Computer Vision-Based Autonomous Method for Quantitative Detection of Loose Bolts in Bolted Connections of Steel Structures

Wulve Lao, Chuang Cui, Dengke Zhang, Qinghua Zhang, Yi Bao

Research output: Contribution to journalArticlepeer-review

11 Scopus citations

Abstract

In this study, an autonomous computer vision-based method is presented to quantitatively detect loose bolts. The method integrates keypoint detection via YOLOv5 and PIPNet, distortion correction via perspective transformation, and rotation angles quantification via geometric imaging. Distortion correction is incorporated to address skewed angles and improve the accuracy of rotation angles. A representative experiment on bolted connection of steel structures is conducted to evaluate the presented approach. The effects of the focal distance, skewed angle, and lighting conditions on the detection and quantification performance are evaluated by varying the imaging conditions. The results demonstrate that the presented approach automatically detects all bolts and their corners, irrespective of the imaging conditions. No false detection occurs, and the quantification errors are lower than 1°. The proposed method can be deployed for automatic detection of loose bolts and quantification of rotation angles in bolted connections under different imaging conditions.

Original languageEnglish
Article number8817058
JournalStructural Control and Health Monitoring
Volume2023
DOIs
StatePublished - 2023

Fingerprint

Dive into the research topics of 'Computer Vision-Based Autonomous Method for Quantitative Detection of Loose Bolts in Bolted Connections of Steel Structures'. Together they form a unique fingerprint.

Cite this