Most of the products produced today are the results of several process stages. With the emphasis in industry on improved quality, control charts are widely used for process monitoring. However, conventional SPC techniques focus mostly on individual stages in a process and do not consider disseminating information throughout the multiple stages of the process. They are shown to be ineff'ective in analyzing multistage processes. A different approach to this problem is the cause-selecting chart (CSC), proposed by Zhang (1980, 1982, 1984, 1985a, 1985b, 1989a, 1989b, 1992). The CSC based on the output adjusted for the effect of the incoming quality shows promise for increasing the ability to analyze multistage processes. It starts a new field of control charting and much more work is require...[ Read more ]

Most of the products produced today are the results of several process stages. With the emphasis in industry on improved quality, control charts are widely used for process monitoring. However, conventional SPC techniques focus mostly on individual stages in a process and do not consider disseminating information throughout the multiple stages of the process. They are shown to be ineff'ective in analyzing multistage processes. A different approach to this problem is the cause-selecting chart (CSC), proposed by Zhang (1980, 1982, 1984, 1985a, 1985b, 1989a, 1989b, 1992). The CSC based on the output adjusted for the effect of the incoming quality shows promise for increasing the ability to analyze multistage processes. It starts a new field of control charting and much more work is required on this subject.

In practice, the model relating input and output measures often needs to be estimated before the CSC is implemented. Little is known about the performance of the CSC when the model parameters are estimated. In this thesis, the effect of parameter estimation is investigated. To get a better understanding of the performance of CSCs with estimated parameters, their run-length distributions are analytically derived. A numerical procedure based on Gaussian quadrature is used to evaluate the run-length distribution.

The simple linear regression model widely discussed in the CSC is insufficient to capture the stochastic behavior of the output. Taking the process dynamics and the autocorrelation structure of the disturbance into account, a more realistic CSC model is described. Like the conventional residual-based charting methods, the autocorrelation is removed by filtering the output with an inverse filter. However, by doing so, the resulting mean shift in the residual is varying over time, which has been referred to as the fault signature. In an attempt to make use of the valuable information contained in the fault signature, the cumulative score (Cuscore) chart and the triggered Cuscore chart are proposed. It is shown that the triggered Cuscore chart performs better than the standard Cuscore chart and the residual-based CUSUM chart.

The multiple CSC (MCSC) is more adaptable than the CSC, which deals with the case with multiple uncontrollable assignable causes. The design and implementation of the MCSC is discussed in this thesis when the model parameters are estimated. Two estimation procedures are considered: the least squares estimation and principal components regression (PCR). It is shown that using prediction limits is quite effective in terms of maintaining a desired false-alarm rate under both procedures. It is expected that this research will greatly expand the scope of Conventional quality research.