Process Optimization Using Statistical Methods
Prof. Attilio Citterio
Dipartimento CMIC “Giulio Natta”
http://iscamap.chem.polimi.it/citterio/dottorato//
PhD
IN INDUSTRIAL CHEMISTRY AND
CHEMICAL ENGINEERING (CII)
Correlation Effect Temperature & pH
0 Temperature (°C) 100
Y ield (%)
100
pH 10
pH 7
Efficiency in Research Development and
Production:
The Statistical Design and Analysis of Chemical Experiments
Leslie Davies
Department of Chemistry and Applied Chemistry,
University of Salford
ROYAL SOCIETY OF CHEMISTRY
Some Difficult Decisions
Design and
Optimisation in Organic Synthesis Rolf Carlson
Department of Organic Chemistry, Umea University, S-901 37 Umea, Sweden
ELSEVIER 1992
EXAMPLE
• Reaction is thought to depend on 3 variables - Temperature (T)
- Concentration (C) - Catalyst (K)
• Set an initial range for each variable:
- Temperature 160-180°C - Concentration 20-40%
- Catalyst A or B
Factorial Design
Data from a 2
3Factorial Design
a) Original units of variables
Test Condition
Number
Temperature (°C)
T
Concentration (%)
C
Catalyst (A or B)
K
Yield (grams)
y
1 160 20 A 60
2 180 20 A 72
3 160 40 A 54
4 180 40 A 68
5 160 20 B 52
6 160 20 B 83
7 160 40 B 45
8 180 40 B 80
b) Coded units of variables
Test Condition
Number
Temperature (°C)
T
Concentration (%)
C
Catalyst (A or B)
K
Yield (grams)
y
1 - - - 60
2 + - - 72
3 - + - 54
4 + + - 68
5 - - + 52
6 + - + 83
7 - + + 45
8 + + + 80
Data from a 23 Factorial Design
Effect of Temperature Change
Condition at which Comparison is made Individual measure of the effect of Concentration Catalyst
Changing concetration from 20 to 40 C K
y
2– y
1= 72 - 60 = 12 20 A
y
4– y
2= 68 - 54 = 14 40 A
y
7– y
5= 83 – 52 = 31 20 B
y
8– y
6= 80 - 45 = 35 40 B
Main effect of concentration C = 23
Effect of Concentration Change
Condition at which Comparison is made Individual measure of the effect of Temperature Catalyst Changing temperature from 160 to 180°C T K
y
2– y
1= 54 - 60 = -6 160 A
y
4– y
2= 68 - 72 = -4 180 A
y
7– y
5= 45 – 52 = -7 160 B
y
8– y
6= 80 - 83 = -3 180 B
Main effect of temperature T = -5
Effect of Catalyst Change
Condition at which Comparison is made Individual measure of the effect of Temperature Conc Changing from catalyst A to catalyst B T C
y
5– y
1= 52 - 60 = -8 160 20
y
6– y
2= 83 - 72 = 11 180 20
y
7– y
3= 45 – 54 = -9 160 40
y
8– y
4= 80 - 68 = 12 180 40
Main effect of catalyst K = 1.5
b) Coded units of variables
Test Condition
Number
Temperature (°C)
T
Concentration (%)
C
Catalyst (A or B)
K
Yield (grams)
y
1 - - - 60
2 + - - 72
3 - + - 54
4 + + - 68
5 - - + 52
6 + - + 83
7 - + + 45
8 + + + 80
Calculation of Interaction Effects
Average
Concentration Temperature effect
(+) B -5.0 C × K interaction = 0/2 = 0
(-) A -5.0
difference 0.0 Average
Concentration Temperature effect
(+) 40% 24.5 T × C interaction = 3/2 = 1.5
(-) 20% 21.5
difference 3.0
Average
Catalyst Temperature effect
(+) B 33 T × K interaction = 20/2 = 10
(-) A 13
difference 20
Interaction Effects
Effect Estimate +/- standard error Average 64.25 ± 0.7 *
Main effects
Temperature T 23.0 ± 1.4 * Concentration C -5.0 ± 1.4 * Catalyst K 1.5 ± 1.4 Two-factor interactions
T × C 1.5 ± 1.4 T × K 10 ± 1.4 * C × K 0.0 ± 1.4 Three-factor interactions
T × C × K 0.5 ± 1.4
Estimated Effects
• Effect of concentration is to reduce yield by ca. 5 units, regardless of other variables
• Effects of temperature and catalyst cannot be interpreted separately - large T × K interaction
- Difference in sensitivity to temperature change for each catalyst - Catalyst A : 13 units; catalyst B : 33 units
• Catalysts obtained from different suppliers, but were supposed identical
• Yield from catalyst B at 180°C highest so far seen
• Careful study of catalyst in further iterations
Conclusions
A + B + C D + other products Solvent E
• Object - Maximize yield of D with respect to A
• Known conditions give ca. 45% yield
• Reactions of this type should give ca. 75%
• Experimental error < 1%
• Five factors varied
- Amount of solvent E x
1- Amount of C x
2- Concentration of C x
3- Reaction time x
O.L. Davies, “The design and Analysis of Industrial Experiments”, 2nd Edn, Longmans, London, 1979
Example of Steepest Ascent
Factor Levels for 1
stExperiment
Factor ----Factor
- 1
Level----
+ 1
X
1Amount of solvent E (ml) 200 250
X
2Amount of C (mol/mol A) 4.0 4.5
X
3Concentration of C (%) 90 93
X
4Reaction time (hours) 1 2
X
5Amount of B (mol/mol A) 3 3.5
Selection of Initial Trials
• Full factorial requires 25 (32) trials
• Analysis will give 32 parameters
- Average yield (1)
- Main effects (5)
- 2-variable interactions (10) - 3-variable interactions (10) - 4-variable interaction (5) - 5-variable interaction (1)
• Many effects will be insignificant
• Should be able to get important information from fewer
trials
Trial Factor level Yield (%)
X
1X
2X
3X
4X
5y
1 - 1 - 1 - 1 - 1 - 1 34.4
2 - 1 - 1 - 1 1 1 51.6
3 - 1 1 - 1 1 1 31.2
4 - 1 1 1 - 1 - 1 45.1
5 1 - 1 - 1 1 - 1 54.1
6 1 - 1 1 - 1 1 62.4
7 1 1 - 1 - 1 1 50.2
8 1 1 1 1 - 1 58.6
Fractional Factorial Design - First
Experiment
Calculation of Parameters - (1)
b
048.5 β
0−β
145−β
235−β
1234b
17.9 β
1−β
45β
234−β
1235b
2-2.2 β
2−β
35β
0134−β
1245b
36.0 β
3−β
25β
124−β
1345b
40.4 β
4−β
15β
123−β
2345b
50.4 β
5−β
14−β
23β
12345b
6-1.8 β
13β
24−β
125−β
345b
70.2 β
12β
34−β
135−β
245Real para- meters
Estimated parameters
Path of Steepest Ascent
x
1(ml)
x
2(mol)
x
3(%)
x
4(hr)
x
5(mol)
Yield (%)
6 Path of steepest ascent
represented by a series of possible trials on it
225 4.25 91.5 1.5 3.25 235 4.22 92.0 1.5 3.25 245 4.19 92.4 1.5 3.26 255 4.17 92.9 1.5 3.26 265 4.14 93.3 1.5 3.27
Trial (9) 275 4.11 93.8 1.6 3.27 80 285 4.08 94.2 1.6 3.28
Trial (10) 275 4.06 94.7 1.6 3.28 79.4
395 4.03 95.1 1.6 3.29
Factor ----Factor - 1
Level----
+ 1
X
1Amount of solvent E (ml) 280 310
X
2Amount of C (mol/mol A) 3.85 4.15
X
3Concentration of C (%) 94 96
X
4Reaction time (hours) 2 4
X
5Amount of B (mol/mol A) 3.5 5.5
Factor Levels for 2
ndExperiment
Trial Factor level Yield (%)
X
1X
2X
3X
4X
5y
11 - 1 - 1 - 1 - 1 1 77.1
12 - 1 - 1 1 1 1 69.0
13 - 1 1 - 1 1 - 1 75.5
14 - 1 1 1 - 1 - 1 72.6
15 1 - 1 - 1 1 - 1 67.9
16 1 - 1 1 - 1 - 1 68.4
17 1 1 - 1 - 1 1 71.5
18 1 1 1 1 1 63.4
Second Experiment
b
0+ β
0(+ β
11+ β
22+ β
33+ β
44+ β
55) = 70.7 b
1+ β
1(+ β
23) = -2.9
b
2+ β
2(+ β
15) = 0.1
b
3+ β
3(+ β
45) = -2.3
b
123+ β
4(+ β
35) = -1.7
b
12+ β
5(+ β
12+ β
34) = -0.4 b
130 (+ β
13+ β
24) = 0.4 b
230 (+ β
23+ β
14) = -0.4
Calculation of Parameters
Enamine synthesis
R. Carlson, L. Hansson & T. Lundstedt, Acfa. Chem. Stand., 1986 (40), 444-452
Works well for most ketones
Self-condensation with hindered ketones
Carlson’s philosophy: Optimise reaction BEFORE searching for new reagents
Design for Competing Reactions
O
HN O
N
O O
heat
+
TiCl
3Response Surface Design
Variables Levels
-1.414 -1 0 1 1.414 x
1: Ratio morpho-
line/chetone (mol/mol)
3.00 3.59 5.00 6.41 7.00
x
1: Ratio TiCl
3/chetone (mol/mol)
0.50 0.57 0.75 0.93 1.00
x
3: Temperature (°C) 60 60 80 100 108
Entry X1 X2 X3 Y1 Y2
1 -1 -1 -1 41.6 14.6
2 1 -1 -1 45.1 6.7
3 -1 1 -1 51.7 26.2
4 1 1 -1 64.7 17.7
5 -1 -1 1 47.8 11.9
6 1 -1 1 57.1 17.5
7 -1 1 1 63.6 26.1
8 1 1 1 77.8 11.0
9 1.414 0 0 66.7 8.1
10 -1.414 0 0 49.5 22.2
11 0 1.414 0 70.4 18.9
12 0 -1.414 0 43.9 80.0
13 0 0 1.414 66.4 9.0
14 0 0 -1.414 52.4 17.3
15 0 0 0 56.5 13.6
16 0 0 0 60.0 12.3
17 0 0 0 56.6 12.6
18 0 0 0 57.2 13.6
Response Surface Design - Results
Ratio Molpholine/ketone
(3) (5) (7)
0
-1 1 2
-2
20
30
40
40 50 60 70 80
10
Yield of enamine
Yield of
by-products
Isocontour Projection
purified by batch distillation to remove various by-products
Factorial Design in a Continuous Process
N O
Et
HS NHCOEt
N HN
OH
N HN
S
NHCOEt O N
Et
NHCOEt
RO S NHCOEt
By-products H
2S, 180-220°C
BMEP
BMEP
R
Factorial Design in a Continuous Process
• Because process continuous, experiment had to be run on plant
• Factors varied - Flow rates
- Temperatures at various sections of reactor - Length of reactor
• Optimised for yield and quality
- Complex because some impurities react in 2nd step to give correct product
• Yield increased from 85% to 91% in 1 week
x
1(ml)
x
2(mol)
x
3(%)
x
4(hr)
x
5(mol)
Yield (%) 295 4.0 95.0 3.0 4.5
Trial (19) 285 4.0 94.5 2.6 4.4 80.8 275 4.0 93.9 2.2 4.3
Trial (20) 265 4.0 93.4 1.8 4.2 84.0 Trial (21) 255 4.0 92.8 1.4 4.1 81.5
Path of Steepest Ascent -
2nd Experiment
Ratio Molpholine/ketone
(3) (5) (7)
0
-1 1 2
-2 -1 0 1 2
20 50
60 70
10
Yield of enamine
Yield of
by-products
Isocontour Projection
Temperature
Ratio TiCl
3/ketone (52)
(80) (108)
0
-1 1 2
-2
(0.5) (1.0)
-1 0 1 2
20 40
50 60 70 80
10
Yield of enamine
Yield of
by-products
0
Isocontour Projection
Preparative Scale Run
To 1L 3-neck flask, Herschberg stirrer, condenser, dropping funnel, cooled in ice-bath
180 ml (2.05 mol, 10 eq.) morpholine added in 220 ml petr. ether
22.5 ml (0.206 mol, 1 eq.) TiCI
3in 80 ml pet ether added over 10 min with vigorous stirring
Heated to reflux (ca. 120°C)
20.09 (0.20 mol, 1 eq.) MTBK in 80 ml petr. ether added rapidly
Refluxed one hour; mixture became very thick
100 ml petr. ether added to help agitation
Mixture cooled, filtered through sintered glass
Solvent & excess morpholine removed under reduced pressure
Crude product fractionated over 25 cm Vigreux column
Yield 28.9 g enamine (85%) at 70-71°C/8 mmHg
Two Pats for OVAT Optimization
20 40 60 80 100
Conc. (g/g)
20 40 60 80 100
Temperature °C 40 60 Path 1 80
Path 2
50 70 90
Number of factors
k
Treatment Three-level
factorials 3
kCombinations Composite
2
k+ 2k + 1
2 9 9
3 27 15
4 81 25
5 243 43
5 81 (1/3 fraction) 27 (1/2 fraction)
6 729 (1/3 fraction) 77
6 243 (1/3 fraction) 45 (1/2 fraction)
Experiments Required for 3k Factorial
& Central Composite Designs
Star points &
centre points
(-1,1,1) (1,1,1)
Factorial points
(1,1,-1)
(1,-1,-1)
(s,0,0)
(0,-s,0) (0.0,-s)
(0,s,0)
A B C
Central Composite Design
3 (desired product)
Case Study - X-ray Contrast Agent
N O OH O
I
HN OH
OH
O O
O
O I HN
O
I
HN OH
OH
O I
HO
OH
O I
OH HO
O
N O OH O
I N
HO
OH OH
O
O HO
OH
O
I N
O OH O
I
HN OH
O O
OH
O HO
OH
O I MeOH
NaOH CaCl2
+ etc.
+
I
I I
Determine which variables influence yield
Determine which variables influence selectivity
Increase yield (Y
1) to > 95%
Decrease total by-products (Y
2) to < 2.5%
Find optimum conditions
Objectives
Response Surface Design
Variables Levels
-2 -1 0 1 2
x
1: equiv. NaOH 1.15 1.30 1.45 1.60 1.75
x
1: equiv. CaCl
20.45 0.80 1.15 1.40 1.85
x
3: Temperature (°C) 17.5 25.0 32.5 40.0 47.5
x
4: subs. conc (g/ml) 0.65 1.00 1.35 1.70 2.05
x
5: equiv. reagent 1.25 1.40 1.55 1.70 1.85
Entry X1 X2 X3 X4 X5 Y1 Y2
NaOH CaCl2 Temp. Substrate Reagent (3) (by-products)
1 -1 -1 -1 -1 1 81.3 1.39
2 1 -1 -1 -1 -1 90.8 1.49
3 -1 1 -1 -1 -1 89.3 2.24
4 1 1 -1 -1 1 94.1 3.62
5 -1 -1 1 -1 -1 87.9 2.42
6 1 -1 1 -1 1 93.9 3.21
7 -1 1 1 -1 1 90.9 2.21
8 1 1 1 -1 -1 93.4 4.41
9 -1 -1 -1 1 -1 66.2 1.58
10 1 -1 -1 1 1 83.9 2.02
11 -1 1 -1 1 1 85.8 1.43
12 1 1 -1 1 -1 90.0 2.30
13 -1 -1 1 1 1 58.0 1.29
14 1 -1 1 1 -1 89.8 4.06
15 -1 1 1 1 -1 89.8 2.23
16 1 1 1 1 1 93.6 3.41
Response Surface Experiments & Results - (1)
Entry X1 X2 X3 X4 X5 Y1 Y2
NaOH CaCl2 Temp. Substrate Reagent (3) (by-products)
17 -2 0 0 0 0 73.4 1.12
18 2 0 0 0 0 91.1 3.27
19 0 -2 0 0 0 59.2 1.26
20 0 2 0 0 0 92.7 2.59
21 0 0 -2 0 0 79.8 0.96
22 0 0 2 0 0 94.0 3.36
23 0 0 0 -2 0 91.2 2.53
24 0 0 0 2 0 83.1 1.39
25 0 0 0 0 -2 90.0 2.73
26 0 0 0 0 2 86.1 1.59
27 0 0 0 0 0 88.7 2.43
28 0 0 0 0 0 89.8 2.57
29 0 0 0 0 0 90.4 2.49
30 0 0 0 0 0 88.8 3.48
Response Surface Experiments &
Results performed in random order - (2)
Y
1Y
2Average 88.21 2.40
Linear effects x
1(NaOH) -4.87 +0.75 x
2(CaCl
2) +5.86 - x
3(temp) - +0.50 x
4(Substrate) -3.41 -0.21 x
5(Reagent) - - Interactions x
1x
2-3.02 - x
2x
4+2.89 - x
2x
5- +0.32 x
3x
5-2.00 - Quadratic
Effects x
2x
2-2.75 -
Response Surface Analysis -
Significant Effects
Temperature (C) = 32.50 Substrate (g/ml) = 1.350 alkyl agent = 1.550
P r o d u c t
Ca Cl
2(e q )
1.2 1.4 1.6
1.0 1.5
0.5
90
95
Example of Contour Plot - Product
Temperature (C) = 32.50 Substrate (g/ml) = 1.350 alkyl agent = 1.70
Ca Cl
2(e q )
NaOH (eq)
1.2 1.4 1.6
1.0 1.5
0.5
3
2
1
0
Example of Contour Plot - Product
Optimum Conditions
• NaOH 1.70 equiv.
• CaCl
21.40 equiv.
• Temperature 10°C
• Substrate 0.91 g/ml
• Reagent 1.60 equiv.
• Yield of product (calcd) 99.2%
• Yield of product (obs) 96.8%
• By-products (calcd) 2.49%
• By-products (obs) 2.31%
Temperature (C) = 10.00 Substrate (g/ml) = 0.910 alkyl agent = 1.600
P r o d u c t
Ca Cl
2(e q )
NaOH (eq)
1.2 1.4 1.6
1.0 1.5
0.5
90
95
100
Chosen conditions
Response Surface Around Chosen
Conditions
Temperature (C) = 10.00 Substrate (g/ml) = 0.910 alkyl agent = 1.600
B y - p r o d u c t s
Ca Cl
2(e q )
1.2 1.4 1.6
1.0 1.5
0.5
90
100
Chosen conditions
Response Surface Around Chosen
Conditions
NaOH (eq) = 1.700 CaCl
2(eq) = 1.400 alkyl ag. = 1.600 P r o d u c t
S ub s tr a te ( g/ ml )
Temperature (°C)
10 20 30
1.2 1.5
0.9
95 96
Chosen conditions 1.8
+
97 98
99.5
Response Surface Around Chosen
Conditions
Evolutionary Operations
IS the process operating under OPTIMUM conditions?
• production processes
• continuous or batch (large number of batches)
DELIBERATE programme of replacing the STATIC operation of a plant by CONTINUAL PROCESS MODIFICATION
LEADS EVENTUALLY TO IMPROVED
PROCESS
Optimize on: Time 19-31 h Responses: % Product
temp 10-30°C % SM
Stoichiom 1-1.68 moles % By-products
conc. 3-7 vol. of solvent
impurity
%
time lactone
Starting material
product
Silyl Group Hydrolysis
Et3N×3HF O N
R
OSiMe2But
R1
O N R
OSiMe2But
R1 HN
O O
R OH
R1
NMP +
Product
Impurity Time
86% 89%
92%
95%
92%
Time
34 1 2
0.5
Temp Temp
Temp
SM
After optimization 94.2% yield 1.9% impurity Solvent 3.5 vol
TEA•HF 1.18 Time 23 h Temp 10°C
Analysis
20
40 60
80
60
40
A B 20
C
Temperature Concentration
Simplex
20
40 60
80
60
40
A B 20
C Concentration
D F E
G H
Simplex (2)
F. L. Chubb, J.T. Edward & S.C. Wong J.Org. Chem. 1980, 45, 2315-2320
Bucherer-Bergs Reaction
OH
C N O
NH3
NH
HCN
C N N
O SH H
C N
OH NH3
S N H
O
N H
NH N
H
O
N H NH3
NH2 S
N C O
NH N
H
O
S
Yield in Bucherer-Bergs Reaction
Experiment N° 1 2 3 4 5 6 7
Solvent (% EtOH) 50 75 50 50 50 50 50
[NH
3] 4 4 6 4 4 4 4
[COS] 2 2 2 1 2 2 2
[HCN] 2 2 2 2 4 2 2
Temperature (°C) 53 53 53 53 53 43 53
Time (hr) 4 4 4 4 4 4 2
Yield (%) 49 69 57 31 32 36 59
Calculation of Next Vertex
Variable 1 2 3 4 5 6
(i) Sum of values (excluding rejected)
325 26 12 14 308 22
(ii) Centroid of retained levels = sum/n*
54.16 4.33 2 2.33 51.33 3.67 (iii) Rejected vertex
50 4 1 2 53 4
(iv) Displacement = (ii)-(iii)
4.16 0.33 1 0.33 -1.67 -0.33 (vi) New vertex (8) = (ii) + (iv)
58.32 4.66 3 2.66 49.84 3.34
Second Simplex
Experiment N° 1 2 3 4 5 6 7 8
Solvent (% EtOH) 50 75 50 50 50 50 50 58
[NH
3] 4 4 6 4 4 4 4 4.7
[COS] 2 2 2 1 2 2 2 3
[HCN] 2 2 2 2 4 2 2 2.7
Temperature (°C) 53 53 53 53 53 43 53 50
Time (hr) 4 4 4 4 4 4 2 3.3
Yield (%) 49 69 57 31 32 36 59 65
Progress of Simplex
30 50 70 90
5 10 15 20
Experiment N°
Y ield (%)
Initial trials
Subsequent
trialsBucherer-Bergs Reaction - Conclusions
Effect of [C] on yield
- Organic chemist’s instinct is to increase concentration to improve yield
- Simplex shows this would not be effective
Reaction time
- Increasing reaction time gives lower yield
- Despite unreacted cyanohydrin and ammonia
- Degradation to imino compound?
C E
3E
2E
1B A D
Modified Simplex Method
Reflected point gives best response
- Move further in same direction - E
1 Reflected point worse than all retained points
- Move a little in reverse direction - E
2 Reflected point worse even than discarded point
- Move back towards discarded point - E
3Calculation of Next Vertex
A B C D E F G
(i) Sum of levels (excluding rejected)
36 195 766 470 2.7 10,000 465
(ii) Centroid of retained levels = sum/n*
5.14 27.9 109 67.1 0.386 1430 66.4 (iii) Rejected vertex
4 45 122 50 0.5 1200 55
(iv) Displacement = (ii)-(iii)
1.14 -17.1 -13.0 17.1 -0.114 230 11.4 (vi) New vertex (8) = (ii) + (iv)
6.28 10.8 96 84.2 0.272 1600 77.8
Case Study - Boots (UK)
Example
A +B +C product A+B by-product B is expensive
Object - to reduce cast without jeopardising quality
Current operating conditions
- Solvent 7500 kg
- Reagent C 1 equivalent
- Quality < 0.8% residual C - Cost £3.10 per kilo
Experimental design
- 16 experiments
- 2 variables (equivalents of C, amount of solvent)
- 2 responses (residual C, material cost)
Expt Solvent Kg
Reagent C Quality (res. C)
Cost (£/Kg)
1 7000 1 0.41 3.078
2 5000 1 1.65 3.161
3 7000 2 3.50 3.099
4 5000 2 7.01 3.314
5 5207 1.5 4.19 3.233
6 6793 1.5 1.46 3.056
7 7500 1.207 0.46 3.260
8 7500 0.793 0.61 3.132
9 7500 1.5 1.60 3.080
10-16 repeats of 1 & 9
1 repeats 0.39-0.51 3.076-3.084
Boots Case Study - Experimental Design
Boots Case Study - Quality Response Surface
6000 7000 8000 9000
Kg of solvent 0.5
1.0 1.5 2.0
M oles of C
0.5 1.0 2.0 3.5 5.0 6.0
Boots Case Study - Cost Response Surface
6000 7000 8000
Kg of solvent 0.5
1.0 1.5 2.0
M oles of C
3.05 3.10 3.15 3.20
Original
Operating
Conditions
Solvent C(mol) %C
predicted actual
7000 1.75 2.37 1.99
7000 1.75 2.37 2.10
7000 1.25 0.72 0.69
6750 2.21 4.75 5.81
6750 1.75 2.12 2.29
6750 1.75 2.12 2.15
Boots Case Study - Conclusion
Optimum conditions
- 6800 kg solvent
- 1.4 equivalents reagent C
- extra cast of reagent C was taken into account in the design
Confirmatory Experiments
Boots Case Study - Final Conditions
Solvent 6750 kg
Reagent C 1.5 equivalents
Cost € 2.46 / kg
Quality 1.3% C
Saving € 0.065 /kg
(€ 81,000 per year)
Evolutionary Operation
IS the process operating under OPTIMUM conditions?
• production processes
• continuous or batch (large number of batches)
DELIBERATE programme of replacing the STATIC operation of a plant by CONTINUAL PROCESS MODIFICATION
LEADS EVENTUALLY TO IMPROVED
PROCESS
Evolutionary Operation
Box 1956
Manufacturing process should yield INFORMATION as well as PRODUCT
Effect of process variables on PERFORMANCE
Carried out on plant
Requires process operators to change the process for
EVERY batch
Evolutionary Operation
Process operating at 85% yield
Assume that this is NOT optimum
Don’t know contours
Choose 4 new conditions
Run A → D + X several times
Some runs give >85%, some less
- (plant manager note!)
Soon established STATISTICALLY that conditions B are
better than X
170 180 190 190 190 190 190 190 1.0
1.01 1.02 1.03 1.04 1.05 1.06
S toic hiom et ry
A B
X
C
E F
G D
93 92 90 85
80
70
Response Surface
Evolutionary Operation
• Repeat cycle with B and X and 3 new points
• E gives best results
• Repeat cycle with E (can use 4 new points if desired)
• Continue until optimum (93%) is reached
• May take up to 8 cycles of 5 × (say) 5 experiments = 200 batches
• Once optimized on these 2 parameters, choose 2 more to
optimize
Response Surface
170 180 190 190 190 190 190 190 1.0
1.01 1.02 1.03 1.04 1.05 1.06
S toic hiom et ry
B X E
93 92 90 85
80
70
Disadvantages/Advantages
Disadvantages:
• Cost in time / money
- Training - record keeping
• Changes can only be small to avoid “failed batches”: thus large number of batches required to statistically prove results
• More COMPLEX production routine; errors may occur
• If process already at optimum changes will lead to poorer performance
Advantages
• Uses actual plant equipment / scale
• Observed changes are real!
• Gain improved knowledge of process
• Process operators become MORE INVOLVED and motivated
Which Processes are Suitable?
High volume production
Where potential benefits are large
Process NOT at optimum
Variables can be altered readily
Process stabilises rapidly after a process change
Response (yield, cast, quality, throughput) can be rapidly obtained and measured
G. P. Rangaiah (Ed.), Multi-Objective Optimization : Techniques and Applications in Chemical