Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

INTERNATIONALJOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL.

37,927-941 (1994)

THIN-LAYER METHOD: FORMULATION IN


THE TIME DOMAIN

EDUARDO KAUSEL.
Department of Civil Engineering, Massachusetts Institute of Technology, Room 1-271, Cambridge, M A 02139, U.S.A.

SUMMARY
The thin-layer method is a semi-discretenumerical technique that may be used for the dynamic analysis of
laminated solids or fluids. In its classical implementation, the method is normally formulated in the
frequencydomain and requires the solution of a complex-valuedquadratic eigenvalue problem; in this paper
we present an alternative time-domain formulation which can offer advantages in some cases, such as
avoiding the use of complex algebra. The proposed method entails expressing the governing equations in the
frequency-wavenumberdomain, solving a linear real-valued eigenvalue problem in the frequency variable,
carrying out an analytical integration over frequencies, and performing a numerical transform over
wavenumbers. This strategy allows obtaining the Green’s functions for impulsive sources directly in the time
domain, even when the system has little or no damping. We first develop the algorithm in its most general
form, allowing fully anisotropic materials and arbitrary expansion orders; then we consider a restricted class
of anisotropic materials for which the required linear eigenvalue problem involves only real, narrowly
banded symmetricmatrices and finally, we demonstrate the method by means of a simple problem involving
a homogeneous stratum subjected to an antiplane impulsive source.

INTRODUCTION
The thin-layer method is a versatile and efficient tool for the dynamic (or static) analysis of
mechanical systems that have material properties that change at most in one co-ordinate
direction. It can be used, for example, to analyse the propagation of waves in layered soils and in
laminated (or thick) plates when they are subjected to dynamic loads. In essence, the method
consists in a partial discretization of the wave equation, namely one in the direction of layering;
thus, a finite element solution is used for that co-ordinate direction, while closed-form solutions
(or other numerical approaches) are used for the remaining co-ordinate directions. This strategy
leads to powerful numerical methods that are currently being used for multiple purposes, such as
the analysis of soil-structure and fluid-structure interaction problems, wave diffraction and
scattering problems, and non-destructive evaluation of pavements.
The method was first used by Lysmer’ to study the propagation of seismic Rayleigh waves in
layered earth strata. Since then, the method has found application in the formulation of
transmitting (or absorbing) in the analysis of laminated plates and soils,5 in the
dynamic analysis of circular and annular foundations,6 and perhaps most importantly, as a tool
to obtain the dynamic displacements and stresses anywhere in a layered medium when it is acted
upon by dynamic loads having arbitrary spatial distributions, especially point loads.’** The latter
are referred to as the Green’s functions of the medium, which are essential in the application of

* Professor

CCC 0029-5981/94/060927-15$9.00 Received 6 April 1993


0 1994 by John Wiley & Sons, Ltd. Revised 28 October 1993
928 E. KAIJSEL

boundary integral methods to laminates with irregularities such as cavities or inclusion^.^ More
recently, the method has also been applied to study Rayleigh waves in fluids," to assess the
seismic response of earth dams," to evaluate the response of coupled solid-fluid systems" or to
analyse poroelastic media.I3
In all of the studies referred to previously, the displacement functions have been obtained in the
frequency domain; the response in time, when required, was computed by numerical transforma-
tions involving the Fast Fourier Transform (FFT) algorithm. In particular, the Green's func-
tions for layered media were obtained by (1) formulating the equations of motion in the
frequency-wavenumber domain; (2) solving a complex-valued quadratic eigenvalue problem in
the wavenumbers; (3) integrating analytically over wavenumbers; (4) integrating numerically over
frequencies by means of the FFT.
An alternative formulation that can offer advantages in some situations consists in reversing
the order of integration described. This requires solving instead a linear real-valued eigenvalue
problem in the frequency variable, carrying out an analytical integration over frequencies, and
performing a numerical transform over wavenumbers. This strategy permits avoiding the use of
complex algebra and allows obtaining the Green's functions directly in the time domain, even
when the system has little or no damping. The price paid, however, is that the last step, namely the
wavenumber integral, must necessarily be carried out numerically; this inescapable fact implies
that both the loads--the sources-as well as the solution are spatially periodic. Hence, the
response functions computed are valid only during a limited time duration, namely until the
fastest waves from the neighbouring sources arrive at the region of interest and contaminate
the solution. The separation between the source-and thus the time until the contamination
begins-is inversely proportional to the sampling rate in the wavenumber domain, so the larger
the time duration (or distance to the source) for which the response is needed, the greater the
numerical effort that will be required. Nonetheless, this strategy can lead to very effective
computations of transients in layered media, particularly because it allows for the most part to
avoid the use of complex variables and involves only a real, linear eigenvalue problem. In this
paper we present the outlined time-domain formulation of the thin-layer method in detail.

WAVE EQUATlON IN CARTESIAN CO-ORDINATES


Consider a horizontally stratified, locally homogeneous and anisotropic, linearly elastic body of
infinite lateral extent which is acted upon by dynamic loads at some location. At any arbitrary
point in the medium, the dynamic equilibrium equation, the stress-strain relation, and the
strain displacement relation are given by
pii - L'ra= b(x, y , z, r ) (1)
a= DE (2)
& = Lu (3)
In these equations,
p = the mass density, (44
u = [u, u, u , ] ~ (the displacement vector)
b = [h, by h,IT (the body load vector)
a = [a, by a, zYt T~~ T ~ , ] . ~(the stress tensor) (44
THIN-LAYER METHOD 929

E = [E, E, E , yyz yxz yx,IT (the strain tensor) (44


D = { d i j } i, j = 1,2,. . . , 6 (the constitutive matrix, and) (4f)

1 4 ;;$1
a
(a, *

LT =
a (a differential operator) (4g)

as ay ax
A superscript T in these equations denotes a transposed vector or matrix, and the double dot
indicates the second derivative with respect to time.
Substituting equations (2) and (3) into equation (1) we are lead to the wave equation in
Cartesian co-ordinates:
pii - LTDLu = b (5)
On the other hand, the differential operator L can be written as
d d 2
L = L,-
dx
+ L,
2y
+
- L,:
c'z

with matrices L,, L, and L, which are trivially composed of zeroes and ones; they are obtained
from equation (4g) by simple inspection. It follows that the expansion of the product LTDL is of
the form

with material matrices D,, defined by


D,, = L,TDL,, Z, p = X, y, z (8)
These matrices depend only on the material parameters, and can be readily evaluated for any
anisotropic medium. Substituting equations (7) and (8) into equation ( 5 ) we obtain the general
wave equation for an anisotropic solid in Cartesian co-ordinates.
On the other hand, the vector of internal stresses in any horizontal plane can be written as
s = L,Tu = LTDLu (9)
whose components are
s = CTXZ Tyz gzlT (10)
Finally, the boundary conditions at regions where external tractions t are prescribed are of the
form
t-s=O (1 1)
with s, being the vector of internal stresses at the boundary with normal v. For an upper
horizontal boundary s, = s, while for a lower horizontal boundary s, = - s, with s given in either
case by equations (9) and (10).
930 E. KAUSEL

THIN-LAY ER METHOD
To solve the wave equation, we begin by dividing the physical domain into layers that are thin in
the finite element sense, each of which may in turn be composed of sublayers; the numbers of
sublayers, if any, will depend on the interpolation order m chosen for the thin-layer formulation,
as will be seen. We proceed next to cut out an individual thin layer, we label the sublayers with
indices, I, I + 1,. . . ,I + m (from the top down), we preserve equilibrium by applying tractions
tl, tl +n on the exposed surfaces such that they match the internal stresses there (Figure l), and we
approximate the displacement field within the layer by means of an interpolation
u = NU (12)
with N = N(z) being a matrix containing the interpolation polynomials, and U = U(x, y, t )
a column vector composed of the interface displacement vectors
u = [Uf u:+, . . . U Tl + m ] T (13)
For example, for linear (m = 1) and quadratic (m = 2) interpolations, the matrix N is
N = [[I (1 - 011 (linear element) (144
N = [[(2[ - 1)1 4[(1 - [)I ( 1 - [)(l - 2[)I] (quadratic element) (14b)
where I is the 3 x 3 identity matrix, and C = z/h, with h being the thickness of the layer.
Clearly, equation (1 2) represents a partial discretization of the displacement field, namely one
in the direction of layering. Thus, when we substitute the displacement expansion (12) into both
the wave equation (5). and into the boundary conditions (1 l), we find that these equations are not
satisfied identically, but exhibit instead residual body forces r = c(x, y, z, t ) and residual boundary
tractions q(x, y, z, t ) , i.e.
b - pii + LTDLu= r (1 5 4
t-%=q (15b)
To derive the discrete equations of motion and dispose of these unbalanced forces, we apply the
method of weighted residuals to the layer, and require the virtual work done by the residul forces
throughout an elementary volume of the layer (of width dx, depth dy, and height h) to be zero:

Gufq, + 6u:+,ql+, +

Figure 1. Discrete thin layer as free body


THIN-LAYER METHOD 93 1

The first two terms represent the virtual work performed by the residual tractions at the upper
and lower boundaries of the layer, while the third term corresponds to the virtual work done by
the residual body forces. The reader will notice that there is no term associated with internal
stresses along the lateral boundaries of the volume, because their virtual work cancels identically
with the work of the equal and opposite stresses acting on neighbouring elementary volumes.
Substituting equations (12) and (15) into equation (16) and discarding the product dxdy, we
obtain
h,t, + Su,+,t,+, + 6UT
J-: NTbdz = 6UT {1:NTNpdz} U + 6u:sl
{
- duf+,s,+, - 6UT lONTLTDLNdz}U
h

When we require this expression to be valid for arbitrary variations 6U,we obtain the dynamic
equilibrium equation for the thin layer as

(18)
The left-hand side contains the consistent external tractions p acting on the sublayer interfaces
which result from the interface tractions t and body loads b. The right-hand side, on the other
hand, contains the inertial loads as well as the elastic loads of deformation. Evaluation of this
expression for a given interpolation order and material constitution requires tedious, but
straightforward algebra, which includes integrating by parts some elements of the last term. After
carrying out this process we obtain an equation of the form

P = M U - A,,
azU
;2 - AXy-azu - AyY-
azu- B,-au - By-au + G U (19)
ox ax ay aY ax dY
where
M= { 1: NTNpdr}

G = {I:N”D,,N’dz I
In equations (22) and (23) N’ denotes the first derivative of the interpolation matrix with respect
to z. Notice that M, A,, and G matrices are symmetric, while the B, matrices are skew-symmetric.
These matrices are listed in the appendix.
932 E. KAUSEL

At this stage, we proceed to overlap the results for a single layer with those of all other layers,
like the element stiffness matrices in a finite element formulation, so as to generate the system
matrix. Since the overlapping process is straightforward and the resulting system of equations has
the same form as equation (Is),we need not introduce additional symbols for this purpose.
Instead, equation (19) can be understood as representing the complete system of layers, relating
the consistent interface tractions P = P(x, y, t ) applied at the layer interfaces with the displace-
ments U = U(x, y, t ) observed there. Such a system is narrowly banded and has a total of 3N
degrees of freedom, with N being the number of active interfaces (which depends on the number of
layers, the expansion order and the boundary conditions at the top and bottom surfaces).

SOLUTION OF THE DISCRETE EQUATIONS OF MOTION


As can be seen, partial discretization of the wave equation in the direction of layering eliminates
the functional dependence on the vertical co-ordinate z, and yields a system of partial differential
equations in the horizontal co-ordinates and time. To solve this equation, we begin by performing
a spatial Fourier transform in the two horizontal co-ordinates. The result is an equation of the
form
P=ME+KO (24)
with
K = kf A,, + k,k,A,, + k;Ayy + G + i(k,B, + k,By) (254

? = P(k,, k,. t) (Fourier transform of P) (25b)


0 = O(k,, k,, t ) (Fourier transform of U) (25c)

where k,, k, are the horizontal wavenumbers (spatial frequencies), and i = /-7 . (24)
Equation
is analogous to the dynamic equation of a conventional discrete system with mass matrix M and
stiffness matrix K, except that the latter is not real, but is instead Hermitian (ie. symmetric real
part, skew-symmetric imaginary part). This analogy suggests that equation (24) can be solved by
a modal superposition involving the eigenvalue problem

having a Rayleigh quotient

in which $* = conjgJIT. Since K is Hcrmitian, while M is real, symmetric and positive-definite, it


follows that both quadratic forms in Rayleigh’s quotient are real. Hence, all eigenvalues E are real;
however, the eigenvectors may still be complex. On the other hand, the numerator and denomin-
ator in equation (27) can be shown to be proportional to the strain and kinetic energies in the
system, respectively; since these energies cannot be negative, we conclude that all eigenvalues
must be non-negative. If, in addition, the system of layers does not admit rigid-body translations
(e.g. a soil stratum supported by rigid rock), then zero eigenvalues cannot occur; in such a case,
only positive eigenvalues exist.
While it is possible to solve the eigenvalue problem posed by equation (26) without much
difficulty, we choose instead to restrict from now on our attention to a more special class of
problems, namely those systems that have a (symmetric, positive-definite) constitutive matrix of
THIN-LAYER METIIOD 933

the form
dll d12 d13 '

dZl d22 d23 .

dbl d62 .
d63 d54
' d5
' 5

This constitutive matrix corresponds to a medium that is somewhat more general than an
.
d66 I
orthotropic material. When this condition is satisfied, it is possible to reduce equation (24) to
a fully real and symmetric form. This simply requires modifying the vertical components of both
the load and displacement vectors by an i = factor, which changes the problem into
P =MU+ KtJ (294
K = k:A,, + k,k,A,, + kiAyy + k,S, + kyBy+ G (29b)
p = [ji pj ip: F: . . ipr]T (294
fj = [UX' ti; 1%
'-1 u,' . . . iti;]' (294
These changes are accomplished in equation (24) by multiplying every third row by i and every
third column by - i. Inspection of the structure of all matrices involved (see the appendix) reveals
that this transformation affects only the iB, terms, which become both real and symmetric. The
modified 8, matrices are obtained from the B, by simply changing the sign of every third row. It
should be noted that since the modification described constitutes a similarity transformation, the
eigenvalues of equation (26) are not changed by it, but remain real and non-negative. Further-
more, since B is now real and symmetric, then all eigenvectors are real, i.e. the eigenvalue problem
is of the form
R + j = ~ f M @ j j = 1,2, . . . , 3N (30)
These eigenvectors satisfy the standard orthogonality conditions, and can be assumed to be
normalized with respect to the mass matrix, i.e.
OTMQ,= I, mTR@= R2
with
Q, = {4Bj}, R = diag { o j } j = 1,2,. . . , 3 N
Finally, if is assumed to be of the separable form
B = P,(k,, k,)f(t) (33)
then conventional modal superposition yields the solution to equations (28) in the time-wave-
~~

number domain as
3N
0= C1 y j h j * f + j
j=
with
-jj= (modal participation factor)
1 .
hj = - sin (oil) (modal impulse response function) (35W
"j
934 E. KAUSEL

and the symbol * denoting a convolution. Note that the modal mass implied by equations (35) is
unity because of the normalization of the eigenvectors.
At this stage, it is possible to generalize the solution so as to include an arbitrary ("propor-
tional") viscous damping law; this is accomplished by simply changing the modal impulse
response functions into

with
wdj = wj Jm (37)
with ti being the fraction of viscous modal damping. On the other hand, it is well known that the
impulse response functions for viscously and hysteretically damped one-degree-of-freedom sys-
tems are nearly identical; hence, we can also simulate hysteretic damping in the thin-layer method
by simply taking identical fractions of damping in each mode.
Finally, the solution in the spatial domain is obtained by applying (numerically) an inverse
Fourier transformation to equation (34), after accounting for the inverse similarity transforma-
tion, which requires multiplying every third component of 0 by - i.

IMPULSE RESPONSE OF LAYERED SYSTEM


Of particular interest in engineering applications is the response of the layered system to an
impulsive load (traction) placed at some elevation. Such a force is of the form
fl: = g ( k x ,k Y ) 6 ( t ) Q = x, y (horizontal loads) (384
fi: = ig(k,, k,)6(t) (vertical load) (3W
with n designating the layer interface number (0 < n < N ) , a the direction of the load and g the
spatial Fourier transform of the load.
From equations (34), (38) and considering the inverse similarity transformation, it follows that
the response of the mth layer interface in direction a( = x, y, z) for an impulsive load in direction
fl( = x, y, z) placed at the nth layer interface is (superindices identify the location and subindices
the direction, respectively, of input-output):

(a) Horizontal loads:


3N
k,, t ) = g(k,, k,)
iiz(kx, ~
cp,m'q$jhj, a,/?= x, y (horizontal response) (39a)
j= I
3N
- ig(k,, k,) 1
j= 1
@q$jhj, fl = x, y (vertical response) (394

3N
ig(k,, k,) 2 q@cp:'h,,
j= I
r = x, y (horizontal response) (39c)
3N
g(k,, k,) 2 q?&hj,
j= 1
(vertical response) (394
THIN-LAYER METHOD 935

In these equations, qrj denote the components of the jth eigenvector in direction a in the mth
layer, i.e. the eigenvector is written as
$j = cq;j q;j qfj q;j . . . qri]T (40)
Notice that except for the i factor in equation (39b) and (39c), all other terms are real. Thus, the
impulse response functions are either purely real or purely imaginary functions of the wavenum-
bers. As before, these impulse response functions are Fourier-transformed numerically into the
spatial domain, and since the functions are known explicitly with respect to time, it follows that
sequential time steps are not required.

PLANE STRAIN RESPONSE


The algorithm described is particularly effective for plane strain problems, since in such case the
spatial Fourier transform required in the last step of the computations is in one dimension only,
and the number of degrees of freedom is reduced by one-third. The governing equations are like
those in the preceding pages, except that all rows and columns associated with one of the
horizontal degrees of freedom (say y ) are deleted. In addition, the expressions become simpler, for
example, equation (29a) changes into
U = k,ZA,, + k,S, + G (41)
For a material in plane strain obeying equation (28), the coupling between the horizontal and
vertical directions is through the F), matrix only. Also, the structure of this matrix is such that
a reversal in the sign of the horizontal wavenumber (k,) does not affect the eigenvalues in (30),
but merely changes the sign of the vertical components in the eigenvectors, i.e.
cp,"'( - k,) = cp,"'(k,),cpF'( - k,) = - cp,"'(k,). It follows that equations (39a) and (39d) are real
even functions and equations (39b) and (39c) are purely imaginary odd functions, of the
horizontal wavenumber. Hence, the spatial Fourier transforms of these equations is real, as
expected.

EXAMPLE ANTIPLANE SH SOURCE IN HOMOGENEOUS ISOTROPIC STRATUM


To illustrate the methodology presented, we apply it to solve a very simple problem, namely that
of an antiplane impulsive source in a homogeneous and isotropic stratum; this problem involves
displacements in the y direction only.
The eigenvalue problem implied by equation (30), when using a linear expansion, and after
removing all degrees of freedom save for those in the y direction, has a simple tridiagonal
structure that is amensable to analytical treatment. Indecd, it can be shown that the solution to
this eigenvalue problem is (with k = k,)
n
sin2 -(2j - 1)
2Hk 4N
)]I" J = 1, 2 , . . . , N (42a)
2H - 32s i .n 2 GR( 2 j - 1)

1)(2j- 1)
1 m , j = 1,2 . . . . , N (42b)

sj = [$ 2
3
n
(1 - - sin2 -(2j - 1)
4N
(normalization factor)
936 E. KAUSEI.

where N is the number of layers, H is the depth of the stratum and C, is the shear wave velocity.
For this case, the impulse response in the wavenumber domain is

The response in the spatial domain is obtained by an inverse Fourier transformation of this
equation, i.e.
mn ( x , t ) = -
uyy
’ I‘
2n - s
u r ( k , t)e-ikXdk (44)

Since the modal components do not depend (in this case!) on the horizontal wavenumber, the
inverse transform involves only the product of g and hi, which can be evaluated efficiently.
In particular, let us consider an antiplane impulsive load of unit intensity that has a bell-shaped
distribution in space in the form of a Hanning window, i.e.
I xx
g ( x ) = - cos2 -, - a<x <a (44)
a 2a
This function has a Fourier transform
sin ka
(45)
g ( k ) = ka [ I - ( k ~ / n ) ~ ]
The amplitude of this transform is rather small above the limit
k,,, = 2 x 1 ~ (46)
so it suffices to carry out computations only up to this point. On the other hand, if x,, and
t,,, are, respectively, the farthest point from the source and the longest time for which the
response calculation is needed (see Figure 2), it follows that the spatial period of the load should
be no smaller than
L 2 C,t,,, + x,, + a (47)

Figure 2. Example: stratum subjected to spatially periodic load


THIN-LAYER METHOD 937

Hence, the sampling rate in the wavenumber domain should be at most

A k = -2n
< 2n
L 'C s t m a x + x m a x + a
and the number of points necessary in the wavenumber domain can be no less than

As a specific example, consider a stratum subjected to an antiplane impulsive Hanning source


of unit amplitude applied at the free surface, with the following characteristics (see Figure 2):
H = 1, C, = 1, a = 0.25, xmaX= 8, t,,, = 4 (equal to the fundamental period of the stratum). We
conclude that the least number of points that can be used in the wavenumber domain is
M = 1+(1~4+8)/0.25=49.
Should one use the FFT algorithm to transform the results into space? The answer to this
question depends on the number and spacing of the receivers (x points at which the response is
needed). If time histories are needed only at a few receivers, then a direct transformation without
the FFT may be more effective. If so, one would begin by evaluating the trigonometric terms in
the transform and placing them into a lookup table for later use. Such course of action has several
advantages: (1) the number of points in the wavenumber domain need not be a power of two; (2)
the real-valuedness and evenness of the response functions with respect to wavenumber indicates
that only a real cosine transform is necessary (i.e. only real algebra is needed); (3) the receivers
need not be equally spaced; (4) receiver spacing is not tied to the maximum wavenumber or the
spatial period of the load. On the other hand, if output is required at many receivers, then the
inversion should be carried out with a conventional FFT algorithm.
The algorithm described (i.e. equations (42)-(44) was applied to the example defined pre-
viously, using a spatial period of 16 units, and carrying out the inversion with a conventional FFT
having 512 points (so as to achieve a good spatial resolution). Figure 3 shows the time history at

-.a
0 2 4 6 a 10

Time (1)
Figure 3. Response at x = 3.75
938 E. KAUSEL

-.8
0 4 8 12 16

Distance (x)

Figure 4. Response at t = 5 (‘photograph’)

a receiver with co-ordinate x = 3.75.Notice that the response is zero until t = 3.50,which is the
time required for waves to propagate from the right edge of the load (x = 0.25) to the receiver.
Figure 4, on the other hand, shows a ‘photograph’ at t = 5, i.e. the displacements as a function of
x at the indicated time. As can be seen, there are two wave trains travelling towards each other in
opposite directions, and the ground is quiescent beyond the wave fronts; both trains will ‘collide‘
after t = 7.75, which is the time needed for waves emanating from the edges of the two
neighbouring loads to travel to the midpoint between them. An evaluation of this problem with
our program PUNCH (which operates in the frequency domain) yielded nearly identical results,
and thus confirmed the accuracy of the results depicted in Figures 3 and 4.

CONCLUSION
The time-domain formulation of the thin-layer method presented in the previous sections can be
an effective numerical tool for the computation of wave propagation problems in laminated
isotropic or anisotropic media. It is particularly attractive when the response is required at only
a few receivers; when the dynamic loads do not vary sharply in space (i.e. not many wavenumbers
are needed to describe them); when the loads are applied impulsively (no convolution is needed);
and when the system has little or no damping. By contrast, the classical frequency-domain
formulation is more attractive when many spatial points are involved, when the loads are
spatially sharp, and when they do not change rapidly in time (e.g. harmonic loads). While
a solution in the frequency domain generally requires some damping, it is possible to consider
also undamped media by recourse to the exponential window method.14
A significant advantage of the time-domain formulation over the frequency-domain formula-
tion is that the former involves in most cases only real-symmetric, narrowly banded eigenvalue
problems for which effective off-the-shelves routines can be used; by contrast, the frequency-
domain formulation leads to quadratic eigenvalue problems involving complex eigenvalues,
special solvers, and more computational effort.
Extensions of the formulation presented herein for problems involving layered media with
cylindrical symmetry can be carried out by starting from the wave equation in cylindrical
co-ordinates, and performing a Hankel transform in the radial co-ordinate in place of the Fourier
THIN-LAYER METHOD 939

transforms in the two horizontal co-ordinates;after carrying out this modification, the reader will
find that the matrices required for cylindrical co-ordinates are closely related to those given in
these pages.

APPENDIX
All layer matrices, except the B,, are of the form
911H 912H ...
= {gijH)
9m+ 1.m+ 1 H

Table 1. Matrices H and coefficients g i j for different layers matrices

h{i+cl
I. +p
}
A,, h

1
-
G h h
1. + 2p
Coefficients g,,
--
Linear expansion Quadratic expansion

l
30 {
2 16
4

-1
2

2
-i)
G
940 E. KAUSEL

Table 11. Matrices D and coefficients gij for the layer matrix B,
Anisotropic Isotropic
.- -- -

Dxz

d55 4 s 4 5

D Y Z

Coefficients g,,
-- ~~ -

Linear expansion Quadratic expansion


3 4 1
4 0 -4}
S{i I;] i{
-! \ - 4 4

~ -1
~ -4 -3

with coefficients gij and matrices H as shown in Table I. The dij are the elements of the
constitutive matrix, D, and I, p are the Lame constants.
On the other hand, the B, matrices (Table 11) are of the form

Examples (linear expansion)

h
A =-
xx 6

d45 - d36
J
TIIIN-LAYER METHOD 94 1

REFERENCES
1. J. Lysmer. 'Lumped mass method for Rayleigh waves', Bull. Seism. SOC. Am., 60,89-104 (1970).
2. J. Lysmcr and G. Waas, 'Shear waves in plane infinite structures', J. Eng. Mech. Dic. A S C E , 18, 85- 105 1972.
3. G. Waas, 'Linear two-dimensional analysis of soil dynamic problems in semi-infinite layer media', P h D . Thesis,
University of California, Berkeley, (1972).
4. E. Kausel, 'Forced vibrations ofcircular foundations on layered media', MITResearch Report R74-I I , Department of
Civil Engineering, MIT, Cambridge, MA 02139, 1974.
5. E. Kauscl and J. M. Roesset, 'Semi-analytic hyperelement for layered strata', J . Eng. Mech. Diu. A S C E . 569-588 (1977).
6. J. Tassoulas, 'Elements for the numerical analysis of wave motion in layered media', M I T Research Report R8I-2,
Department of Civil Engineering, Cambridge, MA 02139, 1981.
7. E. Kausel. 'An explicit solution for the Green's functions for dynamic loads in layered media', M I T Reseurch Report
RBI-13, Department of Civil Engineering, MIT, Cambridge, MA 02139, 1981.
8. E. K a u x l and R. Peck, 'Dynamic loads in the interior of a layered stratum: an explicit solution', Bull. Stism. SOC. Am.,
72, 145Y 1981 (1982a); see also the Errata in BSSA 79,4, p. 1508.
9. E. Kausel and R. Peek,'Boundary integral methods for stratified soils', M I T Research R e p r i R82-50. Department of
Civil Engineering, MIT. Cambridge. MA 02139, 1982b.
10. H. H. Tan, 'Displacement approach for generalized Rayleigh waves in layered solid-fluid media', Bull. Seism. SOC.Am..
79, 1251-1263 (1989).
11. S. Bougacha and J. Tassoulas, 'Seismic analysis of gravity dams. 1: modeling of sediments', J . Eng. Mech. A S C E , 117,
1826 I837 (1991).
12. R. N. Ghibril. 'On the partial discretization ofcoupled plane stratified systems', Ph.D. Thesis, Massachusetts Institute
of Technology, Cambridge, MA, 1992.
13. S. Bougacha, J. M. Roesset and J. Tassoulas. 'Analysis of foundations on a fluid-filled poroelastic stratum', J . Eng.
Mech. ASCE 119(8). 1632 1662 (1993).
14. E. Kausel and J. M. Roesset, 'Frequency domain analysis of undamped systems'. J. Eng. Mech. A S C E 118, 721-734
1991.

You might also like