Thin-Walled Structures 105 (2016) 135146

Contents lists available at ScienceDirect

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

Full length article

Shell element for constrained nite element analysis of thin-walled

structural members
Sndor dny
Budapest University of Technology and Economics, Department of Structural Mechanics, 1111 Budapest, Megyetem rkp. 3, Hungary

art ic l e i nf o a b s t r a c t

Article history: In this paper a novel shell nite element is introduced, specically proposed for constrained shell nite
Received 7 January 2016 element analysis. The proposed element is derived from the nite strips used in the semi-analytical nite
Received in revised form strip method. The new nite element shares the most fundamental feature of the nite strips, namely:
6 April 2016
transverse and longitudinal directions are distinguished. Moreover, the new element keeps the trans-
Accepted 10 April 2016
verse interpolation functions of nite strips, however, the longitudinal interpolation functions are
changed from trigonometric functions (or function series) to classic polynomials. It is found that the
Keywords: proper selection of the polynomial longitudinal interpolation functions makes it possible to perform
Constrained nite element method modal decomposition similarly as in the constrained nite strip method (cFSM). This requires an unusual
Shell nite element
combination of otherwise well-known shape functions. If the so-constructed shell nite elements are
used to model a thin-walled member, (hence, with using discretization in both the transverse and the
longitudinal directions,) modal decomposition can be done essentially identically as in cFSM, whilst the
practical applicability of the method is signicantly extended (e.g., various restraints, holes, certain cross-
section changes can easily be handled). In this paper the focus is on the derivation of the novel shell nite
element. Constraining capability is illustrated by some basic examples. Practical application of the novel
element will be presented in subsequent papers.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction calculation, as well as simpler pre- and post-processing. The price

The nite strip method (FSM) can be regarded as a special
version of nite element method (FEM) in which special nite
element-s are used. The most essential feature of FSM is that
there are two pre-dened directions, and the base functions (or:
interpolation functions) are different in the two directions. (Ty-
pically, though not necessarily and not always, the two char-
acteristic directions are perpendicular to each other). In classical
(sometimes also referred as to semi-analytical) FSM, as in [14] the
structural member to be analyzed is discretized only in one di-
rection (say: transverse direction), while in the other direction
(say: longitudinal direction) there is no discretization, i.e., in this
direction there is only one element along the member. This is why
the dimensions of this nite element are typically distinctly
different in the two directions, and that is why such an element is
called nite strip rather than nite element.
This special nite strip discretization has various consequences.
An advantageous consequence is that the total number of ele-
ments, therefore the total number of degrees of freedom is much
smaller than in case of a classic FEM, which means faster
E-mail address: sadany@epito.bme.hu
of the calculation efciency is restricted generality: the analyzed
member must be highly regular (e.g., typically it must be straight,
prismatic, etc.). A major restriction of the classical semi-analytical
FSM is that a certain longitudinal shape function can properly be
applied for only a certain problem with given boundary conditions,
since in the lack of longitudinal discretization accurate solution
can be expected only if the longitudinal interpolation function well
represent the real behavior (i.e., if the applied shape function sa-
tises the differential equation and boundary conditions of the
problem).This restriction can (partially) be released if either tri-
gonometric series or splines are used for the longitudinal inter-
polation. In either case the problem size is signicantly increased
(compared to FSM), while practical applicability is still limited
(compared to FEM).
The original idea of constraining a shell-model is proposed in
[5,6] then in [710]. The idea is to dene special constraints, based
on some pre-dened mechanical criteria, the introduction of
which enables modal decomposition. Modal decomposition
transforms the original displacement eld into a set of modal
displacements that can solve two basic problems: calculation in a
reduced but practically meaningful space (e.g., calculating global
buckling directly, by using only a few degrees of freedom), and
modal identication (e.g., assigning participation percentages from

http://dx.doi.org/10.1016/j.tws.2016.04.012
0263-8231/& 2016 Elsevier Ltd. All rights reserved.
136 S. dny / Thin-Walled Structures 105 (2016) 135146
the modal deformations to a general deformation eld).
The constrained nite strip method (cFSM) was rst proposed
and developed for the semi-analytical FSM with sine-cosine
longitudinal shape functions that correspond to locally and glob-
ally pinned-pinned end restraints of the thin-walled beam or
column. Later other end conditions have also been considered
[11,12], but still within the semi-analytical FSM. The method was
then generalized to be able to handle general cross-sections
[13,14], which also required a more systematic denition of the
deformation modes [15]. An attempt to constraining a spline FSM
is presented in Ref. [16]. In Refs. [17,18] the constraining technique
is applied to shell FEM in the context of a commercial nite ele-
ment code.
Though all these above-mentioned constrained methods share
the same basic mechanical background, there are distinct differ-
ences. The advantageous features of cFSM are as follows: (a) the
method provides a full modal decomposition (i.e., the whole dis-
placement led is transformed into a modal system), (b) the me-
chanical criteria of the modes are exactly satised, and (c) adding
constraints reduces the DOF number of the problem. However,
since cFSM is based on FSM, it has all the restrictions of FSM: for
example the member has to be regular, or, FSM is efcient only if
the longitudinal shape function is dened specic to the end re-
straints, which means that arbitrary boundary conditions cannot
be handled in an efcient way. The so-far proposed constrained
FEM has potentially more general applicability than cFSM, but
(a) it provides only partial modal decomposition (that is why
modal identication is not readily be handled), (b) the mechanical
criteria are satised only approximately, and (c) adding constraints
increases the DOF number of the problem.
It can be concluded that all of the existing constrained methods
possess limitations and/or disadvantages. A better constrained
method should provide full decomposition, should satisfy the
mechanical criteria exactly, should be generally applicable as
much as possible (as far as loading, boundary conditions, etc. are
concerned). In this paper the rst step toward such a method is
presented. A novel shell nite element is introduced. The appli-
cation of the novel shell element leads to a method which shares
(most of) the advantageous features of the original cFSM, while
providing signicantly extended general practical applicability.
The proposed element has a major geometrical limitation of being
rectangular and that its local coordinate system must match the
longitudinal and transverse direction of the member. Otherwise,
no further restrictions are included, that is, the proposed element
can be used similarly to any other shell nite elements.
By using the proposed novel shell element various problems
can readily be handled: various end and intermediate restraints,
nearly arbitrary loading, arbitrary buckling modes including shear
buckling or web crippling, certain cross-section changes along the
Fig. 1. Coordinates and DOF in nite strip method.
member length, holes. Some of these problems can be solved by
other methods, some not. For example, various loading and re-
straints can be handled by the generalized beam theory (GBT)
together with modal decomposition, see e.g. [1922]. Though at-
tempts to extend GBT for sections with holes are recently reported
[23,24], it is still believed that the cFEM based on the here pro-
posed nite element is powerful, since it integrates the advanta-
geous features of all the available modal decomposition methods,
and it is based on the shell nite element method which is widely
used in research and even in design practice.
In this paper the derivation of the novel shell nite element is
presented in detail. Then the constraining capability of the ele-
ment is illustrated by some basic semi-analytical examples. The
examples justify the applicability. The details of the constrained
nite element method (cFEM) will be presented in subsequent
papers together with various examples to illustrate the advanta-
geous features of the new method.
2. Derivation of the proposed nite element
2.1. Short overview on existing semi-analytical FSM and cFSM
In nite strip method a member is discretized into longitudinal
strips, instead of nite element method, which applies dis-
cretization in both the longitudinal and transverse directions. In
Fig. 1a single strip is shown, along with the typically used local
coordinate system and the degrees of freedom (DOF) for the strip,
the dimensions of the strip, and the applied end tractions.
By using the simplest longitudinal trigonometric functions, the
displacements are approximated as follows.
x x u1 my
u (x, y) = 1 sin
b b u2 a (1)
x x v1 my
v (x, y) = 1 cos
b b v2 a (2)
3x 2 2x 3 2x 2 x 3 3x 2 2x 3 x 2 x 3
w (x, y ) = 1 2 + 3 x+ 2 2 3 2
b b b b b b b b
w1
1 sin my
w2 a
2 (3)
The above formulae represent pinned-pinned boundary con-
ditions. Two important features of the longitudinal shape func-
tions are that the same longitudinal function is used for u and w,
and the longitudinal function for v is the derivative of that used for
 S. dny / Thin-Walled Structures 105 (2016) 135146
u and w. Therefore, the above formulae can be generalized as
follows:
x x u1
u (x, y) = 1 f (y)
b b u2 (4)
x x v1 f (y)
v (x, y) = 1
b b v2 g (y) (5)
3x2 2x3 2x2 x3 3x2 2x3 x2 x3 M.
w (x, y) = 1 2 + 3 x+ 3 2 Modal identication, i.e. categorization of a general deforma-
b b b b2 b2 b b b
w1
1 f (y)
w2
2 (6)
By appropriately selecting the functions, various end conditions
can be described, as in [2,11,12]. The proposed functions are all
trigonometric functions. Depending on the boundary condition,
the functions might be given in series form. Note, in Eq. (5) f(y) is
the longitudinal base function used for the transverse in-plane and
out-of-plane displacements, while g(y) is usually dened as the
rst derivative of the internal function of f(y), see e.g., Eq. (2).
The local elastic and geometric stiffness matrices can be con-
structed by following conventional FEM steps, by considering the
2D generalized Hooke's law (for the elastic stiffness matrix) and by
considering the second-order strain terms (for the geometric
stiffness matrix). The stiffness matrices can be determined analy-
tically. From the local stiffness matrices the member's (global)
stiffness matrices (elastic and geometric, Ke and Kg) can be com-
piled as in FEM, by transformation to global coordinates and
assembly.
For a given distribution of edge tractions on a member the
geometric stiffness matrix scales linearly, resulting in the classic
eigen-buckling problem, namely:
K e K g = 0 (7)
with
= diag 1 2 ... nDOF and = 1 2 ... nDOF (8) FSM. The longitudinal interpolation function should be changed,
where i is the critical load multiplier and i is the mode shape
vector.
The constrained FSM (cFSM) is an extension to FSM that uses
mechanical assumptions to enforce or classify deformations to be
consistent with a desired set of criteria. The method is originally
presented in Refs. [510], and implemented in Refs. [25,26]. The
cFSM constraints are mechanically dened, and are utilized to
formally categorize deformations into global (G), distortional (D),
local (L), and other (i.e., shear and transverse extension, S T)
deformations. The mechanical criteria are mostly given by setting
certain displacement and displacement derivatives to zero. Once
the mechanical criteria are transformed into constraint matrices,
any FSM displacement eld d (e.g. an eigen-buckling mode is an
important special case) may be constrained to any modal dM de-
formation space via:
d = R M dM
where RM is a constraint matrix, the derivation of which can be in
found Refs. [510] for open cross-sections, and M might be G, D, L,
S and/or T.
Though modal decomposition is not restricted to eigen-buck-
ling solution, this is the problem where modal decomposition is
mostly used. It can be completed by introducing the desired
constraint matrix RM, the columns of which can be interpreted as
137
base vectors for the given M space. Applying RM for the intended
space (M G, D, L, S, and/or T) Eq. (7) becomes:
R M TK e R M M M R M TK g R M M = 0 (10)
K eM M M K gM M = 0 (11)
where KeM and KgM are reduced-size elastic and geometric stiff-
ness matrices for the eigen-buckling solution constrained to space
tion into the M spaces, is also possible, due to the fact that
G D L ST spans the entire FSM space. As such, the RGDLST
constraint matrix represents an alternative basis for the FSM
space, in which deformations are categorized. This basis transfor-
mation of displacement vector d may be expressed as:
d = R G R D R L R ST c (12)
where c now provides the deformations within each class: cG, cD,
cL, cST. The values of c are dependent on the normalization of the
base vectors within R. A full discussion of the normalization se-
lection for R is provided in Ref. [27]. Once c is determined,
pi participation of an individual mode or pM participation of a
deformation space M can also be determined as follows:
pi = ci /c or pM = cM /c (13)
The cFSM method has recently been generalized [13,14]. An
important feature of the generalization is the more rened de-
nition of the modes, (e.g., introduction of primary and secondary
mode, as well a practically meaningful handling of the shear
modes), and the extension of the method to arbitrary cross-
sections.
2.2. New longitudinal shape functions
Our goal here is to transform the nite strip into a shell nite
element. Since the above-summarized semi-analytical FSM uses
classic polynomials in the transverse direction, the new shell
element can inherit the transverse interpolation functions from
however, by keeping some important characteristics of the func-
tions of FSM. Namely, the new longitudinal shape functions must
have the following features:
 they must be able to exactly satisfy the constraining criteria for
mode decomposition (no-shear criterion, no-transverse-exten-
sion criterion, etc.),
 the transverse in-plane displacements must be interpolated by
using the same shape functions as used for the out-of-plane
displacements,
(9)
Fig. 2. Coordinates and DOF in nite strip method.
138 S. dny / Thin-Walled Structures 105 (2016) 135146

 they must provide C(1) continuous interpolation for the out-of- (2) 3x 2x2
plane displacements (which is practically useful for dening
Nx,1 =1 + 2
a a (20)
various end restraints).
(2) 4x 4x2
Moreover, we change the local coordinate system, see Fig. 2, in Nx,2 = 2
a a (21)
order to be more conform with the common practice:
(2) x 2x2
 The longitudinal axis will be the global X and local x axis, Nx,3 = +
a a2 (22)
 the coordinate system will be right-handed.
Third-order shape functions in x:
The notations for the displacement functions and nodal dis- 3x2 2x3
(3)
placements will be changed accordingly, as detailed below. Nx,1 =1 + 3
a2 a (23)
The proposed interpolation functions are summarized as fol-
lows: 2x2 x3
(3)
Nx,2 =x +
a a2 (24)
y y u1
u (x, y) = 1
b b u2
(3) 3x2 2x3
3x 2x2 4x 4x2 x 2x2 Nx,3 = 3
1 a2 a (25)
+ 2 2 + 2
a a a a a a
cu1 (3) x2 x3
c Nx,4 = +
a a2 (26)
u2
cu3 (14) First-order shape functions in y:
(1) y
y y v1 N y,1 =1
v (x, y) = 1 b (27)
b b v2
3x2 2x3 2x2 x3 3x2 2x3 x2 x3 (1) y
N y,2 =
1 2 + 3 x + 3 + 2 b (28)
a a a a2 a2 a a a
cv1 Third-order shape functions in y:
c
v2 (3) 3y 2 2y 3
N y,1 =1 + 3
cv3 b2 b (29)
cv4 (15)
(3) 2y 2 y 3
N y,2 =y + 2
b b (30)
3y 2 2y 3 2y 2 y 3 3y 2 2y 3 y2 y3
w (x , y ) = 1 + y + +
b2 b3 b b2 b2 b3 b b 2
(3) 3y 2 2y 3
N y,3 = 3

3x 2 2x 3 2x 2 x 3 3x 2 2x 3 x2 x 3 b2 b (31)
1 + x+
a2 a3 a a2 a2 a3 a a2
cw1 (3) y2 y 3
c N y,4 = +
w2 b b2 (32)
cw 3
cw 4 (16) The proposed longitudinal shape functions and its coefcients
are illustrated in Fig. 3.
In a shorter form:
cu1 2.3. Finite element interpretation
u1
u (x, y) = N y(1,1) N y(1,2) Nx(2,1) Nx(2,2) Nx(2,3) cu2
u2
cu3 (17) The above formulae include separate sets of coefcients for the
transverse and longitudinal directions. However, these coefcients
can easily be exchanged by classic nite element nodal
cv1 displacements.

(1) (1) v1 (3) (3) (3) (3) cv2 As an example, the in-plane longitudinal displacement is ex-
v (x, y) = N y,1 N y,2 Nx,1 Nx,2 Nx,3 Nx,4
v2 cv3 pressed as follows, see Eq. (17):
cv4 (18)
u (x, y)=u1cu1N y(1,1) Nx(2,1)+u2 cu1N y(1,2) Nx(2,1)+u1cu2 N y(1,1) Nx(2,2)+u2 cu2 N y(1,2)
Nx(2,2)+u1cu3 N y(1,1) Nx(2,3)+u2 cu3 N y(1,2) Nx(2,3) (33)
w1

w (x, y ) = N (y3,1) N (y3,2) N (y3,3) (3 ) 1
N y,4
The in-plane longitudinal DOF are the ui cuj constants, where
w2 i1...2 and j1...3. Thus, nally, there are 6 such DOF, all of them
2
are translational, and will be denoted here as in Fig. 4. The inter-
cw1 polation with the nite element DOF:

(3 ) (3 ) (3 ) (3) cw 2
N x,1 N x,2 N x,3 N x,4
cw 3 u (x, y)=u11N y(1,1) Nx(2,1)+u13 N y(1,2) Nx(2,1)+u21N y(1,1) Nx(2,2)+u23 N y(1,2) Nx(2,2)+u31N y(1,1)
cw 4 (19)
Nx(2,3)+u33 N y(1,2) Nx(2,3) (34)
with the shape functions as follows.
Second-order shape functions in x: Similarly, the in-plane transverse DOF are the vi cvj constants,
 S. dny / Thin-Walled Structures 105 (2016) 135146 139

see Eq. (18), where i 1...2 and j1...4. Thus, nally, there are notations as in Fig. 4):
8 such DOF, which will be denoted here as in Fig. 4. Therefore, the
u dNx(2,1) dNx(2,1) dNx(2,2) dNx(2,2)
v displacement function is interpolated as follows: =u11N y(1,1) +u13 N y(1,2) +u21N y(1,1) +u23 N y(1,2) +u31
x dx dx dx dx
v (x, y)=v11N y(1,1) Nx(3,1)+v13 N y(1,2) Nx(3,1)+v31N y(1,1) Nx(3,3)+v33 N y(1,2) Nx(3,3)+z11N y(1,1) (2)
dNx,3 dNx,3(2)
N y(1,1) +u33 N y(1,2)
Nx(3,2)+z13N y(1,2) Nx(3,2)+z 31N y(1,1) Nx(3,4) +z 33N y(1,2) Nx(3,4) dx dx (38)
(35)

The out-of-plane displacement function can be expressed si- dNx(2,1) dNx(2,2)

milarly from Eq. (19), by using nite element nodal displacement (u11N y(1,1)+u13 N y(1,2) ) + (u21N y(1,1)+u23 N y(1,2) )
dx dx
DOF, as follows:
dNx(2,3)
+ (u31N y(1,1)+u33 N y(1,2) ) = 0
w (x, y)=w11N y(3,1) Nx(3,1)+w13 N y(3,3) Nx(3,1)+w31N y(3,1) Nx(3,3)+w33 N y(3,3) dx (39)

Nx(3,3)+x11N y(3,2) Nx(3,1)+x13N y(3,4) Nx(3,1)+x31N y(3,2) Nx(3,3)+x33N y(3,4) Considering the shape functions and its derivatives, it is easy to
conclude that the actual strain function is linear both in x and y.
Nx(3,3)y11N y(3,1) Nx(3,2)y13N y(3,3) Nx(3,2)y31N y(3,1) Nx(3,4) y33N y(3,3)
Therefore, the function can be expressed in the form:
Nx(3,4) xy11N y(3,2) Nx(3,2)xy13N y(3,4) Nx(3,2)xy31N y(3,2)
C11xy +C10 x+C01y +C00=0 (40)
Nx(3,4) xy33N y(3,4) Nx(3,4) (36)
with the C coefcients as follows:
Therefore, the proposed element has 30 DOF: 6 for u, 8 for v. C11= 4 (u11u132u21+2u23+u31u33 )/b/a2 (41)
and 16 for w. 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,
C10=4 (u112u21+u31)/a2 (42)
y) (a/2,b) with one DOF per node for the u displacement. The DOF
are illustrated in Fig. 4.
C01=(3u113u134u21+4u23+u31u33 )/b/a (43)

C00= (3u114u21+u31)/a (44)

3. Constraints
The longitudinal strain is zero for any x-y if (and only if) all the
The constraints that are embedded in cFSM are given in [13,14]. C coefcients are zero. This is satised if:
It can be observed that the constraints are formulated by setting
u11 = u21 = u31 (45)
various displacement derivatives to zero. It can also be observed
that the criteria are practically independent of the longitudinal
u13=u23=u33 (46)
shape functions, therefore, the same criteria can be used for the
here-proposed nite element that are used for nite strips. It is This means that rigid-body longitudinal displacement is al-
also important that the introduction of the mechanical criteria lowed, otherwise the longitudinal displacement must be zero.
must lead to simple relationships in between the nodal displace- The implementation of the various criteria requires the re-
ment DOF, since this is a necessary condition if the mechanical petition of essentially the same steps, therefore, the other criteria
criteria are intended to be exactly satised. will not be discussed in detail. Nevertheless, the resulting re-
In case of the no-longitudinal-extension criterion, the criterion lationships are summarized as follows.
is: In case of the no-transverse-extension criterion, the criterion
u is:
x= =0
x (37) v
y= =0
Using the assumed shape functions (with using the general y (47)

Fig. 3. Longitudinal shape functions.

140 S. dny / Thin-Walled Structures 105 (2016) 135146

Fig. 4. Nodal displacement DOF of the nite element.

Using the assumed shape functions, the criterion is satised for x13 x33
=xy13 (or=xy33)
any x-y if: a (64)
v11=v13 (48) In case of the no- transverse-curvature criterion, the criterion
is:
v31=v33 (49) 2w
y= =0
y2 (65)
z11=z13 (50)
Using the assumed shape functions, the criterion is satised for
any x-y if:
z 31=z 33 (51)
x11=x13 (66)
In case of the no-shear criterion, the criterion is:
u v w13 w11
xy= + =0 =x11(or=x13)
y x (52) b (67)

Since in the constraining procedure the no-shear criterion is

x31=x33 (68)
applied always together with the no-transverse extension criter-
ion, it is more straightforward (and leads to simpler conclusions)
w33 w31
to immediately consider the results of no-transverse-extension =x31(or=x33)
b (69)
criterion, which nally leads to the following equations:
u11 u13 xy11=xy13 (70)
=z11(or = z13)
b (53)

y13 y11
u21 u23 3 v31 v11 z11 + z 31 =xy11(or=xy13)
= b (71)
b 2 a 4 (54)

xy31=xy33 (72)
u31 u33
=z 31(or = z 33)
b (55)
y33 y31
In case of the no- longitudinal-curvature criterion, the criterion =xy31(or=xy33)
b (73)
is:
In case of the no- mixed-curvature criterion, the criterion is:
2w
x= =0
x2 (56) 2w
xy= =0
xy (74)
Using the assumed shape functions, the criterion is satised for
any x-y if: Using the assumed shape functions, the criterion is satised for
y11=y31 any x-y if:
(57)
xy11=xy13=xy31=xy33=0 (75)
w11 w31
=y11(or=y31)
a (58) x11=x31 (76)

y13=y33 (59) x13=x33 (77)

w13 w33
=y13 (or=y33) y11=y13 (78)
a (60)

xy11=xy31 (61) y31=y33 (79)

x11 x31 w31w11=w33w13 (80)

=xy11(or=xy31)
a (62)
In case of no-warping-shear criterion the shear strain is al-
lowed to be non-zero, but only due to transverse displacements.
xy13=xy33 (63)
The criterion therefore is:
 S. dny / Thin-Walled Structures 105 (2016) 135146 141

u above proposed nite elements and constraining equations, but

=0
y (81) the problems are solved analytically in order to provide a better
comparison to exact analytical solutions.
Using the assumed shape functions, the criterion is satised for The problem is a simple straight column or beam member with
any x-y if:
a (narrow) rectangular cross-section with width and thickness b
u11=u13 (82) and t, respectively. The length is L. It is discretized by two (equal)
nite elements along the length, as shown in Fig. 5.
u21=u23 (83) The applied FE model thus has 10 nodes, 6 nodes having 7 DOF
and 4 nodes having 1 DOF. Altogether the model has 46 DOF,
u31=u33 (84) which is reduced due to constraining and end restraints, as dis-
cussed as follows.
In case of no-transverse-shear criterion the shear strain is al- Since our goal is to calculate critical forces to global modes, the
lowed to be non-zero, but only due to longitudinal displacements. constraints characterizing global modes must be applied. Namely:
The criterion therefore is: (i) transverse (membrane) extension is zero, (ii) (membrane) shear
v deformations are zero, and (iii) there is no transverse curvature.
=0
x (85) No-transverse-extension criteria means that the v and z , DOF
are halved, thus, instead of having 6 different v and 6 different z ,
Using the assumed shape functions, the criterion is satised for
there are only 3 independent v and 3 independent z .
any x-y if:
Furthermore, no-shear criterion means that the u (warping)
z11=z13=z 31=z 33 (86) displacements are dependent on v-s and z -s. Therefore, instead of
having 10 u-s, we have only 5 independent u DOF.
v11=v31 (87) Finally, no-transverse-curvature criterion means:

v13=v33 (88)  instead of having 6, there are only 3 independent x

 instead of having 6, there are only 3 independent xy ,
 instead of having 6, there are only 3 independent w
 instead of having 6, there are only 3 independent y
4. Demonstrative examples
There are 23 independent DOF, shown in Fig. 6, The con-
4.1. Model, restraints, constraints straining equations are given in Table 1:
Two classic end restraint conditions will be considered here:
A few examples are presented here. The examples are simple S-S, i.e. simple-simple (or: pinned-pinned) and C-C, i.e., clamped-
classical buckling problems to which analytical solution (for the clamped.
critical load) is known. The examples are solved by using the In case of S-S, fork supports are assumed at both ends, i.e.,

Fig. 5. Beam or column member and its discretization.

142 S. dny / Thin-Walled Structures 105 (2016) 135146

u (x, y)=u1N y(1,1) Nx(2,1)+u2 N y(1,2) Nx(2,1)+u3 N y(1,1) Nx(2,2)+u4 N y(1,2) Nx(2,2)+u5 N y(1,1)
Nx(2,3)+u6 N y(1,2) Nx(2,3) (89)

Substituting the constraint equations (for the notations, see

Fig. 6):

u (x, y)=(ua +z, a b/2) N y(1,1) Nx(2,1)+(ua z, a b/2)

1. 5va z, a 1. 5vc z, c b
N y(1,2) Nx(2,1)+ ub + +
a 4 a 4 2
1. 5va z, a 1. 5vc z, c b
N y(1,1) Nx(2,2)+ ub + + +
a 4 a 4 2
N y(1,2) Nx(2,2)+(uc +z, c b/2) N y(1,1) Nx(2,3)+(uc z, c b/2) N y(1,2) Nx(2,3) (90)

In case of C-C restraints, uc =0, va=0 and z, a=0, therefore:

1. 5vc z, c b
u (x, y)=(ua ) N y(1,1) Nx(2,1)+(ua ) N y(1,2) Nx(2,1)+ ub +
a 4 2
1. 5vc z, c b (1) (2)
N y(1,1) Nx(2,2)+ ub + + N N +(z, c b/2)
a 4 2 y,2 x,2
Fig. 6. Nodal DOF of the constrained column or beam member. N y(1,1) Nx(2,3)+(z, c b/2) N y(1,2) Nx(2,3) (91)

1. 5vc z, c b
va=0, ve=0, wa=0, we=0, x, a=0 and x, e=0. In addition, one cross- u (x, y)=ua N y(1,1) Nx(2,1)+ua N y(1,2) Nx(2,1)+ ub +
a 4 2
section in the longitudinal direction should be restrained, e.g.,:
uc =0. Thus, in case of S-S end restraints there are 16 independent 1. 5vc z, c b (1) (2) b
N y(1,1) Nx(2,2)+ ub N N +z, c N y(1,1)
DOF. a 4 2 y,2 x,2 2
In case of C-C, the following DOF are known to be zero: va=0, b
Nx(2,3)z, c N y(1,2) Nx(2,3)
ve=0, wa=0, we=0, x, a=0, x, e=0, y, a=0, y, e=0, z, a=0, z, e=0, xy, a=0 2 (92)
and xy, e=0. In addition, one cross-section in the longitudinal di-
rection should be restrained, e.g.: uc =0. Thus, in case of C-C end Finally, in case of symmetrical displacements z, c=0, therefore:
restraints there are 10 independent DOF. 3b (1) (2) 3b
Since we are aiming analytical solution, the number of DOF u (x, y)=ua N y(1,1) Nx(2,1)+ua N y(1,2) Nx(2,1)+ ub +vc N N + ub vc
4a y,1 x,2 4a
must further be reduced. It is known that the buckling shapes (for
the considered end restraints) are either symmetrical or point- N y(1,2) Nx(2,2) (93)
symmetrical.
As it can be seen, the u(x,y) displacement function of the rst
If they are symmetrical, this means further reduction of the
nite element is expressed by three displacement parameters,
independent DOF as follows: y, c=0, z, c=0, xy, c=0, ue=ua , ud=ub ,
y, e=y, a , xy, e=xy, a and z, e=z, a . namely: ua, ub and vc. The other displacement functions can be
If they are point-symmetrical, the following equations must be expressed similarly.
satised: vc =0, wc =0 and x, c=0, ue=ua , ud=ub , y, e=y, a , xy, e=xy, a Dependency on z can be considered as follows:
and z, e=z, a . (Note, in case of C-C, the last 3 conditions are auto- w (x, y)
matically satised, since the involved displacement components u (x, y , z )=u (x, y)z
x (94)
are zero.).
Thus, utilizing symmetry or point-symmetry condition, the w (x, y)
independent (non-zero) DOF number reduces to 8 in case of S-S, v (x, y , z )=v (x, y)z
y (95)
while to 5 in case of C-C. With these 8 or 5 parameters the whole
displacement eld is determined. These parameters are listed for
w (x, y , z )=w (x, y) (96)
the considered cases, as follows:
The rst-order strains can be determined as follows:
 S-S, symmetric: ua , y, a , z, a , xy, a , ub , vc , wc , x, c .
 S-S, point-symmetric: ua , y, a , z, a , xy, a , ub , y, c , z, c , xy, c . u (x, y , z )

 C-C, symmetric: ua , ub , vc , wc , x, c . x (x, y , z ) x
 C-C, point-symmetric: ua , ub , y, c , z, c , xy, c . v (x, y , z )
= y (x, y , z ) =
y
(x, y , z )
4.2. Solution procedure xy
u (x , y , z ) v (x , y , z )
+
y x (97)
To have the buckling solution, the energy method is followed.
The main steps are summarized as follows. The material follows Hooke's law:
The u(x,y), v(x,y) and w(x,y) displacement functions are given E
E
above. As an illustration, let us express v(x,y) for the rst element, 2 2
0
in case of C-C restraints and symmetrical displacements. The dis- 1 1
D= E E
placement function for one nite element is given in general by 1 2 1 2 0
Eq. (34). By using the notations of Fig. 5, it can be written as fol-
0 0 G (98)
lows:
 S. dny / Thin-Walled Structures 105 (2016) 135146 143

Table 1
Summary of constraint equations for the example.

Actual constraint equations Reference to

eqs.

u1=ua +z, a b/2 u2=ua z, a b/2 Eq. (53)

v1=va v2=va Eq. (48)
w1=wax, a b/2 w2=wa+x, a b/2 Eq. (67)
x,1=x, a x,2=x, a Eq. (66)
y,1= y, axy, a b/2 y,2= y, a+xy, a b/2 Eq. (71)
z,1=z, a z,2=z, a Eq. (50)
xy,1=xy, a xy,2=xy, a Eq. (70)
Eq. (54)
u3=ub ( 1.5va z , a 1.5vc z , c
a
+
4

a
+
4 ) b
2
u4=ub + ( 1.5va z , a 1.5vc z , c
a
+
4

a
+
4 ) b
2
u5=uc +z, c b/2 u6=uc z, c b/2 Eq. (53) (55)
v5=vc v6=vc Eq. (48) (49)
w5=wc x, c b/2 w6=wc +x, c b/2 Eq. (67)(69)
x,5=x, c x,6=x, c Eq. (66) (68)
y,5= y, c xy, c b/2 y,6= y, c +xy, c b/2 Eq. (71) (73)
z,5=z, c z,6=z, c Eq. (50) (51)
xy,5=xy, c xy,6=xy, c Eq. (70)(72)
Eq. (54)
u7=ud ( 1.5vc z , c 1.5ve z , e
a
+
4

a
+
4 ) b
2
u8=ud + ( 1.5vc z , c 1.5ve z , e
a
+
4

a
+
4 ) b
2
u9=ue +z, e b/2 u10=ue z, e b/2 Eq. (55)
v9=ve v10=ve Eq. (49)
w9=wex, e b/2 w10=we+x, e b/2 Eq. (69)
x,9=x, e x,10=x, e Eq. (68)
y,9= y, exy, e b/2 y,10= y, e+xy, e b/2 Eq. (73)
z,9=z, e z,9=z, e Eq. (51)
xy,9=xy, e xy,10=xy, e Eq. (72)

The internal potential, that is the strain energy is expressed as which the values of critical load can be determined.
follows:
t /2 L /2 b /2 4.3. Flexural and torsional buckling of columns
2
1 T
int = i D i dxdydz
2 The loading is a concentric axial force, which is assumed to
i = 1 t /2 0 0 (99)
distribute uniformly over the end sections.
The external energy is the negative of the work done by the
F
external loading on the second-order displacements. The external px =
bt (104)
loading is either uniformly distributed over the cross-section or
linearly changing with y. In case of S-S restraints and symmetric displacements, there
The second-order strain can be expressed as: are 8 independent displacements parameters. This means that Eq.
(103) leads to a (homogeneous) system of linear equations con-
1 v (x, y , z ) 2 w (x, y , z ) 2 sisting of 8 equations, to which maximum 8 eigen-values, there-
x II = +
2 x x (100) fore, maximum 8 buckling modes might belong. However, if the
coefcient matrix is rank-decient, the number of existing modes
or
might be less than 8, too. Similarly, there are maximum 8 point-
1 u (x, y , z ) 2 v (x, y , z ) 2 w (x, y , z ) 2 symmetric buckling modes for the S-S problem. Since the sym-
x II = + +
2 x x x (101) metric and point-symmetric displacements are certainly linearly
independent of one another, this nally means that maximum 16
The rst formula, i.e. option a) is used in classical (beam- buckling modes might exist.
model-based) stability solutions, while the second formula is ty- In case of C-C restraints there are 5 independent displacement
pically used in shell models (such as FSM or shell FEM). These parameters, which means that Eq. (103) leads to 5 equations for
options are discussed in detail in [28,29]. Here, both options will both the symmetric and point-symmetric cases, therefore, the
be considered. maximum number of buckling modes is 10.
The external potential thus:
t /2 b L /2 4.3.1. Option a)
2
II In case of S-S with option a) the equation system determined
ext = x, i px dxdydz
by Eq. (103) is rank-decient, and there are only 6 solutions for the
i = 1 t /2 0 0 (102)
symmetrical modes and 6 solutions for the point-symmetrical
In equilibrium the total potential is stationary, thus: modes as follows:
(int + ext )
=0
(103)
 2 symmetrical and 2 point-symmetrical solutions in minor-axis

buckling mode
which leads to a homogeneous system of linear equations. Non-  2 symmetrical and 2 point-symmetrical solutions in major-axis
trivial solution exists if the coefcient matrix is stationary, from buckling mode
144 S. dny / Thin-Walled Structures 105 (2016) 135146
 2 symmetrical and 2 point-symmetrical solutions in pure tor-
sional buckling mode.
Therefore, 2 symmetric and 2 point-symmetric modes are non-
existent in S-S with option a). It can be observed from the so-
lutions, that the existing modes are independent of the long-
itudinal displacements. In other words, in option a) the long-
itudinal displacements do not lead to buckling.
In case of C-C with option a) there are 6 solutions as follows:
 1 symmetrical and 1 point-symmetrical solution in minor-axis
buckling mode
 1 symmetrical and 1 point-symmetrical solution in major-axis
buckling mode
 1 symmetrical and 1 point-symmetrical solution in pure tor-
sional buckling mode.
Therefore, 2 symmetric and 2 point-symmetric modes are non-
existent in C-C with option a). Again, the non-existing modes are
associated with the longitudinal displacements.
The modes are illustrated in Fig. 7, where the distribution of the
characteristic displacement is shown along the member length.
The characteristic displacement is v in case of major-axis buckling,
w in case of minor-axis buckling, and x in case of torsional
buckling.
In case of exural buckling the solution (i.e., the expression for
the critical force) can be written as:
EI
Fcr, F =c
L2 (105)
where c is a constant depending on the end restraint conditions
and the buckling shape, while EI is the exural rigidity about the
relevant axis.
In case of pure torsional buckling the solution (i.e., the ex-
pression for the critical force) can be written as:
1 cEI
Fcr, T = GIT + 2w
r02 L (106)
where c is a constant depending on the end restraint conditions
and the buckling shape, while EIw is the warping rigidity, GIt is the
(St Venant) torsion rigidity and r0 is the polar radius of gyration.
Considering both the symmetric and point-symmetric cases,
there are altogether 4 solutions for minor-axis and 4 solutions for
major-axis exural buckling, and 4 solutions for pure torsional
buckling in case of S-S, and 2 minor-axis solutions, 2 major-axis
solutions, and 2 pure torsional solutions in case of C-C. The
Fig. 7. Distribution of the characteristic displacements for exural and torsional buckling.
relevant c constants are summarized in Table 2, where the exact c
values are also given from the well-known classic analytical
solutions.
Here are some comments.
 The applied discretization provides only the 4 lowest critical
forces (and corresponding buckled shapes) for a certain mode
(e.g. minor-axis mode) in case of S-S, while only the lowest
2 critical forces in case of C-C end restraint conditions. Ob-
viously, higher-order modes exist and could be found by ap-
plying mode elements (i.e., a ner discretization).
 The lowest critical forces can reasonably well approximated
even by the applied two nite elements, both in S-S and C-C.
The error of approximation of the lowest critical values is less
than 1%. For critical forces associated with higher modes the
tendency is obvious: the higher the mode, the larger the error.
 The error is caused solely by the longitudinal shape functions. It
is known that the sine-cosine functions belong to the exact
solutions, which are approximated by the Hermite cubic poly-
nomials in case of FEM.
 It is to note that solution for the lowest critical force for exural
buckling in case of S-S with using exactly the same approximate
longitudinal function can be found in the literature, with exactly
the same results.
 The EI exural rigidity in classical solutions is equal to Ebt3/12
and Eb3t/12 for minor-axis and major-axis buckling, respec-
tively. In case of the presented FEM solution the rigidities are
Ebt3/12/(1-2) and Eb3t/12(1-2), respectively, where is the
Poisson's ratio. The 1/(1-2) term is due to the fact that the
presented FEM solution is based on shell model, and the global
modes are dened with no transverse (membrane) extension,
which is resulted in a (virtual) increase of axial and bending
rigidities. The same phenomenon has already been reported
and discussed e.g. in [28,30].

The EIw warping rigidity in classical solutions is equal to
Eb3t3/144 (for the given rectangular cross-section, taking the
second-order warping into account). In case of the presented
FEM solution the rigidity is Eb3t3/144/(1-2). The 1/(1-2) term
is due to the restrained (membrane) transverse strain, as dis-
cussed above.
4.3.2. Option b)
The consideration of the (u (x, y, z ) /x )2 term in the second-
order longitudinal strain will change the equations systems from
 S. dny / Thin-Walled Structures 105 (2016) 135146 145

Table 2
Critical forces for exural and pure torsional buckling.

End restr. Symmetry nr of half waves Exact c FEM approximation c

S-S Sym 1 2 9.8696 9.9438

S-S Point-sym 2 42 39.478 48.00
S-S Sym 3 92 88.826 128.72
S-S Point-sym 4 162 157.91 240.00
C-C Sym 1 42 39.478 40.00
C-C Point-sym 2 80.763 120.00

Eq. (103). It turns out that the system is not rank-decient any
more, therefore, the number of buckling modes is equal to the
number of equations: 16 in case of S-S, and 10 in case of C-C. The
4 additional buckling modes (e.g., two symmetrical, two point-
symmetrical) involve longitudinal displacements only, that is why
they can readily be referred as to axial modes. These axial modes
Fig. 8. Distribution of the characteristic u displacement for axial buckling.
are independent of the end restraints, due to the fact C support is
dened so that the longitudinal displacement is not restrained,
 In case of the analytical solution for the axial mode presented in
therefore, S-S and C-C cases are practically identical from the
[28], the longitudinal shape function is assumed to be cosine
longitudinal displacements point-of-view.
(with one or multiple waves). In case of FEM, there are two
The axial modes are illustrated in Fig. 8, where the distribution
identical critical forces, which means that the buckled shape
of the characteristic displacement, i.e., u displacement is shown
can be any arbitrary linear combination of the corresponding
along the member length.
shape functions (with the ua and ub and ud and ue displacement
The critical forces that belong to the additional (axial) modes
parameters).
are identical for all the 4 modes, and can be given as follows:
 The option b) critical force formulae for F and T buckling are in
EA full accordance with the earlier ndings of [28]. Note, the here
Fcr, A=
1 2 (107) presented option b) formulae correspond to the yyy case of [28].
 However, it is to highlight that the earlier formulae have been
where A is the cross-sectional area (i.e., in this case bt), and is
derived by using the exact longitudinal shape functions, while
the Poisson's ratio.
here the same formulae are found for approximate longitudinal
The consideration of the (u (x, y, z ) /x )2 term will have some
shape functions.
effect on the critical force formulae to exural and pure torsional
 Again, in [28] only S-S end restraint has been considered,
modes, too, as follows.
however, here the formulae are found to be valid for other
The exural modes are changed in accordance with the fol-
(namely C-C) cases, too.
lowing formula:
1 4.4. Lateral-torsional buckling of beams
Fcr, F (b)= 1 1
+
Fcr , F (a) Fcr , A (108)
The loading is a concentric end moment (about the strong axis
The pure torsional modes are changed in accordance with the of the beam), which is assumed to distribute over the end sections,
following formula: linearly changing with y.
1 M b
Fcr, T (b)= px = y
1
+
1 Fw
b3t /12 2 (112)
Fcr , T (a) Fcr , A F w + Ft (109)
For the sake of simplicity, only option a) is discussed here.
where
In case of S-S there are only 8 existing solutions, 4 plus/minus
EIw pairs. In case of C-C there are only 4 existing solutions, 2 plus/
Fw=c
( 12) L2 (110) minus pairs. The critical moments can be written as follows:

cEI GIT + cEIw
Mcr = Fcr, F ( Ft + Fw ) =
Ft =GIt (111) ( 1 ) L
2 2
( 1 ) L
2 2
(113)
where Iw is the warping constant (with considering secondary where Fcr , F is the exural critical force to minor-axis buckling, and
warping) and It is the torsion inertia and c is the same constant the c constants are exactly identical to those given in Table 2.
listed in Table 2. Here are some comments.
Some comments:
 The applied discretization provides only the 4 lowest critical
 The additional modes are characterized by no exural dis- moments (and corresponding buckled shapes) in case of S-S,
placement, that is the axis of the member remains straight while only the lowest 2 critical moments in case of C-C end
during the buckling, while the longitudinal displacements are restraint conditions. Obviously, higher-order modes exist and
non-zero. This type of mode is discussed in [28, 29], called axial could be found by applying more elements (i.e., a ner
mode. discretization).
 In case of axial mode the value of the critical force is in-  The lowest critical moments can reasonably well be approxi-
dependent of whether the end restraint is S-S or C-C. mated even by the applied two nite elements, both in S-S and
146 S. dny / Thin-Walled Structures 105 (2016) 135146
C-C. The error of approximation of the lowest critical values is
less than 1%. For critical moments associated with higher modes
the tendency is obvious: the higher the mode, the larger the
error.
 The error is caused solely by the longitudinal shape functions. It
is known that the sine-cosine functions belong to the exact
solutions, which are approximated by the Hermite cubic poly-
nomials in case of FEM.
 The appearance of the 1/(1-2) term is due to the restrained
(membrane) transverse strain, as discussed above.
5. Concluding remarks
In this paper a novel shell nite element is introduced. The
introduced element is specically intended to be used in con-
strained nite element method. The novel nite element is derived
from the semi-analytical nite strip method, by keeping the
transverse interpolation functions of FSM, but changing the long-
itudinal interpolation functions into classic polynomials. As it is
proved, the proper selection of the longitudinal polynomial in-
terpolation functions makes it possible to apply essentially the
same constraining technique that is used in constrained nite strip
method.
This paper focuses on the novel nite element. The derivation
of the new element is shown in detail. It is proved that the im-
portant mechanical criteria of the constraining procedure can di-
rectly and easily be implemented into the newly proposed ele-
ment. Some basic semi-analytical examples are provided to illus-
trate the applicability of the proposed shell nite element.
The new nite element can be used to model thin-walled
structural members. If a highly regular mesh is used for the dis-
cretization, the constraining can be done, that is modal decom-
position of the deformation modes can be completed, essentially
identically to how cFSM solves the modal decomposition. Hence,
the method that uses the here-introduced nite element can
properly be referred as to constrained nite element method
(cFEM).
Since the so-dened cFEM method is essentially a shell FEM, it
can handle a wide range of practical problems together with
modal decomposition: members with holes, various loading and
restraint conditions, shear buckling, web crippling, certain cross-
section changes along the length. To all these problems rst- or
second-order static analysis, linear buckling analysis, dynamic
analysis, etc. can be completed, with or without modal decom-
position. Some preliminary results of such cFEM calculations have
already been reported in [31]. More details and various applica-
tions are going to be published in subsequent papers in the near
future.
Acknowledgments
The work was conducted with the nancial support of the
OTKA K108912 project of the Hungarian Scientic Research Fund.
References
[1] Y.K. Cheung, Finite strip method in the analysis of elastic paltes with two
opposite ends simply supported, Proc. Inst. Civ. Eng. 40 (1968) 17.
[2] Y.K. Cheung, Finite Strip Method in Structural Analysis, Pergamon Press, Uni-
ted Kingdom, 1976.
[3] G.J. Local Hancock, Distortional, and lateral buckling of I-beams, ASCE J. Struct.
Eng. 104 (11) (1978) 17871798.
[4] B.W. Schafer, Cold-Formed Steel Behavior and Design: Analytical and Nu-
merical Modeling of Elements and Members with Longitudinal Stiffeners (PhD
Dissertation), Cornell University, Ithaca, NY, 1997.
[5] S. dny, Buckling mode classication of members with open thin-walled
crosssections by using the Finite Strip Method, Research Report, Johns Hop-
kins University, 2004. (available at) http://www.ce.jhu.edu/bschafer.
[6] S. dny, B.W. Schafer, Buckling mode classication of members with open
thin-walled cross-sections, Proceedings of the Fourth International Conference
on Coupled Instabilities in Metal Structures, (Rome,Italy); Sept 2729, 2004.
pp. 467476.
[7] S. dny, B.W. Schafer, Buckling mode decomposition of single-branched open
cross-section members via Finite Strip Method: derivation, Thin-Walled
Struct. 44 (5) (2006) 563584.
[8] S. dny, B.W. Schafer, Buckling mode decomposition of single-branched open
cross-section members via Finite Strip Method: application and examples,
Thin-Walled Struct. 44 (5) (2006) 585600.
[9] B.W. Schafer, S. dny, Buckling analysis of cold-formed steel members using
CUFSM: conventional and constrained nite strip methods, Proceedings of
18th International Specialty Conference on Cold-Formed Steel Structures,
October 2628, 2006, Orlando, USA, pp. 3954.
[10] 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.
[11] Z. Li, B.W. Schafer, Finite Strip Stability Solutions for General Boundary Con-
ditions and the Extension of the Constrained Finite Strip Method, in Trends in
Civil and Structural Engineering Computing.. Stirlingshire, UK: Saxe-Coburg
Publications, 2009.
[12] Z. Li, B.W. Schafer, Buckling analysis of cold-formed steel members with
general boundary conditions using CUFSM: Conventional and constrained -
nite strip methods, Proceedings of the 20th International Specialty Conference
on Cold-Formed Steel Structures - Recent Research and Developments in Cold-
Formed Steel Design and Construction, p. 17, 2010.
[13] 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.
[14] S. dny, B.W. Schafer, Generalized constrained nite strip method for thin-
walled members with arbitrary cross-section: secondary modes, orthogon-
ality, examples, Thin-Walled Struct. 84 (2014) 123133.
[15] S. dny, Decomposition of in-plane shear in thin-walled members, Thin-
Walled Struct. 73 (2013) 2738.
[16] M. Djafour, H. Dib, M. Djelli, N. Djafour, M. Matallah, D. Zendagui, Buckling
Mode Decomposition of Thin-Walled Members using a Constrained Spline
Finite Strip Method, Proceeding of the 6th International Conference on Cou-
pled Instabilities in Metal Structures, (Glasgow, Scotland), Dec 35, 2012. (Eds:
J Loughlan, D H Nash, J Rhodes), pp 467474.
[17] M. Casafont, F. Marimon, M.M. Pastor, Calculation of pure distortional elastic
buckling loads of members subjected to compression via the nite element
method (June-July), Thin-Walled Struct. 47 (67) (2009) 701729.
[18] M. Casafont, F. Marimon, M.M. Pastor, M. Ferrer, Linear buckling analysis of
thin-walled members combining the Generalised Beam Theory and the Finite
Element Method, Comput. Struct. 89 (2122) (2011) 19822000.
[19] R. Schardt, Verallgemeinerte Technische Biegetheorie, Springer Verlag, Berlin,
Heidelberg, 1989.
[20] P.B. Dinis, D. Camotim, N. Silvestre, GBT formulation to analyse the buckling
behaviour of thin-walled members with arbitrarily branched open cross-
sections (January), Thin-Walled Struct. 44 (1) (2006) 2038, ISSN 02638231.
[21] 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, Comput. Mech. 46 (5) (2010) 759781.
[22] R. Bebiano, R. Gonalves, D. Camotim, A cross-section analysis procedure to
rationalise and automate the performance of GBT-based structural analyses
(July), Thin-Walled Struct. 92 (2015) 2947, ISSN 02638231.
[23] M. Casafont, J. Bonada, M.M. Pastor, F. Roure, GBT calculation of distortional
and global buckling of cold-formed steel channel columns with multiple
perforations, Eighth International Conference on Advances in Steel Structures,
Lisbon, Portugal, July 2224, 2015, paper no: 87, p. 18.
[24] J. Cai, C.D. Moen, Generalized beam theory buckling analysis for members
with holes, Eighth International Conference on Advances in Steel Structures,
Lisbon, Portugal, July 2224, 2015, paper no: 86, p. 18.
[25] CUFSM: Elastic Buckling Analysis of Thin-Walled Members by Finite Strip)
Analysis. CUFSM v3.12, 2006. http://www.ce.jhu.edu/bschafer/cufsm.
[26] CUFSM: Elastic Buckling Analysis of Thin-Walled Members by Finite Strip)
Analysis. CUFSM v4.05, 2012. http://www.ce.jhu.edu/bschafer/cufsm.
[27] Z. Li, M.T. Hanna, S. dny, B.W. Schafer, Impact of basis, orthogonalization,
and normalization on the constrained nite strip method for stability solu-
tions of open thin-walled members, Thin-Walled Struct. 49 (9) (2011)
11081122.
[28] S. dny, Global buckling of thin-walled columns: analytical solutions based
on shell model, Thin-Walled Struct. 55 (2012) 6475.
[29] S. dny, D. Visy, Global buckling of thin-walled columns: numerical studies,
Thin-Walled Struct. 54 (2012) 8293.
[30] S. dny, N. Silvestre, B.W. Schafer, D. Camotim, GBT and cFSM: two modal
approaches to the buckling analysis of unbranched thin-walled members, Int.
J. Adv. Steel Constr. 5 (2) (2009) 195223.
[31] S. dny, Constrained nite element method: demonstrative examples on the
global modes of thin-walled members, Proceedings of the 22nd International
Speciality Conference on Cold-Formed Steel Structures, Nov 56, 2014, (St.
Louis, USA (eds: R.A. LaBoube, W.W. Yu), 2014, pp. 6782.