PSI - Issue 84

Gianluca Quinci et al. / Procedia Structural Integrity 84 (2026) 199–206

200

1. Introduction Italy’s transportation infrastructure is facing a critical turning point, where ensuring the structural safety of existing bridges and viaducts has become a top national priority. A large portion of this asset stock was constructed during the rapid development decades between the 1950s and the 1980s, when seismic design provisions were either missing or still immature and the prevailing design philosophy differs substantially from present-day requirements. As these structures continue to operate far beyond their original service expectations, they are increasingly exposed to deterioration and aging processes, heavier and more frequent traffic actions, and the effects of seismic hazard. In addition, many assets still lack systematic structural monitoring, which further reinforces the urgency of implementing robust tools for risk evaluation and network-level management Hao et al. 2020 and Skokandi´c et al. 2022. Despite their reliability, standard engineering procedures—such as extensive field inspections combined with high fidelity numerical analyses—are inherently resource-intensive. When transferred from the single-structure scale to the national scale, they become difficult to sustain: the number of bridges to be examined would imply prohibitive requirements in terms of working time, expert manpower, and economic investment. For this reason, the use of these refined approaches as a widespread screening strategy is generally unrealistic, El-Maissi et al. 2021. This gap has stimulated the search for rapid preliminary methodologies capable of identifying potentially critical bridges and supporting the definition of priorities for detailed assessment and intervention. Artificial Intelligence (AI) is increasingly considered a promising enabler in this direction, with Machine Learning (ML) offering a particularly practical contribution. ML models, once trained on well-structured datasets, can learn relationships between heterogeneous variables describing a bridge and its context—such as geometry, material information, and seismic exposure indicators—and can provide a first estimate of seismic vulnerability without requiring dedicated simulations for each asset. The advantages of ML-based strategies for seismic risk applications have already been reported in the literature, where significant reductions in computational demand are achieved while retaining acceptable predictive accuracy, Quinci et al. 2022, Quinci et al. 2025, Quinci et al. 2023. From an operational standpoint, ML outputs can be used to support classification and ranking of infrastructures according to expected seismic risk. This can assist decision-makers in planning maintenance actions, organizing inspection campaigns, and selecting candidates for strengthening measures, ultimately targeting limited resources toward the most critical needs. Importantly, ML is not conceived as a substitute for established engineering practice; rather, it should be interpreted as an additional decision-support layer that increases efficiency and helps move toward more proactive and data-driven asset management, Xie et al. 2020. The motivations that guide this research can be summarized as follows: • Public safety: improving the capability to rapidly identify the most vulnerable bridges and reduce the likelihood of critical failures during earthquakes; • Efficient resource allocation: providing tools that enable better prioritization of funds for maintenance and strengthening interventions; • Technological advancement: fostering the integration of artificial intelligence into engineering workflows, enhancing both speed and reliability of assessments; • Scalability: proposing a streamlined, generalizable methodology applicable to extensive infrastructure networks at regional and national scales. Accordingly, this study investigates a Machine Learning-based method for the preliminary seismic assessment of bridge structures. The approach is applied to a case study located in the Sicilian Region, an area of strategic relevance due to its seismic activity and infrastructure density, in order to evaluate the feasibility and practical applicability of the methodology in a real-world context. 2. Methodological Framework for ML-Based Fragility Assessment In seismic risk assessment of bridge networks, fragility curves are widely used to quantify the probability that a structure will exceed a given performance limit state as the level of seismic intensity increases. When fragility functions are derived from high-fidelity numerical procedures—such as nonlinear time-history analyses or pushover-