J ournal of K or ean
D a ta & I nf orma t ion S cience S ocie ty 2 002 , Vol. 13, N o.1 p p . 105~ 112
B iv ari at e E W M A C o n t ro l Ch art s f o r A u t o c o rre lat e d P ro c e s s e s
Gy o - Y ou n g Ch o 1) an d Y oun g S un A h n 2 )
A b s t ra c t
In t his p aper w e e st ablish biv ar iat e ex pon en tially w eig ht ed m ov in g av er ag e (E W M A ) cont r ol ch ar t s for aut ocor r elat ed pr oces ses u sin g r esidu al v ect or s . W e fir st deriv e t h e r esidu al v ect or s , th eir ex pect ation , v ar ian ce - cov ar ian ce m atr ix , t h en ev alu at e t h e cont r ol ch ar t b a sed on t h e av er ag e r un len gt h (A RL ).
K ey w ords : Biv ariat e EWMA Control Chart , Residual Vect ors , ARL
1. In trodu c tion
In st at istical pr oce s s cont r ol, it is u su ally a s su m ed on t h e pr oces s ou tput at differ en t t im es ar e IID . H ow ev er , for m any pr oces s es t h e ob ser v at ion s ar e cor r elat ed an d con tr ol ch ar t s for m onit or in g t h e se pr oces s es h av e r ecen t ly r eceiv ed m u ch at t ent ion . M any aut h or s su g g est ed fitt in g on appr opr iat e t im e ser ie s m odel t o t h e pr oces s an d u sin g r esidu als a s con tr ol st at istic on a con tr ol ch ar t (A br ah am an d K ar th a (1979 ), A lw an an d Rob er t s (1988 ), Y our st on e an d M ont g om er y (1989 ), H ar r is an d Ros s (1991), B ox an d R am ir ez (1992), L on gn eck er an d Ry an (1992), Y a schin (1993 ), W ar dell et al.(1993 ), an d Run g er et al.(1995 )).
T h er e ar e m an y sit u ation s in w h ich t h e sim ult an eou s con tr ol of t w o or m or e r elat ed qu alit y ch ar act er ist ics is n eces sar y so, m any aut h or s su g g est ed m ult iv ar iat e con tr ol ch ar t s (A lt (1985 ), J ack son (1985 ), L ow r y an d M ont g om er y (1995 ), M a son et al (1997 )). L ow r y et al (1992) stu died m u lt iv ar iat e E W M A ch art un der pr oces s
1. A ssociat e Profes sor , Departm ent of st atistics , Kyun gpook National Univ ersity , Daegu , 702- 701, Korea
E - m ail : gycho@knu .ac .kr
2. P art - tim e In struct or , Departm ent of St atistics , Kyungpook Nation al Univer sity , Daegu ,
702- 701, Kor ea
out pu t at differ ent tim es ar e IID .
A u t ocor r elat ion is quit e com m on in in ch em ical an d pr oces s in du st r ies an d oft en sev er al pr oce s s v ar iable s ar e m ea sur ed an d st or ed sim u lt an eou sly . Run g er (1996 ), M a son et al (1996), Bak shi (1998 ), N egiz an d Cin ar (1998) st u died m u lt iv ar iat e cont r ol ch ar t s for aut ocor r elat ed pr oces s es .
Ou r obj ectiv e is t o ev alu at e th e pr op er ties of m u lt iv ar iat e E W M A ch ar t w h en t h e r esidu al v ect or s ar e u s ed t o m on it or an aut ocor r elat ed pr oces s . A sim ple v ect or tim e s er ies m odel is u sed h er e t o r epr esen t t h e ob ser v at ion s fr om an aut ocor r elat ed pr oces s .
2 . M o de lin g an d Controllin g th e P roc e s s
2 .1 V e c t or T im e S e rie s P ro c e s s M e as u re d w ith R an dom Error V e c t or
In t his section , a v ect or t im e ser ie s pr oces s w it h r an dom er r or v ect or w ill b e dev eloped . L et
x
t=
t+
t, (2.1)
w h er e
tis t h e ob s er v at ion v ect or at tim e t,
tis th e pr oce s s m ean v ect or at t im e t, an d
tis t h e m ea sur em en t er r or v ect or at t im e t.
It is also a s su m ed t h at
tcan b e descr ib ed a s a v ect or fir st or der aut or eg r es siv e (v ect or A R (1)) pr oces s w it h pr oces s lev el , t h at is (Box an d J en kin s (1976 )),
t
= ( - ) +
t - 1+
t, (2.2)
w h er e
t,, t h e r an dom sh ock v ect or at t im e t, is a s su m ed t o b e m u lt iv ariat e n or m ally dist r ibu t ed m ean v ect or 0 an d v ar ian ce - cov an an ce m at rix , in dep en den t of m ea sur em en t er r or v ect or , an d in dep en den t of t h e r an dom sh ock v ect or at any ot h er t im e.
T h e v ect or A R (1) pr oce s s ob s er v ed w it h m ea su r em ent er r or v ect or is equ iv alen t t o a v ect or fir st - or der an t or eg r e s siv e fir st - or der m ov in g av er ag e (v ect or A RM A (1,1)) pr oces s (Box an d J en kin s (1976 )). T h e A RM A (1, 1) pr oces s can b e w r itt en a s
( - )
t= ( - ) + ( - ) a
t ,(2.3)
w h er e B is a b a ck shift op er at or su ch t h at B x
t= x
t - 1,an d a
tis u n cor r elat ed
an d m ult iv ar iat e n or m ally distr ibut ed w ith m ean v ect or 0 an d v ar ian ce - cov ar ian ce m at r ix a an d is t h e m ov in g - av er a g e par am et er m at r ix an d is t h e au t or egr es siv e par am et er m atr ix .
2 .2 S t e p Ch an g e in th e P ro c e s s Lev e l
W h en t h e pr oces s is in - con tr ol ( =
0) , m in im um M S E for eca st v ect or m ade at tim e t for t + 1 is
x
t=
0+ ( x
t-
0) - e
t ,w h er e e
t= x
t- x
t - 1= x
t-
0- ( x
t - 1-
0) + e
t - 1an d e
tis th e r e sidu al v ect or at tim e t. S u ppose t h at th er e is a st ep ch an g e fr om
0t o
1in t h e pr oces s lev el b et w een t im e t = k - 1 an d k . T h en th e pr oces s can b e w r it t en a s
x
t= {
0-
0+ x
t - 1+ a
t- a
t - 1, t = k - 1, k - 2 ,
1
-
0+ x
t - 1+ a
t- a
t - 1, t = k
1
-
1+ x
t - 1+ a
t- a
t - 1, t = k + 1, k + 2 ,
an d t h e ex pect ation of th e ob s er v at ion v ect or x
tis
E ( x
t) = {
0 ,t = k - 1, k - 2 ,
1 ,
t = k , k + 1, k + 2 , T h er efor e t h e r esidu al v ect or s ar e
e
t= { ( [ ( I + a
t,
1- +
0) +
2 +a
t ,+
l - 1)( I - ) +
l](
1-
0) + a
t,t< k t = k t = k + l, l = 1, 2 , 3 , S in ce a
tis w hit e n oise, w e h av e
E ( e
t) = { 0 , t = k - 1, k - 2 ,
1