Research On The Detection Method Of Nozzle Based On Machine Vision

In recent years,electronic products are gradually developing towards the direction of high density.With the fast development of microelectronics manufacturing,the requirements of electronic assembly technology for the detection technology based on machine vision in the direction of intelligence and automation are getting higher and higher.In the practical project,due to the long working time of the system and the influence of the working environment,dust and other sundries may mix into the equipment,resulting in different degrees of blockage of the nozzle.In addition,the nozzle may be damaged or missing due to mechanical friction and vibration loss,which may have a great impact on the ability of the nozzle,and may greatly reduce the mounting accuracy.The objective of this subject is to test the various types of nozzles used by the mounter.Aiming at fast,accurate location and detection of nozzle,this paper designs a nozzle detection system,studies and optimizes the detection method of nozzle based on machine vision,and improves the intelligence of nozzle detection.The efficiency and accuracy of nozzle detection have been improved to some extent.The main research work is summarized as follows:1.Deeply investigated the characteristics and research status of the application of modern visual inspection system in various fields,learned and got familiar with the visual system inspection process.2.Based on the analysis of the structure and advantages and disadvantages of the current vision detection system,a set of vision system scheme for nozzle detection was proposed after discussion with the engineer.The visual part of the nozzle detection system is determined.3.In view of the possible influence of lighting and other environmental factors on the image collected by cameras in industry,theoretical analysis was carried out on the typical image preprocessing methods in digital image processing,such as image filtering,threshold segmentation,edge detection,feature extraction,etc.,and the algorithm suitable for this subject was selected by experimental comparison of calculation time and effect.4.The object detection algorithms commonly used in the field of visual detection are analyzed,and a new detection algorithm of fusion of importance sampling and generalized hough transform is proposed to solve the problem of misdetection of scaled and rotated images by the traditional generalized hough transform algorithm,which has a large amount of computation,low efficiency and template matching algorithm.The generalized hough transform is placed in the importance sampling frame,the edge points are screened according to the significance measurement value,and the weighted multiresolution analysis method is adopted to reduce the calculation greatly.Experimental results show that this method is effective when the object is under the influence of rotation,shrinkage or noise.5.In order to improve the detection speed of vision system,a new detection algorithm based on generating convex hull to determine the region of interest is proposed.The convex hull is generated by filtering feature points which is used to extract the region of interest to narrow the detection range,and a fast search strategy is designed,which greatly reduces the computation.The ROI region determined by the convex hull can contain most of the key feature points,so as to contain the object to be detected better.Moreover,local features and global features are integrated,and it has a good effect in the case of image shrinkage and rotation transformation through multiple experiments.6.Based on the results of the nozzle location detection,through the calculation of the nozzle internal and external contour,cross-sectional area and other relevant physical parameters,according to a series of algorithms such as Open Cv software library and regional growth method,the integrity and quality reliability of the nozzle components are calculated according to the evaluation criteria,and the quality of the nozzle is analyzed through experiments.