Research On PCB Defect Recognition And Classification Based On Eddy Current Testing

Printed circuit boards(PCBs)support information technology,therefore there are a series of high-precision industrial production standards for the appearance quality,circuit connection,and other aspects of PCBs.Especially,defects such as abnormal holes and wire breakage on the surface of PCBs during the production process can affect the functionality and quality of PCBs.It is very important to select defective PCBs to avoid entering the market.At present,the main defect detection technologies for PCBs include mechanical testing,radiation testing,and manual testing,but these testing technologies still have their own limitations.This thesis focuses on the defect recognition of PCB eddy current scanning images based on non-destructive testing,and the research content includes the analysis of PCB eddy current testing technology,construction and analysis of PCB eddy current scanning simulation model,design and hardware selection of eddy current testing system,generation,image processing,and feature extraction of eddy current pseudo color images,construction and optimization of target detection models.The specific research content of this thesis is as follows:(1)Aiming at the problem of verifying the feasibility of eddy current testing PCB,the mathematical model of finite Metacomputing is derived based on Maxwell equations,and the electromagnetic finite element simulation of PCB is established,Receive the real part voltage and imaginary part voltage of the coil,extract the characteristics of the signal,analyze and process the signal,and then obtain the optimal operating frequency required for eddy current scanning.(2)In order to better evaluate PCB damage and locate defects in PCB images,eddy current C-scan was used to collect real and imaginary part data of the signal,and the data was calculated in the complex plane.Time domain image processing methods were used to study the staining methods of different pixel gradient characteristics.Combined with two-dimensional Fourier transform,pseudo color images were obtained,intuitively characterizing the PCB eddy current C-scan image and defect location.(3)In response to the problem of low accuracy in PCB defect recognition models,Faster R-CNN was selected as the basic model to perform dataset matching processing on eddy current images.Based on the CNN algorithm,defects in the pre processed eddy current images were further classified.The original feature extraction network VGG16 was replaced with a deeper Resnet50 network to improve algorithm accuracy.Resnet-50 was selected as the backbone feature extraction network and basic classification model,Optimizing the partial structure of the model,improving the overall performance of the model,and enhancing the ability to classify PCB defects.The characteristics of this research work are reflected in the development of a three-dimensional displacement table,phase-locked amplifier,signal generator,signal acquisition card,and LabVIEW eddy current testing system;Construction and improvement of defect target detection algorithms.