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

This article was downloaded by: [Universitat Politècnica de València]

On: 27 October 2014, At: 06:55


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number:
1072954 Registered office: Mortimer House, 37-41 Mortimer Street,
London W1T 3JH, UK

International Journal of
Production Research
Publication details, including instructions for
authors and subscription information:
http://www.tandfonline.com/loi/tprs20

Machine selection in
flexible manufacturing cell:
A fuzzy multiple attribute
decision-making approach
Tai-Yue Wang , Chia-Fong Shaw & Yueh-Li Chen
Published online: 14 Nov 2010.

To cite this article: Tai-Yue Wang , Chia-Fong Shaw & Yueh-Li Chen (2000)
Machine selection in flexible manufacturing cell: A fuzzy multiple attribute
decision-making approach, International Journal of Production Research, 38:9,
2079-2097

To link to this article: http://dx.doi.org/10.1080/002075400188519

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all
the information (the “Content”) contained in the publications on our
platform. However, Taylor & Francis, our agents, and our licensors
make no representations or warranties whatsoever as to the accuracy,
completeness, or suitability for any purpose of the Content. Any
opinions and views expressed in this publication are the opinions and
views of the authors, and are not the views of or endorsed by Taylor
& Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information.
Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities
whatsoever or howsoever caused arising directly or indirectly in
connection with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study
purposes. Any substantial or systematic reproduction, redistribution,
reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access
and use can be found at http://www.tandfonline.com/page/terms-and-
conditions
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014
INT. J. PROD. RES., 2000, VOL. 38, NO. 9, 2079 ± 2097

Machine selection in ¯ exible manufacturing cell: a fuzzy multiple


attribute decision-making approach

TAI-YUE WANG*, CHIA-FONG SHAW{ and YUEH-LI CHEN

Flexible manufacturing cells (FMC) have been used as a tool to implement


¯ exible manufacturing processes to increase the competitiveness of manufacturing
systems. In implementing an FMC, decision-makers encounter the machine selec-
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

tion problem including attributes, e.g. machine type, cost, number of machines,
¯ oor space and planned expenditures. This paper proposes a fuzzy multiple attri-
bute decision-making (FMADM) model to assist the decision-maker to deal with
the machine selection problem for an FMC realistically and economically. In
addition, the membership functions of weights for those attributes are determined
in accordance with their distinguishability and robustness when the ranking is
performed.

1. Introduction
For a manufacturer, the following factors are considered in the machine selection
process: purchasing cost, ¯ oor space, working environment, productivity, operation
conditions, task and operating preference. However, the methods to determine the
weights of those factors are controversial. One may subjectively assign weights
according to experience or expertize. Agrawal et al. (1991, 1992) determined the
weights of attributes for selecting robots and jigs by using the Satty eigenvector
method. Other researchers treated weights as linguistic representations and used
fuzzy weights in multiple objective decision-making and multiple criteria decision-
making problems (Yager 1977, Stewart 1992).
Most research on the requirements and purchasing problems of automated facil-
ities has focused on the cost measurement. Knott and Getto (1982) suggested a
model to evaluate di€ erent robotic systems under uncertainty. In their model, they
evaluated di€ erent alternatives by computing the total net present values of cash
¯ ows of investments, labour components and overheads. Miller and David (1977)
reviewed a large amount of literature on the machine requirements problem and thus
deduced some important factors, e.g. production output requirements, machine
output rates, machine scrape factors and work time available. In their research
they conducted only simple arithmetic operations composed of the above factors,
although they did recommend a mathematical programming direction. Kusiak
(1987) built an M1 model subjected to the machining type, machine dimensions
and horsepower. Later, other facility constraints , e.g. the available AGVs number
and time were added to the original model to become an M2 model. Subsequently ,
Behnezhad and Khoshnevis (1988) enhanced the machine requirement model by

Revision received October 1999.


{ Department of Industrial Management Science, National Cheng Kung University,
Tainan, Taiwan, Republic of China.
* To whom correspondence should be addressed. e-mail: tywang@ mail.ncku.edu.tw

International Journal of Production Research ISSN 0020± 7543 print/ISSN 1366± 588X online # 2000 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
2080 T .-Y . W ang et al.

adding market demand and inventory amount. However, in the above models for the
machine requirement problem, only limited factors are considered and the model has
been oversimpli® ed. In addition, important factors, e.g. the number of machines,
product quality, available space and interrelation among operating processes were
not considered. In addition, the most important factors for machine attributes, e.g.
spindle speed, type of control and accuracy were also neglected.
In the study of the decision-making problem, the multiple objective, multiple
attribute and multiple criteria decision models are useful. The multiple objective
decision-making (MODM) consists of a set of con¯ icting goals that cannot be satis-
® ed simultaneously. It also usually involves solving a problem on continuous space
via a mathematical programming model, while multiple attribute decision-making
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

(MADM) deals with the problem of choosing an alternative from a set of candidate
alternatives which are characterized in terms of some attributes. Our machine
selection problem is indeed a multiple attributes decision problem. The multiple
criteria decision-making (MCDM) model is generally used when the problem
includes multiple objective or multiple attribute or both. Besides, obtaining an
exact number to represent the importance of attribute is usually di cult and expen-
sive, and the linguistic representation is preferable in practice. Thus, a fuzzy MADM
is a suitable decision model for our machine selection problem. Bass and
Kwakernaak (1977) and Yager et al. (1978) established some fuzzy models to deal
with imprecise decision-making problems. They used fuzzy sets or fuzzy number to
represent the alternative objective function, utility function or other performance
measures (Bellman and Zedah 1970, Tong and Bouissone 1986 , Dong and Wong
1987). Then the fuzzy ranking process is used to generate preference relations, and
those preference relations can be used as the suggested solutions for decision-makers
(Chang 1981, Nakamura 1986). Also, Zimmermann (1991) introduced a two-stage
model for the multiple attribute decision-making problems.
There are two popular ranking methods used among researchers. The ® rst one is
to use crisp relations; i.e. de® ning a ranking function F which maps fuzzy numbers to
the real numbers and then ranks the fuzzy numbers according to the real numbers.
Some approaches, e.g. Adamo’s ¬-cut approach (Adamo 1980), Yager’s centroid
index, area measurement (Yager 1980, 1981), and the approach of Murakami et al.
(1983) belong to this method. This method can rank di€ erent alternatives easily but
may lose most of the information. The other method is to rank those fuzzy numbers
by using fuzzy relations and subsequently explain these grades of membership via
linguistic relations. For this method, Yuan (1991) introduced a Q fuzzy preference
relation, de® ning four criteria: fuzzy preference representation; rationality of fuzzy
ordering; distinguishability ; and robustness to evaluate fuzzy ranking methods.
In this research, we present a FMADM approach to evaluate and rank alternatives.
This approach allows the decision-makers to assign the di€ erent importance of those
attributes in the rating phase. And moreover, the decision-makers need not provide a
crisp number or the related importance between the attributes. Thus, the imprecise
information can be used. To obtain an exact number to represent the importance is
usually di cult and expensive in practice. In some cases, it is impossible to obtain the
above information for the manager due to the incomplete or unobtainabl e data. Thus,
our approach with linguistic importance can `economically’ and `realistically’ help
the decision-makers performing machine selection of an FMC. Also, we conduct a
simulated experiment to decide the membership function distribution of weights
which presents the best performance within the ranking process.
Flexible manufacturing cells 2081

2. Machine selection model


The machine selection problem in FMC can be characterized as the FMADM
model. The selection procedure is shown in ® gure 1 and can be summarized as
below.
Step 1. Screen machine database to choose suitable machine types by inputting the
threshold values of key attributes for each machine.
Step 2. Form alternatives composed of those machine types from step 1 and denote
xij as the jth attribute value from alternative i (where xij has been normal-
ized, i.e. xij 2 ‰0;1Š).
Step 3. Obtain the fuzzy weight wj for each j, and represent the importance of
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

attribute j from the decision-maker.

Decide the Evaluation


Criteria
(section 2.1)
Screen Machine
Attributes
Analyze the
Membership Function
for Weights
(section 2.2 and 2.3.3)

Conduct
Alternatives
Simulate and Choose
the Membership
Function for Weights
(section 2.4)

Establish Decision
Matrix

Rating Model
(section 2.3.1)

Ranking Model
(section 2.3.2)

Determine
Dominant
Alternative

Figure 1. The FMADM model for machine selection problem.


2082 T .-Y . W ang et al.

Step 4. Obtain the weighted average rating of alternative i by


X
wj xij
j
Ri ˆ X ; 8i; …1†
wj
j

where Ri represents a fuzzy set or fuzzy number.


Step 5. Rank those alternatives’ fuzzy numbers and suggest the higher ranked alter-
natives to the manager responsible for purchasing according to the ordered
sequence. Allow the manager to make a proper decision in accordance with
practical criteria.
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

2.1. Evaluation criteria


In most studies, two categories of costs have been considered in the machine
selection problem: the direct cost (machine price) and the indirect cost (the machine
¯ oor space). These two kinds of costs are a€ ected by the number of machines. In
addition, the factor of machine type should be taken into account. The machine
types include single-function machines and multi-function machines, depending on
the manager’ s preference and the requirements of the manufacturing environment.
Another important factor concerning machine selection is productivity.
However, the derivation of productivity is di cult because it includes the measure-
ment of manufacturing process, product material, operating area, characteristic of
surface and operating speed. In this research we use the machine speed to estimate
machine productivity due to the fact that the part family or machine family is
known. Therefore, to establish the alternatives for an FMC, we refer the slowest
speed among machine tools as the lower limit of productivity if the speed of robots
or other material-handlin g facilities is not considered. Thus, the four criteria for the
alternative rating process are:
(1) total purchasing cost is minimum;
(2) total machine ¯ oor space is minimum;
(3) total machine number is minimum;
(4) productivity of the constructed FMC is maximum.

2.2. Representation of weight


Weights play the key role in the MADM method. Regardless of whether they are
assigned subjectively or assigned indirectly by pair-wise comparison, it is di cult to
have reasonable and suitable values due to incomplete or unobtainable information.
Thus, a fuzzy set or fuzzy number is a feasible representation when the importance of
an attribute is linguistic or imprecise. Thus, one can use the membership function as
the base to quantify the importance of attributes, e.g. cost, ¯ oor space, machine
number and productivity. Those membership functions also can be used in further
analysis.
In order to use a reasonable, objective and practical membership function, we
extend the membership function ·w~j …wj † from Bass and Kwakernaak (1977) into the
basic triangular function distribution shown in ® gure 2.
Subsequently , one can compare the extended membership function with other
ordinary membership functions, e.g. sharp Gamma distribution, normal distri-
Flexible manufacturing cells 2083

m ~ (
wj w j)

important
1. 0

more-or-less very
important important

wj
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

0. 0

0. 2 0.4 0.6 0. 8 1.0

Figure 2. Triangular membership function.

bution, etc. Finally, we test these functions via the four criteria proposed by Yuan
(1991) .
(1) Fuzzy preference presentation, indicating that a fuzzy ranking method
should be able to represent preference relations in linguistic or fuzzy terms.
(2) Rationality of preference ordering, this indicates that a ranking method
should be able to represent rational human behaviour in terms of consistency
and coherence in preference ordering.
(3) Distinguishability, which is the ability to distinguish or discriminate fuzzy
numbers in terms of the degree of preferences.
(4) Robustness, this requires that the degree of the preference relationship
between two fuzzy utilities should not change dramatically if the change of
the utility membership function is su ciently small.
Generally, an inadequate membership function will mislead the decision result. A
ranking process with low distinguishability can not discriminate those similar alter-
natives. On the other hand, huge di€ erences will result if the robustness is not
su cient, even though the alternatives are similar.
To date, most researchers choose the membership function from the following
functions: rectangular distribution; triangular distribution; trapezoidal distribution;
normal distribution; sharp Gamma distribution; symmetric convex-concave distri-
bution; Cauchy distribution; and symmetric sine distribution. One can also choose
the membership function subjectively. To obtain the membership function of weight
wj , we use the triangular distribution function shown in ® gure 2 as the basic scheme.
Then one can ® nd other membership functions which pass through the same three
points. That is, one can ® x one point such that ·w~j …wj † ˆ 1:0 and the other two
points cj and dj are decided by the ¬k value, where ¬k ˆ ·w~j …wj †. Figure 3 is an
illustration of a Cauchy distribution which passes through the three points (cj , ¬k ),
(dj , ¬k ) and (a, 1.0). Then other functions can be obtained by the same procedures.
Those functions are shown as below.
(1) Normal distribution
2
·…x† ˆ e¡k…x¡a† ; k>0 …2†
2084 T .-Y . W ang et al.

m ~
wj
(w j)

1.0

a k

0.0 wj
cj a dj
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

Figure 3. A Cauchy distribution passing the three points: …cj , ¬k †, …dj , ¬k † and …a; 1:0†.

(2) Sharp Gamma distribution


(
ek…x¡a† ; xµa
·…x† ˆ …3†
e¡k…x¡a† ; x>a

(3) Cauchy distribution


1
·…x† ˆ ; k>0 …4†
1 ‡ k…x ¡ a†2

(4) Symmetric convex-concave distribution


8
>
> 1
>
> 0; x µ a ¡p
k

>
> a
>
>
>
> 1 k 1
>
> …
< a x ¡ a ‡p
> k

 ;
a
† a ¡p
k

a
< x µ a
·…x† ˆ …5†
>
> 1
k
1
>
>
>
>
>
>
a…¡ x ‡ a ‡ p 
k a
 †
; a < x µ a ‡p 
k a

>
>
>
> 1
>
: 0; x > a ‡p 
k a

where k < 1.
(5) Symmetric sine distribution
8
> 0; x µa¡k
>
>
>
> 1 1 º k
>
< ‡ sin
>
2 2 k
x¡a‡… †
; a¡k < xµ a
2
·…x† ˆ …6†
1 1
>
> ¡ sin º
>
>
>
>
> 2 2 k
…x ¡ a ¡ k2 †; a <x µa‡k
:
0; x >a‡k

In addition to the ® ve fuzzy distributions above, some linguistic variables can


help researchers to enhance the representation of weight and make it more ¯ exible.
This will also make the ranking method become more objective. Those linguistic
Flexible manufacturing cells 2085

variables, e.g. are:


·not A~ …x† ˆ 1 ¡ ·A~ …x† …7†
4
·extremely A~ …x† ˆ ·A~…x† …8†
2
·very A~ …x† ˆ ·A~…x† …9†
0:5
·more-or-less A~ …x† ˆ ·A~…x† …10†

2.3. Mathematical model for machine selection


Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

Zimmermann (1991) proposed that the fuzzy multiple attribute decision-making


model should include a two-phase process. We can apply his two-phase process as
the core model for the machine selection problem. The ® rst phase is to ® nd the fuzzy
utilities or fuzzy ® nal ratings of di€ erent alternatives under given weights and deci-
sion matrices. The second phase is to apply the fuzzy ranking methods to determine
the ranking order of the alternatives. The following three subsections explain the
machine selection model for an FMC.

2.3.1. Rating model


Bass and Kwakernaak (1977) were the ® rst researchers to identify the computa-
tional problem of calculating a fuzzy utility function using ¬-cut representation and
interval analysis. However, in their method only those extreme values are abstracted.
This will lose some useful information and may distort the result. To avoid these
disadvantages , we need to obtain the utility function Ri using weighted average
computation,
n
X
wj xij
jˆ1
Ri ˆ n ; nˆ4 …11†
X
wj
jˆ1

where xij denotes the jth attribute value of alternative i, and xij is a crisp and
normalized value; wj denotes the fuzzy weight of the jth attribute and wj is a
fuzzy number; n represents the number of criteria.
The improved fuzzy weighted average method suggested by Liou et al. (1992) is
used to determine the utility function Ri . In our case, however, the input xij is a crisp
value, for which adjustment is made. After adjustment , this model will hold all the
useful information compared to the method of Bass and Kwakernaak (1977). This
method will also be especially useful when one needs to consider the e€ ects of all
factors. The related de® nitions and theorems are shown below.
(1) De® nition
Three functions f …w1 ;w2 ;. . . ;wn †, f …w1 ;w2 ; . . . ;wn jck † and f …w1 ;w2 ; . . . ;wn jdk †
are de® ned.
De® nition 1
w1 x1 ‡ w2 x2 ‡ ¢ ¢ ¢ ‡ wn xn
f …w1 ;w2 ;. . . ;wn † ˆ …12†
w1 ‡ w2 ‡ ¢ ¢ ¢ ‡ wn
2086 T .-Y . W ang et al.

De® nition 2
f …w1 ;w2 ; . . . ;wn jck † ˆ f …w1 ;w2 ;. . . ;wk¡1 ;ck ;wk‡1 ;. . . ;wn † …13†
f …w1 ; w2 ;. . . ;wnjdk † ˆ f …w1 ;w2 ;. . . ;wk¡1 ;dk ;wk‡1 ;. . . ;wn † …14†
where ck and dk are the interval values of wk , and ck < dk .
(2) L emma
Given real numbers a, b, c, d, and a > 0, c > 0:
. if b=a > d=c, then b=a > b ‡ d=a ‡ c > d=c;
. if b=a < d=c, then b=a < b ‡ d=a ‡ c < d=c.
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

From the Lemma we can derive the following two theorems.


Theorem 1
Assume xk < f …w1 ;w2 ;. . . ; wn jck †, i.e.
…dk ¡ ck †xk w1 x1 ‡ ¢ ¢ ¢ ‡ wk¡1 xk¡1 ‡ ck xk ‡ wk‡1 xk‡1 ‡ ¢ ¢ ¢ ‡ wn xn
xk ˆ <
…dk ¡ ck † w1 ‡ ¢ ¢ ¢ ‡ wk¡1 ‡ ck ‡ wk‡1 ‡ ¢ ¢ ¢ ‡ wn
then f …w1 ;w2 ;. . . ;wn jdk † < f …w1 ; w2 ;. . . ;wnjck †.
Using the same reason, we can obtain the following.
Theorem 2
Assume xk > f …w1 ;w2 ;. . . ;wnjck †, then
f …w1 ;w2 ; . . . :;wn jdk † > f …w1 ;w2 ;. . . ;wnjck †
(3) Final rating algorithm
The calculation of alternatives’ ® nal rating is based on the ¬-cut representation
and interval analysis. The algorithm is described in the following steps.
Step 1. Divide the membership range [0, 1] into a ® nite number of values, denoted
by ¬1 ;¬2 ; . . . ;¬k . The degree of accuracy depends on the value of k.
Step 2. Find the corresponding intervals ‰cj ;dj Š of weight wj for each membership
value ¬k . When wj is non-convex , the resulting interval is combined by
conjunction.
Step 3. Evaluate L ˆ H ˆ f …c1 ;c2 ;. . . ;cn † and let BL ˆ BH ˆ F ,
I ˆ fijxi < L ;i ˆ 1;2;. . . ;ng ; J ˆ f jjxj > H ; j ˆ 1;2;. . . ;ng:
Step 4. If I ˆ F , then let L ˆ min f …w1 ; w2 ;. . . ;wn † and go to step 5; else
Step 4.1. Let wj ˆ dj for j 2 BL and wj ˆ cj for j 2 = BL .
Step 4.2. Evaluate li ˆ f …w1 ;w2 ;. . . ; wn jdi † for i 2 I.
Step 4.3. Let L ˆ lm ˆ min li , i 2 I, insert m into BL and subtract m from I.
Step 4.4. 6 F , then repeat step 4; else stop and let
If I ˆ
L ˆ min f …w1 ;w2 ;. . . ;wn †:
Step 5. If J ˆ F , then let H ˆ max f …w1 ;w2 ; . . . ;wn † and go to step 6 ; else
Step 5.1. Let wj ˆ dj for j 2 BH and let wj ˆ cj for j 2 = BH .
Step 5.2. Evaluate hj ˆ f …w1 ;w2 ;. . . ;wn jdj † for j 2 J.
Step 5.3. Let H ˆ hm ˆ max hj , j 2 J, insert m into BH and subtract m
away from J.
Step 5.4. If J 6ˆ F , then repeat step 5, else stop and let
Flexible manufacturing cells 2087

H ˆ max f …w1 ;w2 ;. . . ;wn †:


Step 6. If ‰L ;HŠ is the interval of ¬k -cut of Ri , then stop.
Repeat the above steps to evaluate other intervals of Ri with other ¬k -cut.

2.3.2. Ranking model


After the implementation of the rating process, we will encounter the situations
of choosing the best alternative from the alternatives to which fuzzy utilities are
assigned. Assuming that t alternatives are considered, a preference relation
Q…Ri ; Rj † may be used for all the pairs …Ri ; Rj †, i; j ˆ 1;2;. . . ;t in order to compare
these t alternatives. Then one needs to compute the degree of membership for all
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

paired preference relations. Finally, we can order those alternatives which satisfy the
consistency and coherence in preference according to the fuzzy reciprocity and fuzzy
transitivity properties. To retain the fuzzy preference representation , rationality of
fuzzy ordering, distinguishability and robustness, we introduce the Q preference
relation proposed by Yuan (1991). The related de® nition and theorem can be
described as below.
(1) De® nition
First, we introduce some notations and de® nitions S R …­ †, lR …­ †, »R …­ †, fuzzy
binary relation P, fuzzy reciprocity and fuzzy transitivity.
De® nition 3
Let R be a fuzzy subset of U and de® ned by a membership function ·R …r†,
·R :U ! ‰0;1Š is piecewise continuous, where ·R …r†, 8r 2 U indicates the degree to
which r belongs to R.
. The level ­ set of R is the crisp set denoted by S R …­ †, for any ­ > 0, and
S R …­ † ˆ frj·R …r† ¶ ­ g …15†
. lR …­ † and »R …­ † are de® ned as the lower bound and upper bound of the level ­
set of R separately, i.e.
lR …­ † ˆ Inf r2SR…­ † …r† …16†

»R …­ † ˆ Sup r2SR…­ † …r† …17†


De® nition 4
A fuzzy binary relation P is a fuzzy subset on U £ U with the membership
function ·P …Ri ;Rj †, 8Ri , Rj 2 U, where ·P …Ri ;Rj † represents the strength or truth
of the relation between Ri and Rj in an ordered pair …Ri ; Rj †. A value of ·P ˆ 1:0
implies that this relation is the strongest relation, and ·P ˆ 0:0 indicates that this
relation is the weakest one.
De® nition 5
. Fuzzy reciprocity:
·P …Ri ;Rj † ˆ 1 ¡ ·P …Rj ;Ri †; 8Ri ;Rj 2 U …18†
. Fuzzy transitivity:
if ·P …Ri ;Rj † ¶ 1=2 and ·P …Rj ;Rk † ¶ 1=2
then ·P …Ri ;Rk † ¶ 1=2; 8Ri ;Rj 2 U
2088 T .-Y . W ang et al.

(2) Ranking method


First of all, let fuzzy set Ri ¡ Rj be the fuzzy di€ erence between two fuzzy utilities
Ri and Rj . Then, one can denote its membership function as:
·Ri ¡Rj …z† ˆ Sup …·Ri …ri † ^ ·Rj …rj †† …19†
fri ;rj :ri ¡rj ˆzg

ri , rj and z 2 U, and the symbol ^ represents the min operator , ·Z0 …z† is the member-
ship function of real number 0,
1; zˆ0
·Z0 …z† ˆ …20†
0; z 6ˆ 0
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

Subsequently , one can de® ne the fuzzy preference relation Q…Ri ;Rj †. The relation
between Q…Ri ;Rj † and fuzzy binary relation P has been shown in equation (21).
Thus, one can use equation (22) to compute ·Q …Ri ; Rj † to represent the degree of
preference of Ri over Rj .

·Q …Ri ;Rj † ˆ ·P …Ri ¡ Rj ; z0 † …21†


8
> D 1
>
< ; D >0
) ·Q …Ri ;Rj † ˆ
D
…22†
>
: 1; D ˆ0
>
2
where
D ˆD 1 ‡D 2 …23†
D 1 ˆ S1 ‡ S2 …24†
D 2 ˆ S3 ‡ S4 …25†

and

S1 ˆ »Ri ¡Rj …­ †d­ …26†
f­ : »Ri¡Rj …­ †>0g

S2 ˆ lRi ¡Rj …­ †d­ …27†
f­ : lRi¡Rj …­ †>0g

S3 ˆ »Ri ¡Rj …­ †d­ …28†
f­ : »Ri¡Rj …­ †<0g

S4 ˆ lRi ¡Rj …­ †d­ …29†
f­ : lRi¡Rj …­ †<0g

»Ri ¡Rj …­ † ˆ Sup …z† ˆ »Ri …­ † ¡ lRj …­ † …30†


·Ri ¡Rj …z† ¶­

lRi ¡Rj …­ † ˆ Inf …z† ˆ lRi …­ † ¡ »Rj …­ † …31†


·R i ¡Rj …z† ¶­

Intuitively, S 1 indicates the portion where Ri is preferred to Rj in the most


favourable situation. S 2 indicates the portion where Ri is preferred to Rj in the
most unfavourable situation. S 3 indicates the portion where Rj is preferred to Ri
Flexible manufacturing cells 2089

in the most favourable situation. S 4 indicates the portion where Rj is preferred to Ri


in the most unfavourable situation.
(3) T heorem
Theorem 3
The fuzzy preference relation Q is reciprocal, i.e.

·Q …Ri ;Rj † ˆ 1 ¡ ·Q …Rj ;Ri † …32†


·Q …Ri ;Rj † ˆ 1=2; i ˆ j …33†
Theorem 4
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

The fuzzy preference relation Q is transitive , i.e. for any Ri , Rj and Rk ,

if ·Q …Ri ;Rj † ¶ 1=2 and ·Q …Rj ;Rk † ¶ 1=2


then ·Q …Ri ; Rk † ¶ 1=2

By using the preference relation de® ned for each ordered pair, one can easily rank
the t alternatives fR1 ;R2 ;. . . ;Rt g. The procedure is described as follows.
(1) Calculate ·Q …Ri ;Rj †;i; j ˆ 1;2;. . . ; t, which consists of a t £ t matrix.
However, we only need to calculate …t £ …t ¡ 1††=2 membership values, i.e.
i ˆ 1;2; . . . ;t; j ˆ i ‡ 1;i ‡ 2; . . . t according to Theorem 3.
(2) Sort fR1 ;R2 ;. . . ;Rt g into fRk1 ;Rk2 ; . . . ;Rkt g , by using the transitivity
described in Theorem 4 such that for any i < j, ·Q …Ri ;Rj † ¶ 1=2.
(3) Apply the linguistic interpretation , which has been de® ned as a fuzzy set on
the calculated degree of preference to the fuzzy preference relation. Also the
membership of the interpretation set indicates the truth level of the interpret-
ation. An illustration is shown in ® gure 4. In this ® gure, the L 9 denotes that
Ri is de® nitely better than Rj ; L 8 denotes that Ri is much better than Rj ; and
L 7 denotes that Ri is better than Rj ; L 6 denotes that Ri is weakly better than
Rj ; however, the L 5 denotes that Ri and Rj are indi€ erent; whereas the L 1 , L 2 ,
L 3 , L 4 have contrary meaning to L 9 , L 8 , L 7 , L 6 separately. Further, we may
conclude that Ri is slightly better than Rj , with a 75% truth level, and Ri is
indi€ erent from Rj with a 25% truth level if ·Q …Ri ;Rj † ˆ 0:575.

truth level
of explanation
L1 L 2 L3 L4 L5 L6 L7 L8 L9
100%

0. 1 0.2 0. 3 0.4 0.5 0.6 0.7 0. 8 0.9 1.0


Figure 4. Illustration for the ·Q …Ri ; Rj † linguistic representation.
2090 T .-Y . W ang et al.

2.3.3. Analysis of membership function


To select the membership function, we simulate ® ve distributions based on
triangular distribution, then choose one in accordance with the properties of
distinguishabilty and robustness, as de® ned in equations (34) and (35).
j·Q …Ri ;Rj † ¡ 0:5j ¶ " …34†
j·Q …Ri ;Rj † ¡ ·Q …Ri0 ;Rj †j µ ²; 8Ri0 ; Ri ;Rj 2 U;and d…Ri0 ;Ri † < ¯ …35†
where ", ², ¯ > 0, Ri0 is an approximation of Ri , and d…Ri0 ;Ri †
is the maximum
di€ erence between Ri0 and Ri . The di€ erence can be computed by using equation (36).
d…Ri0 ;Ri † ˆ max …jlRi0 …­ † ¡ lRi …­ †j;j»Ri0 …­ † ¡ »Ri …­ †j† …36†
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

­ >0

The procedure for membership function selection can be described in the follow-
ing steps.
Step 1. Divide the range of membership [0, 1] into a ® nite number of values, denoted
by ¬1 ;¬2 ; . . . ;¬k .
Step 2. Find the corresponding intervals ‰cj ;dj Š of weight wj , and use the three points
…cj ;¬k †, …dj ;¬k † and …a;1:0† for each membership value ¬k to construct ® ve
membership distributions: normal distribution; sharp Gamma distribution;
symmetric convex± concave distribution ; Cauchy distribution; and sym-
metric sine distribution.
Step 3. Generate all the attribute values of di€ erent alternatives using random
numbers and transfer the attribute value into xij , xij 2 ‰0;1Š for all i, j.
Then, randomly generate all weights wj , j ˆ 1;2;3;4 and modify them using
equations (7) ± (10).
Step 4. Find the utility function Ri for all alternatives under the same fuzzy weight
and modi® cation for di€ erent distribution functions.
Step 5. Rank those utility functions by using a di€ erent distribution.
Step 6. Compare the results from step 5 with the triangular distribution function in
distinguishability and robustness properties. Then choose the one with the
best performance.
Due to the fact that the distinguishability and robustness are in contrast, the
former needs more sensitivity but the latter does not. As for a ranking method, these
two are equally important to our model and should be treated equally. Therefore, we
assign equal weights (0.5) to both characteristics to evaluate the performance of each
ranking.

2.4. An experiment for membership function


Both the distinguishability and robustness test were simulated on an IBM RISC/
6000 workstation, and tested 1000, 1500 and 2000 times under di€ erent conditions.
The results show that they do have the same trends. So the following analysis and
discussion are concluded from the data with 2000 simulations.
Divide the grade of membership into 50 values such that ¬1 ˆ 0:02,
¬2 ˆ 0:04;. . . ; ¬50 ˆ 1:0. Then perform the distinguishability simulation and observe
the percentage of times that satisfy equation (34) for each ¬k , letting " ˆ 0:05. Also,
by letting ² ˆ 0:1 we observe the percentage of times that satis® es equation (35) after
robustness simulation. Then, an average score is obtained. The above tests are
Flexible manufacturing cells 2091

Normal
Sharp-Gamma
Cauchy
Symmetric convex concave
Symmetric sine
0.91

0.9
integrated performan

0.89

0.88

0.87
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

0.86

0.85

0.84

0.83
0 0.2 0.4 0.6 0.8 1
alpha

Figure 5. The integrated performance of distinguishability and robustness for the ® ve


di€ erent distributions.

implemented for the ® ve di€ erent distributions and the results are shown in ® gure 5,
except for the triangular distribution, whose ® nal average score was 85.46%.
From the simulation results above, we can conclude that the normal distribution or
symmetric sine distribution, which are approximate d by the interval with an ¬-cut
near 1.0, can simultaneously satisfy the requirement of distinguishability and robust-
ness in our situation.

3. Numerical example
This section provides a real case to explain the complete process of the proposed
machine selection model, where a normal distribution (k ˆ 1262:67) approximated
by the ¬49 ˆ 0:98 is used.
Assuming that a factory manager has decided to purchase some machine facilities
after su cient discussion and complete evaluation, one can use group technology
analysis to form the parts and machines families. Thus, an FMC, including two CNC
milling machines, one CNC lathe and a robot for material handling is con® gured.
The purchasing constraints are described as below.
Constraint 1: the total purchasing cost can not exceed 600 000 dollars.
Constraint 2: for CNC milling m/c
Vertical/horizontal : Horizontal
Spindle speed : 4500 rpm
X/Y/Z axis travel : 630/630/500 mm
Feed rate : 5000 mm/min
Tool capacity : 40
Maximum tool diameter : F 130 mm
Constraint 3: for CNC lathe
Maximum swing :F 520 mm
2092 T .-Y . W ang et al.

Maximum turning diameter : F 350 mm


Maximum turning length : 500 mm
Hole through spindle : F 70 mm
Chuck size : 8 00
Spindle speed : 4500 rpm
Feed rate : 4500 mm/min
Constraint 4: for robot
Con® guration : Armlike
Maximum load capacity at wrist : 60 kg
Allowable load moment of wrist : 36 kg-m
Horizontal reach : 150 cm
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

Repeatability : 1.0
Drive method : Electrical
Furthermore, in the allowance for the operating procedure, the two milling
machines can be replaced with a multi-functional machining centre.

3.1. Data processing


To determine the most suitable machines, one ® rst inputs threshold values from
constraint 2 to constraint 4. Then, those suitable machines are composed into poss-
ible alternatives. By dropping the alternatives which do not meet constraint 1, we
obtain 10 alternatives shown in table 1. Then, the data from the 10 alternatives are
normalized and compiled into table 2. Finally, the fuzzy weights for the four criteria
described in section 2.1 are given by decision-makers, as follows.
2
Criterion 1: very important …·w~1 …w1 † ˆ e¡1262:67…w1 ¡1:0† , 0 µ w1 µ 1:0†
2
Criterion 2: important …·w~2 …w2 † ˆ e¡1262:67…w2 ¡0:8† , 0 µ w2 µ 1:0†
2
Criterion 3: medium importance …·w~3 …w3 † ˆ e¡1262:67…w3 ¡0:6† , 0 µ w3 µ 1:0†
2
Criterion 4: very important …·w~4 …w4 † ˆ ‰e¡1262:67…w4 ¡0:8† Š2 , 0 µ w4 µ 1:0†

Attributes

Total
purchasing Total ¯ oor Total machine Productivity*
Alternatives cost (dollars) space (m2 ) number (mm/min)

Alternative 1 581 818 54.49 3 5500


Alternative 2 595 454 49.73 3 4500
Alternative 3 586 060 51.24 3 5000
Alternative 4 522 727 45.71 3 5800
Alternative 5 561 818 52.66 3 5200
Alternative 6 543 030 74.46 4 5600
Alternative 7 522 727 75.42 4 5800
Alternative 8 486 970 62.62 4 5600
Alternative 9 509 394 65.87 4 6400
Alternative 10 513 333 70.67 4 6000

* Productivity (mm/min): the value corresponds to the machine with the slowest feedrate in FMC.
Table 1. Raw attribute values for all alternatives.
Flexible manufacturing cells 2093

i 1 2 3 4

1 0.854 0.839 1.0 0.859


2 0.835 0.919 1.0 0.703
3 0.848 0.892 1.0 0.781
4 0.951 1.0 1.0 0.906
5 0.885 0.868 1.0 0.812
6 0.915 0.614 0.750 0.875
7 0.951 0.606 0.750 0.906
8 1.0 0.730 0.750 0.875
9 0.976 0.694 0.750 1.0
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

10 0.968 0.647 0.750 0.938

Table 2. Normalized attribute values.

3.2. Computation of rate function


After obtaining the normalized value xij and fuzzy weight
wj …i ˆ 1;2;. . . ;10; j ˆ 1;2;3;4†, one can calculate the rate function Ri for each
alternative using the procedure shown in section 2.3.1.
Step 1. Divide the degree of membership [0, 1] equally into 49 values denoted by
¬1 ;¬2 ;. . . ;¬49 .
Step 2. Find the ¬k -cut interval for each weight wj … j ˆ 1; 2; 3; 4†. For instance, for
alternative 1 and ¬25 ˆ 0:5, one can use the data …x 11 ;x12 ;x13 ; x14 † ˆ
…0:854 ;0:839 ;1:0;0:859† from table 2 and w1 , w2 , w3 , w4 from section 3.1.
So, we have intervals: [0.977, 1.0], [0.777, 0.823], [0.577, 0.623], [0.799,
0.801], and let BL ˆ BH ˆ F .
Step 3. Compute L ˆ H ˆ f …w1 ; w2 ;. . . ; wn † ˆ f …c1 ;c2 ;. . . ;cn †
2:75
L ˆ H ˆ f …0:977 ;0:777 ;0:577 ;0:799† ˆ ˆ 0:8786
3:13
I ˆ fijx1i < L ;i ˆ 1; 2; 3; 4g ˆ f1;2;4g
J ˆ f jjx1j > H ; j ˆ 1; 2; 3; 4g ˆ f3g
Step 4. For I ˆ f1;2;4g 6ˆ F , BL ˆ F , so
Step 4.1. w1 ˆ c1 , w2 ˆ c2 , w3 ˆ c3 , w4 ˆ c4 .
Step 4.2. l1 ˆ f …0:977 ;0:777 ;0:577; 0:799jd1 † ˆ f …1:0;0:777 ;0:577 ;0:799†
2:75 ‡ …d1 ¡ c1 †x11 2:75 ‡ …1:0 ¡ 0:977†*0:854 2:770
ˆ ˆ ˆ
3:13 ‡ …d1 ¡ c1 † 3:13 ‡ …1:0 ¡ 0:977† 3:153
ˆ 0:8785 and
2:789
l2 ˆ f …0:977; 0:777 ;0:577; 0:799jd2 † ˆ ˆ 0:8781
3:176
2:752
l4 ˆ f …0:977; 0:777 ;0:577; 0:799jd4 † ˆ ˆ 0:8787
3:132
Step 4.3. L ˆ l2 ˆ min li , i 2 I, insert 2 into BL and subtract from I, i.e.
BL ˆ f2g , I ˆ f1;4g.
Step 4.4. I ˆ f1; 4g 6ˆ F , so let w1 ˆ c1 , w2 ˆ d2 , w3 ˆ c3 , w4 ˆ c4 and
repeat step 4, until I ˆ F .
2094 T .-Y . W ang et al.

The ® nal value of


2:810
L ˆ f …1:0;0:823 ;0:577 ;0:801† ˆ ˆ 0:8779
3:201
Step 5. For J ˆ f3g 6ˆ F , BH ˆ F , so
Step 5.1. w1 ˆ c1 , w2 ˆ c2 , w3 ˆ c3 , w4 ˆ c4 .
Step 5.2. h3 ˆ f …0:977 ;0:777 ;0:577 ;0:799jd3 † ˆ f …0:977 ;0:777 ; 0:623 ;0:799†
2:75 ‡ …d3 ¡ c3 †x13 2:796
ˆ ˆ ˆ 0:8804
3:13 ‡ …d3 ¡ c3 † 3:176
Step 5.3. H ˆ h3 ˆ max hj , j 2 J,
insert 3 into BH and subtract that from J, i.e. BH ˆ f3g , J ˆ F .
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

Step 5.4. If J ˆ F , then obtain H and stop.


2:796
H ˆ f …0:977 ;0:777 ;0:623 ;0:799† ˆ ˆ 0:8804
3:176
Step 6. The interval of R1 over 0.5-cut is [0.8779, 0.8804].
By repeating steps 1± 6, we can obtain other intervals of R1 over other ¬-cuts.
Then the approximated utility function of alternative 1 is determined. One continues
to evaluate utility functions of other alternatives with the same procedure.

3.3. Ranking model


After all the Ri values are determined, we can pair these utility functions
Ri ;i ˆ 1;2;. . . ;10, into the form of …Ri ;Rj †. Then the degree of fuzzy preference
relations for i ˆ 1; 2; . . . ;10, j ˆ i ‡ 1;i ‡ 2;. . . ;10 are computed. The other degrees
of fuzzy preference relations can be found by the fuzzy reciprocity property. Finally,
all alternatives are ordered according to the property of fuzzy transitivity. The
following example is to illustrate, by calculating ·Q …R1 ;R8 †, the degree of fuzzy
preference relation of pair …R1 ;R8 †.
Step 1. Use equations (26) ± (31) described in section 2.3.2 to compute S 1 , S 2 , S 3
and S 4 .

S1 ˆ »R1 ¡R8 …­ †d­ ˆ 1:524
f­ : »R1¡R8 …­ †>0g

S2 ˆ lR1 ¡R8 …­ †d­ ˆ 0:009
f­ : lR1¡R8 …­ †>0g

S3 ˆ »R1 ¡R8 …­ †d­ ˆ 0:001
f­ : »R1¡R8 …­ †>0g

S4 ˆ lR1 ¡R8 …­ †d­ ˆ 0:972
f­ : lR1¡R8 …­ †>0g

Step 2. Calculate D 1, D 2 and D using equations (25) ± (27) introduced in section


2.3.2.
D 1 ˆ S 1 ‡ S 2 ˆ 1:533
D 2 ˆ S 3 ‡ S 4 ˆ 0:973
D ˆD 1 ‡D 2 ˆ 2:506
Flexible manufacturing cells 2095

i R1 R2 R3 R4 R5 R6 R7 R8 R9 R10

R1 0.500 0.508 0.497 0.430 0.436 0.552 0.608 0.612 0.624 0.639
R2 0.492 0.500 0.478 0.375 0.399 0.555 0.618 0.619 0.630 0.645
R3 0.503 0.522 0.500 0.239 0.370 0.605 0.671 0.665 0.673 0.686
R4 0.570 0.625 0.761 0.500 0.734 0.921 0.952 0.928 0.927 0.933
R5 0.564 0.601 0.630 0.266 0.500 0.894 0.860 0.803 0.789 0.787
R6 0.448 0.445 0.395 0.079 0.106 0.500 0.478 0.405 0.396 0.405
R7 0.392 0.382 0.329 0.048 0.140 0.522 0.500 0.358 0.384 0.409
R8 0.388 0.381 0.335 0.072 0.197 0.595 0.642 0.500 0.553 0.576
R9 0.376 0.370 0.327 0.073 0.211 0.604 0.616 0.447 0.500 0.547
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

R10 0.361 0.355 0.314 0.067 0.213 0.595 0.591 0.424 0.453 0.500

Table 3. The degree of fuzzy preference relation: ·Q …Ri ; Rj †.

D
Step 3. Insert D and D into equation (24) ; ·Q …R1 ;R8 † ˆ 1
ˆ 0:6125.
1
D
Then, repeating these steps until all the pairs are measured, one obtains table 3
for all degree values of this example.
From table 3, we ® nd the ® nal ranked order for all of the 10 alternatives as: 4, 5,
3, 1, 2, 8, 9, 10, 7, 6. When we compare the information shown in table 3 with the
representation of linguistic variables in ® gure 3, we ® nd that Alternative 4 sig-
ni® cantly dominates all the other alternatives (see row 5). Also, Alternative 5 is
better than the others except for Alternative 4 (see row 6) , and Alternative 6 is the
worst (see row 7). As for the other preference relations, e.g.: ·Q …R1 ;R2 † ˆ 0:508
and ·Q …R1 ;R3 † ˆ 0:497 means that Alternatives 1± 3 are almost the same;
·Q …R2 ;R8 † ˆ 0:619 means that Alternative 2 is somewhat better than Alternative
8; whereas ·Q …R8 ;R9 † ˆ 0:553 means that Alternative 8 is slightly better than
Alternative 9; and ·Q …R9 ;R10 † ˆ 0:547 means that Alternative 9 is slightly better
than Alternative 10.
Finally, the dominant alternatives 4, 5 and 3 have been obtained.

4. Conclusion
A Fuzzy MADM model for the machine selection problem is proposed in this
paper. This model simultaneously considers di€ erent purchasing factors on the com-
ponents of an FMC. Also, we collect the related attributive information on the
necessary components for FMC including machine centre, lathe, milling machine
and robot. At the same time, rating and ranking models are implemented. In addi-
tion, simulation has been used to choose appropriate membership distributions for
the weights of attributes. From this research, we ® nd some characteristics of our
model.
(1) Flexible and real weight. In most traditional methods, decision-makers may
be confused by the representation of the importance of attributes. To avoid
confusion, lingual and changeable linguistic variables are introduced, which
both increases ¯ exibility and also provides a better representation of reality.
(2) More useful information from the original data are obtained, because the
fuzzy weighted average method for the rating process retains the whole useful
2096 T .-Y . W ang et al.

message. Thus, the fuzzy weighted average method is available when the
consideration for the e€ ects of all factors are needed.
(3) Clear representation of dominance between alternatives. Using the ranking
process, the degree of fuzzy preference relation ·Q can clearly represent the
preference between two alternatives and can detect the order of all alterna-
tives as e€ ectively as the traditional method. However, our model provides
the bene® ts of more completed and detailed output than that from traditional
methods.
(4) Changeable number of criteria. The changeable number of criteria intro-
duced in this model allows the decision-makers to reduce or increase the
criterion to meet their needs. For example, the accuracy of production may
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

be added to the model when it is needed. This bene® t can enlarge the appli-
cation scope for our model.
(5) Distinguishable and robust membership function. The selection of member-
ship functions for the weights in the machine selection problem is simulated.
From the test procedure of distinguishability and robustness of ® ve member-
ship distributions, we conclude that the symmetric sine distribution and the
normal distribution approximated by the triangular type interval over a near
1.0 ¬-cut are more suitable than other distributions.
Therefore, our FMADM approach is suitable for the machine selection problem
in FMC. In sum, this model can provide e cient and satisfying results for decision-
makers, although it may not be the best one in other decision-making problems.

References
A DAMO, J . M ., 1980, Fuzzy decision trees. Fuzzy Sets and Systems, 4, 207± 219.
A GRAWAL, V. P ., K OHLI, V. and G UPTA, S., 1991, Computer aided robot selection: the mul-
tiple attribute decision making approach. International Journal of Production Research,
29, 1629 ± 1644.
A GRAWAL, V. P ., VERMA, A . and A GRAWAL, S., 1992, Computer aided evaluation and selec-
tion of optimum grippers. International Journal of Production Research, 30, 2713 ± 2732.
BASS, S. M . and K WAKERNAAK, H ., 1977, Rating and ranking of multiple alternatives using
fuzzy sets. Automatica, 13, 47± 58.
BEHNEZHAD, A . R . and K HOSHNEVIS, B., 1988, The e€ ects of manufacturing progress function
on machine requirements and aggregate planning problems. International Journal of
Production Research, 26, 309 ± 326.
BELLMAN, R . E . and Z EDAH, L . A ., 1970, Decision in a fuzzy environment. Management
Science, 17, 141 ± 164.
C HANG, W . , 1981, Ranking of fuzzy utilities with triangular membership functions. In
Proceedings International Conference on Policy Analysis and Information Systems, pp.
263 ± 272.
C HEN, S. J ., H WANG, C. L . and H WANG, F . P ., 1992, Fuzzy Multiple Attribute Decision
Making: Methods and Application (Berlin: Springer).
D ONG, W . M . and W ONG, F . S., 1987, Fuzzy weighted averages and implementation of the
extension principle. Fuzzy Sets and Systems, 21, 183 ± 199.
K NOTT, K . and G ETTO, R . D ., 1982, A model for evaluating alternative robot systems under
uncertainty. International Journal of Production Research, 20, 155 ± 165.
K USIAK, A ., 1987, The production equipment requirements problem. International Journal of
Production Research, 25, 319 ± 325.
L IOU, T . S., J IUN, M . and W ANG, J ., 1992, Fuzzy weighted average: an improved algorithm.
Fuzzy Sets and Systems, 49, 307± 315.
M ILLER, V. M . and D AVIS, R . P ., 1977, The machine requirements problems. International
Journal of Production Research, 15, 219 ± 231.
Flexible manufacturing cells 2097

M URAKAMI, S., M AEDA, S. and I MAMURA, S., 1983, Fuzzy decision analysis on the develop-
ment of centralized regional energy control system. In IFAC Symposium on Fuzzy
Information, Knowledge Representation and Decision Analysis, pp. 363 ± 368.
N AKAMURA, K ., 1986, Preference relation on set of fuzzy utilities as a basis for decision
making. Fuzzy Sets and Systems, 20, 147 ± 162.
STEWART, T . J ., 1992, A critical survey on the status of multiple criteria decision making
theory and practice. OMEGA International Journal of Management Science, 20, 569 ±
586.
T ONG, R . M . and BOUISSONE, P . P ., 1986 , A linguistic approach to decision making with fuzzy
sets. IEEE T ransactions on Systems, Man and Cybernetics, 20, 147 ± 162.
Y AGER, R . R ., 1977, Multiple objective decision making using fuzzy sets. International Journal
of Man-Machine Studies, 9, 375± 382.
Y AGER, R . R ., 1978, Fuzzy decision making including unequal objectives. Fuzzy Sets and
Downloaded by [Universitat Politècnica de València] at 06:55 27 October 2014

Systems, 1, 87± 95.


Y AGER, R . R ., 1980, On a general class of fuzzy connectives. Fuzzy Sets and Systems, 4, 235 ±
242.
Y AGER, R . R ., 1981, A procedure for ordering fuzzy subsets of the unit interval. Information
Science, 24, 143 ± 161.
Y UAN, Y ., 1991, Criteria for evaluating fuzzy ranking methods. Fuzzy Sets and Systems, 43,
139 ± 157.
Z IMMERMANN, H . J ., 1991, Fuzzy Sets T heory and its Applications, second edition (Kluwer
Academic) , pp. 265± 287.

You might also like