Research On Quality Prediction Method For SMT Production Line

The electronics manufacturing industry is a strategic and basic industry that provides key technology and equipment for the construction of the countrys new generation of information technology industry.The Surface Mount Technology(SMT)production line is a typical production line in the electronics manufacturing industry.Nowadays,electronic products are developing in the direction of diversified functions,refined sizes,and complex devices,which puts new requirements on the quality of SMT production lines.At present,the production process of SMT production line generally has problems such as low level of refined management of product quality,poor correlation between product design and R&D data,production data,and quality inspection data,and poor product quality analysis and poor predictability.Traditional statistical analysis methods cannot effectively extract knowledge and laws from massive amounts of disordered data.At present,methods such as machine learning and deep learning have outstanding performance in solving predictive problems in multiple fields,providing new opportunities and new directions for the quality prediction for SMT production line.Therefore,this article combines machine learning,deep learning and other methods to carry out the research research on quality prediction method for SMT production line.The main research contents are as follows:(1)Construct an overall framework for SMT production line quality prediction.First,the process flow of the SMT production line is introduced,and the production data of each stage of the SMT production line is sorted out in detail,and the data characteristics of the SMT production line are summarized.On this basis,the SMT production line quality prediction index is determined,and whether the solder paste on the pad is defective after printing is used as the first-level quality prediction index of the SMT production line quality prediction,and the specific defect corresponding to the solder paste that will be defective type is used as the second-level quality prediction index for SMT production line quality prediction.The batch qualification rate is used as the third-level quality prediction index.And finally determined the overall framework of SMT production line quality prediction based on the analysis results.(2)Construct a printing quality prediction model for SMT production line.Build an integrated DNN printing quality prediction model based on ensemble learning and deep neural networks to predict the first-level quality prediction indicators,integrate the prediction results of the DNN weak learner through the Boosting integration idea,and determine whether the solder paste on the printed pad is defective Make predictions.(3)Construct a prediction model for printing defects in SMT production lines.Based on the improved CNN printing defect prediction model,the second-level quality prediction index is predicted.First,feature expansion is performed on the production data where defects occur through feature interaction,and the expanded data is used as the input of the CNN defect prediction model to predict the specific types of defects that occur.(4)Construct a prediction model for the qualified rate of SMT production lines.Based on the XGBoost model,the qualification rate of the SMT production line is predicted.First,the random forest algorithm is used to select the feature of the SMT production line qualification rate,and then the PCA algorithm is used to extract the feature of the SMT production line qualification rate.The feature selection and feature extraction results are used as the input of the XGBoost model to predict the qualified rate of SMT production line products.Through the establishment of multiple comparison models,the actual production data of the production line is used to verify the method proposed in this paper,which proves the effectiveness and practicability of this method.