An adaptive Generalized Likelihood Ratio control chart for detecting unknown patterned mean shifts

Traditional statistical process control techniques have been focused on detecting a constant mean shift. In practice, the mean of the monitored variable can exhibit a time-varying behavior after the fault. When the mean shift pattern is known in advance, Generalized Likelihood Ratio (GLR) and Cumulative Score (CUSCORE) control charts can be used for change point detection. Further, some adaptive CUSCORE schemes have been recently developed for detecting an unknown patterned mean shift. However, these adaptive CUSCOREs assume that either an upward or downward shift is present. Hence, their performance can be poor in the case of an oscillatory mean behavior. In the talk, under a normal assumption, we describe a novel adaptive GLR control chart. This method first estimates the dynamic mean using a combination of EWMA and wavelet smoothing, then uses the estimates in a conventional GLR test. A self-starting version and an extension to autocorreleted data are also discussed. Simulations demonstrate the efficiency of the proposed scheme for a wide variety of mean patterns.