Optimisers
Optimisers
(OP)
I guess we've all seen Optistruct. I can't say I was especially impressed by some aspects of it.
Does anyone have any experience using any other optimisers (including those built into post processors)?
Has anyone used an optimiser with beam elements, and if so, can you give me a web site or a rough description of how it worked? I am particularly interested in optimising for static stiffnesss for a minimum mass while working within tight geometrical constraints. This is very closely related to increasing flexural modal frequencies, but subtly different.
Reason I ask is that I have written one in matlab, and, to some extent it works, but having written a few other optimisers for other problems I'm aware that it is (a) inefficient (one change at a time, no memory) and (b) susceptible to finding local optima only. I can't see a nice way of using GA in this, unless I have as many genes as there are elements, which is possible, but rather ungainly.
It basically works by randomly changing properties (selected from a library) for a random element and seeing if k/m increases or decreases, if it gets worse it goes back to the previous version and tries again. There's also some skullduggery to encourage restarts and to eliminate dead ends.
Having said that it does solve reasonable size problems quickly and seems quite sensible in its results.
Does anyone have any experience using any other optimisers (including those built into post processors)?
Has anyone used an optimiser with beam elements, and if so, can you give me a web site or a rough description of how it worked? I am particularly interested in optimising for static stiffnesss for a minimum mass while working within tight geometrical constraints. This is very closely related to increasing flexural modal frequencies, but subtly different.
Reason I ask is that I have written one in matlab, and, to some extent it works, but having written a few other optimisers for other problems I'm aware that it is (a) inefficient (one change at a time, no memory) and (b) susceptible to finding local optima only. I can't see a nice way of using GA in this, unless I have as many genes as there are elements, which is possible, but rather ungainly.
It basically works by randomly changing properties (selected from a library) for a random element and seeing if k/m increases or decreases, if it gets worse it goes back to the previous version and tries again. There's also some skullduggery to encourage restarts and to eliminate dead ends.
Having said that it does solve reasonable size problems quickly and seems quite sensible in its results.
Cheers
Greg Locock
Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.





RE: Optimisers
RE: Optimisers
Garland E. Borowski, PE
Borowski Engineering & Analytical Services, Inc.
www.borowskiengineering.com
RE: Optimisers
I noticed that, since you were using Matlab and its functions, you were working with a gradient based optimization routine (like bi-conjugate gradient).
As known those methods converge to a local stationary point (min/max) since those methods are based on a taylor approximation of second order.
To overcome this limit one should check this method from any arbitrary point. Among others, genitic algorithms or simulated annealing algorithms should give better answers.
ModeFrontier is tought able to do such a kind of genetic like optimazation.
Regards,
Mastro
RE: Optimisers