Research On Key Techniques Of The Industrial Parts’ Surface Inspection Based On 3D Vision

With the development of computer vision technology,the research and application of automatic and contactless inspection of industrial part surface quality have attracted more and more attention.Due to the richer object surface structure data and more texture data that 3D vision sensors can provide,traditional vision-based technologies in the industry are gradually starting to adopt inspection data provided by 3D vision-based sensors to meet the inspection requirements of higher accuracy and precision.This paper designs and implements a surface quality inspection system based on high-performance 3D vision sensors for industrial parts with simple and complex shape structures and apply them to quality inspection of high-speed brake pad and oil pipe threads through cooperation with companies.The main contributions of this paper are as follows.1.For industrial parts with simple shape structures,this paper proposes a surface quality inspection method,including dimensional measurement and defect detection.This method can be targeted to finish the specified inspection and measurement requirements according to the geometric characteristics of industrial parts.The geometric characteristics include simple contours,simple geometries,and flatter surfaces of industrial parts.In addition,the solution designs and implements more general algorithms,including point cloud data pre-processing,plane extraction,and effective point set extraction algorithms,which facilitate the targeted implementation of surface quality inspection tasks for industrial parts with simple geometric structures and shorten the development cycle of the quality inspection process.2.For industrial parts with more complex shapes and structures,this paper proposes a set of surface quality inspection methods based on its standard model data.The method designs and implements a system including one-time manual labeling,point cloud registration,searching for labeled point sets,and a specific dimensional calculation strategy.To complete the quality inspection,we perform one-time manual labeling of in the standard model,then use the point cloud registration and search for labeled point sets to complete the positioning of the parts to be inspected and measured in the scanned point cloud,and follow specific rules with the calculated object already labeled.Moreover,this method simplifies the design logic of computational objects in surface quality inspection,so it is scalable for different inspection needs.3.To verify the feasibility and efficiency of the proposed methods,the mentioned methods were applied to the surface quality inspection tasks of high-speed brake pads and oil pipe threads.Because of the simple surface structure of the high-speed brake pads,the inspection method of the surface quality of industrial parts with simple shape structures was applied,and completed the inspection of each parameter of the surface of the high-speed brake pads.The surface quality inspection solution for industrial parts with complex shapes was applied to the oil pipe thread because of its complex curved spiral structure,and finally,the surface parameters of the oil pipe thread were successfully inspected.The experimental results show that the accuracy of the surface quality inspection method of industrial parts proposed in this paper meets the actual production requirement.Meanwhile,the efficiency and accuracy of defect detection are significantly higher than that of manual inspection.