A variable-size local domain approach for increased design confidence in simulation-based optimization

摘要: Simulation-based design optimization utilizes computational models that rely on assumptions and approximations. There is a need therefore, to ensure the obtained designs will exhibit desired behavior as anticipated given model predictions. The common approach accomplish validate utilized prior process. However, this practically an impossible task especially for problems with high-dimensional parameter spaces. We have recently proposed different maximizing confidence in generated during sequential simulation-based process based calibrating when necessary within local subdomains of space. In work, size domains was held fixed not linked uncertainty, quantified using Bayesian hypothesis testing. article, we present improved methodology. Specifically, use statistical methodology account uncertainty determine at each stage parametric bootstrapping involves maximum likelihood estimators parameters. continues until domain does change from process, ensuring convergence optimal design. illustrated thermal insulator one-dimensional, linear heat conduction solid slab flux boundary conditions.