Int J Adv Manuf Technol (2009) 41:12081220

DOI 10.1007/s00170-008-1567-5
ORIGINAL ARTICLE
A two-level distribution inventory system with stochastic
lead time at the lower echelon
A. Thangam R. Uthayakumar
Received: 28 August 2007 / Accepted: 12 May 2008 / Published online: 14 June 2008
Springer-Verlag London Limited 2008
Abstract We consider a two-level distribution inven-
tory system with a number of identical retailers at
the lower echelon and a single supplier at the upper
echelon. The replenishment policy is continuous review
policy (R, Q) at all installations. We assume indepen-
dent Poisson demands with stochastic lead time for the
retailers and a constant transportation time for replen-
ishing supplier orders from an external warehouse. Un-
satised demands are assumed to be lost in the retailers
and unsatised retailer orders are backordered in the
supplier. We develop an approximate cost function to
nd optimal reorder points for given batch sizes in
all installations, and the related accuracy is assessed
through simulation. We present numerical examples for
the gamma distributed lead time for the retailers orders
at the supplier.
Keywords Multi-echelon Stochastic lead time
Poisson demand Continuous review policy
1 Introduction
Large, multi-echelon inventory systems usually con-
sist of hundreds of thousands of stock keeping units
(SKUs). These SKUs can be classied into two main
categories: consumables and repairables. Multi-echelon
A. Thangam (B) R. Uthayakumar
Department of Mathematics,
Gandhigram Rural University, Dindigul 624 302,
Tamil Nadu, India
e-mail: [email protected]
inventory systems are important to large corporations
and to the military to support their operations.
In large supply networks like Wal-Mart and the
US-Navy, thousands of SKUs are stocked at different
inventory holding points (IHPs). These holding points
might be at different echelons, where the higher ech-
elons supply the lower echelons. Each of these IHPs
might follow different stocking policies resulting in
decentralized control of the supply network. This case
is most likely to occur when each of the locations that
constitute the supply network are owned by different
owners who are not willing to give control of their in-
ventories to external parties. Under this case, each loca-
tion might not take into consideration interactions with
the other locations that might have a signicant effect
on the efciency of the whole supply network, as well
as on each single location. On the other hand, if all of
these locations are owned or managed by a centralized
management system, a single inventory control system
might be implemented. Tremendous improvements are
attainable if a centralized inventory management sys-
tem is considered for the entire supply network. This
motivated building inventory models that consider the
entire supply network and the interactions between
their constituent IHPs. Most of these models have their
own assumptions and characteristics. Some of these
models are built for a special class of supply networks,
such as slow-moving and expensive spare part supply
networks. Other models are built for a particular struc-
ture of a supply network that might not be applicable
to other supply networks. Hence, modeling two-level
distribution inventory systems is still a rich area for
research.
Lead time plays an important role and has been a to-
pic of interest for many authors in two-level distribution
Int J Adv Manuf Technol (2009) 41:12081220 1209
inventory management. In general, the control parame-
ters (Q and R) depend on both the demand process
and a replenishment lead time. Although many studies
have treated lead time as constant, focusing solely on
demand, a two-stage inventory model with stochastic
lead time could be a building block in supply chain man-
agement. Variability in lead time between successive
stages is often what disturbs supply chain coordination.
Our work is motivated by situations where lead time,
while random, can be inuenced by the decision maker.
As one example, the inventory manager of a just-in-
time system could invest in decreasing the lead time in a
stochastic order sense. This is the case where (symmet-
rically) a rebate would be given to the customer who
accepts an above-average lead time.
In most of the early literature dealing with two-level
distribution inventory models, either in deterministic or
probabilistic models, lead time for the retailers orders
at the supplier is viewed as a constant. However, in
real inventory systems, it is not so. Moreover, the lead
time for the retailer orders is equivalent to the service
time for the supplier to meet demand from retailer
without considering the transportation time. We know
from queueing theory that the service time can follow
any probabilistic distribution. Hence, we have assumed
that the lead time for the retailer orders at the supplier
follows a general probabilistic distribution.
In this paper, we consider a two-level supply chain
(see Fig. 1) consisting of a sole supplier at the higher
echelon and many identical retailers at the lower
echelon demanding for a consumable item at the
supplier. The replenishment policy is a continuous
review policy (R, Q) at all installations, which means
that, when the inventory position reaches a predeter-
mined value of R, an order size Q is placed. The
demand process is assumed to followa Poisson distribu-
tion. The use of a Poisson process to model the demand
for both high-value capital goods and spare parts is
reasonable because, in both cases, demands will almost
certainly be for a single unit. The supplier is notied
immediately when a retailer order is placed. The out-
side warehouse is assumed always to have sufcient
stock to meet the supplier orders.
Fig. 1 A two-level distribution inventory system
Unsatised demand at a retailer is lost. Technically,
this may mean either that a demand is lost to the system
or that it is expedited (satised, but by means external
to the normal replenishment system). Orders on the
supplier are met on a rst-come, rst-served basis. We
assume throughout that 0 R < Qfor all installations,
with the consequence that not more than one order
from any given retailer may be outstanding at any time.
The technical reason for making this assumption is that
the continuous review lost sales model with 1 Q < R
has no known analytical solution. Our assumption is
reasonable if Qis much greater than the mean demand
in lead time at a retailer.
We allow the procurement lead time for the re-
tailer orders to be a random variable with a general
probabilistic distribution. The transportation time for
replenishing the supplier orders from an external ware-
house is assumed to be constant, and we assume that
the unsatised retailers orders are backordered in the
supplier and all backordered orders lled according to
a rst-in-rst-out policy. The reorder point and batch
size of the supplier are assumed to be integer multiples
of the retailers identical batch size.
1.1 Literature review
One of the oldest papers in the eld of continuous
review multi-echelon inventory systems is a basic and
famous one written by Sherbrooke [16] in 1968. He
assumed (S 1, S) policies in a depot-base system for a
repairable item in the American air force and approxi-
mated the average unit years of inventory and stockout
in the bases. Continuous review models of multi-
echelon inventory systems in the 1980s concentrated
more on repairable items in a depot-base system than
on consumable items. For example, Graves [9] worked
on determining stocking levels in such a system, Moin-
zadeh and Lee [13] considered the issue of determining
the optimal order batch size and stocking levels at the
stocking locations by using a power approximation, Lee
and Moinzadeh [12] generalized previous models on
multi-echelon repairable inventory systems to cover the
cases of batch ordering and batch shipment. On con-
sumable items, Deuermeyer and Schwarz [8] proposed
a simple approximation for a complex, multi-echelon
system (one warehouse and multiple retailers) assum-
ing backordering of unsatised demands in all installa-
tions with a batch ordering policy. Svoronos and Zipkin
[18] proposed several renements on the latter paper
by considering a second-moment information (standard
deviation as well as mean) in their approximations.
In the 1990s, Axster [3] provided a simple recursive
procedure for determining the holding and stockout
1210 Int J Adv Manuf Technol (2009) 41:12081220
costs of a system consisting of one central warehouse
and multiple retailers with (S 1, S) policy, indepen-
dent Poisson demands in the retailers, and backordered
demand during stockouts in all installations and con-
stant lead times. Axster [4] proposed exact and ap-
proximate methods for evaluating the previous system
for the case of a general batch size in all installations
but with identical retailers. For the case of nonidentical
retailers and a general batch size, Axster [5] proposed
methods for the exact evaluation of a two-retailers
case and approximate evaluation for the case of more
than two retailers. The common assumptions of the
above papers are that demand during stockout in the
retailers is backordered. However, on some situations,
demands may be lost. Anderson and Melchiors [2]
have proposed an approximate method for the case of
lost sales when inventory control policy is (S 1, S) in
all installations (one warehouse and multiple retailers)
and unsatised demands are lost in the retailers. Yang
and Wee [21] have developed a single-vendor, multiple-
buyers production inventory model for deteriorating
items. They obtained that optimal policy for the inte-
grated system results in an impressive cost reduction
when it is compared with the independent decisions
made by the vendor and the buyers. Axster [6] has
developed an approximate method for a two-level dis-
tribution inventory system with normal approximations
for both the retailer demand and the demand at the
warehouse, i.e., the orders from the retailers. Wee and
Yang [19] have obtained optimal and heuristic solu-
tions for a multi-echelon distribution network using the
principle of strategic partnership. Here, supply chain
integration involves joint decision making among the
producer, the distributors, and the retailers. Axster [7]
has developed a decentralized, two-echelon inventory
control for the paper Axster [6]. Wee et al. [20] have
examined and pointed out two possible aws in the
total cost function of Yang and Wees [21] model. They
illustrated a proposal to eradicate the aws. Hill et al.
[11] considered a two-echelon inventory model with
(SQ, (S 1)Q) policy at the warehouse and continuous
review (R, Q) policy with lost sales at the retailers.
Olsson and Hill [15] developed an inventory model with
a number of retailers that operate under base stock
policy and supplier functions as an M/D/1 queue.
2 Preliminaries and notations
The batch replenishment quantity Q is xed for all
installations, having been determined by packaging and
handling constraints. For example, Q may represent
the number of units packed into a carton or a shrink-
wrapped palletthe carton or pallet representing the
product unit of purchase, storage, handling, and ship-
ment purposes. The objective is to nd the optimal
reorder points by minimizing the total holding costs
(of the supplier and retailers) and the stockout costs
of retailers. The results that are needed for solving the
proposed model are presented in Sections 2.1 and 2.2.
Let us introduce the following notations.
N Number of retailers

r
Demand rate at a retailer

0
Demand rate at the supplier

r
Random variable represents the lead time for
deliveries of retailers orders from the supplier

0
Constant transportation time for deliveries of
supplier orders from an external warehouse
Q
0
Batch size of the supplier
Q
r
Batch size of a retailer
R
0
Reorder point of the supplier (integer value, mul-
tiple of Q
r
)
R
r
Reorder point of a retailer (integer value, since
demand is one at a time)
h
r
Holding cost per unit per unit time at a retailer
h
0
Holding cost per unit per unit time at the supplier

r
Penalty cost per unit of lost sale at a retailer
C
r
Cost per unit time of a retailer in steady state
C
0
Cost per unit time of the supplier in steady state
TC Total cost of the inventory system per unit in
steady state
2.1 Exact solution for the backordering problem
with Poisson demands
Considering a single-echelon inventory system with
continuous review policy, reorder point of R and batch
size of Q, constant lead time for replenishing orders,
demand generated by a Poisson process, and backo-
rdered demand during a stockout, Hadley and Whitin
[10] developed formulae for the average stock level
D(Q, R) and for the average stockout level B(Q, R).
Assuming linear unit costs of holding and stockout,
they obtained the related annual cost. We briey review
their results and introduce the parameters that they
have used in their formulae since we will use them later.
B(Q, R) =
1
Q
_
(R) (R + Q)
_
(1)
(v) =
(L)
2
2
P(v 1; L) (L)vP(v; L)
+
v(v + 1)
2
P(v + 1; L) (2)
Int J Adv Manuf Technol (2009) 41:12081220 1211
D(Q, R) =
Q+ 1
2
+ R L + B(Q, R) (3)
P(x; L) =

i=x
e
L
(L)
i
/i! x = 0, 1, 2, 3....
where
Q Ordering batch size of continuous review policy
R Reorder point of continuous review policy
Demand rate (mean of Poisson demand
distribution)
2.2 Exact solution for lost sales problem with Poisson
demand and stochastic lead time
Consider an inventory system with continuous review
control policy, reorder point of R, and batch size of
Q. Demands are assumed to be Poisson with rate
> 0, and one unit is demanded at a time. Demands
not covered immediately from the inventory are lost.
The replenishment lead times are independent and
distributed as a generic random variable L. Denote
E[L] by and let X be a random variable, which
is distributed as a lead time demand, i.e., X takes
the value x with probability p(x) = E[ p(x; L)], where
p(x, t) =

i=x
e
t
(t)
i
i!
, x = 0, 1, 2, ... and t > 0. The in-
ventory cycles are dened as the time period between
two successive replenishment orders, and we assume
that, at most, one order is outstanding at a time (i.e.,
R < Q ). We review the exact solutions for this model
from Johansen and Thorstenson paper [17].
When R is the xed reorder point, the expected
number of lost sales during one cycle
U(R) =

x=R
Pr [X > x] =
R1

x=0
Pr [X > x] (4)
The last term of Eq. 4 vanishes if R = 0. Expected time
during which the system is in out of stock is given by

T =
U(R)

, > 0 (5)
Expected number of cycles is

Q+

T
(6)
Average number of lost sales incurred per year is
E =

Q+

T
_

T
_
(7)
Average stock level is
D =
Q
Q+U(R)
_
R +
Q+ 1
2
+U(R)
_
(8)
3 Modeling
In this section, we are going to approximate the demand
process in the higher echelon and retailers lead time in
the lower echelon.
3.1 Approximating the demand process
in the higher echelon
The average number of cycles per unit time in a con-
tinuous review inventory system when demand is lost
during a stockout is

Q+

T
(Hadley and Whitin [10])
without any special assumption concerning the nature
of stochastic processes generating demands and lead
times in steady state. Formula Eq. 6 is just a special case
of this relation when the stochastic process generating
demands is Poisson and the lead times are stochastic.
Since a batch size of Q is ordered in each cycle, the
mean rate of demand (from this inventory system to a
higher echelon) will be

Q+

T
in terms of the identical
batch size of Q.
As Moinzadeh and Lee [13] mentioned, when the
stockout is backordered in the retailers and the demand
process at each retailer is Poisson, the arrival process
of orders at the supplier is a superposition of N arrival
processes in the case of one supplier and N retailers,
each interarrival time is Erlang distributed with shape
parameter Q. When the number of retailers in the
model is large, the arrival process can be well approxi-
mated by a Poisson process with mean rate

N
i=1

i
Q
(
i
is the demand rate at the retailer i and Q is the iden-
tical batch size of the retailers). They also stress that
such an approximation has been used or suggested by
Muckstadt [14], Deuermeyer and Schwarz [8], Albin
[1], and Zipkin [22]. Using the spirit of this nice approx-
imation and extending it for the case of lost sales (with
identical retailers) when the number of retailers in the
model is large, the arrival process can be well approx-
imated by a Poisson process with mean rate
N
r
Q
r
+
r

T
r
in
terms of the identical batch size of Q
r
, noting that

T
r
is the expected length of time per cycle that a retailer
is out of stock. The accuracy of this approximation is
assessed through simulation in the numerical example.
3.2 Retailers lead time approximation
As mentioned before, retailers at the rst echelon of
the model experience independent Poisson demand
processes. Demand during a stockout is assumed to be
lost. Each order, i.e., placed at the supplier by each
retailer, will have a minimum lead time, which is equal
to E[L
r
] =
r
. Since some of the retailers orders are
1212 Int J Adv Manuf Technol (2009) 41:12081220
placed when there is a stockout at the supplier, the
lead time may be more than
r
. So, the lead time of
the retailer can be approximated by adding an addi-
tional waiting time, which results from a stockout in
the supplier to
r
. This waiting time does not have
any clear distribution, and we just know that it is zero
when orders do not incur stockouts in the supplier and
has a positive value when they are backordered in the
supplier.
Based on the approximation of demand at the sup-
plier described in Section 3.1, the supplier behaves
just like an inventory system of the type described in
Section 2.1. From Littles formula in queuing theory
(as Anderson and Melchiors [2] used it in their approx-
imation), we can use the expression for the average
stockout level given by Eq. 1 to obtain the average
waiting time of each retailer order as given by Eq. 9.
It should be noticed that formula Eq. 1 is valid when
customer demands occur one at a time. Since each
retailer orders a batch size of Q
r
, we can still use the
formula Eq. 1 if we make the additional assumption
that the batch size and reorder point of the supplier
(Q
0
, R
0
) are integer multiples of identical batch size of
the retailer (Q
r
). The average waiting time of the orders
placed by the identical retailers is
w =
B(Q
0
/Q
r
, R
0
/Q
r
)

0
(9)

0
=
N
r
Q
r
+
r

T
r
(10)
In the above formula, w is the average waiting time
of the orders placed by the identical retailers. Based on
our approximation, w is added to
r
of each retailer
to approximate lead time for the orders, and it can
be used for evaluating the retailers costs (holding and
stockout costs). Formula Eq. 10 follows directly from
the result in Section 3.1. Our experience may show
that the effect of this approximation is negligible on
0
.
However, our numerical results will show how accurate
this approximation is.
4 Total cost of the two-echelon inventory system
Based on the results of the previous sections, the total
cost of holding and shortage in the retailers and holding
cost in the supplier are as follows:
TC = C
0
+ NC
r
(11)
The supplier cost consists of holding cost as follows:
C
0
= h
0
D(Q
0
/Q
r
, R
0
/Q
r
)Q
r
(12)
In the above formula, D(Q
0
/Q
r
, R
0
/Q
r
) is the average
stock level in the supplier and is obtained as follows,
using formula Eq. 3 in Section 2.1 and noting that Q
0
should be an integer multiple of Q
r
:
The average stock level in the supplier is
D(Q
0
/Q
r
, R
0
/Q
r
) =
1
2
_
Q
0
Q
r
+ 1
_
+
R
0
Q
r

0
L
0
+B(Q
0
/Q
r
, R
0
/Q
r
) (13)
In formula Eq. 13,
0
is obtained through formula
Eq. 14 (as was explained in Section 3.1). B(Q
0
/Q
r
,
R
0
/Q
r
) is the average backorder level in the sup-
plier in terms of Q
r
and is obtained as in formula
Eq. 15 using formulae Eqs. 1 and 2 from Section 2.1,
noting again that Q
0
should be an integer multiple
of Q
r
:

0
=
N
r
Q
r
+
r

T
r
(14)
B(Q
0
/Q
r
, R
0
/Q
r
) =
1
Q
0
/Q
r
_
(R
0
/Q
r
)

_
R
0
+ Q
0
Q
r
__
(15)
where
(v) = (
0
L
0
)
2
P(v 1;
0
L
0
) (
0
L
0
)vP(v;
0
L
0
)
+
v(v + 1)
2
P(v + 1;
0
L
0
) (16)
Knowing
0
and B(Q
0
/Q
r
, R
0
/Q
r
), we can determine
w from formula Eq. 17, as explained in Section 3.2,
w =
B(Q
0
/Q
r
, R
0
/Q
r
)

0
(17)
An important point in our approximation is that w
is dependent on
0
and
0
itself depends on

T
r
=
U(R
r
)/
r
as in the formula Eq. 5 in Section 2.2.
Each retailer cost consists of shortage cost and hold-
ing cost as follows:
C
r
=
r
E
r
+h
r
D
r
In the above formula, E
r
is the average number of
lost sales incurred per unit time in a retailer and D
r
is the average stock level in a retailer. Based on the
approximation in Section 3.2,
r
+ w is approximated
Int J Adv Manuf Technol (2009) 41:12081220 1213
lead time in a retailer. Hence, using formulae Eqs. 7
and 8 in Section 2.2, we have
E
r
=

r
Q
r
+
r

T

r
_

r

T

r
_
(18)
D
r
=
Q
r
Q
r
+
r

T

_
R
r
+
Q
r
+ 1
2

r
(
r
+ w) +
r

T

r
_
(19)

r
= (
r
+ w)
1

r
R
r
1

x=0
Pr [X > x] (20)
and
p(x) = E
_
p(x; L
r
+ w)
_
p(x; t) =
e
t
(t)
x
x!
, x = 0, 1, 2, ....

r
is the average length of time per cycle for which a
retailer is out of stock when expected lead time is
r
+ w.
It is clear that we should nd the optimal values
of the reorder points of all installations through min-
imizing TC, which is nonlinear function with integer
variables of R
0
and R
r
. It is necessary to state that
the reorder point of a retailer is bound by zero and
Q
r
, i.e., 0 R
r
< Q
r
. There should not be more than
one order outstanding in each retailer at any time, and
this constraint satises this condition for a continuous
review inventory system with lost sales (Hadley and
Whitin [10]). Furthermore, since there are N retailers
in our model and none of them can have more than
one order outstanding, we have R
0
NQ
r
. This is
because if R
0
< NQ
r
, then the reorder point is never
reached in the supplier. Summarizing the above, our
optimization problem is
Minimize TC(R
0
, R
r
)
subject to
0 R
r
< Q
r
R
0
NQ
r
(21)
Noting the above explanations, we can easily nd the
local optimal of the approximate total cost TC by using
the optimization tool box in Matlab 7.0.
5 Numerical examples
In order to determine the power of our approximation,
we have designed a set of 18 numerical problems. To
the best of our knowledge, in a two-echelon inventory
system with (R, Q) policy in all installations, no work
has been done on the case of lost sales and stochastic
lead times for the retailers orders in the supplier with
the policy of batch ordering in all installations. Since
we did not have any previous numerical problems as
a reference to compare our approximated results with
existing research work, we developed a problem set
that offered a reasonable range of model parameters. It
is necessary to mention that the optimal reorder points
of the supplier and retailers were found by solving the
optimization problem via the optimization tool box in
Matlab 7.0.
We also simulate each numerical problem 20 times,
for the optimal reorder points obtained from the ap-
proximate model, using GPSS/H simulation software.
The simulation time length of each run is 110,000 unit
times as a run inperiod. Different starting random
number seeds were employed for each problem. All
of the results show that this length of time is sufcient
for the system to reach a steady state. This is also clear
from the standard deviation of the total system cost.
The cost error is obtained by the following relation:
Cost error =
|simulated total cost - approximated total cost|
simulated total cost
The average stock in transit between supplier and
retailer is not of interest in the backorder model be-
cause all demand is met, and so, this represents a xed
overhead that does not depend on the control policy. It
is, however, of interest in the lost sales model because
it is directly proportional to service level. We report
the retailers service levels as their ready rate (the
fraction of time with positive stock on hand). This can
be obtained through Hadley and Whitin [10] as
Service level =1
Average number of lost sales per unit time
Average number of demands per unit time
The above relation has been employed in the ap-
proximation, and also in the simulation, model to nd
the service levels. The numerical problems are as in
Table 1. The number of retailers is considered 10 and
20 (a large enough number to approximate the demand
distribution as Poisson in the supplier as Moinzadeh
and Lee [13]) to compare the results when the number
of retailers is changed. The holding costs of the supplier
and retailers per unit time are assumed to be 1, h
0
=
h
r
= 1 and L
0
= 1.
We assume that the lead time L
r
of retailer is gamma
distributed with mean
r
and variance (
r
)
2
/ where
is positive. The parameter affects the shape of the
lead time distribution. For example, = 1, exponential
distribution is obtained and for , the degenerate
1214 Int J Adv Manuf Technol (2009) 41:12081220
Table 1 Input data
No
r
Q
0
Q
r

r
1 50 16 8 0.5
2 50 16 8 1.0
3 50 16 8 1.5
4 50 16 16 0.5
5 50 16 16 1.0
6 50 16 16 1.5
7 50 32 8 0.5
8 50 32 8 1.0
9 50 32 8 1.5
10 50 32 16 0.5
11 50 32 16 1.0
12 50 32 16 1.5
13 50 64 8 0.5
14 50 64 8 1.0
15 50 64 8 1.5
16 50 64 16 0.5
17 50 64 16 1.0
18 50 64 16 1.5
case of the distribution for constant lead time results.
The Erlang distribution is obtained when is an inte-
ger. Since the demands in the retailers are Poisson with
rate
r
, the lead time demand (X) in the retailer has
a negative binomial distribution. Hence, X takes the
value x with probability
p(x) =
_
+ x 1
x
__

+
r

r
_

_

r

r
+
r

r
_
x
,
x = 0, 1, 2, .... and > 0 (22)
The mean lead time demand in the retailer is E[X] =

r
, and variance of X is

r

r
1+
r

r
/
. Since the geometric
distribution is the special case of the negative binomial
distribution with = 1, exponentially distributed lead
times result in X having a geometric distribution with
mean
r

r
and variance

r

r
1+
r

r
. The limiting case of
, corresponding to constant lead time, implies
that X has the Poisson distribution with mean
r

r
and
variance
r

r
. Using Eq. 22, we can nd out U(R
r
) as in
formula Eq. 4.
Tables 2, 3, 4, and 5 give the optimum results for
total cost, service level, and average waiting time for
the cases of 10 and 20 retailers. The tables show the
results when = 1, 3, 5, and 10. In order to evaluate
the supply chain structures, we consider the following
parameters (consider problem no. 7 in Table 1): Q
0
=
32, Q
r
= 8,
r
= 0.5,
r
= 50, h
0
= h
r
= 1, L
0
= 1, and
{1, 3, 5, 10, 100, 1000, 10000}, and numerical results
are obtained in Table 6.
5.1 Comment on the numerical examples
With regards to the obtained results from Tables 2 to 6,
we observe the following phenomena:
1. As can be seen from Tables 2 to 5, the errors in
the approximate total cost, average waiting time,
and service level are small in comparison with the
simulated values.
2. As N increases, the mean error decreases in each
table. This implies that if the number of retailers
increases, then the demand process at the supplier
can be well approximated by Poisson process (it is
also very clear from Moinzadeh and Lee [13]).
3. When
r
increases, TC increases. It implies that
a higher value of demand rate at the retailers,
resulting in a shorter replenishment time, increases
the optimal total cost of the supply chain.
4. When Q
0
increases, TCincreases. This implies that,
although the increase in ordering quantity of the
supplier decreases stockout level at a retailer, be-
cause of a relatively high penalty cost (
r
) com-
pared to unit holding cost, the expected total cost
increases.
5. From Table 6, we observe that, as decreases, TC
increases. This implies that, as expected variations
(
2
r
/ ) in lead time at a retailer increase, the total
cost of the supply chain increases. The reason for
this phenomenon is as follows: When lead time
parameter decreases, it increases the expected
variations in the order processing time (i.e., lead
time) at a retailer. Simultaneously, it increases the
stockout level at the lower echelon, and hence, the
total cost increases.
6. Table 6 shows that as increases the service level
increases, but the average waiting time of a retailer
decreases. This implies that shortage cost per unit
lost sale has to be raised at a retailer if the service
level is to be kept constant when the variation in
lead time demand increases due to an increase in
waiting time variation. The reason for this effect
is as follows: When increases, it decreases the
stockout level at the retailers. Decrease in stockout
level at the lower echelon decreases the average
waiting time of orders, and therefore, service level
increases.
From observations 5 and 6, we see that the varia-
tion in lead time of retailer orders disturbs the supply
chain structures. The following observation is made
intuitively:
7. The lost sales model can give a higher service level
than the backorder level because the loss of sales
Int J Adv Manuf Technol (2009) 41:12081220 1215
T
a
b
l
e
2
T
o
t
a
l
c
o
s
t
,
s
e
r
v
i
c
e
l
e
v
e
l
,
a
n
d
w
a
i
t
i
n
g
t
i
m
e
r
e
s
u
l
t
s
w
h
e
n

r
=
5
0
a
n
d

=
1
N
o
.
A
p
p
r
S
i
m
u
E
r
r
o
r
(
%
)
S
e
r
v
i
c
e
l
e
v
e
l
(
%
)
E
r
r
o
r
(
%
)
A
W
T
E
r
r
o
r
(
%
)
R
0
R
r
T
o
t
a
l
c
o
s
t
M
e
a
n
o
f
t
o
t
a
l
c
o
s
t
S
t
.
d
e
v
.
A
p
p
r
S
i
m
u
A
p
p
r
S
i
m
u
1

8
0
6
6
.
0
9
6
7
.
0
2
0
.
1
0
1
.
3
8
7
6
9
0
.
7
9
9
0
.
0
8
0
.
7
8
8
2
0
.
6
2
0
0
0
.
6
2
1
7
0
.
2
7
3
4
2
8
1
1
0
3
.
3
2
1
0
4
.
0
6
0
.
0
5
0
.
7
1
1
1
9
1
.
1
2
9
2
.
4
4
1
.
4
2
8
0
0
.
0
8
0
1
0
.
0
8
1
2
1
.
3
5
4
7
3
1
6
2
1
5
2
.
6
7
1
5
6
.
2
4
0
.
0
8
2
.
2
8
4
9
8
7
.
6
0
8
6
.
1
4
1
.
6
9
4
9
0
.
0
4
7
8
0
.
0
4
6
2
3
.
4
6
3
2
4

1
6
3
9
4
.
7
0
9
5
.
7
2
0
.
0
7
1
.
0
6
5
6
9
4
.
1
1
9
4
.
9
7
0
.
9
0
5
5
1
.
0
0
0
0
1
.
0
0
2
3
0
.
2
2
9
5
5

8
1
1
2
1
.
4
3
1
2
8
.
4
2
0
.
1
2
5
.
4
4
3
1
9
3
.
4
1
9
4
.
3
3
0
.
9
7
5
3
0
.
5
0
8
8
0
.
5
1
0
4
0
.
3
1
3
5
6
8
3
1
5
1
.
0
0
1
5
7
.
2
6
0
.
0
5
3
.
9
8
0
7
9
2
.
6
4
9
2
.
9
9
0
.
3
7
6
4
0
.
1
5
7
6
0
.
1
5
3
1
2
.
9
3
9
3
7

8
5
6
8
.
3
6
6
9
.
8
8
0
.
0
9
2
.
1
7
5
2
9
5
.
9
9
9
6
.
5
2
0
.
5
4
9
1
2
.
3
2
5
9
2
.
3
2
4
9
0
.
3
0
7
8
8
0
6
1
0
6
.
1
1
1
0
9
.
3
8
0
.
0
6
2
.
9
8
9
6
9
0
.
6
0
9
1
.
4
7
0
.
9
5
1
1
0
.
1
4
3
0
0
.
1
4
5
6
1
.
7
8
5
7
9
8
3
1
5
5
.
5
9
1
5
8
.
9
8
0
.
0
3
2
.
1
3
2
3
8
7
.
2
3
8
6
.
9
8
0
.
2
8
7
4
0
.
0
7
8
6
0
.
0
7
7
6
1
.
2
8
8
7
1
0

1
6
0
9
8
.
6
9
9
9
.
7
8
0
.
0
6
1
.
0
9
2
4
9
5
.
3
2
9
7
.
2
9
2
.
0
2
4
9
0
.
5
6
8
7
0
.
5
6
6
7
0
.
3
5
2
9
1
1

8
5
1
2
3
.
7
3
1
2
5
.
8
6
0
.
0
5
1
.
6
9
2
4
9
4
.
3
8
9
1
.
0
6
3
.
6
4
5
9
0
.
2
9
3
1
0
.
3
0
3
7
3
.
4
9
0
3
1
2
0
6
1
5
3
.
9
4
1
5
7
.
0
6
0
.
0
2
1
.
9
8
6
5
9
2
.
2
9
9
5
.
6
8
3
.
5
4
3
1
0
.
2
0
9
6
0
.
2
0
8
1
0
.
7
2
0
8
1
3

2
4
3
7
7
.
0
7
7
9
.
2
4
0
.
0
7
2
.
7
3
8
5
9
2
.
9
5
9
5
.
2
6
2
.
4
2
4
9
1
.
0
2
9
0
1
.
1
9
1
8
1
3
.
6
6
0
0
1
4

8
4
1
1
6
.
4
9
1
1
9
.
2
1
0
.
1
2
.
2
8
1
7
9
0
.
1
6
8
8
.
8
3
1
.
4
9
7
2
0
.
1
9
6
7
0
.
1
9
3
2
1
.
8
1
1
6
1
5
0
2
1
6
5
.
9
3
1
7
3
.
9
2
0
.
0
9
4
.
5
9
4
1
8
6
.
9
4
9
0
.
0
1
3
.
4
1
0
7
0
.
1
0
3
3
0
.
1
0
3
2
0
.
0
9
6
9
1
6

4
0
3
1
0
5
.
7
9
1
0
9
.
2
8
0
.
1
3
3
.
1
9
3
6
9
3
.
6
9
9
4
.
4
2
0
.
7
7
3
1
2
.
1
8
5
4
2
.
3
2
6
1
6
.
0
4
8
8
1
7

2
4
1
1
3
2
.
0
2
1
3
6
.
2
4
0
.
0
6
3
.
0
9
7
5
9
2
.
9
6
8
9
.
9
4
3
.
3
5
7
8
0
.
6
0
6
7
0
.
6
0
3
1
0
.
5
9
6
9
1
8

8
3
1
6
1
.
3
5
1
6
3
.
0
6
0
.
0
8
1
.
0
4
8
7
9
2
.
3
6
9
2
.
8
7
0
.
5
4
9
2
0
.
1
9
9
2
0
.
1
9
3
2
3
.
1
0
5
6
F
o
r
t
h
e
c
a
s
e
o
f
1
0
r
e
t
a
i
l
e
r
s
M
e
a
n
2
.
4
3
8
6
9
1
.
9
1
8
9
9
2
.
2
9
3
3
1
.
6
2
1
3

2
.
3
2
4
4
S
t
.
d
e
v
i
1
.
2
8
6
6
2
.
6
5
1
7
3
.
2
1
5
0
1
.
1
7
0
6

3
.
2
5
1
1
1

4
8
1
1
5
3
.
9
7
1
5
4
.
6
8
0
.
0
8
0
.
4
5
9
0
9
2
.
3
0
9
2
.
8
7
0
.
6
1
3
8
1
.
1
8
2
9
1
.
1
8
3
0
0
.
0
0
8
5
2

1
6
5
2
3
1
.
3
9
2
3
3
.
0
8
0
.
0
6
0
.
7
2
5
1
8
9
.
9
3
9
2
.
8
2
3
.
1
1
3
6
0
.
2
2
3
9
0
.
2
2
2
9
0
.
4
4
8
6
3
8
6
3
2
7
.
4
8
3
2
8
.
5
1
0
.
0
3
0
.
3
1
3
5
8
7
.
4
7
9
0
.
4
2
3
.
2
6
2
6
0
.
0
5
9
0
0
.
0
5
9
1
0
.
1
6
9
2
4

8
0
3
2
1
1
.
5
3
2
1
2
.
8
8
0
.
1
0
0
.
6
3
4
2
9
2
.
5
3
8
6
.
9
8
6
.
3
8
0
8
2
.
6
8
9
9
2
.
6
8
8
7
0
.
0
4
4
6
5

3
2
4
2
6
2
.
6
5
2
6
3
.
5
6
0
.
1
2
0
.
3
4
5
3
9
3
.
7
7
9
2
.
8
7
0
.
9
6
9
1
0
.
4
2
9
2
0
.
4
2
8
8
0
.
0
9
3
3
6

8
5
3
1
8
.
6
5
3
1
9
.
3
7
0
.
0
2
0
.
2
2
5
4
9
2
.
6
0
9
3
.
9
8
1
.
4
6
8
4
0
.
1
6
3
4
0
.
1
6
3
2
0
.
1
2
2
5
7

9
6
5
1
9
2
.
3
1
1
9
4
.
2
8
0
.
0
1
1
.
0
1
4
0
8
8
.
5
3
9
1
.
9
4
3
.
7
0
8
9
2
.
0
8
7
7
2
.
0
8
3
6
0
.
1
9
6
8
8

4
0
1
2
8
0
.
5
7
2
8
1
.
4
3
0
.
0
7
0
.
3
0
5
6
8
9
.
0
8
8
8
.
8
6
0
.
2
4
7
6
0
.
3
2
8
3
0
.
3
2
7
1
0
.
3
6
6
9
9

1
6
1
3
7
9
.
6
8
3
8
0
.
5
4
0
.
1
1
0
.
2
2
6
0
8
6
.
7
0
8
9
.
6
4
3
.
2
7
9
8
0
.
1
2
3
6
0
.
1
2
3
7
0
.
0
8
0
8
1
0

1
6
0
7
2
3
5
.
6
0
2
3
6
.
4
9
0
.
0
7
0
.
3
7
6
3
8
6
.
9
7
8
7
.
1
4
0
.
1
9
5
1
5
.
2
0
3
9
5
.
2
0
6
1
0
.
0
4
2
3
1
1

8
0
1
3
0
2
.
1
6
3
0
3
.
2
0
0
.
0
3
0
.
3
4
3
0
9
1
.
7
7
8
7
.
2
6
5
.
1
6
8
5
0
.
8
7
2
3
0
.
8
7
2
1
0
.
0
2
2
9
1
2

4
0
6
3
6
3
.
6
3
3
6
4
.
5
6
0
.
0
4
0
.
2
5
5
1
9
1
.
8
3
9
2
.
1
6
0
.
3
5
8
1
0
.
2
7
7
9
0
.
2
7
8
1
0
.
0
7
1
9
1
3

1
6
0
7
2
7
2
.
7
4
2
7
3
.
8
2
0
.
0
8
0
.
3
9
4
4
8
5
.
8
4
8
5
.
8
7
0
.
0
3
4
9
2
.
7
5
9
9
2
.
7
5
3
1
0
.
2
4
7
0
1
4

9
6
1
3
8
0
.
9
7
3
8
1
.
7
8
0
.
0
9
0
.
2
1
2
2
8
6
.
4
8
8
6
.
9
5
0
.
5
4
0
5
0
.
6
5
4
8
0
.
6
5
3
2
0
.
2
4
4
9
1
5

4
8
2
4
8
8
.
2
1
4
8
9
.
3
6
0
.
1
2
0
.
2
3
5
0
8
5
.
9
1
8
5
.
9
3
0
.
0
2
3
3
0
.
1
9
0
5
0
.
1
9
0
6
0
.
0
5
2
5
1
6

3
2
0
3
2
8
2
.
9
5
2
8
4
.
8
7
0
.
1
3
0
.
6
7
4
0
7
7
.
0
8
7
6
.
9
8
0
.
1
2
9
9
1
0
.
2
6
2
1
1
0
.
2
5
1
2
0
.
1
0
6
3
1
7

1
7
6
1
3
8
3
.
2
5
3
8
4
.
6
3
0
.
1
5
0
.
3
5
8
8
8
7
.
6
9
8
6
.
6
7
1
.
1
7
6
9
1
.
7
9
9
0
1
.
7
9
6
2
0
.
1
5
5
9
1
8

1
0
4
3
4
5
8
.
2
0
4
5
9
.
4
1
0
.
0
7
0
.
2
6
3
4
9
0
.
0
1
9
0
.
0
4
0
.
0
3
3
3
0
.
5
4
8
5
0
.
5
4
7
8
0
.
1
2
7
8
F
o
r
t
h
e
c
a
s
e
o
f
2
0
r
e
t
a
i
l
e
r
s
M
e
a
n
0
.
4
0
8
9
8
8
.
6
9
3
9
8
8
.
8
5
4
4
1
.
7
0
5
8

0
.
1
4
4
6
S
t
.
d
e
v
i
0
.
2
1
7
3
3
.
8
8
5
2
4
.
0
2
2
2
1
.
9
5
9
9

0
.
1
1
9
4
A
p
p
r
=
A
p
p
r
o
x
i
m
a
t
e
d
,
S
i
m
u
=
S
i
m
u
l
a
t
e
d
,
A
W
T
=
A
v
e
r
a
g
e
w
a
i
t
i
n
g
t
i
m
e
1216 Int J Adv Manuf Technol (2009) 41:12081220
T
a
b
l
e
3
T
o
t
a
l
c
o
s
t
,
s
e
r
v
i
c
e
l
e
v
e
l
,
a
n
d
w
a
i
t
i
n
g
t
i
m
e
r
e
s
u
l
t
s
w
h
e
n

r
=
5
0
a
n
d

=
3
N
o
.
A
p
p
r
S
i
m
u
E
r
r
o
r
(
%
)
S
e
r
v
i
c
e
l
e
v
e
l
(
%
)
E
r
r
o
r
(
%
)
A
W
T
E
r
r
o
r
(
%
)
R
0
R
r
T
o
t
a
l
c
o
s
t
M
e
a
n
o
f
t
o
t
a
l
c
o
s
t
S
t
.
d
e
v
.
A
p
p
r
S
i
m
u
A
p
p
r
S
i
m
u
1
0
1
6
1
.
7
9
6
2
.
1
2
0
.
0
7
0
.
5
3
1
2
9
7
.
1
3
9
9
.
2
4
2
.
1
2
6
2
0
.
5
6
9
0
0
.
5
6
3
2
1
.
0
2
9
8
2
1
6
2
7
9
.
2
9
8
0
.
3
4
0
.
0
3
1
.
3
0
6
9
9
7
.
5
6
9
6
.
3
1
1
.
2
9
7
9
0
.
2
2
7
5
0
.
2
2
7
1
0
.
1
7
6
1
3
2
4
4
9
5
.
7
4
9
6
.
8
6
0
.
0
4
1
.
1
5
6
3
9
8
.
0
8
9
9
.
1
4
1
.
0
6
9
2
0
.
2
5
4
1
0
.
2
5
8
1
1
.
5
4
9
8
4

8
0
9
6
.
5
6
9
6
.
5
7
0
.
1
3
0
.
0
1
0
4
9
4
.
9
6
9
3
.
9
7
1
.
0
5
3
5
1
.
0
3
0
8
1
.
0
4
0
8
0
.
9
6
0
8
5
0
2
1
1
3
.
4
8
1
1
3
.
6
5
0
.
1
5
0
.
1
4
9
6
9
6
.
6
5
9
6
.
9
8
0
.
3
4
0
3
0
.
8
5
4
6
0
.
8
5
8
2
0
.
4
1
9
5
6
1
6
3
1
2
9
.
2
8
1
3
0
.
0
2
0
.
0
7
0
.
5
6
9
1
9
7
.
2
7
9
8
.
3
3
1
.
0
7
8
0
0
.
4
9
5
7
0
.
4
8
1
3
2
.
9
9
1
9
7

8
1
6
4
.
3
1
6
8
.
9
4
0
.
1
2
6
.
7
1
6
0
9
6
.
2
5
9
7
.
1
0
0
.
8
7
5
4
1
.
9
0
3
9
1
.
9
1
1
7
0
.
9
6
0
9
8
8
2
8
1
.
3
5
8
4
.
1
8
0
.
0
2
3
.
3
6
1
8
9
7
.
1
8
9
6
.
8
9
0
.
2
9
9
3
0
.
2
9
7
1
0
.
2
9
4
2
0
.
9
8
5
7
9
1
6
4
9
7
.
6
9
1
0
1
.
0
9
0
.
0
1
3
.
3
6
3
3
9
7
.
8
2
9
5
.
8
1
2
.
0
9
7
9
0
.
2
9
9
4
0
.
2
9
1
8
2
.
6
0
4
5
1
0

4
0
1
9
8
.
6
7
9
9
.
9
2
0
.
0
7
1
.
2
5
1
0
9
4
.
3
4
9
0
.
2
6
4
.
5
2
0
3
1
.
2
8
4
8
1
.
2
5
2
1
0
.
6
7
5
4
1
1

8
2
1
1
5
.
2
5
1
1
6
.
0
5
0
.
0
3
0
.
6
8
9
4
9
6
.
3
4
9
8
.
4
8
2
.
1
7
3
0
0
.
4
3
3
4
0
.
9
3
3
9
0
.
9
7
5
5
1
2
1
6
4
1
3
0
.
6
5
1
3
1
.
7
8
0
.
0
4
0
.
8
5
7
5
9
7
.
1
1
9
5
.
1
0
2
.
1
1
3
6
0
.
2
6
5
4
0
.
5
3
0
6
1
.
4
8
5
3
1
3

6
4
1
7
2
.
6
2
7
8
.
5
3
0
.
0
8
7
.
5
2
5
8
9
3
.
3
7
9
0
.
0
7
3
.
6
6
3
8
2
.
3
6
1
2
1
.
5
1
0
8
0
.
1
2
5
6
1
4

2
4
2
8
9
.
3
5
9
9
.
0
8
0
.
0
9
9
.
8
2
0
3
9
7
.
4
9
9
4
.
3
2
3
.
3
6
0
9
0
.
7
2
2
9
0
.
5
6
0
7
0
.
3
0
2
3
1
5
0
3
1
0
4
.
8
6
1
1
5
.
4
6
0
.
0
8
9
.
1
8
0
7
9
7
.
9
3
9
8
.
2
4
0
.
3
1
5
6
0
.
4
6
4
4
0
.
2
8
0
5
3
.
2
0
0
9
1
6

1
0
4
2
1
0
3
.
9
1
1
0
6
.
6
4
0
.
0
9
2
.
5
6
0
0
9
3
.
3
1
9
3
.
8
3
0
.
5
5
4
2
2
.
8
4
5
5
2
.
9
1
0
4
0
.
5
3
6
6
1
7

5
6
3
1
2
0
.
4
0
1
2
6
.
8
9
0
.
1
2
5
.
1
1
4
7
9
6
.
8
4
9
5
.
2
1
1
.
7
1
2
0
0
.
8
5
6
8
0
.
8
0
5
0
1
.
2
9
9
7
1
8

2
4
4
1
3
5
.
8
8
1
4
4
.
3
2
0
.
1
3
5
.
8
4
8
1
9
7
.
4
5
9
5
.
4
7
2
.
0
7
3
9
0
.
4
4
5
3
0
.
4
5
9
5
3
.
7
7
1
5
F
o
r
t
h
e
c
a
s
e
o
f
1
0
r
e
t
a
i
l
e
r
s
M
e
a
n
3
.
3
3
4
0
9
6
.
5
0
4
4
9
5
.
8
1
9
4
1
.
7
0
6
9

1
.
3
3
6
2
S
t
.
d
e
v
i
3
.
2
3
9
1
1
.
5
0
2
0
2
.
6
6
7
2
1
.
2
0
1
2

1
.
0
9
1
5
1

4
0
1
1
4
3
.
7
6
1
4
3
.
9
9
0
.
0
3
0
.
1
5
9
7
9
4
.
3
1
9
4
.
8
3
0
.
5
4
8
3
1
.
2
8
4
8
1
.
2
8
3
5
0
.
1
0
1
3
2

8
2
1
7
5
.
3
2
1
7
6
.
0
4
0
.
0
6
0
.
4
0
9
0
9
6
.
3
6
9
6
.
0
4
0
.
3
3
3
2
0
.
4
3
3
4
0
.
4
3
2
2
0
.
2
7
7
6
3
1
6
4
2
0
4
.
4
4
2
0
4
.
9
8
0
.
0
5
0
.
2
6
3
4
9
8
.
0
2
9
8
.
8
7
0
.
8
5
9
7
0
.
2
6
5
4
0
.
2
6
8
1
1
.
0
0
7
1
4

6
4
1
2
0
6
.
9
1
2
0
7
.
4
2
0
.
0
2
0
.
2
4
5
9
9
4
.
6
0
9
3
.
9
2
0
.
7
2
4
0
2
.
3
6
1
2
2
.
3
6
1
3
0
.
0
0
4
2
5

2
4
2
2
3
7
.
2
9
2
3
9
.
0
6
0
.
0
5
0
.
7
4
0
4
9
7
.
1
4
9
7
.
7
8
0
.
6
5
4
5
0
.
7
2
2
9
0
.
7
2
1
5
0
.
1
9
4
0
6
0
3
2
6
4
.
1
4
2
6
5
.
1
6
0
.
0
9
0
.
3
8
4
7
9
7
.
4
2
9
7
.
9
5
0
.
5
4
1
1
0
.
4
6
4
4
0
.
4
6
2
1
0
.
4
9
7
7
7

1
0
4
2
1
7
7
.
5
5
1
7
9
.
0
6
0
.
0
6
0
.
8
4
3
3
9
1
.
8
1
9
1
.
0
3
0
.
8
5
6
9
1
.
9
4
5
5
1
.
9
2
8
3
0
.
6
0
8
1
8

5
6
3
2
1
4
.
6
3
2
1
4
.
8
9
0
.
0
3
0
.
1
2
1
0
9
5
.
9
7
9
6
.
4
2
0
.
4
6
6
7
0
.
8
5
6
8
0
.
8
5
3
1
0
.
4
3
3
7
9

2
4
4
2
4
5
.
3
7
2
4
6
.
0
3
0
.
0
6
0
.
2
6
8
3
9
6
.
9
1
9
7
.
1
2
0
.
2
1
6
2
0
.
4
4
5
3
0
.
4
4
1
6
0
.
8
3
7
9
1
0

1
5
2
2
2
3
0
.
8
2
2
3
1
.
6
2
0
.
0
8
0
.
3
4
5
4
9
0
.
5
1
4
9
0
.
7
4
0
.
2
5
3
5
5
.
3
0
8
3
5
.
3
0
1
0
0
.
1
3
7
7
1
1

8
8
3
2
7
0
.
3
1
2
7
0
.
9
3
0
.
0
9
0
.
2
2
8
8
9
5
.
9
7
9
5
.
0
2
0
.
9
9
9
8
1
.
4
6
3
7
1
.
4
6
1
0
0
.
1
8
4
8
1
2

5
6
4
2
9
9
.
5
0
3
0
0
.
4
2
0
.
0
8
0
.
3
0
6
2
9
6
.
9
4
9
7
.
1
4
0
.
2
0
5
9
0
.
8
1
7
4
0
.
8
1
3
6
0
.
4
6
7
1
1
3

1
6
0
2
2
5
3
.
9
0
2
5
4
.
0
8
0
.
0
8
0
.
0
7
0
8
9
1
.
0
7
9
2
.
0
4
1
.
0
5
3
9
3
.
0
4
5
1
3
.
0
4
3
6
0
.
0
4
9
3
1
4

1
6
0
5
2
9
5
.
7
6
2
9
6
.
7
8
0
.
0
1
0
.
3
4
3
7
9
4
.
2
1
9
4
.
8
1
0
.
6
3
2
8
1
.
8
9
1
6
1
.
8
9
1
4
0
.
0
1
0
6
1
5

1
1
2
6
3
3
1
.
7
6
3
3
3
.
6
3
0
.
0
7
0
.
5
6
0
5
9
6
.
4
1
9
6
.
1
2
0
.
3
0
1
7
0
.
9
3
4
4
0
.
9
5
0
1
1
.
6
5
2
5
1
6

3
2
0
3
2
7
4
.
8
3
2
7
4
.
9
0
0
.
0
3
0
.
0
2
5
5
8
1
.
9
0
8
2
.
4
4
0
.
6
5
5
0
1
0
.
7
9
2
1
1
0
.
8
1
1
6
0
.
1
8
0
4
1
7

2
0
8
4
3
3
7
.
5
4
3
3
7
.
9
8
0
.
1
3
0
.
1
3
0
2
9
2
.
5
5
9
3
.
0
2
0
.
5
0
5
3
2
.
8
4
2
6
2
.
8
3
2
6
0
.
3
5
3
0
1
8

1
5
2
5
3
7
4
.
5
3
3
7
5
.
4
8
0
.
1
5
0
.
2
5
3
0
9
5
.
5
0
9
6
.
0
8
0
.
6
0
3
7
1
.
3
8
4
9
1
.
3
3
2
7
3
.
9
1
6
9
F
o
r
t
h
e
c
a
s
e
o
f
2
0
r
e
t
a
i
l
e
r
s
M
e
a
n
0
.
3
1
6
7
9
4
.
3
1
1
1
9
4
.
5
2
0
6
0
.
5
7
8
5

0
.
6
0
6
3
S
t
.
d
e
v
i
0
.
2
1
6
4
3
.
8
2
7
5
3
.
8
1
5
1
0
.
2
5
7
7

0
.
9
2
4
6
A
p
p
r
=
A
p
p
r
o
x
i
m
a
t
e
d
,
S
i
m
u
=
S
i
m
u
l
a
t
e
d
,
A
W
T
=
A
v
e
r
a
g
e
w
a
i
t
i
n
g
t
i
m
e
Int J Adv Manuf Technol (2009) 41:12081220 1217
T
a
b
l
e
4
T
o
t
a
l
c
o
s
t
,
s
e
r
v
i
c
e
l
e
v
e
l
,
a
n
d
w
a
i
t
i
n
g
t
i
m
e
r
e
s
u
l
t
s
w
h
e
n

r
=
5
0
a
n
d

=
5
N
o
.
A
p
p
r
S
i
m
u
E
r
r
o
r
(
%
)
S
e
r
v
i
c
e
l
e
v
e
l
(
%
)
E
r
r
o
r
(
%
)
A
W
T
E
r
r
o
r
(
%
)
R
0
R
r
T
o
t
a
l
c
o
s
t
M
e
a
n
o
f
t
o
t
a
l
c
o
s
t
S
t
.
d
e
v
.
A
p
p
r
S
i
m
u
A
p
p
r
S
i
m
u
1

8
1
5
9
.
8
7
6
1
.
0
6
0
.
1
2
1
.
9
4
8
9
9
6
.
1
6
9
4
.
8
2
1
.
4
1
3
2
0
.
6
2
6
4
0
.
6
2
1
6
0
.
7
7
2
2
2
0
2
7
4
.
9
2
7
5
.
0
6
0
.
0
2
0
.
1
8
6
5
9
6
.
6
2
9
7
.
1
4
0
.
5
3
5
3
0
.
2
8
2
5
0
.
2
8
2
1
0
.
1
4
1
8
3
8
3
8
8
.
5
2
8
9
.
4
6
0
.
0
1
1
.
0
5
0
7
9
6
.
9
8
9
5
.
8
7
1
.
1
5
7
8
0
.
1
6
8
5
0
.
1
6
7
0
0
.
8
9
8
2
4

1
6
1
7
7
.
3
0
7
8
.
4
1
0
.
0
7
1
.
4
1
5
6
9
2
.
3
6
9
2
.
8
7
0
.
5
4
9
2
1
.
0
0
0
0
1
.
0
1
3
1
1
.
2
9
3
1
5

8
2
1
0
7
.
2
1
1
0
9
.
9
8
0
.
0
3
2
.
5
1
8
6
9
7
.
5
1
9
8
.
1
4
0
.
6
4
1
9
0
.
5
1
8
6
0
.
5
1
2
1
1
.
2
6
9
3
6
8
1
1
1
8
.
9
6
1
1
9
.
0
3
0
.
0
4
0
.
0
5
8
8
9
6
.
4
3
9
6
.
9
9
0
.
5
7
7
4
0
.
1
6
7
6
0
.
1
6
2
7
3
.
0
1
1
7
7

8
1
5
9
.
9
6
6
3
.
9
2
0
.
0
1
6
.
1
9
5
2
9
8
.
0
6
9
8
.
7
2
0
.
6
6
8
6
1
.
7
3
5
1
1
.
7
4
2
1
2
.
0
4
6
2
8

8
3
7
7
.
4
7
8
4
.
0
3
0
.
0
7
7
.
8
0
6
7
9
7
.
9
9
9
5
.
8
1
2
.
2
7
5
3
0
.
4
0
2
0
.
4
1
0
1
1
.
9
7
5
1
9
0
4
9
0
.
0
8
9
5
.
0
6
0
.
0
3
5
.
2
3
8
8
9
8
.
0
3
9
4
.
2
3
4
.
0
3
2
7
0
.
2
2
9
6
0
.
2
1
1
6
8
.
5
0
6
6
1
0

4
0
1
8
0
.
3
1
8
7
.
9
3
0
.
0
4
8
.
6
6
6
0
9
1
.
4
6
9
2
.
8
2
1
.
4
6
5
2
1
.
4
1
4
3
1
.
4
1
2
1
0
.
1
5
5
8
1
1

4
0
3
1
0
9
.
4
0
1
1
2
.
0
5
0
.
0
8
2
.
3
6
5
0
9
7
.
1
3
9
5
.
1
0
2
.
1
3
4
6
0
.
6
2
7
2
0
.
6
3
1
7
0
.
7
1
2
4
1
2

1
6
4
1
2
1
.
5
4
1
2
4
.
4
8
0
.
0
1
2
.
3
6
1
8
9
7
.
4
9
9
8
.
4
3
0
.
9
5
5
0
0
.
3
9
6
2
0
.
3
3
0
2
1
9
.
9
8
7
9
1
3

7
2
1
7
5
.
0
6
7
6
.
5
3
0
.
0
7
1
.
9
2
0
8
9
5
.
1
8
9
3
.
1
6
2
.
1
6
8
3
0
.
9
6
6
2
1
.
8
4
4
2
4
7
.
6
0
8
7
1
4

4
8
2
8
5
.
9
8
8
7
.
0
8
0
.
0
3
1
.
2
6
3
2
9
6
.
3
6
9
7
.
3
8
1
.
0
4
7
4
0
.
7
5
4
6
0
.
7
5
1
6
0
.
3
9
9
1
1
5

1
6
2
9
8
.
5
7
1
0
4
.
4
6
0
.
0
4
5
.
6
3
8
5
9
7
.
9
7
9
5
.
8
6
2
.
2
0
1
1
0
.
2
4
1
2
0
.
2
3
3
2
3
.
4
3
0
5
1
6

8
8
1
1
0
3
.
5
3
1
0
4
.
4
2
0
.
0
1
0
.
8
5
2
3
9
9
.
7
3
9
6
.
6
1
3
.
2
2
9
5
2
.
1
0
3
4
2
.
2
5
3
1
6
.
6
4
4
2
1
7

9
6
4
1
1
6
.
9
2
1
1
7
.
9
9
0
.
0
7
0
.
9
0
6
9
9
7
.
2
1
9
8
.
2
3
1
.
0
3
8
4
0
.
6
0
4
0
0
.
6
7
3
1
1
0
.
2
6
5
9
1
8

6
4
5
1
2
8
.
3
7
1
3
4
.
3
4
0
.
0
8
4
.
4
4
3
9
9
7
.
6
3
9
7
.
8
1
0
.
1
8
4
0
0
.
3
6
5
1
0
.
3
6
3
9
0
.
3
2
9
8
F
o
r
t
h
e
c
a
s
e
o
f
1
0
r
e
t
a
i
l
e
r
s
M
e
a
n
3
.
0
4
6
6
9
6
.
6
8
3
3
9
6
.
1
1
0
6
1
.
4
5
9
7