dimensionless group as an indication of the degree of crystal overlap as a way of comparing potential image capture techniques. 16 Other researchers have also investigated methods to determine size distribution. Caillet et al.,17 for example, used a CCD camera on an internal probe to record images during crystallization and then determine the size distribution offline. However this also had problems with crystal overlap at high concentrations suggesting a limit of 8-10 wt % crystals. Eggers et al.18 produced size distribution algorithms based on the axial length distribution showing a high correlation to traditional size distribution methods. More recently Kempkes et al.19 expanded this by utilizing a second camera at a different angle to produce a 3D particle size distribution as well as quantitative information on particle shape; however, it was limited by having to compare with images from electron microscopy. It should be noted that all the aforementioned work by Larsen,13-16 Caillet,17 and Kempkes18,19 implemented a camera that only focused on a small portion of the sample, containing only a few crystals. Recent work by Simon et al.20 has applied bulk imaging to sample a larger portion of a crystallizer. They have also compared the use of bulk video imaging to the more widely used focused beam reflectance measurement (FBRM) for the detection of nucleation.21 However, images from the bulk in that work had only been used to determine the point of nucleation, not any growth-related parameters. Another common technique for measuring crystal mass or solid concentration is turbidimetry22,23 (light that is transmitted down a fiber optic reflected by a mirror positioned in the sample is compared with that returning from a second fiber). However, same as the aforementioned work, this only focuses on a local area of sample as small as the diameter of the probe, either within the system under investigation22 or outside the system by pumping the mixture out and into a measuring cell that contains turbidimetry.23 In order to work out the crystal mass growth rate indirectly, the measurement of local solute concentration will be used to represent the entire system on the assumption that the system is well mixed. Such an assumption may be valid for small laboratory crystallizers, would fail badly for larger laboratory or pilot scale vessels. This will lead to considerable errors in determining the mass growth rate. The significant advance and novelty of our work in comparison to the past publications is that we used bulk images covering either one cell of interest or the entire system to directly obtain the mass growth rate of crystals in a
crystallizer. This provides much more accurate mass growth data and is independent of scales. In addition, our external imaging system is nonintrusive. In this paper, we report, for the first time, our work on the development of such an online measurement of crystal mass growth rate during antisolvent (acetone/water system) crystallization of paracetamol in an oscillatory baffled crystallizer (OBC) utilizing a video imaging system that is illuminated by a laser light sheet. The idea of using a light sheet was stemmed from our previous work on particle image velocimetry (PIV)24,25 and laser-induced fluorescence (LIF)26,27 for measurement of velocity and tracer profiles. Our method could potentially be considered as a further extension of the bulk video imaging method of Simon et al.20 combined with the idea of estimating solid concentration from turbidity measurements.23 The purpose of using an OBC is to provide uniform mixing and a highly consistent environment28-31 with improved heat/mass transfer,32 which allows crystals of consistent quality33,34 that otherwise cannot be produced in traditional stirred tank crystallizers. 2. Experimental Setup, Operation, and Analysis Procedures Experimental Setup. An OBC of 630 mL (32 mm internal diameter) was used in this investigation, the details of which can be found elsewhere.35 The OBC consisted of a square jacketed column with five equally spaced stationary baffles (the provision of the square jacket is to minimize the curvature effect of the crystallizer when using optical equipment). Fluid oscillation was applied via a diaphragm driven by an electric motor at the base. Images were recorded direct to a PC hard drive by a gray CCD camera (SensiCam). Illumination for the images was supplied by a continuous 4 W argon-ion laser (Spectra Physics) that generates two light sheets, 1 mm thick by 60 mm, using a single beam splitter, five optical mirrors (Thor Lab), and a pair of laser line lenses (Edmund Optics). In order to minimize light gradient across the crystallizer both sides of the crystallizer were simultaneously illuminated through the center of the OBC, see Figure 2. Experimental Procedure. Distilled water (230 mL, antisolvent), acetone (Fisher Scientific, solvent), and paracetamol were prepared to their compositions as detailed in Table 1. Each sample was heated to 28 °C and well mixed to ensure a complete dissolution of the paracetamol crystals before being cooled to the operating temperature of 23 °C and then added to the OBC. Prior to imaging, the camera was positioned perpendicular to the crystallizer, covering one baffled cell, that is, the space between a pair of baffles. The OBC was then oscillated at a frequency of 2 Hz and an amplitude of 15 mm to promote uniform mixing in the OBC. Simultaneously images were captured continuously by the CCD camera at 7.5 frames per second directly onto a hard drive. After 5 min of recording, the addition of the antisolvent (water) began. This was fed into the OBC by a peristaltic pump (Watson Marlow) at the required rate as indicated in Table 1 until a maximum volume of 630 mL had been reached. When the antisolvent addition was stopped, the recording continued until very little change in images showed, typically 5-7 min after the addition had been terminated. In this way, the images
Article
Crystal Growth & Design, Vol. 11, No. 3, 2011
721
Table 1. Run Compositions and Parameters mass % run
water
acetone
solvent/antisolvent, mL/mL
paracetamol solubility, g/mL
solution concentration, g/mL
addition rate, mL/min
1 2 3 4 5 6 7 8
30 40 50 60 70 50 50 50
70 60 50 40 30 50 50 50
3.0 1.9 1.3 0.8 0.5 1.3 1.3 1.3
0.37 0.33 0.28 0.20 0.13 0.28 0.28 0.28
0.35 0.31 0.26 0.19 0.12 0.26 0.26 0.26
50 50 50 50 50 133 25 13
Figure 3. (A) Initial OBC image showing baffle support rods. (B) Image during low concentration crystal growth. (C) Final image showing high concentration of crystals. Shaded areas represent increase in image area due to crystal growth. Column summation = sum of pixel values along the image height; column number = pixel position. captured cover the precrystallization, current crystallization, and postcrystallization stages. Once the recording had been completed samples were filtered, dried for 24 h, and weighed to obtain the recovered paracetamol mass. It should be noted that the experimentally recovered mass will always be less than the theoretical value that is determined from the solubility curve due to loss of crystals in filtration as well as the retention of crystals in the vessel. It is this experimental recovered value that is taken as the maximum crystal mass for each test and for the subsequent evaluation of crystal growth rates. Image Analysis. Each recording was then loaded into Matlab for further processing. Due to the utilization of 8 bit gray scale images, each pixel can have an intensity value ranging from 0 (black denoting solution) to 255 (white denoting crystals). Initially before any antisolvent addition, the baseline image is black except pixels with a high value (white) representing the position of the baffle support rods, as shown in Figure 3A. As the antisolvent addition progressed, crystals began to appear. Note that the smallest detectable crystal size for the given camera positioning and resolution is 43 μm and is determined by: The smallest detectable crystal ¼
Captured image width 32 mm ¼ ¼ 43 μm Camera horizontal resolution 736 pixel
ð1Þ
With this in mind, the focus of our work was on crystal growth. With the help of better camera resolution and smaller area of
Table 2. Camera Position and Resolution Effects on Smallest Detectable Crystal (μm) Smallest Detectable Crystal (μm) horizontal resolution (px) captured area width (mm)
736
1024
1392
1600
10 20 32 40 50
14 27 43 54 68
10 20 31 39 49
7 14 23 29 36
6 13 20 25 31
investigation, much smaller crystal sizes can be detected, as shown in Table 2. This system could potentially be exploited in probing nucleation. In this work, crystals above this minimum size are shown in the images as pixels with an increased value (ranging from gray to white), see Figure 3B. Further increase in crystal concentration (due to the increased crystal sizes and numbers) led to an increase in the value and number of gray to white pixels, Figure 3C. For each individual image, it is divided into 736 columns of pixels (each consisting of 1024 pixels for a total of 753 664 pixels per image). Each column of pixels was then summed and plotted against the corresponding column number36 resulting in an intensity distribution curve. Since the summation of pixels is related to the distribution and variation of the crystals in the system, the integral under the intensity distribution would thus correspond to the area of
722
Crystal Growth & Design, Vol. 11, No. 3, 2011
Brown and Ni
the image with the presence of crystals under illumination. Before any antisolvent addition, the image graph shows two peaks (Figure 3A) corresponding to the white pixels caused by the presence of the baffle support rods. As the antisolvent addition progresses, crystals begin to form. There is an increase in the intensity distribution across the entire width of the column (Figure 3B) (the light gray area represents the increase due to crystal presence). Further increase in the crystal concentration leads to further increase in the integral distribution, see Figure 3C (dark gray area represents the increase due to further crystal formation with respect to that shown in Figure 3B). Calculating Crystal and Solution Concentration. Once the mass of the recovered paracetamol was measured experimentally, as mentioned earlier, it was then divided by the crystallizer volume; this was taken to be the maximum concentration of crystals, denoting to the maximum intensity distribution in Figure 3C. Similarly the minimum intensity distribution was equated to a crystal concentration of 0 g/mL in Figure 3A. The crystal concentration can then be calculated from ðarea - areamin ÞPr areamax - areamin Cc ¼ ð2Þ V
Figure 4. Typical solution concentration curve showing the start of crystallization, tc1, and the end, tc2.
where Cc = crystal concentration (g/mL), area = integral of intensity distribution curve (pixel2), areaMin = minimum integral of intensity distribution curve (pixel2), that is, corresponding to crystal concentration of 0 g/mL, areaMax = maximum integral of intensity distribution curve observed over the entire recording (pixel2), Pr = mass of recovered paracetamol as crystals (g), and V = volume of reactor (mL). If the initial mass of paracetamol (Pi) in solution is known, the concentration of paracetamol remaining in solution (Cs) can similarly be calculated from ðarea - areamin ÞPr Pi areamax - areamin ð3Þ Cs ¼ V Calculating Instantaneous Crystal Growth Rates from Image Data. Since the crystal concentration can be calculated at any point during the image recording, with eq 2, the instantaneous crystal growth rate, Gi (g/min), can then be calculated by C c2 - C c1 V avg ð4Þ Gi ¼ t2 - t1 where Cc1 = crystal concentration at time 1 (g/mL), Cc2 = crystal concentration at time 2 (g/mL), t1 = time 1 (min), t2 = time 2 (min), and Vavg = average reactor volume between times 1 and 2 (mL). Calculating Overall Average Growth Rate from Recovered Mass. By plotting the solution concentration, eq 2, against time in Figure 4, one can observe how the concentration of paracetamol evolves over time. From Figure 4, the starting time of crystal growth (tc1) is the point at which the solution concentration rapidly decreases. Similarly, the ending time of crystal growth is the point at which the solution concentration reaches a new low level, shown as tc2 in Figure 4. The slope of the solution concentration over the difference in time (tc2 tc1) in Figure 4 is the overall average crystal growth rate, shown as Pr Gavg, exp ¼ ð5Þ t c2 - t c1 Calculating Overall Average Growth Rate from Image Data. By averaging the instantaneous growth rates, eq 4, over the same period of (tc2 - tc1) at which the crystal growth is known to occur, one can also calculate an overall average growth rate, where n is the number of data points between tc1 and tc2: t c2 X Gi Þ=n Gavg;image ¼ ð
ð6Þ
t c1
The comparison of the growth rates between eqs 5 and 6 would validate the data generated from the direct image methodology. It can be seen that no preprocessing was applied to the images before determining the image areas. The reflections of the baffle support
Figure 5. Determined solubility data compared with that of Granberg and Rasmuson37 varying with (A) solvent/antisolvent ratio and (B) water mass %. rods are present in all images and analyzed accordingly. This is another advantage of this system.
3. Results and Discussion Shown above in Table 1 are the initial solvent/antisolvent compositions, paracetamol concentration and antisolvent addition rates for the eight trials carried out. Runs 1 through 5 were to investigate the effect of the starting composition on the recovery and crystal formation rates at a constant addition rate of 50 mL/min. Whereas the runs 3 and 6 through 8 were used to determine the effect of the addition rate on the recovery and crystal formation rates. A 50/50 w/w% water/ acetone composition was chosen for these runs as this composition gave a higher percentage recovery as detailed later.
Article
Crystal Growth & Design, Vol. 11, No. 3, 2011
723
Figure 7. Deviation between sampling and image results.
Figure 6. Comparison of overall average crystal growth rates from (O) standard experimental method and (b) recorded image data varying with (A) initial solvent composition and (B) antisolvent (water) addition rate.
Figure 5 shows the solubility for various ratios of solvent/ antisolvent solution determined using eq 3, as well as for different percentages of water content. As one can see, there is a high correlation among the data from this work and that of Granberg and Rasmuson,37 suggesting that our optical method of determining crystal concentration is quite accurate over this range of crystal concentrations tested. Further agreement between our image and experimental data can be seen in the high P values from the results of t tests shown in later figures. As the solubility curve is the key to antisolvent crystallization, the good agreement of our image data with the preart provides the confidence in applying the optic methodology to evaluate other crystallization parameters, such as the rate of crystal growth and the mass of recovered crystals. Figure 6 shows how the overall average crystal growth rate varied with the initial solution composition and the antisolvent addition rate for the experimental method (open cirlces) and the image method (filled circles). Note that error bars for the latter were calculated from the standard deviation of the image compared with that of 10 frames before and after and were found to be on average (0.5 g/min (compared with (0.25 g/min for the results from experimental sampling). P values from an unpaired t test between the experimental and image results are also given in Figure 6. For the effect of different mass composition (Figure 6A), both sets of data followed a similar trend, that is, initially an increase in the growth rate with an increase in the initial water up to 50 mass % and decrease thereafter. The trend is consistent with the work by Granberg and Rasmuson1 in which they showed the rate of crystal face growth (m/s) reached a maximum at around 60 mass % water. At the lower initial water fractions (30, 40 mass %), a higher overall mass of
crystals is expected due to higher initial solubility as indicated in Figure 5B. This would lead to crystal overlap, resulting in the evaluated growth rate being smaller than that by the experimental method. The growth rate reached a maximum at 50 mass % of the initial water; further increase in the initial water content led to a decrease in the growth rate, due to first the rate at which solubility changes with water content being less; second the change in solubility between the initial and final water fractions being smaller, for example, a change in solubility of 0.34 g/mL for the 30 mass % trial compared with a change of 0.11 g/mL for the 70 mass % trial. In terms of the effect of the addition rate (Figure 6B) on the average crystal growth rate, an increase trend is seen. This is expected with rapid addition of antisolvent a larger change in solubility is realized, hence faster growth of crystals. An interesting point to note is that the deviation of the growth rate for high final crystal concentrations under 30 or 40 initial water mass % and at high addition rate of 133 mL/min determined from the image analysis is significantly larger than from the sampling method. By plotting the growth rate obtained by the image method against that by the experimental method in Figure 7, we see that the deviation clearly increases toward the right side of graph. This is consistent with those shown in Figure 6A,B. This increase in deviation would be caused by crystal overlapping within the image observation area. Although the thickness of the illuminated plane is 1 mm, while paracetamol crystals are an order of magnitude smaller, it would not be unreasonable for multiple crystals to overlap within this thin layer, resulting in lower reading than expected. This effect is clearly illustrated in Figure 8, which shows how the intensity distribution integral increases with solid concentration up to a maximum value before leveling off. It is after this maximum that the effect of crystal overlap is most significant. To correct for such an effect in the addition rate runs a linear trend was fitted to the trials with the smallest deviation between the sampling and image results, for example, 13, 25, and 50 mL/ min. This trend was then used to calculate new growth rates and the resulting percentage of crystal recovery, as in Figure 9B. Figure 9 shows how the percentage recovery of paracetamol crystals varied with the experimental conditions for both the imaging and experimental methods; P values from results of the unpaired t test are also provided. For the effect of different mass composition (Figure 9A), both sets of data followed a similar trend as that observed for the growth rate (Figure 6A), that is, initially an increase in the recovery with an increase in the initial water up to 50 mass % and decrease thereafter.
724
Crystal Growth & Design, Vol. 11, No. 3, 2011
Figure 8. Variation of intensity distribution integral with solid concentration.
Brown and Ni
average rate of crystal mass growth. When compared with the solubility data from the standard sampling method and previously published data,37 the video imaging results exhibited a high degree of agreement. This was also confirmed by high P values from the results of the unpaired t tests comparing the experimental sampling and the image analysis. This study also showed that the rate of the overall crystal mass growth in the OBC varied with the initial sample composition in a similar trend to that previously observed in stirred batch reactors,1 while the rate of antisolvent addition had little influence on the rate of crystal mass growth. It should be noted that our system currently measures crystal growth on a mass basis rather than size as mentioned for all other optic methods; measuring the mass does allow the setup to be less sensitive to aspects such as image resolution. There are some limitations for this setup, such as the ability to detect smaller crystals (
