Applied Mathematical Modelling 37 (2013) 3340–3354

Contents lists available at SciVerse ScienceDirect

Applied Mathematical Modelling

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

An economic production quantity model with a positive resetup point

under random demand
Shine-Der Lee ⇑, Chin-Ming Yang
Department of Industrial and Information Management, National Cheng Kung University, 1 University Road, Tainan 70101, Taiwan, ROC

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

Article history: In this paper, an extended economic production quantity (EPQ) model is investigated,
Received 31 March 2011 where demand follows a random process. This study is motivated by an industrial case
Received in revised form 15 June 2012 for precision machine assembly in the machinery industry. Both a positive resetup point
Accepted 3 July 2012
s and a fixed lot size Q are implemented in this production control policy. To cope with ran-
Available online 16 August 2012
dom demand, a resetup point, i.e., the lowest inventory level to start the production, is
adapted to minimize stock shortage during the replenishment cycle. The considered cost
Keywords:
includes setup cost, inventory carrying cost, and shortage cost, where shortage may occur
Lot sizing
Economic production quantity
at the production stage and/or at the end of one replenishment cycle. Under some mild
Resetup point conditions, the expected cost per unit time can be shown to be convex with respect to deci-
Safety stock sion parameters s and Q. Further computational study has demonstrated that the proposed
Random demand model outperforms the classical EPQ when demand is random. In particular, a positive
resetup point contributes to a significant portion of this cost savings when compared with
that in the classical lot sizing policy.
Ó 2012 Elsevier Inc. All rights reserved.

1. Introduction

Economic production quantity (EPQ) model is an extension of the well-known economic order quantity (EOQ) model
where demand is assumed to be constant, e.g., see Hadley and Whitin [1]. Each replenishment cycle in EPQ consists of
two stages: the duration to produce a fixed lot size where production rate is continuous and finite, and the duration to de-
plete remaining inventory before a new batch is launched. When the remaining inventory drops to zero, a new replenish-
ment cycle will resume. An analytical model is developed to determine the optimal EPQ so that the sum of setup cost
and inventory carrying cost is minimized. In practice, demand from customers can be dynamic or random. Donaldson [2]
established an optimal replenishment schedule when demand is a linear function. Later, Deb and Chaudhuri [3] extended
this work to allow for stock shortage. Hariga [4] developed a heuristic to find a lot sizing policy with linear demand when
shortage is not permitted. An optimal replenishment schedule with linear or exponential time-varying demand has been
studied by Hariga and Benkherouf [5], where an iterative numerical procedure is developed to find optimal schedule. The
demand has been treated as deterministic with various functions of time in [2–5].
Economic order quantity models with imperfect quality and quantity discount can be found in Lin [6], where both vendor
and buyer are considered. Chang [7] derived the exact expression for inventory carrying cost to improve this model, where
simple and close-form formulas for computing the optimal order quantity are found. The optimal lot size and the maximum
backorder level in an EOQ/EPQ model with fixed and linear backorder cost are studied in Cárdenas-Barrón [8]. Later, Cárd-
enas-Barrón [9] used a simple formula to determine the optimal number of shipments in Chang’s [7] model. Sufficient and

⇑ Corresponding author. Tel.: +886 6 275 7575x53146; fax: +886 6 236 2162.
E-mail address: [email protected] (S.-D. Lee).

0307-904X/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.apm.2012.07.008
 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354 3341

necessary conditions of the optimal replenishment policy and backorder level can be found in Chung and Cárdenas-Barrón
[10]. The above studies have focused on the EOQ/EPQ policies for deterministic and constant demand process.
When faced with the random demand, inventory control policies with a positive reorder or resetup point s are the most
often studied and adopted. For examples, the well known (reorder point s, fixed lot size Q Þ and (reorder point s, order-up-to-
level S) policies, where the reorder point provides safety stock to cope with demand uncertainty. It has been shown that the
control policies with reorder point outperform the classical EOQ or EPQ policy, where the reorder point for procurement or
resetup point for production is s = 0, e.g., see Hadley and Whitin [1]. Rabinowitz et al. [11] studied an economic order quan-
tity model when demand follows a Poisson distribution and shortage is treated as partial backorder. Namit and Chen [12]
analyzed an (s; Q Þ inventory system when demand is Gamma distributed. Simulation is used to find optimal replenishment
quantity that minimizes expected cost per unit time.
In the EPQ context, Gavish and Graves [13] considered a production lot sizing problem with Poisson demand, where a
(resetup point s, production up-to-level S) policy has been studied. The production continues until stock level reaches S;
and production is resumed when stock level falls to or below s in this control policy. Note that the lot size varies from
one replenishment cycle (current batch) to the next cycle. More recently, Rempala [14] studied a production lot sizing policy
for a compound Poisson demand process, where partial backorder and bi-level inventory control are considered. These anal-
yses have focused on theoretical aspects of the control policy. There are only a few papers that incorporate both a resetup
point and a fixed lot size in the production control policies, e.g., see Groenevelt et al. [15] and Palar [16], due to the intrac-
tability of the analytical models and computational difficulty to find the optimal policies.
Another stream of production policies has focused on the effects of imperfect production process. Rosenblatt and Lee [17]
first developed an economic production cycle model that deals with deteriorating production process and defective items. An
EPQ model in two-stage production system is analyzed in Lee and Rong [18], where both mean time to failure and mean time
to repair are random variables. Later, Goyal and Giri [19] studied an optimal lot sizing model for a deteriorating item with
time-varying demand. The optimal production cycle time for an unreliable production system with rework is considered in
Chiu et al. [20]. Konstantaras et al. [21] developed an adapted lot sizing model when the learning effect on imperfect quality
of the production process and shortage are considered. These studies have focused on production policies for deterministic
and known demand process.
This research was motivated by a case study for the precision machine assembly of an international supplier. These pre-
cision machines are used to, for an example, fabricate turbine blades for the aircraft or parts for high-precision products. The
setup cost is extremely high in the assembly system. An order from customers usually consists of one unit for a specific mod-
el of machine; and average annual orders can be reasonable estimated. The time between orders, however, is unpredictable.
The shortage cost is difficult to quantify and shortage is not allowed for this industry. Due to random demand, a shortage
may occur during the production period and/or at the end of the replenishment cycle. For a shortage that occurred during
the production period, it is backordered and immediately filled with the finished product from the assembly line. This is de-
fined as a transient shortage. For a shortage occurred at the end of the replenishment cycle, it will be backordered and sat-
isfied with new stock. The later is defined as a cycle shortage. In addition, due to high production cost, fierce competition,
and rapid market change, both excessive inventory and stock shortage are to be avoided. This production lot sizing problem
is modeled and analyzed with a control policy that includes a positive resetup point s and a fixed lot size EPQ, i.e., (s; Q Þ
policy.
In the following sections, we first formulate this extended lot sizing problem when a production system faces random
demand, then derive and analyze the expected cost model. Section 3 gives an efficient search procedure to find the optimal
resetup point, the production lot size, and its minimal cost value, based on the analytic results. Numerical experiments are
conducted to gain understanding of the extended production policy; and some parametric analyses are reported. A brief con-
cluding summary is given in the last section.

2. Problem description and model formulation

An extended production lot sizing policy with a positive resetup point s and a fixed lot size Q is formulated and analyzed
in this section. The development of the analytical model is a culmination of a case study for the assembly of precision ma-
chines. Most of the times, the order received from customers is one unit, due to high procurement cost of machines. Time
between orders exhibits no pattern, and is infrequent and difficult to observe, although the average orders per year can
be reasonably estimated. Due to the uniqueness of precision machining industry, shortage is prohibited. A resetup point s
is used to cope with such a stock shortage. A fixed production quantity Q is also implemented in the system, which serves
as the most commonly found policy for operation and scheduling in production systems, enabling the minimization of work-
load fluctuation in production lines and the simplification of planning and control.
We consider a production system that processes one item at a time with constant production rate l. Demand of custom-
ers follows a stationary Poisson process with rate k, that is, PfQ d ð0; tÞ ¼ kg ¼ ekt ðktÞk =k!; k ¼ 0; 1; . . ., where l > k > 0, and
Q d ð0; tÞ denotes total demand from time 0 to t. Each replenishment cycle can be partitioned into two stages: duration of pro-
duction period T 1 ¼ ½0; Q =lÞ, where lot size Q is processed with constant production rate l, and duration of depletion period
T 2 ¼ ½Q =l; Þ, where remaining inventory at time Q =l is used to meet future demand. A replenishment cycle starts when the
inventory level falls to or below s and the production system is idle, where Ii ðtÞ; i ¼ 1; 2 denotes the inventory level at time t
3342 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354

during the production stage and the depletion stage, respectively. During T 1 , products are produced one by one to satisfy
demand, and inventory level is expected to increase at mean rate l  k. At any time t; 0 6 t < Q =l, the inventory level
I1 ðtÞ is the sum of resetup point s and the total production Q p ð0; tÞ minus the total demand Q d ð0; tÞ, where Q p ð0; tÞ denotes
total production from time 0 to t. Production continues until the lot size Q has been completed. Unlike classical EPQ policy,
shortage may occur in the interval ½0; Q =lÞ when an order is received and there is no on-hand stock to meet this demand
immediately. The occurrence of such a shortage is intermittent, since new stock is completed with rate l, which can be used
to fill the order when it is available. Such incident is called a transient shortage and the number of transient shortage at any
time t is defined as NI1 ðtÞ; 0 6 t < Q =l.
When the lot size Q is completed, the depletion period starts, and remaining inventory is used to satisfy future demand. If
on-hand inventory at time Q =l is greater than s, the production system is switched to standby mode or to process other
items. When on-hand inventory falls to or below s, a new batch will resume. In this case, the duration of depletion stage
T 2 ¼ 0, and the replenishment cycle time is Q =l. Note that cycle shortage occurs when on-hand inventory at the end of
one replenishment cycle is negative. That is, the number of units short NI2 is the total demand during the production period
minus s þ Q . The total number of cycle shortage can only be known at time Q =l. Thus, the shortage should be backordered
and filled with new stock. The detailed inventory profiles in one replenishment cycle are illustrated in Fig. 1 below, where AIi
denotes the accumulated inventory over the two stages i; i ¼ 1; 2, respectively. Fig. 1a depicts the net inventory for the case
that the replenishment cycle time is the sum of production time and depletion time, while Fig. 1b and c depict the net inven-
tory for the case that the replenishment cycle time is equal to the production time. Note that only transient shortage occurs
in Fig. 1b, while both transient and cycle shortages occur in Fig. 1c.
The considered cost includes: setup cost A when a lot size Q is launched, inventory carrying charge h per unit product per
unit time when on-hand inventory is positive, transient shortage cost b3 per unit time when a transient shortage occurred
during the production period, and cycle shortage cost b2 per unit product per unit when the on-hand stock is negative at the
end of the replenishment cycle. Note that the cost of a transient shortage is treated as an expedient shipping/opportunity
cost, or a penalty cost of a delayed shipment in the production system, which is time dependent. For the cycle shortage,
the number of units short can only be known at the end of the cycle, thus such a shortage cost can be treated as those found
in the classical inventory models.
In the following analysis, the most common assumptions in deriving (s; Q Þ inventory model are used, e.g., see Hadley and
Whitin [1] or Silver et al. [22] These include: (1) crossing order (lot size) is not permitted, i.e., they must be delivered in the
same order in which they are launched, (2) cycle shortage costs are so high that a practical production policy will result in
the average level of shortage being negligible small when compared with the average level of on-hand stock. Under these
conditions, we assume that there is at most one lot size to be launched in the production system. It implies that when
on-hand inventory falls to or below s, there is no outstanding batch to be processed (since l > kÞ and the inventory position
is equal to the net inventory.
The expected cycle time is the sum of production time Q =l to produce the lot size Q and expected depletion time to con-
sume remaining inventory. When total demand during the production stage is greater than or equal to Q, the cycle time is
equal to production time Q =l (see Fig. 1b and c). Otherwise, the cycle time is equivalent to the time to deplete the lot size Q,
i.e., the arrival time of Qth demand W Q (see Fig. 1a). Hence, the replenishment cycle time T can be expressed as below:

Q =l; if Q d ð0; Q=lÞ P Q ;
T¼
WQ ; if Q d ð0; Q=lÞ < Q :
Since demand is a Poisson process with mean rate k; W Q is Gamma distributed with parameters (Q ; k). Hence:

Q =l; if Q d ð0; Q =lÞ P Q;
E½T ¼
E½W Q  ¼ Q =k; if Q d ð0; Q =lÞ < Q :
The expected duration of one replenishment cycle is

X
Q1     X 1            
Q Q Q Q Q Q Q Q
E½T  ¼ P Q d 0; ¼k þ P Q d 0; ¼ k ¼ P Q d 0; 6 Q  1 þ P Q d 0; PQ
k¼0
k l k¼Q
l l k l l l
     
Q Q Q Q
¼   P Q d 0; PQ : ð1Þ
k k l l
Based on the high shortage cost in the case study, the expected units short of cycle shortage
n willbe small
o and occurred for
only a very small fraction of the time, and it can be assumed negligible. It implies that P Q d 0; Ql P Q is also negligible.
The expected duration of one replenishment cycle can be approximated by:

E½T  ffi E½W Q  ¼ Q =k: ð2Þ

Using the well-known renewal-reward theorem (e.g., see Ross [23]), the expected cost per unit time is the ratio of ex-
pected cost in one replenishment cycle and expected replenishment cycle time. In the following sections, the expected costs
of the two stages in one replenishment cycle are derived and analyzed.
 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354 3343

Inventory level Ii(t)

Q

I2(t)

I1(t)
s
AI1 AI2
Time, t
w1 w2 2/µ 4/µ 6/µ wQ-1 wQ
T1(production period) T2(depletion period)
Q/µ
T
1a. Inventory profile with production and depletion stages
Inventory level Ii(t)

I1(t)
s
AI1 Time, t
Q/µ
1b. Inventory profile with only transient shortage
Inventory level Ii(t)

s
I1(t) AI1 Time, t
Q/µ
Transient shortage Cycle shortage

1c. Inventory profile with both transient and cycle shortages

Fig. 1. Inventory profiles in one replenishment cycle.

2.1. The production stage

The time to process the lot size Q is Q/l, since production rate l is a constant. The relevant cost includes setup cost in the
production system, inventory carrying cost, and transient shortage cost. On-hand stock I1 ðtÞ at any time t; 0 6 t < Q =l, is
the difference of s þ Q p ð0; tÞ and the total demand Q d ð0; tÞ. If on-hand stock is greater than 0 at time t, inventory carrying
cost is incurred. Otherwise, transient shortage cost is incurred. The on-hand inventory I1 ðtÞ at time t; 0 6 t < Q =l, can be
expressed as:
(
0; if Q d ð0; tÞ > s þ Q p ð0; tÞ;
I1 ðtÞ ¼
s þ Q p ð0; tÞ  Q d ð0; tÞ; if Q d ð0; tÞ 6 s þ Q p ð0; tÞ;
3344 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354

where PfQ d ð0; tÞ ¼ kg ¼ pðk; ktÞ ¼ ðktÞk ekt =k!.

 ktÞ ¼ 1  Pðk  1; ktÞ ¼ P1 pðj; ktÞ, the expected on-hand inventory at any time t; 0 6 t < Q =l, is:
Let Pðk; j¼k
X
sþi X
1
E½I1 ðtÞ ¼ ðs þ i  kÞpðk; ktÞ þ 0  pðk; ktÞ; i=l 6 t < ði þ 1Þ=l; i ¼ 0; 1; 2; . . . ; Q  1
k¼0 k¼sþiþ1
 þ i; ktÞ  ðs þ iÞP ðs þ i þ 1; kt Þ;
¼ s þ i  kt þ kt Pðs i=l 6 t < ði þ 1Þ=l; i ¼ 0; 1; 2; . . . ; Q  1:
The expected accumulated on-hand inventory (the area under on-hand inventory, see also Fig. 1a) over production period
E½AI1  can be expressed as:
Z Q 1 Z
X ðiþ1Þ=l 
E½AI1  ¼ E½I1 ðtÞdt ¼  ðs þ i; kt Þ  ðs þ iÞPðs
s þ i  kt þ kt P  þ i þ 1; ktÞ dt: ð3Þ
i¼0 i=l

Using the following two expressions for further simplification (see pp. 441–442, Hadley and Whitin, [1]):
Z T
Pðr;  kTÞ  r Pðr
 ktÞdt ¼ T Pðr;  þ 1; kTÞ; and
0 k
Z T
 ktÞdt ¼ T nþ1  1 1 ðn þ rÞ! 
t n Pðr; Pðr; kTÞ  nþ1 Pðn þ r þ 1; kT Þ; n ¼ 0; 1; 2; . . .
0 nþ1 k n þ 1 ðr  1Þ!
Eq. (3) can be rewritten as:
X
Q 1
sþi kð2i þ 1Þ
E½AI1  ¼ 
i¼0
l 2l2
"    #
X
Q 1
ði þ 1Þ2  kði þ 1Þ ðs þ iÞðs þ i þ 1Þ  kði þ 1Þ
þ k P s þ i;  P s þ i þ 2;
i¼0
2l2 l 2k2 l
 "   #
X
Q 1 2
i  ki ðs þ iÞðs þ i þ 1Þ  ki
 k P s þ i;  P s þ i þ 2;
i¼0
2l 2 l 2k2 l
X
Q 1    
ði þ 1Þ  kði þ 1Þ sþiþ1 kði þ 1Þ
 ðs þ iÞ P s þ i þ 1;  P s þ i þ 2;
i¼0
l l k l
X
Q 1    
i  ki sþiþ1 ki
þ ðs þ iÞ P s þ i þ 1;  P s þ i þ 2; : ð4Þ
i¼0
l l k l
PQ 1 hsþi i
Let E½AI1  ¼ i¼0 l  kð22iþ1
l2
Þ
þ f1 ðs; Q Þ, where
"    #
X
Q 1
ði þ 1Þ2  kði þ 1Þ ðs þ iÞðs þ i þ 1Þ  kði þ 1Þ
f1 ðs; Q Þ ¼ k P s þ i;  P s þ i þ 2;
i¼0
2l2 l 2k2 l
"    #
X
Q 1 2
i  ki ðs þ iÞðs þ i þ 1Þ  ki
 k P s þ i;  P s þ i þ 2;
i¼0
2l 2 l 2k2 l
X
Q 1    
ði þ 1Þ  kði þ 1Þ sþiþ1 kði þ 1Þ
 ðs þ iÞ P s þ i þ 1;  P s þ i þ 2;
i¼0
l l k l
X
Q 1    
i  ki sþiþ1 ki
þ ðs þ iÞ P s þ i þ 1;  P s þ i þ 2;
i¼0
l l k l
Q 
X    X Q    
i sþi  ki sþi ki
¼  P s þ i; þ p s þ i;
i¼1
l k l i¼1
k l
" #  
kQ 2 Q ðs þ Q Þ ðs þ Q Þðs þ Q þ 1Þ  kQ
þ  þ P s þ Q;
2l 2 l 2k l
 
Q ðs þ Q Þ ðs þ Q Þðs þ Q þ 1Þ kQ
þ  p s þ Q;
2l 2k l
 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354 3345

Q    
sðs þ 1Þ Q ð2s þ Q  1Þ kQ 2 X sþi i ki
¼  þ 2þ  P s þ i;
2k 2l 2l i¼1
k l l
  " #  
XQ
i ki kQ 2 ðs þ Q Þðs þ Q þ 1Þ kQ
þ p s þ i;  þ P s þ Q;
i¼1
l l 2l 2 2k l
  " #  
Qðs þ Q Þ kQ kQ 2 Q ðs þ QÞ kQ
þ P s þ Q; þ  p s þ Q ; : ð5Þ
l l 2l 2 2l l
Using Eq. (5), Eq. (4) can be rewritten as:

Q ð2s þ Q  1Þ kQ 2 sðs þ 1Þ Q ð2s þ Q  1Þ kQ 2

E½AI1  ¼  2þ  þ 2
2l 2l 2k 2l 2l

X sþi i
Q    XiQ  
ki ki
þ  P s þ i; þ p s þ i;
i¼1
k l l i¼1
l l
" #    
kQ 2 ðs þ Q Þðs þ Q þ 1Þ kQ Q ðs þ Q Þ kQ
 þ P s þ Q ; þ P s þ Q;
2l 2 2k l l l
" #  
kQ 2 Q ðs þ QÞ kQ
þ  p s þ Q ;
2l 2 2l l
Q     X  
sðs þ 1Þ X sþi i ki Q
i ki
¼ þ  P s þ i; þ p s þ i;
2k i¼1
k l l i¼1
l l
" #    
kQ 2 ðs þ Q Þðs þ Q þ 1Þ kQ Q ðs þ Q Þ kQ
 þ P s þ Q ; þ P s þ Q;
2l 2 2k l l l
" #  
kQ 2 Q ðs þ Q Þ kQ
þ  p s þ Q; : ð6Þ
2l2 2l l
Therefore, the expected inventory carrying cost during production period is hE½AI1 . In addition, on-hand inventory at time
t; 0 6 t < Q =l, may be negative, i.e., the total demand is larger than s þ Q p ð0; tÞ during the production stage, transient short-
age cost is incurred. See also Fig. 1a, where a transient shortage occurs. The number of transient shortage at any time
t; NI1 ðtÞ; 0 6 t < Q =l, can be represented below:
(
Q d ð0; tÞ  s  Q p ð0; tÞ; if Q d ð0; tÞ > s þ Q p ð0; tÞ;
NI1 ðtÞ ¼ ð7Þ
0; if Q d ð0; tÞ 6 s þ Q p ð0; tÞ:

The expected number of transient shortage at time t is:

X
sþi X
1
E½NI1 ðtÞ ¼ 0  pðk; ktÞ þ ðk  ðs þ iÞÞpðk; ktÞ; i=l 6 t < ði þ 1Þ=l; i ¼ 0; 1; 2; . . . ; Q  1;
k¼0 k¼sþiþ1

¼ktPðs þ i; ktÞ  ðs þ iÞP

 ðs þ i þ 1; kt Þ; i=l 6 t < ði þ 1Þ=l; i ¼ 0; 1; 2; . . . ; Q  1:
Hence, the expected accumulated transient shortage over production stage E½ANI1  is:
"    #
X
Q 1
ði þ 1Þ2  kði þ 1Þ ðs þ iÞðs þ i þ 1Þ  kði þ 1Þ
E½ANI1  ¼ k P s þ i;  P s þ i þ 2;
i¼0
2l2 l 2k2 l
"    #
X
Q1 2
i  ki ðs þ iÞðs þ i þ 1Þ  ki
 k P s þ i;  P s þ i þ 2;
i¼0
2l2 l 2k2 l
X
Q1    
ði þ 1Þ  kði þ 1Þ sþiþ1 kði þ 1Þ
 ðs þ iÞ P s þ i þ 1;  P s þ i þ 2;
i¼0
l l k l
X
Q1    
i ki sþiþ1 ki
þ ðs þ iÞ P s þ i þ 1;  P s þ i þ 2; : ð8Þ
i¼0
l l k l
3346 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354

Note that the above equation is identical to f1 ðs; Q Þ of Eq. (5), it can be rewritten as:
Q    
sðs þ 1Þ Qð2s þ Q  1Þ kQ 2 X sþi i ki
E½ANI1  ¼  þ 2þ  P s þ i;
2k 2l 2l i¼1
k l l
XQ   " 2 #  
i ki kQ ðs þ QÞðs þ Q þ 1Þ kQ
þ p s þ i;  þ P s þ Q;
i¼1
l l 2l 2 2k l
  " 2 #  
Q ðs þ Q Þ kQ kQ Q ðs þ Q Þ kQ
þ P s þ Q; þ  p s þ Q; : ð9Þ
l l 2l2 2l l
The expected transient shortage cost is b3 E½ANI1 . Hence, the expected total cost at this stage is:
(
Q     X  
sð s þ 1Þ X sþi i ki Q
i ki
E½TC 1 ðs; Q Þ ¼A þ h þ  P s þ i; þ p s þ i;
2k i¼1
k l l i¼1
l l
" #  
2
kQ ðs þ Q Þðs þ Q þ 1Þ kQ
 þ P s þ Q;
2l2 2k l
  " 2 #  )
Qðs þ Q Þ kQ kQ Q ðs þ QÞ kQ
þ P s þ Q; þ  p s þ Q;
l l 2l 2 2l l
(
sðs þ 1Þ Q ð2s þ Q  1Þ kQ 2 X sþi i 
Q  
ki

þ b3  þ 2þ  P s þ i;
2k 2l 2l i¼1
k l l
  " #  
X Q
i ki kQ 2 ðs þ Q Þðs þ Q þ 1Þ kQ
þ p s þ i;  þ P s þ Q;
i¼1
l l 2l2 2k l
  " 2 #  )
Qðs þ Q Þ kQ kQ Q ðs þ QÞ kQ
þ P s þ Q; þ  p s þ Q ; : ð10Þ
l l 2l 2 2l l

2.2. The depletion stage

At the beginning of the depletion stage, i.e., at time Q =l, if total demand in the interval ½0; Q =lÞ is greater than Q, a new
lot size is launched (see Fig. 1b and c). Hence, both the duration and the inventory carrying cost of the depletion stage are
zero. Furthermore, when total demand during the production stage is greater than s þ Q , cycle shortage occurred (see
Fig. 1c). The expected units short E½NI2  can be expressed as:
 
Q
E½NI2  ¼ E E NI2 Q d 0; ¼ i; i ¼ s þ Q þ 1; s þ Q þ 2;   
l
 
Q
¼ E i  s  Q Q d 0; ¼ i; i ¼ s þ Q þ 1; s þ Q þ 2;   
l
X
1
¼ ði  s  Q Þpði; kQ =lÞ
i¼sþQþ1

   
kQ  kQ  s þ Q þ 1; kQ :
¼ P s þ Q;  ðs þ Q ÞP ð11Þ
l l l
Hence, the expected cycle shortage cost is b2 E½NI2 .
When on-hand inventory at time Q =l is greater than resetup point s, i.e., the total demand during production stage is less
than Q, the production system is switched to standby mode or to produce another product. Thus, inventory carrying cost has
incurred. The initial stock at depletion period is the ending inventory of the last stage, i.e., s þ Q minus the total demand
during the production period. When total demand during production period is k, the conditional accumulated on-hand
inventory (expected area under on-hand inventory, see Fig. 1a) over depletion stage is:
" "       ##
X
Qk
Q Q Q
E½AI2  ¼ E E W kþi  þ s WQ  Q d 0; ¼ k; k ¼ 0; 1;    ; Q  1 ; ð12Þ
i¼1
l l l
where W kþi is the arrival time of (k þ iÞth demand. Since demand follows Poisson process with mean rate k; W kþi is Gamma
distributed with parameters (k þ i; k). The expected accumulated on-hand inventory over this period can be rewritten as:
 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354 3347

" "       ##
X
Qk
Q Q Q
E½AI2  ¼ E E W kþi  þ s WQ  Q d 0; ¼ k; k ¼ 0; 1; . . . ; Q  1
i¼1
l l l
"       #
X
Qk
kþi Q Q Q Q
¼E  þs  Q d 0; ¼ k; k ¼ 0; 1; . . . ; Q  1
i¼1
k l k l l
 
ðQ  kÞðQ þ k þ 1Þ Q ðQ  k þ sÞ sQ Q
¼E  þ Q d 0; ¼ k; k ¼ 0; 1; . . . ; Q  1
2k l k l
    !  
Q ðQ þ 1Þ Q ðs þ Q Þ sQ kQ Q kQ 2 kQ
¼  þ P Q  1; þ  þ 2 P Q  2;
2k l k l l l l
 
kQ 2 kQ
 P Q  3;
2l2 l
!  
Q ðQ þ 1Þ Q ðs þ Q þ 1Þ sQ kQ 2 kQ
¼  þ þ 2 P Q;
2k l k 2l l
!  
Q sQ Q 2 sQ kQ 2 kQ
þ þ þ   2 p Q;
l l 2l k 2l l
!  
Q ðQ þ 1Þ Q ðs þ Q þ 1Þ sQ kQ 2 kQ
¼  þ þ 2 P Q  1;
2k l k 2l l
!  
Q ð Q þ 1Þ Q 2 kQ
þ  p Q; : ð13Þ
2k 2l l
The expected inventory carrying cost at the second stage is hE½AI2 . Hence, the expected cost at this stage is:
(" #   " #  )
Q ðQ þ 1Þ Qðs þ Q þ 1Þ sQ kQ 2 kQ Q ð Q þ 1Þ Q 2 kQ
E½TC 2 ðs; QÞ ¼ h  þ þ 2 P Q  1; þ  p Q;
2k l k 2l l 2k 2l l
   
kQ  kQ  s þ Q þ 1; kQ :
þ b2 P s þ Q;  ðs þ Q ÞP ð14Þ
l l l
Thus, the expected total cost in one replenishment cycle is the sum of Eqs. (10) and (14):
E½TC ðs; Q Þ ¼E½TC 1 ðs; Q Þ þ E½TC 2 ðs; Q Þ
(
Q     X  
sðs þ 1Þ X sþi i ki Q
i ki
¼A þ h þ  P s þ i; þ p s þ i;
2k i¼1
k l l i¼1
l l
" #    
kQ 2 ðs þ Q Þðs þ Q þ 1Þ kQ Qðs þ QÞ kQ
 þ P s þ Q; þ P s þ Q ;
2l2 2k l l l
" #  
Q ðQ þ 1Þ Q ðs þ Q þ 1Þ sQ kQ 2 kQ
þ  þ þ 2 P Q  1;
2k l k 2l l
" #   " #  )
kQ 2 Q ðs þ QÞ kQ Q ð Q þ 1Þ Q 2 kQ
þ  p s þ Q; þ  p Q;
2l 2 2l l 2k 2l l
(
Q    
sðs þ 1Þ Q ð2s þ Q  1Þ kQ 2 X sþi i ki
þ b3  þ 2þ  P s þ i;
2k 2l 2l i¼1
k l l
  " #  
XQ
i ki kQ 2 ðs þ Q Þðs þ Q þ 1Þ kQ
þ p s þ i;  þ P s þ Q;
i¼1
l l 2l2 2k l
  " #   )
Q ðs þ Q Þ kQ kQ 2 Q ðs þ Q Þ kQ
þ P s þ Q; þ  p s þ Q ;
l l 2l2 2l l
   
kQ  kQ  s þ Q þ 1; kQ :
þ b2 P s þ Q;  ðs þ Q ÞP ð15Þ
l l l
3348 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354

Using Eq. (2) as an approximation, the expected cost per unit time can be expressed as below:
(
Q      
k sðs þ 1Þ k X sþi i ki kXQ
i ki
E½C ðs; Q Þ ffi A þh þ  P s þ i; þ p s þ i;
Q 2Q Q i¼1 k l l Q i¼1 l l
" #    
k2 Q ðs þ Q Þðs þ Q þ 1Þ kQ kðs þ Q Þ kQ
 þ P s þ Q ; þ P s þ Q;
2l2 2Q l l l
" #  
ðQ þ 1Þ kðs þ Q þ 1Þ k2 Q kQ
þ  þ s þ 2 P Q  1;
2 l 2l l
" #    )
k2 Q kðs þ Q Þ kQ ðQ þ 1Þ kQ kQ
þ  p s þ Q; þ  p Q;
2l2 2l l 2 2l l
(    
sðs þ 1Þ kð2s þ Q  1Þ k2 Q k X Q
sþi i ki
þ b3  þ 2þ  P s þ i;
2Q 2l 2l Q i¼1 k l l
  " #  
kX Q
i ki k2 Q ðs þ Q Þðs þ Q þ 1Þ kQ
þ p s þ i;  þ P s þ Q ;
Q i¼1 l l 2l2 2Q l
  " #   )
kðs þ Q Þ kQ k2 Q kðs þ Q Þ kQ
þ P s þ Q; þ  p s þ Q ;
l l 2l2 2l l
"    #
k2  kQ kðs þ QÞ  kQ
þ b2 P s þ Q;  P s þ Q þ 1; : ð16Þ
l l Q l

Theorem 1. The lower bound and the upper bound of the expected cost per unit time of Eq. (16) are convex with respect to control
parameters s and Q.

Proof. For the Poisson distributed demand with parameter k, the probability mass function is skewed to the right; hence the
median of the distribution Q md is less than the expected value of the distribution, i.e., E½Q d ð0; t ¼ Q =lÞ ¼ kQ =l (see pp. 44,
McClave and Sincich [24]). That is, Q md < kQ =l < Q , since 0 6 k=l < 1. Based on the definition of median, we have:

 Q m ; kQ=l > PðkQ=l; kQ =lÞ > P
1=2 ¼ P  ðQ; kQ =lÞ > pðQ ; kQ =lÞ P 0:
d

The following two inequalities can be obtained:

0 6 P ðQ ; kQ=lÞ < 1=2 for all Q P 1; ð17Þ

and 0 6 pðQ ; kQ =lÞ < 1=2 for all Q P 1: ð18Þ

Using Eqs. (17) and (18), the upper bound and lower bound of E½Cðs; Q Þ in Eq. (16) can be rewritten as:
(
Q      
k sðs þ 1Þ k X sþi i ki kX Q
i ki
E½C ðs; Q Þ ¼ A þ h þ  P s þ i; þ p s þ i;
Q 2Q Q i¼1 k l l Q i¼1 l l
" #    
k2 Q ðs þ QÞðs þ Q þ 1Þ kQ kðs þ QÞ kQ
 þ P s þ Q ; þ P s þ Q ;
2l2 2Q l l l
" #  
2
ðQ þ 1Þ kðs þ Q þ 1Þ k Q kQ
þ  þ s þ 2 P Q  1;
2 l 2l l
" #    )
k2 Q kðs þ Q Þ kQ ðQ þ 1Þ kQ kQ
þ  p s þ Q ; þ  p Q ;
2l2 2l l 2 2l l
(    
sðs þ 1Þ kð2s þ Q  1Þ k2 Q k X Q
sþi i ki
þ b3  þ 2þ  P s þ i;
2Q 2l 2l Q i¼1 k l l
 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354 3349

  " 2 #  
kX Q
i ki k Q ðs þ Q Þðs þ Q þ 1Þ kQ
þ p s þ i;  þ P s þ Q;
Q i¼1 l l 2l 2 2Q l
  " #   )
kðs þ Q Þ kQ k2 Q kðs þ Q Þ kQ
þ P s þ Q; þ  p s þ Q;
l l 2l 2 2l l
"     #
k2  kQ kðs þ Q Þ  kQ
þ b2 P s þ Q;  P s þ Q þ 1;
l l Q l
(
Q  
k sðs þ 1Þ k X sþi i kX Q
i 1
<A þh þ  1þ 
Q 2Q Q i¼1 k l Q i¼1 l 2
" #  
k2 Q ðs þ Q Þðs þ Q þ 1Þ kQ kðs þ Q Þ
 þ P Q  1; þ 1
2l 2 2Q l l
" #    
ðQ þ 1Þ kðs þ Q þ 1Þ k2 Q kQ k2 Q kQ
þ  þ s þ 2 P Q  1; þ 2 p s þ Q;
2 l 2l l 2l l
 
ðQ þ 1Þ kQ kQ
þ  p Q;
2 2l l
(
Q  
sðs þ 1Þ kð2s þ Q  1Þ k2 Q k X sþi i
þ b3  þ 2þ  1
2Q 2l 2l Q i¼1 k l
" #  
kX Q
i 1 k2 Q ðs þ Q Þðs þ Q þ 1Þ kQ
þ   þ P Q  1;
Q i¼1 l 2 2l2 2Q l
!   ) " #
kðs þ QÞ k2 Q kQ k2 1 kðs þ Q Þ
þ 1þ p s þ Q; þ b2   0
l 2l2 l l 2 Q
(  
k ð2s þ Q þ 1Þ kðQ þ 5Þ sðs þ 1Þ kðs þ Q þ 1Þ  kQ
<A þh  þ þ P Q; :
Q 2 4l 2Q l l
!    )
k2 Q kQ ðQ þ 1Þ kQ kQ k2
þ p Q ; þ  p Q ; þ b2
2l 2 l 2 2l l 2l
( " #   !  )
kðQ þ 1Þ k2 Q ðs þ QÞðs þ Q þ 1Þ  kQ k2 Q kQ
þ b3 þ þ P Q; þ p Q;
4l 2l2 2Q l 2l2 l
(
k ð2s þ Q þ 1Þ kðQ þ 5Þ sðs þ 1Þ kðs þ Q þ 1Þ 1
<A þh  þ þ  :
Q 2 4l 2Q l 2
! )
k2 Q 1 ðQ þ 1Þ kQ 1 b2 k2
þ  þ   þ
2l 2 2 2 2l 2 2l
( " # ! )
kðQ þ 1Þ k2 Q ðs þ QÞðs þ Q þ 1Þ 1 k2 Q 1
þ b3 þ þ  þ 
4l 2l2 2Q 2 2l2 2
" #
k ð4s þ 3Q þ 3Þ kð2s  3Þ k2 Q sðs þ 1Þ b2 k2
<A þh þ þ 2þ þ
Q 4 4l 4l 4Q 2l
" #
kðQ þ 1Þ k2 Q ðs þ QÞðs þ Q þ 1Þ
þ b3 þ 2þ
4l 2l 4Q

¼ EU ½C ðs; Q Þ ¼ f2 ðs; Q Þ
and
Ak ð2s þ Q þ 1Þ kðQ þ 5Þ b3 kðQ þ 1Þ
E½Cðs; QÞ > þh  þ ¼ EL ½C ðs; Q Þ ¼ f3 ðs; Q Þ:
Q 2 4l 4l
3350 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354

For the upper bound f2 ðs; Q Þ, the Hessian matrix is:

2Ak
Q3
þ sðhþb2Q3 Þ3ðsþ1Þ ðhþb3 Þð2sþ1Þ
4Q 2
ðhþb3 Þð2sþ1Þ ðhþb3 Þ
4Q 2 2Q

2Ak sðh þ b3 Þðs þ 1Þ ðh þ b3 Þ ðh þ b3 Þð2s þ 1Þ ðh þ b3 Þð2s þ 1Þ

¼ þ 
Q3 2Q 3 2Q 4Q 2 4Q 2
2 2 2
Akðh þ b3 Þ sðh þ b3 Þ ðs þ 1Þ ðh þ b3 Þ ð2s þ 1Þ
¼ þ 
Q4 4Q 4 16Q 4
2
Akðh þ b3 Þ ðh þ b3 Þ
¼ 
Q4 16Q 4
ðh þ b3 Þð16Ak  h  b3 Þ
¼ P 0; 8Q P 1; s P 0; if 16Ak P h þ b3 :
16Q 4
For the lower bound f3 ðs; Q Þ, the Hessian matrix is:
2Ak
Q3
0
¼ 0 P 0; 8Q P 1; s P 0:
0 0
The Hessian matrices of f2 ðs; Q Þ and f3 ðs; Q Þ are positive definite. Hence, both the upper bound f2 ðs; Q Þ and lower bound
f3 ðs; Q Þ are convex with respect to s and Q when 16Ak P h þ b. h

In addition, the second partial derivatives of both f2 ðs; Q Þ and f3 ðs; Q Þ are non-negative for any s and Q, respectively. These
results imply that the conditional expected cost per unit time is convex with respect to s (or Q Þ for a given Q (or sÞ. Using the
above properties, a fast algorithm for finding the optimal production (s; Q Þ policy is developed in the following section.

3. The optimal lot sizing policy and numerical study

A solution procedure to compute the optimal (s; Q Þ production policy and the minimum cost per unit time is proposed in
Section 3.1. In Section 3.2, we report some computational results for the proposed lot sizing policy under Poisson demand
process. Important parameters that affect the expected cost and the production control policy are then summarized.

3.1. The solution procedure

Based on the analytical results presented in the last section, a search procedure is developed to find the optimal resetup
point s, the fixed lot size Q, and the minimal expected cost per unit time. Details of this search procedure are described
below:
 jqffiffiffiffiffiffiffiffiffiffiffiffiffiffik
2Ak
Step 1: Compute the classical EPQ, i.e., ðs; Q Þ ¼ 0; hð1k=lÞ
and the expected cost per unit time E½C ðs; Q Þ by Eqs. (1)
and (15). Let C  ¼ E½C ðs; Q Þ, and go to the next step.
Step 2: For a given Q,
Step 2.1: Compute E½C ðs  1; Q Þ; E½C ðs þ 1; Q Þ, respectively for s > 0. If MinfE½C ðs  1; Q Þ; C  ; E½C ðs þ 1; Q Þg ¼ C  ,
then the optimal solution is s with the minimal cost C  . Go to step 3.
Step 2.2: Else if MinfE½C ðs  1; Q Þ; C  ; E½C ðs þ 1; Q Þg ¼ E½C ðs þ 1; Q Þ, then s s þ 1, C  ¼ E½C ðs; Q Þ, and D ¼ 1. Go to
step 2.4.
Step 2.3: Else, let s s  1, C  ¼ E½C ðs; Q Þ, and D ¼ 1. Go to Step 2.4.
Step 2.4: Let s s þ D and compute E½C ðs; Q Þ. If E½C ðs; Q Þ < C  , then C  ¼ E½C ðs; Q Þ. Repeat this step until
E½C ðs; Q Þ P C  . The optimal solution is s  D with the minimal cost C  ¼ E½C ðs  D; Q Þ. Let s s  D and
go to step 3.
Step 3: For a given s found at the last step, perform the below search:
Step 3.1: Compute E½C ðs; Q  1Þ; E½C ðs; Q þ 1Þ, respectively for Q > 1. If MinfE½C ðs; Q  1Þ; C  ; E½C ðs; Q þ 1Þg ¼ C  ,
then the optimal solution is Q with the minimal cost C  . Go to step 4.
Step 3.2: Else if MinfE½C ðs; Q  1Þ; C  ; E½C ðs; Q þ 1Þg ¼ E½C ðs; Q þ 1Þ, then Q Q +1, C  ¼ E½C ðs; Q Þ, and D ¼ 1. Go to
step 3.4.
Step 3.3: Else, let Q Q  1, C  ¼ E½C ðs; Q Þ, and D ¼ 1. Go to Step 3.4.
Step 3.4: Let Q Q þ D and compute E½C ðs; Q Þ. If E½C ðs; Q Þ < C  , then C  ¼ E½C ðs; Q Þ. Repeat this step until
E½C ðs; Q Þ P C  . The optimal solution is Q  D with the minimal cost C  ¼ E½C ðs; Q  DÞ. Let Q Q D
and go to step 2.
Step 4: The optimal production policy is (s; Q Þ with minimal cost C⁄.

For illustration purpose, data for a particular type of CNC machine from a precision machinery assembly line is collected.
Production rate is 2.0 per week; and mean demand rate is 1.2 per week, which follows a Poisson process (an order of size
 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354 3351

C(s, Q)

12,000
11,000
10,000
9,000
8,000
7,000
6,000
Q
5,000 11 16
1 6
0
1
2
3
4
5
s

Fig. 2. The illustrated expected cost per unit time.

one is the most frequent, and average time between orders is 5.8 days). The cost parameters are: setup cost A = $4,000,
inventory carrying charge h = $500 per unit per week, transient shortage cost b3 ¼ $15; 000 per unit per week and cycle
shortage cost b2 ¼ $5; 000 per unit. The search process is given below:
 jqffiffiffiffiffiffiffiffiffiffiffiffiffiffik
Step 1: The initial production policy is ðs; Q Þ ¼ 0; 2Ak
hð1k=lÞ
¼ ð0; 6Þ. C  ¼ E½C ðs ¼ 0; Q ¼ 6Þ ¼ 3; 814:0.
Step 2: For the given Q = 6,
Step 2.2: Since E½C ðs þ 1 ¼ 1; Q ¼ 6Þ= 2,524.1 < C ⁄, s ¼ s þ 1 ¼ 1, C  ¼ 2; 524:1 and D ¼ 1. Go to step 2.4.
Step 2.4: Since E½C ðs þ 1 ¼ 2; Q ¼ 6Þ ¼ 2; 417:7 < C  ; s ¼ 2 and C  ¼ 2; 417:7. No further improvement, s = 2 and go
to step 3.
Step 3: For a given s = 2,
Step 3.2: E½C ðs ¼ 2; Q  1 ¼ 5Þ ¼ 2; 463:9; E½C ðs ¼ 2; Q þ 1 ¼ 7Þ ¼ 2; 411:8. Since MinfE½C ðs; Q  1Þ; C  ;

E½C ðs; Q þ 1Þg ¼ E½C ðs; Q þ 1Þ; Q ¼ Q þ 1 ¼ 7; C ¼ 2; 411:8 and D ¼ 1. Go to step 3.4.
Step 3.4: Since E½C ðs ¼ 2; Q þ 1 ¼ 8Þ ¼ 2; 430:4 > C  ; Q ¼ 7 and C  ¼ 2; 411:8. No further improvement, Q = 7 and go
to step 2.
Step 2: For the given Q = 7,
Step 2.1: E½C ðs  1 ¼ 1; Q ¼ 7Þ ¼ 2; 490:0; E½C ðs þ 1 ¼ 3; Q ¼ 7Þ ¼ 2; 712:3. Since MinfE½C ðs  1; Q Þ; C  ;
 
E½C ðs þ 1; Q Þg ¼ C , then the optimal solution is s = 2 with the minimal cost C ¼ 2; 411:8. Go to step 3.
Step 3: For a given s = 2,
Step 3.1: E½C ðs ¼ 2; Q  1 ¼ 6Þ ¼ 2; 417:7; E½C ðs ¼ 2; Q þ 1 ¼ 8Þ ¼ 2; 430:4. Since MinfE½C ðs; Q  1Þ; C  ;
 
E½C ðs; Q þ 1Þg ¼ C , then the optimal solution is Q = 7 with the minimal cost C ¼ 2; 411:8. Go to step 4.
Step 4: The optimal lot sizing policy is ðs; Q Þ ¼ ð2; 7Þ with minimal cost C  ¼ E½C ð2; 7Þ ¼ 2; 411:8.

The optimal production policy is ðs; Q Þ ¼ ð2; 7Þ with the minimal expected cost $2,411.8. Using classical EPQ model, the
optimal policy is ðs; Q Þ ¼ ð0; 6Þ with cost $3,814.0. The cost reduction is 37% when the proposed model is implemented in-
stead of that in the classical EPQ model. The expected cost per unit time for a wide range of resetup point s and lot size Q is
shown in Fig. 2 above. The expected cost function appears convex with respect to decision parameters s and Q.

3.2. Additional numerical study

The computational experiment is performed with a 25 factorial design, where effects of various parameters on the pro-
posed production policy are studied. The experimental settings in the production systems are described below. The produc-
tion rate l is 2.0 per week in all of 32 experiments. The remaining five factors are as follows: setup cost A, mean demand rate
k, inventory carrying charge h, transient shortage cost b3 , and cycle shortage cost b2 . Levels of factors are denoted as ‘‘low’’
and ‘‘high’’ to give alternative configurations of the production systems. For an example, run 1 in the table corresponds to a
production system with production rate 2.0 per week, setup cost $1,000 (low), mean demand rate 1.0 per week (low),
3352 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354

Table 1
Experimental results in 25 factorial design.

Run Cost parameter Classical EPQ (s; Q Þ; C  Proposed model (s; Q Þ; C  Cost improvement (%)
A k h b3 b2
1 1000 1 400 12,000 4000 (0, 3); 2,055.6 (1, 3); 1,253.9 39.0
2 1000 1 400 12,000 6000 (0, 3); 2,115.5 (1, 3); 1,270.0 40.0
3 3000 1 400 12,000 4000 (0, 5); 2,221.5 (1, 6); 1,678.0 24.5
4 3000 1 400 12,000 6000 (0, 5); 2,246.3 (1, 6); 1,685.7 25.0
5 1000 1.4 400 12,000 4000 (0, 5); 4,070.6 (2, 3); 1,537.7 62.2
6 1000 1.4 400 12,000 6000 (0, 5); 4,210.1 (2, 3); 1,567.2 62.8
7 3000 1.4 400 12,000 4000 (0, 8); 4,205.5 (2, 7); 2,084.4 50.4
8 3000 1.4 400 12,000 6000 (0, 8); 4,283.3 (2, 7); 2,103.5 50.9
9 1000 1 600 12,000 4000 (0, 3); 2,194.7 (1, 3); 1,556.6 29.1
10 1000 1 600 12,000 6000 (0, 3); 2,254.6 (1, 3); 1,572.6 30.2
11 3000 1 600 12,000 4000 (0, 4); 2,590.1 (1, 5); 2,099.7 18.9
12 3000 1 600 12,000 6000 (0, 4); 2,627.7 (1, 5); 2,107.7 19.8
13 1000 1.4 600 12,000 4000 (0, 4); 4,283.0 (2, 3); 1,920.3 55.2
14 1000 1.4 600 12,000 6000 (0, 4); 4,460.3 (2, 3); 1,946.7 56.3
15 3000 1.4 600 12,000 4000 (0, 7); 4,493.6 (2, 6); 2,571.8 42.8
16 3000 1.4 600 12,000 6000 (0, 7); 4,586.6 (2, 6); 2,592.8 43.4
17 1000 1 400 18,000 4000 (0, 3); 2,717.8 (2, 3); 1,379.6 49.2
18 1000 1 400 18,000 6000 (0, 3); 2,777.7 (2, 3); 1,383.3 50.2
19 3000 1 400 18,000 4000 (0, 5); 2,779.1 (1, 6); 1,810.7 34.8
20 3000 1 400 18,000 6000 (0, 5); 2,803.9 (1, 6); 1,816.4 35.2
21 1000 1.4 400 18,000 4000 (0, 5); 5,700.5 (3, 4); 1,677.1 70.6
22 1000 1.4 400 18,000 6000 (0, 5); 5,839.9 (3, 4); 1,685.1 71.1
23 3000 1.4 400 18,000 4000 (0, 8); 5,762.4 (3, 7); 2,212.7 61.6
24 3000 1.4 400 18,000 6000 (0, 8); 5,840.2 (3, 7); 2,220.3 62.0
25 1000 1 600 18,000 4000 (0, 3); 2,856.9 (1, 3); 1,698.1 40.6
26 1000 1 600 18,000 6000 (0, 3); 2,916.8 (1, 3); 1,714.2 41.2
27 3000 1 600 18,000 4000 (0, 4); 3,198.0 (1, 5); 2,236.4 30.1
28 3000 1 600 18,000 6000 (0, 4); 3,255.6 (1, 5); 2,244.4 31.1
29 1000 1.4 600 18,000 4000 (0, 4); 5,922.0 (2, 3); 2,046.9 65.4
30 1000 1.4 600 18,000 6000 (0, 4); 6,099.3 (2, 3); 2,073.2 66.0
31 3000 1.4 600 18,000 4000 (0, 7); 6,079.3 (2, 5); 2,766.0 54.5
32 3000 1.4 600 18,000 6000 (0, 7); 6,172.3 (2, 6); 2,789.2 54.8
Average cost improvement 45.9

inventory carrying charge $400 per unit per week (low), transient shortage cost $4,000 per unit per week, and cycle shortage
cost $12,000 per unit (low). Table 1 above gives the experimental results.
The optimal production policy and the expected cost per unit time for both classical EPQ and proposed models are shown
in the 7th and 8th columns, respectively. Column 9 depicts the percentage of cost reduction when the proposed model is
used instead of that in the classical EPQ model. It is clear that the proposed model outperforms EPQ, with an average cost
reduction of nearly 46%. The most influential factors are: mean demand rate, setup cost, transient shortage cost, inventory
carrying charge, and cycle shortage cost in the decreasing order. In addition, the cost improvement is significant when de-
mand rate and transient shortage cost are at high level.
Table 2 below depicts the details of shortage statistics in the above numerical experiments. Note that the probability of
cycle shortage and the expected units short for cycle shortage is 0.0173 and 0.0252, respectively, which are both negligi-
ble. It indicates that the assumptions made earlier are realistic enough and reasonable. In addition, the implementation of
resetup point contributes to a significant portion of the cost savings since the lot size Q is very close to the classical EPQ. It
also appears that the use of resetup point significantly reduces both the number of backorders and the probability of
shortage.

4. Concluding summary

An extend EPQ model that includes a positive resetup point s and a fixed lot size Q is studied in this paper, where demand
of customers follows a Poisson process. The expected production inventory cost model is analyzed. Under some mild con-
ditions, the expected cost function can be shown to be convex with respect to the two decision parameters: resetup point
and lot size. A fast search algorithm is developed to find the optimal resetup point and the production quantity. Our numer-
ical experiment has demonstrated that the cost improvement is significant when the proposed model is used, instead of
using the classical EPQ policy, when faces random demand. The implementation of resetup point in the proposed model
has significant impact on the cost improvement. In addition, both the number of backorders and the probability of shortage
decrease dramatically under the proposed production policy.
 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354 3353

Table 2
Detailed shortage statistics in the numerical study.

Run Cost parameter Probability Expected units Expected shortage

of cycle shortage short E½NI2  cost per unit
A k h b3 b2 n   o
P Q d 0; Ql > s þ Q time b2 E½NI2 =E½T 

1 1000 1 400 12,000 4000 0.0186 0.0242 33.31

2 1000 1 400 12,000 6000 0.0186 0.0242 49.96
3 3000 1 400 12,000 4000 0.0119 0.0172 11.66
4 3000 1 400 12,000 6000 0.0119 0.0172 17.49
5 1000 1.4 400 12,000 4000 0.0204 0.0282 55.36
6 1000 1.4 400 12,000 6000 0.0204 0.0282 83.03
7 3000 1.4 400 12,000 4000 0.0283 0.0475 39.48
8 3000 1.4 400 12,000 6000 0.0283 0.0475 59.22
9 1000 1 600 12,000 4000 0.0186 0.0242 33.31
10 1000 1 600 12,000 6000 0.0186 0.0242 49.96
11 3000 1 600 12,000 4000 0.0142 0.0199 16.29
12 3000 1 600 12,000 6000 0.0142 0.0199 24.43
13 1000 1.4 600 12,000 4000 0.0204 0.0282 55.36
14 1000 1.4 600 12,000 6000 0.0204 0.0282 83.03
15 3000 1.4 600 12,000 4000 0.0279 0.0451 43.84
16 3000 1.4 600 12,000 6000 0.0279 0.0451 65.76
17 1000 1 400 18,000 4000 0.0045 0.0056 7.70
18 1000 1 400 18,000 6000 0.0045 0.0056 11.55
19 3000 1 400 18,000 4000 0.0119 0.0172 11.66
20 3000 1 400 18,000 6000 0.0119 0.0172 17.49
21 1000 1.4 400 18,000 4000 0.0081 0.0114 16.77
22 1000 1.4 400 18,000 6000 0.0081 0.0114 25.16
23 3000 1.4 400 18,000 4000 0.0120 0.0192 15.93
24 3000 1.4 400 18,000 6000 0.0120 0.0192 23.90
25 1000 1 600 18,000 4000 0.0186 0.0242 33.31
26 1000 1 600 18,000 6000 0.0186 0.0242 49.96
27 3000 1 600 18,000 4000 0.0142 0.0199 16.29
28 3000 1 600 18,000 6000 0.0142 0.0199 24.43
29 1000 1.4 600 18,000 4000 0.0204 0.0282 55.36
30 1000 1.4 600 18,000 6000 0.0204 0.0282 83.03
31 3000 1.4 600 18,000 4000 0.0267 0.0413 48.36
32 3000 1.4 600 18,000 6000 0.0279 0.0451 65.76
Average 0.0173 0.0252 38.38

This work serves as a basic framework for more advanced study, such as different shortage costs, lead time to resume the
production, or other types of control policies. The performance of different control policies can be compared and analyzed to
identify the best production planning or control policy.

Acknowledgements

The authors thank the anonymous referees and the editor, for their constructive input and comments, which served to
improve the manuscript. This work was supported in part by the National Science Council of Taiwan under Grant NSC95-
2416-H006-025-MY2.

References

[1] G. Hadley, T.M. Whitin, Analysis of Inventory Systems, Prentice Hall Inc., New Jersey, 1963.
[2] W.A. Donaldson, Inventory replenishment policy for a linear trend in demand – an analytical solution, Oper. Res. Quart. 28 (3) (1977) 663–670.
[3] M. Deb, K. Chaudhuri, A note on the heuristic for replenishment of trended inventories considering shortages, J. Oper. Res. Soc. 38 (5) (1987) 459–463.
[4] M. Hariga, The inventory replenishment problem with a linear trend in demand, Comput. Indust. Eng. 24 (2) (1993) 143–150.
[5] M.A. Hariga, L. Benkherouf, Optimal and heuristic inventory replenishment models for deteriorating items with exponential time-varying demand, Eur.
J. Oper. Res. 79 (1) (1994) 123–137.
[6] T.Y. Lin, An economic order quantity with imperfect quality and quantity discounts, Appl. Math. Model. 34 (10) (2010) 3158–3165.
[7] H.C. Chang, A comprehensive note on: an economic order quantity with imperfect quality and quantity discounts, Appl. Math. Model. 35 (10) (2011)
5208–5216.
[8] L.E. Cárdenas-Barrón, The derivation of EOQ/EPQ inventory models with two backorders costs using analytic geometry and algebra, Appl. Math. Model.
35 (5) (2011) 2394–2407.
[9] L.E. Cárdenas-Barrón, A complement to a comprehensive note on: an economic order quantity with imperfect quality and quantity discounts, Appl.
Math. Model. 36 (12) (2012) 6338–6340.
[10] K.J. Chung, L.E. Cárdenas-Barrón, The complete solution procedure for the EOQ and EPQ inventory models with linear and fixed backorder costs, Math.
Comput. Model. 55 (11–12) (2012) 2151–2156.
[11] G. Rabinowitz, A. Mehrez, C.W. Chu, B.E. Patuwo, A partial backorder control for continuous review (r; Q Þ inventory system with Poisson demand and
constant lead time, Comput. Oper. Res. 27 (7) (1995) 689–700.
3354 S.-D. Lee, C.-M. Yang / Applied Mathematical Modelling 37 (2013) 3340–3354

[12] K. Namit, J. Chen, Solutions to the (Q ; rÞ inventory model for Gamma lead-time demand, Int. J. Phys. Distribut. Logist. Manage. 29 (2) (1999) 138–154.
[13] B. Gavish, S.C. Graves, A one-product production/inventory problem under continuous review policy, Oper. Res. 28 (5) (1980) 1228–1236.
[14] R. Rempala, A continuous production-inventory problem with regeneration cycles, Int. J. Prod. Econ. 93–94 (8) (2005) 447–454.
[15] H. Groenevelt, L. Pintelon, A. Seidmann, Production batching with machine breakdowns and safety stocks, Oper. Res. 40 (5) (1992) 959–971.
[16] M. Palar, Continuous-review inventory problem with random supply interruptions, Eur. J. Oper. Res. 99 (2) (1997) 366–385.
[17] M.J. Rosenblatt, H.L. Lee, Economic production cycles with imperfect production processes, IIE Trans. 18 (1) (1986) 48–55.
[18] S.D. Lee, J.M. Rong, Production lot sizing in failure prone two-stage serial systems, Eur. J. Oper. Res. 123 (1) (2000) 42–60.
[19] S.K. Goyal, B.C. Giri, The production–inventory problem of a product with time varying demand, production and deterioration rates, Eur. J. Oper. Res.
147 (3) (2003) 549–557.
[20] S.W. Chiu, S.L. Wang, Y.S.P. Chiu, Determining the optimal run time for EPQ model with scrap, rework, and stochastic breakdowns, Eur. J. Oper. Res. 180
(2) (2007) 664–676.
[21] I. Konstantaras, K. Skouri, M.Y. Jaber, Inventory models for imperfect quality items with shortages and learning in inspection, Appl. Math. Model. 36
(11) (2012) 5334–5343.
[22] E.A. Silver, D.F. Pyke, R. Peterson, Inventory Management and Production Planning and Scheduling, third ed., John Wiley & Sons Inc., New York, 1998.
[23] S.M. Ross, Introduction to Probability Models, ninth ed., Academic Press, Inc., San Diego, 2007.
[24] J.T. McClave, T. Sincich, Statistics, eighth ed., Prentice-Hall, Inc., London, 2000.