A control chart for a general Gaussian process

It has been repeatedly demonstrated that X-bar quality control charts perform poorly when the process subgroups being monitored are correlated. In this paper, we propose and investigate the performance of a control chart that accounts for subgroup correlations in a general Gaussian process. The time-series innovations algorithm is used to construct the desired chart from a set of one-step ahead predictions and prediction variances. The chart is applicable in both stationary and nonstationary settings. A simulation study shows that this 'innovations' chart performs as a traditional X-bar chart even when the correlation structure of the process must be estimated from a small number of subgroups. The innovations chart is then used to study a data set of motor shaft diameters which has correlated subgroups. The results here show that erroneous conclusions can be reached if subgroup correlations are ignored. (C) 1998 Elsevier Science B.V. All rights reserved.