Optimization and nondeterministic analysis with large simulation models: Issues and directions

摘要

Economic and political demands are driving computational investigation of systems and processes like never before. It is foreseen that questions of safety, optimality, risk, robustness, likelihood, credibility, etc. will increasingly be posed to computational modelers. This will require the development and routine use of computing infrastructure that incorporates computational physics models within the framework of larger meta-analyses involving aspects of optimization, nondeterministic analysis, and probabilistic risk assessment. This paper describes elements of an ongoing case study involving the computational solution of several meta-problems in optimization, nondeterministic analysis, and optimization under uncertainty pertaining to the surety of a generic weapon safing device. The goal of the analyses is to determine the worst-case heating configuration in a fire that most severely threatens the integrity of the device. A large, 3-D, nonlinear, finite element thermal model is used to determine the transient thermal response of the device in this coupled conduction/radiation problem. Implications of some of the numerical aspects of the thermal model on the selection of suitable and efficient optimization and nondeterministic analysis algorithms are discussed.