prr.hec.gov.pkprr.hec.gov.pk/jspui/bitstream/123456789/205/1/1680s.pdf · iii development and...
TRANSCRIPT
Development and Analysis of Batch andContinuous Crystallization Models
By
Iltaf Hussain
CIIT/SP09-PMT-004/ISB
PhD Thesis
In
Mathematics
COMSATS Institute of Information TechnologyIslamabad - Pakistan
Spring, 2012
ii
COMSATS Institute of Information Technology
Development and Analysis of Batch and ContinuousCrystallization Models
A Thesis Presented to
COMSATS Institute of Information Technology, Islamabad
In partial fulfillmentof the requirement for the degree of
PhD (Mathematics)
By
Iltaf Hussain
CIIT/SP09-PMT-004/ISB
Spring, 2012
iii
Development and Analysis of Batch and ContinuousCrystallization Models
A Post Graduate Thesis submitted to the Department of Mathematics as partialfulfillment of the requirement for the award of Degree of PhD Mathematics.
Name Registration Number
Iltaf Hussain CIIT/SP09-PMT-004/ISB
Supervisor
Dr. habil. Shamsul Qamar
Associate ProfessorDepartment of MathematicsIslamabad CampusCOMSATS Institute of Information Technology,Islamabad.June, 2012
iv
Final Approval__________________________________
This thesis titled
Development and Analysis of Batch and ContinuousCrystallization Models
By
Iltaf Hussain
CIIT/SP09-PMT-004/ISB
Has been approved
For the COMSATS Institute of Information Technology, Islamabad
External Examiner 1:___________________________________________Dr. Sohail Nadeem Associate Prof., QAU, Islamabad
External Examiner 2:___________________________________________Dr. Khalid Saifullah Syed, Assistant Prof., BZU, Multan
Supervisor: ________________________________________________Dr. habil. Shamsul QamarAssociate Professor CIIT, Islamabad.
HoD:_____________________________________________________Professor Dr. Moiz-ud-Din KhanHoD (Mathematics/Islamabad)
Dean, Faculty of Sciences: ______________________________Professor Dr. Arshad Saleem Bahatti
v
Declaration
I, Iltaf Hussain, CIIT/SP09-PMT-004/ISB, hereby declare that I have produced the workpresented in this thesis, during the scheduled period of study. I also declare that I havenot taken any material from any source except referred to wherever due that amount ofplagiarism is within acceptable range. If a violation of HEC rules on research hasoccurred in this thesis, I shall be liable to punishable action under the plagiarism rules ofthe HEC.
Date: _________________ Signature of the student:
____________________________Iltaf Hussain
CIIT/SP09-PMT-004/ISB
vi
Certificate
It is certified that Iltaf Hussain, registration number CIIT/SP09-PMT-004/ISB has carriedout all the work related to this thesis under my supervision at the Department ofMathematics, COMSATS Institute of Information Technology, Islamabad and the workfulfills the requirement for award of PhD degree.
Date: _________________
Supervisor:
___________________________Dr. habil. Shamsul QamarAssociate Professor
Head of Department:
_____________________________Dr. Moiz-ud-Din Khan
ProfessorDepartment of Mathematics
vii
DEDICATION
I dedicate this thesis tomy parents
ACKNOWLEDGEMENTS
I owe my thanks to the Almighty ALLAH the most Gracious, the most Merciful, the unique
supreme power of the whole universe, Who blessed me the courage, potential and insight
that enabled me to complete this thesis. My all admirations to the last Holy Prophet
Hazrat Muhammad (PBUH) whose life is a candle for us in the darkness of life.
I feel honored to acknowledge my deepest gratitude to my affectionate and devoted super-
visor Dr. habil. Shamsul Qamar whose guidance and keen interest enabled me to complete
this research work. His cooperation and invigorating encouragement will always remain a
source of inspiration for me.
I would like to express my deep thank to our research collaborators Prof. Dr. Andreas
Seidel-Morgenstern and Prof. Dr. Martin Peter Elsner from the Max Planck Institute
for Dynamics of Complex Technical Systems Magdeburg, Germany for their guidance and
research contributions. I would also like to thank Prof. Dr. Moiz ud Din Khan, Head of
Department of Mathematics, CIIT, Islamabad, Pakistan for his cooperation and support.
I pay special thanks to the University of Engineering and Technology, Peshawar
(Mardan Campus), Khyber Pakhtunkhwa, Pakistan for study leave.
Finally, I am very much thankful to Higher Education Commission (HEC), Islam-
abad, Pakistan for financial support under Indigenous PhD 5000 Fellowship
Program Phase-IV under grant number 074-0714-Ps4-213.
Iltaf Hussain
CIIT/SP09-PMT-004/ISB
viii
ABSTRACT
Development and Analysis of Batch and Continuous
Crystallization Models
This thesis presents the development and simulation of batch and continuous crystallization
models. Especially, models are derived for simulating batch and continuous enantioselec-
tive preferential crystallization processes in single and coupled crystallizers. Such processes
are highly important in chemical and pharmaceutical industries. The effects of nucleation,
growth, and fines dissolution phenomena on the crystal size distribution (CSD) are inves-
tigated. For the first time continuous preferential crystallization is investigated and the
effects of different seeding and operating strategies on the process are analyzed. To judge
the quality of the process some goal functions are used, such as purity, productivity, yield
and mean crystal size of the preferred enantiomer. The semi-discrete high resolution finite
volume schemes (HR-FVS) and the discontinuous Galerkin (DG) finite element method
are proposed for solving these models. The resulting systems of ordinary differential equa-
tions (ODEs) are solved by using explicit and nonlinearly stable high order Runge- Kutta
method. The schemes satisfy the total variation bounded (TVB) property which guarantees
the positivity of the schemes, for example the non-negativity of CSD in the present case.
The suggested methods have capabilities to capture sharp discontinuities and narrow peaks
of the CSD. In DG-schemes, the accuracy of the method can be improved by introducing
additional nodes in the same solution element and, thus, avoids the expansion of mesh
stencils which is normally observed in high order finite volume schemes. For that reason,
the method can be easily applied up to boundary cells without loosing accuracy. It was
found that the proposed numerical schemes have the capability to solve the given models
more efficiently and accurately. The results support process design and optimization.
ix
Contents
1 Fundamentals of Crystallization Process 1
1.1 Crystallization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Types of Crystallization . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Batch Crystallization . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Continuous Crystallization . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Enantiomers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Preferential Crystallization . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Ternary Phase Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Historical Background and Motivation . . . . . . . . . . . . . . . . 6
1.7 New Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.8 Layout of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.9 Publications Related to the Research Work . . . . . . . . . . . . . 12
2 Population Balance Models 13
2.1 Population Balance Models . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Growth and Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Dissolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Aggregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Breakage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Goal Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
x
TABLE OF CONTENTS
3 Batch Crystallization 19
3.1 Batch Crystallization Model . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Numerical Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.1 Implementation of Discontinuous Galerkin finite element
method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.2 Implementation of Finite Volume Scheme . . . . . . . . . . 27
3.3 Test Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4 Batch Preferential Crystallization 45
4.1 Batch Preferential Crystallization Model . . . . . . . . . . . . . . . 46
4.2 Coupled Batch Preferential Crystallization Model . . . . . . . . . 48
4.3 Numerical Approximation of the Models . . . . . . . . . . . . . . . 51
4.4 Test Problems for Single Crystallizer . . . . . . . . . . . . . . . . . 54
4.5 Test Problems for Coupled Crystallizers . . . . . . . . . . . . . . . 59
5 Continuous Preferential Crystallization 79
5.1 Continuous Preferential Crystallization Model . . . . . . . . . . . 81
5.2 Numerical Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2.1 Implementation of HR-FVS . . . . . . . . . . . . . . . . . . . 88
5.2.2 Implementation of DG-Scheme . . . . . . . . . . . . . . . . . 90
5.3 Test Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6 Coupled Continuous Preferential Crystallization 128
6.1 Coupled Continuous Crystallization Model . . . . . . . . . . . . . 131
6.2 Implementation of Finite Volume Scheme . . . . . . . . . . . . . . 138
6.3 Test Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.3.1 Numerical Test Problem . . . . . . . . . . . . . . . . . . . . . 142
xi
7 Conclusion and Future Recommendations 161
7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.2 Future Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . 164
8 References 168
xii
List of Figures
1.1 Amino acid enantiomers [61]. . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Ternary phase diagram [61]. . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1 Batch crystallizer with fines dissolution [61]. . . . . . . . . . . . . . . . . . 21
3.2 Problem 1: Size-independent growth at t = 100 s. . . . . . . . . . . . . . . 32
3.3 Problem: Errors and CPU times of DG-scheme. . . . . . . . . . . . . . . . 32
3.4 Problem 1: Comparison of DG-schemes of different orders . . . . . . . . . . 33
3.5 Problem 2: Size-independent growth at t = 0.5 min. . . . . . . . . . . . . . 34
3.6 Problem 3: Size-dependent growth at t = 1000 s. . . . . . . . . . . . . . . . 37
3.7 Problem 4: Size-dependent growth at t = 180min. . . . . . . . . . . . . . . 39
3.8 Problem 5: CSD without fines dissolution. . . . . . . . . . . . . . . . . . . 43
3.9 Problem 5: Fines dissolution without delay . . . . . . . . . . . . . . . . . . 43
3.10 Test 5: Fines dissolution with delay. . . . . . . . . . . . . . . . . . . . . . . 44
3.11 Problem 5: Solute masses. . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.1 Crystallization in coupled vessels. . . . . . . . . . . . . . . . . . . . . . . . 49
4.2 Test 1: Preferred enantiomer CSD with HR-FVS and DG scheme. . . . . . 63
4.3 Test 1: Preferred enantiomer CSD for with and without FD. . . . . . . . . 63
4.4 Test 1: CSD for counter enantiomer with and without FD. . . . . . . . . . 64
4.5 Test 1: Mass fraction for preferred enantiomer with and without FD. . . . 64
4.6 Test 1: Mass fraction for counter enantiomer. . . . . . . . . . . . . . . . . 65
xiii
LIST OF FIGURES
4.7 Test 1: Supersaturation for target enantiomer with and without FD. . . . . 65
4.8 Test 1: Growth rate for target enantiomer with and without FD. . . . . . . 66
4.9 Test 1: Nucleation rate for target enantiomer with and without FD. . . . . 66
4.10 Test 1: Purity and mean crystal size of preferred enantiomer. . . . . . . . . 67
4.11 Test 1: Productivity of preferred enantiomer. . . . . . . . . . . . . . . . . . 67
4.12 Test 2: CSD for target enantiomer with HR-FVS and DG scheme. . . . . . 68
4.13 Test 2: CSD for target enantiomer. . . . . . . . . . . . . . . . . . . . . . . 68
4.14 Test 2: Mass fraction for preferred enantiomer with and without FD. . . . 69
4.15 Test 2: Mass fraction for counter enantiomer. . . . . . . . . . . . . . . . . 69
4.16 Test 2: Supersaturation for preferred enantiomer with and without FD. . . 70
4.17 Test 2: Growth rate for preferred enantiomer. . . . . . . . . . . . . . . . . 70
4.18 Test 2: Productivity of preferred enantiomer. . . . . . . . . . . . . . . . . . 71
4.19 Test 2: Purity and mean crystal size of preferred enantiomer. . . . . . . . . 71
4.20 Test 3: CSD for both enantiomers with HR-FVS and DG scheme. . . . . . 72
4.21 Test 3: Mass fraction for both enantiomers with HR-FVS and DG scheme. 72
4.22 Test 3: Purity and mean crystal size for preferred enantiomer. . . . . . . . 73
4.23 Test 3: Productivity of preferred enantiomer. . . . . . . . . . . . . . . . . . 73
4.24 Test 4: Preferred CSD with FD for single and coupled crystallizers. . . . . 74
4.25 Test 4: Mass fraction with FD for single and coupled crystallizers. . . . . . 74
4.26 Test 4: Supersaturation with FD for single and coupled crystallizers. . . . 75
4.27 Test 4: Growth rate with FD for single and coupled crystallizers. . . . . . . 75
4.28 Test 4: Productivity for single and coupled crystallizers without FD. . . . . 76
4.29 Test 4: Mean crystal size for single and coupled crystallizers without FD. . 76
4.30 Test 5: Preferred CSD for single and coupled crystallizers. . . . . . . . . . 77
4.31 Test 5: Purity and productivity of p-enantiomer for single and coupled crys-
tallizers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
xiv
4.32 Test 5: Mean crystal size of p-enantiomer for single and coupled crystallizers. 78
5.1 Principle of continuous enantioselective crystallization [66]. . . . . . . . . . 82
5.2 Problem 1 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. 97
5.3 Problem 1 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. 98
5.4 Problem 1 (case A): Error in mass balance and CSDs. . . . . . . . . . . . 99
5.5 Schematic diagram for periodic seeding. . . . . . . . . . . . . . . . . . . . . 100
5.6 Problem 1 (Case B): Periodic seeding: Left τS/τL = 1, Right: τS/τL = 2. . . 103
5.7 Problem 1 (Case B): Periodic seeding: Left τS/τL = 1, Right: τS/τL = 2. . . 104
5.8 Problem 2 (Case A): CSDs for the different mass of seed crystal. . . . . . . 107
5.9 Problem 2 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. 108
5.10 Problem 2 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. 109
5.11 Problem 2 (Case B): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. . 111
5.12 Problem 2 (Case B): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. . 112
5.13 Problem 2 (Case C): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. . 114
5.14 Problem 2 (Case C): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. . 115
5.15 Problem 3 (Case A): CSDs for the different mass of seed crystal. . . . . . . 118
5.16 Problem 3 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. 119
5.17 Problem 3 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. 120
5.18 Problem 3 (Case B): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. . 122
5.19 Problem 3 (Case B): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. . 123
5.20 Problem 3 (Case C): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. . 125
5.21 Problem 3 (Case C): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0. . 126
6.1 Principle of coupled continuous enantioselective crystallization. . . . . . . . 132
6.2 Case I: LS: Single crystallizer results. RS: Coupled crystallizer results. . . . 151
6.3 Case I: LS: Single crystallizer results. RS: Coupled crystallizer results. . . . 152
xv
6.4 Case I: LS: Single crystallizer results. RS: Coupled crystallizer results. . . . 153
6.5 Case I: LS: Single crystallizer results. RS: Coupled crystallizer results. . . . 154
6.6 Schematic diagram for periodic seeding. . . . . . . . . . . . . . . . . . . . . 154
6.7 Case II: LS: Single crystallizer results. RS: Coupled crystallizer results. . . 155
6.8 Case II: LS: Single crystallizer results. RS: Coupled crystallizer results. . . 156
6.9 Case II: LS: Single crystallizer results. RS: Coupled crystallizer results. . . 157
6.10 Case II: LS: Single crystallizer results. RS: Coupled crystallizer results. . . 158
6.11 Case III: LS: Single crystallizer results. RS: Coupled crystallizer results. . . 159
6.12 Case III: LS: Single crystallizer results. RS: Coupled crystallizer results. . . 160
7.1 Left: without counter enantiomer; Right: with counter enantiomer. . . . . 167
xvi
List of Tables
3.1 Problem 1: L1-errors and CPU times. . . . . . . . . . . . . . . . . . . . . . 31
3.2 Parameters for Problem 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Parameters for Problem 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Problem 5: Errors in mass balances for size-independent growth rate . . . 42
4.1 Parameters for Problem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2 Parameters for Problems 1, 2 and 4 . . . . . . . . . . . . . . . . . . . . . . 57
4.3 Parameters for Problem 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4 Parameters for Problems 3 and 5 . . . . . . . . . . . . . . . . . . . . . . . 59
4.5 Parameters for Problem 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.1 Physicochemical parameters related to threonine-water system ([17, 18]). . 95
5.2 Problem 1 (Case A): Values of goal functions at τL = 112.41 min. . . . . . 96
5.3 Problem 1 (case B): Periodic seeding: m(p)seeds = 4 g and τL = 112.4 min. . . 102
5.4 Problem 1 (case C): Results with FD (yes) and without FD (no) . . . . . . 105
5.5 Problem 2 (Case A): Values of goal functions at τL = 112.41 min. . . . . . 107
5.6 Problem 2 (Case B): Results with FD (yes) and without FD (no) . . . . . 110
5.7 Problem 2 (Case C): Periodic seeding: m(p)seeds = 4 g and τL = 112.4 min. . . 116
5.8 Problem 3 (Case A): Values of goal functions at τL = 112.41 min. . . . . . 118
5.9 Problem 3 (Case B): Results with FD (yes) and without FD (no) . . . . . 121
5.10 Problem 3 (Case C): Periodic seeding: m(p)seeds = 4 g, τL = 112.4 min. . . . . 127
xvii
LIST OF TABLES
6.2 Case I: Goal functions of single crystallizer without FD, τL = 112.41 min. . 149
6.3 Case I: Goal functions of coupled crystallizers without FD, τL = 112.41 min. 149
6.4 Case II: Results of single crystallizer with FD: FD (yes), without FD (no). 149
6.5 Case II: Results of coupled case with FD: FD (yes) and without FD (no). . 149
6.6 Case III : Periodic seeding in single crystallizer: m(p)seeds = 4 g, τL = 112.4 min.150
6.7 Case III: Periodic seeding in coupled case: m(p)seeds = 4 g, τL = 112.4 min. . . 150
xviii
LIST OF ABBREVIATIONS
Abbreviations
CFL courant-friedrichs-lewyCPU central processing unitCSD crystal size distributionDG discontinuous GalerkinFD fines dissolutionFDM finite difference schemeFEM finite element methodFV S finite volume schemeHR − FV S high resolution finite volume schemeLS left sideLPs legendre polynomialsMSMPR mixed suspension mixed product removalODE ordinary differential equationPBE population balance equationPBM population balance modelPC preferential crystallizationPDE partial differential equationPSD particle size distributionRS right sideS solidTV B total variation boundedTV D total variation diminishingPC preferential crystallizationPDE partial differential equationPSD particle size distributionRS right sideTV B total variation boundedTV D total variation diminishing
xix
Subscripts and superscripts
agg aggregationbreak breakagec crystallizercrz crystallizercryst crystallizerdiss dissolutioneff effectiveeq equilibriumexch exchangeGodu Ggodunov fluxin inflow stream to crystallizerk enantiomer k ∈ p, c (p : preferred, c : counter)L liquidLF Lax friedrichs fluxLLF Local Lax friedrichs fluxmin minimummax maximumnuc nucleationout outlet stream from the crystallizerprim primaryR Roe fluxS solidsolv solventsec secondarystd short notation for steady statesat saturated
xx
Symbols units
A1 constant of surface area dependency [m−2]a1 fitted parameter for log-normal distribution function [−]a2 fitted parameter for log-normal distribution function [−]a3 fitted parameter for log-normal distribution function [−]aASL parameter for size dependency of crystal growth [m−1]aeq temperature dependent parameter for solubility [−]aprim lumped constant of primary nucleation in Eq. (5.25) [−]B0 nucleation rate at negligible particle size z = 0 [s−1]
B(k)0 nucleation rate of k-th enantiomer at negligible particle size z = 0 [s−1]
B0,prim nucleation rate for primary heterogeneous mechanism [s−1]B0,prim,α nucleation rate for primary heterogeneous mechanism in tank α [s−1]B0,sec nucleation rate for secondary mechanism [s−1]B0,sec,α nucleation rate for secondary mechanism in tank α [s−1]B0,α nucleation rate at negligible particle size in tank α z = 0 [s−1]b nucleation order [−]
xxi
symbols (continued)
beq temperature dependent parameter for solubility [−]bsec exponent (secondary nucleation) [−]C solubility concentration [g/g of solvent]Ceq solubility concentration at equilibrium temperature [g/g of solvent]Ceq,α solubility concentration at equilibrium temperature for α crystallizer [kg/m3]
C(k)eq solubility concentration of k-th enantiomer at equilibrium temperature [g/g of solvent]
C0 Initial solubility concentration [g/g of solvent]D−
dis dissolution rate [m−1s−1]dASL parameter for size dependency of crystal growth [−]EA,b activation energy for secondary nucleation [J/mol]EA,g activation energy for crystal growth [J/mol]F number density function [m−1]Fi mean value of number density function [m−1]F0 CSD of initial seed crystals [m−1]
F(k)seeds incoming particle number flux due to seeding of k-th enantiomer [m−1s−1]
F(k)out particle number flux due to product removal of k-th enantiomer [m−1s−1]
F(k)pipe particle number flux due to fines dissolution of k-th enantiomer [m−1s−1]
xxii
symbols (continued)
F(k)0 Initial CSD of k-th enantiomer [m−1]
F (k) number density function of k-th enantiomer [m−1]
F(k)α number density of k-th enantiomer in tank α [m−1]
Fseeds number density function for seed crystals [m−1]Fin number density function of incoming particles [m−1]Fout number density function of outgoing particles [m−1]Fnuc number density function of nuclei [m−1]Fh numerical CSD [m−1]f (k) numerical flux for the k-th enantiomer [m−1]
f(k)α numerical flux for the k-th enantiomer in tank α [m−1]G crystal growth rate [m/s]G(k) crystal growth rate of k-th enantiomer [m/s]
G(k)α crystal growth rate of k-th enantiomer in tank α [m/s]
g exponent in crystal growth term [−]h numerical flux function [m−1]hGodu Godunov-flux [m−1]hLF Lax-friedrichs-flux [m−1]hLLF Local Lax-friedrichs-flux [m−1]hR Roe-flux [m−1]Ij jth mesh interval [m]K1 (fitted) constant for density [m3/kg]K2 (fitted) constant for density [m3/(kg]K3 (fitted) constant for density [m3/kg]KT (fitted) parameter for temperature dependence [K]KW (fitted) parameter for mass fraction dependence [−]kg growth rate constant [−]
xxiii
symbols (continued)
kb nucleation rate constant [m−3 s−1]
k(k)b growth rate constant of the k-th enantiomer [−]
kb,prim lumped constant for nucleation [ (kg m−3)−7/3
min K]
kb,sec rate constant for nucleation rate [m−3/min]kb,sec,0 pre-exponent factor for nucleation [m−3/min]kg,eff effective crystal growth constant [m/min]kg,eff,0 pre-exponent factor for growth [m/min]kv volume shape factor [−]m solute mass in the liquid [kg]m(k) solute mass in the liquid of k-th enantiomer [kg]m(p) solute mass in the liquid of preferred enantiomer [kg]m(c) solute mass in the liquid of counter enantiomer [kg]m mass flow rate [g s−1]
m(p)seeds mass flow rate of seeds of preferred enantiomer [kg/min]
m(k)in incoming mass flow rate of k-th enantiomer to the crystallizer [kg/min]
m(k)out outgoing mass flow rate of k-th enantiomer from the crystallizer [kg/min]
msat saturated mass [g]msolv mass of the solvent [g]mseeds mass flow of seeds [kg/min]min incoming mass flow rate to the crystallizer [kg/min]mout outgoing mass flow rate from the crystallizer [kg/min]min,c incoming mass flow rate of fresh solution [kg/min]
xxiv
symbols (continued)
mout,c outgoing stream of exhausted solution [kg/min]min,pipe incoming stream from the fines dissolution unit [kg/min]mL,out,pipe outgoing stream to the fines dissolution unit [kg/min]
m(p)S,out mass of preferred enantiomer [kg]
m(k)L,α solute mass in the liquid for k-th enantiomer in tank α [kg]
mwater mass of water [kg]
m(k)seeds,α mass flow of seeds for k-th enantiomer in tank α [kg/min]
m(k)L,in,α incoming mass flow rate to the crystallizer [kg/min]
m(k)L,out,α outgoing mass flow rate from the crystallizer [kg/min]
m(k)L,in,c,α incoming mass flow rate of fresh solution [kg/min]
m(k)L,out,c,α outgoing stream of exhausted solution [kg/min]
m(k)L,in,pipe,α incoming stream from the fines dissolution unit [kg/min]
m(k)L,out,pipe,α outgoing stream to the fines dissolution unit [kg/min]
m(c)S,out mass of counter enantiomer [kg]
mm minmod limiter function [−]N number of grid points [−]nsec exponent in secondary nucleation [−]Pr(p) productivity per unit volume [kg/(minmin3)]
Pr(p)α productivity per unit volume in tank α [kg/(minm3)]
Pu(p) purity [−]
Pu(p)α purity in tank α [−]
Pl legendre polynomial of order l [−]Q source term [m−1s−1]Qi mean value of source term [m−1s−1]Q+
nuc length based nucleation term [m−1s−1]
xxv
symbols (continued)
Q−diss dissolution rate [m−1s−1]
Q±agg length based aggregation term [m−1s−1]
Q±break length based breakage term [m−1s−1]
Qin inflow rate [m−3s−1]Qout outflow rate [m−3s−1]R gas constant [J/(mol K)]ri+ 1
2(t) upwind ratio of the consecutive flux gradients [−]
S supersaturation degree [−]S(k) supersaturation degree of k-th enantiomer [−]
S(k)α supersaturation degree for k-th enantiomer in tank α [−]
T temperature [K]Teq equilibrium temperature [min]t time [min]tend final time [min]ton on time of periodic seeding [min]tmax maximum time [min]toff off time of periodic seeding [min]tstd the time at which steady state is achieved [min]∆t time step [s]Vpipe volume of the pipe [m3]VL volume of the liquid in the crystallizer [m3]
Vpipe volumetric flow rate to the pipe [m3/min]
V volumetric flow rate [m3/min]
Vin inflow rate [m3/min]
Vout outflow rate [m3/min]
xxvi
symbols (continued)
VS overall volume of the solid phase [m3]Vc volume of the single crystal [m3]Vcrz volume of the crystallizer [m3]
Vexch volumetric exchange flow rate [m3/min]VL,α volume of the liquid in tank α [m3]
VL,α volumetric flow rate in tank α [m3/min]w mass fraction [−]wsat saturated mass fraction [−]w(k) mass fraction of k-th enantiomer [−]
w(k)α mass fraction for k-th enantiomer in tank α [−]
w(p) mass fraction of preferred enantiomer [−]w(c) mass fraction of counter enantiomer [−]weq mass fraction at equilibrium temperature [−]
w(k)eq mass fraction of k-th enantiomer at equilibrium temperature [−]
w(k) mass fraction of k-th enantiomer [−]wtot total mass fraction of preferred and counter enantiomers [−]Y (p) yield [−]
Y(p)α yield in tank α [−]
z particle size [m]∆z mesh size [m]∆zj width of mesh interval Ij [m]z(p) mean crystal size [m]zcrit critical size [m]z0 minimum crystal size [m]zmax maximum crystal size [m]z constant of initial CSD Eq. (3.69) [m]
xxvii
symbols (continued)
µ(p)0 zero moment: total number of crystals [m−1]
µ(p)1 first moment: total length of crystals [m]
ϕl local basis function of order l
ℑ(k)α numerical flux for the k-th enantiomer in tank α [−]
µi,α i-th moment of the CSD in tank α [−]ρL,α density of liquid phase in tank α [kg/m3]ρL,0,α density of fresh liquid at t = 0 in tank α [kg/m3]ψ selection (death) function for fines dissolution [−]φ flux limiting function [m−1]ε small number to avoid division by zero [−]ρc crystal density [g m−3]ρL density of liquid phase [kg/m3]ρL,0 density of fresh liquid at t = 0 [kg/m3]ρS density of solid phase [kg/m3]ρsolu density of solution [kg/m3]ρo density of water [kg/m3]ρf density of fresh solution [kg/m3]τpipe residence time in the dissolution unit [min]τL mean liquid phase residence time [min]τS mean solid residence time [min]τcryst ratio of crystallizer volume to the volumetric flow rate [min]τL,α mean liquid phase residence time in tank α [min]τS,α mean solid residence time in tank α [min]τpipe,α residence time in the pipe for tank α [min]σ constant of initial CSD Eq. (3.69) [m]ρL density of liquid phase [kg/m3]ρL,0 density of fresh liquid at t = 0 [kg/m3]
xxviii
1
Chapter 1
Fundamentals of Crystallization Process
This chapter introduces different crystallization processes which are widely used in chemi-
cal, pharmaceutical and food industries.
1.1 Crystallization
Crystallization is the process of solid crystals formation from a supersaturated homoge-
neous solution. Crystallization is a common process in nature and everyday life. Since
millions of years, it has played a vital role in the formation of stalactites within caves and
precious stones such as diamonds and sapphires. It is the best and cheapest method for the
production of pure solids from impure solutions. In chemical engineering, crystallization
process is one of the oldest unit operation. For several decades, the process is being used
and the production of sodium chloride is one of its examples. It is widely used in other
fields of chemical engineering as well. For example, the process is used in petro-chemical
industry for separation and purification of hydrocarbons, as well as for the manufacturing
of polymers and high valued chemicals and household products.
The crystallization process is driven by supersaturation of the solution [52]. This can be
achieved by cooling or evaporating the solvent. Normally, two processes are active in a
supersaturated solution, i.e., nucleation to form nuclei due to gathering of solute molecules
into clusters and the crystal growth in which the already existing crystals grows [59]. One
very important variable in industrial crystallizers is the crystal size distribution (CSD).
A control of crystal size and shape is important for achieving the desired goals and for
improving the product quality.
1.2 Types of Crystallization
Crystallization processes are either batch or continuous. The main difference between the
two is consistency of supersaturation and temperature. In batch operation both tempera-
2
ture and supersaturation cannot be kept constant at the same time, but this can be done
in the continuous crystallizers [33].
1.2.1 Batch Crystallization
Batch crystallization is an important unit operation for the production of high-value-added
chemicals at laboratory scale [33]. However, its use in industry is very low. Batch crys-
tallizer is also used if several products are to be crystallized in the same crystallizer. It is
an unsteady process and does not attain steady state [61]. Batch crystallizer is a closed
tank filled with a saturated solution. Afterwards, the solution is made supersaturated by
cooling the crystallizer but still the solution is free of particles. The crystallizer is then
seeded with seed crystals to promptly start the crystallization process. During the process,
new crystals are formed and existing crystals grow, thus, forming a CSD.
1.2.2 Continuous Crystallization
It is the process in which solution is continuously fed to the crystallizer and product is
continuously withdrawn. The continuous processes are usually very lengthy and are used
for the production of large amount of materials. These processes are operated under steady
state condition [70].
1.3 Enantiomers
Enantiomers are organic molecules which are mirror images of each other (chiral com-
pounds). They are stereoisomers just like our left and right hands whose molecules are
non-superimposable mirror images [61]. Figure 1.1 gives the illustration of amino acid
enantiomers as an example. Enantiomers are symmetric organic molecules with identical
chemical and physical properties but with different properties regarding metabolism.
3
Figure 1.1: Amino acid enantiomers [61].
A solution in which both the enantiomers are present in equal amounts is known as a
racemic solution.
1.4 Preferential Crystallization
Separation of enantiomers is an important process in chemical industry because a lot of bio-
organic molecules possess chiral properties. Mostly, one enantiomer has desired properties
while other has undesired. Due to its great importance in chemical and pharmaceutical
industries, there is a need to improve the performance of already established separation
processes [61].
To separate the enantiomeric mixtures, several methods can be applied like chromatog-
raphy, biological and non-biological asymmetric synthesis but one of the most efficient
technique used for separation and purification of enantiomers is preferential crystallization
[3, 11, 12, 13, 17, 18, 31]. In this method, one of the enantiomers is seeded in a super-
saturated racemate solution, resulting in a slight excess of the same isomer to make an
asymmetric environment, so that this isomer is crystallized out from the solution.
4
1.5 Ternary Phase Diagram
This diagram illustrates the principle of preferential crystallization process in which the
saturated solution with initial temperature Tcryst + ∆T is cooled down to Tcryst, see Figure
1.2. Thus, the solution becomes supersaturated. This process takes place in a metastable
zone in which no spontaneous nucleation occurs. In this, two enantiomers and a solvent
are mixed which is represented by point A. At this point the crystals of enantiomer E1 are
seeded. Due to supersaturated solution, the seeds grow and secondary nucleation starts. If
we assume that primary nucleation of enantiomer E2 does not occur, the process would end
at point M, representing the equilibrium point for enantiomer E1 only. But experiments
show that after certain time there is an induction of primary nucleation of E2 and therefore,
after very long time, the common equilibrium point, i.e., point E is achieved for E1 and
E2. At the end, both the enantiomers crystallize out at the same time, but the counter-
enantiomer crystallizes with a higher rate (the desired enantiomer has already been almost
crystallized).
Figure 1.2: Ternary phase diagram [61].
5
1.6 Historical Background and Motivation
Population balance equations (PBEs) are widely used for the simulation of crystallization
processes. These equations are similar to our well known mass and energy balance equa-
tions. They describe a balance law for the number of individuals of a population, such as
crystals, droplets, bacteria etc. These equations are generally important in almost all sci-
entific and engineering disciplines. For instance, biophysicists, pharmacists, food scientists,
chemical engineers, civil (environmental) engineers and many others make use of these bal-
ances for simulating different processes in their respective fields of research. Particularly,
in chemical engineering population balance models (PBMs) are employed for describing a
variety of dispersed processes namely, flocculation, crystallization, granulation, polymer-
ization, and combustion [69]. In their pioneer work, Hulburt and Katz [27] introduced
PBMs in chemical engineering and later on were fully studied by Randolph and Larson
[71]. PBMs are used to simulate different phenomena in the process, such as aggregation,
breakage, growth, nucleation, dissolution, as well as inlet and outlet streams. Mathemat-
ically, the PBEs of these models are nonlinear integro-partial-differential equations of the
number density function. In practical problems, the PBE is normally coupled to one or
more ODEs for the mass and energy balances of the liquid phase. These equations have
no analytical solutions except for few simple cases. Therefore, for most practical problems,
accurate and efficient numerical techniques are applied.
This fact has motivated researchers in this field to develop new and efficient algorithms for
solving population balance equations (PBEs), see for example Ramkrishna [70], Nicmanis
and Hounslow [57], and references therein. Several efficient schemes have been formulated
for the solution of PBEs, such as the quadrature method of moments (QMOM), the method
of characteristics (MOC), the Monte Carlo method (MCM), the finite difference methods
6
(FDMs), and the finite volume methods (FVMs).
Hulburt and Katz [27] were the first who applied the method of moments to PBMs and
pointed out the closure problem associated with this method. Afterwards, different meth-
ods were used to solve the closure problem raised by Hulburt and Katz [27] as considered by
Diemer and Olson [15]. The QMOM was introduced by McGraw [50] for modeling aerosol
evolution. In this method, quadrature approximation was used for solving integrals of PSD
and the required abscissas and weights were obtained by using the product difference (PD)
algorithm of Gordon [22]. Further work in this direction can be found in Fan et al. [20],
Marchisio [48, 49], McGraw [51], Qamar et al. [63], and references therein.
In method of characteristics, a new coordinate system is generated that reduces the PBE
into an ODE with some characteristic curves in the x-t plane [36], where t and x are used
for time and property coordinates. This method gives high accuracy for simple popula-
tion balance equations but for more complex problems the method does not work efficiently.
The Monte Carlo method gives good computational efficiency for complex stochastic PBMs
[70]. However, the computational cost of this method is generally high and accuracy is of
lower order.
In the past few decades, the finite difference methods (FDMs) were frequently used to solve
PBEs [36]. This is a domain discretization method in which the derivatives are replaced
by difference quotients so that the differential equation is represented by algebraic equa-
tions. Various discretization techniques of different accuracy orders have been suggested
by several researchers for PBMs [14, 21, 29, 30, 36, 47, 55].
7
The finite volume schemes (FVS) were initially used for gas dynamics and for the approx-
imation of PBEs, see Ma et al. [44, 45], Gunawan et al. [24], Qamar et al. [62], and
reference therein. The FVS refers to small volumes surrounding the nodal points in the
domain. The method has been used efficiently in different fluid flows and gas dynamics
applications. An advantage of using the FVS is that they can be applied even if the details
of physical characters and behaviors of the flow are not provided. Different researchers have
used these schemes to solve the problems in gas dynamics, astrophysical flows, multi-phase
fluid flows and detonation waves [5, 6, 28, 38, 40, 56, 77].
1.7 New Results
This work is related to the development and simulation of batch and continuous crystal-
lization models. Especially, models are derived for simulating batch and continuous enan-
tioselective preferential crystallization processes in single and coupled crystallizers. Such
processes are highly important in chemical and pharmaceutical industries. The effects of
nucleation, growth, fines dissolution, seeding strategies and residence time characteristics
on the CSD are analyzed.
For the first time, a TVB Runge-Kutta discontinuous Galekrin (DG) finite element method
is applied to solve batch crystallization model [65]. For comparison, the high resolution
finite volume scheme (HR-FVS) is applied to solve the same model [32, 59]. The schemes
satisfy the total variation bounded (TVB) property which assures the positivity of the
schemes, e.g., the non-negativity of CSD in the present case. The suggested methods have
capabilities to capture narrow peaks and sharp discontinuities of the CSD. The perfor-
mance of the DG-scheme can further be improved by inserting additional nodes in the
same solution element and, thus, avoids the expansion of mesh stencils which is normally
observed in high resolution finite volume scheme (HR-FVS). For that reason, the scheme
8
can be easily implemented up to boundary cells without compromising on accuracy and
efficiency. It was found that the proposed numerical schemes have the capability to solve
the given models more efficiently and accurately.
The above schemes were extended to simulate batch preferential crystallization of enan-
tiomers in single and double crystallizers equipped with fines dissolution units. The models
are further elaborated by considering the isothermal and non-isothermal conditions. Firstly,
the crystallization of preferred enantiomer is assumed to take place in a single crystallizer
equipped with a fines dissolution unit. The extracted solution is sieved by filters and
supposed to be free of bigger crystals. Thus, only tiny particles are withdrawn to the dis-
solution unit. To assure a crystal-free liquid exchange, it is assumed that all particles in the
dissolution loop are dissolved before re-entering back to the crystallizer by using a heat ex-
changer. Furthermore, before re-entering into the crystallizer, the liquid in the dissolution
loop is cooled down again. The breakage and agglomeration processes are not considered in
this study. Secondly, the model is extended for a coupled batch preferential crystallization
process with isothermal and non-isothermal conditions. In this setup, the crystallization of
two enantiomers is assumed to take place in two separate crystallizers which are coupled
through exchange pipes. There are two main advantages of considering coupled crystalliz-
ers. The first one is that both enantiomers are crystallized out simultaneously in separate
crystallizers. Secondly, due to the liquid exchange between the crystallizers, the growth
process enhances in both crystallizers and, thus, crystals of large mean size are obtained.
Both HR-FVS and DG-method are used to solve these models. The DG-method is applied
for the first time to simulate such models.
For the first time continuous preferential crystallization is investigated and the effects of
different seeding strategies and residence time characteristics are analyzed on the dynam-
9
ics of a Mixed Suspension Mixed Product Removal (MSMPR) crystallizer equipped with a
fines dissolution unit [66]. The fines dissolution is included as recycle streams around the
MSMPR crystallizer. Moreover, primary heterogeneous and secondary nucleation mecha-
nisms along with size-dependent growth rates are taken into account. The model is then
extended for a coupled continuous preferential crystallization process. In this setup, the
crystallization of two enantiomers is assumed to take place in two separate crystallizers
which are coupled through exchange pipes. To judge the quality of the process some goal
functions are used, such as purity, productivity, yield and mean crystal size of the preferred
enantiomer. Both HR-FVS and DG-method are applied to solve these models [65, 66]. Sev-
eral numerical case studies are carried out. These results could be used to find the optimum
operating conditions for improving the product quality and for reducing the operational
cost of continuous preferential crystallization. Altogether, the process appears to possess
large potential and deserves practical realization.
1.8 Layout of the Thesis
The remaining part of the thesis is organized as follows:
In chapter 2, the population balance model (PBM) is introduced and major phenomena
which include growth, nucleation, dissolution, breakage, aggregation, as well as inflow and
outflow of particles are explained. Furthermore, some goal functions are introduced, such
as purity, productivity, yield and mean crystal size of the preferred enantiomer to judge
the quality of the process. These goal functions give detailed information about the success
and potential of preferential crystallization processes.
Chapter 3 introduces the mathematical model of a batch crystallizer equipped with fines
dissolution unit. This dissolution produces an improvement in the crystal size and product
10
quality. A TVB Runge-Kutta discontinuous Galerkin finite element method is derived to
solve the model. For comparison, the HR-FVS is applied to the same model [59]. The
comparison verifies the robustness of the proposed methods.
Chapter 4 introduces preferential batch crystallization models for single and coupled crys-
tallizers. The same DG-method and HR-FVS are applied to solve these models. Several
case studies are carried out. The comparison verifies the efficiency and accuracy of the
proposed method.
In Chapter 5, a mathematical model is derived for studying the effects of different seeding
strategies and residence time characteristics on the dynamics of a MSMPR crystallizer
equipped with a fines dissolution unit [66]. A semi-discrete HR-FVS is employed for dis-
cretizing the derivatives with respect to the length coordinate. The resulting ordinary
differential equations (ODEs) are solved by a Runge-Kutta method of order four. After-
wards, a DG-method is implemented to solve the same model [65]. Several numerical case
studies are carried out. The results support process design and optimization.
In Chapter 6, a mathematical model is formulated for coupled continuous preferential crys-
tallization which incorporates fines dissolution unit with time-delay. The high resolution
finite volume scheme (HR-FVS) is applied to approximate the model.
Chapter 7 contains the conclusion of the entire research work and future recommendations.
Finally, Chapter 8 contains the references used in the thesis.
11
1.9 Publications Related to the Research Work
Journal papers
Qamar, S., Hussain, I., Morgenstern, A. S., Application of discontinuous Galerkin scheme
to batch crystallization models. Industrial and Engineering Chemistry Research, 50 (2010),
4113-4122.
Qamar, S., Elsner, M.P., Hussain, I., Morgenstern, A. S., Effects of seeding strategies and
different operating conditions on preferential continuous crystallizers. Chemical Engineer-
ing Science, 71 (2012), 5-17.
Mukhtar, S., Qamar, S., Hussain, I., Jan, A., A Quadrature method of moments for solv-
ing volume-based univariate and bivariate population balance models. Positive reviews are
received from the Brazilian Journal of Chemical Engineering, (2012).
Conference proceeding
Qamar, S., Mukhtar, S., Hussain, I., Morgenstern, A. S., An efficient numerical technique
for solving a batch crystallization model with fines dissolution. Proceedings of 4th Interna-
tional Conference on population balance modeling, September 15-17, 2010 Berlin Germany,
pp. 819-833.
12
13
Chapter 2
Population Balance Models
This chapter introduces the length based population balance model which incorporates the
phenomena of crystal nucleation, growth, aggregation, breakage, dissolution, and inflow
and flow streams. The model provides an essential background for understanding the
crystallization process in forth coming chapters. Moreover, the goal functions, such as
purity, productivity, yield, and mean are introduced to analyze the product quality of
preferential crystallization in Chapters 4, 5 and 6.
2.1 Population Balance Models
In this study a well mixed system is considered, i.e., all the particles are homogeneously
distributed in the whole region of study. Thus, the population balance model may be
integrated out over all points in space and the resulting population balance equation (PBE)
describes the dynamics of the size-distribution function F := F (t, z) ≥ 0 of particles of
size z > 0 at time t ≥ 0. The population balance equation (PBE) is expressed as [30, 71]
∂F (t, z)
∂t= − ∂[G(t, z)F (t, z)]
∂z+ Q+
nuc(t, z) −Q−diss(t, z)
+ Q±agg(t, z) + Q±
break(t, z) + Qin(t, z) −Qout(t, z) , (t, z) ∈ R2+, (2.1)
where R+ = (0,∞). The term G(t, z) represents the size-dependent growth rate, Q+nuc(t, z)
is responsible for the nucleation of particles, and Q−diss(t, z) accounts for the dissolution of
particles. The terms Q±agg and Q±
break, represent the birth and death of particles during
the aggregation and breakage processes, respectively. The last two terms Qin(t, z) and
Qout(t, z) on the right hand side of above equation represent the inflow and outflow rates
of particles from the given system, respectively. Mathematically, the inflow and outflow
streams are defined as
Qin(t, z) =Vin
Vcrz
Fin(t, z), Qout(t, z) =Vout
Vcrz
Fout(t, z), (2.2)
where Vin and Vout represent the volumetric inlet and outlet flow rates from the system of
volume Vcrz, while Fin(t, z) and Fout(t, z) are the number densities of incoming and outgo-
14
ing particles.
Generally, a continuous crystallization process has inflow and outflow of particles, while
the batch crystallization process has no inflow or outflow of particles, therefore the last
two terms on the right hand side of Eq. (2.1) does not appear in the batch crystallization
process. This work is concerned with both batch and continuous crystallization processes.
2.2 Growth and Nucleation
During the growth process the existing particles of the dispersed system grow by consum-
ing the solute mass from the solution. Therefore, the total number of particles remains
the same but volume continuously increases. At microscopic level the particle growth rate
is unpredictable but is predictable on macroscopic level. Generally, the growth rate is a
complex phenomenon that is not completely understood when one is dealing with size-
dependent growth rates. Especially, it is difficult to measure the size-dependent growth
rate during the experiments. The adsorption layer theory and mass transfer theory are
usually used for modeling the size-dependent growth rate of particles [78].
The nucleation process introduces new particles in the system by mixing two or more non-
particulate matters. Due to this process, the population of particles increases in the system.
Nucleation can be divided into two categories, such as primary and secondary nucleations,
and the primary nucleation is further subdivided into two categories namely, homogeneous
and heterogeneous nucleations. Homogeneous nucleation occurs in the particle-free liquid
phase. Large amount of small particles can be produced with this type of nucleations. It is
commonly observed in salting out, in precipitation, and in crystallization. The second type
of primary nucleation is heterogeneous nucleation which happens due to several reasons,
for example tubes are not cleaned due to any previous particulate process or any pore
15
in the tube has some dust inside, etc. The secondary nucleation is the heterogeneous
nucleation induced by existing particles, for example due to seeding of batch crystallizer.
The nucleation term is defined as
Q+nuc(t, z) = Fnuc(t, z)B0(t).
Here, Fnuc(t, z) denotes the number density function of nuclei and B0(t) represents the rate
of nucleation. In the case of batch crystallization Fnuc(t, z) = δ(z − z0), where δ is the
Dirac delta distribution and z0 is the minimum crystal size. This means that nucleation
introduces only small nuclei of size z0 in the system. In other words, nucleation terms acts
like a point source at the left boundary of the size domain.
2.3 Dissolution
Generally, particles are not stable below a certain size, usually called critical size. Small
particles with negative growth rate will finally become smaller than the critical size and,
thus, disappear from the population. This phenomenon can be described as
Q−dis(t, z) = D−
dis F (t, z) , (2.3)
where D−dis represents the dissolution rate. If the particles are assumed to disappear im-
mediately when reaching the critical size, the dissolution rate has to be infinite.
2.4 Aggregation
The merging of two or more particles to make a bigger one is called aggregation. In this
process the total volume remains conserved while the number of particles reduces. It takes
place in a variety of processes that includes crystallization, production of dry powders, and
fluidized beds etc. [70].
16
2.5 Breakage
In this process, the bigger particles split into two or more pieces. This results in a rapid
increase of particles. It has a wide range of applications in crystallization, granulation and
environmental sciences. It is obvious to see that the total number of particles increases in
a breakage process while the total volume of the particles remains the same throughout
the process [70].
2.6 Goal Functions
In order to judge the quality of continuous preferential crystallization some goal functions
can be used, such as purity, productivity, yield, and mean crystal size of the preferred
enantiomer [61]. These goal functions give detailed information about the success and
potential of continuous preferential crystallization discussed in Chapters 4, 5 and 6.
Purity: It is a ratio of the mass of preferred enantiomer to the sum of the masses of
preferred and counter enantiomers. It can be described by the following equation
Pu(p) =m
(p)S,out
m(p)S,out + m
(c)S,out
, (2.4)
where m(p)S,out and m
(c)S,out are the masses of preferred and counter enantiomers, respectively.
Productivity: The productivity of a continuous crystallization process can be defined as
the mass flow of solid produced per unit size. It can be described by the following equation
Pr(p) = mass flow of solid produced (preferred) per unit volume of fluid =m
(p)S,out − m
(p)seeds
VL
,
(2.5)
where m(p)S,out, m
(p)seeds and VL denote the mass of preferred enantiomer, mass flow of seeds
and volume of the liquid in the crystallizer, respectively.
17
Yield: The yield can be defined as the ratio of the mass flow of solid target particles (p)
produced over mass flow of this enantiomer in the feed solution introduced
Y (p) =m
(p)S,out − m
(p)seeds
m(p)in.c
, (2.6)
where m(p)S,out, m
(p)seeds and m
(p)in.c denote the mass of preferred enantiomer, mass flow of seeds
and incoming mass of preferred enantiomer, respectively.
Mean crystal size: Besides the purity, the crystal size also plays an important role
in most industrial applications. For that reason, the mean crystal size of the preferred
enantiomer is instructive. It is defined as:
z(p) =µ
(p)1
µ(p)0
, (2.7)
where µ(p)1 is total length of crystals and µ
(p)0 is total number of crystals.
18
19
Chapter 3
Batch Crystallization
Batch crystallization is an important unit operation for the production of high-value-added
chemicals at laboratory scale [33]. Due to its widespread use, finding an efficient working
strategy is mandatory to enhance the product quality and process efficiency. Therefore, it
is essential to understand the process itself and the impact of process variables. The math-
ematical modeling of the process is helpful to attain the required goals and to investigate
the effect of different operating conditions. The observed data can be used to improve the
product quality and to optimize the operational cost.
In this chapter, a new class of scheme is introduced for solving the batch crystallization
models. In 1973, Reed and Hill [73] for the first time introduced the discontinuous Galerkin
(DG) finite element method for solving hyperbolic equations. Since then, various DG meth-
ods were developed for solving hyperbolic, elliptic and parabolic problems [4, 26, 68]. The
technique was further developed by Cockburn and his co-authors by introducing Runge-
Kutta DG-scheme in a series of papers for solving hyperbolic conservation laws [8, 9, 10].
The scheme employs DG-scheme in space-coordinates that converts the given partial dif-
ferential equation (PDE) to a system of ODEs which is computed by using high order
Runge-Kutta method. The scheme satisfies the total variation bounded (TVB) property
that assures the scheme positivity, for instance, the non-negativity of crystal size distribu-
tion (CSD) in the present case. The same DG-scheme is implemented for solving the batch
crystallization models in this dissertation. Besides different classes of schemes, DG-schemes
are more stable and high-order accurate, are capable to handle complex geometries and
irregular meshes with hanging nodes, and can incorporate arbitrary degrees of polynomial
approximations in different elements [8, 68]. The test cases verify the good performance of
the method. For authentication, the numerical results of the suggested method are com-
pared with the flux-limiting HR-FVS of Koren [32, 59]. It was observed that the proposed
method produces better results and is more efficient as compared to other flux-limiting
20
schemes for solving batch crystallization models [59, 64].
3.1 Batch Crystallization Model
Here, a mathematical model is presented for batch crystallizer connected with a fines
dissolution unit. This dissolution process may give an improvement in the crystal size and,
hence, the product quality. The simplified sketch of a batch unit is given in Figure 3.1.
LoopFines Dissolution
dissolution pipe
Settling ZoneAnnular
Figure 3.1: Batch crystallizer with fines dissolution [61].
Assumptions
We make use of the following assumptions while considering this model:
• Ideally mixed crystallizer.
• Constant overall volume.
• Size dependent/independent growth rate.
• No breakage, attrition and agglomeration.
• Isothermal and non-isothermal conditions.
21
• All fines dissolve completely in the dissolution pipe.
• Before liquid comes back to the crystallizer from the pipe, it is cooled again to avoid
negative effect of warm liquid on crystals of the crystallizer.
The rate of change in CSD is described by the PBE [27]
∂F (t, z)
∂t= − ∂[G(t, z)F (t, z)]
∂z+ Q(t, z) , (3.1)
with initial and boundary conditions of the form
F (0, z) = F0(z) , F (t, z0) =B0(t)
G(t, z0). (3.2)
Here, F0(z) represents the CSD of seed crystals added at the beginning of the batch process,
and Q(t, z) is any sink or source term. In addition to the solid phase, a balance law for the
liquid phase is also required in the crystallization process. The expressions for the liquid
balance law, growth rate, and nucleation rate will be presented in the test problems. In
the present case, the term Q(t, z) represents the dissolution of small nuclei below a certain
critical size and is defined in Problem 5.
3.2 Numerical Schemes
In this section, the numerical solution of Eq. (3.1) is presented by using the discontinuous
Galerkin (DG) finite element method and high resolution finite volume scheme (HR-FVS).
3.2.1 Implementation of Discontinuous Galerkin finite elementmethod
For the numerical approximation of PBE in Eq. (3.1), a TVB Runge-Kutta DG-method
[8, 68] is implemented. In this approach, the DG discretization is adopted in the size
22
variable only. The time derivatives are discretized by using a TVB Runge-Kutta method.
Let us define
f(G, F ) = G(t, z)F (t, z). (3.3)
Then, the PBE in Eq. (3.1) can be re-written as
∂F (t, z)
∂t= − ∂[f(G, F )]
∂z+ Q(t, z). (3.4)
To discretize the size-domain [z0, zmax], we proceed as follows. For j = 0, 1, 2, ....N , let zj+ 12
be the cells partitions, Ij =(
zj− 12, zj+ 1
2
)
be the domain of cell j, ∆zj = zj+ 12− zj− 1
2be
the width of cell j, and I =⋃
Ij are the particle size range partitions. We are looking for
an approximation Fh(t, z) to F (t, z) so that for all t ∈ [0, tmax], Fh(t, z) is belonging to the
space of finite dimensions
Vh =
v ∈ L1(I) : v|Ij∈ P k(Ij), j = 0, 1, 2, ....N
, (3.5)
where P k(Ij) stands for the polynomials space in I of at most k degree. Note that in Vh,
discontinuities of the functions are allowed at the cell interface zj+ 12. To find the numerical
solution Fh(t, z), a weak formulation is needed which is usually obtained by multiplying
Eq. (3.4) with a smooth function v(z) and by integrating over the interval Ij. After using
integration by parts, the weak formulation becomes
∫
Ij
∂F (t, z)
∂tv(z)dz = −
(
f(Gj+ 12, Fj+ 1
2)v(zj+ 1
2) − f(Gj− 1
2, Fj− 1
2)v(zj− 1
2))
+
∫
Ij
(
f(G, F )∂v(z)
∂z+ Q(t, z)v(z)
)
dz . (3.6)
Selecting Legendre polynomials (LPs), Pl(z), of order l as local basis functions is one way
to formulate Eq. (4.35). Here, the L2-orthogonality property of LPs can be used, such as
1∫
−1
Pl(s)Pl′(s)ds =
(
2
2l + 1
)
δll′ . (3.7)
23
For each z ∈ Ij, the solution Fh is defined as
Fh(t, z) =k
∑
l=0
F(l)j ϕl(z) , (3.8)
where
ϕl(z) = Pl (2(z − zj)/∆zj) . (3.9)
It can be easily verified that
(
1
2l + 1
)
F(l)j (t) =
1
∆zj
∫
Ij
Fh(t, z)ϕl(z)dz . (3.10)
Due to above Legendre polynomials as local basis functions, the test function ϕl ∈ Vh has
to be used instead of the smooth function v(z) and the approximate solution Fh in place
of the exact solution F . Moreover, the function Fj+ 12
= F (t, zj+ 12) is not known at the cell
interface zj+ 12. Therefore, the flux f(G, F ) has to be approximated by a numerical flux
that depends on two values of Fh(t, z), i.e.,
f(Gj+ 12, Fj+ 1
2) ≈ hj+ 1
2= h(Gj+ 1
2, F−
j+ 12
, F+j− 1
2
) . (3.11)
Here,
F−
j+ 12
= Fh(t, z−
j+ 12
) =k
∑
l=0
F(l)j ϕl(zj+ 1
2) , F+
j− 12
= Fh(t, z+j− 1
2
) =k
∑
l=0
F(l)j ϕl(zj− 1
2) . (3.12)
By utilizing the above expressions, the weak formulation in Eq. (4.37) simplifies to
dF(l)j (t)
dt= − 2l + 1
∆zj
(
hj+ 12ϕl(zj+ 1
2) − hj− 1
2ϕl(zj− 1
2))
+2l + 1
∆zj
∫
Ij
(
f(G, Fh)dϕl(z)
dz+ Q(t, z)ϕl(z)
)
dz . (3.13)
According to Eqs. (3.2) and (4.41), the initial data for this equation is given as
F(l)j (0) =
2l + 1
∆zj
∫
Ij
F0(z)ϕl(z)dz . (3.14)
24
The next step is to select the suitable numerical flux function h. The scheme is monotone
if the flux function h(G, a, b) is consistent, h(G, F, F ) = f(G, F ), and fulfills the Lipschitz
condition [41, 80]. In the literature, the following numerical fluxes are normally used
satisfying the above properties [38, 41, 80].
(i) The Godunov-flux:
hGodu(G, a, b) =
mina≤s≤b f(G, s) if a ≤ b ,maxa≥s≥b f(G, s) if a > b .
(3.15)
(ii) The Lax-Friedrichs-flux:
hLF (G, a, b) =1
2[f(G, a) + f(G, b) − C(b − a)], (3.16)
C = maxinf F (0)(z)≤s≤sup F (0)(z)
|f ′(G, s)| . (3.17)
(iii) The Local Lax-Friedrichs-flux:
hLLF (G, a, b) =1
2[f(G, a) + f(G, b) − C(b − a)] , (3.18)
C = maxmin (a,b)≤s≤max (a,b)
|f ′(G, s)| . (3.19)
(iv) The Roe-flux with entropy fix:
hR(G, a, b) =
f(G, a), if f ′(s) ≥ 0 for s ∈ [min(a, b), max(a, b)] ,f(G, b), if f ′(s) ≤ 0 for s ∈ [min(a, b), max(a, b)] ,hLLf (G, a, b) otherwise .
(3.20)
The above fluxes were tested and all of them produced the same results for considered test
problems. In this chapter, the local Lax-Friedrichs flux was considered in all test problems.
The 10th order Gauss-Lobatto quadrature formula was used for approximation integrals
on the right-hand-side of Eq. (3.13).
A limiting procedure is required to achieve local maximum principle with respect to the
means. For that reason, it is required to modify the interface values F±
j± 12
in Eq. (4.42) by
some local projection limiter. To this end, Eq. (4.43) can be written as [8, 9]
F−
j+ 12
= F(0)j + Fj , F+
j− 12
= F(0)j − Fj , (3.21)
25
where
Fj =k
∑
l=1
F(l)j ϕl(zj+ 1
2) , Fj = −
k∑
l=1
F(l)j ϕl(zj− 1
2). (3.22)
Next, Fj and Fj can be modified as
F(mod)j = mm
(
Fj, ∆+F(0)j , ∆−F
(0)j
)
, F(mod)j = mm
(
Fj, ∆+F(0)j , ∆−F
(0)j
)
, (3.23)
where ∆± = ±(Fi±1 − Fi) and mm is the minmod function which is given as
mm(a1, a2, a3) =
s · min1≤i≤3
|ai| if sign(a1) = sign(a2) = sign(a3) = s ,
0 otherwise .(3.24)
Then, Eq. (3.21) modifies to
F−(mod)
j+ 12
= F(0)j + F
(mod)j , F
+(mod)
j− 12
= F(0)j − F
(mod)j , (3.25)
and Eq. (4.42) is replaced by
hj+ 12
= h(Gj+ 12, F
−(mod)
j+ 12
, F+(mod)
j− 12
) . (3.26)
This limiter corresponds to adding the minimum amount of numerical diffusion while pre-
serving the stability of the scheme. The DG-method combined with the above stated slope
limiter is proven to be stable [23]. Finally, a Runge-Kutta method is required for solving
the ODE-system. Let us rewrite Eq. (3.13) in a compact form as
dFh
dt= Lh(t, Fh) . (3.27)
Then, the r-order TVB Runge-Kutta method can be used to approximate Eq. (3.27)
(Fh)k =
k−1∑
l=0
[
αkl(Fh)(l) + βkl∆tLh((Fh)
(l), tn + dl∆t)]
, k = 1, 2, · · · , r , (3.28)
where based on Eq. (4.45)
(Fh)(0) = (Fh)
n , (Fh)(r) = (Fh)
n+1 . (3.29)
26
For the second order TVB Runge-Kutta method the coefficient are given as [8]
α10 = β10 = 1 , α20 = α21 = β21 =1
2, β20 = 0; d0 = 0 , d1 = 1 . (3.30)
While, for the third order TVB Runge-Kutta method the coefficient are given as
α10 = β10 = 1 , α20 =3
4, β20 = 0 , α21 = β21 =
1
4, α30 =
1
3,
β30 = α31 = β31 = 0 , α32 = β32 =2
3; d0 = 0 , d1 = 1 , d2 =
1
2. (3.31)
To guarantee stability and numerical convergence of the proposed method, the time step
is chosen according to the following Courant-Friedrichs-Lewy (CFL) condition [8, 41]
∆t ≤(
1
2k + 1
)
min(∆zj)
max(|f ′(G, F )|) , (3.32)
where k = 1, 2 for the second and third order methods, respectively. This time step is
adaptive which reduces for the case of large variations (large slopes) in the solution and
increases otherwise.
Boundary conditions: Consider the boundary at z− 12
= z0. Since, G(t, z) ≥ 0, the left
boundary condition described by Eq. (3.2) can be formulated as [8, 9]
F−
− 12
(t) =B0(t)
G(t, z0), F
(mod)0 = mm
(
F0, ∆+F(0)0 , 2
(
F(0)0 − B0(t)
G(t, z0)
))
,
F(mod)0 = mm
(
F0, ∆+F(0)0
)
. (3.33)
Outflow boundary conditions were used on the right end of the domain,∂F
(l)h
∂z
∣
∣
∣
∣
z=zmax
= 0.
3.2.2 Implementation of Finite Volume Scheme
In this section, the high resolution finite volume scheme of Koren [32, 59] is applied for
discretizing the derivative of the length coordinate in Eq. (3.1).
Before applying the finite volume scheme, we first discretize the computational domain
which is the crystal size z in the present case. Let N be a large integer, and denote by
27
(zi− 12), i = 1, 2, · · · , N + 1 the partitions of cells in the domain [z0, zmax], where z0 is the
minimum and zmax is the maximum crystal length of interest. For each i = 1, 2, · · · , N , ∆z
represents the cell width, the points zi refer to the cell centers, and the points zi± 12
denote
the cell boundaries. The integration of Eq. (3.1) over the cell Ωi =[
zj− 12, zj+ 1
2
]
yields the
following cell centered semi-discrete finite volume schemes for fi± 12
= (GF )i± 12
∫
Ωi
∂F (t, z)
∂tdz = − (fi+ 1
2(t) − fi− 1
2(t)) +
∫
Ωi
Q(t, z)dz. (3.34)
Let Fi and Qi represent the mean values of the number density and source term in each
cell Ωi, i.e.
Fi(t) =1
∆z
∫
Ωi
F (t, z)dz , Qi(t) =1
∆z
∫
Ωi
Q(t, z)dz . (3.35)
Then, Eq. (3.34) can be written as
dFi(t)
dt= −
fi+ 12(t) − fi− 1
2(t)
∆z+ Qi(t), i = 1, 2, · · · , N . (3.36)
Here, the computational domain is divided in N mesh points. The performance of the
finite volume discretization is found by the way in which the cell face fluxes are calculated.
Assuming that the flow is in the positive z−direction, a first order accurate upwind scheme
is found by taking the backward differences [59]:
fi+ 12(t) = (G(t)F (t))i , fi− 1
2(t) = (G(t)F (t))i−1 . (3.37)
To get high order accuracy of the scheme, one has to use a better approximation of the
cell interface fluxes. According to the high resolution finite volume scheme of Koren [32]
the flux at the right boundary zi+ 12
is approximated as
fi+ 12(t) = fi(t) +
1
2φ(ri+ 1
2(t))(fi(t) − fi−1(t)) . (3.38)
Similarly, one can approximate the flux at the left cell boundary. The flux limiting function
φ is defined as [32]
φ(
ri+ 12(t)
)
= max
(
0, min (2ri+ 12(t), min (
1
3+
2ri+ 12(t)
3, 2))
)
. (3.39)
28
Here, ri+ 12(t) is the so called upwind ratio of the consecutive flux gradients:
ri+ 12(t) =
fi+1(t) − fi(t) + ε
fi(t) − fi−1(t) + ε, (3.40)
where ε is a small number to avoid division by zero. It has been observed by Koren [32]
that this method is second order accurate in the coordinates of size. Therefore, the second
order ODE-solver is required for the overall size-time accuracy of the scheme. In this work,
we have used the same TVB Runge-Kutta method as discussed above. In the case of fines
dissolution with time-delay, the residence time in the pipe was taken as an integer multiple
of the time step. This facilitates to keep the old values in memory and avoids the linear
interpolation. The same procedure was also used in the DG-scheme.
The current HR-FVS is not applicable up to the boundary cells because it needs values of
the cell nodes which are not present. To overcome this problem, the first order approx-
imation of the fluxes is used at the interfaces of the first two cells on the left-boundary
and at the interfaces of the last cell on the right-boundary. At the remaining interior cell
interfaces, the high order flux approximation of Eq. (3.38) is used. It should be noted that,
the first order approximation of the fluxes in the boundary cells does not effect the overall
accuracy of the proposed high resolution scheme [32].
3.3 Test Problems
In this section, some case studies are presented. For authentication, the numerical results
of the proposed method are compared with those obtained from the HR-FVS of Koren
[32, 59, 64]. Except the test problem 5, the source term Q(t, z) is set equal to zero. The
linear basis functions are used in all test problems for the DG-scheme which gives second
order accurate results in the length coordinate. The ODE-system was solved by a third-
29
order Runge-Kutta method given by Eqs. (3.28) and (3.31). The adaptive time step was
calculated according to Eq. (3.32) in all test problems.
Problem 1
The initial data are given as
F (0, z) =
0.0 if z ≤ 2.0 µm ,1010 if 2.0 µm < z ≤ 10 µm ,0.0 if 10 µm < z ≤ 18 µm ,
1010cos2(π(z−26)64
) if 18 µm < z ≤ 34 µm ,0.0 if 34 µm < z ≤ 42µm ,
1010
√
1 − (z−50)2
64if 42 µm < z ≤ 58 µm ,
0.0 if 58 µm ≤ z ≤ 66 µm ,
1010e−(z−70)2
2σ2 if 66 µm < z ≤ 74 µm ,0.0 if z > 74 µm .
(3.41)
Here, G = 0.1 µm/s and B0 = 0 s−1. The analytic solution for F (0, z) = F0(z) is given as
F (t, z) = F0(Gt−z). The maximum crystal length is 100µm that is divided into 100 equal
sub-intervals and the simulation time is 100 seconds. The first distribution in the initial
data represents a square step function of width 8µm. This step function is considered
for testing the ability of the schemes in resolving sharp discontinuities. The second one
is a cosine-squared function of width 16µm representing a smooth behavior. The third
one denotes a semi ellipse of width 16 µm which is challenging due to its combined rapid
and slow variations in gradients. The last distribution is a narrow Gaussian distribution
with σ = 0.778∆z for testing the capability of the scheme in resolving sharp peaks. Figure
3.2 shows a comparison of the DG-method with the finite volume schemes. The figure
shows that the DG-method resolves the jump discontinuities and gives a better approxi-
mation of the peak of the Gaussian function as compared to Koren’s scheme. Moreover,
the DG-scheme captures the right discontinuities more efficiently as compared to the left
discontinuities. The first order upwind scheme gives a smeared solution. Table 3.1 presents
30
the absolute errors and CPU times of the schemes. It can be seen that the DG-scheme
gives less error in the solution and its computational cost is comparable to Koren’s scheme.
Figure 3.3 shows absolute errors and CPU times of the DG-scheme at different numbers
of grid points. Finally, Figure 3.4 gives a comparison of different orders of DG-schemes.
There is no drastic improvement in the solution with high order basis functions. This is
due to the more restricted minmod limiter which produces wiggles and more smearing in
the solution of high order basis functions [9]. However, the accuracy of the scheme can be
improved by using WENO limiters [9, 67].
Table 3.1: Problem 1: L1-errors and CPU times.Scheme L1-error CPU (s)DG-scheme 3.57 × 1010 0.046Koren-scheme 4.39 × 1010 0.045First-order FVS 1.32 × 1011 0.039
Problem 2
In this problem the numerical scheme is tested for a stiff nucleation phenomenon. Assume
that a stiff nucleation phenomenon takes place at the minimum crystal size (z0 = 0) as a
function of time [42]
F (t, 0) = 100 + 106 exp(−104(t − 0.215)2) . (3.42)
The crystal size and time ranges are given as 0 ≤ z ≤ 2.0 mm and 0 ≤ t ≤ 0.5 min,
respectively. The initial CSD is taken as
F (0, z) =
100 for 0.4 mm ≤ z ≤ 0.6 mm,0.01 elsewhere .
(3.43)
31
0 20 40 60 80 1000
2
4
6
8
10
12x 10
9
length [µm]
CS
D [1
/µm
]exact DG−scheme FVS first order
Figure 3.2: Problem 1: Size-independent growth at t = 100 s.
0 1000 2000 3000 40000
1
2
3
4
5
6
7
number of grid points
CPU time (s)
absolute error ( × 1010 )
Figure 3.3: Problem: Errors and CPU times of DG-scheme.
32
0 20 40 60 80 1000
2
4
6
8
10
12x 10
9
length [µm]
CS
D [1
/µm
]
DG−schemes of different ordersexact 2nd order 3rd order 4th order
Figure 3.4: Problem 1: Comparison of DG-schemes of different orders .
Here, a constant growth rate, G = 1.0 mm/min, is considered. The analytical solution is
given as [42]
F (t, z) =
100 + 106 exp(−104((Gt − z) − 0.215)2) for 0.0 ≤ z ≤ Gtmm ,100 for 0.4 mm ≤ z − Gt ≤ 0.6 mm,0.01 elsewhere .
(3.44)
In the solution, a square step discontinuity and a narrow peak which is originated from
nucleation move along the propagation path-line, z = z0 +Gt. The numerical test is carried
out on 200 grid points. The numerical results are shown in Figure 3.5. It can be observed
that the DG-method gives a better approximation of the solution than the Koren’s scheme.
Once again, the scheme shows better performance in resolving right discontinuities. The
stiff nucleation at the left boundary, which produces a sharp peak and a second step profile,
makes this problem much harder than the previous problems. The absolute error in the
numerical solution of the DG-scheme is equal to 1.37×104, while in the Koren scheme it is
equal to 1.83 × 104. Moreover, the simulation times of the DG-scheme and Koren scheme
were found to be 0.063 and 0.060 seconds, respectively.
33
0 0.5 1 1.5 210
−2
100
102
104
106
length [mm]
CS
D [1
/mm
]
exactDG−schemeFVS
Figure 3.5: Problem 2: Size-independent growth at t = 0.5 min.
Problem 3
This problem is selected for verifying the accuracy and efficiency of the DG-scheme for a
real batch process. The initial CSD is given as [45, 46]
F (0, z) =
−3.48 × 10−4z2 + 0.136z − 13.3 , if 180.5 µm ≤ z ≤ 210.5 µm ,0 , elsewhere .
(3.45)
A balance law for the liquid phase is needed which takes into account the depletion of
material from the solution due to crystal growth. In this problem, we take as variable
concentration instead of mass
dC(t)
dt= −3ρc
∞∫
0
z2 G(t, z) F (t, z) dz , C(0) = C0 , (3.46)
where C denotes the solute concentration in the solution, ρc is the crystal density and
the volume shape factor is denoted by kv so that the volume of a crystal with length z is
kvz3. The kinetic parameters reported for the crystallization of potassium nitrate (KNO3)
34
crystals were used as given in Table 3.2. The growth rate is given as [2, 53]
G(t, z) = kg [S(t) − 1]g (1 + aASLz)dASL , (3.47)
where kg is the growth rate constant. The exponent g denotes the growth order, while aASL
and dASL are constant given in Table 3.2. Moreover, S(t) denotes the supersaturation of
the dissolved component
S(t) =C(t)
Csat(T ). (3.48)
The saturated concentration quantifying the amount of solute per gram of the solvent is
given as [45]
Csat(T ) = 1.721 × 10−4T 2 + 5.88 × 10−3T + 0.1286 , (3.49)
where the temperature T follows an exponentially decaying trajectory of the form
T (t) = 32 − 4(1 − e−t
18600 ) . (3.50)
The nucleation rate is described as [2, 53]
B0(t) = kb [S(t) − 1]b µ3(t) , (3.51)
where kb and b are the nucleation rate constant and the nucleation order respectively, given
in Table 3.2. The third moment µ3 is defined as
µ3(t) =
∞∫
0
z3F (t, z) dz . (3.52)
The problem was solved by using DG method and Koren’s scheme on 2200 mesh points.
The resulting CSDs after 103 seconds are shown in Figure 3.6. The numerical solutions
of both schemes are in good agreement. However, a discrepancy is visible in the part
of distribution coming from the nucleation phenomenon. The reason is the first order
35
approximation of the Koren’s scheme in the first two left boundary cells. A disadvantage
of the Koren’s scheme is that it cannot be used in these boundary cells. As a result,
use of first order scheme introduces numerical errors in the solution that propagate to
the neighboring cells with the passage of time. However, the DG-scheme is free of such
limitations and, hence, gives better approximation of the solution in the boundary cells.
The absolute error in the mass balances of the DG-scheme is found to be 0.9× 10−4, while
in the Koren’s scheme it is equal to 2.4× 10−3. Moreover, the simulation times of DG and
Koren schemes are 88.6 and 85.4 seconds, respectively.
Table 3.2: Parameters for Problem 3Description Symbol Value UnitMaximum crystal size zmax 1100 µmMesh size ∆z 0.5 µmSimulation time t 1000 sNumber of grid points N 2200 −Growth rate constant kg 1.16 × 102 µm
s
Growth rate exponent g 1.32 −Nucleation rate constant kb 4.64 × 10−7 1
µm3s
Nucleation rate exponent b 1.78 −Volume shape factor kv 1.0 −Initial concentration C(0) 0.493 gDensity of crystals ρc 2.11 × 10−12 g
µm3
Constant (Eq. (3.47)) aASL 0.1 µm−1
Constant (Eq. (3.47)) dASL 1.0 −
Problem 4
The aim of this test problem is to illustrate the applicability of the DG-scheme for the case
of discontinuous crystal growth rate. Here, the simulation of potassium sulfate (K2SO4 −
H2O) is considered. The initial seed distribution is taken as [42, 46]
F (0, z) =
5.472 × 107 , if 5.0 × 10−4m ≤ z ≤ 6.0 × 10−4 ,0 , elsewhere .
(3.53)
36
100
101
102
103
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
length [µm]
CS
D [1
/µm
]
initial CSDDG−schemeFVS
Figure 3.6: Problem 3: Size-dependent growth at t = 1000 s.
The size range of interest is 5 × 10−5 m ≤ z ≤ 2 × 10−3 m and the final simulation time is
180 minutes. The growth rate, that replaces Eq. (3.47), is described as
G(t, z) =
86.4 exp(
− 40400R(T+273.15)
)
S2(t) (1 + 2(z × 106)2/3) , if z ≤ 7 × 10−4 m,
G(700)(
z×106
700
)
, otherwise ,
(3.54)
where R = 8.314 J K−1 mol−1 is the universal gas constant and S(t) is given by Eq. (3.48).
The nucleation rate, that replaces Eq. (3.51), is given as
B0(t) =
1.56 × 109 exp (−9300/R(T + 273.15)) S(t) µ3(t) , if 0.002 ≤ V (t) ≤ 0.02 ,
1.56 × 109 exp (−9300/R(T + 273.15)) S(t)√
0.02µ3(t) , otherwise .
(3.55)
Here, µ3 is given as
µ3(t) = ρc
∞∫
0
z3kv(z)F (t, z) dz , (3.56)
and ρc = 2660.0 kg/m3. The volume shape factor is defined as
kv(z) =
0.898 exp (168z0.5 − 8234z) , if z ≤ 1 × 10−4 m,4.460 exp (−0.0797z0.5 + 676z) , otherwise .
(3.57)
37
The saturated concentration quantifying solute mass per gram of solvent is given as
Csat(T ) = 6.0 × 10−6T 2 + 2.3 × 10−3T + 6.66 × 10−2 . (3.58)
The temperature profile used to maintain a constant supersaturation (C −Csat = 0.00732)
is given as
T (t) = 70 − 45e−0.008356(180−t) . (3.59)
Finally, the concentration balance in Eq. (3.46) is replaced by the following equation
dC(t)
dt= −3ρc
∞∫
0
z2 kv G(t, z) F (t, z) dz , C(0) = C0 . (3.60)
The numerical results at 400 grid points are shown in Figure 3.7. The reference solution
was obtained from the DG-method on 3000 grid points. It can be seen that the DG-method
gives better approximation of the solution and correct positions of the discontinuities as
compared to the Koren scheme. The absolute error in the mass balances of the DG-scheme
is 7.9× 10−3, while in the Koren scheme it is equal to 8.3× 10−3. Moreover, the simulation
times of DG-scheme and Koren scheme are 2.81 and 2.79 seconds, respectively.
Problem 5
Here, a mathematical model for batch crystallizer loaded with a dissolution loop of fines
is considered [63]. The fines removal and their subsequent dissolution in an external unit
has potential to improve the quality of the product and to facilitate the down stream
processing. For further improving the process, a time delay in the dissolution unit(pipe) is
also included in the model . Here, the source term Q(t, z) of Eq. (3.1) is given as [63]
Q(t, z) = − V
Vcrz
ψ(z)F (t, z) , (3.61)
where V represents the volumetric flow rate and Vcrz is the volume of the crystallizer.
Moreover, ψ(z) represents a dimensionless cutoff (death) function. Thus, ψ(z)F (t, z) de-
notes that part of the size distribution which has been taken out from the crystallizer to
38
0.5 1 1.5 2
x 10−3
104
106
108
1010
1012
length [m]
CS
D[1
/m/k
g]
reference (DG)DG−schemeFVSinitial CSD
Figure 3.7: Problem 4: Size-dependent growth at t = 180min.
the dissolution unit. The following cutoff function is considered in this problem
ψ(z) =1√
2πσ1
e−( z
ασ1)2
, σ1 =1√
2πFmax
, (3.62)
where, α = 1.1547 × 103m and Fmax = 0.6. The balance law for the liquid phase modifies
to
dm(t)
dt= min(t) − mout(t) − 3ρc kv
∞∫
0
z2 G(t, z) F (t, z) dz , m(0) = m0 , (3.63)
where m denotes the mass of the solute in the solvent.The above equation has two liquid
streams due to the fines dissolution unit. The first one, mout, denotes the liquid stream
containing fines which is being taken out from the crystallizer to the pipe. The second one,
min, denotes the incoming pure and particle-free liquid stream to the crystallizer from the
pipe. They are described as
mout(t) = w(t)ρsolu(T )V , (3.64)
min(t) = mout(td) +kvρcV
Vcrz
∞∫
0
z3ψ(z)F (td, z)dz − kvρcz30B0(td) , (3.65)
39
where, td = t − τpipe. Here, w denotes the mass fraction
w(t) =m(t)
m(t) + msolv(T )
, (3.66)
where, msolv(T ) defines the mass of the solvent and ρsolu(T ) is the density of the solution
as a function of temperature T . Moreover, τpipe ≥ 0 denotes the residence time in the pipe.
It is defined as
τpipe =Vpipe
Vpipe
. (3.67)
The size-dependent crystal growth rate and nucleation rate are given by Eqs. (3.47) and
(3.51). Moreover, the supersaturation S(t) of the dissolved component is given as
S(t) =w(t)
weq(t), (3.68)
where, weq(t) is the mass fraction at equilibrium. This definition of supersaturation is dif-
ferent from the one given in Eq. (3.48). It can be observed that the above model reduces
to the case of fines dissolution without time delay when τpipe = 0. Moreover, the model
reduces to the case without dissolution of fines when the second last term on the right-hand
side of Eq. (3.1) and the first two terms on the right-hand side of Eq. (3.63) are zero.
Then Eqs. (3.64) and (3.65) are not needed.
The following initial seed distribution is considered
F (0, z) =mseeds
kv ρc µ3(0)√
2πσexp
(−(z − z)2
2σ2
)
. (3.69)
The smallest and largest crystal sizes are taken as z0 = 0 and zmax = 0.005m, respectively.
The interval [0, zmax] is divided into 100 mesh points and the final simulation time is 800
minutes for size-independent growth rate and 400 minutes for size-dependent growth rate.
For the size-dependent growth rate, aASL = 100m−1 and dASL = 2 in Eq. (3.47). For the
size-independent growth rate, aASL = 0. The constants and other parameters are given
40
in Table 3.3. The crystallizer was kept at a constant temperature of 33 oC. Figures 3.8
and 3.9 show the final CSDs due to size-independent and size-dependent growth rates,
respectively. In these figures, the final CSDs of the models without fines dissolution and
with fines dissolution as well as with and without (τpipe = 0) time-delay are compared.
The reference solution was obtained from the DG-scheme on a refined mesh. The fines
dissolution unit dissolves small crystals below a certain critical size and therefore reduces
the number of small crystals in the crystallizer as shown in Figures 3.9 and 3.10. Due to
dissolution, the solute mass in the solution increases as shown in Figure 3.11. However, the
dissolution has less effect on the growth rate of seeds crystals in the case of fines dissolution
without time-delay as can be observed in Figure 3.9. In this case, small nuclei get no time
to grow and instantaneously dissolve back in the solution as soon as they achieve a stable
size. As a result, the net effect on the solution concentration is negligible. On the other
hand, the fines dissolution with time-delay allows nuclei to grow in the crystallizer for
a certain time and the concentrated solution of the dissolution unit comes back to the
crystallizer with certain time-delay. Therefore, the concentration of the solution increases
in the crystallizer. In other words, the seed crystals and those introduced by nucleation
grow at a faster rate as can be observed in Figure 3.10. The numerical results of the DG-
method and FVS are almost the same. However, the DG-method better resolves the sharp
discontinuities as compared to FVS.
Table 3.4 gives a comparison of absolute and relative errors in the mass balances as well
as CPU time in the case of size-independent growth rate. No significant changes were
observed in the case of the size-dependent growth rate. The observed data show that the
proposed scheme is efficient and preserves the mass balance.
41
Table 3.3: Parameters for Problem 5Description Symbols Value Unitgrowth rate constant kg 1.37 × 10−5 m
min
growth rate exponent g 0.73 −nucleation rate constant kb 3.42 × 107 1
m3 min
nucleation rate exponent b 2.35 −density of crystals ρc 1250 kg
m3
volume shape factor kv 0.029 −initial solute mass m(0) 0.09915 kgsaturated mass fraction wsat 0.09068 −mass of seeds mseeds 2.5 × 10−3 kgmass of solvent msolv 0.8017 kg
density of solution ρsolu 1000 kgm3
volume of the crystallizer Vcrz 10−3 m3
volumetric flow rate Vpipe 2.0 × 10−5 m3
min
volume of the pipe Vpipe 2.4 × 10−4 mconstant of initial CSD (Eq. (3.69)) σ 3.2 × 10−4 mconstant of initial CSD (Eq. (3.69)) z 1.4 × 10−3 m
Table 3.4: Problem 5: Errors in mass balances for size-independent growth rateScheme Absolute error Relative error CPU time (s)DG-scheme 3.29 × 10−6 3.86 × 10−5 4.12Koren-scheme 4.29 × 10−6 4.21 × 10−5 4.10
42
0 1 2 3 4 5
x 10−3
0
0.5
1
1.5
2
2.5
3x 10
7
length [m]
CS
D [1
/m]
size−independent growth, t=800 minreferenceDG−schemeFVS
0 1 2 3 4 5
x 10−3
0
0.5
1
1.5
2
2.5x 10
7
length [m]
CS
D [1
/m]
size−dependetn growth, t=400 minreferenceDG−schemeFVS
Figure 3.8: Problem 5: CSD without fines dissolution.
0 1 2 3 4 5
x 10−3
0
0.5
1
1.5
2
2.5
3x 10
7
length [m]
CS
D [1
/m]
size−indep. growth, fines, no delayreferenceDG−schemeFVS
0 1 2 3 4 5
x 10−3
0
0.5
1
1.5
2
2.5x 10
7
length [m]
CS
D [1
/m]
size−dep. growth, fines diss., no delayreferenceDG−schemeFVS
Figure 3.9: Problem 5: Fines dissolution without delay
43
0 1 2 3 4 5
x 10−3
0
0.5
1
1.5
2
2.5
3x 10
7
length [m]
CS
D [1
/m]
size−ind. growth, fines, delayreferenceDG−schemeFVS
0 1 2 3 4 5
x 10−3
0
0.5
1
1.5
2
2.5x 10
7
length [m]C
SD
1/m
]
size−dep. growth, fines, delayreferenceDG−schemeFVS
Figure 3.10: Test 5: Fines dissolution with delay.
0 200 400 600 800
0.08
0.085
0.09
0.095
0.1
time [min]
mas
s [k
g]
without fines dissolutionfines diss. without delayfines diss. with delay
0 100 200 300 4000.08
0.085
0.09
0.095
0.1
time [min]
mas
s [k
g]
without fines dissolutionfines diss. without delayfines diss. with delay
Figure 3.11: Problem 5: Solute masses.
44
45
Chapter 4
Batch Preferential Crystallization
This chapter deals with the introduction of single and coupled batch preferential crystal-
lization models along with their numerical solutions by using the DG-scheme. Moreover,
the numerical results of the proposed method are compared with the HR-FVS [59].
4.1 Batch Preferential Crystallization Model
Here, a mathematical model is provided for simulating preferential crystallization of enan-
tiomer in a single batch crystallizer connected with a fines dissolution unit. This dissolution
process may give an improvement in the crystal size and, hence, the product quality. The
simplified sketch of a batch unit is shown in Figure 3.1. This setup preferentially crystal-
lizes only one enantiomer.
We make use of the same assumptions as discussed in section (3.2). In this case, the PBE
for the solid phase is given as [61].
∂F (κ)(t, z)
∂t= −∂[G(κ)(t, z)F (κ)(t, z)]
∂z− 1
τcryst
ψ(z)F (κ)(t, z), κ ∈ p, c , (4.1)
subject to the conditions
F (κ)(0, z) = F(κ)0 (z) , F (κ)(t, z0) =
B(κ)0 (t)
G(κ)(t, z0), (4.2)
where (t, z) ∈ R2+. Here, p and c stand for preferred and counter enantiomers, respectively.
Moreover, F (κ)(t, z) ≥ 0 is the CSD of a corresponding enantiomer with size z ≥ 0 at t ≥ 0,
and G(κ) represents the growth rate. The initial seed distribution is denoted by F(κ)0 , and
B(κ)0 represents the rate of nucleation at smallest size z0. In Eq. (4.1), the time of residence
in the vessel is denoted by τcryst which is defined as a ratio of crystallizer volume Vcrz to
the volumetric flow rate V i.e.
τcryst =Vcrz
V. (4.3)
46
The last term in Eq. (4.1) represents the particles’ death rate below a certain size. The
death function ψ(z) is given as a unit step function in which the critical size zcrit is taken
to be 2 × 10−4m and is defined as
ψ(z) =
0.6 for z ≤ zcrit
0.0 otherwise(4.4)
The moments of CSD are given as
µ(κ)i (t) :=
∞∫
0
zi F (κ)(t, z) dz. (4.5)
The balance laws for the liquid phase are described as
dm(κ)(t)
dt= m
(κ)in (t) − m
(κ)out(t) − 3ρckvG(κ)(t, z)µ
(κ)2 (t), t ∈ R≥0 , (4.6)
where µ(κ)2 denote the second moment, kv > 0 is the volumetric factor and ρc > 0 is the
crystal density. Because of dissolution, there are two mass flow rates given by Eq. (4.6).
The m(κ)out is the outer flux of mass from the vessel whereas m
(κ)in is the re-entering flux from
the pipe. These are given as
m(κ)out(t) = ω(κ)(t)ρsolu(T )V , (4.7)
m(κ)in (t) = m
(κ)out(t − τpipe) +
kvρc
τcryst
∞∫
0
z3 ψ(z) F (κ)(t − τpipe, z)dz. (4.8)
Here, ω(κ) represents the mass fractions and τpipe is the time of residence in the pipe as
expressed below
τpipe =πr2L
Vpipe
=Vpipe
Vpipe
, (4.9)
where the radius of the pipe is denoted by r and the length by L. In this model, the
nucleation is taken at z = z0,
F (κ)(t, z0) =B
(κ)0 (t)
G(κ)(t), κ ∈ p, c . (4.10)
47
The growth rate kinetics for preferential and counter enantiomers are given by [16, 17]
G(κ)(t) = (S(κ)(t) − 1)gkg, (4.11)
where kg ≥ 0 and g ≥ 1. The supersaturation denoted by S(κ) is given by
S(κ)(t) =ω(κ)(t)
ω(κ)eq (t)
. (4.12)
In the above equation, ω(κ) and ω(κ)eq represent mass fraction and mass fraction in equilib-
rium respectively, where
ω(κ)(t) :=m(κ)(t)
m(p)(t) + m(c)(t) + msolv(t), κ ∈ p, c , (4.13)
in which msolv(t) is the solvent mass (water here). Nucleation rates of preferential and
counter enantiomers are given by
B(p)0 (t) = (S(p))b(p)
k(p)b µ
(p)3 (t), (4.14)
B(c)0 (t) = k
(c)b exp
(
− b(c)
(ln(S(c)(t) + 1))2
)
. (4.15)
Without fines dissolution, the model reduces to
∂F (κ)(t, z)
∂t= −∂[G(κ)(t, z)F (κ)(t, z)]
∂z, (4.16)
dm(κ)(t)
dt= −3ρckvG(κ)(t, z)µ
(κ)2 (t) , (4.17)
along with initial and boundary conditions given by Eq. (4.2). Hence, Eqs. (4.7) and (4.8)
are not needed in this case.
4.2 Coupled Batch Preferential Crystallization Model
This section presents a mathematical model for ideally-mixed two-coupled batch prefer-
ential crystallizers connected through exchange pipes and equipped with fnes dissolution
48
units. The residence time and volume are assumed to be the same for both crystallizers.
Similarly, the volume and residence time for the two dissolution pipes are also considered to
be identical. The pipes dissolve the fines completely by providing sufficient heat to them.
Figure 4.1 is showing the coupled preferential crystallization process. In the beginning of
this process, two enantiomers and a solvent are mixed together. Vessel 1 is seeded with
the preferred enantiomer of L-threonine whereas Vessel 2 is seeded with the counter enan-
tiomer of D-threonine. Thus, both enantiomers are simultaneously crystallized in separate
crystallizers [19, 81].
Figure 4.1: Crystallization in coupled vessels.
The PBEs in the coupled case are given below [61]
∂F(κ)α (t, z)
∂t= − ∂[G(κ)
α (t, z)F(κ)α (t, z)]
∂z− 1
τcryst,αψα(z)F (κ)
α (t, z). (4.18)
Here, κ ∈ p, c and α ∈ 1, 2. In Eq. (4.18), τcryst,α can be expressed as
τcryst,α =Vcrz,α
Vα
. (4.19)
The death function ψα(z) is defined as
ψα(z) =
0.6 if z ≤ zcrit ,0 otherwise .
(4.20)
49
The liquid balances are given by
dm(κ)1 (t)
dt= m
(κ)in,2(t) − m
(κ)out,1(t) − 3ρckvG(κ)
1 (t, z)µ(κ)2,1(t), t ∈ R≥0. , (4.21)
dm(κ)2 (t)
dt= m
(κ)in,1(t) − m
(κ)out,2(t) − 3ρckvG(κ)
2 (t, z)µ(κ)2,2(t), t ∈ R≥0. (4.22)
Here, µ(κ)2,1(t) represents second moment of the CSD F
(κ)α (t, z) in Vessel 1. These equations
have four mass fluxes given by Eqs. (4.21) and (4.22) due to fines dissolution. Moreover,
m(κ)in,2(t) denotes the mass flux which is coming from the dissolution unit and going into
vessel 2 whereas m(κ)out,1(t) is the flux which is going out from Vessel 1 and going into the
pipe. Similarly, m(κ)in,1(t) represent the inner mass flux to vessel 1 and m
(κ)out,2(t) represent
the outer flux from Vessel 2. These are given by
m(κ)out,α(t) = ρsolu(T )ω(κ)
α (t)Vα , (4.23)
m(κ)in,α(t) = m
(κ)out,α(t − τpipe,α) +
kvρc
τcryst,α
∞∫
0
z3 sα(z) F (κ)α (t − τpipe,α, z)dz. (4.24)
Here, ω(κ)α denote mass fractions and τpipe,α is expressed as
τpipe,α =πr2
αLα
Vα
. (4.25)
The new particles born during the process are incorporated as
F (κ)α (t, z0) =
B(κ)α (t)
G(κ)α (t)
, κ ∈ p, c , (4.26)
and the growth rate is described as
G(κ)α (t) = (S(κ)
α (t))gkg, (4.27)
where kg ≥ 0 and g ≥ 1. The supersaturation in each crystallizer is given by
S(κ)α =
ω(κ)α
ω(κ)eq,α
− 1, κ ∈ p, c . (4.28)
50
We define the mass fractions for both the enantiomers in each crystallizer as
ω(κ)α (t) =
m(κ)α (t)
m(p)α (t) + m
(c)α (t) + msolv(t)
, κ ∈ p, c , (4.29)
and the nucleation rates by
B(p)α (t) = (S(p)
α (t))b(p)
k(p)b µ
(p)3,α(t), (4.30)
B(c)α (t) = k
(c)b exp
(
− b(c)
(ln(S(c)α (t) + 1))2
)
. (4.31)
4.3 Numerical Approximation of the Models
For the numerical approximation of PBE in Eq. (4.1), a TVB Runge-Kutta DG-method
[8, 68] is implemented. In this approach, the DG discretization is adopted in the size
variable only. The time derivatives are discretized by using a TVB Runge-Kutta method.
For simplicity, it is convenient to consider the simplified PBE for the scheme derivation
∂F (k)(t, z)
∂t= − ∂[G(k)(t, z)F (k)(t, z)]
∂z+ Q(k)(t, z) , (4.32)
where
Q(k)(t, z) = − 1
τcryst
ψ(z)F (κ)(t, z), κ ∈ p, c . (4.33)
Let us define
f(G(k), F (k)) := G(k)(t, z)F (k)(t, z) . (4.34)
To discretize the size-domain [z0, zmax], we proceed as follows. For j = 0, 1, 2, ....N , let zj+ 12
be the cells partitions, Ij =(
zj− 12, zj+ 1
2
)
be the domain of cell j, ∆zj = zj+ 12− zj− 1
2be
the width of cell j, and I =⋃
Ij are the particle size range partitions. We are looking for
an approximation F(k)h (t, z) to F (k)(t, z) so that for all t ∈ [0, tmax], F
(k)h (t, z) is belonging
to the space of finite dimensions
Vh =
v ∈ L1(I) : v|Ij∈ P k(Ij), j = 0, 1, 2, ....N
, (4.35)
51
where P k(Ij) stands for the polynomials space in I of at most k degree. In order to compute
the approximate solution F(k)h (t, z), we require a weak formulation that is usually obtained
by multiplying Eq. (4.32) with a smooth function v(z) and by integrating over the interval
Ij.
∫
Ij
∂F (k)(t, z)
∂tv(z)dz = −
∫
Ij
∂[f(G(k), F (k))]
∂zv(z)dz +
∫
Ij
Q(k)(t, z)v(z)dz . (4.36)
After using integration by parts, the weak formulation has the form
∫
Ij
∂F (k)(t, z)
∂tv(z)dz = −
(
f(G(k)
j+ 12
, F(k)
j+ 12
)vl(zj+ 12) − f(G
(k)
j− 12
, F(k)
j− 12
)vl(zj− 12))
+
∫
Ij
f(Gk, F (k))∂v(z)
∂zdz +
∫
Ij
Q(k)(t, z)v(z)dz . (4.37)
Selecting Legendre polynomials (LPs), Pl(z), of order l as local basis functions is one way
to formulate Eq. (4.35). Here, the L2-orthogonality property of LPs can be used, such as
1∫
−1
Pl(s)Pl′(s)ds =
(
2
2l + 1
)
δll′ . (4.38)
For each z ∈ Ij, the solution F(k)h can be expressed as
F(k)h =
k∑
l=0
F(l,k)j ϕl(z) , (4.39)
where
ϕl(z) = Pl (2(z − zj)/∆zj) . (4.40)
It can be easily verified that
(
1
2l + 1
)
F(l,k)j (t) =
1
∆zj
∫
Ij
F kh (t, )ϕ(z)dz . (4.41)
Then, the test function ϕl ∈ Vh replaces the smooth function v(z) and the approximate
solution F(k)h replaces the exact solution F (k) . Moreover, the function F
(k)
j+ 12
= F (t, z(k)
j+ 12
)
52
is not defined at the cell interface zj+ 12. Therefore, the f(G(k), F (k)) will be replaced with
a numerical flux that further be depends upon two values of F(k)h (t, z), i.e.,
f(G(k)
j+ 12
, F(k)
j+ 12
) ≈ hj+ 12
= h(G(k)
j+ 12
, F(k,−)
j+ 12
, F(k,+)
j− 12
) . (4.42)
Here
F(k,−)
j+ 12
:= F(k)h (t, z−
j+ 12
) =k
∑
l=0
F(l,k)j ϕ(zj+ 1
2) , F
(k,+)
j− 12
) := F(k)h (t, z+
j− 12
) =k
∑
l=0
F(l,k)j ϕl(zj− 1
2) .
(4.43)
Using the Eq. (4.42) and Eq. (4.43), the weak formulation Eq. (4.37) simplifies to
dF(l,k)j (t)
dt= − 2l + 1
∆zj
(
hj+ 12ϕl(zj+ 1
2) − hj− 1
2ϕl(zj− 1
2))
+2l + 1
∆zj
∫
Ij
(
f(G(k), F(k)h )
dϕl(z)
dz
)
dz +2l + 1
∆zj
∫
Ij
Q(k)(t, z)ϕl(z)dz . (4.44)
According to Eqs (5.2) and (4.41), the initial data for this equation is given as
F(l,k)j (0) =
2l + 1
∆zj
∫
Ij
F(k)0 (z)ϕ(z)dz . (4.45)
The next step is to select the suitable numerical flux function h. The scheme is monotone
if the flux function h(G(k), a, b) is consistent, h(G(k), F (k), F (k)) = f(G(k), F (k)), and fulfills
the Lipschitz condition [41, 80]. Here we have used the same numerical fluxes and minmod
limiter which are discussed in Chapter 3. Moreover, for solving the resulting ODE system,
the same Runge-Kutta method is used.
The derivation of the DG-scheme for the coupled batch preferential crystallization model
follows the same procedure as discussed above and is therefore omitted here.
In the following, some case studies are carried out for both single and coupled batch
crystallizers to validate our numerical scheme.
53
4.4 Test Problems for Single Crystallizer
Problem 1
The initial distribution function of preferred enantiomer is
F (p)(0, z) =1√
2πσDa
.1
z. exp
(
−1
2
(
ln(z) − z
σ
)2)
, Da = kvρc
mseedsµ
(p)3 (0) . (4.46)
The counter enantiomer is not seeded initially, i.e.
F (c)(0, z) = 0. (4.47)
Here, it is assumed that σ = 0.3947m, z = −6.8263m, while mseeds is the mass of initial
seeds. Tables 4.1 and 4.2 show the kinetic parameters. For the isothermal case, the
Table 4.1: Parameters for Problem 1Symbols Values Units
zmax 0.005 mz0 1 × 10−8 mt 600 minN 500 −
temperature is kept constant at 33oC whereas the temperature trajectory in the of non-
isothermal case is defined as
T (t)[oC] = −12.407 × 10−8t3 + 45.09 × 10−6t2 − 40.556 × 10−4t + 33, (4.48)
whereas the constants kg(t) and k(k)b (t) are given by
kg(t) = e−EA,g
(T+273.15)R kg,0, k(k)b (t) = e−
EA,b(T+273.15)R k
(k)b,0 . (4.49)
Moreover,
ω(p)eq (t,m(p),m(c)) =
2∑
j=0
T j(t)(aj + bjω(c)(t)), (4.50)
54
ω(c)eq (t,m(p),m(c)) =
2∑
j=0
T j(t)(cj + djω(p)(t)). (4.51)
The values of the constants are listed in Table 4.2. For the isothermal case, the same Eqs.
(4.50) and (4.51) are used.
Figures 4.2 compare the number densities of the preferential enantiomer obtained from the
HR-FVS and the DG-scheme when the fines dissolution is active. Figures 4.3 show the
CSD plots with and without fines dissolution. In the isothermal case, the temperature
is kept constant at 33oC whereas it is given by Eq. (4.48) for non-isothermal condition.
The part of the CSD coming from nucleation is smaller for the fines dissolution case as
compared to the case when fines dissolution is off. This is because the fines are dissolved
in the dissolution unit before they are sent back to the crystallizer, as a result of which
supersaturation of the solution increases which, in turn, diminishes the secondary nucle-
ation.
The non-isothermal case results are totally dependent upon the time depending tempera-
ture. As we reduce the temperature, the number density is increased. This is due to the
increase in nucleation rate. It is observed that the results are better for the non-isothermal
case with fines dissolution as we obtain bigger sized crystals as compared to the other cases.
Figures 4.4 show the CSDs of counter enantiomer which appears because of the nucleation.
The plots for mass fraction of preferred enantiomer are shown in Figures 4.5. Because of
the seeding of preferred enantiomer, its mass fraction decreases sharply in the isothermal
case. With fines dissolution (FD), the curve is slightly above than for without (FD). This
is because of an increase in supersaturation in the fines dissolution case.
55
Figures 4.6 compares the counter enantiomer mass fraction for with and without fines dis-
solution. The plot for supersaturation of the preferred enantiomer is shown in Figures 4.7.
Since the mass fraction of the preferred enantiomer decreases sharply, the supersaturation
of the solution reduces significantly.
The results for the non-isothermal case are influenced by the temperature profile defined
by Eq. (4.48). The change in growth rate and the nucleation rate are due to the change
in mass fraction, see Figures 4.8 and 4.9. Figures 4.10 explain the purity and the mean
crystal size of the crystal during the process, whereas Figure 4.11 shows the productivity
of the preferred enantiomer.
Problem 2
The initial number density for the seeds is
F (p)(0, z) =107
√2πσDa
exp
(
−1
2
(
z − z
σ
)2)
, Da =kv. ρc
mseeds
µ(p)3 (0) , (4.52)
while
F (c)(0, z) = 0. (4.53)
Here we assume σ = z15
, z = 4.10−4m, whereas mseeds is the mass of initial seeds. The
other parameters are given in Tables 4.2 and 4.3 . The comparison of the DG-scheme and
the HR-FVS is given in Figures 4.12. Both the schemes are in accord with each other.
The crystal size distribution of the preferred enantiomer is given in Figures 4.13 which is
compared for both with and without fines dissolution. The first peak in the isothermal
case is due to the nucleation which is very small.
The mass fractions of both the enantiomers (preferential and counter) are displayed in
Figures 4.14 and 4.15.
56
Table 4.2: Parameters for Problems 1, 2 and 4Symbols Values Unitskg,0 4.62 × 108 m
min
k(p)b,0 32.4 × 1024 1
m3min
k(c)b,0 3.84 × 106 1
min
g 1 −EA,g 7.56 × 104 kJ
mol
EA,b 7.87 × 104 kJmol
b(p) 4 −R 8.314 J
mol.K
b(c) 0.6 × 10−1 −ρc 1250 kg
m3
ρsolu 103 kgm3
kv 0.0248 −m(c)(0) 0.100224 kgm(p)(0) 0.100224 kgmseeds 25 × 10−4 kgmsolv 0.7995 kgVcrz 10 × 10−4 m3
τcryst 60 minr 10 × 10−3 mτpipe 10 minL 53 × 10−2 ma1 0.00143838 −a2 −3.41777 × 10−6 −
57
Table 4.3: Parameters for Problem 2Symbols Values Unitszmax 0.002 mz0 1 × 10−8 mt 600 minN 400 −
Figures 4.16 shows the supersaturation of preferred enantiomer with and without fines dis-
solution; the changes in growth rate is shown in Figures 4.17. The results obtained by the
DG-scheme are the same as obtained by the HR-FVS.
The productivity is displayed in Figures 4.18 which increases with time. The purity of the
product in the whole process can be seen in Figures 4.19. It is found that the final product
is 99.7% pure. The mean crystal size is also given in this figure.
Problem 3
The initial number density for the preferred enantiomers is given by
F (p)(0, z) =107
√2πσDa
exp
(
−1
2
(
z − z
σ
)2)
, Da =kv. ρc
mseeds
µ(p)3 (0) , (4.54)
while
F (c)(0, z) = 0. (4.55)
We assume σ = z15
, z = 10−4m. Table 4.4 shows the kinetic parameters for this problem.
In the first plot of Figure 4.20, the CSD of the preferred enantiomer obtained by the DG-
scheme is compared with that of HR-FVS, whereas in the second plot, the CSD of the
counter enantiomer is compared.
The results for mass fraction of the preferential and counter enantiomers are compared
with that of HR-FVS in the case of without fines in Figure 4.21. It is seen that both the
58
Table 4.4: Parameters for Problems 3 and 5Description Symbols Values Unitsmaximum crystal size zmax 0.002 mminimum crystal size z0 10−8 mfinal time tend 700 mintotal mesh points N 400 −growth rate constant kg 15.0 × 108 m
min
exponent of Growth rate g 1 −constant k
(p)b 0.5 × 102 1
m3min
exponent of Nucleation rate b 2 −constant k
(c)b k
(p)b /5 1
min
density of crystals ρc 1250 kgm3
density of liquid ρsolu 1000 kgm3
volume shape factor kv π/6 −solvent mass msolv 0.15004 kginitial preferential mass m(p) 0.01932 kginitial counter mass m(c) 0.01856 kgsaturation mass fraction ωsat 0.05220 kgseeds mass mseeds 50 × 10−5 kgpipe residence time τpipe 10 mincrystallizer residence time τcryst 60.0 minvolume of crystallizer Vcrz 0.001 m3
schemes give the same result. Figure 4.22 shows that the crystal obtained at the end is
99.9% pure, whereas Figure 4.23 displays the productivity with time.
4.5 Test Problems for Coupled Crystallizers
Here, two test cases are presented for the crystallization of enantiomers in coupled crystal-
lizers and the results are compared with those of single batch crystallizers.
Problem 4
The initial CSD of the target enantiomer in both crystallizers is given as
F (p)α (0, z) =
1
z√
2πσDα
exp
(
−1
2
(
ln(z) − z
σ
)2)
, (4.56)
59
with Dα = kv. ρc
mseedsµ
(p)3,α(0). The initial distribution of unwanted enantiomer is taken to be
zero in each crystallizer
F (c)α (0, z) = 0. (4.57)
Here, σ = 0.3947m, z = −6.8263m, while mseeds,α represents the seeds mass in tank
α ∈ 1, 2. Tables 4.2 and 4.5 show the kinetic parameters considered for this problem. It
Table 4.5: Parameters for Problem 4Symbols Values Units
zmax 0.005 mz0 10−8 mt 600 minN 500 −
is to be noted that in crystallizer 1, the preferred enantiomer is denoted by E1 whereas in
crystallizer 2, the preferred enantiomer is denoted by E2. The constants kg(t) and k(k)b (t)
are given by Eq. (4.49). The equilibrium mass fraction of both the enantiomers are defined
as
ω(p)eq,α =
2∑
j=0
T j(aj + bjω(c)α ), (4.58)
ω(c)eq,α =
2∑
j=0
T j(cj + djω(p)α ). (4.59)
The constants of the above equations are presented in Table 4.2.
In Figures 4.24, the number densities of preferred enantiomer are compared for single and
coupled crystallizers. For this test problem, the enantiomer E1 plots in tank 1 and those of
enantiomer E2 in tank 2 are identical. It can be observed that the crystal size distribution
in the coupled crystallizers is better than in the single crystallizer, as large crystals are
60
obtained in the coupled case. For isothermally operated crystallizer, the initially large nu-
cleation rate gives a large peak of preferred enantiomer, while the second curve is small due
to initial crystal growth. However, for the non-isothermal case the peaks are dependent
entirely on the temperature profile.
The mass fractions of the preferred enantiomer are compared for both crystallizers in
Figures 4.25. Because of the seeding of the preferential enantiomer, it crystallizes out
sharply resulting in a decrease in mass fraction. The change in supersaturation and growth
rate is due to the change in mass fraction. These changes can be seen in Figures 4.26 and
4.27 for the isothermal and non-isothermal cases. Comparisons of the productivity and
mean crystal size are made in Figures 4.28 and 4.29 for both single and coupled crystallizers.
Problem 5
The initial distribution functions of the preferred and counter enantiomers are given by the
representations
F (p)α (0, z) =
107
√2πσDα
exp
(
−1
2
(
z − z
σ
)2)
, Dα =kv. ρc
mseeds
µ(p)3,α(0) , (4.60)
F (c)α (0, z) = 0. (4.61)
The values of constants and kinetic parameters considered here are the same as those of
Problem 3.
Here the results of a single crystallizer and for coupled crystallizers are compared. Figure
4.30 makes a comparison of single and coupled crystallizers for the cases of fines and
without FD. It can be observed that the growth in the coupled case is enhanced and, thus,
the distribution sifts to the right. In other words, large size crystals are obtained in the
coupled case. Therefore, the results with the coupled crystallizers are better than with a
61
single crystallizer. Moreover, purity, productivity and mean crystal size are also compared
in Figures 4.31 and 4.32. One can also observe improvements in the purity, productivity
and mean crystal size of the coupled crystallization.
62
0 1 2 3 4 5
x 10−3
0
2
4
6
8
10
12x 10
8
crystal size [m]
part
icle
den
sity
[1/m
]
Isothermal Case
HR FVSDG scheme
0 1 2 3 4 5
x 10−3
0
1
2
3
4
5
6x 10
9
crystal size [m]pa
rtic
le d
ensi
ty [1
/m]
Non−isothermal case
HR FVSDG scheme
Figure 4.2: Test 1: Preferred enantiomer CSD with HR-FVS and DG scheme.
0 1 2 3 4 5
x 10−3
0
2
4
6
8
10
12x 10
8
crystal size [m]
part
icle
den
sity
[1/m
]
Isothermal Case
without fineswith fines (no delay)with fines (with delay)
0 1 2 3 4 5
x 10−3
0
2
4
6
8
10x 10
9
crystal size [m]
part
icle
den
sity
[1/m
]
Non−isothermal Case
without fineswith fines (no delay)with fines (with delay)
Figure 4.3: Test 1: Preferred enantiomer CSD for with and without FD.
63
0 1 2 3 4 5
x 10−3
0
1
2
3
4
5
6
7x 10
−6
crystal size [m]
part
icle
den
sity
[1/m
]
Isothermal Casewithout fineswith fines (no delay)with fines (with delay)
0 1 2 3 4 5
x 10−3
0
0.5
1
1.5
2
2.5
3
3.5x 10
−3
crystal size [m]
part
icle
den
sity
[1/m
]
Non−isothermal Case
without fineswith fines (no delay)with fines (with delay)
Figure 4.4: Test 1: CSD for counter enantiomer with and without FD.
0 100 200 300 400 500 6000.092
0.093
0.094
0.095
0.096
0.097
0.098
0.099
0.1
0.101
0.102
time [min]
mas
s fr
actio
n [−
]
Isothermal Case
without fineswith fines (without delay)with fines (with delay)
0 100 200 300 400 500 6000.084
0.086
0.088
0.09
0.092
0.094
0.096
0.098
0.1
0.102
time [min]
mas
s fr
actio
n [−
]
Non−isothermal Case
without fineswith fines (no delay)with fines (with delay)
Figure 4.5: Test 1: Mass fraction for preferred enantiomer with and without FD.
64
0 100 200 300 400 500 600
0.1004
0.1006
0.1008
0.101
0.1012
0.1014
0.1016
time [min]
mas
s fr
actio
n [−
]
Isothermal Case
without fineswith fines (no delay)with fines (with delay)
0 100 200 300 400 500 6000.1
0.1005
0.101
0.1015
0.102
0.1025
0.103
0.1035
time [min]
mas
s fr
actio
n [−
]
Non−isothermal Case
without fineswith fines (no delay)with fines (with delay)
Figure 4.6: Test 1: Mass fraction for counter enantiomer.
0 100 200 300 400 500 6000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
time [min]
supe
rsat
urat
ion
[−]
Isothermal Case
without fineswith fines (without delay)with fines (with delay)
0 100 200 300 400 500 6000.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
time [min]
supe
rsat
urat
ion
[−]
Non−isothermal Case
without fineswith fines (no delay)with fines (with delay)
Figure 4.7: Test 1: Supersaturation for target enantiomer with and without FD.
65
0 100 200 300 400 500 6000
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5x 10
−6
time [min]
grow
th r
ate
[m/m
in]
Isothermal Case
without fineswith fines (without delay)with fines (with delay)
0 100 200 300 400 500 6000.5
1
1.5
2
2.5
3
3.5
4
4.5
5x 10
−6
time [m]
grow
th r
ate
[m/m
in]
Non−isothermal Casewithout fineswith fines (no delay)with fines (with delay)
Figure 4.8: Test 1: Growth rate for target enantiomer with and without FD.
0 100 200 300 400 500 6000
1000
2000
3000
4000
5000
6000
time [min]
nucl
eatio
n ra
te [m
/min
]
Isothermal Case
without fineswith fines (no delay)with fines (with delay)
0 100 200 300 400 500 6000
2000
4000
6000
8000
10000
12000
14000
time [min]
nucl
eatio
n ra
te [m
/min
]
Non−isothermal Case
without fineswith fines (no delay)with fines (with delay)
Figure 4.9: Test 1: Nucleation rate for target enantiomer with and without FD.
66
0 100 200 300 400 500 60099.7
99.75
99.8
99.85
99.9
99.95
100
time [min]
purit
y [%
]
0 100 200 300 400 500 6000.2
0.4
0.6
0.8
1
1.2
time [min]
mea
n cr
ysta
l siz
e [m
m]
isothermal casenon−isothremal case
Figure 4.10: Test 1: Purity and mean crystal size of preferred enantiomer.
0 100 200 300 400 500 6000
2
4
6
8
10
time [min]
prod
uctiv
ity [k
g/m
in m
3 ]
Isothermal Case
0 100 200 300 400 500 6000
5
10
15
20
time [min]
prod
uctiv
ity [k
g/m
in m
3 ]
Non−isothermal Case
Figure 4.11: Test 1: Productivity of preferred enantiomer.
67
Figure 4.12: Test 2: CSD for target enantiomer with HR-FVS and DG scheme.
0 0.5 1 1.5 2
x 10−3
0
2
4
6
8
10
12
14
16
18x 10
9
crystal size [m]
part
icle
den
sity
[1/m
]
Isothermal Case
without fineswith fines
Figure 4.13: Test 2: CSD for target enantiomer.
68
0 100 200 300 400 500 6000.092
0.093
0.094
0.095
0.096
0.097
0.098
0.099
0.1
0.101
0.102
time [min]
mas
s fr
actio
n [−
]
Isothermal Case
without fineswith fines
0 100 200 300 400 500 6000.08
0.085
0.09
0.095
0.1
0.105
0.11
time [min]
mas
s fr
actio
n [−
]
Non−isothermal case
without fineswith fines
Figure 4.14: Test 2: Mass fraction for preferred enantiomer with and without FD.
0 100 200 300 400 500 600
0.1004
0.1006
0.1008
0.101
0.1012
0.1014
0.1016
time [min]
mas
s fr
actio
n [−
]
Isothermal Case
without fineswith fines
0 100 200 300 400 500 6000.1
0.1005
0.101
0.1015
0.102
0.1025
0.103
0.1035
0.104
time [min]
mas
s fr
actio
n [−
]
Non−isothermal case
without fineswith fines
Figure 4.15: Test 2: Mass fraction for counter enantiomer.
69
0 100 200 300 400 500 6000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
time [min]
supe
rsat
urat
ion
[−]
Isothermal Case
without fineswith fines
0 100 200 300 400 500 6000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
time [min]
supe
rsat
urat
ion
Sp −1
[−]
Non−isothermal Case
without fineswith fines
Figure 4.16: Test 2: Supersaturation for preferred enantiomer with and without FD.
0 100 200 300 400 500 6000
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5x 10
−6
time [min]
grow
th r
ate
[m/m
in]
Isothermal Case
without fineswith fines
Figure 4.17: Test 2: Growth rate for preferred enantiomer.
70
Figure 4.18: Test 2: Productivity of preferred enantiomer.
0 100 200 300 400 500 60099.7
99.75
99.8
99.85
99.9
99.95
100
time [min]
purit
y [%
]
0 100 200 300 400 500 6000.2
0.4
0.6
0.8
1
1.2
time [min]
mea
n cr
ysta
l siz
e [m
m]
isothermal casenon−isothremal case
Figure 4.19: Test 2: Purity and mean crystal size of preferred enantiomer.
71
0 0.5 1 1.5 2
x 10−3
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2x 10
8
crystal size [m]
part
icle
den
sity
[1/m
]
Without Fines Dissolution
HR FVSDG−scheme
Figure 4.20: Test 3: CSD for both enantiomers with HR-FVS and DG scheme.
0 100 200 300 400 500 600 7000.06
0.065
0.07
0.075
0.08
0.085
0.09
0.095
0.1
0.105
time [min]
mas
s fr
actio
n [−
]
Without Fines Dissolution
HR FVSDG− scheme
0 100 200 300 400 500 600 7000.098
0.099
0.1
0.101
0.102
0.103
0.104
0.105
0.106
time [min]
mas
s fr
actio
n [−
]
Without Fines (counter)
HR FVSDG−scheme
Figure 4.21: Test 3: Mass fraction for both enantiomers with HR-FVS and DG scheme.
72
0 100 200 300 400 500 600 70099.88
99.9
99.92
99.94
99.96
99.98
100
100.02
100.04
time [min]
purit
y [%
]
0 100 200 300 400 500 600 7000.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
time [min]
mea
n cr
ysta
l siz
e [m
m]
Figure 4.22: Test 3: Purity and mean crystal size for preferred enantiomer.
Figure 4.23: Test 3: Productivity of preferred enantiomer.
73
0 1 2 3 4 5
x 10−3
0
1
2
3
4
5
6
7x 10
8
crystal size [m]
part
icle
den
sity
[1/m
]
Isothermal Case
single crystallizercoupled crystallizers
0 1 2 3 4 5
x 10−3
0
2
4
6
8
10
12
14
16
18x 10
9
crystal size [m]pa
rtic
le d
ensi
ty [1
/m]
Non−isothermal Case
single crystallizercoupled crystallizers
Figure 4.24: Test 4: Preferred CSD with FD for single and coupled crystallizers.
0 100 200 300 400 500 6000.092
0.093
0.094
0.095
0.096
0.097
0.098
0.099
0.1
0.101
0.102
time [min]
mas
s fr
actio
n [−
]
Isothermal Case
single crystallizercoupled crystallizers
0 100 200 300 400 500 6000.084
0.086
0.088
0.09
0.092
0.094
0.096
0.098
0.1
0.102
time [min]
mas
s fr
actio
n [−
]
Non−isothermal Case
single crystallizercoupled crystallizers
Figure 4.25: Test 4: Mass fraction with FD for single and coupled crystallizers.
74
0 100 200 300 400 500 6000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
time [min]
supe
rsat
urat
ion
[−]
Isothermal Case
single crystallizercoupled crystallizers
0 100 200 300 400 500 6000
0.02
0.04
0.06
0.08
0.1
0.12
0.14
time [min]
supe
rsat
urat
ion
[−]
Non−isothermal case
coupled crystallizerssingle crystallizer
Figure 4.26: Test 4: Supersaturation with FD for single and coupled crystallizers.
0 100 200 300 400 500 6000
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5x 10
−6
time [min]
grow
th r
ate
[m/m
in]
Isothermal Case
single crystallizercoupled crystallizers
0 100 200 300 400 500 6000.5
1
1.5
2
2.5
3
3.5
4
4.5
5x 10
−6
time [min]
grow
th r
ate
[m/m
in]
Non−isothermal Case
single crystallizercoupled crystallizers
Figure 4.27: Test 4: Growth rate with FD for single and coupled crystallizers.
75
0 100 200 300 400 500 6000
2
4
6
8
10
12
14
16
18
20Isothermal Case
time [min]
prod
uctiv
ity [k
g/m
in m
3 ]
single crystallizercoupled crystallizers
0 100 200 300 400 500 6000
5
10
15
20
25
30
35
time [min]
prod
uctiv
ity [k
g/m
in m
3 ]
Non−isothermal Case
single crystallizercoupled crystallizers
Figure 4.28: Test 4: Productivity for single and coupled crystallizers without FD.
0 100 200 300 400 500 6000.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2Isothermal Case
time [min]
mea
n cr
ysta
l siz
e [m
m] single crystallizer
coupled crystallizers
0 100 200 300 400 500 6000.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2Non−isothermal Case
time [min]
mea
n cr
ysta
l siz
e [m
m]
single crystallizerdoubled crystallizers
Figure 4.29: Test 4: Mean crystal size for single and coupled crystallizers without FD.
76
0 0.5 1 1.5 2
x 10−3
0
2
4
6
8
10
12
14
16
18x 10
7
crystal size [m]
part
icle
den
sity
[1/m
]
Without Fines Dissolution
single crystallizercoupled crystallizers
0 0.5 1 1.5 2
x 10−3
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5x 10
7
crystal size [m]
part
icle
den
sity
[1/m
]
With Fines Dissolution
coupled crystallizerssingle crystallizer
Figure 4.30: Test 5: Preferred CSD for single and coupled crystallizers.
77
Figure 4.31: Test 5: Purity and productivity of p-enantiomer for single and coupled crys-tallizers.
mea
n cr
yst
al
size
[mm
]
Figure 4.32: Test 5: Mean crystal size of p-enantiomer for single and coupled crystallizers.
78
79
Chapter 5
Continuous Preferential Crystallization
Preferential crystallization (PC) is an attractive process for gaining pure enantiomers from
racemic mixtures, which has been realized up to now in a discontinuous operation mode,
see for example [3, 11, 12, 13, 17, 18, 31]. For elucidating the principle of a continuous
enantioselective process one might consider a suspension crystallizer revealing MSMPR
characteristics, i.e. a perfectly mixed tank (concerning both phases), which is continuously
fed with a solution possessing a racemic composition of two enantiomers, see Figure 5.1.
Solid particles and liquid phase are continuously withdrawn. By a continuous supply of ho-
mochiral seed crystals of the preferred target enantiomer the preferential crystallization of
only this enantiomer is initialized, i.e. growth of the seed crystals and possibly secondary
nucleation of crystals of the seeded enantiomer, provided the crystallization takes place
within the metastable zone where spontaneous, uncontrolled primary nucleation is kineti-
cally inhibited. During a starting-up period, which strongly depends on the properties of
the system as well as on the process parameters, the concentration of the target enantiomer
is decreasing until a steady state is reached where the composition is determined by the
mean residence time. Due to different kinetic mechanisms and their inherent different time
constants, a different depletion of the supersaturation for each enantiomer can be realized
by an appropriate choice of the process conditions. As long as a critical mean residence
time, where primary nucleation may appear, is not exceeded, the concentration of the un-
desired counter enantiomer remains constant during the whole time. This fact reveals a
benefit of this continuous process in comparison to the batch process. An optimal selection
of the process conditions allows a constant production of the goal enantiomer at a high
purity level. The mathematical modeling of PC as well as the optimization of the essential
operating conditions requires high numerical precision [60].
In this chapter, the semi-discrete flux-limiting finite volume scheme [32] and discontinu-
ous Galerkin finite element method are used to solve a dynamic mathematical model for
80
continuous preferential enantioselective crystallization. The developed model includes the
phenomena of primary heterogeneous nucleation, secondary nucleation and size-dependent
growth. A recycle pipe is attached to the crystallizer for the fines dissolution and it is
assumed that fines are completely dissolved at the other end of the pipe. The effects of
different seeding and operating strategies are investigated in the simulations. The model
and its parameters are based on experimental data obtained from batchwise crystallization
and correspond to the enantiomers of the amino acid threonine and the solvent water, see
[17, 18]. The numerical results demonstrate the high order accuracy, efficiency and potential
of the proposed numerical methods for solving models describing continuous preferential
crystallization of enantiomers and the potential of the process.
5.1 Continuous Preferential Crystallization Model
The model of the continuous crystallization process considers a two-phase dispersed sys-
tem. The first phase is a continuous liquid phase containing the dissolved enantiomers and
the solvent. The conditions in this phase determine the growth and nucleation rates of
crystals during the crystallization process and have significant influence on the crystal size
distributions (CSDs). The second phase is the dispersed solid phase of individual crystals.
The growth of crystals strongly depends on the degree of supersaturation of the liquid
phase.
Description of the CSD for a dispersed system in a crystallizer needs an adequate differential
equation accounting for the effects of nucleation, growth, and other phenomena involving
the change of crystals population, for instance fines dissolution in the current study. The
population balance modeling is a suitable approach for predicting CSD in the solid phase
of a disperse system.
81
(preferred) goal enantiomer: p(unwanted) counter enantiomer: c
dissolution pipe
(p)outS,m&
(p)seedsm&
LV&
LV&
pipeV&
Sτ
pipeτ
Lτ
Residence time in pipe =
Mean residence time of solid =
Mean residence time of liquid =
LV
)()( cp ww =
heat
exc
hang
er
Figure 5.1: Principle of continuous enantioselective crystallization [66].
A one-dimensional mathematical model for an ideally-mixed continuous crystallizer is
given by the following population balance equation (PBE) for the number density function
F (k)(t, z) [2, 27, 54]
∂F (k)
∂t= − ∂(G(k)F (k))
∂z+ F
(k)seeds − F
(k)out − F
(k)pipe , k ∈ p, c , (5.1)
where p stands for the preferred enantiomer and c for the counter enantiomer. In Eq.
(5.1) the term on the left hand side denotes the accumulation of crystals of size z. The
first term on the right hand side represents the convective transport in the direction of the
property coordinate z due to size-dependent crystal growth rate G(k). The term F(k)seeds de-
notes the incoming particle number flux due to seeding, F(k)out is the particle number flux due
to product removal, and F(k)pipe denotes the particle number flux to the fines dissolution unit.
82
The initial conditions for the PBE in Eq. (5.1) are based on the assumption that the
crystallizer initially (at time t = 0) contains only pure liquid phase. In other words
F (k)(t = 0, z) = 0 , k ∈ p, c . (5.2)
Assuming further, that primary nucleation leads to a crystal nuclei of minimum size z0
and that the number density function F (k) vanishes for an arbitrary large crystal size zmax,
holds
F (k)(t, z = z0) =B
(k)0 (t)
G(t, z0), F (k)(t, z = zmax) = 0 , (5.3)
where B(k)0 denotes the nucleation rate. The crystallizer is only seeded with preferred
enantiomer, thus
F(k)seeds(z) =
F(p)seeds(z) for k = p ,
0 for k = c .(5.4)
For the operation of the configuration shown in Figure 5.1, three types of residence times
are relevant:
1. The liquid mean residence time in the perfectly mixed crystallizer, τL = VL/VL, where
VL denotes the volume of the liquid phase in the crystallizer and VL is the volumetric
flow rate of the liquid phase.
2. A characteristic solid mean residence time in a perfectly mixed crystallizer, τS =
VS/(ρS/m(p)seeds), where VS represents the overall volume of the solid phase, ρS is the
density of the solid phase and m(p)seeds is the mass flow rate of the seeds of preferred
enantiomer.
3. The residence time in the pipe (plug flow), τpipe = Vpipe/Vpipe. Here, Vpipe denotes
volume of the pipe and Vpipe is the volumetric flow rate to the pipe.
83
The number density function in the withdrawn crystal stream is assumed to be equal to
the number density function inside the crystallizer. Therefore, the right hand sided second
and third terms of Eq. (5.1) are given as
F(k)seeds(z) =
F(k)seeds(z)
τS
, F(k)out(t, z) =
F (k)(t, z)
τS
, k ∈ p, c . (5.5)
The term corresponding to the dissolution unit in Eq. (5.1) is expressed by
F(k)pipe(t, z) =
ψ(z)F (k)(t, z)
τpipe
, (5.6)
where ψ(z) represents the classification (death) function.
The i-th moment of the CSD is defined as
µ(k)i (t) =
∞∫
0
ziF (k)(t, z) dz , i = 0, 1, 2, · · · , N, k ∈ p, c . (5.7)
According to Figure 5.1, the corresponding mass balances for the liquid phase are given as
dm(k)(t)
dt= m
(k)in (t) − m
(k)out(t) − 3ρS kv
∞∫
0
z2G(k)(t, z)F (k)(t, z) dz , k ∈ p, c (5.8)
with the initial conditions
m(k)(t = 0) = m(k)0 = ρfVL , (5.9)
where ρf = ρ(t = 0) is the density of fresh (initially) supersaturated solution. The incoming
and outgoing mass flow rates are defined as
m(k)in (t) = m
(k)in,c + m
(k)in,pipe(t), m
(k)out(t) = m
(k)out,c(t) + m
(k)out,pipe(t) . (5.10)
The inflow rate m(k)in is the sum of two incoming streams to the crystallizer, the first
one m(k)in,c stands for the incoming flux of fresh solution to the crystallizer and the second
one m(k)in,pipe denotes the incoming flux of particle-free solution from the dissolution pipe.
Similarly, the mass outflow rate m(k)out is the sum of two outgoing streams, i.e. mass flow
84
rate of the utilized solution from the crystallizer and the mass flow rate of the solution
containing the fraction of fines being taken out from the crystallizer to the dissolution
unit. The mass flow rate of the incoming fresh supersaturated solution is defined as
m(k)in,c = w(k)(t = 0)ρf VL , k ∈ p, c , (5.11)
where w(k) represents the mass fraction of the k-th enantiomer which is defined as
w(k)(t) =m(k)(t)
m(p)(t) + m(c)(t) + msolv
, k ∈ p, c . (5.12)
The terms m(k), m(p) and m(c) represent the masses of the k-th, preferred, and counter
enantiomers, respectively. Moreover, msolv is mass of the solvent (water). The composition
dependent solution density is defined as
ρL(t) = 1000(ρ0 + K3wtot(t)) , (5.13)
where ρ0 is the density of water
ρ0 =1
K1 + K2T 2. (5.14)
The parameters K1, K2, K3 are given in the Table 5.1 and T denotes the temperature
which is kept constant at 33 oC. The total mass fraction wtot is given as
wtot(t) = w(p)(t) + w(c)(t) . (5.15)
The outgoing mass fluxes of the solution from the crystallizer are defined as
m(k)out,c(t) = w(k)(t)ρ(t)V , k ∈ p, c , (5.16)
while the outgoing mass flux from the crystallizer to the dissolution unit is given as
m(k)out,pipe(t) = w(k)(t)ρ(t)Vpipe , k ∈ p, c . (5.17)
85
Respecting the fines dissolution and the residence in the tube, the incoming mass flux from
the dissolution unit to the crystallizer is given as
m(k)in,pipe(t) = m
(k)out,pipe(t − τpipe) +
kvρS
τpipe
∞∫
0
z3 h(z) F (k)(t − τpipe, z) dz . (5.18)
Due to the racemic solution, the initial masses of both enantiomers are the same and the
mass fluxes of both enantiomers are equal in the feed stream, so,
m(p)0 = m
(c)0 , m
(p)in,c = m
(c)in,c . (5.19)
A size-dependent growth rate approach is applied ([52, 18])
G(k)(t, z) = kg,eff(T )(S(k)(t) − 1)g(1 + aASLz)dASL . (5.20)
The exponent g denotes the growth order and the constants aASL and dASL represent the
size dependency. The temperature dependence of the growth rate constant kg,eff is given
by an Arrhenius type relation,
kg,eff = kg,eff,0 exp
(
−EA,g
RT
)
. (5.21)
Here, the symbol kg,eff,0 is the pre-exponential factor of the growth rate constant, EA,g
is the activation energy, and R is the universal gas constant. Their values are given in
Table 5.1. The symbol S(k)(t) denotes the supersaturation of the k-th enantiomer which is
defined as
S(k)(t) =w(k)(t)
w(k)eq (t)
, (5.22)
where w(k)eq is the saturated mass fraction of the k-th enantiomer defined as
w(p)eq (t) = aeq + beqw
(c)(t) , w(c)eq (t) = aeq + beqw
(p)(t) . (5.23)
The terms aeq and beq are solubility constants given in the Table 5.1. The nucleation rate
is defined as the sum of primary (heterogeneous) nucleation rate and secondary nucleation
86
rate ([52, 18])
B(k)0 (t) = B
(k)0,prim(t) + B
(k)0,sec(t) , k ∈ p, c . (5.24)
The primary nucleation rate is given by a semi-empirical equation derived from the Mers-
mann model (see [52, 18])
B(k)0,prim(t) = η(k)(t) exp
−aprim ln(
ρS/C(k)eq
)3
(ln S(k)(t))2
, (5.25)
where
η(k)(t) = kb,primTe−KTT exp
(−∑
k w(k)(t)
KW
)
√
√
√
√ln
(
ρS
C(k)eq (t)
)
(
S(k)(t)C(k)eq (t)
)73 . (5.26)
Here, C(k)eq denotes the concentrations of the k-th enantiomer at equilibrium as defined by
C(k)eq (t) = ρ(t)w(k)
eq (t) . (5.27)
The secondary nucleation rate is given by an overall power law expression
B(k)0,sec(t) = kb,sec
(
S(k)(t) − 1)bsec
(
µ(k)3 (t)
)nsec
, (5.28)
where bsec is the secondary nucleation rate exponent and nsec is the third moment exponent.
The secondary nucleation rate constant is given as
kb,sec = kb,sec,0 exp
(
−EA,b
RT
)
. (5.29)
The symbol kb,sec,0 is the pre-exponential factor of the secondary nucleation rate constant
and EA,b is the corresponding activation energy.
Although the steady state results are crucial for an assessment of the process, a better
experimental realization can be achieved by investigating the process dynamically.
5.2 Numerical Techniques
In this section, HR-FVS and DG-schemes are proposed to solve the PBE for continuous
preferential crystallization given by Eq. (5.1).
87
5.2.1 Implementation of HR-FVS
The high resolution finite volume scheme [32] is implemented for discretizing the derivative
of length coordinate in Eq. (5.1). The scheme derivation follows the same procedure as
presented in Chapter 3. However, for completeness, we present the derivation again. It is
convenient to re-write Eq. (5.1) in the following form:
∂F (k)(t, z)
∂t= − ∂[G(k)(t, z)F (k)(t, z)]
∂z+ Q(k)(t, z) , (5.30)
where
Q(k)(t, z) = F(k)seeds(z) − F
(k)out(t, z) − F
(k)pipe(t, z) . (5.31)
In order to apply a numerical scheme, the first step is to discretize the computational
domain which is here the crystal size z. Let N be a large integer, and denote by (zi− 12),
i ∈ ( 1, 2, · · · , N + 1), the partitions of cells in the domain [z0, zmax], where z0 is the
minimum and zmax is the maximum crystal length of interest. For each i = 1, 2, · · · , N , ∆z
represents the cell width, the points zi refer to the cell centers, and the points zi± 12
denotes
the cell boundaries. The integration of Eq. (5.30) over the cell Ωi =[
zj− 12, zj+ 1
2
]
yields the
following cell centered semi-discrete finite volume schemes for f(k)
i± 12
= (G(k)F (k))i± 12
∫
Ωi
∂F (k)(t, z)
∂tdz = − (f
(k)
i+ 12
(t) − f(k)
i− 12
(t)) +
∫
Ωi
Q(k)(t, z)dz. (5.32)
Let F(k)i and Q(k)
i denote the average values of the number density and source term in each
cell Ωi, i.e.
F(k)i (t) =
1
∆z
∫
Ωi
F (k)(t, z)dz , Q(k)i (t) =
1
∆z
∫
Ωi
Q(k)(t, z)dz . (5.33)
Then, Eq. (5.32) can be written as
dF(k)i (t)
dt= −
f(k)
i+ 12
(t) − f(k)
i− 12
(t)
∆z+ Q(k)
i (t), i = 1, 2, · · · , N , k ∈ p, c . (5.34)
88
Here, N denotes the total number of mesh elements in the computational domain. The
accuracy of finite volume discretization is mainly determined by the way in which the cell
face fluxes are computed. Assuming that the flow is in the positive z-direction, a first order
accurate upwind scheme is obtained by taking the backward differences,
f(k)
i+ 12
(t) = (G(k)(t)F (k)(t))i , f(k)
i− 12
(t) = (G(k)(t)F (k)(t))i−1 . (5.35)
To obtain high order accuracy of the scheme, one has to use better approximation of the
cell interface fluxes. According to the high resolution finite volume scheme of Koren the
flux at the right boundary zi+ 12
is approximated as
f(k)
i+ 12
(t) = f(k)i (t) +
1
2φ(r
(k)
i+ 12
(t))(f(k)i (t) − f
(k)i−1(t)) . (5.36)
Similarly, one can approximate the flux at the left cell boundary. The flux limiting function
φ according to [32] is defined as
φ(
r(k)
i+ 12
(t))
= max
0, min (2r(k)
i+ 12
(t), min (1
3+
2r(k)
i+ 12
(t)
3, 2))
. (5.37)
Here, r(k)
i+ 12
(t) is the upwind ratio of the consecutive flux gradients,
r(k)
i+ 12
(t) =f
(k)i+1(t) − f
(k)i (t) + ε
f(k)i (t) − f
(k)i−1(t) + ε
, (5.38)
where ε is a small number to avoid division by zero. This scheme is not applicable up
to the boundary cells because it needs values of the cell nodes which are not present. To
overcome this problem, the first order approximation of the fluxes was used at the inter-
faces of the first two cells on the left-boundary and at the interfaces of the last cell on the
right-boundary. At the remaining interior cell interfaces, the high order flux approximation
of Eq. (5.36) was used. It should be noted that, the first order approximation of the fluxes
in the boundary cells does not effect the overall accuracy of the proposed high resolution
89
scheme.
The resulting system of ordinary differential equations (ODEs) in Eq. (5.34) together with
Eqs. (5.36)-(5.38) can be solved by a standard ODE-solver. In this study a Runge-Kutta
method of order four was used. In the case of fines dissolution with time-delay the residence
time in the pipe was taken as an integer multiple of the time step. This facilitates to keep
the old values in memory and to avoid linear interpolation. The computer program is
written in Matlab 7.9.1 (R2009b).
5.2.2 Implementation of DG-Scheme
The derivation of the Runge-Kutta discontinuous Galerkin (DG) scheme for approximat-
ing Eq. (5.1) follows the same procedure as presented in Chapter 3. Moreover, the same
Runge-Kutta method is used to solve the resulting ODE system. Therefore, we omit the
derivation of the scheme in this chapter.
5.3 Test Problems
In this section the simulation results of a MSMPR preferential crystallizer are presented
for different operating strategies. Firstly, the crystallizer is operated without fines disso-
lution unit and is either seeded continuously or periodically with seeds of the preferred
enantiomer. In this case the last terms on the right hand side of Eqs. (5.31) and (5.10) are
neglected. Secondly, the crystallizer is equipped with a fines dissolution unit and is contin-
uously seeded with seeds of the preferred enantiomer. In this process, it is assumed that
all fines are dissolved in the pipe before re-entering the solution back to the crystallizer. In
order to judge the quality of the process some goal functions can be used such as product
purity, productivity, yield and mean crystal size of the preferred enantiomer. These goal
90
functions give detailed information about the success and potential of continuous prefer-
ential crystallization.
The mean residence time of the liquid phase remained fixed at 112.41 min, while the res-
idence time of the solid phase was considered as either τS = τL or τS = 2τL. In an actual
operation, it may be difficult to decouple the residence times of the liquid and solid phases.
Thus, it is most likely to set τS = τL . However, larger residence times of the solid phase
compared to the liquid phase could eventually be realized by installing a filter for the solid
particles at the outlet, thus, allowing a longer residence τS of the solid phase.
In the numerical calculations, the steady state was identified when the relative deviation
error in the supersaturation was below 5 × 10−7.
Both the HR-FVS and the DG-scheme were applied to solve the model. However, due to
the smoothness of the solution and because of the arrival of steady state both schemes were
found to produce almost the same results. Therefore, to avoid unnecessary repetitions, only
the results of one scheme are presented at a time.
Problem 1
Case A: Continuous seeding without fines dissolution
In this case, the crystallizer is continuously seeded with the preferred enantiomer. The
corresponding seeds size distribution is given as
F(p)seeds =
a1
zAa
exp
(
−0.5
(
1
a3
ln
(
z
a2
))2)
, (5.39)
91
where the values of constants a1, a2 and a3 are given in the Table 5.1. Moreover, the
normalization factor is given as
Aa =kvρS
m(p)seeds · τS
µ(p)3 . (5.40)
The initial masses of the preferred and counter enantiomers in the crystallizer are taken
as m(p)0 = m
(c)0 = 0.0478 kg. Both HR-FVS and DG-scheme produced the same results.
Therefore, the results of HR-FVS are presented only.
In this test problem, the minimum and maximum crystal sizes of interest are taken as
z0 = 1.0 × 10−10 m and zmax = 1.0 × 10−2 m, respectively. The corresponding computa-
tional domain is subdivided into 200 grid points. The remaining parameters are given in
Table 5.1. The physicochemical parameters correspond to the enantiomers of the amino
acid threonine and the solvent water, see [17, 18]. The temperature of the crystallizer was
assumed to be constant at 33 oC.
The simulation results corresponding to steady state conditions are displayed in Table 5.2
for different mass flow of seeds. It can be observed that an increase in the ratio of solid to
the liquid residence times, τS/τL, diminishes the purity of the preferred enantiomer due to
the production of an increasing amount of counter enantiomer as an impurity, while the
productivity, yield, and mean crystal size of the preferred enantiomer are improved. As τS
increases, the crystals of the preferred enantiomer have more time to grow and, hence, the
overall productivity, yield and the mean crystal size are enhanced. The productivity and
yield along with the investment of seeds are increased by increasing the mass flow of seeds.
It is clear that with large amount of seeding more crystals are produced, providing the
potential for high productivity. However, the mean crystal size reduces on increasing the
mass of seeds. The reason is obvious, by increasing the mass flow (amount) of seeds the
92
existing supersaturation is consumed by a large number of crystals and, therefore, the mean
crystal size becomes smaller. However, by investing less seeds, the existing supersaturation
is utilized by only a fewer crystals, leading to larger mean size of the crystals. Finally, the
steady-state is reached after a longer time when the ratio τS/τL is larger. However, the
time needed for steady state reduces by increasing the mass flow of seeds.
Figures 5.2 and 5.3 also justify and support the above discussion. They show a compari-
son of goal functions for different mass flows of seeds and different residence times under
isothermal condition. It can bee seen that supersaturation decreases with increasing mass
flow of seeds. As explained above, an increase in the mass flow (amount) of seeds enhances
the consumption rate of the solute mass (in the solution). Moreover, the time needed to
achieve steady state becomes shorter by increasing the mass flow of seeds which is also
clear from the Table 5.2. The left and right hand side figures show the same behavior but
an increase in the residence time of the solid phase reduces the steady state value of the
supersaturation. As the residence time increases, the crystals remain in the crystallizer for
a longer time and the surface area of the crystals increases, resulting in a high depletion
of the supersaturation. It can be further observed that purity increases by increasing the
mass flow of seeds. The mean crystal size, productivity, and yield plots also agree with
the data in the tables. Initially, overshoots can be observed in the results when the ratio
of residence time is 0.3. It means that, the productivity increases for a certain time and
then bounces back. The reason can be the high initial supersaturation which is depleted
with time before reaching to the steady state condition. As the yield depends on the
productivity, it shows a similar behavior. At the startup, the values of productivity and
yield remain negative for a certain time and then become positive. During that period,
the crystallizer is seeded with the seeds of preferred enantiomer which stay for a certain
time before being taken out as a product. In other words, the productivity does not start
93
instantaneously due to the residence time of the solid phase. Moreover, in Figure 5.2 purity
drops temporarily before reaching the steady state due to the stronger relative (negative)
impact of primary nucleation of counter enantiomer. This effect reduces gradually when a
steady state condition is approaching.
Finally, Figure 5.4 presents the percentage relative errors in the mass balances at steady
state condition and the crystal size distributions for different mass flows of seeds. It can
be observed that the error in the mass balances is moderately increasing until m(p)seeds ≈
0.04 g/min and then bounces back. Moreover, bigger crystals are produced for lower mass
flow rate of seeds which also justifies the results of Table 5.2.
In summary, the purity diminishes by increasing the ratio between residence times of the
solid to the liquid phases, while the productivity, yield, and mean crystal size are enhanced.
An increase in this ratio reduces the mass flow of seeds and increases the time needed to
achieve steady state. Moreover, the supersaturation and mean crystal size are reduced
with increasing the mass flow of seeds, while productivity, yield and purity are improved.
The errors in mass balances are below 0.15% for a wide range of mass flow of seeds.
94
Table 5.1: Physicochemical parameters related tothreonine-water system ([17, 18]).
Parameters Symbols Value UnitVolume shape factor kv 0.122 [−]Density of solid phase ρS 1250 [kg/m3]Crystal growth rate exponent kg,eff,0 2.98 × 109 [m/min]Nucleation rate exponent kb,sec,0 2.38 × 1026 [m−3/min]Volume of the liquid phase VL 4.496 × 10−4 [m3]
Volumetric flow rate VL 0.4 × 10−5 [m3/min]Volume of the pipe Vpipe 4.0 × 10−5 [m3]
Volumetric flow rate Vpipe 0.4 × 10−5 [m3/min]Mass of the solvent msolv 0.3843 [kg]Constant for density K1 1.00023 [cm3g−1]Constant for density K2 4.68 × 10−6 [cm3g−1K−2]Constant for density K3 0.3652 [cm3g−1]Density of water ρ0 0.9947 [g cm−3]Density of fresh solution ρf 1.067 × 103 [kg m−3]Parameter for temperature dependence KT 1874.4 [K]Parameter for mass fraction dependence KW 0.290 [−]Universal gas constant R 8.314 J mol/KGrowth rate exponent g 1.1919 [−]Parameter for crystal growth aASL 2.0209 × 104 [m−1]Parameter for crystal growth dASL −4.066 × 10−1 [−]Activation energy for crystal growth EA,g 75.54 × 103 [J mol−1]Activation energy for nucleation EA,b 63.83 × 103 [J mol−1]Secondary nucleation exponent bsec 4.80 [−]Exponent for third moment nsec 3.0258 [−]
Nucleation constant kb,prim 3.847 × 10−2 [ (kg m−3)−7/3
min K]
Constant for exponential law aprim 4.304 × 10−3 [−]Constant of surface area dependency A1 -6.9584 [m−2]Solubility constant (Eq.(5.23)) aeq 9.83 × 10−2 [−]Solubility constant (Eq.(5.23)) beq −7.45 × 10−2 [−]Seeds distribution constant (Eq.(5.39)) a1 0.014 [−]Seeds distribution constant (Eq.(5.39)) a2 0.0009 [m]Seeds distribution constant (Eq.(5.39)) a3 0.288 [−]
95
Table 5.2: Problem 1 (Case A): Values of goal functions at τL = 112.41 min.τSτL
tstd [min] m(p)seed [ g
min] Pu(p) [%] Pr(p) [ kg
min m3 ] Y (p) [%] z(p) [mm]
1.0 646 0.0089 99.74 0.037 3.91 1.312.0 1411 0.0044 98.95 0.043 4.54 1.541.0 676 0.0356 99.92 0.060 6.29 1.132.0 1315 0.0178 99.54 0.062 6.60 1.271.0 721 0.0623 99.94 0.067 7.08 1.082.0 1318 0.0311 99.68 0.069 7.29 1.18
.
96
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 300098.8
99
99.2
99.4
99.6
99.8
time [min]
purit
y [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 300098.8
99
99.2
99.4
99.6
99.8
100
time [min]
purit
y [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
time[min]
prod
uctiv
ity [k
g/m
in m
3 ]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
time [min]
prod
uctiv
ity [k
g/m
in m
3 ]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
Figure 5.2: Problem 1 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
97
0 500 1000 1500 2000 2500 30000
1
2
3
4
5
6
7
time [min]
yiel
d [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30000
1
2
3
4
5
6
7
time [min]
yiel
d [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30000.9
0.95
1
1.05
1.1
1.15
1.2
1.25
1.3
1.35
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30000.9
1
1.1
1.2
1.3
1.4
1.5
1.6
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
Figure 5.3: Problem 1 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
98
0 0.05 0.1 0.15 0.20.1
0.11
0.12
0.13
0.14
0.15
m(p)seeds
rela
tive
erro
r in
mas
s ba
lanc
e [%
]
0 1 2 3 4 5
x 10−3
0
1
2
3
4
5
6
7
8
9
10x 10
6
crystal size [m]
CS
D [1
/m]
seeds distribution
m(p)seeds = 0.005 g/min
m(p)seeds = 0.0067 g/min
m(p)seeds = 0.01 g/min
m(p)seeds = 0.02 g/min
Figure 5.4: Problem 1 (case A): Error in mass balance and CSDs.
Case B: Periodic seeding without fines dissolution
In this test case, continuous crystallization along with periodic seeding of preferred enan-
tiomer is analyzed. The model equations are exactly the same as for the continuous seeding
case in Case A. In Table 5.3, ton represents the time at which seeding is switched on and
toff denotes the period of time at which seeding is switched off. The seed flow rate can
be varied in contrast to the continuous seeding. In this table, a comparison of simulation
results for continuous and periodic seeding strategies are presented. A schematic diagram
of the periodic seeding is shown in Figure 5.5. In that figure, the shaded areas represent the
seeding times (ton) and the non-shaded areas depict the periods when seeding is switched
off (toff ).
In Table 5.3, the simulation results are presented for different ratios of the residence times.
The increase in τS reduces the purity of the preferred enantiomer in the periodic case. As
residence time increases, more nuclei of the counter enantiomer are produced as an impurity
of the product and, thus, the purity is lower. It can also be observed that the purity has a
99
seeding off
seeding on
time [min]
mas
s flo
w o
f see
ds [g
/min
]
Figure 5.5: Schematic diagram for periodic seeding.
decreasing behavior for increasing toff . On the other hand, as τS increases the crystals of
the preferred enantiomer gets more time to grow and, thus, the overall productivity and
yield enhances. Additionally, the productivity and yield are decreased by increasing toff .
It is obvious that when seeding is switched off for a longer time, the mean crystal size will
increase by increasing the residence time τS due to longer stay of seeds in the crystallizer
and because of utilizing the existing supersaturation by fewer crystals. The mean crystal
size also increases by increasing toff for all considered residence times due to the utiliza-
tion of existing supersaturation by fewer crystals. Table 5.3 also shows that the feed flow
rate of seeds decreases by increasing the residence time. A comparison of the continuous
seeding case in Table 5.2 with the current periodic case in Table 5.3 shows that continuous
seeding gives better purity, productivity and yield, while the mean crystal size is smaller
and investment (mass flow of seeds) remains the same in all considered residence times.
On the other hand, the periodic seeding gives larger crystals at low investment (feed flow
rate) of seeds per residence time. Finally, the time needed to achieve steady-state becomes
longer on increasing τS, while keeping the liquid residence time τL fixed. This behavior is
exactly similar to constant, time independent seeding case.
100
Figures 5.6 and 5.7 show the goal functions at different periods of seeding and for different
ratios of residence times. It can be observed that supersaturation decreases with reducing
time period of seeding, because a decrease in toff of seeds speeds up the consumption of
solute mass. Both figures show the same trends. However, a comparison of the plots in
Figures 5.6 and 5.7 shows that, an increase in the residence time of solid phase gives a
decrease in the steady state supersaturation. The reason is clear, as the solid residence
time increases, the stay of crystals in the crystallizer for a longer time and, thus, the surface
area of crystals increases by consuming more supersaturation. Further, the steady-state
time delays as the residence time of the solid phase increases. Oscillations are also visible
which indicate the conditions of periodic seeding. These oscillations are less prominent in
the right hand side plots of Figures 5.6 and 5.7 due to an increase in the residence time
of the solid phase. The plots show that the purity decreases by increasing toff . Initially,
the decrease in purity is due to increasing the amount of counter enantiomer but after
a certain time the amount of preferred enantiomer starts increasing which improves the
purity. It can also be observed that an increase in toff produces large size crystals which
is also clear from Table 5.3. The behavior of productivity is also similar to that in Table
5.3. It is evident that larger toff gives lower productivity. Yield is directly related to the
productivity, thus, it shows a similar behavior as the productivity. Moreover, in Figure 5.6
the purity drops temporarily before reaching the steady state due to the stronger relative
impact (negative) of primary nucleation of the counter enantiomer. This effect reduces
gradually when the steady state conditions are approached.
In summary, the periodic seeding reduces the productivity and yield, while it improves the
mean crystal size. The purity is slightly reduced with periodic seeding (approx. 0.1%).
Moreover, the investment of seed crystals is reduced.
101
Tab
le5.3:
Prob
lem1
(caseB
):Perio
dic
seedin
g:m
(p)
seeds=
4g
and
τL
=112.4
min
.τS
τL
seedin
gty
pe
tstd
[min
]ton
[min
]toff
[min
]m
(p)
seed[
gm
in]
Pu
(p)[%
]P
r(p
)[kg
min
m3 ]
Y(p
)[%]
z(p
)[mm
]
1.0P
67010
50.0244
99.870.051
5.501.18
1.0P
67010
100.0189
99.840.045
4.671.23
1.0P
67010
200.0126
99.790.040
4.261.27
1.0C
676-
-0.0356
99.920.060
6.291.13
2.0P
132010
50.0125
99.410.044
4.671.34
2.0P
132110
100.0089
99.300.033
3.471.40
2.0P
132110
200.0051
99.100.021
2.251.46
2.0C
1315-
-0.0178
99.540.062
6.601.27
102
0 500 1000 1500 20001.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
0 500 1000 1500 2000 2500 3000 3500 40001.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
0 500 1000 1500 200099
99.1
99.2
99.3
99.4
99.5
99.6
99.7
99.8
99.9
100
time [min]
purit
y [%
]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
0 500 1000 1500 2000 2500 3000 3500 400099
99.1
99.2
99.3
99.4
99.5
99.6
99.7
99.8
99.9
100
time [min]
purit
y [%
]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
0 500 1000 1500 20000
0.01
0.02
0.03
0.04
0.05
0.06
time [min]
prod
uctiv
ity [k
g/m
in m
3 ]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
0 500 1000 1500 2000 2500 3000 3500 40000
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
time [min]
prod
uctiv
ity [k
g/m
in m
3 ]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
Figure 5.6: Problem 1 (Case B): Periodic seeding: Left τS/τL = 1, Right: τS/τL = 2.
103
0 500 1000 1500 20000
1
2
3
4
5
6
time [min]
yiel
d [%
]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
0 500 1000 1500 2000 2500 3000 3500 40000
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
time [min]
yiel
d [%
]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
0 500 1000 1500 20000.95
1
1.05
1.1
1.15
1.2
1.25
1.3
1.35
time [min]
mea
n cr
ysta
l siz
e [m
m]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
0 500 1000 1500 2000 2500 3000 3500 4000
1
1.1
1.2
1.3
1.4
1.5
1.6
time [min]
mea
n cr
ysta
l siz
e [m
m]
ton
=10, toff
=5
ton
=10, toff
=10
ton
=10, toff
=20
Figure 5.7: Problem 1 (Case B): Periodic seeding: Left τS/τL = 1, Right: τS/τL = 2.
104
Case C: Continuous seeding with fines dissolution
In this problem, a continuous seeding with fines dissolution is investigated. Fines particles
below the critical size zcrit = 6.0 × 10−4 m are taken out from the crystallizer along with
the solution into the dissolution unit (the pipe). After a certain time delay the particle-free
solution re-enters to the crystallizer. The fines dissolution unit is usually equipped with a
heat exchanger for dissolving fines followed by a heat sink which brings back the solution
temperature to that of the crystallizer. A complete model Eq. (5.30) is employed in this
problem. The selection (death) function is taken as
ψ(z) =
0.6, if z ≤ zcrit ,0, otherwise .
(5.41)
In this case, the simulation results with continuous seeding and fines dissolution are ana-
lyzed. Table 5.4 shows a comparison of fines dissolution (yes) and without fines dissolution
(no) by considering different ratios of residence times and fixed mass flow of seeds. It can
be seen that the purity is higher in the case of fines dissolution compared to that with-
out fines dissolution, as the fines dissolution diminishes the number of counter enantiomer
crystals in the crystallizer. From the Table 5.4, it can be realized that the productivity and
yield are slightly lower in the case of fines dissolution for both ratios of residence times.
The fines dissolution enhances the supersaturation of the solution which in turn increases
the growth rate and thus, improves the mean crystal size.
Table 5.4: Problem 1 (case C): Results with FD (yes) and without FD (no)τSτL
FD tstd [min] m(p)seed [ g
min] Pu(p) [%] Pr(p) [ kg
min m3 ] Y (p) [%] z(p) [mm]
1.0 no 676 0.0356 99.92 0.060 6.29 1.131.0 yes 648 0.0356 99.99 0.060 6.25 1.152.0 no 1315 0.0178 99.54 0.062 6.60 1.272.0 yes 1303 0.0178 99.97 0.062 6.56 1.29
105
Problem 2
The idea behind choosing this problem is its practical considerations, see [16-19] and ref-
erences therein. Here, a different seeds distribution of preferred enantiomer is considered,
F(p)seeds(0, z) =
1√
(2π)σIa
.1
zexp(−0.5)
(
ln (z) − z
σ
)2
, (5.42)
where
Ia =kv.ρc
mseeds
µ(p)3 (0) (5.43)
We do not seed the crystallizer with counter enantiomer, thus
F(c)seeds(0, z) = 0. (5.44)
Here we assumed σ = 0.3947m, z = −6.8263m, while mseeds is the mass of seeds. The
maximum crystal size that was expected is zmax = 0.005m which is subdivided into 400
grid points. The final simulation time was taken as 3000 minutes. Both, HR-FVS are
applied to solve the problem and they produced the same results. Here, we present the
results of the DG-scheme only.
Case A: Continuous seeding without fines dissolution
Figure 5.8 shows a comparison of CSDs for different mass flows of seeds. The goal functions
are presented in Figures 5.9 and 5.10 and show the same behavior that was seen in Case
A of Problem 1. After analysis of the figures we conclude that the purity diminishes by
increasing the ratio between residence times of the solid to the liquid phases, while the
productivity, yield, and mean crystal size are improved. An increase in this ratio reduces
the mass flow of seeds and increases the time needed to achieve steady state. Moreover,
the supersaturation and mean crystal size are reduced with increasing mass flow of seeds,
while productivity, yield and purity are improved.
Moreover, bigger crystals are produced for greater masses of the seed crystals. The figures
also justify the results of Table 5.5
106
Table 5.5: Problem 2 (Case A): Values of goal functions at τL = 112.41 min.τSτL
tstd [min] m(p)seed [ g
min] Pu(p) [%] Pr(p) [ kg
min m3 ] Y (p) [%] z(p) [mm]
1.0 872 0.0089 99.03 0.0005 0.051 3.762.0 1331 0.0044 93.30 0.0008 0.084 3.491.0 832 0.0356 99.74 0.0014 0.0145 5.152.0 1420 0.0178 98.05 0.0015 0.154 5.291.0 841 0.0623 99.84 0.0021 0.225 5.422.0 1503 0.0311 98.82 0.0022 0.234 5.39
0 0.002 0.004 0.006 0.008 0.010
0.5
1
1.5
2
2.5x 10
4
Crystal Size [m]
CS
D [1
/m]
Preferred Enantiomer Number Density
mseeds(p) = 1g
mseeds(p) = 4g
mseeds(p) = 7g
0 0.002 0.004 0.006 0.008 0.010
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2x 10
4
Crystal Size [m]
CS
D [1
/m]
Counter Enantiomer Number Density
mseeds(p) = 1g
mseeds(p) = 4g
mseeds(p) = 7g
Figure 5.8: Problem 2 (Case A): CSDs for the different mass of seed crystal.
107
0 500 1000 1500 2000 2500 30001.075
1.08
1.085
1.09
1.095
1.1
Time [min]
Sup
ersa
tura
tion
[−]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30001.075
1.08
1.085
1.09
1.095
1.1
Time [min]
Sup
ersa
tura
tion
[−]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30002.5
3
3.5
4
4.5
5
5.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30001.5
2
2.5
3
3.5
4
4.5
5
5.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
Figure 5.9: Problem 2 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
108
0 500 1000 1500 2000 2500 300099
99.1
99.2
99.3
99.4
99.5
99.6
99.7
99.8
99.9
100
Time [min]
Pur
ity [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 300093
94
95
96
97
98
99
100
Time [min]
Pur
ity [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−0.14
−0.12
−0.1
−0.08
−0.06
−0.04
−0.02
0
0.02
Time [min]
Pro
duct
ivity
[kg/
min
m3 ]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
Time [min]
Pro
duct
ivity
[kg/
min
m3 ]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−16
−14
−12
−10
−8
−6
−4
−2
0
2
Time [min]
Yie
ld [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−8
−7
−6
−5
−4
−3
−2
−1
0
1
Time [min]
Yie
ld [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
Figure 5.10: Problem 2 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
109
Case B: Continuous seeding with fines dissolution
In this case, a continuous seeding with fines dissolution is investigated. Fines particles
below the critical size zcrit = 6.0 × 10−4 m are taken out from the crystallizer along with
the solution into the dissolution unit (the pipe). After a certain time delay the particle-free
solution re-enters to the crystallizer. The fines dissolution unit is usually equipped with a
heat exchanger for dissolving fines followed by a heat sink which brings back the solution
temperature to that of the crystallizer. A complete model Eq. (5.1) is employed in this
problem. In this case, the simulation results with continuous seeding and fines dissolution
are analyzed. Table 5.6 shows a comparison of fines dissolution (yes) and without fines
dissolution (no) by considering different ratios of residence times and fixed mass flow of
seeds. It can be seen that the purity is greater in the case of fines dissolution compared
to that without fines dissolution, as the fines dissolution reduces the number of counter
enantiomer crystals in the crystallizer. From Table 5.6, it can be seen that the productivity
and yield are slightly enhanced in the case of fines dissolution for both ratios of residence
times. Supersaturation enhances in the fines dissolution that is the reason why we see
that as increase in the growth rate improves the mean crystal size. Figures 5.11 and 5.12
present the plots of goal function with respect to time which also justify the results of the
Table 5.6.
Table 5.6: Problem 2 (Case B): Results with FD (yes) and without FD (no)τSτL
FD tstd [min] m(p)seed [ g
min] Pu(p) [%] Pr(p) [ kg
min] Y (p) [%] z(p) [mm]
1.0 no 832 0.0356 99.74 0.0014 0.145 5.151.0 yes 962 0.0356 99.77 0.0014 0.147 5.172.0 no 1315 0.0178 98.05 0.0015 0.154 5.292.0 yes 1700 0.0178 98.28 0.0015 0.157 5.32
110
0 500 1000 1500 2000 2500 30001.075
1.08
1.085
1.09
1.095
1.1
Time [min]
Sup
ersa
tura
tion
[−]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30001.075
1.08
1.085
1.09
1.095
1.1
Time [min]
Sup
ersa
tura
tion
[−]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30002.5
3
3.5
4
4.5
5
5.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30001.5
2
2.5
3
3.5
4
4.5
5
5.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
Figure 5.11: Problem 2 (Case B): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
111
0 500 1000 1500 2000 2500 300099.1
99.2
99.3
99.4
99.5
99.6
99.7
99.8
99.9
100
100.1
Time [min]
Pur
ity [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 300093
94
95
96
97
98
99
100
Time [min]
Pur
ity [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−0.14
−0.12
−0.1
−0.08
−0.06
−0.04
−0.02
0
0.02
Time [min]
Pro
duct
ivity
[kg/
min
m3 ] m
seeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
Time [min]
Pro
duct
ivity
[kg/
min
m3 ]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−16
−14
−12
−10
−8
−6
−4
−2
0
2
Time [min]
Yie
ld [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−8
−7
−6
−5
−4
−3
−2
−1
0
1
Time [min]
Yie
ld [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
Figure 5.12: Problem 2 (Case B): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
112
Case C: Periodic seeding without fines dissolution
In this case, continuous crystallization along with periodic seeding of preferred enantiomer
is examined. The model equations are exactly the same as for the continuous seeding case
in test problem 1. In Table 5.7, ton represents the time at which seeding is switched on and
toff denotes the period of time at which seeding is switched off. The seed flow rate can
be varied in contrast to the continuous seeding. In this table, a comparison of simulation
results for continuous and periodic seeding strategies are presented.
The same result as seen in case C of test problem 1 can be observed here. In Table 5.7,
the simulation results are presented for different ratios of the residence times. The increase
in τS reduces the purity of the preferred enantiomer in the periodic case. The longer the
residence time, the larger the formation of the counter enantiomer as an impurity of the
product and, thus, the lower the purity. It can also be observed that the purity has a
decreasing behavior for increasing toff . On the other hand, as τS increases the crystals of
the preferred enantiomer get more time to grow and, thus, the overall productivity and
yield enhances.
Figures 5.13 and 5.14 show the goal functions at different periods of seeding and for dif-
ferent ratios of residence times. It can be observed that supersaturation decreases with
reducing time period of seeding, because a decrease in toff of seeds speeds up the con-
sumption of solute mass. Both figures show the same trends. However, a comparison of
the plots in Figures 5.13 and 5.14 shows that, an increase in the residence time of the solid
phase results in a decrease in the steady state supersaturation.
Keeping all the discussions, we can conclude that the periodic seeding reduces the produc-
113
tivity and yield, while it improves the mean crystal size. The purity is slightly reduced
with periodic seeding (approx. 0.1%). Moreover, the investment of seed crystals is reduced.
0 500 1000 1500 20001.09
1.091
1.092
1.093
1.094
1.095
1.096
1.097
Time [min]
Sup
ersa
tura
tion
[−]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 40001.09
1.091
1.092
1.093
1.094
1.095
1.096
1.097
Time [min]
Sup
ersa
tura
tion
[−]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 20001.5
2
2.5
3
3.5
4
4.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 40000.5
1
1.5
2
2.5
3
3.5
4
4.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
Figure 5.13: Problem 2 (Case C): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
114
0 500 1000 1500 200097.5
98
98.5
99
99.5
100
Time [min]
Pur
ity [%
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 400082
84
86
88
90
92
94
96
98
100
Time [min]
Pur
ity [%
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000−0.055
−0.05
−0.045
−0.04
−0.035
−0.03
−0.025
−0.02
−0.015
Time [min]
Pro
duct
ivity
[kg/
min
m3 ]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 4000−0.03
−0.025
−0.02
−0.015
−0.01
−0.005
Time [min]
Pro
duct
ivity
[kg/
min
m3 ] t
on = 10 , t
off = 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000−6
−5.5
−5
−4.5
−4
−3.5
−3
−2.5
−2
−1.5
Time [min]
Yie
ld [%
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 4000−3
−2.5
−2
−1.5
−1
−0.5
Time [min]
Yie
ld [%
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
Figure 5.14: Problem 2 (Case C): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
115
Tab
le5.7:
Prob
lem2
(Case
C):
Perio
dic
seedin
g:m
(p)
seeds=
4g
and
τL
=112.4
min
.τS
τL
seedin
gty
pe
tstd
[min
]ton
[min
]toff
[min
]m
(p)
seed[
gm
in]
Pu
(p)[%
]P
r(p
)[kg
min
m3 ]
Y(p
)[%]
z(p
)[mm
]
1.0P
67010
50.0244
99.430.0198
2.0884.45
1.0P
67010
100.0189
99.070.021
2.243.71
1.0P
67010
200.0126
98.080.018
1.922.59
1.0C
832-
-0.0356
99.740.0014
0.01455.15
2.0P
132010
50.0125
95.750.011
1.0774.33
2.0P
132010
100.0089
92.660.010
1.0693.33
2.0P
132010
200.0051
84.200.007
0.7752.282
2.0C
1420-
-0.0178
98.050.0015
0.1545.29
116
Problem 3
The parameter of this problem has been used in the experiments, see [16-19] and references
therein. Another seed distribution of preferred enantiomer is considered:
F(p)seeds(0, z) =
107
√
(2π)σIa
.1
zexp
(
ln (z) − z
σ√
2
)2
, (5.45)
where
Ia =kv.ρc
mseeds
µ(p)3 (0) (5.46)
Crystals of the counter are not initially present, i.e.
F(c)seeds(0, z) = 0. (5.47)
Here we assumed σ = z15
, z = 4.10−4m, while mseeds is the mass of the initial seeds.
The maximum time simulation is 3000 minutes for crystal size zmax = 0.002m which is
subdivided into 400 grids points.
Case A: Continuous seeding without fines dissolution
Figure 5.15 shows a comparison of CSDs for different mass flow of seeds. Analysis of the
figures 5.16 and 5.17 the same behavior of the graphs is seen as in case A of test problems 1
and 2. After analysis of the figures we conclude that the purity diminishes by increasing the
ratio between residence times of the solid to the liquid phases, while the productivity, yield,
and mean crystal size are improved. An increase in this ratio reduces the mass flow of seeds
and increases the time needed to achieve steady state. Moreover, the supersaturation and
mean crystal size are reduced with increasing the mass flow of seeds, while productivity,
yield and purity are improved.
Moreover, bigger crystals are produced for greater mass of the seed crystals both for the
preferential and counter which also justifies the results of Table 5.8.
117
Table 5.8: Problem 3 (Case A): Values of goal functions at τL = 112.41 min.τSτL
tstd [min] m(p)seed [ g
min] Pu(p) [%] Pr(p) [ kg
min m3 ] Y (p) [%] z(p) [mm]
1.0 874 0.0089 99.03 0.0005 0.052 3.762.0 1332 0.0044 93.29 0.0008 0.085 3.491.0 832 0.0356 99.74 0.0014 0.145 5.152.0 1650 0.0178 98.03 0.0015 0.158 5.291.0 842 0.0623 97.42 0.0021 0.225 5.222.0 15768 0.0311 98.81 0.0023 0.237 5.39
0 0.002 0.004 0.006 0.008 0.010
0.5
1
1.5
2
2.5x 10
4
Crystal Size [m]
CS
D [1
/m]
Preferred Enantiomer Number Density
mseeds(p) = 1g
mseeds(p) = 4g
mseeds(p) = 7g
0 0.002 0.004 0.006 0.008 0.010
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2x 10
4
Crystal Size [m]
CS
D [1
/m]
mseeds(p) = 1g
mseeds(p) = 4g
mseeds(p) = 7g
Figure 5.15: Problem 3 (Case A): CSDs for the different mass of seed crystal.
118
0 500 1000 1500 2000 2500 30001.075
1.08
1.085
1.09
1.095
1.1
Time [min]
Sup
ersa
tura
tion
[−]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30001.075
1.08
1.085
1.09
1.095
1.1
Time [min]
Sup
ersa
tura
tion
[−]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30002.5
3
3.5
4
4.5
5
5.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
] mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30001.5
2
2.5
3
3.5
4
4.5
5
5.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
Figure 5.16: Problem 3 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
119
0 500 1000 1500 2000 2500 300099
99.1
99.2
99.3
99.4
99.5
99.6
99.7
99.8
99.9
100
Time [min]
Pur
ity [%
] mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 300093
94
95
96
97
98
99
100
Time [min]
Pur
ity [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−0.14
−0.12
−0.1
−0.08
−0.06
−0.04
−0.02
0
0.02
Time [min]
Pro
duct
ivity
[kg/
min
m3 ] m
seeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
Time [min]
Pro
duct
ivity
[kg/
min
m3 ] m
seeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−16
−14
−12
−10
−8
−6
−4
−2
0
2
Time [min]
Yie
ld [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−8
−7
−6
−5
−4
−3
−2
−1
0
1
Time [min]
Yie
ld [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
Figure 5.17: Problem 3 (Case A): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
120
Case B: Continuous seeding with fines dissolution
In this case, a continuous seeding with fines dissolution is investigated. Fines particles
below the critical size zcrit = 6.0 × 10−4 m are taken out from the crystallizer along with
the solution into the dissolution unit (the pipe). After a certain time delay the particle-free
solution re-enters to the crystallizer. The fines dissolution unit is usually equipped with a
heat exchanger for dissolving fines followed by a heat sink which brings back the solution
temperature to that of the crystallizer. A complete model Eq. (5.1) is employed in this
problem. In this case, the simulation results with continuous seeding and fines dissolution
are analyzed. Table 5.9 shows a comparison of fines dissolution (yes) and without fines
dissolution (no) by considering different ratios of residence times and fixed mass flow of
seeds. The same behavior is seen as in case C of test problems 1 and 2, concluding that the
purity is greater in the case of fines dissolution compared to that without fines dissolution,
as the fines dissolution reduces the number of counter enantiomer crystals in the crystallizer.
From Table 5.9, it can be seen that the productivity and yield are slightly lower in the case
of fines dissolution for both ratios of residence times. Supersaturation enhances in the fines
dissolution that is the reason why we see increase in the growth rate improves the mean
crystal size. Figures 5.18 and 5.19 present the plots of goal function with respect to time
which also justify the results of the Table 5.9.
Table 5.9: Problem 3 (Case B): Results with FD (yes) and without FD (no)τSτL
FD tstd [min] m(p)seed [ g
min] Pu(p) [%] Pr(p) [ kg
min] Y (p) [%] z(p) [mm]
1.0 no 832 0.0356 99.74 0.0014 0.145 5.151.0 yes 961 0.0356 99.77 0.0014 0.147 5.182.0 no 1650 0.0178 98.03 0.0015 0.158 5.292.0 yes 1700 0.0178 98.28 0.0014 0.157 5.32
121
0 500 1000 1500 2000 2500 30001.075
1.08
1.085
1.09
1.095
1.1
Time [min]
Sup
ersa
tura
tion
[−]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30001.075
1.08
1.085
1.09
1.095
1.1
Time [min]
Sup
ersa
tura
tion
[−]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30002.5
3
3.5
4
4.5
5
5.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 30001.5
2
2.5
3
3.5
4
4.5
5
5.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
Figure 5.18: Problem 3 (Case B): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
122
0 500 1000 1500 2000 2500 300099.1
99.2
99.3
99.4
99.5
99.6
99.7
99.8
99.9
100
100.1
Time [min]
Pur
ity [%
] mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 300093
94
95
96
97
98
99
100
Time [min]
Pur
ity [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−0.14
−0.12
−0.1
−0.08
−0.06
−0.04
−0.02
0
0.02
Time [min]
Pro
duct
ivity
[kg/
min
m3 ] m
seeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
Time [min]
Pro
duct
ivity
[kg/
min
m3 ]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−16
−14
−12
−10
−8
−6
−4
−2
0
2
Time [min]
Yie
ld [%
]]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
0 500 1000 1500 2000 2500 3000−8
−7
−6
−5
−4
−3
−2
−1
0
1
Time [min]
Yie
ld [%
]
mseeds(p) =1g
mseeds(p) =4g
mseeds(p) =7g
Figure 5.19: Problem 3 (Case B): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
123
Case C: Periodic seeding without fines dissolution
In this case, continuous crystallization along with periodic seeding of preferred enantiomer
is examined. The model equations are exactly the same as for the continuous seeding case
in Problem 1. In Table 5.10, ton represents the time at which seeding is switched on and
toff denotes the period of time at which seeding is switched off. The seed flow rate can
be varied in contrast to the continuous seeding. In this table, a comparison of simulation
results for continuous and periodic seeding strategies are presented.
The figures 5.20 and 5.21 show results for the comparison of different residence times
taken. Keeping all the discussions, we can conclude that the periodic seeding reduces the
productivity and yield, while it improves the mean crystal size. The purity is slightly
reduced with periodic seeding (approx. 0.1%). Moreover, the investment of seed crystals
is reduced.
124
0 500 1000 1500 20001.09
1.091
1.092
1.093
1.094
1.095
1.096
1.097
Time [min]
Sup
ersa
tura
tion
[−]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 40001.09
1.091
1.092
1.093
1.094
1.095
1.096
1.097
Time [min]
Sup
ersa
tura
tion
[−]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 20001.5
2
2.5
3
3.5
4
4.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 40000.5
1
1.5
2
2.5
3
3.5
4
4.5
Time [min]
Mea
n C
ryst
al S
ize
[mm
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
Figure 5.20: Problem 3 (Case C): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
125
0 500 1000 1500 200097.5
98
98.5
99
99.5
100
Time [min]
Pur
ity [%
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 400082
84
86
88
90
92
94
96
98
100
Time [min]
Pur
ity [%
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000−0.055
−0.05
−0.045
−0.04
−0.035
−0.03
−0.025
−0.02
−0.015
Time [min]
Pro
duct
ivity
[kg/
min
m3 ]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 4000−0.03
−0.025
−0.02
−0.015
−0.01
−0.005
Time [min]
Pro
duct
ivity
[kg/
min
m3 ]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000−6
−5.5
−5
−4.5
−4
−3.5
−3
−2.5
−2
−1.5
Time [min]
Yie
ld [%
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
0 500 1000 1500 2000 2500 3000 3500 4000−3
−2.5
−2
−1.5
−1
−0.5
Time [min]
Yie
ld [%
]
ton
= 10 , toff
= 5
ton
= 10 , toff
= 10
ton
= 10 , toff
= 20
Figure 5.21: Problem 3 (Case C): Goal functions: Left τS/τL = 1.0, Right: τS/τL = 2.0.
126
Tab
le5.
10:
Pro
ble
m3
(Cas
eC
):Per
iodic
seed
ing:
m(p
)se
eds=
4g,τ L
=11
2.4m
in.
τ S τ Lse
edin
gty
pe
t std
[min
]t o
n[m
in]
t off
[min
]m
(p)
seed
[g
min
]P
u(p
)[%
]P
r(p)[
kg
min
m3]
Y(p
)[%
]z(
p)[m
m]
1.0
P67
010
50.
0244
99.4
30.
0195
2.06
4.45
61.
0P
670
1010
0.01
8999
.07
0.02
12.
243.
721.
0P
670
1020
0.01
2698
.07
0.18
1.92
2.59
1.0
C83
2-
-0.
0356
99.7
40.
0014
0.14
55.
152.
0P
1320
105
0.01
2595
.75
0.01
01.
077
4.33
2.0
P13
2010
100.
0089
92.6
50.
010
1.07
03.
332.
0P
1320
1020
0.00
5184
.20
0.00
70.
775
2.28
2.0
P13
20−
−0.
0178
98.0
30.
0015
0.15
85.
29
127
128
Chapter 6
Coupled Continuous Preferential Crystallization
Separation of chiral molecules is of great interest in chemical and pharmaceutical indus-
tries as many (bio)organic molecules are chiral. In most of the cases only one enantiomer
has desired properties for therapeutic activities or metabolism, while the other one may
be inactive or may produce undesired effects. Several special techniques can be applied
to separate enantiomeric mixtures, such as chromatography, classical non-biological reso-
lutions via the formation of diastereomers, biological methods, non-biological asymmetric
synthesis, and membrane technologies. An attractive alternative and energy efficient way
to separate chiral substances is the enantioselective preferential crystallization. This tech-
nique is normally applied to conglomerates, a physical mixture of enantiomerically pure
crystals. A lot of work has been done on preferential crystallization in batch mode, see for
example [3, 11, 12, 13, 17, 18, 31].
Recently, Qamar et al. [66] have adopted the concept of continuous mixed-suspension,
mixed product removal (MSMPR) crystallizers to obtain pure enantiomers from racemic
mixtures in a single crystallizer. The drawback of decoupled mode (single crystallizer) is
that only a small amount of pure substance can be recovered from the racemic solution.
To overcome this limitation, in this work, two MSMPR crystallizers are connected through
the liquid phase and operated under continuous exchange of crystal free solution. These
perfectly mixed tanks are continuously fed with the racemic solutions of two enantiomers
and solid particles and liquid phase are continuously withdrawn, see Figure 6.1. On seeding
both enantiomers, one in each vessel, the preferential crystallization of only those enan-
tiomers could be initiated in the corresponding vessels, provided that crystallization takes
place within the metastable zone where spontaneous, uncontrolled primary nucleation is
kinetically diminished. Due to the exchange, the liquid phase shows a higher overall con-
centration of the preferred enantiomer in the vessel in which it is seeded. As a result,
the supersaturation level is higher during the whole process in comparison to the case
129
without an exchange. Additionally, the concentration of the respective counter-enantiomer
in the liquid phase decreases in each tank. For the theoretical case of an infinitely high
exchange flow rate, racemic composition is reached in the liquid phase of both vessels
which corresponds to the maximum supersaturation for both preferred enantiomers. Since
the concentration of the respective counter-enantiomer is decreased, the probability for its
crystallization through primary nucleation is lowered. This leads to higher product purity
and enhances the productivity. During a starting-up period, depending on the properties
of system and process parameters, the concentration of the target enantiomer is decreasing
until a steady state is arrived where the composition is determined by the mean residence
time. Due to different kinetic mechanisms and their inherent different time constants, a
different depletion of the supersaturation for each enantiomer can be realized by an appro-
priate choice of the process conditions. As long as a critical mean residence time, where
primary nucleation may appear, is not exceeded, the concentration of the undesired counter
enantiomer maintains constant during the whole time. This fact reveals a benefit of this
continuous process in comparison to the batch one. An optimal selection of the process
conditions allows a constant production of the goal enantiomer at a high purity level.
The population balance based models are frequently used for simulating crystallization
processes. The theory of population balances began in 1960s when [27] introduced it in the
field of chemical engineering. A comprehensive overview on population balance modeling,
nucleation and growth kinetics terms and methods of solution can be found in the books
by [52, 58, 70]. During the last decades many efficient methods were developed for solv-
ing population balance models (PBMs) such as the method of characteristics introduced
in various fields by [74] and adopted for PBMs by [42, 62, 70], the method of weighted
residual or orthogonal collocation by [72], the Monte Carlo simulation by [75, 76], the fixed
and moving pivot techniques by [34, 35], and the high resolution finite volume schemes by
130
[24, 59, 66].
This work is an extension of our recent work on continuous preferential crystallization in a
single vessel, see [66]. A dynamic mathematical model is derived for simulating two-coupled
continuously operated ideally mixed MSMPR crystallizers applied for continuous prefer-
ential enantioselective crystallization. Both crystallizer are connected through exchange
pipes and are equipped with fines dissolution loops for dissolving small crystals below cer-
tain critical size. It is assumed that fines are completely dissolved at the other end of the
dissolution pipes. The developed model includes the phenomena of primary heterogeneous
nucleations, secondary nucleations, and size-dependent growth rates in each crystallizer.
The effects of different seeding and operating strategies are investigated in simulations.
The model and its parameters are based on experimental data obtained from batchwise
crystallization, see [13, 17, 18]. The semi-discrete flux-limiting finite volume scheme of
[32] is implemented to solve the model equations. To judge the quality of process some
goal functions are used, such as purity, productivity, yield and mean crystal size of the
preferred enantiomer. These goal functions provide useful information about the success
and potential of the process. To realize the advantages of coupled process, its results are
compared with those of single crystallizer obtained by [66]. It was found that coupled
process has considerably improved the values of goal functions and crystallizes both enan-
tiomers simultaneously in separate vessels. The numerical results also demonstrate the
high order accuracy, efficiency and potential of the proposed numerical method for solving
such models.
6.1 Coupled Continuous Crystallization Model
A mathematical model is presented for simulating ideally-mixed two-coupled continuous
preferential crystallizers connected through exchange pipes and equipped with fines disso-
131
L,AV&
L,AV&
L,BV&
L,BV&
Fin
es d
iss.
Fin
es d
iss.
Tank A Tank B
Figure 6.1: Principle of coupled continuous enantioselective crystallization.
lution units. The model contains four population balance equations (PBEs) for the solid
phase, two PBEs for two enantiomers in each tank, coupled with four ordinary differential
equations for the balance of solute masses of both enantiomers in both tanks, a set of al-
gebraic equations for liquid and solid mass flow rates, as well as the growth and nucleation
kinetics. The balance law for the solid phase is represented by PBEs of the form [2, 27, 54]
∂F(k)α
∂t= − ∂(G
(k)α F
(k)α )
∂z+ F
(k)seeds,α − F
(k)out,α − F
(k)pipe,α , k ∈ p, c , α ∈ A,B , (6.1)
where p stands for preferred enantiomer, c denotes counter enantiomer, and A and B rep-
resent crystallizers. The CSD of the k-th enantiomer in tank α is represented by F(k)α . In
Eq. (6.1) the term on the left hand side denotes the accumulation of crystals of size z.
The first term on the right hand side represents the convective transport in the direction
of the property coordinate z due to size-dependent crystal growth rate Gα(k). The term
F(k)seeds,α denotes the incoming particle number flux due to seeding, F
(k)out,α is the particle
number flux due to product removal, and F(k)pipe,α denotes the particle number flux to the
fines dissolution unit in each crystallizer.
132
The initial conditions for the PBEs in Eq. (6.1) are based on the assumption that initially
(at time t = 0) the solutions of both vessels are particle free. In other words
F (k)α (t = 0, z) = 0 , k ∈ p, c , α ∈ A,B . (6.2)
Assuming further, that primary nucleation leads to a crystal nuclei of minimum size zmin
and that the number density function F (k) vanishes for a arbitrary large crystal size zmax,
holds
F (k)α (t, z = zmin) =
B(k)0,α(t)
Gα(t, zmin), F (k)
α (t, z = zmax) = 0 , (6.3)
where B(k)0,α denotes the nucleation rate in the corresponding vessel. Each crystallizer is
only seeded with preferred enantiomer, thus
F(k)seeds,α(z) =
F(p)seeds,α(z) for k = p ,
0 for k = c .(6.4)
For the operation of configuration shown in Figure 6.1, the following assumption are taken
into account.
1. Both tanks are of MSMPR type and are equipped with fines removal systems.
2. The volumetric flow rates of crystallizers and dissolution units are kept constant.
3. Both vessels have the constant temperature and volume.
4. The particle-free liquid exchange flow rates to both crystallizers are the same and
constant.
5. Both primary and secondary nucleations take place at size zmin.
6. The feed flows are crystal-free.
7. All fines are dissolved in the dissolution units.
133
8. The removal of the fines are size-dependent and can be described by the product of
a classification function ψ(z) and the crystal size distributions in each crystallizer.
9. The liquid mean residence time is given by τL,α = VL,α/VL,α, where VL,α denotes the
volume of the liquid phase in the tank α and VL,α is the volumetric flow rate of the
liquid phase to that tank.
10. The characteristic solid mean residence time is given as, τS,α = VS,α/(ρS/m(p)seeds,α),
where VS,α represents the overall volume of the solid phase, ρS is the density of the
solid phase and m(p)seeds,α is the mass flow rate of the seeds of preferred enantiomer to
tank α.
11. The residence time in the pipe (plug flow) is defined as, τpipe,α = Vpipe,α/Vpipe,α. Here,
Vpipe,α denotes volume of the pipe and Vpipe,α is the volumetric flow rate to the pipe
for tank α.
For each vessel, the number density function of withdrawn crystals as a product is assumed
to be equal to the number density function inside that vessel. Therefore, the right hand
sided second and third terms of Eq. (6.1) are given as
F(k)seeds,α(z) =
F(k)seeds,α(z)
τS,α
, F(k)out,α(t, z) =
F(k)α (t, z)
τS,α
, k ∈ p, c , α ∈ A,B . (6.5)
The term corresponding to the dissolution unit in Eq. (6.1) is expressed as
F(k)pipe,α(t, z) =
ψ(z)F(k)α (t, z)
τpipe,α
, (6.6)
where ψ(z) represents the classification (death) function. The i-th moment of the CSD is
defined as
µ(k)i,α(t) =
∞∫
0
ziF (k)α (t, z) dz , i = 0, 1, 2, · · · , N, k ∈ p, c , α ∈ A,B. (6.7)
134
According to Figure 6.1, the corresponding mass balances for the liquid phase are given as
dm(k)L,α(t)
dt= m
(k)L,in,α(t) − m
(k)L,out,α(t) − 3ρS,α kv
∞∫
0
z2G(k)α (t, z)F (k)
α (t, z) dz , (6.8)
with initial data
m(k)L,α(t = 0) = m
(k)L,0,α = w(k)
α (0)ρL,0,αVL,α , k ∈ p, c , α ∈ A,B , (6.9)
where ρL,0,α = ρL,α(t = 0) is the density of fresh (initially) supersaturated solution.
Each crystallizer has six liquid streams. The incoming liquid streams include the continues
feeding of fresh solution, the solution coming from exchange pipe, and the solution coming
from the dissolution pipe. The outer streams include the outgoing liquid stream of the
utilized solution, and the solutions going to exchange and dissolution pipes. The inner
fluxes to both tanks are defined as
m(k)L,in,A(t) = m
(k)L,in,c,A(t) + m
(k)L,out,pipe,A(t) + m
(k)L,in,exch,B(t) , (6.10)
m(k)L,in,B(t) = m
(k)L,in,c,B(t) + m
(k)L,out,pipe,B(t) + m
(k)L,in,exch,A(t) , (6.11)
where the incoming flux of fresh solution to each crystallizer is given by
mL,in,c,α(t) = w(k)α (0)ρL,0,αVL,α , (6.12)
and the mass fluxes from the dissolution pipes and exchange pipes are given as
m(k)L,out,pipe,α(t) = w(k)
α (t)ρL,α(t)Vpipe,α , (6.13)
m(k)L,in,exch,B(t) = w
(k)A (t)ρL,A(t)Vexch , m
(k)L,in,exch,A(t) = w
(k)B (t)ρL,B(t)Vexch . (6.14)
Here, w(k)α denotes the mass fraction of k-th enantiomer in tank α and is defined as
w(k)α (t) =
m(k)α (t)
m(p)α (t) + m
(c)α (t) + mwater
, (6.15)
where m(k)α , m
(p)α , m
(c)α denote the mass of k-th, preferred and counter enantiomers, respec-
tively. Moreover, mwater denotes the mass of water, VL,α is the flow-rate to tank α, Vpipe,α
135
is flow-rate to the dissolution pipe of tank α, Vexch is flow-rate to the exchange pipes of
both tanks, and ρL,α represents the solution density at any time t which is defined as
ρL,α(t) = 1000(ρwater + K3wtotα (t) . (6.16)
Here
wtotα (t) = w(p)
α (t) + w(c)α (t) , (6.17)
and ρwater is defined by
ρwater =1
K1 + K2T 2. (6.18)
The symbols K1,K2 and K3 are the density parameter given in Table 6.1. The outer mass
flux from tank α is defined as
m(k)L,out,α(t) = m
(k)L,out,c,α(t) + m
(k)L,in,pipe,α(t) + m
(k)L,out,exch,α(t) . (6.19)
Here, the outer flux to the dissolution pipe is formulated as
m(k)L,in,pipe,α(t) = m
(k)L,out,pipe,α(t − τpipe,α) +
kvρcVα
Vpipe,α
∞∫
0
z3ψ(z)F (k)α (t − τpipe,α, z) dz , (6.20)
and the outer fluxes of the utilized solution from the tank α and to the exchange pipe are
given as
m(k)L,out,c,α(t) = w(k)
α (t)ρL,α(t)Vα , mL,out,α,exch(t) = w(k)α (t)ρL,α(t)Vexch . (6.21)
Because of racemic solution, the mass of both enantiomers in the feed stream of each vessel
are the same. The size-dependent growth rates are defined as
G(k)α (t, z) = kg(S
(k)α (t) − 1)g(1 + aASLz)dASL , k ∈ p, c , α ∈ A,B . (6.22)
The exponent in above equation denotes the growth order and the constants aASL and
dASL represent the size dependency. The growth rate constant, kg, is given by
kg = kg,0 exp
(
−EA,g
RT
)
. (6.23)
136
Here kg,0 denote the pre-exponent growth rate constant, EA,g represent the activation
energy, and R is the universal gas constant. There values are given in Table 6.1 which
are assumed to be the same for both vessels. Here, S(k)α (t) is the super saturation of k-th
enantiomer
S(k)α (t) =
w(k)α (t)
w(k)sat,α(t)
, (6.24)
where w(k)sat,α is the saturated mass fraction which under isothermal condition is defined as
w(p)sat,α = asat + bsatw
(c)α , w
(c)sat,α(c) = asat + bsatw
(p)α . (6.25)
Here, asat and bsat are the constant of the solubility given in Table 6.1. The nucleation rate
is defined as the sum of primary (heterogeneous) and secondary nucleation rates ([18, 52])
B(k)0,α(t) = B
(k)0,prim,α(t) + B
(k)0,sec,α(t) , k ∈ p, c , α ∈ A,B . (6.26)
The primary nucleation rate is given by a semi-empirical equation derived from Mersmann
model ([18, 52])
B(k)0,prim,α(t) = η(k)
α (t) exp
−aprim ln(
ρS/C(k)eq,α
)3
(ln S(k)α (t))2
, (6.27)
where
η(k)α (t) = kb,primTe−
KTT exp
(
−∑
k w(k)α (t)
KW
)
√
√
√
√ln
(
ρS
C(k)eq,α(t)
)
(
S(k)α (t)C(k)
eq,α(t))
73 . (6.28)
Here, C(k)eq,α denotes the concentrations of the k-th enantiomer at equilibrium as defined
below
C(k)eq,α(t) = ρL,α(t)w
(k)sat,α(t) . (6.29)
The secondary nucleation rate is given by an overall power law expression
B(k)0,sec,α(t) = kb,sec
(
S(k)α (t) − 1
)bsec(
µ(k)3,α(t)
)nsec
, (6.30)
137
where bsec is the secondary nucleation rate exponent and nsec is the third moment exponent.
The secondary nucleation rate constant is given as
kb,sec = kb,sec,0 exp
(
−EA,b
RT
)
. (6.31)
The symbol kb,sec,0 is the pre-exponential factor of the secondary nucleation rate constant
and EA,b is the corresponding activation energy.
Although the steady state results are crucial for an assessment of the process, a better
experimental realization can be achieved by investigating the process dynamically.
6.2 Implementation of Finite Volume Scheme
In this section, the high resolution finite volume scheme of [32] is implemented for discretiz-
ing the derivative of length coordinate in the population balance model. For the derivation
and explanation of the scheme, it is convenient to re-write Eq. (6.1) in the following form
∂F(k)α (t, z)
∂t= − ∂[G
(k)α (t, z)F
(k)α (t, z)]
∂z+ Q(k)
α (t, z) , (6.32)
where
Q(k)α (t, z) = F
(k)seeds,α(z) − F
(k)out,α(t, z) − F
(k)pipe,α(t, z) . (6.33)
Before implementing the proposed numerical scheme, the first step is to discretize the
computational domain which is the crystal size z in the present study. Let N be a large
integer, and denote by (zi− 12), i = 1, 2, · · · , N + 1, the partitions of cells in the domain
[zmin, zmax], where zmin is the minimum and zmax is the maximum crystal length of interest.
For each i = 1, 2, · · · , N , ∆z represents the cell width, the points zi refer to the cell centers,
and the points zi± 12
denotes the cell boundaries. The integration of Eq. (6.32) over the cell
Ωi =[
zj− 12, zj+ 1
2
]
yields the following cell centered semi-discrete finite volume schemes for
138
f(k)
i± 12,α
= (G(k)α F
(k)α )i± 1
2
∫
Ωi
∂F(k)α (t, z)
∂tdz = −
(
f(k)
i+ 12,α
(t) − f(k)
i− 12,α
(t))
+
∫
Ωi
Q(k)α (t, z)dz. (6.34)
Let F(k)i,α and Q
(k)i,α denote the average values of the number density and source term in each
cell Ωi, i.e.
F(k)i,α (t) =
1
∆z
∫
Ωi
F (k)α (t, z)dz , Q
(k)i,α(t) =
1
∆z
∫
Ωi
Q(k)α (t, z)dz . (6.35)
Then, Eq. (6.34) for i = 1, 2, · · · , N can be written as
dF(k)i.α (t)
dt= −
f(k)
i+ 12,α
(t) − f(k)
i− 12,α
(t)
∆z+ Q
(k)i,α(t) , k ∈ p, c, α ∈ A,B . (6.36)
Here, N denotes the total number of mesh elements in the computational domain. The
accuracy of finite volume discretization is mainly determined by the way in which the cell
face fluxes are computed. Assuming that the flow is in positive z-direction, a first order
accurate upwind scheme is obtained by taking the backward differences
f(k)
i+ 12,α
(t) = (G(k)α (t)F (k)
α (t))i , f(k)
i− 12,α
(t) = (G(k)α (t)F (k)
α (t))i−1 . (6.37)
To get high order accuracy of the scheme, one has to use better approximation of the cell
interface fluxes. According to the high resolution finite volume scheme of [32] the flux at
the right boundary zi+ 12
is approximated as
f(k)
i+ 12,α
(t) = f(k)i,α (t) +
1
2φ(r
(k)
i+ 12,α
(t))(f(k)i,α (t) − f
(k)i−1,α(t)) . (6.38)
Similarly, one can approximate the flux at left cell boundary. The flux limiting function φ
according to [32] is defined as
φ(
r(k)
i+ 12,α
(t))
= max
0, min (2r(k)
i+ 12,α
(t), min (1
3+
2r(k)
i+ 12,α
(t)
3, 2))
. (6.39)
139
Here, r(k)
i+ 12,α
(t) is the so-called upwind ratio of the consecutive flux gradients
r(k)
i+ 12,α
(t) =f
(k)i+1,α(t) − f
(k)i,α (t) + ε
f(k)i,α (t) − f
(k)i−1,α(t) + ε
, (6.40)
where ε is a small number to avoid division by zero. This scheme is not applicable up
to the boundary cells because it needs values of the cell nodes which are not present. To
overcome this problem, the first order approximation of the fluxes was used at the inter-
faces of the first two cells on the left-boundary and at the interfaces of the last cell on the
right-boundary. At the remaining interior cell interfaces, the high order flux approximation
of Eq. (6.38) was used. It should be noted that, the first order approximation of the fluxes
in the boundary cells does not effect the overall accuracy of the proposed high resolution
scheme.
The resulting system of ordinary differential equations (ODEs) in Eq. (6.36) together with
Eqs. (6.37)-(6.40) can be solved by a standard ODE-solver. In this study a Runge-Kutta
method of order four was used. In the case of fines dissolution with time-delay the residence
time in the pipe was taken as the integer multiple of the time step. This facilitates to keep
the old values in memory and to avoid the linear interpolation. The computer program is
written in the Matlab 7.9.1 (R2009b).
6.3 Test Problems
In this section the simulation results of two-coupled MSMPR preferential crystallizers are
presented for different operating strategies. Firstly, both crystallizers are operated without
and with fines dissolution units and are seeded continuously. Afterwards, both tanks are
seeded periodically with seeds of the preferred enantiomers. In the case of no fines disso-
lution, the last terms on the right hand side of Eq. (6.1) is neglected.
In order to analyze the quality of process some goal functions can be used, such as product
140
purity, productivity, yield and mean crystal size of the preferred enantiomers. These goal
functions give detailed information about the success and potential of continuous prefer-
ential crystallization.
Purity: It is typically of key importance in enantioselective crystallization
Pu(p)α =
m(p)S,out,α
m(p)S,out,α + m
(c)S,out,α
, where m(k)S,out,α = kvρS
∫ ∞
0
z3F (k)α (t, z)dz . (6.41)
It is a ratio of the mass flow rate of preferred enantiomer to the sum of the mass flow rates
of preferred and counter enantiomers.
Productivity: The productivity of a continuous crystallization process can be defined as
the mass flow of solid produced per unit size. It can be described by the following equation
Pr(p)α = mass flow of solid produced (preferred) per unit volume =
m(p)S,out,α − m
(p)seeds,α
VL,α
.
(6.42)
Yield: The yield can be defined as the ratio of the mass flow of solid target particles (p)
produced over mass flow of this enantiomer in the feed solution introduced
Y (p)α =
m(p)S,out,α − m
(p)seeds
m(p)L,in,c,α
, α ∈ A,B . (6.43)
Mean crystal size: Besides purity, the crystal size also plays an important role in most
industrial applications. For that reason, the mean crystal size of the preferred enantiomer
is instructive. It is defined as:
z(p)α =
µ(p)1,α
µ(p)0,α
, (6.44)
where µ(p)1,α is total length of crystals and µ
(p)0,α is total number of crystals in the correspond-
ing crystallizers (c.f. Eq. (6.7)).
141
In the following three test problems, the minimum and maximum crystal sizes of interest
are taken as zmin = 1.0×10−10 m and zmax = 1.0×10−2 m, respectively. The corresponding
computational domain is subdivided into 200 grid points. The remaining parameters are
given in Table 6.1 and are assumed to be the same for both crystallizers. Moreover, the
volumes of both tanks and liquid volumetric flow rates are taken the same. The physico-
chemical parameters correspond to the enantiomers of the amino acid threonine and the
solvent water, see [17, 18]. The temperatures of both crystallizers were assumed to be
constant at 33 oC.
The mean residence time of the liquid phase was kept fixed at τL,α = 112.41 min in both
tanks, while the residence time of the solid phase was either τS,α = τL,α or τSτL,α= 2τL,α. In
real process, it may be difficult to decouple the residence times of liquid and solid phases.
Thus, it is most likely to set τS,α = τL,α. However, a larger residence times of the solid
phase compared to the liquid phase could eventually be realized by installing a filter for
the solid particles at the outlet to allow a longer residence τS,α of the solid phase. In the
numerical simulations, the steady state was identified when the relative deviation error in
the supersaturation was below 5 × 10−7.
6.3.1 Numerical Test Problem
This case study deals with the continuous enantioselective preferential crystallization pro-
cess in two-coupled vessels under isothermal condition. Both vessels are seeded with the
seeds of corresponding preferred enantiomers. The seeds distributions are taken to be the
same in both crystallizers
F(p)seeds,α =
a3
zAa,α
exp
(
−0.5
(
1
a4
ln
(
z
a5
))2)
, (6.45)
142
where a4, b4 and c4 are constants and there values are given in the Table 6.1. The normal-
ization factor is defined as
Aa,α =kvρS
m(p)seeds,α
µ(p)3,α . (6.46)
The initial masses of the preferred and counter enantiomers in each crystallizer are taken
as m(p)0,α = m
(c)0,α = 0.0478 kg for α ∈ A,B.
Case I: Continuous seeding without fines dissolution
In this case both vessels are seeded continually and the fines dissolution units are not
connected to them. The simulation results at steady state conditions are shown in Tables
6.2 and 6.3 for both single and coupled continuous crystallization models at different mass
flows of seeds and ratios of residence times τs,α/τL,α. The results of single crystallizer are
taken from the recent article by [66].
Moreover, Figures 6.2-6.5 show the comparison of results for single and coupled crystallizers
at different mass flow-rates of seeds and different ratios of residence times. In the coupled
case, only the results of tank A are presented which are exactly the same for tank B due
to the same operating conditions. It is evident from tables and figures that an increase
in the ratio of solid to the liquid residence times, τs,α/τL,α, reduces the purity of preferred
enantiomer due to the production of increasing amount of counter enantiomer as an im-
purity, while the productivity, yield, and mean crystal size of the preferred enantiomer are
enhanced. As τs,α increases, the crystals of the preferred enantiomer have more time to
grow and, hence, the overall productivity, yield and the mean crystal size are improved.
The productivity, yield and investment of seeds increases by increasing the mass flow rate
of seeds. It is understood that large amount of seeding produces more crystals and, thus,
enhances productivity. However, the mean crystal size reduces on increasing the mass of
seeds because the existing supersaturation is consumed by large number of crystals. On the
143
other hand, by low investment of seeds, the available supersaturation is consumed by only a
fewer crystals to obtain larger mean size of the crystals. Finally, the arrival of steady-state
delays when the ratio of residence times is larger and arrives earlier on increasing the mass
flow of seeds.
A comparison of the results for single and coupled crystallizers shows that all goal functions
have improved values in the coupled case. Thus, coupled process not only improves the
product quality but also crystallizes both enantiomers simultaneously in separate crystal-
lizers. This is a big advantage of the couple process over the decoupled (single) one.
Case II: Continuous seeding with fines dissolution
In this case both vessels are connected with fines dissolution units and are seeded continu-
ously. Fines particles below the critical size zcrit = 6.0×10−4 m are removed from the vessel
along with the solution into the recycle unit (the pipe). The selection (death) function is
given as
ψ(z) =
0.6 if z ≤ zcrit ,0 otherwise .
(6.47)
It is assumed that all particles in the dissolution unit are dissolved before re-entering the
solution to the vessel. Tables 6.4 and 6.5 show the results of single and coupled crystalliz-
ers with and without fines dissolution. Moreover, Figures 6.7-6.10 show the comparison of
results for single and coupled crystallizers at different mass flow rates of seeds and different
ratios of residence times. It can be seen that fines dissolution has improved the mean crys-
tal size of the crystals. However, effects on other goal functions are negligible. The results
of the figures reflect the same behavior observed in the tables. Moreover, the advantages of
coupled process are clearly visible these figures and tables. The results show that all goal
function have improved values in the coupled case. Thus, coupled process has enhanced
the process potential and product quality.
144
Case III: Periodic seeding without fines dissolution
In this case continuous crystallization process along with the periodic seeding is investi-
gated for both single and coupled crystallizers. The model equations are exactly the same
to Case I for continuous seeding. The numerical results at steady-state level are given in
Tables 6.6 and 6.7 for single and coupled crystallizers at different ratios of residence times.
In these tables ton represents the time at which seeding is switched on and toff denotes the
period of time at which seeding is switched off. The seed flow rate varies compared to the
continuous seeding.
A schematic diagram of the periodic seeding is shown in Figure 6.6. In that figure, the
shaded regions represent the seeding times (ton) and the non-shaded regions depict the pe-
riods when seeding is switched off (toff ). In both tables it can be observed that an increase
in τS reduces the purity of the preferred enantiomer. As residence time increases, more
nuclei of the counter enantiomer are produced as an impurity of the product and, thus, the
purity is lower. It can also be observed that the purity has decreasing behavior for increas-
ing toff . On the other hand, as τS increases the crystals of the preferred enantiomer get
more time to grow and, thus, the overall productivity and yield increases. Additionally, the
productivity and yield are decreased by increasing toff . It is obvious that when seeding is
switched off for a longer time, the mean crystal size will increase by increasing the residence
time τS due to longer stay of seeds in the crystallizer and because of utilizing the existing
supersaturation by fewer crystals. The mean crystal size also increases by increasing toff
for all considered residence times due to the utilization of existing supersaturation by fewer
crystals. It is also evident from the tables that the seeds flow rate decreases by increasing
the residence time.
145
A comparison of continuous seeding case with the current periodic case shows that con-
tinuous seeding gives better purity, productivity and yield, while the mean crystal size is
smaller and investment (mass flow of seeds) remains the same in all considered residence
times. On the other hand, the periodic seeding gives larger crystals at low investment (feed
flow rate) of seeds per residence time. The arrival of steady-state time delays on increasing
τS, while keeping the liquid residence time τL fixed. This behavior is exactly similar to the
continuous seeding case. Moreover, a comparison of the results in Tables 6.6 and 6.7 shows
that all goal functions are improved in the coupled case.
Figures 6.11-6.12 show the goal functions at different periods of seeding. It can be observed
that supersaturation decreases with reducing time period of seeding, because a decrease in
toff of seeds speedup the consumption of solute mass. All figures show the same trends.
Oscillations are also visible which indicate the conditions of periodic seeding. The purity
plots show that it decreases by increasing toff . Initially, a decrease in purity is due to
increasing amount of counter enantiomer but after a certain period of time the amount of
preferred enantiomer starts increasing which improves the purity. It can also be observed
that an increase in toff produces large size crystals which is also clear from the Tables.
The behavior of productivity is also similar to the corresponding Tables. It is evident that
larger toff gives lower yield and productivity. Moreover, a temporary drop in the purity,
before reaching the steady state, is due to the stronger relative impact (negative) of pri-
mary nucleation of counter enantiomer. This effect reduces gradually when steady state
condition is approaching.
In summary, the periodic seeding diminishes the productivity and yield, while improves
the mean crystal size. The purity is slightly reduced with periodic seeding (approx. 0.1%).
146
Moreover, the investment of seed crystals is reduced. In overall, the periodic seeding
has improved the process and productivity, if one carefully compare the investment and
productivity. In other words, more product is achieved at low investment. Once again,
the coupled process has improved the values of goal functions and, thus, justify the use of
coupled process.
147
Table 6.1: Physicochemical parameters of threonine-water system ([17]).
Parameters Symbols Value UnitVolume shape factor kv 0.122 [−]Density of solid phase ρS 1250 [kg/m3]Crystal growth rate exponent kg,0 2.98 × 109 [m/min]Nucleation rate exponent kb,sec,0 2.38 × 1026 [m−3/min]Volume of the liquid phase VL,α 4.496 × 10−4 [m3]
Volumetric flow rate VL,α 0.4 × 10−5 [m3/min]Volume of the pipe Vpipe,α 4.0 × 10−5 [m3]
Volumetric exchange flow rate Vexch 0.4 × 10−5 [m3/min]Mass of water mwater 0.3843 [kg]Constant for density K1 1.00023 [cm3g−1]Constant for density K2 4.68 × 10−6 [cm3g−1K−2]Constant for density K3 0.3652 [cm3g−1]Density of water ρwater 0.9947 [g cm−3]Density of fresh solution ρL,0,α 1.067 × 103 [kg m−3]Parameter for temperature dependence KT 1874.4 [K]Parameter for mass fraction dependence KW 0.290 [−]Universal gas constant R 8.314 J mol/KGrowth rate exponent g 1.1919 [−]Parameter for crystal growth aASL 2.0209 × 104 [m−1]Parameter for crystal growth dASL −4.066 × 10−1 [−]Liquid mean residence time τL,α 112.4116 [min]Activation energy for crystal growth EA,g 75540 J
mol
Activation energy for nucleation EA,b 63830 Jmol
Secondary nucleation exponent bsec 4.80 [−]Exponent for third moment nsec 3.0258 [−]
Nucleation constant kb,prim 3.847 × 10−2 [ (kg m−3)−7/3
min K]
Constant for exponential law aprim 4.304 × 10−3 [−]Seeds distribution constant a1 0.014 [−]Seeds distribution constant a2 0.0009 [m]Seeds distribution constant a3 0.288 [−]
148
Table 6.2: Case I: Goal functions of single crystallizer without FD, τL = 112.41 min.τSτL
tstd [min] m(p)seed [ g
min] Pu(p) [%] Pr(p) [ kg
min m3 ] Y (p) [%] z(p) [mm]
1.0 646 0.0089 99.74 0.037 3.91 1.312.0 1411 0.0044 98.95 0.043 4.54 1.541.0 676 0.0356 99.92 0.060 6.29 1.132.0 1315 0.0178 99.54 0.062 6.60 1.271.0 721 0.0623 99.94 0.067 7.08 1.082.0 1318 0.0311 99.68 0.069 7.29 1.18
Table 6.3: Case I: Goal functions of coupled crystallizers without FD, τL = 112.41 min.τS,A
τL,Atstd [min] m
(p)seed,A [ g
min] Pu
(p)A [%] Pr
(p)A [ kg
min m3 ] Y(p)A [%] z
(p)A [mm]
1.0 642 0.0089 99.99 0.045 4.79 1.352.0 1410 0.0044 99.99 0.054 5.71 1.621.0 676 0.0356 99.99 0.079 8.32 1.182.0 1313 0.0178 99.99 0.084 8.90 1.331.0 720 0.0623 99.99 0.091 9.62 1.112.0 1316 0.0311 99.99 0.095 10.02 1.24
Table 6.4: Case II: Results of single crystallizer with FD: FD (yes), without FD (no).τSτL
FD tstd [min] m(p)seed [ g
min] Pu(p) [%] Pr(p) [ kg
min m3 ] Y (p) [%] z(p) [mm]
1.0 no 676 0.0356 99.92 0.060 6.29 1.131.0 yes 648 0.0356 99.99 0.060 6.25 1.152.0 no 1315 0.0178 99.54 0.062 6.60 1.272.0 yes 1303 0.0178 99.97 0.062 6.56 1.29
Table 6.5: Case II: Results of coupled case with FD: FD (yes) and without FD (no).τS,A
τL,AFD tstd [min] m
(p)seed,A [ g
min] Pu
(p)A [%] Pr
(p)A [ kg
min m3 ] Y(p)A [%] z
(p)A [mm]
1.0 no 676 0.0356 99.99 0.079 8.32 1.181.0 yes 650 0.0356 99.99 0.078 8.27 1.202.0 no 1314 0.0178 99.99 0.084 8.90 1.332.0 yes 1301 0.0178 99.99 0.084 8.85 1.36
149
Tab
le6.6:
Case
III:Perio
dic
seedin
gin
single
crystallizer:
m(p
)seed
s=
4g,τL
=112.4
min
.τS
τL
seedin
gty
pe
tstd
[min
]ton
[min
]toff
[min
]m
(p)
seed[
gm
in]
Pu
(p)[%
]P
r(p
)[kg
min
m3 ]
Y(p
)[%]
z(p
)[mm
]
1.0P
67010
50.0244
99.870.025
2.591.18
1.0P
67010
100.0189
99.840.0071
0.721.23
1.0P
67020
100.0252
99.870.025
2.611.19
1.0C
676-
-0.0356
99.920.060
6.291.13
2.0P
132010
50.0125
99.410.04
4.21.34
2.0P
132110
100.0089
99.320.028
2.981.40
2.0P
132120
100.0102
99.40.0372
3.931.34
2.0C
1315-
-0.0178
99.540.062
6.601.27
Tab
le6.7:
Case
III:Perio
dic
seedin
gin
coupled
case:m
(p)
seeds=
4g,τL
=112.4
min
.τS
,A
τL
,Aseed
ing
type
tstd
[min
]ton
[min
]toff
[min
]m
(p)
seed,A
[g
min
]P
u(p
)A
[%]
Pr(p
)A
[kg
min
m3 ]
Y(p
)A
[%]
z(p
)A
[mm
]
1.0P
67210
50.0244
99.990.041
4.291.24
1.0P
67210
100.0189
99.990.021
2.1671.27
1.0P
67220
100.0252
99.990.04
4.261.24
1.0C
678-
-0.0356
99.990.079
8.321.18
2.0P
131810
50.0125
99.990.062
6.551.42
2.0P
131810
100.0089
99.990.049
5.191.49
2.0P
131920
100.0102
99.990.062
6.531.43
2.0C
1312-
-0.0178
99.990.084
8.901.33
150
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 300098.8
99
99.2
99.4
99.6
99.8
time [min]
purit
y [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 300098.8
99
99.2
99.4
99.6
99.8
100
time [min]
purit
y [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30000
1
2
3
4
5
6
7
time [min]
yiel
d [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 3000−15
−10
−5
0
5
10
time [min]
yiel
d [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
Figure 6.2: Case I: LS: Single crystallizer results. RS: Coupled crystallizer results.
151
0 500 1000 1500 2000 2500 30000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
time[min]
prod
uctiv
ity [k
g/m
in m
3 ]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 3000−0.15
−0.1
−0.05
0
0.05
0.1
time [min]
prod
uctiv
ity [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30000.9
0.95
1
1.05
1.1
1.15
1.2
1.25
1.3
1.35
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30000.9
1
1.1
1.2
1.3
1.4
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
Figure 6.3: Case I: LS: Single crystallizer results. RS: Coupled crystallizer results.
152
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 300098.8
99
99.2
99.4
99.6
99.8
100
time [min]
purit
y [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 300098.8
99
99.2
99.4
99.6
99.8
100
time [min]
purit
y [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30000
1
2
3
4
5
6
7
time [min]
yiel
d [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 3000−10
−5
0
5
10
time [min]
yiel
d [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
Figure 6.4: Case I: LS: Single crystallizer results. RS: Coupled crystallizer results.
153
0 500 1000 1500 2000 2500 30000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
time [min]
prod
uctiv
ity [k
g/m
in m
3 ]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 3000−0.08
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
0.08
0.1
time [min]
prod
uctiv
ity [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30000.9
1
1.1
1.2
1.3
1.4
1.5
1.6
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30000.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
Figure 6.5: Case I: LS: Single crystallizer results. RS: Coupled crystallizer results.
seeding off
seeding on
time [min]
mas
s flo
w o
f see
ds [g
/min
]
Figure 6.6: Schematic diagram for periodic seeding.
154
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 300099.98
99.985
99.99
99.995
100
time [min]
purit
y [%
] m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 300099.98
99.985
99.99
99.995
100
time [min]
purit
y [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 3000−15
−10
−5
0
5
10
time [min]
yiel
d [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 3000−15
−10
−5
0
5
10
time [min]
yiel
d [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
Figure 6.7: Case II: LS: Single crystallizer results. RS: Coupled crystallizer results.
155
0 1000 2000 3000−0.15
−0.1
−0.05
0
0.05
0.1
time [min]
prod
uctiv
ity [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 3000−0.15
−0.1
−0.05
0
0.05
0.1
time [min]
prod
uctiv
ity [%
]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30000.9
1
1.1
1.2
1.3
1.4
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
0 500 1000 1500 2000 2500 30000.9
1
1.1
1.2
1.3
1.4
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0089 g/min
m(p)seeds = 0.0356 g/min
m(p)seeds = 0.0623 g/min
Figure 6.8: Case II: LS: Single crystallizer results. RS: Coupled crystallizer results.
156
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30001.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 300099.93
99.94
99.95
99.96
99.97
99.98
99.99
100
time [min]
purit
y [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 300099.93
99.94
99.95
99.96
99.97
99.98
99.99
100
time [min]
purit
y [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 3000−10
−5
0
5
10
time [min]
yiel
d [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 3000−10
−5
0
5
10
time [min]
yiel
d [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
Figure 6.9: Case II: LS: Single crystallizer results. RS: Coupled crystallizer results.
157
0 500 1000 1500 2000 2500 3000−0.1
−0.05
0
0.05
0.1
time [min]
prod
uctiv
ity [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 3000−0.1
−0.05
0
0.05
0.1
time [min]
prod
uctiv
ity [%
]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30000.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
time t [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
0 500 1000 1500 2000 2500 30000.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
time [min]
mea
n cr
ysta
l siz
e [m
m]
m(p)seeds = 0.0044 g/min
m(p)seeds = 0.0178 g/min
m(p)seeds = 0.0311 g/min
Figure 6.10: Case II: LS: Single crystallizer results. RS: Coupled crystallizer results.
158
0 500 1000 1500 2000 2500 30001.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
ton
=10, toff
=5
ton
=10, toff
=10
0 500 1000 1500 2000 2500 30001.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
time [min]
supe
rsat
urat
ion
[−]
ton
=10, toff
=5
ton
=10, toff
=10
0 500 1000 1500 2000 2500 300099.8
99.85
99.9
99.95
100
time [min]
purit
y [%
]
ton
=10, toff
=5
ton
=10, toff
=10
0 500 1000 1500 2000 2500 300099.8
99.85
99.9
99.95
100
time [min]
purit
y [%
]
ton
=10, toff
=5
ton
=10, toff
=10
0 500 1000 1500 2000 2500 3000−10
−5
0
5
time [min]
yiel
d [%
]
ton
=10, toff
=5
ton
=10, toff
=10
0 500 1000 1500 2000 2500 3000−10
−5
0
5
time [min]
yiel
d [%
]
ton
=10, toff
=5
ton
=10, toff
=10
Figure 6.11: Case III: LS: Single crystallizer results. RS: Coupled crystallizer results.
159
0 500 1000 1500 2000 2500 3000−0.08
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
time [min]
prod
uctiv
ity [%
]
ton
=10, toff
=5
ton
=10, toff
=10
0 500 1000 1500 2000 2500 3000−0.08
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
time [min]
prod
uctiv
ity [%
]
ton
=10, toff
=5
ton
=10, toff
=10
0 500 1000 1500 2000 2500 30000.9
0.95
1
1.05
1.1
1.15
1.2
1.25
1.3
time [min]
mea
n cr
ysta
l siz
e [m
m]
ton
=10, toff
=5
ton
=10, toff
=10
0 500 1000 1500 2000 2500 30000.9
0.95
1
1.05
1.1
1.15
1.2
1.25
1.3
time [min]
mea
n cr
ysta
l siz
e [m
m]
ton
=10, toff
=5
ton
=10, toff
=10
Figure 6.12: Case III: LS: Single crystallizer results. RS: Coupled crystallizer results.
160
161
Chapter 7
Conclusion and Future Recommendations
7.1 Conclusion
In this thesis, batch and continuous crystallization models were developed and numerically
investigated. Especially, models were derived for simulating batch and continuous enan-
tioselective preferential crystallization processes in single and coupled crystallizers. The
effects of nucleation, growth, fines dissolution, seeding strategies and residence time char-
acteristics on the CSD were analyzed.
A TVB Discontinuous Galerkin (DG) finite element method was proposed for solving batch
crystallization models. The method is explicit in time and, hence, can be coupled with
a high-order TVB Runge-Kutta time discretization method. The local projection limiter
avoids spurious oscillations in the vicinity of a discontinuity that does not affect the accu-
racy of the scheme in smooth regions. Thus, the limiter preserves the non-negativity of the
CSD and assures the TVB property of the scheme. In contrast to the finite difference and
finite volume methods, the high order accuracy of the scheme can be achieved by using
more information within a cell rather than widening the mesh stencil. For that reason,
the method can be used to boundary cells without loosing accuracy and the application of
the boundary conditions are easier. On the other hand, the high resolution finite volume
schemes (HR-FVS) are not applicable up to the boundary cells and, hence, the order of
the scheme has to be lowered in the boundary cells. This produces significant errors in the
part of distribution coming from nucleation as illustrated in the numerical test problems
of Chapter 3. Therefore, the DG-scheme is more flexible and accurate as compared to the
HR-FVS. The obtained results support the process design and optimization.
Mathematical models were derived for batch preferential single and coupled crystallizers
connected with pipes (fines dissolution units). The models were further elaborated by
162
considering the isothermal and non-isothermal conditions. Firstly, the crystallization of
preferred enantiomer was assumed to take place in a single crystallizer equipped with a
fines dissolution unit. In this setup, the extracted solution was sieved by filters and was
supposed to be free of bigger crystals. Therefore, only small particles were withdrawn
to the fines dissolution loop. In order to assure a crystal-free liquid exchange, the with-
drawn liquid in the fines dissolution loop was heated for dissolving small particles. Before
re-entering into the crystallizer, this liquid was cooled again. The breakage and agglom-
eration processes were not considered in this study. Secondly, the model was extended
for a coupled batch preferential crystallization process with isothermal and non-isothermal
conditions. In this setup, the crystallization of two enantiomers was assumed to take place
in two separate crystallizers which were coupled through exchange pipes. There were two
main advantages of considering two coupled crystallizers. The first one was that both
enantiomers were crystallized out at the same time in separate crystallizers. Secondly,
because of liquid exchange between the crystallizers, the growth process was enhanced in
both crystallizers and, thus, crystals of large mean sizes were obtained. Both HR-FVS
and DG-method were used to solve these models. The DG-method was applied for the
first time to model such processes. The results of both schemes were found comparable.
However, the DG-scheme produces better approximations in the part of the distribution
coming from nucleation. Moreover, the results are in good agreement with the experimental
results of our collaborating colleagues from the Max Planck Institute Magdeburg, Germany.
For the first time continuous preferential crystallization was investigated and the effects of
different seeding strategies and residence time characteristics were analyzed on the dynam-
ics of a Mixed Suspension Mixed Product Removal (MSMPR) crystallizer equipped with
a fines dissolution unit. The fines dissolution was included as recycle streams around the
MSMPR crystallizer. Moreover, primary heterogeneous and secondary nucleation mech-
163
anisms along with size-dependent growth rates were taken into account. The model was
then extended for a coupled continuous preferential crystallization process. In this setup,
the crystallization of two enantiomers was assumed to take place in two separate crystal-
lizers which were coupled through exchange pipes. To judge the quality of the process
some goal functions were used, such as purity, productivity, yield and mean crystal size of
the preferred enantiomer. The HR-FVS and DG-method were applied to solve the models.
The results of the DG-method for the single continuous preferential crystallization model
were compared with the results of HR-FVS. The supersaturation and mean crystal size
were reduced on increasing the mass flow of seeds, while productivity, yield and purity
were improved. The errors in mass balances were found below 0.15% for a wide range
of mass flow of seeds. The periodic seeding slightly diminished the purity, productivity
and yield but improved the mean crystal size and the investment (mass flow) of seeds was
also smaller. The fines dissolution improved the mean crystal size and purity but slightly
reduced the productivity and yield. These results could be used to find the optimum oper-
ating conditions for improving the product quality and for reducing the operational cost of
continuous preferential crystallization. For the case study carried out in a single crystallizer
(resolution of an amino acid enantiomers) a production of 100 kg/(day ·m3) appears to be
feasible.
7.2 Future Recommendations
In this dissertation, a DG-method was used to solve the models of different crystallization
processes. However, the accuracy of the DG-scheme can be further improved by using basis
functions of higher order and by using better slope limiters, for example WENO limiters
[67]. In future, we also intend to implement adaptive mesh refinement techniques in the
current DG-scheme. This technique is particularly effective when the PBEs are coupled
with computational fluid dynamic (CFD) codes for non-ideally mixed systems. Moreover,
164
many crystallization processes require more than one property variable to characterize the
crystals. Besides that, the multivariate population balance model (PBM) has various ap-
plications in many scientific, medical, and industrial research disciplines. Thus, it will be
interesting to extend the current DG-scheme for solving bivariate and multivariate PBMs.
The current models can be further extended by including the attrition, aggregation and
breakage phenomena during the crystallization process. In such cases, one has to solve the
resultant integro-differential equations by using HR-FVS or DG-scheme. These schemes
are expensive for such simultaneous processes, especially on regular grids. However, the
use of the adaptive mesh refinement technique and the combination of the current schemes
with other schemes can considerably reduce the overall computational cost for the same
desired accuracy.
In the current study of preferential crystallization, a model system of amino acid-H2O
was considered for our computational study. In this model no significant interdependence
between the preferred and counter enantiomers on crystallization kinetics has been ob-
served. The same was also experimentally observed by the process engineering group of
Max Planck Institute Magdeburg, Germany. However, in their experiments for the man-
delic acid-H2O model system, they observed a strong influence of the counter-enantiomer
on the growth rate of the preferred-enantiomer, see [43] for further details. Moreover, it
can be shown that in the latter case the counter-enantiomer may even change growth rates
of particular crystal faces which results in different crystal shapes, see Figure 7.1. Hence,
a detailed study on the influence of the counter-enantiomer on the growth rate and mor-
phology of the seeded enantiomer is required. To model such processes and to study the
different evolution of crystal faces a multidimensional population balance model has to be
derived along with correct growth and nucleation rate kinetics whose results matches with
165
the experimental results. This in turn also leads inevitably to a higher requirement of the
efficient numerical discretization techniques. Currently, work is in progress in this direction.
Furthermore, this work can be extended to the study and design of control strategies in
crystallization processes. In this direction, it will be more interesting to concentrate on the
less restrictive models of crystallization processes and to derive control strategies for both
batch and continuous crystallizers.
Also, this work can be extended to the study when temperature varies in the vessels, then
the solutions may well be inhomogeneous and will require an energy equation determining
the temperature and crystallization rates for the p and c crystals.
Finally, the current schemes can be extended to other particulate processes, such as pre-
cipitation, polymerization, food processes, pollutant formation in flames, particle size dis-
tribution of crushed material and rain drops, and growth of microbial and cell populations.
166
Figure 7.1: Left: without counter enantiomer; Right: with counter enantiomer.
167
168
Chapter 8
References
Bibliography
[1] Aamir, E., Nagy, Z. K., Rielly, C. D., (2010). Optimal seed recipe design for crystal
size distribution control for batch cooling crystallisation processes. Chem. Eng. Sci. 65,
3602-3614.
[2] Abegg, C. F., Stevens, J. D., Larson, M.A., (1968). Crystal size distributions in con-
tinuous crystallizers when growth rate is size dependent. AIChE J. 14, 118-122.
[3] Alvarez-Rodrigo, A., Lorenz, H., Seidel-Morgenstern, A., (2004). On-line monitoring of
preferential crystallization of enantiomers. Chirality 16, 499-508.
[4] Arnold, D. N., Brezzi, F., Cockburn, B., Marini, L. D., (2002). Unified analysis of
discontinuous Galerkin methods for elliptic problems. SIAM J. Numer. Anal. 39, 1749-
1779.
[5] Chang, S. C., (1995). The method of space time conservation element and solution
element -A new approach for solving the Navier Stokes and Euler equations. J. Comput.
Phys. 119, 295-324.
[6] Chang, S. C., Wang, X. Y., Chow, C. Y., (1999). The space-time conservation element
and solution element method: A new high resolution and genuinely multidimensional
paradigm for solving conservation laws. J. Comput. Phys. 156, 89-136.
[7] Chung, S. H., Ma, D. L., Braatz, R.D., (1999). Optimal seeding in batch crystallisation.
Can. J. Chem. Eng. 77, 590-595.
169
[8] Cockburn, B., Shu, C. -W., (1989). TVB Runge-Kutta local projection discontinuous
Galerkin finite element method for conservation laws II. Gerneral framework. Mathe-
matics of Computations. 52, 411-435.
[9] Cockburn, B., Lin, S. -Y., (1989). TVB Runge-Kutta local projection discontinuous
Galerkin finite element method for conservation laws III. One-dimensional systems. J.
Comput. Phy. 84, 90-113.
[10] Cockburn, B., Shu., C. -W., (2001). Runge Kutta discontinuous Galerkin methods for
convection-dominated problems. J. of Sci. Comput. 16, 173-261.
[11] Collet, A., (1999). Separation and purification of enantiomers by crystallization. Enan-
tiomer 4, 157-172.
[12] Coquerel, G., (2007). Preferential crystallization. In: Novel optical resolution tech-
nologies. K. Sakai, N. Hirayama, R. Tamura (Eds.). Topics in Current Chemistry 269,
1-51.
[13] Czapla, F., Haida, H., Elsner, M. P., Lorenz, H., Seidel-Morgenstern, A., (2009).
Parameterization of population balance models for polythermal auto seeded preferential
crystallization of enantiomers. Chem. Eng. Sci. 64, 753-763.
[14] David, R., Villermaux, J., Marchal, P., Klein, J. P., (1991). Crystallization and pre-
cipitation engineering-IV. Kinetic model to adipic acid crystallization. Chem. Eng. Sci.
46, 1129-1136.
[15] Diemer, R. B., Olson, J. H., (2002). A moment methodology for coagulation and
breakage problems: Part I - analytical solution of the steady-state population balance.
Chem. Eng. Sci. 57, 2193-2209.
170
[16] Elsner, M. P., Lorenz, H., and Seidel-Morgenstern, A., (2003). Preferential crystal-
lization for enantioseparation - New experimental insights indispensable for a theoreti-
cal approach and an industrial application. 10th International Workshop on Industrial
Crystalliza- tion BIWIC 2003, Mainz Verlag ISBN-10:3861301989, 18-25.
[17] Elsner, M. P., Fernt’andez Ment’endez, D., Alonso Muslera, E., and Seidel-
Morgenstern, A., (2005). Experimental study and simplified mathematical description
of preferential crystallization, Chirality 17, 183-195.
[18] Elsner, M. P., Ziomek, G., Seidel- Morgenstern, A., (2011). Simultaneous preferential
crystallization in a coupled batch operation mode. Part II: Experimental study and
model refinement. Chem. Eng. Sci. 66, 1269-1284.
[19] Elsner, M.P., Ziomek, G., Seidel-Morgenstern, A. Efficient Separation of Enantiomers
by Preferential Crystallization in Two Coupled Vessels, DOI 10.1002/aic. 11719.
[20] Fan, R., Marchisio, D. L., Fox, R. O., (2004). Application of the direct quadrature
method of moments to polydisperse gas-soliduidized beds. Powder Technol. 139, 7-20.
[21] Gelbard, F., Tambour, Y., and Seinfeld, J. H., (1980). Sectional representations for
simulating aerosol dynamics. J. of Colloid and Interface Sci. 76, 541-556.
[22] Gordon, R. G., (1968). Error bounds in equilibrium statistical mechanics. J. Math.
Phys. 9, 655-663.
[23] Gowda, V.; Jaffr, J. A., (1993). A discontinuous finite element method for scalar
nonlinear conservation laws. Rapport de recherche INRIA 1848.
[24] Gunawan, R., Fusman, I., Braatz, R. D., (2004). High resolution algorithms for mul-
tidimensional population balance equations. AIChE J. 50, 2738-2749.
171
[25] Hermanto, M. W., Braatz, R.D., Chiu, M.-S. (2008). High-order simulation of poly-
morphic crystallization using weighted essentially nonoscillatory methods. AIChE J. 55,
122-131.
[26] Hesthaven, J. S., Warburton, T. (2008). Nodal discontinuous Galerkin methods: Algo-
rithms, analysis, and applications. Springer Texts in Applied Mathematics. 54, Springer
Verlag: New York.
[27] Hulburt, H. M., Katz, S. (1964). Some problems in particle technology. Chem. Eng.
Sci. 19, 555-574.
[28] Hundsdorfer, W. H, Verwer, J. G., (2003). Numerical solution of time-dependent ad-
vection diffusion- reaction equations, First Edition, Springer-Verlag, New York.
[29] Hounslow, M. J., Ryall, R. L., Marshall, V. R., (1988). A discretized population
balance for nucleation, growth, and aggregation. AIChE J. 34, 1821-1832.
[30] Hounslow, M. J., (1990). A discretized population balance for continuous systems at
steadystate. AIChE J. 36, 106-116.
[31] Jacques, J., Collet, A., Wilen, S.H., (1981). Enantiomers, racemates and resolutions,
Wiley, New York.
[32] Koren, B., (1993). A robust upwind discretization method for advection, diffusion and
source terms, CWI report.
[33] Korovessi, E., Linninger, A., (2006). Batch Processes.
[34] Kumar, S., Ramkrishna, D., (1996). On the solution of population balance equations
by discretization -I. A fixed pivot technique. Chem. Eng. Sci. 51, 1311-1332.
172
[35] Kumar, S., Ramkrishna, D., (1996). On the solution of population balance equations
by discretization -I. A moving pivot technique. Chem. Eng. Sci. 51, 1333-1342.
[36] Kumar, S., Ramkrishna, D., (1997). On the solution of population balance equations
by discretization -III. Nucleation, growth and aggregation of particles. Chem. Eng. Sci.
52, 4659-4679.
[37] Kumar, J., Peglow, M., Warnecke, G., Heinrich, S., Morl, L., (2006). A discretized
model for tracer population balance equation: Improved accuracy and convergence.
Comp. & Chem. Eng. 30, 1278-1292.
[38] Kurganov, A., Tadmor, E., (2000). New high-resolution central schemes for nonlinear
conservation laws and convection-diffusion equations. J. Comput. Phys. 160, 241-282.
[39] Lee, K., Matsoukas, T., (2000). Simultaneous coagulation and breakage using constant
-N Monte Carlo. Powder Technology 110, 82-89.
[40] LeVeque, R. J., (2002). FVM for hyperbolic problems, Cambridge Univ. Press, New
York, NY.
[41] LeVeque, R. J., (2003). Finite volume methods for hyperbolic systems, Cambridge
University Press, Cambridge.
[42] Lim, Y. I., Lann, J-M. L., Meyer, X. M., Joulia, X., Lee, G., Yoon, E. S., (2002). On
the solution of population balance equation (PBE) with accurate front tracking method
in practical crystallization processes. Chem. Eng. Sci. 57, 3715-3732.
[43] Lorenz, H., Perlberg, A., Sapoundjiev, D., Elsner, M. P., Seidel-Morgenstern, A.,
(2006). Crystallization of enantiomers, Chem. Eng. Proc. 45, 863-873.
[44] Ma, A., Tafti, D. K., Braatz, R. D., (2002). Optimal control and simulation of multi-
dimensional crystallization processes. Comp. and Chem. Eng. 26, 1103-1116.
173
[45] Ma, D.L., Tafti, D.K., Braatz, R.D., (2002). High-resolution simulation of multidi-
mensional crystal growth. Ind. Eng. Chem. Res. 41, 6217-6223.
[46] Majumder, A., Kariwala, V., Ansumali, S., Rajendran, A., (2010). Fast high-resolution
method for solving multidimensional population balances in crystallization. Ind. Eng.
Chem. Res. 349, 3862-3872.
[47] Marchal, P., David, R., Klein, J. P., Villermaux, J., (1988). Crystallization and precip-
itation engineering I: An efficient method for solving population balance in crystallization
with agglomeration. Chem. Eng. Sci. 43, 59-67.
[48] Marchisio, D.L., Pikturna,J.T., Fox, R.O., Vigil, R.D., Barresi, A.A., (2003). Quadra-
ture method of moments for population balance equations. AIChE J. 49, 1266-1276.
[49] Marchisio, D.L., Vigil, R.D., Fox, R.O., (2003). Quadrature method of moments for
aggregation-breakage processes. J. Colloid Interface Sci. 258, 322-334.
[50] McGraw, R., (1997). Description of Aerosol Dynamics by the quadrature method of
moments. Aerosol Sci. Tech. 27, 255-265.
[51] McGraw, R., Nemesure S., E. S., Stephen (1998). Properties and evolution of Aerosol
with size distributions having identical moments. Aerosol Sci. Tech. 29, 761-772.
[52] Mersmann, A. (2001). Crystallization Technology Handbook, Marcel Dekker, Inc.,
Second Edition.
[53] Miller, S. M., Rawlings, J. B., (1994). Model identification and control strategies for
batch cooling crystallizers. AIChE J. 40, 1312-1327.
[54] Motz, S., Mitrovic, A., Gilles, E.-D., Vollmer, U., Raish, J., (2003). Modeling, simula-
tion and stabilizing-control of an oscillating continuous crystallizer with fines dissolution.
Chem. Eng. Sci. 58, 3473-3488.
174
[55] Muhr, H., David, J., Villermaux, J., and Jezequel, P. H., (1996). Crystallization and
precipitation engineering VI: Solving population balance in the case of the precipitation
of Silver Bromide crystals with high primary nucleation rate by using first order upwind
differentiation. Chem. Eng. Sci. 51, 309-319.
[56] Nessyahu, H., Tadmor, E., (1990). Non-oscillatory central differencing fo hyperbolic
conservation Laws. SIAM J. Comput. Phys. 87, 408-448.
[57] Nicmanis, M., Hounslow, M.J. (1998). Finite element methods for steady-state popu-
lation balance equations. AIChE J. 44, 2258-2272.
[58] Nyvlt, J.O., Sohnel, O., Matuchova, M., Broul, M., (1985).
[59] Qamar, S., Elsner, M. P., Angelov, I., Warnecke, G., Seidel-Morgenstern, A. (2006).
A comparative study of high resolution schemes for solving population balances in crys-
tallization. Comp. & Chem. Eng. 30, 1119-1131.
[60] Qamar, S., Ashfaq, A., Angelov, I., Elsner, M.P., Warnecke, G., Seidel-Morgenstern,
A., (2008). Numerical solutions of population balance models in preferential crystalliza-
tion. Chem. Eng. Sci 63, 1342-1352.
[61] Qamar, S. (2008). Modeling and simulation of PB in particulate processes, Habilitation
thesis, Department of Maths, Otto-von-Guericke University Magdeburg.
[62] Qamar, S., Mukhtar, S., Seidel-Morgenstern, A., Elsner, M. P., (2009). An efficient
numerical technique for solving one-dimensional batch crystallization models with size-
dependent growth rates. Chem. Eng. Sci. 64, 3659-3667.
[63] Qamar, S., Mukhtar, S., Seidel-Morgenstern, A., (2010). Efficient solution of a batch
crystallization model with fines dissolution. J. Crystal Growth 312, 2936-2945.
175
[64] Qamar, S., Noor, s., Rehman, M., Seidel-Morgenstern, A., (2010). Numerical solution
of a multi-dimensional batch crystallization model with fines dissolution. Comput. Chem.
Eng. accepted.
[65] Qamar, S., Hussain, I., Seidel-Morgenstern, A. (2011). Application of discontinuous
Galerkin scheme to batch crystallization models. Ind. Eng. Chem. Res. 50, 4113-4122.
[66] Qamar, S., Elsner, M.P., Hussain, I., Seidel-Morgenstern, A.(2012), Effects of seed-
ing strategies and different operating conditions on preferential continuous crystallizers.
Chemical Engineering Science, 71 , 5-17.
[67] Qiu, J. X., Shu, C. W., (2004). Hermite WENO schemes and their application as
limiters for Runge-Kutta discontinuous Galerkin method: one dimensional case, J.
Comput. Phys. 193, 115-135.
[68] Qiu,J. X., Khoo, B. C., Shu, C. -W., (2006). A numerical study for the performance of
the Runge-Kutta discontinuous Chem. Eng. Sci. 1996, 51, 1311-1332. Galerkin method
based on different numerical fluxes. J. Comput. Phy. 212, 540-565.
[69] Ramkrishna, D., Kumar, S., (1985). The status of population balances. Chem. Eng.sci.
3(1), 49-95.
[70] Ramkrishna, D., (2000). Population Balances: Theory and applications to particulate
systems in engineering. Academic Press, San Diego, CA.
[71] Randolph, A., Larson., M. A. (1988). Theory of particulate processes. Academic Press,
Inc. San Diego, CA, Second Edition.
[72] Rawlings, J. B., Witkowski, W. R., Eaton, J. W., (1992). Modelling and control of
crystallizers. Powder Technology 69, 3-9.
176
[73] Reed, W. H., Hill, T. R., (1973). Triangular mesh methods for the neutron transport
equation, Tech. Report LA-UR-73-479, Los Alamos Scientifc Laboratory.
[74] Rhee, H.-K., Aris, R., Amundson, N.R., (1986). First order partial fifferential equa-
tions,
[75] Smith, M., Matsoukas, T., (1998). Constant-number Monte Carlo simulation of pop-
ulation balances. Chem. Eng. Sci. 53, 1777-1786.
[76] Tandon, P., Rosner, D. E., (1999). Monte Carlo simulation of particle aggregation and
simultaneous restructuring. Colloid and Interface Science 213, 273-286.
[77] Toro, E. F., (1999). Riemann solvers and numerical method for fluid dynamics, Second
Edition, Springer-Verlag.
[78] Wankat, P. C., (1995). Rate-Controlled Separations. Springer.
[79] Woo, X. Y., Tan, R. B. H., Braatz, R. D., (2011). Precise tailoring of the crystal size
distribution by controlled growth and continuous seeding from impinging jet crystallizers,
Cryst. Eng. Comm. 13, 2006-2014.
[80] Zhang, P., Liu, R. -X., (2005). Hyperbolic conservation laws with space-dependent
fluxes: II.General study of numerical fluxes. J. Comput. Appl. Math. 176, 105U129.
[81] Ziomek, G., Elsner, M. P., Seidel-Morgenstern, A., (2007). Simultaneous preferential
crystallization in a coupled, batch operation mode. Part I: Theoretical analysis and
optimization. Chem. Eng. Sci. 62, 4760-4769.
177