[BANANA] LA/Opt seminar Wed May 7 (Brian Roth)

[Michael A. Saunders]

NOTE: Special room: 380-380C (Math Corner of quad)
       Refreshments served at 4pm

   Linear Algebra and Optimization Seminar (CME510)
   iCME, Stanford University
   http://icme.stanford.edu/seminars/seminars.php

   4:15pm Wed May 7, 2008
   Math Corner 380-380C

   Brian Roth
   Aircraft Aerodynamics and Design Group
   Dept of Aeronautics and Astronautics, Stanford University
   bdroth at stanford.edu
   Group Website: aero.stanford.edu/adg.html


   Aircraft Family Design Using Enhanced Collaborative Optimization

Significant progress has been made toward the development of
multidisciplinary design optimization (MDO) methods that are
well-suited to practical design problems. However, opportunities
exist for further progress. This talk describes the development
of enhanced collaborative optimization (ECO), a new
decomposition-based MDO method. In decomposition-based methods,
the design problem is divided along natural boundaries between
disciplinary groups. Each disciplinary team is responsible for
satisfying their own design constraints while seeking to match
design targets. Currently, coordination between disciplinary
teams is limited to shared design targets. In ECO, the
disciplinary optimization problems are augmented with constraint
models from the other disciplinary teams. These constraint
models, which are constructed through the use of post-optimality
sensitivity analysis, aid each disciplinary team in making
choices that are better aligned with the preferences of other
disciplinary teams.

The ECO method offers several significant contributions. First,
it resolves several troubling computational inefficiencies that
are associated with current methods. Second, it enhances
communication between disciplinary design teams without
increasing the dimensionality of coupling between them. As a
result, ECO provides significant computational savings for the
test cases and aircraft design problems presented in this talk.