ARTICLE IN PRESS

Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137

www.elsevier.com/locate/rcim

Multiple-attribute decision making methods

for plant layout design problem
Taho Yang, Chih-Ching Hung
Institute of Manufacturing Engineering, National Cheng Kung University, Tainan 701, Taiwan, ROC
Received 19 August 2005; received in revised form 8 December 2005; accepted 20 December 2005

Abstract

The layout design problem is a strategic issue and has a significant impact on the efficiency of a manufacturing system. Much of the
existing layout design literature that uses a surrogate function for flow distance or for simplified objectives may be entrapped into local
optimum; and subsequently lead to a poor layout design due to the multiple-attribute decision making (MADM) nature of a layout
design decision. The present study explores the use of MADM approaches in solving a layout design problem. The proposed
methodology is illustrated through a practical application from an IC packaging company. Two methods are proposed in solving the
case study problem: Technique for order preference by similarity to ideal solution (TOPSIS) and fuzzy TOPSIS. Empirical results showed
that the proposed methods are viable approaches in solving a layout design problem. TOPSIS is a viable approach for the case study
problem and is suitable for precise value performance ratings. When the performance ratings are vague and imprecise, the fuzzy TOPSIS
is a preferred solution method.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Plant layout; Fuzzy set theory; Multiple-attribute decision making; TOPSIS

1. Introduction Additional algorithmic approaches could model the

layout design problem as a mixed integer programming
Layout design invariably has a significant impact on the formulation [5–8]. These approaches use the flow distance
performance of a manufacturing or service industry as the surrogate function, and are often computationally
system, and consequently has been an active research area prohibitive.
for several decades [1,2]. Much of the plant layout design The procedural approach, such as the systematic layout
literature is either algorithmic or of a procedural type. The planning procedure [9], has the flexibility to incorporate a
former approach, such as Spirals [3] and MULTIPLE [4], variety of design objectives but is often lacking sound
can efficiently generate alternative layout designs, but the theoretical foundation and credence to be a quality
design objectives are often over-simplified. For example, solution [10].
the resulting departmental shapes often deviate from The layout decision is usually based on both quantitative
practical constraints. For another example, the flow and qualitative performance ratings pertaining to the
distance, either measured in Euclidean or rectilinear desired closeness or closeness relationships among the
distance, and so may not represent the physical flow facilities. The ‘closeness’ is a vague notion that captures
distance. This is particularly important when there are issues such as the material flow and the ease of employee
qualitative design criteria; causing the resulting layout supervision [11]. Clearly, the evaluation of critical criteria
design to lack functionality and credence for a quality for a layout design is often a challenging and complex task
solution. [12,13].
The present study focuses on the evaluation of alter-
Corresponding author. Tel.: +886 6 2090780; fax: +886 6 2085334. native layout designs by considering both qualitative and
E-mail address: [email protected] (T. Yang). quantitative design criteria. It simultaneously evaluates all

0736-5845/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.rcim.2005.12.002
 ARTICLE IN PRESS
T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137 127

the desired criteria for design alternatives. This will permit department placement and departmental dimensions are
the desired design criteria to be better incorporated and not explicitly considered.
evaluated. In addition, the direct evaluation of a design Badiru and Arif [19] proposed a fuzzy linguistic expert
alternative in lieu of incomplete design, e.g., an improve- system in solving a layout design problem. It incorporated
ment type layout design algorithm, will increase the level of an existing layout algorithm, BLOCPLAN [20], to
confidence in searching for a quality solution. It solves a efficiently create design alternatives. Their proposed expert
layout design problem using multiple-attribute decision system is an integrated system with three major compo-
making (MADM) methods. It seeks to evaluate a large nents—fuzzy algorithm, BLOCPLAN and expert system
number of layout design alternatives generated by an (knowledge-based rules). The interactions among the three
efficient layout design algorithm. The evaluation of a large components have the merits of computational efficiency
number of design alternatives will thereby reduce the risk and fuzzy linguistic modeling capability for a layout design
of missing a high-quality solution. problem. The system is fundamentally an improvement-
We propose two MADM methods in solving a plant type layout design algorithm.
layout design problem. They are: technique for order In the study of decision making, terms such as multiple
preference by similarity to ideal solution (TOPSIS) and objective, multiple attribute and multiple criteria are often
fuzzy TOPSIS. A case study from an integrated-circuit (IC) used interchangeably. Here, we provide the conceptual
packaging plant is adopted for the empirical testing. distinctions leading to the definition of the proposed
The remainder of this paper is organized as follows. The MADM methods. However, more detailed information
pertinent literature is reviewed in Section 2. Section 3 can be found in Ribeiro [21–23].
provides the background information for the case study Multiple objective decision making (MODM) consists of
problem. The theories and empirical description for the a set of conflicting goals that cannot be achieved
two methods are discussed in detail sequentially in Sections simultaneously. It invariably concentrates on continuous
4 and 5. The discussion that summarizes the empirical decision spaces and can be solved with mathematical
results is given in Section 6. Finally, conclusions and future programming techniques. MODM generally deals with (i)
research opportunities are drawn together in Section 7. preferences relating to the decision maker’s objectives and
(ii) the relationships between objectives and attributes. An
alternative could be described whether in terms of its
2. Literature review attributes or in terms of the attainment of the decision
maker’s objectives.
Karray et al. [11] proposed an integrated methodology MADM deals with the problem of choosing an option
using the fuzzy set theory and genetic algorithms to from a set of alternatives which are characterized in terms
investigate the layout of temporary facilities in relation to of their attributes. MADM is a qualitative approach due to
the planned buildings in a construction site. It identifies the the existence of criteria subjectivity. It requires information
closeness relationship values between each pair of facilities on the preferences among the instances of an attribute, and
in a construction site using fuzzy linguistic representation. the preferences across the existing attributes. The decision
Grobelny [14,15] explored the use of a fuzzy approach to maker may express or define a ranking for the attributes as
facilities layout problems using a fuzzy criterion to importance/weights. The aim of the MADM is to obtain
determine the closeness relationship among departments; the optimum alternative that has the highest degree of
and then to determine the final optimum design. Evans satisfaction for all of the relevant attributes.
et al. [16] and Dweiri and Meier [17] used a similar concept TOPSIS and fuzzy TOPSIS have been applied to solve a
that employed the theory of fuzzy sets to solve a block variety of applications, and are proven methodology in
layout design problem. solving MADM problems [24,25]. The present study
Raoot and Rakshit [18] proposed a construction-type explores the use of TOPSIS and fuzzy TOPSIS to solve
layout design heuristic based on fuzzy set theory. the proposed layout design problem since we are not aware
A linguistic variable was used to model various qualitative of a similar application. Details of the proposed case and
design criteria, and then to determine the closeness methodology are discussed sequentially in the following
relationship among departments. The resulting closeness sections.
relationship matrix was used to construct a layout design.
This approach allowed, in a qualitative manner, for the
systematic treatment of uncertainty due to fuzziness. 3. The case
All of the above fuzzy-based layout design algorithms
modeled the fuzzy or linguistic closeness relationship The layout design problem presented in Yang and Kuo
among departments. The resulting fuzzy scores that [10] is adopted for the present study. It is an IC packaging
represent the desired closeness are then used for a layout plant. The detail of IC fabrication process is not discussed
design criterion along as part of the layout improvement in this paper for a concise presentation. Interested readers
process. In these methods, the fuzzy closeness determines are referred to Xiao [26] for a detailed discussion of the IC
the order of entry of departments into the layout; but the fabrication process.
 ARTICLE IN PRESS
128 T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137

The IC packaging plant usually adopts the process quantitative and three qualitative design attributes. The
layout strategy that clusters the same tool type to form a quantitative attributes included material handling distance
workstation. A product traverses all the workstations in (in ‘meters’), adjacency score and shape ratio, which are the
the same sequence. For the case study problem, there are direct outputs of Spirals. They are referred to as C1, C2
ten departments (workstations) whose names and area size and C3, respectively, hereafter.
information are depicted in Table 1. The handling distance was measured by the sum of the
The case study is based on an existing layout design. products of flow volume and rectilinear distance between
Understandably, the company would like to know whether the centroids of two departments. The adjacency score is
the existing design is an effective one. The experience the sum of all positive relationships between adjacent
learned from this study will provide guidelines for the departments. There is a positive relationship between each
company’s future layout design problem, as well as for two consecutive departments along the process routing.
identifying potential layout improvement opportunities. Shape ratio is defined as the maximum of the depth-to-
In Yang and Kuo [10], a set of potential ‘good’ layout width and width-to-depth ratio of the smallest rectangle
alternatives were generated by commercial software, that completely encloses the department. The shape ratio is
entitled Spirals [3]. According to the flow distance always greater or equal to one. For a layout design
criterion, the top 17 layout design alternatives were problem, we endeavour to minimize both the shape ratio
generated and selected for further analysis. The existing and flow distance, while maximizing adjacency score.
layout design was the 18th alternative choice. A prelimin- There are three qualitative attributes—flexibility, acces-
ary study was conducted to determine the design criteria sibility and maintenance. They are referred to as C4, C5 and
among the area experts that subsequently led to three C6, respectively, hereafter. Flexibility involves two aspects:
the first is the capability to perform a variety of tasks under
Table 1 a variety of operating conditions; second is the flexibility of
Layout data future expansion. Accessibility involves material handing
and operator paths. Finally, the maintenance issue involves
No. Department name Size (m2)
the required space for maintenance engineers and tool
1 Wafer sawing 89.21 movement. The qualitative attributes are evaluated using
2 Die bond 181.51 an analytic hierarchy process (AHP). Readers are referred
3 Wire bond 577.38
to Yang and Kuo [10] for details of the AHP evaluation
4 Molding 599.57
5 Dejunk/trimming & curing 183.71 process, as well as the block layout figures of the 17 design
6 Electro deflash/solder platting 500.13 alternatives. The performance ratings for the 18 alter-
7 Marking 199.94 natives with respect to the six attributes are summarized in
8 Forming and singulation 186.40 Table 2: decision matrix.
9 Lead scanning/inspection 110.78
Yang and Kuo [10] adopted a data envelopment analysis
10 Packaging 51.09
(DEA) approach in solving the case study problem. DEA is

Table 2
Decision matrix

No. C1 C2 C3 C4 C5 C6

A1 185.9500 8.0000 8.2800 0.0119 0.0260 0.0690

A2 207.3700 9.0000 3.7500 0.0595 0.0260 0.0575
A3 206.3800 8.0000 7.8500 0.0714 0.0519 0.0345
A4 189.6600 8.0000 8.2800 0.0714 0.0779 0.0460
A5 211.4600 8.0000 7.7100 0.0714 0.0390 0.0460
A6 264.0700 5.0000 2.0700 0.0357 0.0519 0.0690

A7 228.0000 8.0000 14.0000 0.0476 0.0390 0.0230

A8 185.5900 9.0000 6.2500 0.0476 0.0130 0.0575
A9 185.8500 9.0000 7.8500 0.0357 0.0260 0.0575
A10 236.1500 8.0000 7.8500 0.0595 0.0779 0.0690
A11 183.1800 8.0000 2.0000 0.0952 0.1169 0.0920
A12 204.1800 8.0000 13.3000 0.0357 0.0390 0.0575

A13 225.2600 8.0000 8.1400 0.0714 0.0390 0.0345

A14 202.8200 8.0000 8.0000 0.0357 0.0779 0.0575
A15 170.1400 9.0000 8.2800 0.0952 0.1169 0.0920
A16 216.3800 9.0000 7.7100 0.0476 0.0519 0.0690
A17 179.8000 8.0000 10.3000 0.0476 0.0779 0.0345
A18 185.7500 10.0000 10.1600 0.0595 0.0519 0.0345
 ARTICLE IN PRESS
T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137 129

a nonparametric approach that does not require any shown in Eq. (1):
assumptions about the functional form of the production C1 C2 C3


Cn
function. Assume that there are n decision-making units 2 3
A1x11 x12 x13


x1n
(DMUs) to be evaluated. Each DMU consumes varying
amount of m different inputs to produce s different A2 6 7
6 x21 x22 x23


x2n 7
6 7
6 7 (1)
outputs. Let D ¼ A3 6 x31 x32 x33


x3n 7:
DMUk ¼ the kth decision making unit (DMU), k ¼ 1, .. 6 . . . . . 7
2,y, n; . 6
4 .
. .. .. .. .. 7
5
Xik ¼ the ith input for the kth DMU, i ¼ 1, 2, y, m and Am xm1 xm2 xm3


xmn
k ¼ 1, 2,y, n;
Yrk ¼ the rth output for the kth DMU, r ¼ 1, 2, y, s Hwang and Yoon [30] developed TOPSIS based on the
and k ¼ 1, 2,y, n; concept that the chosen alternative should have the
vi ¼ the associated weight for the ith input, i ¼ 1, 2, y, shortest distance from the positive ideal solution and the
m; longest distance from the negative ideal solution. More
ur ¼ the associated weight for the rth output, r ¼ 1, detailed information can be found in Yoon and Hwang
2, y, s; and [24]. The terms used in the present study are briefly defined
hk ¼ efficiency score (hk p1). as follows:
Specifically, DMUk consumes amount Xik of input i and
produces amount Yrk of output r, that can be incorporated  Attributes: Attributes (Cj, j ¼ 1, 2, y, n) should provide
into an efficiency measure—the weighted sum of the a means of evaluating the levels of an objective. Each
outputs alternative can be characterized by a number of
P divided P by the weighted sum of the inputs
hk ¼ urYrk/ viXik. This definition requires a set of attributes.
factor weights ur and vi, which are the decision variables.  Alternatives: These are synonymous with ‘options’ or
Each DMUk is assigned the highest possible efficiency ‘candidates’. Alternatives (Ai, i ¼ 1, y, m) are mutually
score (hk p1) by choosing the optimal weights for the exclusive of each other.
outputs and inputs [27]. DEA has been applied to a variety  Attribute weights: Weight values (wj) represent the
of applications for choosing performance frontiers [28]. relative importance of each attribute to the others.
There are constraints for the application of the DEA in W ¼ {wj|j ¼ 1, 2, y, n}.
solving a layout design problem. First, it requires at least  Normalization: Normalization seeks to obtain compar-
two design alternatives or decision-making units for each able scales, which allows attribute comparison. The
input or output measure [29]. Thus, it may become a vector normalization approach divides the rating of each
constraint when there are many performance measures. attribute by its norm to calculate the normalized value
Second, the idea of performance frontiers often generates of xij as defined in Eq. (2):
several popular choices that all lie along the DEA frontier xij
rij ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pm 2ffi ; i ¼ 1; . . . ; m; j ¼ 1; . . . ; n. (2)
line; it is difficult to realize the discrepancy among those
i¼1 ijx
top choices.
This research explores the use of TOPSIS and fuzzy
TOPSIS in solving the proposed layout design problem. Given the above terms, the formal TOPSIS procedure is
The TOPSIS uses specific values for MADM problem, defined as follows:
while the fuzzy TOPSIS is applied to the instances of Step 1: Calculate normalized rating for each element
imprecise and fuzzy performance ratings. in the decision matrix.
Step 2: Calculate weighted normalized ratings. The
weighted normalized value vij is calculated by Eq. (3).
vij ¼ wj rij ; i ¼ 1; . . . ; m; j ¼ 1; . . . ; n. (3)
4. TOPSIS
Step 3: Identify positive ideal (A*) and negative ideal
4.1. Principles of TOPSIS (A) solutions. The A* and A are defined in terms of the
weighted normalized values, as shown in Eqs. (4) and (5),
A MADM problem can be concisely expressed in a respectively:
matrix format, in which columns indicate attributes n o
considered in a given problem; and in which rows list the A ¼ v1 ; v2 ; . . . ; vj ; . . . ; vn
competing alternatives. Specifically, a MADM problem ¼ fðmax vij jj 2 J 1 Þ; ðmin vij jj 2 J 2 Þji ¼ 1; . . . ; mg, ð4Þ
with m alternatives (A1, A2, y, Am) that are evaluated by n i i

attributes (C1, C2, y, Cn) can be viewed as a geometric n o

system with m points in n-dimensional space. An element A ¼ v   
1 ; v2 ; . . . ; vj ; . . . ; vn
xij of the matrix indicates the performance rating of the ith
alternative, Ai, with respect to the jth attribute, Cj, as ¼ fðmin vij jj 2 J 1 Þ; ðmax vij jj 2 J 2 Þji ¼ 1; . . . ; mg, ð5Þ
i i
 ARTICLE IN PRESS
130 T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137

where J1 is a set of benefit attributes (larger-the-better type) normalized decision matrix for the TOPSIS analysis is
and J2 is a set of cost attributes (smaller-the-better type). shown as Table 3.
Step 4: Calculate separation measures. The separation The second step requires the attribute weight informa-
(distance) between alternatives can be measured by the n- tion to calculate the weighted normalized ratings. The
dimensional Euclidean distance. The separation of each decision for the attribute weighting is often ambiguous;
alternative from the positive ideal solution, A*, is given by therefore we adopt the numeric scale method proposed by
Eq. (6): Ribeiro [21]. It uses a five grade scale from ‘‘extremely
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi important (the grade of 5)’’ to ‘‘extremely unimportant (the
uX
u n  2
grade of 1)’’. The calculation algorithm is shown as Eq. (9):
S i ¼ t vij  vj ; i ¼ 1; . . . ; m. (6)
j¼1
gradej
wj ¼ Pn ; j ¼ 1; 2; . . . ; n, (9)
Similarly, the separation from the negative ideal solu- j¼1 gradej
tion, A, is given by Eq. (7):
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where gradej is the grade scale for attribute Cj. Accord-
uX
u n  2 ing to experts’ opinion, the grade scales for the six
S ¼ t v ij  v  ; i ¼ 1; . . . ; m. (7)
i j attributes are {4, 4, 3, 2, 4, 3}. We collected a pretty
j¼1
unanimous conclusion during the weight data collection
Step 5: Calculate similarities to ideal solution. This is process, and thus, do not feel the compelling need to
defined in Eq. (8): develop a more sophisticated approach. Then, the result-
ing numeric scale weights using Eq. (9) are shown as
S
i
C i ¼ ; i ¼ 1; . . . ; m. (8) Eq. (10).
S i þ S
i

Note that 0pC i p1, where C i ¼ 0 when Ai ¼ A , and W ¼ f4=20; 4=20; 3=20; 2=20; 4=20; 3=20g
C i
¼ 1 when Ai ¼ A . ¼ f0:20; 0:20; 0:15; 0:10; 0:20; 0:15g. ð10Þ
Step 6: Rank preference order. Choose an alternative
with maximum Ci* or rank alternatives according to Ci* in The third step finds the weighted normalized decision
descending order. matrix. The analysis then proceeds to Steps 4 and 5. The
results are summarized in Table 4.
4.2. Empirical illustrations for TOPSIS method Finally, the sixth step ranks the alternative according to
Table 4 results as follows:
The decision matrix from Table 2 is used for the TOPSIS
analysis. Based on the first step of the TOPSIS procedure, A11 4A15 4A10 4A4 4A14 4A6 4A17 4A16 4A2 4A3
each element is normalized by Eq. (2). The resulting 4A18 4A5 4A8 4A13 4A9 4A1 4A12 4A7 .

Table 3
Normalized decision matrix for TOPSIS analysis

No. C1 C2 C3 C4 C5 C6

A1 0.2137 0.2277 0.2311 0.0472 0.0978 0.2771

A2 0.2383 0.2562 0.1047 0.2362 0.0978 0.2309
A3 0.2372 0.2277 0.2191 0.2835 0.1952 0.1385
A4 0.2180 0.2277 0.2311 0.2835 0.2931 0.1847
A5 0.2430 0.2277 0.2152 0.2835 0.1467 0.1847
A6 0.3035 0.1423 0.0578 0.1417 0.1952 0.2771
A7 0.2620 0.2277 0.3908 0.1890 0.1467 0.0924
A8 0.2133 0.2562 0.1745 0.1890 0.0489 0.2309
A9 0.2136 0.2562 0.2191 0.1417 0.0978 0.2309
A10 0.2714 0.2277 0.2191 0.2362 0.2931 0.2771
A11 0.2105 0.2277 0.0558 0.3780 0.4398 0.3694
A12 0.2347 0.2277 0.3713 0.1417 0.1467 0.2309
A13 0.2589 0.2277 0.2272 0.2835 0.1467 0.1385
A14 0.2331 0.2277 0.2233 0.1417 0.2931 0.2309
A15 0.1955 0.2562 0.2311 0.3780 0.4398 0.3694
A16 0.2487 0.2562 0.2152 0.1890 0.1952 0.2771
A17 0.2066 0.2277 0.2875 0.1890 0.2931 0.1385
A18 0.2135 0.2847 0.2836 0.2362 0.1952 0.1385

wj 0.2000 0.2000 0.1500 0.1000 0.2000 0.1500

 ARTICLE IN PRESS
T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137 131

Table 4
TOPSIS analysis results

vi1 vi2 vi3 vi4 vi5 vi6 Si* Si Ci*

A1 0.0427 0.0455 0.0347 0.0047 0.0196 0.0416 0.0034 0.0010 0.2273

A2 0.0477 0.0512 0.0157 0.0236 0.0196 0.0346 0.0027 0.0017 0.3864
A3 0.0474 0.0455 0.0329 0.0283 0.0390 0.0208 0.0022 0.0013 0.3714
A4 0.0436 0.0455 0.0347 0.0283 0.0586 0.0277 0.0013 0.0021 0.6176
A5 0.0486 0.0455 0.0323 0.0283 0.0293 0.0277 0.0025 0.0011 0.3056
A6 0.0607 0.0285 0.0087 0.0142 0.0390 0.0416 0.0022 0.0021 0.4884
A7 0.0524 0.0455 0.0586 0.0189 0.0293 0.0139 0.0042 0.0005 0.1064
A8 0.0427 0.0512 0.0262 0.0189 0.0098 0.0346 0.0036 0.0013 0.2653
A9 0.0427 0.0512 0.0329 0.0142 0.0196 0.0346 0.0032 0.0011 0.2558
A10 0.0543 0.0455 0.0329 0.0236 0.0586 0.0416 0.0011 0.0023 0.6765
A11 0.0421 0.0455 0.0084 0.0378 0.0880 0.0554 0.0001 0.0060 0.9836
A12 0.0469 0.0455 0.0557 0.0142 0.0293 0.0346 0.0034 0.0007 0.1707
A13 0.0518 0.0455 0.0341 0.0283 0.0293 0.0208 0.0028 0.0010 0.2632
A14 0.0466 0.0455 0.0335 0.0142 0.0586 0.0346 0.0013 0.0020 0.6061
A15 0.0391 0.0512 0.0347 0.0378 0.0880 0.0554 0.0004 0.0052 0.9286
A16 0.0497 0.0512 0.0323 0.0189 0.0390 0.0416 0.0018 0.0016 0.4706
A17 0.0413 0.0455 0.0431 0.0189 0.0586 0.0208 0.0019 0.0018 0.4865
A18 0.0427 0.0569 0.0425 0.0236 0.0390 0.0208 0.0025 0.0013 0.3421
vj* 0.0391 0.0569 0.0084 0.0378 0.0880 0.0554
vj 0.0607 0.0285 0.0586 0.0047 0.0098 0.0139
W ¼ (0.2, 0.2, 0.15, 0.1, 0.2, 0.15)

Eventually, only the ‘best’ will be adopted as the final are shown by Eq. (11) [32]:
design. Our attention should focus on the top few 8
> 0; xpa1 ;
choices if there is a need for further investigation or >
>
< xa1 ; a1 oxpa2 ;
discussion. a a
ma~ ðxÞ ¼ a23 x1 (11)
>
> a a ; a2 oxpa3 ;
>
: 3 2
0; x4a3 ;
5. Fuzzy TOPSIS

5.1. Fuzzy TOPSIS model Definition 5.2. Let a~ ¼ ða1 ; a2 ; a3 Þ and b~ ¼ ðb1 ; b2 ; b3 Þ be
two triangular fuzzy numbers, then the vertex method is
It is often difficult for a decision-maker to assign a defined to calculate the distance between them, as Eq. (12):
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
precise performance rating to an alternative for the 1 
~
~ bÞ ¼
dða; ða1  b1 Þ2 þ ða2  b2 Þ2 þ ða3  b3 Þ2 . (12)
attributes under consideration. The merit of using a fuzzy 3
approach is to assign the relative importance of attributes
using fuzzy numbers instead of precise numbers. This
Property 5.1. Assuming that both a~ ¼ ða1 ; a2 ; a3 Þ and b~ ¼
section extends the TOPSIS to the fuzzy environment. This
ðb1 ; b2 ; b3 Þ are real numbers, then the distance measurement
method is particularly suitable for solving the group ~ is identical to the Euclidean distance [33].
~ bÞ
dða;
decision-making problem under fuzzy environment. We
briefly review the rationale of fuzzy theory before the
development of fuzzy TOPSIS; as follows: Property 5.2. Let a, ~ and c~ be three triangular fuzzy
~ b,
numbers. The fuzzy number b~ is closer to fuzzy number a~
than the other fuzzy number c~ if, and only if, dða; ~
~ bÞodð ~ c~Þ
a;
Definition 5.1. A fuzzy set a~ in a universe of discourse X is [33].
characterized by a membership function ma~ ðxÞ which
associates with each element x in X, a real number in the The basic operations on fuzzy triangular numbers are as
interval [0, 1]. The function value ma~ ðxÞ is termed the grade follows:
of membership of x in a~ [31].
a~  b~ ¼ ða1  b1 ; a2  b2 ; a3  b3 Þ for multiplication; (13)
The present study uses triangular fuzzy numbers.
A triangular fuzzy number a~ can be defined by a triplet
(a1, a2, a3). Its conceptual schema and mathematical form a~ þ b~ ¼ ða1 þ b1 ; a2 þ b2 ; a3 þ b3 Þ for addition: (14)
 ARTICLE IN PRESS
132 T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137

The fuzzy MADM can be concisely expressed in matrix procedure. For this instance, the D~ defined by Eq. (15) is
format as Eqs. (15) and (16). equivalent to the R~ defined by Eq. (17).
Step 2: Construct the weighted normalized fuzzy
C1 C2 C3 . . . Cn decision matrix. The weighted normalized value V~ is
2 3 calculated by Eq. (18).
A1 x~ 11 x~ 12 x~ 13 . . . x~ 1n
6 7 Step 3: Identify positive ideal (A*) and negative ideal
A2 6 7
6 x~ 21 x~ 22 x~ 23 . . . x~ 2n 7 (A) solutions. The fuzzy positive-ideal solution (FPIS, A*)
6 7
6 7 and the fuzzy negative-ideal solution (FNIS, A) are
D~ ¼ A3 6 x~ 31 x~ 32 x~ 33 . . . x~ 3n 7, ð15Þ
6 7 shown as Eqs. (19) and (20):
.. 6 .. .. .. .. .. 7  
. 6
4 . . . . . 7
5 A ¼ v~1 ; v~2 ; . . . ; v~n
Am x~ m1 x~ m2 x~ m3 . . . x~ mn ¼ fðmax vij ji ¼ 1; 2; . . . ; mÞ; j ¼ 1; 2; . . . ; ng, ð19Þ
i
 
~ ¼ ½w~ 1 ; w~ 2 ; . . . ; w~ n ,
W (16) A ¼ v~  
1 ; v~2 ; . . . ; v~n

where x~ ij , i ¼ 1, 2,y, m, j ¼ 1, 2,y, n and w~ j , j ¼ 1, 2,y, ¼ fðmin vij ji ¼ 1; 2; . . . ; mÞ; j ¼ 1; 2; . . . ; ng. ð20Þ
i
n are linguistic triangular fuzzy numbers, x~ ij ¼ ðaij ; bij ; cij Þ
Step 4: Calculate separation measures. The distance of
and w~ j ¼ ðwj1 ; wj2 ; wj3 Þ. Note that x~ ij is the performance
each alternative from A* and A can be currently
rating of the ith alternative, Ai, with respect to the jth
calculated using Eqs. (21) and (22).
attribute, Cj and w~ j represents the weight of the jth
attribute, Cj. Xn  
d i ¼ d v~ij ; v~j ; i ¼ 1; 2; . . . ; m (21)
The normalized fuzzy decision matrix denoted by R~ is
j¼1
shown as Eq. (17):
X
n  
R~ ¼ ½~rij mn . (17) d d v~ij ; v~
i ¼ j ; i ¼ 1; 2; . . . ; m (22)
The weighted fuzzy normalized decision matrix is shown j¼1

as Eq. (18): Step 5: Calculate similarities to ideal solution. This

2 3 step solves the similarities to an ideal solution by Eq. (23):
v~11 v~12


v~1j


v~1n
6 7 d
6 v~21 v~22


v~2j


v~2n 7 CC i ¼ i
. (23)
6 7 d i þ d 
6 7 i
6 . . . . . . 7
6 .. .. .. .. .. .. 7 Step 6: Rank preference order. Choose an alternative
6 7
V~ ¼ 6
6 v~
7
7 with maximum CCi* or rank alternatives according to
6 i1 v~i2


v~ij


v~in 7 CCi* in descending order.
6 7
6 . .. .. .. .. .. 7 The proposed fuzzy TOPSIS is then applied to the case
6 .. . . . . . 7
6 7 study as shown in Sections 5.2 and 5.3.
4 5
v~m1 v~m2


v~mj


v~mn
2 3 5.2. Fuzzy membership function
w~ 1 r~11 w~ 2 r~12


w~ j r~1j


w~ n r~1n
6 7
6 w~ 1 r~21 w~ 2 r~22


w~ j r~2j


w~ n r~2n 7 The decision makers use the linguistic variables to
6 7
6 7 evaluate the importance of attributes and the ratings of
6 . . . . . . 7
6 .. .. .. .. .. .. 7 alternatives with respect to various attributes. The present
6 7
¼66 w~ r~
7.
7 ð18Þ study has only precise values for the performance ratings
6 1 i1 w~ 2 r~i2


w~ j r~ij


w~ n r~in 7 and for the attribute weights. In order to illustrate the idea
6 7
6 . .. .. .. .. .. 7 of fuzzy MACD, we deliberately transform the existing
6 .. . . . . . 7
6 7 precise values to five-levels, fuzzy linguistic variables—very
4 5
w~ 1 r~m1 w~ 2 r~m2


w~ j r~mj


w~ n r~mn low (VL), low (L), medium (M), high (H) and very high
(VH). The purpose of the transformation process has two
Given the above fuzzy theory, the proposed fuzzy folds as: (i) to illustrate the proposed fuzzy TOPSIS
TOPSIS procedure is then defined as follows: method and (ii) to benchmark the empirical results with
Step 1: Choose the linguistic ratings (x~ ij , i ¼ 1, 2,y, other precise value methods in the later analysis.
m, j ¼ 1, 2,y, n) for alternatives with respect to criteria Among the commonly used fuzzy numbers, triangular
and the appropriate linguistic variables (w~ j , j ¼ 1, 2,y, n) and trapezoidal fuzzy numbers are likely to be the most
for the weight of the criteria. adoptive ones due to their simplicity in modeling and easy
The fuzzy linguistic rating ðx~ ij Þ preserves the property of interpretation. Both triangular and trapezoidal fuzzy
that the ranges of normalized triangular fuzzy numbers numbers are applicable to the present study. We feel that a
belong to [0,1]; thus, there is no need for a normalization triangular fuzzy number can adequately represent the
 ARTICLE IN PRESS
T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137 133

five-level fuzzy linguistic variables and thus, is used for the Table 6
analysis hereafter. Normalized decision matrix for fuzzy TOPSIS analysis
As a rule of thumb, each rank is assigned an evenly No. C1 C2 C3 C4 C5 C6
spread membership function that has an interval of 0.30 or
0.25. Based on these assumptions, a transformation table A1 0.76 0.60 0.12 0.10 0.18 0.71
can be found as shown in Table 5. For example, the fuzzy A2 0.49 0.80 0.46 0.60 0.18 0.57
A3 0.51 0.60 0.13 0.70 0.36 0.25
variable—Very Low has its associated triangular fuzzy
A4 0.71 0.60 0.12 0.70 0.64 0.43
number with minimum of 0.00, mode of 0.10 and A5 0.45 0.60 0.14 0.70 0.27 0.43
maximum of 0.25. The same definition is then applied to A6 0.00 0.00 0.96 0.40 0.36 0.71
the other fuzzy variables—Low, Medium, High and Very
A7 0.29 0.60 0.00 0.50 0.27 0.00
High. Fig. 1 illustrates the fuzzy membership functions. A8 0.77 0.80 0.21 0.50 0.00 0.57
A9 0.76 0.80 0.13 0.40 0.18 0.57
A10 0.21 0.60 0.13 0.60 0.64 0.71
5.3. Empirical illustrations A11 0.80 0.60 1.00 1.00 1.00 1.00
A12 0.53 0.60 0.01 0.40 0.27 0.57
Table 2 numeric performance ratings are adopted again A13 0.31 0.60 0.12 0.70 0.27 0.14
for the fuzzy TOPSIS analysis. In order to transform the A14 0.55 0.60 0.13 0.40 0.64 0.57
performance ratings to fuzzy linguistic variables as A15 1.00 0.80 0.12 1.00 1.00 1.00
discussed in the previous section, the performance ratings A16 0.40 0.80 0.14 0.50 0.36 0.71
A17 0.76 0.60 0.12 0.10 0.18 0.71
in Table 2 are normalized into the range of [0,1] by Eqs. A18 0.49 0.80 0.46 0.60 0.18 0.57
(24) and (25) [34]:
wj 0.2000 0.2000 0.1500 0.1000 0.2000 0.1500

(i) The larger the better type:

   
rij ¼ xij  minfxij g = maxfxij g  minfxij g . (24)
Table 7
(ii) The smaller the better type: Decision matrix using fuzzy linguistic variables
   
rij ¼ maxfxij g  xij = maxfxij g  minfxij g . (25) No. C1 C2 C3 C4 C5 C6

A1 H M VL VL VL H
For the present study, C1 and C3 are the smaller-the- A2 M H M M VL M
better type, the others belong to the larger-the-better type. A3 M M VL H L L
A4 H M VL H M M
Then, Table 2 can be transformed into Table 6. A5 M M VL H L M
A6 VL VL VH L L H
Table 5 A7 L M VL M L VL
Transformation for fuzzy membership functions A8 H H VL M VL M
A9 H H VL L VL M
Rank Attribute grade Membership A10 VL M VL M M H
functions A11 VH H VH VH VH VH
A12 M M VL L L M
Very low (VL) 1 (0.00, 0.10, 0.25)
Low (L) 2 (0.15, 0.30, 0.45) A13 L M VL H L VL
Medium (M) 3 (0.35, 0.50, 0.65) A14 M M VL L M M
High (H) 4 (0.55, 0.70, 0.85) A15 VH H VL VH VH VH
Very high (VH) 5 (0.75, 0.90, 1.00) A16 L H VL M L H
A17 H M VL M M VL
A18 VH VH VL H M VL
Weight H H M L H M
VL L ML H VH
1

The next step uses the fuzzy membership function

0.5 discussed in Section 5.2 to transform Table 6 into
Table 7 as explained by the following example. If the
numeric rating is 0.64, then its fuzzy linguistic variable is
0 ‘‘H’’. This transformation is also applied to the attribute
weight W ¼ {0.20, 0.20, 0.15, 0.10, 0.20, 0.15} for C1, C2,
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
C3, C4, C5, C6, respectively. Then, the resulting fuzzy
Fig. 1. Fuzzy triangular membership functions. linguistic variables are show as Table 7.
 ARTICLE IN PRESS
134 T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137

The fuzzy linguistic variable is then transformed into a ranges belong to the closed interval [0,1]. Thus, we can
fuzzy triangular membership function as shown in Table 8. define the fuzzy positive-ideal solution (FPIS, A*) and the
This is the first step of the fuzzy TOPSIS analysis. The fuzzy negative-ideal solution (FNIS, A) as: v~j ¼ ð1; 1; 1Þ
fuzzy attribute weight is also collected in Table 8. and v~
j ¼ ð0; 0; 0Þ, j ¼ 1, 2, y, n. This is the third step of the
The second step in the analysis is to find the weighted fuzzy TOPSIS analysis.
fuzzy decision matrix. Using Eq. (13), the fuzzy multi- For the fourth step, the distance of each alternative from
plication equation, the resulting fuzzy weighted decision A* and A can be currently calculated using Eqs. (21) and
matrix is shown as Table 9. (22). The fifth step solves the similarities to an ideal solution
According to Table 9, we know that the elements v~ij ; 8i; j by Eq. (23). The resulting fuzzy TOPSIS analyses are
are normalized positive triangular fuzzy numbers and their summarized in Table 10.

Table 8
Fuzzy decision matrix and fuzzy attribute weights

No. C1 C2 C3 C4 C5 C6

A1 (0.55, 0.70, 0.85) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.00, 0.10, 0.25) (0.00, 0.10, 0.25) (0.55, 0.70, 0.85)
A2 (0.35, 0.50, 0.65) (0.55, 0.70, 0.85) (0.35, 0.50, 0.65) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.35, 0.50, 0.65)
A3 (0.35, 0.50, 0.65) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.55, 0.70, 0.85) (0.15, 0.30, 0.45) (0.15, 0.30, 0.45)
A4 (0.55, 0.70, 0.85) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.55, 0.70, 0.85) (0.55, 0.70, 0.85) (0.35, 0.50, 0.65)
A5 (0.35, 0.50, 0.65) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.55, 0.70, 0.85) (0.15, 0.30, 0.45) (0.35, 0.50, 0.65)
A6 (0.00, 0.10, 0.25) (0.00, 0.10, 0.25) (0.75, 0.90, 1.00) (0.15, 0.30, 0.45) (0.15, 0.30, 0.45) (0.55, 0.70, 0.85)
A7 (0.15, 0.30, 0.45) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.35, 0.50, 0.65) (0.15, 0.30, 0.45) (0.00, 0.10, 0.25)
A8 (0.55, 0.70, 0.85) (0.55, 0.70, 0.85) (0.15, 0.30, 0.45) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.35, 0.50, 0.65)
A9 (0.55, 0.70, 0.85) (0.55, 0.70, 0.85) (0.00, 0.10, 0.25) (0.15, 0.30, 0.45) (0.00, 0.10, 0.25) (0.35, 0.50, 0.65)
A10 (0.15, 0.30, 0.45) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.35, 0.50, 0.65) (0.55, 0.70, 0.85) (0.55, 0.70, 0.85)
A11 (0.75, 0.90, 1.00) (0.55, 0.70, 0.85) (0.75, 0.90, 1.00) (0.75, 0.90, 1.00) (0.75, 0.90, 1.00) (0.75, 0.90, 1.00)
A12 (0.35, 0.50, 0.65) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.15, 0.30, 0.45) (0.15, 0.30, 0.45) (0.35, 0.50, 0.65)
A13 (0.15, 0.30, 0.45) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.55, 0.70, 0.85) (0.15, 0.30, 0.45) (0.00, 0.10, 0.25)
A14 (0.35, 0.50, 0.65) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.15, 0.30, 0.45) (0.55, 0.70, 0.85) (0.35, 0.50, 0.65)
A15 (0.75, 0.90, 1.00) (0.55, 0.70, 0.85) (0.00, 0.10, 0.25) (0.75, 0.90, 1.00) (0.75, 0.90, 1.00) (0.75, 0.90, 1.00)
A16 (0.15, 0.30, 0.45) (0.55, 0.70, 0.85) (0.00, 0.10, 0.25) (0.35, 0.50, 0.65) (0.15, 0.30, 0.45) (0.55, 0.70, 0.85)
A17 (0.75, 0.90, 1.00) (0.35, 0.50, 0.65) (0.00, 0.10, 0.25) (0.35, 0.50, 0.65) (0.55, 0.70, 0.85) (0.00, 0.10, 0.25)
A18 (0.55, 0.70, 0.85) (0.75, 0.90, 1.00) (0.00, 0.10, 0.25) (0.55, 0.70, 0.85) (0.15, 0.30, 0.45) (0.00, 0.10, 0.25)
Weight (0.55, 0.70, 0.85) (0.55, 0.70, 0.85) (0.35, 0.50, 0.65) (0.15, 0.30, 0.45) (0.55, 0.70, 0.85) (0.35, 0.50, 0.65)

Table 9
Fuzzy-weighted decision matrix

No. C1 C2 C3 C4 C5 C6

A1 (0.30, 0.49, 0.72) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.00, 0.03, 0.11) (0.00, 0.07, 0.21) (0.19, 0.35, 0.55)
A2 (0.19, 0.35, 0.55) (0.30, 0.49, 0.72) (0.12, 0.25, 0.42) (0.05, 0.15, 0.29) (0.00, 0.07, 0.21) (0.12, 0.25, 0.42)
A3 (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.08, 0.21, 0.38) (0.05, 0.15, 0.29)
A4 (0.30, 0.49, 0.72) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) (0.12, 0.25, 0.42)
A5 (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.08, 0.21, 0.38) (0.12, 0.25, 0.42)
A6 (0.00, 0.07, 0.21) (0.00, 0.07, 0.21) (0.26, 0.45, 0.65) (0.02, 0.09, 0.20) (0.08, 0.21, 0.38) (0.19, 0.35, 0.55)
A7 (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.08, 0.21, 0.38) (0.05, 0.15, 0.29)
A8 (0.30, 0.49, 0.72) (0.30, 0.49, 0.72) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.00, 0.07, 0.21) (0.12, 0.25, 0.42)
A9 (0.30, 0.49, 0.72) (0.30, 0.49, 0.72) (0.00, 0.05, 0.16) (0.02, 0.09, 0.20) (0.00, 0.07, 0.21) (0.12, 0.25, 0.42)
A10 (0.00, 0.07, 0.21) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.19, 0.35, 0.55) (0.19, 0.35, 0.55)
A11 (0.41, 0.63, 0.85) (0.30, 0.49, 0.72) (0.26, 0.45, 0.65) (0.11, 0.27, 0.45) (0.41, 0.63, 0.85) (0.26, 0.45, 0.65)
A12 (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.02, 0.09, 0.20) (0.08, 0.21, 0.38) (0.12, 0.25, 0.42)
A13 (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.08, 0.21, 0.38) (0.00, 0.05, 0.16)
A14 (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.02, 0.09, 0.20) (0.19, 0.35, 0.55) (0.12 ,0.25, 0.42)
A15 (0.41, 0.63, 0.85) (0.30, 0.49, 0.72) (0.00, 0.05, 0.16) (0.11, 0.27, 0.45) (0.41, 0.63, 0.85) (0.26, 0.45, 0.65)
A16 (0.08, 0.21, 0.38) (0.30, 0.49, 0.72) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.08, 0.21, 0.38) (0.19, 0.35, 0.55)
A17 (0.30, 0.49, 0.72) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16)
A18 (0.41, 0.63, 0.85) (0.41, 0.63, 0.85) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16)
 ARTICLE IN PRESS
T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137 135

Table 10
Fuzzy TOPSIS analysis

No. v^i1 v^i2 v^i3 v^i4 v^i5 v^i6 dþ

i
d
i CC i

A1 (0.30, 0.49, 0.72) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.00, 0.03, 0.11) (0.00, 0.07, 0.21) (0.19, 0.35, 0.55) 4.635148 1.606324 0.257363
A2 (0.19, 0.35, 0.55) (0.30, 0.49, 0.72) (0.12, 0.25, 0.42) (0.05, 0.15, 0.29) (0.00, 0.07, 0.21) (0.12, 0.25, 0.42) 4.425966 1.823517 0.291787
A3 (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.08, 0.21,0.38) (0.05, 0.15, 0.29) 4.654412 1.581345 0.253592
A4 (0.30, 0.49, 0.72) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) (0.12, 0.25, 0.42) 4.298028 1.958154 0.312995
A5 (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.08, 0.21, 0.38) (0.12, 0.25, 0.42) 4.558833 1.681219 0.269424
A6 (0.00, 0.07, 0.21) (0.00, 0.07, 0.21) (0.26, 0.45, 0.65) (0.02, 0.09, 0.20) (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) 4.730627 1.510169 0.241983
A7 (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.08, 0.21, 0.38) (0.05, 0.15, 0.29) 4.843391 1.510169 0.221744
A8 (0.30, 0.49, 0.72) (0.30, 0.49, 0.72) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.00, 0.07, 0.21) (0.12, 0.25, 0.42) 4.483565 1.769500 0.282981
A9 (0.30, 0.49, 0.72) (0.30, 0.49, 0.72) (0.00, 0.05, 0.16) (0.02, 0.09, 0.20) (0.00, 0.07, 0.21) (0.12, 0.25, 0.42) 4.54085 1.705947 0.273092
A10 (0.00, 0.07, 0.21) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) 4.646129 1.591422 0.255136
A11 (0.41, 0.63, 0.85) (0.30, 0.49, 0.72) (0.26, 0.45, 0.65) (0.11, 0.27, 0.45) (0.41, 0.63, 0.85) (0.26, 0.45, 0.65) 3.22343 3.112578 0.491252
A12 (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.02, 0.09, 0.20) (0.08, 0.21, 0.38) (0.12, 0.25, 0.42) 4.672235 1.553481 0.249527
A13 (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.08, 0.21, 0.38) (0.00, 0.05, 0.16) 4.877236 1.350266 0.216823
A14 (0.19, 0.35, 0.55) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.02, 0.09, 0.20) (0.19, 0.35, 0.55) (0.12, 0.25, 0.42) 4.539392 1.690642 0.27137
A15 (0.41, 0.63, 0.85) (0.30, 0.49, 0.72) (0.00, 0.05, 0.16) (0.11, 0.27, 0.45) (0.41, 0.63, 0.85) (0.26, 0.45, 0.65) 3.586443 2.728874 0.432104
A16 (0.08, 0.21, 0.38) (0.30, 0.49, 0.72) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) 4.52647 1.721120 0.275485
A17 (0.30, 0.49, 0.72) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) (0.05, 0.15, 0.29) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) 4.539705 1.700176 0.272469
A18 (0.41, 0.63, 0.85) (0.41, 0.63, 0.85) (0.00, 0.05, 0.16) (0.08, 0.21, 0.38) (0.19, 0.35, 0.55) (0.00, 0.05, 0.16) 4.127193 2.150711 0.342584
A v~1 ¼ ð1; 1; 1Þ v~2 ¼ ð1; 1; 1Þ v~3 ¼ ð1; 1; 1Þ v~4 ¼ ð1; 1; 1Þ v~5 ¼ ð1; 1; 1Þ v~6 ¼ ð1; 1; 1Þ
A v~1 ¼ ð0; 0; 0Þ v~2 ¼ ð0; 0; 0Þ v~3 ¼ ð0; 0; 0Þ v~4 ¼ ð0; 0; 0Þ v~5 ¼ ð0; 0; 0Þ v~6 ¼ ð0; 0; 0Þ

W ¼ {(0.55, 0.70, 0.85), (0.55, 0.70, 0.85), (0.35, 0.50, 0.65), (0.15, 0.30, 0.45), (0.55, 0.70, 0.85), (0.35, 0.50, 0.65)}

In order to illustrate Steps 4 and 5 calculation, CC1 d 1

calculation is used as an example as follows: CC 1 ¼
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi d 1
þ d 1
1 1:606324

d1 ¼ ½ð1  0:3Þ2 þ ð1  0:49Þ2 þ ð1  0:72Þ2  ¼ ¼ 0:257363.
3 4:635148 þ 1:606324
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 Based on Table 10, the last step found the preference for
þ ½ð1  0:19Þ2 þ ð1  0:35Þ2 þ ð1  0:55Þ2 
3 the 18 design alternatives as follows:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 A11 4A15 4A18 4A4 4A17 4A8 4A10 4A14 4A2 4A16
þ ½ð1  0Þ2 þ ð1  0:05Þ2 þ ð1  0:16Þ2 
3
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4A9 4A5 4A1 4A3 4A12 4A6 4A7 4A13 .
1
þ ½ð1  0Þ2 þ ð1  0:03Þ2 þ ð1  0:11Þ2  In this section, we deliberately transform the existing
3
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi precise values to fuzzy linguistic variables in order to
1 illustrate the concept of the proposed fuzzy-based method.
þ ½ð1  0Þ2 þ ð1  0:07Þ2 þ ð1  0:21Þ2 
3 It is the aim of this section to illustrate the feasibility of the
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 fuzzy-based method for the instance of fuzzy inputs, which
þ ½ð1  0:19Þ2 þ ð1  0:35Þ2 þ ð1  0:55Þ2 
3 is justified by the empirical results.
¼ 4:635148,
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 6. Discussion
1

d1 ¼ ½ð0  0:3Þ2 þ ð0  0:49Þ2 þ ð0  0:72Þ2 
3 The resulting layouts vary among the different design
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 methods to some extent. The top five design alternatives,
þ ½ð0  0:19Þ2 þ ð0  0:35Þ2 þ ð0  0:55Þ2  according to the two proposed design methods, as well as
3
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi the results by DEA [10], are summarized in Table 11.
1
þ ½ð0  0Þ2 þ ð0  0:05Þ2 þ ð0  0:16Þ2  All methods lead to the choice of A11 a priori as the final
3 layout design. A15 is apparently the second choice. Other
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 than these two alternatives, the preferences vary between
þ ½ð0  0Þ2 þ ð0  0:03Þ2 þ ð0  0:11Þ2 
3 methods. The fuzzy TOPSIS concludes with the same top
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 three alternatives as those DEA. The TOPSIS method
þ ½ð0  0Þ2 þ ð0  0:07Þ2 þ ð0  0:21Þ2  concludes the same top two alternatives as the ones from
3
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi DEA and fuzzy TOPSIS. Due to the MADM nature of the
1 proposed problem, an optimal solution may not exist;
þ ½ð0  0:19Þ2 þ ð0  0:35Þ2 þ ð0  0:55Þ2 
3 however, the systematic evaluation of the MADM problem
¼ 1:606324, can reduce the risk of a poor design.
 ARTICLE IN PRESS
136 T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137

Table 11 viable approaches in solving the proposed layout design

Top five design alternatives from different methods problem. TOPSIS is a viable method for the proposed
Preference 1 2 3 4 5
problem and is suitable for the use of precise performance
order ratings. When the performance ratings are vague and
inaccurate, then the fuzzy TOPSIS is the preferred
TOPSIS A11 A15 A10 A4 A14 technique.
Fuzzy A11 A15 A18 A4 A17
Each layout design application is unique in nature, i.e.,
TOPSIS
DEA A*11 A*15 A*18 A2 A16 there are different attributes associated with different
applications; thus, the success of the present study has no
*Tie. guarantee for its applicability to other applications.
Judicious use of a design method is advised in solving a
specific application. In addition, there exists other worth
When precise performance ratings are available, the investigating MADM methods for a layout design pro-
TOPSIS method is considered to be a viable approach in blem. This becomes one of the future research opportu-
solving a layout design problem. The DEA method from nities in this classical yet important research area.
existing literature is also a viable approach. However, it has
the constraints in the number of decision making units and Acknowledgements
in the limitation to the discrepancy between performance
frontiers. For the instance of imprecise or vague perfor- This work is supported, in part, by the National Science
mance ratings, the fuzzy TOPSIS is a preferred choice in Council of Taiwan, Republic of China, under Grant
solving the proposed design problem. NSC93-2212-E006-065 and NSC94-2212-E006-007.
Although it is the aim of the proposed methodology to
solve the optimal layout design problem, this goal is not
References
achievable due to the nature of a layout design problem. In
addition to the MADM nature, there are other constraints [1] Apple JM. Plant layout and material handling. 3rd ed. New York:
form the shop floor limitations and from the management Wiley; 1997.
philosophy. For instance, the qualitative performance [2] Meller RD, Gau YK. The facility layout problem: recent and
measures are mainly collected from the industry experts. emerging trends, and perspectives. J Manuf Syst 1996;15:351–66.
The group decision making can be complicated, and [3] Goetschalckx M. An interactive layout heuristic based on hexagonal
adjacency graphs. Eur J Oper Res 1992;63:304–21.
perhaps, may not have a unanimous result. Many practical [4] Bozer YA, Meller RD, Erlebacher SJ. An improved-type layout
limitations are not quantifiable. For another example, the algorithm for single and multiple-floor facilities. Manage Sci
management may have their preference against the analysis 1994;40:918–32.
results. Nevertheless, the proposed methodology provides a [5] Heragu SS, Kusiak A. Efficient models for the layout design problem.
Eur J Oper Res 1991;53:1–13.
systematic approach to narrow down the number of
[6] Peters BA, Yang T. Integrated facility layout and material handling
alternatives, and to facilitate the decision making process. system design in semiconductor fabrication facilities. IEEE Trans
In fact, the management of the case study company has Semicond Mater 1997;10:360–9.
adopted the analysis results as the basis to their future [7] Yang T, Peters BA, Tu M. Layout design for flexible manufacturing
layout planning problem. The practical contribution of this systems considering single-loop directional flow patterns. Eur J Oper
research can, thus, be justified. Res 2005;164:440–55.
[8] Chan FTS, Lau KW, Chan PLY, Choy KL. Two-stage approach for
machine-part grouping and cell layout problems. Robot CIM-Int
7. Conclusions Manuf 2006;22:217–38.
[9] Muther R. Systematic layout planning. 2nd ed. Boston, MA: Cahners
The layout design problem is a strategic issue and has Books; 1973.
[10] Yang T, Kuo C. A hierarchical AHP/DEA methodology for the
significant impacts to the efficiency of a manufacturing
facilities layout design problem. Eur J Oper Res 2003;147:128–36.
system. Much of the existing layout design literature that [11] Karray F, Zaneldin E, Hegazy T, Shabeeb AHM, Elbeltagi E. Tools
uses a surrogate function for flow distance or for simplified of soft computing as applied to the problem of facilities layout
objectives may be entrapped into local optimum; and planning. IEEE Trans Fuzzy Syst 2000;8:367–79.
therefore lead to a poor layout design due to the MADM [12] Lin LC, Sharp GP. Quantitative and qualitative indices for the plant
nature of a layout design problem. layout evaluation problem. Eur J Oper Res 1999;116:100–17.
[13] Lin LC, Sharp GP. Application of the integrated framework for the
The present study explored the use of TOPSIS and fuzzy plant layout evaluation problem. Eur J Oper Res 1999;116:118–38.
TOPSIS in solving a layout design problem. A practical [14] Grobelny J. The fuzzy approach to facility layout problems. Fuzzy
application from an IC packaging company was adopted Set Syst 1987;23:175–90.
for empirical testing. This study aimed at searching an [15] Grobelny J. On one possibly ‘fuzzy’ approach to facility layout
improved solution to an existing design. Moreover, the problems. Int J Prod Res 1987;25:1123–41.
[16] Evans GW, Wilhelm MR, Karwowski W. A layout design heuristic
methods and experiences learned from the study can be employing the theory of fuzzy sets. Int J Prod Res 1987;25:1431–50.
valuable to the company’s future strategic planning. [17] Dweiri F, Meier FA. Application of fuzzy decision-making in
Empirical results showed that the proposed methods are facilities layout planning. Int J Prod Res 1996;34:3207–25.
 ARTICLE IN PRESS
T. Yang, C.-C. Hung / Robotics and Computer-Integrated Manufacturing 23 (2007) 126–137 137

[18] Raoot AD, Rakshit A. A ‘fuzzy’ approach to facilities layout [26] Xiao H. Introduction to semiconductor manufacturing technology.
planning. Int J Prod Res 1991;29:835–57. New Jersey: Prentice Hall; 2000.
[19] Badiru AB, Arif A. FLEXPERT: facility layout expert system using [27] Sinuany-Stern Z, Mehrez A, Hadad Y. An AHP/DEA methodology
fuzzy linguistic relationship codes. IEE Trans 1996;28:295–308. for ranking decision making units. Int Trans Oper Res 2000;7:
[20] Donaghey CE, Pire VF. Solving the facilities layout problem with 109–24.
BLOCPLAN. Working paper. Houston, Texas: Industrial Engineer- [28] Seiford LM. Data envelopment analysis: the evolution of the state-of-
ing Department, University of Houston; 1990. the-art 1978–1995. J Prod Anal 1996;7:99–137.
[21] Ribeiro RA. Fuzzy multiple attribute decision making: a review and [29] Bowlin WF. Evaluating the efficiency of US Air Force
new preference elicitation techniques. Fuzzy Set Syst 1996;78:155–81. real-property maintenance activities. J Oper Res Soc 1987;38:
[22] Bellman RE, Zadeh LA. Decision-making in a fuzzy environment. 127–35.
Manage Sci 1970;17:B141–64. [30] Hwang CL, Yoon KP. Multiple attribute decision making: methods
[23] Zimmermann HJ. Fuzzy sets, decision making, and expert systems. and applications. New York: Springer; 1981.
International series in management science/operations research. [31] Zadeh LA. Fuzzy sets. Inform Control 1965;8:338–53.
Dordrecht: Kluwer Academic; 1987. [32] Kaufmann A, Gupta MM. Introduction to fuzzy arithmetic: theory
[24] Yoon KP, Hwang CL. Multiple attribute decision making. Thousand and applications. New York: Van Nostrand Reinhold; 1985.
Oaks, CA: Sage Publication; 1995. [33] Chen CT. Extensions of the TOPSIS for group decision-making
[25] Yang T, Chou P. Solving a multiresponse simulation–optimization under fuzzy environment. Fuzzy Set Syst 2000;114:1–9.
problem with discrete variables using a multiple-attribute decision- [34] Cheng CH. Evaluating weapon systems using ranking fuzzy numbers.
making method. Math Comput Simulat 2005;68:9–21. Fuzzy Set Syst 1999;107:25–35.