Study of the bearing systems of existing bridges for structural monitoring

This thesis delves into the bearing systems of existing bridges to enhance monitoring techniques. It covers theoretical and practical aspects, focusing on the types, maintenance, and damage analysis of bridge bearings. Extensive data collection and analysis demonstrate common challenges and propose solutions using advanced technologies such as computer vision and deep learning algorithms. Bearings allow controlled movements like expansion, contraction, and rotation due to various forces acting on the bridge. These components enable different bridge parts to work together, resisting various forces while maintaining alignment. Over time, bearings wear out due to environmental factors like moisture, temperature changes, and chemical reactions, as well as mechanical fatigue from traffic loads. Proper maintenance, including regular inspections, cleaning, lubrication, and replacement of damaged ones, is vital for their functionality. The research compiled a database of over 8,200 bridge bearings from inspection reports across Italy, covering data from 2021, 2022, and 2023. This extensive dataset allows for a detailed analysis of damage types and severity over time. The study identifies common damages in bridge bearings, particularly in steel bearings, which showed high damage rates. Notable findings include 100% damage rates in double pendulum bearings and 74.4% in single pendulum bearings in 2023, highlighting the need for targeted maintenance strategies. Imaging technology has paved the way for computer vision techniques, though human visual inspection is still crucial for assessing bearings. Limited attempts have been made to use computer vision for automatic condition assessment of bridge bearings. This research aims to develop an intelligent Structural Health Monitoring (SHM) system for steel bridge bearings, particularly pendulum steel bearings and steel connections. The proposed SHM system utilizes Deep Learning (DL), specifically Convolutional Neural Networks (CNNs), for image-based condition assessment, with the model trained to detect bearings and assess their condition from images. The practical application of computer vision models involves creating a dataset of more than 3600 annotated images showing defected conditions of steel bearings. The CNN models are trained to perform component detection and damage classification. Initially, the model identifies the presence and typology of bearings in collected images, then classifies the detected bearings and type of damage, providing a quantitative measure of structural health. Techniques such as object detection and instance segmentation are used for the SHM system's implementation. In conclusion, the thesis provides insights into the types, maintenance, and damage analysis of bridge bearings, underscoring the need for advanced monitoring techniques. Integrating AI and deep learning in structural health monitoring presents a promising path for enhancing the safety and performance of bridge structures. This research significantly contributes to bridge engineering, offering practical solutions for monitoring infrastructures.