Академический Документы
Профессиональный Документы
Культура Документы
VG
of the output c of the function (c"y(x
)) is given by
VG
(c"y(xi))"sup"
(min
G
(
VG
(y(xi)))!R(y(xi)(R, 1)i)N, (4)
where sup is the supernum operator that gives the least upper bound. Eq. (4) can be interpreted as
follows. For a crisp value of the output c, there exist zero, one or more combinations of crisp values
96 U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111
of input variables x
G
such that y(x
G
)"c. The possibility of each combination is given by the smallest
possibility value for all fuzzy input variables. The possibility that y(x
G
))"c is given by the
maximum possibility value for all these combinations. In probability theory, the probability of an
outcome is equal to the product of the probabilities of the constituent events, whereas in fuzzy set
theory, the possibility of an outcome is equal to the minimum possibility of the constituent events
(note that the events in both cases are independent). When the outcome can be reached in many
ways, then in probability theory, the probability of an outcome is given by the sum of the
probabilities of all the ways, while in fuzzy theory, the possibility of the outcome is given by the
maximum possibility of all the possibilities [5].
In "nite element based analysis the function, y, is the "nite element transformation function. It
transforms the input "nite element data to a desired output response c. The "nite element model
must be formulated using basic principles of mechanics and fuzzy transformation operations. The
direct solution of Eq. (4) is intractable except for very simple fuzzy "nite element problems
involving at most one or two fuzzy variables. For a general function y, the number of operations
needed to solve equation (1) exactly is of the order R, [5].
3.2.2. Fuzzy arithmetic
Finite element modelling involves arithmetic operations (addition, subtraction, multiplication
and divisions) and when the input data is fuzzy, then fuzzy arithmetic operations have to be
performed. Fuzzy arithmetic operations represent a class of simple functions for which the
extension principle reduces to a simple tractable solution. This solution can be cast in a very
compact form using constrained interval arithmetic and -level representation [8].
Research in fuzzy modelling has been based on the use of interval arithmetic operations and
-level representation of the fuzzy variables [12}14]. However, in general, the fuzzy output of
a function that involves a combination of fuzzy arithmetic operations (for example, "nite element
model) cannot be evaluated using the result of individual fuzzy arithmetic operations. Dependence
relations between the fuzzy input and intermediate variables must be taken into consideration.
Therefore, constrained fuzzy arithmetic and not ordinary fuzzy arithmetic operations must be used
[7,15]. If ordinary fuzzy arithmetic operations are used instead of constrained fuzzy arithmetic
operations, then the fuzzy output has a width that increases with the number of fuzzy operations
and this can lead to unrealistic results [3,13,14,16,18,20}23].
3.2.3. Approximate numerical techniques for implementing extension principle
The descretization method [11] and the vertex method [16,19] are two approximate numerical
techniques for computing with fuzzy parameters through conversion from fuzzy to crisp numbers.
In the discretization method, each input variable is discretized into a certain domain D
G
of discrete
values. The number of evaluations of the function y needed in this case is ,
G
D
G
. For example for
three fuzzy variables and 10 discrete values for each variable (D
G
"10), the number of evaluations
of y is 1000. The method is too expensive for most practical "nite element numerical modelling
applications.
In the present study, the vertex method is used to numerically implement the extension principle
because it is computational cheaper than the discretization method. The steps involved in the
implementation of the vertex method are:
(1) Representation of fuzzy input variables using the -level method;
U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111 97
(ii) Formation of all binary combinations of extreme left and right value of all the fuzzy variables at
-level;
(iii) Computation of all values of the response for all the binary combinations; and
(iv) Selection of the maximum and minimum values of the response as the extreme values of the
fuzzy response at the -level.
The following discussion is a brief summary of the steps. As noted above, the "rst step in the
implementation of the vertex method is the -level representation of the fuzzy input variables. An
-level representation of a fuzzy variable x
G
denoted by [x?
G*
, x?
G0
], is the interval in which the
possibility of x
G
is at least equal to . This is illustrated in Fig. 1, where x?
G*
and x?
G0
are the extreme
left and right values, respectively, of the fuzzy variable, x
G
, at level . Thus, at an -level, each fuzzy
variable is discretized into two crisp values, x?
G*
and x?
G0
Once the fuzzy variables are discretized then all the binary combinations of the extreme left and
right values of all the fuzzy variables are formed. The binary combinations are then fed sequentially
into a numerical model ("nite element model) and the required response quantity y for each
combination is obtained. Assuming there are N-fuzzy variables and denoting the binary combina-
tions at an -level by C
?H
( j"1,2,
2
,)N
A?
), where N
A?
is the total number of binary combination,
then the fuzzy response corresponding to C
?H
can be written in the abbreviated form y(C
?H
). For
example for N fuzzy variables, the total number of binary combinations at an -level, N
A?
, is 2,.
The extreme values of the fuzzy response at an -level are then given by
[y?
*
, y?
0
]"[min"
HH
(y(C
HH
)), max"
HH
(y(C
HH
))], *; j"1,2,
2
, N
A?
, (5)
where y?
*
, and y?
0
, are the minimum (left) and maximum (right) values of the response quantity,
respectively. Eq. (5) represents an approximate implementation of the extension principle through
vertex method and must be repeated for all the -levels of interest ranging from 0 to 1. The results
obtained from Eq. (5), for a sweep of -levels, are used to construct the possibility distribution (y)
of the response. The accuracy of the possibility distribution of the response quantity depends on the
number of -levels that is used. The higher the number of -levels the greater the accuracy of the
possibility distribution of the response. The computational cost of constructing the possibility
distribution of the response quantity using Eq. (5) depends on the number of fuzzy variables and
-levels and this could be very expensive. The total number of "nite element runs that is involved is
A;2,, where A is the number of -levels and N is the number of fuzzy input variables. For example,
a fuzzy "nite element problem with 5 -levels and 3 fuzzy variables requires 40 "nite element runs
and a fuzzy "nite element problem with 5 -levels and 5 fuzzy variables requires 160 runs. Thus, it
can be seen that the number of "nite element runs required grows exponentially with the number of
fuzzy variables. For very large practical "nite element models with many fuzzy parameters, a large
number of runs might not be realistic. This fact has been recognized by most of the reported studies.
To cut down on the high cost of computing, most of the studies [3,11] restricted the total number of
fuzzy parameter used in analysis. The practical implication of placing restriction on the number of
fuzzy variables is that the approaches that have been reported in the literature are applicable only
to a small number of fuzzy variables. In this study, an e$cient response surface-based optimization
method that minimizes the computational cost and is valid for any number of fuzzy variables is
advanced.
98 U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111
4. Response surface solution strategy
The response surface method is a classical statistical technique in which a complex (computer)
model is approximated by a simple functional relationship between the output quantity and the
input quantities. In the context of fuzzy analysis, the input quantities are the values of the fuzzy
variables while the output quantity is the response function. Since several "nite elements runs are
required to implement the extension principle (Eq. (5)), adopting a simple response function
approximation will lead to a substantial reduction in the number of "nite element runs and increase
computational e$ciency. Within the context of fuzzy "nite element analysis and prior to the
implementation of the extension principle, numerical experiments are carried out about the normal
points of all the fuzzy variables. Output values for various sets of input values are then obtained
and used to develop explicit functional approximations between the input variables and the
response quantity. For example, for the quadratic approximation, the explicit functional relation-
ship takes the form
y (x)"a#
,
G
b
G
x
G
#
,
GH
c
GH
x
G
x
H
, (6)
where N is the number of the fuzzy variables x, y (x) is the approximation of the response function,
and the coe$cients a, b, and c are determined by application of regression analysis on results from
numerical experiments. Eq. (6) is the model used to approximate the fuzzy response function, y(x)
Once an approximate fuzzy function y (x) is developed then a combinatorial optimization is carried
out at an -level. The purpose of the combinatorial optimization is to determine the binary
combinations of the fuzzy variables that will result in extreme response quantities. In particular,
two combinatorial optimisations are required at an -level and can be stated as
Given an -level
Find binary combination of x?
G*
and x?
G0
, C
?
(i"1, 2,
2
, N) that minimizes y (x) (7)
and
"nd the binary combination of x?
G*
and x?
G0
, C
?
'
(i"1, 2,
2
, N) that will maximizes y (x)
(8)
The true values of the maximum and minimum responses at an -level are determined by feeding
C
?
and C
?
, respectively, into the "nite element model. The process is repeated for all the
-levels of interest. This strategy requires 2 "nite element runs at an -level as opposed to 2, "nite
element runs with vertex method. This is extremely e$cient especially for large number of fuzzy
input parameters and large "nite element models. The #ow chart for the solution strategy that has
been developed in this paper is shown in Fig. 2.
5. Applications
The procedure discussed in the preceding section has been implemented with the deterministic
"nite element programVAST [17] as the "nite element solver. Three example problems are used to
illustrate the methodology. These are described below.
U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111 99
Fig. 2. Response surface based solution strategy for fuzzy analysis.
5.1. Three stepped bar
The "rst example is a 3 stepped bar, shown in Fig. 3, that was previously considered by Rao and
Sawyer, [2], and Chen and Rao, [4]. The dimensions and material properties of the bar are
presented in Table 1. In the table, the middle values represent the nominal values of the fuzzy
variables. The static response (due to the static load P
`
) and the natural frequencies of the structure
were computed.
5.2. Static analysis
For the static analysis, three cases of fuzzy variables were considered. In Case A only the load
P
`
was fuzzi"ed. In Case B the load as well as the elastic modulus were fuzzi"ed and in Case C the
load, elastic modulus, cross sectional areas and the lengths of the bar were taken as fuzzy variables.
All the fuzzy variables were assumed to have triangular membership functions and the triplet
number for each fuzzy variable [x
'
, x
`
, x
'
] de"nes a triangular membership function, where x
'
and
x
'
are the lower and upper bounds at "0 and x
`
is the nominal value at "1. The fuzzy input
variables for the three cases are shown in Table 1.
The possibility distributions for the fuzzy axial displacement of the free end and the strain energy
of the system were computed using both the combinatorial and response surface approaches.
In this context combinatorial method refers to the crude method in which all possible combina-
tions of the fuzzy variables are considered, whereas the response surface approach refers to the
100 U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111
Fig. 3. Con"guration of a 3-stepped bar.
Table 1
Data for 3-stepped bar example as triangular fuzzy numbers
Parameters Case A Case B Case C
Data for fuzzy static analysis of 3-stepped bar
A
(in.`) [3.00, 3.00, 3.00] [3.00, 3.00, 3.00] [2.99, 3.00, 3.01]
A
`
(in.`) [2.00, 2.00, 2.00] [2.00, 2.00, 2.00] [1.99, 2.00, 2.01]
A
`
(in.`) [1.00, 1.00, 1.00] [1.00, 1.00, 1.00] [0.99, 1.00, 1.01]
L
(in.) [12.00, 12.00, 12.00] [12.00, 12.00, 12.00] [11.95, 12.00, 12.05]
L
`
(in.) [10.00, 10.00, 10.00] [10.00, 10.00, 10.00] [9.95, 10.00, 10.05]
L
`
(in.) [6.00, 6.00, 6.00] [6.00, 6.00, 6.00] [5.95, 6.00, 6.05]
E
, E
`
, E
`
(psi) [3.0e7, 3.0e7, 3.0e7] [3.0e7, 3.0e7, 3.0e7] [2.8e7, 3.0e7, 3.1e7]
P
, P
`
(lb) [0.00, 0.00, 0.00] [0.00, 0.00, 0.00] [0.00, 0.00, 0.00]
P
`
(lb) [2000, 10000, 18000] [2000, 10000, 18000] [2000, 10000, 18000]
Data for fuzzy eigenvalue analysis of 3-stepped bar
A
(in.`) [3.00, 3.00, 3.00) [3.00, 3.00, 3.00] [2.99, 3.00, 3.01]
A
`
(in.`) [2.00, 2.00, 2.00] [2.00, 2.00, 2.00] [1.99, 2.00, 2.01]
A
`
(in.`) [1.00, 1.00, 1.00] [1.00, 1.00, 1.00] [0.99, 1.00, 1.01]
L
(in.) [12.00, 12.00, 12.00] [12.00, 12.00, 12.00] [11.95, 12.00, 12.05]
L
`
(in.) [10.00, 10.00, 10.00] [10.00, 10.00, 10.00] [9.95, 10.00, 10.05]
L
`
(in.) [6.00, 6.00, 6.00] [6.00, 6.00, 6.00] [5.95, 6.00, 6.05]
E
, E
`
, E
`
(psi) [3.0e7, 3.0e7, 3.0e7] [3.0e7, 3.0e7, 3.0e7] [2.8e7, 3.0e7, 3.1e7]
1, 2, 2 (lb/in.`) [0.27, 0.28, 0.29] [0.27, 0.28, 0.29] [0.27, 0.28, 0.29]
methodology developed in this study, whereby the response surface approximation is used to
reduce the number of computations. The results are presented in Figs. 4 and 5. It is seen that as the
number of fuzzy variables increases the bounds of the possibility distributions for both the strain
energy and the displacements are also increased. Both plots are skewed to the right for Case B and
Case C because the displacement and the strain energy are inversely proportional to the modulus
of elasticity, which is skewed to the left. For the cases considered the response surface strategy and
the combinatorial approach yield exactly the same results for all the possibility distributions of the
responses.
U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111 101
Fig. 4. Displacement at the end of 3-stepped bar under fuzzy static load.
Fig. 5. strain energy of 3-stepped bar system under static load.
5.3. Natural frequency analysis
The same 3-stepped bar was analyzed to determine the e!ect of input parameter fuzzi"cation on
the natural frequencies. Three fuzzy response cases were also considered. Case A considered only
the density of the bars as fuzzy. In case B the density and the modulus of elasticity were taken as
fuzzy variables and in Case C the density, modulus of elasticity and cross-sectional area, and
length, of each section were considered as fuzzy. The triangular fuzzy numbers for each parameter
are shown in Table 1.
The fuzzy frequencies for modes 1}3 are shown in Fig. 6. Again, for all the cases considered the
response surface strategy and the combinatorial approach produce exactly the same results for the
possibility distribution of the responses. Trends for the possibility distributions of the fuzzy natural
frequencies were similar to the possibility distributions of the static responses. As the number of
input fuzzy variables increases from Case A to Case C the spread of fuzzy frequencies also increases.
The fuzzy frequencies are skewed to the left as is the fuzzy number for the modulus of elasticity,
since the frequency solution is directly related to the modulus of elasticity.
102 U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111
Fig. 6. First three fuzzy frequencies of 3-stepped bar.
U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111 103
Fig. 8. Displacement at the mid-span of the "xed end beam under fuzzy static load.
Fig. 7. Con"guration of beam "xed at both ends.
5.4. Fixed ended beam
The second example problem is a beam 400 in.;18 in.;6.17 in., "xed at both ends as shown in
Fig. 7. This problem was "rst considered by Rao and Sawyer [2]. A uniformly distributed load of
400 lb/in. is applied to the beam. The beam had a moment of inertia of 3000 in." and a modulus of
elasticity of 30;10" psi.
A static fuzzy response of the structure was computed using both the combinatorial approach
and the response surface methodology. Three combinations of triangular fuzzy parameters were
considered for this beam. In Case A only the load was considered fuzzy with the load represented
by the triplet [0.90,1.00,1.10] normalized with respect to the nominal values. In Case B the modulus
of elasticity represented by the normalized triplet [0.98, 1.00, 1.02] was considered as the only fuzzy
variable. In Case C both the load and the modulus of elasticity with the same fuzzy numbers as in
Case a and B were considered. The possibility distributions of the maximum displacement at the
mid-span of the beam and the strain energy of the system are shown in Figs. 8 and 9 respectively. It
is evident in both "gures that an increase in the number of fuzzy parameters results in increase in
the boundaries of the possibility distributions. It is also evident from the plots that fuzzi"cation of
the modulus of elasticity has a lesser e!ect on the displacement and strain energy solution than
does fuzzifying the load. For all the cases considered, the response surface strategy and the
combinatorial approach both yielded exactly the same result for the possibility distributions of the
response quantities.
104 U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111
Fig. 9. Strain energy of "xed end beam under fuzzy static load.
5.5. Four story}three bay rigid frame
The last example that is considered is a four story, three bay rigid frame. This is the central frame
of a typical steel-framed structure shown in Fig. 10. All connections in the frame were considered to
be "xed and capable of transferring moments. The dimensions and geometry of the frame
are shown in Fig. 10 and the member sizing, material properties and dimensions are presented in
Table 2.
5.5.1. Static analysis
Typical structural design loads were calculated assuming the building was located in Halifax,
Nova Scotia using the National Building Code of Canada as a guide [24] Fig. 11 shows the
distributed loads considered and Table 3 presents the magnitudes of each of these loads. In Fig. 11
DL represents the dead load of the #oor or roof system and any mechanical additions; LL
represents the occupancy loads (this structure was assumed to be o$ce building); SL represents the
design snow load; and WL represents the design wind load. The following load combination was
used to determine the factored loads shown in Table 3.
P
'''^'
"0.85DL#0.7((1.5LL#1.5SL)#1.5WL). (9)
For this example the wind load, snow load and live load were each considered to be triangular
fuzzy numbers with the triplet of [0.75, 1.00, 1.25]P and the modulus of elasticity was also
considered to be fuzzy with the triplet of [0.90, 1.00, 1.10]E, where P is the nominal value of the
particular load component, and E the nominal value of the modulus of elasticity. The deformed
shape of the frame under static loading is shown in Fig. 12 and the possibility distributions of the
fuzzy horizontal displacement of the top left corner of the frame and the fuzzy strain energy of the
system are shown in Figs. 13 and 14, respectively. Results obtained with the response surface
methodology are again identical to the results from the combinatorial approach. However, by
considering WL, SL, LL and modulus of elasticity as independent fuzzy variables and using linear
response surface approximation and 4 levels for example, the response surface approach requires
U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111 105
Fig. 10. (a) Plan view of a typical steel-framed structure showing location of rigid frame used for analysis. (b) Elevation
view of rigid frame used for analysis.
only 13 "nite element runs compared to 65 runs that are required by the direct combinatorial
approach. This clearly demonstrates the computational e$ciency of the response surface approach.
The e$ciency and robustness of the response surface approach is expected to increase with increase
in the complexity of the structure, number of fuzzy variables and number of -cuts.
5.5.2. Natural frequency analysis
Fuzzy natural frequency analysis of the frame was also performed to determine the possibility
distributions of the "rst three natural frequencies. Three fuzzy cases were considered. Only the
density of the steel was fuzzi"ed in Case A. In Case B only the modulus of elasticity of the steel was
fuzzi"ed and in Case C both the density and the modulus of elasticity were fuzzi"ed. A triangular
fuzzy distribution with the triplet [0.90, 1.00, 1.10] normalized with respect to the nominal values
was used for both parameters.
Fig. 15 shows the fuzzy frequency and mode shapes for the "rst three eigenvalues. Again it is
evident that the higher the number of fuzzy input variables the wider the bounds of the possibility
106 U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111
Fig. 11. Typical structural design loads used for static analysis.
Table 3
Summary of unfactored and factored loads used in the static analysis of rigid frame
Load case Unfactored (kN/m) Factored (kN/m)
Dead, DL (#oor/roof#mechanical) 4.00 3.40
Snow, SL (halifax) 7.44 7.81
Live, LL (o$ce occupancy) 19.20 20.16
Wind, WL (windward lower) 1.40 1.47
Wind, WL (windward upper) 1.60 1.68
Wind, WL (leeward lower) 1.00 1.05
Wind, WL (leeward upper) 1.20 1.26
Table 2
Summary of section properties and material properties used for analysis of rigid frame
Property Member
Floor/roof beam Lower beam column Upper beam column
Section W250;33 W250;45 W200;27
Cross-sectional area (mm`) 4170 5720 3390
Moment of inertia, I
X
(1;10" mm") 48.90 71.10 25.80
Moment of inertia, I
W
(1;10" mm") 4.73 7.03 3.30
Torsional constant, J (1;10" mm") 0.0985 0.2610 0.0713
Depth, d (mm) 129.0 133.0 104.0
Material Structural steel
Yield stress (MPa) 350
Modulus of elasticity, E (GPa) 200
Density, (kg/m`) 7850
Poisson's ratio, 0.300
U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111 107
Fig. 12. Deformed shape of rigid frame under static loading conditions.
Fig. 13. Fuzzy horizontal displacement of top windward corner of frame under static loading conditions.
Fig. 14. Fuzzy strain energy of system under static loading conditions.
108 U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111
Fig. 15. Fuzzy frequencies and mode shapes of rigid frame when fuzzifying: E [0.90, 1.00, 1.10] and [0.90, 1.00, 1.10].
distributions of the fuzzy responses. Also as in previous cases, the response surface approach has
identical results as the combinatorial approach.
6. Summary and conclusions
A robust, practical and versatile framework for fuzzy "nite element analysis of structures has
been developed, implemented and demonstrated in this study. The approach involves integrated
U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111 109
"nite element modelling, response surface methodology and vertex method fuzzy analysis meth-
odology. The response surface methodology has been used to approximate the fuzzy "nite element
response quantity. Combinatorial optimization was then performed on the approximated quantity
to determine the binary combinations of the fuzzy variables that result in extreme responses at an
-level. The determination of the true values of the fuzzy responses at an -cut was accomplished by
feeding the extreme value of the fuzzy input variables to the "nite element model. The merits of the
proposed methodology which include computational e$ciency without compromise on accuracy
and ability to use well-tested existing deterministic "nite element software (example VAST) have
been demonstrated through some example problems.
Acknowledgements
The authors are grateful to Martec Limited for providing support under its corporate research
program (Project C 99-505-04). Partial funding was also provided by NRC Canada, through its
IRAP program (Project C 99-705-02).
References
[1] S. Vallippan, T.D. Pham, Constructing the membership function of a fuzzy set with objective and subjective
information, Microcomput. Civil Eng. 8 (1993) 75}82.
[2] S.S. Rao, J.P. Sawyer, A fuzzy "nite element approach for the analysis of imprecisely-de"ned systems, AIAA J. 33
(1995) 2264}2370.
[3] R.-J. Chao, B. Ayyub, Finite element analysis with fuzzy variables, Building an International Community of
Structural Engineers, in: S.K Ghosh, R. Mohammadi (Eds.), Proceedings of Structures Congress XIV, 1996,
pp. 643}650.
[4] L. Chen, S.S. Rao, Fuzzy "nite element approach for the vibration analysis of imprecisely-de"ned systems, Finite
Element Anal. Des. 27 (1997) 69}83.
[5] T.M. Wasfy, A.K. Noor, Application of fuzzy sets to transient analysis of space structures, AIAA 98-1833 (1998)
1172}1182.
[6] R.L. Mullen, R.L. Muhanna, Bounds of structural response for all possible loading combinations, ASCE J. Struct.
Eng. 125 (1) (1999) 98}106.
[7] B. Ayyub, Personal Communication, 1999.
[8] T.J. Ross, Fuzzy Logic with Engineering Applications, McGraw-Hill, New York, 1995.
[9] D. Cook, D.S. Malkus, M.E. Plesha, Concepts and Applications of Finite Element Analysis, Wiley, 3rd Edition,
1989.
[10] L.A. Zadeh, Fuzzy sets, Inform. Control 8 (3) (1965) 338}352.
[11] T.D. Pham, Fuzzy "nite element analysis of engineering problems, Ph.D. Dissertation, University of New South
Wales, Australia, 1994.
[12] R. Zhao, R. Govind, Solution of algebraic equations involving generalized fuzzy numbers, Inform. Sci. 56 (1}3)
(1991) 199}243.
[13] Q. Shen, R. Leitch, Fuzzy qualitative simulation, IEEE Trans. Systems, Man Cybernet. 24 (4) (1993) 1038}1061.
[14] A. Kaufman, M.M. Gupta, Introduction to Fuzzy Arithmetic: Theory and Application, Van Nostrand Reinhold
Company, Inc New York, 1985.
[15] G. Klir, T. Folger, Fuzzy Sets, Uncertainty and Information, Prentice-Hall, Englewood, Cli!s, NJ, 1988.
[16] R.-J. Chao, Structural and reliability analysis of ships with fuzzy and random variables, Ph.D. Dissertation,
University of Maryland, USA, 1995.
110 U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111
[17] Vibration and Strength Analysis Program (VAST): Users Manual, version 7.3, Martec Limited, Halifax, NS, 1997.
[18] M. Mares, Computation Over Fuzzy quantities, CRC Press, Boca Raton, 1994.
[19] W. Dong, H. Shah, Vertex method for computing function of fuzzy variables, Fuzzy Sets and Systems 2 (1987)
201}208.
[20] J.J. Buckley, Y. Qu, Solving systems of linear fuzzy equations, Fuzzy Sets and Systems 39 (1991) 33}43.
[21] H.U. Koyluoglu, A.S. Cakmak, S.R.K. Nielsen, Interval algebra to deal with pattern loading and structural
uncertainties, J. Eng. Mech. 121 (11) (1995) 1149}1157.
[22] S.S. Rao, L. Berke, Analysis of uncertain structural systems using interval analysis, AIAA J. 35 (4) (1997) 727}736.
[23] M. Kwiesielewicz On explicit, implicit solutions of fuzzy linear equations, In. J. System Sci. 26 (12) (1995)
2277}2286.
[24] National Building Code of Canada, 1995.
U.O. Akpan et al. / Finite Elements in Analysis and Design 38 (2001) 93}111 111