Toward a structured approach to simulation-based engineering design under uncertainty

This dissertation deals with the design and development of structure approaches to the effective and efficient simulation-based engineering design (SBED) under uncertainty. Specifically, it addresses the challenging issue of how to build an effective surrogate model in an efficient manner for engineering design optimizations. Along with the advancements in computer technology, computer simulations in lieu of physical experiments have been increasingly used in engineering design analysis. Physics-based high-fidelity numerical models that are safe to operate, easy to modify, and can be automated for design optimizations are typically employed. The use of such numerical models, however, can be computationally prohibitive in optimization scenarios. As a countermeasure, cost-effective surrogate models built as various statistical approximations have been applied for better efficiency. In spite of their popularity and increasing usages in the design process, surrogate models nonetheless demand a substantial number of expensive simulation runs in order to obtain adequate data for building accurate approximations. Moreover, the limited sampling schemes are very likely inadequate for capturing completely the critical features of a to-be-designed product, since there is no a priori knowledge about the unknown system. As a result to this challenging uncertainty, the SBED process could either remain time-consuming, or result in sub-optimal designs due to modeling errors. To this end, this dissertation has developed PREference-based Surrogate Modeling (PRESM) method, Drowning Flood method, Clustering-based Multi-Location Search (CMLS) method, and CMLS-integrated Hybrid method. It is based on the hypothesis that by integrating designer\u27s preference in the building and/or validation of surrogate models, their subsequent implementations will lead to significant time reduction in SBED optimization without any compromise of the expected optimal design outcomes. The associated case studies demonstrate the improved efficiency and enhanced effectiveness of SBED from implementing these innovative methods. Not only can the developed methods be used to overcome SOBP defects, but also they can help to find multiple local/global optimal points simultaneously. The outcome of the dissertation advances the state of knowledge on the understanding of the importance of engineering model selection and its impact to both the optimal design decisions and the overall modeling cost, leading to the identification of a methodical approach to validate and enhance the relevance of simulation-based predictive models