Polar Device Intelligent Identification Method Towards SMT Technological Process

In the SMT(Surface-Mount Technology)process,there is an inefficient but extremely important self-test and error-prevention mechanism.This mechanism requires the inspector to use the PCB(Printed Circuit Board)bit-map as the benchmark to inspect the first product model produced,the focus is on the polarity of the components.In the actual inspection process,the staff can only perform effective inspection with a relatively small number of components.In order to enable the inspectors to perform the inspection of the product model more efficiently,this dissertation proposes a method to intelligently identify the polar components in the PCB bit-map based on the deep learning theory.Firstly,the original PCB engineering bit-map is preprocessed,then we constructed the PCB Bit-map Dataset and enhanced the data by two-dimensional rotation.By analyzing and comparing the three classic target inspection algorithms often used in engineering applications,Faster R-CNN,SSD,and YOLOv3,and performing performance comparison experiments of three algorithms in the Bit-map Dataset built for the first time,we finally selected YOLOv3.In order to improve the ability of the network model to inspect relatively small components in the bit map,two improvements have been made to the original YOLOv3 network.One is to use the feature-fusion method to extract the fine-grained features of the deep layer of the network and the location features extracted from the shallow layer.The other is to use K-means algorithm to cluster and analyze the dataset,and select the number and size of anchor frames suitable for this dataset to increase the convergence speed of the model.In order to improve the inspection accuracy of the network model,this dissertation adjusts the input resolution of the training picture and optimizes the network training parameters,and obtains the best network model for polar device inspection.The AP value of the final version of the improved network model reaches to97.5% in the Verification set.After testing multiple actual engineering bitmaps,the average polarity device inspection rate can reach 95.13%.On this basis,this dissertation builds an intelligent online detection system for polar devices by using HTML,Javascripts and Flask framework,and deploys the intelligent detection algorithm on the system to provide the production personnel with the polarity of the online detection bit number map.The platform of the device improves the execution efficiency of the self-test and error prevention mechanism in the SMT process.