Study On Identification And Three-Dimensional Modeling Of Disc And Sleeve Parts Based On Machine Vision

When using traditional manual modeling methods for reverse modeling of disc and sleeve parts,size measurement,model establishment,and adjustment of drawings are required,which are relatively complex and need to be repeated continuously.In order to improve design efficiency and shorten the R&D cycle,this thesis proposes a machine vision-based method for identifying and three-dimensional modeling of disc and sleeve parts.By combining machine vision with parametric design,identification,measurement,and modeling of parts are completed,ensuring high model accuracy while maintaining modeling efficiency.This thesis takes three types of disc and sleeve parts,namely gears,shaft sleeves,and flanges,as the research objects.By combining machine vision with parametric modeling,identification,measurement,and modeling of disc and sleeve parts are achieved,and the main research contents are as follows:(1)Overall design of part recognition and measurement.Firstly,the requirements for part recognition and measurement were analyzed,and the selection of key equipment was determined for the hardware part.Secondly,modular thinking was used to design the system software,and Pyqt5 was used to complete the interface design of the software part.Finally,the functions of each module were studied to complete the overall design of the part recognition and measurement system.(2)Design and experimental research of part recognition network model.Based on the YOLOv5 model,the Ghost module was first used to replace the backbone network to reduce the number of model parameters.Then,a new spatial pyramid pool SPP＿F was constructed to accelerate the training speed of the model and strengthen the model’s ability to extract information.The detection layer for large target scales was removed to further speed up the training and detection of the model,and the SE attention mechanism was introduced to give higher weight to channel information of small targets.Finally,the CIo U function in the YOLOv5 model was replaced with the α-SIo U function to improve the location accuracy of the bounding box.Experimental results showed that the improved model achieved an average precision value(m AP)of 99.5% and had a volume of only 3.3MB.Compared with the original YOLOv5 model,the inference speed was increased by35%,meeting the deployment requirements in low-computing-power devices.(3)Measurement method and experimental research of part size.Firstly,the camera was calibrated using a chessboard pattern,and the obtained parameter matrix was used for distortion correction.Secondly,to reduce the computational load of the algorithm,a weighted average method was used to grayscale the image,and the denoising performance of mean,Gaussian,median,and bilateral filtering was analyzed.The median filtering,which had the best denoising and edge-preserving effect,was selected.In order to facilitate the extraction of edge contours,the Otsu’s threshold processing was used to separate the target image from the background.To improve the location accuracy of edge extraction,a coarseto-fine extraction method combining Canny operator with sub-pixel operator of polynomial interpolation was used.In order to accelerate the fitting speed,a least-squares fitting circle algorithm based on equidistant sampling was proposed.Finally,the parameter characteristics of gear,axle sleeve,and flange parts were studied to complete the measurement process design,and the experimental results of the measurement parameters were analyzed.The experimental results showed that the key measurement errors of the three types of parts were within ± 0.1 mm,verifying the accuracy of the measurement method proposed in this article.The stability of the measurement method was verified according to the 3σ criterion as an evaluation index.(4)Design of part modeling system and case analysis.Firstly,the geometric models of gear,shaft sleeve,and flange parts were analyzed,and based on the structural characteristics of each part,the part template model and 2D engineering drawing library were established.Secondly,Visual Studio 2019 platform was used with VB.net as the programming language to develop Solidworks 2018 for secondary development,and a parameterized modeling system for parts was constructed.Finally,the key dimensions from visual measurement were used as the core variables to drive the part parameterized modeling system,completing the modeling of the parts.The 3D model of the parts and 2D engineering drawings were updated,facilitating the closed-loop of product inspection,design,production,and assembly,achieving comprehensive automation,and greatly improving industrial production efficiency.