The Algorithm Research Of Automatic Optical Inspection For Solder Joints Based On Pattern Recognition Technology

With the development of the microelectronics technology, electronic components rapidalydevelop towards ultra thin, high density, small spacing and lead-free.This makes a higherrequirement to the welding quality of printed circuit board, visual inspection of solder joint onthe printed circuit board is difficult to meet the production requirements. And Automatic OpticalInspection (AOI) system can inspection high density of micro components pasted on the PCB(Printed Circuit Board, PCB), and the this Inspection method is standard unified, fast speed,online management to product quality, but the existing AOI system has the following severalaspects of the problems in the practical application, the programming is relatively complex,especially the debugging time is too long; AOI system has high faults missed and false alarm; Asan online examination system, the AOI test speed still needs to improve. In this disserrtation,therefore, study on the three aspects of optimization and improvement AOI: positioning,programming, and defect detection. Main research contents are as follows:(1) A new position method based on three layers of MARK point is developed. the first layeris the whole board alignment which compensate the position error of the PCB loading process,the second layer is the single board alignment which compensate the position error of the singleboard assembly, the last layer is the partial FOV(field of view) alignment which compensate theposition error of the partial uneven surface. The experiment result proved the proposed positionmethod improved the position accuracy obviously.The path planning problem of AOI can be modeled as a standard travelling salesman problem(TSP), to minimize the overall working time, a new method based on Hopfield net algorithm isproposed to optimize the path planning problem. The experiments show the Hopfield netoptimize the planning path.(2) A new method is presented of image matching based on statistic modeling. a standardtemplate image was formed through statistic the gray value of good samples. The testingcomponent image and the model image were compared after alignment, then the comparisionresult determines whether the testing component image qualified or not.The experimental resultsshow that: the proposed method meets practical application requirementsin the rate of false alarmand faults missed And effectively simplifies the programming and debugging, reduces thehuman factors that influence the result of the test, improved the AOI accuracy and efficiency. At present, the algorithm has been successfully applied in actual production, to give users a greatconvenience and benefits.(3) In the process of solder joint inspection, the loss of faults missed is greater than the loss offalse alarm, so this paper proposes a new method based on the minimum risk bayes classifier forsolder joint qualified/defect detection. Adaboost was introduced as a strong classifier.Experimental show that the minimum risk bayes classifier in the process of classfying solderjoint has the best inspection result, the inspection result of Adaboost algorithm is furthervalidated the good performance of minimum risk bayes classifier.(4) A neural network combined with genetic algorithm for the diagnosis of solder joint ispresented.14features are extracted as input features for the classification. The neural network iseasily become over-fitting because these input features are not independent of each other, so thegenetic algorithm is introduced to select and remove redundant input features. The experimentalresults have proved that the neural network combined with genetic algorithm reduced the numberof input feature and had a satisfying recognition rate.(5) A feature selection and a two-stage classifier for solder joint inspection have beenproposed. The images of solder joint can be obtained. The color features including the averagegray level and the percentage of highlights and the template matching feature are extracted. Afterfeature selection of information gain, based on the algorithm of Bayes, each solder joint isclassified into by its qualification. If the solder joint fail in the qualification test, the solder jointis classified into one of pre-defined types based on the support vector machine (SVM), thechoice of the second stage classifier is based on the performance evaluation of various classifiers.The experimental results showed that the proposed scheme not only more efficient, but alsoincreasing the recognition rate, since it reduces the number of needed extracted features.Finally, the future research direction of the AOI is prospected based on the summarizationof the whole paper.