Research On Detection Technology Of Bolt Looseness Of State Grid Tower Based On Machine Vision

High-voltage transmission towers are connected by bolts in large numbers.Although a small number of loose bolts will not endanger the safety of the transmission tower itself,after many transmission lines galloping,the number of loose bolts gradually increases,and the overall load-bearing capacity of the tower decreases.Tower risk increases accordingly.Under normal circumstances,maintenance personnel regularly climb the tower to overhaul the bolts one by one,but this method requires high skills for the maintenance personnel and is time-consuming and labor-intensive.Therefore,the use of a new type of bolt looseness detection technology is of great significance to improve the safety of the tower and reduce the maintenance cost.In order to reduce the inspection cost,this paper proposes to use machine vision method to detect the loosening of the tower bolts,but the visual inspection of the tower bolts is faced with the following difficulties.First of all the position of the lens cannot be completely fixed.Secondly the rotation angle of the loose bolts relative to the tightening is small.According to field experience,the angle is usually between10°～30°.On the other hand the po sture of each bolt is different in the pre-tightened state and the number of bolts is huge.In view of the above difficulties,this paper proposes to use the Scale-Invariant Feature Transformation algorithm(SIFT algorithm)as the core image registration technology to process the reference image and the floating image,and perform an image registration on the images taken under different lens positions.The reference image is the image taken when the bolts are fully tightened,and the floating image is the image taken when part of the bolts are loosened.In order to improve the effect of image registration,this paper proposes a bolt center matching algorithm to further correct the bolt position,and finally judge whether the bolt is loose according to the difference image generated by the reference image and the loose bolt image after registration.In order to be able to quickly output the accurate position of the loose bolt in the huge inspection work,this paper introduces the target detection technology in deep learning between the two image registrations.The Faster R-CNN target detection algorithm is used to train the target detector to identify the position of the bolt in the image,and the target detection result is applied to the difference image,which simplifies the determination process of the loose bolt.A series of detection experiments were carried out at the end of the thesis.The results showed that the accuracy of the algorithm in this paper can reach more than90% under conventional detection conditions,and it has certain adaptability to changes in distance and brightness,and can be used for the looseness detection of regular hexagon nuts.jobs.