Directional MEWMA Schemes for Multistage Process Monitoring and Diagnosis

Abstract

Statistical process control, monitoring, and diagnosis of a multistage process remain challenging problems in both the manufacturing and service industries. This paper proposes a directional multivariate exponentially weighted moving average (directional MEWMA) scheme integrating the EWMA scheme with the generalized likelihood ratio test (GLRT) scheme that can incorporate directional information based on the multistage state-space model and effectively monitor the process mean shift. The proposed directional MEWMA scheme not only provides a statistical process control (SPC) solution that incorporates both interstage and intrastage correlations but also resolves the confounding issue in monitoring and diagnosis due to the cumulative effects from stage to stage. In addition, a systematic diagnostic approach is provided to isolate and identify an out-of-control stage and to locate its change point. Our simulation results show that the proposed monitoring and diagnostic scheme consistently outperforms almost all existing approaches to multistage processes. A sensitivity analysis and discussion on performance indicators for multistage process monitoring are also presented.

• Average Run Length
• EWMA Charts
• Generalized Likelihood Ratio Test
• Multivariate Control Charts
• Recursive Residuals

Additional information

Notes on contributors

Changliang Zou

Mr. Zou is a doctoral student in the LPMC and School of Mathematical Sciences. His email is [email protected].

Fugee Tsung

Dr. Tsung is a Professor in the Department of Industrial Engineering and Logistics Management. He is a Senior Member of ASQ. His email is [email protected].