Article information:
To cite this document:
Yicha Zhang Alain Bernard , (2014),"An integrated decision-making model for multi-attributes decision-making (MADM)
problems in additive manufacturing process planning", Rapid Prototyping Journal, Vol. 20 Iss 5 pp. 377 - 389
Permanent link to this document:
http://dx.doi.org/10.1108/RPJ-01-2013-0009
Downloaded on: 01 July 2015, At: 08:21 (PT)
References: this document contains references to 27 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 238 times since 2014*
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service
information about how to choose which publication to write for and submission guidelines are available for all. Please visit
www.emeraldinsight.com/authors for more information.
IRCCyN, Ecole Centrale de Nantes, LUniversit Nantes Angers Le Mans, Nantes, France
Abstract
Purpose The purpose of this paper is to propose an integrated decision-making model for multi-attributes decision-making (MADM) problems in
additive manufacturing (AM) process planning and for related MADM problems in other research areas.
Design/methodology/approach This research analyzed the drawbacks of former methods and then proposed two sub-decision-making models,
deviation model and similarity model. The former sub-model aimed to measure the deviation extent of each alternative to the aspired goal
based on analyzing Euclidean distance between them, whereas the latter sub-model applying grey incidence analysis was used to measure the
similarity between alternatives and the expected goal by investigating the curve shape of each alternative. Afterwards, an integrated model based
on the aggregation of the two sub-models was proposed and verified by a numerical example and simple case studies.
Findings The calculating results of the cited numerical example and the comparison to former related research showed that this proposed model
is more practical and reasonable than former methods applied in MADM problems of AM. In addition, the proposed model can be applied in other
fields where MADM problems exist.
Originality/value This proposed integrated model not only considered the deviation extent of alternatives to the aspired goal but also
investigated the similarity between alternatives and the expected goal. The similarity analysis compensates the drawbacks of traditional
distance-based models or methods that cannot distinguish alternatives which have the same distance-based index value.
Keywords Multi-attribute decision-making, Process planning, Additive manufacturing, Grey incidence analysis, Similarity measuring
Paper type Research paper
1. Introduction
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1355-2546.htm
377
Cd
D
i
/
i1
(1)
i1
Vector-b: Vb timeb/s.
Db
Vector-a: Va timea/s,
(3)
Vector-goal: Vg time/s 2s ,
Da
a
g
g
1
1 / eA A / A 100%
eD
(2)
(5)
Vj j1,j2,j3,. . .jn
(6)
(7)
(8)
(11)
(12)
When the two vectors have been processed, the grey incidence
of the two vectors can be calculated by:
ij
1 si sj
1 si sj si sj
(13)
n1
si
k2
n1
sj
k2
n1
si sj
k2
1
oi k oi n ,
2
(14)
1
oj k oj n and
2
(15)
1
oi koj k oi noj n , (16)
2
(17)
where C*i denotes the general index value of the ith alternative,
Cdi
, and
(18)
d
i
i1
i'
(19)
i1
d
i
d
i
A (accuracy)
R (surface roughness)
S (tensile strength)
E (elongation)
C (cost/part)
B (built time)
SLA3500
SLS2500
FDM8000
LOM1015
Quadra
Z402
120
150
125
185
95
600
6.5
12.5
21
20
3.5
15.5
65
40
30
25
30
5
5
8.5
10
10
6
1
Very high
Very high
High
Slightly high
Very high
Very very low
Medium
Medium
Very high
Slightly low
Slightly low
Very low
382
VS S1, S2, S3, S4, S5, S6 65, 40, 30, 25, 30, 5
SLA3500
SLS2500
FDM8000
LOM1015
Quadra
Z402
120
150
125
185
95
600
6.5
12.5
21
20
3.5
15.5
65
40
30
25
30
5
5
8.5
10
10
6
1
0.745
0.745
0.665
0.59
0.745
0.135
0.5
0.5
0.745
0.41
0.41
0.225
VB B1, B2, B3, B4, B5, B6 0.5, 0.5, 0.745, 0.41, 0.41, 0.255
dAi
Ai
Ai
6
Ae
1
A
6 i1 i
(20)
W1 11, 12, 13 . . . . . . 16
0.1113, 0.1113, 0.0634, 0.0634,0.3253,0.3253
W2 21, 22, 23 . . . . . . 26
0.319,0.319,0.129,0.129,0.052,0.052
......
and
DVB dB1, dB2, dB3, dB4, dB5, dB6
1.0638, 1.0638, 1.5851, 0.8723, 0.8723, 0.5426
The fifth step is to reconstruct the alternative vectors using the
elements of processed attribute vectors, and the results are
VA A1, A2, A3, A4, A5, A6 120, 150, 125, 185, 95, 600
VR R1, R2, R3, R4, R5, R6 6.5, 12.5, 21, 20, 3.5, 15.5
Table III Ranking results of the RP processes
Cd
RP process
C /
C /
1/
2/
C /
SLA3500
SLS2500
FDM8000
LOM1015
Quadra
Z402
0.1439
0.1145
0.0789
0.1404
0.1778
0.3366
0.5062
0.2569
0.1263
0.1234
0.6287
0.0486
0.8368
0.8492
0.7985
0.9113
0.8566
0.6397
0.9120
0.8649
0.8815
0.8229
0.8953
0.5708
0.1581
0.1445
0.1214
0.1639
0.1772
0.2350
d
1
d
2
383
*
1
C* (0.5)
Rank
C*2/2
4
5
6
3
2
1
0.2419
0.1634
0.1265
0.1197
0.2765
0.0721
Rank
2
3
4
5
1
6
Table V The RPSI values for different RP systems with two set of
weights
Table 5RP
systems
RPSI A and R
are more
important
SLA3500
SLS2500
FDM8000
LOM1015
Quadra
Z402
9.4606
8.0812
7.5736
7.6081
10.7085
6.6198
Rank
RPSI C and B
are more
important
Rank
2
3
5
4
1
6
9.4849
8.1792
7.6659
7.7134
10.5126
6.5914
2
3
5
4
1
6
......
wVZ402' 1dA6', 2dR6', 3dS6', 4dE6', 5dC6', 6dB6'
where iw, i 1, 2, 3. . .6 ,w 1, 2, 36 is the
assigned set of weights for the related attributes. Likewise, the
goal vector can be interpreted by all of the best values of each
attribute from the reconstructed alternatives, as
Vgoal jidA5', jidR5', jidS1', jidE3', jidC6', jidB6'
where, i 1, 2 . . . . . . 6; j 1, 2.
The following step is calculating each alternatives grey
incidence values by using the recovered alternative vectors and
goal vectors through the similarity model, and the final step is
to aggregate the two sets of sub-index values from the results of
the two sub-models into one set of general index values by
applying the integrated model. When using the models to
calculate, the coefficient of the integrated model can be set as
0.5 in this case. The obtained results are presented in Table III
according to the two sets of assigned weights. The ranking of the
RP process is based on the general index values, which means
that the one who has the largest index value would be the most
recommended alternative among the finite choices.
To set a comparison against the methods of former studies
based on the same numerical example, two sets of former
research results are also cited and given in Tables IV and V.
Obviously, the results presented in this paper are consistent with the
results of the original former research conducted by Byun and Lee
RP process
C*i
Rank
C*i
Rank
SLA3500
SLS2500
FDM8000
LOM1015
Quadra
Z402
0.4516
0.4250
0.3009
0.5686
0.5039
0.6640
4
5
6
2
3
1
0.8353
0.7060
0.5799
0.5552
0.8427
0.1721
2
3
4
5
1
6
384
385
two parts, and those scenarios are outlined in Tables VIII and
IX. To make comparison, a group of the same six scenarios are
chosen for the two parts. During consulting the database, no
preference or specification was provided to KARMA to
enlarge the number of available alternatives. For simplicity
sake, only five key attributes, part cost, surface roughness,
build time, tensile or yield strength and part density, are
selected to represent those proposed scenarios as indicated in
the two tables. The values in the table are the average
prediction quantities based on AM expert databases
evaluation and prediction by applying several empirical
algorithms or models. Therefore, to select an optimal choice
from the available alternatives means to analyze these
multi-attributes and make a trade-off among them based on
specific preferences and application requirements.
In this case, only the general application requirements are
offered. For the frame structure part (part 1), surface
roughness is the foremost important factor, and cost and build
time follow in the second place. Requirements for the
mechanical properties of the part are medium because this
part would not bear large loads but has a high requirement for
assembly. While for the ratchet part (part 2), mechanical
properties are the most important, as well as the surface
roughness and density. Build time and cost are not so
important in consideration due to the direct production and
customized design. Hence, related weights should be assigned
to each of the selected attributes to express the application
requirements while executing decision-making. In this case,
1
50 in/s
150 in/s
103 mW
65.5 in/s
1.0028/1.0030/1
1
0.9843 in/s
29.3 C
0.1 mm
22.1 in/s
144.3 in/s
5
Average
Minimum
Maximum
1.2
61.38
2128
10.9
92.9
2560
30
80.6 [Shore D]
0.69
50.9
58.45
0.04 [Inclination: 0]
18.89 [Inclination: 30]
10.06 [Inclination: 60]
2.38 [Inclination: 90]
1.2
59.9
2080
8.2
91.6
2480
14
80 [Shore D]
0.6
50.9
58.3
0.04 [Inclination: 0]
18.89 [Inclination: 30]
10.06 [Inclination: 60]
2.38 [Inclination: 90]
1.21
62.8
2200
15
94.1
2600
56
82 [Shore D]
0.79
50.9
58.6
0.04 [Inclination: 0]
18.89 [Inclination: 30]
10.06 [Inclination: 60]
2.38 [Inclination: 90]
Cost/()
Roughness/(m)
Build
time/h
Tensile or yield
strength/(MPa)
Density/(g/cm3)
47.7
35.08
211.42
146.14
146.14
481.78
3.49
7.8
3.12
19.03
19.81
24.96
5.4
2.32
6.66
3.4
3.4
9.02
61.38
55.10
475
47.22
37.92
936.6
1.2
1.17
7.8
1.05
1.32
4.42
386
Cost/()
Roughness/(m)
Build
time/hours
Tensile or yield
strength/(MPa)
Density/(g/cm3)
231.34
157.47
1006.84
413.75
404.09
1217.47
6.44
10.62
4.25
21.13
21.26
26.53
20.48
13.64
37.52
7.76
7.33
23.79
61.38
50.84
475
47.22
39.24
936.6
1.2
1.17
7.8
1.05
1.32
4.42
When the goal vectors are set, the following calculation steps
are the same with those procedures introduced in the former
sections of this paper. After calculating through the integrated
model, the general index values of these scenarios can be
obtained and outlined in Table XI. The coefficient, of the
integrated model used in this case can also be set as 0.5.
By referring to the index values, the best scenario for Parts
1 and 2 is the same scenario, 3D system Viper Protogen
18420. According to AM production experience, the results
presented here are reasonable. And obviously, the ranking
results of the two sub-models are different according to their
different index values, as one is a distance-based model and
the other investigates the similarities between alternatives and
aspired goals. Hence, as proposed in this paper, integrating the
two sub-models together is more reasonable and can give
more available decision support.
Vgoal1 C2, R3, B2, T6, D4 35.08, 3.12, 2.32, 936.6, 1.17 and
5. Discussion
Vgoal2 C2, R3, B5, T6, D4 157.47, 4.25, 7.33, 936.6, 1.17)
Cost
Build
time
Surface
roughness
Tensile/yield
strength
Density
0.15
0.15
0.5
0.1
0.1
0.05
0.05
0.15
0.6
0.15
Cd1
0.6570
0.4251
0.1779
0.0412
0.0354
0.0021
0.6505
0.4877
0.2744
0.2136
0.2092
0.1646
387
Rank
Cd2
1
2
3
4
5
6
0.4618
0.4264
0.1763
0.2865
0.2747
0.1797
Part 2 (ratchet)
2
C*2
0.6362
0.6403
0.8045
0.6401
0.6387
0.9940
0.4019
0.3833
0.2825
0.3057
0.2989
0.3278
Rank
1
2
6
4
5
3
choices, the best one may not be the one he or she wants.
Then, the user can choose what he or she needs by setting a
special aspired goal which chooses medium attribute values
but not always the best or the limits. And the application of
this model is simple and easy without the need of importing
complicated parameters with uncertainty, such as relative
importance which is usually difficult to obtain without the
subjective and bias effect. Furthermore, it has great potential
to be used as reasoning engines for knowledge-based expert
systems which use knowledge vectors as data format due to its
convenience, and to be adopted in other application areas
where MADM problems exist.
6. Conclusion
To tackle the MADM problems in process planning for
AM, a new integrated decision-making model is proposed
in this paper. This model compensates the drawbacks of
some former distance-based models or methods by
measuring the similarity between alternatives and the
aspired goal. And the best choice among a set of finite
alternatives has been redefined as that the optimal choice
among a finite set of alternatives is the one that has the least
distance deviation and the largest similarity to the aspired
goal. Hence, it is more useful to help users make better
decisions while encountering MADM problems. This
model has great potential to be applied as a reasoning
engine in a generic AM process planning expert system that
would be investigated further. Further research would
explore how this model can be used or integrated in the
decision-making while setting detailed parameters for
knowledge-based process planning in AM.
References
Allen, S. and Dutta, D. (1995), Determination and
evaluation of support structures in layered manufacturing,
Journal of Design and Manufacturing, Vol. 5 No. 3,
pp. 153-162.
Bernard, A. and Deglin, A. (1999), Knowledge-based system
for the choice of rapid prototyping process, Proceedings of
the 10th Solid Freeform Fabrication Symposium, Austin, TX,
pp. 39-45.
Byun, H.S. and Lee, K.H. (2005), A decision support system
for the selection of a rapid prototyping process using the
modified TOPSIS method, The International Journal of
Advanced Manufacturing Technology, Vol. 26 Nos 11/12,
pp. 1338-1347.
Chen, S.J. and Hwang, C.L. (1992), Fuzzy Multiple Attributes
Decision Making Methods and Application, Springer, Berlin
Heidelberg, New York, NY.
Helsinki University of Technology (2002), RP selector,
available at: http://ltk.hut.fi/RP-Selector (accessed 12
December 2012).
Herrmann, A. and Allen, J.K. (1999), Selection of rapid
tooling materials and processes in a distributed design
environment, ASME Design for Manufacturing Conference,
Las Vegas, pp. 12-15.
KARMA (2010), available at: www.femeval.es/proyectos/
karma/Paginas/InicioKarmaNoLogo.aspx (accessed 16
January 2013).
388
Yang, Y., Loh, H.T., Fuh, F.Y.H. and Wang, Y.G. (2002),
Equidistant path generation for improving scanning
efficiency in layered manufacturing, Rapid Prototyping
Journal, Vol. 8 No. 1, pp. 30-37.
Corresponding author
389