Computers and Structures 200 (2018) 68–85

Contents lists available at ScienceDirect

Computers and Structures

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

An explicit algorithm for geometrically nonlinear transient analysis of

spatial beams using a corotational total Lagrangian finite element
formulation
Chu Chang Huang a, Fumio Fujii b, Kuo Mo Hsiao a,⇑
a
Department of Mechanical Engineering, National Chiao Tung University, Hsinchu, Taiwan
b
Department of Mechanical Engineering, Gifu University (Emeritus), Gifu, Japan

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

Article history: An explicit method for nonlinear transient dynamic analysis of spatial beams with finite rotations using a
Received 16 June 2017 corotational total Lagrangian finite element formulation is presented. The kinematics of the beam ele-
Accepted 21 January 2018 ment is described in the current element coordinate system constructed in the current configuration
of the beam element. The element deformation and inertia nodal forces are derived by the virtual work
principle, the d’Alembert principle, and the consistent linearization of the geometrically nonlinear beam
Keywords: theory. A nodal rotation vector is used to represent the finite rotation of a base coordinate system rigidly
Explicit method
attached to each node of the discretized structure. A numerical procedure of explicit method is proposed
Corotational total Lagrangian formulation
Spatial Euler beams
for the solution of the nonlinear equations of motion. The standard central difference method is applied
Dynamic analysis to the incremental displacement vector and the incremental rotation vector, and the time derivatives of
Finite rotation displacement vector and rotation vector. The nodal orientations are updated by the incremental nodal
rotation vectors. The values of nodal rotation vectors are reset to zero in the current configuration.
In order to assess the efficiency and the accuracy of the proposed method, numerical examples are
studied and compared with the results obtained using the implicit method based on the Newmark
method.
Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction that the deformation forces and the inertia forces may be derived
by adopting different formulations.
The implicit methods based on the Newmark direct integration The CR-TL formulation is an approach blending the TL and CR
method have been extensively employed in nonlinear transient descriptions [19]. A good description of the CR formulation and
dynamic analysis of three-dimensional beam structures undergo- its relation to the TL formulation is given by Mattiasson and
ing large displacements and finite rotations using the total Lagran- Samuelsson [20]. In the TL formulation, the reference configuration
gian (TL) formulation [1–13], the corotational total Lagrangian (CR- is the initial undeformed element configuration. The reference con-
TL) formulation [14], the corotational (CR) formulation [15–17], figuration used in the CR formulation is a corotated configuration,
and the combination of CR and TL formulation [18]. In [14], a con- which is obtained as a rigid body motion of the reference configu-
sistent CR-TL formulation is proposed to derive the deformation ration of the TL formulation from the initial to the current (or
nodal force vector and the inertia nodal force vector. In [16,17], a neighboring) element configuration. The reference configurations
consistent CR framework is used to derive the internal force vector of the CR and CR-TL formulation are coincident at the current state.
and the tangent stiffness matrix, and the inertia force vector and However, the reference configuration of the CR formulation is a
the tangent dynamic matrix. In [18], the CR formulation is used moving frame, while that of the CR-TL formulation is an inertial
to develop expressions of the deformation force vector and the tan- frame which is kept fixed at the current state. Newton’s laws could
gent stiffness matrices, while the inertia force vector and the tan- hold in their simplest form in the inertial frame. Thus, it may be
gent inertia matrix are derived using the TL formulation. It seems easier to derive the inertia forces using the CR-TL formulation than
using the CR formulation.
Generally, implicit algorithms are very effective for structural
dynamics problems in which the response is predominantly gov-
⇑ Corresponding author.
erned by a relatively small number of low frequency modes, while
E-mail address: [email protected] (K.M. Hsiao).

https://doi.org/10.1016/j.compstruc.2018.01.011
0045-7949/Ó 2018 Elsevier Ltd. All rights reserved.
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 69

explicit algorithms are efficient for dynamics problems in which needed and not developed. A numerical procedure of explicit
the response is more influenced by higher modes. However, the method based on the central difference method is proposed here
application of the explicit methods in the nonlinear transient for the solution of the nonlinear equations of motion. The standard
dynamic analysis of three-dimensional beams with finite rotations central difference method is applied to the incremental displace-
has been rather limited [21–23]. Hence, the object of this paper is ment vector and the incremental rotation vector, and the time
to present an explicit integration method based on the central dif- derivatives of displacement vector and rotation vector. The nodal
ference method for nonlinear transient analysis of spatial beams orientations of the discretized structure are updated by the incre-
with finite rotations using a CR-TL finite element formulation. mental nodal rotation vectors. Then, the values of nodal rotation
In [24], a CR-TL formulation of beam element for the static non- vectors are reset to zero in the current configuration. To assess
linear analysis of three-dimensional beam structures is proposed. the efficiency and accuracy of the proposed method, numerical
To treat arbitrarily large rotation of node in space, the orientation examples are studied and compared with results obtained from
of the node is described by a base coordinate system rigidly the implicit numerical procedure based on the Newmark method
attached to each node of the discretized structure, and a nodal [14] and the results reported in the literature.
rotation vector is used to represent the finite rotation of the base
coordinate system. The coordinate system associated with the ref- 2. Finite element formulation
erence configuration of the beam element is chosen to be the cur-
rent element coordinate system constructed at the current The kinematics of the beam element and the corotational total
configuration of the element. Therefore, the current element coor- Lagrangian finite element formulation proposed in [14,24,26,27]
dinate system is an inertial local coordinate system, not a moving are adapted and employed here.
coordinate system. Three rotation parameters are defined in the
current element coordinates to determine the orientation of ele- 2.1. Basic assumptions
ment cross section. The element deformation nodal force is derived
by the virtual work principle and the consistent second order lin- The following assumptions are made in the derivation of the
earization of the fully geometrically nonlinear beam theory. This beam element behavior.
formulation is extended for the nonlinear dynamic analysis of
the beam structures in [14]. Because the coordinate system associ- (1) The beam is prismatic and slender, and the Euler-Bernoulli
ated with the reference configuration is regarded as an inertial hypothesis is valid.
local coordinate system, the first and second time derivatives of (2) The cross section of the beam is doubly symmetric.
the position vector defined in the current element coordinates (3) The unit extension and the twist rate of the centroid axis of
are the absolute velocity and acceleration. A numerical procedure the beam element are uniform.
based on the Newmark direct integration method and the (4) The cross section of the beam element does not deform in its
Newton-Raphson method is proposed for the solution of the non- own plane, and strains within this cross section can be
linear equations of motion. The standard Newmark method with neglected.
the stiffness matrices and mass matrices of the elementary beam (5) The out-of-plane warping of the cross section is the product
element and bar element was applied to the incremental displace- of the twist rate of the beam element and the Saint Venant
ment vector and incremental rotation vector, and their time warping function for a prismatic beam of the same cross
derivatives. The formulation and numerical procedure were proven section.
to be very effective by numerical examples studied in [14]. Note (6) The deformation displacements and rotations of the beam
that the nodal orientations of the discretized structure are updated element are small.
by the incremental nodal rotation vectors at each time step and (7) The strains of the beam element are small.
iteration, which entails a large storage pool and extra computa-
tional operations. Thus, an efficient procedure for orientation In conjunction with the CR-TL formulation, the sixth assump-
update may be indispensible. It can be noted that for incremental tion can always be satisfied if the element size is properly chosen.
rotation vector most of the formulation details have been clarified The assumption of small strains may not be required in the CR-TL
in [25], both in spatial and material version, including the dynam- formulation. However, only the linear elastic material is considered
ics. In order to include the nonlinear coupling among the bending, in this study. The yield strains for most engineering materials are
twisting, and stretching deformations, the terms up to the second small. Thus, the assumption of small strains is used here.
order of rotation parameters and their spatial derivatives are all
retained in the element deformation nodal forces in [14,24]. How-
2.2. Coordinate systems
ever, it was reported that the third order term of the twist rate
should be retained in the element deformation nodal forces for
In order to describe the system, we define three sets of right-
the geometric nonlinear analysis of thin-walled beams in [26,27].
handed rectangular Cartesian coordinate systems:
It was also reported in [26] that the second-order terms of the ele-
ment deformation nodal force containing the twist angle and
(1) A fixed global set of coordinates, X Gi (i = 1, 2, 3) (see Fig. 1);
slopes the beam element will converge to zero with the decrease
the nodal coordinates, displacements, rotations, velocities,
of element size, and if these terms are removed from the element
and accelerations, and the equations of motions of the sys-
nodal forces, the convergence rate of the solution may be
tem are defined in these coordinates.
increased. It seems that the complexity of the element nodal forces
could also be significantly reduced by dropping the second-order (2) Element cross section coordinates, xSi (i = 1, 2, 3) (see Fig. 1);
terms containing the twist angle and slopes the beam element. a set of element cross section coordinates is associated with
Hence, the CR-TL formulation proposed in [14,24,26,27] is adapted each cross section of the beam element. The origin of this
and used in this paper. The element deformation and inertia nodal coordinate system is rigidly attached to the centroid of the
forces are derived by the virtual work principle, the d’Alembert cross section. The xS1 axis is chosen to coincide with the
principle, and the consistent second order linearization of the fully normal of the unwrapped cross section, and the xS2 and xS3
geometrically nonlinear beam theory. Since an explicit integration axes are chosen to be the principal directions of the cross
method is used in this paper, the tangential stiffness matrix is not section.
70 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x2S x1S x1 where / ¼ /21 þ /22 þ /23 is the angle of rotation about the axis of
x2 P
2 rotation, e is the unit vector along the axis of rotation. I3 is the iden-
s v tity matrix of order 3  3; U is the skew matrix associated with the
x3S (l ,0,0)
w rotation vector /, and R/ is a rotation matrix.In this paper, the sym-
P x3S
:
u bol ð Þ denotes time derivative. The time derivative of Eq. (1) may be
1
x z Q x2S
expressed as
xP Py
db
¼ R_ / b ¼ R_ / Rt/ b ¼ x  b ð3Þ
x3 dt
X 3G
where x is the angular velocity vector. From Eqs. (1)–(3), the rela-
X 2G
tion between x and /_ may be expressed by [14,29]

X 1G
x ¼ Cu/ /_ ð4Þ

Fig. 1. Coordinate systems. Cu/ ¼ Cu/ ð/Þ ¼ ½I3 þ pU þ qUUÞ ð5Þ

 
where p ¼ 1cos
/2
/
, and q ¼ /12 1  sin/ / . When / ! 0; p ¼ 12 ; q ¼ 16.
(3) Element coordinates, xi (i = 1, 2, 3) (see Fig. 1); a set of ele-
ment coordinates is associated with each element, which is Let d/ denote the variation of the rotation vector /, and dui
constructed in the current configuration of the beam ele- (i = 1, 2, 3) denote the corresponding infinitesimal spatial rotation
ment. The origin of this coordinate system is located at node about xi axes. The vector of infinitesimal spatial rotation
1, and the x1 axis is chosen to pass through two end nodes du ¼ fdu1 ; du2 ; du2 g is related to d/ through the same relation-
(centroid of two end sections) of the element; the x2 and ship that exists between x and /_ [14,29], i.e.:
x3 axes are determined by the method proposed in [24]. Note
du ¼ Cu/ d/ ð6Þ
that the current element coordinate system is chosen to be
the coordinate system associated with the reference coro- The time derivative of Eq. (4) may be expressed by
tated configuration of the element, and is just a local inertial
coordinate system not a moving coordinate system. The x_ ¼ C_ u/ /_ þ Cu/ /€ ð7Þ
deformations, deformation nodal forces, inertia nodal forces,
and mass matrix of the element are defined in terms of these C_ u/ ð/Þ ¼ pU _
_ þ qUU _ þ qUU
þ pU _ þ qUU
_ ð8Þ
coordinates.
where x _ is the angular acceleration, and /€ is the second time
2.3. Rotation vector derivative of /.
It is well known that du; x, and x_ are vector quantities, while
For the convenience of the later discussion, the term ‘rotation _ €
d/; /, and / are not. However, function uðtÞ does not exist. In this
vector’ is used to represent a finite rotation. In this paper, the sym- paper the orientation of each node of the discretized structure is
bol { } denotes column matrix. Fig. 2 shows that a vector b is described by a triad of orthonormal vectors (a base coordinate sys-
rotated into the new position b by the application of a rotation vec- tem) rigidly attached to each node and a nodal rotation vector is
tor / ¼ f/1 ; /2 ; /3 g ¼ /e to vector b. The relation between b and b used to represent the finite rotation of the base coordinate system.
may be expressed as [28,29] The value of rotation vector / corresponding to a known base coor-
dinate system can be set arbitrarily. In order to alleviate the com-
b ¼ b cos / þ ð1  cos /Þðe  bÞe þ sin /ðe  bÞ plexity of relations among d/; /; _ /€ and du; x; x _ , in this paper,


sin / 1  cos / the value of each nodal rotation vector / corresponding a known
¼ I3 þ Uþ UU b ð1Þ
/ /2 configuration of the discretized structure is set to zero. For a given
¼ R/ b increment of rotation vector d/, the corresponding base coordinate
system is updated using / ¼ d/ and Eq. (1). Then, the value of /
2 3 defined at each node of the discretized structure is reset to zero
0 /3 /2
6 7 in the current configuration of the structure. From Eqs. (2) and
U ¼ 4 /3 0 /1 5 ð2Þ _ and C_ u/ ð/Þ/_ ¼ 0,
(5), it can be seen that Cu/ ð/Þ ¼ I; C_ u/ ð/Þ ¼ 1 U
2
/2 /1 0
when / ¼ 0. As a result, du ¼ d/; x ¼ /_ and x € when / ¼ 0.
_ ¼ /,
The rules of vector addition thus can also be applied to the addition
_ and /
of d/; /, € in this study.

2.4. Kinematics of beam element

φ
In this study only the doubly symmetric cross section is consid-
e ered. The deformations of the beam element are described in the
current element coordinate system. From the kinematic assump-
b b tions made in this paper, the deformations of the beam element
may be determined by the displacements of the centroid axis of
the beam element, orientation of the cross section (element cross
section coordinates), and the out-of-plane warping of the cross sec-
tion. Let Q (Fig. 1) be an arbitrary point in the beam element, and P
Fig. 2. Rotation vector. be the point corresponding to Q on the centroid axis. The position
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 71

vector of point Q in the undeformed and deformed configurations hi are called rotation parameters and h ¼ fh1 ; h2 ; h3 g is called the
may be expressed as [24,27] vector of rotation parameters in this study. The relation between
r0 ¼ xe1 þ ye2 þ ze3 ð9Þ (a) dh and the vector of infinitesimal spatial rotation du, (b) h_ and
the angular velocity vector x, and (c) € h and the angular acceleration
and vector x_ will be given in Section 2.5.
r ¼ fr 1 ; r2 ; r 3 g Using Eqs. (15)–(17) and the assumptions of small strains and
small deformations made for the beam element, the relationship
¼ xp ðx; tÞe1 þ v ðx; tÞe2 þ wðx; tÞe3 þ h1;x xeS1 þ yeS2 þ zeS3 ð10Þ among xp ðx; tÞ; v ðx; tÞ; wðx; tÞ in Eq. (10) may be approximated by
Z
where y and z are also the xS2 and xS3 coordinates of point Q referred x
1 2 1 2
xp ðx; tÞ ¼ u1 þ 1 þ ec  v  w dx ð21Þ
to the element cross section coordinates, respectively. 0 2 ;x 2 ;x
xp ðx; tÞ; v ðx; tÞ,and wðx; tÞ are the x1 ; x2 and x3 coordinates of point
P, respectively, in the deformed configuration. h1;x ¼ h1;x ðx; tÞ is the where u1 is the displacement of node 1 in the x1 direction. Note that
twist rate of the deformed centroid axis. x ¼ xðy; zÞ is the Saint due to the definition of the element coordinate system, the value of
Venant warping function for a prismatic beam of the same cross u1 is equal to zero. However, the variation and time derivatives of u1
section, and ei and eSi (i = 1, 2, 3) denote the unit vectors associated are not zero. The axial displacements of the centroid axis,
uðx; tÞ ¼ xp ðx; tÞ  x may be determined from the lateral deflections
with the xi and xSi axes, respectively. Note that ei and eSi are coinci-
and the unit extension of the centroid axis by using Eq. (21).
dent in the undeformed state. Here, the triad eSi in the deformed
Making use of Eq. (21) and u2 ¼ uðL; tÞ ¼ xp ðL; tÞ  L, one obtains
state is assumed to be achieved by the successive applications of
Z L
the following two rotation vectors to the triad ei : 1 2 1 2
‘ ¼ L þ u2  u1 ¼ 1 þ ec  v  w dx ð22Þ
hn ¼ hn n ð11Þ 0 2 ;x 2 ;x

and in which ‘ is the current chord length of the centroid axis of the
beam element, L is the length of the undeformed beam axis, and
ht ¼ h1 t ð12Þ u2 is the displacement of node 2 in the x1 direction. Making use of
where the assumption of uniform unit extension, ec may be calculated
n o by using Eq. (22) and expressed by
1=2 1=2
n ¼ 0; h2 =ðh22 þ h23 Þ ; h3 =ðh22 þ h23 Þ ¼ f0; n2 ; n3 g ð13Þ
ð23Þ
t ¼ fcos hn ; h3 ; h2 g ð14Þ
ua ¼ fu1 ; u2 g ð24Þ
 1=2
cos hn ¼ 1  h22  h23 ð15Þ
Ga ¼ f1; 1g ð25Þ
@wðx; tÞ @wðx; tÞ @x w0 h2 The values of v ;x and w;x will converge to zero with the decrease
h2 ¼  ¼ ¼ ¼ ð16Þ
@s @x @s 1 þ ec 1 þ ec of the element size. Thus, the underlined term in Eq. (23) may be
negligible, and will not be considered in this study. However,
@ v ðx; tÞ @ v ðx; tÞ @x v0 h3 the values of v_ ;x and w _ ;x will converge to constants with the
h3 ¼ ¼ ¼ ¼ ð17Þ
@s @x @s 1 þ ec 1 þ ec decrease of the element size. Thus, the underlined terms in
Eq. (23) are retained in the derivation process.
@s
ec ¼ 1 ð18Þ Substituting Eqs. (23)–(25) into Eq. (21), xp may be expressed by
@x Z Z
x L   1 x 2 
in which n is the unit vector perpendicular to the vectors e1 and eS1 , xp ðx; tÞ ¼ x þ Nta ua þ v 2;x þ w2;x dx  v ;x þ w2;x dx
2L 0 2 0
and t is the tangent unit vector of the deformed centroid axis of the
beam element. Note that the orientation of eS1 coincides with that of ð26Þ
t. h1 is the rotation about vector t. hn is the rotation about vector n, n x xo
measured from x1 axis to xS1 axis; h2 ¼ w0 ; h3 ¼ v 0 ; ec is the unit Na ¼ 1  ; ð27Þ
L L
extension of the centroid axis, and s is the arc length of the
deformed centroid axis measured from node 1 to point P. In this Here, the lateral deflections of the centroid axis v ðx; tÞ and
paper, the symbol ð Þ0 denotes ð Þ;x ¼ @ð Þ=@x . wðx; tÞ are assumed to be the Hermitian polynomials of x, and the
Using Eqs. (1) and (11)–(15), the relation between the vectors ei rotation about the centroid axis h1 ðx; tÞ is assumed to be the linear
polynomials of x. v ðx; tÞ; wðx; tÞ and h1 ðx; tÞ may be expressed by
and eSi (i = 1, 2, 3) in the element coordinate system may be
obtained as [24] v ðx; tÞ ¼ fN1 ; N2 ; N3 ; N4 gt fv 1 ; v 01 ; v 2 ; v 02 g ¼ Ntb ub ð28Þ
eSi ¼ Rh ei ð19Þ
wðx; tÞ ¼ fN1 ; N2 ; N3 ; N4 gt fw1 ; w01 ; w2 ; w02 g ¼ Ntc uc ð29Þ
2 3
cos hn h3 h2
6 7 h1 ðx; tÞ ¼ fN 1d ; N2d gt fh11 ; h12 g ¼ Ntd ud ð30Þ
Rh ¼ 4 h3 cos hn þ ð1  cos hn Þn22 ð1  cos hn Þn2 n3 5
h2 ð1  cos hn Þn2 n3 cos hn þ ð1  cos hn Þn23 where v j and wj (j = 1, 2) are nodal values of v and w at nodes j,
2 3
1 0 0 respectively; v 0j ¼ h3j and w0j ¼ h2j (j = 1, 2) are nodal values of v ;x
6 7 (h3 ) and w;x (h2 ) at nodes j, respectively and h1j (j = 1, 2) are nodal
 4 0 cos h1  sin h1 5 ð20Þ
values of h1 at nodes j. Note that, due to the definition of the
0 sin h1 cos h1
element coordinates, the values of v j and wj (j = 1, 2) are zero. How-
where Rh is a rotation matrix, and both n2 and n3 are defined in ever, their variations and time derivatives are not zero. Ni (i = 1–4)
Eq. (13). The rotation matrix Rh is determined by hi (i = 1, 2, 3). Thus, and Nid (i = 1, 2) are shape functions and are given by
72 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85

1 1 1 1 1
N1 ¼ ð2x3  3x2 L þ L3 Þ; N2 ¼ ðx3 L  2x2 L2 þ xL3 Þ ð31Þ e11 ¼ ec  yv ;xx  zw;xx þ e2c þ ðy2 þ z2 Þh21;x þ y2 v 2;xx
L3 L3 2 2 2
1
1 1 þ z2 w2;xx  yh1 w;xx þ zh1 v ;xx þ yzv ;xx w;xx ð44Þ
N3 ¼ ð2x þ 3x LÞ;
3 2
N4 ¼ ðx L  x L Þ
3 2 2 2
3 3
L L
1 1
x x e12 ¼ ðx;y  zÞh1;x þ zðv ;x w;xx  w;x v ;xx Þ
N1d ¼ 1  ; N2d ¼ ð32Þ 2 4
L L 1
þ x;y ec h1;x þ ðx  yx;y Þh1;x v ;xx  zx;y h1;x w;xx ð45Þ
Substituting Eqs. (19), (13)–(18) into Eq. (10), using the approx- 2
imation cos h1  1  12 h21 ; cos hn  1  12 h22  12 h23 ; sin h1  h1 , and
1 1 1
retaining all terms up to the second order of deformation parame- e13 ¼ ðx;z þ yÞh1;x þ yðw;x v ;xx  v ;x w;xx Þ þ ½x;z ec h1;x
2 4 2
ters, the position vector r may be approximated by
8 9 8 9 þ ðx  zx;z Þh1;x w;xx  yx;z h1;x v ;xx  ð46Þ
> xp þ yðh2 h1  h3 Þ þ zðh2 þ h3 h1 Þ þ h1;x x
= >
< r1 > < >
=
r ¼ r 2 ¼ v þ y½1  12 ðh3 þ h1 Þ þ zð12 h2 h3  h1 Þ þ h1;x h3 x
2 2
>
: > ; >
: >
; 2.5. Time derivatives and variation of rotation parameters
r3 w þ yð12 h2 h3 þ h1 Þ þ z½1  12 ðh22 þ h21 Þ  h1;x h2 x
ð33Þ The time derivative of Eq. (19) may be expressed as

As mentioned above, the values of rotation parameters hi (i = 1, deSi

¼ R_ h ei ¼ R_ h Rth eSi ¼ x  eSi ð47Þ
2, 3) will converge to zero, while their time derivatives h_ i and € hi (i = dt
1, 2, 3) will converge to constants with the decrease of the element where x is the angular velocity vector.
size. Thus, the coupling between rotation parameters and their From Eqs. (19) and (47), the relation between x and h_ may be
time derivatives are not considered in this study. expressed by [14]
Using Eq. (33) and the approximation 1 þ ec  1, and neglect-
ing all terms containing hi (i = 1, 2, 3) and h1;x in the first and second h_ ¼ Chu x ð48Þ
time derivative of the position vector r, one may obtain 2 3
8 9 8 9 1 a b
>
> r_ 1 > > x_ p  yh_ 3 þ zh_ 2 þ h_ 1;x x > 6 7
< > = >
< >
= Chu ¼ 4 h3 cos hn 0 5 ð49Þ
r_ ¼ r_ 2 ¼ _v  zh_ 1 ð34Þ h2 0 cos hn
>
> > > >
: > ; >
: >
;
r_ 3 w_ þ yh_ 1
where a ¼ ð1cos hn Þh3
h2 þh2
, and b ¼  ð1cos hn Þh2
h2 þh2
.
2 3 2 3
8 9 8 9 Note that in this study, the rotation parameters h2 and h3 are
€ _ _ € € _ _ €
< €r 1 >
> < xp þ yð2h1 h2  h3 Þ þ zðh2 þ 2h1 h3 Þ þ h1;x x >
= > = much smaller than unity. Thus, a  h3 =2; b  h2 =2, and the fol-
€r ¼ €r 2 ¼
>
v
€  yðh_ 23 þ h_ 21 Þ þ zðh_ 2 h_ 3  €h1 Þ þ 2h_ 1;x h_ 3 x ð35Þ lowing approximation of Chu are used throughout the paper.
:€ > ; >
: >
;
r3 € þ yðh_ 2 h_ 3 þ €h1 Þ  zðh_ 2 þ h_ 2 Þ  2h_ 1;x h_ 2 x
w 2 3
2 1
1 h3 =2 h2 =2
where 6 7
Chu ¼ 4 h3 1 0 5 ð50Þ
x_ p ¼ Nta u_ a ð36Þ h2 0 1

Z L Z x
The vector of infinitesimal spatial rotation
x du ¼ fdu1 ; du2 ; du2 g is related to dh, the variation of the vector
€a þ
€xp ¼ Nta u ðv_ 2;x þ w
_ 2;x Þdx  ðv_ 2;x þ w
_ 2;x Þdx ð37Þ
L 0 0 of rotation parameter h, through the same relationship that exists
between h_ and x [14,24], i.e.:
h_ 2 ¼ w
_ ;x ð38Þ
dh ¼ Chu du ð51Þ
h_ 3 ¼ v_ ;x ð39Þ The time derivative of Eq. (48) may be expressed by

€h2 ¼ 2e_ c w €h ¼ Chu x

_ þ C_ hu x ð52Þ
_ ;x  w
€ ;x ð40Þ
where x _ is the angular acceleration, and h € is the second time
€h3 ¼ 2e_ c v_ ;x þ v€ ;x ð41Þ derivative of h.
Note that the values of hi (i = 1, 2, 3) will converge to zero with
e_ c ¼ Gta u_ a ð42Þ the decrease of the element size, whereas the values of their time
derivatives will converge to constants. Thus, the coupling between
Note that in this study the current element coordinate system is
hi (i = 1, 2, 3) and their time derivatives are not considered in this
just a local inertial coordinate system, not a moving coordinate
study, and the approximations
system. Thus, r_ and €r are the absolute velocity and acceleration,
respectively. h_  x ð53Þ
If x, y and z in Eq. (9) are regarded as the Lagrangian coordi-
and
nates, the Green strain e11 ; e12 and e13 are given by [30]
€h  x
_ þ C_ hu x ð54Þ
1 1 1
e11 ¼ ðrt;x r;x  1Þ; e12 ¼ rt;x r;y ; e13 ¼ rt;x r;z : ð43Þ
2 2 2 2 3
0 h_ 3 =2 h_ 2 =2
Substituting Eq. (33) into Eq. (43), using h1;xx ¼ 0, and retaining 6 7
all terms up to the second order of deformation parameters
C_ hu ¼ 4 h_ 3 0 0 5 ð55Þ
yield h_ 2 0 0
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 73

are used in this study. the values of /_ j and /

€ j are equal to the values of the corresponding
angular velocity vectors xj and angular acceleration vectors x _ j at
2.6. Element nodal parameters and forces nodes j (see Eqs. (4) and (7)), respectively. Therefore, the rules of
vector addition can also be applied to the addition of m/ ; d/ ; /_ j ,
Let duj ¼ fduj ; dv j ; dwj g and duj ¼ fdu1j ; du2j ; du3j g (j = 1, 2)
j j
€ j in this study. dh are not infinitesimal spatial rotations about
and / ij
denote the virtual displacement vectors and the vectors of virtual
spatial rotation, respectively, applied at the element nodes j (j = the xi axes, thus mhij are not moments about the xi axes at nodes j.
1, 2); d/j ¼ fd/1j ; d/2j ; d/3j g denote the corresponding virtual The values of mhij are not equal to the values of corresponding mu
ij ,

variation of rotation vectors /j ¼ f/1j ; /2j ; /3j g at nodes j, and because the values of dhij are not equal to the values of correspond-
n o n o
dhj ¼ dh1j ; dh2j ; dh3j ¼ dh1j ; dw0j ; dv 0j denote the correspond- ing duij at deformed state (see Eqs. (50) and (51)). Consequently,
n o the rules of vector addition cannot be applied to the addition of
ing virtual variation of hj ¼ h1j ; h2j ; h3j ¼ fh1j ; w0j ; v 0j g at nodes dhj ; mhj ; h_ j , and €
hj .
j. duj ; duj ; d/j , and dhj (j = 1, 2) are all referred to the current ele- To assemble the element equations into the global equations,
ment coordinates. the element nodal parameters and the element nodal forces should
The element nodal force vector is obtained from the virtual be consistent with the global nodal parameters and the global
work principle and the d’Alembert principle in the current element nodal forces. Here, the global nodal parameters for the system of
coordinates. The virtual work principle requires that equations corresponding to the element local nodes j (j = 1, 2) are
chosen to be the displacement vectors uGj and the rotation vectors
dqtu f u ¼ dqt/ f / ¼ dqth f h
Z Z /Gj at nodes j, respectively, referred to the global coordinates X Gi (i
¼ ðr11 de11 þ 2r12 de12 þ 2r13 de13 ÞdV þ qdrt €rdV ð56Þ = 1, 2, 3); the corresponding element nodal parameters are the dis-
V V
placement vectors uj and the rotation vectors /j at nodes j, respec-
dqu ¼ fdu1 ; du1 ; du2 ; du2 g ð57Þ tively, referred to the current element coordinates xi (i = 1, 2, 3).
Therefore, q/ and f / are chosen to be the element nodal displace-
ment vector and the element nodal force vector in this study. As
dq/ ¼ fdu1 ; d/1 ; du2 ; d/2 g ð58Þ
just mentioned, the values of the rotation vectors at nodes j, /j ,
 are reset to zero in the current configuration of the structure in this
dqh ¼ du1 ; dh1 ; du2 ; dh2 ð59Þ
study. Thus, dq/ and f / can be transformed from the element coor-
u u  dinate system to the global coordinate system by the standard pro-
f u ¼ f u þ f u ¼ f 1 ; mu u
D I
1 ; f 2 ; m2 ð60Þ cedure of vector transformation.
n o Using Eqs. (6), (16), (17), (50), (51), (23), and retaining deforma-
D I / /
f / ¼ f / þ f / ¼ f 1 ; m/1 ; f 2 ; m/2 ð61Þ tion parameters up to the first order, the relations among dqu ; dq/
and dqh in Eqs. (57)–(59) then may be expressed as
n o
D I h h
f h ¼ f h þ f h ¼ f 1 ; mh1 ; f 2 ; mh2 ð62Þ dqh ¼ Thu dqu ð63Þ
n o n o n o
/ / / / u u u u
where f j ¼ f 1j ; f 2j ; f 3j , m/j ¼ m/1j ; m/2j ; m/3j , f j ¼ f 1j ; f 2j ; f 3j , dqu ¼ Tu/ dq/ ð64Þ
n o n o n o
mu u u u h h h h
j ¼ m1j ; m2j ; m3j , f j ¼ f 1j ; f 2j ; f 3j , mhj ¼ mh1j ; mh2j ; mh3j
2 3
(j = 1, 2). f / ; f u , and f h are generalized element internal nodal force
I3 03 03 03
6T Ta1 Tb1 03 7
vectors that are conjugate to dq/ ; dqu , and dqh , respectively. f a and
D 6 b1 7
Thu ¼ 6 7 ð65Þ
I 4 03 I3 03 03 5
f a in f a (a ¼ /; u; h) are generalized deformation nodal force
vectors and generalized inertia nodal force vectors corresponding Tb2 03 Tb2 Ta2
R R
to V ðr11 de11 þ 2r12 de12 þ 2r13 de13 ÞdV and V qdrt €rdV, respectively. 2 3
V is the volume of the undeformed beam element. de1j (j = 1, 2, 3) I3 03 03 03
60 Tc1 03 03 7
are the variation ofe1j in Eqs. (44)–(46) corresponding to dqh . r1j 6 3 7
Tu/ ¼ 6 7 ð66Þ
(j = 1, 2, 3) are the second Piola-Kirchhoff stress. For linear elastic 4 03 03 I3 03 5
material, r11 ¼ Ee11 ; r12 ¼ 2Ge12 and r13 ¼ 2Ge13 , where E is 03 03 03 Tc2
Young’s modulus and G is the shear modulus. q is the density, dr
is the variation of r in Eq. (33) corresponding to dqh , and €r is the where
absolute acceleration given in Eq. (35). Note that all terms contain-
ing hi (i = 1, 2, 3) and h1;x in €r are not considered. Thus, all terms con-
taining hi (i = 1, 2, 3) and h1;x in dr are also not considered for the
sake of consistency. ð67Þ
Note that duij are infinitesimal rotations about the xi axes, thus
mu
1j are moments about the xi axes at nodes j, respectively. Thus,
duj and mu
j are vector quantities. d/ij are not infinitesimal rota- 2 3
tions about the xi axes, and thus m/ij are not moments about the 1  12 /3j 1
/
2 2j
6 7
xi axes at nodes j. Thus, d/j and m/j are not vector quantities. How- Tcj ¼ 4 1
/
2 3j
c  12 /1j 5 ðj ¼ 1; 2Þ: ð68Þ
ever, the values of the rotation vectors at nodes j, /j , are reset to  12 /2j 1
/
2 1j
1
zero in the current configuration of the structure in this study. Note that the underlined terms of Taj in Eq. (67) will converge to
Hence, the values of d/j and m/j are equal to the values of the cor- zero with the decrease of the element size, and will not be consid-
responding duj (see Eq. (6)) and mu
j (see Eq. (56)), respectively, and ered in this study; the value of /ij (i = 1, 2, 3) in Eq. (68) are reset to
74 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85

n o n o
zero in the current configuration of the structure in this study, D/ D/ D/ D/ Dh Dh/ Dh Dh
fj ¼ f 1j ; f 2j ; f 3j ¼ f 1j þ f 1j ; f 2j ; f 3j ð81Þ
thus, the value of Tu/ in Eq. (66) is equal to I12 . However, the values
of h_ 2j ; h_ 3j , and /_ ij will converge to constants with the decrease of n o
mD/ ¼ mD/ D/
1j ; m2j ; m3j
D/
the element size, and thus, the underlined terms in Eq. (67) and j
n o
/ij (i = 1, 2, 3) in Eq. (68) are retained in the derivation process.
¼ mDh
1j ; ð1 þ ec Þm2j ; ð1 þ ec Þm3j
Dh Dh
ð82Þ
In view of Eqs. (4), (6), (48), and (51), q_ u is related to q_ h and q_ /
through the same relationship that exists between dqu and dq/ in Dh/
where j = 1, 2, f 12 ¼ f 11 ¼ h21
Dh/
21 þ L m31 þ L m22 þ L m32 .
h31 h22 h32
L
mDh Dh Dh Dh

Eq. (63), and between dqu and dqh in Eq. (64), i.e.: Using Eqs. (65)–(67), (79), (80), Tu/ ¼ I12 , and the approximation
q_ h ¼ Thu q_ u ð69Þ 1 þ ec  1, and neglecting all terms containing hi (i = 1, 2, 3) in
I I
Eq. (67), the relation between f / and f h may be approximated by
q_ u ¼ Tu/ q_ / ð70Þ I I
f/ ¼ fh ð83Þ
where q_ u ¼ fu_ 1 ; x1 ; u_ 2 ; x2 g; xj (j = 1, 2) are the angular velocity
 From Eqs. (56), (44)–(46), (33)–(42), and may be obtained.
D
fh
I
fh
vectors at nodes j, q_ / ¼ fu_ 1 ; /_ 1 ; u_ 2 ; /_ 2 g, and q_ h ¼ u_ 1 ; h_ 1 ; u_ 2 ; h_ 2 ,
n o n o If the element size is chosen to be sufficiently small, the values of
_ 0j ; v_ 0j .
h_ j ¼ h_ 1j ; h_ 2j ; h_ 3j ¼ h_ 1j ; w the rotation parameters of the deformed element defined in the
The time derivative of Eqs. (69) and (70) may be expressed by current element coordinate system may always be much smaller
than unity. Thus the higher-order terms of rotation parameters in
q € u þ T_ hu q_ u
€ h ¼ Thu q ð71Þ
the element internal nodal forces may be neglected. However, in
order to include the nonlinear coupling among the bending, twist-
q € / þ T_ u/ q_ /
€ u ¼ Tu/ q ð72Þ ing, and stretching deformations, the terms up to the second order
where of rotation parameters and their spatial derivatives are retained in
2 3 element deformation nodal forces by consistent second-order lin-
03 03 03 03 earization of Eq. (56). With the decrease of element size, the values
6 T_ T_ a1 T_ b1 03 7
6 b1 7 of rotation parameters will converge to zero, and the twist rate,
T_ hu ¼6 7 ð73Þ
4 03 03 03 03 5 curvatures and unit extension will converge to constants. However,
T_ b2 03 T_ b2 T_ a2 the convergence rates of the rotation parameters may be slower
than those of the twist rate, curvatures and unit extension with
2 3 the decrease of element size. Thus, if the terms containing the twist
03 03 03 03
60 angle and slopes of the beam element are removed from the ele-
6 3 T_ c1 03 03 7 7
T_ u/ ¼ 6 7 ð74Þ ment nodal forces, the convergence rate of the solution may be
4 03 03 03 03 5 increased for numerical studies. This belief was confirmed through
03 03 03 T_ c2 numerical examples in [26]. In [26], it was suggested that these
2 3 second-order terms should be removed. In [27], it was reported
2 3 1 _
0 a b 0 h
2 3j
 12 h_ 2j that the third-order term of the twist rate h1;x should be retained
6 6 7
T_ bj ¼ 4 h_ 2j =L c 0 7
5; T_ aj ¼ 6 
4 h3j
_ e_ c 0 7 ðj ¼ 1; 2Þ
5
in the element deformation nodal forces. Thus, in this study, the
second-order terms containing the twist angle and slopes of the
h_ 3j =L 0 0 _h2j 0 e_ c beam element are removed for all sizes of element in the element
ð75Þ deformation nodal forces. However, the second-order terms con-
taining the twist rate, curvatures and unit extension, and the
2 3
0 /_ 3j /_ 2j third-order term of the twist rate h1;x are retained in the element
_Tcj ¼ 6
1 _
4 /3j 0
7
/_ 1j 5 ðj ¼ 1; 2Þ ð76Þ deformation nodal forces. In the element inertia nodal forces, the
2 coupling between the deformation parameters and their time
/_ 2j /_ 1j 0 derivatives is not considered, and the terms up to the second order
Substitute Eq. (70) into Eq. (69), and Eq. (72) into Eq. (71), of time derivatives of deformation parameters are retained.
g gh
respectively, using Tu/ ¼ I12 ; T_ u/ q_ / ¼ 0, and the approximation For convenience, f h (g ¼ D; I) are divided into four vectors f i
1 þ ec  1 and Thu  I12 , we obtain (i = a, b, c, d) and given by
n o
Dh Dh Dh
q_ h ¼ q_ u ¼ q_ / ð77Þ f a ¼ f 11 ; f 12

Z Z Z 
3 EIp EIy EIz
€ u þ T_ hu q_ u ¼ q
€h ¼ q € / þ T_ hu q_ /
¼ AEec 1 þ ec þ h21;x dx þ w2;xx dx þ v 2;xx dx Ga ð84Þ
q ð78Þ 2 2L 2L 2L

From Eqs. (56), (63), and (64), the relation between f h ; f u , and
g g n o Z
N00b v ;xx dx
Dh Dh Dh
g f b ¼ f 21 ; mDh
31 ; f 22 ; mDh
32 ¼ ð1 þ ec ÞEI z ð85Þ
f / (g ¼ D; I) may be given by
g
f u ¼ Tthu f h
g
ð79Þ n o Z
¼ ð1 þ ec ÞEIy N00c w;xx dx
Dh Dh Dh
f c ¼ f 31 ; mDh
21 ; f 32 ; m22
Dh
ð86Þ
g g
f / ¼ Ttu/ f u ð80Þ
Dh 
g g f d ¼ mDh
11 ; m12
Dh
Note that Tu/ ¼ I12 , thus the value of f u and f / in Eq. (80) are Z Z
1
equal.Using Eqs. (65)–(67), and Tu/ ¼ I12 , and neglecting all under- ¼ ðGJ þ EIp ec Þ N0d h1;x dx þ EK I N0d h31;x dx ð87Þ
D 2
lined terms in Eq. (67), the relation between components of f/ and
n o
€ a þ f Iav
D
f h in Eqs. (80) and (79) may be given by Ih Ih Ih
f a ¼ f 11 ; f 12 ¼ ma u ð88Þ
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 75

Z
element nodal displacement vector q/ . However, the present for-
ma ¼ qA Na Nta dx ð89Þ
mulation is intended for explicit integration procedures, thus, ele-
Z
Z Z x  ment stiffness matrices are not needed and not developed here.
Iv x
f a ¼ qA Na ðv_ 2;x þ w
_ 2;x Þdx  ðv_ 2;x þ w
_ 2;x Þdx dx ð90Þ
L 0 2.7. Equations of motion
n o
Ih Ih
f b ¼ f 21 ; mIh
Ih
31 ; f 22 ; m32
Ih € b þ f Ibv
¼ mb u ð91Þ The nonlinear equations of motion may be expressed by
€ ¼ PðtÞ  FD  Fv I
MQ ð104Þ
Z Z
mb ¼ qA Nb Ntb dx þ qI z N0b N0tb dx ð92Þ where M is the mass matrix of the structure; Q€ is the second time
derivative of the vector of global nodal parameters Q ; P(t) is the
Z Z
Iv external nodal force vector; FD is the deformation nodal force vec-
f b ¼ 2qIz N0b e_ c v_ ;x dx þ 2qIz N0b h_ 1 w
_ ;x dx ð93Þ
tor, and Fv I is the inertia nodal force vector corresponding to the
n o velocity coupling terms of the element inertia nodal force. M, Fv I
Ih
fc
Ih
¼ f 31 ; mIh
Ih
21 ; f 32 ; m22
Ih € c þ f Icv
¼ mc u ð94Þ and FD are assembled from the element mass matrix mh , the vector
Iv
of velocity coupling terms of element inertia nodal forces f / (Eq.
Z Z D
(103)), and the element deformation nodal force vector f/ , respec-
mc ¼ qA Nc Ntc dx þ qIy N0c N0tc dx ð95Þ
Iv Iv
tively. mh ; f/ ,
and f/
are all calculated first in the current element
Z Z coordinates and then transformed from the element coordinate sys-
Iv
f c ¼ 2qIy N0c e_ c w
_ ;x dx  2qIy N0c h_ 1 v_ ;x dx ð96Þ tem to the global coordinate system before assemblage by the stan-
dard procedure of coordinate transformation.
 Iv
Ih
f d ¼ mIh €
11 ; m12 ¼ md ud þ f d
Ih
ð97Þ
3. Numerical procedure
Z Z
md ¼ qIp Nd Ntd dx þ qIx N0d N0td dx ð98Þ An incremental method based on the central difference method
is proposed here for the solution of the nonlinear equations of
Z motion. The basic steps involved in the numerical solution of Eq.
Iv
f d ¼ qðIy  Iz Þ Nd v_ ;x w
_ ;x dx ð99Þ (104) are outlined as follows. Let Dt denote the time step size,
and time t n ¼ tn1 þ Dt (n P 1). Let Q ; Q_ n1 , and Q
n1
€ n1 denote
Z Z Z the vector of global nodal parameters, velocity, and acceleration
Iy ¼ z2 dA; Iz ¼ y2 dA; Ip ¼ Iy þ Iz ; Ix ¼ x2 dA ð100Þ of the structure at time t n1 (n P 1), respectively, and
DQ n1 ¼ Q n  Q n1 :
Z n o Z eSG
Let n ij (i = 1, 2, 3, j = 1, 2) denote the unit vectors associated
2 2
J¼ ðz þ x;y Þ2 þ ðy þ x;z Þ2 dA; KI ¼ ðy þ z Þ dA
2
with the xSi axes of the element cross section coordinates referred
to the global coordinates at the element local nodes j and at time
R
in which the range of integration for the integral ( )dx in Eqs. (84)– tn (n P 0).
(99) is from 0 to L; A is the cross section area; Ni (i = a, b, c, d) are Let Q ; Q ; DQ ; Q_ n1 ; Q
n1 n n1
€ n1 ;
n1 ijeSG denote the known val-
given in Eqs. (27)–(30), and ( )0 = d( )/dx. mi (i = a, b, c, d) are the con-
ues at time tn (n P 1). The values of Q nþ1 ; DQ n ; Q_ n and Q
€ n ; eSG
sistent mass matrices of elementary beam element. u € i (i = a, b, c, d) n ij
may be obtained by the following incremental procedure.
are the second time derivative of ui defined in Eqs. (24), and (28)–
Iv
(30), and f i (i = a, b, c, d) are the vectors of velocity coupling terms
R (a) Extract D/Gj , the global incremental nodal rotation vector
of the element inertia nodal forces. The integrals ( )dx in Eqs. (84)–
corresponding to each element local nodes j (j = 1, 2), from
(99) are calculated using four points Gaussian quadrature.In view of
I DQ n1 for each element. Extract uGj , the global nodal dis-
Eqs. (88)–(99), the element inertia nodal force vector f h may be
placement vector corresponding to each element local nodes
expressed by
j (j = 1, 2), from Q n for each element. Determine n eSG
ij by
Iv
I
€h þ fh
f h ¼ mh q ð101Þ applying the incremental rotation vector D/Gj to n1 eSG
ij . Then
determine the current element coordinates, the element
where mh is the element mass matrix assembled by the submatrices
Iv
deformation nodal parameters hij , and the unit extension ec
mi (i = a, b, c, d) given in Eqs. (89), (92), (95), and (98), and f h is the
ij ; uj , and the method
(Eq. (23)) for each element using n eSG G
vector of velocity coupling terms of the element inertia nodal forces
Iv described in [24]. Then, calculate the element deformation
assembled by f i (i = a, b, c, d) given in Eqs. (90), (93), (96), and (99). nodal force vector using Eqs. (84)–(87), Eqs. (80) and (81),
From Eqs. (77)–(78), (83), and (101), we may obtain and assemble the structural deformation nodal force vector
Iv FDn at time tn .
I
€/ þ f/
f / ¼ mh q ð102Þ
(b) Reset the values of elements in Q n corresponding to the
Iv Iv nodal rotation vectors to zero.
f / ¼ mh T_ hu q_ / þ f h ð103Þ (c) Extract q_ G/ , the element nodal velocity vector referred to the
The element deformation nodal force vector f / may be
D
global coordinates, for each element from Q_ b ¼ DQ =Dt
n n1

assembled by using Eqs. (81), (82), and (84)–(87). The element (an approximation for Q_ n ), and then transform q_ G/ to q_ / ,
stiffness matrix may be obtained by differentiating the the element nodal velocity vector referred to the current
D element coordinates, using the standard procedure. Then
element deformation nodal force vector f / with respect to the
76 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85

Iv time step sizes and the element sizes, until the difference between
calculate the element inertia nodal force vector f / using Eq.
Iv two successive results is negligible.
(103). Then, transform f/ from the current element coordi-
nates to the global coordinates using the standard procedure
4.1. Simple pendulum
to assemble Fvn I at time tn .
(d) Transform the element mass matrix mh (Eq. (102)) from the
The example considered is a simple stiff pendulum shown in
current element coordinates to the global coordinates using
Fig. 3. The pendulum, which was in a vertical position at rest under
the standard procedure. Then assemble the mass matrix of
a gravity field, is set in motion by applying an initial horizontal
the structure M using the element mass matrices.
€ n using MQ € n ¼ Pn  FD  Fv I (Eq. (104)). velocity V_ 0 ¼ 1:695 m=s to the lumped mass attached at the free
(e) Calculate Q n n
end of the pendulum. This example was introduced by Ibrahimbe-
(f) Calculate the vector of incremental global nodal parameters
govic and Mamouri [31] to test the performance of the proposed
€ n ; let Q
DQ n ¼ DQ n1 þ Dt2 Q and Q_ n ¼
nþ1 ¼ Q n þ DQ n , mid-point scheme [31] in the case of stiff problem. The geometry
ðDQ n þ DQ n1 Þ=2Dt. of the initial configuration, inertia properties, and material proper-
ties of the flexible beam are given in Fig. 3. Note that the dis-
When n ¼ 1; Q n1 ¼ Q 0 and Q_ n1 ¼ Q_ 0 can be obtained from tributed mass of the beam and its rotary inertia are considered
€ n1 ¼ Q
the initial conditions, Q € 0 can be calculated using Eq. here, while are neglected in [31]. The lumped mass is modeled
(104), and Q 1 can be calculated by by a one-node element with a point mass M = 1 kg. The flexible
beam is discretized by one beam element. The time step sizes
Dt €
2
Q 1 ¼ Q 0 þ DtQ_ 0 þ Q 0: Dt ¼ 2  106 s is used for the CD method, Dt ¼ 103 ; 105 and
2
5  106 s are used for the Newmark method, and Dt ¼ 102 and
103 s are used for the Newmark-D method.
4. Numerical studies The present results are shown in Figs. 4–9. Time histories for
the displacement components at point B obtained by using the
In order to assess the performance of the proposed method, five CD method, the Newmark method, and the Newmark-D method
numerical examples are studied in this section. The results
obtained using the proposed method are compared with
those obtained using the implicit numerical procedure present
in [14], which is based on the Newmark method with EA = 1010 N
b¼ 1
ð1 2
 aÞ ; c ¼ 1
ð1 2aÞ; a 6 0 [2]. When a ¼ 0, this method A
4 2
EI z = 103 N ⋅ m 2
corresponds to the constant average acceleration method. When
a < 0, this method limits the influence of high frequencies by ρA = 10−6 kg / m
introducing a numerical damping. However, a numerical damping g
gives also a dissipation of the total energy, which may affect the ρ I z = 10−6 kg ⋅ m
results for large time intervals. It is noted that for beam structures,
L
L = 0.5m
more robust methods have been proposed in [13]. As suggested in
[14], the standard Newmark method with the stiffness matrices g = 9. 8m / s 2
and mass matrices of the elementary beam element and bar ele- B . M = 1kg
ment are applied to the incremental nodal displacement vector V0
and incremental nodal rotation vector, and the time derivatives X 2G , V
of nodal displacement vector and nodal rotation vector. The ele- M
ment internal nodal forces given in Eqs. (81)–(99) are also used
for the Newmark method. The nodal position vectors and nodal X1G , U
orientations of the discretized structure and time derivatives of
nodal displacement vector and rotation vector are updated by Fig. 3. Simple pendulum.

the increment or corrector of nodal displacement vectors and

nodal rotation vectors at each increment and iteration. Then, the
values of nodal rotation vectors are reset to zero in the current con-
CDM Δt = 2×10-6s Newmark Δt = 1×10-3s
figuration. Therefore, the rules of vector addition can also be
0.4 [31] Δt = 0.1s Newmark-D Δt = 1×10-3s
applied to the addition of m/ ; d/ ; /_ j , and /
j j
€ j for the Newmark
method in this study. The convergence criterion used for the
pffiffiffiffi
Tip displacement (m)

Newmark method is given by kWk= N 6 etol , where kWk is the 0.2 V

B
Euclidean norm of the unbalanced force vector, N is the number
of equilibrium equations, and etol is a prescribed error tolerance.
etol ¼ 106 is used unless otherwise stated. 0.0
For brevity, the proposed explicit algorithm is referred to as the
CD (Central Difference) method herein, while the implicit numeri-
cal procedure presented in [14] is referred to as the Newmark
-0.2
method, when a ¼ 0, and the Newmark-D method, when
a ¼ 0:05. Both methods are implemented using Fortran 90. If UB
the results with different time steps and different number of ele-
-0.4
ments are given for an example, the results with the smallest time 0 1 2 3 4 5 6 7 8 9 10
step and the largest number of elements of each method will be Time (s)
regarded as the convergence results corresponding to each
method. The convergence results are obtained by decreasing the Fig. 4. Time history of tip displacements (Simple pendulum).
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 77

Fig. 5. Time history of tip horizontal acceleration (Simple pendulum, t 2 ½0; 1 s). Fig. 8. Time history of axial force (Simple pendulum, t 2 ½0; 10 s).

Fig. 6. Time history of tip horizontal acceleration (Simple pendulum, t 2 ½0; 10 s). Fig. 9. Time history of the total energy (Simple pendulum, t 2 ½0; 10 s).

Dt ¼ 103 s, and the Newmark-D method with Dt ¼ 102 s are all

CDM Δt = 2×10-6 s in excellent agreement with those reported in [31]. The tip hori-
zontal accelerations are presented in Fig. 5 for the time interval
12
t 2 ½0; 1 s, and in Fig. 6 for the time interval t 2 ½0; 10 s. It can
be seen from Fig. 5 that the results obtained using the CD method
Axial force (N)

with Dt ¼ 2  106 s are in excellent agreement with those

8
obtained using the Newmark method with Dt ¼ 5  106 s, while
the results obtained using the Newmark-D method with

4
Dt ¼ 103 s and those reported in [31] with Dt ¼ 102 s are nearly
indistinguishable. It can also be seen from Fig. 5 that when
t > 0:5 s, the higher frequency acceleration corresponding to the
Newmark-D method with Dt ¼ 102 s is damped out, and the
0
acceleration responses of the Newmark-D method with Dt ¼ 102
0.32460 0.32465 0.32470 0.32475 0.32480 and 103 s are nearly identical. However, the discrepancy between
Time (s) the tip horizontal accelerations responses corresponding to the
Newmark method/the CD method and the Newmark-D method is
Fig. 7. Time history of axial force (Simple pendulum, t 2 ½0:3246 s; 0:3248 s). remarked. The results shown in Figs. 5 and 6 indicate that the con-
tribution of the high frequency axial vibration is remarked for the
tip horizontal accelerations responses corresponding to the New-
are shown in Fig. 4 together with the solution given in [31], which mark method and the CD method. It can be seen from the axial
was obtained by using one element with the time step size force responses, depicted in Fig. 7, for the time interval
Dt ¼ 101 s. Fig. 4 demonstrates that the results obtained using t 2 ½0:3246 s; 0:3248 s that the period of the axial force response
the CD method with Dt ¼ 2  106 s, the Newmark method with obtained using the CD method with Dt ¼ 2  106 s is
78 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85

4:4  105 s, which is in excellent agreement with the analytical Case (a) P(t )
solution of the period of the linear axial vibration
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi L P(t )
T ¼ 2p= EA=ML ¼ 4:443  105 s. It can be seen from Fig. 8 that
for the time interval t 2 ½0; 10 s, the axial force response obtained B A
50
using the CD method with Dt ¼ 2  106 s are in excellent agree- X 3G
ment with those obtained using the Newmark method with X 2G L
t
Dt ¼ 105 s. However, similar to the tip horizontal accelerations 0 1 2
responses, the discrepancy between the axial force responses cor- X1G Time history of load
responding to the Newmark method/the CD method and the
Newmark-D method with Dt ¼ 103 s is remarked. The maximum
Case (b)
and minimum values of the axial force are F max ¼ 21:346 N and
F min ¼ 1:1317 N for the CD method and the Newmark method with L P(t )
Dt ¼ 105 s, while are F max ¼ 15:5405 N and F min ¼ 6:92831 N for B A
P(t ) 1500
the Newmark-D method with Dt ¼ 103 N. The maximum and X 3G
X 2G 0.75 L
minimum values of the axial force for the corresponding rigid pen-
t
F max ¼ Mg þ M V_ 2 =L2 ¼ 15:546 N C 0 0 . 1 0. 2
dulum are and D 0.25 L
pffiffiffi X 1G Time history of load
F min ¼ 2Mg=2 ¼ 6:92965 N. It seems that when the higher fre-
quency axial deformation of the stiff pendulum is damped out,
the axial force responses corresponding to the Newmark-D method L = 10, EI y = EI z = GJ = 103 , EA = 106
with Dt ¼ 103 s will converge to that of a corresponding rigid ρA = 1, ρI x = 20, ρI y = ρI z = 10
pendulum.
The time history of total energy is depicted in Fig. 9 for case (b). Fig. 10. Right angle cantilever beam.
The total energy is the sum of the potential energy of the external
forces, and the strain energy and the kinetic energy of the struc- located at the quarter of the leg length from the fixed end. The
tural system. The expression of the strain energy and kinetic geometry, inertia properties and material properties of the right
energy for the beam element used here are given in Appendix A. angle are given in Fig. 10.
It can be seen that the values of total energy is equal to the initial Case (a) of this example was introduced by Simo and Vu-Quoc
kinetic energy of the lump mass, 12 mV_ 20 ¼ 1:4365 N  m, on the time [4] and has often been used in order to check the ability of various
interval [0, 10 s]. This indicates that the strain energy and the nonlinear dynamic formulations involving large rotations
potential energy of gravity are negligible comparing with the initial [3,7,9,10,14,17]. The bending shear deformation is considered in
kinetic energy of the lump mass for this stiff problem. [4], but not considered in this study. The frame is molded with 8
The CPU needed for the CD method with Dt ¼ 2  106 s and is equal elements for case (a), and with 8, 32, and 64 equal elements,
834.25 s (5  106 increments), while the Newmark-D method with for case (b). For the CD method, the time step Dt ¼ 1  103 is used
Dt ¼ 0:01 s requires 1.14 s (4066 iterations), and the Newmark in cases (a), and Dt ¼ 1  103 (8 elements), Dt ¼ 2  104 (8, 32
method with Dt ¼ 103 ; 105 and 5  106 s require 10.03 s elements), and Dt ¼ 1  104 (64 elements) are used in case (b).
(31,786 iterations), 421.08 s (2,048,982 iterations), and 857.49 s The Newmark method and the Newmark-D method with
(4,097,198 iterations), respectively. Dt ¼ 0:15 and 0.05 are used in case (a), while the Newmark
For the displacement response of stiff pendulum, the contribu- method with Dt ¼ 0:001 and the Newmark-D method with
tion of the low swinging mode is predominant and the contribu- Dt ¼ 0:01 and 0.001 are used in case (b). In order to assess the per-
tion of the high frequency axial vibration is negligible. If the formance of this formulation for a long duration of analysis, the
displacement response is the only concern, the Newmark-D computation is carried out for 75 and 50 times duration of the
method, which can take larger time steps and still produce a very pulse for case (a) and case (b), respectively. The present results
accurate displacement response, is very effective. However, for the are shown in Figs. 11–14 for case (a). Fig. 11 shows the time histo-
acceleration and axial force responses, the contributions of the
high frequency axial vibration and the low swinging mode are of
the same order of magnitude. If the accuracy of the acceleration
and axial force response is considered, the CD method or the New- 8
WB
mark method with very small time step (Dt < 105 s) is required.
The CD method requires the time step size employed to be inver- 4 WA
sely proportional to the highest frequency of the discrete system.
Displacement

However, this restriction is also related to accuracy. It seems that

0
the time step size restriction is as severe as accuracy considera-
tions require for this example.
-4
4.2. Right angle cantilever beam
-8
The example considered is a right angle cantilever beam sub- CDM, Δt =1×10-3
jected to an out-of-plane concentrated triangular pulse force with Newmark, Δt =0.15
-12
different points of application, amplitude, and duration as shown 0 5 10 15 20 25 30
in Fig. 10. Two loading conditions are considered: (a) the point
Time
impulse load with amplitude 50 and duration 2 is applied at point
A, the elbow of the right angle, (b) the point impulse load with Fig. 11. Time history of displacement in the X G3 direction at point A and B (Right
amplitude 1500 and duration 0.2 is applied at point C, point angle cantilever beam Case (a)).
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 79

10 ries of the out-of-plane displacements of the elbow and the tip of

CDM Δt = 1×10
-3
case (a) on the time interval [0, 30]. It can be seen that the results
8 Newmark Δt = 0.15 Newmark Δt = 0.05 obtained using the CD method are in excellent agreement with
Newmark-D Δt = 0.15 Newmark-D Δt = 0.05 those obtained using the Newmark method with Dt ¼ 0:15. The
6
present results of case (a) on the time interval [0, 30] are also in
Displacement WA

4 very good agreement with the results of Iura and Atluri [3] (not
shown). Figs. 12 and 13 show the time histories of the out-of-
2 plane displacements of the elbow and the tip of case (a), respec-
0 tively, on the time interval [0, 150]. The results obtained using
the CD method are in excellent agreement with those obtained
-2 using the Newmark method and the Newmark-D method with
Dt ¼ 0:05. It can be seen that the discrepancy between the results
-4
of the Newmark method with Dt ¼ 0:15 and 0.05 is not negligible
-6 when t > 50. The results of Newmark method with Dt ¼ 0:15 even-
tually becomes unstable at t ¼ 94:45, while the results of
-8 Newmark-D method with Dt ¼ 0:15 shows that no numerical
0 20 40 60 80 100 120 140
instability has appeared during the time period under considera-
Time tion. However, the use of larger time steps together with numerical
dissipation could lead to inaccurate results for long duration
Fig. 12. Time history of displacement in the X G3 direction at point A (Right angle
cantilever beam Case (a)).
response. Fig. 14 shows the time histories of acceleration in the
X G3 direction at point A of case (a) on the time interval [0, 150].
The results obtained using the CD method are in excellent agree-
ment with those obtained using the Newmark method with smal-
ler time step size Dt ¼ 0:05. It can be seen that the discrepancy
CDM Δt = 1×10
-3
16 between the results of the Newmark method and the Newmark-
Newmark Δt = 0.15 Newmark Δt = 0.05 D method with Dt ¼ 0:05 is not negligible when t > 50. From
12 Newmark-D Δt = 0.15 Newmark-D Δt = 0.05 Figs. 12 and 14, it is also shown that the acceleration response is
much more affected by the higher modes of vibration than the dis-
Displacement WB

8 placement response.
The time histories of the out-of-plane displacements of the
4 elbow, the tip and the loading point for case (b) on the time inter-
val [0, 10], depicted in Figs. 15–17, respectively, show that the
0
results of case (b) obtained using the CD method are in excellent
agreement with those obtained using the Newmark method. How-
-4
ever, for case (b) with 8 elements, the discrepancy between the
-8 results of Newmark-D method with Dt ¼ 0:01 and the results of
Newmark method with Dt ¼ 0:001 is not negligible when t > 5:5.
-12 It also can be seen from Figs. 15–17 that the discrepancy among
0 20 40 60 80 100 120 140
Time 32 Elements
8 Elements CDM Δt =2×10
-4
Fig. 13. Time history of displacement in the X G3 direction at point B (Right angle
CDM Δt =1×10
-3
Newmark Δt =1×10
cantilever beam Case (a)). -3

CDM Δt =2×10
-4
64 Elements
Newmark-D Δt =1×10 CDM Δt =1×10
-2 -4

Newmark Δt =1×10 Newmark Δt =1×10

-3 -3
6
CDM Δt = 1×10
-3
5
Acceleration of A in X3 direction

Newmark Δt = 0.05
2
4 Newmark-D Δt = 0.05
3 1
Displacement WA
G

2
1 0
0
-1 -1
-2
-3 -2
-4
0 20 40 60 80 100 120 140 0 2 4 6 8 10
Time Time
Fig. 14. Time history of acceleration in the X G3 direction at point A (Right angle Fig. 15. Time history of displacement in the X G3 direction at point A (Right angle
cantilever beam Case (a)). cantilever beam Case (b)).
80 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85

32 Elements
1.4
8 Elements CDM Δt =2×10
-4

CDM Δt =1×10 1.2

-3
Newmark Δt =1×10
-3

CDM Δt =2×10
-4
64 Elements 1.0

Total energy (103)

Newmark-D Δt =1×10 CDM Δt =1×10
-2 -4
CDM Δt =1×10 (8 ele)
-3
0.8
Newmark Δt =1×10 Newmark Δt =1×10 CDM Δt =2×10 (8 ele)
-3 -3 -4

2 0.6
CDM Δt =2×10 (32 ele)
-4

0.4 CDM Δt =1×10 (64 ele)

-4

-3
0.2 Newmark Δt =10 (8 ele)
-3
1 Newmark Δt =10 (32 ele)
Displacement WB

0.0 -3
Newmark Δt =10 (64 ele)
-0.2
0 2 4 6 8 10
0 Time

Fig. 18. Time history of total energy (Right angle cantilever beam Case (b)).

-1 the Newmark method. Thus, the results of 8 elements for case

(b) are not accurate, and the results of the Newmark method with
64 elements and Dt ¼ 0:001 may be regarded as convergence
0 2 4 6 8 10 results.
The results shown in Figs. 11–18 indicate that the fundamental
Time
mode is predominant for the displacement response in case (a), but
Fig. 16. Time history of displacement in the X G3 direction at point B (Right angle the contribution from the first few higher modes is of the same
cantilever beam Case (b)). order as that of the fundamental mode for the displacement
response in case (b). This may explain why the time increment
and the element size required to obtain the convergence results
32 Elements for case (b) are much smaller than that for case (a). In case (a),
CDM Δt =2×10 the CPU needed for the CD method is 208.3861 s (150,000 incre-
-4
ments), while the Newmark method with Dt ¼ 0:05, and the
Newmark Δt =1×10
8 Elements -3
Newmark-D method with Dt ¼ 0:15 and 0.05 require 22.4953 s
CDM Δt =1×10
-3
64 Elements (14,834 iterations), 10.5925 s (6276 iterations) and 27.4718 s
Newmark-D Δt =1×10 CDM Δt =1×10
-2 -4 (16,435 iterations), respectively. In case (b) with 8 elements, the
CPU needed for the CD method with Dt ¼ 1  103 and
Newmark Δt =1×10 Newmark Δt =1×10
-3 -3
Dt ¼ 2  104 are 17.3317 s (10,000 increments) and 88.6398 s
(50,000 increments), respectively, while the Newmark method
2 with Dt ¼ 0:001, and the Newmark-D method with Dt ¼ 0:01 and
Dt ¼ 0:001 require 65.5360 s (39,144 iterations), 9.2665 s (5844
1 iterations), and 82.5401 s (49,621 iterations), respectively. In case
Displacement WC

(b) with 32 elements, the CPU needed for the CD method with
Dt ¼ 2  104 is 284.51 s (50,000 increments), while the Newmark
0 method with Dt ¼ 0:001 requires 250.35 s (40,046 iterations). In
case (b) with 64 elements, the CPU needed for the CD method with
Dt ¼ 1  104 is 1169.37 s (105 increments), while the Newmark
-1 method with Dt ¼ 0:001 requires 497.88 s (39,997 iterations).
Explicit integration is perhaps not a good approach to case (a),
for the time step required for numerical stability is very small,
-2
comparing to the period of the fundamental mode of the structure.
Nevertheless, case (a) may be used to demonstrate that the pro-
0 2 4 6 8 10 posed method is capable of handling the nonlinear transient
Time dynamic analysis of spatial beams with finite rotations. The contri-
bution of higher frequency structural modes to the response is
Fig. 17. Time history of displacement in the X G3 direction at point C. (Right angle important for case (b). It seems that the time step size required
cantilever beam Case (b)).
for numerical stability and accuracy are of the same order of mag-
nitude for case (b) of this example. The CD method may be as effi-
the results obtained using 8, 32, and 64 elements, respectively is cient as or even more efficient than the Newmark method with
remarked. However, the agreement between the results of 32 further decrease in the period of impulse and the increase in the
and 64 elements is very good when t < 4. Fig. 18 shows the time closeness of the applied load to the fixed end for this example.
histories of total energy for case (b) on the time interval [0, 10].
The total energy is constant for case (b) when t > 0:2. It can be seen 4.3. Free–free flexible beam with disks
from Fig, 18 that when t > 0:2, the values of total energy may con-
verge to a constant with the increase of the number of elements The flight of a flexible beam with rigid disks attached to its ends
and the decrease of time increment for both the CD method and shown in Fig. 19 is analyzed. This problem has been simulated first
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 81

6 4.4. Lee’s frame

A
P3G (t) The example considered is a Lee’s frame with uniform rectangu-
lar cross-section subjected to two out-of-plane forces as shown in
Fig. 21. The translations of points A and B along the X Gi (i = 1, 2, 3)
XG ,V axes are restrained. This problem has been simulated first by Le,
2 8 Battini, and Hjiaj [17] using a corotational beam element based
on the classical Euler–Bernoulli beam theory. The geometry of
X1G , U P3G (t) the initial configuration, inertia properties and material properties
of the flexible beam, and the load conditions are given in Fig. 21.
P1G (t) The Lee’s frame is discretized by using 12, 16, and 24 equal beam
X3G, W B elements, respectively. The time step size Dt ¼ 1:6  104 s with
12 and 16 elements, and 1  104 s with 24 elements are used
P1G (t) = 2P3G (t)
P1G (t) for the CD method. The time step size Dt ¼ 103 s and error toler-
EA = 104
ance etol ¼ 105 are used for the Newmark method.
EI y = EI z = GJ = 500 20 Unless otherwise stated, the results obtained using 16 elements
ρA = 1 are depicted for this example. Time histories for the displacement
components at point C obtained by using the CD method and the
ρI x = 20 t
0 2.5 5 Newmark method are shown in Figs. 22–24 together with the solu-
ρI y = ρI z = 10 tion given in [17], which was obtained by using 40 cubic elements
Time history of load
with the time step size Dt ¼ 103 s. Figs. 22–24 demonstrate that
Fig. 19. Free–free flexible beam with disks. the displacement response obtained using the CD method and
those obtained using the Newmark method are all in excellent
by Ibrahimbegović and Mikdad [7] using a beam element based on
agreement with those reported in [17]. Fig. 23 also shows that
the Reissner beam theory. The geometry of the initial configura-
the agreement among the present displacement responses
tion, inertia properties, and material properties of the flexible
obtained using 12, 16, and 24 elements, respectively, is very good.
beam and the load conditions are given in Fig. 19. The rigid disks
are modeled by a one-node element with a point mass M = 10 The time history of bending strain eEb ¼ 0:15w;xx , calculated at
and inertia qIx ¼ 200; qIy ¼ qIz ¼ 100. The flexible beam is dis- points E on the cross section passing through the last Gaussian
cretized by 10 equal beam elements. The time step sizes point of the 12th element counted from node A (see Fig. 21), are
depicted in Fig. 25. It can be seen that the bending strain response
Dt ¼ 5  103 and Dt ¼ 0:1 are used for the CD method and the
obtained using the CD method are in excellent agreement with
Newmark method, respectively.
those obtained using the Newmark method. From Figs. 24 and
Time histories for the displacement components at (initially)
25, it is also shown that the bending strain response is much more
the lower right end of the system are shown in Fig. 20 together
affected by the higher modes of vibration than the displacement
with the solution given in [7], which was obtained by using 10 lin-
response. The time history of total energy depicted in Fig. 26 shows
ear elements with Dt ¼ 0:1. As showed in Fig. 20, the present
results obtained using the CD method and the Newmark method,
respectively, are nearly indistinguishable. Qualitatively, an agree-
ment between the present results and those given in [7] can be 6m 6m
found. Quantitatively, there are considerable differences. These
may be attributed at least in part to that the bending shear defor- C B
mation is considered in [7], but not considered in this study. The 6m P
CPU time needed for the CD method is 12.1369 s (6000 incre- P
ments), and that needed for the Newmark method is 2.9952 s D X 3G , W
(1381 iterations). 6m X 2G , V
A
X 1G , U
50

CDM Δt = 5×10
-3
40
Newmark Δt = 0.1
[7] E = 2.1×1011 N m 2 , ν = 0.3, ρ = 7.85 ×103 kg m3
30
Displacement

UB
X 3S
20
P( MN )
E
10
3
VB
0 0.3 m X 2S
WB
-10
0 10 20 30
0.2 m
0 0 .1 0 .2 t ( s)
Time
Cross section Time history of load
Fig. 20. Time history of displacements at point B (Free–free flexible beam with
disks). Fig. 21. Lee’s frame.
82 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85

0.4 2
-4
CDM Δt = 1.6x10 s Ref [17]
-3
Newmark Δt = 10 s
0.3

Bending strain εb (10 )

-2
1
Displacement UC (m)

E
0.2
0

0.1
-1
CDM Δt = 1.6×10 s
-4

0.0
Newmark Δt = 1×10 s
-3

-2
0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
Time (s) Time (s)

Fig. 22. Time history of displacement in the X G1 direction at point C (Lee’s frame). Fig. 25. Time history of bending strain at point E (Lee’s frame).

No of ele Δt (s) 8
-4
CDM 12 1.6x10
-4 6
CDM 16 1.6x10
-4
CDM 24 1x10
Total energy (106N· m)

Newmark 16 1x10
-3 4
Ref [17]
0.4 2

0 CDM Δt = 1.6×10 s
-4
0.2
Newmark Δt = 1×10 s
-3
Displacement VC (m)

-2
0.0
-4
0.0 0.2 0.4 0.6 0.8
-0.2 Time (s)

Fig. 26. Time history of the total energy (Lee’s frame).

-0.4
that the values of total energy nearly keep constant with negligible
0.0 0.2 0.4 0.6 0.8 fluctuations when t > 0:2 s. It seems that the time step size
Time (s) required for numerical stability and accuracy are of the same order
of magnitude of this example.
Fig. 23. Time history of displacement in the X G2 direction at point C (Lee’s frame). The CPU needed for the CD method with 16 elements and
Dt ¼ 1:6  104 s is 11.65 s (5000 increments), while the Newmark
2 method with 16 elements and Dt ¼ 103 s requires 18.55 s (5899
iterations). The CPU needed for the CD method with 24 elements
and Dt ¼ 1  104 s is 27.14 s (8000 increments), while the New-
1 mark method with 24 elements and Dt ¼ 103 s requires 27.30 s
Displacement WC (m)

(5681 iterations).

4.5. Fixed end beam of cruciform cross section subjected to an

0
eccentric force

This example considered is a fixed end beam of cruciform cross

-1 section subjected to an eccentric rectangular pulse force
-4 PðtÞ ¼ 105 N; 0 6 t 6 103 s at point D on the cross section through
CDM Δt = 1.6x10 s Ref [17]
-3
midpoint of the beam as shown in Fig. 27. The geometry and mate-
-2 Newmark Δt = 10 s rial properties of the beam and the load history are also given in
0.0 0.2 0.4 0.6 0.8 Fig. 27. Forty and eighty equal elements are used for discretization,
respectively. For both of the CD method and the Newmark method,
Time (s)
the time step Dt ¼ 2  105 s and Dt ¼ 2  106 s are used. The
Fig. 24. Time history of displacement in the X G3 direction at point C (Lee’s frame). time histories for the displacement components in X G2 and X G3 direc-
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 83

X 3G , W 12

X 2G , V

Displacement WC (10 m)
10

-2
8
C (D)
A B X 1G , U 6
Δt = 2×10 s
-6
P
40 elements
4
CDM Newmark
L/2 L/2 80 elements
2 CDM Newmark
X 3S
0
E P(105 N ) 0 10 20 30 40 50 60 70 80
-3
Time (10 s)

t Fig. 29. Time history of displacement in the X G3 direction at point C. (Fixed end
C
h D X 2S
beam of cruciform cross section).
1
P
t 2.0

1 t (10 −3 s ) 1.5
b
Cross section Time history of load Displacement φC (rad)
1.0

0.5
−3
L = 12m, b = 0.06m, h = 0.1m, t = 5 × 10 m 0.0
11 2 3 3
E = 2.1× 10 N m , ν = 0.3, ρ = 7.8 × 10 kg m -0.5
×
Δt = 2 10 s
-6

Fig. 27. Fixed end beam of cruciform cross section. -1.0 40 elements
CDM Newmark
-1.5 80 elements
tions, and the rotation about X G1
direction at C, the centroid of the CDM Newmark
cross section through midpoint of the beam, are shown in Figs. 28– -2.0
0 10 20 30 40 50 60 70 80
30, respectively. Figs. 28–30 show that the results obtained using -3
the CD method are in excellent agreement with those obtained
Time (10 s)
using the Newmark method. For both methods, there are some dif- Fig. 30. Time history of rotation about the X G1 direction at point C. (Fixed end beam
ferences between the results obtained by Dt ¼ 2  105 s and of cruciform cross section).
Dt ¼ 2  106 s. It seems that the time step size required for accu-
racy is smaller than that required for numerical stability for this
example. Figs. 28–30 also show that the results obtained using The time histories of internal nodal force in the X G3 direction at
40 and 80 elements have the same tendency. However, their differ- point B are depicted in Fig. 31. The time histories of bending strain
ences are not negligible. eEb ¼ 0:05w;xx , calculated at points E on the cross section passing
through the last Gaussian point of the 40th element counted from
4 the node A (See Fig. 27), are shown in Fig. 32. From Figs. 31 and 32,
40 elements it can be seen that the results obtained using the CD method are in
CDM Δt = 2×10 s CDM Δt = 2×10 s
-5 -6
3 excellent agreement with those obtained using the Newmark
Newmark Δt = 2×10 s Newmark Δt = 2×10 s method. From Figs. 29, 31, and 32, it can be seen again that the
-5 -6
Displacement VC (10 m)

2 80 elements internal nodal force response and the bending strain responses
-2

are much more affected by the higher modes of vibration than

CDM Δt = 2×10 s
-6
1 the displacement response.
Newmark Δt = 2×10 s
-6
For cases with 40 elements, it only takes 28.94 s (4000 incre-
0 ments) and 296.04 s (40,000 increments) of CPU time for the CD
method with Dt ¼ 2  105 s and Dt ¼ 2  106 s, respectively,
-1
while it takes 151.60 s (20,183 iterations) and 1242.67 s (159,518
-2 iterations) of CPU time for the Newmark method with
Dt ¼ 2  105 s and Dt ¼ 2  106 s, respectively. For cases with
-3 80 elements and Dt ¼ 2  106 s, it only takes 590.51 s (40,000
increments) of CPU time for the CD method, while it takes
0 10 20 30 40 50 60 70 80
-3 2429.31 s (159,807 iterations) of CPU time for the Newmark
Time (10 s)
method. The time step size required for accuracy is smaller than
Fig. 28. Time history of displacement in the X G2 direction at point C. (Fixed end that required for numerical stability for this example. Thus, the
beam of cruciform cross section). CD method is more efficient than the standard Newmark method.
84 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85

6 second time derivative of the nodal rotation vector are equal to

80 Elements Δt = 2×10 s
-6
the values of the spatial nodal angular velocity and acceleration in
(10 N)

CDM Newmark the current configuration. Five numerical examples are studied
4

4 and compared with the results obtained using the numerical proce-
dure based on the Newmark method [14] and the results reported in
Internal nodal force F3B
G

2 the literature to assess the efficiency and accuracy of the proposed

method. Numerical results demonstrate that the present method
can be applied to nonlinear transient dynamic analyses of three-
0 dimensional beam structures undergoing large displacements and
finite rotations. From the examples studied here, it is also shown
-2 that the acceleration response, internal nodal force response and
strain response are much more affected by the higher modes of
vibration than the displacement response. For problems, in which
-4 the location of application of the load is near the fixed end and the
duration of the time of the applied impulse is much smaller than
0 10 20 30 40 50 60 70 80 the fundamental period of the structure, the contribution of higher
-3
Time (10 s) frequency modes to the displacement response, acceleration
response and internal nodal force response are important. The time
Fig. 31. Time history of internal nodal force in the X G3 direction at point B. (Fixed step size required for accuracy may be of the same order of magni-
end beam of cruciform cross section).
tude as that required for numerical stability for examples Lee’s
frame, and right angle cantilever beam Case (b). The time step size
required for accuracy is smaller than that required for numerical
6
stability for the example fixed end beam of cruciform cross section.
The proposed explicit procedure is very efficient for such kind of
4 problems that the time step size required for accuracy is smaller
than required for numerical stability, while the implicit method
Bending strain εb (10 )
-3

based on the Newmark method is effective for examples in which

2
the response is predominantly governed by a relatively small num-
E

ber of low frequency modes.

0 The element tangent stiffness matrix and the inertia matrices
are not needed and not derived in this paper. However, the element
stiffness matrix and the element inertia matrices may be easily
-2
obtained by differentiating the element nodal force vector with
respect to the element nodal parameters and their time derivatives.
80 elements Δt = 2×10 s
-6
-4
CDM Newmark
Appendix A. Expressions of strain energy and kinetic energy for
0 20 40 60 80 the beam element
-3
Time (10 s)
The strain energy of the beam element may be expressed as
Fig. 32. Time history of bending strain at point E (Fixed end beam of cruciform Z
cross section). 1 2
U¼ Ee þ 2Ge212 þ 2Ge213 dV ðA:1Þ
V 2 11
5. Conclusions
Substituting Eqs. (44)–(46) into Eq. (A.1) and neglecting the
An explicit method for nonlinear transient dynamic analysis of terms containing the twist angle and slopes of the beam element,
spatial beams with finite rotations using a corotational total Lagran- one may obtain
gian (CR-TL) finite element formulation is presented. A nodal rota- Z
1 1 1
tion vector is used to represent the finite rotation of the base U¼ AELe2c ð1 þ ec Þ þ EIz ð1 þ ec Þ v 2;xx dx þ EIy ð1 þ ec Þ
2 2 2
coordinate system rigidly attached to each node of the discretized Z Z Z
1 1
structure. The proposed element formulation is set in a fixed (iner-  w2;xx dx þ ðGJ þ EIp ec Þ h21;x dx þ EK I h41;x dx ðA:2Þ
2 8
tial) local frame constructed in the current configuration of the
beam element. Thus, Newton’s laws could hold in their simplest The kinetic energy of the beam element may be expressed as
form, and the element displacements and rotations could always Z
1
be small relative to the local element axes. The complexity of the T¼ qðr_ 21 þ r_ 22 þ r_ 23 ÞdV ðA:3Þ
2 V
inertia terms could be significantly reduced if the coupling between
the deformation parameters and their time derivatives is not con- Substituting Eq. (33) into Eq. (A.3), using the approximation
sidered. The element deformation nodal force vector and the inertia 1 þ ec  1, and retaining all terms up to the first order of the twist
nodal force vector are consistently derived by the virtual work prin- angle and slopes of the beam element, one may obtain
ciple, the d’Alembert principle, and the consistent linearization of
the fully geometrically nonlinear beam theory. The standard central
difference method is applied to the incremental displacement vec-
tor and the incremental rotation vector, and time derivatives of
the displacement vector and the rotation vector. After the nodal ori-
entations are updated by the increment of the nodal rotation vec-
tors, the values of the nodal rotation vectors are reset to zero in
ðA:4Þ
the current configuration. Thus, the values of the first and the
 C.C. Huang et al. / Computers and Structures 200 (2018) 68–85 85

The underlined terms in Eq. (A.4) may be negligible, and will [15] Crisfield MA, Galvanetto U, Jelenić G. Dynamics of 3-D co-rotational beams.
Comput Mech 1997;20:507–19.
not be considered in this study. However, these terms are relevant
[16] Le TN, Battini J-M, Hjiaj M. A consistent 3D corotational beam element for
to the velocity coupling terms of the element inertia nodal forces. nonlinear dynamic analysis of flexible structures. Comput Methods Appl Mech
Thus, the underlined terms in Eq. (A.4) are retained. Eng 2014;269:538–65.
[17] Le TN, Battini J-M, Hjiaj M. Corotational formulation for nonlinear dynamics of
beams with arbitrary thin-walled open cross-sections. Comput Struct
References 2014;134:112–27.
[18] Le TN, Battini J-M, Hjiaj M. Dynamics of 3D beam elements in a corotational
[1] Bathe KJ, Ramm E, Wilson EL. Finite element formulations for large context: a comparative study of established and new formulation. Finite Elem
deformation dynamics. Int J Numer Method Eng 1975;9:353–86. Anal Des 2012;61:97–111.
[2] Cardona A, Geradin M. A beam finite element non-linear theory with finite [19] Felippa CA, Haugen B. A unified formulation of small-strain corotational finite
rotation. Int J Numer Method Eng 1988;26:2403–38. elements: I. Theory. Comput Methods Appl Mech Eng 2005;184:2285–335.
[3] Iura M, Atluri SN. Dynamic analysis of finitely stretched and rotated three [20] Mattiasson K, Samuelsson A. Total and updated Lagrangian forms of the co-
dimensional space-curved beams. Comput Struct 1988;29:875–89. rotational finite element formulation in geometrically and materially
[4] Simo JC, Vu-Quoc L. On the dynamics in space of rods undergoing large nonlinear analysis. In: Tailor C, Hinton E, Owen DRJ, editors. Numerical
motions-a geometrically exact approach. Comput Methods Appl Mech Eng methods for non-linear problems, vol. 2. Swansea, UK: Pineridge Press; 1984.
1988;66:125–61. p. 135–51.
[5] Geradin M, Cardona A. Kinematics and dynamics of rigid and flexible [21] Belytschko T, Hsieh BJ. Nonlinear transient finite element analysis with
mechanisms using finite elements and quaternion algebra. Comput Mech convected coordinates. Int J Numer Method Eng 1973;7:255–71.
1989;4:115–35. [22] Belytschko T, Schwer L. Large displacement, transient analysis of space frames.
[6] Geradin M, Cardona A. Modeling of superelements in mechanism analysis. Int J Int J Numer Method Eng 1977;11:65–84.
Numer Method Eng 1991;32:1565–93. [23] Zupan E, Saje M, Zupan D. Quaternion-based dynamics of geometrically
[7] Ibrahimbegović A, Mikdad MA. Finite rotations in dynamics of beams and nonlinear spatial beams using the Runge-Kutta method. Finite Elem Anal Des
implicit time-stepping schemes. Int J Numer Method Eng 1998;41:781–814. 2012;54:48–60.
[8] Jelenić G, Crisfield MA. Interpolation of rotational variables in nonlinear [24] Hsiao KM. Corotational total Lagrangian formulation for three-dimensional
dynamics of 3D beams. Int J Numer Method Eng 1998;43:1193–222. beam element. AIAA J 1992;30:797–804.
[9] Jelenić G, Crisfield MA. Geometrically exact 3D beam theory: implementation [25] Ibrahimbegovic A. On the choice of finite rotation parameters. Comput
of a strain-invariant element for statics and dynamics. Comput Methods Appl Methods Appl Mech Eng 1997;149:49–71.
Mech Eng 1999;171:141–71. [26] Chen HH, Lin WY, Hsiao KM. Co-rotational finite element formulation for thin-
[10] Mäkinen J. Total Lagrangian Reissner’s geometrically exact beam element walled beams with generic open section. Comput Methods Appl Mech Eng
without singularities. Int J Numer Method Eng 2007;70:1009–48. 2006;195:2334–70.
[11] Zupan E, Saje M, Zupan D. Dynamics of spatial beams in quaternion description [27] Lin WY, Hsiao KM. Co-rotational formulation for geometric nonlinear analysis
based on the Newmark integration scheme. Comput Mech 2013;51:47–64. of doubly symmetric thin-walled beams. Comput Methods Appl Mech Eng
[12] Bourihane O, Braikat B, Jamal M, Mohri F, Damil N. Dynamic analysis of a thin- 2001;190:6023–52.
walled beam with open cross section subjected to dynamic loads using a high- [28] Goldstein H. Classical mechanics. Reading, MA: Addison-Wesley; 1980.
order implicit algorithm. Eng Struct 2016;120:133–46. [29] Borri M, Mello F, Atluri SN. Variational approaches for dynamics and time-
[13] Ibrahimbegovic A, Mamouri S. Energy conserving/decaying implicit time- finite-elements: numerical studies. Comput Mech 1990;7:49–76.
stepping scheme for nonlinear dynamics of three-dimensional beams [30] Chung TJ. Continuum mechanics. NJ: Prentice-Hall; 1988.
undergoing finite rotations. Comput Methods Appl Mech Eng [31] Ibrahimbegovic A, Mamouri S. Nonlinear dynamics of flexible beams in planar
2002;191:4241–58. motion: formulation and time-stepping scheme for stiff problems. Comput
[14] Hsiao KM, Lin JY, Lin WY. A consistent co-rotational finite element formulation Struct 1999;70:1–21.
for geometrically nonlinear dynamics analysis of 3-D beams. Comput Methods
Appl Mech Eng 1999;169:1–18.