Spatial correlation metamodels for global approximation in structuraldesign optimization

Timothy W. Simpson, Janet K. Allen, Farrokh Mistree

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Despite the steady and continuing growth of computingpower and speed, the complexity and computational expenseof engineering analysis codes maintains pace. Statisticaltechniques are becoming widely used in engineering design toconstruct approximations or metamodels of these analysiscodes which are then used in lieu of the actual codes,facilitating optimization and concept exploration. Ourpurpose in this paper is to report results of ongoing researchaimed at increasing the efficiency of computer-basedengineering design through the use of spatial correlationmetamodels to build global approximations ofcomputationally expensive computer analyses. Threestructural design examples are presented to test the predictivecapability of these metamodels for use in design optimization.The reported results confirm that these spatial correlationmetamodels can produce sufficient accuracy for optimizationwhen used as global approximations.

Original languageEnglish (US)
Title of host publication24th Design Automation Conference
PublisherAmerican Society of Mechanical Engineers (ASME)
ISBN (Electronic)9780791880326
DOIs
StatePublished - 1998
EventASME 1998 Design Engineering Technical Conferences, DETC 1998 - Atlanta, United States
Duration: Sep 13 1998Sep 16 1998

Publication series

NameProceedings of the ASME Design Engineering Technical Conference
Volume2

Conference

ConferenceASME 1998 Design Engineering Technical Conferences, DETC 1998
CountryUnited States
CityAtlanta
Period9/13/989/16/98

All Science Journal Classification (ASJC) codes

  • Mechanical Engineering
  • Computer Graphics and Computer-Aided Design
  • Computer Science Applications
  • Modeling and Simulation

Fingerprint Dive into the research topics of 'Spatial correlation metamodels for global approximation in structuraldesign optimization'. Together they form a unique fingerprint.

Cite this