Rare event simulation for large-scale structures with local nonlinearities

When modeling large-scale structures, knowledge of the systems and their environment is often incomplete, i.e., wind or wave loads, and possible defects in the structural components, often referred to as uncertainty parameters. Conventional simulations are deterministic and do not take such uncertainty into consideration, making structural reliability analysis, i.e., rare event simulation, feasible. In this work, we use parametrized models to account for the uncertainty. As a result, a rare event simulation becomes a quantication of the impact of all uncertainty parameters on a quantity of interest of the structure, and an essential part of risk assessment. However, traditional parametrized simulations, i.e., through the finite element method (FEM), for large-scale infrastructures, possibly with a moderate to high dimensional parameter domain, is computationally intractable due to multi-query sampling necessitated during the simulation. We adopt non-intrusive reduced order modeling and machine learning techniques as an enabling technique for high-fidelity rare event simulations. We consider dierent approaches for risk assessment for large-scale structures with local nonlinearities with moderate to high dimensional parameter spaces, and provide a comparative study through several numerical examples to demonstrate the feasibility and effectivity of the presented approach.


 Record created 2019-08-18, last modified 2020-01-07

Download fulltext

Rate this document:

Rate this document:
(Not yet reviewed)