PSI - Issue 78

Mirko Calò et al. / Procedia Structural Integrity 78 (2026) 710–717

711

Nomenclature D pier

Pier diameter

Mean concrete compressive strength

f c f y

Mean steel tensile strength

Pier height

H pier L AVG

Average span length

Amount of corrosion of reinforcement in terms of weight loss

Q S ρ L ρ T

Pier longitudinal reinforcement ratio Pier transverse reinforcement ratio

1. Introduction Bridges are critical components of transportation networks, and their functionality and operability must be ensured by transportation management companies through accurate safety assessments and timely maintenance to avoid collapses and fatalities as the ones in the past chronicles (Calvi et al., 2018). This occurrence is due to environmental conditions as well as material degradation and dangerous natural events, such as earthquakes, that bridges undergo during their service life. Seismic assessment at the portfolio level has historically posed a significant challenge for engineers, irrespective of the assets considered (e.g., buildings or bridges). One of the common strategies employed in this field are the indirect typological approaches (Abarca et al., 2022; Nettis et al., 2024; Ruggieri et al., 2023), especially when dealing with the lack of specific structure-level information in the evaluation. In these approaches, structures are classified according to taxonomies (e.g., SYNER-G described by Pitilakis et al., 2014) which are proposed by considering the key structural parameters considered to influence structural capacity and seismic response. For each taxonomy branch (combination of categories), multiple representative structures are analyzed with the intention of capturing the variability of the behavior in each class and defining an average fragility curve. The latter is used to evaluate the probability of overcoming a predefined limit state (LS) conditioned to the value of the intensity measure (IM). The accuracy of typological seismic assessment methods depends on aspects like the complexity and efficiency of the classification scheme, (Stefanidou and Kappos, 2019) on the quality and quantity of available data (as in the case of prioritization; MIT, 2020), and on the time at disposal. In this case, simplified modelling and analysis approaches (e.g., Capacity Spectrum Method, CSM, by Freeman, 2004) are still preferable for the low computational effort, although providing a lower accuracy in the evaluation compared to more advanced methodologies based on complete bridge finite-element (FE) models and nonlinear time-history analysis (NLTHA). On the other hand, Machine Learning (ML) algorithms, such as artificial neural networks and gradient boosting algorithms (e.g., eXtreme Gradient Boosting) are increasingly being used in the field of seismic assessment due to their ability to model complex nonlinear relationships between structural parameters, seismic intensity measures, and damage outcomes (Neduri et al., 2024; Mangalathu et al., 2017). Furthermore, another limitation related to the use of taxonomy-based approaches is the assumption that the average fragility curve is representative of the entire taxonomy branch, which neglects the possibility of further exploring the influence of structural and geometric characteristics that vary across representative structures analyzed within each taxonomy branch (Abarca et al., 2022). Considering this open problem in the field of taxonomy-based approaches, an ML-based framework for assessing the seismic fragility of existing bridges is proposed in Section 2, along with a new taxonomy formulated in accordance with the information collected in the prioritization levels of the multilevel approach described in the Italian guidelines (MIT, 2020). The eXtreme Gradient Boosting (XGBoost) algorithm (Chen and Guestrin, 2016) is trained using a supervised classification approach based on the results obtained from CSM analysis. The latter is applied to a large sample of single concrete cylindrical piers generated according to the proposed taxonomy and accounting for the seismic capacity reduction induced by the corrosion of steel reinforcement. The results of the application to a case study are presented in Section 3 where average fragility curves for each taxonomy branch are discussed, along with their variability with respect to other structural and geometrical parameters not accounted for in the classification. The proposed strategy demonstrated considerable promises for the implementation of seismic risk-informed prioritization and has the potential to be expanded to encompass other bridge typologies.