00224 q3d0bb694843da3e03139b45685cb12

226

Baxley

Time Order

LECEND - rt-Torget — -Zl

4

2

Xt    .

-2

-4 -L. 0

(b)

X*=.4 g-1.0 o,* 1.0 S,-0, istsioo X_t-.6 g,* 1.5

EWMA Limils* ±Lo_t* ± 1.5il655- i .98

—---1-----1-•---•-1-------1->-----r

20    40    60    80    100

Time Order

Figurę Sb. Simulation of the EWMA Algorithm with Over-Control.

The process adjustments are pre-determined for each nile as shown along the right-hand vertical axis. These adjustments are based on the assumption that when a rule is violated the current process mean is in the involved zonę. There are two significant problems with this approach. First, the modified Shewhart chart is slow to signal a drift away from target as shown in Figurę 6, and there is no convenient way to tune the chart to correct this. Second, and morę importantly, adjustments based on Rule 3 and Rule 4 violations tend to unnecessarily add variability to the product. An example of this is shown on Figurę 7, where a Rule 4 violation occurred even though a morę reasonable estimate of process level was in Zonę B.

Chatto (1989) compared the performance of the modified Shewhart and EWMA algorithms in his simulation study. His study differs from the previous sections of this paper in that he doesn’t use an IMA disturbance