Improving the sensitivity of the T-2 statistic in multivariate process control

The T-2 statistic associated with an observation from a multivariate proces s is a function of the residuals taken from a set of linear regressions exi sting among the various process variables. These residuals are contained in the conditional T-2 terms of the orthogonal decomposition of the statistic . It is shown in this paper that a large residual in one of these fitted mo dels can be due to incorrect model specification. By improving model specif ication at the time that the historical data set is constructed, it may be possible to increase the sensitivity of the T-2 statistic to signal detecti on. The resulting regression residuals also can be used to improve the sens itivity of the T-2 statistic to small but consistent process shifts. This i s accomplished using plots that are similar to cause-selecting control char ts.