|
|
Type of Document Master's Thesis Author Chandila, Parveen Kumar Author's Email Address p_chandila@yahoo.co.in URN etd-04142004-144454 Title Strategy for Global Optimization and Post-Optimality Using Local Kriging Approximations Degree Master of Science in Mechanical Engineering Department Aerospace and Mechanical Engineering Advisory Committee
Advisor Name Title Dr. John E. Renaud Committee Chair Dr. Mihir Sen Committee Member Dr. Stephen M. Batill Committee Member Keywords
- global optimization
- Kriging
- approximation
- response surface
Date of Defense 2004-04-07 Availability unrestricted Abstract A strategy for global optimization based uponlocal approximate optimization has been developed
in this research. The global design space is
divided into local regions. Accurate Kriging
approximations are constructed in the local
regions to provide an inexpensive formulation
of the system behavior. Each Kriging approximation
is validated by using a cross-validation scheme.
Gradient based local optimizers are used to
identify local optima in each of the regions.
A reduced trust region around the local optimum
is further analyzed to provide the global
optimum solutions. This approach possesses the
capability of identifying global optimum solutions
within reduced time frames as compared to
existing methods.
A second investigation focused on the formulation and development of a
nonlinear post-optimality analysis. The
non-linear response of the system is captured
through local approximations. A cumulative
approximation is constructed from the local
response surface approximations via a blending
function. The post-optimal solution is obtained
by performing optimization over the cumulative
approximation of the objective function and the
constraints.
Files
Filename Size Approximate Download Time (Hours:Minutes:Seconds)
28.8 Modem 56K Modem ISDN (64 Kb) ISDN (128 Kb) Higher-speed Access ChandilaP042004.pdf 640.95 Kb 00:02:58 00:01:31 00:01:20 00:00:40 00:00:03