Design Of Experiments Made Easy - Cdn.statease

Design of ExperimentsMade EasyBefore we begin All attendees will be muted during the webinar. Please send questions to mark@statease.comwho will answer questions after the webinar. The webinar will be available for viewingafter the presentation.

Synergies between design of experiments and multivariateanalysis: Sum is larger than the partsDesign of experiments Systematic testingof effects Optimised samplingplans to solve definedproblem Number of experimentslarger than number offactors Hypothesis checkingMultivariate analysis Description of data –data driven Identifying main variationsources in data Handles interactionamong factors andamong responses Hypothesis generationsynergynounThe interaction or cooperation of two ormore organizations, substances, or otheragents to produce a combined effectgreater than the sum of theirseparate on/synergy

Together we cover the whole processfrom design to pilot to productionDesign ofexperiments(DOE)Multivariateanalysis (MVA)ProcessAnalyticalTechnology(PAT)

Design of Experiments (DOE)Made Easy (and more powerful) with Version 12of Design-Expert SoftwareBy Mark J. Anderson, PE, CQEStat-Ease, Inc., Minneapolis, Minnesota, USAHosted byTo prevent audio disruptions, all must be muted.Please address questions to mark@statease.com.

References*3rd edition 2015If you see me,please press theraise hand button.DOE Made Easy with DX122nd edition 20161st edition 2018* Taylor & Francis/CRC/Productivity PressNew York, NY.5

The WIIFM* for this Webinar*(What’s in it for me)This demonstration (‘demo’) of Design-Expert features an array ofpowerful DOE tools for quickly converging on the ‘sweet spot’—themost desirable combination of process parameters and productattributes.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 yourresearch.DOE Made Easy with DX126

Multi-Factorial (VS OFAT)(life from accelerated test)1612885B 2619Start point forOne Factor ata Time (OFAT)17B-A-21C Relativeefficiency 16/8 2 to 1!25A C-"To make knowledge work productivewill be the great management task of this century."-- Peter DruckerDOE Made Easy with DX127

Screening/CharacterizationPurpose: Quickly sift through a large numberof potential factors to discard the trivialmany. Then follow-up with an experimentthat focuses on the vital few.Tool: Two-level factorial designs:1. Medium resolution fractional forscreening main effects in minimal runs.2. High resolution full (or less fractional)to resolve two-factor interactions.DOE Made Easy with DX128

Screening/CharacterizationCase StudyThe biggest client of a large pie-maker confronted them with ofunsightly pitting on the bottom crust. Their food scientists ran atrouble-shooting study on these six factors via a two-level fractionaldesign:A. Dough temperature,B. Amount of shortening,C. Shortening temperature,D. Rework,E. Aging,F. Conditioner.They expected the factors to interact. However, time being of theessence, the experiment could not exceed 24 runs.DOE Made Easy with DX129

Standard (Classical) Two-Level DesignsRunsFactorsOnly can afford a screening (yellow)design on 6 factors, which resolvesmain effects, not interactions. Design-Expert to the rescue with modern DOE options!DOE Made Easy with DX1210

Modern Minimum-Run Designs (up to 50 factors)Considerable savings over standard fractionsScreeningCharacterizationFactorsStd Res VMR5*632227648Factors Std Res IV 176428* 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.pdfDOE Made Easy with DX1211

Screening/CharacterizationCase StudyUsing Design-Expert software let’s rebuild this MR5 design so youcan see how it’s done, re-open the file to collect the data, analyze itand, finally, search out the optimal settings (aka “sweet spot”) tominimize pitting (most important!), raw spot and bake shrink (notvery important) to less than 20, 15 and 1.5; respectively. Hide Presenter View (blocks DX projection)Pies-aRebuild, note transformatons, show cube for pittingOptimizeDOE Made Easy with DX1212

Strategy of ExperimentationRSMDOE Made Easy with DX1213

RSM vs OFATDOE Made Easy with DX1214

RSM Case StudyA chemist studied three process factors:A. Time (minutes)B. Temperature (degrees C)C. Catalyst (percent)To optimize two key responses:1. Conversion (%) Maximize (80% or better)2. Activity Target 63 ( 3 allowable)For convenience, the experiment is run in two blocks via a“central composite design” (CCD):1. Two-level factorial with center points.2. Axial runs (star points) plus more center points.DOE Made Easy with DX12RSMRebuild—Show CCD layoutOptimize & Confirm15

Mixture Design*Considerations: Factors are ingredients of a mixture. The response is a function of proportions, not amounts. Given these two conditions, fixing the total (an equalityconstraint) facilitates mixture modeling as a function ofcomponent proportions.Let’s try forcing a factorial design onto a mixture.*(Pioneered by Henry Scheffé, U Cal., 1957)DOE Made Easy with DX1216

Forcing (squeezing?) factorial design on a mixture:Glasses of sugar waterLemonade211DOE Made Easy with DX12Lemons217

Mixture Design and Modeling (sweet!)Two components: Quadratic (synergistic)Ŷ 1x1 2 x 2 12 x1x 2 12 0Response1 4 12 2Lemons plus watertaste better thaneither one alone . 1X1 13/41/21/40X2 01/41/23/41DOE Made Easy with DX1218

Three-Component MixtureBFactorialBACMixtureAC Raise hand if you have used a triangular (ternary) graph.DOE Made Easy with DX1219

Ternary Diagram for Mixture Composition(for example, stainless steel flatware)x1 x2 x3 1X1907050303010103050507070X2DOE Made Easy with DX12This geometry is called a simplex.909010X320

Mixture Case StudyHoping to hit their target for viscosity, while keeping their productfrom becoming cloudy (low turbidity), detergentchemists varied three components:A. Water, 3-5%B. Alcohol, 2-4%C. Urea, 2-4%constraining the total of these active components to 9% while holdingthe 91% of other ingredients constant.Mix-aRebuild, Run, Analyze, Optimize Numerical & GraphicalDOE Made Easy with DX1221

Optimal (Custom) DesignIn this study a paint chemist working for anautomobile manufacturer was tasked to choose: Monomer vendor M1 or M2. Crosslinker type CL1, CL2 or CL3. The optimal mix ofA. Monomer, 5 - 20 %B. Crosslinker, 25 - 40 %C. Resin, 55 - 70 %With these goals for two key response measures:1. Knoop hardness 10.2. Solids content 50%.DOE for Chem & BiochemAutocoatRebuild, Run, Analyze, Optimize Numerical & Graphical22

Conclusion Trim out the OFAT!By making use of multifactor design of experiments(DOE) starting with simple two-level factorials andgraduating to response surface methods (RSM) forprocesses and products (mixture design), you willgreatly accelerate product development and processoptimization. That’s the key. Design-Expert software makes DOE easy, yet powerful.Experimenters do well by this DOE dedicated toolversus a general statistical package. Why use aSwiss Army Knife when you need a screwdriver?DOE Made Easy with DX1223

Much appreciation to Camo Analytics for hostingand thank you for listening!Markmark@statease.com24

By making use of multifactor design of experiments (DOE) starting with simple two-level factorials and graduating to response surface methods (RSM) for processes and products (mixture design), you will greatly accelerate product development and process optimization. Thats the key. Design-Expert