Quality Identification Of Solder Joints Based On Optimized Limit Learning Machine Algorithm

Production of PCB(Printed Circuit Board)Board in the process of welding is an important part of electronic components,solder joint quality also directly affects the quality of the PCB.In order to ensure the high quality PCB applied to electronic products,improve product percent of pass and avoid unnecessary losses,achieve PCB solder joint defect detection is also becoming more and more important,With the development of electronic components in the direction of refinement,the more and more subtle defects also make the stability and reliability of the quality of electronic products difficult.At present,many practical methods have been put forward in academic circles to solve the problem of welding spot defect in PCB plate,and the pattern recognition mode of extreme learning machine is attractive.Therefore,based on the existing experimental equipment,we design a practical and effective microscopic image welding spot defect detection system.From the specific process,it is mainly divided into the following parts:First,based on the purpose of experiments,we introduce and analyze the software facilities of hardware devices and host devices,such as microscope,CCD,auxiliary light source,image acquisition card,etc.Secondly,the features are extracted,and the core features of the welding spots are extracted from the background,and the LBP feature vectors are extracted by using the OSTU image segmentation technique.Finally,the ELM(limit learning machine algorithm)is used to classify the quality of welding spots.In order to ensure the stability of ELM and improve the extreme learning machine,we use the global optimization ability of adaptive differential evolution algorithm to get the input weight matrix and hidden layer bias matrix of extreme learning machine with relatively small training error,and accurately identify the results.In order to prove the effectiveness of the optimization,the traditional ELM detection and improved DE_ELM detection are compared through a large number of welding spot images with different defects.