PCB Solder Paste Defect Detection Based On Machine Vision

Solder paste defect detection refers to the inspection of printing quality after the completion of solder paste printing on the production line.By detecting solder paste printing defects,not only can the defective workpiece be reworked in time to reduce losses,but also the defect can optimize the solder paste printing parameters and improve production efficiency.This thesis is based on the actual production project of solder paste printing leakage and multicoating defects,based on machine vision technology,using two methods of image processing and deep learning to achieve the detection of defects in solder paste printing.The main work of this thesis is as follows.In view of the characteristics and detection requirements of solder paste printing defects,combined with the actual situation of the production site environment,this thesis select the vision system hardware,such as the ac A2440-20 gm industrial camera of the Basler,the MFA230-S25 lens of the coolens,the RI9090-W ring light source of the OPT.The above equipment was used to build the solder paste inspection hardware system.As the number of workpieces in a single shot image is six,the workpieces need to be extracted for unified processing.In this thesis,the six workpiece images are extracted separately by contour template matching to exclude the interference of the background and facilitate the defect detection by subsequent image processing methods and deep learning methods.To further improve the efficiency and accuracy of contour matching,an image pyramid hierarchical search strategy is used to improve the search speed and a sub-pixel precision optimization algorithm to improve the matching accuracy.In the image processing-based method for detecting solder paste defects,the maximum interclass variance method is used for image segmentation to extract the solder paste region from the workpiece background and transform it into a binary map,and then the second scan method based on the runs is used to label the binary map with connected blocks.The first scan of the secondary scan method first obtains the runs information of the image,and the second scan performs the connected block labeling on the runs,and this thesis uses the equivalence table method to deal with the redundancy and conflict of block labeling.After the marking is completed,the coordinates of the center position,length and width,area and other parameters of each connected block are calculated,and the pixel units are converted into actual units through the calibration matrix,and then the solder paste defects are detected according to the process requirements on the production line,and the accuracy rate of this method is 89%.In the deep learning-based method for detecting solder paste defects,this thesis first produces a dataset of qualified and unqualified images of workpiece products,and then uses a Swin Transformer neural network model based on a selfattentive mechanism for training,and the completed model is tested with an accuracy of99.23%.The deep learning method is used as the final solution because of its higher accuracy than the image processing method.According to the above method,this thesis implements the modules of image acquisition,camera calibration,defect detection,file management and communication under VS2022 integrated development environment using C++ language to complete the solder paste defect detection software system.After experimental testing,the detection system designed in this thesis can accurately and reliably detect PCB solder paste leakage and overcoating defects,which helps to improve production efficiency and automation of the production line.