Support Vector Data Description Based Multivariate Exponentially Weighted Moving Average Control Chart

For many years, Statistic Process Control is the main method of the qualitymanagement. In the initial stage, the univariate control charts were widely used. Thenthey could hardly meet the requirement with times past by. As a result, themultivariate control charts were proposed. Those control charts are almost restrictedby the distribution of the dataset and limited in use. With the development of scienceand technology, machine learning brought the new dynamic to the control charts. Theclassification methods of machine learning are combined with the control charts suchas artificial neuro network, decision tree and support vector machine (denoted asSVM). However, there are also weaknesses of these methods. Some scientistsproposed support vector data description (denoted as SVDD) on the basis of SVM. Itis a one-class classification method with all the advantages of SVM, suitable for thesmall samples and high dimensional problems. It can use the normal data only and beappropriate for the unbalanced condition. In this paper, we do research on this methodand build a SVDD-based MEWMA (denoted as S-MEWMA) control chart. By doingsimulation experiences, we decide the values of parameters and discuss another newmethod about confirming them. Then we compared S-MEWMA control chart withanother SVDD-based one, and it is called D2chart. Simulation results show that theS-MEWMA chart outperforms the D2control chart no matter whether the processfollows a normal or non-normal distribution.