Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
KHADIJEHYOUSEFIPOUR
MAHMOUD ALBORZI
REZA RADFAR
JEDDI
Department of Management
Department of Management
Science and Research branch
Science and Research branch
Department of Management
Islamic Azad University
Islamic Azad University
Science and Research branch
Tehran, Iran
Tehran, Iran
Islamic Azad University
Tehran, Iran
Abstract: Many enterprises have been devoting a significant portion of their budget to product
development in order to distinguish their products from those of their competitors and to make them
better fit the needs and wants of customers. An important source of competitive advantage, survival and
renewal for firms is the successful new product development (NPD). Quality function deployment (QFD)
aims to facilitate the NPD process from product conceptualization to production requirements. This
paper proposes a decision system to support product development and used from first matrix of QFD. A
case study of a scale producer firm to select the Specification of new product based on data collected from
designed questionnaires is given which includes identification & determination of customers’ needs, while
looking for technical and engineering characteristics related to their needs. The results are then used to
construct houses of quality for QFD, which is incorporated by fuzzy AHP. A feed forward back
propagationANN is designed and trained based on the houses of quality. The result of neural net
simulation is compared with the results from fuzzy QFD model. It is concluded that the proposed decision
support system bycombining the fuzzy QFD and ANN models can be used as a powerful tool to select the
most suitable specificationfor new productsto satisfy customer's needs.
Keywords: New product development (NPD) – Quality function deployment (QFD) – Fuzzy AHP–
Decision Support Systems (DSS) – Artificial Neural Network (ANN)
I.
Introduction
Technological advances and increasing customer expectations have resulted in new products appearing on the
market at an ever increasing pace. The products are becoming more complex and the product life cycles are
getting shorter. Under a globally competitive business environment, technological innovation and satisfaction of
customer needs are the keys to survival and success for firms, especially for high-tech firms. The success of new
product development (NPD) is important to maintain a competitive edge and to make a decent profit for a firm
because new products are usually a source of new sales and profits (Lee and Lin, 2011). It is clear that NPD and
customer satisfaction are associated. In the NPD process, decision making at the front end is particularly
important as it helps companies concentrate on developing competitive and customer-focused products instead
of investing in worthless ones (Chan, Sze-ling, 2011). In order to support product development, there is a
requirement to capture the knowledge of the manufacturing processes within the organization, which includes
the process, materials, resource, design rules, capacity and other constraints that may limit the capabilities of the
organization.
The outcome is an effective decision support providing information and knowledge in the place, time and format
required, thus ultimately reducing product development costs and improving quality.
Companies that are able to bring new products, that satisfy the expectations of the customer fast and efficient to
the market, will manage to succeed in the intense and dynamic global environment in which it operate. The US
based Product Development & Management Association (PDMA) defines New Product Development as “a
disciplined and defined set of tasks and steps that describe the normal means by which a company repetitively
converts embryonic ideas into saleable products or services” (Østeras, .et al , 2004).
Therefore, relationship between marketing and production departments in any organization seems necessary in
order to provide goods, which can meet customer’s satisfaction in the best possible form (Armoun, et al, 2012).
The important relationship between production and marketing has been known in the late 1960s, i.e., when
Japanese used QFD to design products with features and taste of customers. They believe that constant changes
resulting from the globalization process and innovation in technology, which affects business environment
severely, force organizations to create new competitive advantages in order to maintain their situation in the
market.
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 663
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
The integration of fuzzy AHP-QFD for new product's Specification selection and by using the trained neural
network to predict the specification in this paper is a significant contribution of the proposed approach in
comparison to others in the literature.
Given the above subjects, we are in great need to design one DSS in order to apply comments and tastes of
customers in designing new products, on one hand and consider requirements and restrictions of producer to
meet demands of customer according to the features as expected by customer from product on the other hand.
It was decided to use this model in a company as producer of Digital Scales in IRAN.
The remainder of this paper is structured as follows. Section 2 presents a research background review focused
on the new product development using QFD. Section 3 presents a research framework and analysis procedure.
Section 4 presents data preparation and analysis and finally our concluding remarks are provided in Section 5.
II.
LITERATURE REVIEW
The literature on NPD contains several alternative NPD process models (e.g., Wesner et al, 1995, Wind, 1982,
Sounder, 1987, Pugh, 1991, Pahl and Beitz, 1988, Belliveau et al, 2002, and IEC 60300-1, 1991). It is possible
to recognize the similarities between the different models. What they have in common is that the NPD process
begins with an idea to build a product that meets specific needs (or create new needs for radically innovative
products) defined by customers and/or the manufacturer, and ends when the product is launched on the market.
This involves six phases as illustrated in Fig.1(Østeras, et al, 2004).
Fig.1: Six phase of NPD
Before a product is designed, most companies perform marketing studies. The goal of these studies is to
understand the customers’ expectations. Online analytical processing tools are used to extract relevant customer
information from multi-dimensional database. Classical statistical tools are used to compute various models
(e.g. regression models) and parameters (e.g. mean, confidence intervals) based on the collected data.
Hypotheses can be validated in support of decision-making (Bae& Kim, 2011).
In recent times, the decision support systems are implemented for new product development. These frameworks
applied modeling technique, several decision support techniques (such as the analytic hierarchy process and
discriminant analysis) (Liberatore and Stylianou,1995, JieL , et al, 2008, Oduoza and Harris,2011).
Quality function deployment (QFD) is a popular tool to fulfill the task in different phases in NPD.Despite of its
benefits, QFD does have some drawbacks (Büyüközkan et al. 2004; Ertay et al. 2005; Kahraman et al. 2006;
Kumar et al. 2007; Lee et al. 2010a). Therefore, some improved QFD approaches have been proposed to tackle
the issues such as the generalization of the opinions of multiple decision makers, the large amounts of subjective
data, the burden of a large dimensional comparison, and the ambiguity and uncertainty in human decision
making. While there are many hybrid QFD models available, the incorporation of fuzzy sets theory and analytic
hierarchy process (AHP) in QFD is one of the new trends (Bakshiet al.2012, Shahin et al.2011).
III.
PROPOSED MODEL
The structure of the proposed model is shown in Fig.2. The required data are initially prepared and make the
QFD Matrix. The central relation matrix of QFD is calculated by Fuzzy AHP. The results are used to design,
train, and after that to simulate the artificial neural network (ANN) model. The approval of these results and
final decision is made by the decision maker (DM).
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 664
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
The basic accepted customer attributes (Table 1) in the model are extracted from reviewed literature and the
interviews for the Digital Scale experts. Unit Price is one of important attributes for customers but we not
consider it in this paper;
Table 1: Customer attributes for purchasing Digital Scales
1-Max Weight
10- PLU numbers
2-Accuracy
11- Protect against liquid
3-Display (Type and position) 12- Count Function
4-Keypad (Type and position) 13- backup Battery
5-Printer Width
14- Reports
6-Printer type
15- Product guide and Learning CD
7-Plate dimension
16- Ability to Transfer data
8-Color
17- After-sales services
9-Proper packaging
For engineering characteristics, the specification of 20 Digital Scales collected from Manual and website of
Iranian producer and main specification is shown in Table. 2.
Table 2: Digital ScalesSpecifications
1.


2.



3.










4.
Printer
5. Keypad
Type : receipt , Label
Width : 57,80 mm
Display
Type : LED , LCD
Digit Numbers
Position : Up, Down of plate
Function
Type : Flat , Raised

Max : 30 , 40,50,75 kg
Accuracy: 5, 10,20 g

Position : Up, Down of plate
6. Weight

7. I/O interface
Count
Tare
Zero
PLU
Purchase
Refund
Storage control
Multi Customer
Calculator
Report
PLU









8. General
Max Qty of PLU: 100,200,500,1000






IV.
Cash drawer
Scanner
USB
Memory Card
PC software
Ethernet
External Printer
Power Supply
Rechargeable battery
Operating temperature
Environment humidity
Weight
Dimension
AN INTEGRATED FUZZY QFD – ANN MODEL FOR NPD
4.1. QFD
Generally QFD process consists of two stages: capturing customer requirement and the waterfall decomposition
process of the customer requirements. In the first phase, the QFD team members (experts) use all technique
available to gather or collect the customer requirements i.e., what the customer really wants and expects for the
product or model being designed, and then analyze them. This phase is the most critical and also the most
difficult part of the QFD process. In the second phase, the QFD team members use a graphic- based tool known
as House of Quality (HOQ) to translate the customer’s requirements into technical requirements. House of
Quality (HOQ) used in each translation is a chart that is made up of information on “What to do in relation to
customer attributes (CAs)”, “How CAs are related to engineering characteristics (ECs)” and relationships
between CAsand ECs. It includes among the ECs, benchmarking data, attributes’ data and prioritization
information (Bakshi, et al, 2012). In a complete QFD system, there are typically four phases: product planning,
part deployment, process planning and production planning. Each phase contains a matrix, HOQ. The
systematic procedure for the first HOQ contains seven steps (Lee and Lin, 2011):
1- Obtain customer attributes (CAs). A structured list of the Customer needs for different guilds use
Digital scales
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 665
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
2- Develop engineering characteristics (ECs). To represent the evaluating criteria that should be
considered to satisfy the customer needs.
3- Construct the pair wise comparison matrices between CAs and ECs usingSatty’s nine –point scale.
the relationship between the ithcustomer attributes (CAs) and the jthengineering
4- Compute
characteristics (ECs) in the central relationship matrix. isgeometric mean ofWeights of Fuzzy AHP
for all decision-maker participated in the decision making process.
5- Input importance of CAs(Input value is 1~5 )
6- Wj , the importance of ECs is calculated by using Eq (1) :
(1)
7- Normalize the degree of importance of technical criteria ,Wj:
(2)
QFD has been applied abundantly. For example, Armoun et al (2012) used from first matrix of QFD leading to
estimation of engineering & technical characteristics in order to enter to the quality deployment matrix. They
designed and distributed a questionnaire, which includes identification & determination of customers’ needs and
investigation of their satisfaction of manufactured products. Jaririet al (2006) suggested the use of QFD to
compare between company solution and mathematical model solution and shows the quality of the mathematical
model solution for platform of automobile design.
Because of the interrelationships among CAs and among ECs, AHP is used in some works (Chuang 2001,
Bakshi et al2012). This paper, a general super matrix approach, can tackle more complicated problems
(Saaty1996). Nevertheless, the input variables are assumed to beprecise and are treated as numerical data under
the two approaches. However, human decision making often contains ambiguity and uncertainty, and
conventional AHP is inadequate to explicitly capture the importance assessment of CAs and ECs. To confront
this problem, a new trend of studies is to incorporate fuzzy set theory into AHP-QFD .In this paper, design and
calculation of QFD matrix is done by QFD 2000 application trial version;
4.2. Fuzzy AHP
In the conventional AHP method first is developed by Saaty(Taha, Rostam, 2011). Pair wise comparisons for
each level with respect to the goal of the best alternative selection are conducted using a nine-point scale.
Different types of fuzzy membership functions have been used in fuzzy logic. However, three types are most
common: monotonic, triangular, and trapezoidal. Because the fuzzy set is a convex function, the trapezoidal
function or triangular function approaches the convex function well. The triangular fuzzy numbers are more
convenient in applications due to their computational simplicity, and they are useful in promoting representation
and information processing in a fuzzy environment.
The characteristics and membership function of the triangular fuzzy number are expressed by Eq. (3):
μ(x)
(3)
By introducing the α-cut and defining the interval of confidence at confidence level α, the triangular fuzzy
number can be characterized by Eq. (4).The α-cut is known to incorporate the experts or decision maker(s)
confidence over his/her preference or the judgments.
(4)
The degree of satisfaction can be estimated from the decision maker by index of optimism λ by Eq. (5), where
its value range is 0< λ <1. The larger the index λ, the higher the degree of satisfaction:
(5)
equationabove, and the degree of satisfaction can be estimated
The matrix is reconstructed by using the
setting the index of optimism λ and fixing α. Therefore:
=
(6)
The five triangular fuzzy numbers and their reciprocal scale are defined with the corresponding membership
function as shown in Table 3.
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 666
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
Table 3 : Definition and membership functions of fuzzy numbers
Fuzzy
numbers
Definition
Membership
function
(1,1,2)
(2,3,4)
Reciprocal scale
Equally important
(1/2,1,1)
Moderately
(1/4,1/3,1/2)
important
Strongly important
(4,5,6)
(1/6,1/5,1/4)
Very
Strongly (6,7,8)
(1/8,1/7,1/6)
important
Extremely important (8,9,10)
(1/10,1/9,1/8)
Intermediate
, , ,
values
Calculating the overall priority weight for each alternative (AW) by multiplying the vector of criteria weight
(CW) by the matrix of alternative evaluation weights (AEW) using the equation below:
(7)
Where n=number of criteria, m=number of alternatives, and k=1, 2,…, m.
In order to identify the consistency ratio of a matrix, first, the matrix consistency index (CI) is found by:
λ
(8)
The consistency index of a randomly generated reciprocal matrix with reciprocal forces is called the random
index (RI) and is calculated using the matrix order (n) and the table explained by Saaty.
So, the matrix consistency ratio (CR) is calculated using:
(9)
A consistency ratio of 0.1 or less is considered acceptable. Then compute geometric mean of Weights of Fuzzy
AHP for group decision making. To find the weights of the QFD relationship matrix by the fuzzy AHP model, a
program is developed using MATLAB.
4.3. ANN
The Artificial Neural Networks are known as the ‘‘universal approximators’’ and ‘‘computational models’’ with
particular characteristics such as the ability to learn or adapt, to organize or to generalize data (Rady, 2011). Upto-date designing a (near) optimal network architecture is made by a human expert and requires a tedious trial
and error process. On the other hand, they are simplified mathematical approximations of biological neural
networks in terms of structure as well as function.
ANNs can be classified into two major categories: supervised and unsupervised ANNs (Taha Z, Rostam S,
2011). In supervised ANNs, There is usually a decision maker who can provide some feedback in terms of
evaluating the given set of training patterns, while the unsupervised ANNs do not require the external evaluator.
Supervised learning systems are generally more flexible in the design of hidden layers. ANN architecture is
generally described as an arrangement of interconnected nodes organized into three groups input, hidden, and
output. The most commonly used approach to ANN learning is the feed-forward back propagation algorithm.
A Feed Forward ANN consists of an input layer, an output layer, and a variable number of hidden layers
(Vishwa A, .et al , 2011). The input layer is not counted normally, because it is only formally present, in the
sense that it does not do any processing: for example, a two-layer net- work consists of the input, hidden, and
output layers. All connections between the layers are allowed. Connections between the nodes of the same layer,
as well as the auto connections (loops), are prohibited. By using Back Propagation algorithm the weights in the
networks are adjusted in each iteration so as to reduce the error.
In this paper, a supervised feed forward back propagation ANN is designed, and values from QFD Model,
where the importance of CAs and ECs are determined, are then used in training stage. The designed ANN
consists of three layers: an input layer, a hidden layer, and an output layer.
The algorithm Levenberg-Marquardt back propagation (trainlm) is used to design network model for training in
MATLAB® software. This algorithm appears to be the fastest method for training moderate-sized feed forward
neural networks (up to several hundred weights). It also has an efficient implementation in MATLAB®
software, because the solution of the matrix equation is a built-in function, so its attributes become even more
pronounced in a MATLAB environment.
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 667
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
The Back propagation algorithm may be described with the following three steps, which have to be applied
several times in iteration (Rady, 2011).
1. Forward computation of input signal of training sample and determination of neural network response.
2. Computation of an error between desired response and neural network response.
3. Backward computation of the error and calculation of corrections to synaptic weights and biases.
The prepared 450 input–output sets mentioned above are then split into two parts: part one (250 sets) is used for
training the neural network and the second part (200 Sets) where they are not used in training stage, is used to
test the ANN model. At the end, Results from the ANN model are then compared with QFD model.
In the proposed model, the ANN model is used to verify the results of the QFD model and to predict the
importance of ECs. On the other hand, once the neural network is trained, it can be used to predict the ECs
ranking with any input–output set of judgments from decision maker(s).The ANN is applied using MATLAB
7.4 (R2007a) software.
V.
CASE STUDY OF DIGITAL SCALE NPD
Digital scale is a measuring device to weigh of an object or material. Digital scales are often small-scale,
sustainable, and more accurate than other types of scales.
As global demand of information technology increases, the demand of Digital Scales with high accuracy, fast
computation, with a proper keyboard and display and large data storage also increases. However, as more and
more firms enter Scale production, an extremely competitive and cost-cutting war is foreseeable.
This industry is currently one of the brilliant industries in IRAN. NPD is essential for these manufacturers to
maintain a competitive edge and to make a decent profit in a longer term. Thus, developing products that deliver
the quality and functionality customers demand while generating the desired profits becomes an important task
for the manufacturers. The case study is carried out in one Scale manufacturer in IRAN. Seven experts from the
firm are asked to contribute their expertise in the study.
5.1. Stage I: The selection of QFD factors (CAs & ECs)
To recognize the main needs of customers (CAs), some questionnaires were distributed among 30 of them. As
Table.1, only 11 main needs were propounded out of 17 presented needs. These 11 needs were considered more
important in designing new products by experts and the results were obtained as you see in Fig3.
According to the results, 7 needs are very significant for customers about choosing and buying the scales. Oneof
them, Manual book or learning CD has no influence on determining the attributes of the new products. The 6
important needs of the customers to determine the technical features of the new product are as follows: The
most suitable weigh, high accuracy, clear and readable display, appropriate and resistant keyboard, enough
Memory (PLU), immediate and accurate after sales service.
Determin the main needs of customers
18
16
14
12
10
8
6
4
2
0
Unimportant
Slightly important
Moderately important
Strongly important
Extermely important
Fig3: The Histogram of comparing importance of CAs about scale
To determine the engineering characteristics (ECs)in providing the customer's needs, other questioners were
prepared and distributed among 25 active sales agents who were chosen by the manager of the company. Only
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 668
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
15 sale agents replied. According to the most chosen items, the following results to determine the characteristics
of the product were obtained. These features are presented in table 4.
Table 4 : The product 's features
Max weight
30,50 kg
Accuracy
5,10 g
Display
LED Up, LED down
Keyboard
Press Down, Flat Down
PLU
100, 1000
5.2. Stage II: The ranking of factors using the FAHP-QFD
The Combination of CAs and ECs made QFD matrix and it's shown in Table 5. Seven experts on Scale
production participated in the filling Fuzzy comparison matrix. A questionnaire based on the proposed hierarchy
structure was formulated using fuzzy numbers. The next Step for the decision makers are assigning the
preference score for the evaluation ECs with respect to each evaluation
After the data have been collected from the decision makers, 48 matrices for the ECs' comparisons are built.
The priority weights of the selected features based on fuzzy AHP model are determined using the program in
MATLAB Software as follows:
Table 5: QFD ( house of quality )
engineering characteristics
Max
weight
50
30 kg
kg
MatrixQFD
Accuracy
5g
Display
LED
Up
10g
Keyboard
LED
down
Press
Down
PLU
Flat
Down
100
1000
Suitable Max weight
High accuracy
Readable display
appropriate and resistant
keyboard
enough Memory (PLU)
After sales service
Step1. Preparing the input data to program, the first decision maker's preference scores are presented. (Table 6)
Table 6: comparison matrix for the ECs with respect to the first Customer attribute-Suitable Max weight
Max weight
50
kg
30 kg
Max
weight
Accuracy
Display
Keyboard
PLU
Accuracy
5g
10g
Display
LED
Up
LED
down
Keyboard
PLU
Press
Down
Flat
Down
100
1000
30 kg
1
1
5
5
5
5
5
5
5
5
50 kg
5g
1
1
5
5
5
5
5
5
5
5
0
0
1
1
1
1
1
1
1
1
10g
0
0
0
1
1
1
1
1
1
1
LED Up
0
0
0
0
1
1
1
1
1
1
LED down
0
0
0
0
0
1
1
1
1
1
Press Down
0
0
0
0
0
0
1
1
1
1
Press Down
0
0
0
0
0
0
0
1
1
1
100
0
0
0
0
0
0
0
0
1
1
1000
0
0
0
0
0
0
0
0
0
1
Step2. The weights ofECs for the first decision maker are shown in Table.7. If CI
showed error, the related matrix would refill by the decision maker
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
0.1, the program is
Page | 669
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
Table 7: ECs' weight for the first decision maker's judgment with respect to the first Customer attribute-Suitable Max weight
LED
LED
Press
Flat
100
1000
Up
down
Down
Down
0.5
0.5
0
0
0
0
0
0
0
0
Step3. Repeating the steps from (1) to (2) for the remaining decision makers (number 2 to number 7).
30 kg
50 kg
5g
10g
The final results for the ECs' weights for the seven decision makers are shown in Table 8.In this investigation,
the results obtained from the previous steps (fuzzy AHP model) are used to denote the relationship between
Whats and Hows.
Table 8: ECs' weights of for all decision makers with respect to the first Customer attribute-Suitable Max weight
Max weight
30 kg
50 kg
DM1
0.4693
DM2
0.5360
DM3
Accuracy
Display
LED
Up
Keyboard
LED
down
Press
Down
Flat
Down
PLU
100
1000
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
5g
10g
0.5307
0
0
0
0
0
0.4640
0
0
0
0
0
0.5300
0.4700
0
0
0
0
DM4
0.5325
0.4675
0
0
0
DM5
0.4671
0.5329
0
0
DM6
0.5
0.5
0
DM7
Overall
weight
0.5
0.5
0.5042
0.4942
To use the QFD 2000 software, we determined and entered the quantities in the Software as table 9. The HoQ
matrix is then developed as depicted in Fig.4.
Table 9: Determine importance of HoQ matrix according to results of fuzzy AHP model
Relation
Mark
Overall weight of fuzzy AHP model
Strong
0.75
~
1

Medium
 0. 5 ~ 0.75
Weak
 0. 25 ~ 0. 5
No relation
0 ~ 0. 25
Fig 4: HoQ matrix for the evaluation alternatives
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 670
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
The results from the HoQ matrix are used to design and train the proposed ANN model. The next steps are to
find the alternatives weights and ranking using the ANN model as follows:
Step 1. The importance weights of 6 CAs are used for input values of the ANN model and the relative weight
total of the 4 alternatives are used as a target.
Step 2. Designing the model using Levenberg-Marquardt back propagation algorithm with different number of
hidden nodes, seven (7), and ten (10).
Step 3. Training the model with same training parameters learning rate, and Validation rate using different of epoch
(Table 10). The mean square error (MSE) value is used as the stop criteria.
Table 10: ANN model parameters
Input
Nodes
6
6
1
2
Hidden
Nodes
7
10
Output
Nodes
10
10
Training
Function
Trainlm
Trainlm
Learning
rate
0.7
0.7
Validation
rate
0.15
0.15
Test
rate
0.15
0.15
Epoches
Performance
160
28
0.9973
0.9985
Step 4. Use 250 samples for training and 200 samples for testing.
From Table 10, it is clear that the best performance for (6-10-10) model. The used models has been achieved
with the Trainlm Training function. As Fig 5, 6, the regression line is closed to a line with a one by one slope in
a high degree. The degree of closeness in training data in much higher than testing and evaluating data, and it
shows the accuracy of the designed network for HoQ matrix.
Fig 5: Regression of Training for (6-10-10) model
Fig 6: Regression of Testing for (6-10-10) model
The comparison of the outputs from the Fuzzy QFD model and from the ANN model is made for 5 input
samples, and the result is shown in Table 11. From the table 11, one can obviously observe that the ranking of
the alternatives for both fuzzy AHP-QFD model and ANN model are the same. These results clearly show the
accuracy and power of the proposed fuzzy AHP –QFD which is based on the developed a Fuzzy AHP program
and the ANN model. So, the proposed decision support system by combining the fuzzy QFD and ANN in this
work can be used as an active tool to select the most suitable Specification for new product to satisfy customer's
needs.
Table 11: Comparison between combined fuzzy AHP–QFD, and ANN methods for 5 input samples
Max weight
No
1
2
Output
Accuracy
Display
Keyboard
PLU
30 kg
50 kg
5g
10g
LED Up
LED down
Press Down
Flat Down
100
1000
QFD
9.00
4.00
9.00
4.00
17.00
6.00
12.00
33.00
4.00
4.00
Net
9.99
3.95
10.87
4.14
16.39
6.00
10.75
30.65
3.82
3.82
QFD
15.00
6.00
8.00
4.00
15.00
6.00
10.00
28.00
4.00
4.00
Net
15.34
6.03
8.89
3.77
15.04
5.91
10.12
27.72
3.86
3.86
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 671
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
3
4
5
QFD
19.00
8.00
8.00
4.00
13.00
6.00
10.00
25.00
4.00
4.00
Net
19.42
7.72
8.00
3.74
13.66
5.77
9.53
24.98
3.94
3.94
QFD
22.00
9.00
7.00
4.00
12.00
6.00
9.00
22.00
4.00
4.00
Net
22.75
9.12
7.64
3.84
12.29
5.56
9.00
22.61
4.02
4.02
QFD
25.00
10.00
7.00
4.00
12.00
6.00
9.00
21.00
4.00
4.00
Net
26.76
10.57
6.40
3.45
10.97
5.23
8.59
20.97
4.00
3.82
VI.
CONCLUSIONS
Customer needs and wants are sensitive and complex. If a firm can understand them and make efforts to fulfill
customer demands and provide friendly service, then customers will be more supportive and loyal to the
enterprise. During the process of development from the product concept to the actual product, the customer can
only passively receive new information, and can only select from the products that are currently on sale in the
market. No matter which type of product, the customer cannot individually come up with a product concept and
then develop it.
Furthermore, buying what is available on the market does not mean that customers are satisfied with the current
product, because the customer’s experiences and preferences were not considered in developing the product so
they can only accept the product as it is. As a result, a business should develop products that fulfill the
customer’s needs and wants, since this will increase the enterprise’s competitiveness and it is an essential
criterion to earning higher loyalties and profits.
In this research, a systematic framework that incorporates Fuzzy AHP into QFD is proposed for new product
development and learning machine made for this frame work by ANN model. Through literature review and
interview with experts, lists of factors are prepared, including customer attributes, engineering characteristics,
parts characteristics, key process operations and production requirements.
The results are used to construct the first houses of QFD, and the priorities of the factors can be calculated
through FAHP. Then training and testing ANN with input-output of HoQ matrix to verify the results of the QFD
model and to predict the importance of ECs. The proposed framework is examined by a case study in a digital
scale manufacturer in IRAN. The results show the important customer attributes, engineering characteristics,
parts characteristics, key process operations and production requirements in designing a new product. Those
factors with higher priorities should especially be focused on. In conclusion, the proposed model can help a firm
systematically consider relevant NPD information and effectively determine key specifications for designing
and manufacturing of new products.
REFERENCES
[1]
Armoun Z, Javidnia M, Nikkhah Farkhani Z, Nasiri S,2012,Utilizing QFD model to determine quality
characteristics of the products and priority needs of customers in the medical industry products (Case
Study: Plasma seat product in mashhad`s Sahateb medical equipment company), Management Science
Letters 2, 2525–2536
[2]
Bae J.K, Kim J, 2011, Product development with data mining techniques: A case on design of digital
camera, Expert Systems with Applications 38, 9274–9280
[3]
Bakshi T, Sarkar B, Sanyal K, 2012, A Novel Integrated AHP-QFD Model for Software Project Selection
under Fuzziness, International Journal of Computer Applications (0975 – 8887), c 54– No.7
[4]
Belliveau, P., Griffin, A. and Somermeyer, S. (2002) The PDMA toolbook for new product development
,Wile, New York.
[5]
Büyüközkan G, Kahraman C, Ruan D (2004) A fuzzy multi-criteria decision approach for software
development strategy selection. Int J Gen Syst 33(2–3):259–280
[6]
Chan, Sze-ling,2011, An integrated decision support system for new product development with customer
satisfaction, Pao Yue-kong Library Electronic Theses Database, Dept. of Industrial and Systems
Engineering Pages: xiii, 196 p. : ill. ; 30 cm.
[7]
Chuang, 2001, Combining the Analytic Hierarchy Process and Quality Function Deployment for a
Location Decision from a Requirement Perspective, International Journal Advance Manufauctring
Technology, 18,842–849, 2001 Springer-Verlag London Limited
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 672
Khadijehyousefipourjeddi* et al. / (IJITR) INTERNATIONAL JOURNAL OF INNOVATIVE TECHNOLOGY AND RESEARCH
Volume No.2, Issue No. 1, December – January 2014, 663 - 673.
[8]
Ertay T, Büyüközkan G, Kahraman C, Ruan D (2005) Quality function deployment implementation
based on analytic network process with linguistic data: an application in automotive industry. J Intell
Fuzzy Syst 16:221 232
[9]
IEC 60300-1/ISO 9000-4 (1991) Dependability management; Part 1: Dependability assurance of
products, International Electrotechnical Commission, Geneva.
[10]
Lee AHI, Wang WM, Lin TY (2010c) An evaluation framework for technology transfer of new
equipment in high technology industry. Technol Forecast Soc Change 77:135–150
[11]
Lee and Lin ,2011, An integrated fuzzy QFD framework for new product development, Flexible Services
and Manufacturing Journal ,Springer Science Business Media, LLC 2011 , 10.1007/s10696-011-90765
[12]
Jariri F , Zegordi S.H, 2006, Quality function deployment planning for platform design, Int J Adv Manuf
Technol 36:419–430 , DOI 10.1007/s00170-006-0853-3
[13]
Jie L, Yijun Z, Xianyi Z, Koehl L, Jun M, Guangquan Z,2008,A Fuzzy Decision Support System for
Garment New Product Development, AI 2008 Advances in Artifical Intelligence 21st Australasian Joint
Conference on Artificial Intelligence Issue: Mcdm, Pages: 532-543
[14]
Kahraman C, Ertay T, Büyüközkan G (2006) A fuzzy optimization model for QFD planning process
using analytic network approach. Eur J Oper Res 171:390–411
[15]
Kumar A, Said G, Arnold R (2007) Mass customization research: trends, directions, diffusion intensity,
and taxonomic frameworks. Int J Flex Manuf Syst 19:637–665
[16]
Liberatore and Stylianou, 1995, toward a framework for developing knowledge-based decision support
systems for customer satisfaction assessment: An application in new product development, Expert
Systems with Applications, Volume: 8, Issue: 1, Pages: 213-228
[17]
Oduoza C.F., Harris A,2011, A Decision System to Support Product Development: Decision Support for
Product Development, LAP LAMBERT Academic Publishing (November 11, 2011)
[18]
Østeras T, D.N.P. Murthy and M. Rausand, 2004, Reliability Performance and Specifications in New
Product Development, Available at: http://www.ntnu.no/ross/reports/ReliabilitySpecifications.pdf
Pahl, G. and Beitz, W. (1988) Engineering design, The Design Council, Springer Verlag, London.
[19]
Pugh, S. (1991) Total design. Integrated methods for successful product engineering, Addison-Wesley,
Wokingham
[20]
Rady H.A.K, 2011, Shannon Entropy and Mean Square Errors for speeding the convergence of
Multilayer Neural Networks: A comparative approach, Egyptian Informatics Journal, 12, pp: 197–209
[21]
Taha Z, Rostam S, 2011, A fuzzy AHP–ANN-based decision support system for machine tool selection
in a flexible manufacturing cell, Int J Adv Manuf Technol, 57:719–733, DOI 10.1007/s00170-011-33235
[22]
Saaty TL (1996) Decision making with dependence and feedback: the analytic network process. RWS
Publications, Pittsburgh
[23]
Shahin A, Poormostafa M, 2011, Facility Layout Simulation and Optimization: an Integration of
Advanced Quality and Decision Making Tools and Techniques, Modern Applied Science, Published by
Canadian Center of Science and Education, Vol. 5, No. 4;page 95-111
[24]
Sounder, W.E. (1987) Managing new product innovations, Lexington Books, Lexington, MA.
[25]
Vishwa A, Vishwa Alka, Sharma A,2011, Pre-Diagnosis Of Lung Cancer Using Feed Forward Neural
Network And Back Propagation Algorithm, International Journal on Computer Science and Engineering
(IJCSE), Vol. 3 No. 9 , pp: 3313-3319
[26]
Wesner, J.W., Hiatt, J.M. and Trimble, D.C. (1995) Winning with quality: Applying quality principles in
product development, Addison-Wesley, Reading, MA.
[27]
Wind, Y.J. (1982) Product policy: Concepts, methods and strategy, Addison-Wesley, Reading, MA.
ISSN 2320 –5547
@ 2013 http://www.ijitr.com All rights Reserved
Page | 673