© ESTEVE QUÍMICA (2019)
The Use of Simulation Tools
for Crystallisation Process
Development.”Steve Winter
Crystallisation Workshop
Barcelona, April 11, 2019
© ESTEVE QUÍMICA (2019)
ESTEVE GROUP
ESTEVE PHARMACEUTICALS, S. A. (SPAIN)/ ESTEVE FARMA
(SPAIN), LDA (PORTUGAL) / PENSA PHARMA, S. A. (SPAIN)/
PENSA PHARMA, sPa (ITALY) / toLife (PORTUGAL) / PENSA
PHARMA ILAÇ (TURKEY) / PENSA PHARMA AB (SPAIN)/ ESTEVE
QUÍMICA, S. A. (SPAIN) / SINTENOVO, S.A. (MEXICO) /
ZHEJIANG HUAYI PHARMACEUTICAL, CO. LTD. (CHINA) /
ESTEVE HUAYI PHARMACEUTICAL CO., LTD. (CHINA) / ISDIN,
S.A. (SPAIN)/ ESTEVE TEIJIN HEALTHCARE (ETH) (CHINA)
2
Family owned business with international presence
Esteve QuímicaCorporate HQ and R&D - Barcelona
Manufacturing – Celrà (Girona), Banyeres de Penedes (Tarragona), Mexico, China
Process Development and Manufacture of APIs
© ESTEVE QUÍMICA (2019)
Lab optimization
PAR studies Scale-up Manufacturing
Process Development: old approach
Sequential activities with the objective to demonstrate that the
process works.
3
Why use simulation?
© ESTEVE QUÍMICA (2019)
Process Development: QbD approach
Lab optimization
PAR studies Scale-up Manufacturing
4
Build the quality by understanding the process and be able to predict
the possible impact of the changes
How to integrate the different phases of a
development into a solid knowledge platform
Using simulation tools
QbD
© ESTEVE QUÍMICA (2019)
Process Development: QbD approach
Models enable systematic analysis of experimental data and
quantification of effects on process change
Less experimental trials
Less resources
Deeper knowledge
Global knowledge platform
Good customer perception
Future trend for regulatory submission
5
© ESTEVE QUÍMICA (2019)
Simulation Tools
Several different commercially available
packages
Crystallization
•Solvent selection
•Optimization
Reaction kinetics
•Process optimization
•PAR studies
Unit operations
•Distillation
•Solvent swap
•Extractions
Scale up attributes
•PSD control
•Filterability
•Drying
•Reactor design
Safety studies
•Safety tests
•Safe scale up
6
© ESTEVE QUÍMICA (2019)
Solid Form Development Workflow7
API
Solid form
screening
Solvent
screening
Normally some limited solubility data generated
Basic solubility data for a range of solvents
Crystallisation
process
development
Thermodynamic modelling (accurate solubility data)
Kinetic modelling
© ESTEVE QUÍMICA (2019)
Solvent Selection Workflow8
Solid form
knowledge
Solubility
modelSolubility
experiments
Solubility
predictions
Thermal data (m.p. DH0f)
solubility info. fromsolid form screen
XRPD recovered solid
Solvent
systems for
development
Solubility model (e.g. UNIFAC) allows for predictions in solvent systems
outside of the experimental range (e.g. temp., mixtures, different
solvent with similar functional groups)
© ESTEVE QUÍMICA (2019)
Regressed UNIFAC Solubility Model9
UNIFAC model Main groups = 51
The UNIFAC method (UNIQUAC Functional-group Activity Coefficients) is
a semi-empirical system for the prediction of non-electrolyte activity in
non-ideal mixtures.
UNIFAC uses the functional groups present on the molecules that make
up the liquid mixture to calculate activity coefficients.
Each group is characterized by a Van der Waals surface area (Q) and
volume (R) and binary interaction parameter (ij).
Regressed UNIFACUNIFAC Main groups + solute, the API
51 + 1= 52 groups
© ESTEVE QUÍMICA (2019)
UNIFAC – Experimental screen10
Define solvents and enter measured data Vol frac 1 Vol frac 2 Temp Measured
Solvent1 Solvent2 - C wt%
Ethanol <select solvent> 1 0 25 0,13
acetonitrile <select solvent> 1 0 25 0,06
acetone <select solvent> 1 0 25 0,01
THF <select solvent> 1 0 25 0,06
1_4_Dioxane <select solvent> 1 0 25 0,01
MEK <select solvent> 1 0 25 0,00
Acetic_acid <select solvent> 1 0 25 14,42
water <select solvent> 1 0 25 0,27
Ethanol water 0,5 0,5 25 2,39
acetonitrile water 0,5 0,5 25 10,02
acetone water 0,5 0,5 25 3,32
THF water 0,5 0,5 25 6,70
1_4_Dioxane water 0,5 0,5 25 3,10
MEK water 0,5 0,5 25 5,16
Acetic_acid water 0,5 0,5 25 13,21
Include relevant functional groups to allow predictions outside the
experimental range
© ESTEVE QUÍMICA (2019)
Screen solvent mixtures Select your 10 solvents using row 3, then click 'Screen' button Temperature 20 C
Solubility (g/L) vs comp (vfrac) Acetonitrile MTBE Acetone 2_MeTHF i_propyl_acetate Isopropanol MIBK n_heptane n_pentane
Acetonitrile
MTBE VERDADERO
Acetone VERDADERO VERDADERO
2_MeTHF VERDADERO VERDADERO VERDADERO
i_propyl_acetate FALSO FALSO FALSO FALSO
Isopropanol VERDADERO VERDADERO VERDADERO VERDADERO FALSO
MIBK VERDADERO VERDADERO VERDADERO VERDADERO FALSO VERDADERO
n_heptane VERDADERO VERDADERO VERDADERO VERDADERO FALSO VERDADERO VERDADERO
n_pentane VERDADERO VERDADERO VERDADERO VERDADERO FALSO VERDADERO VERDADERO VERDADERO
Water VERDADERO VERDADERO VERDADERO VERDADERO FALSO VERDADERO VERDADERO VERDADERO VERDADERO
Max solubility (g/L) Acetonitrile MTBE Acetone 2_MeTHF i_propyl_acetate Isopropanol MIBK n_heptane n_pentane
Acetonitrile
MTBE 0,31 Maximum solubilities over the composition range
Acetone 0,43 0,05
2_MeTHF 0,33 0,01 0,06
i_propyl_acetate 6,23 1,01 1,50 1,01
Isopropanol 1,37 0,48 0,89 0,76 11,80
MIBK 0,31 0,00 0,05 0,01 1,01 0,31
n_heptane Immiscible 0,00 0,05 0,01 1,01 0,24 0,00
n_pentane Immiscible 0,00 0,05 0,01 1,01 0,24 0,00 0,00
Water 684,52 Immiscible 363,35 Immiscible Immiscible 93,76 Immiscible Immiscible Immiscible
Solvent Selection Model –Regressed UNIFAC
11
https://dcresources.scale-up.com/#q=SOLUBILITY___173
https://dcresources.scale-up.com/#q=SOLUBILITY___284
Predicted solubilities used
for guideline purposes
Not 100% accurate!
© ESTEVE QUÍMICA (2019)
Examples Solvent Selection Model12
0
100
200
300
400
500
600
700
800
900
10 20 30 40 50 60
So
lub
ility
(g
/L D
is)
Temperature (ºC)
Solubility Vs Temperature
Ethanol
Methanol
Single solvent
S Methanol > S Ethanol
Binary mixtures
Solvent-water
S Methanol-water < S Ethanol-water
0
10
20
30
40
50
60
70
80
0 0.2 0.4 0.6 0.8 1
So
lub
ility
(g
/L)
Volume fraction
Solubility Vs Volume fraction
Ethanol
Methanol
Process efficiency improved from 100 to 30 volumes
© ESTEVE QUÍMICA (2019)
Solvent Selection Workflow13
Solid form
knowledge
Solubility
modelSolubility
experiments
Solubility
predictions
Thermal data (m.p. DH0f)
solubility info. from
solid form screen
XRPD recovered solid
Solvent
systems for
development
What are the objectives?• Yield
• Purity
• Polymorph
• Particle Size Distribution
© ESTEVE QUÍMICA (2019)
Particle Size Considerations14
As long as it filters OK
and dries OK, we don’t
mind.
I’ve got a crystallization process. The
yield is 90%, the purity is 99.5%, but….
What PSD do we need?
We don’t know yet. But it
will be VERY important!
We need all plant batches to have
the same PSD as that first lab batch
you gave us!
Some months later….
Solid form
scientist
Plant
scientist
Formulation
scientist
© ESTEVE QUÍMICA (2019)
Scale-up Challenges15
Objectives
1. Correct polymorphic form
2. Optimise yield, purity
3. Correct particle size distribution
mg g kg
Requirements
1. Thermodynamics (solubility)
2. Kinetics of crystallisation
© ESTEVE QUÍMICA (2019)
Some Words of Wisdom16
“Trying to develop or troubleshoot a solution
crystallisation process without knowledge of the solubility
curve and metastable zone width is akin to hiking though
the wilderness without a map or compass.”
Chris Price
ChemEng Prog 34-43 (1997)
© ESTEVE QUÍMICA (2019)
Solubility Curves or Models17
Co
nce
ntr
atio
nUnstable Zone
Stable Zone
Meta-stable Zone
spontaneousnucleation
Temperature
What if we have a solvent mixture? – Better to have a solubility model.
© ESTEVE QUÍMICA (2019)
Solubility Models for Binary Mixtures18
Van’t Hoff Equation:
Solubility models:
where x1 is solvent fraction
R
ΔS
RT
ΔHnKl
Model Equation
0
1
2
3
4
RT
xDxC
RT
BAC 10
100
0lnln
RT
xDxC
RT
BAC 1
1111
1 explnln
)exp(ln 122122
12 xDRT
C
RT
xBxAC
RT
xDxC
RT
BAC 13
133
3 )exp(lnln
44142
14ln DRT
C
RT
xBxAC
1. Experimental data
2. Data regression
3. Choose best fitting model
2/3 temperatures × 4/5 compositions = 8 – 15 experiments
© ESTEVE QUÍMICA (2019)
Solubility Model
Binary Mixture Example19 Dynochem Simulation of Solubility Curves
Variation of Crude 4093 Solubility with Water Vol. Fraction
0
50
100
150
200
250
300
350
400
10 20 30 40 50 60 70 80 90
Temperature (ºC)
Co
ncen
trati
on
(g
4093 /
L o
f so
lven
t)
0.21 vol fraction
0.25 vol fraction
271 g/L concentration
308 g/L concentration
bp 655mmHg 75ºC
bp 760mmHg 80ºC
65-70ºC seeding range
So
lub
ility
(g
/Kg
)
Temperature (ºC)Solubility model aids basic process design
But how can it help us control PSD?
© ESTEVE QUÍMICA (2019)
Crystallisation Process Development
First step – solubility curve / model20
Co
nce
ntr
atio
n
Unstable Zone
Stable Zone
Meta-stable Zone
spontaneousnucleation
Temperature
Conventional wisdom: for large PSD seed close to solubility curve and cool slowly.
Is this always true? And what if we don’t want large particles?
We need to understand crystal growth and nucleation.
© ESTEVE QUÍMICA (2019)
Some Words of Wisdom21
“Remember that all models are wrong; the
practical question is how wrong do they
have to be to not be useful.”
George E. P. Box (1919 – 2013)
“Trying to develop or troubleshoot a solution crystallisation
process without knowledge of the solubility curve and
metastable zone width is akin to hiking though the wilderness
without a map or compass.”
Chris Price, ChemEng Prog 34-43 (1997)
© ESTEVE QUÍMICA (2019)
Crystal Growth Modelling22
Growth rate
Δc = (CL-C*)
B = KB · Δcb
G = KG · Δcg
Nucleation rate
Supersaturation
To model crystallization kinetics:
Need to know KG, g, KB, b and the solubility (C*)
Simplified approach: Turn off nucleation!
Peter Clark, DynoChem ACT, 16th Larson Workshop October 5, 2009
© ESTEVE QUÍMICA (2019)
Crystallisation kinetics23
Supersaturation
Growth rate
Nucleation rate
Growth rate
Δc = (CL-C*)
B = K · Δcb
G = K · Δcg
Nucleation rate
Supersaturation
Use growth only model, vary process conditions
and track supersaturation
If the model shows high supersaturation
small particles due to nucleation
Rahn McKeown
https://dcresources.scale-
up.com/#q=Crystallization___2982
Typically g << b
PSD is determined by the relative rates of
growth, nucleation, attrition,
agglomeration
© ESTEVE QUÍMICA (2019)
Simplified Supersaturation Model24 G = K·Δcg
*)CC(AKdt
dMLSL
t
SS V·aA
*)CC(VKdt
dMLSG
t
dt*)CC(VKM L
t
0t
SGt
G growth rate
Δc supersaturation
K constants
Mt mass of formed crystals
AS area of crystals
VS volume of crystals
CL concentration
C* solubility
g ~ 1
This model is wrong – but is it useful?
Determine KG and solubility model (C*) experimentally
Follow crystal mass over time, but more importantly we can track DC
© ESTEVE QUÍMICA (2019)
Example: Crystallisation Development
Compound A25
Tem
pera
ture
Charge base, IPA
and water
Addition of acidSeeding
Addition IPA Filtration
Polish filtration
Parameters:
Seeding T, charge, PSD
Maturation time
Cooling ramp
Final T
Antisolvent addition rate
Agitation
Time
© ESTEVE QUÍMICA (2019)
Experimentation vs Simulation26
Many Parameters:Seeding T, charge, PSD
Solvent volumes
Maturation time
Cooling ramp
Final TAntisolvent addition rate
Agitation
Computer simulationMany variables, many levels
Virtual DoE
Hundreds of simulations in a few minutes
Semi-empirical modelPredictions not limited by design space
Experimental DoEMany experiments
Only practical to study a few levels
Lots of analytical time
Statistical model based on exp. dataOnly valid within design space
Complementary approaches
© ESTEVE QUÍMICA (2019)
Supersaturation model27
dt*)CC(VKM L
t
0t
SGt
Solubility model (C*)
Crystal growth equation
RT
xDxC
RT
BAlnCln 10
100
0
So far the only experimentation has been determination of the solubility model
Calculate the maximum supersaturation (CL-C*)
experienced during the crystallisation process
Virtual DoE Full factorial
three levels
Factors Seeding temperature
Cooling ramp
Volumes water
Volumes IPA
-0,4
-0,3
-0,2
-0,1
-0,0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1,0
1,1
1,2
maxsupratio maxsuper
Investigation: sita_kg0.01_3 (MLR)
Normalized Coefficients
N=81 Cond. no.=6,203 DF=66
Temp
ram
H2O
IPA
Temp*Temp
ram*ram
H2O*H2O
IPA*IPA
Temp*ram
Temp*H2O
Temp*IPA
ram*H2O
ram*IPA
H2O*IPA
MODDE 9.1 - 2013-03-13 16:31:59 (UTC+1)
-0,4
-0,3
-0,2
-0,1
-0,0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1,0
1,1
1,2
maxsupratio maxsuper
Investigation: sita_kg0.01_3 (MLR)
Normalized Coefficients
N=81 Cond. no.=6,203 DF=66
Temp
ram
H2O
IPA
Temp*Temp
ram*ram
H2O*H2O
IPA*IPA
Temp*ram
Temp*H2O
Temp*IPA
ram*H2O
ram*IPA
H2O*IPA
MODDE 9.1 - 2013-03-13 16:31:59 (UTC+1)
KG = 0.01 s-1
slow growth kinetics
Factors:
ramp,
ramp*ramp
-1,2
-1,0
-0,8
-0,6
-0,4
-0,2
0,0
0,2
0,4
0,6
0,8
1,0
maxsupratio maxsuper
Investigation: sita_kg0.1_3 (MLR)
Normalized Coefficients
N=81 Cond. no.=6,203 DF=66
Temp
ram
H2O
IPA
Temp*Temp
ram*ram
H2O*H2O
IPA*IPA
Temp*ram
Temp*H2O
Temp*IPA
ram*H2O
ram*IPA
H2O*IPA
MODDE 9.1 - 2013-03-13 16:36:32 (UTC+1)
KG = 0.2 s-1
fast growth kinetics
Factors:
water,
seeding temp
© ESTEVE QUÍMICA (2019)
Effect of Seeding T and Ageing Time
KG = 0.01s-128Virtual DoE Factors
Seeding temperature (4 levels)
Ageing time (2 levels)
Lowest max.
in supersat
Highest max.
in supersat
How does this compare with experimentation?
© ESTEVE QUÍMICA (2019)
Experimental DoE29
IPA H2OTemp. seed
Cooling Ramp
- - - -
+ + - -
- + + -
+ - + -
- + - +
+ - - +
- - + +
+ + + +
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
FactorCentre Point
Lower Value
Upper Value
Volumes IPA (ml/g) 3.43 3.31 3.81
Volumes water (ml/g) 0.89 1.00 1.06
Seeding Temperature (ºC) 73 65 70
Cooling Ramp (ºC/min) 0.3 0.1 0.5
Resolution IV fractional factorial design
24-1, 8 experiments + 4 centre point
Analysis by laser diffraction
(Malvern Mastersizer)
© ESTEVE QUÍMICA (2019)
Comparison of Models:
Simulation vs DoE30
Two main effects are statistically significant:
Seed temperature(marginal) and cooling ramp.
The effect of cooling ramp is non-linear.
-0,30
-0,20
-0,10
-0,00
0,10
0,20
0,30
ram
ram
*ram
Tem
p
Effects
Effects for D01~
N=12 R2=0,963 RSD=0,02695
DF=8 Q2=0,917 Conf. lev.=0,95
Investigation: sita (MLR)
MODDE 9.1 - 2013-03-13 16:46:56 (UTC+1)
ramp
ramp * ramp
temp
-1,2
-1,0
-0,8
-0,6
-0,4
-0,2
0,0
0,2
0,4
0,6
0,8
1,0
maxsupratio maxsuper
Investigation: sita_kg0.1_3 (MLR)
Normalized Coefficients
N=81 Cond. no.=6,203 DF=66
Temp
ram
H2O
IPA
Temp*Temp
ram*ram
H2O*H2O
IPA*IPA
Temp*ram
Temp*H2O
Temp*IPA
ram*H2O
ram*IPA
H2O*IPA
MODDE 9.1 - 2013-03-13 16:36:32 (UTC+1)
-0,4
-0,3
-0,2
-0,1
-0,0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1,0
1,1
1,2
maxsupratio maxsuper
Investigation: sita_kg0.01_3 (MLR)
Normalized Coefficients
N=81 Cond. no.=6,203 DF=66
Temp
ram
H2O
IPA
Temp*Temp
ram*ram
H2O*H2O
IPA*IPA
Temp*ram
Temp*H2O
Temp*IPA
ram*H2O
ram*IPA
H2O*IPA
MODDE 9.1 - 2013-03-13 16:31:59 (UTC+1)
KG = 0.01 s-1
Factors:
ramp,
ramp*ramp
KG = 0.2 s-1
Factors:
water
seeding temp,
The experimental PSD results correlate inversely with
the supersaturation simulation for slow growth kinetics
(lower KG).
D[0.5]
© ESTEVE QUÍMICA (2019)
Plant Batches31
Batch A particle size distribution was
too fine.
Batch B was better. Please provide
future batches like this.
Formulation
scientist
Batch ScaleSeed Temp
Seed load
Time after seeding
Cooling time
Agitation D[0.1] D[0.5] D[0.9]
Kg ºC wt% min hours rpm µm µm µm
A 50 69 0.5 30 5 75 4.7 20.6 89.7
B 50 69 0.5 15 5.1 70 8.5 49.4 162.3
Your process isn’t robust
enough!
Plant
scientist
DoE suggested cooling rate is main factor
Can our crystal growth / supersaturation model be improved?
© ESTEVE QUÍMICA (2019)
Determination of KG32
Cooled crystallization
Samples taken, filtered and analysed (HPLC or gravimetrically)
© ESTEVE QUÍMICA (2019)
Determination of KG33
RT
xDxC
RT
BAlnCln 10
100
0
© ESTEVE QUÍMICA (2019)
Determination of KG34
dt*)CC(VKM L
t
0t
SGt
© ESTEVE QUÍMICA (2019)
35
Determination of KG
dt*)CC(VKM L
t
0t
SGt
© ESTEVE QUÍMICA (2019)
36
dt*)CC(VKM L
t
0t
SGt
Determination of KG
KG = 0.08 s-1
© ESTEVE QUÍMICA (2019)
Virtual DoE with KG = 0.08 s-1
37
65 66 67 ºC6463
0.1
0
.3
0.5
ºC
/min
Co
olin
g r
ate
Seeding Temperature
Maximum supersaturation levelVariables:
Seed temperature (9 levels)
Cooling rate (3 levels)
© ESTEVE QUÍMICA (2019)
Particle Size Considerations38
You must be joking! We need a
seeding temperature range of
at least 5ºC, preferably 10ºC
Well… we can try.
OK, the formulation scientists want
larger particles, so we should seed
at 64-66ºC and cool at 0.1ºC/min
Solid form
scientist
Plant
scientist
Batch ScaleSeed Temp
Seed load
Time after seeding
Cooling time
Agitation D[0.1] D[0.5] D[0.9]
Kg ºC wt% min hours rpm µm µm µm
A 50 69 0.5 30 5 75 4.7 20.6 89.7
B 50 69 0.5 15 5.1 70 8.5 49.4 162.3
© ESTEVE QUÍMICA (2019)
Plant Batches39
Erm, sorry…. But the PSD of
batches C and D are actually a
little too large. Please can we
have all batches like batch B?
Formulation
scientist
Batch ScaleSeed Temp
Seed load
Time after seeding
Cooling time
Agitation D[0.1] D[0.5] D[0.9]
Kg ºC wt% min hours rpm µm µm µm
A 50 69 0.5 30 5 75 4.7 20.6 89.7
B 50 69 0.5 15 5.1 70 8.5 49.4 162.3
C 42.2 66 0.6 20 10 70 10.1 68.6 211.1
D 60 66 0.5 15 10 75 10.3 70.1 236.3
###*****!
Solid form
scientist
Seeding at lower temperature, slower cooling
larger particles
Still room for improvement!
© ESTEVE QUÍMICA (2019)
Some Words of Wisdom40
“Remember that all models are wrong; the
practical question is how wrong do they
have to be to not be useful.”
George E. P. Box (1919 – 2013)
“Trying to develop or troubleshoot a solution crystallisation
process without knowledge of the solubility curve and
metastable zone width is akin to hiking though the wilderness
without a map or compass.”
Chris Price, ChemEng Prog 34-43 (1997)
“Everything should be made as simple
as possible, but not simpler.”
Albert Einstein (1879-1955)
© ESTEVE QUÍMICA (2019)
Particle Size Distribution Simulations41
Requirements:
1. Improved crystal growth modelling
2. Add equations for nucleation
(and attrition and agglomeration)
3. Particle distribution tracking
© ESTEVE QUÍMICA (2019)
More Sophisticated Growth Equation42
Temperature dependent growth,
non-linear with respect to supersaturation
Previous approximation with g = 1
gCRT
EakG D
exp
g
ref
ref CTTR
EakG D
)
11exp( ln_
gCKG D
g ≠ 1
© ESTEVE QUÍMICA (2019)
More Sophisticated Growth Equation43
Experiments performed at low supersaturation to avoid nucleation
Large amount of seed to increase area for reasonable growth
Data fitted to determine Ea, Kref_ln and growth_order
Takes into account temperature dependence of growth constant
g
ref
ref CTTR
EakG D
)
11exp( ln_
© ESTEVE QUÍMICA (2019)
Particle Size Distribution Simulations44
1. Improved crystal growth modelling
2. Add equations for nucleation
(and attrition and agglomeration)
3. Particle distribution tracking
© ESTEVE QUÍMICA (2019)
Secondary Nucleation Equation45
DC = supersaturation (calculated)
MT = slurry density (calculated)
Power per unit mass (calculated)
Secondary nucleation rate constant (fitted experimentally)
Secondary nucleation order (fitted experimentally)
N_exp = exponent on agitation (fitted experimentally)
Power per unit mass is our equipment dependent variable
that effects nucleation
J. Mullin, Crystallization, 4th edition, Butterworth-Heinemann, 2001, p.249H-H. Tung et al, Crystallization of Organic Compounds: An Industrial Perspective, John Wiley & Sons, 2009. p.86
© ESTEVE QUÍMICA (2019)
Calculation of Power per Unit Mass46
https://dcresources.scale-up.com/#q=solid-
liquid+Liquid+Mixing___5578
© ESTEVE QUÍMICA (2019)
Particle Size Distribution Simulations47
1. Improved crystal growth modelling
2. Add equations for nucleation
(and attrition and agglomeration)
3. Particle distribution tracking
© ESTEVE QUÍMICA (2019)
Population Balance Modelling
Method of Moments48
Approximation assumes a lognormal
distribution
© ESTEVE QUÍMICA (2019)
Determination of nucleation parameters
Cooled crystallization with FBRM data49
Fit data (concentration
and moments)
to determine
sec_nuc_rate_ln,
N_exp and
sec_nuc_order
© ESTEVE QUÍMICA (2019)
Crystallisation model with growth,
nucleation and population modelling50
Problem: only one lab experiment with FBRM data
no variation in power per unit mass over data set
N_exp and sec_nuc_rate_ln not properly resolved
Solution: second lab experiment changing scale and agitation
Or alternatively…
use data from one of the plant batches (Malvern PSD only)
© ESTEVE QUÍMICA (2019)
Summary of Plant Batches51
BatchScale Seed temp Seed load
Time at T after seeding
Cooling time AgitationPower per unit
mass
Kg ºC wt% min hours rpm W/Kg
A 50 69 0.5 30 5 75 0.35
B 50 69 0.5 15 5.10 70 0.31
C 42.2 66 0.6 20 10 70 0.45
D 60 66 0.5 15 10 75 0.32
E 90 68 0.5 15 10 75 0.21
Experimental PSD Dynochem Models
Batch D[0.1] D[0.5] D[0.9]max
supersatwith nucleation and population modelling
µm µm µm g/KgD[0.1]
µmD[0.5]
µmD[0.9]
µm
A 4.7 20.6 89.7 39 12.0 44.9 166.9
B 8.5 49.4 162.3 39 12.9 47.2 172.4
C 10.1 68.6 211.1 28 21.0 67.1 214.2
D 10.3 70.1 236.3 30 21.9 70.0 223.3
E 10.6 55.4 168.0 30 21.9 69.9 223.1
Still room for improvement, but getting close!
© ESTEVE QUÍMICA (2019)
Crystallisation Development52
Solubility
Model
Crystal growth /
supersaturation
model
PSD model with
nucleation
One or two experiments allows for model that can
guide effects of process parameters on
supersaturation (and therefore PSD trends!)
Experiments with good PSD and concentration
data and varying power / unit mass (agitation,
scale).
Scale up
predictions
Essential first step for any crystallization model!
Based on good experimental data
Process Knowledge
© ESTEVE QUÍMICA (2019)
Acknowledgments53
Guillem Molas Andrew Bird
Peter Clark
Joe Hannon
Victor Nicolau
Cristina Renau