Chang, Tsong-how (1991, ASQC) University of Wisconsin, Milwaukee, WI
This abstract is an edited version of the author's original.
This paper presents control charts for bivariate and trivariate applications, where the process quality is characterized by two or three correlated variables.
The method of principal components is employed to transform the correlated sample data into mutually independent sample principal components, each of which is a weighted linear function of the original variables. A control chart is then constructed for each sample principal component, and it is interpreted as an ordinary Shewhart chart. The set of two (for bivariate) or three (for trivariate) principal component charts must be analyzed together to detect any out-of-control signals. When examined together, these principal component charts produce a set of unique distributional patterns, one for each type of the shifts in the means of the variables. This feature is shown to have an advantage over some conventional multivariate control charts in that it aids in the identification of the assignable causes as they occur. Tables of these unique patterns are developed for bivariate and trivariate processes to facilitate chart interpretations in practice.
Control charts,Process control,Shewhart control chart,Statistics