J. Eng. Technol. Manage.

27 (2010) 110–124

Contents lists available at ScienceDirect

Journal of Engineering and

Technology Management
journal homepage: www.elsevier.com/locate/jengtecman

Risk analysis models and risk degree determination

in new product development: A case study
Hoo-Gon Choi *, Jungon Ahn
Department of Systems Management Engineering, Sungkyunkwan University, 300 Cheoncheon-dong, Jangan-gu, Suwon,
Republic of Korea

A R T I C L E I N F O A B S T R A C T

Article history: This paper proposes a risk analysis model to determine the risk
Available online 3 April 2010 degrees of the risk factors occurring in product development
processes. The model uses both fuzzy theory and Markov processes
Keywords: on a concurrent engineering (CE) basis. Fuzzy models determine the
Risk impact values of the risk factors, and Markov processes determine the
Risk degrees probability of risk occurrences. The analysis model is used to compute
Concurrent engineering the risk degrees by multiplying the probability of risk occurrences by
Fuzzy theory
the impact value. This study can be utilized for analyzing the
Markov process
influences of risk factors on product development projects and will
contribute toward the development of a risk management frame-
work (RMF) to defend against various risk factors. Implications and
directions for future research are discussed.
ß 2010 Elsevier B.V. All rights reserved.

1. Introduction

The product development process determines whether a company survives or fails in competitive
markets because a product life is generally determined by its market share (Fig. 1). In order to ensure
sustainable competitiveness, innovative efforts to develop new products must focus on substituting
existing products or entering new markets. As new products must be introduced to the market
periodically, the development process is a critical strategic issue at business level because the product
life-cycle is becoming shorter (Kim, 2003a,b). However, about 80% of new product development (NPD)
efforts have failed before project completion and more than 50% of the efforts have made no returns on
the investment of money and time (Cooper, 2003). In other words, the product development process
for new products is a complex and difficult business decision-making process because of the high

* Corresponding author.
E-mail address: hgchoi@skku.edu (H.-G. Choi).

0923-4748/$ – see front matter ß 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.jengtecman.2010.03.006
 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124 111

Fig. 1. Product life-cycle and innovation cycle.

capital investment required and exposure to low success probability. The critical explanations for the
difficulty of the product development process are unexpected risks and their impact, and the inability
of the firm to defend against those risks effectively and efficiently.
This paper suggests a new systematic risk management framework (RMF), as shown in Fig. 2. RMF
determines risk degrees for risk factors and total risk degrees of the product development project, and
shows effective and efficient responding activities. Especially, RMF suggests a risk analysis model
under a concurrent engineering (CE) environment. CE is an approach to link all functional areas such as
manufacturing, financing and marketing with the design process (Savic and Kayis, 2006). There is a
multidirectional exchange of information among all functional areas for better, easier, and more
economical product development. Therefore, either a high degree of collaboration or a high
concurrency level (CL) is desirable to construct the CE environment. Furthermore, the fluent

Fig. 2. The entire structure of the risk management framework (RMF) (Ahn et al., 2008).
112 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124

information exchange under high CL enables the functional areas to handle the related risk factors
more effectively and efficiently. The increasing difficulties of a product development project require a
higher CL. In addition, a full understanding of the pitfalls and risks is required for successful CE
implementation. The famous pitfalls include unobtainable-schedule, change-product-ineffective-
team, requirements, business-as-usual vendoring, automate-everything, supplier dependent lead-
time, teams unsupported by reward systems, lack of information technology support project
development instead of process improvement, and discontinued change (Willaert et al., 1998). In this
case study, the risk analysis model is used to determine CL on a CE basis, and the model quantifies the
influence of a risk factor to the development project as an impact value. When a risk factor occurs in a
functional area, process, department or project, its impact value or the amount of influence differs.
Therefore, the impact value has to be defined precisely. This paper proposes the use of fuzzy theory for
this purpose. Because a risk factor occurs probabilistically or stochastically, a Markov process can be
adopted to determine the probability of occurrences for a given risk factor. Finally, the developed risk
analysis model is used to compute the risk degrees by multiplying the probability of occurrences with
the impact value.

2. Determination of concurrency levels (CL)

CL is the degree of collaboration among all functional areas, processes or departments for attaining
better quality in economical ways, achieving faster delivery, and satisfying customer needs under a CE
environment. In this paper, a CE matrix (CEM) is utilized to determine CL. CEM is further classified into
an influencing dimensions matrix (IDM) consisting of 9 major dimensions, as shown in Table 1, and an
environment assessment matrix (EAM) composed of 4 major environments, as presented in Table 2
(Singh, 1996). CEM has been used to design a proper CE environment for developing or producing a
new product. IDM identifies both the influencing dimensions and IDM’s levels on individual
dimension basis using information on the product features, product requirements, and other aspects.
The influencing levels are categorized as A, B, C, and D, respectively, for the increasing order of levels.
The completion of this matrix defines the desired CE environment for a given product development
project. EAM identifies the required elements and sub-elements as well as their levels to support the
desired environment. EAM provides a basis for assessing specific capabilities within individual project
to address the CE approach to product and process development. This paper does not discuss EAM any
further because it focuses solely on risk analysis.
If both IDM and EAM are constructed and aggregated successfully for all departments joining a
development project, then CL is obtained for a firm to develop a product, and the risks can be analyzed
and managed under a given CL of the enterprise. CL is quantified as a scale for a firm to measure
whether it is capable of developing and producing its new product. In addition, CL is affected by the
risk factors and their degrees. Therefore, the project manager needs to decide which actions are
required for the product development under both CL and risks.

Table 1
Influencing dimension matrix (IDM).

Dimension Level Point

a a a a
A B C D

Product complexity O 2
Product technology O 1
Program structure O 2
Program futures O 3
Competition O 3
Business relationship O 4
Team scope O 1
Resource tightness O 1
Schedule tightness O 4

Total 21
a
A: low; B: medium; C: high; D: very high.
 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124 113

Table 2
Environment assessment matrix (EAM).

Environment Level

A B C D

Organization
Team membership O
Team leadership O
..a
.

Requirements
Definition O
Schedule types O
..a
.

Communication
Data management O
Data acquisition O
..a
.

Product development methodology

Optimization O
Data libraries O
..a
.

A: low; B: medium; C: high; D: very high.

a
Reference (Singh, 1996) for more information.

The difficulty of a product development project defined by IDM determines the form of
membership function in fuzzy theory. Different projects have different difficulties due to different
environments. This leads to the adjustment of risk occurrence frequencies, as described below in
Section 3.4.
When the weight is assigned to each level in IDM as 1 for A, 2 for B, 3 for C, and 4 for D, then the
difficulty ranges from a minimum of 9 to a maximum of 36 because there are 9 dimensions. For
example, Table 1 has the project difficulty of 21. However, any values can be assigned to those levels
for defining the difficulty differently in a project.

3. Risk analysis model

The risk analysis techniques can be powerful tools to manage the uncertainty occurring in various
fields. A thorough risk analysis method can support valuable decision-making for estimating and
evaluating the uncertainty. However, many risk analysis techniques have been used mostly for
financial risks.
Risk management experts usually evaluate the risk factors with the simple criteria shown in
Table 3. However, this renders difficult the meticulous classification of the impact values because it
applies only five criteria for evaluating a given risk factor. If tiny value differences exist among risk

Table 3
Evaluation criteria for impact value of risk factorsa.

Value Severity Technical performance Schedule Cost

0.1 Very little Minimal or no impact Minimal or no impact Minimal or

no impact
0.3 Little Minor reduction or supportability Able to meet key dates <5%
0.5 Moderate Moderate reduction with limited Able to meet key milestones 5–7%
impact with no float
0.7 Severe Significant degradation or may Program critical path affected >7–10%
jeopardize program success
0.9 Very severe Key technical threshold; will Cannot meet key program >10%
jeopardize program success milestones
a
DoD (1998).
114 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124

factors, the method cannot classify them effectively. Therefore, fuzzy set theory is an appropriate
choice for better classification. The theory is highly subjective and works well with either inexact or
vague information. In this paper, fuzzy theory is used to decide the impact values of risk factors
because the experts usually evaluate the risk factor inexactly.

3.1. Fuzzy set theory

Zadeh (1965) developed fuzzy set theory to deal with fuzzy events. The theory is based on fuzzy
logic which is a superset of conventional (Boolean) logic. The characteristic function of a conventional
set is defined to be either zero or one, and has been extended to handle the concept of partial truth in
the fuzzy logic. Zadeh used a membership function to represent the intensity of the partial truth. A
stronger or weaker intensity corresponds to a membership function closer to one or to zero,
respectively.
There are two forms of a fuzzy proposition: atomic fuzzy proposition and compound fuzzy
proposition. For example, if a compound fuzzy proposition uses ‘AND’ to combine two atomic fuzzy
propositions such as ‘x is in set A’ and ‘y is in set B’, then the compound fuzzy proposition becomes
mA \ B ðx; yÞ ¼ minðmA ðxÞ; mB ðyÞÞ; ðx; yÞ 2 X  Y

where mA(x) and mB(y) are the membership functions for x and y, respectively. Zadeh also proposed
fuzzy inference to deal with contiguous sets that cannot be covered in the traditional binary logic
inference. In this paper, the direct fuzzy inferencing method is used to obtain fuzzy values for the
contiguous sets. Finally, the clipped center of gravity method as a defuzzification process is applied for
converting the fuzzy value to a nominal non-fuzzy value.
A fuzzy inference system would be an intelligent system if it were developed by using experts’
knowledge, operators’ experience, membership function and logical rules (Hsieh et al., 2007; Dereli
et al., 2007).

3.2. Markov process

A Markov process is a stochastic process in which the next state is determined only by the current
state. Therefore, it is called a memory less process. Because the probability of a given risk factor occurring
in the current project only affects the probability of the same risk factor occurring in next project, the risk
analysis model utilizes the Markov process concept in this paper. If X(t) follows a Markov process at time
steps t1 < t2 < . . . < tk < tk+1, the Markov property can be succinctly stated as follows;
P½Xðt kþ1 Þ ¼ xkþ1 jXðt k Þ ¼ xk ; . . . ; Xðt 1 Þ ¼ x1  ¼ P½Xðt kþ1 Þ ¼ xkþ1 jXðt k Þ ¼ xk :

If the Markov process has a discrete value, it is called a Markov chain. The Markov process consists
of a set of states, the initial probability of each state and the transition probability between states.

3.3. Modeling for impact value of a risk factor

In this paper, the membership function for evaluating the impact value of a risk factor is
characterized by a triangular membership function, as shown in the typical plot of Fig. 3. The triangle
membership function is simple and can be used for the exclusive definition of fuzzy control
mechanisms, managerial decision-making in finance areas, and social science areas. The triangular
membership function is defined as follows:

Lðu; a; b; g Þ ¼ 0; ua
¼ ðu  aÞ=ðb  aÞ; a<u  b (1)
¼ ða  uÞ=ða  bÞ; b<u  g
¼ 0; u>g

Eq. (1), however, is affected by the difficulty of a product development project. The IDM described in
Section 2 is used to determine the project difficulty and forms a membership function. IDM generates
28 different membership functions because the minimum and maximum project difficulties are 9 and
 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124 115

Fig. 3. Typical form of the triangular membership function.

36, respectively. A membership function is designed to be located in a less severe area when the
difficulty is low while another membership function must be located in a very severe area for the most
difficulty project. That is, a membership function can be selected according to the difficulty of a given
project for a fuzzy process. The 30 different types of difficulty give an equivalent number of
membership functions. In this paper, the fuzzy process uses the selected membership function on the
basis of a specified project difficulty.
In general, the dynamic behaviour of the fuzzy system is characterized by a set of fuzzy rules. The
fuzzy rule can be stated simply in the following general form:

if antecedent(s), then consequent(s).

The antecedents of a fuzzy rule are a combination of fuzzy sets defined by logic operators (i.e.,
complement, intersection, and union). The consequent part of a rule is usually a single fuzzy set with a
corresponding membership function. Fig. 4 shows the fuzzy process based on a union operator when
the impact value of a risk factor is given.

Fig. 4. Fuzzy process based on a union operator.

116 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124

The membership function value obtained from the union operator is still a fuzzy value, and thus
requires a defuzzification process to convert the output from the fuzzy rules into a scalar, or non-fuzzy
value. There are several defuzzification methods such as the clipped center of gravity method, mean of
maximum method, and maximum criterion method. The first of these methods has been regarded as
the most powerful for defuzzification (Lee, 1990). For this approach, the centroid of the composite area
is calculated when the horizontal axis represents various outputs. The centroid of the trapezoidal
areas is calculated according to Eq. (2).
R
xmðxÞdx
output ¼ Rx 2 X (2)
x 2 X mðxÞdx

where X is the universe of discourse. The impact value of a risk factor can be calculated by Eq. (2).

3.4. Modeling for probability of risk occurrences

In this paper, a Markov process is adapted to determine the probability of risk occurrences that
follows the stochastic process. The Markov process has three elements: a set of states, the
transition probability matrix (TPM), and the initial state probability. The initial state probability is
obtained by risk occurrence data in the most recent projects, and is represented by row vector, as
shown in Eq. (3).

a ¼ ðak Þ (3)

where ak is the initial state probability for the state k. The set of states is defined as a range of
frequencies for a given risk factor recognized in past projects. S is a set of states, as shown in
Eq. (4).

S ¼ fS1 ; S2 ; . . . ; Sn g (4)

A large, reliable amount of past risk data is required for the Markov process (Kim, 2006). Especially,
the frequency of risk factors occurring in each project is collected for computing the probability of risk
occurrences in the Markov process. The 28 different difficulty values, ranging from 9 to 36, are denoted
as Ci, where i indicates a project. As the median of this range is 22.5, projects with difficulties ranging
from 9 to 22 are easier to perform than those between 23 and 36. Eq. (5) determines the frequency of a
given risk factor occurring in a project on the basis of the difficulty, Ci. The entire difficulty values are
divided into two groups on either side of the median value of 22.5: Group 1 has those difficulty values
of 9  Ci1  22 and Group 2 is defined by the values of 23  Ci2  36. High difficult projects experience
various risk factors and more frequent risk occurrences, which necessitate more effective or efficient
activities to prevail over those risk factors compared to easier projects. This necessitates the
adjustment of risk occurrence frequencies for easier projects which need less effort to handle various
risk factors. Suppose that a risk factor occurs twice in the easiest project (Ci = 9) and the same factor
occurs twice in the median difficult project (Ci = 22). If the average frequency in Group 1 is 4, then the
frequency for the easiest project is increased to 5.7 or 6 (rounded up to next higher integer). The

Fig. 5. Prorated risk occurrences.

 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124 117

frequency for the median difficult project remains at still 2. Refer to Eq. (5). Fig. 5 presents the prorated
risk occurrence in each difficulty value.

2  
Ib  C i1
6 Fi þ Ia
 F av1 ; 9  C i1  22
6  
0 6
F i ¼ 6 C i2  Ig (5)
6 Fi   F av2 ; 23  C i2  36
4 Ia
0; F i ¼ 0

where Ci1: difficulty of project i in Group 1, Ci2: difficulty of project i in Group 2, Cmax: the maximum
difficulty of Ci (=36 in this paper), Cmin: the minimum difficulty of Ci (=9 in this paper), Ia: total number
of Ci in each group, Ib = ((Cmax + Cmin  1)/2): adjusted median of Ci (a rounded off integer),
Ig = ((Cmax + Cmin + 1)/2): adjusted median of Ci (a rounded up integer), Fi = frequency of risk occurrences
in project i, Fav1: average frequency of risk occurrences in Group 1, Fav2 = average frequency of risk
occurrences in Group 2, F0 i: adjusted frequency of risk occurrence in project i.
After the frequency of risk occurrences for a risk factor under a given difficulty value is modified by
Eq. (5), the probability of risk occurrences is computed by a Markov process. A state is defined by a
range of frequencies measured in past projects. The transition probability between states is
computed in a Markov chain. It is important to reduce the size of TPM or the number of states for
simplification. This can be done by determining the range of risk frequencies. The number of
transitions between states must be normalized to convert the total sum of probabilities in each row
to 1. As a result, Pkl, which is TPM, is determined by Qkl, the number of transitions between states k and
l. Refer to Eq. (6).
0 1
P 11 P12 P 13   P1n
B P 21 P22 P 23   P2n C
B CX n Xn Xn
B       C
P¼B B 
C P1l ¼ 1;
C P2l ¼ 1; . . . ; P nl ¼ 1 (6)
B  Pkl    C
l¼1 l¼1 l¼1
@       A
Pn1 P n2 Pn3   P nn

where
Q
Pkl ¼ Pn kl ; k; l ¼ states
l¼1 Q kl

Finally, the probability of risk occurrence for a risk factor, P0 , is obtained by multiplying the initial
state probability with the TPM shown in Eq. (7).

P0 ¼ a  P (7)

where a: the initial state probability.

3.5. Modeling for risk degrees

The previous sections presented theoretical backgrounds for fuzzy models, frequency
adjustment methods, and the Markov process to determine the impact value, the adjusted
frequency, and the probability of risk occurrences for a risk factor, respectively. As the final step,
the risk analysis model computes the total risk degrees for the entire product development project,
as shown in Eqs. (8)–(11). Eq. (8) indicates that the risk degree is determined by multiplying the
probability of risk occurrences with its impact value. Furthermore, if Rmj is aggregated properly,
PRm, DRk, and TR can be computed, and the optimized set of responding activities may be obtained
to minimize the total risk degrees for the entire project, as shown in Eq. (12). At present, Eqs. (9)–
(12) are in the conceptual stage, and this aggregation method is now under investigation for a
complete form of RMF.

Rm j ¼ Pm j  Im j (8)
118 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124

Table 4
Expert’s evaluation for the given example.

Dimension Level

A(1) B(2) C(3) D(4)

Product complexity O
Product technology O
Program structure O
Program futures O
Competition O
Business relationship O
Team scope O
Resource tightness O
Schedule tightness O
Project difficulty 30
Target risk factor Manufacturing dimensional and geometrical
tolerances may affect product performance.
Expert’s evaluation for the impact value of the target risk factor 0.7

where Rmj: risk degrees for risk factor j in the mth process, Pmj: probability of risk occurrence for risk
factor j in the mth process, Imj: impact value of risk factor j for the mth process

PRm ¼ f ðRm j Þ (9)

DRk ¼ gðPRm Þ (10)

TR ¼ hðDRk Þ (11)

argmin TRa (12)

a

where PRm: total risk degrees in the mth process, DRk: total risk degrees in the kth department, TR: total
risk degrees for the product development project, argmina TRa : the set of responding activities (a) to
minimize total risk degrees for the product development project.

3.6. An illustrated example of risk analysis models

In this example, the target risk factor is described such that manufacturing dimensional and
geometrical tolerances may affect product performance (Anderson, 2008). Given the 30 projects
performed in past and under the assumption that the same risk factor can occur in the future projects,
we can estimate the risk degree of the specified risk factor for the 31st project. Table 4 shows the
expert’s evaluation for the project difficulty using IDM, and for the target risk factor. Table 5 presents
the data (frequencies) of the past 30 projects. In this illustrated example, all evaluations and data were
assumed by the authors.
Because the project difficulty is 30 in Table 4, the 22nd membership function is selected, as shown
in Fig. 6. Since the expert evaluates the impact value of the target risk factor as 0.7, its corresponding
value is 0.3684 on the very severe line and 0.8889 on the severe line in the selected membership
function, as shown in Fig. 7. Finally, the impact value of the target risk factor is calculated by the

Table 5
The sets of difficulties and the modified frequencies of the target risk factor in each project.
Project no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Difficulty (Ci) 27 20 18 9 15 36 26 12 30 9 15 25 20 28 10
Frequency (Fi) 0 0 0 1 2 2 0 1 3 3 5 4 2 1 1
Project no. 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Difficulty (Ci) 24 12 33 16 14 30 25 13 15 22 20 25 36 32 28
Frequency (Fi) 1 1 0 0 1 2 2 0 0 1 1 2 4 2 0
 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124 119

Fig. 6. Selection of a membership function among 30 different membership functions.

Fig. 7. Membership function according to an expert’s evaluation.

defuzzification process with the clipped center of gravity method. This produces 0.8386 as the impact
value of the risk factor, as shown in Figs. 8 and 9. This indicates that the proposed fuzzy theory changes
the expert’s evaluation into a high impact value because the project is relatively more difficult than the
others.
To compute the probability of risk occurrences, the set of states is defined by S1, S2, S3 and S4,
depending on the adjusted risk frequency (F0 i). S1 is the state for no occurrences of the target risk factor
and S4 is that occurring more than 4 times. In addition, the frequency of the target risk factor for each
project presented in Table 6 is adjusted by using Eq. (5).
120 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124

Fig. 8. Distribution of the impact value through fuzzy inferencing.

Fig. 9. Defuzzification using the clipped center of gravity method.

In order to calculate TPM, the number of transitions between states k and l, Qkl, is determined first.
Q 11 ¼ 4; Q 12 ¼ 3; Q 13 ¼ 1; Q 14 ¼ 0; Q 21 ¼ 4; Q 22 ¼ 4; Q 23 ¼ 3; Q 24 ¼ 0;
Q 31 ¼ 0; Q 32 ¼ 4; Q 33 ¼ 2; Q 34 ¼ 1; Q 41 ¼ 0; Q 42 ¼ 0; Q 43 ¼ 1; Q 44 ¼ 1:

Q11 means how many times S1 for a project is maintained or is not changed for the next consecutive
project. Qkl should be used to determine Pkl, which is TPM, as shown in Eq. (13). Then, TPM can be
represented as a state transition diagram (Fig. 10).

Table 6
Adjusted frequency (F0 i) of the risk factor and the set of states.
Project no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Difficulty (Ci) 27 20 18 9 15 36 26 12 30 9 15 25 20 28 10
Frequency (Fi) 0 0 0 1 2 2 0 1 3 3 5 4 2 1 1
Adjusted frequency (F0 i) 0 0 0 2.2 2.7 0.4 0 1.9 2.1 4.2 5.7 3.6 2.3 0.3 2.1
Sk S1 S1 S1 S3 S3 S2 S1 S2 S3 S4 S4 S3 S3 S2 S3
Project no. 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Difficulty (Ci) 24 12 33 16 14 30 25 13 15 22 20 25 36 32 28
Frequency (Fi) 1 1 0 0 1 2 2 0 0 1 1 2 4 2 0
Adjusted frequency (F0 i) 0.8 1.9 0 0 1.8 1.1 1.6 0 0 1.1 1.3 1.6 2.4 0.8 0
Sk S2 S2 S1 S1 S2 S2 S2 S1 S1 S2 S2 S2 S3 S2 S1

S1: none (F  0), S2: little (0 < F  2), S3: likely (2 < F  4), S4: very likely (4 < F).
 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124 121

Fig. 10. State transition diagram.

The initial state probability for the target risk factor is required and obtained by using the most
recent 8 projects from the 23rd project to the 30th project for S1 through S4 (refer to Eq. (14)). That is,
the number of each state is counted; for example, S2 occurs 4 times among the 8 projects.

(13)

(14)

The probability of risk occurrences can be calculated by Eq. (15). For example, the probability that
the target risk factor does not occur, or that it occurs once or twice, or that it occurs more than once, in
the 31st project is 0.3695, or 0.394, or 0.6305, respectively. These probabilities are used for computing
risk degrees.
0 1
0:5 0:375 0:125 0
B 0:364 0:364 0:272 0 C
P0 ¼ ð 0:375 0:5 0:125 0 ÞB @ 0
C
0:571 0:286 0:143 A
0 0 0:5 0:5
¼ ð 0:3695 0:394 0:2186 0:0179 Þ (15)

We have computed both the impact value of risk factor, 0.8386, and the probability of risk
occurrence, 0.6305, which, when multiplied together give the risk degree of 0.5287.

4. Discussion and implications

Various risks render the success of a NPD project somewhat difficult due to their unexpected
occurrences and the less effective responses taken against them. The CE paradigm has been introduced
and adapted by many firms as a proven and powerful concept to reduce the cost and time needed for
product development and to respond effectively and efficiently against various risks that arise in the
development phases. However, few studies have followed a systematic approach to handling the risks
under the CE environment. A new and systematic RMF needs to be developed and activated in the near
future for successful product development. The framework has to be capable of handling, analyzing,
and defending against the risks.
This paper has suggested the methodologies such as fuzzy model and Markov process to analyze
the impact of risk factors and to estimate the number of risk occurrences in a collaboration-based
product development project. These methodologies can be validated by the characteristics of risk
factors such as uncertain impact on the project and random occurrences. The fuzzy set theory can be
used for defining the impact of a risk factor on the project. Because the risk factor impact is often
uncertain, it is estimated or measured by using an expert’s experience, and the historical records in
terms of time and cost. While such a result can be reliable, it may not be consistent because different
122 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124

experts or records are used for estimating the impact and each project has unique difficulties. In this
paper, 28 different, triangular-type membership functions were defined by various project difficulties,
which are generated by IDM. Then, an expert’s evaluation of the impact caused by a risk factor
becomes an input value for the fuzzy process.
A risk factor randomly occurs during the project. The number of occurrences of a risk factor can be
obtained probabilistically from the firm’s historical records. The Markov process is used to estimate
how many occurrences of a risk factor can be expected for a new project. Then, the expected risk
degree is computed according to both the impact and the probability of risk occurrence.
The contributions made by this paper can be summarized as follows:

1. The key contribution is the logic development in order to compute the risk degree of a risk factor.
With this development, the RMF entire structure for the NPD can become a complete form, as
shown in Fig. 2. Although this paper presents the logic for only one risk factor type, the same logic
can easily be expanded to accommodate all kinds of risk factor types that occur or will occur in the
NPD project. The logic performs the critical role in the framework as a computer program to
determine the risk degree.
2. Recently, the CE philosophy has become widely accepted by industry for enhancing NPD success by
reducing the impact of risks through collaboration among all related departments. This paper
provides a model to determine the risk degree of a risk factor under a given project difficulty. The
project difficulty estimation is one of the major concerns of the CE philosophy. The same model can
easily be applied to the case of more risk factors with different types.
3. Any difference between the experts’ opinion in evaluating the impact of a risk factor will seriously
affect the decision-making on the project’s future. The fuzzy process provides a membership
function under a given project difficulty to prevent any ambiguity over the expert’s evaluation on a
risk factor.
4. The number of risk factors occurring at any time in the NPD project should be estimated logically for
successful development of a new product. The Markov process is a state-based process to estimate
probabilistically the number of risk factors as a state.

RMF still requires several additions to produce a complete form. First, the risk factors should be
classified according to their types and the design phases. The risk types include those risk factors
occurring in the areas of project management, marketing, finance, manufacturing, resources,
suppliers, and technology, while the design phases can be divided into the following six steps:
planning, concept development, system-level design, detail design, testing refinement, and
production ramp-up. Although those risk factors based on the types and phases were not mentioned
in this paper, the authors have so far collected more than 200 risk factors and classified them according
to the types and phases.
Second, the responding activities against a specific risk factor should be added to reduce the
risk degree, and also their efficiencies should also be measured logically. Again, the authors have
found more than 200 different responding activities by which the risk degree of a risk factor can be
reduced. Third, if all the risk degrees have been obtained by the types, phases, and departments,
they have to be integrated into the framework in order to determine the total risk degree of an
NPD project, as demonstrated in Eqs. (9)–(12). The integration can be done through either the
harmonic mean or the arithmetic mean. Other logical methods have to be investigated for better
integration.
The risk analysis model proposed here to determine the risk degree needs three inputs from the CE
team. One is the CL or project difficulty level specified for an NPD project. Especially, the project
difficulty can be obtained from IDM and decides its membership function corresponding to CL. The
second input is the impact value of a risk factor evaluated by an expert. This impact value can be
estimated by any numerical value in terms of time delay and cost increase due to the risk factor, but it
should be normalized to a value between 0 and 1. Then, the value is converted into an unambiguous
value through the membership function. The last input is the historical frequency data of a specified
risk factor that has occurred in past NPD projects. The expected number of the risks that will occur in a
new project can be obtained from the data.
 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124 123

Table 7
Possible responding activities to the target risk factor.
Ensure manufacturing dimensions and tolerance stack-up calculations are undertaken early and monitored
Design for manufacturing and assembly (DFMA) applied to all assemblies where tolerances are a potential issue

5. Conclusion and further research

This paper has proposed a new risk analysis model using fuzzy theory and Markov processes on a
CE basis. We determined the project difficulty by using IDM. This level is then used to select the
appropriate membership function in modeling the impact value of a risk factor, after which the
frequency adjustment process is used to model the probability of risk occurrences. The impact value of
a risk factor is calculated by the fuzzy theory, and the probability of risk occurrences is obtained by the
Markov processes. Finally, the risk degree of the risk factor that will occur in a future project is
computed by multiplying the probability of risk occurrences by the impact value of the target risk
factor. This risk degree can be used for selecting the appropriate responding activities to remove or
minimize the possible effects of the risk factor in the product development project. The possible
responding activities shown in Table 7 can be suggested for the target risk factor used in the example.
Few studies have investigated risk analysis for product development. Therefore, it is hoped that
this study will stimulate wide-ranging discussions about risk management in product development
on the CE basis. Traditional risk analysis has been used to evaluate a certain risk factor by calculating
the probability of risk occurrences according to an expert’s subjective point of view. However, the
proposed risk analysis model can be applied for various risk factors that are realized in product
development projects.
Further study is required to construct a robust RMF. New algorithms must be developed to
compute the total risk degrees of the entire system. Especially, all possible functions and algorithms
must be investigated for optimized application to and development of Eqs. (9)–(12) in the near future.
The algorithm should also include a method to determine the minimum total risk degree in a
development project by using the proper decision-making criteria. Furthermore, the performance of
the risk analysis model must be verified by application to real industries. Finally, a huge volume of
accurate data from real industries is needed to develop a more reliable, Markov-based model.

Acknowledgements

This study was supported by the Brain Korea 21 program of the Korea Research Foundation, 2008
(2008-0084-000), and by a Korea Science and Engineering Foundation (KOSEF) grant funded by the
Korea government (MEST) (R01-2006-000-10303-0, R01-2008-000-10646-0).

References

Ahn, J.O., Jeung, H.S., Kim, J.S., Choi, H.G., 2008. A framework for managing risks on concurrent engineering basis. In:
Proceedings of the IEEE International Conference on Management of Innovation and Technology, Bangkok, Thailand.
Anderson, David M., 2008. Design Guidelines. Design for Manufacturability & Concurrent Engineering. CIM Press.
Cooper, L.P., 2003. A research agenda to reduce risk in new product development through knowledge management: a
practitioner perspective. Journal of Engineering and Technology Management 20, 117–140.
Dereli, T., Baykasoglu, A., Sena Das, G., 2007. Fuzzy quality-team formation for value added auditing: a case study. Journal of
Engineering and Technology Management 24, 366–394.
DoD, 1998. Risk Management: DSMC Risk Management Workshop.
Hsieh, K.L., Tong, L.I., Wang, M.C., 2007. The application of control chart for defects and defect clustering in IC manufacturing
based on fuzzy theory. Expert Systems with Application 32, 765–776.
Kim, Y.G., Lee, T., In, H.P., 2006. A probabilistic approach to estimate the damage propagation of cyber attack. Lecture Notes in
Computer Science 3935, 175–185.
Kim I, J., 2003a. Loss and weakness of enterprise risk management. LG Weekly Economy 750, 18–24.
Kim II, J., 2003b. Risk management. How to do it. LG Weekly Economy 744–745, 28–32.
Lee, C.C., 1990. Fuzzy logic in control systems: fuzzy logic controller. IEEE Transactions on Systems, Man, and Cybernetics 20 (2),
404–435.
Savic, S., Kayis, B., 2006. Knowledge elicitation for risk mapping in concurrent engineering projects. International Journal of
Production Research 44 (9), 1739–1755.
124 H.-G. Choi, J. Ahn / Journal of Engineering and Technology Management 27 (2010) 110–124

Singh, N., 1996. Concurrent Engineering. System Approach to Computer-Integrated Design and Manufacturing. John Wiley &
Sons, pp. 103–147.
Willaert, S.S., Graaf, R., Minderhoud, S.d., 1998. Collaborative engineering: a case study of concurrent engineering in a wider
context. Journal of Engineering and Technology Management 15, 87–109.
Zadeh, L.A., 1965. Fuzzy sets. Information and Control 8, 338–353.