design of experiments for real world chemical systems

17
6/17/2019 1 Design of Experiments for Real World Chemical Systems James N Cawse Cawse and Effect LLC www.cawseandeffect.com 1 Chemistry offers unique challenges. Mixtures Kinetics Reactor systems Nonlinear relationships Nonquantitative factors and responses 2

Upload: others

Post on 25-Dec-2021

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Design of Experiments for Real World Chemical Systems

6/17/2019

1

Design of Experimentsfor

Real World Chemical Systems

James N Cawse

Cawse and Effect LLC

www.cawseandeffect.com

1

Chemistry offers unique challenges.

• Mixtures

• Kinetics• Reactor systems

• Nonlinear relationships• Nonquantitative factors and responses

2

Page 2: Design of Experiments for Real World Chemical Systems

6/17/2019

2

Why use Design of Experiments?

• Coverage

• Efficiency

• Signal to Noise

• Interactions

• Mathematical models

[a‐,b‐,c‐] A

C

[a+,b‐,c‐]

[a‐,b+,c‐]B

[a‐,b‐,c+]

BC [a‐,b+,c+] ABC[a+,b+,c+]

AC [a+,b_,c+]

AB[a+,b+,c‐]

3

Why use Design of Experiments?

• DOE is great for many scientific disciplines. Because…

• Factors are ordered

• Simple interactions 

• Not too many dimensions

• Space is relatively smooth

• Linear models hold!Y = B0 + B1x1 + B2x2 + B12x1x2 + B11x1

2 +…

[a‐,b‐,c‐] A

C

[a+,b‐,c‐]

[a‐,b+,c‐]B

[a‐,b‐,c+]

BC [a‐,b+,c+] ABC[a+,b+,c+]

AC [a+,b_,c+]

AB[a+,b+,c‐]

4

Page 3: Design of Experiments for Real World Chemical Systems

6/17/2019

3

Why…?

• DOE is great for many scientific disciplines. Because…

• Factors are ordered

• Simple interactions 

• Not too many dimensions

• Space is relatively smooth

• Linear models hold!Y = B0 + B1x1 + B2x2 + B12x1x2 + B11x1

2 +…

5

Mixtures

6

Page 4: Design of Experiments for Real World Chemical Systems

6/17/2019

4

Mixtures

• Chemists make mixtures of “stuff”• Catalysts• Drugs• Personal care• Other chemicals…

Active Ingredients     Citric Acid 1000 mgAspirin 325 mgSodium Bicarbonate 1916 mg

Active Ingredients     Sodium Fluoride      0.24%Triclosan 0.30%

Inactive Ingredients     water, hydrated silica, glycerin, sorbitol, PVM/MA copolymer, sodium lauryl sulfate, cellulose gum, flavor, sodium hydroxide, carrageenan, propylene glycol, sodium saccharin and titanium dioxide,

EO catalyst Ingredients     SilverChlorideSodiumPotassiumCesium…

Black ink composition WO 2016048234 A1A black ink composition for post print and preprint, obtainable by adding (colour concentrate mixture) solid red concentrate, blue concentrate, and green concentrate to a carbon black ink …, and the colour pigments  are made up 75‐90% by weight.                                                            

7

BUT... Real Components may…

• Not range 0 to 100%

• have proportional constraints

• have one component which is a solvent

Extreme VerticesA

B

CC

B

B < 60%

A < 80% C < 75%

C > B

8

Page 5: Design of Experiments for Real World Chemical Systems

6/17/2019

5

Mixture Systems may….• have an effect proportional to amount

• be in a system where process variables count

9

Mixture Systems may….• have an effect proportional to amount

• be in a system where process variables count

Low

High

Fertilizers or Insecticides

10

Page 6: Design of Experiments for Real World Chemical Systems

6/17/2019

6

Catalyst Mixture Study

• Screening: 8 components• B, Fe, Ga, La, Mg, Mn, Mo, and V

• Select in groups of 4 at constant total concentration• e.g.  {B, Fe, Ga, La};  {Ga, La, Mg, Mn};   {La, Mg, Mn, Mo}

• Combin(8,4) = 70

• xMixture(4) ≈ 20

• Total runs     ≈ 1,400

11

Catalyst Study

100% Mg

100% Ga

100% V

100% Mo

12

In Design‐Expert:

Page 7: Design of Experiments for Real World Chemical Systems

6/17/2019

7

Basic Chemistry: Kinetics

13

Basic Chemistry

• What is being made?

• Where is the reaction going?

• How fast is it getting there?

• Fundamental kinetics:

• Rate = f(concentrations, temp)

• Order• Rate = k*C1

• Rate = k*C1*C2 …

• Synthesis

• Equilibrium

• Kinetics

Multiplicative, not additive!

Remember: Linear Models!:Y = B0 + B1x1 + B2x2 + B12x1x2 + B11x1

2 +…14

Page 8: Design of Experiments for Real World Chemical Systems

6/17/2019

8

“Simple” Chemical Kinetics

When do you take your data??

0

0.2

0.4

0.6

0.8

1

0 1 2 3

fraction complete

Reaction time (hrs)

Second Order reaction

0

0.2

0.4

0.6

0.8

1

0 1 2 3

Fraction complete

Reaction Time (hrs)

First Order Reaction

15

“Simple” Chemical Kinetics

When do you take your data??

0

0.2

0.4

0.6

0.8

1

0 1 2 3

fraction complete

Reaction time (hrs)

Second Order reaction

0

0.2

0.4

0.6

0.8

1

0 1 2 3

Fraction complete

Reaction Time (hrs)

First Order Reaction

Reaction too fast! Noisy data!

Reaction over! No useful data!

Fraction complete

Time

ABC

Two Stage reactionA   B  C

What (and when) do you sample?

16

Page 9: Design of Experiments for Real World Chemical Systems

6/17/2019

9

Kinetics in a Reactor

• Batch

• Tubular

• Continuously stirred tank

Feed

Product

Cooling JacketFeed Product

Feed

Feed

Product

Product

Cooling Jacket

Cooling Jacket

Flow

Flow

17

Nonlinear Reactor kinetics

• Batch

• Tubular

• Continuously stirred tank

190        230        270

Feed Product

Flow

0

0.2

0.4

0.6

0.8

1

0 1 2 3

completion

Time (hrs)

Temperature

Da = reaction rate    Flow

Conversion =    Da1+ Da 

18

Page 10: Design of Experiments for Real World Chemical Systems

6/17/2019

10

Tubular Example

• Butyrolactone hydrogenation

H2, catPBT polyester

Reactant

FLOW RATE

Product

Residence time 

Completion

Reactant

Mass Transfer

Mass Transfer

Product

Diffusion Zone

19

Hard Core Gas‐Solid Kinetics

• Langmuir‐Hinshelwood Equation

• Standard linear regression  hopeless interactions

• Better bet: modified exponential model regression• Ln(rate) = f(ln PA, lnPb, ln PA* lnPb…)

A

BAB

Catalyst particle

krKA

KB

Rate(A + B ‐> AB)  =kr KA KB PA PB

(1+ KA PA + KB PB ) 

… Jean Cropley

20

Page 11: Design of Experiments for Real World Chemical Systems

6/17/2019

11

Basic Chemistry: Catalysis

21

Catalyst Chemistry: Increasing Complexity and Decreasing Returns

• Ethylene Oxide Reaction

22

Page 12: Design of Experiments for Real World Chemical Systems

6/17/2019

12

Ethylene Oxide: Chemistry and History

U.S Patent April 10, 1990 4,916,243

1980’s: Silver, Potassium, Cesium%K

%Cs

Two‐factor interaction

1940 1950 1960 1970 1980 1990 2000 2010 2020

Catalyst Selectivity100

90

80

70

60 Ag

Theoretical limit?

+Na

+Cl

+Na/K+Mn… +Mn+1…+Re

23

2016: High Level Interactions

• Modeling five cocatalysts!!• All main effects

• 7 of 10 2‐way interactions• But…

Term t Ratio Prob>|t|

Intercept 94.33 <.0001*

M1(M2) 3.25 0.0030*

M2 4.71 <.0001*

M3 0.61 0.5484

M4 9.71 <.0001*

M5 ‐9.4 <.0001*

(M1)*(M3) ‐7.73 <.0001*

(M1)*(M4) 7.73 <.0001*

(M1)*(M5) ‐11.94 <.0001*

(M2)*(M3) ‐2.56 0.0163*

(M2)*(M5) ‐1.61 0.1182

(M3)*(M4) ‐3.13 0.0041*

(M3)*(M5) 9.19 <.0001*

M1 M1*M3

M3                     M3*M5                     M1*M3*M5        

M5                     M5*M1

(M1)*(M3)*(M5) 7.48 <.0001*

(M2)*(M3)*(M5) 9.27 <.0001*

(M1)*(M1) 3.59 0.0013*

(M3)*(M3) 3.23 0.0032*Ignore 3‐way interactions….!!??

24

Page 13: Design of Experiments for Real World Chemical Systems

6/17/2019

13

It gets worse…The Law of diminishing returns

• As you get closer to the theoretical limit – any improvement disappears in the noise!

Remaining improvement!

25

Multistep Reactor Study

• New Zeolites:

• Every vessel is a reactor... and an experiment!

Mix

Zeolite sub‐typeBinder type

Exchange

TimeTemperatureNo. of exchanges

Calcination

Order

Steaming

Mix

Zeolite

Binder

Exchange

Temperature

No. Exchan

ges

Order Calcination Steaming

Loss of Information!

Measurement

26

Page 14: Design of Experiments for Real World Chemical Systems

6/17/2019

14

Nested Design

Mix

Zeolite sub‐typeBinder type

TimeTemperatureNo. of exchanges

Order

Z1B1 Z1B2 Z2B1 Z2B2

Temp

No. Exc

Temp

No. Exc

Temp

No. Exc

Temp

No. Exc ….

….

4 runs x 4 runs x 2 runs = 32 Run Experiment

Exchange

Calcine

27

Basic ChemistryQualitative vs. Quantitative

28

Page 15: Design of Experiments for Real World Chemical Systems

6/17/2019

15

Numbers = Continuous Factors

• Soap side chains

• What’s wrong here?• This is not a continuous factor!• These are distinct compounds – hence qualitative!

• In Design‐Expert: “Discrete Numeric”

/

CH2‐CH2‐CH3 CH2‐(CH2)3‐CH3 CH2‐(CH2)5‐CH3 CH2‐(CH2)7‐CH3

Factor (units) Type(qualitative, quantitative, formulation)

Normal level & range

Measurement precision & setting error

Proposed settings Ease of Adjust.

Chain Length Quantitative 3, 5, 7, 9

Control Factors

29

Osha.gov

Complex Natural Products

Gasoline

30

Page 16: Design of Experiments for Real World Chemical Systems

6/17/2019

16

Factors from “natural”Mixtures• Petroleum impurities problem

• Sample and model an experimental space in two chemical factors using blends of existing oils. All oils must be used!• Oils are clustered at low levels

• Spread out at high levels

0.00

1.00

2.00

3.00

4.00

5.00

0.00 1.00 2.00 3.00 4.00 5.00 0.01

0.10

1.00

10.00

0.01 0.10 1.00 10.00Impurity 1

Impurity 2

1. Transform to a log scale!2. Fit a RSM design3. Perform an iterative search for blends

Impurity 1

Impurity 2

31

MoralBeware force fitting chemistry

into a statistical box!

• In Chemistry:

3+ factor interactions

Nonlinear models

Reactor dependence

Natural product factors

A B

C

32

Page 17: Design of Experiments for Real World Chemical Systems

6/17/2019

17

References

Mixtures Kinetics Reactors Chemistry

[email protected]

Questions?33