This action might not be possible to undo. Are you sure you want to continue?
Interval Finite Elements
M. V. Rama Rao
1
, R. L. Mullen
2
, and R. L. Muhanna
3
1
Vasavi College of Engineering, Hyderabad  500 031 INDIA, dr.mvrr@gmail.com
2
Case Western Reserve University, Cleveland, OH, 441067201, USA, rlm@case.edu
3
Georgia Institute of Technology, Atlanta, GA, 31407 USA, rafi.muhanna@gtsav.gatech.edu
Abstract: This paper addresses the main challenge in interval computations which is to minimize the
overestimation in the target quantities. When sharp enclosures for the primary variables are achievable in a
given formulation such as the displacements in Interval Finite Elements (IFEM) the calculated enclosures
for secondary or derived quantities such as stresses usually obtained with significantly increased
overestimation. One should follow special treatment in order to decrease the overestimation in the derived
quantities see Muhanna, Zhang, and Mullen (2007), Neumaier and Pownuk (2007). In this work we
introduce a new formulation for Interval Finite Element Methods where both primary and derived quantities
of interest are included in the original uncertain system as primary variables. The formulation is based on
the variational approach and Lagrange multiplier method by imposing certain constraints that allows the
Lagrange multipliers themselves to be the derived quantities. Numerical results of this new formulation are
illustrated in a number of example problems.
Keywords: Interval; Uncertainty; Dependent Variables; Finite Elements.
1. Introduction
Since the early development of Interval Finite Element Methods (IFEM) during the mid nineties of last
century (Koyluoglu, H. U., Cakmak, A. S., and Nielson, S. R. K. 1995, Muhanna, R. L. and Mullen, R. L.
1995, Nakagiri S. and Yoshikawa, N. 1996, Rao, S. S. and Sawyer, P. 1995, Rao, S. S. and Berke, L. 1997,
Rao, S.S., and Chen Li 1998.) researchers have focused among other issues on two major problems; the first
is how to obtain solutions for the resulting linear interval system of equations with reasonable bounds on
the system response that make sense from practical point of view, or in other words with the least possible
overestimation of their bounding intervals, the second is how to obtain reasonable bounds on the derived
quantities that are functions of the system response. For example, when the system response is the
displacement, the derived quantities might be forces or stresses which are functions of the displacements.
Obtaining tight bounds on the derived quantities has been a tougher challenge due to the existing
4th International Workshop on Reliable Engineering Computing (REC 2010)
Edited by Michael Beer, Raﬁ L. Muhanna and Robert L. Mullen
Copyright © 2010 Professional Activities Centre, National University of Singapore.
ISBN: 9789810851187. Published by Research Publishing Services.
doi:10.3850/9789810851187 019
129
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
dependency of these quantities on the primary dependent variables which are already overestimated. So far,
the derived quantities are obtained with significantly increased overestimation.
A significant effort has been made in the work of Zhang (2005) to control the additional overestimation in
the values of the derived quantities; the derived quantities have been calculated by an implicit substitution
of the primary quantities. In addition to calculating rigorous bounds on the solution of the resulting linear
interval system, a special treatment has been developed to handle the overestimation in the derived
quantities. Instead of first evaluating the primary quantities and then substituting the obtained values in the
expression for the derived quantities, the expression for the primary quantities has been substituted before
its evaluation in the derived quantities expression and both were evaluated simultaneously preventing a
large amount of overestimation in the values of derived quantities. In spite of the advancement provided by
this approach, still it is conditioned by the original IFEM formulation and the special treatment of required
transformations.
A significant improvement in the formulation of IFEM with application to truss problems has been
introduced in the work of Neumaier and Pownuk (2007). This work has presented an iterative method for
computing rigorous bounds on the solution of linear interval systems, with a computable overestimation
factor that is frequently quite small. This approach has been demonstrated by solving truss problems with
over 5000 variables and over 10000 interval parameters, with excellent bounds for up to about 10% input
uncertainty. Although, no calculated derived quantities have been reported in this work, a formulation has
been introduced for the calculation of derived quantities by intersecting the simple enclosure z = Z(u),
where z depends linearly or nonlinearly on the solution u of the uncertain system with another enclosure
obtained from the centered form (Neumaier and Pownuk, 2007, Eq. 4.13, pp 157). In spite of the provided
improvement in this formulation, the twostep approach will result in additional overestimation when
evaluating the derived quantities.
It is quite clear that among other factors, the issue of obtaining tight enclosures for the primary variables as
well as for the derived quantities is conditioned by IFEM formulation and the methods used for the
evaluation of the derived quantities. In this work we introduce a new mixed formulation for Interval Finite
Element Methods where the derived quantities of the conventional formulation are treated as dependent
variables along with the primary variables. The formulation uses the mixed variational approach based on
the Lagrange multiplier method. The system solution provides the primary variables along with the
Lagrange multipliers which represent the derived quantities themselves. Numerical results of this new
formulation are illustrated in a number of example problems.
2. Formulation
In the current formulation, our focus will be on two major issues:
1. Obtaining the secondary variables (derived) such as forces and stresses of the conventional
displacement FEM along with the primary variables (displacements) and with the same accuracy of
the primary ones.
130 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
2. Reducing of overestimation in the bounds on the system response due to the coupling and
transformation in the conventional FEM formulation as well as due to the nature of used interval
linear solvers (Muhanna and Mullen, 2001).
We will begin the formulation with a short theoretical background with the hope that it will facilitate a
clearer understanding of the procedure followed in the present formulation. Interval quantities will be
introduced in boldface nonitalic font.
2.1. SECONDARY VARIABLES
Mixed or hybrid variational formulations are those where secondary variables of the conventional
formulation are treated as dependent variables along with the primary variables. Most often these
formulations are developed with the objective of determining the secondary variables, which are often
quantities of practical interest, directly rather than from postcomputations. Mixed formulations are based
on stationary principles. A stationary principle is one in which the functional attains neither a minimum nor
a maximum in its argument. In fact, a functional can attain a maximum with respect to one set of variables
and a minimum with respect to another set of variables involved in the functional. An example of such
functionals is provided by the functional based on the Lagrange multiplier method (Reddy, 2002). The
Lagrange multiplier method, which forms the basis for the present mixed formulation method, will be
introduced briefly in the next section.
2.2. LAGRANGE MULTIPLIER METHOD
The Lagrange multiplier method is one in which the minimum of a functional with linear equality
constraints is determined. If we consider the problem of finding the minimum of a functional I (u, v),
³
b
a
dx v v u u x F v u I ) , , , , ( ) , (
' '
(1)
subjected to the constraint
0 ) , , , (
' '
v v u u G (2)
where u , v , ' u and ' v are the dependent variables and their first derivatives, respectively. The necessary
condition for the minimum of ) , ( v u I is 0 I G . We have
dx u
v
F
v
v
F
u
u
F
u
u
F
I
b
a
³
¸
¹
·
¨
©
§
w
w
w
w
w
w
w
w
'
'
'
'
0 G G G G G
(3)
Since u and v must satisfy the constraint condition given by Eq. (2), the variation u G and v G are related by:
dx u
v
G
v
v
G
u
u
G
u
u
G
G
b
a
³
¸
¹
·
¨
©
§
w
w
w
w
w
w
w
w
'
'
'
'
0 G G G G G
(4)
4th International Workshop on Reliable Engineering Computing (REC 2010) 131
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
The Lagrange multiplier method consists of multiplying Eq. (4) with an arbitrary parameter O , integrating
over the interval (a, b), and adding the results to Eq. (3). The multiplier O is called the Lagrange multiplier.
The Lagrange’s method can be viewed as one of determining u , v andO by setting the first variation of the
modified functional
³ ³
{
b
a
b
a
dx G F dx v v u u G v u I v u L ) ( ) , , , ( ) , ( ) , , (
' '
O O O (5)
to zero.
We have
°
°
°
°
¿
°
°
°
°
¾
½
¿
¾
½
¯
®
»
¼
º
«
¬
ª
¸
¹
·
¨
©
§
w
w
w
w
w
w
w
w
»
¼
º
«
¬
ª
¸
¹
·
¨
©
§
w
w
w
w
w
w
w
w
»
¼
º
«
¬
ª
¸
¹
·
¨
©
§
w
w
w
w
¸
¹
·
¨
©
§
w
w
w
w
¸
¹
·
¨
©
§
w
w
w
w
¸
¹
·
¨
©
§
w
w
w
w
³
³
³
³
dx G v
v
G
v
F
dx
d
v
G
v
F
u
u
G
u
F
dx
d
u
G
u
F
dx G v
v
G
v
F
v
v
G
v
F
u
u
G
u
F
u
u
G
u
F
dx ) G G F (
dx ) G F ( L
b
a
' ' ' '
b
a
'
' '
'
' '
b
a
b
a
GO G O O G O O
GO G O G O G O G O
OG GO G
O G 0
(6)
The boundary terms vanish because ( ) u a G , ( ) u b G , ( ) v a G , ( ) v b G = 0. Suppose that u G is independent
and v G is related to u G by Eq. (4). We choose O such that the coefficient of v G is zero. Then by the
fundamental lemma of variational calculus, it follows that (because u G is arbitrary) the coefficient of u G is
also zero. Thus we have:
°
°
°
¿
°
°
°
¾
½
»
¼
º
«
¬
ª
w
w
w
w
»
¼
º
«
¬
ª
w
w
w
w
0 ) , , , (
0 ) ( ) (
0 ) ( ) (
' '
'
'
v v u u G
G F
v dx
d
G F
v
G F
u dx
d
G F
u
O O
O O
(7)
Equations (7) are the Euler equations of the functional
³
{
b
a
dx G F v u L ) ( ) , , ( O O from which the
dependent variables u, v, and O can be determined at the same time. In general, one can introduce the
secondary variable z = Z(u) as the constraint G = {[zZ(u)]
2
} = 0. Alternatively, we have found that the
Lagrange multiplier O is the pursued secondary variable by judicious choice of the constraint condition G(u,
132 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
uc, v ,vc) = 0. In the next section we will illustrate the use of the Lagrange multiplier method in discrete
structural models.
2.3. DISCRETE STRUCTURAL MODELS
In steadystate analysis, the variational formulation for a discrete structural model within the context of
Finite Element Method (FEM) is given in the following form of the total potential energy functional
(Gallagher 1975, Bathe 1996)
P U KU U
T T
3
2
1
(8)
with the conditions
i all for
U
i
0
w
3 w
(9)
where 3, K, U, and P are total potential energy, stiffness matrix, displacement vector, and load vector
respectively. Assume that we want to impose onto the solution the m linearly independent discrete
constraints V CU where C is a matrix of order m n u . In the Lagrange multiplier method we amend the
righthand side of Eq. (8) to obtain
) (
2
1
*
V CU P U KU U
T T T
3 O (10)
where O is a vector of m Lagrange multipliers. Invoking the stationarity of 3
*
, that is G3
*
=0, we obtain
¸
¸
¹
·
¨
¨
©
§
¸
¸
¹
·
¨
¨
©
§
¸
¸
¹
·
¨
¨
©
§
V
P U
C
C K
T
O 0
(11)
The solution of Eq. (11) will provide the values of dependent variable U and O at the same time.
The present interval formulation, which will be presented in the next section, is based on the ElementBy
Element (EBE) finite element technique developed in the work of Muhanna and Mullen 2001. In the EBE
method, each element has its own set of nodes, but the set of elements is disassembled, so that a node
belongs to a single element. A set of additional constraints is introduced to force unknowns associated with
coincident nodes to have identical values. Thus, the constraint equation CU=V takes the form
0 CU (12)
where C is the constraint matrix, and equation (11) takes the form:
4th International Workshop on Reliable Engineering Computing (REC 2010) 133
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
¸
¸
¹
·
¨
¨
©
§
¸
¸
¹
·
¨
¨
©
§
¸
¸
¹
·
¨
¨
©
§
0 0
P
Ȝ
U
C
C K
T
(13)
2.4. PRESENT INTERVAL FORMULATION
The main sources of overestimation in the formulation of IFEM are the multiple occurrences of the same
interval variable (dependency problem), the width of interval quantities, the problem size, and the problem
complexity, in addition to the nature of the used interval solver of the interval linear system of equations.
While the present formulation is valid for the FEM models in solid and structural mechanics problems, the
truss model will be used here to illustrate the applicability and efficiency of the present formulation of
without any loss of generality.
To illustrate the present formulation, let us consider a typical two dimensional truss bar finite element as
shown in Figure 1. According to finite element formulation (Bathe, 1996, Gallagher, 1995, Zienkiewicz and
Taylor, 2000) the global finite element model of a truss system is given in the following form:
P KU (14)
where K is the assembled global stiffness matrix, P is the global load vector, and U is the unknown global
displacement vector.
Using boldface nonitalic font for interval quantities, the interval form of Eq. (14) will be
P KU (15)
M
E, A, L
F
1X
, u
1X
F
2Y
, u
2Y
2
1
F
2
, u
2
F
1
, u
1
y
Y
X
x
F
1Y
, u
1Y
F
2X
, u
2X
Figure 1. A typical truss bar element, local and global coordinates.
134 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
where K, U, and P are the interval global stiffness matrix, interval global displacement vector, and interval
global load vector, respectively. The interval solution of Eq. (15) results in a significant overestimation in
the system response; a comprehensive discussion can be found in (Muhanna and Mullen 2001). In addition,
internal forces and stresses are quantities of practical interest in design. Usually interval element forces can
be obtained as:
U k F
e e e
L (16)
where F
e
, k
e
, L
e
are global interval vector of element forces, global interval element stiffness matrix, and
element Boolean matrix, respectively. Once again, an additional overestimation in the values of forces is
obtained due to the dependency between U and k
e
. Frequently, element forces are pursued in local
coordinate system that will require the transformation from the global coordinates to the local ones in the
form:
U k F
e e e local e,
L T (17)
where F
e, local
and T
e
are the local vector of interval element forces and the corresponding transformation
matrix, respectively. The transformation procedure will provide an additional overestimation.
The current formulation is attempting to reduce overestimation due to coupling in the FEM assembling
process, multiple occurrences of interval quantities, transformation, and solving the final system of interval
linear equations. In addition this formulation will introduce the derived quantities such as forces and
stresses as dependent variables which will be obtained along with displacements when the system is solved.
2 2
2
2
Element (m)
1
P
Y
Node (n)
(a)
Element (m)
u
Y
u
X
F
2m
, u
2m
F
1m
, u
1m
P
Y
2 2
1
2 1
2
1
1
Free node (n)
(b)
Figure 2. A typical node of a truss problem. (a) Conventional formulation. (b) Present formulation.
4th International Workshop on Reliable Engineering Computing (REC 2010) 135
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
In the conventional formulation of FEM Figure 2 (a), after deriving the local elements’ stiffness matrices
along with the local elements’ load vectors the system will be transformed to the global system and
assembled based on compatibility requirements resulting in the equilibrium system given by Eq. (14). In the
present formulation the following steps are followed:
1. Considering a typical node of the truss system Figure 2 (a), elements and nodes are disassembled as
in Figure 2 (b). The typical node is called a free node and is given along with all pertinent variables
in the global coordinate system. Displacements are
X
u and
Y
u and applied forces are
X
P and
Y
P .
The free node displacements are considered as independent of those of the elements.
2. All coinciding elements at the free node along with pertinent variables are given in local coordinate
system. For example, element m has the end nodes 1 and 2, the local displacements
1m
u and
2m
u ,
and the local forces
1m
F and
2m
F . By doing that, each element is treated as having independent
degrees of freedom in its own local coordinate system.
3. The system will be assembled imposing the discrete constraints C
mi
to ensure the equality between
the free node displacements and those of the elements. Where i is the number of constraints per
element.
This procedure will result in the following system of equations:
¸
¸
¹
·
¨
¨
©
§
¸
¸
¹
·
¨
¨
©
§
¸
¸
¹
·
¨
¨
©
§
0 0
P
Ȝ
U k
C
C
T
(18)
where k is an interval matrix consists of the individual elements’ local stiffness and zeros at the bottom
corresponding the free nodes’ degrees of freedom and have the following structure:
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¹
·
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
©
§
mY
mX
Y
X
n n
n n
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
1
1
1 1
1 1
k k
k k
k k
k k
k (19)
and
136 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
i
i
i
L
A E
k
i
= (20)
where E
i
, A
i
, and L
i
are the interval modulus of elasticity, the interval crosssectional area, and the length of
each element, respectively.
Matrix C has the dimensions (k × l), where k = number of elements’ degrees of freedom (2 × number of
elements in the truss bar element case), and l = total number of the system’s degrees of freedom. The entries
of the matrix are equality constraints of the following type
0
1
= + +
i jY i jX i
sin cos m m u u u (21)
Where u
1i
is the local displacement of the node 1 that belongs to i
th
element, u
jX
and u
jY
are the X and Y
global displacements of j
th
free node coinciding with the 1
st
node of the i
th
element. Elements of C
T
are
shown in Eq. (22)
U is a vector of size 1 s × where s is the number of elements’ local degrees of freedom + number of free
nodes’ global degrees of freedom. The entries of the vector are the interval local displacements of elements
followed by interval global displacements of the free nodes as shown in Eq. (23).Vector ì has the
dimension of total number of elements’ local degrees of freedom and given in Eq. (24). The entries of the
vector are the interval Lagrange multipliers that represent minus the local element forces in this case.














.

\

=
1
1
1
1
sin 0
cos 0
0 sin
0 cos
0 0
0 0
1 0
0 1
m
m
m
m
T
C
(22)














.

\

=
mY
mX
Y
X
n
n
u
u
u
u
u
u
u
u
U
1
1
2
1
21
11
(23)






.

\

=
n
n
2
1
21
11
Ȝ
Ȝ
Ȝ
Ȝ
Ȝ
(24)
Vector P is the interval load vector and has the dimension equal to the sum of elements degrees of freedom
and the free nodes degrees of freedom. The entries of the vector are given in Eq. (25).
( )
mY mX Y X
T
P P P P P
1 1
0 0 0 0 = (25)
The accuracy of the system solution depends mainly on the structure of Eq. (18) and on the nature of the
used solver. The solution of the interval system (18) provides the enclosures of the values of dependent
4th International Workshop on Reliable Engineering Computing (REC 2010) 137
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
variables which are the interval displacements U and interval element forces ì. An iterative solver is
discussed in the next section.
2.5. ITERATIVE ENCLOSURES
The best known method for obtaining very sharp enclosures of interval linear system of equations that have
the structure introduced in Eq. (26) is the iterative method developed in the work of Neumaier and Pownuk,
(2007). The current formulation results in the interval linear system of equations given in (18) which can be
introduced in the same structure of the following system:
b u ) D ( F a A B K + = + (26)
with interval quantities in D and b only. The quality of the enclosures is known to be superior only in the
special case where D is diagonal, however It is expected that the enclosures are also good in case D is block
diagonal with diagonal blocks small compared to the matrix size. The solution is obtained by performing
the following iterative scheme:
d v ) D {( d , v } d ) ( b ) ( ) { v · ÷ = · + + =
0
D ACB ACF ACa (27)
until some stopping criteria, and then the following enclosure is obtained:
d ) ( b ) ( ) ( u CB CF Ca + + = (28)
Where
v D d
d b v
d b u
) (
) ( :
0
1
÷ =
+ + =
+ + =
+ =
÷
D
ACB ACF ACa
CB CF Ca
A BD K C
0
(29)
In Neumaier’s work only excellent enclosures of the interval displacements are obtained. However for the
derived quantities such as forces he suggested an improved enclosure by intersecting the simple enclosure
z = Z(u) with the following enclosure:
d ) S ( ) b b )( S ( ) b ( z CB mid CF mid CF Z + ÷ + = (28)
which results in additional significant overestimation
The current formulation allows obtaining the interval displacement U and the accompanied interval derived
quantities ì with the same accuracy. A number of examples are introduced in the following sections that
illustrate the excellent accuracy of the developed method.
138 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
Figure 3. Eleven bar truss.
15 kN
3. Example Problems
Three example problems are chosen to illustrate the present approach and also to demonstrate its ability to
obtain sharp bounds to the displacements and forces even in the presence of large uncertainties and large
number of interval variables.
The first example chosen is an eleven bar truss (Muhanna and Mullen, 2001) as shown in Figure 3. The
displacements and forces of this statically indeterminate truss are dependent on the uncertainty present in
the modulus of elasticity and load. The results of this example allow us to investigate the effect of load and
stiffness uncertainty on the displacements and forces using various approaches presented. A cantilever
truss as shown in Figure 8 is chosen as second example. This truss structure is a benchmark problem
adopted from the website of the Center for Reliable Computing
(http://www.gtsav.gatech.edu/rec/benchmarks.html). The objective of choosing this example is to
demonstrate the applicability, computational efficiency and scalability of the present approach for structures
with complex configuration with a large number of interval parameters. The third example problem is a
fifteen bar truss as shown in Figure 11 (Zhang, 2005). This truss is internally indeterminate but externally
determinate. Thus, the support reactions as well as axial forces in elements 1,2,14 and 15 are independent of
structural stiffness although structural stiffness is uncertain while the axial forces in the remaining elements
are dependent on structural stiffness. The objective of choosing this example is to verify the ability of
Neumaier’s approach and present approach to capture this phenomenon.
Considering the first example problem, the eleven bar truss is subjected to a concentrated load of 15 kN,
applied at the middle lower joint. The deterministic value of Young’s modulus of each element is
E
i
=2u10
11
N/m
2
, i=1,2,…,11, while the cross sectional area is 0.01m
2
. The modulus of elasticity of each
element is assumed to vary independently. The solution is computed using four approaches viz. a
combinatorial approach, Krawczyk Fixed Point Iteration (FPI)this approach uses the same system structure
as given by Neumaier and a fixed point iterative solver, Neumaier’s approach and the present approach.
Tables 1, 2 and 3 show the computed values of selected displacements (vertical displacement V
2
at node 2,
horizontal displacement U
4
and vertical displacement V
4
at node 4) using four approaches mentioned above
for uncertainties of 1 percent, 12 percent and 25 percent (r0.5%, r6% and r12.5% from the mean value of
4th International Workshop on Reliable Engineering Computing (REC 2010) 139
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
E, respectively. The error in width is computed as
¸
¸
¹
·
¨
¨
©
§
u 1 100
solution ial combinator of width
solution computed of width
. It is
observed from Table 1 that the displacements obtained using the present approach at 1 percent uncertainty
provides sharp enclosure to the displacements obtained using the combinatorial approach and also agree
very well with the displacements obtained using Neumaier’s approach and Krawczyk FPI . It is observed
from Table 2 that for a large uncertainty of 12% (and beyond), Krawczyk FPI fails to provide any solution
(no enclosure is reached). However the present approach and Neumaier’s approach still provide solution
with reasonable sharpness for this level of uncertainty. For a comparison, the error in width of vertical
displacement at node 2 (V
2
) varies from 0.71 at 1% uncertainty to 9.02 at 12% uncertainty, while the error
in width is 20.24 at 25% uncertainty. Thus it is observed that both Neumaier’s approach and the present
approach provide guaranteed bounds on the combinatorial approach.
Table 1 Eleven bar truss  displacements for 1% uncertainty in the modulus of elasticity (E)
V
2
u10
5
U
4
u10
5
V
4
u10
5
Lower Upper Lower Upper Lower Upper
Combinatorial approach 15.024443 14.874946 2.923326 3.003075 0.748810 0.732849
Krawczyk FPI 15.024603 14.874039 2.922944 3.003347 0.748841 0.732732
Error %(width) 0.71 0.81 0.93
Neumaier’s approach 15.024602 14.874039 2.922943 3.003347 0.748841 0.732731
Error %(width) 0.71 0.81 0.93
Present approach 15.024603 14.874039 2.922944 3.003347 0.748841 0.732732
Error %(width) 0.71 0.81 0.93
Table 2 Eleven bar truss displacements for 12% uncertainty in the modulus of elasticity (E)
V
2
u10
5
U
4
u10
5
V
4
u10
5
Lower Upper Lower Upper Lower Upper
Combinatorial approach 15.903532 14.103133 2.490376 3.451843 0.843182 0.650879
Krawczyk FPI      
Neumaier’s approach 15.930764 13.967877 2.431895 3.4943960 0.848475 0.633096
Error %(width) 9.02 10.50 11.99
Present approach 15.930764 13.967877 2.431895 3.494396 0.848475 0.633096
Error %(width) 9.02 10.50 11.99
Table 3 Eleven bar truss displacements for 25% uncertainty in the modulus of elasticity (E)
V
2
u10
5
U
4
u10
5
V
4
u10
5
Lower Upper Lower Upper Lower Upper
Combinatorial approach 17.084938 13.288285 1.983562 4.013033 0.971261 0.565208
Neumaier’s approach 17.231940 12.666701 1.705818 4.220473 0.999561 0.482011
Error %(width) 20.24 23.90 27.45
Present approach 17.231940 12.666701 1.705818 4.220473 0.999561 0.482011
Error %(width) 20.24 23.90 27.45
Figure 4 shows the computed interval values of vertical displacement V
2
at node 2. The figure depicts the
variation of the width of the present approach and the combinatorial approach with the variation of modulus
of Elasticity (E) from its mean value. It is observed from this figure that the present solution encloses the
combinatorial solution at all values of variation from 0 percent to 25 percent. A similar behaviour is
observed in the plot for variation of width of axial force N
3
in element 3 in Figure 5.
140 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
Axial forces are computed for the eleven bar truss using the following approaches
a) Simple enclosure z
1
(u)
b) z
2
(u), the intersection of z
1
(u) with the enclosure obtained using Eq. (28) of Neumaier.
c) Present approach
The interval values of axial forces in elements 3 and 9 (N
3
and N
9
) are presented in Table 4. It is clearly
observed from this table that the present method provides very sharp enclosure to the forces in comparison
7.3
6.8
6.3
5.8
5.3
4.8
0% 5% 10% 15% 20% 25%
Percentage variation of E about the mean
A
x
i
a
l
f
o
r
c
e
N
3
(
k
N
)
N3 comb
N3 present
Figure 5. Eleven bar truss – variation of axial force in element 3 with uncertainty of E. Comb. solution vs. present solution.
Figure 4. Eleven bar truss  variation of vertical displacement at node 2 with uncertainty of E. Comb. solution vs. present solution.
1.8E04
1.7E04
1.6E04
1.5E04
1.4E04
1.3E04
0% 5% 10% 15% 20% 25%
Percentage variation of E about the mean
V
e
r
t
i
c
a
l
d
i
s
p
l
a
c
e
m
e
n
t
V
2
(
m
)
V2 comb
V2 present
4th International Workshop on Reliable Engineering Computing (REC 2010) 141
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
with both the enclosures suggested by Neumaier, even at an uncertainty as large as 10%. This illustrates the
ability of the present approach to obtain sharp bounds to displacements and forces even at larger values of
uncertainty.
The eleven bar truss mentioned above is analysed once again at various levels of uncertainty of Young’s
modulus and load and the results are tabulated. Table 5 presents the computed values of selected
displacements for the second case study with uncertainty of Young’s modulus and load being 1 percent.
Table 6 presents the corresponding values of axial forces in elements 3 and 9. It is observed from these
tables that the present solution gives very sharp enclosure to the values of displacements and forces.
Further, the results of displacements obtained using the present approach agree very well with the results
obtained using Neumaier’s approach. It is observed that Krawczyk FPI fails to provide an enclosure at
11.5 percent uncertainty of both load and modulus of elasticity (E). Figure 6 shows the variation of vertical
displacement at node 4 with the variation of uncertainty of Young’s modulus and load. Figure 7 shows the
variation of axial force in element 9 with the variation of uncertainty of Young’s modulus and load. It is
observed from these figures that the present solution encloses the combinatorial solution at all levels of
uncertainty.
Table 4 Eleven bar truss  comparison of axial forces for 10% uncertainty in the modulus of elasticity (E) for various
approaches
3
( ) N kN
3
( ) N kN
9
( ) N kN
9
( ) N kN
Combinatorial approach 6.28858 5.57152 10.54135 9.73966
Simple enclosure z
1
(u) 7.89043 3.96214 11.89702 8.39240
Error %(width) 447.83 337.15
Intersection z
2
(u) 6.82238 5.08732 11.32576 9.02784
Error %(width) 141.97 186.63
Present approach 6.31656 5.53601 10.58105 9.70837
Error %(width) 8.85 8.85
Table 5 Eleven bar truss displacements for 1% uncertainty in the modulus of elasticity (E) and load
V
2
u10
5
U
4
u10
5
V
4
u10
5
Lower Upper Lower Upper Lower Upper
Combinatorial approach 15.09956 14.80057 2.90870 3.01809 0.75255 0.72918
Neumaier’s approach 15.09972 14.79891 2.90792 3.01836 0.75258 0.72898
Error %(width) 0.60 0.96 0.98
Present approach 15.09972 14.79891 2.90792 3.01836 0.75258 0.72898
Error %(width) 0.60 0.96 0.98
Table 6 Eleven bar truss  Axial forces for 1% uncertainty in the the modulus of elasticity (E) and load
N
3
(kN) N
9
(kN)
Lower Upper Lower Upper
Combinatorial approach 5.99198 5.861027 10.23567 10.05406
Intersection z
2
(u) 6.03902 5.81438 10.30685 9.98368
Error %(width) 71.53 77.95
Present approach 5.99224 5.86033 10.23604 10.05338
Error %(width) 0.72 0.58
142 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
The cantilever truss structure as described earlier is shown in Figure 8. This structure has 101 elements with
interval modulus of elasticity (E). The numerical values adopted to analyse the truss are P= 1000 N, L = 1
m, A = 0.01 m
2
, and E = 2u10
11
N/m
2
. Maximum uncertainty allowed in modulus of Elasticity (E) is
5 percent (r2.5% from the mean value of E). Four different solutions are presented for the problem using
the following approaches
Figure 7. Eleven bar truss – Variation of axial force in element 9 with uncertainty of load and E. Comb. solution vs, present
solution.
13
12
11
10
9
8
7
0% 5% 10% 15% 20% 25%
Percentage variation of E and load about the mean
A
x
i
a
l
F
o
r
c
e
N
9
(
k
N
)
N9 Comb
N9 Present
Figure 6. Eleven bar truss – Variation of vertical displacement at node 4 with uncertainty of load and E. Comb. solution
vs present solution.
1.2E05
1.1E05
1.0E05
9.0E06
8.0E06
7.0E06
6.0E06
5.0E06
4.0E06
3.0E06
0% 5% 10% 15% 20% 25%
Percentage variation of E and load about mean
V
e
r
t
i
c
a
l
d
i
s
p
l
a
c
e
m
e
n
t
V
4
V4 comb
V4 Present
4th International Workshop on Reliable Engineering Computing (REC 2010) 143
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
Figure 8. Benchmark problem – Cantilever truss.
Elementbyelement (EBE) method (Muhanna, Zhang and Mullen, 2007)
Neumaier’s method (Neumaier and Pownuk, 2007)
Pownuk’s sensitivity analysis (Pownuk, 2004) and
Present approach.
The results for the horizontal and vertical displacement at the right upper corner (node D) of the truss are
computed. These values are used for the computation of nondimensional constants
xD
C and
yD
C defined
as
§ ·
¨ ¸
© ¹
xD D
AE
C U
PL
and
§ ·
¨ ¸
© ¹
yD D
AE
C V
PL
. Values of
xD
C and
yD
C are computed for all the approaches
and are presented in Tables 3 and 4. It is observed from Tables 7 and 8 that widths of enclosures obtained
using the present approach agree very well with the Neumaier’s approach as well as sensitivity analysis.
The variation of the width of the enclosure with uncertainty of Young’s modulus for displacements U
D
and
V
D
can be observed from the Figures 9 and 10. It is further observed that the EBE method gives
overestimated results compared to the other three methods. Further, it is to be noted that sensitivity analysis
provides an inner bound solution to the displacements.
Table. 7 Cantilever truss nondimensional width
xD
C of displacement
D
U
Uncertainty (%) 0.0 1.0 2.0 3.0 4.0 5.0
EBE approach 0.0 218.50 535.30 1002.7 1700.2 2747.80
Neumaier’s approach 0.0 183.16 368.34 555.71 745.3 937.20
Present approach 0.0 183.12 368.33 555.67 745.19 936.95
Sensitivity analysis 0.0 182.10 364.20 546.4 728.65 911.10
Table.8 Cantilever truss  nondimensional width
yD
C of displacement
D
V
Uncertainty (%) 0.0 1.0 2.0 3.0 4.0 5.0
EBE approach 0.0 14.48 46.03 104.84 206.84 376.89
Neumaier’s approach 0.0 8.41 16.92 25.54 34.28 43.14
Present approach 0.0 8.40 16.91 25.53 34.26 43.11
Sensitivity analysis 0.0 8.35 16.70 25.06 33.41 41.78
144 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
0
500
1000
1500
2000
2500
3000
0% 1% 2% 3% 4% 5%
Percentage uncertainty
N
o
n

d
i
m
e
n
s
i
o
n
a
l
w
i
d
t
h
o
f
s
o
l
u
t
i
o
n
C
EBE approach
Neumaier's approach
Present approach
Sensitivity approach
Figure 9. Cantilever truss  variation of nondimensional widths of inte val solution of
D
U w.r.t. uncertainty of E.
0
50
100
150
200
250
300
350
400
0% 1% 2% 3% 4% 5%
Percentage uncertainty
N
o
n

d
i
m
e
n
s
i
o
n
a
l
w
i
d
t
h
o
f
i
n
t
e
r
v
a
l
s
o
l
u
t
i
o
n
C
EBE approach
Neumaier's approach
Present approach
Sensitivity approach
Figure 10. Variation of nondimensional widths of interval solution of
D
V w.r.t. uncertainty of E.
4th International Workshop on Reliable Engineering Computing (REC 2010) 145
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
The fifteen element truss shown in Figure 11 is subjected to vertical point loads of P
1
=200 kN, P
2
=100
kN,P
3
=100 kN and a horizontal point load P
4
=90 kN applied at the joints 5, 2, 6 and 3 respectively. Cross
section areas of elements 1,2,3,13,14 and 15 are 10.0u10
4
m
2
while for the rest of the elements is the cross
sectional area is 6.0u10
4
m
2
. The deterministic value of Young’s modulus of each element is
E
i
=2u10
11
N/m
2
, i =1, 2 ,…15, while the cross sectional area is 0.01 m
2
. The modulus of elasticity of each
element is assumed to vary independently. Results are computed using combinatorial approach, Neumaier’s
approach and present approach. The following two case studies are taken up to demonstrate the
effectiveness of the present approach.
x Case A: 10 percent uncertainty in modulus of Elasticity (E) while loads are deterministic.
x Case B: 10 percent uncertainty in both modulus of Elasticity (E) and loads
Table 9 presents the axial forces in elements 1,2,14 and 15 for case A. It is observed that the forces in these
elements computed using the present approach as well as combinatorial solution are thin intervals and
match exactly with each other. This is because the axial forces in these elements are independent of
structural stiffness. However, it is observed from Table 9 that forces in these elements obtained using
Neumaier’s approach have interval values. Thus Neumaier’s approach fails to capture the deterministic
nature of axial forces in these elements. Error in bounds is computed because error in width can not be
computed owing to zero width of combinatorial solution. Table 10 shows the corresponding axial forces in
the elements 3, 4 and 5. It is further observed that the solution obtained by present approach matches
exactly with the combinatorial solution while Neumaier’s approach gives an overestimated solution.
Table 9 Forces (kN) in elements 1,2, 14 and 15 of fifteen element truss for 10% uncertainty in the modulus of Elasticity
N
1
(kN) N
2
(kN) N
14
(kN) N
15
(kN)
Lower Upper Lower Upper Lower Upper Lower Upper
Combinatorial
approach
267.500 267.500 251.022 251.022 222.500 222.500 314.662 314.662
Neumaier’s
approach
240.750 295.660 277.450 225.920 200.250 245.92 347.780 283.200
Error %(bounds) 10.00 10.53 10.53 10.00 10.00 10.53 10.52 10.00
Present approach 267.500 267.500 251.022 251.022 222.500 222.500 314.662 314.662
Error %(bounds) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Figure 11. Fifteen bar truss.
146 4th International Workshop on Reliable Engineering Computing (REC 2010)
Primary and Derived Variables with the Same Accuracy in Interval Finite Elements
Table 10 Forces (kN) in elements 3,4 and 5 of fifteen element truss for 10% uncertainty in the modulus of
Elasticity (E)
N
3
(kN) N
4
(kN) N
5
(kN)
Lower Upper Lower Upper Lower Upper
Combinatorial approach 115.221 126.900 329.778 318.099 38.043 21.526
Neumaier’s approach 103.379 139.929 361.014 288.885 41.594 18.910
Error %(width) 114.609 330.390 38.858
Present approach 114.609 127.476 330.390 317.523 38.858 20.661
Error %(width) 10.17 10.17 10.17
Table 11 depicts the horizontal and vertical displacements U
5
and V
5
of node 5 for case B. It is observed
from Table 11 that, even for a large uncertainty of 10 percent, the displacements obtained using the present
approach give excellent bounds to combinatorial solution and also compare very well with the results of
Neumaier’s approach.
Table 11 Fifteen element truss –displacements for 10% uncertainty i the modulus of elasticity (E) and load
U
5
u10
2
V
5
u10
2
Lower Upper Lower Upper
Combinatorial approach 1.51430 1.87214 6.56423 5.37344
Neumaier’s approach 1.49353 1.87429 6.57150 5.30663
Error %(width) 6.40 6.22
Present approach 1.49353 1.87429 6.57150 5.30663
Error %(width) 6.40 6.22
Table 12 presents the axial forces in selected elements for case B. It is observed from Table 12 that the
widths of the axial forces in elements 1,2,14 and 15 computed using Neumaier’s approach are quite large in
while the corresponding errors in widths are zero in the case of forces obtained using present approach.
Also, the axial forces in elements 3, 4 and 5 computed using the present approach provide a sharp enclosure
to combinatorial solution while Neumaier’s approach provides overestimated bounds to combinatorial
solution. Thus it is concluded that the present approach provides very sharp enclosures to the axial forces
obtained using combinatorial solution while Neumaier’s solution provides overestimated bounds.
Table 12 Forces (kN) in elements of fifteen element truss for 10% uncertainty in modulus of elasticity (E) and load
Element Combinatorial approach Neumaier’s approach Present approach
LB UB LB UB
%Error
in width LB UB
%Error
in width
1 254.125 280.875 227.375 310.440 210.53 254.125 280.875 0.000
2 266.756 235.289 294.835 210.187 169.01 266.756 235.289 0.000
3 108.385 134.257 95.920 148.174 101.97 107.098 134.987 7.797
4 346.267 302.194 379.167 272.461 142.12 347.003 300.909 4.585
5 43.854 16.275 48.143 12.985 27.48 44.975 14.543 10.344
14 211.375 233.625 189.125 258.217 210.53 211.375 233.625 0.000
15 330.395 298.929 365.174 267.463 210.53 330.395 298.929 0.000
4. Conclusions
A new formulation for Interval Finite Element Methods is introduced. In this approach, both
primary and derived quantities of interest are included in the original uncertain system as primary
4th International Workshop on Reliable Engineering Computing (REC 2010) 147
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
variables. The formulation is based on the variational approach and Lagrange multiplier method
involving imposition of certain constraints that allow the Lagrange multipliers themselves to be
the derived quantities. Numerical results of this new formulation are illustrated in a number of
example problems. It is observed that the displacements obtained by present approach provide a
sharp enclosure to combinatorial solution and agree very well with the results obtained by
Neumaier at all uncertainties. However, Krawczyk’s Fixed Point Iteration fails to provide any
enclosure to the solution at large uncertainties. Further, the forces computed using the present
approach provide sharp enclosure to the combinatorial solution while forces computed using the
approach suggested by Neumaier are found to yield significantly overestimated results. The
present approach captures exactly the behaviour of statically determinate structures where the
internal forces in elements are independent of material properties while all previous methods do
not.
The present method addresses the basic issue of eliminating the additional overestimation in the
derived quantities by adopting the mixed formulation that makes possible the simultaneous
computation of primary and derived variables at the same level of accuracy. The present approach
can find further application in the area of nonlinear problems of structural mechanics involving
large uncertainties of structural parameters.
References
Bathe, K. Finite Element Procedures, Printice Hall, Englewood Cliffs, New Jersey, 1996.
Gallagher, R. H. Finite Element Analysis Fundamentals, Printice Hall, Englewood Cliffs, New Jersey, 1995.
Köylüoglu, U., Cakmak, S., Ahmet, N., and Soren, R. K. Interval Algebra to Deal with Pattern Loading and Structural Uncertainty.
Journal of Engineering Mechanics, 121(11): 1149 1157, 1995.
Muhanna, R. L. and Mullen, R. L.: Uncertainty in Mechanics Problems—IntervalBased Approach, J. Engrg. Mech. 127 (6), pp.
557–566, 2001.
Muhanna, R. L. and Mullen, R. L. Development of Interval Based Methods for Fuzziness in Continuum Mechanics. In Proceedings
of ISUMANAFIPS’95: 145150, September 1720, 1995.
Nakagiri, S. and Yoshikawa, N. Finite Element Interval Estimation by Convex Model. In Proceedings of 7th ASCE EMD/STD Joint
Specialty Conference on Probabilistic Mechanics and Structural Reliability, WPI, MA, 278281, August 79, 1996.
Neumaier, A. and Pownuk, A. Linear Systems with Large Uncertainties, with Applications to Truss Structures. Reliable
Computing, 13(2): 149172, 2007.
Pownuk, A.: Calculation of the Extreme Values of Displacements in Truss Structures with Interval Parameters, 2004,
http://webapp.math.utep.edu:8080/~andrzej/php/ansys2interval/
Rao, S. S. and Sawyer, P. Fuzzy finite element approach for analysis of imprecisely defined systems. AIAA J., 33(12): 2364–2370,
1995.
Rao, S. S. and Berke, L. Analysis of uncertain structural systems using interval analysis. AIAA J., 35(4): 727–735, 1997.
Rao, S. S. and Chen, Li. Numerical solution of fuzzy linear equations in engineering analysis. Int. J. Numer. Methods in Engrg.,43:
391–408, 1998.
Reddy, J. N. Energy principles and variational methods in applied mechanics, John Wiley & Sons, Inc. 2
nd
edition, 2002.
Zhang, H. Nondeterministic Linear Static Finite Element Analysis: An Interval Approach. Ph.D. Dissertation, Georgia Institute of
Technology, School of Civil and Environmental Engineering, 2005.
Zienkiewicz,O.C. and Taylor,R.L The Finite Element Method, Butterworth Heinemann, Oxford, UK, 2000.
148 4th International Workshop on Reliable Engineering Computing (REC 2010)
M. V. Rama Rao, R. L. Mullen, and R. L. Muhanna
dependency of these quantities on the primary dependent variables which are already overestimated. So far, the derived quantities are obtained with significantly increased overestimation. A significant effort has been made in the work of Zhang (2005) to control the additional overestimation in the values of the derived quantities; the derived quantities have been calculated by an implicit substitution of the primary quantities. In addition to calculating rigorous bounds on the solution of the resulting linear interval system, a special treatment has been developed to handle the overestimation in the derived quantities. Instead of first evaluating the primary quantities and then substituting the obtained values in the expression for the derived quantities, the expression for the primary quantities has been substituted before its evaluation in the derived quantities expression and both were evaluated simultaneously preventing a large amount of overestimation in the values of derived quantities. In spite of the advancement provided by this approach, still it is conditioned by the original IFEM formulation and the special treatment of required transformations. A significant improvement in the formulation of IFEM with application to truss problems has been introduced in the work of Neumaier and Pownuk (2007). This work has presented an iterative method for computing rigorous bounds on the solution of linear interval systems, with a computable overestimation factor that is frequently quite small. This approach has been demonstrated by solving truss problems with over 5000 variables and over 10000 interval parameters, with excellent bounds for up to about 10% input uncertainty. Although, no calculated derived quantities have been reported in this work, a formulation has been introduced for the calculation of derived quantities by intersecting the simple enclosure z = Z(u), where z depends linearly or nonlinearly on the solution u of the uncertain system with another enclosure obtained from the centered form (Neumaier and Pownuk, 2007, Eq. 4.13, pp 157). In spite of the provided improvement in this formulation, the twostep approach will result in additional overestimation when evaluating the derived quantities. It is quite clear that among other factors, the issue of obtaining tight enclosures for the primary variables as well as for the derived quantities is conditioned by IFEM formulation and the methods used for the evaluation of the derived quantities. In this work we introduce a new mixed formulation for Interval Finite Element Methods where the derived quantities of the conventional formulation are treated as dependent variables along with the primary variables. The formulation uses the mixed variational approach based on the Lagrange multiplier method. The system solution provides the primary variables along with the Lagrange multipliers which represent the derived quantities themselves. Numerical results of this new formulation are illustrated in a number of example problems.
2. Formulation In the current formulation, our focus will be on two major issues: 1. Obtaining the secondary variables (derived) such as forces and stresses of the conventional displacement FEM along with the primary variables (displacements) and with the same accuracy of the primary ones.
130
4th International Workshop on Reliable Engineering Computing (REC 2010)
A stationary principle is one in which the functional attains neither a minimum nor a maximum in its argument. will be introduced briefly in the next section. 2. The Lagrange multiplier method. directly rather than from postcomputations. The necessary condition for the minimum of I ( u . u ' and v ' are the dependent variables and their first derivatives.Primary and Derived Variables with the Same Accuracy in Interval Finite Elements 2. u. Reducing of overestimation in the bounds on the system response due to the coupling and transformation in the conventional FEM formulation as well as due to the nature of used interval linear solvers (Muhanna and Mullen. v ) subjected to the constraint b a F ( x. v ' ) 0 (2) where u . SECONDARY VARIABLES Mixed or hybrid variational formulations are those where secondary variables of the conventional formulation are treated as dependent variables along with the primary variables. If we consider the problem of finding the minimum of a functional I (u. Mixed formulations are based on stationary principles. In fact. 2. 2001). which forms the basis for the present mixed formulation method. LAGRANGE MULTIPLIER METHOD The Lagrange multiplier method is one in which the minimum of a functional with linear equality constraints is determined.1. v ' )dx (1) G (u. Most often these formulations are developed with the objective of determining the secondary variables. I ( u. Interval quantities will be introduced in boldface nonitalic font. 2002). v. We will begin the formulation with a short theoretical background with the hope that it will facilitate a clearer understanding of the procedure followed in the present formulation. u ' . which are often quantities of practical interest. We have 0 I b a F u u F u' u' F v v F u ' dx v' (3) Since u and v must satisfy the constraint condition given by Eq. a functional can attain a maximum with respect to one set of variables and a minimum with respect to another set of variables involved in the functional. An example of such functionals is provided by the functional based on the Lagrange multiplier method (Reddy. u ' . v . v). (2). v ) is I 0 . v. respectively. the variation u and v are related by: 0 G b a G u u G u' u' G v v G u ' dx v' (4) 4th International Workshop on Reliable Engineering Computing (REC 2010) 131 .2.
b). integrating over the interval (a. Thus we have: u v (F (F G) G) d dx d dx u' v' (F (F G) G) 0 0 (7) G ( u. V. L. Alternatively. (4) with an arbitrary parameter . one can introduce the secondary variable z = Z(u) as the constraint G = {[zZ(u)]2} = 0. 132 4th International Workshop on Reliable Engineering Computing (REC 2010) . v(b) = 0. Rama Rao. (4). (3).M. Mullen. v. Muhanna The Lagrange multiplier method consists of multiplying Eq. ) b a (F G )dx from which the dependent variables u. v(a) . v ) b a G (u. Suppose that u is independent and v is related to u by Eq. v . and adding the results to Eq. ) to zero. We choose such that the coefficient of v is zero. The Lagrange’s method can be viewed as one of determining u . v ' )dx b a (F G )dx (5) 0 L a b a b (F G )dx G G u G u G )dx u d dx F u' F u' G u' G u' u' u F v F v G v G v v d dx F v' F v' G v' G v' v' v G G dx dx (6) ( F F u F u b a b a The boundary terms vanish because u (a ) . u ' . and can be determined at the same time. and R. u (b) . v ' ) 0 Equations (7) are the Euler equations of the functional L(u. v. R. v and by setting the first variation of the modified functional L( u . L. We have I ( u. v. u ' . The multiplier is called the Lagrange multiplier. v . In general. Then by the fundamental lemma of variational calculus. it follows that (because u is arbitrary) the coefficient of u is also zero. we have found that the Lagrange multiplier is the pursued secondary variable by judicious choice of the constraint condition G(u.
U.3. (11) The solution of Eq.Primary and Derived Variables with the Same Accuracy in Interval Finite Elements u . A set of additional constraints is introduced to force unknowns associated with coincident nodes to have identical values. is based on the ElementByElement (EBE) finite element technique developed in the work of Muhanna and Mullen 2001. K. but the set of elements is disassembled. the constraint equation CU=V takes the form CU 0 (12) where C is the constraint matrix. and equation (11) takes the form: 4th International Workshop on Reliable Engineering Computing (REC 2010) 133 . Thus. (8) to obtain * 1 T U KU U T P 2 T (CU V ) * (10) . and load vector respectively. stiffness matrix. so that a node belongs to a single element. Assume that we want to impose onto the solution the m linearly independent discrete constraints CU V where C is a matrix of order m n . In the next section we will illustrate the use of the Lagrange multiplier method in discrete structural models. In the EBE method. which will be presented in the next section. Bathe 1996) 1 T U KU U T P 2 with the conditions (8) Ui 0 for all i (9) where . the variational formulation for a discrete structural model within the context of Finite Element Method (FEM) is given in the following form of the total potential energy functional (Gallagher 1975. we obtain K C CT 0 U P V at the same time. DISCRETE STRUCTURAL MODELS In steadystate analysis. and P are total potential energy. (11) will provide the values of dependent variable U and The present interval formulation. each element has its own set of nodes.v ) = 0. In the Lagrange multiplier method we amend the righthand side of Eq. displacement vector. 2. Invoking the stationarity of =0. v . that is * where is a vector of m Lagrange multipliers.
According to finite element formulation (Bathe. While the present formulation is valid for the FEM models in solid and structural mechanics problems. Gallagher. u2 F2X. u2Y 2 E. and R. u1X 1 y x X F1. To illustrate the present formulation. R. the interval form of Eq. 2000) the global finite element model of a truss system is given in the following form: KU P (14) where K is the assembled global stiffness matrix. PRESENT INTERVAL FORMULATION CT 0 U P 0 (13) The main sources of overestimation in the formulation of IFEM are the multiple occurrences of the same interval variable (dependency problem). u1 F1Y. Mullen. 1996. local and global coordinates. Using boldface nonitalic font for interval quantities. P is the global load vector. L. the width of interval quantities. Muhanna K C 2. A typical truss bar element. u1Y Figure 1. u2X Y F1X. (14) will be KU P (15) 134 4th International Workshop on Reliable Engineering Computing (REC 2010) . V. the truss model will be used here to illustrate the applicability and efficiency of the present formulation of without any loss of generality. A. the problem size. F2Y. L.4. let us consider a typical two dimensional truss bar finite element as shown in Figure 1. in addition to the nature of the used interval solver of the interval linear system of equations. L F2.M. Zienkiewicz and Taylor. Rama Rao. 1995. and U is the unknown global displacement vector. and the problem complexity.
local Te k e Le U (17) where Fe. an additional overestimation in the values of forces is obtained due to the dependency between U and ke. 4th International Workshop on Reliable Engineering Computing (REC 2010) 135 . U. The current formulation is attempting to reduce overestimation due to coupling in the FEM assembling process. multiple occurrences of interval quantities. Usually interval element forces can be obtained as: Fe k e Le U (16) where Fe. The interval solution of Eq. and solving the final system of interval linear equations. and element Boolean matrix.Primary and Derived Variables with the Same Accuracy in Interval Finite Elements where K. transformation. global interval element stiffness matrix. and interval global load vector. 2 Element (m) 1 2 2 2 (a) PY Node (n) F2m. In addition this formulation will introduce the derived quantities such as forces and stresses as dependent variables which will be obtained along with displacements when the system is solved. u1m 1 2 Figure 2. respectively. and P are the interval global stiffness matrix. respectively. a comprehensive discussion can be found in (Muhanna and Mullen 2001). local and Te are the local vector of interval element forces and the corresponding transformation matrix. u2m 2 1 uY 2 1 Free node (n) (b) PY uX Element (m) 2 1 F1m. Le are global interval vector of element forces. internal forces and stresses are quantities of practical interest in design. interval global displacement vector. In addition. (15) results in a significant overestimation in the system response. A typical node of a truss problem. The transformation procedure will provide an additional overestimation. element forces are pursued in local coordinate system that will require the transformation from the global coordinates to the local ones in the form: Fe. (a) Conventional formulation. (b) Present formulation. ke . Once again. respectively. Frequently.
and the local forces F1m and F2m . L. For example. By doing that. after deriving the local elements’ stiffness matrices along with the local elements’ load vectors the system will be transformed to the global system and assembled based on compatibility requirements resulting in the equilibrium system given by Eq. element m has the end nodes 1 and 2. (14). Mullen.M. The typical node is called a free node and is given along with all pertinent variables in the global coordinate system. 3. and R. Displacements are u X and uY and applied forces are PX and P . Considering a typical node of the truss system Figure 2 (a). V. elements and nodes are disassembled as in Figure 2 (b). Rama Rao. This procedure will result in the following system of equations: k C CT 0 U P 0 (18) where k is an interval matrix consists of the individual elements’ local stiffness and zeros at the bottom corresponding the free nodes’ degrees of freedom and have the following structure: k1 k1 0 0 k 0 0 0 0 0 0 and k1 k1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 kn kn 0 0 0 0 0 0 0 0 kn kn 0 0 0 0 0 0 0 0 0 0 01 X 0 0 0 0 0 0 0 0 0 0 01Y 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0mX 0 0 0 0 0 0 0 0 0 0 0mY (19) 136 4th International Workshop on Reliable Engineering Computing (REC 2010) . the local displacements u1m and u2m . L. each element is treated as having independent degrees of freedom in its own local coordinate system. Y The free node displacements are considered as independent of those of the elements. Muhanna In the conventional formulation of FEM Figure 2 (a). The system will be assembled imposing the discrete constraints Cmi to ensure the equality between the free node displacements and those of the elements. All coinciding elements at the free node along with pertinent variables are given in local coordinate system. 2. R. Where i is the number of constraints per element. In the present formulation the following steps are followed: 1.
The entries of the vector are given in Eq. Ai. 1 0 0 C T 0 1 0 0 1 1 u11 u 21 u1n (22) 11 21 0 cos sin 0 0 0 0 cos sin U u 2n u1 X u1Y (23) 1n 2n (24) 1 1 u mX u mY Vector P is the interval load vector and has the dimension equal to the sum of elements degrees of freedom and the free nodes degrees of freedom. and the length of each element.Vector has the dimension of total number of elements’ local degrees of freedom and given in Eq.Primary and Derived Variables with the Same Accuracy in Interval Finite Elements ki Ei A i Li (20) where Ei. The entries of the vector are the interval local displacements of elements followed by interval global displacements of the free nodes as shown in Eq. (25) The accuracy of the system solution depends mainly on the structure of Eq. ujX and ujY are the X and Y global displacements of jth free node coinciding with the 1st node of the ith element. and l = total number of the system’s degrees of freedom. the interval crosssectional area. (25). The entries of the matrix are equality constraints of the following type u1i u jX cos i u jY sin i 0 (21) Where u1i is the local displacement of the node 1 that belongs to ith element. The entries of the vector are the interval Lagrange multipliers that represent minus the local element forces in this case. (22) U is a vector of size s 1 where s is the number of elements’ local degrees of freedom + number of free nodes’ global degrees of freedom. (18) and on the nature of the used solver. Elements of CT are shown in Eq. and Li are the interval modulus of elasticity. where k = number of elements’ degrees of freedom (2 number of elements in the truss bar element case). respectively. (24). Matrix C has the dimensions (k l). (23). The solution of the interval system (18) provides the enclosures of the values of dependent PT 0 0 0 0 P1 X P1Y PmX PmY 4th International Workshop on Reliable Engineering Computing (REC 2010) 137 .
However for the derived quantities such as forces he suggested an improved enclosure by intersecting the simple enclosure z = Z(u) with the following enclosure: z Z (CF mid b) (SCF )(b mid b) (SCB)d (28) which results in additional significant overestimation The current formulation allows obtaining the interval displacement U and the accompanied interval derived quantities with the same accuracy. 138 4th International Workshop on Reliable Engineering Computing (REC 2010) . 2. (2007). Rama Rao. A number of examples are introduced in the following sections that illustrate the excellent accuracy of the developed method. L. The solution is obtained by performing the following iterative scheme: v { ACa ) ( ACF )b ( ACB )d} v. L. Mullen. and then the following enclosure is obtained: u (Ca) (CF )b (CB)d Where (28) C : (K u v d BD0 A) 1 Ca CFb CBd ACa ACFb ACBd ( D0 D) v (29) In Neumaier’s work only excellent enclosures of the interval displacements are obtained. Muhanna variables which are the interval displacements U and interval element forces discussed in the next section. The current formulation results in the interval linear system of equations given in (18) which can be introduced in the same structure of the following system: (K B D A)u a F b (26) with interval quantities in D and b only. d {( D0 D) v d (27) until some stopping criteria. ITERATIVE ENCLOSURES .M. and R. R. An iterative solver is The best known method for obtaining very sharp enclosures of interval linear system of equations that have the structure introduced in Eq. however It is expected that the enclosures are also good in case D is block diagonal with diagonal blocks small compared to the matrix size. V. The quality of the enclosures is known to be superior only in the special case where D is diagonal.5. (26) is the iterative method developed in the work of Neumaier and Pownuk.
The results of this example allow us to investigate the effect of load and stiffness uncertainty on the displacements and forces using various approaches presented. This truss structure is a benchmark problem adopted from the website of the Center for Reliable Computing (http://www. 15 kN Figure 3. The objective of choosing this example is to verify the ability of Neumaier’s approach and present approach to capture this phenomenon. The third example problem is a fifteen bar truss as shown in Figure 11 (Zhang.gtsav. 12 percent and 25 percent ( 0. Tables 1.11.01m2. The deterministic value of Young’s modulus of each element is Ei=2 1011N/m2. Example Problems Three example problems are chosen to illustrate the present approach and also to demonstrate its ability to obtain sharp bounds to the displacements and forces even in the presence of large uncertainties and large number of interval variables.5%. Neumaier’s approach and the present approach. The displacements and forces of this statically indeterminate truss are dependent on the uncertainty present in the modulus of elasticity and load. i=1. The solution is computed using four approaches viz. The objective of choosing this example is to demonstrate the applicability. This truss is internally indeterminate but externally determinate.Primary and Derived Variables with the Same Accuracy in Interval Finite Elements 3. 2 and 3 show the computed values of selected displacements (vertical displacement V2 at node 2.…. Krawczyk Fixed Point Iteration (FPI)this approach uses the same system structure as given by Neumaier and a fixed point iterative solver.14 and 15 are independent of structural stiffness although structural stiffness is uncertain while the axial forces in the remaining elements are dependent on structural stiffness. computational efficiency and scalability of the present approach for structures with complex configuration with a large number of interval parameters. 2005).html). Eleven bar truss. applied at the middle lower joint. a combinatorial approach. 2001) as shown in Figure 3.edu/rec/benchmarks. while the cross sectional area is 0. The modulus of elasticity of each element is assumed to vary independently. Considering the first example problem.5% from the mean value of 4th International Workshop on Reliable Engineering Computing (REC 2010) 139 .2. A cantilever truss as shown in Figure 8 is chosen as second example.2. the eleven bar truss is subjected to a concentrated load of 15 kN. Thus.gatech. the support reactions as well as axial forces in elements 1. 6% and 12. horizontal displacement U4 and vertical displacement V4 at node 4) using four approaches mentioned above for uncertainties of 1 percent. The first example chosen is an eleven bar truss (Muhanna and Mullen.
the error in width of vertical displacement at node 2 (V2) varies from 0. L. A similar behaviour is observed in the plot for variation of width of axial force N3 in element 3 in Figure 5.930764 13.732849 0.748841 0.650879 0.874039 0.90 1. L.874039 0.003347 3.732732 0.81 2.983562 4.874946 15. Krawczyk FPI fails to provide any solution (no enclosure is reached).24 U4 105 Lower Upper 1.93 0.71 15.220473 23.81 3. Muhanna E. respectively. The figure depicts the variation of the width of the present approach and the combinatorial approach with the variation of modulus of Elasticity (E) from its mean value.81 2.999561 0.99 Table 3 Eleven bar truss displacements for 25% uncertainty in the modulus of elasticity (E) V2 105 Combinatorial approach Neumaier’s approach Error %(width) Present approach Error %(width) Lower Upper 17.490376 3.24 17.732731 0.45 0. and R.494396 10. The error in width is computed as 100 width of computed solution 1 width of combinatorial solution .02 at 12% uncertainty.848475 0.220473 23.003347 Upper 3.45 Figure 4 shows the computed interval values of vertical displacement V2 at node 2. However the present approach and Neumaier’s approach still provide solution with reasonable sharpness for this level of uncertainty.103133 15.024443 14.748841 0.90 V4 105 Lower Upper 0.748841 0.288285 17.93 Table 2 Eleven bar truss displacements for 12% uncertainty in the modulus of elasticity (E) V2 105 Combinatorial approach Krawczyk FPI Neumaier’s approach Error %(width) Present approach Error %(width) Lower Upper 15.431895 3.705818 4.999561 0.50 2. For a comparison. Table 1 Eleven bar truss .922944 0.084938 13.874039 0.903532 14.482011 27.732732 0.M.024602 14.930764 13.93 0. Mullen.666701 20.71 at 1% uncertainty to 9.922944 0.482011 27.705818 4.922943 0.024603 14. It is observed from this figure that the present solution encloses the combinatorial solution at all values of variation from 0 percent to 25 percent.967877 9. while the error in width is 20.748810 0. It is observed from Table 1 that the displacements obtained using the present approach at 1 percent uncertainty provides sharp enclosure to the displacements obtained using the combinatorial approach and also agree very well with the displacements obtained using Neumaier’s approach and Krawczyk FPI .displacements for 1% uncertainty in the modulus of elasticity (E) V2 105 Combinatorial approach Krawczyk FPI Error %(width) Neumaier’s approach Error %(width) Present approach Error %(width) Lower Upper 15.451843 2.99 0.50 V4 105 Lower Upper 0.024603 14.231940 12.923326 2.003075 3.24 at 25% uncertainty.633096 11.02 15.003347 V4 105 Lower Upper 0.565208 0.013033 1.02 U4 105 Lower Upper 2.971261 0. Thus it is observed that both Neumaier’s approach and the present approach provide guaranteed bounds on the combinatorial approach. V.967877 9.431895 3.848475 0. Rama Rao.231940 12. 140 4th International Workshop on Reliable Engineering Computing (REC 2010) . R.71 U4 105 Lower 2.843182 0.71 15.666701 20.4943960 10.633096 11. It is observed from Table 2 that for a large uncertainty of 12% (and beyond).
solution vs. Comb. Axial forces are computed for the eleven bar truss using the following approaches a) Simple enclosure z1(u) b) z2(u).8 5.3 Axial force N3 (kN) 5. present solution. Eleven bar truss . solution vs. present solution.5E04 1.Primary and Derived Variables with the Same Accuracy in Interval Finite Elements 1. It is clearly observed from this table that the present method provides very sharp enclosure to the forces in comparison 4th International Workshop on Reliable Engineering Computing (REC 2010) 141 . 4. Comb.3E04 Vertical displacement V2 (m) 1.8E04 0% 5% 10% 15% 20% Percentage variation of E about the mean 25% V2 comb V2 present Figure 4.8 7.4E04 1.8 N3 comb 6.3 N3 present 6.3 0% 5% 10% 15% 20% 25% Percentage variation of E about the mean Figure 5. Eleven bar truss – variation of axial force in element 3 with uncertainty of E. the intersection of z1(u) with the enclosure obtained using Eq. (28) of Neumaier.7E04 1.variation of vertical displacement at node 2 with uncertainty of E. c) Present approach The interval values of axial forces in elements 3 and 9 (N3 and N9) are presented in Table 4.6E04 1.
861027 5.96 3.32576 186.5 percent uncertainty of both load and modulus of elasticity (E).28858 5.99198 6.M.31656 5.72 5.01836 V4 105 Lower 0. Figure 7 shows the variation of axial force in element 9 with the variation of uncertainty of Young’s modulus and load. even at an uncertainty as large as 10%.89043 3.72918 0.09972 0.90792 0.58 10.08732 141.72898 Upper 0.96 2. The eleven bar truss mentioned above is analysed once again at various levels of uncertainty of Young’s modulus and load and the results are tabulated.63 10. R.58105 8.83 6. It is observed from these figures that the present solution encloses the combinatorial solution at all levels of uncertainty.89702 337.99224 0.97 6.85 Table 5 Eleven bar truss displacements for 1% uncertainty in the modulus of elasticity (E) and load V2 105 Combinatorial approach Neumaier’s approach Error %(width) Present approach Error %(width) Lower 15.98 0. Further.85 N 9 ( kN ) 9. Muhanna with both the enclosures suggested by Neumaier.23567 10.70837 6.comparison of axial forces for 10% uncertainty in the modulus of elasticity (E) for various approaches N 3 ( kN ) Combinatorial approach Simple enclosure z1(u) Error %(width) Intersection z2(u) Error %(width) Present approach Error %(width) N 3 ( kN ) N 9 ( kN ) 10. Rama Rao.53 5.09972 0. Table 4 Eleven bar truss .80057 14.60 15.75258 0.75258 0.72898 Table 6 Eleven bar truss .05338 Upper 10.95 10.75255 0.90870 2.96214 447.82238 5.81438 N9 (kN) Lower 10.23604 0. Table 6 presents the corresponding values of axial forces in elements 3 and 9.05406 9.90792 0.57152 7.39240 9. Figure 6 shows the variation of vertical displacement at node 4 with the variation of uncertainty of Young’s modulus and load. Mullen. L.79891 U4 105 Lower 2.03902 71. This illustrates the ability of the present approach to obtain sharp bounds to displacements and forces even at larger values of uncertainty.79891 Upper 14.02784 9. and R.86033 Upper 5.98 0.73966 8.53601 8. It is observed that Krawczyk FPI fails to provide an enclosure at 11.09956 15.30685 77.60 14.Axial forces for 1% uncertainty in the the modulus of elasticity (E) and load N3 (kN) Combinatorial approach Intersection z2(u) Error %(width) Present approach Error %(width) Lower 5.01836 Upper 3.15 11. Table 5 presents the computed values of selected displacements for the second case study with uncertainty of Young’s modulus and load being 1 percent. It is observed from these tables that the present solution gives very sharp enclosure to the values of displacements and forces.01809 3. V. the results of displacements obtained using the present approach agree very well with the results obtained using Neumaier’s approach. L.98368 142 4th International Workshop on Reliable Engineering Computing (REC 2010) .54135 11.
Comb. The numerical values adopted to analyse the truss are P= 1000 N. Eleven bar truss – Variation of axial force in element 9 with uncertainty of load and E. This structure has 101 elements with interval modulus of elasticity (E). 7 8 Axial Force N9 (kN) 9 10 11 12 13 0% 5% 10% 15% 20% 25% Percentage variation of E and load about the mean N9 Comb N9 Present Figure 7.1E05 1.0E06 4. L = 1 m.0E06 1. solution vs present solution. and E = 2 1011N/m2. Eleven bar truss – Variation of vertical displacement at node 4 with uncertainty of load and E.5% from the mean value of E).2E05 0% 5% 10% 15% 20% 25% Percentage variation of E and load about mean V4 comb V4 Present Figure 6. present solution. A = 0. Maximum uncertainty allowed in modulus of Elasticity (E) is 5 percent ( 2. The cantilever truss structure as described earlier is shown in Figure 8.0E06 9.0E05 1.0E06 8.0E06 7. solution vs.01 m2.0E06 Vertical displacement V4 6. Four different solutions are presented for the problem using the following approaches 4th International Workshop on Reliable Engineering Computing (REC 2010) 143 .Primary and Derived Variables with the Same Accuracy in Interval Finite Elements 3.0E06 5. Comb.
30 368.0 0. Values of C xD and C yD are computed for all the approaches PL and are presented in Tables 3 and 4.53 25.65 Table. V.nondimensional width Uncertainty (%) EBE approach Neumaier’s approach Present approach Sensitivity analysis 0. L. L.19 728.71 555.M.89 43. Zhang and Mullen.0 535.0 206. Benchmark problem – Cantilever truss.03 16. Table.34 368.70 3.10 CxD of displacement 3.0 218.0 1.8 Cantilever truss .40 8.0 0. it is to be noted that sensitivity analysis provides an inner bound solution to the displacements.26 33.35 C yD of displacement VD 5.10 2.50 183. and R.92 16. 2007) Pownuk’s sensitivity analysis (Pownuk.41 8. R.2 745.20 4.0 0.0 0.41 144 4th International Workshop on Reliable Engineering Computing (REC 2010) .0 0.0 1002.nondimensional width Uncertainty (%) EBE approach Neumaier’s approach Present approach Sensitivity analysis 0.16 183.0 0.0 1700.0 0. Further.84 34.95 911.14 43.0 1. Rama Rao.4 UD 5.0 2747.84 25. The results for the horizontal and vertical displacement at the right upper corner (node D) of the truss are computed. These values are used for the computation of nondimensional constants C xD and C yD defined as C xD UD AE PL and C yD VD AE . It is further observed that the EBE method gives overestimated results compared to the other three methods.67 546.0 46. 2004) and Present approach. Figure 8.80 937. Muhanna Elementbyelement (EBE) method (Muhanna. 7 Cantilever truss.0 14. It is observed from Tables 7 and 8 that widths of enclosures obtained using the present approach agree very well with the Neumaier’s approach as well as sensitivity analysis.11 41.54 25.91 16.3 745.0 104.20 936.7 555. The variation of the width of the enclosure with uncertainty of Young’s modulus for displacements UD and VD can be observed from the Figures 9 and 10.33 364. 2007) Neumaier’s method (Neumaier and Pownuk.48 8.28 34.06 4. Mullen.0 376.12 182.0 0.78 2.
t.Primary and Derived Variables with the Same Accuracy in Interval Finite Elements 3000 Nondimensional width of solution C 2500 2000 1500 1000 500 0 0% 1% 2% 3% 4% 5% Percentage uncertainty Figure 9.r.r.variation of nondimensional widths of inte val solution of EBE approach Neumaier's approach Present approach Sensitivity approach U D w. uncertainty of E. Variation of nondimensional widths of interval solution of EBE approach Neumaier's approach Present approach Sensitivity approach VD w. 400 350 Nondimensional width of interval solution C 300 250 200 150 100 50 0 0% 1% 2% 3% 4% 5% Percentage uncertainty Figure 10.t. 4th International Workshop on Reliable Engineering Computing (REC 2010) 145 . Cantilever truss . uncertainty of E.
Mullen.0 104 m2 while for the rest of the elements is the cross sectional area is 6.00 Upper 251.00 N14 (kN) Lower 222.500 0.00 Upper 222. it is observed from Table 9 that forces in these elements obtained using Neumaier’s approach have interval values. It is further observed that the solution obtained by present approach matches exactly with the combinatorial solution while Neumaier’s approach gives an overestimated solution.500 245. The modulus of elasticity of each element is assumed to vary independently.920 10.P3=100 kN and a horizontal point load P4=90 kN applied at the joints 5. L.250 10.662 347.022 0.0 104 m2. L.3.500 295.022 277. Rama Rao. and R.52 314.022 225.00 Upper 314.00 N2 (kN) Lower 251.M.500 200.660 10. The fifteen element truss shown in Figure 11 is subjected to vertical point loads of P1=200 kN.14 and 15 are 10. Fifteen bar truss.00 146 4th International Workshop on Reliable Engineering Computing (REC 2010) .200 10.00 Upper 267. i =1.780 10. V.500 0.2.53 251. Error in bounds is computed because error in width can not be computed owing to zero width of combinatorial solution. 14 and 15 of fifteen element truss for 10% uncertainty in the modulus of Elasticity N1 (kN) Combinatorial approach Neumaier’s approach Error %(bounds) Present approach Error %(bounds) Lower 267.662 283. Muhanna Figure 11.01 m2. Thus Neumaier’s approach fails to capture the deterministic nature of axial forces in these elements. Neumaier’s approach and present approach. It is observed that the forces in these elements computed using the present approach as well as combinatorial solution are thin intervals and match exactly with each other. 2. Case B: 10 percent uncertainty in both modulus of Elasticity (E) and loads Table 9 presents the axial forces in elements 1. P2=100 kN. This is because the axial forces in these elements are independent of structural stiffness. Case A: 10 percent uncertainty in modulus of Elasticity (E) while loads are deterministic.500 0.00 314.00 251.00 267. while the cross sectional area is 0.53 267.00 N15 (kN) Lower 314.662 0. The following two case studies are taken up to demonstrate the effectiveness of the present approach.450 10. Table 10 shows the corresponding axial forces in the elements 3.662 0.750 10. Table 9 Forces (kN) in elements 1.2.13.14 and 15 for case A.500 0. Results are computed using combinatorial approach.53 222. The deterministic value of Young’s modulus of each element is Ei=2 1011N/m2. 4 and 5.022 0.500 240. 6 and 3 respectively.92 10.…15. 2 . Cross section areas of elements 1.2. However.00 222. R.
289 294.37344 5.929 365.221 126.835 210.40 1.043 21.987 7.929 0.194 379.585 5 43.929 114.985 27.87429 V5 102 Lower 6.17 Table 11 depicts the horizontal and vertical displacements U5 and V5 of node 5 for case B.390 317.17 N4 (kN) Lower Upper 329.526 41. It is observed from Table 11 that.40 1.395 298.875 0.523 10. 4 and 5 computed using the present approach provide a sharp enclosure to combinatorial solution while Neumaier’s approach provides overestimated bounds to combinatorial solution.000 4.48 44.57150 6.125 280.875 227.275 48.975 14.125 280.51430 1.543 10.22 6. Conclusions A new formulation for Interval Finite Element Methods is introduced.2.375 233.56423 6.797 4 346.854 16.099 361.4 and 5 of fifteen element truss for 10% uncertainty in the modulus of Elasticity (E) N3 (kN) Combinatorial approach Neumaier’s approach Error %(width) Present approach Error %(width) Lower Upper 115.609 127. Table 11 Fifteen element truss –displacements for 10% uncertainty i the modulus of elasticity (E) and load U5 102 Combinatorial approach Neumaier’s approach Error %(width) Present approach Error %(width) Lower 1.30663 Upper 5.375 233.440 210.87214 1.174 101. Table 12 Forces (kN) in elements of fifteen element truss for 10% uncertainty in modulus of elasticity (E) and load Element Combinatorial approach Neumaier’s approach Present approach %Error %Error in width in width LB UB LB UB LB UB 1 254.12 347.01 266.000 2 266.885 330.97 107.910 38.661 10.900 103. both primary and derived quantities of interest are included in the original uncertain system as primary 4th International Workshop on Reliable Engineering Computing (REC 2010) 147 .57150 6.267 302.756 235. the displacements obtained using the present approach give excellent bounds to combinatorial solution and also compare very well with the results of Neumaier’s approach.461 142.49353 6.87429 Upper 1.476 10.187 169.17 N5 (kN) Lower Upper 38.909 4.395 298.143 12.174 267.379 139.257 95.000 3 108.344 14 211.14 and 15 computed using Neumaier’s approach are quite large in while the corresponding errors in widths are zero in the case of forces obtained using present approach. It is observed from Table 12 that the widths of the axial forces in elements 1.920 148. Also.014 288.53 254. even for a large uncertainty of 10 percent.098 134.000 15 330.22 5.858 38.49353 6.217 210.003 300.625 0.858 20. the axial forces in elements 3. In this approach.289 0.375 310.390 330.594 18.53 211.125 258.778 318.463 210.53 330.385 134.756 235.30663 Table 12 presents the axial forces in selected elements for case B.609 114.167 272. Thus it is concluded that the present approach provides very sharp enclosures to the axial forces obtained using combinatorial solution while Neumaier’s solution provides overestimated bounds.Primary and Derived Variables with the Same Accuracy in Interval Finite Elements Table 10 Forces (kN) in elements 3.625 189.
Nondeterministic Linear Static Finite Element Analysis: An Interval Approach. School of Civil and Environmental Engineering. Methods in Engrg. 13(2): 149172. MA. Journal of Engineering Mechanics. Zhang. Finite Element Interval Estimation by Convex Model.1157. and R. L. L. Zienkiewicz. Cakmak. The formulation is based on the variational approach and Lagrange multiplier method involving imposition of certain constraints that allow the Lagrange multipliers themselves to be the derived quantities.C. R. Oxford.. R. New Jersey. http://webapp. The present approach captures exactly the behaviour of statically determinate structures where the internal forces in elements are independent of material properties while all previous methods do not. Ph. Li. AIAA J. Rao. pp. Pownuk. WPI. Numerical results of this new formulation are illustrated in a number of example problems. S. H. 2007. and Yoshikawa. S. John Wiley & Sons.D. 1996. 2002. However. R. P. Rao.O. S. Finite Element Analysis Fundamentals.L The Finite Element Method.. Analysis of uncertain structural systems using interval analysis. Muhanna variables. Köylüoglu. Nakagiri. K.. Inc. New Jersey. Energy principles and variational methods in applied mechanics. H. Muhanna.edu:8080/~andrzej/php/ansys2interval/ Rao. Reliable Computing. and Sawyer. 2004. Mullen. Printice Hall. Development of Interval Based Methods for Fuzziness in Continuum Mechanics. Rama Rao. A. S.. It is observed that the displacements obtained by present approach provide a sharp enclosure to combinatorial solution and agree very well with the results obtained by Neumaier at all uncertainties. Printice Hall. Numerical solution of fuzzy linear equations in engineering analysis. UK. Linear Systems with Large Uncertainties. Int.R. K. 557–566. 1998. and Pownuk. The present method addresses the basic issue of eliminating the additional overestimation in the derived quantities by adopting the mixed formulation that makes possible the simultaneous computation of primary and derived variables at the same level of accuracy. Georgia Institute of Technology. L. L. AIAA J. 33(12): 2364–2370. Neumaier. Mech. S. A. 278281.utep. A. U. J. 127 (6). 2001. 1995. 2005. Englewood Cliffs. 121(11): 1149. August 79. 1995.43: 391–408. 148 4th International Workshop on Reliable Engineering Computing (REC 2010) . 1996. L. Fuzzy finite element approach for analysis of imprecisely defined systems. 35(4): 727–735. and Mullen.math. Ahmet. Numer. S. 2000. 2nd edition. Gallagher. 1995. Muhanna. Dissertation. R. R.. R. Further. In Proceedings of 7th ASCE EMD/STD Joint Specialty Conference on Probabilistic Mechanics and Structural Reliability. N. J. and Mullen. Finite Element Procedures. and Chen. J. and Berke. Interval Algebra to Deal with Pattern Loading and Structural Uncertainty. with Applications to Truss Structures. The present approach can find further application in the area of nonlinear problems of structural mechanics involving large uncertainties of structural parameters. References Bathe.. L. S. 1995. S. Krawczyk’s Fixed Point Iteration fails to provide any enclosure to the solution at large uncertainties. V.: Calculation of the Extreme Values of Displacements in Truss Structures with Interval Parameters. the forces computed using the present approach provide sharp enclosure to the combinatorial solution while forces computed using the approach suggested by Neumaier are found to yield significantly overestimated results.: Uncertainty in Mechanics Problems—IntervalBased Approach. R.M. Butterworth Heinemann. and Soren. Reddy. Englewood Cliffs. N. N. L. Engrg. In Proceedings of ISUMANAFIPS’95: 145150. September 1720. and Taylor. 1997.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.