Date of Award

2019

Degree Type

Dissertation

Degree Name

Doctor of Philosophy in Civil and Environmental Engineering

Department

Civil and Environmental Engineering

First Advisor

Mayrai Gindy

Abstract

Statistics by the Federal Highway Administration (FHWA)’s report indicates that 11% of the bridges in the United States are rated as "structurally deficient" and over 30% of existing bridges have exceeded their 50-year design life, meaning that condition assessment and repair programs will require substantial budget in the near future. Current observation-based bridge inspection techniques consist largely of time consuming and subjective measures for quantifying deterioration of bridges. During bridge inspection, simplistic methods for assessing deterioration in concrete bridge decks are only capable of detecting deterioration in its moderate to severe stages. To provide a more thorough assessment of deterioration in concrete bridge decks, advanced technologies should be incorporated into bridge inspection. Some advanced nondestructive testing methods such as Ground Penetrating Radar (GPR), are being implemented that provide sub-surface information. GPR has been successfully used in a wide range of applications. Using advanced technologies like ground penetrating radar (GPR), deterioration hidden from the naked eye or missed using traditional methods, like Chain Dragging and Hammering Sounding, can be more accurately detected.

Automatic rebar detection in GPR data is the basic step in an automatic system for GPR-based condition evaluation of bridge decks. Achieving real-time performance on Acorn Reduced Instruction Set Computing Machine (ARM) based platforms for onsite applications still remains a challenge. Development an accurate and cost-effective system for real-time onsite rebar detection in GPR images is goal of this study. The authors proposed a novel computer vision-based method for automatic detection of rebars in complex GPR images in highly deteriorated concrete bridge decks. Extensive experiment performed to develop a reference for selecting a deep learning-based detection architecture that provides the right accuracy, speed, and memory usage balance for real-time detection of rebars on the latest version of ARM-based platforms. A deep learning-based detector is presented that can be deployed on the latest version of ARM-based platforms. State of the art results is obtained on GPRDETN detection task by implementing rebar detection model using Faster R-CNN with ResNet 101 CNN backbone.

Available for download on Friday, December 17, 2021

Share

COinS