Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Engineering Applications of Artificial Intelligence

journal homepage: www.elsevier.com/locate/engappai

AI-based methodology of integrating affective design, engineering,

and marketing for defining design specifications of new products
Kwong C.K.a,n, Jiang Huimina, Luo X.G.a,b
a
Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hong Kong, PR China
b
Department of Systems Engineering, School of Information, Northeastern University, Shenyang, PR China

art ic l e i nf o a b s t r a c t

In the early stage of product design, particularly for consumer products, affective design, engineering,
Keywords: and marketing issues must be taken into considerationand they are commonly performed respectively
Affective design by product designers, engineers, and marketing personnel. However, they have different concerns and
Marketing focuses with regard to the new product design. Thus, these three processes are commonly conducted
NSGA-II separately, leading to a sub-optimal and even sub-standard design. Such scenario indicates the need to
Fuzzy regression incorporate the concerns of the three processes in the early stage of product design. However, no study
Chaos optimization algorithm has explored the incorporation of the concerns of the three processes into the product design. In this
paper, an artificial intelligence (AI)-based methodology for integrating affective design, engineering, and
marketing for defining design specifications of new products is proposed by which the concerns of the
three processes can be considered simultaneously in the early design stage. The proposed methodology
mainly involves development of customer satisfaction and cost models using fuzzy regression,
generation of product utility functions using chaos-based fuzzy regression, formulation of a multi-
objective optimization model and its solving using a non-dominated sorting genetic algorithm-II (NSGA-
II). A case study was conducted for electric iron design to evaluate the effectiveness of the proposed
methodology.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction customers and influence their choices and preferences, such as

loyalty to the company and joy of use (Creusen and Schoormans,
Affective design, engineering, and marketing concerns are 2005; Noble and Kumar, 2008).
always considered in the early design stage of consumer products. Various engineering concerns have to be considered by design
Affective design is commonly performed by product designers, engineers during the product development stage such as product
which is about the analysis of customer reactions toward candi- functionality, technical specification, structural performance,
date designs and the quantification of such reactions and their material selection and design for manufacture. Among the engi-
integration into physical product design parameters for maximiz- neering concerns, one of the major concerns that needs to be
ing customer affective satisfaction on the new product to be addressed in the early product design stage is to define technical
developed (Barnes and Lillford, 2009). It can help designers specifications of new products such that customer satisfaction on
generate designs that better appeal to the markets. Affective the new products can be maximized. To define the specifications,
design has been shown to excite customers’ psychological feelings design engineers have to consider various issues, such as engi-
and can help improve customer satisfaction in terms of emotional neering performance of competitive products and difficulties in
aspects. It involves the processes of identifying, measuring, ana- attaining high target values of engineering requirements. Compe-
lyzing, and understanding the relationship between the affective titive product benchmarks are commonly used to help determine
needs of the customer domain and the perceptual design attri- the technical specifications of new products in industries. On the
butes in the design domain (Lai et al., 2005a). Design attributes, other hand, various marketing concerns relating to new product
such as shape and color, evoke the affective responses of custo- development, such as market opportunities, new product posi-
mers to products. Products with good affective design can attract tioning, competitors’ performance, price positioning, and custo-
mer needs, have to be considered by marketing personnel in the
early product design stage who are mainly concerned with the
n
Corresponding author. Tel.: þ 852 27666610; fax: þ852 23625267. market share, profit and degree of customer satisfaction to be
E-mail address: [email protected] (C.K. Kwong). obtained by launching a new product (Crawford and Benedetto,

http://dx.doi.org/10.1016/j.engappai.2015.04.001
0952-1976/& 2015 Elsevier Ltd. All rights reserved.

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
2 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎

2006; Luo et al., 2005). Therefore, product designers, engineers, One important task of the Kansei engineering framework is to
and marketing personnel have different goals and concerns with evaluate the relationships between defined affective dimensions
regard to new product development and some of their concerns and design attributes. Previous studies on Kansei engineering have
are interrelated. For example, in the determination of the screen applied various regression analyses, including multiple linear
size of a new smartphone, the product designer considers the regression (Han et al., 2000), quantification theory I (You et al.,
issues of portability, ergonomics and user-interface and subse- 2006; Chang, 2008), ordinal logistic regression (Barone et al.,
quently, sets the screen size as 4.5 in. Marketing personnel found 2007), partial least squares analysis (Nagamachi, 2008), and
that their new smartphones with screen size 5 in. are essential to multilevel regression (Seva et al., 2007). Various computational
compete with the competitive smartphones, whereas the design intelligence techniques have been attempted to model the rela-
engineer considers the constraints of product cost and weight and tionships such as artificial neural networks (Hsiao and Huang,
concludes that the screen size should be set as 4 in. In view of the 2002; Lai et al., 2005b), radial basis function neural networks
different concerns of designers, marketing staff and engineers, (Chen et al., 2003), fuzzy rule-based modeling (Park and Han,
therefore, a coordinating mechanism/framework or a methodol- 2004), fuzzy expert system with gradient descent optimization
ogy for the simultaneous consideration of their concerns is (Lau et al., 2006), and fuzzy neural networks (Sun et al., 2000; Tsai
required in the early product design stage such that the best et al., 2006). Lin et al. (2007) presented a new fuzzy logic approach
setting of design variables can be determined. However, no for consumer-oriented product form design in a case study of
previous study has proposed such framework or methodology. mobile phones. Tanaka’s fuzzy regression (Sekkeli et al., 2010) and
To fill the research gap, an artificial intelligence (AI)-based meth- genetic programming based fuzzy regression (Chan et al., 2011)
odology for integrating the affective design, engineering, and have been proposed to model affective relationships. Recently, the
marketing for defining design specifications of new products is adaptive neural fuzzy inference system (ANFIS) has been applied
proposed by which the concerns of affective design, engineering, to generate nonlinear and explicit customer satisfaction models
and marketing can be considered simultaneously in the early and fuzzy rules based on market survey data for new product
product design stage. The proposed methodology mainly involves design (Kwong et al., 2009).
a fuzzy regression (FR) approach for modeling customer satisfac- In the early product design stage, one of the key tasks in
tion and developing cost models, a chaos-based FR approach for undertaking affective design is to determine the optimal settings
generating product utility functions, and a non-dominated sorting of the design attributes for the affective aspects of the products to
genetic algorithm-II (NSGA-II) for solving multi-objective optimi- achieve maximum customer satisfaction. Various techniques have
zation problems. The rest of this paper is organized as follows. A been attempted to determine the optimal settings such as conjoint
review of related research is presented in Section 2. The proposed analysis (Shi et al., 2001), multiple response surfaces methodology
methodology is described in Section 3. Section 4 describes a case (Hong et al., 2008), ordinal logistical regression (Aktar Demirtas
study of defining the design specification of a new electric iron et al., 2009) and genetic algorithms (Hsiao and Tsai, 2005; Kim and
based on the proposed methodology. Finally, the conclusion is Cho, 2000). Although numerous studies were conducted on
given in Section 5. affective design, simultaneous consideration of affective design
and marketing/engineering issues was not addressed.

2. Literature review 2.2. Integrated marketing and engineering for product design

Although no publication was found thus far regarding the Some recent research attempting to link up marketing and
integration of affective design, engineering, and marketing con- engineering concerns for new product development have been
cerns for product design, quite a number of previous studies have conducted. The studies mainly involved market surveys, demand
explored affective design as well as the integration of engineering modeling techniques, cost modeling, product performance model-
and marketing for product design. ing, and optimization model formulation and solution. The objec-
tive of the optimization is to maximize profits and/or product
2.1. Affective design performance. Some techniques were employed to coordinate
marketing and engineering design problems in order to yield a
Nagamachi (1995) proposed Kansei engineering, also known as joint optimal solution such as analytical target cascading
affective engineering, which is a product development methodol- (Michalek et al., 2005), multi-objective genetic algorithms (Luo
ogy that uses quantitative methods to acquire and transform et al., 2005; Besharati et al., 2006) and game-theoretic model
customer affections into design attribute settings using quantita- (Shiau and Michalek, 2009) Kang et al. (2007) proposed a
tive methods. It can be performed by analyzing customers’ Kansei methodology of marketing and R&D integration for new product
and translating how the design matches the Kansei, collecting development that involves the methods of quality function
customers’ Kansei experience and establishing mathematical pre- deployment (QFD), multivariate statistical analysis, conjoint ana-
diction models that relate the Kansei to the design attributes lysis, and the Taguchi method. Kwong et al. (2011a) proposed a
(Lokman, 2010; Marghani et al., 2013). Surveys are always required methodology of integrating marketing with engineering for defin-
in Kansei engineering, which is used to analyze the affective ing design specifications of new products using factor analysis,
meanings related to a product domain based on semantic differ- Kano model, and genetic algorithm (GA). Williams et al. (2011)
ential (SD) method (Chuang and Ma, 2001). Kansei engineering proposed a strategic framework, which involves marketing, stra-
has been applied in various affective product designs, such as tegic design and engineering design to understand how different
automobiles (Zhang and Wang, 2013), drink bottles (Barnes and retail channel structures impact the engineering design of the new
Lillford, 2009), and surface tactility of plastic products (Choi and product and to determine the optimal design under different
Jun, 2007). The framework of Kansei engineering encompasses channel structure conditions. All the above studies only address
four tasks (Barnes and Lillford, 2007; Nagamachi, 1995), namely, the integration issue for a single product design.
definition of the product domain, determination of the dimensions Quite a few previous studies were conducted to integrate
of customer affections, determination of design attributes and marketing with engineering concerns for the design of multi-
attribute options, and evaluation of relationships between custo- products. Some previous studies developed methodologies of
mer affections and design attributes. integrating marketing with engineering concerns for product

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

family design (Kumar et al., 2009; Wang et al., 2013) while some 3.2. Modeling of customer satisfaction using fuzzy regression
others aimed to consider marketing and engineering concerns
simultaneously for product line design (Michalek et al., 2006, In this research, Tanaka’s FR (Tanaka, 1987; Liu et al., 2015) is
2011; Luo, 2011). Jiao and Zhang (2007) employed conjoint applied to model customer satisfaction. Before conducting a
analysis and multi-nomial logit choice models to formulate a survey for affective design, product samples need to be identified
product portfolio planning problem, which links the marketing and collected first. Then, affective dimensions and design attri-
concerns with the determination of product specifications and butes are defined. The semantic differential (SD) method is
manufacturing cost for product portfolio planning. It can be noted adopted in this research to design an SD questionnaire for
that various methodologies were developed in previous studies to collecting the affective responses of customers on products. Based
integrate marketing with engineering concerns for the design of on the survey data, FR approach is employed to model the
single product and multi-products. However, development of relationships between the affective dimensions and design attri-
methodologies for simultaneous consideration of affective design, butes. To model the relationships between the dimensions of
marketing and engineering concerns for product design was not customer satisfaction and engineering requirements, a competi-
addressed. tive product benchmark has to be conducted first. Based on the
benchmark results, FR is introduced to model the relationships. In
the following, a brief description of Tanaka’s FR is provided. In
3. Proposed methodology Tanaka’s FR, fuzzy coefficients with the central point ac and the
spread value as are determined by solving the following linear
In this paper, an AI-based methodology of integrating affective programming (LP) problem:
design, engineering, and marketing for defining design specifica- !
Xn X
M  
tions of a new product is proposed. In the proposed methodology, Min J ¼ as xij  ð1Þ
j
customer satisfaction models and cost model are generated based j¼0 i¼1
on FR approach. A conjoint survey is conducted and utility
functions are then developed using a chaos-based FR approach. where J is the objective function that represents the total fuzziness
Subsequently, the market share model is developed using the of the system, 1 þ n is the number of terms of the fuzzy polynomial
utility functions and the MNL model. The profit model is devel- model, M is the number of data sets; xij is the jth independent
oped based on the market demand and cost model. Then, a multi- variable of the ith data, and j:j refers to absolute value of the
objective optimization model is formulated with the objectives of independent variable. The constraints can be formulated as
maximizing the market share and maximizing profit. Finally, follows:
NSGA-II is adopted to solve the optimization problem and a set X
n X
n  
of nondominated solutions of design specifications can be acj xij þ ð1  hÞ asj xij  Z yi þ ð1  hÞei ð2Þ
j¼0 j¼0
obtained. Fig. 1 shows a flowchart of the proposed methodology.
X
n X
n  
3.1. Conjoint survey and lead user survey acj xij  ð1  hÞ asj xij  r yi  ð1  hÞei
j¼0 j¼0

Rating-, ranking-, and choice-based conjoint surveys are the asj Z 0; acj A R; j ¼ 0; 1; 2; …n
three types of conjoint survey designs. The rating-based conjoint xi0 ¼ 1 for all i; i ¼ 1; 2; …; M and 0 r hr 1 ð3Þ
survey is widely used in previous studies and requires a set of
product profiles with respect to pre-defined attributes and attri- where h, referring to the degree to which the fuzzy model fits the
bute levels (Kazemzadeh et al., 2009). In this paper, a rating-based given data, is between 0 and 1; yi is the value of the ith
conjoint survey is conducted which contains two parts. The first dependent variable in the data sets; and ei is the spread value of
part is to study the consumer perception of various dimensions of the ith dependent variable. Constraints (2) and (3) set the upper
customer satisfaction on products, and the second part is to study and lower boundaries of the estimated data, respectively.
the affective satisfaction of consumers on products. To design the To evaluate the performance of FR, two criterions, the mean
first part of the survey, the dimensions of customer satisfaction, absolute percentage error (MAPE) and index of confidence (IC), are
such as quality and functionality, have to be identified first. Then, a introduced. MAPE is calculated as follows:
survey questionnaire is designed based on orthogonal arrays, 
M ~

1 X y i  yi 
which contain a number of product profiles. Consumers are then MAPE ¼ n100 ð4Þ
M i ¼ 1 yi
asked to rate the product profiles. The results of the survey are

used to generate product utility functions based on chaos-based where yi is the ith predictive output based on the genera-
FR. For the other part, the process of survey design is similar to ted model.
that of the former one, except that the design attributes of For the fuzzy outputs, an IC was introduced by Wang and Tsaur
products and affective dimensions have to be identified instead (2000) and is defined by Eq. (5).
of dimensions of customer satisfaction of products. The results of
SSR SSE
this survey are used to generate the affective customer satisfaction IC ¼ ¼ 1 ð5Þ
SST SST
models using FR.
Besides the conjoint survey, a lead user survey is also con- where SST is the total sum of squares, which considers the
ducted to perform competitive product benchmarking and gen- measure of the variation between the upper bound and lower
erate price models. In the survey, the product specifications of bound of the prediction h-certain interval;
competitive products are shown and lead users are asked to rate 0 0 112
X
M X
n
individual ones with respect to various dimensions of customer SST ¼ @ yi  @ ðaj  ð1  hÞaj Þxij AA
c s
satisfaction. Lead users are also asked to express their views on i¼1 j¼0
several price levels from a scale of 1 to 4, which means low, 00 1 12
medium, high, and very high, respectively. Using the collected data X
M X
n
þ @@ ðacj þ ð1 hÞasj Þxij A  yi A ð6Þ
sets, the utility function of price can be developed using poly-
i¼1 j¼0
nomial modeling.

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
4 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Information of competitive products

Design and conduct a conjoint Conduct a lead user survey

3.1 3.1
survey for affective design

Generate customer Generate customer satisfaction models, Generate

satisfaction models and and correlation a price
using fuzzy regression for models, , using fuzzy regression based model,
affective design on competitive product benchmark
3.2
3.2

Generate a cost
Generate a product utility function, , using
3.3 model, , using 3.5
chaos-based fuzzy regression approach
fuzzy regression

Develop a market share Develop a profit

Develop a market 3.5
model, , using MNL 3.4 model,
demand model, 3.4

Formulate a multi-objective optimization model with the

3.6
objectives of maximizing and
Estimated market
potential,

Solve the optimization problem using NSGA-П 3.7

Optimal settings of design

specification of new products

Fig. 1. AI-based methodology of integrating affective design, engineering, Note: Sections 3.1–3.7 are the section numbers of this paper.

SSR is the regression sum of squares, which is the variation of judgments, leading to a high degree of fuzziness of the survey data.
the prediction interval with respect to the center regression line; In this paper, a chaos-based FR approach is introduced to develop
00 1 0 112 utility functions. In the proposed approach, chaos optimization
XM X n X
n
algorithm (COA) is introduced to generate the nonlinear polynomial
SSR ¼ @@ acj xij A  @ ðacj  ð1  hÞasj Þxij AA
i¼1 j¼0 j¼0
structures of utility functions that could contain second- and/or
00 1 0 112 higher-order and interaction terms. COA employs chaotic dynamics
X
M X
n X
n to solve the optimization problem, which does not rely on learning
þ @@ ðacj þ ð1  hÞasj Þxij A  @ acj xij AA ð7Þ factors and has been demonstrated to have faster convergence and
i¼1 j¼0 j¼0
can search more accurate solutions than the conventional optimiza-
SSE is the error sum of squares, which is formulated by Eq. (8). tion methods. The FR method is employed to determine the fuzzy
0 0 11 2 coefficients for all the terms of the utility function. An example of a
XM X
n product utility function developed based on the chaos-based FR
SSE ¼ 2n @ yi  @ c AA
aj xij ð8Þ approach is shown as follows:
i¼1 j¼0

Also, U ¼ A~ 0 þ A~ 1 x1 þ A~ 2 x1 x2 þ A~ 3 x3 2 þ ⋯ þ A~ n  1 ðx1 x2 ⋯xn  1 Þ þ A~ n xn n
ð10Þ
SSE ¼ SST  SSR ð9Þ
or
IC is similar to the determinant confident (R2 ) in classical
regression. The higher value of IC implies a better estimation of         
U ¼ ac0 ; as0 þ ac1 ; as1 x1 þ ac2 ; as2 x1 x2 þ ac3 ; as3 x3 2 þ ⋯
yi (Azadeh et al., 2011).  c   
þ an  1 ; asn  1 ðx1 x2 ⋯xn  1 Þ þ acn ; asn xn n ð11Þ

3.3. Generation of product utility functions using chaos-based fuzzy 

where U is the dependent variable, which  arethe ratings
 of the
regression respondents on the product profiles; A~ 0 ¼ ac0 ; as0 , A~ 1 ¼ ac1 ; as1 , ⋯,
 
A~ n ¼ acn ; asn are the fuzzy coefficients in which ac and as are the
In the conjoint survey to study the dimensions of the customer central value and the spread of fuzzy numbers, respectively; x1  xn
satisfaction of products, respondents are invited to assess all the and are the independent variables. Details of the chaos-based FR
product profiles using the different ratings. Unavoidably, respon- approach to generate the utility function can be found in the
dents’ ratings on the product profiles involve their subjective authors’ publication (Jiang et al., 2013).

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5


3.4. Market share model eUi
Objective 2 : Max Pf ¼ MpP  PK   ðPrc  CÞ
T
The market share model is developed based on the generated t¼1 eU t þ k¼1 eU k þ e U i
product utility function and MNL model. In the recent years, ð17Þ
probabilistic choice rules have turned out to be more realistic in Subject to:
representing the customer behavior of purchase decision-making.
Some probabilistic choice rules can offer flexibility in calibrating yA ¼ f A ðxÞ ð18Þ
the actual choice behavior, such as the option of mimicking the
yB ¼ f B ðxÞ ð19Þ
first choice rule (Kaul and Rao, 1995). A widely-used probability
rule is the MNL choice rule. The estimation of the probability of yP ¼ f P ðPrcÞ ð20Þ
choosing the ith product among the company’s existing and

competitive products is obtained as follows (Aydin et al., 2014): U ¼ f U  ðyA ; yB ; yp Þ ð21Þ


eUi C ¼ f c ðxÞ ð22Þ

Pr i ¼ P  PK   ð12Þ
T
t¼1 eU t þ k ¼ 1 e þe
Uk Ui
xj min rxj r xj max ð23Þ
where Pr i is the choice probability,

indicating how a customer will
likely choose

the ith product; U i is the utility value of the ith xj ¼ gðxj1 Þwhere j a j1 ð24Þ
product; U t is the utility of the tth competitive product; and U k is
where (18) are the customer satisfaction models for affective
the utility of the kth company’s existing product. The utility
design. (19) are the models to relate customer satisfaction with
function values and the choice probabilities are considered iden-
engineering requirements based on a competitive product bench-
tical across all customers. Therefore, the market share, Ms, can be
mark. Both (18) and (19) are generated based on FR approach. (20)
treated as the individual choice probability.
is the utility function of price. (21) is the product utility function

developed based on the chaos-based FR approach. (22) is the cost
eUi
Ms ¼ P  PK   ð13Þ model generated using FR approach. xj is the jth design variable;
T
t¼1 eU t þ k¼1 eU k þe U i (23) are the ranges of the settings of the design variables. (24) is
Hence, the market demand of a new product, Md, can be the correlation models for relating the jth design variable and
estimated as follows: some other design variables. For example, in the electric iron
design, the weight has a correlation with the water tank volume


eUi and soleplate material.

Md ¼ MpnMs ¼ MpP  PK   ð14Þ
T
t¼1 eU t þ k¼1 eU k þe U i
3.7. Solving the optimization model using a NSGA-II approach
where Mp is market potential that is commonly estimated by
marketing personnel based on their knowledge and judgments. NSGA-II is an elitist genetic algorithm and is commonly used to
solve multi-objective optimization problems. The major features of
NSGA-II include low computational complexity, parameter-less
3.5. Cost and profit models diversity preservation, elitism, and real-valued representation
(Deb et al., 2002; Haghighi and Asl, 2014). NSGA-II uses a real-
In the developed competitive product benchmark, competitive coded simulated binary crossover (SBX) operator and a real-coded
products and their engineering performance are identified. Cost- polynomial mutation operator to support the crossover and
ing engineers of companies are invited to estimate the product mutation operations directly for real-valued decision variables.
costs of individual competitive products assuming that the pro- Deb et al. (2002) found that NSGA-II was able to maintain a better
ducts are designed and produced by their own companies. The spread of solutions and had better convergence than other multi-
estimated costs together with the engineering performance data objective genetic algorithms, such as Pareto-archived evolution
can be used to generate a cost model, C, using FR. Thus, the profit, strategy (PAES) and strength-Pareto evolutionary algorithm
Pf , of the new product can be estimated as follows: (SPEA). A flowchart and the algorithms of NSGA-II are shown in
 Appendix.
eUi
Pf ¼ MdnðPrc  CÞ ¼ MpP  PK   ðPrc CÞ ð15Þ
T
t¼1 eU t þ k¼1 eU k þ e U i
4. Case study
where P rc is the price of the new product.
A case study of electric iron design was conducted based on the
proposed methodology to evaluate its effectiveness.
3.6. Formulation of an optimization model
4.1. Conjoint survey and generation of product utility function
The two prime objectives, namely, maximizing the market
share and maximizing the profit, are commonly considered in
A conjoint survey was conducted to study the consumer
new product development projects for determining the optimal
perception on the dimensions of customer satisfaction of electric
product design (Kwong et al., 2011b; Deng et al., 2014). In this
irons. Table 1 shows the survey questionnaire, which contains
research, the two objectives are also considered. Based on the
sixteen product profiles. Four scales labeled “below average,”
equations described in the previous sections, a multi-objective
“average,” “good,” and “very good” (scale from 1 to 4, respectively)
optimization model can be formulated as follows.
are used to describe the five dimensions of the customer satisfac-
Objectives:
tion of electric irons, which are attractiveness A, quality Q ,

functionality F, user-friendliness Uf , and price P. The conjoint
eUi
Objective 1 : Max Ms ¼ P  PK   ð16Þ survey was conducted in a university. Respondents were invited to
T
t¼1 eU t þ k¼1 eU k þ e U i assess all the product profiles by filling out the survey

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
6 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎

questionnaires using a scale from 1 to 10, where 10 means highly the value of IC is larger than 0.7, the corresponding model fits the
preferred and 1 means not preferred at all. Totally, eighty-four data sets well.
valid questionnaires were received. All the respondents are adults
and have more than five-year experience of using electric irons. 4.1.1. Modeling the relationships between customer satisfaction and
Based on the collected survey data, the average rating for each engineering requirements
product profile was calculated. After that, a chaos-based FR A lead user survey was conducted to perform a competitive
approach was used to generate the product utility function. In product benchmark for electric irons. Five lead users were
the proposed approach, the number of iterations for chaotic involved in the survey who all have more than fifteen-year
searching is set as 500 to ensure that the least number of iterations experience of using electric irons and also have more than three
and the smallest error are obtained. The value setting of h was times of selection and purchase of electric irons. In this case study,
determined using different values within the range of ½0 ; 1. eight major competitive electric irons were identified and were
After a number of trials, h was set as 0.2 because it yielded the denoted as A–H, respectively. In the survey, the product specifica-
smallest training error of the FR models. The number of elements tions of the eight electric irons are shown in Table 2. Lead users of
in the chaos variable is set as nine to guarantee that the generated electric irons were invited to assess each electric iron with respect
model has the chance to include all the five terms. The proposed to the three dimensions of customer satisfaction, Q , F, and Uf . The
approach to generate the product utility functions was implemen- scale of 1 to 4 was used to describe the degrees of customer
ted using Matlab programming software. The following shows a satisfaction. The average values of the dimensions of customer
product utility function, which was generated using the survey satisfaction were calculated and are shown in Table 2. The product
results. specifications contain ten engineering requirements, namely, anti-

calcium, weight, heat-up time, power, water tank volume, sole-
U ¼ ð1:7108; 0:5048Þ þ ð0:1073; 0:0569ÞAnUf þð0:5477; 0:0165ÞF plate material, self-cleaning, variable steam setting, auto shut-off,
þð1:1307; 0:2878ÞQ þð  0:5655; 0ÞP ð25Þ and price, which are denoted as x1  x10 , respectively.
In the benchmark, x2 ; x3 ; x4 ; x5 ; and x10 are quantitative vari-
The values of MAPE and IC for the utility function are 7.79% and ables, which are real numbers. x1 ; x6 ; x7 ; x8 ; and x9 are qualitative
0.85, respectively. A model with the value of MAPE is smaller than variables. In this research, dummy variables were introduced for
10%, which normally indicates that the model has good prediction coding the qualitative variables, which may normally have the
accuracy. Referring to the categorization of the strength of the value one or zero. One indicates the occurrence, while zero
correlations (Dancey and Reidy, 2011), the absolute values of the otherwise. If an attribute has ki levels, it is coded in terms of
correlation coefficients between 0.7 and 0.9 are categorized as ki  1 dummy variables. Taking the soleplate material as an exam-
strong correlations and 1 means a perfect correlation. Therefore, if ple, it can be coded as shown in Table 3.

Table 1
Survey questionnaire for the electric iron.

Product profiles Attractiveness (A) Quality (Q ) Functionality (F) User-friendliness (Uf ) Price (P) Rating (1–10)n

1 4 4 4 4 4
2 3 4 3 3 3
3 2 4 2 2 2
4 1 4 1 1 1
5 3 3 4 2 1
6 4 3 3 1 2
7 1 3 2 4 3
8 2 3 1 3 4
9 2 2 4 1 3
10 1 2 3 2 4
11 4 2 2 3 1
12 3 2 1 4 2
13 1 1 4 3 2
14 2 1 3 4 1
15 3 1 2 1 4
16 4 1 1 2 3

Table 2
Competitive benchmark of electric irons.

Comp. Anti- Weight Heat-up time Power Water tank vol. Soleplate Self- Variable steam Auto shut- Price Dimensions
products calc (kg) (min) (W) (ml) material cleaning setting off of customer
satisfaction

x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 Q F Uf

A Yes 1.2 3 2200 300 Ceramic No Yes Yes 478 3.6 3.8 3.8
B No 0.45 0.5 1000 40 Non-stick No Yes No 338 2.6 3.2 3.2
C No 0.45 1 420 40 Non-stick No No No 230 2.6 1.6 2.5
D No 1.03 1 800 70 Stainless steel No Yes No 155 1.8 2.8 1.5
E Yes 1.24 3 2200 310 Ceramic Yes Yes Yes 435 3.9 4 3.9
F No 1.3 3.3 2000 300 Stainless Steel No Yes No 350 1.8 3.1 1.2
G Yes 1.35 2.5 1800 160 Stainless Steel Yes Yes Yes 428 3.7 3.7 3.1
H No 0.65 0.7 700 40 Ceramic No No No 210 2.9 2.3 1.8

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

The engineering requirements x1 ; x6 ; x7 ; x8 ; and x9 are all Table 10 shows the values of MAPE and R2 of the utility functions
qualitative variables and they are coded using the dummy vari- generated based on the five functions.
ables shown in Table 4. The table shows that the values of MAPE and R2 based on the
Based on the benchmark, the customer satisfaction models of Gaussian function are the best compared with those based on the
Q , F and Uf were developed using FR approach. The generated other four functions. Therefore, the model generated based on the
models, the values of MAPE and IC are shown in Table 5. Gaussian function was selected as the utility function of price.
2
P ¼ 3:828e  ððPrc  708:5Þ=532:6Þ ð27Þ
4.1.2. Modeling customer satisfaction for affective design
For the affective design, five design attributes were defined to Based on model (27), the P values for the competitive products
model the customer satisfaction “Attractiveness”. Table 6 shows A–H are calculated as shown in Table 11.
the five design attributes: body color tone, soleplate, handle Based on the results of Sections 4.1.1–4.1.3, the customer
design, spray button design, and water level indicator, which are satisfaction models of Q , F, Uf , and A and the price model P are
denoted as x11 , x12 , x13 , x14 , and x15 , respectively. The design substituted into Eq. (25) and the product utility function can be
profiles of the eight competitive products are shown in Table 7. obtained as follows:
Dummy variables were used to code the qualitative variables, and 
the dummy coded design variables are shown in Table 8. A survey U ¼ ð1:7108; 0:5048Þ þð0:1073; 0:0569ÞAnUf þ ð0:5477; 0:0165ÞF
was conducted using a questionnaire in which four scales were þð1:1307; 0:2878ÞQ þ ð  0:5655; 0ÞP
used to assess the attractiveness of the eight competitive electric ¼ ð1:7108; 0:5048Þ þ ð0:1073; 0:0569Þnfð3:0382; 0:8Þ
irons. The means of the affective responses of respondents to the þð  0:4206; 0Þx11 þð  0:8206; 0Þx12 þ ð2:4412; 0Þx131
“Attractiveness” are shown in the last column of Table 8. þð0:9618; 0Þx132 þ ð1:2412; 0Þx133 þ ð  0:6588; 0Þx141
Based on the data sets shown in Table 8, the customer
þð  1:2794; 0Þx142 þ ð0:2206; 0Þx15 gnfð1:0134; 0Þ
satisfaction model of “Attractiveness”A was developed based on
FR as follows: þð0:8440; 0Þx1 þð  0:0201; 0Þx2 þ ð  0:1220; 0Þx3
þð  0:0001; 0:02Þx5 þ ð0:8898; 0Þx61 þ ð1:6223; 0Þx62
A ¼ ð3:0382; 0:8Þ þ ð  0:4206; 0Þx11 þð  0:8206; 0Þx12
þ ð2:4412; 0Þx131 þ ð0:9618; 0Þx132 þ ð1:2412; 0Þx133 þð0:1093; 0Þx7 þð0:6390; 0Þx8 þð0:8440; 0Þx9 g
þ ð  0:6588; 0Þx141 þ ð 1:2794; 0Þx142 þ ð0:2206; 0Þx15 ð26Þ þð0:5477; 0:0165Þnfð1:2870; 0Þ þ ð0:0172; 0Þx1 þ ð  0:0519; 0Þx3

The values of MAPE and IC for model (26) are 7.72n10  14 and 1, þð0:0010; 0:0019Þx4 þ ð  0:0033; 0Þx5 þð0:4314; 0Þx61
respectively. þð0:0788; 0Þx62 þ ð0:2462; 0Þx7 þ ð0:9965; 0Þx8 þ ð0:0172; 0Þx9 g

þð1:1307; 0:2878Þnfð1:8; 0:8Þ þ ð0:35; 0Þx1 þ ð1:1; 0Þx61

4.1.3. Development of price models
To develop the utility function of price for electric irons, in the þð0:8; 0Þx62 þ ð0:75; 0:5Þx7 þð0:35; 0Þx9 g þ ð  0:5655; 0Þ
2
lead user survey, lead users were asked to express their views on nf3:828e  ððPrc  708:5Þ=532:6Þ g ð28Þ
the prices of household electric irons using the scale from 1 to 4,
which denote low, medium, high, very high, respectively. After the
survey, the means of ratings were obtained as shown in Table 9. 4.2. Formulation of an optimization model
Based on the above results, the utility function of price was
generated based on polynomial modeling. In this research, several In this research, two objectives, namely, maximizing the
common mathematical functions, including exponential function, market share and maximizing the profit, are considered in the
logarithm function, linear polynomial, power function, and Gaus- optimization. In the former one, the objective function can be
sian function, were employed to generate the utility function. developed based on the product utility function (28) and Eq. (16).
In the formulation of the objective function of the latter one, a cost
model needs to be developed first. To develop the cost model, an
experienced product engineer, who has more than ten-year
Table 3
Dummy coded for the soleplate material.
experience in costing and development of electrical appliance
products, was invited to estimate the product costs of individual
x61 x62 competitive products based on the assumption that the products
were designed and produced by his own company. The estimated
Ceramic 1 0
product costs by the engineer for products A–H were 312, 182, 124,
Non-stick 0 1
Stainless steel 0 0 104, 338, 234, 273, and 111, respectively. Based on the settings of
x1  x9 shown in Table 4, the cost model C can be generated as

Table 4
Coded engineering requirements with dummy variables.

Comp. products Anti-calc Weight (kg) Heat-up time (min) Power (W) Water tank vol.(ml) Soleplate Self-cleaning Variable steam setting Auto shut-off Price
material

x1 x2 x3 x4 x5 x61 x62 x7 x8 x9 x10

A 1 1.2 3 2200 300 1 0 0 1 1 478

B 0 0.45 0.5 1000 40 0 1 0 1 0 338
C 0 0.45 1 420 40 0 1 0 0 0 230
D 0 1.03 1 800 70 0 0 0 1 0 155
E 1 1.24 3 2200 310 1 0 1 1 1 435
F 0 1.3 3.3 2000 300 0 0 0 1 0 350
G 1 1.35 2.5 1800 160 0 0 1 1 1 428
H 0 0.65 0.7 700 40 1 0 0 0 0 210

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
8 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Table 5
Developed customer satisfaction models and their training results.

Models MAPE (%) IC

Quality (Q ) Q ¼ ð1:8; 0:8Þþ ð0:35; 0Þx1 þ ð1:1; 0Þx61 þ ð0:8; 0Þx62 2.96 0.94
\  t þ ð0:75; 0:5Þx7 þ ð0:35; 0Þx9

Function (F) F ¼ ð1:2870; 0Þþ ð0:0172; 0Þx1 þ ð  0:0519; 0Þx3 0.39 1

þ ð0:0010; 0:0019Þx4 þ ð  0:0033; 0Þx5 þ ð0:4314; 0Þx61
þ ð0:0788; 0Þx62 þ ð0:2462; 0Þx7 þ ð0:9965; 0Þx8
þ ð0:0172; 0Þx9

User-friendliness (Uf ) Uf ¼ ð1:0134; 0Þþ ð0:8440; 0Þx1 þ ð  0:0201; 0Þx2 0.19 1

þ ð  0:1220; 0Þx3 þ ð 0:0001; 0:02Þx5 þ ð0:8898; 0Þx61
þ ð1:6223; 0Þx62 þ ð0:1093; 0Þx7 þ ð0:6390; 0Þx8
\  t þ ð0:8440; 0Þx9

Table 6
Morphological analysis on the eight electric irons.

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

Table 7
Design matrix of the eight competitive products.

Design attributes Category Steam iron

A B C D E F G H

Body colour tone (x11 ) (1) Warm √

(2) Cold √ √ √ √ √ √ √
Soleplate (x12 ) (1) Sharp tip √ √ √ √ √ √
(2) Round tip √ √
Handle design (x13 ) (1) Embedded type √ √ √
(2) “╗” Type √ √
(3) “╔” Type √ √
(4) “T” Type √
Spray button design (x14 ) (1) Flat √ √ √ √
(2) Convex √ √ √
(3) Handle-shape √
Water level indicator (x15 ) (1) Transparent √ √ √ √
(2) Sandblasted √ √ √ √

Table 8
Dummy coded design attributes for affective design.

Comp. products Dummy coded design attributes Attractiveness (A)

Body colour tone Soleplate Handle design Spray button design Water level indicator

x11 x12 x131 x132 x133 x141 x142 x15

A 1 1 1 0 0 1 0 1 3.8
B 0 0 0 0 1 0 1 0 3
C 0 0 0 0 0 1 0 1 2.6
D 0 1 0 1 0 0 1 0 1.9
E 0 1 1 0 0 1 0 0 4
F 0 1 0 0 1 1 0 0 2.8
G 0 1 1 0 0 0 1 1 3.6
H 0 1 0 1 0 0 0 1 3.4

Table 9 follows using FR approach.

Survey results for the price.
C ¼ ð31:9734; 0Þ þ ð16:0864; 0Þx1 þð  11:5390; 0Þx2 þ ð  5:0097; 0Þx3
Price Average rating
þ ð0:0885; 0:1114Þx4 þð0:2157; 0Þx5 þð18:6070; 0Þx61
600 3.7
500 3.2 þ ð56:4465; 0Þx62 þ ð20:1316; 0Þx7 þ ð3:4440; 0Þx8 þ ð16:0864; 0Þx9
400 2.8
ð29Þ
300 2.2
200 1.4
The values of MAPE and IC for model (29) are 1.92% and 1,
100 1.1
respectively. Based on Eqs. (28), (29), and (17), the objective
function for maximizing the profit is obtained.
Table 10
The constraints, (23) and (24), are necessary for the formula-
Comparison results based on the five functions. tion of the multi-objective optimization model. The first constraint
is the ranges of the design variables, which are [0.45, 1.35], [0.5,
Exponential Logarithm Linear Power Gaussian 3.3], [420, 2200], [40, 310], and [155, 600] for x2 , x3 , x4 , x5 , and x10 ,
function function polynomial function function
respectively, and [0, 1] for other design variables. The second
MAPE(%) 12.44 12.86 7.95 6.97 4.09 constraint is the technical correlation among the design variables.
R2 0.90 0.93 0.96 0.97 0.99 Referring to the competitive product benchmark (Table 2), weight
x2 has technical correlations with the water tank volume x5 and
soleplate material x6 ; heat-up time x3 has technical correlations
with power x4 and soleplate material x6 . The two technical
Table 11 correlations were generated using FR approach as shown in (30)
P values for the products A–H.
and (31), respectively.
Comp. products Price (x10 ) Utility
x2 ¼ ð0:9121; 0:1403Þ þ ð0:0021; 0:0010Þx5 þ ð  0:3448; 0Þx61
A 478 3.2 þ ð 0:5448; 0Þx62 ð30Þ
B 338 2.4
C 230 1.7
D 155 1.3 x3 ¼ ð  0:3346; 0:4104Þ þð0:0017; 0:0002Þx4 þ ð  0:1500; 0Þx61
E 435 2.9
þð  0:0612; 0:8209Þx62 ð31Þ
F 350 2.4
G 428 2.9
H 210 1.6
The values of IC for models (30) and (31) are 0.91 and 0.82,
respectively.

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
10 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎

4.3. Determination of design specifications number of generations were set as 100. The distribution index for
crossover and mutation were both set as 20. The crossover
Based on the formulated multi-objective optimization model, a probability and mutation probability were set as 0.9 and 0.1,
NSGA-II was introduced to solve the optimization problem and respectively. The above parameter settings, suggested by Deb
determine the optimal settings of the design variables for the et al. (2002), have been adopted widely by other researchers.
electric iron design. In this paper, both the population size and the The tournament size and the size of the mating pool are com-
monly set as two and one half of the population size, respectively.
The optimization model and its solving were implemented using
Matlab programming software. Fig. 2 shows the Pareto optimal
solutions of the multi-objective optimization problem solved by
the NSGA-II.
Each optimal solution contains the settings of the design
variables. The company can perform a tradeoff between the two
objectives and select the best/most preferred one with reference
to their business objectives and competitive strategies. If the
company would like to develop a new electric iron with the
objective of maximizing the profit while achieving a 25% or above
market share, the solution, as shown in a black rectangle in Fig. 2,
can be considered as the optimal solution. The market share and
profit of the optimal solution are 25% and $3.88n105, respectively.
The optimal settings of the design variables of the new electric
iron are x1 ¼ 1, x2 ¼ 0:74, x3 ¼ 2:07, x4 ¼ 1500, x5 ¼ 80, x61 ¼ 1,
x62 ¼ 0, x7 ¼ 0, x8 ¼ 1, x9 ¼ 1, x10 ¼ 277:27, x11 ¼ 1, x12 ¼ 1, x131 ¼ 1,
x132 ¼ 0, x133 ¼ 0, x141 ¼ 0, x142 ¼ 1, and x15 ¼ 0. Thus, the design
specification of the new electric iron can be obtained in Table 12 .

Fig. 2. Pareto solutions.

5. Conclusion

In the early design stage of consumer products, the concerns of

Table 12
affective design, engineering, and marketing have to be considered
Design specification of the new electric iron.
simultaneously to generate an optimal product design. However,
Design variables The optimal settings of design variables publications on the simultaneous consideration of those concerns
in the early design stage have not been found thus far. To fill the
Anti-calc x1 Yes research gap, this paper proposes and describes an AI-based
Weight (kg) x2 0.74
methodology of integrating affective design, engineering, and
Heat-up time (min) x3 2.07
Power (W) x4 1500 marketing for defining the design specifications of new products
Water tank vol. (ml) x5 80 by which concerns of affective design, engineering, and marketing
Soleplate material x6 Ceramic can be considered simultaneously. In the proposed methodology,
Self-cleaning x7 No
customer satisfaction models of attractiveness, quality, function-
Variable steam setting x8 Yes
Auto shut-off x9 Yes
ality, and user-friendliness are developed based on FR. The product
Price x10 277.27 utility function is generated using a chaos-based FR approach. The
Body colour tone x11 Warm tone market share model is then developed based on the product utility
Soleplate x12 Sharp tip function and MNL model. The profit model is developed based on
the market demand and cost model, which the cost model is
generated using FR approach. After that, a multi-objective optimi-
zation model is formulated for maximizing the market share and
profit. The optimization model is solved using a NSGA-II and the
optimal settings of the design variables of a new product can be
determined. A case study of electric iron design was conducted to
evaluate the effectiveness of the proposed methodology.
Future work could further extend the proposed methodology to
Handle design x13 embedded type
perform product line design, which involves the design of multi-
products to satisfy the needs of various market segments. Further-
more, customer preference and market status are in static forms in
the development of the proposed methodology. In reality, they
could be quite dynamic. Future work could consider their dynamic
effects in the proposed methodology.
Spray button design x14 convex

Acknowledgement

The work described in this paper was fully supported by a grant

Water level indicator x15 Sandblasted
from the Research Grant Council of the Hong Kong Special
Administrative Region, China (Project no: PolyU 517113).

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 11

Initialization

Generation = 1

Calculate the objective functions of

each chromosome

Rank the population according to

non-dominating criteria

Selection
No
Crossover

Mutation

Final generation? Calculate the objective

Yes functions of the new population

Pareto optimal Combine parent population and

solutions off spring population

Non-dominating ranking on the

Obtain the optimal combined population
settings of design
specifications of a
new product Calculate crowding distance of
all the solutions

Getthe Nmember from the

combined population on the
basis of rank and crowding
distance

Replace parent population by

the better members of the
combined population

Fig. 3. Flowchart of NSGA-II.

Appendix domination, and the crowding distances are calculated for each
individual. The generation is set as t ¼ 1.
Fig. 3 shows a flowchart of NSGA-II and algorithms of the Step 3: The crowded tournament selection is applied to create a
NSGA-II are as follows: mating population. In the selection process, two individuals
from the parent population P t are selected at random for a
Step 1: The initialization of the parameters is first conducted, tournament. The winners chosen are inserted in the mating
including the population size N, the number of generations, the pool for reproduction and the selection is repeated until the
distribution index for crossover ηc , the crossover probability, size of the mating pool reaches the predefined value.
the distribution index for mutation ηm , the mutation probabil- Step 4: The SBX operator and the polynomial mutation are
ity, the size of the mating pool, and the tournament size. conducted in the mating pool to create the offspring pop-
The description of the objective functions is given, such as ulation Q t . A combined population Rt is generated by combin-
the number of objective functions, the number of design ing the parent population P t and the offspring population Q t ,
variables V, and the range of the design variables j, ½xj min ; Rt ¼ P t [ Q t .
xj max , 1 r jr V. Step 5: The combined population Rt is sorted based on non-
Step 2: A parent population P 1 is initialized randomly. The domination, and different fronts F i , i ¼ 1; 2; …, are identified.
values of the objective functions of each individual are calcu- The new population P t þ 1 with size N is obtained based on the
lated. The parent population P 1 is then sorted based on non- process of combination and selection.

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i
12 C.K. Kwong et al. / Engineering Applications of Artificial Intelligence ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Step 6: The generation counter is increased by t þ 1-t. The Kim, H.S., Cho, S.B., 2000. Application of interactive genetic algorithm to fashion
algorithm is again executed from Step 3 and stops after the design. Eng. Appl. Artif. Intell. 13 (6), 635–644.
Kumar, D., Chen, W., Simpson, T.W., 2009. A market-driven approach to product
number of generations reaches the predefined value. Finally, family design. Int. J. Prod. Res. 47 (1), 71–104.
the Pareto front is obtained and the individuals belonging to Kwong, C.K., Chen, Y., Chan, K.Y., 2011a. A methodology of integrating marketing
the Pareto front are the optimal settings of the design with engineering for defining specifications of new products. J. Eng. Des. 22 (3),
201–213.
specifications. Kwong, C.K., Luo, X.G., Tang, J.F., 2011b. A multiobjective optimization approach for
product line design. IEEE Trans. Eng. Manage. 58 (1), 97–108.
Kwong, C.K., Wong, T.C., Chan, K.Y., 2009. A methodology of generating customer
satisfaction models for new product development using a neuro-fuzzy
References approach. Expert Syst. Appl. 36 (8), 11262–11270.
Lai, H.H., Chang, Y.M., Chang, H.C., 2005a. A robust design approach for enhancing
Aktar Demirtas, E., Anagun, A.S., Koksal, G., 2009. Determination of optimal product the feeling quality of a product: a car profile case study. Int. J. Ind. Ergon. 35 (5),
styles by ordinal logistic regression versus conjoint analysis for kitchen faucets. 445–460.
Int. J. Ind. Ergon. 39 (5), 866–875. Lai, H.H., Lin, Y.C., Yeh, C.H., 2005b. Form design of product image using grey
Aydin, R., Kwong, C.K., Ji, P., Law, H.M.C., 2014. Market demand estimation for new relational analysis and neural network models. Comput. Oper. Res. 32 (10),
product development by using fuzzy modeling and discrete choice analysis. 2689–2711.
Neurocomputing 142 (2014), 136–146. Lau, T.W., Hui, P.C.L., Ng, F.S.F., Chan, K.C.C., 2006. A new fuzzy approach to improve
Azadeh, A., Seraj, O., Saberi, M., 2011. An integrated fuzzy regression–analysis of fashion product development. Comput. Ind. 57 (1), 82–92.
variance algorithm for improvement of electricity consumption estimation in Lin, Y.C., Lai, H.H., Yeh, C.H., 2007. Consumer-oriented product form design based
uncertain environments. Int. J. Adv. Manuf. Technol. 53 (5-8), 645–660. on fuzzy logic: a case study of mobile phones. Int. J. Ind. Ergon. 37 (6), 531–543.
Barnes, C., Lillford, S.P., 2007. Affective design decision-making—issues and oppor- Liu, Y., Chen, Y., Zhou, J., Zhong, S., 2015. Fuzzy linear regression models for QFD
tunities. CoDesign: Int. J. CoCreation Des. Arts 3 (Suppl. 1), 135–146. using optimized h values. Eng. Appl. Artif. Intell. 39 (2015), 45–54.
Barnes, C., Lillford, S.P., 2009. Decision support for the design of affective products. Lokman, A.M., 2010. Design & Emotion: the Kansei engineering methodology.
J. Eng. Des. 20 (5), 477–492. Malays. J. Comput. 1 (1), 1–14.
Barone, S., Lombardo, A., Tarantino, P., 2007. A weighted logistic regression for Luo, L., 2011. Product line design for consumer durables: an integrated marketing
conjoint analysis and Kansei engineering. Qual. Reliab. Eng. Int. 23 (6), and engineering approach. J. Marketing Res. 48 (1), 128–139.
689–706. Luo, L., Kannan, P.K., Besharati, B., Azarm, S., 2005. Design of robust new products
Besharati, B., Luo, L., Azarm, S., Kanan, P.K., 2006. Multi-objective single product under variability: marketing meets design. J. Prod. Innovat. Manage. 22 (2),
robust optimization: an integrated design and marketing approach. J. Mech. 177–192.
Des. 128 (4), 884–892. Marghani, V.G.R.E., da Silva, F.C., Knapik, L., Verri, M.A., 2013. Kansei engineering:
Chan, K.Y., Kwong, C.K., Dillon, T.S., Fung, K.Y., 2011. An intelligent fuzzy regression methodology to the project oriented for the customers. In: Fukuda, S. (Ed.),
approach for affective product design that captures nonlinearity and fuzziness. Emotional Engineering, vol. 2. Springer, London, pp. 107–125.
J. Eng. Des. 22 (8), 523–542. Michalek, J.J., Ceryan, O., Papalambros, P.Y., Koren, Y., 2006. Balancing marketing
Chang, C.C., 2008. Factors influencing visual comfort appreciation of the product and manufacturing objectives in product line design. J. Mech. Des. 128 (6),
form of digital cameras. Int. J. Ind. Ergon. 38 (11–12), 1007–1016. 1196–1204.
Chen, C.H., Khoo, L., Yan, W., 2003. Evaluation of multicultural factors from elicited Michalek, J.J., Ebbes, P., Adiguzel, F., Feinberg, F.M., Papalambros, P.Y., 2011.
customer requirements for new product development. Res. Eng. Des. 14 (3), Enhancing marketing with engineering: optimal product line design for
119–130. heterogeneous markets. Int. J. Res. Marketing 28 (1), 1–12.
Choi, K., Jun, C., 2007. A systematic approach to the Kansei factors of tactile sense Michalek, J.J., Feinberg, F.M., Papalambros, P.Y., 2005. Linking marketing and
regarding the surface roughness. Appl. Ergon. 38 (1), 53–63. engineering product design decisions via analytical target cascading. J. Prod.
Chuang, M.C., Ma, Y.C., 2001. Expressing the expected product images in product Innovat. Manage. 22 (1), 42–62.
design of micro-electronic products. Int. J. Ind. Ergon. 27 (4), 233–245. Nagamachi, M., 1995. Kansei engineering: a new ergonomic consumer-oriented
Crawford, M., Benedetto, A.D., 2006. New Products Management. McGraw-Hill. technology for product development. Int. J. Ind. Ergon. 15 (1), 3–11.
Creusen, M.E.H., Schoormans, J.P.L., 2005. The different roles of product appearance Nagamachi, M., 2008. Perspectives and the new trend of Kansei/affective engineer-
in consumer choice. J. Prod. Innovat. Manage. 22 (1), 63–81. ing. TQM J. 20 (4), 290–298.
Dancey, C., Reidy, J., 2011. Statistics without Maths for Psychology,. Prentice Hall, Noble, C.H., Kumar, M., 2008. Using product design strategically to create deeper
Harlow. consumer connections. Bus. Horiz. 51 (5), 441–450.
Deb, K., Pratap, A., Agarwal, S., Meyarivan, T., 2002. A fast and elitist multiobjective Park, J., Han, S.H., 2004. A fuzzy rule-based approach to modeling affective user
genetic algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6 (2), 182–197. satisfaction towards office chair design. Int. J. Ind. Ergon. 34 (1), 31–47.
Deng, S., Aydin, R., Kwong, C.K., Huang, Y., 2014. Integrated product line design and Sekkeli, G., Koksal, G., Batmaz, I., Bayrak, O.T., 2010. Classification models based on
supplier selection: a multi-objective optimization paradigm. Comput. Ind. Eng. Tanaka’s fuzzy linear regression approach: the case of customer satisfaction
70 (2014), 150–158. modeling. J. Intell. Fuzzy Syst. 21 (5), 341–351.
Haghighi, A., Asl, A.Z., 2014. Uncertainty analysis of water supply networks using Seva, R.R., Duh, H.B.L., Helander, M.G., 2007. The marketing implications of affective
the fuzzy set theory and NSGA-II. Eng. Appl. Artif. Intell. 32 (2014), 270–282. product design. Appl. Ergon. 38 (6), 723–731.
Han, S.H., Yun, M.H., Kim, K.J., Kwahk, J., 2000. Evaluation of product usability: Shi, L., Olafsson, S., Chen, Q., 2001. An optimization framework for product design.
development and validation of usability dimensions and design elements based Manage. Sci. 47 (12), 1681–1692.
on empirical models. Int. J. Ind. Ergon. 26 (4), 477–488. Shiau, C.S.N., Michalek, J.J., 2009. Should designers worry about market systems.
Hong, S.W., Han, S.H., Kim, K.J., 2008. Optimal balancing of multiple affective J. Mech. Des. 131 (1), 011011–1-011011-9.
satisfaction dimensions: a case study on mobile phones. Int. J. Ind. Ergon. 38 (3– Sun, J., Kalenchuk, D.K., Xue, D., Gu, P., 2000. Design candidate identification using
4), 272–279. neural network-based fuzzy reasoning. Rob. Comput. Integr. Manuf. 16 (5),
Hsiao, S.W., Huang, H.C., 2002. A neural network based approach for product form 383–396.
design. Des. Stud. 23 (1), 67–84. Tanaka, H., 1987. Fuzzy data analysis by possibilitic linear model. Fuzzy Sets Syst. 24
Hsiao, S.W., Tsai, H.C., 2005. Applying a hybrid approach based on fuzzy neural (3), 363–375.
network and genetic algorithm to product from design. Int. J. Ind. Ergon. 35 (5), Tsai, H.C., Hsiao, S.W., Hung, F.K., 2006. An image evaluation approach for
411–428. parameter-based product form and color design. Comput.-Aided Des. 38 (2),
Jiang, H., Kwong, C.K., Ip, W.H., Chen, Z., 2013. Chaos-based fuzzy regression 157–171.
approach to modeling customer satisfaction for product design. IEEE Trans. Wang, H.F., Tsaur, R.C., 2000. Insight of a fuzzy regression model. Fuzzy Sets Syst.
Fuzzy Syst. 21 (5), 926–936. 112 (3), 355–369.
Jiao, J., Zhang, Y., 2007. Product portfolio planning with customer-engineering Wang, J., Yannou, B., Alizon, F., Yvars, P.A., 2013. A usage coverage-based approach
interaction. IIE Trans. 37 (9), 801–814. for assessing product family design. Eng. Comput. 29 (4), 449–465.
Kang, N., Kim, J., Park, Y., 2007. Integration of marketing domain and R&D domain Williams, N., Kannan, P.K., Azarm, S., 2011. Retail channel structure impact on
in NPD design process. Ind. Manage. Data Syst. 107 (6), 780–801. strategic engineering product design. Manage. Sci. 57 (5), 897–914.
Kaul, A., Rao, V.R., 1995. Research for product positioning and design decisions: an You, H., Ryu, T., Oh, K., Yun, M.H., Kim, K.J., 2006. Development of customer
integrative review. Int. J. Res. Marketing 12 (4), 293–320. satisfaction models for automotive interior materials. Int. J. Ind. Ergon. 36 (4),
Kazemzadeh, R.B., Behzadian, M., Aghdasi, M., Albadvi, A., 2009. Integration of 323–330.
marketing research techniques into house of quality and product family design. Zhang, F., Wang, J., 2013. Application of Kansei engineering in electric car design.
Int. J. Adv. Manuf. Technol. 41 (9–10), 1019–1033. Appl. Mech. Mater. 437 (2013), 985–989.

Please cite this article as: Kwong, C.K., et al., AI-based methodology of integrating affective design, engineering, and marketing for
defining design specifications of new products. Eng. Appl. Artif. Intel. (2015), http://dx.doi.org/10.1016/j.engappai.2015.04.001i