Sie sind auf Seite 1von 15

*Corresponding author. Tel.

: #46 13 281 504; fax: #46 13


288 975; e-mail: rwg@ipe.liu.se.
Int. J. Production Economics 59 (1999) 361375
A net present value approach to safety stocks in a multi-level MRP
system
Robert W. Grubbstro m*
Department of Production Economics, Linko( ping Institute of Technology, S-581 83 Linko( ping, Sweden
Abstract
The Laplace transform and inputoutput Analysis have been used for formulating a basic theory for material
requirements planning (MRP) in a sequence of recent papers. The transform has been used for describing time
developments and lags of the relevant production, demand and inventory variables in a compact way including eects of
order ows and lead times. Secondly, the transform has functioned as a generating function, and thirdly, the transform
has been applied for assessing cash ows adopting the net present value (NPV) principle (or the annuity stream which is
a variation of NPV). Input-output Analysis, in particular, the input matrix, has been applied for describing multi-level
product structures. This has made the analysis compact and distinct.
In the current paper the analysis of determining optimal safety stock levels in MRP systems is extended from
a single-level model recently investigated to a multi-level system assuming production to take place according to
a lot-for-lot (L4L) policy when applying the NPV as the objective criterion and disregarding the opportunities for joint
set-ups. Relaxing the L4L assumption is also discussed. 1999 Elsevier Science B.V. All rights reserved.
Keywords: MRP; Inputoutput analysis; Laplace transform; Multi-period; Production; Safety stock
1. Introduction
This paper follows up a recent paper treating
a similar problem for a single-level MRP system
[1]. The main question addressed, is to nd the
amounts to produce together with their timing con-
cerning end items having multi-level product struc-
tures, when the demand for these end items is
stochastic. The objective is to choose a best produc-
tion plan maximising the net present value (NPV) of
the cash ow associated with production and de-
mand. By deciding on the produced amounts and
their timing, safety stock decisions will be incorpor-
ated automatically.
Production on dierent levels is assumed to take
place in batches of dierent sizes at dierent points
in time. Cumulative production of an item will then
follow a staircase function. The safety stock is then
dened as the dierence between cumulative pro-
duction and expected cumulative demand immedi-
ately prior to a replenishment. This dierence may
be negative in such cases when a positive expected
backlog is favourable. This might typically happen
for an early batch in the sequence. Safety stocks are
0925-5273/99/$- see front matter 1999 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 5 - 5 2 7 3 ( 9 8 ) 0 0 0 1 6 - 4
Fig. 1. Inventory and backlog relationships.
only considered for top-level items, i.e. for nished
goods having an external demand.
2. Fundamental equations
Let there be N items in the system altogether.
Demand D, stock S and production P (including
purchases when appropriate) are represented by
N-dimensional column vectors each being a func-
tion of time. These vectors are rates with the dimen-
sion units per time unit and they are turned into
Laplace transforms denoted by tildes or by )
Time integrals (cumulative values) of functions are
denoted by bars and inverse transforms by ) .
The notation i is reserved for the imaginary unit
(!1.
For the production (assembly) of one unit of item
j, there is a need in the amount of h
IH
of item k, and
there is a lead time t
H
ahead of the completion of the
production at which the components are needed.
The h
IH
are arranged into the square input matrix
H describing the product structures of all relevant
products. The lead times t

, t
`
,
2
, t
,
, create inter-
nal demands and are represented by a diagonal
matrix, tJ the lead time matrix, having eQOH in its jth
diagonal position, where s is the complex Laplace
frequency. HI"HtJ is the generalised input matrix
and it captures component requirements together
with their required timing.
Consider Fig. 1 explaining the main equations to
follow. The following denitions are adopted. otal
inventory SI accounts for all items in the system and
is initial stock plus cumulative net production
(I!H)PI/s less cumulative deliveries FI(s)/s:
SI(s)"
S(0)#(I!H)PI(s)!FI(s)
s
, (1)
where S(0) is the vector of initial inventory and I the
identity matrix. The factor 1/s is an operator ac-
counting for a time integration.
Deliveries concern items that are demanded ex-
ternally. Deliveries from inventory occur as soon as
there is a demand of an item not previously satised
and there is a supply to meet this demand, i.e. that
this demand can be covered from available inven-
tory.
Available inventory RI(s) is cumulative production
PI(s)/s less cumulative deliveries FI(s)/s and cumulat-
ive internal demand HPI(s)/s, which evaluated at the
time the components are reserved is HtJPI(s)/s, and
RI(s) (as a time function) must never be negative,
otherwise the production plan will be infeasible:
RI(s)"
R(0)#(I!HtJ)PI(s)!FI(s)
s
. (2)
Internal demand must always be met, but if ex-
ternal demand cannot be met, we assume that this
demand is backlogged and satised at the time
available inventory starts to become positive once
again (when a batch is completed). The non-
negativity of the inverse transform of RI(s) in each
component is the main set of constraints of this
class of problems.
362 R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375
Allocated component stock AI(s) is dened as the
dierence between (total) inventory and available
inventory. This stock is reserved for production
and is part of work-in-process.
AI(s)"SI(s)!RI(s)"
A(0)#H(tJ!I)PI(s)
s
, (3)
where A(0)"S(0)!R(0). Backlogs BI(s) only
concern external demand and are given by initial
backlogs plus cumulative external demand less
deliveries:
BI(s)"
B(0)#DI(s)!FI(s)
s
. (4)
Backlogs are necessarily non-negative. If Eqs. (2)
and (4) are subtracted eliminating FI(s), we obtain
RI(s)"
R(0)!B(0)#(I!HtJ)PI(s)!DI(s)
s
#BI(s). (5)
For any item, its available inventory and its back-
log cannot be positive at the same time, since a de-
livery takes place from available inventory as soon
as there is an unsatised external demand. Hence, if
for any component R
H
(t)'0 at time t, then
B
H
(t)"0, and vice versa. Therefore R(t) and B(t),
both being non-negative, may be written as
R(t)"

R(0)!B(0)#(I!HtJ)PI(s)!DI(s)
s
>
,
(6)
B(t)"

B(0)!R(0)!(I!HtJ)PI(s)#DI(s)
s
>
,
(7)
where [ ) ]> is the maximum operator Max0, )
operating on a time function, and in transform
terms
RI (s)"

R(0)!B(0)#(I!HtJ)PI(s)!DI(s)
s
>
, (8)
BI (s)"

B(0)!R(0)!(I!HtJ)PI(s)#DI(s)
s
>
, (9)
where [ ) ]
>
is the transform of the non-negative
truncation of the time function corresponding to
the argument, since cumulative production as well
as cumulative external demand are zero at time
t"0. Total inventory, being the sum of available
inventory and allocated component stock then
obeys
SI(s)"

R(0)!B(0)#(I!HtJ)PI(s)!DI(s)
s
>
#
A(0)#H(tJ!I)PI(s)
s
, (10)
or in the time domain
S(t)"

R(0)!B(0)#(I!HtJ)PI(s)!DI(s)
s
>
#

A(0)#H(tJ!I)PI(s)
s
"Max

A(0)#H(tJ!I)PI(s)
s
,

S(0)!B(0)#(I!H)PI(s)!DI(s)
s
. (11)
The equations above dening the development of
SI, RI, BI and AI, we call the fundamental equations.
3. Solution to stockout function
We assume demand to be a renewal process, i.e.
demand is created by unit events separated by
independent stochastic time intervals, all having
the same probability distribution f (t), t*0.
A basic problem is to nd an expression for
E(B(t)) and E(BI(s)) in terms of decision variables
and stochastic properties of demand.
Theorem. If external demand for an individual item
follows a renewal process and cumulative production
at time t is PM (including initial stock), then, during the
time that cumulative production remains at PM, the
level of expected stockouts for the item B(t) will be
part of the function
E(B(t))"E(BI(s))"

fI.
M
>(s)
s(1!fI(s))
, (12)
for the item in this interval.
R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375 363
Fig. 2. Illustration to stockout function for four dierent levels of cumulative production PM.
Therefore, the expected stockout as a function of
time can be written as the sum
E(B(t))"
.
M
[E(BI(s))]
.
M
,
.
M
"
.
M

fI.
M
>(s)
s(1!fI(s))
.
M
,
.
M
, (13)
where ,
.
M
is the characteristic function being unity
within the intervals where cumulative production
of the item is PM and zero elsewhere, and where the
summation takes place over all intervals.
Proof of this theorem is given in Refs. [2,3].
As an illustration, we choose demand as a Pois-
son process having the parameter z (per time unit).
The resulting stockout function E(B(t)) is shown in
Fig. 2 for four levels of cumulative production.
The height of the dashed curves above the hori-
zontal lines in Fig. 2 indicates the expected level of
stockouts over time for the corresponding given
cumulative production. For instance, if PM"10 dur-
ing the interval from t"7 till t"12 (indicated by
broken vertical lines), then E(B(t)) follows the part
of the curve pointed out with an arrow.
4. Lot-for-lot policy (L4L)
The L4L lot-sizing policy involves production
decisions on lower levels in exactly the amounts
that are required by internal demand at the time
this demand occurs. If production of externally
demanded items is given, then so is internal de-
mand creating additional internal demand on still
lower product structure levels.
The L4L policy requires that the product struc-
ture is of the assembly type. By an assembly struc-
ture we mean that the product structure is such that
for each higher-level item, the items immediately
required are all on a lower level (no items are split
into several later-stage items and there is no feed-
back from higher to lower levels). If the items are
suitably enumerated in such a case, then this en-
ables the input matrix to have elements only be-
neath its main diagonal (a triagonal matrix with
a zero-valued diagonal). Higher-level items then
have lower item indices than lower-level items.
Let us denote the number of end items by N'. If
the production of these end items is set to PI '(s)
(which contains zeros in all other positions, i.e. the
lower NN' positions of PI'(s)), then the immediate
requirements of components will be HtJPI '(s). If
these are produced L4L, they will create an internal
demand (HtJ)`PI'(s), etc. Total production will then
amount to
PI(s)"(I#HtJ#(HtJ)`#2)PI '(s)
"
I"
(HtJ)IPI '(s)
"(I!HtJ)PI '(s). (14)
364 R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375
The summation of the geometric series
I"
(HtJ)I
converges if the absolute values of the eigenvalues
of the matrix are all less than unity (in this case they
are all zero). However, with H and therefore HtJ
being triagonal matrices with zeros along their di-
agonals, the sum converges in a nite number of
terms (exactly N terms in this type of problem,
including the identity matrix for k"0).
The inverse matrix (I!HtJ) has the same prin-
cipal structure as the Leontief inverse in inputout-
put analysis. Since HtJ is a generalised input matrix
capturing required amounts as well as their timing,
the inverse written HI*"(I!HtJ) may be con-
sidered to be a generalised eontief inverse.
5. Extensions to capacity requirement
considerations
By extending the input matrix H with additional
rows, we may also take capacity requirements into
consideration, cf. Ref. [[4], Ch. 3]. Let there be
M work stations which may be used for the produc-
tion of the items considered. Units of capacity used
when manufacturing each item j"1, 2,
2
, N may
then be described by elements h
IH
, where
k"N#1, N#2,
2
, N#M, now making the
input matrix rectangular rather than square having
the dimension (N#M);N. Let the original
N;N-dimensional input matrix be denoted by H'
and the extended matrix by H. The direct materials
as well as capacity requirements for the production
of end items PI ' will then be HtJPI ', of which H'tJPI '
are the direct materials requirements. These, in
their turn create an internal demand for materials
and capacity amounting to HtJH'tJPI '. Assuming an
L4L lot-sizing policy, the total materials and capa-
city requirements (including nal end items) for the
production of PI ' will then become
PI(s)"(I'#(HtJ#HtJH'tJ#HtJ(H'tJ)`#2)PI '(s)
"

I '#HtJ
I"
(H'tJ)I

PI '(s)
"(I'#HtJ(I!H'tJ)PI '(s), (15)
where I ' is an [(N#M);N]-dimensional matrix
consisting of an (N;N)-dimensional identity
matrix extended with M rows of zeros. Since
I#H'tJ(I!H'tJ)"(I!H'tJ), we may also
write Eq. (15) as
PI(s)"

I
H"tJ(s)

(I!H'tJ(s))PI '(s), (16)


where H" contains the lower M rows of H. The
capacity requirements have here been allocated at
the beginning of the respective lead times. This can
be modied if necessary.
Available capacity may be included in the funda-
mental equations as negative demand, cf. Ref. [[4],
Ch. 3].
An example of a product structure taking capa-
city requirements X and Y (two work stations) into
account is given in Fig. 3.
The extended input matrix applicable to this
product structure may be written as
H"
A
E
B
C
D
X
Y
A E B C D
0 0 0 0 0
0 0 0 0 0
4 1 0 0 0
2 0 1 0 0
0 3 2 2 0
0 0 1 0 0
0 0 0 0 1
.
Items A and E are end items, B, C and D are
subordinate items and X and Y are capacities. We
therefore have N"5, N'"2 and M"2.
To illustrate Eqs. (15) and (16) we choose a hor-
izon of 100, lead times of A, E, B, C and D to be 14,
12, 10, 8, 6, and 10 units of end item A to be
produced at time 50 and 5 units of end itemE to be
produced at time 60. We then have
tJ"
e"Q 0 0 0 0
0 e`Q 0 0 0
0 0 e"Q 0 0
0 0 0 e`Q 0
0 0 0 0 e'Q
,
R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375 365
Fig. 3. Example of a product structure with two end items and
taking capacity requirements into account.
(I!H'tJ)"
1 0 0 0 0
0 1 0 0 0
4e"Q e`Q 1 0 0
2e"Q#4e`"Q e``Q e"Q 1 0
8e`"Q#4e``Q#8e``Q 3e`Q#2e``Q#2e`"Q 2e"Q#2e`Q 2e`Q 1
.

4e`"Q e``Q e"Q 0 0


8e`"Q#4e``#8e``Q 3e`Q#2e``Q#2e`'Q 2e'Q#2e`"Q 2e"Q e'Q
.
PI '(s)"
10e`"Q
5e'"Q
0
0
0
.
The generalised Leontief inverse will be
To obtain the whole coecient matrix in Eq. (16),
the generalised Leontief inverse is augmented with
the following two rows:
The resulting cumulative production/capacity
levels PI
I
(s)/s, k"1, 2,
2
, N#M, are shown
in Fig. 4.
6. Assumptions in current study
We make the following assumptions in our cur-
rent study:
1. All external demands are created by renewal
processes, one for each end item.
2. There is no external demand for an item that is
part of a higher-level item (no external demand
for spare parts, etc.).
3. All product structures are given including ca-
pacity requirements, these parameters being
represented by the (extended) H matrix. All
lead times are given and are represented by the
matrix tJ .
4. Capacity constraints are not considered.
5. A nite horizon K is assumed.
6. All end items are produced in batches of pos-
sibly varying size.
7. The decisions to be made concern the end items
as to their amounts and timing. This includes
deciding on the number of batches to produce
until the horizon.
8. The objective function is the net present value
(NPV) (or the annuity stream proportional to
the NPV). The payments involved are xed
out-payments K
H
, j"1, 2,
2
, N, for each set-
up assumed to occur at the beginning of the
corresponding lead time, a unit production
out-payment for each item produced located at
the completion time c
H
, j"1, 2,
2
, N#M,
where the last M components refer to unit
capacity costs, and a unit in-payment r
H
, j"1,
2,
2
, N#M, for each unit sold at the time of
completion. The r
H
are zero for all items that are
not sold externally.
366 R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375
Fig. 4. Total production of seven items in example (including
two capacities used).
9. Backlogging is permitted only for end items.
The penalty for a backlog is the delay in rev-
enue in-payments. Other penalties could also
be introduced easily.
10. All items on lower levels (only having internal
demand) are produced L4L. Relaxing this
assumption is discussed in Section 10.
11. The opportunities for joint set-ups for lower-
level items of the same kind are disregarded.
This means that the internal demand for
a batch of each lower-level item is generated by
the production decision concerning each end
item individually. If two end items need the
same type of component at exactly the same
time, then two separate batches will be produc-
ed simultaneously (with the consequence of two
set-up charges).
7. Decision variables
As stated above, the decisions to be made are the
amounts of end items and the timing of their pro-
duction, which is all captured in the vector PI' (s).
Each component of this vector we write as a sum of
Dirac impulses multiplied by a batch quantity. Then
PI ' (s) will be written as
PI '(s)"
PM

eQR#PM
`
eQR`#2#PM
'"''
eQR'"''
PM
`
eQR#PM
``
eQR``#2#PM
`L'`'
eQR`L'`'

PM
,
eQR,#PM
,`
eQR,`#2#PM
,L','
eQR,L','
0

0
,
(17)
where n(k) is the number of batches of end item
k produced during K, PM
IJ
the size of the lth batch of
item k, l "1, 2,
2
, n(k), t
IJ
the completion time of
lth batch of end item k. Completion times of lower-
level items (including capacities) are then given by
the L4L policy, see below.
Associated with the end-item production vector
PI ' (s) we dene a vector of completion times I(s) for
end items as
J '(s)"
eQR#eQR`#2#eQR'"''
eQR`#eQR``#2#eQR`L'`'

eQR,#eQR,`#2#eQR,L','
0

0
. (18)
These are decision variables. Low-level completion
times are then obtained from the relation (16). Ex-
amining the NN' last rows of the generalised
R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375 367
Leontief inverse HI*"(I!H'tJ(s)), cf. the
example of Section 5, an element in row k and
column j may be written as the sum
hI*
IH
(s)"
K
g
IHK
eQOIHK, (19)
where the g
IHK
are constants and the exponents are
made up of cumulative lead times t
IHK
formed as
sums of the original lead times. The internal de-
mand and therefore the production of a lower-level
item is then
PI
I
(s)"
,

H
hI*
IH
(s)PI
H
(s)
"
,

K
g
IHK

R
PM
HJ
eQ'OIHKRHJ', k'N'. (20)
Lower-level completion times for item k, due to
internal demand from end item j, are thus formed
by relations of the type t
IJ
( j)"t
HJ
!t
IHK
, for posit-
ive coecients g
IHK
, where l '"1, 2,
2
. Whether or
not the index m takes on more than one value,
depends on the structure of the input matrix H. In
the example above for k"4, j"1 (item C as inter-
nally demanded for the production of item A), we
have m"1, 2 and for k"5, j"1 (item D de-
manded for A) then m "1, 2, 3. The greater the
number of occurrences of an item in a product
structure tree, the greater the number of set-ups of
this item will be when following an L4L policy.
Having determined the completion times of all
lower-level items, when given the completion times
of end items, the vector J '(s) may have its NN'
lower components exchanged, forming the vector
of all completion times J (s):
J (s)"
eQR#eQR`#2#eQR'"''
eQR`#eQR``#2#eQR`L'`'

eQR,#eQR,`#2#eQR,L','
, (21)
where a lower-level component of I(s) is determined
by
J (s)"
,

H
L'H'

K
eQ'OIHKRHJ', k'N'. (22)
Both vectors J '(s) and J (s) contain the transforms
of sequences of unit impulses corresponding to the
completion times of items (excluding capacities).
The lower-level completion times are not decision
variables, but depend on the end-item completion
times through the L4L policy and the associated
lead times.
8. Objective function and available inventory
constraints
We are now in a position to formulate the
objective function in terms of all decision variables.
Taking the net present value as the objective and
using j for the continuous interest rate, we may
write
E(NPV)"r(E(DI(j))!jE(BI(j))!E(B(I))eM2

)
!cPI(j)!KtJ(j)J (j), (23)
where r and K are N-dimensional row vectors con-
taining the components r
H
(sales prices) and K
H
(set-
up costs), respectively, and c an (N#M)-dimen-
sional row vector with components c
H
(unit
incremental production costs). The lead-time matrix
tJ(j) entering the last terms accounts for allocating
the set-up costs at the beginning of the lead time,
since J (j) is dened from the completion dates. The
production costs have been allocated at the com-
pletion dates by using the production vector PI(j).
A more detailed description might have been to
allocate production costs somewhere within the
lead time interval. This timing is only a question as
to the denition of the components of c which may
be individually adjusted for taking care of such
timing considerations.
In this NPV expression we have used the NPV
short-cut by exchanging the complex frequency
s for the continuous discount rate j [5].
In Eq. (23), E(DI(j)) is an N-dimensional column
vector of components representing expected ex-
ternal demand discounted to time zero with no
consideration to stockouts, jE(BI(j)) is a similar
vector of discounted changes in backlogs, sE(BI(s))
being the transformof the time derivative of E(B(t)),
and E(B(K))eM2

contains correction terms for the


sake of backlogs not being satised at the horizon
K, since the process then terminates and no more
368 R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375
cPI(j)"c

I
H"tJ(j)

(I!H'tJ(j))PI '(j)
"c

I
H",tJ(j)

(I!H'tJ(j))
PM

eMR#PM
`
eMR`#2#PM
'"''
eMR'"''
PM
`
eMR`#PM
``
eMR``#2#PM
`L'`'
eMR`L'`'

pN
,
eMR,#PM
,`
eMR,`#2#PM
,L','
eMR,L','
0

0
, (27)
production is planned (turning the stockouts there
into lost sales). Since we are dealing with renewal
processes for demand, the expression E(B(K))eM2

only depends on the nal cumulative production


levels PM
IL'I'
and on no other decision variables. For
items not externally demanded, the corresponding
components are zero-valued. Since we are not
allowing any external demand for internally de-
manded items, the vectors E(DI(j)) and E(BI(j)) have
zero-valued elements in their NN' lower positions.
The (N#M)-dimensional column vector PI(j)
contains discounted values of all production (in-
cluding capacity usage) and is obtained from
Eq. (16) inserting j instead of s, where PI' contains
the decision variables concerning amounts and
their timing regarding externally demanded items.
We evaluate the terms in Eq. (23), each in turn.
For a renewal process, the external demand of
item k (which is truncated at the horizon t"K)
obeys
E(DI
I
(j))"
1
2i
@>'`

U@'`
fI
I
(w)
1!fI
I
(w)
)
1!e' MU'2

j!w
dw,
(24)
where fI
I
(j) is the transform of the probability den-
sity function of the stochastic time interval between
consecutive demand events. This formula applies to
any renewal process. The term is void of decision
variables.
Using the stockout function (13), the expected
discounted backlog follows:
E(BI
I
(j))"
1
2i
L'I'

J"
@>'`

U@'`
fI.
M
IJ
I
(w)
w(1!fI
I
(w))
)
e'MU'RIJ!e' MU'RIJ>
j!w
dw. (25)
The nal stockout is evaluated as
E(B
I
(K))"

fI.
M
IL'I'>
I
(s)
s(1!fI(s))
R2
K
"
1
2i
@>'`

U@'`
fI.
M
IL'I'>
I
(s)
s(1!fI
I
(s))
eQ2
K
ds. (26)
The discounted production cost term in vector
form may be written as
where all elements of the coecient vector
c

I
H"tJ

(I!H'tJ)
are void of decision variables. Finally, the set-up
cost term follows:
KtJ(j)J (j)"
K
eM'RO'#eM'R`O'#2#eM'R'"''O'
eM'R`O`'#eM'R``O`'#2#eM'R`L'`'O`'

eM'R,O,'#eM'R,`O,'#2#eM'R,L','O,'

(28)
R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375 369
These equations show that when considering an
L4L lot-sizing policy with no joint set-ups, there
will be a substantial degree of independence be-
tween decisions regarding the amounts of end items
and their timing. Hence, the optimisation of the
production of any end item can be carried out
completely independently of the optimisation of
every other end item.
For the rst batch of each item there might be
a lead time constraint in this multi-level case. In the
one-level case treated in Ref. [1] we have been able
to disregard the eects that lead times impose on
the feasible set of values of the decision variables. In
the current multi-level system, lead times might
impose constraints due to necessarily avoiding pro-
duction decisions to be taken in past time. Such
constraints are due the non-negativity of available
inventory.
Disregarding capacity requirements at rst,
available inventory is given by
RI(s)
"

R(0)!B(0)#(I!H'tJ(s))PI(s)!DI(s)
s
>

,
(29)
where PI(s) has its original denition as total pro-
duction of all items excluding capacity. In this
expression, all vectors are N-dimensional. Regard-
ing end items (the N' rst components), the non-
negativity is taken care of by the backlogging op-
portunity. We may therefore conne our interest to
the intermediate set of indices, N'#1, N'#2,
2
,
N, which concern lower-level items. For this set of
indices, there is no external demand nor any back-
log and the corresponding components of both DI(s)
and B(0) are zero-valued. Furthermore, from the
L4L policy we have PI(s)"(I!H'tJ(s))PI '(s),
which implies
RI(s)"

R(0)!B(0)#PI '(s)!DI(s)
s
>

,
(30)
with the consequence that there is no net produc-
tion of lower-level items, i.e. that available inven-
tory of each of these stays the same. Eq. (16) means
that total production is the weighted sum of the N'
rst columns of the generalised Leontief inverse
(I!H'tJ(s)), the weights being end item produc-
tion variables of the kind PM
HJ
eQRHJ for the jth end
item produced in its lth batch. As discussed in
Section 7, the generalised Leontief inverse
(I!H'tJ(s))"I#H'tJ#(H'tJ)`#
2
will in-
clude a sum of weighted exponential functions

K
g
IHK
eQOIHK where the g
IHK
are constants and the
t
IHK
are cumulative lead times made up as sums of
various lead times in the exponents appearing in
t` (s). For production to take place in present or
future time we must therefore require that t
HJ
*t
IHK
,
for all k, j, l and m, which means that
t
HJ
*Max
IK
t
IHK
"t*
H
, for all j, where t*
H
is an ab-
breviation.
This result means that the rst batch of each item
cannot be completed before the cumulative lead
time (for all its components and sub-components)
has elapsed. This sets minimum constraints on the
timing decisions of the rst batches.
If we also wish to take the timing constraints for
utilising capacity into account, the N-dimensional
PI(s) above is exchanged for the (N#M)-dimen-
sional extended vector
PI(s)"

I
H"tJ(s)

(I!H'tJ(s))PI '(s)
and the same reasoning may be applied to the
exponents entering the components of this vector,
possibly prolonging the cumulative lead times com-
pared to when not taking capacity requirements
into account.
9. Necessary optimisation conditions
Forming a Lagrangean function
"E(NP)#
,

I
z
I
(t
I
!t*
I
), (31)
where the z
I
are non-negative multipliers, the ne-
cessary optimisation conditions for a maximum of
the NPV are derived as
c
ct
IJ
"j eMRIJ

!r
I
([E(B
I
(t
IJ
))]
J
![E(B
I
(t
IJ
))]
J
)#a
I
(j)PM
IJ
#K
I
eMO)
370 R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375
Table 1
Economic parameters
Item Revenue r Unit production
cost c
Set-up
cost K
A 5000 100 50
E 2000 200 100
B 150 50
C 50 20
D 10 10
Table 2
Demand parameters for end items
Item Mean of interval j Variance of
interval o`
A 1 1
E 2 4
Table 3
Optimal production decisions for end items A and E
Item A Item E
Batch k PM
I
t
I
SS
I
PM
I
t
I
SS
I
0 0 0 !32.00 0 0 !15.00
1 49 32.00 0.96 27 30.00 !0.09
2 65 48.04 2.16 41 54.18 0.32
3 80 62.84 2.96 57 81.37 7.00
4 95 77.04 3.37
5 111 91.63 11.00
#
H,
K
H
eMOH
K
eMOHK

#z
I
"0,
k)N', l)n(k), (32)
c
ct
IJ
"!r
I
j

[E(BI(j))]
.
M
IJ>
![E(BI
I
(j))]
.
M
IJ
!

J
eM2
K
j

fI.
M
IL'I'>(s)
s
R2
K

!a
I
(j)(eMRIJ!eMRIJ>#
J
eM2
K
))0,
k)N', l)n(k), (33)
where PM
IJ
denotes cumulative production after the
lth batch has been completed, where z
I
in Eq. (32) is
only included for l"1, where
a
I
(j)"

I
H"tJ(j)

(I!H'tJ(j))

'"" I
,
where the inequality in Eq. (33) accounts for the
condition that the PM
IJ
are integers, and where
J
is
unity for l"n(k) and zero otherwise. The basic
expressions in Eqs. (32) and (33) have been taken
over from Ref. [1]. If z
I
'0, then t
IJ
"t*
I
.
The corresponding sucient conditions become
rather involved. However, simple numerical exam-
ples indicate that in normal cases the optimum will
be unique.
10. A numerical example
We use the major part of the previous example
from Section 5 (cf. Fig. 3) to illustrate the optimisa-
tion procedure proposed. The capacity require-
ments are ignored below. Numerical values for the
economic parameters used in the example are listed
in Table 1. Revenues are generated only from the
end items A and E. Stochastic demand parameters
are also assumed to be dierent for these two nish-
ed items. The time interval of demand follows
exponential distributions for A and E.
Using the economic and demand parameters
listed in Tables 1 and 2, we can solve the problem
from Eqs. (31)(33). Since only the production of
nished items are decision variables, we rst calcu-
late the optimal production decision for items
A and E following a similar procedure to the one
proposed in Ref. [1]. The results are listed in
Table 3. The expression SS
I
refers to the optimal
safety stock immediately prior to receiving batch
k#1.
The optimal batch number is 5 and 3 for A and E,
respectively. As soon as Table 3 is obtained,
production decisions for the lower-level items are
immediately calculated according to Eq. (16). The
results are listed in Tables 46.
Since an L4L policy is applied for the lower-level
items, the optimal decisions are independent be-
tween dierent nished items. This brings simpli-
city when solving the safety stock problem based
R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375 371
Table 4
Optimal production decisions for item B
Source of requirement
Item A Item E
Batch k PM
I
t
I
Batch k PM
I
t
I
1 196 18.00 2 27 18.00
3 260 34.04 4 41 42.18
5 320 48.84
6 380 63.04 7 57 69.37
8 444 77.63
Table 5
Optimal production decisions for item C
Source of requirement
Item A Item E
Batch k PM
I
t
I
Batch k PM
I
t
I
1 196 8.00 2 27 8.00
3 98 18.00
4 260 24.04 5 41 32.18
6 130 34.04
7 320 38.84
8 160 48.84
9 380 53.04 10 57 59.37
11 190 63.04
12 444 67.63
13 222 77.63
Table 6
Optimal production decisions for item D
Source of requirement
Item A Item E
Batch k PM
I
t
I
Batch k PM
I
t
I
1 392 0.00 2 54 0.00
3 392 8.00 4 54 8.00
5 196 10.00
6 520 16.04 7 81 18.00
8 82 24.00
9 520 24.04
10 260 26.04
11 640 30.84 12 82 32.18
13 640 38.84
14 320 40.84 15 123 42.18
16 760 45.04 17 114 51.37
18 760 53.04
19 380 55.04 20 114 59.37
21 888 59.63
22 888 67.63 23 171 69.37
24 444 69.63
on the optimisation conditions (31)(33). The solu-
tion for the production decision for lower-level
items is also straightforward. On the other hand,
the application of an L4L policy prevents addi-
tional possible improvements of the production
decisions. For example, we lose the opportunity to
combine batches 8 and 9 of ItemD (Table 6) which
would reduce the combined set-up costs involved.
11. Relaxing the lot-for-lot policy
As a nal topic discussed in this paper, we con-
sider the relaxation of the L4L policy for producing
lower-level items, assuming an assembly system
structure. Internal demand for a lower-level item
may be generated from one or several down-stream
products. If the down-stream production is given,
then the internal demand would be given as a stair-
case function possibly made up as the weighted
sum of staircase functions generated by each
down-stream item acting as a source of these re-
quirements.
Below we show that when internal demand is
given, then there is a nite number of staircase
structures for the production of the lower-level item
that need be investigated in order to cover all
candidates for an overall optimum when applying
the NPV criterion. This result is alike, but slightly
more general than the similar result discussed in
Ref. [6].
In Ref. [6] the following remark was made for
the lower-level production in a serial two-level sys-
tem applying an average cost principle:
Whatever staircase the upper-level cumulative
production follows, at the optimum there can
never be a gap between the lower-level cumulat-
ive function and the highest of the steps of the
upper function within a particular lower-level
production interval. If there were such a gap, the
lower-level production level could be lowered
reducing inventory holding costs with no other
eects. Also, whatever staircase the upper-level
cumulative production follows, at the optimum
there can never be a gap in time between the next
replenishment of the lower level and the next
replenishment of the upper level, because if this
372 R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375
Fig. 5. A two-level serial system with the left staircase showing
cumulative lower-level production and the right showing cumu-
lative upper-level production (which generates internal de-
mand"solid left staircase curve). Inner corners are marked
with circles.
Fig. 6. Lower-level production opportunities and internal
demand. Moving a to the right or b downwards creates in-
crements to the net present value.
was the case, then the lower-level replenishment
could be postponed reducing holding costs with
no other eects. Therefore in the two-level serial
case, we note:
In the optimal solution, the lower-level replenish-
ments always must take place at the same time as
some upper-level replenishment and at that time
(immediately prior to the replenishment in ques-
tion) both cumulative levels coincide.
Geometrically, this was described by considering
variations from the upper-level staircase as illus-
trated in Fig. 5. The two staircases t into each
other at the inner corners when lower-level re-
plenishments take place (indicated with circles in the
gure). Thus, the optimal number of batches on the
lower-level cannot be higher than on the upper level.
We use the left-hand side of the same gure in
our current multi-level case in order to ascertain the
following similar conclusion. The solid left-hand side
staircase is interpreted as the sum of all internal
demand from given down-stream production.
In the optimal solution, the lower-level replenish-
ments must always take place at the same time as
some upper-level item generates internal demand and
at that time (immediately prior to the replenishment
in question) cumulative production of the item con-
sidered must coincide with cumulative internal
demand.
To prove this, we rst note that cumulative pro-
duction always must be to the left and above inter-
nal demand according to the non-negativity of
available inventory with the possibility that the
curves follow each other as a whole or partly. If
a certain replenishment were to take place before
a vertical step in internal demand (cf. a in Fig. 6),
then the replenishment could be postponed until
the same time as the step with the only consequence
that the discounted cost of production and set-up
decreased. On the other hand, if the replenishment
were to correspond to a step reaching a point in an
interval strictly between two cumulative levels of
internal demand (b in Fig. 6), then lowering the
batch quantity would have no other consequence
than decreasing an earlier production cost and rais-
ing a later production cost by the same nominal
amount. Taking the discount rate into considera-
tion, this would raise the resulting NPV.
Hence, the lower-level production staircase must
have inner corners that coincide with its internal
demand generated from the given down-stream
production. The lower-level staircase structures
that may act as candidates for an overall optimal
production are therefore nite in number for each
set of given down-stream production staircases.
For a two-level case, the number of lower-level
structures to be considered when internal demand
has n steps will be 2'L' (as already shown in Ref.
[6]) and in a three-level case, assuming the third
level only to have the second level as a source for
internal demand (a serial structure in this respect),
we obtain altogether 3'L' structures to examine.
In the general serial system of m levels (the upmost
level being internal demand generated from end
items), the total number of lower-level structures
will be m'L'.
R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375 373
This principle also immediately results in an inte-
grality condition between batch sizes on dierent
levels for a serial system (which has been extensive-
ly discussed in the literature). If consecutive batches
on an upper level are equally sized, then batches on
the next level down will be equally sized and will
amount to an integral multiple of the upper level size.
12. Conclusions
Our main ndings are that the single-level results
[1] are applicable in all essential respects also in the
multi-level case when the L4L lot-sizing policy is
applied to all subordinate items of an assembly
system on condition that joint set-up opportunities
on lower levels are not taken into account. The
optimisation conditions for the multi-level system
treated above have been shown to be simple gener-
alisations of those of the single-level system.
The next step in this line of work would be to
examine further numerical examples based on the
optimisation conditions derived. Some work in this
direction has already been carried out in Ref. [21].
As regards the relaxation of the L4Lpolicy discussed
in Section 11, some future interest could be devoted
to whether there might exist myopic conditions that
could be derived for rejecting various alternatives
with the aim to decrease the number of cases created
by the subordinate itemlot-sizing alternatives. A fur-
ther but more complex issue raised above concerns
the question of joint set-ups. This would rst need
to be handled in more theoretical detail.
13. Further Reading
The following references are also of interest to
the reader: [720] and [2226].
Acknowledgements
The author is deeply indepted to Ou Tang, Dept
of Production Economics, Linko ping Institute of
Technology, for many valuable discussions held
with him, and for his work in preparing the numer-
ical example of Section 10.
References
[1] R.W. Grubbstro m, A net present value approach to safety
stocks in planned production, International Journal of
Production Economics 56/57 (1998) 213229.
[2] R.W. Grubbstro m, Stochastic properties of a production-
inventory process with planned production using trans-
form methodology, International Journal of Production
Economics 45 (1996) 407419.
[3] L.E. Andersson, R.W. Grubbstro m, Asymptotic Behaviour
of a Stochastic Multi-Period Inventory Process with Plan-
ned Production, Working paper WP-210, Dept. of Pro-
duction Economics, Linko ping Institute of Technology,
April 1994.
[4] A. Segerstedt, Multi-level production and inventory con-
trol related to MRP and cover-time planning, Produc-
tion-Economic Research in Linko ping, PROFIL 14,
Linko ping, 1995.
[5] R.W. Grubbstro m, On the application of the Laplace
transform to certain economic problems, Management
Science 13 (7) (1967) 558567.
[6] R.W. Grubbstro m, A. Molinder, Safety production plans
in MRP-systems using transform methodology, Interna-
tional Journal of Production Economics 4647 (1996)
297309.
[7] J.A. Aseltine, Transform Method in Linear System Analy-
sis, McGraw-Hill, New York, 1958.
[8] R.V. Churchill, Complex Variables and Applications,
McGraw-Hill, New York, 1960.
[9] R.V. Churchill, Operational Mathematics, McGraw-Hill,
New York, 1958.
[10] R.W. Grubbstro m, The z-transform of tI, The Mathemat-
ical Scientist 16 (1991) 118129.
[11] A. Vazsonyi, The use of mathematics in production and
inventory control, Management Science 1 (1) (1955) 7085.
[12] R.W. Grubbstro m, A principle for determining the correct
capital costs of work-in-progress and inventory, Interna-
tional Journal of Production Research 18 (2) (1980)
259271.
[13] R.W. Grubbstro m, Y. Jiang, A survey and analysis of the
application of the Laplace transform to present value
problems, Revista di Matematica per le Scienze Eco-
nomiche e Sociale 12 (1) (1990) 4362.
[14] A. Thorstenson, Capital costs in inventory models a dis-
counted cash ow approach, Production-Economic Re-
search in Linko ping, PROFIL 8, Linko ping, 1988.
[15] A. Thorstenson, R.W. Grubbstro m, Theoretical analysis
of the dierence between the traditional and the
annuity stream principles applied to inventory evaluation,
in: R. Flavell (Ed.), Modelling Reality and Personal
Modelling, Physica-Verlag, Heidelberg, 1993, pp.
296326.
[16] R.W. Grubbstro m, O. Tang, Further developments on
safety stocks in an MRP system applying Laplace trans-
forms and inputoutput analysis, Proceedings 14th Inter-
national Conference on Production Research, Osaka, 48
August 1997, pp. 17181721.
374 R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375
[17] L. Bogataj, L. Horvat, Stochastic considerations of Grubb-
stro mMolinder model of MRP, inputoutput and multi-
echelon inventory systems, International Journal of Pro-
duction Economics 45 (1996) 329336.
[18] L. Bogataj, Lead times and other delays in inventory
systems, in: L. Bogataj (Ed.), Inventory Modelling, Lecture
Notes of the International Postgraduate Summer School,
vol. 1, ISIR, Budapest, 1995, pp. 2335.
[19] L. Bogataj, L. Horvat, The application of game theory to
MRP, inputoutput and multi-echelon inventory systems
with exponentially distributed external demand, Proceed-
ings of the International Symposium on Operations Re-
search SOR 95, Portorozy , Slovenia, 1995, pp. 2536.
[20] R.W. Grubbstro m, Material requirements planning and
manufacturing resource planning, in: M. Warner (Ed.),
International Encyclopedia of Business and Management,
Routledge, London, 1996.
[21] R.W. Grubbstro m, A. Molinder, Further theoretical consid-
erations on the relationship between MRP, Inputoutput
analysis and multi-echelon inventory systems, International
Journal of Production Economics 35 (1994) 299311.
[22] R.W. Grubbstro m, P. Ovrin, Intertemporal generalization
of the relationship between material requirements plann-
ing and inputoutput analysis, International Journal of
Production Economics 26 (1992) 311318.
[23] L. Horvat, L. Bogataj, MRP, inputoutput analysis and
multi-echelon inventory systems with exponentially dis-
tributed external demand, Proceedings of GLOCOSM
Conference, Bangalore, India, 1996, pp. 149154.
[24] L. Horvat, L. Bogataj, Game theory aspects of MRP,
inputoutput and multi-echelon inventory systems,
Pre-prints, vol. 3 Ninth International Working Seminar
on Production Economics, Igls/Innsbruck, 1996, pp.
112.
[25] L. Horvat, L. Bogataj, Multi-attribute decision making
upgrading of Grubbstro mMolinder model of the rela-
tionship between MRP and inputoutput analysis, to
appear.
[26] A. Molinder, Material requirements planning employing
inputoutput analysis and Laplace transforms, Produc-
tion-Economic Research in Linko ping, PROFIL 14,
Linko ping, 1995.
R.W. Grubbstro m/Int. J. Production Economics 59 (1999) 361375 375

Das könnte Ihnen auch gefallen