Multi-Scale Characterization of Anti-solvent Crystallization

by

Des OGrady, B.E.

A Thesis presented to the National University of Ireland in fulfilment of the requirements for the degree of Doctor of Philosophy (Ph.D.)

Under the supervision of Dr. Brian Glennon

December 2007

School of Chemical and Bioprocess Engineering, College of Engineering, Mathematical and Physical Sciences, University College Dublin.

Page 1

ACKNOWLEDGEMENTS

First and foremost I would like to thank my parents Dorothy and Dermot. Their love and support over the years has been unstinting and in one way or another they have taught me everything I know. For that I will always be grateful.

Thanks to my wonderful sisters, Caoilfhionn, Fiona and Elva, for constantly endeavouring to figure out what it was that I actually did during my PhD. Theyre inquisitive nature has hopefully rubbed off on me over the years.

I would like to thank Dr. Brian Glennon for his talented and dedicated guidance. He has an uncanny ability to get to the heart of a problem and convey the solution in a clear and meaningful way. My PhD studies were made possible and have been a valuable and extremely enjoyable experience thanks to him.

I would like to thank my extended family of aunts, uncles and cousins. A special thanks to Joe and Teresa for taking such a keen interest in my studies over the years.

Countless friends have made my PhD studies all the more enjoyable. They offered a unique brand of humour that provided the perfect distraction from the lab. I have ninety-two reasons to thank Skellig and Marko along with Bud, Miller and Mac. Not to concentrate on that small group I have to also thank many others from college and school including Ed, Hutchy, Dots, Robbie, Ale, Karol, Marsha, Nikki, and Sarah.

Thanks to my fellow postgrads for making UCD a fun place to be. Special thanks to Ale, Brian and Paul who have been were there from the start. Thanks to an influx of new faces the last couple of years have been very enjoyable. Thanks to Eoin, Jessica, Kate, Aisling, Barry, and John for starting it off. Thanks to Mark for all his help and the CFD work in this thesis. Mairtin, good luck!

Thanks to all the lecturers in the School of Chemical and Bioprocess Engineering. A special thanks to Dermot Malone, Patricia Kieran, Eoin Casey and Frank MacLoughlin for all their help.

Thanks to everyone else in the school who helped me out along the way, including, Pat, Liam, Oliver, Jim, Brid, Sinead, Pat and Brian. A special thanks to Aoife.

Thanks to everyone at Mettler Toledo for giving me a great opportunity during my PhD. A special thanks to Paul Barrett for opening that door.

Finally, I would like to thank Anne for her brilliant friendship and unwavering loyalty and love. Heres to NYC Anneski!!!

TABLE OF CONTENTS
LIST OF PUBLICATIONS ABSTRACT CHAPTER 1: INTRODUCTION CHAPTER 2: LITERATURE REVIEW 2 1 2 8

2.1 ANTI-SOLVENT CRYSTALLIZATION 8 2.1.1 INTRODUCTION 8 2.1.2 VARIABLES TO CONSIDER IN ANTI-SOLVENT CRYSTALLIZATION 9 2.1.2.1 Choice of Anti-solvent 9 2.1.2.2 Anti-solvent Addition Rate 11 2.1.2.3 Agitation Intensity 13 2.1.2.4 Concentration Effects 14 2.1.2.5 Conclusion 17 2.2 FOCUSSED BEAM REFLECTANCE MEASUREMENT (FBRM) 19 2.2.1 INTRODUCTION 19 2.2.2 MODE OF OPERATION 20 2.2.3 VALIDATION OF FBRM TECHNIQUE INSTRUMENT PARAMETERS 25 2.2.3.1 Probe Position 26 2.2.3.2 Focal Point Position 27 2.2.3.3 Measurement Duration 29 2.2.4 VALIDATION OF FBRM TECHNIQUE - PROCESS PARAMETERS 30 2.2.4.1 Solids Concentration 30 2.2.4.2 Agitation Rate 32 2.2.4.3 Particle Material 33 2.2.4.4 Surrounding Medium 35 2.2.4.5 Temperature 36 2.2.5 CORRELATING FBRM MEASUREMENTS WITH OTHER PARTICLE SIZE ANALYSERS 37 2.2.6 THEORETICAL MODELLING OF FBRM DATA 40 2.2.6.1 Modelling Spherical Particles 41 2.2.6.2 Modelling Non-Spherical Particles 44 2.2.7 FBRM FOR CRYSTALLIZATION CHARACTERIZATION 47 2.2.7.1 Solubility Curve and Metastable Zone Width Determination 49 2.2.7.2 Nucleation Kinetics 50 2.2.7.2 Monitoring Crystal Size 51 2.2.7.3 Temperature Cycling 54 1.2.7.4 Polymorphic Transitions 55 2.2.7.5 Effect of Impurities 56 2.3 PROCESS VIDEO MICROSCOPE (PVM) 57 2.3.1 INTRODUCTION 57 2.3.2 CHARACTERIZATION OF PARTICULATE SYSTEMS USING PVM 61 2.4 ATTENUATED TOTAL REFLECTANCE FOURIER TRANSFORM INFRARED SPECTROSCOPY (ATR-FTIR) 62 2.4.1 INTRODUCTION 62 2.4.2 INITIAL WORK USING VARIOUS CALIBRATION MODELS 63 2.4.3 THE USE OF ATR-FTIR TO CHARACTERIZE CRYSTALLIZATION PROCESSES 67 2.4.3.1 Seeding 67

2.4.3.2 Oiling Out

68

CHAPTER 3: SOLUBILITY MEASUREMENT FOR AN ANTI-SOLVENT SYSTEM USING GRAVIMETRIC ANALSYSIS, ATR-FTIR AND FBRM 69
3.1 ABSTRACT 3.2 INTRODUCTION 3.3 EXPRESSION OF SOLUBILITY 3.4 EXPERIMENTAL WORK AND ANALYSIS 3.4.1 GRAVIMETRIC ANALYSIS 3.4.2 POLYTHERMAL METHOD USING FBRM 3.4.3 SOLUBILITY MEASUREMENT USING ATR-FTIR TECHNIQUE 3.4.3.1 Calibration of Probe 3.4.3.2 ATR-FTIR Solubility Measurement 3.4.4 OVERALL SOLUBILITY MEASUREMENT 3.4 DISCUSSION 69 69 71 74 74 77 81 81 87 90 93

CHAPTER 4: THE EFFECT OF MIXING ON THE METASTABLE ZONE WIDTH IN ANTI-SOLVENT CRYSTALLIZATION 95
4.1 ABSTRACT 4.2 INTRODUCTION 4.3. EXPERIMENTAL METHODS 4.4 RESULTS AND DISCUSSION 4.4.1 COMPARISON OF FBRM AND ATR-FTIR FOR THE DETECTION OF NUCLEATION 4.4.2 IMPACT OF PROCESS PARAMETERS ON THE MSZW 4.4.3 NUCLEATION KINETICS 4.5 DISCUSSION 95 96 98 100 100 104 106 111

CHAPTER 5: THE USE OF FBRM AND ATR-FTIR TO MONITOR ANTISOLVENT CRYSTALLIZATION AND ESTIMATE GROWTH RATE KINETICS 113
5.1 ABSTRACT 5.2 INTRODUCTION 5.3 MATERIALS AND METHODS 5.4 RESULTS AND DISCUSSION 5.4.1 FBRM RESULTS 5.4.2 ATR-FTIR RESULTS 5.4.3 GROWTH RATE KINETICS ESTIMATION 5.5 CONCLUSIONS 113 113 115 116 116 123 125 130

CHAPTER 6: SCALE-UP OF ANTI-SOLVENT CRYSTALLIZATION

6.1 ABSTRACT 6.2 INTRODUCTION 6.3 MATERIALS AND METHODS 6.3 RESULTS AND DISCUSSION 6.3.1 FBRM RESULTS 6.4 DISCUSSION 6.5 CONCLUSION

131
131 132 133 136 136 145 147

7. THESIS CONCLUSIONS 8. NOMENCLATURE 9: REFERENCES APPENDIX A

1. ABSTRACT 2. INTRODUCTION 3. COMPUTATION FLUID DYNAMICS MODEL 4. RESULTS 4.1 500ML SCALE 4.2 70 L SCALE 6. REFERENCES

149 153 155 171

171 172 173 174 174 177 179

LIST OF PUBLICATIONS

D. O'Grady & B. Glennon, Measurement of Nucleation Kinetics in an Anti-solvent System, Conway Institute Conference, Dublin, Ireland, February, 2005

P. Barrett, B. Smith, B OSullivan, J. Worlitschek, V. Bracken, & D. O'Grady, A Review of the Use of Process Analytical Technology for the Understanding and optimization of Batch Crystallization Processes, Organic Process Research and Development,9 (3), 348-355, 2005

D. O'Grady & B. Glennon, Growth and Nucleation Kinetics for an Anti-solvent System Using In Situ Instrumentation', Industrial Symposium on Industrial Crystallization, Dresden, Germany, September 2005

D. O'Grady & B. Glennon, 'Characterisation of Anti-solvent Addition Crystallization Using In-Line Tools', British Association for Crystal Growth Annual Conference, Sheffield, UK, September 2005

D. O'Grady & B. Glennon, 'Use of In-Situ Instrumentation to Characterise Anti-solvent Addition Crystallization', AIChE Annual Meeting, Cincinnati, USA, October, 2005

D. OGrady & B. Glennon, Comparing a Fast and Slow Crystallization UCD 150 Engineering Exhibition, Dublin, Ireland, January, 2006

M. Barrett, D. O'Grady, B. Glennon, 'Characterisation of Anti-solvent Addition Crystallization Using In Situ Tools and Computational Fluid Dynamics', 2006 International Real Time Analytics Users' Conference, Barcelona, Spain, February 2006

D. O'Grady, M. Barrett, E. Casey & B. Glennon, 'PAT and the Crystallization Toolkit', Pharmaceutical Manufacturing, 5 (6), 44-47, 2006

D. O' Grady, M. Barrett & B. Glennon, 'Scale-up of Anti-Solvent Crystallization Using in Situ Tools and Computational Fluid Dynamics', AIChE Annual Meeting, San Francisco, USA, November, 2006

M. Barrett, D. O'Grady, B. Glennon & E. Casey, 'The Application of CFD to the Multi-Scale Characterization of Anti-Solvent Addition Crystallization', AIChE Annual Meeting, San Francisco, USA, November, 2006

D. OGrady & B. Glennon, Fundamentals of Anti-solvent Crystallization, Mettler Toledo AutoChem Seminar Series New Jersey, USA, July 2006

D. O'Grady, M. Barrett, E. Casey & B. Glennon, 'The Effect of Mixing on the Metastable Zone Width and Nucleation Kinetics In the Anti-solvent Crystallization of Benzoic Acid, Chemical Engineering Research and Design, Transactions IChemE part A, 85 (7) 945-953, 2007

D. O'Grady & B. Glennon, Solubility Measurement for an Anti-solvent Crystallization System Using Gravimetric Analysis, ATR-FTIR and FBRM', Crystal Growth and Design, in press

ABSTRACT

The anti-solvent crystallization of benzoic acid from ethanol-water solutions using water as the anti-solvent, at scales ranging from 500 mL to 70 L, is presented. A thorough review of the literature focusing on anti-solvent crystallization and the use of in situ tools for crystallization characterization was undertaken prior to experimental studies. A novel method for measuring solubility in organic systems, based on the FBRM technique, is presented. This method has been compared to other solubility measurement techniques and found to be reliable over a wide range of concentrations. The observed trends were found to be consistent with the UNIQUAC liquid activity coefficient model. Using the FBRM technique, novel methods for the determination of nucleation and growth kinetics were developed. By measuring the metastable zone width under a range of process conditions, including addition rate, agitation intensity and feed location, a kinetic expression for nucleation in organic anti-solvent systems that incorporates the influence of agitation, was developed and is presented for the first time. A kinetic expression for growth was elicited from growth rate data, gathered using the FBRM technique, and liquid phase concentration data, gathered using the ATR-FTIR technique. This novel method serves to highlight the possibility of using in situ tools to measure growth rate kinetics in process and in real time. The extensive characterisation work performed at the laboratory scale provided the basis for an investigation of system performance upon scale-up to a geometrically dissimilar 70 L pilot plant crystallizer. Suitable operating parameters were chosen based on the small scale characterization and the scale-up was deemed successful as product of a similar size and yield was produced at the 70 L scale using the same cycle time. 1

CHAPTER 1: INTRODUCTION

In recent years the pharmaceutical industry has shown a renewed interest in the field of crystallization. Crystallization is seen within the industry as a key process bottleneck during API drug development and manufacture. Through effective characterization at the laboratory scale, development times can be reduced, effective and efficient scale-up can be ensured and cycle times can be improved. At the manufacturing scale, improved crystallization monitoring and optimization can reduce the number of failed batches, ensure regulatory compliance and enhance downstream processing operations such as filtration, drying and formulations. Such work ultimately leads to reduced costs, improved margins and a shorter time to market for new drugs.

A key element of the process development lifecycle is the scale-up of the crystallization operation to large-scale, whilst maintaining product and process compatibility. A major challenge to address is the impact of scale-related effects on the process. In this work, the characterization of a crystallization process on a variety of scales is presented. An anti-solvent crystallization process is chosen for study as it is a technique that has been relatively neglected in the literature, in favour of cooling, despite its prevalence in API drug development and manufacture. Effective antisolvent crystallization scale-up also presents specific mixing-related problems worthy of investigation.

In situ tools are employed for the measurement of key crystallization parameters such as crystallization kinetics and supersaturation profile. Focused Beam Reflectance 2

Measurement (FBRM) is used to study the degree and rate of change of the size, and number of crystals in suspension and Attenuated Total Reflectance Fourier Transform Infrared Spectroscopy (ATR-FTIR) is used to measure solution concentration. Process Video Microscope (PVM) is used to image crystals as they exist in process. The use of in situ tools has gained much attention within industrial and academic crystallization groups for their ease of use, ability to minimize or remove sampling and real time response to changes in the crystallization system.

Chapter 2 of this thesis is a review of the literature in the field of anti-solvent crystallization and use of in situ tools for crystallization characterization. Attention is paid to areas of the field that have been neglected, for example anti-solvent addition location and the use of organic crystallization systems. A complete review of the FBRM technique is presented, with the aim being to outline its evolution from novel technology for particulate systems characterization to the industrial standard for in situ crystallization characterization.

Chapter 3 will focus on the solubility curve for anti-solvent crystallization systems. A novel method for the determination of solubility in anti-solvent systems was developed using the FBRM technique. To assess the reliability of this method two other techniques, gravimetric analysis and ATR-FTIR, were employed to measure the solubility. The results for each method were consistent over a full range of concentrations. The solubility data measured using the FBRM technique also proved consistent with the UNIQUAC liquid activity coefficient model. Solubility data for anti-solvent systems is typically presented in terms of grams of solute per gram of solution. This can make interpretation of key crystallization parameters such as the

metastable zone width and supersaturation difficult. In this case the solubility is expressed on an anti-solvent free basis (grams of solute per gram of solvent) making subsequent characterization of the system simple and effective.

A method for the estimation of nucleation kinetics in antisolvent systems is presented in Chapter 4. An extensive study aimed at characterizing the impact of mixing conditions on the metastable zone width was undertaken and focused on the impact of addition rate, agitation intensity and feed location. As expected, a widening of the metastable zone width at higher addition rates was observed. The feed location proved to be an extremely important parameter with extreme variation in the metastable zone width observed for two different feed point locations. The impact of agitation intensity on the metastable zone width proved to be heavily dependant on the feed location chosen. By modifying traditional nucleation kinetics equations for the antisolvent system, a kinetic expression for nucleation in organic anti-solvent systems that accounts for the impact of agitation was developed and is presented here for the first time. The formulation of such an expression was not possible under certain condition, specifically a feed location close to the wall of the vessel. This was due to a nonrepeatable point of nucleation and in some cases nucleation prior to the attainment of bulk solubility. In order to better understand the mechanism by which this occurred computational fluid dynamics (CFD) was employed to model mixing conditions in the vessel. Poor incorporation of the anti-solvent into the bulk solution, when it was added close to the wall, proved the important factor and the results of this investigation can be found in Appendix A.

A quick, efficient and novel method for estimating growth kinetics for organic crystallization systems is presented in chapter 4. Growth rate kinetics can be estimated by finding the relationship between the crystal growth rate and its driving force, supersaturation. ATR-FTIR can be used to monitor supersaturation and FBRM can be used to measure a relative growth rate, hence the combination of these in situ tools can, in theory, be used to generate real time, in situ, growth rate information.

For this study a specific set of process conditions is chosen to ensure a low nucleation rate and repeatable crystallization with consistent yield. The choice of these conditions is made based on the findings of chapter 3. The experiment is conducted in triplicate and the results averaged to elicit the desired growth rate information. The results indicate the ease with which reliable and representative growth rate information may be estimated, in situ and in real time.

The implementation of such a technique for crystallization control may prove extremely useful. Much of the research published to date focuses on controlling the prevailing level of supersaturation in the crystallizer to achieve the desired product and process performance. While such work is valuable, in order to truly control the crystallization process the crystal nucleation and growth rates must also be known. With this information better control can be achieved that allows specific particle size distributions to be targeted in a specific batch time. The ability to measure the crystal growth rates in situ may prove extremely useful for tacking this difficult but potentially rewarding problem.

The scale-up of a successful crystallization process from the laboratory to production is extremely challenging and is tackled in Chapter 6. Increasing the crystallizer size impacts the mixing regime, heat and mass transfer properties as well as the surface area to volume ratio and crystal suspension profile. A combination of these factors can change the physical and chemical characteristics of the system that may have been established at smaller scales, resulting in a very different process and product. In the pharmaceutical industry there are added pressures associated with large scale crystallization operations due to its time consuming nature and expense, in terms of operational and product costs. In situ tools can play a vital role in addressing these many and varied challenges. The direct comparison of continuous and representative data between scales can be used to identify the source of variation and suggest a suitable course of action.

For this study scale up from the 500mL laboratory scale to a 70L pilot scale is performed. The ultimate goal of the scaled up process is to produce a similar yield, in a similar time and for the crystals to be of a similar size. The extensive

characterization work, outlined in chapters 3-5, is used to determine the set of process conditions that were most likely to achieve this goal with computational fluid dynamics employed to identify a suitable feed location. The scale-up ultimately proved successful with the yield, batch time and crystal size being within acceptable limits.

Despite this, scale-up success the in situ FBRM data raises concerns over the mechanism by which the results were achieved. The FBRM data clearly indicates that while the growth and nucleation rates at both scales were similar, the nucleation point

is different and the pilot scale batch nucleates prior to the attainment of bulk saturation. The FBRM data also indicates that there is significant segregation in the vessel.

CHAPTER 2: LITERATURE REVIEW

2.1 ANTI-SOLVENT CRYSTALLIZATION

2.1.1 Introduction

That a solute can have different solubilities in different solvents has long being recognised. Anti-solvent crystallization exploits this fact to crystallize solid product from solution. Solution supersaturation is generated by the addition of a second solvent (anti-solvent) that reduces the solubility of the solute and induces crystallization. The solvent with the lower solubility is often called an anti-solvent. The process is often referred to as drowning-out, salting-out, solventing-out or extractive crystallization.

The industrial use of an anti-solvent for the separation and purification of solid product was first demonstrated by Gee et al., (1947). Iron impurities were separated from the desired aluminium sulphate in a 2-ton-per-day continuous pilot plant using ethyl alcohol as an anti-solvent. This method had apparently first been suggested by Vittorf (1924). The process relied on the lower solubility of the aluminium sulphate compared to the iron impurities in alcohol solutions but ultimately proved uneconomical due to a low cost differential between the crude and finished product.

In the pharmaceutical industry, where the final solid product is extremely valuable and the product thermal stability is often low, anti-solvent crystallization provides an important alternative to cooling and evaporative crystallization. 8

As in a cooling or evaporative crystallization there are a number of issues to consider when designing an anti-solvent crystallization. The size, number and shape of the particles produced are important, as these factors can impact on downstream processing operations such as filtration, drying and formulation. The yield of a batch is important as the solute in the pharmaceutical industry is extremely valuable and it is desirable to keep losses at a minimum. It is very important to ensure the correct polymorphic form is crystallized. In an anti-solvent system supersaturation is generally high which can lead to a non-stable polymorph. Additionally, because a second solvent is being added to the crystallization system solvate and hydrate issues are common. Much research has been carried out to investigate how various process variables can affect these crystallization characteristics.

2.1.2 Variables to Consider in Anti-solvent Crystallization

2.1.2.1 Choice of Anti-solvent

The first step in the design of an anti-solvent crystallization is the identification of a suitable anti-solvent. Important characteristics include low solute solubility, high miscibility in the primary solvent, ease of recovery and potential environmental impact. Apart from the obvious operational constraints associated with choosing an anti-solvent, outlined above, there are a number of important crystallization parameters that can be affected.

The choice of anti-solvent can affect particle size and population, through the different mixing conditions associated with the addition of anti-solvents of different density and viscosity. Takiyama et al., (1998) found that the number of crystals produced by mixing a saturated aqueous solution and a saturated ethanol solution was greater, by two orders of magnitude, than that produced by mixing two saturated ethanol aqueous solutions having different concentrations. The reason proposed was that pure water and pure ethanol mixtures took far longer to mix fully leading to areas of high supersaturation and elevated nucleation rates. Barata and Serrano, (1998b) showed that the choice of anti-solvent had some impact on crystal size, for the crystallization of potassium dihydrogen phosphate (KDP) using three different aqueous alcohol solutions, but did not propose a mechanism for this effect. However, further work by this group, using the same system, showed that nucleation kinetics (Barata & Serrano, 1996a) and growth kinetics (Barata & Serrano, 1998a) were affected by the choice of anti-solvent. It appeared the nucleation kinetics were influenced by the dielectric constant of the solution and the size of the alcohol molecules in the anti-solvent. The growth kinetics were influenced by the size of the alcohol molecules, in that, large molecules increase the physical barrier to solute deposition. However it was noted that larger alcohol molecules could be less adsorbed on the crystal surface, thus inducing higher growth rates.

Finally, product yield can be improved by choosing an anti-solvent in which the solute is most insoluble. This has been shown by Oosterhof et al., (1999), for the crystallization of sodium carbonate from aqueous solutions. Diethylene glycol was deemed the most suitable anti-solvent as the as it provided the highest recovery of sodium carbonate from solution.

10

Another important factor to consider when choosing a suitable anti-solvent is phase separation. Due to the high level of supersaturation typically encountered during antisolvent addition the precipitation of a second liquid phase prior to nucleation of the solute is not uncommon (Barata & Serrano, 1996a; Oosterhof et al., 1999). Care should be taken to ensure phase separation is not an issue when selecting a suitable anti-solvent.

2.1.2.2 Anti-solvent Addition Rate

The anti-solvent addition rate is an important process variable that must be controlled in order to produce a suitable crystal product. High addition rates typically result in a wider metastable zone and consequently high supersaturation generation rates. This has been shown and characterized by Pina et al., (2001) for the crystallization of single and double sulphates from aqueous solutions using methanol as an anti-solvent. Similarly, high supersaturation levels can be encountered in regions close to the addition location, especially at elevated addition rates (Takiyama et al., 1998). Thus, high addition rates often lead to elevated nucleation rates and the formation of fine crystals. This has been shown by Holmbck & Rasmuson (1999) for the benzoic acidethanol-water system and Charmoloue and Rousseau (1991) for the crystallization of L-serine using methanol as an anti-solvent. In both of these cases the crystal morphology was also influenced by the addition rate. In some cases, very high addition rates may lead to an increase in the final product size due to elevated agglomeration rates (Granberg et al., 1999; Jones et al., 1987)

11

Budz et al., (1986) experimentally determined the most suitable anti-solvent addition rate, for the crystallization of cocarboxylase hydrochloride from its aqueous solution by the addition of acetone, by correlating the final product size with addition rate. The same addition rate produced maxima in the average size and bulk density of the crystals produced via an unseeded process. A similar result was observed for the seeded process, but the flow rate that produced the result was double that for the unseeded process. This indicates the beneficial impact seeding may have, by maintaining low levels of supersaturation even at high addition rates.

The impact of the addition rate on crystallization is not limited to the crystal size and shape. High supersaturation levels, resulting from elevated addition rates, can lead to the formation of undesirable solvates, hydrates and solvates. Kitamura & Sugimoto, (2003) investigated the crystallization of polymorphs of thiazole-derivative (BPT) by the addition of water to methanol solutions. When a very low addition rate was used, nucleation of the solvated methanol crystal form (D) occurred preferentially. At high water addition rates the proportion of the hydrated crystal form (BH) increased. Beckmann (1999) studied the effect of addition rate on the crystallization of Abecarnil from isopropyl acetate by the addition of hexane. At slow anti-solvent addition rates the most stable polymorphic form, form C, was generally crystallized but at very high addition rates the unstable B form was observed.

In general, high addition rates have a negative impact on the properties of the final crystal product, either through poor crystal size distribution, incorrect form or both. However, at the production scale a balance must be struck between implementing a

12

slow enough addition rate to ensure a suitable crystal product and ensuring the batch time is not excessive.

2.1.2.3 Agitation Intensity

The agitation intensity is an extremely important variable in any crystallization. It has increased importance for anti-solvent crystallization where two liquid phases must be mixed to create supersaturation. The influence of agitation intensity on crystallization processes is often conflicting. Charmoloue & Rousseau (1991) showed that increasing agitation intensity reduced agglomeration levels and improved the impurity profile, but this in turn led to elevated breakage rates. Similarly, Budz et al., (1986) showed that a high agitation intensity prevented scaling on the walls of the crystallizer but this was at the expense of excessive secondary nucleation.

Crystallization kinetics are also influenced in conflicting ways by the agitation intensity. Barata & Serrano (1998a) showed that increasing agitation intensity improved growth rates because more crystals were formed initially, and the surface area available for growth increased. However, the growth rate was limited at very high agitation intensities. This may have been due to crystal breakage but was not proved. It has also been reported that agitation intensity has no impact on growth kinetics. Jones & Mylardz (1989) studied the crystallization of potassium sulphate from aqueous and aqueous acetone solutions using acetone and acetone-water mixtures as an anti-solvent in an MSMPR. Crystal growth rates were largely independent of agitation intensity. This may indicate that the crystallization was

13

surface diffusion controlled rather than by the bulk transport of the solute molecules to the crystal.

Agitation intensity has also been shown to affect nucleation kinetics. Barata & Serrano, (1996b) studying the KDP-alcohol-water system, showed that, between 50 and 400 rpm, induction periods did not depend on the agitation intensity but above 400 rpm the induction periods became smaller. Takiyama et al. (1998) studied the number of sodium chloride crystals formed by mixing saturated aqueous ethanol solutions, under different agitation conditions. The number of crystals formed exhibited a dependence on the agitation condition leading to the conclusion that nucleation rates were strongly influenced by local mixing.

Further examples of the conflicting influences of agitation intensity, and mixing in general, have been presented by Paul et al., (2005). For example, increasing the agitation can improve heat and mass transfer, minimise settling and improve the purity profile. However the same increase in agitation may cause attrition, secondary nucleation and entrain gas from the headspace.

2.1.2.4 Concentration Effects

The concentration of the solution to be crystallized is another important variable that must be considered in an anti-solvent crystallization. Typically the solution concentration is chosen to maximise the yield per batch; however, there may be advantages to crystallizing from a solution of different concentration. Crystallizing from a dilute solution can avoid regions of high supersaturation and improve product

14

quality, as shown by Budz et al., (1986), for the crystallization of cocarboxylase hydrochloride by the addition of acetone to dilute and concentrated aqueous solutions. Crystallizing from concentrated solutions typically results in higher supersaturation and a smaller final crystal product, due to higher nucleation rates. This has been shown by a number of researchers including, Holmbck & Rasmuson (1999), Beckmann (1999) and Kaneko et al., (2002). Kaneko et al., (2002) studied the crystallization of sodium chloride from aqueous ethanol solutions using ethanol as the anti-solvent. At high ethanol solution concentrations the supersaturation was reduced and monodispersed unagglomerated crystals were formed. As the concentration of solute in the starting solution increased the number of crystals formed increased.

The influence of solution concentration may also be independent of supersaturation. Jones & Mylardz (1989), studying the crystallization of potassium sulphate from aqueous and aqueous acetone solutions in an MSMPR, showed that the growth rate was strongly affected by the acetone concentration in the solution. At high acetone concentration the growth rate was significantly reduced for the same level of supersaturation. Similarly, Granberg et al., (2001) showed that for similar thermodynamic driving forces the induction time, for the nucleation of paracetamol from acetone-water mixtures using water as the anti-solvent, increased with increasing water content in the solution.

The solution concentration may also impact on the polymorphic form of the crystal product and the rate of transformation from a less stable form to a more stable form. Kitamura & Sugimoto (2003) studied the crystallization and subsequent transformation of polymorphs of thiazole-derivative (BPT) by the addition of water to

15

methanol solutions of the solute. At low initial concentrations the hydrated crystal form (BH) crystallized independent of the addition rate. At higher initial concentrations the solvated methanol form crystal (D) nucleated. This may have been due to the hindrance of the hydration of the BPT molecule. At low initial concentration, transformation of the BH form to the other polymorphic form (A) was observed and the transformation rate appeared to decrease with initial concentration. At higher initial concentrations both the BH and D forms transformed to the A form. At yet higher initial concentrations the D form transformed to the BH form.

Anti-solvent addition can sometimes result in the production of a large number of very fine crystals. This is typically due to elevated nucleation rates, encountered at the point of primary contact between solution and anti-solvent. In cases where the formation of these fine crystals is undesirable, dilution of the anti-solvent with the system solvent can be beneficial. The usefulness of this technique has been shown for the crystallization of potassium sulphate from aqueous solution using aqueous acetone as an anti-solvent (Mullin et al., 1989) and for the crystallization of benzoic acid from ethanol solutions using aqueous ethanol as an anti-solvent (Holmbck & Rasmuson, 1999). In the latter case the anti-solvent feed concentration also influenced the crystal morphology. Diluting the anti-solvent with the system solvent has shown its use in other crystallization systems, at two different scales (Jones et al., 1987), and in a mixed-suspension, mixed product removal (MSMPR) crystallizer (Jones and Mylardz, 1989). Barata and Serrano (1998b) also showed that anti-solvent concentration influences the yield, due to a dilution effect, and the growth rate, due to the formation of smoother crystals and a reduced mass deposition rate.

16

2.1.2.5 Conclusion

It is clear that a wealth of research has been conducted on the factors affecting antisolvent crystallization. However there are some areas that warrant further investigation especially in respect of their application to pharmaceutical crystallization. Nucleation kinetics for a large number of anti-solvent systems have been calculated using induction time experiments. However, pharmaceutical

crystallizations are typically performed by continuously generating supersaturation until nucleation occurs, rather than generating a given level of supersaturation and then waiting for nucleation to occur. For this reason there is value in measuring the metastable zone width under various process conditions and applying this data to calculate nucleation kinetics.

The effect of process conditions, on various crystallization parameters have been investigated and outlined above. Much of this research has surmised that supersaturation is a key variable in defining the final crystal product. However, little research has been done to actually measure the supersaturation and provide an understanding of the underlying mechanisms involved in the crystallization. The ATR-FTIR probe provides the opportunity to monitor supersaturation in situ.

One important process variable that has not been characterized adequately is the feed point location. Since mixing is a vital aspect of the anti-solvent system the position of the anti-solvent feed warrants investigation.

17

Much of the work characterising the particle size has been performed using off line techniques such as microscopy. While useful, these methods rely on sampling which can be time consuming and unrepresentative. With in line techniques such as FBRM and PVM now available it is possible to characterize anti-solvent crystallization without the need to sample.

Finally much of the initial work on anti-solvent crystallization used inorganic compounds as model systems. Pharmaceutical compounds are in the main organic compounds and it is best to characterize the anti-solvent system using organic compounds.

Thus the scope for this study is to characterize anti-solvent addition crystallization using an organic system with a view to applying the data gathered to pharmaceutical crystallizations. The solubility of the model system will be measured using a number of techniques. Metastable zone width data will be measured at various process conditions including varying feed location. Nucleation kinetics will be calculated from this data. The crystallization will be examined using in-line tools. FBRM and PVM will be used to monitor the particle size and shape and ATR-FTIR will monitor supersaturation. Attempts will be made to scale up the process.

18

2.2 FOCUSSED BEAM REFLECTANCE MEASUREMENT (FBRM)

2.2.1 Introduction

In recent years, Focused Beam Reflectance Measurement (FBRM) has emerged as a powerful technique for the characterization of particulate systems. FBRM is a probe based instrument that measures a function of the size, shape and population of a particulate system in-process and in real time. It can measure the degree of change to these properties as well as the rate of change making it a useful tool for the study of dynamic particulate systems. Its advantage over other off-line techniques is that it measures the particles in situ, eliminating the need to sample and ensuring that the particles are measured under process conditions. Off line techniques rely on a potentially unrepresentative sampling procedure and for some it is necessary to prepare the sample before analysis through for example sonication or dilution. Even in the case of off-line microscopy, preparing a slide may alter the particles significantly through breakage. In the case of crystallization, nucleation, growth or dissolution may occur after sampling thus providing unrepresentative data.

There is a growing body of literature outlining the use of FBRM for the characterization of a diverse array of particulate applications. These include granulation (Sistare et al., (2005)), polymerisation (Hukkanen, (2003)), flocculation (Blanco et al., (2002)), biofilm (Choi (2003)), gas hydrate systems (Clarke & Bishnoi, (2004)), wastewater treatment (de Clercq (2004)) plant cell suspensions (Jeffers et al., (2003)), and filamentous organism fermentations (Pearson, (2003)). It is in the field of crystallization however, that FBRM has received most attention.

19

2.2.2 Mode of Operation

FBRM operates by shining a monochromatic laser beam, of wavelength 790 nm, generated by a class 1 laser source, via a fibre optic conduit, to an optical assembly housed within a probe shaft. This optical assembly (Figure 1.1) consists of a lens mounted eccentrically, and this entire assembly rotates in a circular motion at high speeds. The circular motion is generated mechanically in the models used for this work (M400L and S400A) however for use in explosive environments the assembly can be rotated pneumatically. The user may change the speed of rotation (scan speed) in the software to values of 2 m s-1, 4 m s-1, 6 m s-1 and 8 m s-1. For the duration of this work the scan speed was held constant at 2 ms-1 at which the measurement range is between 0.5 and 1000 m. At higher scan speeds the measurement range can be extended at the upper level but in doing so the sensitivity at the lower end of the range is compromised. As the monochromatic beam passes through the lens it is focused to a fine point (d ~ 4m) just inside the probe window. The circular motion of the assembly traces the fine point of the laser over the circumference of a circle (Figure 1.1).

When the beam comes in contact with a particle, light is reflected in all directions. Some of the light is reflected back up the probe to a detector. The detector measures the time duration for which the light is reflected (Figure 1.2). This time duration can be used to measure a chord length across the particle according to equation 2.1.

l=tv

Eq. 2.1

20

Where l is the chord length (m), t is the time duration of the backscatter (s) and v is the speed of rotation of the laser beam (m s-1). For a chord to be accepted the intensity of backscattered light must exceed a certain threshold. Additionally the rise time of the electronic pulse generated by the backscattered light must be short. The theory is that only particles that pass directly through the focal point will have a short rise time. Thus particles that are too far from the focal point and are out of focus will not be considered (Sparks & Dobbs 1993).

Since the laser rotates at high speed and a chord length is an entirely random measurement there is the possibility to measure thousands of chord lengths per second.

21

Figure 2.2: Chord Length Measurement (Mettler Toledo Users Site)

These individual measurements can be combined to produce a chord length distribution (CLD) that provides a robust fingerprint of the particulate system. This distribution is sensitive to changes in the size, number and shape of particles in the system. Importantly it is sensitive not only to the degree of change but also the rate of change making the FBRM technique extremely powerful for studying particulate systems.

The measurement duration is the length of time for which the instrument scans before the data are logged and can be altered using the software, to between 1 s and 1 hr. A long measurement duration ensures a large number of counts and therefore statistically robust data. However, if the measurement duration is too long, key process changes that occur quicker than the measurement duration will be missed. Mettler Toledo recommends that a measurement duration of 10 s is used in the 22

laboratory and 1 min is used in the plant. This gives a good balance between sensitivity to process changes and statistically robust data.

A typical FBRM installation consists of a probe, which is connected via a fibre optic cable to an electronics box known as the field unit. The field unit is connected to a PC for instrument operation, data acquisition and data review (Figure 1.3). A more detailed description of how FBRM operates can be found in the literature (Sparks & Dobbs, 1993; Tadayyon &Rohani, 1998; Barrett & Glennon, 2002).

An important attribute of the FBRM software is its ability to trend important statistics of the chord length distribution over time (Figure 1.4). For example the number of particles in a small size range, e.g. 1 10 m, can be trended over time to assess the change in the fine end of the distribution. Additionally the same can be done for a large size range, e.g. 100 250 m. In fact it is possible to focus on any size range or statistic (i.e. mean, median, or mode) of the distribution and trend it over time to improve understanding of the process at hand. It is also possible to weight the statistics to provide the most relevant information. This is especially useful for monitoring a dynamic process such as attrition where there is a decrease in the size of the coarse particles and an increase in the number of fine particles (Kougoulos 2005c). Suitable statistics for measuring such a process, apart from counts in individual size ranges, may be the median for fine particles and a square weighted mean for the coarse particles. By trending suitable statistics over time it is possible to not only confirm that agglomeration is taking place but also assess the rate at which it is occurring and when it has stopped.

23

Figure 2.3: Typical FBRM Setup with probe, field unit and computer (Mettler Toledo Users Site)

Figure 2.4: Chord Length Distribution Trended over time (Mettler Toledo Users Site)

24

The chord length distribution itself can be manipulated to analyse key particle parameters. A variety of weightings may be applied to the chord length distribution in order to highlight the fine and coarse end of the distribution. A 1/length weighting is applied by dividing the number of counts in each size range by the midpoint of that size range. This emphasises small particles. A square weighting is applied by multiplying the number of counts in each size range by the square of the midpoint. In this way the large particles are emphasised. Heath et al., (2002) provides an excellent overview of the difference in these distributions and how they may be used and Fujiwara et al., (2002) compares how different weightings can be used to identify the onset of nucleation in the aqueous batch cooling crystallization of paracetamol.

2.2.3 Validation of FBRM Technique Instrument Parameters

Much work has been carried out in order to validate the FBRM technique and assess how process and instrumental parameters affect the chord length measurement. Instrumental parameters that have been investigated include probe position, focal point position, scan speed and measurement duration. Process parameters investigated include solids concentration, agitation intensity, surrounding medium and particle material.

It should be noted that the FBRM technique existed under a different name in the 1990s. The Par Tec 100 and Par Tec 200 were precursors to the current FBRM probe. The Par Tec probes operated on the same principle of backscattered light as the current FBRM probe. It was possible to display the raw chord data, a spherical equivalent diameter or a volume distribution. A lot of early validation work was

25

performed on the Par Tec probe that still applies to the current FBRM models. For the following sections all examples are for the current FBRM probe unless otherwise stated.

2.2.3.1 Probe Position

Probe positioning plays a vital role in ensuring that a representative sample is presented to the window. Poor probe position can lead to unrepresentative data and poor process understanding. It is imperative to position the probe so the particles are flowing into the window. Tadayyon & Rohani (1998) identified this need and also suggested the probe window should be sufficiently far away from any light reflective object such as the stirrer blades to avoid the generation of signal noise, especially at low concentrations. Barrett and Glennon (1999) assessed the effect of probe position, on the CLD of alkaline frit in water, at two different scales (1.5 L and 70 L stirred tanks). The most statistically robust data were achieved when the count data were highest. This was achieved, at both scales, with the probe mounted from the side at an angle to the dominant flow direction. The manufacturer provides advice on positioning the probe in a reactor and a pipeline. In both cases the probe must be placed at an angle to the dominant flow direction (Figure 1.5).

26

Figure 2.5: Suitable probe position (circled) Mettler Toledo Lasentec Users Site

2.2.3.2 Focal Point Position

The choice of focal point is extremely important to ensure representative data is gathered. If the focal point is set far from the probe window there will be a reduced sensitivity to fine particles. This is due to the deterioration of the light signal as it leaves and re-enters the probe. This fact has been identified by a number of researchers including Worlitschek and Mazzotti (2003), for the study of a single particle in water and in a dispersion of fine particles and Heath et al. (2002), for the study of the CLD of aqueous calcite suspensions. Heath et al., (2002) also showed that setting the focal point far from the window increased sensitivity to coarse particles. The reason given for this was that large particles are less likely to enter the 27

measurement zone when the focal point is set close the window. Monnier et al., (1996) studied the effect of the focal point position on the measured mean size of latex and orgasol particles in various solvents using the Par Tec device. The mean size varied depending on the focal point chosen. For small particles the most accurate size data was achieved when the focal point was set close to window. For large particles the opposite was true with the optimal focal point being between 1.5 and 2 mm from the window. Ruf et al., (2000) monitored how the distance of a particle from the focal point affected the measured chord length of a single particle. They found that as the particle moved farther and farther from the window the size of the measured chord length decreased. This was due to the weakening and broadening on the laser beam that makes it harder for the FBRM signal-processing unit to detect precisely the boundary of the particle. For the same reason, a number of chord lengths much smaller that the particle itself were measured. Worlitschek and Mazzotti (2003) found similar results for the single particle systems and their findings suggested that when the whole particle size distribution is of interest, the best focal point to detect large and small particles was at the probe window. Furthermore, they suggest that if a specific known size range of particles is of interest then the focal point should be positioned half the diameter of the smallest particle away from the window. This eliminates the risk of not detecting the smallest particles.

In each of these studies the optimum focal point position for fine material was determined to be close to the window. Since fine crystals typically are of most interest in crystallization applications the focal point is usually set at the window. While some sensitivity to larger crystals is lost, this set-up ensures maximum sensitivity to the more important fine crystals. Some of the newer laboratory-based FBRM systems

28

(S400 range) have a fixed focal point at the window. In the case where larger particles are of interest, for example in granulation, a probe with a variable focal point may be preferable.

2.2.3.3 Measurement Duration

The choice of measurement duration is a balance between ensuring the collection of statistically robust data and ensuring key process events are not missed. Additionally, applying a short measurement duration results in a large amount of data that can be difficult to handle. Pearson (2004) assessed the effect of measurement duration on the CLD of filamentous bacteria measurement measured offline. Values between 1 and 60 s were investigated and a measurement duration of 30 s was chosen as it showed only minor variation in all the statistics of the CLD examined and allowed relatively rapid collection of data. Monnier et al., (1996) investigated the effect of measurement duration on the size distribution measured using the Par Tec device for glass beads between 125 and 250 m (measured by sieving) in water. The cycle time was varied between 0.2 s and 3.2 s for the same suspension of glass beads. For a measurement duration of 0.2 s the total count data was too small to allow correct statistical processing of the data. When the measurement duration was changed to 3.2 s the reproducibility of the data increased significantly and the standard deviation halved indicating a more robust measurement.

29

2.2.4 Validation of FBRM Technique - Process Parameters

2.2.4.1 Solids Concentration

An increase in solids concentration leads to an increase in total counts measured by the FBRM. This is simply due to the presence of more particles in the measurement zone at higher concentrations. A linear relationship between solids concentration and counts has been observed by (Pearson et al., 2004), for a filamentous bacterial system, and, by Monnier (1996), for aqueous suspensions of glass beads. Other studies have noted a similar linear relationship but at very high concentration count data has been observed to stop increasing with increasing solids concentration. Barrett and Glennon (1999) investigated the relationship between the CLD and solids concentration aqueous alkaline fit suspensions at scales ranging from 500 mL to 70 L. At all scales the relationship between counts and concentration was linear at low concentration but tapered off at high concentrations. Similar results were found by Heath et al., (2002) investigating the change in total counts of aluminium and calcite particles in water due to an increase in solids concentration.

The reason for this non-linearity is due to an increase in the instrument dead time at high concentrations. At low concentration only a small fraction of the distance traversed by the laser is used to measure a particle. However at high concentration a significant percentage of the distance is used reducing the time available to detect other particles. This dead time is slightly larger than the time spent traversing the particle because the reflected light must return to zero for a short period between particles. Additionally, the dead time is increased at high concentrations because

30

particles outside the viewing zone are measured and then rejected if the rate at which the signal increases is not fast enough. Furthermore, overlapping particles may be counted as one large particle exacerbating the problem. It is possible to calculate this dead time but it is system specific. Heath et al., (2002) measured the value for aluminium and calcite particles.

Another possible reason for the non-linear relationship between counts and concentration is that at high concentration, the intensity of light at the focal point can be reduced by blocking of the laser path by other particles. Therefore, the intensity of backscattered light may not be sufficient to generate pulses with sufficient amplitude to be detected by the instrument (Tadayyon & Rohani 1998).

The impact of solids concentration on the size of the particles measured has also being investigated by a number of researchers. In some cases a strong influence has been observed but in others the impact has been negligible. Dowding et al., (2001) showed that increase in solids concentration, of poly(vinyl chloride) beads in water, resulted in a decrease in length weight mean chord length and an increase in the cube weighted mean chord length. It was proposed that at high solids concentration the mean chord length increased due to clustering of the PVC beads and the cube weighted chord length decreased due to the reduced probability of seeing the largest particles. Law et al., (1997) analysed the size distribution measured by the Par Tec 100 of various particles (pollen, sand, algal cells etc) in water at solids concentrations ranging from 10 to 50,000 mg L-1. As concentration increased the mean and median particle size remained relatively constant. The mean standard deviation was about 12% and the median standard deviation was about 5%. A slight trend to a larger mean particle size

31

at higher concentrations was attributed to overlapping of particles leading to a larger particle size. Monnier et al., (1996) made similar investigations for glass beads in water using the Par Tec instrument but found the mean size to be independent of solids concentration in the range 0 and 450 g L-1

2.2.4.2 Agitation Rate

Effective agitation is critical to ensure that a representative sample of the particulate system is presented to the window. Heath et al., (2002) noted a reduction in the average measured size as agitation intensity was reduced. Settling of the particles was visibly noted confirming that the large particles were falling out of the measurement zone impacting on the measured size.

Statistical robustness may also be improved by agitating at a sufficient level. Pearson et al., (2004) examined the effect of agitation rate on the measured CLD of a filamentous bacteria system. Agitation was varied between 100 and 500 rpm at a constant solids concentration. As agitation increased the total number of counts increased as more of the mycelial aggregates were presented to the probe window. The mean and median remained constant indicating the new particles presented to the window were no different in size. However by increasing the count data a more robust measurement was achieved.

Care must also be taken not to entrain air into the system at elevated agitation rates. Dowding et al. (2001) investigated the influence of agitation intensity on the CLD of

32

PVC in water at a fixed concentration (2.3 w/w%). The average chord length increased with increasing agitation and further investigation showed that at higher agitation levels air bubbles were been entrained in the suspension. These bubbles were measured by the FBRM as particles in the system and the average chord length increased. Monnier et al., (1996) investigated the effect of agitation intensity on the size distribution of glass beads measured by the Par Tec 100 instrument in toluene and water. In water the total counts increased 10-fold when the agitation intensity was increased from 250rpm to 750 rpm. However, at higher agitation levels (up to 2000rpm) the total counts actually dropped. This was not the case for the toluene system where total counts increased with increasing agitation from 500 rpm to 1500 rpm. At the highest values the total counts increased significantly. This was put down to bubbles being entrained at this high level of agitation. The mean size of particles measured increased with increasing agitation. This may have been due to a more representative sample being presented to the window.

An increase in agitation intensity does not always result in changes to the chord length distribution. Law et al., (1997) found no effect on the size distribution of various particles (pollen, sand, algal cells etc) generated by the Par Tec 100 instrument for a change in the agitation intensity. The agitation intensity was varied between 500 and 1000 rpm and no effect on the mean or median size was observed.

2.2.4.3 Particle Material

Since the FBRM technique is based on reflection the optical properties of the particles under investigation are extremely important. Sparks & Dobbs (1993) investigated a

33

number of different materials using the Par Tec instrument. They surmised that particle shape, or surface characteristics (spherical, flat, porous), absorbing properties (colour), refractive index (relative to suspension fluid) and high or low reflectivity may influence the measurement. They found that large transparent particles in general displayed very low count rates while opaque highly reflective particles displayed the highest count rates. It has been noted that in some cases FBRM is a poor technique for measuring extremely reflective particles such as TiO2. This is due to specular reflection in which light is reflected in every direction except back to the probe. For most particulate systems, however, this is not a problem

Ruf et al., (2000) showed the importance of the optical properties of the particle material by measuring a single particle of ceramic material and a single particle of quartz in toluene. For the ceramic material a single chord length was returned when the particle was measured at a specific position. However for the quartz particle a number of chord lengths were measured. This was due to a splitting of the signal due to different backscattering properties of different faces of the crystal and a weakening of the signal. This effect was not observed when the particle was measured in air. This was due to the more favourable refractive index difference between the fluid and the particle. Heath et al., (2002) compared the CLD generated by various size fractions of aluminium and calcite particles. These particles were chosen because they had similar densities and therefore a similar volume fraction when made up by weight leading to similar flow properties to maintain suspension. The calcite particles showed a larger tail in the fine end of the unweighted chord length distribution and this was attributed to abrasion of fine particles from the particles and the finer edges of the particles producing more fine counts. The aluminium particles were not subject to this abrasion

34

and had more round edges leading to a more symmetrical unweighted distribution. The square weighted distributions proved remarkably similar despite dissimilar morphologies and a presumed difference in reflectivities. This was not expected to be applicable to all materials and was supposed to be more a consequence of the relatively low aspect ratio of the two materials.

2.2.4.4 Surrounding Medium

The medium in which particles are suspended may also influence the FBRM measurement. This is typically down to differences in refractive index associated with different solvents. Monnier et al., (1996) investigated the influence of solvent on the size distribution of glass beads in water, THF, toluene and isoctane using the Par Tec instrument. For the organic liquids the total counts increased at all concentrations as refractive index decreases. This was not the case for water however where counts were higher compared to those in isoctane despite having a higher refractive index. This indicates that there is a parameter other than refractive index that influences the total counts. The size distribution of the measured glass beads was very similar in all solvents. Ruf et al., (2000) also showed the impact of refractive index by measuring the chord length of a single 200m ceramic bead in water, toluene and air. They showed that when measuring the ceramic bead in solvent the actual focal point is different to the nominal focal point. This is due to the difference in refractive index of air, found inside the probe and the solvent, found outside the probe. It was suggested that the focal point should be kept at or inside the window to minimise the number of times the light passes through a medium of differing refractive index.

35

2.2.4.5 Temperature

In many particulate systems, specifically crystallization, the temperature of the system will change. Monnier et al., (1996) investigated the effect of temperature on the particle size distribution measured by the Par Tec 100 instrument by measuring the evolution of mean size and total counts for glass beads in water as the suspension was subjected to a series of heating and cooling ramps. Total counts decreased form 15000 s-1 to 12000 s-1 for an increase in temperature of 60 C. The counts returned to their original value on cooling back to the original temperature. Mechanical effects on the probe, i.e. expansion of the probe, were discounted due to the small variation in temperature. The refractive index of the suspension changes with increasing temperature and this may have an effect but the refractive index of the solvent decreases with an increase in temperature and this should lead to an increase in counts at elevated temperatures. No parameter was identified for this phenomenon. The mean size was unaffected by any change in temperature.

36

2.2.5 Correlating FBRM measurements with other particle size analysers

In order to further validate the FBRM measurement a number of researchers have sought to compare the chord length distribution measured by FBRM and the particle size distribution measured by other size measurement techniques. Correlation between the FBRM data and other particle size analysis techniques will depend on a number of factors, including, technique investigated, particle system, and sampling mechanism. FBRM data has been compared to a number of different techniques for various systems and Heath et al., (2002) provides a comprehensive review of much of the research carried out in this area.

Laser diffraction is an industry standard particle size analysis tool and is used extensively in the pharmaceutical industry for product release. For this reason much work has been devoted to comparing it to the FBRM measurement. Alfano et al., (2002) produced a correlation between the mean chord length measured by FBRM and the median particle size measured by laser diffraction for microcrystalline cellulose. However, it was noted that when assessing the validity of the correlation two important issues come into play. Firstly, laser diffraction yields a true spherical equivalent particle size distribution whereas FBRM provides a chord length distribution that is related in a complex way to the true particle size and shape. Secondly, to measure a particle size distribution using laser diffraction it is necessary to pull a sample of the particulate system from the reaction vessel whereas the FBRM probe measures the system in-situ. They noted that the FBRM probed the microcrystalline cellulose slurry in the high-shear region of the impeller tip region, whereas the laser diffraction measures sample aliquots taken from the low-shear bulk

37

region, transferred to a low-shear sample cell. Additionally, the laser diffraction method required a 30-fold dilution of the sample. These two points clearly highlight the difficulty associated with comparing these techniques. However, despite these difficulties a number of researchers have achieved useful correlations between FBRM and laser diffraction for spherical or near spherical particulate systems.

Heath et al., (2002) found a linear relationship between the d50 measured using laser diffraction and the median chord length measured by the FBRM for aqueous aluminium suspensions. For large particles the correlation was improved by applying a length weighting to the FBRM data. A similar linear relationship between the techniques was observed by Abbas et al., (2002), for aluminium oxide particles. In this case, the volume weighted mean measured by FBRM and the volume based diameter measured using laser diffraction were used and the study only concentrated on particles less than 25m. It was noted that a more critical evaluation of the any correlation would only be possible if the size range studied was increased to 500m or more.

Another common particle size analysis technique that has been compared to the FBRM measurement is sieving. McDonald et al., (2001) found a correlation between the mean cube weight of the chord length distribution and the particle size measured by sieving for plant cell suspension cultures. The correlation was dependent on the system studied with wild Chinese cucumber cells producing a linear correlation and rice cells producing a non-linear fit. Tadayyon & Rohani (1998) compared the size distribution of ion exchange resin generated by the Par Tec 100 and sieving. For spherical particles there was a good correlation between the volume weighted chord

38

length distribution and the unweighted particle size distribution generated by the sieve. However the difference between the actual particle size distribution and the chord length distribution increased as the particle shape deviated from a sphere and the size distribution became non-normal..

Comparisons between the size distribution measured using the Par Tec 100 and various measurement techniques, including laser diffraction, image analysis and coulter counter have also been made. In general it has been observed that the Par Tec device overestimated the size of the smallest particles and underestimated the size of the largest particles. This was the case for a number of systems including fluvial suspended sediments, Philips & Walling (1998), sand particles, Law et al., (1997), and a variety of spherical particles (coulter microspheres, Bayer resin, orgasol and micronized pharmaceutical compound), Monnier (1996). In the case of Philips & Walling (1998), and Law et al., (1997), a calibration model was derived that accounted for this deviation.

Sparks & Dobbs (1993) compared the spherical equivalent diameter and volume diameter generated by the Par Tec device with a Microtrac laser forward scatter based technique. For opaque lightly coloured material such as aluminium hydrate results between the two methods correlated well. For large transparent particles the correlation was poor.

39

Comparing size distributions using different techniques is difficult. First and foremost the various techniques produce a measure of a different particle size property. Abbas et al., (2002) noted that when comparing the size distribution of aluminium oxide using laser diffraction Turbiscan and FBRM the Malvern returned a volume based measure D[4,3], the Turbiscan measured a mean D[3,2] based on the volume fraction and the FBRM reported a weighted mean based on a chord length distribution. Also, as noted above, Alfano et al., (2002) saw the difficulties associated with correlating between two techniques with different measurement principles as well as the inherent sampling issues associated with off-line techniques. Despite these issues it is possible to form useful correlations between the FBRM measurement and other size analysis techniques. These correlations will vary greatly depending on the measurement technique used and the system under investigation Heath et al., (2002). . It is clear that a large volume of research has been carried out on the comparison of the FBRM method with other particle size analysis techniques. A number of empirical correlations have been found that typically work well for spherical particles with suitable optical properties. In addition to empirical methods theoretical studies have been carried out to investigate whether the chord length distribution measured by FBRM is a true representation of the particulate system in question.

2.2.6 Theoretical Modelling of FBRM Data

The use of theoretical methods to convert chord length data into particle size data has gained much attention in recent years. Empirical methods have been used to compare the chord length measurement with other particle size analysers and the degree of

40

success in these correlations depends greatly on the shape of the particle. A suitable model that can convert the measured chord length of a variety of particle sizes and shapes into a usable particle size distribution would prove useful in further validating the FBRM measurement.

In a more practical sense a correlation between the FBRM measurement and another particle size measurement device may prove useful industrially. The standard for particle size measurement across many industries, most notably the pharmaceutical, is laser diffraction. A calibration between the chord length data measured using FBRM and the mean size measured by laser diffraction would prove extremely useful in predicting whether a crystal slurry, for example, was within specification.

In addition to this practical application the conversion of chord length data into particle size data is an extremely difficult problem and is mathematically complex. The difficulty of the problem and the challenges associated with it have attracted many researchers. Initial study in this area concentrated on modelling the FBRM measurement of spherical particles. The projected area of a spherical particle is always a circle no matter what the orientation of the particle. This makes conversion of the chord length data into particle size data less challenging than for other morphologies.

2.2.6.1 Modelling Spherical Particles

For spherical particulate systems, e.g. bubbles and droplets, a number of researchers have had success converting chord length distributions into particle size distributions.

41

Liu et al., (1998) analysed the relationship between bubble sizes and chord lengths in a heterogeneous bubbling system. They developed a forward transform to calculate a chord length distribution from a known bubble size distribution and three backward transforms to infer bubble sizes from a chord length distribution. The techniques were tested and compared using Monte-Carlo simulations. Simmons et al., (1999) presented two methods for estimating droplet size distributions from chord length measurements. The first is a probability apportioning method (PAM), that assumes particles are randomly cut and calculates the diameter probability distribution from each chord size detected. The distribution is accumulated over all chord measurements. The second is a finite element method (FEM) that solves simultaneously the equations relating the chord data and the diameter distribution. Both methods were tested on sets of chord data developed from ideal probability functions, a distinct element simulation and actual experimental data. It was found that the PAM method was fairly robust when the particle diameters were known, but the FEM method was generally more applicable when there was a wide range of unknown particle diameters. Both models predicted the chord length distribution of droplets in a 63mm pipeline measured by FBRM with a high degree of accuracy. The data returned by the FEM model proved slightly better. It was concluded that the FEM model was better for engineering applications where actual particle sizes are unknown. Langston et al., (2001) improved the PAM method proposed by Simmons et al., (1999) to take into account particles of unknown size distribution by incorporating Bayes theorem and an iterative procedure (PAM 2). The improvement comes from the implementation of the chord length data in a collective manner rather than in isolation. A range of particle size distributions were modelled accurately where the size of the particles was known and unknown. Experimental studies were

42

carried out on droplets of oil in water and the model predicted the measured data accurately.

Barrett & Glennon (1999), moved away from perfectly spherical systems and successfully modelled FBRM chord length data of dilute aqueous alkaline frit suspensions of known size distribution, using a method similar to that of Simmons et al., (1999). They assumed that the suspension was dilute enough to ensure particle interactions were negligible and a good agreement between experimental results and the model was achieved. It was concluded that the good agreement was only expected if the particles were spherical, or at least if the dominant projected area was circular in profile.

The research outlined above indicates that there has been some success in modelling the chord length distribution of spherical particles. When modelling spherical particles a 2-dimensional approach is sufficient since the orientation of the particle does not affect the projected area. However for non-spherical particles, i.e. ellipses, the orientation of the particles plays a very important role since it affects the projected area on which a chord length is measured.

Tadayyon and Rohani (1998) developed a model to predict the response of the Par Tec instrument in measuring the CLD of a suspension of spherical and ellipsoidal particles and to infer the actual PSD using the measured CLD output. The model showed that the measured CLD was reasonably accurate for spherical particles however the measurement progressively deteriorated as particles became more

43

ellipsoidal. For spherical particles the error in the mean sizes reported by both methods was improved by employing a volume weighted mean on the Par Tec data. A further aspect of this research used the Par Tec instrument to examine a suspension of ion exchange resin of normal size distribution in water. The chord length data was transformed into particle size data using a model and compared to the actual particle size distribution measured by sieving. For a normal size distribution the model predicted the particle size distribution well. However for a non-normal size distribution the model performed poorly.

Tadayyon and Rohani (1999) showed this when the model they proposed was unsuitable for ellipsoidal particles and became more and more unsuitable as the ellipse became more elongated and less spherical. For real particulate systems, especially crystal systems, the particle morphology is typically highly non-spherical. In order to model the chord length distribution of a real particulate it is absolutely necessary to analyse non-spherical particles.

2.2.6.2 Modelling Non-Spherical Particles

Modelling non-spherical particles is difficult due to the difference in the projected area depending on the orientation of the particle. A number of researchers have developed methods to account for this that are usually computationally intensive. Langston et al., (2001) attempted to analyse non-spherical particles using a 2dimensional approach by developing the PAM 3 model, an improvement on the PAM 2 method, Langston et al., (2001). This model was used to predict particle size distributions from chord length measurements in a 2-dimensional assessment of non-

44

circular shapes, which included circles, ellipses and blocky shapes. They supposed that the model was applicable to 3-dimensional assessments but conceded that the random orientation of the particles would have to be accounted for. Wynn (2003) derived a relationship between the underlying particle size distribution and the measured chord length distribution measured using FBRM. A number of assumptions regarding the measurement technique were taken into account. The principle of the method is similar to earlier peeling methods, but an improvement is made here in that the method is shown to be stable in some cases, due to finer spacing of the size intervals.

In order to account for random particle orientation a 3-dimensional approach is needed. Ruf et al., (2000) proposed a 3-dimensional model to transform particle size distributions into chord length distributions. The shapes of different particles were defined mathematically and the 2-dimensional projection of each particle was calculated for every possible orientation in 3-dimensional space. The chord length distribution of each 2-dimensional projection was calculated and by summing these individual distributions and applying suitable weightings the overall chord length distribution was calculated. Theoretical chord length distributions were calculated for a variety of particle morphologies including ellipsoids and cuboids. This model represents an improvement on the 2-dimensional model proposed by Tadayyon & Rohani (1999). It can account not only for particles of non-spherical morphology but it can also account for particle populations of the same shape but different size. Li & Wilkinson (2005) and Li et al. (2005) presented a theoretical and experimental study of the conversion CLD to PSD and PSD to CLD for non-spherical particles. Analytical solutions were used to calculate the PSD-CLD models for spherical and

45

ellipsoidal particles and numerical solutions were used to non-spherical particles. To convert CLD to PSD a non-negative least squares (NNLS) method that is insensitive to the measurement noise and the particle shape was employed. For the model to work it is necessary to give it some information on the shape of the particles in question. The effectiveness of the proposed methods was validated by extensive simulations. To test the validity of these models on a real experimental system the particle size distributions of ceramic beads, plasma aluminium and zinc dust were measured using image analysis and compared to the restored particle distribution calculated from experimental chord length data. Image analysis was used to define the shape of the particles and values of sphericity, circularity and chunkiness were used in the model. The results compared well for each of the test particles. The accuracy of the translation depended heavily on particle shape, which determines the optimal aspect ratio that is used in the model.

Worlitschek et al., (2005) developed a method to restore the PSD of a particulate system from the measured CLD. The restoration constitutes a mathematically illposed problem and its solution is based on a two-step procedure. Firstly, the computation of a matrix that converts the PSD of a population of particles with given shape into the corresponding CLD using a 3-dimensional geometric model and secondly the solution of the resulting linear matrix equation for the PSD. The restoration of the PSD of spherical particles represented the easiest case whereas for non-spherical shapes such as octahedral and needles lead to highly ill-posed inversion problems. However, experimental investigations were carried out on populations of paracetamol crystals (octahedral shape) and the model predicted measured data accurately.

46

It is clear that the goal of producing a particle size distribution from a measured chord length distribution is becoming a reality. The restoration of the PSD of a spherical population of particles from a measured CLD is a relatively simple problem and recent work has shown that this is now possible to restore the PSD of populations of ellipses and octahedra. Further work is needed to model highly non-spherical morphologies such as needles.

2.2.7 FBRM for Crystallization Characterization

While the possibility of inferring actual particle size distributions from chord length distributions may be possible in the future it should be noted that all of the current models require some information on the shape of the particles. For a well-behaved system of spherical particles, such as oil in water, this is possible, but for a crystallization system this is extremely difficult and may be impossible. In a pharmaceutical crystallization system, for example, the particle morphology is typically needle-like and the aspect ratio is usually non-uniform. Attrition can lead to numerous small crystals with a low aspect ratio and agglomeration can lead to large disordered particles with no recognisable shape. For a polymorphic system the shape of the crystals can change over the course of the crystallization. These factors make it extremely difficult to decide on the shape of the crystals under investigation.

FBRM avoids this problem by assuming no shape. The chord length distribution is a function of the size, shape and number of particles that enter the measurement zone. In a crystallization system, the chord length distribution will change if the size, shape

47

or number of crystals entering the measurement zone changes. Consequently key crystallization phenomenon such as nucleation, growth, attrition, agglomeration, dissolution and polymorphic transition can be identified and monitored. To facilitate identification and monitoring of these dynamic crystallization processes, PVM may be implemented as c complementary technique. When one considers that this can be done in situ in real time the power of FBRM for monitoring crystallization systems become apparent. When changes to the chord length distribution can be related to process parameters, such as cooling rate, anti-solvent addition rate, agitation or seed loading, it is possible to design a crystallization that will produce a suitable particle size distribution that behaves appropriately during downstream processing operations such as filtration, drying and formulation.

The vast majority of the research carried out on crystallization using FBRM is performed in the pharmaceutical industry. Due to the secretive nature of much of the research it is rare for work to be published. In fact the number of published articles documenting the use of FBRM for the characterization of crystallization is surprisingly few. Some of the research undertaken industrially is presented at conferences, the proceedings of which can be useful. Of particular use is the Mettler Toledo Users Site www.mt.com/lasentec where papers presented at the annual Mettler Toledo Real Time Analytics Users Forum (formerly the Lasentec Users Forum) are posted. While one must be wary of studying in too much detail the research presented by the vendor itself, it nonetheless provides a wealth of information on how FBRM has been used to design, develop and optimise crystallization processes in many of the major pharmaceutical companies.

48

2.2.7.1 Solubility Curve and Metastable Zone Width Determination

The starting point for the characterization of any crystallization system is the solubility curve. Barrett and Glennon (2002) implemented the polythermal method, Nyvlt (1968), to generate solubility data for aluminium potassium sulphate in water and used FBRM to monitor the point of dissolution. They chose counts between 50m and 250m to monitor the dissolution as large particles have a large surface area to volume ratio, and hence dissolve at a slower rate, compared to smaller crystals. The results obtained were in excellent agreement with other published data, Mullin et al., (1965). In addition to solubility data, metastable zone width information was generated for the same system. FBRM was used to detect the onset of nucleation when solutions of aluminium potassium sulphate were cooled at various rates. The number of counts between 0 and 20 m was used to detect the onset of nucleation. Since the instrument was set to measure every 10 s this size range was chosen since published data reported that aluminium potassium sulphate may grow up to 15 m in 10 s. Fujiwara et al., (2002) performed similar work, on an aqueous paracetamol system, but compared the three techniques for the measurement of the MSZW, FBRM, ATR-FTIR and visual observation. FBRM proved to be the most effective technique as it reliably identified the onset of nucleation earlier than the other methods. Various weightings on the chord length distributions were examined to see which provided the best data. A 1/length weighting was tested to concentrate on the small particles but increased noise and drift of the background signal was observed. A square weighting was also tested but resulted in a slightly delayed detection. The most reliable response was from the unweighted chord length distribution and this parameter was chosen to identify the onset of nucleation. Liotta and Sabesan (2004)

49

showed that this technique could be automated, reducing the time and manpower associated with measuring solubility and MSZW information. To do this required a link that enabled the crystallizer to respond in real time to the FBRM data. A solution of high concentration was held at a temperature above the solubility point and then cooled until crystals nucleated. The FBRM detected the onset of nucleation (point on the metastable zone) and when the total counts exceeded a certain threshold the cooling ramp was terminated and a heating ramp was initiated. This heating ramp continues until the total counts fall to the baseline (solubility point). At this point the solution is held at an elevated temperature and a solvent is added to dilute the system. The process is then repeated to gather further solubility and metastable zone width data. An automated technique such as this significantly reduces the time needed to generate solubility and metastable zone width data.

2.2.7.2 Nucleation Kinetics

FBRM is very useful for tracking nucleation as it can focus on specific size ranges. By focussing on the smallest chords and trending them over time it is possible to characterize nucleation behaviour. This is also facilitated by the sensitivity FBRM displays to small particles. Shi et al., (2003) used FBRM to monitor the evaporative crystallization of burkeite. Using FBRM trends the crystallization process was divided into three regimes each dependant on the dominant mechanism of nucleation involved. Unusual nucleation behaviour was observed and the inhibitory effect of specific impurities on the nucleation of burkeite was identified. Monnier et al., (1997) used the Par Tec instrument to calculate nucleation rates for the batch crystallization of adipic acid in water. Global nucleation rates were calculated using Par Tec and by

50

subtracting the primary nucleation rates from this it was possible to calculate secondary nucleation parameters. This information was then used in a model that predicted the final size distribution of the adipic acid crystals. The PSD calculated using the model compared well image analysis.

Nucleation kinetics can also be inferred from MSZW information gathered using FBRM. Barrett & Glennon (2002) used such information to calculate nucleation kinetics for the aluminium potassium sulphate. The results agreed with literature data

2.2.7.2 Monitoring Crystal Size

Even though FBRM does not report the true particle size it is sensitive to the size, shape and population of the crystal slurry. FBRM is useful for comparing these parameters under different process conditions. Loan et al., (2002) showed this by comparing ferrihydrite precipitation at two different pH values. They found that at low pH smaller crystals were formed and that poor filtration characteristics for the low pH process were as a result of the unsuitable particle size distribution. FBRM was used for a similar purpose by Fujiwara et al., (2002) for the study of the aqueous batch cooling crystallization of paracetamol. FBRM indicated an increase in the number of crystals during a run where supersaturation was high due to nucleation. For a second batch where supersaturation was kept at a constant low level there was no increase in the number of crystals indicating the absence of nucleation. FBRM has an advantage over off-line techniques die the fact that sampling is not necessary. Loan et al. (2002) noted that the FBRM chord length measurement was a superior indicator of the particle characteristics, compared with scanning electron

51

microscopy, since there was no sample removal, dilution or drying needed. A similar advantage was noted by Abbas et al., (2002) in that FBRM was the only technique suitable for reactive crystallization, since sampling was not needed. (Off-line sampling of the nickel hydroxide particle resulted in the alteration of the physiochemical structure).

Another advantage FBRM has over other techniques is that it can measure a lot of data in real time. This allows dynamic crystallization processes such as growth, breakage and agglomeration to be studied in real time. By focussing on size ranges of interest it is possible to identify these crystallization mechanisms. For example, breakage is identified by a decrease in the number of chords counted in large size ranges, and an increase in the number of chords counted in small size ranges, as well as a decrease in the mean chord length. Such analysis can be facilitated through the use of PVM to identify particle morphology as well as differentiate between mechanisms that are sometimes confused, such as growth and agglomeration. Abbas et al., (2002) used FBRM to monitor the reaction crystallization of nickel hydroxide and the cooling crystallization of ammonium sulphate. In each case FBRM monitored the size of the crystals over the course of the batch. For the reactive crystallization an initial increase in particle size was identified as agglomeration and the subsequent reduction in particle size was identified as the breakage of these agglomerates due to the action of the agitator. For the cooling crystallization the size of the particles increased steadily over the course of the batch. Kougoulous et al., (2005b) used FBRM to monitor the cooling crystallization of an organic fine chemical in a MSMPR crystallizer. Upon heating a decrease in particle counts was observed indicating dissolution of the organic fine chemical. Upon cooling counts increased indicating

52

nucleation. After a certain period of time the trends reached steady state and a recirculation loop was initiated. The FBRM trends remained steady indicating breakage due to the action of the peristaltic pump was not an issue (Kougoulos 2005a). After a certain period a secondary loop was opened and the MSMPR cooling crystallization process commenced. Crystals were removed from the system and FBRM successfully identified the size range of crystals removed. There was a significant decrease in coarse crystals and a significant increase in fine crystals. This indicated that large particles were being removed while the nucleation and growth of finer particles became more evident. Complementary video images provided by process video imaging (PVI) confirmed that this was the case. Steady state was identified after about 6-9 residence times by identifying the point at which the FBRM trends flat-lined. The same researchers used FBRM to monitor particulate attrition and breakage of an organic fine chemical in a turbulently agitated system Kougoulos et al., (2005c). A dilute solution of the organic fine chemical in isopropyl alcohol and water was agitated for 150 min at 300rpm. The impeller frequency was then increased to 400 rpm for 150 min and then to 500 rpm for a further 150 min. An FBRM probe placed in the vessel monitored the chord length distribution over the course of the experiment. The FBRM data indicated that that the number of fine particles (0-10 m) increased and the number of coarse particles (100-300 m) decreased. The number of intermediate particles (50-100 m) initially increased and then decreased. FBRM was successful in monitoring the death and birth of crystals by disruption for different size classes and provided evidence of micro-attrition effects. The advantage of FBRM, over other size measurement techniques for monitoring such a process was noted. This was because FBRM can monitor, in situ, the crystal size distribution for different particle size classes (fine, intermediate, coarse). Information on particle attrition and

53

disruption proved extremely useful for the prediction of particle behaviour upon scale up.

The previous examples have shown some of the utility FBRM has for the design, development and optimization of crystallization processes. In each case the chord length distribution was sufficient to characterize the system and there was no need to assume a shape or convert to a crystal size distribution. However, this can be achieved using some of the modelling work outlined in the previous section. Worlitschek and Mazzotti (2004) used FBRM, coupled with a model-based method to calculate particle size distribution from the raw chord length data (Worlitschek et al., 2005), to monitor particle size in the batch cooling crystallization of paracetamol in ethanol. An optimal temperature trajectory that minimised the difference between the final PSD (measured using FBRM) and a given optimal monomodal PSD was calculated using a non-linear constrained minimization algorithm. The size distributions of the crystals produced using optimal cooling profile and a simple linear cooling profile were compared. The FBRM data combined with the CLD-PSD model showed that the linear cooling profile produced a significant amount of small crystals while the optimal cooling profile provided a PSD close to the desired monomodal distribution.

2.2.7.3 Temperature Cycling

Temperature cycling is a useful technique used to increase crystal size and reduce the number of fines at the end of a batch. FBRM is useful here because the cycling regime (number of cycles, cycle duration temperature change) may be optimised by focussing on the removal of small chords and the increase in the mean chord length.

54

Doki et al., (2004) applied this technique from the start of a crystallization to ensure a large final product crystal size and minimise the number of fine crystals. FBRM was used to monitor the number of crystals formed in the aqueous batch cooling crystallization of glycine. The goal of the research was to produce crystals of a suitable particle size distribution and of the correct polymorphic form by alternating the temperature profile and the termination temperature. Initial work showed that there was a linear correlation between the number of crystals of glycine suspended in a saturated solution and the number of counts measured by the FBRM. This correlation was applicable for a number of size ranges. During the batch experiments seed crystals were added at the saturation temperature and FBRM monitored the number of crystals in solution as cooling progressed. When the number of crystals in solution exceeded a threshold value (due to nucleation) cooling was terminated and a heating step was initiated to dissolve the fine crystals. When the number of crystals measured returned to the original value the cooling step was restarted. In this way large particles of suitable size distribution were formed.

1.2.7.4 Polymorphic Transitions

In certain cases different polymorphs have different morphologies. Since FBRM is sensitive to changes in shape for these certain instances it can be used to track polymorphic transitions. O Sullivan et al., (2003) used FBRM to monitor the polymorphic transition of metastable -form mannitol to the stable -form. A small amount of -form mannitol was added to a saturated solution of -form mannitol in water. FBRM trends were used to track the appearance of the -form as it was charged, the dissolution of the -form and the subsequent growth and nucleation of

55

the form. This - transition was corroborated using off-line X-Ray Powder Diffraction (XRPD) and differential scanning calorimetry (DSC) as well as in process imagery provided by PVM.

2.2.7.5 Effect of Impurities

FBRM is useful for measuring the rate of crystallization. Impurities often effect the crystallization and there impact can be assessed using FBRM. Scott & Black (2005) used FBRM to monitor the effect of impurities on the crystallization of urea and an in-house system. FBRM successfully identified a difference in the crystal growth rate when impurities were absent and present. Crystallization in the presence of the impurity was 7-times slower for the urea system and ~20-times slower for the inhouse system. The work identified the potential use of FBRM in the plant for the monitoring of impurities. In commercial manufacture impurity levels often change during process development and on scale-up. This can affect morphology and thus downstream processing operations. Additionally if impurity levels increase but the crystallization time remains the same the yield can be affected due to the retarded growth rate. Alternatively, if the impurity level decreases and the crystallization time remains the same, crystallization may be complete long before the end of the batch resulting in a decrease in productivity. This can be avoided by in-process tests or by the use of in-line technology.

56

2.3 PROCESS VIDEO MICROSCOPE (PVM)

2.3.1 Introduction

Imaging can provide useful information on the size and morphology of particles. These variables are important in any particulate system. In a crystallization system the size and shape of the crystals can influence downstream processing operations such as filtration, drying and formulation. For example, fine particles have a large surface area and can take a long time to dry. Large flat crystals may pack on top of each other leading to long filtration times. In polymorphic systems imaging can be extremely useful as it is often possible to distinguish between two polymorphic forms according to their shape.

The traditional imaging method used to gather information on size, and more specifically shape, has been off-line microscopy. Microscopy is a useful and relatively cheap method to gather data but there are some problems associated with it. To gather statistically robust data on the size of the particles hundreds of images must be taken and processed. In addition obtaining a representative sample can be difficult and is time consuming and labour intensive. Once a sample is isolated the slurry must be dried and prepared for analysis. This can alter the particles and lead to data that is not representative of what is actually going on in the process. For example brittle particles can break upon isolation, generating a large umber of fine particles. This may lead one to conclude that the process is producing the fine particles rather than the sampling method.

57

For a crystallization system there are some additional problems. Sampling is undesirable because once a sample is taken changes to the crystals can occur due to further crystallization phenomena e.g. nucleation, growth, breakage agglomeration dissolution or polymorphic transformation. In some situations it may not be possible to isolate a sample at all due to process or safety constraints.

The problems associated with off line imaging can be avoided by imaging within the process environment. The Particle Vision and Measurement (PVM) (Figure 1.6) system is a probe-based, high-resolution video microscope that can perform such a task. PVM uses six independent lasers of wavelength 905 m to illuminate the particulate system. The lasers source is located at the back of the probe and the laser light is sent to the probe tip via a fibre optic connection. At the tip there are six lenses arranged hexagonally, each of which focus the light from one laser onto an area with a fixed area of approximately 2 mm2. A micrometer positioned at the back of the probe can focus this illuminated area into or out of the system under investigation. As particles intercept the illuminated area light is reflected in all directions and some of it is reflected back up the probe. The backscattered light passes through an objective lens, focussing it onto a light sensitive charge coupled device (CCD) that produces the image. A recording of this image is made using a CCD camera at the back of the probe.

The PVM software allows 10 images per second to be recorded. Settings such as the gain and the offset of the CCD can be changed using the software to optimise the image. If the material under investigation backscatters too much light the laser intensity can be reduced. A useful feature of the software is that the laser intensity,

58

gain and offset can be altered automatically at specified time intervals to optimise the images obtained in a dynamic process, such as crystallization. This means the probe can be left unattended while images of sufficient quality are recorded.

A typical PVM unit consists of a probe, a field unit (which contains the electronics) and a computer for image storage and review. An FBRM 700L was used for this research and is 25 mm in diameter, has a wetted length of 300 mm and is made from stainless steel. The window of the probe is made from sapphire and a kalrez o-ring is used to seal the window to the probe tip. A more detailed description of the mode of operation of PVM is available elsewhere (Barrett (2002)).

PVM is an extremely useful tool in for the study of particulate systems as it can provide qualitative information on the size and shape of the particles in situ and in real time (Figure 1.7). Furthermore analysis of a sufficient number of PVM images can provide qualitative information on the size of the particles being studied. Additionally when PVM is used in conjunction with the quantitive information that the FBRM can provide, it is possible to achieve an excellent understanding of the particulate system under investigation.

59

Figure 2.6: PVM internals (Mettler Toledo Users Site)

Figure 2.7: PVM images

60

2.3.2 Characterization of Particulate Systems Using PVM

O Rourke & MacLoughlin (2005) PVM to analyse the evolving droplet size distributions in lean silicone oil-water dispersions in an agitated system of standard geometry. A satisfactory level of agreement between results obtained using PVM and the more traditional sampling method was achieved. PVM held an advantage over the sampling method in that it was less labour intensive and was suitable when there was a large density difference between the two phases. However a minimum of three minutes was required to acquire the large number of images needed for a representative sample. This made PVM unsuitable form monitoring very rapid changes in the droplet size distribution.

Barrett and Glennon (2002) used PVM to monitor the crystallization and dissolution of potassium aluminium sulphate. PVM was used to validate the rapid increase in the number of coarse crystals identified by the FBRM as the solution was cooled. PVM images highlighted the presence of fine crystals, but also the presence of large single crystals and agglomerates thus validating the FBRM observation. PVM images taken during the heating period showed the complex nature of the dissolution process. A clear transition between the distinctive octahedral shape of the crystals to a more rounded morphology was observed.

McDonald et al., (2001) used PVM to study rice, tobacco and wild Chinese cucumber cell suspensions. The PVM clearly identifies the different morphologies of the different suspensions without the need to sample.

61

2.4 ATTENUATED TOTAL REFLECTANCE FOURIER TRANSFORM INFRARED SPECTROSCOPY (ATR-FTIR)

2.4.1 Introduction

Supersaturation is a non-equilibrium state where there is more solute dissolved in solution than in the equilibrium state at a given temperature. A solute will remain in solution until there is a sufficient level of supersaturation to induce crystal formation. Supersaturation is the driving force for subsequent crystal nucleation, growth and agglomeration. As such it impacts on the final crystal size distribution. By measuring supersaturation the underlying mechanism that governs the final crystal size distribution can be controlled. In doing so the opportunity to produce crystals of a suitable size, number and morphology is afforded. A method for the direct measurement of supersaturation has been proposed by Lfflemann and Mersmann, (2002), however the most common technique for measuring supersaturation is to measure the solution concentration and infer the prevailing supersaturation from solubility data. In this case it is more accurate to say that supersaturation is being monitored, rather than measured.

Attenuated Total Reflection Fourier Transform Infrared (ATR-FTIR) spectroscopy has emerged as a useful technique for the measurement of solution concentration in crystallizations. It has an advantage over other concentration measurement techniques, as sampling is not required. This eliminates the problems associated with temperature-controlled external sampling loops and removes the need for phase separation devices. Additionally, multiple solute and multiple solvent crystallizations

62

can be characterised. This can be useful for anti-solvent and reactive crystallization as well as crystallizations where impurities are present. With accurate solubility data the measured solution concentration can be used to infer the prevailing level of supersaturation.

Briefly, ATR-FTIR spectroscopy measures solution concentration by irradiating the solution with infrared light to produce an infrared spectrum. This spectrum is characteristic of the vibrational structure of the substance in immediate contact with the ATR probe and can be described as a unique fingerprint of the liquid phase. An ATR crystal is chosen so the depth of penetration of the infrared energy is smaller than the liquid phase barrier between the probe and the solid crystal particles. Hence when the probe is inserted into a crystal slurry the probe should be in immediate contact with the liquid phase and interference from the solid phase should be negligible. Despite this assumption some researchers have noted interference of the spectrum by the solid phase in crystallization systems (O Sullivan, 2005). The liquid phase concentration is a function of the infrared spectrum generated and a calibration model is used to inter-relate the data. . A more detailed description of spectroscopy in general and the mode of operation of the ATR-FTIR probe is available elsewhere, (Dunuwilla et al., 1994; Togkalidou et al., 2001; Lewiner et al., 2001a; O Sullivan, 2005).

2.4.2 Initial Work Using Various Calibration Models

Initial studies using ATR-FTIR to measure solution concentration provide a useful insight into the evolution of the technique, the problems associated with using it and

63

perhaps most usefully the various calibration models that relate the infrared spectrum to the solution concentration.

Dunuwilla et al., (1994) were one of the first researchers to examine the use of ATRFTIR for the measurement of solution concentration. The solubility of citric acid in aqueous solution was measured by taking a sample from a crystallization vessel and placing it in a micro circle open boat cell equipped with an ATR rod. Relatively good agreement between the measured data and literature values was achieved. However at high temperatures the experimental solubility value was overestimated. Subsequent research by Grn & Roberts (1999) repeated this work in situ and showed results with increased accuracy at all temperatures. This indicates that the sampling method used by Dunuwilla et al., (1994) may have been poor.

Dunuwilla & Berglund (1997) were the first researchers to use the ATR-FTIR technique to monitor supersaturation in situ. The crystallization of maleic acid from aqueous solutions was examined using a parabolic, intermediate parabolic and linear cooling profile. The supersaturation was monitored for each of the cooling profiles and the final crystal size distribution was measured at the end of each batch by sieving. The ATR-FTIR successfully monitored the supersaturation in each case. Supersaturation for the parabolic cooling profile remained low and constant throughout the batch and large crystals were formed. For the linear cooling profile supersaturation was higher over the course of the crystallization resulting in an average crystal size that was half that produced using the parabolic cooling profile.

64

In this work supersaturation, solubility and the metastable zone width were represented in terms of the ratio of the transmittance of two maleic acid peaks, however a calibration between the transmittance ratio, concentration and temperature was supplied and the solubility data generated was within 3% of literature values.

Lewiner et al., (2001a) used ATR-FTIR to study the crystallization of three fine chemical products. For this study a different calibration model was used that calibrated the height of one relevant solute peak with the concentration of the solute and the temperature. The crystallization of Bifenox (a weed killer) from methanol solutions by cooling was examined. Combining supersaturation data with information on the size of the product crystals (measured by SEM and laser diffraction) it was concluded that the final product size was mainly controlled by agglomeration and secondary nucleation. The second system studied was the cooling crystallization of another weed killer, isoproturon (IPU), from ethanol solutions. The supersaturation measurements combined with information on the final crystal size distribution gathered using image analysis indicated a different crystallization behaviour. In this case there were large variations in the metastable zone width and low levels of supersaturation due to very fast crystal growth. Finally the cooling crystallization of an active pharmaceutical ingredient, compound F was studied. Compound F exhibited four polymorphic forms and the ATR-FTIR successfully monitored the transition from one form to the other over the course of a cooling crystallization.

Togkalidou et al., (2001) combined ATR-FRIR spectroscopy with robust chemometrics to produce a very accurate calibration model for the estimation of the solubility of potassium dihydrogen phosphate (KDP) in water. A number of

65

chemometric models were applied and the one that gave the most accurate predictions was selected. In this case the most accurate model arose from choosing a whole region of the spectrum rather than a single peak or a number of peaks. This improved accuracy may have been due to a significant shift in one of the peaks meaning more accurate predictions were made when all the data in the region was used or, it may have been due to an increase in the apparent signal-to-noise ratio resulting from the random noise in the absorbance being averaged over many more frequencies. The solubility of KDP in water was measured using a number of models and the results were compared. The solubility data gathered using a single peak chemometric model proved inaccurate, while the data produced using the whole spectral region proved this most accurate. This indicates that multiple absorbances and chemometrics produce the most accurate calibration models. The most accurate model produced concentration estimates with and accuracy of 0.12 wt%.

The research outlined above outline three different methods for the calibration of the infrared spectrum with concentration. Lewiner et al., (2001b) commented on the accuracy of the data presented by Togkalidou et al., (2001), describing it as remarkable. However they noted that in industry there is a reluctance to use, so called chemometric methods since a calibration using well-defined absorption bands has a clearer chemical and physical meaning. However many of the ATR-FTIR probes available for commercial use include useful chemometric packages that can be used simply and effectively to generate an accurate calibration model. For the purpose of this study the chemometric software package QuantIR, supplied with the Mettler Toledo ReactIR 4000 ATR-FTIR probe will be used to generate calibration models.

66

2.4.3 The Use of ATR-FTIR to Characterize Crystallization Processes

2.4.3.1 Seeding

Lewiner et al., (2001b) continued previous studies of the cooling crystallization of isoproturon (IPU) from ethanol solutions using a calibration model based on absorption at a single peak. Unseeded crystallization resulted in large variations in the metastable zone width, to the extent that no identifiable metastable zone width could be identified at a constant cooling rate. This variability resulted in a highly variable final crystal size. Supersaturation measurements indicated that the point of nucleation was followed by a dramatic decrease in the supersaturation regardless of the cooling rate used, indicating that supersaturation was minimal during the growth period. By introducing a seeded crystallization the variability in the metastable zone width was reduced and the final mean crystal size was increased and the CSD narrowed. A suitable seeding temperature was identified by monitoring the supersaturation for runs where the seed was introduced at different temperatures. When the seed was introduced close to the solubility curve supersaturation continued to increase until there was a burst of secondary nucleation. When the seed was introduced close to the metastable zone width a rapid decrease in the supersaturation was observed similar to the unseeded case indicating the introduction of the seeds was not effective. An intermediate seeding temperature proved the most effective as the final crystal size increased and the coefficient of variation decreased. With a suitable seeding temperature identified it was possible to assess the effect of the seed mass and the cooling rate after seeding. Supersaturation was measured for different cooling rates after seeding. Higher cooling rates led to increased supersaturation and hence a higher nucleation rate that was confirmed by a decreased final product size. With a suitable 67

cooling rate identified the effect of the seed mass was assessed. An increase in the mass of seeds added resulted in a decrease in the prevailing level of supersaturation due to the increased surface area available for growth.

2.4.3.2 Oiling Out

Groen & Roberts, (2001) combined ATR-FTIR supersaturation measurements with optical turbidity measurements to identify liquid-liquid phase separation (also known as oiling out) in the cooling crystallization of citric acid. Batch cooling experiments were conducted and a reduction in optical transmittance was observed a significant length of time prior to the onset of nucleation measured by a reduction in the measured supersaturation. This change in the optical properties of the system without crystallization indicated the formation of non-crystalline phase. Oiling out was suggested as a possible explanation for the phenomenon as the supersaturation level was extremely high and previous researchers had noted similar results for the citric acid water system. The actual reduction in the solute concentration due to the phase separation was estimated to be about 1 % (w/w), enough to facilitate nucleation. This value was similar to the peak-to-peak noise of the ATR-FTIR signal. Hence no change in the supersaturation was observed.

68

CHAPTER 3: SOLUBILITY MEASUREMENT FOR AN ANTISOLVENT SYSTEM USING GRAVIMETRIC ANALSYSIS, ATRFTIR AND FBRM

3.1 ABSTRACT

The solubility of benzoic acid in ethanol-water mixtures was measured using gravimetric analysis (solid analysis and liquid analysis), Focused Beam Reflectance Measurement (FBRM) and Attenuated Total Reflectance Fourier Transform Infra Red Spectroscopy (ATR-FTIR). A suitable method that describes the solubility curve for an anti-solvent addition system is presented. By expressing the solubility on an anti-solvent free basis, the dilution effect may be eliminated and expression of the metastable zone width (MSZW) and supersaturation is simplified. The FBRM, ATRFTIR and solid analysis methods produced similar solubility data validating the techniques and highlighting the accuracy of the data. The liquid analysis underestimated the solubility. This may have been due to the esterification of the benzoic acid whilst drying. A correlation between the measured data and the predictions of the UNIQUAC liquid activity coefficient model is also presented.

3.2 INTRODUCTION

Key to the successful design and optimisation of a batch crystallization process is knowledge of the solubility over the range of operating conditions likely to be encountered during the batch. In processes involving the crystallization of a solute through the addition of an anti-solvent to a saturated, or near saturated, solution the 69

solute solubility should be known over a wide range of solvent mixture concentrations. Gravimetric analysis is the traditional standard method for the determination of solubility (Mullin, 1993) and its use has been widely reported for a range of anti-solvent systems (e.g. Hoijati & Rohani, 2006a; Granberg and Rasmuson, 2000). Perhaps its principle drawback is the time-consuming nature of the method. The use of in ATR-FTIR has become an increasingly popular technique for investigation of solubility characteristics in crystallization systems including antisolvent processes (Hoijati & Rohani, 2006a). ATR-FTIR requires a substantial effort to initially calibrate the system, but having developed the calibration data set, subsequent application is relatively quick. Investigations based on the Focussed Beam Reflectance Measurement (FBRM) probe have generally involved cooling crystallization processes through the application of an automated polythermal method for solubility determination (Barrett and Glennon, 2002). FBRM-based methods have the advantage of requiring no calibration and of rapid measurement duration. However, measurement of solubility in anti-solvent systems is not immediately amenable to automation.

A priori prediction of solubility has been less successfully demonstrated to date, although the use of liquid activity coefficient models has facilitated good correlation of available experimental data (Gracin et al, 2002; Hoijati & Rohani, 2006b).

In this paper, a comparison of these three techniques for the measurement of solubility in anti-solvent systems is presented, using benzoic acid-ethanol-water as the model system. The correlation between the measured data and the predictions of the UNIQUAC liquid activity coefficient model is also presented.

70

3.3 EXPRESSION OF SOLUBILITY

In an anti-solvent system, supersaturation is generated by a combination of reduced solubility due to the action of the anti-solvent, and dilution. Typically, solubility is expressed in terms of the mass of solute per total mass (or volume) of solvent and anti-solvent. The combination of these factors makes expression of key crystallization variables, specifically supersaturation and the metastable zone width, difficult. To overcome this problem researchers have generally chosen to depict the path of antisolvent addition on the solubility curve as a diagonal line that accounts for the dilution and reduced solubility (Granberg et al., 2001; Pina et al., 2001; Takiyama et al., 1998; Kaneko et al., 2002). Sometimes dubbed a mixing line this diagonal line represents the apparent concentration of the solution once the solution and anti-solvent have been mixed, in the absence of crystallization. In this case expression of the supersaturation and metastable zone width becomes potentially complex.

Figure 3.1 is a traditional solubility curve with the solubility expressed in terms of the total mass of solvent and anti-solvent. A diagonal line on the solubility curve represents the change in concentration as anti-solvent is added. Expression of the supersaturation generation rate is, for such systems, cumbersome (Pina et al., 2001).

71

0.9 Solubility Data 0.8

g solute / (g solvent + anti-solvent)

Solute Concentration 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 20 40 anti-solvent % 60 80

Figure 3.1: Change in concentration, as anti-solvent is added.

Solubility is generally given as a function of anti-solvent concentration, i.e. cs = cs(m). Therefore,

dcs dt dcs dm dm dt

Eq. 3.1

When concentrations are reported in terms of g/g solution, and a constant anti-solvent addition rate is used:

m t

a Rt w a Rt
R s s w a w Rt
2

Eq. 3.2

dm dt

Eq. 3.3

c t
dc dt

s s w a Rt
sR s w a Rt
2

Eq. 3.4

Eq. 3.5

72

Therefore, the supersaturation generation rate is given by

sR s dcs R s w dm 2 w a Rt

d c dt

dc dt

dcs dt

Eq. 3.6

When concentrations are reported in terms of g/g solvent (i.e. on an anti-solvent-free basis):
m t
dm dt c t s w

Rt w
R w

Eq. 3.7

Eq. 3.8

Eq. 3.9

dc dt

Eq. 3.10

Therefore, the supersaturation generation rate is given by

d c dt dcs R dm w

Eq. 3.11

By defining all concentrations on an anti-solvent free basis, the solute concentration follows a horizontal path as anti-solvent is added and is analogous to cooling on a temperature-concentration curve (3.2). In this way the metastable zone width is clearly defined as the difference in anti-solvent concentration at the point of nucleation and at the point of saturation. The supersaturation is the difference (or ratio) between the actual concentration and the equilibrium concentration.

73

0.9 Solubility Data 0.8 0.7

g solute / g solvent

Solute Concentration

MSZW (g/g)

0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 g anti-solvent / g solvent

C (g/g)

Figure 3.2: Change in concentration (anti-solvent free basis), as anti-solvent is added.

3.4 EXPERIMENTAL WORK AND ANALYSIS

3.4.1 Gravimetric Analysis

An ethanol-water mixture of known composition is charged to a 500ml glass jacketed vessel and the temperature allowed to equilibrate at 25 C. A known excess of benzoic acid is added making the total solution mass about 150 g. The solution is then stirred at 300rpm for 2 hours. Ideally the solution should be left for 24 hours, however experiments conducted using the ATR-FTIR indicate that equilibrium is reached after as little as 10 minutes. At this point the slurry is filtered and the filtrate and filter cake placed in separate conical flasks and covered with filter paper. The filter paper allows 74

the evaporation of the solvent but prevents the apparent sublimation of the solute, a phenomenon observed during initial runs. The conical flasks is placed in an oven at 80 C and left for a number of hours. The mass of the filtrate and filter cake are measured at regular intervals to identify when drying is complete. When three consecutive measurements yield three similar masses the sample is deemed to be dry. This typically occurred after 24 hrs for the filter cake and 60 hours for the filtrate (Figure 3.3).

Two methods were used to determine the solubility. Firstly, the mass of solute remaining once all of the solvent from the filtrate had evaporated provided liquid analysis solubility. Secondly, the mass of undissolved solute remaining in the filter cake once it had been dried was subtracted from the initial charge of solute to the vessel to provide solid analysis solubility (3.4).

It is clear that there is a systematic difference between the solubility values generated from the solid analysis and the liquid analysis. The liquid analysis method appears to underestimate the solubility over all the water concentrations. There are a number of possible reasons for this. It was difficult to transfer all of the slurry from the crystallizer into the filter. Every effort was made to ensure all the saturated solution and solute was transferred but inevitably some was left in the crystallizer. It is also possible that there may have been an esterification reaction between the benzoic acid and the alcohol at the elevated temperatures in the oven. It is possible that an ester formed and evaporated with the rest of the solvent mixture. By drying samples at a low temperature in a vacuum oven it may be possible to avoid this problem.

75

160 Filtrate Filter Cake

16 14 12 10

120

80

8 6

40

4 2

0 0 20 40 60 80 Time (hrs) 100 120 140

0 160

3.3: Typical drying curve for gravimetric analysis run

0.6 Solid Analysis 0.5

(g benzoic acid/ g ethanol)

Liquid Analysis

0.4

0.3

0.2

0.1

0.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 (g w ater/ g ethanol)

Figure 3.4: Solubility of benzoic acid in ethanol water at 25 C Gravimetric Analysis

Filter Cake Mass (g)

Filtrate Mass (g)

76

3.4.2 Polythermal Method Using FBRM

An ethanol-water mixture of known composition, and a mass of about 1200g, is added to a 2-litre stainless steel jacketed vessel fitted with a Julabo chiller, to provide temperature control, and a condenser to prevent evaporation. A known mass of benzoic acid is charged to the vessel and the solution heated slowly. The slurry is monitored using FBRM and the dissolution temperature noted. At this point the solution is cooled to below the saturation temperature and another known mass of benzoic acid is charged. The slurry is reheated and the point of dissolution noted giving a new saturation temperature. In this way the solubility of benzoic acid in a given mixture of ethanol and water is calculated for a range of temperatures close to 25 C and the actual solubility at 25 C can be interpolated. Once a sufficient number of data points are gathered to allow interpolation more ethanol is added and the process repeated so the solubility for another concentration can be calculated. In this way the solubility for a full range of ethanol-water compositions at 25 C can be gathered.

Figure 3.5 is a plot of a typical run for the dissolution of the benzoic acid. Counts between 100 and 1000 microns are used to identify the point of dissolution. This is because large particles are the last to dissolve due to their high surface area to volume ratio. The point at with the counts dropped to zero was clear for all the runs making identification of the saturation temperature simple.

It is important to assess the effect of heating rate on the dissolution temperature measured by FBRM. If the heating rate is too high, dissolution may be observed at a

77

temperature above that of the actual saturation temperature. A number of runs were carried out to assess the saturation temperature measured at different heating rates (Figure 3.6). A heating rate above 0.3 C/min is too fast and the saturation temperature is overestimated. A heating rate of 0.2 C/min was chosen for all the runs since this rate garnered the same saturation temperature as for a heating rate of 0.1 /min but allows a shorter batch time.

Figure 3.7 shows the solubility of benzoic acid in four ethanol-water mixtures close to 25 C. As expected the solubility increases with increasing temperature and decreasing water concentration. By interpolating these curves the solubility at 25 C in various ethanol-water mixtures can be evaluated (Figure 3.8).

78

30 Temperature #/s (100-1000 microns)

300

250

150

20

100

50

15 180

205

230 Time (min)

255

0 280

Figure 3.5: Nucleation and dissolution of benzoic acid monitored using FBRM
23.6

23.5
Saturation Temperature (C)

23.4

23.3

23.2 0 0.1 0.2 0.3 Heating Rate (C/m in) 0.4 0.5

Figure 3.6: Heating rate vs. saturation temperature

#/s (100-1000 microns)

25

200

Temperature (C)

79

0.7 0.90 (g w ater/g ethanol) 0.6 1.37 (g w ater/g ethanol) 2.08 (g w ater/g ethanol)
g benzoic acid/g ethanol

0.5

3.37 (g w ater/g ethanol)

0.4

0.3

0.2

0.1

0.0 15 20 Tem perature (C) 25 30

Figure 3.7: Temperature vs. Solubility in various water concentrations

0.7

0.6

Solubility (g benzoic acid/ g ethanol)

0.5

0.4

0.3

0.2

0.1

0.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Water Concentration (g water/ g ethanol)

Figure 3.8: Solubility of benzoic acid in ethanol water mixtures measured using FBRM

80

3.4.3 Solubility Measurement Using ATR-FTIR Technique

3.4.3.1 Calibration of Probe

ATR-FTIR spectroscopy can be used to measure the concentration of a dissolved solute. The IR absorbance measured by the probe (Mettler Toledo, React IR 4000) is a function of concentration of the dissolved solute and a calibration is needed to convert the absorbance data into concentration. Operation at a constant temperature of 25 C eliminated any temperature effect on the absorbance.

The method of partial least squares (PLS) has been commonly used to perform such calibrations. Although the use of PLS does not produce a physical understanding of the system, it is well suited for the primary goal of prediction and estimation (Liotta & Sabesan 2004). A PLS model developed by Mettler-Toledo (Quant- IR) was used to perform the calibration. A region to two-point baseline model was used as this produced the lowest error. The spectral region used was between wave numbers 1003 and 1814 cm-1 (Figure 3.9).

40 standards were used to calibrate the ATR-FTIR probe and 12 standards were used to validate the model (Figure 3.10). Spectra for all the standards were taken in a 500ml glass jacketed vessel at a temperature of 25 C. A concerted effort was made to get standards inside the metastable zone. However the system exhibits a very narrow metastable zone width and often nucleation would occur before a measurement could be taken. This was especially the case at low water concentration where the solubility curve is steep and high supersaturation is generated for a small increase in water

81

concentration. It was relatively easy to gather standards inside the metastable zone width at higher water concentrations where the solubility curve is less steep.

1.2

0.8
Absorbance
4.0

0.6

0.4

0.2

1800

1700

1600

1500 1400 1300 Wavenum ber (cm -1)

1200

1100

0 1000

Figure 3.9 Spectral region chosen for quant

0.6 Calibration Points 0.5 Validation Points

(g benzoic acid/ g ethanol)

0.4

0.3

0.2

0.1

0.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 (g water/ g ethanol)

Figure 3.10: Standards used to calibrate and validate the ATR-FTIR probe. Weak validation points arrowed.

82

Figure 3.11 shows the comparison between the known and predicted water concentration calculated using the PLS model. Despite the excellent correlation coefficient, there is a slight offset of about 4% between the known and predicted concentration. Closer investigation of the validation set indicates that this offset is due to an unusually high error for two of the low water concentration experiments (indicated in Figure 3.10). For these validation points the error is 5.2% and 5.0%, while for the other 10 validation points the error is always less than 3%. If these values are removed from the validation set the offset is reduced to below 0.1%. It is possible that the calibration model is less accurate at these low water concentration values. Inspection of Figure 3.10 indicates that the validation solutions that produce the largest error are relatively isolated compared to the calibration set. The relationship between known and predicted benzoic acid concentration is excellent with no offset (Figure 3.12). This indicates that if there is a weakness in the calibration model at low water concentrations, the prediction of the benzoic

acid concentration is not affected. The average error was 1.8% of the measured water concentration and 1.4% of the measured benzoic acid concentration.

83

3.0 y = 0.9642x + 0.0422 R2 = 0.9979

2.5
Water Concentration Calculated (g/g)

2.0

1.5

1.0

0.5

0.0 0.0 0.5 1.0 1.5 2.0 Water Concentration Know n (g/g) 2.5 3.0

Figure 3.11: Predicted vs. Known water concentration

0.5 y = 1.0003x - 0.001 R2 = 0.9993 0.4

Benzoic Acid Concentration Calculated (g/g)

0.3

0.2

0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 Benzoic Acid Concentration Know n (g/g)

Figure 3.12: Predicted vs. Known benzoic acid concentration

84

In order to further validate the model a number of experiments were performed under different conditions and the predicted water concentration and benzoic acid concentration was compared to the measured values. Table 3.2 compares known and calculated water and benzoic acid concentrations across the three separate validation standards. In each case the predicted concentration is within an acceptable margin of error. Liotta & Sabesan (2004) noted the level of agitation can affect the measurement of spectra. To assess the effect of agitation intensity, a standard solution was measured at various levels of agitation. Table 3.2 shows the effect an increase in the agitation rate has on the water and benzoic acid concentration calculated by the calibration model. In each case the predicted concentration was acceptable. Since the crystallization under investigation is an anti-solvent addition and the total volume in the vessel will increase over time, the effect of total volume on the calibration model was assessed. Standards of the same composition but of different total volumes were made up and measured. Table 3.3 shows the difference between the known and calculated values. Once again the concentrations were within acceptable limits.

85

Table 3.1: Error across three standards for (a) water concentration and (b) benzoic acid concentration (a) Water Concentration (g/g) Standard # Known Calculated Error % 1 2 3 1.292 1.292 1.292 1.255 1.292 1.263 -2.90 0.00 -2.2 (b) Benzoic Acid Concentration (g/g) Standard # Known Calculated Error % 1 2 3 0.188 0.188 0.188 0.193 0.189 0.186 +2.55 +0.34 -0.96

Table 3.2: Effect of agitation on quant: (a) water concentration and (b) benzoic acid concentration (a) Water Concentration (g/g) Known Calculated Error % 1.292 1.292 1.292 1.292 1.277 1.271 0.00 -1.20 -1.46 (b) Benzoic Acid Concentration (g/g) Known Calculated Error % 0.188 0.188 0.188 0.189 0.187 0.187 +0.34 -0.55 -0.44

rpm 200 300 400

rpm 200 300 400

Table 3.3: Effect on volume on quant: (a) water concentration and (b) benzoic acid concentration (a) Water Concentration (g/g) Volume (ml) Known Calculated Error % 110 220 330 1.292 1.292 1.292 1.277 1.259 1.287 -1.20 -2.55 -0.40 (b) Benzoic Acid Concentration (g/g) Volume (ml) Known Calculated Error % 110 220 330 0.188 0.188 0.188 0.186 0.188 0.187 -0.55 -1.72 -0.80

Finally, the effect of a hold period on the solution was investigated. The purpose of this was two-fold. Firstly it gives an indication of the error associated between two different measurements. Secondly it indicates if evaporation plays a role. A standard solution was made up and spectra were taken every three minutes for one hour. 3.13 shows the plot of water and benzoic acid concentration against time.

86

2.5 w ater concentration - calculated w ater concentration - know n 2.0 benzoic acid concentration - calculated benzoic acid concentration - know n

0.5

0.4

1.5

0.3

1.0

0.2

0.5

0.1

0.0 0 10 20 30 Tim e (m in) 40 50 60

0.0

Figure 3.13: Effect of a hold period on the quant model

The validation experiments combined with the low average error across 12 validation samples of different concentration indicate that the calibration model is suitable for the range of concentrations and operating conditions encountered for this study.

3.4.3.2 ATR-FTIR Solubility Measurement

The method applied in this case was similar to that outlined for the gravimetric analysis but instead of weighing the mass of undissolved solute at the end of the hold period, a direct measure of the concentration was taken using the ATR-FTIR.

Benzoic acid was ground in a pestle and mortar for a number of minutes to ensure as small a particle size as possible. An excess of the milled benzoic acid was added to an ethanol-water mixture, of known composition and mass roughly equal to 150g, that had been stirring for 30 minutes in a 500ml glass lined, temperature controlled vessel. 87

benzoic acid concentration (g/g)

water concentration (g/g)

The temperature was allowed to equilibrate at 25 C and the slurry was then stirred at 300rpm for 30mins. Spectra were taken every 30s and the calibration model was applied so the benzoic acid concentration and water concentration could be calculated. At the end of the 30-minute hold period a known mass of ethanol was added to the slurry to alter the solubility and the procedure was repeated. This allowed a number of solubility measurements to be gathered over time. Figure 3.14 shows the results from one of the solubility determination experiments using ATR-FTIR. Six ethanol additions were made, allowing seven points on the solubility curve to be measured. A hold period of 30 mins was deemed sufficient since the concentration of benzoic acid in solution increased to an equilibrium level, and remained constant at that level, after as little as 10 mins (Figure 3.14). In viscous solutions and at low temperatures it can take a number of hours or even days for a system to reach equilibrium. However in this case it appears that 30mins is sufficient to allow the system to equilibrate. Unlike increasing temperature, which takes time to increase solubility, the addition of the solvent (in this case ethanol) is instantaneous which leads to a rapid increase in solubility. Additionally, the benzoic acid had been milled to a small size prior to addition. It is possible that this improved its dissolution properties. The water concentration and benzoic acid concentration at the end of each 30-minute hold period were used to generate the solubility data. In all 18 points on the solubility curve were calculated (Figure 3.15).

88

0.6 3 Water Concentration Benzoic Acid Concentration 2.5

Water Concentration (g/g)

0.5
Benzoic Acid Concentration (g/g)
4.0

0.4 2 0.3 1.5 0.2

0.5

0.1

0 0 2000 4000 6000 Tim e (s) 8000 10000

0 12000

Figure 3.14: Time vs. water concentration and benzoic acid concentration

0.6

Solubility (g benzoic acid/ g ethanol)

0.5

0.4

0.3

0.2

0.1

0.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 Water Concentration (g water/ g ethanol)

Figure 3.15: Solubility curve of benzoic acid in ethanol-water using ATR-FTIR

89

3.4.4 Overall Solubility Measurement

The solubility data gathered using the four different methods compares well (Figure 3.16). For the FBRM, ATR-FTIR and solid analysis techniques the solubilities measured similar data for the solubility. Interestingly, the solubility increases for a small increase in water concentration at high ethanol concentration. This is not unusual and solute systems have been shown to exhibit increased solubility in mixed solvents even if one of them is an anti-solvent (Granberg & Rasmuson, 2000; Hojjati & Rohani, 2006a).
0.70 0.60
g benzoic acid / g ethanol
ATR-FTIR Solubility FBRM Solubiliy

0.50
Solid Analysis

0.40 0.30 0.20 0.10 0.00 0.00

Liquid Analysis

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

Water Concentration (g water / g ethanol)

Figure 3.16. Solubility of benzoic acid in ethanol-water using four different methods

To establish the thermodynamic consistency of the measured data, the recorded solubility can be compared with the theoretical predictions of the UNIQUAC model for liquid activity coefficients (Hoijati & Rohani, 2006b).

At equilibrium, 90

fiS T, P
where
ln f iL T, P f iS T, P

xi

T, P, x i fiL T, P

Eq. 3.12

h fus Tm , P R

1 T

1 Tm

Eq. 3.13

The UNIQUAC model for prediction of the liquid activity coefficients is given by
ln
i

ln

c i

ln

r i

Eq. 3.14

where
ln
c i

ln

xi

z q i ln 2

i i

li

xi

xj lj
j

Eq. 3.15

and

ln
with
i

r i

qi 1

ln
j

j j ji j k k

ij kj

Eq. 3.16

xi qi xj qj
j

Eq. 3.17

x i ri x j rj
j

Eq. 3.18

and
li z ri 2 qi ri 1

Eq. 3.19

The biggest challenge in the use of the UNIQUAC model is the determination of the binary interaction parameters,
ij.

To establish the consistency of the measured data

reported here with the model, the binary interaction parameters are estimated by nonlinear regression of the measured data. The sum of squares of the difference between the experimental and model data was minimized by changing the binary interaction parameters using SOLVER in Microsoft Excel. The goodness of fit can then be used

91

as an indication of the consistency. A comparison of the model predictions with the experimental data is shown in Figure 3.17, using the pure component data given in Table 3.4 and the regressed binary interaction parameters reported in Table 3.5. As shown, the UNIQUAC model follows the observed trends, with the measured increase in the solubility of benzoic acid in ethanol for low water concentrations well described by the model.

Table 3.4.: Pure component data for UNIQUAC model (Yaws, 2003, Reid et al., 1987)

hfus (J/mol) benzoic acid ethanol water 16,230 -

Tm (K) 395 -

ri 4.323 2.1055 0.920

qi 3.344 1.972 1.400

Table 3.5. Estimated binary interaction parameters for benzoic acid (1) - ethanol (2) - water system.
12 13 21 23 31 32

0.359

8.774

0.000

9.214

0.000

2.465

0.7

0.6

solute conc (g/g ethanol)

0.5

0.4

0.3

0.2

0.1

0 0 0.5 1 1.5 2 2.5 3 3.5 4 water conc (g/g ethanol)

Figure 3.17. Comparison of measured FBRM solubility data with predictions of fitted UNIQUAC model.

92

3.4 DISCUSSION

Three methods for the determination of the solubility of an anti-solvent system have been presented. Gravimetric analysis provides a cheap and simple method for estimating the solubility. However this method can be time consuming with the recommended hold time being up to 24 hours. While the solid analysis provided accurate data the liquid analysis underestimated the solubility either through the saturated solution remaining in the crystallizer or an esterification reaction occurring whilst drying. FBRM provided accurate data and lends itself to automation. While a turbidity probe would measure the solubility in a similar way the FBRM will provide further information after the nucleation event and is more sensitive. ATR-FTIR provided much information about the dissolution process. It was clear that a 30minute hold time was sufficient. To gather quantitive data it is necessary to build a calibration model. While this is time consuming, once it is complete, it is possible to monitor supersaturation during crystallization by measuring the liquid phase concentration in situ.

A suitable method to express solubility for an anti-solvent system has been presented that allows for clear depiction of the metastable zone width and supersaturation in terms of simple units. The solubility of benzoic acid at 25 C has been measured for a full range of ethanol-water mixtures. The ATR-FTIR, FBRM and solid analysis methods all produced similar data while the liquid analysis systematically underestimated the solubility. The trends observed are in agreement with theoretical

93

considerations and are similar to trends reported for other ternary anti-solvent systems.

94

CHAPTER 4: THE EFFECT OF MIXING ON THE METASTABLE ZONE WIDTH IN ANTI-SOLVENT CRYSTALLIZATION

4.1 ABSTRACT

The effects of anti-solvent addition rate and location, and agitation speed on the metastable zone width (MSZW) of an anti-solvent system were investigated using Focused Beam Reflectance Measurement (FBRM) and Attenuated Total Reflectance Fourier Transform Infra-Red Spectroscopy (ATR-FTIR). Benzoic acid in ethanolwater mixtures, with water acting as the anti-solvent, was chosen as the model system and was studied at a 500 mL scale.

FBRM proved to be the more sensitive method for the detection of nucleation. In general, the MSZW widened with increasing addition rate, with the effect most pronounced when the anti-solvent was added close to the impeller. At this location, an increase in agitation intensity resulted in a narrower MSZW for all addition rates. For an addition location close to the vessel wall, the MSZW was narrower and the impact of addition rate and agitation were less pronounced. Substantial variation in the MSZW was also observed, with nucleation occasionally occurring at bulk concentrations less than the saturation level. It is proposed that the MSZW is influenced by the differing degrees of anti-solvent incorporation at each addition location. Close to the impeller anti-solvent is rapidly incorporated leading to consistent results, but, close to the vessel wall, incorporation is hindered by

95

unfavourable mixing conditions leading to premature nucleation and more variability*. Using the MSZW information nucleation kinetics at two different agitation intensities were estimated. Using this data, an agitation dependent expression for the nucleation rate was generated.

4.2 INTRODUCTION

The addition of an anti-solvent to a concentrated solution to induce crystallization of the solute through a reduction in its solubility in the combined solvent system is a power isolation and purification technique particularly where the temperature coefficient of solubility is low or where the solute is unstable at elevated temperatures. Its use in the pharmaceutical industry is, therefore common. Particular problems associated with anti-solvent crystallization include wide batch-to-batch variability, fine and irregularly shaped product crystals and solvate/hydrate formation. The effects of various process parameters, including type of anti-solvent (Takiyama et al., 1998; Oosterhof et al., 1999), feed concentration (Holmbck & Rasmuson, 1999; Barata & Serrano 1998(b)) solution concentration (Kaneko et al., 2002; Kitamura & Sugimoto, 2003), anti-solvent addition rate (Beckmann, 1999; Holmbck & Rasmuson, 1999) and agitation intensity (Takiyama et al. 1998; Yu et al., 2005), on the size, number, shape, degree of agglomeration and polymorphic form of product crystals have been reported. However, less attention has been paid to the metastable zone width and its relationship with the process conditions (Guo et al., 2005; Pina et al., 2001).
*

This proposal has been confirmed by the application of computational fluid dynamics (CFD) to characterise mixing conditions at both locations. A full treatment of the application of CFD to this system can be found in Appendix A.

96

The metastable zone width is an extremely important parameter in the design and optimisation of crystallization processes. A solute will remain in solution until a sufficiently high level of supersaturation is generated to induce spontaneous nucleation. Typically, it is desirable to operate away from this metastable limit so as to ensure reliable process performance. Metastable zone information can be used to optimise crystallization processes (Ulrich & Strege, 2002) and calculate nucleation kinetics (Nyvlt, 1968). The MSZW may be affected by various process parameters, such as supersaturation generation rate (Barrett & Glennon, 2002), agitation speed (OSullivan, 2005) and the presence of impurities (Myerson & Jang, 1995; Sayan & Ulrich, 2001).

Recent research aimed at modeling the impact of mixing on crystallization has focused on reactive crystallization (Tavare, 1995; Phillips et al., 1999; David, 2001; Torbacke & Rasmusson, 2001 & d2004). Little work has been conducted studying the impact of mixing on the MSZW for anti-solvent systems. Mixing is critical for all crystallization systems but in the case of anti-solvent addition adequate mixing is needed to incorporate the anti-solvent into the bulk solution and maintain a constant level of supersaturation throughout the crystallizer, over a potentially wide volume range. This must be achieved while considering many other mixing sensitive parameters such as solids suspension, attrition, impurity profile, agglomerate formation/break-up, of entrainment of gas/vapor from the headspace and rate of heat transfer (Paul, 2005). In this work, the effect of addition rate, agitation speed and feed point location on MSZW are investigated at a 500 mL scale. FBRM and ATR-FTIR are implemented to detect nucleation and the results are compared. Appendix A

97

highlights the use of Computational Fluid Dynamics (CFD) software is used to model the mixing behaviour of the systems under investigation and to provide an explanation of the observed results.

With accurate MSZW information it is possible to estimate nucleation kinetics. This is achieved my modifying classical nucleation theory (Nyvlt, 1968) for an anti-solvent system. Much of the literature focuses on static induction time experiments to estimate nucleation kinetics for an anti-solvent system (Barata & Serrano, 1996(b)). However useful nucleation kinetics can be gathered using dynamic experiments and this is the method will be used here.

4.3. EXPERIMENTAL METHODS

The solute in this work, benzoic acid, is soluble in ethanol (58.36 g/100g @ 25 C) and essentially insoluble in water (0.25g/100g @ 25 C) (Perry). All metastable zone width measurements were carried out at 25 C.

An initially undersaturated solution, containing 75g water, 75g ethanol and 21g benzoic acid, was held at 25 C in a 500 mL glass jacketed vessel with a Julabo chiller fitted for temperature control. A motor driven, pitched blade impeller provided agitation. FBRM (S400 A; Mettler Toledo), ATR-FTIR (ReactIR 4000; Mettler Toledo) and temperature probes were placed in the vessel to monitor the crystallization and as a consequence provided baffling (Figure 4.1). Water at 25 C was fed to the vessel at a variety of rates (0.05 gs-1, 0.14 gs-1, 0.24 gs-1, 0.34 gs-1 and 98

0.48 gs-1). Two addition locations (Figure 4.1) and two agitation intensities (325 rpm and 475 rpm) were investigated. All experiments were performed in triplicate and the reported values are the mean of the three experiments. The error bars reported for the standard deviation of the mean. FBRM and ATR-FTIR were used to detect nucleation and the MSZW is reported in terms of concentration (g anti-solvent / g solvent) as outlined in Chapter 2. Solubility data for the system can also be found in Chapter 2.

Figure 4.1: Crystallizer configuration

99

4.4 RESULTS AND DISCUSSION

4.4.1 Comparison of FBRM and ATR-FTIR for the detection of nucleation

FBRM and ATR-FTIR detect the formation of crystals in different ways. FBRM detects the point at which crystals are present in the solution, whereas ATR-FTIR detects the point at which the solution concentration decreases, indicating solute is being forced out of solution and crystals are being formed. To measure the actual solution concentration a calibration model is needed, but for the purpose of identifying the point of nucleation such a model is not necessary, since it is sufficient to examine peaks in the spectrum associated with benzoic acid.

Benzoic Acid Peak

Figure 4.2: ATR-FTIR waterfall plot for standard MSZW experiment

Figure 4.2 shows a waterfall plot for a typical crystallization. The peak at 1275 cm -1 represents the C-O bond present in the benzoic acid molecule. The peak at 1480 cm-1 100

is representative of a C-C bond indicating an aromatic group. After a 10-minute hold period to equilibrate the temperature, water is added, in this case, at 0.045 g/s (and agitation speed of 475 rpm). Examining the peak at 1275 cm-1 the concentration of benzoic acid in solution decreases at a steady rate due to dilution. After about 5 minutes there is a sharp decrease in this peak, indicating that the concentration of benzoic acid in solution is decreasing at a faster rate, due to the formation of benzoic acid crystals. In this way the onset of nucleation can be identified using ATR-FTIR with no need for a calibration model.

The height of a specific peak can be trended against time so the onset of nucleation can be identified easily. Figure 4.3 (a) shows the comparison between FBRM counts (10-100 m), and the height of the benzoic acid peak at 1275 cm-1. It was noted that the counts per second data in the region 0-10 m exhibited more noise than the counts data in the 10-100 m region. Therefore, the count rate in the latter region was used for nucleation detection. The chosen range also ensured that agglomerates of newly formed crystals were also detected. In general, however, it is advisable to choose the most sensitive detection range by inspection. The FBRM indicates clearly the point at which the first crystals are observed while the decrease in intensity of the benzoic acid peak indicates the point at which solute comes out of solution. Figure 4.3(b) focuses on the region where nucleation occurs and shows that FBRM detects the onset of nucleation before the ATR-FTIR. For all addition rates the FBRM detected the point of nucleation before the ATR-FTIR. (Figure 3.4) with the difference most pronounced at higher addition rates. FBRM has an advantage over ATR-FTIR in detecting the point of nucleation due to its much shorter sampling time of 2 seconds rather than 30 seconds. It is possible to scan faster with the ATR-FTIR probe, but this increases the

101

likelihood of errors in the measurement. For this reason FBRM was chosen to indicate the point of nucleation.

3500

0.12

3000

0.1

2500

#/s 10-100 microns

0.08 2000 FBRM 1500 ATR-FTIR 0.04 1000 0.02 0.06

500

0 0 5 10 TIme (min) 15 20 25

Figure 4.3 (a): Time vs. FBRM counts and peak height
20

0.1

FBRM ATR-FTIR
#/s 10-100 microns
15

0.08

Relative Abs

0.06
ATR-FTIR Nucleation FBRM Nucleation

10

0.04

5 12 13 14 15 16 17 18

0.02 TIme (min)

Figure 4.3 (b): Time vs. FBRM counts and peak height: Point of nucleation

Relative Abs

102

0.4

ATR-FTIR FBRM

0.3
MSZW (g/g)

0.2

0.1

0 0 0.1 Addition Rate (g/s) 0.2 0.3

Figure 4.4: Comparison of MSZW measured by FBRM and ATR-FTIR

103

4.4.2 Impact of Process Parameters on the MSZW

Figure 4.5 shows the impact of addition rate and agitation intensity on the MSZW when anti-solvent is added close to the impeller (addition point 1 in Figure 3.1). At a given agitation intensity the MSZW becomes wider as addition rate increases. The increase in MSZW with increasing supersaturation generation rate is typical of most crystallization systems.

For each addition rate studied an increase in the agitation intensity resulted in a natrower MSZW. Agitation can have two effects on the metastable zone width. Increased agitation can narrow the MSZW by increasing the probability of solute molecules contacting to form the critical sized nuclei necessary for nucleation. However this increase in agitation can also break up these clusters widening the MSZW. For an anti-solvent system there is the added factor of mixing between the solvent and anti-solvent.

104

0.4 475 rpm - impeller addition location 325 rpm impeller addition location

0.3

MSZW (g water/g ethanol)

0.2

0.1

0.0 0 0.1 0.2 0.3 0.4 0.5

Addition Rate (g water/s )

Figure 4.5: Addition Rate vs. MSZW

addition location close to the impeller

Similar experiments were performed with an addition point close to the wall of the vessel (addition point 2 in Figure 4.1). In this case the addition rate had substantially less effect on the MSZW (Figure 4.7). In general the MSZW is significantly narrower when the anti-solvent is added close to the wall. This can be explained by considering the unfavourable mixing conditions at the wall. Addition close to the impeller ensures rapid incorporation of the anti-solvent into the bulk solution, whereas addition close to the wall leads to areas of locally high supersaturation close to the addition location. There was also a far greater variability in the MSZW, especially at 325 rpm. At high addition rates it becomes harder to incorporate the anti-solvent into the bulk solution. This leads to an area of local supersaturation close to the feed point. This can result in premature nucleation often at bulk concentrations below the saturation level. Increasing the agitation facilitates dissipation of the local supersaturation. This

105

explains why increasing the agitation intensity results in a wider MSZW as well as reduced variability.
0.4 475 rpm - wall addition location 325 rpm - wall addition location 0.3

MSZW (g water/g ethanol)

0.2

0.1

0.0 0 0.1 0.2 0.3 0.4 0.5

-0.1 Addition Rate (g water/s )

Figure 4.7: Addition Rate vs. MSZW addition location close to the vessel wall

4.4.3 Nucleation Kinetics

With sufficient metastable zone width data it is possible to calculate nucleation kinetics for the anti-solvent system. It is important to identify an operating regime where kinetics may be reliably calculated. In the situation where anti-solvent was added at the wall location, and at 325 rpm, it is probable that true nucleation kinetics cannot be estimated. This is because the observed nucleation occurs due to excessive local supersaturation near the addition point rather than true supersaturation in the bulk of the liquid. When the addition location is close to the impeller, the nucleationdoes not appear to be so dominated by hydrodynamic conditions. In this case useful kinetic information may be estimated. 106

Classical nucleation theory (Nyvlt, 1968) can be applied to the MSZW data as used with modifications for anti-solvent addition rather than cooling.

dN dt

ko C n

4.1

During anti-solvent addition, the rate of supersaturation generation can be expressed as a function of the specific anti-solvent addition rate,

d c dt

R.

dcs dA

4.2

At the point of nucleation, the supersaturation is related to the metastable zone width by

C max

Amax .

dcs dA

4.3

where the MSZW is given by

Amax

As

Anuc

4.4

Using the solubility data reported in Chapter 3 dcs/dA is equal to 0.2231 g solute g-1 solvent g-1 anti-solvent for the benzoic acid-ethanol-water system. At the point of

107

nucleation, it may be assumed that the rate of supersaturation generation equals the rate of formation of new crystals which is given by

dM dt

ko

3 rnuc Cn

kn C n

4.5

Combining equations 4.2, 4.3 and 4.4 yields,

ln R

n 1 ln

dcs dA

ln k n

n ln Amax

3.7

Using equation 4.7 a plot of ln (R) verses ln (Amax), will yield a line of slope n. (Figure 4.8). Fitting a simple trend line to the data allows n and kn to be calculated.

Figure 4.8: ln (MSZW) vs. ln (addition rate) with trend-lines

108

Figure 4.8 clearly indicates that agitation has an impact on the nucleation kinetics. In order to study the effect of agitation on the kinetics it is proposed that the nucleation rate expression can be modified to include an expression for the agitation intensity (),

kn N

kn

dcs dA

Amax

dcs dA

4.8

Taking logarithms again,

ln R

n 1 ln

dcs dA

ln k n

n ln Amax

ln( N )

4.9

In this case a plot of ln(R) vs. ln( Amax ) will once again yield a straight line. In order to solve for n, kn and a non-linear regression was applied to the data using the known values of
dc * , N and dA

Amax . The regression analysis yielded the following

expression for the nucleation rate (Equation 4.10).

2.78 N 1.1 C 2.5

4.10

From equation 4.9, the metastable zone width can be obtained thus

109

Amax

R ' kn N

1 n

dcs dA

1 1 n

4.11

Good agreement is obtained between the predictions of equation 4.11 and the experimental data (see Figure 4.9), supporting the contention that impeller speed plays a critical role in determining the level of nucleation in the vessel, and thus providing a correlation to facilitate quantification of the role of impeller speed in the determination of nucleation rates. This is consistent with the observations made based on the measured trends in metastable zone width.

Figure 4.9: Metastable zone widths predicted form Equation 4.11

110

4.5 DISCUSSION

The effect of addition rate and location, along with agitation intensity, on the metastable zone width has been investigated for an anti-solvent crystallization. An addition location close to the impeller results in a repeatable crystallization and a positive correlation between the addition rate and the MSZW. An addition location close to the wall of the vessel results in a significantly narrower MSZW, and more random nucleation behaviour.

The effect of agitation intensity depends on the addition location. When anti-solvent is added close to the impeller, an increase in agitation intensity results in a narrower MSZW, possibly due to the increased probability of contact between solute molecules. When anti-solvent is added close to the wall, an increase in agitation results in a wider MSZW and a significant improvement in the batch-to-batch repeatability. These results can be explained in terms of mixing conditions at each of the addition locations. Close to the impeller, mixing conditions allow for the rapid incorporation of the anti-solvent and a homogenous mixture of solution and antisolvent. However, close to the wall, mixing conditions are less suitable and areas of supersaturation build up leading to narrower MSZWs and a reduction in the batch-tobatch repeatability. In this situation, when the agitation is increased, the local areas of supersaturation can be dissipated, to some degree, and the MSZW is wider and the batch-to-batch repeatability improves.

By modifying classical nucleation theory for an anti-solvent system, it was possible to estimate the nucleation order and the nucleation constant at both agitation intensities,

111

and for the addition location close to the impeller. Only this location was studied as addition close to the wall resulted in unrepeatable experiments and nucleation was judged to be a result of local areas of supersaturation close to the addition location rather than as a result of supersaturation in the bulk solution. Agitation clearly impacted on the nucleation kinetics and to account for this the nucleation rate expression was modified, to account for agitation, and solved using a non-linear regression. The modified equation fitted the experimental data adequately.

112

CHAPTER 5: THE USE OF FBRM AND ATR-FTIR TO MONITOR ANTI-SOLVENT CRYSTALLIZATION AND ESTIMATE GROWTH RATE KINETICS

5.1 ABSTRACT

The unseeded, anti-solvent crystallization of benzoic acid, from ethanol-water mixtures using water as an anti-solvent, is presented. Focussed Beam Reflectance Measurement (FBRM) and Attenuated Total Reflectance Fourier Transform Infrared spectroscopy (ATR-FTIR) were used to monitor the solid and liquid phase respectively. FBRM was used to track crystal growth and ATR-FTIR was used to measure the water concentration and the benzoic acid concentration. In combination with previously gathered solubility data (Chapter 3) the measured concentration data was used to monitor the supersaturation. Using previously gathered MSZW information (Chapter 4), optimal mixing conditions were chosen to ensure a repeatable crystallization with a low nucleation rate. Growth rate kinetics were estimated by combining suitable statistics, from the FBRM chord length distribution, with the supersaturation data.

5.2 INTRODUCTION

In recent years the use of in situ tools for the study of crystallization systems has gained much attention. This has been driven by the need to rapidly gather representative, real time data, without sampling. Sampling can be time and labour intensive, unrepresentative, and may require the sample crystals to be prepared in

113

some way, typically through sonication or dilution. Also, the FDAs Process Analytical Technology (PAT) initiative (FDA website) encourages the use of in situ tools to monitor processes such as crystallization.

FBRM and ATR-FTIR have emerged as excellent tools for the study of crystallization with much work in the literature devoted to their application. FBRM is a probe-based instrument that measures a chord length distribution of the crystal slurry, which is a function of the dimension, population and shape of the crystals in the solution. It is extremely sensitive to fine crystals and can measures tens of thousands of crystals per second allowing a very robust real-time measurement to be gathered. FBRM has, amongst other things, been used to measure solubility and the MSZW (Barrett & Glennon 2002), study crystal size (Kougoulos et al., 2005), and assess the effect of impurities on batch time (Scott and Black, 2005).

ATR-FTIR is another probe-based tool that allows the liquid phase concentration to be measured once a calibration model has been created. Importantly, for the antisolvent system the ATR-FTIR probe offers the opportunity to measure both the solute and anti-solvent concentration. In combination with accurate solubility data it is then possible to monitor supersaturation. ATR-FTIR has been used to optimise crystallization processes (Lewiner et al., 2001) and monitor and control supersaturation (Liotta & Sabesan 2004).

The mode of operation of both of these instruments is outlined in detail elsewhere (Sparks & Dobbs 1994; Dunwilla & Berglund 1997) and a detailed review of each of these instruments can be found in Chapter 2.

114

By combining supersaturation data, gathered using ATR-FTIR, with growth rate data, gathered using FBRM, it is possible to estimate growth rate kinetics according to Equation 5.1 (Mullin 1993).

G = kgCg

5.1

A plot of supersaturation against growth rate allows the growth coefficient and the growth order to be estimated. This information is useful for the estimation of batch time and calculation of suitable anti-solvent addition rates. By assessing the growth rate and supersaturation under different mixing conditions in the laboratory the impact of scale on the growth kinetics can be identified.

5.3 MATERIALS AND METHODS

A solution consisting of 75 g of water 75 g of ethanol and 21 g of benzoic acid was made up and placed in the 500 mL glass reactor. FBRM, ATR-FTIR and temperature probes provided baffling and a pitched blade impeller provided agitation. The temperature was allowed to equilibrate at 25 C and the solution was held for 20 minutes to ensure all the benzoic acid had dissolved. Previous work identified that a slow addition rate close to the impeller, with high agitation produced the most robust crystallization in terms of consistent metastable zone width and low nucleation rate (Chapter 4). With this in mind, water was added close to the impeller at a rate of 0.065 g/s for 45 mins, and the agitation was set at 475 rpm. The contents of the vessel were held for a further 15mins after the end of the addition period to ensure the supersaturation was consumed and the crystallization complete. Three similar 115

experiments were performed (Run 1, Run 2, Run 3). In the case of Run 3 the addition period was 40 mins and the hold period was 5 mins.

5.4 RESULTS AND DISCUSSION

5.4.1 FBRM Results

A comparison of the three runs was made in order to assess the repeatability of the crystallization under the chosen process parameters. One useful aspect of the FBRM measurement technique is its ability not only to measure the degree of change of the size, number and shape of the crystals, but to also measure their rate of change. Figure 5.1 shows three FBRM statistics plotted against time, for each run.

Figure 5.1 focuses on #/s (1-10 microns) and is useful for tracking nucleation, as it highlights the shortest chord lengths. It is clear that the nucleation rate for each run is similar and that there is no secondary nucleation event apparent, a phenomenon typically characterised by a sudden increase in fine chords after the primary nucleation event.

Figure 5.2 focuses on the larger crystals and tracks #/s (100-1000 microns) and highlights the longer chord lengths, making it useful for tracking growth. The initial increase in this statistic occurs after the increase in #/s (1-10 microns) indicating that this statistic is less sensitive to the primary nucleation event, and that it takes some time for the crystals to grow into larger size ranges.

116

The initial rate of increase in #/s (100-1000 microns) is rapid, when supersaturation is high, and gradually plateaus as supersaturation is consumed. A similar result is observed for the mean square weighted chord length (MSQW) (Figure 5.3). A square weighting emphasises the long chords and de-emphasises the fine chords. This makes the MSQW a very useful statistic for tracking crystal growth. Prior to the nucleation event there is significant noise in the MSQW measurement. Even though count data are low the MSQW is influenced by small changes in the chord length distribution.

1400 run 1 1200 run 2 run 3

1000
#/s (1-10 microns)

800

600

400

200

0 0 600 1200 1800 Time (s) 2400 3000 3600

Figure 5.1: FBRM data for three similar experiments: (a) #/s 1-10 microns

600 500
a
)

Run 1

117

600 run 1 500 run 2 run 3

#/s (100-1000 microns)

400

300

200

100

0 0 600 1200 1800 Time (s) 2400 3000 3600

Figure 5.2: FBRM data for three similar experiments: #/s 100-1000 microns

400 run 1

Square Weighted Mean Chord Length (microns)

350

600 500
a
) ro n s ic 0 m 0 0 -1 0 0 / s (1 #

run 2 run 3

300

250

200

150

100

50

400
1800 Time (s)

Run 1
2400 3000 3600

0 0 600 1200

Figure 5.3: FBRM data for three similar experiments: square weighted mean chord length (microns)

300 200 600

Run 2

118

The chord length distribution itself can also be used to assess repeatability and is very useful for identifying crystallization mechanisms. Figure 5.4 shows the (a) unweighted and (b) square weighted chord length distribution for each run at three different points in time. Immediately it is clear that the chord length distribution is similar for each run at each time point.

Figure 5.4 also shows that the crystallization is growth dominated. After 30 minutes there is little change in the unweighted distribution, but a large increase in the square weighted distribution. This indicates that after the primary nucleation event, growth dominates the crystallization. Slight differences in the trends and chord length distributions between the first two batches and the third batch may be due to the small difference in operating conditions.

It is sometimes difficult, based on FBRM data alone, to discern between crystal growth and agglomeration. Typically FBRM measures agglomeration as a reduction in fine chords and a corresponding increase in large chords as small crystals come together to form large agglomerates. However, if growth and agglomeration are occurring simultaneously this may not be the case and fine counts may actually increase as the likelihood of measuring a short chord increases as the crystals get larger. PVM images taken during a similar crystallization* show that the benzoic acid forms large well formed elongated platelets. It also appears that the crystals come together to form loosely bound agglomerates or flocs (Figure 5.5) PVM is advantageous for this analysis as taking samples and analyzing the crystals under a

It was not possible to insert the FBRM, ATR-FTIR and PVM probes into the crystallizer at the same time. PVM images were taken from an experiment performed under the same conditions but without the FBRM and ATR-FTIR probes present. It is assumed that the different mixing conditions encountered do not affect the crystallization appreciably.

119

microscope can lead to misleading information as a result of agglomeration during filtration or drying, or simply because crystals lie on top of each other and resemble agglomerates under the microscope.

120

140 run1 t = 20 mins run2 t = 20mins 100 run3 t = 20 mins run1 t = 30 mins 80 run2 t = 30 mins run3 t = 30 mins 60 run1 t = 50 mins run2 t = 50 mins 40 run 3 t = 50 mins

120

#/s
20 0 1 10 100 Chord Length (microns)

1000

10000

14

run1 t = 20 mins run2 t = 20mins

12

run3 t = 20 mins run1 t = 30 mins run2 t = 30 mins

10

run3 t = 30 mins run1 t = 50 mins

#/s
6

run2 t = 50 mins run 3 t = 50 mins

0 1 10 100 1000

Chord Length (microns)

Figure 5.4: Chord length distributions at three time stamps for each run (a) unweighted; (b) square weighted

121

Figure 5.5: PVM images taken after 30 minutes of crystallization

122

5.4.2 ATR-FTIR Results

Figure 5.6(a) shows water concentration and benzoic acid concentration, for the three runs, gathered using ATR-FTIR. The linear increase in the mass ratio of water is evident, as is the end of the addition period at 2700s for runs 1 and 2 and 2400s for run 3. Prior to nucleation the benzoic acid concentration remains essentially constant, with a slight decrease apparent. This subtle change is most likely due to a weakness in the calibration model in the region close to the starting point of the crystallization. It may also be a result of variations in temperature due to a heat of mixing. The point at which the concentration decreases dramatically represents a point on the MSZW and coincides with the increase in #/s (1-10 microns) (Figure 5.1). In general FBRM is more sensitive to the nucleation event than ATR-FTIR (Chapter 4).

Supersaturation is calculated by combining accurate solubility data (Chapter 3) with the measured concentrations (Figure 5.46(a)) and is shown in Figure 5.4 (b). The point of nucleation is observed when the supersaturation drops and it is clear that all of the supersaturation is consumed soon after the end of the addition period (2700s). This indicates that a short hold period is sufficient to ensure the crystallization is complete and the yield is maximised.

123

2.5

0.4

2.0

0.3

1.5

0.2 1.0 Water Run 1 Benzoic Run 1 0.5 0 600 1200 1800 Time(s) 2400 3000 Water Run 2 Benzoic Run 2 Water Run 3 Benzoic Run 3 0.1 3600

0.06

0.03

Supersaturation (g/g)

0.00 0 600 1200 1800 2400 3000 3600

-0.03

-0.06 run 1 -0.09 run 2 run 3 -0.12 Time (s)

Figure 5.6: ATR-FTIR data for three similar runs: (a) Water concentration and benzoic acid concentration; (b) Supersaturation

Benzoic Acid Concentration (g/g)

Water Concentration (g/g)

124

5.4.3 Growth Rate Kinetics Estimation

Section 5.4.1 shows that nucleation dominates only in the early stages of the crystallization and that growth with some agglomeration is dominant for the remainder. Growth rate kinetics are estimated once the primary nucleation event is complete. This exact point is open to interpretation, however, close examination of Figure 5.1 shows that the initial rapid increase in #/s (1-10 microns) is probably over after about 1300 s. With the region of interest identified, it is necessary to choose a statistic that will suitably track the growth. As outlined in section 4.4.1 #/s (100-1000 microns) and the MSQW are track the growth phase of the crystallization in a similar fashion. For the purpose of this study the MSQW will be used to estimate the growth rate kinetics as it is measured in units of dimension rather than population-per-second facilitating interpretation.

The measurement of accurate growth rate kinetics in this case is difficult due to the initial uncontrolled nucleation event and the subsequent loose agglomeration outlined in the previous section. While pure crystal growth is not apparent in this system the method described will serve to highlight the possibility of rapidly measuring growth rates in-process and in real-time. For a seeded crystallization with no agglomeration the accuracy of this method for estimating the growth kinetics maybe improved.

The growth rate is measured by applying the first derivative to the MSQW in the FBRM software according to Eq. 5.2 and is plotted in Figure 5.7.

' vk

vk tk

vk 1 tk 1

Eq. 5.2

125

0.3 run 1 0.25 0.2

MSQW (microns)

run 2 run 3

0.15 0.1 0.05 0 0 -0.05 -0.1 Time (s) 500 1000 1500 2000 2500 3000 3500 4000

Figure 5.7: Growth rate data gathered using FBRM data after 1300s: square weighted mean chord length

The FBRM and ATR-FTIR data were combined, according to Equation 5.1, to measure growth rate kinetics. By plotting supersaturation (Figure 5.4(b)) against growth rate (Figure 5.7) the kinetic parameters g and kg can be estimated. Figure 5.8 shows that, as expected, high growth rates occur at high supersaturation. A non-linear least squares fit was applied to the data and is also shown in Figure 5.8. In this case the growth order is close to unity.

126

0.3 Model 0.25

Growth Rate (microns/s)

G = 3.61C0.91

Run 1 Run 2

0.2

Run 3

0.15

0.1

0.05

0 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035

Supersaturation (g./g)

Figure 5.7: Growth rate kinetics using three experiments

To eliminate scatter from the data a second approach was taken where the growth rate and supersaturation data was averaged over the three experiments (Figure 5.7) and then combined. Figure 5.9 shows a plot of the averaged supersaturation against the averaged growth rate. Clearly this approach results in less scatter to the data. Since the experiments were repeatable this approach was deemed suitable for calculating the average growth rate over a number of crystallizations. The calculated value of 1.1 for the growth order is in the acceptable range for organic crystallization processes.

127

0.04

a
0.02

0.00 0
Supersaturation (g/g)

600

1200

1800

2400

3000

3600

-0.02

-0.04

-0.06

-0.08

-0.10

-0.12 Time (s)

b
0.2

Avergaed Growth Rate - SQWM (microns/s)

0.15

0.1

0.05

0 0 600 1200 1800 2400 3000 3600

-0.05

-0.1 Time (s)

Figure 5.8: (a) Supersaturation average; (b) FBRM growth rate average

128

0.25

Model

G = 7.8C1.1

Growth Rate (MSQW - microns/s)

0.2

Experimental

0.15

0.1

0.05

0 0 0.005 0.01 0.015 0.02 0.025 0.03

Supersaturation (g/g)

Figure 5.9: Growth rate kinetics estimated using averaged values

129

5.5 CONCLUSIONS

FBRM and ATR-FTIR have been successfully used to monitor the anti-solvent crystallization of benzoic acid from ethanol water mixtures using water as the antisolvent. FBRM showed that the nucleation rate for each experiment was similar and that under the chosen process conditions the crystallization was robust and repeatable. ATR-FTIR successfully tracked the water concentration and the benzoic acid concentration, and in conjunction with previously gathered solubility data the supersaturation was monitored.

Growth rate data were calculated by combining the supersaturation data and growth rate data, calculated using the square weighted mean from the FBRM data. A growth rate order of 1.1 was calculated, a suitable number for an organic crystallization system.

The accuracy of the measured kinetics is questionable due to the uncontrolled nucleation events prior to growth and the agglomeration during the growth phase; however this technique highlights the ease with which growth rate kinetics may be estimated in-process. This also opens the opportunity to measure the growth rate kinetics in real time by combining the FBRM and ATR-FTIR data as the crystallization is being performed. By choosing a system where agglomeration is minimal and nucleation can be minimised, and by using seed, the accuracy of the growth rate kinetics may be improved.

130

CHAPTER 6: SCALE-UP OF ANTI-SOLVENT CRYSTALLIZATION

6.1 ABSTRACT

The pilot-scale crystallization of benzoic acid, from ethanol-water mixtures, using water as the anti-solvent is presented. With extensive small-scale characterisation performed it was possible to increase the scale of the crystallization from the laboratory scale (500 mL) to the pilot plant scale (70 L) with a view to assessing crystallization behaviour on scale-up. Prior to the experimental work the pilot scale vessel was modelled using computational fluid dynamics (CFD) in order to assess the mixing regime.

A scale-up strategy based on the laboratory scale experiments was implemented and three batches were run at the pilot-scale under various operating conditions to assess which parameters most closely met the requirements of producing crystals of a similar size. CFD was used to choose a suitable anti-solvent addition point. Each run was monitored in-line using Focussed Beam Reflectance Measurement (FBRM) and Particle Vision and Measurement (PVM).

Similar chord length distributions were achieved at the laboratory-scale and the pilotscale indicating that the operating conditions chosen were suitable in terms of conserving the particle size on scale-up.

131

6.2 INTRODUCTION

While extensive characterisation work is possible at the laboratory scale there is almost always difficulty associated with reproducing similar results at a large-scale. This is most often due to differences in mixing conditions between scales. An intermediate, pilot-plant scale is often used to try and predict how a crystallization may behave upon scale-up without the expense, time and labour associated with scaling directly to the plant. With adequate characterisation at the small scale a small number of experiments can be performed at the pilot scale giving valuable information regarding potential plant scale issues. In the pharmaceutical industry this is often performed in tandem with the production of material for use in clinical trials.

In order to assess the impact of scale on the crystallization, key parameters must be measured. In situ tools provide the opportunity to do this without the need for sampling and can continuously monitor in real time the progression of the crystallization. Critically they can be implemented at a variety of scales allowing a direct comparison to be made between experiments conducted in the laboratory, pilot plant and in production.

In this study, data gathered at the small scale along with CFD models were used to choose suitable process conditions at the pilot scale to produce a robust crystallization with a particle size similar to that produced at the laboratory scale. The laboratory studies indicated that agitation intensity, addition location and addition rate were the critical parameters that needed to be controlled. At this scale a robust and repeatable crystallization was achieved when the agitation intensity was high, the addition rate

132

was low and the addition location was optimized to ensure rapid incorporation of the anti-solvent into the bulk solution. With this in mind a scale-up protocol was devised. The addition rate was chosen to ensure a similar batch time at the lab and pilot scales. The agitation intensity was chosen as the maximum that would not result in entrainment of bubbles into the system. The addition location was chosen based on CFD results for the large scale vessel .these indicated that the optimal addition location may change during the batch so experiments were designed to take advantage of that.

6.3 MATERIALS AND METHODS

For the laboratory-scale experiment 75g of ethanol, 75g of water and 21g of benzoic acid were added to a 500 mL glass vessel, fitted with a pitched blade impeller set at 475 rpm, and held at 25C. Water was added at 0.065 g/s above the liquid surface close the impeller for 45 mins. This was followed by a 15 minute hold period. These experimental conditions resulted in a repeatable crystallization based on the FBRM trends (Chapter 5).

The same solution concentration was employed at the pilot-scale where 12.5 kg of ethanol, 12.5 kg of water and 3.5 kg of benzoic acid were added to a 70 L glass stirred tank reactor fitted with a Rushton turbine and four baffles. Figure 6.1 shows the vessel dimensions and the position of the probes and internals. The contents were stirred at 130 rpm for 30 mins to ensure complete dissolution of the benzoic acid. This

agitation intensity was the highest possible without entraining air into the system. Water was then added at 10 gs-1 for 45 mins in order to achieve the same batch time

133

as the experiments conducted at the laboratory-scale. At the end of the addition period the solution was held for a further 15 minutes to ensure complete crystallization.

Three pilot-scale batches were run under different mixing conditions and addition locations. For batch A the agitation was set at 130 rpm, the highest agitation intensity that did not result in a vortex, and the anti-solvent was added at point 2 (Figure 6.1). This addition location was chosen based on CFD simulations showing the highest downward velocity at this point (Appendix A). For batch B the same addition location was chosen (point 2) but in this case the agitation was increased from 157 rpm to 250 rpm so as to maintain the same power per unit volume for the duration of the batch (Figure 6.2). The starting agitation of 157rpm was calculated based on a constant power per unit scale up from the laboratory vessel. At the start of the batch vortexing and air entrainment were noticeable but dissipated as the volume of the batch increased. For batch C the same agitation profile as that used for batch B was implemented but the addition point was changed five minutes into the experiment from point 1 to point 2. This decision was based on CFD simulations indicating the optimal addition location changed as the volume in the vessel increased (Appendix A). Table 6.1 summarizes the operating conditions used for the three pilot-scale batches.

134

Figure 6.1: Pilot-scale vessel and internals; (1) Addition Location 1; (2) Addition Location 2; (3) PVM probe; (4) FBRM probe; (5) Overflow
600 400 350 300 400

500

P/V (W/m )

250 200 150 100

300

200
P/V Rpm

100

50 0

0 0 10 20 30 40 Time (mins) 50 60

Figure 6.2: Time vs. agitation intensity and power-per-unit-volume for batches B and C 135

Agitation Intensity (rpm)

Table 6.1: Summary of Pilot-Scale Batches Addition Rate Batch A Batch B Batch C 10gs-1 10gs-1 10gs-1 Agitation Intensity 130 rpm (157 rpm 200rpm) (157 rpm 200rpm) Addition Location 2 2 Points 2 & 1

6.3 RESULTS AND DISCUSSION

6.3.1 FBRM Results

Figure 6.3 shows the chord length distributions measured at the end of batches A, B and C. Clearly batches B and C, with constant power-per-unit-volume, are very similar at the fine (unweighted) and coarse (square weighted) ends of the distribution. The change in addition location implemented in Batch C appears to have had no effect.

For Batch A, a slight increase in fine and coarse particles compared to the other two batches is observed. In this case it is difficult to ascertain whether this is a function of the crystallization mechanism or a function of the presentation of the crystals to the FBRM window. At 130 rpm the mixing was visibly inadequate to fully suspend the crystal slurry and there were a number of dead zones at the top, bottom and sides of the reactor. Evidence of this segregation is shown in Figure 6.4 where samples taken from the top and bottom of the pilot-scale vessel are compared off-line using FBRM. The chord length distributions show significantly more fine material for the sample

136

taken from the top and significantly more coarse material for the sample taken at the bottom. Smaller crystals floated to the top of the vessel and were not re-incorporated into the bulk of the slurry while larger crystals sank to the bottom and were not resuspended. Thus, for batch A, the chord length distribution measured by the FBRM in a turbulent zone close to the impeller may not be representative of the crystallization as a whole.

120 Batch A Batch B Batch C

100

80

#/s

60

40

20

0 1 10 100 1000

Chord Length (microns)

137

18 16 14 12 10 8 6 4 2 0 1 10 100 1000
Batch A Batch B Batch C

#/s

Chord Length (microns)

Figure 6.3: Chord Length Distributions for Batches A, B and C; (a) unweighted, (b) square weighted
3000 Bottom Sample 2500 Top Sample

2000

#/s

1500

1000

500

0 1 10 100 1000

Chord Length (microns)

138

6 Bottom Sample 5 Top Sample

#/s

0 1 10 100 1000

Chord Length (microns)

Figure 6.4: Offline FBRM measurements taken from samples from the top and bottom of the pilot plant vessel at the end of Batch A

The large degree of segregation observed during batch A indicates that agitating at 130 rpm for the duration of the batch is not adequate to ensure a robust and repeatable crystallization. Batches B and C during which the agitation was increased to maintain a constant power per unit volume resulted in much better mixing and no segregation. Additionally these batches produced an almost identical chord length distribution indicating repeatability, assuming the impact of changing the addition location (batch C) was negligible.

In order to assess the effectiveness of the scale-up in terms of the crystal size predicted, chord length distributions at the end of batch B are compared with the

139

standard laboratory experiment. In order to effectively compare the chord length distributions the FBRM data are normalized to allow for the differences in scale.

Figure 6.5 indicates that in general the chord length distribution measured at each scale is very similar. There are small differences at the fine end of the distribution where there is slightly more fine material produced at the pilot scale. At the coarse end of the distribution the distributions are essentially identical with the lab laboratory scale producing slightly larger material. There is an increase in the number of fine crystals for the pilot-scale crystallization compared to the laboratory-scale experiment. Additionally the squared weighted chord length distribution indicates the size of the largest crystals is essentially identical with the laboratory-scale crystallization producing slightly larger crystals. Statistics from each distribution are compared in Table 6.2 and indicate the similarity between the crystallization conducted at both scales.

140

3 Standard Laboratory 2.5 Batch B

#/s

1.5

0.5

0 1 10 100 1000

Chord Length (microns)

5 Standard Laboratory 4 Batch B

#/s
2 1 0 1 10 100 1000

Chord Length (microns)

Figure 6.5: Comparison of standard laboratory experiment with batch B (a) unweighted and (b) square weighted

141

Table 6.3: Summary of Chord Length Distribution Statistics

Median (m) Standard 20.53 Laboratory Batch B 19.54 41.12 208.9 15.5 73.4 11.1 43.13 222.52 11.9 76.2 11..9 Mean (m) MSQW (m) % counts (1-5 m) % counts (5-100 microns) % counts (100-1000 microns)

The data shown in Table 6.3 is complemented by the PVM images taken for each crystallization, show in Figure 6.6. The images indicate the similarity in crystal size for each batch. They also help validate the FBRM measurement and indicate that the mean square weight is probably a good indicator of the length of the crystals while the unweighted mean is probably a good indicator of the width of the crystals.

Clearly the scale-up has been effective in terms of producing crystals of a similar size at both scales at the end of the batch. However the in-situ tools offer the opportunity monitor the crystallization in its entirety. Figure 6.7 compares FBRM trends in two size ranges for the laboratory scale experiment and Batch B.

Comparison of the trends show that even though the chord length distributions are similar at the end of the batch the mechanism through which this is achieved is different depending on the scale. By studying #/s (1-10 microns) (Figure 6.7) it is clear that nucleation occurs earlier at the pilot scale. It is difficult to see if the rate of nucleation was similar at both scales as some probe coating was observed at the pilot scale just after nucleation. The probe was manually cleaned soon after and the counts returned to normal. The rate of increase of the large counts (100-1000 microns) is 142

very similar for each batch. Undoubtedly agglomeration is playing some part in the crystallization as shown in the PVM images (Figure 6.7 (b)). The fact that the FBRM trends are similar at both scales indicates suitable process parameters have been chosen and the change in mixing conditions between the scales does not impact the kinetics of the crystallization.

Figure 6.6: PVM images of (a) Pilot Scale (b) Laboratory Scale.

143

3000 Pilot-Scale 2500 Lab-Scale

2000
#/s (1-10)

1500

1000

500

0 0 500 1000 1500 2000 Time (s) 2500 3000 3500 4000

Figure 6.7: Comparison of #/s (1-10 microns) between batch B and lab scale

600
Pilot-Scale

500

Lab-Scale

400

#/s (100-1000)

300

200

100

0
0 500 1000 1500 2000 Time (s) 2500 3000 3500 4000

Figure 6.8: Comparison of #/s (100-1000 microns) between batch B and lab scale

144

6.4 DISCUSSION

Crystallization scale-up requires knowledge of every impact the change in scale has on the crystallization. There are liquid-liquid interactions to be taken into account i.e. how the antisolvent is incorporated into the bulk solution. This is influenced by scale, geometry, type of impeller, agitation intensity, addition location, addition rate, feed pipe diameter and solution and antisolvent concentration (i.e. liquid viscosities and densities). All of these factors can influence the micro-, meso- and macromixing properties of the system and hence the supersaturation profiles within the system. In turn the supersaturation profile impacts the nucleation, dissolution, growth and agglomeration rates. These in turn impact on the yield, purity and size of the crystal product. With this in mind it becomes clear that to try and scale an antisolvent crystallization it is impossible to do so based on a single traditional scale up parameter such as blend time or power per unit volume as no single parameter can take into account all of the factors.

Equally important is the fact that a change in scale impacts on the physical aspects of the crystallization. Thus, segregation is more likely to occur at larger scales where the mixing is not sufficient to fully suspend the crystal slurry (as was the case in Batch A of this study). Similarly, attrition and secondary nucleation rates will be affected by a change in scale. In the lab where the surface area to volume ratio is high, attrition and secondary nucleation rates are more sometime significantly elevated compared to larger scales where this ratio is low.

145

With these points in mind care should taken when scaling crystallization processes, specifically antisolvent crystallization, not to make assumptions based on a single mixing parameter. In many cases by combining a sound understanding of the crystallization system with a degree of common sense, effective scale-up can be achieved, as observed in this study. For example, by choosing a pitched blade impeller better crystal suspension may be achieved but this may be at the cost of entraining air into the system for a prolonged period. By studying the impact of mixing conditions on attrition levels in the laboratory the choice of impeller may become easier.

In order to scale-up effectively, intimate knowledge of every impact the change in scale will have on the crystallization is necessary. Current research in this area focuses on the use of CFD coupled with the population balance equations. This approach has the ability to take into account almost every parameter that can affect the liquid-liquid interactions. However it remains some way off in terms of predicting crystallization mechanisms such as growth rates, nucleation rates and agglomeration rates that are influenced in competing directions by the mixing effects. Despite this the approach appears to be the only one available that can fully take into account all of the variables that impact crystallization scale-up.

Before this becomes a reality there are still some common sense approaches that can be taken to enhance the likelihood of scale-up success. For antisolvent crystallization homogeneity is vital. By eliminating areas of local supersaturation problems such as elevated nucleation rates, incorrect polymorph nucleation and phase separation can be avoided. This can be achieved be adding slowly, at high agitation in a good addition

146

location. However, care must be taken to ensure that these parameters are suitable. For example the addition rate must not be so slow that the batch-time is compromised and throughput reduced. The agitation rate must not be so high as to introduce significant attrition and/or secondary nucleation. If this is a problem there are other techniques that can be used to improve dispersion such as adding anti-solvent through a narrow pipe in the form of a jet facilitating dispersion. Finally the choice of addition point can be aided by employing CFD to model how the anti-solvent will be dispersed.

6.5 CONCLUSION

The pilot-scale crystallization of benzoic acid from ethanol-water mixtures using water as an anti-solvent has been studied. Adequate scale-up in terms of particle size was achieved. CFD was used to assess the ideal addition location and it was observed that this point changed with the increase in the liquid volume. Changing the addition location during a batch however had no effect on the final crystal product. Effective scale-up in terms of particle size was achieved by using extensive data gathered at the laboratory scale to design a scale-up strategy. Results at the lab scale showed that a slow addition rate, intense agitation and a precise addition location (based on CFD) allowed for a robust and repeatable crystallization. These conditions were mimicked at the pilot scale and results showed that the final product particle size, in terms of chord length distribution and in process images, was almost identical. This was achieved for a vessel that was geometrically dissimilar to laboratory scale vessel where the initial characterization studies were conducted. This study shows that scaleup is possible without relying on traditional scale-up parameters such as power per

147

unit volume. And it is proposed that for anti-solvent crystallization using such parameters may not be the most useful approach for crystallization given the numerous mixing interactions that impact on crystallization. Current research aimed at modelling the complete crystallization process using computational fluid dynamics coupled with the population balance will in the future be useful but is still limited. Currently, common sense allied with sound knowledge of the crystallization mechanisms for a given system are the most useful tools for effective crystallization scale-up.

148

7. THESIS CONCLUSIONS

The anti-solvent crystallization of benzoic acid from ethanol-water solutions using water as the anti-solvent at multiple scales has been presented. A thorough review of the literature indicated that anti-solvent crystallization has been neglected, in favour of the study of cooling crystallization, and, in the cases where anti-solvent crystallization is studied the model system used is often inorganic. While some crystallization process parameters have been studied in detail (addition rate, concentration and agitation intensity) many have been neglected, including, antisolvent addition location and supersaturation.

The starting point for crystallization characterization is the solubility curve and metastable zone width. For this study, the solubility was measured using three separate techniques gravimetric analysis, FBRM, and ATR-FTIR and the results were in good agreement. The gravimetric method proved a simple and cost effective technique, however, it can be time consuming and there was some disagreement between the results of the liquid and solid analysis. The ATR-FTIR technique required the development of a calibration model before the solubility could be measured. However, once the model was developed and validated the solubility was measured rapidly. The ATR-FTIR method can also be used in other ways to study crystallization experiments most notably to monitor supersaturation (Chapter 5) and in this respect can offset the significant investment required for implementation. The FBRM technique can be used to rapidly measure solubility by quickly identifying the point of crystal dissolution. It does not require any calibration and its application as the standard in process tool for crystallization characterization makes it an obvious 149

choice for measuring solubility. In order to validate the experimental results garnered using the three techniques the UNIQAC method was used to model the system. The model results correlated well with the experimental values even in the region at low anti-solvent concentration where solubility increased with anti-solvent concentration. This solubility measurement work highlighted the varied methods available to investigators wishing to gather accurate solubility information and also provided the platform necessary for further characterization of this system including the metastable zone width and nucleation kinetics under varied process conditions.

The metastable zone width was measured at the 500 mL scale, under various process conditions including addition rate and location, and, agitation intensity. As expected increasing the addition rate resulted in a wider metastable zone width, however, the influence of addition location proved to be extremely important with variable and sometimes negative metastable zone widths occurring when anti-solvent was added close the wall of the vessel. This problem was exacerbated at low agitation intensities. The inadequate incorporation of the antisolvent into the bulk solution was identified as a possible cause of this variability and a study of the vessel using computational fluid dynamics confirmed this to be the case. Nucleation kinetics were estimated for the well mixed case where anti-solvent was added close to the impeller and the influence of agitation intensity on the kinetics was quantified. A unique expression relating the nucleation rate to the agitation intensity was developed by modifying classic nucleation theory for the anti-solvent case. This study highlighted the sensitive and sometimes counterintuitive nature of crystallization processes. By moving the addition location a very short distance a significant difference in the repeatability and robustness of the metastable zone width was observed. The sensitivity of this system

150

to such small changes highlights some of the difficulties associated with scale-up and achieving a consistent process and product performance.

In order to facilitate efficient scale-up a series of crystallizations experiments were undertaken at the 500 mL scale to identify suitable operating parameters that would ensure a repeatable and robust crystallization. The in situ tools offered the opportunity to monitor each crystallization in real time for the full duration of the experiment and allowed the comparison of each crystallization over the entirety of each run. High agitation intensity and a slow addition rate resulted in a very repeatable crystallization in terms of supersaturation, monitored using ATR-FTIR, and particle size, monitored using FBRM. The in situ data also allowed the elucidation of growth rate kinetics for the system. By combining the ATR-FTIR supersaturation information with growth rate data, generated using the FBRM, a kinetic expression for growth was rapidly estimated. This method highlights the opportunity to monitor growth rate kinetics in situ and in real time for any crystallization process. The ability to do this may prove useful for the control of crystallization processes in the lab as well as at production scales. The ability to modify the growth rate in real time by tuning process parameters would be a challenging but potentially valuable alternative to current production scale crystallization practices.

Current crystallization practice requires the efficient and effective scale-up of the crystallization from the lab to production, typically with the aid of an intermediate pilot scale. In order to study the sale-up behaviour of this system three 70 L pilot scale batches were conducted in a geometrically dissimilar vessel under different process conditions. The process conditions implemented were chosen by combining the

151

detailed information gathered over the course of this study - accurate solubility measurement, the impact of agitation on nucleation kinetics, and the choice of suitable process parameters to ensure repeatability. Computational fluid dynamics was also used to model mixing patterns in the vessel allowing suitable addition locations to be chosen. A successful scale-up was achieved based on repeatability, short batch time, and a similar yield and particle size as the laboratory scale process. The FBRM distributions and PVM images clearly showed the similarity in particle size and morphology at both scales. However, significant segregation and premature nucleation in the pilot scale vessel were observed indicating further work would be needed before scaling up to the full production scale.

152

8. NOMENCLATURE

A c cs f g G i j J k kn kg n N m P R s T Tm t w

anti-solvent concentration (g antisolvent/ g solvent) concentration (g/g) saturation concentration fugacity growth rate order growth rate component in UNIQUAC method component in UNIQUAC method nucleation rate component in UNIQUAC method nucleation rate constant growth rate constant nucleation order agitation intensity (rpm) mass ratio of anti-solvent to solvent (-) pressure anti-solvent addition rate (gs-1) mass of solute (g) temperature melting point time (s) mass of solvent (g)

153

Symbols

shape factor agitation parameter for nucleation kinetics supersaturation (g/g)

Amax metastable zone width Amax latent heat of fusion liquid activity coefficient density binary interaction parameter

154

9: REFERENCES
Abbas, A., D. Nobbs, and J. A. Romagnoli, "Investigation of on-line optical particle characterization in reaction, and cooling crystallization systems. Current state of the art," Measurement Science & Technology 13 (3): 349-356 (2002) Akiti, O. and P. M. Armenante, "Experimentally-validated micromixing-based CFD model for fed-batch stirred-tank reactors," Aiche Journal 50 (3): 566-577 (2004) Alfano, J. C. et al., "Polyelectrolyte-induced aggregation of microcrystalline cellulose: Reversibility and shear effects," Journal of Colloid and Interface Science 223 (2): 244-254 (2000) Barata, P. A. and M. L. Serrano, "Salting-out precipitation of potassium dihydrogen phosphate (KDP) .2. Influence of agitation intensity," Journal of Crystal Growth 163 (4): 426-433 (1996) Barata P.A., Serrano M.L., Salting-out precipitation of potassium dihydrogen phosphate (KDP). III. Growth Process, Journal of Crystal Growth, 194, 1, 101-108 (1998) Barata P.A., Serrano M.L., Salting-out precipitation of potassium dihydrogen phosphate (KDP). IV. Characterization of Final Product, Journal of Crystal Growth, 194, 1, 109-118 (1998) Barrett, P. and B. Glennon, "In-line FBRM monitoring of particle size in dilute agitated suspensions," Particle & Particle Systems Characterization 16 (5): 207-211 (1999)

155

Barrett, P. and B. Glennon, "Characterizing the metastable zone width and solubility curve using lasentec FBRM and PVM," Chemical Engineering Research & Design 80 (A7): 799-805 (2002) Barrett, P. et al., "A review of the use of process analytical technology for the understanding and optimization of production batch crystallization processes," Organic Process Research & Development 9 (3): 348-355 (2005) Beckmann, W., "Nucleation phenomena during the crystallisation and precipitation of Abecarnil," Journal of Crystal Growth 199: 1307-1314 (1999) Blanco, A. et al., "Focused beam reflectant measurement as a tool to measure flocculation," Tappi Journal 1 (10): 14-20 (2002) Bohlin, M. and A. C. Rasmuson, "Importance of macromixing in batch cooling crystallization," Aiche Journal 42 (3): 691-699 (1996) Borissova, A. et al., "Examination of the semi-batch crystallization of benzophenone from saturated methanol solution via aqueous antisolvent drowning-out as monitored in-process using ATR FTIR spectroscopy," Crystal Growth & Design 4 (5): 10531060 (2004) Charmolue, H. and R. W. Rousseau, "L-Serine Obtained by Methanol Addition in Batch Crystallization," Aiche Journal 37 (8): 1121-1128 (1991) Choi, YC., Morgenroth, E., Monitoring biofilm detachment under dynamic changes in shear stress using laser-based particle size analysis and mass fractionation, Water Sci Technol., ;47, 5, 69-76 (2003)

156

Clarke, M. A. and P. R. Bishnoi, "Determination of the intrinsic rate constant and activation energy of CO2 gas hydrate decomposition using in-situ particle size analysis," Chemical Engineering Science 59 (14): 2983-2993 (2004) David, R., "General rules for prediction of the intensity of micromixing effects on precipitations," Powder Technology 121 (1): 2-8 (2001) De Clercq, B., P. A. Lant, and P. A. Vanrolleghem, "Focused beam reflectance technique for in situ particle sizing in wastewater treatment settling tanks," Journal of Chemical Technology and Biotechnology 79 (6): 610-618 (2004) Derdour, L. et al., "Real-time evaluation of the concentration of impurities during organic solution crystallization," Powder Technology 129 (1-3): 1-7 (2003) Doki, N. et al., "Process control of seeded batch cooling crystallization of the metastable alpha-form glycine using an in-situ ATR-FTIR spectrometer and an in-situ FBRM particle counter," Crystal Growth & Design 4 (5): 949-953 (2004) Dowding, P. J., J. W. Goodwin, and B. Vincent, "Factors governing emulsion droplet and solid particle size measurements performed using the focused beam reflectance technique," Colloids and Surfaces A-Physicochemical and Engineering Aspects 192 (1-3): 5-13 (2001) Dunuwila, Dilum D., I. I. Carroll, and Kris A. Berglund, "An investigation of the applicability of attenuated total reflection infrared spectroscopy for measurement of solubility and supersaturation of aqueous citric acid solutions," Journal of Crystal Growth 137 (3-4): 561-568 (1994)

157

Dunuwila, Dilum D. and Kris A. Berglund, "ATR FTIR spectroscopy for in situ measurement of supersaturation," Journal of Crystal Growth 179 (1-2): 185-193 (1997) Feng, L. L. and K. A. Berglund, "ATR-FTIR for determining optimal cooling curves for batch crystallization of succinic acid," Crystal Growth & Design 2 (5): 449-452 (2002) Franke, D. and W. Gosele, "Modeling precipitation with a hydrodynamic approach," Chemie Ingenieur Technik 71 (11): 1245-1252 (1999) Fujiwara, M. et al., "Paracetamol crystallization using laser backscattering and ATRFTIR spectroscopy: Metastability, agglomeration, and control," Crystal Growth & Design 2 (5): 363-370 (2002) Gee, E.A., Cunnigham, William K., Heindl, R., Pilot Plants: Production of Iron-Free Alum, Ind. Eng. Chem., 39, 9, 1178 - 1188 (1947) Gong, J. B. et al., "Simulation and scale-up of barium sulphate precipitation process using CFD modeling," Chinese Journal of Chemical Engineering 13 (2): 167-172 (2005) Gong, J. B. et al., "Simulation and scale-up of barium sulphate precipitation process using CFD modeling," Chinese Journal of Chemical Engineering 13 (2): 167-172 (2005) Granberg, Roger A. et al., "Primary nucleation of paracetamol in acetone-water mixtures," Chemical Engineering Science 56 (7): 2305-2313 (2001)

158

Granberg R. A., Rasmuson, A.C., Solubility of Paracetamol in Pure Solvents, J. Chem. Eng. Data, 44 (6), 1391 -1395 (1999) Groen, H., Roberts, K.J., Application of ATR FTIR spectroscopy for on-line determination of solute concentration, Proceedings of the 14th International Symposium on Industrial Crystallization (1999) Groen, H. and K. J. Roberts, "Nucleation, growth, and pseudo-polymorphic behavior of citric acid as monitored in situ by attenuated total reflection Fourier transform infrared spectroscopy," Journal of Physical Chemistry B 105 (43): 10723-10730 (2001) Heath, A. R. et al., "Estimating average particle size by focused beam reflectance measurement (FBRM)," Particle & Particle Systems Characterization 19 (2): 84-95 (2002) Heffels, C. et al., "Control of particulate processes by optical measurement techniques," Particle & Particle Systems Characterization 15 (5): 211-218 (1998) Holmback, X. and A. C. Rasmuson, "Size and morphology of benzoic acid crystals produced by drowning-out crystallisation," Journal of Crystal Growth 199: 780-788 (1999). Hukkanen, E. J. and R. D. Braatz, "Measurement of particle size distribution in suspension polymerization using in situ laser backscattering," Sensors and Actuators B-Chemical 96 (1-2): 451-459 (2003)

159

Jeffers, P. et al., "Focussed beam reflectance measurement (FBRM) monitoring of particle size and morphology in suspension cultures of Morinda citrifolia and Centaurea calcitrapa," Biotechnology Letters 25 (23): 2023-2028 (2003) Kaneko, S. et al., "Effect of supersaturation on crystal size and number of crystals produced in antisolvent crystallization," Journal of Chemical Engineering of Japan 35 (11): 1219-1223 (2002) Kitamura, M. and M. Sugimoto, "Anti-solvent crystallization and transformation of thiazole-derivative polymorphs--I: effect of addition rate and initial concentrations," Journal of Crystal Growth 257 (1-2): 177-184 (2003) Kougoulos, E., A. G. Jones, and M. W. Wood-Kaczmar, "Modelling particle disruption of an organic fine chemical compound using Lasentec focussed beam reflectance monitoring (FBRM) in agitated suspensions," Powder Technology 155 (2): 153-158 (2005) Kougoulos, E., A. G. Jones, and M. W. Wood-Kaczmar, "Estimation of crystallization kinetics for an organic fine chemical using a modified continuous cooling mixed suspension mixed product removal (MSMPR) crystallizer," Journal of Crystal Growth 273 (3-4): 520-528 (2005) Kougoulos, E. et al., "Use of focused beam reflectance measurement (FBRM) and process video imaging (PVI) in a modified mixed suspension mixed product removal (MSMPR) cooling crystallizer," Journal of Crystal Growth 273 (3-4): 529-534 (2005) Kougoulos, E., A. G. Jones, and M. Wood-Kaczmar, "CFD modelling of mixing and heat transfer in batch cooling crystallizers - Aiding the development of a hybrid

160

predictive compartmental model," Chemical Engineering Research & Design 83 (A1): 30-39 (2005) Kougoulos, E., A. G. Jones, and M. W. Wood-Kaczmar, "A hybrid CFD compartmentalization modeling framework for the scaleup of batch cooling crystallization processes," Chemical Engineering Communications 193 (8): 10081023 (2006) Langston, P. A. et al., "Particle and droplet size analysis from chord measurements using Bayes' theorem," Powder Technology 116 (1): 33-42 (2001) Law, D. J., A. J. Bale, and S. E. Jones, "Adaptation of focused beam reflectance measurement to in-situ particle sizing in estuaries and coastal waters," Marine Geology 140 (1-2): 47-59 (1997) Lewiner, F. et al., "Improving batch cooling seeded crystallization of an organic weed-killer using on-line ATR FTIR measurement of supersaturation," Journal of Crystal Growth 226 (2-3): 348-362 (2001) Lewiner, F. et al., "On-line ATR FTIR measurement of supersaturation during solution crystallization processes. Calibration and applications on three solute/solvent systems," Chemical Engineering Science 56 (6): 2069-2084 (2001) Lewiner, F. et al., "An Online Strategy To Increase the Average Crystal Size during Organic Batch Cooling Crystallization," Industrial & Engineering Chemistry Research 41 (5): 1321-1328 (2002)

161

Li, M., D. Wilkinson, and K. Patchigolla, "Comparison of particle size distributions measured using different techniques," Particulate Science and Technology 23 (3): 265-284 (2005) 168. Li, M. Z., D. Wilkinson, and K. Patchigolla, "Determination of non-spherical particle size distribution from chord length measurements. Part 2: Experimental validation," Chemical Engineering Science 60 (18): 4992-5003 (2005) Li, M. Z. and D. Wilkinson, "Determination of non-spherical particle size distribution from chord length measurements. Part 1: Theoretical analysis," Chemical Engineering Science 60 (12): 3251-3265 (2005) Liotta, V. and V. Sabesan, "Monitoring and feedback control of supersaturation using ATR-FTIR to produce an active pharmaceutical ingredient of a desired crystal size," Organic Process Research & Development 8 (3): 488-494 (2004) Liu, Weidong, Nigel N. Clark, and Ali Ihsan Karamavruc, "Relationship between bubble size distributions and chord-length distribution in heterogeneously bubbling systems," Chemical Engineering Science 53 (6): 1267-1276 (1998) Loan, M. et al., "Iron oxy-hydroxide crystallization in a hyd ro metallurgical residue," Journal of Crystal Growth 235 (1-4): 482-488 (2002) Lffelmann, M., Mersmann, A., A New Method for Supersaturation Measurement, International Journal of Modern Physics B, 16, 01-02, 391-398 (2002) Madras, G. and B. J. Mccoy, "Mixing effects on particle precipitation," Industrial & Engineering Chemistry Research 44 (14): 5267-5274 (2005)

162

Marcant, B. and R. David, "Experimental-Evidence for and Prediction of Micromixing Effects in Precipitation," Aiche Journal 37 (11): 1698-1710 (1991) Marchisio, D. L. et al., "Effect of fluid dynamics on particle size distribution in particulate processes," Chemical Engineering & Technology 29 (2): 191-199 (2006) McDonald, K. A., A. P. Jackman, and S. Hurst, "Characterization of plant suspension cultures using the focused beam reflectance technique," Biotechnology Letters 23 (4): 317-324 (2001) Monnier, O. et al., "Model identification of batch cooling crystallizations through calorimetry and image analysis," Chemical Engineering Science 52 (7): 1125-1139 (1997) Mori, Y. et al., "CFD study of local crystal growth rate in a continuous suspension crystallizer," Journal of Chemical Engineering of Japan 35 (11): 1178-1187 (2002) Mullin, J.W., Garside, J., Unahabhokha, R., Diffusivities of Ammonium and Potassium Alums in Aqueous Solutions, Journal of Applied Chemistry of the USSR, (1965) Mullin, J., Teodossiev, N., Shnel, O., Potassium sulphate precipitation from aqueous solution by salting-out with acetone, Chemical Engineering and Processing, 26, 2, 93-99 (1989) Mydlarz, J. and A. G. Jones, "Crystallization and Agglomeration Kinetics During the Batch Drowning-Out Precipitation of Potash Alum with Aqueous Acetone," Powder Technology 65 (1-3): 187-194 (1991)

163

Nyvlt. J., Kinetics of nucleation in solutions, Journal of Crystal Growth, 3, 377383, (1968)
OGrady D., et al, 'The Effect of Mixing on the Metastable Zone Width and Nucleation Kinetics In the Anti-solvent Crystallization of Benzoic Acid, Chemical Engineering Research and Design, Transactions IChemE part A, 85 (7) 945-953, 2007

OGrady D. & Glennon B.,, Solubility Measurement for an Anti-solvent Crystallization System Using Gravimetric Analysis, ATR-FTIR and FBRM', Crystal Growth and Design, in press

O'Rourke, A. M. and P. F. MacLoughlin, "A comparison of measurement techniques used in the analysis of evolving liquid-liquid dispersions," Chemical Engineering and Processing 44 (8): 885-894 (2005) O'Sullivan, B. et al., "In situ monitoring of polymorphic transitions," Organic Process Research & Development 7 (6): 977-982 (2003) O'Sullivan, B. and B. Glennon, "Application of in situ FBRM and ATR-FTIR to the monitoring of the polymorphic transformation of D-mannitol," Organic Process Research & Development 9 (6): 884-889 (2005) Oosterhof, H., G. J. Witkamp, and G. M. van Rosmalen, "Some antisolvents for crystallisation of sodium carbonate," Fluid Phase Equilibria 155 (2): 219-227 (1999) Paul, E. L., H. H. Tung, and M. Midler, "Organic crystallization processes," Powder Technology 150 (2): 133-143 (2005)

164

Pearson, A. P., B. Glennon, and P. M. Kieran, "Comparison of morphological characteristics of Streptomyces natalensis by image analysis and focused beam reflectance measurement," Biotechnology Progress 19 (4): 1342-1347 (2003) Pearson, A., Glennon, B., Kieran, P., Monitoring of cell growth using the focused beam reflectance method, Journal of Chemical Technology & Biotechnology, 79, 10, 1142 -1147 (2004) Pettersson, M. and A. C. Rasmuson, "Hydrodynamics of suspensions agitated by pitched-blade turbine," Aiche Journal 44 (3): 513-527 (1998) Pina, C. M., L. FernandezDiaz, and M. Prieto, "Crystallization of beta''-LiNH4SO4 and (NH4)(2)SO4 in gels: Growth morphology and epitaxy phenomena," Journal of Crystal Growth 177 (1-2): 102-110 (1997) Pina, C. M. et al., "Metastability in drowning-out crystallisation: precipitation of highly soluble sulphates," Journal of Crystal Growth 222 (1-2): 317-327 (2001) Pohorecki, R. and J. Baldyga, "New Model of Micromixing in Chemical Reactors .1. General Development and Application to a Tubular Reactor," Industrial & Engineering Chemistry Fundamentals 22 (4): 392-397 (1983) Rohani, S. and J. Baldyga, "Micromixing Described in Terms of Inertial Convective Disintegration of Large Eddies and Viscous Convective Interactions Among Small Eddies .2. Semibatch and Continuous Stirred Tank Reactors," Chemical Engineering Science 42 (11): 2611-2619 (1987) Rohani, S. and J. Baldyga, "Mathematical-Modeling of Micromixing in Liquid Liquid-Systems," Hungarian Journal of Industrial Chemistry 16 (1): 85-97 (1988)

165

Ruf, A., J. Worlitschek, and M. Mazzotti, "Modeling and experimental analysis of PSD measurements through FBRM," Particle & Particle Systems Characterization 17 (4): 167-179 (2000) Scott, C. and S. Black, "In-line analysis of impurity effects in crystallization," Organic Process Research & Development 9 (6): 890-893 (2005) Sha, Z. et al., "Application of CFD simulation to suspension crystallization-actors affecting size-dependent classification," Powder Technology 121 (1): 20-25 (2001) Shi, B., W. J. Frederick, and R. W. Rousseau, "Effects of calcium and other ionic impurities on the primary nucleation of burkeite," Industrial & Engineering Chemistry Research 42 (12): 2861-2869 (2003) Simmons, M. J. H., P. A. Langston, and A. S. Burbidge, "Particle and droplet size analysis from chord distributions," Powder Technology 102 (1): 75-83 (1999) Sistare, F., St. Pierre Berry, L., Mojica, C., Process Analytical Technology: An Investment in Process Knowledge, Org. Process Res. Dev., 9, 3, 332 -336 (2005) Sparks, R.G., Dobbs, C.L., The use of laser backscatter instrumentation for the online measurement of the particle size distribution of emulsions 10, 5, 279-289 (1993) Tadayyon, A., Rohani, S., Determination of Particle Size Distribution by Par-Tec 100: Modeling and Experimental Results Particle and Particle Systems Characterization, 15, 3, 127-135 (1998) Tadayyon, A. and S. Rohani, "Control of fines suspension density in the fines loop of a continuous KCI crystallizer using transmittance measurement and an FBRM (R) probe," Canadian Journal of Chemical Engineering 78 (4): 663-673 (2000) 166

Takiyama, H., T. Otsuhata, and M. Matsuoka, "Morphology of NaCl crystals in drowning-out precipitation operation," Chemical Engineering Research & Design 76 (A7): 809-814 (1998) Tavare, N. S., "Mixing, Reaction, and Precipitation - Interaction by Exchange with Mean Micromixing Models," Aiche Journal 41 (12): 2537-2548 (1995) Togkalidou, T. et al., "Solute concentration prediction using chemometrics and ATRFTIR spectroscopy," Journal of Crystal Growth 231 (4): 534-543 (2001) Togkalidou, T. et al., "Solution concentration prediction for pharmaceutical crystallization processes using robust chemometrics and ATR FTIR spectroscopy," Organic Process Research & Development 6 (3): 317-322 (2002) Torbacke, M. and A. C. Rasmuson, "Influence of different scales of mixing in reaction crystallization," Chemical Engineering Science 56 (7): 2459-2473 (2001) Torbacke, M. and A. C. Rasmuson, "Mesomixing in semi-batch reaction crystallization and influence of reactor size," Aiche Journal 50 (12): 3107-3119 (2004) Vittorf, N.M. Separation of aluminium-sulphate from iron by means of alcohol, Institut prikladnoi mineralogii i petrografii, Moscow. Transactions (1924) Wei, H et al., "Computational fluid dynamics modeling of the precipitation process in a semibatch crystallizer," Industrial & Engineering Chemistry Research 40 (23): 5255-5261 (2001)

167

Woo, X. Y. et al., "Simulation of mixing effects in antisolvent crystallization using a coupled CFD-PDF-PBE approach," Crystal Growth & Design 6 (6): 1291-1303 (2006) Worlitschek, J. and M. Mazzotti, "Choice of the focal point position using lasentec FBRM," Particle & Particle Systems Characterization 20 (1): 12-17 (2003) Worlitschek, J. and M. Mazzotti, "Model-based optimization of particle size distribution in batch-cooling crystallization of paracetamol," Crystal Growth & Design 4 (5): 891-903 (2004) Worlitschek, J., Hocker, T., Mazzotti, M., Restoration of PSD from Chord Length Distribution Data using the Method of Projections onto Convex Sets, Particle & Particle Systems Characterization, 22, 2, 81-98 (2005) Wynn, E. J. W., "Relationship between particle-size and chord-length distributions in focused beam reflectance measurement: stability of direct inversion and weighting," Powder Technology 133 (1-3): 125-133 (2003) Yu, Z. Q., R. B. H. Tan, and P. S. Chow, "Effects of operating conditions on agglomeration and habit of paracetamol crystals in anti-solvent crystallization," Journal of Crystal Growth 279 (3-4): 477-488 (2005) Zauner, R. and A. G. Jones, "Scale-up of continuous and semibatch precipitation processes," Industrial & Engineering Chemistry Research 39 (7): 2392-2403 (2000) Zauner, R. and A. G. Jones, "Scale-up of Continuous and Semibatch Precipitation Processes," Industrial & Engineering Chemistry Research 39 (7): 2392-2403 (2000)

168

Zauner, R. and A. G. Jones, "Mixing effects on product particle characteristics from semi-batch crystal precipitation," Chemical Engineering Research & Design 78 (A6): 894-902 (2000). Zauner, R. and A. G. Jones, "On the influence of mixing on crystal precipitation processes - application of the segregated feed model," Chemical Engineering Science 57 (5): 821-831 (2002) Zhu, Y. et al., "Study of the solubility and the metastable zone of 1,3dihydroxyacetone for the drowning-out process," Journal of Crystal Growth 257 (34): 370-377 (2003). Zijlema, T. G. et al., "The production of pure NaCl by the antisolvent crystallisation of NaCl center dot 2H(2)O and a consecutive recrystallisation step," Journal of Crystal Growth 199: 789-795 (1999) Zijlema, T. G. et al., "Antisolvent crystallization as an alternative to evaporative crystallization for the production of sodium chloride," Industrial & Engineering Chemistry Research 39 (5): 1330-1337 (2000)

169

170

APPENDIX A

APPLICATION OF COMPUTATION FLUID DYNAMICS TO EXPLAIN VARIATION IN THE MSZW FOR ANT-SOLVENT ADDITION

CRYSTALLIZATION

Des O Grady, Mark Barrett, Eoin Casey, Brian Glennon

Chapter 3 focuses on the study of the impact of process parameters on the MSZW of an anti-solvent system and chapter 5 looks at the scale up of the crystallization to the pilot-plant. This technical note will focus on the application of velocity profiles within crystallization vessels at both scales, predicted using Computational Fluid Dynamics (CFD), to model crystallization behaviour. This work was carried out by Mark Barrett and Eoin Casey in conjunction with Des OGrady and Brian Glennon. A more in depth analysis of the use of CFD to model mixing for this crystallization system can be found elsewhere (OGrady et al., 2007)

1. Abstract

Computational Fluid Dynamics is used to model mixing conditions in a 500 mL vessel and a 70L vessel used to crystallize benzoic acid from ethanol water mixtures using water as the anti-solvent. At the 500mL scale, anti-solvent addition close to the impeller results in consistent MSZWs at all addition rates. However, when the addition location is close to the vessel wall the MSZW is variable and premature nucleation is common. CFD modelling indicated high downward velocities close to

171

the impeller but upward velocities close to the wall. This suggests that anti-solvent added close to the impeller is more rapidly incorporated into the vessel, limiting the potential for locally high supersaturation. However, close to the wall the anti-solvent is less rapidly incorporated and local regions of high supersaturation may be expected. This provides an explanation for the variations in MSZW behaviour observed. Similar velocity profiles were calculated for the 70 L pilot plant vessel in order to predict optimal addition location prior to crystallization. The profiles indicate that the optimal addition location changes as the volume in the reactor increases.

2. Introduction

Computational Fluid Dynamics have been used to model semi-batch reactive crystallizations and has been used to predict the effect of mixing on the supersaturation distribution and the resulting size of the crystals (Wei et al., 2001, Baldyga & Orciuch, 2001). It has also been used to study the effect of mixing on product yield (Akiti & Armenante, 2004). CFD has also been used in conjunction with crystallization kinetics and solubility data to simultaneously solve the mass and population balances for a reactive crystallization, allowing the impact of feed rate, agitation intensity, feed point and feed tube diameter on nucleation rate and crystal size to be illustrated (Zauner and Jones, 2001). While the focus of CFD research has been on reactive crystallization there is clearly scope for the use of the technique to model mixing in anti-solvent crystallization processes.

In this study, CFD is used to generate theoretical velocity profiles for the vessel prior to nucleation, in an effort to understand the observed trends in metastable zone width.

172

Velocity in the z-direction was chosen as a suitable parameter for assessing the mixing regime, as movement of the liquid in this plane is vital for the incorporation of anti-solvent into the bulk solution.

3. Computation Fluid Dynamics Model

To investigate the velocity profiles prior to nucleation, the system is treated as a single phase and the possible solid-liquid interactions after nucleation are not considered. Initially, the 500 ml vessel and its complex baffling system (the FBRM, REACT-IR and temperature probes) are constructed within GAMBIT 2.1.6. GAMBIT is a software package designed to aid in the construction and meshing of systems for computational fluid dynamics (CFD). The precise geometry and dimensions of the system (pitch blade impeller, probes etc.) are created in the graphical user interface allowing for the meshing of the vessel and its internals, along with the assigning of zone types and system specifics (i.e. fluid viscosity, internal reactor temperature and agitation speed).

A large number of individual volumes created in the construction of the geometry are then meshed. The system contains 823 individual volumes and 475,000 cells. An

unstructured hexahedral meshing scheme is applied, as less numerical diffusion errors are evident then in a tetrahedral-based mesh. The mesh created in GAMBIT is then exported to FLUENT 6.1.22 for solution of the momentum and continuity equations for the turbulent flow within the crystallizer. The Multiple Reference Frame (MRF) approach is applied to the system and the flow equations solved using the SIMPLE algorithm. The Reynolds number for the agitation speeds assessed lie between 12,500 (325 rpm) and 18,500 (475rpm). The standard k-epsilon turbulence model 173

was chosen to model the flows. In addition, a no-slip boundary condition was imposed on all walls and the free liquid surface is modelled with no vortex, a zeroflux and zero-stress conditions. To validate the model the tip speed of the impeller is calculated and compared to the tip speed calculated using the model. The model underestimated the tip speed by about 3% at 325 and 475 rpm indicating the model is suitable.

4. Results

4.1 500mL Scale

The difference in the behaviour of the MSZW can be explained in terms of mixing at different regions of the vessel. Figures 1 and 2 depict velocity in the z-direction through a cross section of the vessel at 475 rpm and 325 rpm respectively. Velocity in this direction was chosen, as it will impact on the incorporation of the anti-solvent into the bulk of the vessel. It is clear from Figure 1 that close to the impeller the velocity is in the downward direction, but close to the wall the velocity is in the upward direction. This means that when anti-solvent is added close to the impeller it is incorporated easily into the bulk solution. However, when it is added close to the wall this incorporation is more difficult. This may lead to an area of local supersaturation close to the feed point and premature, variable nucleation as observed experimentally. Furthermore, analysis of the same velocity profiles at lower agitation levels indicate that velocity in the downward direction is less intense and velocity in the upward direction is more pronounced. This explains the greater variability and

174

narrower MSZWs when anti-solvent is added close to the wall at lower agitation levels.

175

Figure 1: Velocity profiles in the z-direction at 475 rpm through a central cross section of the vessel

Figure 2: Velocity Profiles in the z-direction at 325 rpm through a central cross section of the vessel

176

4.2 70 L Scale At the 70 L scale CFD models indicate that the optimum feed location changes over the course of the crystallization. Initially, when the liquid level is close to the plane of the impeller the most intense downward velocities, and hence optimal addition location, are found close vessel wall (Figure 3). However, as the volume in the vessel increases the optimal location moves closer to the impeller shaft (Figure 4). These results assisted in the choice of addition location in Chapter 5. Two batches were conducted with the feed point close to the wall for the entire run and for the final batch the feed point was changed after some time to take advantage of the change in the optimal feed location.

177

Figure 3: Velcoity profiles in the z-direction at (a) 32 L and (b) 64 L showing change in optimum feed location 178

5. Discussion

The experimental results outlined in Chapter 3 can be readily accounted for by considering the different mixing regimes depending on the process conditions. CFD is a very useful tool for modelling mixing conditions within a crystallizer and in this case the CFD model explains the experimental results. Close to the impeller, mixing conditions favour anti-solvent incorporation leading to consistent MSZW results, whereas close to the wall anti-solvent incorporation is hindered leading to variability in the nucleation mechanism and non-repeatable MSZW measurements. This effect is exacerbated at lower agitation intensities where mixing conditions become even less favourable.

6. References

Akiti O., & Armenante P., Experimentally-validated micromixing-based CFD model for fed-batch stirred-tank reactors, AIChE, 50 (3) 566-577 (2004)

Bakldyga J., & Orciuch W., Barium Sulphate precipitation in a pipe and experimental study and modelling, Chemical Engineering Science, 58 (7) 2435-2444 (2001)

Wei H., et al., "Computational Fluid Dynamics Modelling of the Precipitation Process in a Semibatch Crystallizer," Industrial & Engineering Chemistry Research 40 (23): 5255-5261 (2000)

179