Description of Course in
Design of Experiments
Audience: The course is targeted for an audience of degreed
engineers, most of whom have some familiarity with the basic concepts of
Statistical Process Control. I specify degreed engineers for mathematical
maturity. Exceptions are feasible, but a judgment must be made on whether a
given person can handle the computational and conceptual work. The background
in Statistical Process Control is helpful because of the relationship of the
two areas.
Duration: The course is best offered as two 2.5 day sessions,
though it can be done as a 5 day session.
Software and computing
options: I teach this course using
either JMP 4.0 or Statistica. Ideally, we should have
one computer per person in the class, though a ratio of two people per computer
is workable. (Learning in this latter case is not as good!)
Class size: I expect a class size of roughly 10 – 25.
Deliverables: Each participant receives one copy of course notes
which contain the overheads used in delivery of the class, and which
incorporate basic concept discussion (mainly in bulletized/Powerpoint
format) as well as extensive screen shots.
Content: Course content follows below. This is highly
adaptable based on specific needs.
Content for One Week Course:
Introduction
to concepts of Design of Experiments --- Comparing processes and recipes using
experiments. Why experiment? -- .5 days
Introduction
to JMP or Statistica --- basic data manipulation
using JMP or Statistica. (about an hour)
Components of variation ---
two and three level components of variation, with and without fixed effects. How to assess process capability. Impact
of components of variation on control charts. -- .5 days
Simple multifactor
experiments – a simple 2 x 2 designed experiment (two factors, two levels each)
to illustrate a variety of concepts – how to design an experiment, how to
analyze the results using ANOVA and graphically, followed by a 2x2x2 experiment
-- 1 days
Fractional factorial Designs
– in this section we explore how to design an experiment in many factors. We
stress 8 and 16 run designs, but show how to use software to design more
general experiments. Foldover designs, follow-up
experiments. Analysis via normal probability plot, as well as other methods, is
discussed in detail. – 2 days
Experimental strategy – In
this section we explore the strategy of experimentation. Specifically we
discuss selection of factors (which ones), levels (how far apart? Centered on
what?), as well as choice of design, all in the context of looking at the
purposes of the experiment -- .5 days
Wrap-up for
Week 1 – Summarization of concepts, discussion of issues of pressing interest. As possible, a small “simulation” in which
participants get to try out their skills of design and analysis of experiments
in a computer simulated world -- .5 days
Week 2 of the course focuses
on the following topics, with emphasis among the topics being chosen based on
interests and needs of the participants
More on fractional
factorials to produce robust processes and products
Use of blocking
as a powerful tool
Split plots as a
means of getting a lot of information from a few lots
Introduction to
response surface methods
Other topics
based on interest.