Title: The Essentials of 2-Level Design of Experiments Part I: The Essentials of Full Factorial Designs
1The Essentials of 2-Level Design of
ExperimentsPart I The Essentials of Full
Factorial Designs
- Developed by Don Edwards, John Grego and James
Lynch Center for Reliability and Quality
SciencesDepartment of StatisticsUniversity of
South Carolina803-777-7800
2Part I The Essentials of Full Factorial Designs
- Some Motivation and Background
- Two Important Advantages of Factorial
Experiments - The Essentials of 2-Cubed Designs
- Full Factorial Designs
3I.1 Some MotivationArno PenziasChief Scientist
and VP for Research, Bell Labs Nobel
Laureate-Physics Teaching Statistics to
Engineers, Science Editorial, June 2 1989
- Statistical Tools Are Needed In Industry
- Competitive Position Demands It
- Optimizing Complex Technological Manufacturing
Processes Requires It
4I.1 Some Motivation
- Leaders In Quality
- Use Statistics At All Process Stages For Quality
and Optimization Purposes - Provide The Necessary Statistical Training To Do
This
5I.1 Some Motivation QS9000
- QS9000 required that The supplier shall
demonstrate knowledge in Design of Experiments
(DOE) and use it as appropriate.
6I.1 Some MotivationExamples of DOE Applications
- NCR has used factorial designs in a variety of
situations, e.g., to analyze computer performance
and to compare different soldering methods. - Sara Lee Hosiery Division has used simple designs
in a number of settings. Several have resulted
in considerable annual savings.
7I.1 Some MotivationExamples of DOE Applications
- Ohio Brass has conducted several fractional
factorial designs which have had significant
impact. One study resulted in an annual savings
of 25K by modifying an existing process and
avoided a capital investment of a 1/4 to 1/2
million dollars in new equipment. Another
enabled them to reduce the dimensions of two key
components which resulted in annual savings of
50K.
8I.1 Some MotivationExamples of DOE Applications
- Michelin has used designs to determine
maintenance programs for some of their machinery.
9I.1 BackgroundWhy Should You Use DOE?
- Intelligent Decisions Should Be Based On
"Informed Observation And Directed
Experimentation" (George Box) - It is consistent with the Scientific Method which
is fundamental to the quality management
philosophy (The Deming-Shewhart PDSA Cycle) - DOE is a formalism that forces you to organize
your thoughts (so you don't do things haphazardly)
10I.1 BackgroundWhy Should You Use DOE?
- DOE Concentrates Your Efforts
- Screening designs aid in identifying the
vital/critical factors that may affect the
(process) response of interest - More refined design techniques determine the
factor levels that optimize the response
11I.1 BackgroundWhy Should You Use DOE?
- DOE Concentrates Your Efforts
- DOE helps you to understand how factors affect
the process. This knowledge helps to choose
factor settings that are cost effective but dont
compromise quality (constrained optimization).
12I.1 BackgroundQuality Management Philosophy
- Some Tenets Related to These Components
- All processes have variation
- Different types of variation
- e.g., common cause system verses special causes
being present - Management needs predictable/stable processes to
make decisions (process needs to be in control,
i.e., a common cause variation system)
13I.1 BackgroundQuality Management Philosophy
- Implications for DOE
- The smaller the effects you are trying to detect
relative to the background variation, the more
replication you need or a different design
(blocking) - Data from an out-of-control process is suspect
14I.1 BackgroundContrasting SPC and DOE
- Statistical Process Control (SPC)
- SPC is used to determine if a process is in
control - An Out-of-Control process that is brought into
control is not process improvement (Juran)
15I.1 BackgroundContrasting SPC and DOE
- Design of Experiments (DOE)
- A methodology useful for determining
- what factors may affect a response
- what factor settings are feasible
- SPC Lets You Listen to the ProcessDOE Allows
You to Converse With It
William Hunter
16I.1 Background Experimentation
- Experiment
- A series of trials or tests which produce
quantifiable outcomes. - Quantifiable Outcome
- Some Outcome Measurement of Interest
- Response Variable (y)
17I.1 Background Examples of Responses
- Yield
- Viscosity
- Computer Performance
- Breaking Strength of Fiber
- Smoothness of Polyurethane Sheets
- Bowing of a Molding
- Chain Length in Polymer
- Number of Flaws
18I.1 Background Responses- Bowing of a Molding
- Three Moldings
- Top - Most Severe Bowing
- Bottom - Flat, No Bowing
19I.1 Background Responses- Bowing of a Molding
True versus Substitute Quality Characteristics
- The Displacement D
- Substitute Quality Characteristic for Bowing
- Measurable
20I.1 Background Factors
- Experimental (Variable) Conditions That May
Affect the Response. - A. Flow rate of a raw material
- B. Process temperature
- C. Presence/Absence of a Catalyst
- D. Raw Material Supplier (e.g. 1,2, or 3)
21I.1 Background Factors
- Factors May Be
- Continuous (A and B Above)
- Discrete (C and D Above)
22I.1 Background First Motivation To Experiment
- To Improve The Response.....
- Optimize average response
- Minimize variability in response
- Minimize susceptibility to uncontrollable noise
factors
23I.1 Background Best Motivation
- To Understand The Response! (George Box)
- Levels of Understanding
- Which?
- How?
- Why?
24I.1 Background Levels of UnderstandingAn
Example - Yellowfin Tuna Growth
- Traditional Theoretical Growth Models Allow For
Only One Point of Inflection(Two Growth Stages)
25I.1 Background Levels of Understanding How
StageAn Example - Yellowfin Tuna Growth
- Lowess Fit Suggests
- Two Points of Inflection
- Rethink Theory
26I.1 Background Levels of Understanding How
StageAn Example - Yellowfin Tuna Growth
- More Pronounced In The Atlantic Ocean Yellowfin
Tuna
27I.1 Background Levels of Understanding Why
Stage An Example - Yellowfin Tuna Growth