From: John Larkin
Subject: Re: Design of Experiments Workshop in NYC, SF, Miami
Date: Thu, 12 Dec 2002 19:42:43 -0800
Organization: Posted via Supernews, http://www.supernews.com
References: <firstname.lastname@example.org> <3DF8C323.4428B09B@mindspring.com>
X-Newsreader: Forte Agent 1.91/32.564
On Thu, 12 Dec 2002 12:10:59 -0500, L Smith
>John Larkin wrote:
>> since you actually use DOE, what do you think of my rant, declaring
>> DOE to be the natural enemy of proper first-time design? I try very
>> hard to make me and my engineers do solid, quantitative hardware and
>> software designs that can be put into production with a minimum of
>> tweaking and debugging, and I think that optimization-by-experiment is
>> not only a nasty habit to form, it actually doesn't work in real life.
> I'm late to this thread since I've been without power for the last week
>(part of the Carolina ice-storm you might have heard about recently).
> If I have the attributions right, I believe you earlier mentioned that this
>thread was being posted in an electronics group, and that you believed
>anyone who didn't design his/her circuits right the first time didn't know
>their subject. I'd just like to point out that the thread is also appearing in
>a chemical engineering group where the situation is not as clear-cut as
>you would have it seem.
> In various parts of the 'chemical' industry, if you wait to build a plant
>you understand the process, you might as well not bother building the plant.
>This is because someone else, who is willing to accept a partial understanding
>of the process, will have built their plant first and will have captured the
>market. This is especially true in basic consumer goods (think Proctor & Gamble
>and laundry detergents, for example). In these cases, you want to run the
>fewest tests of the process you can get away with and still get some idea
>of the process conditions you need to run at. You also couple this with your
>understanding of basic principles to make an educated guess about optimum
>conditions, but you are always aware that there may be unexpected
>going on that you don't know about. You also figure there will be a chance
>later to optimize the process - assuming the product is successful.
> This does not mean we can't design processes from first principles. Most
>chemical companies don't even bother to test 'simple' distillations in the lab
>anymore, the simulations are very accurate for these systems.
OK, for a process this makes sense. You can always tweak temperature
controller settings, process times, catalyst concentrations, a million
things like that, and gradually optimize the process (or, more
specifically, the process you *have*) If it takes two years, that's
okay... it just keeps getting better.
Electronics design is very different. We design something very
complex, release the specs, drawings, test procedures, and firmware to
Manufacturing, and then we ship (if we're lucky) a jillion units. Any
change requires ECOs, procedure edits, maybe hard tooling (like PC
board) changes. Some engineers *do* design by futzing with a
breadboard or a simulation until it works, but that is dicey.