00016 Üeb564a5f5b77c28047d13120a3d2db

15

An Overview and Perspective on Control Charting

Types of Charts

As Box and Rami'rez (1992) point out, the type of chart to use depends on the expected behavior when the process is affected by an assignable cause. The Shewhart chart is most appropriate for large shocks or shifts, measured relative to the standard error of the plotted statistic, and for detection of assignable causes resulting in unexpected pattems. Cumulative sum (CUSUM) charts [see Lucas (1976) and Woodall and Adams (1993)] and exponentially weighted moving average (EWMA) charts [see Crowder (1989) and Lucas and Saccucci (1990)] have much better performance in detecting smali sustained shifts in the level of a process. The use of runs rules with Shewhart charts, however, narrows this performance gap [Champ and Woodall (1987)]. The zonÄ™ control chart of Jaehn (1991) is very similar to the Shewhart chart with runs rules, but is claimed to be somewhat easier to use. Runger and Pignatiello (1992) and Kenett and Zacks (1992) propose new approaches that employ changepoint techniÄ…ues to estimate when a detected process change took place.

Much research effort has been devoted to developing charts that can morÄ™ Ä…uickly detect shifts that are smali relative to measurement and process variability. In some applications the increased sensitivity may not be desirable because it is either not possible or not economical to remove the corresponding assignable causes. This situation is analogous to using an hypothesis test with unnecessarily high power. In such cases one needs to desensitize the control chart by, for example, widening the standard 3-sigma Shewhart control limits or increasing the CUSUM reference value. These issues are discussed by Woodall (1985).

Modified and acceptance control charts have control limits based on specification limits and allow the possibility of a specified proportion of nonconforming items to be produced by a highly capable process. We consider the routine use of these charts to be inconsistent with the philosophy of continuous improvement because they allow increased variability about the target value. Montgomery (1991, pp. 317-320) describes modified and acceptance charts in detail. Wheeler (1991b) advises strongly against their use and Deming (1986, p. 369) refers to such charts as "bear traps."

Short-Run Methods

In short production runs little data is available to obtain tentative estimates of parameters necessary to construct usual control chart limits. Thus, most short-run charting methods detect significant deviations from a specified target value as opposed to a lack of statistical control. See, for example,