Date of Award
January 2025
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Civil Engineering
First Advisor
Sattar Dorafshan
Abstract
Ensuring the safety and functionality of infrastructure is crucial for maintaining public welfare and economic stability. Traditional inspection methods are often labor-intensive and time-consuming, leading the engineering field to seek advancements in non-destructive techniques to improve inspection speed, accuracy, and safety. Artificial intelligence (AI) has emerged as a key tool in recent studies for interpreting nondestructive (ND) tests, as manual classification is often infeasible due to the complexity and volume of data.
In this research, AI was integrated with various ND technologies to detect defects, such as carbonation in concrete and steel or autonomous measurement in infrastructure. The first objective of this dissertation combines impact echo (IE) and Ground Penetrating Radar (GPR) with AI, to identify various subsurface defects in reinforced concrete bridge decks. Additionally, a feature-selection-based Support Vector Machine (SVM) model was developed using the physics of IE signals to detect defect areas in slabs, and its results were compared with other deep learning algorithms, including 1D-CNN and 2D-CNN as first objective of my dissertation.
Applying computer science and machine learning techniques to drone-acquired data supports autonomous defect detection and condition assessments, significantly reducing time and labor. The second objective explores the use of visual data for inspecting ancillary structures and detecting defects, such as cracked or defective bolts. Building on promising results from previous studies, this research proposes comprehensive algorithms to further enhance inspection capabilities. The role of infrared thermography is also discussed, particularly for its effectiveness in enhancing deep learning performance by removing background noise. Combining thermal images with visual data (data fusion) significantly improves the accuracy of models in locating ancillary structures and identifying defects. Data fusion of infrared and visual images shows potential in non-contact methods for detecting issues like loosened or missing bolts.
In the third objective, hyperspectral imaging (HSI) is introduced as a novel non-destructive tool for detecting carbonation defects in concrete structures, eliminating the need for traditional techniques like XRD, SEM, or phenolphthalein solution. This study uses hyperspectral cameras to detect carbonation defects on concrete surfaces, with machine learning techniques and image processing approaches applied to detect damaged areas following chemical exposure.
Finally, this dissertation presents an AI‑driven photogrammetric workflow that leverages UAS imagery to measure stockpile volumes and monitor construction progress. Deep‑learning point‑cloud classification and change‑detection algorithms achieve 5 % volume error across multiple flight altitudes and capture layer‑by‑layer pavement thickness changes in near real time, providing contractors with actionable information for material management and schedule control.
This dissertation advances infrastructure monitoring in the U.S. by integrating artificial intelligence with non-destructive technologies such as Impact Echo, GPR, UAS-based imaging, and Hyperspectral Imaging with using AI. These contributions enable faster, safer, and more accurate inspections, enhancing the reliability and resilience of critical infrastructure systems.
Recommended Citation
Jafari, Faezeh, "AI-Enhanced Nondestructive Testing For Defect Evaluation And Performance Monitoring In Infrastructure Systems" (2025). Theses and Dissertations. 7515.
https://commons.und.edu/theses/7515