Complex Weld Image Recognition And 3D Reconstruction

In the manufacturing industries such as containers,pressure vessels,shipbuilding,etc.,which have a large amount of welding,due to the complicated structure of the welded components,the thickness of the welded base metal varies,resulting in different weld shapes,varying lengths and widths,and flatness.Shaped,but also three-dimensional shape,for the identification of the weld type has brought challenges.Different welding types have different settings for automated welding parameters.Incorrect welding parameter settings can easily cause the occurrence of welding defects,which affects the entire production.In general,China's large-scale welding process production mainly has the following problems: 1)low degree of automation,the main welding method is still manual welding;2)multiple welds and irregular shape,different types of welds,changes Variety and complexity;3)For the non-weld structure in complex components,it is easy to mistakenly identify the weld,and feature extraction is difficult.4)The spatial structure of the weld is relatively complex,which is not conducive to the acquisition and reconstruction of the three-dimensional structural information of the weld.In the research of this paper,in order to solve the problem of automated and intelligent welding in these large-scale welding environments,an automatic and intelligent welding robot platform was established by image recognition and three-dimensional reconstruction to increase the automation productivity of the welding industrial environment and reduce manual operations.,Reduce labor costs and complete the transition from traditional manufacturing methods to automation and intelligence.This research mainly completes the following tasks: 1)Analysis of complex weld types: In order to study the characteristics of complex welds more comprehensively,a commonality study of various complex weld properties is proposed to analyze the structural properties and 3D information of complex welds,and to establish features.Extraction of vector and classification models to achieve complex weld recognition classification;2)Complex weld feature extraction,positioning,classification and three-dimensional reconstruction: Study of welding component types and structural characteristics,identification of components in the weld and non-welding characteristics,established Feature vector,using decision tree algorithm,support vector machine(SVM)algorithm to classify complex welds,establish dual-vision image acquisition model,use the principle of depth image restoration to reconstruct the three-dimensional structure;3)build experimental platform verification: use Yaskawa MA1440 robot as an experiment The welding robot was verified and the image acquisition was performed using a Keyence cv-x200 camera.Halcon was used as image processing and 3D reconstruction software.The classification model was established using the method proposed in this paper,and the effectiveness and feasibility of the proposed method were verified by experiments..The experimental results show that the research in this dissertation can effectively identify a variety of complex welds in complex workshops,and the recognition rate can reach over 93%.Moreover,3D images of complex welds can be successfully reconstructed to provide automatic welding robots with automatic welding.Parameter setting data.This article is applied to the actual welding work environment,and the research meets the actual welding needs,improves the industry's automated welding technology,reduces labor costs,and has practical significance.