Component damage models for detailed seismic risk analysis using structural reliability methods

Abstract

A probabilistic approach to modeling the seismic damage incurred by structural and nonstructural components is developed in this study. The resulting models are intended for use in detailed risk analysis of structures through structural reliability methods. The proposed modeling approach employs Bayesian linear regression to characterize the epistemic uncertainty in the model. In this approach, regressors are selected and constructed based on the mechanics of damage. In turn, the probability distributions of the model parameters and the model error are quantified by Bayesian linear regression using real-life data from laboratory experiments and/or post-earthquake inspections of damaged components. The paper presents in detail the stepwise procedure of collecting the observations, quantifying the repair costs, developing model forms, performing model diagnostics, and conducting model reduction, i.e., eliminating the inconclusive terms from the model. Subsequently, a library of Bayesian damage models are developed using the proposed approach for structural components, such as concrete moment frame joints, and nonstructural components, such as partition walls and traction elevators. For this purpose, the data of a host of past experimental studies are collected for Bayesian regression analysis. The proposed Bayesian damage modeling approach provides a number of insights into the parameters that are influential on damage that are discussed for the proposed models.