Vous êtes sur la page 1sur 55

Caughey Absorbing Layer Method -

Validation and Comparison with Lysmer Absorbing Boundaries

Finite Elements in Structural Engineering

Prof. Corneliu Cismaiu

Doctoral Program in Civil Engineering

Andr Filipe da Silva Rodrigues

Monte da Caparica, 2012


2
Table of Contents

Table of Contents .......................................................................................................................................... 3


Abstract ......................................................................................................................................................... 5
Introduction .................................................................................................................................................. 6
The Problem .............................................................................................................................................. 6
Solutions ................................................................................................................................................... 6
Caughey Absorbing Layer Method............................................................................................................ 7
Rayleigh Damping ..................................................................................................................................... 7
One-Dimensional Absorbing Layer ............................................................................................................. 10
One-Dimensional Homogeneous Absorbing Layer ................................................................................. 10
Non-homogeneous Discretized Absorbing Layer ................................................................................... 14
Non-homogeneous Continuous Absorbing Layer ................................................................................... 17
Optimization of the Non-homogeneous Absorbing Layer ...................................................................... 19
Optimum Loss Factor as a Function of the Layers Length ..................................................................... 23
Detuned Absorbing Layer ....................................................................................................................... 26
Two-Dimensional Absorbing Layer ............................................................................................................. 27
The Model ............................................................................................................................................... 27
Lysmer Absorbing Boundary Conditions ................................................................................................. 29
Results ..................................................................................................................................................... 30
Optimization of the Absorbing Layer ...................................................................................................... 34
Computing Time ...................................................................................................................................... 36
Conclusions ................................................................................................................................................. 38
Future Work ................................................................................................................................................ 39
References .................................................................................................................................................. 40

Appendix A Results for the Optimization of the Non-homogeneous Absorbing Layer ........................... 42
Appendix B Results for the One-Dimensional Model for Various Materials ........................................... 49
Appendix C Results for the Two-Dimensional Plane Stress Model .......................................................... 53

3
4
Abstract

The goal of this report is to present the results of a study on the efficiency and applicability of the
Caughey Absorbing Layer Method for finite element analysis of elastic wave propagation in unbounded
media.

The Caughey Absorbing Layer Method (or CALM) was proposed by Jean-Franois Semblant et. al, from
the Paris-Est University. Is consists in applying Rayleigh/Caughey damping to the outer layers of a
numerical model to absorb the elastic waves before they reflect at the boundaries.

This method is applied for one-dimensional and two-dimensional case studies, and compared to the
more traditional Lysmer boundaries. The dependency on material parameters, damping coefficients and
load frequency is also tested.

5
Introduction

The Problem

In the numerical study of elastic wave propagation in solids, particularly in the use of Finite Element
Models, one of the biggest drawbacks is the difficulty to simulate a semi-infinite or unbounded domain.

This is the case in the analysis of soil vibrations: although the area of interest may be relatively small, it
is neither confined nor isolated from the surrounding soil. This means that modeling just the area of
interest will cause spurious reflections of the elastic waves to occur at the mesh boundaries.

In an analytical analysis, it is common to admit the soil as a semi-infinite medium. Since this is not
possible in standard Finite Element formulation, other approaches must be employed.

Solutions

The most common solutions (apart from the Boundary Element Method [1]) are the so called Non-
Reflecting Boundary Conditions (NRBCs), and include:

Absorbing Boundary Conditions (ABCs) [2]: specific conditions at the model boundaries that
approximate the radiation condition for the elastic waves.

They have the drawback of usually working in one direction only (or a set of orthogonal directions), so
they are less effective when dealing with various wave types and different angles of incidence.

Lymser boundaries [3] are an example of such ABCs, being among the earliest to be developed.

Infinite Elements [4]: similar to finite Elements, but with their end nodes moved to infinity. They
approximate the decaying laws governing the waves radiation process at infinity.

They suffer from essentially the same drawbacks as ABCs.

Absorbing Boundary Layers (ABLs) [5]: they consist of a layer of finite thickness located at the medium
boundaries with properties that enable it to attenuate the wave propagation.

Among the most successful are the Perfectly Matched Layers (PMLs) [6, 7], a particular ABL formulation
with a field that describes the attenuation of the waves in a specific direction. They are more efficient
than ABCs, but may lead to instabilities, although improved formulations have been and continue to be
proposed [8, 9].

6
Caughey Absorbing Layer Method

The method proposed by Jean-Franois Semblant et. al [10] was to employ absorbing layers with multi-
direction attenuating properties. Simplicity of the formulation was also a factor on its development.

The chosen approach was to define a finite elastic medium with an absorbing layer at its boundaries.
This absorbing layer is modeled with the same element technology and material properties as the
interior of the medium, but it will include damping properties that will ensure that no spurious wave
reflections occur at the boundaries.

To ensure that the absorbing layer exhibits the desired properties, some considerations about the
damping parameters to employ must be made.

Rayleigh Damping

One of the simplest ways to define damping in finite element analysis is to consider the Rayleigh
formulation [11]. In this damping formulation one assumes that the damping matrix (C) is a linear
combination of the stiffness (K) and mass (M) matrixes:

C M K (1)

The coefficients and are known as the Rayleigh coefficients.

This approach is very convenient because it can be easily computed, since finite element methods
already require the assembly of the mass and stiffness matrices.

It is well known [12] that the loss factor (, the ratio of energy dissipated from the system to the energy
stored in the system for every oscillation) is approximately double the damping ratio (, the ratio of the
damping coefficient in the system's differential equation to the critical damping coefficient), which, for
Rayleigh damping, relates to the frequency of excitation () and to the Rayleigh coefficients in the
following way:

2 (2)

Another quantity for accessing the efficiency of the damping is the magnification factor Q, also known as
the amplification or quality factor. It is defined as the ratio of the response amplitude at resonance to
the static response [13]. The amplification factor relates to the hysteretic loss factor through the
equation

Q 1
2
(3)

However, for weak to moderate Rayleigh damping, there is a simpler relation between these two
quantities:

7
Q1 2 (4)

Therefore, Q-1 can be referred as the loss factor.

According to equation (2), for the Rayleigh damping the loss factor will be minimum when the frequency
of excitation is

d
0 (5)
d

If one calls this frequency r and the corresponding loss factor Q-1min, the curve shown on Figure 1 can
be obtained.

Figure 1 Loss factor curve for the Rayleigh damping (based on Semblat [10])

It is clear that the loss factor tends do infinity when the frequency of excitation approaches zero or
infinity. This suggests that, in theory, if the Rayleigh coefficients is defined to get a desired loss factor
Q-1min for a certain frequency r, all excitations will be damped at least as much as the value defined. As
the frequency of excitation moves away from r, the more the damping will be felt.

From equations (2), (4) and (5) it is easy to obtain the desired values for the Rayleigh coefficients as a
function of the target frequency and the desired minimum loss factor:

8
Q
1
Qmin 1r 2
(6)
r Qmin 2r
1

It is straightforward to apply these conclusions to the absorbing layers in study: to define Rayleigh
damping for those layers, the coefficients to adopt can be calculated using equation (6), by defining the
minimum loss factor as a desired value of absorption and the target frequency as the expected
frequency of excitation for the problem at hand.

This methodology will first be applied to a one-dimensional wave propagation problem, to assess its
efficiency, and then tested in a two-dimensional problem.

9
One-Dimensional Absorbing Layer

One-Dimensional Homogeneous Absorbing Layer

The following example was modeled in Ansys [14] as a transient analysis with implicit integration.
Preliminary studies have suggested that the CALM is not suitable for the explicit integration scheme
provided by Ansys: it required a time-step size five to six orders of magnitude lower than what would be
recommended for the same problem without the desired value of stiffness proportional damping.

The model is analogous to a bar with only axial deformation, but it was modeled as a mesh of
quadrilateral plane stress elements, in which the degrees of freedom perpendicular to the direction of
wave propagation are restrained, as depicted in Figure 2.

Figure 2 One-dimensional model with CALM (based on Semblat [10])

The length of the model is equal to four times the wavelength to absorb, plus the length of the
absorbing layer, which is also a multiple of the wavelength.

The material properties chosen are those of a hard soil, but with the Poisson ratio equal to zero, as
detailed on Table 1.

Table 1 Material properties for the model (E Young modulus, Poisson ratio, density, cp pressure waves
speed)

E cP

[MPa] [kg/m3] [m/s]
Hard soil 200,0 0 2 000 316,228

The reason to choose a Poisson ratio equal to zero was to avoid extensions in the direction transversal
to the waves propagation, which could lead to wave reflection in the upper and lower boundaries of the
model. Those extra waves would be difficult to differentiate from the ones reflecting at the end of the
model, therefore interfering with ones ability to assess the efficiency of the absorbing layer.

10
It should be noted that the pressure waves speed is a derived property, being a function of the other
three properties:


cP E 1 2 2 1 (7)

The wavelength is dependent not only on the material properties, but also on the excitation applied. In
this case, the chosen excitation is an imposed displacement at the left side of the elastic medium, with
the time history equal to a second-order Ricker Wavelet, like the one used by Semblat [10]:

2 t ts 2 2 t t 2
R2 t U 0 2
t 2p
1 e s
, (8)
t 2p

where U0 is the maximum amplitude of the wave, t is the time coordinate, tp is the fundamental period
of the wavelet and ts is the time shift and t is the time coordinate.

Assuming circular frequency (r) of 500 rad/s for the load, the period of the wavelet is

2
tp 0.012566 s (9)
r

The time shift was chosen to be equal to the fundamental period, and the maximum amplitude equal to
one millimeter, resulting in the excitation shown on Figure 3.

Figure 3 Second-order Ricker Wavelet

11
The wavelength can then be calculated according to

P 2 cP r , (10)

which yields 3.974 meters for the hard soil.

As a first test, the hard soil was chosen, and the absorbing layer was defined as having a length equal to
the wavelength. The size of the elements was chosen to be /24, to reduce numerical wave dispersion to
a minimum.

Figure 4 shows the displacement at the middle of the elastic medium (not counting the absorbing layer)
over the first 0.12 seconds of the analysis, with the minimum loss factor ranging from 0 to 2 (which
represents 100% of critical damping).

It can be seen that the excitations reaches the chosen point around the 0.025 seconds mark, which is
consistent with the theoretical wave speed. In the model without any sort of damping, there are no
displacements after the passage of the wavelet, until at the 0.1 seconds mark, when the waves reflected
at the fixed boundary return, exhibiting amplitudes close to that of the original passage.

With damping in the absorbing layer, however, there is some perturbation starting at the 0.075 seconds
mark. This instant matches the time that the waves take to get to the interface between the elastic
medium and the absorbing layer, and then travel to the middle of the elastic medium, as shown on
Table 2.

Table 2 Time required for the Ricker Wavelet to travel a distance multiple of the wavelength

Distance 2 4 6 8 10
Time [s] 0.0126 0.0251 0.0503 0.0754 0.1005 0.1257

One can conclude that the perturbation at t = 0.075 s is due to reflection at the interface between the
elastic medium and the absorbing layer. By observing the detailed view at the bottom of Figure 4, its
easy to conclude that, as the loss factor increases, the amplitude of the waves reflected at the boundary
decreases, but the amplitude of the waves reflected at the interface increases.

A simple solution would be to choose the value for the loss factor that would lead to the amplitude of
both waves being equal, thus minimizing the maximum amplitude of both reflections. There are,
however, other alternatives to this problem, like the one presented in the following section.

12
Figure 4 Displacement at the middle of the elastic medium due to the Ricker Wavelet excitation for various
values of the minimum loss factor of the absorbing layer

13
Non-homogeneous Discretized Absorbing Layer

Increasing the loss factor of the absorbing layer reduces the amplitude of the waves reflected at the
boundary (the wave is almost completely absorbed while it travels inside the layer), but leads to
increased reflection at the interface.

A possible solution is to guarantee that the overall damping of the absorbing layer is enough to absorb
the waves, but the sudden change of damping properties at the interface between the elastic medium
and the absorbing layer doesnt occur. This can be achieved by varying the damping from zero to a
maximum value along the length of the layer, as shown if Figure 5.

Figure 5 Non-homogeneous absorbing layer with loss factor as a linear function of the axial coordinate

Three variation schemes were tested: linear, quadratic and hyperbolic tangent. As a first approach, the
variation was discretized, so each element had a constant damping, but the damping changed from
element to element, up to a maximum value equal to the defined Qmin-1.

To compare the results between the various approaches, three parameters were used. In the following
expressions, the function u(x,t) represents the displacement in the elastic medium only, as a function of
the horizontal coordinate ( x [ 0 , 4 ] ) and the time coordinate ( t [ 0 , 0.2 ] ). The constant tW stands
for the time instant when the wavelet has left the elastic medium.

Two of the parameters are based on the L2 norm which, for an arbitrary function f(x) defined in the
space [ xi , xf ] can be defined as

L2 f x f x dx
xf

xi
(11)

The first parameter is the maximum of u(x,t) after the wavelet has left the elastic medium (t > tW),
expressed as a percentage of the amplitude of the applied excitation:

max u x, t
x ,t tW
umax (12)
U0

14
The second parameter was the maximum value of the L2 norm of u(x,t) for t > tW, expressed as a
percentage of the maximum value of the L2 norm of u(x,t) for t < tW:


max L2 u x, t
t tW

L 2
(13)
max L u x, t
2
t tW

The last parameter is the integral over time of the L2 norm of u(x,t) for t > tW, expressed as a percentage
of the maximum value of the L2 norm of of u(x,t) for t < tW. Since the denominator is not an integral over
time, the numerator is normalized by dividing it by the time interval:

tf

L u x, t dt t tp
2
f
tp
L2t (14)

max L2 u x, t
t tW

In Tables 3 to 5 the results are shown for various values of the minimum loss factor, using the various
variations of the absorbing layer damping.

Table 3 Maximum displacement of the reflected waves

Q-1min 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75
Homog. 29.7% 9.5% 11.2% 15.7% 20.4% 25.1% 29.7% 34.2% 38.6% 42.8% 46.8%
Linear 60.8% 28.4% 14.4% 7.6% 4.2% 3.5% 4.1% 4.8% 5.4% 6.0% 6.6%
Quadratic 80.9% 46.5% 28.4% 17.9% 11.5% 7.4% 5.4% 4.4% 3.7% 3.4% 3.9%
Hyperbolic 64.9% 31.6% 16.3% 8.5% 5.1% 3.9% 4.6% 5.3% 6.0% 6.5% 7.0%

Table 4 Maximum L2 norm of the displacement of the reflected waves

Q-1min 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75
Homog. 29.6% 10.8% 9.9% 13.3% 17.1% 21.1% 25.0% 28.8% 32.5% 36.1% 39.5%
Linear 57.4% 28.2% 15.0% 8.3% 5.4% 4.5% 5.0% 5.6% 6.3% 6.9% 7.6%
Quadratic 74.0% 44.4% 28.0% 18.1% 11.9% 8.5% 6.5% 5.5% 5.1% 5.2% 5.6%
Hyperbolic 60.6% 30.8% 16.4% 9.1% 6.2% 5.2% 5.7% 6.4% 7.1% 7.7% 8.3%

Table 5 Time integration of the L2 norm of the displacement of the reflected waves

Q-1min 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75
Homog. 19.1% 7.0% 5.9% 7.3% 9.2% 11.3% 13.4% 15.6% 17.9% 20.1% 22.3%
Linear 36.8% 18.2% 9.7% 5.6% 3.6% 2.9% 3.0% 3.2% 3.6% 4.0% 4.4%
Quadratic 47.3% 28.8% 18.4% 12.2% 8.4% 6.0% 4.5% 3.7% 3.3% 3.3% 3.4%
Hyperbolic 38.8% 20.1% 11.1% 6.5% 4.2% 3.4% 3.5% 3.8% 4.3% 4.7% 5.1%

15
As can be seen, the three considered approaches lead to a considerable improvement over the
homogeneous absorbing layer. They all result in very similar results, although the hyperbolic variation is
somewhat worse than the other two. Between the linear and quadratic approaches, the first shows
lower values of the L2 norm, but the second as an inferior value for the maximum amplitude of the
reflected waves.

In Figure 6 the displacement at the middle of the elastic medium is shown once again, now comparing
the best results from the homogeneous, linear and quadratic approaches. It can be seen that the main
difference between these last two is that the reflected waves arrive later in the case with the quadratic
variation of the damping, and exhibits less peaks.

Figure 6 Displacement at the middle of the elastic medium due to the Ricker Wavelet excitation for various
functions of the minimum variation of the absorbing layer

Although there is no clear advantage between one and the other, the quadratic variation was chosen to
implement the continuous integration of the loss factor (in the next section), since it has a smoother
variation of damping in the first elements, thus hopefully helping to attenuate reflected waves at the
interface that may turn out to be more relevant in further analysis. However, the linear discretized
variation was still analyzed in detail in the optimization section.

16
Non-homogeneous Continuous Absorbing Layer

Since some reflection still exists when the waves cross adjacent elements with different values of the
loss factor, elements with a continuous variation of damping were assembled.

The first approach was to use a single element the size of the absorbing layer (with length equal to the
wavelength), with the quadratic variation of the loss factor accounted for in the integration of the
damping matrix. The element had the following definition for the damping matrix:

Ce eMe eK e (15)

where the superscript e stands for the element in question, and the Rayleigh quotients are calculated
according to equation (6). The loss factor in that equation was defined as a function of the horizontal
coordinate of the element (, see Figure 7 for reference):

1
2

Q 1
Qmin 1
(16)
2

Figure 7 Quadrilateral finite element local coordinates

The damping matrix was then obtained by exact integration of the formulation in equation (15).

Although the resulting matrix is exact, there was no appreciable reduction of the reflected waves
compared to the model without an absorbing layer.

To determine the reason for the ineffectiveness of the absorbing element, two models were compared:
the first used a discretized layer with the same length as the absorbing layer, but no damping was
defined; the second also had a layer with the same length, but instead of being discretized, it was
composed of a single element, also without damping.

The results are shown in Figure 8, which shows the displacement in the middle of the elastic medium for
both cases.

17
Figure 8 Displacement at the middle of the elastic medium for a discretized layer and for a continuous (one-
element) layer, both with no damping

In the model with a single-element layer, the waves are reflected much sooner, just before the 0.08
seconds mark, which is the time that waves take to travel to the interface and return to the middle
section of the elastic medium (see Table 2). Therefore, this reflection occurs at the interface.

It stands to reason that the reflected waves that occurred in the model with the continuous absorbing
layer werent due to the inadequacy of the damping, but due to the sudden variation in element size.

Since reducing the element size leads to a very short absorbing layer that is incapable of absorbing the
majority of the incident elastic waves, the second approach was to have a discretized layer, with the
variation of the loss factor implemented in each element. For that effect it is necessary to introduce
additional variables that can describe the value and variation of the loss factor for each element:

1
2

Q 1
Qmin 1
d f , (17)
2

where d is the normalized distance from the interface between the elastic medium and the layer to the
closest side of the element in question (x), and f is the normalized size of the element (ex) relative to the
length of the absorbing layer (habs). They can be expressed as:

d x habs , f ex habs (18)

18
Once again, the elements were integrated to obtain the exact damping matrix. The results are shown
and compared to the discretized approach on Figure 9.

Figure 9 Displacement at the middle of the elastic medium for a absorbing layer with quadratic variation, both
discretized for each element and continuous inside each element

It can be seen that the results are very similar. In the following section, the influence of the loss factor
and the length of the absorbing layer will be tested, and a more thorough comparison of the efficiency
of both approaches (discretized and continuous variation) will be presented.

Optimization of the Non-homogeneous Absorbing Layer

To better understand the influence of the minimum loss factor and the length of the absorbing layer, a
parametric study was carried. The length varied from one to five times the wavelength. The loss factor
varied from 0 to 3 in steps of 0.25. For each case, the value of the maximum displacement and of the
time integration of the L2 norm of the displacement was obtained.

Figures 10 to 13 show these results for linear discretized, quadratic discretized and quadratic continuous
variation of the loss factor, represented as a function of the minimum loss factor. The results shown are
for an absorbing layer with length equal to the wavelength and five times the wavelength.

Figures 14 and 15 show the optimum parameter value attained as a function of the length of the layer.

19
Figure 10 Maximum displacement, length of the layer equal to wavelength

Figure 11 Time integration of the L2 norm of the displacement, length of the layer equal to wavelength

20
Figure 12 Maximum displacement, length of the layer five times the wavelength

Figure 13 Time integration of the L2 norm of the displacement, length of the layer five times the wavelength

21
Figure 14 Optimum maximum displacement as a function of the length of the layer

Figure 15 Optimum time integration of the L2 norm of the displacement as a function of the length of the layer

22
The values that lead to these figures are presented on Appendix A, along with their visual representation
for different layers length.

In Figures 10 to 13 it can be seen that theres a clear local minimum for all cases, but the exact value
depends on the variation of the loss factor (linear or quadratic) and on the length of the absorbing layer.
In the particular case of the layer with length equal to five times the wavelength, the minimum loss
factor that minimizes umax is not the same as the one that leads to the minimum Lt2. It should be noted
that the vertical axis of these figures is in logarithmic scale, so the dependence of the studied
parameters on the minimum loss factor is steeper than what may appear.

Figures 14 and 15 show that the minimum value attainable for both umax and Lt2 improves asymptotically
with the layers length. There is a considerable improvement when the length goes from being equal to
the wavelength to twice the wavelength.

From analyzing all cases (see Appendix A), it can be concluded that:

The quadratic variation of the loss factor generally leads to better results when compared to the
linear variation (the only exception being the time integration of the L2 norm for the shortest
absorbing layer, and even it that case only by less that 0.5%);
There is no advantage in considering continuous variation of the loss factor for each different
element (most likely, the fact that there are 24 elements per wavelength assures that the
discreet variation is smooth enough);

Since the continuous variation requires each element to be correctly oriented, and there was no
advantage in its use, it was abandoned in further tests.

Optimum Loss Factor as a Function of the Layers Length

Since the time integration of the L2 norm gives a more general idea of the total reflection of the incident
waves, the optimum loss factor was chosen as the value that leads to the minimum value of the norm.
This was done only for the quadratic discretized variation.

To express the optimum loss factor as a function of the layers length, some different regression
techniques were tried. The one with the best fit was the power law, which yielded the following
expression:

0.706
1 h
Qmin 2.540 abs (19)

The power law is a good approach from a theoretical point of view: as the length of the layer tends to
zero, the loss factor needed to absorb the incident elastic waves grow to infinity; as the length grows to
infinity, the loss factor diminishes until no damping is needed at all.

23
The optimum values of loss factor obtained in the parametric analysis and the ones obtained with this
regression are summarized in table 6.

Table 6 Optimum loss factor as a function of the layers length

habs [m] habs / Qmin-1* Qmin-1


4 1.007 2.50 2.528
8 2.013 1.50 1.550
12 3.020 1.25 1.164
16 4.026 1.00 0.950
20 5.033 0.75 0.812

Since the properties of the material are already taken into account in the calculation of the wave speed
(and therefore the wavelength), it seems reasonable to expect this results to be independent of the
material parameters.

To test this assumption, four other soils were tested: the same hard soil, but with a Poisson ratio equal
to 0.3 (a common value for soils), 0.4 and 0.45, and a soft soil, but with Poisson ratio equal to 0. Table 7
shows the five soils and respective wavelengths.

Table 7 Material properties for the different materials considered (see Table 1) and respective wavelength

E cP P

[MPa] [kg/m3] [m/s] [m]
Hard soil 1 200.0 0 2 000 316.228 3.974
Hard soil 2 200.0 0.3 2 000 366.900 4.611
Hard soil 3 200.0 0.4 2 000 462.910 5.817
Hard soil 3 200.0 0.45 2 000 615.882 7.739
Soft soil 48.1 0 1 850 161.246 2.026

Figures 16 and 17 show the value of the optimization parameters used before, when using the optimum
loss factors estimated by equation (19), for the four different materials. The values that served as the
basis for these figures are presented on Appendix B.

It can be seen that the estimated values for the optimum loss factor do lead to very good results for the
other materials, so it is very likely that are only dependent on the length of the layer relative to the
wavelength.

It is also of note that, as the Poisson ratio increases, the amplitude of the reflected waves decreases
considerably. The same tests were carried out for a plane strain model (see appendix B), which yield
results very close to the hard soil with =0, and no variation in the results when the Poisson ratio
changes. It is still not clear why the Poisson ratio affects the efficiency of the absorbing layer (see the
Future Work section)

24
Figure 16 Predicted optimum maximum displacement as a function of the length of the layer for the 5 materials

Figure 17 Predicted optimum time integration of the L2 norm of the displacement as a function of the length of
the layer for the 5 materials

25
Detuned Absorbing Layer

Since its not always possible to define so clearly the frequency content of the dynamic loads, and even
if it is, it may happen that theres a wide range of relevant frequencies, it is important to assure that
underestimating or overestimating the prevailing frequency doesnt lead to a drastic drop in the
efficiency of the absorbing layer.

To that effect, the value assumed as the load frequency for the absorbing layer (now dubbed CALM) was
changed to take different values from the Ricker wavelet (r). Figure 18 shows the maximum
displacement as a function of the ratio of the load frequency to the layer frequency.

Figure 18 Amplitude of the reflected waves as a function of the ratio of the load to layer frequency

From the analysis of the results, one can confirm that the damping parameters defined in equation (6)
do lead to maximum absorption of the desired frequency.

Furthermore, it becomes evident that the efficiency of the absorbing layer suffers the most when the
frequency of the load is lower than what the absorbing layer is prepared for. As will be shown in the
next section, the absorbing layer is not as efficient in absorbing low frequency elastic waves, which
explains these results. It is therefore preferable to underestimate the dominating frequency of the loads
than overestimating it.

26
Two-Dimensional Absorbing Layer

After concluding that the Caughey Absorbing Layer Method works properly for one-dimensional
problems, a two-dimensional model was assembled. A plane stress analysis was considered, but the
principle would be the same for plane strain.

The Model

The model created represents an elastic half-space, with a horizontal free-surface where a dynamic
vertical point load is applied. The intensity of the load as a function of time follows the Ricker wavelet
(equation (8)). The maximum intensity is 1 kN, and the frequency is once again equal to 500 rad/s.

In a two-dimensional problem, particularly one with a free-surface, besides the already mentioned
pressure waves, two other types of waves can be observed: shear waves (which exhibit displacement in
the direction perpendicular to their direction of propagation) and Rayleigh waves [15] (which cause an
elliptical retrograde motion in the neighborhood of the surface, as depicted in Figure 19).

Figure 19 Rayleigh wave propagation (adapted from en.wikipedia.org)

The speed of the shear waves can be expressed as

cS E 2 1 (20)

For the Rayleigh waves speed there isnt an explicit formulation, but it can be estimated using the
following relation to the shear waves speed [16]

cR 0.87 1.12 1 cS (21)

For the hard soil used initially, but with a Poisson ratio of 0.3 (a typical value for soils), equations (20)
and (21) yield the results presented in Table 8.

27
Table 8 Wave speed and wavelength for the various types of waves, for the hard soil with = 0.3

c [m/s] [m]
Pressure waves 366.900 4.611
Shear waves 196.116 2.464
Rayleigh waves 181.935 2.286

It must be noted that the pressure waves have a greater wavelength than the other two types of wave,
so it is to expect that, if the absorbing layer is wide enough to absorb the pressure waves, it will also be
able to absorb the shear and Rayleigh waves.

Since the line of action of the applied force defines a plane of symmetry for the half-space, only one half
needs to be modeled. The resulting boundary must have symmetry conditions: in this case, horizontal
displacement and rotation arent allowed.

To implement this problem in finite elements, some sort of absorbing boundary or layer must be
implemented in the remaining sides of the model, like the Caughey Absorbing Layer illustrated in Figure
20.

Figure 20 Two-dimensional model with CALM (adapted from Semblat [10])

The size of the elastic medium was restrained to only one multiple of the pressure waves wavelength,
due to the considerable increase in the number of degrees of freedom when compared with the one-
dimensional model. The layers length was again considered to be a multiple of the wavelength. Like in
the one-dimensional model, the absorbing layer itself has fixed boundary conditions at its end.

28
Lysmer Absorbing Boundary Conditions

To understand how the CALM compares to other classical proposed in the literature to absorb the
incident waves, the model was also implemented with Lysmer boundaries [3], a type of Absorbing
Boundary Conditions widely used in similar problems.

As the Lysmer boundaries are viscous in nature, they are implemented in the Finite Element Model by
assigning dampers to each degree of freedom at the boundary (which is not a layer, like CALM, but a
surface), as depicted on Figure 21.

Figure 21 Two-dimensional model with Lysmer Absorbing Boundary Conditions (adapted from Semblat [10])

The value of damping depends on the direction of the degree of freedom relative to the surface of the
boundary: degrees of freedom normal to the surface have damping proportional to the speed of the
pressure waves,

cn aAinf cP (22)

The degrees of freedom tangent to the boundary surface have damping proportional to the speed of the
shear waves

ct bAinf cS (23)

In the above equations, Ainf stands for the area of influence of the node to which the degree of freedom
is associated. Since the model is based on plane stress, the area of influence of each node is equal to the

29
sum of half the side of the adjacent finite elements, multiplied by the thickness of the model. For a
lateral interior node (one that isnt in a corner), it yields

Ainf e y 2 e y 2 ez e y ez (24)

For corner nodes (like the ones at the surface of the model), the area of influence is

Ainf ey ez 2 (25)

The dimensionless parameters a and b depend on the angle of incidence of the waves, usually taken as
unitary for angles of incidence lower than 30 degrees.

These dampers are arguably enough for the lateral boundaries, at least as long as all the loads are of
transient nature [3], but the bottom boundaries require additional consideration: in the absence of fixed
boundary conditions or some sort of elastic boundary, there would be no limit to the displacement of
the model, which would allow it to display rigid body translation.

The implemented boundary must be viscoelastic. The damping parameters are the same as above. The
elastic component is achieved by introducing springs (as shown of Figure 21). As before, the value of
stiffness depends on the orientation of the degree of freedom. For degrees of freedom normal to the
surface, the stiffness can be expressed as

E 1
kn Ainf (26)
1 1 2 H
The degrees of freedom tangent to the boundary surface have stiffness given by

E
kt Ainf (27)
2 1 H

The parameter H is the height of the soil that would be below the boundary. A semi-infinite medium
cant actually be modeled, since an infinite height would lead to zero stiffness, but various values of H
were tested.

As stated before, it is arguable if the lateral boundaries require an elastic component for transient loads,
so this approach was also tested, since the springs may better simulate the elastic behavior of the solid
that would exist beyond the boundary.

Results

In total, four kinds of model were tested: without any sort of absorbing boundary or layer (just fixed
boundaries), with Lysmer boundaries (viscous lateral boundaries and viscoelastic bottom boundaries),

30
with Lysmer boundaries and springs (both boundaries are viscoelastic) and with CALM (using quadratic
discretized variation, and estimating the optimum loss factor with equation (19)).

To assure that the absorption is affecting the various types of waves, the chosen displacements are the
vertical and horizontal ones at the surface of the elastic medium, which is where the Rayleigh waves
have a bigger influence (pressure and shear waves are important throughout the model).

The parameters used to characterize the efficiency of the ABCs and absorbing layers were similar to the
ones used before, with some adaptations. First, there is no longer a single direction of displacement, but
two. It must also be noted that, due to the contribution of the Rayleigh waves to the behavior in the
area of interest, it is important to also compare the total displacement of the nodes at the surface:

utotal u x2 u y2 (28)

The other main difference from the one-dimensional case is that the maximum displacement after the
waves have left the elastic medium is now expressed as a percentage of the maximum displacement
before that instant, since the action is no longer a prescribed displacement:

max u x, t
x ,t t p
umax (29)
max u x, t
x ,t t p

Results are presented in Figures 22 and 23, but only for total displacement. The values for horizontal,
vertical and total displacement are presented in Appendix C.

It is clear that, although the Lysmer boundaries do provide a good solution to the reflecting elastic
waves problem, the CALM approach leads to better results. The lateral springs dont seem to improve
the Lysmer boundaries efficiency, at least for the analyzed parameters. It is possible that the
displacements at the boundary are more correctly modeled with the addition of the springs, but the
parameters used are only concerned with the maximum value overall and the norm of the
displacements across the surface of the medium. It is also clear that the stiffness of the springs at the
bottom boundary doesnt affect the chosen parameters in a meaningful way.

For the absorbing layers, however, the length of the boundary has a very clear influence in the results. A
layer with length equal to the wavelength leads to better results than the Lysmer boundaries, but by
increasing the length, there is a considerable improvement in the absorption of the elastic waves.
Naturally, the CALM approach has the disadvantage of requiring a bigger model, which requires more
degrees of freedom, increasing considerably the solution time (as discussed in Computing Time section).

In Figure 24 the vertical displacement at the point where the load is applied is shown, for the model
with no boundaries, Lysmer boundaries and CALM. It can be seen that both the Lysmer and CALM
approaches absorb most of the incident waves, but the latter not only reduces the amplitude of the
reflection, it also cuts off the higher frequency content of those waves.

31
Figure 22 Maximum displacement of the reflected waves, two-dimensional model

Figure 23 Time integration of the L2 norm of the displacement of the reflected waves, two-dimensional model

32
Figure 24 Vertical displacement of the elastic medium for various approaches, two-dimensional model

33
In Figure 25 the same results are shown only for a larger model with no boundaries and for CALM with
different lengths and loss factors. Once again it is clear that the frequency of the reflected waves goes
down with the increase of the length, as does their amplitude.

Figure 25 Vertical displacement of the elastic medium for different lengths of the absorbing layer, two-
dimensional model

Optimization of the Absorbing Layer

As before, a detailed optimization of the response as a function of the loss factor was done, but only for
the absorbing layer with length equal to the wavelength. The results are show on Figures 26 and 27, and
summarized in Table 9.

Table 9 Optimum loss factor values for the various parameters considered, compared to the predicted value

umax umax L2 L2 L t2 L t2
Qmin-1 Qmin-1* Qmin-1 Qmin-1* Qmin-1* Qmin-1*
[%] [%] [%] [%] [%] [%]
ux 1.13% 1.9 0.87% 1.74% 1.75 1.13% 3.72% 1.8 2.22%
uy 2.528 0.48% 2.3 0.48% 2.528 1.78% 2.9 1.76% 2.528 4.64% 2.6 4.64%
utotal 0.48% 2.35 0.48% 1.81% 2.4 1.80% 4.38% 2 4.26%

34
Figure 26 Maximum displacement of the reflected waves with CALM, two-dimensional model

Figure 27 Time integration of the L2 norm of the displacement of the reflected waves with CALM, two-
dimensional model

35
It is easy to verify that the optimum loss factor is not the same for the horizontal, vertical and total
displacement, neither for all the parameters considered. The optimum value for the horizontal
displacement is lower than the predicted value. The optimum value for the vertical and total
displacement is closer to the prediction, even though the absorbing layer was optimized for the pressure
waves which, at the surface of the elastic medium, are expected to travel in the horizontal direction.

It is however clear that the predicted values dont lead to results much worse than the optimum value,
and from the figures it can be seen that exceeding the actual optimum value is preferable than to
underestimate it.

Computing Time

To assess the usefulness of the Caughy Absorbing Boundary Method, it is important to be aware of the
solution time for its implementation, both as a consequence of the increase in model complexity (due to
the increase of the number of degrees of freedom), and of the potential increase in the number of
iterations for the solution.

The first is straightforward to evaluate. The additional layer requires a number of extra degrees of
freedom based on its length (and the geometry of the problem). In the particular case study of the plane
stress model with or without the absorbing layer, the number of degrees of freedom can be computed
from the following equation:

ngdl n 1 e ,
2
(30)

where n stands for the length of the absorbing layer relative to the wavelength , and e is the size of the
elements (assuming they all have the same length and width).

In the case of the model with the Lysmer Absorbing Boundary Conditions, since the outer nodes arent
fixed, the number of degrees of freedom is

ngdl e 1
2
(31)

In the particular model analyzed, the number of degrees of freedom is summarized in Table 10.

Table 10 Number of degrees of freedom for the two-dimensional model

n CALM Lysmer
0 529
0.52 1225
1 2116
576
2 4761
3 8464
4 13225

36
The computing time of the two-dimensional model for the various approaches was determined and
compared, taking into account the number of degrees of freedom for each approach. The results are
presented in table 11.

Table 11 Number of degrees of freedom for the two-dimensional model

No boundaries Lysmer CALM


n total time [s] time/dof [s] total time [s] time/dof [s] total time [s] time/dof [s]
0 168 0.318 - - - -
0.52 336 0.274 178 0.309 647 0.528
1 480 0.227 175 0.304 1429 0.675
2 969 0.204 196 0.340 3322 0.698
3 1868 0.221 165 0.286 6536 0.772
4 2829 0.214 169 0.293 9485 0.717

It is clear that the CALM approach is taking much more computation time per degree of freedom than
the other approaches. At first, this result is not particularly unexpected: as was stated before, the
implementation of the CALM in an explicit dynamics analysis was abandoned because the beta damping
required a much smaller step size than what would be needed without damping.

However, analyzing the Ansys output, the total number of steps (and substeps) is the same for all
approaches. It would be then reasonable to expect solution time to be a function of the complexity of
the model (the number of degrees of freedom), so the solution time should be about the same for the
approaches with the same number of degrees of freedom.

The explanation that seems more likely is because of the way the damping was applied: since the Ansys
software doesnt allow defining the alpha damping for each element independently, arbitrary damping
elements where added, connecting the desired nodes. Even though these dont change the solution
time, during post-processing, for each time step, their stress, strain and energy information is stored,
leading to an overall slower process.

Since Ansys control output settings didnt solve the problem, a way to circumvent this problem would
be to implement finite elements with independent Rayleigh damping, which in turn would allow for a
more flexible implementation of the absorbing layers.

37
Conclusions

The Caughey Absorbing Layer Method has been show to work effectively to mitigate the problem of the
spurious wave reflections at the boundaries, for one and two-dimensional models. The absorbing layer
not only greatly reduces the amplitude of the reflected waves, but also filters their high frequency
content.

It was shown that the quadratic variation of the loss factor leads to better results, although there was no
clear benefit of considering a continuous variation of the loss factor for each element of the absorbing
layer, as opposed to a discretized variation, with each individual element having constant damping.

An estimated value for the optimum loss factor as a function of the length of the layer in relation to the
wavelength to absorb was proposed, and shown to lead to good results in all further analysis.

It is important to note that, although the CALM is more efficient than the Lysmer boundaries, it couldnt
be implemented with standard explicit time integration, and it has the disadvantage of requiring a
higher number of degrees of freedom that make the analysis slower when compared to absorbing
boundary methods.

The referred technical difficulties in the Ansys implementation have prevented the assessment of how
computationally efficient the method really is, so it is still difficult to comment on its advantage relative
to other approaches.

38
Future Work

The next logical step would be to test the two-dimensional model with plain strain elements. Since the
results in the one-dimensional model with plain strain were very close to the ones obtained with plane
stress, it is expected that the same will happen for the two-dimensional model.

The influence of the Poisson ratio in the plane stress results is still unexplained, lacking further analysis
and theoretical research. Performing a parametric optimization for different values of the Poisson ratio
may provide some insight to this phenomenon.

The optimization process that led to equation (19) (the optimum minimum loss factor as a function of
the layers length relative to the wavelength), although successful, was relatively coarse, so a more
refined analysis could lead to a more exact approximation. This optimization could also be done for the
two-dimensional model, in which it was carried out only for a layer with length equal to the wavelength.

An important development to pursue would be to implement the finite elements with independent
Rayleigh damping in the Ansys software. Then it would be possible to analyze how efficient the CALM is
when compared with other absorbing boundary and layer techniques, and apply it to more complex
problems.

39
References

[1] L. Gaul, M. Kgl, M. Wagner. "Boundary element methods for engineers and scientists", Springer-
Verlag Berlin Heidelberg, Germany, 2003.

[2] B. Engqui, A. Majda. "Absorbing boundary conditions for the numerical simulation of waves",
Mathematics of Computation, Vol. 31, No. 139, pp. 629-651, American Mathematical Society, 1977.

[3] J. Lysmer, R. L. Kuhlemeyer. Finite dynamic model for infinite media, Journal of the Engineering
Mechanics Division, Vol. 95, No. 4, pp. 859-878, 1969.

[4] P. Bettess, O. C. Zienkiewicz. "Diffraction and refraction of surface waves using finite and infinite
elements", International Journal for Numerical Methods in Engineering, Vol. 11, No. 8, pp. 1271-1290,
1977.

[5] H. Lane, P. Kettil, M. Enelund, T. Ekevid, N Wiberg. "Absorbing boundary layers for elastic wave
propagation", VIII International Conference on Computational Plasticity (COMPLAS VIII), Barcelona,
2005.

[6] J. Berenger. "A perfectly matched layer for the absorption of electromagnetic waves", Journal of
Computational Physics, Vol. 114, No. 2, pp. 185200, 1994.

[7] U. Basu, A.K. Chopra. "Perfectly matched layers for time-harmonic elastodynamics of unbounded
domains: theory and finite-element implementation", Computer Methods in Applied Mechanics and
Engineerinhg, Vol. 192, pp. 13371375, 2003.

[8] M. Chevalier. "Advances in the perfectly matched layer absorbing boundary condition and a
technique for efficiently modeling long path propagation with applications to finite difference grid
technique", PhD Dissertation, Department of Electrical Engineering of Stanford University, 2003.

[9] K.C. Meza-Fajardo, A.S. Papageorgiou. "A nonconvolutional, split-field, perfectly matched layer for
wave propagation in isotropic and anisotropic elastic media: stability analysis", Bulletin of the
Seismological Society of America, Vol. 98, pp. 1811-1836, 2008.

[10] J-F. Semblat, L. Lenti, A. Gandomzadeh. "A simple multi-directional absorbing layer method to
simulate elastic wave propagation in unbounded domains", International Journal for Numerical Methods
in Engineering, Vol. 85, pp. 1543-1563, 2011.

[11] Man Liu, D.G. Gorman. "Formulation of Rayleigh damping and its extensions", Computers and
Structures, Vol. 57, No. 2, pp. 277-285, 1995.

[12] R.W. Clough, J. Penzien. "Dynamics of structures", Mc Graw-Hill, United States of America, 1993.

[13] P. Macioce. "Viscoelastic Damping 101", Roush Industries, Inc., Livonia, Michigan, United States of
America.

40
[14] E. Madenci, I. Guven. "The finite element method and applications in engineering using Ansys",
Springer, United States of America, 2005.

[15] W.M. Telford, L.P. Geldart, R.E. Sheriff. "Applied geophysics", Cambridge University Press, United
States of America, 1990.

[16] Y.B. Yang, H.H. Hung. Wave propagation for train-induced vibrations: a finite/infinite element
approach, World Scientific Publishing, 2009.

41
Appendix A Results for the Optimization of the Non-homogeneous Absorbing Layer

In this appendix, the results of the parametric study of the optimal minimum loss factor for the one-
dimensional wave propagation problem are presented.

As stated before, the length of the absorbing layer varied from one to five times the wavelength. The
minimum loss factor varied from 0 to 3 in steps of 0.25. For each case, the value of the maximum
displacement (umax) and of the time integration of the L2 norm of the displacement (Lt2) was obtained.

Tables A.1 to A.6 show the results for linear discretized, quadratic discretized and quadratic continuous
variation of the loss factor, for the considered values of the minimum loss factor. The same information
is presented in graphical form in Figures A.1 to A.10.

Table A.1 Maximum displacement of the reflected waves for the linear discretized variation of the loss factor

Qmin-1 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3
habs=4 m 154.29% 60.84% 28.40% 14.37% 7.63% 4.18% 3.48% 4.14% 4.78% 5.40% 6.00% 6.61% 7.19%
habs=8 m 155.55% 28.60% 7.67% 2.50% 1.03% 1.29% 1.60% 1.92% 2.24% 2.57% 2.89% 3.22% 3.55%
habs=12 m 156.46% 14.38% 2.41% 0.57% 0.61% 0.79% 0.99% 1.19% 1.39% 1.60% 1.81% 2.03% 2.24%
habs=16 m 94.07% 4.98% 0.65% 0.32% 0.44% 0.57% 0.70% 0.84% 0.99% 1.14% 1.29% 1.44% 1.60%
habs=20 m 92.51% 2.66% 0.25% 0.28% 0.34% 0.44% 0.54% 0.65% 0.76% 0.87% 0.99% 1.11% 1.23%

Table A.2 Maximum displacement of the reflected waves for the quadratic discretized variation of the loss factor

Qmin-1 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3
habs=4 m 154.29% 80.88% 46.46% 28.38% 17.87% 11.53% 7.45% 5.39% 4.37% 3.71% 3.36% 3.87% 4.34%
habs=8 m 155.55% 47.16% 17.87% 7.67% 3.53% 1.69% 0.86% 0.97% 1.10% 1.22% 1.35% 1.47% 1.58%
habs=12 m 156.46% 28.69% 7.62% 2.44% 0.89% 0.39% 0.39% 0.46% 0.52% 0.58% 0.64% 0.70% 0.76%
habs=16 m 94.07% 11.63% 2.24% 0.55% 0.22% 0.19% 0.23% 0.26% 0.30% 0.33% 0.37% 0.41% 0.44%
habs=20 m 92.51% 7.47% 1.05% 0.20% 0.19% 0.18% 0.17% 0.17% 0.19% 0.22% 0.24% 0.26% 0.29%

Table A.3 Maximum displacement of the reflected waves for the quadratic continuous variation of the loss factor

Qmin-1 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3
habs=4 m 154.29% 83.78% 49.37% 30.88% 23.93% 13.13% 8.67% 6.13% 4.93% 4.17% 3.69% 3.71% 4.20%
habs=8 m 155.55% 48.64% 18.84% 8.25% 3.86% 1.88% 0.94% 0.97% 1.10% 1.22% 1.35% 1.47% 1.59%
habs=12 m 156.46% 29.54% 8.00% 2.61% 0.96% 0.41% 0.39% 0.46% 0.52% 0.58% 0.64% 0.70% 0.76%
habs=16 m 94.07% 11.95% 2.34% 0.58% 0.23% 0.19% 0.23% 0.26% 0.30% 0.33% 0.37% 0.40% 0.44%
habs=20 m 92.51% 7.66% 1.10% 0.21% 0.19% 0.18% 0.17% 0.17% 0.19% 0.22% 0.24% 0.26% 0.29%

42
Table A.4 Time integration of the L2 norm of the displacement for the linear discretized variation of the loss
factor

Qmin-1 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3
habs=4 m 83.54% 36.76% 18.22% 9.75% 5.59% 3.62% 2.94% 2.95% 3.24% 3.62% 4.01% 4.38% 4.73%
habs=8 m 71.08% 15.52% 4.65% 1.82% 1.19% 1.24% 1.44% 1.65% 1.87% 2.09% 2.30% 2.51% 2.72%
habs=12 m 50.98% 5.94% 1.25% 0.59% 0.65% 0.79% 0.95% 1.10% 1.26% 1.41% 1.57% 1.72% 1.87%
habs=16 m 31.24% 2.08% 0.41% 0.36% 0.48% 0.60% 0.71% 0.83% 0.95% 1.07% 1.18% 1.30% 1.41%
habs=20 m 11.19% 0.50% 0.21% 0.28% 0.38% 0.48% 0.57% 0.67% 0.76% 0.86% 0.95% 1.04% 1.14%

Table A.5 Time integration of the L2 norm of the displacement for the quadratic discretized variation of the loss
factor

Qmin-1 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3
habs=4 m 83.54% 47.31% 28.77% 18.42% 12.24% 8.39% 5.96% 4.48% 3.66% 3.31% 3.27% 3.41% 3.62%
habs=8 m 71.08% 24.47% 10.09% 4.62% 2.32% 1.35% 1.05% 1.06% 1.15% 1.26% 1.37% 1.47% 1.57%
habs=12 m 50.98% 11.12% 3.31% 1.20% 0.59% 0.47% 0.50% 0.56% 0.62% 0.68% 0.74% 0.80% 0.86%
habs=16 m 31.24% 4.48% 1.02% 0.36% 0.24% 0.26% 0.30% 0.34% 0.38% 0.42% 0.46% 0.50% 0.54%
habs=20 m 11.19% 1.11% 0.24% 0.13% 0.14% 0.17% 0.20% 0.23% 0.26% 0.29% 0.31% 0.34% 0.37%

Table A.6 Time integration of the L2 norm of the displacement for the quadratic continuous variation of the loss
factor

Qmin-1 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3
habs=4 m 83.54% 48.80% 30.42% 19.89% 15.16% 9.40% 6.76% 5.07% 4.07% 3.56% 3.40% 3.46% 3.63%
habs=8 m 71.08% 25.17% 10.60% 4.94% 2.51% 1.45% 1.09% 1.06% 1.15% 1.26% 1.37% 1.47% 1.57%
habs=12 m 50.98% 11.42% 3.46% 1.27% 0.62% 0.48% 0.50% 0.56% 0.62% 0.68% 0.74% 0.80% 0.86%
habs=16 m 31.24% 4.59% 1.07% 0.37% 0.24% 0.26% 0.30% 0.34% 0.38% 0.42% 0.46% 0.50% 0.54%
habs=20 m 11.19% 1.13% 0.25% 0.13% 0.14% 0.17% 0.20% 0.23% 0.26% 0.28% 0.31% 0.34% 0.37%

43
Figure A.1 Maximum displacement, length of the layer equal to wavelength

Figure A.2 Time integration of the L2 norm of the displacement, length of the layer equal to wavelength

44
Figure A.3 Maximum displacement, length of the layer twice the wavelength

Figure A.4 Time integration of the L2 norm of the displacement, length of the layer twice the wavelength

45
Figure A.5 Maximum displacement, length of the layer three times the wavelength

Figure A.6 Time integration of the L2 norm of the displacement, length of the layer three times the wavelength

46
Figure A.7 Maximum displacement, length of the layer four times the wavelength

Figure A.8 Time integration of the L2 norm of the displacement, length of the layer four times the wavelength

47
Figure A.9 Maximum displacement, length of the layer five times the wavelength

Figure A.10 Time integration of the L2 norm of the displacement, length of the layer five times the wavelength

48
Appendix B Results for the One-Dimensional Model for Various Materials

Tables B.1 to B.5 show the results of the optimization parameters analyzed for the five different
materials tested (see Table 7), as a function of the layers length, using the minimum loss factor
predicted by equation (19), for a plane stress model.

Tables B.6 to B.9 show the same results (except for the soft soil), for a plane strain model.

Table B.1 Results for the hard soil 1 ( = 0), plane stress model

habs [m] habs / Qmin-1 umax Lt2


4.00 1.007 2.528 3.42% 3.22%
5.12 1.288 2.124 2.31% 2.08%
6.08 1.530 1.881 1.65% 1.62%
8.00 2.013 1.550 0.87% 1.10%
10.08 2.537 1.317 0.60% 0.76%
12.00 3.020 1.164 0.49% 0.54%
14.08 3.543 1.040 0.39% 0.38%
16.00 4.026 0.950 0.24% 0.27%
18.08 4.550 0.872 0.19% 0.19%
20.00 5.033 0.812 0.15% 0.14%
22.08 5.556 0.757 0.09% 0.09%

Table B.2 Results for the hard soil 2 ( = 0.3), plane stress model

habs [m] habs / Qmin-1 umax Lt2


4.50 0.976 2.584 3.39% 3.19%
5.76 1.249 2.171 2.19% 2.08%
6.84 1.483 1.923 1.52% 1.59%
9.00 1.952 1.584 0.78% 1.01%
11.34 2.459 1.346 0.48% 0.64%
13.50 2.928 1.190 0.34% 0.44%
15.84 3.435 1.063 0.24% 0.30%
18.00 3.904 0.971 0.18% 0.22%
20.34 4.411 0.891 0.12% 0.15%
22.50 4.880 0.830 0.07% 0.11%
24.84 5.387 0.774 0.05% 0.08%

49
Table B.3 Results for the hard soil 3 ( = 0.4), plane stress model

habs [m] habs / Qmin-1 umax Lt2


6.00 1.031 2.485 2.52% 2.48%
7.68 1.320 2.088 1.44% 1.57%
9.12 1.568 1.849 0.94% 1.11%
12.00 2.063 1.523 0.47% 0.61%
15.12 2.599 1.294 0.26% 0.35%
18.00 3.094 1.144 0.16% 0.24%
21.12 3.631 1.022 0.11% 0.16%
24.00 4.126 0.934 0.08% 0.12%
27.12 4.662 0.857 0.06% 0.08%
30.00 5.157 0.798 0.04% 0.06%
33.12 5.694 0.744 0.03% 0.05%

Table B.4 Results for the hard soil 4 ( = 0.45), plane stress model

habs [m] habs / Qmin-1 umax L t2


7.75 1.001 2.538 1.77% 1.94%
9.92 1.282 2.132 0.84% 1.14%
11.78 1.522 1.888 0.43% 0.77%
15.50 2.003 1.556 0.18% 0.40%
19.53 2.523 1.321 0.10% 0.22%
23.25 3.004 1.168 0.06% 0.14%
27.28 3.525 1.044 0.04% 0.09%
31.00 4.005 0.954 0.03% 0.07%
35.03 4.526 0.875 0.02% 0.05%
38.75 5.007 0.815 0.02% 0.04%
42.78 5.528 0.760 0.01% 0.03%

Table B.5 Results for the soft soil ( = 0), plane stress model

habs [m] habs / Qmin-1 umax Lt2


2.00 0.987 2.564 3.53% 3.27%
2.56 1.263 2.154 2.38% 2.09%
3.04 1.500 1.907 1.72% 1.63%
4.00 1.974 1.571 0.91% 1.12%
5.04 2.487 1.335 0.61% 0.78%
6.00 2.961 1.180 0.50% 0.56%
7.04 3.474 1.054 0.40% 0.40%
8.00 3.948 0.963 0.25% 0.29%
9.04 4.461 0.884 0.20% 0.21%
10.00 4.935 0.823 0.16% 0.15%
11.04 5.448 0.767 0.11% 0.10%

50
Table B.6 Results for the hard soil 1 ( = 0), plane strain model

habs [m] habs / Qmin-1 umax Lt2


4.00 1.007 2.528 3.42% 3.22%
5.12 1.288 2.124 2.31% 2.08%
6.08 1.530 1.881 1.65% 1.62%
8.00 2.013 1.550 0.87% 1.10%
10.08 2.537 1.317 0.60% 0.76%
12.00 3.020 1.164 0.49% 0.54%
14.08 3.543 1.040 0.39% 0.38%
16.00 4.026 0.950 0.24% 0.27%
18.08 4.550 0.872 0.19% 0.19%
20.00 5.033 0.812 0.15% 0.14%
22.08 5.556 0.757 0.09% 0.09%

Table B.7 Results for the hard soil 2 ( = 0.3), plane strain model

habs [m] habs / Qmin-1 umax Lt2


4.50 0.976 2.584 3.62% 3.30%
5.76 1.249 2.171 2.43% 2.09%
6.84 1.483 1.923 1.76% 1.62%
9.00 1.952 1.584 0.94% 1.13%
11.34 2.459 1.346 0.62% 0.79%
13.50 2.928 1.190 0.51% 0.58%
15.84 3.435 1.063 0.41% 0.41%
18.00 3.904 0.971 0.26% 0.30%
20.34 4.411 0.891 0.20% 0.21%
22.50 4.880 0.830 0.16% 0.16%
24.84 5.387 0.774 0.12% 0.11%

Table B.8 Results for the hard soil 3 ( = 0.4), plane strain model

habs [m] habs / Qmin-1 umax Lt2


6.00 1.031 2.485 3.30% 3.17%
7.68 1.320 2.088 2.21% 2.08%
9.12 1.568 1.849 1.57% 1.60%
12.00 2.063 1.523 0.82% 1.08%
15.12 2.599 1.294 0.58% 0.73%
18.00 3.094 1.144 0.47% 0.51%
21.12 3.631 1.022 0.33% 0.35%
24.00 4.126 0.934 0.23% 0.26%
27.12 4.662 0.857 0.18% 0.18%
30.00 5.157 0.798 0.13% 0.12%
33.12 5.694 0.744 0.08% 0.09%

51
Table B.9 Results for the hard soil 4 ( = 0.45), plane strain model

habs [m] habs / Qmin-1 umax L t2


7.75 1.001 2.538 3.45% 3.23%
9.92 1.282 2.132 2.33% 2.08%
11.78 1.522 1.888 1.67% 1.63%
15.50 2.003 1.556 0.88% 1.11%
19.53 2.523 1.321 0.60% 0.77%
23.25 3.004 1.168 0.49% 0.55%
27.28 3.525 1.044 0.39% 0.39%
31.00 4.005 0.954 0.25% 0.28%
35.03 4.526 0.875 0.19% 0.20%
38.75 5.007 0.815 0.15% 0.14%
42.78 5.528 0.760 0.10% 0.10%

52
Appendix C Results for the Two-Dimensional Plane Stress Model

The value of the optimization parameters for the horizontal and vertical displacements, as well as the
total displacement at the surface of the plane-stress two-dimensional model are presented in Tables C.1
to C.9

Table C.1 Maximum horizontal displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 139.70% - - -
1 81.25% 7.21% 6.45% 1.13%
2 8.52% 7.57% 7.11% 0.24%
3 0.06% 7.65% 7.33% 0.06%
4 0.00% 7.68% 7.44% 0.02%

Table C.2 Maximum vertical displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 32.92% - - -
1 14.93% 1.85% 1.94% 0.48%
2 1.57% 1.85% 1.86% 0.13%
3 0.04% 1.86% 1.85% 0.05%
4 0.04% 1.88% 1.86% 0.02%

Table C.3 Maximum total displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 33.44% - - -
1 20.34% 1.85% 1.96% 0.48%
2 5.70% 1.86% 1.87% 0.14%
3 0.04% 1.87% 1.87% 0.05%
4 0.04% 1.89% 1.87% 0.02%

53
Table C.4 Maximum L2 norm of the horizontal displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 150.02% - - -
1 83.31% 7.41% 7.14% 1.74%
2 11.41% 7.52% 7.29% 0.36%
3 0.03% 7.56% 7.39% 0.06%
4 0.00% 7.58% 7.45% 0.02%

Table C.5 Maximum L2 norm of the vertical displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 79.15% - - -
1 36.59% 4.79% 4.91% 1.78%
2 5.01% 4.88% 4.79% 0.65%
3 0.14% 4.91% 4.84% 0.27%
4 0.14% 4.93% 4.87% 0.13%

Table C.6 Maximum L2 norm of the total displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 86.66% - - -
1 62.27% 5.44% 5.46% 1.81%
2 27.21% 5.52% 5.40% 0.63%
3 0.14% 5.55% 5.46% 0.25%
4 0.14% 5.57% 5.49% 0.12%

54
Table C.7 Time integration of the L2 norm of the vertical displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 293.59% - - -
1 162.44% 12.94% 12.42% 3.72%
2 17.50% 13.03% 12.68% 0.74%
3 0.00% 13.11% 12.87% 0.15%
4 0.00% 13.16% 12.98% 0.04%

Table C.8 Time integration of the L2 norm of the vertical displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 192.58% - - -
1 92.43% 11.22% 11.84% 4.64%
2 9.96% 11.16% 11.19% 1.70%
3 0.27% 11.23% 11.20% 0.77%
4 0.27% 11.27% 11.23% 0.30%

Table C.9 Time integration of the L2 norm of the total displacement of the reflected waves, plane stress model

No Lysmer
habs/ Lysmer CALM
boundaries with springs
0 223.35% - - -
1 137.85% 11.52% 11.84% 4.38%
2 27.95% 11.51% 11.44% 1.51%
3 0.23% 11.58% 11.50% 0.66%
4 0.23% 11.62% 11.55% 0.26%

55

Vous aimerez peut-être aussi