Probabilistic Seismic Capacity Model of Pier Columns: A Semiparametric Regression Approach

Abstract

Piers are usually the most vulnerable components in a bridge structure and generally undergo excessive deformation, which will lead to damage and even whole structural collapse. This paper investigates the probabilistic seismic deformation capacities of reinforced concrete piers under different limit states for two engineering demand parameters, i.e., the drift ratio and displacement ductility. Based on sample data from the UW-PEER database, a penalized generalized additive model is used for predictor variable selections and to determine whether the mechanism of each predictor on the seismic capacity is linear or nonlinear. The influence of a predictor that illustrated a nonlinear pattern is modeled by a Gaussian process, and Bayesian semiparametric regression is conducted in the R environment to obtain posteriori estimations of the capacity measures. The results indicate that the ratios of the model predictions to the experimental observations are all around 1.0, which proves the unbiasedness of the models. Compared with previous seismic capacity models, the prediction of seismic capacity measures shows higher accuracy, lower dispersion, and better portrayal of uncertainties. The proposed model based on Bayesian semiparametric regression provides a performance improvement in the seismic capacity evaluation of the bridge structures, which can be used for the subsequent bridge seismic fragility and risk assessment.