Research On PCB Solder Joint Inspection Based On Rough Set Theory

Nowadays,the development trend of the electronic component manufacturing industry is very rapid,and the automatic optical inspection technology for PCB solder joints has become particularly important.Since the quality of PCB solder joints directly affects and relates to the quality of the corresponding electronic products,how to improve the accuracy of solder joint defect detection,how to quickly complete the on-line inspection of solder joints with high quality and quantity,these problems have already been It has become a research hotspot in the field of electronic component manufacturing,and it is also an important issue to be solved urgently.Therefore,this paper takes the detection of PCB solder joint defects using automatic optical inspection as the research background,and solves the problems ofpoor real-time caused by the cumbersome data miscellaneous in the automatic optical inspection system during solder joint inspection[1-2]The following research work:(1)Microvision vision platform and camera calibration.The Microvision vision platform was researched and used to study and analyze the light source environment of the experiment,and compare the effects of ring light source and other light sources on the image acquisition of solder joints.The camera was calibrated using an industrial area array camera,and the matrix parameters and distortion parameters of the camera were calculated.(2)Preprocessing and feature extraction of PCB solder joint image.The industrial camera is used to collect the solder joint image of the circuit board.Through research and several different image denoising methods are adopted,the adaptive median filtering is finally selected for image denoising.Then,using the segmentation method of global threshold and Canny operator,the edge of the solder joint image is extracted.Finally,the extracted edge features of the solder joint are counted.(3)Rough set attribute reduction features.There are many problems with the welding spotfeature data,such as redundancy.The characteristics of training samples are reduced by rough set attribute reduction method,and the combination offeature attributes is selected through experiments.The simplified solder joint feature also has good recognition efficiency and reduces solder joint feature matching complexity.(4)Solder joint defect detection and comparison experiment.BP neural network,rough set feature reduction algorithm and similarity comparison algorithm were compared.By comparing the correct rate of solder joint detection,false detection rate,missed detection rate and detection time.The rough set feature reduction algorithm can effectively solve the problem of long detection time caused by feature redundancy,the detection accuracy is 94.9%} and the detection time is shortened from 4.1s to 3.9s.