Sie sind auf Seite 1von 16

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng 2003; 57:669684 (DOI: 10.1002/nme.696)


An algorithm using co-ordinate transformation for free
vibration analysis of fully incompressible materials
Yuji Kohsetsu
,
National Space Development Agency of Japan, Tsukuba Space Center, 2-1-1 Sengen, Tsukuba,
Ibaraki 305-8505, Japan
SUMMARY
This paper presents an algorithm for conducting eigenvalue analysis of fully incompressible isotropic
materials (a Poissons ratio of exactly 0.5). This algorithm is based on the concept of co-ordinate
transformation dened by eigenvectors with positive eigenvalues of the system stiness matrix. The
transformation to the new co-ordinates of orthogonal bases in the system potential energy can reduce
the size of the system equations. The original eigenvalue problem is reduced to a smaller one that can
be solved. This algorithm possesses two drawbacks. One is the increase of system degrees of freedom
(DOF) due to the mixed method for incompressible problems; the other is due to additional eigen-
value calculation. However, the present algorithm is inherently able to compensate for the above two
drawbacks by reducing the DOF. Numerical demonstrations indicate that the exact transformation yields
accurate results that converge to analytical results. They also indicate that an approximated transfor-
mation retaining only several per cent of all the eigenvectors produces accurate results in the lower
frequency range that is generally of interest in engineering analyses. Therefore, the proposed algorithm
is judged to be an ecient procedure for eigenvalue problems of fully incompressible materials. This
co-ordinate-transformation scheme, derived from the system stiness matrix, will be applicable to static
incompressible problems. Copyright ? 2003 John Wiley & Sons, Ltd.
KEY WORDS: incompressible materials; free vibration analysis; co-ordinate transformation algorithm;
size reduction by approximation; mixed method nite element
1. INTRODUCTION
Solid-rocket motor propellants are incompressible isotropic materials used in aerospace engi-
neering. Solid propellants play a major role in structural dynamics of the total solid-rocket
motor. The present algorithm for eigenvalue analysis of incompressible materials was for-
mulated in free-vibration analysis of solid propellants. The conventional displacement-based
nite element method has a signicant limitation originating from the stress-to-strain relations

Correspondence to: Yuji Kohsetsu, National Space Development Agency of Japan, Tsukuba Space Center, 2-1-1
Sengen, Tsukuba, Ibaraki 305-8505, Japan.

E-mail: kohsetsu.yuji@nasda.go.jp
Received 16 October 2001
Revised 2 July 2002
Copyright
?
2003 John Wiley & Sons, Ltd. Accepted 13 August 2002
670 Y. KOHSETSU
in incompressible materials. In such circumstances, various nite element formulations for in-
compressible materials have been developed, as reviewed by Gadala [1], Sussmann and Bathe
[2], Hughes [3], Szabo and Babu ska [4], and Zienkiewicz and Taylor [5].
These methods deal with fully incompressible problems in an asymptotic manner, formulat-
ing them as nearly incompressible problems with a Poissons ratio of 0.4999 as an example.
The convergence of the solution is then tested when the value of Poissons ratio converges
asymptotically toward 0.5. Asymptotic methods require verifying the numerical solution when
the submitted Poissons ratio diers from exactly 0.5. Therefore, these procedures can be
considered as indirect algorithms in terms of asymptotic properties. A typical asymptotic pro-
cedure is shown in Appendix A.
This paper proposes a direct algorithm, which implies the capability to directly calculate
fully incompressible problems, i.e. a Poissons ratio of exactly 0.5 is input into the calculation.
The present algorithm is based on the Lagrange multiplier method by Herrmann [6] and
consists of the following three processes.
(i) Formulate equations of free-vibration motion of incompressible isotropic materials by
the mixed nite element method.
(ii) Transform the co-ordinates of the system equations of motion using a matrix constructed
from eigenvectors of the system stiness matrix; during this process, the system equa-
tions of motion can reduce the degrees of freedom (DOF).
(iii) Calculate a conventional eigenvalue problem in the transformed co-ordinates with re-
duced DOF.
The second process plays the most important role in the present algorithm. This role is estab-
lished by using a specic co-ordinate transformation from the original physical co-ordinates to
certain generalized co-ordinates with orthogonal bases of the potential energy functional. The
transformation matrix is dened by combining eigenvectors with positive eigenvalues in the
system stiness matrix. This requires additional eigenvalue calculation, which will increase
computing loads. However, co-ordinate transformation has the potential to reduce the system
DOF by introducing an approximation. The approximation scheme truncates less signicant
eigenvectors while retaining more signicant ones. Truncation of less signicant DOFs will
enable reduction of the column size of the co-ordinate-transformation matrix. If the approxi-
mation scheme is adopted, the eigenvalue analysis will become a much smaller problem than
the original. Kolata [7] has presented another co-ordinate-transformation scheme associated
with the system mass matrix, but does not investigate the reduction in the size of the system
equations.
Numerical examples of an axisymmetic natural vibration analysis of a fully incompressible
isotropic cylinder are presented to demonstrate the capability of the proposed algorithm. This
paper describes the process below.
(a) Formulate system equations of motion based on the Herrmann method [6] using an
axisymmetric ring with a four-node section of the bilinear element.
(b) Create co-ordinate-transformation matrices of exact (fully retained) and approximated
(truncated) terms in order to transform the above equations of motion.
(c) Calculate eigenvalues of the transformed system equation to obtain natural frequencies
and natural modes, both by full transformation and by approximated transformation.
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
FREE VIBRATION ANALYSIS OF FULLY INCOMPRESSIBLE MATERIALS 671
(d) Compare the exact and approximated results obtained by the above proposed algorithm
with the analytical solutions.
In this demonstration study, a standard bilinear shape function is applied both to the dis-
placement eld and the pressure eld; resulting pressure is therefore discontinuous between
the elements. We note that the InfSup condition has been established for the mixed nite
element method [8]. A four-node element with equal order of displacement and pressure
eld (4}4 u}p element) does not satisfy the InfSup condition. However, this is permissible
because a structural dynamics study does not necessarily require pressure continuity. The nu-
merical examples demonstrate that the results obtained by an exact co-ordinate-transformation
scheme converge to the analytical solutions. Other examples suggested that degradation of
the accuracy due to truncation is negligible for engineering analyses. Based on these ndings,
we believe the proposed algorithm is an ecient one for analysing eigenvalue problems of
incompressible materials.
2. A CO-ORDINATE-TRANSFORMATION ALGORITHM
2.1. Formulation of equations of motion
Free-vibration equations of motion of fully incompressible solids of revolution are formulated
below using the Lagrange multiplier, which is a mixed nite element method.
_
_
K
uu
K
up
K
T
up
0
_
_
_
u
p
_
+
_
M
uu
0
0 0
__
u
0
_
=0 (1)
Here, u is the nodal displacement vector, {u
ri
, u
0i
, u
:i
} (see Figure 1); p the nodal hydrostatic
pressure vector, {
i
}; u a vector of nodal acceleration; K
uu
the (n n) matrix; K
up
the (n !)
u

r
u
r
uz
z
=
uu
==

0
Axisymmetric
displacement;

z
Figure 1. Cylinder and co-ordinate system.
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
672 Y. KOHSETSU
matrix; M
uu
the (n n) matrix; m the number of nodes; n the degrees of freedom of u
(n =3m) and ! the degrees of freedom of p.
2.2. Denition of co-ordinate transformation
The rst term of Equation (1) is traced back to the potential energy (PE) dened below, which
has a quadratic form in the physical co-ordinates of the displacement and stress determined
by the mixed nite element method.
PE=
1
2
_
u
p
_
T
_
K
uu
K
up
K
T
up
0
__
u
p
_
(2)
A quadratic form is transformed to a diagonal form by orthogonal transformation as
x
T
Ax =

i
z
i
q
2
i
(3)
Here, z
i
is an eigenvalue of A, and q
i
is the orthogonal co-ordinate dened by x =q
where is an orthogonal matrix.
Therefore, potential energy (2) can be transformed to a diagonal form as follows:
PE=
1
2
_
u
p
_
T
_
K
uu
K
up
K
T
up
0
__
u
p
_
=
1
2
n
[

i=1
z
i
q
2
i
, (z
i
0) (4)
where z
i
(i =1, 2, . . . , n
[
) is a positive eigenvalue of the stiness matrix and q
i
is its cor-
responding orthogonal base. The number of positive eigenvalues is equal to n
[
, the rank of
the stiness matrix, which is equal to n the rigid bodys DOF. It is essential to retain
only the positive eigenvalues in (4) for physical interpretation. The non-positive eigenvalues
of the stiness matrix are thus ignored. The ignored eigenvalues in (4) are zero eigenvalues
and negative eigenvalues. The number of non-positive eigenvalues is equal to l, the DOF of
{p}, which traces back to the Lagrange multiplier.
Equation (4) is rewritten when eigenvalues are rearranged in increasing order as follows:
PE=
1
2
n
[

i=1
z
i
q
2
i
, (0z
1
z
2
z
i
z
n
[
) (5)
Thus, (5) leads to the co-ordinate transformation dened by
_
u
p
_
=[[
1
, [
2
, . . . , [
i
, . . . , [
n
[
]{q
i
} =[]{q} (6)
where [] is the orthogonal matrix whose ith column is [
i
, the ith eigenvector of the system
stiness matrix. Transforming the system equation of (1) by (6) yields (7), whose DOF is
n
[
. The DOF of (7) is equal to that of the displacement-based nite element method if it
were applied.
[

K]{q} + [

M]{ q} =0 (7)
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
FREE VIBRATION ANALYSIS OF FULLY INCOMPRESSIBLE MATERIALS 673
where
[

K] =[]
T
_
K
uu
K
up
K
T
up
0
_
[] (reduced stiness matrix: n
[
n
[
)
and
[

M] =[]
T
_
M
uu
0
0 0
_
[] (reduced mass matrix: n
[
n
[
)
2.3. Approximation algorithm
An approximation algorithm can be introduced by truncating the higher-order eigenvectors in
the co-ordinate transformation (6). This truncation is based on the concept that engineering
analysis is generally interested only in a relatively small number of lower frequencies. The
lower frequencies will relate to the lower stiness of the system. Eigenvalue z
i
in (5) means
the generalized stiness, so the higher-order eigenvalues have less signicant eects on the
lower frequencies of the system. Consequently, we truncate the higher-order eigenvectors in
(6) to obtain the approximated transformation matrix in
_
u
p
_

=[[
1
, [
2
, . . . , [
i
, . . . , [
k
]{q
i
} =[
k
]{q}, (kn
[
) (8)
Accordingly, when the equations of motion (1) are transformed using (8), the DOF of the
new equations becomes much smaller than that of (7) (kn
[
).
2.4. Implementation of the algorithm
Implementing this algorithm in a computer program is described next. Boundary conditions
are applied to (1), and the constrained DOFs are removed from (1). For simplicity of notation,
the reduced system equations of motion are expressed by (1). Accordingly, n
[
and ! become
DOFs of {u} and {p}, respectively. The eigenvalue analysis to obtain positive eigenvalues
only from the smallest of the system stiness matrix in (1) leads to a transformation matrix
expressed by
_
u
p
_
=[]{q} (9)
where [] is an (n
[
+ !) n
[
matrix, whose ith column is [
i
, the ith eigenvector with
positive eigenvalue in increasing order of the system stiness matrix. The number of positive
eigenvalues is equal to rank n
[
of the system stiness matrix of (1).
Equation (1) is co-ordinate-transformed to (10) by (9). The DOF of (10) is n
[
.
[[]{q} + [

M]{ q} =0 (10)
where
[

M] =[]
T
_
M
uu
0
0 0
_
[]
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
674 Y. KOHSETSU
Table I. Geometry and mechanical properties of example cylinder.
Geometry Mechanical properties
Radius (mm) 1000 Poissons ratio, v 0.5
Length (mm) 2000 Shear modulus, j (MP
a
) 27.0
Density, j (kg}m
3
) 1.8 10
3
Another co-ordinate transformation is dened as
{q} =[[]
1}2
{r} (11)
Equation (10) is transformed to (12) using (11) as follows:
[I]{r} + [

M]{ r} =0 (12)
where
[

M] =[[]
1}2
[

M][[]
1}2
To formulate the approximation scheme, less signicant (higher-order) eigenvectors are
truncated in (9). If the number of retained eigenvectors in (9) is k, the DOF of (12) becomes
k (kn
[
).
3. NUMERICAL EXAMPLES
3.1. Cylinder to be analyzed
Axisymmetric free vibrations of a fully incompressible cylinder are calculated analytically and
numerically in order to demonstrate the capability of the algorithm presented in
Section 2. Figure 1 shows the cylinder and its co-ordinate system. The cylinder geometry
and its mechanical properties are shown in Table I. Boundary conditions are specied as
follows:
r = a, u
r
(a, :) = 0 and t
r :
(a, :) = 0
: = 0, u
:
(r, 0) = 0 and t
:r
(r, 0) = 0
: = !, u
:
(r, !) = 0 and t
:r
(r, !) = 0
where t
r:
and t
:r
are traction stress.
3.2. Analytical solutions
Analytical solutions for axisymmetric natural vibrations of an incompressible cylinder are ob-
tained from a constrained problem of null volumetric strain based on the classical longitudinal
vibration of a cylinder [9]. The partial dierential equations of c
0
for the cylinder are derived
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
FREE VIBRATION ANALYSIS OF FULLY INCOMPRESSIBLE MATERIALS 675
0
z

r
(-1,-1)
1
(1,-1)
4
(-1, 1)
2

(1,1)
3
Figure 2. Axisymmetric ring element with a four-node section.
below (details of the solution are shown in Appendix B):
_
c
2
cr
2
+
1
r
c
cr
+
c
2
c:
2

1
r
2
_
c
0

j
j
c
2
c
0
ct
2
=0 (13)
where
c
0
(r, :, t) =
1
2
_
cu
r
c:

cu
:
cr
_
The analytical solutions are obtained as
[
mn
=
1
2
=
1
2
_
j
j
_
z
2
n
+ (:
m
}a)
2
u
r
(r, :) = A}z
n
J
1
(k
n
r) cos z
n
:
u
:
(r, :) = A}k
n
J
0
(k
n
r) sin z
n
:
(14)
where [
mn
is the natural frequency, :
m
the mth root of J
1
(:
m
) =0, Bessel function of the rst
order, k
2
n
=j
2
}j z
2
n
and z
n
=n}!.
3.3. Results by exact transformation scheme
The axisymmetric ring element used in this calculation is depicted in Figure 2. This ring
element has a four-node section of the bilinear shape function. The stiness matrix and mass
matrix for this element are formulated as (15) based on the modied HellingerReissner
Principle [10], which is equivalent to the Herrmann method. The derivation is shown in
Appendix C is as follows:
_
K
uu
K
up
K
T
up
K
pp
__
u
p
_
+
_
M
uu
0
0 0
__
u
p
_
=0 (15)
We note that [K
pp
] becomes a null matrix when Poissons ratio is 0.5. The total system equa-
tions of motion are constructed as in (1). After boundary conditions specied in Section 3.1
are applied to (1), the rst eigenvalue analysis of the stiness matrix is performed to generate
the co-ordinate transformation matrix of (6) without approximation. The Householder method,
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
676 Y. KOHSETSU
Number of elements
anti-symmetric
u
= 1000 mm r
x 32 32
40 x 40
16
25
20
x 25
x
x
16
20
n n
CL C
x
10
x 8 8
x 10
x 5 5
half model
for
symmetric
conditon
boundary
boundary
condition
CCL
r
u
z
l

=

2
0
0
0

m
m
Figure 3. Models for numerical calculation.
Table II. Natural frequencies (Hz) of axisymmetric vibration of incompressible cylinder (v =0.5).
Numerical solution (number of elements)
Mode Analytical
no. (m, n) 5 5 8 8 10 10 16 16 20 20 25 25 32 32 40 40 solution
1 (1, 1) 2.5996 2.5719 2.5649 2.5570 2.5551 2.5539 2.5531 2.5526 2.5526
2 (1, 2) 3.1294 3.0829 3.0723 3.0609 3.0582 3.0565 3.0554 3.0547 3.0542
3 (1, 3) 3.9371 3.8143 3.7881 3.7604 3.7541 3.7501 3.7473 3.7458 3.7437
4 (2, 1) 4.8433 4.6057 4.5456 4.4771 4.4608 4.4503 4.4429 4.4388 4.4315
5 (1, 4) 4.8602 4.6571 4.6127 4.5655 4.5547 4.5479 4.5431 4.5404 4.5363
6 (2, 2) 5.1946 4.9178 4.8537 4.7833 4.7670 4.7564 4.7491 4.7450 4.7382
7 (2, 3) 5.7828 5.4193 5.3418 5.2600 5.2414 5.2295 5.2214 5.2168 5.2094
8 (1, 5) 6.1228 5.6433 5.5470 5.4473 5.4250 5.4108 5.4010 5.3955 5.3866
9 (2, 4) 6.5237 6.0622 5.9663 5.8663 5.8438 5.8295 5.8196 5.8140 5.8052
10 (1, 6) 7.2729 6.6460 6.5079 6.3620 6.3289 6.3077 6.2932 6.2850 6.2712
11 (3, 1) 7.6755 6.8852 6.7347 6.5059 6.4484 6.4114 6.3856 6.3710 6.3452
12 (2, 5) 8.1315 6.9711 6.7500 6.5824 6.5487 6.5275 6.5128 6.5047 6.4914
Note: (m, n) means the mode shape; m: Order of root in the radial direction and n: Order of root in the axial
direction.
the bisection method, and the inverse iteration method are used in eigenvalue analysis. The
co-ordinate-transformed equations of motion of (7) are solved as ordinary eigenvalue prob-
lems, whose schemes are the same as that for the eigenvalue analysis of the stiness matrix.
Numerical calculations are conducted for eight models, which are divided into 5 5, 8 8,
10 10, 16 16, 20 20, 25 25, 32 32, and 40 40 elements as depicted in Figure 3.
Table II compares natural frequencies of the numerical results and the analytical solutions.
Figure 4 depicts convergence of the natural frequencies produced by the numerical analysis to
the analytical solutions. In the lower frequency range, the numerical results agree well with the
analytical solutions. Figure 5 shows the rst natural vibration modes obtained numerically by a
40 40 model. The mode shapes of numerical and analytical results are compared in Table III.
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
FREE VIBRATION ANALYSIS OF FULLY INCOMPRESSIBLE MATERIALS 677
Number of Elements
f
,
n
u
m
/
f
,
t
h
e
Mode
1 (1,1)
2 (1,2)
3 (1,3)
4 (2,1)
5 (1,4)
6 (2,2)
20 2000 1000 100
1.00
1.01
1.02
1.03
1.04
1.05
1.00
1.01
Figure 4. Convergence of natural frequencies by the presnet analysis.
Table III. Comparison of rst mode shapes between numerical
(40 40 model) and analytical results.
Mode ratio Numerical Analytical
u
:
(r =1000)}u
:
(r =0) at : =1000 2.4863 2.4828
u
:
(: =1500)}u
:
(: =1000) at : =0 0.70773 0.70710
u
r
(: =2000)}u
r
(: =1500) at r =500 1.4133 1.4142
These agreements indicate that the proposed algorithm using the co-ordinate transformation
of the stiness matrix is accurate enough for eigenvalue problems with full incompressibility.
3.4. Results by approximated transformation scheme
The approximation procedure is incorporated in the eigenvalue analysis of the stiness matrix.
The k eigenvectors with positive eigenvalues from the lowest are retained and generate an
approximated co-ordinate transformation matrix (8) with a reduced column size of k. Conse-
quently, the eigenvalue problem of (7) can decrease the DOF from n
[
of the original system
to k (kn
[
). The number of eigenvectors retained is varied as a parameter in the numerical
calculations of the two models. One has 20 20 elements, and the other has 40 40 elements.
There are a total of 800 eigenvectors for the anti-symmetric case and 798 for the symmetric
case for the 20 20 elements. There are a total of 3200 eigenvectors for the anti-symmetric
case and 3198 for the symmetric case for the 40 40 elements. Tables IV(a) and IV(b) show
the natural frequencies obtained by this approximation scheme for the 20 20 elements and
the 40 40 elements. Figures 6(a) and 6(b) show the accuracy of the approximated lower nat-
ural frequencies when the number of retained eigenvectors is varied for the 20 20 elements
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
678 Y. KOHSETSU
Figure 5. Mode vector of rst mode ([
1
=2.5526 Hz) by 40 40 model.
and the 40 40 elements. The resulting several per cent of the total number of eigenvectors
can produce the lower six natural frequencies within 1% relative errors. Figures 6(a) and
6(b) suggest that our proposed algorithm can greatly reduce the size of the problem. Based
on this numerical demonstration, the proposed scheme is judged to be an ecient algorithm
for analysing eigenvalue problems with full incompressibility.
4. CONCLUSIONS
This paper presents an algorithm for the mixed nite element method based on the concept
of a co-ordinate transformation in order to analyse eigenvalue problems of fully incompress-
ible materials. The essentials of this algorithm presented are a co-ordinate transformation
dened by the system stiness matrix and its approximation scheme. The proposed proce-
dure has two drawbacks, increased system DOF in its formulation by the mixed method and
additional eigenvalue calculation for the co-ordinate transformation. However, this algorithm
is able to compensate for these drawbacks through the co-ordinate transformation process
by reducing the systems DOF. Therefore, this algorithm is judged to be an ecient and a
common procedure for eigenvalue problems with incompressibility based on the co-ordinate
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
FREE VIBRATION ANALYSIS OF FULLY INCOMPRESSIBLE MATERIALS 679
Table IV(a). Approximated natural frequencies (Hz) by 20 20 element model.
Approximated solution with number of the retained eigenvectors
Mode Exact solution
no. (m, n)

20 25 30 40 80 200 400 640 800}798

1 (1, 1) 2.5648 2.5627 2.5559 2.5564 2.5563 2.5561 2.5553 2.5551 2.5551
2 (1, 2) 3.0773 3.0771 3.0619 3.0603 3.0598 3.0592 3.0587 3.0582 3.0582
3 (1, 3) 4.1696 3.8161 3.7634 3.7581 3.7565 3.7553 3.7547 3.7541 3.7541
4 (2, 1) 5.0926 4.6941 4.5414 4.4736 4.4648 4.4625 4.4612 4.4609 4.4608
5 (1, 4) 5.3307 5.3307 4.5836 4.5677 4.5600 4.5566 4.5554 4.5547 4.5547
6 (2, 2) 5.5389 5.5386 4.8843 4.7852 4.7719 4.7693 4.7674 4.7670 4.7670
7 (2, 3) 6.7759 6.3861 5.4550 5.2810 5.2480 5.2444 5.2442 5.2414 5.2414
8 (1, 5) 11.096 6.7221 5.7474 5.4965 5.4325 5.4273 5.4259 5.4250 5.4250
9 (2, 4) 12.539 7.4449 6.1776 5.9117 5.8548 5.8480 5.8449 5.8438 5.8438
10 (1, 6) 13.652 8.0014 6.8420 6.4874 6.3508 6.3332 6.3298 6.3289 6.3289
11 (3, 1) 14.358 9.6685 8.0369 6.7697 6.4620 6.4537 6.4493 6.4485 6.4484
12 (2, 5) 14.700 14.365 8.9429 7.1317 6.5642 6.5541 6.5503 6.5488 6.5487

(m, n) means the mode shape. m: Order of root in the radial direction and n: Order of root in the axial
direction.

The total number of eigenvectors is 800 for the anti-symmetric case and 798 for the symmetric case.
Note: Axisymmetric vibration of incompressible cylinder (v =0.5).
Table IV(b). Approximated natural frequencies (Hz) by 40 40 element model.
Approximated solution with number of the retained eigenvectors
Mode Exact solution
no. (m, n)

30 40 60 80 160 320 800 1600 3200}3198

1 (1, 1) 2.5546 2.5544 2.5536 2.5536 2.5535 2.5535 2.5526 2.5526 2.5526
2 (1, 2) 3.0580 3.0580 3.0559 3.0557 3.0556 3.0555 3.0551 3.0547 3.0547
3 (1, 3) 3.7980 3.7552 3.7480 3.7471 3.7468 3.7467 3.7464 3.7458 3.7458
4 (2, 1) 4.5453 4.5385 4.4532 4.4400 4.4397 4.4395 4.4391 4.4388 4.4388
5 (1, 4) 4.6770 4.6770 4.5461 4.5432 4.5422 4.5417 4.5411 4.5406 4.5404
6 (2, 2) 4.8813 4.8812 4.7685 4.7469 4.7465 4.7462 4.7454 4.7451 4.7450
7 (2, 3) 5.7193 5.4596 5.2585 5.2215 5.2187 5.2184 5.2174 5.2169 5.2168
8 (1, 5) 6.2923 5.7124 5.4544 5.3999 5.3975 5.3970 5.3964 5.3958 5.3955
9 (2, 4) 7.4390 7.4390 5.8933 5.8306 5.8172 5.8163 5.8151 5.8142 5.8140
10 (1, 6) 7.9029 7.9024 6.4320 6.2980 6.2897 6.2877 6.2862 6.2854 6.2850
11 (3, 1) 8.1891 8.1643 6.7565 6.3906 6.3735 6.3729 6.3718 6.3712 6.3710
12 (2, 5) 9.0422 8.9771 6.9979 6.5382 6.5082 6.5073 6.5060 6.5049 6.5047

(m, n) means the mode shape. m: Order of root in the radial direction and n: Order of root in the axial
direction.

The total number of eigenvectors is 3200 for the anti-symmetric case and 3198 for the symmetric case.
Note: Axisymmetric vibration of incompressible cylinder (v =0.5).
transformation of the system stiness matrix. The co-ordinate-transformation scheme will en-
dow the present algorithm with a common capability for static problems of incompressible
materials and is extendable to incompressible uid problems.
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
680 Y. KOHSETSU
Number of retained eigenvectors
R
a
t
i
o

(
f
,
a
p
p
/
f
,
t
h
e
)
Mode
1 (1,1)
2 (1,2)
3 (1,3)
4 (2,1)
5 (1,4)
6 (2,2)
7 (2,3)
40 400 1000 10 100
1.00
1.01
1.02
1.03
1.04
1.05
1.00
1.01
40 400 4000
R
a
t
i
o
(
f
,
a
p
p
/
f
,
t
h
e
)
Number of retained eigenvectors
Mode
1 (1,1)
2 (1,2)
3 (1,3)
4 (2,1)
5 (1,4)
6 (2,2)
7 (2,3)
8 (1,5)
10 100 1000
1.00
1.01
1.02
1.03
1.04
1.05
1.00
1.01
(a) (b)
Figure 6. Accuracy of approximation analysis by: (a) 20 20
element model; and (b) 40 40 element model.
APPENDIX A: A TYPICAL ASYMPTOTIC PROCEDURE
A typical asymptotic procedure of eigenvalue analysis for the fully incompressible materials
is shown for reference. This procedure considers a ctitious compressibility for the system
stiness matrix.
Free-vibration equations of motion of fully incompressible material is given as (A1) (see
Equation (1)).
_
K
uu
K
up
K
T
up
0
__
u
p
_
+
_
M
uu
0
0 0
_
_
u
0
_
=0 (A1)
A ctitious compressibility matrix [10] is introduced as follows:
_

_
K
uu
K
up
K
T
up

1
zj
I
_

_
_
u
p
_
+
_
M
uu
0
0 0
__
u
0
_
=0 (A2)
where zj( K) is a bulk modulus of a compressible material and z =2(1 + v)}3(1 2v).
Substituting {p}, obtained from the second equation of (A2), into the rst equation of (A2)
yields
[

K]{u} + [M
uu
]{ u} =0 (A3)
where
[

K] =[K
uu
+ zjK
up
K
T
up
]
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
FREE VIBRATION ANALYSIS OF FULLY INCOMPRESSIBLE MATERIALS 681
Table AI. Natural frequencies (Hz) by asymptotic procedure (20 20 elements).
z First Second Third Fourth
50 2.5559 9880 3.0589 1626 3.7549 0409 4.4611 5399
500 2.5559 9904 3.0589 1626 3.7549 1023 4.4611 5403
5000 2.5559 9904 3.0589 1625 3.7549 1085 4.4611 5402
Present method (Table II) 2.5551 3.0582 3.7541 4.4608
Eigenvalue analysis of (A3) can be conducted with an appropriately large value of z. When
z becomes larger (for example 50, 500, and 5000), Equation (A3) approaches Equation (A1)
asymptotically. The numerical calculation results for 20 20 elements in Figure 3, obtained
by the above asymptotic procedure are shown in Table AI.
APPENDIX B: ANALYTICAL SOLUTIONS OF AXISYMMETRIC VIBRATIONS
OF AN INCOMPRESSIBLE CYLINDER
The partial dierential equation for axisymmetric vibrations of an incompressible cylinder
is derived below from the longitudinal vibration of a cylinder [9] with constraints of null
volumetric strain.
_
c
2
cr
2
+
1
r
c
cr
+
c
2
c:
2

1
r
2
_
c
0

j
j
c
2
c
0
ct
2
=0 (B1)
where
c
0
=
1
2
_
cu
r
c:

cu
:
cr
_
(B2)
By separating variables in c
0
as follows:
c
0
(r, :, t) =c
0
(r)Z(:)e
ipt
(B3)
Equation (B1) is separated into two ordinary dierential equations of
_
d
2
dr
2
+
1
r
d
dr

1
r
2
+
_
j
2
j
z
2
__
c
0
(r) =0 (B4)
d
2
Z(:)
d:
2
+ z
2
Z(:) =0 (B5)
The solution of (B5) is obtained as follows:
Z(:) =C
1
sin z
n
: + C
2
sin z
n
:, (z
n
=n}!) (B6)
In the next step, a change of variable of (B7) is applied to (B4)
_
j
2
}j z
2
r =x (B7)
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
682 Y. KOHSETSU
Thus, this change yields (B8) of the Bessels dierential equation:
_
d
2
dx
2
+
1
x
d
dx
+
_
1
1
x
2
__
c
0
(x) =0 (B8)
The solution of (B8) is obtained as
c
0
(x) = A
1
J
1
(x)
c
0
(r) = c
0
(x}
_
j
2
}j z
2
) =1}
_
j
2
}j z
2
c
0
(x)
= A
1
}
_
j
2
}j z
2
J
1
(
_
j
2
}j z
2
r)
(B9)
Consequently, the solution of (B1) is expressed as follows:
c
0
(r, :, t) =A
1
J
1
(
_
j
2
}j z
2
r)(C
1
cos z : + C
2
sin z :)e
ipt
(B10)
Equation (B2) and boundary conditions are taken into account, and natural frequencies and
natural modes for u
r
and u
:
are obtained as (B11) from (B10).
[
nm
=
1
2
=
1
2
_
j
j
{k
2
n
+ (z
m
}a)
2
}
u
r
= A
1
C
2
}z
n
J
1
(k
n
r) cos z
n
: = A}z
n
J
1
(k
n
r) cos z
n
:
u
:
= A
1
C
2
}k
n
J
0
(k
n
r) sin z
n
: =A}k
n
J
0
(k
n
r) sin z
n
:
(B11)
where [
nm
is the natural frequency; A is a constant; z
m
: mth root of J
1
(z
m
) =0, Bessel function
of rst order; k
2
n
=j
2
}j z
2
n
and z
n
=n}!.
APPENDIX C: DERIVATION OF ELEMENT STIFFNESS AND MASS MATRIX
BY HERRMANN METHOD
A strain energy function in cylindrical co-ordinates (see Figure 1) is given as (C1) based on
the Modied HellingerReissner Principle [11], and is equivalent to that by Herrmann.
A(c
r
, c
0
, c
:
,
r:
, H) = j
_
c
2
r
+ c
2
0
+ c
2
:
+
1
2

2
r:

+ 2vj(c
r
+ c
0
+ c
:
)H v(1 2v)jH
2
(C1)
where H is a Lagrange multiplier, which is multiplied by 2vj for convenience.
Axisymmetric strains are expressed as
{U} =
_

_
c
r
(r, :)
c
0
(r, :)
c
:
(r, :)

r :
(r, :)
_

_
=
_

_
c
cr
0
1
r
0
0
c
c:
c
c:
c
cr
_

_
_
u
r
(r, :)
u
:
(r, :)
_
(C2)
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
FREE VIBRATION ANALYSIS OF FULLY INCOMPRESSIBLE MATERIALS 683
A standard bilinear shape function is applied to the displacement and pressure elds and the
co-ordinates of the element shown in Figure 2 is as follows:
N
i
(, p) =
1
4
(1 +
i
)(1 + pp
i
) (C3)

i
and p
i
are the normalized co-ordinates at node i (i =1, 2, 3, 4) in Figure 2.
By using (C3), the axisymmetric strains (C2) are expressed by nodal displacements as
{U} =
_

_
c
cr
0
1
r
0
0
c
c:
c
c:
c
cr
_

_
_
N
1
N
2
N
3
N
4
N
1
N
2
N
3
N
4
__
u
ri
u
:i
_
=[B]
_
u
ri
u
:i
_
(C4)
The Lagrange multiplier {H} is also expressed as
{H} =[N
1
N
2
N
3
N
4
]{p
i
} (C5)
The strain energy (C1) for the element is rewritten in matrix form using (C4) and (C5) as
A(c
r
, c
0
, c
:
,
r :
, H) =
1
2
_
u
p
_
T
_
K
uu
K
up
K
T
up
K
pp
__
u
p
_
(C6)
where
[K
uu
] =
_
J
[B]
T
[D
uu
][B] dJ, [K
up
] =
_
J
[B]
T
[D
up
][N
i
] dJ
[K
pp
] =
_
J
[N
i
]
T
[D
pp
][N
i
] dJ
[D
uu
] = j
_

_
2 0 0 0
0 2 0 0
0 0 2 0
0 0 0 1
_

_
, [D
up
] =j[2v 2v 2v 0]
T
[D
pp
] = 2jv(1 2v)
The kinetic energy of the element is given by
1 =
1
2
{ u}
T
[M
uu
]{ u} (C7)
where
[M
uu
] =j
_
J
[N
i
]
T
[N
i
] dJ
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684
684 Y. KOHSETSU
Using (C6) and (C7), Lagranges equation yields equations of motion of the element as
_
K
uu
K
up
K
T
up
K
pp
__
u
p
_
+
_
M
uu
0
0 0
__
u
0
_
=0 (C8)
REFERENCES
1. Gadala MS. Numerical solutions of nonlinear problems of continua-II. Survey of incompressibility constraints
and software aspect. Computers and Structures 1986; 22:841855.
2. Sussman T, Bathe K. A nite element formulation for nonlinear incompressible elastic and inelastic analysis.
Computers and Structures 1987; 26:375409.
3. Hughes TJR. The Finite Element Method. Dover: New York, 2000; 192231.
4. Szabo B, Babu ska I. Finite Element Analysis. Wiley: New York, 1991; 205215.
5. Zienkiewicz OC, Taylor RL. The Finite Element Method, Volume 1 The Bases (5th edn). Butterworth: London,
2000; 305343.
6. Herrmann LR. Elasticity equations for incompressible and nearly incompressible materials by a variational
theorem. AIAA Journal 1965; 3(10):18961900.
7. Kolata WG. Eigenvalue approximation by the nite method: the method of Lagrange multipliers. Mathematics
of Computation 1979; 33(145):6376.
8. Bathe KJ. Finite Element Procedures. Prentice-Hall: Englewood Clis, NJ, 1996; 300335.
9. Love AEH. A Treatise on the Mathematical Theory of Elasticity (4th edn). Dover: New York, 1944;
287291.
10. Zienkiewicz OC, Taylor RL. The Finite Element Method, Volume 1 The Bases (5th edn). Butterworth: London,
2000; 323325.
11. Washizu K. Variational Methods in Elasticity (2nd edn). Pergamon Press: New York, 1975; 358360.
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 57:669684

Das könnte Ihnen auch gefallen