Stochastics and Statistics

An extension of Banerjee and Rahims model for economic

design of moving average control chart for a
continuous ow process
Yun-Shiow Chen
*
, Yit-Ming Yang
Department of Industrial Engineering and Management, Yuan-Ze University, 135 Yuan-Tung Road, Nei-Li 32026, Taiwan, ROC
Received 14 June 2000; accepted 22 August 2001
Abstract
In this paper, we propose a model of a moving average control chart (MA control chart) with a Weibull failure
mechanism from an economic viewpoint. When the process-failure mechanism follows a Weibull model or other models
having increasing hazard rates, it is desirable to have the decreasing sampling interval with the age of the system. The
MA control chart is used to monitor quality characteristics of raw material or products in a continuous process. A cost
model utilizing a variable scheme instead of xed sampling lengths in a continuous ow process is studied in this re-
search. The variable sampling scheme is used to maintain a constant integrated hazard rate over each sampling interval.
Optimal values for the design parameter, the moving subgroup size, the sampling interval, and the control limit co-
ecient are determined by minimizing the loss-cost model. The performance of the loss cost with various Weibull
parameters is studied. A sensitivity analysis shows that the design parameters and loss cost depend on the model pa-
rameters and shift amounts.
2002 Elsevier Science B.V. All rights reserved.
Keywords: Economic design; Moving average control charts; Weibull shock model; Continuous ow process
1. Introduction
Duncan (1956) was the rst researcher to design
an xx-control chart from an economic viewpoint,
and the methodology he used was optimal for
solving the design parameters. The three design
parameters, the subgroup size (n), the sampling
interval h, and the control limit width (L), con-
sidered in the model were selected to optimally
minimize the average cost of a process. Duncan
(1956) assumed that the process had only a single
assignable cause and that the occurrence time of
an assignable cause was exponentially distributed.
This assumption has been widely used in subse-
quent work on the subject. Chiu and Wetherill
(1974) developed a simple approximate procedure
to optimize Duncans model, Taylor (1968) pre-
sented an economic design model for Cusum
charts, Saniga (1977) considered the joint eco-
nomic design of

XX and R charts, Duncan (1971)
developed an economic design for the cases in
European Journal of Operational Research 143 (2002) 600610
www.elsevier.com/locate/dsw
*
Corresponding author. Tel.: +886-3-463-8800; fax: +886-3-
463-8907.
E-mail address: [email protected] (Y.-S. Chen).
0377-2217/02/$ - see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0377- 2217( 01) 00341- 1
which multiple assignable causes, and Ho and
Case (1994) proposed the economically based
EWMA control charts. Banerjee and Rahim
(1987, 1988) studied a non-Markovian model, and
they applied a renewal theory to establish an
economic model when the process failure mecha-
nism follows a Weibull distribution. Banerjee and
Rahim (1988) incorporated Lorenzen and Vances
cost structure into a random sampling interval (cf.
Lorenzen and Vance, 1986). Rahim and Banerjee
(1993) extended Banerjee and Rahims model to
allow the possibility of age-dependent replacement
before failure when such a replacement yields
economic benets (cf. Banerjee and Rahim (1988)).
Parkhideh and Case (1989) developed a six deci-
sion variables in their economic model for

XX-
control chart. Ohta and Rahim (1997) proposed an
alternative and simplied design methodology that
reduces the number of six design variables to three.
Rahim and Costa (2000) presented the joint eco-
nomic design of

XXR control charts when the oc-
currence time of assignable causes follow Weibull
distribution. Koo and Case (1990) rst proposed
an economic design of

XX-control charts for use in
monitoring a continuous ow process for the ex-
ponential process failure mechanism. Chen and
Yang (1999) extended Koo and Cases model to
cases following a Weibull shock model (cf. Koo
and Case, 1990). All of the above work is either
focused on a piece part process for dierent kinds
of control charts or on a continuous ow process
for

XX charts; little attention has been focused on
the moving average control chart (MA control
chart). The MA control chart is widely used in
continuous ow processes; for example, a chemical
plant may collect data, periodically, on the results
of analysis made to determine the percentages of
certain chemical constituents.
In this paper, we adopt Banerjee and Rahims
cost structure to develop an economic design of an
MA control chart under a Weibull shock model in
a continuous ow process.
2. Denitions and assumptions
The moving average can be determined from a
time series of individual data values by nding the
mean of the rst n consecutive values, then drop-
ping the oldest value and subsequently adding one
sample value from a new sample to form the suc-
cessive averages. The features of the model con-
sidered in this article are as follows:
(1) The time that the process remains in an in-
control state follows a Weibull distribution. The
p.d.f. is given by
f t ktt
t1
e
kt
t
; t > 0; t=1; k > 0; 1
where k is the scale parameter and t is the shape
parameter.
The Weibull hazard rate is derived as follows in
Eq. (2):
rt ktt
t1
; t > 0; t=1: 2
(2) The process is monitored by drawing a
random sample of size 1 at times, h
1
, h
1
h
2
,
h
1
h
2
h
3
; . . . ; let W
j

j
i1
h
i
for j 1; 2; . . .
W
0
0, where j is the number of samples of size 1
taken from the process. Fig. 1 depicts the sampling
scheme in a continuous ow process for the MA
chart. For non-uniform sampling, the optimal
sampling interval h
j
is chosen such that the prob-
ability of a process shift in any interval, given no
shift until the start of the interval, is a constant
for all intervals. According to this denition, the
integrated hazard rate over each sampling interval
is a constant, that is,
_
W
j1
W
j
rt dt
_
h
1
0
rt dt: 3
The h
j
and W
j
can easily be obtained as follows:
h
j
bj
1=t
j 1
1=t
ch
1
; 4
and
W
j
j
1=t
h
1
: 5
The h
j
j 1; 2; . . . is a function of h
1
and
possesses the following two properties: (I) h
1
P
h
2
; and (II) lim
m!1

m
j1
h
j
1.
(3) The time to sample and chart one item is
negligible, and production ceases during the sear-
ches and repair.
(4) The process is normally distributed and
characterized by an in-control state. An assignable
cause occurring at random by a magnitude d re-
sults in a shift in the mean from u
0
to either u
0
dr
Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610 601
or u
0
dr, where u
0
, r, and d are, respectively, the
process mean, the process standard deviation, and
the shift parameter.
Dene P
j
j 1; 2; . . . as the conditional
probability that an assignable cause will occur
during the sampling interval h
j
, given that the
process is in an in-control state at time W
j1
, that is,
P
j

_
W
j
W
j1
f t dt
_
1
W
j1
f t dt
1 e
kh
t
1
: 6
The P
j
s are the same for j 1; 2; . . . For conve-
nience, let P
j
P
0
for j 1; 2; . . .
Dene q
j
(j 1; 2; . . .) as the unconditional
probability that an assignable cause occurs during
the sampling interval h
j
, and
q
j

_
w
j
w
j1
f t dt 1 P
0

j1
P
0
for j 1; 2; . . .
7
3. Model components and cycle length
A complete description of the cost structure
model was presented by Lorenzen and Vance
(1986). The time factors in our model involve the
following terms:
1. Average search time per false alarm, denoted as
Z
0
.
2. Average time to discover the assignable cause
once an assignable cause detected, denoted as Z
1
.
3. Average time to repair the assignable cause
once the assignable cause discovered, denoted
as Z
2
.
Cost factors in the model are considered as fol-
lows:
1. Average quality cost per unit time when the
process is in-control, denoted as D
0
.
2. Average cost per false alarm when the process is
in-control, denoted as Y.
3. Average quality cost per unit time when the
process is out of control, denoted as D
1
.
4. Cost to locate and repair the assignable cause,
denoted as w.
5. The xed sample cost and cost per unit sam-
pled, denoted as a and b, respectively.
A complete average cycle length is illustrated in
Fig. 2. The objective is to nd the optimal design
parameters, including moving subgroup size (n),
the rst sampling interval h
1
, and control limit
coecient (L), which can minimize the loss cost for
the given cost and time parameters.
Since the time for occurrence of an assignable
cause follows a Weibull distribution, the average
process in-control time (denoted as AVGICT) is
Fig. 1. Sampling scheme and plotting of an MA control chart for a continuous ow process.
602 Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610
AVGICT

1
j0
_
W
j1
W
j
tf t dt

1
k
_ _
1=t
C 1
_

1
t
_
; 8
where Cy is the gamma function and y P1.
Dene s
j
(j 1; 2; . . .) as the expected duration
of the in-control period within the sampling in-
terval h
j
, given that the shock occurred during
sampling interval, that is,
s
j

_
W
j
W
j1
t W
j1
f t dt
q
j
: 9
Thus, the expected in-control time during a sam-
pling interval in which the transition from an in-
control state to an out-of-control state occurred is
obtained using Eq. (10) as follows:
s

1
j1
s
j
q
j

1
k
_ _
1=t
C 1
_

1
t
_
h
1
P
0
1 P
0
A1 P
0
;
10
where Ax

1
l0
l 1
1=t
x
l
, jxj < 1.
When the process is out of control, the process
mean will shift to u
0
dr. Suppose that the shift
occurs in the (j 1)th sampling interval, and let
the mean of this moving subgroup be u. Let P
ji
denote the probability that the process mean is
shifted in the sampling interval h
j1
and that the
assignable cause is detected at the subsequent ith
moving subgroup. Then, j i g is the number
of samples before the assignable cause is detected.
Thus, Q
ji
1 P
ji
will be the probability that the
assignable cause will not be detected in the fol-
lowing moving subgroups. Three dierent situa-
tions for u and P
ji
, can be obtained as follows:
u
u
0

id
g
r if g < n;
u
0

id
n
r if g Pn and i < n;
u
0
dr if g Pn and i Pn;
_

_
11
and
P
ji

1 U L
id

g
p
_ _
U L
id

g
p
_ _
if g < n;
1 U L
id

n
p
_ _
U L
id

n
p
_ _
if g Pn and i < n;
1 UL d

n
p
UL d

n
p

if g Pn and i Pn;
_

_
12
where UX is the CDF of the standard normal
distribution and L is the control limit. However,
when g Pn and i Pn in Eq. (12), P
ji
is independent
of j or i, let P
ji
P and Q
ji
Q 1 P.
From the three situations mentioned in Eq.
(12), the average time from a shift occurring in the
Fig. 2. A complete average cycle length.
Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610 603
(j 1)th sampling interval to the assignable cause
detected in a later subgroup is derived and denoted
as E
j
t. E
j
t consists of two parts, the average
times for the assignable cause to be detected in the
subsequent ith sample for i < n (denoted as Pt
j
1
)
and for i Pn (denoted as Pt
j
2
), i.e., E
j
t Pt
j
1
Pt
j
2
.
Thus,
Pt
j
1
W
j1
W
j
P
j1

n1
i2
W
ij
W
j
P
ji

i1
l1
Q
jl
_ _
h
1
j
_
1
1=t
j
1=t
_
P
j1
h
1

n1
i2
i
_
j
1=t
j
1=t
_
P
ji

i1
l1
Q
jl
_ _
;
13
Pt
j
2

1
in
W
ij
_
W
j
_

n1
l1
Q
jl
_ _
Q
in
P
h
1

n1
l1
Q
jl
_ _
An
_
j; Q j
1=t
1
1 Q
_
;
14
where Ak; x

1
m0
m k
1=t
x
m
for jxj < 1.
Let Et be the expected time from an occur-
rence of a shift in the subgroup to the detection of
the assignable cause, and let it be formulated as
follows:
Et

1
j0
E
j
tq
j1

1
j0
Pt
j
1
Pt
j
2
1 P
0

j
P
0
: 15
Combining Eqs. (10), (13)(15), we obtain the
expected time to detect the assignable cause after
the process shift, denoted as AVGOOCT. Thus,
AVGOOCT Et s: 16
The expected number of false alarms (ENF)
depends on Type I error a and the number of
sampled moving subgroups before an assignable
cause occurred. The ENF can be obtained as fol-
lows:
ENF a

1
j1
jq
j
_
1
_
a
1 P
0
P
0
_ _
; 17
where a 21 UL.
The average time to search false alarms is
Z
0
ENF. If we let ATENF Z
0
ENF, then the ex-
pected length of a cycle, ET, can be obtained
from Eqs. (9)(17). The result is as follows:
ET AVGICT AVGOOCT
Z
1
Z
2
ATENF
Et h
1
P
0
1 P
0
A1 P
0

Z
1
Z
2
Z
0
a
1 P
0
P
0
_ _
: 18
The expected total cost incurred during a cycle
consists of the following cost components:
1. The expected cost to search false alarms is C
1
,
and C
1
Y ENF.
2. The expected quality cost for in-control process
is C
2
, and C
2
D
0
AVGICT.
3. The expected quality cost for out-of-control
process is C
3
, and C
3
D
1
AVGOOCT.
4. The cost to locate and repair the assignable
cause is C
4
, and C
4
w.
5. The expected sampling cost is C
5
, and C
5

a br
1
r
2
, where r
1
is the number of
sampled units before an assignable cause oc-
curs, r
2
is the number of sampled units before
the assignable cause is detected after a cause
occurs. Thus, r
1
and r
2
are derived as follows:
r
1

1
j1
jq
j

1
P
0
19
and
r
2

1
j0

n1
i2
i
_
1P
ji

i1
l1
Q
jl
_ _

n1
l1
Q
jl
_ _
n
_
1
Q
P
_
_
1 P
0

j
P
0
:
20
Combining the cost components mentioned
above, we obtain the expected total cost, EC, as
follows:
604 Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610
EC C
1
C
2
C
3
C
4
C
5
aY
1 P
0
P
0
_ _
D
1

1
j0
Pt
j
1
_
Pt
j
2
1 P
0

j
P
0
_
D
0
D
1
s D
0
h
1
P
0
1 P
0
A1 P
0

a b r
2
_

1
P
0
_
w: 21
The optimum values for design parameters n, h
1
and L can be determined by minimizing the loss
cost function EA, that is, EA EC=ET.
4. Determination of optimal design parameters
Basing our classication on the characteristics
of the variables or parameters in the expected loss-
cost model, the parameters can be classied into
ve categories. There are cost parameters Y ; D
0
;
a; b; D
1
; w; time parameters Z
0
; Z
1
; Z
2
; shift para-
meter d; Weibull distribution parameters k; t and
Table 1
The eect on the optimal design as t changes
Optimal design
t k n h
1
L Loss cost
1 0.02 2 0.76 2.51 410.392
1.5 0.002425 2 2.96 2.65 378.730
1.8 0.0007078 2 4.99 2.65 375.228
2 0.000314 2 6.46 2.65 372.730
3 0.0000057 2 13.71 2.65 359.989
Note: The mean duration of time to failure for each pair k; t is same.
Table 2
The eect on the optimal design as k changes
Optimal design
k t n h
1
L Loss cost
0.00001 2.859 2 12.74 2.65 351.750
0.0001 2.284 2 8.57 2.65 369.141
0.001 1.716 2 4.40 2.65 376.347
0.015 1.067 2 0.81 2.65 382.222
Note: The mean duration of time to failure for each pair k; t is same.
Table 3
The eect on the optimal design as either k (or t) changes while t (or k) is xed
Optimal design
k t Mean n h
1
L Loss cost
0.0001 2 88.62 2 9.8 2.66 336.257
0.001 2 28.03 2 4.24 2.63 416.556
0.015 2 7.24 2 1.60 2.59 524.906
0.001 1.5 90.33 2 4.34 2.66 339.338
0.001 1.8 41.29 2 4.36 2.65 389.346
0.001 3 8.93 2 3.41 2.60 482.747
Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610 605
Table 4
The sensitivity results of changing one parameter from the base as t 3, k 0:001
Quantity n h
1
L Loss cost
d 1 5 3.05 2.49 609.857
1.5 3 3.23 2.57 532.597
1.8 3 3.25 2.70 502.427
2.2 2 3.42 2.67 466.810
2.5 2 3.45 2.77 448.753
3 2 3.51 2.87 428.870
Z
0
0.625 2 3.39 2.66 487.813
0.9375 2 3.40 2.63 485.369
1.125 2 3.40 2.61 483.819
1.375 2 3.41 2.58 481.642
1.5625 2 3.42 2.56 479.915
1.875 2 3.43 2.52 476.835
Z
1
Z
2
1.625 2 3.43 2.58 552.136
2.4375 2 3.42 2.59 515.126
2.925 2 3.41 2.59 495.199
3.575 2 3.40 2.60 470.090
4.0625 2 3.40 2.60 454.185
4.875 2 3.39 2.61 428.806
Y 1000 2 3.53 2.17 454.623
1500 2 3.45 2.45 472.036
1800 2 3.42 2.55 478.909
2200 2 3.40 2.64 486.160
2500 2 3.38 2.69 490.679
3000 2 3.36 2.76 497.058
w 500 2 3.40 2.61 444.046
750 2 3.40 2.60 463.399
950 2 3.41 2.60 478.878
1100 2 3.41 2.59 490.484
1250 2 3.41 2.59 502.089
1500 2 3.42 2.58 521.425
D
0
105 2 3.39 2.62 410.154
157.5 2 3.40 2.61 446.460
189 2 3.40 2.60 468.234
231 2 3.41 2.59 497.256
262.5 2 3.42 2.59 519.014
315 2 3.43 2.58 555.260
D
1
2000 2 3.96 2.60 399.862
3000 2 3.62 2.60 445.088
3600 2 3.48 2.60 468.363
4400 2 3.34 2.59 496.384
5000 2 3.26 2.59 515.659
6000 2 3.14 2.59 545.226
a b 20 3 2.81 3.02 428.956
30 2 3.19 2.70 459.629
36 2 3.33 2.64 474.021
44 2 3.48 2.56 490.906
50 2 3.59 2.51 502.256
60 2 3.74 2.44 519.268
Note: Base case: d 2, Z
0
1:25 h, Z
1
Z
2
3:25 h, Y $2000, w $1000, D
1
$4000, a b $40, D
0
$210.
606 Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610
design parameters n; h
1
; L. Two numerical exam-
ples are used to verify the capability of the model.
We modify Rahims BASIC computer program
to meet our research needs and solve the optimal
design parameters to reach the minimum of the
EA (cf. Rahim, 1993).
Example 1. The cost, time, shift, Weibull distri-
bution parameters are set-up as follows: Y
$2000, w $1000, D
1
$4000, D
0
$210, a $20,
b $20, Z
0
1:25 h, Z
1
1:25 h, Z
2
2 h, k
0:02, t 1, and d 2. (Note: h represents hours.)
The optimal economic design parameters solved
from the program are n 2, h
1
0:59 h and
L 2:65, and the minimal loss cost is $383:011.
Example 2. Reset the parameters in Example 1 as
follows: Y $500, w $1100, D
1
$950, D
0

$50, a $20, b $20, Z
0
0:25 h, Z
1
Z
2
1:0
h, k 0:002, t 3 and d 2. The optimal design
parameters are n 2, h
1
4:13 h, L 2:08 and
the minimal loss cost is $297:972.
To study the eect on the design parameters
and loss cost of the various combinations of
Weibull parameters, we let the mean time to failure
be a constant 50 and the cost, time, and shift pa-
rameters be the same as in Example 1. Numerical
results for the parameters n; h
1
; L and the loss
cost under dierent pairs of k; t are shown in
Tables 1 and 2. Table 1 shows that an increasing t
results in an increasing sampling interval h
1
, the
decreasing of the loss cost, and no signicant
change in moving subgroup size n. The shortest
sampling interval and largest loss cost occur when
the failure time follows an exponential distribu-
tion, i.e., t 1. Table 2 reveals that as k increases,
the h
1
decreases, the loss cost increases and there
is little eect on L.
The numerical results of using the same set-up
of model parameters as in Example 1, except that
dierent pairs of k; t are used, are shown in
Table 3. Table 3 indicates that the loss cost in-
creases when one of the k; t is xed and the other
increases. Also, the higher the value of mean time
to failure, the smaller the value of loss cost.
However, we cannot conclude that the loss cost is
a decreasing function of mean time to failure. An
increasing k (xed t), results in a decreasing h
1
and
L. There is a signicant eect on n, and no extreme
change in L when t increases and k is xed.
5. Sensitivity analysis
In this section, we discuss the robustness of the
model when the time, cost, and shift parameters
vary. If we use the parameter values given in Ex-
ample 1 and t 3, k 0:001 as our base case,
each of the following parameters d; Z
0
; Z
1

Z
2
; Y , w; D
0
; D
1
, and a b are perturbed by
10%, 25%, and 50% from the base case. The
optimal values of n, h
1
, L and the loss cost for all
48 cases are presented in Table 4. Several points
can be observed from an analysis of Table 4.
1. An increasing value of d leads to a decreasing n
and loss cost, but an increasing h
1
and L.
2. Z
0
; Z
1
Z
2
and w have little eect on n, h
1
and
L; however, the parameters Z
1
Z
2
and w have
more impact on the loss cost.
3. Y has a signicant eect on h
1
and L, but a less
signicant eect on the loss cost.
4. When D
1
is small, h
1
is large and the loss cost is
small.
5. When D
0
varies, there is no signicant eect on
n, h
1
and L, but a signicant eect on the loss
cost.
6. A large sample cost a b leads to a smaller
moving subgroup size n, but has a signicant ef-
fect on the h
1
, L, and loss cost.
6. Conclusion
This paper is a presentation of a detailed devel-
opment of an economically based MAcontrol chart
where the process-failure mechanism follows a
Weibull distribution. The loss-cost model is well
established and formulated, and the optimal design
parameters for the cost model are solved using a
BASIC program. The numerical and sensitivity
analyses are performed to reveal how the loss cost,
the design parameters, and the model parameters
relate. In this paper, we successfully extend the
economically based MAcontrol chart design froma
Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610 607
uniform sampling scheme to a non-uniform sam-
pling scheme for use in a continuous ow process.
Acknowledgements
Part of this work was supported by the Na-
tional Science Council of Taiwan under grant No.
NSC89-2213-E-155-066.
Appendix A. Proof of ET and EC
The proposed procedure considers the system at
the end of the rst sampling moving subgroup.
Systems of equations can be formulated by taking
into account all possible states of the system, the
residual times and cost in the cycle, and associated
probabilities. Let T and C, respectively, be the
residual time and cost in the rst sampling moving
subgroup interval h
1
whether an assignable cause
has or has not occurred. Let T
1
and C
1
be the re-
sidual time and cost in the second moving sub-
group interval h
2
whether an assignable cause
has or has not occurred, given that no assignable
cause has occurred at the rst moving subgroup
interval. In a similar fashion, we may dene T
2
;
T
3
; . . . and C
2
; C
3
; . . .
Let T
j
and C
j
(j 0; 1; 2; . . .) be the residual
times and cost in the cycle beyond W
j
, given that
the process is in the in-control state at time W
j
.
Clearly, T
0
T and C
0
C.
The proofs of ET and EC are illustrated in
the following lemmas.
Lemma A.1. The following statements are true:
ET P
0
Pt
0
1
Pt
0
2
Z
1
Z
2

1 P
0
h
1
aZ
0
ET
1
; A:1
ET
1
P
0
Pt
1
1
Pt
1
2
Z
1
Z
2

1 P
0
h
2
aZ
0
ET
2
A:2
for j 2; 3; . . . ; and
ET
j1
P
0
Pt
j1
1
Pt
j1
2
Z
1
Z
2

1 P
0
h
j
aZ
0
ET
j
: A:3
From (A.1)(A.3), we can obtain
ET P
0
Pt
0
1
Pt
0
2
Z
1
Z
2

1 P
0
fh
1
aZ
0
P
0
Pt
1
1
Pt
1
2

Z
1
Z
2
1 P
0
h
2
aZ
0
ET
2
g
P
0
Pt
0
1
Pt
0
2
1 P
0
Pt
1
1
Pt
1
2

1 P
0

2
Pt
2
1
Pt
2
2

P
0
Z
1
Z
2
1 1 P
0
1 P
0

2

aZ
0
1 P
0
1 P
0

2

1 P
0
h
1
1 P
0

2
h
2
1 P
0

3
h
3

1
j0
Pt
j
1
Pt
j
2
1 P
0

j
P
0
Z
1
Z
2
aZ
0
1 P
0
P
0
1 P
0
P
0
h
1
A1 P
0
;
A:4
where Pt
j
1
and Pt
j
2
are the same as Eqs. (13) and (14).
Lemma A.2. The following statements are true:
EC a b
P
0
a
_
b

n1
i2
i
_
1P
0i

i1
l1
Q
0l
_ _

i1
l1
Q
0l
_ _
n
_
1
Q
P
_
_
D
0
D
1
s
1
D
1
Pt
0
1
Pt
0
2
w
_
1 P
0
D
0
h
1
aY EC
1
A:5
for j 2; 3; . . . ; and
EC
j1
a b P
0
a
_
b

n1
i2
i
_
1
P
j1i

i1
l1
Q
j1l
_ _

i1
l1
Q
j1l
_ _
n
_
1
Q
P
_
_
D
0
D
1
s
j
D
1
Pt
j1
1
Pt
j1
2
w
_
1 P
0
D
0
h
j
aY EC
j
; A:6
608 Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610
From (A.5) and (A.6), we can obtain
EC a b
P
0
a
_
b

n1
i2
i
_
1P
0i

i1
l1
Q
0l
_ _

n1
l1
Q
0l
_ _
n
_
1
Q
P
_
_
D
0
D
1
s
1
D
1
Pt
0
1
Pt
0
2
w
_
1 P
0
D
0
h
1
_
aY a b
P
0
a
_
b

n1
i2
i
_
1P
1i

i1
l1
Q
1l
_ _

n1
l1
Q
1l
_ _
n
_
1
Q
P
_
_
D
0
D
1
s
2
D
1
Pt
1
1
Pt
1
2
w
_
1 P
0
D
0
h
2
aY EC
2

_
a b1 1 P
0
1 P
0

2

a bP
0

n1
i2
i
_
1P
0i

i1
l1
Q
0l
_ _

n1
l1
Q
0l
_ _
n
_
1
Q
P
_
1 P
0

n1
i2
i
_
1P
1i

i1
l1
Q
1l
_ _

n1
l1
Q
1l
_ _
n
_
1
Q
P
_
_

_
D
0
D
1
P
0
s
1
1 P
0
s
2
1 P
0

2
s
3

wP
0
1 1 P
0
1 P
0

2

aY 1 P
0
1 1 P
0
1 P
0

2

D
1
P
0
Pt
0
1
Pt
0
2
1 P
0
Pt
1
1
Pt
1
2

D
0
h
1
1 P
0
h
2
1 P
0

2

a b
1
P
0
a b

1
j0

n1
i2
i
_
1P
ji

i1
l1
Q
jl
_ _

n1
l1
Q
jl
_ _
n
_
1
Q
P
_
_
1 P
0

j
P
0
D
0
D
1
s w aY
1 P
0
P
0
_ _
D
1

1
j0
Pt
j
1
Pt
j
2
1 P
0

j
P
0
D
0
1 P
0
P
0
h
1
A1 P
0
: A:7
References
Banerjee, P.K., Rahim, M.A., 1987. The economic design of
control charts: A renewal theory approach. Engineering
Optimization 12, 6673.
Banerjee, P.K., Rahim, M.A., 1988. Economic design of xx
control charts under Weibull shock models. Technometrics
30, 407414.
Chen, Y.S., Yang, Y.M., 1999. Economic design of xx control
charts for continuous ow process with Weibull in-control
time. In: The 13th Asia Quality Symposium 99 Osaka, pp.
1720.
Chiu, W.K., Wetherill, G.B., 1974. A simplied scheme for the
economic design of xx-charts. Journal of Quality Technology
6, 6369.
Duncan, A.J., 1956. The economic design of xx charts used to
maintain current control of a process. Journal of the
American Statistical Association 51, 228242.
Duncan, A.J., 1971. The economic design of xx-charts
when there is a multiplicity of assignable causes.
Journal of the American Statistical Association 66,
107121.
Ho, C., Case, K.E., 1994. The economically-based EWMA
control chart. International Journal of Production Research
32, 21792186.
Koo, T.Y., Case, K.E., 1990. Economic design of x-bar control
charts for use in monitoring continuous ow processes.
International Journal of Production Research 28, 2001
2011.
Lorenzen, T.J., Vance, L.C., 1986. The economic design of
control charts: A unied approach. Technometrics 28, 310.
Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610 609
Ohta, H., Rahim, M.A., 1997. A dynamic economic model for
an xx-control chart design. IIE Transactions 29 (6), 481486.
Parkhideh, B., Case, K.E., 1989. The economic design of a
dynamic xx-control chart. IIE Transactions 21, 313323.
Rahim, M.A., 1993. Economic design of xx control charts
assuming Weibull in-control times. Journal of Quality
Technology 25, 296305.
Rahim, M.A., Banerjee, P.K., 1993. A generalized model for
the economic design of xx-control charts for production
system with increasing failure rate and early replacement.
Naval Research Logistics 40 (6), 787809.
Rahim, M.A., Costa, A.F.B., 2000. Joint economic design of xx
and R charts under Weibull shock models. International
Journal of Production Research 29 (3), 845873.
Saniga, E.M., 1977. Joint economically optimal design of xx and
R control charts. Management Science 24, 420431.
Taylor, H.M., 1968. The economic design of cumulative sum
control charts. Technometrics 10, 479488.
610 Y.-S. Chen, Y.-M. Yang / European Journal of Operational Research 143 (2002) 600610