Large-sample normal approximation and relevant sensitivity in linear seismic fragility analysis with hierarchically modeled stiffness properties

•Hierarchically modeled stiffness properties.•Large-sample Bayesian techniques in linear seismic fragility analysis.•Comprehensive quantification work relevant to the stochastic characteristics.•Detailed sensitivity analysis with respect to the number of observations.•Illustrated convenient and straightforward application of the formulated approximate model.

This paper investigates the potential of the large-sample normal approximation to Bayesian posterior distributions in linear seismic fragility analysis. With the stiffness properties modeled hierarchically, the prior information on the parameters involved in the relevant probability distributions can be updated based on the latest stiffness data acquired, leading to the posterior distributions of these parameters. Using large-sample Bayesian techniques, the posterior distributions may be approximated by normal distributions. The stochastic characteristics of some parameters in the normal distributions are first presented. Specifically, the relationships of the pertinent coefficients of variation and percentage points to the number of the observations for the stiffness properties are examined; a sensitivity analysis with respect to the number of observations is implemented; and the effect of some included distribution types on the stochastic characteristics is analyzed through defined factors. The seismic fragility analysis of a shear frame is then carried out, and comparisons are made between the fragilities from the underlying model and those from the approximate model. The results of the study could be useful in formulating related structural design strategies, provided that adequate amount of the latest stiffness data can be made available through techniques such as structural health monitoring.