PSI - Issue 62

ScienceDirect Available online at www.sciencedirect.com ScienceDirect Structural Integrity Procedia 00 (2022) 000 – 000 Available online at www.sciencedirect.com ScienceDirect Structural Integrity Procedia 00 (2022) 000 – 000 Available online at www.sciencedirect.com Procedia Structural Integrity 62 (2024) 832–839

www.elsevier.com/locate/procedia

www.elsevier.com/locate/procedia

II Fabre Conference – Existing bridges, viaducts and tunnels: research, innovation and applications (FABRE24) A Bayesian network-based framework for SHM data fusion supporting bridge management II Fabre Conference – Existing bridges, viaducts and tunnels: research, innovation and applications (FABRE24) A Bayesian network-based framework for SHM data fusion supporting bridge management Laura Ierimonti a, *, Francesco Mariani a , Ilaria Venanzi a , Filippo Ubertini a a) Department of Civil and Environmental Engineering, Via G. Duranti, 06125, Perugia, Italy Abstract Guidelines for Risk Classification and Management, Safety Assessment and Monitoring of Bridges, issued in 2020 by the Italian Ministry of Sustainable Mobility, promote the use of Structural Health Monitoring (SHM) systems for risk assessment of bridges. Prestressed or ordinary reinforced concrete bridges are designed to maintain their functionality over a long period of time. However, during their service life, such structures may be exposed to ever- increasing traffic volumes, extreme weather conditions and/or marine environments and so on. Therefore, constant maintenance and timely interventions are crucial to ensure the functionality of the transportation network. The current management process is mainly based on visual inspections, while the aggregation of information coming from multiple sources is still a major challenge. In this context, the main objective of the present work is to develop a general Bayesian framework capable of processing the different sources of information based on a Bayesian network (BN), a probabilistic graphical model that represents a set of variables and their conditional dependencies. A BN is a useful tool for the management of bridges since it allows the prediction of damage states and failure conditions based on the knowledge by visual inspections and SHM. The combined use of SHM and Bayesian approaches can reduce the overall risk of failure increasing the efficiency of the infrastructure system. The procedure is exemplified and validated with reference to a simply supported post-tensioned case study bridge. © 2024 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Scientific Board Members Laura Ierimonti a, *, Francesco Mariani a , Ilaria Venanzi a , Filippo Ubertini a a) Department of Civil and Environmental Engineering, Via G. Duranti, 06125, Perugia, Italy Abstract Guidelines for Risk Classification and Management, Safety Assessment and Monitoring of Bridges, issued in 2020 by the Italian Ministry of Sustainable Mobility, promote the use of Structural Health Monitoring (SHM) systems for risk assessment of bridges. Prestressed or ordinary reinforced concrete bridges are designed to maintain their functionality over a long period of time. However, during their service life, such structures may be exposed to ever- increasing traffic volumes, extreme weather conditions and/or marine environments and so on. Therefore, constant maintenance and timely interventions are crucial to ensure the functionality of the transportation network. The current management process is mainly based on visual inspections, while the aggregation of information coming from multiple sources is still a major challenge. In this context, the main objective of the present work is to develop a general Bayesian framework capable of processing the different sources of information based on a Bayesian network (BN), a probabilistic graphical model that represents a set of variables and their conditional dependencies. A BN is a useful tool for the management of bridges since it allows the prediction of damage states and failure conditions based on the knowledge by visual inspections and SHM. The combined use of SHM and Bayesian approaches can reduce the overall risk of failure increasing the efficiency of the infrastructure system. The procedure is exemplified and validated with reference to a simply supported post-tensioned case study bridge. Keywords: Structural health monitoring; Bayesian networks; Bayesian model class selection; Decision making, Bridges.

Keywords: Structural health monitoring; Bayesian networks; Bayesian model class selection; Decision making, Bridges.

* Corresponding author. E-mail address: laura.ierimonti@unipg.it

2452-3216 © 2024 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Scientific Board Members 10.1016/j.prostr.2024.09.112 2452-3216 © 2024 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license ( https://creativecommons.org/licenses/by-nc-nd/4. 0 ) Peer-review under responsibility of Scientific Board Member s 2452-3216 © 2024 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license ( https://creativecommons.org/licenses/by-nc-nd/4. 0 ) Peer-review under responsibility of Scientific Board Member s * Corresponding author. E-mail address: laura.ierimonti@unipg.it

Laura Ierimonti et al. / Procedia Structural Integrity 62 (2024) 832–839 Ierimonti etal. / Structural Integrity Procedia 00 (2019) 000 – 000

833

2

1. Introduction In the domain of bridge construction and management, the escalating integration of SHM systems marks a transformative data-driven era (Laflamme et al., 2023), a trend further supported by national codes, such as the Italian Guidelines on Risk Classification and Management of Bridges (LLGG, 2020). However, the growing prevalence of SHM introduces a new challenge: effectively managing the complexity of the generated data. Leveraging this wealth of information requires a high level of expertise, placing a substantial burden on bridge management systems. Among data-driven approaches, which rely solely on measured data to detect damage (Jiang et al., 2023; Bunce et al., 2023), model-based techniques are spreading out, which necessitates a deep understanding of the structural characteristics to reconstruct the mathematical model of the bridge. Recent advancements in SHM have embraced Bayesian model updating techniques, addressing various sources of uncertainty (Behmanesh and Moaveni, 2016; Ierimonti et al., 2021; Mao et al., 2023; Ierimonti et al., 2023;). Alongside SHM-based techniques, visual inspections of bridges are crucial for aging process tracking, safety assurance, risk mitigation and long-term maintenance planning. Hence, innovative, and user-friendly methodologies are essential, with a focus on optimizing the integration of diverse data sources to facilitate prompt and effective decision-making in maintenance practices. Data fusion, deployed at various levels, presents a promising avenue for enhancing the precision and dependability of SHM information. The recent surge in the adoption of Artificial Intelligence (AI)-based methodologies, though requiring substantial volumes of training data, has become notable. Given the restricted availability of labelled data for monitoring bridges, alternative techniques have gained prominence. Bayesian Networks (BNs) distinguish themselves as pivotal tools, thanks to their capacity to accommodate a wide range of uncertainties (Tubaldi et al., 2022). Notably, Dynamic Bayesian Networks (DBNs) are particularly compelling due to their integration of the time dimension (Xu et al., 2022), facilitating the efficient analysis of evolving system states. Despite being in their nascent stages within the field, DBNs have a huge potential to revolutionize the landscape of infrastructure monitoring. In the above depicted context, this paper introduces an innovative DBN-based unified framework for post-tensioned bridges, integrating SHM static data and results from visual inspections related to prestress losses. Encompassing key stages, including the selection of damage scenarios for SHM monitoring, numerical modeling, post-processing of SHM data, Bayesian model selection, and DBN-based data fusion, this framework aims to assess the risk of bridge failure. More in depth, DBNs enable handling incomplete or uncertain information, incorporating prior knowledge. In evaluating the performance of the proposed method, a single-span bridge serves as the testbed, simulating prestress- dependent damaging scenarios, environmental noise, and instrumental errors. To achieve this, a simple monitoring system based on inclinometers is employed, ensuring cost-effectiveness and ease of use. The inclinometers focus on measuring the inclination angle at the designated points. The data collected from these sensors can be utilized to estimate the deformed shape through analytical functions that establish a connection between the rotation and deflection. This testing scenario allows for a comprehensive assessment of the method's efficacy under varying conditions, providing insights into its robustness and reliability in real-world applications. The simplicity and affordability of the monitoring system further enhance the method's practicality and applicability across a range of bridge structures. The subsequent sections of this paper undertake an in-depth exploration of the framework, delving into its theoretical insights (Section 2), comprehensive framework description (Section 3), details of the case study (Section 4), main results (Section 5) and concluding insights. 2. Theoretical insights 2.1. Visual inspection-based damage assessment for post-tensioned bridges The visual inspection process conducted on existing bridges aims to evaluate the presence of surface defects, which may necessitate further examination. It is crucial to emphasize that the visual assessment of a bridge is fundamentally qualitative, providing a foundation for guiding repair and maintenance actions based on reports and photographic evidence of any structural anomalies. Aligned with LLGG 2020, the level of structural defectiveness (DL) is characterized by assigning discrete defect levels to each structural component. Each defect is defined by several

Laura Ierimonti et al. / Procedia Structural Integrity 62 (2024) 832–839 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

834

3

attributes: G (1-5), indicating the significance of the defects, which is pre-defined; K1 (0.2-0.5-1), denoting the extent or spread of the defect; K2 (0.2-0.5-1), reflecting the intensity or severity of the defect. In this context, according to LLGG (2020) and to the document on post-tensioned bridges special inspections (FABRE, 2022), the deteriorating defects related to the prestressing system can be categorized into two macro groups: apparent defects and latent defects. An example of common apparent defects is the degradation of sheaths, and the oxidation of wires. Cracks exhibiting a longitudinal pattern along the beam serve as a classic example of latent defects. To determine the level of defects within structural components and the overall structure, all the attributes are combined, resulting in the following tags: (i) low, (ii) medium, (iii) high. Subsequently, in conjunction with defining the defectiveness level, it is crucial to correlate the inspected defects with the monitored damage scenarios, leveraging expertise knowledge. 2.2. SHM-based bridge deflection reconstruction Ensuring the robust health of a bridge and implementing timely maintenance measures are essential actions to mitigate latent risks in bridge safety and prevent potential safety issues. Given that deflection serves as an informative indicator reflecting prestress losses, the measurement of deflection stands as a pivotal task in the comprehensive ever- evolving realm of bridge health monitoring. Experimentally, the bridge deflection can be reconstructed from monitoring data (Lan etal., 2019) by means of measured rotations ∗ at selected positions along the bridge, by means of an appropriate calibrated polynomial function: ሺ ǡ ሻ ൌ ሺ ሻሾ ͳ ሺ ሻ ͵ ൅ ʹ ሺ ሻ ʹ ൅ ͵ ሺ ሻ ൅ Ͷ ሺ ሻሿ (1) where ሺ ሻ is a function that respects the boundary conditions, in the present case of a simply supported beam, defined as ሺ ሻ ൌ ሺ − ሻ ; the terms ሺ ሻ are the unknown coefficients to be calibrated through experimental data at each time step according to the following equation: ∗ ሺ ǡ ሻ ൌ ሺ ǡ ሻȀ (2) where represents the partial derivative function and w ( x , t ) is the vertical deflection of the beam. The optimal solutions for coefficients ሺ ሻ can be obtained by using the least square method (Lan etal., 2019). A minimum of 5 inclinometers should be used to find out accurate solutions (Hou etal., 2005). 2.3. Novelty detection Novelty detection is here performed by using Hotelling’s Control charts involving monitoring structural properties or sensor measurements over time (Hotelling, 1947). The Hotelling’s Control T -squared (T²) chart is a statistical tool that helps to identify deviations from the expected behaviour in multivariate data. In the context of SHM, Hotelling's T² assesses whether the mean vector of a multivariate dataset is significantly different from a reference or baseline mean vector and it can be evaluated as follows: ʹ ൌ ሺ − ሻ Σ − ͳ ሺ − ሻ (3) where is the analyzed data set; is the mean value of the residual error R between the deflection time series and the predicted values evaluated within the range ; and Σ are the mean value and the covariance matrix of R evaluated in the training period (typically one year of measurements), respectively. The chart's space is divided vertically into two sub-regions using a specific upper control limit (UCL). The separation into these sub-regions is guided by statistical considerations, and the chosen UCL serves as a threshold to identify and distinguish normal behaviour from potential anomalies in the monitored data. The Hotelling Control chart provides a way to visually and quantitatively assess whether the observed data points fall within the expected range, highlighting potential anomalies that warrant further investigation in the monitored structure.

Laura Ierimonti et al. / Procedia Structural Integrity 62 (2024) 832–839 Ierimonti etal. / Structural Integrity Procedia 00 (2019) 000 – 000

835

4

2.4. Bayesian model class selection Bayesian model class selection (BMCS) is a probabilistic-based technique employed to discern the best fitting model from a pool of candidates, relying on input-output data. In the realm of SHM, these competing models often encapsulate distinct damage parametrizations linked to various damaging mechanisms. The essence of BMCS lies in harnessing Bayesian statistics (Yuen, 2010), a methodology that integrates prior knowledge to refine the probability of a hypothesis at a specific time. For the selection of the best fitting model among the candidates, the Bayesian Information Criterion (BIC) can be adopted. The computation of BIC is expressed as: (ℳ) ൌ −2 ⋅ Ž‘‰ ሺ ሺ ℳ)) ൅ ⋅ Ž‘‰ ሺ ሻ (4) where: ℳ denotes the selected mathematical model class; ሺ ℳ ሻ represents the likelihood function which assesses the agreement between measured and predicted data at time ; denotes the number of model parameters, and stands for the number of data points. The BIC criterion achieves a balance between model fit (likelihood) and complexity, penalizing the inclusion of parameters and accounting for sample size. This approach provides a principled method to select a model that effectively fits the data, preventing overfitting. In the context of BMCS, the model with the lowest BIC value is selected as the most suitable, striking a compromise between the goodness of fit and model simplicity. 2.5. Dynamic Bayesian networks Dynamic Bayesian Networks represent a powerful class of probabilistic graphical models, uniquely suited for tackling a diverse array of problems (Koller and Friedman 2009). Their structured design facilitates the systematic computation of conditional probabilities grounded in available evidence, offering a transparent and intuitive understanding of how objects interact over time. Consequently, DBNs find extensive application in domains requiring automated decision-making. At their core, a BN serves as a model for capturing the joint probabilities of the events it represents. The construction of a Bayesian Network typically involves defining three key components: (i) nodes, which depict random variables; (ii) edges, which establish relationships between variables, also designated as parent and child; (iii) conditional probabilities, which relate the nodes. The behaviour of each child node, i.e., of the variables representing the node, is primarily influenced by its parents. These connections are pivotal in comprehending how different factors interact and are indispensable for making well- informed decisions based on available evidence. Let ℊ ሺ ሻ be a Bayesian Network graph over the variables ͳ ሺ ሻǡ … ǡ ሺ ሻ . It is possible to evaluate the distribution ሺ ͳ ǡ … ǡ ሻ according to the chain rule for BNs (Koller and Friedman 2009): ሺ ǡ … ǡ ǡ ሻ ൌ ሺ ሺ ሻ ∣ ƒ ℊ ǡ ሻ ͳ ൌͳ (5) where ƒ represents the parents of and ሺ ሺ ሻ ∣ ƒሻ are the conditional probability distributions of , which may vary over time based on evidence. Indeed, results based on a BN can be updated in two primary ways: by adjusting conditional probabilities, reflecting new information or prior knowledge, and by making inferences on node attributes, allowing for updates and refinements as new data becomes available. This can be achieved using various algorithms, such as exact or approximate inference algorithms. In the present paper, the variable elimination algorithm is used which consists of marginalizing out the variables that are not in the subset of the variables involved in the evidence process: ሺ ǡ ሻ ൌ ∑ ∖ ሺ ∪ ሻ ሺ ሻ (6) where represents the remaining variables among those in , ∑ ∖ ሺ ∪ ሻ denotes the summation over all possible values of the variables in , excluding those in the sets and . Specifically, the variable elimination process

Laura Ierimonti et al. / Procedia Structural Integrity 62 (2024) 832–839 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

836

5

involves summing over the unwanted variables to obtain the marginal distribution of interest (in this case, E and Z). This method is a key step in simplifying computations in BN, especially when dealing with a large number of variables, making the calculations more efficient. Finally, BN allows forward and backword propagation to update the probabilistic information of any node given evidence. 3. The methodology The core of the proposed framework relies on continuously updating the knowledge regarding the health status of the bridge over time, utilizing the outcomes of the DBN. The amalgamation of all risks associated with each node can offer insights into the overall risk assessment and the most probable Damage Model (DM) for the monitored Damage Scenario (DS) within the bridge context. In a broader sense, the suggested framework encompasses the following procedural steps (Fig. 1): 1) Selection of potential Damage Scenarios (DS) and corresponding damage models (node DM) that can be effectively monitored utilizing the existing SHM system. It is noteworthy that the same sensors may have the capability to monitor one or multiple DS. 2) Novelty detection (node NY, with attributes yes or no). Subsequently, the acquired data undergoes a comprehensive analysis and preprocessing stage. This involves removing environmental effects and interferences from the original signals, enabling the anomaly detection (Novelty), i.e., data-driven detection of abnormal or unusual observations.

Fig. 1. The methodology: a schematic representation.

3) Bayesian model class selection for each DS j . In this phase, the most likely DM is determined, leading to evidence or to update conditional probabilities in the DS-dependent DM-SHM node. 4) Establish the severity of damage (low, medium, high) on the basis of the results of the model updating, i.e., node S-SHM. 5) If an on-site inspection is not performed, proceed to step 8. If an on-site inspection is conducted, along the inspected DL, it is essential to establish a correlation between the observed defects and the monitored DM (see Section 2.1). This correlation is based on expert knowledge and may lead to prioritizing DMs by inferring the node DM-VI. 6) Establish the severity of damage (low, medium, high) on the basis of the results of the visual inspection, i.e., node S-VI. 7) Risk evaluation as a consequence of model based SHM and VI evidence (node RI2). A risk index with attributes low, medium and high is evaluated on the basis of the results obtained from steps 4-6. 8) Risk evaluation (Node RI1). A risk index endowed with attributes low, medium and high is evaluated based on the results obtained from steps 1 and 7.

Laura Ierimonti et al. / Procedia Structural Integrity 62 (2024) 832–839 Ierimonti etal. / Structural Integrity Procedia 00 (2019) 000 – 000

837

6

9) Decision making. Decisions, i.e., do nothing or repair, can be updated on the evidence unveiled from the network at time t . 10) Update damage probabilities in the Decision-Making process; if a repair action is performed, restore the health condition to an undamaged state. Otherwise, maintain conditional probabilities equal to those of the previous time step. 4. Description of the case study The analyzed case study involves a simply supported, post-tensioned single-span concrete bridge. The deck comprises a concrete slab with a thickness of 0.2 m and is supported by three longitudinal post-tensioned I-shaped beams. The total span of the bridge is 32 m. The post-tensioning system (Figure 2b) comprises 5 tendons with varying numbers of strands, each characterized by a diameter of 7 mm. Specifically, cables number 1 and 2, positioned at the beam intradox, are composed of 32 strands, while the remaining tendons (number 3, 4, and 5) consist of 42 strands. The SHM system is composed by 5 inclinometers, equally distributed along the exterior beam (Figure 2a).

Fig. 2. The case study: a) SHM system and reconstructed deflection.; b) Beam’s sections with the indication of tendons T1-T5.

For the numerical application, 9 months of data are artificially simulated, taking into account measurement uncertainties associated with sensor precision and daily traffic flow. Data are recorded with a frequency of one data point every 30 minutes. The deflection of the exterior beam, evaluated according to Eq. 1, is depicted in Fig. 2. For the numerical application, potential losses in the prestress system are selected as DS and 4 different models are considered and associated with node DM: 1) N0, undamaged model; 2) MP1, malfunctioning of cable T5; 3) MP3, malfunctioning of cables T3, T4, T5; 4) MP5, malfunctioning of alle the cables. 5. Main results This section offers an overview of the primary research findings. The main objective is to assess the impact of prestress losses on deflection. In Fig. 3a), the results of a parametric analysis of normalized maximum deflection concerning the undamaged case are presented. This analysis considers the intensity of the damage, ranging from 0.2 to 1. For instance, a value of 1 signifies the absence of a cable, indicating a prestress loss of 100%. The figure highlights that Model MP3 is the most influential, demonstrating a substantial increase in maximum deflection. Subsequently, a damage is simulated based on Model MP3, with an intensity set at 0.5. Fig. 3b) illustrates the Hotelling’s control Chart evaluated within the monitored period, including both undamaged and damaged states. The presence of outliers, i.e., novelty detection, implies the evidence in NY node, as can be depicted from the figure. Fig. 3c) illustrates the results of BMCS in terms of BIC relative differences, i.e., ∆ ൌ ሺ ℳ ሻ − ‹ ሺ ሺ ℳ ͳ ǡ Ǥ Ǥ ǡ ℳ ሻሻ , where the term ‹ ሺ ሺ ℳ ͳ ǡ Ǥ Ǥ ǡ ℳ ሻሻ ൌ ሺ ͵ሻ , which implies an evidence in node DM-SHM. Finally, Fig. 4 provides a summary of the DBN results, drawing inferences on: (i) Novelty detection; (ii) possible DM identified through the simulated model-based SHM and VI; (iii) severity of damage based on VI results; and (iv) repair action.

Laura Ierimonti et al. / Procedia Structural Integrity 62 (2024) 832–839 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

838

7

The results indicate that in the absence of evidence and with the model remaining undisturbed, the most probable damage model is N0 (undamaged), the Decision-Making favors the action of doing nothing, and the risk index RI1 has a higher probability for the low level.

Fig. 3. Damage models: a) Parametric analysis as a function of intensity of damage; b) Hoteling’s Control Chart.

This suggests that the conditional probabilities in the Bayesian Network, selected by engineering expertise, have been fine-tuned to reflect a low likelihood of damage when no evidence is available. Subsequently, novelty detection updates the network, impacting both Decision-Making and risk RI1, resulting in an increased probability of repair and a shift in risk level from low to medium-low. Conversely, DM is minimally affected by novelty evidence, aligning with the data-driven nature of the information. Subsequently, based on the BIC results, the node DM-SHM is inferred. Consequently, from the DM node, the model MP3 emerges as the most feasible one. This information, in turn, triggers an increased need for a repair action and a slight elevation of index RI1.

Fig. 4. DBN results over time considering different information: a) node DM; b) node Decision Making; c) node RI1.

To underscore the potential influence of visual inspections, the results are then updated taking into account possible VI outcomes: evidence on model MP1 (as visual inspections may find it challenging to correctly identify damage models with latent defects) and evidence on the severity of damage (medium and high). The results reveal that if different DMs are identified from SHM and VI, the probabilities are adjusted accordingly, giving higher importance to both determined models. Moreover, the increase in damage severity does not affect the probabilities of the DM node but leads to an escalation in the suggested action (repair) and risk. Finally, evidence pointing to a repair action restores the structure to its undamaged condition.

Laura Ierimonti et al. / Procedia Structural Integrity 62 (2024) 832–839 Ierimonti etal. / Structural Integrity Procedia 00 (2019) 000 – 000

839

8

6. Conclusions In the context of this study, an innovative unified framework for bridges, built upon DBN integrating Structural Health Monitoring (SHM) data and findings from visual inspections is proposed. This comprehensive framework covers the selection of DS and DM, the numerical modeling, the post-processing of SHM data, Bayesian model class selection, correlation between inspected defects and monitored damaged scenarios, and the updating of the belief DBN to support decision making. To validate the effectiveness of the approach, a single span post-tensioned bridge was selected as a case study, and DS related to the prestress system were simulated. DBNs are probabilistic graphical models that excel in representing and reasoning about dynamic systems over time. Their ability to model uncertainty and integrate prior knowledge is crucial in the field of SHM where data may be incomplete or subject to uncertainty. This makes DBNs a particularly effective choice for data fusion in bridge monitoring systems. Acknowledgements The first author acknowledges support by the PNRR project “STRIC - Centro internazionale per la ricerca sulle scienze e tecniche della ricostruzione fisica, economica e sociale” (in Italian) and by University of Perugia via the funded project “Study of multi-risk scenarios for natural disasters in the central-southern Italy and Sicily area: Understanding the past and present to protect the future” within the program “Fondo Ricerca di Ateneo, 2021. The second author acknowledges funding by FABRE – “Research consortium for the evaluation and monitoring of bridges, viaducts and other structures” ( www.consorziofabre.it/en) within the activities of the FABRE-ANAS 2021-2024 research program. The third and fourth authors acknowledge funding by the Italian Ministry of Education, University and Research (MIUR) through the project of national interest “TIMING – Time evolution laws for IMproving the structural reliability evaluation of existING post- tensioned concrete deck bridges” (Protocol No. P20223Y947). Any opinion expressed in the paper does not necessarily reflect the view of the funders. Behmanesh, I. and Moaveni, B., 2016. Accounting for environmental variability, modeling errors, and parameter estimation uncertainties in structural identification, Journal of Sound and Vibration, 374: 92-110. Brincker, R. and Ventura, C. E., 2015. Introduction to Operational Modal Analysis, John Wiley & Sons, Ltd. Bunce, A. and Hester, D. and Taylor, S. and Brownjohn, J. and Huseynov, F. and Xu, Y., 2023. A robust approach to calculating bridge displacements from unfiltered accelerations for highway and railway bridges, Mechanical Systems and Signal Processing, 200. H. Hotteling, 1947. Multivariate quality control, illustrated by the air testing of sample bombsights, Techniques of statistical analysis.111 – 184. Ierimonti, L., Cavalagli, N., Venanzi, I., García-Macías, E., Ubertini, F., 2023. A Bayesian-based inspection-monitoring data fusion approach for historical buildings and its post-earthquake application to a monumental masonry palace. Bulletin of Earthquake Engineering, 21 (2), pp. 1139- 1172 Ierimonti, L., Cavalagli, N., Venanzi, I., García-Macías, E., Ubertini, F., 2021, A transfer Bayesian learning methodology for structural health monitoring of monumental structures. Engineering Structures, 247, art. no. 113089. Jiang, H., Ge, E., Wan, C., Li, S., Quek, S.-T., Yang, Ding, K. Y., Xue, S., 2023. Data anomaly detection with automatic feature selection and deep learning, Structures, 57. Koller, D. and Friedman, N., 2009. Probabilistic Graphical Models: Principles and Techniques, MIT Press. Laflamme S., Ubertini F., Di Matteo A., Pirrotta A., Perry M., Fu Y., Li J., Wang H., Hoang T., Glisic B., Bond L.J., Pereira M., Shu Y., Loh K.J., Wang Y., Ding S., Wang X., Yu X., Han B., Goldfeld Y., Ryu D., Napolitano R., Moreu F., Giardina G., Milillo P., 2023. Roadmap on measurement technologies for next generation structural health monitoring systems, Measurement Science and Technology, 34 (9), art. no. 09300. Lan, R., Wang, Y. & Chi, Q. 2019. Reconstitution of Static Deflections of Suspension Bridge Based on Inclinometer Data. IOP Conference Series: Earth and Environmental Science. Hou, X., Yang, X. & Huang, Q. 2005. Using inclinometers to measure bridge deflection. Journal Of Bridge Engineering. LLGG, 2020. Ministry of Infrastructure, CSLP: Guidelines on risk classification and management, safety assessment and monitoring of existing bridges. Mao, J., Su, X., Wang, H. and Li, J., 2023. Automated Bayesian operational modal analysis of the long-span bridge using machine-learning algorithms, Engineering Structures, 289. FABRE, 2022. Special inspections on existing post-tensioned bridges according to the LLGG: classification and accurate assessment. Tubaldi, E. and Turchetti, F. and Ozer, E. and Fayaz, J. and Gehl, P. and Galasso, C., 2022. A Bayesian network-based probabilistic framework for updating aftershock risk of bridges, Earthquake Engineering and Structural Dynamics, 51(10): 2496-2519. Xu, Y. and Zhu, B. and Zhang, Z. and Chen, J., 2022. Hierarchical dynamic Bayesian network-based fatigue crack propagation modelling considering initial defects, Sensors, 22(18): 6777. Yuen, K.V., 2010. Bayesian methods for structural dynamics and civil engineering, John Wiley & Sons Ltd. References

ScienceDirect Available online at www.sciencedirect.com ScienceDirect Structural Integrity Procedia 00 (2022) 000 – 000 Available online at www.sciencedirect.com ScienceDirect Structural Integrity Procedia 00 (2022) 000 – 000 Available online at www.sciencedirect.com Procedia Structural Integrity 62 (2024) 73–80

www.elsevier.com/locate/procedia

www.elsevier.com/locate/procedia

II Fabre Conference – Existing bridges, viaducts and tunnels: research, innovation and applications (FABRE24) A BIM-based approach for risk assessment and management in roadway bridges: preliminary results from the application to a post tensioned concrete box girder bridge Andrea Meoni a, *, Matteo Castellani a , Francesco Mariani a , Marco Ceccobelli a , Ilaria Venanzi a , Filippo Ubertini a a Department of Civil and Enviromental Engineering, University of Perugia, Via G. Duranti 93, 06125 Perugia (PG), Italy Abstract The Guidelines for Risk Classification and Management, Safety Assessment, and Monitoring of Existing Bridges, released in 2020 by the Italian Ministry of Infrastructures and Transport, promote the use of Building Information Modeling (BIM) environment as one of the main tools to achieve informed risk mitigation of infrastructures. To date, BIM environments are often only used as repositories of information from different entities involved in bridge design, management, and retrofitting. However, to achieve informed risk mitigation of bridges, the current BIM paradigms should be expanded by integrating new typologies of data, such as those from Structural Health Monitoring (SHM) systems, and functionalities, such as the ability to process stored data directly within the BIM environments. In this light, this paper presents a BIM-based approach conceived to achieve comprehensive assessment and management of the risk conditions and structural performance of bridges. To this end, specific algorithms have been implemented into the selected BIM environment to perform tasks such as assessing the preliminary risk level of a bridge according to the methodology proposed by the Italian Guidelines, evaluating the outcomes from visual inspections to estimate its defect level, storing and processing data from SHM systems installed on the structure, and more. The paper also presents the first application of the proposed BIM-based approach to a post-tensioned box girder bridge belonging to the Italian road network. II Fabre Conference – Existing bridges, viaducts and tunnels: research, innovation and applications (FABRE24) A BIM-based approach for risk assessment and management in roadway bridges: preliminary results from the application to a post tensioned concrete box girder bridge Andrea Meoni a, *, Matteo Castellani a , Francesco Mariani a , Marco Ceccobelli a , Ilaria Venanzi a , Filippo Ubertini a a Department of Civil and Enviromental Engineering, University of Perugia, Via G. Duranti 93, 06125 Perugia (PG), Italy Abstract The Guidelines for Risk Classification and Management, Safety Assessment, and Monitoring of Existing Bridges, released in 2020 by the Italian Ministry of Infrastructures and Transport, promote the use of Building Information Modeling (BIM) environment as one of the main tools to achieve informed risk mitigation of infrastructures. To date, BIM environments are often only used as repositories of information from different entities involved in bridge design, management, and retrofitting. However, to achieve informed risk mitigation of bridges, the current BIM paradigms should be expanded by integrating new typologies of data, such as those from Structural Health Monitoring (SHM) systems, and functionalities, such as the ability to process stored data directly within the BIM environments. In this light, this paper presents a BIM-based approach conceived to achieve comprehensive assessment and management of the risk conditions and structural performance of bridges. To this end, specific algorithms have been implemented into the selected BIM environment to perform tasks such as assessing the preliminary risk level of a bridge according to the methodology proposed by the Italian Guidelines, evaluating the outcomes from visual inspections to estimate its defect level, storing and processing data from SHM systems installed on the structure, and more. The paper also presents the first application of the proposed BIM-based approach to a post-tensioned box girder bridge belonging to the Italian road network. Keywords: Bridges; Building Information Modeling; Risk assessment; Structural Health Monitoring; Data management; Data processing. © 2024 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Scientific Board Members

Keywords: Bridges; Building Information Modeling; Risk assessment; Structural Health Monitoring; Data management; Data processing.

* Corresponding author. Tel.: +39-075-5853996 E-mail address: andrea.meoni@unipg.it

2452-3216 © 2024 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Scientific Board Members 10.1016/j.prostr.2024.09.018 2452-3216 © 2024 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license ( https://creativecommons.org/licenses/by-nc-nd/4. 0 ) Peer-review under responsibility of Scientific Board Member s 2452-3216 © 2024 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license ( https://creativecommons.org/licenses/by-nc-nd/4. 0 ) Peer-review under responsibility of Scientific Board Member s * Corresponding author. Tel.: +39-075-5853996 E-mail address: andrea.meoni@unipg.it

Andrea Meoni et al. / Procedia Structural Integrity 62 (2024) 73–80 Meoni et al/ Structural Integrity Procedia 00 (2019) 000 – 000

74 2

1. Introduction Bridge management is a challenging task on the part of management authorities and bridge owners, yet of paramount importance to ensure the proper operation and level of safety of the infrastructure network. In 2020, in Italy, the Italian Ministry of Infrastructures and Transport released the Guidelines for Risk Classification and Management, Safety Assessment, and Monitoring of Existing Bridges (hereinafter referred to as Italian Guidelines), to provide specifications to standardize the management of bridges, starting from the assessment of their risk and structural defective conditions to the evaluation of their structural performance, the implementation of monitoring systems, and more (Consiglio Superiore dei Lavori Pubblici 2020). The Italian Guidelines also promote the use of Building Information Modeling (BIM) environments as a key tool for bridge management, especially in the case of bridges of strategic importance for the road network or characterized by critical structural conditions. BIM platforms ensure high interoperability between different tasks and entities involved in bridge design, management, and retrofitting, yet, in practical applications, they are often used solely as repositories of information rather than active operating environments in which to process and analyze data from different sources, such as risk assessments and Structural Health Monitoring (SHM) systems. In recent years, several methodologies and approaches have been proposed by the scientific community to implement new functionalities for the management of bridges and other structures in BIM environments (Delgado et al 2018, Ciccone et al 2022, Deng et al 2022, Li et al 2022, Meoni et al 2022). In this context, this paper presents the first results regarding a new BIM-based approach conceived to achieve comprehensive assessment and management of the risk conditions and structural performance of bridges. The paper also reports a practical application of the proposed framework on a post-tensioned concrete box girder bridge included in the Italian road network to demonstrate its feasibility and potentialities. The rest of the paper is organized as follows: Section 2 outlines the proposed BIM-based approach, while Section 3 presents the application case study and the methodologies adopted to perform risk and structural performance assessment. Section 4 illustrates the obtained results; hence Section 5 concludes the paper with final comments and remarks. 2. Informed bridge management The proposed approach defines a framework for informed management of bridges based on the assessment of their risk conditions and structural performance. Its implementation into the BIM environment ensures high interoperability between the diverse analyses included in the framework. Figure 1 schematically illustrates the proposed approach. Once the digital 3D model of the selected bridge is constructed, data to perform risk and structural performance evaluations can be gathered from different sources, such as technical drawings and reports, site inspections, SHM systems, and more, hence stored/rendered accessible within the adopted BIM platform. Risk evaluations can be performed by following the provisions of a selected standard for bridge risk assessment, the latter integrated into the BIM environment via specific programming codes, then processing the required information by selecting it directly from the dataset constructed into the BIM platform. Concurrently, at a first level of knowledge, structural performance evaluations can be carried out by assessing the extent of the structural defects detected on the bridge under assessment during its visual inspection. In this regard, outcomes from laser scanner and photogrammetric surveys, possibly taken at each consecutive visual inspection, can be also stored into the BIM platform to create visual time histories containing information regarding the onset and progression of the structural damage revealed on the bridge at every inspection. More refined structural performance evaluations can be obtained when the bridge is equipped with tailored SHM systems. In this case, a digital replica of the monitoring system, for instance including information on sensor placement and connectivity, datasheets, and more, can be represented in the 3D model of the bridge. Then, according to the monitored features (i.e., strains, displacements, accelerations, tilts, and more), specific algorithms can be implemented within the adopted BIM platform by using programming codes, thus allowing SHM data processing and result visualization. The outcomes from the proposed approach contribute to the informed management of the bridge under evaluation through (i) understanding the risk conditions to which the asset is exposed and to what extent, (ii) detailed monitoring of its structural defectiveness, and (iii) assessment of its structural performance under operational conditions.

Andrea Meoni et al. / Procedia Structural Integrity 62 (2024) 73–80 Meoni et al/ Structural Integrity Procedia 00 (2019) 000 – 000

75

3

Fig. 1. Flowchart of the proposed approach for informed bridge management in BIM environments.

3. Methodology 3.1. Description of the case study bridge

The case study bridge, shown in Fig. 1(a) and located on the central Italy road network, consists of a ten-span post-tensioned box girder constructed in 1984 by using the balanced cantilever method. The structure has a length of 630 m, with a maximum span length of 70 m. Piers and abutments are made of cast-in-place reinforced concrete. The height of the piers varies from approximately 7.5 m to a maximum of 18.5 m depending on the slope of the ground. The bridge exhibits a continuous-beam static scheme, with vertically prestressed internal joints ensuring the continuity between the cantilever beams at spans no. 3, 5, 7, and 9 (see Fig. 1(b)). In particular, the upper and lower segments of every cantilever beam are connected by means of two systems of longitudinal sliding supports and 30 post-tensioned vertical Dywidag bars, these last embedded into the structural joints at the end of the construction phase. With this configuration, structural joints allow the transferring of shear stresses and bending moments between the connected cantilever beams, as well as thermal dilatation (Dywidag bars are located inside plastic pipes that allow for their movement along the longitudinal direction of the bridge). The post-tensioned box girder is connected to the abutments by means of longitudinal sliding supports, while hinges and longitudinal sliding supports alternatively ensure the connection between the box girder and the piers. The bridge prestressing system also includes post-tensioning tendons deployed at different positions of the box girder. The digital 3D model of the case study bridge was constructed by using the BIM design platform Revit 2022 by Autodesk (Seidler 2021) based on information from the original design drawings and photogrammetric surveys carried out by the Authors.

Fig. 2. Case study bridge: (a) aerial view; (b) detail of a structural joint.

3.2. Risk evaluation The Italian Guidelines were selected as a reference standard to assess the risk conditions of the case study bridge (Santarsiero et al 2021, Meoni et al 2023). The assessment procedures to evaluate the Structural-Foundational Class of Attention (CoA), Seismic CoA, Landslides CoA, and Hydraulic CoA of bridges (i.e., structural, seismic, landslides, and hydraulic risks) were implemented into the selected BIM platform by using custom Python scripts. Data required for the assessment were collected in Excel worksheets and then processed directly into the BIM environment.

Andrea Meoni et al. / Procedia Structural Integrity 62 (2024) 73–80 Meoni et al/ Structural Integrity Procedia 00 (2019) 000 – 000

76 4

3.3. Structural performance evaluation

The case study bridge was recently visually inspected by some of the Authors to assess its structural defectiveness. Given the height of the piers, a by-bridge platform was adopted to properly inspect the bridge superstructure. Defects revealed during the inspection were noted in digital worksheets and then uploaded to the adopted BIM platform. Subsequently, the defect level of the case study bridge was evaluated according to the provisions of the Italian Guidelines through the implementation of the evaluation procedure in the BIM platform by using customized Python scripts. A photogrammetric survey of the case study bridge was also carried out during visual inspection. To do that, a professional drone, model DJI Mavic 3 Enterprise series, equipped with a 20MP camera and a 4/3 CMOS sensor, was used to scan the structure from different points of view. The resulting point cloud was first cleaned via the Metashape Pro software, then imported into the adopted BIM platform and overlaid on the digital 3D model of the bridge. The point cloud can then be used as a reference for future defect monitoring activities. Ambient Vibration Tests (AVTs) were carried out on the case study bridge on October 19th, 2022, hence Operational Modal Analysis (OMA) was used to determine the modal features characterizing the dynamic response of the structure. These can be considered as reference modal features for future SHM activities. In fact, as it is well known, possible changes in modal features over time can be indicative of modifications in the global structural performance of a structure due to the occurrence of damage (Brincker et al 2001). Figure 3 illustrates the measurement setup adopted to test span no. 8 and half of span no. 7 and 9. Each measurement station consisted of a high-sensitivity (10 V/g) uniaxial accelerometer, model PCB393B12, oriented along the z-axis by using stabilizing steel support deployed on the concrete curbs of the deck of the bridge. The data acquisition system was a NI cDAQ 9188, equipped with four NI 9234 modules for dynamic signal acquisition. The duration of each acceleration measurement was 35 minutes, while the sampling rate was set to 1653 Hz. Acceleration signals were detrended and resampled at 40 Hz before using the Frequency Domain Decomposition (FDD) technique to extract the modal features of the bridge. Processing operations were carried out directly within the BIM platform by using custom Python scripts to implement the dynamic identification algorithm.

Fig. 3. Measurement setup adopted to perform ambient vibration tests on the case study bridge.

4. Results The BIM-based approach for informed management of bridges proposed in Section 2 was implemented in the Revit BIM platform through the development of “BridgeBIM”, a n add-in application based on custom Python scripts, powered by the pyRevit plugin (Iran-Nejad 2020), and accessible via the Revit ribbon panel. Figure 4 shows the digital 3D model of the case study bridge together with the main panel of BridgeBIM. 4.1. Module for risk evaluation Figure 5(a) illustrates the main panel of the Risk Evaluation Module. The extent of every risk condition affecting the bridge under evaluation was assessed by considering the five risk levels provided by the Italian Guidelines, namely high, medium-high, medium, medium-low, and low risk. Figures 5(b) and 5(c) show exemplifications of a worksheet collecting defects detected on a structural element of the case study bridge during the performed visual inspection (defect level is a key parameter for the evaluation of structural and seismic risk conditions) and of a worksheet gathering information required in the assessment procedures of the diverse risk conditions encompassed