Probabilistic seismic analysis of concrete dry cask structures

摘要

Dry casks are vertical reinforced concrete cylinders, which are used to store spent nuclear fuel. In a seismic event, impact between adjacent casks, due to excessive horizontal displacements or large rocking angles, might cause damage to the structure, potentially leading to release of radioactive material. This paper analyzes the displacement and rocking response of dry casks subjected to seismic loads and proposes a two-layer probabilistic seismic demand modeling strategy to predict cask response. To enable this probabilistic modeling, virtual experimental data for seismic cask response is derived using a validated 3D finite element model. Latin Hypercube Sampling is used to generate 160 cask-pad-soil configurations across typical ranges of structural, geometric, and material properties, and nonlinear time history analyses are performed with a suite of ground motion records selected representative of regions where storage facilities are located or with moderate to high seismicity. Fivefold cross-validated response surface models are used to predict the probabilistic horizontal and vertical accelerations of the cask center of gravity. Then in the second step of the metamodeling strategy, probabilistic models of cask horizontal displacement and rocking are developed through stepwise regression in which the cask accelerations are tested as predictors in addition to structural, geometric and material parameters. The two-layer approach proposed herein offers an advance over traditional probabilistic seismic demand modeling methods, improves the performance of the developed metamodels and can be employed to study the seismic response of other rigid-body-type structures. The resulting probabilistic models are used in seismic fragility analyses, and parametric studies are conducted to explore the influence of different parameters on the fragility. Finally, the annual probability of failure, defined as the probability of seismic displacement or rocking response exceeding a prescribed limit, is evaluated for different locations in the United States.