PSI - Issue 78

Andrea Nettis et al. / Procedia Structural Integrity 78 (2026) 1404–1411

1405

1. Introduction Transportation networks are vital for economic growth and social mobility, with their performance heavily reliant on the structural integrity of key components such as bridges. These structures are particularly vulnerable to long-term deterioration caused by environmental exposure, increasing service demands, and aging. In many developed nations, a significant share of the bridge inventory dates back to the period between the 1950s and 1990s, constructed under outdated codes that did not fully consider long-term durability or seismic resilience (Miluccio et al., 2021; Borzi et al., 2015). As a result, many of these assets now show signs of pronounced degradation, requiring timely evaluation and intervention, especially in seismically active regions. Among the deterioration mechanisms affecting reinforced concrete (RC) bridges, reinforcement corrosion is one of the most detrimental. It compromises the mechanical properties of the structural elements, reducing both flexural and shear capacity (Biondini et al., 2014; Pinto et al., 2024). Corrosion of longitudinal bars undermines flexural strength, while degradation of transverse reinforcement impacts shear resistance and can shift the failure mode from ductile to brittle. Additionally, stirrup corrosion impairs concrete confinement, reducing the ultimate compressive strength of confined concrete and leading to lower ductility — an especially critical issue for bridge piers that serve as primary energy-dissipating elements during seismic events (Vu et al., 2016; Vu et al., 2018). The widespread need for condition assessment presents a significant challenge to infrastructure managers, particularly given limited resources. Traditional seismic evaluation methods, which often rely on detailed finite element (FE) models calibrated against experimental data, are computationally intensive and thus unsuitable for network-level applications. Furthermore, many existing seismic risk studies overlook the influence of corrosion, potentially underestimating the true vulnerability of aging bridges (Stefanidou et al., 2019). Recent developments in artificial intelligence (AI) and machine learning (ML) offer promising avenues to streamline the seismic assessment of corroded RC structures. ML-based tools have demonstrated the ability to predict structural behavior, estimate residual strength, and identify likely failure modes with reduced computational demands (Xu et al., 2024). Nevertheless, the specific application of AI techniques to evaluate corroded RC bridge piers remains underexplored, highlighting a notable research gap (Di Mucci et al., 2024). In this context, the present study introduces an AI-powered computer vision (CV) framework aimed at automating the classification of corrosion severity in RC bridge piers. The objective is to support rapid seismic assessments by incorporating visual damage indicators into a simplified yet informative evaluation workflow. The proposed method accounts for both observable defects and uncertainties in material properties — commonly referred to as knowledge based uncertainties — to enhance the accuracy of risk prioritization for bridge maintenance (Nettis et al., 2024). Using advanced image analysis, corrosion severity is categorized into three discrete levels, each linked to a probabilistic range of mass loss values. These classifications inform automated structural analyses, including fragility simulations, which are conducted while incorporating variability in mechanical parameters to derive the expected annual losses (EAL). By integrating CV techniques with streamlined structural modeling, this approach aims to improve decision making in bridge management, allowing for efficient identification of critical structures requiring further investigation. 2. AI-Driven Corrosion Assessment A novel AI-based framework (Fig. 1) has been developed to support the classification of corrosion severity on reinforced concrete (RC) bridge piers by analyzing images of exposed steel reinforcement collected during visual inspections. The method categorizes corrosion into three levels — Low, Medium, and High — following the definitions provided in the Italian technical guidelines (Ministero delle Infrastrutture e dei Trasporti, 2020). Each classification level is associated with a corresponding range of steel mass loss, which can subsequently be integrated into structural modeling and analysis procedures.

2.1. Dataset Construction

The process begins with assembling a dataset of annotated images, which involves two main steps: