Sunteți pe pagina 1din 20

III European Conference on Computational Mechanics

Solids, Structures and Coupled Problems in Engineering


C.A. Mota Soares et.al. (eds.)
Lisbon, Portugal, 59 June 2006

TOTAL AND UPDATED LAGRANGIAN GEOMETRICALLY EXACT


BEAM ELEMENTS
Jari Mkinen1, Heikki Marjamki2
1
Tampere University of Technology,
Applied Mechanics and Optimization
P.O. Box 589, FIN-33101 Tampere, FINLAND
e-mail:jari.m.makinen@tut.fi,
2
Tampere University of Technology,
Applied Mechanics and Optimization
P.O. Box 589, FIN-33101 Tampere, FINLAND
e-mail: heikki.marjamaki@tut.fi

Keywords: Geometrically Exact Beams, Computational Mechanics, Multibody Dynamics

Abstract. We consider a Reissners geometrically exact beam element, which is based on a


total Lagrangian updating procedure. The beam formulation has several benefits such as all
the unknown vectors belong to the same tangential vector space, no need for secondary stor-
age variables, the path-independence in the static case, any standard time-integration algo-
rithm may be used, and the symmetric stiffness tensor. The total Lagrangian formulation
preserves the path-independence and it can be regarded as a consistent updating formulation.
The major drawback is the singularities at the rotation angle 2 and its multiples, but we

overcome this difficulty by the complement rotation vector that is the change of parametriza-
tion. Numerical examples are also given
J. Mkinen and H. Marjamki

1 INTRODUCTION
In the paper [1] is given finite element implementations for a geometrically exact beam
element with different updating procedures. The formulations are named Eulerian, total La-
grangian, and updated Lagrangian. The updated Lagrangian formulation can bypass the well-
known singularity problem of the total Lagrangian formulation which is singular at the rota-
tion angle 2 and its multiples. The updated Lagrangian formulation has additional benefits

such as a fully symmetrical stiffness tensor when applying a conservative loading. In addition,
any time integration algorithm can be used because the changes of the rotation vector belong
to the same tangent space of the rotation manifold SO(3). The updated Lagrangian formula-
tion requires some secondary storage variables for the curvature and rotation vectors, at every
spatial integration point.
A total Lagrangian formulation in static cases with the consistent stiffness tensor is given
in [2]. Generally, a total Lagrangian formulation in a static case with a conservative loading
has an important property that is path-independence, whereas an updated Lagrangian formu-
lation is path-dependent. Lagrangian formulations have a consistent interpolation while in
Eulerian formulations the interpolation has to apply within an approximate, inconsistent, way.
As we have noted earlier, total Lagrangian formulations have singularity at the rotation angle
2 and its multiples that are a remarkable restriction, especially in dynamic cases.

2 ROTATION MANIFOLD
In Fig. 1 is shown a basic idea of differentiable manifold. A differentiable manifold can be
mapped from a chart in a parameter space into a chart of manifold in an embedding space.
The change of parametrization is differentiable for differentiable manifolds.

manifold n

n
embedding space
1 1 2 2
2
1 change of
parametrization
1
21 1 2

d
parameter space

Fig. 1 A geometric interpretation for a parametrization of a manifold, when n = 3 and d = 2 .

Let () be a parametrized curve in the manifold through the base point x such
that ( = 0) = x . The tangent of curve () at = 0 to the manifold is defined as,

2
J. Mkinen and H. Marjamki

() (0)
x = lim , where (0) = x, () (1)
0
The tangent vector x belongs to a tangent space of the manifold, namely x Tx , The tan-
gent (vector) space Tx is a set of tangent vectors at x .
A rotation motion can be represented by rotation operators R that form a group. This spe-
cial noncommutative Lie-group of the proper orthogonal linear transformations is defined as
SO(3): = R : 3
3
R T R = I, det R = +1 (2)

where 3 indicates three-dimensional Euclidean vector space. Rotation tensor can be repre-
sented minimally by three parameters, which parametrize rotation tensor only locally. It is
well known that there exist no a single three-parametric global presentation of rotation tensor
because the rotation group is a compact group [3]. The rotation group is also a three-
dimensional manifold (i.e. Lie-group) with differentiable structure. Euler angles are the most
extensive three-parametric presentations in the literature of analytical dynamics. However,
simpler and more useful parametrization can be obtained if the parameters are canonical, i.e.
the rotation vector parametrization.

2.1 Rotation Vector


We are trying to find an expression for the rotated vector p1 in the terms of the original vector
p 0 , the unit rotation axis n , and the non-negative rotation angle about the rotation axis,
see Fig. 2. The original projection vector r0 and the rotated projection vector r1 in the rota-
tion plane are given in Fig. 2, where denotes the cross product on 3 . Now, the rotated
vector p1 can be expressed by
p1 = p 0 r0 + r1 = p0 + 1 cos n n p0 + n p0 sin
(3)
= Rp 0 p1 , p 0 , n, r0 , r1 3
, +

Now, the rotation operator R can be written in terms of the rotation vector that is defined
by : = n, n 3 , + . This yields the expression of the rotation operator
sin ~ 1 cos ~ 2 ~
R:= I ++ = exp( ), = (4)
2

~ ~
where the skew-symmetric rotation tensor , is defined by formally : = .

r1

r0

3
p1 r1 e p 0 sin
e p0


r0

Fig. 2 A rotational motion about e -axis where p0 is the original vector and p1 is the rotated vector.

3
J. Mkinen and H. Marjamki

2.2 Compound Rotation


Compound rotation can be defined by two different, nevertheless, equivalent ways: the
material description, and the spatial description. We define the material description of a com-
pound rotation by
~
mat
RR inc = R exp( ) mat
R inc , R SO(3) (5)
mat
where R inc is an incremental material rotation operator, and is an incremental material
rotation vector with respect to the base point R SO(3) . This description is called material
since the incremental rotation operator acts on a material vector space. Respectively, we could
define the spatial description of the same compound but we neglect this for shortness.

2.3 Tangent Spaces of Rotation Manifold


~
Differentiating the expression (): = exp( ) with respect to the parameter t at t = 0
gives the tangent of the rotation tensor at the identity, yielding
~
d exp ~
= (6)
d
=0
~
Thus, the skew-symmetric tensor belongs to the tangential space of the rotation manifold,
~
notation TI SO(3) , where the identity I SO(3) represents the base point of the rotation
manifold.

~ T SO(3)
exp mat I

~

I
T SO(3)
mat R
R
~
R

SO(3)

Fig. 3 A geometric representation of the material tangent space on the rotation man ifold.

~
Differentiating the material expression of the compound rotation R exp( ) with respect
to the parameter and setting = 0 , yields the material tangent space at the base point
R SO(3) , defined by
~ ~ ~
T SO(3): = R : = ( R , ) for any so(3)
mat R (7)
~
where an element of the material tangent space R matTR SO(3) and is a skew-symmetric
~
tensor. The notation (R , ) , the pair of the rotation operator R and the skew-symmetric ten-
~
sor , represents the material skew-symmetric tensor at the base point R SO(3) , see Fig.
3.

4
J. Mkinen and H. Marjamki

Let us consider the material form of compound rotation, given in (5), with the aid of -
parametrized exponential mappings
~ ~ ~ ~
exp + = exp exp R , (8)
~
where we are finding an incremental rotation tensor, the virtual rotation tensor , such that
~
it belongs to the same tangent space as the rotation tensor , i.e. such that
~ ~
, matTI SO(3) with the identity as a base point omitted for simplicity.
Taking the derivatives of (8) with respect to the parameter at = 0 gives
R = T
sin 1 cos ~ sin
T: = I + (9)
2 3
~
: = , R = exp( ), lim T( ) = I
0

where the material tangential transformation T = T( ) is a linear mapping between the vir-
tual material tangent spaces mat TI SO(3) matTR SO(3) . Now, we could make a verification
that the virtual rotation vector R and the virtual total rotation vector belong to differ-
ent vector spaces on the manifold. This is because the tangential transformation T is equal to
the identity only at = 0 . Note that the transformation T has an effect on the base points,
changing the base point I into R.
For convenience, we define a material vector space on the rotation manifold at any point
R as
3 ~
mat TR : = R R matTR SO(3) (10)

where an element of the material vector space is R matTR , which is an affine space with the
rotation vector as a base point and the incremental rotation vector as a tangent vector,
then T: matTI matTR , see [4]. Definition (10) gives a practical notation for sorting rotation
vectors in different tangent spaces.
Material angular velocity (skew-symmetric) tensors are defined with the following formu-
las
~
R : = R T R matTR SO(3),. (11)
The material angular vector is given in terms of total rotation vector
R = T( ) , R matTR ; , matTI , (12)
A material angular acceleration tensor and vector are defined as the time derivative of corre-
sponding angular velocity terms, giving:
~ ~ ~
R : = R , R matTR SO(3)
(13)
R = R , R matTR
where A R is the material angular acceleration vector at the base point R .
Note that the material incremental rotation vector R , the material angular velocity vector
R and the material angular acceleration vector R belong to the same material vector space

5
J. Mkinen and H. Marjamki

~
on the rotation manifold, i.e. R , R , R mat TR with the base point R = exp( ) . At sepa-
rate moments, these vectors, however, occupy different vector spaces because the rotation op-
erator depends on time, namely R = R (t ) . The base point is moving in process of time.
Vector quantities of this kind may be called spin vectors. Spin vectors are rather tricky in nu-
merical sense as they always occupy a distinct vector space on a manifold.
The material angular acceleration and time derivatives of total rotation vectors are related
by
R = T + T , R matTR ; ,, mat TI , (14)
where the tangential transformation depends on the total rotation vector, and the rotation op-
~
erator is R = exp( ) . Time derivative of the tangential transformation is given in Appendix.

2.4 Complement Rotation Vector


Let a rotation vector with a rotation angle larger than zero and less than full-angle, i.e.
0 < < 2 , then its complement rotation vector C is defined

C : = 2 , := . (15)

We could represent the rotation manifold globally with these two parametrization charts, see
Fig. 4. When a rotation angle exceeds straight-angle ( > ), we accomplish the change of
parametrization according to (15), giving a new rotation angle smaller than straight-angle.
Thus, we never get into trouble with singularity at = 2 .
As it is illustrated in Fig. 4, the change of parametrization maps rotation angle outside of
straight-angle into inside of straight angle. Note that there exists no other canonical parame-
trization with rotation less than perigon such as those parametrizations given in (15).

rotation manifold SO(3)

parametrization parametrization
~ ~C
mapping exp( ) mapping exp( )
2 change of 2
parametrization

parametrization chart complement parametrization chart

Fig. 4 The change of parametrization in the parameter space 3 for the rotation manifold.

6
J. Mkinen and H. Marjamki

When the rotation vector matTI is switched to the complement rotation vector
matTI C , in dynamic analysis we need its time derivatives that are
C

C = B C = B + B , (16)

where the symmetric kinematic operator is defined by B: = D C and its time derivative are

B = (1 2 ) I + 2 n n ,

B = 22 ( n )I + ( n + n ) 3( n )n n ,

(17)
B = 22 ( n ) + ( n ) I + n + n + n + n +

62 ( n ) + ( n ) n n + ( n )( n n + n n ) 2 ( n )B

where the rotation axis is n = / .

3 GEOMETRICALLY EXACT BEAMS


In this Chapter, we apply the principle of virtual work for deriving the resulting equations
of Reissners beam theory. The virtual work may be decomposed into three terms: external,
internal and inertial virtual works with the equation
W = Wext Wint + Winert , (18)
where the subscripts correspond to external, internal and inertial. In the virtual work of
internal forces, the minus sign indicates that internal forces work against the virtual displace-
ments. In addition, the inertial virtual work Winert includes a minus sign in its form. Some-
times it is convenient to avoid additional minus signs by introducing the virtual work of
acceleration forces by the formula Wacc := Winert .
Therefore, when we apply the principle of virtual work ( W = 0 ), governing equations
state
WaccA = Wext Wint + WaccB , (19)
where WaccA is the virtual work of acceleration forces, which depends on acceleration, and
WaccB corresponds to the other acceleration virtual works like the virtual work of centrifugal
forces.

3.1 Virtual Work of Reissners Beam


In this section, we derive the virtual work for Reissners beam when the kinematic as-
sumptions are applied. The virtual work of acceleration forces Wacc reads with using used the
central line conditions
~
Wacc = x c ( A 0 x c ) ds + ( J + J ) ds , (20)
L L

where x c is the beam central line placement vector and the cross section vector
E := X 2E 2 + X 3E3 . If the principal axes of the inertial tensor J are parallel to the base vectors

7
J. Mkinen and H. Marjamki

{E 2 , E 3} , then the matrix of the inertial tensor is diagonal. In following, we will use this as-
sumption for simplicity, having J = diag( J1 , J 2 , J 3 ), J1 = J 2 + J 3 , where the lower indexes
correspond to the base vectors {E1 , E 2 , E 3} .
As for the virtual work of acceleration forces, we could write the virtual work of external
forces, giving
Wext = x c n ds + R M R ds , (21)
L L

where the external force vector n and the external material moment vector M R .
We can also give the internal virtual work in its material representation
Wint = N + R M R d s , (22)
L

~
where the material curvature tensor is defined by R : = R TR and where the material internal
force vector N and the material internal moment vector M R are denoted by the constitutive
formulas. We introduce a linear constitutive relation in the material representation between
material strain and curvature vectors and internal force and m oment vectors by writing
N = Cn , M R = Cm R . (23)
The work conjugate of the material vector N is the variation of the material strain vector ,
defined by the formula and its variation:
~
: = R T x c E1 , = R T x c R T x c . (24)

4 LAGRANGIAN FORMULATIONS
Next we consider a total Lagrangian updating formulation for a material rotation vector
and compare it with an updated Lagrangian formulation. We also discuss some difficulties
about an Eulerian formulation. We exploit only the material representation but the spatial ver-
sion may be considered similarly.
Updating formulations of the material rotation vector can be divided into three classes [1]:
1) The total Lagrangian formulation where the total rotation vector matTI is the
dependent variable
2) The updated Lagrangian formulation where the updated rotation vector
I ref matTI ref is the dependent variable
3) The Eulerian formulation where the incremental rotation vector R matTR is the
dependent variable
In the total Lagrangian formulation, we have updating procedure similar to displacement
vector
(i +1) = (i ) + ( i ) matTI , (25)

where the rotation vector change (i ) always occupies the vector space mat TI since the base
point I SO(3) remains fixed. The superscripts in brackets correspond to iterative steps.
We may consider the total Lagrangian updating procedure as a linear space updating proce-
dure. This formulation preserves the path-independent property (in static case) and can be re-
garded as a consistent updating formulation. The major drawback is the singularities at the

8
J. Mkinen and H. Marjamki

rotation angle = 2 and its multiples but we overcome this difficulty by the complement
rotation vector with the change of parametrization. We will study this updating procedure in
later.
The updated Lagrangian formulation is closely related with the total Lagrangian but the
base point R ref SO(3) , the reference placement, is updated occasionally. Usually, we up-
date the base point at every incremental step in static analysis, and at every time step in tran-
sient dynamic analysis. Updated Lagrangian procedure reads
( i +1)
ref = ref
(i )
+ ( i ) matTI ref , at every iteration,
~ ( n) (26)
ref = R ref exp ref , ref = 0 , when updating the base point,
R new old ( 0)

where the second formula corresponds to updating the base point which we execute occasion-
ally. The updated Lagrangian formulation does not preserve the path-independent property (in
static case). This is because the new base point R new
ref depends on the previously computed
solution and then the error of the updated rotation vector ref(n)
causes a cumulative error to
the base point R new
ref .
The Lagrangian formulations (total and updated) are geometrical integration methods
where the constraint manifold is parametrized. The total rotation vector, likewise the updated
rotation vector, forms a local parametrization for the rotation manifold SO(3) via the expo-
nential mapping, but the incremental rotation vector (Ri ) matTR(i ) at each iteration step (i )
belongs to a different vector space. We called quantities like incremental rotation vectors by
spin vectors. Spin vectors have a vector character at a point but not in local sense, on the con-
trary to total and updated rotation vectors that are vectors in local sense, i.e. a local parametri-
zation of the manifold.
Next, we will derive the virtual work form in terms of the total material rotation vector
and its virtual displacement in order to give a total Lagrangian formulation. We need to
express spin vector like the virtual incremental rotation vector R , the angular velocity
vector R , the angular acceleration vector R and the curvature vector R in terms of the
total rotation vector and , giving

R = T , R = T , A R = T + T , R = T , (27)

where the tangential transformation is given in (9). We note that , , , matTI whereas
the spin vectors R , R , R matTR (subscripts for base points are omitted for simplicity).
Now we could write the virtual work of acceleration forces Wacc , (20), in terms of the to-
tal material vectors , , , , yielding

Wacc = x c ( A 0x c ) ds + T T JT + T T JT + T T (T ) ~ JT ds , (28)
L L

that can be decomposed into the acceleration dependent part WaccA and the velocity depend-
ent part WaccB

WaccA = x c ( A 0 x c ) ds + T T JT ds,
L L
(29)
WaccB = T JT + T ( T ) ~ JT ds .
T T

9
J. Mkinen and H. Marjamki

The virtual work of external forces reads in T -formulation


mat I

Wext = x c n ds + T T M R ds , (30)
L L

where the external material moment vector M R matTR so it is a spin vector. According to the
above formula, we may define a new external material moment vector by setting
M I := T T M R matTI , (31)

where mat TI is the initial vector space of rotation in the material description.
Finally the virtual work of internal forces, (22), can be written in the total Lagrangian for-
mulation

Wint = ( x c R N) d s + T T (R T x c ) ~ N + C1T ( , ) M R + T T M R d s , (32)


L L

where tensor C1 is given in Appendix. It is clear that the virtual work of internal forces is
more coupled than the other virtual work forms. This is mainly because of the choice of de-
pendent variable, the rotation vector.

5 CONSISTENT TANGENTIAL TENSORS OF REISSNERS BEAM


In this Section, we will derive consistent tangential tensors for geometrically exact Reiss-
ners beam. The total and updated Lagrangian formulations yield identical tangential tensors.
However the updated Lagrangian formulation requires secondary storage variables like the
curvature vector and the rotation operator at the previously converged solution, see [1] and
[5]. In the total Lagrangian formulation, secondary storage variables are completely avoided
since the reference placement is always the initial placement. The consistent stiffness tensor
of Reissners beam with the total Lagrangian formulation has been firstly introduced in [2].
That formulation suffers singularity at the rotation angle = 2 and its multiples, and is
therefore restricted with the rotation angle < 2 .
The virtual work of acceleration forces WaccA can be written via the bilinear form

WaccA (q0 , q; q) = M ( 0 ) : ( q q) ds , (33)


L

where q that denotes the generalized displacement field includes the translational displace-
ment and rotational quantities, i.e. q : = (d, ) . We have denoted M( 0 ) for generalized
mass tensor that depends on the current rotation field 0 at the time moment t = t0 . Accord-
ing to Eqn (29a) we have the generalized mass tensor M( ) for Reissners beam

A 0 I O
M ( ): = T
. (34)
O T JT
Next, we give the linearization for the virtual work forms Wacc , Wext and Wint , separately.
The virtual work of acceleration forces WaccA is already in a linear form with respect to the
generalized acceleration field q . The generalized mass tensor given in (34) however depends
on the generalized displacement field q . In addition, the form WaccB does not depend on the
translational displacement field d , which simplifies the linearization procedure considerably.

10
J. Mkinen and H. Marjamki

3
We denote the rotation field by C ( matTI ) and the translational displacement field by C( ),
hence the displacement manifold is equal to C ( matTI ) . 3

The linearization of the virtual work of acceleration forces Wacc , (20), at the state point
(q 0 , q 0 ) yields

Wacc (q, q, q; q) = Wacc (q0 , q0 , q; q) + D Wacc + D Wacc , (35)

where the linear virtual forms D Wacc and D Wacc may be given by

D Wacc = K cent :( ) ds , D Wacc = Cgyro :( ) ds , (36)


L L

and where the centrifugal and gyroscopic tensors K cent and Cgyro are denoted
~ ~
K cent : = C2 (0 J0 + J 0 , 0 ) + T T J ( J ) ~ C1( 0 , 0 ) +
+ T T J C5(0 ,0 ) + C1( 0 ,0 ) , (37)
~
Cgyro : = T T 0 J ( J0 ) ~ T + JC4 ( 0 , 0 ) .

Above we have utilized formulas given in Appendix. The angular velocity field 0 C ( matTR )
and the angular acceleration vector 0 C ( matTR ) are computed by (27) via the total material
rotation vector and its time derivatives 0 , 0 , 0 C ( matTI ) at the fixed time moment t = t0 .
The centrifugal and gyroscopic tensors are in general non-symmetrical tensors. This non-
symmetry arises from the kinetic energy that depends on the generalized displacement field q
as well as the generalized velocity field q .
Correspondingly, the linearization of the virtual work of external forces Wext at the point
q 0 yields

Wext (q; q) = Wext (q0 ; q) + D qWext (q0 , q) q , (38)


where the linear virtual form D q Wext q is denoted with the aid of the loading tensor K load

D q Wextq = K load : ( q q) ds . (39)


L

We have above assumed that the external forces and moments do not depend on the gene r-
alized velocities q . External force or moment fields are called conservative if they are ob-
tained via a generalized displacement dependent potential functional. On a linear space (i.e.
on a flat manifold), the conservative loading and the symmetry of the loading stiffness tensor
are equivalent issues. This symmetry of the loading stiffenss tensor is because of the local
parametrization of the rotation operator, see details in [6].
Next, we linearize the virtual work form of internal forces Wint . Now the internal work
form Wint can be given by

Wint = q B T Fint ds , (40)


L

where q : = (d, , ) . The generalized internal force field Fint and the kinematic tensor B
are

11
J. Mkinen and H. Marjamki

N RT O ( R Tx c )~ T
Fint : = , B := . (41)
MR O T C1 ( , )
Moreover, the field Fint can be given in the term of the material strain and curvature vec-
tors with the aid of the linear constitutive relations. Now, we could linearize the internal work
form Wint (40) at the point q0 = (d0 , 0 ) in the vector direction q = ( x c , , ) giving

Wint (q; q) = Wint (q0 ; q) + D qWext (q0 , q) q , (42)

where the linear virtual form D q Wint q is denoted with the aid of the material stiffness ten-
sor K mat and the geometric stiffness tensor K

D q Wint q = ( K mat + K ) : ( q q ) ds, K mat : = B T0 C NM B 0 ,


L
~
O O RNT (43)
K := O O C 2 ( M R , 0 ) ,
T~ ~ ~
T NR T C T2 ( M R , 0 ) C3 ( M R , 0 , 0 ) + C 2 ( NR T x c , ) + T T N ( R T x c ) ~ T

where the material stiffness tensor K mat arises from the linearization of the vector Fint with
the aid of kinematic relation, and the geometric stiffness tensor K arises from the lineariza-
tion of the kinematic operator B . The tensors C2 and C3 are given in Appendix.
We note that the material stiffness tensor K mat is clearly symmetric because of the sym-
metry of the elasticity tensor CNM . In addition, the geometric stiffness K is also a symmet-
ric tensor. This symmetry of the stiffness tensor is due to the local parametrization of the
rotation operator, see details in [6].

5.1 Switching Beam Element


The method of resolving singularity problems is basically similar to Cardona & Gradin
[1], however they have not solved the problem where the rotation vector at one node of the
element has been switched but the rotation vector at another node of the same element re-
mains unaltered. In large rotational analysis, there often occurs a situation where the rotation
vector has been switched into its complement rotation vector at the one end of the beam ele-
ment, and the rotation vector has remained the same at the other end, since the rotation angle
is less than straight angle. In Fig. 5 we give an example of this situation: at the node 1 the ro-
tation vector 1C mat TI C has been switched, and at the node 2 the rotation vector 2 mat TI
still less that straight angle, hence it is non-switched. The virtual work at the node 2 reads in
the spaces matTI C and matTI

2C f 2C ( 2C , 2C , q, q, q ) M C2 2C = 2 f 2 ( 2 , 2 , q, q, q ) M 2 2 , (44)

where f 2C := f 2Cext f 2Cint f 2CaccB corresponds the nodal force vector at the node 2 including the
external, internal and velocity dependent forces, M C2 is the mass matrix of the node 2, and
q, q, q are the displacement, velocity and acceleration vectors for the other nodes than the
node 2. Note that the element matrices and the nodal force vectors for the ordinary element 2,
in Fig. 5, are known and these nodal vectors and matrices are given in terms of the rotation

12
J. Mkinen and H. Marjamki

vectors 2 , 2 , 2 matTI . Substituting 2C = B 2 and (16) for (44), we have the force
vector and the mass matrix at the node for all 2

f 2 = B T f 2C ( 2C , 2C , q, q, q ) M C2 B2 matTI ,
(45)
M 2 = B T M C2 B matTI matTI .
Linearized form of the force vector f 2 reads

K 2 : = D 2 f 2 = B T K C2 B + C C2 B + M C2 B + K 2 ( f 2C + M C2 2C ) matTI matTI ,
(46)
C 2 : = D 2 f 2 = B T C C2 B + 2M C2 B matTI matTI ,

where K C2 : = D C f 2C , C C2 : = D C f 2C , and the geometric stiffness matrix K 2 yields


2 2

K 2 ( f ) = 22 ( f n )I + f n + n f 3( f n )n n . (47)

We have only considered the virtual work at the node 2. Of course, the other nodes of the
same element depend on the rotation 2C and its time derivatives. When linearizing 2C and
its time derivative 2C , we may just substitute following formulas

2C = B 2 , 2C = B 2 + B 2 .(1)
and the linearizations of the acceleration C .
The interpolation in the switching beam element Fig. 5 is carried out as follows
C ( s ) = N 1 ( s )1C + N 2 ( s )2C matTI C (48)
where linear shape functions N i ( s ) are used. We note that the interpolation (48) is fully de-
fined and it is a genuine formula where the nodal rotation vectors iC and the interpolated
value C ( s ) belong to the same vector space of rotation, namely matTI C .
Finally, we note that the complement of complement rotation vector is equal to the rotation
vector itself, i.e. CC = , hence the switching beam element in Fig. 5 is equivalent to the
case where at the node 2 the rotation vector 2C has been switched, and at the node 1 the ro-
tation vector 1 ( = 1CC ) is still less that straight angle hence it is non-switched.

Switching beam Ordinary beam


2 2
1C 3 Cantilever beam system
under loadings

2C
1C
Ordinary beam

Fig. 5 A typical situation when switching beam element is needed.

13
J. Mkinen and H. Marjamki

6 NUMERICAL RESULTS
In this Section, we present several numerical examples to illustrate the performance of the
proposed total Lagrangian beam element. We choose the simplest finite element approxima-
tions based on 2-node isoparametric beam element with linear interpolation functions. Two-
point Gaussian quadrature is used in the numerical computations of the acceleration vectors
and their tangential tensors, and avoiding locking phenomena, a one-point quadrature is used
for the internal vectors and their tange ntial tensors.

6.1 Cantilever 45-degree bend

F = 600
e2
EA = 10 7 , GA2 = GA3 = 5 10 6
GJ = 7.05 105 , EI 2 = EI 3 = 8.33 105 F
L = 25 e1

e3

Fig. 6 The cantilever 45-degree bend beam under the free-end force.

The first example presents a geometrically nonlinear analysis of a cantilever 45-degree


bend (with the radius R = 100 , yielding the length of the cantilever beam L = 25 ) placed in
the horizontal plane with a vertical static force applied at the free end, see Fig. 6. This exam-
ple was introduced by [7], and it is the one of the most frequently used example. The length of
the rotation vector remains smaller than 2 , so that a single rotation parametrization chart is
used during computations. The cantilever beam is divided into eight 2-node beam elements
with the cross-section properties given in Fig. 6.
The free-end displacements are given in Table 1 compared with the other authors results.
The comparing results in Table 1, we found a minor difference between each other that is due
to the difference of the constitutive relations, interpolation functions, rotation parametrization
or even cross-section properties. However, our model and the model presented in [2] give
practically the same response because of the identical rotation parametrization, i.e. the total
Lagrangian updating formulation. The final solution is obtained by applying the load in six
equal load step with the full Newton-Raphson iteration method. We also observed the identi-
cal quadratic convergence rates as in [2].

Reference Displacement Displacement Displacement


d1 d2 d3
Present 23.696 53.497 13.668
[7] 23.5 53.4 13.4
[8] 23.48 53.37 13.5
[1] 23.67 53.50 13.73
[9] 23.87 53.71 13.63
[10] 23.746 53.407 13.601
[2] 23.697 53.498 13.668
Table 1. The cantilever bend free-end displacement components Helical beam

14
J. Mkinen and H. Marjamki

6.2 Helical beam

n = 50 m = 200
e2 EA = GA2 = GA3 = 10 4
GJ = EI 2 = EI 3 = 10 2 m

L = 10 e1

e3
n

Fig. 7 Straight cantilever beam under the force n and the moment m R spatTR load.

This example was introduced by [11] where the straight cantilever beam is at the free-end
under the force vector n and the spatial moment vector m R spatTR load both in the direction
e 3 , see Fig. 7. This type of moment load is nonconservative. The purpose of the example is to
show how our switching beam element, described in [5], works in the situation when the free-
end rotation angle goes over 2 . If the moment load acts only, the free-end rotation angle is
equal to 20 that corresponds ten revolutions.
During the computation, the spatial moment vector m R spatTR , which is a spin vector, is
mapped into the initial vector space of rotation T
mat I via Eqns (31) giving
M I = T R m R = Tm R . This mapping is necessary in order to get the external nodal load
T T

vector of the total Lagrangian beam element where the rotation vector ( s ) matTI .
At the free-end, the out-of-plane displacement component (in the direction e 3 ), when
loading is increased into its final value, is shown in
Fig. 8. We have very similar but not identical results compared with [11], where out-of-
plane displacement is almost symmetrical around zero. We observe the decrease of finite ele-
ment error when the cantilever beam is divided into 200 elements compared with the model of
100 elements.
1

0.9 100 elements

0.8 200 elements

0.7

0.6
Loading

0.5

0.4

0.3

0.2

0.1

0
1 0.5 0 0.5 1 1.5 2 2.5 3 3.5
Displacement

Fig. 8 The free-end out-of-plane displacement curve versus loading.

15
J. Mkinen and H. Marjamki

6.3 Fast Symmetrical Top

e 3 , E3 EA = GA2 = GA3 = 106


e3 , E3
GJ = EI 2 = EI 3 = 103
J1 = 1 J 2 = J 3 = 0.5
mgl = 20
A 0 = 13.5 L =1
J1 = J 2 = 5 mg = 40
J3 = 1 mg
mg

O e2 , E2 O e2 , E2
e1 , E1 e1 , E1

Fig. 9 The fast Symmetrical tops at the initial position, modeled by a rigid body (on the left) and a Reissners
beam element (on the right).

Next, we examine how our beam element acts as a rigid body in a dynamical case. We
consider, in this example, the motion of the fast symmetrical top with one point fixed origi-
nally given in [12], Fig. 9. The Newmark time integration scheme is used with scheme pa-
rameters = 1 / 4 and = 1 / 2 , i.e. the trapezoidal rule.
We obtain the acceleration force vector and its tangent tensors, and the mass tensor of the
rigid body element in the total Lagrangian formulation from Eqns (29), (34), and (37) by
keeping the unknown variables, ( x, ) , constant and integrating over the beam length. The
top is under the gravitation load mge 3 , see Fig. 9, that cause the moment load in the rigid
body element and the force load in the beam element. The numerical constants, the initial
conditions, and the external moment M I in the material frame {E1 , E 2 , E 3 } are for the rigid
body element as follows:
J = 5 E1 E1 + 5 E 2 E 2 + 1 E3 E3
M I = mgl T ( ) E 3 ( R T E 3 )
T
matTI (49)
0 = 0.3 E1 , 0 = 50 E 3 , 0 = T 1 ( 0 ) 0

where m, g , l are the mass of the top, the gravity constant, and the distance of the top gravity
center with respect to the origin O. The initial conditions mean that the angular velocity of the
top is about eight revolutions per time unit; therefore, a switching beam element is required.
The numerical results of time history, the nutation and precession angles, for the top mod-
eled by the rigid body element and by the beam element are shown in Fig. 10. Two beam
models are exploited: Stiff beam where constants are given in Fig. 9 and Flexible beam
where the stiffness constants are reduced into the 100th part. Fig. 10 indicates the identical be-
havior for the Rigid body and the Stiff beam model as expected. The results are identical
with the computation given in [4] where a different rotation updating scheme, an updated La-
grangian formulation, is used.

16
J. Mkinen and H. Marjamki

0.335 0.6 Rigid body


Rigid body
0.33 Stiff beam Stiff beam
Flexible beam 0.5
Flexible beam

Angle of precession [rad]


Angle of nutation [rad]

0.325
0.4
0.32
0.3
0.315

0.2
0.31

0.305 0.1

0.3
0
0 0.5 1 1.5 0 0.5 1 1.5
Time t [s] Time t [s]

Fig. 10 Time history for the notation and precession angles.

6.4 Two-Component Robot Arm

EA = GA2 = GA3 = 10 6
GJ = EI 2 = EI 3 = 10 5
(t )
J1 = 2 J 2 = J 3 = 1
15
.
A 0 = 1 L = 5
e2
0.5 t
spherical
L joint L e1

(t )
e3 w( t ) = 2 (t )

Fig. 11 Two-component robot arm: problem data.

This example, introduced in [13], considers a simple multibody system composed of two
flexible beams connected by a spherical joint. The system is brought in motion by imposing
simultaneously the forced displacement in the out-of-plane direction e 3 and a rotation about
this axis, see Fig. 11. Two finite element models are built: one model has in total 8 equal lin-
ear beam elements and another has in total 40 elements. The computations are carried by the
Newmark scheme with the constant time step h = 0.01 .
The computed time history response for the tip horizontal, vertical, out-of-plane displacement
components and the length of the tip displacement vector are shown in Fig. 12. These results
correspond the results given in [13].

17
J. Mkinen and H. Marjamki

0 12

40 elements
2 40 elements
10 8 elements
8 elements
Tip horizontal displacement

Tip vertical displacement


4
8
6

6
8

10 4

12 2

14
0
16
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time t [s] Time t [s]

9 20
40 elements
8 40 elements 18
8 elements
8 elements
Tip outofplane displacement

7 16

Tip displacement norm


6
14
5
12
4
10
3
8
2
6
1

0 4

1 2

2 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time t [s] Time t [s]

Fig. 12 Robot arm: Time history of tip displacement components and its length for 8 and 40 element mesh.

7 CONCLUSIONS
We have introduced the new Reissners geometrically exact beam element that is based on
the total Lagrangian updating procedure. The element has the total rotation vector as the de-
pendent variable and the singularity problems at rotation angle 2 and its multiples are han-
dled by the change of parametrization on the rotation manifold SO(3) . This change of
parametrization is applied via the complement rotation vector. The kinematic assumptions of
the beam element are based on the Timoshenko-Reissner-hypothesis. The beam formulation
has several benefits like all the unknown vectors belong to the same tangential vector space,
no need forfor secondary storage variables, the path-independence property in static case, any
standard time-integration algorithm may be used, the symmetric stiffness tensor, the simple
form of the kinetic energy and all the nonlinear effects are included. Several numerical exam-
ples illustrate a good efficiency of the introduced element.

8 APPENDIX
The tensors C i are defined by directional derivatives

C1 ( V , ) : = D T V , C 2 ( V, ) : = D T T V ,
C3 ( V, , ) : = D C1T ( , ) V , V 3
, (50)
C 4 ( , ) : = D T , C 5 ( , ) : = D T ,

18
J. Mkinen and H. Marjamki

where tensors Ci read explicitely


~ ~
C1 ( V, ) = c1 V c2 ( V ) + c3 ( V ) c4 V + c5 ( V )I + V ,
~ ~
C 2 ( V, ) = c1 V + c2 ( V ) + c3 V + c4 V + c5 ( V )I + V ,
~ ~
C 3 ( V , , ) = c1 V + c2 ( V ) + c3 ( )( V ) I + c2 ( V ) +
S

1 ~
+ c3 ( ) ( V ) + c1 V + c2 ( V ) + c3 ( )( V ) ( ) +
S
+ c3 ( V ) ( ) + c5 ( V ) , (51)
S S

C 4 ( , ) = (c1 + c5 )( )I + 2c3( )( ) + 2c5( ) + (c1 + c5 )( ) +


~ ~
c2( ) c2 ( ),
c3
C5 ( , ) = c3 ( ) 2 + c5 ( ) I + ( ) 2 + c3 ( ) ( ) + 2c3( )( )+

c1 c ~ ~ ~
+ + c3 ( )( ) 2 ( )( ) c2( ) + c2( ) + (c1 + c5 )( ),

and where denotes the symmetric tensor product and coefficients ci and their derivatives
S
are given by
( a b) : = (a b) + ( b a ), a, b 3
,
S

cos sin sin + 2 cos 2


c1: = , c2 : = ,
3 4
3 sin 2 cos cos 1 sin
c3: = , c4 : = , c5 : = , (52)
5 2
3
3 sin 2 sin 3 cos 2 cos 5 sin 8 cos + 8
c1 = , c2 = ,
4 5
7 cos + 8 + 2 sin 15 sin
c3 = .
6
The limit processes for the tensors Ci give
~ ~
lim C1 ( V , ) = 1 V , lim C 2 ( V, ) = 1 V,
0 2 0 2
lim C 3 ( V , , ) = 1 ( V )I + 1 ( V ), , (53)
0 3 6 S

lim C 4 ( , ) = O, lim C 5 ( , ) = 1 ( ) + 1 ( )I .
0 0 6 6
We also need the time derivative of the transformation T , giving
~ ~
T( , ) = c1 I c2 + c3 + c4 + c5 + , (54)

where the coefficients are given in ( 52). The limit value of the tensor T is
~
lim T( , ) = 1 . (55)
0 2

19
J. Mkinen and H. Marjamki

ACKNOWLEDGEMENTS
Financial support for this research was provided by the Academy of Finland under the
project number 206020. This support is gratefully acknowledged.

REFERENCES
[1] A. Cardona and M. Gradin, A Beam Finite Element Non-Linear Theory with Finite
Rotations, International Journal for Numerical Methods in Engineering, 26, 2403-2438,
1988.
[2] A. Ibrahimbegovi, F. Frey and I. Kozar, Computational Aspects of Vector-Like

Parametrization of Three-Dimensional Finite Rotations, International Journal for Nu-


merical Methods in Engineering, 38, 3653-3673, 1995.
[3] J. Stuelpnagel, On the Parametrization of the Three-Dimensional Rotation Group. SIAM
Review, 6, 422-430, 1964.
[4] J. Mkinen, Critical Study of Newmark-Scheme on Manifold of Finite Rotations, Comp.
Meth. Appl. Mech. Engng, 191, 817-828, 2001.
[5] J. Mkinen, A Formulation for Flexible Multibody Mechanics Lagrangian Geometri-
cally Exact Beam Elements using Constraint Manifold Parametrization, Doctoral The-
sis, TUT, 2004. URL: http://www.tut.fi/~jmamakin/vk.pdf
[6] J. Makowski and H. Stumpf, On the Symmetry of Tangent Operators in Nonlinear Me-
chanics, ZAMM Appl. Mathematics and Mechanics, 75, 189-198, 1995.
[7] K.-J. Bathe and S. Bolourchi, Large Displacement Analysis of Three-Dimensional Beam
Structures, Int. J. Num. Meth. Engrg, 14, 961-986, 1979
[8] J. Simo and L. Vu-Quoc, On the Dynamics in Space of Rods Undergoing Large Motion
- A Geometrically Exact Approach, Comp. Meth. Appl. Mech. Engng, 66, 125-161,
1988.
[9] M. Crisfield, A Consistent Co-Rotational Formulation for Non-linear, Three-
Dimensional, Beam Elements, Comp. Meth. Appl. Mech. Engng., 81, 131-150, 1990.
[10] A. Ibrahimbegovi, An Finite Element Implementation of Geometrically Nonlinear
Reissners Beam Theory: Three-Dimensional Curved Beam Elements, Comp. Meth.
Appl. Mech. Engng, 122, 11-26, 1995.
[11] A. Ibrahimbegovi, On the Cho ice of Finite Rotation Parameters, Comp. Meth. Appl.
Mech. Engng, 149, 49-71, 1997.
[12] J. Simo and K. Wong, Unconditionally Stable Algorithms for Rigid Body Dynamics
that Exactly Preserve Energy and Momentum, Int. J. Num. Meth. Engng, 31, 19-52,
1991.
[13] A Ibrahimbegovi and M. Al-Mikdad, Finite Rotations in Dynamics of Beams and I m-
plicit Time-Stepping Schemes, Int. J. Num. Meth. Engng, 41, 781-814, 1998.

20

S-ar putea să vă placă și