Application of Support Vector Machine for Diagnosing the Out-of-Control Signal in Multivariate Mean Vector and Covariance Matrix

• Main Authors: Yun Tai, 戴筠
• Other Authors: Chuen-Sheng Cheng
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/35273226574094795943

碩士 === 元智大學 === 工業工程與管理學系 === 101 === Control charts are a useful tool in attaining process stability. It is helpful in determining whether a process is behaving as intended or if there are some unnatural causes of variation, corresponding to out-of-control situations. In multivariate process control, Hotelling T2 control chart and generalized variance, S , control chart are often used to monitor mean vector and covariance matrix of multivariate process, respectively. Out-of-control signals in multivariate control charts may be caused by one or more variables. Therefore, a diagnostic method is needed to determine the source of out-of-control signal. The diagnostic task involves the identification of shifted variable(s) responsible for the signal. The correct identification of shifted variable(s) might provide detailed diagnostic information and facilitate the planning of corrective action. In this study, we consider the case of monitoring the process mean vector and covariance matrix simultaneously. The diagnosis of out-of-control signal is formulated as a classification problem. Once an out-of-control signal is generated by T2 control chart or S control chart, a support vector machine (SVM) based classification system is applied to identify the shifted variable(s) and determines either mean or variance has shifted. We propose a set of relevant features to further enhance the classification ability of SVM. Extensive simulation study and comparison with previous research have demonstrated the superiority of the proposed approach. Several examples are used to illustrate the implementation of the proposed method.