Multibody System Dynamics 11: 185–208, 2004.

185
© 2004 Kluwer Academic Publishers. Printed in the Netherlands.

Large Deflection Analysis of a Thin Plate:

Computer Simulations and Experiments

WAN-SUK YOO1 , JEONG-HAN LEE1, SU-JIN PARK1,

JEONG-HYUN SOHN1, DMITRY POGORELOV2 and
OLEG DMITROCHENKO2
1 Pusan National University, Kumjung-Ku, Busan 609-735, Korea; E-mail: [email protected]
2 Bryansk State Technical University, Bulvar 50-letiya Oktyabrya 7, Bryansk 241035, Russia;
E-mail: [email protected]

(Received: 7 July 2003; accepted in revised form: 18 December 2003)

Abstract. Many previous studies have conducted computer-aided simulations of elastic bodies un-
dergoing large deflections and deformations, but there have not been many attempts to validate their
numerical results. The subject of this paper is a thin clamped plate undergone large vibration due to
attached end-point weight. The main aim of this paper is to show the validity of the absolute nodal
coordinate formulation (ANCF) by comparing to the real experiments. Large oscillations of thin
plates are studied in the paper with taking into account effects of an attached end-point weight and
aerodynamic damping forces. The physical experiments are carried out using a high-speed camera
and data acquisition system. For numerical modeling of the plate, the absolute nodal coordinate
formulation is used.

Key words: plates, large displacements, real experiments, absolute nodal coordinate formulation,
rigid-flexible multibody dynamics.

1. Introduction
There are many approaches for finite-element simulation of large displacement
problems in flexible multibody dynamics such as floating reference frame formula-
tion, different incremental approaches or large rotation vector formulations [1]. All
these approaches suffer from the high nonlinear terms in the equations of motion
because of using a local reference frame fixed to a flexible body [2, 6]. The recently
introduced ANCF(absolute nodal coordinate formulation) is quite remarkable from
this point of view.
The ANCF produces finite elements that can represent arbitrary large displace-
ments relative to the global reference frame directly. In the ANCF, in contrast to
other large deformation formulations, the equations of motion contain a constant
mass matrix and a constant vector of generalized gravity forces as well as zero
centrifugal and Coriolis forces [2]. So the only nonlinear term of equations of
motion is the vector of elastic forces, however, which is quite cumbrous one to
calculate. Several models of elastic forces for two or three-dimensional beams have
186 W.-S. YOO ET AL.

been proposed [3, 4, 11–13]. Several kinds of plate elements were proposed in
different ways [5, 8].
Investigation [5] presents the plate element as a solid body introducing three-
dimensional shape functions and applying solid mechanics theory for computation
of elastic forces. This approach allows to account for Mindlin’s plate theory.
In [8], thin plates are considered only and Kirchhoff theory is applied. ANCF
beam and plate elements are treated as generalizations of ordinary finite elements
used in structural mechanics. Several kinds of rectangular and triangular elements
are proposed in this manner.
Preceding papers [5, 8] have paid main attention to theoretical background and
building equations of motion of the elements. The validity of the elements has been
shown on simple abstract examples without any experimental proofs. This paper is
devoted to comparison of simulation and experiments with thin plates. It is the
second one in a series, which was initiated with a beam [13].
The subject of the current paper is a rectangular clamped plate undergoing large
vibration due to a heavy rigid-body weight attached to one of the free edges of the
plate. The Kirchhoff model based on ANCF [8] is employed to simulate the plate.
We describe the plate element in Section 2 very explicitly.
Next important part of the paper is devoted to modeling of a rigid body attached
to the plate, in Section 3. We derive differential-algebraic equations for simulation
of the rigid-flexible multibody system with ‘plate + weight’. The developed for-
mulation allows to introduce various kinds of links between the plate and the rigid
body such as revolute, spherical or pin joint.
Section 4 explains how to include air resistance forces in the simulation. It was
found that the mass matrix proportional damping by Rayleigh [14, 22], which was
used in a preceding paper [13] for beam elements, works for plates badly. That
is why we derived and applied a model of damping forces that are quadratic with
respect to velocities.
Experimental equipment used to perform experimental measurements is dis-
cussed in Section 5. Finally in Section 6, we have got large deflections of the
plate from real experiments to verify the computational results with ANCF. To
the authors’ knowledge, this paper is the first one to compare large deformation of
plates with real experiments.

2. Formulation of the Plate Element Using ANCF

2.1. NODAL COORDINATES AND SHAPE FUNCTIONS
In [8], the shape functions and nodal coordinates for this element were introduced
by analogy with the ordinary finite element method. In this paper, we introduce
them explicitly.
Let us consider a plate element of size a × b × h (length × width × thickness)
as shown in Figure 1. Since it is a thin plate, displacements are represented by its
middle surface.
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 187

Figure 1. Beam × beam model of a plate FE.

The surface of the plate is parameterized by values p1 and p2 . Let O1 be the

origin of the configuration space of the element. Then let us imagine a coordinate
curve (bold in Figure 1) parallel to the p1 axis (p1 -beam) so that we may define
the position r(p1 , p2 ) of an arbitrary point on the plate relative to the origin O
of the inertial reference frame. Extending of the idea of absolute nodal coordinate
formulation [3] for three-dimensional beams gives us the following equation:
 

 ρ1 

 
τ1
r(p1 , p2 ) = [ŝ1 I ŝ2 I ŝ3 I ŝ4 I] , (1)

 ρ 
 2 
τ2
where ŝk = sk (p1 , a) are the Hermite shape functions [17] for the p1 -beam:
s1 (p, l) = s3 (l − p, l) = 1 − 3ξ 2 + 2ξ 3 , s3 (p, l) = 3ξ 2 − 2ξ3 ,
p
s2 (p, l) = −s4 (l − p, l) = l(ξ − 2ξ 2 + ξ 3 ), s4 (p, l) = l(ξ 3 − ξ 2 ), ξ= ,
l
and where I is the 3×3 identity matrix and ρ k , τ k are the absolute nodal coordinates
of the p1 -beam (global displacements and slopes of the end points).
The latter vectors ρ k (p2 , b) and τ k (p2 , b) depend on the parameter p2 , which
defines the p1 -beam. They can be defined analogously. For example, ρ 1 is obtained
from the bold-faced p2 -beam shown in Figure 2 (cf. Figure 1 and expression (1)):
 00 

 r00 


 r01 


ρ 1 = ŝˆ 1 I ŝˆ 2 I ŝˆ 3 I ŝˆ 4 I 00
,

 r00 


 01 
0b

r0b

where ŝˆ k = sk (p2 , b) are the p2 -beam’s shape functions. The vectors
∂ i+j r
rijuv = j
(2)
∂p1i p2 p1 =u
p2 =v
188 W.-S. YOO ET AL.

Figure 2. Set of nodal coordinates of a plate FE.

play a part of nodal coordinates of the p2 -beam. The next step involves using the
same beam functions to interpolate slopes:
 

 r10 


 r11 
00


τ 1 = ŝˆ 1 I ŝˆ 2 I ŝˆ 3 I ŝˆ 4 I 00
. (3)

 r10 

 11 

0b 
r0b

Note that we use slopes r10 10

00 and r0b here like we used the displacements in Equation
(1); that is why the second-order slopes r11 11
00 and r0b are introduced.
The expressions for ρ 2 and τ 2 are obtained analogously:
 00   

 ra0 
 
 r10 


 
 

a0 

 r01  r11
ˆ ˆ ˆ ˆ
ρ 2 = ŝ 1 I ŝ 2 I ŝ 3 I ŝ 4 I a0
, τ 2 = ˆ
ŝ 1 I ˆ
ŝ 2 I ˆ
ŝ 3 I ˆ
ŝ 4 I a0
.

 rab 
00
 
 rab 
10


 01  
 11 
rab rab
After substitution of ρ k and τ k into (1), we obtain the radius vector
r = Se, (4)
along with the matrix of global shape functions
S = [S11 I, S12 I, S13 I, S14 I; . . . ; . . . ; S41 I, S42 I, S43 I, S44 I],
Sij = si (p1 , a)sj (p2 , b) = ŝi ŝˆ j (5)
and the vector of absolute nodal coordinates

00 , r00 , r0b , r0b , r00 , r00 , r0b , r0b , ra0 , ra0 , rab , rab , ra0 , ra0 , rab , rab } . (6)
e = {r00 01 00 01 10 11 10 11 00 01 00 01 10 11 10 11 T

ij
In the latter expression, we omit the transpose signs T over the elements ruv .
Thus, the set of two-dimensional shape functions we introduce for the plate
element is a cross Cartesian product of sets of one-dimensional shape functions
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 189

for beams. Since the expressions are decoupled in their parameters p1 and p2 , it
offers certain advantages. For example, it allows us easily to reduce their double
integration to a single one when developing the expressions for elastic forces.
The disadvantage of using these shape functions is that they require the use of
second-order slopes that increase the number of degrees of freedom up to 48, and
may not have any clear geometrical meaning. Dmitrochenko and Pogorelov [8]
proposed a way to eliminate higher-order slopes from the set of nodal coordinates
(6).

2.2. EQUATIONS OF MOTION AND MASS MATRIX

To derive dynamical equations for the plate element, we employ Lagrange equa-
tions in a matrix representation

d ∂T ∂T ∂U δW
− + = ,
dt ∂ ė ∂e ∂e δe


with kinetic energy

T = (1/2) P µṙT ṙ dP , internal strain energy U and virtual
work δW = P δrT µg dP resulting from the external gravity force µg. We use

a b
the symbols P . . . dP = 0 0 . . . dp1 dp2 to denote double integration over the
plate surface.
Taking Equation (4) into account leads to the following equations of motion:

Më + Qe = Qg , (7)


with a constant mass matrix M = P µST S dP , plate surface mass density µ,
Qe = ∂U/∂e and gravity Q = S̄¯ µg generalized forces, where S̄¯ =
T
g
elasticity


P S dP . Note that centrifugal and Coriolis inertia forces are absent from this
equation, as is usually the case when one uses the absolute nodal coordinate for-
mulation.
An explicit expression for the mass matrix can be obtained in block matrix form
using the definition of the shape function matrix (5):
 
M11 M12 M13 M14
 M21 M22 M23 M24 
M=  M31 M32 M33 M34  ,


M41 M42 M43 M44

where each block Mij is also a block matrix

 
Mij 11 Mij 12 Mij 13 Mij 14
 Mij 21 Mij 22 Mij 23 Mij 24 
Mij =  
 Mij 31 Mij 32 Mij 33 Mij 34 
Mij 41 Mij 42 Mij 43 Mij 44
190 W.-S. YOO ET AL.

with block-elements

Mij kl = Mij kl I,
 
Mij kl = µSik Sj l dP = µ ŝi ŝˆ k ŝj ŝˆ l dP
P P

a b 00 ˆ 00
= µ ŝi ŝj dp1 ŝˆ k ŝˆ l dp2 = µS̄ˆ ij S̄ˆ kl . (8)
0 0

The matrices with hats are similar to the mass matrices for the beam elements [3]:
 
156 sym
a  2 
S̄ˆ ij00 =  22a 4a .
420  54 13a 156 
−13a −3a −22a 4a
2 2

Matrix S̄¯ is easy to calculate explicitly, too:


S̄¯ = S̄¯ 11 I, S̄¯ 12 I, S̄¯ 13 I, S̄¯ 14 I; . . . ; . . . ; S̄¯ 41 I, S̄¯ 42 I, S̄¯ 43 I, S̄¯ 44 I ,
  a b
ˆ
S̄¯ ij = Sij dP = ŝi ŝˆ j dP = ŝi dP ŝˆ j dp2 = S̄ˆ i0 S̄ˆ j0 . (9)
P P 0 0

Thus, all the terms in Equation (7) are obtained except for the elastic forces Qe ,
which present the most calculation difficulties to due to the complexity of the strain
energy involved.

2.3. STRAIN ENERGY AND ELASTIC FORCES

According to the Kirchhoff theory, the strain energy of an orthotropic plate can be
decomposed into one due to the longitudinal and shear deformations in the mid-
plane, and one due to bending and twisting [17, 20]:

U = U ε + U κ,
 
  2  2
6 
Uε = 2 Dij εij2 + 2D22
11
ε11 ε22  dP , (10)
h i=1 j =1
P
 
 
2 
2
1
Uκ =  Dij κij2 + 2D22
11
κ11 κ22  dP . (11)
2 i=1 j =1
P
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 191

This strain energy equation contains, firstly, the elastic parameters of the mater-
ial of the plate: flexural rigidities D11 , D22 and twist stiffness D12 :
E11 h3 E22 h3 E12 h3
D11 = , D22 = , D12 = D21 = ,
12(1 − ν12 ν21 ) 12(1 − ν12 ν21 ) 6
as well as an additional stiffness coefficient D22 11
= 0.5(D11 ν21 + D22 ν12 ). The
latter expressions depend on the Young moduli E11 , E22 and a shear modulus E12
as well as on the Poisson ratios ν12 and ν21 . These ratios satisfy the condition
E11 ν21 = E22 ν12 .
Secondly, the strain energy equation contains the geometrical parameters of the
plate: the longitudinal deformations ε11 , ε22 and shear deformations ε12 = ε21 as
well as the transverse curvatures κ11 , κ22 and the twist κ12 = κ21 .
In the case of small deflections [20], the deformed surface of the plate is defined
by three scalar functions such as the mid-plane displacements u(x, y), v(x, y)
and the transverse displacement w(x, y). In such a situation, the afore-mentioned
deformations and curvatures are calculated as follows:
∂u ∂v 1 ∂u ∂v
ε11 = , ε22 = , ε12 = + ,
∂x ∂y 2 ∂y ∂x
∂ 2w ∂ 2w ∂ 2w
κ11 = , κ22 = , κ12 = . (12)
∂x 2 ∂y 2 ∂x∂y
In our case, where the plate is oriented in an arbitrary way and specified in a
parameterized form r = r(p1 , p2 ), we should use relationships from differential
geometry of surfaces [18]. By doing this, we obtain the following expressions for
deformations in the mid-plane of the plate
1
εij = (rTi rj − δij ) (13)
2
with Kronecker symbols δij as well as for transverse curvatures and twist
rTij n
κij = (14)

n
3
with the normal vector n = r1 × r2 . Other notations are derivatives
∂r ∂S ∂ 2r ∂ 2S
ri = = e, rij = = e. (15)
∂pi ∂pi ∂pi pj ∂pi ∂pj
It can be shown that Equation (13) expresses Green’s non-linear strain-displace-
ment relationships. Indeed, let us imagine that the deformed position of a plate is
defined in parameterised form as follows:
   
 p1   u1 (p1 , p2 ) 
r = u∗ + u = p2 + u2 (p2 , p2 ) .
   
0 u3 (p1 , p2 )
192 W.-S. YOO ET AL.

Here u∗ specifies the initial flat configuration of the plate while u corresponds
to its deflections. If we apply Equations (13), we obtain the following mid-plane
deformation quantities:
 
∂uj  ∂uk ∂uk
3
1 ∂uj
εij = + + , i, j = 1 . . . 2.
2 ∂pi ∂pj k=1
∂pi ∂pj

As one can see, the expressions contain both linear and quadratic terms. Note
that the linear part of the latter expression corresponds to the values in Equation
(12).
Analogously, one can ensure that expression (14) gives the same curvatures and
twist as in the linear case in Equation (12) when the normal vector n is supposed
to be fixed at n ≡ {0, 0, 1}T .
After computing the strain energy, the vector of generalized elastic forces can
be found as its gradient:
∂U
Qe = = K(e)e, (16)
∂e
where K(e) is the nonlinear stiffness matrix of the plate element.
In [8], these cumbrous computations were performed and several models of
mid-plane and transverse generalized elastic forces were obtained.

3. Modeling of the Rigid Body Attached to the Plate

First let us consider a free rigid body whose motion is described by Newton–Euler
equations
mj aj = mj g,

Jj εj + ω̃j Jj ωj = 0,

with mass mj and the 3×3 symmetric inertia tensor Jj in its centroidal axes, vector
ωj and the 3 × 3 skew-symmetric tensor ω̃j of angular velocity of the body, as well
as mass center acceleration aj and angular acceleration εj of the body. The vector
g represents the acceleration of gravity force applied to the body center of mass,
that is why its moment in the second equation is equal to zero.
Numerical integration of the dynamical equations assumes representing them in
terms of generalized coordinates
 
rj
ej = , (17)
α

where rj = {xC , yC , zC }T is the radius vector of the body mass center and α =
α{α1 , α2 , α3 }T is a set of orientation angles. Possible singularities of the orientation
angles can be avoided by employing quaternion parameters instead.
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 193

The vectors of linear and angular velocities, as well as the accelerations of the
body, are then introduced as follows:
vj = Dj ėj ⇒ aj = v̇j = Dj ëj + aj , where aj = Ḋj ėj ,

ωj = Bj ėj ⇒ εj = ω̇j = Bj ëj + εj , where εj = Ḃj ėj , (18)

where Dj and Bj are the Jacobian matrices discussed in Section 3.3.
After substitution for these dependencies, the dynamical equations of motion of
the rigid body take the form of Euler–Lagrange equations
Mj ëj = Qj , (19)
where
   
DTj mj Dj O DTj mj (g − aj )
Mj = , Qj = .
O BTj Jj Bj Bj (−ω̃j Jj ωj − Jj εj )
T

3.1. EQUATIONS OF MOTION FOR CONSTRAINED SYSTEM

In order to simulate a rigid-body weight attached to a plate, one can use the sub-
system technique. Wherein the hybrid ‘plate+body’ system is decomposed into two
subsystems – an ANCF plate subsystem and another one containing the rigid body
only. The equations of motion of the two disconnected subsystems can be written
as follows:
Mi ëi = Qi , (20)

Mj ëj = Q. (21)

Equation (20) resembles the equations of structural dynamics (7) for an elastic
plate, while Equation (21) is the Euler–Lagrange equation (19) for the rigid body.
However, the motion of the two subsystems is constrained by
f(ei , ej ) = 0. (22)
The explicit expression is derived in Section 3.2. These constraints add corres-
ponding reaction forces to the dynamic equations in (20) and (21). After double-
differentiating the constraint equation (22) with respect to time,
f̈ = i ëi + j ëj + f = 0, (23)
the equations of motion of the constrained system can be represented in augmented
matrix form [2, 21, 22]
    
Mi O Ti  ëi   Qi 
 O Mj Tj  ëj = Qj ,
    
i j O −λ −f
194 W.-S. YOO ET AL.

Figure 3. Connection of elastic plate and rigid body.

where i and j are Jacobian matrices of the constraints (22) with respect to the
coordinate sets ei and ej , respectively, while λ is a column of Lagrange multipliers.

3.2. DETAILS OF THE CONSTRAINT EQUATIONS

Constraint equation (22) consists of two parts. The first expresses the closure con-
dition of radius vectors, as it can be seen in Figure 3:
f1 = ri + rij − rj = 0. (24)
The second equation requires the same for orientations. We can write the equa-
tion for orthogonal rotation matrices as follows:
 = vAi Aij A−1
j − I = O, (25)
where Ai and Aj are orientation matrices of coordinate systems xi , yi , zi and xj ,
yj , zj . Aij is a rotation matrix specifying the angular shift, if any, between the two
systems. In our case, there is no such a shift so Aij ≡ I.
Yet, Equation (25) is redundant because it contains nine scalar equations al-
though only three of them are independent. The latter equation should be vectorized
using the well-known transformation
1
3
f2 = ι̃k ιk = 0, (26)
2 k=1
where ιk are unit basis vectors of the fixed reference frame and ι̃k are their 3 × 3
skew-symmetric tensors, k = 1 . . . 3.
After a direct computation, we obtain the following vector equation:
 
 (32 − 23 )/2 
f2 = (13 − 32 )/2 = 0 (27)
 
(21 − 12 )/2
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 195

as the second part of constraint equation (22).

The first derivative of these equations gives the constraint equations for linear
and angular velocities:
ḟ1 = vi + ω̃i + ω̃i rij + vij − vj = 0,

ḟ2 = ωi + ωij − ωj = 0.

These are used for computation of initial conditions for velocities. The second
derivative expresses the constraints for linear and angular accelerations
f̈1 = ai + ε̃i rij + ω̃i ω̃i rij + 2ω̃i vij + aij − aj = 0,

f̈2 = εi + ω̃i ωij + ε ij − εj = 0.

Expressed in terms of generalized coordinates, the constraint equations for ve-

locities and accelerations take the forms
ḟ = i ėi + ij ėij + j ėj + f = 0 (28)
and
f̈ = i ëi + ij ëij + j ëj + f = 0, (29)
respectively. The newly introduced matrices may be written as follows:
     
Di − r̃ij Bi Dij Dj
i = , ij = , j = − , f = 0,
Bi Bij Bj
  
 ai + ε̃ i rij + ω̃i ω̃i rij + 2ω̃i vij + aij − aj
f = .
εj + ω̃i ωij + εij − εj

The linear and angular velocities vi and ωi , and the accelerations ai and εi for
the elastic plate, as well as the related Jacobian matrices Di and Bi are defined
exactly as they were for the rigid body in Equation (18).
In order to obtain the constraint equation (23) we should eliminate the term
from Equation (29). This can be done in all cases, but in our case, ij ëij = 0 (see
the note above), and vij = ωij = aij = εij = 0.
The following subsections are devoted to numerical calculation of Di and Bi
matrices for the elastic plate, as well as the Dj and Bj matrices for the rigid body.

3.3. CALCULATION OF MATRICES DJ AND BJ FOR RIGID BODY

From one perspective, the velocity vector of the center of mass of the body is
vj = ṙj , but it may also be calculated in Equation (18) as vj = Dj ėj . Taking into
account the structure of vector ej shown in (17), we find that
ėj = {ṙTj α̇ T }T
196 W.-S. YOO ET AL.

and obtain the value of the 3 × 6Dj matrix in the form

Dj = [I O] (30)
with identity and zero 3 × 3 matrices I and O, respectively.
The structure of the matrix Bj as well as of the vector of angular velocity ωj
depends on which set of orientation angles α is used. For example, let us consider
the Cardan angles α1 , α2 , α3 . The matrix of direct cosines expressed in terms of
these angles is a product of three simple rotation matrices:
   
1 0 0 c2 0 s2 c3 −s3 0
Aj =  0 c1 −s1   0 1 0   s3 c3 0  ,
0 s1 c1 −s2 0 c2 0 0 1
where the symbols ck = cos αk and sk = sin αk are used.
The skew-symmetric tensor of angular velocity ω̃j is defined by deriving of the
rotation matrix with respect to time, after which the vector of angular velocity can
be found:
 
0 −ω3 ω2
ω̃j = Ȧj ATj =  ω3 0 −ω1 
−ω2 ω1 0
    
 ω1  1 0 s2  α̇1 
⇒ ωj = ω2 =  0 c2 −s1 c2  α̇2 = Bα α̇.
   
ω3 0 s1 c1 c2 α̇3
To express this vector in the form ωj = Bj ėj , as required in Equation (18), we
introduce the 3 × 6 matrix
Bj = [O Bα ]. (31)
The last step requires us to calculate the vectors a and ε contained in Equa-
tion (18):
aj = Ḋj ėj = 0,
 
 c2 α̇2 α̇3 
ε j = Ḃj ėj = Ḃα α̇ = −s1 α̇1 α̇2 + s1 c2 α̇2 α̇3 − c1 c2 α̇1 α̇3 .
 
+c1 α̇1 α̇2 − c1 s2 α̇2 α̇3 − s1 c2 α̇1 α̇3

3.4. CALCULATION OF D AND B MATRICES FOR ELASTIC PLATE

Since the position ri of the body attachment point in Figure 3 is defined by Equation
(4), it is evident that vi = Sėi ; hence
Di = S.
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 197

Calculating the matrix Bi is more cumbrous, so we should start from definition

of the rotation matrix Ai . It can be constructed using three orthonormal vectors
along the xi , yi , and zi axes in Figure 3. We can compute two slope vectors
∂r ∂S
τi = = Si e, i = {1, 2}, where Si = . (32)
∂pi ∂pi
Although these vectors are, in general, not orthogonal and not one unit in length,
we can still apply the Gram–Schmidt orthogonalization to them:

τ T1 τ 2
τ ∗1 = τ 1 , τ ∗2 = τ 2 − τ 1.
τ T1 τ 1

The latter two vectors τ ∗1 and τ ∗2 and the normal vector τ ∗3 = τ 3 = τ̃ 1 τ 2 combine
to form the orthogonal rotation matrix
 ∗ 
τ1 τ ∗2 τ ∗3
Ai = .

τ ∗1

τ ∗2

τ ∗3

This value can be used easily in constraint equations (25) and (26), but it is too
bulky to derivate it in order to obtain angular velocity. However, we can assume
that the slope vectors τ 1 and τ 2 are virtually unit-length and orthogonal, because
the plate material has a very high Young’s modulus of approximately 1011 Pa. That
is why we approximate the value of the rotation matrix for its further derivation:

Ai ≈ [τ 1 τ 2 τ 3 ].

Thus, the skew-symmetric tensor of angular velocity can be expressed as

 T 
τ1
ω̃i = Ȧi ATi = [τ̇ 1 τ̇ 2 τ̇ 3 ]  τ T2  = τ̇ 1 τ T1 + τ̇ 2 τ T2 + (τ̃ 1 τ̇ 2 − τ̃ 2 τ̇ 1 )τ T3 ,
!" #
τ T3 τ̇ 3

using the newly introduced time-derivatives of the slope vectors (32):

τ̇ i = Si ė, i = {1, 2}.

This tensor is then vectorized the same way as in Equation (26):

1
3
ωi = ι̃m ω̃i ιm
2 m=1
 
1 3
 
= ι̃m τ̇ 1 τ T1 ιm +τ̇ 2 τ T2 ιm +(τ̃ 1 τ̇ 2 − τ˜2 τ̇ 1 ) τ T3 ιm  .
2 m=1 !" # !" # !" #
τ 1m τ 2m τ 3m
198 W.-S. YOO ET AL.

Each of the marked scalar terms τ km is the& m-th component of vector τ k . Fur-
ther, after placing ιm in brackets, we find that 3m=1 ι̃m τkm = ι̃k , so the expression
becomes
1
ωi = ((τ̃ 1 − τ̃ 3 τ̃ 2 )τ̇ 1 + (τ̃ 2 + τ̃ 3 τ̃ 1 )τ̇ 2 ).
2
In order to obtain the final result, we simplify the latter formula by applying the
well-known identity ãb̃ = baT − (aT b)I and recalling the orthogonal properties
τ T3 τ 2 = τ T3 τ 1 = 0. The angular velocity may then be rewritten as

1
ωi = ((τ̃ 1 − τ 2 τ T3 )τ̇ 1 + (τ̃ 2 + τ 1 τ T3 )τ̇ 2 ).
2
The Jacobian matrix of angular velocity is expressed as follows:
1
Bi = ((τ̃ 1 − τ 2 τ T3 )S1 + (τ̃ 2 + τ 1 τ T3 )S2 ).
2
The vector of angular acceleration is also written as

ε i = ω̇i = Bi ëi + εi ,

1 ˙
ε i = Ḃi ėi = ((τ̃ 1 − τ̇ 2 τ T3 )τ̇ 1 + (τ̃˙ 2 + τ̇ 1 τ T3 + τ 1 τ̇ T3 )τ̇ 2 )
2
1
= (−(τ̇ 2 τ T3 + τ 2 τ̇ T3 )τ̇ 1 + (τ̇ 1 τ T3 + τ 1 τ̇ T3 )τ̇ 2 ).
2

4. Modeling of Damping Forces

In a preceding paper [13] devoted to verifying the ANCF for beams and com-
paring the results with experimental data, a mass matrix proportional damping
by Rayleigh [14, 22] was used to account air resistance forces. The vector of
generalized damping forces was calculated as follows:

Qd = αMė, (33)

with the mass matrix M and coefficient α, which was estimated from experimental
data.
When this model of damping forces was applied to our plate problem, it did not
lead to a good agreement with the experimental curves. In the authors’ opinion, this
is due to high influence of air resistance forces that are not linear in velocities as in
Equation (33) but they are quadratic in velocities. That is why we have to develop
the corresponding model of damping forces.
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 199

Figure 4. Cantilever beam under uniform loading.

4.1. AUXILIARY PROBLEM

Consider small vibrations of the clamped plate and represent its motion by one
degree of freedom e that is the deflection of the end point, in Figure 4. Then the
displacement fields of the beam centerline as well as the velocity field are described
by

z = z(p1 , p2 , e) = S(p1 , p2 )e,

v = ż(p1 , p2 , ė) = S(p1 , p2 )ė,

where S is the shape function, which corresponds to a static deflection of the beam
under uniform load:
1
S(p1 , p2 ) = (6(p1 /a)2 − 4(p1 /a)3 + (p1 /a)4 ).
3
The expression for kinetic energy of the plate can now be written as follows:

1 1
T = µ S 2 ė2 dP = M ė2
2 2
P

with the area mass density µ in kg/m2 and a mass coefficient


104
M =µ S 2 dP = M0 ,
405
P

where M0 = µab is the whole mass of the plate.

Assume that the damping force has a linear and a quadratic part in velocities

f (v) = α1 v + α2 v|v| (34)

with unknown coefficients α1 and α2 that should be identified from experimental

data.
200 W.-S. YOO ET AL.

The generalized damping force is calculated as follows:

 
Q =
d
Sf dP = S(α1 S ė + α2 S ė|S ė|) dP
P P

1
= α1 M ė + α2 S 2 |S| dP ė|ė|.
µ
P

Since S > 0 for ∀p1 ∈ [0 . . . a], we can easily calculate the integral
 
2336 0.19 0.74
S |S| dP =
2
S 3 dP = ab ≈ M0 ≈ M.
12285 µ µ
P P

The equation of motion of our one d.o.f. plate is

M ë + Ce + β1 M ė + β2 M ė|ė| = 0, (35)

where C is a stiffness coefficient and

α1 0.74α2
β1 = , β2 = . (36)
µ µ
After reducing Equation (35) in the mass coefficient M, we obtain equation of
a linear oscillator with nonlinear damping (ω02 = C/M)

ë + ω02 e + β1 ė + β2 ė|ė| = 0,

which should be solved using the initial conditions e(0) = A0 , ė(0) = 0.

Following Van der Pol’s method, we represent the solution in the form

e(t) ≈ A(t) cos ω0 t

with a variable amplitude

3πβ1 A0 e−β1 t /2
A(t) = .
8ω0 β2 A0 (1 − e−β1 t /2) + 3πβ1
In order to identify the coefficients β1 and β2 we should measure the amplitudes
Ak as well as time points tk from an experimental curve as shown in Figure 5.
Then we use the method of least squares


N
(β1 , β2 ) = (A(tk ) − Ak )T → min
k=0

and find the values β1 , β2 at first and calculate α1 , α2 after that from Equation (36).
Following this algorithm, parameters of the model of damping forces were
obtained as shown in Table I.
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 201

Figure 5. Damped oscillations.

Table I. Parameters of the model of damping forces for a free clamped plate.

Plate length, cm β1 , s−1 β2 , m−1 α1 , kg·s−1 m−2 α2 , kg·m−3

40 0.17 0.85 0.53 3.61

Thus, the structural equations of motion of the plate with damping forces taken
into account assume the form similar to (7):

Më + Qe + Qd = Qg ,

where Qd is the vector of generalized damping forces. To calculate it applying the

model (34) of damping forces for our plate element, the following steps are used:

1. calculate the velocity of the mass center of the element using formula
¯
vC = S̄ė,

see the value of matrix S̄¯ in Equation (9);

2. calculate average damping force

f = (α1 + α2
vC
)vC ,

using the coefficients α1 , α2 obtained as explained above; and

3. distribute this force to nodal variables

Qd = S̄¯ f.
T

Another variant of calculation generalized damping forces is to apply the rela-

tion

Qd = (β1 + β2
vC
)Mė

analogously to Equation (35) for the one d.o.f. model.

202 W.-S. YOO ET AL.

Figure 6. Scheme of experimental equipment.

5. Experimental Setup
5.1. EXPERIMENTAL SETUP AND RESULTS
The setup for this plate oscillation experiment is shown in Figure 6. The camera
follows the path of the target attached to the tip, but since the target is moving in
a three-dimensional space, the distance from the camera to the target is changing
when the deflection occurs. To minimize this kind of ‘visual distance error’, the
camera is placed as far away as possible from the target. With the camera positioned
10 m from the target, the maximum error is less than 2.5% when the deflection is
about 25 cm.

5.2. EXPERIMENTAL RESULTS

The front view and the side view of the plate oscillation are shown in Figure 7. The
position of the tip in space with a 260 g mass attached to it is shown in Figure 8,
and its x, y, and z positions with respect to time are presented in Figure 9.

5.3. CALCULATION OF YOUNG ’ S MODULUS

Geometrical parameters of the plate as well as its inertia parameters can be meas-
ured directly. On the contrary, the stiffness parameters were identified indirectly
using the data obtained from experimental measurement of frequencies of small
oscillations of the plate. Firstly, the plate is cut into several strips to make slender
beams as shown in Figure 10. From the free vibration test of the clamped beam,
the first frequency of the strip is measured.
In this case, its motion is approximately the same as for the cantilever beam
[13]. The stiffness of the plate (i.e., Young’s modulus E), which is the most im-
portant material property, is calculated by an indirect method rather than a tensile
test.
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 203

Figure 7. Motion of the 0.4 × 0.2 m plate with 0.26 kg weight.

Figure 8. Position of the tip in space.

Figure 9. Positions of the tip with respect to time.

Figure 10. Strips of plates to measure Young’s modulus.

204 W.-S. YOO ET AL.

Table II. Geometrical, inertia and stiffness parameters of the plate.

Length, width and Mass Density of Cross Inertia First Young’s

thickness of the plate, of the the material section area moment of eigen- modulus
mm plate ρ = m/abh, of the cross the cross frequency E, MPa
m, kg kg/m3 section section ω, Hz
J = bh3 /12, J = bh3 /12,
mm4 mm4
a b h

400 200 0.4 0.242 7554 80 1.07 2.02 189

Table III. Inertia parameters of the weights.

Plate sizes, Mass of Shift of the center Principal moments of

a × b, m weight, kg of mass, mm inertia, kg·mm2
m x y z Jx Jy Jz

0.4 × 0.2 0.26 1.5 1.5 18.7 82.1 72.9 149

The first natural frequencies of a cantilever beam are given by

'
EJ
ωk = βk2 , (37)
ρAl 4
where E is Young’s modulus to be found from this formula, A and J are cross
section area and inertia moment of the beam cross section; ρ is the density of the
plate material and l is the beam length. For our plate l = a. The coefficients βk
have the values β1 = 1.875 for the first frequency and β2 = 4.694 for the second
one.
Table II presents the values of the parameters, which were obtained for the plate
employed in the experiment. Note that in this experiment no weight was attached
to the plate so that we observed free oscillations of the clamped plate.
The values of the Poisson ratio were not measured in the experiments. It was
just assumed that v21 = 0.3 in the longitudinal p1 -direction.

5.4. INERTIA PROPERTY OF THE ATTACHED MASS

In order to obtain large oscillations of the plate, the steel weight shown in Figure 11
was used. Inertia parameters of the weight employed in the experiments are listed
in Table III.
The principal moments of inertia were computed with the help of AutoCAD
program package using a three-dimensional model of the weights shown in Fig-
ure 11.
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 205

Figure 11. Scheme of attaching the weight to the plate.

Figure 12. Test of convergence.

6. Comparison of Results
Numerical results presented here were obtained with the help of the program pack-
age Universal Mechanism (UM), which can be found in [9] as well as online at
www.umlab.ru.

6.1. TEST ON CONVERGENCE

In this section, we studied influence of the amount of finite elements on conver-
gence of simulation results. The subject is the plate of length 0.4 × 0.2m with
attached weight of mass 0.26 kg. The plate was subdivided into 2 × 2, 4 × 2, 6 × 2
and 8 × 2 elements and simulation results are presented in Figure 12. One can
see that the two latter schemes show almost the same results that is why all other
simulations below will employ 6 × 2 elements.
206 W.-S. YOO ET AL.

Figure 13. Free oscillation of the 0.4 × 0.2 m plate.

Figure 14. Forced oscillation of 0.4 × 0.2 m plate with 0.26 kg weight.

6.2. FREE OSCILLATIONS OF A PLATE

Figure 13 demonstrates experimental data and simulation data concerning vertical
displacements of endpoint of free clamped plate vs. time.
We can comment that agreement of the results is quite good. Discrepancies in
static and dynamic behavior are not significant and may be explained by inexact
description of initial conditions.

6.3. FORCED OSCILLATION OF A PLATE WITH WEIGHT

Results of experiments and simulations for a clamped plate with attached weight
are presented in Figure 14. They show the vertical displacements of the attach-
ment point of the plate vs. time. The agreement of the results can also be nicely
characterized for the 0.4 × 0.2 m plate.
LARGE DEFLECTION ANALYSIS OF A THIN PLATE 207

7. Conclusion
We considered large oscillation of a thin clamped plate with the attached end-point
rigid-body weight. We performed both experiments and numerical modeling of the
problem with taking into account resistance forces. To the authors’ knowledge, it
is the first paper to compare experiments and simulations in the absolute nodal
coordinate formulation. In our opinion, some new results have been obtained in
this investigation:

• The proposed beam × beam plate element showed reasonable results for large
displacement plate problem with ANCF.
• New formulation of constraint equations that allow simulating hybrid systems
containing elastic ANCF plate elements and rigid bodies was suggested.
• Computer simulation of a large deformation plate was well matched to the
experimental results with accurate material properties measured from the ex-
periments.

Acknowledgements
The authors would like to thank the Ministry of Science and Technology of
Korea for its financial support through a grant (M1-0203-00-0017) under the
NRL (National Research Laboratory) project as well as the Russian Founda-
tions RFBR/RFFI (02-01-00364-A, 02-01-06098-MAC, 03-01-06487-MAC) and
URBR/UniRos (UR.04.01.046).

References
1. Shabana, A.A., ‘Flexible multibody dynamics: Review of past and recent developments’,
Multibody System Dynamics 1, 1997, 189–222.
2. Shabana, A.A., Dynamics of Multibody Systems, 2nd edition, Cambridge University Press,
Cambridge, 1998.
3. Berzeri, M. and Shabana, A.A., ‘Development of simple models for the elastic forces in the
absolute nodal co-ordinate formulation’, Journal of Sound and Vibration 235(4), 2000, 539–
565.
4. Omar, M.A. and Shabana, A.A., ‘A two-dimensional shear deformation beam for large rotation
and deformation’, Journal of Sound and Vibration 243(3), 2001, 565–576.
5. Mikkola, A.M. and Shabana, A.A., ‘A new plate element based on the absolute nodal coordinate
formulation’, in Proceedings of ASME 2001 DETC, Pittsburgh, PA, 2001.
6. Dmitrochenko, O.N., ‘Methods of simulating dynamics of hybrid multibody systems with tak-
ing into account geometrical nonlinearity’, in Dynamics, Strength and Reliability of Transport
Machines, B.G. Keglin (ed.), Bryansk State Technical University, Bryansk, 2001, 24–34 [in
Russian].
7. Dmitrotchenko, O.N., ‘Efficient simulation of rigid-flexible multibody dynamics: Some imple-
mentations and results’, in Proceedings of NATO ASI on Virtual Nonlinear Multibody Systems,
Vol. 1, W. Schiehlen and M. Valášek (eds.), Prague, 2002, 51–56.
208 W.-S. YOO ET AL.

8. Dmitrochenko, O.N. and Pogorelov, D.Yu., ‘Generalization of plate finite elements for absolute
nodal coordinate formulation’, Multibody System Dynamics 10(1), 2003, 17–43.
9. Pogorelov, D., ‘Some developments in computational techniques in modeling advanced mech-
anical systems’, in Proceedings of IUTAM Symposium on Interaction between Dynamics
and Control in Advanced Mechanical Systems, D.H. van Campen (ed.), Kluwer Academic
Publishers, Dordrecht, 1997, 313–320.
10. Pogorelov, D., ‘Differential-algebraic equations in multibody system modeling’, Numerical
Algorithms 19, 1998, 183-194.
11. Shabana, A.A. and Yakoub, R.Y., ‘Three dimensional absolute nodal coordinate formulation
for beam elements: Theory’, Journal of Mechanical Design 123, 2001, 606–621.
12. Takahashi, Y. and Shimizu, N., ‘Study on elastic forces of the absolute nodal coordinate
formulation for deformable beams’, in ASME Proceedings of Design Engineering Technical
Conference, VIB-8203, Las Vegas, NV, 1999.
13. Yoo, W.S., Lee, J.H., Sohn, J.H., Park, S.J., Dmitrotchenko, O.N. and Pogorelov, D., ‘Large
oscillations of a thin cantilever beam: Physical experiments and simulation using the absolute
nodal coordinates’, Nonlinear Dynamics 34, 2003, 3–29.
14. Takahashi, Y., Shimizu, N. and Suzuki, K., ‘Introduction of damping matrix into absolute nodal
coordinate formulation’, in Proceedings of the 1st Asian Conference on Multibody Dynamics,
Iwaki, Fikushima, 2002, 33–40.
15. Schwertassek, R., ‘Flexible bodies in multibody systems’, Computational Methods in Mechan-
ical Systems 161, 1997, 329–363.
16. Szilard, R., Theory and Analysis of Plates. Classic and Numerical Methods, Prentice-Hall,
Englewood Cliffs, NJ, 1974.
17. Palii, O.M. (ed.), Handbook on Ship Structural Mechanics 2, Shipbuilding Publishers, Lenin-
grad, 1982 [in Russian].
18. Bronstein, I.N. and Semendyayev, K.A. (eds.), Handbook on Mathematics for Engineers and
Students, State Techn.-Theor. Publ. (GITTL), Moscow, 1957 [in Russian].
19. Zienkiewicz, O.C. and Taylor, R.L., The Finite Element Method, 4th edition, Volume 2: Solid
and Fluid Mechanics, McGraw-Hill, New York, 1991.
20. Timoshenko, S.P. and Woinowsky-Krieger, S., Theory of Plates and Shells, 2nd edition,
McGraw-Hill, New York, 1991.
21. Craig, R.R., Structural Dynamics.
22. Bathe, K.-J., Finite Element Procedures, Prentice Hall, Englewood Cliffs, NJ, 1996.