Professional Documents
Culture Documents
A Method For Optimal Material Selection Aided With Decision Making Theory
A Method For Optimal Material Selection Aided With Decision Making Theory
A Method For Optimal Material Selection Aided With Decision Making Theory
199᎐206
Abstract
As a CAE ŽComputer Aided Engineering. tool to help design engineers, the procedure of material selection should be
objective so as to minimize personal bias and it should be able to be coded into the software of ‘expert system’. In this work, we
have utilized the theories of decision-making. One of them is the concept of entropy; to evaluate the weight factor for each
material property or performance index, and the other is TOPIS; to rank the candidate materials, for which several requirements
are considered simultaneously. As an example, the procedure to select the optimal material for a flywheel has been developed.
We can grade the candidate materials for given subjective weight. In the cases for which fatigue strength or fracture toughness is
weighted, the chosen materials coincide with the ones utilized for a commercial flywheel in support of our approach. 䊚 2000
Elsevier Science Ltd. All rights reserved.
0261-3069r00r$ - see front matter 䊚 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 2 6 1 - 3 0 6 9 Ž 9 9 . 0 0 0 6 6 - 7
200 D. Jee, K. Kang r Materials and Design 21 (2000) 199᎐206
until every adjoining process is complete, they sug- 2. Material selection procedure
gested an object model of knowledge-based material
selection in which materials are chosen iteratively in 2.1. Quantifying the requirements
parallel with the requirements of the link. The relation-
ship between linked components is stored in a software
structure. First, the material properties that should be taken
Garton et al. w4x used Fatigue Property Charts to into account are identified and the requirements of
select the optimal class or subclass of material in each material property are formulated. The require-
minimum weight design for infinite fatigue life. Design ments of a number of material properties are quantita-
criteria were displayed in the charts and provide a tively formulated. Other properties Žfor example, corro-
systematic basis to the design process. Takuma et al. w5x sion resistance, aesthetics, etc.. can be formulated only
constructed a neural network which learned the pat- qualitatively, such as YesrNo or PoorrGoodrExcel-
tern sets dealing with the interaction between various lent. Table 1 shows typical examples of qualitative and
mechanical strengths and material ingredients. They quantitative material properties. We use a bi-polar
argue that the neural network, a kind of artificial method w9x to grade a material for each qualitative
intelligence, can be used successfully to evaluate an requirement. For example, regarding degree of 0᎐10,
optimal material for the given performance of a 10 means the most desirable while 0 means the most
product. undesirable; and 5 means the medium between them.
Generally a number of criteria or requirements But in the case of material properties concerning pro-
should be simultaneously considered to select a mate- fit, a degree of 9 is three times as desirable as 3.
rial for a given component. Few studies except for
2.2. Combination of requirements
Sandstrom w1x, however, have been done to provide a
basis on managing several criteria or requirements and
ranking the candidate materials. In this paper, an ap-
In selecting materials for a given component one
proach to choose the optimal material for a given should consider all requirements simultaneously. For
component is described, and material properties are convenience of procedure it is beneficial to combine
classified into qualitative and quantitative properties some of them. The performance indices proposed by
according to Waterman and Ashby w6x. Performance Waterman and Ashby w6x are a useful tool. Only for
indices w6,7x are evaluated for a component under de- requirements which can be quantitatively formulated,
sign. They are used as functional parameters in a the material properties are combined to be a perfor-
rational procedure of material selection. To decide the mance index, using physical relation or equations
relative ranking among candidate materials according among them.
to several requirements, Yoon’s w8x algorithm TOPSIS
ŽTechnique for Order Preference by Similarity to Ideal 2.3. Building a material properties matrix
Solution., a kind of decision making theory, is utilized.
The compatibility with other components linked with in
a larger assembly is not considered. The approach The values of material properties or performance
adopted here is meant to help a design engineer make indices for each candidate material are expressed in a
sensible decision on materials selection. material property matrix as
Table 1
Typical examples of quantitative and qualitative material properties
X1 ⭈⭈⭈ Xj ⭈⭈⭈ Xn sj j
jU s n . Ž4.
A1
x 11 ⭈⭈⭈ x1 j ⭈⭈⭈ x1 n Ý sj j
⭈⭈ js1
⭈⭈ ⭈⭈ ⭈⭈
D s Ai
x i1 ⭈⭈⭈ xi j ⭈⭈⭈ xin
⭈⭈ It would be reasonable that the weight s j is assigned a
Am ⭈⭈ ⭈⭈ ⭈⭈ default by an expert who has experience or knowledge
x m1 ⭈⭈⭈ xm j ⭈⭈⭈ xmn enough about materials and design of the component
under question. But if the design circumstance is un-
usually changed Žfor example, a sudden rise of the
where, A1 ,..., A m are the kinds of candidate materials price or new regulation for environment protection.,
and X 1 ,..., X n are the material properties or perfor- one may adjust s j at his option. With the weight
mance indices. Hence, x i j is the value or degree of the factors, the material properties matrix is reconstructed
j-th property or index X j for the i-th material A i . as:
2.4. Normalization
¨ i j s jU ri j Ž no sum on j.. Ž5.
The various material properties and performance
indices have different physical dimensions from each If the subjective weight s j is not assigned, jU s j .
other. In order to make a comparison between proper-
ties, the elements of the matrix are normalized as: 2.6. E¨ aluation of the separation measure and relation
closeness
m
r i j s x i jr Ý x i j Ž1.
is1
We apply a kind of decision making theory ᎏ TOP-
SIS ŽTechnique for Order Preference by Similarity to
Ideal Solution. ᎏ proposed by Yoon w8x, to rank candi-
where ri j is in the range of 0᎐1.
date materials. The basic idea of TOPSIS is that the
best decision should be made to be closest to the ideal
2.5. Determination of weight factor
and farthest from the non-ideal. In the material
properties matrix whose elements are normalized and
To prescribe the relative priority among the material
weighted according to Eq. Ž1. and Eq. Ž5., the element
properties and performance indices, a weight factor is
with the most preferred value Ži.e. the highest value in
given to each of them. The weight factors are evaluated
most except for costs. for the j-th material property or
using the concept of entropy proposed by Shannon and
performance index is defined as the ideal ¨ q j , and the
Weaver w10x. The entropy Ej of the normalized values
element with the least preferred value is defined as the
of a j-th material property or index is defined as
non-ideal ¨ yj . Hence, the matrices V
q
and V y which
follows: q y
consist of ¨ j and ¨ j , respectively, are expressed as the
m
equations:
Ej s yk Ý ri j log ri j Ž2.
min
¨i j < j g J . , Ž i ¨ i j < j g J U . < i s 1, . . . ,m
max
is1
Vqs ½Ž i 5
where k s 1rlog m. Ej is also in the range of 0᎐1. The s Ž ¨q y y.
1 , ⭈⭈⭈ ,¨ j , ⭈⭈⭈ ,¨ n
Ž6.
weight factor j for the j-th material property or index
is defined as:
min max
Vy ½ž ¨i j < j g J , /ž ¨ i j < j g J U < i s 1, ⭈⭈⭈ ,m
/ 5
1 y Ej i i
j s n . Ž3.
Ý Ž 1 y Ej . s Ž ¨y y y.
1 , ⭈⭈⭈ ,¨ j , ⭈⭈⭈ ,¨ n
js1
n
Sq
i s ( Ý Ž ¨ i j y ¨qj .
js1
2
Ž7.
n
Sy
i s ( Ý
js1
Ž ¨i j y ¨y
j
.2 ,
Cq y Ž q y.
i s Si r Si q Si . Ž8.
Table 2
Candidate materials for a flywheel and their properties a
ture stress scales with the fracture toughness of the fragmentability, while the fragmentability of the metals
material K IC . Thus, upon substituting K IC for S in Eq. is assumed poor. As mentioned above, we use the
Ž9., a second performance index K IC r is obtained. A bi-polar method w9x to quantitatively grade the frag-
third important requirement is to minimize the cost of mentability of the materials. Because of lack of recent
the material, manufacturing and fabrication. The latter information, most of the material price refers to 1980
should be taken into account in design but it is not w18x, and the rest are assumed in comparison with the
considered in this paper. Thus the third index is the recent price.
material price per weight. The fourth requirement is The values of the four material properties and per-
fragmentability. If a flywheel breaks into small pieces formance indexes chosen above are listed in Table 3.
at final failure, the hazard should be much reduced. Also the values of ri j normalized by Eq. Ž1. are listed.
The fragmentability is an important property for safety. The entropy Ej and the weight factor j are calcu-
The candidate materials are simply screened on the lated according to Eq. Ž2. and Eq. Ž3., respectively.
basis of working temperature Tw , that is, the constraint Four cases of the subjective weight s j are considered.
is Tw G 100⬚C. Most materials available in the database In case 1, limitr is most weighted Ž s1 s 0.4., and
of Kang et al. w11x survive except for thermoplastic K IC r , price per unit mass and the fragmentability
polymers. As the final candidates we chose 10 materials follows Ž s2 s 0.3, s3 s 0.2, s4 s 0.1.. In cases 2 and 3,
comprising four metals and six unidirectional fiber-re- K IC r and price per unit mass are most weighted,
inforced-epoxy composites. Hereafter the materials se- respectively. In case 4, limitr and the fragmentability
lection procedure is demonstrated only for the 10 ma- are most weighted equally Ž s1 s s4 s 0.3., while in cases
terials. The properties of the materials are listed in 1, 2 and 3 the fragmentability is least weighted Ž s4 s 0.1.
Table 2 where limit and K IC of composites are for because the possibility of catastrophic failure is as-
loading parallel to the fiber direction. The frag- sumed to be minimal. Ej , j , s j and the revised weight
mentability of the composites is determined on basis of factor jU are listed in Table 4. Because the third
the shear bond strength of fiber-to-matrix, that is, the performance index Žprice per unit mass. has the largest
higher bond strength is regarded to give the worse scatter, its objective weight factor 3 is much larger
Table 3
Material properties and performance indexesrthe values normalized by Eq. Ž1. for candidate materials for a flywheel
Table 4
Entropy and weight factors about each material properties or performance index
U
j Case 1 0.296369 0.059602 0.622679 0.021350
Žmodified Case 2 0.235021 0.084026 0.658379 0.022574
weight factor. Case 3 0.100499 0.040422 0.844600 0.014480
Case 4 0.234286 0.041882 0.656321 0.067511
than others. As the result, even in the cases where the multiple thin ring flywheel of Garrett Air Research
subjective weight s3 is small, the revised weight factor which is described as ‘ . . . the most successful design’ by
U3 is still the largest. The normalized values ri j are Lewis w17x. The flywheel consists of a Kevlar 49-epoxy
weighted by jU according to Eq. Ž5. to give ¨ i j . Then FRP outside ring, a Kevlar 29-epoxy FRP interior ring
the ideal ¨ q j and the non-ideal ¨ yj for each material and an S glass᎐epoxy FRP inside ring. It is somewhat
property or performance index is obtained by Eq. Ž6.. fortuitous that the sequence coincides with the ranking
Eq. Ž7. and Eq. Ž8. give the separation measures Sq i , of the materials selected in cases 1 and 2. Probably the
Syi and the relative closeness Cq i for each material, design is a result of long research and experience. Here
the values of which are listed in Table 5. we can understand that the materials of the design
We rank the candidate materials on the basis of Cq i . were chosen with the strength being most weighted and
In both cases 1 and 2 for which the subjective weight s j then the best material is utilized for the most heavily
is greatest for limitr and K ICr , respectively, Kevlar loaded outside ring.
49-epoxy FRP is the best, and then Kevlar 29-epoxy In both cases 3 and 4 for which the subjective weight
FRP and S glass᎐epoxy FRP follow. Fig. 2 shows a s j of price per unit mass and fragmentability is given
Table 5
Separation measures and relative closeness of each candidate material for four cases a
Acknowledgements
the most, respectively, S glass᎐epoxy FRP is evaluated
as the best. Also, 300M and 7050-T73651 are ranked
high. 300M is a highly refined Ni᎐Cr᎐Mo alloy steel The authors are grateful to Professor N.A. Fleck of
Žvacuum-arc melted, silicon-modified 4340 steel. whose Cambridge University for valuable discussion and cor-
fatigue strength limit and fracture toughness K IC are rections of the English.
superior but its density is high. 7050-T73651 is an
aircraft alloy whose fatigue strength limit is especially
References
high among aluminum alloys and also the price per
unit mass is the least of the 10 candidate materials.
Both the metals can be recommended for cases where w1x Sandstrom R. An approach to systematic materials selection.
cost or machinability is important. Mater Des 1985;6:328᎐337.
w2x Ashby MF. Overview No.80: on the engineering properties of
materials. Acta Metall Mater 1989;37Ž5.:1273᎐1293.
w3x Bamkin RJ, Piearcey BJ. Knowledge-based material selection
5. Concluding discussion in design. Mater Des 1990;11Ž1.:25᎐29.
w4x Garton DA, Kang KJ, Fleck NA, Ashby MF. Materials selec-
tion for minimum weight fatigue design. Theoretical Concept
In the procedure of material selection, knowledge of and Numerical Analysis of Fatigue 1992:359᎐376.
multi-disciplines is needed. For a complex system which w5x Takuma M, Shibasaka T, Teshima T, Iwai Y, Honda T. Study
are composed of many parts Žfor example, a passenger on support system for materials selection in the design process.
car is composed of approx. 15 000 parts. no one can Trans JSME ŽC. 1994;60Ž574.:294᎐300 Žin Japanese..
w6x Waterman NA, Ashby MF. Elsevier material selector. Elsevier
provide all of the information enough to select the Applied Science, 1992.
optimum material for each part. As the time given to w7x Fleck NA, Kang KJ, Ashby MF. The cyclic properties of
design a new product is getting shorter, a CAE ŽCom- engineering materials. Acta Metall Mater 1994;42Ž2.:365᎐381.
puter Aided Engineering. tool to help a design engi- w8x Yoon KS. Systems selection by multiple attributes decision
neer with material selection is needed. The procedure making. Kansas Univ, 1980.
w9x Saati TL. The analytic hierarchy process. McGraw-Hill, 1980.
should be objective so as to minimize personal bias and w10x Shannon CE, Weaver W. The mathematical theory of commu-
it should be able to be coded into the software of nication. The University of Illinois Press, 1947.
‘expert system’. In this work, we have utilized the w11x Kang KJ, Garton DA, Fleck NA. Compilation of fatigue design
theories of decision-making. One of them is the con- data. Internal Report, Cambridge University Engineering De-
cept of entropy; to evaluate the weight factor for each partment, 1992.
w12x Mallick PK. Fiber-reinforced composites. Marcel Dekker Inc,
material property or performance index and the other 1988.
is TOPIS; to rank the candidate materials, for which w13x Friendrich K. Application of fracture mechanics to composite
several requirements are considered simultaneously. materials. Elsevier, 1989.
206 D. Jee, K. Kang r Materials and Design 21 (2000) 199᎐206
w14x Vinson JR, Sierakovski RL. The behavior of structures com- w17x Lewis G. Selection of engineering materials. Prentice Hall,
posed of composite materials. Martinus Nijhoff Publishers, 1990:10᎐19.
1986. w18x Materials data. Cambridge University Engineering Depart-
w15x Bader MG. Molding processes ᎏ an overview. Delaware com- ment, 1981.
posite design encyclopedia, vol. 3. Technomic Pub. Inc, 1990. w19x Suh NP. The principles of design. The Oxford University Press,
w16x Reifsnider KL. Fatigue of composite materials. Elsevier, 1991. 1990.