Bolt looseness detection based on deep learning and image registration

DOI:	10.11809/bqzbgcxb2023.07.002
Keywords:	bolt loosening; image registration; deep learning; object detection
Abstract:	Aiming at the problem that the repeated vibration of a wind turbine for a long time may cause its own bolt loosening and affect its normal operation, this paper proposes a bolt loosening detection method based on deep learning and image registration to avoid manual intervention. YOLOv5s is improved, and ShuffleNetV2 is used as Backbone to realize the location of the bolt area to be detected in the image, in which process the detection speed increases by 69.8% without reducing the accuracy. After that, the to be detected image and the template image are registered through an intensity based method, and the registered grayscale images are processed by means of feature enhancement, thereby locating the loose bolts. According to the experimental results, the method can accurately detect the looseness of the bolt when the bolt rotates more than 5°.
Published:	2023-07-28
PDF HTML