Research On Machine Vision Based Broken Wire Damage Detection Method For Steel Wire Rope

As an important load-bearing component in the industrial field,steel wire ropes often suffer from various types of damage such as broken wires and wear on their surfaces due to their harsh working environment,which leads to a reduction in their load-bearing capacity.Efficient and reliable damage detection of wire ropes is of great significance since failure to timely replace damaged wire ropes can easily lead to safety accidents,causing property damage and casualties.This paper researches the damage detection method of wire rope surface broken wire based on machine vision.The main contents are as follows.(1)To obtain surface image data of wire ropes,an experimental platform was designed that can automatically acquire panoramic images of wire ropes in the circumferential direction.The platform consists of four parts: an experimental platform frame,power control section,image acquisition system,and information feedback device.The image acquisition system uses slider rail components to capture surface images of wire ropes with different diameters;adds a roller device to reduce wire rope shaking and improve imaging quality;and employs a bar-shaped light source with a diffuse reflector plate to reduce the influence of surface reflection on imaging.One to four broken wire damage types within the length range of 1-5mm,5-10 mm,10-15 mm,and 15-20 mm were respectively created on a 22mm-diameter wire rope,totaling 16 damages,and surface images of the wire rope specimens were acquired using the image acquisition experimental platform.(2)To comprehensively detect wire rope damage in the circumferential direction,a panoramic image generation method for the surface of wire ropes was studied.Firstly,in the image preprocessing stage,images collected by the image acquisition experimental platform were sequentially corrected in the circumferential and axial directions,and Gaussian filtering was used to denoise the corrected images.Secondly,after comparing and analyzing common image registration methods,a fixed-region-based image registration method was used for image registration.Then,after comparing and analyzing common image registration methods,an improved pixel average weighted fusion method was used for image fusion.Finally,a panoramic image of the surface of the wire rope was generated.(3)To identify the number of broken wires on the surface of wire ropes,a wire rope broken wire damage pattern recognition method based on an improved residual network was studied.Firstly,a residual network was used to avoid network degradation caused by deep convolutional neural networks.Secondly,an attention mechanism was incorporated to enhance the network’s focus on targets and recognition performance.Then,transfer learning was introduced to reduce model iteration times,accelerate model convergence,and improve recognition accuracy.Finally,the recognition accuracy of the proposed improved residual network model for identifying wire rope broken wire damage was compared with that of other three network models.The results showed that the proposed improved residual network model had the highest recognition accuracy for wire rope broken wire damage,at 97.4%.(4)To obtain contour and position information of broken wire damage on the surface of wire ropes,a wire rope broken wire damage object detection method based on an improved YOLOv5 was studied.Firstly,the backbone network of the YOLOv5 n model was replaced with a Mobile Net V3 structure that uses depth-separable convolution.Then,the transition layer of the YOLOv5 n model was replaced with a Slim Neck structure that uses GS convolution modules.Finally,the detection effect of the proposed model for wire rope broken wire damage was compared with that of other three network models in the CPU environment.The results show that compared with the original YOLOv5 n model,the m AP value of the proposed model decreases from 98.4% to 98%,which is reduced by 0.4%.The number of parameters decreased from 1.76 million to 0.69 million,a decrease of 61%;The detection frame rate increased by 21.6% from 104.1 to 126.6.