PSI - Issue 84

Vincenzo Mario Di Mucci et al. / Procedia Structural Integrity 84 (2026) 521–528

522

1. Introduction Transportation infrastructure is fundamental for economic growth and social resilience, with bridges representing critical assets whose structural integrity is increasingly challenged by aging, environmental degradation, and evolving service demands. In many industrialized countries, a substantial portion of bridges was constructed between the 1950s and 1990s, often according to outdated design codes that did not fully account for long-term durability or seismic performance. Consequently, a lot of structures now exhibit signs of significant deterioration, making timely condition assessment and risk-informed maintenance planning essential, particularly in seismically active regions (Miluccio et al., 2021; Borzi et al., 2015; Nettis et al., 2025). Among the degradation phenomena affecting reinforced concrete (RC) bridges, corrosion of steel reinforcement is particularly dangerous. Corrosion of longitudinal bars reduces flexural capacity, while deterioration of transverse reinforcement impairs shear strength and confinement, potentially shifting failure modes from ductile to brittle (Xu et al., 2020; Biondini et al., 2014). In bridge piers, loss of confinement reduces both ultimate compressive strength and ductility, critically affecting energy dissipation under seismic loading (Bartolozzi et al., 2022; Vu et al., 2016; Pinto et al., 2024). Many seismic vulnerability studies neglect corrosion effects, potentially underestimating the actual risk posed by aging bridges (Stefanidou et al., 2019). In response to these issues, transportation agencies are progressively adopting prioritization frameworks that rely on risk-based metrics to support maintenance planning when budgets and resources are limited (Ministero delle Infrastrutture e dei Trasporti, 2020), adopting a life-cycle–oriented perspective that integrates current condition, future deterioration, and residual service life to optimize interventions over the structure’s remaining lifespan. Such prioritization schemes should solve two challenging tasks: (i) recognizing the current degradation state as observed through visual inspections and (ii) predicting its evolution over time to prioritize interventions and allocate infrastructure management budgets effectively. Recent advances in artificial intelligence (AI) and machine learning (ML) provide new opportunities to automate deterioration assessment. Computer vision (CV) and deep learning techniques—including convolutional neural networks (CNNs), attention-based architectures (Ruggieri et al., 2025), and image-segmentation methods—enable extraction of relevant information from inspection imagery and automated classification of defect severity with reduced computational effort (Di Mucci et al., 2024). Building on these developments, this study proposes a unified AI-based framework addressing both challenges in the context of corroded RC bridge piers. First, the method quantifies corrosion intensity through CV-based processing of images captured on each surface of the structural elements, providing an objective measure of the current degradation state. This parameter is then directly incorporated into probabilistic structural models to compute fragility curves and the mean annual frequency of exceedance (MAFE) representative of the structure’s present condition. Second, by integrating evolutionary corrosion laws, the framework extends the analysis to future states of deterioration. Projecting the observed corrosion level forward in time enables forecasting of time-dependent MAFE, providing a fast and low-cost basis for estimating residual service life. Overall, the approach couples CV-based image assessment with structural modelling and probabilistic analysis, delivering a scalable tool for both immediate seismic risk evaluation and long-term maintenance planning. 2. CNN with attention for corrosion severity assessment in RC bridge piers 2.1. BriCANet architecture This study introduces a deep learning framework designed to automate the evaluation of corrosion severity in RC bridge piers through image-based analysis. The approach employs a dedicated CNN to classify images of exposed steel reinforcement into three predefined corrosion levels (Low, Medium, and High) consistent with Italian technical guidelines (Ministero delle Infrastrutture e dei Trasporti, 2020). These discrete categories correspond to specific ranges of steel mass loss, facilitating direct integration into structural models that explicitly account for corrosion-induced degradation. The methodology begins with the development of a curated image dataset, combining high-resolution inspection photos with publicly available sources (e.g., dacl10k - Flotzinger et al., 2024) to enhance both dataset volume and