Research On Analysis Method Of SMT Product Quality Prediction Based On Big Data

Manufacturing companies are enjoying the dividends of automation and information integration,but also facing new challenges--"data explosion,lack of information".Massive data storage of various information systems has not been conducted in depth data mining,failed to make full use of the hidden information of massive data,is a waste of resources.Especially the Surface Mounted Technology(SMT)production line,its automation and information level is more advanced and productive than other manufacturing industries,traditional quality prediction methods can no longer meet the data processing needs of the big data era.From the perspective of data mining,this paper adopts the core technology of industrial big data and focuses on the prediction of the quality of SMT solder paste printing.The main research work is as follows:1.SMT process flow and SMT big data resources are combed in detail to formulate the SMT product quality forecasting process,which mainly includes feature engineering,data package construction,data preprocessing,quality prediction model construction and model optimization,and combines the mechanism analysis and feature selection methods to determine the paste printing quality critical factors and predicted quality characteristics.2.A feature reconstruction method based on timing characteristics is proposed,using the product quality characteristics at time t-1 as the input of the product to be predicted at time t,minimizing the information loss of some important uncontrollable and uncollectible factors.The use of controllable and collectable discrete qualitative and discrete quantitative small elements to segment the original data is proposed,the basis for dividing the data packet according to the length of the PCB is determined,and setting the time window T in the SMT quality prediction process is proposed to dynamically update the data in real-time and finally the time series data package is build to improve the model prediction accuracy.3.The improved learning strategy that the optimization of clustering cluster values is transformed into finding the best cluster distance of the Agglomerative NESting(AGNES)algorithm is proposed in this paper.The use of AGNES algorithm to analyze the data characteristics to determine the hidden layer neural number and center of Radial Basis Function(RBF)neural network is proposed.Using Particle Swarm Optimization(PSO)algorithm to optimize the key parameters in the clustering algorithm is proposed.The formation and optimization of the quality prediction model based on AGNES and PSO optimized RBF neural network have been comprehensively formed.4.In the end,the analysis ideas and methods proposed in this paper are verified by the data of an enterprise's SMT production line.The results show that this method has the best prediction effect and the accuracy and feasibility of the methods and ideas are verified.