A Multi-Chart Approach for Mean Shift Detection

In this paper we consider a multi-chart for detecting a unknown shift in the mean of an identically distributed process. It is shown that the multi-chart has usually two advantages: one is in that it can much reduce computational complexity compared to the GLR (generalized likelihood ratio) and GEWMA (generalized exponentially weighted moving average) control charts when the in-control ARL (average run length) is large; the other is that it can quickly detect the size of the mean shift. Moreover, the numerical simulations show that the multi-chart can not only perform better than its constituent charts which consist of the multi-chart in the sense that the average of the ARLs of the constituent charts is large than that of the multi-chart, but also be superior on the whole to a single CUSUM, EWMA, EWMA multi-chart and GLR control charts in detecting the various mean shifts when the in-control ARL is not large.