Robotic Inspection Using Learning Based Impact Echo For Civil Infrastructure

<p>Concrete infrastructure often develops a variety of internal flaws that cannot be detected through visual inspection alone, and must be regularly inspected with other methods to maintain structural integrity. Impact echo (IE) is an acoustic non-destructive evaluation (NDE) technique essential for assessing the structural integrity of concrete structures, such as bridge decks. IE data interpretation has traditionally been manual, while automated frequency analysis methods, though faster, require expertise and careful calibration. This led to a growing shift towards learning-based methods for the autonomous interpretation of IE signals. However, learning-based IE methods lack generalization and robustness. These methods are highly sensitive to changes in experimental parameters, such as geometry of the structures, defect characterization, sensor configuration, data acquisition settings, noise and non-ideal data. Additionally, it has been demonstrated through previous research and experience that relying solely on a single NDE method can be insufficient in providing a comprehensive evaluation of the structure's condition. In addition, manual NDE data collection can be labor-intensive for on-site engineers.</p> <p>Therefore, in this thesis we propose a concrete inspection system that consists of five main components. First, we present three robotic-based inspection systems for efficient NDE data collection, and 2D defect mapping of shallow and deep surface defects. Each robot integrates visual-inertial positioning system, and tags collected NDE data with pose information. The first robot employs a microphone for impact sounding (IS) assessment, the second utilizes an accelerometer-based impact echo (IE) sensor, and the third incorporates a ground penetrating radar (GPR) antenna. Second, we present an unsupervised learning IS analysis system for autonomous detection and characterization of shallow concrete defects. Third, a learning-based IE defect detection and mapping coupled with GPR for comprehensive evaluation of concrete structures. Fourth, a supervised and self-supervised learning IE method to bridge the generalization gap of learning-based IE for defect mapping. Finally, we present an uncertainty quantification system for learning-based IE which further enhances the quality of the inspection.</p> <p>In this thesis, we review previous work and inspection method solutions with a focus on IE, present and compare our results with other studies, showcase the proposed solutions and their improvements, and finally outline future work. The developed methods have the potential to significantly impact the field of NDE by increasing the efficiency and quality of concrete structures inspection.</p>

Name thesis:degree_name

Doctor of Philosophy (Ph.D.)

Level thesis:degree_level

Dissertation

Year dc:date.available

2025

Author dc:creator

• Hoxha, Ejup

Contributors dc:contributor

• Jizhong Xiao

• Robotics
• Impact Echo
• Artificial Intelligence
• Artificial Intelligence and Robotics
• Civil Engineering
• Signal Processing

Repository record dc:identifier

https://academicworks.cuny.edu/cc_etds_theses/1255

OAI identifier oai:identifier

oai:academicworks.cuny.edu:cc_etds_theses-2363