Detection and quantification of cracks in concrete bridges using drone-image inspection
Time: Fri 2024-06-14 10.00
Location: M108, Brinellvägen 23, Stockholm
Video link: https://kth-se.zoom.us/j/62274505390
Language: English
Subject area: Civil and Architectural Engineering, Structural Engineering and Bridges
Doctoral student: Juan Camilo Avendaño , Bro- och stålbyggnad
Opponent: Associate Professor Jelena Ninic, University of Birmingham, UK
Supervisor: Docent John Leander, Bro- och stålbyggnad; Professor Raid Karoumi, Bro- och stålbyggnad
QC 20240521
Abstract
The assessment of civil infrastructure plays an important role in ensuring the safety of the general public and the durability of structures. Traditional inspection methods often involve manual labour and subjective evaluations, resulting in limitations in efficiency and accuracy. In recent years, there has been an increasing interest on using advanced technologies, such as unmanned aerial vehicles (UAVs), image analysis and machine learning techniques, to establish them as alternatives for the inspection process. These techniques provide different advantages compared with the manual method in terms of time, objectivity, and safety. The results of these techniques can allow the engineers in charge of the assessment and maintenance planning to obtain detailed results that can improve their efficiency but they are not without challenges. This research project aims to evaluate different methods used for damage detection and quantification and their integration with UAVs as an alternative to structural inspections. The proposed methodology combines image analysis techniques, Convolutional Neural Networks (CNNs) with drones to address the different aspects of inspections, from the data gathering to the damage detection and quantification. This methodology focuses on detecting and quantifying small cracks as narrow as 0.1 mm on concrete structures, aiming to achieve results comparable to those of traditional inspection. Furthermore, an application demonstrating the feasibility of the methodology in inside environments is also presented, focusing on the inspection of the internal section of a box girder bridge, including the creation of 3Dphotogrammetrical models to improve the inspection process. The results of the thesis highlight the methodology’s capabilities in detecting small cracks with a high probability and the possibility to use it for inside inspections without ideal conditions. This demonstrates the potential of integrating the different methods to transform structural inspections toward more efficient methodologies. Furthermore, the analysis of the results evaluates the drawbacks encountered and outlines future research directions aimed at advancing image-based inspections and their practical application.