XANSYS Message: 21360 [Go back to message list]
[bookmark on del.icio.us]
No rating yet
Rate item:

Subject: Re: Poll results for xansys - optimization
Author: Shen-Yeh Chen
Date: 2001-02-26 11:28:00

I think the numerical optimization is, just like all other computer
programs, a tool to help you solve the problem, but it is not going
to solve ALL your problems. This is just like FEA. FEA will help you
solve your design problems, but not ALL your design problems. Today
there are still a lot of companies believe FEA is not critical at
all, and it is not worth the effort. It may be true, and may not
be true. Same as numerical optimization.

Optimization is focusing on solving problem with complicate
interaction between constraints/objective/etc.. Optimization WILL
make you life easier if you know how to use it. For example, perhaps
everyone already agrees today, that the parametric model is the way
to go. Ok, we have a parametric model, we can iterate it easier,
but, except for making mistake faster, are we doing anything
better ? I doubt.

Optimization was originated from mathematical programming.
The name "programming" implies its nature very well - it is a
mathematical model and methodology, you got to figure out the best
way to fit your problem into the formulation.

I have seen a lot of failed examples from the optimization process.
From my own limited experience, 90% of them are due to improper
modeling and problem formulation. And, of course a good optimizer
is very important. It is really sad for a researcher in the
field of optimization, to see some software vendors claiming they
have such and such capability, and understand the software is doing
something definitely wrong but there is nothing he can do about it.
Some optimization software vendors just try to attract as many
customers as possible, without even considering if their package
can do it. After these users get discouraged, or get the wrong
education, they usually become the attackers of this technology.

Believe it or not there are some very famious software vendors
doing something definitely wrong, which makes them only able to
solve very simple cases. The users usually get excited about the
advertisement, and begin to use it. But as the software doesn't
get them anywhere for practical problems, they get disappointed,
and jump to the conlusion that optimization is not practical at all.
Or, even worse, the user believe that the program is correct and
think there is no way to improve the design anymore. I am not
talking about ANSYS. The optimizer in ANSYS is not very strong
but ANSYS defintely knows its limitation and never tried to make
the user very excited. I imagine that is the reason ANSYS leaves
a programable interface to other optimizers. And, the architecture
of ANSYS is almost for optimization.

For the issue of parametric model, it was heavily studied in
the 80's, and today several softwares are already able to rebuild
the parametric model upon existing FE mesh (only). Usually this is
then used to do preliminary optimization and give a good idea
about what the shape should look like. Then the information will
be given to the designer, and after the prliminary design it is
given to the analysts again for checking and very detailed
design. At this stage, usually there is not a lot you can change,
and there will be a lot of geometric constraints. Then, if the
analysts want to use optimization, the skills will really come
to play now. I submitted a article regarding this issue to the
ANSYS solution magazine in the end of last year, but I am not
quite sure what happens since then.........

This is only based my own experience though.

Shen-Yeh Chen
Honeywell Engines & Systems
Phoenix, Arizona

--- In xansys@y..., "Altidis, Paris (Bellwood)" wrote:
> The ideal scheme would be :
> a) Knowledge base Engineering (Not necessarily in electronic form) -
to weed
> out unsuccessful scenarios. Experience may still be under your
nose.
> Problem is, we don't look there.
> b) Use your optimizer for the remainder.

> Paris Altidis
> Staff Engineer
> Borg Warner Automotive
> 708-547-2719
> paltidis@a...

> > -----Original Message-----
> > From: Christopher Wright [SMTP:chrisw@s...]
> > Sent: Friday, February 23, 2001 11:30 AM
> > To: XANSYS
> > Subject: Re: [xansys] Poll results for xansys - optimization

> > >I think that I can overcopme the "fiddle factor" with
optimization
> > >if I see that the payoff is there.
> > I may as well chip in here. There's really not much general value
in
> > optimization because it's restricted to a very narrow area.
Problems like
> > 'optimal' fillet radii are trivial and not worth learning the
necessary
> > vocabulary, let alone waiting around for results. Real engineering
optima
> > involve matters such as cost, schedule and producibility. Weight
savings
> > is sueful, but only if it comes with a cost or time savings. I
once
> > worked on a rib-stiffened cylindrical shell where the 'optimized'
> > structure really saved a lot of weight by adding a great many
small,
> > closely spaced stiffeners. But the fabrication costs were
monstrous--it
> > wasn't an optimum anything.

> > Engineering designs often require specific types of
components--you're
> > stuck with the off-optimum just because plate and fasteners and
rolled
> > shapes in any old size you want.

> > The biggest problem (apart form inability to import parameterized
CAD
> > geometry) is that by the time you have enough information to
'optimize' a
> > design, it's already pretty much cast in concrete, particularly if
you're
> > waiting on a CAD model. A good example is a pressure vessel with a
flat
> > head--you can figure out the 'optimum' combination of head and
shell
> > thickness for the least material, but FEA won't explain that a
flat head
> > is a rotten idea from the outset.

> > OTOH, I daresay situations involving proof of concept lend
themselves to
> > both parameterization and optimization but the use of small models
for
> > studying physics seems to have gone out of fashion.

> > Christopher Wright P.E. |"They couldn't hit an elephant from
> > chrisw@s... | this distance" (last words of Gen.
> > ___________________________| John Sedgwick, Spotsylvania 1864)
> > http://www.skypoint.com/~chrisw

Posts possibly associated with message #21360AuthorDateScore
21301Poll results for xansys - optimizationDan Bohlen2001/02/23 
21303Re: Poll results for xansys - optimizationDanny Levine2001/02/23 
21311Re: Poll results for xansys - optimizationParis Altidis2001/02/23 
21321Re: Poll results for xansys - optimizationChristopher Wright1991/02/23 
21323Re: Poll results for xansys - optimizationMark Tate2001/02/23 
21325Re: Poll results for xansys - optimizationParis Altidis2001/02/23 
21334Re: Poll results for xansys - optimizationChristopher Wright1991/02/23 
21342Re: Poll results for xansys - optimizationFrank Exius2001/02/24 
21360Re: Poll results for xansys - optimizationShen-Yeh Chen2001/02/26 
21371Re: Poll results for xansys - optimizationParis Altidis2001/02/26 
25378Re: Poll results for xansys - optimizationChris Rogers2001/06/28