Study On Multivariate Process Control Based On Penalized Logistic Regression

In modern manufacturing and service industries,the quality control problems actually are composed of numerous quality characteristics rather than a single one.With the development of science and technology,the data dimension and complexity are getting higher and higher,and the traditional multivariate control charts can no longer meet the practical application requirements.Some researchers have combined the research of big data analysis technology and control chart field,and the proposed control charts' performance are better than the traditional control chart,but research in this area is still insufficient.This thesis proposes a multivariate control chart based on logistic regression model in machine learning.The model is very effective in solving the two-class classification problem.It can not only accurately classify the sample categories,but also obtain sample classification probability.This result can provide more information for the quality status of the sample,which is not achievable by other classifiers;In order to solve the computational complexity caused by the large sample dimension,this thesis chooses to use Lasso regularization to select the model parameters and leave only a few non-zero variables,which can achieve sparse processing of highdimensional data;This thesis also uses the RTC monitoring method to integrate the new observation and historical sample to solve the problem of one-class classification in statistical process control,and improve the accuracy of the classifier and the sensitivity of the control chart.Then,we selected the charting statistic and model parameters through the simulation.The simulation results show that the multivariate control chart based on the penalty logistic regression model proposed in this thesis is superior to the RTC-RF control and D-SVM control proposed by previous researchers.Finally,the new control chart proposed in this paper is applied to the actual production case.The results show that the control chart also has good monitoring performance in practical applications.