Research On Automated Optical Inspection System Based On Extreme Learning Machine For THT Solder Joint In PCB

PCB is essential component of electronic products.The quality of solder joint in PCB can affect stability and service life of electronic products directly.Thus,quality detection of solder joint is very important procedure in production process of PCB.There are some disadvantages such as slow speed and low accuracy in traditional manual inspection so that manual inspection can't meet requirement of production.Thus,automated optical inspection has been widely used in quality detection of solder joint in PCB.At present,for quality detection of solder joint,domestic and international researchers are mainly aimed at SMT solder joint.Little research work about THT solder joint has been done.Aimed at quality detection system of THT solder joint,which is integrated into industrial robot,this article uses automated optical inspection and extreme learning machine to determine whether the quality of solder joint is qualified,so that efficiency of detection can be improved and reliability of detection results can be assured.Research contents and achievements of this article are as follows.(1)Overall scheme design of automated optical inspection system has been achieved.This scheme design includes model selection and calculation about industrial camera,lens,and light source.Besides,software platform and process of solder joint detection algorithm have been determined.(2)Calibration of automated optical inspection system has been achieved.Correspondence between points in image and in threedimensional space has been obtained.Then position of marks in PCB have been measured.According to position relationship between marks and solder joints,positions of solder joints in image has been obtained.And positions of solder joints in three-dimensional space can also be determined.(3)In order to obtain input of machine learning algorithm,plenty of features in image of solder joint have been extracted,and information gain is used in feature selection.K-means clustering algorithm can achieve discretization of feature data during calculating information gain.Best number of cluster is determined by CH index.(4)Cost-sensitive extreme learning machine is used to detect the quality of solder joint.Experimental result shows that its effect is better than general extreme learning machine in classification of imbalanced samples.Then artificial fish swarm algorithm is used for parameter optimization and performance of cost-sensitive extreme learning machine has been improved dramatically.