You are on page 1of 19

INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719


Published online 25 June 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/nag.643

Calculation of vibration transmission over bedrock using


a waveguide finite element model

A. T. Peplow, and S. Finnveden


Marcus Wallenberg Laboratory for Sound and Vibration Research, Department of Aeronautics and Vehicle
Engineering, Kungliga Tekniska Hogskolan (KTH), S-100 44 Stockholm, Sweden

SUMMARY
A finite element method is developed for the study of elastic wave propagation in layered ground
environments. The formulation is based on a spectral finite-element approach using a mixture of high-
order element shape functions and wave solutions. The numerical method provides solutions to vibration
transmission on and within layered elastic waveguides. Examples of its use include the theoretical analysis
of transmission of vibrations in the vicinity of the surface of the ground. The mathematical model is two
dimensional, and the interior of the ground is modelled as an elastic layer overlying a rigid foundation. An
analysis of the natural modes of free vibration in a single layer and two layers is presented and compared
with known results. In addition the forced response of the layers, for which the surface is assumed to
be subjected to a harmonic point force load is shown. These results also include an illustration of the
attenuation of surface vibration due to wave impedance blocks in the near field of the source up to a
frequency of 200 Hz for two soil types. Copyright q 2007 John Wiley & Sons, Ltd.

Received 17 August 2006; Revised 19 May 2007; Accepted 19 May 2007

KEY WORDS: wave propagation; layered media; spectral methods; linear elastic material; spectral finite
elements

1. INTRODUCTION

The spectral finite element method (SFEM) is applied to an elastic waveguide problem consisting
of layered material strata. The term spectral is used here to emphasize the underlying nature of the
finite element basis functions; determined from solutions of an eigenvalue problem. Although the
approximate functions are a mixture of polynomials and spectral wave functions, the finite element
approach may be placed alongside classical finite formulations and more modern approaches.

Correspondence to: A. T. Peplow, Marcus Wallenberg Laboratory for Sound and Vibration Research, Department
of Aeronautics and Vehicle Engineering, Kungliga Tekniska Hogskolan (KTH), S-100 44 Stockholm, Sweden.

E-mail: atpeplow@kth.se

Contract/grant sponsor: Marie Curie Individual Fellowship; contract/grant number: HPMF-CT-2000-01080

Copyright q 2007 John Wiley & Sons, Ltd.


702 A. T. PEPLOW AND S. FINNVEDEN

The transmission of vibrations in the near field of the surface of the layered ground strata
is investigated theoretically. The mathematical model is two dimensional, and the interior of the
ground is modelled as homogeneous, isotropic elastic layers over a rigid foundation (or bedrock).
Firstly, stiffness matrices for the layer are derived; the free vibration in undamped layers are
considered next; then validation of forced responses for damped layers, and lastly the attenuation
of vibration due to wave impedance blocks (WIBs) is investigated. For the forced response, the load
is modelled as an infinite strip, so that the problem is plane. For this reason, the free vibration is
also treated as a plane problem. The purpose of the present study is to evaluate a new computational
method for predicting vibration transmission, in particular its attenuation through uniform stratified
and non-uniform stratified layers.
A possible vibration attenuation device that has showed some promise is the WIB. The principle
of this is to modify the modal wave propagation regime of the ground by introducing an artificial
stiffened layer (inclusion) near the load or receiving structure. The performance of the inclusion in
impeding wave transmissions at a number of receiver positions is studied and measured in terms
of insertion loss as an illustration of the model capabilities. Takemiya and Fujiwara [1] and Petyt
et al. [2] have studied the idea of a WIB using a boundary element (BE) model. Moreover in [1] the
response on the surface of the ground in the time domain with a Ricker wavelet type loading was
studied and the authors concluded that an effective width of the WIB should be chosen in relation
to the wavelength of the given load. More recently Takemiya [3] has considered a honeycomb
style WIB placed in the soil to mitigate high-speed train induced vibrations. In this article, an
alternative numerical model to the previous reference is presented which enables the wave field in
the region of the inclusion above the bedrock to be determined accurately.
Previous workers have studied free vibration for the bedrock model. Waas [4] provided a
simple method of predicting the natural frequencies of the layer for zero wavenumber. Tolstoy
and Usdin [5] used the interference principle of ray theory to derive the period equations giving
the wavenumbers of the natural frequencies. The interference principle is based on the equality of
amplitude at a given depth of each of the compression and shear waves, which is the requirement for
unattenuated propagation. The authors also considered the variation of wavenumber or wavespeed
with frequency, known as dispersion curves, for the first two propagating modes in the layer using
the interference principle, similar to the work by Kuhlemeyer [6]. Kobori et al. [7] also studied
the free vibration bedrock model as part of their three-dimensional work. Dispersion curves, in
common with those found here were presented.
For the forced response of a bedrock model, Warburton [8] looked at the direct receptance
of a rigid disc load by approximating the stress distribution under the load by a static solution.
For the rigid disc problem, Waas [4] used a finite element method to study the direct receptance
of a forced rigid strip or disc and Segol et al. [9] used the method to study the reduction of
surface waves due to trenches. The now well-known method by Kausel and Roesset [10] calcu-
lates the three-dimensional dynamic stiffness matrix, relating stress to displacement, via Fourier
transform methods and has been applied by Kausel and Peek [11] to find the vertical direct re-
ceptance for a disc load. Until Jones and Petyt [12] none of the workers referenced above have
considered the variation of surface displacement with distance of the bedrock model. Some of
the results of Jones and Petyt have been repeated here for direct comparison for free and forced
response results.
The spectral finite element method has been applied, in various forms, to a number of acoustic
and structural waveguide problems by the authors. In particular, the method used by Finnveden
[13, 14] can be viewed as a merger of the dynamic stiffness method and the finite element method.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 703

The principle of the method is based on a variational formulation for non-conservative motion in
the frequency domain.
The SFEM has also been used to study vibration in beam frame works [13], for fluid-filled
pipes [14], and recently for acoustic ducts [15]. Use of a variational formulation for the spectral
method provides a natural basis for approximations and a straightforward method for combination
with standard finite elements. Recently, the method has been applied in various guises to generate
dispersion properties of elastic waveguides in three dimensions by Taweel et al. [16], Widehammar
et al. [17], and Adamou and Craster [18]. The SFEM approach here, applied to layered media,
has particular similarities to the thin-layer method (TLM) first used by Lysmer [19] and developed
by Kausel [20] for the time domain and Park and Kausel [21] for investigation of the numerical
dispersion properties. In this article, the solutions of the dispersion relations are used in the set of
finite element basis functions. As the user may observe the wave dispersion curves before turning
to a full solution, this allows flexibility in both modelling and choosing design parameters. The
paper is organized as follows. In Section 2, the variational principle for non-conservative acoustic
motion is stated. In Section 3, a description of the trial functions is given and associated dynamic
stiffness matrices are presented. Dispersion relations and elastic responses for various complicated
geometries are shown in Section 4 and some conclusions are given in Section 5.

2. THEORETICAL MODEL

In the sections that follow, the governing equations are stated and the new numerical approach
is written in detail. To throw a little light on the new method, a short section on general finite
element modelling is given below.

2.1. Finite element modelling for elastic layered media


The basic field equations of elasticity are given by the momentum balance equations, the strain
displacement equations, and the constitutive equations. The finite element method is based on a
variational form of the fundamental equations and to formulate a variational statement a weak form
is generated by considering a balance between potential and kinetic energies, see Petyt [22]. The
finite element approximation now follows by discretizing the domain into nodes and elements so
that the solution space and the weighting functions are defined on a finite number of elements. First,
the solution space and the weighting function space are replaced by finite dimensional subspaces.
Second, these subspaces are employed in the weak form to generate matrix equations which are
then used to solve for the unknown displacements at the element nodes. For dynamic problems,
where short wavelengths are often seen a fine mesh with many nodes is often required to capture the
wave-type solutions. Thus the number of degrees of freedom and finite element matrices increase
in size as wavelengths become shorter. Equally if the domain size is large then a large number
of nodes are also required across the original domain. Consider now a horizontally stratified,
elastic waveguide such as layered bedrock ground subjected to dynamic forces somewhere on the
surface of the medium. Within each layer of constant thickness the elastic material is isotropic. The
medium is bounded by an upper and lower interface at which either stresses or displacements are
prescribed. If the layers are not thin, with respect to horizontal length, and are of finite length then
it is possible to carry out a finite element discretization for the displacement fields in the vertical
and longitudinal directions. A finite element solution for the unknown displacements is well known

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
704 A. T. PEPLOW AND S. FINNVEDEN

and easily computable. However, if the domain is extremely long the number of unknowns required
becomes unfeasible. Here a new (spectral) finite element which takes into account the waviness
of the solution and accommodating the arbitrary length of the domain becomes desirable.
The basic idea of employing the spectral finite element method for elastic solid waveguides
is to discretize the waveguide into a finite number of rectangular elements. For each of these
elements, it is possible to define the virtual work as a function of frequency for each local element.
For plane strain conditions, without a force, this modifies the wave equations involving partial
derivatives with respect to x, z into a system of differential equations. Subsequently, a combination
of approximating functions form a complete set for the finite element subspace; polynomials in
the vertical direction and moreover wave influence functions in the horizontal direction. This is
the subject of Section 3.1 and the derivation of the numerical wave functions is in Section 3.2.
By assembling local dynamic stiffness matrices, a system of equations representing the elastic
waveguide is derived, see Equation (31).
The spectral finite element method for elastic wave propagation through a layered medium is
described in detail in the following sections.

2.2. Governing equations and the variational statement


A typical elastic rectangular waveguide in the xz plane is illustrated in Figure 1. It is assumed that
individual rectangular regions of the waveguide, as illustrated in Figure 2, may possess different
ambient material properties, similar to standard finite element practice. Elastic wave propagation

F x1 x2
x0

L0 H0
z H1
D Block

W H2

Figure 1. Schema for a waveguide application including a wave impedance blocking mass.

Force

Block
j

Figure 2. Finite element geometry for a waveguide application in Figure 1. Solid


circles denote finite element nodes.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 705

through the inhomogeneous waveguide, , with various geometries and material properties is
then sought.
Displacements, stresses and strains in a particular rectangular region, Figure 2, may be denoted
by u(x, t), r(x, t), and e(x, t), respectively, and x = (x, z). To represent a wave propagating in
the direction of the waveguide, displacements of the form: u(x, t) = Re{U(z)ei(tx) }, where U
represents a cross-sectional mode shape,  the propagation constant and  the angular frequency.
The formulation of the governing equations is based on two-dimensional linear elasticity, see
Petyt [22] and Szabo and Babuska [23]. The mechanical variables are the displacement u, stress r,
and strain e with components

{u} = (u, w)T , {r} = (x x , zz , x z )T , {e} = (x x , zz , x z )T (1)

The stressstrain constitutive relations for each sub-region material are given by

{r} = [C]{e} (2)

where [C], the symmetric (3 3) matrix, contains the rectilinear elastic moduli. The linear strain
displacement equations are written in split-operator form according to the coordinates, (x, z):
 
*
{e} = L{u} = z + {u} (3)
*x

where



1
*

z = *z , = (4)

* 1

*z
The governing equations of motion for the elastic material are derived by Hamiltons principle [23]:
t1  
 T (U + V ) d dt = 0 (5)
t0 V

where T is the kinetic energy, U the strain energy and V the potential energy due to external
forces. The kinetic energy T in terms of the velocity vector u, and strain energy U in terms of the
stress, are given by

1 1
T= {u}T q{u} d, U = {e}T [C]{e} d (6)
2  2 

where q is the mass density matrix,





q= (7)


Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
706 A. T. PEPLOW AND S. FINNVEDEN

and the stiffness matrix, [C] is, for a plane-strain problem, given by

(1 )   + 2 
E
C=  (1 ) =   + 2
(1 + )(1 2)
(1 2)/2

where denotes shear modulus, E Youngs modulus,  Lame constant, and  the Poisson ratio.
The potential energy V henceforth consists of applied traction acting on the boundary

V = {u}T {rn } dx (8)
j

where rn denotes normal and shear tractions at the boundary. A modified version of Hamiltons
principle in the frequency domain will be used here for the dissipative motion, see Reference [13],
since material is modelled through a complex form of Youngs modulus, E(1 + i
) where
>0
represents damping as a loss factor. Performing the variational operation on Equation (5), and
carrying out a partial integration with respect to time, and assuming harmonic time dependence,
eit , one obtains

(L) = (2 {ua }T q{u} (L{ua })T [C]L{u}) d = 0 (9)


where superscript .adenotes complex conjugate in the adjoint system having negative damping
and virtual quantities are denoted by .
It is understood that the stiffness matrix [C] takes a complex form in the following analysis and
calculations. In Equation (9), only the adjoint of the displacement functions and their derivatives
are being varied. During the variational process, the lateral and vertical displacements u and w and
their derivatives are viewed as being independent from their complex conjugates. The underlying
differential equation is linear and hence this approach is in effect equivalent to varying the real
and imaginary parts of the functions independently.

3. SPECTRAL FINITE ELEMENT MODELLING

In this section, the wave influence functions are derived and the local dynamic stiffness matrix is
presented.

3.1. Numerical discretization


In elastic waveguides with constant geometrical cross-section, the solutions of the governing
equations of motion are a combination of exponential terms describing the wave propagation in
the horizontal direction with cross-sectional modes in the vertical direction.
Thus, the axial x-dependence is separable from the cross-sectional z-dependencies. In this case,
the application of the finite element method, the displacement field of the cross-section of a
rectangular element may be represented by high-order polynomials, N j (z), j = 1, . . . , M, Peplow
and Finnveden [15]. For in-plane motion a combination of polynomials is assumed. In the finite
element discretization, the cross-section C j = [z j,1 , z j,M+1 ] is divided so that material parameters

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 707

are constant within each interval. Within one super-spectral finite element,  j for example, the
in-plane displacement field may be represented in the co-ordinate system such that a dimensionless
2 2m real-valued matrix [N(z)] with the given co-ordinate functions as elements, and a 2m 1
vector u(x) have the dimension of length,
u(x, z) = [N(z)]{u(x)} (10)
where the displacement function, u(x, z) = [u(x, z), w(x, z)]T ,

u(x, z) = [N1 (z), 0, N2 (z), . . . , 0, N M (z), 0]{u(x)}


w(x, z) = [0, N1 (z), 0, . . . , N M1 (z), 0, N M (z)]{u(x)}

using M piecewise polynomial shape functions,


compactly supported over C j , in the vertical
direction. Elemental shape functions u(x) will be derived later in the analysis.
The separation of variables leads to the virtual quantities u = [N]u. With the assumed operator
splitting in (3) this gives

du and u = ( z N)u + (N)


u = ( z N)u + (N) du (11)
dx dx
By inserting the displacement functions according to Equations (10) and (11) into the virtual work
Equation (9) yields

2 {ua }T NT qN{u}

 
d{u}
{u } (z N) [C] (z N){u} + (N)
a T T
dx
 
d{ua }T T d{u} d
(N) [C] ( z N){u} + (N) (12)
dx dx

Partial integration in the x-direction, assuming one super-spectral element rectangle takes the
form  = H T , leads to an expression:

{ua }T 2 NT qN{u} ( z N)T [C]( z N){u}
T H

d{u} + (N)
( z N)T [C](N) T [C]( z N) d{u}
dx dx

d {u} dH dx
2
T [C](N)
+ (N)
dx 2
  
a T d{u}
{u } (N) [C] (z N){u} + (N)
T
dx = 0 (13)
T dx H

The final integral expression in Equation (13) provides boundary conditions at the cross-section
extremities. It is clear that this is the natural boundary condition describing a stress-free condition.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
708 A. T. PEPLOW AND S. FINNVEDEN

Since u is arbitrary and u is a function of the axial direction only, Equation (13) yields a system
of second-order differential equations:
 

{ N qN + (z N) [C](z N)} dH {u}
2 T T
H
 
T [C]( z N) ( z N)T [C](N)}
d{u}
+ {(N) dH
H dx
 
d2 {u}
{(N) [C](N)}
T
dH
H dx 2
  
d{u}
+ (N) [C] (z N){u} + (N)
T
=0 (14)
dx H

Hence the formulation involves evaluating the derivatives and the integrals over z, and then
integration by parts for the x-dependence. Written as an analogy to Equation (9), assuming a
stress-free boundary:
 
d2 d
L = {ua }T [K2 ] 2 + [K1 ] + [K0 ] 2 [M] {u} dx = 0 (15)
dx dx
The system of equations, derived from (15), possesses constant matrix coefficients in the form
of square (2m 2m) matrices, [K2 ], [K1 ], [K0 ], and [M]. These are given by

[K2 ] = {(N) T [C](N)}
dH (16)
H

[K1 ] = T [C]( z N) ( z N)T [C](N)}
{(N) dH (17)
H

[K0 ] = {( z N)T [C]( z N)} dH (18)
H

[M] = 2 {NT qN} dH (19)
H

and are conveniently partitioned into m m matrices



2
0
Kuu Kuu
[K2 ] = , [K0 ] =
K2ww K0ww

(20)
Muu
[M] =
Mww

1
1
Kuu Kwu
=
{(N) [C](z N)} dH,
T
= {( z N)T [C](N)}
dH (21)
1
Kwu H K1ww H

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 709

Hence, the solutions of the linear homogeneous system, {u}l = (U (x), W (x))lT , may be written as

{U (x)}l = {1..m }l ei l x , l = 1, . . . , 2m
(22)
{W (x)}l = {m+1..2m }l ei l x , l = 1, . . . , 2m

where {}l is a vector representing the nodal amplitudes (1,l , m,l ), (1+m,l , 2m,l ), and interior
amplitudes ( j,m ), 2 jm 1, ( j+m,m ), 2 j2m 1. Under this assumption, a non-linear
eigenvalue problem arises, K( )U = 0, which may be cast, simply, as a polynomial eigenvalue
problem,

[K( )]U = { 2 [K2 ] i [K1 ] [K0 ] + 2 [M]}U = 0 (23)

of order m for the variable .


The eigenvalue problem relates the wavenumber to the angular frequency , one of them
being given and the other being the eigenvalue to be sought. If R is given, Equation (23) is a
generalized eigenvalue problem with 2m real eigenvalues 2m . If instead the angular frequency is
given, Equation (23) is a quadratic eigenvalue problem (QEP) with 4m eigenvalues m , see Tisseur
and Meerbergen [24] for discussion and review of numerical evaluation of QEP. In the latter
case, the eigenvalues are generally complex valued and come in conjugate pairs and correspond
to propagating and evanescent waves in the positive and negative longitudinal directions. For real
values of the wavenumber eigenvalues, , corresponding to propagating waves, it is possible to
describe dispersion relations with zero damping. The results are discussed in Section 4.1.
The eigenvalue problem may be reduced to first-order form by introducing an auxiliary parameter,
= i and variable {v} = {u}. With this change of variables, the polynomial eigenvalue problem
(23) may be recast as a generalized (4m 4m) linear eigenvalue problem known as a second
companion form in Tisseur and Meerbergen [24]:

[A]{V} = [B]{V} (24)

where




K0 + 2 M K1 K2 {u}
[A] = , [B] = , {V} =
I2m I2m {u}

The resolution of the matrix eigenvalue problem, Equation (24) is itself easily achieved by a
number of standard computational routines where damping is included in the model, i.e. the value
of the loss factor
is positive. This produces a complete set of eigenvalues l , l = 1, . . . , 4m and
corresponding cross-sectional mode shapes or waveforms. Note that for zero damping
= 0, the
polynomial eigenvalue problem is of Hamiltonian type and care must be taken in evaluating the
eigenvalues as discussed in [24]. It has been found that some eigenvalues, , violate the complex
conjugate pairing rule especially near the imaginary axis. This lead to instability issues for zero
and very small values of damping
0.1%. This may be overcome using other methods such as
Van Loans square-reduced algorithm [25], but is yet to be resolved.
This section considered the formulation with basis functions N (z). The next section describes
the construction of basis functions in the x-direction and the assemblage of local dynamic stiffness
matrices.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
710 A. T. PEPLOW AND S. FINNVEDEN

3.2. Wave influence functions for super-spectral elements


Final construction of the set of basis functions, Equation (10) and local dynamic stiffness matrix,
defined in local coordinates over a region  j := H T , where T := {x : Dx+D}, will now
be described. By consideration of expressions (22), it is clear that each wave influence function
may be written as a linear combination of the solutions, written in an alternative form

2m
u ejk (x) =  jl Ell (x)Alk u k , j = 1, . . . , 2m, k = 1, . . . , 4m (25)
l=1

where entries  jl and Ell take the values of eigenvectors and exponential functions, respectively,
and superscript .e represents local element definition. Coefficients Alk are to be determined by
scaling the set of wave influence functions to unity, to be described shortly. Also the vector !,
formed by the wavenumber eigenvalues, (23) is given by
! = {i 1 , . . . , i 4m }T (26)
and introducing the vector ! p defined by

p l , Re(l )0
l = (27)
l , Re(l )<0
l = 1, . . . , 4m, basis functions (10) may be scaled in the longitudinal direction. These now take
the form:
p
Ell (x) = el xl D , |x|D, l = 1, . . . , 4m (28)
Projecting the basis functions, (25), onto the end points of Te :
u ejk (D) =  jk u k , j = 1, . . . , m, k = 1, . . . , 2m and
(29)
u e( jm)k (+D) =  jk u k , j = m + 1, . . . , 2m, k = 2m + 1, . . . , 4m
gives a system of 4m 4m equations determining the basis functions coefficients in (25), Alk ,

[X ][A] = [I ]

where


E(D)
[X ] =
E(+D)

and [I ] is the identity matrix and  represents a diagonal matrix created from the original
vector. Substitution of Equation (10) into Equation (9), over the local region  j , and using
the same matrices as in Equations (16)(19), yields a system of 4m 4m linear equations. It
is necessary to partition the matrix (2m 4m) of eigenvectors into equal (m 4m) lateral and
vertical displacement partitions such that


u
=
w

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 711

3.3. Local dynamic stiffness matrix


Now the approximating functions have been defined the local dynamic stiffness matrix for a certain
element is constructed by considering Equation (12) and may be written as
L  j = uT [L]loc u (30)
where
[L]loc = [A]T ([L]uu + [L]uw + [L]wu + [L]ww ). [E I ][A] (31)
where . denotes the Schur product and

[L]uu = !Tu [K2uu ]u ! + Tu ([K0uu ] 2 [Muu ])u

[L]uw = !Tu [K1uu ]w + Tu [K1uw ]w !

[L]wu = !Tw [K1wu ]u + Tw [K1ww ]u !

[L]ww = !Tw [K2ww ]w ! + Tw ([K0ww ] 2 [Mww ])w

Within the computations of local dynamic stiffness matrices use is made of the matrix generating
function
D
[E I (!, D)] = exp(!x ! p D) exp(!x ! p D)T dx (32)
D

which has an analytic form detailed in Finnveden [14]. The convention that the exp function
preserves the dimension of column vectors has been adopted. Generating the E I matrix was
performed only once thus decreasing overall dynamic stiffness matrix computation substantially.
For problems with short waveguides, (small D) or small valued entries in {!} a Taylor expansion
series for each integrated expression was performed.
Evaluation of the dynamic stiffness matrix above applies to a single super spectral finite element.
The matrix is full and complex. For a combination of finite elements, describing a general elastic
domain, the corresponding dynamic stiffness matrix has a block diagonal structure; each block
derived from the expression in Equation (31). The description above and in the previous section
applies to a layered waveguide where a cross-section is discretized by locating appropriate interface
nodes. Wave functions are subsequently found for the composed layers, by imposing displacement
continuity at the interface, thus increasing the size of the matrix eigenvalue problem. Construction of
stiffness matrices follows as with construction of stiffness matrices for a single layer waveguide.
Hence, it is possible to solve multi-layered waveguide problems using super-spectral elements
bearing in mind the large generalized matrix eigenvalue problems to be solved.
Point forces may be easily included into the model if defined at nodal points. The complete ele-
ment formulation, for an arbitrary element length, element, including assembling element matrices,
solving dispersion relations, and evaluating the dynamic stiffness matrix required only 0.351 s on
a 1.3 GHz PC for a single layer eleven degree-of-freedom element.
For blocked boundary conditions, i.e. an elastic layer on a rigid foundation, displacements are
forced to zero value at the lower boundary. Associated linear eigenvalue problem reduces in size
to 4(m 1) 4(m 1) and the subsequent dynamic stiffness matrix for a simple single element
problem reduces in size similarly.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
712 A. T. PEPLOW AND S. FINNVEDEN

4. RESULTS

Results are presented for material properties and principal wavespeeds as shown in Tables IIII;
c1 , c2 and c R are the compression, shear and Rayleigh wavespeeds, respectively. The properties
in Table III were measured at a British Rail test site named Clarborough, which has a layer depth
of 7 m, and the material properties given in Tables I and II were used in the analysis by Petyt
et al. [2].

Table I. Physical parameters for single-layer model. Clarborough test site.


Upper layer Lower layer
Material propertyClarborough site (H1 = 7 m) (H2 = 0 m)

Density,  1550 kg/m3


Youngs modulus 2.69 108 N/m2
Shear modulus 1.07 108 N/m2
Poisson ratio 0.257
c1 compression wavespeed 459.4 m/s
c2 shear wavespeed 262.7 m/s
c R Rayleigh wavespeed 241.9 m/s
Loss factor 0.0

Table II. Physical parameters for two-layer model. site A.


Upper layer Lower layer Block
Material propertysite A (H1 = 2 m) (H2 = 2 m) (W = 5 m), (D = 1 m)

Density,  1517 kg/m3 1759 kg/m3 1600 kg/m3


Youngs modulus 1.57 108 N/m2 1.063 109 N/m2 1.0 1010 N/m2
Shear modulus 6.65 107 N/m2 4.25 108 N/m2 4.17 109 N/m2
Poisson ratio 0.18 0.25 0.2
c1 compression wavespeed 335 m/s 852 m/s 2635 m/s
c2 shear wavespeed 209 m/s 492 m/s 511 m/s
c R Rayleigh wavespeed 190 m/s 452 m/s 466 m/s
Loss factor 0.1 0.1 0.01

Table III. Physical parameters for two-layer model. site B.


Upper layer Lower layer Block
Material propertysite B (H1 = 2 m) (H2 = 2 m) (W = 5 m), (D = 1 m)

Density,  1500 kg/m3 2000 kg/m3 1600 kg/m3


Youngs modulus 2.6 107 N/m2 3.57 108 N/m2 1.0 1010 N/m2
Shear modulus 8.84 106 N/m2 1.2 108 N/m2 4.17 109 N/m2
Poisson ratio 0.47 0.49 0.2
c1 compression wavespeed 323 m/s 1748 m/s 2635 m/s
c2 shear wavespeed 77 m/s 245 m/s 511 m/s
c R Rayleigh wavespeed 73 m/s 233 m/s 466 m/s
Loss factor 0.1 0.1 0.01

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 713

Wavenumber, (m1)
(2)

4 (10)
(7) (9)
(8)
(3)
3
(6) (5)
(4)
2

1 (1)

0
0 50 100 150 200 250 300 350
Frequency (Hz)

Figure 3. Variation of wavenumber with frequency of the first 10 propagating modes for the Clarborough
test site, Table I. Annotations refer to mode number.

5
Wavenumber, (m1)

0
0 50 100 150 200 250 300 350
Frequency (Hz)

Figure 4. Variation of wavenumber with frequency of the first 10 propagating modes for the two material
parameter sets: Tables II and III. Dash-dotted lines: site A. Solid lines: site B.

4.1. Free vibration


Figures 3 and 4 all feature results for free vibration. The results were obtained by solving
Equation (24) by using a generalized eigenvalue solver readily available in software such as
MATLAB 6.1 used in this article. Note that since the propagating modes are of interest, the results
obtained here are for zero damping,
= 0. For this free vibration analysis, a particular set of values

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
714 A. T. PEPLOW AND S. FINNVEDEN

of the material constants E,  and  were used. These were measured at a particular site and are
given in Table I. Figure 3 shows a set of dispersion curves, the variation of wavenumber with
frequency of the first 10 propagating modes. The set of curves is in excellent agreement with Jones
and Petyt [12] Figure 2.
It has been shown in Waas [4] that for wavenumber = 0 the natural frequencies of the prop-
agating modes produced by the interaction of shear and compression waves are given by the
one-dimensional free-fixed rod natural frequencies, which in terms of the body wavespeeds and
height h = H1 = 7 m are f n = {(2n 1)/28}(ci ), i = 1, 2. These frequencies are easy to calculate
and are close to the roots of the dispersion curves, with = 0. Kobori et al. [7] have noticed
that crossing of two dispersion curves for single elastic layers corresponds to roots of the dis-
persion curves, f n above. Another feature observed by Kobori et al. is the behaviour at the
fourth root in Figure 3 being many valued for a small range of frequency, between 45 and 49 Hz.
A careful description and illustration of this exceptional behaviour is given in Jones and Petyt [12],
Section 6.1.
The features above are not just peculiar for a single layer of elastic material over bedrock.
However, it is not possible to derive such a simple formula for natural frequencies where two
or more layers make up the elastic layer. Figure 4 shows the dispersion curves for data given in
Tables II and III. It can be seen that the two phenomena described above do indeed occur for the
two test sites. For site A, the seventh and sixth curves cross at the root corresponding to the eighth
mode. It is important to note that these resonance frequencies also have an effect on direct or
transfer receptances due to a concentrated load, producing peaks in the response. One of the aims
of this work is to investigate the inclusion of a block in the elastic layers in order to shift these
resonant frequencies under certain circumstances. Also, Figure 4 shows the occurrence of multiple
eigenvalues for material parameters given for site B. The value of this frequency is around 25 Hz.
Notice that the slope of one of the dispersion curves is negative in this local region. Also, it is
important to observe that waves begin to propagate at 20 Hz for site A and at around 6 Hz for
site B.

4.2. Computational validation of model


The spectral finite element method was tested on the two-layered waveguide rectangular region
of total depth 4.0 m for the two sites with parameter values given in Tables I and II. The layers
are bounded below by bedrock and are stress-free at the surface. A point source was located at
the origin. Ninth-order polynomials were chosen as a basis for each of the five sub-element layers
across all four element sectors. Within the bounds of accuracy for the model this equated to 10
degrees of freedom for just under one and a half shear wavelengths. For the five-layer system, this
equated to m 1 = 45 degrees of freedom across each layer totalling 450 degrees of freedom for
the complete model.
Numerical experiments were performed for the five element layer models, see Figure 2, three
for top layer and two for the lower layer. The numerical solutions were compared with a semi-
analytical solution presented in Jones and Petyt [12] computed using accurate numerical quadrature
techniques.
As can be observed in Figures 5 and 6, the results are encouraging. Aside from generally
acceptable levels of accuracy, the spectral finite element method computed the wave influence
functions by solving 90 90 matrix polynomial eigenvalue problem (23) solution in a clock time
of 2.083 s per frequency.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 715

70 70

65 65

Attenuation, |V| ( dB )
Attenuation, |W| ( dB )

60 60

55 55

50 50

45 45

40 40
40 60 80 100 120 140 160 180 200 40 60 80 100 120 140 160 180 200
(a) Frequency ( Hz ) (b) Frequency ( Hz )

Figure 5. Attenuation of (a) vertical and (b) lateral motion to 200 Hz at three receiver positions for site A
versus exact solution, computed using methods in [12]. Lines (exact), and marks (SFEM). Locations
x0 = 5 m (solid and ), x1 = 30 m (dash-dot and ), and x2 = 50 m (dashed and ).

65
55
60
50
Attenuation, |W| ( dB )

Attenuation, |V| ( dB )

55
45
50
40

35 45

30 40

25 35
10 20 30 40 50 60 70 80 90 100 110 120 10 20 30 40 50 60 70 80 90 100 110 120
(a) Frequency ( Hz ) (b) Frequency ( Hz )

Figure 6. Attenuation of (a) vertical and (b) lateral motion to 200 Hz at three receiver positions for site B
versus exact solution, computed using methods in [12]. Lines (exact), and marks (SFEM). Locations
x0 = 5 m (solid and ), x1 = 30 m (dash-dot and ), and x2 = 50 m (dashed and ).

Total clock time for generating dynamic stiffness matrices and solving set of linear simultaneous
equations took 10.175 clock seconds per frequency. Notice that the accuracy of results seem to
decline for site B results in Figure 6. This is due to site B having a very soft upper layer and a
high velocity contrast between this layer and the substratum.

4.3. Wave transmission through an impeding block


The introduction of a region of stiffened material near the source modifies the layer depth and
stiffness properties over a finite part of the surface near the source. Its effect is therefore to raise
the frequency at which propagation in the upper layer occurs and hence lowers the response in the

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
716 A. T. PEPLOW AND S. FINNVEDEN

30
25
20
15

Insertion Loss ( dB )
10
5
0
-5
10
15
20
25
20 40 60 80 100 120
Frequency ( Hz )

Figure 7. Insertion loss for vertical motion for block at site A at unit frequency steps (H0 = 1 m and
L 0 = 5 m). Solid represents 5.0 m, dashed line 30.0 m, and dash-dot 50.0 m.

low frequency range. For this reason, the efficiency of the block is dependent on the properties of
the soil layers into which it is introduced and also on the stiffness of the material. The performance
of the block at two different sites for which the substratum is a half-space for which the modelling
parameters have been determined is presented and discussed in Petyt et al. [2]. A small number
of techniques exist to increase the stiffness of soil in situ. These have been developed to increase
the bearing strength of soils for building and pavement foundations. Two mechanisms of losses
geometric spreading and material damping occur in the work of Petyt et al. [2] but the lossy
medium is the only vehicle for damping mechanism here.
The spectra of insertion loss, calculated at unit frequencies illustrate the effect of adding an
artificial 5.0 m wide stiffened layer placed to one side of the load between the lower layer and
the weathered layer for site A, as shown in Figure 7. The insertion loss is presented for three
distances (5, 30 and 50 m) covering a practical range of near and far-field receiver distances. The
onset of modal wave propagation in each case has been shifted upwards in frequency producing
large insertion losses at 25 and 50 Hz as illustrated in Figure 7. However, large insertion losses
come at a price; dips follow peaks as a consequence of the wave propagation modal behaviour.
The nature of the responses in the frequency range above the dip are difficult to characterize due
to the complex interaction of wave fields but at site A insertion loss tends to zero as frequency
increases. Results for site B are not shown as modal behaviour of the system overwhelms the
discrete frequency content.
Figure 8 shows the spectra of insertion loss for the same 5.0 m wide 1.0 m thick block positioned
close to the surface. It is clear that this minor geometrical modification has an overall significant
effect on insertion losses for site A. This analysis was carried out to indicate whether it is necessary
to treat the soil directly under the surface or whether treatment on the surface would be sufficiently
effective. In this case, the inclusion interrupts the propagation of the Rayleigh wave to the
observation points thus causing large peaks in the results. Again it is difficult to make clear

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 717

30
25
20
15

Insertion Loss ( dB )
10
5
0
-5
10
15
20
25
20 40 60 80 100 120
Frequency ( Hz )

Figure 8. Insertion loss for vertical motion for block sitting close to surface at site A at unit frequency
steps (H0 = 0.0 m and L 0 = 5.0 m). Solid represents 5.0 m, dashed line 30.0 m, and dash-dot 50.0 m.

x 108 x 108
4 5

3.5 4.5
Vertical amplitude |W|, (m)
Lateral amplitude, |V|, (m)

4
3
3.5
2.5 3
2 2.5
1.5 2
1.5
1
1
0.5 0.5
0 0
10 0 10 20 30 40 50 10 0 10 20 30 40 50
(a) x (m) (b) x (m)

Figure 9. (a) Lateral and (b) vertical motion along the surface of the constrained layered ground at a certain
excitation frequency. Predictions by the spectral FE method for configuration in Figure 1, L 0 = 5.0 m,
W = 5.0 m, D = 0.0 m; two elastic layers (solid line), two elastic layers and surface block (dashed line,
H0 = 0.0 m), two layers and lower block (dot-dashed line, H0 = 1.0 m); frequency 32 Hz for site B.

conclusions here as the onset of modal interaction between compressive and shear waves becomes
complicated.
Figure 9 demonstrates the complicated nature of the wave field on the soil surface in the vicinity
of the block due to a concentrated vertical force at one frequency, 32 Hz. Figure 9(a) presents
the lateral motion for the three cases, (without block/block situated within soil/block resting

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
718 A. T. PEPLOW AND S. FINNVEDEN

under surface) and Figure 9(b) the corresponding vertical motion. Both figures show exceptionally
oscillatory behaviour at the soil surface trapped between the block and source. This example
illustrates situations where approximations by general BE or finite element software may not be
sufficiently accurate.

5. CONCLUSIONS

A new spectral method in the form of a finite element scheme has been used to treat the problem
of wave transmission in non-uniform linear elastic waveguides. These configurations are not easily
amenable to analytical treatment and the use of a numerical method is appropriate here. The finite
element formulation in this investigation is, however, sufficiently general that it can be extended
to any non-uniform rectangular waveguide with varying soil parameters, and may be coupled with
a standard finite element code.
A unique feature of the super-spectral finite element approach is the use of basis functions
generated from linear eigenvalue calculations. The basis functions, themselves solutions to the
homogeneous reduced wave equations, may be defined over regions of arbitrary length with
stress-free or rigid boundary conditions. This new finite element technique extends the family of
computational methods for predicting elastic wave transmission and may possess wider applications
in general wave propagation analysis for solids and in soilstructure interaction.
The model predicts the frequency response of a layered ground over a bedrock at a distance
from a concentrated point load. Material parameters used have been measured at a number of
sites and have been shown to be suitable in terms of reproducing the correct response against
well-developed semi-analytic methods.
Due to the finite element structure of the technique, it is worth mentioning here that it is possible
to extend the model geometry to include a number of different features; namely a long embankment
where a rail force may sit and filled or empty trenches. For modelling half-spaces, this approach
would require either a non-reflecting boundary condition model or a novel approach using infinite
elements.

REFERENCES
1. Takemiya H, Fujiwara A. Wave-propagation impediment in a stratum and wave impeding block (WIB) measured
for SSI response reduction. Soil Dynamics and Earthquake Engineering 1994; 13(1):4961.
2. Petyt M, Peplow AT, Jones CJC. Surface vibration propagation over a layered half-space with an inclusion.
Applied Acoustics 1999; 56:283296.
3. Takemiya H. Field vibration mitigation by honeycomb WIB for pile foundations of a high-speed train viaduct.
Soil Dynamics and Earthquake Engineering 2004; 24(1):6987.
4. Waas G. Linear Two-dimensional Analysis of Soil Dynamics in Semi-infinite Media. University of California:
Berkeley, 1972.
5. Tolstoy I, Usdin E. Dispersive properties of stratified elastic and liquid media: a ray theory. Geophysics 1953;
18(4):844870.
6. Kuhlemeyer R. Vertical Vibrations of Footings Embedded in a Layered Media. University of California: Berkeley,
1969.
7. Kobori T, Minai R, Suzuki T. The dynamical ground compliance of a rectangular foundation on a viscoelastic
stratum. Bulletin of the Disaster Prevention Research Institution, Kyoto University 1981; 20:289329.
8. Warburton GB. Forced vibrations of a body on an elastic stratum. Journal of Applied Mechanics (ASME) 1957;
24:5558.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag
VIBRATION TRANSMISSION THROUGH LAYERED BEDROCK 719

9. Segol G, Lee PCY, Abel JF. Amplitude reduction of surface waves by trenches. Journal of the Engineering
Mechanics Division (ASCE) 1978; 104:621641.
10. Kausel E, Roesset JM. Stiffness matrices for layered soils. Bulletin of the Seismological Society of America
1981; 71(6):17431761.
11. Kausel E, Peek E. Dynamic loads in the interior of a layered stratum: an explicit solution. Bulletin of the
Seismological Society of America 1982; 72(5):14591481.
12. Jones DV, Petyt M. Ground vibration in the vicinity of a strip load: an elastic layer on a rigid foundation.
Journal of Sound and Vibration 1992; 152(3):501515.
13. Finnveden S. Exact spectral finite element analysis of stationary vibrations in a railway car structure. Acta
Acustica 1994; 2:461482.
14. Finnveden S. Spectral finite element analysis of the vibration of straight fluid-filled pipes with flanges. Journal
of Sound and Vibration 1997; 199:125154.
15. Peplow AT, Finnveden S. A super-spectral finite element method for sound transmission in waveguides. Journal
of the Acoustical Society of America 2004; 116:13891400, 1972; 739745.
16. Taweel H, Dong SB, Kazic M. Wave reflection from the free end of a cylinder with an arbitrary cross-section.
International Journal of Solids and Structures 2000; 37:17011726.
17. Widehammar S, Gradin PA, Lundberg B. Approximate determination of dispersion relations and displacement
fields associated with elastic waves in barsmethod based on matrix formulation of Hamiltons principle. Journal
of Sound and Vibration 2001; 246:853876.
18. Adamou ATI, Craster RV. Spectral methods for modelling guided waves in elastic media. Journal of the Acoustical
Society of America 2004; 116(3):15241535.
19. Lysmer J. Lumped mass method for Rayleigh waves. Bulletin of the Seismic Society of America 1970; 43:1734.
20. Kausel E. Thin-layer method: formulation in the time-domain. International Journal for Numerical Methods in
Engineering 1994; 37:927941.
21. Park J, Kausel E. Numerical dispersion in the thin-layer method. Computers and Structures 2004; 82:607625.
22. Petyt M. Introduction to Finite Element Vibration Analysis. Cambridge University Press: Cambridge, U.K., 1991.
23. Szabo B, Babuska I. Finite Element Analysis. Wiley: New York, 1991.
24. Tisseur F, Meerbergen K. The quadratic eigenvalue problem. SIAM Review 2001; 43:235286.
25. Van Loan CF. A symplectic method for approximating all the eigenvalues of a Hamiltonian matrix. Linear
Algebra 1984; 61:233251.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2008; 32:701719
DOI: 10.1002/nag

You might also like