Research On Machine Vision Inspection For SMT Assembly Products Defects

With the improvement of production technology and the use of the surface mounted technology, SMT Assembly Products are proned to be more layers,smaller volume and more density. which make the installation and produce realized automation in the electronic manufacturing industry,but it makes the quality check work become more difficult.The traditional examination method can not satisfy the production's needs both in measuring ability and speed already. The study on the automatic machine vision inspection of SMT Assembly Products defects is popular in the defects inspection of SMT Assembly Products, which is based on the computer and the image processing technology.This thesis studied the character choosen and classifier design during the automatic examination for the SMT Assembly Products. Chapter1 made the review and analysis of the surface mounted technology, introduced the domestic and foreign situations of the trade, and explained the positive meaning of the study; Chapter2 introduced the general situation of automatic vison inspection technology and system which are applied in surface mounting technology of micro-electronics,the structure of the inspection system of SMT Assembly Products and the character, The thesis studied on the choice of some equipment at the same time;Chapter3 introduced several kinds of image processing technology and character choosen methods on simulation platform with MATLAB, testing the defects such as missing component, misalignment of component, component in wrong orientation and so on adopting to image processing technology and character choosen methods; Chapter4 had inspected component type, improved character choosen method with wavelet transform and coarse gridding of character, wavelet transform has ability of detecting mutation point and more scale analysis, We used this ability to recognise character, The experiments testified the validity of this method; Chapter5 bringed forward a method of SMT Assembly Products defects recognition using AdaBoost algorithm, which based on multi-image features and a fast delamination approach for defects inspection of imbalance distributing, Adaboost could construct a highly accurate classifier by combining many weak classifiers that just had slightly better accurate than random prediction, The thesis trained the weak classifiers with BP neural network, trained the accurate classifiers with AdaBoost to inspect defects. Experiment proved that the SMT Assembly Products defects recognition arithmetic based on AdaBoost had higher practicability, lower error rate and good ability of generalization.