Automated Direct Nucleation Control in Continuous Mixed Suspension

Article pubs.acs.org/crystal

Automated Direct Nucleation Control in Continuous Mixed Suspension Mixed Product Removal Cooling Crystallization Yang Yang,† Liangcheng Song,†,‡ and Zoltan K. Nagy*,† †

School of Chemical Engineering, Purdue University, West Lafayette, Indiana 47907, United States School of Chemical Engineering & Technology, Harbin Institute of Technology, Harbin 150001, P. R. China

‡

ABSTRACT: Automated direct nucleation control (ADNC) is a focused beam reflectance measurement-based feedback control approach that is used to produce large and uniform crystals in batch cooling crystallization. In this work, this feedback control approach is modified to work for continuous cooling crystallization processes, including single-stage and multistage mixed suspension mixed product removal crystallizers. The proposed continuous ADNC approach is observed to provide quick startup, high quality control of crystal size distribution, as well as automated and effective disturbance suppression.

1. INTRODUCTION Crystallization is a commonly used purification technique in the pharmaceutical industry. The United States Food and Drug Administration (FDA) introduced guidance related to the use of quality-by-design (QbD) and process analytical technology (PAT) to encourage innovations in development and manufacturing of pharmaceutical products.1−3 In the QbD concept, typically a process is first designed based on the understanding of the system to achieve a desired product quality, and then monitored and controlled.4 Unlike the “design then control” strategy applied in QbD, we would like to propose a quality-by-control (QbC) concept, which encourages the “design via control” strategy by using closed-loop (feedback) control approaches to determine operating trajectories. The suitable closed-loop control approaches can reduce the influence of disturbances and decrease process and product variations and therefore increase the applicable design space. Because of the development of online PAT tools,3,5−9 some closed-loop control approaches have been developed, and the QbC concept has been implemented in batch crystallization in recent years.10−19 The critical quality attributes (CQA) of a crystallization process include yield, purity, polymorphic form, crystal shape, and crystal size distribution (CSD).20 The control of CSD is particularly important since CSD can significantly influence the efficiency of downstream operations such as filtration and drying.21,22 Typically large and uniform crystals are desired, whereas fine particles are unwanted since they can cause problems in filtration and drying. A well-controlled CSD may even help avoid some downstream procedures such as milling or granulation. However, precise control of the CSD poses significant challenges. One main challenge is due to the stochastic nature of nucleation. On the basis of the QbC concept and application of PAT tools, two model-free closed© XXXX American Chemical Society

loop control approaches have been proposed to solve this challenge and control CSD in batch cooling crystallization. The first approach is focused beam reflectance measurement (FBRM)-based feedback control, which is known as automated direct nucleation control (ADNC).12,15,16 ADNC uses in situ particle chord counts information measured by FBRM to automatically detect nucleation and actuate adaptive heating or cooling rate to maintain particle chord counts in a desired setpoint range during the crystallization process. This approach typically ends up with a few thermo cycles that provide controlled dissolution of fine particles during the crystallization. The second approach is called supersaturation control (SSC), in which UV or IR spectroscopy is used to measure concentration in real-time.14,17,23,24 This information is then used in a feedback control loop to control the cooling temperature in order to maintain the supersaturation level at a desired set-point value after seeds are introduced. SSC is able to ensure that the system is operated within the metastable zone so that nucleation can be avoided or minimized. Both the above two approaches have been found to be able to avoid formation of fine particles and produce large and uniform crystals in batch cooling crystallization. These control approaches are also more robust than simple open-loop temperature control which cannot overcome the effects of disturbances. In addition, these two control approaches do not require any complicated first principle mathematical model, such as population balance model (PBM), and therefore are simple to design and implement in batch crystallization. Continuous crystallization is considered as a promising technology for the production of pharmaceuticals and fine Received: August 24, 2015 Revised: November 5, 2015

A

DOI: 10.1021/acs.cgd.5b01219 Cryst. Growth Des. XXXX, XXX, XXX−XXX

Crystal Growth & Design

Article

Figure 1. Experimental setup of continuous two-stage MSMPRC.

either without process control or with open-loop control only. In addition, the vast majority of works described above that focus on optimization and control aspects related to MSMPRC are theoretical simulation studies only. Therefore, the experimental implementation of some practical control methodology is necessary in the development of continuous crystallization, to bring theoretical advances in the realm of possibilities. According to our knowledge, this paper is the first that implements a closed-loop (feedback) control in a continuous crystallization process in practice using PAT tool. In this work, a FBRM-based feedback control approach, which is modified from batch ADNC12,15,16 to accommodate the requirement for steady-state operation in continuous crystallization, is proposed and implemented in single-stage and two-stage continuous MSMPRC. An active pharmaceutical ingredient (API), paracetamol, is used as the model compound in ethanol solvent in a continuous cooling crystallization process.

chemicals since it can overcome the intrinsic disadvantages of batch crystallization, including batch to batch inconsistency, low process and equipment efficiency, and high operating cost.25−40 This is because continuous crystallization works with constant operating conditions (e.g., temperature) at steadystate, or more accurately quasi-steady operation with small and controlled variations in output quality. Although steady-state is never achievable during any functioning unit operation due to disturbances, this term will be used in the paper as a shorthand. Because of the steady-state (quasi-steady state) operation, continuous processes are in general simpler to control compared to batch processes. Mixed suspension mixed product removal crystallizer (MSMPRC) is a term defined for a continuous stirred tank reactor (CSTR) that is specially used for crystallization. It is simple to operate and only requires a small technology change from batch.34 It can work either as a single-stage or as part of a multistage cascade. Some research studies were carried out to investigate the influence of process operation conditions (e.g., temperature, antisolvent addition rate, residence time) on steady-state yield, purity, and CSD in MSMPRC in experimental25−31,39,40 and simulation studies.32−35 Vetter et al. calculated the attainable regions of average particle size at steady-state under various residence times in MSMPRC of different number of stages via PBM.34 Yang and Nagy proposed a nonlinear model predictive control approach based on the PBM of the cascade and proved the feasibility of its application in continuous MSMPRC in simulation.32 Besides the design and control of steady-state operation, startup procedure is another important challenge in continuous crystallization since time and product are wasted during startup. However, only a few works have investigated the startup of MSMPRC.26,33,35 Hou et al. observed that seeding the crystallizer with crystals generated from the previous run could shorten the startup durations.26 Su et al. developed a general and rigorous mathematical model for continuous MSMPRC and proposed a concentration-control strategy to facilitate the convenient design of the startup procedure.35 Yang and Nagy demonstrated via simulation that the startup duration can be significantly reduced by using a dynamic operating profile using open-loop optimization techniques.33 At the end of this paper, a closed-loop control approach to reduce startup duration was proposed but not implemented. In general, the works that have been done so far in continuous MSMPRC are

2. EXPERIMENTAL SECTION The experiments were carried out in a continuous two-stage MSMPRC (Figure 1). Three 1 L jacketed round-bottom crystallizers were used as feed vessel, first stage MSMPRC, and second stage MSMPRC, respectively. The jackets of the three vessels were connected to three temperature circulators (Ministat 125 with Pilot ONE, Huber, USA), in which a jacket control (internal control) model was selected. Each MSMPRC stage was equipped with a focused beam reflectance measurement (FBRM) probe (S400, Lasentec; G400, Mettler-Toledo, USA) and a Pt100 thermocouple. The particle chord counts and process temperature in each stage were measured every 15 s and transferred into a Labview based software Crystallization Monitoring and Control (CryMOCO) or Crystallization Process Informatics System (CryPRINS), which was installed on a supervisory computer. The real-time data were then used in the ADNC approach to automatically modify temperature set-points in the temperature circulators through CryMOCO. The data were communicated via an RS232 interface. A 38 °C saturated paracetamol (purity >98%; Alfa Aesar, USA) − ethanol (purity >99%; KOPTEC, USA) solution (0.26 g/g ethanol) was used as feed solution.41 This clear feed solution was transferred from the feed vessel into the first stage MSMPRC through a peristaltic pump (L/S digital pump, Masterflex, USA). The slurry obtained in the first stage was then transferred into the second stage semicontinuously via a transfer unit,26 which consists of a transfer zone, an inlet valve, an outlet valve, a nitrogen valve, and a vacuum valve. The slurry could be B



Article

first drawn in from the previous stage into the transfer zone by a vacuum, and then transferred into the next stage via high pressure nitrogen gas. Another transfer unit was used to transfer the product slurry from the second stage MSMPRC back into the feed vessel for recycle in order to save materials.31 The solution in the feed vessel was kept at around 1 L and 55 °C. The recycled crystals were observed to dissolve immediately once they entered into the feed vessel. In addition, the feed solution was used as the initial solution in both MSMPRC vessels. Therefore, theoretically the recycled slurry should have the same concentration as the feed after all solids dissolve, and the influence of recycle on feed concentration was negligible. The transfer intervals of the two transfer units were controlled in a controller box to maintain the slurry volumes constant in both stages (i.e., 300 mL). At the beginning of every experiment, 0.10 g of paracetamol fine powders (1100 counts/s) were introduced into the first stage to trigger secondary nucleation. ADNC was switched on at around 0.25 RT, right after seed crystals were introduced. The purpose is to ensure that the ADNC approach controls secondary nucleation rather than primary nucleation, since secondary nucleation is generally dominant in MSMPRC systems.

Figure 3. CryMOCO user interface for ADNC implementation.

point will remain constant once the measured total chord counts falls between the upper and the lower bounds. In addition, maximum and minimum temperature set-points allowed can also be defined in ADNC in CryMOCO (Figure 3). In this paper, a series of experiments were carried out in single-stage and two-stage MSMPRC, with and without using ADNC approach. In the two-stage MSMPRC the use of ADNC in both stages were also investigated and compared to when ADNC is only used in the second stage. The detailed operating conditions and ADNC parameters are summarized in Table 1.

3. CONTINUOUS ADNC METHODOLOGY In the proposed continuous ADNC approach, each stage is controlled individually by an ADNC loop. For a certain MSMPRC stage, an upper bound and a lower bound of particle total chord counts set-point are defined by the user in the ADNC setup in CryMOCO. During the continuous crystallization, once the particle total chord counts measured by the FBRM in that MSMPRC stage is found higher than the upper bound, a fast heating will be actuated automatically by the ADNC to reduce nucleation rate or trigger controlled dissolution (Figure 2) in that stage. The fast heating will

Table 1. Operating Conditions and ADNC Parameters Used in Every Experiment ADNC parameters

Figure 2. ADNC set-point and heating/cooling algorithm.

change to slow heating once the measured total counts drops below upper bound. Similarly, fast cooling or slow cooling is implemented automatically by ADNC if the measured total chord counts is below lower bound or increases above lower bound. The CryMOCO user interface for ADNC implementation is presented in Figure 3, from which the total chord counts set-point upper bound, lower bound, fast heating rate, and fast cooling rate can be specified. Slow heating and slow cooling rates are defined from heating gain (Kh) and cooling gain (Kc) (Figure 3) based on the following equations: slow heating rate =

fast heating rate Kh

(1)

slow cooling rate =

fast cooling rate Kc

(2)

no.

no. of stages

residence time in each stage [min]

1 2 3 4 5

single-stage single-stage two-stage two-stage two-stage

20 20 20 20 75

6

two-stage

20

ADNC is used in which stage no ADNC first no ADNC second first and second second

fast heating/ cooling rates [°C/min]

minimum/ maximum temperatures [°C]

± 0.1

3−50

± 0.1 ± 0.05

3−50 15−26

± 0.1

3−50

4. RESULTS AND DISCUSSION 4.1. Startup and Control of CSD. 4.1.1. Continuous ADNC in Single-Stage MSMPRC. Experiments 1 and 2 are from single-stage MSMPRC. In experiment 1, ADNC is turned off and the jacket temperature of MSMPRC is kept constant at 25 °C. In experiment 2, the initial jacket temperature of MSMPRC is also 25 °C, but ADNC is switched on immediately after seed crystals are introduced. The process temperature, ADNC jacket set-point, FBRM total chord counts, ADNC total counts setpoint, and square weighted mean chord length (SWMCL) of these two experiments are compared in Figure 4. It is observed that the total counts increase slowly and 6 residence time (RT) is required to reach steady-state in experiment 1 when ADNC is not used. However, in experiment 2 when ADNC is turned on, the temperature of MSMPRC decreases automatically at the beginning of the startup to force the system to nucleate faster and reach the desired total counts set-point sooner. As a result, only 4 RT is spent to reach steady-state in experiment 2 (Figure 4b). Therefore, ADNC is able to reduce significantly

Since steady-state operation is desired in continuous crystallization, extremely large heating and cooling gains (i.e., 255) are used in this work so that slow heating and slow cooling rates become practically zero. In other words, the temperature setC



Article

Figure 4. Process temperature, ADNC jacket set-point, FBRM total chord counts, ADNC total counts set-point and square weighted mean chord length (SWMCL) in single-stage MSMPRC: (a) without ADNC (experiment 1), (b) with ADNC (experiment 2).

mixing conditions. In our setup, the slurry is transferred using high pressure nitrogen gas, which can cause short spikes and variations in the FBRM measurements even under practically steady-state operation. In addition, in experiment 2 the ADNC set-point is modified in CryMOCO after the system reaches steady-state 1. The purpose is to characterize the ADNC performance when a setpoint change is required, as well as to compare the crystal chord length distribution (CLD) obtained at steady-states at different set-points. It is found that only 0.7 RT is required for ADNC to move the system from steady-state 1 into steady-state 2 (Figure 4b). The CLD obtained in both steady-states are compared in Figure 5. The results indicate that when ADNC is used, larger ADNC total counts set-point results in smaller average crystal size and more fine particles. This is probably because paracetamol is a nucleation dominated compound. This phenomena agrees with a simulation study in the literature which shows that there is a trade-off between average crystal size and yield within fixed MSMPRC setup and operating constraints.32 This observation also provides an approach to control product CSD according to formulation requirements by manipulating the ADNC set-point. It is considered that once the experimental setup, feed concentration and residence times are fixed, the average particle size is a function of particle number only. In order to reach the target average particle size, the first step is to find out a target particle number, which corresponds to the target average particle size. This would require a few preliminary experiments to determine a correlation or calibration relationship, which then can be used to obtain the target particle count corresponding to a desired size. Once the target particle number is found out, the

Figure 5. Crystal chord length distribution (CLD) at steady-state 1 and steady-state 2 in experiment 2 when ADNC is used in single-stage MSMPRC.

the startup duration in a single-stage MSMPRC. It should be noted that the steady-state is identified qualitatively based on both the total counts and mean chord length. For experiments without ADNC, steady-state is considered reached when both total counts and mean chord length do not show any clear increasing or decreasing trend. For the experiments with ADNC, steady-state is considered reached when the total counts is kept closely between or around the maximum and minimum values of the total counts set-points, and mean chord length does not show any clear increasing or decreasing trend. The steady-state criteria cannot be mathematically quantified here due to the nature of FBRM measurement, which shows high sensitivity to noise or disturbance, such as fouling and particles adhered on probe window, as well as variations in the D



Article

Figure 6. Crystal microscope images of (a) experiment 1, (b) experiment 2 steady-state 1, and (c) experiment 2 steady-state 2.

Figure 7. Process temperature, ADNC jacket set-point, FBRM total chord counts, ADNC total counts set-point and SWMCL in the second stage: (a) without ADNC (experiment 3), (b) with ADNC implemented in the second stage (experiment 4). The variation in total counts between 3.5 and 4 RT is due to probe fouling and cleaning.

E



Article

steady-state is reached. This is because the ADNC thermo cycles can dissolve fine particles and produce some very large particles,12,15,16 which are very difficult to be washed out. 4.1.2. Continuous ADNC in Two-Stage MSMPRC. Twostage MSMPRC is used in experiment 3, 4, and 5. Experiment 3 is the base case in which ADNC is not used and the jacket setpoints of the first and the second MSMPRC are kept constant at 25 and 19 °C, respectively. Since the advantages of ADNC on the first stage have already been observed in section 4.1.1, the effects of applying ADNC on the second stage is investigated in experiment 4, in which the jacket set-point of the first stage is kept constant at 25 °C and the initial jacket setpoint of the second stage is set as 19 °C before turning on ADNC. The second stage of these two experiments is compared in Figure 7. It is observed that the startup duration can be reduced from 5 RT to only 2.5 RT when ADNC is used in the second stage. In addition, sometimes sharp jumps of total counts or mean chord length may be noticed, such as total counts between 3.5 RT to 4 RT in Figure 7b. This part of the data is actually a false signal caused by fouling and crystals adherring to the window of the FBRM probe. The total counts signal returns to its original value immediately after the probe window is cleaned at 4 RT. Since the ADNC is based on the FBRM total counts measurement only, the false signal of mean chord length will not influence the control performance; however the false signal of total counts can disrupt the control algorithm. Therefore, it is very important to ensure that the FBRM window is clean during operation. The window must be cleaned as soon as abnormal data jump is observed in the signal. A set-point change in ADNC is also implemented in experiment 4 in the second stage (Figure 7b). Around 2.3 RT is used for the system to move from steady-state 1 to

Figure 8. CLD at steady-state 1 and steady-state 2 in experiment 4 when ADNC is used in the second stage.

target average particle size can be achieved by controlling the particle number using the ADNC approach. In general, for the paracetamol system it is observed that a relatively small ADNC set-point should be used to avoid strong nucleation when fine crystals are not desired, and vice versa. The crystal images (Figure 6) confirm that more fine particles are generated in steady-state 2 than steady-state 1. In addition, theoretically as long as the final operating temperatures or particle numbers are fixed, whether the continuous ADNC approach is used or not during startup, or how the steady-state is reached will not have any effect on the final steady-state CSD or average crystal size, since all the crystals are continuously washed out, and once the ADNC set-point stabilized the steady-state CSD will be reached. However, under certain conditions, in practice it might be observed that using the ADNC during startup leads to larger average size than without ADNC (Figure 6) when

Figure 9. Crystal microscope images of (a) experiment 3, (b) experiment 4 steady-state 1, and (c) experiment 4 steady-state 2. F



Article

Figure 10. Process temperature, ADNC jacket set-point, FBRM total chord counts, ADNC total counts set-point and SWMCL in (a) first stage with ADNC (experiment 5), (b) second stage with ADNC (experiment 5).

simultaneously in more than one stage. The RT of each stage in experiment 5 is 75 min. Two ADNC loops are used in the first and the second stage simultaneously. The FBRM and temperature data of both stages are presented in Figure 10. The startup durations spent in the first and the second stage are only 2.4 RT and 3 RT, respectively. This observation once again shows the ability of the ADNC approach to provide quick startup. Two steady-state operating conditions are evaluated in experiment 5 by changing the ADNC set-point in the second stage. In this case it takes only about 0.3 RT for the system to go from steady-state 1 to steady-state 2 (Figure 10b). The ability of the ADNC approach to control the CSD is also observed here based on the comparisons of CLD (Figure 11) and microscope images (Figure 12) in the two steady-states. The results indicate that large and uniform crystals (Figure 12a) can be obtained by the ADNC if the set-point is properly chosen, whereas fine particles may be generated (Figure 12b) if the set-point is selected to be at larger counts/sec. The experiments previously discussed indicate that the length of residence time will not limit the performance of ADNC. In addition, ADNC can be used in any one of the stages or in multiple stages simultaneously. Although only twostage MSMPRC is implemented here, in principle there should be no problem to apply ANDC in MSMPRC of more than two stages. Additionally, this approach allows control of nucleation quantitatively in every stage if ADNC is applied in multiple stages and the FBRM data are calibrated suitably. For example, nucleation and the generation of fine particles in the second stage might be completely avoided by selecting the same ADNC set-point in both the first and the second stages. More

Figure 11. CLD at steady-state 1 and steady-state 2 in experiment 5 when ADNC is used in both stages.

steady-state 2. The CLD obtained from these two steady-states are compared in Figure 8, which indicates that in the ADNC approach, smaller ADNC set-point results in larger average crystal size and less fine particles for paracetamol. Therefore, similarly as in the case of the single-stage system, the CSD can be controlled in two-stage MSMPRC by manipulating the ADNC set-point in the second stage. Crystal images in Figure 9 also show that average crystal size is increased from steady-state 1 to steady-state 2 due to a decreased ADNC set-point. In previous experiments, the RT in each stage is relatively short (i.e., 20 min), and ADNC is used only in one of the two stages. Therefore, experiment 5 is performed in order to test the feasibility of applying ADNC in systems that have long residence time, as well as the possibility of using ADNC G



Article

Figure 12. Crystal microscope images of (a) experiment 5 steady-state 1, (b) experiment 5 steady-state 2.

Figure 13. Process temperature, ADNC jacket set-point, FBRM total chord counts, ADNC total counts set-point and SWMCL in experiment 6 with disturbances introduced artificially.

stages should be as large in order to maximize nucleation and minimize growth. 4.2. Disturbance Suppression. Since ADNC is a feedback control approach, in principle it should be able to help suppress disturbances. Experiment 6 is a follow up of experiment 3 after steady-state is reached. As shown in Figure 13, disturbance 1 and disturbance 2 are the same artificial pulse disturbances of adding 0.5 g of paracetamol fine powders into the second stage to disrupt the steady-state. This could be for example a practical disturbance that may occur from accidental seeding from an encrust layer that may form on the walls or stirrer shaft during the continuous crystallization operation. ADNC is turned off when disturbance 1 is introduced but turned on before introducing disturbance 2. It is found that 1.7 RT is required to wash out the fine particles without ADNC. However, when ADNC is used, the MSMPRC can automatically heat up to dissolve the fine particles; therefore they are removed much more quickly. Only 0.9 RT is needed to completely suppress the same disturbance by using ADNC. The CLD in Figure 14 shows that ADNC can provide exactly the same good performance as the simple washing out approach to adjust the CLD back into the original steady-state. But less time is required and less products are wasted if ADNC is used. In addition, ADNC should have advantages on suppressing not only pulse disturbance but also continuous disturbance (e.g., step disturbance). For example, if a feed solution that has higher concentration than nominal concentration is used, ADNC should be able to detect the increased nucleation rate and therefore increase the MSMPRC temperature accordingly to maintain the total particle chord counts within the ADNC

Figure 14. CLD at different times in experiment 6 with disturbances introduced artificially.

generally, in a multistage MSMPRC with the ADNC applied and FBRM data calibrated in all stages, for a given starting solution concentration or saturation temperature, if large particle size is desired, same or similar maximum/minimum total counts set-points should be used in all stages so that nucleation after the first stage can be minimized or completely avoided. The values of these total counts set-points can be then tuned based on the yield requirement. On the other hand, if small particle size and high yield are required, the maximum/ minimum total counts set-points of the second or subsequent H



■

set-point and as a result maintain the average crystal size. However, the system without ADNC cannot react to the feed concentration increase. It will end up with a smaller crystal size than desired. The operating temperature has to be redesigned manually in order to work for a different feed concentration without ADNC. Therefore, ADNC can significantly improve the robustness of continuous MSMPRC. Additionally, this study shows that the proposed closed-loop ADNC approach is stable, and total counts can converge to the desired set-point under selected control parameters. However, it is observed that this closed-loop approach becomes unstable when too large heating and cooling rates are used. There are two potential causes for instability under such aggressive parameters. First of all, the actual process temperature will have significant delay compared to the calculated temperature set-point. Additionally, the temperature change will be too fast compared to the crystallization kinetics for the process to be able to respond. Therefore, a few preliminary experiments are required to tune the control parameters when a new experimental setup or compound is introduced for ADNC application. Finally, as already described in introduction, SSC is another widely used feedback control approach that uses a concentration measurement to adjust the temperature set-point in batch cooling crystallization. Since in this work ADNC is observed to have superior performance in continuous crystallization, we anticipate that SSC can be used in a similar way to reduce startup duration, control CSD, and suppress disturbance in continuous MSMPRC. Low cost PAT imaging tools can also be used in the process as a supplement for FBRM or UV/vis spectroscopy to monitor and control CSD.42

REFERENCES

(1) Guidance for Industry PAT  A Framework for Innovative Pharmaceutical Development, Manufacturing and Quality Assurance; U.S. Food and Drug Administration/Department of Health and Human Services; Rockville, MD, 2004. (2) Icten, E.; Giridhar, A.; Taylor, L. S.; Nagy, Z. K.; Reklaitis, G. V. J. Pharm. Sci. 2015, 104, 1641−1649. (3) Simon, L. L.; Pataki, H.; Marosi, Gy.; Meemken, F.; Hungerbühler, K.; Baiker, A.; Tummala, S.; Glennon, B.; Kuentz, M.; Steele, G.; Kramer, H. J. M.; Rydzak, J. W.; Chen, Z.; Morris, J.; Kjell, F.; Singh, R.; Gani, R.; Gernaey, K. V.; Louhi-Kultanen, M.; O’Reilly, J.; Sandler, N.; Antikainen, O.; Yliruusi, J.; Frohberg, P.; Ulrich, J.; Braatz, R. D.; Leyssens, T.; von Stosch, M.; Oliveira, R.; Tan, R. B. H.; Wu, H.; Khan, M.; O’Grady, D.; Pandey, A.; Westra, R.; Delle-Case, E.; Pape, D.; Angelosante, D.; Maret, Y.; Steiger, O.; Lenner, M.; Abbou-Oucherif, K.; Nagy, Z. K.; Litster, J. D.; Kamaraju, V. K.; Chiu, M.-S. Org. Process Res. Dev. 2015, 19, 3−62. (4) Pharmaceutical Quality for the 21st Century: A Risk-Based Approach; U.S. Food and Drug Administration/Department of Health and Human Services: Silver Springs, MD, 2007. (5) Hermanto, M. W.; Chow, P. S.; Tan, R. B. H. Cryst. Growth Des. 2010, 10, 3668−3674. (6) Hermanto, M. W.; Chow, P. S.; Tan, R. B. H. Ind. Eng. Chem. Res. 2012, 51, 13773−13783. (7) Simone, E.; Saleemi, A. N.; Nagy, Z. K. Org. Process Res. Dev. 2015, 19, 167−177. (8) Simone, E.; Saleemi, A. N.; Nagy, Z. K. Chem. Eng. Res. Des. 2014, 92, 594−611. (9) Simone, E.; Saleemi, A. N.; Nagy, Z. K. Chem. Eng. Technol. 2014, 37, 1305−1313. (10) Nagy, Z. K.; Fevotte, G.; Kramer, H.; Simon, L. L. Chem. Eng. Res. Des. 2013, 91, 1903−1922. (11) Nagy, Z. K.; Braatz, R. D. Annu. Rev. Chem. Biomol. Eng. 2012, 3, 55−75. (12) Abu Bakar, M. R.; Nagy, Z. K.; Saleemi, A. N.; Rielly, C. D. Cryst. Growth Des. 2009, 9, 1378−1384. (13) Ma, C. Y.; Wang, X. Z. J. Process Control 2012, 22, 72−81. (14) Saleemi, A. N.; Onyemelukwe, I. I.; Nagy, Z. K. Front. Chem. Sci. Eng. 2013, 7, 79−87. (15) Saleemi, A. N.; Rielly, C. D.; Nagy, Z. K. CrystEngComm 2012, 14, 2196−2203. (16) Saleemi, A. N.; Rielly, C. D.; Nagy, Z. K. Cryst. Growth Des. 2012, 12, 1792−1807. (17) Sanzida, N.; Nagy, Z. K. Comput. Chem. Eng. 2013, 59, 111− 121. (18) Simone, E.; Saleemi, A. N.; Tonnon, N.; Nagy, Z. K. Cryst. Growth Des. 2014, 14, 1839−1850. (19) Simone, E.; Zhang, W.; Nagy, Z. K. Cryst. Growth Des. 2015, 15, 2908−2919. (20) Acevedo, D.; Nagy, Z. K. J. Cryst. Growth 2014, 394, 97−105. (21) Acevedo, D.; Tandy, Y.; Nagy, Z. K. Ind. Eng. Chem. Res. 2015, 54, 2156−2166. (22) Yang, Y.; Nagy, Z. K. Cryst. Growth Des. 2014, 14, 687−698. (23) Nagy, Z. K.; Chew, J. W.; Fujiwara, M.; Braatz, R. D. J. Process Control 2008, 18, 399−407. (24) Fujiwara, M.; Nagy, Z. K.; Chew, J. W.; Braatz, R. D. J. Process Control 2005, 15, 493−504. (25) Alvarez, A. J.; Singh, A.; Myerson, A. S. Cryst. Growth Des. 2011, 11, 4392−4400. (26) Hou, G.; Power, G.; Barrett, M.; Glennon, B.; Morris, G.; Zhao, Y. Cryst. Growth Des. 2014, 14, 1782−1793. (27) Quon, J. L.; Zhang, H.; Alvarez, A.; Evans, J.; Myerson, A. S.; Trout, B. L. Cryst. Growth Des. 2012, 12, 3036−3044. (28) Ferguson, S.; Morris, G.; Hao, H.; Barrett, M.; Glennon, B. Chem. Eng. Sci. 2013, 104, 44−54. (29) Wong, S. Y.; Tatusko, A. P.; Trout, B. L.; Myerson, A. S. Cryst. Growth Des. 2012, 12, 5701−5707. (30) Zhang, H.; Quon, J.; Alvarez, A. J.; Evans, J.; Myerson, A. S.; Trout, B. Org. Process Res. Dev. 2012, 16, 915−924.

5. CONCLUSIONS In this work, a FBRM-based feedback control approach, named ADNC, is implemented in single-stage and two-stage MSMPRC. On the basis of the in situ FBRM information, ADNC can automatically actuate heating or cooling to control the total particle chord counts within the ADNC set-point range during both startup and steady-state operations. It is found that the proposed continuous ADNC approach can significantly reduce the startup duration. In addition, the steady-state CSD can be controlled by adjusting the ADNC setpoint. Large crystals can be obtained if the ADNC set-point is properly selected to avoid nucleation in the second stage. Finally, ADNC is observed to provide very quick and effective disturbance suppression in continuous MSMPRC.

■

Article

AUTHOR INFORMATION

Corresponding Author

*Tel: +1-765-494-0734. Fax: +1-765-494-0805. E-mail: [email protected]. Web: https://engineering.purdue.edu/ ChE/People/ptProfile?id=79574. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We acknowledge Eli Lilly, Indiana, USA, for providing the MSMPRC transfer units and partial financial support. We are grateful to Christopher Burcham, Christopher Polster, and Daniel Jarmer for input on the rig configuration and discussions. I



Article

(31) Power, G.; Hou, G.; Kamaraju, V. K.; Morris, G.; Zhao, Y.; Glennon, B. Chem. Eng. Sci. 2015, 133, 125−139. (32) Yang, Y.; Nagy, Z. K. Chem. Eng. Sci. 2015, 127, 362−373. (33) Yang, Y.; Nagy, Z. K. Ind. Eng. Chem. Res. 2015, 54, 5673−5682. (34) Vetter, T.; Burcham, C. L.; Doherty, M. F. Chem. Eng. Sci. 2014, 106, 167−180. (35) Su, Q.; Nagy, Z. K.; Rielly, C. D. Chem. Eng. Process. 2015, 89, 41−53. (36) Lawton, S.; Steele, G.; Shering, P.; Zhao, L.; Laird, I.; Ni, X. W. Org. Process Res. Dev. 2009, 13, 1357−1363. (37) Yang, Y.; Nagy, Z. K. American Control Conference (ACC) 2015, 4282−4287, July 1−3, 2015.. (38) Koswara, A.; Nagy, Z. K. IFAC-PapersOnLine 2015, 48, 193− 198. (39) Peña, R.; Nagy, Z. K. Cryst. Growth Des. 2015, 15, 4225−4236. (40) Yang, Y.; Song, L.; Gao, T.; Nagy, Z. K. Cryst. Growth Des.. DOI: http://dx.doi.org/10.1021/acs.cgd.5b01290.201510.1021/ acs.cgd.5b01290 (41) Mitchell, N. A.; Frawley, P. J. J. Cryst. Growth 2010, 312, 2740− 2746. (42) Simon, L. L.; Merz, T.; Dubuis, S.; Lieb, A.; Hungerbuhler, K. Chem. Eng. Res. Des. 2012, 90, 1847−1855.

J


Automated Direct Nucleation Control in Continuous Mixed Suspension

Recommend Documents