Design of Experiments

Principles and Applications ISBN 91-973730-0-1

L. Eriksson, E. Johansson, N. Kettaneh-Wold, C. Wikström, and S. Wold

How to Optimize Products and Processes

“Which combination of factors will give the best car engine at low cost, with low fuel consump-
tion and minimal pollution?”
A usually complex and costly experimental situation is easily resolved with Design of Experi-
ments (DOE). All factors are considered in a minimal number of experiments, and the results
are verified with recognized statistical methods. The book describes the method in its simple
basis and adds valuable examples from a variety of application areas. The authors provide their
detailed analyses and offer solutions, with the graphical presentation that is the trademark of
Umetrics software MODDE.
“For newcomers in the field of experimental design
“Design of Experiments: Principles and Appli- the book is very useful.”
cations” is 329 pages and the contents range “The book gives a detailed introduction to the princi-
from beginner’s level with initial screening all ples of experimental design and gives excellent illus-
the way up to complex mixtures. The authors are trations of the use of the MODDE software as an in-
tegrated tool in design and analysis of experiments.”
experts in Design of Experiments and have a
Rolf Carlson in Journal of Chemometrics, 15, 2001,
vast experience of application areas from years 495-496
of consulting and lecturing at Umetrics.
Content of “Design of Experiments: Principles and Applications”
Computation of main effects in the Benefits of response specification/Significant model
0 Preface 1 22 case transformation Fourth limiting case – Outside
Why Design of Experiments (DOE)
A second method of understanding Cause of poor model. 2. Curvature specification/Non-significant
is used
main effects Cause of poor model. 3. Bad model
Sectors where DOE is used
A quicker by-hand method for replicates
Three primary experimental computing effects Cause of poor model. 4. Deviating 18 D-optimal design
objectives D-optimal design
Plotting of main and interaction experiments
effects Introduction to PLS regression When to use D-optimal design
1 Introduction (Part 1)
Interpretation of main and Features of the D-optimal
When and where DOE is useful
interaction effects in the 23 case approach
What is DOE?
Computation of effects using least 13 Screening (Part 1) Evaluation criteria
General Example 1: Screening squares fit Selection of experimental objective
General Example 2: Optimization
Fractional Factorial Designs 19 Mixture design
General Example 3: Robustness A working strategy for mixture
A geometric representation of
testing 9 Factorial designs (Part 3) fractional factorial designs
design
The Cake Mix application Introduction to least squares Advanced mixture designs
Resolution of fractional factorial
analysis designs
2 Introduction (Part 2) Least squares analysis applied to 20 Multi-Level qualitative
Laser welding application I
Variability the CakeMix data factors
What to do after screening
Reacting to noise The proper way of expressing Example – Cotton cultivation
Gradient techniques
Focusing on effects regression coefficients Regression analysis of Cotton
Gradient techniques applied to the
Illustration: Cake Mix Use of coefficient and effect plots application – Coefficient plot
laser welding application
Consequence of variability Regression analysis of Cotton
Automatic search for an optimal
application – Interaction plot
3 Introduction (Part 3) point
10 Analysis of factorial
The model concept designs (Part 1) 14 Screening (Part 2) 21 Taguchi approach to
Conceptual basis of semi-empirical Introduction to the analysis of Confounding pattern robust design
modeling factorial designs The Taguchi approach –
Generators
Evaluation of raw data – Replicate Introduction
4 Problem formulation (Part 1) Defining relation
plot Arranging factors in inner and
Stages in the experimental process Resolution
Regression analysis – The R2 /Q2 outer arrays
Specification of factors Plackett-Burman designs
diagnostic tool The classical analysis approach
Specification of responses R2 /Q2 pointing to a poor model 15 Optimization (Part 1) 22 Models of variability:
Selection of model Model interpretation – Coefficient Background to General Example 2
Generation of design plot Distributions
Introduction to response surface
Creation of experimental worksheet Use of model – Response contour methodology (RSM) designs Models of variability - Distributions
plot Overview of composite designs The normal distribution
5 Problem formulation (Part 2) Model interpretation and use in The t-distribution
Qualitative factors at two levels Truck engine application
the case of several responses Confidence intervals
Qualitative factors at many levels What to do after RSM
Regular and irregular experimental Automatic search for an optimal
point
23 Analysis of variance,
regions 11 Analysis of factorial ANOVA
Metric of factors MODDE optimizer applied to the
designs (Part 2) truck engine data Introduction to ANOVA
Metric of responses Evaluation of raw data – Condition ANOVA – Regression model
6 Problem formulation (Part 3)
number 16 Optimization (Part 2) significance test
Evaluation of raw data – Scatter Three-Part full factorial designs ANOVA – Lack of fit test
Overview of models plot Box-Behnken designs
Geometry of models Evaluation of raw data – Histogram 24 PLS
A comparison of Composite, Three-
of response Part factorial, and Box-Behnken When to use PLS
7 Factorial designs (Part 1)
Evaluation of raw data – designs The LOWARP application
Introduction to full factorial
Descriptive statistics of response Properties of classical RSM PLS model interpretation
designs
Regression analysis – Analysis of designs
The 22 full factorial design –
construction & geometry
variance (ANOVA) Statistical Appendix
Regression analysis – Normal 17 Robustness testing References
The 23 full factorial design –
probability plot of residuals Introduction to robustness testing
construction & geometry
Use of model – Making predictions HPLC application
The 24 and 25 full factorial designs
First limiting case – Inside
Pros and cons of two-level full
specification/Significant model
factorial designs
12 Analysis of factorial Second limiting case – Inside
8 Factorial designs (Part 2) designs (Part 3) specification/Non-significant
Main effect of a factor Cause of poor model. 1. Skew model
response distribution Third limiting case – Outside

In all countries: In USA and Canada: In UK:

Umetrics AB Umetrics Inc. Umetrics UK Ltd.
Box 7960 17 Kiel Ave. Woodside House
SE-90719 Umeå Kinnelon NJ 07405 Winkfield, Windsor
Sweden USA Berkshire, SL4 2DX, UK
Phone: +46 (0)90 184800 Phone: +1 973 492 8355 Phone: +44 (0)1344 885605
Fax: +46 (0)90 184899 Fax: +1 973 492 8359 Fax: +44 (0)1344 885410
www.umetrics.com Email: info.se@umetrics.com Email: info.us@umetrics.com Email: info.uk@umetrics.com