14 SHARAD S.PRABHU AND GEORGE C RUNGER TABLE 4.Worst-State Average Run Lengths for Multivariate EWMA Control Schemes (H=15.1g元CL=±0.89 =±1.16) (H=40.s220 =士1.47 ZRaeARorH 3. 107.05 1424 220 14 088 464 115 557 二4 Note:ARLo=500,r=0.1,and the vertical location is zero for all cases. ieAaRwAoeo 4-VI Example Although.the matrix Po is the transition matrix etween all the transient state sults in Prabhu and Runger (1996).the large.two 0.7 88 0. 03 82 dimensional transition matrix can again be avoided 01 but slightly less than,the results in Table 1.When That is.thre and the EWM a state lations betweens the two sets are low to medium. the ARL i If the proce sonable choice for this size shift.Ther Now when the lewMA is in u state near a con 17.51 provides ARLo=200 and the (zer can be degra dodane 3 If the proce shifts tou(111 000)ther are considered in Table 4.By checking each state 6=('∑-1)/2=1.21.From Table1,r=0.1o 0.2 are reasonable choices.The performance can the les. e AR andr=0.1.In all cases in the table,the vertical is easily determined.If u is changed to u then changes to k6.for any constant k. location of the worst state is the zero axis.The hor Conclusions creases,the worst state is closer to-UCI This is pected becaus Our results indicate that good choices forrdepend e as d increa es.the probab ility tha in the cont state is nearer UCL.Table also provides the cor- teady-state analys not diferent responding zero-state ARLs for comparison.Similar than those for zero-state. The worst-state results Vol.29,No.1.January 1997 Reproduced with permission of the copyright owner.Further reproduction prohibited without permission.Reproduced with permission of the copyright owner. Further reproduction prohibited without permission