Mathematical and Computer Modelling 48 (2008) 581–593

www.elsevier.com/locate/mcm

A fuzzy nonlinear model for quality function deployment

considering Kano’s concept
Liang-Hsuan Chen ∗ , Wen-Chang Ko
Department of Industrial and Information Management, National Cheng Kung University, Tainan, Taiwan, ROC

Received 8 December 2005; received in revised form 28 May 2007; accepted 22 June 2007

Abstract

Quality function deployment (QFD) is a customer-driven approach for processing new product developments in order to maxi-
mize customer satisfaction. Each engineering design characteristic is maximized for product performance according to the level of
customer satisfaction. To cope with the vague nature of product development processes, fuzzy approaches are used to represent the
importance scores of customer requirements (CRs), the relationship between CRs and design requirements (DRs) and relationship
between the DRs themselves. Considering Kano’s category of design requirements, this paper extends Chen and Weng’s model
(2003), and presents a fuzzy nonlinear model to determine the performance level of each DR for maximizing customer satisfaction,
under the same group of constraints as that in Chen and Weng’s model. The results, from an illustrative example, indicate that the
total degree of customer satisfaction achieved by the proposed nonlinear model is greater than that by the original fuzzy model.
c 2007 Elsevier Ltd. All rights reserved.

Keywords: Mathematical models; Quality function deployment (QFD); Fuzzy set; Fuzzy nonlinear model; Kano’s concept

1. Introduction

Quality function deployment (QFD) is a customer-driven approach that allows the needs of the customer to be
communicated through the various stages of product planning, design, engineering and manufacturing into a final
product. It has been successfully introduced in many industries to improve design processes, customer satisfaction,
and to create a competitive advantage [1]. A QFD team collects and treats a set of customer requirements (CRs) for
new product development. A number of design requirements (DRs) that affect CRs are also identified and worked out
to maximize customer satisfaction. In practice, a QFD team is organized to determine the importance levels of DRs by
computing the relationships between CRs and DRs and relationship between the DRs themselves with the importance
score of each customer requirement [2,3].
The importance scores of CRs, the relationships between CRs and DRs and relationship between the DRs them-
selves are usually determined subjectively by ambiguous or vague judgments. However, they are usually treated as
crisp variables [4,5]. For example, in traditional practice, the importance score of each CR is set as a crisp value,
although linguistic terms seem more adequate for evaluating the CR’s importance. Furthermore, the degree to which
a DR affects a CR is expressed on a scale system such as 1-3-9, or 1-5-9, representing linguistic expressions such as

∗ Corresponding author. Fax: +886 6 236 2162.

E-mail address: [email protected] (L.-H. Chen).

c 2007 Elsevier Ltd. All rights reserved.

0895-7177/$ - see front matter
doi:10.1016/j.mcm.2007.06.029
582 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593

Fig. 1. The QFD relationship matrix (HOQ).

“weak”, “moderate”, and “strong”. However, design engineers usually do not have sufficient knowledge and informa-
tion about the influence of engineering responses on CRs, due to the lack of information or language hedge from the
customer [6]. These considerations have made the applications of fuzzy approaches significant in addressing diver-
sified and imprecise problems in the importance score of each CR and the relationships between CRs and DRs and
relationship between the DRs themselves. In applying fuzzy approaches, several studies [7–11] have attempted to deal
with the vague nature of linguistic terms by using fuzzy numbers to express the importance score of CRs. Chen and
Weng [6] proposed a fuzzy linear programming model to determine the optimal fulfilment degree of each DR at each
α-cut for achieving the optimal customer satisfaction. Wang [5] presented a fuzzy outranking approach to model the
imprecise preference relationships between DRs to achieve customer satisfaction and balanced product design. Tem-
poni et al. [12] developed a fuzzy logic-based extension to QFD for capturing vague requirements. Some models have
also been formulated for determining the levels of CRs or DRs based on fuzzy sets theory [8,13–16]. However, these
existing studies still suffer from drawbacks in approaches and methodologies, as mentioned by Chen and Weng [6].
In practice, effective design leads to the achievement of maximum customer satisfaction. DRs are able to achieve
better design effectiveness when their performance is highly correlated with customer satisfaction. However, the per-
formance of DRs is not necessarily proportionally consistent with customer satisfaction. Kano et al. [17] proposed
that quality performance of a product or service has three different relationships with customer satisfaction. Some re-
searchers have incorporated Kano’s idea into QFD for assigning weights to different CRs to explain how CRs impact
customer satisfaction in different ways [1,2,18–20]. Unlike the existing literature, in this study we apply Kano’s con-
cept to classify DRs into three categories based on their importance to customer satisfaction, since the performance of
various DRs achieve different levels of customer satisfaction. This approach is mainly used to extend the fuzzy linear
QFD model in Chen and Weng’s study [6], and a fuzzy nonlinear mathematical programming model is constructed to
increase customer satisfaction under the same conditions, such as budget limitations, market competition and technical
difficulty constraints. An example of a semiconductor packing case is given to illustrate the proposed model.
In the following section, the fuzzy linear model is introduced. In Section 3, a fuzzy nonlinear model is developed
to figure out the performance degrees of DRs that produce maximum customer satisfaction. The way in which DRs
are categorized into three classes based on the Kano’s concept is also described. An example is presented to illustrate
our approach in Section 4. Finally, the concluding remarks are provided in Section 5.

2. Fuzzy linear model

In practice, an important task in the QFD processes is to determine the importance levels (or scores) of DRs based
on the relationships between CRs and DRs and relationship between the DRs themselves, referring to the importance
of each CR. The achievement priority or level of DRs is determined based on their importance levels. The associated
relationships in QFD are represented in the matrix form as shown in Fig. 1, which is also called the House of Quality
 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593 583

(HOQ). In the figure, Ri j denotes the relation level in terms of score between the CRi and DR j , and r j J is the
correlation score between DR j and
P DR J . The notations ki and W j represent the importance score and rating for the
CRi and DR j , respectively, and i ki = 1.
Considering the correlations among DRs, Wasserman [21] proposed a normalization formula (1) to calculate the
normalized relationship value between CRs and DRs:
J
P
Ril rl j
l=1
Ri0 j = , (1)
J P
P J
Ril rl j
j=1 l=1

where Ri0 j is the normalized relationship value between CRi and DR j , i = 1, 2, . . . , I ; j = 1, 2, . . . , J , and
P 0
j Ri j = 1 for each i. In existing literature [1–3,6,21], the relationships are quantified through the use of a 1-3-
9 or 1-5-9 scale to denote low, medium and high relationship degrees between CRs and DRs and relationship between
the DRs themselves. In order to reflect the imprecise nature of the relationships, Chen and Weng [6] proposed a
fuzzified quantitative formulation as in the following equation:
J
P
R̃il ⊗ r̃l j
l=1
R̃i0 j = , (2)
J P
P J
R̃il ⊗ r̃l j
j=1 l=1

where R̃il and r̃l j are described by linguistic terms and defined as the fuzzy numbers of [0,1], and the symbol of ⊗
represents the fuzzy multiplication operator.
Actually, a fuzzy set can fully and uniquely be represented by its α-cuts [22]. The α-cut of the fuzzy set R̃il at the
α level, α ∈ [0, 1], can be denoted by its lower and upper bounds as [(Ril )αL , (Ril )U
α ], which is defined by

(Ril )αL = inf {x|µ R̃il (x) ≥ α}, (3a)

x∈[0,1]

and
(Ril )U
α = sup {x|µ R̃il (x) ≥ α}, (3b)
x∈[0,1]

where µ R̃il (x) is the membership degree of x belonging to R̃il . Based on α-cuts and the extension principle [22–24],
the membership function of the fuzzy normalized relationship can be defined by the lower and upper bounds of each
α-cut of R̃il and r̃l j .
In addition, Chen and Weng [6] further proposed a modified formulation to obtain a more precise representation of
the fuzzy normalized relationship, of which the lower and upper bounds of the membership function at each α-cut are
formulated as:
J
(Ril )αL (rl j )αL
P
l=1
m(Ri0 j )αL = ,
J P
J J
(Ril )U
α (rlm )α + (Ril )αL (rl j )αL
P U
P
m=1 l=1 l=1
m6= j
(4)
J
(Ril )U
α (rl j )α
U
P
l=1
m(Ri0 j )U
α = .
J P
J J
(Ril )αL (rlm )αL (Ril )U
α (rl j )α
P P U
+
m=1 l=1 l=1
m6= j
584 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593

The above lower and upper bounds of α-cuts of fuzzy normalized relationship are aggregated with the importance
of each CR to determine the fuzzy technical importance ratings W̃ j for the jth DR in the form of α-cuts. Unlike
Chen and Weng’s study [6], considering the importance weight of customer requirements as a crisp number, fuzzy
numbers are used in this paper, since the fuzzy version seems more appropriate to deal with the inherent imprecision
and ambiguity of QFD processes. In determining the fuzzy importance score of CRs, the consensus of customers is
important since the DRs should satisfy the CRs maximally. For doing this, Karsak [8] used the fuzzy Delphi method
to decide the fuzzy importance weight. This method is also adopted here to determine the fuzzy importance score of
CRs as a triangular fuzzy number. Appendix A simply describes the procedures of the fuzzy Delphi method.
k̃i as a fuzzy subset in [0, 1] represents the final average of the fuzzy importance score of CRi by experts’ evaluation
in the fuzzy Delphi method. In the α-cuts aspect, k̃i can also be denoted as [(ki )αL , (ki )Uα ]. Once the lower and upper
bounds of α-cuts of m( R̃i0 j ) and k̃i are obtained, the fuzzy technical importance ratings W̃ j for the jth DR can be
determined in the form of α-cuts as
 I I

i  (ki )α · m(Ri j )α (ki )α · m(Ri j )α 
L 0 L U 0 U
P P
h
 i=1 i=1
(W j )α = (W j )αL , (W j )U = , , (5)

α  I I
(ki )α (ki )α
 P 
U
P L
i=1 i=1

based on the division operation of positive fuzzy numbers described by the interval arithmetic [22,25]. The resulting
W̃ j is then employed to figure out the optimal fulfilment level of each DR to maximally fulfil the CRs, since the fuzzy
set of W̃ j represents the overall customer satisfaction that can be achieved by DR j .
Using the fuzzy technical importance ratings of DRs, Chen and Weng [6] proposed a fuzzy linear programming
model, in which the decision variables, x j , are defined in percentages to denote the level of fulfilment percentage of the
DR j , j = 1, 2, . . . , J , i.e. x j ∈ [0, 1]. The x j = 0, which implies that the DR has a basic design requirement so that
no more effort and cost are needed. In Chen and Weng’s model, the increment unit cost to achieve the fulfilment level
is represented as a fuzzy number, C̃ j , to reflect its fuzzy nature at the design stage. With the fulfilment percentage of
DR j , a corresponding percentage of the increment unit cost is required to enhance the quality of the product or service.
The total incremental unit cost cannot exceed a cost constraint. Besides the cost limitation, the impact on customer
satisfaction of various DRs is prioritized, and business competitions and technological difficulties are also considered
in the model. If DRs is preferred to DR p in terms of customer satisfaction, a constraint of W̃s · xs − W̃ p · x p ≥ 0 is
needed. In order to solve the fuzzy linear programming model, the lower and upper bounds of α-cuts of W̃ j and C̃ j
are placed in the model to find the lower and upper bounds of x j and the total customer satisfaction at each α-cut.
Thus, a fuzzy linear programming model is formulated as:
J
X
(Z )αL = max (W j )αL · x j ,
j=1
J
X
s.t. (C j )U
α · x j ≤ B,
j=1

(Ws )αL · xs − (W p )U
α · x p ≥ 0,
0 ≤ ε j ≤ x j ≤ η j ≤ 1, ∀ j,
s, p ∈ {1, 2, . . . , J }, (6a)
XJ
(Z )U
α = max (W j )U
α · xj,
j=1
J
X
s.t. (C j )αL · x j ≤ B,
j=1

(Ws )U
α · x s − (W p )α · x p ≥ 0,
L

0 ≤ ε j ≤ x j ≤ η j ≤ 1, ∀ j,
 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593 585

Fig. 2. Kano’s conceptual model.

s, p ∈ {1, 2, . . . , J }, (6b)
where B is the budget limitation, and (Z )αL as well as (Z )U
α represent the lower and upper bounds of objective values,
i.e. total customer satisfaction, at each α-cut, respectively. In addition, ε j and η j denote the possible range of the
fulfillment level of one DR, indicating the minimum required level due to the business competition and the maximum
level due to technical difficulty.

3. A fuzzy nonlinear model

The solutions of Model (6) are the fulfilment levels of DRs in order to maximize total customer satisfaction.
From this perspective, the fulfilment levels of DRs are related to product performance, and DRs with the higher
technical importance rating will have a greater contribution to total customer satisfaction. In other words, the DRs
with equal fulfilment levels achieve different levels of customer satisfaction. For classifying DRs, this paper employs
Kano’s concept to categorize DRs into different classes. Kano et al. [13] proposed a useful method in quantifying
the relationship between quality performance of a product or service and customer satisfaction. Three kinds of
conceptual relationships were presented: attractive, one-dimensional, and must-be relationships. Fig. 2 demonstrates
these relationships. For the DRs with an attractive relationship, increasing performance of the product in terms of
the fulfilment levels can produce more customer satisfaction than the other DRs. The one-dimensional relationship is
a linear function between performance and satisfaction. The must-be relationship is described as when an essential
performance of product is enough to satisfy the customers’ need.
Since the fulfilment level of each DR is related to product performance and its fuzzy technical importance rating W̃ j
is closely associated with customer satisfaction, it is reasonable to categorize DRs into three Kano’s classes based on
the fuzzy technical importance ratings. The membership function of the fuzzy technical importance rating of each DR
is determined by (5). Therefore, the categorization of DRs can be achieved by ranking the fuzzy technical importance
ratings W̃ j of DRs. W̃ j is a fuzzy number, since it is convex and normal based on (4) and (5).
Accordingly, for the categorization, a fuzzy number ranking approach, proposed by Chen and Lu [26], is employed
by this paper to rank the fuzzy ratings. The approach uses the lower and upper bounds of several α-cuts of two fuzzy
numbers to calculate the average difference between them. For ranking the j and l fuzzy technical importance ratings,
the measure for the average difference D j,l by (n + 1) α-cuts is formulated as follows [26]:
(" # " #)
n n n n
1 X X X X
D j,l (β) = β (W j )U
m + (1 − β) (W j )m
L
− β (Wl )U
m + (1 − β) (Wl )m
L
, (7)
n+1 m=0 m=0 m=0 m=0

where (W j )m
L and (W )U are the lower and upper bounds of the mth α-cut of W̃ ; the index β is used to reflect a
j m j
decision maker’s attitude to the lower or upper bound. The value of β equal to 0.5 is adopted for a neutral decision.
The greater the measure D j,l (>0) is, the greater the average difference that the fuzzy number jis larger than l. In
586 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593

general, a fuzzy number is selected as the reference when there are m fuzzy numbers to be compared. Obviously, the
measure D j,l = 0 when the comparison happens at the reference itself, i.e. j = l. Once the measure D j,l for all fuzzy
numbers based on the specified reference is obtained, the rank of the fuzzy numbers can be determined easily. In this
study, the number of fuzzy technical importance ratings to be ranked is equal to that of DRs.
After ranking the fuzzy technical importance ratings W̃ j , we categorize all DRs into three classes in accordance
with Kano’s concept. Let 1-class DRs be ones with the higher measure value with respect to the reference, and 3-
class be the reverse. Obviously, the performance achieved by the 1-class DRs can increase customer satisfaction by
more than the DRs in the other classes. In other words, increasing the performance level by 1-class DRs, and even
reducing that by 3-class DRs due to budgetary limitations, can augment the total customer satisfaction. Based on this
consideration, we treat the 1-class DRs as attractive requirements, while 3-class DRs are must-be ones. As to the
2-class DRs, they are considered as one-dimensional, playing the same role as in (6).
In order to incorporate the above concepts into constructing the fuzzy model, the decision variable x j representing
q
the performance level of DR j is modified as x j j , where q j = 0.5, 1, and 2, when the DR has the attractive, one-
dimensional and must-be feature, due to x j ∈ [0, 1]. Referring to the model (6), a fuzzy nonlinear programming
model to maximize customer satisfaction can be formulated as
J
X q
(Z 0 )αL = max (W j )αL · x j j ,
j=1
J
X q
s.t. (C j )U j
α · x j ≤ B,
j=1
q q
(Ws )αL · xs s − (W p )U p
α · x p ≥ 0,
q
0 ≤ ε 0j ≤ x j j ≤ η0j ≤ 1, ∀ j,
s, p ∈ {1, 2, . . . , J }, (8a)
J
X qj
(Z 0 )U
α = max α · xj ,
(W j )U
j=1
J
X q
s.t. (C j )αL · x j j ≤ B,
j=1
q q
(Ws )U
α · x s − (W p )α · x p ≥ 0,
s L p

q
0 ≤ ε 0j ≤ x j j ≤ η0j ≤ 1, ∀ j,
s, p ∈ {1, 2, . . . , J }, (8b)

It is noted in (8) that the possible range of the performance level of one DR can be changed in accordance with
the modified decision variable. From the application end, the possible performance level of one DR with the attractive
feature can be increased, while that with the must-be feature can be reduced for the increment of the total customer
satisfaction under the budgetary limitation. After solving the above model at various α ∈ [0, 1], the lower and upper
q
bounds of the objective function and each decision variable x j j can be obtained, and then the associated membership
functions are determined, which is useful for the QFD team in decision making.

4. An illustrative example

In this section, a semiconductor packing case of the turbo thermal ball grid array (T 2 -BGA) package is used
to exemplify the applicability of the proposed models. T 2 -BGA is a member of the ball grid array (BGA) family
in microelectronics packaging, and it can provide excellent thermal and electrical performance since a heat slug
is inserted into the moulding compound of a plastic BGA (PBGA). The heat slug has the advantages of high
heat dissipation and good ground shielding to reduce the electromagnetic interference (EMI) and electromagnetic
compatibility (EMC) effects between adjacent traces; however, it also produces some problems due to the more
 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593 587

Fig. 3. The cross-sections of PBGA and T 2 -BGA.

Fig. 4. The HOQ for a T 2 -BGA package.

complex structure of package when compared to PBGA [27]. Fig. 3 displays the cross-sections of PBGA and T 2 -
BGA.
Based on the above considerations of T 2 -BGA, a QFD team collects five customer requirements (CRs) and
proposes five design requirements (DRs) in the earlier design process. The fuzzy relations R̃il between CRs and
DRs and the relevant relations r̃l j among DRs are shown in the house of quality (HOQ) in Fig. 4. In HOQ, the CRs
are “package profile” (CR1 ), “thermal performance” (CR2 ), “electrical performance” (CR3 ), “reliability” (CR4 ), and
“co-planarity” (CR5 ), and the five DRs are “heat slug exposed area” (DR1 ), “heat slug attached material” (DR2 ),
“height of heat slug” (DR3 ), “copper pattern” (DR4 ), and “molding flow” (DR5 ). The importance levels of CRs and
relationships are denoted as “very low (VL)”, “low (L)”, “medium (M)”, “high (H)” or “very high (VH)”, which are
translated into the triangular fuzzy numbers as a 3-element set, (0, 0, .2), (0, .2, .4), (.3, .5, .7), (.6, .8, 1), (.8, 1, 1),
respectively, with membership functions in Fig. 5. As an example, R̃il = (.6, .8, 1) is formulated as
(Ril − .6)

 , .6 ≤ Ril ≤ .8,
(.8 − .6)


µ R̃il (Ril ) =
(1 − Ril )
, .8 ≤ Ril ≤ 1.


(1 − .8)


The middle value in the 3-element set representing the membership function of a fuzzy number is the most likely
one with the membership degree equivalent to 1, such as µ H (.8) = 1. In this study, the system of 0-.2-.5-.8-1 is
used to express the most likely value of the linguistic terms on VL-L-M-H-VH. Based on this system, the fuzzy
Delphi method is used to obtain the fuzzy importance score k̃i , and finally they are calculated as (.038, .238, .438),
(.263, .463,.663), (.113, .313, .513), (.55, .75, .925), (0, .15, .35) for the five CRs, respectively. The relevant data and
computational procedures are listed in Appendix B. Furthermore, considering the linguistic relationships in Fig. 4, the
fuzzy normalized relationship at each α-cut can be calculated using (4). For doing this, the lower and upper bounds of
588 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593

Fig. 5. The membership functions of linguistic variables in the HOQ.

Table 1
The lower and upper bounds of fuzzy technical importance ratings at different α levels

α W1L W1U W2L W2U W3L W3U W4L W4U W5L W5U
1 .168 .168 .233 .233 .052 .052 .315 .315 .094 .094
0.8 .116 .280 .182 .354 .037 .092 .238 .517 .063 .159
0.6 .074 .419 .138 .498 .025 .140 .169 .745 .039 .243
0.4 .040 .583 .102 .668 .015 .198 .110 .998 .021 .349
0.2 .016 .775 .073 .862 .008 .264 .059 1.278 .008 .476
0 .000 .992 .050 1.081 .002 .339 .017 1.584 .000 .625

Fig. 6. The membership functions of fuzzy technical importance ratings.

each α-cut of R̃il and r̃l j should be determined beforehand based on their membership functions. Actually, the lower
and upper bounds of each α-cut can be represented as the functions of α(∈ [0, 1]). As an example, the α-cut of the
membership function (.6, .8, 1) of “H” is expressed as [(Ril )αL , (Ril )U
α ] = [.6+.2α, 1−.2α]. Without any biases, the α
levels in this paper are evenly distributed in [0, 1]. Let αm denote the mth α level, and αm = m/n, m ∈ {0, 1, . . . , n},
such that the distance between each two adjacent α levels is equal, i.e. αm − αm−1 = 1/n, m ≥ 1.
The lower and upper bounds of the fuzzy technical importance rating (W̃ j )α , j = 1, 2, . . . , J , at each α level, is
obtained by (5) to determine the importance priority of each DR. The lower and upper bounds of (W̃ j )α at different
α levels are listed in Table 1, and their membership functions are shown in Fig. 6. They are applied to (6) and (8) to
solve fuzzy linear and nonlinear models.
As described previously, the DRs are categorized into three basic classes based on Kano’s concept. For
categorizations, (7) is applied to calculate the measure of difference between two fuzzy technical importance ratings.
Visually, W̃3 has the least contribution to customer satisfaction, so it is treated as the reference in the comparisons.
Suppose that the neutral decision factor is set, i.e. β = 0.5. After applying (7) based on 11 α-cuts that α levels are
distributed evenly in [0, 1], the values of the difference measure D j,3 , j = 1, . . . , 5, are 0.198, 0.269, 0.0, 0.424, and
0.077, respectively. In order to assign each DR into the associated class based on the measure, a relative scale is used
 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593 589

Table 2
The cost, business competition, and technical difficulty for each DR

DR j C̃ j α-cut εj ηj
DR1 (.2, .3, .4) [.2 + .1α, .4 − .1α] .5 1.0
DR2 (.8, .9, 1.) [.8 + .1α, 1. − .1α] .3 1.0
DR3 (.4, .5, .6) [.4 + .1α, .6 − .1α] .1 1.0
DR4 (.3, .4, .5) [.3 + .1α, .5 − .1α] .1 1.0
DR5 (.1, .2, .3) [.1 + .1α, .3 − .1α] .2 1.0

in the following:
D j,l − min D j,l
D 0j,l = , j = 1, 2, . . . , J, (9)
max D j,l − min D j,l
where l is set as the reference. In this case, the minimum among the difference measures is zero, so that the resulting
scales are the percentages relative to the maximum, i.e. 46.7%, 63.4%, 0%, 100%, 18.2%. For the assignments, a
three-interval system, i.e. 0-33.3%-66.7%-100%, is adopted in this paper based on the assumption that the quantity
of the relative difference is closely related to a DR’s feature in the interval manner. Following this rule, the fourth
requirement DR4 is considered the one with the attractive feature; DR1 and DR2 have one-dimensional features; DR3
and DR5 have must-be features.
In addition to incorporating the Kano’s concept, the fuzzy nonlinear model also considers the business competition
and technical difficulty of each DR as well as the required cost when one DR’s performance level is fully achieved,
i.e. 100%. As with the fuzzy linear model [6], the required increment unit cost is represented as a fuzzy number. For
comparisons, Table 2 lists the associated data for the fuzzy linear model. To be in accordance with the above Kano’s
categorization, the corresponding business competition is changed as ε10 = ε1 = .5, ε20 = ε2 = .3, ε30 = .01, ε40 = .32,
and ε50 = .04, in constructing the fuzzy nonlinear model. The budgetary total increment cost is limited to 1 unit,
i.e. B = 1.0 in (6) and (8). Furthermore, we suppose that the QFD team has prioritized DRs as DR4  DR1  DR2
and DR3  DR5 for the design preference, where “” means “is more preferred than”.
Applying fuzzy technical importance ratings, fuzzy increment cost, and the associated constraints to (6) and (8) at
several α levels, the performance level of each DR and the total customer satisfaction degree (the objective function
value) can be obtained from each model.
Both models (Model (6) and (8)) are solved by Lingo 9.0 software in a Microsoft Windows XP environment running
on a laptop with a 1.5 GHz processor as well as a 512 MB random access memory. The efficiency is satisfactory due
to the software runtime close to zero seconds for each α-cut. In this example, the settings of α levels less than 0.9 and
0.85 will produce infeasible regions in solving fuzzy linear and nonlinear models, respectively. From the property of a
fuzzy number, this means that the confidence level of fuzzy inputs should be high to avoid producing the wider ranges
of α-cuts. Fig. 7 shows the membership functions of the total customer satisfaction degree for the two models based
on three α levels (0.9, 0.95, 1.). Obviously, the membership function constructed from the fuzzy nonlinear model is
greater than that from the fuzzy linear model, indicating that incorporating the Kano’s concept into the model can
enhance the total customer satisfaction. The range of the performance level of each DR at the corresponding α level
can also be obtained from (6) and (8). However, for application purposes, the crisp performance level should be taken
as the action by the QFD team in the design.
For doing this, a crisp performance level belonging to the α-cut of each DR at a particular α level that the QFD team
accepts can be adopted. In fact, with the fuzzy sense, the α level can be interpreted as the confidence degree [28–31].
Based on the acceptable confidence degree, the crisp performance level of DR j from the fuzzy nonlinear model can
be determined as
q q
(x 0j )α = β(x j j )αL + (1 − β)(x j j )U
α, (10)
q q
where (x j j )αL and (x j j )U
α are the lower and upper bounds of DR j at the α level, respectively; the index β is used to
reflect the QFD team’s attitude to the lower or upper bound. Suppose that the QFD team’s attitude is neutral and α
level = 0.9 is accepted for the design purpose. Then the crisp performance levels of the five DRs at α = 0.9, (x 0j )α=0.9 ,
j = 1, . . . , 5, are 0.905, 0.314, 0.075, 1.0, and 0.04, respectively; while, based on the fuzzy linear model (6), they are
590 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593

Fig. 7. The membership functions of fuzzy linear and nonlinear models.

0.65, 0.3, 0.392, 0.751, and 0.2. It is obvious that the crisp performance levels of DR3 and DR5 from (8) are less than
those from (6) by 0.317 and 0.16 respectively, since they are classified as the must-be features. With the attractive
feature, the DR4 level is increased by around 0.25 from 0.751 to 1.0. As to DR1 and DR2 with the one-dimensional
feature, they are also increased, in comparison with those from (6), to make up the less performance levels of DR3 and
DR5 in order to maximize the total customer satisfaction, even though the DR4 level is larger than that from (6). It is
worth noting that DR1 is increased more than DR2 , because DR1 has higher priority in design preference and a lower
increment unit cost than DR2 . In fact, the change of the performance levels of DRs with the one-dimensional feature
is dependent on the quantity change of those DRs belonging to the other two Kano’s categories in maximizing the
total customer satisfaction. Based on the modification of the performance level, the total customer satisfaction degree
is increased, shown in Fig. 7.

5. Concluding remarks

Determining the performance levels of DRs is an important decision problem in QFD applications. Considering
the Kano’s concept, this paper proposes a fuzzy nonlinear model to determine the performance levels of DRs based
on the existing fuzzy linear model. The DRs can be categorized into three classes based on the Kano’s concept. The
attractive class has the higher impact on customer satisfaction, so that the performance level of one DR in this class
is preferably increased. Meanwhile, the must-be class makes less contribution to customer satisfaction, and therefore
the performance level can be reduced. As to the DRs in the one-dimensional class, the change of their performance
levels is dependent on the quantity change of those DRs belonging to the other two Kano’s categories in maximizing
total customer satisfaction. Through the modification of the decision variables denoting the performance level, a fuzzy
nonlinear model can be constructed. The illustration of the proposed model with a numerical example indicates that
the total customer satisfaction degree is greater than that from the fuzzy linear model. By comparing the performance
levels of DRs from the two models, the changes actually reflect the categorizations based on Kano’s concept. Besides
Kano’s concept, in future research, the other considerations, such as design risk concept, the integration of more
phases of QFD activity, and so on, could be modelled in the QFD problems.

Acknowledgements

The authors appreciate the anonymous reviewer’s useful comments to make this paper more comprehensive. This
research was financially supported by the National Science Council, Republic of China, under contract NSC95-2416-
H-006-028-MY3.

Appendix A

The fuzzy Delphi method consists of four steps for determining the importance score of CRs, and is described as
follows [8,32].
 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593 591

Step 1. The current and potential customers, acting as an expert role Custm , m = 1, 2, . . . , n, provide the possible
importance score of CRi , i = 1, 2, . . . , I . The importance score assigned by each expert Custm is presented in the
form of a triangular fuzzy number as
h i
k̃i,m = (ki,m )a , (ki,m )b , (ki,m )c , m = 1, 2, . . . , n; i = 1, 2, . . . , I, (A.1)

where k̃i,m denotes the importance score of CRi by the expert m, the indexes “a”, “b”, “c” represent the lowest, the
most likely and the highest level of k̃i,m .
Step 2. Computing the average of all k̃i,m , the formulation is expressed as
" #
n n n
h i 1 1 1
k̄˜ i = (k̄i )a , (k̄i )b , (k̄i )c =
X X X
(ki,m )a , (ki,m )b , (ki,m )c , (A.2)
n m=1 n m=1 n m=1

where i = 1, 2, . . . , I . And then, for each expert Custm the differences between k̃i,m and k̄˜ i are calculated and returned
to each expert for the next reevaluation run.
Step 3. Each expert takes the differences in the previous step as a reference, and then provides a revised linguistic
variable (if necessary) of the importance scores as
h i
0
k̃i,m = (ki,m
0
)a , (ki,m
0
)b , (ki,m
0
)c , m = 1, 2, . . . , n; i = 1, 2, . . . , I. (A.3)

Then the average k˜¯0 i,m can also be obtained from (A.2) and its difference to k̄˜ i is calculated using a distance measure,
proposed by Bojadziev and Bojadziev [33], as follows:
n h   i  o
d(k̄˜ i , k˜¯0 i ) = 0.5 max (k̄˜ i )a − (k˜¯0 i )a  , (k̄˜ i )c − (k˜¯0 i )c  + (k̄˜ i )b − (k˜¯0 i )b  . (A.4)
     

This process could be repeated until two adjacent averages become reasonably close. The consensus criterion is
assumed to be a small distance, such as d ≤ 0.2.
Step 4. If there is any change or update information that may lead to a reevaluation of the importance scores of CRs
at a later time, Step 1 should start the next-round process.

Appendix B

The fuzzy Delphi method is applied to determine the importance score of CRs in T 2 -BGA package development.
The computational procedure is described as follows:
Step 1. It is assumed that a QFD team forwards a questionnaire with linguistic evaluation levels, “very low (VL)”,
“low (L)”, “medium (M)”, “high (H)” or “very high (VH)”, to eight current and potential customers in order to obtain
the importance scores of five CRs, as mentioned in Section 4. As the experts, the eight customers use those five
linguistic terms to express the importance score of each CR. Their first evaluations are listed in Table B.1. Then, the
QFD team translated those linguistic terms into the triangular fuzzy numbers as a 3-element set, namely (0, 0, .2),
(0, .2, .4), (.3, .5, .7), (.6, .8, 1), and (.8, 1, 1), respectively. For example, the importance score of the fifth CR evaluated
by the second expert is expressed as k̃5,2 = (0, .2, .4).
Step 2. Computing the average of all k̃i,m by (A.2), the producing results of k̄˜ i to each CR are (.075, .225, .425),
(.263, .463, .663), (.113, .263, .463), (.438, .638, .813), and (.038, .138, .338), respectively. Then, the differences
between k̃i,m and k̄˜ i are calculated and sent back to each corresponding expert for reevaluation.
Step 3. As Step 1, each expert provides a revised evaluation, as listed in Table B.2. The average of k̃i,m 0 can
˜
also be obtained from (A.2), and then the k¯0 i of each CR is determined as (.038, .238, .438), (.263, .463, .663),
(.113, .313, .513), (.55, .75, .925), and (0, .15, .35), respectively. The distance between k˜¯0 and k̄˜ is calculated by
i i
(A.4) as .025, 0, .05, .113, .025, respectively. Since the distances between two adjacent averages are less than or equal
to 0.2, this means that the consensus condition is achieved. The resulting fuzzy number k˜¯0 i will be applied to determine
the fuzzy technical importance ratings. If the consensus condition was not obtained, the reevaluation process would
be asked to perform the next round until the consensus condition is achieved.
592 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593

Table B.1
The first round evaluation of importance score of the each CR

Expert CR1 CR2 CR3 CR4 CR5

Cust1 L M L H VL
Cust2 L M VL H L
Cust3 VL H M M L
Cust4 M L L H M
Cust5 L L L VH VL
Cust6 L M H M VL
Cust7 M M L L VL
Cust8 VL M VL M L

Table B.2
The second round evaluation of importance score of each CR

Expert CR1 CR2 CR3 CR4 CR5

Cust1 L M L H L
Cust2 L M L H L
Cust3 L M M M L
Cust4 L L L H L
Cust5 L M L VH L
Cust6 L M H H VL
Cust7 M M L M VL
Cust8 L M L H L

References

[1] J.R. Hauser, D. Clausing, The house of quality, Harvard Business Review 66 (1988) 63–73.
[2] L. Cohen, Quality Function Deployment: How to Make QFD Work for You, Addison Wesley, Reading, MA, 1995.
[3] J. Terninko, Step-by-step QFD: Customer-driven product design, 2nd edition, St. Lucie Press, Boca Raton, FL, 1997.
[4] M. Zhou, Fuzzy logic and optimization models for implementing QFD, Computer Industrial Engineering 35 (1998) 237–240.
[5] J. Wang, Fuzzy outranking approach to prioritize design requirements in quality function deployment, International Journal of Production
Research 37 (4) (1999) 899–916.
[6] L.H. Chen, M.C. Weng, A fuzzy model for exploiting quality function deployment, Mathematical and Computer Modeling 38 (2003) 559–570.
[7] L.K. Chan, M.L. Wu, A systematic approach to quality function deployment with a full illustrative example, Omega 33 (2) (2005) 119–139.
[8] E.E. Karsak, Fuzzy multiple objective decision making approach to prioritize design requirements in quality function deployment,
International Journal of Production Research 42 (18) (2004) 3957–3974.
[9] C.K. Kwong, H. Bai, Determining the importance weights for the customer requirements in QFD using a fuzzy AHP with an extent analysis
approach, IIE Transactions 35 (2003) 619–626.
[10] L.K. Chan, H.P. Kao, A. Ng, M.L. Wu, Rating the importance of customer needs in quality function deployment by fuzzy and entropy methods,
International Journal of Production Research 37 (11) (1999) 2499–2518.
[11] L.P. Khoo, N.C. Ho, Framework of a fuzzy quality function deployment system, International Journal of Production Research 34 (2) (1996)
299–311.
[12] C. Temponi, J. Yen, W.A. Tiao, House of quality: A fuzzy logic-based requirements analysis, European Journal of Operational Research 117
(1999) 340–354.
[13] K.J. Kim, H. Moskowitz, A. Dhingra, G. Evans, Fuzzy multi-criteria models for quality function deployment, European Journal of Operational
Research 121 (2000) 504–518.
[14] L.V. Vanegas, A.W. Labib, A fuzzy quality function deployment (FQFD) model for deriving optimum targets, International Journal of
Production Research 39 (1) (2001) 99–120.
[15] L.V. Vanegas, A.W. Labib, Application of new fuzzy-weighted average (NFWA) method to engineering design evaluation, International
Journal of Production Research 39 (6) (2001) 1147–1162.
[16] E.E. Karsak, Fuzzy multiple objective programming framework to prioritize design requirements in quality function deployment, Computers
& Industrial Engineering 47 (2004) 149–163.
[17] N. Kano, N. Seraku, F. Takahaashi, S. Tsuji, Attractive quality and must-be quality, Hinshitsu: The Journal of the Japanese Society for Quality
Control (1984) 39–48.
[18] K. Matzler, H.H. Hinterhuber, How to make product development projects more successful by integrating Kano’s model of customer
satisfaction into quality function deployment, Technovation 18 (1) (1998) 25–38.
[19] K.C. Tan, X.X. Shen, Integrating Kano’s model in the planning matrix of quality function deployment, Total Quality Management 11 (8)
(2000) 1141–1151.
 L.-H. Chen, W.-C. Ko / Mathematical and Computer Modelling 48 (2008) 581–593 593

[20] S.B. Han, S.K. Chen, M. Ebrahimpour, M.S. Sodhi, A conceptual QFD planning model, The International Journal of Quality & Reliability
Management 18 (8–9) (2001) 796–812.
[21] G.S. Wasserman, On how to prioritize design requirements during the QFD planning process, IIE Transactions 25 (3) (1993) 59–65.
[22] G. Klir, B. Yuan, Fuzzy sets and fuzzy logic: Theory and Application, 3rd impression, Pearson Education, Taiwan, 2003.
[23] L.A. Zadeh, Fuzzy set as a basis for a theory of possibility, Fuzzy Sets and Systems 1 (1978) 3–28.
[24] H.J. Zimmermann, Fuzzy Set Theory and its Applications, 2nd edition, Kluwer-Nijhoffg, Boston, MA, 1991.
[25] C.H. Cheng, Evaluating weapon systems using ranking fuzzy numbers, Fuzzy Sets and Systems 107 (1999) 25–35.
[26] L.H. Chen, H.W. Lu, An approximate approach for ranking fuzzy numbers based on left and right dominance, Computers and Mathematics
with Applications 41 (2001) 1589–1602.
[27] K.M. Chen, K.H. Horng, K.N. Chiang, Coplanarity analysis and validation of PBGA and T 2 -BGA packages, Finite Elements in Analysis and
Design 38 (2002) 1165–1178.
[28] H.C. Wu, Linear regression analysis for fuzzy input and output data using the extension principle, Computers and Mathematics with
Applications 45 (2003) 1849–1859.
[29] J. Bondia, J. Picó, Analysis of linear systems with fuzzy parametric uncertainty, Fuzzy Sets and Systems 135 (2003) 81–121.
[30] P.T. Chang, E.S. Lee, A generalized fuzzy weighted least-squares regression, Fuzzy Sets and Systems 82 (1996) 289–298.
[31] D.L. Mon, C.H. Cheng, H.C. Lu, Application of fuzzy distributions on project management, Fuzzy Sets and Systems 73 (1995) 227–234.
[32] A. Kaufmann, M.M. Gupta, Fuzzy Mathematical Models in Engineering and Management Science, Elsevier Science Inc., New York, NY,
1988.
[33] G. Bojadziev, M. Bojadziev, Fuzzy Sets, Fuzzy Logic, Applications- Advances in Fuzzy Systems- Applications and Theory, vol. 5, World
Scientific, Singapore, 1995.