doe made easy, yet powerful, with design-expert® software ... · a. culture-medium temperature b....
TRANSCRIPT
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 1
1
Making the most of this learning opportunityStat-Ease worldwide webinars attract many attendees, so, to prevent audio disruptions, all must be muted by the presenter. Also, to avoid interruptions and keep the presentation to about one hour, please hold all questions until afterwards and address them to [email protected].
– Mark
P.S. Find slides posted now at www.statease.com/webinar.html and, barring technical issues, a recording put up afterwards.
By Mark J. Anderson, PE, CQEStat-Ease, Inc., Minneapolis, MN
DOE Made Easy, Yet Powerful, with Design-Expert® Software
DOE Made Easy with DX
0.00 3.60 7.21 10.81 14.42 18.02 21.63
0
10
20
30
50
70
80
90
95
99
Half-Normal Plot
|Standardized Effect|
Half-N
orm
al %
Pro
bability
A-Temperature
C-ConcentrationD-Stir Rate
AC
AD
0.00
1.40
2.80
4.20
5.59
6.99
8.39
9.79
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Pareto Chart
Rank
t-V
alu
e o
f |E
ffe
ct|
Bonferroni Limit 3.82734
t-Value Limit 2.22814
A-Temperature
AC
AD
D-Stir Rate
C-Concentration
The WIIFM* for this Webinar*(What’s in it for me)
This demo of Design-Expert features factorials—the core tool for design of experiments (DOE), plus a few peeks at the peak technology for DOE: response surface methods (RSM).
See how you can quickly converge on the ‘sweet spot’—the most desirable combination of process parameters and product attributes.
Some WIFFMs for you to take home will be:
❖ Appreciation for multifactor testing.
❖ A tried-and-true strategy of experimentation.
❖ Inspiration to use stat tools that can greatly accelerate your research.
DOE Made Easy with DX 2
Please press the raise hand now if you are with me.
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 2
3
Multi-Factorial (VS OFAT)(life from accelerated test)
"To make knowledge work productive will be the great management task of this century."
-- Peter Drucker
A- A+B-
B+
C-
C+
17
19
26
16
25
21
85
128Relative efficiency = 16/8
2 to 1!Start point for One Factor at a Time (OFAT)
Multifactor TestingMany experimenters cannot believe it
What did you do today?
No way! You tested more than one factor at a time?
DOE Made Easy with DX 4
I learned how to juggle, I tamed a wild horse and I did a DOE.
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 3
5
Strategy of Experimentation
DOE Made Easy with DX
6
Screening/Characterization
Purpose: Quickly sift through a large number
of potential factors to discard the trivial
many. Then follow-up with an experiment
that focuses on the vital few.
Tool: Two-level factorial designs:
1. Fractional for resolving main effects in
minimal runs.
2. Full (or less fractional) to resolve two-
factor interactions.
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 4
7
Screening/CharacterizationCase Study
A research scientist at a biotechnology firm reports* the use of DOE
in pilot studies on a pharmaceutical made via recombinant E. coli
fermentation. He screened six factors:
A. Culture-medium temperature
B. Fermentation-medium pH
C. Dissolved oxygen concentration (DO)
D. Induction optical density (OD) – a measure of cell density
E. Feed rate of proprietary complex solution
F. Feed rate of inducer solution
The primary response was harvest OD – a measure of cell-growth.
*(Arun Tholudur, et al, “Using Design of Experiments to Assess Escherichia Coli Fermentation Robustness,” BioProcess – posted at http://www.statease.com/pubs/fermentationrobustness.pdf.)
DOE Made Easy with DX
8
Full and Fractional Two-Level Designs
Factors
Run
s
A ‘happy medium’ for screening 6 factors: adequate resolution (IV) with
decent power (n=8 high vs low).
DOE Made Easy with DX
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 5
9
Experimental design template (1/4th fraction of 26 via std design with E=ABC and F=ABD)
Std A:
Temp.
B:
pH
C:
DO
D:
In. OD
E:
Soln 1
F:
Soln 2
1 Lo (–) Lo (–) Lo (–) Lo (–) Lo (–) Lo (–)
2 Hi (+) Lo (–) Lo (–) Lo (–) Hi (+) Hi (+)
3 Lo (–) Hi (+) Lo (–) Lo (–) Hi (+) Hi (+)
4 Hi (+) Hi (+) Lo (–) Lo (–) Lo (–) Lo (–)
5 Lo (–) Lo (–) Hi (+) Lo (–) Hi (+) Lo (–)
6 Hi (+) Lo (–) Hi (+) Lo (–) Lo (–) Hi (+)
7 Lo (–) Hi (+) Hi (+) Lo (–) Lo (–) Hi (+)
8 Hi (+) Hi (+) Hi (+) Lo (–) Hi (+) Lo (–)
9 Lo (–) Lo (–) Lo (–) Hi (+) Lo (–) Hi (+)
10 Hi (+) Lo (–) Lo (–) Hi (+) Hi (+) Lo (–)
11 Lo (–) Hi (+) Lo (–) Hi (+) Hi (+) Lo (–)
12 Hi (+) Hi (+) Lo (–) Hi (+) Lo (–) Hi (+)
13 Lo (–) Lo (–) Hi (+) Hi (+) Hi (+) Hi (+)
14 Hi (+) Lo (–) Hi (+) Hi (+) Lo (–) Lo (–)
15 Lo (–) Hi (+) Hi (+) Hi (+) Lo (–) Lo (–)
16 Hi (+) Hi (+) Hi (+) Hi (+) Hi (+) Hi (+)
DOE Made Easy with DX 10
Effects: Half-Normal Plot
Half-N
orm
al %
Pro
babili
ty
|Effect|
0 8 16 24 32
015
10
Hi
30
50
70
80
90
95
99
A
D
AD
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 6
How Half-Normal Plots* Help Experimenters*One of Design-Expert’s greatest strengths!
11DOE Made Easy with DX
Half-normal (aka “Daniel”) plots provide many advantages for factorial analysis:
❖ Easy to produce❖ Gives a quick visual of important effects❖ Effects can be compared to a reference—the near-zero
contrasts that emanate from the origin❖ Adapts to split-plot experiments [for hard-to-change factors]
“The greatest single advantage of the Daniel plot is its ability to stimulate discussion by encapsulating, in a single display, such a variety
of information.”
* (David Steinberg, “Discussion: On Daniel Plots”, Journal of Quality Technology, V47, N2,April 2015, p110.)
12
Effects: Ranked
Rank
T-v
alu
e o
f |E
ffect|
0.00
2.38
4.76
7.14
9.51
T Value Limit 2.16037
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
D
AD
A
DOE Made Easy with DX
Design-Expert makes it far easier to interpret this ordered bar chart of effects by it’s innovative enhancements:➢ Limit lines ➢ Color coding*➢ Interactivity*
* Also on half-normal
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 7
13
Interaction Effect
Lo Mid Hi
A: Temperature
Ha
rve
st O
D
48
66
84
101
119
D- Induction OD =Lo
Arun Tholudur's Fermentation
DOE Made Easy with DX
14
Minimum-Run Designs (up to 50 factors)Considerable Savings Over Standard Fractions
Factors Std Res V MR5* Factors Std Res IV MR4**
6 32 22 9 32 18
7 64 30 10 32 20
8 64 38 11 32 22
9 128 46 12 32 24
10 128 56 13 32 26
11 128 68 14 32 28
12 256 80 15 32 24
13 256 92 16 32 26
14 256 106 17 64 28
* Oehlert & Whitcomb, “Small, Efficient, Equireplicated Resolution V Fractions of 2k designs …”, Fall Technical Conference, 2002: www.statease.com/pubs/small5.pdf
** Anderson & Whitcomb, “Screening Process Factors In the Presence of Interactions,” Annual QualityCongress, American Society of Quality, Toronto, 2004: www.statease.com/pubs/aqc2004.pdf
Characterization Screening
DOE Made Easy with DX
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 8
15
RSM
Strategy of Experimentation
DOE Made Easy with DX
16
Response Surface Methods (RSM)When to Apply It
1. Fractional factorials for screening such as MR4
2. High-resolution fractional, such as MR5, or full factorial for characterization of interactions (add center points at this stage to test for curvature)
3. Response surface methods (RSM) for optimization.
Contour maps (2D) and 3D surfaces guide you to the peak.
DOE Made Easy with DX
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 9
17
RSM: When to Apply It
Region of Operability
Region of InterestUse factorial design to get close to the peak. Then RSM to climb it.
DOE Made Easy with DX
18
RSM vs OFAT
DOE Made Easy with DX
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 10
19
RSM: Process Flowchart
Process
Vital Few Factors (x’s)
Measured Response(s) (y(s))
Subject Matter Knowledge(Plus Factorial Screening)
Polynomial Model
Fitting*
Response Surface
“All models are wrong, but some are useful.” - George BoxDOE Made Easy with DX
20
Carbon Monoxide EmissionsCase Study – RSM Design & Analysis
Via a response surface method (RSM) design aimed at minimizing carbon monoxide (CO) emission, Ford Motor UK varied 5 factors :
A. Engine loadB. Engine speedC. Spark advanceD. Air-to-fuel ratioE. Exhaust gas recycle
They used two test engines and set this up in two blocks.
Emission
Rebuild, analyze and optimize
DOE Made Easy with DX
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 11
21
Mixture DesignRSM Tailored to Formulation Space
Considerations:
➢ Factors are ingredients of a mixture.
➢ The response is a function of proportions, not amounts.
❖Given these two conditions, fixing the total (an equality
constraint) facilitates modeling of the response as a
function of component proportions.
DOE Made Easy with DX
22
A (1)
A (0)
B (0)
C (1)
C (0)
260
460
660
860
1060
E
lasticity
B (1)
Mixture Design:Rocket Science
Mixture model (special cubic):
Elasticity=
+ 351 * A
+ 446 * B
+ 653 * C
- 6 * AB
+1008 * AC
+1597 * BC
+6141 * ABCFile: RocketDOE Made Easy with DX
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 12
23
Mixture Case Study
Formulators* measured the effects of three mixture components on
an aqueous dispersion of polymeric nanospheres:
A. Poloxamer 188 NF (“P 188”)
B. Polyoxyethylene 40 monostearate NF (“Pe 40 Ms”)
C. Polyoxyethylene sorbitan fatty acid ester NF (“P.S.F.A.E.”)
They also studied the film-forming properties of this pharmaceutical
preparation. The measured responses were:
1. Particle size, nanometers (< 250 nm desired)
2. Glass transition, degrees C (< 17.5 deg desired)
Is there a sweet spot? See next slide!*(Frisbee, S.E., McGinity, J.W., “Influence of Nonionic Surfactants on the Physical and Chemical Properties of a
Biodegradable Pseudolatex,” European Journal of Pharmaceutics and Biopharmaceutics, V 40, No. 6 (Dec 1994).)
DOE Made Easy with DX
Framing the Sweet Spot for QbD
➢ Graphical optimization now frames the functional “design space” where all modeled responses for a unit operationfall within confidence intervals: Ideal tool for quality by design (QbD), for example, to see the real sweet spot in this pharmaceutical recipe.
24
A: P 188 1.000
B: Pe 40 Ms
1.000
C: P.S.F.A.E.
1.000
0.000 0.000
0.000
Overlay Plot
Particle size: 260.000
Particle size CI High: 260.000
Glass transition: 20.000
Glass transition CI High: 20.000
Glass transition CI High: 20.000
2
File: CEP-Mix
DOE Made Easy with DX
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 13
Plus free DOE appetizersonline!
252525
Stat-Ease Training: Sharpen Up via Computer-Intensive Workshops
Shari Kraber,Workshop Manager & Master [email protected]
Experiment DesignMade Easy
Response Surface Methods for Process Optimization
Mixture and CombinedDesigns for
Optimal Formulations
Designed Experiments forPharma, Life Sciences,
Medical Devices,Assay Optimization,
Oilfield Services,Food Science
Modern DOE for Process Optimization
DOE Made Easy with DX
26
References*
2nd edition 2016
* Taylor & Francis/CRC/Productivity Press
New York, NY.
3rd edition 2015 1st edition 2018
DOE Made Easy with DX
Copyright © 2018 Stat-Ease, Inc. Do not copy or redistribute in any form. 14
The WIIFM* for this Webinar*(What’s in it for me)
This demo of Design-Expert features factorials—the core tool for design of experiments (DOE), plus a few peeks at the peak technology for DOE: response surface methods (RSM).
See how you can quickly converge on the ‘sweet spot’—the most desirable combination of process parameters and product attributes.
Some WIFFMs for you to take home will be:
❖ Appreciation for multifactor testing.
❖ A tried-and-true strategy of experimentation.
❖ Inspiration to use stat tools that can greatly accelerate your research.
PS (per bonus material): Formulators will do best by using mixture design.
DOE Made Easy with DX 27
Now you know.
Statistics Made Easy®
DOE Made Easy with DX 28
Best of luck for your experimenting!
Thanks for listening!
-- Mark
“Statistics used as a catalyst to engineering creation will always result in the fastest and most economical progress.”
- George Box