The loosening and detachment of high-strength bolts in steel bridges have become a hot topic in the steel structure industry. The existing manual inspection methods are laborious, time-consuming, and inefficient. Machine vision is used for detecting loose high-strength bolts in steel bridges, which can greatly improve detection efficiency, reduce detection costs, and ensure detection accuracy. However, this method still needs improvement in terms of bolt recognition and detection reliability, and has not yet been applied in engineering practice. To meet the demand for detecting the loosening angle of high-strength bolts in steel bridges, a high-strength bolt loosening detection algorithm based on the Hourglass network and digital image processing has been proposed. The algorithm consists of two parts: feature point recognition and angle calculation. The recognition algorithm consists of two parts: the key point localization network and the key point coordinate generation module. The key point localization network is a ResNet-50 feature extraction network designed to generate heatmaps containing the key points of the bolts. The key point coordinate generation module, based on the heatmaps, uses the localization network to detect the coordinates of the key points. The loosening angle algorithm works as follows: after obtaining the key point heatmaps, the bolt key points are transformed into an envelope diagram using the K-Means clustering algorithm. Once the bolt envelope diagram is obtained, the six corner points are numbered sequentially in a counterclockwise manner. The angle between corner point 1 and the positive x-axis is taken as the current angle of the bolt. By comparing the current angles at different times, the loosening angle of the bolt can be determined. Multiple photos of high-strength bolt connections on highway steel bridges have been collected, with the outer six corner points and center point of the nuts labeled. The bolt dataset has been expanded by applying geometric transformations, color transformations, adding Coarse Dropout noise, color perturbations, and other operations. A model training was conducted using the TensorFlow deep learning framework, and the recognition performance of the model at different distances was tested. The recognition results of the network in this paper were compared with CPMs and Two-Stage networks. The tests showed that the network model in this paper had excellent recognition performance, high computational efficiency, and good robustness. The APCK and AACC were as high as 97.6% and 99.5%, respectively. The APCK, AACC, and detection speed were superior to CPMs and Two-Stage networks.
Cha Y J, You K, Choi W. Vision-based detection of loosened bolts using the Hough transform and support vector machines[J]. Automation in Construction, 2016, 71:181-188.
[5]
Ramana L, Choi W, Cha Y J. Fully automated vision-based loosened bolt detection using the Viola-Jones algorithm[J]. Structural Health Monitoring, 2019, 18(2):422-434.
[6]
Zhao X, Zhang Y, Wang N. Bolt loosening angle detection technology using deep learning[J]. Structural Control and Health Monitoring, 2019, 26(1), e2292.
[7]
Zhang Y, Sun X, Loh K J, et al. Autonomous bolt loosening detection using deep learning[J]. Structural Health Monitoring, 2020, 19(1):105-122.
[8]
Huynh T C. Vision-based autonomous bolt-looseness detection method for splice connections:design, lab-scale evaluation, and field application[J]. Automation in Construction, 2021, 124, 103591.
He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Chicago:2016.
[11]
Ioffe S, Szegedy C. Batch normalization:accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the International Conference on Machine Learning. Boston:2015.
[12]
Khan A H, Cao X, Li S, et al. BAS-ADAM:an ADAM based approach to improve the performance of beetle antennae search optimizer[J]. IEEE/CAA Journal of Automatica Sinica, 2020, 7(2):461-471.
[13]
Dong X, Yu S I, Weng X, et al. Supervision-by-registration:an unsupervised approach to improve the precision of facial landmark detectors[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City:2018.
[14]
Lyu J, Shao X, Xing J, et al. A deep regression architecture with two-stage re-initialization for high performance facial landmark detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu:2017.
Login
Register
DownLoad: