This action might not be possible to undo. Are you sure you want to continue?
PLEASE SCROLL DOWN FOR ARTICLE
This article was downloaded by: [Bon Consortium  2007]
On: 6 June 2010
Access details: Access Details: [subscription number 919435512]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37
41 Mortimer Street, London W1T 3JH, UK
Communications in Statistics  Theory and Methods
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713597238
The Design and Performance of the Multivariate SyntheticT
2
Control
Chart
Francisco Aparisi
a
; Marco A. de Luna
b
a
Departamento de Estadística e Investigación Operativa Aplicadas y Calidad, Universidad Politécnica
de Valencia, Valencia, España
b
Departamento de Ingeniería Industrial y Mecánica, ITESM,
Guadalajara, México
To cite this Article Aparisi, Francisco and de Luna, Marco A.(2009) 'The Design and Performance of the Multivariate
SyntheticT
2
Control Chart', Communications in Statistics  Theory and Methods, 38: 2, 173 — 192
To link to this Article: DOI: 10.1080/03610920802178413
URL: http://dx.doi.org/10.1080/03610920802178413
Full terms and conditions of use: http://www.informaworld.com/termsandconditionsofaccess.pdf
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, redistribution, reselling, loan or sublicensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
Communications in Statistics—Theory and Methods, 38: 173–192, 2009
Copyright © Taylor & Francis Group, LLC
ISSN: 03610926 print/1532415X online
DOI: 10.1080/03610920802178413
The Design and Performance of the Multivariate
SyntheticT
2
Control Chart
FRANCISCO APARISI
1
AND MARCO A. DE LUNA
2
1
Departamento de Estadística e Investigación Operativa Aplicadas y
Calidad, Universidad Politécnica de Valencia, Valencia, España
2
Departamento de Ingeniería Industrial y Mecánica, ITESM,
Guadalajara, México
One of the objectives of research in statistical process control is to obtain control
charts that show few false alarms but, at the same time, are able to detect quickly the
shifts in the distribution of the quality variables employed to monitor a productive
process. In this article, the syntheticT
2
control chart is developed, which consists
of the simultaneous use of a CRL chart and a Hotelling’s T
2
control chart. The
ARL is calculated employing Markov chains for steady and zerostate scenarios.
A procedure of optimization has been developed to obtain the optimum parameters
of the syntheticT
2
, for zero and steady cases, given the values of incontrol ARL
and magnitude of shift which needs to be detected rapidly. A comparison between
(standard T
2
, MEWMA, T
2
with variable sample size, and T
2
with double sampling)
charts reveals that the syntheticT
2
chart always performs better than the standard
T
2
chart. The comparison with the remaining charts demonstrate in which cases the
performance of this new chart makes it interesting to employ in real applications.
Keywords ARL; Multivariate; Quality control; SPC; Synthetic.
Mathematics Subject Classiﬁcation 62P30.
1. Introduction
Statistical quality control has the objective of detecting shifts in the distribution of
the monitored quality variables. In the majority of production processes the possible
spontaneous changes in this distribution will worsen the quality of the product
manufactured. Therefore, the quick detection of these shifts is very important to
maintain quality. The statistical process control (SPC) has shown itself to be a
powerful tool in achieving this objective.
It is possible to employ two strategies when it is desired to control concurrently
¡ variables from one component or from a production process: to use simultaneously
Received August 21, 2007; Accepted May 2, 2008
Address correspondence to Francisco Aparisi, Departamento de Estadística e
Investigación Operativa Aplicadas y Calidad, Universidad Politécnica de Valencia, Valencia
46022, España; Email: faparisi@eio.upv.es
173
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
174 Aparisi and de Luna
¡ univariate control charts or to employ only one multivariate control chart
(Montgomery, 2005). Over the years, much of research has been made in order
to design control charts capable of fast detection of small shifts in the process.
The problem is that the standard control charts (X in the univariate case and
Hotelling’s T
2
in the multivariate case) are not efﬁcient at detecting small shifts.
The efﬁcacy of a quality control chart is normally measured using the ARL (Average
Run Length), the average number of points in the chart until the ﬁrst outofcontrol
signal appears.
Many strategies have been proposed with the aim of improving the ARL values
when the process is out of control while keeping a desired ARL value when the
process is in control. From our point of view, the more efﬁcient strategies are:
1. Cumulative sums, in the univariate case the CUSUM charts (Jones et al., 2004;
Luceno and PuigPey, 2000) and in the multivariate case the MCUSUM charts
(Khoo and Quah, 2002; Woodall and Ncube, 1985).
2. Exponentially weighted averages, the EWMA charts in the univariate case
(Knoth, 2005; Lucas and Saccucci, 1990) and MEWMA charts for the
multivariate case (Aparisi and GarcíaDíaz, 2004; Lowry et al., 1992; Prabhu and
Runger, 1997).
3. Variable sample size (VSS) univariate charts (Costa, 1994; Costa and Rahim,
2001; Reynolds and Arnold, 2001) and VSS multivariate charts (Aparisi, 1996;
Aparisi and Haro, 2003).
4. Double sampling (DS) univariate charts (Daudin, 1992; He and Grigoryan, 2006)
and DS multivariate charts (Champ and Aparisi, 2006; He and Grigoryan, 2005).
5. Synthetic charts for the univariate case (Chen and Huang, 2005a; Wu and
Spedding, 2000).
The performance of the SPC has been improved employing simultaneously two
control charts. Examples of this strategy are: the ShewhartCUSUM chart (Lucas,
1982), the ShewhartEWMA chart (Klein, 1997), and the synthetic chart, which is a
ShewhartCRL (conforming run length) chart. The syntheticX chart was introduced
by Wu and Spedding (2000) as a control chart useful for detecting small shifts in
the mean of a process.
In this article, a procedure will be developed for ﬁnding the parameters of
the multivariate synthetic Tsquare chart, Ghute and Shirke (2008), that minimizes
the ARL for a given size of shift, given a desired incontrol ARL, for the steady
and zero states. In addition, a comparison versus other multivariate control charts
is made. Section 2 shows a summary of multivariate SPC and the methodologies
which enable more efﬁcient control charts to be obtained. In Sec. 3, the syntheticT
2
chart is deﬁned. The optimization of the parameters of this chart can be found in
Sec. 4. Section 5 shows the comparison between several multivariate control charts,
in order to ﬁnd out in which cases the syntheticT
2
chart shows better performance.
The conclusions can be found in Sec. 6.
2. Multivariate Quality Control
Multivariate quality control consists of monitoring in one chart ¡ variables of the
same production process or component. The ﬁrst proposal to control possible shifts
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
The SyntheticT
2
Control Chart 175
in the mean vector was Hotelling’s T
2
control chart (Hotelling, 1947) which consists
of plotting the values of the statistic T
2
i
,
T
2
i
= n
X
i
−
0
T
−1
0
X
i
−
0
. (1)
where n is the sample size, X
i
is the mean vector of the ¡ sample means,
0
is the
mean vector when the process is in control, and
0
is the incontrol covariance
matrix. The distribution of the statistic T
2
i
, if the process is in an incontrol state
and
0
and
0
are known, is a chisquare with ¡ degrees of freedom. Therefore, it
is easy to obtain the control limit (CL) ﬁxing the probability of the Type I error, :.
Given the value of :, the control limit is CL = X
2.:
¡
.
As the points plotted in the Hotelling’s T
2
chart are independent, the incontrol
ARL is given by ARL
0
=
1
:
. The performance of this chart to detect shifts
in the mean vector depends on the non centrality parameter z = nJ
2
, where
J is the Mahalanobis’ distance, J =
(
1
−
0
)
T
−1
0
(
1
−
0
), where
1
is the
outofcontrol mean vector. The value of ARL when J = 0 is ARL(J = 0) =
1
1−[
(where [ is the probability of the Type II error) is easily calculated, knowing the
distribution of the T
2
i
statistic when the process is out of control, a non central
chisquare, T
2
≈ X
2
¡
(z = nJ
2
).
Commonly, the mean vector and the covariance matrix are estimated. Lowry
and Montgomery (1995) and Champ et al. (2005) presented guides for the design
of the T
2
control chart with estimated parameters. Vargas (2003) and Williams
et al. (2006) studied alternatives to estimate the covariance matrix for the case of
individual multivariate observations.
The T
2
chart has a poor performance to detect small shifts in the mean vector.
In this article, the use of a syntheticT
2
control chart (Ghute and Shirke, 2008)
is proposed to reduce the values of ARL to detect process shifts computing the
ARL employing Markov chains. The analysis is done employing the zerostate and
steadystate approaches (Champ, 1992; Davis and Woodall, 2002; Zhang and Wu,
2005) and considering that
0
and
0
are known. Ghute and Shirke (2008) did not
show results for small shifts and the ARL is only computed for the zerostate case.
Davis and Woodall (2002) noted the importance of studying the synthetic charts for
the steadystate case. Subsequently, there will be a description of the multivariate
control charts that are going to be used in the comparison of performance shown
in Sec. 5.
2.1. Variable Sample Size T
2
Control Chart
Aparisi (1996) and Aparisi and Haro (2003) improved the sensibility of the T
2
chart
employing variable sample size (VSS). The strategy consists of determining the value
of a warning limit (n). The sample size at time i depends on the value of T
2
i−1
.
If T
2
i−1
≤ n, a sample of size n
1
at the time i is taken. Otherwise, if n  T
2
i−1
 CL
the sample size to be used is n
2
. When T
2
i
> X
2
¡.:
, the process is considered to be
out of control, otherwise, it is decided that only random causes are present in the
process. It is possible, for a given incontrol ARL, to ﬁnd the values of CL. n. n
1
,
and n
2
that minimizes the outofcontrol ARL for a speciﬁed size of magnitude J,
considering the restrictions n
1
+n
2
≤ n
max
and a desired mean sample size, n
0
,
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
176 Aparisi and de Luna
E(n) = n
0
. A software tool has been employed to ﬁnd the optima parameters of this
chart, for comparison purposes.
2.2. MEWMA Control Chart
The multivariate EWMA control chart, MEWMA (Lowry et al., 1992), is a chart
where the plotted statistic takes into account the information from past samples.
The MEWMA control chart shows an outofcontrol signal when the T
2
i
statistic
is larger than the control limit (h), selected to obtain a desired incontrol ARL.
In order to compute the T
2
i
statistic it is necessary to employ a smoothing coefﬁcient,
r (0  r ≤ 1). The MEWMA vector is obtained through Z
i
= rX
i
+(1 −r)Z
i−1
and
the statistic to be plotted is given by T
2
i
= Z
T
i
−1
z
i
Z
i
, being
z
i
the covariance
matrix of Z
i
, where Z
0
=
0
. The exact covariance matrix of Z
i
is computed
according to
z
i
=
r1−(1−r)
2i
]
2−r
.
It is a common practice to use the asymptotic approximation of the matrix
z
i
.
When r = 1, the resultant chart is the Hotelling’s control chart. If the value of r
is reduced the weight of the past samples is more important. Aparisi and García
Díaz (2004) developed genetic algorithm based software which ﬁnds the parameters
r and h (for univariate and multivariate cases) that produces the minimum ARL for
a given size of shift and for a given incontrol ARL.
2.3. Double Sampling T
2
Control Chart
The strategy of double sampling (DS) consists of taking two samples of sizes n
1
and n
2
from the process at the same time. With the statistical information obtained
in the ﬁrst sample it is determined whether the process is in control, out of control,
or whether it is required to analyze the second sample, and combine it with the
ﬁrst one in order to make a ﬁnal analysis. Champ and Aparisi (2006) developed an
analytical method to obtain the ARL of the DST
2
chart and, employing genetic
algorithms, the parameters of the optimum DST
2
chart to detect a shift of given
magnitude are found.
There are ﬁve parameters to deﬁne the DST
2
chart: h, h
1
, n
1
, n
1
, and n
2
.
The T
2
i.1
statistic is computed only considering the ﬁrst sample of size n
1
, employing
T
2
i.1
= n
1
(X
i.1
−
o
)
T
−1
(X
i.1
−
o
). If T
2
i.1
> h
1
, the process is deemed to be out of
control; if T
2
i.1
 n
1
, it is accepted that the process is in an incontrol state. However,
if n
1
≤ T
2
i.1
 h
1
it is not possible to make a decision and the second sample of size
n
2
is studied. The two samples are combined employing X
i
=
1
n
1
+n
2
(n
1
X
i.1
+n
2
X
i.2
).
Hence, T
2
i
is calculated using the expression T
2
i
= (n
1
+n
2
)(X
i
−j
0
)
T
−1
(X
i
−j
0
).
If T
2
i
> h, it is decided that the process is outofcontrol, otherwise the process is
deemed to be in an incontrol state.
3. The SyntheticT
2
Control Chart
The univariate synthetic chart was introduced by Wu and Spedding (2000) as an
alternative to improve the performance of the Shewhart control chart to detect
process shifts. It is the result of combining a Shewhart chart and CRL chart (a chart
originally designed to detect increments in the percentage of defective units). The
syntheticX chart for the zerostate scenario shows better ARL values to detect
process shifts, for any shift magnitude, than the X control chart. In some cases,
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
The SyntheticT
2
Control Chart 177
Figure 1. Subcharts of the syntheticT
2
control chart: T
2
subchart and CRL subchart.
especially for moderate and large shifts, the syntheticX chart exhibits a better
performance than the EWMA control chart (Wu and Spedding, 2000).
The synthetic chart has been also applied to monitor the variability of a process
(Chen and Huang, 2005a,b, 2006; Costa and Rahim, 2006) and the percentage of
defective units of a process (Wu et al., 2001). With the objective of improving
the performance of the synthetic control charts, Chen and Huang (2005b) applied
variable sampling intervals (VSI) and Davis and Woodall (2002) improved the
performance of the syntheticX control chart by adding the concept of sidesensitive.
The synthetic T
2
control chart (syntheticT
2
) is a chart for monitoring
simultaneously two or more quality characteristics. It consists of two subcharts:
a T
2
subchart and a CRL subchart. Figure 1 shows the concept of the synthetic
T
2
chart. The T
2
subchart has a unique control limit, LCsynt. The CRL subchart
has a low control limit, L, L ≥ 1. The value of LCsynt is the criteria to classify a
sample as conforming or non conforming. The value of L is the criteria to decide if
the process is in control or out of control.
The CRL concept assumes that in t = 0 there is a point above the LCsynt limit
(a non conforming sample in t = 0). This characteristic, called head start, is very
important for the performance of the synthetic charts. When this assumption is
ruled out, the performance of synthetic charts worsens (Davis and Woodall, 2002).
For a syntheticT
2
chart, unlike the T
2
control chart, T
2
i
> LCsxnt does not
mean that an outofcontrol state has to be assumed, but the inspected sample
must be classiﬁed as non conforming. Only when the sample is classiﬁed as non
conforming and CRL ≤ L will the process be considered to be out of control.
Like the T
2
control chart, the syntheticT
2
chart shows an outofcontrol signal,
indicating that there probably is a shift in the process. However, this signal does
not inform us about the variable or variables that have produced the shift. Several
methods for the interpretation of the outofcontrol signal of the T
2
control chart
have been developed. These methods can also be applied to the syntheticT
2
chart
(Aparisi et al., 2006; Atienzas et al., 1998; Kourti and MacGregor, 2004; Mason
and Young, 1999; Mason et al., 1997). Applying these techniques to the synthetic
T
2
chart the user may obtain a controlling procedure that, although it is not the
fastest in comparison versus other multivariate charts (see the comparison later),
may be more useful in a real application, due to the help provided in identifying the
variables that have shifted.
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
178 Aparisi and de Luna
4. Optimization of the Parameters of the SyntheticT
2
Chart
Two values of ARL are important for the design and performance of the synthetic
T
2
chart, the incontrol ARL [ARL
S−T
2 (J = 0)] and the outofcontrol ARL
[ARL
S−T
2 (J = 0)]. The value of the incontrol ARL is selected taking into account
the frequency of false alarms. The value of ARL
S−T
2 (J = 0) is important in order
to rapidly detect a shift in the mean vector of magnitude J.
4.1. Obtaining the ARL Values
When the ARL needs to be computed, two approaches are possible: zero state
and steady state (Champ, 1992; Zhang and Wu, 2005). The zerostate scenario
implies that the shift in the process occurred just before the ﬁrst sample was taken.
The steadystate case means that the process was in an incontrol state for a
long period, before the outofcontrol situation happened. Wu and Spedding (2000)
showed the formulas to obtain the ARL value of the syntheticX chart for the
zerostate case. Davis and Woodall (2002) found that the syntheticX chart may
be modelled employing Markov chains, like a standard X chart with a run rule of
2 points of (L +1) outside control limits.
In a similar way, it is possible to obtain the ARL of the multivariate synthetic
T
2
control chart (ARL
S−T
2 ) employing a Markov chain approach. The syntheticT
2
chart may be modeled like a T
2
control chart with a run rule that consists of 2 points
of (L +1) outside the upper control limit; see the Appendix. This approach is also
useful for calculating the ARL for the zerostate case.
4.2. Optimization of the SyntheticT
2
Control Chart
A software tool for Windows
®
has been developed in order to ﬁnd the optima
parameters of the syntheticT
2
control chart. The software tool is available upon
request from the authors. The output of this software tool is shown in Fig. 2.
Once the zero or steady state is selected, the optimization process ﬁnds the values
of L and LCsynt that minimizes the value of ARL
S−T
2 (J = 0) for a given shift
of magnitude J. sample size, and number of variables, taking into account the
restriction of the speciﬁed incontrol ARL, ARL
S−T
2 (J = 0). The chart that fulﬁls
this design is considered to be optimum to detect a shift of size J in the mean vector.
As exposed in Fig. 2, the output from the software tool shows the improvement of
ARL against Hotelling’s T
2
control chart and it draws the ARL curves of T
2
and
syntheticT
2
charts.
Figure 2 shows the solution for the problem ARL
S−T
2 (J = 0) = 200, ¡ = 3,
n = 4, J = 0.75, and steady state. The parameters of the optimum syntheticT
2
chart are: LCsxnt = 9.163 and L = 9, with ARL
S−T
2 (J = 0.75) = 13.58. The outof
control ARL value for the standard T
2
chart is 20.41. Therefore, the reduction in
the value of ARL for a shift J = 0.75 employing the syntheticT
2
chart is 33.5%.
5. Comparison Versus Other Multivariate Charts
5.1. Comparison Between Optimum Steady and Zero States SyntheticT
2
Charts
A comparison of performance between the optimum design for the zerostate
syntheticT
2
and the steadystate synthetic T
2
control charts has been carried out.
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
The SyntheticT
2
Control Chart 179
Figure 2. Main window of the inhouse software tool used to ﬁnd the optimum
syntheticT
2
and syntheticX control charts.
The optimum design of the syntheticT
2
charts has been utilized in the comparison
for different cases of ARL
S−T
2 (J = 0) = ARL
T
2 (J = 0), n, ¡, and J. The ARL
values are shown in the second and third columns of Tables 1–6.
For the same design parameters (same incontrol ARL, number of variables and
sample size), the optimum value of L of the syntheticT
2
chart has the same value
for the steady state and zerostate scenarios, if the magnitude of shift J is moderate
or large. However, for small shifts, the optimum parameter L for the steadystate
case is always lower than the optimum for the zerostate case. For all the cases, the
optimum value of LCsynt for zero state will always be larger than the optimum value
for steady state.
The performance of the optimum zerostate syntheticT
2
chart is always better
than the optimum chart for the steadystate case. This is a predictable result,
a consequence of the inertia that the steady state produces in the detection of the
shift. The largest differences (50%) are found for large shifts, although these shifts
are detected quickly for both schemes. For small shifts (J ≈ 0.5), the differences in
the ARL values are around 20%. However, these differences tend to be larger as the
sample size increases, and tend to decrease as the number of variables, ¡. increase.
5.2. Comparison Versus Hotelling’s T
2
Chart
Acomparison of performance between the zerostate syntheticT
2
and the standard T
2
control charts [ARL
T
2 (J)] has been carried out. The zerostate case of the syntheticT
2
chart has been selected for this comparison, and for the next one, because for the
rest of charts the ARL has been computed employing the zerostate approach.
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
T
a
b
l
e
1
A
R
L
o
f
t
h
e
o
p
t
i
m
u
m
m
u
l
t
i
v
a
r
i
a
t
e
c
o
n
t
r
o
l
c
h
a
r
t
s
f
o
r
a
s
h
i
f
t
J
.
A
R
L
(
J
=
0
)
=
2
0
0
,
¡
=
2
,
n
=
2
S
y
n
t
h
e
t
i
c

T
2
S
y
n
t
h
e
t
i
c

T
2
V
S
S

T
2
D
S

T
2
T
2
(
L
C
s
y
n
t
,
L
)
(
L
C
s
y
n
t
,
L
)
M
E
W
M
A
(
C
L
.
n
.
n
1
.
n
2
)
(
h
.
h
1
.
n
1
.
n
1
.
n
2
)
J
(
C
L
)
Z
e
r
o
s
t
a
t
e
S
t
e
a
d
y
s
t
a
t
e
(
h
.
r
)
n
1
+
n
2

=
2
0
n
1
+
n
2

=
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
1
0
.
5
9
6
6
0
.
2
5
1
4
8
.
3
4
1
2
9
.
2
7
1
3
7
.
7
1
4
1
.
7
5
1
3
8
.
2
8
1
6
.
1
1
(
8
.
9
4
2
,
5
0
)
(
8
.
1
5
5
,
2
9
)
(
7
.
3
7
2
,
0
.
0
5
)
(
1
0
.
5
9
7
,
5
.
7
5
,
1
,
1
9
)
(
1
1
.
2
5
,
1
9
.
1
6
,
1
.
3
9
,
1
,
2
)
0
.
5
7
6
.
8
6
5
1
.
3
7
6
1
.
3
8
1
7
.
0
3
3
9
.
9
5
2
1
.
3
6
(
8
.
4
9
9
,
3
0
)
(
7
.
8
2
6
,
1
9
)
(
9
.
0
6
7
,
0
.
1
3
)
(
1
0
.
5
9
7
,
5
.
7
8
,
1
,
1
9
)
(
6
.
0
0
,
1
3
.
9
6
,
5
,
7
3
,
1
,
1
8
)
0
.
7
5
3
7
.
0
1
1
9
.
8
8
2
6
.
0
9
9
.
2
4
1
2
.
5
5
9
.
4
8
(
7
.
9
8
7
,
1
7
)
(
7
.
4
5
1
,
1
2
)
(
9
.
3
6
6
,
0
.
1
6
)
(
1
0
.
5
9
7
,
5
.
4
1
,
1
,
1
6
)
(
6
.
6
9
,
1
4
.
7
7
,
5
.
1
1
,
1
,
3
)
1
1
8
.
4
8
8
.
8
0
1
2
.
2
8
6
.
0
3
6
.
2
2
5
.
7
9
(
7
.
4
9
6
,
1
0
)
(
7
.
1
0
8
,
8
)
(
9
.
9
5
,
0
.
2
6
)
(
1
0
.
5
9
7
,
4
.
3
9
,
1
,
1
0
)
(
9
.
5
9
,
1
1
.
2
7
,
4
.
7
1
,
1
,
1
1
)
1
.
2
5
9
.
9
1
4
.
5
9
6
.
6
3
4
.
3
4
3
.
9
3
3
.
6
2
(
7
.
1
6
1
,
7
)
(
6
.
8
5
9
,
6
)
(
1
0
.
2
8
,
0
.
3
8
)
(
1
0
.
5
9
7
,
3
.
8
9
,
1
,
8
)
(
9
.
8
7
,
1
2
.
6
2
,
3
.
2
0
,
1
,
5
)
1
.
5
5
.
7
6
2
.
7
9
4
.
1
0
3
.
2
7
2
.
8
6
2
.
6
4
(
6
.
8
4
2
,
5
)
(
6
.
6
9
8
,
5
)
(
1
0
.
3
6
,
0
.
4
3
)
(
1
0
.
5
9
7
,
3
.
2
2
,
1
,
6
)
(
9
.
4
5
,
1
1
.
5
8
,
4
.
3
4
,
1
,
9
)
1
.
7
5
3
.
6
5
1
.
9
3
2
.
8
6
2
.
6
1
2
.
2
8
1
.
9
6
(
6
.
6
3
0
,
4
)
(
6
.
5
0
0
,
4
)
(
1
0
.
5
1
,
0
.
5
9
)
(
1
0
.
5
9
7
,
2
.
7
7
,
1
,
5
)
(
1
1
.
2
7
,
1
1
.
1
2
,
3
.
8
3
,
1
,
7
)
2
2
.
5
1
1
.
4
9
2
.
2
2
2
.
1
0
1
.
9
2
1
.
5
0
(
6
.
3
5
5
,
3
)
(
6
.
2
4
2
,
3
)
(
1
0
.
5
4
8
5
,
0
.
6
7
)
(
1
0
.
5
9
7
,
2
.
2
0
,
1
,
4
)
(
1
3
.
2
5
,
1
1
.
0
1
,
2
,
7
4
,
1
,
4
)
2
.
2
5
1
.
8
7
1
.
2
6
1
.
8
7
1
.
7
3
1
.
6
8
1
.
2
6
(
5
.
9
6
6
,
2
)
(
5
.
8
7
3
,
2
)
(
1
0
.
5
7
,
0
.
7
4
)
(
1
0
.
5
9
7
,
1
.
3
9
,
1
,
3
)
(
1
3
.
5
9
,
1
0
.
9
5
,
2
.
7
4
,
1
,
4
)
2
.
5
1
.
5
0
1
.
1
3
1
.
6
8
1
.
4
7
1
.
4
9
1
.
2
0
(
5
.
9
6
6
,
2
)
(
5
.
8
7
3
,
2
)
(
1
0
.
5
8
,
0
.
7
9
)
(
1
0
.
5
9
7
,
1
.
3
9
,
1
,
3
)
(
1
4
.
1
2
,
1
0
.
8
2
,
3
.
1
8
,
1
,
5
)
180
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
T
a
b
l
e
2
A
R
L
o
f
t
h
e
o
p
t
i
m
u
m
m
u
l
t
i
v
a
r
i
a
t
e
c
o
n
t
r
o
l
c
h
a
r
t
s
f
o
r
a
s
h
i
f
t
J
.
A
R
L
(
J
=
0
)
=
2
0
0
,
¡
=
2
,
n
=
7
S
y
n
t
h
e
t
i
c

T
2
S
y
n
t
h
e
t
i
c

T
2
V
S
S

T
2
D
S

T
2
T
2
(
L
C
s
y
n
t
,
L
)
(
L
C
s
y
n
t
,
L
)
M
E
W
M
A
(
C
L
.
n
.
n
1
.
n
2
)
(
h
.
h
1
.
n
1
.
n
1
.
n
2
)
J
(
C
L
)
Z
e
r
o
s
t
a
t
e
S
t
e
a
d
y
s
t
a
t
e
(
h
.
r
)
n
1
+
n
2

=
2
0
n
1
+
n
2

=
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
1
0
.
5
9
6
6
0
.
2
5
8
4
.
3
5
5
8
.
3
2
6
8
.
6
9
1
8
.
2
7
6
6
.
1
7
1
0
.
3
6
(
8
.
5
5
6
,
3
2
)
(
7
.
8
6
6
,
2
0
)
(
8
.
4
7
1
,
0
.
0
9
)
(
1
0
.
5
9
7
,
2
.
2
0
,
1
,
1
9
)
(
1
2
.
0
2
,
1
1
.
9
0
,
0
.
5
6
,
1
,
8
)
0
.
5
2
1
.
9
8
1
0
.
7
0
1
4
.
7
3
6
.
6
5
9
.
0
6
7
.
8
1
(
7
.
6
6
6
,
1
2
)
(
7
.
2
0
9
,
9
)
(
9
.
7
6
9
,
0
.
2
2
)
(
1
0
.
5
9
7
,
2
.
1
9
,
1
,
1
9
)
(
9
.
7
3
,
1
5
.
2
7
,
2
.
9
2
,
4
,
1
3
)
0
.
7
5
7
.
0
4
3
.
3
3
4
.
8
6
3
.
6
2
3
.
3
0
2
.
8
5
(
6
.
8
4
2
,
5
)
(
6
.
6
9
8
,
5
)
(
1
0
.
1
9
5
,
0
.
3
4
)
(
1
0
.
5
9
7
,
2
.
3
3
,
2
,
1
8
)
(
1
0
.
1
9
,
1
1
.
8
8
,
5
.
0
7
,
6
,
1
3
)
1
3
.
0
2
1
.
6
8
2
.
5
0
2
.
3
6
2
.
1
0
1
.
5
2
(
6
.
3
5
5
,
3
)
(
6
.
2
4
2
,
3
)
(
1
0
.
4
1
,
0
.
4
7
)
(
1
0
.
5
9
7
,
3
.
0
1
,
5
,
1
4
)
(
9
.
6
3
,
1
4
.
3
4
,
4
.
5
9
,
6
,
1
0
)
1
.
2
5
1
.
7
2
1
.
2
0
1
.
7
9
1
.
6
3
1
.
5
4
1
.
1
8
(
5
.
9
6
6
,
2
)
(
5
.
8
7
3
,
2
)
(
1
0
.
5
7
,
0
.
7
4
)
(
1
0
.
5
9
7
,
3
.
8
9
,
6
,
1
3
)
(
1
0
.
8
9
,
1
2
.
3
6
,
2
.
5
8
,
4
,
1
1
)
1
.
5
1
.
2
4
1
.
0
5
1
.
5
7
1
.
2
5
1
.
2
5
1
.
0
4
(
5
.
9
6
6
,
2
)
(
5
.
8
7
3
,
2
)
(
1
0
.
5
9
,
0
.
8
7
)
(
1
0
.
5
9
7
,
0
.
0
0
,
6
,
7
)
(
1
0
.
2
5
,
1
5
.
3
6
,
3
.
8
8
,
6
,
7
)
1
.
7
5
1
.
0
7
1
.
0
1
1
.
5
1
1
.
0
8
1
.
0
8
1
.
0
2
(
5
.
9
6
6
,
2
)
(
5
.
8
7
3
,
2
)
(
1
0
.
5
9
4
,
0
.
9
1
)
(
1
0
.
5
9
7
,
1
.
3
9
,
6
,
8
)
(
1
7
.
9
4
,
1
0
.
6
3
,
5
.
1
5
,
6
,
1
4
)
2
1
.
0
2
1
.
0
0
1
.
5
0
1
.
0
2
1
.
0
3
1
.
0
0
(
5
.
9
6
6
,
2
)
(
5
.
8
7
3
,
2
)
(
1
0
.
5
9
,
0
.
8
6
)
(
1
0
.
5
9
7
,
3
.
2
2
,
6
,
1
1
)
(
1
5
.
2
3
,
1
0
.
7
5
,
3
.
5
3
,
5
,
1
2
)
2
.
2
5
1
.
0
0
1
.
0
0
1
.
5
0
1
.
0
0
1
.
0
1
1
.
0
0
(
5
.
9
6
6
,
2
)
(
5
.
8
7
3
,
2
)
(
1
0
.
5
9
,
0
.
9
0
)
)
(
1
0
.
5
9
7
,
2
.
2
0
,
6
,
9
)
(
1
2
.
5
6
,
1
1
.
3
3
,
2
.
7
4
,
5
,
8
)
2
.
5
1
.
0
0
1
.
0
0
1
.
5
0
1
.
0
0
1
.
0
0
1
.
0
0
(
5
.
9
6
6
,
2
)
(
5
.
2
3
2
,
1
)
(
1
0
.
5
9
,
1
.
0
0
)
(
1
0
.
5
9
7
,
3
.
8
9
,
6
,
1
3
)
(
1
3
.
5
4
,
1
0
.
9
4
,
3
.
0
4
,
4
,
1
4
)
181
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
T
a
b
l
e
3
A
R
L
o
f
t
h
e
o
p
t
i
m
u
m
m
u
l
t
i
v
a
r
i
a
t
e
c
o
n
t
r
o
l
c
h
a
r
t
s
f
o
r
a
s
h
i
f
t
J
.
A
R
L
(
J
=
0
)
=
2
0
0
,
¡
=
5
,
n
=
2
S
y
n
t
h
e
t
i
c

T
2
S
y
n
t
h
e
t
i
c

T
2
V
S
S

T
2
D
S

T
2
T
2
(
L
C
s
y
n
t
,
L
)
(
L
C
s
y
n
t
,
L
)
M
E
W
M
A
(
C
L
.
n
.
n
1
.
n
2
)
(
h
.
h
1
.
n
1
.
n
1
.
n
2
)
J
(
C
L
)
Z
e
r
o
s
t
a
t
e
S
t
e
a
d
y
s
t
a
t
e
(
h
.
r
)
n
1
+
n
2

=
2
0
n
1
+
n
2

=
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
1
6
.
7
4
9
6
0
.
2
5
1
6
8
.
9
6
1
5
5
.
5
1
1
6
1
.
0
3
5
2
.
8
2
1
6
4
.
1
1
7
.
2
8
(
1
4
.
7
1
9
,
4
8
)
(
1
3
.
6
5
7
,
2
5
)
(
1
3
.
4
4
,
0
.
0
6
)
(
1
6
.
7
5
0
,
1
0
.
7
9
,
1
,
1
9
)
(
1
8
.
8
5
,
1
7
.
3
3
,
0
.
0
4
,
1
,
1
,
)
0
.
5
1
0
9
.
2
0
8
2
.
5
8
9
2
.
8
3
2
0
.
9
6
6
6
.
1
7
3
1
.
0
7
(
1
4
.
3
8
6
,
3
5
)
(
1
3
.
4
4
1
,
2
0
)
(
1
4
.
1
,
0
.
0
8
)
(
1
6
.
7
5
0
,
1
0
.
6
2
,
1
,
1
9
)
(
1
0
.
1
7
,
2
7
.
1
3
,
1
0
.
9
4
,
1
,
1
9
)
0
.
7
5
6
1
.
4
3
3
7
.
5
3
4
6
.
1
6
1
1
.
6
9
2
0
.
2
7
1
5
.
1
3
(
1
3
.
9
2
8
,
2
3
)
(
1
3
.
1
5
3
,
1
5
)
(
1
5
.
3
1
,
0
.
1
5
)
(
1
6
.
7
5
0
,
1
0
.
7
9
,
1
,
1
9
)
(
1
2
.
4
3
,
2
0
.
1
5
,
9
.
6
8
,
1
,
1
2
)
1
3
3
.
1
1
1
7
.
1
5
2
2
.
7
3
7
.
5
8
9
.
4
6
7
.
6
1
(
1
3
.
4
4
8
,
1
5
)
(
1
2
.
8
3
3
,
1
1
)
(
1
5
.
8
7
,
0
.
2
2
)
(
1
6
.
7
5
0
,
9
.
7
3
,
1
,
1
3
)
(
1
2
.
8
0
,
2
4
.
2
6
,
8
.
6
2
,
1
,
8
)
1
.
2
5
1
8
.
0
7
8
.
5
3
1
1
.
9
1
5
.
3
8
5
.
5
2
5
.
0
7
(
1
2
.
9
8
0
,
1
0
)
(
1
2
.
4
9
4
,
8
)
(
1
6
.
1
9
,
0
.
2
9
)
(
1
6
.
7
5
0
,
8
.
6
2
,
1
,
9
)
(
1
3
.
7
3
,
1
8
.
8
4
,
9
.
1
8
,
1
,
1
0
)
1
.
5
1
0
.
2
8
4
.
7
5
6
.
8
4
4
.
0
7
3
.
7
8
3
.
2
2
(
1
2
.
5
6
1
,
7
)
(
1
2
.
1
8
0
,
6
)
(
1
6
.
3
6
6
,
0
.
3
5
)
(
1
6
.
7
5
0
,
7
.
8
2
,
1
,
7
)
(
1
4
.
2
6
,
2
2
.
2
0
,
7
.
8
0
,
1
,
6
)
1
.
7
5
6
.
2
0
2
.
9
7
4
.
3
6
3
.
2
2
2
.
8
7
2
.
6
1
(
1
2
.
1
5
9
,
5
)
(
1
1
.
9
7
7
,
5
)
(
1
6
.
5
1
,
0
.
4
2
)
(
1
6
.
7
5
0
,
7
.
2
8
,
1
,
6
)
(
1
6
.
9
9
,
1
7
.
6
6
,
7
.
7
6
,
1
,
6
)
2
3
.
9
9
2
.
0
7
3
.
0
7
2
.
6
3
2
.
3
7
2
.
0
6
(
1
1
.
8
9
0
,
4
)
(
1
1
.
7
2
4
,
4
)
(
1
6
.
6
0
9
,
0
.
5
1
)
(
1
6
.
7
5
0
,
5
.
7
3
,
1
,
4
)
(
1
4
.
6
0
,
1
9
.
2
5
,
8
.
2
2
,
1
,
7
)
2
.
2
5
2
.
7
6
1
.
5
9
2
.
3
6
2
.
1
8
2
.
0
8
1
.
4
7
(
1
1
.
5
3
8
,
3
)
(
1
1
.
3
9
3
,
3
)
(
1
6
.
6
6
1
,
0
.
5
8
)
(
1
6
.
7
5
0
,
4
.
3
5
,
1
,
3
)
(
1
8
.
8
1
,
1
7
.
6
9
,
5
.
6
9
,
1
,
3
)
2
.
5
2
.
0
4
1
.
3
2
1
.
9
7
1
.
8
2
1
.
7
4
1
.
2
2
(
1
1
.
5
3
8
,
3
)
(
1
1
.
3
9
3
,
3
)
(
1
6
.
7
3
4
,
0
.
7
9
)
(
1
6
.
7
5
0
,
4
.
3
5
,
1
,
3
)
(
2
2
.
9
6
,
1
6
.
9
2
,
5
.
6
9
,
1
,
3
)
182
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
T
a
b
l
e
4
A
R
L
o
f
t
h
e
o
p
t
i
m
u
m
m
u
l
t
i
v
a
r
i
a
t
e
c
o
n
t
r
o
l
c
h
a
r
t
s
f
o
r
a
s
h
i
f
t
J
.
A
R
L
(
J
=
0
)
=
2
0
0
,
¡
=
5
,
n
=
7
S
y
n
t
h
e
t
i
c

T
2
S
y
n
t
h
e
t
i
c

T
2
V
S
S

T
2
D
S

T
2
T
2
(
L
C
s
y
n
t
,
L
)
(
L
C
s
y
n
t
,
L
)
M
E
W
M
A
(
C
L
.
n
.
n
1
.
n
2
)
(
h
.
h
1
.
n
1
.
n
1
.
n
2
)
J
(
C
L
)
Z
e
r
o
s
t
a
t
e
S
t
e
a
d
y
s
t
a
t
e
(
h
.
r
)
n
1
+
n
2

=
2
0
n
1
+
n
2

=
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
1
6
.
7
4
9
6
0
.
2
5
1
1
6
.
7
4
9
0
.
8
1
1
0
0
.
8
9
2
3
.
0
0
1
0
1
.
1
1
6
.
3
6
(
1
4
.
4
4
5
,
3
7
)
(
1
3
.
4
4
1
,
2
0
)
(
1
3
.
8
,
0
.
0
7
)
(
1
6
.
7
5
0
,
5
.
7
3
,
1
,
1
9
)
(
1
5
.
9
5
,
2
8
.
5
1
,
0
.
0
2
,
1
,
6
)
0
.
5
3
8
.
8
6
2
0
.
8
7
2
7
.
1
8
8
.
3
8
1
6
.
8
1
1
4
.
4
7
(
1
3
.
5
9
0
,
1
7
)
(
1
2
.
9
2
4
,
1
2
)
(
1
5
.
7
4
5
,
0
.
2
0
)
(
1
6
.
7
5
0
,
5
.
7
3
,
1
,
1
9
)
(
1
5
.
6
3
,
2
0
.
2
3
,
8
.
0
2
,
5
,
1
3
)
0
.
7
5
1
2
.
7
1
5
.
8
8
8
.
3
8
4
.
5
1
4
.
5
1
3
.
9
5
(
1
2
.
7
1
9
,
8
)
(
1
2
.
3
4
9
,
7
)
(
1
6
.
3
4
,
0
.
3
4
)
(
1
6
.
7
5
0
,
5
.
7
3
,
1
,
1
9
)
(
1
5
.
2
4
,
2
3
.
0
7
,
8
.
2
4
,
5
,
1
4
)
1
4
.
9
7
2
.
4
7
3
.
6
3
2
.
9
2
2
.
6
2
1
.
9
5
(
1
1
.
8
9
0
,
4
)
(
1
1
.
7
2
4
,
4
)
(
1
6
.
6
1
8
,
0
.
5
2
)
(
1
6
.
7
5
0
,
5
.
7
3
,
2
,
1
7
)
(
1
5
.
2
3
,
2
0
.
1
1
,
8
.
4
3
,
5
,
1
5
)
1
.
2
5
2
.
4
6
1
.
4
8
2
.
2
0
2
.
0
4
1
.
8
6
1
.
3
9
(
1
1
.
5
3
8
,
3
)
(
1
1
.
3
9
3
,
3
)
(
1
6
.
7
,
0
.
6
6
)
(
1
6
.
7
5
0
,
6
.
9
8
,
5
,
1
4
)
(
1
6
.
0
8
,
2
8
.
5
2
,
7
.
0
8
,
4
,
1
4
)
1
.
5
1
.
5
5
1
.
1
4
1
.
7
1
1
.
5
2
1
.
4
4
1
.
1
0
(
1
1
.
0
3
7
,
2
)
(
1
0
.
9
1
7
,
2
)
(
1
6
.
7
1
,
0
.
6
9
)
(
1
6
.
7
5
0
,
7
.
2
9
,
6
,
1
1
)
(
1
6
.
7
7
,
2
0
.
1
3
,
6
.
6
1
,
5
,
8
)
1
.
7
5
1
.
1
9
1
.
0
4
1
.
5
5
1
.
2
2
1
.
1
9
1
.
0
2
(
1
1
.
0
3
7
,
2
)
(
1
0
.
9
1
7
,
2
)
(
1
6
.
7
2
,
0
.
7
3
)
(
1
6
.
7
5
0
,
0
.
0
0
,
6
,
7
)
(
2
3
.
3
9
,
1
6
.
9
0
,
5
.
6
9
,
5
,
6
)
2
1
.
0
5
1
.
0
1
1
.
5
1
1
.
0
6
1
.
0
5
1
.
0
1
(
1
1
.
0
3
7
,
2
)
(
1
0
.
9
1
7
,
2
)
(
1
6
.
7
5
,
0
.
9
1
)
(
1
6
.
7
5
0
,
0
.
0
0
,
6
,
7
)
(
1
8
.
4
6
,
1
7
.
5
7
,
7
.
5
1
,
5
,
1
1
)
2
.
2
5
1
.
0
1
1
.
0
0
1
.
5
0
1
.
0
1
1
.
0
1
1
.
0
0
(
1
1
.
0
3
7
,
2
)
(
1
0
.
9
1
7
,
2
)
(
1
6
.
7
5
,
0
.
9
7
)
(
1
6
.
7
5
0
,
0
.
0
0
,
6
,
7
)
(
1
7
.
4
4
,
1
8
.
1
7
,
0
.
4
9
,
3
,
4
)
2
.
5
1
.
0
0
1
.
0
0
1
.
5
0
1
.
0
0
1
.
0
0
1
.
0
0
(
1
1
.
0
3
7
,
2
)
(
1
0
.
9
1
7
,
2
)
(
1
6
.
7
5
,
0
.
9
2
)
(
1
6
.
7
5
0
,
0
.
0
0
,
6
,
7
)
(
2
3
.
5
3
,
1
6
.
9
3
,
5
.
1
0
,
5
,
5
)
183
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
T
a
b
l
e
5
A
R
L
o
f
t
h
e
o
p
t
i
m
u
m
m
u
l
t
i
v
a
r
i
a
t
e
c
o
n
t
r
o
l
c
h
a
r
t
s
f
o
r
a
s
h
i
f
t
J
.
A
R
L
(
0
)
=
2
0
0
,
¡
=
1
0
,
n
=
2
S
y
n
t
h
e
t
i
c

T
2
S
y
n
t
h
e
t
i
c

T
2
V
S
S

T
2
D
S

T
2
T
2
(
L
C
s
y
n
t
,
L
)
(
L
C
s
y
n
t
,
L
)
M
E
W
M
A
(
C
L
.
n
.
n
1
.
n
2
)
(
h
.
h
1
.
n
1
.
n
1
.
n
2
)
J
(
C
L
)
Z
e
r
o
s
t
a
t
e
S
t
e
a
d
y
s
t
a
t
e
(
h
.
r
)
n
1
+
n
2

=
2
0
n
1
+
n
2

=
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
5
.
1
8
8
2
0
.
2
5
1
7
9
.
2
4
1
6
9
.
4
3
1
7
3
.
2
4
6
4
.
1
5
1
7
6
.
6
0
(
2
2
.
6
8
8
,
4
5
)
(
2
1
.
3
3
8
,
2
2
)
(
1
7
.
2
1
4
,
0
.
0
2
)
(
2
5
.
1
8
8
,
1
7
.
9
6
,
1
,
1
9
)
0
.
5
1
3
2
.
2
8
1
0
8
.
0
1
1
1
7
.
0
7
2
6
.
5
8
9
2
.
5
8
4
2
.
1
6
(
2
2
.
4
3
9
,
3
7
)
(
2
1
.
1
6
3
,
1
9
)
(
2
1
.
8
,
0
.
0
7
)
(
2
5
.
1
8
8
,
1
7
.
5
0
,
1
,
1
9
)
(
1
7
.
2
1
,
3
5
.
7
6
,
1
8
.
0
5
,
1
,
1
9
)
0
.
7
5
8
5
.
1
5
5
7
.
7
5
6
7
.
5
0
1
4
.
7
3
3
1
.
8
4
1
9
.
7
6
(
2
2
.
0
2
9
,
2
7
)
(
2
0
.
8
7
5
,
1
5
)
(
2
2
.
7
1
,
0
.
1
)
(
2
5
.
1
8
8
,
1
7
.
9
5
,
1
,
1
9
)
(
1
7
.
7
1
,
2
9
.
9
8
,
1
8
.
0
8
,
1
,
1
9
)
1
5
0
.
7
8
2
8
.
9
4
3
6
.
4
2
9
.
4
3
1
3
.
5
4
1
0
.
8
4
(
2
1
.
5
5
7
,
1
9
)
(
2
0
.
5
9
5
,
1
2
)
(
2
3
.
8
1
,
0
.
1
7
)
(
2
5
.
1
8
8
,
1
7
.
3
7
,
1
,
1
6
)
(
1
9
.
7
7
,
4
8
.
5
4
,
1
5
.
9
8
,
1
,
1
0
)
1
.
2
5
2
9
.
4
3
1
4
.
7
2
1
9
.
7
1
6
.
6
8
7
.
6
3
7
.
1
1
(
2
1
.
0
3
4
,
1
3
)
(
2
0
.
3
6
1
,
1
0
)
(
2
4
.
0
9
,
0
.
2
)
(
2
5
.
1
8
8
,
1
5
.
9
9
,
1
,
1
1
)
(
2
0
.
8
0
,
2
7
.
5
4
,
1
6
.
5
6
,
1
,
1
2
)
1
.
5
1
7
.
1
4
7
.
9
8
1
1
.
1
7
5
.
0
1
4
.
9
7
4
.
3
9
(
2
0
.
5
1
3
,
9
)
(
1
9
.
8
9
0
,
7
)
(
2
4
.
7
5
4
,
0
.
3
4
)
(
2
5
.
1
8
8
,
1
4
.
7
2
,
1
,
8
)
(
2
0
.
8
7
,
3
2
.
6
6
,
1
5
.
2
2
,
1
,
8
)
1
.
7
5
1
0
.
2
7
4
.
7
1
6
.
7
8
3
.
9
1
3
.
6
2
3
.
5
3
(
2
0
.
1
5
0
,
7
)
(
1
9
.
6
8
2
,
6
)
(
2
4
.
8
7
2
5
,
0
.
3
9
)
(
2
5
.
1
8
8
,
1
3
.
4
4
,
1
,
6
)
(
2
0
.
1
9
,
4
2
.
6
4
,
1
5
.
6
1
,
1
,
9
)
2
6
.
4
2
3
.
0
5
4
.
4
7
3
.
1
9
2
.
8
4
2
.
4
0
(
1
9
.
6
5
7
,
5
)
(
1
9
.
4
3
2
,
5
)
(
2
5
.
0
2
2
,
0
.
4
9
)
(
2
5
.
1
8
8
,
1
2
.
5
5
,
1
,
5
)
(
2
2
.
8
6
,
3
0
.
7
6
,
1
3
.
4
3
,
1
,
5
)
2
.
2
5
4
.
2
3
2
.
1
6
3
.
1
9
2
.
6
5
2
.
3
5
1
.
9
3
(
1
9
.
3
2
4
,
4
)
(
1
9
.
1
2
0
,
4
)
(
2
5
.
0
5
2
,
0
.
5
2
)
(
2
5
.
1
8
8
,
1
2
.
5
5
,
1
,
5
)
(
2
3
.
8
8
,
2
7
.
8
8
,
1
3
.
4
0
,
1
,
5
)
2
.
5
2
.
9
6
1
.
6
6
2
.
4
7
2
.
3
0
2
.
0
1
1
.
4
5
(
1
8
.
8
9
0
,
3
)
(
1
8
.
7
0
9
,
3
)
(
2
5
.
1
7
,
0
.
7
7
)
(
2
5
.
1
8
8
,
1
1
.
3
2
,
1
,
4
)
(
2
5
.
6
4
,
2
7
.
6
2
,
1
1
.
2
9
,
1
,
3
)
184
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
T
a
b
l
e
6
A
R
L
o
f
t
h
e
o
p
t
i
m
u
m
m
u
l
t
i
v
a
r
i
a
t
e
c
o
n
t
r
o
l
c
h
a
r
t
s
f
o
r
a
s
h
i
f
t
J
.
A
R
L
(
J
=
0
)
=
2
0
0
,
¡
=
1
0
,
n
=
7
S
y
n
t
h
e
t
i
c

T
2
S
y
n
t
h
e
t
i
c

T
2
V
S
S

T
2
D
S

T
2
T
2
(
L
C
s
y
n
t
,
L
)
(
L
C
s
y
n
t
,
L
)
M
E
W
M
A
(
C
L
.
n
.
n
1
.
n
2
)
(
h
.
h
1
.
n
1
.
n
1
.
n
2
)
J
(
C
L
)
Z
e
r
o
s
t
a
t
e
S
t
e
a
d
y
s
t
a
t
e
(
h
.
r
)
n
1
+
n
2

=
2
0
n
1
+
n
2

=
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
5
.
1
8
8
2
0
.
2
5
1
3
8
.
8
3
1
1
5
.
9
1
1
2
4
.
5
1
2
9
.
3
6
1
2
7
.
6
0
6
.
9
1
(
2
2
.
4
7
4
,
3
8
)
(
2
1
.
1
6
3
,
1
9
)
(
2
0
.
8
0
,
0
.
0
5
)
(
2
5
.
1
8
8
,
1
1
.
3
2
,
1
,
1
9
)
(
2
5
.
9
6
,
2
9
.
0
5
,
0
.
0
4
,
2
,
5
)
0
.
5
5
8
.
2
9
3
4
.
6
3
4
2
.
8
0
1
0
.
5
4
3
1
.
9
7
2
4
.
5
2
(
2
1
.
6
9
3
,
2
1
)
(
2
0
.
6
9
6
,
1
3
)
(
2
3
.
2
9
6
,
0
.
1
3
)
(
2
5
.
1
8
8
,
1
1
.
6
0
,
2
,
1
8
)
(
2
3
.
5
5
,
3
3
.
5
5
,
1
5
.
6
1
,
6
,
9
)
0
.
7
5
2
1
.
0
9
1
0
.
0
3
1
3
.
8
1
5
.
5
8
7
.
0
5
7
.
5
9
(
2
0
.
7
9
9
,
1
1
)
(
2
0
.
0
6
8
,
8
)
(
2
4
.
2
9
4
,
0
.
2
3
)
(
2
5
.
1
8
8
,
1
1
.
6
0
,
2
,
1
8
)
(
2
6
.
2
7
,
2
6
.
1
6
,
1
6
.
1
8
,
6
,
1
1
)
1
8
.
1
3
3
.
7
7
5
.
4
8
3
.
7
0
3
.
3
4
2
.
9
4
(
1
9
.
9
2
5
,
6
)
(
1
9
.
4
3
2
,
5
)
(
2
4
.
4
8
,
0
.
2
6
)
(
2
5
.
1
8
8
,
1
1
.
6
7
,
3
,
1
6
)
(
2
5
.
9
3
,
2
6
.
8
8
,
1
2
.
5
0
,
4
,
1
2
)
1
.
2
5
3
.
7
0
1
.
9
5
2
.
8
9
2
.
4
9
2
.
2
1
1
.
7
7
(
1
9
.
3
2
4
,
4
)
(
1
9
.
1
2
0
,
4
)
(
2
5
.
1
,
0
.
5
8
)
(
2
5
.
1
8
8
,
1
2
.
1
9
,
4
,
1
5
)
(
2
5
.
8
6
,
2
6
.
5
6
,
1
2
.
3
3
,
3
,
1
5
)
1
.
5
2
.
0
7
1
.
3
3
1
.
9
8
1
.
8
2
1
.
7
0
1
.
2
2
(
1
8
.
8
9
0
,
3
)
(
1
8
.
7
0
9
,
3
)
(
2
5
.
1
7
,
0
.
7
7
)
(
2
5
.
1
8
8
,
1
4
.
1
4
,
6
,
1
2
)
(
3
7
.
2
7
,
2
5
.
2
4
,
1
3
.
3
4
,
5
,
1
0
)
1
.
7
5
1
.
4
1
1
.
1
0
1
.
6
5
1
.
3
9
1
.
3
7
1
.
1
2
(
1
8
.
2
6
6
,
2
)
(
1
8
.
1
1
5
,
2
)
(
2
5
.
1
7
,
0
.
7
8
)
(
2
5
.
1
8
8
,
1
4
.
1
4
,
6
,
1
2
)
(
2
1
.
7
9
,
3
1
.
3
2
,
1
7
.
1
1
,
6
,
1
4
)
2
1
.
1
4
1
.
0
3
1
.
5
4
1
.
1
4
1
.
1
4
1
.
0
5
(
1
8
.
2
6
6
,
2
)
(
1
8
.
1
1
5
,
2
)
(
2
5
.
1
8
,
0
.
8
5
)
(
2
5
.
1
8
8
,
0
.
0
0
,
6
,
7
)
(
2
3
.
9
1
,
2
7
.
7
8
,
1
4
.
9
1
,
5
,
1
5
)
2
.
2
5
1
.
0
4
1
.
0
1
.
5
1
1
.
0
4
1
.
0
4
1
.
0
1
(
1
8
.
2
6
6
,
2
)
(
1
8
.
1
1
5
,
2
)
(
2
5
.
1
9
,
0
.
9
7
)
(
2
5
.
1
8
8
,
0
.
0
0
,
6
,
7
)
(
2
6
.
2
4
,
2
6
.
5
8
,
1
3
.
7
4
,
5
,
1
1
)
2
.
5
1
.
0
1
1
.
0
0
1
.
5
0
1
.
0
1
1
.
0
1
1
.
0
0
(
1
8
.
2
6
6
,
2
)
(
1
8
.
1
1
5
,
2
)
(
2
5
.
1
8
7
,
0
.
9
4
)
(
2
5
.
1
8
8
,
0
.
0
0
,
6
,
7
)
(
2
6
.
4
9
,
2
8
.
2
8
,
0
.
1
4
,
3
,
4
)
185
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
186 Aparisi and de Luna
The optimum design of the syntheticT
2
chart has been utilized in the comparison
for different cases of ARL
S−T
2 (J = 0) = ARL
T
2 (J = 0), n, ¡, and J. The ARL
values are shown in the two ﬁrst columns of Tables 1–6. The percentage of
improvement for the ARL is calculated as:
% of improvement =
ARL
T
2 (J = 0) −ARL
S−T
2 (J = 0)
ARL
T
2 (J = 0)
∗ 100. (2)
A list of the key conclusions will be discussed below. To observe some of the
conclusions, more cases (tables) may be needed. A very large number of cases have
been studied, however, due to size limitations, only six tables are shown in this
article. The rest of tables are available from the authors.
1. The syntheticT
2
control chart optimum for a shift J is always faster
to detect than the Hotelling’s T
2
chart. For large shifts, J > 2, the percentage
of improvement of the syntheticT
2
chart is better when the sample size is
small. For large sample sizes the performance of both charts is similar for
large shifts. In some cases, the improvement is quite signiﬁcant. For example,
when ARL
S−T
2 (J = 0) = ARL
T
2 (J = 0) = 200, ¡ = 2, n = 2, and J = 1.25 the
improvement in the ARL is 53.66%, with ARL values of ARL
T
2 (J = 1.25) = 9.91
and ARL
S−T
2 (J = 1.25) = 4.59. For the same case the percentage of improvement
is larger than 40% for the range 0.75 ≤ J ≤ 2.
2. If the syntheticT
2
chart is designed to minimize the ARL for a given shift J,
the optimized chart is also faster than the standard T
2
chart for the rest of values
of J, 0  J ≤ 3.
3. The percentage improvement for the syntheticT
2
, given the same set of
values of ¡, n and J, increases as the value of ARL(J = 0) is larger.
4. For the same values of ¡ and ARL(J = 0), the optimum parameters of the
syntheticT
2
charts for J are constants, independently of the sample size (if n ≥ 2)
and magnitude of shift, J, if the value of J is large. The parameter L takes small
values if the chart is designed for large values of J. but L tends to be small if the
chart is optimized for large values of J.
5. Inﬂuence of the sample size. For n = 1, 2, or 3, the optimum syntheticT
2
for a shift J is always faster than the standard T
2
for detecting shifts of magnitude
0  J ≤ 3. For sample sizes n ≥ 4, the performance of the optimum syntheticT
2
chart is better for small and moderate shifts, but the improvement is marginal
for large shifts. For example, if n = 7 the improvement is negligible if J ≥ 2.25.
This behavior is logical because the standard T
2
shows ARL values close to one.
As the sample size increases, the range of values of J where the syntheticT
2
chart
has better performance is narrower. Figure 3 shows the ARL values of the optimum
syntheticT
2
charts for a shift J. The curves shown are for the cases ARL(J = 0) =
200, ¡ = 2, n ≤ 7, and J = 0.25. 0.5. . . . . 2.
For small and moderate shifts (J ≤ 0.75), and for the same set of values of
ARL(J = 0), ¡ and J, a larger sample size increases the percentage of improvement.
However, when J ≥ 1.75 the percentage improvement for the syntheticT
2
chart is
smaller when the sample size increases.
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
The SyntheticT
2
Control Chart 187
Figure 3. ARL of the optimum syntheticT
2
chart for a shift J. ARL
S−T
(J = 0) = 200, ¡ = 2.
6. Inﬂuence of the number of variables. The percentage improvement for the
syntheticT
2
chart depends of the number of variables, ¡. For the same values
of n and J the percentage improvement is smaller when more variables are to
be controlled if J ≤ 0.75. However, when J ≥ 1.5 and n ≥ 3, the percentage of
improvement is larger when the number of variables increases.
5.2.1. Comparison versus MEWMA, VSST
2
, and DST
2
Charts. In this section, a
comparison of the performance (ARL) of the optimum zerostate syntheticT
2
chart
for a shift J, versus the optimum MEWMA, VSST
2
, and DST
2
charts for the same
J is carried out. For the VSST
2
and DST
2
charts, the restrictions E(n) = n
0
and
n
1
+n
2
≤ 20 must be fulﬁlled. Given these restrictions, ﬁrstly it is guaranteed that,
on average, the same sample size is employed. Secondly, it is very frequent that
in order to ﬁnd the optimum charts to minimize ARL(J = 0) for the VSST
2
and
DST
2
charts, large values of n
2
are obtained, producing solutions that may not be
feasible in real applications.
Tables 1–6 show the optimum parameters of the control charts that are used
in the comparison and the value of ARL(J = 0) for J = 0.25. 0.5. . . . . 2.5 and for
some combinations of ¡ and n when ARL(J = 0) = 200. The authors have obtained
more tables for the following cases: ¡ = 2, 3, 4, 5, and 10, n = 1. 2. 3. 5, and 7,
ARL(J = 0) = 200 and 400 and J = 0.25. 0.5. . . . . 2.5.
The optimum syntheticT
2
chart for a shift J is faster than the MEWMA and
VSST
2
charts when ¡ = 2 and n is large. For example, when ¡ = 2 and n = 7 the
syntheticT
2
chart is superior or equal than the MEWMA chart if J ≥ 0.75. If, on
the other hand, considering the same case, the syntheticT
2
chart is superior or
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
188 Aparisi and de Luna
similar than the VSST
2
chart when J ≥ 1. However, the range where the optimum
syntheticT
2
chart is superior to these charts worsens as the sample size decreases
and the number of variables increases. For example, if ¡ = 10 and n = 1, the
ARL
S−T
2 (J = 0) is lower than the ARL of MEWMA if J ≥ 2.75.
There are some cases where the MEWMA and VSST
2
optimized charts for a
shift J have an ARL value larger or equal than the standard T
2
chart. However, the
syntheticT
2
and DST
2
optimized charts for a shift J always have lower or equal
ARL values in comparison with the standard T
2
chart. If the syntheticT
2
chart is
designed to minimize the ARL for a shift J, for the rest of size of shifts, 0  J ≤ 3,
it will be always faster to detect these shifts than the T
2
chart. This behavior does
not happen sometimes with the MEWMA and VSST
2
charts.
In general, the performance of the MEWMA and VSST
2
charts is better for
detecting small shifts than the syntheticT
2
chart, but this is may not be true
for moderate and large shifts. For a given value of ARL(J = 0) and number of
quality variables, the range where it is better to use the syntheticT
2
chart instead of
the MEWMA control chart increases as the sample size increases. The syntheticT
2
chart is faster than the VSST
2
chart when J = 0.25. However, for this same value,
the lowest ARL value is obtained by the DST
2
chart. For very small shifts the
DST
2
chart has better performance than the MEWMA.
When n ≥ 3 and ¡ ≤ 10, the syntheticT
2
chart is faster than the MEWMA and
VSST
2
charts for shifts of magnitude J ≥ 1.75, although the range of J where the
performance of the syntheticT
2
is superior is incremented when the sample size
increases and the number of variables is smaller.
The DST
2
chart has better performance than the syntheticT
2
for J ≤ 1,
independently of the sample size and the number of variables. The performance of
these two charts, for moderate and large shifts, is again very similar as the sample
size increases and the number of variables decreases.
Figure 4 shows the ARL curves of the optimum charts designed to have the
best performance for J = 0.75, for the case ARL(J = 0) = 200, ¡ = 2, and n = 7
(Table 2). It is possible to see that the ARL curve of the syntheticT
2
chart
is always below the curve of the Hotelling’s T
2
chart. However, this does not
happen with the curves of the MEWMA and VSST
2
charts. The ARL of the
syntheticT
2
is lower than the ARL of the MEWMA and VSST
2
charts when
J ≥ 0.8. The DST
2
control chart is the chart that exhibits better performance in
detecting simultaneously very small and large shifts. For moderate and large shifts,
the performance of the syntheticT
2
chart is similar to the DST
2
control chart.
6. Conclusions
In this article, the syntheticT
2
control chart has been developed, ﬁnding a procedure
to obtain the optimum parameters of this chart to minimize the outofcontrol ARL,
given a value of incontrol ARL value and a magnitude of shift in the mean vector.
The ARLs have been calculated employing the Markov chain approach for the
steady and zero states.
It may be advisable to employ the syntheticT
2
control chart as an alternative to
the standard T
2
chart, because the former is always faster than the latter in detecting
the shifts in the mean vector for all the values of magnitude of shift. One important
property of the zerostate syntheticT
2
chart is that an optimized chart for a given
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
The SyntheticT
2
Control Chart 189
Figure 4. ARL curves of the optimum control charts for a shift J = 0.75. ARL(J = 0) =
200, ¡ = 2, n = 7.
magnitude of shift always gives a better performance, for all the values of shift, than
the equivalent Hotelling’s control chart.
Although its real application can be more complicated than the standard
T
2
chart, the improved performance, and the possibility of continuing employing
the tools for the interpretation of the outofcontrol signal of the Hotelling’s T
2
chart, make the utilization of the syntheticT
2
control chart an interesting one.
For example, a user interested in ﬁnding out the variables that have produced
the outofcontrol signal can apply the MTY decomposition method (Mason and
Young, 1999; Mason et al., 1997) or the use of neural networks (Aparisi et al., 2006).
The syntheticT
2
chart may provide a good alternative to the DST
2
chart in
order to detect moderate and large shifts in the mean vector, having the advantage
of being easier to employ in real applications. For the detection of very small
shifts (J = 0.25), we recommend the DST
2
. In general, for very small shifts the
syntheticT
2
chart cannot compete with the MEWMA, VSST
2
, and DSST
2
charts,
but it has a much better performance in comparison with the Hotelling’s T
2
chart.
Given moderate and large shifts, these are cases where the syntheticT
2
chart is
superior to that of the MEWMA or VSST
2
charts.
Appendix: Markov Chain Model
Let A = P(T
2
i
 LCsxnt) and B = 1 −A. As Davis and Woodall (2002) suggested,
the following transition matrix would govern the Markov chain for the multivariate
syntheticT
2
control chart:
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
190 Aparisi and de Luna
1. The ﬁrst row contains A in the ﬁrst column and B in the second column.
2. The last row contains A in the ﬁrst column.
3. In all the other rows, the entry above the diagonal is A.
4. In all the other locations, the entry is zero.
Therefore, for example, the transition matrix for the syntheticT
2
chart when
L = 5 is:
00000 00001 00010 00100 01000 10000 Signal
00000 A B 0 0 0 0 0
00001 0 0 A 0 0 0 B
00010 0 0 0 A 0 0 B
00100 0 0 0 0 A 0 B
01000 0 0 0 0 0 A B
10000 A 0 0 0 0 0 B
Signal 0 0 0 0 0 0 1
When the process is in control, B=1 −P(X ≤LCsxnt) =1 −
LCsxnt
0
1
!(
¡
2
)
√
2
¡
·
x
¡
2
−1
e
−
x
2
Jx, where ¡ is the number of variables, LCsynt is the control limit of the T
2
subchart, and !(o) is the gamma of Euler.
In order to obtain the outofcontrol ARL, B is computed employing the
cumulative distribution function of the non central chisquare with ¡ degrees
of freedom and non centrality parameter z = nJ
2
, where n is the sample
size and J is the Mahalanobis’ distance. Hence, B = 1 −P(T
2
i
≤ LCsxnt) =
P(,
2
¡
(z = nJ
2
) ≤ LCsxnt) and A = 1 −B.
With this Markov chain model the ARL for the zerostate case is ARL
zero
=
s
(I −R)
−1
1, where 1 is a column vector of “ones” of size (L +1), I is the identity
matrix of size (L +1)(L +1), s is the vector (L +1) of initial probabilities, 1 for the
initial state and 0 for the rest of cases, s
= 0. 1. 0 . . . 0. 0], and the R submatrix
(L +1)*(L +1) has the transition probabilities without considering the absorbent
state.
The ARL of the multivariate syntheticT
2
for the steadystate case is
obtained considering SS = ¬
0
. ¬
1
. . . . ¬
L
]
T
and ARL
steady
= SS ×(I −R)
−1
1, where
the matrices I, R, and 1 are the aforementioned and ¬
i
is the steadystate probability
for the state i,
L
i=0
¬
i
= 1.
In order to calculate the SS vector, it is recommended employing a cyclic
probability vector proposed by Lucas and Saccucci (1990) and simpliﬁed by Champ
(1992). It is also possible to use an approximation obtaining the probabilities
following the procedure employed by Celano et al. (2006). Following this last
approach, it is possible to prove that the steadystate probabilities are a function of
the parameters L and LCsynt of the syntheticT
2
control chart.
Acknowledgments
This work has been supported by the Ministry of Education and Science of Spain,
research project number DPI200606124 including European FEDER funding, and
the support of the ITESMFoundation Carolina agreement.
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
The SyntheticT
2
Control Chart 191
References
Aparisi, F. (1996). Hotelling’s T
2
control chart with adaptive sample sizes. Int. J. Product.
Res. 34(10):2853–2862.
Aparisi, F., Haro, C. L. (2003). A comparison of T
2
control charts with variable sampling
schemes as opposed to MEWMA chart. Int. J. Product. Res. 41(10):2169–2182.
Aparisi, F., GarcíaDíaz, C. (2004). Optimization of univariate and multivariate
exponentially weighted movingaverage control charts using genetic algorithms.
Comput. Operat. Res. 31(9):1437–1454.
Aparisi, F., Avendaño, G., Sanz, J. (2006). Interpreting T
2
control signal: effectiveness of
MTY decomposition vs. a neural network. IIE Trans. 38(8):647–657.
Atienza, O. O., Tang, L. C., Ang, B. W. (1998). Simultaneous monitoring of univariate and
multivariate SPC information using boxplots. Int. J. Qual. Sci. 3(2):194–204.
Celano, G., Costa, A., Fichera, S. (2006). Statistical design of variable sample size and
sampling interval X control charts with run rules. Int. J. Adv. Manufact. Technol.
28:966–977.
Champ, C. W. (1992). Steadystate run length analysis of a Shewhart quality control chart
with supplementary runs rules. Commun. Statist. Theor. Meth. 21(3):765–777.
Champ, C. W., Aparisi, F. (2006). Double sampling Hotelling’s T
2
charts. Qual. Reliab. Eng.
Int. 24:153–166.
Champ, C. W., JonesFarmer, L. A., Rigdon, S. (2005). Properties of the T
2
control chart
when parameters are estimated. Technometrics 47(4):437–445.
Chen, F. L., Huang, H. J. (2005a). A synthetic control chart for monitoring process
dispersion with sample standard deviation. Comput. Industr. Eng. 49(2):221–240.
Chen, F. L., Huang, H. J. (2005b). A synthetic control chart for monitoring process
dispersion with sample range. Int. J. Adv. Manufact. Technol. 26(7–8):842–851.
Chen, F. L., Huang, H. J. (2006). Variable sampling interval synthetic control charts for
jointly monitoring process mean and standard deviation. Int. J. Industr. Eng. 13(2):
136–146.
Costa, A. F. B. (1994). X charts with variable sample size. J. Qual. Technol. 28(3):155–163.
Costa, A. F. B., Rahim, M. A. (2001). Economic design of X charts with variable parameters:
the Markov chain approach. J. Appl. Statist. 28(7):875–885.
Costa, A. F. B., Rahim, M. A. (2006). A synthetic control chart for monitoring the process
mean and variance. J. Qual. Maint. Eng. 12(1):81–88.
Daudin, J. J. (1992). Double sampling X charts. J. Qual. Technol. 24(2):78–87.
Davis, R. B., Woodall, W. H. (2002). Evaluating and improving the synthetic control chart.
J. Qual. Technol. 34(2):200–208.
Ghute, V. B., Shirke, D. T. (2008). A Multivariate synthetic control chart for monitoring
processes mean vector. Commun. Statist. 37:2136–2148.
He, D., Grigoryan, A. (2005). Multivariate multiple sampling charts. IIE Trans. 37(6):
509–521.
He, D., Grigoryan A. (2006). Joint statistical design of double sampling X and S charts.
Eur. J. Operat. Res. 168(1):122–142.
Hotelling, H. (1947). Multivariate quality control. In: Eisenhart, C., Hastay, M. W., Wallis,
W. A., eds. Techniques of Statistical Analysis. New York: McGrawHill, pp. 111–184.
Jones, L. A., Champ, C. W., Rigdon, S. E. (2004). The run length distribution of the
CUSUM with estimated parameters. J. Qual. Technol. 36(1):95–108.
Khoo, M., Quah, S. (2002). Computing the percentage points of the runlength distributions
of multivariate CUSUM control charts. Qual. Eng. 15(2):299–310.
Klein, M. (1997). Modiﬁed Shewhartexponentially weighted moving average control charts.
IIE Trans. 29(12):1051–1056.
Knoth, S. (2005). Fast initial response features for EWMA control charts. Statist. Pap.
46(1):47–64.
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
192 Aparisi and de Luna
Kourti, T., MacGregor, J. F. (2004). Multivariate statistical process control and industrial
applications. Qual. Congr. ASQ World Conf. Qual. Improve. Proc. 58:449–452.
Lowry, C. A., Montgomery, D. C. (1995). A review of multivariate control chart. IIE Trans.
27(6):800–810.
Lowry, C. A., Woodall, W. H., Champ, C. W., Rigdon, S. E. (1992). A multivariate
exponentially weightes moving average control chart. Technometrics 34:46–53.
Lucas, J. M. (1982). Combined ShewhartCUSUM quality control schemes. J. Qual.
Technology 14(2):51–59.
Lucas, J. S., Saccucci, M. S. (1990). Exponentially weighted moving average control schemes:
properties and enhancements. Technometrics 32(1):1–12.
Luceno, A., PuigPey, J. (2000). Evaluation of the runlength probability distribution for
CUSUM charts: assessing chart performance. Technometrics 42(4):411–416.
Mason, R. L., Young, J. C. (1999). Improving the sensitivity of the T
2
statistic in
multivariate process control. J. Qual. Technol. 31(2):155–165.
Mason, R. L., Tracy, N. D., Young, J. C. (1997). A practical approach for interpreting
multivariate T
2
control chart signals. J. Qual. Technol. 29(4):396–406.
Montgomery, D. C. (2005). Introduction to Statistical Quality Control. 5th ed. New York: John
Wiley & Sons.
Prabhu, S. S., Runger, G. C. (1997). Designing a multivariate EWMA control chart. J. Qual.
Technol. 29(1):8–15.
Reynolds, Jr., M. R., Arnold, J. C. (2001). EWMA control charts with variable sample sizes
and variable sampling intervals. IIE Trans. 33(6):511–530.
Vargas, J. A. (2003). Robust estimation in multivariate control charts for individual
observations. J. Qual. Technol. 35(4):367–376.
Williams, J. D., Woodall, W. H., Birch, J. B., Sullivan, J. H. (2006). Distribution of
Hotelling’s T
2
statistic based on the successive differences estimator. J. Qual. Technol.
38(3):217–229.
Woodall, W. H., Ncube, M. M. (1985). Multivariate CUSUM qualitycontrol procedures.
Technometrics 27(3):285–292.
Wu, Z., Spedding, T. A. (2000). A synthetic control chart for detecting small shifts in the
process mean. J. Qual. Technol. 32(1):32–38.
Wu, Z., Yeo, S. H., Spedding, T. A. (2001). A synthetic control chart for detecting fraction
nonconforming increases. J. Qual. Technol. 33(1):104–111.
Zhang, S., Wu, Z. (2005). Designs of control charts with supplementary runs rules. Comput.
Industr. Eng. 49(1):76–97.
D
o
w
n
l
o
a
d
e
d
B
y
:
[
B

o
n
C
o
n
s
o
r
t
i
u
m

2
0
0
7
]
A
t
:
2
2
:
3
0
6
J
u
n
e
2
0
1
0
Communications in Statistics—Theory and Methods, 38: 173–192, 2009 Copyright © Taylor & Francis Group, LLC ISSN: 03610926 print/1532415X online DOI: 10.1080/03610920802178413
The Design and Performance of the Multivariate SyntheticT 2 Control Chart
Downloaded By: [Bon Consortium  2007] At: 22:30 6 June 2010
FRANCISCO APARISI1 AND MARCO A. DE LUNA2
1
Departamento de Estadística e Investigación Operativa Aplicadas y Calidad, Universidad Politécnica de Valencia, Valencia, España 2 Departamento de Ingeniería Industrial y Mecánica, ITESM, Guadalajara, México
One of the objectives of research in statistical process control is to obtain control charts that show few false alarms but, at the same time, are able to detect quickly the shifts in the distribution of the quality variables employed to monitor a productive process. In this article, the syntheticT 2 control chart is developed, which consists of the simultaneous use of a CRL chart and a Hotelling’s T 2 control chart. The ARL is calculated employing Markov chains for steady and zerostate scenarios. A procedure of optimization has been developed to obtain the optimum parameters of the syntheticT 2 , for zero and steady cases, given the values of incontrol ARL and magnitude of shift which needs to be detected rapidly. A comparison between (standard T 2 , MEWMA, T 2 with variable sample size, and T 2 with double sampling) charts reveals that the syntheticT 2 chart always performs better than the standard T 2 chart. The comparison with the remaining charts demonstrate in which cases the performance of this new chart makes it interesting to employ in real applications. Keywords ARL; Multivariate; Quality control; SPC; Synthetic. Mathematics Subject Classiﬁcation 62P30.
1. Introduction
Statistical quality control has the objective of detecting shifts in the distribution of the monitored quality variables. In the majority of production processes the possible spontaneous changes in this distribution will worsen the quality of the product manufactured. Therefore, the quick detection of these shifts is very important to maintain quality. The statistical process control (SPC) has shown itself to be a powerful tool in achieving this objective. It is possible to employ two strategies when it is desired to control concurrently p variables from one component or from a production process: to use simultaneously
Received August 21, 2007; Accepted May 2, 2008 Address correspondence to Francisco Aparisi, Departamento de Estadística e Investigación Operativa Aplicadas y Calidad, Universidad Politécnica de Valencia, Valencia 46022, España; Email: faparisi@eio.upv.es
173
much of research has been made in order to design control charts capable of fast detection of small shifts in the process. The problem is that the standard control charts (X in the univariate case and Hotelling’s T 2 in the multivariate case) are not efﬁcient at detecting small shifts. Lucas and Saccucci. 5. The ﬁrst proposal to control possible shifts . Reynolds and Arnold. 1997). 1992. 2004. 2006) and DS multivariate charts (Champ and Aparisi. Woodall and Ncube. in the univariate case the CUSUM charts (Jones et al. Many strategies have been proposed with the aim of improving the ARL values when the process is out of control while keeping a desired ARL value when the process is in control. Ghute and Shirke (2008).. Multivariate Quality Control Multivariate quality control consists of monitoring in one chart p variables of the same production process or component. The syntheticX chart was introduced by Wu and Spedding (2000) as a control chart useful for detecting small shifts in the mean of a process. He and Grigoryan. In Sec. 2. Luceno and PuigPey. Wu and Spedding. the average number of points in the chart until the ﬁrst outofcontrol signal appears. 2000) and in the multivariate case the MCUSUM charts (Khoo and Quah. From our point of view.2007] At: 22:30 6 June 2010 1. a procedure will be developed for ﬁnding the parameters of the multivariate synthetic T square chart. 1982). for the steady and zero states. Lowry et al. 1990) and MEWMA charts for the multivariate case (Aparisi and GarcíaDíaz. the ShewhartEWMA chart (Klein. 2001. Cumulative sums. Examples of this strategy are: the ShewhartCUSUM chart (Lucas. 2005. 3. In addition. 4. Exponentially weighted averages. Variable sample size (VSS) univariate charts (Costa.174 Aparisi and de Luna p univariate control charts or to employ only one multivariate control chart (Montgomery. Over the years. 1994. Synthetic charts for the univariate case (Chen and Huang. a comparison versus other multivariate control charts is made. 4. the EWMA charts in the univariate case (Knoth. In this article. in order to ﬁnd out in which cases the syntheticT 2 chart shows better performance. and the synthetic chart. Section 2 shows a summary of multivariate SPC and the methodologies which enable more efﬁcient control charts to be obtained. Prabhu and Runger. which is a ShewhartCRL (conforming run length) chart. The efﬁcacy of a quality control chart is normally measured using the ARL (Average Run Length). 1997). 1985).. The performance of the SPC has been improved employing simultaneously two control charts. the more efﬁcient strategies are: Downloaded By: [Bon Consortium . 6. Section 5 shows the comparison between several multivariate control charts. 2002. 1996. that minimizes the ARL for a given size of shift. 2003). 2005). 2001) and VSS multivariate charts (Aparisi. 2006. He and Grigoryan. The optimization of the parameters of this chart can be found in Sec. given a desired incontrol ARL. 3. the syntheticT 2 chart is deﬁned. 2. Double sampling (DS) univariate charts (Daudin. 2000). 2004. Aparisi and Haro. 1992. The conclusions can be found in Sec. 2005a. Costa and Rahim. 2005).
d = 1 − 0 1 − 0 . The strategy consists of determining the value 2 of a warning limit (w). 2 Given the value of . is a chisquare with p degrees of freedom. 2 2 If Ti−1 ≤ w. The distribution of the statistic Ti2 . The analysis is done employing the zerostate and steadystate approaches (Champ. chisquare. Xi is the mean vector of the p sample means. 2008) is proposed to reduce the values of ARL to detect process shifts computing the ARL employing Markov chains. The performance of this chart to detect shifts in the mean vector depends on the non centrality parameter = nd2 . a sample of size n1 at the time i is taken. The T 2 chart has a poor performance to detect small shifts in the mean vector. Therefore. . to ﬁnd the values of CL w n1 . When Ti2 > Xp . 2 As the points plotted in the Hotelling’s T chart are independent.2007] At: 22:30 6 June 2010 where n is the sample size. 2. the control limit is CL = Xp . where −1 T d is the Mahalanobis’ distance. it is easy to obtain the control limit (CL) ﬁxing the probability of the Type I error. 2005) and considering that 0 and 0 are known. Variable Sample Size T 2 Control Chart Aparisi (1996) and Aparisi and Haro (2003) improved the sensibility of the T 2 chart employing variable sample size (VSS). Ti2 = n Xi − T 0 0 −1 Xi − 0 (1) Downloaded By: [Bon Consortium . Zhang and Wu. Otherwise.1. there will be a description of the multivariate control charts that are going to be used in the comparison of performance shown in Sec. it is decided that only random causes are present in the process. otherwise. Lowry and Montgomery (1995) and Champ et al. where 1 is the 0 1 outofcontrol mean vector. 1992. (2005) presented guides for the design of the T 2 control chart with estimated parameters. Subsequently. for a given incontrol ARL. (2006) studied alternatives to estimate the covariance matrix for the case of individual multivariate observations. Vargas (2003) and Williams et al. if w < Ti−1 < CL 2 the sample size to be used is n2 . 1947) which consists of plotting the values of the statistic Ti2 . . Ghute and Shirke (2008) did not show results for small shifts and the ARL is only computed for the zerostate case. a non central 2 = nd2 . T 2 ≈ Xp Commonly. knowing the distribution of the Ti2 statistic when the process is out of control. It is possible. Davis and Woodall (2002) noted the importance of studying the synthetic charts for the steadystate case. the mean vector and the covariance matrix are estimated. The sample size at time i depends on the value of Ti−1 . The value of ARL when d = 0 is ARL d = 0 = 1− (where is the probability of the Type II error) is easily calculated. if the process is in an incontrol state and 0 and 0 are known. In this article. the process is considered to be out of control. Davis and Woodall. 0 is the mean vector when the process is in control. n0 . 5. and 0 is the incontrol covariance matrix. the incontrol ARL is given by ARL0 = 1 . and n2 that minimizes the outofcontrol ARL for a speciﬁed size of magnitude d. considering the restrictions n1 + n2 ≤ nmax and a desired mean sample size. the use of a syntheticT 2 control chart (Ghute and Shirke. 2002.The SyntheticT 2 Control Chart 175 in the mean vector was Hotelling’s T 2 control chart (Hotelling.
The SyntheticT 2 Control Chart The univariate synthetic chart was introduced by Wu and Spedding (2000) as an alternative to improve the performance of the Shewhart control chart to detect process shifts. where Z0 = 0 . n1 . It is the result of combining a Shewhart chart and CRL chart (a chart originally designed to detect increments in the percentage of defective units). 1992). If Ti2 > h. the parameters of the optimum DST 2 chart to detect a shift of given magnitude are found. Double Sampling T 2 Control Chart The strategy of double sampling (DS) consists of taking two samples of sizes n1 and n2 from the process at the same time. the resultant chart is the Hotelling’s control chart.3. MEWMA (Lowry et al.. it is decided that the process is outofcontrol. The two samples are combined employing Xi = n +n n1 Xi 1 + n2 Xi 2 . is a chart where the plotted statistic takes into account the information from past samples. for any shift magnitude. Downloaded By: [Bon Consortium . otherwise the process is deemed to be in an incontrol state. r (0 < r ≤ 1). if w1 ≤ Ti21 < h1 it is not possible to make a decision and the second sample of size 1 n2 is studied.2. The exact covariance matrix of Zi is computed 1−r 2i according to zi = r 1−2−r . Ti2 is calculated using the expression Ti2 = n1 + n2 Xi − 0 T −1 Xi − 0 . In order to compute the Ti2 statistic it is necessary to employ a smoothing coefﬁcient. With the statistical information obtained in the ﬁrst sample it is determined whether the process is in control. . w1 . 1 2 Hence. and n2 . There are ﬁve parameters to deﬁne the DST 2 chart: h. the process is deemed to be out of control. for comparison purposes. employing genetic algorithms. it is accepted that the process is in an incontrol state. employing Ti21 = n1 Xi 1 − o T −1 Xi 1 − o . MEWMA Control Chart The multivariate EWMA control chart. If Ti21 > h1 . 2. 2. However. Champ and Aparisi (2006) developed an analytical method to obtain the ARL of the DST 2 chart and. if Ti21 < w1 .176 Aparisi and de Luna E n = n0 . The MEWMA vector is obtained through Zi = rXi + 1 − r Zi−1 and the statistic to be plotted is given by Ti2 = ZT −1 Zi . selected to obtain a desired incontrol ARL. The Ti21 statistic is computed only considering the ﬁrst sample of size n1 .2007] At: 22:30 6 June 2010 3. or whether it is required to analyze the second sample. and combine it with the ﬁrst one in order to make a ﬁnal analysis. If the value of r is reduced the weight of the past samples is more important. Aparisi and GarcíaDíaz (2004) developed genetic algorithm based software which ﬁnds the parameters r and h (for univariate and multivariate cases) that produces the minimum ARL for a given size of shift and for a given incontrol ARL. In some cases. out of control. It is a common practice to use the asymptotic approximation of the matrix zi When r = 1. than the X control chart. A software tool has been employed to ﬁnd the optima parameters of this chart. being zi the covariance zi i matrix of Zi . The syntheticX chart for the zerostate scenario shows better ARL values to detect process shifts. The MEWMA control chart shows an outofcontrol signal when the Ti2 statistic is larger than the control limit (h). h1 .
The synthetic T 2 control chart (syntheticT 2 ) is a chart for monitoring simultaneously two or more quality characteristics. However. These methods can also be applied to the syntheticT 2 chart (Aparisi et al. the syntheticX chart exhibits a better performance than the EWMA control chart (Wu and Spedding. With the objective of improving the performance of the synthetic control charts. due to the help provided in identifying the variables that have shifted. Applying these techniques to the syntheticT 2 chart the user may obtain a controlling procedure that. The value of LCsynt is the criteria to classify a sample as conforming or non conforming. Ti2 > LCsynt does not mean that an outofcontrol state has to be assumed.. The value of L is the criteria to decide if the process is in control or out of control. When this assumption is ruled out.b. 2005a. 2006. is very important for the performance of the synthetic charts. unlike the T 2 control chart. Only when the sample is classiﬁed as non conforming and CRL ≤ L will the process be considered to be out of control. 1998. Costa and Rahim. although it is not the fastest in comparison versus other multivariate charts (see the comparison later).. . Several methods for the interpretation of the outofcontrol signal of the T 2 control chart have been developed. 2006) and the percentage of defective units of a process (Wu et al. 2004. L. LCsynt. Subcharts of the syntheticT 2 control chart: T 2 subchart and CRL subchart. 2002). L ≥ 1. 2001). Mason and Young. Mason et al. 1997). It consists of two subcharts: a T 2 subchart and a CRL subchart. The T 2 subchart has a unique control limit. this signal does not inform us about the variable or variables that have produced the shift.2007] At: 22:30 6 June 2010 especially for moderate and large shifts. Figure 1 shows the concept of the syntheticT 2 chart. Like the T 2 control chart. but the inspected sample must be classiﬁed as non conforming. the syntheticT 2 chart shows an outofcontrol signal. called head start.The SyntheticT 2 Control Chart 177 Figure 1. This characteristic. For a syntheticT 2 chart. Chen and Huang (2005b) applied variable sampling intervals (VSI) and Davis and Woodall (2002) improved the performance of the syntheticX control chart by adding the concept of sidesensitive. Downloaded By: [Bon Consortium . indicating that there probably is a shift in the process. 1999. Kourti and MacGregor. The CRL subchart has a low control limit. the performance of synthetic charts worsens (Davis and Woodall.. Atienzas et al. The CRL concept assumes that in t = 0 there is a point above the LCsynt limit (a non conforming sample in t = 0). 2006. may be more useful in a real application. 2000).. The synthetic chart has been also applied to monitor the variability of a process (Chen and Huang.
The steadystate case means that the process was in an incontrol state for a long period. The value of the incontrol ARL is selected taking into account the frequency of false alarms.2007] At: 22:30 6 June 2010 5. 1992. As exposed in Fig. Obtaining the ARL Values When the ARL needs to be computed. The zerostate scenario implies that the shift in the process occurred just before the ﬁrst sample was taken. before the outofcontrol situation happened. 4. p = 3. with ARLS−T 2 d = 0 75 = 13 58. Comparison Versus Other Multivariate Charts 5. In a similar way.41. Optimization of the Parameters of the SyntheticT 2 Chart Two values of ARL are important for the design and performance of the syntheticT 2 chart. and number of variables. 2. Optimization of the SyntheticT 2 Control Chart A software tool for Windows® has been developed in order to ﬁnd the optima parameters of the syntheticT 2 control chart. The chart that fulﬁls this design is considered to be optimum to detect a shift of size d in the mean vector. the incontrol ARL [ARLS−T 2 d = 0 ] and the outofcontrol ARL [ARLS−T 2 d = 0 ]. the reduction in the value of ARL for a shift d = 0 75 employing the syntheticT 2 chart is 33. Once the zero or steady state is selected.5%. The outofcontrol ARL value for the standard T 2 chart is 20. d = 0 75. 2005). Davis and Woodall (2002) found that the syntheticX chart may be modelled employing Markov chains.1.178 Aparisi and de Luna 4. and steady state. the output from the software tool shows the improvement of ARL against Hotelling’s T 2 control chart and it draws the ARL curves of T 2 and syntheticT 2 charts. . The software tool is available upon request from the authors. Figure 2 shows the solution for the problem ARLS−T 2 d = 0 = 200. Downloaded By: [Bon Consortium . The value of ARLS−T 2 d = 0 is important in order to rapidly detect a shift in the mean vector of magnitude d. ARLS−T 2 d = 0 . like a standard X chart with a run rule of 2 points of (L + 1) outside control limits. the optimization process ﬁnds the values of L and LCsynt that minimizes the value of ARLS−T 2 d = 0 for a given shift of magnitude d sample size.1. taking into account the restriction of the speciﬁed incontrol ARL.2. The output of this software tool is shown in Fig. Comparison Between Optimum Steady and Zero States SyntheticT 2 Charts A comparison of performance between the optimum design for the zerostate syntheticT 2 and the steadystate synthetic T 2 control charts has been carried out. 2. 4. two approaches are possible: zero state and steady state (Champ. This approach is also useful for calculating the ARL for the zerostate case. Zhang and Wu. The parameters of the optimum syntheticT 2 chart are: LCsynt = 9 163 and L = 9. Therefore. n = 4. see the Appendix. The syntheticT 2 chart may be modeled like a T 2 control chart with a run rule that consists of 2 points of (L + 1) outside the upper control limit. Wu and Spedding (2000) showed the formulas to obtain the ARL value of the syntheticX chart for the zerostate case. it is possible to obtain the ARL of the multivariate syntheticT 2 control chart (ARLS−T 2 ) employing a Markov chain approach.
The ARL values are shown in the second and third columns of Tables 1–6. The zerostate case of the syntheticT 2 chart has been selected for this comparison. and d. p increase. . the optimum value of LCsynt for zero state will always be larger than the optimum value for steady state.2. For all the cases. This is a predictable result. number of variables and sample size). and for the next one.2007] At: 22:30 6 June 2010 Figure 2. the optimum parameter L for the steadystate case is always lower than the optimum for the zerostate case. The optimum design of the syntheticT 2 charts has been utilized in the comparison for different cases of ARLS−T 2 d = 0 = ARLT 2 d = 0 . the optimum value of L of the syntheticT 2 chart has the same value for the steady state and zerostate scenarios. Comparison Versus Hotelling’s T 2 Chart A comparison of performance between the zerostate syntheticT 2 and the standard T 2 control charts [ARLT 2 d ] has been carried out. these differences tend to be larger as the sample size increases. and tend to decrease as the number of variables. although these shifts are detected quickly for both schemes. For the same design parameters (same incontrol ARL. However. Main window of the inhouse software tool used to ﬁnd the optimum syntheticT 2 and syntheticX control charts. because for the rest of charts the ARL has been computed employing the zerostate approach. The largest differences (50%) are found for large shifts. For small shifts (d ≈ 0 5). if the magnitude of shift d is moderate or large. 5. a consequence of the inertia that the steady state produces in the detection of the shift. However. p. The performance of the optimum zerostate syntheticT 2 chart is always better than the optimum chart for the steadystate case. for small shifts. n. the differences in the ARL values are around 20%.The SyntheticT 2 Control Chart 179 Downloaded By: [Bon Consortium .
1.1.1.0.9) 1.19) 12.18) 9.4) 2.3) 1.95.57.3) 1.14.1. L ) Zero state MEWMA h r SyntheticT 2 (LCsynt.77.58.64 (9.75.3.4.630.987.19) 26.26 (5.1.59.500.27.6) 2.76 1.5.4.59) 2.29) 61.48 (6.71 (8.0.93 (6.5) 1.62 (9.1.1.26) 4.1.28.58.11) 3.27.28 (7.38 (7.2) 41.24 (9.39.75 (7.16) 6.28 (10. 5.6) 4.1.49 (10.5 76.96.2.86 0.03 (9.95 (10.39.10 ( 6.8) 2.3) 1.67) 1.597.7) 1.499.10.5) 2.1.01.93 (10.13.5 5.55 (10.2.966.86 (10.75 37.92 (10.11.4) 1.10 (10.01 180 1 18.5.3.5) 2.10.87 ( 5.4) 1.842.2007] At: 22:30 6 June 2010 Table 1 ARL of the optimum multivariate control charts for a shift d ARL d = 0 = 200.96 (11.03 ( 9.89.597.1.1.82.067.1.17) 8.12.5) 1.68 (10.50 129. n = 2 2 d 200 200 200 T CL SyntheticT 2 (LCsynt.45.37 (8.12.18.11.48 1.79 (9.873.13 (5.50) 51.355.51 2.5.74.597.91 1.11.2) 21.61 (10. 3.79) 138.34 (10.Downloaded By: [Bon Consortium . p = 2.372.10) 4.27 (10.942.73 (10.69.25 9.36.11. 0.11 (11.0.19.0.155.19) 39.43) 2.87 2.25.5. L ) Steady state VSST 2 (CL w n1 n2 ) n1 + n2 <= 20 DST 2 h h1 w1 n1 n2 n1 + n2 <= 20 200 0 200 0.09 (7.79 (6.20.20 (14.2) 137.00.59 (7.27 (8.0.8) 6.1.13) 9.2.47 (10.74.1.5 1.25 200 10.2.597.3.4.39.2) 1.25 1.83.3) 5.0.698.1.597.62.22 (10.1.80 (7.1.1.5966 148.4) 1.859.63 (6.50 (13.22 (6.30) 19.36 (6.496.38) 3.88 (7.74) 1.2) 1.34.11.451. 0.1.597.16) 6.49 (6.597.7) 2.87.597.22.5) .4) 1.28 (10.1.59.597.3) 16.161.873.12.0.86 (6.95.26 (13.12) 12.16.73.5485.0.68 (5.597.826.108.39.51.41.71.78.3.1.34 0.65 2 2.366.77.05) 17.25.242.10) 3.1.75 3.966.20.
68 (6.597.94.72 1.98 0.00 58.1.89.73 (7.2) 68.4.23.597.59.00 (10.20.25 (10.91) 1.00 (10. L ) Steady state VSST 2 (CL w n1 n2 ) n1 + n2 <= 20 DST 2 h h1 w1 n1 n2 n1 + n2 <= 20 200 0 200 0.2) 1.32 (8.47) 1.01 (5.1.355.13) 1.63 (10.19.0.90)) 1.04.842.00 (15.2.10.2.0.666.73.4.11.7) 1.33 (6.59.81 (9.59.5) 1.02 (17.02.25 1.20 (5.00) 66.89.0.00 (12.195.6.6.00 2.7) 1.0.11) 1.10 (10.9) 4.88. 0.966.2) 1.0.5) 2.8) 1.58.769.03 (10.6.2007] At: 22:30 6 June 2010 Table 2 ARL of the optimum multivariate control charts for a shift d ARL d = 0 = 200.54.5966 84.19) 9.65 (9.20.597.63.75.873. L ) Zero state SyntheticT 2 (LCsynt.5.11.3) 1.00 (10.2) 1.15.50 ( 5.698.18) 2.20) 14.1) 18.2) 1.19.6.3.0.05 ( 5.13) 10.4.2) 1.04 (10.597.62 (10.01.88.07 2 1.6.69 (7.54 (10.5.33.00 (5.86) 1.14) 1.41.0. 2.14.74) 1.1.12) 3.2) 1.3.02 1.22.3) 1.1.2.27.2) 1.75 7.5.25 200 10.25 1.3.15.51 (5.6.33.01 (10.3.2.85 (10.14) 1.471.597.24 1.94.594.30 (10.5.873.5.07.232.966.6.17 (10.5 1.32) 10.3.57.8) 7. n = 7 d 200 200 200 T2 CL MEWMA h r SyntheticT 2 (LCsynt.06 (10.873.966.63.873.25.5 1.08 (10.9) 1.79 (5.34.11) 1.52 (9.2) 1.02 (10.10) 1.12.12) 1.209.59.56.74.87) 1.597.89.04 181 1 3.25 (10.19) 3.242.966.36.56.34) 2.00 (5.27 (8.0.90.873.75 1.597.10.0. p = 2.13) 2.2.5 21.2) 1.02 2.866.00.4.50 ( 6.597.597.70 (7.2.59.8) 1.2.53.556.11.10.1.92.57 (5.597.966.86 (6.08 (10.14) .3.6.Downloaded By: [Bon Consortium .22) 3.13) 1.50 (5.39.18 (10.0.2) 1.3.00 (13.36.35 0.36 (10.6.36 (12.15.50 (5.00 (5.966.6.09) 6.
23) 17.75 (12. L ) Steady state VSST 2 (CL w n1 n2 ) n1 + n2 <= 20 DST 2 h h1 w1 n1 n2 n1 + n2 <= 20 200 0 200 0.3) 161.494.22.Downloaded By: [Bon Consortium .441.68.46 (16.9.69.3) 52.25.2007] At: 22:30 6 June 2010 Table 3 ARL of the optimum multivariate control charts for a shift d ARL d = 0 = 200.26.5 10.5.20.42) 2.1.0.61 (12.07 (16.1.07 (10. 8.0.1.06) 20.9) 3.78 ( 16.15 (13.561.0.20 2 3.1.11 1.750.393.657.3) 7.26.84.5.750.0.62.60.38 (16.27.19) 9.16.04 155.87.1.9.51 (14.82 (13.1.94.3) .36 (11. 0.10) 3.97 (12.58 (15.661.1.22.17 (16.51) 2.73.1.35.20 0.6) 2.69.25 200 16.25 18. p = 5.28 1.890.24.11) 11.73.07 (11.53 (13.19.750.10) 4.63 (16. L ) Zero state MEWMA h r SyntheticT 2 (LCsynt.7) 2.43 182 1 33.1.928.9.7.1.59 (11.79.87 (16.1.10.734.08) 11.1.11 (16.153.80.5 109.99.17.13.0.19.58 (14.20) 46.22) 5.15) 22.) 31.1.96.20.4.17. n = 2 2 d 200 200 200 T CL SyntheticT 2 (LCsynt.0.0.15) 7.6) 2.66.1.91 (12.1.43.8) 6.750.81.96 (14.7.719.1.83 (13.1.13) 5.19) 15.5) 2.3) 1.366.15.7496 168.8) 5.25 2.386.0.29) 4.18.07 1.448.4) 1.17.04.5 2.538.36 (11.12) 7.47 (18.3) 1.73.1.1.10.10.28 (18.13 (12.79.7) 2.99 2.35) 3.4) 2.07 (13.10.750.977.19) 20.18 (16.8.75 6.393.76.0.44.159.53 (12.03 (13.58) 1.833.08 (16.06 (14.1.3) 1.07 (11.750.609.750.5.18.75 61.27 (16.22 (16.0.724.980.28.73 ( 12.82 (16.52 (16.1.33.62.85.8.35) 37.74 (16.3) 1.7.97 (11.4) 2.69 (15.82.750.22 (22.48) 82.15) 8.22 (14.84 (12.7) 1.61 (16.76 2.69.17.538.37 (16.96 0.19) 66.180.92.750.31. 7.51.62.25) 92.750.4.16 (13.1.32 (11.6) 2.79) 164.6) 4.35.5) 3.80.
7496 116.01 (16.17.2) 1.5.719.38 (12.19) 4.73) 1.1.0.618.00 (23.5.0.89 (13.08.73.2) 1.0.2007] At: 22:30 6 June 2010 Table 4 ARL of the optimum multivariate control charts for a shift d ARL d = 0 = 200.43.6.00 90.7) 1.5 1.98.11) 1.55 (10.0.39 (16.55 1.11) 1.02.5.57.23. p = 5.745.73.90.6) 14.14) 1.44 (16.71 183 1 4.8.4) 1.20.5 38.11.73.02.5.07.0.750. 0.590.07) 8.13.87 (13.3) 1.0.01 2.17.52.445.0.00 (11.86 (16.5.0.04 (11.14) 1.0.917.36 (15.75 1.19) 16.00 (17.77.037.5.01 (16.00 (11.38 (15.51 (16.20.97 1.05 (16.724.037.750.0.7.50 (10.97) 1.7) 1.890.6.51.00 (13.01 (18.25 2.7.71.23.75 12.34.7.2) 23.53.5.917.29.73.750.95.8) 1.91) 1.47 (11.72.05 2.6.0.25 1.81 (14.8.23.10 (16.6.349.Downloaded By: [Bon Consortium .037.037.917.95 (15.74 0.11 (16.441.19 2 1.2) 100.538.0.19 (16.7.00.00 (16.08.44.14) 1.24.0.037.02 (23.46.750.75.00 (16.50 (10.2.01 (11.13) 3.17) 5.917.7) 1.86 0.00.81 (16. L ) Zero state SyntheticT 2 (LCsynt.37) 20.750.750.51.3.6.0.8) 2.16.5.7) 3.8.20.1.750.6.5.39.12) 8.750.92 (16.750.5.00.2) 1.61.14 (11.8.69.1.917.69) 1.1.34) 2.28.4) 1.19) 2.5 1.2) 1.5.92) 101.46 1.62 (16.4. n = 7 d 200 200 200 T2 CL MEWMA h r SyntheticT 2 (LCsynt.06 (16.52) 2.5.2) 1.0.71 (10.2) 1.750.2) 1.95 (15.6.18.10.5.24.63 (11.4) 2.47 (15.2) 1.5.51 (10.49.18 (12.48 (11.75.7) 6.52 (16.20) 27.6) 1.66) 1.28.04 (16.75.63.00.25 200 16.924.16.15) 1.93.20) 4.393.3) 1.5) .88 (12.22 (16.20 (11. L ) Steady state VSST 2 (CL w n1 n2 ) n1 + n2 <= 20 DST 2 h h1 w1 n1 n2 n1 + n2 <= 20 200 0 200 0.17) 1.51 (16.
42.338.84 (19.6) 2.66.01 (25.25 200 25.11.754.78 1.27.32.16 (19.81.65 (25.29.5) 1.13.71 (20.0.48.07 (21.98.5 17.9) 2.875.45 (25.1.35.64.1.1.30.15.27) 28.50 (20.30 (25.27.15.19) 92.96.188.43 (22.73 (22.5 132.40.37) 57.07) 14.40 (22.1882 179.15 184 1 50.7) 6.3) 173.14.9) 4.01 (24.8725.18.62 (25.88.35 (25.42 2.15.01 (22.50.43 1.45) 108.7) 3.55.1.93 (23.76.99.Downloaded By: [Bon Consortium .17.595.21.78 (19.657.18.58 (25.77.17 (19.75 (22.56.12) 19.1.60 (25.890.8) 3.361.1.68 (24.58 (21.24 0.19) 13.0.54.163.1.17) 6.44.709.3) .95.120.72 (21.3) 64.0.8) 3.15) 36.5) 1.4) 42.5 2.54.324.188.05.188.1.98 (20.688.43 (23.150.16.15.22) 117.0.5) 3.432.84 (25. L ) Zero state MEWMA h r SyntheticT 2 (LCsynt.2007] At: 22:30 6 June 2010 Table 5 ARL of the optimum multivariate control charts for a shift d ARL 0 = 200.11.5) 2.19 (19.94 (21.32.39 (20.11) 4.8. n = 2 2 d 200 200 200 T CL SyntheticT 2 (LCsynt.439.47 (19.71 (20.034.557.022.052. p = 10.4) 1.6) 4.13) 7.13.02) 26.22.12.1.39) 3.0.1.88.14 1.91 (24.43.029.61.42 (20.72.0.2) 5.890.75 10.96 169.188.17.188.11 (20.86.0.87.19) 67.97 (25.98.1.19) 10.66 (18.1. L ) Steady state VSST 2 (CL w n1 n2 ) n1 + n2 <= 20 DST 2 h h1 w1 n1 n2 n1 + n2 <= 20 200 0 200 0.17.84 (25.09.19) 19.4) 2.13.188.12) 4.513.214.47 (18.63 (25.08.37.27 2 6.1.5) 2.1.1.27.0.55.62.15 (17.0.49) 2.5) 2.34) 3.10) 7.1.25 4.80.71.188.19 (25.71.29.1.16 (17.1.23 2.19) 31.188.17.64.682.05 (19.28 0.1) 9.76.19) 14.17.24 (21.75 85.53 (20.188.1.52) 2.10) 11.25 29.16) 7.77) 176.54 (25.188.19.12.76 (17. 0.
05) 10.74.6.7) 1.38) 34.266.04 (25.16.266.97) 1.23) 3.03 (18.3.55.78) 1.3.8) 5.5 58.70 (24.696.3) 1.26.266.14.14 (25.13 1.19) 31.34 (25.95 (19.2.6.85) 1.709.188.188.70 (25.2) 1.48.26) 2.0.55.5) 24.33 (18.29 0.17.12) 1.05 (25.15) 1.17.98 (18.60 (25.00.5) 2.17.51 (18.61.05 (23.4) 1.115.60.77 (25.187.86.79.09 185 1 8.26.2.56.324.19) 42.26.00 (26.188.27.24.0.6.0.188.6.11.21 (25.6.11.13.19.93.15) 1.37 (25.0.60.28.3) 1.0.432.91 (22.5.3.4) .266.068. L ) Zero state 200 SyntheticT 2 (LCsynt.12.18.188.18) 7.22 (37.07 1.7) 6.16) 2.54 (18.12 (21.14.01 (26.296.14.0. L ) Steady state VSST 2 (CL w n1 n2 ) n1 + n2 <= 20 DST 2 h h1 w1 n1 n2 n1 + n2 <= 20 0 200 0.51 (21.21) 10.50.67.96.52 (23. n = 7 d 200 200 200 T2 CL MEWMA h r SyntheticT 2 (LCsynt. 0. p = 10.5.115.6.163.39 (25.00 (18. 0.34.19.63 (21.2) 1.1.14.25.13.2) 1.36 (20.91 (25.77) 1.59 (26.188.04.41 2 1.14.49.11.54 (23.94 (25.32.75 1.14 2.Downloaded By: [Bon Consortium .80.188.925.12) 1.01 (25.0.2) 29.693.11.6.82 (25.10) 1.14 (25.11) 2.97 (25.50 (18.12.00.88.27.12.16.9) 7.33.01 (25.5 2.11.14) 1.05.5 1.188.5.799.0.13) 5.12) 1.78.6) 1.6.26.294.115.4) 1.1.49 (25.0.77 (19.75 21.32.04 (25.89 (19.11) 3.25 200 25.03 (20.188.65 (18.2007] At: 22:30 6 June 2010 Table 6 ARL of the optimum multivariate control charts for a shift d ARL d = 0 = 200.10 (18.14.2.120.00.01 115.58.0.4.2) 124.7) 1.15) 1.890.15.29.58) 1.2) 1.24.0 (18.25 3.48 (19.0.91.2) 1.31.1882 138.25 1.80 (20.28.18) 3.11) 1.33.115.04 2.81 (20.474.2) 1.27.83 0.94) 127.18.13) 13.58 (24.188.91.4.0.0.70 1.
For example. only six tables are shown in this article. A very large number of cases have been studied. with ARL values of ARLT 2 d = 1 25 = 9 91 and ARLS−T 2 d = 1 25 = 4 59. The rest of tables are available from the authors. In some cases. the range of values of d where the syntheticT 2 chart has better performance is narrower. The ARL values are shown in the two ﬁrst columns of Tables 1–6. If the syntheticT 2 chart is designed to minimize the ARL for a given shift d. 2. given the same set of values of p. the optimum syntheticT 2 for a shift d is always faster than the standard T 2 for detecting shifts of magnitude 0 < d ≤ 3. 1. p = 2. For the same values of p and ARL d = 0 . when d ≥ 1 75 the percentage improvement for the syntheticT 2 chart is smaller when the sample size increases. For n = 1.186 Aparisi and de Luna The optimum design of the syntheticT 2 chart has been utilized in the comparison for different cases of ARLS−T 2 d = 0 = ARLT 2 d = 0 . more cases (tables) may be needed. For large shifts. d. n = 2. increases as the value of ARL d = 0 is larger. For large sample sizes the performance of both charts is similar for large shifts. 5. independently of the sample size (if n ≥ 2) and magnitude of shift. For example. or 3. As the sample size increases. due to size limitations. The syntheticT 2 control chart optimum for a shift d is always faster to detect than the Hotelling’s T 2 chart. but the improvement is marginal for large shifts. the improvement is quite signiﬁcant. To observe some of the conclusions. 0 < d ≤ 3. the optimum parameters of the syntheticT 2 charts for d are constants. The parameter L takes small values if the chart is designed for large values of d but L tends to be small if the chart is optimized for large values of d. The curves shown are for the cases ARL d = 0 = 200. and d = 0 25 0 5 2. For sample sizes n ≥ 4. if the value of d is large. n ≤ 7. p. n and d. The percentage improvement for the syntheticT 2 .66%. For the same case the percentage of improvement is larger than 40% for the range 0 75 ≤ d ≤ 2. The percentage of improvement for the ARL is calculated as: % of improvement = ARLT 2 d = 0 − ARLS−T 2 d = 0 ∗ 100 ARLT 2 d = 0 (2) Downloaded By: [Bon Consortium . d > 2.2007] At: 22:30 6 June 2010 A list of the key conclusions will be discussed below. . Inﬂuence of the sample size. p and d. p = 2. if n = 7 the improvement is negligible if d ≥ 2 25. For small and moderate shifts (d ≤ 0 75). This behavior is logical because the standard T 2 shows ARL values close to one. and for the same set of values of ARL d = 0 . and d = 1 25 the improvement in the ARL is 53. the percentage of improvement of the syntheticT 2 chart is better when the sample size is small. However. however. n. the performance of the optimum syntheticT 2 chart is better for small and moderate shifts. the optimized chart is also faster than the standard T 2 chart for the rest of values of d. 2. and d. Figure 3 shows the ARL values of the optimum syntheticT 2 charts for a shift d. when ARLS−T 2 d = 0 = ARLT 2 d = 0 = 200. a larger sample size increases the percentage of improvement. 4. 3.
In this section. VSST 2 . p. However. 4. considering the same case. For the same values of n and d the percentage improvement is smaller when more variables are to be controlled if d ≤ 0 75. when p = 2 and n = 7 the syntheticT 2 chart is superior or equal than the MEWMA chart if d ≥ 0 75. ﬁrstly it is guaranteed that. and 7. Inﬂuence of the number of variables. VSST 2 . p = 2. producing solutions that may not be feasible in real applications. Tables 1–6 show the optimum parameters of the control charts that are used in the comparison and the value of ARL d = 0 for d = 0 25 0 5 2 5 and for some combinations of p and n when ARL d = 0 = 200. the syntheticT 2 chart is superior or . the same sample size is employed. the percentage of improvement is larger when the number of variables increases. The percentage improvement for the syntheticT 2 chart depends of the number of variables. 5. Secondly. it is very frequent that in order to ﬁnd the optimum charts to minimize ARL d = 0 for the VSST 2 and DST 2 charts. when d ≥ 1 5 and n ≥ 3. ARL of the optimum syntheticT 2 chart for a shift d. For example. The optimum syntheticT 2 chart for a shift d is faster than the MEWMA and VSST 2 charts when p = 2 and n is large. 5. 3. Comparison versus MEWMA. The authors have obtained more tables for the following cases: p = 2. on the other hand.2007] At: 22:30 6 June 2010 Figure 3.The SyntheticT 2 Control Chart 187 Downloaded By: [Bon Consortium . the restrictions E n = n0 and n1 + n2 ≤ 20 must be fulﬁlled. 6. If.1. ARLS−T d = 0 = 200. Given these restrictions. large values of n2 are obtained. and 10. a comparison of the performance (ARL) of the optimum zerostate syntheticT 2 chart for a shift d. versus the optimum MEWMA. and DST 2 charts for the same d is carried out.2. ARL d = 0 = 200 and 400 and d = 0 25 0 5 2 5. on average. n = 1 2 3 5. and DST 2 Charts. For the VSST 2 and DST 2 charts.
It is possible to see that the ARL curve of the syntheticT 2 chart is always below the curve of the Hotelling’s T 2 chart. although the range of d where the performance of the syntheticT 2 is superior is incremented when the sample size increases and the number of variables is smaller. independently of the sample size and the number of variables. for moderate and large shifts. However. for the rest of size of shifts. However. However. the syntheticT 2 chart is faster than the MEWMA and VSST 2 charts for shifts of magnitude d ≥ 1 75. The performance of these two charts.2007] At: 22:30 6 June 2010 6. the syntheticT 2 control chart has been developed. the range where it is better to use the syntheticT 2 chart instead of the MEWMA control chart increases as the sample size increases. is again very similar as the sample size increases and the number of variables decreases. The ARLs have been calculated employing the Markov chain approach for the steady and zero states. There are some cases where the MEWMA and VSST 2 optimized charts for a shift d have an ARL value larger or equal than the standard T 2 chart. The ARL of the syntheticT 2 is lower than the ARL of the MEWMA and VSST 2 charts when d ≥ 0 8. the lowest ARL value is obtained by the DST 2 chart. because the former is always faster than the latter in detecting the shifts in the mean vector for all the values of magnitude of shift. ﬁnding a procedure to obtain the optimum parameters of this chart to minimize the outofcontrol ARL. The DST 2 control chart is the chart that exhibits better performance in detecting simultaneously very small and large shifts. for this same value. In general. For moderate and large shifts. 0 < d ≤ 3. the range where the optimum syntheticT 2 chart is superior to these charts worsens as the sample size decreases and the number of variables increases. For very small shifts the DST 2 chart has better performance than the MEWMA. However. this does not happen with the curves of the MEWMA and VSST 2 charts. the syntheticT 2 and DST 2 optimized charts for a shift d always have lower or equal ARL values in comparison with the standard T 2 chart. Downloaded By: [Bon Consortium . the performance of the MEWMA and VSST 2 charts is better for detecting small shifts than the syntheticT 2 chart. Figure 4 shows the ARL curves of the optimum charts designed to have the best performance for d = 0 75. It may be advisable to employ the syntheticT 2 control chart as an alternative to the standard T 2 chart. it will be always faster to detect these shifts than the T 2 chart. The DST 2 chart has better performance than the syntheticT 2 for d ≤ 1. the performance of the syntheticT 2 chart is similar to the DST 2 control chart. and n = 7 (Table 2). The syntheticT 2 chart is faster than the VSST 2 chart when d = 0 25. One important property of the zerostate syntheticT 2 chart is that an optimized chart for a given . given a value of incontrol ARL value and a magnitude of shift in the mean vector. For example. p = 2. If the syntheticT 2 chart is designed to minimize the ARL for a shift d. but this is may not be true for moderate and large shifts. for the case ARL d = 0 = 200. For a given value of ARL d = 0 and number of quality variables. This behavior does not happen sometimes with the MEWMA and VSST 2 charts. if p = 10 and n = 1. Conclusions In this article. When n ≥ 3 and p ≤ 10. the ARLS−T 2 d = 0 is lower than the ARL of MEWMA if d ≥ 2 75.188 Aparisi and de Luna similar than the VSST 2 chart when d ≥ 1.
In general.. the improved performance. but it has a much better performance in comparison with the Hotelling’s T 2 chart. for very small shifts the syntheticT 2 chart cannot compete with the MEWMA.The SyntheticT 2 Control Chart 189 Downloaded By: [Bon Consortium . these are cases where the syntheticT 2 chart is superior to that of the MEWMA or VSST 2 charts. a user interested in ﬁnding out the variables that have produced the outofcontrol signal can apply the MTY decomposition method (Mason and Young. For example. n = 7. and the possibility of continuing employing the tools for the interpretation of the outofcontrol signal of the Hotelling’s T 2 chart. 1999. ARL curves of the optimum control charts for a shift d = 0 75. Although its real application can be more complicated than the standard T 2 chart. and DSST 2 charts. As Davis and Woodall (2002) suggested. VSST 2 . magnitude of shift always gives a better performance. Appendix: Markov Chain Model Let A = P Ti2 < LCsynt and B = 1 − A. we recommend the DST 2 . 2006).. ARL d = 0 = 200. 1997) or the use of neural networks (Aparisi et al. the following transition matrix would govern the Markov chain for the multivariate syntheticT 2 control chart: . than the equivalent Hotelling’s control chart. having the advantage of being easier to employ in real applications. Given moderate and large shifts. The syntheticT 2 chart may provide a good alternative to the DST 2 chart in order to detect moderate and large shifts in the mean vector. p = 2. make the utilization of the syntheticT 2 control chart an interesting one.2007] At: 22:30 6 June 2010 Figure 4. for all the values of shift. For the detection of very small shifts (d = 0 25). Mason et al.
for example. 1 for the initial state and 0 for the rest of cases. and 1 are the aforementioned and i is the steadystate probability for the state i. where p is the number of variables. Therefore. With this Markov chain model the ARL for the zerostate case is ARLzero = s I − R −1 1. The last row contains A in the ﬁrst column. The ARL of the multivariate syntheticT 2 for the steadystate case is T obtained considering SS = 0 1 and ARLsteady = SS × I − R −1 1. 3. the transition matrix for the syntheticT 2 chart when L = 5 is: 00000 00000 00001 00010 00100 01000 10000 Signal A 0 0 0 0 A 0 00001 B 0 0 0 0 0 0 00010 0 A 0 0 0 0 0 00100 0 0 A 0 0 0 0 01000 0 0 0 A 0 0 0 10000 0 0 0 0 A 0 0 Signal 0 B B B B B 1 LCsynt 0 1√ · 2p 2 Downloaded By: [Bon Consortium . i=0 In order to calculate the SS vector. LCsynt is the control limit of the T subchart. In all the other locations.2007] At: 22:30 6 June 2010 When the process is in control. s = 0 1 0 0 0 . In order to obtain the outofcontrol ARL. research project number DPI200606124 including European FEDER funding. and the R submatrix (L + 1)*(L + 1) has the transition probabilities without considering the absorbent state. where L the matrices I. R. It is also possible to use an approximation obtaining the probabilities following the procedure employed by Celano et al. B = 1 − P X ≤ LCsynt = 1 − p x x 2 −1 e− 2 dx. and the support of the ITESMFoundation Carolina agreement. Hence. (2006). . where n is the sample size and d is the Mahalanobis’ distance. where 1 is a column vector of “ones” of size (L + 1). p 2 Acknowledgments This work has been supported by the Ministry of Education and Science of Spain. 2. and a is the gamma of Euler. In all the other rows. B = 1 − P Ti2 ≤ LCsynt = 2 P p = nd2 ≤ LCsynt and A = 1 − B. the entry above the diagonal is A. B is computed employing the cumulative distribution function of the non central chisquare with p degrees of freedom and non centrality parameter = nd2 . s is the vector (L + 1) of initial probabilities. the entry is zero. it is possible to prove that the steadystate probabilities are a function of the parameters L and LCsynt of the syntheticT 2 control chart.190 1. it is recommended employing a cyclic probability vector proposed by Lucas and Saccucci (1990) and simpliﬁed by Champ (1992). 4. L i = 1. Following this last approach. Aparisi and de Luna The ﬁrst row contains A in the ﬁrst column and B in the second column. I is the identity matrix of size (L + 1)(L + 1).
J. Qual.. (2005). Qual. Haro. J. Multivariate multiple sampling charts.. Adv. Int. Qual. D. C. Comput. 37(6): 509–521. W. Maint. J. The run length distribution of the CUSUM with estimated parameters. Huang. 12(1):81–88. C. (2005). Variable sampling interval synthetic control charts for jointly monitoring process mean and standard deviation. J. S. Fichera. J. Manufact. A. Operat.. H.. Operat. Quah. Technol. (2003). F. Statist. F. W. M. 49(2):221–240. Champ. Grigoryan A. S. J. Product. J. L. Technol. Reliab. J. Int. Statist. F. Double sampling Hotelling’s T 2 charts. (2006).. 34(2):200–208. H. Eng. IIE Trans. F. Aparisi. F.The SyntheticT 2 Control Chart 191 References Aparisi. 46(1):47–64. Industr. J. (1992). Hotelling. O. IIE Trans. He. J. (1992). pp. C. T. 37:2136–2148. W. (1994). Huang. A. 24(2):78–87. (2006). C. IIE Trans. (2006). W. Champ. Int. Klein. (2008). 24:153–166. 13(2): 136–146. Eng. (2004). New York: McGrawHill. F. W. A. 15(2):299–310. H. 28:966–977. Grigoryan. L. W. Hotelling’s T 2 control chart with adaptive sample sizes. Wallis. H. Product.. Qual. Avendaño. L. Commun. Technol. F... Ghute. C. B. Chen. 26(7–8):842–851. Fast initial response features for EWMA control charts. J.. (1947). F. F. A. Costa. Int. R. 111–184. L. 41(10):2169–2182. J. Sanz. Tang. M. In: Eisenhart. Technol. Ang. Woodall. Aparisi. Steadystate run length analysis of a Shewhart quality control chart with supplementary runs rules. He. Eng. G.. B. Khoo. Theor. L. Qual. 28(3):155–163. A. Daudin. Eng. Champ. 28(7):875–885. A Multivariate synthetic control chart for monitoring processes mean vector. Costa. Qual. (2006). Statist. H. B.. S. (1996). Modiﬁed Shewhartexponentially weighted moving average control charts. F. M. Davis. Industr. Hastay. Technometrics 47(4):437–445. A synthetic control chart for monitoring process dispersion with sample range. C. J. B. Double sampling X charts. JonesFarmer. 36(1):95–108. Optimization of univariate and multivariate exponentially weighted movingaverage control charts using genetic algorithms.. Champ. Int. GarcíaDíaz. C. 38(8):647–657.. X charts with variable sample size. S. Adv. 3(2):194–204.. (2002). Sci.. O. B. F. (2005b). Downloaded By: [Bon Consortium . W. Comput.2007] At: 22:30 6 June 2010 . Statist.. Chen. C. Int. Knoth. (2002). D. (1998). Res. J. J. L. Techniques of Statistical Analysis. G. Rigdon. Rigdon. Properties of the T 2 control chart when parameters are estimated.. Eur. A synthetic control chart for monitoring the process mean and variance.. Rahim. Costa. Aparisi. Interpreting T 2 control signal: effectiveness of MTY decomposition vs. Int. (2005). Jones. Simultaneous monitoring of univariate and multivariate SPC information using boxplots. Qual. Celano. 29(12):1051–1056. Multivariate quality control. Computing the percentage points of the runlength distributions of multivariate CUSUM control charts. M. (2006). A. Atienza. Qual. Res. (2005a). Evaluating and improving the synthetic control chart.. Statistical design of variable sample size and sampling interval X control charts with run rules. A. Chen. A.. D. (2006).. Economic design of X charts with variable parameters: the Markov chain approach. eds. V. A.. Aparisi. A synthetic control chart for monitoring process dispersion with sample standard deviation. S. B. Eng. J. a neural network. 34(10):2853–2862. Rahim. Commun. Res. (2004). W. 31(9):1437–1454.. E. A comparison of T 2 control charts with variable sampling schemes as opposed to MEWMA chart. J.. Joint statistical design of double sampling X and S charts.. M. J. Appl. J. (1997). (2001). Technol. Meth. 21(3):765–777. 168(1):122–142. Shirke. Res. Technol. A. Costa. L. Huang.. Pap. Manufact.
29(1):8–15. (1985). J. Technology 14(2):51–59.. Young. Zhang. H. Eng. Birch. (2005). 49(1):76–97. Qual. Mason. M. J. Rigdon. 33(1):104–111. MacGregor. Woodall. J. 32(1):32–38. (2001). S. Qual. S. T.. W. Designs of control charts with supplementary runs rules. Young. W. 31(2):155–165. J. M. New York: John Wiley & Sons. Technol. Luceno. (1997). Saccucci. W. 5th ed. C. Jr. D. Technol. Spedding... Congr. 27(6):800–810. Downloaded By: [Bon Consortium . Mason. J. (1990). (2004). J. (1992). A. Runger. S.. A synthetic control chart for detecting small shifts in the process mean. 33(6):511–530. E. M. Multivariate statistical process control and industrial applications. A. C. J.. C. Arnold. Prabhu. Montgomery.. R. Qual.. J. Lucas. C. Proc. G. (1997). 58:449–452..... Z. D.. Robust estimation in multivariate control charts for individual observations. A.. Z. Distribution of Hotelling’s T 2 statistic based on the successive differences estimator. Spedding. Technol. Comput. IIE Trans. A synthetic control chart for detecting fraction nonconforming increases. Champ. C. Z. S. Qual. A. Qual. Montgomery. Multivariate CUSUM qualitycontrol procedures. J. (1999). H. Technometrics 34:46–53. Qual. (2006). Lucas. Wu. Technol. (2000). C. J. (2003). Introduction to Statistical Quality Control. W. A. Vargas. Lowry. J. (2001). Improve. 35(4):367–376. Combined ShewhartCUSUM quality control schemes. A review of multivariate control chart.. S. IIE Trans. D.. Technol. Ncube. 29(4):396–406. S. Williams. J. (1982). B. Reynolds. M. EWMA control charts with variable sample sizes and variable sampling intervals. Qual. L. Tracy. Industr. ASQ World Conf. Qual. Qual. A multivariate exponentially weightes moving average control chart. (2000). S.. N. PuigPey. D. J. A practical approach for interpreting multivariate T 2 control chart signals. J. R. Technometrics 42(4):411–416. F. J. (2005). J.. H. Wu. M.2007] At: 22:30 6 June 2010 . Wu. Exponentially weighted moving average control schemes: properties and enhancements. C. A. Improving the sensitivity of the T 2 statistic in multivariate process control. Woodall. 38(3):217–229. Qual. Designing a multivariate EWMA control chart. T. Yeo.. Evaluation of the runlength probability distribution for CUSUM charts: assessing chart performance. C. Sullivan. H. J.. Technometrics 27(3):285–292. Technol.. Technometrics 32(1):1–12. J. C. R. L. J. Lowry. (1995). Technol. Woodall. H. T.192 Aparisi and de Luna Kourti.