by
Doctor of Philosophy
in
Metallurgical Engineering
Metallurgical Engineering
January 2005
Copyright
c Byung Sang Choi 2005
of a dissertation submitted by
This dissertation has been read by each member of the following supervisory committee
and by majority vote has been found to be satisfactory.
Chair: H. Y. Sohn
Terry A. Ring
Jan D. Miller
Siva Guruswamy
Edward Trujillo
THE UNIVERSITY OF UTAH GRADUATE SCHOOL
I have read the dissertation of Byung Sang Choi in its final form
and have found that (1) its format, citations, and bibliographic style are consistent and
acceptable; (2) its illustrative materials including figures, tables, and charts are in place;
and (3) the final manuscript is satisfactory to the Supervisory Committee and is ready
for submission to The Graduate School.
Date H. Y. Sohn
Chair: Supervisory Committee
Jan D. Miller
Chair/Director
David S. Chapman
Dean of The Graduate School
ABSTRACT
v
To my wife, Misun, for her dedication...
CONTENTS
ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx
CHAPTERS
1. FUNDAMENTALS OF CRYSTALLIZATION . . . . . . . . . . . . . . . 1
1.1 Crystallization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Solutions to Population Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 CMSMPR Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Homogeneous Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.2 Heterogeneous Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.3 Secondary Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Crystal Growth Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4.1 Surface Energy Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.2 Adsorption Layer Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.2.1 Surface Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.4.2.2 Two-dimensional Growth of Surface Nuclei . . . . . . . . . . . . 32
1.4.2.3 Surface Diffusion and Dislocation . . . . . . . . . . . . . . . . . . . 35
1.4.3 Diffusion-reaction Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.4.3.1 Two-component Diffusion with Different Concentrations . 44
1.4.4 Chemical Reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
1.4.5 Impurity Effects on Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.4.6 Summary of Crystal Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.4.7 Size-dependent Crystal Growth . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.5 Crystal Shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.5.1 Heat Flow and Interface Stability . . . . . . . . . . . . . . . . . . . . . . . 55
1.6 Aggregation of Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
1.7 Attrition of Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2. EXPERIMENTAL DESIGN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.1 General Consideration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.2 Process Chemistry of Crystallization . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.3 Layout and Design of a CMSMPR Crystallizer . . . . . . . . . . . . . . . . . 63
2.4 Reactor Agitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.5 Automatic System Control and Data Acquisition . . . . . . . . . . . . . . . 65
2.6 Summary of Overall Design Parameters . . . . . . . . . . . . . . . . . . . . . . 67
2.7 Statistical Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.7.1 Statistics in Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.7.2 Statistical Design of Experiments . . . . . . . . . . . . . . . . . . . . . . . 69
2.8 Design of Batch Breakage and Aggregation Experiments . . . . . . . . . 70
3. EXPERIMENTAL METHODS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.1 Measurement of Solution Properties . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.1.1 Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.1.2 Viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.1.3 Element Analysis of Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.2 Residence Time Distribution Experiment . . . . . . . . . . . . . . . . . . . . . 80
3.3 Crystallization Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3.1 Precipitation Point Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.3.2 Determination of Reactor Steady State . . . . . . . . . . . . . . . . . . . 84
3.3.3 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.3.4 Calculation of Yield in a Crystallization Process . . . . . . . . . . . . 87
3.3.5 Kinetic Data Measurement and Utilization . . . . . . . . . . . . . . . . 87
3.4 Characterization of Particle Properties . . . . . . . . . . . . . . . . . . . . . . . 88
3.4.1 Element Analysis of Solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.4.2 Crystal Size and Shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.4.2.1 SEM and Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.4.2.2 Crystal Size Distribution (CSD) . . . . . . . . . . . . . . . . . . . . 89
3.4.3 Crystal Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.4.3.1 XRD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.5 Measurement of Particle Size Distribution (PSD) . . . . . . . . . . . . . . . 91
3.5.1 Stabilization of NaCl Particles for ex situ PSD measurement . . 93
4. EXPERIMENTAL RESULTS AND DISCUSSION . . . . . . . . . . 94
4.1 Reactor Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.1.1 RTD measurement as a Reactor Characterization Tool . . . . . . . 94
4.2 Solution Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.1 Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.2 Viscosity measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2.3 Element Analysis of Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.3 Crystallization Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.3.1 Precipitation Point Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.3.2 Crystallization of NaCl from CaCl2 , KCl, and NaCl solution . . 99
4.3.3 Crystallization Condition Monitoring . . . . . . . . . . . . . . . . . . . . 101
4.4 Characterization of Particle Properties . . . . . . . . . . . . . . . . . . . . . . . 107
4.4.1 Elemental Analysis of Solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.4.2 Crystal Size and Shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.4.2.1 SEM and Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.4.3 Crystal Size Distribution (CSD) . . . . . . . . . . . . . . . . . . . . . . . . 114
viii
4.5 Characterization of CSD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.5.1 Reproducibility of CSD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.5.2 Effect of Solution Concentration . . . . . . . . . . . . . . . . . . . . . . . . 119
4.5.3 Effect of Crystallization Temperature . . . . . . . . . . . . . . . . . . . . 122
4.5.4 Effect of Flow Rate (Mean Residence Time) . . . . . . . . . . . . . . . 123
4.5.5 Effect of Mixing Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.5.6 Effect of Different Mixer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.5.7 Overall Effect on Yield and Lmean . . . . . . . . . . . . . . . . . . . . . . . 129
4.6 Batch Breakage and Aggregation Experiments . . . . . . . . . . . . . . . . . 140
4.6.1 Temperature Effect on Breakage and Aggregation . . . . . . . . . . . 141
4.6.2 Mixing Speed Effect on Breakage and Aggregation . . . . . . . . . . 142
4.6.3 Effect of Different Mixer on Breakage and Aggregation . . . . . . . 144
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5. ANALYSIS AND MODEL VALIDATION . . . . . . . . . . . . . . . . . . 152
5.1 Evaluation of Nucleation and Growth Rates . . . . . . . . . . . . . . . . . . . 153
5.2 Application of Hounslows Equation . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.3 Statistical Experimental Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.3.1 Analysis of T samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.3.2 Analysis of TS samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.3.3 Analysis of IS sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.3.4 Influence of Different Mixer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
5.4 Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.4.1 Phenomenological Models for Crystallization Modeling . . . . . . . 197
5.4.2 Crystallization Model Validation . . . . . . . . . . . . . . . . . . . . . . . . 198
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6. GENERAL CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
6.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
APPENDICES
A. CRYSTALLIZER GEOMETRY AND CONFIGURATION . . . 212
ix
G. THERMODYNAMIC MODEL (OLI SYSTEMS) . . . . . . . . . . . . 286
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
INDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
x
LIST OF FIGURES
xii
4.21 Effect of crystallization temperature, feed flow rate, and mixing speed
on the mean size of particles (Lmean ) at steady state for T samples . . . 129
4.22 Linear regression statics on the mean size of particles (Lmean ) vs.
mixing speed at steady state for T samples . . . . . . . . . . . . . . . . . . . . . 131
4.23 Yield of NaCl product at 10 for TS samples . . . . . . . . . . . . . . . . . . . 133
4.24 Effect of temperature, flow rate, and mixing speed on the mean size
of particles (Lmean ) at steady state for TS samples . . . . . . . . . . . . . . . 133
4.25 Linear regression statics on the mean size of particles (Lmean ) vs.
mixing speed at steady state for TS samples . . . . . . . . . . . . . . . . . . . . 134
4.26 Yield of NaCl product at 10 for IS samples . . . . . . . . . . . . . . . . . . . . 136
4.27 Linear regression statics on the product yield (g/cm3 ) vs. flow rate
(ml/min) at steady state (10 ) for IS samples . . . . . . . . . . . . . . . . . . . 137
4.28 Effect of temperature, flow rate, and mixing speed on the mean size
of particles (Lmean ) at steady state for IS samples . . . . . . . . . . . . . . . . 137
4.29 Linear regression statics on the product yield (g/cm3 ) vs. flow rate
(ml/min) at steady state (10 ) for IS samples . . . . . . . . . . . . . . . . . . . 138
4.30 CSDs in differential volume % as a function of time (hours) . . . . . . . . 147
4.31 Effect of reaction temperature on mean size of particles as a function
of time (hours) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.32 Effect of mixing speed on mean size of particles as a function of time
(hours) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.33 Effect of different mixer on CSDs as a function of time (hours) and
differential volume % . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.34 Effect of different type of mixer on mean size of particles as a function
of time (hours) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.1 A plot of population number density distribution . . . . . . . . . . . . . . . . 154
5.2 FEM predictions of number density function (n(v)) for aggregation
and growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.3 A typical plot of nonlinear regression data fit to the particle number
density (N ) using the Hounslows constant aggregation rate equation
at steady state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.4 The order of importance in main and interaction effects on growth
rate (G) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.5 The order of importance in main and interaction effects on nucleation
rate (B0 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.6 The order of importance in main and interaction effects on aggregation
rate (Ka ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
xiii
5.7 Influence of crystallization temperature and mixing speed only (main
effect) on dependent variables at steady state for T samples . . . . . . . . 175
5.8 Nucleation rate (B0 ) vs. number of nuclei (n0 ) at steady state for T
samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
5.9 Nucleation rate (B0 ) vs. particle mean size (Lmean ) at steady state
for T samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.10 Influence of feed flow rate and mixing speed only (main effect) on
number of nuclei(n0 ), growth rate (G), and nucleation rate (B0 ), as
well as aggregation rate (Ka ) at steady state for TS samples . . . . . . . . 179
5.11 Nucleation rate (B0 ) vs. number of nuclei(n0 ) at steady state for TS
samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
5.12 Nucleation rate (B0 ) vs. reciprocal of number of nuclei (n0 ) at steady
state for T samples in . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.13 Particle mean size (Lmean ) vs. nucleation rate (B0 ) at steady state
for TS samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.14 Predicted B0 vs. measured B0 at steady state for TS samples . . . . . . . 185
5.15 Predicted G vs. measured G at steady state for TS samples . . . . . . . . 186
5.16 Predicted Ka vs. measured Ka at steady state for TS samples . . . . . . 188
5.17 Number of nuclei(n0 ) vs. nucleation rate (B0 ) at steady state for IS
samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.18 Nucleation rate (B0 ) vs. growth rate (G) at steady state for IS samples190
5.19 Influence of mixing speed only (main effect) on number of nuclei(n0 ),
growth rate (G), and nucleation rate (B0 ), as well as aggregation rate
(Ka ) at steady state for IS samples . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
5.20 Particle mean size (Lmean ) vs. nucleation rate (B0 ) at steady state
for IS samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.21 Influence of different type of mixer on population number density
distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.22 Measured and predicted volume fraction of particle size distribution
for T 16 sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
5.23 Measured and predicted volume fraction of particle size distribution
for TS 16 sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.24 Measured and predicted volume fraction of particle size distribution
for IS 16 sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
A.1 Geometry of a laboratory scale crystallizer (1.4 ` working volume)
with baffles and a turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
A.2 Dimension and configuration of a 1.4 ` crystallizer (sideview) . . . . . . 214
xiv
A.3 Dimension and configuration of a 1.4 ` crystallizer (topview) . . . . . . 215
A.4 Dimension and configuration of a 6-bladed flat disc turbine (Rushton
type) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
A.5 Dimension and configuration of a 6-bladed pitch blade impeller . . . . . 218
B.1 Concentration profile measured with conductivity, H+ concentration,
and temperature. Operating condition: flow rate = 110 ml/min,
mixing speed = 0 rpm, and reactor volume = 1,300 ml for a tank
without baffles and with a turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
B.2 Concentration profile measured with conductivity, H+ concentration,
and temperature. Operating condition: flow rate = 110 ml/min,
mixing speed = 287 rpm, and reactor volume = 1,300 ml for a tank
without baffles and with a turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
B.3 RTD measured with H+ concentration. Operating condition: flow
rate = 110 ml/min, mixing speed = 287 rpm, and reactor volume =
1,300 ml for a tank without baffles and with a turbine . . . . . . . . . . . . . 229
B.4 Comparison of tm vs. mixer rpm at different flow rate for 1,300 ml
laboratory reactor without baffles and with a turbine . . . . . . . . . . . . . 229
B.5 Comparison of tm vs. mixer rpm at different flow rate for 1.4 `
laboratory reactor with baffles and a turbine . . . . . . . . . . . . . . . . . . . . 230
B.6 Comparison of tm vs. mixer rpm at different flow rate for 1.4 `
laboratory reactor with baffles and an impeller . . . . . . . . . . . . . . . . . . 230
B.7 Comparison of /tm vs. mixer rpm at different flow rates for 1.3 `
laboratory reactor without baffles and with a turbine . . . . . . . . . . . . . 231
B.8 Comparison of /tm vs. mixer rpm at different flow rate for 1.4 `
laboratory reactor without baffles and an impeller . . . . . . . . . . . . . . . . 231
B.9 Comparison of tm /(V /Q) vs. mixer rpm at different flow rate for 1.3
` laboratory reactor without baffles and with a turbine . . . . . . . . . . . . 232
B.10 Comparison of tm /(V /Q) vs. mixer rpm at different flow rate for 1.4
` laboratory reactor with baffles and a turbine . . . . . . . . . . . . . . . . . . 232
B.11 Comparison of tm /(V /Q) vs. mixer rpm at different flow rate for 1.4
` laboratory reactor with baffles and an impeller . . . . . . . . . . . . . . . . . 233
B.12 Velocity vector profile for turbulent flow in the 1.4 ` laboratory reactor
operating at 40 ml/min feed flow rate and a mixer speed of 80 rpm . . . 234
B.13 Output concentration as a function of time for the Fluent prediction
of the residence time distribution for various mixer speeds (i.e., 0, 20,
40, 80 and 200 rpm) and a feed flow rate of 40 ml/min . . . . . . . . . . . . . 236
B.14 Comparison of experimental result of residence time distribution func-
tion, E(t), experimental measurements, and calculated ideal E(t) for
a perfectly mixed CSTR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
xv
B.15 Comparison of experiment and Fluent simulation result for mean
residence time ( ) vs. mixing speed (rpm) at various flow rate for
a 1.4 ` laboratory reactor with baffles and a turbine . . . . . . . . . . . . . . 238
B.16 Comparison of experiment and Fluent simulation result for standard
deviation (/tm ) of RTD vs. mixing speed (rpm) at various flow rate
for a 1.4 ` laboratory reactor with baffles and a turbine . . . . . . . . . . . 239
C.1 Operating flow and control diagram for a CMSMPR crystallization
system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
D.1 A plot of differential Number % vs. particle size, L (m) . . . . . . . . . . . 255
D.2 A plot of population density distribution vs. particle size, L (m) . . . . 256
D.3 A plot of differential Number % vs. particle size, L (m) . . . . . . . . . . . 257
D.4 3D plot of PSDs for TS 04, TS 05, and TS 06 . . . . . . . . . . . . . . . . . . . 258
D.5 3D plot of PSDs for TS 07, TS 08, and TS 09 . . . . . . . . . . . . . . . . . . . 259
D.6 3D plot of PSDs for TS 10, TS 11, and TS 12 . . . . . . . . . . . . . . . . . . . 260
D.7 3D plot of PSDs for TS 13, TS 14, and TS 15 . . . . . . . . . . . . . . . . . . . 261
D.8 3D plot of PSDs for TS 16, TS 17, and TS 18 . . . . . . . . . . . . . . . . . . . 262
D.9 3D plot of PSDs for TS 19, TS 20, and TS 21 . . . . . . . . . . . . . . . . . . . 263
D.10 3D plot of PSDs for TS 22, TS 23, and TS 24 . . . . . . . . . . . . . . . . . . . 264
D.11 3D plot of PSDs for TS 25, TS 26, and TS 27 . . . . . . . . . . . . . . . . . . . 265
D.12 3D plot of PSDs for IS 04, IS 05, and IS 06 . . . . . . . . . . . . . . . . . . . . . 266
D.13 3D plot of PSDs for IS 07, IS 08, and IS 09 . . . . . . . . . . . . . . . . . . . . . 267
D.14 3D plot of PSDs for IS 10, IS 11, and IS 12 . . . . . . . . . . . . . . . . . . . . . 268
D.15 3D plot of PSDs for IS 13, IS 14, and IS 15 . . . . . . . . . . . . . . . . . . . . . 269
D.16 3D plot of PSDs for IS 16, IS 17, and IS 18 . . . . . . . . . . . . . . . . . . . . . 270
D.17 3D plot of PSDs for IS 19, IS 20, and IS 21 . . . . . . . . . . . . . . . . . . . . . 271
D.18 3D plot of PSDs for IS 22, IS 23, and IS 24 . . . . . . . . . . . . . . . . . . . . . 272
D.19 3D plot of PSDs for IS 25, IS 26, and IS 27 . . . . . . . . . . . . . . . . . . . . . 273
F.1 A schematic showing the pinning of an advancing step by absorbed
immobile impurities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
F.2 Dependence of step velocity on supersaturation in the presence of Fe3+
contamination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
F.3 Result of PSD measurements without the addition of CdCl2 . . . . . . . . 282
F.4 Result of PSD measurements with the addition of CdCl2 . . . . . . . . . . 283
F.5 SEM image of particles showing the crystal habit and size after crys-
tallization at 30 C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
xvi
LIST OF TABLES
xix
ACKNOWLEDGMENTS
FUNDAMENTALS OF CRYSTALLIZATION
1.1 Crystallization
Crystallization is an important separation and purification process used in many
areas such as chemical, pharmaceutical, petrochemical, and biotechnological, as well
as electronic industries. Its wide use is due to the highly purified and attractive
form in which the compounds can be obtained from relatively impure solutions
by means of a single processing step. Crystallization can be performed at high or
low temperatures, and generally requires much less energy for separation of pure
materials than other commonly used methods of purification. While crystallization
may be carried on from vapor or a melt, the most common industrial method is
from a solution.
Crystallization may be defined as a phase change in which a crystalline product
is obtained from a solution. When a substance is transformed from one phase to
at constant pressure and
another, the change in the molar Gibbs free energy, G,
temperature is given by
= (2 1 )
G (1.1)
where is the chemical potential phase 1 (solute) and phase 2 (solid). When
G < 0, the transition from 1 to 2 is a spontaneous process. When G
> 0, it is
2
Figure 1.1. Solubility curves for various types of crystallization: Line A, isother-
mal solubility; Line B, positive temperature coefficient of solubility; Line C, negative
temperature coefficient solubility [1].
where Rg is the gas constant, T is the absolute temperature, a is the activity of the
solute, and a0 is the activity of the pure solute in equilibrium with a macroscopic
3
becomes equivalent to
crystal. Assuming the activity coefficients are one, G
Rg T ln(S), where S is the saturation ratio given by
C
S= (1.4)
Ceq
where C is the actual concentration in solution and Ceq is the solubility at the tem-
perature and pressure of the system. For ionic crystal precipitation, the solubility
is given by the solubility product as shown in the following example,
2
A+2 + 2B AB2 (s), Ksp = A+2 0 B 0
(1.5)
The crystal population density, n (number of crystals per unit size per unit
volume of system) is defined by
N
lim =n (1.7)
L0 L
where N is the number of crystals in the size range L per unit volume. The
value of n depends on the value of L at which the interval dL is taken, i.e., n is a
function of L (Figure 1.3 (a)). The number of crystals in the size range L1 to L2 is
thus given by Z L2
N = ndL (1.8)
L1
3. Particles formed only by nucleation and increase in size only through growth
(negligible breakage, attrition, and agglomeration).
Figure 1.3. Population plots characterizing the crystal size distribution, and the
nucleation and the growth kinetics for a CMSMPR crystallizer [3].
5. All crystals of the same shape, and shape factors for particles are not a
function of size (use single characteristic size as the dimension to describe
the entire size range).
Using the same approach as is used to write material and energy balances, a
population balance can also be constructed. Considering an arbitrary size within
the crystallizer, for size range L1 to L2 with population densities n1 and n2 ,
respectively, the mass balance is
Particles may enter or leave the size range via either growth or flow. If G (= dL/dt)
is denoted as the growth rate of the characteristic dimension, V is the crystallizer
volume, and Q is the volumetric flow rate, the number balance equation yields,
V n1 G1 + Qi ni L = V n2 G2 + Q
nL (1.10)
L = L2 L1 , and Q
nL is the number of particles leaving the size range by flow.
Upon arrangement,
n 2 G2 n i G2
V = Qi ni Q
n (1.11)
L
As L approaches zero,
d(Gn)
V= Qi ni Qn (1.12)
dL
the mean retention time, = V /Q, and assuming no particles in feed stream, i.e.,
ni = 0,
d(Gn)
+n=0 (1.13)
dL
Apply McCabes L law, G 6= G(L),
dn n
= (1.14)
dL G
which is an ordinary differential equation that can be separated and integrated.
If the boundary condition of n0 representing zero-sized particles is employed, i.e.,
n(0) = n0 , the result is
L
n = n0 exp (1.15)
G
Equation 1.15 is the fundamental relationship between crystal size L and population
density n characterizing the CSD. The quantity n0 is the population density of
nuclei (zero-sized crystals). A plot of log n versus L should give a straight line with
an intercept at L = 0 equal to n0 and a slope of 1/G (Figure 1.3 (b)). Therefore,
if the residence time is known, the crystal growth rate, G, can be calculated.
Furthermore, CSD data can be used to give information on the nucleation and
growth kinetics. If the nucleation rate, B0 , is the rate of appearance of near zero-
sized particles, B0 can be expressed as a function of the supersaturation, c.
dN
B0 = = k1 cb (1.16)
dt L0
Therefore, the nucleation rate may be expressed in terms of the growth rate by
B0 = n0 G (1.18)
or
B0 = k3 Gi (1.19)
where
b
i= (1.20)
g
Consequently
n0 = k4 Gi1 (1.21)
so a plot of log n0 versus log G should give a straight line of slope i 1 (Figure 1.3
(c)) or a plot of log B0 versus log G should give a line of slope i. Thus the kinetic
order of nucleation, b, may be evaluated if the kinetic order of growth, g, is known.
The moments of the size distribution can be calculated for the exponential
number density distribution (Equation 1.15) obtained with the idealized CMSMPR
model. These are shown in Table 1.1. The nucleation rate B0 based on the
1.3 Nucleation
A supersaturated solution is required for crystallization to occur. A supersat-
urated solution is not at equilibrium. In order to relieve the supersaturation and
move towards equilibrium, the solution crystallizes. Once crystallization starts,
however, the supersaturation can be relived by a combination of nucleation and
crystal growth. It is the relation of the degree of nucleation to crystal growth that
controls the product crystal size and size distribution, and is, therefore, a crucial
aspect of industrial crystallization processes. The fundamentals of nucleation and
crystal growth will be discussed in this and the next section.
During crystallization new particles are born in the size distribution by nucle-
ation processes. The nucleation rate, which appears as a boundary condition at size
L = L 0 in the population balance, generally has a dominating influence on the
particle size distribution. Nucleation is also the least understood of the various
rate processes in precipitation. There are three main categories of nucleation,
i.e., primary homogeneous, primary heterogeneous, and secondary nucleation. The
mechanisms governing the various types of primary and secondary nucleation are
different and result in different rate expressions. Also the relative importance of
each type of nucleation varies with the crystallization conditions.
It is desirable to classify the various mechanisms of nucleation as shown in
Figure 1.4. Primary nucleation occurs in the absence of crystalline surfaces, whereas
secondary nucleation involves the active participation of these surfaces. Homoge-
neous nucleation rarely occurs in practice, however, it forms the basis of several
10
a + a = a2
a2 + a = a3
a3 + a = a4
a2 + a2 = a4
..
.
acm + am = ac (1.23)
11
that continues until a critical size is reached. The rate of nucleus formation by this
mechanism is given by an Arrhenius type of expression as
G
B0 = A exp (1.24)
kB T
where A is the preexponential factor and has a theoretical value of 1030 nuclei/cm3 s.
The thermodynamic considerations for homogeneous nucleation were developed
by Gibbs (1928), Volmer (1939), and others. The overall free energy change per
embryo, G[erg/embryo] , of the aggregate is a result of two terms, the free energy
due to the new surface (a positive quantity) and the free energy due to the formation
of new solid (a negative quantity) [8].
V
G = G + A (1.25)
V
V HT T
= 1 + A (for cooling)
V T0
where
= kB T ln(S)
G (1.26)
or with cooling,
HT 1 1
ln(S) =
kB T T0
where HT is the enthalpy of the phase transition (condensation for a gas or
fusion for a melt) occurring at the temperature T below the equilibrium temperature
T0 . kB is Boltzmanns constant. S = a/a0 is the saturation ratio, where a is the
activity of the condensing species and a0 is the activity in equilibrium with the
12
For a sphere,
4
V =
3
A = 4
V = 8
A = 24
When the supersaturation, S < 1.0, G(r) is always positive and cluster formation
is non-spontaneous. When the supersaturation, S > 1.0, G(r) has a positive
maximum at the critical size, r , with a maximum Gibbs Free energy, Gmax =
G(r ) which is the energy barrier for nucleation as shown in Figure 1.5. Cluster
larger than the critical size will decrease their free energy by further growth, giving
stable nuclei which grow to form macroscopic particles. Below the critical size,
clusters will decrease their free energy by dissolving. The critical size, r , is obtained
by setting dG(r)/dr = 0 giving
2A V
r = (1.28)
3V kB T ln(S)
Nielsen [7] has adopted the terminology that an embryo is subcritical and a
nuclei is supercritical in size. This critical size corresponds to a value of the free
energy at the maximum of
A r2
Gmax = (1.29)
3
In general, the maximum free energy of a critical cluster, Gmax , includes strain
energy, magnetic energy, electrical energy, etc., in addition to the volume free-energy
13
Figure 1.5. Classical nucleation theory, dependence of nuclei size on Gibbs free
energy as a function of saturation ratio, S [1].
and surface free-energy terms. If they are relevant, these contributions should be
added to obtain the maximum free energy of the critical cluster. Figure 1.6 shows
the critical nuclei size as a function of saturation ratio, S. The standard critical size
of nuclei is given by
2A V
rstd = (1.30)
3V kB T
which occurs in the limit of S = e (=2.718).
For a given value of S, all particles with r r will grow and all particles
with r < r will dissolve. This phenomena is referred to as ripening [9]. At high
supersaturation, the critical size, r , approaches the size of an individual molecule,
where the theory is invalid. At such large supersaturation, the rate of nucleation
is limited by the collision of molecules by diffusion. Homogeneous nucleation is
difficult to observe in practice due to the presence of dissolved impurities and
physical features such as crystallizer walls, stirrers, and baffles.
Embryo Concentrations:
14
Using a Boltzman distribution and these free energy concepts, the equilibrium
number density of embryos of size, r, is given by
G(r) NA G(r)
Ne (r) = N0 exp = exp (1.31)
kB T V kB T
where n0 is number of molecules per unit volume of condensing material in the gas
phase, NA is Advogadros number, and kB is Boltzmanns constant. This expression
is only valid over a range of G values where G is greater than zero. This embryo
size distribution function is shown in Figure 1.7. Detailed experiments have been
carried out studying the nucleation of liquid droplets in clean vapor [10, 11]. The
surface tension of the liquid can be measured readily, and the clusters of liquid in
the vapor phase are likely to be spherical. The experimental data on these systems
has verified the validity of the nucleation equations.
Figure 1.7. Embryo size distribution function for various values of the saturation
ratio, S. Plot generated for a sphere [1].
# 1 dN (r)
B0Homo (s) =
cm3 mmin r dt r=rs
where
The classical nucleation rate, J = dN (r )/dt, was divided by a critical nucleus size
of r = 2A V /3V kB T ln(S) to make use of the classical theory of nucleation rate
with population balance equation. It is clear that the nucleation rate increases
with increasing supersaturation and temperature, and decreases with an increase
in surface energy.
of nuclei. The size of the critical nucleus decreases with increasing supersaturation.
Thus, the probability of the nuclei surviving to form crystals is higher. As the su-
persaturation is increased, the micro-roughness of the crystal surface also increases,
resulting in a larger nuclei population. In general, the secondary nucleation rates
are enhanced with increasing supersaturation. However, the nucleation exponent
is found to be lower than that for primary nucleation.
Stirring the solution leads to a thinning of the absorbed layer and hence should
lead to lower nucleation rates. However, it was found out that the nucleation rate
increased with the degree of agitation for smaller crystals of magnesium sulfate. The
effect of hardness of the contact material and the crystal hardness also found out
to be important factor on the generation of secondary nuclei; a harder material is
more effective in enhancing the nucleation rates. In situ experiments have indicated
that micro-attrition is also an important source of nuclei. Crystal hardness also
affects nucleation behavior, i.e., a hard, smooth crystal is less effective. Irregular
crystals with some roughness are generally more active. Larger crystals yield more
secondary nuclei. Smaller crystals generally follow the path of the liquid and have
a smaller probability of coming in contact with the stirrer or walls to generate
secondary nuclei.
It is well known that a small amount of impurity can profoundly affect the
nucleation rate, however, it is impossible to predict the effect prior. The presence
of additives can either enhance or inhibit the solubility of a substance. Enhanced
solubilities would lead to lower supersaturations and lower growth rates. If it is
postulated that the impurity adsorbs on the crystal surface, then two opposing
effects come into play the presence of an additive would lower the surface tension
and lead to higher growth rates, however, the impurity adsorption blocks potential
growth sites and lowers nucleation rates. Thus, the effect of impurities is complex
and unpredictable.
Break-up of the crystals has been difficult to predict and it is hard to avoid.
It affects the CSD and the crystal growth rate by reducing the size of the large
crystals and increasing the number of small crystals. Unstable secondary nucleation
19
Typical values of b lie between 0.5 and 2.5. These values are much lower than
the typical values for primary nucleation by either homogeneous or heterogeneous
mechanisms where B0 values between 6 12 are more common. The importance
of MT is first order, i.e., j = 1, suggesting that contact of the crystals with the
walls and impeller is the important phenomenon. However, some systems, i.e.,
K2 SO4 [20] and KCl [21], have much lower values of j 0.4. Typical values
of h range from 0 to 8 but most fall between 2 and 4, which are expected from
semi-theoretical models [22].
Gahn and Mersmann [23, 24] have presented a mechanical model for crystal
break-up due to crystal-impeller collisions. Their model describes the effect of the
impact energy and the material properties of the crystals. Impeller rotation speed
and the geometries of the impeller and the draft tube were used to obtain the
collision velocities and the probability of collision.
Secondary nucleation occurs when the impact energy in collision of a crystal
with a crystallizer or other crystals reaches above a critical value. Small fragments
are born and the volumes of the original crystals are reduced. The size of the
smallest fragments, La,min (Equation 1.37), depends only on the material properties
20
of the crystal. The size of the largest fragments, La,max (Equation 1.38), depends
additionally on the impact energy [23]:
32
La,min = (1.37)
3 H 2 Kr
1/3
1 H 2/3 Kr
La,max = Wp4/9 (1.38)
2
where
It has been shown that only fragments with La > La,ef f possess such a high
statistically mean growth rate that they are effective fragments or secondary nuclei.
The ratio Na,ef f /Na,tot of effective fragments to the total number of fragments is
given [23] by
Z Z La,ef f
Na,ef f
= n(La )dLa = 1 n(La )dLa (1.39)
Na,tot La,ef f 0
The effective fragment size, La,ef f , depends on supersaturation because the mean
growth rate of a size interval increases with supersaturation. It also has been shown
that only fragments larger than La,ef f 25m contribute significantly to growth
max . The effective size, La,ef f , is smaller when supersaturation is
G(La ) > 0.1G
higher and vice versa.
1. Transport of solute from the bulk solution to the crystal surface through an
interface between the crystals and the growth medium.
4. Attachment to a step.
Figure 1.8. The energetics of crystal growth from solution. (a) movement
of the solvated solute molecule and (b) corresponding energy changes for each
transformation [12].
and thus limit the growth rate. The solute does not become a part of the crystal
structure until the enthalpy of crystallization has been librated and desolvation is
complete. Figure 1.8 schematically presents this process along with the energetics
of each step [12].
The growth rate is determined by the rate at which the crystals pass through
each of these processes. The process that has the largest impact on the overall
rate of growth is clearly the process in which the rate of growth is the slowest.
This particular process is therefore known as the rate-determining process. Exactly
which of the processes becomes the rate-determining process depends on the growth
environments and the growth conditions.
If the growth medium is solution, many of the crystals in the medium will
grow as polyhedrons. In this case the growth interface is flat on an atomic scale.
Therefore, the rate-determining step is the interface kinetic process because the
orientation time for this particular process is sufficiently long compared with both
the atom-supply process and the heat removal process. For the crystal growth
in a solution, the concentration of solute atoms in the growth medium is lower
than that in the crystals, but the density of the solution itself is almost equal to
that of the crystal. Therefore, the volume diffusion coefficient is estimated to be
smaller by two orders of magnitude than that of the mixed gas phase. Because the
density of solute atoms in a solution phase is sufficiently low compared to that in
crystals, solution phase is called thin growth media. However, the density of atoms
in the environment surrounding the crystals is significantly different from that of
the atoms in crystals, thus one phase can be clearly distinguished from the other
at a boundary (surface).
The growth units can either remain at their initial point of contact, diffuse across
the surface, and eventually integrate into the structure of the crystal, or return to
solution. The nature of the crystalsolution interface is of critical importance. A
simple representation of both atomically smooth and rough surface is shown in
Figure 1.9. This figure illustrates two types of atomic structure of crystal surfaces,
i.e., smooth (atomically flat close-packed interface) and rough (atomically diffuse
24
(a) (b)
interface) surfaces. On the smooth surface, all the atoms shown as a cube are
identical. Inside a crystal, any atom will have six neighbors with a binding energy
of three times the atomic bond energy (i.e., 3Eaa ), since each bond is shared
between two atoms; when the binding energy per atom pair is (1/2)Eaa and only
nearest-neighbor interactions are considered. If a single extra atom is to be added
to the smooth crystal interface, it forms a bond with only one nearest neighbor with
the atomic pair bond energy. Single or multiple atoms can be added to form extra
bonds with the first adatom stabilizing the cluster, but the bonding energy of the
first atom to the surface is clearly a major barrier to the crystal growth. Therefore,
in this case the transition from liquid to solid occurs over a rather narrow transition
zone approximately one atom layer thick. On the other hand, any atom added to
a rough surface has a greater probability of becoming part of that surface than a
crystal with a smooth surface since the probabilities of the binding energies will
depend on through terms exp[((1/2)Eaa )/kB T ], exp[(Eaa )/kB T ], and etc., where
T is the interface temperature and kB is Boltzmanns constant. An atom added
to a rough surface has several possible sites of attachment with different binding
energies as shown in Table 1.2. The growth rate of a crystal with a rough surface
will be higher than that of a crystal with a smooth surface due to the transition from
25
a
These sites are equivalent
to the sites in Figure 1.9
b
This is the same as the ad-
dition to an atomically smooth
surface
liquid to solid occurs over several atom layers. Rough surfaces tend to maintain
their roughness during growth; atoms attach themselves at sites which are preferred
sites for subsequent additions. On a smooth surface, the rate-limiting growth step
is the addition of the first atom to the surface, since the subsequent addition of
other atoms to the newly created corners is energetically favored. Consequently,
this new layer is quickly completed, yielding another smooth surface with a higher
activation energy. Therefore the crystal growth on a smooth surface is referred to
as a lateral growth whereas a continuous growth refers to the addition of atoms
to a rough surface. As well defined crystal faces are considered to be planar as
shown in Figure 1.8, new solute molecules arriving on these planes must find a
suitable incorporation site. This sequence then allows for surface attachment with
subsequent surface diffusion to a step and incorporation into a kink-site. Growth
then occurs by a flow of steps across the surface. With a step velocity of , a
step spacing , and step height of a (equivalent to a cubic embryo of height), it is
immediately evident that the growth rate normal to the surface, R, is given simply
as:
a
R= (1.41)
26
Temkin [32] and Jackson [33] have characterized the roughness of a crystal
surface with a surface entropy factor, , defined as
2
= (Ess + Ef f Esf ) (1.42)
kB T
where E is the bond energy. Subscripts f and s correspond to the fluid and
solid, respectively. The free energy change due to solidification, Gs , is given
in Figure 1.10 as a function of the fractional occupation of a single layer, x, and
the surface entropy factor, . From this plot it is shown that when < 2, the
Figure 1.10. Dependence of fraction of the surface occupied on the Gibbs free
energy for various degrees of surface roughness [33].
where ai is the area of ith face of a crystal bounded by n faces, and gi is the surface
free energy per unit area of the ith face. Therefore, if a crystal is allowed to growth
in a supersaturated medium, it should develop into a equilibrium shape, i.e., the
development of the various faces should be in such a manner as to ensure that
the whole crystal has a minimum total surface free energy for a given volume. Of
course, a liquid droplet is very different from a crystalline particle; in the former
the constituent atoms or molecules are randomly dispersed, whereas in the latter
they are regularly located in a lattice structure. The surface energy and the rate of
growth of a face, however, should be inversely proportional to the reticular or lattice
density of the respective lattice plane, so that faces having low reticular densities
would grow rapidly and eventually disappear. In other words, high index faces grow
faster than low. So far there is no general acceptance of the surface energy theories
of crystal growth, since there is little quantitative evidence to support them; their
failure to explain the well-known effects of supersaturation and solution movement
on the crystal growth rate.
equilibrium is established between this layer and the bulk solution. The adsorption
layer, or third phase, plays an important role in crystal growth and secondary
nucleation. The thickness of the adsorption layer probably dose not exceed 10 nm,
and may even be nearer 1 nm.
term in Equation 1.44 corresponds to the change in free energy due to the volume
of the embryo. The second term represents the free energy due to the length of the
edge of surrounding the embryo. Therefore, for the case of spontaneous nucleation
(i.e., S > 1), Gs has a maximum value of
e2 L2 v e2 L2 a2
Gsmax = = (1.45)
4A akB T ln(S) 4A kB T ln(S)
occurring at
e L v e L a2
rS = = (1.46)
2A akB T ln(S) 2A kB T ln(S)
The last term in the previous two equations was obtained by noting that v ' a3 .
Therefore, at high temperature and solution supersaturation, low value of rS is
expected. The number of molecules in a critical nucleus, n , can also be written as
aA (rS )2 A (rS )2 e2 L2 a2
n = = = (1.47)
v a2 4A (kB T ln(S))2
The mole fraction of surface embryos, Xr , of size r is given by
Gs (r)
Xr = exp (1.48)
kB T
Equation 1.48 can also be written in terms of the surface concentration of embryo
by dividing by the unit area of the crystal. Recall that in three-dimensional
nucleation theory the nucleation rate was on a per volume of the embryo basis,
but for two-dimensional nucleation theory JS is on an embryo per surface basis
(i.e., embryo/cm2 s). From this, the simplified surface nucleation rate [7], JS , on a
unit area per unit time is estimated as
Gsmax
D
JS = exp (1.49)
d4 kB T
critical value of the supersaturation ratio, Sc . These critical saturation ratios are
presented in decreasing order as
This order of critical saturation ratios is indication to the order of the nucleation
activation energy which controls the nucleation rates. A plot of each entity in the
total nucleation rate equation is given in Figure 1.11. It shows that heterogeneous
# dN (r)
Js =
cm2 min dt r=rs
Gs (rs )
= 4DN0 rs Nc V 1/3 exp
kB T
!
2 L2 V 2/3
! " #
L V 2/3 2A RT ln(S)
= 4DN0 Nc V 1/3 exp
A RT ln(S) kB T
(1.51)
where
(1.52)
dr
= d2 NA Ceq D(S 1) (1.53)
dt
where r is the size of the growing surface nuclei and d is the molecular diameter
of the diffusing species. The two-dimensional growth rate must be compared to
the surface nucleation rate, JS , to have further insight into the rate-determining
step of crystal growth. To make this comparison, the time between two nucleation
events, (JS A)1 , on a given surface area, A, is compared to the time necessary to
grow a two-dimensional layer over the whole surface, [L/(dr/dt)s ], where L is the
maximum length from the surface nuclei to the edge of the crystal.
A R2 D GSmax
dR
= dAJS = exp (1.54)
dt m d3 kB T
A being a crystal shape factor given by A/R2 for a cylinder and d is the mean
molecular diameter. This is extraordinary simple concept that the rate limiting
(a) (b)
step is the formation of a critical size of nuclei. The odd result of this theory is
that (dR/dt) A R2 . Therefore, for crystals comprised of [100] planes, even if
no symmetric initially, should grow eventually into perfect cubes. No experiments
have not carried out yet to show any relation between these variables.
each layer of the crystal is the result of the intergrowth from many individual surface
nuclei. This rate of crystal growth is described by a polynuclear crystal growth (see
Figure 1.12 (b)). Again with a nucleation rate per unit area and time the change
in crystal size, R,as
GSmax
dR Dd 2/3 2/3
= (NA Ceq ) (S 1) exp (1.55)
dt P 3 kB T
Between the above two extreme cases, the nucleation occurs and the nuclei
can grow and spread at finite velocities [7] as illustrated in Figure 1.13. The
Figure 1.13. Two-dimensional growth of surface nuclei by birth and spread model.
exact rules relating lateral spreading velocities to supersaturation and other system
parameters represent the basic difference between the different birth and spread
models. However, the above two growth mechanisms are of equal importance when
the time between two nucleation events is 60% of the time necessary to complete
the entire surface. That is
L
(JS A)1 = 0.6 (1.56)
dr
dt S
35
dR 0.2V DCeq (S 1)
= (1.57)
dt R
which is 0.2 times the diffusion-controlled growth rate, which will be discussed later.
V is the molar volume of the precipitate in Equation 1.57. This result indicates
that when the crystal growth rate is less than 20% of the diffusion-controlled growth
rate, the monocular layer growth rate dominates; above 20% of the rate of diffusion-
controlled growth, polynuclear layer growth dominates.
In general, the final equation shows that R does not depend upon crystal area
but is a complex function of supersaturation and temperature, i.e., R increases
both with S and T . However, no maximum of R with S is generally predicted.
These three models briefly discussed above are not satisfactory in predicting or
correlating crystal growth rates from solution. Some of the unusual dependencies
of rate with crystal area or supersaturation have already received comment. Also,
with reasonable values of the constants and system parameters, it is found that
at low values of S, predicted rates are much less than those generally observed
experimentally.
(a) (b)
into the crystal at the monocuclear step AB. At all points on the curve AB, the
step moves such that the angular velocity decreases with the distance away from
the starting point A. This produces a screw or spiral growth pattern on the crystal
surface, as illustrated in Figure 1.14. This type of growth has been shown in a
large variety of crystals growth from vapor [41], from aqueous solution [42], and
from melt [12]. The presence of growth spirals gives evidence for Franks screw
dislocation model [43].
Burton, Cabrera, and Frank [39] and Bennema and Gilmer [44] developed a
theory to predict the crystal growth rate for screw dislocations, where the growth
rate depends on the shape of the growth spiral. For an Archimedian spiral, as
shown in Figure 1.14, the distance between the steps of the spiral, y0 , is given as
y0 = 4rS (1.58)
19e L h2
y0 = 19rS = (1.59)
2A kB T ln S
37
The base angle, , of the growth cone formed by the spiral is given by
1 h
= tan (1.60)
y0
where h is the new layer thickness deposited with each turn of the spiral. The rate
of solute deposition is a result of a volume diffusion flux, jv , through the bulk, and
surface diffusion, js , to the step, where integration into the crystal takes place. For
simplicity, suppose that the curvature of the spiral can be neglected and that the
growth of the spiral is slow compared to surface diffusion. Figure 1.15 is a schematic
representation of this surface. At steady state, the two fluxes must be equal, giving
Figure 1.15. Schematic representation of the solute flux to a step site [36].
djs (y)
jv = 0 (1.61)
dy
where the surface flux, js , can also be expressed in terms of a surface diffusion
coefficient, Ds , and a surface concentration, nS .
dnS ds
js = Ds = Ds nse (1.62)
dy dy
where nse is the equilibrium surface concentration and s is the local saturation
ratio at the step surface. The volumetric flux, jv , can be expressed in terms of a
38
volume diffusion coefficient, Dv , a solute surface concentration (CS = nse s /h) and
a bulk solution concentration (C = nse S/h), as follows:
Dv nse (S s ) Dv (C CS )
jv = = (1.63)
h
where again h is the thickness of each layer, is the diffusion boundary layer
thickness and S is the saturation ratio in the bulk solution. Substituting these
two fluxes into the flux balance Equation 1.61, the following differential equation
results:
d2 s
yS2 + s = S (1.64)
dy 2
where ys = (Ds h/Dv )1/2 is the mean distance traveled by the solute molecules on
the surface to arrive at a step, which is always less than or equal to y0 (the average
distance between steps). This second-order ordinary differential equation has the
solution as
cosh (y/ys )
S s (y) = (S) (1.65)
cosh (y0 ys )
for the following boundary conditions:
S s = S at y = y0
s = S at y = 0
where = 1 s /S at y = y0 .
Note that the above solution is only valid in the regime where the distance
between each kink site is much smaller than the distance between the steps (i.e.,
x0 ys , as shown in Figure 1.15. If x0 > ys , it is necessary to introduce an extra
factor into this equation to account for the nonplanar growth fields around the
kinks. For this condition, the flux of molecules to the step may be written as
ds Ds nse S tanh(y0 /ys )
js |y0 = Ds nse = (1.66)
dy y0 ys
The crystal growth rate (in the z-direction), dR/dt, can be calculated from the flux
of steps in the y-direction times the step height, h. The step flux, jstep (i.e., the
flux of one step per unit time), is described by
js |y0
jstep = (1.67)
y0 d
39
where is the crystal density. Making these substitutions, the crystal growth rate
can be shown to be
dR Ds nse S y0
= tanh
dt y0 ys ys
2
Ds nse S S1
= 2
tanh (1.68)
ys S1 S
where
y0 9.5e d
S1 = S= (1.69)
ys ys kB T
Due to complications associated with parameter determination, the first term in
the above equation is usually reduced to an experimentally determined constant.
This theory predicts a second-order dependence of supersaturation on the particle
growth rate (i.e., dR/dt S 2 ) when S S1 (low supersaturation condition), and a
linear dependence (i.e., dR/dt S) when S S1 (high supersaturation condition).
If many screw dislocations emerge on the same dislocation line of the crystal
surface, multiple spirals will result. To account for the cooperative effect of multiple
spirals, a factor, , must be introduced into the growth equation, giving
2
dR Ds nse S S1
= 2
tanh (1.70)
dt ys S1 S
On the assumption that there would be a thin stagnant film of liquid adjacent to the
growing crystal face, through which molecules of the solute would have to diffuse,
Nernst (1904) modified Equation 1.71 to the form
dm D
= A(C C ) (1.72)
dt
41
where D is the coefficient of diffusion of the solute and is the length of the diffusion
path.
The thickness of the stagnant film would obviously depend on the relative solid-
liquid velocity, i.e., on the degree of agitation of the system. Film thicknesses up
to 150 m have been measured on stationary crystals in stagnant aqueous solution,
but values rapidly drop to virtually zero in vigorously agitated systems. As this
implies an almost infinite rate of growth in agitated systems, it is obvious that the
concept of film diffusion alone is not sufficient to explain the mechanism of crystal
growth. Furthermore, crystallization is not necessarily the reverse of dissolution.
A substance generally dissolves at a faster rate than it crystallizes under the same
conditions of temperature and concentration. Another important finding was made
by Miers [48], who determined, by refractive index measurements, the solution
concentrations near the faces of crystals of sodium chlorate growing in aqueous
solution; he showed that the solution in contact with a growing crystal face is not
saturated but supersaturated.
In the light of these facts, a considerable modification was made to the diffusion
theory of crystal growth by Berthoud [49] and Valeton [50], who suggested that there
were two steps in the mass deposition (i.e., the diffusion process), whereby solute
molecules are transported from the bulk of the fluid phase to the solid surface,
followed by a first-order reaction when the solute molecules arrange themselves
into the crystal lattice. These two stages, occurring under the influence of different
concentration driving forces, can be represented as
dm
= kd A(C Ci ) (diffusion) (1.73)
dt
and
dm
= kr A(Ci C ) (reaction) (1.74)
dt
where kd is a coefficient of mass transfer by diffusion, kr is a rate constant for the
surface reaction (integration) process, and Ci is the solute concentration in the
solution at the crystal-solution interface. Figure 1.17 shows these two stages with
the various concentration driving forces. It must be clearly understood that this is
42
only diagrammatic: the driving forces will rarely be of equal magnitude, and the
concentration drop across the stagnant film is not necessarily linear. Furthermore,
there appears to be some confusion in recent crystallization literature between this
hypothetical film and the more fundamental boundary layers.
Equations 1.73 and 1.74 are not easy to apply in practice because they involve
interfacial concentrations that are difficult to measure. It is usually more convenient
to eliminate the term Ci by considering overall concentration driving force, C C ,
which is quite easily measured. A general equation for crystallization based on this
overall driving force can be written as
dm
= KG A(C C )g (1.75)
dt
where
1 1 1
= + (1.77)
KG kd kr
or
kd kr
KG = (1.78)
kd + kr
For the cases of extremely rapid reaction, i.e., large kr , KG kd and the
crystallization process is controlled by the diffusional operation. Similarly, if the
value of kd is large, i.e., if the diffusional resistance is low, KG kr , and the process
is controlled by the surface integration. It is worth pointing out that whatever the
relative magnitude of kd and kr they will always contribute to KG .
The diffusional step (Equation 1.73) is generally considered to be linearly de-
pendent on the concentration driving force, but the validity of the assumption of a
first-order surface reaction (Equation 1.74) is highly questionable. Many inorganic
salts crystallizing from a aqueous solution give an overall growth rate order, g, in
the range 1 to 2. The rate equations, therefore, can be written as
1 dm
RG = = kd (C Ci ) (diffusion) (1.79)
A dt
= kr (Ci C )r (reaction) (1.80)
= KG (C C )g (overall) (1.81)
RD = KD (C C)d (1.82)
However, apart from such simple cases, Equation 1.84 cannot be solved explicitly
for RG and the relationship between the coefficients KG , kd , and kr remains obscure.
Recently, however, Sobczak [51] has proposed an integral method, based on a
linearization of Equation 1.84, which allows reasonable values of kd and kr to be
estimated.
aA+ + bB
Aa Bb (s) (1.88)
For the crystal Aa Bb (s) being formed, the flux of A (i.e., JA ) and the flux of B
(i.e., JB ) must be related to the crystal stoichiometry by
A A
B B
JA JB DA C Ceq DB C Ceq
= = = (1.90)
a b aR bR
A B
However, both Ceq and Ceq are related through the solubility product. For the case
of a 1:1 (a : b) stoichiometrywith similar diffusion coefficients (i.e., DA = DB =
45
A B
D, an analytical solution is obtained by eliminating Ceq and Ceq in Equation 1.90,
giving s
B 2
A B
A
D C + C C C
JA = JB = Ksp + (1.91)
R 2 4
A B A B A B
For the case of C C (or equivalently, C C Ksp = Ceq Ceq ) this reduces
to !
dR V DC
B
= (1.93)
dt R
which is the limiting single component case when the equilibrium concentration
B B
is much less than the bulk (i.e., C Ceq ). However, Equation 1.92 accounts
for the varying concentrations of two reactants within a solution. It should be
A B
noted here that the situation of C C , corresponding to the solution given in
Equation 1.93, relates to high saturation ratios.
where r is the order of the surface integration process and Da is the Damkohler
number for crystal growth, which represents the ratio of the pseudo-first-order rate
coefficient at the bulk conditions to the mass transfer coefficient, defined by
(1 w)
Da = kr (C C )r1 (1.96)
kd
the interfacial free energy of the solute crystal. The theories developed from
either two-dimensional nucleation or surface diffusion bases would predict a growth
retardation since an increase in would increase the size of the critical nucleus
and decrease the nucleation rate J, the step spacing y0 , and the step velocity js .
A more interesting idea assumes that impurities adsorb tightly on the surface (in
a regular array) and hinder the smooth flow of steps [54]. That is, an advancing
step is somehow forced to flow between impurity atoms. The steps can then have a
local, high curvature. This results consequently a decrease in step velocity. In fact,
if two impurity atoms are separated by a distance less than 2c , the curvature is so
large that step movement completely ceases (refer to Figure F.1 in Appendix F).
Comparing growth rates for two solution, with and without the impurity, it
can be shown that in the impurity case, there is a critical supersaturation (i.e., d
shown in Figure F.2 in Appendix F) below which no growth should occur. If there
is an impurity flux Ji to the surface, the rate of growth without impurity has to
exceed about 142c Ji h in order for any finite growth to occur in the impurity case.
The critical supersaturation for an impure solution, d , is:
Jia
d (1.97)
Tb
where a is roughly 1/3 to 1/4 and b, 3/4 to 2/3 and the proportionality constant
is given in terms of the system parameters. The significance of Equation 1.97 is
that growth in an impure solution should not occur if < d . When > d , two
equally good answers are predicted, that is, there appear to be two mechanisms for
growth through the impurity array [37]. If both routes are equally likely (or nearly
so), unstable growth could result.
With respect to the movement of the steps, Chernov [46] suggests that there are
two effects of impurities: (1) reduction of the number of kink sites, if the impurities
are relatively small and mobile, and (2) action as an obstacle for step movement
(i.e., step pinning) if the impurities are large and immobile. Reducing the number
of kink sites will also effectively decrease crystal growth. Chernov estimates that an
impurity concentration of 103 M will drastically reduce the crystal growth rate.
48
Slavnova [55, 56, 57] has observed qualitative confirmation of Chernovs theory.
The effects of poisons on nucleation kinetics and the spiral shape are discussed by
Sears [58].
The effect of additives on the rates of nucleation can be accounted for with
the combination of the Gibbs adsorption equation and the Langmuir adsorption
isotherm as they effect the interfacial energy, , resulting in [59]:
= m kB T ln (1 + KLang aa ) + 0 (1.98)
where
kB = Boltzmanns constant
T = temperature, and
dl h m i
D(l, S) = = K f (S) g(l) (1.100)
dt s
49
Effect of Supersaturation:
For all the models except for those included in the two-dimensional nucleation,
the prediction of the derivative ( ln R/ ln S) is either 1 or 2. For the Chernov
bulk diffusion and the BCF surface diffusion (screw dislocation) models, the slope
of this derivative changes from 2 at low values of S to 1 at high S (For the BCF
bulk diffusion the slope is slightly greater than 2 at low S and decreases to 1 at
high S.). For other models, it remains equal to unity at all values of S.
For the two-dimensional nucleation theories, the slope changes with S and
the key group is 1/T 2 . For any reasonable supersaturation, the two-dimensional
nucleation can be reduce to:
ln R 1
' 2 (1.101)
ln S T S
The predicted variation of ln R with ln S is quite large. For example, for KCl-water
if S ' 0.01, ln R/ ln S ' 3, 300 if = 100 erg/cm2 or 200 if = 25 erg/cm2 .
Variations of this magnitude are not observed.
Most models predict a dependence of R on S as either proportional to S or
S 2 . The exponential dependence of the two-dimensional nucleation theories seems
50
GS L2 e2 d2 9.5e d2 R T Hf
, S1 = , yS = DS deads , ln S = T eq
p
max = dT
4a kB T ln S kB T yS Rg T 2
where D is the diffusion coefficient (sub-s surface),
Vb is the molar volume of the crystal,
Nav is Avogadros number,
Ceq is the equilibrium concentration [mole/volume],
Hf is the heat of fusion,
is the Damkohler number,
kH is the thermal conductivity,
a is the area shape factor for surface nuclei,
L is the length shape factor for the nuclei,
is the ratio of the surface to the bulk supersaturation ratio,
ys is the mean distance traveled by solute molecules at the surface,
ys2 = Ds deads where deads = the time for de-adsorption of the solute
molecule from the surface,
e is the edge energy per unit length, h i
nse is the surface equilibrium concentration gm/cm2 ,
is the crystal density,
Teq is the equilibrium freezing temperature.
51
incorrect. However, it also does not seem possible to delineate a particular theory
if only isothermal rate data as a function of supersaturation are available.
Fluid velocities relative to the crystal affect growth rates only when mass trans-
fer controls. At low relative velocities, the rate is independent of velocity. At high
relative velocities, R U 1/2 where U is the relative velocity between the crystal
and the bulk (length/time).
Effect of Temperature:
dn n
G = (1.103)
dL
(nG) n
= (1.104)
L
53
which, when combined with equation 1.102 and integrated, equating with 1/G0 ,
gives
1 (1 + L)1b
b
n = n0 (1 L) exp (1.105)
1b
For the case b = 0, equation 1.105 reduces to equation 1.15 and the population
density plot is linear (Figure 1.3). For b > 0 the plot begins to exhibit curvature
which becomes very pronounced at values of b > 0.5. Considerable care must be
taken, however, when attempting to interpret curvature in population density plots
based on experimental data since it is impossible to distinguish the independent ef-
fects of growth rate dispersion and size-dependent growth. Furthermore, curvature
at the extremities of a population density plot can indicate a departure from ideal
CMSMPR behavior. These and other aspects of the problem have been discussed
in some detail by Jan
ci
c and Grootscholten [63] and Randolph and Larson [2].
are held in place by electronic forces. The sodium chloride structure is face centered
cubic and the unit cell contains four sodium ions and four chloride ions. Because the
unit cells contain two types of atoms some additional constraints on the structure
exit. For example, a symmetry operation on the crystal must superimpose atoms
of the same type.
The external appearance of the crystals is also of great importance and is
referred to as the crystals habit, crystal shape, or crystal morphology. Crystal
habit refers to the external appearance of the crystal. A quantitative description of
a crystal means knowing the crystal faces present, their relative areas, the lengths of
the axes in the three directions, the angles between the faces, and the shape factor
of the crystal. Shape factors are a convenient mathematical way of describing the
geometry of a crystal. If the size of a crystal is defined in terms of a characterization
dimension, L, two shape factors can be defined: the volume shape factor (V ) and
the area shape factor (A ).
V = V L3 (1.106)
A = A L2 (1.107)
Body V A F = A /V
Sphere 0.524 3.142 6.00
Tetrahedron 0.182 2.309 12.7
Octahedron 0.471 3.464 7.35
Six-sided prism 2.60 11.20 4.31
Cube 1.00 6.00 6.00
Prism 10.00 2.00 5.00
Platelet 0.20 2.80 14.00
Needle 10.00 42.00 4.20
a
Data adapted from N`
y vlt et al., 1985 [31].
The habit of crystals obtained from an industrial crystallization process can have
a major impact on a number of important properties relating to the slurry and the
dry product. Crystal habit will affect the rheological properties of the suspension,
the filtration or centrifugation efficiency, the bulk density of the solid, and the
flow properties of the solid. The control of crystal habit (along with crystal size
distribution) is, therefore, an important part of industrial crystallization processes.
Crystal habit can vary dramatically with the rate of crystal growth and nu-
cleation. Also changes in the solvent used or the presence of an impurity can
profoundly affect the crystal habit. Figure 1.20 demonstrates the effect of urea on
sodium chloride crystals.
(a) (b)
Figure 1.20. Sodium chloride crystal grown from pure solution and from a solution
containing 10% urea.
Figure 1.21. (a) Temperature distribution for solidification when heat is con-
ducted into the liquid. Isotherms (b) for a planar S/L interface, and (c) for a
protrusion [66].
where K is the thermal conductivity, T 0 is the temperature gradient (dT /dx), the
subscripts S and L stand for solid and liquid, is the rate of growth of the solid,
and Lv is the latent heat of fusion per unit volume. If a protrusion forms on the
solid the negative temperature gradient in the liquid becomes even more negative.
Therefore heat is removed more effectively from the tip of the protrusion than from
the surrounding regions allowing it to grow preferentially. A solid/liquid interface
advancing into supercooled liquid is thus inherently unstable.
Heat flow into the liquid arise if the liquid is supercooled below crystallization
temperature (Tc ). Such a situation can arise at the beginning of crystallization if
57
nucleation occurs at impurity particles in the bulk of the liquid. Since a certain
supercooling is required before nucleation can occur, the first solid particles will
grow into supercooled liquid and the latent heat of solidification will be conducted
away into the liquid. An originally spherical solid particle will therefore develop
arms in many directions as shown in Figure 1.22. It is found experimentally that
Figure 1.22. The development of thermal dendrites. (a) a spherical nucleus, (b)
the interface becomes unstable, and (c) primary arms develop in crystallographic
directions [67].
the primary arms are always in certain crystallographic directions, each of which
is the axis of a pyramid whose sides are the most closely packed planes with which
a pyramid can be formed: e.g., < 100 > in cubic materials, and < 1100 > in hcp
materials [68]. Dendritic growth only continues until the heat of fusion removes
the constitutional supercooling. When this happens, freezing is completed by the
filling in of the spaces between the dendrite arms.
where the aggregation kernel, or kinetic collision frequency for different character-
istic conditions are given in Tables 1.5 and 1.6. (ai , aj , s) is a measure of the
frequency with which particles of size ai and aj collide, adhere, and form stable
aggregates, s.
speed impeller of the pump or the stirrer. Attrition occurs also as the result of
collisions of crystals with each other and with the wall. When the impact energy
exceeds the crystal strength, the consequence is the fracture of the crystals. Thus
the crystal geometry changes substantially and it seldom grows to its geometric
shape. Ang and Mullin [71] measured the volume shape factor of nickel ammonium
sulphate crystals and found a value of 0.58 0.02. In an industrial crystallizer a
wide variation in shape factor may be expected. This impact may occur with crystal
faces, edges, and corners. Because of the fluctuating turbulence in the liquid phase,
the particles are almost randomly oriented in the vicinity of the impeller. The
contacts with corners are therefore most likely to result in high local stress due
to the small area of contact. Hence repeated contacts would result in rounded-off
crystals of reduced size, provided that there is no competing mechanism of crystal
growth.
The shape factor of the crystals may vary widely depending on the attrition
process and the nature of the substance [72]. The detailed procedure of generating
random samples of volume shape factor from the normal density function is given
by Dey and Sen Gupta [73].
Particle breakage on a mass basis
Z
BB DB = S(l, s)n(l, t) S(l, s)P (l, x, s)n(x, t)dx (1.110)
l
where breakage rate term, S(l, s), is given in Table 1.7 and primary progeny
function, P (x, y) is given in Table 1.8 with different theories and references.
60
Reference P (x, y)
1 2
x1 x1
Prasher [77] j + (1 j )
y y
x
1 exp
y
Broadbent and Callcott [78]
1 exp(1)
y
Klimpel and Austin [79] P1 > 0, not normalized
x
EXPERIMENTAL DESIGN
which should result in the precipitation of NaCl particles out from the solution
leaving small amount of Na+ in aqueous solution.
63
1. The vessel must be small enough so that minimum feed is required, but
large enough that sampling does not result in an appreciable disturbance to
the system.
2. The crystallizer should be designed so that both the suspension and mother
liquor are well mixed, i.e., the liquor composition and the crystal size
distribution must be uniform throughout the volume of the crystallizer.
This perfect mixing condition should be accomplished with a minimum
power input so as to minimize particle breakage.
vessel. PolyscienceT M digital temperature controller and bath were used to cool
or heat the solution, making it possible to keep the temperature constant within
0.1 C. The production rate will be established once the choice of the mean
residence time, , and solution concentrations, before and after crystallization, are
determined. The choice of the mean residence time, , will be made based on
reasonable run times when steady state is imposed. Also one needs to keep in
mined that is a function of the specific reaction kinetics associated with each
process. Randolph and Larson [20] have shown that longer residence time can
often be detrimental to the resultant particle size distribution.
The crystallized material can be conserved by recycling the product to a redis-
solving storage container maintained at 70 C and recycling it to the crystallizer.
When this is done over long periods, care must be taken to ensure that evaporation
does not take place, that no crystals remain undissolved, and that extraneous
impurities do not build up in the feed. However, product recycle to reconstitute
the feed may often be advantageous or necessary for high value products.
The feed concentration will be measured beforehand for each of these experi-
ments since the resultant CSD will depend on the level of supersaturation at certain
temperature. Also care must be taken so that no crystallization takes place before
the feed enters the crystallizer. This could be checked with measuring temperature
at the end-tip of feed-point inside the crystallizer.
The geometries of the laboratory scale crystallizer with working volume (1.4 `)
are shown in Figure A.1 through A.3 in Appendix A. These figures present the
strategic placement of all the important flow lines and baffles, also the position
of impeller. The feed is injected 4/5 of the way down the crystallizer to aid well
mixing before going out to the outlet at the top surface. The ratio of liquid level
to the crystallizer diameter (L/D) of the vessel in Figure A.2 shows 1.27 which is
within the general range of 1.0 to 1.5 for most industrial precipitator [20, 86].
To reduce the energy input to the system while maintaining uniformity, a
standard baffle design [87] is used. This device induces turbulence at a lower
energy input while decreasing the nucleation zone resulting in narrower nuclei
size distribution. The use of standard baffle consisting of four flat vertical plates,
radially-directed (i.e., normal to the vessel wall), spaced at 90 around the vessel
periphery running the length of the vessels straight side. Standard baffle width
is 1/10 or 1/12 of the vessel diameter (D/10 or D/12). The gaps with the vessel
wall and base are left to allow the flow to clean the baffles. Recommended gaps are
equal to 1/72 of the vessel diameter (D/72) between the baffles and the vessel wall,
and 1/4 to one full baffle width between the bottom of the baffles and the vessel
base. More of the detailed baffling information could be found in Myers et al. [88].
The dimension of standard baffle (1/12 width of the vessel diameter) and vessel is
given in Table A.1 along with the dimension of mixer in Appendix A. Stirring was
achieved by a 6-bladed flat disc turbine (Rushton type) or a 6-bladed pitch blade
impeller.
the computer-interface by one simple click and modification. Only thing that is not
controllable is the temperature for storage container which is maintained constant
at 70 C with the aid of drum heater (see more details in Figure C.1 for an operating
flow and control diagram and in Table C.2 for equipment list and specification of a
CMSMPR crystallization system in Appendix C).
The feeding and withdrawing of solution and product from the crystallizer,
respectively, are established by peristaltic pumps controlled by OPTO 22T M . Flow
rate was obtained by the careful calibration of the speed of pump and tube size,
therefore ensuring the desired feed and withdrawal rates as well as a constant
residence time. To ensure constant solution level (within 1.5 mm) inside the
reactor, a Small Tank Alphasonic Level TransmitterT M was used to control the
outlet pump speed while maintaining constant feed flow rate.
Also, the OPTO 22 unit system measures the temperature (solution feed and
outlet point, solution at the end-tip of feed, solution in the tank, coolant in and
outlet of the jacked vessel, coolant in circulator, water bath, and source solution),
pH, conductivity, ionic concentration, mixing speed (rpm), flow rate, and solution
67
level inside the tank on-line in this system. All of these measurements were used
to monitor the reaction and the steady state condition inside the crystallizer.
The calibration for all of these systems was done before each run to increase the
measurement accuracy, see equipment calibration procedures in Appendix C.
a
H: Vessel Height.
b
D: Vessel Diameter.
c
The feed solution concentration, C01 = CaCl2 (47.631.76 wt%), KCl (1.070.16
wt%), NaCl (0.670.08 wt%) with 95% confidence level was obtained by AAS.
d
6-bladed flat disc turbine (Rushton type).
68
precipitation of NaCl was established after the completion of trials, see Appendix C
for more details of experimental setup and operating procedures.
solution concentration of C02 = CaCl2 (46.88 wt%), KCl (0.93 wt%), NaCl (0.50
wt%) as compared to C01 = CaCl2 (47.63 wt%), KCl (1.07 wt%), NaCl (0.67 wt%).
All other experimental conditions are same as stated in each table. The comparison
of design in Table 2.4 with respect to Table 2.3 is to see the mechanical effect of
different type of mixer (turbine and impeller) on the resultant structure of particles
and CSD. Also from the comparison among Tables 2.2, 2.3, and 2.4 one can easily
have an access to see the every possible effects of three variables (TC , Q, and MS )
in addition to the effect of solution concentrations (C01 , C02 , and C03 ) and the
different type of agitator (turbine and impeller) on the resultant particle size and
shape as a result of different nucleation and crystal growth kinetics.
A total of 81 experiments were performed with some experiments taking as much
as 12 hours to run from the setup to the completion. All the rpms chosen give well
mixed behavior as determined by the residence time distribution measurements, see
Section 4.1. For the impeller, higher rpms than those for the turbine are needed to
give well-mixed behavior.
concentration will be ignored since the starting CSD for this experiment will be
known from a CMSMPR crystallization which will be performed ahead of each
breakage and aggregation experiment as mentioned earlier.
72
Experimental Condition
Test No. TC b Qc MS d
( C) (ml/min) (rpm)
T 01 25 20 300
T 02 25 20 400
T 03 25 20 500
T 04 25 40 300
T 05 25 40 400
T 06 25 40 500
T 07 25 60 300
T 08 25 60 400
T 09 25 60 500
T 10 30 20 300
T 11 30 20 400
T 12 30 20 500
T 13 30 40 300
T 14 30 40 400
T 15 30 40 500
T 16 30 60 300
T 17 30 60 400
T 18 30 60 500
T 19 35 20 300
T 20 35 20 400
T 21 35 20 500
T 22 35 40 300
T 23 35 40 400
T 24 35 40 500
T 25 35 60 300
T 26 35 60 400
T 27 35 60 500
a
Constant experimental parameters:
feed solution concentration (C01 ) =
CaCl2 (47.631.76 wt%), KCl (1.070.16
wt%), NaCl (0.670.08 wt%) with 95%
confidence level, feed solution tempera-
ture (TS ) = 70 C, mixing design = stan-
dard baffle + turbine.
b
TC : Crystallization temperature
c
Q: Flow rate
d
MS : Mixing speed
73
Experimental Condition
Test No. TC b Qc MS d
( C) (ml/min) (rpm)
TS 01 25 20 300
TS 02 25 20 400
TS 03 25 20 500
TS 04 25 40 300
TS 05 25 40 400
TS 06 25 40 500
TS 07 25 60 300
TS 08 25 60 400
TS 09 25 60 500
TS 10 30 20 300
TS 11 30 20 400
TS 12 30 20 500
TS 13 30 40 300
TS 14 30 40 400
TS 15 30 40 500
TS 16 30 60 300
TS 17 30 60 400
TS 18 30 60 500
TS 19 35 20 300
TS 20 35 20 400
TS 21 35 20 500
TS 22 35 40 300
TS 23 35 40 400
TS 24 35 40 500
TS 25 35 60 300
TS 26 35 60 400
TS 27 35 60 500
a
Constant experimental parameters:
feed solution concentration (C02 ) =
CaCl2 (46.881.73 wt%), KCl (0.930.15
wt%), NaCl (0.500.05 wt%) with 95%
confidence level, feed solution tempera-
ture (TS ) = 70 C, mixing design = stan-
dard baffle + turbine
b
TC : Crystallization temperature
c
Q: Flow rate
d
MS : Mixing speed
74
Experimental Condition
Test No. TC b Qc MS d
( C) (ml/min) (rpm)
IS 01 25 20 500
IS 02 25 20 700
IS 03 25 20 900
IS 04 25 40 500
IS 05 25 40 700
IS 06 25 40 900
IS 07 25 60 500
IS 08 25 60 700
IS 09 25 60 900
IS 10 30 20 500
IS 11 30 20 700
IS 12 30 20 900
IS 13 30 40 500
IS 14 30 40 700
IS 15 30 40 900
IS 16 30 60 500
IS 17 30 60 700
IS 18 30 60 900
IS 19 35 20 500
IS 20 35 20 700
IS 21 35 20 900
IS 22 35 40 500
IS 23 35 40 700
IS 24 35 40 900
IS 25 35 60 500
IS 26 35 60 700
IS 27 35 60 900
a
Constant experimental parameters:
feed solution concentration, (C03 ) =
CaCl2 (47.090.725 wt%), KCl
(1.030.04 wt%), NaCl (0.440.034
wt%) with 95% confidence level, feed
solution temperature (TS ) = 70 C,
mixing design = standard baffle +
impeller.
b
TC : Crystallization temperature
c
Q: Flow rate
d
MS : Mixing speed
75
Experimental Experimental
Condition Condition
Test No. Test No.
RT c MS d RT MS
( C) (rpm) ( C) (rpm)
BT 01 25 300 BI 01 25 500
BT 02 25 400 BI 02 25 700
BT 03 25 500 BI 03 25 900
BT 04 30 300 BI 04 30 500
BT 05 30 400 BI 05 30 700
BT 06 30 500 BI 06 30 900
BT 07 35 300 BI 07 35 500
BT 08 35 400 BI 08 35 700
BT 09 35 500 BI 09 35 900
a
Constant experimental parameters: Starting solution con-
centration (C01 ) = CaCl2 (47.631.76 wt%), KCl (1.070.16
wt%), NaCl (0.670.08 wt%) with 95% confidence level, stan-
dard baffle + turbine configuration.
b
Constant experimental parameters: Starting solution con-
centration (C03 ) = CaCl2 (47.091.76 wt%), KCl (1.030.16
wt%), NaCl (0.440.08 wt%) with 95% confidence level, stan-
dard baffle + impeller configuration.
c
RT : Reactor temperature
d
MS : Mixing speed
CHAPTER 3
EXPERIMENTAL METHODS
3.1.1 Density
Density is a significant part of measurement and instrumentation. Density mea-
surements are made for at least two important reasons: (1) for the determination
of mass and volume of products, and (2) the quality of the product. In many
industrial applications, density measurement ascertains the value of the product.
The density of the solution is often needed for mass balance, flow rate, and product
yield calculations.
Pycnometers are static devices. A fixed volume vessel (10 ml) is used and be
filled with the sample liquid. The density of the fluid is measured by weighing the
sample. The simplest version consisting a vessel in the shape of a bottle with a long
stopper containing a capillary hole is shown in Figure 3.1. The capillary is used to
determine the exact volume of the liquid, thus giving high resolution when filling
the pycnometer. The bottle is first weighed empty, and then with distilled-aerated
water to determine the volume of the bottle. The bottle is then filled with the
process fluid and weighed again. The density is determined by dividing the mass
by the volume. The specific gravity of the liquid is found by the ratio of the fluid
mass to water mass.
77
Figure 3.1. A pycnometer with fixed volume (10 ml) is filled with liquid and
weighted accurately; capillary is used to determine the exact volume of the liquid.
When pycnometers are used, for good precision, ultimate care must be exercised
during the measurements; that is, the bottle must be cleaned after each measure-
ment, the temperature must be kept constant, and precision balances must be used.
Commonly accepted sets of conditions are normal temperature and pressure (NTP)
and standard temperature and pressure (STP).
The density measurement for the solution containing the mixture of CaCl2 , KCl,
and NaCl was performed by first taking the solution sample at 70 C which was
heated more than 24 hours and then the solution sample was diluted with distilled
water (1:1 by vol%) since the temperature at which the density measurement of
solution should be made causes particles formation. After the measurement at
20 C with the aid of precision balance, the mass and volume of water added were
subtracted to get the solution density of interest.
3.1.2 Viscosity
The design of any equipment that involves the flow or stirring of liquids re-
quires a knowledge of the fluids viscosity. Since crystallization operations involve
the stirring and movement of suspensions of particles in fluids, the viscosity of
suspensions is important in crystallization design and operation. Viscosity is a
property of particular material defined as the ratio of the shear stress and the shear
78
measurement accuracy within the range of 0.45 cP, i.e., 1% error, which confirms
the equipment accuracy. However, the measurement of solution of interest with
varying temperature will precipitate particles at lower temperature thus lowering
measurement accuracy. The solution sample was taken at 70 C from the source
tank the same way as solution sampling for the density measurement except for
dilution.
The SI physical unit of dynamic viscosity is the Pascal-second (Pas), which is
identical to 1 Ns/m2 or 1 kg/ms. The cgs physical unit for dynamic viscosity is
the poise (P). It is more commonly expressed, particularly in ASTM standards, as
centipoise (cP).
element is measurable at a time. AAS requires that the analytic atoms be in the
gas phase. Ions or atoms in a sample must undergo desolvation and vaporization
in a high-temperature source such as a flame or graphite furnace. Flame AAS can
only analyze solutions, while graphite furnace AAS can accept solutions, slurries, or
solid samples. Flame AAS uses a slot type burner to increase the path length, and
therefore to increase the total absorbance. Sample solutions are usually aspirated
with the gas flow into a nebulizing/mixing chamber to form small droplets before
entering the flame. AA spectrometers use monochromators and detectors for uv and
visible light. The main purpose of the monochromator is to isolate the absorption
line from background light due to interferences. Photomultiplier tubes are the most
common detectors for AA spectroscopy.
detectable, in one shot into the feed stream entering the reactor at some time
t = 0 and then measuring the tracer concentration, C, in the outlet stream as
a function of time. On-line analyzers detected each of the species in the tracer;
blue dye with the aid of a video camera and UV-vis spectrometer, temperature
with a thermocouple, salt with conductivity, and H+ concentration with pH meter
were measured simultaneously with various conditions such as different flow rates
and mixer rpm values. Great care was used in synchronizing the pulse input with
the initiation of data accumulation for RTD analysis and to calibrate the pumps
to ensure that the reactor space-time [= Vtank /Q] was accurately measured. The
reactor control system has the capability of controlling the feed source and reactor
temperature to 0.25 C, the feed and product flow rate to 0.5 ml/min, liquid
level to 3.9 mm and mixing speed to 3 rpm.
For detailed RTD measurement method along with results and simulations refer
to Appendix B. The configuration of reactor with turbine and impeller are shown in
Figure A.2 through A.3. Also for the details of dimension of reactor configuration
and type of mixer see Table A.1 in Appendix A.
1. Source Temperature, TS = 70 C.
2. Crystallization Temperature, TC = 25 35 C.
(a)
(b)
Figure 3.3. (a) Dominant aqueous species and (b) precipitation point for the
dominant solid species crystallizing out from an aqueous solution (100 g) con-
taining CaCl2 (47.63 wt%), KCl (1.07 wt%) and NaCl (0.67 wt%) as a function
of temperature at 1 atm pressure [Calculated and plotted using Stream Analyzer
(OLI Systems, Inc.)].
84
a
Feed solution concentration (C01 ) = CaCl2
(47.631.76 wt%), KCl (1.070.16 wt%), NaCl
(0.670.08 wt%) with 95% confidence level, feed so-
lution temperature (TS ) = 70 C, and total solution
mass = 100 g. The calculation was performed at 1
atm pressure.
t
SS = 1 exp 100 (3.1)
For the non-ideality of a particular system, each reactor performance should be
measured. In order to characterize steady state for our system, pH, conductivity,
and ionic concentration as well as temperature are monitored with time. Figure 3.4
shows a result obtained with a crystallization experiment at 34.9 C, 40 ml/min,
and 380.5 rpm. From the figure it is possible to determine a reasonable steady
state condition for the system. The relationship for steady state given in Equa-
tion 3.1 predicts 99.62% of steady state at 5.6 . The time t was set to 0 when the
crystallization temperature reaches 35 C in this case.
3.3.3 Sampling
For continuous processes, the output flow rate must be equal to the input flow
rate such that no mass accumulates in the reactor with time. Allen [94] reviewed
86
P = weight of product,
of solvent,
Wo = weight of solute in feed,
E = evaporation.
the industrial level (> 100 kg/m3 ). Large working volumes, however, can present
problems with the maintenance of good mixing throughout the vessel and the need
to store large volumes of solution at the required feedstock entry conditions. Since
steady-state conditions are not normally achieved in much less than ten residence
times, even the minimum recommended sized vessel would need 30 ` of solution
to be stored to enable meaningful runs to be carried out.
After reaching steady state, which usually obtained after a period of 8 to 15
residence times, product can be used to determine the crystal size distribution and
kinetic parameters, crystal growth rate, and nucleation rate rate. There should
be no crystal size classification at the discharge and sampling points. Every effort
should be made to sample isokinetically, i.e., with the sample withdrawal velocity
equal to that of the solution flowing at the point of removal. For crystallization
vessels operating with solution densities around industrial values it is helpful, for
design and scale-up purposes, if measurements are made of the power input to
the stirrer. Once all the kinetic data measurements are made with precision for
every set of crystallization and batch breakage and aggregation experiments stated
in Tables 2.2, 2.3, 2.4, and 2.5, its utilization and characterization are attempted
with the application of well-known theoretical approach for better understand of
crystallization kinetics.
Figure 3.5. Schematic of the PIDS assembly in the Beckman Coulter LS230
particle size analyzer.
tion that differentiates between sub-micron particle sizes and dramatically increases
resolution.
Although PIDS uses a second light source split into flavors, the scattering of these
light beams by particles is described by the same Mie theory as laser scattering,
so all scattering information is converted to particle size using the same algorithm
in a single operation. The performance of the Beckman Coulter LS230 Particle
Size Analyzer relative to traceable standards indicates that the instrument can be
used with confidence, and that the particle size distributions it produces accurately
reflect the sample material.
The theories to calculate particle size based on the Fraunhofer and Mie light
scattering are beyond the scope of this research but the principle is; when a particle
is illuminated by a laser light source, scattering from this beam occurs. The pattern
resulting from light scattering from a suspension of solid particles can be interpreted
to produce a size distribution of the particles present. High suspensions cannot
be handled due to multiple scatter, that is, light scattering off of more than one
particle surface. Multiple scatter is a function of particle number, and thus the
allowable slurry density (above which multiple scatter invalidates the measurement)
91
3.4.3.1 XRD
To find out the crystallographic information of NaCl crystals, X-ray diffraction
(XRD) of dried particles was performed. Below is the brief description of X-ray
diffraction principle.
When X-ray radiation passes through matter, the radiation interacts with the
electrons in the atoms, resulting in scattering of the radiation. If the atoms are
organized in planes (i.e., the matter is crystalline) and the distances between the
atoms are of the same magnitude as the wavelength of the X-rays, constructive
and destructive interference will occur. This results in diffraction where X-rays are
emitted at characteristic angles based on the spaces between the atoms organized
in crystalline structures called planes. Most crystals can have many sets of planes
passed through their atoms. Each set of planes has a specific interplanar distance
and will give rise to a characteristic angle of diffracted X-rays. The relationship
between wavelength, atomic spacing (d) and angle was solved as the Bragg Equa-
tion, see Figure 3.6. If the illuminating wavelengh is known (depends on the type
of X-ray tube used and if a monochromator is employed) and the angle can be
measured (with a diffractometer) then the interplanar distance can be calculated
from the Bragg equation. A set of d-spaces obtained from a single compound will
be represent the set of planes that can be passed through the atoms and can be
used for comparison with sets of d-spaces obtained from standard compounds.
allow for particle stabilization against aggregation were ones that also dissolved the
particles. Those liquids that were inert (e.g., no dissolution) did not disperse the
particles well enough for PSD analysis.
Alternatively, a slurry sample ( 10 ml) was taken directly from the reactor
to the sample bottle with vacuum pump in a very short time (< 1 s) for a PSD
measurement. The temperature was kept constant during sampling to avoid any
change in the crystals. This sample was diluted in the LS230 to lower the number
density of particles so that light scattering measurement could be made without
multiple scattering. A saturated solution at the temperature of the crystallizer was
used for dilution. It was prepared by filtering the product of the same crystallization
experiment with a 0.2 m Supor-200 GelmanScienceT M filter making sure that
the temperature remained constant. Thus the temperature of the solution and
suspension for sampling, diluting, and measuring of sample was kept constant at
all times at 30 C, the same temperature as the slurry in the crystallizer. This is
feasible for all of the steps except measurement in the LS230 where the sample
is heated by 2 to 5 C by the light source and mixer inside the sample chamber.
Due to this heating the measurement of the PSD was not stable over a 10-minute
period as is shown in Figure F.3 see Section F for more details. The particles tend
to dissolve slowly thus requiring a special technique to make the particles stable
and the PSD measurements successful. This will be discussed later.
93
4.2.1 Density
The feed solution sample was taken from the solution in preheating solution
container at 730.4 C and the solution after crystallization was a filtrated solution
from the crystallizer outlet. Due to the precipitation of solution at 20 C, the
solution was first diluted with water (1:1 by vol%) and measurements were done at
constant temperature for more than five times to increase accuracy. With a simple
calculation the desired solution density was obtained and given in Table 4.1. Only
one sample was chosen for each set of experimental design from Tables 2.2, 2.3,
and 2.4, simply to see the difference in solution density due to the crystallization of
NaCl. The results clearly showed the loss of solution weight after the crystallization.
This result is verified with other analyses, e.g., AAS, EDS, and XRD. For the
properties of stock solution, i.e., element concentration, pH, and density, as well as,
viscosity, refer to Section C.2 in Appendix C. Although, the density measurement
here was to show the difference in solution before and after crystallization, the
density of feed solution requires special attention since it will be used in Section 5.2
for the interpretation of crystallization process.
96
Crystallization
Test No.b
Before After
T 14 1.479 1.474
Densityc
TS 14 1.475 1.470
(g/cm3 )
IS 14 1.476 1.471
T 14 1.480 1.475
Specific Gravity
TS 14 1.476 1.471
of Liquid
IS 14 1.477 1.472
a
As described in Section 3.1.1
b
For detailed experimental conditions of the test
number, see Table 2.2, 2.3, and 2.4
c
The uncertainties of all density measurements
were within 0.003 with confidence level of 95%.
Figure 4.1. Viscosity of the solution with varying temperature, allowable error =
5%. Analysis of solution concentration (C01 ) is given in Table C.1, see Appendix C.
a
The uncertainties in concentration of CaCl2 , KCl, and NaCl were
1.73, 0.15, and 0.05, respectively, with confidence level of 95% for
the AAS analysis.
b
The crystallization condition for TS 13 is given in Table 4.5, see
Section 4.3.3.
the results of that study and discusses the implications of these results on the
crystallization process based on the proof from the analytical experiments and later
in Chapter 5, more details with the theoretical application and model validation
will be discussed.
in Figure 3.3 (see Section 3.3.1) showed that as the temperature is cooled from
70 C to 30 C increasing amount of NaCl crystallizes out of solution.
The results of experiments show that indeed there was increasing amount of
crystals as the solution is cooled as shown in Figure 4.2. These particles are
end of experiment for the run, i.e., typically 10 times of mean residence time (10 ).
After the completion of trials for crystallization obtaining necessary conditions
and modifications of experimental set-up, all the crystallization experiments were
performed following statistically designed experimental conditions in Tables 2.2, 2.2,
and 2.2 (see Section 2.7).
a
All measured values were taken and averaged from the
time when the crystallization temperature reached steady
state to the end of crystallization experiment. The uncer-
tainty was obtained with confidence level of 95%.
b
Reaction pH was measured during the crystallization
at the crystallization temperature (TC ) with a pH probe
temperature compensation.
c
The value for the solution level is the distance between
solution level indicator and level of solution.
inside feed tube should be avoided since the difference in crystallization environ-
ment, e.g., different solution mixing pattern (laminar and turbulent) and different
solution temperature (level of supersaturation), will make enough difference in crys-
tallization phenomena between the feed tube and the crystallizer. Tables 4.5 and 4.6
are the results of experimental conditions from CMSMPR crystallization which was
designed from Tables 2.3 and 2.4, respectively. The experimental condition for IS
104
a
Constant experimental parameters: feed solution concentration (C01 ) = CaCl2
(47.631.76 wt%), KCl (1.070.16 wt%), NaCl (0.670.08 wt%) with 95% con-
fidence level, feed solution temperature (TS ) = 70 C or above, mixing design =
standard baffle + turbine.
b
TS : Source solution temperature.
c
Tcoolin : Coolant inlet temperature to the tank jacket.
d
Tcoolout : Coolant outlet temperature from the tank jacket.
e
TC : Crystallization temperature.
f
Q: Flow rate.
g
MS : Mixing speed.
105
a
Constant experimental parameters: feed solution concentration (C02 ) = CaCl2
(46.881.73 wt%), KCl (0.930.15 wt%), NaCl (0.500.05 wt%), feed solution temperature
(TS ) = 70 C or above, mixing design = standard baffle + turbine.
b
TS : Source solution temperature.
c
TS1 : feed solution temperature at the inlet end.
d
Tcoolin : Coolant inlet temperature to the tank jacket.
e
Tcoolout : Coolant outlet temperature from the tank jacket.
f
TC : Crystallization temperature.
g
Q: Flow rate.
h
MS : Mixing speed.
106
a
Constant experimental parameters: feed solution concentration (C03 ) = CaCl2
(47.090.725 wt%), KCl (1.030.04 wt%), NaCl (0.440.034 wt%), feed solution tempera-
ture (Ts ) = 70 C or above, mixing design = standard baffle + impeller.
b
TS : Source solution temperature.
c
TS1 : Feed solution temperature at the inlet end.
d
Tcoolin : Coolant inlet temperature to the tank jacket.
e
Tcoolout : Coolant outlet temperature from the tank jacket.
f
TC : Crystallization temperature.
g
Q: Flow rate.
h
MS : Mixing speed.
107
Figure 4.4. EDS result for the comparison of NaCl particles after the crystalliza-
tion to the powder produced by evaporation of solution. The crystallization was
done at 25 C, 40 ml/min, and 283 rpm.
108
Element analysis
Element Dried Powdera NaCl Particlesb
(wt%) (at%) (wt%) (at%)
Na 0.97 1.55 37.29 47.86
Cl 65.52 67.77 62.09 51.68
K 1.26 1.18 0 0
Ca 32.25 29.51 0.63 0.46
a
Dried powder was prepared by evaporation of
solution.
b
NaCl particles were produced by the crystalliza-
tion at 25 C, 40 ml/min, and 283 rpm in a CMSMPR
crystallizer.
to be residual from the filtration. These experimental results are in rather good
agreement with the AAS analysis of solution when the experimental error such
as loosing CaCl2 during filtration and analytical errors in AAS and ICP analysis
are considered. However, there was a discrepancy in concentration of CaCl2 when
compared to the precipitation analysis (shown in Table 3.1 in Section 3.3.1) which
109
showed there should be CaCl2 6H2 O particles crystalizing out starting from at 28 C
as the solution cools down to 25 C. This might be the reason arising from the
difference between an ideal thermodynamic system and our system that is close to
ideal mixing but not the same as that in an ideal system.
(a) (b)
(c) (d)
(e) (f)
Figure 4.5. Microscopic photos taken (a) and (b) at 2 , (c) and (d) at 5 , and
(e) and (f) at 10 for T 05 sample. Note: The scale of 10 in figure equals 130 m.
Figure 4.6. NaCl particles showing the crystal habit and size (SEM).
size. The spherical bodies with molecules or solvated ions are believed to be in a
state not too different from that in the bulk fluid, with no clearly defined surface.
It also shows chains formed initially and eventually, as the crystallization precedes,
112
KL TL0 KL T
= ' (4.1)
LV LV r
Thus for a given T , rapid growth will be favored by small values of r due to the
increasing effectiveness of heat conduction as r diminishes. However T is not
independent of r. As a result of the GibbsThomson effect equilibrium across a
curved interface occurs at an undercooling Tr below Tm given by
2Tm
Tr = (4.2)
LV r
The minimum possible radius of curvature of the tip is when Tr equals the total
undercooling T0 = Tm T . This is just the critical nucleus radius r given
114
particles are forming, growing, and evolving in a CMSMPR crystallizer with varying
experimental conditions with time. With many trials and with the application
of CdCl2 impurity, stabilizing NaCl particles in a saturated solution during CSD
measurement was successful within the equipments allowable error, i.e., 1% re-
producibility, see Appendix F for more details. Also, the sampling and the CSD
measurement in timely manner should be emphasized the same as the importance
in the crystallization experiment and the CSD measurement themselves. If any
of the processes during a crystallization experiment fails to meet the procedure
requirements (see Appendix C for the procedures of crystallization, sampling, and
CSD measurement), the whole experiment was discarded and reproduced again
following all the requirements.
Figures 4.11 and 4.12 give CSDs measured at 1.5 and 10.5 , respectively, by
using a LS230 PCSA. With these CSDs measured at every mean residence time ( ),
Figure 4.11. Differential and cumulative volume % of CSD at 1.5 for T 05 sample
crystallization temperature = 250.16 C, feed flow rate = 400.91 ml/min,
mixing speed = 3826.14 rpm, and crystallizer volume = 14009.4 ml. The mass
of particles was 2.6 g/` at 1 . The uncertainty was obtained with confidence level
of 95%.
116
the 3D plot of CSD was constructed to illustrate the evolution of CSDs with time
in a CMSMPR crystallizer. The particle size distributions (x-axis) at the outlet of
the crystallizer as a function of mean residence time (y-axis) and particle volume %
(z-axis) after the crystallizer contents reaches the operating temperature are shown
in Figures 4.13 and 4.14. The 3D plot of CSDs in Figure 4.14 is shown with the
view from z-direction. Initially the particles have dentritic growth arms, which with
time fill in to give bipyramidal particles, see Figure 4.5. The bipyramidal particles
aggregate as the time increases as well. The 3D plots of CSDs in Figures 4.13
and 4.14 show the particle diameter in m (x-axis) taken at different times measured
in units of the mean residence time, (y-axis), and differential volume % (z-axis).
The crystallization experiment for this particular plot was done at crystallization
temperature = 25 C, feed flow rate = 40 ml/min, and mixing speed = 382 rpm with
crystallizer volume = 1400 ml. The CSD measurements were performed every mean
117
residence time, i.e., (V /Q = 35 min.), for more than 10 times of mean residence time
( ). At times longer than 10 , there was no change in CSDs. Comparing the CSDs
with the microscopic photos in Figure 4.5 and the SEM in Figure 4.6 shows that
the CSDs accurately represent the particle size distribution for the crystallization
process in a CMSMPR crystallization. In Figure 4.13 the bi/tri modal (small
humps below 50 m in size) is an indication of the aggregation of small particles,
i.e., mostly from the secondary nucleation and the broken particles. No change in
CSDs from 8 to 10 is the indication that the steady state has been reached.
With Figure 4.13 and photos in Figure 4.5, it is clear that the crystal morpholo-
gies and size distributions are evolving with time therefore indicating the combined
effects of nucleation, growth, and breakage, as well as aggregation mechanisms
during the crystallization. Particles with size smaller than 5 m were not observed
118
with LS230 even though the instruments size resolution is 0.04 m. Particles
smaller than 5 m in size were also not observed with the microscope even though
it has a size resolution limited to 1.3 m at this magnification. However, the
SEM figure in Figure 4.8 shows particles in size ranging from 2 to 50 m of
spheres and octahedrons. These are dried particles showing the halted process
of crystallization. Some shows agglomerating to larger octahedral particles and
others show aggregating each other. By the comparison of this SEM figure with
microscopic figures and PSD measurements, it can be concluded that the wet
particles smaller than 5 m are not stable due to either redissolution of small
particles into the solution or agglomeration into the larger particles.
For details in the crystallization operation, sampling, and CSD measurement
procedures refer to Section C.6 in Appendix C.
119
(a) (b)
(c) (d)
Figure 4.15. Reproducibility of CSDs. The CSDs were measured and plotted as
a function of mean residence time ( = 70 min.) (a) and (b) for T 10 sample,
and (c) and (d) for T 10-01 sample (replica of T 10 sample) with a 6-bladed flat
disc turbine (Rushton type) at mixing speed = 284.6 and 284.8 rpm, crystallization
temperature = 30.1 and 30.3 C, respectively, with same feed flow rate = 20 ml/min
and crystallizer volume = 1400 ml. Steady state was achieved after 9 . The plots
in (b) and (d) are shown with a view from z-direction. See Table 4.8 for detailed
experimental conditions.
that the difference in CSD also depends on crystallization temperature and feed
flow rate in addition to the concentration difference. Figure 4.16 illustrates one
of the comparisons with same experimental conditions except for the solution
concentration. The figure shows two resultant CSDs measured and plotted as a
function of mean residence time; (a) and (b) for T 10 sample, and (c) and (d) for
TS 10 sample with a 6-bladed flat disc turbine (Rushton type) in the middle of
121
a
Constant experimental parameters: feed solution concentration (C01 ) = CaCl2
(47.631.76 wt%), KCl (1.070.16 wt%), NaCl (0.670.08 wt%) with 95% confidence
level, feed solution temperature (TS ) = 70 C, mixing design = standard baffle +
turbine, and crystallizer volume = 1.4 `.
(a) (b)
(c) (d)
Figure 4.16. Effect of feed solution concentration on CSDs. The CSDs were
measured and plotted as a function of mean residence time ( = 70 min.) (a)
and (b) for T 10 sample, and (c) and (d) for TS 10 sample with a 6-bladed flat
disc turbine (Rushton type) at mixing speed = 284.8 and 281.9 rpm, crystallization
temperature = 30.3 and 29.8 C with same flow rate = 20 ml/min and crystallizer
volume = 1400 ml. Steady state was achieved after 9 . The plots in (b) and (d) are
shown with a view from z-direction. See Table 4.4 and 4.5 for detailed experimental
conditions, also for resultant yield and mean size of particles in Table 4.9 and 4.10.
time due to the breakage of large particles and the aggregation of small particles
resulting in distinct transition of CSDs between 3 6 . Figure 4.17 (c) and (d) at
30.6 C show a moderate change with time and Figure 4.17 (e) and (f) at 35.5 C
shows the least change in size with time. Consulting microscopic photos, some
broken particles were found in (a) and hardly any in (c). Steady state is clearly
evident from 8 to 10 for all Figure 4.17 cases.
(a) (b)
(c) (d)
(e) (f)
(a) (b)
(c) (d)
(e) (f)
Figure 4.18. Flow rate effect on CSDs. The CSDs were measured and plotted
as a function of mean residence time ( ) Flow rate (a) and (b) 20 ml/min (
= 70 min.) (T 11 sample), (c) and (d) 40 ml/min ( = 35 min.) (T 14 sample),
and (e) and (f) 60 ml/min ( = 23.3 min.) (T 17 sample) at same crystallization
temperature = 30 C, mixing speed = 385 rpm, and crystallizer volume =
1400 ml. Steady state was achieved after 8, 9 and 9 , respectfully. The plots in
righthand side are shown with a view from z-direction. See Table 4.4 for detailed
experimental conditions.
127
(a) (b)
(c) (d)
(e) (f)
Figure 4.19. Mixing speed effect on CSDs. The CSDs were measured and plotted
as a function of mean residence time ( = 35 min.) Mixing speed (a) and (b) at
287.7 rpm (T 13), (c) and (d) at 387.1 rpm (T 14), and (e)and (f) at 482.2 rpm (T
15) at same crystallization temperature = 30 C, feed flow rate = 40 ml/min, and
crystallizer volume = 1400 ml. Steady state was achieved after 8 . The plots in
righthand side are shown with a view from z-direction. See Table 4.4 for detailed
experimental conditions.
128
(a) (b)
(c) (d)
Figure 4.20. Effect of different mixer on CSDs. The CSDs were measured and
plotted as a function of mean residence time ( = 23.3 min.) (a) and (b)
with a 6-bladed flat disc turbine (Rushton type) (TS 18) and (c) and (d) with
a 6-bladed pitch blade impeller (IS 16) at mixing speed = 479 and 482 rpm,
crystallization temperature = 30.0 and 30.2 C with same feed flow rate = 60
ml/min and crystallizer volume = 1400 ml. Steady state was achieved after 9 and
8 , respectively. The plots in righthand side are shown with a view from z-direction.
See Table 4.5 and 4.6 for detailed experimental conditions.
129
Figure 4.21. Effect of crystallization temperature (TC ), feed flow rate (Q), and
mixing speed (M S) on the mean size of particles (Lmean ) at steady state (10 ) for
T samples (see Table 4.4).
a
The uncertainties for yield and mean
size of particles were obtained with 95%
confidence level.
b
Yield of NaCl particles were obtained
from the results of AAS analysis.
c
Lmean : Mean Size of Particles (diame-
ter) obtained from the PSD measurements
by LS230.
data was increased with increase in mixing speed. With this regression, the mean
particle size at steady state (10 ) can be predicted within the standard error of
8.587.
131
Figure 4.22. Linear regression statics on the mean size of particles (Lmean ) vs.
mixing speed at steady state (10 ) for T samples (see Table 4.4 and 4.9). The
regression was performed with confidence level of 95%.
Table 4.10 shows the yield and mean particle size at steady state for TS samples.
The yield in TS samples also shows relatively large uncertainty comparable to yield
in T samples and has shown a weak relationship with crystallization temperature
(TC ), flow rate (Q), and mixing speed (M S) as shown in Figure 4.23. This was most
likely the result of difficulties inherent in obtaining accurate liquid compositions.
The sample solution containing solid particles at room temperature requires further
treatments such as making dilution with distilled water and adding 5% HCl. In
addition to the difficulties obtaining precise concentration of aqueous electrolyte so-
lutions, the accuracy of solution concentration was violated during such treatments
for T and TS samples.
The mean size of particles in TS samples shows a strong relationship with
mixing speed (MS) and weak relationships with flow rate (Q) and crystallization
temperature (TC ) as M S > Q > TC shown in Figure 4.24 same as in the case of
T samples which though showed less clear effect of crystallization temperature in
Figure 4.21. The detailed statistical analysis for the effects (main and interaction)
132
a
The uncertainties for yield and mean
size of particles were obtained with 95%
confidence level.
b
Yield of NaCl particles were obtained
from the results of AAS analysis.
c
Lmean : Mean Size of Particles (diame-
ter) obtained from the PSD measurements
by LS230.
Figure 4.24. Effect of temperature, flow rate, and mixing speed on the mean size
of particles (Lmean ) at steady state (10 ) for TS samples (see Table 4.5).
134
Figure 4.25. Linear regression statics on the mean size of particles (Lmean ) vs.
mixing speed at steady state (10 ) for TS samples (see Table 4.5 and 4.10). The
regression was performed with confidence level of 95%.
yield and mean size of particles data obtained for the IS samples will be attempted
for further analysis. Figure 4.26 shows the yield (g/`) of NaCl particles vs. sample
a
The uncertainties for yield and mean
size of particles were obtained with 95%
confidence level.
b
Yield of NaCl particles were obtained
from the results of AAS analysis.
c
Lmean : Mean Size of Particles (diame-
ter) obtained from the PSD measurements
by LS230.
136
number. This figure shows the effects of crystallization temperature and feed flow
rate on the yield of NaCl at 10 . As the feed flow rate and the crystallization
temperature increase the yield of NaCl particles decreased. The effect of mixing
speed on yield was small. Figure 4.27 shows the yield of NaCl vs. feed flow rate
with temperature groupings. The linear regression was poor due to the effects of
crystallization temperature and mixing speed. As mentioned earlier, this analysis
along with the data shown in Figure 4.26 provides justification for the effect of feed
flow rate on the yield of NaCl particles, i.e., the particles having long residence
time (space time in crystallizer) will grow more as a result of particles spending
more time in crystallizer. Figure 4.28 shows the overall effect of crystallization
temperature, feed flow rate, and mixing speed on the mean size of particle (Lmean )
at 10 for IS samples. At 25 C, the experiment with feed flow rate of 20 ml/min
was not successful due to crystallization in the feed before entering the crystallizer.
Figure 4.28 together with Figure 4.29 indicates the effects of all three variables, i.e.,
137
Figure 4.27. Linear regression statics on the product yield (g/cm3 ) vs. flow
rate (ml/min) at steady state (10 ) for IS samples (see Table 4.6 and 4.11). The
regression was performed with confidence level of 95%.
Figure 4.28. Effect of temperature, flow rate, and mixing speed on the mean size
of particles (Lmean ) at steady state (10 ) for IS samples (see Table 4.6).
138
crystallization temperature, flow rate, and mixing speed, on resultant mean size of
particles. Figure 4.29 shows the regression analysis of mean particle size on mixing
Figure 4.29. Linear regression statics on the product mean size of particles (Lmean )
vs. mixing speed (rpm) at steady state (10 ) for IS samples (see Table 4.6 and 4.11).
The regression was performed with confidence level of 95%.
speed for IS samples. As shown in these two Figures 4.28 and 4.29, the particle
size decreases with increase in mixing speed which showed the most profound effect
on resultant particles, and increases with increase in crystallization temperature
within the experimental conditions, i.e., 25 35 C. The effect of feed flow rate
shows somehow interesting results in this system; the particle size increases with
increase in flow rate from 20 to 40 ml/min and decreases with further increase to
60 ml/min. This feed flow rate effect was large at lower mixing speed but small at
higher mixing. This was the indication showing the dominant effect of mixing on
resultant particles at higher mixing reducing other effects.
A correlation analysis was performed to measure the relationship between two
data sets for all three experimental designs. The two data sets are independent
of the unit of measurement in this analysis; a variable and a product or between
139
two products. The population correlation calculation returns the covariance of two
data sets divided by the product of their standard deviations based on the following
formulas.
cov(X, Y )
X,Y = (4.4)
X Y
where
1
x2 = (Xi x )2
n
1
y2 = (Yi y )2
n
This analysis is to determine whether two ranges of data move together, i.e.,
whether large values of one set are associated with large values of the other (positive
correlation), whether small values of one set are associated with large values of the
other (negative correlation), or whether values in both sets are unrelated (correla-
tion near zero). The results are shown in Table 4.12. The analysis indicates a strong
Sample TC a Qb MSc
Yieldd 0.3289 -0.0872 -0.1563
T
Lmean 0.2246 -0.1881 -0.8989
Yield 0.3320 -0.2081 0.2870
TS
Lmean 0.3732 -0.6084 -0.7292
Yield -0.1869 -0.6500 0.0747
IS
Lmean 0.3217 -0.2947 -0.8516
a
TC : Crystallization temperature ( C).
b
Q: Flow rate (ml/min).
c
MS: Mixing speed (rpm).
d
Yield of NaCl particles (g/cm3 ).
correlation between mixing speed and mean size of particles. The yield dependence
on feed flow rate and crystallization temperature should have strong function as
discussed and showed previously in precipitation analysis in Section 3.3.1. The
140
explanation for the weak correlation in T and TS samples might be the large
uncertainty in yield which was obtained by AAS analysis. As a result, the yield
data in Table 4.9 and 4.10 are used to show merely the reduction of NaCl and
cannot be used in further analysis.
a
Constant experimental parameters: Feed solution concen-
tration (C01 ) = CaCl2 (47.631.76 wt%), KCl (1.070.16
wt%), NaCl (0.670.08 wt%) with 95% confidence level, mix-
ing design = standard baffle + turbine. All the measured
values are given with confidence level of 95%.
b
TR : Reaction temperature.
c
MS : Mixing speed.
a
Constant experimental parameters: Feed solution concentration (C03 ) =
CaCl2 (47.091.76 wt%), KCl (1.030.16 wt%), NaCl (0.440.08 wt%) with
95% confidence level, mixing design = standard baffle + impeller. All the
measured values are given with confidence level of 95%.
b
TR : Reaction temperature.
c
MS : Mixing speed.
d
MP : Mixing power.
Table 4.15. Mean size of particles measured as a function of time (hour) for
BT samplesa .
a
Constant experimental parameters: mixing design = standard baffle + turbine. All the
measured values are given with confidence level of 95%.
Table 4.16. Mean size of particles measured as a function of time (hour) for BI
samplesa .
a
Constant experimental parameters: mixing design = standard baffle + impeller. All the
measured values are given with confidence level of 95%.
result in prevention of particles from aggregation. Even with high speed of mixing
for BI sample, the effect of mixing with an impeller was negligible compared to a
turbine.
144
4.7 Summary
The crystallization experiment at 30 C showed the reduction of NaCl only which
was comparable with a OLI analysis showing a reasonable prediction of reduction in
NaCl after the crystallization. However, the precipitation analysis with C01 solution
concentration showed CaCl2 6H2 O crystallizing out from 28 C. The EDS analysis
of solids which were produced by the crystallization at 25 C showed no CaCl2 . The
discrepancy between experiment and OLI analysis was due to the difference in real
and ideal system of kinetics and thermodynamics.
The realtime crystallization conditions were monitored for each experiment with
on-line data acquisition measuring all parameters in every second till the completion
of crystallization experiment typically 10 times of mean residence time (10 ). The
SEM figures with the aid of microscopic figures demonstrate the detailed particle
shape and size to explain the possible phenomena of how these particles are evolving
in a CMSMPR crystallizer. Initially the mean particle size was 180 m and
with time the mean size shifts to 120 m. In other experiments, this shift of
mean size with operating time was always observed and it was affected mostly by
crystallization temperature, feed flow rate, mixer rpm, and different type of mixer.
In addition, a small satellite peak at 20 m in Figure 4.13 developed with time in
all of our experiments. These particle size distribution results were the indicative
of the shape change from dendritic to bipyramidal particles and more breakage of
145
the aggregates with time. The particle size distribution stabilized at long operating
times as would be expected for a CMSMPR crystallizer operating at steady state.
The experimental results with the 3D plots of CSDs clearly showed the proof
of CSD reproducibility which demonstrated the precision of crystallization process
and CSD measurement. With the accuracy in CSDs measured and microscopic
photos taken at every mean residence time during the crystallization showed distinct
characterizations, i.e., the CSD at the beginning of crystallization and the evolution
of CSDs with time, of NaCl crystals depending on various experimental parameters,
such as, feed solution concentration, crystallization temperature (TC ), feed flow rate
(Q), mixing speed (M S), and different type of mixer on resultant CSD with vivid
graphical representations.
Overall effects of crystallization temperature, feed flow rate, and mixing speed
on products, such as mean size of particles and yield, were discussed in Section 4.5.7.
However, due to the result of difficulties inherent in obtaining accurate liquid
compositions the effects on yield were not clear. The sample solution at room
temperature containing solid particles required further treatments such as making
dilution with distilled water and adding 5% HCl. In addition to the difficulties
obtaining precise concentration of aqueous electrolyte solutions, the accuracy of
solution concentration was violated during such treatments for T and TS samples.
However, with more careful treatments for IS samples it was possible to show the
effects of crystallization temperature and feed flow rate on the yield of NaCl at
10 . It was found out that as the feed flow rate and the crystallization temperature
increase the yield of NaCl particles decreased. The effect of mixing speed on yield
was small. The mean particle size decreased with respect to increase in feed flow
rate and mixing speed and increased with temperature increase. This clearly showed
the mean particle size dependence on M S > Q > TC . The analysis along with the
data shown in Figure 4.26 provided justification for the effect of feed flow rate on
the yield of NaCl particles, i.e., the particles having long residence time (space
time in crystallizer) will grow more as a result of particles spending more time in
crystallizer. For the crystallization temperature at 25 C, the experiment with feed
146
flow rate of 20 ml/min was not successful due to crystallization in the feed before
entering the crystallizer.
Additional experiments performed focusing on aggregation and attrition without
the influence of nucleation and crystal growth being active. In this experiment, the
effect of reaction temperature on mean size of particles as a function of time showed
in the order of 30 > 25 > 35 C for both of experimental designs (BT and BI) in
which it clearly showed the competing mechanism of breakage and aggregation.
With lower speed of mixing the lager particle was obtained as the high speed of
mixing resulted in more breakage and less aggregation of particles. The mixing
speed effect with an impeller was negligible. As expected, the turbine creating
biaxial mixing of solution below and above the mixer was resulted in almost 2-fold
higher particle growth by aggregation than the impeller.
As described in Section 1.2, all of the collected conditions, parameters, and
product data will be analyzed with the application of known model in next Chapter.
Purification of CaCl2 from a solution, CaCl2 + KCl + NaCl, involved nucleation
and growth of NaCl crystals from the solution without much aggregation. As a
result, we should be able to run a validation of the OLI model with this system.
147
(a) (b)
(c) (d)
(e) (f)
(a)
(b)
(a) (b)
(c) (d)
Figure 4.33. Effect of different mixer on CSDs as a function of time (hours) and
differential volume (%) (a) and (b) with a 6-bladed flat disc turbine (Rushton
type) for BT 06, and (c) and (d) with a 6-bladed pitch blade impeller for BI 04
at mixing speed = 482.16.0 and 479.78.49 rpm, crystallization temperature =
31.00.38 and 29.910.23 C starting with same reactor volume = 1400 ml. The
plots in righthand side are shown with a view from z-direction. See Table 4.13
and 4.14 for detailed experimental conditions.
151
The CMSMPR models described in Section 1.2.1 to evaluate the results from
CMSMPR experiments are only valid if the assumptions defined in that section
hold. Deviations from this idealized CMSMPR model may be caused by the
following effects:
1. kinetics of crystal growth (e.g., size dependent growth, growth rate disper-
sion) [63, 96].
Usually any operating condition will produce some of these effects. For instance
increase of stirrer speed causes better mixing but more attrition. Figure 5.21 in
Section 5.3.3 illustrates qualitatively how some of these deviations from the idealized
CMSMPR model can influence the steady state crystal size distribution plotted in
a semi-logarithmic population density plot. The straight solid line in Figure 5.21
represents the population density expected according to the idealized CMSMPR
model. In most cases it is difficult to evaluate the specific reason for the deviation
from the linear plot because the same type of deviation can result from several
153
causes. Therefore the shape of a population density plot does not explain the
particular effect that has caused the deviation in a particular experiment. The
purpose of this chapter is to illustrate the particular effects to the resultant CSD
focusing on the effect of independent variables, i.e., feed solution concentration,
crystallization temperature, feed flow rate, mixing speed, and different type of
mixer, to the dependent variables, i.e., number of nuclei, nucleation rate, growth
rate and aggregation rate.
Figure 5.1. A plot of population number density distribution at 1.7 and 10.3
from the starting of a CMSMPR crystallizer.
I1 (x)
n(v) = 2n0 exp(px) (5.1)
x
where
r
1
p = 1+
20 n0 G0 2
v p
x = 20 n0 G0
G0
n(0) = n0 is the number density of nuclei in the product stream. The constants
and n0 were set to unity and three simulations were performed with the 0
(size-independent aggregation kernel) and G0 (growth rate constant) values given
in Figure 5.2. This figure clearly shows the effects of aggregation and growth to
Figure 5.2. FEM predictions of number density function (n(v)) for aggregation
and growth [99].
constant aggregation rate kennel and a constant crystal growth rate [99]. With
measured CSD in number fraction (nL ), total mass of particles (MT ) was obtained
as Z
nL
MT = L3 dL (5.2)
0 100
where = 0.471 is shape factor for octahedron, = 2.170 g/cm3 is the density of
NaCl. Then the mass of particles in size L is given as
mvL
ML = (5.3)
MT
where vL is total volume of particles in size L and m is the mass of particles per
unit suspension volume. The number density, n, is given as
ML
n= (5.4)
(L3 ) L
With the nonlinear regression data fit adjusting initial variables, i.e., A, B, and
G (approximate G was obtained with the application of a MSMPR model), to the
experimental result of particle number density (N ), n0 , G, A and B are known, so
does B0 (= n0 G). Therefore, the aggregation rate constant is obtained.
B
Ka = (5.6)
G 2 n0
A typical nonlinear regression data fit is shown in Figures 5.3 and the resulting
coefficients for all the experimental runs for T samples are shown in Table 5.1. In
Figure 5.3, the little humps below 50 m in particle size were not used in the data fit.
Only the larger particle sizes where there is a clear indication of aggregation were
used. This analysis also does not account for breakage which is evident in Figure 5.3
due to lower trend in the data than the fit line for large particles. While this fitting
methodology does not account for all phenomena, i.e., breakage is missing, it will
157
Figure 5.3. A typical plot of nonlinear regression data fit to the particle number
density (N ) using the Hounslows constant aggregation rate equation at steady
state. The data points (2s) filled with s were used for the fit which is the solid
line. The particle population at zero size (n0 ) according to the fit is shown as a .
be used for further analysis since it is the only method available in the literature to
account for all possible phenomena in CMSMPR crystallizer. There is no analytical
solution to the population balance in a CMSMPR crystallizer to deal with the
nucleation, growth, aggregation, and breakage, altogether. Hounslows solution is
only for nucleation, growth, and aggregation, and does not account for breakage.
As a result, the results for the aggregation rate constant must be considered an
effective aggregation rate one that combines the dynamic balance of aggregation
with breakage.
With the application of Hounslows solution into the measured CSD at 10 for
all the experimental runs, the number of nuclei (n0 ), growth rate (G), nucleation
rate (B0 ), and aggregation rate (Ka ) were obtained and tabulated for each set of
designs in Tables 5.1, 5.2, and 5.3 which also give the standard error of estimate
indicated as a fitting error. The experimental conditions, such as crystallization
temperature (TC ), feed flow rate (Q), and mixing speed (M S), for each set of data
158
Analytical Data
Test No. n0 G B0 Ka Fitting
(#/m4 ) (m/s) (#/m3 s) (m3 /s) error c
T 04 4.02E+13 8.21E-03 3.30E+05 1.03E-15 0.023
T 05 1.21E+14 6.43E-03 7.77E+05 4.38E-16 0.022
T 06 3.52E+14 5.84E-03 2.06E+06 1.65E-16 0.063
T 07 4.53E+13 1.10E-02 5.19E+05 1.47E-15 0.02
T 08 1.03E+14 9.02E-03 9.29E+05 7.69E-16 0.046
T 09 2.63E+14 5.11E-03 1.35E+06 4.93E-16 0.055
T 10 3.11E+13 4.47E-03 1.39E+05 6.12E-16 0.033
T 11 7.22E+13 3.60E-03 2.60E+05 3.27E-16 0.041
T 12 1.17E+14 3.19E-03 3.73E+05 2.28E-16 0.048
T 13 4.25E+13 8.30E-03 3.53E+05 9.64E-16 0.003
T 14 9.52E+13 6.68E-03 6.36E+05 5.35E-16 0.022
T 15 2.44E+14 5.63E-03 1.37E+06 2.48E-16 0.045
T 16 3.99E+13 1.30E-02 4.99E+05 1.53E-15 0.026
T 17 9.71E+13 9.93E-03 9.63E+05 7.94E-16 0.033
T 18 2.39E+14 8.68E-03 2.07E+06 3.69E-16 0.037
T 19 3.09E+13 4.59E-03 1.42E+05 5.99E-16 0.058
T 20 4.72E+13 4.04E-03 1.91E+05 4.46E-16 0.07
T 21 8.72E+13 3.42E-03 2.98E+05 2.85E-16 0.053
T 22 3.54E+13 8.49E-03 3.00E+05 1.13E-15 0.038
T 23 7.67E+13 7.07E-03 5.42E+05 6.28E-16 0.045
T 24 1.45E+14 6.07E-03 8.81E+05 3.86E-16 0.054
T 25 3.60E+13 1.30E-02 4.52E+05 1.69E-15 0.036
T 26 9.34E+13 1.00E-02 9.38E+05 8.16E-16 0.019
T 27 2.04E+14 8.09E-03 1.65E+06 4.63E-16 0.023
a
Constant experimental parameters: feed solution concentration
(C01 ) = CaCl2 (47.631.76 wt%), KCl (1.070.16 wt%), NaCl (0.670.08
wt%) with 95% confidence level, feed solution temperature (TS ) = 70 C
or above, mixing design = standard baffle + turbine.
b
For experimental condition of each test no., see Table 4.4 in Sec-
tion 4.3.3.
c
The analytical data with fitting error bigger than 0.07 was discarded
in this set of experiments.
were tabulated previously in Tables 4.4, 4.5, and 4.6 in Section 4.3.3. As shown in
Figure 5.3, the aggregation of particles was noticed in size range less than 50 m
159
Analytical Data
Test No. n0 G B0 Ka Fitting
(#/m4 ) (m/s) (#/m3 s) (m3 /s) error c
TS 04 4.87E+13 7.24E-03 3.53E+05 9.64E-16 0.021
TS 05 1.18E+14 6.05E-03 7.16E+05 4.75E-16 0.039
TS 06 2.95E+14 4.87E-03 1.43E+06 2.37E-16 0.071
TS 07 5.56E+13 1.10E-02 6.04E+05 1.27E-15 0.022
TS 08 1.21E+14 9.08E-03 1.10E+06 6.99E-16 0.006
TS 09 3.38E+14 7.08E-03 2.39E+06 3.20E-16 0.085
TS 10 2.18E+13 4.51E-03 9.81E+04 8.67E-16 0.025
TS 11 4.56E+13 3.75E-03 1.71E+05 4.98E-16 0.051
TS 12 9.32E+13 3.19E-03 2.97E+05 2.86E-16 0.048
TS 13 3.72E+13 7.74E-03 2.88E+05 1.18E-15 0.026
TS 14 6.20E+13 7.04E-03 4.36E+05 7.80E-16 0.023
TS 15 1.34E+14 5.88E-03 7.87E+05 4.32E-16 0.041
TS 16 4.75E+13 1.10E-02 5.39E+05 1.42E-15 0.02
TS 17 1.12E+14 9.26E-03 1.04E+06 7.39E-16 0.033
TS 18 2.29E+14 7.73E-03 1.77E+06 4.32E-16 0.059
TS 19 1.59E+13 4.70E-03 7.46E+04 1.14E-15 0.036
TS 20 2.92E+13 4.21E-03 1.23E+05 6.92E-16 0.051
TS 21 5.79E+13 3.60E-03 2.09E+05 4.08E-16 0.067
TS 22 2.43E+13 8.64E-03 2.10E+05 1.62E-15 0.034
TS 23 5.27E+13 7.35E-03 3.88E+05 8.77E-16 0.019
TS 24 1.09E+14 6.17E-03 6.74E+05 5.05E-16 0.038
TS 25 4.78E+13 1.10E-02 5.40E+05 1.42E-15 0.018
TS 26 9.03E+13 9.68E-03 8.75E+05 8.75E-16 0.034
TS 27 1.58E+14 8.44E-03 1.34E+06 5.73E-16 0.033
a
Constant experimental parameters: feed solution concentration
(C02 ) = CaCl2 (46.881.73 wt%), KCl (0.930.15 wt%), NaCl (0.500.05
wt%) with 95% confidence level, feed solution temperature (TS ) = 70 C
or above, mixing design = standard baffle + turbine.
b
For experimental condition of each test no., see Table 4.5 in Sec-
tion 4.3.3.
c
The analytical data with fitting error bigger than 0.085 was discarded
in this set of experiments.
Analytical Data
Test No. n0 G B0 Ka Fitting
(#/m4 ) (m/s) (#/m3 s) (m3 /s) error c
IS 04 5.36E+13 6.80E-03 3.64E+05 9.34E-16 0.018
IS 05 1.48E+14 5.56E-03 8.22E+05 4.14E-16 0.032
IS 06 4.55E+14 4.21E-03 1.92E+06 1.77E-16 0.05
IS 07 5.96E+13 9.90E-03 5.90E+05 1.30E-15 0.016
IS 08 1.24E+14 8.28E-03 1.03E+06 7.44E-16 0.025
IS 09 2.60E+14 6.79E-03 1.76E+06 4.34E-16 0.041
IS 10 4.38E+13 3.81E-03 1.67E+05 5.09E-16 0.025
IS 11 9.63E+13 3.11E-03 2.99E+05 2.84E-16 0.019
IS 12 1.78E+14 2.72E-03 4.86E+05 1.75E-16 0.019
IS 13 3.23E+13 8.05E-03 2.60E+05 1.31E-15 0.002
IS 14 6.53E+13 6.85E-03 4.47E+05 7.61E-16 0.017
IS 15 1.45E+14 5.56E-03 8.08E+05 4.21E-16 0.023
IS 16 5.05E+13 1.10E-02 5.44E+05 1.41E-15 0.024
IS 17 1.01E+14 8.75E-03 8.83E+05 8.67E-16 0.02
IS 18 2.69E+14 7.12E-03 1.92E+06 3.99E-16 0.025
IS 19 2.53E+13 4.26E-03 1.08E+05 7.91E-16 0.047
IS 20 5.89E+13 3.44E-03 2.02E+05 4.20E-16 0.056
IS 21 1.19E+14 2.93E-03 3.48E+05 2.44E-16 0.033
IS 22 2.13E+13 8.54E-03 1.82E+05 1.87E-15 0.018
IS 23 5.24E+13 7.02E-03 3.68E+05 9.24E-16 0.021
IS 24 1.05E+14 5.91E-03 6.18E+05 5.50E-16 0.021
IS 25 3.98E+13 1.10E-02 4.52E+05 1.69E-15 0.023
IS 26 8.65E+13 9.31E-03 8.05E+05 9.51E-16 0.023
IS 27 3.31E+14 6.70E-03 2.22E+06 3.45E-16 0.061
a
Constant experimental parameters: feed solution concentration
(C03 ) = CaCl2 (47.091.74 wt%), KCl (1.030.15 wt%), NaCl (0.440.04
wt%), feed solution temperature (Ts ) = 70 C or above, mixing design =
standard baffle + impeller.
b
For experimental condition of each test no., see Table 4.6 in Sec-
tion 4.3.3.
c
The analytical data with fitting error bigger than 0.061 was discarded
in this set of experiments.
Dependent Variable
Independent
n0
Variable
Reg. Coef.b tstat c 2 d
Rad F Fcrit
0 3.23E+141.36E+14 2.3674
TC 0.0597 2.4605 1.31E-01
1 -6.89E+124.39E+12 -1.5686
0 5.22E+135.13E+13 1.0164
Q 0.0203 1.4767 2.37E-01
1 1.38E+121.13E+12 1.2152
0 -2.15E+145.34E+13 -4.0286
MS 0.6223 38.8904 2.82E-06
1 8.53E+111.37E+11 6.2362
G
0 8.60E-034.65E-03 1.8469
TC -0.0414 0.0856 7.73E-01
1 -4.38E-051.50E-04 -0.2926
0 1.03E-039.80E-04 1.0507
Q 0.6598 45.6131 8.69E-07
1 1.46E-042.17E-05 6.7537
0 1.35E-022.53E-03 5.3310
MS 0.1897 6.3831 1.92E-02
1 -1.64E-056.49E-06 -2.5265
B0
0 1.98E+069.05E+05 2.1889
TC 0.0368 1.8783 1.84E-01
1 -4.00E+042.92E+04 -1.3705
0 -7.95E+042.92E+05 -0.2721
Q 0.2623 9.1774 6.16E-03
1 1.95E+046.45E+03 3.0294
0 -9.97E+054.41E+05 -2.2624
MS 0.4016 16.4331 5.29E-04
1 4.57E+031.13E+03 4.0538
Ka
0 6.96E-167.00E-16 0.9938
TC -0.0454 0.0003 9.87E-01
1 -3.68E-192.25E-17 -0.0163
0 1.25E-162.25E-16 0.5547
Q 0.2085 7.0583 1.44E-02
1 1.32E-174.96E-18 2.6567
0 2.26E-152.64E-16 8.5575
MS 0.6109 37.1131 3.93E-06
1 -4.11E-186.75E-19 -6.0921
a
Regression statistics was performed with confidence level of 95%.
b
Regression model with one independent variable: y = 0 + 1 x1 .
c
T statistics provide the relative importance of each variable in the model.
d 2
Rad : R is the correlation between the observed and predicted values of the dependent
2
variable. Rad is the square of this correlation and adjusted by the number of variables in the
model and the sample size to compensate for the optimistic bias of R2 .
163
Dependent Variable
Independent
n0
Variable b
Reg. Coeff. tstat c 2 d
Rad F Fcrit
0 2.50E+141.61E+14 1.5534
TC Q 1 -5.90E+124.56E+12 -1.2945 0.0495 1.5989 2.26E-01
2 1.01E+121.15E+12 0.8739
0 -2.79E+145.66E+13 -4.9289
Q MS 1 1.46E+126.47E+11 2.2582 0.6816 25.6185 2.32E-06
2 8.58E+111.26E+11 6.8332
0 -1.17E+128.74E+13 -0.0134
M S TC 1 8.54E+111.18E+11 7.2122 0.7167 30.0910 6.82E-07
2 -6.96E+122.41E+12 -2.8864
G
0 -2.53E-033.08E-03 -0.8187
TC Q 1 1.06E-048.75E-05 1.2143 0.6670 24.0356 3.72E-06
2 1.53E-042.21E-05 6.9140
0 7.16E-031.15E-03 6.2123
Q MS 1 1.45E-041.32E-05 10.9983 0.8744 81.0756 1.33E-10
2 -1.59E-052.56E-06 -6.2124
0 1.48E-024.89E-03 3.0307
M S TC 1 -1.64E-056.63E-06 -2.4729 0.1551 3.1105 6.56E-02
2 -4.25E-051.35E-04 -0.3149
Table continues on next page.
a
Regression statistics was performed with confidence level of 95%.
b
Regression model with two independent variables: y = 0 + 1 x1 + 2 x2 + .
c
T statistics provide the relative importance of each variable in the model.
d 2
Rad : R is the correlation between the observed and predicted values of the dependent
2
variable. Rad is the square of this correlation and adjusted by the number of variables in the
model and the sample size to compensate for the optimistic bias of R2 .
a
Regression statistics was performed with confidence level of 95%.
b
Regression model with two independent variables: y = 0 + 1 x1 + 2 x2 + .
c
T statistics provide the relative importance of each variable in the model.
d 2
Rad : R is the correlation between the observed and predicted values of the dependent
2
variable. Rad is the square of this correlation and adjusted by the number of variables in the
model and the sample size to compensate for the optimistic bias of R2 .
90% accuracies in the case of T samples shown in Table 5.6. The linear regression
analyses of single, two, and three factor interaction were performed for all samples,
i.e., T, TS, and IS samples. However, for the illustration purpose only the result
of T samples is given.
The correlation analysis for each one of independent variables such as crystalliza-
tion temperature (TC ), feed flow rate (Q), and mixing speed (M S), to each depen-
dent variable was performed for T, TS, and IS samples as shown in Table 5.7. The
negative sign indicates that as an independent variable increases the corresponding
dependent variable tends to decrease. As shown in Table 5.7, this analysis presents
165
Dependent Variable
Independent
n0
Variable b
Reg. Coeff. tstat c 2 d
Rad F Fcrit
0 -8.19E+139.39E+13 -0.8718
1 -5.89E+122.36E+12 -2.4945
TC Q M S 0.7450 2.34E+01 9.49E-07
2 1.09E+125.97E+11 1.8257
3 8.58E+111.12E+11 7.6342
G
0 3.61E-031.98E-03 1.8281
1 1.06E-044.97E-05 2.1337
TC Q M S 0.8926 6.47E+01 1.79E-10
2 1.51E-041.26E-05 12.0499
3 -1.59E-052.36E-06 -6.7160
B0
0 -1.13E+066.43E+05 -1.7631
1 -2.21E+041.62E+04 -1.3646
TC Q M S 0.7216 2.09E+01 2.26E-06
2 1.86E+044.09E+03 4.5465
3 4.65E+037.70E+02 6.0353
Ka
0 1.25E-153.48E-16 3.5888
1 1.33E-178.75E-18 1.5234
TC Q M S 0.8521 4.52E+01 4.29E-09
2 1.36E-172.21E-18 6.1559
3 -4.07E-184.16E-19 -9.7691
a
Regression statistics was performed with confidence level of 95%.
b
Regression model with three independent variables: y = 0 + 1 x1 + 2 x2 + 3 x3 + .
c
T statistics provide the relative importance of each variable in the model.
d 2
Rad : R is the correlation between the observed and predicted values of the dependent variable.
2
Rad is the square of this correlation and adjusted by the number of variables in the model and
the sample size to compensate for the optimistic bias of R2 .
the dominant interactions (the number within a box) of feed flow rate (Q) and
mixing speed (M S) on resultant product data in all designs, also shows the weak
interactions of crystallization temperature (TC ). The strong interaction of mixing
speed (M S) and weak interaction of crystallization temperature (TC ) on number of
nuclei (n0 ), nucleation rate (B0 ), and aggregation rage (Ka ) could be explained by
secondary nucleation which occurs in the presence of a solute-particle interface or
166
a
TC : Crystallization temperature ( C).
b
Q: Flow rate (ml/min).
c
MS: Mixing speed (rpm).
causes mostly by collisions of crystals with other crystals, crystallizer, and mixer.
Once supersaturation creates solid particles by primary nucleation in a CMSMPR
crystallizer, the nucleation rate (B0 ) mostly depends on stirrer rotation rate (rpm)
in the presence of such solid particles. This is called secondary or contact nucle-
ation as discussed in Section 1.3.3. Compared to the strong contribution of mixer
rpm, the effect of supersaturation on secondary nucleation was weak as shown in
Equation 1.36, thus explaining the weak interaction of crystallization temperature
(TC ) and the strong interaction of mixer rpm on products (dependent variables).
The large correlation of dependent variables with mixing speed indicates that the
secondary nucleation and growth play the dominant crystallization mechanism for
this system, as most industrial crystallizer does. In addition to the large number of
small particles created by secondary nucleation and growth, particle breakage and
aggregation were certainly resulted in deviation of CSD from linearity as indicated
167
in Figure 5.3 in previous section. The humps in small particle size range (below 50
m) were the indication of aggregation among these small particles and relatively
large particles. The deviation of CSD from the linear line in large particle size
region (above 230 m) was the result of large particle breakage. The growth rate
(G) shows relatively weak correlation with mixing speed due to such dominant
particle interaction. The flow rate effect, i.e., Q = V / where = mean residence
time, shows strong correlation with growth rate (G) and nucleation rate (B0 ) which
induced mostly by supersaturation. Also the bold numbers in Table 5.7 indicate
rather strong correlations between dependent variables, i.e., B0 and n0 in B0 = n0 G.
The aggregation rate (Ka ) shows relatively strong correlation with the number of
nuclei (n0 ), growth rate (G), and nucleation rate (n0 ) indicating the importance of
secondary nucleation and the crystallization mechanism involving growth, breakage,
and aggregation together.
The individual and interaction effects of independent variables, i.e., crystalliza-
tion temperature (TC ), feed flow rate (Q), and mixing speed (M S), on each one
of dependent variables, i.e., number of nuclei (n0 ), growth rate (G), nucleation
rate (B0 ), and aggregation rate (Ka ), were attempted to compare T, TS, and IS
sample designs to show the degree of importance among processing variables to the
differences in product variables due to the difference in designs which were shown
previously in Tables 2.2, 2.3, and 2.4 in Section 2.7.2. Tables 5.8, 5.9, and 5.10
present the results obtained with the application of factorial design analysis. In
this analysis however, all the observations should be used to supply information on
each of the individual and interaction effects. However, as explained in Section 4.3.3,
the crystallization experiments at 25 C with 20 ml/min were not carried out for all
experimental designs. Therefore, the measures of individual and interaction effects
were performed disregarding all of experimental results at 25 C. The regression and
correlation analyses showed previously that the crystallization temperature effect
on the dependent variables was small. This leads the choice made for this factorial
design analysis was reasonable to show the individual and interaction effects of
independent variables focusing the effects of feed flow rate and mixing speed on
168
product variables. In Tables 5.8, 5.9, and 5.10, the main effects of crystallization
temperature (TC ), feed flow rate (Q), and mixing speed (M S) are denoted by A,
B, and C, respectively. The two or three-factor interactions are indicated as AB,
BC, and CA, or ABC. Figures 5.4, 5.5, and 5.6 illustrate the effects of individuals
and interactions on growth rate (G), nucleation rate (B0 ), and aggregation rate
(Ka ), respectively for T, TS, and IS samples. The effect on number of nuclei (n0 )
is omitted here since the relation, i.e., B0 = n0 G, is known. Figure 5.4 clearly
Figure 5.4. The order of importance in main and interaction effects on growth
rate (G). Crystallization temperature (TC ), feed flow rate (Q), and mixing speed
(M S) are denoted by A, B, and C, respectively. Dotted lines indicate the standard
error of estimate.
shows the importance of main and interaction effects on growth rate (G). The
feed flow rate (Q) provides the largest positive effect on growth rate. The main
effect of mixing speed (M S) and the two-factor interaction effect (BC) of feed
flow rate and mixing speed show the second and the third effects as negative to
growth rate. All other effects are within the standard error of estimation. The
difference in feed solution concentration between T and TS samples (C01 and C02 )
169
clearly stands out in all three figures with lower effects of flow rate and mixing
speed on growth, nucleation, and aggregation rates for TS samples as a result of
lower NaCl concentration in C02 compared to C01 . Figures 5.5 and 5.6 present the
effects on nucleation rate (B0 ) and aggregation (Ka ), respectively. The main and
Figure 5.5. The order of importance in main and interaction effects on nucleation
rate (B0 ). Crystallization temperature (TC ), feed flow rate (Q), and mixing speed
(M S) are denoted by A, B, and C, respectively. Dotted lines indicate the standard
error of estimate.
interaction effects of feed flow rate and mixing speed on nucleation rate explain
the importance of secondary nucleation caused by high supersaturation and high
turbulence which easily washes nuclei away from the existing crystal surface in
solution. By comparing three figures, the effects on IS samples show almost same
trends as on T and TS samples. The effects on growth and aggregation rate were
almost comparable to T sample. The effects on nucleation rate was even higher than
the T and TS samples even though the concentration (C03 ) of IS sample was the
lowest among the designs and the speed of mixing was higher compare to T and TS
samples. The mixing pattern due to the different type of mixer, i.e., an impeller,
170
was single axial mixing rather than bi-axial mixing. In Figure 5.6, the effect of
Figure 5.6. The order of importance in main and interaction effects on aggregation
rate (Ka ). Crystallization temperature (TC ), feed flow rate (Q), and mixing speed
(M S) are denoted by A, B, and C, respectively. Dotted lines indicate the standard
error of estimate.
mixing shows negative effect on aggregation rate meaning that the mixing plays an
important role on secondary nucleation which is clearly indicated in Figure 5.5.
Table 5.8. Factorial design analysis of independent variables (TC , Q, and M S) to dependent
variables (n0 , G, B0 , and Ka ) for T samplesa,b .
n0 G
Interaction
Reg. Coef. F Fcrit S.E.E.c Reg. Coef. F Fcrit S.E.E.
A -9.21E+12 15.16 12.25 -3.03E-05 0 12.25
B 3.42E+13 208.52 12.25 3.52E-03 3784.61 12.25
C 6.88E+13 845.91 12.25 -1.52E-03 702.61 12.25
AB -1.12E+12 0 12.25 4.73E+12 -1.23E-04 4.58 12.25 1.15E-04
BC 3.04E+13 165.16 12.25 -8.80E-04 236.24 12.25
CA -8.12E+12 11.77 12.25 -6.25E-05 1.19 12.25
ABC -1.42E+11 0.00 12.25 -9.03E-05 2.49 12.25
B0 Ka
Interaction
Reg. Coef. F Fcrit S.E.E. Reg. Coef. F Fcrit S.E.E.
A -7.23E+04 5.09 10.04 3.93E-17 3.59 10.04
B 5.02E+05 245.56 10.04 3.10E-16 222.70 10.04
C 4.26E+05 177.20 10.04 -4.11E-16 391.13 10.04
AB -5.30E+04 0 10.04 6.41E+04 2.75E-17 0 10.04 4.16E-17
BC 3.21E+05 100.56 10.04 -2.25E-16 117.57 10.04
CA -6.05E+04 3.57 10.04 6.57E-19 0 10.04
ABC -3.96E+04 0 10.04 -1.79E-17 0 10.04
a
Regression statistics was performed with confidence level of 95%.
b
Crystallization temperature (TC ), feed flow rate (Q), and mixing speed (M S) are denoted by A, B, and
C, respectively.
c
S.E.E.: Standard error of estimation.
171
Table 5.9. Factorial design analysis of independent variables (TC , Q, and M S) to dependent
variables (n0 , G, B0 , and Ka ) for TS samplesa,b .
n0 G
Interaction
Reg. Coef. F Fcrit S.E.E.c Reg. Coef. F Fcrit S.E.E.
A -1.52E+13 43.51 12.25 1.71E-04 18.70 12.25
B 3.99E+13 299.80 12.25 2.87E-03 5239.55 12.25
C 5.51E+13 571.58 12.25 -1.07E-03 726.56 12.25
AB -4.01E+12 3.03 12.25 4.61E+12 1.42E-05 0 12.25 7.92E-05
BC 2.43E+13 110.75 12.25 -4.42E-04 124.36 12.25
CA -1.37E+13 35.24 12.25 1.21E-04 9.31 12.25
ABC -5.70E+12 6.11 12.25 6.47E-05 2.67 12.25
B0 Ka
Interaction
Reg. Coef. F Fcrit S.E.E. Reg. Coef. F Fcrit S.E.E.
A -7.42E+04 15.32 12.25 7.00E-17 69.55 12.25
B 4.76E+05 630.69 12.25 1.50E-16 320.71 12.25
C 3.20E+05 285.91 12.25 -4.13E-16 2424.23 12.25
AB -4.37E+04 5.31 12.25 3.79E+04 -3.37E-17 16.10 12.25 1.68E-17
BC 2.30E+05 147.20 12.25 -6.83E-17 66.16 12.25
CA -6.81E+04 12.91 12.25 -9.79E-19 0 12.25
ABC -5.04E+04 7.06 12.25 3.88E-17 21.35 12.25
a
Regression statistics was performed with confidence level of 95%.
b
Crystallization temperature (TC ), feed flow rate (Q), and mixing speed (M S) are denoted by A, B, and
C, respectively.
c
S.E.E.: Standard error of estimation.
172
Table 5.10. Factorial design analysis of independent variables (TC , Q, and M S) to dependent
variables (n0 , G, B0 , and Ka ) for IS samplesa,b .
n0 G
Interaction
Reg. Coef. F Fcrit S.E.E.c Reg. Coef. F Fcrit S.E.E.
A -3.69E+12 1.22 12.25 3.09E-05 0 12.25
B 4.39E+13 172.68 12.25 2.87E-03 3816.85 12.25
C 9.97E+13 892.45 12.25 -1.38E-03 877.88 12.25
AB 1.74E+13 27.30 12.25 6.68E+12 -1.39E-04 9.02 12.25 9.29E-05
BC 3.81E+13 130.46 12.25 -7.49E-04 260.41 12.25
CA 4.23E+12 1.60 12.25 -8.49E-05 3.35 12.25
ABC 1.53E+13 21.04 12.25 -2.36E-05 0 12.25
B0 Ka
Interaction
Reg. Coef. F Fcrit S.E.E. Reg. Coef. F Fcrit S.E.E.
A 1.39E+03 0 12.25 7.75E-17 36.82 12.25
B 5.46E+05 477.79 12.25 2.83E-16 489.79 12.25
C 5.03E+05 404.08 12.25 -4.31E-16 1135.56 12.25
AB 5.49E+04 4.83 12.25 5.0E+04 -1.58E-17 1.53 12.25 2.56E-17
BC 3.51E+05 196.66 12.25 -1.96E-16 236.18 12.25
CA 4.25E+04 2.89 12.25 -7.35E-17 33.10 12.25
ABC 6.37E+04 6.49 12.25 -1.69E-17 1.75 12.25
a
Regression statistics was performed with confidence level of 95%.
b
Crystallization temperature (TC ), feed flow rate (Q), and mixing speed (M S) are denoted by A, B, and
C, respectively.
c
S.E.E.: Standard error of estimation.
173
174
(a)
(b)
Figure 5.7. Influence of crystallization temperature and mixing speed only (main
effect) on number of nuclei(n0 ), growth rate (G), and nucleation rate (B0 ), as well
as aggregation rate (Ka ) at steady state for T samples. (a) T 04, T 13, and T 22
samples at 285 rpm and (b) T 06, T 15, and T 24 samples at 481 rpm (see Tables 4.4
and 5.1 for experimental conditions and analytical data). The feed flow rate was
40 ml/min for both (a) and (b).
176
proved here again in Figure 5.8. As the strong correlation (0.9228) between n0 and
Figure 5.8. Nucleation rate (B0 ) vs. number of nuclei(n0 ) at steady state for T
samples (see Tables 4.4 and 5.1 for experimental conditions and analytical data).
B0 was indicated in Table 5.7, the correlation along with the interaction of mixing
speed, in addition the main interaction of flow rate on n0 and B0 are shown in
Figure 5.8. This plot shows as mixing speed increases from 283 to 385 rpm and to
483 rpm the effect of mixing on n0 and B0 was increased linearly as 2 and 4-fold
magnitude, respectively. With respect to increase in feed flow rate, the effect on
B0 tends to increase more while the effect on n0 increases less. This is showing the
relation of mean residence time on both of n0 and B0 meaning that the short mean
residence time creates high supersaturation resulting in high nucleation and growth
rate meanwhile particles spending less time in crystallizer reducing the effect of feed
flow rate on n0 meaning less secondary nucleation. In other words, lower feed flow
rate creates easy of relaxation for supersaturation resulting in lower nucleation rate,
but particles spending more time in crystallizer will experience more interactions,
i.e., particle-particle, particle-mixer, and particle-baffle-crystallizer wall. As a result
177
of those interactions more nuclei were created at low feed flow rate. The feed flow
rate along with mixing speed having major effect on the growth rate shows increase
in slope of line in the order of 3, 5, and 8 as feed flow rate increases in the order of
20, 40, and 60 ml/min.
The power law curve fit of particle mean size (Lmean ) to nucleation rate (B0 )
shown in Figure 5.9 presents the relation as higher nucleation rate mostly caused
by high supersaturation creating large number of nuclei results in smaller size of
particles or vice versa.
Figure 5.9. Nucleation rate (B0 ) vs. particle mean size (Lmean ) at steady state
for T samples (see Tables 4.4 and 5.1 for experimental conditions and analytical
data).
(TC ) showed the linear decrease in number of nuclei (n0 ) and nucleation rate
(B0 ) while growth rate (G) and aggregation rate (Ka ) showed linear increase.
TS samples showed the same trends as T samples for the case of crystallization
temperature effect on dependent variables. The linearity showed almost same
order of magnitude in both samples. In both sample designs the interaction of
crystallization temperature to the product data was weak. In T samples, mixing
speed effect was significant on n0 , B0 , and Ka while feed flow rate was interacting
strongly with growth rate (G). However in TS samples, the nucleation rate (B0 )
interacted more with feed flow rate instead of interacting with mixing speed. Also
in both designs, the correlation of product analytical data, i.e., between dependent
variables, showed somewhat different results. The interaction between n0 and B0
was significant for both designs. Meanwhile the interaction between Ka and G was
significant in T samples, Ka correlated more with n0 in TS samples.
When comparing the main effects of feed flow rate and mixing speed only to
the analytical product data shown in Figure 5.10 plotted as a function of feed flow
rate, all dependent variables tend to show increase in values as the feed flow rate
increases. However, the difference was minor when compared to T samples. In both
T and TS samples, the increase in growth and aggregation rate was small compared
to the number of nuclei and nucleation rate as the flow rate increases.
For the comparison of three-factor interaction on the number of nuclei (n0 ) and
nucleation rate (B0 ), the effects of crystallization temperature (TC ), feed flow rate
(Q), and mixing speed (M S) all together were plotted in Figure 5.11. Figure 5.11
shows the same pattern that the growth rate was almost constantly decreased with
respect to increase in mixing speed and decrease in crystallization temperature at
same feed flow rate when compared to Figure 5.8. However, there are two little
differences that can be pointed out between these two figures. The experimental
conditions for T and TS samples were almost same except for only a small difference
in feed solution concentration (C01 and C02 , see Appendix C). First, the slope of
linear line to 60 ml/min flow rate was smaller by 2 orders of magnitude when
compared to Figure 5.9 meaning that the growth rate was smaller in TS samples
179
(a)
(b)
Figure 5.10. Influence of feed flow rate and mixing speed only (main effect)
on number of nuclei(n0 ), growth rate (G), and nucleation rate (B0 ), as well as
aggregation rate (Ka ) at steady state for TS samples. (a) TS 10, TS 13, and TS
16 samples at 282 rpm and (b) TS 12, TS 15, and TS 18 samples at 479 rpm (see
Tables 4.5 and 5.2 for experimental conditions and results). The crystallization
temperature was 30 C for both (a) and (b).
than T samples at feed flow rate of 60 ml/min. Second, the mixing effect at feed
flow rate of 40 ml/min in TS samples was small compared to T samples. The growth
180
Figure 5.11. Nucleation rate (B0 ) vs. number of nuclei(n0 ) at steady state for TS
samples (see Tables 4.5 and 5.2 for experimental conditions and analytical data).
rate (G) at feed flow rate of 20 ml/min was almost same for both T and TS samples
due to the major effect of mixing at low feed flow rate. T samples with C01 having
slightly higher concentration than TS samples in all three components, i.e., CaCl2 ,
KCl, and NaCl, produced more particles which in turn created a suitable condition
for the particles to be exposed more to the effects of mixing speed (M S) and feed
flow rate (Q). As was the case with the T samples, the crystallization temperature
effect to the products in TS sample was small when compared to the effects of mixing
speed and feed flow rate. To see overall effect on growth rate, Figure 5.12 is plotted
B0 vs. 1/n0 . The growth rate within the same feed flow rate follows power law
relationship due to the effect from mixing speed and crystallization temperature.
This called growth rate dispersion which is more evident at low feed flow rate and
high mixing speed.
The correlation analysis at each of constant crystallization temperature, con-
stant feed flow rate, and constant mixing speed was performed to explore the
details of main effects to the product data as shown in Table 5.11. This is to
181
show the relationship between two variables while one is set constant. The crys-
tallization temperature interaction was small when compared with mixing speed,
i.e., the interactions of mixing speed with dependent variables were dominant over
the crystallization temperature in all cases. When compared the correlations of
crystallization temperature (TC ) and feed flow rate (Q) to the dependent variables,
Q was mostly dominant over TC at constant mixing speed, however, as mixing
speed increases from 281.8 to 479.0 rpm the interaction of TC to the products
is increased. At 479.0 rpm, the interactions of TC to number of nuclei (n0 ) and
aggregation rate (Ka ) are dominant over Q. This could be attributed to more
particle breakage and less aggregation as mixer rpm increases. Also, high rpm will
induce uniformity of the temperature distribution in the crystallizer which in turn
will increase temperature gradient between feed and slurry in crystallizer, therefore
resulting in high nucleation rate (B0 ) and aggregation rate (Ka ) due to the effect
of both temperature and mixing. The correlations of feed flow rate (Q) and mixing
182
rate (B0 ) also verifies further the interaction of variables on products as shown in
Figure 5.13.
Figure 5.13. Particle mean size (Lmean ) vs. nucleation rate (B0 ) at steady state
for TS samples (see Tables 4.5 and 5.2 for experimental conditions and results).
Figure 5.14. Predicted B0 vs. measured B0 at steady state for TS samples (see
Tables 4.5 and 5.2 for experimental conditions and results). Multiple regression
model of independent variables to B0 (i.e., y = 0 + 1 x1 + 2 x2 + 3 x3 + where
x1 , x2 , and x3 are crystallization temperature, feed flow rate, and mixing speed,
respectively.) was performed with 95% confidence level and showed 91% prediction
accuracy.
size smaller than 50 m cannot be used in the data fit and from the fitting errors
shown in Tables 5.1, 5.2, and 5.3 for T, TS, and IS samples. The predictions of G
and Ka to the measured show 95 and 96% accuracies, respectively, with confidence
level of 95% in Figures 5.15 and 5.16.
186
Figure 5.15. Predicted G vs. measured G at steady state for TS samples (see
Tables 4.5 and 5.2 for experimental conditions and results). Multiple regression of
independent variables to G (i.e., y = 0 + 1 x1 + 2 x2 + 3 x3 + where x1 , x2 , and
x3 are crystallization temperature, feed flow rate, and mixing speed, respectively.)
was performed with 95% confidence level and showed 99% prediction accuracy.
Table 5.11. Correlation analysis of independent variables to the dependent variables at each
of constant crystallization temperature (TC ), constant feed flow rate (Q), and constant mixing
speed (M S) for TS samples.
a
TC : Crystallization temperature ( C).
b
Q: Feed flow rate (ml/min).
c
MS: Mixing speed (rpm)
187
188
Figure 5.16. Predicted Ka vs. measured Ka at steady state for TS samples (see
Tables 4.5 and 5.2 for experimental conditions and results). Multiple regression of
independent variables to Ka (i.e., y = 0 + 1 x1 + 2 x2 + 3 x3 + where x1 , x2 , and
x3 are crystallization temperature, feed flow rate, and mixing speed, respectively.)
was performed with 95% confidence level and showed 96% prediction accuracy.
189
Figure 5.17. Number of nuclei(n0 ) vs. nucleation rate (B0 ) at steady state for IS
samples (see Table 5.3).
with an impeller mixer shows almost same trends even with higher mixing speed
than the TS samples with a turbine mixer. At low feed flow rate, i.e., 20 and
40 ml/min, the slope of regression line fit in IS samples shows lower magnitude
compared to TS samples. As in all cases (T, TS, and IS samples), the growth rate
190
shows almost constant behavior with constant feed flow rate meaning that the feed
flow rate is the major controlling variable for the growth rate.
Figure 5.18 illustrates the power law relation of nucleation rate (B0 ) to growth
rate (G). The analysis of the unseeded CMSMPR crystallizer with size-independent
Figure 5.18. Nucleation rate (B0 ) vs. growth rate (G) at steady state for IS
samples (see Table 5.3).
growth leads to an exponential CSD [2]. In terms of population density n(L) and
particle size L, this is expressed as:
L
n(L) = n0 exp (5.7)
L
= G is the product of the growth rate G and the mean
where the mean size L
residence time . The nuclei population density (n0 ) equals B0 /G, where B0 is the
nucleation rate. Alternatively, the distribution can be expressed in terms of the
cumulative number oversize distribution, N (L), given by:
191
L
N (L) = N0 exp (5.8)
L
where N0 = B0 is the total number of crystals. The material balance relation
links the parameters of the above distribution to the suspension density, MT , by
MT
= 3 = n0 (G )4 = N0 (G )3 (5.9)
6c kv
where c is the crystal density, kv the volumetric shape factor, and 3 the unnor-
malized third moment of the distribution [101]. Strictly, the term material balance
relation applies when the suspension density is related to the change between inlet
and outlet solution concentrations. The above relation is really a self-consistency
check; it tests that the measured quantities satisfy the third moment for the
expected exponential distribution. The last two equalities in Equation 5.9 result
from the direct integration of the distribution of Equation 5.7. The estimate of
B0 is correlated with the estimate of G, according to B0 G3 . Figure 5.18
shows the results of several independent estimates. The 3 power dependence is
obvious therefore proving the accuracies in CSD measurements and in analytical
data obtained from those CSD measurements.
The correlation analysis with one independent variable constant and the other
two are changing was performed to explore the details of main effects to the
dependent variables, i.e., n0 , B0 , G, and Ka , shown in Table 5.12. When looking
at individual effects of independent variables to products, IS samples follow almost
the same behavior as explained previously for TS samples except for the constant
mixing speed case which shows dominant effects of crystallization temperature on n0
in all cases. However, as mixing speed increases the effects of TC on n0 , B0 , and Ka
tend to decrease contrary to the case in TS samples which showed increasing effect of
TC (except for G) with respect to mixing speed. Overall, the effect of crystallization
temperature was evident at 40 ml/min and at 677.8 rpm (479 rpm). The effects of
feed flow rate dominated on dependent variables at 30 C (35 C) and at lower
mixing speed. The mixing speed effect was dominating at lower crystallization
temperature and higher feed flow rate (lower). The indications in parentheses
are for TS samples. All of above differences between TS and IS samples can be
192
attributed to the different types of mixer, i.e., turbine and impeller. However, the
difference in speed of mixing and concentration cannot be ignored.
As an illustration for the effects discussed previously, the main effect of mixing
speed and feed flow rate in Figure 5.19 show an increase in number of nuclei and
nucleation rate, but decrease in growth and aggregation rate. The rapid increase
in number of nuclei (n0 ) and nucleation rate (B0 ) with respect to increase in mixer
rpm is mostly due to secondary nucleation. The linear decrease in aggregation rate
(Ka ) and growth rate (G) with increasing mixer rpm is expected due to turbulent
mixing of the crystallizers contents resulting in more breakage and less aggregation.
Figure 5.20 shows a plot of nucleation rate (B0 ) vs. mean size of particles (Lmean ).
The mean size of particles was smaller and the slope of power law curve fit showed
more gradual in IS samples when compared to TS samples mainly due to different
type of mixing pattern, i.e., uniaxial and biaxial mixing.
3. the possibility that preordered species or surface nuclei may be removed from
the crystal surface by shear forces.
Mechanical stress can contribute to secondary nucleation and may also influence the
growth rate of large crystals. In Figure 5.21 the experimental population density
distributions are compared to illustrate the effect of different type of mixer on
resultant population number density with indications of the possible explanation
for the deviation from idealized MSMPR model, i.e., a solid straight line on this
plot. The experimental conditions were same at 30 C, 40 ml/min, and 480 rpm
193
(a)
(b)
Figure 5.19. Influence of mixing speed only (main effect) on number of nuclei(n0 ),
growth rate (G), and nucleation rate (B0 ), as well as aggregation rate (Ka ) at steady
state for IS samples. (a) IS 04, IS 05, and IS 06 samples and (b) IS 07, IS 08, and
IS 09 samples (see Tables 4.6 and 5.3 for experimental conditions and analytical
data).
with C02 feed solution concentration and with different types of mixer, i.e., turbine
and impeller. Although possible explanations are indicated in Figure 5.21 for the
194
Figure 5.20. Particle mean size (Lmean ) vs. nucleation rate (B0 ) at steady state
for IS samples (see Tables 4.6 and 5.3 for experimental conditions and analytical
data).
Figure 5.21. Influence of different type of mixer, i.e., turbine and impeller,
on population number density distribution at steady state. The experimental
conditions were same at 30 C, 40 ml/min, and 480 rpm with C02 feed solution
concentration except for the different types of mixer. The straight solid line
indicates the population density expected according to the idealized MSMPR
model.
Table 5.12. Correlation analysis of independent variables to the dependent variables at one
independent variable constant (other two are changing) for IS samples.
a
TC : Crystallization temperature ( C).
b
Q: Feed flow rate (ml/min).
c
MS: Mixing speed (rpm)
196
197
As a result of the work done in this project, it is clear that the models for
secondary nucleation and breakage are lacking a basis on first principles that is
needed for accurate prediction based upon first principles. However, the models
incorporated into the new OLI and Fluent software products to date are found
out to be useful for developing empirically based crystallization models that are
significantly advanced compared to previous crystallization models.
So far the models that have been implemented in the products of this project:
OLI Crystallization Compartment Modeler (CCModeler)
For crystallization validation, Table 5.13 presents inlet and outlet conditions
and compositions for T, TS, and IS samples involving three salt solution of CaCl2 ,
KCl, and NaCl. The precipitation analyses of inlet and outlet solutions are obtained
using OLI stream analyzer. The Figure 3.3 and Table 3.1 in Section 3.3.1 illustrate
one of solution concentration (C01 ). As the solution is cooled, NaCl is the first
to precipitate at a temperature of approximately 45, 35, and 31 C for C01 , C02 ,
and C03 , respectively. The next salt is a CaCl2 6H2 O staring to crystallize out at
approximately 28, 27, and 27 C for the same order of solution as before. The KCl
was not crystallized out at temperature above 20 C for all samples. The experi-
mental results confirmed that there was no CaCl2 crystallizing out at crystallization
temperature of 25 C or above. With the analysis being done, preliminary kinetic
model parameter estimates were determined for combined kinetic phenomena using
CCModeler by OLI systems. Parameter estimates were obtained by trial and error
while monitoring the following variables.
3. the lower and upper interval where the weight fraction was 1%.
4. suspension density.
Figures 5.22 through 5.24 show the measured vs. predicted discrete volume fraction
of particle size distributions as an example for each of T, TS, and IS samples.
201
The experimental conditions and results are given previously for each of sample
Figure 5.22. Measured and predicted volume fraction of particle size distribution
for T 16 sample considering nucleation, growth, aggregation, breakage kinetics (see
Table 5.13).
designs. From these very preliminary results, it appears that nucleation and growth
were the dominant influences for the crystallization of NaCl. The high degree of
correspondence in the downward slope between 100 and 200 m particle size range
was noticed for T and TS samples. The IS sample with higher mixing speed (481
rpm) requires much more rigorous works should be done in this model. This model
has not been accommodated in the different type of mixer and speed of mixing,
yet. Table 5.14 shows the estimated values for the kinetic model parameters. These
results and the ease of performing this preliminary analysis demonstrate the promise
of this modeling technology.
202
Figure 5.23. Measured and predicted volume fraction of particle size distribution
for TS 16 sample considering nucleation, growth, aggregation, breakage kinetics
(see Table 5.13).
Figure 5.24. Measured and predicted volume fraction of particle size distribution
for IS 16 sample considering nucleation, growth, aggregation, breakage kinetics (see
Table 5.13).
Table 5.13. Inlet and outlet conditions and composition for T, TS, and IS samples.
T TS IS
Inlet Outlet Inlet Outlet Inlet Outlet
Pressure (kPa) 101.325 101.325 101.325 101.325 101.325 101.325
Temperature (K) 343 303 343 303 343 303
Stream
Mass flow (g/min) 87.9024 87.9024 87.9024 87.9024 87.9024 87.9024
Volume Flow (ml/min) 60 59.0963 60 59.0628 60 59.0564
Mass flow (g/min) 88.6884 88.545 87.9024 87.8313 88.0944 88.0373
Liquid Mass density (g/ml) 1.4781 1.5 1.4650 1.4879 1.4682 1.4914
Volume flow (ml/min) 60 59.03 60 59.03 60 59.03
Mass flow (g/min) 0 0.1434 0 0.0711 0 0.0571
Solid Mass density (g/ml) 0 2.1638 0 2.1638 0 2.1638
Volume flow (ml/min) 0 0.0663 0 0.0328 0 0.0264
Water 50.63 50.84 51.69 51.91 51.44 51.66
Composition CaCl2 47.63 47.82 46.88 47.08 47.09 47.29
(wt%) KCl 1.07 1.07 0.93 0.93 1.03 1.03
NaCl 0.67 0.41 0.5 0.42 0.44 0.42
203
Table 5.14. Kinetic model parameter estimation for T, TS, and IS samples.
5.5 Summary
With the application of Hounslows solution into the measured CSD at 10 for
all the experimental runs, the number of nuclei (n0 ), growth rate (G), nucleation
rate (B0 ), and aggregation rate (Ka ) were obtained for each set of designs. However,
the aggregation of particles was noticed in size range less than 50 m in most of
crystallization experiments, therefore, the line fitting was unsuccessful with particle
size less than 50 m. The deviation of CSD from the linear line in large particle
size region due to particle breakage was also not included in this analysis.
To show the relative importance of independent variables, i.e., crystallization
temperature (TC ), feed flow rate (Q), and mixing speed (M S), affecting a crys-
tallization process, as well as the importance of their interactions to the research
experiments, the analytical data obtained with the application of Hounslows con-
stant rate equation was analyzed following the statistical design of experiments
according to factorial design analysis and other various statistical analyses. With
these analyses it was possible to explore the detailed crystallization phenomena
and to find out the controlling variables for the purpose of crystallization at certain
conditions.
CHAPTER 6
GENERAL CONCLUSION
6.1 Conclusions
The following general conclusions are drawn from the results of this research
project.
Reactor characterization
A comprehensive experiment was performed to characterize a laboratory scale
reactor and ultimately to find out a suitable condition for the crystallization system
built in our Lab.
1. The mean and variance of the residence time distribution for a stirred tank
deviating from ideal values were observed at low mixer rpm. As the mixer
rpm increased, the mean and variance of the residence time distribution
were observed approaching the ideal values1 .
2. For the nonbaffled tank, ideal behavior was observed at 100 rpm or an
mixer Reynolds number of 3,877 and above, while for the baffled tank ideal
behavior was being observed at 60 rpm or an mixer Reynolds number of
2,326 and above using a turbine. The rpms where ideal mixing occurred
take place for impeller Reynolds numbers in the transition range.
3. Predictions of the RTD and its mean and variance using CFD with a k
model of turbulence showed good agreement with experiment for
transitional turbulent flow regimes in the tank2 .
1
Appendix B.
2
The prediction of RTD to compare with experimental results was provided by an affiliation
of this research project team and with their contributions.
207
Crystallization of NaCl
1. The SEM figures with the aid of microscopic photos demonstrate the
detailed particle shape (octahedron) and size (1 to 300 m) to explain the
3
Section 2.5
4
Section 4.3.3.
5
URL: http://www.che.utah.edu/ring/CrystallizationWeb.
6
Section 3.3.1, 4.2.3, 4.4.1, and Appendix F.
7
Section 4.3.2.
208
2. Even with the careful treatments of slurry sample to measure the CSD in
LS230, the change in temperature of 2 to 5 C during the measurement was
enough to alter the particle characteristics by either growth or dissolution.
Therefore, a special technique was required to make the particles stable and
the CSD measurements successful. The application of CdCl2 impurity to
limit zero-growth rate and to stabilize NaCl particles was discussed with
underlying theory and showed promising result of CSD measurement
without changing the characterization of particles9 .
3. The CSDs measured and microscopic photos taken at every mean residence
time during the crystallization showed distinct characterizations, i.e., the
CSD at the beginning of crystallization and the evolution of CSDs with
time, of NaCl crystals depending on various experimental parameters, such
as, solution concentration, crystallization temperature, feed flow rate,
mixing speed, and type of mixer10 .
5. The factorial design analysis to nucleation rate, growth rate and aggregation
rate without considering experimental results at 25 C showed the
importance of feed flow rate and mixing speed as main effects and their
interaction to the resultant analytical data16 .
12
Section 4.5.7.
13
Section 5.1.
14
Section 5.1.
15
Section 2.7.2 and 5.3.
16
Section 5.3.
210
6. More rigorous analyses for T, TS, and IS samples were performed to show
how the independent variables are interacting together or alone onto the
product variables to find out the controlling variables at certain condition of
crystallization and also, to show how the role of independent variables on
products are changing with change in crystallization conditions17 .
9. As a result of the work done in this project, it is clear that the models for
secondary nucleation and breakage are lacking a basis on first principles that
is needed for accurate prediction based upon first principles. However, the
models incorporated into the new OLI and Fluent software products to date
are found out to be useful for developing empirically based crystallization
models that are significantly advanced compared to previous crystallization
models20 .
17
Section 5.3.1, refsec:aofts, and 5.3.3.
18
Section 5.3.2.
19
Section 4.5.6 and 5.3.4.
20
Section 5.4.
211
6.2 Recommendations
1. The analytical values, i.e., number of nuclei (n0 ), growth rate (G),
nucleation rate (B0 ), and aggregation rate (Ka ), obtained with the
application of Hounslows model are not the absolute ones, but they tend to
show accurately enough the possible phenomena for a crystallization in a
CMSMPR crystallizer. However, much more rigorous work should be done
to account for the breakage and secondary nucleation into the model.
2. The ex situ CSD measurements performed in this project account for the
NaCl particles with size ranging from 1 to 300 m with currently available
particle size analyzer. The treatment for NaCl particles to stabilize with the
application of Cd2+ impurity was required for this ex situ CSD measurement
due to the different environments. For the study of crystallization however,
it is important to be able to monitor the reactions and kinetics of
crystallization with high accuracy without any alterations and modifications
to the crystallization phenomena in a CMSMPR crystallizer. With current
advance in extracting large amounts of information from a single
multiwavelength measurement, the in situ application of multiwavelength
ultraviolet/visible (UV-VIS) spectroscopy could be one of possible
candidates with added benefits of being rapid, inexpensive, and relatively
simple to operate; the distribution of intensities as a function of wavelength
depends on the size, shape, and chemical composition of the sample.
APPENDIX A
Figure A.1. Geometry of the laboratory scale crystallizer (1.4 ` working volume)
with baffles and a turbine.
wall of the crystallizer, a 6-bladed flat disc turbine (Rushton type) or 6-bladed
213
pitch blade impeller and three internal pipes, two feed ports and a thermal well
protruding into the reactor from its top. One of the feed tubes dead ends at the
approximate height of the impeller and the other is bent to feed directly below
the impeller. Also protruding ever so slightly into the top of the liquid are a
pH probe, conductivity probe, thermocouple, potassium specific ion electrode and
level indicator. All of these instruments as well as the feed pump, product pump,
the jacket feed temperature, thermal well temperature and the rpm and torque on
the stirrer are connected to an OPTO 22T M data acquisition and control system
collecting data at time intervals of 1 s. The crystallizer control system has the
capability of controlling the feed source and crystallizer temperature to 0.25 C,
the feed and product flow rate to 0.5 ml/min, liquid level to 3.9 mm and mixing
speed to 3 rpm.
Figures A.1 through A.3 show crystallizer geometry with baffles and a turbine
in place along with the location of inlet and outlet. The scales in figure are all in
mm. Also the dimension of standard baffle width 1/12 of the vessel diameter is
given in Table A.1 along with the dimension of mixer. Figure A.4 and A.5 show
stirrer geometry of a 6-bladed flat disc turbine (Rushton type) and a 6-bladed pitch
blade impeller.
214
Figure A.4. Dimension and configuration of a 6-bladed flat disc turbine (Rushton
type).
218
REACTOR CHARACTERIZATION
These higher parameter models consider by-pass, dead zones, recycle, ideal back
mixed regions, and plug flow regions. These models have been used to predict
the nonideal RTD [87, 93, 102, 103, 104, 105, 106, 107] for the performance of
reactors operating with various chemical reactions [108, 109, 110] and nonideal
reactor scale-up [104, 111].
This work measures the RTD of a laboratory reactor both with and without
baffles and makes comparisons of these measurements with predictions for the RTD
using computational fluid dynamics (CFD) with a k turbulence model coupled
with a dynamic two-species, fluid and tracer, mass balance operating with a step
change of the concentration of tracer in the feed.
The ratio of liquid level to the tank diameter (H/D) of the vessel in Figure A.1
is 1.27 that is within the general range of 1.0 to 1.5 for most industrial stirred
tank precipitator [20, 86]. To reduce the energy input to the system while main-
taining mixing uniformity, a standard baffle design [87] was used consisting of four
flat vertical plates, radially-directed (i.e., normal to the vessel wall), spaced at
90 around the vessel periphery running the length of the vessels straight side.
Standard baffle widths are between 1/10 and 1/12 of the vessel diameter (D/10 or
D/12), see Table A.1 in Appendix A) for details. The gaps with the vessel wall and
base are left to allow the flow to clear the baffles. Recommended gaps are equal to
1/72 of the vessel diameter (D/72) between the baffles and the vessel wall, and 1/4
to one full baffle width between the bottom of the baffles and the vessel base. More
of the detailed baffling information could be found in Myers, et al. [88]. Stirring
was achieved by a 6-bladed flat disc turbine (Rushton type) or a 6-bladed pitch
blade impeller. The dimensions and configurations of a turbine and an impeller are
shown in Figures A.4, A.5, and Table A.1 in Appendix A.
The rpm of peristaltic pumps was controlled by an OPTO 22T M control system
establishing the feeding and removal of product solution from the tank. Flow rate
was obtained by the careful and repeated calibration of the speed of pump and tube
size to ensure the desired feed and withdrawal rates as well as a constant residence
time. The OPTO 22 system also measured the on-line sensor quantities; tempera-
ture, pH, absorbance, and conductivity, as well as potassium ion concentration. All
of these measurements were used to monitor the concentration of the tracer inside
and exit of the reactor. The calibration for all of these systems is done before each
run to assure measurement accuracy.
The RTD measured was the combination of process dynamics and sensor dynam-
ics [112]. The sensor dynamics are much faster than that of the process. The sensor
time constants are the following: conductivity, which has a measurement time of 1
s; K + and pH electrodes, which have measurement times of 4 8 s; absorbance,
which has a measurement time related to the flow rate through the absorbance
222
Qa MS b tm c Vd V /Q Re e
/tm f
(ml/min) (rpm) (min.) (ml) (min.) FT Mg
0 18.90 0.636 1,290 11.73 0
10.2 15.94 0.665 1,290 11.73 395
30 12.28 0.964 1,290 11.73 1,163
49.7 11.75 0.978 1,290 11.73 1,927
110 1,676
69.5 11.66 1.011 1,290 11.73 2,695
90.6 11.61 1.021 1,292 11.75 3,513
190 11.65 1.003 1,294 11.76 7,367
286.7 11.76 1.026 1,298 11.80 11,120
0 8.09 0.746 1,290 5.73 0
9.8 7.26 0.819 1,290 5.73 380
29.7 6.26 0.915 1,290 5.73 1,152
50.2 5.93 0.965 1,290 5.73 1,946
225 3,428
70 5.96 0.988 1,290 5.73 2,714
91.6 5.67 1.011 1,292 5.74 3,551
190.6 5.78 1.015 1,294 5.75 7,390
288 6.10 1.007 1,298 5.77 11,170
0 5.32 0.863 1,290 4.37 0
11 4.70 0.939 1,290 4.37 426
30.2 4.24 1.017 1,290 4.37 1,171
50.5 4.41 0.974 1,290 4.37 1,958
295 4495
69.9 4.42 1.016 1,290 4.37 2,710
90 4.32 1.047 1,292 4.38 3,480
188.5 4.27 1.024 1,294 4.39 7,308
290 4.42 1.041 1,298 4.40 11,240
Table continues on next page.
a
Q: flow rate.
b
MS : mixer speed.
c
tm : mean residence time.
d
V : reactor working volume.
e
Re: Reynolds number.
f
F T : feed tube.
g
M : mixer.
abruptly in an instantaneous rise at time zero and decay exponentially with time
thereafter. The RTD is obtained from these experimental data by normalization in
typical manner [93].
224
a
Q: flow rate.
b
MS : mixer speed.
c
tm : mean residence time.
d
V : reactor working volume.
e
Re: Reynolds number.
f
F T : feed tube.
g
M : mixer.
C(t) C(t = 0)
E(t) = R (B.1)
0
[C(t) C(t = 0)]dt
where C(t) represents the concentration (or temperature) of tracer, e.g., Figures B.1
and B.2. An example of the RTD is plotted in Figure B.3. Even for this good mixing
case there are deviations from ideal mixing at short residence times where the initial
rise is not immediate and is not a smooth curve. This figure also contains a plot of
the ideal RTD for the stirred tank for comparison purposes. The mean residence
time (tm ) was calculated by integrating the RTD as follows [91],
Z
tm = tE(t)dt (B.2)
0
The variance, or square of the standard deviation of the RTD is calculated using:
225
Qa MS b tm c V V /Q Re d
/tm MPN g
(ml/min) (rpm) (min.) (ml) (min.) FT e Mf
0.0 106.66 0.741 1400 82.35 0
8.6 111.73 0.819 1,400 82.35 376 1,194,000
28.5 87.51 0.928 1,400 82.35 1,148 48,290
47.8 86.40 0.939 1,400 82.35 1,904 10,990
17 307
67.1 86.84 0.936 1,400 82.35 2,714 3,649
87.5 81.77 0.937 1,400 82.35 3,416 1,668
185.1 82.87 0.938 1,400 82.35 7,289 249
283.8 81.98 0.916 1,400 82.35 11,040 108
0.0 50.32 0.766 1,400 40.58 0
8.8 54.08 0.833 1,400 40.58 446 729,500
28.5 44.97 0.985 1,400 40.58 1,221 40,370
48.2 44.08 1.036 1,400 40.58 1,973 9,866
34.5 614
41.1 37.71 0.930 1,400 40.58 2,710 3,666
87.4 40.79 0.931 1,400 40.58 3,551 1,524
185.3 40.77 0.906 1,400 40.58 7,367 243
283.8 40.29 0.885 1,400 40.58 11,090 107
0.0 31.37 0.866 1,400 27.18 0
9.5 30.46 0.838 1,400 27.18 446 729,500
29.5 27.68 0.896 1,400 27.18 1,221 40,370
48.6 27.26 0.912 1,400 27.18 1,973 9,866
51.5 614
67.7 28.38 0.952 1,400 27.18 2,710 3,666
88.0 28.06 0.948 1,400 27.18 3,551 1,524
186.2 28.05 0.940 1,400 27.18 7,367 243
284.9 27.07 0.924 1,400 27.18 11,090 107
a
Q: Flow rate
b
MS : Mixer speed
c
tm : Mean residence time
d
Re: Reynolds number
e
FT : Feed tube
f
M : Mixer
g
MPN : Mixer power number
Z
2
= (t tm )2 E(t)dt (B.3)
0
The magnitude of this 2nd moment is an indication of the spread of the RTD.
226
Qa MS b tm c V V /Q Re d
/tm MPN g
(ml/min) (rpm) (min.) (ml) (min.) FT e Mf
0 111.82 0.752 1,400 78.65 0
9.1 101.73 0.767 1,400 78.65 376 1,194,000
28.4 85.18 0.962 1,400 78.65 1,148 48,290
17.8 47.9 80.44 0.987 1,400 78.65 307 1,904 10,990
67.1 81.86 1.021 1,400 78.65 2,714 3,649
86.8 78.94 0.990 1,400 78.65 3,416 1,668
185.0 78.39 0.990 1,400 78.65 7,289 249
0 50.32 0.766 1,400 37.33 0
9.0 48.05 0.835 1,400 37.33 446 729,500
29.1 41.91 0.907 1,400 37.33 1,221 40,370
37.5 47.7 39.10 0.940 1,400 37.33 614 1,973 9,866
67.6 37.71 0.969 1,400 37.33 2,710 3,666
86.8 37.87 0.992 1,400 37.33 3,551 1,524
185.1 37.44 0.982 1,400 37.33 7,367 243
0 29.29 0.871 1,400 25.09 0
9.4 28.87 0.834 1,400 25.09 446 729,500
28.6 29.06 0.897 1,400 25.09 1,221 40,370
55.8 47.4 27.31 0.947 1,400 25.09 614 1,973 9,866
67.1 25.73 0.960 1,400 25.09 2,710 3,666
87.0 25.30 0.978 1,400 25.09 3,551 1,524
185.1 25.09 0.985 1,400 25.09 7,367 243
a
Q: Flow rate
b
MS : Mixer speed
c
tm : Mean residence time
d
Re: Reynolds number
e
FT : Feed tube
f
M : Mixer
g
MPN : Mixer power number
In Figure B.4, the mean resident time obtained from the experimental data
is plotted as a function of mixer rpm for various feed flow rates (high flow rate
case) for the tank without baffles and a turbine. Figures B.5 and B.6 show the
experimental data plotted as a function of mixer rpm for various feed flow rates
(low flow rate case) for the tank with baffles and a turbine and for the tank with
227
baffles and an impeller, respectively. Here we see that the mean residence time
decreases with increasing rpm until a constant value (solid line) is reached. This
constant value is approximately Vtank /Q, the space-time for the reactor, where Vtank
is the reactor volume and Q is the volumetric flow rate. The error in the calculation
of the mean residence time, tm , is 8% of Vtank /Q for all experiments and was
determined by experiments done in duplicate and some in triplicate. These errors
are due to the variations in tank volume caused by the liquid level control (3.9
mm corresponding to 41 ml) and to the variations in the flow rate caused by
a small calibration drift and pulsating peristaltic pump flow. By comparison of
these three figures (Figures B.4, B.5, and B.6), one can easily see the difference in
RTD as a function of mixer rpm, flow rate, and type of mixer. In high flow rate
case (Figure B.4), the constant value (Vrank /Q) has reached at below 100 rpm of
mixing speed compared to above 100 rpm with low flow rate case (Figure B.5). Also
228
due to the different mixing pattern between a turbine and an impeller, Figure B.6
with axial mixing shows smooth decrease in tm with little longer space time while
Figure B.5 shows fluctuation of tm due to bi-axial mixing.
In Figures B.7 and B.8, the variance of the RTD divided by the space-time is
plotted as a function of mixer rpm for various feed flow rates for the tank without
baffles and with a turbine and for the tank with baffles and with an impeller,
respectively. The error in the variance of the residence time divided by the mean
residence time is 12% at all rpms and is larger than that of the mean residence
time because two integrals with their inherent errors are involved in the variance
divided by mean residence time calculation. For an ideal reactor the value of /tm
should be 1.0, which is observed, within the experimental error, at higher impeller
rpms. With the Figures B.7 and B.9 one can see that mixer speed of 100 rpm was
enough to approach perfect mixing for the unbaffled tank. With baffles, the results
229
Figure B.4. Comparison of tm vs. mixer rpm at different flow rate for 1,300 ml
laboratory reactor without baffles and with a turbine.
230
Figure B.5. Comparison of tm vs. mixer rpm at different flow rate for 1.4 `
laboratory reactor with baffles and a turbine.
Figure B.6. Comparison of tm vs. mixer rpm at different flow rate for 1.4 `
laboratory reactor with baffles and an impeller.
are shown in Figures B.8 and B.10 (low flow rate case). With baffles (low flow rate
case), the results are similar to those without baffles (high flow rate case), however,
231
Figure B.7. Comparison of /tm vs. mixer rpm at different flow rates for 1.3 `
laboratory reactor without baffles and with a turbine.
Figure B.8. Comparison of /tm vs. mixer rpm at different flow rate for 1.4 `
laboratory reactor with baffles and an impeller.
Figure B.9. Comparison of tm /(V /Q) vs. mixer rpm at different flow rate for 1.3
` laboratory reactor without baffles and with a turbine.
The high flow rate case with baffles and the low flow rate case without baffles could
Figure B.10. Comparison of tm /(V /Q) vs. mixer rpm at different flow rate for
1.4 ` laboratory reactor with baffles and a turbine.
Figure B.11. Comparison of tm /(V /Q) vs. mixer rpm at different flow rate for
1.4 ` laboratory reactor with baffles and an impeller.
flow rate of 40 ml/min, and 0.17 m/s for a feed flow rate of 20 ml/min. The 0.17 m/s
average feed velocity is larger than the impeller tip velocity in the 0, 20, 40, and 80
rpm simulations and the 0.33 m/s average feed velocity is larger than the impeller
tip velocity in the 0, 20, and 40 rpm simulations. The tank output was given a
pressure outlet boundary condition with no slip. The walls of the tank, baffles, and
the other tank internals were assigned standard wall function boundary conditions,
the top surface was assigned a symmetry boundary condition, the surface of the
moving zone was assigned an interface boundary condition, and the surface of the
impeller was assigned a (stationary) standard wall function inside the mesh that is
rotating. The model was allowed to run using double precision calculations until
all the scaled residuals reached a value of 104 . This level of convergence took
1, 100 iterations. The resulting velocity profile is given in Figure B.12 and shows
that the steady-state solution contains two major circulation cells - one above and
one below the impeller. The overall flow pattern in each circulation cell is that of
Figure B.12. Velocity vector profile for turbulent flow in the 1.4 ` laboratory
reactor operating at 40 ml/min feed flow rate and a mixer speed of 80 rpm.
235
a helical path on the surface of a torus, circulating from the impeller to the tank
wall, up or down the tank wall (up for the upper circulation cell and down for
the lower circulation cell), back into the center of the tank and into the impeller
again. This overall flow pattern is interrupted by the flow around and behind the
baffles. The flow behind the baffles plays an important role in passing fluid from
the top circulation cell to the bottom circulation cell as there is a minor circulation
cell of cylindrical form behind each baffle in which the material can enter from the
top circulation cell and exit into the bottom circulation cell or vice versa. There
is also some mixing of material between the two circulation cells at the plane of
the impeller. As the flow moves radially out some of the fluid is exchanged from
the upper circulation cell to the lower circulation cell and vice versa. This later
mechanism operates in the tanks both with and without baffles.
The behavior of the tracer is modelled by fixing the fluid flow field and adding a
user defined scalar to model the concentration of tracer with a diffusion coefficient of
105 cm2 /s. No source term was used for the user-defined scalar nor any turbulent
diffusivity. The boundary conditions for the tracer consist of a feed of mass fraction
of 1.0 and an output of whatever concentration is at the outlet tube located at
the top of the tank. All the tank internals use a zero flux boundary condition.
The initial condition for this unsteady simulation was to fill the tank with zero
concentration of tracer and feed the tank with a solution with a mass fraction of
1.0 from the feed tube located just below the impeller. The simulation was allowed
to precede using time steps of 20 s with the same convergence criterion. An overall
mass balance was verified to be accurate to within the convergence criterion. The
convective flux of the tracer at outlet is collected from this simulation and plotted
against time as shown in Figure B.13 where runs for several rpm values at a flow
rate of 40 ml/min are plotted. Here we see that initially the concentration of tracer
is zero and increases with time until the tracer concentration approaches that of the
feed, a value of 1.0. This type of plot is characteristic of the response of a stirred
tank reactor to a step input. The residence time distribution is obtained from this
plot of tracer concentration by differentiating the curve with respect to time, i.e.,
236
Figure B.13. Output concentration as a function of time for the Fluent prediction
of the residence time distribution for various mixer speeds (i.e., 0, 20, 40, 80 and
200 rpm) and a feed flow rate of 40 ml/min.
dC(t)
E(t) = (B.4)
dt
The RTD determined in this way is normalized since the feed tracer concentration
was 1.0. This step input is a different way to determine the RTD than that
used in the experiments, which was a pulse input. In attempting to use a pulse
input method in the simulations, the concentrations were very small and round
off errors were sufficiently large to invalidate the overall mass balance for the
reactor significantly beyond the convergence criterion. Mass balance errors have
been shown to cause drastic errors in RTD measurements [114]. For this reason,
the step method of determining the RTD is used in these simulations.
A comparison of the Fluent-predicted RTD and that measured experimentally
is shown in Figure B.14 for different impeller rpm values for 40 ml/min flow rate
with the baffled tank. The predictions show similar trends in that there is a delay
before the RTD increases after time zero and there is an initial roughness in the
curve. The RTD then decreases exponentially with time. The prediction typically
237
(a)
(b) (c)
over predicts in the initial spike or two in the first tens of seconds, the experimental
data, and the perfectly mixed theoretical curve.
The predictions for various flow rates and impeller rpm values were analyzed
for the mean residence time and the variance of the RTD. These results (filled
238
data points) are also plotted in Figures B.15 and B.16 for comparison with the
experimental data. The error bars associated with the theoretical points were
Figure B.15. Comparison of experiment and Fluent simulation result for mean
residence time ( ) vs. mixing speed (rpm) at various flow rate for a 1.4 ` laboratory
reactor with baffles and a turbine. The larger filled symbols are from the Fluent
simulation results.
Figure B.16. Comparison of experiment and Fluent simulation result for standard
deviation (/tm ) of RTD vs. mixing speed (rpm) at various flow rate for a 1.4 `
laboratory reactor with baffles and a turbine. The larger filled symbols are for the
Fluent simulation results.
fluid flow profile will depend on the location of the vanes of the stopped impeller
relative to the other tubes inside the tank and this location may change until the
flow becomes stabilized. The location of the impeller vanes with respect to the
feed tube was not noted in the experiments and could not be used for a fixed solid
boundary condition in the CFD prediction. As a result the Fluent CFD prediction
is inaccurate when the impeller is not rotating.
No simulations of the unbaffled tank were performed. We were focusing our
work on the baffled tank since it is most commonly used in industry. Furthermore,
our nonbaffled tank has three inlet tubes in an odd geometry that act as crude and
poorly characterized baffles. In retrospect, due to more deviations in concentration
from an ideal stirred tank, the simulation of the nonbaffled tank may have been a
better choice to validate the CFD predictions of concentration. We did not predict
the pulse concentration with time leaving the reactor in our simulations but the
step concentration. So direct comparisons would not have been possible. In further
240
B.1.4 Conclusion
The mean and variance of the residence time distribution for a stirred tank
deviate from ideal values at low impeller rpm. As the impeller rpm is increased, the
mean and variance of the residence time distribution approach the ideal values. For
the nonbaffled tank ideal behavior is observed at 100 rpm or an mixer Reynolds
number of 3,877 and above, while for the baffled tank ideal behavior is observed
at 80 rpm or an mixer Reynolds number of 2,326 or above using a turbine and
an impeller. With an impeller, a smooth decrease in constant space time with
respect to mixer speed was observed while with a turbine showed fluctuation before
reaching the constant values due to different pattern of mixing. Predictions of the
RTD and its mean and variance using CFD with a k model of turbulence show
good agreement with experiment for transitional turbulent flow regimes in the tank.
B.1.5 Nomenclature
C(t): Concentration as a function of time
E(t): Residence time distribution function
t: time (sec)
Vtank : Volume of the reactor
Q: Flow rate into and out of the reactor (ml/min)
APPENDIX C
OPERATING PROCEDURE
temperature at the end-tip of feed-point inside the crystallizer. Also, the crystal-
lizer control system has the capability of controlling the feed source and reactor
temperature to 0.25 C, the feed and product flow rate to 0.5 ml/min, liquid
level to 1.5 mm and mixing speed to 3 rpm.
The power input per unit volume, crystallizer geometry and layout should be
carefully considered to achieve the vessel to be well mixed so that powders are
suspended uniformly. An empirical correction for the additional power needed for
homogeneity, proposed by Yamasaki et al. [85], was incorporated to estimate the
minimum power requirement for our 1.4 ` working volume crystallizer vessel with
the configuration of baffles and a turbine or an impeller. The geometry of laboratory
scale crystallizer was shown in Figure A.1, see Appendix A. This figure presented
the strategic placement of all the flow lines and baffles, also the position of mixer.
The feed is injected 4/5 of the way down the crystallizer to aid well mixing before
going out to the outlet at the top surface.
The ratio of liquid level to the crystallizer diameter (H/D) of the vessel is 1.27
which is within the general range of 1.0 to 1.5 for most industrial precipitator [20,
86]. To reduce the energy input to the system while maintaining uniformity, a
standard baffle design [87] is used. The detailed baffling information could be
found in Myers et al. [88]. The dimension of standard baffle was given in Table A.1
along with the dimension of impeller.
The continuous stirred crystallizer is used ubiquitously in the chemical process
industry for mixing, reaction and crystallization. The mixing in a continuous stirred
crystallizer is often not ideal. The RTD is one of the ways to characterize the
non-ideal mixing in the crystallizer. Comparison of the measured RTD with that
of an ideal reactor allows the process engineer to diagnose the ills of the crystallizer
and mixer design. The engineer can then use an appropriate mixing model for
the crystallizer in combination with the kinetics of the reaction to be performed in
the crystallizer to develop an appropriate model for the reactor [91]. Experimental
procedure and result with modeling could be found elsewhere [92].
243
a
As described in Section 3.1.1.
b
The analysis was done by the Dow Chemical, Ludington, Michigan. The
solution samples were taken at 50 C when the solution was drummed.
c
Solution samples were taken at 70 C after heating the solution more than
24 hrs. Analysis was done by AAS and ICP as described in Section 4.2.3
d
pH of solution was measured at 70 C with a pH probe using temperature
compensation.
244
chosen to study in this work is the purification involving precipitation of NaCl from
the aqueous solution mixture of CaCl2 , KCl and NaCl based upon its characteristic
nucleation and crystal growth mechanism with less aggregation as described in
Section 2.2.
Samples of these starting solutions for the precipitation of NaCl were extracted
in the following manner: First, the density and pH of the starting stock solution were
measured to verify that a representative sample was being taken from the storage
drum (D1). The solution density was measured as described in Section 3.1.1. A 100
ml sample was also taken from this stock solution and stored for future analysis.
These values were monitored in order to maintain daily consistency in the sampling
of the stock solution.
Equipment
Equipment Specifications
Symbol
B Thermo NESLABT M EX-510D Bath with Digital Controller
Cole ParmerT M Conductivity Meter with TopacT M 4 Electrodes
C
Epoxy body Platinum plates Conductivity Probe
C1 Preheating Solution Container (4 `)
C2 Preheating Solution Container (4 `)
Continuous Stirred Jacked Tank Reactor (1.4 `) with
CSTR
Turbine and Baffles
D1 Storage Drum (200 `)
D2 Overflow Collection Drum (30 `)
DH Thermostat Controlled Drum Heater (Off to 290 C)
ApplikonT M Mixer operating between 1 1250 rpm
M
with Digital Controller
OPTO 22T M Controller. Reads and Controls Temperature,
OPTO
pH, Conductivity, Level Indicator, Mixer Speed, Pump Speed
P1 Source Refill Pumpa
P2 Preheating Solution Refill Pump
P3 Inlet Solution Refill Pump
P4 Outlet Solution Refill Pump
P5 Rotary Vane Vacuum Pump with 1/2 HP
PC Personal Computer used to commmunicate with and control OPTO 22
OaktonT M pH/mV/Rel mV/ C Benchtop Meter
PH
with Cole-ParmerT M precision pH electrode, epoxy body
PH1 Thermo Heating or Cooling Bath for Crytallizerb
PH2 Thermo Heating or Cooling Bath for PH1
Table continues on next page.
a
P1, P2, P3, and P4 use MasterflexT M Console Analog L/S Pump Drive operating between
6 600 rpm using a Masterflex L/S Easy-LoadT M II Pump Head with NorpreneT M tubing.
b
PH1 and PH2 use PolyScienceT M Digital Temperature Controller with a solution of water
and ethylene glycol mixture controlling temperature between 30 and 130 C.
C.4.2 pH Calibration
pH meter and probe with automatic temperature compensation were calibrated
at 25 C using OrionT M buffer solution pH of 4.00.01, 7.010.01, and 10.000.02,
then incorporation with OPTO measuring DC mV from pH meter once again they
246
1. Connect the indicator into OPTO 22 unit to read the current output.
2. Verity that as the distance from the liquid to the indicator increases, the
current signal decreases.
3. Verity that as the distance from the liquid to the indicator decreases, the
current signal increases.
A calibration result with 99.99% accuracy was obtained with water as a solution in
a crystallizer without mixing.
1. Before starting any equipments make sure that the crystallizer is filled with
1.4 ` of distilled water.
4. Turn on Pumps (P1, P2, P3 and P4) and verify that these pumps are all in
the external mode.
5. Fill the Thermo NESLABT M EX-510D bath (B) with distilled water.
6. Turn on Thermo NESLABT M EX-510D bath with digital controller (B) and
set the operating temperature to 75 C.
1
Brand name of Cole-parmer.
248
10. Empty the water contents of the 1.4 ` crystallizer as soon as the
temperature reaches about 50 C. Refill with 1.4 ` of warm distilled water
and repeat this flushing step two additional times. This cleaning step is to
remove any contaminations as much as possible without dismounting the
crystallizer. On the last flush of the crystallizer leave the water in the
crystallizer and momentarily start the overflow pump (P4) so that the liquid
level is just at the bottom of the overflow pipe.
11. Wait for about 3 hrs to reach the desired temperature for all equipments.
1. When the desired temperatures for all equipments are reached after the last
steps in section C.5, turn on P1 to transfer solution from D1 to C1.
3. When the solution in C2 reaches same level with solution in B, stop P1 and
P2.
5. When the solution in crystallizer reaches 1,400 ml, turn on P4 and open V1
valve such that the solution overflows to D2 storage container.
6. Turn on mixer (M) to a desired rpm and wait about 30 min to have steady
state at 70 C.
The controlling and reading of all equipments can be done with computer-interfaced
experimental set-up as showed in Figure 2.1 and explained in Section 2.5 except for
the valves which require manual controls.
1. Make sure that the temperature of slurry sampling device (S1) is kept at the
same as that of crystallization.
2. One minute before the required sampling time, turn on vacuum pump (P5)
and open valve V2 such that the sampling line is ready.
5. The filter sample was labelled CMSMPR mm/dd/yyyy and kept in dry for
further analysis such as SEM, EDX, and XRD. Also the filtrate solution was
labelled CMSMPR mm/dd/yyyy AF (where AF means after filtration) and
kept for ICP and AAS analysis.
7. After finishing all the above steps, follow the shut down procedure in C.7.
2. One minute before the required sampling time, turn on vacuum pump (P5)
and open valve V2 such that the sampling line is ready. Make sure valve V4
is closed.
4. Place an extra bottle in S1 and lift the sampling tube out from the solution
in the crystallizer and then open valve V3 quickly such a way that the slurry
inside the tube can be removed for the next run.
5. Remove the bottle from S1, empty and flush it with distilled water for
couple of times.
6. Remove any moisture inside the bottle with lint free paper and then place
the bottle in S1 for the next run. This should be done at least 5 min. before
the next sampling so that the temperature of bottle is kept at the same as
that of crystallization.
252
2. Add small amount of highly concentrated CdCl2 solution into the saturated
solution making 100 ppm of Cd+ . This solution was used to stabilize
particles during the CSD measurement. The details with underlying theory
for the application of particle stabilization refer to Appendix F [117].
3. When LS230 is ready after the startup procedure (Section E.1), flush two
times with the saturated solution. After the last flush, fill the Cd+ modified
saturated solution ( 150 ml) and follow the sample analysis procedure in
Section E.2. This should be done at least 15 min. before the required time
for every CSD measurements. During this step, pay every attention to
remove bubbles from the solution inside LS230 performing de-bubbling
because bubbles are highly detrimental to the CSD measurement giving a
false result.
4. The slurry sample from step 3 in Section C.6.3.1 was added instantly to the
solution in LS230 right ahead of CSD measurement to obtain an allowable
concentration of particles. Extreme care should be taken to assure that the
temperature for sampling, diluting, and measuring of sample must be the
same as the slurry in the crystallizer at all time.
3. Empty both preheating solution containers (C1 and C2) into the storage
drum (D1) and drain water in B.
4. The reactor contents are emptied into an overflow collection drum (D2) by
using an external tube and a pump.
5. Fill the reactor with distilled water and follow the cleaning step 10 in C.5.
6. Shut down OPTO and PC. Make sure that all hardware are turned off.
7. Transfer the contents from D2 into D1 and keep the solution in D1 for more
than 12 hr for the next run. The temperature of solution in storage drum
(D1) is kept at 70 75 C with the thermostat controlled drum heater (DH).
8. Make sure all the solution flow lines are cleaned with distilled water. Keep
them clean and dry for the next run.
Figure C.1. Operating flow and control diagram for a CMSMPR crystallization system.
254
APPENDIX D
CHARACTERIZATION OF NaCl
CRYSTALS
Figure D.1. A plot of differential Number % vs. particle size, L (m). After 1
hr from starting CMSMPR crystallizer. Note: scale 10 = 130 m in microscopic
photo.
256
Figure D.2. A plot of population density distribution vs. particle size, L (m).
After 1 hr from starting a CMSMPR crystallizer.
For the experimental number and condition, refer to Table 4.4 through 4.6 in
Section 4.3.3. In 3D plot of PSDs, x = particle diameter (m), y = mean residence
time ( ), and z = differential volume (%). For all experiments the volume of
crystallizer was 1400 ml.
Figure D.3. A plot of differential Number % vs. particle size, L (m). After 6
hr from starting CMSMPR crystallizer. Note: scale 10 = 130 m in microscopic
photo.
(a) (b)
(c) (d)
(e) (f)
Figure D.4. 3D plot of PSDs, (a) and (b) for TS 04, (c) and (d) for TS 05, and
(e) and (f) for TS 06.
259
(a) (b)
(c) (d)
33
(e) (f)
Figure D.5. 3D plot of PSDs, (a) and (b) for TS 07, (c) and (d) for TS 08, and
(e) and (f) for TS 09.
260
(a) (b)
(c) (d)
(e) (f)
Figure D.6. 3D plot of PSDs, (a) and (b) for TS 10, (c) and (d) for TS 11, and
(e) and (f) for TS 12.
261
(a) (b)
(c) (d)
(e) (f)
Figure D.7. 3D plot of PSDs, (a) and (b) for TS 13, (c) and (d) for TS 14, and
(e) and (f) for TS 15.
262
(a) (b)
(c) (d)
(e) (f)
Figure D.8. 3D plot of PSDs, (a) and (b) for TS 16, (c) and (d) for TS 17, and
(e) and (f) for TS 18.
263
(a) (b)
(c) (d)
33
(e) (f)
Figure D.9. 3D plot of PSDs, (a) and (b) for TS 19, (c) and (d) for TS 20, and
(e) and (f) for TS 21.
264
(a) (b)
(c) (d)
(e) (f)
Figure D.10. 3D plot of PSDs, (a) and (b) for TS 22, (c) and (d) for TS 23, and
(e) and (f) for TS 24.
265
(a) (b)
(c) (d)
(e) (f)
Figure D.11. 3D plot of PSDs, (a) and (b) for TS 25, (c) and (d) for TS 26, and
(e) and (f) for TS 27.
266
(a) (b)
(c) (d)
(e) (f)
Figure D.12. 3D plot of PSDs, (a) and (b) for IS 04, (c) and (d) for IS 05, and
(e) and (f) for IS 06.
267
(a) (b)
(c) (d)
(e) (f)
Figure D.13. 3D plot of PSDs, (a) and (b) for IS 07, (c) and (d) for IS 08, and
(e) and (f) for IS 09.
268
(a) (b)
(c) (d)
(e) (f)
Figure D.14. 3D plot of PSDs, (a) and (b) for IS 10, (c) and (d) for IS 11, and
(e) and (f) for IS 12.
269
(a) (b)
(c) (d)
(e) (f)
Figure D.15. 3D plot of PSDs, (a) and (b) for IS 13, (c) and (d) for IS 14, and
(e) and (f) for IS 15.
270
(a) (b)
(c) (d)
(e) (f)
Figure D.16. 3D plot of PSDs, (a) and (b) for IS 16, (c) and (d) for IS 17, and
(e) and (f) for IS 18.
271
(a) (b)
(c) (d)
(e) (f)
Figure D.17. 3D plot of PSDs, (a) and (b) for IS 19, (c) and (d) for IS 20, and
(e) and (f) for IS 21.
272
(a) (b)
(c) (d)
(e) (f)
Figure D.18. 3D plot of PSDs, (a) and (b) for IS 22, (c) and (d) for IS 23, and
(e) and (f) for IS 24.
273
(a) (b)
(c) (d)
(e) (f)
Figure D.19. 3D plot of PSDs, (a) and (b) for IS 25, (c) and (d) for IS 26, and
(e) and (f) for IS 27.
APPENDIX E
E.1 STARTUP
It is recommended that the power remain on for day to day operation for LS230
small volume module plus (SVM+).
1. From LS230 Control Program main menu select Run Use Optical Module.
7. (Optional) Run a control. For the first run of the day check that:
(b) Alignment produces a pattern similar to the dfr.$ file in the calfile
folder.
Power Up
275
(b) Turn on the power switches at the LS bench, printer, monitor, and
computer.
2. Prepare the sample. (If existing, run settings can be loaded for type of
sample: select Run Load Run Setting. . .).
3. From main menu select Run Run Cycle. . . New Sample Start or from
the toolbar select Run Cycle. . . icon then click on New Sample Start.
4. When Measure Loading screen appears, slowly add sample and observing
the title bar for correct obscuration of 8% to 12%.
For PIDS samples, < 0.5 m, check for obscuration of 45 to 55%.
Measure Loading: Obscuration = 0%; PIDS = 0% Low, add sample
Add sample until prompt after % level says OK (8-12%)
The runfile title bar will appear as follows:
Measure Loading: Obscuration = 10%; PIDS = 50% OK
8. The runs start and continue until the last runs data appears on the screen.
9. Always rinse sample from module when finished with run, and fill with fresh
suspension fluid.
10. To run additional samples that use this same sample and run info:
276
(a) From the main menu select Run Start or from the toolbar select
Start icon.
E.3 SHUTDOWN
1. On SVM+ keypad, select Off to turn pump off.
2. Close all open RunFiles by clicking on the control menu item until you are
at the LS main menu.
Power Down
(b) Click on the control menu icon on the LS main window, and from the
Windows taskbar, select:
Start Shut Down. . . select the radio button for shut down and select
OK.
(c) Turn off the power switches at the LS bench, printer, monitor and
computer.
APPENDIX F
F.1 Introduction
Cabrera and Vermilyea [54] hypothesized that strongly absorbing, immobile im-
purities can be adsorbed on terraces of crystal surfaces and pin steps, thus drastically
impeding the step movement relative to that expected without impurities present.
The mechanism of pinning and impeding steps has been widely practiced in many
metal alloys, e.g., precipitation hardening in aluminum copper and aluminum silver
alloys and others [66]. Figure F.1 is a schematic picture showing the qualitative
effect of immobile impurities on step movement. The step movement, and hence,
growth rate, would be halted altogether when the distance (d) between impurities
278
on the surface was < 2rc apart, where rc was the medium radius curvature of a
step. For spacings greater than this, the step may squeeze past the adsorbed species
if the step growth velocity is high enough. The impurity spacing is also called a
critical distance, whose magnitude is approximately given by the Gibbs-Thomson
critical diameter [54, 118].
effect also appears to decline above a certain critical impurity concentration. This
was shown in the behavior of the nucleation exponent, which is unaffected at low
contaminant concentration (ppm) but changes erratically at higher concentrations.
The use of 100 ppm CdCl2 was chosen to stabilize NaCl particles since Cd2+
is known to adsorb on the crystal surface [121, 122, 123], pinning and immo-
bilizing the steps in the surface of the NaCl crystal. Cadmium was chosen as
the divalent impurity because its ionic radius is within five percent of that of
sodium [122, 123, 124, 125]. Also the value of is determined not only by impurity
concentration (number of impurity ions per mole of NaCl), but by its type as well.
The strongest hindering action was observed with a smaller ionic radius [120]. In the
solution containing small amounts of Cd2+ and NaCl crystals, the Cd2+ ions enter
the N a+ lattice substitutionally on the surface of NaCl crystals. Since the crystal
must be electrically neutral, each divalent impurity ion replaces two sodium ions
with the result that an excess concentration of positive ion vacancies equal to the
concentration of impurity ion is introduced into the crystal [122]. However, at low
enough temperature the concentration of vacancies is so low that the impurity ions
are mostly unassociated and permanently fixed at their positions in the lattice.
The unassociated impurity ions have no mobility by themselves. However, an
impurity ion may move whenever there is an adjacent vacancy into which it can
jump, although not otherwise.
The use of step pinning Cd2+ additive to stabilize NaCl crystals, to measure PSDs
using LLS, in nearly saturated solutions undergoing temperature variations that
under or super saturate the solution is discussed below.
F.2 Experiments
F.2.1 Crystallization Experiments
A feed solution of (C0 ) = CaCl2 (46.83 wt%), KCl (0.92 wt%), and NaCl
(0.49 wt%) which contains a large amount of crystals at room temperature can
be dissolved by heating to 70 C. To ensure complete dissolution, the solution
was maintained at 70 C for more than 24 hr. Then the solution is fed to the
281
crystallizer and cooled to 30 C to induce crystallization. The CaCl2 and KCl are
soluble at around 30 C, and NaCl crystallizes out of solution in a continuous mixed
suspension, mixed product removal (CMSMPT) crystallizer with four standard
baffles radially-directed around the reactor periphery running the length of the
reactors straight side and a turbine in the middle of reactor. A standard baffle
design and detailed baffling information as well as experimental set-up can be
found elsewhere [87, 88, 92]. The solute concentration of solutions before and
after crystallization was determined by atomic adsorption spectroscopy (AAS)
and inductively coupled plasma-atomic emission spectrometry (ICP-AES). Also
scanning electron microscopy (SEM), energy-dispersive spectrometry (EDS), and
X-ray diffraction (XRD) were performed on solid particles for crystal morphology,
elementary, and phase analyses.
Figure F.3. Result of PSD measurements without the addition of CdCl2 . Each of
the measurements was done for 1 min..
Figure F.4. Result of PSD measurements with the addition of CdCl2 . Each of the
measurements was done for 1 min..
Figure F.5. SEM image of particles showing the crystal habit and size after
crystallization at 30 C.
The measured PSD was validated with SEM (Figure F.5) and microscopic photos,
which show detailed particle size and morphology, that do not change with time in
the cadmium modified saturated solution.
284
Figure 2 Figure 3
(without Cd2+ ) (with Cd2+ )
0 min. 118 108
Dv (m) 5 min. 108 106
10 min. 103 106
C.V. (%) 6.8 1.1
0 min. 38.9 38.6
Dv10 (m) 5 min. 36.9 38.4
10 min. 36.1 38.3
C.V. (%) 3.8 0.4
0 min. 114 103
Dv50 (m) 5 min. 104 101
10 min. 99.6 101
C.V. (%) 6.7 1.0
0 min. 200 184
Dv90 (m) 5 min. 183 181
10 min. 175 180
C.V. (%) 6.9 1.13
285
Using this experimental protocol with its step immobilization additive and
careful temperature control, accurate particle size distributions can be measured
for materials that are unstable due to crystal growth/dissolution from/to the back-
ground solution of the light scattering apparatus. We have used this experimental
protocol to measure the effluent of a crystallizer for various materials [59]. For
practical purposes, particle stabilization with impurities could be a useful industrial
practice when particle size measurement is required during crystallization if no inert
solvent is available to stabilize the particles. Step immobilizing additives and crystal
habit modifiers for other particle systems are tabulated in Myersons book [89].
F.4 Conclusions
Measuring the PSD produced by crystallizers is often difficult due to the dis-
solution or growth of the particles during analysis in traditional light scattering
instruments. If an inert liquid can not be found into which the particles can be
dispersed the only technique available is to use a saturated solution. Even with
dispersion in a saturated solution, the PSD measured can be effected by dissolution
or growth of the particles due to slight variations in the temperature of the sample
in the instrument. Dissolution and growth of the particles in a slightly under
or over saturated solution can be arrested by the addition of step immobilizing
additives. These additives effectively stabilize the crystals by opening up a window
of saturation where the steps of the surface of the crystal lattice are immobile.
APPENDIX G
ion/ion and ion/molecule) that are not accounted for by the long-range term. The
rationale and derivation of Equation G.1 was discussed in detail by Wang et al. [129].
The long-range interaction contribution is calculated from the Pitzer-Debye-
Hckel formula (Pitzer [130, 131]) expressed in terms of mole fractions and symmet-
rically normalized, i.e.,
! !
1/2
Gex
DH
X 4Ax Ix 1 + Ix
= ni ln P 0 1/2
(G.2)
RT i
i xi 1 + (Ix,i )
where the sum is over all species, Ix is the mole fraction-based ionic strength,
0
Ix,i represents the ionic strength when the system composition reduces to a pure
component i, i.e., i0x,i = 1/2; is related to the hard-core collision diameter ( = 14)
and the Ax parameter is given by
3/2
e2
1
Ax = (2NA ds )1/2 (G.3)
3 40 s KB T
where ds and s are the molar density and the dielectric constant of the solvent,
respectively. The dielectric constant is calculated from an expression developed
previously at OLI Systems for mixed solvents (Wang and Anderko [132]).
For the short-range interaction contribution, the UNIQUAC equation (Abrams
and Prausnitz [133]) is used. The middle-range interaction contribution is calcu-
lated from an ionic strength-dependent, symmetrical second coefficient-type expres-
sion (Wang et al. [129]):
!
Gex
MR
X XX
= ni xi xj Bij (Ix ) (G.4)
RT i i j
288
where Bij (Ix ) = Bji (Ix ), Bii = Bjj = 0 and the ionic strength dependence of Bij is
given by
p
Bij (Ix ) = bij = Cij exp Ix + a1 (G.5)
where bij and cij are adjustable parameters and a1 is set equal to 0.01. In general,
the parameters bij and cij are calculated as functions of temperature as
b2,ij
bij = b0,ij + b1,ij T + (G.6)
T
c2,ij
cij = c0,ij + c1,ij T + (G.7)
T
In practice, the middle-range parameters are used to represent ion-ion and ion-
neutral molecule interaction. The short-range parameters are used primarily for
interaction between neutral molecules.
To account for speciation, the chemical effects due to the formation of ion pairs
and complexes are explicitly taken into account using chemical equilibria. Also,
the same chemical equilibrium formalism is used to calculate solid-liquid equilibria.
For this purpose, the standard-state chemical potential 0i needs to be computed
for all species together with the activity coefficients. For solids, the values of 0i are
calculated using the reference-state Gibbs energy of formation, entropy and heat
capacity according to standard thermodynamic relationships. For aqueous species,
the standard-state chemical potentials are calculated as functions of temperature
and pressure using a comprehensive model developed by Helgeson and cowork-
ers (Helgeson et al. [134, 135, 136, 137]), commonly referred to as the Helgeson-
Kirkham-Flowers equation of state. The parameters of this model are available for
a large number of aqueous species including ions, associated ion pairs, and neutral
species (Shock and Helgeson [138, 139, 140], Shock et al. [141, 142]). It should be
noted that standard-state property data and the model of Helgeson et al.are based
on the infinite-dilution reference state and on the molality concentration scale.
At the same time, the activity coefficient model is symmetrically normalized and is
expressed in terms of mole fractions. To make the two reference systems consistent,
the activity coefficients calculated from Equation G.1 are converted to those based
on the unsymmetrical reference state, i.e. at infinite dilution in water, as described
289
by Wang et al. [129]. At the same time, the molality-based standard-state chemical
potential is converted to a corresponding mole fraction-based quantity.
A similar procedure is used for enthalpy and heat capacity calculations. In the
unsymmetrical normalization, the total enthalpy is expressed as
X
h= xi hi = hex, (G.8)
i
[18] J. Estrin, Preparation and Properties of Solid State Materials, vol. 2, Dekker,
New York, 1976.
[19] J. Garside and R. J. Davey, Secondary contact nucleation:kinetics, growth
and scale-up, Chem. Eng. Commun., vol. 4, pp. 393424, 1980.
[20] A. D. Randolph, J. R. Bechman, and K. Kraljevich, Crystal size distribution
dynamics in a classified crystallizer: Part I. experimental and theoretical
study of cycling in a potassium chloride crystallizer, AIChE J., vol. 23, no.
4, pp. 500510, 1977.
[21] B. C. Shah, W. L. McCabe, and R. W. Rousseau, Polyethylene vs. stainless
steel impellers for crystallization processes, AIChE J., vol. 19, no. 1, pp.
194196, 1973.
[22] J. Garside and M. B. Shah, Crystallization kinetics from msmpr crystalliz-
ers, Ind. Eng. Chem. Process. Des. Dev., vol. 19, pp. 509514, 1980.
[23] C. Gahn and A. Mersmann, Brittle fracture in crystallization processes:
Part a. attrition and abrasion of brittle solids, Chem. Eng. Sci., vol. 54, no.
9, pp. 1273, 1999.
[24] C. Gahn and A. Mersmann, Brittle fracture in crystallization processes:
Part b. growth of fragments and scale-up of suspension crystallizers, Chem.
Eng. Sci., vol. 54, no. 9, pp. 1283, 1999.
[25] A. F. Wells, Crystal growth, in Annual Reports on the progress of
Chemistry, London, 1946, The Chemical Society, vol. 43, pp. 6287.
[26] H. E. Buckley, Crystal Growth, Chapman and Hall, London, 1952.
[27] R. F. Strickland-Constable, Kinetics and Mechanism of Crystallization,
Academic Press, London, 1968.
[28] B. Lewis, Nucleation and growth theory, chapter In Crystal Growth, pp.
2363, Pergamon Press, Oxford, 1980.
[29] A. A. Chernov, Modern Crystallography, Springer-Verlag, Berlin, 1980.
[30] A. A. Chernov, Formation of crystals in solution, Contemporary Physics,
vol. 30, pp. 251276, 1989.
[31] J. N`
y vlt, O. Sohnel, M. Matuchova, and M. Broul, The Kinetics of Industrial
Crystallization, Elsevier, Amsterdam, 1985.
[32] D. E. Temkin, Crystallization Processes, Consultant Bureau, New York,
1964, p.15.
[33] K. A. Jackson, Liquid Metals and Solidification, Am. Soc. Metals, Cleveland,
1958, p.174.
292
[129] P. Wang, A. Anderko, and R. D. Young, , Fluid Phase Equilibria., vol. 141,
pp. 203, 2002.
[130] K. S. Pitzer, Thermodynamics of electrolytes. i. theoretical basis and general
equations, J. Phys. Chem., vol. 77, no. 2, pp. 268277, 1973.
[131] K. S. Pitzer, Electrolytes. from dilute solutions to fused salts, J. Am.
Chem. Soc., vol. 102, no. 9, pp. 29022906, 1980.
[132] P. Wang and A. Anderko, , Fluid Phase Equilibria., vol. 186, pp. 103, 2001.
[133] D. S. Abrams and J. M. Prausnitz, , AIChE J., vol. 21, pp. 116, 1975.
[134] H. C. Helgeson, D. H. Kirkham, and G. C. Flowers, , Amer. J. Sci., vol.
274, pp. 1089, 1974.
[135] H. C. Helgeson, D. H. Kirkham, and G. C. Flowers, , Amer. J. Sci., vol.
274, pp. 1199, 1974.
[136] H. C. Helgeson, D. H. Kirkham, and G. C. Flowers, , Amer. J. Sci., vol.
276, pp. 97, 1976.
[137] H. C. Helgeson, D. H. Kirkham, and G. C. Flowers, , Amer. J. Sci., vol.
281, pp. 1241, 1981.
[138] E. L. Shock, H. C. Helgeson, and D. A. Sverjensky, , Geochim. Cosmochim.
Acta, vol. 52, pp. 2009, 1988.
[139] E. L. Shock, H. C. Helgeson, and D. A. Sverjensky, , Geochim. Cosmochim.
Acta, vol. 53, pp. 2157, 1989.
[140] E. L. Shock, H. C. Helgeson, and D. A. Sverjensky, , Geochim. Cosmochim.
Acta, vol. 54, pp. 915, 1990.
[141] E. L. Shock, D. C. Sassani, M. Willis, and D. A. Sverjensky, , Geochim.
Cosmochim. Acta, vol. 61, pp. 907, 1997.
[142] D. A. Sverjensky, E. L. Shock, and H. C. Helgeson, , Geochim. Cosmochim.
Acta, vol. 61, pp. 1359, 1997.
INDEX
Activity BCF model, 35
in equilibrium with the condensate, chemical reaction, 45
12 continuous growth, 25
of the condensing species, 11 diffusion-reaction theories, 40
Aggregation kernel, 58 impurity effect, 46
size-independent, 0 , 155 lateral growth, 25
Aggregation rate, Ka , 157 surface energy theories, 27
Atomic absorption spectroscopy (AAS), Crystal shape, 53
79 Crystallization, 1
Attrition, 192 Crystallization temperature, TC , 157
Crystallography, 53
CMSMPR, 1
CSTR, 4
Computational fluid dynamics (CFD),
Cumulative number oversize distribu-
94
tion, N (L), 190
Contact angle, 16
Contact nucleation rate, 19 Damkohler number, 50
Conversion factor for crystal growth, Da, 46
surface area, A , 12 Density, 76
volume, V , 12 suspension density, 4
Correlation analysis, 164 Diffusion
Crystal coefficient, 50n
habit, 54 EDS, 88
population density, 5 Effectiveness factor, 45
size distribution for crystal growth rate, c , 45
CSD, 3 Element analysis
Crystal growth kinetics, 20 of solid, 88, 107
adsorption layer theories, 27 of solution, 79, 97
300