PSI - Issue 64

Francesco Pentassuglia et al. / Procedia Structural Integrity 64 (2024) 254–261 F. Pentassuglia et al./ Structural Integrity Procedia 00 (2019) 000 – 000

258

5

Element (FE) model updating is a critical aspect where the parameters influencing the deflection of the bridge within the FE model are automatically adjusted. These parameters, such as the loss of prestressing in tendons, are fine-tuned to ensure a good match between the computed deflections and the deflections measured on-site. This alignment is essential for closely determining the damage state of the bridge based on its deflected shape. Once the relevant parameters are identified, the gradient, which quantifies the variance between the predicted behaviour of the FE model and the actual behaviour observed in the real structure, is carefully evaluated. If this assessment reveals a significant deviation, indicating a substantial difference between the model predictions and real-world observations, an optimisation process is initiated. This optimisation process automatically refines the parameters of the FE model to bring its deflected shape into closer agreement with the actual measurements. To manage this optimisation process effectively, an optimisation function, such as those implemented within MATLAB, should be employed. Previous research efforts (Ferrari et al., 2019; Yang and Wang, 2022) have typically focused on identifying the global minimum of the objective function or determining the point with the minimum value among the maximum errors on natural frequencies across various local minima (known as the "minimax" criterion). However, this paper proposes a novel approach by suggesting the use of deflections rather than natural frequencies in the optimisation process. The optimisation process concludes once an optimal solution is reached, and the behaviour of the updated FE model aligns closely with that of the real structure under investigation. The process ends by validating the efficiency and capacity of the k-NN algorithm in classifying bridges into different damage states. This validation entails comparing the results (i.e., the damage state prediction) obtained from the FE model updated through the optimisation process with the results derived from the k-NN algorithm. If the machine learning algorithm demonstrates comparable performance to the FE model updating in accurately identifying the damage state of the bridge asset, it validates the k-NN methodology as a reliable approach for bridge damage classification. In cases where the results are not satisfactory, further training or exploration of more advanced ML methods might be needed. This could involve investigating alternative techniques such as the General Regression Neural Network (GRNN), point cloud processing networks, or convolutional neural networks to enhance the accuracy and efficiency of classifying damage states in bridges. This iterative process of validation and potential refinement aims to optimise the methodology for robust and accurate bridge damage state assessment using ML techniques. Lastly, for each bridge type, drift-based fragility functions are built by exploiting deflections and bridge spans. Fragility functions play a crucial role as valuable tools for risk assessment and decision-making in bridge engineering. Fragility functions are graphical representations that illustrate the likelihood of a structure reaching or surpassing a specific level of damage or performance limit given a certain level of maximum vertical drift (Nazri, 2018) . Overall, the convergence of cutting-edge technologies, Μ L algorithms, and sophisticated methodologies represents a pivotal moment in the evolution of bridge engineering and asset management practices. By leveraging the capabilities of these advanced tools, there exists a profound opportunity to revolutionize the way in which bridges are designed, monitored, and maintained. This transformative approach not only enhances the safety and structural integrity of concrete bridges on a global level but also contributes significantly to the overall sustainability and efficiency of critical infrastructure systems. The integration of advanced technologies such as ML algorithms enables engineers and decision-makers to harness vast amounts of data and extract valuable insights to make informed decisions regarding bridge health and maintenance strategies. By utilising these tools, the industry can proactively address potential issues, optimise maintenance schedules, and extend the lifespan of bridges, ultimately leading to cost savings and improved operational performance. Furthermore, the adoption of these innovative methodologies not only enhances the safety and resilience of concrete bridges but also plays a crucial role in ensuring the long-term sustainability of infrastructure networks. By embracing these advancements, stakeholders can effectively manage risks, prioritise maintenance activities, and enhance the overall efficiency of bridge management practices, thereby contributing to the longevity and reliability of critical infrastructure systems worldwide.