Professional Documents
Culture Documents
Finite-Element Methods
Rafi L. Muhanna, M.ASCE1; Hao Zhang2; and Robert L. Mullen, F.ASCE3
Abstract: This paper presents a new approach for the treatment of parameter uncertainty for linear static structural mechanics problems.
Parameter uncertainties are introduced as interval numbers. Interval arithmetic is applied to finite-element method to analyze the structural
responses due to uncertain loading, axial and bending stiffness. However, a naïve use of interval arithmetic in the formulation of
finite-element method, i.e., replacing the deterministic parameters with corresponding interval ones, will result in meaningless wide results
due to the so-called dependency problem. In the present approach, an element-by-element technique is used to reduce the overestimation
and compatibility conditions are ensured by a penalty method. With the newly developed overestimation control, most sources of
overestimation are eliminated and a very sharp enclosure for the system response is obtained. A number of numerical examples are
introduced.
DOI: 10.1061/共ASCE兲0733-9445共2007兲133:12共1700兲
CE Database subject headings: Bending; Stiffness; Finite element method; Parameters.
Introduction ries. For example, a fuzzy number 共Zadeh 1978兲 can be viewed as
a set of valued intervals with different confidence of given level
Real life engineering practice often deals with variables and pa- of presumptions 共␣ cuts兲. Thus fuzzy arithmetic can be performed
rameters of uncertain values. For a proper performance assess- as interval arithmetic on ␣ cuts. A Dempster–Shafer structure
ment, these uncertainties must be accounted for appropriately. 共Dempster 1967; Shafer 1976兲 with interval focal elements can be
Probability theory is the traditional approach to handle uncer- viewed as a set of intervals with probability mass assignments. In
tainty. This approach is computationally expensive, and requires this sense, the developed approach in the present work is also
sufficient statistical data to justify the assumed statistical distribu- applicable to the uncertain systems handled by other set-based
tions. An alternative interval approach, especially when data are uncertainty theories.
incomplete, has been employed 共Moore 1966兲. The uncertainties Recently, interval arithmetic has been used with the finite-
in parameters will be introduced as interval numbers, i.e., the element method to analyze structures with bounded parameters,
uncertain variables are known to lie between two prescribed such a method is called the interval finite-element method
bounds 共endpoints兲, but no assumption about the statistical distri- 共IFEM兲. A number of developments in IFEM can be reviewed.
bution inside the interval is made. For example, if the yield Rao and Berke 共1997兲 used a combinatorial approach 共based on
strength is known to range from 240 to 250 MPa, it can be rep- an exhaustive combination of the extreme values of the interval
resented as 关240,250兴 MPa. Interval arithmetic is very convenient parameters兲 to compute interval bounds on system response. For
to represent uncertain variables with known bounds when statis- linear structural analysis, this approach gives an exact solution.
tical information is not available. Nevertheless, interval arithmetic However, it is computationally tedious and expensive, and is lim-
can also be combined with any available statistical information to ited to the solutions of small-scale problems only. A search-based
represent different levels of uncertainty 共Ferson and Ginzburg algorithm was developed in the work of Rao and Chen 共1998兲.
1995; Joslyn 1996; Joslyn and Kreinovich 2002兲. Interval arith- The algorithm is inefficient especially in large size problems, and
metic is closely connected with other set-based uncertainty theo- the accuracy of the results is limited to narrow intervals only.
Koyluoglu and Elishakoff 共1998兲 have introduced a comparison
1
Associate Professor, School of Civil and Environmental Engineering, of stochastic and interval finite-element method applied to shear
Georgia Institute of Technology, Atlanta, GA 30332. E-mail: frames exhibiting uncertain modulus of elasticity. The work was
rafi.muhanna@gtsav.gatech.edu restricted to narrow intervals and approximate results. Qiz and
2
Graduate Student, School of Civil and Environmental Engineering, Elishakoff 共1998兲 used the first-order perturbation method and
Georgia Institute of Technology, Atlanta, GA 30332. E-mail: hao.zhang@ interval arithmetic to determine the bounds of static displace-
ce.gatech.edu ments of structures under interval modulus of elasticity and inter-
3
Professor, Dept. of Civil Engineering, Case Western Reserve Univ., val loads. This method does not consider the dependence that
Cleveland, OH 44106. E-mail: rlm@case.edu exists between the interval coefficients of the stiffness matrix and
Note. Associate Editor: Elisa D. Sotelino. Discussion open until May load vector, therefore, the obtained result is overly conservative.
1, 2008. Separate discussions must be submitted for individual papers. To
Pantelides and Ganzerli 共2001兲 compared convex model and
extend the closing date by one month, a written request must be filed with
the ASCE Managing Editor. The manuscript for this paper was submitted fuzzy set finite-element method. Convex superposition approach
for review and possible publication on September 9, 2004; approved on was proposed to analyze load uncertainty, and exact solution was
March 19, 2007. This paper is part of the Journal of Structural Engi- obtained. However, the superposition is only applicable to load
neering, Vol. 133, No. 12, December 1, 2007. ©ASCE, ISSN 0733-9445/ uncertainty. Combinatorial approach was used in Ganzerli and
2007/12-1700–1709/$25.00. Pantelides 共1999兲 to treat interval modulus of elasticity, and this
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
approach is of exponential complexity. Chen et al. 共2002兲 have definition interval operations. It is easy to see that the set of all
used matrix perturbation to calculate bounds on displacements of possible results for x 苸 x and y 苸 y forms a closed interval 共for 0
structural static problems. Obtained results were approximate and not in a denominator interval兲, and the end points can be calcu-
not guaranteed to contain the exact bounds. McWilliam 共2000兲 lated by
proposed two methods for determining bounds on static displace-
ments of structures with interval parameters. The first method is a
modified version of perturbation analysis. The second method is x ⴰ y = 关min共xi ⴰ y i兲,max共xi ⴰ y i兲兴 for ⴰ 苸兵+ ,− , · ,/其 共2兲
based on the assumption that the displacement surface is mono-
tonic. However, for the general case, the validity of monotonicity
is difficult to verify. Möller et al. 共2000兲 have developed an opti- Dependency Problem in Interval Arithmetic
mization algorithm combining evolution strategy, gradient
method and Monte Carlo method. The algorithm was applied to The interval-system quality is measured by the width of the inter-
both static and dynamic nonlinear analysis. The accuracy of the val results, and a sharp enclosure of the exact solution is the goal.
approach is not clear as the exact solutions were not presented in However, the width of results may be unnecessarily wide in some
the paper. Dessombz et al. 共2001兲 have introduced an IFEM in occasions due to dependency effect. For example, if the interval
which the interval parameters were factorized during the assem- function f共x兲 = x − x is evaluated with x = 关a , b兴 = 关1 , 2兴, the inter-
blage process of the stiffness matrix, then Rump’s iterative algo- val subtraction rule 共Appendix兲 gives the result: f共x兲 = 关a − b , b
rithm 共Rump 1983兲 was employed for solving the linear interval − a兴 = 关−1 , 1兴, which contains the exact solution 关0 , 0兴, but is much
equation. In this work, the overestimation control becomes more wider. The interval arithmetic implicitly made the assumption that
difficult with the increase of the number of the interval param- all intervals are independent, namely it treats x − x as if evaluating
eters, which does not lead to useful results for practical problems. the intervals x − y, and x , y are two independent interval quantities
In the works of Muhanna and Mullen 共1995兲, Mullen and that happen to have the same bounds. This phenomenon is re-
Muhanna 共1996, 1999兲, an interval-based fuzzy finite element has ferred as overestimation due to “dependency” of the variables
been developed for treating uncertain loads in static structural 共Moore 1979; Neumaier 1990; Hansen 1992兲. Reducing the over-
problems. Load dependency was eliminated and the exact solu- estimation is a central issue to successful engineering application
tion was obtained. Also, Muhanna and Mullen 共2001兲 have devel- of interval analysis. In general, sharp results are obtained when
oped an IFEM based on element-by-element technique and
proper bracketing is used, dependency is reduced and the physical
Lagrange multiplier. Uncertain modulus of elasticity was consid-
nature of the problem is not violated. In the previous example, the
ered. Most sources of overestimation were eliminated, and a sharp
dependency problem could be avoided in evaluating x − x as
enclosure for the system response was obtained.
x · 共1 − 1兲 = 0.
In this paper we consider finite-element analysis of frame
structures with interval loading, axial stiffness and bending stiff-
ness. A brief review of interval arithmetic is presented, the for-
mulation is described, and numerical examples are given. Interval Vectors and Matrices
Basic Definition A linear interval equation with coefficient matrix A 苸 IRn⫻n and
right-hand side b 苸 IRn is defined as the family of linear equations
An interval number is a closed set in R that includes the possible
range of an unknown real number, where R denotes the set of real
numbers. Therefore, a real interval is a set of the form Ax = b 共A 苸 A, b 苸 b兲 共3兲
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
The solution set ⌺共A , b兲 usually is not an interval vector, and Theorem 1 共Rump 2001兲: Let A 苸 IRn⫻n , R 苸 Rn⫻n , b , x
does not need to be convex; in general, ⌺共A , b兲 has a very com- 苸 IRn be given, if
plicated structure. In order to guarantee that the solution set
⌺共A , b兲 is bounded, it is required that the matrix A be regular, i.e.
that every matrix A 苸 A is nonsingular. The hull of the solution Rb + 共I − RA兲x 債 int共x兲 共12兲
set ⌺共A , b兲 is an interval vector which has the narrowest possible
interval components, denoted as then R and every matrix A 苸 A is nonsingular, and
where
where int共x兲 denotes the interior of x. Expression 共13兲 provides a
guaranteed enclosure to the solution set of the linear interval
AHb = 〫 兵A−1b兩A 苸 A,b 苸 b其 for b 苸 IRn 共6兲 equation Ax= b. The residual form of Eq. 共12兲 can be given in the
form 共Neumaier 1990兲
In fact, the exact hull of the solution set for the general case is not
computationally achievable. The solution of interest is seeking an
enclosure, i.e., an interval vector x containing AHb, while narrow Rb − RAx0 + 共I − RA兲x* 債 int共x*兲 共14兲
enough to be useful
Condition 共9兲 is a range determination problem, and can be re- Interval Finite-Element Analysis
duced to the following interval expression:
Overestimation in IFEM
Rb + 共I − RA兲x 債 x 共11兲
A naïve use of interval arithmetic in FEM 共naïve IFEM兲, i.e.,
If an interval vector x satisfying Eq. 共11兲 can be found, then x replacing deterministic numbers in conventional FEM with inter-
contains the solution of Ax = b. The result can be extended to find val numbers and solving the system as a nonparametric interval
the enclosure of the solution set of linear interval equation equation will result in meaningless wide results 共Muhanna and
Ax= b 共Neumaier 1990; Rump 2001兲. The following theorem can Mullen 2001; Dessombz et al. 2001兲. Let us consider the simple
be presented. frame shown in Fig. 1. The frame is subjected to a unit moment
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Fig. 2. Two-step bar
冉 关2.97,3.03兴
关0.99,1.01兴
关0.99,1.01兴
关1.98,2.02兴
冊冉 冊 冉 冊
1
2
=
1
0
共18兲
a significant improvement in the solution convergence will be
illustrated.
The element-by-element technique can be illustrated by the
two-step bar problem shown in Fig. 2. The elements are disjointed
Solving Eq. 共18兲 using Theorem 1, the values of 1 and 2 are as shown in Fig. 3, thus the system stiffness matrix K takes a
obtained as block-diagonal structure with dimension of n ⫻ n, and n⫽degrees
of freedom per element ⫻ number of elements in the structure.
1 = 关0.3926,0.4074兴 Such a structure of the matrix avoids the element coupling. K is a
singular matrix, conditions of compatibility and equilibrium will
be ensured using the penalty method.
2 = 关− 0.2078,− 0.1922兴 共19兲
1 1
1 = = = 关0.3960,0.4040兴
k1 + 0.75k2 关2.475,2.525兴
− 0.5
2 = = 关− 0.2020,− 0.1980兴 共20兲 Fig. 3. EBE two-step bar model
k1 + 0.75k2
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Fig. 4. EBE two-step bar model with penalty method
1
⌸ = 2 uTKu − uT p 共21兲 Fig. 5. Two-bay two-story frame
1 1
⌸* = 2 uTKu − uT p + 2 tT␣t 共23兲 The coefficient matrix in Eq. 共26兲 represents the sum of two
matrices: The interval element-by-element stiffness matrix K and
Invoking the stationarity of ⌸*, that is ␦⌸* = 0, we obtain deterministic penalty matrix Q.
For a frame element with interval parameters of cross-section
共K + CT␣C兲u = p 共24兲 area A, moment of inertia I and modulus of elasticity E, the
required coordinate transformation can result in the coupling of
or EA and EI, thus makes it more difficult to handle the dependency
effect. In the present formulation, the coordinate transformation
共K + Q兲u = p 共25兲 that usually occurs in the conventional FEM is delayed. The
matrix Q can be called the penalty matrix. As ␣ grows, u changes interval parameters 共A, I, and E兲 are factorized out from the
in such a way that the constraint equations are more nearly satis- element stiffness matrix in a local element coordinate system.
fied 共Cook et al. 1989兲. The physical interpretation of ␣ is that it Consider the ith finite element in the structure, the element stiff-
acts as large spring stiffness to force the adjacent nodes to have ness matrix Ki⬘ in local element coordinate can be expressed in
identical displacements, as shown in Fig. 4. Penalty numbers the form Ki⬘ = DiSi = SiDi
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
冢 冣
1 1
0 0 − 0 0
L L
12 6 12 6
冢 冣
0 0 −
L3 L2 L3 L2 EA 0 0 0 0 0
6 4 6 2 0 EI 0 0 0 0
0 0 − 2
L2 L L L 0 0 EI 0 0 0
K = S iD i = · 共27兲
1 1 0 0 0 EA 0 0
− 0 0 0 0
L L 0 0 0 0 EI 0
12 6 12 6 0 0 0 0 0 EI
0 − 3 − 2 0 − 2
L L L3 L
6 2 6 4
0 0 − 2
L2 L L L
where Si⫽deterministic matrix and Di⫽interval diagonal matrix. and the residual form 共14兲 is
Applying the coordinate transformation, the element stiffness ma-
trix Ki in global coordinate can be expressed as
Rp⬘ − RAu⬘0 + 共I − RA兲u⬘* 債 int共u⬘*兲 共33兲
Ki = TTi Ki⬘Ti = TTi 共SiDi兲Ti 共28兲 where A = 共SD + Q⬘兲; R⫽inverse of midpoint of A; u⬘ = u⬘0 + u⬘*;
and u⬘0 = R mid共p⬘兲. Assigning z = Rp⬘ − RAu⬘0 and C = 共I − RA兲, Eq.
共33兲 can be rewritten as
in which Ti⫽element coordinate transformation matrix. Similarly,
the system stiffness matrix K constructed by an EBE model can
be expressed as z + Cu⬘* 債 int共u⬘*兲 共34兲
the first term of Eq. 共34兲, z, is a constant interval vector, and it can
K = TTK⬘T = TT共SD兲T 共29兲 be introduced as
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
冢 冣冢 冣 冢 冣
EA 0 0 0 0 0 u⬘01 u⬘01 0
0 EI 0 0 0 0 u⬘02 0 u⬘02
0
0
0
0
EI
0
0
EA 0
0 0
0
·
u⬘03
u⬘04
=
0
u⬘04
u⬘03
0
· 冉冊
A
I
· 共E兲 共38兲
0 0 0 0 EI 0 ⬘
u05 0 ⬘
u05
0 0 0 0 0 EI u⬘06 0 u⬘06
Table 2. Bounds of Selected Nodal Displacement for the Frame in Fig. 5 w3 = 关49.255,52.905兴 kN/m, w4 = 关49.255,52.905兴 kN/m
Monte Carlo sampling Present IFEA 共43兲
Displ. LB UB LB UB The geometric and material properties of each member are
u5 共cm兲 −0.83282 −0.59153 −0.88035 −0.53708
considered uncertain as well. The variations are 1% for the cross-
sectional area and the moment of inertia, and 4% for the modulus
v5 共cm兲 −0.25154 −0.22315 −0.25212 −0.22174
of elasticity. The intervals used are summarized in Table 1. It is
5 共rad兲 −0.00420 −0.00347 −0.00426 −0.00339
assumed that all interval variables vary independently within their
u9 共cm兲 −1.62616 −1.25484 −1.73034 −1.13262
bounds.
v9 共cm兲 −0.21269 −0.188541 −0.21385 −0.18683 There are 34 interval variables involved in this example. The
9 共rad兲 0.00510 0.00613 0.00497 0.00626 combinatorial method requires 234 deterministic finite-element
Note: LB⫽lower bound and UB⫽upper bound. analysis 共FEA兲, which is computationally infeasible. Monte Carlo
sampling method is used to evaluate the quality of the results
Table 3. Bounds of Selected Member Nodal Forces for the Frame in Fig. 5
Monte Carlo sampling Present IFEA
Member 共node兲 Nodal force LB UB LB UB
B2 Axial 共kN兲 215.23 244.11 210.43 248.54
共left node兲
Shear 共kN兲 832.81 892.79 831.95 893.56
Moment 共kN m兲 1834.51 1989.67 1826.20 1995.95
C5 Axial 共kN兲 −619.25 −573.12 −620.40 −571.93
共bottom node兲
Shear 共kN兲 −291.27 −258.63 −294.86 −254.82
Moment 共kN m兲 −692.45 −612.55 −702.28 −600.71
Note: LB⫽denotes lower bound and UB⫽upper bound. Axial force: tension 共+兲 and moment: counter clockwise 共+兲.
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Table 6. Displacement Bounds of Node 80 for the Frame in Fig. 6
Present IFEM Monte Carlo sampling
Displacement LB UB LB UB
x 共cm兲 3.6581 7.7758 5.2014 6.2237
y 共cm兲 −1.1235 −0.6643 −0.9902 −0.7996
Note: LB⫽denotes lower bound and UB⫽upper bound.
one from the Monte Carlo sampling method. This suggests that
the overestimation of the bounds obtained by the present interval
FEA is small, and sharp bounds are obtained.
The second example is intended to illustrate the capability of
developed method in solving large size problems with large num-
Fig. 6. Four-bay 15-story frame ber of uncertain parameters. The presented problem is a four-bay
15-story frame as shown in Fig. 6. The member sections are listed
in Table 4. All beams are subjected to a distributed vertical load
obtained by the present interval FEA. One million realizations are with an unknown distribution but bounded between two constant
used with assumed uniform probability distribution for the inter- functions, and the left-hand side of the frame is subject to con-
val parameters. centrated forces, one concentrated force at each story. The loads
The displacements at nodes 5 and 9, and the member nodal are unknown but bounded by the following interval values:
force of member B2 共left node兲 and C5 共bottom node兲 are sum- Distributed load intensity
marized in Tables 2 and 3, respectively. As seen from the tables,
the solution obtained by the present method sharply encloses the w = 关42, 56兴 kN/ m 共关240,380兴兲 lb/ in.
Each concentrated force
p = 关17.8, 22.24兴 kN 共关4000, 5000兴兲 lb
Table 4. Section Properties for the Members of the Frame in Fig. 6
Element No. Section Element Number Section The geometric and material properties of each member are
1 , 6 , 11, 16, 5 , 10, 15, 20 W14⫻ 426 62–64,67–69, W14⫻ 176
considered uncertain as well. The variations are 2% of the nomi-
42–44,47–49,52–54,57–59 共columns兲 72–74 共columns兲 nal values for the cross-sectional area, the moment of inertia, and
the modulus of elasticity. The intervals used are summarized in
Table 5. It is assumed that all interval variables vary indepen-
2–4,7–9,12–14,17–19 W14⫻ 808 76–91 W24⫻ 84
共columns兲 共beams兲 dently within their bounds. The total number of the interval vari-
21,26,31,36,25 30,35,40 W14⫻ 311 92–107 W21⫻ 68
ables involved in this example is 480 共3 ⫻ 135+ 60+ 15兲. The
共columns兲 共beams兲 combinatorial method requires 2480 deterministic FEA analyses,
which is computationally infeasible and usually provides only an
inner enclosure in frame problems. A Monte Carlo integration
22–24,27–29,32–24,37–39 W14⫻ 605 108–123 W18⫻ 55
共columns兲 共beams兲 method is used to evaluate the quality of the results obtained by
the present interval FEA. Five million Monte Carlo realizations
41,46,51,56,45 50,55,60 W14⫻ 211 124–135 W18⫻ 40
共columns兲 共beams兲 are made assuming a uniform probability distribution for the un-
certain variables. The displacement at node 80 共right top corner兲,
and the member nodal forces of Column 1 are summarized in
61,66,71,65,70,75 W14⫻ 120
共columns兲 Tables 6 and 7, respectively. As seen from the tables, the solution
obtained by the present method represents an enclosure for that
obtained by the Monte Carlo sampling method. As it is clear, the
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Table 7. Bounds of Selected Member Nodal Forces for the Frame in Fig. 6
Present IFEM Monte Carlo sampling
sample space of this problem is huge and a more realistic Monte er’s fixed point theory with Krawczyk’s operator and a newly
Carlo simulation would require a computationally prohibited developed overestimation control. The numerical examples show
number of realizations to capture the worst case bounds on the the naïve interval FEM produces meaningless wide results. The
systems response. present approach eliminates most sources of overestimation and a
very sharp enclosure for the system response, due to loading and
stiffness uncertainty in linear static structural mechanics prob-
lems, is obtained. The method is computationally efficient.
Conclusion
if c ⱖ 0 and d ⱖ 0 if c ⬍ 0 ⬍ d if c ⱕ 0 and d ⱕ 0
if a ⱖ 0 and b ⱖ 0 关ac,bd兴 关bc,bd兴 关bc,ad兴
if a ⬍ 0 ⬍ b 关ad,bd兴 关min共ad,bc兲,max共ac,bd兲兴 关bc,ac兴
if a ⱕ 0 and b ⱕ 0 关ad,bc兴 关ad,ac兴 关bd,ac兴
•
Division: x / y • Subdistributivity
if c ⬎ 0 and d ⬎ 0 if c ⬍ 0 ⬍ d x ⫻ 共y ± z兲 債 x ⫻ y ± x ⫻ z; 共x ± y兲 ⫻ z 債 x ⫻ z ± y ⫻ z
if a ⱖ 0 and b ⱖ 0 关a/d,b/c兴 关b/d,a/c兴 • Subcancellation
if a ⬍ 0 ⬍ b 关a/c,b/c兴 关b/d,a/d兴
if a ⱕ 0 and b ⱕ 0 关a/c,b/d兴 关b/c,a/d兴 x − y 債 共x + z兲 − 共y + z兲; x/y 債 共x ⫻ z兲/共y ⫻ z兲;
On the other hand, only some of the algebraic laws valid for 0 苸 x − x; 1 苸 x/x
real numbers remain valid for intervals; other laws only hold in a
weaker form. The following laws hold for intervals x, y, and z.
• Commutativity References
x+y=y+x
Alefeld, G., and Herzberger, J. 共1983兲. Introduction to interval computa-
tions, Academic, New York.
x⫻y=y⫻x Apostolatos, N., and Kulisch, U. 共1968兲. “Grundzüge einer intervallrech-
• Associativity tung für matrizen und einige anwwendungen.” Elektron. Rechenan-
lagen, 10, 73–83 共in German兲.
共x + y兲 ± z = x + 共y ± z兲 Bathe, K. 共1996兲. Finite element procedures, Prentice-Hall, Upper Saddle
River, N.J.
Buonopane, S. G., Schafer, B. W., and Igusa, T. 共2003兲. “Reliability im-
共x ⫻ y兲 ⫻ z = x ⫻ 共y ⫻ z兲
plications of advanced analysis in design of steel frames.” Proc.,
The following laws represent weak forms of several laws from ASSCCA’03, Sydney, Australia.
the real arithmetic: Chen, S. H., Lian, H. D., and Yang, X. W. 共2002兲. Interval static displace-
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
ment analysis for structures with interval parameters.” Int. J. Numer. based methods for fuzziness in continuum mechanics.” Proc.,
Methods Eng., 53, 393–407. ISUMA-NAFIPS’95, 23–45.
Cook, R. D., Malkus, D. S., and Plesha, M. E. 共1989兲. Concepts and Muhanna, R. L., and Mullen, R. L. 共1999兲. Formulation of fuzzy finite
applications of finite element analysis, Wiley, New York. element methods for mechanics problems.” Comput. Aided Civ. Infra-
Dempster, A. P. 共1967兲. “Upper and lower probabilities induced by a struct. Eng., 14, 107–117.
multi-valued mapping.” Ann. Math. Stat., 38, 325–339. Muhanna, R. L., and Mullen, R. L. 共2001兲. “Uncertainty in mechanics
Dessombz, O., Thouverez, F., Laîné, J.-P., and Jézéquel, L. 共2001兲. problems—Interval-based approach.” J. Eng. Mech., 127共6兲, 557–
“Analysis of mechanical systems using interval computations applied 566.
to finite elements methods.” J. Sound Vib., 238共5兲, 949–968. Mullen, R. L., and Muhanna, R. L. 共1996兲. “Structural analysis with
Ferson, S., and Ginzburg, L. 共1995兲. “Hybrid arithmetic.” Proc., ISUMA-
fuzzy-based load uncertainty.” Proc., 7th ASCE EMD/STD Joint Spe-
NAFIPS’95, IEEE Computer Society Press, 619–623.
cialty Conf. on Probabilistic Mechanics and Structure Reliability,
Gallagher, R. H. 共1975兲. Finite element analysis fundamentals, Prentice-
310–313.
Hall, Englewood Cliffs, N.J.
Ganzerli, S., and Pantelides, C. P. 共1999兲. “Load and resistance convex Mullen, R. L., and Muhanna, R. L. 共1999兲. “Bounds of structural re-
models for optimum design.” Struct. Optim., 17, 259–268. sponse for all possible combinations.” J. Struct. Eng., 125共1兲,
Gay, D. M. 共1982兲. “Solving interval linear equations.” SIAM (Soc. Ind. 98–106.
Appl. Math.) J. Numer. Anal., 19共4兲, 858–870. Neumaier, A. 共1987兲. “Overestimation in linear interval equations.” SIAM
Hansen, E. 共1965兲. “Interval arithmetic in matrix computation. SIAM (Soc. Ind. Appl. Math.) J. Numer. Anal., 24共1兲, 207–214.
(Soc. Ind. Appl. Math.) J. Numer. Anal., 1共2兲, 308–320. Neumaier, A. 共1989兲. “Rigorous sensitivity analysis for parameter-
Hansen, E. 共1992兲. Global optimization using interval analysis, Marcel dependent systems of equations.” J. Math. Anal. Appl., 144, 14–25.
Dekker, New York. Neumaier, A. 共1990兲. Interval methods for systems of equations, Cam-
Jansson, C. 共1991兲. “Interval linear system with symmetric matrices, bridge University Press, Cambridge, Mass.
skew-symmetric matrices, and dependencies in the right hand side.” Pantelides, C. P., and Ganzerli, S. 共2001兲. “Comparison of fuzzy set and
Computing, 46, 265–274. convex model theories in structural design.” Mech. Syst. Signal Pro-
Joslyn, C. 共1996兲. Hybrid methods to represent incomplete and uncertain cess., 15共3兲, 499–511.
information.” Proc., 1996 Interdisciplinary Conf. on Intelligent Sys- Qiz, Z., and Elishakoff, I. 共1998兲. “Antioptimization of structures with
tems: A Semiotic Perspective, Vol. 2, R. Trappl, ed., Vienna, Austria, large uncertain-but-non-rand parameters via interval analysis.” Com-
905–910. put. Methods Appl. Mech. Eng., 152, 361–372.
Joslyn, C., and Kreinovich, V. 共2002兲. “Convergence properties of an Rao, S. S., and Berke, L. 共1997兲. “Analysis of uncertain structural sys-
interval probabilistic approach to system reliability estimation.” Rep. tems using interval analysis.” AIAA J., 35共4兲, 727–735.
No. LA-UR-02-6261, Los Alamos National Laboratory, Los Alamos, Rao, S. S., and Chen, L. 共1998兲. “Numerical solution of fuzzy linear
N.M. equations in engineering analysis.” Int. J. Numer. Methods Eng., 43,
Koyluoglu, H. U., and Elishakoff, I. 共1998兲. A comparison of stochastic 391–408.
and interval finite elements applied to shear frames with uncertain Rump, S. M. 共1983兲. “Solving algebraic problems with high accuracy.” A
stiffness properties.” Comput. Struct., 67, 91–98.
new approach to scientific computation, U. Kulisch and W. Miranker,
Mayer, O. 共1970兲. Algebraische und metrische strukturen in der interval-
eds., Academic, New York.
lrechung und eingine anwendungen.” Computing, 5, 144–162
Rump, S. M. 共1992兲. “On the solution of interval linear systems.” Com-
共in German兲.
McWilliam, S. 共2000兲. “Anti-optimisation of uncertain structures using puting, 47, 337–353.
interval analysis.” Comput. Struct., 79, 421–430. Rump, S. M. 共2001兲. “Self-validating methods.” Linear Algebr. Appl.,
Möller, B., Graf, W., and Beer, M. 共2000兲. “Fuzzy structural analysis 324, 3–13.
using level-optimization.” Comput. Mech., 26共6兲, 547–565. Shafer, G. 共1976兲. A mathematical theory of evidence, Princeton Univer-
Moore, R. E. 共1966兲. Interval analysis, Prentice-Hall, Englewood Cliffs, sity Press, Princeton, N.J.
N.J. Sun Microsystems. 共2002兲. Interval arithmetic in high performance tech-
Moore, R. E. 共1979兲. Methods and applications of interval analysis, nical computing, Sun Microsystems 共A White Paper兲.
SIAM, Philadelphia. Zadeh, L. A. 共1978兲. “Fuzzy sets as a basis for a theory of possibility.”
Muhanna, R. L., and Mullen, R. L. 共1995兲. “Development of interval Fuzzy Sets Syst., 1, 3–28.
Downloaded 17 Aug 2010 to 130.54.110.32. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org