| IC(Integrated Circuit)have been widely used in various electronic products due to their superior performance.As the rapid innovation of surface mounted technology(SMT),Highdensity flexible integrated circuit substrates(FICS)have entered an era with 10μm processing precision,which has the advantages of being thinner,bendable and allows for denser mounted components.However,the FICS manufacturing with high mounted precision inevitably leads to higher risk of defects and difficulty of defect detection on their surface.Also,the harsh requirements were gave for real-time and accuracy of the defect detection algorithms related.Most of the existing defect inspection algorithms for high-density FICS have been improved based on those methods applied to low-density printed circuit board(PCB),which need to be interfered by personnel in FICS’ surface defects and SMT solder paste detection.In summary,aiming to the key technologies and challenges in the quality control process of high-density FICS packaging,related researches were conducted,it consists of the following innovations:(1)Focused on the quality inspection of high-density FICS packaging lines,a substrate vision imaging module at microscopic and multi-scale and a SMT solder paste vision imaging module at stereo and microscopic are proposed separately in this paper.In detail,the proposed multi-scale imaging module is able to output the global field of view images and the local defect images for FICS at the low and high magnification respectively.The defect detection algorithm could choose the input images at different scales according to the task under handle,which improved its accuracy and efficiency.Also,the proposed imaging module can suppress the interference of FICS background texture structure and the influence of industrial noises in the local substrate imaging at the high magnification.(2)Focused on the diversity of high-density FICS surface defects and the low accuracy,inefficiency,and high omission rate using single network model,a deep convolutional neural network(DCNN)based hierarchical cascaded model is proposed applicable for high-density FICS quality inspection.The complicated FICS surface quality inspection tasks are decoupled into several sub-tasks and deploys them at different stages of the proposed cascaded structure.And the adopted hierarchical decision-making mechanism allows qualified and unqualified FICS types to have different workflows,which could improve the efficiency and accuracy of quality detection for high-density FICS.(3)Focused on the problem of the imbalanced datasets due to the difficulty in obtaining a large and balanced defective samples in the actual high-density FICS packaging processes,an improved YOLOv4 detector in the second stage of our cascaded structure is proposed.It’s considered to replace classifiers in the existing methods to perform FICS quality classification,which could reduce the impact of imbalanced datasets on defect detection results.Also,the third stage of our model uses a classifier with the selective activation mode to perform defect severity pattern classification,which changes the existing scheme that FICS is discarded directly as soon as defects are found on its surface.It thus boosts FICS production efficiency and raw material utilization.(4)Paid attention to the difficulty in reaching good balance between inspection precision and efficiency when applying image feature analysis based low-density solder joint methods to inspect high-density SMT solder joint,a differential geometry tool based quality detection method is proposed for the SMT solder joint paste.Considering more key sub-region feature extraction windows and introducing 3D crest line features,a series of predefined logical rules are utilized to quickly match the feature ranks of different solder joints based on decision tree search strategy in our algorithm.Therefore,the SMT solder joint to which a type belongs is obtained rapidly,which could raise the success rate and accuracy for SMT solder joint defect detection. |
PDF Full Text Request