[Ipopt] ipopt in structural optimization
Clausen Peter M.
peter.clausen at fe-design.de
Wed May 27 12:06:32 EDT 2009
Hello
I am working at FE-DESIGN GmbH (www.fe-design.com) and our
program TOSCA Structure is a structural optimization
program. We use the Method of Moving Asymptotes (MMA of
Krister Svanberg) as optimizer which in most cases does a
really good job. In some cases though it has its drawbacks
(linear object function, constraints which become active
later in optimization) and we would like to test an
alternative.
Our problems may exceed 1 mio. design variables and we
usually only need around 30-50 iterations to converge. The
primal solution is a FEM system solved by an external
program, the sensitivity analysis (1. order) is done by
ourselves. We have no 2. order information.
My questions:
1) How would you expect Ipopt to perform in a benchmark
against MMA ? Has anybody got experiences?
2) We warmstart the optimization each iteration and
leave the optimization to calculate the new primary
solution. Thus we have changed the MMA so that we call it
with objective function, constraints and sensitivities of
the actual iteration plus some saved information from
former iterations, e.g. the last asymptotes and other
values. Could this be easily done with Ipopt ?
We would prefer to use the fortran version, but
interfacing to C++ is nothing new to us. It would simply
mean that Ipopts objects would be created and destroyed
each iteration (would this be a performance problem?).
Best regards
Peter M. Clausen, developer at FE-DESIGN GmbH
_______________________________________________
Peter M. Clausen
Email: peter.clausen at fe-design.de
Tel. : +49 721 96467 - 236
_______________________________________________
FE-DESIGN GmbH, Haid-und-Neu-Str. 7, 76131 Karlsruhe,
Germany
Registration Court: D-Mannheim HRB 107382
Director: Dr. Juergen Sauter
Fax: +49 (0)721 96467-290
Web: www.fe-design.com
_______________________________________________
--
This email was Anti Virus checked by Astaro Security Gateway. http://www.astaro.com
More information about the Ipopt
mailing list