College of Graduate Studies: Theses & Dissertations

Term of Award

Spring 2026

Degree Name

Master of Science, Mechanical Engineering

Document Type and Release Option

Thesis (open access)

Copyright Statement / License for Reuse

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Department

Department of Mechanical Engineering

Committee Chair

Jinki Kim

Committee Member 1

Doyun Lee

Committee Member 2

Bishal Silwal

Abstract

This thesis presents an autonomous mobile robotic system that integrates real-time visual crack detection with laser-based three-dimensional profiling for building inspection. The system coordinates SLAM-based navigation, deep learning perception, and manipulator-guided measurement within a unified ROS 2 framework, enabling scan-on-demand workflows where laser profiling occurs selectively at visually detected defect locations. The detection pipeline employs a two-stage architecture combining U-Net semantic segmentation with Pix2Pix residual refinement, achieving 73.9% mean intersection-over-union and 76.4% F1-score on the CrackSeg9k benchmark while maintaining 16 frames-per-second real-time throughput. Real-world validation demonstrated 95% detection success on 80 campus images under challenging conditions. The profiling subsystem utilizes an LMI Gocator 2618 laser scanner mounted on a Yaskawa HC10 manipulator to measure crack geometry, with width measurements agreeing with ground truth within 0.04 mm mean absolute error on three laboratory specimens. However, the HC10 controller's three-phase power requirement prevented field profiling deployment, limiting validation to controlled laboratory conditions. The research establishes methodologies for residual learning-based segmentation refinement, scan-on-demand profiling workflows, and autonomous integration of navigation, detection, and measurement subsystems for structural inspection applications.

Research Data and Supplementary Material

No

Download

Share

COinS