You are on page 1of 15

Comput Mech (2007) 39:309322

DOI 10.1007/s00466-006-0029-x


A co-rotational formulation for 3D beam element using vectorial

rotational variables
Z. X. Li

Received: 14 April 2005 / Accepted: 23 December 2005 / Published online: 31 August 2006
Springer-Verlag 2006

Abstract Based on a co-rotational framework, a 1 Introduction

3-noded iso-parametric element formulation of 3D
beam was presented, which was used for accurate mod- Developing an efficient beam element formulation for
elling of frame structures with large displacements and large displacement analysis of frame structures has been
large rotations. Firstly, a co-rotational framework was an issue of many researchers (Crisfield 1996, Hsiao et
fixed at the internal node of the element, it translates al. 1987). There already exist various formulations to
and rotates with the node rigidly; then, vectorial rota- meet this requirement, Hsiao et al. (1987) had divided
tional variables were defined, they are three smaller them into three categories: Total Lagrangian formula-
components of the cross-sectional principal vectors at tion (Bathe and Bolourchi 1979, Kwak et al. 2001, Pai et
each node, sometimes they represent different compo- al. 2000, Schulz and Filippou 2001). Updated Lagrang-
nents of the cross-sectional principal vectors in incre- ian formulation (Bathe and Bolourchi 1979, Cardona
mental solution procedure so as to avoid the occurrence and Geradin 1988, Chen and Blandford 1991, Misra et
of ill-conditioned tangent stiffness matrix; thereafter, al. 2000, Teh and Clarke 1999) and co-rotational for-
the internal force vector and tangent stiffness matrix in mulation (Battini and Pacoste 2002, Crisfield 1990, Cr-
local system was derived from the strain energy of the isfield and Moita 1996, Hsiao et al. 1987, Hsiao and Lin
element as its first partial derivative and second partial 2000a, Teh and Clarke 1998), certainly, there also ex-
derivative with respect to local variables, respectively, ist some mixed type formulations of them (Jiang and
and a symmetric tangent stiffness matrix was achieved; Chernuka 1994, Hsiao and Lin 2000b, Lin and Hsiao
finally, several examples were analysed to illustrate the 2001). In addition, Simo and Vu-Quoc developed a class
reliability and accuracy of this procedure. of geometrically-exact beam formulation, this formula-
tion demonstrates its computational efficiency in large
Keywords Co-rotational procedure Vectorial displacement analyses of frame structures and benefit
rotational variable Large rotation Large in solving dynamic problems of flexible beam or beams
displacement 3D beam element system subject to large overall motions (Simo and Vu-
Quoc 1986a,b, 1988, 1991, Vu-Quoc and Deng 1995,
Vu-Quoc and Ebcioglu 1995, 1996, Vu-Quoc and Simo
1987). For convenience, these formulations can also be
This work is supported by National Natural Science Foundation classified into two groups: formulations with asymmet-
of China (50408022), and the Scientific Research Foundation for
the Returned Overseas Chinese Scholars, State Education
ric element tangent stiffness matrices and formulations
Ministry and Zhejiang Province. with symmetric element tangent stiffness matrices. Due
to the non-commutativity of spatial rotations, most
Z. X. Li (B) co-rotational formulations belong to the first group, and
Department of Civil Engineering,
the geometrically-exact beam formulation proposed by
Zhejiang University,
Hangzhou 310027, China Simo and Vu-Quoc also falls into this category (Simo and
e-mail: Vu-Quoc 1986a,b, 1988, 1991, Vu-Quoc and Deng 1995,
310 Comput Mech (2007) 39:309322

Vu-Quoc and Ebcioglu 1995, 1996, Vu-Quoc and Simo like beam and shell elements in bionic structural mod-
1987). For an asymmetric tangent stiffness matrix, more elling of dragonflys wing.
storage is occupied so as to store all its components.
Simo and Vu-Quoc (1986c) denoted that in a conser-
vative system, although their developing tangent stiff-
ness matrix is always asymmetric, it will become sym- 2 Description of the co-rotational framework
metric once the incremental loading process arrives at
an equilibrium level, Crisfield and his co-worker (Cr- In this beam element formulation, several basic assump-
isfield 1990, 1996, Crisfield and Moita 1996) had also tions were adopted: (1) all elements are straight at the
found this phenomenon, so they symmetrized element initial configuration; (2) the shape of the cross-section
tangent stiffness matrix by excluding the non-symmet- does not distort with element deforming; (3) restrained
ric term (Simo and Vu-Quoc 1986c, Crisfield 1990, 1996, warping effects are ignored. Certainly, this co-rotational
Crisfield and Moita 1996). This treatment can improve procedure and vectorial rotational variables can also
the computational efficiency greatly. Simo (1992) pre- be easily extended to solve some complicated beam
sented a rigorous justification for the symmetrization of or shell problems, such as open cross-section beams,
non-symmetric tangent stiffness matrix. Simo (1992) and curved beam, multi-layered composite beam and lami-
Crisfield (1996) also predicted that a symmetric tangent nated curved shell element etc.
stiffness matrix in the co-rotational framework could be Local coordinate system and global coordinate sys-
achieved if a certain set of additive rotational variables tem are illustrated in Fig. 1, both of them are Cartesian
were adopted. coordinate systems, where the local coordinate system
Up to now, numerous theoretical models of beams is fixed at the internal node of element, and translates
have been developed and applied to various practical and rotates with the element rigid-body translation and
circumstances. No single theory has proven to be gen- rotation, but does not deform with the element.
eral and comprehensive enough for the entire range of In order to define the initial orientation of the local
applications. Some beam formulations address better coordinate axes, an auxiliary node is prescribed, which is
performance over certain class of physical problems with located in one of the symmetry plane of the element (see
greater accuracy and efficiency rather than their general- Point A in Fig. 1). Vectors v120 and v3A0 are calculated
ity, while, other models tend to a wider range of practical from,
engineering problems, and the accuracy of the formula-
tions has been somewhat sacrificed. In this paper, the v120 = X20 X10 v3A0 = XA0 X30
author defined a set of vectorial rotational variables and
developed an advanced co-rotational formulation for where, Xi0 (i = 1, 2, 3, A) is the global coordinates of
3D beam element. In contrast with other existing beam Node i, then the orientation vectors of local axes are
element formulations for large displacement and large defined as,
rotation analysis of frame structures, this formulation
v120 v120 v3A0
has several advantages: (1) all the variables are additive ex0 = ez0 = ey0 = ez0 ex0
in an incremental solution procedure, this renders great |v120 | |v120 v3A0 |
simplification in updating vectorial rotational variables
in incremental loading; (2) a quite simple relationship is where, ex0 , ey0 , ez0 (see Fig. 1) are the normalized ori-
established between the local variables and the global entation vectors of x-axis, y-axis and z-axis in global
variables, and the transformation matrix can be derived coordinate system, respectively.
from this relationship conveniently; (3) symmetric tan- The orientation vectors eix , eiy , eiz of Node i at the
gent stiffness matrices are achieved both in local system deformed configuration are calculated from the rota-
and global system; (4) total variables are used in calcu- tional variables directly in incremental solution pro-
lating tangent stiffness matrix in local system and global cedure. In particular, at Node 3 (the internal node),
system, this ensures the accuracy and reliability of beam e3x , e3y , e3z are coincident with the orientations of lo-
element formulation. Considering the merits of the pro- cal coordinate axes,
posed co-rotational procedure and vectorial rotational
variables, the author and his co-worker (Izzuddin and LI e3x = ex e3y = ey e3z = ez
2004, Li and Izzuddin 2005) have also extended them to
2D beam element, curved shell element, multi-layered and at the initial configuration,
tube-like beam element, laminated curved shell element
and several super-elements consisting of multiple tube- e3x0 = ex0 e3y0 = ey0 e3z0 = ez0
Comput Mech (2007) 39:309322 311

Fig. 1 Definition of the y

initial configuration

e 2y 0
A r2x0
Y e3y 0 e2x0
r3y 0 x
r3x0 2 r2z0
e 1y 0 3
e3x0 e2z0
r1y 0
o r3z 0
e3z 0 z
1 e1x0
e1z 0 r1z 0


however, the initial orientation vectors of two end nodes eiy,l eiz,l + eiy,m eiz,m + eiy,n eiz,n = 0 (1a)
are defined as e2iy,l + e2iy,m + e2iy,n = 1 (1b)
eix0 = {1, 0, 0} eiy0 = {0, 1, 0} e2iz,l + e2iz,m + e2iz,n
=1 (1c)
eiz0 = {0, 0, 1}T i = 1, 2        
Firstly, assumed that eiy,l  eiy,m , eiy,l  eiy,n  (l, m,
In global coordinate system, there are 18 degree of free- n {1, 2, 3}, and l = m = n) at the end of the current
doms per element, and each node six freedoms, incremental loading or iterating step:

uG = U1 V1 W1 e1y,n1 e1y,m1 e1z,n1 , . . . ,        
T Case 1 If eiz,l  eiz,m  and eiz,l  eiz,n , then three
U3 V3 W3 e3y,n3 e3y,m3 e3z,n3 rotational variables at the next incremental loading or
where, Ui , Vi and Wi are the displacements of Node i, iterating step are eiy,n , eiy,m
 , e 
iz,n , where n m l are the
eiy,ni , eiy,mi and eiz,ni are the vectorial rotational vari- cyclic permutations of 1 2 3 , according to Eq. (1),
ables, they are three smaller components of eiy and other components of eiy and eiz are calculated from
eiz . This definition can eliminate the possibility of the these rotational variables as,
denominators approaching to zero in the partial deriv- 
eiy,l = s1 1 e2iy,n e2iy,m (2a)
atives of the rest components of eiy and eiz with respect

to eiy,ni , eiy,mi and eiz,ni [referred to Eqs. (2) and (3)], eiy,m eiy,n eiz,n + s2 eiy,l 1 e2iy,n e2iz,n
accordingly, avoid the occurrence of ill-conditioned tan- eiz,m = (2b)
gent stiffness matrix in global system. 1 e2iy,n

In local coordinate system, there are 12 freedoms per
eiz,l = s3 1 e2iz,m e2iz,n (2c)
element, and each end node 6 degree of freedoms,
 T where, s1 , s3 are the sign flags of eiy,l and eiz,l at the start
uL= u1 v1 w1 r1y,n1 r1y,m1 r1z,n1 u2 v2 w2 r2y,n2 r2y,m2 r2z,n2
of the incremental loading or iterating step, respectively,
where, ui , vi , wi are the local displacements of Node i, they are one of the numeric values of 1 or 1, s2 is also
and riy,ni , riy,mi and riz,ni are the local vectorial rotational such a constant, and it is conditioned on eT iy eiz = 0. Note
variables, which are three smaller components of riy and that n, m, l may have different values at different node
riz . This definition can avoid the occurrence of ill-condi- or at different incremental loading or iterating step.
tioned tangent stiffness matrix in local system.        
Considering that eiy and eiz are always orthogonal Case 2 If eiz,m  eiz,l  and eiz,m  eiz,n  at the end
eTiy eiz = 0, and both of them are unit vectors, the rest of the current incremental loading or iterating step, then
components of eiy and eiz can be calculated from the three rotational variables are defined as eiy,n , eiy,m , eiz,n ,
rotational variables eiy,ni , eiy,mi and eiz,ni according to the according to Eq. (1), other components of eiy and eiz are
following equations, calculated from them as,
312 Comput Mech (2007) 39:309322

Fig. 2 Diagram of the

co-rotational framework r2x e2x
r2y0 r2x0 e'2x0 r2y r2z
y e'2y0 e2y
r2z0 2
2 y x
r3y0 x e2z
e'3y0 r3x0 e'3x0e'2z0 z e 3y r3x
r3y e3x r z
e'3z0 3z
3 r3z0 3 3 e 3z

r1y r
e'1y0 r1y0 r1x0 e'1x0 e1y 1x
y 1
e1z Y
e'1z0 A e 2y0 r2y0
e3y0 r3y0 r2x0
e3x0 x 2
e1y0 3 r3x0 e2z0 r2z0
r1y0 o r3z0
1 r1x0 e3z0 z
1 e1x0
e1z0 o X

eiy,l = s1 1 e2iy,n e2iy,m (3a) and from (2) to (3), it suffers pure deformation. In the
  co-rotational framework presented in Fig.2, the process
s1 1e2iy,me2iy,n eiy,n eiz,n +s2 eiy,m 1e2iy,ne2iz,n from (1) to (2) is excluded, and only the process from
eiz,l = (2) to (3) is considered, where, (2) is treated as a pseudo
initial configuration of (3), so the relationships of local
(3b) variables and global variables are given as,

eiz,m = s3 1 e2iz,n e2iz,l (3c)
ti = R(di d3 + vi0 ) R0 vi0 (5a)
where, s1 , s2 , s3 are the same kind of constants as those riy = RRT
i ey0 (5b)
in case 1.
riz = RRT
i ez0 (5c)
Vector eix is the cross-product of vectors eiy and eiz , T T
ex0 ex
eix = eiy eiz (4) ui
where, ti = vi , R0 = ey0 , R = ey , Ri =
The definition of local vectorial rotational variables
wi eT eT
riy,ni , riy,mi and riz,ni (they are three smaller components T z0 z
of riy and riz ) follows the same route as that of global T
vectorial rotational variables (eiy,ni , eiy,mi and eiz,ni are eiy (i = 1, 2). At the right side of Eq. (5a), the first
three smaller components of eiy and eiz ). eT
Rigid-body motion contributes nothing to strain, so it term is the local coordinates of Node i at the deformed
is excluded in advance so as to achieve an element-inde- configuration, while, the second term is its initial local
pendent co-rotational formulation. In Fig. 2, (1) denotes coordinates. At the right side of Eq. (5b,5c), e iy = RT
i ey0
the initial configuration; (3) represents the current con- and e iz = RTi ez0 are the cross-sectional principal vec-
figuration. From (1) to (3), the element experiences both tors of Node i at the deformed configuration (they are
rigid-body motion and pure deformation. (2) is an inter- coincident with ey0 and ez0 at undeformed configuration
mediate configuration between (1) and (3). From (1) for a straight beam element, and the nodal orientation
to (2), the element experiences pure rigid-body motion, matrix Ri0 at initial configuration is a unit matrix, while,
Comput Mech (2007) 39:309322 313

at the deformed configuration, it becomes Ri . Note that where,

Ri e iy = Ri0 ey0 and Ri e iz = Ri0 ez0 , accordingly, e iy =
1 u0 u0 u0 x
RTi ey0 and e iz = RTi ez0 ). vi0 is the relative vector from
2 x x +  x x 
Node 3 to Node i at the initial configuration, it is calcu- (0) = u0
x r y + x
x r y ry0

u0 x

lated from, r
x z + x z rz0 )

1 u ry x (ry ry0 )
2 x0 x + x x
vi0 = Xi0 X30 i = 1, 2 (6) ry0
(1) = ry
r r

y x

Especially, at the internal node, t3 = {0, 0, 0}T , r3x0 = x r z x r z0

r3x = {1, 0, 0}T , r3y0 = r3y = {0, 1, 0}T , r3z0 = r3z =

{0, 0, 1}T . u r (rz rz0 )
12 x0 xz + x
x x
(2) = rz rz0
x ry x ry0
rz rz0
x rz x rz0
ry r r
3 Kinematics of 3-noded iso-parametric beam element 12 x xz xy0 rxz0
(3) = 0
For this 3-noded iso-parametric beam element, Lagrang-
ian interpolation functions are introduced to describe
the coordinates, displacements and vectorial rotations  
1 ry ry ry0 ry0
at any point of element in local coordinate system.
2 x x

x x
The local coordinates at any point of element are (4) = 0

depicted as 0
1 r r rz0 rz0

3 2 xz xz x x
x= hi ( ) xi0 + yl riy0 + zl riz0 (7) (5) = 0

i=1 0
3 3
where, x = {x, y, z}T ; hi ( ) is Lagrangian interpolation and u = ) ti ; ry =
i=1 hi ( riy ; ry0 =
i=1 hi ( )
3 0 3 3
function at Node i; xi0 = {xi0 , yi0 , zi0 }T are the local coor- i=1 ih ( ) r ;
iy0 zr = i=1 i h ( ) r ; r
iz z0 = i=1 hi ( )
dinates of Node i; yl and zl are the relative coordinates of riz0 ; x = / i=1 hi ( )xi0 , the prime appended above
a point to the central line of element along its two cross- and to the right of hi ( ) represents its first derivative
sectional principal vectors, respectively, in the proposed with respect to .
formulation, yl = y, zl = z.
The displacements at any point of element can be
expressed as, 4 Definition of element tangent stiffness matrix

    The strain energy of element can be calculated as below,
u= hi ( ) ti + yl riy riy0 + zl (riz riz0 ) (8)

i=1 1 T
U= DdV (10)
Considering the possibility of large displacements and V
large rotations, Green strain measure is introduced to where, D is the elastic constant matrix, D
describe the straindisplacement relationship of this
E 0 0
beam element formulation, = 0 k0 G 0 ; E and G are the elastic modulus and
0 0 k0 G
1  (u+x) (u+x) x x 
2 x x x x shear modulus, respectively; k0 is the shear factor of
= xy = (u+x) (u+x) x x cross-section; V the volume of element.
y x y

xz (u+x) (u+x) x x The internal force vector in local system is the first
x z x z derivatives of the strain energy with respect to local
variables, it is calculated from
For convenience, Eq. (9) is rewritten as, 
f= = BT DdV (11)
= (0) + yl (1) + zl (2) + yl zl (3) + y2l (4) + z2l (5) uL
314 Comput Mech (2007) 39:309322

where, where, uLj and uLk are the jth and kth components of
uL , respectively. Due to the commutativity of uLj and
(0) (1) (2) (3)
B= = + yl + zl + yl zl uLk in the differentiation of Eq. (12), kt is symmetric.
uL uL uL uL uL Equation (12) can be rewritten as
(4) (5)
+y2l + z2l   % &
uL uL T T B
kt = B DB + D dAdx
= B(0) + yl B(1) + zl B(2) + yl zl B(3) + y2l B(4) + z2l B(5) uL

Eq. (11) can be rewritten as 
= b0 A + b1 Sy + b2 Sz + b3 Syz + b4 Iy + b5 Iz

f = BT DdAdx 
+b6 Swyz + b7 Swzy + b8 Iwyz + b9 Iwy +b10 Iwz dx

= a0 A + a1 Sy + a2 Sz + a3 Syz + a4 Iy + a5 Iz
L B(0)
b0 = B(0) DB(0) + (0) D

+a6 Swyz + a7 Swzy + a8 Iwyz + a9 Iwy + a10 Iwz dx uL
B (1)
= B(0) DB(1) + (0) D + B(1) DB(0)
where, A and L are the cross-sectional
" area and"length b1
of beam element, respectively; A = dA; Sy = yl dA; uL
B (0)
" " "A "A
+(1) D
Sz = zl dA; Syz = yl zl dA; Iy = y2l dA; Iz = z2l dA; uL
A " A " A "A
Swyz = y2l zl dA; Swzy = z2l yl dA; Iwyz = y2l z2l dA; B(2)
= B(0) DB(2) + (0) D + B(2) DB(0)
"A " A A uL
Iwy = y4l dA; Iwz = z4l dA, for a beam element with
+(2) D
bisymmetric cross-section, Sy = Sz = Syz = Swyz = uL
Swzy = 0; a0 a10 can be calculated as below, B(3)
= B(0) DB(3) + (0) D + B(3) DB(0)
a0 = B(0) D(0)

B(0) B(2)
+(3) D + B(1) DB(2) + (1) D
a1 = B(1) D(0) + B(0) D(1)

uL uL
a2 = B(2) D(0) + B(0) D(2)
B (1)
+B(2) DB(1) + (2) D
a3 = B(3) D(0) + B(0) D(3) + B(2) D(1)
+B(1) D(2)
B (4)
= B(0) DB(4) + (0) D + B(4) DB(0)
a4 = B(4) D(0) + B(0) D(4) + B(1) D(1)
(0) B(1)
a5 = B(5) D(0) + B(0) D(5) + B(2) D(2) B

+(4) D + B(1) DB(1) + (1) D


a6 = B(3) D(1) + B(1) D(3) + B(4) D(2)

T T T uL uL
B (5)
+B(2) D(4)
= B(0) DB(5) + (0) D + B(5) DB(0)
(5)T (1) (1)T (5) (3)T (2)
a7 = B D +B D +B D
B(0) B(2)
+(5) D + B(2) DB(2) + (2) D
(2)T (3)
+B D uL uL
(5)T (4)
+ B(4) D(5)
a8 = B D B (3)
= B(1) DB(3) + (1) D + B(3) DB(1)
a9 = B(4) D(4)
a10 = B(5) D(5) B(1) B(4)

+(3) D + B(2) DB(4) + (2) D


uL uL
The tangent stiffness matrix in local system is the sec- (2)
+B(4) DB(2) + (4) D
ond partial derivative of the strain energy U with respect
to local variables, it is given as, uL
B (5)
# $  % &
= B(1) DB(5) + (1) D + B(5) DB(1)
2U B b7
kt = = T T
B DB + D dV uL
uLj uLk 1212 uL (1)
B B(3)
+(5) D + B(2) DB(3) + (2) D

(12) uL uL
Comput Mech (2007) 39:309322 315

B(2) The global tangent stiffness matrix is derived from fG

+B(3) DB(2) + (3) D

uL as below,
B(5) fG f TT TT
b8 = B(4) DB(5) + (4) D + B(5) DB(4)

uL ktG = = TT + f = TT kt T + f (17)
uG uG uG uG
B (4)
+(5) D
uL where, ktG is an 18 18 matrix; u G
is a 12 18 18
matrix, it is calculated from
b9 = B(4) DB(4) + (4) D
R R R16

uL uG 0 0 0 u G uG
b10 = B(5) DB(5) + (5) D
T T T 0 R 22
0 0 0 R 26
= uG uG
uL uG R R R36
0 0 u G
0 u G uG
Gaussian integral procedure is adopted to calculate 0 0 0 R 44
uG 0 R 46
the internal force vector and tangent stiffness matrix,
n0 #
the nonzero sub-matrices of uGij (i = 1, 2, 3, 4; j = 2, 4, 6)
 are given in Appendix A.2. Considering that
f= a0 A + a1 Sy + a2 Sz + a3 Syz + a4 Iy + a5 Iz
# $
i=1 T 2 uLi
+a6 Swyz + a7 Swzy + a8 Iwyz + a9 Iwy = (19)
$ uG uGj uGk 121818
+a10 Iwz wt (i)J (13) and the commutativity of the uGj and uGk in the differ-
entiation of Eq. (19), it is obvious that the second term
in the right-hand side of Eq. (17) is symmetric, further-
n0 #
 more, ktG is symmetric.
kt = b0 A + b1 Sy + b2 Sz + b3 Syz + b4 Iy + b5 Iz In incremental solution procedure, the equilibrium
equation at the start of ith loading increment is given as,
+b6 Swyz + b7 Swzy + b8 Iwyz + b9 Iwy
$ ktG0 i uG1 i = i1 P (20)
+b10 Iwz wt (i)J (14)
i where, kitG0 is the global tangent stiffness matrix at the
start of incremental loading Step i; uG1 i and i1 are
where, n0 is the number of Gaussian integral points
the increments of the global variables and the load-
along the central axis of element, n0 = 2 in solving
ing parameter, respectively; P is the prescribed external
the examples below; i and wt (i) are the dimensionless
force vector.
coordinate and weight factor at Gaussian point i, respec-
The equilibrium equation at jth iterating step of ith
tively; J is the jacobian, J = 3i=1 hi ( ) xi0 .
loading increment is given as,
The relationship of the global internal force vector fG
and the local internal force vector f is given as, ktG ij1 uG ij = ij P + Pres j2 (21)
fG = T f (15) where, ktG ij1 is the tangent stiffness matrix at the end
of (j 1)th iterating step; uG ij and ij are the incre-
where, T is the transformation matrix from global coor-
ments of the global variables and the loading parameter
dinate system to local coordinate system, it is calculated
achieved at jth iterating step; Pres is the unbalanced load-
ing vector at the end of last iterating step; i1 and ij
uLi are calculated in accordance with general displacement
Ti,j = (16a)
uGj controlling procedure (Yang and Shieh 1990).
For convenience, Eq. (21) can be rewritten as
R 0 0 0 R R16
0 R22 0 0 0 R26
0 0
(16b) ktG ij1 u1 ij = P (22a)
R 0 R R36
0 0 0 R44 0 R46 ktG ij1 u2 ij = Pres (22b)

T is a 12 18 matrix, R and Rij (i = 1, 2, 3, 4; j = 2, 4, 6) and the increments of the global variables can be given
are its sub-matrices. R represents the same matrix as as
that in Eq. (5); Rij is a 3 3 matrix (see Appendix A.1),
and 0 is a 3 3 zero matrix. uG ij = ij u1 ij + u2 ij (23)
316 Comput Mech (2007) 39:309322

Table 1 Comparison of tip displacements under different load levels

Load level (lb) Tip displacement (in)

Present study Bathe and Bolourchi Simo and Vu-Quoc

u v w u v w u v w

300 7.01 11.93 40.15 6.8 11.5 39.5 6.97 11.86 40.08
450 10.73 18.47 48.46 10.68 18.38 48.39
600 13.55 23.56 53.43 13.4 23.5 53.4 13.51 23.47 53.37

Considering that all the global variables are additive in Z 1

an incremental solution procedure, the global variables
at the end of ith loading increment are updated by


uiG = uG + uG ij (24)
j=1 O Y

where, uG is the values of the global variables at the
end of (i 1)th loading increment, and n is the number P
of iterations at ith increment.

Iterating process will be terminated if the following X

converge criterion is satisfied,
u w
 u2 i T Pres 
  err (25)
 uG i1 T i1 P 
Fig. 3 A cantilever 450 -bend with a concentrated tip load

where, err is a small constant, in analysing the following

examples, err = 105 . 5.2 A cantilever subject to an end moment

5 Examples An initially straight cantilever beam is subjected to an

end bending moment M = 2 EI/L(see Fig. 4), its width
5.1 Analysis of a cantilever 450 -bend with large and height are b = 0.5 and t = 0.1, respectively, and
displacements and large rotations its length is 100; its youngs modulus is 2.1 107 , the
Poissons ratio is 0.3, and the cross-sectional shear
A bend beam lies in XY plane, it is fixed at one end factor is 5/6.
and free at another end (see Fig. 3), a concentrated load Considering that the components of nodal external
is applied at the free end in Z-direction. The bend has force vector with respect to vectorial rotational vari-
an average radius of 100 in, and its cross-section is a ables are not moment in the proposed beam element
square with an area of 1 in2 , and its elastic modulus E formulation, the end moment is transformed into equiv-
and Poissons ratio are 107 psi and 0.0, respectively. alent components with respect to vectorial rotational
This bend is subdivided into eight beam elements
equally, and these elements are idealized as straight
beams. The tip displacements under different load lev-
els are given in Table 1. To verify the reliability and
accuracy of the procedure, the results from Bathe and
Bolourchi (1979) and Simo and Vu-Quoc (1986c) are
also presented in Table 1, it is shown that the results
from present studies can fit in well with them. X
In calculating the tip displacements under 600 lb load- M
ing level, 8 loading increments are required, and their
iteration numbers are 7, 5, 5, 4, 4, 4, 3, 3, respectively. Fig. 4 A cantilever subject to an end moment
Comput Mech (2007) 39:309322 317

Y and will run into computational difficulty once locking

phenomena in thin beam element occur.

-20 M=0
M=21.42 5.3 A portal frame subject to a concentrated load
-40 M=43.98 A portal frame is subjected to a concentrated load (see
M=54.95 Fig. 6). the cross sections of its three members are rect-
-60 angular, and their sizes are b = 0.5 and h = 0.1, respec-
tively. The material properties are E = 2.1 107 and
= 0.3, and the cross-sectional shear factor is 5/6.
-80 X
-25 0 25 50 75 100 In order to check the convergence of the proposed
procedure, each member is divided into 4, 8 and 16 equal
Fig. 5 Deformed shapes of the cantilever under different end beam elements, respectively. The curve of load against
moment levels
displacement at loading point is depicted in Fig. 7. It
is shown that four elements per member are enough
to achieve satisfying accuracy. Considering the effects
variables of end node firstly. The relationship of the rota- of shear locking and membrane locking phenomena in
tion iz and the vectorial rotational variables at end node thin beam element, reduced integration procedure can
i is given as below: alleviate or eliminate locking problems of this portal
If eiy,1 0 and eiy,2 > 0, then iz = arcsineiy,1 ; frame effectively.
If eiy,1 < 0 and eiy,2 > 0, then iz = 2 + arcsineiy,1 ; The deformed shapes of portal frame under differ-
If eiy,2 0, then iz = arcsineiy,1 . ent loading levels are depicted in Fig. 8. It demonstrates

If 22 eiy,1 22 , then eiy,1 , eiy,3 and eiz,1 are three that the proposed procedure is reliable and efficient in
vectorial rotational variables at the end node, solving large displacement and large rotation problems.

W = Miz = eiy,1 P

the equivalent load component with respect to eiy,1 is

eiy,2 , and another two components with respect to eiy,3

and eiz,1 are equal tozero.
If 22 < eiy,2 < 22 , then eiy,2 , eiy,3 and eiz,2 are three
vectorial rotational variables at the end node,
W = Miz = eiy,2
eiy,1 Fig. 6 A portal frame subject to a concentrated load

the equivalent load component with respect to eiy,2 is

eMi , and another two components with respect to eiy,3 -80
and eiz,2 are equal to zero.
This cantilever is divided into five beam elements
equally. The deformed shapes of the cantilever at differ-
ent end moment levels are depicted in Fig. 5, it experi-
ences large displacement and large rotation, and its end -40 16-elements
rotation arrives at 2 under M = 2LEI , however, the
proposed procedure still demonstrates satisfying effi-
ciency and reliability. Urthaler and Reddy (2005) and -20
Lee (1997) had also solved similar problems (but they
did not present the geometry and material properties of
0 V
the problems), and Lees procedure (Lee 1997) seems to 0 -50 -100 -150 -200
be quite efficient in solving similar problems, however, it
can not cope with buckling and post-buckling problem, Fig. 7 Response of portal frame subject to a concentrated load
318 Comput Mech (2007) 39:309322

100 -10


Present study
0 Wen & Rahimzadeh(1983)
Hsiao et al(1987)
-50 P=-44.94 -2
P=-70 w
0 0.3 0.6 0.9 1.2
-100 X
-30 0 30 60 90 120
Fig. 10 Response of space arc frame under ultimate concentrated
Fig. 8 Deformed shapes of portal frame under different load loading

5.5 Space dome subject to a concentrated load

5.4 Analysis of a space arc frame under vertical and at the apex
horizontal concentrated loading
The space dome is described in Fig. 11, all its mem-
This frame is shown in Fig. 9, it consists of two groups bers have the same rectangular cross sections, b h =
of members. For the members in the arc frame planes, 0.76 m1.22 m. The material properties are E = 2.069
the cross-section property are A1 = 0.5, Iy1 = 0.4 and 1010 and G = 8.83 109 N/m2 , respectively. This space
Iz1 = 0.133, respectively, and for the rib members, A2 =
0.1, Iy2 = 0.05 and Iz2 = 0.05, respectively. The material
properties are E = 4.32 105 and G = 1.66 105 . This 0.76
frame is pinned at four boundary nodes. In addition to
four vertical concentrated loads P, the structure is also

subjected to two lateral concentrated loads 0.001P (see
Fig. 9).
In large displacement analysis of this space arc frame,
each leg is divided into two equal beam elements, while, X
the rest members are treated as one element, respec-
tively. The curve of nodal displacement w against load P

is given in Fig. 10. It is shown that this curve is in close 21.115

agreement with the solution given by Hsiao et al. (1987)
and Wen and Rahimzadeh (1983).

P P 12.19

2 1

1 2 P

4.55 1.55

1 1 Z
6.285 X
69.28 61.44 69.28

Fig. 11 Geometry of space dome subject to an apex concentrated

Fig. 9 Space arc frame load
Comput Mech (2007) 39:309322 319




-60 Present study 2

1 X
Teh & Clarke(1999)

0 -1 -2 -3 -4 -5

Fig. 12 Load-displacement curve at the peak of space dome

dome is fixed at its six boundary nodes, and a concen-

trated load is exerted at the apex. P
In numerical analysis, each member is modelled by

one beam element. The load-displacement curve at the

apex is depicted in Fig. 12. Comparison of the results Z
from the proposed procedure and those from Teh and X
Clarke (1999) and Izzuddin (2001) is given in Fig. 12. It 25 25
is shown that they are in agreement with each other very
43.3 43.3
Fig. 13 Twenty four member star-shaped dome subject to a con-
centrated load at the apex

5.6 Twenty-four-member star-shaped shallow dome

The geometry of a 24-member star-shaped dome is -2.5
shown in Fig. 13, its member cross-section properties are
A = 3.17 cm2 , Iy = 0.837 cm4 , Iz = 0.837 cm4 , and the -2.0
elastic modulus and shear modulus are E = 3.03 105
and G = 1.096 105 N/cm2 , respectively. This dome -1.5
is pinned at six boundary nodes, and is loaded with a
concentrated load at the apex. -1.0
Present study
In numerical analysis of this dome, each member is Meek & Tan (1984)
divided into two equal beam elements. The load-dis- Hsiao et al(1987)
placement curve of this dome at the apex is depicted in
Fig. 14, where the results from Hsiao et al. (1987) and w(cm)
Meek and Tan (1984) are also given, it is shown that the 0 -1 -2 -3 -4 -5
curve from present studies can fit in well with them.
Keeping the same geometry of the dome, the same Fig. 14 Load-displacement curve of the star-shaped dome at the
loading case, and the same material properties and mem-
ber cross-sectional area, while adjusting the cross section
sizes along two principal inertia axes -y and -z: A =
3.17 cm2 , Iy = 0.295 cm4 , Iz = 2.377 cm4 , each member nomenon occurs after the concentrated load arrives at
is subdivided into two beam elements equally, the load- an ultimate level.
displacement curve at the apex is presented in Fig. 15. The results from Meek and Tan (1984) are also
It illustrates that the response of the dome is greatly depicted in Fig. 15. Meek and Tans results are very
different from that in the last case, snap-through phe- close to those from present studies.
320 Comput Mech (2007) 39:309322

P(KN) where, i = 1, 3, k = i+1 2 .

R Te R Te R Te
R y0 R y0 R y0
e3y,n k ,nr e3y,m k ,nr e3z,n k ,nr
Present study R RT e R T R
R ey0
Ri6 = e3y,n k y0 R ey0
Meek & Tan(1984) ,mr e3y,m k ,mr e3z,n k ,mr
e3y,n Rk ez0 e3y,m Rk ez0
,nr e3z,n Rk ez0
,nr ,nr
where, i = 2, 4, k = i
RTi RTi RT i
w(cm) R eiy,n ey0 R eiy,m ey0 R eiz,n ey0
0.0 ,nr ,nr ,nr
0 -1 -2 -3 -4 -5
RTi e RT
R eiz,n ey0
Fig. 15 Load-displacement curve of the dome at the apex R(2i)(2i) = R eiy,n y0
R eiy,m ey0 i

,mr ,mr ,mr

RTi RTi RT i
R eiy,n ez0 R eiy,m ez0 R eiz,n ez0
6 Conclusions ,nr ,nr ,nr

where, i = 1, 2.
Compared with the existing 3-D beam element formu-
Note that the subscripts nr and mr denote the nr th and
lations for frame structures with large displacements
mr th components of the related vectors, respectively.
and large rotations, there are several advantages in the
proposed procedure: (1) the element tangent stiffness
matrix is symmetric, so it ensures the computational A.2 Sub-matrices of uG
efficiency and the saving of storage source; (2) vec-
torial rotational variables are defined so all the vari-   
ables of freedoms are additive, and correction matrix T  R R R
= s
is avoided; (3) these variables can be used to describe uG i=i1 :i3 ,j=16:18,k=k1 :k3 e3y,n e3y,m e3z,n

large member deformation. Through several examples

where, if i1 : i3 = 1 : 3 and k1 : k3 = 1 : 3, or i1 : i3 = 7 : 9
test, the proposed procedure demonstrates satisfying
and k1 : k3 = 7 : 9, then s = 1; if i1 : i3 = 1 : 3 and
accuracy and efficiency in large displacement analyses
k1 : k3 = 13 : 15, or i1 : i3 = 7 : 9 and k1 : k3 = 13 : 15,
of frame structures. T
then s = 1. The subscripts at the right side of u G
represent the location of this sub-matrix in uG .

uG i=i1 :i3 ,j=16:18,k=16:18

2R 2R 2R
(d d3 + vl0 ) e3y,n e3y,m (dl d3 + vl0 ) e3y,n e3z,n (dl d3 + vl0 )
e23y,n l

2R 2R 2R
d3 + vl0 )
= e3y,n e3y,m (dl d3 + vl0 ) (d
e23y,m l
d3 + vl0 ) e3y,m e3z,n (dl

2R 2R 2R

e3y,n e3z,n (dl d3 + vl0 ) e3y,m e3z,n (dl d3 + vk0 )
(dl d3 + vl0 )

where, if l = 1, then i1 : i3 = 1 : 3; if l = 2, then
A. 1 Sub-matrices of T i1 : i3 = 7 : 9.

 R T 
Ri6 = (dk d3 + vk0 ) (dk d3 + vk0 ) uG i=i1 :i3 ,j=16:18,k=16:18
e3y,n e3y,m
(dk d3 + vk0 )
Comput Mech (2007) 39:309322 321

 2R T
2R T
2R T

e3y,n uGS Rl ey0 ,n e3y,m uGS Rl ey0 ,n e3z,n uGS Rl ey0 ,n
  r   r   r
2R 2R 2R
= e3y,n uGS RTl ey0 e u R Te
y0 e u R Te
 ,mr  3y,m2 GS l ,mr  3z,n2 GS l ,mr
2R Te R Te R Te
e3y,n uGS l z0 R
e3y,m uGS l z0 R
e3z,n uGS l z0
,nr ,nr ,nr

where, uGS = e3y,n , e3y,m , e3z,n T , if l = 1, then i1 : i3 = Crisfield MA (1996) Nonlinear finite element analysis of solid and
4 : 6; if l = 2, then i1 : i3 = 10 : 12. structures, vol 2. Wiley, Chichester
  Crisfield MA, Moita GF (1996) A unified co-rotational framework
T  T  for solids, shells and beams. Int J Solids Struct 33:29692992
uG i=i1 :i2 ,j=16:18,k=k1 :k3 uG i=i1 :i2 ,k=k1 :k3 ,j=16:18 Hsiao KM, Horng HJ, Chen YR (1987) A co-rotational proce-
dure that handles large rotations of spatial beam structures.
% & % & % & Comput Struct 27:769781
R Rl R Rl R Rl
T T T Hsiao KM, Lin WY (2000a) A co-rotational finite element formu-
e3y,n uGS ey0 e3y,m uGS ey0 e3z,n uGS ey0
lation for buckling and postbuckling analyses of spatial beams.
% &,nr % &,nr % &,nr Comput Methods Appl Mech Eng 188:567594
e3y,n uGS ey0
e3y,m uGS ey0
e3z,n uGS ey0

Hsiao KM, Lin WY (2000b) Co-rotational formulation for thin-
% & r%
,m & r%
,m & r
,m walled beams with monosymmetric open section. Comput
R Rl
T Methods Appl Mech Eng 190:11631185
e3y,n u ez0
GS e3y,m u ez0 GS e3z,n u ez0 GS
Izzuddin BA (2001) Conceptual issues in geometrically nonlin-
,nr ,nr ,nr ear analysis of 3D framed structures. Comput methods Appl
 T Mech Eng 191:10291053
where, if l = 1, then uGS = e1y,n , e1y,m , e1z,n , i1 : Izzuddin BA, LI Z (2004) A co-rotational formulation for large
i3 = 4 : 6 and k1 : k3 = 4 : 6; if l = 2, then uGS = displacement analysis of curved shell. In: Deeks, Hao (eds)
 T The 18th Australasian conference on the mechanics of struc-
e2y,n , e2y,m , e2z,n , i1 : i3 = 10 : 12 and k1 : k3 = 10 :
tures & materials. Taylor & Francis Group, London, pp 1247
12. 1253

T  Jiang L, Chernuka MW (1994) Co-rotational formulation for geo-
uG i=i1 :i3 ,j=j1 :j3 ,k=k1 :k3
metrically nonlinear finite element analysis of spatial beams.
 Trans Can Soc Mech Eng 18:6588
R e Ru R e Ru R e Ru
2 T 2 T 2 T Kwak HG, Kim DY, Lee HW (2001) Effect of warping in geomet-
l l l
ey0 ey0 ey0
 ly,n GS
,nr  ly,m GS
,nr  lz,n GS
,nr ric nonlinear analysis of spatial beams. J Construct Steel Res
2 Rl T 2 Rl T 2 Rl T 57:729751

= ely,n uGS y0
R e R e R e
e u y0 e u y0 Lee K (1997) Analysis of large displacements and large rotations
 2 R T ,mr  ly,m ,mr  lz,n2 TGS

R e Ru R e Ru
2 T of three-dimensional beams by using small strains and unit
l l l
R e u ez0 ez0 ez0 vectors. Commun Numer Methods Eng 13:987997
ly,n GS ,nr ly,m GS ,nr lz,n GS ,nr
  Li ZX, Izzuddin BA (2005) Application of vectorial rotational
where, if l = 1, then uGS = e1y,n , e1y,m , e1z,n T , i1 : variables in large displacement analysis of structures. In: Shen
ZY, Li GQ, Chan SL (eds) The fourth international conference
i3 = 4 : 6, j1 : j3 = 4 : 6 and k1: k3 = 4 : 6; if l = 2, on advances in steel structures. Elsevier, Oxford, pp 15011506
then uGS = e2y,n , e2y,m , e2z,n T , i1 : i3 = 10 : 12, Lin WY, Hsiao KM (2001) Co-rotational formulation for geomet-
j1 : j3 = 10 : 12 and k1 : k3 = 10 : 12. ric nonlinear analysis of doubly symmetric thin-walled beams.
Comput Methods Appl Mech Eng 190:60236052
Meek JL, Tan HS (1984) Geometrically nonlinear analysis of space
frames by an incremental iterative technique. Comput Meth-
References ods Appl Mech Eng 47:261282
Misra A, Behdinan K, Cleghorn WL (2000) Implementation of
Bathe KJ, Bolourchi S (1979) Large displacement analysis of consistent updated Lagrangian formulation for static analysis
three-dimensional beam structures. Int J Numer Methods Eng of geometrically nonlinear beam. Trans Can Soc Mech Eng
14:961986 24:135142
Battini JM, Pacoste C (2002) Co-rotational beam elements with Pai PF, Anderson TJ, Wheater EA (2000) Large-deformation
warping effects in instability problems. Comput Methods Appl tests and total-Lagrangian finite-element analyses of flexible
Mech Eng 191:17551789 beams. Int J Solids Struct 37:29512980
Cardona A, Geradin M (1988) Beam finite element non-linear the- Schulz M, Filippou FC (2001) Non-linear spatial timoshenko beam
ory with finite rotations. Int J Numer Methods Eng 26:2403 element with curvature interpolation. Int J Numer Methods
2438 Eng 50:761785
Chen H, Blandford GE (1991) Thin-walled space frames. I. Large- Simo JC (1992) (Symmetric) Hessian for geometrically nonlinear
deformation analysis theory. J Struct Eng ASCE 117:2499 models in solid mechanics. Intrinsic definition and geometric
2520 interpretation. Comput Methods Appl Mech Eng 96:189200
Crisfield MA (1990) Consistent co-rotational formulation for non- Simo JC, Vu-Quoc L (1986a) On the dynamics of flexible beams
linear, three-dimensional, beam-elements. Comput Methods under large overall motions the plane case: part I. J Appl
Appl Mech Eng 81:131150 Mech ASME 53:849854
322 Comput Mech (2007) 39:309322

Simo JC, Vu-Quoc L (1986b) On the dynamics of flexible beams Vu-Quoc L, Deng H (1995) Galerkin projection for geometrically
under large overall motions the plane case: part II. J Appl exact sandwich beams allowing for ply drop-off. J Appl Mech
Mech ASME 53:855863 ASME 62:479488
Simo JC, Vu-Quoc L (1986c) A three-dimensional finite-strain rod Vu-Quoc L, Ebcioglu IK (1995) Dynamic formulation for geo-
model. Part II: computational aspects. Comput Methods Appl metrically-exact sandwich beam and 1-plates. J Appl Mech
Mech Eng 58:79116 ASME 62:756763
Simo JC, Vu-Quoc L (1988) On the dynamics in space of rods Vu-Quoc L, Ebcioglu IK (1996) General multilayer geometri-
undergoing large motions a geometrically exact approach. cally-exact beams and 1-D plates with piecewise linear sec-
Comput Methods Appl Mech Eng 66:125161 tion deformation. Zeitschrift fuer Angewandte Math Mech
Simo JC, Vu-Quoc L (1991) A geometrically-exact rod model 76:391409
incorporating shear and torsion-warping deformation. Int J Vu-Quoc L, Simo JC (1987) Dynamics of earth-orbiting flexile
Solids Struct 27:371393 satellites with multibody components. J Guid Control Dyn
Teh LH, Clarke MJ (1998) Co-rotational and Lagrangian formu- 10:549558
lations for elastic three-dimensional beam finite elements. J Wen RK, Rahimzadeh J (1983) Nonlinear elastic frame analysis
Construct Steel Res 48:123144 by finite element. J Struct Eng ASCE 109:19511971
Teh LH, Clarke MJ (1999) Symmetry of tangent stiffness matrices Yang YB, Shieh MS (1990) Solution method for nonlinear prob-
of 3D elastic frame. J Eng Mech ASCE 125:248251 lems with multiple critical points. AIAA J 28:21102116
Urthaler Y, Reddy JN (2005) A corotational finite element for-
mulation for the analysis of planar beams. Commun Numer
Methods Eng 21:553570