Multi-Channel PCB Quality Inspection Based On Machine Vision

China is a big country of PCB(Printed Circuit Board)production.In the industrial production of printed circuit boards,the real-time inspection of PCB chip welding quality is an important procedure.The welding quality is directly related to the quality of products,so the realization of PCB welding quality testing technology is an urgent problem for many PCB board manufacturers.This paper mainly focuses on the defects in PCB patch area.The defects include component dislocation,polarity reversal,defective parts,solder joint defects and so on.The main contents of this paper are as follows.Firstly,through analyzing the quality inspection standards of various component welding,a reasonable PCB quality inspection scheme is designed,and a reasonable camera model is selected according to the parameters of various image acquisition hardware devices.At the same time,through analyzing the image requirements of the detection scheme,a set of reasonable image pre-processing approach is proposed.Secondly,in view of the increasing variety of electronic components,it is more and more difficult to judge whether there are welding defects at the same time,and it is also difficult to adapt to complex detection simply using area and appearance.But also it is difficult to make a good judgment on that weld quality according to the common single characteristic.In this paper,a multi-feature extraction scheme is proposed,and good features are selected using principal component analysis.Experiments show that the proposed method in this paper is not only to extract more complete features,but also has advantages to speed up training and recognition time.Finally,by studying the classification algorithms of various types of image recognition,the neural network for recognition classification has excellent effect on various types of recognition,but it requires a large number of sample sets and machine calculation support.Considering the manufacturer's equipment performance and the limitation of the number of data sets provided,it is difficult to use neural network,so this paper adopts SVM(Support Vector Machine)classification scheme according to the actual situation.In order to solve the problem that different types of component information will interfere with the judgment of welding quality,this paper proposes a multi-channel classification scheme.The PCB is segmented structurally in advance,and a SVM single classification channel is designed for each type of component welding area to determine whether the welding is qualified or not,so as to determine the welding quality of the PCB.Experiments show that the detection effect of multi-channel classification is better than that of one-time classification.According to the statistics of experimental data,the multi-channel PCB quality detection proposed in this paper has achieved 94.5% of the overall detection effect.The proposed method in this paper can also be applied to other product testing fields,such as printing quality testing applications.