Probabilistic Model Based on Bayesian Model Averaging for Predicting the Plastic Hinge Lengths of Reinforced Concrete Columns

Abstract

A probabilistic model is devised for predicting the plastic hinge lengths (PHLs) of RC columns. Seven existing parametric models are evaluated first using a comprehensive database comprising PHL measurements from 133 RC column tests. It is observed that the performances of these seven models are fair (as opposed to strong), and their predictions bear significant uncertainties. A novel technique is devised to combine them into a weighted-average supermodel wherein the weights are determined via Bayesian inference. This approach naturally produces the weights’ statistical moments, and thus, the resulting model is a probabilistic one that is amenable for performance-based seismic design and assessment analyses. Prediction comparisons indicate that the proposed supermodel has a higher performance than all prior models. The new model is easily expandable should more test data become available.