Abstract
Multistage process monitoring and fault identification are currently receiving considerable attention. This article focuses on detecting common faults in a multistage process that affect the process covariance matrix. The process covariance matrix monitoring problem is formulated into a multiple hypotheses testing problem. The proposed method is an exponentially weighted moving average chart built on vectors that are transformed from sample covariance matrices of the collected observations. Extensive simulation analysis shows that, compared to alternative methods for multistage process covariance monitoring and diagnosis, the proposed method is capable of not only detecting variation changes quicker but also identifying faults with higher accuracy.
Acknowledgements
The authors deeply thank the editor, the department editor, and the anonymous referees for their many constructive comments that have significantly improve the quality of the article. The present study was co-funded by Project National Natural Science Foundation of China 70902070, 50875168 and 51075277. This research was also supported by the RGC Competitive Earmarked Research Grants 620707 and 620010.