Bridge seismic hazard resilience assessment with ensemble machine learning

Abstract

Recent years have seen a paradigm shift in assessing the performance of assets in response to disruptive hazards, in that resilience is seen as a more inclusive and over-arching decision variable. This shift in decision drivers provides a better picture of asset behavior in response to hazardous events such as earthquakes and hurricanes. Since highway bridges are among the most critical and vulnerable components of transportation networks, evaluating their functionality under extreme events leads to well-informed decision-making. Whilst there is an ever-growing interest in resilience-based hazard assessment in a wide range of infrastructure sectors, there is limited attention on identifying resilience drivers as a function of hazard and asset characteristics. To this end, this paper presents a framework for probabilistic resilience assessment of a cohort of common highway bridges subjected to a wide range of ground acceleration intensities. This study presents the first ensemble learning-based predictive model using bagging and boosting techniques to predict resilience index as a function of seismic events and asset characteristics on bridge resilience. The hypermeters and input structure of the predictive model are optimized to reduce complexity and maximize efficiency. The findings show that the proposed model performs with a 75–95% success rate in predicting resilience as a function of structural characteristics and peak ground acceleration. This model provides useful insights on the impact of various parameters and drivers of resilience in concrete box-girder bridges.