Beruflich Dokumente
Kultur Dokumente
Developing an integrated model for the evaluation and selection of six sigma
projects based on ANFIS and fuzzy goal programming
Abbas Saghaei *, Hosein Didehkhani
Science and Research Branch, Islamic Azad University, Tehran, Iran
a r t i c l e
i n f o
Keywords:
Six sigma
Project selection
Adaptive neuro fuzzy inference systems
Fuzzy goal programming
a b s t r a c t
Six sigma is one of the most popular tools to eliminate waste in organizations, reduce the cost and
improve quality. The process of creating and evaluating projects is an initial activity in implementing
six sigma. This paper aims at proposing a comprehensive methodology for the evaluation and selection
of the six sigma projects. For the evaluation of projects, reviewing the literature and decision teams opinion, we identied three main categories of criteria including business criteria, technological & process criteria and nancial criteria which contain eight sub-criteria. For deriving the overall utility of projects, we
designed an adaptive neuro fuzzy inference system which is capable to consider interrelations among criteria. Then, applying a fuzzy weighted additive goal programming model, we obtained the optimal portfolio of projects which should be implemented. Finally, we applied the proposed model in a leading
company in Iran to illustrate the applicability of the model.
2010 Elsevier Ltd. All rights reserved.
1. Introduction
Nowadays, the environment of manufacturing and service is
completely competitive. Companies are trying to reach and sustain
a competitive competency to protect their market share. But
reaching this aim is not possible without applying the concept of
a continuous improvement. The Japanese were pioneer in applying
quality concepts and continuous improvement. They have reported their impressive benets in both nancial on non-nancial
parts.
Recently, we have faced with concepts such as quality management, customer satisfaction, cost of poor quality, quality improvement and as likes. These concepts relate to a wide range category
named Total Quality Management (TQM). Six sigma is one of the
most applicable approaches helps organizations to implement
TQM concepts. Since its innovation by the late Bill Smith of Motorola, six sigma concepts have been widely utilized in the world of
industry. Using statistical techniques and approaches, six sigma
tries to eliminate subjectivity of decision making process and propose a systematic way to identify the problems and defects drivers.
Furthermore, six sigma presents some solutions for eliminating
these defects and improves the process capability to achieve a
business excellence and a competitive edge. Nowadays six sigma
has been considered as the most applicable and useful approach
722
Table 1
Review of six sigma project selection criteria.
Criteria
Snee (2002)
Banuelas et al. (2005)
Kahraman and
Bykzkan (2008)
Adam et al. (2003)
Yang and Hsieh (2008)
723
3. Model construction
In this section, we describe stages of the proposed methodology
elaborately. The model includes two stages: The evaluation stage
and the selection stage. In the rst stage, we develop an adaptive
neuro fuzzy inference system to derive the total utility of each project considering project selection criteria. In the second stage, we
develop a fuzzy binary weighted additive goal programming model
for deriving the optimal portfolio of projects.
3.1. Evaluation Stage: designing an ANFIS for deriving overall utility of
projects
fuzzy reasoning. Fuzzy reasoning, also known as approximate reasoning, is an inference procedure that derives conclusions from a
set of fuzzy if-then rules and known facts. The fuzzy inference system is a popular computing framework based on the concepts of
fuzzy set theory, fuzzy if-then rules, and fuzzy reasoning. It has
been applied successfully in a wide range of science and engineering such as control, function approximation, signal processing,
simulation, data clustering and data mining and decision support
systems. In literature, we can nd some other names such as fuzzy
rule-based system, fuzzy expert system, fuzzy model, fuzzy associative memory, fuzzy logic controller, and simply (and ambiguously) fuzzy system. Fuzzy IF-THEN rules are the basic on fuzzy
inference systems (Jang, Sun, & Mizutani, 1996). A fuzzy rule is
in the following form:
IF premise (antecedents), THEN conclusion (consequents).
In application cases, there is always more than a one rule. There
has been developed some inference methods such as Mamdani
inference system (1977), and Takagi and Sugeno inference systems
(1985). Conventional inference systems, in spite of their applicability, have some shortages such as: there is no systematic procedure
for rule generation and validation, and also there is no systematic
approach for optimizing the system parameters such as membership function parameters to achieve more accurate solutions. Jang
(1993) developed a network-based inference system which was
capable to be trained from experiments. He called that ANFIS,
which is the abbreviation of adaptive neuro fuzzy inference system. He applied least square estimator (LSE) and back-propagation
(BP) to train the system. The most common inference system used
in ANFIS is a rst order Sugenu system which is in the form of:
e 1 and x2 is A
e 1 . . . and xn is A
e 1 then y f1 x1 ; x2 ; . . . ; xn
If x1 is A
1
1
2
n
e m and x2 is A
e m . . . and xn is A
e m then y fm x1 ; x2 ; . . . ; xn
If x1 is A
1
Table 2
Six sigma project selection criteria.
Criteria
Sub-Criteria
Financial benets
Process improvement
1
Where, xi; i = 1, . . . , n, are crisp inputs of system and yj is a crisp linear combination of inputs. The output of the system is calculated as
the weighted average of output functions as below:
Pm
j1 yj
y
Pm
j1
n
Q
i1
n
Q
i1
lAi xi
2
lAi xi
In the rst layer, the rates of each project with respect to criteria are
fed into the network. These ranks are derived by a decision team.
The layer 2 includes fuzzy partitions of each input. The model is
capable to use different types of membership functions. Also, we
can choose different numbers of membership functions (MFs) to
partition the criteria space. Each MF has some parameters and
ANFIS during the training process will tune these parameters into
present desirable outputs with the minimum error. In layer 3, we
have some rules that can be derived by manager or by partitioning
of the criteria (input) space. Clustering the inputs is another
efcient way for a rule generation. Traditional K-means and fuzzy
c-means are criticized because we should impose the number of
clusters. Mountain clustering developed by Yager and Filev (1994)
is an efcient clustering approach which approximates the center
of clusters by using a density function called mountain function.
This approach uses the grid points as the alternatives of cluster centers. Chiu (1994) used data points as clusters center alternative instead of grid points in mountain clustering and called the method
subtractive clustering. In this paper, we use subtractive clustering
for rule generation. In the last layer, we aggregate the consequents
(yj; j = 1, . . . , m) to obtain the output of the system. In a hybrid learning method, consequent parameters are tuned too.
724
Up
N
X
~ g
ui xi Pu
i1
3.2. Selection Stage: obtaining the portfolio of six sigma project using a
fuzzy weighted additive goal programming model
Goal programming has been a widely discussed and applied
technique for solving decision problems with multiple criteria. In
classical GP models, unpleasant deviations from goal values and
constraints dened by decision maker are minimized to derive a
desirable solution. The unwanted deviations are measured using
positive and negative deviation variables that are dened for each
and they represent overachievement of the goal, respectively. After
introducing the concept of fuzzy decision making by Bellman and
Zadeh (1970), researchers applied fuzzy logic with a Goal Programming. In classical goal programming models, all goals have equal
importance. Weighted additive models developed to consider different importance weights of goals and constraints. These models
use a single utility function that is the weighted summation of
goals and constraints satisfaction level (membership functions).
The weighted additive model, proposed by Tiwari, Dharmar, and
Rao (1987) is:
lD x
p
X
wj lzj x
wj
h
X
br lgr x
r1
j1
p
X
h
X
br 1
r1
j1
Parameters wj and br are normalized weighting coefcients represent relative importance among objectives and constraints. In fuzzy
goal programming, objectives and constraints are treated in the
same way. But, by varying weighting parameters, we can prioritize
objectives and constraints and emphasize on goals should be satised more strictly. By identifying variables kj and cr, the model can
be converted into this crisp single objective model:
Max
p
X
wj kj
h
X
cr 6 lgr x r 1; 2; . . . ; h
j1
wj
h
X
N
X
~
C i xi 6B
i1
Fp
N
X
~ g
NPV i xi PF
i1
So, by combining (5), (6) and (7), the FGP model for deriving portfolio of projects is:
p h 1; . . . ; m
ku 6 lu X
kc 6 lc X
kf 6 lf X
xi 2 f0; 1g
4
kj ; cr 2 0; 1 j 1; 2; . . . ; q and r 1; 2; . . . ; h
p
X
Cp
s:t:
br cr
s:t : kj 6 lzj x j 1; 2; . . . ; q
g p x 6 bp
Max wu ku wc ku wf kf
r1
j1
br 1; wj ; br P 0
r1
Where, X = (x1, x2, . . . , xN) and wu, wf and wc are the relative importance weights of goals.
4. Model implementation
For proving the applicability of the model and illustration, we
applied the model in one of the leading companies in Iran. The rst
step to apply the model was to construct the decision team. We
constructed the decision team including top and middle managers:
strategic planning manager, nancial manager, engineering manager, quality control and insurance manager and process manager.
Then, we asked them to identify the assessment criteria and rate
the potential projects with respect to each criterion in the range
of [0, 10]. Decision team identied the criteria as mentioned in
Table 1.
4.1. Training ANFIS and obtaining the overall utility of projects
For training the ANFIS, we needed some experiences about the
system behavior. For this aim, we designed a questionnaire including different combinations of criteria and asked the decision team
to give a utility score to them if possible at all, based on their
knowledge about the system. But where the number of antecedents (criteria) is large, it is not practical to ask the decision team
725
Layer 1: Criteria
Layer 4: Utilities
R1
Business
criteria (BC)
R2
Fuzzy partitions
Financial
criteria (FB)
Utility Score = w i * f i
i =1
Aij
Rn
Process
criteria (PI)
Fig. 2. (a) Trained sub-ANFIS (I) surface: BC and business excellence-customer satisfaction. (b) Trained sub-ANFIS (II) surface: BC and ROI-Cost of appraisal.
726
Fig. 3. Trained sub-ANFIS (III) surface: PI and process variability and non value-added.
Table 3
Inputs (rates of projects) and outputs of ANFIS and sub-ANFISs.
Project code
CS
BE
SP
sub-ANFIS (1)
BC
ROI
RS
SP
sub-ANFIS (2)
FB
VC
NV
sub-ANFIS (3)
PI
ANFIS
Overall utility (OU)
A
B
C
D
E
F
G
5
3
8
5
4
7
2
5
8
6
6
6
6
4
6
3
7
2
7
7
4
5.2
4.5
7.2
4.7
5.2
6.7
3
5
4
6
5
3
5
6
4
5
5
6
6
6
3
3
2
7
2
8
7
2
4.45
3.95
5.9
4.8
4.5
5.55
4.65
7
9
4
4
4
6
5
8
3
8
7
8
5
7
7.3
7.2
5.2
4.9
5.2
5.7
5.6
4.72
4.23
5.65
4.19
4.23
5.17
3.51
Table 4
Initial and optimal data of projects.
Project
NPV
Initial investment
Utility score
Value
Status
a
b
c
d
e
f
g
50000
33000
45000
75000
38000
55000
25000
11500
8000
12000
12700
10000
11500
10000
4.72
4.23
5.65
4.19
4.23
5.17
3.51
0
1
0
0
1
1
0
Rejected
Selected
Rejected
Rejected
Selected
Selected
Rejected
727
lc X
8
>
>
>
1
>
>
>
>
>
>
>
<
1
>
>
>
>
>
>
>
>
>
>
:0
if
N
P
tuned. In ANFIS, it is not important which category the criteria belong to, in spite of MCDM models. In this paper, we considered a
wide range of subjective and objective criteria affect the utility of
a six sigma project. We categorized them into three main categories. We used these criteria to derive the overall utility of projects
by ANFIS. Also, we developed a FGP model considering three main
goals to derive the optimal portfolio of project should be done. We
applied the proposed model in a leading company in Iran to illustrate the applicability of model. By using model results, company
could select an efcient portfolio of six sigma projects to handle
its problems considering the constraints of company such as budget and desired nancial benets level. For further researches,
model parameters such as NPV of projects and initial capital cost
can be considered fuzzy, as they are in nature, and by using a possibilistic models instead of FGP models, derive the efcient portfolio of projects.
Appendix A
See Fig. A1.
c i xi 6 B
i1
N
P
Appendix B
ci xi B
i1
ac
if
B6
N
P
ci xi 6 B ac
i1
if
N
P
ci xi P B ac
i1
OU =
OU =
OU =
OU =
OU =
OU =
OU =
OU =
OU =
OU =
OU =
728
Appendix C
Crisp LP model for deriving the optimal portfolio of projects
Kahraman, C., Ertay, T., & Buyukozkan, G. (2006). A fuzzy optimization model for
QFD planning process using analytic network approach. European Journal of
Operational Research, 171, 390411.
Kumar, U. D., Nowicki, D., Ramirez-Marquez, J. E., & Verma, D. (2008). On the
optimal selection of process alternatives in a six sigma implementation.
Internationl Journal of Production Economics, 111, 456467.
References
Adam, W. C., Gupta, P., & Wilson, C. E. (2003). Six sigma deployment. USA:
Butterworth-Heinemann.
Banuelas, R., Antony, J., & Brace, M. (2005). An application of six sigma to
reduce waste. Quality and Reliability Engineering International, 21(6),
553570.
Bellman, R., & Zadeh, L. A. (1970). Decision-making in a fuzzy environment.
Management Science, 17, 141164.
Chao, T. S., & Chia, J. C. (2008). A systematic methodology for the creation of six
sigma projects: A case study of semiconductor foundry. Expert Systems with
Applications, 34, 26932703.
Chiu, S. (1994). Fuzzy model identication based on cluster estimation. Journal of
Intelligent and Fuzzy Systems, 2(3), 267278.
Gryna, F. M., Chua, R. C. H., & Defeo, J. A. (2005). Jurans quality planning and analysis:
for enterprise quality. McGraw-Hill.
Jang, R. (1993). ANFIS: Adaptive-network-based fuzzy inference systems. IEEE
Transactions on Systems, Man, and Cybernetics, 23, 665685.
Jang, R., Sun, C. T., & Mizutani, E. (1996). Neuro-fuzzy and soft computing. Prentice
Hall.
Kahraman, C., & Bykzkan, G. (2008). A combined fuzzy AHP and fuzzy goal
programming approach for effective six-sigma project selection. Journal of
Multiple-Valued Logic and Soft Computing, 14(6).
Kumar, U. D., Saranga, U., Ramirez-Marquez, J. E., & Nowicki, D. (2007). Six sigma
project selection using data envelopment analysis. TQM magazine, 19(5),
419441.
Lanyon, S. (2003). At Raytheon six sigma works, too, to improve HR management
processes. Journal of Organizational Excellence, 22(4), 2942.
Mamdani, E. H. (1977). Application of fuzzy logic to approximate reasoning using
linguistic systems. Fuzzy Sets and Systems, 26, 11821191.
Robinson, B. (2005). Build a management system based on six sigma. ASQ six sigma
Forum Magazine, 5(1), 2834.
Roger, G. S., Linderman, K., Liedtke, C., & Choo, A. S. (2008). Six sigma: Denition and
underlying theory. Journal of Operations Management, 26, 536554.
Snee, R. D. (2002). Dealing with the Achilles heel of six sigma initiatives project
selection is key to success. Quality Progress, 34(3), 6669.
Takagi, T., & Sugeno, M. (1985). Fuzzy identication of systems and its application
to modeling and control. IEEE Transactions on Systems Man and Cybernetics,
15(1), 116132.
Tiwari, R. N., Dharmar, S., & Rao, J. R. (1987). Fuzzy goal programming An additive
model. Fuzzy Sets and Systems, 24, 2734.
Yager, R. R., & Filev, D. P. (1994). Approximate clustering via the mountain method.
IEEE Transactions on Systems Man and Cybernetics, 24(8), 12791284.
Yang, T., & Hsieh, T. H. (2008). Six-sigma project selection using national quality
award criteria and fuzzy multiple criteria decision-making method. In
Proceeding of WiCOM08 fourth international conference.