Computers & Industrial Engineering 102 (2016) 374–382

Contents lists available at ScienceDirect

Computers & Industrial Engineering

journal homepage: www.elsevier.com/locate/caie

Integration of production quantity and control chart design

in automotive manufacturing
E.E. Gunay ⇑, U. Kula
Department of Industrial Engineering, Sakarya University, Sakarya, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: This study presents a two-stage stochastic programming model for the determination of control limits in
Available online 16 June 2016 p-charts when a production process produces above a certain quantity. Consideration of production
quantity needed along with control limit determination is important for the following competing two
Keywords: reasons: (1) Wider control limits make it difficult to detect the changes in the process, therefore produc-
Attribute chart (p-chart) design ing excessive number of cars with paint defects. (2) Narrower control limits, on the other hand, increase
Stochastic models the number of unnecessary interventions even if there is no deterioration in the process so that inspec-
Sample average approximation
tion costs increase. In both cases, quantity produced reduces due to defective products and unnecessary
Quality management
interventions. Therefore, it is important to design a control chart for proportion of defects that takes pro-
duction quantity requirements into account. We consider the problem in an automotive manufacturing
setting in which the cars are inspected for paint defects after paint operations.
We formulate the problem as a two-stage stochastic programming model. In the first stage, control
limit parameter k is decided for the p-chart and in the second stage, production quantity is determined
that minimizes total quality-related and production costs. We solve the model by sample average
approximation algorithm (SAA). In a numerical study, we investigate the effect of various factors on con-
trol limit parameter k and the total cost. Our numerical study shows that (i) an increase on the mean
defect rate increases both the total cost and the total production quantity, (ii) effect of an increasing pro-
cess variance to the control limit parameter k is significantly small, (iii) frequency of special cause occur-
rences affects the total cost significantly and (iv) all the experiments show that the commonly used 3r
control limits in practice are wider than required.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction
In automobile manufacturing, after body shop operations,
welded bodies are directed to the paint shop. Vehicle painting
operation includes several steps such as cleaning, primer coat
painting, top coating and polishing. Once painting is completed,
the vehicles are inspected for paint defects before they are released
to the Final Assembly (FA) department. Since paint defects may be
as much as 40%, some manufacturers use control charts to monitor
the paint process and determine whether the current painting pro-
cess is out-of-control. An out-of-control situation means that there
is a special cause increasing the proportion of paint defects. In
vehicle painting operations, defects are categorized as either minor
or major ones. Minor defects just need small touchups, usually
done on the line, which does not cause any delays in assembly
⇑ Corresponding author.
E-mail addresses: ekabeloglu@sakarya.edu.tr (E.E. Gunay), ukula@sakarya.edu.tr
(U. Kula).
operations. On the other hand, vehicles with major defects must
be taken off the line and sent back to paint shop for re-painting.
Since minor defects do not add any significant cost and do not dis-
turb the vehicle flows in the line, the scope of this study is confined
to major paint defects.
Fig. 1 shows the simplified inspection procedure used in the
paint shop of a major car manufacturer’s plant located in Turkey.
A sample of size n is drawn in a sampling interval of h hours at
the end of epoch t from the batch size of X t vehicles. Number of
the defective vehicles in the sample is counted and estimated aver-
age defect rate of the process p^, is calculated. The painting process
is deemed out-of-control if the defect rate of the sample p ^, is above
the upper control limit (UCL). When the p-chart signals an out-of-
control situation, the engineering team searches for special causes
that might increase the proportion of paint defects. If a cause for
the defect rate increase is found, the painting process is restored
to the in-control state at some cost and all the newly painted vehi-
cles in the sampling interval are inspected. Since no action is taken
if the defect rate of the sample is below UCL, determination of
lower control limit in p-charts is of no practical importance.

http://dx.doi.org/10.1016/j.cie.2016.06.016
0360-8352/Ó 2016 Elsevier Ltd. All rights reserved.
 E.E. Gunay, U. Kula / Computers & Industrial Engineering 102 (2016) 374–382 375

Randomly take sample of size n from a

batch of Xt vehicles at the end of
production period t

Does the chart signal an out-of- Yes Inspect all the newly painted
t t 1
control in the tth production period? vehicles in the tth production period

No

Send defective vehicles for re-painting.

Fig. 1. Simplified procedure of paint control process.

Setting an optimal UCL is important to reduce the production
and quality costs: A widely set UCL decreases the sensitivity of a
control chart to detect the special cause occurrences. On the other
hand, a narrow UCL increases the number of false out-of-control
signals in the process even if there is no deterioration in the
process.
As Eq. (1) suggests, setting an optimal UCL involves finding the
optimal k that would minimize quality related costs. Determina-
tion of optimal control limits along with optimal sampling size
and sampling frequency that minimize quality related costs is
known as economical design of chart parameters and is commonly
studied in the literature.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p0 ð1  p0 Þ
UCL ¼ p0 þ k
n costs simultaneously.
Ladany (1973) is the first paper that considers economical
design of p-chart parameters. His model includes cost of sampling,
cost of not detecting a change in the process (Type 2 error cost), cost
of false indication of change (Type 1 error cost) and cost of re-
adjusting detected change.
Montgomery, Heikes, and Mance (1975) consider the same
problem when there are several out-of-control states. Chiu
(1975a) presents a model which minimizes loss cost function in
an np-control chart and Chiu (1975b) investigate the effects of vari-
ation in cost factors by drawing loss cost surface as contour plots.
Chiu (1976) considers the case where there are several out-of-
control states in an np-control chart. More recent studies on eco-
nomical design of chart parameters consider variable sample size,
variable sampling interval and both variable sample size/sampling
interval and present the advantages of these models rather than
traditional p-chart design (Aslam, Azam, Khan, & Jun, 2015;
Inghilleri, Lupo, & Passannanti, 2015; Kooli & Limam, 2011; Wu
& Luo, 2004).
Recognizing the effect of the monitoring policy on the produc-
tion capacity, Lee and Rosenblatt (1987) develop a model address-
ing the problem of joint determination of optimal production run
time, number of inspections to minimize quality-related costs in
an X chart. Lee and Park (1991) consider the same problem by
focusing on the difference between rework cost before sale and
warranty cost after sale. Rahim (1994) considers the same joint
problem by developing a non-Markovian shock model under pro-
duction setup, inventory holding and maintenance cost. Rahim
and Ben-Daya (1998) extend this model to consider the case in
which the production is halted not only if there is true alarm but
also there is a false alarm. Ben-Daya and Rahim (2000) consider
the joint-optimization of preventive maintenance actions and X
chart parameters when in-control state follows a general probabil-
ity distribution with increasing hazard rate. Pan, Jin, Wang, and
Cang (2012) develop a model to minimize the total expected pro-
duction costs while jointly determining the optimal parameters
of control chart and the maintenance decision policy whereas
Bouslah, Gharbi, and Pellerin (2015) consider joint design of pro-
duction, quality and maintenance control policy problem in c-
chart.
In vehicle painting process, since defects in the paint shop occur
randomly, it is not possible to know before how many vehicles to
paint in order to meet the FA demand. Therefore, a control chart
policy should take random paint defect occurrences into account
ð1Þ to meet assembly line demand and to minimize quality related
Economic design problem of control charts, including p-chart
design, involve determination of three important parameters: Con-
trol limit width, sampling interval, and sample size. In a p-chart,
narrowly determined UCL increases the number of false alarms
and when a p-chart reports an alarm, all the vehicles painted in
between previous and the current sampling epochs are inspected
one by one, which increases the unnecessary cost of inspection.
This inspection cost depends on the number of vehicles painted
in the batch. Therefore, if the number of painted vehicles is more
than the optimal number needed, extra inspection costs occur.
On the other hand widely determined UCL makes difficult to detect
the shift in the process so defective vehicles will be sent to the FA
department.
In this paper, we develop a two-stage mathematical model that
jointly determines the optimal upper control limit in a p-chart and
the paint batch size X t . In the first stage, the control limit parame-
ter k is set and after observing the random paint defects, the paint
batch size X t is decided. The two-stage decision making framework
optimizes the first stage decision k given the fact that vehicles are
repeatedly painted and paint defects are experienced over and over
again. In fact, in statistical process control applications, the control
chart parameters are determined first and random process shifts
occur repeatedly as process is monitored over time. Therefore,
two-stage stochastic programming provides an appropriate model-
ing framework for incorporating various constraints such as
demand, service level, and maximum inventory level extra into
control chart design problem. Bouslah, Gharbi, and Pellerin
(2013) built a stochastic mathematical model and use a simulation
based optimization approach for joint determination of the pro-
duction quantity, hedging level and the sample size in X charts.
376 E.E. Gunay, U. Kula / Computers & Industrial Engineering 102 (2016) 374–382

There are also some other studies that both decide on quality and pout ¼ 1  ek ð2Þ
maintenance decisions to both lower the cost of quality and main-
where pout is the probability that the painting process defect rate
tenance. These models aim to both improve quality of items pro-
would shift from p0 to p1 . Due to sampling error, it is possible to
duced also lower maintenance costs (Chan & Wu, 2009; Mehdi,
get a false out-of-control signal even if the process defect rate has
Nidhal, & Anis, 2010; Mehrafrooz & Noorossana, 2011; Pandey,
not shifted from p0 to p1 . This error is called as Type 1 error and
Kulkarni, & Vrat, 2011).
is denoted by Pðp^ P UCLjp1 ¼ p0 Þ.
After developing the two-stage stochastic model, we perform a
Since number of defectives D in a sample of size n is a binomial
numerical study that investigates the effect of several cost compo-
random variable with mean np0 , Type 1 error probability, the prob-
nents and random disturbances on first stage decision variable k
ability of having more than n(UCL) defects in the sample given that
and the total cost. Lam and Rahim (2002) present an extensive sen-
the process mean stays as p0 , a ¼ PðD P nUCLjp1 ¼ p0 Þ is given as
sitivity analysis similar to ours to investigate the effects of cost and
follows:
inspection parameters on production and quality costs in X charts.  
Chakraborty, Giri, and Chaudhuri (2009) study the joint effect of Xn
n d nd
a¼ p0 1  pd0 ð3Þ
process shift, machine breakdown and inspection decisions on d
dPnðUCLÞ
production quantity for deteriorating production system where
p-chart is used to control the process. which can be approximated by normal distribution. Therefore we
To our knowledge, our study is the first that models determina- have,
tion of p-chart parameter k when a certain production quantity   Z 1
nUCL  np0 1 1 xl 2
needs to be produced in a planning horizon of length T with the a¼P ZP ¼ PðZ P kÞ ¼ pffiffiffiffiffiffiffi e2ð r Þ dx;
r k r 2p
objective of minimizing total production and quality-related costs.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
In the first stage two-stage stochastic model, we decide on the where r is equal to np0 ð1  p0 Þ and l equals to np0 . When a Type
p-chart control limit parameter k which determines the UCL for 1 error occurs, an unnecessary inspection cost is incurred since the
p-chart. In the second stage of the model, the production quantities true mean of the process is still p0 . Total expected Type 1 error cost
X t are determined for each period t for a planning horizon of length in a production period of length T is given in Eq. (4), where c1 shows
T. In addition to the demand constraint, a maximum inventory the Type 1 error cost per vehicle and Xt shows the number of vehi-
level constraint that guarantees the total production quantity does cles painted in production period t.
not exceed a certain percentage of the quantity produced, i.e. X
T
safety stock level SSL, is added to the second stage model. X t c1 ð1  pout Þa ð4Þ
This paper is organized as follows. Section 2 formulates the t¼1
problem and presents the two-stage stochastic model. In Section 3 When a special cause occurs and the defect rate of the process
we give the details of the sample average approximation (SAA) shifts to p1 , the probability of not detecting this shift is,
algorithm used to solve the developed model. A numerical study is b ¼ Pðp^ 6 UCLjp1 > p0 Þ, called as Type 2 error. Type 2 error occurs
performed to offer managerial insights in Section 4. Section 5 con- when the defect rate of the process shifts to p1 but the sample
cludes the paper and explores possible future research directions. ^ is less than the UCL so the deterioration in the process
defect rate p
is not detected. By Normal approximation to binomial distribution,
2. Two stage stochastic model we can calculate the Type 2 error probability b as follows:
Z  2
UCL xp1
1 1
Consider a painting process that is controlled by a p-chart and b¼ pffiffiffiffiffiffiffi e 2 rp1
dx ð5Þ
let the upper control limit of the p-chart is set at 1 r p1 2p
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
UCL ¼ p0 þ k p0 ð1  p0 Þ=n, where p0 is the in-control defect rate pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where l is equal to p1 and r equals to p1 ð1  p1 Þ=n.
of the process and k is the sigma level of the process. Once the When the process defect rate shifts to p1 , a p-chart will detect
value of k is decided, the painting process would be monitored at this shift with probability ð1  bÞ and all the newly painted vehi-
each discrete time epoch t. At each epoch a sample of size of n vehi- cles are inspected one by one. Since this inspection is different than
cles are inspected for paint defects and proportion of defectives in the inspection of sample at the end of each production period t and
the sample p ^ is calculated. If p ^ is below UCL, the paint nozzles are
requires inspection of all the vehicles painted in that production
purged and the next batch of vehicles are painted until the next period, we call it as re-inspection cost. Let c0 shows the cost of
inspection epoch t þ 1. If p ^ is above the UCL, it means that the p-
re-inspection per vehicle, then the expected total cost of re-
chart signals a shift in process defect rate from p0 to p1 , where inspection is,
p1 > p0 .
When p ^ is above the UCL, a search for the special cause is per- X
T
X t c0 pout ð1  bÞ ð6Þ
formed. If a special cause is found resulting in a higher than accept- t¼1
able defect rate UCL, the process is brought back to in-control state
at some cost and no cessation in the production. We assume that On the other hand, when the process defect rate shifts to p1 , a p-
the special causes in the painting process occur according to a Pois- chart will not detect this shift with probability b. If the shift in the
son Process with rate k vehicles per inspection interval, ½t; t þ 1. average defect rate of the process is not detected, the defective vehi-
Without loss of generality, we assume that inspection period cles will be sent to following department, FA. In this case FA workers
length is one time units. Therefore, time between special cause will be busy to take the vehicles off the line which may cause delays
occurrences follows an exponential distribution with mean 1=k. and line stoppages. In some cases, defective vehicles may not be
Modeling special cause occurrences as a Poisson Process is a rea- noticed at the final inspection after the FA and may be shipped to
sonable and a common assumption since multiple sources con- the customer. In practice, usually this cost is much higher than the
tribute to special cause occurrences. (Lam & Rahim, 2002; Lee & cost of false alarm, Type 1 cost. Let c2 presents the cost of Type 2 error
Rosenblatt, 1987; Montgomery et al., 1975). Since the process per vehicle. Then the total cost of Type 2 error is,
shifts to an out-of-control state when there is one or more special X
T
cause occurrences, the probability that there is one or more special X t c2 pout b ð7Þ
cause occurrences in each period ½t; t þ 1 is, t¼1
 E.E. Gunay, U. Kula / Computers & Industrial Engineering 102 (2016) 374–382 377

8
Once the value of k is decided, i.e., once UCL is set, the defective > z ¼ min fE½Q ðk; nÞg
< k2K
number of vehicles in the sample at epoch t is observed, which ðFSÞ k 6 B ð10Þ
depends on whether the process is in-control or out-of-control. >
:
Let the random variable It be an in-control or in an out-of- kP0
control state, In Eq. (10), E½Q ðk; nÞ is taken with respect to random vector n,
which is a vector of size T, whose ith element shows whether ith
1; if an out-of-control signal is not reported
It ¼ production period is in-control or out-of-control. First stage (FS)
0; if an out-of-control signal is reported
decision is usually associated with a cost coefficient cT, and the
For example if production period is limited with 6 periods, objective function in Eq. (10) is written as minfcT k þ E½Q ðk; nÞg.
T ¼ 6 and the realization of the sequence of random variables It In our model since no cost is associated with FS decision k, the cost
is as follows nðxÞ ¼ ð1; 0; 1; 1; 0; 1Þ. Then out-of-control signal is coefficient of the k is zero.
reported in production period 2 and 5 while an out-of-control sig- The second stage (SS) model determines minimum total cost
nal is not reported in other periods 1, 3, 4 and 6. If out-of-control Q ðk; nÞ, i.e. sum of all the costs in Eqs. (4) and (6)–(9), according
signal is not reported, i.e. p ^ 6 UCL, number of defective vehicles to the FS decision k and random realization nðxÞ of the random
in the sample ytin , are sent back to paint shop for re-painting, where vector n. Once the status of the processes nðxÞ are observed, then
ytin is a binomial random variable with mean np0 and variance the SS decisions on production quantity for the next production
np0 ð1  p0 Þ. When an out-of-control signal is reported, i.e. period X tþ1 is decided at the end of production period t so the FA
^ P UCL, then all the vehicles painted in that production period
p demand should be met. Production quantity X t shows the number
the are inspected and number of defects out of all the vehicles of vehicles to be painted in production periods between t and t + 1.
painted in period t, ytout are also sent back to paint for re- Before, introducing the second stage (SS) model, we list the
painting, where ytout is a binomial random variable with mean parameters, indices, variables, subscripts and decision expressions
X t p1 and variance X t p1 ð1  p1 Þ. However since ytout depends on used in the model.
the decision variable X t , we replace ytout by its expected value
X t p1 and obtain a fat solution. Then the expected cost due to re- Input parameters
painting can be written as, n Sample size
" # p0 Average defect rate of the process when in-control
X
T X
T
^
p Average defect rate of the sample
cp ytin ðIt Þ þ cp E X t p1 ð1  It Þ ð8Þ
t¼1 t¼1
p1 The average defect rate of the process after special cause
occurrence
where cp is cost of re-painting a vehicle. Note that if It ¼ 1, i.e., if p- k Rate of the occurrence of the special cause
chart does not signal an out-of-control situation, the second cost c Painting cost of each vehicle
component vanishes, and if It ¼ 0, i.e., if p-chart signals an out-of- cp Re-painting cost of a vehicle
control situation, the first cost component vanishes. The probability c0 Inspection cost
distribution of It , which is given by c1 Type 1 error cost
c2 Type 2 error cost
PðIt ¼ 0Þ ¼ PðIt ¼ 0jp1 > p0 ÞPðp1 > p0 Þ þ PðIt ¼ 0jp1 ¼ p0 ÞPðp1 ¼ p0 Þ
: D Re-painting time of defective vehicles
PðIt ¼ 1Þ ¼ PðIt ¼ 1jp1 > p0 ÞPðp1 > p0 Þ þ PðIt ¼ 1jp1 ¼ p0 ÞPðp1 ¼ p0 Þ FD Final assembly demand
Cap Capacity of paint process for each production period
Since Pðp1 > p0 Þ ¼ pout ¼ ek , we have BM Very big integer number
SSL Safety stock level in percentage
^ > UCLjp1 > p0 Þek þ Pðp
PðIt ¼ 0Þ ¼ Pðp ^ > UCLjp1 ¼ p0 Þð1  ek Þ
^ 6 UCLjp1 > p0 Þe þ Pðp
PðIt ¼ 1Þ ¼ Pðp k ^ 6 UCLjp1 ¼ p0 Þð1  ek Þ Indices
t Length of production period also sampling interval,
Since the costs given in Eq. (8) is due to repainting at the end of t ¼ 1; . . . ; T
each inspection epoch is a random variable, a realization of (8) may Decision variables
be calculated by obtaining random drawing from the probability The distance of the control limit from the process
k
distribution of It by the following routine: (1) first generate an average defect rate, control limit parameter
uniform random variable RN in ½0; 1, (2) If RN 6 pout ¼ ek , then Xt Number of vehicles painted in period t
the process is out-of-control, i.e. p1 > p0 . If RN > pout ¼ ek ,
then the process is in-control, i.e. p1 ¼ p0 . (3) If p1 > p0 , decide Decision expression
on the magnitude of shift d by drawing a uniform random number a Probability of exceeding UCL given the process is in-
RN in ½0; 1. control
Let c is a unit cost of painting a vehicle. Then the total cost of b Probability of not exceeding UCL given the process is
painting vehicles is, out-of-control
z Standard difference between UCL and p1 , z-statistic
X
T
value
cX t ð9Þ
1; If variable z is negative
t¼1 bv ¼
0; Otherwise
In our two-stage stochastic programming model, the first stage
decision is to decide on control limit parameter, k. Control limit
parameter k is decided before status of the process It as
 
in-control or out-of-control is known. The optimal k is one that Once the first stage decision k is determined and the random vec-
minimizes the expected total cost E½Q ðk; nÞ, respect to pre- tor n is observed, the following SS is solved in order to minimize the
determined limit B, and the solution of the first stage (FS) problem. total cost Q ðk; nÞ. Note that the production quantity X t and the
Objective function of the FS problem E½Q ðk; nÞ in Eq. (10) is the status of the process It depend on the random defect occurrences.
average sum of all the costs in Eqs. (4) and (6)–(9). To show this dependency, they should be typed as X t ðnÞ and It ðnÞ.
378 E.E. Gunay, U. Kula / Computers & Industrial Engineering 102 (2016) 374–382

However, to simplify the presentation of the SS model, we drop n zero, variable bv in Eq. (11f) and (11g) takes a value of 1 and Type 2
from X t and It . error probability equals to,

8 " #
>
> X T X T XT XT X T XT
>
> min Qðk; nÞ ¼ X c ð1  p Þ a þ X c p ð1  bÞ þ X c p b þ c y t
ðI Þ þ c E X p ð1  I Þ þ cX t ðaÞ
>
> t 1 out t 0 out t 2 out p in t p t 1 t
>
>
>
>
t¼1 t¼1 t¼1 t¼1 t¼1 t¼1
>
>
>X
>
T
>
> X t 6 FDð1 þ SSLÞ ðbÞ
>
> t¼1
>
>
>
>
>
> X t þ yinmaxf0;tD1g
ðImaxf0;tD1g Þ þ yout
maxf0;tD1g
ð1  ðImaxf0;tD1g Þ 6 Cap; 8t ¼ 1; . . . ; T ðcÞ
>
>
>
>
>
> 8
>
> a ¼ ð1 þ 0:09979271k þ 0:04432014k þ 0:0096992k3  0:00009862k4 þ 0:0058155k5 Þ =2
2
ðdÞ
>
>  qffiffiffiffiffiffiffiffiffiffiffiffiffiffi  qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
>
>
> p0 ð1p0 Þ p1 ð1p1 Þ
> z ¼ p0 þ k
> n
 p1 n
ðeÞ
>
>
>
>
< z 6 BM  ð1  bv Þ
> ðfÞ
SS ¼ z 6 BM  bv ðgÞ ð11Þ
>
>
>
> 8
>
>
> 1  ð1 þ 0:09979271jzj þ 0:04432014jzj2 þ 0:0096992jzj3  0:00009862jzj4 þ 0:0058155jzj5 Þ =2  b 6 BM  bv ðhÞ
>
>
>
> 8
>
> ð1 þ 0:09979271jzj þ 0:04432014jzj2 þ 0:0096992jzj3  0:00009862jzj4 þ 0:0058155jzj5 Þ =2  b 6 BM  ð1  bv Þ ðiÞ
>
>
>
> þ
> X t 2 Z ; 8t ¼ 1; . . . ; T
>
>
ðjÞ
>
> 06a61 ðkÞ
>
>
>
>
>
> 06b61 ðlÞ
>
>
>
>
>
> k 2 Rþ ðmÞ
>
>
>
>
> bv 2 f0; 1g ðnÞ
>
>
>
> y0in ¼ 0 ðoÞ
>
>
: 0
yout ¼ 0 ðpÞ

Objective function of the model in Eq. (11a) minimizes the total PðZ 6 zÞ ¼ UðzÞ;
cost. Total cost is the sum of costs as presented in Eqs. (4) and (6)– 8
UðzÞ ¼ ½ð1 þ a1 jzj þ a2 jz2 j þ a3 jz3 j  a4 jz4 j þ a5 jz5 jÞ =2 þ 2eðzÞ;
(9). Constraint (11b) ensures that the difference between total pro-
duction and FA demand must not exceed the safety stock level, SSL. jeðzÞj < 2  105 ; z60
Constraint (11c) guarantees that number of vehicles painted and ð13Þ
repainted must not exceed the production capacity, Cap.
We use normal approximation to binomial distribution for the When the variable z in Eq. (11e) is greater than zero, variable bv
calculation of Type 1 error in Eq. (3). There are several polynomial in Eq. (11f) and (11g) takes a value of 0 and Type 2 error probability
approximation methods to calculate the area under the normal is calculated as in Eq. (14),
curve that is greater than k, i.e. PðZ P kÞ, easily (Patel & Read,
PðZ 6 zÞ ¼ UðzÞ;
1982). One of the polynomial rational methods which has good
8
approximation with sufficiently small error, i.e. jeðkÞj < 2  10 , 5
UðzÞ ¼ 1  ½ð1 þ a1 jzj þ a2 jz2 j þ a3 jz3 j  a4 jz4 j þ a5 jz5 jÞ =2 þ 2eðzÞ;
is presented in Eq. (12). Approximation given in Eq. (12) is written jeðzÞj < 2  105 ; z P 0 ð14Þ
as Eq. (11d) in the SS problem.
According to the values of variable bv, constraint (11h) and (11i)
PðZ P kÞ ¼ 1  UðkÞ; calculates the Type 2 error probability accurately as presented in
2 3 4 5 8 Eqs. (13) and (14). The types of the variables used in the model
1  UðkÞ ¼ ½ð1 þ a1 k þ a2 k þ a3 k  a4 k þ a5 k Þ =2 þ 2eðkÞ;
are defined in Eq. (11j)–(11n). Constraints (11o) and (11p) satisfy
jeðkÞj < 2  105 ; kP0 that at the beginning of the production i.e., t ¼ 0, no defective
a1 ¼ 0:09979271; a2 ¼ 0:04432014; vehicles from previous day is observed.
a3 ¼ 0:0096992; a4 ¼ 0:00009862;
a5 ¼ 0:00581551 ð12Þ 3. SAA algorithm
For the polynomial approximation to Type 2 error probability in
Eq. (5), some modifications to Eq. (12) is required: If the new defect The main difficulty of solving stochastic optimization problems
rate of the process p1 is less than UCL, Type 2 error is greater than by conventional techniques is the number of possible realizations
0.5 and less than 0.5 vice versa. Since the approximation in Eq. (12) (scenarios) of the random variables. If the number of possible
is defined for the calculation of the right hand tail of the normal realizations of the random variable considered in the problem is
curve and maximum value of this approximation equal to 0.5, we large, evaluation of the expectation E½Q ðk; nÞ, becomes extremely
introduce a binary indicator variable bv to calculate Type 2 accu- difficult sometimes not possible. For example, in our problem

 
rately. The standard difference between UCL and p1 is calculated QT PnðUCLÞ n
the number of scenarios jXj is t¼1 d¼0  It þ
with variable z, in Eq. (11e). If the variable z in Eq. (11e) is less than d
 E.E. Gunay, U. Kula / Computers & Industrial Engineering 102 (2016) 374–382 379

  
PX t Xt Step 3: Increase the number of samples R to R0 , where R0  R.
 ð1  It Þ . Since the calculation of the exact
ytout ¼nðUCLÞ
ytout ^ , solve the second stage problem
Fix the first stage decision to k
expected value of the total costs E½Q ðk; nÞ, in FS problem in Eq. ^ Þ correspond to k
^
and calculate the expected total cost ^zR0 ðk
(15) is impossible due to the increase on number of scenarios.
0
The sample average approximation (SAA) which is a simulation X
R
^ Þ ¼ 1
^zR0 ðk ^ ; nr Þ:
Q ðk
based method developed by Kleywegt, Shapiro, and Homem- 0
R r¼1
de-Mello (2001) is used to solve two-stage stochastic optimization
problems. ^ Þ  zM and
Step 4: Calculate the optimality gap estimate, ^zR0 ðk
X
jXj the estimated variance of the gap, r ^ ^
2
¼r^ 2 ^ þ r^ 2z
^z ðk Þz ^z ðk Þ
R0
M
R0
E½Q ðk; nÞ ¼ pf Qðk; nf Þ ð15Þ
M

where,
f ¼1

XR0
The idea of the SAA algorithm is to approximate the expectation 1 ^ ÞÞ2
^ ; nr Þ  ^z 0 ðk
r^ ^2z 0 ðk^ Þ ¼ 0 0 ðQ ðk R and
in (15). A sample size R of n is taken from the scenario space X and R R ðR  1Þ r¼1
P
the expectation E½Q ðk; nÞ is replaced by 1=R Rr¼1 Q ðk; nr Þ and an 1 XM
ð^zm  zM Þ :
2
approximate first stage problem (AFS) is developed. It is shown r^ 2zM ¼
MðM  1Þ m¼1
by Kleywegt et al. (2001), that the optimal objective value ^z and
optimal solution k ^ to AFS, converge to the true optimal objective ^ Þ  zM is less than 1% of the costs, then stop.
Step 5: If gap ^zR0 ðk

value z and true optimal solution k of the stochastic program as Otherwise go to Step 1.
the size of the scenario R increases. In our model, we assume that
there is no cost associated with k so the cost coefficient cT of the 4. Numerical analysis
first stage decision variable all zero. Therefore, we can write the
AFS problem correspond to Eq. (10) as below, In this part of study, we present a set of numerical examples to
8 ! provide insights on how the magnitude of the shift, special cause
>
> X
R
>
< ^z ¼ mink
> Q ðk; nr Þ
1 occurrence rate, k and ratio of Type 1 error cost to Type 2 error cost
R
ðAFSÞ r¼1
ð16Þ affect total inspection and production costs, i.e. total cost. We solve
>
> k6B the problem in GAMS 23.5.2 (version 18.08.2010) with KNITRO sol-
>
>
: ver. The numerical examples are run on an Intel(R) Core(TM)2 Duo
kP0
2.67 GHz CPU PC, with 2 GB of memory. It takes approximately
Let AFSm be the mth solution to AFS, where m ¼ 1; 2; . . . ; M and 20 min to solve the problem for a production horizon consisting
^m be the optimal total cost and optimal control limit
let ^zm and k of 24 production periods, which correspond to a day’s production.
parameter to AFSm , then ^z ¼ minm¼1;2;...;M f^zm g is a point estimate The automobile company that we worked with in this project
for the optimal value z of the true total cost in FS problem, and randomly inspects 10 vehicles at the end of each hour. The average
^ ¼ arg min^m f^zm g, i.e. k
k ^ such that zðk
^ Þ ¼ minm¼1;2;...;M f^zm ðk
^m Þg, defect rate of the process p0 , is 0.3 and rp0 is 0.15. The values of the
k
 parameters involved in the model are as follows: painting cost of
is an estimated control limit parameter k of the true FS problem.
each vehicle c = 200 $, re-painting cost of a defective vehicle
It is shown that average of M problems of AFSm problems,
P cp = 600 $, inspection cost of vehicles c0 = 10 $, Type 1error cost
zM ¼ M1 M ^m is an estimated lower bound on the optimal aver-
m¼1 z c1 = 1500 $. Type 2 error cost is considered in six different levels
age total cost z (Kleywegt et al., 2001). (1500 $, 3000 $, 4500 $, 7500 $, 15,000 $, 30,000 $) in the experi-
An estimated upper bound on z is calculated by fixing any ments. Re-paining time of defective vehicles takes 4 h. Capacity
^ , the control
feasible solution to AFS. A usual practice is to fix k of the paint process is 80 vehicles per hour. At the end of each
limit parameter that satisfy minimum expected total cost out of day, 1200 painted vehicles should be ready to meet FA demand.
m solutions, to solve the SS problem given in Eqs. (11a)–(11p). The safety stock level allowed by the company is 10% of the daily
An estimated upper bound on minimum average total cost z demand. In the numerical study, we assume that the average
of the FS problem is calculated by solving the SS problem for R0 defect rate of the process is same for every color. If the defect rate
different scenarios of n where R0 > R as follows, ^zR0 ðk ^ Þ ¼ of the process highly depends on the color, it is also possible to
PR0 ^  r design new charts for different colors. We solve the two stage
1
R0 r¼1Q ðk ; n Þ.
stochastic model by SAA. The parameters used in SAA algorithm
The estimated gap for z is the difference between ^zR0 ðk ^ Þ and zM .
are: M ¼ 10; R ¼ 100 and R0 ¼ 200.
Note that the estimated gap could be negative due to sampling We model the magnitude of the shift, d as a random variable in
error since it is possible to obtain ^zR0 ðk ^ Þ < zM in some iterations our model, and it represents the percent change in the process
of the algorithm. Therefore, the estimated gap is set average defect rate when the process runs out-of-control:
^ Þ  zM ; 0g. The algorithm is stopped when the gap is suf-
maxf^zR0 ðk p1 ¼ p0 þ d. We assume that magnitude of shift d is normally dis-
ficiently small. The steps of the algorithm are detailed below: tributed with mean ld and variance r2d . Hence the new defect rate
of the process is p1 ¼ p0 þ drp0 . In our numerical study we consider
Step 1: Generate M independent samples, each of size R, i.e., effect of four different parameters on control limit parameter k,
ðnR1 ; . . . ; nRm Þ for each instance of AFSm . For each m ¼ 1; 2; . . . ; M, Type 1 and Type 2 error probability and total cost. These four
solve the AFSm problem. Let ^zm be the optimal objective value parameters are (i) mean of the shift ld , (ii) variance of the shift
(total cost) and k ^m be the optimal control limit parameter (FS r2d , (iii) occurrence of the special cause rate k and (iv) cost ratio
decision) for the mth problem. of Type 2 error cost to Type 1 error cost.
Step 2: Calculate the average total cost zM of M problems solved
P ^ 4.1. Effect of the mean shift on k and costs
in step 1: zM ¼ M1 M ^m
m¼1 z and find the minimum of solutions, k
which is the estimated optimal control limit parameter that
Table 1 shows the effect of the mean shift, ld . Mean shift ld , is
^ ¼ arg min^m f^zm g.
minimizes the average total cost, i.e., k k considered at five different levels and these levels are reported in
380 E.E. Gunay, U. Kula / Computers & Industrial Engineering 102 (2016) 374–382

Table 1 Fig. 2(a) shows that even though Type 1 and Type 2 error costs
Effect of mean shift on k, Type 1 and Type 2 probability. decrease, the total cost function increases due to the increase on
ld r2d k Cost k UCL Type 1 Type 2 average defect rate. The increase on average defect rate ld
ratio error a error b increases the re-painting and defective production costs which
1.00 0.03 0.05 2 2.110 0.617 0.015 0.866 cannot be compensated by the decrease in Type 1 and Type 2 error
1.40 0.03 0.05 2 1.960 0.594 0.023 0.697 costs.
1.60 0.03 0.05 2 1.660 0.549 0.060 0.620
1.80 0.03 0.05 2 1.880 0.582 0.030 0.504
2.20 0.03 0.05 2 1.881 0.582 0.030 0.343 4.2. Effect of the shift variance on k and the costs
2.30 0.03 0.05 2 1.889 0.583 0.030 0.339
2.40 0.03 0.05 2 1.901 0.585 0.030 0.323
In Table 2, we investigate the effect of the increase in variance
2.50 0.03 0.05 2 1.927 0.589 0.026 0.306
2.60 0.03 0.05 2 1.933 0.589 0.026 0.278 of the shift d. Table 2 shows that as shift variance r2d increases, vari-
ation in process defect rate also increases. Therefore lower k values
hence lower UCL, are determined to detect the special cause occur-
the first column of Table 1. In experiments in Table 1, we assume rences, which in turn decreases Type 2 error probability and
that r2d ¼ 0:03 and Type 2 to Type 1 error cost ratio is two. Table 1 increases Type 1 error probability.
shows mean shift of the random variable d increases k initially. As As Fig. 3(a) shows, total cost increases as the variance of the
mean of d keeps increasing k starts to decrease in order to balance shift increases as expected. Fig. 3(b) shows that as shift variance
out Type 1 and Type 2 error costs. r2d increases, Type 1 error cost is increased in order to minimize
Fig. 2(a) shows that total cost increases as the magnitude of the the Type 2 error cost.
average shift ld increases. The reason for this is that an increase on In experiments given in Table 3, we changed the mean and the
average defect rate increases the number of non-conforming units variance of the shift simultaneously to see their effect on chart
therefore more vehicles needed to be painted to meet the FA parameters and the total cost. Table 3 shows that as the mean
demand, which increases painting costs. The increase on number and the variance of the shift d increase, k decreases to 1.80 and
of vehicles painted increases the painting cost. Also, since all the the corresponding UCL decreases to 0.57. As the mean and the vari-
newly painted vehicles are inspected when an out-of-control signal ance of the shift continue to increase, higher k values are deter-
is reported, the increase on the total number of vehicles painted mined to decrease Type 1 error probability and its associated costs.
increase re-inspection cost. In addition to these two cost
components, the increase on the number of defects increases Table 2
re-painting cost. Effect of shift variance on k, Type 1 and Type 2 probability.
Fig. 2(b) shows that since Type 2 error cost is twice the Type 1
r2d ld k Cost k UCL Type 1 Type 2
error cost, the p-chart parameter k is optimized to give a priority to ratio error a error b
reduce Type 2 error cost over Type 1 error cost. Both Table 1 and
0.003 1 0.05 2 2.110 0.617 0.015 0.866
Fig. 2(b) shows that as ld increases to 1.6, Type 1 error cost 0.03 1 0.05 2 2.040 0.606 0.018 0.803
increases since UCL initially increases to reduce the k, Type 2 error 0.12 1 0.05 2 1.937 0.591 0.024 0.654
probability and its cost. As the mean shift ld continue to increase 0.24 1 0.05 2 1.890 0.584 0.030 0.500
above 1.6, both Type 1 and Type 2 error costs decrease since both 0.33 1 0.05 2 1.870 0.581 0.030 0.432

Type 1 and Type 2 error probabilities decrease.

(a) 405000
(a) 410000
400000
405000 395000
390000
Total Cost

400000
385000
Total Cost

395000 380000
375000
390000
370000
385000 365000
360000
380000
0.003 0.03 0.12 0.24 0.33
1 1.4 1.6 1.8 2.2 2.3 2.4 2.5 2.6
Shift variance
Mean shift
(b) 80000
(b)
65000
60000 70000
55000
50000 60000
Cost

45000
Cost

50000
40000
35000 40000
30000
25000 30000
20000
15000 20000
1 1.4 1.6 1.8 2.2 2.3 2.4 2.5 2.6 0.003 0.03 0.12 0.24 0.33
Mean shift Shift variance
Type1 cost Type2 cost Type1 cost Type2 cost

Fig. 2. Effect of mean shift on total cost, Type 1 and Type 2 cost. Fig. 3. Effect of shift variance of on total cost, Type 1 and Type 2 cost.
 E.E. Gunay, U. Kula / Computers & Industrial Engineering 102 (2016) 374–382 381

Table 3
Effect of changing mean and variance of shift on k, Type 1, Type 2 probability and total
(a) 580000
cost.
530000

Total Cost
ld r2d k Cost k UCL Type 1 Type 2 Total
ratio error a error b cost
480000
0.5 0.10 0.05 2 2.50 0.68 0.0007 0.900 394,205
0.7 0.15 0.05 2 2.27 0.64 0.0020 0.851 396,828
430000
0.9 0.20 0.05 2 2.07 0.61 0.0070 0.780 405,570
1.1 0.25 0.05 2 2.04 0.61 0.0075 0.730 408,755
1.3 0.30 0.05 2 2.02 0.60 0.0080 0.672 411,064 380000
1.5 0.35 0.05 2 1.99 0.60 0.0090 0.613 421,267 0.05 0.07 0.1 0.15 0.2
1.7 0.40 0.05 2 1.80 0.57 0.0200 0.606 426,146 Special cause rate
1.9 0.45 0.05 2 2.00 0.60 0.0090 0.590 430,574
2 0.50 0.05 2 2.11 0.62 0.0050 0.570 434,681 (b) 160000
2.1 0.55 0.05 2 2.27 0.64 0.0030 0.560 436,086
2.2 0.60 0.05 2 2.32 0.65 0.0020 0.550 439,305 140000
2.3 0.65 0.05 2 2.43 0.66 0.0010 0.530 440,343 120000

Cost
100000
80000
60000
40000
20000
0.05 0.07 0.1 0.15 0.2
440000
Special cause rate
Type1 cost Type2 cost
Total cost 420000
Fig. 5. Effect of special cause rate k on total cost, Type 1 and Type 2 cost.
0,6
400000
0,4
Variance of shift Table 5
0,5 0,2 Effect of changing the cost ratio of Type 2 error cost to Type 1 error cost.
1,0
1,5
2,0
Mean of shift Cost ld r2d k k UCL Type 1 Type 2
ratio error a error b
Fig. 4. Effect of changing mean and variance of shift on total cost.
1 1 0.03 0.05 2.350 0.652 0.001 0.918
2 1 0.03 0.05 2.110 0.617 0.015 0.866
3 1 0.03 0.05 2.090 0.614 0.018 0.860
Fig. 4 plots the change in total cost as the mean and the variance 5 1 0.03 0.05 1.990 0.599 0.023 0.838
of the shift increase: As mean and the variance of the shift increase, 10 1 0.03 0.05 1.931 0.590 0.026 0.727
20 1 0.03 0.05 1.629 0.544 0.072 0.642
the number of non-conforming vehicles and hence the number of
re-painted vehicles, therefore total painting and re-painting cost
increases.
(a)
800000
4.3. Effect of special cause rate k on k and the costs
700000
In experiments given in Table 4, we investigate the effect of spe-
Total Cost

cial cause rate k. Experiments in Table 4 show that as k increases, 600000

the control limit parameter k decreases. The reason for this
decrease in k can be explained as follows: As the probability of 500000
the process running out-of-control increases with k, the lower val-
ues of k must be set to detect the shift, hence to decrease Type 2 400000
error probability. For instance when k is 0.05, control limit param-
eter k is set to 2.11 and Type 2 error probability is 0.866. As k 300000
1 2 3 5 10 20
increases to 0.07, the control limit parameter k decreases to
1.816 and Type 2 error probability decreases to 0.76. Cost ratio
Fig. 5(a) shows that as k increases, total cost increases since the (b)
401700
number of times the process runs out-of-control state increases on
351700
average. Fig. 5(b) shows that as k increase both Type 1 and Type 2
301700
error probabilities increase.
251700
Cost

201700
Table 4
151700
Effect of special cause rate on k, Type 1 and Type 2 probability.
101700
k ld r2d Cost k UCL Type 1 Type 2
51700
ratio error a error b
1700
0.05 1 0.03 2 2.110 0.617 0.015 0.866 1 2 3 5 10 20
0.07 1 0.03 2 1.816 0.572 0.035 0.760
0.10 1 0.03 2 1.688 0.553 0.044 0.738
Cost ratio
0.15 1 0.03 2 1.560 0.534 0.060 0.710 Type1 cost Type2 cost
0.20 1 0.03 2 1.460 0.519 0.070 0.670
Fig. 6. Effect of cost ratio on total cost, Type 1 and Type 2 costs.
382 E.E. Gunay, U. Kula / Computers & Industrial Engineering 102 (2016) 374–382
4.4. Change on the cost ratio of Type 2 error cost to Type 1 error cost,
cost ratio
Table 5 shows how the ratio of Type 2 error cost to Type 1 error
cost affects the chart parameter k and the costs. As the Type 2
error cost increase, lower k values are selected to decrease Type
2 error cost. Since usually Type 2 error cost is higher than Type 1
error cost, our examples show that commonly used 3r control
limits are quite far from being optimal.
Exponential increase on cost ratio exponentially increase total
cost and Type 2 error cost as seen in Fig. 6(a) and (b).
5. Conclusion
In our study we present a two-stage stochastic model to
determine control chart parameter, upper control limit (UCL) and
production quantity of vehicles painted for each sampling period
Xt. The model is solved by sample average approximation (SAA)
algorithm. A numerical study is performed to investigate the
behavior of the model according to various parameters: mean of
the shift, variance of the shift, occurrence rate of the special cause,
cost ratio of Type 2 error to Type 1 error.
Our numerical study provides the following insights: (i) An
increase on the mean defect rate increases both the total cost
and the total production quantity. (ii) Effect of an increasing pro-
cess variance to the control limit parameter k is insignificant. (iii)
Frequency of special cause occurrences significantly affects the
total cost. (iv) All the experiments show that the commonly used
3r control limits in practice are wider than required, which
increases Type 1 error risk.
As a future study, determination of optimal sample size n and
sampling interval h along with parameter k may be an interesting
research direction. Another research direction is to study effect of
several types of special cause occurrences on the control chart
parameter k and the total quality and production related costs.
Also, since total cost and total production quantity are very sensi-
tive to the special cause occurrence rates, effect of different prob-
ability distributions of special cause occurrences may be explored.
References
Aslam, M., Azam, M., Khan, N., & Jun, C.-H. (2015). A mixed control chart to monitor
the process. International Journal of Production Research, 53(15), 4684–4693.
http://dx.doi.org/10.1080/00207543.2015.1031354.
Ben-Daya, M., & Rahim, M. A. (2000). Effect of maintenance on the economic design
of x-control chart. European Journal of Operational Research, 120(1), 131–143.
http://dx.doi.org/10.1016/S0377-2217(98)00379-8.
Bouslah, B., Gharbi, A., & Pellerin, R. (2013). Joint production and quality control of
unreliable batch manufacturing systems with rectifying inspection.
International Journal of Production Research (April), 1–15. http://dx.doi.org/
10.1080/00207543.2012.746481.
Bouslah, B., Gharbi, A., & Pellerin, R. (2015). Integrated production, sampling quality
control and maintenance of deteriorating production systems with AOQL
constraint. Omega. http://dx.doi.org/10.1016/j.omega.2015.07.012.
Chakraborty, T., Giri, B. C., & Chaudhuri, K. S. (2009). Production lot sizing with
process deterioration and machine breakdown under inspection schedule.
Omega, 37(2), 257–271. http://dx.doi.org/10.1016/j.omega.2006.12.001.
Chan, L.-Y., & Wu, S. (2009). Optimal design for inspection and maintenance policy
based on the CCC chart. Computers & Industrial Engineering, 57(3), 667–676.
http://dx.doi.org/10.1016/j.cie.2008.12.009.
Chiu, W. K. (1975a). Economic design of attribute control charts. Technometrics, 17
(1), 81–87. http://dx.doi.org/10.1080/00401706.1975.10489275.
Chiu, W. K. (1975b). Minimum cost control schemes using np charts. International
Journal of Production Research, 13(4), 341–349.
Chiu, W. K. (1976). Economic design of np charts for processes subject to a
multiplicity of assignable causes. Management Science, 23(4), 404–411.
Inghilleri, R., Lupo, T., & Passannanti, G. (2015). An effective double sampling
scheme for the c control chart. Quality and Reliability Engineering International,
31(2), 205–216. http://dx.doi.org/10.1002/qre.1572.
Kleywegt, A. J., Shapiro, A., & Homem-de-Mello, T. (2001). The sample average
approximation method for stochastic discrete optimization. SIAM Journal on
Optimization, 12(2), 479–502. http://dx.doi.org/10.1137/S1052623499363220.
Kooli, I., & Limam, M. (2011). Economic design of an attribute np control chart using
a variable sample size. Sequential Analysis, 30(2), 145–159. http://dx.doi.org/
10.1080/07474946.2011.563703.
Ladany, S. P. (1973). Optimal use of control charts for controlling current
production. Management Science, 19(7), 763–772. http://dx.doi.org/10.1287/
mnsc.19.7.763.
Lam, K. K., & Rahim, M. A. (2002). A sensitivity analysis of an integrated model for
joint determination of economic design of x-control charts, economic
production quantity and production run length for a deteriorating production
system. Quality and Reliability Engineering International, 18(4), 305–320. http://
dx.doi.org/10.1002/qre.463.
Lee, J. S., & Park, K. S. (1991). Joint determination of production cycle and inspection
intervals in a deteriorating production system. Journal of the Operational
Research Society, 42(9), 775–783. http://dx.doi.org/10.1057/jors.1991.148.
Lee, H. L., & Rosenblatt, M. J. (1987). Simultaneous determination of production
cycle and inspection schedules in a production systems. Management Science, 33
(9), 1125–1136.
Mehdi, R., Nidhal, R., & Anis, C. (2010). Integrated maintenance and control policy
based on quality control. Computers & Industrial Engineering, 58(3), 443–451.
http://dx.doi.org/10.1016/j.cie.2009.11.002.
Mehrafrooz, Z., & Noorossana, R. (2011). An integrated model based on statistical
process control and maintenance. Computers & Industrial Engineering, 61(4),
1245–1255. http://dx.doi.org/10.1016/j.cie.2011.07.017.
Montgomery, D. C., Heikes, R. G., & Mance, J. F. (1975). Economic design of fraction
defective control charts. Management Science, 21(11), 1272–1284. http://dx.doi.
org/10.1287/mnsc.21.11.1272.
Pan, E., Jin, Y., Wang, S., & Cang, T. (2012). An integrated EPQ model based on a
control chart for an imperfect production process. International Journal of
Production Research, 50(23), 6999–7011. http://dx.doi.org/10.1080/
00207543.2011.642822.
Pandey, D., Kulkarni, M. S., & Vrat, P. (2011). A methodology for joint optimization
for maintenance planning, process quality and production scheduling.
Computers & Industrial Engineering, 61(4), 1098–1106. http://dx.doi.org/
10.1016/j.cie.2011.06.023.
Patel, J. K., & Read, C. B. (1982). Handbook of the normal distribution. Marcel Dekker
(chapter 3).
Rahim, M. (1994). Joint determination of production quantity, inspection schedule
and control chart design. IIE Transactions, 26(6), 2–11.
Rahim, M. A., & Ben-Daya, M. (1998). A generalized economic model for joint
determination of production run, inspection schedule and control chart design.
International Journal of Production Research, 36(1), 277–289. http://dx.doi.org/
10.1080/002075498194047.
Wu, Z., & Luo, H. (2004). Optimal design of the adaptive sample size and sampling
interval np control chart. Quality and Reliability Engineering International, 20(6),
553–570. http://dx.doi.org/10.1002/qre.566.