Automated Crack Detection for Underwater Inspection of Miter Gates with Unmanned Underwater Vehicle

Abstract

The emergence of Unmanned Underwater Vehicles (UUVs) as tools for underwater inspection tasks presents promising potential. This study presents a novel automatic damage detection framework in large-scale underwater structures using a physics-based graphics model (PBGM). A high-fidelity finite element model of the Greenup Miter gate on the Ohio River is utilized to provide the fundamental information for the graphical model development. An open-source software is used as a graphic-based observational system to render images given different inspection distances and environmental conditions, such as light conditions and water quality. A deep neural network for crack detection with segmentation from the existing literature is adopted and then trained using transfer learning, adapting it to the unique conditions of underwater circumstances. Results indicate that the proposed method provides highaccuracy damage detection amidst the unique background noise and uncertainties presented in the underwater environment, contributing significantly to the field of UUV inspection of large-scale structures.