Analytical seismic performance and sensitivity evaluation of bridges based on random decision forest framework

Abstract

Various bridge portfolios and modeling parameters will influence the seismic response of bridges differently. These features are typically fixed prior to modeling bridges, while there is inherentuncertainty associated withchoosing them. Sensitivity analysis of analytical seismic demands with respect to the changing bridge attributes helps to improve estimated seismic demand models which are eventually used in the reliability assessment of bridges. To this end, the current study implements statistical approaches such as analysis of covariance to evaluate the impact of common bridge portfolios such as abutment types on the primary engineering demand parameters such as deck displacement. Moreover, this paper proposes a machine learning algorithm, Random Forest ensemble learning method, to assess the level of importance of modeling parameters on estimating seismic demands. The framework is presented for analyzing concrete box-girder bridges with tall piers that are typically constructed in response to the complex topography of the construction site such as mountain or valley regions. However, the proposed framework is applicable to other types of bridges. Furthermore, although previous research revealed distinctive seismic performance for bridges with tall piers compared to the bridges with ordinary configurations, there is still a lack of understanding of the variability of their seismic demands. Thereby, the findings of this study provide a better understanding of the seismic performance of this class of bridge.