Robust on-line monitoring for univariate processes based on two sample goodness-of-fit test

On-line monitoring of quality variables and data streams raises attention in fields of quality management and statistical process control. Though much work has been dedicated on it in the literature, some challenges associated with designing distribution-free on-line control schemes are yet to be addressed well. This paper proposes a new distribution-free control chart for detecting both mean shifts and variance shifts in a univariate process based on nonparametric two sample goodness-of-fit test. It also integrates the self-starting feature by using data-dependent control limits set on-line instead of predetermined limits. Simulation study shows that the chart has satisfactory in-control run length performance given desired ARL and robust detection capability for general out-of-control changes, and is especially useful in short-run processes or start-up stages.