PSI - Issue 62

Lorenzo Principi et al. / Procedia Structural Integrity 62 (2024) 89–96 Principi L. / Structural Integrity Procedia 00 (2019) 000 – 000

90 2

1. Introduction Bridges are important components of the road networks and they significantly contribute to ensuring the well-being of countries and communities. Hence, it is crucial to ensure their functionality throughout their service life. In recent decades, there have been numerous cases of bridge collapses worldwide leading to social and economic losses. Italy has also experienced several bridge failures during the last few years, due to different combinations of causes. In fact, bridges are prone to various sources of risks, necessitating a comprehensive procedure to assess their safety (Pinto and Franchin, 2010). Two problems arise from this scenario. Firstly, a significant number of existing bridges were realized during the 1970s. (Pinto and Franchin, 2010), consequently, the aging process has significantly compromised their structural integrity. Secondly, a substantial increase in traffic volume and weight, coupled with a lack of adequate maintenance over the years, has exacerbated the degradation phenomena (Bień and Salamak, 2020) . Moreover, bridges are susceptible to multiple hazards, not only associated with the structural deficiencies but also due to the natural events related to environment in which they are situated. Thus, there is a need for a procedure that considers all different scenarios that could lead to a critical situation for a given bridge. In recent decades, several methodologies (Pregnolato, 2019; Pellegrino et al. 2011; Chase et al., 2016; Mangalathu et al., 2017; Whelan et al., 2019; Valenzuela et al., 2009; Padgett and DesRoches, 2007; Franchin et al., 2015; Chen et al., 2021; Minnucci et al., 2021; Tubaldi et al., 2021) have been proposed to assess the health condition of bridges. A limitation of those procedures is that they are mostly structural and seismic risk-focused (Santarsiero et al., 2021) and as result, they do not combine other sources of risk. To address this problem, in 2020 the Ministry of Infrastructure and Transportation (MIT) has enacted the New Guideline (MIT, 2020), promoting a more holistic multilevel and multi-risk approach that consider multiple aspects: structure-foundational, seismic, landslide and flood risk. Within this procedure, the method is conceived on six levels of increasing degree of details. Levels 0 to Level 2 must be applied to all bridges of the stock, while Level 3 to Level 5 are devoted to bridges which present a higher level of risk, as derived from the previous levels of analysis (Levels 0-2). Level 0 consists in a Census of all bridges, where all existing data about these structures is gathered and collected in a protocolized way. In Level 1 visual inspections are performed, this activity is fundamental to define the Level of Defectiveness (LoD) of the bridges and the condition of the surrounding areas. Building on the earlier levels of analysis, Level 2 involves a risk-based classification of bridges, which is categorized according to four primary hazard risk sources: Structural and Foundational, Seismic, Flooding, and Landslides. Each risk is evaluated as a combination of three aspects: exposure, vulnerability, and hazard, each of which contains multiple primary and secondary parameters. In this framework, the LoD of the Guidelines is a key element in the risk classification, defining the primary Structural and Foundational vulnerability parameter. Given the fact that Levels 0-2 are mandatory for all bridges, inspection activities must be yearly scheduled over all the national bridges inventory. Two main issues arise: inspections are time and cost consuming activities; and the number of bridges that need to be classified for the first time and yearly inspected is elevated. Moreover, it is expected that several structures present critical situations that need to be quickly notified to the relative management body. To address this problem, modern Machine Learning (ML) algorithms could offer effective support in planning inspection and conducting Level 2 risk assessment. Recently, ML techniques have been effectively utilized in various civil engineering scenarios and, Artificial Neural Networks (ANNs) demostrated high versatility and the ability to capture complex non-linear pattern, (Karim et al. 2020, Xie et al. 2020, Cattan and Mohammadi 1997, Elhag and Wang 2007; Fan et al. 2021) across various domains within civil engineering. Also in the context of forecasting the LoD of bridges, many authors have used the ANNs which have demonstrated the ability to predict the current state of preservation of structures (Assaad and El-Adaway 2020a, 2020b; Xia et al. 2021; Alogdianakis et al. 2022). Despite the high performance achieved by these models, they were designed to predict the LoD, limiting their application to vulnerability analysis and focusing on hazards related to material degradation or structural damage. As a result, they do not provide information regarding the risk level, which should consider information about the hazard (e.g., traffic load intensity) and the potential extent of consequences in the event of failure. As a result, an additional step is required to gain insight into the overall Structural and Foundational Attention Class. Thus, the objective of this paper is to develop an ANN capable of rapidly estimate the LoD and the Structural and Foundational Attention Class, using limited information selected from the Census data. The model will be applied to a dataset of existing bridges to assess its effectiveness. Only minimal information will be employed for making predictions.