See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228942419

Buckling analysis of structures with uncertain properties and loads using an

interval finite element method

Article · January 2004

CITATIONS READS

8 440

4 authors, including:

Robert Mullen Rafi Muhanna

University of South Carolina Georgia Institute of Technology
170 PUBLICATIONS   3,760 CITATIONS    82 PUBLICATIONS   1,316 CITATIONS   

SEE PROFILE SEE PROFILE

Hao Zhang
The University of Sydney
70 PUBLICATIONS   1,340 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Reliability and full-range analysis of joints in steel frameworks View project

Aerospace Seals, Aeropropulsion, Lubrication View project

All content following this page was uploaded by Rafi Muhanna on 28 February 2014.

The user has requested enhancement of the downloaded file.

 1

Buckling Analysis of Structures with Uncertain Properties and Loads Using

an Interval Finite Element Method

Mehdi Modares
Department of Civil Engineering, Case Western Reserve University, Cleveland, Ohio 44106

Robert Mullen
Department of Civil Engineering, Case Western Reserve University, Cleveland, Ohio 44106
e-mail: [email protected]

Rafi L. Muhanna
Department of Civil Engineering, Georgia Institute of Technology, Savannah, Georgia 31407
e-mail: [email protected]

Hao Zhang
Department of Civil Engineering, Georgia Institute of Technology, Savannah, Georgia 31407
e-mail: [email protected]

Abstract. In order to ensure the safety of a structure, one must provide for adequate strength of
structural elements. In addition, one must prevent large unstable deformations such as buckling.
In most analyses of buckling, structural properties and applied loads are considered certain. This
approach ignores the fact that imperfections and unknown changes in properties, albeit small, are
required for onset of buckling. In this paper, we extend the interval finite element methods
developed by the authors to solve for the possible values of loads that will result in a structural
stability failure. The analysis requires that interval axial element forces in each frame element in
a structure be calculated. These values are calculated from a linear system of interval equations
resulting from the static structural analysis. Using the calculated axial loads, a subsequent
interval eigenvalue problem is solved for the buckling loads. For both solutions of the linear
system of equations and the eigenvalue problem, the unique properties of the finite element
methods result in sharp solutions. Several structural problems are presented as exemplars. The
sharpness of the solution is demonstrated by comparing to combinatorial solutions.

1. Introduction

In order to ensure the safety of a structure, one must provide for adequate strength of structural
elements. In addition, one must prevent large unstable deformations known as buckling. In
determining adequate strength as well as adequate stability, the finite element method has
become the standard of practice for predicting a structure’s behavior.
In current practice, uncertainty in system parameters is not considered during the analysis.
Uncertainty is accounted for in a design by a combination of load amplification and strength
reduction factor.
Such factors are based on probabilistic models of historic data. Thus, consideration of the
impact of changing uncertainty on a design has been removed from current analysis tools.

REC2004
2

In order to mitigate this problem, the authors among others (REF IFEM) have developed an
interval based finite element method (IFEM). IFEM allows a structural analyst to calculate the
impact of uncertainty in parameters on the structure’s predicted behavior. To our knowledge,
IFEM has only been applied to analysis addressing the strength criterion. In this paper, we
extend IFEM to linear stability analysis of structures.

The method presented in this paper requires that interval internal element axial forces in each
element in a structure be calculated. These values are calculated from a linear system of interval
equations resulting from the static structural analysis. Using the calculated internal forces, a
subsequent interval eigenvalue problem is formulated. The solution of the interval eigenvalue
problem is then used to calculate the bounds on the critical buckling load. For both the solution
of the linear system of equations and the eigenvalue problem, the unique properties of the finite
element method are employed to achieve sharp results.

In the following, a brief review of IFEM for calculation of internal element forces is presented.
Section 3 describes the formulation of the interval linear stability problem. In section 4, a
method for calculation of exact bounds on the resulting eigenvalue problem is then given. An
example problem is presented in section 5. Observations and conclusions are given in section 6.

2. Review of Static Interval Finite Element Methods

The linear stability analysis of structures requires the element forces to be determined as the first
step in the analysis. For problems with interval values for the stiffness or loads, one needs an
interval solution to the underlying statics problem. For the solution of interval finite element
(IFEM) problems, Muhanna and Mullen (2001) introduced an Element-by-Element interval finite
element formulation, in which a guaranteed enclosure for the solution of interval linear systems
of equations was achieved. This method accounts for the parametric representation of element
properties and a very sharp enclosure for the solution set due to loading, material and geometric
uncertainty in solid mechanics problems. Element matrices were formulated based on the physics,
and Lagrange multiplier or penalty methods were applied to impose the necessary constraints for
compatibility and equilibrium.

For example, a two-element finite element construct is shown in figure (1). In this example, (E)
is Young’s modulus, (A) the cross-sectional area, and (L) the length of each element. Subscripts
here indicate element number. Nodal loads are denoted by (P), and nodal displacements are
denoted by (u).

The conventional finite element formulation results in a global stiffness matrix as given in Eq. (1)

1 2 3
P1 P2 P3

L1, E1, L2, E2,

Figure (1) : Two connected linear truss elements.

REC2004
 3

 E1 A1 E1 A1 
 − 0 
 L1 L1  u1   P1 
 E1 A1 E1 A1 E A E2 A2     (1)
− + 2 2 −  u2  =  P2 
 L1 L1 1 L2 L2    
 0 E2 A2 E2 A2  u3   P3 
 −
 L2 L2 

When the parameters E, A, or L are interval quantities, the resulting interval matrix allows for
independent interval values for elements of this matrix which is not physically possible. The
element-by-element method generates a global stiffness matrix in the form shown in Eq. (2).

 E1 A1 E1 A1 
 − 0 0 
 L1 L1 
 χ1 0 0 0  E1 A1 E1 A1 
  − 0 0 
 0 χ1 0 0  L1 L1 (2)
K = DS =   
0 0 χ2 0  E 2 A2 E A 
  0 0 − 2 2
 0 0 0 χ 2  L2 L2
 
 0 E 2 A2 E 2 A2 
 0 −
 L2 L 2 

where χi is the interval multiplier of the ith finite element obtained due to uncertainty in Ei, Ai,
and li. Such a form (i.e.DS) allows for factoring out the interval multiplier, resulting with an
exact inverse for (DS).

To ensure compatibility (unique displacements for all elements connecting to a node), one adds
constraint conditions in the form of Eq. (3). The resulting system of linear interval equations
becomes Eq. (3) and (4)

Equation (1) can be introduced in the following equivalent form:

~
CU = 0 (3)

~
KU + C T λ = P (4)
~
If we express K ( n × n) in the form DS and substitute in equation (4):

~ ~
DS U = P − C T λ (5)

where D (n × n) is interval diagonal matrix, where its diagonal entries are the positive interval
multipliers associated with each element, and n is the multiplication of degrees of freedom per
~
element and the number of elements in the structure. S (n × n) is a deterministic singular matrix
~
(fixed point matrix). If we multiply equation (3) by DC T and add the result to equation (5), we
get:

~ ~ ~ ~
D ( S U + C T CU ) = ( P − C T λ ) (6)

REC2004
4

or:
~ ~ ~
D ( S U + QU ) = ( P − C T λ )
~ ~ ~
D ( S + Q )U = ( P − C T λ )
~ ~
DR U = ( P − C T λ ) (7)

~
where (R ) is a deterministic positive definite matrix, and the displacement vector U can be
obtained from equation (7) in the following form:

~ ~
U = R −1 D −1 ( P − C T λ ) (8)

~ ~
where ( R −1 D −1 ) is an exact inverse of the interval matrix ( DR ) . Equation (8) can be presented
in the form:
~
U = R −1 M δ (9)

Matrix (M) has the dimensions (n × number of elements), and its derivation has been discussed in
the previous works of the authors (Mullen and Muhanna 1999, Muhanna and Mullen 1999). The
vector δ is an interval vector that has the dimension of (number of elements × 1) and its elements
are the diagonal entries of D −1 with the difference that every interval value associated with an
element is occurring only once.
If the interval vector λ can be determined exactly, the solution of Eq. (8) will represent an
exact hull for the solution set of the general interval FE equilibrium equation.

More details on optimal implementation of the above concepts for static finite element solutions
is presented in another paper in this proceedings. (Muhanna, Mullen and Zhang 2004).

3. Problem Definition

Deterministic Buckling Analysis:

As discussed in the previous section, the buckling analysis using the linear finite element method
is carried out in two main steps. First, a parametric static analysis is performed using an arbitrary
ordinate of applied load.

[ Ke]{u} = {P} (10)

The solution output includes the internal axial forces in terms of the load ordinate. Using these
results, the geometric stiffness of the structure is developed which represents the pre-compression
load’s effects on the total stiffness of the structure (McGuire, Gallagher and Ziemian 2000).

Second, a generalized eigenvalue problem is performed between the elastic and geometric
stiffness matrices of the structure in order to find the critical buckling loads in terms of the
geometric and material characteristics of the structure.

([ Ke] − λ [ Kg ]){u} = {0} (11)

REC2004
 5

Buckling Analysis for Structures with Bounded Uncertainty:

For structures with bounded uncertainty present in the stiffness characteristics, the buckling
analysis procedure requires the modifications on the following: First, the representation of
stiffness characteristics must consider the presence of uncertainty using interval numbers. Second,
the static analysis must be performed using the obtained interval stiffness matrix; hence, the
calculated element axial forces are interval values. Third, using the obtained element interval
axial forces, the interval geometric stiffness matrix can be established. Fourth, the interval
eigenvalue problem must be solved in order to obtain the bounds on the critical buckling loads.

I. Interval Representation of Uncertainty

Interval Number:
A real interval is a closed set of the form:
~
Z = [ z l , z u ] = {z ∈ ℜ | z l ≤ z ≤ z u } (12)

In this work, the symbol (~) represents an interval quantity.

Interval Formulation:
The structure’s global stiffness can be viewed as a summation of the element contributions to the
global stiffness matrix:

n
[ Ke] = ∑ [ Li ][ Kei ][ Li ]T (13)
i =1

where [ Li ] is the element Boolean connectivity matrix and [ K i ] is the element stiffness matrix
in the global coordinate system. Considering the presence of uncertainty in the stiffness
properties, the non-deterministic element stiffness matrix is expressed as:
~
[ Kei ] = ([li , ui ])[ Kei ] (14)

in which [li , u i ] is an interval number that pre-multiplies the deterministic element stiffness
matrix. Therefore, the structure’s global stiffness matrix in the presence of any uncertainty is the
linear summation of the contributions of non-deterministic interval element stiffness matrices:

n n
~
[ Ke] = ∑ ([li , ui ])[ Li ][ Kei ][ Li ]T = ∑ ([li , ui ])[ K ei ] (15)
i =1 i =1

in which, [ K ei ] is the deterministic element stiffness contribution to the global stiffness matrix.

II. Interval Geometric Stiffness Matrix

Using the obtained interval axial forces by IFEM (Section 2), the interval geometric stiffness
matrix can be set up. The structure geometric stiffness can be viewed as a summation of the
element contributions to the global geometric stiffness matrix:

REC2004
6

n
[ Kg ] = ∑ [ Li ](( f i )[ Kˆ g i ])[ Li ]T (16)
i =1

where, ( f i ) is the element axial force and [ Kˆ gi ] is the force independent matrix of geometric
stiffness. Considering the axial force as an interval quantity, the interval structure’s geometric
stiffness matrix can be established as:

n n
~ ~ ~
[ Kg ] = ∑ ( f i )[ Li ][ Kg i ][ Li ]T = ∑ ( f i )[ K g i ] (17)
i =1 i =1

~
where [ K g i ] = [ Li ][ Kˆ g i ][ Li ]T and ( f i = [ f min i , f max i ] ) is the element interval axial load.

III. Interval Eigenvalue Problem for Buckling Analysis

Hollot and Barlett (1987) studied the spectra of eigenvalues of an interval matrix family which
are found to depend on the spectrum of its extreme sets. Dief (1991) presented a method for
computing interval eigenvalues of an interval matrix based on an assumption of invariance
properties of eigenvectors.

The concept of interval eigenvalue problem has been used in structures with interval uncertainty.
Modares and Mullen (2004) have introduced a method for the solution of the interval eigenvalue
problem which determines the exact bounds of the natural frequencies of a structure using IFEM
formulation.

In order to obtain the bounds on the critical buckling loads, a generalized interval eigenvalue
problem must be performed between the interval elastic and interval geometric stiffness matrices
as:

n
~ n ~
( ∑ ([li , ui ])[ K ei ]){u} = (λ )( ∑ ( f i )[ K g i ]){u} (18)
i =1 i =1

Interval Eigenvalue Problem Definition:

The eigenvalue problems for matrices containing interval values are known as the interval
~ ~
eigenvalue problems. If [ A] is an interval matrix ( A ∈ IR n×n ) and [ A] is a member of the
~
interval matrix ( A ∈ A) , the interval eigenvalue problem is shown as:

~
([ A] − λ[ I ]){u} = 0, ( A ∈ A) (19)

The solution of interest to the real interval eigenvalue problem is defined as an inclusive set of
~
real values (λ ) such that for any member of the interval matrix, the solution to its eigenvalue
problem is a member of the solution set shown as:
~ ~
{λ ∈ λ = [λl , λu ] | ∀A ∈ A : ([ A] − λ[ I ]){u} = 0} (20)

REC2004
 7

which is the enclosure of all possible solutions. A sharp enclosure is defined as the solution with
the smallest radius as:
~ ~
{∃λ ∈ λ = [λl , λu ] | ∀A ∈ A : ([ A] − λ [ I ]){u} = 0} (21)

4. Solution for Interval Eigenvalue Problem

The following concepts must be considered in order to bound the non-deterministic interval
eigenvalue problem, Eq.(18).

The classical linear eigenpair problem for a symmetric matrix is:

Ax = λx (22)

with the solution of real eigenvalues ( λ1 ≤ λ 2 ≤ ... ≤ λ n ) and corresponding eigenvectors

( x1 , x 2 ,..., x n ). This equation can be transformed into a ratio of quadratics known as the Rayleigh
quotient:

x T Ax
R( x) = (23)
xT x

The Rayleigh quotient for a symmetric matrix is bounded between the smallest and the largest
eigenvalues (Bellman 1960 and Strang 1976).

x T Ax
λ1 ≤ R( x) = ≤ λn (24)
xT x

Thus, the first eigenvalue ( λ1 ) can be obtained by performing an unconstrained minimization on

the scalar-valued function of Rayleigh quotient:

x T Ax
minn R ( x ) = minn ( ) = λ1 (25)
x∈R x∈R xT x

For finding the next eigenvalues, the concept of maximin characterization can be used. This
concept obtains the kth eigenvalue by imposing (k-1) constraints on the minimization of the
Rayleigh quotient:

λ k = max[min R( x)] (26)

(subject to constrains ( x z i = 0), i = 1,...k − 1, k ≥ 2 )
T

Bounding the Critical Buckling Loads:

REC2004
8

Using the concepts of minimum and maximin characterizations of eigenvalues for symmetric
matrices, the solution to the generalized interval eigenvalue problem for the critical buckling
loads of a structure with uncertainty in the stiffness characteristics (Eq.(18)) for the first
eigenvalue can be shown as:

n
~ x T ( ∑ ([li , ui ])[ K ei ]) x
~ x T [ Ke] x
λ1 = min( ~ ) = min ( n
i =1
) (27)
x∈R n x T [ Kg ]x x∈R n
x ( ∑ ([ f min i , f max i ])[ K g i ]) x
T

i =1

for the next eigenvalues:

n
~ x T ( ∑ ([li , ui ])[ K ei ]) x
~ x T [ Ke] x
λk = max[ min ~ ] = max[ x . z =min ( n
i =1
)] (28)
x . zi =0 ,i =1,...,k −1 x T [ Kg ] x i 0 ,i =1,...,k −1
x ( ∑ ([ f min i , f max i ])[ K g i ]) x
T

i =1

Bounding Deterministic Eigenvalue Problems for the Critical Buckling Loads:

Since the matrices [ K ei ] and [ K g i ] are non-negative definite, the terms ( x T ( K ei ) x )
and ( x T ( K g i ) x ) are non-negative. Therefore, the upper bounds on the eigenvalues in Eqs.(18)
and (19) are obtained by considering maximum values of interval coefficients of uncertainty for
all elements in the elastic stiffness matrix and the lower values of axial force in the geometric
stiffness matrix . Similarly, the lower bounds on the eigenvalues are obtained by considering
minimum values of interval coefficients of uncertainty for all elements in the elastic stiffness
matrix and the upper values of axial force in the geometric stiffness matrix.
Also, it can be observed that any other element stiffness selected from the interval sets will yield
eigenvalues between the upper and lower bounds. Using these concepts, the deterministic
eigenvalue problems corresponding to the maximum and minimum critical buckling loads are
obtained as:

n n
( ∑ (ui )[ K i ]){u} = (λmax )∑ ( f min i )[ K g i ]{u} (29)
i =1 i =1
n n
( ∑ (li )[ K i ]){u} = (λmin )∑ ( f max i )[ K g i ]{u} (30)
i =1 i =1

5. Example

The bounds on the critical buckling load for a 2D statically indeterminate truss with interval
uncertainty present in the modulus of elasticity of each element are determined (Figure (2)).

REC2004
 9

Figure (2): The structure of 2-D truss

The cross-sectional area A , the length for horizontal and vertical members L , the Young’s
~
moduli E for all elements are E = ([0.99,1.01]) E .

The problem is solved using the method presented in this work. First, a static analysis on the
structure with uncertainty is performed using IFEM, and the bounds on obtained element axial
forces are obtained. Second, two deterministic eigenvalue problems are performed to obtain the
bounds on the critical buckling load.

For comparison, a combinatorial analysis has performed which considers lower and upper values
of uncertainty for each element i.e. solving ( 2 n = 210 = 1024 ) deterministic problems.

The static analysis results obtained by IFEM and the brute force combination solution are
summarized in Table (1).

REC2004
10

Lower Upper Lower Bound Upper Bound

Bound Bound
Combination Combination
IFEM IFEM Method Method

f1 -0.7943 -0.7863 -0.7945 -0.7862

f2 -0.3021 -0.2908 -0.3023 -0.2905

f3 -0.3021 -0.2908 -0.3023 -0.2905

f4 -0.7943 -0.7863 -0.7945 -0.7862

f5 0.3887 0.4013 0.3882 0.4018

f6 -0.8182 -0.8108 -0.8187 -0.8104

f7 -0.2674 -0.2569 -0.2679 -0.2563

f8 -0.2674 -0.2569 -0.2679 -0.2563

f9 -0.8182 -0.8108 -0.8187 -0.8104

f10 0.1817 0.1891 0.1812 0.1895

Table (1): Static Analysis of the example problem using IFEM and combination method

Second, a buckling analysis is performed using the method presented in this work. Also, the
solution to a combinatorial buckling analysis is obtained, and the results for the fundamental
critical buckling load is summarized in Table (2).

Lower Lower Lower Relative Relative

Bound Bound Bound Uncertainty Uncertainty
with
Present Exact Combination Present Combination
Method Forces Method Method Method

Pcr1 / AE 0.1080 0.1081 0.1093 0.021 0.010

Table (2): Buckling of the example problem using the present method and comparison with the
combinatorial analysis results

In practice, the lowest buckling load is the only value of interest. As such, we have compared
only the lower bound in Table 2. The example problem shows an overestimation of the width of
the interval results of the proposed method compared to a combinatorial solution. The
overestimation could be attributed to three possible sources: overestimation in the interval values

REC2004
 11

of the static solution of internal forces, overestimation in the eigenvalue solution or

overestimation from uncoupling of the element forces Eq. (10) and the critical load Eq. (11).
The internal forces calculated by the interval method and the exact combinatorial results are
correct in the first three digits. The eigen solution has been proved to be sharp. Therefore the
uncoupling of the static solution from Eq. (10) and the stability equation (11), is the most likely
cause of the overestimation of the width seen in the solution. This can be seen by examining a
solution where the exact combinatorial internal forces are used in Eq. (11) to find the critical
buckling load. In this calculation, the critical load in (0.1081), just slightly above the results
from the proposed method of 0.1080 (See Table 2).

6. Discussion

In this paper, we have introduced a method for linear stability analysis of a structure with
stiffness properties expressed as an interval quantity. To our knowledge, this is the first
treatment of interval methods for structural stability. The conventional two step method
consisting of solving the linear static problem for internal forces and subsequent solution of an
eigen problem for the critical buckling load has been adapted from the non-interval approach.
The method presented provides a lower bound for the minimum buckling load. Dependency of
the interval internal forces and interval stiffness parameters have not been included in the method;
this is the expected cause of loss of sharpness in the interval results.

References

1. Bellman, Richard, 1960, Introduction to Matrix Analysis, McGrawHill.

2. Dief, A.,1991, Advanced Matrix theory for Scientists and Engineers,pp.262-281. Abacus
Press (1991)
3. Hollot, C. and Bartlett, A., “ On the eigenvalues of interval matrices”, Technical Report,
Department of Electrical and Computer Engineering, University of Massachusetts, Amherst,
MA (1987).
4. McGuire, W. , Gallagher, R. and Ziemian ,R. 2000. “Matrix Structural Analysis”, John Wiley
& Sons.
5. Muhanna, Rafi L. & Mullen, Robert L., 2001. “Uncertainty in Mechanics Problems-Interval-
Based Approach”. Journal of Engineering Mechanics June-2001, pp.557-566.
6. Modares, Mehdi and Mullen, Robert L., 2004. “ Free Vibration of Structures with Interval
Uncertainty ”. 9th ASCE Specialty Conference on Probabilistic Mechanics and Structural
Reliability.
7. Muhanna, Rafi L., Mullen, Robert L., Zhang, Hao, 2004 “Interval Finite Elements as a Basis
for Generalized Models of Uncertainty in Engineering Mechanics”. NSF workshop on
Reliable Engineering Computing
8. Strang, Gilbert, 1976.“ Linear Algebra and its Applications ”, Massachusetts Institute of
Technology

REC2004

View publication stats