| Baby carriage is one of the indispensable children's products for family parenting.As the main parts of the baby carriage,the quality of metal pipe fittings directly affects the safety of the baby carriage.Thus,the quality control of the baby carriage metal pipe fittings is the top priority for the enterprise.According to the actual needs of some well-known brand of baby stroller manufacturer,in this paper,machine vision inspection technology was applied to the processing quality inspection of baby stroller metal pipe fittings,also,the revelent detection algorithms were studied and the corresponding testing equipment was developed,which can promote the realization of high-precision,high-efficiency and intelligent detection of baby stroller metal pipe fittings.The thesis is divided into three aspects,which are generalized visual inspection,multi-precision inspection,and neural network-based defect classification technology,including the following research contents:(1)According to the task requirements of one machine with multiple functions put forward by the baby carriage manufacturer,a universal design of visual inspection equipment was carried out.In terms of hardware,for the problem of large differences in the types of parts,considering the universality of the testing equipment interface end,starting from the positioning jig of the baby carriage fittings,a positioning jig that can be flexibly disassembled and accurately positioned was designed;in terms of software,the inspection items of the metal pipe fittings of the baby carriage were classified and summarized,and the achievable inspection items can be divided into 6 categories,including 17kinds of size inspection functions.(2)The feature positioning method and image processing technology were studied.The distribution of the features to be tested on the metal pipe fittings of the stroller was scattered.In order to improve the detection efficiency of parts and remove the interference of irrelevant areas,reference points were introduced in combination with the structural characteristics of each part,and the ROI of the features to be tested was accurately positioned by the method of offsetting the position of the reference points.The combination between threshold value segmentation and morphological image processing technology can remove oil stains,indentations,burrs and other interferences surrounding the features.(3)The round hole fitting algorithm required for multi-precision detection was studied.The round holes of the metal pipe fittings of the baby carriage may come from machining or stamping,and the detection accuracy of round holes of different processing methods is quite different.To this end,an ellipse fitting algorithm based on geometric symmetry with adjustable precision was proposed,which can solve the problems of circular hole deformation and difficulty in achieving multiple precision detection with a single device.(4)The classification method of metal pipe processing defects based on neural networkwas studied.Aiming at the problems of high similarity of defects in baby carriage pipe fittings and low accuracy(approximately 60%)of traditional machine vision recognition,neural network technology was applied to the defect recognition of baby carriage metal pipe fittings.The Google Net model was modified to achieve the defect recognition accuracy rate of above 97%,which can meet the actual production needs of the enterprise(above 96%).(5)The application software of the baby carriage metal pipe fitting detection system was developed.Based on the development of user operation interface and deep learning framework,the user operation software was constructed.Correlative experiments have verified that the measurement accuracy of this set of equipment is up to 0.06 mm,and the accuracy of defect recognition has reached above 97%. |
PDF Full Text Request