Research On Assembly Quality Inspection Technology Of High-speed Rail Cable Clamps Based On Convolutional Neural Network

At present,the assembly quality inspection of high-speed rail cable hanging string clamps mainly depends on human visual judgment.This traditional method not only can guarantee the accuracy but also can make the detection efficiency low,so it greatly hinders the level of automation and intelligence of the production of clamp assembly.With the rapid development of computer technology,machine vision detection technology based on convolutional neural networks which can work under non-contact conditions and obtain relevant data in real time to ensure the accuracy and efficiency of detection.The technology is widely used in face recognition and medical imaging detection areas to achieve the maximum benefits of multidisciplinary integration.So this paper improves and optimizes the related algorithms by machine vision detection technology which is based on the convolutional neural network,to achieve the intelligent detection of the assembly quality of high-speed rail cable hanging string clamps,which is conducive to further improve its production efficiency and assembly precision.The main research work and relevant conclusions of the thesis are as follows:(1)Overall design of wire clamp assembly quality inspection system.Accorrding to the functional requirements and inspection index requirements of the assembly inspection,the system combines the experimental conditions with inspection index requirements to complete the selection of key hardware such as industrial cameras,lenses,and light sources etc.Then this paper not only builds a hardware platform for the clamp assembly quality inspection,but also designs the corresponding software algorithm flow.(2)Detection for error assembly of the clamps.Based on the convolutional neural network algorithm,this paper detects the missing or wrong assembly during the assembly of the clamp.Firstly the acquired sample data are expanded by image processing technology to complete the construction of the data sets.Then this paper analyzes the R-CNN series network models through the structure and working principles,and proposes an improved Mask R-CNN algorithm based on the loss function and candidate window classification structure.Finally detection results are compared and analyzed from this algorithm and Faster R-CNN and Mask R-CNN algorithm,which show that the improved Mask algorithm hashigher detection accuracy and can be used for error detection of wire clamp assembly.(3)Detection for the tightness of the clamp connection assembly,which mainly converts the tightness of the bolt connection into the length of the bolt extension.The method is a kind of non-contact indirect measurements.In order to realize the measurement of the size of the two-dimensional image,the image is divided into regions,edge detection,corner detection and straight line fitting.The principles of common algorithms are analyzed and the corresponding parameters are continuously adjusted through experiments.After comparing the speeds,paper finally chooses several ways to pre-measure the clamp images,including threshold segmentation,Canny algorithm,Harris corner detection and Hough transformation.The key corner points are determined by traversing all possible points.The length of the bolt extension is measured by calculating the average distance between the points and the fitted line.(4)Test and analysis of wire clamp assembly inspection.Firstly,the functions of the software system are clarified,and related functional tests are carried out to meet the actual detection requirements.Then the software is used in combination with the experimental platform and the clamp test set images to conduct the test experiments,which verifies the feasibility and effectiveness of the algorithm in this paper.In summary,through in-depth research on error detection of wire clamp assembly and bolt connection tightness detection,a wire clamp assembly quality detection system is developed based on the convolutional neural network algorithm,which realizes the intelligent inspection of high-speed railway cable.The related research results have a good reference value for the research and development of the assembly inspection technology of the workpiece,which is based on the image processing.