Research On The Geometric Parameter Measurement And The Surface Defect Recognition For The Weld Of Steel Structural Components Based On Machine Vision

The steel structural component is an important part of buildings,bridges,and other infrastructure,and its structural performance determines the quality of the project.Welding is one of the main connection methods in the manufacturing process of steel structural components,and welding quality will directly affect the bearing capacity and stability of steel structural components.As for the problem of weld quality detection,it is currently mainly manual,which is difficult to meet the needs of modern production in terms of detection efficiency,accuracy,and stability.In recent years,machine vision technology has developed rapidly,and using machine vision instead of artificial vision can achieve efficient and accurate intelligent detection of weld quality.Therefore,this article takes the weld of steel structural components as the research object,designs the overall scheme of the visual detection system,builds a visual detection platform,proposes a weld geometric parameter measurement method based on active vision and a weld surface defect recognition method based on passive vision,and conducts in-depth research on related technologies.The main research contents are as follows:(1)Aiming at the problem of the quality detection for the weld of steel structural components based on machine vision,a visual detection scheme is designed.Based on the visual sensor structure layout of oblique incidence and vertical reception type,a visual detection platform is designed,which includes an active vision module and a passive vision module.Finally,a visual model is established.(2)Aiming at the problem of image acquisition and processing for the weld of steel structural components,firstly,using the active vision to collect laser stripe images.By analyzing and comparing image filtering and threshold segmentation techniques,the optimal scheme is selected to preprocess the laser stripe images to accurately extract the stripe target area.Secondly,using the passive vision to collect weld surface defect images,performing data enhancement after preprocessing to construct a weld surface defect dataset WSDD.(3)Aiming at the problem of weld geometric parameter measurement based on active vision,a center of gravity method based on gray value square weighting is proposed for centerline extraction of segmented fringe target regions.Then,based on the extracted subpixel center lines of the laser stripe,this paper proposes a feature point extraction method based on the RANSAC algorithm and distance search method.Finally,based on the extracted feature points,the definition and calculation method of weld geometric parameters are studied,and the measurement of weld geometric parameters is realized.(4)Aiming at the problem of weld surface defect recognition based on passive vision,the lightweight Mobile Net V2 network architecture is improved in accuracy and speed dimensions by embedding the CBAM module and reducing the width factor.After studying the training methods of neural network models,training parameters such as cross-entropy loss function and Adam optimization algorithm are selected to perform the model training process by using the WSDD dataset.Finally,the recognition accuracy rate on the testing dataset reached98.23%,and the weld images to be tested correctly predicted the defect category with a high classification confidence.The accuracy and superiority of the proposed weld surface defect recognition model are verified.