• 검색 결과가 없습니다.

Multivariate Cumulative Sum Control Chart for Dispersion Matrix

N/A
N/A
Protected

Academic year: 2021

Share "Multivariate Cumulative Sum Control Chart for Dispersion Matrix"

Copied!
9
0
0

로드 중.... (전체 텍스트 보기)

전체 글

(1)

2 002 , V ol. 13, N o.2 p p . 21~2 9

M u ltiv ari at e Cu m u lativ e S u m Con tro l Ch art for D i s pe rs ion M atrix 1 )

D uk - Jo o n Ch an g 2 ) , Jae - K y o un g S hin 3 )

A b s trac t

S ev er al differ en t cont r ol st at istics t o sim ult an eou sly m onit or disper sion m atrix of s ev er al qu alit y v ariables ar e pre sen t ed sin ce differ en t cont r ol st at ist ics can b e u sed t o de scrib e v ariability . M ultiv ariar e cu m ulativ e su m (CU S UM ) con tr ol ch art s are pr oposed an d t h e perform an ces of t h e pr op os ed CU S U M ch art s ar e ev alu at ed in t erm s of av er ag e ru n len gt h (A RL ). M ultiv ariat e S h ew h art ch art s ar e also pr op osed t o com p ar e th e pr opert ies of th e pr opos ed CU S UM ch art s . T h e n um erical r esult s sh ow th at m u lt iv ariat e CU S UM ch art s ar e m ore efficien t t h an m ultiv ariat e S h ew h art ch art s for sm all or m oderat e shift s . A n d w e also foun d t h at sm all r efer en ce v alu e of t h e CU S U M ch art is m or e efficien t for sm all shift .

K e y w o rd s : cont r ol st atist ic , ex pect ed t im e t o sign al, false alarm

1 . In tro du c ti on

Con t rol ch art is a r ep etit iv e st at istical pr ocedu re for cont in u ou sly m on it orin g p ar am t er s of t h e pr odu ction pr oces s , an d is also u s ed t o quickly det ect w h en th e pr oces s ch an g es an d elim in at e a s sign able cau s es th at m ay pr odu ce an y det erior at ion in t h e qu alit y of t h e m anu fa ctu r ed pr odu ct s . T o m aint ain a cont r ol ch art , s am ple s of size n ar e t ak en at r eg ular sam plin g t im e int erv al of len gt h d . T h e ab ilit y of a cont r ol ch art is det erm in ed by t h e len g t h of t im e r equ ir ed for th e ch art t o sig n al w h en th e pr oce s s is ou t - of- cont r ol, an d th e r at e of false alarm

1. T his r es earch is fin ancially s upport ed by Changw on Nat ional Univ er sity in 2000.

2. Pr ofes s or , Dept . of St atist ics , Changw on National Univer sity , Changw on , 641- 773, Kor ea.

E - m ail : djchan g @s ar im .changw on .ac.kr

3. Ass ociat e Profess or , Dept . of St atistics , Changw on National University, Changw on, 641- 773,

Korea.

(2)

w h en th e proces s is in - cont r ol. T h er efor e a g ood ch art sh ou ld qu ick ly det ect ch an g es in t h e produ ct ion proces s w hile pr odu cin g few fals e alarm s .

T h e b a sic S h ew h art ch art , alth ou gh sim ple t o u n der st an d an d apply , u s es on ly t h e in form ation in t h e curr en t s am ples an d is th u s r elat iv ely in efficien t in det ect in g sm all ch an g es on t h e pr oces s par am et er s . M odificat ion s su ch a s t h e u s e of supplem ent ary ru n s ru le s m ay im pr ov e th e efficien cy of S h ew h art ch art s om ew h at , bu t ot h er appr oach es w hich t ak e full adv ant ag es of th e in form ation in p a st sam ples ar e n eeded .

T h e m ultiv ariat e procedur e t o qu ality cont r ol w a s fir st in t rodu ced by H ot ellin g (1947 ) an d b ecam e popu lar in r ecen t y ear s . J ack son (1959 ) an d Gh ar e an d T or g er sen (1968) pr esen t ed m ultiv ariat e S h ew h art ch art b a sed on H ot ellin g ' s T

2

st atist ic. W oodall an d N cu b e (1985 ) ex t en ded t h e un iv ariat e CU S UM pr ocedur e t o t h e m ult iv ariat e ca se for m on it orin g m ean v ect or of qu ality v ariable s . T h ey oper at ed p t w o - sided un iv ariat e CU S UM sch em es sim ult an eou sly , an d ev alu at ed t h e p erform an ce of t h e collect ion of t h e s ch em e. Cr osier (1988 ) an d P ign atiello an d Run g er (1990) con sider ed n ew m u lt iv ariat e CU S UM pr ocedur es t h at accu m ulat e pa st s am ple in form at ion for each p ar am et er an d t h en form a u niv ariat e CU S UM st atist ic fr om t h e m ultiv ariat e dat a for m on it orin g t h e m ean v ect or .

Up t o th e pr esen t , m ultiv ariat e cont r ol pr ocedur es h av e b een w idely u sed for m onit orin g pr oces s m ean v ect or . But , r elat iv ely lit t le at t ent ion h a s b een giv en t o t h e u se of CU S UM ch art s for m onit orin g disper sion m at rix . In t his paper , w e pr opose sev er al differ en t cont r ol st at ist ics an d CU S UM ch art s for m on it orin g t h e disper sion m at rix of correlat ed qu ality v ariab les w h er e th e t ar g et pr oce s s m ean v ect or r em ain ed k n ow n con st an t .

2 . E v alu atin g Con trol S t ati s ti c s

T h e qu alit y of a pr odu ct is oft en ch ara ct erized by j oint lev els of sev er al qu alit y ch ar act eristics . B ecau se it is in appr opriat e t o u se in div idu al ch art s t o det ect any ch an g es of each pr oces s p ar am et er s in th is ca s e, a m ultiv ariat e qu ality con t rol pr ocedu re for sim ult an eou sly m onit orin g correlat ed v ariables is n eeded . S in ce th e a s sign able cau ses w h ich pr odu ce r an dom v ariat ion of th e qu alit y ar e t r eat ed a s inh er ent t o th e pr oces s in it s curr en t st at e, t h e se qu ality ch ar act erist ics ar e r an dom v ariables .

A s sum e t h at t h e pr odu ct ion pr oces s of int er est h a s p ( p 2) correlat ed qu alit y

v ariables r epr esent ed by t h e r an dom v ect or X = ( X

1

, X

2

, X

3

, , X

p

) ' an d w e

obt ain an in depen dent sequ en ce of ran dom v ect or s X

1

, X

2

, X

3

, , w h er e th e

v ect or X

i

= ( X '

i1

, , X '

in

) ' is a s am ple of ob serv at ion s at each sam plin g t im e

i an d X

ij

= ( X

ij 1

, X

ij2

, , X

ijp

) ' . T h e u n derly in g prob abilit y dist ribu t ion of th e p

qu alit y v ariables is a s sum ed t o b e m ultiv ariat e n orm al distribut ion w it h m ean

(3)

v ect or an d disp er sion m at rix .

In pr act ice , it m ay b e n eces s ary in m any ca s es t o e st im at e b oth an d from t h e pa st dat a , but for sim plicity w e a s su m e th at

0

an d

0

ar e kn ow n . Let

0

= (

0

,

0

) b e th e k n ow n t arg et v alu es for th e pr oces s par am et er s of p qu alit y ch ar act eristics an d

0

is r epr esen t ed a s

0

=

10 20

p0

an d

0

=

2

10 120 10 20 1p0 10 p0

120 10 20 2

20 2p0 20 p0

1p0 10 p0 2p0 20 p0

2 p0

,

w h er e t h e t arg et cov arian ce com p on ent of X

r

an d X

s

is

rs0

=

r s0 r0 s0

for r , s = 1, 2 , , p .

In u niv ariat e ca se , t h e pr oce s s disp er sion can b e m on it ored by S

2

ch art or r an g e ch art , w h er e S

2

den ot e an u nbia sed sam ple v arian ce for a r an dom sam ple of size n fr om a pr oces s . T h e S

2

ch art sig n als for lar g e v alu es of S

2i

or equ iv alen t ly for lar g e v alu es of T

i

= ( n - 1) S

2i

/

20

w h er e

20

is t ar g et v alu e of t h e pr oces s disper sion

2

an d S

2i

is obt ain ed at sam plin g t im e i . W h en t h e pr oces s is in - con t rol, th e st atist ic T

i

h a s a ch i- squ ar ed dist rib ut ion w it h ( n - 1) deg r ees of fr eedom .

F or m u lt iv ariat e ca s e, on e pos sible m u lt iv ariat e v er sion of T

i

is V

i

=

n

j = 1

( X

ij

- X

i

) '

0- 1

( X

ij

- X

i

) = tr ( A

i 0- 1

) (2.1)

w h er e A

i

=

n

j = 1

( X

ij

- X

i

) ( X

ij

- X

i

) ' an d t h e p p s am ple disper sion m at rix

S

i

is A

i

/ ( n - 1) . W h en th e pr oces s is in - cont r ol, th e disp er sion m atrix is

0

an d th e cont r ol st atist ic V

i

h a s a ch i- squ ar ed dist ribu tion w it h ( n - 1) p degr ees of freedom . H ot ellin g (1947 ) pr op osed th at th e st at ist ic V

i

can b e u sed t o m on it or t h e pr oces s disp er sion m at rix of p qu alit y v ariables .

T h e g en er al m ultiv ariat e st atist ical qu ality cont r ol ch art can b e con sider ed a s a r ep et it iv e t est s of sign ifican ce w h er e each qu alit y ch ar act erist ic is defin ed by p qu alit y v ariables X

1

, X

2

, , X

p

. T h er efor e, w e can obt ain an ot h er ch art st at istic for m om it orin g b y u sin g th e lik elih ood r atio t e st (LRT ) st atist ic for t est in g H

0

: =

0

v s H

1

:

0

w h er e t ar g et m ean v ect or of th e qu alit y v ariables

0

is k n ow n . T h e r eg ion s ab ov e t h e U CL corr esp on ds t o t h e LRT r ej ection r eg ion .

F or t h e i t h s am ple , lik elih ood r at io can b e ex pr es sed a s

(4)

= n

- np

2

|A

i

- 1

0

|

n

2

ex p [ - 2 1 tr (

0- 1

A

i

) + 1 2 np ] .

Let T V

i

b e - 2 ln . T h en T V

i

= tr ( A

i

- 1

0

) - n ln |A

i

| + n ln |

0

| + np ln n - np . (2.2) H en ce, t h e st atist ic T V

i

can b e u s ed a s th e sam ple st at ist ics for m on it orin g . A lt (1982) pr op os ed t h e u se of sam ple g en er alized v arian ce | S

i

| t o m onit or disper sion m at rix . A n d H ui (1980) also u s ed th e sam ple g en er alized v arian ces for m onit orin g t h e pr oces s disper sion b y u sin g t h e follow in g st at istic n - 1 ( | S

i

| / |

0

| - 1 ) . T his st at istic is a sy m pt ot ically n orm ally dist ribu t ed w it h m ean 0 an d v arian ce 2p .

If t h e st at istic V

i

or T V

i

plot s ab ov e t h e upper con tr ol lim it s , t h e pr oces s disper sion m at rix is deem ed ou t - of- cont r ol st at e an d a s sig n able cau ses ar e s ou ght . T h e st atist ic V

i

h a s a chi - s qu ar ed distribut ion w it h ( n - 1) p deg r ees of fr eedom an d th e per cen t a g e poin t of t h e V

i

can b e obt ain ed from chi - s qu ar e distribut ion w h en th e proces s is in - cont r ol. But , if th e pr oces s sh ift s fr om

0

th en it is difficult t o obt ain t h e ex act dist ribu t ion of V

i

. T hu s in or der t o ob t ain th e p er cen t ag e p oint s of V

i

w h en t h e pr oes s is ou t - of - con tr ol st at e, it is n eces sary t o u se com pu t er sim ulat ion s . A n d , it is difficu lt t o obt ain t h e ex act dist ribu t ion of T V

i

w h en t h e pr oces s is in - con t rol or ou t - of- cont r ol st at es . T hu s , in or der t o ev alu at e t h e p erform an ce s of t h e ch art s b a sed on t h e st atist ic s V

i

or T V

i

for it is n eces s ary t o carry out com put er sim u lation s .

3 . M ultiv ari at e S h e w h art Ch art

S h ew h art ch art is w idely u sed t o display s am ple dat a fr om a pr oces s for t h e purpose of det erm inin g w h et h er a pr odu ct ion proces s is in - cont r ol, for brin gin g an out - of - cont r ol pr oce s s int o in - cont r ol, an d for m on it orin g a pr oces s t o m ak e sur e t h at it st ay s in - cont r ol. A S h ew h art ch art h a s a g ood ab ilit y t o det ect larg e ch an g es in m onit or ed p ar am et er quickly an d is ea sy t o im plem en t t h e pr oces s . H ow ev er , th e S h ew h art ch art is slow t o sign al sm all or m oderat e ch an g es in t h e pr oces s par am et er s .

T h e ru n len gt h N is defin ed a s t h e r an dom n um b er of s am ple s r equ ir ed for t h e ch art t o sign al an d th e t im e r equir ed t o sig n al T is d N w h ere d is t h e len g th of th e sam plin g in t erv al. T h e av er ag e run len g th (A RL ) is E ( N ) an d t h e ex pect ed t im e t o sig n al E ( T ) is sim ply t h e pr odu ct of t h e A RL an d d . T h er efor e, th e A RL can b e t h ou g ht of a s t h e ex pect ed t im e t o sign al.

Let q b e th e pr ob abilit y th at a ch art st atist ic falls out - of - con t rol lim it s , t h en N

(5)

is g eom etrically dist ribu t ed w it h p ar am et er q w h en t h e pr oces s is in - cont r ol. T h e ex p ect ed tim e t o sign al E ( T ) an d t h e v arian ce of t h e t im e t o sig n al V ( T ) can b e repr esen t ed a s

E ( T ) = d E ( N ) = d

q an d V ( T ) = d

2

( 1 - q) q

2

.

S in ce t h e cont r ol lim it s for a m ultiv ariat e S h ew h art ch art b a s ed on t h e sam ple st atist ic V

i

w ould b e s et a s { 0 ,

21 -

[ ( n - 1)p ] } , a S h ew h art ch art b a s ed on V

i

sig n als w h en ev er

V

i 21 -

[ ( n - 1) p ] . (3.1)

A n d b ecau se t h e cont r ol lim it s for a m u lt iv ariat e S h ew h art ch art b a sed on t h e st atist ic T V

i

w ou ld b e set by u sin g p er cent ag e poin t of T V

i

, a S h ew h art ch art b a sed on T V

i

sig n als w h en ev er

T V

i

h

T V ( S )

(3.2)

w h er e h

T V ( S )

can b e obt ain ed t o sat isfy a specified in - con tr ol A RL by sim ulat ion . [R esu lt 3.1] Let X

ij

= ( X

ij 1

, X

ij2 ,

, X

ijp

) ' b e dist ribu t ed a s N

p

(

0

,

0

) an d X

ij

's b e in depen dent w h en t h e pr oces s is in - cont r ol. A s su m e t h at m u lt iv ariat e S h ew h art ch art b a s ed on th e st at ist ic V

i

in (2.1) is u sed a s st at ed ab ov e. If th e pr oces s par am et er s of th e dist ribu tion shift ed a s N

p

(

0

, c

0

) w h er e c is a con st ant , th en

A R L = 1

1 - F ( h

V

/ c ) (3.3)

w h er e h

V

=

21 -

[ ( n - 1)p ] is th e con tr ol lim it of t h e ch art in (3.1) an d F ( ) is a chi- s qu ar ed dist ribu tion fu n ction w it h ( n - 1) p deg rees of freedom .

4 . M u ltiv ari at e CU S U M Ch art

T h e CU S U M ch art is a g ood alt ern at iv e t o t h e S h ew h art ch art an d is oft en u sed in st ea d of st an dar d S h ew h art ch art w h en det ect ion of sm all sh ift s in a pr odu ct ion pr oces s is im p ort ant . A CU S UM ch art dir ect ly in corp or at es all of th e inform at ion in th e sequ en ce of sam ple v alu e s by plot t in g t h e cum ulat iv e sum of t h e dev iat ion of t h e sam ple v alu es from th e t ar g et v alu e.

A m ult iv ariat e CU S UM ch art b a sed on th e st at ist ic V

i

in (2.1) is g iv en by

Y

V , i

= m ax { Y

V , i - 1

, 0 } + ( V

i

- k

V

) (4.1)

(6)

w h er e Y

V , 0

=

V

(

V

0) an d r efer en ce v alu e k

V

0 . T his ch art for disp er sion m at rix sign als w h en ev er Y

V , i

h

V

.

W h en th e pr oces s par am et er s ar e on - t ar g et , decision int erv al h

V

can b e ev alu at ed by th e M ark ov ch ain or in t egr al equ at ion appr oach t o s at isfy a specified in - cont r ol A RL . A n d w h en t h e proces s par am et er s in h av e ch an g ed, th e p erform an ce s an d pr opert ies of t his ch art can b e ev alu at ed by sim ulation .

T h e CU S UM pr ocedu re can b e con sider ed a s a sequ en ce of in depen dent t est s w h er e each t est is act u ally equ iv alen t t o a sequ ent ial pr ob ability r at io t est (S P RT ) for t estin g H

0

: =

0

v s H

1

:

0

. T h is sequ en ce of S P RT ' s is equiv alent t o u sin g th e CU S UM st atist ic

Y

i

= m ax { Y

i - 1

, 0 } + ( L

i

- k ) (4.2)

w h er e t h e lik elih ood r at io L

i

= max {L ( , ) } / m ax {L ( ,

0

) } an d L ( , ) is lik elih ood fu n ction of r an dom v ect or X

i

= ( X '

i1

, X '

i1

, , X '

in

) ' an d k 0 . T his ch art sign als w h en ev er Y

i

> c . T h er efore , a CU S UM procedur e b a s ed on t h e st atist ic T V

i

in (2.2) can also b e con st ru ct ed a s

Y

T V , i

= m ax { Y

T V , i - 1

, 0 } + ( T V

i

- k

T V

) (4.3)

w h er e Y

T V , 0

=

T V

(

T V

0) an d k

T V

0 . T his ch art sign als w h en ev er Y

T V , i

h

T V ( C)

.

S in ce it is difficult t o obt ain t h e ex act perform an ces of m u lt iv ariat e CU S UM s ch em e b a sed on T V

i

, t h e per cent ag e p oint an d pr opert ies of t his ch art can b e ev alu at ed b y sim u lat ion u n der th e pr oces s p ar am et er s of th e pr oce s s ar e on - t ar g et or ch an g ed .

5 . N u m e ric al P e rf orm an c e s an d Con c lu din g R em ark s

T h e desig n of a CU S UM ch art r equ ir es t h e specification of t h e sam ple size n , t h e sam plin g int erv al d , an d t h e ch art par am et er s h an d k . A g ood ch oice for t h e ch art par am et er s depen ds on th e n um b er of qu alit y v ariables in t h e pr op os ed con tr ol sch em e an d t h e size of shift on in t ere st in g .

In or der t o ev alu at e th e perform an ces an d com pare t h e pr oposed m u lt iv ariat e

CU S U M an d S h ew h art ch art s fairly , it is n ece s sary t o calibr at e ea ch s ch em es so

t h at on - t ar g et A RL E ( N |

0

,

0

) b e t h e sam e for all t h e pr oposed sch em e s . In

our com put at ion , each sch em e w a s calibr at ed so t h at t h e on - t ar g et A RL w a s

appr ox im at ely equ al t o 370.4 an d t h e s am ple size for each ch ar act erist ic w a s fiv e

for p = 3 an d p = 4 . F or conv en ien ce, w e let th at t h e s am plin g int erv al of un it

t im e d = 1 an d k n ow n t ar g et m ean v ect or

0

= 0 . T h e perform an ce of th e

(7)

ch art s for m onit orin g a disper sion m at rix depen d s on t h e com pon en t s of . F or com pu t ation al sim plicity in our com pu t ation , w e a s sum e th at

2r0

= 1,

r s0

= 0 . 3 for

r , s = 1, 2 , , p .

S in ce it is n ot pos sible t o inv estig at e all of t h e differ en t w ay s in w h ich cou ld ch an g e, w e con sider t h e follow in g ty pical ty p es of shift s for com p arison in t h e pr oces s par am et er s :

(1) V

i

:

10

of

0

is in cr ea sed t o [ 1 + ( 4 i - 3) / 10 ] .

(2) C

i

:

120

an d

2 10

of

0

ar e ch an g ed t o [ 0 . 3 + ( 2 i - 1) / 10 ] (3 ) ( V

i

, C

i

) for i = 1, 2 , 3 .

(4 ) S

i

:

0

is ch an g ed t o c

i 0

w h ere c

i

= [ 1 + ( 3 i - 2) / 10 ]

2

.

A ft er t h e r efer en ce v alu e of t h e proposed CU S UM ch art b a sed on t h e cont r ol st atist ic V

i

, decision in t erv al h

V

w a s calculat ed by M ark ov ch ain m et h od w it h t h e nu m b er of t r an sien t st at es r = 100 . A n d t h e p ar am et er s h

T V ( C)

b a sed on T V

i

for CU S U M sch em e , an d th e A RL v alu es for all t h e pr oposed t y pes of shift s for S h ew h art an d CU S UM ch art s b a sed on V

i

or T V

i

w er e obt ain ed by sim ulation w it h 10,000 it er at ion s .

< T able 1> A RL v alu es for disp er sion m at rix ( p = 3 )

t y pes S h ew h art CU S UM

of shift s V

i

T V

i

b a s ed on V

i

b a sed on T V

i

n o shift 370.4 370.4 370.4 370.4 370.4 370.3 370.3 370.3

V

1

177.7 340.6 86.1 91.5 99.5 302.2 311.9 318.0

V

2

15.8 53.9 12.6 10.7 9.8 24.5 25.2 26.4

V

3

4.5 8.8 6.2 5.0 4.5 7.2 6.6 6.3

C

1

419.7 354.4 623.3 564.2 530.6 321.6 329.1 334.2

C

2

403.8 244.7 1663.7 1121.5 880.8 103.5 119.9 136.4

C

3

309.0 101.3 3019.5 1502.3 996.9 18.5 19.1 20.9

( V

1

, C

1

) 203.8 328.2 116.3 123.9 133.6 274.5 288.6 296.8

( V

2

, C

2

) 19.2 46.3 15.6 13.3 12.4 20.3 20.4 21.3

( V

3

, C

3

) 5.6 6.9 7.5 6.2 5.5 5.7 5.2 4.9

S

1

65.3 299.6 26.6 24.5 24.7 213.8 229.0 242.0

S

2

4.5 27.2 5.8 4.7 4.1 12.3 11.9 11.9

S

3

1.7 4.3 3.1 2.6 2.0 4.2 3.8 3.5

S

4

1.2 1.8 2.2 1.8 1.6 2.4 2.1 2.0

kV= 12 . 5 kV= 13 kV= 13 . 5 kT V= 9 kT V= 9 . 5 kV= 10

(8)

T h e perform an ces of t h e pr oposed m u lt iv ariat e S h ew h art an d CU S UM ch art ar e g iv en in T able 1 an d 2. By v ariou s n um erical com put at ion , w e foun d t h e follow in g pr opert ies . W h en a shift in v arian ce com p on ent s h av e occur ed , m ultiv ariat e sch em e b a sed on t h e con tr ol st atist ic V

i

is efficien t . A n d a shift in corr elation coefficien t s h av e occu r ed, con tr ol procedur e b a sed on t h e T V

i

w ill b e r ecom m en ded. A sh ift for b oth v arian ces an d corr elation coefficient s in h a s occurr ed , t h e m ultiv ariat e CU S U M pr ocedu re b a sed on T V

i

w ill b e r ecom m en ded.

W h en sm all or m oder at e ch an g es in t h e pr odu ction pr oces s h av e occu rr ed , CU S U M procedur es are m ore efficien t t h an S h ew h art ch art s in t erm s of A RL . N um erical r esu lt s for v ariou s r efer en ce v alu es of CU S UM sch em es sh ow th at sm all r efer en ce v alu es ar e m or e efficient in det ect in g sm all sh ift s an d lar g e r efer en ce v alu es ar e m or e efficient for lar g e shift s .

H en ce, w e r ecom m en d CU S UM ch art b a sed on th e con tr ol st atist ic T V

i

t o sim u lt an eou sly m onit or b oth v arian ce s an d corr elation coefficient s of in th e m u lt iv ariat e n orm al pr oces s w h en sm all or m oder at e sh ift h a s occu r ed in t h e pr odu ct ion pr oces s .

< T able 2> A RL v alu es for disp er sion m at rix ( p = 4 )

t y pes S h ew h art CU S UM

of shift s V

i

T V

i

b a s ed on V

i

b a sed on T V

i

n o shift 370.4 370.4 370.4 370.4 370.4 370.4 370.4 370.4

V

1

199.9 357.3 97.9 103.6 111.9 316.3 322.8 325.8

V

2

19.3 118.7 14.9 12.6 11.6 43.5 41.8 42.8

V

3

5.3 19.3 7.2 5.9 5.2 13.8 12.0 11.0

C

1

397.6 359.7 540.7 506.5 488.8 331.8 334.7 339.9

C

2

374.3 291.6 1107.9 859.3 721.2 130.8 140.9 153.6

C

3

291.8 166.4 1795.4 1106.6 815.8 33.8 32.1 32.8

( V

1

, C

1

) 2191.1 349.6 123.6 130.5 139.9 290.7 298.0 305.9

( V

2

, C

2

) 22.5 100.7 17.7 15.1 14.0 35.7 33.6 33.9

( V

3

, C

3

) 6.2 14.2 8.5 7.0 6.2 10.7 9.2 8.4

S

1

55.3 326.6 23.2 20.7 20.0 215.1 226.6 237.2

S

2

3.4 47.9 5.2 4.2 3.7 17.9 15.7 14.7

S

3

1.4 5.7 2.9 2.4 2.1 6.3 5.3 4.8

S

4

1.1 2.0 2.0 1.7 1.5 3.5 2.9 2.6

kV= 16 . 5 kV= 17 kV= 17 . 5 kT V= 16 kT V= 16 . 5 kV= 17

(9)

R e f e re n c e s

1. Alt , F .B. (1982). Multiv ariat e Qu ality Cont rol in T he E ncy clop ed ia of S ta tis tical S ciences , eds. S . Kot z and John son , J ohn W iley , New York . 2. Arnold, J .C. and Reynolds, M.R. (2001). CUSUM Control Chart s with Variable

S am ple S izes an d S am plin g Int erv als, J ournal of Quality T echn ology , V ol.

33, 66- 81.

3. Ch an g , T .C. an d Gan , F .F . (1995). A Cum ulativ e S um Control Ch art for M onit orin g P rocee V arian ce, J ournal of Quality T echn ology , V ol. 27, 109- 119.

4. Crosier , R.B. (1988). Multiv ariat e Generalization of Cumulativ e Sum Qu ality - Control S chem e, T echn om etrics, V ol. 30, 291- 303.

5. Gh ar e, P .H . an d T or ger sen , P .E . (1968). T he Multicharact eristic Control Ch art , J ournal of I nd us trial E ng in eering , Vol. 19, 269- 272.

6. Hot ellin g , H .(1947). Multiv ariat e Quality Control, T echniques of S tatis tical A naly s is , McGraw - Hill, New York , 111- 184.

7. Jack son , J .S .(1959). Qu ality Cont rol M et hods for S ev eral Relat ed Variables , T echn om e trics , V ol. 1, 359- 377.

8. Pign at iello, J .J ., Jr . an d Run g er , G.C. (1990). Com parison s of Multiv ariat e CUS UM Chart s , J ournal of Quality T echn ology , V ol. 22, 173- 186.

9. Pr abhu , S .S . and Run g er , G.C.(1997). Designin g a Multiv ariat e EW MA Cont rol Ch art , J ournal of Quality T echn ology , V ol. 29, 8- 15.

10. Reynolds , M .R., Jr . and Ghosh , B.K . (1981). Designing Control Ch art s for Mean s an d V arian ces , A S Q C Quality Cong ress T ransactions , S an F ran cisco, 400- 407.

11. W oodall, W .H . an d N cube, M .M . (1985). Multiv ariat e CUSUM Qu ality Control P rocedure, T echn om e trics , V ol. 27, 285- 292.

[ 2002년 8월 접수, 2002년 10월 채택 ]

참조

관련 문서

Center for Korean studies East Asian Languages &amp; Cultures Korean Concentration in Asian Studies Program Korean American Student Association in UC Santa Cruz

공정이 관리상태에 있다고 판정하기 위해서는 관리한계선을 벗어난 점이 없거나 혹은 점의 배열에 아무런 습성이 없어야

Lab., Hanyang Univ.. Lab., Hanyang Univ.. Lab., Hanyang Univ.. Lab., Hanyang Univ.. Lab., Hanyang Univ.. Lab., Hanyang Univ.. Lab., Hanyang Univ.. Lab., Hanyang Univ..

As for the data used in this study, for the synoptic analysis, the surface weather chart and 500 hPa weather chart, which had been produced by the

치주질환에 있어서 주요한 역할을 한다고 잘 알려진 P.gi ngi val i s ,P.i nt er medi a, A.ac t i nomyc et emc omi t ans ,T.f or s yt hi a및 Tr eponemadent i c ol

경향을 보이고 있다.. Fresh snow cover at each station by the time series on 28 Dec. Surface weather chart on 00UTC 28 Dec.. Surface weather chart on 09UTC 28 Dec. Wind rose

I Kourti and MacGregor: Process analysis, monitoring and diagnosis, using multivariate projection methods (paper 31). I MacGregor and Kourti: Statistical process control

– A fixed lump-sum price based on the quantities provided by the owner for the major components of the project. – Most infrastructure projects and