| The objective of this research is to develop techniques for a multivariate statistical process monitoring scheme when the usual normality assumptions are not met. In particular, this study considers a situation where the process variables consist of a mixture of normally and non-normally distributed variables and the presence of unusual observations. An application of the generalized linear model (GLM) to a model-based control strategy is used to accomplish this objective. Instead of ordinary least squares regression, GLM is used in the regression adjustment scheme. Deviance residuals are used to detect process upset. A robust GLM and a robust deviance is developed and applied to the regression adjustment scheme. This modification is expected to enhance the ability of regression adjustment to perform effectively in an environment where there are both normally and non-normally distributed variables with the presence of unusual observations. A Monte Carlo simulation reveals that this proposed method can detect the mean shift more quickly than the Shewhart control chart for individual responses and the T2 chart based on the U statistic in both a single- and a multiple-stage process. |
