| Surface Mount Technology(SMT)has a wide range of processes,each process may produce a variety of defects.If the defects can not be found early and accurately on the production line,it may reduce the product qualification rate,affect the reliability,even lead to the scrap of the printed sheet,and then increase the life.Production cost.The SMT production line tests the printing quantity of the solder paste by the detection threshold set in the Solder Paste Inspection System(SPI),and controls the defective printed circuit board(PCB)into the subsequent process.At present,the setting of threshold in SPI detection equipment mainly relies on the traditional manual experience,lacks theoretical guidance,and can not give full play to the ability of SPI to eliminate the defects.It often makes SPI misjudged and missed,so that the real defects are released by human.Therefore,scientific and reasonable setting of solder paste SPI detection threshold is an important part of SMT product quality control.On the other hand,all kinds of testing equipment in SMT production line have accumulated huge amounts of data,but the information and value contained in these data have not been fully excavated.As data driven process modeling technology does not require deep process knowledge and mechanism model,a new method of SPI detection threshold estimation is proposed in this paper using data driven process modeling technology.The main research work and innovation in this paper are as follows:(1)In this paper,the process of SMT process and the data resources of production line are introduced in detail.The process of solder paste printing and quality detection is described in detail.On this basis,the overall framework of the SPI detection threshold estimation process is given,including the construction of data packets,the probability density estimator of the feature parameters of the solder paste and so on.(2)The data sampling technique is used to solve the class imbalance problem of data packets.The kernel density estimation(KDE)of non-parameter estimation is introduced to the probability density estimation of the feature parameters of the solder paste,and the optimization problem of the SPI detection threshold is established.The SPI detection threshold is used as the decision variable,and the target function is constructed from the point of minimization of SPI misjudgement and missing judgment,and the optimal SPI detection threshold is solved by genetic optimization algorithm.(3)This method is applied to the actual production line of SMT in a communication company.Compared with the traditional artificial threshold setting method,the new method has a certain theoretical significance and good application value.It can reduce the defect rate of SMT products to a certain extent and provide effective help for the quality control of enterprise SMT products. |
