Estimating Probability of Improvement in Next Cycle of Surrogate Based Design Optimization

University Of Florida, Gainesville FL

Investigators

Abstract

The research objective of this award is to provide accurate estimates of the probability of achieving a target level of improvement in the next surrogate based design optimization cycle, hence increasing the cost-effectiveness of the overall design process. A major challenge in the design optimization of complex engineering systems such as those found in the automotive (e.g., crashwordiness), aerospace (e.g., liquid rocket injector), oil (e.g., chemical flooding) and biomechanical (e.g., knee implants) industries is how to meet the conflicting objectives of improved performance, reduced costs and enhanced safety considering the time constraints present in industrial environments. Surrogate-based design optimization (SBDO), which proceeds in stages or cycles, has shown to be an effective approach to address this challenge. It refers to design optimization that relies on the use of surrogates or metamodels (computationally inexpensive) built/updated on reduced samples of computationally expensive/high fidelity simulations collected at each of the cycles. Each cycle is costly, and in spite of the progress, the current state-of-knowledge regarding SBDO lacks statistics-based stopping criteria for assessing the merit of investing in another cycle of optimization versus accepting the present best solution. If successful, the research will address the shortcomings of present industrial practice through the use of alternative surrogate model estimation and appraisal information to provide a rigorous stopping criterion for SBDO. The research will be validated addressing one of the most significant problems in biomechanical engineering: the design of artificial knees for minimizing wear and hence avoiding the need for a second replacement. Wear analysis of artificial knees is computationally very costly and the limit on number of simulations makes it an ideal application of SBDO. The results are expected to be disseminated through computational tools which will be freely available via the internet, and conference presentations, journal publications and engineering design optimization courses. The research will benefit from an ongoing collaboration with Brazilian, Venezuelan, and French researchers.