Research On Quality Inspection Of Robot Gluing Process Based On Vision

In automobile manufacturing production,glue coating defect refers to the phenomenon of glue leakage,glue breaking,too narrow and too wide due to the contact problem between robot and automobile,glue quantity problem and technical problem in industrial production.Automobile gluing defects are very harmful,even harmful to human life.In this paper,a set of target detection technology of gluing defects based on convolutional neural network is studied for the quality detection of automobile gluing.The purpose is to provide an efficient and practical test method for automobile glue coating in the automobile manufacturing industry,to help enterprises to better control the quality of glue coating,save costs and improve productivity.At present,visual based auto glue defect detection mainly USES artificial feature extraction algorithm combined with classifier for defect classification.However,due to the diversity of adhesive defects and the poor generalization ability of traditional machine vision model,the high rate of missed detection cannot meet the real-time problem.Therefore,in this paper,the target detection algorithm based on convolutional neural network is applied to automobile glue coating defect detection,which can realize automatic feature extraction of automobile glue coating defects and realize defect classification and defect location positioning.In this paper,the target detection algorithm based on deep learning was applied in the detection and recognition of automobile adhesive defects,and the single-stage and two-stage classical algorithms such as Faster R-CNN,SSD and YOLOv3 network detection models were established to study the detection and classification effects of different network structures in automobile adhesive defects data sets.Finally,YOLOv3,whose m AP was 86.3%,was selected as the final network model of this experiment.In order to optimize the glue coating detection model in this paper,the YOLOv3 algorithm was improved for the deficiencies of the glue coating defect data.In order to improve the detection accuracy and reduce the phenomenon of missing detection,this paper applies DIOU to change the loss function of the algorithm,at the same time,it USES DIOU to optimize the NMS algorithm,and applies k-means ++ algorithm to re-cluster anchor box according to the characteristics of the data set in this paper.Finally,in order to improve the detection speed and reduce the number of parameters,the deep separable convolution is used instead of the standard convolution of feature extraction network.The experimental results in this paper show that the improved YOLOv3 algorithm can improve the detection speed,accuracy and positioning accuracy of the glue coating detection model,which can not only reduce the empirical error of manual detection,but also avoid the complex feature construction in the traditional visual method.