PSI - Issue 84

Marianna Crognale et al. / Procedia Structural Integrity 84 (2026) 898–905

899

1. Introduction Modelling of existing bridges and viaducts is a crucial task in modern engineering. It enables accurate assessment of structural behaviour, safety levels, and remaining service life. Advanced numerical models (Di Re et al., 2024) also support maintenance planning, strengthening strategies, and risk mitigation under increasing traffic and environmental demands. A remarkable case is that of aging reinforced-concrete (RC) bridges, increasingly exposed to the combined effects of material deterioration, environmental actions, and seismic demand (Salvatore et al., 2024; Ciminelli et al., 2025). In this context, performance-based assessment frameworks are often adopted, combining nonlinear static (pushover) analyses with vulnerability metrics such as fragility curves to support efficient decision-making at the network level (Gattulli et al., 2017). For bridge structures, pushover analysis remains an attractive tool when global system features are adequately represented and results are interpreted through comparison with dynamic analyses. Several studies have demonstrated that capacity-based nonlinear static procedures can be effectively employed to derive fragility curves for RC bridges, providing vulnerability estimates consistent with nonlinear time-history analyses and supporting their use in seismic assessment and screening applications (Banerjee and Shinozuka, 2007; Mosleh et al., 2015). Fiber-based section modeling provides an effective way to capture nonlinear behavior at piers and girders while maintaining a reasonable computational cost. At the member level, fiber formulations allow the explicit representation of confined and unconfined concrete and reinforcing steel, enabling a realistic description of stiffness and strength evolution under increasing deformation demand. Experimental and numerical studies on RC bridge columns have shown that fiber-based models can accurately reproduce force–deformation hysteretic response, plastic hinge behavior, and strain distributions, provided that appropriate element formulations and modeling choices are adopted (Feng et al., 2014; Bernardini et al., 2024). In the presence of material deterioration, phenomenological fiber-based approaches further allow the incorporation of corrosion-induced degradation of concrete and steel through constitutive laws defined at the fiber level, while preserving predictive accuracy and computational efficiency (De Domenico et al., 2023). Previous investigations have also highlighted that seismic response and vulnerability estimates of RC bridges are sensitive to modeling assumptions, particularly when complex structural configurations or non-uniform deterioration patterns are considered. Advanced three-dimensional nonlinear analyses have shown that corrosion scenarios, pier irregularity, and modeling choices can significantly influence both local failure mechanisms and global seismic capacity, as well as the resulting fragility estimates (Ghazal and Mwafy, 2023; Dizaj et al., 2023). These findings underline the importance of clearly defined and transparent modeling strategies when nonlinear static analyses are used as a basis for seismic vulnerability assessment. Within this context, numerical modeling and analysis are acquiring significant enhancements from AI-based approaches (Paduano et al., 2023). Scope and contribution. This study applies an AI-enhanced, degradation-aware nonlinear static analysis framework to a three-span reinforced-concrete road viaduct in Italy. The proposed approach combines a conventional global pushover analysis with fiber-based section modeling at the pier level through a machine-learning surrogate trained on offline moment–curvature simulations. The objectives of the study are to: (1) evaluate the influence of material degradation on global pushover response; (2) compare conventional fiber-based distributed plasticity modeling with the proposed AI-enhanced deterioration-aware interpretation; and (3) assess the computational efficiency and practical applicability of the framework for bridge seismic assessment. Paper outline. Section 1 introduces the contribution. Section 2 describes the case-study viaduct, the numerical modeling strategy, and the offline fiber-section dataset generation. Section 3 presents the machine-learning surrogate and its use for the post-processing and interpretation of pushover analysis results. Section 4 compares the seismic responses obtained from the conventional and AI-enhanced approaches. Section 5 summarizes the main findings and concludes the paper.