You are on page 1of 16

ARTICLE IN PRESS

Thin-Walled Structures 44 (2006) 585–600


www.elsevier.com/locate/tws

Buckling mode decomposition of single-branched open cross-section


members via finite strip method: Application and examples
S. Ádánya, B.W. Schaferb,
a
Department of Structural Mechanics, Budapest University of Technology and Economics, 1111 Budapest, Müegyetem rkp. 3., Hungary
b
Department of Civil Engineering, Johns Hopkins University, Latrobe Hall 210, Baltimore, MD 21218, USA
Received 17 October 2005; received in revised form 23 March 2006; accepted 30 March 2006
Available online 30 May 2006

Abstract

The objective of this paper is to provide implementation details of, and practical examples for, modal decomposition of the cross-
section stability modes of thin-walled members by constraining a traditional finite strip method (FSM) solution. The theoretical
development of the proposed method is provided in a companion to this paper [Ádány S, Schafer BW. Buckling mode decomposition of
single-branched open cross-section members via finite strip method: derivation. Thin-walled Structures, submitted for publication,
companion to this paper.] The constraint matrix, which is directly applied to the elastic and geometric stiffness matrices of a traditional
FSM solution in order to constrain the deformations, is provided along with all formulae necessary in its construction. In addition, a
completely worked out numerical example is provided to aid in implementing the constrained FSM solution. The authors implemented
the constrained FSM in the open source program CUFSM. This modified version of CUFSM is then used to provide a series of
numerical examples that illustrate (i) the advantages of performing modal decomposition, (ii) the importance of understanding and
defining the deformation fields related to a desired mode, and (iii) the behavior of constrained FSM stability solutions compared with
classical analytical solutions, GBT, and unconstrained FSM. Decomposition of the cross-section buckling classes related to global and
distortional modes is demonstrated. Further, the impact of how to select the deformation fields and perform modal decomposition for
cross-section stability modes within a class, e.g., for the traditional three global modes (weak-axis flexure, strong-axis flexure and
flexural-torsional buckling), is explored and the impact of the deformation field definitions demonstrated. Comparisons of the
constrained FSM solutions with other available solutions demonstrate the importance of properly determining when beam theory and
plate theory should apply to the cross-section stability of thin-walled members.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Stability; Thin-walled; Modal decomposition; Finite strip method

1. Introduction and (iii) which is able to calculate any of the modes


separately from the others as well as handle arbitrary
In thin-walled members usually three basic types of
interactive modes.
instability phenomena are distinguished: global buckling
In Ádány and Schafer [1], the companion to this paper, a
(e.g., flexural, flexural-torsional or lateral-torsional), dis-
new approach is proposed to overcome this problem. The
tortional buckling, and local (or local plate) buckling.
proposed approach is based on appropriately constraining
Although these basic buckling modes are widely agreed
general purpose numerical methods such as the finite
upon, there is still no complete method which can handle
element method or finite strip method (FSM). Within the
all of these modes in a robust and simple way: (i) which is
context of FSM the derivation of the necessary constraint
generally applicable for a wide range of cross-sections, (ii)
matrix is provided in [1]. The full derivations are thought to
which is widely available and supported by software tools,
be important and necessary for the complete understanding
Corresponding author. Tel.: +1 410 516 7801; fax: +1 410 516 7473. of the proposed method; however, the paper does not
E-mail addresses: sadany@epito.bme.hu (S. Ádány), schafer@jhu.edu demonstrate how the theory can be applied to practice. It is
(B.W. Schafer). this latter issue which is covered in this paper.

0263-8231/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tws.2006.03.014
ARTICLE IN PRESS
586 S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600

The focus of this paper is the practical application of


RU1
the buckling mode decomposition method. Here, numerical
examples are provided, along with options for imple-
mentation (e.g., application of various base vector
systems) and the results are discussed in detail. The RU1 is a 1×n submatrix,
examples demonstrate the applicability and potential of
the proposed method, help the interested reader in better RU
RU is a (n-2)×n submatrix,
understanding the idea of constrained FSM, and demon-
strate key differences between proposed and existing
RUn is a 1×n submatrix,
methods.
The paper begins in Section 2 with the calculation of
pure buckling modes. The overall procedure is summar- RUn
ized, including the definition of the RGD constraint matrix,
and a discussion of length-dependency and orthogonality
of the base vectors. Next, in Section 3 a numerical example
is presented. The example provides the final result (i.e., the
critical force) and also all of the internal variables/results, RV is an n×n submatrix,
including the numerical values of the RGD and RGD,H
constraint matrices. Then, in Section 4 the results of studies RV
comparing the proposed method to analytical solutions,
generalized beam theory (GBT), and regular (uncon-
strained) FSM are summarized. The comparisons highlight
some basic differences of the various calculation methods,
which are discussed in Section 5. Finally, conclusions are
RGD =
drawn in Section 6. RW1
An Appendix is also included in which all formulae
necessary to implement the constrained FSM are provided.
Through this Appendix, one can relatively easily repeat
and verify the provided results, even without complete RW1 is a 1×n submatrix,
knowledge of all steps of the derivations.
RW
Similar to the companion paper, two left-handed RW is a (n-2)×n submatrix,
coordinate systems are used in this paper: global and local
(see Fig. 1 of [1]. The global coordinate system is denoted RWn is a 1×n submatrix,
as: X–Y–Z, with the Y-axis parallel with the longitudinal
axis of the member. The local system is denoted as x–y–z.
The local system is always associated with a plate element RWn
(strip) of the member such that the x-axis is parallel with RΘ1
the plate element, and perpendicular to the member
longitudinal axis, the y-axis is parallel with Y, and the RΘ2 RΘ1 is a 1×n submatrix,
z-axis is perpendicular to the x–y plane. In accordance with
the displacement degrees of freedom (DOF) used in
RΘ2 is a 1×n submatrix,
FSM, three translations and a rotation is assigned to each

nodal line. Global translations and rotation are denoted
Rθ is a (n-4)×n submatrix,
by U, V, W and y, and local translations and rotation as
u, v, w and y.
RΘn-1 RΘn-1 is a 1×n submatrix,
2. Pure mode calculation
RΘn RΘn is a 1×n submatrix.
2.1. General
Fig. 1. Submatrices of the RGD constraint matrix.
The main steps for pure mode calculation (also see [1])
can be summarized as follows. First, define the space as follows:
spanned by the global and distortional modes: the GD
d ¼ RGD dGD , (2.1)
space. The GD space contains all deformation modes that
satisfy the GBT basic assumptions. Mathematically, the where dGD is the vector of warping displacements.
GD space is defined by the RGD constraint matrix, so The GD space may be separated into G and D spaces for
that any general deformation vector, d, may be expressed global and distortional deformation modes, respectively,
ARTICLE IN PRESS
S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600 587

by introducing an H ¼ ½HG HD  matrix of base vectors and one torque mode) may be defined as follows:
such that
dGD ¼ HdH (2.2)  
HG ¼ HG1 HG2 HG3 HG4 ¼ (2.8)
from which the pure global and pure distortional
deformation modes are expressed as
where Xi and Zi are global coordinates, and oi stand for
d ¼ RG dH;G ¼ RGD HG dH;G and the sectoral coordinates of the cross-section nodes. Note,
d ¼ RD dH;D ¼ RGD HD dH;D , ð2:3Þ that it is the distribution, not magnitude, of the long-
itudinal (warping) displacements which is relevant, there-
where dH,G and dH,D are the global and distortional vectors
fore the HG vectors can be scaled arbitrarily.
of modal coordinates, respectively.
The pure global and distortional buckling modes may be
calculated by solving the constrained eigenvalue problem 2.3.2. Distortional base vectors
as follows: The warping functions for the distortional modes must
be defined such that the space that they define together
Ke;G UG ¼ KG Kg;G UG and with the base vectors of the G space result in the GD space.
Ke;D UD ¼ KD Kg;D UD , ð2:4Þ As suggested by the GBT, this can be ensured by analyzing
solely the warping functions, namely: the warping func-
where UG and UD are the eigenvectors (i.e., displacement tions that belong to the distortional modes must be
vectors of modal coordinates), Ke,G, Kg,G and Ke,D, Kg,D are orthogonal to any warping functions that belong to the
the stiffness and geometric stiffness matrices of the global modes. (Note that global warping functions are
constrained FSM problem for global and distortional presented in Fig. 2 of the companion paper.)
modes, respectively, defined as Though various orthogonality conditions are possible,
Ke;G ¼ RTG Ke RG and Kg;G ¼ RTG Kg RG , (2.5) here the following condition is adopted:
Z
Ke;D ¼ RTD Ke RD and Kg;D ¼ RTD Kg RD , (2.6) vi ðxÞvj ðxÞtðxÞ dx ¼ 0, (2.9)

and KG and KD are diagonal matrices with the eigenvalues where the integral is over the whole cross-section, t(x) is the
for global and distortional modes, respectively. thickness, while vi(x) and vj(x) are two arbitrary warping
The derivations necessary to construct the RGD matrix functions of the G and D space, respectively. Note, the
are presented in detail in the companion paper, while its above expression is identical with the one used in GBT [2].
practical calculation, as well as the calculation of the H To find HD base vectors for the distortional modes that
matrix is discussed in subsequent sections of this paper. satisfy the orthogonality condition of Eq. (2.9), re-write the
condition in terms of HG, HD base vectors, as follows.
2.2. Compilation of the constraint matrix From the left-hand side of Eq. (2.9) the integral can be
calculated from strip to strip:
RGD is defined by its submatrices, as shown in Fig. 1. n1 Z bðkÞ
X
The derivation of each submatrix of RGD is detailed in the vi vj tðkÞ dx ¼ 0: (2.10)
0
companion paper [1]. The necessary formulae are summar- k¼1

ized in the Appendix to this paper. Note that the assumed


DOF order is identical to the one used throughout the
companion paper and may be described by the following
displacement vector:
 
dT ¼ UT VT WT HT , (2.7)
 
where UT ¼ U 1 U 2 . . . U n is the vector of long-
itudinal (warping) nodal displacements, and VT, WT and
HT similarly described.

2.3. Separation of G and D space

2.3.1. Global base vectors


For the global modes: HG1, HG2, HG3 and HG4, the base
vectors may be calculated from the mode definitions given
in [1]. Thus, the base vectors for the four global
deformation modes (one axial mode, two bending modes, Fig. 2. Cross-section for example (a).
ARTICLE IN PRESS
588 S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600

Considering that the thickness is constant, then sub- Such, length-independent base vectors have potential
stituting the corresponding FSM interpolation functions numerical advantages, include certain similarities to the
(e.g., see [3]): section property calculations performed in GBT, and serve
Z bðkÞ h to highlight the role of member length, or buckling half-
Xn1
x x ðkÞ i rpy wavelength, in the stability calculation.
t ðkÞ
1  vðkÞ i;1 þ vi;2 cos
k¼1 0 b b a The HG (Eq. (2.7)) and HD (Eq. (2.17)) base vectors
h x  x i rpy contain only cross-section data, and are thus
 1  vðkÞ þ vðkÞ cos dx ¼ 0. ð2:11Þ
b j;1 b j;2 a length-independent by definition. Portions of the
Introducing a non-dimensionalized coordinate x ¼ x=b, constraint matrix RGD, as defined in the Appendix, are
and considering that the cosine functions are independent linearly dependent on the length. The RGD submatrices
of x, we may also write that related to the transverse displacements: RU, RW and RH are
Z 1h linearly proportional to the length, while the submatrix
rpy X n1 i related to the warping displacements, RV, is independent of
cos2 tðkÞ bðkÞ ð1  xÞvðkÞ ðkÞ
i;1 þ xvi;2
a k¼1 0 length.
h i A length-independent RGD,0 may be defined by using the
 ð1  xÞvj;1 þ xvðkÞ
ðkÞ
j;2 dx ¼ 0 ð2:12Þ RU, RW and RH sub-matrices without division by kr:
which finally leads to the following expression: RGD ¼ LRGD;0 , (2.18)
1 rpy X
n1
where L is a diagonal matrix as follows:
cos2 AðkÞ ð2vðkÞ ðkÞ ðkÞ ðkÞ
i;1 vj;1 þ vi;1 vj;2
6 a k¼1 * U1 Un V1 Vn W1 Wn Y1 Yn +
þ vðkÞ ðkÞ
j;1 vi;2 þ 2vðkÞ ðkÞ
i;2 vj;2 Þ ¼ 0, ð2:13Þ L¼ 1

1
1  1
1

1 1

1
kr kr kr kr kr kr
where A(k) is the area of the kth plate element (strip) of the ð2:19Þ
cross-section. Noting the equivalence of local v and global
V nodal displacements and using V i;k to denote the kth and
element of the Vi vector of nodal warping displacements,
and then removing constant terms, the orthogonality kr ¼ rp=a, (2.20)
condition may be re-written as follows: where a is the member length, and r is the number of
X
n1  longitudinal half-sine waves. For numerical efficiency this
AðkÞ 2V i;k V j;k þ V i;k V j;ðkþ1Þ þ V j;k V i;ðkþ1Þ definition is advantageous since the stability analysis is
k¼1
 typically completed at many lengths, and RGD,0 may be
þ 2V i;ðkþ1Þ V j;ðkþ1Þ ¼ 0 ð2:14Þ calculated prior to the stability analysis.
With RGD and H known, one can calculate the transverse
or in matrix form:
2 3 displacement components of the base vectors, i.e., the base
2Að1Þ Að1Þ 0 0 0 0 vectors in the original space of FSM nodal degrees of
6 ð1Þ 7 freedom. This is done by applying Eq. (2.2), as follows:
6 A 2Að1Þ þ 2Að2Þ Að2Þ 0 0 0 7
6 7
6 .. .. .. 7
T6
6 . . . 7 RGD;H ¼ RGD H ¼ LRGD;0 H ¼ LRGD;H0 . (2.21)
Vi 6 7Vj ¼ 0,
7
6 0 0 0 7 Here, RGD,H0 is the matrix of base vectors in the original
6 .. .. .. 7
6 7 FSM nodal space, which contains both warping
4 . . . 5
0 0 0 0 Aðn1Þ 2Aðn1Þ and transverse displacements. RGD,H0 is length-
independent, while L represents a transformation that
(2.15) modifies transverse displacements only, without any effect
on the warping components. The distribution of neither
VTi AVj ¼ 0. (2.16)
warping nor transverse displacements is changed as length
For the HD vectors, thus, we have the following condition: changes, only the ratio of transverse and longitudinal
(warping) displacements is influenced by the member
HTD AHG ¼ 0. (2.17)
length.
The necessary mathematical apparatus is not detailed here; The length-independent constraint matrix introduced
however, modern software tools such as Matlab [4] provide above (i.e., RGD,0) and base vectors (i.e. the column
simple way to solve the problem numerically. vectors of RGD,H0) can also be interpreted as the
constraint matrix and base vectors of a ‘‘unit’’ member,
2.4. Length-independent base vectors respectively. Recognizing the longitudinal (sinusoidal)
function employed in FSM, the unit member does
It is possible to construct the base vectors that form the not have unit length, but instead is defined such that
constraint matrix RGD prior to any stability calculation. kr ¼ 1, which means a member length a ¼ rp. If r is taken
ARTICLE IN PRESS
S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600 589

as 1 (traditional in FSM stability applications), the length only one distortional base vector. Thus, if we want a
is p. general means to define HD we must introduce additional
conditions.
2.5. Orthogonal base vectors In GBT the distortional modes are calculated via a
special diagonalization procedure which can be interpreted
Orthogonality of the base vectors in the G space with as a solution of an auxiliary eigenvalue problem defined for
respect to any base vectors in the D space is essential for a unit length member. We propose here a procedure for the
separating the G and D spaces, and enables the desired calculation of HD that involves (i) initial definition of the
decomposition of buckling modes. Further orthogonaliza- distortional base vectors, (ii) definition of a ‘‘unit’’
tion of base vectors within the G or D space is also possible, member, and (iii) calculation of pure distortional buckling
and has some advantages. By applying the orthogonality modes for an axially loaded unit member. This procedure
criterion defined in Eq. (2.9), we can write that results in HD base vectors, hence referred as ‘‘unit-member
Z ( axial’’ base vectors, which are in good agreement with
¼ 0 if iaj; GBT. The procedure is detailed as follows.
vi ðxÞvj ðxÞtðxÞ dx
a0 if i ¼ j; First, create an initial set of HD distortional base vectors:
HG and the orthogonality condition of Eq. (2.17) provide
i; j ¼ 1 . . . nG or i; j ¼ 1 . . . nD , ð2:22Þ this. Next, define a ‘‘unit’’ member, e.g., using the
where vi(x) and vj(x) are two arbitrary warping functions of definition of Section 2.4, that is kr ¼ 1. This ‘‘unit’’
the G or D space (but both from the same space), and nG member of length rp is subjected to pure axial loading
and nD are the number of global and distortional (necessary for calculating the geometric stiffness matrix).
deformation modes, respectively. Alternatively, expressed Finally, calculate the pure distortional buckling modes, by
by means of displacement vectors orthogonality implies: solving the constrained eigenvalue problem:
(
¼ 0 if iaj; Kue;D UuD ¼ KuD Kug;D UuD , (2.25)
T
Vi AVj i; j ¼ 1 . . . nG or i; j ¼ 1 . . . nD ,
a0 if i ¼ j; where Kue;D and Kug;D are the elastic and geometric stiffness
matrices of the constrained FSM problem for distortional
(2.23)
modes, respectively, and are defined as
which leads to the following condition for the H matrix:
Kue;D ¼ RTD Kue RD ¼ HTD RTGD Kue RGD HD , (2.26)
T
H AH ¼ Diag. (2.24)
The HG global base vectors, defined by Eq. (2.8), Kug;D ¼ RTD Kug RD ¼ HTD RTGD Kug RGD HD , (2.27)
automatically satisfy the above condition provided the where RGD is the constraint matrix, Kue and Kug are the
X–Z global coordinate system coincides with the principal elastic and geometric stiffness matrices of the unit member,
axes of the cross section. respectively, KuD is the diagonal matrix of distortional mode
As far as distortional base vectors are concerned there is eigenvalues, and the superscript u denotes unit-member
an infinite number of different HD base systems that satisfy axial modes.
Eq. (2.24). Currently, no clear and physically meaningful If the HD basis is transformed to the basis where the
definition for distortional base vectors exists. However, warping displacements are equal to the buckling modes of
GBT produces a series of orthogonal deformation modes the unit member under axial load:
that have physical appeal, therefore, a methodology
motivated from GBT is discussed in the following section. HuD ¼ HD UuD (2.28)
Although full systems of orthogonal base vectors are not the resulting HuDis then similar to the GBT distortional
crucial for separating the G and D spaces, such a system modes.
may be advantageous when calculating buckling modes Although this procedure requires the solution of an
according to an individual deformation mode (e.g., flexural auxiliary eigenvalue problem, it is completed only once per
buckling of a member with symmetrical cross-section). cross-section, and is of a much smaller size than the full
Moreover, the full orthogonal system is also useful for eigenstability problem. For example, a simple C- or Z-
modal identification when the contribution of a pure mode section has only two distortional modes, therefore solution
to a general deformation mode is provided. This question is of Eq. (2.25) requires only two eigenvectors and eigenva-
not discussed here, but briefly addressed in [5,6]. lues. Further, the procedure involves no greater complexity
than the GBT procedure.
2.6. Unit-member axial modes
2.7. The constrained eigenvalue problem
Orthogonal G base vectors as defined above are identical
with the ones used in GBT. However, the distortional base Having the constraint matrix and base vectors calcu-
vectors, HD, cannot be unambiguously defined solely lated, the pure buckling modes may be calculated by
from HG and the orthogonality condition, unless there is following the procedure summarized in Section 2.1.
ARTICLE IN PRESS
590 S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600

" #
When calculating all the pure global and all the pure 0 0:005 0:005 0
distortional buckling modes, we apply Eqs. (2.4)–(2.6). It is S2 ¼ ,
24 0 0:005 0:005 0
useful to realize that for the calculation of all the modes for " #
a given buckling type it is not necessary to have unit- 0:0125 0:0125 0 0
member axial, nor orthogonal base vectors; the only C2 ¼ ,
24 0 0 0:0125 0:0125
important matter is to define the G and D spaces, which
is included in the RG and RD matrices, independent of the " #
HG and HD vectors. 0 0:005 0:005 0
It is also possible to make calculations for an individual S3 ¼ ,
24 0 0 0 0
buckling mode. This requires the same steps as defined " #
above for calculation of the pure modes. The difference is 0 0 0 0
C3 ¼ ,
that, instead of using the whole RG or RD matrices, we use 24 0 0 0:0125 0:0125
a single column, which may be denoted as RG1, RG2, etc. or
RD1, RD2, etc. Mathematically, the individual mode
calculation may be completed for arbitrary base vectors,    
S^ 3 ¼ 0 0 0 0 ; ^ 3 ¼ 0:0125 0:0125
C 0 0 ,
but physical meaning to an individual distortional mode 14 14
can be assigned only if unit-member axial base vectors are
employed.
   
S~ 3 ¼ 0 0 0 0 ; ~3 ¼ 0
C 0 0:0125 0:0125 ,
14 14

3. Illustrative example
   
S^ 4 ¼ 0:0667 0 0 0 ; ^4 ¼ 0 0
C 0 0 ,
3.1. General
14 14

An illustrative example is presented here to demonstrate


the capabilities of the proposed method of constrained    
S~ 4 ¼ 0 0 0 0:0667 ; ~4 ¼ 0 0
C 0 0 ,
FSM. Complete numerical results are given, including the 14 14
values of internal variables, to allow the reader to follow
the calculations.
 
The first illustrative example (Example (a)) is a simple S^ 5 ¼ 0:0667 0:0667 0 0 0 0 ,
column member with a symmetrical C cross-section, the 16
dimensions of which are presented in Fig. 2. The member
length is a ¼ 650 mm, the modulus of elasticity is  
E ¼ 210 000 MPa, and Poisson’s ratio is n ¼ 0:3. (All ^5 ¼ 0
C 0 0 0 0 0 ,
16
dimensions are mid-line for the cross-section.)

 
S~ 5 ¼ 0 0 0 0 0:0667 0:0667 ,
16
3.2. Base vectors

3.2.1. S, C and B matrices  


~5 ¼ 0
C 0 0 0 0 0 ,
By using the formulae given in the Appendix, the 16
intermediate S, C and B matrices are as follows:

2 3    
560 200 400 200
0 0:0125 0:0125 0 0 0 B1 ¼ ; B2 ¼ ,
6 7 22 200 560 22 0 160
6 0 0:0125 0:0125 0 0 07
6 7
S1 ¼ 6 7,
46 60 0 0 0:0125 0:0125 0 7
4 5    
B^ 2 ¼ 80 0 ; ~2 ¼ 0
B 80 .
0 0 0 0:0125 0:0125 0 12 12
2 3
0:0667 0:0667 0 0 0 0
6 7
6 0 0 0:5 0:5 0 0 7
6 7
C1 ¼ 6 7,
46 6 0 0 0:5 0:5 0 0 7 3.2.2. The RGD matrix
4 5
The RGD matrix follows from equations of the
0 0 0 0 0:0667 0:0667
Appendix.
ARTICLE IN PRESS
S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600 591

RGD ¼

3.2.3. H matrices
To have orthogonal H base vectors, we need the A Note that the columns of the H matrix may be arbitrarily
matrix, defined by Eq. (2.15). scaled since it is the distribution of warping displacements
2 3 that is important. Obviously, normalization is possible but
60 30 0 0 0 0
6 30 380 160 none is applied here.
6 0 0 0 77
6 7 By using the above H matrix the RGD,H matrix can be
6 0 160 1120 400 0 0 7 calculated. Now, the columns of RGD,H correspond to
A¼6 6 7.
60 0 400 1120 160 0 7 7 individual deformation modes in a manner similar to GBT.
6 7
40 0 0 160 380 30 5
0 0 0 0 30 60
The (unit-member axial) orthogonal H base vectors are
2 3
1 85 57:436 6242:4 95:003 92:22
6 1 100 57:436 4520:5 21:748 24:935 7
6 7
6 7
6 1 100 22:564 3479:5 2:2892 7:5348 7
H¼66 1 100 22:564
7.
6 3479:5 2:2892 7:5348 77
6 7
4 1 100 57:436 4520:5 21:748 24:935 5
1 85 57:436 6242:4 95:003 92:22
ARTICLE IN PRESS
592 S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600

RGD;H ¼

(a) RG1 RG2 RG3 RG4 RD1 RD2

(b) RG1 RG2 RG3 RG4 RD1 RD2

Fig. 3. Unit-member axial modes: (a) warping displacements; (b) transverse displacements.
ARTICLE IN PRESS
S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600 593

The modes that are represented by the above vectors are 10


shown in Fig. 3: both warping displacements and 9 D1
transverse displacements are given. D2
8 example (a)
7
3.3. Pure mode calculation
6 D2

Pcr/Py
Pure mode calculation requires solving the constrained 5 D1
eigenvalue problem, defined by Eqs. (2.4)–(2.6). Consider
4
the C-section of example (a) from the previous Section, but
now with member lengths from 50 to 10 000 mm. For the 3
sake of comparison, the buckling modes of the example are 2
calculated with various options, including: (a) pure global
modes, (b) pure distortional modes, (c) interacted modes in 1
the GD space, and (d) individual global deformation 0
modes. 102 103 104
If the constrained eigenvalue problem is solved in the G half-wavelength (mm)
space, three pure global buckling modes can be calculated:
Fig. 5. Pure distortional buckling modes for example (a).
G1, G2 and G3. (Note, the member is subjected to pure axial
compression, therefore the three modes always exist.)
These modes are shown in Fig. 4, where the critical force 5
(Pcr) is plotted as a function of the half-wavelength for the example (a) GD1
4.5
two flexural-torsional (FT) modes (G1, G3) and the one D1
flexural mode (G2). Note, Pcr is normalized by Py, the 4 G1
squash load of the cross-section. In this example, the first 3.5
GD2
FT mode provides the lowest critical force for a length less D2
than 7000 mm (approx.), while for longer columns the pure 3 G2
Pcr/Py

flexural mode governs. Also note, the constrained FSM 2.5


solution has successfully removed local and distortional
2
buckling modes from a regular (unconstrained) FSM
solution. 1.5
If the constrained eigenvalue problem is solved in the D 1
space two pure distortional buckling modes can be
calculated, as shown in Fig. 5. The first pure mode (D1) 0.5
is a symmetrical mode, while the second pure mode is anti- 0
symmetrical (D2). In the example D1 always governs, which 102 103 104
is typically the case for symmetrical C-section columns. half-wavelength (mm)

Fig. 6. G, D and GD buckling modes for example (a).

20
example (a)
G1
G2
18
G2
Fig. 6 presents the modes that are calculated within the
16 G3
GD space. Note, theoretically five modes exist; however,
14 only the first two are shown: GD1 and GD2. For the sake of
12 G3
comparison the curves belonging to the first two pure
global buckling modes (G1, G2) and to the two pure
Pcr/Py

10 G1 distortional buckling modes (D1, D2) are also given. For


8 shorter wavelengths the D1 and GD1 curves are practically
identical, for longer wavelengths G1 and GD1 are nearly
6
identical. Similarly, GD2 covers D2 and G2 for shorter, and
4 longer (wave)lengths, respectively. However, in the region
2 where the pure buckling curves intersect each other, the
presence of an interaction effect can clearly be observed.
0
The interaction takes place between G1 and D2 modes, and
102 103 104
between G2 and D1 modes, since G1–D2 are both modes
half-wavelength (mm)
with some torsion, while G2–D1 modes both only involve
Fig. 4. Pure global buckling modes for example (a). in-plane deformation.
ARTICLE IN PRESS
594 S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600

The difference between pure buckling modes and 4. Validation of the method
individual deformation modes is illustrated in Fig. 7 in the
context of global modes. If the eigenvalue problem is 4.1. Comparison with analytical solution and GBT
solved within the G space, these are the pure global buckling
modes that can be calculated (as already shown in Fig. 4), (a) General: Theoretical considerations and the example
denoted by G1, G2 and G3. However, it is also possible to of the previous section demonstrate the applicability of
solve the eigenvalue problem separately for individual constrained FSM for modal decomposition of global and
deformation modes, namely: the major-axis-bending defor- distortional modes. It is also important to compare the
mation mode (RG2), minor-axis-bending deformation results to existing methods. Analytical formulae (for global
mode (RG3) and the torque deformation mode (RG4). This modes) and GBT results (for both global and distortional
leads to the results for individual buckling modes designated modes) are selected for comparison.
Gmajor, Gminor and Gtorque in Fig. 7. The solution requires Two examples are considered, example (a) from the
three consecutive eigenvalue calculations, but each one for previous section and example (b), a S-section with the
only a single DOF system. dimensions defined in Fig. 8.
As Fig. 7 shows, the G2 buckling mode is identical with (b) Constrained FSM comparison with classical flexural-
the Gminor mode, since G2 is the pure weak-axis flexural torsional analytical solution: From Figs. 9 and 10 it is
mode. However, G1 and G3 are different from the evident that, although the tendencies of the classical
individual buckling modes. In fact, both G1 and G3 can
be constructed as a linear combination of Gmajor and
Gtorque deformations. Note, that for members with doubly
symmetrical cross-sections the pure buckling modes and
individual buckling modes are identical.
It may be interesting to highlight the relationship of unit-
member axial modes to pure buckling modes and individual
buckling modes. For global buckling, unit-member axial
modes are individual buckling modes calculated for a
‘‘unit’’ member of length rp, subjected to an axial force.
Unit-member axial modes are typically different than pure
global buckling modes, see example (a). For distortional
buckling modes, by definition, unit-member axial
distortional buckling modes are identical with both
pure distortional buckling modes and individual distor-
tional buckling modes for ‘‘unit’’ members under axial
compression.

Fig. 8. Cross-section for example (b).

20
20 Theory G1
deformation Gmajor(RG2) 18 example (a)
Theory G2
18
16 Theory G3
16 G1
14
G2
14
12 G3
Pcr/Py

12
10
Pcr/Py

pure G3 mode
10
8
G1
8
G2 6
6 G3 4
4 Gmajor(RG2)
2
Gminor(RG3)
2
Gtorque(RG4) 0
0 102 103 104
102 103 104 half-wavelength (mm)
half-wavelength (mm)
Fig. 9. Example (a): analytical modes vs. pure global modes for
Fig. 7. Individual global modes vs. pure global modes for example (a). constrained FSM.
ARTICLE IN PRESS
S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600 595

20 2
Theory G1 example (b)
18 example (b) Theory G2
1.8

16 Theory G3 1.6
G1
14 1.4
G2
12 G3 1.2

Pcr/Py
Pcr/Py

10 1

8 0.8

0.6 DGBT
6
D1
4 0.4
GGBT
2 0.2 G1

0 0
102 103 104 102 103 104
half-wavelength (mm) half-wavelength (mm)

Fig. 10. Example (b): analytical modes vs. pure global modes for Fig. 12. Example (b): GBT modes vs. pure modes for constrained FSM.
constrained FSM.

2 2
example (a)
(FSM All)1
1.8 1.8
GD1
1.6 1.6 (FSM All)2
1.4 1.4 GD2

1.2 1.2 example (a)


Pcr/Py
Pcr/Py

1 1

0.8 0.8

0.6 DGBT 0.6

0.4 D1 0.4
GGBT
0.2 0.2
G1
0 0
102 103 104 102 103 104
half-wavelength (mm) half-wavelength (mm)

Fig. 11. Example (a): GBT modes vs. pure modes for constrained FSM. Fig. 13. Regular FSM vs. constrained FSM: example (a).

analytical solution for flexural-torsional buckling [7] and 4.2. Comparison with regular FSM
constrained FSM are the same, constrained FSM gives
higher critical forces. The difference is near 10% for any In the previous section constrained FSM results are
buckling lengths of practical importance. This difference is compared to analytical solutions and GBT since those
a direct consequence of the basic assumptions between methods calculate pure buckling modes. In this section we
beam and plate theory as discussed in Section 5. compare with regular (unconstrained) FSM, which is also a
(c) Constrained FSM comparison with GBT results: widely used method in buckling analysis of thin-walled
Figs. 11 and 12 present the critical forces for both global members.
and distortional buckling, as calculated by constrained For examples (a) and (b), Figs. 13 and 14 present the
FSM and GBT. For global buckling, the same nearly 10% results of regular and constrained FSM. The results are
difference as noted above exists. For distortional buckling different in nature for short buckling half-wavelengths.
a difference between 0% and 10% is observed. For Constrained FSM successfully eliminates the short half-
distortional buckling, the longer the buckling length, the wavelength local modes, which are naturally included in a
smaller the difference. At the distortional minimum, the regular FSM analysis. Note, to have accurate results for
difference is 4% for both examples. For further discussion local modes, intermediate nodes must be allowed in the
see Section 5. cross-section.
ARTICLE IN PRESS
596 S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600

2 can be expressed as follows:


(FSM All)1 2 3
1.8
GD1 2 3 E nE 2 3
1.6 (FSM All)2
sx 6 1  n2 1  n2 0 7 e x
6
6 s 7 6 nE 7
E 76 e 7
1.4 GD2 4 y5¼6
0 74 y 5. (5.2)
txy 4 1  n2 1  n2 5 g
xy
1.2 example (b) 0 0 G
Pcr/Py

1
Here we will concentrate on the normal stresses, the shear
0.8 stresses and strains appear in the above equation only for
0.6 the sake of completeness.
It is implicitly assumed in a plate model that the plates
0.4
that form the cross-section are thin enough (compared to
0.2 their width and length) so that the Poisson effect should be
0 considered. Thus, neither sx nor sy is negligible, while sz is
102 103 104 implicitly assumed to be zero.
half-wavelength (mm) If we calculate the pure global or distortional modes by
applying the constraint matrix (RGD), we generate defor-
Fig. 14. Regular FSM vs. constrained FSM: example (b).
mations that must satisfy the conditions that ex and gxy are
zero (GBT assumptions #1 and #2 in the companion
If the buckling length is long enough, global buckling paper). Thus, for constrained FSM the normal stresses can
governs the behavior. For long length the regular and be expressed from Eq. (5.2) as follows:
constrained FSM coincide well, though certain difference
exists (see Section 5).
nE
For intermediate wavelengths both regular and con- sx ¼ ey ,
strained FSM predict distortional buckling modes. Regular 1  n2
FSM gives slightly smaller critical values, suggesting that
other deformation modes (e.g., local) have influence on the E
behavior. sy ¼ ey ,
1  n2

5. Discussion on the definition and calculation of pure modes


sz ¼ 0. (5.3)
5.1. Discussion on the difference between FSM and other The difference between the beam model, Eq. (5.1), and the
methods constrained FSM model, Eq. (5.3), is conspicuous. For
longitudinal stresses the difference is equal to 1/(1-n2).
In classical analytical solutions for global flexural Considering that for steel the Poisson’s ratio is approxi-
buckling it is assumed that the length of the member is mately 0.3, the difference in the longitudinal stresses is
much larger than the cross-sectional dimensions, and, approximately 10%.
neither the width nor height of the cross-section is For flexural (global) buckling the only non-zero stress
considerably larger or smaller than the other. As a component (according to a beam model) is the longitudinal
consequence, only the longitudinal normal stresses are stress, consequently this difference in the longitudinal
considered, while the transverse normal stresses are stresses directly transfers into a difference in the critical
assumed negligible. Thus, if the problem is handled by a loads. The difference between the flexural buckling load
beam model, the stresses can be expressed (in the local calculated by the beam model and the constrained FSM is
x–y–z coordinate system) as follows: 1=ð1  n2 Þ or 10% for n ¼ 0:3, as shown in Figs. 9 and 10.
sx ¼ 0, Flexural-torsional global buckling involves shear stresses
which are not affected by the Poisson’s ratio, therefore, a
sy ¼ Eey , smaller difference (i.e., o10%) occurs between analytical
and constrained FSM solutions.
For distortional buckling, longitudinal (in-plane) and
sz ¼ 0, (5.1)
transverse (bending) stresses are both important. Trans-
which is a 1D form of Hooke’s law. Although rarely verse plate bending involves the Poisson effect which may
considered, transverse strains are not zero, even if they are also be accounted for in GBT. Thus, the difference between
small. longitudinal stresses is reduced, and the difference between
FSM calculations are predicated on a plate model, the pure distortional buckling load calculated by GBT and
therefore, we need the typical 2D form of Hooke’s law as constrained FSM is always less than 1=ð1  n2 Þ, as shown
constitutive equation which, assuming isotropic material, in Figs. 11 and 12.
ARTICLE IN PRESS
S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600 597

5.2. Discussion on global buckling 5.2.2. Regular FSM model


A regular FSM model (which is equivalent with a plate
In the following section options for calculating pure FEM model) considers both the Poisson effect and
buckling modes are discussed from the standpoint of transverse plate flexure. If the analyzed plate elements are
addressing the inherent differences in the constitutive thin enough (which is always the case for cold-formed
equations. steel), and the applied finitization is sufficiently dense, both
effects are considered in the correct way. Although these
models still involve certain approximations (e.g. stresses/
5.2.1. Beam model
strains perpendicular to the plates are neglected), it is fair
Although the beam model is the classical model of
to say that FSM models calculate the elastic buckling of a
structural mechanics, applied for centuries, and at least for
thin-walled member with negligible error.
decades even in buckling problems, its application for thin-
walled members can be regarded as only an approximation.
5.2.3. GBT model
Two important approximations are involved: (i) transverse
Since GBT is an extension of classical beam theory, it
stresses are neglected and (ii) transverse plate flexure is
naturally involves the classical beam model and leads to
neglected.
critical loads identical with those of the classical beam
To assess the consequence of neglecting the transverse
model.
stresses, consider the flexural buckling of Example (a).
Either the column buckles about its minor or major axis.
5.2.4. Constrained FSM model
The majority of the elastic strain energy develops from
The constrained FSM model is a reduction of the regular
membrane strains/stresses. If the plates that make up the
FSM model, therefore, it handles the plate elements as
member are slender enough, the plane stress assumption is
plates, i.e., together with the Poisson effect. However, the
certainly more reasonable than the 1D stress assumption,
transverse strains as well as transverse flexure is completely
which means that the transverse stresses (Poisson effect)
eliminated, consequently a constrained FSM model be-
cannot be neglected. Thus, from this aspect the beam
haves more rigidly than a regular FSM model.
model under-estimates the column rigidity.
At the same time it differs from beam models, too, as
Transverse flexure, though small in extent, always takes
already demonstrated via Eqs. (5.1) and (5.3), because it is
place during flexural buckling. This may easily be demon-
not possible to simultaneously enforce zero transverse
strated in a regular FSM or FEM analysis. It is obvious
strains (which is a requirement from the basic GBT
that transverse flexure provides additional flexibility to the
assumptions) and zero transverse stresses (which is the
column, consequently, from this aspect the beam model
assumption of beam models). Since the constrained FSM
over-estimates the column rigidity.
satisfies the zero transverse strain condition, it gives larger
Thus, the beam model involves two competing approx-
rigidity, consequently higher critical forces than the beam
imations which nearly compensate each other so that the
models, as presented in Fig. 15.
global buckling load provided by the beam model shows
good agreement with more sophisticated models (e.g.,
5.2.5. Conclusion
FSM) for most practical cases, see Fig. 15.
Theoretical elastic critical force can best be assessed by a
regular FSM model, while both the beam models (classical
beam model or GBT) and the constrained FSM model are
1 an idealization of the theoretically exact behavior, both
0.9 example (a) Theory G1 having some limitations. In the end, it is a question of
Theory G2 definition as to what to call a pure buckling mode.
0.8 Currently no definitively clear, nor widely adopted defini-
(FSM All)1
0.7 G1 tions exist.
G1 ν=0 Unconstrained FSM provides the most rigorous solu-
0.6
tion, but does not allow decomposition nor identification
Pcr/Py

0.5 of the buckling modes. Constrained FSM is clearly defined,


0.4 but its lack of agreement with classical analytical solutions
0.1
(the beam model) makes its practical use limited. Exact
0.3
agreement between the beam model and constrained FSM
0.2 is impossible. However, if we artificially assume that
Poisson’s ratio is zero, the constitutive equation for the
0.1 0.05
104
constrained FSM model, Eq. (5.3), reduces to that of the
0 beam model, Eq. (5.1). The two models are not identical,
103 104 since transverse strains are zero in constrained FSM and
half-wavelength (mm)
non-zero in the beam model, but this difference has little
Fig. 15. Global buckling calculation with constrained FSM, n ¼ 0. effect, as Fig. 15 shows.
ARTICLE IN PRESS
598 S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600

It may be interesting to mention that although the mechanical definitions for all of the buckling modes, which
applied constitutive equation has an effect on the critical could be used in the context of modern numerical tools
loads for pure global buckling modes, it has no practical such as FSM or FEM, do not exist. Work to resolve this
influence on the base vectors, or, on the constraint shortcoming is a topic of future research for the authors.
matrices. The only exception would be the case where
materials with considerably different Poisson coefficients Acknowledgments
are joined together within one cross-section.
The presented research was partially supported by the
6. Concluding remarks Korányi Imre Scholarship of the Thomas Cholnoky
Foundation, the OTKA K62970 project of the Hungarian
In this paper, the application of a constrained finite strip Scientific Research Fund and CMS-0448707 of the United
method (FSM) for performing modal decomposition of States National Science Foundation. The authors would
cross-section stability modes is presented. The theoretical also like to express their thanks to Dinar Camotim and
background of the method is summarized in a companion Nuno Silvestre from Technical University of Lisbon,
paper [1]. To aid others in implementing the constraints Portugal, for providing the GBT results presented in this
within the finite strip method detailed examples are paper.
provided in this paper; including intermediate results for
the constraint matrices derived in the companion paper. Appendix
The method is implemented in an existing open source
finite strip application, CUFSM [8]. Submatrices of RGD matrix are listed as follows:
Numerical examples using CUFSM are completed to
demonstrate the capabilities and unique challenges inher- 1 n^ h 
ent in constraining a finite strip solution for modal RU1 ¼ C5  sin að1Þ bð1Þ S^ 3  B^ 2 B1
1 S 2 þ ^
S 4 S1
kr
decomposition in a cross-section stability analysis. The   io
þ C ^3 B^ 2 B1 C2 þ C ^ 4 C1 ,
examples demonstrate that constrained FSM can success- 1
fully decompose the stability modes and eliminate modes
associated with chosen deformation fields, including those
1
of local plate buckling, and modes associated with shear RŪ ¼ S1 ,
and transverse extension (i.e., other modes). The ability to kr
switch on or off deformation modes through constraint 
equations in FSM, is not only useful for decomposing the 1 ~  
RUn ¼ C5  sin aðn1Þ bðn1Þ S~ 3  B~ 2 B1 ~
1 S2 þ S4 S1
solution and performing analysis on pure modes, e.g., pure kr
global, or pure distortional buckling modes, but also 
 
reduces the size of the eigen-stability problem that must be þ C ~3 B
~ 2 B1 C2 þ C ~ 4 C1 .
1
solved.
In this paper constrained FSM results are compared to RV is an n-order unit matrix.
analytical solutions, GBT solutions, and regular (uncon-
strained) FSM solutions. The constrained FSM solutions 1 n^ h 
RW1 ¼ ^ 2 B1 S2 þ S^ 4 S1
S5 þ cos að1Þ bð1Þ S^ 3  B 1
are in good agreement with the other methods, except that kr
  io
small differences are observed in how the various methods þ C ^3 B ^ 2 B1 C2 þ C^ 4 C1 ,
1
handle the Poisson effect. The question of the appropriate-
ness of beam theory (ignoring Poisson effect) versus plate
theory (including Poisson effect) in the stability of thin- 1
walled members is discussed in full detail. While uncon- RW̄ ¼  C1 ,
kr
strained FSM fully includes the Poisson effect and
transverse bending that occurs, analytical solutions, i.e., 
1 ~  
classic Euler buckling, ignore such two-dimensional RWn ¼ S5 þ cos a ðn1Þ ðn1Þ
b S~ 3  B~ 2 B1 ~
1 S2 þ S4 S1
approximations in favor of a one-dimensional beam theory kr

and no Poisson effect. A pragmatic solution where  
~ ~ 1 ~
þ C3  B2 B1 C2 þ C4 C1 ,
constrained finite strip models are additionally modified
to eliminate the Poisson effect, by forcing n to 0, yields
practically identical results with classical global mode 1 h ^    i
^ 3  B^ 2 B1 C2 C1 ,
solutions. Rh1 ¼ S3  B^ 2 B1
1 S 2 S1 þ C 1
kr
The examples in the paper highlight that, the definition
of different deformation/buckling modes has crucial
importance, since different definitions lead to somewhat 1 h ^  
^
 i
Rh2 ¼ S3  B^ 2 B1 ^ 1
1 S2 S1 þ C3  B2 B1 C2 C1 ,
different results. Currently, exact and widely adopted kr
ARTICLE IN PRESS
S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600 599

 
1  1
  1
 In the above expressions b(i) and a(i) denotes the width and
Rh̄ ¼ S3  B2 B1 S2 S1 þ C3  B2 B1 C2 C1 ,
kr angle of the ith strip, respectively, while:
rp
  kr ¼ ,
1 ~    a
Rhn1 ¼ S3  B~ 2 B1 S2 S1 þ ~
C 3  ~
B B
2 1
1
C 2 C 1 ,
kr 1 where r is the number of half-sine waves along the member
length, and a is the member length.
  In addition, the S, C and B matrices are summarized as
1 ~  
~3 B

follows. For notations, see Figs. 1 and 2 of the companion
Rhn ¼ S3  B~ 2 B1
1 S 2 S1 þ C ~ 2 B1 C2 C1 .
1
kr paper.
2 3
sin að2Þ sin að2Þ sin að1Þ sin að1Þ
6 ð1Þ
  0 0 0 0 07
6 Det2 b Det2 bð1Þ Det2 bð2Þ Det2 bð2Þ 7
6 7
6 sin að3Þ sin að3Þ sin að2Þ sin að2Þ 7
6 0   0 0 0 07
6 Det3 b ð2Þ
Det3 b ð2Þ
Det3 bð3Þ Det3 bð3Þ 7
S1 ¼ 6
6 : : : : : : :
7
: 7,
6 7
6 : : : : : : : : 7
6 7
6 7
4 0 0 0 0    05
0 0 0 0 0   

2 3
cos að2Þ cos að2Þ cos að1Þ cos að1Þ
6 ð1Þ
  0 0 0 0 07
6 Det2 b Det2 bð1Þ Det2 bð2Þ Det2 bð2Þ 7
6 7
6 cos að3Þ cos að3Þ cos að2Þ cos að2Þ 7
6 0   0 0 0 07
6 Det3 bð2Þ Det3 bð2Þ Det3 bð3Þ Det3 bð3Þ 7
C1 ¼ 6
6 : : : : : : :
7
: 7,
6 7
6 : : : : : : : : 7
6 7
6 7
4 0 0 0 0    05
0 0 0 0 0   

2 3
sð2Þ sð2Þ þ sð3Þ sð3Þ 0 0 0 0 0
6 0 sð3Þ sð3Þ þ sð4Þ sð4Þ 0 0 0 0 7
6 7
6 .. .. .. 7
6 7
6 : : . . . : : : 7
S2 ¼ 6
6 0 ði1Þ ði1Þ ðiÞ ðiÞ
7,
7
6 0 0 s s þs s 0 0 7
6 .. .. .. 7
6 7
4 : : : : . . . : 5
ðn3Þ ðn3Þ
0 0 0 0 0 s s þ sðn2Þ s ðn2Þ

2 3
cð2Þ cð2Þ þ cð3Þ cð3Þ 0 0 0 0 0
6 0 cð3Þ cð3Þ þ cð4Þ cð4Þ 0 0 0 0 7
6 7
6 .. .. .. 7
6 7
6 : : . . . : : : 7
C2 ¼ 6
6 0 ði1Þ ði1Þ
7,
7
6 0 0 c c þ cðiÞ cðiÞ 0 0 7
6 .. .. .. 7
6 7
4 : : : : . . . : 5
0 0 0 0 0 cðn3Þ cðn3Þ þ cðn2Þ cðn2Þ
ARTICLE IN PRESS
600 S. Ádány, B.W. Schafer / Thin-Walled Structures 44 (2006) 585–600

2 3 2 3
0 sð3Þ sð3Þ 0 0 0 0 0 0 cð3Þ cð3Þ 0 0 0 0 0
6 7 6 7
60 0 sð4Þ sð4Þ 0 0 0 0 7 60 0 cð4Þ cð4Þ 0 0 0 0 7
6 .. .. 7 6 .. .. 7
6 7 6 7
6 . . 7 6 . . 7
S3 ¼ 6
60
7;
7 C3 ¼ 6
60
7,
7
6 0 0 0 sðiÞ sðiÞ 0 0 7 6 0 0 0 cðiÞ cðiÞ 0 0 7
6 .. .. 7 6 .. .. 7
6 7 6 7
4 . . 5 4 . . 5
ðn2Þ ðn2Þ ðn2Þ
0 0 0 0 0 0 s s 0 0 0 0 0 0 c cðn2Þ

   
S^ 3 ¼ sð2Þ sð2Þ 0  0 ; ^ 3 ¼ cð2Þ
C cð2Þ 0  0 ,

   
S~ 3 ¼ 0    0 sðn2Þ sðn2Þ ; ~3 ¼ 0
C  0 cðn2Þ cðn2Þ ,

   
S^ 4 ¼ sð1Þ 0  0 ; ^ 4 ¼ cð1Þ
C 0  0 ,
   
S^ 5 ¼ sð1Þ sð1Þ 0  0 ; ^ 5 ¼ cð1Þ
C cð1Þ 0  0 ,
   
S~ 4 ¼ 0    0 sðn1Þ ; ~4 ¼ 0
C  0 cðn1Þ ,
   
S~ 5 ¼ 0    0 sðn1Þ sðn1Þ ; ~5 ¼ 0
C  0 cðn1Þ cðn1Þ ,
2 3
2bð2Þ þ 2bð3Þ bð3Þ 0 0 0 0
6 7
6 bð3Þ 2bð3Þ þ 2bð4Þ bð4Þ 0 0 0 7
6 .. .. .. 7
6 7
6 . . . 7
B1 ¼ 6
6
7,
7
6 0 0 bði1Þ 2bði1Þ þ 2bðiÞ bðiÞ 0 7
6 .. .. .. 7
6 7
4 . . . 5
ðn3Þ ðn3Þ
0 0 0 0 b 2b þ 2bðn2Þ
2 3
2bð3Þ bð3Þ 0 0 0 0
6 7
6 0 2bð4Þ bð4Þ 0 0 0 7
6 7
6 .. .. 7 derivation. Thin-walled Structures, in press, doi:10.1016/j.tws.
6 . . 7
B2 ¼ 6
6 0
7, 2006.03.013.
6 0 0 2b ðiÞ
b ðiÞ
0 7 7 [2] Silvestre N, Camotim D. First-order generalised beam theory for
6 .. .. 7 arbitrary orthotropic materials. Thin-Walled Structures 2002;40(9):
6 7
4 . . 5 755–89.
0 0 0 0 0 2bðn2Þ [3] Schafer BW. Cold-formed steel behavior and design: analytical and
numerical modeling of elements and members with longitudinal
    stiffeners. PhD dissertation. Cornell University, Ithaca, NY, 1997.
^ 2 ¼ bð2Þ 0    0 ; B
B ~ 2 ¼ 0    0 bð2Þ ,
[4] MathWorks (2005). MATLAB. Release 14 SP3. www.mathworks.com
with (last visited 29 September 2005).
[5] Ádány S, Schafer B. Buckling mode classification of members with
ðiÞ sin aðiÞ ðiÞ cos aðiÞ open thin-walled cross-sections. Proceedings of the fourth interna-
s ¼ ; c ¼ , tional conference on coupled instabilities in metal structures (CIMS
bðiÞ bðiÞ
‘04), Rome, Italy, September 27–29, 2004. p. 467–76.
Deti ¼ sin aðiÞ cos aði1Þ  sin aði1Þ cos aðiÞ , [6] Schafer BW, Ádány S. Understanding and classifying local, distor-
tional and global buckling in open thin-walled members. Proceedings
and b(i) and a(i) denotes the width and angle of the ith strip, of the annual technical session and meeting, structural stability
respectively. research council, Montreal, Quebec, Canada; May 2005. p. 27–46.
[7] Timoshenko SP, Gere JG. Theory of elastic stability. New York: Mc-
Graw Hill; 1936.
References [8] Schafer BW. CUFSM: elastic buckling analysis of thin-walled
members by the finite strip method. Version 3.0. www.ce.jhu.edu/
[1] Ádány S, Schafer BW. Buckling mode decomposition of single- bschafer/cufsm (last visited 29 September 2005).
branched open cross-section members via finite strip method:

You might also like