Research On Defect Detection Method Of Circuit Board Components Based On Machine Vision

As a high-end industrial representative,the all-digital and high-precision servo system is being widely used in various fields of industrial control.As the core component of the servo system product,the driver circuit borad product defects to be found as early as possible is a crucial link to ensure the quality and safety of the driver circuit board,and it is also a powerful guarantee for improving the quality of the servo system product.In order to improve and enhance the inspection efficiency,accuracy and robustness of the driver circuit board surface defect detection process,this paper is mainly based on machine vision technology,taking the circuit board component loading defect as the research object,and carrying out in-depth research on related algorithms in the circuit board component defect detection process,and designing and implementing a defect detection system for driver circuit board components.The main research has following four aspects:(1)According to the actual inspection requirements,the software and hardware design of the driver circuit board component defect detection system is given.Aiming at the problem that it is difficult to obtain the same level of circuit board images in the image acquisition process,the Super Point feature point detection and description network is introduced for image feature point extraction.Compared with traditional feature points in multiple indicators,the Super Point algorithm has the advantage of matching performance in the scene of viewpoint change and illumination change,and finally the feature point extraction method is used to complete the accurate registration of the test image and the standard image.(2)Aiming at the problem that the current component defect detection is generally unable to effectively distinguish and locate different types of components,a circuit board component positioning and identification algorithm based on the improved YOLOv4 network is proposed for component positioning and identification.According to the characteristics of components scale changes greatly and the numerous small targets in the circuit board,the original YOLOv4 network structure has been improved.A detection layer sensitive to small targets has been added on the basis of the original three detection scales to enhance feature fusion,and combined with K-means++ clustering optimization to select prior box size to improve multi-scale target detection.Experiments show that this method can effectively identify and locate circuit board components,which is conducive to the improvement of subsequent defect detection accuracy.(3)Aiming at the problems of low efficiency,high false detection rate and incomplete coverage of defect types in current component defect detection methods,a similarity matching calculation method between component direction,position and polarity feature vector is proposed to quickly detect five types of components defects:missed,wrong device,deviation,skew and polarity reversal.Based on the results of component recognition and positioning,this method extracts and constructs the direction,position and polarity feature vector to express the absolute position of the component body and its polarity mark on the circuit board.And then quickly calculates matching error of matched component feature vectors in the position,angle and polarity identification direction between the standard circuit board and the tested circuit board,and then quickly and effectively detect the five common defects of the components.(4)According to the above research results,including the component positioning and identification algorithm,and defect detection algorithm proposed in this paper,a software system of the driver circuit board component defect detection platform based on machine vision is implemented.And the basic framework,functional module design and software operating interface of the software system have been introduced.