PSI - Issue 84

Lorenzo Principi et al. / Procedia Structural Integrity 84 (2026) 73–80

74

1. Introduction Bridges play a fundamental role in transportation systems, with a direct impact on public safety and economic growth, (2025, Huang et al.). Recent bridge collapses (Wardhana et al., 2003; Zhang et al., 2022) have emphasized their susceptibility to both human-induced and natural hazards, underlining the need for effective maintenance strategies and risk reduction measures. Numerous studies have proposed methods to address different hazard sources, such as environmental deterioration and traffic effects (2013, Li et al., , Nettis et al., 2024, Scozzese et al., 2025), seismic actions (Tubaldi et al., 2021; Minnucci et al., 2022; Ruggieri et al., 2024; Scozzese and Minnucci, 2024), flooding events (Scozzese et al., 2023), and landslides (Fell et al., 2008). However, most of these contributions address individual hazards. To overcome this limitation, the Italian Ministry of Transportation (MIT) issued the national Italian Guidelines (hereafter IG) for the classification and management of risks in existing bridges (MIT, 2020; ANSFISA, 2022; Natali et al., 2023), introducing a multi-hazard framework that accounts for multiple risks. The guidelines have garnered international acclaim for their innovative approach (Gara et al., 2025). A key challenge of this framework lies in the shift from the preliminary phase, typically based on the collection of inventory data (Level 0), to more detailed assessment phases (Levels 2–5). This step requires extensive inspections, including in-depth surveys of structural elements and surrounding contexts (Level 1). Such inspections remain crucial but can become a limiting factor when dealing with large inventories (Yaw Adu-Gyamfi et al., 2016). Within the IG methodology, inspections are essential for seismic risk evaluation, as they provide fundamental vulnerability information. In parallel, recent research has effectively applied machine learning (ML) techniques across different bridge engineering contexts, including the prioritization and optimization of territorial-level investments (Bridgelall, 2025; Yepes-Bellver, 2025), structural health monitoring (Quarchioni et al., 2025; Ierimonti et al., 2025), and rapid bridge assessment applications (Bergmeister et al., 2024; Casassa et al., 2025; Pentassuglia et al., 2025; Ruggieri et al., 2023). In this context, Principi et al. (2025) developed an Artificial Neural Network (ANN)-based framework to estimate both the Class of Damage (CD), describing the degradation state of bridges according to MIT (2020), and the Class of Structural Risk (CSR), related to traffic loads. The framework was successfully implemented on the Italian highway network using data from pilot IG-based assessments (Natali et al., 2023). Building upon this approach, the present study extends the application of ANNs to the prediction of the Class of Earthquake Risk (CER) (MIT, 2020). The proposed model integrates seismic hazard information with bridge vulnerability and exposure data, allowing assessments at a regional scale. The optimized ANN constitutes the core output of the framework and is applied to a case study involving 95 bridges in the Marche Region to predict the CER. The results are also visualized through a GIS-based map, offering an effective tool for territorial-scale seismic risk evaluation. 2. Framework The present work builds upon the innovative framework introduced by Principi et al. (2025) (Figure 1) to predict the CER. The framework is structured into three primary phases: (i) Data Collection and Preprocessing, (ii) Data Processing, and (iii) Optimal Model and Performance Analysis, each articulated into several sequential steps (Figure 1).

Figure 1. General framework.