THE CUSCORE AND HIGH-DIMENSIONAL CONTROL CHARTS FOR STATISTICAL MONITORING OF AUTOCORRELATED PROCESS DATA
Open Access
- Author:
- Chen, Shuohui
- Graduate Program:
- Industrial Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- September 19, 2006
- Committee Members:
- Harriet Black Nembhard, Committee Chair/Co-Chair
Paul H Cohen, Committee Chair/Co-Chair
Runze (Richard) Li, Committee Member
Ling Rothrock, Committee Member - Keywords:
- adaptive Neyman test
profile monitoring
autocorrelation
multivariate process control
generalized minimum variance control
time series models
statistical process control (SPC)
fault signature
engineering process control (EPC)
Cuscore
high-dimensional (HD) control chart
Fourier transform - Abstract:
- A unified likelihood-based Cuscore control approach is developed to monitor the mean shift in both univariate and multivariate autocorrelated processes. Appropriate statistics based on the fault signatures of the signal is derived for the detection of signals in an ARMA noise process adjusted by the generalized feedback control (GMV) scheme. It is shown that the performance of Cuscore charts is independent of the amount of variability transferred from the output quality characteristic to the adjustment actions in the GMV control system. An example with the background of valve leakage detection in flow control illustrates the application of the approach. In extending the Cuscore control chart to monitor the mean vector of autocorrelated multivariate processes, a bivariate time series model is used to illustrate the theory and application of the MCuscore chart. Simulation and the application example with background in the reactive ion etching (RIE) process show that the MCuscore chart outperforms the traditional residual-based MCusum control chart in detecting a mean vector shift signal in autocorrelated bivariate processes. A high-dimensional control chart approach based on the discrete Fourier transform and the adaptive Neyman test is proposed for monitoring processes or products in which the quality can be characterized by a profile or a function between a response variable and a number of explanatory variables. This control chart approach can be used to monitor both linear and nonlinear profiles with either i.i.d. or autocorrelated stationary noise component. An example with the background in the woodboard manufacturing process illustrates the application of this approach.