Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

European Journal of Operational Research 172 (2006) 230–248

www.elsevier.com/locate/ejor

Production, Manufacturing and Logistics

An evaluation approach to engineering design


in QFD processes using fuzzy goal programming models
a,* b
Liang-Hsuan Chen , Ming-Chu Weng
a
Department of Industrial and Information Management, National Cheng Kung University, Tainan, Taiwan 701, ROC
b
Department of Industrial Management, Kun Shan University of Technology, Tainan County 71003, Taiwan, ROC

Received 9 August 2002; accepted 5 October 2004


Available online 8 December 2004

Abstract

Quality function deployment (QFD) is a product development process used to achieve higher customer satisfaction:
the engineering characteristics affecting the product performance are designed to match the customer requirements.
From the viewpoint of QFDs designers, product design processes are performed in uncertain environments, and usually
more than one goal must be taken into account. Therefore, when dealing with the fuzzy nature in QFD processes, fuzzy
approaches are applied to formulate the relationships between customer requirements (CRs) and engineering design
requirements (DRs), and among DRs. In addition to customer satisfaction, the cost and technical difficulty of DRs
are also considered as the other two goals, and are evaluated in linguistic terms. Fuzzy goal programming models
are proposed to determine the fulfillment levels of the DRs. Differing from existing fuzzy goal programming models,
the coefficients in the proposed model are also fuzzy in order to expose the fuzziness of the linguistic information.
Our model also considers business competition by specifying the minimum fulfillment levels of DRs and the preemptive
priorities between goals. The proposed approach can attain the maximal sum of satisfaction degrees of all goals under
each confidence degree. A numerical example is used to illustrate the applicability of the approach.
Ó 2004 Elsevier B.V. All rights reserved.

Keywords: Quality function deployment (QFD); Fuzzy numbers; Fuzzy goal programming

1. Introduction

Quality function deployment (QFD) is a systematic method for translating the voice of customers into a
final product through various product planning, engineering and manufacturing stages in order to achieve

*
Corresponding author. Tel.: +886 6 2757575/53140; fax: +886 6 2362162.
E-mail address: lhchen@mail.ncku.edu.tw (L.-H. Chen).

0377-2217/$ - see front matter Ó 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.ejor.2004.10.004
L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 231

higher customer satisfaction. The method includes both ‘‘customer requirement management’’ and ‘‘prod-
uct development’’ systems, which begin by sampling the desires and preferences of consumers of a product
through marketing surveys or interviews, and organizes them as a set of customer requirements (CRs). A
group of engineering design requirements (DRs) affecting CRs are then identified, analyzed and improved
in order to maximize customer satisfaction. By analyzing the relationships among DRs and between CRs
and DRs, while considering cost and technical constraints as well as organizational strategies, QFD team
members are responsible for determining the fulfillment levels of DRs. In the conventional QFD approach,
decisions are achieved imprecisely in an uncertain environment because customer requirements tend to be
subjective and qualitative. In addition, data availability for product design is often limited, inaccurate, or
vague—particularly when developing an entirely new product. Therefore, engineers usually do not have the
full knowledge necessary to map CRs onto the relevant DRs.
Some authors, such as Park and Kim (1998) and Trappey et al. (1996), have presented some modified
methods for assigning the relationship ratings between CRs and DRs, instead of a conventional relation-
ship rating scale, such as the three point levels of 1, 3, and 9. However, these methods still use crisp mea-
surement data with the result that ambiguous relationships cannot be captured. Some researchers have
applied fuzzy theory in order to quantitatively formulate the problem for optimizing the improvements
of DRs. Fung et al. (1998) proposed a fuzzy inference system of customer requirements which allowed
the product attributes to be mapped out. Moskowitz and Kim (1997) presented a decision support system
for optimizing product designs. The development of these systems usually requires professional knowledge
and experience to establish rules and facts in ensuring that the system works well. Kim et al. (2000) used a
fuzzy theoretical modeling approach to QFD by developing fuzzy multi-objective models, under the
assumption that the function relationships among DRs and between CRs and DRs could be recognized
based on the benchmarking data set of customer competitive analysis. Justifying this assumption in a gen-
eral situation is difficult, particularly when developing an entirely new product. Some researchers, such as
Shen et al. (2001), Vanegas and Labib (2001), Wang (1999), and Zhou (1998), developed some fuzzy ap-
proaches, for example, fuzzy sets, fuzzy arithmetic, and/or defuzzification techniques, to address complex
and often imprecise problems in customer requirement management. However, in these models the inter-
relationships among the engineering design requirements (DRs) were not properly considered.
In this study, we consider both the inherent fuzziness in the relationships among DRs and those between
CRs and DRs. An aggregation of the two kinds of fuzzy relationships based on WassermanÕs (1993) study is
performed to obtain the fuzzy normalized relationship matrix, containing a fuzzy number in each cell.
Using the matrix and fuzzy weights of CRs, the fuzzy importance ratings of DRs are determined, after
which the customer satisfaction function is formulated. In addition, some authors emphasized the need
of conducting cost consideration and/or technical difficulty in the models in accordance with the QFD plan-
ning effort (Fung et al., 2002; King, 1987; Park and Kim, 1998; Trappey et al., 1996; Wasserman, 1993;
Wang, 1999; Zhou, 1998). Therefore, this paper incorporates the costs and technical difficulties of DRs into
the models so as to formulate three objectives for maximizing customer satisfaction, minimizing cost, and
minimizing technical difficulties. Fuzzy goal programming models are formulated to achieve the objectives
in terms of the fulfillment levels of DRs. Moreover, due to some organizational strategies and constraints in
QFD processes, the design team may have a preference order, i.e., a preemptive priority structure, to
achieve the goals. For this reason, we adopt a preemptive priority structure into the formulations based
on the study of Chen and Tsai (2001). Different from existing models, the goals and coefficients in the pro-
posed models are fuzzy, with the objective of achieving maximal total satisfaction of all goals.
In the following section, a fuzzy approach is introduced to determine the fuzzy normalized relationship
matrix of QFD and fuzzy technical importance ratings for DRs. Section 3 formulates the QFD planning
problem as fuzzy goal programming problems with conflicting objectives and a preemptive priority. This
paper applies the concept of a-cut and the extension principle to transform the fuzzy model into a series
of conventional crisp linear programming models to find the fulfillment levels of DRs so as to produce
232 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

the maximal total satisfaction degree of all goals. Section 4 demonstrates our approach using an example,
and discusses the findings. Finally, we give our conclusion in Section 5.

2. A fuzzy QFD approach

The QFD employs the matrix, called House Of Quality (HOQ), to establish the relationships between
CRs and DRs, as shown in Fig. 1. Two dimensions, customer wants and engineering design requirements,
are included in the matrix. We place a triangular-shaped matrix over the engineering design requirements to
indicate the correlation between engineering design requirements. Wasserman (1993) proposed a normal-
ized transformation on the relationship values contained in the relationship matrix to account for the
dependency effects among DRs, as described by the following equation,
Pn
R r
Pn ik kj ,
R0ij ¼ Pn k¼1 ð1Þ
j¼1 k¼1 Rik rkj

where

R0ij normalized relationship between customer requirement i and engineering design requirement j,
i = 1, 2, . . . , m, j = 1, 2, . . . , n,
Rik quantified relationship between customer requirement i and engineering design requirement k,
i = 1, 2, . . . , m, k = 1, 2, . . . , n,
rkj quantified relationship between design requirements, k,j = 1, 2, . . . , n.

The conventional method to quantify the relationships is accomplished using a 1-3-9 or 1-5-9 scale to
denote weak, medium, and strong relationships between CRs and DRs (Fung et al., 2002). However, in
practice the relationships are usually vague and imprecise, and can be described in linguistic terms. In this
study, the relationships are represented as linguistic terms, and fuzzy set theorems are employed to repre-
sent the vagueness of the relationship. Three fuzzy numbers are denoted as R e0 , R
e ik , and ~c , which corre-
ij kj
spond to R0ij , Rik and rkj, respectively, and can be defined as follows:
Pn e
e 0 R ik ~ckj
R ij ¼ Pn k¼1 Pn e , ð2Þ
j¼1 k¼1 R ik ~ckj

r jn

Engineering Design
Degree of
Requirements
Importance
DR1 DR j DR n
CR1 k1

Customer
CRi ki R ij
Wants

CRm k m

Fig. 1. QFD relationship matrix.


L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 233

e ik ¼ fðRik ,l
R ðRik ÞÞ j Rik 2 Rik g 8i, k,
e
R ik
ð3Þ
~ckj ¼ fðrkj ,l~ckj ðrkj ÞÞ j rkj 2 ckj g 8k, j,

where l ðRik Þ and l~ckj ðrkj Þ denote the associated membership functions. The above formulation is not easy
e
R ik
to solve, since it contains the multiplication and addition of two fuzzy numbers in the numerator and
denominator. For dealing with this, we first use the a-cut approach to represent R e ik , and ~c as the several
kj
crisp interval values under different a levels, which can be expressed in the following forms:
 
L U
ðRik Þa ¼ minfRik 2 Rik j l ðRik Þ P ag, maxfRik 2 Rik j l ðRik Þ P ag ¼ ½ðRik Þa ,ðRik Þa ,
Rik e
R ik Rik e
R ik

 
L U
ðckj Þa ¼ minfrkj 2 ckj j l~ckj ðrkj Þ P ag, maxfrkj 2 kkj j l~ckj ðrkj Þ P ag ¼ ½ðckj Þa ,ðckj Þa :
rkj rkj

L U L U
The above crisp interval values, ½ðRik Þa ,ðRik Þa  and ½ðckj Þa ,ðckj Þa , can be considered as the corresponding
ranges of R e ik and ~c , respectively, under a confidence degree. Based on ZadehÕs extension principle (Zadeh,
kj
1978), the membership function of fuzzy normalized relationship, R e 0 , can be defined as
ij
(  P )
 n
0  0 k¼1 Rik rkj
l 0 ðRij Þ ¼ sup min l ðRik Þ,l~ckj ðrkj Þ, 8k,jRij ¼ Pn Pn : ð4Þ
e
R ij R;r e
R ik  j¼1 k¼1 Rik rkj

To find the membership function of l e0


0 , it suffices to find the lower and upper bonds of the a-cuts of R ,
e
R ij ij

which can be solved as (Kao and Liu, 2000)


Pn
R r
L
Pn ik kj
ðR0ij Þa ¼ min R0ij ¼ Pn k¼1
j¼1 k¼1 Rik rkj

s:t: ð5aÞ
L U
ðRik Þa 6 Rik 6 ðRik Þa 8k,
L U
ðckj Þa 6 rkj 6 ðckj Þa 8k, j,
Pn
k¼1 Rik rkj
ðR0ij ÞU
a ¼ max R 0
ij ¼ P n Pn
j¼1 k¼1 Rik rkj

s:t: ð5bÞ
L U
ðRik Þa 6 Rik 6 ðRik Þa 8k,
L U
ðckj Þa 6 rkj 6 ðckj Þa 8k, j:
e 0 , ðR
Mathematically, the lower and upper bounds of a-cuts of R e 0 ÞL and ð R
e 0 ÞU , can be reformulated as
ij ij a ij a
Pn L L
L k¼1 ðRik Þa ðckj Þa
ðR0ij Þa ¼ Pn P n U U
,
j¼1 k¼1 ðRik Þa ðckj Þa
Pn U U
ð6Þ
0 U k¼1 ðRik Þa ðckj Þa
ðRij Þa ¼ Pn Pn L L
:
j¼1 k¼1 ðRik Þa ðckj Þa

Solving Eq. (6) gives us a set of solutions with the possible extreme ranges at each a-cut. For improving the
L
outcomes, Chen and Weng (2003) have proposed new formulations to find more accurate ranges, mðR0ij Þa
0 U
and mðRij Þa , which are formulated as follows:
234 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

Pn L L
k¼1 ðRik Þa ðckj Þa
mðR0ij ÞLa ¼ Pn Pn U U Pn L L
,
l¼1 k¼1 ðRik Þa ðckl Þa þ k¼1 ðRik Þa ðckj Þa
l6¼j
Pn U U
ð7Þ
k¼1 ðRik Þa ðckj Þa
mðR0ij ÞU
a ¼ Pn Pn L L Pn U U
:
l¼1 k¼1 ðRik Þa ðckl Þa þ k¼1 ðRik Þa ðckj Þa
l6¼j

Appendix A lists the derivative processes of the new formulations. The ranges produced by Eq. (7) are obvi-
ously smaller than those by Eq. (6), such that more accurate representations can be obtained. Comparisons
between the above two equations are made in the illustrated example in Section 4. The support of the fuzzy
number is the subset of the universe of discourse [0, 1]. The new a-cuts of fuzzy normalized relationship are
applied to formulate the fuzzy technical importance ratings of DRs.
The fuzzy technical importance rating of design requirement j, W ej , is determined by the fuzzy weighted
average of each fuzzy weight of customer requirement and the jth fuzzy normalized relationship, shown as
e j is used to measure the overall impact of the jth design requirement on customer
Eq. (8). The rating of W
satisfaction. In other words, the fuzzy set of We j represents the overall customer satisfaction that can be
achieved by the jth DR.
Pm e0 ei
ej¼ i¼1 mð R ij Þ  K
W Pm , ð8Þ
e
i¼1 K i

where

ei
K fuzzy weight of customer requirement, i, i = 1, 2, . . . , m,
ej
W fuzzy technical importance rating for engineering design requirement, j, j = 1, 2, . . . , n.

The above formulation is also difficult to solve, since several fuzzy numbers are included. Similarly, the
calculations can be performed via a-cuts of fuzzy numbers. At a specific possibility level a, the lower and
upper bounds of the a-cuts of l Wej can be obtained using Eq. (9). Vanegas and Labib (2001) have also pro-
posed a similar formulation.
Pm 0 L
L i¼1 mðRij Þa  k i
ðW j Þa ¼ min Pm
i¼1 k i
ð9aÞ
s:t:
L U
ðK i Þa 6 k i 6 ðK i Þa , i ¼ 1,2, . . . ,m,
Pm 0 U
U i¼1 mðRij Þa  ki
ðW j Þa ¼ max Pm
i¼1 k i
ð9bÞ
s:t:
L U
ðK i Þa 6 k i 6 ðK i Þa , i ¼ 1,2, . . . ,m:

3. Formulations

In addition to customer satisfaction emphasized by the conventional QFD, some authors also high-
lighted the need to conduct the cost and/or technical difficulty considerations in the QFD planning effort
L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 235

(Fung et al., 2002; King, 1987; Park and Kim, 1998; Trappey et al., 1996; Wasserman, 1993; Wang, 1999;
Zhou, 1998). Similarly, the cost and technical difficulty are also represented in fuzzy terms in order to coin-
cide with the fuzzy nature in the design stage. In this paper, a fuzzy goal programming model is formulated
to assist the design team in selecting a mix of DRs to produce the maximal sum of satisfaction degrees of all
goals. Three goals are considered for maximizing customer satisfaction, minimizing cost, and minimizing
technical difficulty. First, some notations are specified as follows. Let xj be the fulfillment level of engineer-
ing design requirement, j, j = 1, 2, . . . , n. If xj = 100%, it denotes complete fulfillment of the objective targets
for the jth DR; W e j is the descriptive of the overall impact of the fulfillment of the jth DR on customer sat-
isfaction; Ce j represents the fuzzy cost required to the jth DR; and Te j denotes the fuzzy technical difficulty to
the jth DR. Furthermore, considering business competition, a company usually desires some fulfillment
levels (xj) of engineering design requirement better than its competitors (lj), i.e., xj P lj. The model is then
formulated as follows:
X n X n Xn
max e j xj , min
W Ce j xj , min Te j xj
j¼1 j¼1 j¼1

s:t: ð10Þ
xj P lj , j ¼ 1,2, . . . ,n,
0 6 xj , lj 6 1:

3.1. Aspiration levels of goal

According to the above formulation, determining the goal values precisely is difficult for the design
team, since the customersÕ satisfaction, cost, and technical difficulty are not easy to measure exactly.
These goals usually conflict with each other. For dealing with this, the design team first determines
the aspiration levels for each goal, and then finds a set of solutions to achieve the maximum satisfaction
degree of all goals in total. Let Gmins and Gmaxs represent the lower and upper bounds of the aspiration of
Gs as the goal level of customer satisfaction. The design team would be completely dissatisfied with a
design (x) at which Gs ðxÞ 6 Gmin max
s , while the design would be completely satisfied if Gs ðxÞ P Gs , where
x denotes the variable vector. While, if Gp represents the cost or technical difficulty with the smaller-the-
better characteristic, the design team would be completely satisfied with a design (x) at which
Gp ðxÞ 6 Gmin max
p , but it would be completely dissatisfied if Gp ðxÞ P Gp . Here Gs(x)/Gp(x) is the achieve-
ment degree of the sth/pth goal at x. The degree of satisfaction can be formulated linearly as (Zimmer-
mann, 1978, 1983)
8
>
> 0 if Gs ðxÞ 6 Gmin
s ,
< min
ðxÞGs
ls ðxÞ ¼ GGsmax if Gmin 6 Gs ðxÞ 6 Gmax ,
>
> s Gmin
s
s s
:
1 if Gs ðxÞ P Gmaxs

or
8
>
> 1 if Gp ðxÞ 6 Gmin
p ,
>
<
Gmax
p Gp ðxÞ
lp ðxÞ ¼ Gmax min if Gmin
p 6 Gp ðxÞ 6 Gmax
p ,
>
> p Gp
>
:
0 if Gp ðxÞ P Gmax
p :

Based on the above formulations, the lower and upper bounds of the aspiration level of each goal, i.e., Gmin
and Gmax, should be predetermined. However, determining the two bounds is not easy, because Eq. (10)
236 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

contains more than one goal and each goal has fuzzy coefficients. For doing this, a three-step solution pro-
cedure is developed as follows:

Step 1: Set each fuzzy coefficient as the upper (lower) bound at the a-level = 0 for the goal having the max-
imum (minimum) target, such as customerÕs satisfaction (cost). This obtains the largest (smallest)
crisp value of each coefficient.
Step 2: Solve the problem for each single goal, i.e., remove the other goals, under the system constraints.
The optimal solution set and the goal value can be obtained in this step, which are supported by all
the resources. The determined goal values are considered as the upper (lower) bound of the aspi-
ration level.
Step 3: Place the solution set of one goal into other goals to determine their goal values. Find the lower
(upper) bounds of each goal with the maximum (minimum) target using the smallest (largest) goal
values determined in this step.

Once the two bounds of aspiration level are obtained, this study uses an additive model to sum up the
goals for finding the maximal overall satisfaction degree (Tiwari et al., 1987).

3.2. Preemptive priority structure for goals

A design team usually has a preemptive priority in achieving goals. For example, increasing customer
satisfaction may be the main purpose in the QFD process. However, cost expenditure and technical diffi-
culty are also taken into account in the design stage. Let G1, G2 and G3 be the goals of customer satisfac-
tion, cost expenditure, and technical difficulty, respectively. Suppose that G1 and G2 are considered more
important than G3 such that two priority levels are recommended in the QFD process. For simplifying
the computational efforts, a recently proposed model has been adopted in this study (Chen and Tsai,
2001). To illustrate, the three fuzzy goals are ranked as

Priority level 1: G1 and G2.


Priority level 2: G3.

Denoted as membership functions, the preemptive priority structure is represented as

l1 ðxÞ P l3 ðxÞ,
l2 ðxÞ P l3 ðxÞ:

3.3. Fuzzy coefficients in FGP

Based on the three fuzzy goals and their preemptive priority structure, the overall model can be formu-
lated as follows:

X
3
e ¼ max
Z ~h ðxÞ
l
h¼1

s:t:
Pn
e min
j¼1 W j xj  G1
~1 ðxÞ ¼
l ,
Gmax
1  Gmin1
L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 237

Pn
Gmax  e
2 j¼1 C j xj
~2 ðxÞ ¼
l ,
G2  Gmin
max
2
Pn
Gmax  e
3 j¼1 T j xj
~3 ðxÞ ¼
l ,
Gmax
3  Gmin
3

~1 ðxÞ  l
l ~3 ðxÞ,
~2 ðxÞ  l
l ~3 ðxÞ, ð11Þ
~i ðxÞ 6 1,
l
~i ðxÞ P 0, i ¼ 1,2,3,
l
xj P lj , j ¼ 1, . . . ,n,
xj , lj 6 1,
xj , lj P 0,

where ‘‘’’ means that a fuzzy number ‘‘dominates’’ the other fuzzy number, since partial ordering usually
exists between fuzzy numbers.
Note that the coefficients of the above formulation are fuzzy, such that the solutions are difficult to ob-
tain. For solving this problem, we transform the model with fuzzy coefficients to a family of conventional
crisp mathematical programming models by applying the a-cut approach and ZadehÕs extension principle
(Zadeh, 1978). The membership function of the objective value can be defined as
( )
X
3
leZ ðzÞ ¼ sup min l Wej ðwj Þ,le ðc Þ,leT j ðtj Þ, 8j j z ¼
Cj j
lh ðxÞ , ð12Þ
w;c;t h¼1

where w, c, and t are the element values of fuzzy coefficients, and z is the objective value. Applying Eq. (12),
the membership function of leZ can be determined based on membership degrees of all fuzzy coefficients.
Similar to Eq. (4), we separate Model (11) into two crisp sub-problems to find the lower and upper bounds
of leZ by specifying the a-cuts of all fuzzy coefficients as follows:

ðZÞLa ¼ min Z
s:t:
L U
ðW j Þa 6 wj 6 ðW j Þa , 8j, ð13aÞ
ðC j ÞLa 6 cj 6 ðC j ÞU
a , 8j,
ðT j ÞLa 6 tj 6 ðT j ÞU
a , 8j;
U
ðZÞa ¼ max Z
s:t:
L U
ðW j Þa 6 wj 6 ðW j Þa , 8j, ð13bÞ
L U
ðC j Þa 6 cj 6 ðC j Þa , 8j,
L U
ðT j Þa 6 tj 6 ðT j Þa , 8j:
238 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

And therefore, the full form is formulated as follows:


X
3
ðZÞLa ¼ min max lh ðxÞ
ðW j ÞL U
a 6wj 6ðW j Þa ;8j h¼1
ðC j ÞL
a 6cj 6ðC Þ U
j a ;8j
ðT j ÞL U
a 6tj 6ðT j Þa ;8j

s:t:
Pn min
j¼1 wj xj  G1
l1 ðxÞ ¼ ,
Gmax
1  Gmin
1
P
Gmax
2  nj¼1 cj xj
l2 ðxÞ ¼ ,
Gmax
2  Gmin
2
P
Gmax  nj¼1 tj xj ð14aÞ
3
l3 ðxÞ ¼ ,
Gmax3  Gmin
3

l1 ðxÞ P l3 ðxÞ,
l2 ðxÞ P l3 ðxÞ,
li ðxÞ 6 1,
li ðxÞ P 0, i ¼ 1,2,3,
x j P lj , j ¼ 1, . . . ,n,
xj , lj 6 1,
xj , lj P 0;

X
3
ðZÞU
a ¼ max max lh ðxÞ
ðW j ÞL U
a 6wj 6ðW j Þa ;8j h¼1
ðC j ÞL
a 6c j 6ðC Þ U
j a ;8j
ðT j ÞL U
a 6tj 6ðT j Þa ;8j

s:t:
Pn min
j¼1 wj xj  G1
l1 ðxÞ ¼ ,
Gmax
1  Gmin
1
Pn
Gmax
2  j¼1 cj xj
l2 ðxÞ ¼ ,
Gmax
2  Gmin
2
Pn ð14bÞ
Gmax
3  j¼1 tj xj
l3 ðxÞ ¼ ,
Gmax
3  Gmin
3

l1 ðxÞ P l3 ðxÞ,
l2 ðxÞ P l3 ðxÞ,
li ðxÞ 6 1,
li ðxÞ P 0, i ¼ 1,2,3,
x j P lj , j ¼ 1, . . . ,n,
xj , lj 6 1,
xj , lj P 0:
L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 239

When the importance ratings, costs, and technical difficulties vary, the minimum of Z occurs when the coef-
ficients of importance ratings are set to their lower bounds and the coefficients of costs and technical dif-
ficulties are set to their upper bounds; otherwise, the maximum of Z occurs. Therefore, the mathematical
formulations in Eq. (14) can be simplified to conventional linear programming models, shown as Eq. (15).
L U
The membership function leZ can then be constructed from ½ðZÞa ,ðZÞa  at different a levels.

L
X
3
ðZÞa ¼ max lh ðxÞ
h¼1

s:t:
Pn L min
j¼1 ðW j Þa xj  G1
l1 ðxÞ ¼ ,
Gmax
1  Gmin1
Pn U
Gmax
2  j¼1 ðC j Þa x j
l2 ðxÞ ¼ ,
Gmax
2  Gmin
2
Pn U
Gmax
3  j¼1 ðT j Þa xj ð15aÞ
l3 ðxÞ ¼ ,
Gmax
3  Gmin
3
l1 ðxÞ P l3 ðxÞ,
l2 ðxÞ P l3 ðxÞ,
li ðxÞ 6 1,
li ðxÞ P 0, i ¼ 1,2,3,
xj P lj , j ¼ 1, . . . ,n,
xj , lj 6 1,
xj , lj P 0;

U
X
3
ðZÞa ¼ max lh ðxÞ
h¼1

s:t:
Pn U min
j¼1 ðW j Þa xj  G1
l1 ðxÞ ¼ ,
Gmax
1  Gmin1
Pn L
Gmax
2  j¼1 ðC j Þa xj
l2 ðxÞ ¼ ,
Gmax
2  Gmin
2
P
Gmax
3  nj¼1 ðT j ÞLa xj ð15bÞ
l3 ðxÞ ¼ ,
Gmax
3  Gmin
3
l1 ðxÞ P l3 ðxÞ,
l2 ðxÞ P l3 ðxÞ,
li ðxÞ 6 1,
li ðxÞ P 0, i ¼ 1,2,3,
xj P lj , j ¼ 1, . . . ,n,
xj , lj 6 1,
xj , lj P 0:
240 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

4. Illustrations

4.1. A writing instrument example

In order to demonstrate the feasibility of the proposed models, a simple example of a writing instrument
from a related work is adopted in this section (Wasserman, 1993). The design includes four customer
requirements (CRs) and five design requirements (DRs). Fig. 2 illustrates the HOQ. In the figure, four
CRs are ‘‘easy to hold’’ (CR1), ‘‘does not smear’’ (CR2), ‘‘point lasts’’ (CR3), and ‘‘does not roll’’
(CR4), while the important engineering design requirements contain ‘‘length of pencil’’ (DR1), ‘‘time be-
tween sharpening’’ (DR2), ‘‘least dust generated’’ (DR3), ‘‘hexagonal’’ (DR4), and ‘‘minimal erasure res-
idue’’ (DR5). In addition, cost and technical difficulty of DRs are also incorporated in the planning
processes.
e ik ) between CRs and DRs, those (~c ) among DRs, and the relative
Firstly, the fuzzy relationships ( R kj
importance ( Ke i ) of the four CRs must be determined to derive the fuzzy importance ratings of the five
DRs. Owing to the imprecise design information available in the early design stage, it is difficult to assess
the relationship of the specified design variables in design planning accurately. Therefore, linguistic terms
are used to describe the strengths of relationship among DRs and between CRs and DRs, the relative
importance of the four CRs, and the estimated cost and technical difficulty of each DR. In this paper, four
groups of linguistic terms are defined in Table 1 for different descriptions. Each group contains seven lin-
guistic terms. For example, the descriptions of relationship strength are ‘‘weakest’’, ‘‘weak’’, ‘‘fairly weak’’,

Strongest

Strong

Fairly strong

Engineering Design Medium


Relative Requirements
Import- Fairly weak
ance DR1 DR2 DR3 DR4 DR5
CR1 Weak
Customer CR2
Wants Weakest
CR3
CR4
Estimated Cost
Technical Difficulty

Fig. 2. QFD matrix for a writing instrument.

Table 1
The linguistic scales used by design team
Linguistic scale for Linguistic scale for Linguistic scale for Linguistic scale for
relationship strengths relative importance estimated cost technical difficulty
Weakest Very unimportant Very high Very difficult
Weak Unimportant High Difficult
Fairly weak Fairly unimportant Fairly high Fairly difficult
Medium Medium Medium Medium
Fairly strong Fairly important Fairly low Fairly easy
Strong Important Low Easy
Strongest Very important Very low Very easy
L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 241

‘‘medium’’, ‘‘fairly strong’’, ‘‘strong’’, and ‘‘strongest’’. For the subsequent fuzzy operations, these linguis-
tic terms should be translated to fuzzy numbers. Seven trapezoidal fuzzy numbers are used to represent
each group of linguistic terms according to the conversion scale (Chen et al., 1992). Their definitions are
(0, 0, 0.1, 0.2), (0.1, 0.2, 0.2, 0.3), (0.2, 0.3, 0.4, 0.5), (0.4, 0.5, 0.5, 0.6), (0.5, 0.6, 0.7, 0.8), (0.7, 0.8, 0.8, 0.9), and
(0.8, 0.9, 1.0, 1.0), respectively, as shown in Fig. 3.
Using the linguistic terms in Table 1 to represent the various relationships in the QFD matrix, as shown
in Fig. 2, the fuzzy normalized relationship, mð R e 0 Þ, can be calculated using Eq. (7). For obtaining mð R e 0 Þ,
ij ij
the upper and lower bounds of the a-cuts of R e ik and those of ~c should be determined beforehand, based
kj
on their membership functions. The membership function of a trapezoidal fuzzy number is defined by linear
functions. As an illustration, suppose that R e ik is assessed as ‘‘fairly strong’’ ( e S ), and the membership func-
tion of the fuzzy number e S ¼ ð0:5,0:6,0:7,0:8Þ can be expressed as
8
< ðRik  0:5Þ=ð0:6  0:5Þ, 0:5 6 Rik 6 0:6,
>
leS ðRik Þ ¼ 1, 0:6 6 Rik 6 0:7,
>
:
ð0:8  Rik Þ=ð0:8  0:7Þ, 0:7 6 Rik 6 0:8:
Then, the a-cut of the above membership function is
h i
L U
ðRik Þa ,ðRik Þa ¼ ½0:5 þ 0:1a,0:8  0:1a:

Once the a-cuts of all relationships are determined, they are placed into the equations for obtaining the
upper and lower bounds of a-cuts of fuzzy normalization relationships. As mentioned before, the ranges
produced by applying Eq. (7)0 are smaller than those 0when Eq. (6) is used. For comparison purposes, four
membership functions, mð R e 0 Þ, R
e Þ, mð R e 0 , and R
e , are shown in Fig. 4. This justifies the derived for-
14 44 14 44
mulations in Appendix A.
The fuzzy technical importance rating of the jth engineering design requirement, W e j , can be obtained
using Eq. (9) to determine the priority of each design requirement. The membership functions, W e 1 to
e e e e
W 5 , are shown in Fig. 5. Both W 3 and W 5 rank the highest, with a range of 16.2–40.8%, while W 1 ranks
the lowest, with a range of 3.1–16.2%. These ratings are then used to formulate the customer satisfaction
function.
The objective of a QFD planning is not only to maximize customer satisfaction (G1), but also to mini-
mize cost (G2) and technical difficulty (G3), subject to other organizational constraints, such as the fulfill-
ment level of engineering design requirements. These goals have the preemptive priority structure that is the
same as that described in Section 3.2. Cost and technical difficulty are evaluated and illustrated in Fig. 2,

Fairly Fairly
Weakest weak strong Strongest
membership Weak Medium Strong
degree 1

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
strength scale

Fig. 3. The membership functions of linguistic terms for relationships.


242 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

Fig. 4. The membership functions of normalized relationship degrees.

α ~
W1
~
W2
~ ~
W3= W5
1

0.8 ~
W4
0.6
0.4

0.2
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
technical importance degree

Fig. 5. The membership functions of technical importance ratings.

using the seven linguistic scales of Table 1. For formulating fuzzy goal programming models, the range of
aspiration levels of Gh, i.e., ½Gmin max
h ,Gh , should be specified beforehand. Suppose that the minimum fulfill-
ment levels of the five engineering design requirements, x1, . . . , x5, are required as 0.2, 0.5, 0.3, 0.7, and 0.5,
respectively. Following the solution procedure in Section 3.1, the model can be formulated as follows:
X U
max ðW i Þa¼0 xi ¼ 0:62x1 þ 0:311x2 þ 0:408x3 þ 0:325x4 þ 0:408x5
X
min ðC i ÞLa¼0 xi ¼ 0:7x1 þ 0:5x2 þ 0:4x3 þ 0:1x4 þ 0:2x5
X L
min ðT i Þa¼0 xi ¼ 0:1x1 þ 0:4x2 þ 0:7x3 þ 0:1x4 þ 0:4x5
s:t:
x1 P 0:2, ð16Þ
x2 P 0:5,
x3 P 0:3,
x4 P 0:7,
x5 P 0:5,
xi 6 1, i ¼ 1, . . . ,5:
The ranges of aspiration levels of the three goals are determined as [0.245, 1.615], [0.68, 3.1], and [0.7, 2.17],
respectively. And the full model, Eq. (15), can be constructed, subject to the preemptive priority structure
and the required fulfillment levels of DRs.
L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 243

Table 2
e 2 and X
The ranges for the three fuzzy goals and the fulfillment levels of X e 3 at 11 different possibility a values
a ½ðl1 ÞLa ; ðl1 ÞU
a  ½ðl2 ÞLa ; ðl2 ÞU
a  ½ðl3 ÞLa ; ðl3 ÞU
a  ½ðx2 ÞLa ; ðx2 ÞU
a  ½ðx3 ÞLa ; ðx3 ÞU
a 

0.0 0.18 0.72 0.37 0.87 0.18 0.72 0.5 0.79 0.76 1
0.1 0.20 0.70 0.38 0.85 0.20 0.70 0.5 0.79 0.78 1
0.2 0.23 0.67 0.40 0.83 0.23 0.67 0.5 0.78 0.81 1
0.3 0.25 0.65 0.42 0.81 0.25 0.65 0.5 0.78 0.83 1
0.4 0.27 0.62 0.44 0.80 0.27 0.62 0.5 0.77 0.85 1
0.5 0.29 0.60 0.46 0.78 0.29 0.60 0.5 0.75 0.87 1
0.6 0.31 0.57 0.48 0.76 0.31 0.57 0.5 0.74 0.89 1
0.7 0.34 0.55 0.50 0.74 0.34 0.55 0.5 0.73 0.90 1
0.8 0.36 0.52 0.52 0.72 0.36 0.52 0.5 0.71 0.92 1
0.9 0.39 0.50 0.54 0.70 0.39 0.50 0.5 0.67 0.94 1
1.0 0.41 0.47 0.56 0.69 0.41 0.47 0.5 0.67 0.95 1

α µ~1 = µ~3 µ~2


1

0.8

0.6

0.4

0.2

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
satisfaction degree

Fig. 6. The membership functions of goal values.

Solving the model (15) of the example using 11 different a-cuts, i.e., a = 0, 0.1, 0.2, . . . , 1, the satisfaction
degree for each goal and their sum, as well as the fulfillment levels of all DRs, can be acquired at each a
level. Table 2 lists the ranges of three fuzzy goals and those of the fulfillment levels of X e 2 and Xe 3 at 11 dif-
ferent possibility levels. Based on the ranges, Fig. 6 depicts the membership functions of satisfaction degree
of the three goals. The satisfaction degree of G2 is greater than those of G1 and G3 for which G2 locates in
the interval [0.37, 0.87] and G1 as well as G3 in [0.18, 0.72], although G1 has the same priority as G2. Obvi-
ously, achieving the cost objective (G2) is easier than that of customer satisfaction (G1) in this example.
In Fig. 7, x2 and x3 are determined as being fuzzy, X e 2 and Xe 3 , respectively, while x1, x4, and x5 are crisp
with the fulfillment level of 20%, 100%, 100%, respectively. As described before, the decision variable
xj = 100% denotes complete fulfillment of the jth DR. This means that the DRs, x4 and x5, should have
the best quality level in order to achieve the total satisfaction degree. Particularly, x1 and x2 are smaller
than the others in the example, due to the low technical importance ratings and high estimated costs, if
referring to Figs. 2 and 5.

4.2. Discussion

Echoing a common belief that imprecise input data generally produce imprecise output in a decision-
making problem, our example illustrates that the fulfillment levels of some DRs are fuzzy due to the use
of imprecise information, and their ranges at different possibility levels can be obtained by applying the
244 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

X2 X3
1

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
fulfillment level

Fig. 7. The membership functions of decision variables x2 and x3, while x1 = 0.2 and x4 = x5 = 1.

proposed approaches. Consider, for instance, the second DR in the example, i.e., X e 2 in Fig. 7, with the
ranges at the possibility levels a = 1.0 and a = 0 being [0.5, 0.67] and [0.5, 0.79], respectively. In the fuzzy
sense, it is definitely possible that the fulfillment level of x2 is in [0.5, 0.67]; this fulfillment level will never
exceed 0.79 or fall below 0.5. With the fuzzy sense, the possibility level can be interpreted as the confidence
degree (Bondia and Picó, 2003; Chang and Lee, 1996; Mon et al., 1995; Wu, 2003). A designer can adopt a
fulfillment level of one DR from the range produced under an acceptable confidence degree (say 0.7), such
that only one a-cut is needed for the proposed approach.
Alternatively, if a designer desires to utilize more information from QFD processes, the fuzzy fulfillment
level can be defuzzified into a real number in [0, 1], which can then be considered to be the action (fulfill-
ment level) to be taken by the designer. Several defuzzification methods have been developed in the fuzzy
control area, such as the centroid method, the center of maxima method, and the mean of maxima method
(Klir and Yuan, 1995). Among them, the centroid method is the commonly used, and hence has been
adopted in this paper. The centroid method for defuzzifying a fuzzy number X e i is formulated as (Klir
and Yuan, 1995).
Pm ðiÞ ðiÞ
k¼1 le ðx Þxk
Xi k
^xi ¼ Pm ðiÞ
, ð17Þ
k¼1 le ðx Þ
Xi k

where X e i is defined on a finite universal set, and l ðxk Þ is the membership degree (possibility level) of the
ðiÞ
e
Xi
ðiÞ
kth element (fulfillment level in this paper) xk in X e i . Consider again X
e 2 in Fig. 7. Applying the interpola-
ð2Þ
tion method, based on the 11 a-cuts (possibility levels), renders 15 fulfillment levels xk , k = 1, . . . , 15, for
Xe 2 ; these 15 fulfillment levels and the corresponding possibility levels give the defuzzified value ^x2 ¼ 0:62 in
Eq. (17).
In this paper, the a-cut approach is employed in order to determine the fulfillment level of each DR in an
imprecise environment. The proposed approach mainly consists of two sequent models, i.e., Models (9) and
(15). The a value and the resulting range of each fuzzy technical importance rating in Model (9) are taken as
the input of Model (15). Obviously, if a designer requires more information to decide the fulfillment level of
DRs, more a-cuts are needed. As listed in Table 2, different a values lead to different ranges of satisfaction
degrees of the goals and also those of the fulfillment levels of the DRs. The membership functions for fuzzy
goals and fuzzy fulfillment levels are constructed by piecewise linear segments based on different a values
and the resulting ranges in Table 2, as shown in Figs. 6 and 7. Therefore, the number of a-cuts is critical
to the proposed approach.
L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 245

From Fig. 6, the variability of membership function of each goal is not significant if different numbers of
a values are adopted, since their piecewise linear segments connect smoothly. However, as an example, the
membership function of X e 2 in Fig. 7 will somewhat change if the number of a values is small. For illustra-
tions, we can construct the membership functions of X e 2 by using three different numbers of a-cuts, i.e., 2
(a = 0, 1), 3 (a = 0, 0.5, 1), and 6 (a = 0, 0.2, 0.4, 0.6, 0.8, 1), in Fig. 8. Although these membership functions
differ from the membership function of X e 2 in Fig. 7, the actual differences are not significant, especially
those under 3 and 6 a-cuts. Similarly, substituting 15 elements from the membership functions in Fig.
8(a)–(c) into Eq. (17) gives, respectively, the defuzzified values 0.6126, 0.6176, and 0.6158, all of which
are close to the defuzzified value 0.62 in Fig. 7.

α ~
X2
1

0
0 0.2 0.4 0 .6 0.8 1
(a) fulfillment level

α ~
X2
1

0.5

0
0 0.2 0.4 0.6 0.8 1
(b) fulfillment level
α ~
X2
1

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
(c) fulfillment level
e 2 based on three different numbers of a-cuts. (a) The membership function based on two a-cuts
Fig. 8. The membership functions of X
(a = 0, 1). (b) The membership function based on three a-cuts (a = 0, 0.5, 1). (c) The membership function based on six a-cuts
(a = 0, 0.2, 0.4, 0.6, 0.8, 1).
246 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

In summary, for application purposes, a designer can perform one a-cut model-solving process with an
acceptable degree of confidence. If more information is required, more a-cuts are usually needed, and the
interpolation method as well as the defuzzification method is used. With the fuzzy nature of design in the
early stage of product planning, a small number of a-cuts, say four (a = 0, 0.33, 0.67, 1) or six
(a = 0, 0.2, 0.4, 0.6, 0.8, 1), usually can suffice for the designersÕ needs, such as the illustrated example.

5. Conclusions

Ambiguity exists in the QFD planning, since the assessments are imprecise in the relationships between
CRs and DRs as well as among DRs, the relative importance between CRs, cost, and technical difficulty.
Due to the impreciseness in a QFD process, fuzzy approaches are applied in this paper to determine the
required fulfillment levels of DRs for achieving the maximum satisfaction degree of several goals in total
in the product design stage. Three goals are considered: maximizing customer satisfaction, minimizing cost,
and minimizing technical difficulty with respect to each DR. The coefficients in the three goal formulations
are allowed to be fuzzy, and the satisfaction of each goal is also fuzzy. In addition, the minimum fulfillment
degree of each DR can be delimited, and the preemptive priority structure for the goals can be required. In
general, crisp values can be considered as special conditions of fuzzy numbers. Therefore, through the
applications of fuzzy goal programming models on the QFD processes, the formulations in this study
can allow a QFD planning team to make various kinds of assessments under an uncertain environment.
The applicability of our formulations is demonstrated by a simple example from the existing study. Only
a few a-cuts are required to construct the membership functions of fuzzy goals and those of fuzzy fulfillment
levels of DRs in the example. The resulting ranges of satisfaction degree of each goal and the possible
ranges of the fulfillment levels of DRs can provide a QFD team with more useful information. For appli-
cations, a designer can perform one a-cut model-solving process with an acceptable degree of confidence.
Alternatively, performing more a-cuts, the fuzzy fulfillment levels of DRs can be defuzzified into real num-
bers, which can be done by designers.

Appendix A

The new formulations for the upper and lower bounds of fuzzy normalized relationship can be derived as
follows:
Pn Pn
0 k¼1 Rik rkj k¼1 Rik r kj
Rij ¼ Pn Pn ¼ Pn Pn Pn , ðA:1Þ
j¼1 R
k¼1 ik kjr l¼1 k¼1 R ik rkl þ k¼1 Rik r kj
l6¼j

where
L U
0 6 ðRik Þa 6 Rik 6 ðRik Þa 6 1, 8k, i ¼ 1, . . . ,m,

0 6 ðckj ÞLa 6 rkj 6 ðckj ÞU


a 6 1, 8k,j:
Pn Pn Pn
Let / ¼ k¼1 Rik rkj and u ¼ l¼1 k¼1 Rik rkl .
l6¼j
Then Eq. (A.1) is expressed as
/
f ð/Þ ¼ :
uþ/
L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248 247

Since
u
f 0 ð/Þ ¼ P 0,
ðu þ /Þ2
f(/) is an increasing function and
X L L
X U U
ðRik Þa ðckj Þa 6 / 6 ðRik Þa ðckj Þa :
k k

Therefore,
Pn L L
k¼1 ðRik Þa ðckj Þa
min f ð/Þ ¼ Pn L L
u þ k¼1 ðRik Þa ðckj Þa
and
Pn U U
k¼1 ðRik Þa ðckj Þa
max f ð/Þ ¼ Pn :
u þ k¼1 ðRik ÞU U
a ðckj Þa

Furthermore,
Xn Xn
L L
n X
X n
U U
ðRik Þa ðckl Þa 6 u 6 ðRik Þa ðckl Þa ,
l¼1 k¼1 l¼1 k¼1
l6¼j l6¼j

such that the new lower and upper bounds of a-cuts of R e 0 , mðR0 ÞL and mðR0 ÞU , can be formulated as
ij ij a ij a
Pn L L
k¼1 ðRik Þa ðckj Þa
mðR0ij ÞLa ¼ min f ð/Þ ¼ Pn Pn U U Pn L L
,
l¼1 k¼1 ðRik Þa ðckl Þa þ k¼1 ðRik Þa ðckj Þa
l6¼j

Pn U U
U k¼1 ðRik Þa ðckj Þa
mðR0ij Þa ¼ max f ð/Þ ¼ Pn Pn L L P n U U
:
l¼1 k¼1 ðRik Þa ðckl Þa þ k¼1 ðRik Þa ðckj Þa
l6¼j

References

Bondia, J., Picó, J., 2003. Analysis of linear systems with fuzzy parametric uncertainty. Fuzzy Sets and Systems 135, 81–121.
Chang, P.-T., Lee, E.S., 1996. A generalized fuzzy weighted least-squares regression. Fuzzy Sets and Systems 82, 289–298.
Chen, L.-H., Tsai, F.C., 2001. Fuzzy goal programming with different importance and priorities. European Journal of Operational
Research 133, 548–556.
Chen, L.-H., Weng, M.C., 2003. A fuzzy model for exploiting quality function deployment. Mathematical and Computer Modelling
38, 559–570.
Chen, S.-J., Hwang, C.-L., Hwang, F.P., 1992. Fuzzy Multiple Attribute Decision Making Methods and Applications. Springer-
Verlag.
Fung, R.Y.K., Popplewell, K., Xie, J., 1998. An intelligent hybrid system for customer requirements analysis and product attribute
targets determination. International Journal of Production Research 36 (1), 13–34.
Fung, R.Y.K., Tang, J., Tu, Y., Wang, D., 2002. Product design resources optimization using a non-linear fuzzy quality function
deployment model. International Journal of Production Research 40 (3), 585–599.
Kao, C., Liu, S.T., 2000. Fuzzy efficiency measures in data envelopment analysis. Fuzzy Sets and Systems 113, 427–437.
Kim, K.-J., Moskowitz, H., Dhingra, A., Evans, G., 2000. Fuzzy multicriteria models for quality function deployment. European
Journal of Operational Research 121, 504–518.
King, B., 1987. Better Designs in Half the Time: Implementing QFD in America, Goal/QPC. Methuen, MA.
Klir, G.J., Yuan, B., 1995. Fuzzy Sets and Fuzzy Logic—Theory and Applications. Prentice Hall, Englewood Cliffs, NJ.
248 L.-H. Chen, M.-C. Weng / European Journal of Operational Research 172 (2006) 230–248

Mon, D.-L., Cheng, C.-H., Lu, H.-C., 1995. Application of fuzzy distributions on project management. Fuzzy Sets and Systems 73,
227–234.
Moskowitz, H., Kim, K.-J., 1997. QFD optimizer: A novice friendly quality function deployment decision support system for
optimizing product designs. Computers and Industrial Engineering 32 (3), 641–655.
Park, T., Kim, K.-J., 1998. Determination of an optimal set design requirements using house of quality. Journal of Operations
Management 16, 569–581.
Shen, X.X., Tan, K.C., Xie, M., 2001. The implementation of quality function deployment based on linguistic data. Journal of
Intelligent Manufacturing 12, 65–75.
Tiwari, R.N., Dharmar, S., Rao, J.R., 1987. Fuzzy goal programming—an additive model. Fuzzy Sets and Systems 24, 27–34.
Trappey, C.V., Trappey, A.J.C., Hwang, S.-J., 1996. A computerized quality function deployment approach for retail services.
Computers and Industrial Engineering 30 (4), 611–622.
Vanegas, L.V., Labib, A.W., 2001. A fuzzy quality function deployment (FQFD) model for deriving optimum targets. International
Journal of Production Research 39 (1), 99–120.
Wang, J., 1999. Fuzzy outranking approach to prioritize design requirements in quality function deployment. International Journal of
Production Research 37 (4), 899–916.
Wasserman, G.S., 1993. On how to prioritize design requirements during the QFD planning process. IIE Transactions 25 (3), 59–65.
Wu, H.-C., 2003. Linear regression analysis for fuzzy input and output data using the extension principle. Computers and Mathematics
with Applications 45, 1849–1859.
Zadeh, L.A., 1978. Fuzzy sets as a basis for a theory of possibility. Fuzzy Sets and Systems 1, 3–28.
Zhou, M., 1998. Fuzzy logic and optimization models for implementing QFD. Computers and Industrial Engineering 35, 237–240.
Zimmermann, H.J., 1978. Fuzzy programming and linear programming with several objective functions. Fuzzy Sets and Systems 1,
45–55.
Zimmermann, H.J., 1983. Fuzzy mathematical programming. Fuzzy Sets and Systems 10 (4), 291–298.

You might also like