Machine learning based framework for the seismic fragility assessment of reinforced concrete bridges

The evaluation of the seismic vulnerability of existing bridge portfolios is of great concern for transport network managers especially in earthquake-prone countries. The not-negligible number of bridges designed in the past without adequate anti-seismic requirements, the poor structural knowledge of these structures and the material degradation pose challenges in the evaluation. This study proposes a framework for large scale seismic vulnerability assessment of multi-span reinforced concrete girder (RC) bridges considering knowledge-based uncertainties and a novel bridge taxonomy based on risk attributes collected in the prioritization steps of the multi-level approach of the Italian guidelines. It is based on subsequent steps that involve the input of basic knowledge data (e.g., the maximum bridge span length), the simulation of knowledge-based uncertainties, and simplified seismic analysis. An Extreme Gradient Boosting Machine Learning algorithm is investigated to define a binary classificatory able to predict whether the pier failure occurs and is used for the seismic fragility assessment through generated fragility curves. A case-study section demonstrates the application of the framework in the case of circular RC piers showing promising results for seismic risk-informed prioritization of bridges and suggesting the possibility of extending it to other bridge-pier systems.