A robust adaptive exponentially weighted moving average control chart with a distribution-free design strategy

Abstract

The adaptive EWMA control chart can dynamically adjust the smoothing parameter according to the current prediction error. It has good overall performance in monitoring Gaussian processes. However, it often fails under many non-Gaussian processes, and it is hard to select optimal parameters in its design when we do not have enough information about the underlying distribution, especially for heavy-tailed processes. This paper develops a new robust adaptive EWMA control chart, which can discount outliers and achieve excellent detection performance in monitoring most processes for its high robustness in parameter selection, even if the practitioner has little information about the underlying distribution. A nearly optimal and distribution-free parameter design strategy regarding the relative mean index (RMI) is constructed for it. Simulation results show that the chart performs more robustly and is better overall than alternatives in process monitoring. Finally, two real examples are provided to illustrate the broad applicability and superiorities of the proposed control scheme.