Research On Quality Prediction Of Single Small Batch Product Based On Copula-Lasso

Intelligent manufacturing mode can balance the high cost and high added value of personalized customization production,and meet the personalized requirements of consumers,which greatly promotes the development of personalized customization production mode.The production mode of personalized customization often presents the characteristics of multiple varieties,small batch,or even single piece production,which leads to the data characteristics of product quality characteristic value presenting as a few samples or a single sample.Therefore,the traditional statistical process quality control can not be used in the quality assurance of single small batch production.Therefore,this paper explores the quality assurance of single small batch product processing.In the intelligent manufacturing mode,the intelligent manufacturing system can record the processing quality characteristics and other data in the process of equipment processing,and the networked distributed manufacturing system based on "cloud manufacturing" can facilitate the storage and sharing of equipment quality data.With the help of the quality data of the equipment,this paper puts forward the quality assurance method based on the prediction in the processing of single small batch products.That is to say,based on the equipment quality data,a quality prediction model is established for the alternative processing equipment to predict the quality characteristics of the single product processed by the equipment in the future.Through the automatic selection of the optimal processing equipment,the quality assurance of single small batch product processing is realized.At the same time,based on the characteristics of mass,high dimension and low data value of equipment quality data set,this paper constructs a single small batch product quality prediction model based on Copula-lasso under the integrated feature selection.In order to deal with the problems of over fitting and low efficiency caused by high-dimensional features,this model firstly selects three different and complementary methods of Pearson correlation coefficient,maximum information coefficient(MIC)and rrelieff algorithm to reduce the dimension of features.Then,the binary copula function is selected to better represent the complex nonlinear correlation between each characteristic variable and quality characteristic variable,and the nonparametric prediction model of binary copula regression is constructed.Finally,in order to improve the performance of the model,the least absolute shrinkage and selection operator(Lasso)is used to learn the prediction results of multiple binary copula regression models and the true value data of quality characteristics,so as to obtain the appropriate weights for the fusion of binary copula regression models.The model is simulated on the actual production data of semiconductor industry and compared with multiple linear regression,Lasso regression and support vector regression(SVR).The experimental results show that the quality prediction model based on Copula-Lasso is superior to other prediction models in terms of mean square error and mean absolute percentage error,and has strong generalization ability.This study provides new ideas and specific methods for the quality assurance of single small batch products,and promotes the development of intelligent manufacturing to a certain extent.