Poster 22-01

UAS-Assisted Inspection of Bridges for Corrosion Effects

PI: Iris Tien

Co-PI(s): 

Institution(s): Georgia Institute of Technology

Abstract

New technologies, such as UAVs and machine learning, have been used to conduct imagery-based bridge inspections and efficiently evaluate damage on the bridge. However, corrosion detection is still an open problem, and corrosion detection algorithms have only proven adequate in certain environments and conditions. The main goal of this project is to explore the use of UAVs as a proof-of-concept to detect and characterize corrosion on bridges in Georgia. The first objective of this research is to investigate the use of UAVs for bridge inspections. The second objective of this research is to develop and evaluate automatic corrosion detection and evaluation algorithms. As part of this project, imagery data from two bridges, selected in collaboration with GDOT personnel, was collected using the Skydio 2+ drone. After the images were cleaned and labeled, they were used in varying computer vision and machine learning algorithms for corrosion detection. First, texture thresholding and color thresholding methods were implemented. Then, K-Means was investigated as an unsupervised machine learning algorithm using texture and color features. Finally, deep learning methods were investigated with automated featured extraction. It was found that the Skydio 2+ drone can be used to reach places that are hard for inspectors to reach, but the trees in Georgia limit where the drone can fly and how much of the bridge can be assessed. Therefore, pre-flight planning is recommended. Additionally, it was found that no corrosion segmentation technique as of yet works perfectly. K-Means segmentation, coupled with background removal, has the highest recall, but the mIoU is low. Texture thresholding performed with the second highest recall and an mIoU close to 90%. The deep learning techniques, SpotRust and FCN, have the lowest performance. This research shows that these methods must be developed further to be useful in supporting bridge inspections in Georgia.

Please comment below with any statements or questions you may have. Also let GTI if you would be interested webinars or presentations on similar topics.