Date of Submission

7-2024

Document Type

Thesis

Degree Name

Master of Science in Civil Engineering

Department

Civil and Environmental Engineering

Advisor

Dr. Reihaneh Samsami, Ph.D., P.E., CAPM

Committee Member

Dr. Byungik Chang, Ph.D., P.E., M.B.A.

Committee Member

Dr. Goli Nossoni, Ph.D.

Committee Member

Dr. Saida Elmi, Ph.D.

Keywords

Concrete Bridge Deck, Unmanned Aerial Systems (UAS), Machine Learning, Operation and Maintenance, Inspection Specifications

LCSH

Bridges--Floors, Drone aircraft, Machine learning, Bridges--Maintenance and repair

Abstract

Many national infrastructures, particularly bridges, are becoming older and older day by day. Their safe operation and maintenance require routine monitoring and inspection. There are lots of bridge inspection specifications and manuals to guide inspectors to give the proper recommendations about the bridge structure and safety. These manuals and specifications require bridge inspectors to manually inspect the bridge and recommend maintenance and repair tasks referring to the specific criteria available in the manuals. In time, the application of technology makes many Unmanned Aerial Systems (UAS) advancements during the phase of data collection. For the image processing and analysis of collected data using UAS, different machine-learning approaches have been developed. In this thesis, two bridge damage detection models are created using Convolutional Neural Network (CNN) and Vision Transformer (ViT) aiming to improve efficiency and maintain the structural safety of bridges. The test data shows that both models have an accuracy of more than 92% in recognizing damages on a bridge deck with ViT performing slightly better than the CNN model. It is concluded that the illustrated methodology not only improves the structural safety of the infrastructure and maintenance, but also can significantly reduce the misjudgment by inspection personnel and financial cost for these inspections.

Download

Share

COinS