New measures of statistical performance of process control charts
George Nenes
Abstract
George Nenes
Abstract
Traditional measures of statistical performance of SPC control charts are widely used to evaluate the performance of a control chart. Pretty often, even if the control chart is designed from an economic point of view, statistical properties are set as necessary conditions (constraints), to control performance. The usual statistical measure used to evaluate the performance of a chart is either the false alarm rate, α, or the equivalent average run length, ARL0. Usually, the condition is that α should not exceed 0.27% (probability of false alarm for a Shewhart chart with k = 3), or, that the ARL0 (=1/α), should be no less than 370. This paper revisits the issue of statistical properties of a chart, shows that no α constraint suffices to rationalize the performance of the control procedure, in terms of limiting the false alarms to an acceptable level, and proposes new measures of statistical performance that can be used for most commonly used control charts. In addition, a by-product of the paper is a generic model development that can be used for the computation of statistical (and economic) measures of performance for control charts, even for the more advanced, adaptive control charts.
OpenAlex reports 1 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Traditional measures of statistical performance of SPC control charts are widely used to evaluate the performance of a control chart. Pretty often, even if the control chart is designed from an economic point of view, statistical properties are set as necessary conditions (constraints), to control performance. The usual statistical measure used to evaluate the performance of a chart is either the false alarm rate, α, or the equivalent average run length, ARL0. Usually, the condition is that α should not exceed 0.27% (probability of false alarm for a Shewhart chart with k = 3), or, that the ARL0 (=1/α), should be no less than 370. This paper revisits the issue of statistical properties of a chart, shows that no α constraint suffices to rationalize the performance of the control procedure, in terms of limiting the false alarms to an acceptable level, and proposes new measures of statistical performance that can be used for most commonly used control charts. In addition, a by-product of the paper is a generic model development that can be used for the computation of statistical (and economic) measures of performance for control charts, even for the more advanced, adaptive control charts.
Key concepts: Control chart, Statistical process control, Shewhart individuals control chart, Computer science, Chart, EWMA chart, \bar x and R chart, Statistics