Thin-Walled Structures (xxxx) xxxxxxxx

Contents lists available at ScienceDirect

Thin-Walled Structures
journal homepage: www.elsevier.com/locate/tws

Full length article

Constrained shell Finite Element Method, Part 2: application to linear

buckling analysis of thin-walled members

Sndor dny , Dvid Visy, Rbert Nagy
Budapest University of Technology and Economics, Department of Structural Mechanics, Megyetem rkp. 3, 1111 Budapest, Hungary

A B S T R A C T

In this paper a novel method is employed for the buckling analysis of thin-walled members. The method is
basically a shell nite element method, but constraints are applied which enforce the thin-walled member to
deform in accordance with specic mechanical criteria, e.g., to force the member to buckle in exural, or lateral-
torsional or distortional mode. The method is essentially similar to the constrained nite strip method, but the
trigonometric longitudinal shape functions of the nite strip method are replaced by polynomial longitudinal
shape functions, and longitudinal discretization is used, which transform the nite strip into multiple nite
elements, that is why the new method can readily be termed as constrained (shell) nite element method. In the
companion to this paper a band of nite elements is discussed in detail, where band is a segment of the member
with one single nite element longitudinally. In this paper the constraining procedure is applied on thin-walled
members discretized both in the transverse and longitudinal direction. The possible base systems for the various
deformation spaces are demonstrated here, as well as numerous buckling examples are provided to illustrate the
potential of the proposed method.

1. Introduction
Thin-walled members have complicated stability behaviour. Due to
the high slenderness, stability is the governing phenomenon in many
cases. If a thin-walled member is subjected to longitudinal compressive
stresses, three characteristic buckling classes are usually distinguished:
global, distortional, and local-plate buckling. When the eect of shear
stresses is dominant, shear buckling may also occur. Transverse
compressive stresses might lead to instability frequently referred to as
web crippling. In practical situations these buckling modes rarely
appear in isolation, but they are interacted with each other. Still,
current design approaches rst separate these phenomena, and deter-
mine a capacity to each of the potential buckling, and then check the
possible interaction of them. This design approach appears also in
current thin-walled design standards, e.g. [1,2].
Capacity prediction hence requires the critical loads associated with
the various buckling modes. Nowadays, critical load calculation for
thin-walled members can readily be accomplished by some numerical
methods, the most widely used ones being the shell nite element
method (FEM), the generalized beam theory (GBT), and the nite strip
method (FSM).
FEM, by using shell nite elements, is general and can be used to
analyse almost any thin-walled member. However, a general FEM is not
able to separate the various buckling modes, which often makes it
inecient to directly use it in standard capacity calculations. GBT has
shown that buckling deformations may be formally treated in a modal
nature that mechanically separates global, distortional, local, and other
deformations, see e.g. [35]. This formal separation is integral to GBT,
and allows pure buckling mode calculations and measurements of
modal participation in coupled modes. FSM is based on the work of
Cheung [6], but popularized by Hancock [7] who provided the
organizing thrust of today's member design, which later evolved into
the Direct Strength Method (DSM) [8]. Hancock introduced the notion
of the signature curve, from which quasi-pure buckling modes and
associated loads could be determined, at least for typical design. The
mechanical criteria embedded in GBT led to the development of a
special version of FSM, the constrained Finite Strip Method (cFSM), see
[913]. cFSM possesses the ability of modal decomposition as well as
mode identication in a manner similar to GBT.
Both GBT and cFSM is easily available, since they are implemented
into the free-to-use programs GBTUL [14] and CUFSM [15]. It is easy to
understand, however, that either the method or its available imple-
mentation is not general enough. For example, the analysed member
has to be prismatic, no holes are allowed in the members, and there are
restrictions w.r.t boundary conditions as well as loading (at least in the
available software implementations). Recent works aimed at partially

Corresponding author.
E-mail address: sadany@epito.bme.hu (S. dny).

http://dx.doi.org/10.1016/j.tws.2017.01.022
Received 6 October 2016; Received in revised form 10 January 2017; Accepted 12 January 2017
0263-8231/ 2017 Elsevier Ltd. All rights reserved.

Please cite this article as: DNY, S., Thin-Walled Structures (2017), http://dx.doi.org/10.1016/j.tws.2017.01.022
S. dny et al.
removing these limitations, but these attempts have not led to a simple
and easily applicable general method with modal features, as discussed
in a companion paper [16].
In this paper a novel method is used. The proposed method follows
the logic of cFSM, however, the longitudinal shape functions are
changed and applied together with a longitudinal discretization.
Thus, strips are transformed into multiple shell nite elements, there-
fore, the new method can readily be termed as constrained nite
element method, abbreviated as cFEM.
The cFEM has rst been reported in [1719]. The cFEM method is
using a special shell nite element, specically designed for the
method. The most basic feature of this special rectangular shell element
is that it distinguishes the so-called longitudinal and transverse direc-
tions. Otherwise, the shell element uses classic polynomial shape
functions, though in an unusual combination. The element, together
with the derivation of the interpolation functions is presented in detail
in [20]. The elementary stiness matrices (i.e., elastic and geometric)
can be determined analytically. As earlier studies highlighted the
importance of some details of the derivations, the stiness matrices
are derived in various options, as discussed in [21]. Namely: the
second-order eect of the various strain terms can be switched on or
o, as well as it is optional whether the through-thickness variation of
the stresses/strains is considered or disregarded.
In [20] it is proved that the mechanical criteria, which are necessary
for modal decomposition, can exactly be satised by the shape
functions of the special shell element. Furthermore, in [16] the
constraining procedure is discussed for members discretized into
multiple nite elements. More precisely: in [16] multiple nite
elements are assumed along the cross-section line, but only one single
nite element is assumed longitudinally, which in other words means
that one band of nite elements is discussed. Though one band of nite
elements alone is not enough to solve practical problems, the discussion
of a single band is important in order to understand the constraining
procedure and the resulted base systems, and it helps to construct the
modal base systems for a real member consisted of multiple bands. In
fact, real members with multiple bands are in the main focus of this
paper.
In this paper rst the most important features of the new shell nite
element are summarized, based on [20]. Also the constraining proce-
dure in case of one single band of shell nite element is briey
summarized. Then two alternatives are presented for the construction
of the modal base system of a member with multiple bands. Finally,
numerous practical examples are shown for the constrained buckling
analysis of thin-walled members.
As the numerical examples demonstrate, the here presented cFEM is
much more general than any of the existing modal decomposition
methods. Since it is based on a shell nite model, a variety of
engineering problems can be solved. The element is rectangular, which
means a certain limitation, but otherwise the method is general.
Practically there are no limits for the loading and boundary conditions.
Linear or non-linear static or even dynamic analyses can be performed,
though in this paper only elastic linear buckling problems are shown.
Holes can easily be handled, too, once they t into the rectangular
discretization. Though the problem of holes is not the topic of this
actual paper, a few numerical examples with holes are also presented.
Finally, though here the member to be constrained is assumed to be
prismatic, piece-wise prismatic members can also be handled, by
joining prismatic members together. Again, this question is not
discussed here, only numerical examples demonstrate this ability of
the proposed cFEM.
2. Basics of cFEM
2.1. Shell nite element for cFEM
A sample thin-walled member is shown on Fig. 1. It is discretized
2
Thin-Walled Structures (xxxx) xxxxxxxx
into shell nite elements. Local and global coordinate systems as well as
some nodal displacements are also shown. The goal of the constrained
nite element method is model the thin-walled member by shell nite
elements, and to be able perform modal decomposition via enforcing
the member to deform in accordance with pre-dened mechanical
criteria. It is also essential to satisfy these criteria exactly, i.e., not only
in specic locations (e.g., at some nodes), but all over the member: at
any x and y locations within any nite element. As concluded in
[16,20], this requires certain basic features from the interpolation
functions, that is why the otherwise classic shape functions are used in a
relatively unusual combination.
The derivation of the interpolation functions are discussed in [20],
where the displacement functions are given, too. The nodal displace-
ment degrees of freedom (DOF) are illustrated in Fig. 2. As can be seen,
the proposed element has 30 DOF: each corner node has 7 DOF (1 for u,
2 for v, and 4 for w), while there are two additional nodes at (x,y)=(a/
2,0) and (x,y)=(a/2,b) with one DOF per node for the u displacement.
2.2. cFEM in general
When a member is constrained, it is enforced to deform in
accordance with some mechanical criteria. These criteria are identical
to those applied in cFSM [e.g., 1213], as summarized in the table of
Fig. 3. As can be seen from Fig. 3, mostly displacement derivatives (i.e.,
strains) are used, and the important question is whether the given strain
is zero or not (Y or N, respectively). In calculating the strains, all the u, v
and w functions are interpreted at the middle surface of the plates, i.e.,
at z=0. In row transv. eq. it is given whether the cross-section
equilibrium (in the transverse direction) is satised or not (Y or N,
respectively).
The deformation spaces dened in Fig. 3 are described in detail in
[12,13]. It might be interesting to mention here that the mode spaces
are separated into primary and secondary mode spaces. Primary modes
are those deformations which are completely dened by the degrees of
freedom (DOF) associated with the main nodes only, i.e., those nodes at
the junction or end of the at plates comprising the section. Secondary
modes are dened by the DOF of the sub-nodes, i.e., those nodes within
a at plate discretized into multiple strips or elements.
It is to mention that though the table of Fig. 3 (mostly) clearly
dene the listed deformation spaces, these are only the displacement
derivative is equal to zero type criteria that can be transformed into
equations. Therefore, only these so-called null criteria can and will
directly be used in the constraining procedure, while the displacement
derivative is not equal to zero type criteria are used only indirectly in
the construction of the deformation spaces.
As it is shown in [16], and will further be discussed here, the
mechanical criteria can be transformed into constraint matrices. The
application of the constraint matrix enforces to full certain relation-
ship between various nodal degrees of freedom. Another view of
constraint matrix is that the column vectors of the matrix are the
modal base vectors of the displacement eld that is represented by the
constraint matrix.
The d displacement vector may be constrained to any modal
deformation space (i.e., to a dM modal displacement vector) via:
d = RMdM (1)
where RM is a so-called constraint matrix, the derivation of which can
be found in [16] for a single band of nite elements, and will be
presented in this paper for more general cases. M might be G, D, L, S or
T, or, in fact, M might mean any combination of base vectors from any
spaces.
By using the constraint matrix, solution in a reduced, specic
deformation space is possible. For example, rst or second-order static
analysis can be done, the regular (i.e., unconstrained) problem of which
takes the following form:
S. dny et al. Thin-Walled Structures (xxxx) xxxxxxxx

Fig. 1. FEM discretization.

Fig. 2. Nodal degrees of freedom of the shell element.

Kd = f (2) d=[ R G RD RL R S RT ]c (5)

where K is the stiness matrix, d is the displacement vector and f is the where c now provides the deformations within each space, from which
load vector. If deformations are constrained, the problem can be written pi participation of an individual base vector or pM participation of a
as follows: deformation space M can be determined.
KRMdM = f (3)
2.3. Constraining a single nite element
from which

RM TKRMdM = RM Tf KMdM = fM
(4)
where KM and fM can be regarded as the modal version of the stiness
matrix and load vector, respectively, constrained into the given
deformation space.
Modal identication, i.e. categorization of a general deformation
into the modal spaces, is also possible, due to the fact that G+D+L+S
+T spans the entire FEM space. As such, the RGDLST constraint matrix
represents an alternative basis for the FEM space, in which deforma-
tions are categorized. The transformation of displacement vector d may
be expressed as:
To apply the constraints within a nite element, the displacement
functions (see e.g. [20]) must be substituted into the strain criteria of
Fig. 3. The procedure is discussed in [20]. Important conclusions are
that (i) it is possible to have zero strain over the whole element, and (ii)
any of these criteria can be expressed by relationships in between the
nodal DOF. The resulted DOF relationships are summarized in [16].
It is to note, that if relationship exists between the DOF, this also
means that the eective number of DOF is reduced. That is why the
application of constraints reduces the size of the problem, e.g., the
constrained displacement vector dM is smaller than the unconstrained
one. As a general rule: the more strains are forced to be zero, the

Fig. 3. Mechanical criteria for constraining.

3
S. dny et al.
Fig. 4. Longitudinal distribution of the base functions for a single band (a) FEM-like, u
displacement (b) FEM-like, v and w displacements (c) Fourier-like, u displacement (d)
Fourier-like, v and w displacements.
smaller the size of the problem. That is why the size of the constrained
problem might be drastically smaller than that of the unconstrained
one.
2.4. The constraining procedure
For the construction of the base system (i.e., for nding the
constraint matrix RM) for any deformation space, three kinds of criteria
are employed: null, independence and orthogonality. For any space or
sub-space some strain components are set to zero, these dene the null
criteria. These criteria are given in Fig. 3. Since transverse equilibrium
can also be interpreted as the unbalanced nodal forces/moments are
zero, transverse equilibrium is also treated as a special null criterion.
Furthermore, the spaces must be linearly independent of each other,
leading to the independence criteria, while the base functions within a
sub-space must be orthogonal, leading to the orthogonality criteria.
The criteria are enforced on the displacement vector in matrix form.
Matrices for null criteria are denoted by Z and an appropriate subscript.
Matrices for the orthogonality criteria are denoted by O and an
appropriate subscript, where the subscript denotes which displacement
or strain component distribution is orthogonal. The orthogonality
criteria are interpreted for the cross-section.
All Z matrices are constructed such that the column number is equal
to the DOF number, therefore, enforcement of a null criterion on a d
displacement vector simply requires Zd=0. The O matrices are square
matrices, therefore, enforcement of an orthogonality criterion for two
non-identical displacement vectors requires drTOds=0. Finally, the
independence criterion is enforced by requiring the scalar product of
two displacement vectors to be zero, i.e., drTds=0.
The order of constructing the primary mode spaces is not arbitrary,
construction of some of the spaces requires knowledge of other spaces,
in order to be able to apply independence or orthogonality criteria.
The constraining procedure, therefore, rst of all means the
compilation of various Z and O matrices. Then the equations like
Zd=0 must be solved, which mathematically means to nd the null
space of Z. This leads to the given deformation space, but since
4
Thin-Walled Structures (xxxx) xxxxxxxx
deformation spaces are typically multidimensional not to specic base
vectors, since various base vectors are possible within a space. The
selection of an appropriate base system is hence the last step of the
procedure.
2.5. Constraining a band of nite elements
In [16] the constraining procedure is applied for a band of nite
elements, where band is a segment of the member with multiple
elements along the cross-section, but with one single nite element
longitudinally. The most important conclusions are as follows.
It is demonstrated that the proposed constraining procedure means
a transformation within the displacement-deformation eld dened by
the nodal degrees of freedom of the shell model. In other words, the
sum of G, D, L, S and T spans the same displacement-deformation eld
dened by the unconstrained nite element model.
The base vectors/functions of any space can be constructed by
taking the transverse distribution of the characteristic displacements and
adding appropriate longitudinal distribution. The transverse distribu-
tion can also be termed as cross-section mode, and it is found that the
cross-section modes of cFEM can be identical to those of the constrained
nite strip method (cFSM). When the longitudinal distribution is added,
the values of the characteristic displacements must be dened at the
characteristic locations. Two alternatives are mentioned in [16]: FEM-
like and Fourier-like. In case of FEM-like base system the character-
istic displacement is set to unit at one specic location, while set to zero
at all the other locations. In case of Fourier-like base system the
longitudinal distribution of the characteristic displacement contains a
constant term, a linear term, then one, two, three, etc. waves.
It is to observe that the u warping displacement is interpolated by a
quadratic function longitudinally, while the v and w transverse
displacements are interpolated by cubic functions longitudinally. As a
consequence, the FEM-like and Fourier-like longitudinal base systems
are dierent for the u and for the v and w displacements, as illustrated
by Fig. 4 for a single band.
It is also concluded from [16], that rigid-body member displace-
ments (i.e., when the whole member is displaced as a rigid body) satisfy
all the null criteria. So, it is a question of decision how to treat or
classify the rigid-body member displacements.
3. Constraining a member
3.1. General
There are (at least) two approaches. One approach is to follow the
same constraining procedure that is applied for one single band of nite
elements, as discussed in companion paper [16], but now applying the
procedure for all the nite elements (from multiple bands) that are to
be constrained. The other approach is to perform the constraining from
band to band, and then to construct the base vectors/functions of the
member from the base vectors/functions of the bands. Both of these
approaches will be illustrated here. Both of these approaches have
potential advantages and disadvantages.
3.2. Direct construction of the base systems for a member
The procedure is discussed in detail in [16], here only illustrated for
the global axial and global bending spaces.
In case of global axial (GA) space the displacement vectors must
satisfy all the null criteria given by Fig. 3. Mathematically, the criteria
can be summarized into the following matrix equation:
S. dny et al.
Zvy
Z uyvx
Z
eq
Z ux
ZGAd GA = Zw2y2 d GA = 0
Zw2xy
Zw2x2
Z uy
Zvx (6)
where the Z matrices correspond to various null criteria (see the rows of
the table of Fig. 3) necessary for the GA space. It is to observe that this
matrix equation is formally identical to that used in the constraining of
a single band, see [16]. However, the actual content of the blocks of
ZGA is dierent, as illustrated below.
According to the above equation, any dGA displacement vector that
lies in the GA space is orthogonal to ZGA. The GA space, therefore, is the
null space of ZGA. This can formally be expressed as:
R GA=null (ZGA) (7)
where RGA is termed here as the constraint matrix for the GA space. The
column vectors of RGA, therefore, the base vectors for the GA space. It
is important to emphasize here that the above equation unambiguously
denes the space, but since the space is multidimensional not the
base vectors, i.e., various base vectors are possible to nd within the
space.
In case of global bending (GB) space the displacement vectors must
satisfy all the null criteria given by Fig. 3. Mathematically, the criteria
can be summarized into the following matrix equation:
Zvy
Z uyvx
Z
= d = 0
eq
ZGBd GB
Zw2y2 GB
Zw2xy
R TO
GA u (8)
where the Z matrices correspond to various null criteria (see the rows of
the table of Fig. 3) specic for the GB space. In addition, the equation
enforces that the global bending space must be orthogonal to the global
axial space. From this equation the RGB constraint matrix for the GB
space can be determined as the null space of ZGB. It is to observe that
this matrix equation is formally identical to that used in the constrain-
ing of a single band, see [16]. However, the actual content of the blocks
of ZGB is dierent, as illustrated below.
In the above equations each block of a ZM matrix enforces a certain
null criterion. In the compilation of these blocks the procedure applied
in [16] can be followed. Here only one sample is briey shown, as
follows. For example, the null longitudinal strain criterion requires that
u /x = 0 . According to [16], this criterion is satised within a nite
element if (and only if):
u11 = u 21 = u31 and u13 = u 23 = u33 (9)
If there are multiple nite elements, the above DOF relationships
must be enforced for all the nite elements. Applying it for the (k)-th
nite element, and assuming that the nite element nodes of the (k)-th
element are denoted as k11, k13, k21, k23, k31 and k33, this can be
written by the following 4 equations:
uk11 = uk 21 and uk11 = uk 31 and uk13 = uk 23 and uk13 = uk 33 (10)
Utilizing that the corresponding local u and global U displacements
are equal to one another, the same 4 equations can also be expressed by
the global degrees of freedom:
Uk11 = Uk 21 and Uk11 = Uk 31 and Uk13 = Uk 23 and Uk13 = Uk 33 (11)
Thin-Walled Structures (xxxx) xxxxxxxx
This can be expressed in matrix form as follows:
(k )
Z ux d = 0 (12)
where
Applying the same criterion for all the nite elements of the
member leads to:
Z uxd = 0 (14)
where
Z (1)
ux
Z ux (2)
Z ux = : (k )
Z ux
:
(p )
Z ux (15)
As can be understood, the number of columns of Zux is equal to the
number of DOF if the member, while the number of rows of Zux is equal
to 4 times the number of nite elements, i.e., 4p if p is the number of
nite elements in the member.
Other spaces can be constructed similarly, by appropriately extend-
ing the equations presented in [16]. It is to emphasize that although the
above equations are formally identical to the corresponding equations
of [16], here the Z matrices are typically signicantly much larger
matrices (e.g., the row and column number of any ZM for the member is
roughly mband times the row and column number of the corresponding
ZM for a band, if mband is the number of bands in the member). That is
why though the above-presented direct construction of the base system
for the whole member is theoretically possible, but practically (mostly)
inecient, because the calculation of the null space of large matrices is
computationally not eective.
It is to observe that the base vectors/functions of any space can be
constructed by taking the transverse distribution of the characteristic
displacements and adding appropriate longitudinal distribution. When
the longitudinal distribution is added, the values of the characteristic
displacements must be dened at the characteristic locations. For
example, in case of GA the characteristic displacement is uniform
warping of a cross-section (i.e., longitudinal translation of a transver-
sally rigid cross-section), while the characteristic locations are at x=0,
x=a/2 and x=a of any nite element. In case of GB space the
characteristic locations are at the edges of the nite elements, i.e., at
x=0 and x=a, while the characteristic displacements are (rigid) cross-
section translations (in Y and Z) and (rigid) cross-section rotation
(around Y and Z).
3.3. Construction of base systems for a member from bands
Typically a shell nite element model of thin-walled member
requires discretization both along the cross-section and longitudinally.
If the special nite element of cFEM is applied, which is a rectangular
element, the model is necessarily built up from bands. When the
member is constrained, a possible approach is to construct the modal
base systems band by band, then compiling the base systems of the
member from those of the bands. This approach might be ecient if the
base functions of the bands can easily be connected.
In [16] the construction of the base systems for a band is presented
in detail. Any modal base function of a band is combined from cross-
section displacements and longitudinal distributions. As far as long-
itudinal distributions are concerned, modal base functions can easily be
connected if the base functions of the connecting bands are FEM-like,
see Fig. 4(a) and (b). On the other hand, the connection of bands is
problematic if the base functions of the connecting bands are Fourier-
like, see Fig. 4(c) and (d). Hence, if the modal base systems of the whole
member are compiled from those of the bands, it is more practical to
use FEM-like base systems for the bands.
5
S. dny et al.
w_distr = 1 at X = 0 w_distr = 1 at X = a
w_distr = 1 at X = 0 w_distr = 1 at X = a
Fig. 5. Compilation of base functions from a symmetric distortional cross-section mode.
As far as cross-section displacements are concerned, they are
necessarily compatible if the cross-sections of the connecting bands
are identical, i.e., the member is prismatic. (It is to note that bands with
dierent cross-sections can also be connected, but this topic is not
discussed here.)
The compilation of the modal base system is illustrated by Fig. 5,
where the symmetric distortional mode is used as characteristic
displacement for a simple member having only two bands. The length
of a band is a, hence the total member length is 2a. There are three
characteristic locations, namely: X=0, X=a and X=2a. At any location
either the characteristic displacement (w_distr, in this case the local
transverse displacements along the cross-section line) or its derivative
w.r.t. x (w_distr, in this case the warping distribution along the cross-
section line) is set to unit or zero. Therefore, there are 6 native base
functions for the symmetric distortional space. (It is to emphasize that
in Fig. 5 the shown member is discretized into two bands, that is there
are only two nite elements longitudinally. Also, for the distortional
modes cross-section discretization is not necessary at all. So the shown
mesh does not correspond to the nite element discretization. The ne
mesh is used solely to visualize the deformations).
3.4. Comments on the deformation spaces of a member
In accordance with the logic of cFEM, the deformation spaces must
satisfy certain criteria, but since the spaces are typically multidimen-
sional, the space can be determined by various base vectors/functions.
Theoretically, therefore, the nal base systems of the member are not
dependent on the way how the base systems are constructed. If the base
systems of the member are constructed directly, as shown in Section
3.2, an initial base system is achieved during the determination of the
null space of the actual Z matrix. If the base systems of the member are
constructed from those of the bands, as discussed in Section 3.3, an
initial base system is determined as combined from those of the bands.
6
Thin-Walled Structures (xxxx) xxxxxxxx
w_distr = 1 at X = 2a
w_distr = 1 at X = 2a
However, independently from how the initial base system is deter-
mined, this initial base system can be transformed into various other
base systems.
Though the direct and band-based construction of the base functions
for a member with multiple bands is theoretically equivalent, both
might have advantages and disadvantages. The main practical dier-
ence is that even if in either case equations like Eq. (7) must be solved
direct approach requires the solution for the null space of a single, but
large size matrix (to get a specic deformation space), while the band-
based approach requires the solution of small size equations, but
multiple times. Since the null space determination is a relatively
computation-demanding process, the band-based approach is probably
more ecient and more stable numerically. (Other dierences exist,
too, which might be important in specic cases, e.g., if there are holes
in the member, or if the member to analyse is not prismatic. These
questions are not discussed here.)
If there are multiple bands in the nite element model of a prismatic
member (as typical in practical situations), the cross-section modes
(i.e., characteristic displacements) are dependent only on the cross-
section geometry, and also the characteristic locations are dened,
namely: at x=0, x=a/2 and x=a positions of the nite elements for u
displacements, while at x=0 and x=a positions for v and w displace-
ments. Still, various longitudinal distributions are possible. There seems
to be at least two practical ways again: FEM-like and Fourier-like. In
case of a FEM-like base function the characteristic displacement or
displacement derivative is set to unit at one specic location, while set
to zero at all the other locations. In case of Fourier-like base system
there is base function with a constant longitudinal distribution of the
characteristic displacement, then there is another one with linear
distribution, then there are other base functions with one, two, three,
etc. waves (up to as many waves as the number of bands and the
elements longitudinal shape functions allow). The two kinds of base
systems are illustrated in Fig. 6.
S. dny et al.
Fig. 6. Longitudinal distribution of the base functions for a member built up from multiple bands (a) FEM-like, for u displacement (b) FEM-like, for v and w displacements (c) Fourier-like.
Even if there are multiple bands in the model, rigid-body member
displacements are possible and needs special consideration. It is easy to
understand that when null criteria are enforced, rigid-body member
displacements are always appear among the solutions, since any
displacement derivative is zero if the member displaces as a rigid body.
So, it is a question of decision how to treat or classify the rigid-body
member displacements. There seems to be two logical approaches, as
follows.
Approach #1: Rigid-body displacements can be considered as a
separate space, i.e., as not part of any of the spaces dened by the
table of Fig. 3. In fact, this approach is totally consistent with the
criteria given by Fig. 3, since the displacement derivative is not
equal to zero type criteria are never satised by any rigid-body
member displacements.
7
Thin-Walled Structures (xxxx) xxxxxxxx
Approach #2: Rigid-body displacements can also be assigned to
some of the deformation spaces. The trivial candidates are the global
spaces, since rigid-body member displacement certainly means the
displacement of the whole member (and such, can be regarded as
global). Moreover, for each rigid-body member displacement it is
easy to nd the global space is characterized by displacements that
are similar to the given rigid-body member displacements. Namely:
rigid-body translation of the member in X direction can be assigned
to GA, rigid-body translation of the member in Y and rigid-body
rotation of the member around Z can be assigned to one of the GB
space, while rigid-body translation of the member along Z and rigid-
body rotation of the member around Y can be assigned to the other
GB space, nally rigid-body rotation of the member around X can be
assigned to the GT space.
S. dny et al. Thin-Walled Structures (xxxx) xxxxxxxx

Table 1
Deformation space dimensions.

cFEM space size (ncs=no of c.s. modes, mband=no of bands) ncs in case of unbranched open cross-sections

global axial, GA (2mband+1)ncs (incl. 1 rigid) 1

global bending, GB (2mband+2)ncs (incl. 4 rigid) 2
global torsional, GT (2mband+2)ncs (incl. 1 rigid) 1
distortional, D (2mband+2)ncs ni 2
primary local, LP (2mband+2)ncs ni +4
primary shear bending, SBw or SBt (2mband+1)ncs 2
primary shear torsional, STw or STt (2mband+1)ncs 1
primary shear distortional, SDw or SDt (2mband+1)ncs ni 2
primary shear other, SCw or SCt (2mband+1)ncs 0
primary transverse extension, TP (2mband+2)ncs ni +1
secondary local, LS (2mband+2)ncs 2ns
secondary shear, SS (2mband+1)ncs ns
secondary transverse extension, TS (2mband+2)ncs ns

It is to observe that the base functions of the cFEM deformation
spaces can be constructed as a product of cross-section modes and
longitudinal distributions. Moreover, the cross-section modes of cFEM
can be identical to those of cFSM, so, a possible practical way for the
construction of cFEM base vectors/functions is to take the cross-section
modes from cFSM, and then to add the longitudinal distributions. These
longitudinal distributions can be dened by innite number of ways.
FEM-like distributions are advantageous to use since (i) these can most
easily be constructed from the base vectors/functions of the bands, and
(ii) these are totally compatible with the basic idea of the nite element
method. Fourier-like distributions can be advantageous, too, since (i)
these provide with a natural way to handle rigid-body member
displacements, and (ii) these make possible to establish signature
curve for the given problem, i.e., an output similar to that of FSM or
cFSM. It is to note, however, that the examples presented in Section 4 of
this paper are all calculated by using FEM-like modal shape functions
constructed from the base vectors/functions of the bands.
By applying the above summarized constraining procedure, all the
deformation spaces can be constructed, i.e, modal base systems can be
found. The number of base vectors/functions denes the size or
dimension) of the given space, and the sum of the space dimensions
must be equal to the total number of degrees of freedom of the shell
nite element model, which can illustrate that the modal deformation
spaces span the same displacement-deformation eld spanned by the
(unconstrained) nite element model. The space dimensions are
summarized in Table 1, given as the function of the corresponding
cross-section mode numbers (ncs) and number of bands (mband). The
number of cross-section modes for a given space might be dependent on
the cross-section topology (in case of primary spaces) and the cross-
section discretization (in case of secondary spaces). In case of simple
unbranched open cross-sections (with at least three plates), however,
simple rules can be given.
The number of cross-section modes are also given in Table 1 for
unbranched open cross-sections (e.g., unlipped or lipped channel
section), where ni is the number of internal main nodes of the cross-
section (e.g., 4 in case of a lipped channel section), and ns is the number
of sub-nodes, i.e., intermediate nodes in a at part of the cross-section.
Altogether, the number of nodes in a cross-section is (ni+ns+2),
having ni internal main nodes, 2 external main nodes and ns sub-nodes.
Having a regular nite element mesh as shown in Fig. 1, therefore, the
total number of DOF in the shell model is: 7(mband+1)(ni+ns+2)
+ mband(ni+ns+2), where the rst term corresponds to the DOF at
nite element corner nodes, while the second term corresponds to the
DOF at the intermediate nodes of the shell elements. This total DOF
number is equal to the sum of the dimensions of the modal deformation
spaces listed in Table 1.
4. Application for stability analysis
4.1. Stability analysis
When the shell FEM is applied to solve linear buckling problems to
get critical loads and buckling shapes, rst the elastic and geometric
stiness matrices must be constructed for an element. This can be
completed by following conventional FEM steps, by considering the 2D
generalized Hooke's law (in case of the elastic stiness matrix) and by
considering the second-order strain terms (in case of the geometric
stiness matrix). From the local stiness matrices the member's (global)
stiness matrices (elastic and geometric, Ke and Kg) can be compiled
by transformation to global coordinates and assembly.
The geometric stiness matrix is dependent on the distribution of
the internal forces/stresses. In a general case the distribution of stresses
is not obvious, but can be determined by a (rst-order) static analysis.
Once the stiness matrices are established, the linear stability analysis
is formulated mathematically as a generalized eigen-value problem,
namely:
K e K g = 0 (16)
with
=diag<1 2 3 nDOF > and =[123 nDOF] (17)
where i is the critical load multiplier and i is the associated buckling
shape, and nDOF denotes the number of degrees of freedom.
When the constraints are enforced, the displacement vector is
expressed by modal coordinates, according to Eq. (1). This equation
can also be applied to any buckling shape , since the buckling shape
itself is a displacement vector. Therefore:
= RM M (18)
By substituting Eq. (18) to Eq. (16), then multiplying (from the left)
by the transpose of the constraint matrix, the eigen-value problem can
be written as:
RM TK eRM M RM TK gRM M = 0 (19)
which is another generalized eigen-value problem, given in the reduced
M deformation space:
K eM M K gM M = 0 (20)
Solving the above equation leads to the modal eigen-vectors, from
which the buckled shapes can be calculated by solving Eq. (18).
4.2. Calculation options

The derivation of the elastic and geometric stiness matrices is

discussed in [21]. Here only a few practically important characteristics

8
S. dny et al.
of the derivations are highlighted.
In the derivation of the geometric stiness matrix of a shell element
the second-order strain terms are necessary to consider. These terms are
collected in the Green-Lagrange strain matrix. Since the given nite
element is a (thin) shell element, three in-plane stress/strain terms are
considered: longitudinal and transverse normal stress/strain, and shear
stress/strain. The second-order strain terms therefore are as follows:
u 2 v 2 w 2
( x ) ( x ) ( x )
2 v 2 w 2
II = u
1
y y
2 y
u u + u u v v
+
v v w w w w
+ y x
x y y x x y y x x y (21)
Altogether there are 9 second-order strain terms. It is usual to
disregard certain terms, based on their probable importance in a given
application. In the developed cFEM application all the 9 terms are
considered optionally, i.e., any of the 9 terms can be switched on or o
during the determination of the geometric stiness matrix. This feature
of the developed cFEM application is not closely related to the modal
decomposition, but it is found to be useful to be able to investigate the
eect of the various stress/strain terms separately of the others, as will
be demonstrated later by the numerical examples.
Both the elastic and geometric matrix of the nite element requires
integration along all the three coordinate axes, hence, through the
thickness, too (along the z axis). All these integrations can be completed
analytically. Since the shell element is assumed to be thin, it is usual to
slightly simplify the through-thickness integration as follows: in the
derivation of the elastic stiness matrix the through-thickness stress/
strain variation is usually exactly considered, however, in the deriva-
tion of the geometric stiness matrix the through-thickness stress/strain
variation is usually disregarded which means that the average (i.e.,
middle plane) stress/strain is used in considering the second-order
eect. Though this means certain discrepancy, in normal circumstances
it has negligible eect on the results. Earlier studies showed, however,
that when constraints are applied, the exact or simplied through-
thickness stress/strain variation might have visible eect on the
buckling solutions. [22,23] Therefore, both the elastic and geometric
stiness matrix is derived with considering and with disregarding the
through-thickness stress-strain variation.
4.3. Numerical examples
4.3.1. Example 1
Example #1 is presented in Fig. 7 where the pure global and local
buckling solutions of a simple column problem are shown. The
considered column is 500 mm long, doubly symmetrical I-shaped cross
section (4010 mm anges, 2004 mm web). The column is sup-
ported at its end sections, where all the nodes are supported against
transverse translation. This support can be regarded as locally and
globally pinned. The material is isotropic steel (E=210 GPa, =0.3).
The member is loaded by a concentric axial force, uniformly distributed
over the end cross-sections (represented by nodal forces, as required by
FEM technique). The global buckling solutions are in accordance with
the expectations: the rst mode is a minor-axis exural buckling with
one single longitudinal wave, the second mode is a pure torsional mode
with one single wave, while higher modes consist of multiple waves
along the member length. The rst three modes of pure local-plate
buckling are also shown. These results can easily be compared to other
methods. In case of global modes either analytical or numerical
solutions are available. In case of local modes, numerical methods
can be applied, e.g., GBTUL and CUFSM [14,15]. The comparison of the
results shows negligible dierence between the various methods.
4.3.2. Example 2
Example #2 is to demonstrate that various loading can be applied.
Thin-Walled Structures (xxxx) xxxxxxxx
In fact, the loading can be arbitrary as far as it can be handled by shell
nite element analysis, which means that there is no real practical
limitation for the loading. Example #2 is similar to Example #1,
however, the member is loaded by a transverse force (instead of the
axial force). The load is a concentrated vertical force at the middle of
the beam, acting at either the top ange, or the centroid, or the bottom
ange. Note, in all the cases the load is vertical, acting in the plane of
the symmetry of the cross-section. As well known, the load position has
signicant eect on the lateral-torsional (global) critical moment. The
results show that the load application point has pronounced inuence
on the local behaviour and associated critical values, too, see Fig. 8.
4.3.3. Example 3
Since cFEM is based on a shell nite element, various support
conditions can easily be modelled. Example #3 is similar to Example #2
(i.e., simply supported beam with a concentrated transverse force), but
the end supports are somewhat released. The end-sections are sup-
ported, but only the web-to-ange junction points are supported against
lateral translations, and the bottom web-to-ange junction is supported
against vertical translation. This change in the supports has little eect
on the global modes, but has signicant eect on the local-plate modes
and on the associated critical load values. Fig. 9 shows that as the
vertical movement of the top ange is not prevented, the whole web
buckles. The critical values are much smaller compared to the classic
fork-support case of Example #2. It is also to observe that the critical
values are much less inuenced by the load position (compared to
Example #2).
4.3.4. Example 4
In Example #4 lipped channel section members are studied. The
cross-section is mono-symmetric, the mid-line dimensions are as
follows: 200 mm web depth, 20 mm lip length, and 40 or 100 mm
ange width. The plate thickness is 2 mm. The member length is either
200 mm or 1000 mm. The member is supported at its two ends in a
globally and locally pinned way, namely: all the nodes at the end
sections are supported perpendicularly to the plates. The material is
steel-like, with E=200 000 MPa, =0.3 and G=77 000 MPa.
Three load cases are considered (Load #1 and #2 being used in
Example #4, while Load #3 being used in Example #5). All of the load
cases are dened so that shear would be dominant, illustrating that the
here-used cFEM can readily handle shear buckling problems, too. The
load cases are shown in Fig. 10. In case of Load #1 the web is clearly in
pure shear (while the anges and lips are practically free of stresses).
Load #2 can also be considered as a pure shear, but all the plate
elements are stressed. In case of Load #1 and #2 the load is distributed
over the loaded edges of the plates, and the distributed load has the
same intensity on all the edges. Load #3 is an eccentric mid-point
transverse loading, placed on the beam as a uniformly distributed
loading with distribution length of 100 mm (so that the resultant would
be 1 kN). This loading is not pure shear, but due to the short beam
lengths shear has important eect. In case of Load #3 the position of
the load application point is varying. The action line of the load lays in
the plane of the web. The force is acting either at the junction of the
web and the top ange (case top), or at the junction of the web and the
bottom ange (case bottom), or half of the load at the junction of the
top ange, half at the junction of the bottom ange (case topbot).
In Fig. 11 solutions of the 1000-mm-long member with the 40-mm-
wide anges are presented, loaded by Load #1. The buckling problem is
solved in 8 options, considering pure global (G), pure distortional (D),
pure local (L) modes and the various combinations of them. The
unconstrained problem is also solved (i.e., when all the modes are
considered), labelled as all-mode. The buckled shapes and critical load
values are given in the gure for the rst buckling mode (i.e., with the
lowest critical value). Both the buckled shapes and the critical values
are strongly inuenced by the constraints. Since the ange is relatively
narrow, it cannot eectively support the web edges laterally, therefore
9
S. dny et al. Thin-Walled Structures (xxxx) xxxxxxxx

Fcr = 978.6 kN Fcr = 1423 kN Fcr = 3877 kN

pure G, buckl #1 pure G, buckl #2 pure G, buckl #3

Fcr = 956.5 kN Fcr = 964.7 kN Fcr = 1285 kN

pure L, buckl #1 pure L, buckl #2 pure L, buckl #3
Fig. 7. Buckling mode samples for Example #1.

Mcr = 94.68 kNm Mcr = 151.8 kNm Mcr = 241.6 kNm

pure G, buckl #1, top pure G, buckl #1, middle pure G, buckl #1, bottom

Mcr = 39.73 kNm Mcr = 100.3 kNm Mcr = 151.4 kNm

pure L, buckl #1, top pure L, buckl #1, middle pure L, buckl #1, bottom
Fig. 8. Buckling mode samples for Example #2.

10
S. dny et al. Thin-Walled Structures (xxxx) xxxxxxxx

Mcr = 16.30 kNm Mcr = 19.73 kNm Mcr = 21.21 kNm

pure L, buckl #1, top pure L, buckl #1, middle pure L, buckl #1, bottom
Fig. 9. Buckling mode samples for Example #3.

all the G, D and L modes have important contribution. Still, the L mode
is the most important (note, the pure L solution is the closest to the all-
mode solution). The GDL solution is not too far from the all-mode
solution, the experienced approx. 5% dierence between them is due to
the other modes (i.e., shear and transverse extension modes). It is to
note that set of problems identical to those presented in [24], where the
problems are solved by (unconstrained) nite strip method. Thus, the
unconstrained results can directly be compared. The comparison shows
only small deviations between the results of the various methods (but
obviously, cFEM provides with constrained solutions, too).
Similar calculations are carried out with dierent ange width and/
or dierent length. Some of the critical loads are summarized in Table 2
(in N/mm2), focusing on the practically most interesting options: pure
L, GDL and all-mode options. As the results of Table 2 show, the
importance of the various deformation spaces is very much dependent
on the geometric parameters. For example, if the member length is very
short (i.e., the aspect ratio of the web panel is 1), the only practically
important space is the L space, and in these cases the pure L solution
corresponds to the classic analytic solution what is typically called
shear buckling of a plate. This also suggests that pure shear buckling
belongs to the L space.
It is important to highlight that shear mode in the cFSM/cFEM/
GBT terminology is a deformation mode that involves signicant (or
even in many cases: pure) in-plane shear deformations of the plate
elements of the member. On the other hand, shear buckling is a classic
term that describes a plate buckling due to shear stresses. Since shear
mode involves rst-order in-plane shear strains, while classic shear
buckling does not, it is nally not surprising that shear modes have
little eect on classic shear buckling.
4.3.5. Example 5
In Example #5 a 600-mm-long beam is analysed. The cross-section
is the lipped channel section with the 40-mm-wide ange from the
previous example. The beam is loaded by a transverse force as in Load
#3 of Fig. 10. Three load positions are considered, as mentioned
previously: top, bot and topbot, depending on whether the trans-
verse force is acting at the top or bottom ange, or both. The beam is
simply supported at its two ends, just as in Example #4.
Since the loading is not pure shear, it is not only the shear stresses
that contribute to the buckling solution. The problem is solved in
various options, depending on which stress components are considered/
disregarded during the calculations. According to the local coordinate
system of the applied shell element, sigx is the longitudinal normal
stress, sigy is the transverse normal stress, while tauxy is the
membrane shear stress. The considered options are as follows: sigx
only, sigy only, tauxy only, and all the possible combinations of these
three stress components, including all-stress option when all the three
stress components are considered. Thus, this example illustrates how
the calculation options of the here-proposed cFEM implementation can
be utilized.
Some results are presented in Fig. 12, namely: critical force values
and buckled shapes, all of them being rst buckling modes, calculated
by constraining deformations into the pure L space. As the results
suggest, various buckling behaviour modes can be achieved depending
on the considered stress components. (More precisely: all the stress
components are considered in the rst-order solution, but the second-
order eect of a certain stress is considered or disregarded.)
If the load is acting at the top ange, the middle part of the web is in
(transverse) compression, and for this geometry this transverse com-
pression is the governing eect, resulting in a buckling phenomenon
frequently referred to as web crippling. When the transverse stresses
are considered only (sigy option), the behaviour is very similar to the
all-stress. However, even if these transverse stresses are not consid-
ered, buckling is still possible due to either the longitudinal normal
stresses or to shear stresses. In these latter cases the buckled shapes are
the ones that are typical for the plate buckling of a web in bending, or
shear buckling of a web.
When the transverse force is acting at the bottom ange, the middle

Fig. 10. Load cases of Example #4 and #5.

11
S. dny et al. Thin-Walled Structures (xxxx) xxxxxxxx

Fcr = 1085.95 N/mm2 Fcr = 337.178 N/mm2 Fcr = 141.856 N/mm2 Fcr = 173.850 N/mm2
G, buckl #1 D, buckl #1 L, buckl #1 GD, buckl #1

Fcr = 141.260N/mm2 Fcr = 134.822 N/mm2 Fcr = 114.842 N/mm2 Fcr = 109.613N/mm2
GL, buckl #1 DL, buckl #1 GDL, buckl #1 all-mode, buckl #1
Fig. 11. Buckling mode samples for Example #4.

Table 2 position in the sigx and tauxy options. The all-stress solution
Critical loads for selected cases, Example #4. however is greatly dependent on the load position: in the top option
the all-stress solution is dominantly a web crippling, in the bot
ange width, mm length, mm loading L GDL all-mode
position it is a combination of shear buckling and plate buckling (of the
40 200 Load #1 204.934 203.860 203.107 compressed, upper part of the web), while in the topbot position it
40 200 Load #2 204.985 204.136 203.358 seems to be a combination of all the three types of plate buckling.
40 1000 Load #1 141.856 114.842 109.613
40 1000 Load #2 141.846 118.961 114.050
100 200 Load #1 198.329 198.258 197.765 4.3.6. Example 6
100 200 Load #2 198.587 198.518 197.939 Though modal decomposition is worked out only for prismatic
100 1000 Load #1 131.449 130.544 130.544 members, members with dierent cross-sections can be joined. Thus, if
100 1000 Load #2 130.702 130.702 128.812 the member is built up from prismatic segments, the proposed cFEM
can still be applied. This question is not discussed in detail here, but
illustrated by this Example #6. A simply supported member is analysed,
part of the web is in (transverse) tension, therefore crippling of the with I cross-section. The member is loaded either as a column
middle part of the web does not occur. (In the sigy option there is still (concentric compressive force, uniformly distributed over the end
buckling solution, but in this case web-crippling-like phenomenon takes sections), or as a beam (transverse force acting at the middle of the
place at the supports, and at a much higher load level.) beam at the junction of the web and the top ange). The problem,
In case of topbot loading, web crippling of the middle web can be therefore, is very similar to the ones presented by Example #1 and #2.
achieved (in the sigy option) due to the transverse compression at the However, the member is divided into three segments, and the middle,
upper part of the web, but web crippling is not governing if all the stress 300-mm-long segment is replaced by a stronger, mono-symmetric I-
components are considered. section: top ange 8012 mm, bottom ange 4012 mm, web
The load position has little eect on the distribution of the 2006 mm. Some results are given in Fig. 13. In case of the column
longitudinal normal stresses and on the shear stresses, that is why the problem now the rst mode is a exural-torsional buckling, unlike in
buckling patterns and critical values are little aected by the load Example #1, and with an increased critical value. In case of the beam

12
S. dny et al. Thin-Walled Structures (xxxx) xxxxxxxx

Fcr = 106.702 kN Fcr = 44.655 kN Fcr = 105.034 kN Fcr = 40.856 kN

top, sigx top, sigy top, tauxy top, all-stress

Fcr = 103.724 kN Fcr = 138.448 kN Fcr = 115.910 kN Fcr = 72.646 kN

topbot, sigx topbot, sigy topbot, tauxy topbot, all-stress

Fcr = 99.335 kN Fcr = 725.879 kN Fcr = 99.974 kN Fcr = 100.022 kN

bottom, sigx bottom, sigy bottom, tauxy bottom, all-stress
Fig. 12. Buckling mode samples for Example #5 (in all the cases buckl #1 is shown for pure L).

problem the lateral-torsional mode and the rst local mode is similar to
the prismatic case (see Example #2, Fig. 8), but the critical values are
signicantly increased.
4.3.7. Example 7
The handling of holes is not the topic of this actual paper, so it is not
discussed here in detail, only a few examples are shown. Example #7 is
a simply supported beam with C-shaped cross-section similar to the
previously presented ones: web, ange and lip widths are 200, 80 and
20 mm, respectively, plate thickness is 2 mm. The beam is loaded by
mid-point concentrated transverse force, acting at the junction of the
web and bottom ange. Pure global, distortional local-plate modes are
calculated with considering holes, namely: either two large rectangular
holes, or slotted web, or a large elliptical opening is considered. In
Fig. 14 the rst (practically meaningful) modes are shown. The
presence, the dimensions and arrangements of the holes greatly

13
S. dny et al. Thin-Walled Structures (xxxx) xxxxxxxx

Fcr = 1422 kN Mcr = 506.4 kNm Mcr = 109.4 kNm

pure G, buckl #1 pure G, buckl #1 pure L, buckl #1
Fig. 13. Pure G and L buckling mode samples for Example #6.

Mcr = 1352 kNm Mcr = 1089 kNm Mcr = 5.411 kNm

pure G, buckl #1 pure D, buckl #1 pure L, buckl #1

Mcr = 1549 kN Mcr = 69.61 kNm Mcr = 2.354 kN

pure G, buckl #1 pure D, buckl #1 pure L, buckl #1

Mcr = 1160 kN Mcr = 1624 kNm Mcr = 11.34 kN

pure G, buckl #1 pure D, buckl #1 pure L, buckl #1
Fig. 14. Buckling mode samples for Example #7.

inuences the results, especially local and distortional buckling. More
problems with holes are presented in [24].
5. Summary
In this paper a novel method is applied for the modal decomposition
of the deformations of thin-walled members. The method applies
essentially the same constraining technique as the constrained nite
strip method, however, the member is discretized both in the transverse
and longitudinal directions, and the longitudinal base functions are
modied accordingly: from the trigonometric functions of FSM to
polynomial functions that are widely used in the nite element method.
Due to these changes, the new method can readily be termed as
constraint nite element method, which possesses the same modal
features as the constrained nite strip method, but with signicantly
extended practical applicability. The new method requires a rectangu-

14
S. dny et al.
lar mesh, but otherwise it can handle nearly arbitrary problems. The
applicability of the proposed method is demonstrated by numerical
examples. The main conclusions can be summarized as follows.
The examples show that the proposed constrained (shell) nite
element method can be applied for the buckling analysis of thin-walled
members with virtually arbitrary restraints and loading. It can handle
members with holes, and can handle some cross-section changes, too,
though these questions are not discussed in detail in this paper. Without
enforcing constraints, the method can be used as an arbitrary shell
nite element method. With the modal decomposition feature, how-
ever, it allows a better understanding of the behaviour, as well as it
makes possible to have buckling solution in a reduced displacement-
deformation space.
The numerical results suggest that it might be practically useful to
analyse the eect of the stress components even within a specic
deformation space. Though more research is needed in this area, the
rst results show that, for example, classic plate buckling, shear
buckling or web crippling can be separated by considering the second-
order eect of the longitudinal normal, shear, or transverse normal
stresses, respectively, if the analysis is constrained into the local
deformation space (to which all these local-plate buckling phenomena
belong in a geometric sense).
It is believed that the newly proposed method can contribute to the
better understanding of the behaviour of thin-walled members. The
method can further be extended to practical cases where the modal
decomposition technique is not explored yet, e.g. to consider transverse
stieners or end-plates in a thin-walled member, or to analyse trusses or
frames built up from thin-walled members.
Acknowledgements
The presented work was conducted with the nancial support of the
OTKA K108912 project of the Hungarian Scientic Research Fund.
References
[1] NAS, North American Specication for the Design of Cold-formed Steel Structural
Members. 2007 ed. Washington DC, USA, American Iron and Steel Institute, 2007.
[2] Eurocode 3, EN 1993-1-3, General rules, Supplementary Rules for Cold-formed
Members and Sheeting, October 23, 2006.
[3] R. Schardt, Verallgemeinerte Technische Biegetheorie, Springer Verlag, Berlin,
Heidelberg, 1989.
[4] R. Gonalves, M. Ritto-Corra, D. Camotim, A new approach to the calculation of
cross-section deformation modes in the framework of generalized beam theory,
15
Thin-Walled Structures (xxxx) xxxxxxxx
Comput. Mech. 46 (5) (2010) 759781.
[5] N. Silvestre, Generalised beam theory to analyse the buckling behaviour of circular
cylindrical shells and tubes, Thin-Walled Struct. 45 (2) (2007) 185198.
[6] Y.K. Cheung, Finite strip method in the analysis of elastic plates with two opposite
ends simply supported, Proc. Inst. Civ. Eng. 40 (1968) 17.
[7] G.J. Hancock, Local, Distortional, and lateral buckling of I-beams, ASCE J. Struct.
Eng. 104 (11) (1978) 17871798.
[8] DSM, American Iron and Steel Institute. Direct Strength Method Design Guide.
Washington, DC, USA, 2006.
[9] S. dny, B.W. Schafer, Buckling mode decomposition of single-branched open
cross-section members via nite strip method: derivation, Thin-Walled Struct. 44
(5) (2006) 563584.
[10] S. dny, B.W. Schafer, Buckling mode decomposition of single-branched open
cross-section members via nite strip method: application and examples, Thin-
Walled Struct. 44 (5) (2006) 585600.
[11] S. dny, B.W. Schafer, A full modal decomposition of thin-walled, single-branched
open cross-section members via the constrained nite strip method, J. Constr. Steel
Res. 64 (1) (2008) 1229.
[12] S. dny, B.W. Schafer, Generalized constrained nite strip method for thin-walled
members with arbitrary cross-section: primary modes, Thin-Walled Struct. 84
(2014) 150169.
[13] S. dny, B.W. Schafer, Generalized constrained nite strip method for thin-walled
members with arbitrary cross-section: secondary modes, orthogonality, examples,
Thin-Walled Struct. 84 (2014) 123133.
[14] GBTUL 2.0: Buckling and Vibration Analysis of Thin-Walled Members. DECivil/IST,
2013. Technical University of Lisbon, (http://www.civil.ist.utl.pt/gbt).
[15] CUFSM v4.05: Elastic Buckling Analysis of Thin-Walled Members by Finite Strip
Analysis, 2012. http://www.ce.jhu.edu/bschafer/cufsm.
[16] S. dny, Constrained shell Finite Element Method for thin-walled members, Part1:
constraints for a single band of nite elements, submitted to Thin-Walled
Structures.
[17] S. dny, Constrained nite element method: demonstrative examples on the global
modes of thin-walled members, in: Proceedings of the Twenty-Second International
Speciality Conference on Cold-Formed Steel Structures, (eds. R. A LaBoube, W.W.
Yu) St. Louis, Missouri, USA, Nov 5-6, 2014, pp. 6781.
[18] S. dny, Constrained nite element method for the modal analysis of thin-walled
members with holes, Proceedings of the Annual Stability Conference, SSRC,
Orlando, Florida, April 12-15. paper no: 40. p. 20, 2016.
[19] S. dny, D. Visy, R. Nagy, Buckling solutions for thin-walled members by using the
constrained nite element method: demonstrative examples, in: Proceedings of the
International Colloquium on Stability and Ductility of Steel Structures, SDSS 2016,
(eds. D. Dubina, V. Ungureanu), Timisoara, Romania, May 30 June 1, 2016, pp.
51-58.
[20] S. dny, Shell element for constrained nite element analysis of thin-walled
structural members (August)Thin-Walled Struct. 105 (2016) 135146.
[21] D. Visy, S. dny, Local Elastic and Geometric Stiness Matrices for the Shell
Element Applied in cFEM, submitted to Periodica Polytechnica ser. Civil
Engineering, http://dx.doi.org/10.3311/PPci.10111.
[22] S. dny, Global Buckling of thin-walled columns: analytical solutions based on
shell model, Thin-Walled Struct. 55 (2012) 6475.
[23] S. dny, D. Visy, Global buckling of thin-walled columns: numerical studies, Thin-
Walled Struct. 54 (2012) 8293.
[24] C.H. Pham, G.J. Hancock, Shear buckling of thin-walled channel sections, J. Constr.
Steel Res. 65 (2009) 578585.