Вы находитесь на странице: 1из 15

Wave Motion ( )

Contents lists available at ScienceDirect


Wave Motion
journal homepage: www.elsevier.com/locate/wavemoti
Vibration modelling of structural networks using a hybrid
finite element/wave and finite element approach
Jamil M. Renno
a,
, Brian R. Mace
b
a
Institute of Sound and Vibration Research, University of Southampton, Southampton SO17 1BJ, United Kingdom
b
Department of Mechanical Engineering, University of Auckland, Auckland, New Zealand
h i g h l i g h t s
A hybrid finite element/wave and finite element approach is presented.
This approach can be used for modelling structural networks.
The networks comprise arbitrarily complicated waveguides and joints.
The approach leads to small numerical models.
This extends the frequency range of applicability of the finite element method.
a r t i c l e i n f o
Article history:
Received 1 March 2013
Received in revised form 29 August 2013
Accepted 2 September 2013
Available online xxxx
Keywords:
Structural networks
Wave and finite element method
Scattering coefficients
a b s t r a c t
The vibration modelling of waveguide structures is considered. These structures comprise
waveguides connected via joints. Traditionally, analytical models of the wave behaviour
of such structures can be developed if they are simple (beams or rods connected at
point joints, etc.). However, if the waveguides are of complicated constructions (truss-like,
layered media, etc.) or the joints are complicated (e.g. of significant physical dimensions),
obtaining the wave characteristics might be a formidable task. Inthis paper, suchstructures
are modelled using a hybrid finite element/wave and finite element (FE/WFE) approach.
The waveguides are modelled using the WFE method and thus their wave characteristics
are obtained regardless of the complexity of their cross-section. The joints are modelled
using standard FE, and the WFE and FE models are coupled to yield the scattering properties
of the joints. The propagation and scattering models are assembled to describe the
behaviour of the structure using relatively small models, while also providing information
for other applications such as structure-borne sound, statistical energy analysis, etc.
Numerical examples are presented to illustrate the approach.
2013 Elsevier B.V. All rights reserved.
1. Introduction
Many structural networks comprise waveguides which are connected via joints. The vibrational behaviour of such
networks can be described by the waves that travel through the waveguides and the scattering properties of the joints.
For simple structures (where the waveguides are of simple construction and are connected at point joints for example),
the wave characteristics of the waveguides can be obtained analytically along with the scattering properties of the joints
[13]. For complicated structures, analytical solutions might be very difficult to obtain and the finite element (FE) method is
often used. However, FE models become impractical at higher frequencies leading to a multitude of problems (computation

Corresponding author. Tel.: +44 0 23 8059 2344; fax: +44 0 23 8059 3190.
E-mail addresses: renno@isvr.soton.ac.uk (J.M. Renno), b.mace@auckland.ac.nz (B.R. Mace).
0165-2125/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.wavemoti.2013.09.001
2 J.M. Renno, B.R. Mace / Wave Motion ( )
Nomenclature
k Wavenumber, 1/m.
L Length, m.
n Number of degrees of freedom on each side of the waveguide segment.
t Time, s.
Y Mobility, m/(N s).
Length of segment, m.
Excitation frequency, rad/s.


1.
(.) Denotes the real part of.
(.) Denotes the imaginary part of.
Superscript
T Matrix transposition.
+ Quantity pertaining to a positive-going wave.
Quantity pertaining to a negative-going wave.
Subscript
i Interface nodes.
n Non-interface nodes.
w Waveguide.
x, y, z Property in the x-, y- or z-direction.
J Joint.
speed, accuracy, memory consumption, etc.). Additionally, the knowledge of the wave characteristics is advantageous for
many applications such as structure-borne sound, statistical energy analysis, etc.
The analysis of waveguide structures has attracted much attention. The approaches developed can be categorised into
matrix- and wave-based methods. The former uses a matrix formulation to couple the waveguides that constitute the
structure [4]. Examples include the dynamic stiffness method [57], mobility power-flow analysis [8], the wave scattering
approach [9], and the spectral element method [10]. Other researchers presented matrix formulations that incorporate wave
scattering to analyse waveguide structures [3,11,12].
Wave-based methods rely on obtaining the reflection and transmission coefficients of the joints and using the wave
characteristics of the waveguides to describe the vibrational behaviour of the structural network. For thin beams, Cremer
et al. [13] calculated the scattering coefficients of joints due to an incident propagating wave. Mace [14] considered the
scattering due also to an incident evanescent wave, and later studied the properties of the scattering coefficients [15].
Researchers also used wave-based methods to treat three-dimensional frames [9], truss-like structures [16], structural
networks [1,2], and water-filled piping systems [17]. Wave approaches were also used to calculate the reflection and
transmission coefficients of plate/beam junctions [18], bolted joints [19], curved beams [20] and to analyse the vibrational
energy in built-up structures [2123]. Heron [24] used a line impedance wave approach to calculate the transmission
between two plates and Hambric [25] compared various models of a T-joint and showed that at higher frequencies the
finite dimensions of the joint become important and should be included in the model.
The waveguides in the reviewed literature were mostly isotropic and the joints were often idealised (e.g. point
connections). For such cases, analytical models can be developed both for the wave propagation and scattering properties of
the joints. However, for complicated waveguides and/or complicated joints, finding the reflection and transmission matrices
of the joints using a purely analytical formulation can be extremely difficult at best, so that a numerical approach might be
required. In this paper, the vibrational response of structural networks will be analysed using a hybrid finite element/wave
finite element (FE/WFE) approach. The waveguides will be modelled using the WFE method. This relies on post-processing
a standard FE model of a small segment of each waveguide using periodic structure theory. As for the joints, their scattering
properties are obtained using the WFE models of the waveguides and the FE model of the joints. The degrees of freedom
(DOFs) at the interfaces of the WFE and FE models (of the waveguides and the joint respectively) are compatible [26].
The approach is similar to that described by Se-Knudsen and Sorokin [27] where the wave behaviour of straight and
curved pipes is obtained through the boundary integral equations method and the supporting structures (i.e., the joints)
are modelled using standard FE. Earlier, Doyle [10] proposed modelling waveguides by the spectral elements formed from
analytical methods and coupling these wave models to the FE models of joints. However, these approaches suffer from
a possible problem: if the waveguide has a complicated construction (e.g. a laminate or a solid or thin-walled section for
whichhigher order wavemodes are important), thenfinding the spectral element analytically, or using the boundary integral
equations method can be very difficult. Moreover, within the approach of Doyle [10], the models of the waveguides and the
J.M. Renno, B.R. Mace / Wave Motion ( ) 3
Fig. 1. FE model of a segment of a waveguide with the waveguides local coordinate system.
joint are not compatible at their interface. While the models can be coupled (e.g. [10] or by the use of Lagrange multipliers),
there is a discontinuity in the models, which results in spurious, although usually small, additional scattering effects at the
interface [28,29]. In the approach described here, these issues are not relevant: the WFE method is used to determine the
wavemodes so that the complexity of the construction is arbitrary, and the interfaces between the joint and the waveguides
have compatible meshes. The WFE models and the scattering properties of the joints can then be used to obtain the response
of the whole structure.
The WFE method has been used to study the free [30] and forced [31] vibration of waveguides. The WFE method for
waveguides has also been used to study thin-walled structures [32], laminated plates [30] and fluid filled pipes [33,34]. It
has also been extended to two dimensional plane [35] and cylindrical structures [36]. Recently, the WFE method was used
along with contour integration to calculate the response of waveguides [37] and two dimensional plane structures [38] to
arbitrarily distributed loads.
The remainder of this paper is organisedas follows. InSection2, the WFE methodis briefly reviewedandSection3reviews
its application to obtain the scattering properties of joints. The response of structural networks is discussed in Section 4 and
numerical examples are presented in Section 5. Finally, conclusions for this work are drawn in Section 6.
2. Review of the WFE method
The WFE method can be used to model one and two-dimensional piecewise homogeneous/periodic structures. The
overview below is borrowed from [39].
The WFE method starts with obtaining the FE model of a small segment of the waveguide, Fig. 1, using any FE package
with the only constraint being that the nodes and their DOFs are ordered identically on the left and right sides of the
segment; internal nodes can be eliminated via dynamic condensation [39]. Time harmonic dependence of the formexp(t)
is assumed throughout this work and is suppressed for brevity. The governing equation of the segment of Fig. 1 is
Dq = f +e where D = K +C
2
M, (1)
and q, f and e are (2n 1) vectors of nodal DOFs, internal and external nodal forces, respectively. D, M, C and K are the
dynamic stiffness, mass, viscous damping and stiffness matrices of the segment and n is the number of DOFs at each side of
the segment. The dynamic stiffness matrix (DSM), D, can be partitioned to reflect the influence of the left and right nodes of
the segment, and thus Eq. (1) can be expressed as
_
D
LL
D
LR
D
RL
D
RR
_ _
q
L
q
R
_
=
_
f
L
f
R
_
+
_
e
L
e
R
_
, (2)
where the subscripts L and R denote the left and right sides of the segment, respectively.
2.1. Free wave propagation
When a wave propagates freely along the x-axis of the waveguide, i.e., e = 0, the propagation constant = exp(k)
relates the right and left nodal DOFs and forces by [39]
q
R
= q
L
, f
R
= f
L
. (3)
Since waves are only travelling along the x-axis, the propagation constant relates all the DOFs and internal nodal forces on
both sides of the segment. Substituting Eq. (3) into Eq. (2) yields an eigenvalue problem for . This can be cast in a number
of forms, perhaps the simplest being

_
q
L
f
L
_
= T
_
q
L
f
L
_
where T =
_
D
1
LR
D
LL
D
1
LR
D
RL
+D
RR
D
1
LR
D
LL
D
RR
D
1
LR
_
(4)
is the transfer matrix. For complicated waveguides with many DOFs at each node, care needs to be taken when solving this
eigenvalue problem as various numerical problems may arise [39]. The eigenvalue problem is usually then recast into one
of a number of better-conditioned forms [40].
The eigenvalues occur in reciprocal pairs as
+
j
and

j
= 1/
+
j
[40,41], with wavenumbers k
+
j
and k

j
= k
+
j
,
corresponding to positive- and negative-going waves respectively. Associated with these eigenvalues are the positive-
4 J.M. Renno, B.R. Mace / Wave Motion ( )
and negative-going eigenvectors
+
j
and

j
respectively, which are also called the wavemodes. Every wavemode can be
partitioned as

j
=
_

f
_
j
where j = 1, . . . , n.
Positive going waves are characterised by

+
j

1,

_
f
T
L
q
L
_
=
_
f
T
L
q
L
_
< 0 if

+
j

= 1,
(5)
where the overdot indicates a derivative taken with respect to time. Eq. (5) states that if the wave is travelling in the positive
direction, then its amplitude should be decreasing or that if its amplitude remains constant, then there is time average power
transmission in the positive direction. Eq. (5) includes evanescent waves which decay with distance. Although these waves
do not transfer energy individually, they contribute to the input response and when incident at a joint, they could give rise
to propagating and evanescent waves [14].
The wavemodes can be grouped into the matrices

=
_

n
_
, =
_

_
.
The left wavemodes can also be obtained. These are (1 2n) vectors which can be partitioned as

j
=
_

f

q
_
j
where j = 1, . . . , n,
and further grouped into
=
_

_
where

=
_
_
_

1
.
.
.

n
_

_
.
The left and right wavemodes can be normalised [39] such that

= I. The partitions of the left and right eigenvectors


can be used to form the matrices

q
=
_

q,1

q,n
_
,

q
=
_
_
_

q,1
.
.
.

q,n
_

_
.
Similar expressions hold for

f
and

f
. These matrices, together with the orthogonality relations, define transformations
between the physical domain, where the motion is described in terms of q and f, and the wave domain, where the motion
is described in terms of waves of amplitudes a
+
and a

travelling in the positive and negative directions respectively.


Specifically,
_
q
L
f
L
_
= a with a =
_
a
+
a

_
. (6)
In practice, as in modal analysis, only m pairs of (positive- and negative-going) waves might be retained, so that

q,f
and

q,f
are (n m) and (mn) matrices respectively. The number retained can be different at different frequencies [26].
2.2. Forced wave propagation
A point excitation f
e
acting on the waveguide will generate positive- and negative-going waves of amplitudes a
+
and a

which will propagate away from the excitation point in the right and left directions respectively. The amplitudes of these
waves can be obtained as [39]
a
+
=
+
q
f
e
and a

q
f
e
. (7)
In the case of spatially distributed loads, the amplitudes of the excited waves can be obtained analytically by using the WFE
model along with contour integration techniques [37].
3. Scattering properties of joints
The reflection and transmission coefficients of joints in structural networks will nowbe evaluated using a hybrid FE/WFE
approach [26]. First, the vibration of the waveguides will be described in a global coordinate system. Then, the scattering
properties of point joints will be treated, followed by the case of joints of finite dimensions.
J.M. Renno, B.R. Mace / Wave Motion ( ) 5
a b
Fig. 2. Local and global transformation with: (a) joint to the left end of the waveguide and (b) joint to the right end of the waveguide.
3.1. Waveguide motion in the global coordinate system
Consider a waveguide connected to a joint. The local coordinate system(x, y) of the waveguide is defined with the x-axis
along the waveguide as shown in Fig. 2. Thus the DOFs and internal forces vectors in the local coordinate system can be
written as Eq. (6).
The amplitudes of the incoming and outgoing waves at the joint are denoted by a
i
and a
o
, respectively. The superscripts
i and o represent the incoming and outgoing waves, respectively. For the two cases shown in Fig. 2, these are related to the
wave vectors straightforwardly. For example, for a joint at the left end of the waveguide a
i
= a

and a
o
= a
+
.
Suppose now a global coordinate system (X, Y) is defined (Figs. 2 and 3). The DOFs and internal forces vectors in the
global coordinate system can be written in terms of the incoming and outgoing waves as
_
q
g
f
g
_
=
_

i
q

o
q

i
f

o
f
_ _
a
i
a
o
_
, (8)
where
i
q
,
o
q
,
i
f
and
o
f
are matrices that relate the wavemodes in the local coordinate system to the global coordinate
system and the subscript g indicates that the quantity is evaluated in the global coordinate system. The transformation
depends on the location of the joint with respect to the waveguide. If the left end of the waveguide is connected to the joint,
Fig. 2(a), then
a

= a
i
, a
+
= a
o
, q
g
= Rq, f
g
= Rf, (9)
where R is the rotation matrix that relates the local coordinate system (x, y) to the global one. Substitute Eqs. (6) and (9) in
Eq. (8) yields
_

i
q

o
q

i
f

o
f
_
=
_
R

q
R
+
q
R

f
R
+
f
_
.
In the following sections it is convenient to define
o
f
=
+
f
R
T
. Similarly, if the right end of the waveguide is connected to
the joint, Fig. 2(b), then
a
+
= a
i
, a

= a
o
, q
g
= Rq, f
g
= Rf. (10)
Substituting Eqs. (6) and (10) in Eq. (8) yields
_

i
q

o
q

i
f

o
f
_
=
_
R
+
q
R

q
R
+
f
R

f
_
.
3.2. Point joints
Fig. 3 shows p waveguides with zero-dimensional cross-sections connected at a point joint.
1
In the j-th waveguide, the
local coordinate system (x
j
, y
j
) is assigned arbitrarily, and the rotation matrix R
j
transforms the DOFs from the local (x
j
, y
j
)
to the global (X, Y) coordinate system. External excitations may act at the joint, generating waves, and some of the joint
DOFs might be constrained.
The WFE model of each waveguide is obtained in its local coordinate system (x
j
, y
j
). In Fig. 3, waves of amplitudes a
i
j
are
incident on the joint from the j-th waveguide. They give rise to reflected outgoing waves of amplitudes a
o
j
= r
jj
a
i
j
in the
j-th waveguide, and transmitted waves of amplitudes a
o
k
= t
kj
a
i
j
in the k-th waveguide, where r
jj
and t
kj
are matrices of the
reflection and transmission coefficients of the joint. These define the scattering matrix s of the joint, whose partitions relate
the amplitudes of the incident and scattered waves as
a
o
k
= s
kj
a
i
j
for j, k = 1, . . . , p,
where s
jj
= r
jj
and s
kj
= t
kj
for j = k.
1
For simplicity, the waveguides are shown to lie in a plane: if this is not the case, the only difference is related to the rotation matrices R
j
of the
waveguides which appears in Eqs. (9) and (10).
6 J.M. Renno, B.R. Mace / Wave Motion ( )
Fig. 3. Waveguides connected at a point joint with incident and outgoing waves.
3.2.1. Similar waveguides rigidly connected at a point joint
Here, the case of similar waveguides rigidly connected at a point joint is considered. All the waveguides can be modelled
using the same number of DOFs; however, the WFE models need not retain the same number of waves. The equilibriumand
continuity conditions at the joint can be written as
A
_
f
g,1
+ +f
g,p
_
+B

q
J
= F
J
+F
ext
, q
g,1
= = q
g,p
= q
J
, (11)
where A and B

are matrices arising from the equilibrium and continuity conditions for the waveguides respectively,
F
J
= D
J
q
J
is the vector of nodal forces due to the dynamic stiffness of the joint if relevant, q
J
is the vector of DOFs of
the joint, D
J
is the DSM of the joint and F
ext
is the vector of applied external forces. For example, if a mass is attached to the
joint then A = I
nn
, B

= 0
nn
and D
J
=
2
M
ext
where M
ext
is a diagonal matrix of masses and second moments of mass.
The dynamic stiffness of the joint can be included in the matrix B

so that Eq. (11) can be written as


A
_
f
g,1
+ +f
g,p
_
+Bq
J
= F
ext
, q
g,1
= = q
g,p
. (12)
Substituting Eq. (8) for each waveguide in Eq. (12) leads to the relation between the incident and outgoing waves at the
junction
E
o
a
o
= E
i
a
i
+

G
J
where a
i,o
=
_

_
a
i,o
1
.
.
.
a
i,o
p
_

_
, (13)
where
E
i,o
=
_
_
_
_
_
A
i,o
f ,1
+B
i,o
q,1
A
i,o
f ,2
A
i,o
f ,p

i,o
q,1

i,o
q,2
.
.
.
.
.
.

i,o
q,1

i,o
q,p
_

_
and

G
J
=
_

_
F
ext
0
.
.
.
0
_

_
.
Eq. (13) can be solved straightforwardly to yield
a
o
= sa
i
+G
J
,
where s = E
1
o
E
i
is the scattering matrix and G
J
= E
1
o

G
J
is the joint generation matrix.
If only m
j
< n pairs of (positive- and negative-going) waves are retained for the WFE model of the j-th waveguide, then
E
o
is not invertible. In this case, pseudo-inversion is used by, for instance, premultiplying Eq. (13) by
o
f
defined as

o
f
=
_
_
_

o
f ,1
R
T
1
.
.
.

o
f ,p
R
T
p
_

_
,
resulting in s =
_

o
f
E
o
_
1

o
f
E
i
and G
J
=
_

o
f
E
o
_
1

o
f

G
J
.
J.M. Renno, B.R. Mace / Wave Motion ( ) 7
Fig. 4. Schematic of two waveguides attached at a joint with incident, reflected and transmitted waves (the number of waveguides conned to the joint is
arbitrary but here two are shown for simplicity). The joint is shown in dashed lines.
3.2.2. Point joints general case
In the most general case, the number of DOFs in each waveguide might differ (e.g., a string connected to a beam), some
joint DOFs might be zero (e.g., a beam on a simple support) or some external forces might be zero (e.g., a pin joint between
two beams). In such a case, the equilibrium condition of Eq. (12) still holds, but the continuity conditions take different
forms. In general, the continuity and equilibrium conditions at the joint can be written as
f
g
+q
g
=
_
F
ext
0
_
, (14)
where the vectors q
g
and f
g
are now defined by concatenating the relevant vectors for the individual waveguides.
Furthermore, the matrices
i,o
q,f
are defined by concatenating the relevant matrices for the individual waveguides (i.e., the
matrices are block-diagonal). Matrices
i,o
q,f
are defined in a similar manner.
The matrices and include blocks of identity and zero matrices. For example, for the case discussed in Section 3.2.1, the
equilibriumand continuity conditions of Eq. (12) can be written in the formof Eq. (14). The first block-rowof includes the
identity matrices of appropriate dimensions, the remaining elements being zero. The first block-row of contains the DSM
of the joint and a block of null matrices, while the remaining block-rows contain positive and negative identity matrices.
Expressing q
g
and f
g
in the wave domain using the concatenated matrices
i,o
q,f
yields the scattering matrix as
s =
_

o
f
+
o
q
_
1
_

i
f
+
i
q
_
.
The joint generation matrix is
G
J
=
_

o
f
+
o
q
_
1
_
F
ext
0
_
.
Pseudo-inversion can again be used to estimate the scattering matrix if a reduced wave basis is retained.
3.3. Finite joints
Now suppose that p waveguides are attached at a joint of finite dimensions, Fig. 4. The situation is similar to that of
a point joint except that there are now a number of joint nodes, some internal, and normally more than one node across
the cross-section of each waveguide. Again, external forces might be applied and some DOFs might be driven with a given
displacement () to generate waves, and some DOFs might be constrained (thus not generating any waves). The waveguides
are modelled using the WFE method of Section 2, and the joint is modelled using standard FE. The WFE model of each
waveguide is obtained in its local coordinate system (x
j
, y
j
) using n
j
DOFs (on each side of the waveguide segment) and m
j
pairs of positive- and negative-going waves are retained in the WFE model. In the following, it is assumed that the interface
DOFs of the joint are compatible with those of the WFE models.
Time harmonic behaviour of the joint is described through
_

D
ii

D
in

D
ni

D
nn
_ _
Q
i
Q
n
_
=
_
F
i
F
n
_
(15)
where Q and F are vectors of DOFs and internal nodal forces of the joint represented in the global coordinate system (X, Y)
and the subscripts i and n represent interface and non-interface nodes, respectively.
8 J.M. Renno, B.R. Mace / Wave Motion ( )
Vectors q
g
andf
g
andmatrices
i,o
q,f
and
i,o
f
are nowdefinedas inSection3.2.2. The continuity andequilibriumconditions
for the joint are
Q
i
= q
g
, F
i
f
g
= 0. (16)
The displacement and force boundary conditions of the non-interface nodes can be written as
AF
n
+B(Q
n
) = F
J
. (17)
Substituting Eq. (15) in Eq. (17) yields
Q
n
= (AD
nn
+B)
1
_
F
J
+B AD
ni
Q
i
_
. (18)
Now substituting Eqs. (8) and (15) in Eqs. (16) and (18) yields
E
o
a
o
= E
i
a
i
+

G
J
, (19)
which is identical to Eq. (13) and the matrices are now given as
E
o
=
o
f
D
ii

o
q
+(AD
nn
+B)
1
AD
ni

o
q
,
E
i
=
i
f
D
ii

i
q
+(AD
nn
+B)
1
AD
ni

i
q
,

G
J
= D
in
(AD
nn
+B)
1
(F
J
+B).
Eq. (19) can be solved straightforwardly yielding the scattering matrix as s = E
1
o
E
i
and the joint generation matrix
G
J
= E
1
o

G
J
. Once again, if a reduced number of waves is retained, then pseudo-inversion is required. If Eq. (18) is
premultiplied by
o
f
the scattering matrix becomes s =
_

o
f
E
o
_
1

o
f
E
i
.
3.4. Incompatible meshes
The development presented in Section 3.3 is based on the joint/waveguide interface having a compatible mesh which
means that the waveguide will have the same meshthroughout its length. This might not always be possible or desirable (e.g.
for a waveguide connected to two different joints, requiring different interface meshes, on each side). In this case, one can
introduce an artificial joint along the waveguide which in essence connects the two sections of the same waveguide (with
different meshes) and the results of Section 3.2.1 can be used. Few waves are retained for the WFE model of each section
of the waveguide and these can be identified using a wave assurance criterion [39] (analogous to the modal assurance
criterion [42]).
4. Response of structural networks
The WFE models (developed in Section 2) and the scattering properties of the joints (obtained in Section 3) can be
used to obtain the response of structural networks, i.e., structures comprising a number of interconnected waveguides.
The models can be assembled in the physical domain, i.e., in terms of DOFs and forces, as in the dynamic stiffness method.
This approach will be briefly outlined below. Alternatively the models can be assembled in the wave domain, in terms of
waves propagating along each waveguide. The wave domain approach results in smaller models because only a reduced
wave basis might be required and the joints and waveguides are described in terms of relatively fewwaves compared to the
number of physical DOFs. Indeed, assembling the model in the wave domain has connections with the general mathematical
concept of a graph/network which comprises a set of elements that are connected by some relations [43].
4.1. Model assembly in the physical domain
The model of the structural network could be assembled in the physical domain, i.e., in terms of the DOFs and forces.
The WFE model of each waveguide is used to find its DSM using the approach outlined by Duhamel et al. [31]. If the j-th
waveguide comprises N
j
= L
j
/
j
segments, then its DSM is given by [31]
D
j
=
_
[D
LL
]
j
0
0 [D
RR
]
j
_
+
_
[D
LR
]
j
0
0 [D
RL
]
j
_
_
P
N
j
1
L
P
R
P
L
P
N
j
1
R
__
P
N
j
L
I
I P
N
j
R
_
,
where
P
L
=

q
diag
_

+
j
_

q
_

q
_
1
and P
R
=
+
q
diag
_

+
j
_

+
q
_

+
q

+
q
_
1
,
where diag(.) is the diagonal matrix. The above computation does not depend on the number of segments N
j
, which
need not be integral. Moreover, the computation of matrix exponents is straightforward since for example, P
N
j
1
L
=

q
_
diag
_

+
j
__
N
j
1

q
_

q
_
1
involves only raising
+
j
to the desired exponent.
J.M. Renno, B.R. Mace / Wave Motion ( ) 9
Fig. 5. A structural network of four waveguides. The orientation of the waveguides is arbitrary, and any waveguide can be excited at any point.
The DSMs of the individual waveguides are then appropriately assembled together with the DSMs of the joints as in
standard FE to form the DSM of the whole structural network. Despite the computational efficiency of calculating the DSMs
of the individual waveguides, the size of the assembled model of the structural network may be large due to the number of
DOFs used to construct the model. This approach will then be computationally demanding, especially when inverting the
assembled DSM of the whole network.
4.2. Model assembly in the wave domain
The network can contain an arbitrary number of waveguides; Fig. 5 shows an example of a network with four
waveguides,
2
each of length L
j
(j = 1, . . . , 4). The assembly procedure closely follows that of FE analysis [44]. The n
w
waveguides (cf. elements in FE analysis) are numbered, as are the n
J
joints (cf. nodes). A WFE model of each waveguide
is obtained using the waveguides local coordinate system (which can be assigned arbitrarily). The wavemodes are then
transformed into the global coordinates and the scattering matrices of the joints are obtained.
Vectors a
i
and a
o
of incident and outgoing waves at the joints are formed by concatenating the vectors a
i,o
j,k
of the
amplitudes of the incident/outgoing waves at joint j from/to waveguide k. The block-diagonal global scattering matrix s is
formed fromthe scattering matrices s
j
of the joints along the leading diagonal. The amplitudes of the incident and scattered
waves are related through the scattering matrices as
a
o
= sa
i
+G
J
, (20)
where G
J
is the vector of joint generation matrices, found by concatenating the vectors G
J ,j
for each joint. For the example
shown in Fig. 5 the vectors and matrices are
a
i,o
=
_

_
a
i,o
1,1
a
i,o
1,4
a
i,o
2,1
a
i,o
2,2
a
i,o
3,2
a
i,o
3,3
a
i,o
4,3
a
i,o
4,4
_

_
, s =
_
_
_
s
1
s
2
s
3
s
4
_

_
, G
J
=
_

_
G
J ,1
G
J ,2
G
J ,3
G
J ,4
_

_
.
2
For simplicity, a planar structural network withpoint joints is showninFig. 5. However, the developments equally apply to three-dimensional structural
networks with joints of finite dimensions.
10 J.M. Renno, B.R. Mace / Wave Motion ( )
Next, relations of wave propagation are written for each waveguide and assembled. A wave, propagating over a distance x,
changes in phase and magnitude by the factor exp(k
+
x). Suppose that the k-th waveguide is attached to joints m and n.
The propagation relations for the waveguide are then
a
i
m,n
=
k
(L
k
)a
o
n,k
+G
W,m,k
and a
i
n,k
=
k
(L
k
)a
o
m,k
+G
W,n,k
,
where

k
(x) = diag
_
exp
_
k
+
1,k
x
_
, . . . , exp
_
k
+
m
k
,k
x
__
(21)
is the wave propagation matrix that gives the amplitudes of the waves after travelling a distance x in the k-th waveguide. The
terms G
W,m,k
and G
W,n,k
are the waveguide generation vectors for the k-th waveguide. For a point excitation (Section 2.2)
these are simply the waves a

generated by the excitation and propagated from the excitation point to the joints, with
similar expressions holding if the excitation is distributed [37]. The propagation relations for all waveguides are assembled
to give
a
i
= a
o
+G
W
. (22)
Eqs. (20) and (22) can then be solved straightforwardly to give
a
o
= (I s)
1
(G
J
+sG
W
).
Note that the magnitude of every element of
k
(x) (and consequently of ) is less thanor equal to 1, resulting ina formulation
that is well-conditioned. This is in contrast to, for example, transfer matrix methods, where numerical problems arise due
to the presence of terms such as both exp(k
+
L) and exp(+k
+
L), with k
+
L having a large (negative) imaginary part.
Furthermore, in contrast with the DSM approach described in Section 4.1, the size of the resulting model will be much
smaller since it is directly related to the number of waves retained in the wave basis of each waveguide. Assembling the
model in the wave domain directly benefits from the reduced order basis introduced by the WFE model.
As an example, waveguide 1 in Fig. 5 is connected to joints 1 and 2, with the point load being applied at a distance x
e
from joint 1. Hence the propagation relations become
a
i
1,1
=
1
(L
1
)a
o
2,1
+
1
(x
e
)b

and a
i
2,1
=
1
(L
1
)a
o
1,1
+
1
(L
1
x
e
)b
+
, (23)
where b

are the waves generated by the excitation; these can be obtained using Eq. (7). Matrix and vector G
W
of Eq. (22)
follow straightforwardly from Eq. (23).
The response at any point in the k-th waveguide can then be found by propagating the outgoing waves a
o
m,k
and a
o
n,k
from
the joints to the response point using the appropriate wave propagation matrix of Eq. (21). Once the amplitudes of the waves
are known at the response point, the DOFs and forces can be obtained by employing the transformation between the wave
and physical domains using Eq. (6).
5. Numerical examples
In this section, three numerical examples are presented. Analytical solutions for the first example, a rectangular frame
comprising four EulerBernoulli beams, can be obtained. The second example is of four simply supported plate strips also
forming a rectangular frame and connected at line joints. The last example is of a cross-section of a train cabin where the
hybrid approach of Section 3.3 is required. In the following, all material properties and dimensions are in SI units.
5.1. Beam structure
A rectangular beam structure is considered, Fig. 5. The point joining every two beams has a mass of 0.25 kg. Joint 2
cannot move in the x
1
-direction but can freely slide in the y
1
-direction, whereas joint 3 is assumed to be simply supported.
The beams are made of mild steel with a density = 7800, modulus of elasticity E = 200 10
9
, Poisson ratio = 0.3,
and structural damping = 0.01. The four beams are identical and the width, thickness and length are 0.03, 0.01 and 1
respectively. The beams are slender, so EulerBernoulli beam theory is used (but other theories can be equally used).
Since the beams are identical, it is sufficient to obtaina single WFE model. This is achievedhere via the FE model of a single
BEAM3 element of ANSYS
R
( = 0.01). BEAM3 is a two-noded element with 3 DOFs per node. The approach presented in
Section 4 is used to calculate the scattering properties of the joints and subsequently find the response of the structure. For
comparison, the response can be obtained analytically using a dynamic stiffness approach, e.g. [45]. A unit amplitude force
in the vertical direction is applied to the lower beam (waveguide 1 in Fig. 5) at x = 0.25 m from joint 1. Fig. 6 shows the
transverse mobility of the lower beam at the midpoint of the beam computed analytically and using the FE/WFE approach.
The FE/WFE results are in good agreement with the analytical results.
5.2. Plate structure
In this example, consider four plates that are connected at their ends to form a rectangular frame as in Fig. 5 with the
waveguides being plates rather than the beams considered in the previous example. All the edges of the plates that are
J.M. Renno, B.R. Mace / Wave Motion ( ) 11
Fig. 6. Transfer mobility in the y
1
-direction at the midpoint of the lower beam:analytical, o FE/WFE.
Fig. 7. Schematic of a plate strip with simply-supported boundaries and a segment of length for the WFE model.
parallel to the X-axis are simply supported, Fig. 7. The plates are made of mild steel (material properties are given in the
previous example). The plates have the same length, L = 0.6, are 18 cm wide and 1.8 mm thick.
Since the four plates are identical, one WFE model is sufficient to model the waveguides. This was obtained using a
segment (with = 0.02) of nine SHELL63 elements of ANSYS
R
spanning across the width of the plate, Fig. 7. SHELL63 is
a four-noded element with six DOFs per node. The segment has 20 nodes with a total of 120 DOFs. However, five DOFs are
removed from each of the four boundary nodes; these correspond to the three translations and the rotations about the y
1
and z
1
axes. Thus, the segment to be processed (as described in Section 2.1) has 100 DOFs.
For comparison, the whole structure is modelled with FE using elements of the same size. The full FE model has 5992
DOFs. If all the waves in the WFE model are retained, i.e., 100 (positive- and negative-going) waves, then the size of the
wave model will be 400 400. However, this is not advisable: reducing the size of the wave basis partly reduces the size
of the model, but also the calculation of the high-order wavemodes, which decay very rapidly with distance (by orders of
magnitude over the segments length) andthus have a negligible contributiontothe response, is very prone topoor numerical
conditioning [39]. If all the waves with |(k)| 1 are retained, then the WFE model of each plate will have nine pairs of
positive- and negative-going waves. Thus, the overall size of the model will be 72 72 which ensures numerical efficiency.
A force with unit amplitude is applied normal to the lower plate at (0.3, 0.08), and the response of the upper plate in the
transverse direction is observed at (0.5, 0.08). The transfer mobility is stiffness dominated at low frequencies, Fig. 8. The
FE/WFE results are in good agreement with the full FE results throughout the frequency range. Many of the resonances of
the structure are very close to those of a simply supported plate in transverse vibrations. As the frequency increases, the
effect of the axial vibration in the vertical plates grows and further resonances can be observed. At higher frequencies, the
response is seen to start converging towards the response of an infinite plate.
12 J.M. Renno, B.R. Mace / Wave Motion ( )
Fig. 8. Transfer mobility in the transverse direction of the upper plate:full FE, o FE/WFE.
Fig. 9. Schematic of a truss-cored train floor.
5.3. Train cross-section
Some train floor panels comprise aluminium plates with a truss-like core. Such a structure is periodic in the x-direction
and homogeneous in the y-direction, Fig. 9. The walls and ceiling of the train cabin are often of a similar construction.
Numerical values assumed are density = 2700, modulus of elasticity E = 7210
9
, Poisson ratio = 0.34, and structural
damping = 0.03.
To illustrate the applicability of the approach developed above for studying the behaviour of such structures, the model
of a section of a train cabin will be developed, based on an assumption of plane strain in the xz plane, i.e., no motion in the
y-direction. Thus, the section, Fig. 10, can be modelled using beam elements of an equivalent modulus of elasticity given as
E
eq
=
E(1 +)
1
2
.
The results obtained will be valid per unit width (in the y-direction). For simplicity (and without any loss of generality), the
walls and ceiling are assumed to be identical to the floor with the joints (also being identical) displayed in Fig. 10. The floor
and ceiling each comprise 25 cells (i.e., spanning 2.5 m), whereas each of the two walls comprises 20 cells (i.e., spanning
2 m). In the model of the cross-section shown in Fig. 10, the truss-cored sections are connected by joints at the corners, each
joint comprising six plates that are uniform in the y-direction.
The WFE model of the four waveguides (ceiling, floor, and two walls) can be realised by modelling a single cell
(highlighted in Fig. 10) using BEAM3 elements of ANSYS
R
(here with an element length of 1 cm or less). The FE model
J.M. Renno, B.R. Mace / Wave Motion ( ) 13
Fig. 10. Simplified cross-section of a train cabin: the joints and a single unit cell used in the WFE modelling are drawn using thick lines.
Fig. 11. Dispersion curves of the truss cored structure. The real and imaginary parts of each wavenumber are displayed in the same line (or marker) style
and colour.
of the cell has 102 DOFs, but the internal DOFs can be condensed reducing the size of the FE model of the cell to 12 DOFs
only; these are namely the displacements and rotations of the nodes on either sides of the cell. The FE model of the joints
(also made of aluminium) is obtained using BEAM3 elements of ANSYS
R
(with element length of 1 cm or less).
Fig. 11 shows the dispersion curves of the truss structure. The wavenumbers are normalised with respect to /, where
= 10 cm. Since this structure is periodic, rather than being a continuum, the WFE analysis predicts the propagation
constant of the periodic cell. The real part of the propagation constant is ambiguous to within an integral multiple of 2, i.e.,
if k is a valid wavenumber, so is k +2q (q here is an integer). The results shown in Fig. 11 are such that (k) .
At low frequencies (below 1100 Hz), the truss structure behaves like an orthotropic beam with three wave types:
propagating axial, propagating bending and evanescent bending waves. At higher frequencies, the periodicity of the truss
in the x-direction becomes apparent with stop-pass bands and higher-order waves cutting-on. These comprise motions in
the face and core beams and are difficult to divide into such simple types.
The approach presented in Section 3 is now used to compute the scattering properties of the joints. Then, a point force
(in the z-direction) is applied to the upper-face of the lower waveguide at 1.3 m from its left end. The response (in the x-
direction) is observed at the upper-face of the right waveguide at 1.3 m from its bottom end. For comparison, the whole
14 J.M. Renno, B.R. Mace / Wave Motion ( )
Fig. 12. Transfer mobility of the simplified train cabin:full FE, o FE/WFE.
cross-section is modelled using FE with elements of the same size. This yields a model of 4560 DOFs. The results of the
FE/WFE results, Fig. 12, are in very good agreement with the full FE results.
It should be noted here that the full FE solution can be obtained in the frequency domain directly by inverting the DSMof
the whole structure at each frequency. In this case, the FE/WFE approach has a clear advantage with a relatively very small
model to solve at each frequency. However, the full FE model can also be solved in the time domain to get the modes. In
this case, the computational cost of the full FE to obtain the modal solution is high, but relatively little effort is required to
find the response at each frequency. For the modal summation approach, the total cost depends on how many modes are
required to correctly represent the motion of the structure. Thus, a direct comparison between a full FE solution and the
FE/WFE approach is not conclusive.
6. Conclusions
This paper presents a review of the wave and finite element (WFE) method. Free and forced wave propagation were
first reviewed, and then the hybrid finite element/wave and finite element (FE/WFE) approach for obtaining the scattering
properties of complicated joints in arbitrary waveguide structures was presented. Then, an approach for modelling the
response of structural networks is presented. The waveguides are modelled using the WFE method where the FE model of a
small segment of the waveguide is post-processed using periodic structure theory to obtain the wave characteristics of the
waveguide. The joints are modelled using FE, and the WFE and FE models are coupled to calculate the scattering properties
of the joints. Consequently, the vibrational behaviour of the structure is described in the wave domain which has many
advantages. First, the resulting models are smaller in size which is useful at higher frequencies where FE models become
impractical. On the other hand, the knowledge of wave properties is useful for other applications (statistical energy analysis,
structure-borne sound, disturbance propagation, etc.). Numerical examples were presented to illustrate the approach.
Acknowledgements
The authors gratefully acknowledge the financial support provided by the Engineering and Physical Sciences
Research Council under grant number EP/F069391/1 and the European Commission in the context of the collaborative
project Mid-Mod: Mid-frequency vibro-acoustic modelling tools-innovative CAE methodologies to strengthen European
competitiveness (Grant agreement number: 218508).
References
[1] A.H. von Flotow, Disturbance propagation in structural networks, J. Sound Vib. 106 (1986) 433450.
[2] D.W. Miller, A.H. von Flotow, A travelling wave approach to power flow in structural networks, J. Sound Vib. 128 (1989) 145162.
[3] L.S. Beale, M.L. Accorsi, Power flow in two-and three-dimensional frame structures, J. Sound Vib. 185 (1995) 685702.
[4] R.L. Sack, Matrix Structural Analysis, PWS-Kent Publishing Company, 1989.
[5] T.H. Richards, Y.T. Leung, An accurate method in structural vibration analysis, J. Sound Vib. 55 (1977) 363376.
[6] R.S. Langley, Analysis of power flow in beams and frameworks using the direct stiffness method, J. Sound Vib. 136 (1990) 439452.
J.M. Renno, B.R. Mace / Wave Motion ( ) 15
[7] J.R. Banerjee, F.W. Williams, Coupled bending-torsional dynamic stiffness matrix for Timoshenko beamelements, Comput. Struct. 42 (1992) 301310.
[8] J.M. Cuschieri, Structural power-flow analysis using a mobility approach of an l-shaped plate, J. Acoust. Soc. Am. 87 (1990) 11591165.
[9] G.Q. Cai, Y.K. Lin, Wave propagation and scattering in structural networks, J. Engrg. Math. 117 (1991) 15551574.
[10] J.F. Doyle, Wave Propagation in Structures: Spectral Analysis Using Fast Discrete Fourier Transforms, second ed., Springer, 1997.
[11] B. Chouvion, C.H.J. Fox, S. McWilliam, A.A. Popov, In-plane free vibration analysis of combined ring-beam structural systems by wave propagation,
J. Sound Vib. 329 (2010) 50875104.
[12] B. Chouvion, A.A. Popov, S. McWilliam, C.H.J. Fox, Vibration modelling of complex waveguide structures, Comput. Struct. 89 (2011) 12531263.
[13] L. Cremer, M. Heckel, B. Petersson, Structure-Borne Sound, third ed., Springer, 2005.
[14] B.R. Mace, Wave reflection and transmission in beams, J. Sound Vib. 72 (1984) 237246.
[15] B.R. Mace, Reciprocity, conservation of energy and some properties of reflection and transmission coefficients, J. Sound Vib. 155 (1992) 375381.
[16] Y. Young, Y.K. Lin, Dynamic response analysis of truss-type structural networks: a wave propagation approach, J. Sound Vib. 156 (1992) 2754.
[17] S.V. Sorokin, N. Olhoff, O.A. Ershova, Analysis of the energy transmission in spatial piping systems with heavy internal fluid loading, J. Sound Vib. 310
(2008) 11411166.
[18] R.S. Langley, K. Heron, Elastic wave transmission through plate/beam junctions, J. Sound Vib. 143 (1990) 241253.
[19] I. Bosmans, T. Nightingale, Modeling vibrational energy transmission at bolted junctions between a plate and a stiffening rib, J. Acoust. Soc. Am. 109
(2001) 9991010.
[20] S. Walsh, R. White, Vibrational power transmission in curved beams, J. Sound Vib. 233 (2000) 455488.
[21] E.C.N. Wester, B.R. Mace, Wave component analysis of energy flow in complex structurespart I: a deterministic model, J. Sound Vib. 285 (2005)
209227.
[22] E.C.N. Wester, B.R. Mace, Wave component analysis of energy flow in complex structurespart II: ensemble statistics, J. Sound Vib. 285 (2005)
229250.
[23] E.C.N. Wester, B.R. Mace, Wave component analysis of energy flow in complex structurespart III: two coupled plates, J. Sound Vib. 285 (2005)
251265.
[24] K. Heron, Curved laminates and sandwich panels within predictive sea, in: Second International AutoSEA Users Conference, 1997.
[25] S.A. Hambric, Low-frequency measurements and pprediction of the structural-acoustic properties of the ince standard t-beamstructure, Noise Control
Eng. J. 50 (2002) 9099.
[26] J.M. Renno, B.R. Mace, Calculation of reflection and transmission coefficients of joints using a hybrid finite element/wave and finite element approach,
J. Sound Vib. 332 (2013) 21492164.
[27] A. Se-Knudsen, S.V. Sorokin, Analysis of linear elastic wave propagation in piping systems by a combination of the boundary integral equations
method and the finite element method, Contin. Mech. Thermodyn. 22 (2010) 647662.
[28] Y. Waki, On the application of finite element analysis to wave motion in one-dimensional waveguides, Ph.D. Thesis, Institute of Sound and Vibration
Research, University of Southampton, 2007.
[29] S.P. Shone, A flexural wave scattering method for damage detection in beams, Ph.D. Thesis, Institute of Sound and Vibration Research, University of
Southampton, 2006.
[30] B.R. Mace, D. Duhamel, M.J. Brennan, L. Hinke, Finite element prediction of wave motion in structural waveguides, J. Acoust. Soc. Am. 117 (2005)
28352843.
[31] D. Duhamel, B.R. Mace, M.J. Brennan, Finite element analysis of the vibrations of waveguides and periodic structures, J. Sound Vib. 294 (2006) 205220.
[32] L. Houillon, M. Ichchou, L. Jezequel, Wave motion in thin-walled structures, J. Sound Vib. 281 (2005) 483507.
[33] M. Maess, N. Wagner, L. Gaul, Dispersion curves of fluid filled elastic pipes by standard fe models and eigenpath analysis, J. Sound Vib. 296 (2006)
264276.
[34] J.-M. Mencik, M. Ichchou, Wave finite elements in guided elastodynamics with internal fluid, Internat. J. Solids Struct. 44 (2007) 21482167.
[35] B.R. Mace, E. Manconi, Modelling wave propagation in two-dimensional structures using finite element analysis, J. Sound Vib. 318 (2008) 884902.
[36] E. Manconi, B.R. Mace, Wave characterization of cylindrical and curved panels using a finite element method, J. Acoust. Soc. Am. 125 (2009) 154163.
[37] J.M. Renno, B.R. Mace, On the forced response of waveguides using the wave and finite element method, J. Sound Vib. 329 (2010) 54745488.
[38] J.M. Renno, B.R. Mace, Calculating the forced response of two-dimensional homogeneous media using the wave and finite element method, J. Sound
Vib. 330 (2011) 59135927.
[39] Y. Waki, B.R. Mace, M.J. Brennan, Numerical issues concerning the wave and finite element method for free and forced vibrations of waveguides,
J. Sound Vib. 327 (2009) 92108.
[40] W. Zhong, F. Williams, On the direct solution of wave propagation for repetitive structures, J. Sound Vib. 181 (1995) 485501.
[41] W. Zhong, F. Williams, A.Y.T. Leung, Symplectic analysis for periodical electro-magnetic waveguides, J. Sound Vib. 267 (2003) 227244.
[42] D.J. Ewins, Modal Testing: Theory, Practice and Application, in: Mechanical Engineering Research Studies: Engineering Dynamics Series, Wiley, 2001.
[43] S. Gnutzmann, U. Smilansky, Quantum graphs: applications to quantum chaos and universal spectral statistics, Adv. Phys. 55 (2006) 527625.
[44] M. Petyt, Introduction to Finite Element Vibration Analysis, Cambridge University Press, New York, 1990.
[45] J.R. Banerjee, Dynamic stiffness formulation and its application for a combined beamand a two degree-of-freedomsystem, Trans. ASME J. Vib. Acoust.
125 (2003) 351358.

Вам также может понравиться