Sei sulla pagina 1di 9

Dispersion and excitability of guided acoustic waves in isotropic beams with arbitrary

cross section
P. Wilcox, M. Evans, O. Diligent, M. Lowe, and P. Cawley

Citation: 615, (2002); doi: 10.1063/1.1472800


View online: http://dx.doi.org/10.1063/1.1472800
View Table of Contents: http://aip.scitation.org/toc/apc/615/1
Published by the American Institute of Physics
DISPERSION AND EXCITABILITY OF GUIDED ACOUSTIC WAVES
IN ISOTROPIC BEAMS WITH ARBITRARY CROSS SECTION

P. Wilcox, M. Evans, O. Diligent, M. Lowe and P. Cawley

Department of Mechanical Engineering, Imperial College, London SW7 2BX, UK.

Abstract. A finite element (FE) technique for computing the properties of guided waves that can exist
in an isotropic beam of arbitrary cross section is presented. The FE model uses a two-dimensional
mesh to represent a cross section through the beam and cyclic axial symmetry conditions to prescribe
the displacement field perpendicular to the mesh. FE results are presented for plate and angle sections.
Excitability functions are calculated and implications for transducer placement are considered.

INTRODUCTION
Guided acoustic waves may be used to provide a highly efficient method of rapidly
inspecting certain types of structures (see for example the review paper by Chimenti [1]).
The problems associated with using guided waves are well documented [2] and include the
problems of multiple modes and dispersion. The simplest structure in which guided waves
can propagate is a free, flat isotropic plate and the analytic modal solution to wave
propagation in this system was found in 1917 by Lamb [3]. However, the practical
inspection of even a two-dimensional (2D) plate-like structure using guided waves is
further complicated by the fact that guided waves can propagate in an infinite number of
directions within the structure. Much more commercial success [4,5] in guided wave
inspection has been achieved in one-dimensional (ID) structures such as pipes where the
propagation of guided waves is limited to two directions - forwards and backwards.
Paradoxically, the modelling of guided waves in a ID waveguide is generally more
complicated than in a 2D system. For this reason, analytic models of ID waveguides have
so far been limited to axi-symmetric [6] and rectangular ones [7]. Clearly, there are many
beam-like structures such as T beams, channel sections and railway tracks in which guided
waves will propagate that cannot be modelled analytically.
Guided wave propagation in any structure can be modelled using a variety of
numerical methods, such as the time-marching finite element (FE) technique. Such models
are of great practical use in predicting, for example, the reflection coefficient of a guided
wave mode from a defect, when the modal properties (i.e. mode shape, phase velocity and
group velocity) of the guided wave modes are already known. However, the extraction of
the modal properties of guided waves from such models is exceedingly difficult. To date,
there has been surprisingly little work in the NDE community on using finite element
methods to compute the modal properties of guided waves in a structure. This is in marked
contrast to the amount of work devoted to providing analytic solutions to increasingly

CP615, Review of Quantitative Nondestructive Evaluation Vol, 21, ed. by D. O. Thompson and D. E. Chimenti
© 2002 American Institute of Physics 0-7354-006 l-5C702/$ 19.00
203
complicated planar structures incorporating multiple layers, anisotropy and material
damping. However Gavric [8] presents a numerical technique that is used to predict the
modal properties of guided waves that can exist in a ID waveguide of arbitrary cross
section, in that particular case for a railway track.
In this paper, a method of rapidly implementing the numerical technique proposed by
Gavric using a standard FE package is presented. The results from this numerical model are
compared to analytical results from a flat plate. Results are also presented for guided wave
propagation in an angle section beam as a demonstration of the power of the technique for
predicting modal solutions in beams with arbitrary cross section.

MODELLING

Previous Work
The method of calculating the modal properties of guided waves in a beam of
constant but irregular cross section is described in detail in the paper by Gavric [8] and is
only briefly summarised here. The basic idea is to create a 2D mesh of the cross section of
the beam as shown in Figure l(a). Then a special stress-strain condition is specified in the
direction perpendicular to the mesh that corresponds to the spatially periodic stresses and
strains associated with a guided wave propagating in that direction. For a specified spatial
period of stresses and strains, an Eigenvalue technique is used to extract the natural
frequencies of the mesh. These natural frequencies are those at which guided waves with a
wavelength equal to the specified spatial period can exist. The corresponding Eigenvectors
give the displacement mode shape associated with each frequency.
Implementation
A reason why the Gavric approach has not been more widely investigated may be that
it requires a non-standard stress-strain condition to be specified and therefore is not
straightforward to implement on a typical FE package. Gavric wrote his own FE code in
order to solve the problem. However, there is a way to obtain a very close approximation to
the required stress-strain condition that is available on standard FE packages and that it is to
use an cyclic axial symmetry condition as shown in Figure l(b).
(a) 2D mesh of
cross section
through
waveguide

, 2D mesh of cross section


through waveguide I
FIGURE 1. (a) The basic principle of using a 2D FE mesh to model the propagation of guided waves
perpendicular to the mesh, (b) Approximate implementation of (a) using a 2D FE mesh and a cyclic axial
symmetry condition.

204
Both the basic axial symmetry condition and the cyclic axial symmetry condition
mean that a 2D FE mesh represents a cross section through a 3D object that has axial
symmetry about a specified axis that lies in the plane of the mesh. In the current context,
the 2D mesh will be a mesh representing a cross section through the waveguide of interest,
and the axis of curvature will be made to lie as far away from the mesh as the numerical
restrictions of the FE package will allow. Clearly the 3D object modelled by the FE mesh
will be an annulus with the desired cross section, rather than a straight beam. Thus the
approximation in this implementation is that the modal solutions are actually for guided
waves propagating in a curved waveguide rather than a straight one. In practice, the radius
of curvature can be made very large compared to the dimensions of the mesh and the results
converge fairly rapidly to those for a straight waveguide.
On its own, an axial symmetry condition implies that in the 3D object all
deformations are functions only of position in the cross section and are the same at all
angles. Axial symmetry may be regarded as zeroth order cyclic symmetry. In cyclic
symmetry, the displacement distribution is periodic in the angular direction with one or
more periods per revolution. If 9 is the angular direction then the displacement, u, in the
radial, r, and axial, z, directions will have the following form:

u =

where n is the order of cyclic symmetry and Ur and Uz are functions of r and z only. By
using symmetry arguments it can be shown that while the angular displacement UQ is still
periodic, it must be 90 degrees out of phase with the other two displacement components:

ue =U0r,z)smn0 (2)

where U# is again a function of r and z only. For a guided wave propagating in a lossless
isotropic waveguide, the displacements in the direction of wave propagation are always 90
degrees out of phase with the displacements perpendicular to it [9]. Hence a 2D FE model
with cyclic axial symmetry specified automatically has the correct relationship between
displacements in the r, 9 and z directions if the direction of wave propagation is the 9
direction.
The natural frequencies of a 2D model with cyclic axial symmetry of a certain order
can be solved using an Eigensolver routine that most good FE packages will contain. Here
the FE package Finel [10] is used. Knowing the radius of curvature (i.e. the location of the
axis of symmetry) and the circular order specified, it is straightforward to work out the
wavelength of the guided waves in a specific model. Hence a single model yields a number
of frequencies at which guided waves of the specified wavelength can propagate. By
finding the Eigenvalues of a number of different models with different orders of cyclic
symmetry specified, it is possible to build up a number of points in wavelength-frequency
space at which modal solutions to guided wave propagation in the waveguide exist. The
method only returns discrete points in wavelength-frequency space. It does not provide
continuous lines representing modes. The task of joining together points that lie on the
same mode in order to generate dispersion curves must be done externally. Currently a
routine written in Matlab is used that compares mode shapes at different points in order to
work out which points to join together.
It should be noted that this method cannot find non-propagating solutions as the
wavenumber is implicitly forced to be real by the constraints imposed by using the cyclic
axial symmetry approximation. The custom written FE realisation created by Gavric

205
allowed complex wavenumbers to be specified and could therefore also compute points on
the non-propagating branches of dispersion curves.
Excitability
The modal quantities of wavenumber and frequency can be manipulated to plot
dispersion curves of phase and group velocity as functions of frequency. The dispersion
curves are exceedingly useful for identifying the modes that can exist at a certain frequency
in a free waveguide, but they do not give any indication of the degree to which different
modes will be excited by a particular transduction system. The latter information is
essential for the development of a practical guided wave inspection system, as it determines
the type and location of transducers that must be used. To this end, the concept of
excitability is introduced.
Consider the case when a harmonic force in an arbitrary direction is applied at a
particular point on the cross section of a waveguide. The force will cause all the guided
wave modes that can exist at that frequency to be excited to varying degrees. The amplitude
of a particular mode may be defined in terms of its displacement at any one point in the
cross section, since the relative displacement between all the points in the cross section is
fixed and defined by the mode shape. For convenience, the excitability of a particular mode
is defined [11] as the ratio of displacement of that mode to applied force when both
quantities are measured at the same location and direction in the cross section. The
excitability, E9 of a mode at a particular point in the waveguide cross section and in a
particular direction can be obtained from the mode shape using the following relationship:

EocU2f (3)

where / is the frequency and u is the displacement in the power flow normalised
displacement mode shape at the point and direction of interest. The mode shapes that the
FE model yields are not power flow normalised, but this can be done relatively
straightforwardly since in the case of guided waves propagating in a lossless waveguide,
the power flow, P, for a particular mode at a particular frequency,/, is:
nodes

f~/2(»,-»,) (4)
where vgr is the group velocity of the mode, ra, is the lumped mass represented by each
node in the FE model and Ui is the displacement vector of each node in the mode shape.
The excitability in conjunction with the dispersion curves enables a complete guided
wave transduction system to be modelled, as the effect of multiple transmitting transducers
can be incorporated by performing the appropriate superposition of forces applied at
discrete points on the surface of a waveguide.

RESULTS

Flat Plate Validation


To validate the cyclic axial symmetry approach, models of nominally straight crested
guided waves in a free and nominally flat 1 mm thick aluminium plate were compared with
analytic Lamb wave solutions. Two possible FE model configurations were examined, the
difference being in the position of the axis of symmetry with respect to the orientation of
the 2D mesh. A complication of modelling the basic Lamb wave case is that the flat plate
does not have a finite cross section, as the wavefronts are assumed to be infinitely long. To

206
include this in the FE model, a single row of 100 elements, each 0.01 mm square, is used to
model a strip through the thickness of the plate and suitable boundary conditions are then
applied to the nodes on either side of the strip. In the first model, the axis of symmetry is
placed at right angles to the strip of elements and 500 mm away. The boundary condition
imposed on the edge nodes in this case is that uz = 0 as shown in Figure 2(a). In this
configuration, the 3D object that the mesh represents is an infinitely long thin walled pipe
and the guided wave solutions that will be produced are for straight crested guided waves
propagating around the circumference of the pipe. Hence the approximation used here is
that the flat plate is modelled as a lightly curved one. In the second configuration shown in
Figure 2(b), the axis of symmetry is parallel to the strip of elements in the mesh and
500 mm away. The boundary condition ur = 0 is applied at the edge nodes. The 3D object
being modelled here is therefore a flat plate and the approximation is that the wavefronts of
the guided waves that can exist are not parallel. Instead the wavefronts must all converge at
the axis of symmetry and the direction of wave propagation is around that axis.
The two models shown in Figures 2(a) and (b) were used to generate the dispersion
curves shown in Figures 3 (a) and (b) respectively. The circles on the graphs indicate points
generated by the FE models and the solid lines indicate dispersion curves traced using the
analytic solutions using Disperse [12]. It can be seen that over the frequency range shown,
the agreement between both FE models and the analytic solutions is very good.
The FE models were also used to obtain predictions of excitability for the case of a
normal surface force applied at the surface of the plate. Analytic solutions to this case can
also be obtained using the method described by Viktorov [13]. The FE results for both
models and the analytic results are shown in Figures 3(c) and (d). It can be seen that at
frequencies below about 4 MHz, the agreement between both models and the analytic
predictions are good. However, above this frequency, the excitability results from the
curved plate FE model peel away from the analytic results, whereas the ones from the non-
parallel wavefront model remain in good agreement. This is thought to be because the
excitability calculation is very sensitive to the predicted mode shapes. In the case of the
curved plate FE model, symmetry is not preserved about the centreline of the plate, hence
the mode shapes themselves are distorted particularly at high frequencies. This example
illustrates that some care must be taken when using the cyclic axial symmetry model,
especially at high frequencies.
(a) |————i———| (b)

2D mesh

500mm

Boundary conditions:.
u = 0 on these edges
FIGURE 2. Two model geometries used to approximately model Lamb waves in a nominally flat isotropic
plate. The axis of symmetry is arranged so that (a) the plate is very slightly curved and the wavefronts are
straight and (b) so that the plate is flat, but the wavefronts are not parallel.

207
'0 Frequency (MHz) 3 0 Frequency (MHz) 8
FIGURE 3. Comparison of results obtained using FE models (circles) with analytic results (solid lines), (a)
Phase velocity from curved plate model; (b) phase velocity from non-parallel wavefront model; (c)
excitability from curved plate model; (d) excitability from non-parallel wavefront model.

Angle Section
The case of guided waves propagating in an angle section is now examined as an
example of the more complex type of geometry that may be modelled using the FE
technique. The angle section considered is made from aluminium with arms 5 mm long and
1 mm thick as shown in Figure 4. It is meshed using 40 elements each 0.5 mm square and
the axis of symmetry is placed 2 m away from the mesh.
The model is used to predict the phase velocity dispersion curves for the angle section
and these are shown in Figure 5. In order to gain an understanding of the sort of modes that
can propagate, the mode shapes of the five modes that can exist at 150 kHz are also shown.
It can be seen that modes 1 and 3 do not have any significant deformation of the cross
section. These modes are the fundamental extensional and torsional modes respectively.
The remaining modes involve flexure of the arms of the angle section combined with gross
bending or torsion of the entire cross section.

1 mm

1 >
~ | 2D mesh ^^^
E
E
in
*5 1
.<2 i -+H+- ^
<£ 1
1 2000 mm 5mm
^———————i

FIGURE 4. Geometry and mesh used to model angle section.

208
0 Frequency (MHz) 0.4
FIGURE 5. Phase velocity dispersion curves for the angle section shown in Figure 4 and the mode shapes of
the five modes that can exist at 150 kHz. The undeformed mesh is shown in grey.

The remaining modes involve significant deformation of the cross section. The use of
the model to investigate the effect of transducer placement on the angle section has also
been investigated. In the first instance, the optimisation of transducer location to excite a
particular mode is considered. For the purposes of this example, the mode labelled mode 5
at 150kHz is chosen as the guided wave mode of interest. Six candidate transducer
locations around the perimeter of the section are used, and in each of these locations, three
transducer orientations are considered giving a total of eighteen different transducer
positions as shown in Figure 6(a).
(a) Locations:
ff

Orientations:
A-normal
B - in-plane parallel to axis 1
2 ' 3 ' 4
C - in-plane perpendcular to axis Transducer location

(c)

Model Mode 2 Mode 3 Mode 4 Mode 5


FIGURE 6. (a) Candidate transducer locations and orientations on angle section, (b) Relative excitability of
mode 5 at 150 kHz when excited by transducers positions in (a), (c) Relative excitability of all 5 modes at
150 kHz when the angle section is excited by a single transducer at the position shown.

209
The orientations considered are (A) transducer applying normal traction to surface,
(B) transducer applying shear traction to surface in axial direction and (C) transducer
applying shear traction to surface at right angles to axial direction. The relative excitabilites
of the mode of interest for transducers at each of these locations are plotted in Figure 6(b).
Not surprisingly perhaps, the best location for a transducer to excite this mode is at the tip
of one of the arms of the angle section and orientated so that it applies a shear traction at
right angles to the axial direction. It could be argued that this conclusion is at least
qualitatively obvious just be looking at the mode shape of the mode. Where the excitability
proves really useful is when examining the relative excitability between different modes.
As an example of this, the case of a transducer located at the previously identified optimum
position for mode 5 at 150 kHz is considered. The relative excitabilities of all five possible
modes are shown in Figure 6(c). It can be seen that the transducer excites four out of the
five modes reasonably efficiently. In a practical application where it is desirable to excite
only a single mode [2], a single transducer at this location would not be sufficient. In
practice, the same technique is used to optimise the number and location of multiple
transducers in order to achieve the desired level of modal selectivity.

CONCLUSIONS
The power of the 2D FE technique proposed by Gavric has been demonstrated using
an approximation that can enables it to be implement in a standard FE package. It has been
shown how the excitability of different modes in addition to the dispersion curves may also
be extracted from the data. The potential of the 2D FE technique is thought to have
considerable potential not only for isotropic beams but also for more complex anisotropic
structures that cannot be easily modelled analytically.

ACKNOWLEDGEMENTS
We would like to express our thanks to Prof. Richard Weaver at the University of
Illinois who directed our attention towards the work of Gavric.

REFERENCES
1. Chimenti, D. E., Appl. Mech. Rev., 50(5), 247-284 (1997).
2. Alleyne, D. N. and Cawley, P., NOT and E Int., 25, 11-22 (1992).
3. Lamb, H., Proc. Royal Soc., A93, 114-128, (1917).
4. Rose, J. L., Ditri, J. J., Pilarski, A., Rajana, K. and Carr, F., NDT and E Int., 27(6),
307-310(1994).
5. Alleyne, D. Pavlakovic, B., Lowe, M. and Cawley, P., in Review of Progress in
QNDE, Vol. 20A, eds. Thompson, D. O. and Chimenti, D. E. (American Institute of
Physics, New York, 2001), 180-187.
6. Gazis, D. C., /. Acoust. Soc. Am., 31(5), 568-578 (1959).
7. Kastrzhitskaya, E. V. and Meleshko, V. V., Sov. Appl. Mech. 26, 773-781 (1991).
8. Gavric, L., /. Sound and Vib., 185(3), 531-543 (1995).
9. Lowe, M., IEEE Trans. Ultrason., Ferroelec. Freq. Cont., 42(4), 525-542 (1995).
10. Hitchings, D., FE77 User Manual (2.40), Imperial College, Dept. of Aeronautics,
London, UK, 1995.
11. Wilcox, P. , Monkhouse, R., Lowe, M. and Cawley, P., in Review of Progress in
QNDE, Vol. 17A, op. cit. (Plenum Press, New York, 1998), 915-922.
12. Pavlakovic, B., Lowe, M., Alleyne, D. and Cawley, P., in Review of Progress in
QNDE, Vol. 16A, op. cit. (Plenum Press, New York, 1997), 185-192.
13. Viktorov, I. A., Rayleigh and Lamb waves: Physical Theory and Applications,
Plenum Press, New York, 1967.

210

Potrebbero piacerti anche