UNCERTAINTY QUANTIFICATION IN ENGINEERING OPTIMIZATION APPLICATIONS

Abstract

In this dissertation, we propose three novel methodologies for modeling the uncertainties in engineering design problems. The first work proposes a multilevel zero-inflated model to capture the various types of variations in high-quality manufacturing processes. The second work focuses on the development of Bayesian optimal designs for the efficient estimation of the optimum design setting. The developed framework employs a Shannon information utility measure to quantify the reduction in the uncertainty of the optimum setting from an experiment. In the third work, we look into metamodel-based optimization of stochastic computer models where the objective functions are uncertain. We leverage on the flexible and efficient radial basis function metamodel and a novel experimental design approach to model the objective function as a function of both the design factors and the uncertain objective function parameters. These three developed methodologies together contribute to improving the engineering design process and facilitate robust decisions.