engineering
Costs and ~rod~~t~o~ Economics, 20 ( I 990)
1I 3-
113
120
Elsevier
MULTI-OBJECTIVE, MULTI-PER100 PLANNING FOR A
MANUFACTURING
PLANT
T. Ttzaskalik
iffst;tute of Econometrics, Academy of Economics, ui. Bogucicka 74,40-226
Kato~jce,
Poland
ABSTRACT
This paper aims at introducing a formalized
description of work in an industrial enterprise
and using it for laying out optimal multi-period
economic plans. In that description basic input
and output variables, relations and connections
with environments all important technological
dependencies and derision functions can be
taken into consideration.
Industrial enterprises differ, for instance in
size, assortment of production etc. We want to
derive a solution useful for varied types of enterprises and that is why we introduce the concept
of the manufacturing plant.
A new interactive hierarchical dynamic pro-
gramming technique, briefly described in the
paper, is applied to the solution of the problem.
An example problem is formuIated and solved
with the following objectives taken into consideration: (i) minimization of utiIization of deftit materials, {ii) m~~mizat~on of production
size, and (iii) maximization of profits.
We consider muiti-objective, multi-period
planning in the manufacturing plant as a deterministic problem.
INTRODUCTION
crete dynamic pro~amming,
developed by
Brown and Strauch [3], Klatzler (41, Mine
and Fukushima [ 5 1, Yu [ 6 1. A wide review of
literature in the field of multi-criteria dynamic
programming was made by Li and Haimes [ 71.
There are many concepts of describing
mathematical aspects of planning for industrial enterprises. Trzaskalik [ I ,2] introduced
the concept of the manufacturing plant. In the
present article we describe the problem of
choosing the optimal schedule for a manufacturing plant in the fixed space of time in the
future divided into a finite number of periods.
We take into consideration more than one objective and that is why we describe the problem as that of optimal multi-objective, multiperiod planning.
Methods of solving problems of that sort are
naturally connected with multi-criterion dis0167-l 88X/90/$03.50
0 1990-Elsevier
Science Publishers B.V.
MANUFACTURING
PLANT
We define a manufacturing plant as a set of
features describing really existing enterprises.
We assume, that:
( 1) The work of the manufacturing plant is
considered in the fixed space of time in the future divided into T basic periods. During the
considered space of time the manufacturing
114
plant leads the activities aimed at producing
products of specified sorts.
(2 ) Production technologies are used for
laying out products. Production technology is
a logical and usefully well-ordered set of operations, as a result of which we obtain a particular product. It is possible to fix the set of different kinds of products which can be
manufactured and the set of production technologies which can be used in the production
process in the fixed space of time.
(3) Interaction between the manufacturing
plant and the environment takes place. The
manufacturing plant receives new production
technologies, materials used in the production
process and manpower from the environment
and delivers manufactured products to the
environment.
(4) Production possibilities of the manufacturing plant are determined by an activity
condition vector in each period t (t= 1, . ... T).
Components of this vector concern possibilities of:
using production technologies,
storage of materials and products,
putting new production technologies into
service,
buying materials,
utilization of labour resources,
selling manufactured products,
values of technological coefficients,
prices of materials and products,
costs per unit of production.
We consider the simplified version of the
model. We assume that materials bought in a
fixed period are utilized for production in the
same period and products manufactured in a
fixed period are sold in this period. According
to these assumptions we can omit the inventory problem and reduce dimensions of the
problem.
In this simplified model, components of the
state vector at the beginning of each period are
accumulated amounts of products manufactured by the use of separate technologies before the beginning of the fixed period. Coml
l
l
l
l
l
l
l
l
ponents of the decision vector are lots of
products manufactured by the use of particular production technologies in the fixed period.
In the fixed space of time the following objectives may be taken into consideration:
(i) maximization of profits,
(ii) maximization of production size,
(iii) maximization of production quality,
(iv) minimization
of utilization of deficit
materials,
(v) minimization of costs.
It is also possible to consider other objectives.
In further considerations we accept for simplicity that each criterion for the fixed space of
time is a sum of corresponding criteria for the
following periods.
MULTI-CRITERIA,
PROGRAMMING:
APPROACH
DISCRETE DYNAMIC
HIERARCHICAL
We consider a decision process which consists of zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPO
T periods. Let Y, be the set of all possible state variables in period t, and X,( yt) the
set of all permissible decision variables for period t and state yt~ Y,, t = zyxwvutsrqponmlkjihgfedcb
1,...,T. We assume that
all these sets are finite. Let D,, the set of all period realizations in period t, be defined as
o,={d,=(y,,x,):y,EY~,XrEXr(yr))
(1)
We assume that for t= l,...,T there are given
transformations Q,: D,p Y,. A sequence of period realizations
d= (d, ,...dr) = (Y, 7x1 ,...,YT,XT)
(2)
is said to be a process realization if
Y,+1=Q(yrJ,)
Vt= l,...,T
(3)
Let D be the set of all process realizations.
We assume that in each period t there are defined K period criteria functions F f : D, --+R
(t= 1,..., T, k= l,..., K). F is the vector-valued
criteria function for the whole process and its
components F ’ ,...,FK are understood as
Fk=Fk(F:,...,Fk,),
k= l,...,K
(4)
115
period structure of the problem is not taken
into consideration. We describe interactive
modification of this approach excluding these
shortcomings.
There are given: the quasi-hierarchy of multiP(d)= i FF(d,)
(5)
period
criteria, the multi-period tolerance
I= I
coefficients ek for k= I,...,K and the one-peWe postulate maximization of all components
riod tolerance coefficients pf for k= 1,...,K,
Fk.
t=l ,..., T. Let be D’=D. At the kth step
Realization p is called ef~cient if zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCB
(k= l,...,K) the set D“ is known. We find the
value
3&D
F(d)&F(d*)
(6)
Components Fk are said to be rnu~t~-periodcriteria functions. The simplest and most important form for applications is the additive
function
Let D* be the set ofaIi efficient realizations of
the process. Let us determine &D. The set of
ail efficient realizations better than d, written
as D*(d), is defined as
D*(d)={d*cD*:F(d*)aF(d))
(7)
One of the methods used for multi-criteria
problems is the hierarchical approach, which
we apply to the problem of multi-criteria, discrete dynamic programming. We assume that
multi-period criteria are numerated in such a
way that a more important criterion has a lower
number than a less important one. The method
of solution is sequential. At the first step we
maximize the most important criterion using
the one-criterion
dynamic
programming
method and we get a set of realizations. Then,
from the set received at the previous step, we
choose those realizations which maximize a
criterion of lower impo~ance,
using the
method of the full review. On the Kth step we
get the solution of the problem.
We can modify this approach introducing
quasi-hierarchy of criteria. Transition from the
more important multi-period criterion number k to the criterion of lower importance
number k+ 1 is possible, when the value of the
kth criterion reaches the satisfactory (not necessarily optimal) level calculated by using the
maximum value Ek and the given tolerance
coefficient ek for the kth criterion.
There are two shortcomings in the approach
described above. Firstly, the necessity of using
the method of the full review and secondly, the
P”=max{Fk(d):dcDk)
(8)
and the set of realizations Dk (c k, defined as
b”(tk)={dkEDk:Fk(dk)
> ( 1-tk)Ek}
(9)
From this set the decision maker chooses the
model realization dk= (df,...,&)
which is
characterized
by the best period values
Ff(df). Then we generate the set Dkf ’ consisting of these realizations which are the most
similar to the model realization. The set Dk+ ’
is defined as
(10)
Usually dK( eK) c D*, so we propose the decision maker to choose the final realization
among the realizations belonging to B”(eK)
and all efftcient realizations better than realizations belonging to BK( rK). If he is not satisfied, he can repeat the procedure taking other
model realizations at following steps.
EXAMPLE - FORMULATION
FORMALIZATION
AND
We are interested in the work of the manufacturing plant in the fixed space of time divided into two periods. It is possible to manufacture two sorts of products: P, and PZ.
Product P, can be manufactured using technologies Ti and T2 and product P2 by using
technologies T3 and T+ At the beginning of the
present space of time, technologies T1 and T,
can be implemented and the remaining tech-
116
(iii) maximization of profits.
Each multi-period criterion is the sum of the
values of the corresponding one-period criteria.
Characteristics
Technology
The problem of laying out an optimal, multi2
3
4
5
1
6(b)
7
6(a)
period plan for a manufacturing plant can be
described as a multi-criteria, discrete decision
0,3,5
2
1 2
0
7
8
5
T,
process. Decision vector x, (called also produc0,2,4
1
1
1
10
7
8
4
T2
0,4,7
1
2
2
0
10
11
6
tion vector) for the period t ( t= 1,2 ) consists
T3
0,2,4
3
2
1
10
10
11
5
T,
of4 components:
(1) component xlt designs lots of products P,
manufactured in the period t by the use
nologies must be put into service. Values of
of
technology T,,
characteristics for technologies T, to T, are
component
xzI designs lots of products P1
(2)
shown in Table 1. The respective columns of
manufactured
in the period t by the use
this table describe: ( 1) lots of products which
of technology T,,
can be manufactured in one period; (2) lots of
(3) component x3, designs lots of products P2
material M 1 per unit of production; ( 3 ) lots of
manufactured in the period t by the use
material M2 per unit of production; (4) lots of
of technology T3.
labour per unit of production; ( 5 ) costs of put(4) component x41designs lots of products P2
ting technologies into service; (6) income from
manufactured in the period t by the use
selling production (per unit): (a) in the first
of technology T4.
period; zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
(b ) in the second period; and (7 ) costs
State vector yI for the period t (t= 1,2,3) conof manufacturing per unit of production.
sists also of 4 components:
Values of all characteristics, except character(1) component yll designs lots of products P1
istic number 6, are the same in the first and the
manufactured from the beginning of the
second period.
process to the end of the period t - 1 by
It is impossible to manufacture in one pethe use of technology T 1.
riod more than 9 units of products using tech(2) component y2t designs lots of products P,
nologies T1 and Tz, more than I1 units using
manufactured from the beginning of the
technologies T3 and T, and more than 14 units
process to the end of the period t- 1 by
using all technologies. Moreover, it is imposthe use of technology Pz.
sible to manufacture both in the first and seccomponent
y,, designs lots of products P2
(3)
ond period more than 6 units of products using
manufactured from the beginning of the
technology T1, more than 7 units using techprocess to the end of the period t- 1 by
nology TZ, more than 12 units using technolthe use of technology T3.
ogy T3 and more than 10 units using technol(4) component y41designs lots of products P2
ogy T,. Minimum production of product P, in
manufactured from the beginning of the
one period is 4 units; the same holds for the
process to the end of the period t - 1 by
product PZ. In each period material M,, matethe use of technology T4.
rial MZ, resources of labour and funds for preWe accept y1 = [0,zyxwvutsrqponmlkjihgfedcbaZYXWVUTSR
zyxwvutsrqponmlkjihgfedcbaZ
0, 0, O]=. Transition jiuncparing technologies are 18, 17,30 and 10 units,
tions for t= 1,2 have the form:
respectively.
(11)
Yr+,=Yr+&
We consider the following objectives:
For each ae[Wwe define an auxiliary function zyxwvutsr
(i) minimization of utilization of deficit material MI.
1 fora>O
A(a)=
(12)
(ii ) maximization of production size,
{ 0 fora<
TABLE
1
117
According to the accepted assumptions we have
to take into account the following constraints:
l constrains
for lots of products which can
be manufactured in one period by the use of
particular technologies:
(29)
Now we consider the criteria functions. Utilization of deficit material 1 for the period t is
calculated according to the formula:
X,,E{O,3,5)
(13)
~+,2,4)
~r(x,)=2x,,+X2r+Xjr+3X4,
(30)
(14) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH
X3*E {0,4,7)
(15)
~~,~p,2,4~
(16)
The multi-period utilization function for material 1 is calculated as
@(x,7x2) =
l
additional constraints for technologies:
x,t+x2,<7
(17)
X3, +x42, G 11
(18)
@, (XI I+
(31)
@*(x2)
Because we have postulated maximization of
all multi-period components of the vector objective function, a new function
~‘f-u,,x2)=-@(X,,%)
(32)
is defined. Production size for period t is calculated according to the formula
F:(x,)=5 xi,
r=,
XI,+x,z~6
(20)
<7
(21)
x21 +x22
x3, +x32 =s 12
(22)
(23)
.x41+x42 < 10
l constraints
production:
for the
minimum
size of
4ei,,+.x2,,
4<xx,,+x4,
(24)
(25)
constraints for the maximum utilization of
materials:
l
2x,,+.x~,+x,,+3~,,,<18
(26)
x,,+x~,+2.x~,+2x,,~17
(27)
l constraints
concerning the maximum
lization of labour resources:
2x,,+.r,,+2xj,+x4,,<30
uti(28)
l constraints
concerning the possibility of
putting new production
technologies into
service
(33)
The multi-period function of production size is
calculated as follows:
F’(x,,xz)=F:(x,)+F:(“~~?cz)
(34)
Income q(q) for one period is calculated according to the formula:
(y,(~,)=7(~,,
+x21)+10(*~31+x4,)
(35)
~~2(~2)=8(~12
+x22)
(36)
+
11 (X32
+x42)
Costs of production pf(xt) in the period t are
calculated according to the formula:
p,(X,)= 5X,,+4X,,
+6X,,+ %X4,,
(37)
Costs of putting technologies into service
y,(y,,x, ) are calculated according to the
formula:
+lon[~(y,l+X,,)-~(Y41)1
(38)
Profit for period t can be expressed in the following terms:
F:QJ,J,) =%(&)-P,(4) -V,(Y&)
(39)
where (xI,pt and yt are defined according to the
118
3682
(_oooo
Fig. 1.The graph of the example problem.
formulae (35)-( 38). The zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
multi-period profit
6) minimization of utilization of deficit mafunction is calculated as follows:
terial 1,
(ii) maximization
of production size,
F3(yl,X1,y2,XZ)=F:(y,,X,)+F:(yz,X*)
(40)
(iii) maximization
of profit.
The example problem of multi-objective,
We get the following sequence of problems:
multi-period
planning for the manufacturing
plant can be formulated as
Problem 1
“Max”{[F’(d),F2(d),F3(d)lT:deD}
(41)
Max(F’(d):dED’}
where F ‘, F2, F 3 are defined according to
where
(32)) (34), (40) and D is the set of realizations defined according to ( 1 1zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
)- (29 ) .
D'=D
The graph of the example problem is shown
The graph of the process with one-period
in Fig. 1. Nodes of the graph represent states,
values of the first criterion is shown in Fig. 2.
arcs are decisions.
We obtain P ’ = 20. There are two plans corresponding to this value:
APPLICATION
OF THE HIERARCHICAL
x,=[044O]T
~~=[3240]~
METHOD
(42)
Let us assume that the decision maker accepts the following quasi-hierarchy
of criteria:
and
xz=[32401T
X*=[O44O]T
(43)
119
c
0000
Fig. 2. The graph of the process with one-period values of the first criterion.
Fig. 3. The reduced graph of the process with one-period values of the second criterion.
Fig. 4. The reduced graph of the process with one-period values of the third criterion.
120
Let us assume that the decision maker accepts the following values of coefficients:
e’=O, ,ui =0.5, pi =0.25 and takes plan (42)
as the model plan. Hence F 1(dt ) = - 8,
F;(d:)=-12.
If the decision maker accepts e3=O then this
plan is the solution of the problem. It is an efficient plan.
CONCLUSIONS
Problem 2
Max{F2(d):deD2\
where
D’={dcD’:F;(d,)>-12,F:(d,)>-15}
(44)
The reduced graph of the process (according
to (43 ) ) with one-period values of the second
criterion is shown in Fig. 3. We obtain
F2 = 20. There is only one plan corresponding
to this value:
x,=[32401T
xZ=[3440]T
(45)
Let us assume that the decision maker accepts the following values of coefficients:
e2= 0, PU:= 0.2, $ = 0.2. It means that the plan
(45) becomes the model plan. Hence
Ff(d:)=9,F:(6;)=11.
Problem 3
MaxjF3(d):dcD3}
where
D3={dcD2:F:(d,)>8.1,
F:(d,)>,9.8}
(46)
The reduced graph of the process (according
to (46) ) with one-period values of the criterion 3 is shown in Fig. 4. We obtain E3=63.
The only plan corresponding to this value is
_x,= [ 3 2 4 01’
xz=[34401T
(47)
Application of the modified hierarchical approach to multi-criteria, discrete dynamic programming is the way for effective solution of
problems of this sort. This approach has been
used for multi-criteria, multi-period planning
for the manufacturing plant and the optimal
plan corresponding to the chosen quasi-hierarchy of criteria has been obtained.
REFERENCES
Ttzaskalik, T., 1980. Mathematical Models of Multi-Stage
Planning for Industrial Enterprise, Doctoral Dissertation,
Academy of Economics Katowice.
Trzaskalik. T., 1984. Optimal multi-period planning for
manufacturing plant - general problems. In: A. Straszak
(Ed.), Large Scale Systems - Theory and Applications,
Pergamon Press.
Brown, T.A. and Strauch, R.E., 1965. Dynamic programming in multiplicative lattices. Math. Anal. Appl., 12: 364370.
Klotzler, R., 1978. Multi-objective dynamic programming. Math. Operationsforsch. Statist. Ser. Optimization,
9(3): 423-426.
Mine, H. and Fukushima, M., 1979. Decomposition of
multiple criteria dynamic programming problem by dynamic programming. Int. J. Syst. Sci., lO(5): 557-566.
Yu, P.L.. 1978. Dynamic programming in finite stage
multi-criteria decision problems. The University of Kansas, School of Business, Working Paper No 118, 1978.
Li. D. and Haimes, V., 1988. Multi-objective Dynamic
Programming: The State of The Art. VII International
MDCM Conference, Manchester, August 2 l-26.
(Received
1990)
Augusr
1989;
accepred
in revised
form
May