Information Reuse for Importance Sampling in Reliability-Based Design Optimization

摘要: Abstract This paper introduces a new approach for importance-sampling-based reliability-based design optimization (RBDO) that reuses information from past iterations to reduce computational effort. RBDO is two-loop process—an uncertainty quantification loop embedded within an loop—that can be computationally prohibitive due the numerous evaluations of expensive high-fidelity models estimate probability failure in each iteration. In this work, we use existing create efficient biasing densities importance sampling estimates failure. The method involves two levels reuse: (1) reusing current batch samples construct posteriori density with optimal parameters, and (2) designs visited design. We demonstrate benchmark speed reducer problem combustion engine proposed leads savings range 51% 76%, compared building no reuse