Characterizing and Predicting Seismic Repair Costs for Bridges

Abstract

This paper focuses on probabilistic seismic loss estimation of bridges. The results provide several crucial insights on the distribution of seismic repair costs and the reasoning behind their nature—the distribution of seismic repair costs follows a multimodal distribution. Additionally, Gaussian mixture models (GMMs) are proposed to appropriately model the total repair costs in closed form. Furthermore, this study demonstrates that the multimodal distributions for repair costs can be efficiently predicted using neural networks (NNs). To study the repair costs of bridges, this study simulated the responses of 696 bridges, with varying geometry and design parameters, subjected to a suite of synthetic ground motions. Next, uncertainties in component capacities, repair actions, and costs of repair actions were propagated through Monte Carlo simulations (MCS) with 10,000 samples for each bridge. Additionally, for repair-cost estimation, this study also considered the correlation in damage states within all elements of the same component type and among elements belonging to different component types. Studies in the literature usually neglect these correlations and use maximum component damage estimates to assess the repair costs, overestimating the repair costs. The resulting multimodal distributions of repair costs are attributed to uncertainty in repair actions and differences in damage states of elements belonging to the same component types, such as columns, bearings, and abutments. The NN models, along with GMMs, used to predict the multimodal distribution of total repair costs are demonstrated for six cases of study bridges; the predictions are observed to be in good agreement with the actual distributions, underscoring the viability of the proposed strategy to support seismic loss estimation for bridges in seismic zones.