Numerical Assessment of the Seismic Vulnerability of Bridges within the Italian Road Network

Abstract

The safety of existing bridges represents a serious problem in Italy since these structures are fundamental for the national transportation system and, at the same time, can be subject to significant deterioration phenomena linked to the fact that the construction period typically dates back to the 1960s. This study involves the seismic analysis of five case study bridges belonging to the Italian Road Network. Using nonlinear time–history analysis with sets of code-spectrum compatible ground motions, analytical fragility curves have been constructed for each of the five bridges. The results obtained interpreting the analytical fragility curves agree with the fact that the seismic behavior of existing bridges can be problematic and that higher seismicity can be associated with more detrimental behavior. In particular, the results reveal that in regions with higher seismicity, the main problems in bridges are related to bearings and connecting elements located in the piers. Five case studies have also been analyzed to determine the Structural and Foundational Class of Attention and Seismic Class of Attention, following the approach proposed by the 2020 Italian Guidelines. In this way, it is possible to compare two different assessment approaches with different safety levels. The results obtained with the two approaches are in good agreement considering bridges in high seismicity regions, while the procedure of the Guidelines could lead to not reflecting the seismic behavior of bridges when the seismicity of the area is lower.

1. Introduction

The safety of existing bridges represents a serious problem since these structures are fundamental for the transportation system of a nation and, at the same time, they can be subject to significant deterioration phenomena linked to the fact that the construction period typically dates back to the 1960s. Considering that at the time durability was not one of the goals in projects and that the service life of the structures has been, in most cases, already reached or exceeded, proper maintenance and retrofit interventions are needed.

In the last few decades, a significant increase in earthquake-related losses has been observed. The first studies on the effects of earthquakes on structures and the estimation of losses date back to the second half of the 1900s [1]. To effectively deal with a possible seismic event, it is important to identify the areas and structures that will most likely be affected by the earthquake. In this way, it is possible to determine the potential losses associated with the seismic event [2]. The definition of an earthquake loss model for a given region is also fundamental in view of the mitigation of seismic risk and, consequently, of losses. A loss model calculates the seismic hazard at sites of interest and predicts the damage distribution of the structures by combining the hazard with the vulnerability of the exposed structures. Damage ratios are then used to calculate the loss by relating repair and replacement costs to demolition and replacement costs. The seismic vulnerability, which is a fundamental component of an earthquake loss model, indicates the susceptibility to damage of a structure given a certain intensity of ground shaking [3]. Several methods to assess seismic vulnerability have been proposed in the literature over the years. The approaches can be divided into empirical or analytical. Hybrid approaches represent a combination of empirical and analytical methodologies. The work of Calvi et al. [3], in which this classification was proposed, describes the evolution of vulnerability assessment procedures over the years, highlighting the main developments for each methodology.

The behavior of bridge structures is complex and requires attention to a number of important details. To gain a thorough understanding of these systems, researchers have explored various fields in bridge engineering. Tavares et al. [4] studied the bridge network of the Province of Quebec (Canada). Many bridges in the region were not designed for seismic loads and therefore may have a high seismic vulnerability. The authors developed fragility curves related to the bridges in the province to assess the seismic vulnerability of the structures. Considering the variability in terms of structural systems, it was found that concrete girder bridges are more vulnerable than steel girder bridges. The fragility curves developed in the study are useful to determine potential losses from seismic events and prioritize retrofit interventions but can be improved by collecting more information on the response of individual bridge components and integrating the definitions of the limit states with experimental tests. Fragility curves are common representations of the vulnerability of structures. Considering the extension of road networks, which count thousands of bridges, it is not possible to individually study every structure.

Given that every bridge is different, a 2013 study [5] proposed a method to calculate fragility curves based on geometry, materials, and data recorded by a monitoring system. The proposed method allows changes in structural parameters to be tracked during the life of the structure, updating the fragility curves to obtain more accurate vulnerability assessments. It has to be pointed out that the structures involved in the research were not subject to a strong earthquake at the time of the study, so the updating of the fragility curves in post-earthquake condition had not been assessed in the study. Fragility curves have also been studied in the work of Stefanidou and Kappos [6], which proposed a methodology to develop bridge-specific fragility curves in order to improve the reliability of loss assessment in road networks and the prioritization of retrofit interventions. The method, which defines critical limit state thresholds for the components of the bridge, can be applied to a large number of bridges within a network.

The approach presented by Mangalathu and Jeon [7] employs machine learning to generate bridge-specific fragility curves. The method has been tested through a case study consisting of a multi-span concrete bridge class, which can be found in California. The method is reliable and can be employed in risk assessment platforms, but additional studies are required to check if the methodology can be applied to other bridge configurations.

Wang, Wu, and Liu [8] studied the fragility of highway bridges. The research presents a fragility calculation method that takes into account multidimensional performance limit state parameters. Column ductility and transverse deformation in the abutments were considered as performance limit state parameters. The method can be employed to assess the vulnerability of bridges sensitive to multiple response parameters in a more reliable way, allowing also for implementation in loss estimation for the road network.

In 2014, Kameshwar and Padgett [9] presented a parameterized fragility-based Multi-Hazard Risk Assessment procedure for the assessment of a portfolio of highway bridges exposed to earthquakes and hurricanes. The annual risk, measured as the annual probability of damage, is obtained by combining fragility functions and regional hazard data. The framework has been applied to multi-span simply supported concrete girder bridges, allowing the gain of insights on the influence of bridge geometry and hazard exposure on the risk. The framework can be extended to include other hazards, bridge types and regions, and concurrent occurrence along with independent hazards such as scour and collisions.

Kim [10] studied the use of Fiber Reinforced Polymer (FRP) composites in highway infrastructures. The use of these composites started around 1996 and is common in strengthening interventions. The long-term durability of FRP composites still requires additional research. In 2018, Pang and Wu [11] studied the effect of aftershocks on the seismic response of multi-span concrete bridges. Currently, aftershock effects on structures are not taken into account in design codes. Considering reinforced concrete (R.C.) continuous bridges, aftershocks can increase the system fragility, the displacement of the bearings, and the curvature of piers. The research of Kumar and Gardoni [12] shows that the effects of multiple earthquakes during the life of a structure can produce a reduction in its seismic reliability. The study considers highway R.C. bridges. A degradation model has been developed and the results show that the seismic vulnerability of R.C. bridges can increase as a consequence of the seismic degradation in the columns.

A recent work [13] studies multi-span R.C. bridges typical of Iran. Many existing bridges, designed at a time when there were no specific requirements in terms of cap beam-column joints, have poor details. The research analyzed the fragility functions of these bridges, highlighting that the seismic fragility of the structure is highly influenced by the fragility of the joints. The research is limited to regular straight bridges with multi-column bents, not including irregular structures in the fragility analysis.

The work of Thakkar et al. [14] provides a literature review on the seismic behavior of bridges and on the analytical methods that can be employed to assess bridge performance, including the methodology to develop fragility curves. Research by Cao et al. [15] presents the comparison between four common approaches used in the literature in the seismic fragility framework, assessing their accuracy and applicability. Given that the proper selection of the seismic input is fundamental to obtaining a realistic result from the assessment and that the records employed in the fragility analysis must be representative of the site conditions, recent research proposed the use of stochastic earthquake models to generate the seismic records [16] and studied the fragility of structures in near-fault regions [17].

The study of Dilena and Morassi [18] deals with the dynamic testing of a damaged bridge. The bridge is subjected to increasing damage levels. The natural frequencies of the structure vary as the damage progresses. Lower modes can be employed to gain information about the damage location.

The retrofit of R.C. bridges can be performed in different ways. For instance, examples can be found in the literature on retrofit interventions using Ultra High Performance Fiber R.C. [19] or Stainless Steel Wire Mesh [20]. Clearly, these are not the only possible retrofit interventions that can be employed to improve the structural response of a bridge. Upadhyay, Pantelides, and Ibarra [21] investigated the effect of Buckling Restrained Braces and Self Centering Energy Dissipation devices on the seismic response of a bridge, while the study of Chen and Li [22] treated seismic retrofit methods employing lead rubber bearings and rocking foundations. The most common criteria considered in the selection of a retrofit strategy or, in a more general sense, a bridge management strategy are associated cost, resilience enhancement, values of design parameters, risk, and network resilience [23,24,25].

The present research work aims at assessing the structural performance of five bridges belonging to the Italian Road Network, by using different techniques. The five case studies have been built around the same period, between the early 60s and mid-70s. At that time, there was a less developed understanding of the seismic behavior of structures, and crucial facets of the design process, such as bearing design and anti-seismic details, were addressed in a simplified way. For each of the case studies, fragility curves have been computed, by defining the target Limit States and the analysis method. The obtained results have been used then to evaluate the efficiency of the procedure set out by the new “Guidelines for the classification and management of risk, safety evaluation, and monitoring of existing bridges” [26], which were introduced in 2020 and will be herein referred to as “the Guidelines”. In particular, the Level 2 assessment has been applied to the case studies, defining Structural and Foundational and Seismic Classes of Attention, allowing for evaluation of the proposed approach. The obtained results, in terms of fragility curves and Classes of Attention, are finally discussed.

2. Case Studies and Modeling

In this section, the case studies considered in the presented research and the relative modeling techniques are briefly described. Nonlinear modeling is particularly suitable when dealing with existing structures, including bridges, as structures that are not designed to withstand earthquakes can be subjected to displacement demands that exceed the limit that characterizes a linear response from the point of view of geometry and materials behavior.

2.1. Characteristics of the Case Studies

The five case study bridges are located in Italy. The five case study bridges have been selected because they belong to common structural types within the existing bridges of the Italian Road Network. At the same time, the fact that some characteristics of the case studies are different allows for the evaluation of the procedure proposed by the Guidelines in different scenarios. Some common elements are present in the selected case study bridges. All considered bridges were built between 1960 and 1973, like most of the existing bridges in the country [27]. All the considered bridges are beam bridges. The total length ranges from a minimum of 50.00 m (Bridge 5) to a maximum of 421.10 m (Bridge 1). In two cases (Bridge 1 and Bridge 3) the structural scheme involves the presence of Gerber saddles. In terms of piers, Bridge 1 has frame piers with two columns and transverse connecting elements. Bridge 2 has two elements with rectangular sections connected on top by a slab. Each pier of Bridge 3 is characterized by eight columns monolithically connected with the superstructure. Bridge 4 has a scheme in which each pier consists of three columns and a pier cap, and Bridge 5 involves piers having two columns and a pier cap.

Table 1. Main characteristics of the case studies.

Case Study	Year	Structural Scheme	Bearings	Number of Spans	Total Length [m]	Soil Category (NTC2018)
Bridge 1	1960	Continuous beams with Gerber saddles	Neoprene	13	421.1	A
Bridge 2	1969	Simple support beams	Neoprene and steel plates	4	124	A
Bridge 3	1973	Simple support beams with Gerber saddles	PTFE-Steel	5	157	B
Bridge 4	1969	Simple support beams	Neoprene and steel plates	3	87.8	A
Bridge 5	-	Simple support beams	Neoprene and steel plates	4	50	A

summarizes the main characteristics of the case studies. The available documentation provides information about the structural elements, both in terms of sections and reinforcement, allowing accurate modeling of the structures. For the material properties, the documents report information about the original materials. Results of experimental tests performed on the structures have been taken as a reference for concrete properties when available. This is particularly important for the case of Bridge 3, as the available documentation highlights damage to the structural elements of the bridge. In the cases test results were not available, assumptions about the deterioration of materials have been made. For the steel properties, the study of Verderame et al. [28] has been taken as a reference. The available documentation also provides information about the soil category of the sites.

2.2. Modeling

All structures have been modeled using Seismostruct [29,30]. Columns and connecting elements of the piers have been modeled using inelastic force-based frame elements. In particular, the Seismostruct elements employed in this study are the “infrmFB”. This type of element is able to capture the inelastic behavior along the entire length of a structural element and is the most accurate among the inelastic frame elements available in Seismostruct. The sectional stress–strain curve is obtained by integrating the uni-axial response of the individual fibers in which the section has been discretized [30]. The number of fibers and integration sections in the elements, for each model, has been defined after performing a convergence analysis. For each section of inelastic elements, the reinforcement effectively present in the section has been defined to obtain structural models as accurately as possible. The deck has been modeled using elastic frames. This is a common choice and can be justified considering that, in general, bridge decks are less vulnerable than piers and tend to have an elastic response. The same simplification has been made for the modeling of abutments.

The Soil–Structure Interaction was not specifically modeled in this study, and consequently, all the results are based on the assumption of fixed base nodes. Uni-axial constitutive models employed for the fiber section of inelastic elements are based on the Mander et al. [31] concrete model and on the Menegotto–Pinto [32] steel model. In particular, the concrete and steel models employed in the study are defined in Seismostruct as “con_ma” and “stl_mp”, respectively. The confinement effects in the concrete are taken into account assuming that the confining pressure is constant through the entire stress–strain range. The steel model, which includes isotropic hardening rules, is suitable for modeling R.C. structures in presence of complex loading histories where significant load reversals might occur [30].

Bearing devices have been modeled employing link elements. These elements are 3D links with uncoupled axial, shear, and moment actions. The link elements connect two initially coincident structural nodes and require the definition of an independent force–displacement (or moment–rotation) response curve for each of its local six degrees of freedom [30]. For neoprene pad bearings and reinforced neoprene pad bearings, it has been assumed that the rotational stiffness was negligible. The calculation of vertical and lateral stiffness was performed following the indications of Australian Standard AS 5100.4:2017 [33]. The force–deformation law employed for lateral stiffness was linear elastic. This assumption can be justified considering that the energy dissipated by these devices is negligible and that they behave linearly until failure. A different approach has been followed for the PTFE-steel devices. The vertical stiffness has been assumed as extremely high, the rotational stiffness was neglected and the lateral stiffness was modeled using a bilinear force–deformation law. A friction coefficient μ = 2% and a thickness t of the PTFE layer of 1.5 mm were assumed. Vertical reactions on the bearings R_b were estimated, allowing the calculation of the static friction force F_fri as F_fri = R_b·μ. The sliding force was taken as equal to F_fri. Using the static friction force as a yielding force reflects the fact that to start the sliding movement it is necessary to exceed the static friction force. Figure 1, Figure 2, Figure 3, Figure 4 and Figure 5 show the longitudinal view of each structural model considered in the present study.

3. Fragility Curves

The seismic vulnerability of the considered structures has been numerically assessed by computing fragility curves, through Nonlinear Time–History Analyses (NLTHA), as a function of suites of natural earthquake ground motions, related to specific Intensity Measures (IM).

3.1. Introduction to Multiple Stripes Analysis

Different procedures are available to perform the fragility analysis. Incremental Dynamic Analysis (IDA) is a common approach. In the IDA, a set of ground motions is selected and scaled until the structure collapses [34]. Performing the fragility analysis employing the IDA can present challenges when ground motions must be scaled to high IM values to induce collapse. This process is computationally intensive, requiring numerous structural analyses at increasing IM levels. Furthermore, the validity of scaling typical moderate-IM ground motions to represent real instances of extreme IM levels is questionable [35].

The fragility curves associated with each case study have been derived using the so-called Multiple Stripes Analysis (MSA) [36]. This type of approach requires structural analysis to be performed at a discrete set of IM levels, using different ground motions at each level. Target properties of the ground motions vary at different IM levels and the selection of different accelerograms allows the variation of the seismic hazard parameters to be taken into consideration [37,38]. The MSA approach is more efficient than the IDA approach since it targets a limited number of IM levels rather than requiring analysis at IM levels associated with collapse for some ground motions but not as critical for constraining the fragility function. Literature results indicate that fragility functions can be estimated in an efficient way by focusing on IM levels with lower probabilities of collapse improving the quality of the fitting on the most important region for the risk assessment [39]. The fundamental aspects of the methodology used to derive the fragility curves are reported below. A total of 10 code-based elastic spectra have been defined according to NTC2018 [40]. Each elastic spectrum takes into account the topographical and stratigraphical characteristics of the site. The spectra are characterized by an increasing value of return period T_r, from 30 to 2475 years, as reported in

Table 2. Value of T_r for each IM level.

IM Level	1	2	3	4	5	6	7	8	9	10
Tr	30	50	101	201	475	712	975	1462	1950	2475

. The 2475-year T_r for IM level 10 has been selected considering the prescriptions of the NTC2018, which indicates this value as the upper bound for the definition of seismic action. The definition of these spectra is fundamental to ensure that the selected ground motions represent a broad enough range of IM levels.

The number of ground motions that have to be selected to provide reliable results is still an open issue, seven ground motions appear to be the minimum requirement [41]. For each IM level, seven pairs of natural records have been selected using the ground motion selection software REXEL [42,43].

The selection of the ground motions required that seismic hazard disaggregation [44,45] be undertaken first to identify, for each site, the magnitude–epicentral distance pair with the highest contribution in the definition of local seismic hazard. In the matching of the ground motions to the spectra, NTC2018 [40] has been taken as a reference for the issue of spectral compatibility. In particular, given the fundamental period T, the highest value between 2T and 2 s has been selected as the maximum value of the range of periods in which spectral compatibility must be ensured.

The definition of the criteria used to compute capacity and demand for each structural element is crucial. A brief discussion of the criteria related to shear capacity, chord rotation capacity, and bearing displacement capacity is provided to fully describe the framework developed in this study. In terms of shear actions, considering the indications of the 2019 Commentary [46] of NTC2018, the capacity of an existing structure has to be computed taking into account the ductility demand on the elements. The main contributors to shear capacity in the formulation proposed in the NTC2018 Commentary are normal actions N and the interaction between flexural rotation of the element and plastic component of ductility demand.

Considering the chord rotation capacity, in the 2019 Commentary of NTC2018 three different formulations are proposed in relation to Near Collapse Limit State (CLS), Life-Safety Limit State (LLS), and Damage Limit State (DLS). The first two formulations refer to an ultimate capacity, while the last formulation gives a capacity correlated with yielding in tension.

For bearing displacement, the chosen approach consisted of monitoring the horizontal deformations of each bearing through the analyses, defining four Limit States associated with different conditions of the devices. The reaching or exceeding of the four Limit States, the first associated with normal service of the device and the last with the failure of the bearing, has been defined in terms of the ratio between transverse displacement and device height. There is a slight difference between the criteria introduced for elastomeric bearings and PTFE-steel bearings. In the first case, the height, which has been taken as a reference, is the one associated with the elastomer (it is the only part of the device which can be subjected to shear deformations, eventual steel plates would not be deformed in that way). In the second case, the whole height of the device has been considered in the calculations. The definition of Limit States correlated to the seismic response of the bearings is crucial to properly describe the structural behavior of the case studies.

Two Global Limit States, Damage and Collapse, have also been defined to describe the global response of the bridge. The former is associated with an operating condition of the structure while the latter defines a condition where critical failures in piers or bearings are present.

Nonlinear Time–History Analyses have been performed to obtain the data required to construct the fragility curves. The modeling of the seismic energy absorbed during a ground motion is a crucial point of an NLTHA. A viscous damping model proportional to tangent stiffness has been adopted, following the recommendations reported by Priestley and Grant [47]. In particular, a 2% tangent stiffness-proportional damping model has been chosen. This type of equivalent damping has been employed because a viscous damping proportional to initial stiffness would not have physical meaning and would lead to overestimating the damping, while the proportionality to tangent stiffness allows one to consider the progressive anelasticity present in the structure [48].

For each of the considered Limit States, the fraction of analyses that, for a given IM level, reached or exceeded the associated threshold was computed. The fitting procedure, which allowed finally obtaining the fragility curves, was performed by applying the maximum likelihood method, which is the most appropriate for this type of data [39].

3.2. Limit States

The considered Limit States are reported in

Table 3. Shear capacity Limit States.

Limit States	Criteria
Shear failure (piers)	Reaching of shear capacity in at least one pier
Shear failure (transverse elements of frame piers)	Reaching of shear capacity in at least one transverse element of frame piers
Shear failure (Connections between pier elements)	Reaching of shear capacity in at least one connection between pier elements

,

Table 4. Chord rotation Limit States.

Limit States	Criteria
Chord rotation DLS/LLS/CLS (piers)	Reaching of chord rotation capacity, for the considered LS, in at least one pier
Chord rotation DLS/LLS/CLS (transverse elements of frame piers)	Reaching of chord rotation capacity, for the considered LS, in at least one transverse element of frame piers
Chord rotation DLS/LLS/CLS (Connections between pier elements)	Reaching of chord rotation capacity, for the considered LS, in at least one connection between pier elements

,

Table 5. Elastomeric bearings Limit States.

Limit States	Criteria
γ = 0.5	Reaching of horizontal displacement corresponding to γ = 0.5 in at least one bearing device
γ = 1	Reaching of horizontal displacement corresponding to γ = 1 in at least one bearing device
γ = 1.5	Reaching of horizontal displacement corresponding to γ = 1.5 in at least one bearing device
γ = 2	Reaching of horizontal displacement corresponding to γ = 2 in at least one bearing device

,

Table 6. PTFE-steel bearings Limit States.

Limit States	Criteria
50% bearing height	Reaching of horizontal displacement corresponding to 50% of device height in at least one bearing
100% bearing height	Reaching of horizontal displacement corresponding to 100% of device height in at least one bearing
150% bearing height	Reaching of horizontal displacement corresponding to 150% of device height in at least one bearing
200% bearing height	Reaching of horizontal displacement corresponding to 200% of device height in at least one bearing

and

Table 7. Global Limit States.

Limit States	Criteria
Damage	γ = 0.5; Reaching of chord rotation capacity (DLS) in at least a pier; Failures in secondary elements.
Collapse	γ = 2; Reaching of shear capacity in at least a pier; Reaching of chord rotation capacity (LLS) in at least a pier; Reaching or exceeding γ = 1 and failures in secondary elements.

, with some comments to briefly discuss the choices. This is necessary because there is an intrinsic difficulty in the association between physical damage and response thresholds [49]. Shear and chord rotation criteria refer only to inelastic elements. Shear failure is a brittle type of failure that takes place suddenly when the shear capacity is exceeded by the demand. The failure of a pier is a critical event in terms of bridge safety, while failures in transverse beams or connecting elements of the piers refer to localized problems. Nevertheless, their effects must be carefully considered as they can have an impact on the global response of the structure. For chord rotation, three response thresholds have been defined. The CLS refers to the computation of an ultimate capacity θ_u, the LLS involves the calculation of a fraction (3/4) of θ_u, while the DLS requires the definition of a yielding capacity θ_y. The response of bearing elements has been checked through the definition of specific criteria in the analyses. In general, bearings fail due to an excessive displacement demand. The Limit States specified for bearings are slightly different for elastomeric and PTFE-steel bearings. Different approaches have been proposed to assess the behavior of the entire structure [50,51,52]. In this work, two different Limit States, Damage and Collapse, have been considered to describe the global response of the bridge. While the Damage Limit State refers to a condition of service for the structure, the Collapse Limit State is an indicator of critical conditions in terms of safety and is reached when there are failures in piers or bearings. It has been assumed that to reach the Collapse Limit State, it was necessary to exceed the Damage Limit State. It has to be pointed out that the definition of the Collapse Limit State does not include the real collapse mechanism of the structure, since it is not possible to capture it with the regular Finite Element models used for the structural analysis, but it is instead a conventional definition employed in the assessment of the case studies. The criteria listed in Table 7 are alternative possibilities and reaching just one of them is enough to consider the corresponding Limit State as being exceeded.

3.3. Fragility Curves of the Five Case Studies

This paragraph reports the relevant fragility curves obtained for each case study. Each graph provides the probability of exceedance (which is a unitless parameter [#]) as a function of the indication of the a_g·S value corresponding to the reaching of Operational, Damage, Life-Safety, and Near Collapse limit states (OLS, DLS, LLS, CLS, respectively), according to NTC2018.

Figure 6 reports the fragility curves related to bearings, structural elements, and global response of Bridge 1. The response of the bearings is not a major issue for this case study, with no cases of bearing failure observed in the analyses. The pier behavior is acceptable both in terms of shear response and chord rotation. Local problems, mainly in terms of shear failures, are present in the transverse connecting elements of the piers. The global behavior of Bridge 1 is acceptable, but the presence of local problems is relevant in the assessment of the structure.

Figure 7 reports the fragility curves related to bearings, structural elements, and global response of Bridge 2. The response of the bearings plays a major role in this case study, with many bearing failures observed in the analyses. This result can be explained by considering the high seismicity of the site. Analyzing the documentation, it has also been noted that the design, which dates back to 1969, did not take into account seismic actions.

These considerations contribute to explaining the response of the bearings in this case study. The response of the piers is acceptable both in terms of shear response and chord rotation. Critical behavior in the connecting elements between the two parts of each pier has been noted. The predominant failures are shear failures, which, as stated for Bridge 1, must not be underestimated because they can have an impact on the global response of the structure. The response of Bridge 2 is characterized by local problems, the curves related to the global response of the bridge take into account the fact that these local failures influence the response of the whole structure.

Figure 8 shows the fragility curves related to bearings, structural elements, and global response of Bridge 3. The bearings present in Bridge 3 are made of PTFE and steel. This type of device is characterized by high displacement capacity. The seismicity of the region (the ground acceleration on rock a_g corresponding to T_r = 2475 years is equal to 0.191 g) leads to significant displacements of the bearings. In terms of structural elements, the shear response generates some problems, while the response in terms of chord rotation does not highlight particular problems. This type of behavior can be explained considering the large spacing between the stirrups and that concrete properties have been obtained using experimental tests as reference (which showed a state of deterioration with respect to design properties).

Figure 9 contains the fragility curve related to the reaching of γ = 0.5 in the bearings of Bridge 4. This was the only relevant curve for this case study since the other curves were not characterized by enough discrete points to perform the fitting procedure. No significant issues related to the seismic response of structural elements have been detected. The only relevant curve, γ = 0.5, corresponds to a normal condition for the bearings. These results can be explained considering the low seismicity of the zone, with a_g corresponding to T_r = 2475 years equal to 0.081 g.

Figure 10 contains the fragility curves related to the bearings of Bridge 5. These were the only relevant curves for this case study. No significant issues related to the seismic response of structural elements have been detected. The response of the bearings observed in the analyses highlighted many bearing failures. This result can be explained considering the high seismicity of the site and that the seismic actions were probably not considered in the design phase.

4. Bridge Assessment Guidelines

This section discusses the methodology proposed by the Guidelines [26] and reports on the application of Level 2 to the case studies. The application of this method allows readers to understand the structure of the multilevel approach and how it can be applied to existing bridges.

4.1. Proposed Methodology

The Guidelines propose a multilevel multi-risk approach to the assessment of existing bridges, with the aim of identifying the actions to be taken. The structure of the multilevel approach is represented in a schematic way in Figure 11.

The approach is organized into six progressively more detailed levels:

• Level 0 involves a stocktake of bridges, gathering information about, for instance, location, structural system, period of construction, and any past retrofit interventions;
• Level 1 requires visual inspection to detect various forms of damage. At the end of the inspection, a report is produced using a standardized format provided by the Guidelines [26];
• In Level 2, Classes of Attention are defined by combining in a simplified way data related to hazard, vulnerability, and exposure. There are five Classes of Attention (Low, Medium-Low, Medium, Medium-High, High);
• Level 3 consists of a preliminary assessment of structural safety and is performed by comparing the design actions provided by the codes at the time of construction with the design actions provided by NTC2018 [40];
• Level 4 is a detailed safety assessment of the bridge considering current code requirements;
• Level 5, not directly covered by the Guidelines, is related to the resilience of the bridge network.

From level 0 to level 5 the analyses become more demanding in terms of engineering effort, but it is expected that the number of structures that have to be assessed progressively decreases. The Guidelines also introduce the structural monitoring of bridges. Considerations on structural monitoring are beyond the scope of this work, given the extremely broad nature of the topic.

4.2. Assessment of Structural and Seismic Classes of Attention

The assessment of the Class of Attention for a bridge is fundamental since it is used to determine whether the bridge passes to more rigorous levels of evaluation. The Guidelines consider four different types of hazards:

• Structural and Foundational hazard;
• Seismic hazard;
• Geotechnical hazard;
• Hydraulic hazard.

Each hazard is analyzed separately, defining four independent Classes of Attention, which are then combined to assess the overall Class of Attention of the bridge.

The definition of the Structural and Foundational Class of Attention for the case studies is based on the determination of hazard, vulnerability, and exposure classes. In terms of hazard class, the main parameters are the load level of the bridge and the frequency of commercial vehicles. To estimate the vulnerability class, the level of defectiveness, structural scheme, span length, number of spans, materials, design code, and speed of degradation are considered. The exposure class is defined on the basis of mean daily traffic (MDT), the average length of spans, the presence of road alternatives, the type of entity under the bridge, and the transit of dangerous goods. These three classes, defined using decisional trees that combine the parameters introduced above, are then employed to derive the Structural and Foundational Class of Attention using tables in the Guidelines.

In the definition of the Class of Attention, vulnerability is the most important factor because if it is high the Class of Attention will be High independently from the classes related to the other two factors. In fact, the vulnerability class is directly connected with the level of defectiveness of the structure. A bridge with a poor state of conservation has to be associated with a high Structural Class of Attention, which indicates a high priority.

Table 8. Parameters for the assessment of Structural Hazard class.

Structural Hazard Class
Case Studies	Frequency of Passage of Commercial Vehicles	Classification According to Load Limitations	Hazard Class
Bridge 1	High	A	High
Bridge 2	High	A	High
Bridge 3	High	A	High
Bridge 4	High	A	High
Bridge 5	Medium	A	High

,

Table 9. Parameters for the assessment of Structural Vulnerability class.

Structural Vulnerability Class
Case Studies	Defect Level	Period of Construction or of the Last Maintenance Intervention	Design Code	Support Condition, Length, and Material	Vulnerability Class
Bridge 1	Medium	≥1980	A	High	High
Bridge 2	Medium	1945–1980	A	High	Medium-High
Bridge 3	High	≥1980	A	High	High
Bridge 4	Medium	1945–1980	A	Medium-High	Medium-High
Bridge 5	Medium-High	1945–1980	A	Medium-High	High

,

Table 10. Parameters for the assessment of Structural Exposure class.

Structural Exposure Class
Case Studies	Level of MDT and Medium Length of Spans	Presence of Road Alternatives	Type of Institution Bypassed	Exposure Class
Bridge 1	Medium-High	No road alternatives	Medium	High
Bridge 2	Medium-High	No road alternatives	Medium	High
Bridge 3	Medium-High	No road alternatives	High	High
Bridge 4	Medium-High	No road alternatives	Medium	High
Bridge 5	Medium-Low	No road alternatives	Low	Medium-Low

and

Table 11. Assessment of Structural and Foundational Class of Attention.

Structural and Foundational Class of Attention
Case Studies	Hazard Class	Vulnerability Class	Exposure Class	Class of Attention
Bridge 1	High	High	High	High
Bridge 2	High	Medium-High	High	High
Bridge 3	High	High	High	High
Bridge 4	High	Medium-High	High	High
Bridge 5	High	High	Medium-Low	High

report the definition of the parameters introduced earlier and the assessment of the Structural and Foundational Class of Attention for the five case studies. For all the case studies, the Structural and Foundational Class of Attention is High.

The definition of the Seismic Class of Attention follows the same approach presented for the Structural and Foundational Class of Attention.

In terms of the hazard class, the main parameters are a_g, topographic category, and subsoil type. The structural scheme, span length, materials, level of defectiveness, and design criteria are considered to derive the vulnerability class. The exposure class is derived from the level of MDT, medium length of the spans, presence of road alternatives, type of entity under the bridge, transit of dangerous goods, and strategic importance of the infrastructure. The complete definition of each of the listed parameters can be found in the Guidelines [26].

As in the case of the Structural and Foundational Class of Attention, these three classes are employed to derive the Seismic Class of Attention using tables in the Guidelines.

Table 12. Parameters for the assessment of Seismic Hazard class.

Seismic Hazard Class
Case Studies	ag and Ti	Subsoil Category	Hazard Class
Bridge 1	Medium-High	A	Medium-High
Bridge 2	Medium-High	A	Medium-High
Bridge 3	Medium	B	Medium
Bridge 4	Medium-Low	A	Medium-Low
Bridge 5	Medium-High	A	Medium-High

,

Table 13. Parameters for the assessment of Seismic Vulnerability class.

Seismic Vulnerability Class
Case Studies	Support Condition, Length, and Material	Design Criteria	Defect Level	Vulnerability Class
Bridge 1	High	Not seismic	Medium	Medium-High
Bridge 2	High	Not seismic	Medium	Medium-High
Bridge 3	High	Not seismic	High	High
Bridge 4	High	Not seismic	Medium	Medium-High
Bridge 5	Medium-High	Not seismic	Medium-High	High

,

Table 14. Parameters for the assessment of Seismic Exposure class.

Seismic Exposure Class
Case Studies	Structural Exposure Class	Strategic Structure?	Exposure Class
Bridge 1	High	Yes	High
Bridge 2	High	Yes	High
Bridge 3	High	Yes	High
Bridge 4	High	Yes	High
Bridge 5	Medium-Low	Yes	Medium

and

Table 15. Assessment of Seismic Class of Attention.

Seismic Class of Attention
Case Studies	Hazard Class	Vulnerability Class	Exposure Class	Class of Attention
Bridge 1	Medium-High	Medium-High	High	High
Bridge 2	Medium-High	Medium-High	High	High
Bridge 3	Medium	High	High	High
Bridge 4	Medium-Low	Medium-High	High	Medium-High
Bridge 5	Medium-High	High	Medium	High

report the definition of the parameters introduced earlier and the assessment of the Seismic Class of Attention for the five case studies.

The results indicate that for Bridge 1, Bridge 2, Bridge 3, and Bridge 5, which are located in regions with high seismicity, the Seismic Class of Attention is High, while for Bridge 4, which is located in a low-seismicity region, the Seismic Class of Attention is Medium-High.

5. Discussion

This section provides a critical analysis of the results. First, the main issues found in existing bridges are introduced. The results obtained in terms of fragility curves and Classes of Attention are then discussed and a comparison between the two approaches is provided.

5.1. Main Issues of Existing Bridges

Most of the existing Italian bridges have been built around 50–60 years ago [27], mainly using R.C. or prestressed concrete [53]. Considering that at the time durability was not one of the goals in projects and that the service life of the structures has been in most cases already reached or exceeded, proper maintenance and retrofit interventions are needed. In more than 80% of the cases, the collapse of a bridge is due to problems at the foundational level, overloads, and environmental factors. The rise in the frequency of bridge collapses is a clear indicator of the effects of aging and lack of maintenance [54]. Proper maintenance is fundamental to the management of bridge structures. Critical points of structures (R.C. bridges in particular) are not always easy to detect.

Given that the main causes of collapse in existing bridges are related to floods, impacts, and foundational problems [55], the loads considered in the design are responsible only for a small number of collapses, which is aligned with the expected collapse probability. The analysis of the Italian infrastructure network reveals that the railway system is smaller, more uniform, and more organized in terms of maintenance with respect to the roadway system. Shortage of resources for the maintenance of structures and increasing traffic are also relevant factors [56].

5.2. Fragility Curves

Before discussing the results obtained through the fragility curves, it has to be taken into account that this study is oriented toward a large-scale assessment of existing bridges, so some assumptions and simplifications had to be made. The present discussion refers only to the analyzed case studies and is not meant to have general validity.

In regions with higher seismicity, it has been noted that the response of the bearings is critical. This can be explained considering that, at the time of construction, the design of bearings was carried out in a simplified manner, neglecting the fact that their behavior has a direct influence on the global seismic response of the structure. In regions with lower seismicity, the response of the bearings is not particularly critical, as expected, since the actions due to ground motions are not very high.

The ductile response (chord rotation) of piers is generally good. The response in terms of brittle failures does not highlight excessive problems, except in the case of Bridge 3. The reasons that could explain this result are that the bridge is in a high seismicity region (which implies that the piers are subjected to large forces), the structure of the pier is not beneficial since a localized failure can have severe consequences, and also the material properties were defined on the basis of the results of experimental tests performed to investigate the degradation of the structure.

For the connecting elements of frame piers, the main issues are related to the critical response in terms of brittle failures. This phenomenon could be explained considering that the spacing of the stirrups in these elements is quite large, which leads to issues when compared with shear forces. This is an aspect that should not be underestimated as failures in the connection elements can modify the flexural behavior of the frame piers, with direct consequences on the global response of the bridge. As expected, the seismic behavior of existing bridges can show some issues directly related to the seismicity of the region. These results are confirmed by the literature [53,55] where it is reported that bearings and piers are between the most vulnerable parts of bridges under the effect of earthquakes.

Some remarks have to be made on the methodology employed to develop the fragility curves. Using the MSA, the selection of ground motions is usually made by matching a Conditional Mean Spectrum [57], while in this work the matching operation was performed using a code-based spectrum specified by the NTC2018, which is based on the Uniform Hazard Spectrum. There are some specific reasons that motivated this choice. The Guidelines are strongly code-oriented so following the approach proposed by NTC2018 to select the ground motions has been considered consistent with the multilevel approach, in order to discuss the differences between Classes of Attention and Fragility Curves. As stated earlier, this study is oriented toward a large-scale assessment of existing bridges. While matching ground motions to a Conditional Mean Spectrum is less conservative, the uncertainties related to existing bridges are quite high, and given that a code-based spectrum is essentially an envelope of spectral acceleration values at all periods, matching ground motions to a Uniform Hazard Spectrum allows for obtaining a certain degree of conservatism in the assessment.

In the literature, examples of large-scale assessments can be found. The research of Borzi et al. [50] has its focus on the seismic vulnerability evaluation of the Italian bridge network and is here recalled since the approach that was used is essentially equal to the one employed in the present study (with the difference that natural records have been used in the present study). The proposed methodology, part of a framework for the automatic assessment of the Italian bridge network, employed artificial records to perform the NLTHAs needed to develop the fragility curves. It has been reported that collapse fragilities are more conservative if artificial accelerograms are used and that the comparison with natural records shows that the differences, which are non-negligible, are accepted for the purposes of the large-scale vulnerability assessment.

A recent study by Manfredi et al. [41] addresses the selection of ground motions for seismic fragility analyses. The research highlights the importance of ground motion selection in the construction of fragility curves using NLTHAs. The proposed methodology is almost the same as the one presented in this work, with only slight differences. Eight code-based spectra are defined following the indications reported in NTC2018 and EC8. The T_r of the elastic spectra goes from 50 to 10,000 years. This ensures that the selected ground motions represent a broad range of IM levels, gradually increasing from low to high levels. Two sets of 125 natural accelerograms are then selected, for stiff and soft soil conditions, matching the code-based spectra introduced earlier. The ground motions are then employed to perform NLTHAs on a benchmark structural model, analyzing the results to check if the selected sets of natural records are able to cover all the damage levels generally adopted in seismic risk analyses. The results of the study demonstrate that the selected sets are able to populate all the considered damage levels, providing reliable estimates of fragility parameters. The study is oriented, as the present work is, to a large-scale assessment.

5.3. Bridge Assessment Guidelines

The central fulcrum of the multilevel multi-risk approach proposed by the Guidelines is Level 2, where the assessment of the Class of Attention is performed. In the present work, only Structural and Foundational and Seismic Classes of Attention have been assessed, since the assessment of Geotechnical hazard and Hydraulic hazard was beyond the scope of this work. The discussion here reported only refers to Structural and Foundational and Seismic Classes of Attention and is intended as an early evaluation of the approach proposed by the Guidelines. It is important to remark that assumptions on the degradation level have been made when no data were available.

In terms of the Structural and Foundational Class of Attention, each of the case studies has been associated with the High Class of Attention. The assessment procedure appears to be very conservative, even if the limited number of case studies does not make it possible to properly address this aspect. As stated earlier, the assessment of Classes of Attention is based on hazard, vulnerability, and exposure parameters and vulnerability is the most important factor in the assessment. Similar considerations can be made about the Seismic Class of Attention. Four of the case study bridges (Bridge 1, Bridge 2, Bridge 3, and Bridge 5) have been associated with the High Seismic Class of Attention, while Bridge 4 has been associated with the Medium-High Seismic Class of Attention. This result can be explained considering that Bridge 1, Bridge 2, Bridge 3, and Bridge 5 are located in regions with high seismicity, while Bridge 4 is located in a low seismicity region. Definitive evaluations about the conservativeness of the approach cannot be made because of the small sample of structures considered in this study, but it has to be pointed out that some studies are available in the literature about the Guidelines. For instance, the study by Santarsiero et al. [58] applied the Level 2 approach to 48 bridges, making assumptions on the degradation level when no data were available, highlighting that a certain safety level was present in the approach proposed to assess the Class of Attention.

5.4. Comparison between the Two Approaches

The present comparison aims to qualitatively evaluate the differences between two approaches with different safety levels applied to the same case studies. Since the assessment of the Seismic Class of Attention is based on decisional trees and visual inspections, it is expected that it will be more conservative than an analytic approach based on NLTHAs.

Comparing the results obtained in terms of Seismic Class of Attention with those obtained in terms of fragility curves, it can be noted that for bridges in high seismicity regions (Bridge 1, Bridge 2, Bridge 3, and Bridge 5), there is an agreement between the results of the two approaches while for Bridge 4, located in a low seismicity region, there is a discrepancy. The differences in the results obtained for Bridge 4 are quite important. The Seismic Class of Attention of the bridge is Medium-High, while the NLTHAs did not highlight particular issues in terms of the response of bearings and structural elements. Considering that only the curve corresponding to γ = 0.5 was constructed, since even for T_r = 2475 years the seismic action was not particularly intense, it seems that for Bridge 4, the approach of Level 2 gives a result that does not capture in a proper way the seismic behavior of a bridge in a low seismicity region. One possible reason for this difference could be the different levels of conservatism of the methods. An extended sample of bridges has to be considered to improve the reliability of this qualitative comparison and to evaluate in a more accurate way the framework proposed by the Guidelines. Considering that the topic of the seismic vulnerability assessment of bridge structures has been significantly developed in recent years, comparing the results with other approaches based on seismic vulnerability would increase the accuracy in the evaluation of the approach proposed by the Guidelines. For instance, Ozsarac et al. [59] explored the use of simulated records in the assessment of R.C. bridges; Biondini et al. [60] proposed and implemented a procedure based on a two-level structure-network risk assessment for bridge prioritization, while the work of Zhang et al. [61] introduces a technique that employs artificial intelligence for the large scale vulnerability assessment of bridge structures.

6. Conclusions

In the present research, the structural performance of five bridges belonging to the Italian Road Network has been assessed using different techniques. For each of the case studies, fragility curves have been computed, by defining the target Limit States and the analysis method. The obtained results have been then used to evaluate the efficiency of the procedure set out by the Guidelines. It has to be pointed out that the present study is oriented toward a large-scale assessment of existing bridges, so some assumptions and simplifications had to be made. The present results refer only to the analyzed case studies and are not meant to have general validity.

Considering the fragility curves, in regions with higher seismicity, it has been noted that the response of bridge bearings is critical, while in lower seismicity regions, the response of the devices is not particularly detrimental. In terms of ductile mechanisms, a good response of the piers has been noted. Good behavior in terms of brittle mechanisms has been noted as well, apart from Bridge 3. Considering the connecting elements in frame piers, local problems (mainly in terms of brittle mechanisms) have been found. This is an aspect that must not be underestimated as failures in these elements can modify the flexural behavior of the piers and thus the overall response of the structure. The results obtained interpreting the analytical fragility curves agree with the fact that the seismic behavior of existing bridges can be problematic and that a higher seismicity can be associated with more detrimental behavior.

Considering the Guidelines, each case study is associated with the High Structural Class of Attention. The four bridges in higher seismicity zones have been associated with the High Seismic Class of Attention, while Bridge 4 has been associated with the Medium-High Seismic Class of Attention. For bridges in areas with higher seismicity, there is, therefore, a result in agreement with that of the fragility curves, while for Bridge 4, there is a discrepancy that could be due to the different levels of conservatism of the analyses. The procedure of the Guidelines could lead to not reflecting the seismic behavior of bridges when the seismicity of the area is lower.

Since the assessment of Classes of Attention is based on decisional trees and visual inspections, it is expected that it will be more conservative than an analytic approach based on NLTHAs. An extended sample of bridges has to be considered to improve the reliability of the reported qualitative comparison and to evaluate in a more accurate way the framework proposed by the Guidelines. Future developments of this work could consider other approaches focused on the seismic vulnerability of bridges to increase the accuracy in the evaluation of the approach proposed by the Guidelines.