THE MPS PARAMETERIZATION UNDER LEAD TIME UNCERTAINTY

Mohamed-Aly OULD-LOULY and Alexandre DOLGUI

Lab. of Industrial Systems Optimization
University of Technology of Troyes
12, Marie Curie st., BP2060,10010 Troyes cedex, France
ph: +33 (0)325715879, fax: +33 (0)325715649, e-mail: {louly, dolgui}@utt.fr


Abstract. The supply planning of assembly systems under lead times uncertainty is
studied. The used criteria is the sum of the average holding cost for the components, the
average backlogging cost for the finished product, and the setup cost. The studied policy
avoids the nervousness, the decision variables are the planned lead times of components
and the periodicity of the periodic ordering quantity (POQ). A mathematical method,
which gives the optimal values for these parameters, is given.

Keywords: Manufacturing systems, Planning, Inventory Control, Optimization.



1. INTRODUCTION

Many works are devoted to the analysis of the customer
demand uncertainties. The most of these works deals
with nervousness (excessive change in production plans).
Nervousness increases costs and decreases the customer
service level (Mather, 1985). Blackburn et al., (1986)
and Sridharan et al., (1988) proposed two different
metrics to measure this instability. These metrics can be
used to study the impact of safety stock on nervousness
(Sridharan and LaForge, 1989).

Several methods dampen the nervousness. Carlson et al.,
(1982) proposed to extend the forecast beyond the
planning horizon to smooth the forecast when the
horizon length isn't well determined. Sridharan and Berry
(1990a, 1990b) propose to freeze a part of the Master
Production Schedule (MPS) to dampen the instability,
their simulations show that the lot-sizing method is the
most important parameter. However, the Kadipasaoglu
and Sridharan (1995) simulations show that freezing a
part of the MPS gives a better stability. Graves et al.
(1998) propose an interesting mathematical formulation
for the nervousness problem. His model, based on
stochastic process, improves demand forecasting, and
reschedules the plan while keeping a desired stability.
Yeung et al. (1999) show that in assembly systems, the
freezing part of the end-item must not includes the lead
times of components.

More recently Ho and Ireland (1998) studied the
correlation between forecast errors and nervousness.
Their simulations show that lead time uncertainty affects
the instability whatever is the lot-sizing method and
whatever is the forecast error level. Wemmerlov and
Whybark (1984) give a detailed presentation of the lot-
sizing methods used in MRP systems.

The symmetric problem, when demand is constant and
lead time is uncertain, can be formulated adopting
policies without nervousness (Dolgui and Ould-Louly
2001). In fact, the planning change is due to the
cumulative demand uncertainty. When the demand is
constant there is no need to change, and the MPS can be
frozen on the total horizon planning. The problem with
constant demand appears, for example, in mass
production.

MRP begins with the end-product need date and uses the
lead time to calculate the components release date.
Clearly, the calculation doesn't take into account the
actual lead time because this one isn't known at this
moment and can be uncertain. The calculation uses
forecast (planned lead time). Melnyk and Piper (1981)
proposed a forecast method for the lead time which is
issued from the used methods for random demand:

Planned lead time =
= lead time forecast + safety lead time
= lead time mean + k lead time standard deviation.

However the Wemmerlov (1986) study shows that the
errors of forecast increase the inventories and, in the
same time, decrease the customer service level. Molinder
(1997) studies this problem. Instead of forecasting, he
proposes simulated annealing to find good safety stock
and safety lead time. His results show that high planned
lead time gives excessive inventory, and small planned
lead time gives shortages and delays.

Whybark and Williams (1976) found the use of safety
lead time more efficient than safety stock. Grasso and
Tayor (1984) give an opposite conclusion, their
simulations find more prudent safety stocks. The
conceptual analysis of New (1975) shows that safety lead
time is better when the components are ordered from
external suppliers. There is another approach to avoid the

2
uncertainty using a different information structure (Ho et
al., 1995). (Yeung et al., 1998; Dolgui and Ould-Louly,
2000) made exhaustive reviews on the methods studying
the parameter affecting the effectiveness of MRP
systems.

Gupta and Brennan (1995) studied MRP systems using
simulations, they showed that lead time uncertainty has a
large influence on the cost. The statistics done on
simulations by Bragg et al. (1999) show that lead times
substantially influence the inventories.

As shown, there are a lot of simulation studies dealing
with this problem, and there are a lot of studies dealing
with the formulation problem, but in our knowledge,
there is no exact mathematical method giving the optimal
planned lead times.

A particular supply planning problem under lead time
uncertainties concerns assembly systems. In fact, for
assembly systems, several types of components are
needed to produce one finished product. So, the
inventories of the different types of components become
dependent. Then the supply planning becomes more
difficult.

The mathematics formulation of the cost under lead time
uncertainty is difficult. A lot of researchers avoid this
problem using simulations (Banks et al., 2001). There
are some contributions which propose mathematics
resolution, but their studies are limited to a single-period
problem and do not take into account the dependence
between periods. For example the Chu et al. (1993)
model gives optimal values of the planned times in
assembly systems, but only for the single-period
problem. The mathematics formulation of the multi-
period problems under lead times uncertainty is more
difficult. In that case, orders may cross, that is, they may
not be received in the same sequence in which they are
placed (He et al., 1998). Some contributions assume that
orders do not cross, then they solve, under this
assumption, the single-item problem (Graves 1993).

The problem difficulty increases for assembly systems
because of the dependence between inventories. Wilhelm
and Som (1998) studied this problem and showed that a
renewal process can describe end-item inventory level
evolution, but they did not study the dependence between
components. This dependence was studied by Gurnani et
al. (1996) but their assembly system was only with two
components, and the considered lead time distribution
was simple.

The aim of this paper is the study of the supply problems
of assembly systems in which several types of
components are needed to produce one finished product.
It focuses on the components lead times uncertainty. The
target is to find the optimal values of the planned lead
times and the optimal sizes of the lots. A Markov model
that gives the measure of the average cost is used, and a
new generalized newsboy model is proposed to find the
optimal policy.


2. PROBLEM DESCRIPTION

The unit holding cost h
i
of the component i per period,
the unit backlogging cost b of a finished product per
period and the setup cost c are known. The distribution
of the component i lead time L
i
is also known, and its
upper value is equal to u
i
. The discrete random variable
k
i
L is the lead time of the components i ordered at the
beginning of the period k. The demand D of finished
products per period is constant, and a
i
components i are
needed to assemble the finished product. The component
lot-sizes are determined by using the periodic order
quantity (POQ) method, with a periodicity of p periods.
The orders of components are released at the beginning
of the periods kp+1, k=0,1,2,etc, and there is no order
release in the periods kp+r, r=2,3,,p. Then, the supply
orders Q
i
of components are constant Q
i
=a
i
Dp (p is a
decision variable). The finished product demands are
satisfied at the end of each period and unsatisfied
demands are backordered and have to be satisfied during
next periods.

As the lead times are uncertain, the orders have to be
released before the need instant (the planned lead times).
But the ordered quantities are the same, then the planned
lead times are equivalent to initial inventories. So, the
aim is to find the optimal values of the initial inventories
a
i
Dx
i
, where x
i
, i=1,2, n, are the planned lead times,
and the optimal values of the parameter p of the method
POQ.

Given that the maximal value of the component i lead
time is equal to
i
u , only the orders made in the previous
u
i
-1 periods may not be arrived yet. The orders made
before already arrived. The number
m , p
i
N of the
component i expecting deliveries at the end of the period
m=kp+r is easy to calculate.

Let
j m
i
L
+1
, j=r, r+p, r+2p,, r+
(
(
(

p
r u
i
1
p, be the lead
times of the orders made at the beginning of the periods
kp+1, (k-1)p+1,, (k-
(
(
(

p
r u
i
1
)p+1. The order made in
the period m+1-j is delivered after the end of the period
m when j L
j m
i
>
+1
, i.e. if
j L
j m
i
>
+1
1 is equal to 1. Then,
the random variables

=
(
(
(

=
+ >
+

+
p
r
i
u
p ) j k (
i j
r jp L
r kp , p
i
N
1
1
0
1 , n i , , 1 = , (1)

give the global state of the previous orders. In fact,
m p
i
N
,
is the number of the component i orders that are

3
not arrived yet (at the end of the period m, they are
already waited).

Proposition 1. The cost of the period m=kp+r can be
expressed as follows:

) , , , , (
, ,
1
m p
n
m p
k
N N p X C =
1
1
=

r
c
+
+
=
+
n
i
r kp p
i i i i
pN r p x a h D
1
,
) (
+
+ +
=
+ ) ( max
,
, , 1
i
r kp p
i
n i
x p r pN DH

(2)

where ) , , (
1 n
x x X = and + =
=
n
i
i i
a h b H
1
. The value
(Z)
+
is equal to max{Z, 0}.

Proof. The cost is equal to the sum of the holding cost of
components, the backlogging cost of the finished
product, and the setup cost.

There is a shortage at the end of the period m when the
amount of components, arrived in the inventory since the
first period, are not sufficient to satisfy the cumulative
demand D(kp+r) of the finished product.

The amount of components i needed by this cumulative
demand is equal to a
i
D(kp+r). The number of orders
made since the beginning until the end of the period
(kp+r) is equal to k+1, and the number of the delivered
orders is equal to (k+1-
r kp , p
i
N
+
). There is a shortage if
there is a component i for which the initial inventory
i i
Dx a plus the delivered amount Q
i
(k+1-
r kp p
i
N
+ ,
) is
smaller than the cumulative needed amount a
i
D(kp+r).
So, there is a shortage at the end of the period m, if there
is a component i satisfying ) (
,
i
r kp p
i
x p r pN +
+
>0.
The number of backlogged demand is given by the
component i that has the maximal value
) x p r pN (
i
r kp , p
i
+
+
. Then, the backlogging cost of
the period m is ) x p r pN ( max bD
i
r kp , p
i
n ,..., i
+
+
=1
, and
only [kp+r- ) x p r pN ( max
i
r kp , p
i
n ,..., i
+
+
=1
] demands are
satisfied.

The inventory
m
i
S of the component i at the end of the
period m is equal to the initial inventory
i i
Dx a plus the
delivered amount Q
i
(k+1-
r kp p
i
N
+ ,
) without the quantity
used for the satisfied demand:

+ + ) pN r p x ( D a = S
m , p
i i i
m
i


+
=
+ ) x p r pN ( max D a
j
m , p
j
n , , j
i
1
. (3)

Given that there is a setup cost c if r=1, the sum of the
holding cost of the components, the backlogging cost of
the finished product, and the setup cost at the end of the
period m is equal to:

) N , , N , p , X ( C
m , p
n
m , p
m

1
=
1
1
=

r
c
+
+
=
+
n
i
r kp p
i i i i
pN r p x a h D
1
,
) (
+
+ +
=
+ ) ( max
,
, , 1
i
r kp p
i
n i
x p r pN DH

.

This cost is a random variable. To express the holding
cost on an infinite horizon, a Markov chain is proposed
for which a state Z { }
1
1 , 0

i
u
is a binary vector that
describes the orders made in the previous u
i
-1 periods.
The average cost on the infinite horizon is:

) , ( p X C =
=
p
r
r
p
p X C E
1
1
)] , ( [ , (4)

where E(Z) is the expected value of Z, and

) , ( p X C
r
=
1
1
= r
c
+ +
=
+
n
i
r p n p
i i i i
i
pN r p x a h D
1
,
) (
+
+ +
=
+ ) ( max
,
, , 1
i
r p n p
i
n i
x p r pN DH
i

, (5)
p
p r u
i
i
n
+
=
1
. (6)

To simplify the equations, let h
i
be the holding cost of
the amount a
i
D, instead of the unit holding cost. The
problem is the same with this new notation.

3. PERFORMANCE MEASURE

The problem is to minimize the objective function given
by the equation (4). This minimization is rather difficult
because the function is not linear and, in our case, the
decision variables, x
i
, i=1,2,,n, are integer. The
minimization of an objective function with the same
structure was the aim of Chu et al. (1993), but in their
case the lead times of components were continuous
random variables, p equal to 1, and the decision variables
were real, so they didn't have discrete optimization
problem.

Theorem 1. An explicit form for average cost, given by
equation (4), is the following:

) , ( p X C =
p
c
+ ) (
2
1
b H
p

+
=
n
i
p
i i i
N E x h
1
)] ( [
+
= =
+ +
0 1 1
,
1
] ) ( 1 [
k
p
r
n
i
p
p r k x r p
i
p
F H , (7)


4
where ) Pr( ) (
, ,
x N x F
r p
i
r p
i
= and =
=
p
r
r p
i
p
i
N N
1
,
.

Proof. In fact,

)] , ( [ p X C E
r
=
1
1
=

r
c + ) )( ( r p b H
+
=
+
n
i
r p n p
i i i
i
N pE x h
1
,
)] ( [ + ) (Z HE (8)

where
i
n i
Z Z
, , 1
max
=
= , (9)
and
+ +
+ = ) (
,
i
r p n p
i i
x p r pN Z
i
. (10)

Z is a positive discrete random variable with a finite
number of possible values, its expected value is:

) (Z E =

=
0
, , 1
)] max Pr( 1 [
k
i
n i
k Z

. (11)

But the random variables Z
i
are independent, hence:

) (Z E =
= 0 1
] ) ( Pr 1 [
k
n
i
i
k Z
=
=
+ +
0 1
,
] ) ( 1 [
k
n
i
p
p r k x r p
i
F . (12)

Using the equation (12), the equation (8) becomes:

)] , ( [ p X C E
r
=
1
1
=

r
c + ) )( ( r p b H
+
=
+
n
i
r p n p
i i i
i
N pE x h
1
,
)] ( [
+
=
+ +
0 1
,
] ) ( 1 [
k
n
i
p
p r k x r p
i
F . (13)

The proof is obtained by using (13) in (4).

In addition, it is easy to see that the components
i
x of
the optimum X must satisfy 1 0
i i
u x .


4. OPTIMIZATION

The general optimization problem is difficult due to the
no linearity of the cost. In this section, the problem is
solved under the assumption that holding costs of the
quantities Q
i
=a
i
D per period are the same, and the lead
time L
i
of the different components have the same
distribution probability. Then, the costs h
i
, i=1n, can
be noted by h, and the distributions
r p
i
F
,
, i=1n, can
be noted by
r p
F
,
. The obtained problem will be noted
by and an exact method which solves this problem in
polynomial time will be given. The following notations
are used.

E
k
is the set of vectors X, each vector's component is an
integer variable satisfying:
j i
x x = , for k i and k j .

1
k
P and
2
k
P are applications from E
k
to E
k+1
which are
defined as follows:

=
+ =
=
, otherwise , x y
1, k i if , x y
Y ) X ( P
i i
i
k
1
1

=
+ =
=
+
. otherwise , x y
1, k i if , x y
Y ) X ( P
i i
k i
k
1
2

Then two vectors, ) (
1
X P
k
and ) (
2
X P
k
, of the set E
k+1

are associated to every element X of E
k
.

Theorem 2. The following equation is true under the
assumption of the problem : { } 1 , , 2 , 1 n k ,

) p , X ( C Min
k
E X
= ) p , X ( C Min
k
E X
1 +

. (14)

Proof. Note that
k k
E E
+1
. Then it is sufficient to
show that: { } 1 , , 2 , 1 n k ,

) p , X ( C Min
k
E X
1
) p , X ( C Min
k
E X
. (15)

For each vector X

of E
k
, there is a vector of E
k+1
which
gives a smaller cost. Precisely, at least one of the two
vectors, ) (
1
X P
k
and ) (
2
X P
k
, gives a cost less than the
cost C(X,p):

k
E X , { } ] p ), X ( P [ C ], p ), X ( P [ C Min
k k
2 1

) ( p , X C . (16)

In order to proof the theorem, it is sufficient to show the
inequality (16). Let's suppose that (16) is false, then the
two following equations are obtained:

] p ), X ( P [ C ) p , X ( C
k
1
< , (17)
] p ), X ( P [ C ) p , X ( C
k
2
< . (18)

The equation (17) can easily be rewritten as:

) (
k
x x h
1 1 +
>

=
+ + + +
+
0 1
, ,
)] ( ) ( [(
1 1
j
p
r
p
p r j x
r p
p
p r j x
r p
p
H k
F F

+ =
+ + + +
n
k i
p
p r j x
k
r p
p
p r j x
k
r p i
F F
2
, ,
) ( ) (
1
. (19)

In the same time, (18) gives:

<
+
) (
1 1 k
x x h k

5

=
+ + + +
+
0 1
, ,
)] ( ) ( [
1 1
j
p
r
p
p r j x
r p
p
p r j x
r p
p
H k
k k
F F

+ +

+ =
n
k i
i r p
p
p r j x
F
1
,
) ( . (20)

Using that
k k
= ) (
l k
k
l
l

=

1
1
0
for
) (
1 ,
p
p r j x
r p
k
F
+ +
= and ) (
1 ,
p
p r j x
r p k
k
F
+ +
+
= , the
equation (20) becomes:

<
+

+
nh b
x x h k
k
) (
1 1
+ +

+
0
1 1
)] ( ) ( [
j
k
j x F j x F
+ +

=
+

1
0
1
1
1
) ( ) (
k
l
k
l k l
j x F j x F +
+ =
n
k i
i
j x F
1
) ( ,

that can be rewritten as follows :

<
+

+
nh b
x x h k
k
) (
1 1
+ +

=
+
1
0 0
1 1
)] ( ) ( [
k
l j
k
j x F j x F
+ + +
+ =
+

n
k i
i k
l k l
j x F j x F j x F
2
1 1
) ( ) ( ) ( . (21)

Using the fact that
k l k l
) ( ) (

for
) (
1 ,
p
p r j x r p
F
+ +
= and ) (
1 ,
p
p r j x r p k
F
+ +
+
= ,
equation (21) becomes:

) (
1 1 +

k
x x h <

=
+ + + +
+
0 1
, ,
)] ( ) ( [(
1 1
j
p
r
p
p r j x
r p
p
p r j x
r p
p
H k
F F

+ =
+ + + +
n
k i
p
p r j x
r p
p
p r j x
k
r p i
F F
2
, ,
) ( ) (
1
. (22)

The equation (19) and the equation (22) are in
contradiction. Then, the equation (16) is true.

Corollary 1. The problem has an optimal solution in
which the initial inventories have the same value for
every type of components.

Proof. The prove is immediate by transitivity from the
equation (14): ) p , X ( C Min
E X
1

= ) p , X ( C Min
n
E X
.

Then, there is an optimal solution X =
) ..., , ..., , ( x x x E
n
. This optimal solution can be
calculated using the following theorem.

Theorem 3. The optimum X= ) ..., , ..., , ( x x x which
minimizes the problem is such that the integer x
satisfies:

) ( F
H
bp
) ( F
p
r p x
p
r
r , p
p
r p x
p
r
r , p
n n
+
=
+
=

1
1
1
. (23)

Proof. According to corollary1, the cost C(X,p) can be
reduced to a two-variable function ) , ( p x C
e
:

) p , x ..., , x ..., , x ( C = ) p , x ( C
e
=
2
1
+
p
p
c
) b H ( +
)] ( [
p
N E x nh +
=
+ +
0 1
1
1
k
p
r
p
r p k x r , p
p
)] ( F [ H
n
,

where ) ( ) ( ) ( ) (
2 1
p p
n
p p
N E N E N E N E = = = = .

Let G(x,p) be the following function:

) , ( p x G = ) , 1 ( p x C
e
+ - ) , ( p x C
e

= b F
p
r
p
r p x r p
p
H
n

=
+
1
,
) ( . (24)

G(x,p) is an increasing function, because ) (
,
x F
r p
is a
positive increasing function. Then, the inequalities,
) , 1 ( ) , ( p x C p x C
e e
and ) , 1 ( ) , ( p x C p x C
e e
+ ,
give the value of x which minimizes the cost C
e
. These
inequalities can be rewritten as:

) ( ) (
1
, 1
1
,
p
r p x
p
r
r p
p
r p x
p
r
r p
n n
F
H
bp
F
+
=
+
=
.

Note that for n=1, i.e. for one type of components, and
for p=1, i.e. periodicity equal to 1, the obtained model is
the well known newsboy model.


5. CONCLUSIONS

The aim of this paper is the study of a supply problem for
assembly systems, when several types of components are
needed to assemble a finished product. The lead times of
the components are independent random variables, and
the demand of the finished product is constant. The
studied criteria are the sum of the holding cost of
components, the finished product backlogging cost, and
the setup cost.

An explicit form of the average cost as a function of the
MRP parameters is given. A generalized newsboy model
that allows the problem resolution under an assumption
is given. This assumption is that the lead times of the
different types of components follow the same
distribution probability, and that the holding costs per
period of the ordered quantities are the same.

Compared to the classical newsboy model, the proposed
model is not limited to one type of raw materials, and can
be used for assembly systems with several types of
components. This model takes into account the
dependence between inventory levels at different
periods, and also the dependence between the inventories
of the different types of components.


6

REFERENCES

Banks J., J.S. Carson, B.L. Nelson and D.M. Nicol
(2001). Discrete-event system simulation. Prentice-
Hall, New Jersey.
Blackburn J.D., D.H. Kropp and R.A. Millen (1986). A
comparison of strategies to dampen nervousness in
MRP systems. Management Science, 33, 413-429.
Bragg D.J., E.A. Duplaga and C.A. Watis (1999). The
effects of partial order release and component
reservation on inventory and customer service
performance in an MRP environment. International
Journal of Production Research, 37, 523-538.
Carlson R.C., S.L. Beckman and D.H. Kropp (1982).
The effectiveness of extending the horizon in rolling
production scheduling. Decision Science, 13, 129-
146.
Chu C., J.M. Proth and X. Xie (1993). Supply
management in assembly systems, Naval Research
Logistics, 40, 933-949.
Dolgui A. and M.A. Ould-Louly (2000). Supply chain
planning under uncertainties. Proc. of the IFAC
Symposium on Manufacturing, Modeling,
Management and Control (MIM'2000), Patras,
Greece, 291-296.
Dolgui A. and M.A. Ould-Louly (2001). A model for
supply planning under lead time uncertainty.
International Journal of Production Economics, (to
appear).
Grasso E.T. and B.W. Tayor (1984). A simulation based
experimental investigation of supply/timing
uncertainty in MRP systems. International Journal of
Production Research, 22, 485-497.
Graves S.C., 1993. Handbooks in Operations Research
and Management Science. Vol. 4: Logistics of
Production and Inventory, North-Holland,
Amsterdam.
Graves S.C., D.B. Kletter and W.B Hetzel. (1998). A
dynamic model for requirements planning with
application to supply chain optimization. Operations
Research, 46, 35-49.
Gupta S.M. and L. Brennan (1995). MRP systems under
supply and process uncertainty in an integrated shop
floor control. International Journal of Production
Research, 33, 205-220.
Gurnani H., R. Akella and J. Lehoczky (1996). Optimal
order policies in assembly systems with random
demand and random supplier delivery. IIE
Transactions, 28, 865-878.
He X.X., S.H. Xu, J.K. Ord and J.C. Hayya (1998). An
inventory model with order crossover. Operations
research, 46, 112-119.
Ho C., W.K. Law and R. Pampal (1995). Uncertainty-
dampening methods for reducing MRP system
nervousness. International Journal of Production
Research, 33, 483-496.
Ho C.J. and T.C. Ireland (1998). MRP system
nervousness and forecast errors. International
Journal of Production Research, 36, 2285-2299.
Kadipasaoglu S.N. and S.V. Sridharan (1995).
Alternative approach for reducing schedule instability
in multistage manufacturing under demand
uncertainty. Journal Operations Management, 13,
193-211.
Mather H. (1985). Dynamic lot-sizing for MRP: help or
hindrance? Journal of Production and Inventory
Management, 2, 29-35.
Melnyk S.A. and C.J. Piper (1981). Implementation of
material requirements planning. International
Journal of Production and Operations Management,
2, 52-60.
Molinder A. (1997). Joint optimisation of lot-sizes,
safety stocks and safety lead times in an MRP system.
International Journal of Production Research, 35,
983-994.
New C. (1975). Safety stocks for requirements planning.
Production and Inventory Management, 16, 1-18.
Sridharan V., W.L. Berry and V. Udayabanu (1988).
Measuring master production schedule stability under
rolling planning horizon. Decision Science, 19,147-
166.
Sridharan V. and R.L. LaForge (1989). The impact of
safety stock in schedule instability, cost and ervice.
Journal Operations Management, 8, 327-347.
Sridharan V. and W.L. Berry (1990a). Master production
schedule, make-to-stock products: a framework for
analysis. International Journal of Production
Research, 28, 541-558.
Sridharan V. and W.L. Berry (1990b). Freezing the MPS
under demand uncertainty. Decision Science, 21, 97-
120.
Wemmerlov U. (1986). A time-phase order-point system
in environment with and without demand uncertainty,
24, 343-358.
Wemmerlov U. and D.C. Whybark (1984). Lot-sizing
under uncertainty in a rolling schedule environement.
International Journal of Production Research, 22,
467-484.
Wilhelm W.E. and P. Som (1998). Analysis of a single-
stage, single-product, stochastic, MRP-controlled
assembly system. European Journal of Operational
Research, 108, 74-93.
Yeung J.H.Y., W.C.K. Wong and L. Ma (1998).
Parameters affecting the effectiveness of MRP
systems: a review. International Journal of
Production Research, 36, 313-331.
Yeung, J.H.Y., W.C.K. Wong, L. Ma and J.S. Law
(1999). MPS with multiple freeze fences in multi-
product multi-level MRP systems. International
Journal of Production Research, 37, 2977-2996.