Uncertainty Quantification Working Group
August 9, 11:30 AM, CNLS Conf. Room

Lessons and Innovations from a Case Study in Optimization Under Uncertainty

Vicente J. Romero
Validation and Uncertainty Quantification Dept. 9133
Sandia National Laboratories, Albuquerque

Abstract

"Optimization Under Uncertainty" (OUU) is an emerging discipline that combines technologies from uncertainty analysis and optimization to determine optimal designs, processes, and systems when significant uncertainty exists in the system state and/or its representation. OUU provides a quantitative mechanism for infusing uncertainty (which is ubiquitous in the real world) into the optimization process in order to meet or more closely approach risk/reliability/robustness objectives in the final [optimal] solution. This talk will describe an ongoing OUU case study in which uncertain material properties, component failure criteria, and numerical resolution effects are confronted in an optimization problem to find worst-case heating boundary conditions on a firing set. This "industrial-scale" OUU problem involves the use of a transient 3-D finite element conduction/radiation thermal model. Various optimization and uncertainty-propagation techniques have been applied to aspects of this problem with varying degrees of success, as the talk will survey. Lessons learned, fixes and innovations implemented, future directions, and other issues of pertinence will be discussed in this far-ranging talk.

Send e-mail to author at vjromer@sandia.gov