Multivariate statistical process monitoring and diagnosis with grouped regression-adjusted variables

A common theme among the many existing multivariate statistical process monitoring (MSPM) methods is the recommendation that process knowledge be used to select a suitable monitoring procedure. Several methods possess the property of directional invariance, with shift detection performance depending only on the distance of a shift away from the target mean vector. This property is of special importance when characterizing a new process, or when available process knowledge suggests that shifts may occur in virtually any direction away from the target mean. In other cases, it is possible and may be desirable to increase a control scheme's sensitivity by using knowledge of the process structure and possible upset mechanisms to "aim" the control procedure. This paper identifies a potentially common MSPM scenario and extends the idea of using process knowledge to determine an appropriate control statistic for assignable cause detection and identification.