Research On Product Quality Risk Prediction Method Based On Quality Data Analysis

With the development of economy,consumers have higher and higher requirements for the quality of electronic products,and enterprises are more and more strict in the quality risk management of electronic products.Nowadays,electrostatic discharge(ESD)is the main cause of quality problems of electronic products.More than 35% of chip damage is caused by ESD,and soft ESD accounts for 90% of ESD damage.The soft damage caused by electrostatic discharge is difficult to capture,the relationship is complex,and the quality risk is difficult to predict.In order to solve the above problems,this thesis studies the quality risk prediction method of electronic products based on the electrical performance test data and electrostatic monitoring data of electronic products.In addition,for imbalanced data,a multi classifier hierarchical integration algorithm is proposed.The main research contents are as follows:(1)The production process and detection process of electronic products are briefly described.The electrostatic damage phenomenon and the mechanism of electrostatic protection monitoring are analyzed in detail.It also combs the original data sets related to quality of the whole production line.On this basis,aiming at the low quality density and extremely unbalanced data set of industrial big data,the overall framework of quality risk prediction of electronic products is designed.It mainly includes three parts: the analysis of electrical performance drift based on statistical analysis and random forest algorithm,the alarm analysis based on electrostatic monitoring data and the construction of product quality risk prediction model based on multi classifier hierarchical integration algorithm.(2)Based on statistical analysis and random forest algorithm,electrical performance drift analysis.Screening important sensitive parameters in electronic products and extracting corresponding test data.By changing the limitation of the drift range,the ratio of the electrical performance drift of sensitive parameters and the prediction accuracy of random forest algorithm for product quality risk are taken as the standard.The index of whether the sensitive parameters have electrical performance drift is determined,so as to achieve the first balance of positive and negative sample data,so as to screen the products with electrical performance drift for subsequent analysis.(3)Alarm analysis based on electrostatic monitoring data.Based on the products with electrical performance drift,through analyzing the electrostatic monitoring alarm data of each station in the production process,the comprehensive alarm index is defined,and the data without alarm is eliminated to complete the second balance of positive and negative sample data.(4)Product Quality Risk Prediction Model Based on Multi-classifier Hierarchical Integration Algorithm.Extracted the products with electric performance drift and static monitoring alarm after the first two steps.The product quality risk prediction model was established based on the electrical performance Test data of FCT(Functional Circuit Test)and ICT(In Circuit Test)and the comprehensive alarm index of each station.To predict whether the product with electrical performance drift of sensitive parameters will fail due to electrostatic soft damage.In the process of building the model,a multi-classifier layered integration algorithm was proposed,which improved the prediction accuracy compared with the single classifier algorithm.Based on the above research content,this thesis completes the optimization of low quality density of industrial big data,and realizes the quality risk prediction of electronic products.The reliability and accuracy of the model are verified by the result analysis and model comparison.The effectiveness of the proposed algorithm model is verified by many field experiments.