Вы находитесь на странице: 1из 14

International Journal of Solids and Structures 48 (2011) 27642777

Contents lists available at ScienceDirect

International Journal of Solids and Structures

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

Three kinematic representations for modeling of highly exible beams

and their applications
P. Frank Pai
Department of Mechanical and Aerospace Engineering, University of Missouri, Columbia, MO 65211, USA

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

Article history: Presented here are three kinematic representations of large rotations for accurate modeling of highly
Received 3 February 2010 exible beam-like structures undergoing arbitrarily large three-dimensional elastic deformation and/or
Received in revised form 28 April 2011 rigid-body motion. Different methods of modeling torsional deformation result in different beam theories
Available online 12 June 2011
with different mathematical characteristics. Each of these three geometrically exact beam theories fully
accounts for geometric nonlinearities and initial curvatures by using Jaumann strains, exact coordinate
Keywords: transformations, and orthogonal virtual rotations. The derivations are presented in detail, a nite element
Geometrically exact beam theories
formulation is included, fully nonlinear governing equations and boundary conditions are presented, and
Highly exible beams and cables
Jaumann strains
the corresponding form for numerically exact analysis using multiple shooting methods is also derived.
Nonlinear structural mechanics These theories are compared in terms of their appropriate application areas, possible singular problems,
and easiness for use in modeling and analysis of multibody systems. Nonlinear nite element analysis of
a rotating beam and nonlinear multiple shooting analysis of a torsional bar are performed to demonstrate
the capability and accuracy of these beam theories.
2011 Elsevier Ltd. All rights reserved.

1. Introduction without skin incisions and perform common surgical procedures

and retraction. Such surgical tools are required to be small in diam-
Highly exible beams are used in many mechanical, civil, aero- eter and highly exible in order to follow and pass through the gul-
space, and architectural systems (Gantes, 2001; Pai, 2007), such as let, intestine, natural orices and/or endoscope channels easily and
helicopter rotor blades, wings of high-altitude long-endurance safely but do not undergo kinking, permanent deformation, or
aircrafts, aviation propeller blades, wind-turbine blades, robot breaking during surgery.
manipulators, slender space structures for buildings, arm-type Even in molecular biology, nonlinear beam theories are needed
positioning mechanisms of magnetic disk drives, and exible links (Yang et al., 1993; Shi and Hearst, 1994; Schlick, 1995). The super-
for high-speed slider-crank mechanisms. Moreover, NASAs various coiling (or writhing) of DNA is known to affect every physical,
science missions have extensively used deployable/inatable chemical, and biological property of a molecule. Because super-
structures consisting of highly exible beams in order to reduce coiled DNA is an important functional state active in the processes
the stowed volume and weight during launch, minimize extra of replication, transcription, and recombination, it is important to
vehicular activities in space, and/or reduce the operation time model and predict structural properties of DNA in higher-order
and cost (Pai, 2007; Jenkins, 2001). Furthermore, although cables forms (supercoils, knots, catenanes, and protein-DNA complexes)
are beams with very small bending rigidity, study of loops and in order to understand DNAs fundamental functions, including
kinks of cables requires the use of a nonlinear beam model that strand unwinding (replication, transcription) and passage (knot-
can account for dramatic geometric nonlinearities. ting and catenation), looping, and slithering. For modeling and
Todays surgical and medical treatments for many diseases analysis, a double-helical DNA polymer can be modeled as a very
depend on the use of highly exible beam- or cable-like medical thin beam with initial curvatures, and nonlinear buckling analysis
devices and tools. For example, natural orice translumenal endo- is the main task for understanding DNAs fundamental functions.
scopic surgery (NOTES) has recently emerged as a favorable surgi- This has stimulated many theoretical, computational, and experi-
cal technology under worldwide research and development (Sporn mental studies of mechanics and stability of thin beams, including
et al., 2008; Miedema et al., 2008). NOTES uses exible endoscopic the controlled buckling of elastic beams and the ingenious manip-
devices to access body cavities through the mouth, vagina, or anus ulations of DNA strands (Schlick, 1995).
Hence, to advance theoretical structural mechanics for todays
Tel.: +1 573 884 1474; fax: +1 573 884 5090. science and engineering applications, it is important to derive a
E-mail address: paip@missouri.edu geometrically exact beam theory that can be used to investigate

0020-7683/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777 2765

inuences of high-order geometric nonlinearities on static and dy- as shown in Fig. 1a. The abc is a xed global rectangular coordinate
namic characteristics of highly exible beams and cables. However, system used for reference, the xyz is a xed local orthogonal curvi-
difculties exist. linear coordinate system used to describe the undeformed beam
Beam theories are one-dimensional (1D) mathematical models geometry, and the ngf is a moving local orthogonal curvilinear
for structures with one dimension being much larger than the coordinate system used to describe the deformed beam geometry.
other two. Modeling of beams is more difcult than modeling of Moreover, u, v, and w represent the absolute displacements of the
plates and shells because the former transforms a 3D problem in observed reference point O with respect to the x, y, and z axes,
nature into a 1D problem whereas the latter transforms a 3D prob- respectively, and s denotes the undeformed arc length along the
lem into a 2D problem. Beam theories can be divided into three reference line starting from the beam root. Because u, v, and w
groups of different complexity: (a) EulerBernoulli beam theory, are continuous functions of time and s, v0 (@v/@s), w0 , and u0 exist
(b) shear-deformable beam theories (i.e., Timoshenkos theory, and they can exactly describe the reference lines two bending cur-
third-order shear theory, higher-order shear theory, layer-wise vatures under any magnitude of rotations (Pai, 2007). However, to
shear theory, etc.), and (c) 3D beam theories (Pai, 2007). In the Eu- determine the two bending curvatures still requires the use of one
lerBernoulli beam theory, only the axial stress r11 is considered or two rotations in the modeling process although Euler angles do
and a plane cross section perpendicular to the reference axis before not explicitly appear in the nal governing equations. Moreover, a
deformation is assumed to remain plane and perpendicular to the torsional Euler angle / (see, e.g., Fig. 1a) is needed in order to de-
deformed reference axis after deformation. In shear-deformable scribe the rotation of the cross section with respect to the de-
beam theories, transverse shear stresses r12 and r13 and out-of- formed reference line. Unfortunately, nite rotations are not
plane warpings are taken into account. In 3D beam theories, both commutative, different rotation sequences of these two or three
out-of-plane and in-plane warpings are taken into account. In- angles result in three different geometrically exact beam theories,
plane warpings and hence transverse normal strains e22 and e33 and each theory has its own mathematic characteristics and appro-
and in-plane shear strain e23 may result in signicant stresses priate application areas.
r22, r33, and r23. Hence, all six stresses are accounted for in a 3D Mechanical systems are multibody systems. A multibody sys-
beam theory. tem consists of interconnected rigid and deformable components,
For composite and built-up beams, some non-classical effects and each component may undergo large translations and rotations
may become signicant due to material anisotropy, asymmetry (Shabana, 2005; Bauchau, 2010; Ibrahimbegovic et al., 2003). Mod-
of the cross section, and/or different Poissons ratios over the cross eling and analysis of a exible component that undergoes large
section. These effects include transverse shear deformation, tor- rotations is very challenging because geometric nonlinearities exist
sional warping, inplane warpings due to bending and extension, and they couple with the nonlinear ordinary differential equations
transverse normal stresses, in-plane shear stresses, warping re- that govern the motions of rigid components. Hence, nite element
straints at two ends, and free-edge effects, and they cause signi- techniques are often used in the modeling and analysis of exible
cant in-plane and/or out-of-plane warping displacements and a multibody systems. Even with the use of nite elements, however,
3D stress state. Hence, 3D nite element modeling may be the only many challenging problems still exist, and the most serious chal-
way to solve dynamic problems of general composite beams, lenge is how to accurately describe the large rotations of exible
which is too expensive in order to achieve certain accuracy.
Because in-plane and out-of-plane warpings are relative displace-
ments with respect to the deformed cross section and are much
smaller than global displacements, inertial forces caused by war-
pings are negligible. However, because these warpings offer extra
degrees of freedom for the cross-section to deform, they signi-
cantly affect the global stiffnesses of a beam. Starting from 3D elas-
ticity and using a perturbation analysis with the slenderness ratio
as the ordering parameter, Parker (1979), Parker (1979) showed
that a combination of St. Venants warping solutions derived from
linear elasticity and a 1D nonlinear beam model is natural and can
account for 3D stress effects. More specically, Berdichevskii
(1981) stated that the geometrically nonlinear problem of elastic
beams can be decoupled into a nonlinear 1D problem and a linear
2D sectional problem. Consequently, one can neglect inertia forces
due to in-plane and even out-of-plane warpings and only consider
warpings in the calculation of elastic constants of beams. In other
words, a 1D nonlinear beam model with global stiffnesses deter-
mined from a static 2D sectional analysis of warpings is a general
and practical approach in solving nonlinear anisotropic beam prob-
lems (Borri and Merlini, 1986; Hodges, 2006).
Hence the two major tasks in modeling highly exible beams
are: (1) how to describe the cross section deformation, and (2)
how to describe the reference-line deformation. For the derivation
of warping functions that describe the cross section deformation,
the reader is referred to Refs. Borri and Merlini (1986), Borri and
Merlini (1986), Hodges (2006). Here we present three geometri-
cally exact total-Lagrangian beam theories that exactly describe
the reference-line deformation under any order of elastic displace-
ments and rotations.
An initially curved beam undergoing large rigid-elastic defor- Fig. 1. An elastic beam undergoing large rigid-elastic deformation: (a) three
mation requires three coordinate systems to describe its motion, coordinate systems, and (b) displacement components.
2766 P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777

2 0 0 0 3 2 3
components. One way to release this complexity is to derive and n o ix  ix ix  iy ix  iz 0 k3 k2
use total-Lagrangian structural theories. Because the strain-dis- 0 6 0 0 0 7 6 7
ixyz kfixyz g; k  4 iy  ix iy  iy iy  iz 5  4 k3 0 k1 5
placement relations of a total-Lagrangian beam theory account 0 0 0
iz  ix iz  iy iz  iz k2 k1 0
for rigid-body effects, there is no need of separate consideration
of the rigid-body movement of the rigid component that the beam 2c
is connected to. However, the most serious difculty is to derive
objective strains using the absolute displacements u, v, and w where
and the not-well-dened torsional variable / because they contain X
3 X
0 0
rigid-body effects (Pai, 2007). k1  iy  iz T o0 o
2i T 3i ; k2  ix  iz  T o0 o
1i T 3i ;
Here we introduces the use of local displacement measures, Jau- i1 i1
mann strains, exact coordinate transformation, and orthogonal vir- 0

tual rotations to derive nonlinear beam theories that fully account k3  ix  iy T o0 o

1i T 2i 2d
for large rotations, large displacements, initial curvatures, and
extensionality. This work is to derive three possible geometrically
exact total-Lagrangian beam theories, investigate their mathemat- The k1, k2, and k3 are initial curvatures with respect to axes x, y, and
ical characteristics, and compare and point out inappropriate z, respectively, and they can be calculated using Eq. (2d) and the
application areas of each of these three theories. known functions A(s), B(s), C(s), h21(s), h22(s), and h23 (s). Eqs.
(2a)(2d) completely describe the xyz system, the undeformed
2. Derivation of three geometrically exact beam theories beam geometry, and initial curvatures.
The n axis is fully related to the xyz system by displacements u,
In Fig. 1a, the xed local orthogonal curvilinear coordinate sys- v, and w, as shown next. In Fig. 1b, if OP ds, we have
tem xyz is used to describe the undeformed beam geometry and * * * *

initial curvatures, and the moving local orthogonal curvilinear OP 1 edsi1 OP PP  OO dsix uix v iy wiz uix
coordinate system ngf is used to describe the deformed beam v iy wiz 0 ds  uix v iy wiz  dsix uix v iy
geometry and deformed curvatures. Let ia, ib, and ic be the unit vec-
wiz 0 ds 3
tors of the abc system; ix, iy, and iz be the unit vectors of the xyz sys-
tem; i1, i2, and i3 be the unit vectors of the ngf system; and s be the
where e denotes the axial strain on the n axis. It follows from Eqs.
unformed arc length along the reference line starting from the
(3) and (2c) that
beam root. It is for sure that the Lagrangian coordinate s is more
appropriate and convenient than Eulerian coordinates for such OP ix u0 ix v 0 iy w0 iz ui0x v i0y wi0z
i1  T 11 ix T 12 iy T 13 iz 1eds 1e
elastic problems because an elastic beam springs back to its unde-
v k3 wk2
; T 12 v uk1e 2 v k1
0 0 0
formed geometry when it is unloaded. Fig. 1b shows that the dis- T 11 1u 1e 3 wk1
; T 13 w uk 1e
placement vector D of an arbitrary point on the observed cross q
section consists of a rigid-body motion that moves a rectangle of e 1 u0  v k3 wk2 2 v 0 uk3  wk1 2 w0  uk2 v k1 2  1
side lengths y and z on the yz plane to that on the gf plane and a 4a
small local relative displacement vector u with respect to the gf
plane. u accounts for out-of-plane shear and torsional warping where the expression of e is derived from the fact ji1 j
deformations and in-plane warping mainly caused by Poissons q
T 211 T 212 T 213 1. The deformed system n gf can be related to
The undeformed position vector R of any reference point O is as- the undeformed system xyz as
sumed to be known and given by 2 3 8 9
T 11 T 12 T 13 < i1 >
> =
R Asia Bsib Csic 1a 6 7
fi123 g Tfixyz g; T  4 T 21 T 22 T 23 5; fi123 g  i2
: > ;
Then, the unit vector ix can be dened as T 31 T 32 T 33 i3
ix  R0 A0 ia B0 ib C 0 ic 1b 4b
where ()0  d()/ds. Moreover, the angles between the y axis and the
a, b, and c axes are h21(s), h22(s), and h23(s) and are assumed to be However, because u,v, and w only describe the translation from
known. Hence, we have point O to point O, one or two rotations and one torsional Euler an-
iy cos h21 ia cos h22 ib cos h23 ic 1c gle are needed in order to determine the T2i and T3i (i = 1, 2, 3) in [T].
Three different choices for the number and sequence of these rota-
Using Eqs. (1b) and (1c) and the identity iz = ix  iy, we obtain a tions result in different beam theories because their [T] matrices
transformation matrix [To] that relates the coordinate systems abc have different singularity problems and their torsional variables
and xyz as and curvatures are in different forms. Each of these three geometri-
8 9 8 9
< ix >
> = < ia >
> = cally exact beam theories is suitable for modeling and analysis of
o different types of beam-like structures, as shown later in Sections
fixyz g T fiabc g; fixyz g  iy ; fiabc g  ib 2a
: > ; >
: > ; 2.1, 2.2, 2.3, 2.4.
iz ic
After T2i and T3i are derived, one can take spatial derivatives of
2 0
3 Eq. (4b) and uses Eq. (2c) and the identity [T]1 = [T]T to obtain
A B0 C
6 7 0 
T o  4 cos h21 cos h22 cos h23 5 i123 Kfi123 g
0 0 2 0 3 2 3
B0 cos h23  C 0 cos h22 C 0 cos h21  A cos h23 A cos h22  B0 cos h21 0 0
i1  i1 i1  i2 i1  i3 0 q3 q2
6 0 7 6
2b 0 0
K  4 i2  i1 i2  i2 i2  i3 5 4 q3 0 q1 7 0 T
5 T T TkT

0 0 0
Moreover, one can use Eq. (2a) and the orthonormality of ix, iy, and i3  i1 i3  i2 i3  i3 q2 q1 0
iz to obtain 4c
P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777 2767

8 9 8 0 9 8 9
  < de
> >
= < du >
> = < du >
> =
q1  i02  i3 T 02i T 3i T 1i ki
1 edh3 T dv 0  Tk dv 6c
i1 >
: >
; >
: 0; > >
: >
 0  1 edh2 dw dw
q2  i01  i3 T 1i T 3i T 2i ki
One can use Eq. (4d) and the Kirchhoff kinetic analogy between cur-
1  vatures and angular velocities to derive the angular velocity vector
0 0
T 31 u0  v k3 wk2 T 32 v 0 uk3  wk1 x of the ng1 system as (Pai, 2007)
X 3
x  h_ 1 i1 h_ 2 i2 h_ 3 i3 ; x_ h1 i1 h2 i2 h3 i3
T 33 w0  uk2 v k1 T 2i ki 4d
i1 P 3 P 3 P 3
h_ 1  i_2  i3 T_ 2i T 3i ; h_ 2  i_3  i1 T_ 3i T 1i ; h_ 3  i_1  i2 T_ 1i T 2i
  i1 i1 i1
q3  i01  i2 T 01i T 2i T 3i ki
1  where h_ 1  dh1 =dt, and i_k x  ik and x  x = 0 are used. Because
T 21 u0  v k3 wk2 0 T 22 v 0 uk3  wk1 0
1e the exact function forms of hi are unknown, they are also called
 X 3 quasi-coordinates (i.e., not well dened). But, dhi and h_ i have exact
T 23 w0  uk2 v k1 0 T 3i ki function forms, as shown in Eqs. (6b) and (6d).
Because the local displacement vector u in Fig. 1b is due to in-
Here, q1 is the deformed twisting curvature w.r.t. the n axis, and q2 plane and out-of-plane warpings, which are negligibly small for
and q3 are the deformed bending curvatures with respect to the g the calculation of kinetic energy and will be neglected here. With-
and f axes, respectively. Eqs. (4a)(4d) completely describe the out u, the displacement vector D in Fig. 1b and its time derivatives
ngf system, the deformed beam geometry, and deformed curva- and variation are given by
tures. However, the explicit forms of T2i and T3i (i = 1, 2, 3) and q1
D uix v iy wiz yi2 zi3  yiy  ziz
need to be derived later.
After the undeformed and deformed beam geometries are fully _ ui
D _ x v_ iy wi
_ z x  yi2 zi3
described, a geometrically exact beam theory can be derived by u
D ix v iy wi
_  yi2 zi3 x  x  yi2 zi3 
using the extended Hamilton principle, i.e.,
dD duix dv iy dwiz ydh1 i3  dh3 i1 zdh2 i1  dh1 i2
Z t
0 dK e  dP dW nc dt 5
Hence, we have
where t is the time, Ke the kinetic energy, P the elastic energy, and Z t Z t Z L Z
Wnc the non-conservative work due to external loads. Virtual rota- dK e dt _  D=2dAdsdt
dqD _
tions dhi w.r.t. the axes n, g, and f need to be derived in order to de- 0 0 0 A
Z t Z L Z
rive the variations dKe, dP , and dWnc. Virtual rotations dhi result in
variations of ik as
qD_  dDdAdsdt
0 0 A
8 9 2 38 9 Z t Z L Z Z t Z L
< di1 >
= 0 dh3 < i1 >
dh2 > =   dDdAdsdt 
qD du mv dv
6 7 0 0 A 0 0
di2 4 dh3 0 dh1 5 i2 6a
: >
; >
: > ;
mwdw Ah1 dh1 Ah2 dh2 Ah3 dh3 dsdt 7b
di3 dh2 dh1 0 i3
where A is the cross-sectional area, L is the beam length, and
Hence, it follows from Eqs. (4a), (6a), and (4b) that
8 9 8 9 8 9 8 9
_ _ 2 2
de T 11 d1 u0  v k3 wk2 T 12 dv 0 uk3  wk1 < Ah 1 >
> = >
< j1 h1  j2  j3 h2 h3 > = < j1 >
> = Z <y z >
> =
T 13 dw0  uk2 v k1 Ah2  j2 h2  j3  j1 h_ 3 h_ 1 ; j2  z2 qdA;
: >
; >
: >
; >
: > ; A >
: 2 >
3 Ah 3 j3 h3  j1  j2 h_ 1 h_ 2 j3 y
dh1 di2  i3 T 3i dT 2i Z
i1 m qdA 7c
T 33 k2  T 32 k3
dh2 di1  i3  T 3i dT 1i du Here the axes y and z are assumed to be the principal axes of the
1e R
i1 cross section and hence A qfx; y; xygdA f0; 0; 0g. Note that the ro-
T 31 k3  T 33 k1 T 32 k1  T 31 k2 tary inertias jk are often small, especially for exible beams. If rotary
dv dw
1e 1e 6b inertias jk are neglected, there are no nonlinear inertial terms. How-
T 31 T 32 T 33 ever, spinning beams, for example, can have signicant jk hk .
 du0  dv 0  dw0
1e 1e 1e Because Jaumann strains are pointwise co-rotated engineering
X 3
T 22 k3  T 23 k2 strains without inuences of rigid-body movement (Pai, 2007),
dh3 di1  i2 T 2i dT 1i du fully nonlinear Jaumann strains can be derived using the concept
of local displacements relative to the deformed coordinate system
T 23 k1  T 21 k3 T 21 k2  T 22 k1 ngf (Pai, 2007; Nayfeh and Pai, 2004). If out-of-plane and in-plane
dv dw
1e 1e warpings and hence u are neglected, Jaumann strains eij can be de-
T 21 T 22 T 23 rived to be (Pai, 2007)
du0 dv 0 dw0
1e 1e 1e 8 9
8 9 2 3 > e >
> >
Since T2i are not known yet, the explicit form of dh1 needs to be < e11 >
> = 1 0 z y >
<q  k > =
6 7 1 1
derived later. In linear EulerBernoulli beam theories without initial feg Sfwg; feg  e12 ; S  4 0 z 0 0 5; fwg 
: > >
> q2  k 2 >
curvatures, e = u0 , h2 = w0 , and h3 = v0 . In geometrically exact e13 ; 0 y 0 0 >
initially curved beam theories, Eq. (6b) show that de, dh3, and dh2
q3  k 3
are exactly and explicitly related to du0 , dv0 , and dw0 in the form 8a
2768 P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777

In the EulerBernoulli beam theory, only out-of-plane shear war- Here, F1 and Mi are stress resultants, and c1 is a correction factor
pings are neglected, and in-plane warpings are included by using accounting for the decrease of torsional rigidity due to out-of-plane
Jaumann stresses r22 = r33 = 0 to account for Poissons effect in torsional warping and can be calculated using the theory of elastic-
the constitutive equation. Note that the u in Fig. 1b does not include ity (Timoshenko and Goodier, 1970). c1 = 1 for a circular cross sec-
the out-of-plane deformation due to longitudinal stretching be- tion, and c1 < 1 for non-circular cross sections (Timoshenko and
cause Point O represents the deformed location of the undeformed Goodier, 1970). For composite materials, the inuences of non-uni-
Point O. Here we use the simple EulerBernoulli beam theory in or- form material distribution over the cross section on the [D] matrix
der to clearly demonstrate the derivation process and the main need to be calculated using static sectional analysis (Borri and Mer-
characteristics of geometrically exact beam theories without com- lini, 1986; Hodges, 2006).
plex mathematics and notations. For fully nonlinear strain-displace- It follows from Eqs. (4c) and (6a) and the identities
0 0
ment relations and inertial terms that include the inuences of u, i2  i2 i3  i3 0 that q1 is related to dhi as
the derivation steps are essentially the same and the reader is re- Z L Z L Z L
0 0 0
ferred to Ref. Pai (2007) for details. Hdq1 ds Hdi2  i3 ds Hi2  di3 i3  di2 ds
With the use of r22 = r33 = 0 and neglecting r23, Jaumann 0 0 0
strains are related to their work-conjugate stresses, i.e., Jaumann 0 0
Hi2  di3  Hi3 H0 i3  di2 ds Hi3  di2 jL0
stresses rij, as 0
8 9 2 3 Z L
< r11 >
> = E 0 0
Hq3 dh2 Hq2 dh3  H0 dh1 ds Hdh1 jL0
6 7
frg Q feg; frg  r12 ; Q  4 0 G 0 5 8b 0
: >
; Z L
r13 0 0 G
Hq3 dh2 Hq2 dh3 Hdh1 0 ds 9a
where E is Youngs modulus and G is the shear modulus. Again, sim-
ple isotropic materials are assumed here for illustration purpose. where H is an arbitrary function of s. Hence, by similar derivations
For anisotropic materials, the reduced material property matrix we obtain
[Q] is a full matrix (Pai, 2007; Hodges, 2006). The [Q] can be deter- 8 9 8 0
9 8 90 8 9
mined by experiments using small engineering strain and stress < dq1 >
> < dh1  q3 dh2 q2 dh3 >
= > < dh1 >
= > = < dh1 >
> =
measures because Jaumann strains are co-rotated engineering dq2 dh2 0 q3 dh1  q1 dh3 dh2  K dh2
: > >
; : > >
; : >
; >
: >
strains (Pai, 2007). On the other hand, if Green-Lagrange strains dq3 dh3 0  q2 dh1 q1 dh2 dh3 dh3
are used, the [Q] needs to be determined by experiments using sec- 9b
ond PiolaKirchhoff stresses and Green-Lagrange strains, which is
not usually done in experiments and is nonlinear and hence compu- If q1, q2, and q3 are distributed forces along the x, y, and z directions
tationally awkward. and q4 is the distributed torsional load along the n axis, we have
Using Eqs. (8a) and (8b) we obtain Z L
Z L Z Z L dW nc q1 du q2 dv q3 dw q4 dh1 ds 10
dP fdegT frgdAds fdwgT Dfwgds 0

0 A 0
Z Substituting Eqs. (7b), (8c), (9b), and (10) into Eq. (5) and integrat-
F 1 de M1 dq1 M 2 dq2 M 3 dq3 ds 8c ing by parts yields
0 2 3
 q3 gfdu; dv ; dwgT
 q1 ; mv  q2 ; mw
F 1 de fmu
where Z 6 7
L 6 M 01 M 3 q2  M 2 q3  Ah1 q4 dh1 7
2 3 8 9 0 6 7ds
EA 0 0 0 > F1 > 6 M 0  M q M q  A dh 7
Z >
> >
> 0 4 2 3 1 1 3 h2 2 5
60 GI1 0 0 7 < M1 =
D  ST Q SdA 6
7; M03  M 2 q1 M 1 q2 Ah3 dh3
4 0 0 EI2 0 5 >
> M2 >
: >
; M1 dh1 M2 dh2 M 3 dh3 L0 11a
0 0 0 EI3 M3
8 9 Next we dene the internal transverse shear forces F2 and F3 as
> r11 >
Z >>
<r y  r z>
13 12 1  
 dA DfwgfI1 ; I2 ; I3 g F2  M03  M2 q1 M 1 q2 Ah3 ;
A >
> r11 z >
> 1e
: ; 1  0 
r11 y F3  M 2  M 3 q1 M 1 q3  Ah2 11b
Z 1e
 fy2 z2 c1 ; z2 ; y2 gdA 8d
A Replacing the de, dh2, and dh3 in Eq. (11a) with Eq. (6c), using Eq.
(11b), and taking integration by parts yields

Z " #
L fF 1 ; F 2 ; F 3 gTfdu0 ; dv 0 ; dw0 gT  kfdu; dv ; dwgT
0 ds
0 fmu  q3 gfdu; dv ; dwgT  M 01 M3 q2  M2 q3  Ah1 q4 dh1
 q1 ; mv  q2 ; mw

M 1 dh1 M2 dh2 M 3 dh3 L0

Z L " fF ; F ; F gT0 fF ; F ; F gTkfdu; dv ; dwgT #
1 2 3 1 2 3
0 fmu  q3 gfdu; dv ; dwgT  M 01 M3 q2  M2 q3  Ah1 q4 dh1
 q1 ; mv  q2 ; mw

M 1 dh1 M2 dh2 M 3 dh3 fF 1 ; F 2 ; F 3 gTfdu; dv ; dwgT L0

P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777 2769

Setting the coefcients of du, dv, dw, and dh1 to zero yields the fol-
lowing governing equations:
0 8 91 8 9 8 9 8 9
< F1 >
> = < F1 >
> < q1 >
= > = > >
< mu =
@ B C
@s @
TT F 2 A  kTT F 2 q2 mv
: > ; >
: > ; >
: > ; >
: > ; 12a
F3 F3 q3 mw
M 01 M 3 q2  M2 q3 q4 Ah1

The boundary conditions are:

du 0 or F x  F 1 T 11 F 2 T 21 F 3 T 31 const:
dv 0 or F y  F 1 T 12 F 2 T 22 F 3 T 32 const:
dw 0 or F z  F 1 T 13 F 2 T 23 F 3 T 33 const:
dh1 0 or M 1 const:
dh2 0 or M 2 const:
dh3 0 or M 3 const:

The actual implications of boundary conditions dhi = 0 will be ex-

plained later in Section 2.4.
The governing Eqs. (12a) and (11b) can also be derived using a
vector approach based on Newtons second law and the free-body Fig. 2. The rst beam theory: (a) two coordinate systems, and (b) kinematic
diagram of a differential beam element (Pai, 2007). This shows that relations between three Euler angles.
the energy formulation starting from the extended Hamilton prin-
ciple (i.e., Eq. (5)) is fully correlated with the vector formulation,
and governing equations obtained from these two different ap-
proaches are essentially the same. On the other hand, if Green-La- T 11 T 012 T 13  T 011 T 12 T 13
q1 /0 T 11 k1 T 12 k2 T 13 k3
grange strains and second PiolaKirchhoff stresses are used in the 1  T 213
extended Hamilton principle, one can never show that the ob-
T 13
tained governing equations are the same as those from the vector /0 T 11 v 0 uk3  wk1 0
formulation. 1  T 213 1 e
This ends geometrically exact modeling of large rigid-elastic  T 12 u0  v k3 wk2 0  T 11 k1 T 12 k2 T 13 k3
deformation of an initially curved beam in the three-dimensional
T 13
space. Next we derive and explicitly write down the function forms dh1 d/ T 11 k3 du T 12 k3 dv
of T2i, T3i, q1, and dh1 of each geometrically exact beam theory to 1  T 213 1 e
complete the whole derivation process.  T 11 k1 T 12 k2 dw  T 12 du0 T 11 dv 0  13b

2.1. The rst geometrically exact beam theory H321 The [T] matrix reveals that the singular points of this beam theory
happen at T13 = 1.
The rst beam theory, denoted herein by H321, is most appro- If an initially straight beam (i.e., ki = 0) undergoes inextensible
priate for modeling and analysis of rotor blades and other beam- deformation (i.e., e = 0) and v0 and w0 are of order e (a small number
like structures that undergo large rigid-elastic deformations on for bookkeeping), Taylor expansions of the T in Eq. (13b) up to
the xy plane, as shown in Fig. 2a. H321 uses three consecutive rota- terms of order e3 yield
tions a3, a2, and /, as shown in Fig. 2b. The a3 rotates the axes x
2 3
1  v 02 w02 =2 v0 w0
and y to  , the a2 rotates the axes 
x and y x and z to n and z, and the 6 7
T 4 v 0  v 0 w02 =2 1  v 02 =2 0 5 13c
/ rotates the axes y and z to g and f. It follows from Fig. 2b and Eq.
(4a) that w v w =2
0 02 0
v w
0 0 02
1  w =2

iy  sin a3 ix cos a3 iy pT

12 T 11
ix p
iy Substituting Eq. (13c) into Eq. (4d) with / = 0 yields
1T 213 1T 213
13a q1 v 00 w0 ; q2 w00  w00 w02 =2; q3 v 00 v 00 v 02 =2 w00 v 0 w0
iz i1  iy  pT 11
T 13
ix  T 12 T 13
p i y 1  T 13 iz
1T 13 2
1T 13 13d
 and z, respec-
where iy and iz are the unit vectors along the axes y
tively. Hence, we obtain from Eqs. (13a), (4b), (4d), and (6b) that In Eqs. (13c) and (13d) we use the following equation that results
2 3 from Taylor expansion of Eq. (4a) with e = 0 :
1 0 0
6 7 u0 v 02 w02 =2 14
T 6
40 cos / sin / 7
5T; 0 2
Eq. (14) reveals that u is of order e .
0  sin / cos / If the three consecutive rotations are a2, a3, and u (denoted
2 3 herein by H231), we obtain
T 11 T 12 T 13
6 q q 7
6 7 2 3
6 T 12 = 1  T 213 T = 1  T 2
0 7 1 0 0
T  6 11 13 7
6 7 6 7
4 q q q 5 T 4 0 cos u sin u 5 Tb ;
2 2 2
T 11 T 13 = 1  T 13 T 12 T 13 = 1  T 13 1  T 13 0  sin u cos u
2770 P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777

2 3
T 11 T 12 T 13
6 q
6 2
1  T 212 T 12 T 13 = 1  T 212 7
Tb   6 T 11 T 12 = 1  T 12 7
4 q q 5
2 2
T 13 = 1  T 12 0 T 11 = 1  T 12
T 011 T 12 T 13  T 11 T 12 T 013
q1 u0 T 11 k1 T 12 k2 T 13 k3
1  T 212
T 12
dh1 du T 11 k2 du  T 11 k1
1  T 212 1 e
T 13 k3 dv T 13 k2 dw T 13 du0  T 11 dw0  15a

The singular points of this beam theory happen at T12 = 1. If an ini-

tially straight beam undergoes inextensible deformation, Taylor
b  in Eq. (15a) up to terms of order e3 yield
expansions of the T
2 3
1  v 02 w02 =2 v0 w0
6 7
Tb  4 v 0 v 0 w02 =2 1  v 02 =2 v 0 w0 5 15b
w  v w =2
0 02 0
0 1  w =202

Substituting Eq. (15b) into Eq. (4d) with / = 0 yields Fig. 3. The second beam theory: (a) kinematic relations between two Euler angles,
and (b) singularity problem.
q1 v 0 w00 ; q2 w00  w00 w02 =2  v 00 v 0 w0 ; q3 v 00 v 00 v 02 =2
Eqs. (13b) and (15a) show that the different rotation sequences An alternative derivation of T is shown in Ref. Pai (2007). Then, we
cause T and / to be different from T b  and u. The zeros of T
obtain from Eqs. (16c), (4b), (4d), and (6b) that
b  also reveal that / needs to be different from u because
and T 2 3
T 23 0 but Tb 32 0. Note that, without / and u, q1 can still be 1 0 0
6 7
non-zero because the two bendings v and w can result in a non-zero T 4 0 cos / sin / 5T
torsional curvature. 0  sin / cos /
However, there is no need to treat Eq. (15a) as a different beam 1  
theory because using Eq. (15a) for a beam is the same as using Eq. q1 /0 T 13 T 012  T 12 T 013 T 11 k1 T 12 k2 T 13 k3
1 T 11
(13b) for the beam with the coordinate system xyz being rotated 
w.r.t. the x axis by 90. /0 T 13 v 0 uk3  wk1 0
1 e1 T 11

2.2. The second geometrically exact beam theory Ha1 T 12 w0  uk2 v k1 0 T 11 k1 T 12 k2 T 13 k3
T 13 k3 T 12 k2 T 12 k1 T 13 k1
dh1 d/ du  dv  dw
The second beam theory, denoted herein by Ha1 is most appro- Tb 11 Tb 11 Tb 11
priate for modeling and analysis of beam-like structures that un- T 13 0 T 12
dergo weakly nonlinear elastic deformations. Ha1 uses two dv  dw0 ; Tb 11  1 e1 T 11
Tb 11 Tb 11
consecutive rotations a and /, as shown in Fig. 3a. The rotation a
is w.r.t. the n axis that is perpendicular to axes x and n. The a rotates 16d
; and z, and then the / rotates the
the axes x, y, and z to the axes n; y The only singular point of this beam theory happens at T11 = 1.
axes y and z to the axes g and f. Hence, it follows from Fig. 3a that The unique merit of this theory is that the governing equations
q expanded into polynomial forms by Taylor expansion are symmet-
cos a T 11 ; sin a 1  T 211 16a ric in v and w and hence interchangeable when the coordinate sys-
ix  i1 tem is rotated by 90 w.r.t. the x axis, as explained next. If an
in   sin wiy cos wiz ;
jix  i1 j initially straight beam (i.e., ki = 0) undergoes inextensible deforma-
T 13 T 12 tion (i.e., e = 0) and v0 and w0 are of order e (a small number for
sin w  q ; cos w  q 16b bookkeeping), Taylor expansions of the T in Eq. (16c) up to terms
1  T 11 1  T 211 of order e3 yield
2 3
where 0 6 a 6 p is assumed. Eq. (16b) reveals that in is on the yz 1  v 02 w02 =2 v0 w0
plane and has an angle w w.r.t. the z axis. Hence, the rotation of 6 7
T 4 v 0
1  v 02 =2 v 0 w0 =2 5 17a
the xyz system w.r.t. the n axis by a is equivalent to a rotation w
w 0
v w =2
0 0
1  w02 =2
w.r.t. the x axis, a rotation a w.r.t. the n axis, and then a rotation
w w.r.t. the new x axis that is perpendicular to axes y  and z. Hence, Substituting Eq. (17a) into Eq. (4d) with / = 0 yields
we have
2 32 32 3 q1 v 00 w0 =2  v 0 w00 =2; q2
1 0 0 cos a sin a 0 1 0 0
6 76 76 7 w00  w00 w02 =2  v 00 v 0 w0 =2; q3
T  4 0 cos w  sin w 54  sin a cos a 0 54 0 cos w sin w 5
v v00 00
v 02
=2 w 00
v w =2
0 0
0 sin w cos w 0 0 1 0  sin w cos w
2 3
T 11 T 12 T 13 If the coordinate system xyz is rotated by 90 w.r.t. the x axis,
6 T T 11 T 13 =1 T 11 T 12 T 13 =1 T 11 7
displacements u, v, and w become displacements u, w, and v.
4 12 5
T 13 T 12 T 13 =1 T 11 T 11 T 212 =1 T 11 Hence, the following transformation should exist:
P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777 2771

2 3
T 11 T 12 T 13
fq1 ; q2 ; q3 g ) fq1 ; q3 ; q2 g; 4 T 21 T 22 T 23 5
T 31 T 32 T 33
2 3 18
T 11 T 13 T 12
) 4 T 31 T 33 T 32 5
T 21 T 23 T 22
Each of the three exact beam theories satises Eq. (18) because they
are geometrically exact. However, their expanded nonlinear polyno-
mial equations may not satisfy Eq. (18). For example, Eqs. (17a) and
(17b) satisfy Eq. (18), and hence the governing equations of Ha1 is
symmetric in v and w. On the other hand, Eqs. (13c) and (13d) of
H321 and Eqs. (15b) and (15c) of H231 do not satisfy Eq. (18). Note
that jT ij j jT ji j exist for Eqs. (16c) and (17a) of Ha1, but do not exist
for those of H321 and H231. For weakly nonlinear beam vibration
problems, nonlinear analytical solutions can be obtained by pertur-
bation analysis, and the obtained asymptotic solutions can provide
nonlinear dynamic characteristics in functional forms for better
understanding of nonlinear dynamics (Nayfeh and Pai, 2004; Nayfeh
and Mook, 1979). However, for perturbation analysis, the nonlinear
governing equations shown in Eq. (12a) need to be expanded by Tay-
lor expansion into polynomials up to certain orders. Taylor expan-
sion of Ha1 results in governing equations symmetric in v and w,
and hence order-consistent and accurate perturbation analysis can
be performed. Hence, Ha1 is the most appropriate beam theory for
weakly nonlinear analysis of beams.
However, the bending rotation a may result in a discontinuous /
, as explained next. For the initially straight clamped-free rotating
beam on the x axis in Fig. 3b, if the beam is rotated by p and de-
Fig. 4. The third beam theory: (a) a spinning beam, (b) a differential element, and
formed on the xz plane, the beam root is rotated by a = p w.r.t. (c) kinematic relations between three Euler angles.
the axis n1 and the frame xyz is rotated to x1y1z1. On the other hand,
any other point on the deformed reference line is rotated by a < p
w.r.t. the axis n2 and the frame xyz is rotated to x2y2z2. Because
T 11 T 13 T 012  T 12 T 13 T 011
the z2 axis is downward, a torsional angle / = p is needed in order q1 T 11 /0 k1 T 12 k2 T 13 k3
1  T 213
to rotate the z2 axis upward, but the beam root has / = 0. Hence, / is !
not continuous. This makes it impossible to solve for / using any T 13 T 11 v 0 uk3  wk1 0  T 12 u0  v k3 wk2 0 cos /
methods based on the use of continuous functions. For example, 1  T 213 1 e T 11 w0  uk2 v k1 0  T 13 u0  v k3 wk2 sin /
if the nodal values of / in nite element modeling and analysis T 11
/0 T 11 k1 T 12 k2 T 13 k3
are interpolated using polynomial shape functions in s, it results 1  T 213
in an extremely high articial torsional strain that prevents the T 11 T 13
dh1 d/
blade from rotating forward. Moreover, this situation also happens 1  T 213 1  T 213 1 e
at the singular point of the theory because T11 + 1 = 0. Hence, this

T 11 k3 du T 12 k3 dv  T 11 k1 T 12 k2 dw  T 12 du0 T 11 dv 0 cos /
theory Ha1 is not appropriate for analysis of rotor blades. T 11 k2 du T 11 k1 T 13 k3 dv  T 13 k2 dw  T 13 du T 11 dw sin /
0 0

2.3. The third geometrically exact beam theory H132
Note that, if / = 0, the [T] in Eq. (19b) is equal to that in Eq. (13b).
The third beam theory, denoted herein by H132, is most appropri- If two consecutive rotations / and a are used, it follows from
ate for modeling and analysis of spinning shafts and other beam-like Fig. 4c and Fig. 3a and Eq. (16c) that
structures that undergo large rotation about the x axis (e.g., super- 2 3
T 11 T 12 T 13
coiling of DNAs), as shown in Fig. 4a. H132 uses three consecutive 6 7
rotations /, a3, and a2, as shown in Figs. 4b and c. The / rotates axes T 4 T 12 T 11 T 213 =1 T 11 T 12 T 13 =1 T 11 5
y and z to axes y and z, the a3 rotates axes x and y x and g, and the
 to  T 13 T 12 T 13 =1 T 11 T 11 T 212 =1 T 11
2 3 20a
a2 rotates axes  x and z to n and f. If we dene 1 0 0
T 12  T 12 cos / T 13 sin /; T 13  T 12 sin / T 13 cos / 19a 6 7
 4 0 cos / sin / 5
because the rotations a3 and a2 in Fig. 4c are similar to those in 0  sin / cos /
Fig. 2b, we obtain from Fig. 4c and Fig. 2b and Eqs. (19a), (4b),
However, using direct expansion and Eq. (19a) one can show that
(4d), and (6b) that
2 3 Eq. (20a) is equal to
T T 12 T 13 2 3
6 q11 q
7 1 0 0
6 T 12 = 1  T 2 T 11 = 1  T 13 2
0 7 6 7
T 6 13 7 T 4 0 cos / sin / 5
4 q q q 5
T 11 T 13 = 1  T 13 T 12 T 13 = 1  T 13 2
1  T 13 2 0  sin / cos /
2 3 2 3 20b
1 0 0 T 11 T 12 T 13
6 2 7
 4 0 cos / sin / 5  4 T 12 T 11 T 13 =1 T 11 T 12 T 13 =1 T 11 5
0  sin / cos / T 13 T 12 T 13 =1 T 11 T 11 T 12 =1 T 11
2772 P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777

Eqs. (20a) and (20b) show that, when two Euler angles are used in the displacement vector D in Eq. (7a). However, two equations gov-
the modeling, the model is independent of rotation sequence and erning c5 and c6 will be added to Eq. (12a) (Pai, 2007).
the beam theory is the same as the second beam theory Ha1. How-
ever, the problem of discontinuous / shown in Fig. 3b remains, but 3. Finite element formulation
it is not expected for spinning shafts, especially if the shafts two
ends are supported by bearings. The weak forms shown in Eqs. (7b), (8c), and (10) can be used
with Eqs. (9b) and (6b) for nite element formulation using u, v,
2.4. Discussions w, u0 , v0 , w0 , and / as nodal degrees of freedom, as shown next. It
follows from Eq. (8c) that
The singular points of beam theories H321,H231, and Ha1(=H1a) RL RL
are T13 = 1, T12 = 1, and T11 = 1, respectively, and the singular dP 0
fdwgT Dfwgds 0
fdUgT WT Dfwgds
points of beam theory H132 are T13 = 1 when cos / = 1 and fdwg  WfdU g; fUg  fu; u0 ; u00 ; v ; v 0 ; v 00 ; w; w0 ; w00 ; /; /0 gT
T12 = 1 when sin/ = 1. In programming, these singular points
can be easily bypassed by subtracting a very small number (e.g.,
sign(Tij)  1012) from Tij when Tij = 1 happen. Although this The explicit forms of Wij(@ wi/oUj) can be derived from Eqs. (6b)
may cause minor errors in the obtained displacements at the sin- and (9b). Next we use two-node beam elements to discretize the
gular points, these errors will not accumulate like those happen beam into ne elements. The displacements of the ith element are
to updated Lagrangian formulations because these are displace- discretized as
ment-based total-Lagrangian beam theories.
fdg  fu; v ; w; /gT Nfqi g 22a
For a clamped end, it follows from Eqs. (13b), (16d), and (19b)
that dh1 = d/ = 0 for each of the three beam theories because where [N] is a 4  14 matrix of shape functions and {q } is the ele-
T11 = 1 and T12 = T13 = 0. For a clamped end, dh2  3i1 T 3i dT 1i ment displacement vector of the ith element dened as
0 in Eq. (12b) is equivalent to dT13 = 0 because T33 = 1 and
T31 = T32 = 0. Similarly, dh3 3i1 T 2i dT 1i 0 is equivalent to fqi g fuj ; v j ; wj ; /j ; w0j ; v 0j ; u0j ; uk ; v k ; wk ; /k ; w0k ; v 0k ; u0k gT 22b
dT12 = 0 because T22 = 1 and T21 = T23 = 0. dT13 = 0 means that It follows from Eqs. (22a) and (21) that
T13 = constant, which is a nonlinear constraint equation involving
several variables. If the inuences of ki and e are neglected, these fUg @Nfqi g; @N  @N 22c
boundary conditions reduce to d T13 = dw0 = 0 and dT12 = dv0 = 0, where [@] is a 11  4 matrix of differential operators. Then we ob-
which are the same as linear cases. Other boundary conditions tain the variation of elastic energy as
can be similarly determined.
ne Z
In rotordynamics, the inertia-induced torsional uctuation var-
dP fdqi gT @NT WT Dfwgds
iable / is sometimes neglected (because the average spinning Li
speed is much higher than / _ and hence only the governing equa-
tions of u,v, and w (or even just v and w) are used in nonlinear
fdqi gT k fqi g fdqgT Kfqg 23a
modeling and analysis of rotor shafts (Genta, 2005; Muszynska, i1
2005). In that case, it is very important to know that beam theories
H321, H132, and H231 are not appropriate for such analysis because Z
they cannot represent beam deformations with T23 0 or T32 0. i
k fqi g @NT WT Dfwgds 23b
Hence, the beam theory H1a(=Ha1) is most appropriate for such Li
ne is the total number of elements, Li is the ith element length, [K] is
Different sequences of applying a set of loads to a nonlinear
the global stiffness matrix, and {q} is the global displacement vec-
beam may result in different deformed geometries because the ra-
tor. We note that the element stiffness matrix [k(i)] and the element
tio v/w varies differently during different loading sequences. If the
displacement vector {q(i)} are nonlinearly coupled into a vector and
deformation sequence assumed in modeling is different from the
cannot be separated in this fully nonlinear formulation.
actual loading process, the derived beam theory may be incapable
Eqs. (7b), (7c), and (6b) show that the rotary inertial moments
of solving the specic problem. For example, if / = 0 is assumed,
Ahi are nonlinear functions of u, v, w, and / and their spatial deriv-
Eq. (13b) shows that H321 cannot model deformations with
atives. However, because rotary inertias jk of highly exible beams
T23 0. Hence, it is important to include / in the modeling in order
are negligibly small, the inertial moments Ahi are negligible if the
to make [T] a full matrix. For a beam theory with a spatially discon-
vibration frequency is not too high. If inertial moments are ne-
tinuous torsional variable (e.g., Ha1) , spatial discretization of / by
glected, Eq. (7b) reduces to
nite elements is problematic because nite element techniques
are based on the assumption that dependent variables are spatially
dK e  du mv dv mwdw
mu ds
continuous. 0
Because the torsional Euler angle is different for each of the X
ne Z
three beam theories, adding damping in the same form of c/_ with  fdqi gT NT mNf
~ i gds
the same coefcient will result in different dynamic responses. i1 Li

Moreover, the large velocities due to rigid-body motions can cause X


dramatic damping effects if the concepts of proportional damping  fdqi gT mi fq

i g fdqgT Mfq
g 24a
or modal damping are used to construct the damping matrix. If
only material damping due to straining is to be accounted for, it where [M] is the global mass matrix and
_ q_ 1 ; q_ 2 ; and q_ 3 with some coef-
is better to use the straining rates e; 2 3
m 0 0 0
cients from experiments to account for the damping effect. If the Z 60 m 0 07
aerodynamic damping effect due to rigid-body motion is to be ac- mi   NT mNds;
~ ~ 6
m 6
7 24b
counted for, the damping matrix should be velocity-dependent. Li 40 0 m 05
To consider the inuences of transverse shear deformations c5 0 0 0 0
and c6, one just needs to keep the local displacement vector u in
P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777 2773

The variation of non-conservative work due to four external distrib- the accuracy of converged solutions because this is a total-Lagrang-
uted loads is obtained from Eqs. (10) and (13b) (or Eq. (16d) or Eq. ian formulation instead of an updated-Lagrangian formulation. If
(19b)) as nonlinear effects of dh1 on the nodal loading vector {R(i)} and
Z L Ahi dhi on the mass matrix [m(i)] need to be investigated, they can
dW nc q1 du q2 dv q3 dw q4 dh1 ds be similarly treated as those shown in Eqs. (27a)(27d).
fdUgT f Rgds fdqi gT fRi g fdqgT fRg 25a 4. Multiple shooting formulation
0 i1
For nonlinear static problems or steady-state dynamic prob-
Moreover, {R} is the global nodal loading vector and {R(i)} is the ele- lems, these geometrically exact beam theories can be transformed
mental nodal loading vector given by into nonlinear ordinary differential equations (ODEs) with s as the
e only independent variable (Pai, 2007). If these governing equations
fRi g  @NT f Rgds 25b
can be put into a group of nonlinear rst-order ODEs, they can be
solved for numerically exact solutions using a multiple shooting
Substituting Eqs. (23a), (24a), and (25a) into Eq. (5) yields the fol- algorithm based on direct numerical integration using RungeKut-
lowing equation of motion ta methods (Pai, 2007). Because there are six boundary conditions
g Cfqg
Mfq _ Kfqg fRg 26 at each end, this is a 12th-order system. The governing Eqs. (12a)
and (11b) can be arranged into the following 13 rst-order ODEs
where the damping matrix [C] is added and it can be obtained by (Pai, 2007):
using the concept of modal damping or proportional damping
(Ewins, 2000). F 01 q3 F 2  q2 F 3 T 11 mu
 q1 T 12 mv  q2 T 13 mw
For static problems, Eq. (26) can be solved by using an incre- F 02 q1 F 3  q3 F 1 T 21 mu
 q1 T 22 mv  q2 T 23 mw
mental/iterative method based on the modied Riks method (Pai, F 03 q2 F 1  q1 F 2 T 31 mu
 q1 T 32 mv  q2 T 33 mw
2007). For dynamic problems, Eq. (26) can be solved by direct 0
M 1 q 3 M 2  q 2 M 3  q4
numerical integration using the Newmark-b method (Pai, 2007).
If the linear form dh1 = d/ is adopted, the [M] and {R} are constant M 02 q1 M 3  q3 M 1 1 eF 3  q5
and only [K] is a function of displacements. To derive the incre- M 03 q2 M 1  q1 M 2  1 eF 2  q6
mental form of Eq. (23b) for static and dynamic analyses using T 011 q3 T 21  q2 T 31 T 12 k3  T 13 k2
any incremental/iterative methods, rst we dene T 012 q3 T 22  q2 T 32 T 13 k1  T 11 k3
fqi g fq
g fDqi g; fUg fUg fDUg 27a T 013 q3 T 23  q2 T 33 T 11 k2  T 12 k1
g denotes an equilibrium state and {Dq(i)} denotes a dis-
where fq /0 q1  q3 T 22  q2 T 32 T 13 k1  T 11 k3 T 11 T 13 =1  T 213
placement increment vector when the loads increase and/or time q3 T 21  q2 T 31 T 12 k3  T 13 k2
proceeds. Substituting Eq. (27a) into {w}, [W], and Eq. (23b) yields  T 12 T 13 =1  T 213  T 11 k1  T 12 k2  T 13 k3
fwg fwg W W H; u 1 v k3  wk2 1 eT 11

@ Wij @ wi 2 v 0 wk1  uk3 1 eT 12

Hij  DU k DU k
@U k @U j @U k w0 uk2  v k1 1 eT 13
Z 27b
i 28a
k fqi g  @NT WT Dfwg
 @NT HT Dfwg

Because e and qi are functions of F1 and Mi as shown in Eq. (8d),
@NT WT DWfDUg ds there are only 13 unknowns in Eq. (28a), i.e., F1, F2, F3, M1, M2, M3,
T11, T12, T13, /, u, v, and w. However, T11 can be determined by using
By direct expansion, one can show that q
1  v k3 wk2 Du=Ds
 ( 2 ) T 11  1  T 212  T 213 28b
T   @ Wmi  @ wm  1e
H Dfwg fHmi Dmn wn g DU j Dmn wn Dmn wn DU j
@U j @U i @U j
where the second expression obtained from Eq. (4a) is proposed for
@ 2 wm 2
 n @ wm Dmn w
 n Cji determining the sign of T11 using spatial nite difference. Hence,
 CfDUg; Cij  Dmn w
@U i @U j @U j @U i only 12 equations with 12 unknowns need to be solved with 12
27c boundary conditions, indicating a 12th-order system.
The equations of T 011 ; T 012 ; and T 013 are obtained from the alter-
Z Z native form [T]0 = [K][T]  [T][k] of Eq. (4c), and the equations for u0 ,
k fqi g @NT WT Dfwgds
 @NT C v0 , and w0 are obtained from Eq. (4a). The ninth equation of Eq.
Li Li
i ^i fDqi g 27d (28a) is obtained from Eq. (13b) of the rst beam theory H321. If
WT DW@NdsfDqi g k fqi gjfqg k Eq. (16d) of the second beam theory Ha1 is used, we have
Z /0 q1  q3 T 22  q2 T 32 T 13 k1  T 11 k3 T 13 =1 T 11
k @NT C WT DW @Nds
Li q3 T 23  q2 T 33 T 11 k2  T 12 k1 T 12 =1 T 11  T 11 k1
where k^i  is the elemental tangential stiffness matrix and is sym-  T 12 k2  T 13 k3 29
metric because [C] is symmetric. Because {w}, [W], and [C] in Eq.
If Eq. (19b) of the third beam theory H132 is used, we have
(27b) fully account for all geometric nonlinearities, the only differ-
ence in nite element formulation of the three nonlinear beam the- 1  T 213
/0 q1  T 11 k1  T 12 k2  T 13 k3 q3 T 21  q2 T 31 T 12 k3
ories is the computation of them. Since [C] requires second-order T 11
differentiations on {w}, it is more efcient to use nite difference  T 13 k2 T 12 T 13 =T 11  q3 T 22  q2 T 32 T 13 k1  T 11 k3 T 13 cos /
in programming. The use of nite difference for [C] affects the con-
vergence rate of the iteration solution process, but it does not affect  q3 T 23  q2 T 33 T 11 k2  T 12 k1 T 13 sin / 30
2774 P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777

5. Numerical results beam elements based on the rst beam theory H321 without
damping, Fig. 5a shows the time-varying displacements u and w
Two examples are used here to illustrate the capabilities of of nodes 6, 11, and 16 (i.e., at s = L/3, 2L/3, and L), and Fig. 5b shows
these geometrically exact beam theories for huge rigid-elastic the time-varying elastic energy P, kinetic energy Ke, gravitational
deformation analysis of highly exible beams. For more numerical potential energy Eg, and total energy (=P + Ke + Eg). Fig. 5b shows
examples using these beam theories, the reader is referred to Ref. that the total energy keeps at the value of the beginning kinetic en-
Pai (2007) for Ha1, Ref. Qian et al. (2010) for H132, and Ref. Wu ergy, indicating the beam theory and the nite element algorithm
et al. (2010) for H321. are energy conserved. Numerical results indicate that /(x, t) 0
due to bendings v and w, but the value of j/j at any location is less
5.1. Vibration of a rotating exible beam than 1.4 at any time, as expected. Fig. 5c shows 19 consecutive de-
formed geometries with a time interval of 0.1 s, where the beam
We consider a clamped-free titanium alloy beam on the x axis width is plotted to be 10 mm (instead of 50.8 mm) in order to have
being given an initial angular speed X = p rad/s w.r.t. the vertical a better visual effect. The gravity-induced static deection of the
z-axis, as shown in Fig. 2a with r = 0. The beam has a mass density beam tip is calculated from nite element analysis to be
q = 4430 kg/m3, Youngs modulus E = 127 GPa, and dimensions w16 = 12.6 cm and experimental result was w16 = 12.7 cm (Pai,
L  b  h = 479 mm  50.8 mm  0.45 mm. After the beam is ro- 2007). On the other hand, Figs. 5a and c show that the maximum
tated away from the x axis, gravity causes the beam to vibrate. downward deection of the beam tip is w16 = 21.9 cm because
We set u = v = w = / = w0 = 0 for the clamped end. Using 15 equal of inertial as well as gravitational loads. Moreover, Figs. 5ac show

Fig. 5. Finite element analysis of a rotating beam: (a) displacements u and w, (b) different energies, and (c) deformed geometries at different times.
P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777 2775

that the free vertical vibration frequency is about 1.86 Hz. On the E 200 GPa; m 0:32; q 7860 kg=m3
other hand, the rst three linear bending frequencies without grav- L 1:0 m; r 0:5 mm; q3 mgN=m; g 9:81 m=s2
ity are 1.6964 Hz, 10.631 Hz, and 29.769 Hz.
We also perform the same nite element analysis using the
third beam theory H132. The obtained results overlap with those where r is the radius of the circular cross-section and the distrib-
shown in Figs. 5ac. However, numerical results show that the uted load q3 is due to the gravity along the z direction. The bound-
beam tip can have /min = 3.1 but H321 has /min = 1.4, indicat- ary conditions are:
ing the torsional Euler angles of H321 and H132 are different. Be-
cause the main displacements u and v are on the xy plane, it is T 12 T 13 / u v w 0 at s 0
 at s L 31b
better to have the major rotation a3 as the rst rotation and hence F 1 T 12 T 13 v w 0; T 11 1; //
H321 is more appropriate for this problem.
On the other hand, if the second beam theory Ha1 is used, nite Using the multiple shooting formulation shown in Section 4 with 40
element simulations show that the beam cannot rotate beyond an shooting points and the second beam theory Ha1 (=H1a), we solve
 Figs. 6a-f
for static deformed geometries under different values of /.
angle of p because of the problem of discontinuous torsional angle
and the nite element interpolation, as explained using Fig. 3b in show the deformed geometries corresponding to different values of
 and their projections onto the ab, ac, and bc planes. Fig. 6a shows
Section 2.2. When the beam is rotated by about 180, / can have
a value close to p. that the beam is more like a cable with small bending rigidity be-
cause the gravity is able to make it sag by about 1.6 cm. Although
the gravity is signicant for this thin beam, the deformed geome-
5.2. Static twisting of a slender bar tries shown in Figs. 6c, d, and f are bent upward, indicating the
straining is high enough to resist the gravity. Note that the de-
We consider an initially straight xed-sliding bar being twisted formed geometry may jump under a nite increment of /,  as shown
 at s = L, as shown in Fig. 4a with the left end being
by an angle / in Figs. 6be. However, the applied right-end torsional moment M 1
xed. The bar has the following material properties and increases with / almost linearly from zero to M1 = 37.61 GI1/L when
dimensions:  12p 37:70; as shown by the reference straight line in Fig. 7.

 0; b/
Fig. 6. Deformed geometries of a xed-sliding bar under different angles of twisting: a/  3:5p; c/
 4p; d/
 4:5p; e/
 6p, and f /
2776 P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777

problem shown in Fig. 3b does not exist in this system, numerical

simulations indicate that H1a is better. For example, with the use
of H1a and an increment of D/  p=2, all the gures shown in
Fig. 6 can be sequentially obtained without difculty. On the other
hand, if H132 is used, it is almost impossible to obtain Fig. 6c by
starting from Fig. 6b, even using a very small D/. We actually ob-
tained Fig. 6e rst by trial and error and then reduced / to obtain
Figs. 6d and c. In other words, because H132 uses three rotations
and hence is sequence-dependent, difculties exist in continuous
searching for some solutions.
Similar to H132, H321 has difculties in obtaining Fig. 6c by
starting from Fig. 6b because H321 is also sequence-dependent.
For spinning shafts (see Fig. 4a), because the main motion is the
spinning, it is better to have / as the rst rotation and hence
Fig. 7. Load-deection curve. H132 and H1a are more appropriate than H321.

6. Concluding remarks
This indicates that the non-trivial transverse displacements at
points away from the two ends do not have signicant inuences Three geometrically exact beam theories for analysis of beam-
on M 1 because this is a very thin beam and T11 = 1 at s = 0, L. like structures undergoing arbitrarily large rigid-elastic deforma-
Because the problem is highly nonlinear, multiple equilibrium tions are derived. They are different in modeling torsional defor-
paths are expected. Fig. 7 shows that the main equilibrium path mation and have different mathematical characteristics. Detailed
bifurcates at Points 1, 2, and 3. To obtain the rst bifurcated branch derivations are presented, fully nonlinear governing equations
during the multiple shooting solution process, we increase /  be- and boundary conditions are listed, and formulations for nonlinear
yond Point 1, obtain a converged solution there, and then gradually nite element analysis and multiple shooting analysis are also de-
decrease / to move along the bifurcated path. Figs. 8a and b show rived. These theories are compared in terms of their appropriate
the deformed geometries corresponding to Point a and b on Fig. 7. application areas, possible singular problems, and easiness for
The same process is used to obtain the other two branches and use in modeling and analysis of multibody systems. Two numerical
Figs. 8c and d. Figs. 7 and 8 show that the rst, second, and third examples are used to demonstrate the capabilities of these three
solution branches correspond to single-, double-, and triple-loop beam theories. Comparisons show that the second beam theory
deformation geometries, respectively, and they all involve large is more versatile than the other two because it uses only two rota-
bending deformations. tions in modeling and hence is sequence-independent. However,
Because the deformation is mainly caused by a torsional load, because of the use of only two rotations, the torsional variable of
one may think that H132 is more appropriate than H1a for this the second theory can be discontinuous for some problems and
problem. However, because the left end is xed, the transverse invalidates any continuity-based methods (including the nite ele-
deformation may jump (see Figs. 6b-d), and the singularity ment method) for solving such problems.

 2p, (b) /
Fig. 8. Deformed geometries of points a, b, c and d on the bifurcated branches in Fig. 7: (a) /  0, (c) /
 4:4p, and (d) /
P.F. Pai / International Journal of Solids and Structures 48 (2011) 27642777 2777

Acknowledgement Nayfeh, A.H., Pai, P.F., 2004. Linear and nonlinear structural mechanics. Wiley-
Interscience, New York.
Pai, P.F., 2007. Highly Flexible Structures: Modeling, Computation, and
This work is supported by the National Science Foundation un- Experimentation. AIAA, Reston, Virginia.
der Grant CMMI 1039433. This support is gratefully acknowledged. Parker, D.F., 1979. An asymptotic analysis of large deections and rotations of
elastic rods. International Journal of Solids and Structures 15, 361377.
Parker, D.F., 1979. The role of saint Venants solutions in rod and beam theories.
References Journal of Applied Mechanics 46, 861866.
Qian, X., Du, X., Pai, P.F., 2010. Experimental nonlinear dynamics of a highly exible
Bauchau, O.A., 2010. Flexible Multibody Dynamics. Springer Verlag, New York. spinning beam using a 3D motion analysis system. In: 51st AIAA Structures,
Berdichevskii, V.L., 1981. On the energy of an elastic rod. Journal of Applied Structural Dynamics and Materials Conference, Orlando, Florida, April 1215.
Mathematics and Mechanics (PMM U.S.S.R.) 45, 518529. Schlick, T., 1995. Modeling superhelical DNA: recent analytical and dynamical
Borri, M., Merlini, T., 1986. A large displacement formulation for anisotropic beam approaches. Current Opinions in Structural Biology 5, 245262.
analysis. Meccanica 21, 3037. Shabana, A.A., 2005. Dynamics of Multibody Systems, 3rd ed. Cambridge University
Ewins, D.J., 2000. Modal Testing: Theory, Practice and Application, second ed. Press, New York.
Research Studies Press, New York. Shi, Y., Hearst, J.E., 1994. The Kirchhoff elastic rod, the nonlinear Schrodinger
Gantes, C.J., 2001. Deployable Structures: Analysis and Design, WIT Press, equation and DNA supercoiling. Journal of Chemical Physics 101, 51865200.
Southampton, England, UK. Sporn, E., Miedema, B.W., Bachman, S.L., Loy, T.S., Calaluce, R.D., Thaler, K., 2008.
Genta, G., 2005. Dynamics of Rotating Systems. Springer, New York. Endoscopic colotomy closure after full thickness excision: comparison of T-
Hodges, D.H., 2006. Nonlinear Composite Beam Theory. AIAA, Reston, Virginia. fastener versus multi-clip applier. Endoscopy 40 (7), 589594.
Ibrahimbegovic, A., Taylor, R.L., Lim, H., 2003. Nonlinear dynamics of exible Timoshenko, S.P., Goodier, J.N., 1970. Theory of Elasticity, 3rd ed. McGraw-Hill,
multibody systems. Computers & Structures 81, 11131132. New York.
Jenkins, C.H.M. (Ed.), 2001. Gossamer Spacecraft: Membrane and Inatable Wu, G., He, X., Pai, P.F., 2010. Geometrically exact total-Lagrangian modeling and
Structures Technology for Space Applications, AIAA, Reston, Virginia. nite element analysis of highly exible multibody systems. In: 51st AIAA
Miedema, B.W., Astudillo, J.A., Sporn, E., Thaler, K., 2008. NOTES techniques: present Structures, Structural Dynamics and Materials Conference, Orlando, Florida,
and future. Review, European Surgery 40 (3), 103110. April 1215.
Muszynska, A., 2005. Rotordynamics. Taylor & Francis, New York. Yang, Y., Tobias, I., Olson, W.K., 1993. Finite element analysis of DNA supercoiling.
Nayfeh, A.H., Mook, D.T., 1979. Nonlinear Oscillations. Wiley-Interscience, New Journal of Chemical Physics 98, 16731686.