Research On Defect Detection Method Of Stencil Based On Machine Vision

As the electronic manufacturing industry develops faster and faster,the integration and complexity of related products are constantly increasing.Surface Mounted Technology(SMT)is a common process for assembling electronic components on circuit board.Solder paste printing is the first operation of SMT,and its printing quality affects the subsequent craft process.SMT stencil deposits a precise amount of solder paste on the corresponding position of circuit board.Therefore,the correctness of stencil has a direct impact on the printing of solder paste.At present,the inspection method of stencil is mainly manual inspection,but the area and spacing of the stencil mesh are getting smaller and smaller,which increases the difficulty of manual inspection,leads to the accuracy,stability,speed and other performance indicators of the inspection can not meet the requirements of industry.The defect detection method based on machine vision,which combines software and hardware,can effectively deal with the problem of stencil defect detection.This thesis takes the stencil provided by the manufacturer as the main research object.Firstly,it summarizes the research status of domestic and foreign stencil defect detection technology,and combines the needs of the detection scheme,and selects the camera,light source,motion device and other hardware suitable for this system.The equipment has designed detection methods for the six defects in the s stencil that are too large,too small,burrs,hole position offset,multiple holes,and few holes.Secondly,in order to facilitate subsequent research on defect detection,it is necessary to preprocess the image.In this thesis,the noise in the collected stencil image is filtered by the method of superposition of bilateral filtering and median filtering,and then the filtered image will be directly stitched by the feature-based stitching method.Through experiments,the quality indicators of each matching algorithm are counted,and the SURF-based matching algorithm with better experimental results is selected.Finally,the image fusion technology of the weighted average method is used to achieve smooth splicing of the stencil images with an overlap rate of 29.26%.Thirdly,this thesis uses the SURF registration method of loading features to register the standard image and the sample image,and uses the OTSU threshold segmentation method to obtain the binary image.Aiming at the detection of non-burr defects in stencil,this thesis designs two detection methods,namely the defect detection method based on connected domain and the defect detection method based on the contrast method.The connected domain-based defect detection method uses the connected domain labeling method to locate the meshes of the standard graph and the sample graph,and detects the defects according to the defect judgment rules.The defect detection method based on the contrast method obtains the difference part of the standard binary image and the sample binary image through the XOR operation,and then removes the noise through morphological processing,completes the location of the defect,and finally realizes the classification of the defect according to the classification standard.The defect detection method based on the contrast method adopts the contour comparison method based on Hu invariant moment and the enlarged circumscribed rectangle scanning method.Comparing the two defect detection methods,according to the experimental results,the defect detection method based on the contrast method is selected as the non-burr defects detection method.The detection method designed in this thesis makes the missed detection rate and false detection rate of non-burr defects less than 3%,and the average detection time of 300 mm ×400mm stencil is within 30 s,which meets the scheme requirements.Finally,because the detection method based on the reference method is not accurate enough for the detection of burr defects,it is easy to cause missed inspection,so the detection method based on non-reference is carried out for the burr defects.In this thesis,the segmentation algorithm based on IFCM is used to segment the mesh image,and then uses morphological operations and XOR operation to obtain the outer boundary of the mesh.The improved Zhang-Suen algorithm is used to complete the refinement of the mesh image,and finally through the direction-based bone spur removal algorithm detects burr defects.The average detection time for burr defects is13.16 s,which is in line with the design requirements of the scheme and meets the needs of the industry when added to the time used by the defect detection method based on the contrast method.