Engineers are constantly challenged with the task of creating better products. Design optimization is one of several modern tools that have been developed to meet this challenge. A common issue in optimization is that of finding a reliable design. A design may nominally pass all requirements, but when it is put into service stochastic variations in performance may cause the design to fail more often than is acceptable. The preferred solution is to perform reliability-based design optimization (RBDO), which accounts for uncertainty. In order to evaluate the reliability of a given design, considerable computational cost is necessary.The work presented in this thesis outlines a new strategy for performing RBDO where local response surfaces are used to reduce the computational burden of predicting the reliability. This strategy also takes advantage of the fact that the deterministic design optimization (DDO) and RBDO solutions are often spatially nearby each other, resulting in the opportunity to use inexpensive and well-worn DDO techniques to get a head start on finding the RBDO solution. The DDO study also provides ample data that can be used to fit response surfaces during the RBDO stage, without the need for additional evaluations. This new strategy is applied to several problems, including simple analytical functions and real engineering problems.
Read
