The US transportation network relies heavily on bridges, with over 40% being older than 50 years, as reported in the 2021 ASCE infrastructure report card. Traditional bridge monitoring is slow, labor-intensive, struggles with accessibility, and can be inconsistent, hindering proactive condition monitoring. To overcome these limitations, transportation agencies are investigating new technologies for inspections to improve access, reduce costs, and enhance safety. This research explores using drones and deep neural networks to create a system for detecting and tracking infrastructure damage. The study focuses on a bridge deck with patched patched spalling, using drones to capture images and deep learning models to identify and measure the extent of the damage over time. The study’s methodology goes beyond simply locating damage by quantifying changes in the affected surface area. Experiments on simulated damage showed that both U-Net and SAM can generate accurate distress maps, which are crucial for estimating the progression of patched spalling during timeline monitoring. The developed methods and methodology can be expanded and applied to monitor other types of infrastructure damage, enabling proactive maintenance.
Read
