Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Fuzzy Optim Decis Making (2007) 6:29–49

DOI 10.1007/s10700-006-0023-y

Interactive fuzzy linear programming and an


application sample at a textile firm

İrfan Ertuğrul · Ayşegül Tuş

© Springer Science+Business Media, LLC 2007

Abstract The purpose of this study is to examine Interactive Fuzzy Linear Pro-
gramming (IFLP) model by using Zimmermann, Werners, Chanas and Verde-
gay’s approaches that provide best decision-making under fuzzy environments.
In this study, it is used the method which can model the fuzzy structure of the
real world and which operates with the decision maker interactively, which aims
at obtaining the best solution by continuing this interactiveness in the solution
process, which includes fuzziness with more realistic approach to the system. It
is showed that the importance of fuzziness concept for IFLP problems, how it
is applied on real-world problems and its effects.

Keywords Fuzzy sets theory · Fuzzy linear programming · Interactive fuzzy


linear programming

1 Introduction

The conventional quantitative techniques of system analysis are unsuited


for dealing with humanistic systems and other comparable complex systems,
because, as the complexity increases, our ability to make precise and
yet significant statements diminishes until a threshold is reached beyond
which precision and significant relevance become almost mutually exclusive

İ. Ertuğrul (B) · A. Tuş


Department of Bussiness Administration, Major of Quantitative Methods,
The Faculty of Economic and Administrative Sciences,
Pamukkale University, 20070 Kinikli, Denizli, Turkey
e-mail: iertugrul@pamukkale.edu.tr
A. Tuş
e-mail: atus@pamukkale.edu.tr
30 İ. Ertuğrul, A. Tuş

characteristics (Sankar, 1986; Yager & Filev, 1994). It’s generally used the
concepts and methods in probability theory to examine uncertainty. But in
1960s these are renewed and critisized. In the direction of these critics, it has
been studied hard to improve the new methods instead of probability theory
(Ertugrul, 1996).
Zadeh who extends the certainty limited world of mathematics through
uncertainty by degree mechanism, has created a paradigm shift which has taken
great deal of attention all over the world. Fuzzy sets theory providing a more
widely frame than classic sets theory, has been contributing to capability of
reflecting real world and applicability of linear programming (LP) by supplying
a wide moving area to a decision maker.
While obtaining the mathematical models in real-world decision problems
are constituting, two main properties are; objective at problem structure and
fuzziness at problem definition (Arıkan, 1996). In this study, it is used the
method which can model the fuzzy structure of the real world and which oper-
ates with the decision maker interactively, which aims at obtaining the best solu-
tion by continuing this interactiveness in the solution process which includes
fuzziness with more realistic approach to the system.
Fuzzy linear programming (FLP) approach of solving industrial production
planning problem can have feedback within the decision maker, the implemen-
ter and the analyst. In such case this approach can be called as IFLP (Vasant,
2003).The aim of this article is to examine IFLP model by using Zimmermann,
Werners, Chanas and Verdegay’s approaches that provide best decision-making
under fuzzy environments. In this article, after a general introduction about the
subject, it’s mentioned about the fuzzy sets theory, its application areas and
the basic concepts. Then FLP and its models are examined. These models are
analys with the relation of fuzziness and emphasized the aproach differences
with each other. Finally, an IFLP which is the main subject of this study firstly
LP is explained and it has applied to a production planning on household tex-
tile group of a textile firm in Denizli, Turkey and then conclusions and findings
have been interpreted. The aim of this application is to show the importance of
fuzziness concept for IFLP problems, how it is applied on real-world problems
and its effects.

2 Fuzzy sets theory

In the early 20th century, reducing complex real-world systems into precise
mathematical models became the main trend in science and engineering. Since
then, Operations Research (OR) has been applied to real-world decision mak-
ing problems. Traditional OR approaches may not really be suitable for solving
practical decision making problems (Lai & Hwang, 1992). Since fuzzy sets
theory was proposed by Zadeh in 1965, we have been able to handle not
only well-defined, precise data, but also vague or fuzzy data (Kosko, 1997). It
provides to express real world mathematically thus to take part uncertainty
in decison process by passing over certain limitations classical mathematic
Interactive fuzzy linear programming and an application sample at a textile firm 31

create. Common usage of fuzzy sets theory at almost every area of science
and technology expands range traditional OR with new derivations about deci-
sion making at industrial systems. Fuzzy sets theory is suggested in order to
develop for simplified models thus to solve complex systems of real world. Fur-
thermore it helps not only optimize the system by appraising alternatives under
given constraints but also it helps developing new altenatives.
Zadeh defines fuzzy set concept as ranges on a reel line by shaping in a set.
Fuzzy sets are inadequately defined set of objects in which there are not ade-
quate critera for membership. It is impossible to define these objects whether
in the set or not. The set of objects in real life don’t have membership crite-
ria defined clearly. But sets which are defined uncertainly in real life play an
important role in human decision making. So fuzzy sets theory is developed
in order to define and to solve the problems. Fuzzy sets theory which creates
wider frame than classical set theory has a wide application area.
Fuzzy sets theory has a lot of application areas such as OR, management
science, control theory, artificial intelligence/intelligent systems, human behav-
iours and these applications are used commonly in increasing world scale
(Paksoy & Atak, 2002).

3 Fuzzy linear programming

Emprical surveys reveal that LP is one of the most frequently applied OR tech-
niques in real-world problems. However, given the power of LP one could have
expected even more applications. This might be due to the fact that LP requires
much well-defined and precise data which involves high-information costs. In
real-world applications certainty, reliability and precision of data is often illu-
sory. Furthermore the optimal solution of an LP only depends on a limited
number of constraints and, thus, much of the information collected has little
impact on the solution. Being able to deal with vague and imprecise data may
greatly contribute to the diffusion and application of LP (Romelfanger, 1996).
FLP, proposed by Bellman and Zadeh, is an extension of LP with both objective
function(s) and constraints represented by fuzzy sets. The FLP technique has
provided increased flexibility of applying LP to those engineering problems in
which subjective requirements occur in both objective functions and constraints
(Zhao, Govind, & Fan, 1992).
Fuzzy sets theory appears to be an ideal approach to deal with decison
problems that are formulated as LP models but with imprecison parameters.
FLP models are designated for such a purpose. In the literature, FLP has been
classified into different categories, depending on how imprecise parameters
are modeled by possibility distributions or subjective preference-based mem-
bership functions (Guu & Wu, 1999). In this paper, we shall associate FLP
problems with the subjective preference-based membership functions.
The first study on FLP is published by Bellman and Zadeh in 1970 (Maleki
et al., 2001). Then in 1974 Tanaka et al., in 1976 Negoita and Sularia (Romm-
elfanger et al., 1989), in 1976 Zimmermann (Wang & Liang, 2004), in 1978
32 İ. Ertuğrul, A. Tuş

Orlovsky (Delgado et al., 1989) studied on this subject. In 1983 Chanas used
parametric programming for FLP. In 1984 Tanaka and Asai accepted techno-
logical matrix and coefficients of objective function and right side constraints
as fuzzy numbers and the constraints as fuzzy functions. Also they suggested a
method that accepts objective function as a constraint by giving a satisfaction
level to it. In 1986 Carlsson and Korhonen suggested an approach that accepts
all coefficients as fuzzy and presents a parametric solution. In 1987 Werners
(Werners, 1987) studied on an interactive model and in 1990 Inuiguchi et al.
(Inuiguchi et al., 1990) studied a FLP with partial linear membership functions.
In recent years in 2000 Tanaka et al., in 2001 Jamison and Lodwick (Jamison and
Lodwick, 2001), in 2001 Chiang, Liu (Liu, 2001), in 2002 Bector and Chandra
contributed the theory and methodolgy (Paksoy, 2002).

3.1 Fuzzy linear programming models

We can define a LP problem with crisp or fuzzy resource constraints, and a crisp
or fuzzy objective as:

max̃Z = cT x,
s.t.(Ax)i ≤ b̃i , i = 1, 2, . . ., m, (1a)
x ≥ 0,

where fuzzy resources b̃i , ∀i have the same forms of membership functions. We
may also consider the following fuzzy inequality constraints:

max̃Z = cT x,
s.t.(Ax)i ≤ bi , i = 1, 2, . . ., m, (1b)

x ≥ 0.

Even though (1a) and (1b) are different in some points of view, we can use the
same approach to handle them under the pre-assumption of the membership
functions of the fuzzy available resources and fuzzy inequality constraints.
The difference between crisp and fuzzy constraints is that in case of crisp
constraints the decison maker can strictly differentiate between feasibility and
infeasibility; in case of fuzzy constraints he wants to consider a certain degree
of feasibility in the interval (Werners, 1987).

• When the resources can be determined precisely, a traditional LP problem


is considered as:

max Z = cT x,
s.t.(Ax)i ≤ bi , ∀i, (2)
x ≥ 0,
Interactive fuzzy linear programming and an application sample at a textile firm 33

where c, A and bi , ∀i are precisely given. The optimal solution of (2) is a


unique optimal solution.
• Approach of Chanas and Verdegay: A decision maker wishes to make a
postoptimization analysis. Thus, a parametric programming problem is for-
mulated as:

max Z = cT x,
s.t.(Ax)i ≤ bi + θ pi , ∀i x ≥ 0, (3)
θ ∈ [0, 1],

where c, A, bi and pi , ∀i are precisely given and θ is a parameter. pi , ∀i are


maximum tolerances which are always positive. The solutions Z∗ (θ ) of Eq. 3
are functions of θ . That is, for each θ , we can obtain an optimal solution.
On the other hand, the available resources may be fuzzy. Then the LP
problem with fuzzy resources becomes:

max Z = cT x,
s.t.(Ax)i ≤ b̃i , ∀i, (4)
x ≥ 0.

It is possible to determine the maximum tolerances pi of the fuzzy resources


bi , ∀i . Then we can construct the membership functions µi assumed linear
for each fuzzy constraints, as follows:

⎨1 if (Ax)i ≤ bi ,
(Ax)i −bi
µi = 1 − if bi ≤ (Ax)i ≤ bi + pi , (5)
⎩ pi
0 if (Ax)i > bi + pi .

Verdegay and Chanas, propose that Eqs. 4 and 5, however, are equivalent
to Eq. 3, a parametric LP where c, A, bi and pi , ∀i are given, by use of the
λ-level cut concept.
For each λ-level cut of the fuzzy constraint set, Eq.4 becomes a traditional
LP problem. That is,

max Z = cT x,
s.t.x ∈ Xλ ,
Xλ = { x| µi ≥ λ, ∀i, and x ≥ 0, λ ∈ [0, 1]}. (6)

Equation 6 is then equivalent to:

max̃Z = cT x,
s.t.(Ax)i ≤ bi + (1 − λ)pi , ∀i,
λ ∈ [0, 1] and x ≥ 0, (7)
34 İ. Ertuğrul, A. Tuş

where c, A, bi and pi , ∀i are precisely given. Now, if we set λ = 1 − θ , the


equation given by (7) will be the same as Eq.3. Then a solution table is
presented to the decision maker to determine the satisfying solution. Z∗ (θ ),
θ ∈ [0, 1] is the fuzzy solution corresponding to Verdegay’s approach.
• Werners’s approach: A decison maker may want to solve a FLP problem
with a fuzzy objective and fuzzy constraints, while the goal b0 is not given.
That is:

max̃Z = cT x,
s.t.(Ax)i ≤ bi , ∀i,

x ≥ 0, (8)

which is equivalent to:

max̃Z = cT x,
s.t.(Ax)i ≤ bi + θ pi , ∀i,
θ ∈ [0, 1] and x ≥ 0 (9)

where c, A, bi and pi , ∀i are given, but the goal of the fuzzy objective is not
given.
To solve (9) by use of Werners’s approach, let us first define Z0 and Z1 as
follows:

Z0 = inf(max cT x) = Z∗ (θ = 0), (10)


x∈X
Z1 = sup(max cT x) = Z∗ (θ = 1), (11)
x∈X
X = { x| (Ax)i ≤ bi + θ pi , ∀i, θ ∈ [0, 1], and x ≥ 0}.

Then, we can obtain Werners’s membership function µ0 of the fuzzy objec-


tive. That is;


⎨ 1, if cT x > Z1
µ0 = 1 − Z1 −cT x
, if Z0 ≤ cT x ≤ Z1 (12)

⎩ Z1 −Z0
0, if cT x < Z0

The membership functions µi , ∀i , of the fuzzy constraints are defined as (5).


By use of the min-operator proposed by Bellman and Zadeh, we can obtain
the decision space D which is defined by its membership function µD where,

µD = min(µ0 , µ1 , . . ., µm ). (13)
Interactive fuzzy linear programming and an application sample at a textile firm 35

It is reasonable to choose the decision where µD is maximal as the optimal


solution of Eq. 8. Therefore, Eq. 8 is equivalent to:

max λ,
s.t.µ0 ≥ λ,
µi ≥ λ,
λ, µ0 and µi ∈ [0, 1], ∀i,
x ≥ 0, (14)

where c, A, bi and pi , ∀i, are given, and λ = µD = min(µ0 , µ1 , . . ., µm ).


Let λ = 1 − θ . Then the problem given by (14) will be equivalent to:

min θ ,
s.t.cT x ≥ Z1 − θ (Z1 − Z0 ),
(Ax)i ≤ bi + θ pi , ∀i,
θ ∈ [0, 1] and x ≥ 0, (15)

where c, A, bi and pi , ∀i, are given and θ is a fraction of (Z1 − Z0 ) for the first
constraint and a fraction of the maximum tolerance for others. The solution
is a unique optimal solution.
• Zimmermann’s approach: A decision maker may want to solve a FLP prob-
lem with a fuzzy objective and fuzzy constraints, when the goal b0 of the
fuzzy objective and its minimum tolerance are given. That is,

max̃Z = cT x,
s.t.(Ax)i ≤ bi , ∀i,

x ≥ 0, (16)

where c, A, b0 , p0 , bi and pi , ∀i are given. The problem given by (16) is


actually equivalent to:

Find x,
s.t. cT x ≥ b0 ,

(Ax)i ≤ bi , ∀i,

x≥0 (17)

with the membership functions of the fuzzy constraints as previously


described in (5) and the membership function of the fuzzy objective µ0
as follows:
36 İ. Ertuğrul, A. Tuş



⎨ 1, if cT x > b0 ,
b0 −cT x
µ0 = 1 − p0 , if b0 − p0 ≤ cT x ≤ b0 , (18)


0, if cT x < b0 − p0 .

Thus, by use of the maxmin concept, Eq.17 ia actually eqivalent to:

max λ,
s.t.µ0 and µi ≥ λ, ∀i, (19)
 
λ, µ0 and µi ∈ 0, 1 , ∀i,
x ≥ 0,

where c, A, b0 , p0 , bi and pi , ∀i are given. Let λ = 1 − θ . Then Eq.14 will be


equivalent to:

min θ ,
s.t.cT x ≥ b0 − θ p0 ,
(Ax)i ≤ bi + θ pi , ∀i,
 
θ ∈ 0, 1 and x ≥ 0, (20)

where c, A, b0 , p0 , bi and pi , ∀i, are given and θ is a fraction of the maximum


tolerances. The optimal solution of Eq.20 is unique.
When a fuzzy objective is assumed, what Zimmermann and Werner’s
approaches are asumming is essentially a performance function on the
objective,
f (x) = F(cT x) ∈ [0, 1].
Then, in all cases, if Z∗ (θ ), θ ∈ [0, 1], is the fuzzy solution to the problem, the
corresponding point solution for each fuzzy objective(performance func-
tion associated to Zimmermann’s or Werner’s approach) to be considered,
can be obtained by solving the point-fix equation. It is showed that in the
application.
• A decision maker may want to solve a FLP problem with a fuzzy objective
and fuzzy constraints, while only the goal b0 of the fuzzy objective is given,
but its tolerance p0 is not given. That is,

max̃cT x,
s.t.(Ax)i ≤ bi , ∀i,

x ≥ 0, (21)

where c, A, b0 , bi and pi , ∀i, are given, but p0 is not given. While p0 is


not given, we do know that p0 should be in between 0 and b0 − Z0 . For
each p0 ∈ [0, b0 − Z0 ], we can obtain the membership function of the fuzzy
objective as Eq. 18. Since in a high-productivity system the objective value
Interactive fuzzy linear programming and an application sample at a textile firm 37

should be larger then Z0 at θ = 0, there is no meaning to giving a positive


grade of membership for those which are less than Z0 .
The difference between this problem and Zimmermann is that p0 is not
initially given in this problem. Therefore, we may assume a set of p0 s, where
p0 ∈ [0, b0 −Z0 ]. Then, the problem of each p0 given is a Zimmermann problem.
The decision maker may choose a refined p0 among the solutions for this given
set of p0 s. Then a Zimmermann problem with the decision maker’s refined p0
is solved. This solution will be the final optimal solution for Eq. 21.

3.2 Comparison of Zimmermann, Werners, Chanas


and Verdegay approaches

After a survey various models of FLP techniques, we find that Zimmermann’s,


Chanas’s and Verdegay’s approaches and concepts are the most practical ones
among various techniques in the literature. However, each of them solves one
kind of LP problem-solving and decision making task. In order to understand
the differences and contributions of these approaches and concepts, it is neces-
sary to discuss some major points among them.
In Zimmerman approach, decision maker gets rid of defining problem with
certain limitations in contrast to LP problem. In addition it is solved using only
one more variable and constraint. It makes it effective from the point of pro-
cess time (Paksoy, 2002). In this model, goal b0 and its maximum tolerance p0
should be given initially. In real-world problems, it is unrealistic to initially ask
the decison maker to give b0 and p0 without providing any information about
them. Therefore, the membership function of the fuzzy objective is question-
able and so the solution is also questionable. For example, if b0 given is too
large, there will be no solution in this FLP model. At the same time, if p0 given
is too large, then there will be no meaning for the membership function. Thus
the solution is questionable. Instead of asking the decision maker b0 and p0 to
establish the membership function of the fuzzy objective, Wernes provides two
possible extreme points Z0 and Z1 (Lai & Hwang, 1992). Werners, who accepts
objective function as a fuzzy set, constructed an objective membership function
by using Z0 and Z1 values. The satisfaction level will change until it reaches
optimal solution between Z0 and Z1 values (Paksoy, 2002).
While Tanaka and Zimmermann connect fuzzy sets theory and the max-
min operator proposed by Bellman and Zadeh with LP, Verdegay and Chanas
obtain the equivalent relation between parametric programming with paramet-
ric resources and FLP with fuzzy resources defined by assumed linear mem-
bership functions. Thus, for a symmetric FLP problem, we can first consider its
subproblem of nonsymetric FLP with a crsip objective and fuzzy constraints and
then construct and solve its corresponding parametric programming problem.
The solutions will be peresented in a table which not only provides an optimal
solution with respect to a parameter, but also provides the actually consumed
resources. Therefore, the difficulty of providing b0 and p0 initially is over-
come by presenting the solution table of a parametric programming problem.
38 İ. Ertuğrul, A. Tuş

After refering to this solution table, the decision maker can precisely locate his
subjective b0 and p0 . The solution of Zimmermann’s model is reliable.
While Verdegay has connected the parametric concept with FLP, Chanas has
suggested that the membership function of the fuzzy objective be constructed
directly by the parametric optimal solution. Thus, µ0 is a function of the param-
eter instead of the general function of x. However in any real-world problems,
the number of constraints are always rather large, and so are the decision vari-
ables. Therefore Chanas’s approach for formulating the membership function
of the fuzzy objective is not practical (Lai & Hwang, 1992).

4 Interactive fuzzy linear programming

Decision processes are better described and solved using fuzzy sets theory,
rather than precise approaches. However, the decision maker himself always
plays the most important role in using fuzzy sets theory. Therefore, an intereac-
tive process between the decision maker and the decision process is necessary
to solve our problems. That is actually a user-dependent fuzzy LP tecnique.
Furthermore, a problem-oriented concept is also a vitally important concept in
solving practical problems, as noted by Simon.
By use of fuzzy sets theory, and user dependent (interactive) and problem-
oriented concepts, the flexibility and robustness of LP techniques are improved.
An IFLP approach which is a symmetric integration of Zimmermann’s, Wern-
ers’s, Verdegay’s and Chanas’s FLP approaches is developed and additionally it
provides a decision support system for solving a specific domain of a real-world
LP system (Lai & Hwang 1992). Lai ve Hwang suggested “expert decision
support system” that give an aggregate solution to all possible cases.
The system determines fuzzy-efficient extreme solutions and a fuzzy efficient
compromise solution. They are judged by the decision maker and he decides
whether one of them is subjectively the best one or whether modifications are
necessary. In the latter case the decison maker changes membership functions
assisted by the system (Werners, 1987).
The application of FLP implies that the problems will be solved in an inter-
active way. In the first step, the fuzzy system is modelled by using only the infor-
mation which the decision maker can provide without any expensive additional
information acquisiton. Knowing a first “compromise solution” the decision
maker can perceive which further information should be obtained and he is
able to justify the decision by comparing carefully additional advantages and
arising costs. In doing so, step by step the compromise solutions are improved.
This procedure obviously offers the possibility to limit the acquisition and pro-
cessing information to the relevant componenets and therefore information
costs will be distinctly reduced (Rommelfanger, 1996).
The most important element that affects solutions of FLP problems is param-
eters which are used reflecting fuzziness of model. How these parameters define
fuzzy geometry is the most sensitive point. Because the success of solution
depends on the success of reflecting the system of model.
Interactive fuzzy linear programming and an application sample at a textile firm 39

Moreover, the interactive concept provides a learning process about the


system and makes allowance for psychological coergence for the decision
maker, whereby, (s)he learns to recognize good solutions, the relative impor-
tance of factors in the system and then design a high-productivity system, instead
of optimizing a given system.
This IFLP system provides integration-oriented, adaption and learning fea-
tures by considering all possibilities of a specific domain of LP problems which
are integrated in logical order using an IF-THEN rule.
IFLP methods have been studied, since 1980. Typical works are Baptistella
and Ollero, Fabian, Ciobanu and Stoica, Ollero, Aracil and Camacho, Seo
and Sakawa, Slowinski, Werners and Zimmermann. Zimmermann described
some general concepts and modelling methods of decision support systems and
expert systems in a fuzzy enviroment. Others developed interactive approaches
to solve multiple criteria decison making problems (Lai & Hwang, 1992).
With the aim of the solutions for the models like these, there are many stud-
ies on LP models. However the studies of Zimmermann, Chanas, Werners and
Verdegay have become quite efficient for improving LP models with decision
support to solve real-world problems.

4.1 Interactive fuzzy linear programming algorithm

Step 1 Solve a traditional LP problem of Eq. 2 by use of the simplex method.


The unique optimal solution with its corresponding consumed resources is pre-
sented to the decision maker.
Step 2 Does this solution satisfy the decision maker? Consider the following
cases.
1 If solution is satisfied then print out results an stop.
2 If resource i, for some i are idle then reduce available bi and go to Step 1.
3 If available resources are not precise and some tolerances are possible then
make a parametric analysis with and go to Step 3.
Step 3 Solve a parametric LP problem of Eq. 3. The results are depicted in a
table. At the same time, let us identify Z0 = Z∗ (θ = 0) and Z1 = Z∗ (θ = 1) .
Step 4 Do any of these solutions shown in table satisfy the decision maker?
Consider the following cases:
1 If solution is satisfied then print out results ad stop.
2 If resource i, for some i are idle then decrease bi (and change pi ) and then
go to Step 1.
3 If tolerance i, for some i are not acceptable then change pi as desired and
go to Step 3.
4 If the objective should be considered as imprecise then to to Step 5.
Step 5 After reffering to first table, the decision maker is then asked for his
subjective goal b0 and its tolerance p0 for solving a symmetric FLP problem.
40 İ. Ertuğrul, A. Tuş

If the decision maker does not like to give his goal fo the fuzzy objective, go to
Step 6. If b0 is given, go to Step 8.

Step 6 Solve problem of Eq. 15. A unique Werners’s solution is the provided.

Step 7 Is the solution of Eq. 15 satisfying? Consider the following cases:


1 If the solution is satisfied then print out results and stop.
2 If the user has realized his/her goal then give the goal b0 and go to Step 8.
3 If resource i, for some i are idle then decrease bi (and change pi ) and then
go to Step 1.
4 If tolerance i, for some i are not acceptable then change pi as desired and
go to Step 3.

Step 8 Is p0 determined by the decision maker? If the decision maker would


like to specify p0 , we should provide a table to help the decison maker. Then
go to Step 9. If p0 is not given, then go to Step 11.

Step 9 Solve problem of Eq. 20. A unique Zimmermann’s solution is obtained.

Step 10 Is the solution of Eq. 20 satisfying?


1 If solution is satisfied then print out results and stop.
2 If the user has realized beter his/her goal (and its tolerance) then give the
goal b0 (and p0 ) and go to Step 8.
3 If resource i, for some i are idle then decrease bi (and change pi ) and then
go to Step 1.
4 If tolerance i, for some i are not acceptable then change pi as desired and
go to Step 3.

Step 11 Solve last problem. That is, call Step 9 to solve problem of Eq.20 for a
set of p0 s. Then the solutions are depicted in a table.

Step 12 Are the solutions satisfying? If yes, print out the solution and then
terminate the solution prosedure. Otherwise, go to Step 13.

Step 13 Ask the decision maker to specify the refined p0 , and then go to Step
9. It is rather reasonable to ask the decison maker p0 at this step, because he
has a good idea about p0 now Fig. 1.

For implementing the above IFLP, we need only two solution-finding techniques,
the simplex method and parametric method. Therefore, the IFLP approach
proposed here can be easily programmed in a PC system for its simplicity
(Lai & Hwang, 1992).

5 Application sample at a textile firm

In this study, it has been applied on an outstanding firm in Denizli, the centre
of textile city in Turkey that produces fabric, garments and hometextile. In this
Interactive fuzzy linear programming and an application sample at a textile firm 41

Model Formulation

First Step
Efficient Extreme
Solutions

Compromise Solution Local Informations

Yes “Best ”
Solution Acceptable? Compromise
STOP
No

Interactive Step
Modification of membership functions

Local consequences?
yes

No

Fig. 1 Flow chart decision support system (Werners, 1987)

firm, market distribution of production is 15% on domestic market and 85%


on export. Three main products of this firm are garment products/ready made
clothes, hometextile groups, design archive system. We will only focus on home-
textile group of this firm in this study. For simplicity, products are defined as
sheet, pillow case and sheet of a quilt and departments are defined as cutting,
sewing, pleating and packaging.

5.1 The aim of application

The aim of this application is to determine monthly production planning and


profit of hometextile group firstly by LP model with certain data and secondly
to determine interactively by Zimmermann, Chanas, Werners and Verdegay
approaches and finally to have a comparison among these models.

5.2 The data of application

The profit for a unit of sheet sale is 1.05 Euro, pillow case sale is 0.3 Euro
and sheet of a quilt sale is 1.8 Euro. This firm thinks to sale “approximately
20.000 units of sheet, 40.000 units of pillow case and 8.000 units of sheet of a
quilt”. Monthly working capacity and required process time for the production
of sheet, pillow case and sheet of a quilt are given in Table 1.
In this view, let’s determine monthly production planning and profit of home-
textile group. X1 presents the quantity of sheet that will be produced, X2
presents the quantity of pillow case and X3 presents the quantity of a sheet
of a quilt.
42 İ. Ertuğrul, A. Tuş

Table 1 Required Process Time for sheet, pillow case and sheet of a quilt

Departments Required unit time(hour) Working hours


for a month
Sheet Pillow case Sheet of a quilt

Cutting 0.0033 0.001 0.0033 208


Sewing 0.056 0.025 0.1 4368
Pleating 0.0067 0.004 0.017 520
Packaging 0.01 0.01 0.01 780

Z max = 1.05X1 + 0.3X2 + 1.8X3 (profit),


0.0033X1 + 0.001X2 + 0.0033X3 ≤ 208(cutting),
0.056X1 + 0.025X2 + 0.1X3 ≤ 4368(sewing),
0.0067X1 + 0.004X2 + 0.017X3 ≤ 520(pleating),
0.01X1 + 0.01X2 + 0.01X3 ≤ 780(packaging),
X1 ≥ 20000(sheet),
X2 ≥ 40000(pillow case),
X3 ≥ 8000(sheet of a quilt),
X1 , X2 , X3 ≥ 0.

Step 1 Optimal solution is: X* = (27767, 40000, 10233) and Z* = 59575


Resources actually used are: (165.4, 3578.252, 520, 780)

Step 2 The decision maker feels that 165.4 h for cutting, 3578.252 h for sew-
ing is enough to reach 59,575 Euro. So 208 − 165.4 = 42.6 h of cutting and
4368 − 3578.252 = 789.7475 h of sewing are idle resources. The decision maker,
therefore, would like to try to find the solution of a new LP with the resorces
changed from (208, 4368, 520, 780) to (165.4, 3578.252, 520, 780).

Z max = 1.05X1 + 0.3X2 + 1.8X3 (profit),


0.0033X1 + 0.001X2 + 0.0033X3 ≤ 165.4(cutting),
0.056X1 + 0.025X2 + 0.1X3 ≤ 3578, 252(sewing),
0.0067X1 + 0.004X2 + 0.017X3 ≤ 520(pleating),
0.01X1 + 0.01X2 + 0.01X3 ≤ 780(packaging),
X1 ≥ 20000(sheet),
X2 ≥ 40000(pillow case),
X3 ≥ 8000(sheet of a quilt),
X1 , X2 , X3 ≥ 0.
Optimal solutiom is X ∗ = (27767, 40000, 10233), and Z∗ = 59575,
Interactive fuzzy linear programming and an application sample at a textile firm 43

Table 2 The solutions of the


θ Z∗ (θ ) Resources actually used
parametric programming
problem Cutting Sewing Pleating Packaging

0.0 59575 165.4 3578.252 520 780


0.1 60056 167.1160 3603.5960 522.6 785.2
0.2 60538 168.8320 3628.4 525.2 790.4
···
0.8 63428 179.1280 3781.0020 540.8 821.6
0.9 63909 180.844 3806.346 543.4 826.8
1.0 64391 182.56 3831.69 546 832

Resources actually used are: (165.4, 3578.252, 520, 780). Now there are no idle
resources.

Step 3 Let us assume that the decision maker provides the maximum tolerances
for pleating and packaging:

Tolerances

Cutting 0
Sewing 0
Pleating 26
Packaging 52

The parametric programming is:

Z max = 1.05X1 + 0.3X2 + 1.8X3 (profit),


0.0033X1 + 0.001X2 + 0.0033X3 ≤ 208(cutting),
0.056X1 + 0.025X2 + 0.1X3 ≤ 4368(sewing),
0.0067X1 + 0.004X2 + 0.017X3 ≤ 520 + 26θ (pleating),
0.01X1 + 0.01X2 + 0.01X3 ≤ 780 + 52θ (packaging),
X1 ≥ 20000(sheet),
X2 ≥ 40000(pillow case),
X3 ≥ 8000(sheet of a quilt),
X1 , X2 , X3 ≥ 0.

Results in Table 2 are provided to decision maker. Decision maker may choose
a satisfying solution for implementation.

Step 4 The decision maker may consider that traditional LP is not enough to
solve his problem because of the imprecise properties of the resources in nature.
After detailed analysis, he feels maximum tolerances for cutting, sewing, pleat-
ing and packaging:
44 İ. Ertuğrul, A. Tuş

Tolerances pi

Cutting 38 170
Sewing 768 3600
Pleating 52 520
Packaging 104 780

Z max = 1.05X1 + 0.3X2 + 1.8X3 (profit),


0.0033X1 + 0.001X2 + 0.0033X3 ≤ 170 + 38θ (cutting),
0.056X1 + 0.025X2 + 0.1X3 ≤ 3600 + 768θ (sewing),
0.0067X1 + 0.004X2 + 0.017X3 ≤ 520 + 52θ (pleating),
0.01X1 + 0.01X2 + 0.01X3 ≤ 780 + 104θ (packaging),
X1 ≥ 20000(sheet),
X2 ≥ 40000(pillow case),
X3 ≥ 8000(sheet of a quilt),
X1 , X2 , X3 ≥ 0.

Step 5 Present Table 3 to decision maker.

Step 6 (Werners’ approach)

Z0 = 59575,
Z1 = 69207.

Membership function of fuzzy objective;




⎪ 1, if 1.05X1 + 0.3X2 + 1.8X3 > 69207,

µ0 = 1 − 69207−1.05X1 −0.3X2 −1.8X3 , if 59575 ≤ 1.05X1 + 0.3X2 + 1.8X3 ≤ 69207,
⎪⎪ 69207−59575
⎩ 0, if 1.05X1 + 0.3X2 + 1.8X3 < 59575

Table 3 The solutions of the


θ Z∗ (θ ) Resources actually used
parametric programming
problem Cutting Sewing Pleating Packaging

0.0 59575 165.4 3578.252 520 780


0.1 60538 168.8320 3628.4 525.2 790.4
0.2 61501 172.2640 3679.6270 530.4 800.8
···
0.9 67281 192.856 3983.752 561.6 863.2
1.0 69207 199.72 4085.127 572 884
Interactive fuzzy linear programming and an application sample at a textile firm 45

Membership function of fuzzy constraints;




⎨ 1, if 0.0033X1 + 0.001X2 + 0.0033X3 < 170,
µ1 = 1 − 0.0033X1 +0.001X2 +0.0033X3 −170 , if 170 ≤ 0.0033X1 + 0.001X2 + 0.0033X3 ≤ 208,

⎩ 38
0, if 0.0033X1 + 0.001X2 + 0.0033X3 > 208,


⎨ 1, if 0.056X1 + 0.025X2 + 0.1X3 < 3600,
µ2 = 1 − 0.056X1 +0.025X2 +0.1X3 −3600 , if 3600 ≤ 0.056X1 + 0.025X2 + 0.1X3 ≤ 4368,

⎩ 768
0, if 0.056X1 + 0.025X2 + 0.1X3 > 4368,


⎨ 1, if 0.0067X1 + 0.004X2 + 0.017X3 < 520,
µ3 = 1 − 0.0067X1 +0.004X2 +0.017X3 −520 , if 520 ≤ 0.0067X1 + 0.004X2 + 0.017X3 ≤ 572,

⎩ 52
0, if 0.0067X1 + 0.004X2 + 0.017X3 > 572,


⎨ 1, if 0.01X1 + 0.01X2 + 0.01X3 > 780,
0.01X1 +0.01X2 +0.01X3 −780
µ4 = 1− , if 780 ≤ 0.01X1 + 0.01X2 + 0.01X3 ≤ 884,

⎩ 104
0, if 0.01X1 + 0.01X2 + 0.01X3 > 884,

min θ
1.05X1 + 0.3X2 + 1.8X3 ≥ 69207 − 9632θ (profit),
0.0033X1 + 0.001X2 + 0.0033X3 ≤ 170 + 38θ (cutting),
0.056X1 + 0.025X2 + 0.1X3 ≤ 3600 + 768θ (sewing),
0.0067X1 + 0.004X2 + 0.017X3 ≤ 520 + 52θ (pleating),
0.01X1 + 0.01X2 + 0.01X3 ≤ 780 + 104θ (packaging),
X1 ≥ 20000(sheet),
X2 ≥ 40000(pillow case),
X3 ≥ 8000(sheet of a quilt),
X1 , X2 , X3 ≥ 0,
X ∗ = (33825.2, 40000, 9374.761)Z∗ = 64391 at θ = 0.5,

Resources actually used are: (182.54, 3831.68, 546.37, 832).


We can find the fuzzy solution to the problem directly by solving the point-fix
equation:

θ = 0.5 and Z∗ (θ ) = 69207−9632θ . So Z∗ (0.5) = 69207−9632(0.5) = 64391.


69207 − cT x 69207 − 64391
F(cT x) = . So F(64391) = = 0.5.
9632 9632

Step 7 Let us assume that after recalling Table 3, determining b0 is considered.

Step 8 Presenting Table 3, ask the decision maker to determine p0 .

Step 9 Let us assume b0 = 64391 at θ = 0.5 and p0 = 4500, 64391 − 59575 =


4816. So p0 must be between 0 and 4816.
46 İ. Ertuğrul, A. Tuş

Membership function of fuzzy objective;




⎨ 1, if 1.05X1 + 0.3X2 + 1.8X3 > 64391,
µ0 = 1 − 64391−1.05X1 −0.3X2 −1.8X3 , if 59891 ≤ 1.05X1 + 0.3X2 + 1.8X3 ≤ 64391,

⎩ 4500
0, if 1.05X1 + 0.3X2 + 1.8X3 < 59891,

min θ
1.05X1 + 0.3X2 + 1.8X3 ≥ 64391 − 4500θ (profit),
0.0033X1 + 0.001X2 + 0.0033X3 ≤ 170 + 38θ (cutting),
0.056X1 + 0.025X2 + 0.1X3 ≤ 3600 + 768θ (sewing),
0.0067X1 + 0.004X2 + 0.017X3 ≤ 520 + 52θ (pleating),
0.01X1 + 0.01X2 + 0.01X3 ≤ 780 + 104θ (packaging),
X1 ≥ 20000(sheet),
X2 ≥ 40000(pillow case),
X3 ≥ 8000(sheet of a quilt),
X1 , X2 , X3 ≥ 0,
X ∗ = (31896.17, 40000, 9648.04) Z∗ = 62857 θ = 0.3408,

Resources actually used are: (180.3946, 3799.7080, 519.999, 780).


We can find the fuzzy solution to the problem directly by solving the point-fix
equation:

θ = 0.3408 and Z∗ (θ ) = 64391 − 4500θ .


So Z∗ (0.3408) = 64391 − 4500(0.3408) = 62857.
64391 − cT x 64391 − 62857
F(cT x) = . So F(62857) = = 0.3408.
4500 4500

Step 10 Here, let us assume that the solution is satisfied.

Step 11 Take a simulation by choosing a set of p0 s. Let us take six possible


values of p0 ∈ [0, 4816]; (0, 1000, 2000, 3000, 4000, 4816).

Step 12 After referring to Table 4, the decision maker may choose a satisfying
solution and then terminate the solution procedure.

Step 13 The comparison of the solutions of Werners and Zimmermann


θ 0.5 0.3408
Z∗ 64391 62857
X∗ (33825.2, 40000, 9374.761) (31805.86, 40000, 9660.835)
Resources used (182.54, 3831.68, 546.37, 832) (176.8, 3747, 537, 814.67)
Interactive fuzzy linear programming and an application sample at a textile firm 47

Table 4 The optimal solution of a symmetric FLP for a given set of p0 s

p0 Resources actually used

θ Z∗∗ x∗∗ Cutting Sewing Pleating Packaging

0 0.5 64390.999 (33825.16, 40000, 182,560 3831,686 546 832


9374.767)
1000 0.453 63938.031 (33255.39, 40000, 180,946 3807,85 543,554 827,109
9455.484)
2000 0.414 63562.942 (32783.58, 40000, 179,610 3788,112 541,529 823,059
9522.324)
3000 0.3813 63247.242 (32386.47, 40000, 178,484 3771,5 539,825 819,651
9578.582)
4000 0.3533 62977.848 (32047.61, 40000, 177,525 3757,325 538,371 816,742
9626.587)
4816 0.3333 62785.656 (31805.86, 40000, 176,840 3747,211 537,333 814,667
9660.835)

µ0
1
Zimmerman
0.66

0.5
Werners
0.34

0.27

0
59575 59891 62141 62857 64391 69207 cT x

Fig. 2 The comparison of the solutions of Zimmermannn and Werners membership functions µ0

The difference between Wernes’s and Zimmermann’s membership function is


depicted in Fig. 2. In this figure, Zimmermann’s b0 and p0 are assumed to be
rational. If so, the decision maker may consider Zimmemann’s membership
function as a more acceptable one most of the time. The difference between
these two models is caused from the shape of membership functions.

6 Conclusions

As it has been seen, the algorithm will be the optimal result that can be obtained
because of being constructed on the best sides of Chanas, Werners, Verdegay
and Zimmermann approaches.
The FLP is a method that is used when fuzziness seen in many decision pro-
cesses is incorparated in the problems which can be solved by using the certain
LP method. If it is thought that many decision processes are uncertain, it is
understood that FLP is an active and a useful method.
48 İ. Ertuğrul, A. Tuş

As it has been seen in application, FLP and solution processes follow an


approach that allows to be directed by decision maker. So, many situations that
the decision maker can face must be taken into consideration to construct model
and solve this model. Decision maker may be pleased or not. However, with
this algorithm, decision maker can change some situations whenever he wants
to modify original model and he can stop solution process when he reaches the
satisfaction result. If the decision maker is not pleased to results, interactive
solution process can be continued till reaching satisfaction result. Here it is
noted that approach aim of algorithm is not only to solve given LP problem but
also to design high-productivity system.
Decision maker must review different hypothesis of suggested procedures
and compare them with real-world decision problems to make the best choice.
In any situations, the solution must be defined step by step in an interactive
process. So, it can be avoided the inadequate modelling of real problem and
knowledge costs are decreased in general (Rommelfanger, 1996).
The IFLP has an active usage advantage on an area of system improvement
and system analysis. By IFLP, the possible risk factors can be decreased or at
least basic knowledge opportunity can be given to the decision maker about
what the possible risk factors will be.
Although IFLP, decision making process not only finds the optimal solu-
tion of a problem but also provides a high-productivity system. Therefore IFLP
system provides integration-oriented, adaption and learning features by consid-
ering all the possibilities of a specific main of LP problems which are integrated
in logical order using an IF-THEN rule.
We can notice that fuzzy sets theory provides more information to the deci-
sion maker to solve practical decision making problems in the application
studied.
As a result it is noted that FLP models are improved by two different con-
cepts as preference and probability. Here IFLP is based on preference concept.
(Lai & Hwang, 1992).
Even if it is possible to decrease uncertainties of decision problems, human
who has flexible thinking capability must use fuzzy logic and its applications
to the OR techniques in decision making processes actively. According to the
uncertainty and structure in decision problems, using different models, finding
their solutions and defining most suitable one among these solutions can give
an important support to active decisions (Güneş & Umarusman, 2003).

References

Arıkan, F. (1996). Bulanık hedef programlamanın çok amaçlı proje şebekesine uygulanması. Ankara,
Türkiye: Yayınlanmamış Y.L.Tezi, Gazi Üniversitesi.
Delgado, M; Verdegay, J.L, Vila (1989). A general model for fuzzy linear programming. Fuzzy Sets
and Systems, 29, 21–29.
Ertuğrul, I. (1996). Bulanık mantık ve bir üretim planlamasında uygulama örneği. Denizli, Türkiye:
Yayınlanmamış Y.L. Tezi, Pamukkale Üniversitesi.
Güneş, M., & Umarusman, N. (2003). Bir karar destek aracı bulanık hedef programlama ve
yerel yönetimlerde vergi optimizasyonu uygulaması. Review of Social, Economic and Business
Studies, 2, 242–255.
Interactive fuzzy linear programming and an application sample at a textile firm 49

Guu S. M., & Wu, Y. K. (1999). Two phase approach for solving the fuzzy linear programming
problems. Fuzzy Sets and Systems, 107(2), 191–195.
Inuiguchi, M., Ichihashi, H., & Kume, Y. (1990). A solution algorithm for fuzzy linear programming
with piecewise linear membership functions. Fuzzy Sets and Systems, 34, 15–31.
Jamison K. D. & Lodwick W. A. (2001). Fuzzy linear programming using a penalty method. Fuzzy
Sets and Systems, 119(1), 97–110.
Kosko, B. (1997). Fuzzy engineering. NJ: University of Southern California, Prentice Hall.
Lai, Y. J., & Hwang, C. L. (1992). Intereactive fuzzy linear programming. Fuzzy Sets and Systems,
45, 169–183.
Liu, X. (2001). Measuring the satisfaction of constraints in fuzzy linear programming. Fuzzy Sets
and Systems, 122, 263–275.
Maleki, H. R., Tata, M., & Mashinchi, M. (2001). Linear programming with fuzzy variables. Fuzzy
Sets and Systems, 109, 21–33.
Paksoy, T. (2002). Bulanık küme teorisi ve doğrusal programlamada kullanımı: Karşılaştırmalı bir
analiz. Selçuk Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi, 17(1), 1–16.
Paksoy, T., & Atak, M. (2002). Etkileşimli bulanık çok amaçlı doğrusal programlam ile bütünleşik
üretim planlama: Hidrolik pompa imalatçısı firma örnek olayı. Gazi Üniversitesi Fen Bilimleri
Enstitüsü Dergisi, 15(2), 457–466.
Rommelfanger, H. (1996). Fuzzy linear programming and applications. Europan Journal of Oper-
ational Research, 92, 512–527.
Rommelfanger, H., Hanuscheck, R., & Wolf, J. (1989). Linear programming with fuzzy objectives.
Fuzzy Sets and Systems, 29, 31–48
Sankar, K. P. (1986). Fuzzy mathematical approach to pattern recognition. New York, USA
Vasant, P. M. (2003). Application of fuzzy linear programming in production planning. Fuzzy
Optimization and Decision Making, 3, 229–241.
Wang, R. C., & Liang, T. F. (2004). Application of fuzzy multi-objective linear programming o
aggregate production planning. Computers and Industrial Engineering, 46, 17–41.
Werners, B. (1987). An interactive fuzzy programming system. Fuzzy Sets and Systems, 23, 131–147.
Yager, R. R., & Filev, D. P. (1994). Essentials of fuzzy modeling and control. New York: Wiley.
Zhao, R., Govind, R., & Fan, G. (1992). The complete decision set of the generalized symmetrical
fuzzy linear programming. Fuzzy Sets and Systems, 51, 53–65.

You might also like