A SPC Procedure for Detecting Level Shifts of Autocorrelated Processes

Recent advances in manufacturing automation make the collection of data for gauging process condition fast and economical. In most cases, the classical statistical process (SPC) control procedures cannot be directly implemented due to the inherent autocorrelation in the data series. A common approach to monitoring autocorrelated processes is to apply the classical SPC techniques on the residuals of a chosen autoregressive moving average model. However, the sensitivity of the residual-based SPC procedures in detecting process shift deteriorates when the process is highly positively autocorrelated. In this paper, we propose the application of the statistics used for detecting outliers and level shifts in time series for process monitoring. Focusing on level shift detection and using a first order autorregessive (AR(1)) model with the average run length as the criterion for comparing the performance of control charting procedures, we show that the proposed charting scheme has a superior performance in detecting level shifts. The proposed scheme can easily be extended to effectively detect the presence of additive and innovational outliers.