A Bayesian Inference Driven Computational Framework For Seismic Risk Assessment Using Large-Scale Nonlinear Finite Element Analyses

Nuclear engineers are increasingly relying on large-scale simulations particularly for seismic risk assessment. Experimentally validated simulation models are used to consider the effects of uncertainties and evaluate fragilities by conducting multiple nonlinear analyses. However, such an approach becomes computationally prohibitive and care is needed to achieve desired degree of accuracy with a reasonable amount of computational effort. In this paper, a statistical framework is presented to minimize the total computational effort needed in conducting large-scale simulations for seismic risk assessment. The salient features of the framework are: (i) use of Bayesian inference to allow consideration of data from diverse sources like experiments, field data, existing or simplified approaches, and data from large-scale simulations, and (ii) embedment of Bayesian methods within an iterative process to plan and allocate adequate computing resources such that the desired accuracy is achieved using minimum possible simulations. The applicability and efficiency of the proposed framework is illustrated using the example of a box-shaped reinforced concrete shear wall.