| With the development of electronics manufacturing technology,Automated Optical Inspection(AOI)has been widely used in Surface Mounted Technology(SMT)on circuit boards for mount defect detection.At present,AOI equipment basically adopts statistical modeling and template matching detection methods,with high reliability and fast speed.However,the statistical modeling needs to be designed again while each time the product model is replaced.It is necessary to establish a template by performing statistical learning on the currently qualified samples,and then perform automatic detection.The method is time consuming and cumbersome,which results AOI device is not good effective in small-volume and multi-variety electronic manufacturing enterprises.In this paper,we has studied patch components soldering defect detection method based on deep learning.We use the deep learning method to establish a deep convolution network based on the AlexNet network.The network can be trained by the welding image samples of a large number of small components such as resistors and capacitors,and then the welding quality of different types of resistors and capacitors can be effectively detected by the trained network.The universal model can test the welding quality inspection for different types of products by simply judging the position of components such as resistors and capacitors.It avoids the need to re-establish the template after replacing the product model,which improves the production efficiency.The deep learning model not only can extract the classification features directly from the input image,but also can obtain high-level abstract features through layer-by-layer convolution,with the common features of the solder joint defect images is found,the weld different types of soldering defect can be detected accurately and quickly.Finally,it is proved by experiments that the deep learning-based soldering defect detection method proposed in this paper can effectively complete different types of the component solder defect detection.In addition,this paper has studied the problem of component positioning and rapid model detection during the inspection of patch components installation defect detection process and an innovative design was carried out.When it comes to the positioning of the chip components,the combination of PCB board component information extraction and minimum external rectangle detection technology is used to study the componentpositioning and deflection angle detection.The detection speed is increased within the range of permitted detection errors.For the type detection technology of the chip resistor components,a model detection algorithm combining character feature encoding and optical character verification is proposed in the paper.Benefiting from the algorithm performs character recognition in the manner of character grayscale hopping detection and encoding,it just needs to scan at the position of the detection line,and,there is not necessary to perform pixel-by-pixel comparison and single character division.The algorithm not only has good robustness to the micro deflection angle of the component,but also improves the detection speed of the patch component model.In the laboratory tests performed on the PCB surface mounted technology production line demonstrate the effectiveness of the proposed method and will be applied to improve the performance of AOI equipment. |
