Computers and Structures 89 (2011) 2337–2351

Contents lists available at ScienceDirect

Computers and Structures

journal homepage: www.elsevier.com/locate/compstruc

A geometrically exact nonlinear finite element for composite closed section

thin-walled beams
C. Martín Saravia ⇑, Sebastián P. Machado, Víctor H. Cortínez
Centro de Investigación en Mecánica Teórica y Aplicada, CONICET, Universidad Tecnológica Nacional, Facultad Regional Bahía Blanca,
11 de Abril 461, 8000 Bahía Blanca, Argentina

a r t i c l e i n f o a b s t r a c t

Article history: A geometrically nonlinear beam finite element for composite closed section thin-walled beams consider-
Received 17 January 2011 ing arbitrary displacements and rotations is presented. The virtual work equations are written as a func-
Accepted 21 July 2011 tion of nine generalized strain components, which are parametrized in terms of the director field and its
Available online 21 August 2011
derivatives. The presented finite element is valid for both isotropic and anisotropic materials. The pro-
posed approach could be attractive to be used in optimization problems of composite thin-walled beams
Keywords: with finite deformation such as helicopter rotor blades and wind turbine blades. It is shown that the pro-
Thin-walled
posed formulation has an excellent correlation against shell finite elements.
Finite
Rotations
Ó 2011 Elsevier Ltd. All rights reserved.
Composite
Beam
Closed

1. Introduction A geometrically exact beam theory must be capable of provid-

ing exact relations between the configuration and strains, in a fully
The use of thin walled beams in different areas of engineering consistent manner with the virtual work principle regardless of the
increased fast since mathematical methods for determining its magnitude of the kinematic variables chosen to parametrize the
structural behavior were developed. With the use of the finite ele- configuration. It is well known that the treatment of finite rota-
ment method different formulations for beams have been reported, tions constitutes the main difficulty of a geometrically exact non-
but some subjects, such as the inclusion of anisotropic materials in linear formulation. In this work, we choose a set of variables to
geometrically exact thin-walled beam theories, have not been fully represent exactly the kinematic behavior of the thin-walled beam
treated yet. and derive the relationships between the strains and the
With the recent technological advances in composite materials, configuration.
the use of composite structures in the design of mechanical sys- Several authors have studied geometrically exact beam finite
tems has been continuously growing. Due to their outstanding element formulations after the works of Reissner [1] and Antman
engineering properties, such as high strength/stiffness to weight [2] appeared in the early 1970s. Reissner provided a 2D exact beam
ratios and favorable fatigue characteristics, thin-walled beams theory capable of describing arbitrary large displacements and
made of composite materials are widely used in the design of air- rotations and a 3D theory for second order rotations [3].
craft wings, helicopter rotor blades, wind turbine blades and the An extensive and detailed study of finite rotations was made by
like. As a result, the structural design concepts have changed sub- Argyris [4]. Besides describing key aspects of spatial and material
stantially, and a complete understanding of the nonlinear behavior rotations he also clarifies the so called semitangential rotations,
of structures that work under extreme load conditions is essential. for which commutativity holds.
Composite closed section thin-walled beams undergoing large Updated and total Lagrangian formulations valid for large dis-
displacements and rotations can be found in several engineering placements and based on a degenerate continuum concept were
applications. The analysis of the kinematic behavior and also the presented by Bathe and Bolourchi [5].
load carrying capacity of these members demand a deep knowl- It were Simo [6] and Simo and Vu-Quoc [7,8] who first devel-
edge of the methods to treat geometrical nonlinearities. Indeed, oped a 3D geometrically exact formulation for isotropic hyperelas-
in the study of the geometrically nonlinear behavior of beams, tic beams. They used the Reissner relationships between the
the correct understanding of finite rotations plays a crucial role. variation of the rotation tensor and the infinitesimal rotations to
derive the strain-configuration relations, maintaining the geomet-
⇑ Corresponding author. ric exactness of the theory. Simo [6] parametrized the finite rota-
E-mail address: [email protected] (C.M. Saravia). tions with quaternions, aiming to enhance the computational

0045-7949/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2011.07.009
2338 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351

efficiency of the algorithm. He proposed a multiplicative update done by Librescu [32]. He used different analytical approaches to
procedure for the rotational changes, obtaining a non-symmetric treat composite beams undergoing moderate rotations, treating
tangent stiffness. rotation variables in a vectorial fashion. Piovan and Cortı́nez [33]
In contrast, Cardona and Geradin [9] presented a different and Machado and Cortı́nez [34] presented a formulation for com-
parametrization, using the total rotational pseudo-vector to update posite beams undergoing moderate rotations. Both formulations
the 3D rotations on the basis of the initial configuration. They also rely on an assumed displacement field, considering ‘‘vectorial’’
update the configuration on the basis of the last converged config- rotation variables up to a certain order.
uration (in what could be understood as an updated Lagrangian ap- Although approximations to the rotation tensor in ‘‘vectorial’’
proach). This additive treatment of the rotational degrees of rotations formulations can give good results in some cases, the
freedom gives rise to a symmetrical tangent stiffness. An isotropic treatment of rotations variables using vector operations (such
hyperelastic constitutive law was assumed. as additivity) modifies the actual meaning of these variables.
Simo and Vu-Quoc [10] incorporated shear and torsion warping This is, the use of vectorial calculus to obtain the variation of
deformation effects in his geometrically exact model. the rotation tensor, which actually belong to the SO(3) manifold,
An extension of the formulation of Simo to curved beams was introduces and approximation to the formulation. Thus, a
presented by Ibrahimbegovic [11]. He extended the formulation geometrically exact beam theory which has been reduced in
to arbitrary curved space beams maintaining some key aspects of order by using, for instance, a second order rotation tensor
Simo formulation but using hierarchical interpolation. He also pro- differs from the cited theories because additional terms appears
posed an incremental rotation vector formulation [12] to solve the in the evaluation of the variation of the rotation tensor. It can be
nonlinear dynamics of space beams. said that information about the kinematic behavior is lost
The use of the Green–Lagrange strain measures in a geometri- when using the Gateaux derivative instead of the Lie
cally exact finite element formulation for 3D beams seems to have derivative.
been first introduced by Gruttmann et al. [13,14]. Also, Auricchio This work presents an implementation of the classical thin-
et al. [15] reviewed the Simo theory making equivalence between walled beam theory in a geometrically exact (configuration based)
Green–Lagrange strain measures and Reissner strain measures. finite element formulation. The rotation variables are treated ex-
An extensive and detailed review of several aspects of existent actly, and thus the varied configuration is obtained considering
geometrical nonlinear finite element formulations of beams was that the rotation variables belong to a SO(3) manifold.
presented by Crisfield [16], describing essential aspects of geomet- This formulation is based on an assumed displacement field
rical nonlinear beam theories and focusing the attention in the that, as said before, describes the kinematic behavior of the beam
treatment of finite rotations. Additional reviews and clarification regardless of the magnitude of displacements and rotations. Fol-
of concepts of existing beam formulations can also be found in ref- lowing the classical procedure of the theory of thin walled beams,
erences [17–19]. we develop the expressions for the Green–Lagrange (GL) strains
During the last years, great efforts were made to shed light to and then transform these strains to a curvilinear coordinate sys-
the problem of loss of objectivity introduced by the interpolation tem. To ease that process, we first express the GL strains in terms
of rotations variables, a problem first noted by Crisfield and Jelenic of generalized strains by the introduction of a linear transforma-
[20]. Jelenic and Crisfield [21] implemented the ideas proposed in tion. These generalized strains are written in terms of the director
[20] to complete de development of a frame-invariant and path vectors, its derivatives and the derivatives of the position vector of
independent geometrically exact 3D beam element. the center of reduction (the pole). This leads to a remarkable sim-
Also Ibrahimbegovic and Taylor [22] re-examine the geometri- ple expression for the curvilinear strains.
cally exact models to clarify the frame invariance issues concerning Extracting from the GL strains the functions that describe the
multiplicative and additive updates of rotations. Betsch and Stein- cross section shape we can write the virtual work only in terms
mann [23], Armero and Romero [24] and Romero and Armero [25] of generalized strains and generalized beam forces. The equations
further contributed to the subject presenting frame-invariant for- of motion are thus parametrized in terms of the director field
mulations for geometrically exact beams using the director field and their variations.
to parametrize the equations of motion. Additional treatment of The parametrization of finite rotations is done using the spa-
frame invariance can be found in references [26,27]. tial spin, obtaining a natural relationship between the director
The inclusion of anisotropic materials to thin-walled and also field and the configuration. This spin components are then
solid beam finite element formulations was extensively studied interpolated using linear interpolation and the rotation tensor
by Hodges et al. [28], his work is based on the powerful variational is calculated and updated for each iteration only at the integra-
asymptotic method (VAM) and deserves special attention. Besides tion points. The update of the derivatives of the director field is
several interesting developments, he and his coworkers developed performed iteratively on the basis of the last iterative configura-
a geometrically-exact, fully intrinsic theory for the dynamics of tion, eliminating the necessity of calculating the total rotation
curved and twisted composite beams, having neither displacement tensor.
nor rotation variables appearing in the formulation. Using the The present formulation is based on the classical thin-walled
VAM, a generalized Vlasov theory for composite beams based on beam theory and thus can handle any type of composite material
the variational asymptotic beam sectional analysis was also pre- without the necessity of performing a 2D finite element cross sec-
sented by Yu et al. [29]. These developments were helped by the tional analysis to obtain the constitutive matrix. This leads to an
variational asymptotic beam sectional analysis software (VABS) interesting approach for addressing optimization problems.
[30], a tool for obtaining thin-walled composite beams sectional
properties. VABS is based on a 2D finite element analysis of the 2. Kinematics
cross section to obtain the stiffness matrix of the underlying 1D
theory. 2.1. Generalities
Recently Gonçalves et al. [31] presented a large displacement
and finite rotation thin-walled beam formulation for thin-walled Consider two states of the beam, an undeformed reference
isotropic beams considering cross-section deformation. state, denoted by B0 , and a deformed state, denoted by B, as shown
An extensive review on analytical methods for solving geomet- in Fig. 1. Being ai a spatial frame of reference, we define a reference
rically nonlinear problems of composite thin-walled beams was frame Ei and a current frame ei (both frames being orthonormal).
 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351 2339

Remember that, for simplicity, we adopted a reference configu-

ration such that K0 = I (the initial configuration is assumed to be
straight). Therefore, in the reference configuration the axial direc-
tor vector coincides with the reference tangent vector X,1. The con-
sideration of transverse shear strains precludes the possibility of
obtaining the same conclusion for the current tangent vector, so
in general:
e1  x;1 > 0: ð6Þ
The last equation can be understood as the definition of a Timo-
shenko beam. It must be noted that since the present formulation
is thought to be used in cases where the thin-walled beam is slen-
der, has a closed cross section and torsional loads are small com-
pared with flexural loads we assume that the cross-section is
rigid on its own plane, i.e. warping is not considered.
Fig. 1. 3D beam kinematics.
2.2. Parametrization of finite rotations

According to the kinematic description presented previously,

The displacement of any point in the deformed beam measured the beam configuration space consists on a linear space of three-
with respect to the undeformed reference state can be expressed dimensional vectors u0 and a nonlinear manifold of transforma-
in the global coordinate system ai in terms of a vector tions K (which belong to a special orthogonal Lie group SO(3)).
u = (u1, u2, u3). This transformation is described mathematically via the exponen-
The current frame ei is a function of a running length coordinate tial map as:
along the reference line of the beam, denoted as x, and is fixed to
the beam cross-section. For convenience, we choose the reference sin h 1  cos h
KðhÞ ¼ cos hI þ Hþ h  h; ð7Þ
curve C to be the locus of cross-sectional inertia centroids. The ori- h h2
gin of ei is located on the reference line of the beam and is called:
where h is the so-called rotation vector, h its modulus and H is its
pole. The cross-section of the beam is arbitrary and initially normal
skew symmetric matrix (often called spinor [18]), this is:
to the reference line of the beam. 2 3
The relations between the orthonormal frames are given by the 0 h3 h2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
6 7
linear transformations: h ¼ ½h1 h2 h3 T ; H ¼ 4 h3 0 h1 5; h¼ h21 þ h22 þ h23 :
Ei ¼ K0 ðxÞai ; ei ¼ KðxÞEi ; ð1Þ h2 h1 0
ð8Þ
where K0(x) and K(x) are two-point tensor fields 2 SO(3); the spe-
cial orthogonal (Lie) group. Thus, it’s satisfied that: According to Euler’s theorem, when a rigid body rotates from
one orientation to another, which may be the result of a series of
KT0 K0 ¼ I; KT K ¼ I: ð2Þ rotations (with one rotation superposed onto the previous), the to-
tal rotation can be seen as single (compound) rotation about some
Throughout this work we will consider that the beam element is
spatial fixed axis (see e.g. [4]). Therefore, the rotation vector can be
straight, so we set K0 = I, being I the identity matrix.
understood as a compound rotation that globally or totally para-
Considering the relations (1), we express the position vectors of
metrizes the compound rotation tensor via (7).
a point in the beam in the undeformed and deformed configuration
As stated before, rotation transformations are objects belonging
respectively as:
to a non linear manifold. Thus, rotations variables are not commu-
X
3
tative and consequently vector operations, such additivity, cannot
Xðx; n2 ; n3 Þ ¼ X 0 ðxÞ þ ni E i ; be applied.
i¼2
There are different ways to parametrize finite rotations, Euler
X
3
xðx; n2 ; n3 ; tÞ ¼ x0 ðx; tÞ þ ni ei : ð3Þ angles, a four parameter quaternion intrinsic representation
i¼2 [6,11], a three parameter rotational vector [9], etc. These paramet-
rizations can be total or incremental, as well as their combinations,
where in both equations the first term stands for the position of the leading to multiplicative or additive updating procedures.
pole and the second term stands for the position a point in the cross Parametrization with the rotation tensor as done in [7] leads to
section relative to the pole. Note that, x is the running length coor- a non-symmetric tangent matrix, although symmetry is recovered
dinate and n2 and n3 are cross section coordinates. at equilibrium. This kind of parametrization has the advantage of
Also, it is possible to express the displacement field as: giving very simple expressions for the tangent matrix. On the other
X
3 hand, using the rotational vector to parametrize finite rotations
uðx; n2 ; n3 ; tÞ ¼ x  X ¼ u0 ðx; tÞ þ ðK  IÞ ni Ei ; ð4Þ leads to a symmetric tangent matrix but its derivation is much
i¼2 more complicated.
where u0 represents the displacement of the kinematic center of In this work we parametrize finite rotations with the rotation
reduction, i.e. the pole. Note that t is the pseudo-time variable. tensor, using spins as rotation variables. The update of the direc-
The set of kinematic variables is defined by three displacements u tors is performed in a multiplicative way in each iteration. With
and three spins Dw as: this scheme neither the total rotation tensor nor the incremental
rotation tensor are needed, leading to a very simple and effective
V :¼ f/ ¼ ½u; DwT : ½0; ‘ ! R3 g; algorithm.
ð5Þ It is well known the fact that this parametrization of finite
½u; DwT ¼ ½u1 ; u2 ; u3 ; Dw1 ; Dw2 ; Dw3 T : rotations leads to a nonsymmetrical tangent stiffness matrix for
2340 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351

non-equilibrium configuration. This is caused by the use of non- in the cross section (i.e. ni. Therefore, the mentioned transforma-
additive (even in the limit) spin variables, see e.g. [16]. Although tion gives:
different corrections for this problem has been proposed in the lit-
EGL ¼ De; ð13Þ
erature, for example by using additive in the limit spin variables
[16], this discussion is merely academic since ‘‘forced’’ symmetri- where the transformation matrix is:
zation of the tangent matrix leads to the same result. 2 3
1 2 1 2
It is also well known the fact that in the presence of finite three 1 n3 n2 0 0 0 n
2 2
n
2 3
n2 n3
6 7
dimensional rotations the concept of objectivity of strain measures D ¼ 40 0 0 1 0 n3 0 0 0 5: ð14Þ
does not extend naturally from the theory to the finite element 0 0 0 0 1 n2 0 0 0
formulation. Hence, despite being this formulation frame indiffer-
ent, it suffers from interpolation induced non-objectivity. Also, as And the generalized strain vector is:
every formulation based on the interpolation of iterative and 2 3 21 3
 2
x00  x00  X 00  X 00
incremental rotations it was proven to be path-dependent 6j7 6 7
7 6
6 2 x00  e03  X 00  E03 7
[20,21]. However, we show in the numerical investigations section 6 7 6 7
7 6
6 j3 x00  e02  X 00  E02 7
that this shortfall does not affect the accuracy of the solution since 6 7 6 7
6 c 7 6 7
the error that it originates for static problems is negligible. The fact 6 2 7 6 x00  e2  X 00  E2 7
6 7 6 7
that it’s difficult to guarantee that the last assertion is always valid e¼6 7 6
6 c3 7 ¼ 6 x00  e3  X 00  E3 7:
7 ð15Þ
motivates the implementation of an interpolation scheme that 6 7 6 7
6 j1 7 6 e02  e3  E02  E3 7
could lead to a frame-invariant path-independent finite element 6 7 6 7
6v 7 6 e02  e02  E02  E02 7
formulation. This implementation will be presented in a future 6 2 7 6 7
6 7 6 7
work. 4 v3 5 4 e03  e03  E03  E03 5
v23 e02  e03  E02  E03
The generalized strain vector e contains nine generalized beam
3. Strain and stress fields
strains belonging to a material description and expressed in a rect-
angular coordinate system. It must be noted that the above relation-
3.1. The strain tensor
ship was only introduced to simplify the mathematical treatment of
the formulation.
In contrast with most existing formulations for thin-walled
The derivation of strains and stresses measures for thin-walled
beams (see e.g. [32,34]) we express the GL strain tensor in terms
beams is helped by the introduction of an orthogonal curvilinear
of reference and current position derivatives. The derivatives of
coordinate system (x, n, s), see Fig. 2. The cross-section shape will
the position vectors of the undeformed and deformed configura-
be defined in this coordinate system by functions ni(n, s). The coor-
tion give:
dinate s is measured along the tangent to the middle line of the
X ;1 ¼ X 00 þ n2 E02 þ n3 E03 ; x;1 ¼ x00 þ n2 e02 þ n3 e03 ; cross section, in clockwise direction and with origin conveniently
chosen. Also, the thickness coordinate n(e/2 6 e/2) is perpendicu-
X ;2 ¼ E2 ; x;2 ¼ e2 ; ð9Þ
lar to s and with origin in the middle line contour.
X ;3 ¼ E3 ; x;3 ¼ e3 : In order to represent the GL strains in this curvilinear coordi-
Note that we have implicitly made the classical assumption of nate system we make use of the curvilinear transformation tensor:
beam theories of plane cross-sections remaining plane. Proceeding 2 3
1 0 0
with the derivation, we operate in a conventional way by injecting 6 dn2 dn3 7
the tangent vectors X,i and x,i into the GL strain tensor:
P ¼ 40 ds ds 5: ð16Þ
 dn2
0  ddsn3 ds
1
EGL ¼ ðx;i  x;j  X ;i  X ;j Þ: ð10Þ where the functions n i describe the mid-contour of the cross
2
section.
According to the kinematic hypotheses, the non-vanishing com-
Hence, the GL strain vector in the curvilinear coordinate system,
ponents of the GL strain tensor are only three, in vector notation; b GL , is obtained by transforming the rectangular GL strains as:
i.e. E
EGL ¼ ½ E11 2E12 2E13 T ; ð11Þ b GL ¼ ½Exx 2Exs 2Exn T ¼ PEGL ;
E ð17Þ
b GL ¼ PDe ¼ D
E b e: ð18Þ
where:

1  02     
E11 ¼ x0  X 02
0 þ n2 x00  e03  X 00  E03 þ n3 x00  e02  X 00  E02
2
1   1    
þ n22 e022  E2
02
þ n23 e02
3  E3
02
þ n2 n3 e02  e03  E02  E03 ;
2 2
1  
E12 ¼ x00  e2  X 00  E2  n3 e03  e2  E03  E2 ;
2
1  
E13 ¼ x00  e3  X 00  E3 þ n2 e02  e3  E02  E3 :
2
ð12Þ

where we have used the relation: e0i  ei  E0i  Ei ¼ 0.

In order to ease the derivation of the thin-walled beam strains
we introduce a new entity; the generalized strain vector e. The gen-
eralized strain vector is a vector that properly transformed gives
the GL strain vector. This transformation actually ‘‘separates’’ from
the GL strain vector the variables related to the location of a point Fig. 2. Curvilinear transformation schematic.
 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351 2341

Recalling Eqs. (14) and (15), it is found that the GL strain vector in work without modifying the classical thin-walled beam approach.
curvilinear coordinates has a remarkably simple closed expression: In this section we develop the constitutive relations for the com-
2 3 posite thin-walled beam. It should be noted that although the
 þ n2 j3 þ n3 j2 þ 12 n22 v2 þ 12 n23 v3 þ n2 n3 v23 constitutive relations here presented are valid for composite lam-
b GL 6 n0 c þ n0 c þ ðn2 n0  n3 n0 Þj1 7
E ¼4 2 2 3 3 3 2 5; ð19Þ inates with general lamination, a reduction to the isotropic case is
n03 c2 þ n02 c3 þ ðn2 n02 þ n3 n03 Þj1 trivial.
The relationship between the second Piola–Kirchhoff stress ten-
where, with an abuse of notation, we have reused the prime symbol
sor and its energetic conjugate; the GL strain tensor, can be ex-
to denote derivation with respect to the s coordinate.
pressed in curvilinear coordinates in terms of the stiffness
Proceeding in a similar way as done in Eq. (13), it is convenient
b GL the quantities related to the middle-line coor- coefficients Qij of an orthotropic lamina as [35,36]:
to separate from E
2
dinate s and the quantities related to the thickness coordinate n. rxx 3 2 Q 11 Q 12 Q 13 0 0 Q 16
32
xx 3
Before, we can refer to Fig. 2 (see also [34]) to easily verify that 6 r 7 6Q Q 26 7 6 7
6 ss 7 6 12 Q 22 Q 23 0 0 76 ss 7
the location of a point anywhere in the cross-section can be ex- 6 7 6 76 7
6 rnn 7 6 Q 13 Q 23 Q 33 0 0 Q 36 76 nn 7
pressed as: 6 7 6 76 7
6r 7¼6 0 0 0 Q 44 Q 45 0 76 c 7: ð25Þ
6 sn 7 6 76 sn 7
n2 ðn; sÞ ¼ n2 ðsÞ  nn03 ; n3 ðn; sÞ ¼ n3 ðsÞ þ nn02 ; ð20Þ 6 7 6 76 7
4 rxn 5 4 0 0 0 Q 45 Q 55 0 54 cxn 5
where ni locates a point anywhere in the cross section and n i rxs Q 16 Q 26 Q 36 0 0 Q 66 cxs
locates the points lying in the middle-line contour. Introducing
the expression (20) into the matrix D, b defined in Eq. (18), we can In matrix form:
obtain a cross sectional matrix Db that is a function only of the mid- r ¼ Q s : ð26Þ
surface coordinates  ni . The expression of this matrix is given in
Appendix A. In the above equation Qij are components of the transformed consti-
As it will be clarified in the next section, we will use five inde- tutive (or stiffness) matrix defined in terms of the elastic properties
pendent curvilinear strain measures (collected in the vector s to (elasticity moduli, shear moduli and Poisson coefficients) and fiber
describe the strain state of the thin-walled laminate (see [35]). orientation of the ply [35].
Thus, the strain state of the mid cross section will be described by: The shell stress resultants (or shell forces) in a lamina are then
defined according to:
s ¼ ½exx cxs cxn ,xx ,xs T : ð21Þ
Z e=2 Z e=2
To describe the strain state of the beam in terms of the generalized Nij ¼ rij dn; Mij ¼ rij ndn: ð27Þ
strain vector, we found a matrix T to establish the relationship be- e=2 e=2

tween the GL curvilinear strains and the generalized strains as: Employing (25) and (27) and neglecting the normal stress in the
s ¼ Te: ð22Þ thickness (i.e. rnn = 0) it is possible to obtain a constitutive relation
to relate the shell forces and the shell strains as:
Substituting (20) into (19), we obtain:
2 3 2 32
2
1 n3 n2 0 0 0 1 2
n 1 2
n n2 n3 3 Nxx A11 A12 A13 0 0 B11 B12 B16 xx 3
2 2 2 3 6 Nss 7 6 A12 B26 76 7
6
60 0 0 n0 n0 n3 n0  n2 n0 0 0 0 7
7 6 7 6 A22 A23 0 0 B12 B22 76 ss 7
6 2 3 2 3
7 6 7 6 76 7
6 Nxs 7 6 A13 A23 A33 0 0 B16 B26 B66 76 cxs 7
TðsÞ ¼ 6
60 0 0 n03 n02 n2 n0 þ n3 n0 0 0 0 7 7 6 7 6
6N 7 6 0
76 7
0 76 csn 7
76
2 3
6 7 6 sn 7 6 0 0 AH44 AH45 0 0 7
4 0 n02 n0 3 0 0 0 n2 n0
3
n3 n0
2 T49 5 6 7¼6 76 7;
 02  6 Nxn 7 6 0 AH45 AH55 0 7 6 7
0 0 0 0 0  n2 þ n02
 0 0 0 6 7 6 0 0 0 0 76 cxn 7
3 6M 7 6B B12 B16 0 0 D11 D12 D16 76 xx 7
76 j
6 xx 7 6 11 7
ð23Þ 6 7 6 76 7
 
4 M ss 5 4 B12 B22 B26 0 0 D12 D22 D26 54 jss 5
where T49 ¼  n2  n3 
n02   n03 and we have neglected higher order ef- M xs B16 B26 B66 0 0 D16 D26 D66 jxs
fects in the thickness, i.e. terms in n2.
ð28Þ
Remark 3.1. It is important to note that the matrix T can be where: Nxx, Nss, and Nxs are axial, hoop and shear-membrane shell
understood as a double transformation matrix, transforming the forces and Nxn and Nsn are transverse shear shell forces. Also; Mxx,
generalized strains into the curvilinear GL strain, defined in (21), Mss and Mxs are axial bending, hoop bending and twisting shell mo-
without the necessity of using an intermediate transformation. So, ments, respectively (see Fig. A.1 in Appendix A.1 for a graphical rep-
once T has been obtained the transformation matrix D b defined in
resentation of the shell forces and moments). The same
(18) is no longer needed. nomenclature is extended to the shell strain resultants, thus: exx
Now, it is straightforward to obtain the curvilinear strains as a and ess are axial and hoop normal strains, cxs, csn and cxn are shear
function of mid-contour quantities and the generalized strains as: shell strains and ,xx, ,ss and ,xs are axial, hoop and twisting curva-
2 3 tures respectively. The coefficients Aij ; AHij ; Bij and Dij in the constitu-
 þ j3 n2 þ j2 n3 þ 12 v2 n22 þ v23 n2 n3 þ 12 v3 n23
6   7 tive matrix are shell stiffness-coefficients that result from the
6 c2 n02 þ c3 n03 þ j1 n03 n2  n02 n3 7 integration of Qij in the thickness [35].
6   7
6
s ¼ 6    0
c3 n2  c2 n3 þ j1 n2 n2 þ n3 n3
0 0  0  7: ð24Þ Although the last relationships were derived for a single lamina,
7
6 0 7 we can obtain the constitutive relations for a laminate by spanning
4 j2 n2  j3 n3  v2 n3 n2 þ v3 n2 n3 þ v23 ðn2 n2  n3 n3 Þ 5
0 0 0 0 0
 02  the integrals in the thickness of the lamina over the different layers
j1 n2  n023
of the laminate (each layer being a single lamina). Therefore, using
the hypotheses of plane stress in the laminate and rigid cross sec-
3.2. Composite mechanics-constitutive relations tion (it can be seen that according to this hypothesis ess = cns = 0,
but in order to avoid overstiffening effects we set Nss = cns = 0
The main advantage of the present formulation is the capability [35], thus generating a mild inconsistency typical of thin-walled
of handling composite materials in a geometrically exact frame- beam formulations), the relations in (28) simplify to:
2342 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351

2 3 2 32 3
Nxx A11 A16 0 B11 B16 exx Remark 3.3. Note that the evaluation of beam constitutive matrix
6 7 6 76 7 D does not involve a 2D finite element analysis of the cross section
6 Nxs 7 6 A16 A66 0 B16 B66 76 cxs 7
6 7 6 76 7 (as, for example, in the VABS approach [30]). Although the consti-
6 Nxn 7 ¼ 6 0 AH55 0 7 6 7
6 7 6 0 0 76 cxn 7; ð29Þ
tutive constants are not as accurate that the ones obtained with the
6 7 6 b b 7 6 7
4 M xx 5 4 B11 B16 0 D 11 D 16 54 ,xx 5 latter method, the present approach is simpler, faster and it also
M xs B16 B66 0 b 16
D b 66
D ,xs opens the possibility of addressing optimization problems of large
deformation of thin-walled composite beams.
where Aij are components of the laminate reduced in-plane stiffness
matrix, Bij are components of the reduced bending-extension cou- 4. Variational formulation
pling matrix, Dij are components of the reduced bending stiffness ma-
trix and AH55 is the component of the reduced transverse shear The virtual work of external and internal forces is derived in the
stiffness matrix. next subsections. To maintain the variational formulation parame-
We can express the above relation in matrix form as: trized in terms of the director field, its admissible variation must
be found. The expression for the director variations depends on
N s ¼ C s ; ð30Þ the perturbed rotation tensor, which must be found as a function
of the rotation variables chosen to parametrize the finite rotations.
where C is the composite shell constitutive matrix and s is the cur- Once the set of kinematically admissible variations is obtained, the
vilinear shell strain vector defined in (22). generalized virtual strains can be found, so the virtual work of the
Now, it is possible to express the shell forces as a function of the internal and external forces can be derived. Therefore, the objective
generalized strains. Replacing (22) into (30) we obtain: of this section is to express the virtual work principle as a function
of the generalized virtual strain vector and its work conjugate
N s ¼ CTe: ð31Þ
beam forces vector.
The next step in the development of the beam theory is to re- The weak form of equilibrium of a three dimensional body B is
duce the 2D formulation to a 1D formulation. Since we are pursu- given by [37,38]:
ing to formulate the theory in terms of generalized quantities, we Z Z Z
need to find a one dimensional stress (or force) entity such as to be Gð/; d/Þ ¼ r  ddV  q0 b  d/dV  ðp  du þ mdwÞdX;
B0 B0 @B0
work conjugate with the generalized strains. To that purpose, we
first transform the shell forces in (31) back to the ‘‘generalized ð36Þ
space’’ by using the double transformation matrix T (see REMAK where b, p and m are: body forces, prescribed external forces and
3.1). Thus, we obtain a new entity, a sort of transformed back shell prescribed external moments per unit length respectively. d is
strain: the variation of the GL strain vector, work conjugate to the second
Piola–Kirchhoff stress tensor r. Where r could be defined in either a
N Gs ¼ TT N s ¼ TT CTe: ð32Þ rectangular or a curvilinear coordinate system, such a distinction is,
G at least here, unnecessary.
It can be observed that new entity Ns is a vector of generalized
shell stresses defined in the global coordinate system. Since it is a
4.1. Admissible variations and perturbed rotation field
function of the cross section contour, integration over the contour S
gives the vector of generalized beam forces, work conjugate to the
Now it is necessary to define the space of kinematically admis-
generalized strains, as:
sible variations in terms of the independent kinematic variables. To
Z Z

obtain the generalized strains variations we need to find first the

SðxÞ ¼ N Gs ds ¼ TT CTds eðxÞ; ð33Þ
s s
admissible variation of the basis vectors. Remembering that we
SðxÞ ¼ DeðxÞ: ð34Þ set K0 = I and recalling (1), we can write:
dei ¼ dðKðsÞEi Þ ¼ dKðxÞEi : ð37Þ
Note that since the generalized strain vector e is not a function
of the curvilinear coordinate s, see (15), it was taken out of the The admissible variation of the rotation tensor (Lie variation)
integral over the contour. Also, the matrix D was defined as: can be obtained introducing an infinitesimal virtual rotation super-
Z posed onto the existing finite rotation, see e.g. [39,40]. This virtual
D¼ TT CTds: ð35Þ rotation can lie in the tangent space at K (spatial virtual rotation)
s
or in the tangent space at I (material virtual rotation), and is repre-
The matrix D contains functions n i that define the cross section sented by a skew symmetric matrix dW or dW respectively (see
mid-contour and all the anisotropic material constants. Since the Fig. 3). Hereafter these variables will be called: ‘‘spins’’ [16].
functions ni are often defined as piecewise functions, the integral We now construct the perturbed rotation tensor by exponenti-
to evaluate D often needs to be performed in a piecewise manner ating the spatial spin as:
(see e.g. [32]). It’s of crucial importance the correct evaluation and
K ¼ expðdWÞK: ð38Þ
formulation of D since it contains not only all geometrical couplings
but also all material couplings. Now, being K a two point tensor that takes vectors from the tangent
So, the last derivations complete the formulation of the consti- space in the initial configuration to the tangent space in the current
tutive relations of the thin-walled beam theory. configuration, we can use it to relate spatial and material spins
respectively as:
Remark 3.2. It is worth noting that in contrast with most existing
dW ¼ KT dWK; dW ¼ KdWKT : ð39Þ
thin-walled beam formulations [32–34], the beam forces were not
defined but deduced from the shell forces expression. To help the From which we can write the material version of the kinematically
derivation of the beam forces a new stress entity, i.e. the general- admissible perturbed finite rotation tensor as:
ized shell stress vector, was defined (see Fig. A.1 in Appendix A for a
K ¼ KexpðdWÞ: ð40Þ
graphical representation generalized beam forces).
 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351 2343

Considering the last relationship, the set of kinematically admissi-

ble variations is defined as:

dV :¼ fd/ ¼ ½du; dwT : ½0; ‘ ! R3 jd/ ¼ 0 on Sg; ð47Þ

where S describes de boundaries with prescribed displacements
and rotations.
Additionally, we can find the variation of the director’s deriva-
tive as:
de0i ¼ dw0  ei þ dw  e0i : ð48Þ
Note that we still do not know dw0 and e0i . In the finite element
implementation, the derivative of the virtual spins will be found
by interpolation. On the other hand, the derivative of the directors
Fig. 3. Geometric interpretation of the exponential map.
is not interpolated, it is calculated by derivation of (1).
The variations of the directors and its derivatives are now used
It is interesting to note that it can be devised yet another way of to obtain the generalized virtual strains variation. Considering that
constructing the perturbed finite rotation tensor by making use of dEi = 0 and that dX 00 ¼ 0, and performing the variation to (15) we
the rotation vector, this gives obtain:
2 3 2 3
K ¼ expðH þ dHÞ: ð41Þ d x00  du0
6 dj2 7 6 e0  du0 þ x0  de0 7
6 7 6 3 0 37
Note that we are forcing the additivity property to hold. It is well 6 7 6 0 7
6 dj3 7 6 e2  du0 þ x00  de02 7
known that when using additive rotation variables the following 6 7 6 7
6 dc2 7 6 7
transformation emerges (see e.g. [12,16,18]): 6 7 6 e2  du0 þ x00  de2 7
6 7 6 7
de ¼ 6
6 dc3 7 6 0 0
7 ¼ 6 e3  du þ x0  de3 7:
7 ð49Þ
dw ¼ T w ðhÞdh; ð42Þ 6 7 6 7
6 dj1 7 6 de02 e3 þ e02  de3 7
where: 6 7 6  0 0 7
6 dv2 7 6 2 de2  e2 7
6 7 6  0 0 7
sin h 1  cos h h  sin h 6 7 6 7
T w ðhÞ ¼ Iþ Hþ h  h: ð43Þ 4 dv3 5 4 2 de3  e3 5
h h2 h3 dv23 de02  e03 þ e02  de03
The parametrization of the compound rotations using h is com-
Making use of (46) and (48) it is possible to rewrite (49) in the fol-
monly referred as vector-like parametrization and belongs to the
lowing form:
class of total Lagrangian descriptions [9,19].
These different choices in the construction of a kinematically de ¼ Hð!d/Þ; ð50Þ
admissible representation of the perturbed rotation tensor to- where ! is a matrix differential operator and H a matrix of director
gether with the type of algorithm chosen to perform the configura- vectors such as:
tion update lead to different finite element formulations, i.e. total 2 3
Lagrangian, updated Lagrangian and Eulerian formulations. Using x00 0 0
these definitions, the present formulation belongs to the class of 6 e0 e03  x00 e3  x00 7
6 3 7
6 0 7
Eulerian formulations. We will not focus attention on their differ- 6 e2 e02  x00 e2  x00 7
6 7 2 3
ences but in the implementation of the thin-walled beams theory 6e e2  x00 0 7 d
6 2 7 dx
1 0
using a geometrically exact finite element formulation. 6 7 6 7
Since the weak form of the equations of motion was parame- H¼6
6 e3 e3  x00 0 7;
7 !¼4 0 1 5: ð51Þ
6 7 d
trized in terms of the current directors and its derivatives, to ease 60 0 e2  e3 7 0 dx
1
6   7
the derivation of the virtual work it is better to use rotation vari- 60 0 2 e2  e02 7
6   7
ables that belong to the same tangent space as the directors, i.e. 6 7
40 0 2 e3  e03 5
the tangent space at K. Considering the latter, we will use the spa-
0 0 e3  e02 þ e2  e03
tial version of virtual rotations (i.e. dW) to obtain the kinematically
admissible variation of the rotation tensor. We note that this It can be seen that the description of the exact kinematic behavior
parametrization avoids the use of the tangential map, and thus of the beam can be fully described using nine generalized
the linearization of the nonlinear equilibrium equations is consid- quantities.
erably simpler. Since the tangential map, its inverse and its deriv-
atives do not play a role in the tangent stiffness the resulting Remark 4.1. This last conclusion is in contrast with several
algorithm results to be faster. As it will be clarified later, the last approximated thin-walled beam formulations where more beam
benefits are paid with loss of symmetry in the tangent stiffness higher order forces appear and the exact kinematic behavior is still
and path dependency (see. Example 7.5). Recalling (38) we can ex- not represented (see e.g. [32,34]). It seems to be that in a
press the variation of the rotation tensor in terms of the spatial geometrically exact formulation some of this higher order beam
spin as: forces cancel out.
d
dK ¼ ½expðdWÞKj¼0 ¼ dWK: ð44Þ
d Remark 4.2. The presented derivation of the virtual generalized
Again, dW is a skew symmetric matrix such as: strains in terms of the variations of the directors and its derivatives
is independent of the parametrization of finite rotations. This is a
dWa ¼ dw  a: ð45Þ very important fact since any finite rotations algorithm could be
Therefore, we can rewrite (37) as: used with the proposed formulation. Nonetheless, standard time
stepping algorithms could be implemented without any additional
dei ¼ dw  ei : ð46Þ
modifications.
2344 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351

4.2. Internal virtual work Recalling (50) it is possible to re-write the last expression as:
Z Z
Having derived the expressions for the admissible variations of Gð/; d/Þ ¼ ½Hð!d/ÞT Sdx  ðp  du0 þ m  dwÞdx: ð60Þ
‘ l
the basis vectors and strains we develop in this section the expres-
sions for the internal virtual work of the beam. Recalling (36), the Now, the equilibrium equations of the geometrically nonlinear thin-
internal virtual work of a three dimensional body can be written in walled composite beam are available in its variational form.
vector form as:
Z
Gint ð/; d/Þ ¼ dT rdV; ð52Þ 5. Linearization of the weak form
B0
The solution of the nonlinear system of equations requires the
which in the curvilinear coordinate system is:
Z Z Z linearization of variational equilibrium equations with respect to
Gint ð/; d/Þ ¼ ðdT rÞdn ds dx: ð53Þ an increment in the configurations variables. This linearization is
‘ s e obtained through the directional derivative and, assuming conser-
vative loading, its application gives two tangent terms; the mate-
We can now use the definition of the shell resultant forces (27)
rial and the geometric stiffness matrices.
to reduce the 3D formulation to a 2D formulation. Therefore, per-
Being L[G(/, d/)] the linear part of the functional G(/, d/), we
forming integration of (53) in the n direction (see (27)) we can
have:
write the internal virtual work in terms of shell quantities as:
Z Z ^ d/Þ ¼ Gð/;
L½Gð/; ^ d/Þ þ DGð/;
^ d/Þ  D/; ð61Þ
Gint ð/; d/Þ ¼ dTs N s ds dx: ð54Þ
‘ s ^ d/Þ is the unbalanced force at the config-
where the first term Gð/;
^ The Frechet differential in the second term is obtained in
uration /.
The reduction to a 1D formulation is aided by the deduction of
1D beam forces presented in (33). Transforming the virtual curvi- a standard way as:
linear shell strains into virtual generalized strains we can rewrite d
the last expression as: DGð/; d/Þ  D/ ¼ j Gð/ þ D/Þ; ð62Þ
d ¼0
Z Z

Gint ð/; d/Þ ¼ deT TT N s ds dx ð55Þ where D/ fulfills the geometrical boundary conditions. Applying
‘ s the definition (62) and recalling (57) and (49), we obtain the tan-
gent stiffness as:
In which the term in parentheses is the generalized beam forces
Z
vector (see (33)). Using (32) the beam forces vector can be ex-
DGint ð/; d/Þ  D/ ¼ ðdeT DDe þ DdeT SÞdx; ð63Þ
pressed a slightly different form as: ‘
Z
SðxÞ ¼ TT N s ds: ð56Þ where ‘ is the length of the undeformed beam, de is the variation of
s the generalized strain vector and Dde is the linearized variation of
the generalized strain vector (or equivalently: the incremental vir-
The explicit expression of the beam forces can be found in Appendix
tual generalized strain vector). The integral of the first term gives
A.
raise to the material stiffness matrix and from the integral of the
Finally, it is possible write the 1D virtual work in terms of the
second term evolves the geometric stiffness matrix.
generalized strains and the generalized beam forces as:
Z Recalling (50) the first term takes the form:
Z
Gint ð/; d/Þ ¼ deT Sdx: ð57Þ
‘ D1 Gint ð/; d/Þ  D/ ¼ ð!d/ÞT H T DHð!/Þdx: ð64Þ
‘

To proceed with the formulation of the geometric stiffness terms we

4.3. External virtual work
need first to linearize the virtual generalized strain measures in
(49). This linearization gives:
In this section we derive the expression of the external virtual
work. In order to simplify this derivation we neglect the body 2 3
du0  Du0
forces. For this particular case, the external virtual work can be
6 du0  De03 þ de03  Du0 þ x00  Dde03 7
written as: 6 7
6 7
Z 6 du 0
 D e0
þ de 0
 D u 0
þ x0
 D de 0 7
6 2 2 0 2 7
Gext ð/; d/Þ ¼ ðp  du0 þ m  dwÞdx; ð58Þ 6 du 0
 D e þ de  D u 0
þ x0
 D de 7
6 2 2 0 2 7
l 6 7
Dd e ¼ 6
6 du0  De3 þ de3  Du0 þ x00  Dde3 7:
7 ð65Þ
where p is the external forces vector and m the external moments 6 0 0 0 0 7
vector. As first noted by Ziegler [41] moments about fixed axes 6 de2  De3 þ de3  De2 þ e3  Dde2 þ e2  Dde3 7
6  0  7
are non-conservative. Consequently, the work is path-dependent 6 0
2 e2  Dde2 þ de2  De2 0 0 7
6   7
if concentrated moments are applied (see [42] for a clarification
6 0 0 0 0 7
4 2 e3  Dde3 þ de3  De3 5
of concepts). 0 0 0 0 0 0
de2  De3 þ de3  De2 þ e3  Dde2 þ e2  Dde3 0 0

4.4. Weak form of equilibrium Then, the general expression of the geometrical tangent stiffness
operator gives:
The variational equilibrium statement can now be presented in Z
terms of generalized components of 1D forces and strains. Recall- D2 Gint ð/; d/Þ  D/ ¼ DdeT Sdx: ð66Þ
‘
ing (57) and (58) the virtual work of a composite beam is written
in its one dimensional form as: Replacing (65) into (66), recalling the expressions of the incremen-
Z Z tal virtual directors Ddei (see Section A.2) and re arranging some
Gð/; d/Þ ¼ deT Sdx  ðp  du0 þ m  dwÞdx: ð59Þ terms, we can write the geometric stiffness operator as:
‘ l
 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351 2345

Z
6.1. Interpolation
D2 Gint ð/; d/Þ  D/ ¼ ð!d/ÞT Gð!D/Þdx; ð67Þ
‘
Being Nn linear Lagrangian shape functions, we interpolate the
where the matrix G is given in Section A.3. It is interesting to note position vectors in the undeformed and deformed configuration
that the terms in (65) are quite complex since one has to find the as [38]:
second variation of the directors and its derivatives. A simpler
way to derive the tangent stiffness matrix can be devised by using X
nxel X
nxel
b ¼
X NnXn; ^¼
x N n ðX n þ un Þ; ð72Þ
the chain rule to find DdeT as:
n¼1 n¼1

DdeT ¼ D½Hð!d/ÞT ¼ ð!d/ÞT DH; ð68Þ where nxel represent the number of nodes in the element. The same
finite element interpolation is also applied to the configuration, so:
where we invoked D(Ld/) = 0. Thus, the linearization of the opera-
tor H gives:

2 3
D u0 0 0
6 De 0 De03  x00 þ e03  Du0 De3  x00 þ e3  Du0 7
6 3 7
6 0 7
6 De 2 De02  x00 þ e02  Du0 0
De2  x0 þ e2  Du 0 7
6 7
6 De De2  x00 þ e2  Du0 0 7
6 2 7
6 7
DH ¼ 6
6 De 3 De3  x00 þ e3  Du0 0 7
7 ð69Þ
6 7
6 0 0 De 1 7
6   7
6 0 0 2 De2  e02 þ e2  De02 7
6   7
6 0 0 7
4 0 0 2 De3  e3 þ e3  De3 5
0 0 0 0
0 0 De3  e2 þ e3  De2 þ De2  e3 þ e2  De3

It can be verified that: X

nxel X
nxel
b¼
/ N n /n ; ^0 ¼
/ N 0n /n : ð73Þ
Gð!D/Þ ¼ DH S; ð70Þ n¼1 n¼1

which can be used in (67) to obtain the geometric stiffness. Using the last expressions, the derivatives of the configuration can
It should be noted that, since the external loads are assumed to be easily obtained. Also, the variations are interpolated as:
be conservative and not eccentric with respect to the pole, the lin-
X
nxel X
nxel
earization of the external virtual work is zero. As has been men- ^¼
d/ N n d/n ; ^0 ¼
d/ N 0n d/n : ð74Þ
tioned several times in geometrically exact formulations, a n¼1 n¼1
parametrization using spin variables necessarily implies a non con-
Having presented the interpolation functions we can now proceed to
servative concentrated moment as its work conjugate variable (see
derive the discrete version of the variational equilibrium statement.
[42]). Indeed, it is well known the fact that the use of spin variables
leads to a non-symmetric tangent stiffness matrix.
6.2. Tangent stiffness matrix
Finally, recalling (64) and (67) the linearized virtual work can
be written as:
Z In order to formulate the tangent stiffness matrix, we first recall
DGð/; d/Þ  D/ ¼ ð!d/ÞT ðH T DH þ GÞð!D/Þdx ð71Þ the definition of the differential operator ! (see (50)). Using the
‘ interpolation function presented above, the differential operator
! can be replaced by its numeric counterpart in such a way that:
Remark 5.1. When it is guaranteed the symmetry of all sub terms X
nxel
in G, the symmetry of the geometrical stiffness matrix is also !d/ ffi ^ n;
Bn d/ ð75Þ
guaranteed. In the present formulation, only some of the unsym- n¼1
metrical terms in G cancel out when ‘‘assembling’’ the geometrical
where the matrix operator Bn contains shape functions and its
stiffness matrix, hence it results to be unsymmetrical. This conse-
derivatives and plays a crucial role in the finite element formula-
quence is common for all spin parametrized formulations and is
tion. In the expressions presented hereafter, summation over index
due to the fact that the second variation of a functional P that maps
n will be implicitly defined. So:
a manifold C into R is symmetric only at the critical points of P. 2 3
Indeed, the Hessian of P only makes sense at its critical points. N01 0
6 7
B¼4 0 N1 5 ð76Þ
6. Finite element formulation 0 N01
Introducing (75) into (64) and into (67) we can obtain the expres-
The implementation of the proposed finite element is based on sions for the material and geometric stiffness matrices respectively.
linear interpolation and one point reduced integration. This ap- Following this process, we firstly obtain:
proach relies on the interpolation of iterative spin variables. As a
Z
consequence, the orthonormality of the directors is not fulfilled ^ d/Þ
^  D/
^¼ ^ T H T DHðBD/Þdx:
^
D1 Gint ð/; ðBd/Þ ð77Þ
at the integration point, leading to interpolation induced loss of ‘
objectivity of the strain measures. Although the effects caused by
the loss of objectivity of the finite element formulation disappear Then, the element material stiffness matrix is:
as the mesh is refined. Indeed, we show in the numerical investiga- Z
tions section that, for quasi-static problems, the error introduced kM ¼ BT H T DHBdx: ð78Þ
by the path dependency is very small. ‘
2346 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351

Proceeding in a similar way, the geometric stiffness term can be mulations. This implies that we need to enforce the material to be
written as: isotropic. So, we chose to compare our beam model against the
Z classical Simo and Vu-Quoc [7] beam model in a roll-up test. The
^ d/Þ
D2 Gint ð/; ^  D/
^¼ ^ T GðBD/Þdx:
ðBd/Þ ^ ð79Þ example is set with a thin-walled beam with a square cross section
‘
(b = 0.5, h = 0.5 and t = 0.05) and a length of 50. The material con-
Thus, the element geometric stiffness matrix becomes: stants are: E = 144  109 and m = 0.3. Being the Euler formula:
Z
kG ¼ BT GBdx: ð80Þ Ml
h¼ ð84Þ
‘ EI
Following the common steps of the finite element assembly We can obtain the magnitude of the moment that produces a de-
process, the global tangent stiffness matrix is: formed shape of half a circle or full circle. To obtain these deformed
X
els shapes we must apply moments M1 = 3.80761  107 and M2 =
KT ¼ ðkM þ kG Þ: ð81Þ 7.615221  107. Fig. 4 shows the deformed shapes obtained after
e¼1 application of these moments.
With an abuse in notation, the summation operator was used to Tables 1 and 2 present the numerical results obtained for the
represent the finite element assembly procedure. maximum tip displacements for both load cases (M1 and M2).
As it can be seen from the tables, the performance of the pre-
6.3. Directors update algorithm sented finite element is very good. Both the vertical and horizontal
displacements agree very well with the results obtained using the
As mentioned before, the updating procedure for the directors well validated Simo and Vu-Quoc model [7].
and its derivatives is iterative. This means that the current config- The differences between the Simo and Vu-Quoc model [7] and
uration is updated in each iteration. Being n the iteration counter the proposed model are originated because the inertia moments
we can recall (1) to find the new director as: resulting from a thin-walled beam formulation are slightly lower
than those that feed a Timoshenko model (most of the thin-walled
enþ1
i ¼ Keni : ð82Þ beam formulations underestimate the real inertial of the beam);
where K is now the incremental iterative rotation tensor. Note that this difference makes the present model to be slightly more flexi-
although the formulation is based on the iterative spins, the director ble than the Simo and Vu-Quoc beam model.
triads are ‘‘total rotation quantities’’. Therefore, the reference con-
figuration is never updated and only at the first iteration of the first 7.2. Bending of a cantilever beam – isotropic and anisotropic materials
increment we set ei = Ei:
According to (82), we can find the derivative of the directors as: The present example studies the performance of the proposed
formulation for both isotropic and anisotropic materials. First, to
e0nþ1
i ¼ K0 eni þ Ke0n
i ¼K 0
KT enþ1
i þ Ke0n
i ; ð83Þ further test the performance of the presented finite element
T
where the expression for K0 K can be found in the Appendix A.4. against existing geometrically exact beam elements, we propose
Note that K is only at the integration point and its expression is a full three dimensional problem where we study the behavior of
fed with the interpolated spin. an isotropic curved cantilever beam (see. Fig. 5). The curved beam
has a reference configuration given as a 45° circular segment with
radius R = 100 and laying in the x  y plane, the beam is loaded
7. Numerical investigations
with a vertical load (z direction). The properties of the isotropic
material are: E = 1.0  107 and m = 0.3. The cross section is a box
To close the development of the presented beam formulation
with b = 1, h = 1 and t = 0.1.
we compare the performance of the proposed finite element with
Table 3 shows the results of the bending test for P = 100. We
existing finite elements. We investigate both the isotropic and
have used an Abaqus 3D shell model as the reference model. As
the anisotropic cases, choosing some benchmark tests proposed
it can be seen, the present finite element formulation behaves
in the literature.
slightly better than the Simo and Vu-Quoc element [7] available
It must be noted that since most of the reported benchmark
in FEAP and also to the Abaqus B31 beam element.
tests performed with displacement based geometrically exact
beam theories [6–10,16,23,24,26,39,43,44] were performed using
35
solid cross sections, often it is not possible to find an equivalent M1
thin-walled section with the same mass and inertia properties to M2
30
those of the solid sections. Because of that, we have used the re-
search finite element software FEAP [45] to obtain results using
the Simo and Vu-Quoc [7] and Ibrahimbegovic and Al Mikad [45] 25

finite elements (which are included in the package). In these calcu-

lations we have used cross sections with the same inertia moments 20
of the thin-walled cross section.
z

Because of the lack of a geometrically exact thin-walled beam 15

theory to compare the present formulation with, for composite
material study cases the present formulation was compared exclu- 10
sively against Abaqus 3D shell models.
5
7.1. Pure bending of a cantilever isotropic beam
0
Firstly, to validate the formulation and especially the finite rota- −25 −20 −15 −10 −5 0 5 10 15 20 25
x
tion algorithm we set and study case where we can compare the
present formulation against existing geometrically exact beam for- Fig. 4. Roll up test.
 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351 2347

Table 1 Table 4
Displacements Components for M1. Maximum displacements in a 45° arc bending test (P = 400).

Tip vertical Tip Max vertical Elements Tip y Tip x Max z Elements
displacement horizontal displacement displacement displacement displacement
displacement
Abaqus Shell 12.201 21.546 50.997 –
Simo & Vu-Quoc 31.673 50.448 31.673 10 Abaqus B31 12.401 21.311 51.110 50
(FEAP) Abaqus B32 12.416 21.310 51.111 50
Ibrahimbegovic– 31.673 50.448 31.673 10 Simo & Vu-Quoc 12.008 20.692 50.067 50
Al Mikad (FEAP)
(FEAP) Present 12.205 21.015 50.880 50
Analytic 31.831 50.000 31.831 –
Present 31.694 50.405 31.694 10

To increase the complexity of the stress state in the beam we

modify the applied load to have components Px = 4.0  105, Py =
Table 2 4.0  105, Pz = 8.0  105. Fig. 6 presents the curves that describe
Displacements components for M2. the evolution of the displacements along the load path (LPF: load
Tip vertical Tip Max vertical Elements proportional factor). The displacements shown in Fig. 6 are dis-
displacement horizontal displacement placements of the pole.
displacement It can be seen from Fig. 6 that the correlation of the present for-
Simo & Vu-Quoc 0.013 49.545 16.038 10 mulation against the Abaqus shell model is excellent, note that this
(FEAP) good correlation does not deteriorate when the displacements
Ibrahimbegovic– 0.013 49.545 16.038 10
grow.
Al Mikad
(FEAP)
Analytic 0.000 50.000 15.915 –
7.3. Post buckling of curved arc – limit point traversal
Present 0.016 49.494 16.004 10

The problem of traversal of limit points has been used several

times as a benchmark test for geometrical nonlinear theories of
With the present algorithm the final solution is arrived in 5 load isotropic beams, see e.g. [11]. We analyze the behavior a curved
steps using an average of 7 iterations per step. thin-walled arc, as shown in Fig. 7, for both isotropic and aniso-
Increasing the load to P = 400 we obtain also a very good corre- tropic materials. The arc cross section is boxed with b = 1, h = 1
lation with the reference model (see Table 4). We add to the com- and t = 0.1.
parison the Abaqus parabolic beam element B32. Although the arc buckles, the buckling phenomenon studied in
Taking as a reference the Abaqus shell model, the maximum this example is not characterized by a bifurcation in the nonlinear
percentile error of the present model is about 0.25%, slightly better phase-plane but by a limit point traversal. A vertical load creates a
than the other beam models. compressive stress state that causes the buckling of the arc. The
Now, we turn to the anisotropic case. As explained before, for load displacement path was followed using the arc-length method.
anisotropic materials we compare the present beam model only In Fig. 8 the load–displacement curve of the isotropic arc using
against an Abaqus shell model. In order to study the behavior of the present formulation is compared vs. the Simo and Vu-Quoc ele-
the proposed finite element when using anisotropic laminates, ment implemented in FEAP and an Abaqus shell model. The isotro-
we analyze the 45° arc of Fig. 5 laminated with a {45, 45, pic material properties are: E = 144  109 and m = 0.3. It can be
45, 45} configuration. The laminas are made of E-Glass fibers
and an Epoxy matrix [35], the material properties are given in
Table 5. Table 5
Material properties of E-Glass Fiber-Epoxy lamina.

E11 E22 G12 G23 m12

45.0  109 12.0  109 5.5  109 5.5  109 0.3

1
Present
0.9 Abaqus
0.8

0.7
Fig. 5. 45° arc bending.
0.6
LPF

0.5
u3 u1 u2
Table 3 0.4
Maximum displacements in a 45° arc bending test (P = 100).
0.3
Tip y Tip x Max z Elements
0.2
displacement displacement displacement
0.1
Abaqus Shell 2.090 3.641 22.611 –
Abaqus B31 2.574 3.570 22.734 50 0
Simo & Vu-Quoc 1.986 3.325 22.001 50 −50 −40 −30 −20 −10 0 10 20 30 40
(FEAP) Displacements
Present 2.068 3.495 22.366 50
Fig. 6. Displacements vs. load proportional factor.
2348 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351

150
P

100

50

z
0

Fig. 7. 215° Thin-walled arc.

−50
−150 −100 −50 0 50 100 150
x

Fig. 9. Deformed shape of the isotropic arc.

observed that the present formulation gives very good results even
for extremely large displacements. It can also be observed that the
present formulation is slightly more flexible than the Simo and Vu- 4
Quoc finite element [7]. Also, and as expected, the 3D shell model x 10
18
is more flexible that both beam formulations. This fact is ascribed Present
16
mainly to the deformability of the cross section. Abaqus

Fig. 9 shows the deformed shape of the structure after the crit- 14
ical point has been traversed. 12
Fig. 10 shows the load–displacement relation during the
10
collapse of a composite arc. We compare the present model
Load

with an Abaqus 3D shell model. In this example we have used 8

a cross section laminated with 4 layers of E-Glass fibers and 6
an Epoxy matrix, oriented in directions {0, 90, 90, 0}. As expected,
4
the limit point load is slightly overestimated by the beam
formulation. This is actually a good result, since the hypothe- 2
ses made by the beam formulation, i.e. the cross sections 0
remaining plane and undeformable, are expected to stiff the
−2
structure. 0 20 40 60 80 100 120 140
Vertical Displacement

7.4. Lateral buckling of a curved composite arc – bifurcation Fig. 10. Load–displacement curve for the {0, 90, 90, 0} E-FiberGlass-Epoxi beam.

To continue with the study of the proposed theory when

encountering critical points, we analyze the behavior of the same
Fig. 11 shows the excellent correlation of the present theory
arc of the last example in a bifurcation scenario. Since the arc
against an Abaqus 3D shell model. As is can be seen, not only the
has a small flexural rigidity the bifurcation point is found before
bifurcation load is coincident with the shell model but also the full
the limit point. Therefore, it’s necessary to use the correct bracket-
nonlinear post-buckling path is essentially the same.
ing procedure in order to detect the bifurcation point, as well as a
branch switching technique in order to follow the stable path of
equilibrium.

6 4
x 10 x 10
9 14
[7] FEAP Present
8 Abaqus Abaqus
12
Present
7

6 10

5
8
Load
Load

4
6
3

2 4
1
2
0

−1 0
0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 80 90 100
Vertical Displacement Vertical Displacement

Fig. 8. Load–displacement curve for the isotropic arc. Fig. 11. Lateral buckling of a {0, 90, 90, 0} laminated arc.
 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351 2349

Table 6 seen that in any case the maximum percentile error is greater than
Loading scheme. 0.02%.
Step Px Py Pz The path dependency of the presented finite element has shown
1 0 0 400 to be independent of the time stepping scheme (and consequently
2 0 200 0 of the iterative path). Also, we can conclude that path dependency
3 100 0 0 has a minor effect in the accuracy of the solution of the presented
4 0 0 400 theory.
5 100 0 0
6 0 200 0
8. Conclusions

Table 7 A static geometrically exact nonlinear beam finite element for

Path dependency test results. composite closed section thin-walled beams has been presented.
The proposed formulation relies on the parametrization of the
Final displacements and error
weak form of equilibrium in terms of the director field and its
Inc. Elements u % of v % of w % of
derivatives. Spatial spins were used to establish the relationship
max. max. max.
between the director field and the configuration. An iterative up-
5 50 0.00128 0.005 0.00292 0.020 0.00247 0.005
date procedure for the director triad was proposed, having the
25 0.00512 0.019 0.01166 0.081 0.00988 0.019
advantage to be very fast since only one triad per element is re-
10 50 0.00128 0.005 0.00292 0.020 0.00247 0.005 quired. If desired, the presented formulation can be easily modified
25 0.00512 0.019 0.01166 0.081 0.00985 0.019 to embed any existing finite rotation algorithm into the thin-
walled beam theory.
20 50 0.00128 0.005 0.00292 0.020 0.00246 0.005
The evaluation of the weak form of equilibrium was aided by
25 0.00512 0.019 0.01166 0.081 0.00985 0.019
the introduction of generalized strains, resulting from a dual trans-
formation of the rectangular GL strains. The generalized strains
work conjugate variables, i.e. the generalized beam forces, were
7.5. A practical path dependency test deduced from the curvilinear shell stresses before the obtention
of the weak form.
It is known the fact that formulations with iterative update of Representative numerical experiments show that the presented
spins are non-invariant and path-dependent towards the path of thin-walled beam formulation has an excellent correlation against
convergence [21]. However, we show in this example that when existing finite elements for isotropic materials. For the case of com-
using the present formulation with proper mesh refinement these posite materials, the correlation against superior theories, such as
effects are not noticed (see also [20]). 3D shell models, was also excellent.
The path dependency of the present formulation is ascribed to The possibility of using any type of composite material in a geo-
the interpolation of rotation variables that are not additive even metrically exact thin-walled framework represents the main
in the limit. With the present formulation the orthonormality of advantage of the present work. Since the cross section properties
director vectors is guaranteed. Although, the interpolation of spin can be obtained without performing a 2D finite element mesh of
variables leads to a loss of objectivity of the discrete strain the cross section the proposed approach is attractive to be used
measures. in optimization problems of composite beams with finite deforma-
Considering that the proposed formulation relies on the param- tion such as helicopter rotor blades and wind turbine blades. The
etrization of the equations of motion in terms of the director field accounting of all existing geometrical and material couplings
and its derivatives, a natural approach to obtain a path indepen- regardless of the magnitude of displacements and rotations is an
dent formulation would be to interpolate directly the director field important achievement of the present formulation. Thus, the finite
and not the rotation variables (see [23,25]). Defining one triad per element here presented could help in the study of the effects of
node and obtaining the derivatives of the director vector via inter- geometrical nonlinearities and anisotropy in the stability behavior
polation of triads we would not need to interpolate the spins of composite closed section thin-walled beams.
(interpolation of triads is allowed since they belong to a vector
space). However, the implementation of this type of interpolation
requires the use of an additional triad (one triad per node) and con- Acknowledgements
sequently requires an additional iterative update, thus leading to a
slower algorithm. The authors wish to acknowledge the supports from Secretarı́a
Using the same curved beam of the previous example we apply de Ciencia y Tecnologı´a of Universidad Tecnológica Nacional and
six load cases and analyze the resulting displacements at the end- CONICET.
ing of the load cycle. The loading scheme is shown in Table 6.
Table 7 presents the results of this practical path-dependency Appendix A
test. The table shows the displacements of the tip of the beam at
the end of the load cycle for three different time stepping schemes b
A.1. Matrix D
(5, 10 and 20 increments). The maximum error is defined as the
quotient of the remaining displacements and the maximum dis- Introducing (20) into the matrix Db we can obtain a cross sec-
placements (obtained at any moment of the load cycle). It can be tional matrix as a function of the midsurface coordinates, this is:

2 3
1 nn02 þ n3 nn03 þ n2 0 0 0 a1 a2 a3
   0 
b ¼6
D 40 0 0 n0 n0 n n02 02    0 0 0
7
0 5;
2 3 2 þ n3 þ n2 n3  n3 n2 ðA1Þ
0 0 0 n03 n02 n2 n02 þ n3 n03 0 0 0
2350 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351

 
b  Dde0i ¼ b  Dðdw0  ei Þ þ b  D dw  e0i
  
¼ b  ðdw0  ðDw  ei ÞÞ þ b  dw  Dw0  ei þ Dw  e0i
¼ b  ðdw0  ðDw  ei ÞÞ þ ðb  dwÞ  ðDw0  ei Þ
 
þ ðb  dwÞ  Dw  e0i
h i
~~ei Dw þ dw½b
¼ dw0 ½b ~ ~ei Dw0 þ dw b
~ ~e0 Dw ðA4Þ
i

where  denotes the skew symmetric matrix of vector as defined in

(45). Proceeding in a similar way we can obtain all the second
variations.

A.3. Matrix G

Fig. A.1. Graphical representation of shell and beam forces. The full expression of the geometric matrix G gives:

2   3
N  M2 e03 þ M3 e02  ðQ 2 e2 þ Q 3 e3 Þ ðM 2 e3 þ M 3 e2 Þ
6 7
G ¼ 4 M 2 e03 þ M 3 e02 þ Q 2 e2 þ Q 3 e3 M2 x00 e03 þ M 3 x00 e02 þ Q 2 x00 e2 þ Q 3 x00 e3 M 2 x00 e3 þ M 3 x00 e2 5 ðA5Þ
M 2 e3 þ M 3 e2 M 2 x00 e3 þ M3 x00 e2 þ G23 G33

where the coefficients ai are: where, for simplicity, we have eliminated the tilde () and as-
n2 sumed that all the terms ei are now skew symmetric matrices.
1
a1 ¼ n2 n02 0 
3  nn3 n2 þ
2
The coefficients Gij are:
2 2       
1 n2 ðA2Þ G23 ¼ Te1 þ 2 P2 e02 e2 þ e2 e02 þ P3 e03 e3 þ e3 e03
a2 ¼ n2 n02 0 
2 þ nn2 n3 þ
3
;  0        
2 2 þ P23 e2 e3 þ e03 e2 þ e3 e02 þ e2 e03 ðA6Þ
a3 ¼ n n2 n3 þ n2 n3 þ nðn02 n2  n03 n3 Þ
2 0 0  
G33 ¼ 2P2 e2 ðe2 Þ þ 2P3 e3 ðe3 Þ þ P23 ½e3 ðe2 Þ þ P3 e2 ðe3 Þ
Explicitly, the beam forces vector gives:
23 A.4. Derivative of the exponential map
N
6 M2 7
6 7 An alternative expression for (7) can be written as:
6 M3 7
6 7
6 Q2 7
6 7 2
S¼6 7 K¼Iþ ðH þ H2 Þ; ðA7Þ
6 Q3 7  2
1 þ k#k
6 T 7
6 7
6 P2 7  is a scaled pseudo-vector such that:
where #
6 7
4 P3 5

P23  ¼ tan 1 kDwk t;

# t¼
Dw
ðA8Þ
0 1 2 kDwk
Nxx
B M 0
xx n2 þ N xx n3
 C  Derivation of K with re-
and H is the skew symmetric matrix of #.
B   C
B M 0
xx n3 þ N xx n2 C spect to x gives:
B   C
Z B N xs n 0
 N xn n0 C " #
B 
2

3 C  #
 0 ðH þ H2 Þ
¼ B N n 0
þ N n0 Cds: 2 2#
B   xn xs   
n þ n þ Nxs n n2  n n3 þ Nxn n n2 þ n n3 C
2 3 0 0 0 0
s B M 02 02 0 0 0 0 C K ¼ H þ H H þ HH  ; ðA9Þ
B xs 2 3 3 2 2 3 C  2
1 þ k#k 1 þ k#k 2
B M n0 n2 þ 1 Nxx n2 C
B xx 3 2 2 C
@ M xx n02 n3 þ 1
N 
n 2 A After some manipulations and identities it can be found that:
 20 xx 3 0 
   
Nxx n2 n3 þ M xx n2 n2  n3 n3  
2
ðA3Þ K0 KT ¼ ðH0  H0 H þ HH0 Þ ðA10Þ
 2
1 þ k#k
where N is the axial beam force, M2 and M3 are the beam flexural
Recalling (83) we see that obtaining a close expression for K0 KT we
moments, Q2 and Q3 are beam shear forces, T is the beam torsion
can update the derivatives of the directors without the necessity of
moment and P2, P3 and P23 are high order flexural moments. A
obtaining a close expression for K0 . Therefore, the above expression
graphical representation of the shell and beam forces can be seen
is used in (83).
in Fig. A.1.
References
A.2. Second variation of directors
[1] Reissner E. A one-dimensional finite strain beam theory: the plane problem. J
Appl Math Phys 1972;23:307–51.
The second variation of the director field is obtained in a stan-
[2] Antman SS. Ordinary differential equations of one dimensional elasticity:
dard manner. For the sake of brevity we show only how to obtain foundations of the theories of nonlinearly elastic rods and shells. Arch Ration
the second variation of the derivative of the director field (which is Mech 1976;61.
more complex than the second variation of the director field itself). [3] Reissner E. On finite deformations of space-curved beams. Zeitschrift für
Angewandte Math und Phys (ZAMP) 1981;32:734–44.
Being b any vector, we can obtain the second variation of the direc- [4] Argyris J. An excursion into large rotations. Comput Methods Appl Mech Eng
tor’s derivative as: 1982;32:85–155.
 C. Martín Saravia et al. / Computers and Structures 89 (2011) 2337–2351 2351

[5] Bathe K-J, Bolourchi S. Large displacement analysis of three-dimensional beam [25] Romero I, Armero F. An objective finite element approximation of the
structures. Int J Numer Methods Eng 1979;14:961–86. kinematics of geometrically exact rods and its use in the formulation of an
[6] Simo JC. A finite strain beam formulation. The three-dimensional dynamic energy-momentum conserving scheme in dynamics. Int J Numer Methods Eng
problem. Part I. Comput Methods Appl Mech Eng 1985;49:55–70. 2002;54:1683–716.
[7] Simo JC, Vu-Quoc L. A three-dimensional finite-strain rod model. Part II: [26] Ghosh S, Roy D. A frame-invariant scheme for the geometrically exact beam
Computational aspects. Comput Methods Appl Mech Eng 1986;58:79–116. using rotation vector parametrization. Comput Mech 2009;44:103–18.
[8] Simo JC, Vu-Quoc L. On the dynamics in space of rods undergoing large [27] Sansour C, Wagner W. Multiplicative updating of the rotation tensor in the
motions – a geometrically exact approach. Comput Methods Appl Mech Eng finite element analysis of rods and shells – a path independent approach.
1988;66:125–61. Comput Mech 2003;31:153–62.
[9] Cardona A, Geradin M. A beam finite element non-linear theory with finite [28] Hodges DH, Yu W, Patil MJ. Geometrically-exact, intrinsic theory for dynamics
rotations. Int J Numerical Methods Eng 1988;26:2403–38. of moving composite plates. Int J Solids Struct 2009;46:2036–42.
[10] Simo JC, Vu-Quoc L. A Geometrically-exact rod model incorporating shear and [29] Yu W, Hodges DH, Volovoi VV, Fuchs ED. A generalized Vlasov theory for
torsion-warping deformation. Int J Solids Struct 1991;27:371–93. composite beams. Thin-Walled Struct 2005;43:1493–511.
[11] Ibrahimbegovic A. On finite element implementation of geometrically [30] Cesnik CES, Hodges DH. VABS: a new concept for composite rotor blade cross-
nonlinear Reissner’s beam theory: three-dimensional curved beam elements. sectional modeling. J Amer Helicopter Soc 1997;42:27–38.
Comput Methods Appl Mech Eng 1995;122:11–26. [31] Gonçalves R, Ritto-Corrêa M, Camotim D. A large displacement and finite
[12] Ibrahimbegovic A. On the choice of finite rotation parameters. Comput rotation thin-walled beam formulation including cross-section deformation.
Methods Appl Mech Eng 1997;149:49–71. Comput Methods Appl Mech Eng 2010;199:1627–43.
[13] Gruttmann F, Sauer R, Wagner W. A geometrical nonlinear eccentric 3D-beam [32] Librescu L. Thin-walled composite beams. Dordrecht: Springer; 2006.
element with arbitrary cross-sections. Comput Methods Appl Mech Eng [33] Piovan MT, Cortı́nez VH. Mechanics of thin-walled curved beams made of
1998;160:383–400. composite materials, allowing for shear deformability. Thin-Walled Struct
[14] Gruttmann F, Sauer R, Wagner W. Theory and numerics of three-dimensional 2007;45:759–89.
beams with elastoplastic material behaviour. Int J Numer Methods Eng [34] Machado SP, Cortı́nez VH. Non-linear model for stability of thin-walled
2000;48:1675–702. composite beams with shear deformation. Thin-Walled Struct
[15] Auricchio F, Carotenuto P, Reali A. On the geometrically exact beam model: a 2005;43:1615–45.
consistent, effective and simple derivation from three-dimensional finite- [35] Barbero E. Introduction to composite material design. London: Taylor and
elasticity. Int J Solids Struct 2008;45:4766–81. Francis; 2008.
[16] Crisfield MA. Non-linear finite element analysis of solids and structures: [36] Jones RM. Mechanics of composite materials. London: Taylor & Francis; 1999.
advanced topics. John Wiley & Sons Inc.; 1997. [37] Washizu K. Variational methods in elasticity and plasticity. Oxford: Pergamon
[17] Antman SS. Nonlinear problems of elasticity. Springer; 1995. Press; 1968.
[18] Felippa C. Lecture Notes in Nonlinear Finite Element Methods, Report No. CU- [38] Zienkiewicz OC. The finite element method. Oxford: Buttherworth-
CSSC-99-xx, in: C.o.A. Structures (Ed.), University of Colorado, 1999. Heinemann; 2000.
[19] Mäkinen J, Marjamäki, H. Total and updated Lagrangian geometrically exact [39] Betsch P. On the parametrization of finite rotations in computational
beam elements, in: III European Conference on Computational Mechanics; mechanics a classification of concepts with application to smooth shells.
2006, pp. 658–658. Comput Methods Appl Mech Eng 1998;155:273–305.
[20] Crisfield M, Jelenic G. Objectivity of strain measures in the geometrically exact [40] Ritto-Corrêa M, Camotim D. On the differentiation of the Rodrigues formula
three-dimensional beam theory and its finite-element implementation. Proc and its significance for the vector-like parameterization of Reissner–Simo
Royal Soc London Ser A: Math Phys Eng Sci 1999;455:1125–47. beam theory. Int J Numer Methods Eng 2002;55:1005–32.
[21] Jelenic G, Crisfield MA. Geometrically exact 3D beam theory: implementation [41] Ziegler H. Principles of structural stability. Waltham, Massachusetts: Blaisdel
of a strain-invariant finite element for statics and dynamics. Comput Methods Publishing Co.; 1968.
Appl Mech Eng 1999;171:141–71. [42] Ritto-Corrêa M, Camotim D. Work-conjugacy between rotation-dependent
[22] Ibrahimbegovic A, Taylor R. On the role of frame-invariance in structural moments and finite rotations. Int J Solids Struct 2003;40:2851–73.
mechanics models at finite rotations. Comput Methods Appl Mech Eng [43] Crisfield MA, Jeleni G. Objectivity of strain measures in the geometrically exact
2002;191:5159–76. three-dimensional beam theory and its finite-element implementation. Proc
[23] Betsch P, Steinmann P. Frame-indifferent beam finite elements based upon the Royal Soc London Ser A: Math Phys Eng Sci 1999;455:1125–47.
geometrically exact beam theory. Int J Numer Methods Eng 2002;54:1775–88. [44] Taylor R. FEAP Users Manual. In: FEAP Berkeley; 2009.
[24] Armero F, Romero I. On the objective and conserving integration of [45] Ibrahimbegovic A, Al Mikdad M. Finite rotations in dynamics of beams and
geometrically exact rod models. In: Proceedings Trends in computational implicit time-stepping schemes. Int J Numer Methods Eng 1998;41:781–814.
structural mechanics, CIMNE, Barcelona, Spain; 2001.