Computers and Structures 165 (2016) 96–106

Contents lists available at ScienceDirect

Computers and Structures

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

A quadrature element formulation of an energy–momentum conserving

algorithm for dynamic analysis of geometrically exact beams
Run Zhang, Hongzhi Zhong ⇑
Department of Civil Engineering, Tsinghua University, Beijing 100084, PR China

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

Article history: A weak form quadrature element formulation of geometrically exact beams is established and employed
Received 18 May 2015 for dynamic analysis along with the energy and momentum conserving algorithm. In addition to its high
Accepted 16 December 2015 computational efficiency and circumvention of shear locking problems, the formulation also ideally pre-
Available online 7 January 2016
serves strain objectivity and conservation of energy, which are violated in some finite element formula-
tions for rotation interpolations. Several benchmark examples are presented to demonstrate the
Keywords: effectiveness of the formulation.
Geometrically exact beams
Ó 2015 Elsevier Ltd. All rights reserved.
Dynamics analysis
Energy–momentum conserving algorithms
Weak form quadrature element method

1. Introduction interpolation for rotation quaternion to formulate an objective

finite element approximation of geometrically exact beams [10].
Nonlinear dynamic analysis of slender structures has found Many of these procedures usually complicate the implementation
wide applications in many engineering areas. Research in this area or lower the efficiency of the formulation.
mainly focuses on two aspects: geometrically exact formulations of Classical time integration algorithms such as the Newmark
structures and stable time integration algorithms with high accu- method, which can be unconditionally stable in linear cases, exhibit
racy. A geometrically exact formulation of structures like beams instability and divergence for analysis of nonlinear systems. In
or shells does not introduce any approximations when describing addition to mathematical remedies, some researchers also empha-
large displacements and rotations, being recognized as an efficient size the algorithms’ satisfaction of the conservation laws of physical
means for many geometrically nonlinear problems. The initial systems to improve their stability. One of the earliest energy and
finite element formulation of geometrically exact space beams momentum conserving algorithms for nonlinear dynamic analysis
was proposed by Simo and his co-workers [1,2] and then extended was proposed by Simo and Tarnow [11] and later extended to
by many other researchers [3–5]. As indicated by Jelenić and Cris- dynamics of geometrically shells [12] and beams [13]. This algo-
field [6], however, the geometrically exact beam formulation of rithm is second-order accurate and has exerted great influence on
Simo spoiled objectivity of strain measures. Because of the non- the development of new algorithms with conservation of energy
linearity of rotation space, additional efforts have to be made to and momentum since its inception. Kuhl and Crisfield combined
preserve objectivity when interpolating rotations from element the energy–momentum algorithm and the generalized-a method
nodes to integration points in finite element formulations. Jelenić [14] to develop a formulation allowing controllable numerical dis-
and Crisfield [6] suggested a remedy to preserve objectivity sipation of unwanted high frequency response [15]. Laursen and
wherein incremental local rotations were used. Betsch and Stein- Meng corrected the error of uncoupling configuration variables
mann derived a formulation that retains objectivity for geometri- and strain interpolation parameters in Simo’s energy–momentum
cally exact beams [7]. They introduced an isoparametric algorithm [11] for general constitutive models [16]. Ibrahimbegović
interpolation of director vectors, which actually refers to the for- and Mamouri introduced dissipative terms into the energy–mo-
mulation of geometrically exact shells to handle space rotations mentum algorithm for controllable energy decay and applied it to
[8]. Zupan and Saje presented a geometrically exact beam model dynamic analysis of geometrically exact beams [17]. Romero and
based on interpolation of strain measures to ensure strain Armero revealed that Simo’s finite element implementation of
objectivity [9], while Ghosh and Roy adopted a spherical linear energy–momentum algorithm for beams [13] is not exactly
energy-conserving due to the spatial finite element interpolations.
⇑ Corresponding author. Tel.: +86 10 62781891. They adopted the finite element interpolation of the director fields
E-mail address: [email protected] (H. Zhong). to define the beam’s motion and proposed an objective finite

http://dx.doi.org/10.1016/j.compstruc.2015.12.007
0045-7949/Ó 2015 Elsevier Ltd. All rights reserved.
 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106 97

element formulation that exactly conserves energy for geometri-

cally exact beams in dynamic analysis [18]. Lens and Cardona
e3
developed a practical nonlinear beam element model that is simple e1
to be combined with their previous energy-preserving time integra- e01
e2
tion [19], making it feasible for dynamic analysis of constrained
Λ e03
multibody systems [20]. Gams et al. proposed a time integration
scheme for planar geometrically exact beams that conserves energy
by incremental strain update [21]. In dynamic analysis of geomet-
rically exact beams, most dynamic algorithms have been applied r r0 Λ 0 e02
to finite element formulations for spatial discretization. As indi- E1
cated in [18], however, improper finite element discretizations in
space may undermine conserving properties of time integration
schemes apart from the complex implementation and low E2
efficiency. E3 1 2
(ξ , ξ , s )
Differing from the conventional finite element method, the
weak form quadrature element method (abbreviated as the
quadrature element method or QEM) incorporates efficient numer- Fig. 1. Initial and deformed configurations of beam.
ical integration schemes and the differential quadrature analog for
derivative approximations to evaluate integrals in the variational
description of a problem. The QEM starts with numerical integra- length coordinate of the beam, denoted by s, the cross-section
tion for integrals. Thereafter derivatives at the integration points strain measures fc; jg are given as
are approximated using the differential quadrature analog instead
of taking derivatives of shape functions, thus making it feasible for c ¼ KT r0  KT0 r00 ;
ð1Þ
high-order approximation and advantageous in reducing the num- j^ ¼ KT K0  KT0 K00 ;
ber of degrees of freedom. The differential quadrature analog is
established based on a complete polynomial series, which ensures where the prime stands for the first order derivative with respect to
convergence. The coincidence of integration points and nodes s, and j
^ is the skew-symmetric matrix associated with vector j. An
within a quadrature element endows the QEM with not only com- infinitesimal rotation vector du can be defined in the skew-
pactness and conciseness, but also the superiority in geometrically symmetric tensor form by taking variation of the rotation tensor
exact formulations for avoiding some interpolation problems. The K as
QEM has been successfully utilized in many areas of structural
u ¼ dKKT :
dc ð2Þ
analysis [22–26] including geometrically exact planar and spatial
beams [27–30]. Then the variation of strains can be obtained as
On the basis of Simo’s energy–momentum conserving algo-
dc ¼ KT ðdr0 þ r0  duÞ;
rithm, a quadrature element formulation for dynamic analysis of ð3Þ
geometrically exact beams is proposed in the present work. Unlike dj ¼ KT du0 :
some finite element formulations, the spatial discretization of this
The configuration of the beam f at a certain time t is given by
formulation exactly inherits the energy conserving property of the n o
algorithm. Besides its superior efficiency, no additional effort to f ¼ x 2 R3 : xðn1 ; n2 ; s; tÞ ¼ rðs; tÞ þ nðs; tÞ; s 2 ½0; L; t 2 ½0; T 
avoid shear locking problems or preserve strain objectivity is
needed in this formulation. ð4Þ
The rest of paper is organized as follows. In Section 2, a brief with
review of the geometrically exact beam theory and the correspond-
ing weak formulation are given. The energy–momentum conserv- nðs; tÞ ¼ na ea ðs; tÞ; ð5Þ
ing algorithms and the quadrature element formulation are a
where n denote the local position of a certain point at the cross-
introduced in detail in Section 3. Four numerical examples illus- section. The Greek indices range from 1 to 2; L is the length of the
trating the effectiveness of the formulation are provided in Sec- beam.
tion 4, and conclusions are drawn in Section 5. In Appendix A, The linear velocity v and spatial angular velocity w are given as
the expression of the element tangent stiffness matrix in the for- follows
mulation is presented. The preservation of strain objectivity of
the formulation is proved in Appendix B. v ¼ r_ ;
ð6Þ
_ T;
^ ¼ KK
w

2. Geometrically exact beam theory, governing equations and where a superposed dot stands for the first order derivative with
weak formulation respect to time t. For establishment of the governing equations, a
notation for loads applied on the beam is introduced. The dis-
For the convenience of the configuration description of a geo- tributed force and couple are denoted as n and m;
 the concentrated
metrically exact beam, a Cartesian frame fE1 ; E2 ; E3 g is defined force and couple are represented by n ~ and m;
~ the prescribed dis-
as the reference coordinate system, as shown in Fig. 1. A certain ~ while the resultant
placement constraints are designated as ~r and K,
cross-section of the beam is described by the centroidal positions force and couple are denoted by n and m. The translational inertia
at the current and the initial configurations, denoted as r and r0 , Aq and rotational inertia tensor iq are defined as
respectively; and the rotation tensors transforming the reference Z
system to the coherent frame fe1 ; e2 ; e3 g and to the initial frame Aq ¼ qdA;
fe01 ; e02 ; e03 g are denoted as K and K0 , respectively. Throughout ZA   ð7Þ
the paper, a vector term with a superposed hat denotes the iq ¼ q knk2 I  n  n dA;
skew-symmetric matrix associated with the vector. With the arc A
98 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106

where A represents the cross-sectional area and q is the density of while the kinetic energy is constructed as
the beam. The local governing equations are established as Z L Z Z L  
1 1
K kin ¼ q0 x_  xdAds
_ ¼ Aq kv k2 þ wT iq w ds: ð16Þ

p_ ¼ n0 þ n 2 2
ð8Þ 0 A 0
_p ¼ m0 þ m  þ r0  n
For conservative external loads that correspond to a potential
with boundary conditions functional V ext , the total energy of the system can be written as

~ on @ U f;
r ¼ ~r; K ¼ K H ¼ W int  V ext þ K kin ð17Þ
ð9Þ
n¼n ~; m ¼ m~ on @ R f; with the strain energy W int having a quadratic form as
Z L 
where 1 T 1
W int ¼ c DN c þ jT DM j ds: ð18Þ
2 2
p ¼ Aq v ; 0
ð10Þ
p ¼ iq w It can be verified that the total energy is conserved, i.e.

and @ U f and @ R f are displacement-prescribed and force-prescribed dH

¼ 0: ð19Þ
boundaries, respectively. dt
Before discussion of the linear elastic constitutive relationship If the total force and the total torque are equal to zero, i.e.
of the beam, a number of symbols are introduced first. E and G rep- Z L
resent Young’s modulus and shear modulus, respectively; A1 and Fext ¼  ds þ n
n ~ j@ f ¼ 0;
R
A2 are the shear areas corresponding to axes e1 and e2 , while I1 0
ð20Þ
Z L
and I2 are the cross-sectional inertial moments about axes e1 and
Text ¼  þm
½r  n  ds þ ½r  n
~ þm
~ j@ f ¼ 0
e2 , respectively. The torsional inertia moment of the cross- 0
R

section is denoted by J. Define two diagonal matrices

and no displacement constraints are prescribed, conservation of the
DN ¼ diag ½ GA1 GA2 EA ; total linear momentum and total angular momentum holds true
ð11Þ
DM ¼ diag ½ EI1 EI2 GJ : dL dJ
¼ 0; ¼ 0: ð21Þ
The stress resultants of the beam are given by dt dt
  " #       The energy–momentum conserving algorithm proposed by
N KT 033 n DN 033 c c
¼ ¼ ¼D : Simo et al. is a one-step scheme for temporal integration. The time
M 033 K T
m 033 DM j j interval ½0; T  is discretized into a certain number of steps. Eq. (13)
ð12Þ over time interval ½tm ; t mþ1  is approximated as
Z L
For the implementation of the quadrature element method, the 1
pmþ1  pm  dr þ ðpmþ1  pm Þ  du ds
governing equations in Eq. (8) are written in weak form as Dt 0
Z Z L h   i
L
 Þ  dr þ ðp_  m0  m
½ðp_  n0  n   r0  nÞ  duds þ nðmÞ  dr0 þ r0ðmþ1=2Þ  du þ mðmÞ  du0 ds
0
0
Z L Z L  dW extðmþ1=2Þ ¼ 0 ð22Þ
¼ ðp_  dr þ p_  duÞds þ ½N  dc þ M  djds
0
Z
0 with f m ¼ f ðt m Þ; f mþ1 ¼ f ðt mþ1 Þ; f ðmþ1=2Þ ¼ f m þ f mþ1 =2 and
L
 ðn   duÞds  n
  dr þ m ~  duj0
~  drj0  m L L f ðmþ1=2Þ ¼ f ððt mþ1 þ tm Þ=2Þ for any variable f and Dt ¼ tmþ1  t m . The
0 updated schemes for r and K adopt the following approximations:
¼ dK kin þ dW int  dW ext ¼ 0 ð13Þ
1
rmþ1  rm ¼ u ¼ Dtðv mþ1 þ v m Þ;
with 2 ð23Þ
Z 1 ^ðmÞ
L Kmþ1  Km ¼ # ðKmþ1 þ Km Þ;
dK kin ¼ ðp_  dr þ p_  duÞds; 2
0
Z L where the rotation vector # between t m and t mþ1 , denoted as #ðmÞ ,
dW int ¼ ½N  dc þ M  djds; ð14Þ has the form
0
Z L 1
dW ext ¼   duÞds þ n
  dr þ m
ðn ~  duj0 :
~  drj0 þ m L L #ðmÞ ¼ Dt wmþ1 þ cay #ðmÞ wm ð24Þ
2
0
with Cayley transform

1 1
3. Energy–momentum conserving algorithms and quadrature cay½# ¼ I þ #^ þ #^2 : ð25Þ
element formulation 1 þ 1 k#k2
4
2

With the notation in Section 2, the total linear momentum L and Thus, the updated scheme of linear and angular momenta can
the total angular momentum J of the beam are given in the form be derived using Eqs. (23) and (24) as
 
Z Z Z 2
ðrmþ1  rm Þ  v m ;
L L
_ pmþ1 ¼ Aq
LðtÞ ¼ q0 xdAds ¼ pds; Dt
0 A 0  ð26Þ
Z Z Z ð15Þ 2 ðmÞ
L L
pmþ1 ¼ iq #  cay #ðmÞ wm :
JðtÞ ¼ _
x  q0 xdAds ¼ ðr  p þ pÞds; Dt
0 A 0
 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106 99

From the variation of rotation tensor with


d c
Y
n Y
n
dK ¼  cay½ed#K ¼ d#K ð27Þ HðnÞ ¼ n  nj ; H0 ðni Þ ¼ ni  nj : ð35Þ
de e¼0
j¼1 j¼1;j–i

one arrives at the relationship du ¼ d#.

Define the element’s nodal variable vector as
After a lengthy but straightforward manipulation, the following
ðeÞ
duT1 duTi duTn
T
updated schemes for strains can be obtained dd ¼ drT1    drTi    drTn : ð36Þ
 
cmþ1  cm ¼ KTðmþ1=2Þ u0  #ðmÞ  r0ðmþ1=2Þ The utilization of the numerical integration scheme and the dif-
0
ð28Þ ferential quadrature analog in Eq. (22) for a quadrature element
jmþ1  jm ¼ KT
ðmþ1=2Þ #
ðmÞ
yields the identity
with lX
n
wi pmþ1  pm  dr þ ðpmþ1  pm Þ  du
ðeÞ ðeÞ
 dK kin þ dW int  dW ðeÞ
ext ¼
1 1 ðmÞ Dt i

 I þ 2 #^
i¼1
Kðmþ1=2Þ ¼  Km ¼ U #ðmÞ Km : ð29Þ h  
1  ðmÞ 2 X
n
Iþ # 4 þl wi NðmÞ  KTðmþ1=2Þ dr0 þ r0ðmþ1=2Þ  du
i¼1
The mid-point constitutive relationships for linear elastic model
i X
n 
ðmþ1=2Þ du
þ MðmÞ  KT
1
0
 l wi n ðmþ1=2Þ  dr
N ¼ DN cmþ1 þ cm
ðmÞ i
i¼1
2 ð30Þ 
1  ðmþ1=2Þ  du  n
þm ~ ðmþ1=2Þ1  dr1  n
~ ðmþ1=2Þn  drn
M ¼ DM ðjmþ1 þ jm Þ
ðmÞ
i
2
m~ ðmþ1=2Þ1  du1  m ~ ðmþ1=2Þn  dun
are utilized to conserve energy, and the resultant force and couple  
ðeÞT ðeÞ ðeÞ
in Eq. (22) are defined by ¼ dd Gkin þ Gint  GðeÞ
ext ; ð37Þ

nðmÞ ¼ Kðmþ1=2Þ NðmÞ ; where l ¼ ds=dn denotes the initial configuration parameter of the
ð31Þ element. The variable derivatives dr0 and du0 are approximated via
mðmÞ ¼ Kðmþ1=2Þ MðmÞ :
the differential quadrature analog. This actually implements the
With the established time discretization and updated schemes, high-order polynomial interpolation of dr and du. The element
the conservation laws of linear and angular momenta as well as the kinetic, internal and external force vectors in Eq. (37) are given as
total energy can be verified theoretically. For details of the proof, follows
one can refer to [13]. It should be emphasized that these properties
ðeÞ lX
n
~ ðeÞ ;
hold only for linear elastic constitutive model and quadratic exter- Gkin ¼ wi G ð38Þ
nal loading potential. The mid-point scheme is not suitable for gen- Dt i¼1
i

eral models, and additional efforts should be made to find the ( )

X NðmÞ
n
corresponding interpolation points [16]. ðeÞ
In order to implement numerical integration in the weak form Gint ¼ l wi BTi CTi ; ð39Þ
i¼1 MðmÞ i
quadrature element formulation, the local coordinates within a
beam element need to be transformed into a normalized coordi- ðeÞ ðeÞ
nate system n 2 ½1; 1. In the present paper, Lobatto quadrature GðeÞ
ext ¼ Gext þ Gext ; ð40Þ
is employed as the integration scheme. Assume that there are n where
integration points within an element, the sampling point ni is "  T     #T
ði  1Þ th zero of the first order derivative of the ðn  1Þ th order e ðeÞ ¼ d pmþ1  pm pmþ1  pm T pmþ1  pm T
G  d  d ;
Legendre polynomial. The corresponding weighting coefficients i i1
pmþ1  pm i ij
pmþ1  pm i in
pmþ1  pm i
are [31]
ð41Þ
1
w1 ¼ wn ¼ ;
nðn  1Þ Bi ¼ bi1    bij    bin ; ð42Þ
ð32Þ
2 " #
wj ¼ ; j ¼ 2; . . . ; n  1:
nðn  1Þ½Pn1 ðnj Þ2 KTðmþ1=2Þi 033 KTðmþ1=2Þi r^0 ðmþ1=2Þi
Ci ¼ ; ð43Þ
The differential quadrature analog [32] approximates the 033 KT
ðmþ1=2Þi 033
derivative at a grid point with a weighted linear summation of
ðeÞ T
function values at all the grid points within the element, i.e.  T1 w1 m
Gext ¼ l w1 n  T1    wi n
 Ti wi m
 Ti    wn n
 Tn wn m
 Tn ðmþ1=2Þ

d f 
k X
n
ðkÞ
ð44Þ
 ¼ C ij f ðnj Þ; ð33Þ
dnk  n¼ni j¼1 and

ð1Þ  ðeÞ ¼ ½ n
G ~1 ~1
m ~n
0  0 n ~ n Tðmþ1=2Þ
m ð45Þ
where the weighting coefficients C ij for first-order derivatives, ext

which will be used in this paper, are given by with

8 2 3
> H 0 ðni Þ
> ; ði – jÞ; 2
C I33
ð1Þ
033
< ðni nj ÞH0 ðnj Þ
>
6 L ij 7
ð1Þ
C ij ¼ X n
ð1Þ
ð34Þ bij ¼ 6
4 033 2 ð1Þ
C I
7:
5 ð46Þ
>
>  C ik ; ði ¼ jÞ L ij 33
>
:
k¼1;k–i 033 dij I33
100 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106

dij is the Kronecker delta function. For the arbitrariness of the nodal
ðeÞ
variable vector dd in Eq. (37), the element balance vector is
defined as: EA = GA1 = GA2 = 10000,
EI1 = EI 2 = GJ = 500,
ðeÞ ðeÞ
RðeÞ ¼ Gkin þ Gint  GðeÞ
ext ð47Þ Aρ = 1,
8
with its incremental form i ρ = diag [10, 10, 10]

ðeÞ ðeÞ ðeÞ

DR ¼ KT Dd : ð48Þ E1
M 1 (t ) = 0.5 p ( t ) ,
M 2 (t ) = − p (t ) ,
ðeÞ
KT in Eq. (48) is the element tangent stiffness matrix whose expli- E2 E3
F ( t ) = 0.1 p ( t ) ,
cit expression is given in Appendix A. After element assembly, the
6 ⎧80t , 0 ≤ t < 2.5;
global balance vector and its incremental form are obtained as ⎪
F (t ) p ( t ) = ⎨400 − 80t , 2.5 ≤ t < 5;
R ¼ Gkin þ Gint  Gext ¼ 0; M 2 (t ) ⎪0, t ≥ 5.
⎩
ð49Þ
DR ¼ K T Dd M1 (t )

With the Newton–Raphson iteration scheme, Eq. (49) can be solved Fig. 2. Flying flexible beam: geometric, material and loading data.
step by step. A displacement-convergent criterion is adopted in the
ðiÞ
present paper. Define Dd as the incremental displacement vector
at the ith iteration during a certain step. The solution is considered of no kinematic constraints, the linear momentum, the angular
to have converged if momentum as well as the total energy should keep constant after
    fully unloading, making this example a perfect numerical test of
 ðiÞ   ðiÞ 
Dd  6 sd ; ð50Þ the preserving properties of time-integration algorithms.
The energy and momentum conserving properties are verified
 
 ðiÞ  first. One quadrature element with 9 integration points is used for
where d  is the Euclidian norm of the displacement vector of the
the entire beam and the time step is taken as Dt ¼ 0:2. The time his-
step. The tolerance s is a small quantity set to be 1  106 in this tory of total energy in a long time period t 2 ½0; 500 is presented
paper. and compared with that obtained by Simo et al. [13] in Fig. 3. The
Due to the high-order approximation and optimal choice of energy history obtained using the original data in Fig. 2 shows large
sampling nodes, much fewer degrees of freedom are needed in discrepancies with that in Ref. [13]. Hesse et al. indicated that their
the quadrature element formulation, as has been shown in many results match well with those in Ref. [33] if the rotational inertia
previous papers. As compared to low-order methods, a single tensor is taken as iq ¼ diag ½10; 10; 20 and an increased stiffness
quadrature element can depict complex configurations accurately
EIa ¼ GJ ¼ 103 is used [35]. Considering the good agreement of
for its great geometric feasibility, which is especially suitable for
the results between [13] and [33], the total energy history for the
large displacement and rotation analysis of beam-like structures.
modified data is also presented in Fig. 3, which coincides well with
Another advantage of the high-order approximation is the circum-
that in [13]. Therefore, the corresponding examples in [13,33] may
vention of the locking problems, which result from the incapability
have used larger inertia and stiffer material data mistakenly. It is
of approximating some deformation modes in displacement-based
observed that the present formulation guarantees the invariance
low-order elements. The locking-free property of the QEM in geo-
of the total energy up to four-digit accuracy during the whole time
metrically exact beam analysis has been studied in a previous
period after unloading in all cases. The total amounts of energy cor-
quaternion-based quadrature element formulation [30].
responding to the original and modified data after unloading are
It is seen in the present formulation that integration points
910.1 and 613.5, respectively. It should be noted that Dt ¼ 0:2 is
within an element are also the nodes. This feature can ensure the
not small enough to guarantee these energies to be exact. In Table 1,
property of strain objectivity, which is very important in judging
the total amounts of energy for different numbers of integration
a formulation’s effectiveness in geometrically nonlinear analysis
points and time steps under the original data are listed. For compar-
[6]. In other words, strain measures are uniform in any reference
ison, results obtained with the beam modeling with one element
coordinate system in this formulation. The proof for the objectivity
and two equal elements, i.e. N e =1 and N e =2, are presented. It can
of the formulation is given in Appendix B.
be seen that for space discretization, a quadrature element with
As stated before, the finite element formulation given in Ref.
[13] spoils the energy-conserving property of the algorithm. This
defect is incurred by the direct interpolation of rotations on the
nodes to integration points, leaving an error for the rotational part
of the kinetic energy. A detailed proof of this defect can be found in
Ref. [18]. In contrast, the total energy is exactly conserved due to
the coincidence of the integration points and nodes in the present
formulation.

4. Numerical examples

4.1. Example one-Flying flexible beam

As shown in Fig. 2, an initially straight flexible beam without

kinematic constraints undergoing large overall motion is studied.
This problem was first introduced in Ref. [33] to study the dynam-
ics of rods in space and later analyzed by many researchers
[13,21,34–37]. A triangular impulsive loading with force and tor-
ques at the beam’s end causing spatial motion is adopted. Because Fig. 3. Flying flexible beam: time history of total energy.
 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106 101

10 integration points is sufficient for convergence of the total Table 2

energy for any time step, while 7 integration points per element CPU time and total numbers of iterations for different modeling schemes.

are needed to achieve the same accuracy if two quadrature ele- Modeling Ne ¼ 1 Ne ¼ 2 Ne ¼ 3 Ne ¼ 4 Ne ¼ 5
ments are used. It is noted that because of the exponential conver- scheme n = 10 n=7 n=6 n=5 n=5
gent property of quadrature elements, models using large elements CPU time (s) 9.62 11.07 13.50 13.68 21.01
with more integration points require fewer total degrees of freedom N iter 8305 8278 8289 8283 8319
to achieve the same computation accuracy as compared to those
divided into smaller elements. However, high-order elements
may bring about larger condition numbers and bandwidths in stiff- where rc is the position of the mass center. With one quadrature
ness matrices although this would not generally induce solution element of 9 integration points, the coordinates of the beam mass
problems. The corresponding discussions and remedies have been center in the E1 –E3 plane at different times are computed and listed
given in some previous papers about static analysis of beam struc- in Table 3 together with the corresponding analytical results. For
tures based on the quadrature element formulation [24,26]. Discus- time step Dt ¼ 0:01, the present results coincide almost perfectly
sions on pros and cons of such high-order formulations can be with the analytical results. Even for a relatively large time dis-
found in [38] wherein high-order methods are asserted to have a cretization Dt ¼ 0:1 the numerical results are still in agreement
superiority in reducing total computational cost over low-order with the analytical ones to three effective digits. Coordinates of
methods. For time discretization, the energy converges as long as the mass center along axis E2 , which are not listed in Table 3 for
Dt < 0:02. conciseness, are observed to be less than 1  1012 all the time,
For further comparison of the total computational costs under implying that the beam moves in the E1 –E3 plane.
different element-division schemes, the CPU time consumption
and the total number of iterations over the time period
t 2 ½0; 200 with the time step Dt ¼ 0:1 are listed in Table 2. The 4.2. Example two-Vibration of an L-shaped cantilever
total number of iterations is denoted by N iter . Choosing the number
of equal size elements for modeling the beam as 1 to 5, it is found An L-shaped cantilever subjected to an out-of plane impulse
that the least numbers of integration points per element to obtain loading at the elbow is considered in this example, as shown in
the convergent total energy 889.0 are 10, 7, 6, 5 and 5, respectively. Fig. 6. The impulse varies in a triangular pattern over time interval
It can be seen in Table 2 that the CPU time consumption increases t 2 ½0; 2 with its maximum value F max ¼ 50 at t ¼ 1. After fully
significantly for models utilizing lower-order elements to reach the unloading, the right-angle cantilever undergoes large displacement
convergent result, demonstrating the efficiency of high-order vibrations while the total energy should stay invariant because of
approximation. Nevertheless, the total number of iterations stays no damping. This example has been studied in many papers
virtually unchanged for different modeling schemes. [6,33,34,37,39–41].
Fig. 4 shows the time histories of linear and angular momenta In the present analysis, the cantilever is modeled by two
for time step Dt ¼ 0:2. After unloading, the linear momentum quadrature elements meeting at the elbow, and each element
and angular momentum stay invariant over the time period. Theo- employs 6 integration points. A constant time step Dt ¼ 0:25 is
retical values of the linear momentum after unloading, taken throughout the calculation. The vertical displacements at
L1 ¼ L2 ¼ 0; L3 ¼ 50, can be obtained from the loading condition. the elbow and the free tip are displayed in Fig. 7 for time interval
With Dt ¼ 0:2, the third entry of the linear momentum, L3 , after t 2 ½0; 50. Comparison is made with the results in Ref. [37] for
unloadingis found to be 50.080 (see Fig. 4), exhibiting a slight devi- time interval t 2 ½0; 30 where each straight portion of the limb
ation from the theoretical value. This is ascribed to the loose con- was modeled with 10 elements with quadratic interpolation of
trol of time step. It can be asserted that L3 ¼ 50:000 is achievable if strains. With much fewer numbers of degrees of freedom, the pre-
a smaller time step, say Dt ¼ 0:05 or Dt ¼ 0:1, is adopted. sent results reach excellent agreement with those in Ref. [37].
By employing eleven integration points and time step Dt ¼ 0:1, Fig. 8 shows the time history of the cantilever’s total energy
the configurations in the E1 –E3 plane at the early tumbling stage over time interval t 2 ½0; 50 for time steps Dt ¼ 0:25 and
are depicted in Fig. 5. The present configurations resemble those Dt ¼ 0:1. While the numbers of steps in these two cases are 200
in [34,35] closely. and 500, the corresponding numbers of total iteration are 1039
It can be shown easily that the motion of the beam mass center and 2012, respectively. The average iteration number per step is
can be described analytically, i.e. 5.20 for Dt ¼ 0:25 or 4.02 for Dt ¼ 0:1, illustrating that more itera-
8 tions are required for convergence in large time steps. The results
< 4 0 15 t þ 3 ; 0 6 t < 2:5;
2 3
>
in [37] that were obtained employing the classical Newmark time
rc ðtÞ ¼ 4 0 6  5t þ 2t  15 t ; 2:5 6 t < 5; ;
43 2 2 3 ð51Þ integration scheme with Dt ¼ 0:1 are also shown for comparison.
>
:
½ 4 0 15:5 þ 5t ; tP5 The curve for time step Dt ¼ 0:1 matches well with that in [37],
though the latter shows small oscillation after unloading. In fact,
the classical Newmark scheme may experience instability and
Table 1
Total energy of flying flexible beam for different time steps and numbers of
finally fail for larger time step or long-term time integration in this
integration points. example, presenting an oscillation of total energy, as has been
shown in [6,37]. In contrast, the present energy conserving algo-
Dt ¼ 0:01 Dt ¼ 0:02 Dt ¼ 0:05 Dt ¼ 0:1 Dt ¼ 0:2
rithm circumvents such instability. It is observed that the total
Ne ¼ 1 n=7 881.7 881.8 882.6 886.4 908.5 energy is exactly conserved after fully unloading in the present for-
n=8 883.4 883.5 884.5 888.4 909.5
n=9 884.1 884.1 885.1 889.0 910.1
mulation for both time steps in Fig. 8.
n = 10 884.4 884.4 885.4 889.0 910.2 The time history of axial force at the clamped end of the can-
n = 11 884.4 884.4 885.4 889.0 910.2 tilever is displayed in Fig. 9, with the constant time step Dt ¼ 0:1
Ne ¼ 2 n=4 874.0 873.9 874.7 877.5 896.0 and one element of 13 integrations points per straight portion of
n=5 880.4 880.4 881.5 885.2 906.4 the limb for modeling. Due to the lack of numerical dissipation
n=6 884.1 884.1 885.2 888.9 910.0 mechanism in this energy–momentum conserving algorithm, the
n=7 884.4 884.4 885.4 889.0 910.1
unresolved high frequencies would introduce oscillations, which
n=8 884.4 884.4 885.4 889.0 910.2
can be observed in Fig. 9. As has been introduced in the literature,
102 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106

Fig. 4. Flying flexible beam: time history of angular momentum and linear momentum.

Fig. 5. Flying flexible beam: snapshot of configurations in the E1 –E3 plane at early tumbling stage.

Table 3
Flying flexible beam: coordinates of mass center.

Time Dt ¼ 0:1 Dt ¼ 0:01 Analytical

r c1 r c3 r c1 r c3 r c1 r c3
0 4.00000 3.00000 4.00000 3.00000 4.00000 3.00000
0.5 4.00000 3.01700 4.00000 3.01667 4.00000 3.01667
1 4.00000 3.13400 4.00000 3.13334 4.00000 3.13333
1.5 4.00000 3.45100 4.00000 3.45001 4.00000 3.45000
2 4.00000 4.06800 4.00000 4.06668 4.00000 4.06667
2.5 4.00000 5.08500 4.00000 5.08335 4.00000 5.08333
3 4.00000 6.56799 4.00000 6.56668 4.00000 6.56667
3.5 4.00000 8.45100 4.00000 8.45000 4.00000 8.45000
4 4.00000 10.6340 4.00000 10.6333 4.00000 10.6333
4.5 4.00000 13.0170 4.00000 13.0167 4.00000 13.0167
5 4.00000 15.5000 4.00000 15.5000 4.00000 15.5000

EA = GA1 = GA2 = 106 ,

E1 ⎧50t , 0 ≤ t < 1;
Fig. 7. Vertical displacements at elbow and free tip of an L-shaped cantilever.
GJ s = EI1 = EI 2 = 103 , F = ⎨100 − 50t ,1 < t ≤ 2
⎩
O Aρ = 1,
E2 The variation of axial force along the axis of the upper limb at
i ρ = diag [10, 10, 20].
four chosen times t ¼ 10; 20; 30; 40 is studied. The origin of the
arc length coordinate s ¼ 0 is set at the clamped end. Two kinds
of node mesh, n = 13 and n = 15, are employed and the correspond-
L = 10 ing results are displayed in Fig. 10, respectively. It is seen that axial
E3
forces obtained with 13 nodes exhibit visible oscillations, while the
F
15 node model presents smoother results. This indicates that
Fig. 6. Vibration of L-shaped cantilever: geometric, material and loading data. results of higher accuracy demand high-order approximation for
modeling and capturing deformation modes satisfactorily. It is also
observed that the profile of axial force along the upper limb and
many dissipating algorithms have been proposed to guarantee the position where the maximum axial force takes place vary with
more reliable results. Further study on the extension of the quadra- time, indicating that dynamic response involving large deflection
ture element formulation in this area is still needed. complicates the variation of stress resultants.
 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106 103

4.3. Example three- Swing with flexible beam and concentrated mass

This example was first proposed by Bauchau et al. [42,43], it is

well-known as a stiff problem in multi-body dynamic analysis and
has been studied by many researchers thereafter [17,21,37]. A
swing composed of a flexible beam with a concentrated mass mc
at its midspan and two rigid links at its end is analyzed in this
example. The geometric and material data of the swing are given
in Fig. 11. The rigid links are assumed to be weightless. A concen-
trated force F applied at the midspan in the E3 direction is a trian-
gular impulse. It consists of a linear increase from 0 to 2 in the time
period [0, 0.128] and a linear decrease that terminates at t = 0.256,
leaving the swing oscillating in the E1 –E3 plane.
In the present analysis, the beam is equally divided into two
elements and each link is modeled by one element. Eleven integra-
tion points are used in each element. The Young’s modulus of the
links is set to 100 times of that of the flexible beam to ensure
Fig. 8. L-shaped cantilever: time history of total energy. the rigidity. The time interval [0, 1] for observation is discretized
into 2000 time steps, i.e. Dt ¼ 0:0005. A total iteration number of
8635 is recorded for the accomplishment of the computation.
The initial configuration and the sequence of deformed config-
urations of the swing in the E1–E3 plane at each Dt ¼ 0:1 during
the time interval [0, 1] are depicted in Fig. 12. For further illustra-
tion, the horizontal (in E3 direction) and vertical (in E1 direction)
displacements of point A on the flexible beam are presented in
Fig. 13. It can be observed that both displacement components
increase steadily until tc ¼ 0:64, after which fluctuations resulting
from the influence of high frequency vibrations of the flexible
beam turn up.
Because of no energy dissipation, the total energy of the swing
should stay invariant after the removal of the impulse. In Fig. 14,
the time history of total energy, kinetic energy and strain energy
of the swing, along with the total energy results obtained by
Ibrahimbegović and Mamouri with finite element formulation
[17] are displayed. It is seen that the total energy matches perfectly
with those in Ref. [17] and is exactly conserved after the triangular
impulse. The kinetic part dominates the total energy before tc . Both
the kinetic energy and the strain energy begin oscillating after-
wards, reflecting the high frequency response after tc .
The time history of the axial force at point A is depicted in
Fig. 9. L-shaped cantilever: time history of axial force at the clamped end.
Fig. 15. A sharp increase of the axial force is observed at around

(a) (b)
Fig. 10. L-shaped cantilever: axial force curves along the upper limb at chosen times for (a) n = 13 and (b) n = 15.
104 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106

Fig. 11. Swing with flexible beam and concentrated mass: geometric and material
data.

Fig. 14. Swing: time history of total energy, kinetic energy and strain energy.

Fig. 12. Sequence of configurations of swing at each Dt = 0.1 time interval.

Fig. 15. Swing: time history of axial force at point A on the flexible beam.

EA=GA1=GA2=10000,
E1 GJ s=EI 1=EI 2=1000,
Aρ=1,
iρ =diag[10, 10, 20],

M2
E3
F2
E2

Fig. 13. Swing: displacement components at point A on the flexible beam. Fig. 16. Flexible beam in helicoidal motion: geometric and material properties.

tc , followed by high frequency oscillations. The peak of axial force [37,41,44]. One quadrature element with 10 integration points is
is recorded at t ¼ 0:641 for the present time division scheme, coin- utilized for modeling and a constant time step Dt ¼ 0:5 is taken.
ciding with the result obtain in Ref. [43]. For the time period t 2 ½0; 50, the total iteration number is 500.
The time histories of tip displacements and the mass center dis-
4.4. Example four-Flexible beam in helicoidal motion placement along axis E2 are displayed in Fig. 17.
It is found that the linear momentum along axis E2 is exactly
In this example, a flexible straight beam in a helicoidal motion preserved to be 10 after unloading in the calculation. This preser-
is analyzed. As shown in Fig. 16, the beam of length 10 is initially vation can be verified in Fig. 17 by the unaltered slope of the curve
placed along axis E3 . One end of the beam is so constrained that of mass center displacement along E2 . The time-displacement
translation along axis E2 and rotation about axis E2 are allowed. curve along E2 at the tip shows slight oscillations, exhibiting the
A constant force F 2 ¼ 4 and a constant torque M 2 ¼ 80 are applied mutual conversion between the kinetic energy and the strain
over time interval t 2 ½0; 2:5. This problem has been studied in energy. In the meantime, the beam also experiences a periodic
 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106 105

_
Hi ¼ di1 Hi    dij Hi    din Hi ; ðA:2Þ

^

1
^ ðmÞ  #^ðmÞ m
Hi ¼ mðmÞ  #ðmÞ þ m ^ ðmÞ : ðA:3Þ
2 i

The element tangent stiffness matrix in Eq. (48) can be

expressed as
Xn  Xn
ðeÞ
KT ¼ l
1 e i Bi þ l
wi BTi Ni þ CTi Di C e i;
wi H ðA:4Þ
i¼1
2 Dt i¼1

where
2 3
d ðmÞ
033 033  12 Kðmþ1Þi Ni
6 7
6 ^ 7
Ni ¼ 6 033 033 Hi 7; ðA:5Þ
4 5
d ðmÞ 1 ^0 d ðmÞ
Fig. 17. Tip and mass center displacements of flexible beams in helicoidal motion.
1
2
Kðmþ1=2Þi Ni 033 r K N
2 ðmþ1=2Þi ðmþ1Þi i

" #
~i ¼ KTmþ1 033 KTmþ1 r^0 mþ1
C ðA:6Þ
movement in the E1 –E3 plane, which can also be verified from the 033 KTmþ1 033
tip displacements along axes E1 and E3 . The rotation period around and
axis E2 can be worked out analytically to be 13.614 if the beam is
completely rigid. Because of the variation of rotational inertia
h _ _ _
iT
ei ¼
H : ðA:7Þ
incurred by elastic deformation, the period for this deformable di1 HTi    dij HTi    din HTi
beam is found to be around 13.60, which is slightly smaller than
the value for rigid beam.
Appendix B. Proof of strain objectivity

5. Conclusion A formulation is strain objective if arbitrary rigid body motions

do not incur any additional strains. Assume that the configuration
In combination with Simo’s energy–momentum conserving at the ith integration points of an element at time t ¼ t m is defined
algorithm, a weak form quadrature element formulation of geo- by ðrmi ; Kmi Þ, while the configuration at time t ¼ t mþ1 resulting
metrically exact beam model is proposed for dynamic analysis of from an arbitrary rigid body motion ður ; #r Þ superposed onto the
spatial beams undergoing large displacements and rotations. The previous configuration is characterized by rðmþ1Þi ; Kðmþ1Þi , i.e.
high-order approximation property of the QEM enhances compu-
tational efficiency and circumvents shear and membrane locking rðmþ1Þi ¼ cay½#r ðrmi þ ur Þ;
ðB:1Þ
problems. The coincidence of integration points and nodes within Kðmþ1Þi ¼ cay½#r Kmi :
a quadrature element enables the exact preservation of the
strain-objectivity of the beam model and energy-conserving prop- The strain measures correspond to time instants tm and t mþ1 are
 
erty of the algorithm after discretization without extra effort, thus denoted as ðcmi ; jmi Þ and cðmþ1Þi ; jðmþ1Þi , respectively. In the pre-
making the QEM superior in dynamic analysis of geometrically sent quadrature element formulation, the derivatives of the rigid
exact beams. Further work on incoporation of better dynamic algo- body motion are approximated by the differential quadrature ana-
rithms and dynamic analysis of more complex structures such as log. From Eqs. (34) and (35), the following relation holds:
plates and shells undergoing large displacements and rotations is
under way and will be reported in the future. X
n
ð1Þ
C ij ¼ 0: ðB:2Þ
j¼1
Acknowledgment
With Eq. (B.2), it can be shown that
The present investigation was performed with the support of X
n
r0ðmþ1Þi  r0mi ¼ #^r cay½#r ðrmi þ ur Þ C ij
ð1Þ
the National Natural Science Foundation of China (No. 51378294).
j¼1
!
Appendix A. Expression of element tangent stiffness matrix X
n
ð1Þ
þ cay½#r  r0mi þ ur C ij  r0mi
j¼1
Prior to establishment of the element tangent stiffness matrix,
the following expressions are introduced. ¼ ðcay½#r   IÞr0mi ; ðB:3Þ

2 3
2
A I
Dt q 33
033
6   7
Hi ¼ 4 5; ðA:1Þ
033 Kmþ1 Iq KTmþ1 w^ mþ1 þ D2t U 1 #ðmÞ þ cay #d
ðmÞ d
wm  Kmþ1 Iq K T
mþ1 wmþ1
i
106 R. Zhang, H. Zhong / Computers and Structures 165 (2016) 96–106

1^ ^ Xn
ð1Þ
[17] Ibrahimbegovic A, Mamouri S. Energy conserving/decaying implicit time-
#  r0mþ1=2 ¼ #r #r cay½#r ðrmi þ ur Þ C ij stepping scheme for nonlinear dynamics of three-dimensional beams
2 j¼1 undergoing finite rotations. Comput Methods Appl Mech Eng
" ! # 2002;191:4241–58.
1^ Xn
ð1Þ [18] Romero I, Armero F. An objective finite element approximation of the
0 0
þ #r cay½#r  rmi þ ur C ij þ rmi kinematics of geometrically exact rods and its use in the formulation of an
2 j¼1 energy-momentum conserving scheme in dynamics. Int J Numer Methods Eng
2002;54:1683–716.
1^
¼ #r ðcay½#r  þ IÞr0mi : ðB:4Þ [19] Lens E, Cardona A, Geradin M. Energy preserving time integration for
2 constrained multibody systems. Multibody System Dyn 2004;11:41–61.
[20] Lens E, Cardona A. A nonlinear beam element formulation in the framework of
In combination with Eq. (30), the following identities hold after an energy preserving time integration scheme for constrained multibody
some straightforward manipulations: system dynamics. Comput Struct 2008;86:47–63.
  [21] Gams M, Planinc I, Saje M. Energy conserving time integration scheme for
1
cðmþ1Þi  cmi ¼ KTðmþ1=2Þi ðcay½#r   IÞr0mi  #^r ðcay½#r  þ IÞr0mi geometrically
2007;196:2117–29.
exact beam. Comput Methods Appl Mech Eng
2
[22] Zhong H, Yu T. Flexural vibration analysis of an eccentric annular Mindlin
¼ KTðmþ1=2Þi ðcay½#r   IÞr0mi  ðcay½#r   IÞr0mi ¼ 031 ; ðB:5Þ plate. Arch Appl Mech 2007;77:185–95.
[23] Zhong H, Yu T. A weak-form quadrature element method for plane elasticity
problems. Appl Math Model 2009;33:3801–14.
X
n
[24] Zhong H, Gao M. Quadrature element analysis of planar frameworks. Arch Appl
jðmþ1Þi  jmi ¼ KiT
ðmþ1=2Þ #r
ð1Þ
C ij ¼ 031 : ðB:6Þ Mech 2010;80(12):1391–405.
j¼1 [25] Zhong H, Zhang R, Yu H. Buckling analysis of planar frameworks using the
quadrature element method. Int J Struct Stab Dyn 2011;11:363–78.
Therefore, strain measures keep uniform after arbitrary rigid [26] He R, Zhong H. Large deflection elasto-plastic analysis of frames using the
body motions, proving that the present formulation ensures the weak-form quadrature element method. Finite Elem Anal Des
property of strain objectivity. 2012;509:125–33.
[27] Xiao N, Zhong H. Nonlinear quadrature element analysis of planar frames
based on geometrically exact beam theory. Int J Non-linear Mech
References 2012;47:481–8.
[28] Zhang R, Zhong H. Weak form quadrature element analysis of planar slender
[1] Simo JC. A finite strain beam formulation. The three-dimensional dynamic beams based on geometrically exact beam theory. Arch Appl Mech 2013;83
problem. Part I. Comput Methods Appl Mech Eng 1985;49:55–70. (9):1309–25.
[2] Simo JC, Vu-Quoc L. A three-dimensional finite-strain rod model. Part II: [29] Zhang R, Zhong H. Weak form quadrature element analysis of spatial
Computational aspects. Comput Methods Appl Mech Eng 1986;58:79–116. geometrically exact shear-rigid beams. Finite Elem Anal Des 2014;87
[3] Jelenić G, Saje M. A kinematically exact space finite strain beam model- finite (15):22–31.
element formulation by generalized virtual work principle. Comput Methods [30] Zhong H, Zhang R, Xiao N. A quaternion-based weak form quadrature element
Appl Mech Eng 1995;120:131–61. formulation for spatial geometrically exact beams. Arch Appl Mech 2014;84
[4] Zupan E, Saje M, Zupan D. The quaternion-based three-dimensional beam (12):1825–40.
theory. Comput Methods Appl Mech Eng 2009;198:3944–56. [31] Davis PI, Robinowitz P. Methods of numerical integration. 2nd
[5] Sonneville V, Cardona A, Brüls O. Geometrically exact beam finite element ed. Orlando: Academic Press; 1984.
formulated on the special Euclidean group SE(3). Comput Methods Appl Mech [32] Bellman RE, Casti J. Differential quadrature and long term integration. J Math
Eng 2014;268:451–74. Anal Appl 1971;34:235–8.
[6] Jelenić G, Crisfield MA. Geometrically exact 3D beam theory: implementation [33] Simo JC, Vu-Quoc L. On the dynamics in space of rods undergoing large
of a strain-invariant finite element for statics and dynamics. Comput Methods motions – a geometrically exact approach. Comput Methods Appl Mech Eng
Appl Mech Eng 1999;171:141–71. 1988;66:125–61.
[7] Betsch P, Steinmann P. Frame indifferent beam finite elements based upon the [34] Hsiao KM, Lin JY, Lin WY. A consistent co-rotational finite element formulation
geometrically exact beam theory. Int J Numer Methods Eng 2002;54:1775–88. for geometrically nonlinear dynamic analysis of 3-D beams. Comput Methods
[8] Simo JC, Fox DD, Rifai MS. On a stress resultant geometrically exact shell Appl Mech Eng 1999;169:1–18.
model. Part III: The computational aspects of nonlinear theory. Comput [35] Hesse H, Palacios R. Consistent structural linearization in flexible-body
Methods Appl Mech Eng 1990;79:21–74. dynamics with large rigid-body motion. Comput Struct 2012;110–111:1–14.
[9] Zupan D, Saje M. Finite-element formulation of geometrically exact three- [36] Zupan E, Saje M, Zupan D. Quaternion-based dynamics of geometrically
dimensional beam theories based on interpolation of strain measures. Comput nonlinear spatial beams using the Runger–Kutta method. Finite Elem Anal Des
Methods Appl Mech Eng 2003;192:5209–48. 2012;54:48–60.
[10] Ghosh S, Roy D. Consistent quaternion interpolation for objective finite [37] Češarek P, Saje M, Zupan D. Dynamics of flexible beams: finite-element
element approximation of geometrically exact beam. Comput Methods Appl formulation based on interpolation of strain measures. Finite Elem Anal Des
Mech Eng 2008;198:555–71. 2013;72:47–63.
[11] Simo JC, Tarnow N. The discrete energy-momentum method. Conserving [38] Canuto C, Hussaini MY, Quarteroni A, Zang TA. Spectral methods,
algorithms for nonlinear elastodynamics. Zangew Math Phys 1992;43:757–92. fundamentals in single domains. Berlin: Springer; 2006.
[12] Simo JC, Tarnow N. A new energy and momentum conserving algorithm for the [39] Iura M, Atluri SN. Dynamic analysis of finitely stretched and rotated three-
nonlinear dynamics of shells. Int J Numer Methods Eng 1994;37:2527–49. dimensional space-curved beams. Comput Struct 1988;29:875–89.
[13] Simo JC, Tarnow N, Doblare M. Non-linear dynamics of three-dimensional [40] Pimenta PM, Campello EMB, Wriggers P. An exact conserving algorithm for
rods: exact energy and momentum conserving algorithms. Int J Numer nonlinear dynamics with rotational DOFs and general hyperelasticity. Part I:
Methods Eng 1995;38:1431–73. Rods. Comput Mech 2008;42:715–32.
[14] Chung J, Hulbert GM. A time integration algorithm for structural dynamics [41] Ibrahimbegović A, Mikdad MA. Finite rotations in dynamics of beams and
with improved numerical dissipation: the generalized-a method. J Appl Mech, implicit time-stepping schemes. Int J Numer Methods Eng 1988;41:781–814.
Transactions ASME 1993;60:371–5. [42] Bauchau OA, Damilamo G, Theron NJ. Numerical integration of non-linear
[15] Kuhl D, Crisfield MA. Energy-conserving and decaying algorithms in non-linear elastic multi-body systems. Int J Numer Methods Eng 1995;38:2727–51.
structural dynamics. Int J Numer Methods Eng 1999;45:569–99. [43] Bauchau OA, Theron NJ. Energy decaying scheme for non-linear multi-body
[16] Laursen TA, Meng XN. A new solution procedure for application of energy- systems. Comput Struct 1996;59:317–31.
conserving algorithms to general constitutive models in nonlinear [44] Zupan E, Saje M, Zupan D. Dynamics of spatial beams in quaternion description
elastodynamics. Comput Methods Appl Mech Eng 2001;190:6309–22. based on the Newmark integration scheme. Comput Mech 2013;51:47–64.