Graduation Semester and Year
Fall 2026
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Civil Engineering
Department
Civil Engineering
First Advisor
Dr. Nur Yazdani
Second Advisor
Dr. Himan Hojat Jalali
Third Advisor
Dr. Warda Ashraf
Fourth Advisor
Dr. Shouyi Wang
Abstract
Reinforced concrete bridge decks require reliable assessment of reinforcement location, concrete cover, and corrosion to support maintenance decisions. Ground-penetrating radar (GPR) enables rapid subsurface investigation, but manual interpretation is time consuming and dependent on expert judgment. Automated assessment is further complicated by variations in reflection geometry, acquisition conditions, and concrete properties. The primary objective of this dissertation is to develop and evaluate data-driven methods for annotation-efficient rebar detection, adaptive cover estimation, and continuous corrosion mass-loss prediction.
The research comprises three related studies. The first developed a rebar-detection framework combining representative image selection, model-assisted label propagation, and targeted human review. A dataset of 37608 B-scan images from 18 reinforced concrete structures supported model development, while three additional bridges were reserved for independent evaluation. The framework reduced manual annotation effort by 85.48% and achieved a bridge-averaged F1-score of 0.9868 on the excluded bridges. The second study integrated automated detection with hyperbola fitting and machine-learning-guided configuration selection for concrete cover estimation. Measurements from 80 laboratory specimens yielded 2976 fitting-feasible observations. A library of 1176 fitting configurations was reduced to 192 configurations, and regression models guided selection for individual reflections. The best-performing model achieved a mean absolute cover error of 1.777 mm (0.070 in.), with 96.64% of test estimates having absolute errors below 5 mm (0.197 in.). The third study investigated continuous corrosion mass-loss prediction using 3199 GPR A-scan observations, material and geometric information, and physically motivated waveform descriptors. Eight regression models were evaluated using nested cross-validation against gravimetrically measured steel loss. The best-performing model achieved a coefficient of determination of 0.924 and a mean absolute error of 0.668 percentage points of mass loss. The findings demonstrate reduced annotation requirements, improved cover estimation through adaptive fitting, and the feasibility of continuous corrosion prediction under controlled experimental conditions.
Keywords
Ground Penetrating Radar, Machine Learning, Corrosion, Rebar mass loss, Object detection, Reinforcement cover, Dataset diversity
Disciplines
Civil Engineering | Structural Engineering
License

This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Recommended Citation
Pal, Avijit, "AI-DRIVEN ESTIMATION OF CONCRETE COVER DEPTH AND REINFORCEMENT CORROSION MASS LOSS IN BRIDGE DECKS USING GROUND-PENETRATING RADAR" (2026). Civil Engineering Dissertations. 16.
https://mavmatrix.uta.edu/civilengineering_dissertations2/16