Reliability Engineering and System Safety 72 (2001) 327±334

www.elsevier.com/locate/ress

Fuzzy QFD for supply chain management with reliability consideration

So Young Sohn a,*, In Su Choi b
a
Department of Computer Science and Industrial Systems Engineering, Yonsei University, 134 Shinchon-dong, Sudaemon-ku, Seoul, South Korea
b
Korea Panel Research, Seoul, South Korea
Received 27 October 2000; accepted 10 February 2001

Abstract
Although many products are made through several tiers of supply chains, a systematic way of handling reliability issues in a various
product planning stage has drawn attention, only recently, in the context of supply chain management (SCM). The main objective of this
paper is to develop a fuzzy quality function deployment (QFD) model in order to convey fuzzy relationship between customers needs and
design speci®cation for reliability in the context of SCM. A fuzzy multi criteria decision-making procedure is proposed and is applied to ®nd
a set of optimal solution with respect to the performance of the reliability test needed in CRT design. It is expected that the proposed approach
can make signi®cant contributions on the following areas: effectively communicating with technical personnel and users; developing
relatively error-free reliability review system; and creating consistent and complete documentation for design for reliability. q 2001 Elsevier
Science Ltd. All rights reserved.
Keywords: Fuzzy QFD; Reliability; Customer satisfaction; Supply chain management

1. Introduction relation to those of PC monitor makers in the context of

SCM.
Although many products are made through several tiers End users' needs for a monitor typically consist of relia-
of supply chains, a systematic way of handling reliability in bility, cost and aesthetic attributes for the ®nished product.
product planning stage has drawn attention, only recently, in Among them, we focus our attention on reliability. When
the context of supply chain management (SCM) (Garg [1]). managing CRT vendors, set makers should be able to pro-
Consider a monitor production schematic as displayed in perly translate end users' needs for PC monitor in terms of
Fig. 1. One of the major components of PC monitor is cath- necessary reliability attributes of a CRT. This transition
ode ray tube (CRT), which requires numerous processing as however, is not trivial.
well as a series of assembly operation with various sub- For instance, customers of a PC monitor may require the
components such as an electron gun, shadow mask, and brightness of monitor should not fall down to a certain level
glass panels provided by vendors. This kind of an assembled in the lifetime of a monitor. What kind of reliability test has
CRT is then used as part of a PC set. to be required for CRT to keep this level of brightness?
In order to maximize end users' satisfaction, PC set Since the brightness can only be observed in the form of a
makers must understand what needs to be done in order to monitor, reliability assurance test for the brightness cannot
attract customers of PC. Once customers' needs are identi- be directly performed on the CRT itself. Therefore, when
®ed, they, in turn, have to be properly propagated to design CRT is made, instead of directly conducting the degradation
speci®cations of the subsystem such as CRT. Accordingly, test in terms of brightness, voltage test is done on CRT
needs of CRT makers should be known to vendors of CRT samples by taking the average maximum cathode induction
component or raw material, In order to maximize the system (MIK) of red, green and blue beams of electron gun of a
effectiveness, one would expect that vendors of the compo- CRT over various times of age. Although we use the MIK
nent of CRT (e.g. electron gun, shadow mask, and glass level changes as the proxy for the brightness level change,
panel) need to comply the requirements of CRT makers in the relationship between the MIK level and the brightness is
not clear yet (Sohn [2]).
In addition, the critical level of brightness which would
* Corresponding author. Tel.: 182-2-361-4014; fax: 182-2-364-7807. annoy monitor users are often represented in fuzzy verbal
E-mail address: [email protected] (S.Y. Sohn). expression. Therefore, one can easily imagine that the
0951-8320/01/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0951-832 0(01)00022-9
328 S.Y. Sohn, I.S. Choi / Reliability Engineering and System Safety 72 (2001) 327±334

Fig. 1. Design for reliability in a supply chain management.

required relationship for the reliability attributes between in Section 4, apply fuzzy QFD to ®nding design speci®ca-
the CRT and the components of CRT would be fuzzy in tions for reliability in the context of SCM. In Section 5,
many places. This is just a simple example where many study results are summarized.
relationships in SCM could be fuzzy. Therefore, in SCM,
it is important to de®ne customers and their fuzzy needs
properly at every chain. Consequently, customers' fuzzy 2. QFD studies
needs have to be identi®ed in relation to design speci®cation
as well. QFD consists hierarchically of several HOQ. Typically
Quality function deployment (QFD) is often used to the ®rst HOQ represents the relationship between the end
understand customers' needs on the product and to users' needs and product design variables as displayed in
relate them with design speci®cations through several Fig. 2. In the second HOQ, design variables of a product are
hierarchies of house of quality (HOQ). De®ning cus- related to those of components. In the third HOQ, design
tomers and their needs at each stage is not relatively variables of components are related to job attributes.
dif®cult compared to re¯ecting recursively both fuzzy needs Finally, in the last HOQ, job attributes are related to person-
and relationship in several HOQs. Fuzzy QFD has been nel job assignment.
developed mainly in view of fuzzy relationship between Much variation of QFD has been extensively studied
the customers' needs and design speci®cations (Khoo and recently. We review it in terms of application areas and
Ho [3]). However, not much research has been done to deal methodological issues. Main application areas of QFD are
with fuzzy customers' needs for reliability tests in the the new product design (Halbleib et al. [4], Belhe and
context of the SCM. Kuisak [5]). Examples of other application areas include
The main objective of this paper is to develop a fuzzy service improvement (Trappey et al. [6]), software develop-
QFD model in order to convey fuzzy relationship between ment (Haag et al. [7]), business planning (Crowe and Cheng
supply chains in terms of multi-attribute reliability require- [8]), rehabilitation engineering (Logan and Radcliff [9]),
ments. A fuzzy multi criteria decision making procedure is education process (Pitman et al. [10]), construction manage-
proposed and is expected to make signi®cant contributions ment (Mallon and Muligan [11]), and road traf®c accident
on the following areas: effectively communicating with control (Sohn [12]) as well.
technical personnel and users; developing relatively error- Some methodological problems of QFD are associated
free reliability review system; and creating consistent and with dimension problem of the relational matrix, compari-
complete documentation for design for reliability. son method of customer requirements and assessment of
Organization of this paper is as follows. In Section 2, relational matrix.
QFD studies are reviewed. In Section 3, a fuzzy multi In terms of methodology, ®rst, the dimension problem
criteria decision making procedure is illustrated. We then, (Kim and Shin [13]) has been indicated. That is the size
 S.Y. Sohn, I.S. Choi / Reliability Engineering and System Safety 72 (2001) 327±334 329

Fig. 2. Hierarchical stucture of QFD.

of house of quality increases as the dimension of customers' the relationship between the customer requirements and the
needs and the number of potentially related design variables design speci®cations. In order to identify such relationship,
becomes large. One of the main problems associated with a group discussion is often held and the correlation is assigned
large dimension of the relational matrix is the dif®culty of to each combination of customer requirement and design
setting priorities of customers' needs in an accurate manner. speci®cation variables based on the linguistic expression
Accuracy of QFD depends on the comparison results of such as `strong' or `weak' relationship. Linguistic expres-
customers' needs. In order to avoid such accuracy problems sion could often be fuzzy or stochastic (Tanaka and Asai
due to a large dimension, several partition methods have [17]). Nevertheless most of the QFD literature assumes a
been suggested including those applied to group technology crisp relationship, which may not fully re¯ect the reality.
(Kim and Shin [13]). Other suggestions include the use of Tanaka and Asai [17] tried a fuzzy analysis based on the
attribute categorization and evaluation (ACE) matrix (Park linguistic expression on the relationship in QFD. Their
et al. [14]). Instead of putting all needs in one dimension, fuzzy approach, however, is limited to the relational matrix
ACE matrix categorizes them into a three by three matrix and helps setting priorities of design variables but it does not
where each block consists of a combination of attributes provide the way to ®nd the optimal levels of design variables.
from a group of basic, discriminant, and exciting needs When some retrospective data are available, a classical
and a group of positive, negative and neutral needs. In this regression can be utilized in QFD to ®nd the relationship
way, comparison can be done within each block and reduces between the customers' needs and design variables. In view
the burden of a large dimensional comparison. of the fact that the characteristics of customers' needs are
Secondly, in order to assess the relative importance typically observed as frequency, Poisson regression can be a
of customer requirements, analytical hierarchy analysis proper approach to estimate such a relationship, when the
(AHP) and conjoint analysis are often used (Saaty [15]). frequency is relatively low (Sohn [18±20]). When such a
When comparing customer requirements, AHP takes pair- sample based relationship is estimated, most of the QFD
wise comparison while conjoint analysis uses rating or literature set the ranking of design variables based on the
ranking, method, respectively. Various comparison methods point estimate alone, neglecting the in¯uence of the
(pairwise comparison, rating and ranking) are evaluated and sampling error. It is recommended that the interval estima-
the following results are obtained (Elrod and Louviere [16]): tion be used for the comparison purposes, acknowledging
superior estimation ef®ciency of rating model to the ranking the sampling error. Other problem is that the relationship
model; relative informational ef®ciency of rankings to pair- between the customer needs and design variables may not
wise comparison; and rating models being less vulnerable to be ®xed. In this case, a random effects model (Sohn [21])
bias than ranking models. can be applied.
Next methodological problem lies in the assessment of Next, when a regression approach is used to estimate
330 S.Y. Sohn, I.S. Choi / Reliability Engineering and System Safety 72 (2001) 327±334

the relationship, one has to be careful in scaling data, matrix relating customers' needs to manufacturing variables
since the magnitude of the estimated regression coef®cients X. Note that f(X) includes cross product terms of x1 ; ¼; xn in
depends on the unit of dependent variable (customers needs) addition to xj themselves. Selection of cross product terms in
and explanatory variables (design speci®cations). Not f(X) depends on the evaluation of the relationship between
considering this point, erroneously, regression coef®cients the manufacturing variables. For instance, for a positive
could be directly utilized to represent the degree of relation between x1 and x2, a12x1x2 can be added to f(X)
relationship. where a12 is positive.
In QFD, the relationship of our interest is not restricted to For each element of fuzzy set matrix, A, we assume a
that between the customers' needs and design speci®cation. triangular membership function. That is each element of A
Since numerous attributes are involved in design speci- matrix, a ik, has a center point aik and spread cik with
®cation, there could be potential correlation between the membership function
speci®cation variables. Many authors use the relational
maik …a† ˆ 1 2 ua 2 aik u=cik …2†
matrix of design speci®cation variables at the top of the
HOQ but the meaning and the usage of this matrix has for i ˆ 1; ¼; m; k ˆ 1; ¼; n 0 :
scarcely been mentioned. Additionally, in the next hierarchy of HOQ, manu-
In summary, it can be pointed out that the fuzzy concept facturing variables for CRT production (X) can be viewed
has not been fully cooperated in assessing the relation as CRT makers' requirements on the reliability test of CRT.
between the customers' needs and design variables; and CRT makers would try to conduct internal reliability tests to
even the crisp relation is not fully utilized to ®nd the optimal meet the PC monitor makers' requirements on the speci®ed
levels of design variables to best meet the customers' needs. attributes. We assume that the following fuzzy relationship
In applying QFD to ®nding the optimal degree of satis- between the n £ 1 matrix of CRT makers' manufacturing
faction for reliability tests in the SCM context, the following requirements X: an n £ p 0 matrix of fuzzy relational matrix
areas deserve the concern of fuzzy analysis: required level B; and a p 0 £ 1 vector g(Z) which is a function of z1 ; ¼; zp ;
of reliability test for the ®nal product; relational matrix representing the reliability test items for CRT
between the end users' needs and the design speci®cation
of the ®nal product; required level of reliability for the X ˆ Bg…Z†: …3†
subsystem of the ®nal product; relational matrix between
the design speci®cations of the ®nal product and those of By replacing X in Eq. (1) with Eq. (3), we ®nd
the subsystem; required level of reliability for the compo-
Y ˆ Af …Bg…Z††: …4†
nent. Adding these fuzzy relationship and requirement need
to be dealt with in terms of fuzzy multi-criteria decision Our objective is to ®nd the tolerable level of CRT
making problem. In Section 3, we propose a fuzzy multi makers' reliability test results Z, which will eventually
criteria decision making technique applicable to design for satisfy PC monitor users' requirement level ri for each
reliability (DFR) in the context of SCM. attribute yi. That is to ®nd Z such that

R1 , Y ˆ Af …Bg…Z†† , R2 ; …5†
3. Fuzzy multi criteria decision making
where elements of A,B,R1, and R2 are fuzzy with triangular
In this section, we show how QFD can be used not only to membership function speci®ed in Eq. (2). When A and B are
help setting priorities of customers' needs but also for opti- fuzzy, extension principle needs to be applied (Zadeh [22]).
mizing the levels of reliability test when the customer However, a closed form of membership function for Eq. (4)
requirements are fuzzy and the relationship between those cannot be found analytically based on sequential fuzzy set
and the design speci®cation variables are fuzzy as well. For analysis related to Eqs. (1) and (3). Therefore, instead of
illustration purpose, we consider two-stage supply chains of applying fuzzy set concept to A and B sequentially, we
manufacturing, but it can be easily extended to more general apply it to the ®nal coef®cients on Z after recursively ®nd-
cases. ing the relationship in Eq. (4).
Consider the schematic for design for reliability of a PC In order to ®nd the optimal Z which satis®es the fuzzy
monitor in the context of SCM displayed in Fig. 1. Suppose inequality in Eq. (5), we use Zimmerman's approach (Tong
that the end users' requirements on the functions of a PC [23], Lai and Chang [24]).
monitor (yi's) are identi®ed to have the following relation- First, we de®ne Y 2 R2 , 0 as constraint and Y 2 R1 . 0
ship with manufacturing variables of CRT (xj's) as goal. Then we transform all the constraints into goals.
Y ˆ Af …X†; …1† That is Y 2 R2 , 0 is transformed as 2Y 1 R2 . 0. Next
we denote gl be the element of Y 2 R1 and 2Y 1 R2. For
Where Y is an m £ 1 vector of customers' needs for PC l ˆ 1; ¼; 2m; fuzzy inequality gl . 0 shows to be almost
monitor; f(X) is an n 0 £ 1 vector of function of manu- positive. The exact de®nition of gl . 0 is that gl . 0 and
facturing variables of CRT x1 ; ¼; xn ; A is an m £ n 0 fuzzy the membership function of gl at 0 is mgl …0† # 1 2 h;
 S.Y. Sohn, I.S. Choi / Reliability Engineering and System Safety 72 (2001) 327±334 331

where h is the degree of gl . 0. Here the larger the h is, the

stronger the meaning of `almost positive' is.
Therefore, the problem is reduced to obtaining the largest
degree h. Hence, this problem is one of ®nding h and Z so as
to solve the following optimization problem.
max h
such that mgl …0† # 1 2 h for l ˆ 1; ¼; 2m; …6†
where the membership function of gl would be as displayed
in Fig. 5.
In order to accommodate the different degree of importance
assigned to each goal, optimization model can be revised as
followsP
max 2m lˆ1 wl hl

such that mgl …0† # 1 2 hl for l ˆ 1; ¼; 2m; …7†

where wl is the weight representing relative importance of
each goal l.

4. Examples
Fig. 4. Relationship between CRT design process and aging test.
In this section, we show a case based on the author's
consulting experience in order to illustrate how the proposed relationship matrix between the users' requirements and
approach can be applied for QFD in SCM context. CRT manufacturing variables, using one of 9, 3, and 1
As displayed in Figs. 3 and 4, ®rst of all, we summarize values representing very strong, strong and weak relation,
PC monitor users' requirements …yi ; i ˆ 1; ¼; 5† as bright- respectively (Misterek [25]). Intercorrelation between CRT
ness, linearity, focus, white balance, and convergence. manufacturing variables are also assigned re¯ecting their
Corresponding CRT manufacturing variables …xk ; k ˆ mutual impact. So model (1) for HOQ 1 can be written as
1; ¼; 5† are mask operation and welding, black matrix, follows
¯uorescent paint, electron gun sealing, and electron tube HOQ 1
operation which represent the sequence of CRT manufac-
turing process. Cross functional team is formed to assign the y1 ˆ …9x1 1 x3 1 9x4 1 3x5 1 3x3 x2 1 3x3 x4 †=28 …8†

y2 ˆ …x1 1 3x3 1 9x5 1 3x2 x3 1 3x3 x4 †=19

y3 ˆ …3x1 1 9x2 1 x4 1 9x5 1 3x2 x3 1 3x3 x4 †=27

y4 ˆ …3x2 1 3x2 x3 1 3x3 x4 †=9

y5 ˆ …x2 1 3x3 1 9x4 1 3x5 1 3x2 x3 1 3x3 x4 †=22;

where yi is the percentage complaint reported with respect to
user requirement i…0 # yi # 1†; and xk is the percentage
defects occurred in operation k…0 # xk # 1†:
In the second HOQ as displayed in Fig. 4, we consider

Fig. 3. Relationship between monitor users reequirements and CRT design

process variables. Fig. 5. Membership function of each goal.
332 S.Y. Sohn, I.S. Choi / Reliability Engineering and System Safety 72 (2001) 327±334

CRT manufacturing variables …xk ; k ˆ 1; ¼; 5† as custo- y3 ˆ 1:14z1 z3 1 0:1z1 z5 1 1:11z3 z5 1 0:1z3 z4 1 0:03z4 z5
mers' needs and relate them with reliability tests named
such as COEK, SHORT, MIK, Pin-C, and Stem-C …zj ; j ˆ 1 0:01z21 1 0:89z23 1 0:02z21 z3 1 0:09z1 z23 1 0:23z25
1; ¼; 5†: Again cross functional team members assign the rela-
tionship matrix between the CRT manufacturing variables 1 0:01z22 z4 1 0:08z1 z3 z5 1 0:02z1 z2 z3 z4 1 0:09z2 z3 z4
and reliability test variables, 9, 3, and 1 representing very
strong, strong and weak relationship, respectively. Resulting 1 0:08z2 z4 z5 1 0:02z1 z2 z4 1 2:74z1 1 0:21z2 1 2:56z3
model (3) is as follows
HOQ 2 1 0:34z4 1 6:12z5 1 0:9z2 z4 1 0:03z1 z2 1 0:09z2 z3

1 0:01z2 z4 1 0:09z2 z5 1 0:01z1 z2 z3

x1 ˆ …9z1 1 z4 1 3z5 1 z1 z3 1 z2 z4 †=15

y4 ˆ 0:02z1 1 0:02z21 1 0:04z1 z2 1 0:03z1 z5 1 0:02z21 z3

x2 ˆ …z1 1 3z3 1 9z5 1 z1 z3 1 z2 z4 †=15
1 0:02z1 z2 z4 1 0:06z3 1 0:003z1 z3 1 0:11z2 z3
x3 ˆ …3z1 1 9z3 1 z4 1 9z5 1 z1 z3 1 z2 z4 †=24 1 0:33z3 z4 1 0:01z2 z3 z4 1 0:2z5 1 0:34z2 z5

x4 ˆ …3z2 1 9z3 1 z1 z3 1 z2 z4 †=14 1 0:01z4 z5 1 0:0725 1 0:03z1 z3 z5 1 0:04z1 z2 z3

1 0:02z2 z4 1 0:01z1 z2 z3 z4z

x5 ˆ …9z1 1 3z3 1 9z4 1 3z5 1 z1 z3 1 z2 z4 †=26; (9)
y5 ˆ 0:05z1 1 0:11z2 1 0:28z3 1 0:04z4 1 0:08z5
where zj is the percentage of unsuccessful reliability test
j…0 # zj # 1†: 1 0:04z1 z3 1 0:04z2 z4 1 0:01z1 z3 1 0:01z1 z5
As mentioned in Section 3, it is not feasible to ®nd the
closed form membership function for this kind of hier- 1 0:04z23 1 0:01z2 z3 1 0:06z3 z5 1 0:01z1 z23
archical fuzzy relationship. Therefore, we treat each number
representing relationship in both HOQs as crisp until 1 0:01z2 z3 1 0:06z3 z5 1 0:01z1 z23 1 0:01z2 z3 z4
recursive relationship (4) is found as Eq. (10).
1 0:01z2 z5 1 0:03z25 1 0:01z1 z3 z5 1 0:01z2 z4 z5
y1 ˆ 0:2z1 1 0:2z3 1 0:06z4 1 0:09z5 1 0:6z1 z3 1 0:09z2 z4
Next we assume that each coef®cient in Eq. (10) as center
point and c ˆ 0:01 as the spread.
1 0:01z2 z3 1 0:02z23 1 0:01z2 z5 1 0:1z21 1 0:1z1 z5 Now we consider constraints for Y. Since Y represents
complaints, we do not need R1 in Eq. (5). R2 is the maximum
1 0:05z3 z5 1 0:03z25 1 0:01z1 z3 z5 1 0:07z2 z4 z5 percentage of complaints allowed for each user's require-
ment j and is assumed to be `very small' for [y1, y2, y3, y4, y5],
1 0:01z1 z3 z4 1 0:01z1 z2 z3 z4 1 0:01z22 z24 respectively. We then make all the constrains for Y be
2Y 1 R2 . 0.

Table 1
y2 ˆ 0:21z1 1 0:11z3 1 0:17z4 1 0:12z5 1 0:03z1 z2 Fuzzy optimization result

Spread Small Medium Large

1 0:17z1 z3 1 0:04z2 z4 1 0:09z2 z3 1 0:09z2 z5
h hi h hi h hi

1 0:02z3 z4 1 1:11z3 z5 1 0:1z1 z5 1 0:03z4 z5 1 0:01z21 COEK z1 0.00 0.00 0.00 0.00 0.005 0.00
SHORT z2 0.13 0.20 0.00 0.10 0.003 0.17
MIK z3 0.00 0.00 0.00 0.00 0.002 0.00
1 0:08z23 1 0:23z25 1 0:01z1 z2 z3 1 0:02z1 z2 z4 PIN-C z4 0.02 0.00 0.09 0.00 0.00 0.01
STEM-C z5 0.00 0.00 0.00 0.00 0.00 0.00
h 1.00 Na 1.00 Na 1.00 Na
1 0:09z2 z3 z4 1 0:08z1 z3 z5 1 0:08z2 z4 z5 1 0:02z1 z2 z4
Brightness h1 Na 0.18 Na 0.24 Na 0.68
Linearity h2 Na 0.67 Na 0.29 Na 0.48
1 0:09z2 z4 z5 1 0:02z21 1 0:09z1 z23 1 0:81z23 Focus h3 Na 0.54 Na 0.82 Na 0.40
White balance h4 Na 0.10 Na 0.95 Na 0.19
Convergence h5 Na 1.00 Na 1.00 Na 1.00
1 0:01z22 z4 1 0:02z1 z2 z3 z4
 S.Y. Sohn, I.S. Choi / Reliability Engineering and System Safety 72 (2001) 327±334 333

For fuzzy set representation for very small R2, we assume variables in each supply chain. In this paper, we used the
triangular membership functions with [r1, r2, r3, r4, hierarchies of QFD to systematically relate customers'
r5] ˆ [0.05, 0.05, 0.05, 0.05, 0.05] as center points and [c1, needs with design variables in each supply chain of product
c2, c3, c4, c5] ˆ [0.02, 0.02, 0.02, 0.02, 0.02] as spread development process. We then recursively consider those
values. relationships so that end users' requirement can be identi®ed
Based on this structure, we calculate triangular member- as functions of reliability test performance variables in the
ship functions of individual goal m gl (0) for l ˆ 1; ¼; 5 and last chain of the product design process. Next, we proposed
®nally solve Eq. (6). a fuzzy multi criteria decision-making procedure to ®nd a
According to the optimal solution illustrated in Table 1 set of optimal solution with respect to the reliability perfor-
with the medium level of spread we used, four kinds of mance needed. This optimization procedure can be easily
reliability tests (COEK, SHORT, MIK, STEM-C) need to extended to other types of models by accommodating
be perfectly satis®ed while the failure rate of the reliability various fuzzy optimization approaches including a gamma
test of PIN-C could be allowed upto 9% in order to fully operator. Although the model was developed with our appli-
satisfy customer needs where each need has the same degree cation in mind, it can be used to perform a host of various
of importance. levels of fairly general supply chains. We expect that our
Next we relax h to be different for each need in order to approach can make signi®cant contributions on resolving
accommodate different degrees of importance of each need. reliability issues in SCM context, which has scarcely been
In view of the fact that the brightness in considered as the considered previously.
most important requirement for the monitor, we assume that
the relative importance is as follows: w1 ˆ 0:6; w2 ˆ 0:1;
w3 ˆ 0:1; w4 ˆ 0:1; w5 ˆ 0:1; we ®nd the optimal solution Acknowledgements
to be (z1 ˆ 0:00; z2 ˆ 0:10; z3 ˆ 0:00; z4 ˆ 0:00; z5 ˆ 0:00)
as displayed in Table 1. In this case, overall customer satis- This work was supported by Korea Research Foundation
faction can be achieved although the failure rate of reliabil- Grant (KRF-97-005-E00191).
ity test with respect to SHORT is increased up to 10%. It is
interesting to note that the degree of satisfaction for each
goal varies and those goals for Brightness and Linearity are
References
particularly low.
These results are based on the assumed spread values of [1] Garg A. An application of designing products and processes for
fuzzy requirements and relationship. For sensitivity analy- supply chain management. IIE Trans 1996;31:417±29.
sis, we vary spread values. Let the currently used spread [2] Sohn SY. Application of degradation test data to advertisement of
values c ˆ 0:01 and [c1, c2, c3, c4, c5] ˆ [0.02, 0.02, 0.02, consumer electronic products. Reliab Engng Syst Safety
0.02, 0.02] be the medium level of spread. Small and large 2000;67:317±20.
[3] Khoo LP, Ho NC. Framework of a fuzzy quality function deployment
values of spread are set to be by subtracting and adding 0.05 system. Int J Prod Res 1996;34:209±311.
from the medium level, respectively. This would represent [4] Halbleib L, Wormington P, Cieslak W, Street H. Application of qual-
the relative degree of decreasing and increasing fuzziness on ity function deployment to the design of a lithium battery. IEEE Trans
both relationship and requirements on the reliability. As Components, Hybrid, Manufac Technol 1993;16:802±7.
fuzziness increases, strictness of reliability tests tends to [5] Belhe U, Kuisak A. The house of quality in a design process. Int J
Prod Res 1996;34:2119±31.
be spread over several tests, when the customers' needs [6] Trappey CV, Trappey AJC, Hwang SJ. A computerized quality func-
are treated equally important. However, when different tion deployment approach for retail services. Comput Industrial
weight is assigned to each need, this kind of relaxation Engng 1996;30:611±22.
applies to just one test. Most of the cases, reliability test [7] Haag S, Raja MK, Schkade LL. SQFD. Commun ACM 1996;39:42±
with respect to SHORT is considered to be relatively 49.
[8] Crowe TJ, Cheng CC. Using quality function deployment in manu-
relaxed. facturing strategic planning. Int J Ops Prod Mgmt 1996;16:35±48.
[9] Logan GD, Radcliffe DF. Potential for use of a house of quality matrix
technique in rehabilitation engineering. IEEE Trans Rehab Engng
5. Discussions 1997;5:106±15.
[10] Pitman G, Motwani J, Kumar A, Cheng CH. QFD application in an
educational setting Ð a pilot ®eld study. Int J Quality Reliab Mgmt
SCM is an important issue to be considered in the 1996;13:99±108.
customer relation management. We deal with a sequential [11] Mallon JC, Mulligan DE. Quality function deployment Ð a system
manufacturing process for electronic product in terms of for meeting customers' needs. J Const Engng Mgmt 1993;119:516±
supply chains connecting the vendor of raw material, assem- 31.
bly manufacturer, and the set maker. The main issue focused [12] Sohn SY. QFD applied to local traf®c accident reduction. Accident
Anal Prevention 1999;31:751±61.
was to ®nd the proper levels of reliability test performance, [13] Kim KJ, Shin C. Design problems: size reduction in QFD, Korean
which can satisfy various fuzzy requirements of end uses Society of IE, Fall Conference, Seoul, Korea, 1997.
and the relationship between customers' needs and design [14] Park C, Cho K, Jun C. Strategic product concept selection using ACE
334 S.Y. Sohn, I.S. Choi / Reliability Engineering and System Safety 72 (2001) 327±334

matrix in QFD, Korean Society of Quality Management, Fall Confer- [20] Sohn SY. In¯uence of a prior distribution for traf®c intensity estima-
ence Proceeding 313±323 Seoul, Korea, 1996. tion with covariates. J Statist Comput Simulation 1996;55:169±80.
[15] Saaty TL. The analytical hierarchy process. Pittsburgh, PA: RWS [21] Sohn SY, Park CJ. Random effects linear models for both process
Publications, 1990. mean and variance. J Quality Technol 1998;30:33±9.
[16] Elord T, Louviere JJ. An empirical comparison of rating-based and [22] Zadeh LA. The concept of a linguistic variable and its application to
choice-based conjoint models. J Marketing Res 1992;29:368±77. approximate reasoning 1. Inf Sci 1975;8:199±249.
[17] Tanaka M, Asai K. Fuzzy linear programming with fuzzy numbers. [23] Tong S. Interval number and fuzzy number linear programming.
Fuzzy Sets Syst 1984;13:1±10. Fuzzy Sets Syst 1994;66:301±6.
[18] Sohn SY. A comparative study of four estimators for analyzing the [24] Lai YJ, Chang SI. A fuzzy approach for multi response optimization:
random event rate of Poisson process. J Statist Comput Simulation an off-line quality engineering problem. Fuzzy Sets Syst
1994;49:1±10. 1994;63:117±29.
[19] Sohn SY. Empirical Bayesian estimation for the traf®c intensity para- [25] Misterek SDA. The performance of cross-functional quality improve-
meter in M/M/1 queues with covariates. Queuing Syst: Theory Appl ment project teams, University of Minnesota, PhD Thesis, Minnesota,
1996;22:383±401. 1995.