A Study of Monitoring Non-normal Multivariate Process Using Support Vector Machine

摘要

This paper focuses on the monitoring techniques in multivariate processes when the underlying distribution of the quality characteristics differs from normality. Hotelling T~2 control chart is the most common used control chart for multivariate process, however, it is based on the assumption of normality. Normality assumption is not always reasonable. Researchers gradually applied support vector machine (SVM) to monitor non-normal multivariate process. By using SVM, the selection of parameters in SVM will affect the classification accuracy of SVM. It is an important issue of choosing the SVM parameters. The purpose of this research is to apply SVM in statistical quality control. By simulating bivariate t and bivariate gamma distributions, we study the relationship between the distributions and parameters of SVM to obtain the best classification rate. After obtaining the proper parameters of SVM, we will applied SVM to construct a control chart to monitor non-normal process mean and study the performance of the new chart.