Autonomous unmanned aerial vehicles and deep learning-based damage detection

dc.contributor.authorKang, DongHo
dc.contributor.examiningcommitteeKavgic, Miroslava (Civil Engineering) Ho, Carl (Electrical and Computer Engineering Li, Jian (Civil, Environmental & Architectural Engineering, University of Kansas)en_US
dc.contributor.supervisorCha, Young-Jin (Civil Engineering)en_US
dc.date.accessioned2021-11-22T17:38:24Z
dc.date.available2021-11-22T17:38:24Z
dc.date.copyright2021-11-18
dc.date.issued2021-11-18en_US
dc.date.submitted2021-11-18T23:13:18Zen_US
dc.degree.disciplineCivil Engineeringen_US
dc.degree.levelDoctor of Philosophy (Ph.D.)en_US
dc.description.abstractInfrastructure failure causes the loss of human lives and high socio-financial costs. Due to the continuous aging of infrastructure, a proper structural health monitoring (SHM) system is required to ensure the safety of structures and reduce repair costs through the early detection of structural damage. Existing visual inspection methods are not reliable due to the low frequency of inspection, subjective evaluation of structural damage, and vulnerability of inspectors’ safety, along with high costs. Traditional damage detection methods have similar limitations, since they require a large number of sensors to monitor large-scale infrastructure and involve high levels of uncertainty due to environmental noises and sensor malfunctions. Computer vison techniques have been implemented to overcome the limitations mentioned above, relying on image processing algorithms to extract damage-sensitive features. However, it is very difficult to extract a robust damage-sensitive feature. To resolve this limitation, I developed two deep learning-based damage detection methods using computer vision. The first method is a hybrid pixel-level crack segmentation and quantification method for complex cracks on rough scenes. The developed hybrid method provides robust damage detection for images, which addresses the uncertainties of traditional approaches. The second method is a real-time semantic transformer representation network (STRNet) for crack segmentation. The proposed STRNet can process 49 images per second with a mean intersection over union score of 92.6, which represents state-of-the-art performance in this area when it comes to accuracy. Using advanced deep learning methods and computer vision for damage detection still requires a great number of cameras to monitor large-scale infrastructure, which can be expensive. Consequently, another achievement of this thesis is that I developed an autonomous flight method using unmanned aerial vehicles (UAVs) for SHM purposes. Some critical parts of the bridge system, which should be monitored, are located beneath the bridge deck where global positioning system (GPS) signals are very weak or not available. Therefore, a three-dimensional pseudo map was developed using an inexpensive ultrasonic beacon system to replace the GPS signals for the autonomous flight of the UAVs.en_US
dc.description.noteFebruary 2022en_US
dc.identifier.citationKang, D., Benipal, S.S., Gopal, D.L. and Cha, Y.J., 2020. Hybrid pixel-level concrete crack segmentation and quantification across complex backgrounds using deep learning. Automation in Construction, 118, 103291.en_US
dc.identifier.citationKang, D. and Cha, Y.J., 2018. Autonomous UAVs for structural health monitoring using deep learning and an ultrasonic beacon system with geo‐tagging. Computer‐Aided Civil and Infrastructure Engineering, 33 (10), 885-902en_US
dc.identifier.citationKang, D. and Cha, Y.J., 2020. Efficient attention-based deep encoder and decoder for automatic crack segmentation, Structure health monitoring, Sage, (accepted)en_US
dc.identifier.urihttp://hdl.handle.net/1993/36120
dc.language.isoengen_US
dc.rightsopen accessen_US
dc.subjectDeep learningen_US
dc.subjectComputer visionen_US
dc.subjectStructural health monitoringen_US
dc.subjectUnmanned aerial vehicleen_US
dc.titleAutonomous unmanned aerial vehicles and deep learning-based damage detectionen_US
dc.typedoctoral thesisen_US
Files
Original bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
kang_dongho.pdf
Size:
4.9 MB
Format:
Adobe Portable Document Format
Description:
Main article
License bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
license.txt
Size:
2.2 KB
Format:
Item-specific license agreed to upon submission
Description: