Improvement Of RBF Network And The Application In Predicting The Thickness Of Solder Paste Printing

Solder paste printing is the first working procedure of the surface mounted production line, and the thickness of solder paste printing usually reflects the quality of solder paste printing. It affects the following processes of the line directly. Solder paste printing process with the characteristics of complex, multi-variability, it is affected by many factors, which usually have dynamic changes. However, the detection methods for the quality of solder paste printing are the post-test. Therefore, it is necessary to find a more effective means for the quality of solder paste printing.For this reason, this paper presents a combination method by Genetic Algorithm (GA) and Radial Basis Function (RBF) neural network, and use it to predict the thickness of the solder paste printing. Firstly, Factors affect the efficiency are extracted on the basis of deeply studied about the structural characteristics and working process of solder paste printing. Secondly, an improved GA is proposed for the defect of the traditional GA, which is improved on the selection operator and adaptive cross and mutation probability. Thirdly, the improved GA is used to optimize the hidden layer nodes, center and width, and the network weights are calculated by the least mean square algorithm. So the center of RBF can be selected automatically and efficiently. Finally, the improved algorithm is applied to predict the thickness of solder paste printing, and the mathematical model about thickness of solder paste printing prediction is established in the end.The result shows that the model can predict the thickness of solder paste printing effectively, and it can be as a mean to inspect the quality of solder paste printing. It has practical significance and application value to improve the quality of solder paste printing to some degree.However, since the complicated features of solder, it is to study the standards about the quality of solder paste printing. In addition, as the development of optimization algorithm, research newer optimization algorithm to optimize the neural network will be the future study.