PSI - Issue 78

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

716

Fig. 4. Fragility curves for two different taxonomy branches derived considering their variability with respect to the seismic defectiveness level (color) and pier dimensional characteristics ( D pier , H pier ) within ranges used for independent variables definition (squat and slender): (a) 45 80/NS/LS; (b) 03/S/SS. μ is the AvgSa level associated with 50% probability of failure. 4. Conclusions The assessment of seismic vulnerability at the portfolio level is a complex task for engineers, especially when dealing with knowledge-based uncertainties and material degradation. A commonly used method in such cases is the indirect typological approach, where structures are grouped into taxonomies based on key structural characteristics that influence seismic behavior to develop average fragility curves. In this framework, simplified analysis methods such as the Capacity Spectrum Method (CSM), are usually preferred to more complex and computational expensive ones at the expense of the accuracy of the evaluation which, anyway, complies with prioritization scopes. However, the use of average fragility curves prevents exploring the influence of structural and geometric features that vary among the representative structures analyzed within each taxonomy branch. To overcome this limitation, an eXtreme Gradient Boosting (XGBoost) Machine Learning (ML) algorithm is proposed based on results derived from CSM considering a novel bridge taxonomy based on the structural and geometrical data available at the end of the risk prioritization step of the Italian guidelines. The ML algorithm was applied to multi-span reinforced concrete girder (RC) bridges with single cylindrical RC pier showing promising results for seismic risk-informed prioritization of bridges and suggesting the possibility of extending it to other bridge-pier systems. Indeed, the proper calibration of the XGBoost algorithm allowed the computation of average fragility curves of taxonomic branches which were close to the ones obtained with CSM, with a difference ranging between 4% and 7%. Furthermore, given the ability of the XGBoost algorithm to predict whether the failure occurs at different levels of AvgSa , fragility curves for different geometric pier configurations were investigated. As result, among piers belonging to 03/S/SS typologies, slender piers (high H pier / D pier ratios, H pier is the Pier height while D ier is its diameter) characterized by a low (L) seismic defectiveness level are suggested to be investigated before squat ones (low H pier / D pier ratios) with a medium (M) seismic defectiveness level. On the other hand, less variability in the median of the fragility curves was registered for 45 80/NS/LS typologies, suggesting that the seismic defectiveness level plays a crucial role for this bridge typology regardless of the ratio between H pier and D pier . Acknowledgements Andrea Dall ’Asta was supported by the National Recovery and Resilience Plan (NRRP), Mission 4 Component 2 Investment 1.4 - Call for Tender No. 3138 of 16 December 2021 of the Italian Ministry of University and Research, funded by the European Union — NextGenerationEU; Project code: CN00000013, Concession Decree No. 1031 of 17 February 2022 adopted by the Italian Ministry of University and Research, CUP: H93C22000450007, Project title: National Centre for HPC, Big Data and Quantum Computing t n t e r je t “ Fabrich ” .