From: John Larkin
Subject: Re: Design of Experiments Workshop in NYC, SF, Miami
Date: Sun, 08 Dec 2002 17:59:08 -0800
Organization: Posted via Supernews, http://www.supernews.com
References: <3DF00F90.email@example.com> <firstname.lastname@example.org>
X-Newsreader: Forte Agent 1.91/32.564
On 8 Dec 2002 15:40:53 -0800, email@example.com (Brian Teasley)
>It doesn't surprise me that someone ended up with a bad product in the
>course of using a "Taguchi" approach or any other valid approach to
>running experiments. The "optimal" solution (indicated by statistics
>in the experiment) can certainly be a non-feasible solution for any
>number of practical reasons. Mathematical optimization is pure
>mathematics; the "subject matter expert" must factor in other
>The test designs are simply tools that should be used in conjunction
>with expert subject knowledge.
>(Oh, and as for the nice bibliography/resource listing, my point was
>not that there aren't plenty of books and articles out there on the
>topic, my point was that there were not a lot of good posts withing
>the USENET groups themselves. Perhaps somebody would like to quote the
>favorite passage from one of the books - the part they read that was
>applicable to whatever it was they were working on? Then somewhere
>down the road someone might be doing a keyword search within usenet on
>the topic and can find something valuable to them?)
this is actually an interesting issue. I've read a bit of the DOE
stuff, and that seems, to me, to be nearly my natural enemy.
Optimizing a design through a series of solution-space experiments
directly implies that the experimenter doesn't understand the process.
The fundamental question becomes: where did the design come from?
I design things based on theory, experience, and intuition, and I like
to think that I do everything for a reason, that I fully understand
why it is the way it is and how everything works, and I fully
appreciate that, in most designs, 'optimum' is such a complex concept
that it can't really be quanitified. It annoys me when a 'quality' or
'manufacturing' person tries to mess with my design by blindly
frobbing values; more often than not, they make something worse or
create a hazard, because they break something that was carefully
designed for good reasons.
One cannot use DOE to design bridges, skyscrapers, or space shuttles.
They have to be *right* by design, with small margin for ignorance. If
electronics and software are designed with similar levels of theory
and discipline, there's no room or need for DOE or Taguchi techniques.
Only a few very final tweaks to a design should ever be done
experimentally, and they should have been well anticipated. And if the
improvement is *not* small, the designer doesn't know his trade.
Design by simulation, or optimization by fiddling, teach bad habits.