ARTICLE IN PRESS

European Journal of Operational Research xxx (2005) xxx–xxx

www.elsevier.com/locate/dsw

Production, Manufacturing and Logistics

The value of setup cost reduction and process improvement

for the economic production quantity model with defects
Michael Freimer a, Douglas Thomas a,*
, John Tyworth a

a
509 Business Administration Building, Smeal College of Business Administration, The Pennsylvania State University,
University Park, PA 16802, USA

Received 19 January 2004; accepted 29 November 2004

Abstract

This paper investigates the effect of imperfect yield on economic production quantity decisions. The production sys-
tem is assumed to produce some time-varying proportion of defective parts which can be repaired at some unit cost. For
a general defect rate function, we develop results that characterize the optimal run length and expected total cost and
how these objects are affected by the cost parameters. Two kinds of investments in process improvements are consid-
ered: (i) reducing setup costs and (ii) improving process quality. We develop expressions for the marginal value of both
setup cost reduction and process improvement and discuss the relationship between these marginal values and problem
cost parameters. We show that any investment in setup cost reduction will result in a reduction in the number of defects
produced, and the total number of defects can increase or decrease with an investment in quality improvement.
Ó 2005 Elsevier B.V. All rights reserved.

Keywords: Manufacturing; Quality management; Economic production quantity

1. Introduction rates have known, deterministic values. The manu-

facturing process is assumed to be perfect; produc-
An enormous body of research has stemmed tion occurs without defects. One branch of the
from the original economic production quantity literature has accounted for imperfections in the
(EPQ) model. The basic model determines the manufacturing process. Much of this work has fo-
optimal lot size when demand and production cused on the benefits of smaller lot sizes in imper-
fect systems, and the relationship between quality
improvement (improving yield) and reducing setup
*
Corresponding author.
costs. Wright and Mehrez [17] provide a taxonomy
E-mail addresses: [email protected] (M. Freimer), dtho- and survey of the work relating inventory and
[email protected] (D. Thomas), [email protected] (J. Tyworth). quality issues.

0377-2217/$ - see front matter Ó 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.ejor.2004.11.024
 ARTICLE IN PRESS

2 M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx

Perhaps the first paper to relate quality and lot case it is stochastically increasing with the dura-
size was written by Porteus [11]. Porteus describes tion of the run (see [4,11,13]).
a system that begins each production run in con- Yano and Lee [18] also point out that expected
trol (i.e. producing only good units). As each unit cost is the dominant performance measure among
is produced, there is a probability p that the system models described by the literature. In this context,
goes out of control, at which time all subsequent our approach generalizes the three methods de-
units (until the end of the production run) are scribed above, as well as the case in which defects
defective. The time until the process goes out of are produced at a known rate which may vary over
control therefore follows a geometric distribution. time. That is, we allow any random yield function
Porteus used this model to study the optimal setup as long as all defective units are repaired instantly
investment in relation to reducing the probability p at some per unit cost. If all units are not repaired
of the process going out of control. Other instantly, then one must allow for the possibility
researchers have used this model to investigate of lost sales or backorders, or impose conditions
the economic benefits of reducing setup costs (see on the random yield function such that the effec-
[6,5]). More recently, Affisco et al. [1] investigated tive production rate always exceeds the demand
investment in quality improvement and setup cost rate.
reduction in a joint supplier–customer system with It should be noted that while our approach gen-
defects produced at a known constant rate. Rosen- eralizes several prior methods, some recent re-
blatt and Lee [14] present a model where the pro- search has addressed other aspects of the EPQ
cess goes out of control after some exponentially problem such as stochastic demand [16], coordina-
distributed time. Once out of control, the process tion between buyer and seller [3,7] and choosing an
produces units at a constant defect rate a. inspection schedule [8,12].
We extend prior work in this area by providing We focus on systems in which the process dete-
a general method for modeling production yields. riorates over time, although we point out where
Several papers (see [2,1]) model systems in which our results may be generalized (in the opposite
defects are produced at a known constant rate. direction) for a process that improves over time.
Others, such as Porteus, model the production of We develop properties of the optimal production
defects as a random process. Yano and Lee [18] quantity and optimal cost and consider the options
provide a representative survey of research involv- of investing in reducing setup cost and improving
ing lot sizing with random yields, together with a process quality. Expressions for the marginal value
useful description of the modeling issues involved. of setup cost reduction and process improvement
They list several methods for modeling random are also developed.
yield that are represented in the literature, which This paper is organized as follows. In Section 2
include: we present our model formulation and establish
results regarding the optimal solution. In Section
1. modeling defects as a Bernoulli process, where 3 we investigate the percentage cost penalty of
each unit is defective with probability p (see using the standard economic production quantity
[4,19]); even when the process is imperfect. In Sections 4
2. describing a distribution for the overall fraction and 5 we explore optimizing investment in setup
of defective units (see [4,15,9]); cost reduction and process improvement. Con-
3. specifying a distribution for the time in which cluding remarks are offered in Section 6.
the process remains in control (i.e. producing
good units), after which the process is out of
control (a fraction of the units are defective). 2. Model formulation and notation

In the first two methods, the fraction of defec- We consider the production system of a single
tive units is stochastically constant with respect item on a single machine which produces some
to the length of the production run. In the third (time-varying) fraction of defective items. Specifi-
 ARTICLE IN PRESS

M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx 3

cally, we assume that the fraction of defective after an exponential time T with mean b1 after
goods produced is described by u(x) at any time which defects are produced at the constant frac-
x, with 0 6 u(x) 6 1. As noted above, we focus tion a. Let N(h) be the number of defective units
on the situation in which the process deteriorates produced for a run length of h,
over time, u 0 (x) P 0. All defective units are re-
paired instantaneously at a per unit cost s. If repair N ðhÞ ¼ aR maxðh  T ; 0Þ; ð2Þ
takes some constant time then the analysis either Z h
remains the same if units are considered to count E½N ðhÞ ¼ aRðh  tÞbebt dt ð3Þ
in inventory when they are first produced, or the 0
inventory expression is slightly modified if units Z h  
are considered to count in inventory when the fi- ¼R a 1  ebt dt: ð4Þ
0
nal, non-defective unit is produced. In addition,
we define the following notation: This is equivalent to our model with
u(x) = a(1  ebx). As Yano and Lee point out,
R production rate per unit time, the geometric time to failure in PorteusÕ model
D demand rate per unit time, can be viewed as the discrete approximation to
h actual production run time, the Rosenblatt and Lee model.
h* optimal production run time, For notational convenience, define a = KD/R,
v unit production cost, b = h(R  D)/2, c = sD. With these substitutions
h0 non-financial holding cost per unit per
unit time, Z h
a c
i opportunity cost of capital per unit time, CðhÞ ¼ þ bh þ uðxÞ dx; ð5Þ
h h 0
h h0 + iv, holding cost per unit time,
K setup cost for each production run.  Z h 
0 a c
C ðhÞ ¼ 2 þ b þ 2 huðhÞ  uðxÞ dx : ð6Þ
h h 0
The separation of the per unit holding cost into
financial and non-financial pieces facilitates the Note that the first two terms of Eq. (5) form the
discussion of investment in setup cost and process standard EPQ cost function, which pffiffiffiffiffiffiffiffiis convex and
improvements in later sections. has a unique minimizer at b
h a=b with optimal
The total cost per unit time as a function of h is pffiffiffiffiffi
cost per unit time 2 ab. Our first result establishes
given by that with a deteriorating process, the optimal run
Z h
length is shorter than the run length suggested by
KD hðR  DÞh sD the classical EPQ. Furthermore, the faster a pro-
CðhÞ ¼ þ þ uðxÞ dx: ð1Þ
Rh 2 h 0 cess deteriorates, the shorter the optimal run
length. While this is intuitive, we show (a) that
Recall that our approach generalizes the three the result holds for our generalized model, (b)
methods from Yano and Lee mentioned above. how to obtain this run length and (c) that the cost
The first approach (modeling defects as a Bernoulli penalty for using the standard EPQ model can be
process with parameter p) corresponds to a con- significant.
stant defect rate u(x) = p. The second approach
(describing the constant fraction of defective items Proposition 1
with random variable Y) is also equivalent to the 1. There is a unique optimal run length h* that is
constant defect rate case (u(x) = E[Y]) since we less than (greater than) bh for defect rate u(x)
do not incorporate the variance of defective units increasing (decreasing) over time.
into our model. Finally, for the third method 2. The faster a process deteriorates (improves), the
(specifying a random time until the process goes shorter (longer) the optimal run length.
out of control) consider the model in Rosenblatt 3. The optimal cost is given by C(h*) = 2bh* +
and Lee [14] where the process goes out of control cu(h*).
 ARTICLE IN PRESS

4 M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx

Rh
Proof. Define f ðhÞ ¼ ðc=hÞ 0 uðxÞdx. This is the oh oh
1 þ 2bh þ ch u0 ðh Þ ¼ 0: ð12Þ
cost of repairing defective units per unit time. oa oa
When u(x) is non-decreasing, f(h) Ris non-decreas- Recall that the optimal cost is given by
h
ing sinceR f 0 ðhÞ ¼ ðc=h2 ÞðhuðhÞ  0 uðxÞdxÞ and
h Cðh Þ ¼ 2bh þ cuðh Þ; ð13Þ
huðhÞ P 0 uðxÞdx. Furthermore, since 0 6 u(x) 6 1,
0 6 f(h) 6 c. Now, the total cost function, C(h), so after differentiating and substituting (12),
is the sum of a convex function and a bounded, oC  oh oh 1
non-decreasing function, implying that there exists ¼ 2b þ cu0 ðh Þ ¼ : ð14Þ
oa oa oa h
a unique minimizer, h 6 b h.
Differentiating (11) with respect to b yields:
Similarly, when u(x) is non-increasing, f is
bounded and non-increasing, implying that C(h) oh oh
h2 þ 2bh þ ch u0 ðh Þ ¼ 0; ð15Þ
has a unique minimizer h P b h. ob ob
Next, consider two defect rate functions u1, u2 implying
with u01 ðxÞ P u02 ðxÞ P 0 for all x P 0. Define
u1 ðxÞ ¼ u1 ðxÞ  u1 ð0Þ with ~
~ u2 similarly defined. oC  oh oh
¼ 2h þ 2b þ cu0 ðh Þ ¼ h : ð16Þ
Let C1, C2 represent cost functions associated with ob ob ob
defect rate functions u1, u2 respectively. Differentiating (11) with respect to c yields:
" Z h #
 Z h  
a c  oh  
C 01 ðhÞ ¼ 2 þ b þ 2 h~u1 ðhÞ  u1 ðxÞ dx ;
~ 2bh þ h uðh Þ  uðxÞ dx
h h oc 0
0
ð7Þ oh
þ ch u0 ðh Þ
 Z h  oc
a c ¼ 0; ð17Þ
C 01 ðhÞ ¼ 2 þ b þ 2 h~u2 ðhÞ  u2 ðxÞ dx :
~
h h 0
implying
ð8Þ R h
oC  uðxÞ dx
Since u ~01 ðxÞ P ~
u02 ðxÞ P 0, C 01 ðhÞ P C 02 ðhÞ. Simi- ¼ 0  : ð18Þ
oc h
0
u02 ðxÞ 6 0, C 01 ðhÞ 6 C 02 ðhÞ.
u1 ðxÞ 6 ~
larly, if ~
Finally, if the first order condition is satisfied at Rearranging terms in Eq. (12):

h*, setting (6) equal to zero yields: oh 1
¼ P 0: ð19Þ
Z h oa 2bh þ ch u0 ðh Þ
c a
 uðxÞ dx ¼ bh þ cuðh Þ   ; ð9Þ Similarly, rearranging terms in Eqs. (15) and
h 0 h
(17) establishes:
Cðh Þ ¼ 2bh þ cuðh Þ:  ð10Þ
oh h2
¼ 60 ð20Þ
Now we derive expressions for the changes in ob 2bh þ ch u0 ðh Þ


optimal run length h* and optimal cost and

C* C(h*) with respect to the cost parameters h i
 
R h
a, b and c. These expressions will be useful in oh  h uðh Þ  0 uðxÞ dx
establishing later results. The first order condition ¼ 6 0: ð21Þ
oc 2bh þ ch u0 ðh Þ
will be satisfied at h*, implying
" Z #
h
2  
a þ bh þ c h uðh Þ  uðxÞ dx ¼ 0: ð11Þ 3. Cost penalty of using the EPQ
0

Differentiating this expression with respect to a Despite the strong underlying assumptions, the
gives: EPQ can often provide a reasonable solution for
 ARTICLE IN PRESS

M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx 5

Z
more complicated settings (see [16]). We now exam- a c h
ine the percentage cost penalty of using the EPQ CðhÞ ¼ þ bh þ ða þ bxÞ dx ð30Þ
h h 0
when the production process is imperfect. Define a
pffiffiffiffiffi c Z bh ¼ þ ðb þ cb=2Þh þ ca;
h
ð31Þ
b b
C Cð hÞ ¼ 2 ab þ uðxÞ dx ð22Þ a
b
h 0 0
C ðhÞ ¼ 2 þ ðb þ cb=2Þ; ð32Þ
h
and
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
C  Cðh Þ ¼ 2bh þ cuðh Þ: ð23Þ a
h ¼ ; ð33Þ
b þ cb=2
The percentage cost penalty if using the EPQ
rather than the optimal quantity is given by and
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b  C
C cb

D : ð24Þ Cðh Þ ¼ 2 aðb þ Þ þ ca: ð34Þ
C 2
pffiffiffiffiffiffiffiffi
Analyzing D for general u(x) appears to be dif- b
Since h ¼ a=b,
ficult. We first present general expressions for the rffiffiffi
pffiffiffiffiffi cb a
change in cost at EPQ and then analyze the special b
Cð hÞ ¼ 2 ab þ þ ca ð35Þ
case where the production process deteriorates 2 b
linearly: and
rffiffiffi 2 3 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z bh pffiffiffiffiffi pffiffi
oCb b c ob h 4b b 2 ab þ cb2 ab  2 aðb þ cb2 Þ
¼ þ 2 huð hÞ  uðxÞ dx5 ð25Þ D¼ pffiffiffiffiffi pffiffi : ð36Þ
oa a b h oa 0 2 ab þ cb2 ab þ ca
2 0 13
Z bh
14 c @b b The analysis of the percentage cost penalty is
¼ 1 þ pffiffiffiffiffi huð hÞ  uðxÞ dxA5; complicated by the constant term ca. We momen-
b
h 2b
h ab 0
tarily drop this constant term by defining
ð26Þ b0 C
C b  ca, C  C   ca.
2 3 0
rffiffiffi Z bh
b
oC b
a c o h 4b b Proposition 2. For the linear deterioration case
¼ þ 2 huð hÞ  uðxÞ dx5 ð27Þ
ob with no constant term a (u(x) = bx), the percentage
b b h ob 0
2 0 13 cost penalty of using the EPQ is: (1) not affected by
Z bh the ordering cost K, (2) decreasing in the holding
c
¼bh 41  2
@bhuðbhÞ  uðxÞ dxA5; ð28Þ cost h and (3) increasing in the repair cost s.
b
2b h 0

and Proof.
R bh b0
C
b
oC uðxÞ dx D0 1
¼ 0
: ð29Þ C 0
oc b
h sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b b b cb 1
Note that o C=oc P oC  =oc, and that o C=ob ¼ þ  1: ð37Þ
b þ cb=2 4 bðb þ cb=2Þ
can be negative. That is, if a firm is going to use
the EPQ, overall cost could be reduced by increas- Part (1) is obvious from inspection of D0.
ing the holding cost h.
oD0 c2 b2
¼ ½bðb þ cb=2Þ3=2 6 0; ð38Þ
3.1. Linearly increasing defect rate ob 16

For illustrative purposes, we consider the spe-

cial case of a process that deteriorates linearly, oD0 bcb2 3=2
¼ ½bðb þ cb=2Þ P 0:  ð39Þ
u(x) = a + bx with a, b P 0. From (5) we have: oc 16
 ARTICLE IN PRESS

6 M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx

Proposition 3. For the linear deterioration case, the Differentiating (35),

percentage cost penalty of using the EPQ is: (1) pffiffiffi pffiffiffi
b 2 b þ cb=ð2 bÞ b þ cb=4
oC
increasing in the ordering cost K, (2) decreasing in ¼ pffiffiffi ¼ pffiffiffiffiffi : ð45Þ
the holding cost h (3) increasing in the repair cost oa 2 a ab
s and (4) increasing in the rate of deterioration b. Differentiating (34),
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Proof. If we differentiate D with respect to any one oC  b þ cb=2
¼ ð46Þ
of the three cost parameters, we have oa a
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
bðb þ cb=2Þ
b 0  ðC
ðC 0 þ caÞ C b 0 þ caÞC 0 ¼ ð47Þ
D0 ¼ 0 0
ð40Þ ab
2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðC 0 þ caÞ 2 2 2
0 b þ bcb=2 þ ðcb=4Þ  ðcb=4Þ
b C
C 0 C b 0  C 0 Þ
b 0 C 0 þ cað C ¼ ð48Þ
¼ 0 0 0 0
: ð41Þ ab
ðC 0 þ caÞ
2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðb þ cb=4Þ2  ðcb=4Þ2 o C b
Note that the denominator is always positive, ¼ 6 : ð49Þ
ab oa
and we are only interested in the sign of this
expression. From the proof of Proposition 2, the Similarly,
sum of the first two terms in the numerator is zero,
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
negative and positive for derivatives with respect oC  a
¼ ð50Þ
to a, b and c respectively. It is therefore sufficient ob b þ cb=2
to show
and
oCb0 oC 
P 0; ð42Þ rffiffiffi 
oa oa b
oC a cb
¼ 1 : ð51Þ
ob b 4b
oCb0 oC 0
; ð43Þ  b
ob
6
ob Note that when c = 0, oC ob
¼ oobC . Furthermore,
pffiffiffi  3=2
and o2 C 0 b a 1
¼ ð52Þ
oboc 4 b þ cb=2
oCb0 oC 
P 0 ð44Þ and
oc oc pffiffiffi  3=2
b0
o2 C b a 1
to establish our first three results. ¼ ; ð53Þ
oboc 4 b

Table 1
Production quantities, costs and percentage defective units at both the optimal run length and the run length implied by EPQ
Optimal values At EPQ D (%)
Rh* C* % Defective Rb
h b
C % Defective
Base case 242 $1077 2.01 387 $1170 2.61 8.66
K = 90 230 $1034 1.96 367 $1123 2.53 8.55
K = 110 254 $1117 2.06 406 $1215 2.69 8.75
h=7 248 $1056 2.03 414 $1164 2.73 10.24
h=9 236 $1097 1.98 365 $1178 2.52 7.43
s = 22 251 $1016 2.05 387 $1091 2.61 7.44
s = 28 234 $1136 1.97 387 $1248 2.61 9.85
 ARTICLE IN PRESS

M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx 7

establishing From Eq. (14) and the definition a = KD/R, the

b 0 oC  marginal change in optimal cost cost as a function
oC
6 0 ð54Þ of K is
ob ob
oC  oC  oa D
as desired. ¼ ¼ : ð58Þ
oK oa oK Rh
Next,
Eq. (58) together with Eqs. (19)–(21) and Prop-
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
oC  b a osition 1-(2) immediately establish the following
¼aþ ð55Þ results.
oc 2 b þ cb=2
Proposition 4. For a deteriorating process, the
and marginal value of setup cost reduction is higher for
rffiffiffi 1. Smaller setup cost K;
b
oC b a
¼aþ : ð56Þ 2. Larger holding cost h;
oc 2 b
3. Larger repair cost s;
Finally, to establish part (4), 4. Faster deteriorating process (larger u 0 (x)).
pffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffi pffiffiffi pffiffiffiffiffi
oD c2 ab 2a 2b þ cb þ 2b a  2a 2b
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffi pffiffiffi pffiffiffi2 ;
ob 2b þ cb 4 ab þ cb a þ 2ca b Note also, that the absolute number of defective
units produced is increasing in K, so any invest-
ð57Þ ment in setup cost reduction will decrease the
number of defective units.
which is greater
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi pthan
ffiffiffiffiffi or equal to zero since
2a 2b þ cb P 2a 2b. h
4.1. Linear deterioration
Example 1. Consider the numerical example
Following Porteus [10], we consider a logarith-
given in [10]: K = 100, v = 50, h0 = 0.5, i = 0.15
mic investment function, MK(K) = A  B ln (K).
and D = 1000. We assume R = 1200, s = 25 and
That is, an investment of MK(K) is required to re-
u(x) = 0.01 + 0.1x.
duce the setup cost to K from the current level K0.
The quantity given by the standard EPQ is
As described in [10], one could fit this functional
Rbh ¼ 387:3, resulting in a cost of $1170. The
form by estimating the investment x, required to
optimal quantity is Rh* = 241.9, resulting in a
reduce setup cost from K0 to a specific K0(1  d)
cost of $1077, a percentage difference of 8.6%.
and using the fact that MK(K0) = 0. It is easy to
Table 1 shows production quantities, costs, per-
verify that B = x/ln (1  d) and A = B ln (K0).
centage of defective units and the cost penalty for
Since we deal with the undiscounted model, an
this base case and for cases with small changes in
opportunity cost of iMK(K) is charged per unit
the cost parameters K, h and s.
time. With logarithmic investment function MK
and linear deterioration rate u(x) = a + bx the to-
tal cost function is
4. Optimization of setup cost reduction
gðKÞ ¼ iðA  B lnðKÞÞ
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
In this section we consider investing in reduc- 2KDðhðR  DÞ þ sbDÞ
ing the setup cost. First, we develop an expression þ þ asD ð59Þ
R
for the marginal value of setup cost reduction for
general u(x). We analyze how the cost parameters from Eq. (34) and the definitions of a, b and c. Set-
K, h and s affect this marginal value. We then ting the first derivative,
consider the linear deterioration case and a spe- rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cific functional form for investment in setup cost 0 iB DðhðR  DÞ þ sbDÞ
g ðKÞ ¼  þ ; ð60Þ
reduction. K 2KR
 ARTICLE IN PRESS

8 M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx

equal to zero yields in the defect rate u(x) such that the new process
2
produces defects at a rate ~uðxÞ ¼ uðxÞ  , and (2)
2ðiBÞ R a scale factor p such that the new defect rate is
K ¼ : ð61Þ
DðhðR  DÞ þ bsDÞ ~uðxÞ ¼ puðxÞ.
We first develop expressions for the marginal
It is easily verified that g 0 (K) < ( > )0 for values of each kind of process improvement for
K < ( > )K*, implying that g is pseudo-convex, general u(x), analyzing how the cost parameters
thus K* is a unique cost minimizer. Note that un- K, h and s affect these marginal values. We then
less one wants to permit positive compensation for consider the linear deterioration case and a specific
increasing setup time, K* > K0 would imply the functional form for investment in process
optimal investment level is zero. improvement.
Since h* is increasing in K, the marginal value
of setup cost reduction (58) is decreasing in K in
the general case as well. This implies that for any Proposition 5
investment function that has marginally increasing 1. Let  be a constant shift in u(x) such that the new
cost of setup cost reduction, there will be a unique defect rate function is ~uðxÞ ¼ uðxÞ  . (i) The
solution. optimal run length h* is not affected by the shift,
and (ii) the rate of change in optimal cost as a
Example 2. Consider the data from Example 1, function of the (downward) shift  is sD.
now with setup cost reduction function A  2. Let 0 6 p 6 1 define a scaling of u(x) such that
B ln (K). Suppose a 10% setup cost reduction would the new defect rate function is ~uðxÞ ¼ puðxÞ. (i)
cost $200, implying A = 8742 and B = 1898. The The optimal run length h* is decreasing in p
optimal setup cost is K* = 47 with Rh* = 267. (the optimal run length gets longer as the process
Table 2 shows production quantities, costs and is improved), and (ii) the rate of change in optimal
percentage of defective units for the original cost as a function of p is given by the defect repair
optimal solution (K = K0) and when setup cost cost per unit time:
reduction is considered. Cases with small changes R h
in the cost parameters h and s are also shown. oC  sD 0 uðxÞ dx
¼ : ð62Þ
op h
3. For a deteriorating process, the marginal value of
5. Optimization of process quality improvement process improvement (by scaling) is larger for
greater K and smaller h.
In this section we consider investing in improv- 4. The marginal value can increase or decrease with
ing process quality. We examine two kinds of pro- s, and the total number of defects can increase or
cess improvement: (1) a constant downward shift  decrease with p.

Table 2
Optimal production quantities, costs and percentage defective units with and without investment in setup cost reduction
Original values With setup cost reduction
Rh* C* % Defective K* Rh* C* % Defective
Base case 242 $1077 2.01 47 167 $1032 1.69
h=7 248 $1056 2.03 50 175 $1017 1.73
h=9 236 $1097 1.98 45 159 $1045 1.66
s = 22 251 $1016 2.05 51 180 $980 1.75
s = 28 234 $1136 1.97 44 155 $1082 1.65
 ARTICLE IN PRESS

M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx 9

Proof. With a new defect rate u ~ðxÞ ¼ uðxÞ  , the will become t, and the defect repair cost will drop
original cost function (1) can be written as to zero.
Z Now, note that the total number of defects is
KD hðR  DÞh sD h
CðhÞ ¼ þ þ uðxÞ dx  sD: equal to oC*/op divided by the constant s. With a
Rh 2 h 0
constant defect rate, it is clear that the total number
ð63Þ of defects is increasing in p. Now, consider a process
This final term is a constant with respect to h, so that produces no defects for 0 6 x 6 t and has
h* is clearly not affected by . Also, oC/o is clearly constant defect rate pa thereafter. For very large s,
the constant sD. moving p from 0 to a positive value will shift the run
Next, to show part two of the proposition, length from b h to t, decreasing total defects. h
consider the scaling ~uðxÞ ¼ puðxÞ. One can re-write
the original total cost function (1) as 5.1. Linear deterioration
Z h
KD hðR  DÞh psD We consider the case where process improve-
CðhÞ ¼ þ þ uðxÞ dx: ð64Þ
Rh 2 h 0 ment scales u(x) with some constant p. We use a
From Eq. (21) it is clear that Rh* is decreasing in similar logarithmic function /(p) = B ln (p).
h
p. The expression oC  =op ¼ ðsD 0 uðxÞ dxÞ=h fol- There is no constant term A in /(p) since we as-
lows from Eq. (18). sume that keeping the current defect rate (p = 1)
To show part 3, note that when the defect rate is requires no investment, /(1) = B ln (1) = 0.
non-decreasing (u 0 (x) P 0), the defect repair cost Again, one could fit this functional form by esti-
at the optimal solution (which is also equal to oC*/ mating the investment required to scale down the
op) is increasing in h* when s and D are constant. defect rate by a specific p. It is worth noting that gi-
From Eqs. (19) and (20) it is clear that increasing ven this logarithmic investment form, the percent-
K or decreasing h will increase h*, thus increasing o age of defective units at the optimal value for p
C*/op. does not depend on K and h. We seek to minimize
Finally, to establish part 4, consider a process wðpÞ ¼ C  ðpÞ þ /ðpÞ: ð65Þ
with a constant defect rate a. The defect repair cost
Differentiating,
at any run length is the constant asD which is
clearly increasing in s. Next, consider a process oC  ðpÞ iB
w0 ðpÞ ¼  ð66Þ
that is defect free for some very short initial op p
interval, u(x) = 0, 0 6 x 6 t, with a constant R h
sD 0 uðxÞ dx iB
defect rate thereafter. For small enough s, the ¼  : ð67Þ
h p
optimal run length will be greater than t and oC*/
op > 0. As s approaches 1, the optimal run length Setting w 0 (p*) = 0 implies

Table 3
Optimal production quantities, costs and percentage defective units with and without investment in process improvement
Original values With process improvement
Rh* C* % Defective p* Rh* C* % Defective
Base case 242 $1077 2.01 0.52 288 $1011 1.14
K = 90 230 $1034 1.96 0.53 271 $976 1.14
K = 110 254 $1117 2.06 0.50 304 $1045 1.14
h=7 248 $1056 2.03 0.51 300 $987 1.14
h=9 236 $1097 1.98 0.53 277 $1035 1.14
s = 22 251 $1016 2.05 0.59 288 $975 1.29
s = 28 234 $1136 1.97 0.46 288 $1044 1.02
 ARTICLE IN PRESS

10 M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx

R h
p uðxÞ dx iB cost reduction will result in a reduction in the
0
¼ : ð68Þ number of defects produced. Interestingly, the to-
h sD
tal number of defects can increase or decrease with
For example, reducing K would reduce the opti- the scaling p. For the logarithmic investment func-
mal run length, h*, but the optimal p* would be tion investigated, the number of defects at the opti-
higher such that the number of defective units pro- mal run length is unaffected by the holding cost h
duced remained constant. This is illustrated in Ta- and the setup cost K. These results may provide
ble 3. guidance for manufacturing managers regarding
With linear deterioration rate u(x) = a + bx, the relative value of setup cost reduction and pro-
finding the optimal p involves finding the appro- cess improvement investments.
priate root of a cubic equation. As such, we do
not present the algebra.
Example 3. Consider the data from Example 1 References
now with process improvement investment func-
tion /(p) = B ln (p). Suppose a 10% reduction in [1] J.F. Affisco, M.J. Paknejad, F. Nasri, Quality improve-
defect rate would cost $200, implying B = 1898. ment and setup reduction in the joint economic lot size
model, European Journal of Operational Research 142
Table 3 shows production quantities, costs and
(2002) 497–508.
percentage of defective units for the original opti- [2] T.C.E. Cheng, EPQ with process capability and quality
mal solution (K = K0) and when process improve- assurance considerations, The Journal of the Operational
ment is considered. Cases with small changes in the Research Society 42 (8) (1991) 713–720.
cost parameters K, h and s are also shown. [3] S.K. Goyal, V.P. Gupta, Integrated inventory models: The
buyer–vendor coordination, European Journal of Opera-
tional Research 41 (1989) 261–269.
[4] M. Henig, Y. Gerchak, A flexible conceptualization of
random yield and its implications for source selection,
6. Conclusions
Proceedings of the First Conference of the ORSA Technical
Section of Manufacturing Management (1990) 133–139.
In this paper we consider the classical economic [5] J.D. Hong, J.C. Hayya, Joint investment in quality
production quantity model with defects produced improvement and setup reduction, Computers and Oper-
according to some time-varying function u(x). ations Research 2 (1995) 567–574.
[6] S.L. Kim, J.C. Hayya, J.D. Hong, Setup reduction in the
Our results hold for any random yield function
economic production quantity model, Decision Sciences 23
as long as units are repaired instantly. With a dete- (1992) 500–508.
riorating process, the optimal run length is shorter [7] H.L. Lee, M.J. Rosenblatt, A generalized quantity dis-
than the run length implied by the EPQ. Further- count pricing model to increase supplierÕs profits, Man-
more, the faster the process deteriorates the short- agement Science 32 (9) (1986) 1144–1185.
[8] H.L. Lee, M.J. Rosenblatt, Simultaneous determination of
er the optimal run length. For the special case of
production cycles and inspection schedules in a production
linear deterioration, the cost penalty for using system, Management Science 33 (1987) 1125–1136.
the EPQ instead of the optimal quantity is increas- [9] K.L. Mak, Inventory control of defective product when the
ing in the setup cost K, the defect repair cost s and demand is partially captive, International Journal of
the rate of process deterioration b, and decreasing Production Research 23 (3) (1985) 533–542.
[10] E. Porteus, Investing in reduced setups in the EOQ model,
in the holding cost h.
Management Science 31 (8) (1985) 998–1010.
We consider the opportunity to invest in reduc- [11] E. Porteus, Optimal lot sizing, process quality improve-
ing setup cost and improving process quality. For ment and setup cost reduction, Operations Research 34
a general time varying defect rate u(x), the mar- (1986) 137–144.
ginal value of setup cost reduction is inversely pro- [12] M.A. Rahim, Joint determination of production quantity
inspection scheduling and control chart design, IIE Trans-
portional to optimal run length while the marginal
actions 26 (6) (1994) 2–11.
value of process improvement (as modeled by a [13] M.J. Rosenblatt, H.L. Lee, A comparative study of
constant scaling p of the defect rate) is increasing continuous and periodic inspection policies in deteriorating
in the optimal run length. Any investment in setup production systems, IIE Transactions 18 (1) (1986) 2–9.
 ARTICLE IN PRESS

M. Freimer et al. / European Journal of Operational Research xxx (2005) xxx–xxx 11

[14] M.J. Rosenblatt, H.L. Lee, Economic production cycles [17] C.M. Wright, A. Mehrez, An overview of representative
with imperfect production processes, IIE Transactions 18 research of the relationships between inventory and qual-
(1) (1986) 48–55. ity, OMEGA 26 (1) (1998) 29–47.
[15] W. Shih, Optimal inventory policies when stockouts result [18] C.A. Yano, H.L. Lee, Lot sizing with random yields: A
from defective products, International Journal of Produc- review, Operations Research 43 (2) (1995) 311–334.
tion Research 18 (6) (1980) 677–685. [19] X. Zhang, Y. Gerchak, Joint lot sizing and inspection
[16] E.A. Silver, R. Peterson, D.F. Pyke, Inventory Manage- policy in an EOQ model with variable yield, IIE Transac-
ment and Production Planning and Scheduling, third ed., tions 22 (1) (1990) 41–47.
John Wiley & Sons, 1998.