[amorosi] parametric study on seismic ground response by finite element modelling
TRANSCRIPT
-
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
1/14
Parametric study on seismic ground response by finite element modelling
Angelo Amorosi a,*, Daniela Boldini b,1, Gaetano Elia c,2
a Technical University of Bari, Department of Civil and Environmental Engineering, via Orabona 4, 70125 Bari, Italyb University of Bologna, Department of Civil, Environmental and Material Engineering, viale Terracini 28, 40136 Bologna, Italyc Newcastle University, School of Civil Engineering & Geosciences, Drummond Building, NE1 7RU Newcastle upon Tyne, United Kingdom
a r t i c l e i n f o
Article history:Received 22 September 2009
Received in revised form 5 February 2010
Accepted 10 February 2010
Available online 21 March 2010
Keywords:
Seismic ground response analysis
Constitutive models
Numerical modelling
Finite element analysis
a b s t r a c t
In this paper the results of 2D FE analyses of the seismic ground response of a clayey deposit, performedadopting linear visco-elastic and visco-elasto-plastic constitutive models, are presented. The viscous and
linear elastic parameters are selected according to a novel calibration strategy, leading to FE results com-
parable to those obtained by 1D equivalent-linear visco-elastic frequency-domain analyses. The influence
of plasticity on the numerical results is also investigated, with particular reference to the relation
between the hysteretic and viscous damping effects. Finally, different boundary conditions, spatial dis-
cretisation and time integration parameters are considered and their role on the FE results discussed.
2010 Elsevier Ltd. All rights reserved.
1. Introduction
The site response analysis has traditionally been performed
using one-dimensional frequency-domain numerical scheme
based on the equivalent visco-elastic approach [13]. This ap-
proach has successfully been adopted in the last 30 years and it
is widely accepted in the engineering practice, although its limita-
tions are well-known. In particular, concerning this latter aspect, it
is worth remarking the following points:
soil behaviour is controlled by effective stresses while a total
stress approach is implemented in most equivalent visco-elas-
tic schemes, disregarding the soilfluid interaction [4] and
possible build-up of excess pore water pressure during seismic
events;
the mechanical behaviour of soil under cyclic loads is character-
ised by strong non-linearity, dependence on past stress-history,
reduction of shear stiffness with consequent hysteretic dissipa-
tion during the cycles, early irreversibility, etc. e.g. [59]. In
contrast, a fully-reversible soil model, with constant visco-elas-
tic soil properties (i.e. shear stiffness and damping ratio) over
the duration of earthquake shaking, is adopted in the traditional
frequency-domain analysis methods;
many engineering problems cannot be scaled down to the one-
dimensional case but require a soilstructure interaction analy-sis in two- or three-dimensional conditions[10,11].
Time-domain finite element or finite difference schemes are
nowadays available to solve the wave propagation problem in a
more realistic way, accounting for the solidfluid interaction by
means of a fully coupled effective stress formulation [4,12]. In
those schemes, the behaviour of the soil can be described using
either simple or sophisticated non-linear constitutive models of
different level of complexity. These numerical approaches permit
to include in one single analysis the evaluation of the site response
and the corresponding interaction with the existing structurese.g.
[1321].
Such approaches are seldom adopted in engineering practice by
non-expert users because both the model calibration procedures
and the code usage protocols are often unclear or poorly docu-
mented, leading to unrealistic results and, as such, obscuring the
possible benefits of the time-domain numerical analysis. The main
difficulties can be summarised as follow:
in linear finite element or finite difference analyses constant
values of stiffness and viscous properties have to be selected
according to a representative level of strain assumed to occur
during the earthquake. Depending on the characteristics of
the soil deposits these properties can be constant or variable
with depth;
sophisticated constitutive formulations are not yet available
in most commercial finite element or finite difference codes.
0266-352X/$ - see front matter 2010 Elsevier Ltd. All rights reserved.doi:10.1016/j.compgeo.2010.02.005
* Corresponding author. Tel.: +39 080 5963693; fax: +39 080 5963675.
E-mail addresses:[email protected](A. Amorosi), [email protected](D.
Boldini),[email protected](G. Elia).1 Tel.: +39 051 2090233; fax: +39 051 2090247.2 Tel.: +44 191 2227934; fax: +44 191 2225322.
Computers and Geotechnics 37 (2010) 515528
Contents lists available at ScienceDirect
Computers and Geotechnics
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p g e o
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://dx.doi.org/10.1016/j.compgeo.2010.02.005mailto:[email protected]:[email protected]:[email protected]://www.sciencedirect.com/science/journal/0266352Xhttp://www.elsevier.com/locate/compgeohttp://www.elsevier.com/locate/compgeohttp://www.sciencedirect.com/science/journal/0266352Xmailto:[email protected]:[email protected]:[email protected]://dx.doi.org/10.1016/j.compgeo.2010.02.005http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?- -
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
2/14
When implemented, their calibration is not straightforward and
requires non-conventional geotechnical tests, often not
included in standard geotechnical characterisation;
in time-domain schemes there are two sources of damping: vis-
cous damping, generally introduced through the Rayleigh [22]
formulation, and the hysteretic dissipation associated to the
irreversible material response. The amount of hysteretic damp-
ing, which is frequency independent, is strictly related to the
adopted material model. Viscous damping, which in contrast
is frequency dependent, is added to the dynamic equations of
motion to obtain stable numerical solutions and to account
for the soil damping at small strains, if an hysteretic model is
employed, or for the total amount of damping, if a non-dissipa-
tive constitutive formulation is adopted. In this respect, the
main issues are associated to the selection of the appropriate
target viscous damping ratio e.g. [23] and of the frequency
range required by the Rayleigh damping functione.g. [2426],
as they can play a crucial role on FE results;
finite element or finite difference methods use a finite discre-
tised domain to represent the infinite continuous soil medium.
Users are asked to define the extension of the finite domain, the
characteristics of the spatial discretisation (i.e. the dimension
and the type of elements) and the appropriate boundary condi-
tions to artificially simulate the far-field medium. While all
these aspects are well understood in the context of static anal-
yses, the literature concerning numerical analyses in dynamic
conditions is less exhaustive;
the integration of the dynamic equations of motion can be per-
formed adopting time-stepping schemes characterised by dif-
ferent accuracy, stability, algorithmic damping and run-time.
In the paper, some of these features are investigated by compar-
ing a set of 1D ground response numerical analyses performed in
the frequency-domain with the corresponding time-domain based
2D finite element simulations. 1D frequency-domain analyses
were performed modelling the soil as a single phase visco-elastic
equivalent-linear medium. These results, besides the possibledrawbacks they can contain, are taken as target solutions for the
2D finite element analyses based on linear visco-elasticity. This lat-
ter assumption is underpinned by the following hypothesis: the re-
sults of any 1D analysis performed in the frequency-domain and
based on (equivalent) linear visco-elasticity should, in principle,
coincide with the corresponding 2D finite element analysis per-
formed in the time-domain assuming the same constitutive behav-
iour, provided an appropriate calibration of the parameters is
adopted. The above comparison scheme is obviously no longer va-
lid once more complex constitutive laws are adopted in the FE
analysis as, for example, when plasticity is included in the
formulation.
In order to provide a useful framework for standard finite ele-
ment users, the use of advanced constitutive models was avoided.Soil behaviour, in fact, is described either in terms of visco-elastic-
ity, with viscous damping accounting for all the dissipative mate-
rial behaviour, or by means of simple visco-elasto-plasticity
assumptions. Realism is introduced in the investigation by consid-
ering a soil deposit characterised by variable stiffness and damping
ratio with depth.
The first part of the paper outlines the geometrical and geotech-
nical characteristics of the ideal soil deposits under study, the main
features of the adopted seismic motions and the criteria followed
for their selection. It also describes the numerical models em-
ployed for 1D and 2D ground response simulations and summa-
rises the results of equivalent-linear visco-elastic analyses
performed in the frequency-domain. A new procedure for the cal-
ibration of the Rayleigh parameters in FE time-domain analyses is
then proposed and validated.
In the subsequent section, the paper investigates the effect of
the introduction of plasticity in the soil constitutive assumption
and illustrates the results of different strategies adopted in order
to obtain a good matching between frequency and time-domain
analyses.
Finally, the influence of the boundary conditions, spatial dis-
cretisation and the time integration parameters on the results of
the FE simulations is reviewed.
2. Outline of the idealised problem
An ideal deposit of soft clay is assumed as the reference soil pro-
file, characterised by the following physical and mechanical
parameters: plasticity index IP = 44%, unit weight of volume of
the saturated soil c= 17kN/m3, overconsolidation ratio in termsof mean effective stress R= 1.5, small-strain shear stiffness
G0= variable with depth, Poissons ratio t0 = 0.25, small-straindamping ratio D0= 1.0%, coefficient at rest K0 = 0.6, cohesion
c0 = 0 and friction angle u0 = 24. The water table is assumed at
the ground surface.
Fig. 1. Profiles of the small-strain shear stiffness G0 (a) and shear wave velocityVS (b).
516 A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?- -
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
3/14
Three different thicknesses were considered for this ideal soil
deposit, namely 60, 120 and 240 m.
The assumed profile of the small-strain shear stiffness G0 with
depth was calculated adopting the relationship proposed by Viggi-
ani and Atkinson[27]:
G0pr
S p0
pr
nRm 1
wherepris a reference pressure taken equal to 1 kPa,p0 is the mean
pressure (resulting from the lithostatic pressure and the application
of a uniform load at the surface equal to 30 kPa), S, n and m are
parameters depending on the plasticity index IP(here set equal to
550, 0.82 and 0.36, respectively, according to the correlations pro-
posed by Viggiani and Atkinson [27] for fine-grained soils) and R
is the overconsolidation ratio in terms of mean effective stress.
The variation ofG0
with depth and the corresponding shear wave
Table 1
Main characteristics of the adopted input seismic motions.
Station Component Earthquake PGA (g) Duration (s) Dominant frequencyfp (Hz)
Tarcento NS Friuli (Italy), 1976 0.21 16.85 10.10
Gilroy 2 50 Coyote Lake (USA), 1979 0.20 18.00 5.00
Kalamata X Kalamata (Greece), 1986 0.24 29.75 1.63
Port Island 90 Kobe (Japan), 1995 0.28 42.00 0.91
Fig. 2. Plot of the four selected acceleration time histories scaled at 0.35 g: (a) Kalamata, (b) Gilroy 2-050, (c) Tarcento and (d) Port Island.
A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528 517
http://-/?-http://-/?-http://-/?-http://-/?- -
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
4/14
-
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
5/14
-
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
6/14
Base and lateral hydraulic boundaries were assumed as impervious
while drained condition was imposed at the top of the mesh.
Although water flow was thus allowed within the mesh, such
movement was not large enough to be detected, due to the sub-
stantially undrained condition that characterises the dynamic
analyses in relation to the earthquake duration and the assumed
low hydraulic conductivity.
The PLAXIS domain was discretised by 15-node plane strain tri-
angular finite elements, characterised by a reduced integration for-
mulation for the pore water pressures. The boundary conditions
adopted for the static stages were the same as the ones used in
SWANDYNE, while in the dynamic analyses the bottom of the
mesh was assumed to be rigid and the lateral sides were character-
ised by the viscous boundaries proposed by Lysmer and Kuhlemey-
er[37], with parametersa = 1.0 andb = 0.25.
All PLAXIS analyses were performed under undrained condi-
tions. This option, selected due to the incapability of the code to
perform fully coupled dynamic analyses, made the PLAXIS results
being consistent with those obtained by SWANDYNE, as discussed
in the following sections.
The characteristic dimension of the elements h in the SWAN-
DYNE analyses and in the central portion of the domain in the
PLAXIS analyses always satisfies the condition that the spacing of
the finite element nodes, Dlnode, must be smaller than approxi-
mately one-tenth to one-eighth of the wavelength associated with
the maximum frequency componentfmax of the input wave[38]:
Dlnode 6 kmin=810 VS;min=810fmax 4
where VS,min is the lowest wave velocity. An example of the mesh
employed in the PLAXIS analyses is sketched inFig. 6.
It is nowadays well established that the time discretisation can
play a significant role on the accuracy of dynamic finite elementFig. 8. G and D profiles assumed in the FE visco-elastic analyses on the basis of
EERA results (240-m thick deposit and Tarcento earthquake).
Fig. 9. Calibration strategies for the Rayleigh coefficients (240-m thick deposit and Tarcento earthquake).
520 A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?- -
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
7/14
-
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
8/14
5. Calibration of stiffness and viscous parameters in FE analyses
The simulation of the wave propagation problem through FE
analyses employing a linear visco-elastic model variable with
depth requires the appropriate definition of the elastic and viscous
parameters for each sub-layer of the discretised deposit. In fact, it
is well-known that the solution strongly depends on the assumed
profile of the stiffness and damping coefficients with depth.In this paper, a recently developed calibration procedure of the
visco-elastic parameters to be assumed in dynamic FE analyses is
adopted [40]. In each FE analysis, G andD profiles were defined
in order to match the ones resulting from the corresponding EERA
analyses. To this aim, the numerical models in PLAXIS and SWAN-
DYNE were subdivided into the same number of layers employed
in EERA and for each layer a value ofGand Dwas selected with ref-
erence to the shear deformation level resulting from the EERA
analyses at the corresponding depth.Fig. 8shows theGandDpro-
files adopted in the FE analysis for the same case ofFig. 7. Rayleigh
damping introduced in the simulations is defined by selecting the
coefficients aRandbR, which depend onD and on the adopted fre-quency intervalfmfn according to Eq.(3). Different possible cali-
bration procedures were proposed in the literature to identify
the interval fmfn. In particular, a well established one e.g.
[24,25] suggests to selectfmas the first natural frequency of the de-
positf1, whilefnis assumed equal to ntimesfm, where nis the clos-
est odd integer larger than the ratiofp/f1between the predominant
frequency of the input earthquake motion (fp) and the fundamental
frequency of the soil deposit (f1). This latter assumption was based
on the evidence that the higher modes of a shear beam are odd
multiples of the fundamental mode of the beam. Recently, Kwok
et al. [26] proposed to select, as a first approximation, the first
mode of the site and five times this frequency for fmandfn, respec-
tively. More generally, in order to obtain a good matching between
the linear time-domain and the frequency-domain solutions, they
suggested to identify the two frequencies through an iterative
procedure.
For the case of a 240 m thick deposit exited by the Tarcen-
to earthquake, Fig. 9a shows the amplification function of the
signal between the bedrock and the surface obtained through
the frequency-domain analysis, while the Fourier spectrum of
the input motion is reported in Fig. 9b. Assuming, for example,
a target damping ratio of 5%, the standard procedure would
lead to the selection of fm=f1= 0.29 Hz (equal to the funda-
mental frequency of the deposit, represented by the first peak
of the amplification function) and fn = 10.15 Hz, being the ratio
fp/f1 equal to 34.83 (and, therefore, n = 35). The corresponding
Rayleigh damping curve is reported in the same figure with
a solid line: it plots well below D = 5% target line. This condi-
tion leads to a significant under-damped response of the sys-
tem in the frequency range characterised by an amplification
factor larger than one, i.e. in the frequency interval in which
the site effects would be more relevant. Fig. 10a illustrates
the above issue by comparing the Fourier and response spectra
of the acceleration, as obtained at the surface level by the FE
visco-elastic analysis performed by SWANDYNE, with the corre-
sponding EERA results.
Fig. 12. Comparison between peak ground acceleration values obtained with EERA and FE visco-elastic analyses according to the two investigated calibration strategies forthe Rayleigh coefficients.
522 A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?- -
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
9/14
In order to obtain a better match between the linear time-do-
main and frequency-domain solutions, a new procedure for the
selection of the two Rayleigh frequencies is here proposed. The first
natural frequency of the system which results as significantly ex-
ited by the earthquake, fm, should be identified by comparing the
EERA amplification function and the Fourier spectrum of the input
motion. In the case ofFig. 9 is evident that the Tarcento earthquake
is characterised by a very low energy content for the first two nat-
ural frequencies of the deposit, such that the third natural fre-
quency of the system (equal to 1.12 Hz) should be selected asfm.
As regards the second frequency fn, it should be identified consid-
ering the range over which the input motion is amplified during
the propagation process: in particular,fn should be selected equal
to the frequency where the amplification function gets lower than
one. For the case described inFig. 9,the proposed procedure leads
to a value offn equal to 3.86 Hz, significantly lower than the one
obtained by the standard calibration procedure. The Rayleigh
damping curve corresponding to the new values offmandfnis plot-
ted in Fig. 9a (dashed line). The resultingaRandbRprofiles adoptedin the FE analysis are shown inFig. 11a and b, respectively, and
compared to those resulting from the standard calibration ap-
proach. Fig. 10b reports the Fourier and response spectra of the
acceleration recorded at the surface during the FE visco-elastic
analysis, performed with SWANDYNE employing the new proce-
dure: the results are in fair agreement with those obtained for
the same deposit and at the same depth by the frequency-domain
based EERA analysis. More generally, adopting the proposed proce-
dure for the definition of Rayleigh damping coefficient profiles, a
reasonably good matching between the EERA and the FE visco-
elastic analysis results was achieved at each depth for all the inves-
tigated cases, both in terms of frequency response and acceleration
time histories. A comparison between the peak ground accelera-
tions obtained with linear time-domain analyses and those result-
ing from the corresponding frequency-domain solution is reported
in Fig. 12, for all the cases analysed in this work. It can be observed
that the use of the standard procedure may lead to significant er-
rors for increasing values of the ratio fp/f1 and of the soil deposit
thickness (Fig. 12a). On the contrary, the difference between the
results obtained adopting the new procedure and the correspond-
ing EERA solutions is always lower than 10% (Fig. 12b).
Fig. 13. Results of the 1D ground response analysis performed with EERA (60-m thick deposit and Gilroy 2-050 earthquake).
Fig. 14. Shear modulus (a), damping ratio (b) and damping reduction (c) profiles for the different FE visco-elasto-plastic analyses (60-m thick deposit and Gilroy 2-050earthquake).
A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528 523
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?- -
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
10/14
Fig. 15. Comparison between response spectra obtained withEERA and the different investigated FE visco-elasto-plastic analyses at different depths (60-m thickdeposit and
Gilroy 2-050 earthquake).
524 A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528
-
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
11/14
6. Influence of plasticity
In order to investigate the effects of non-linearity on the wave
propagation process, plasticity was added to the FE visco-elastic
analyses through a non-associated visco-elasto-plastic constitutive
assumption, with a MohrCoulomb yield criterion and a null dilat-
ancy angle. In the following, the case of a 60 m thick soil deposit,
exited at the bedrock by the Gilroy 2-050 input motion, is dis-cussed in detail, as considered representative of the entire set of re-
sults obtained with the FE analyses. The frequency-domain
solution obtained by the code EERA for the selected case is summa-
rised in Fig. 13. The wave propagation from the bedrock to the sur-
face leads, in this case, to a peak ground acceleration of 0.42 g, with
a magnification factor of 1.2 over the peak base amplitude. The
amount of viscous damping resulting from the iterative equiva-
lent-linear procedure attains an average value of about 7.5%.
All the analyses performed in the time-domain were carried out
with the code SWANDYNE. The stiffness and damping profile were
selected according to the calibration procedure discussed in the
previous Section. In particular, the Rayleigh parameters assume
in this case the values offm=f1= 0.54 Hz andfn= 4.37 Hz.
Adopting the same stiffness profile resulting from the EERAanalysis (Fig. 14a), three different hypotheses concerning the
amount of viscous damping to be introduced in the non-linear
time-domain analyses were explored. In the first simulation
(named FE_vep_1), the target damping ratio at each depth of
the column was selected equal to the corresponding value ob-
tained by the EERA analysis, i.e. assuming as negligible the plas-
ticity-related hysteretic dissipation provided by the constitutive
model (Fig. 14b). In the second simulation (FE_vep_2), the
amount of viscous damping was set equal to 60% of that adoptedin the previous case (Fig. 14b). As illustrated in Fig. 14c, this im-
plies that the reduction of the target damping ratios is, in this
case, more pronounced in the upper part of the clayey deposit
as compared to the remaining portion of it. Finally, in the third
FE analysis (FE_vep_3) the EERA damping profile was reduced
at each depth by DD= 3%, resulting in the profile also shown in
Fig. 14b.
Fig. 15reports the comparison between the results of the three
FE visco-elasto-plastic analyses and the corresponding EERA simu-
lation in terms of response spectra obtained at different depths
along the deposit. All the plasticity-based analyses show a contrac-
tion of the spectra as compared to the EERA one, this being related
to the additional damping supplied by the MohrCoulomb model.
This effect is more pronounced in the uppermost portion of the de-posit, between 0 and 15 m depth, where the shear strains attain
Fig. 16. Comparison between Fourier and response spectra obtained during SWANDYNE and PLAXIS visco-elastic analyses at surface for different extensions of the mesh
(120-m thick deposit and Kalamata earthquake).
A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528 525
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?- -
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
12/14
Fig. 17. Comparison between Fourier and response spectra obtained with EERA and FE visco-elastic analyses at surface for different values of Newmark parameters (240-m
thick deposit and Tarcento earthquake).
526 A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528
-
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
13/14
their maximum values. The frequency range where this effect is
prominent is between 3.4 Hz and 5.5 Hz. The non-linearity induced
by the plasticity assumption does not significantly modify the fun-
damental modes of vibration of the soil deposit.
None of the three proposed approach for the reduction of the
viscous damping is able to balance the introduction of the hyster-
etic dissipation, at least when the results are compared to those
obtained by EERA. Concerning this latter outcome, it is worth
remarking that the EERA results might not be the right term of
comparison when strong motions induce large strain in a soil de-
posit. In this last circumstance, plasticity might prevail and bias
the picture traditionally obtained by means of visco-elastic analy-
ses. Under these latter conditions, permanent displacement and
corresponding variation of the effective stress state occur, signifi-
cantly modifying the soilstructure interaction in any geotechnical
contexte.g. [41].
7. Influence of boundary conditions and spatial discretisation
The analyses performed with the code SWANDYNE adopting the
5-m wide mesh characterised by tied-nodes boundaries (see Sec-
tions5 and 6) are representative of ideal 1D problems. For 2D and3D problems wider meshes should be employed and the hypothe-
sis of tied horizontal displacements of the lateral boundaries needs
to be abandoned. Therefore questions concerning the appropriate
lateral extension of the FE mesh arise.
A numerical investigation regarding this issue was performed
with the code PLAXIS adopting the viscous boundaries proposed
by Lysmer and Kuhlemeyer [37] and meshes characterised by
different width. The twelve visco-elastic analyses described in
Section 5 were re-simulated assuming the standard values of
the Lysmer and Kuhlemeyer parameters (a= 1.0 and b= 0.25).
The horizontal dimension of the mesh, L, was assumed equal
to 2, 4 and 8 times the thickness H. In this context, therefore,
the results obtained by the code SWANDYNE are assumed as
reference.Fig. 16shows, as an example, the comparison between Fourier
and response spectra at the surface obtained for a 120-m thick
deposit excited by the Kalamata earthquake. The similarity be-
tween the results of the PLAXIS analysis characterised by L= 8H
and the reference analysis is clearly recognizable. A satisfactory
agreement between the analyses is already attained for L= 4H.
This value can be considered as a good compromise between
accuracy and time required to perform the analysis of a 2D
boundary value problem.
The same trend was indeed observed in all the other eleven
investigated cases. In addition, no significant differences were
identified in the numerical results when adopting different values
of the Lysmer and Kuhlemeyer parameters a and b in the range
01.
8. Influence of time integration parameters
According to the Generalised Newmark time-stepping proce-
dure[31], the displacement (u) and velocity ( _u) vectors in a solid
node at timen + 1 are expressed as:
un1 un _unDt 1
21 b2un
1
2b2un1
Dt2 5
_un1 _un 1 b1un b1un1Dt 6
while the pore pressure (p) vector in a fluid node, at the same time
n+ 1, can be obtained from:
pn1 pn 1 b1_pnb1_pn1
Dt 7
The algorithm is unconditionally stable if the following condi-
tions apply:
b1 P1
2; b2 P
1
2
1
2b1
2; b
1P
1
2 8
The choice ofb1= b2= b
1= 0.5 (corresponding to the higher or-
der accurate trapezoidal scheme) guarantees the stability of the
time-stepping scheme for any value ofDt (i.e. the algorithm re-mains implicit) and does not provide any numerical (or algorith-
mic) damping during the integration of the governing equations.
In this case, numerical oscillations may occur during the analysis
if no physical (viscous or hysteretic) damping is present [12]. As
such, some numerical damping is typically introduced adopting
coefficient values larger than 0.5, consistently with condition (8).
All the time-domain simulations illustrated in this note were
performed assuming a set of Newmark parameters which leads
to a small amount of algorithmic dissipation (see Section3). To as-
sess the influence of the numerical damping on the FE results, the
case of a 240 m thick deposit exited by the Tarcento earthquake
was studied with the code SWANDYNE, varying the values of the
parameter b1in the range 0.50.9, setting b2according to condition
(8) and assuming b
1= b1.The comparison between Fourier and response spectra at the
ground surface obtained with the different Newmark parameters
and the corresponding EERA reference results is reported in Fig. 17.
The figure clearly indicates that the numerical dissipation intro-
duced by the time-stepping scheme is more pronounced at high
frequencies. The FE analysis performed with an un-damped time
integration scheme (b1 = 0.5) gives the best agreement with the
frequency-domain result in terms of peak ground acceleration,
but tends to over-predict the energy content in the range 13 Hz.
The simulation characterised by b1= 0.6 (the adopted value in
the analyses discussed in the previous Sections) represents a good
compromise between a satisfactory agreement with EERA in terms
of frequency response and a small under-estimation of the peak
ground acceleration. Increasing values of b1 induce an over-damped response, especially for the high frequency modes, leading
to significantly reduced peak ground accelerations.
9. Conclusions
This paper describes a set of 2D finite element analyses for the
simulation of the seismic ground response of a clayey deposit.
Some of the several factors potentially influencing the numerical
results are highlighted and critically discussed. In particular, the
stiffness values and the amount of viscous damping in visco-elastic
analyses, the hysteretic damping when plasticity is added to the
soil model, the spatial and time discretisation and the nature of
boundary conditions are examined. To generalise the investigation,
a parametric study was carried out using four earthquake signals,three deposits characterised by different heights, two finite ele-
ment codes and two different boundary conditions.
Most of the analyses were performed using a linear visco-elastic
soil model characterised by the Rayleigh formulation for the vis-
cous damping. The calibration of the Rayleigh coefficients as well
as the selection of the appropriate mobilised stiffness represent
critical issues for this kind of simulations. In the note the validation
of finite element visco-elastic analyses is performed comparing the
results with those obtained by equivalent-linear visco-elastic anal-
yses performed in the frequency-domain by the code EERA. Those
latter are thus takenas reference in the validation procedure. Using
this approach, the paper shows that the traditionally adopted pro-
cedures for the calibration of the Rayleigh coefficients can lead to
large overestimation of the peak ground acceleration. A novel cal-ibration procedure is here proposed and discussed: in this case the
A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528 527
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?- -
8/11/2019 [AMOROSI] Parametric Study on Seismic Ground Response by Finite Element Modelling
14/14
results of the FE analyses compare nicely with those obtained by
the frequency-domain approach.
A second set of FE analyses were carried out introducing plastic-
ity in the soil constitutive formulation. The appropriate selection of
the viscous damping to be added in the model was subjected to
further investigation. Different strategies were attempted in order
to optimise the balance between the hysteretic dissipation and the
viscous component of the damping. None of the proposed ap-
proaches allowed to achieve a good matching between the FE anal-
yses and the corresponding frequency-domain ones. Concerning
this latter outcome, it is worth remarking that the EERA results
should not be considered as the right term of comparison when
modelling strong motion earthquakes, as those selected for this
study. In fact, intense shaking results in large and partly irrevers-
ible strains associated with modification of the effective stress
state induced by excess pore pressures build-up. Those features
cannot be accounted for by visco-elasticity based constitutive laws,
as that adopted in EERA, making the plasticity-based time domain
approach more realistic.
The simulations were performed by the finite element code
SWANDYNE, adopting a 5-m wide mesh characterised by tied
nodes at the lateral boundaries, thus limiting the case to the 1D
condition. The possibility of performing 2D finite element simula-
tions was investigated by re-running the numerical analyses with
the finite element code PLAXIS, adopting the Lysmer and Kuhle-
meyer conditions at the lateral boundaries. The match between
the results of the two different geometrical configurations as-
sumed in the two codes were obtained employing 2D meshes char-
acterised by a width-height ratio larger than eight, while
satisfactory results were already achieved for a ratio equal to four.
No influence of the values of the Lysmer and Kuhlemeyer coeffi-
cients was observed in the 2D analyses.
Finally, accuracy and damping characteristics of the time inte-
gration algorithm were analysed. It was found that the standard
values of the time-stepping coefficients for the Generalised New-
mark scheme represent the best compromise to obtain satisfactory
results both in terms of frequency content and peak groundacceleration.
Acknowledgements
The Authors gratefully acknowledge the financial support of the
Italian Ministry of Instruction, University and Research (Grants:
PRIN 2007 Seismic response of slopes, excavations and tunnels
and PRIN 2008 Design of underground constructions in seismic
conditions) and the ReLUIS (Italian University Network of Seismic
Engineering Laboratories) network. The employed accelerograms
were extracted from the SISMA (Site of the Italian Strong Motion
Accelerograms) website.
References
[1] Schnabel PB, Lysmer J, Seed HB. SHAKE: a computer program for earthquake
response analysis of horizontally layered sites. Report no EERC72-12,
Earthquake Engineering Research Center, University of California, Berkeley;
1972.
[2] Idriss IM, Lysmer J, Hwang R, Seed HB. QUAD-4: a computer program for
evaluating the seismic response of soil structures by variable damping finite
element procedures. Report no EERC 73-16, Earthquake Engineering Research
Center, University of California, Berkeley; 1973.
[3] Idriss IM, Sun JI. SHAKE91: a computer program for conducting equivalent
linear seismic response analyses of horizontally layered soils deposits. Center
for Geotechnical Modeling, University of California, Davis; 1992.
[4] Biot MA. General theory of three-dimensional consolidation. J Appl Phys
1941;12:15564.
[5] Sangrey DA, Henkel DJ, Esrig MI. The effective stress response of a saturated
clay soil to repeated loading. Can Geotech J 1969;6(3):24152.
[6] Hardin B, Drnevich VP. Shear modulus and damping in soils: measurements
and parameter effects. J Soil Mech Div (ASCE) 1972;98:60324.
[7] Castro G, Christian JT. Shear strength of soils and cyclic loading. J Geotech Eng
Div (ASCE) 1976;102(GT9):88794.
[8] Sagaseta C, Cuellar V, Pastor M. Cyclic loading. In: Proceedings of the 10th
European conference on soil mechanics and foundation engineering, Florence,
Italy; 1991. p. 98199.
[9] Vucetic M, Dobry R. Effects of the soil plasticity on cyclic response. J Geotech
Eng Div (ASCE) 1991;117(1):89107.
[10] Geli L, Bard PY, Jullien B. The effect of topography on earthquake ground
motion: a review and new results. Bull Seismol Soc Am 1988;78(1):4263.
[11] Semblat JF, Pecker A. Waves and vibrations in soils: earthquakes, traffic,
shocks, construction works. Pavia: IUSS Press; 2009.[12] Zienkiewicz OC, Chan AHC, Pastor M, Schrefler BA, Shiomi T. Computational
geomechanics (with special reference to earthquake engineering). Chichester:
Wiley & Sons; 1999.
[13] Arulanandan K, Scott RF. In: Proceedings of VELACS symposium. Rotterdam:
AA Balkema, 1993.
[14] Dewoolkar MM, Ko HY, Pak RYS. Seismic behaviour of cantilever retaining
walls with liquefiable backfills. J Geotech Geoenviron Eng 2001;127(5):
42435.
[15] Elgamal AW, Parra E, Yang Z, Adalier K. Numerical analysis of embankment
foundation liquefaction countermeasures. J Earthquake Eng 2002;6(4):
44771.
[16] Aydingun O, Adalier K. Numerical analysis of seismically induced liquefaction
in earth embankment foundations. Part I. Benchmark model. Can Geotech J
2003;40(4):75365.
[17] Muraleetharan KK, Deshpande S, Adalier K. Dynamic deformations in sand
embankments: centrifuge modelling and blind, fully coupled analyses. Can
Geotech J 2004;41(1):4869.
[18] Dakoulas P, Gazetas G. Seismic effective-stress analysis of caisson quay walls:
application to Kobe. Soils Found 2005;45(4):13347.
[19] Madabhushi SPG, Zeng X. Simulating seismic response of cantilever retaining
walls. J Geotech Geoenviron Eng 2007;133(5):53949.
[20] Amorosi A, Elia G. Analisi dinamica accoppiata della diga Marana Capacciotti.
Ital Geotech J 2008;4:7896.
[21] Sica S, Pagano L, Modaressi A. Influence of past loading history on the seismic
response of earth dams. Comput Geotech 2008;35(1):6185.
[22] Rayleigh L. Theory of sound, vol. 2. New York: Dover; 1945.
[23] Woodward PK, Griffiths DV. Influence of viscous damping in the dynamic
analysis of an earth dam using simple constitutive models. Comput Geotech
1996;19(3):24563.
[24] Hudson M, Idriss IM, Beikae M. Users manual for QUAD4M. Center for
Geotechnical Modeling, University of California, Davis; 1994.
[25] Hashash YMA, Park D. Viscous damping formulation and high frequency
motion propagation in nonlinear site response analysis. Soil Dyn Earthq Eng
2002;22(7):61124.
[26] Kwok AOL, Stewart JP,Hashash YMA, Matasovic N, Pyke R, Wang Z, et al. Useof
exact solutions of wave propagation problems to guide implementation of
nonlinear seismic ground response analysis procedures. J Geotech GeoenvironEng 2007;133(11):138598.
[27] Viggiani GMB, Atkinson HJ. Stiffness of fine-grained soils at very small strains.
Gotechnique 1995;45(2):24965.
[28] Bardet JP, Ichii K, Lin CH. EERA a computer program for equivalent-linear
earthquake site response analyses of layered soils deposits. User manual;
2000.
[29] Chan AHC. User manual for DIANA-SWANDYNE II. School of Engineering,
University of Birmingham, Birmingham; 1995.
[30] PLAXIS 2D. Reference manual, version 8; 2003.
[31] Katona MC, Zienkiewicz OC. A unified set of single step algorithms. III. The
beta-m method, a generalization of the Newmark scheme. Int J Numer
Methods Eng 1985;21(7):134559.
[32] Li XS, Shen CK, Wang ZL. Fully coupled inelasticsite responseanalysis for 1986
Lotung earthquake. J Geotech Geoenviron Eng 1998;124(7):56073.
[33] Borja RI, Chao HY, Montns FJ, Lin CH. Nonlinear ground response at Lotung
LSST site. J Geotech Geoenviron Eng 1999;125(3):18797.
[34] Elia G, Amorosi A, Chan AHC. Non-linear ground response: effective stress
analyses and parametric studies. In: Proceedings of the Japan-Europe seismic
risk workshop, Bristol, UK; 2004.[35] Arslan H, Siyahi B. A comparative study on linear and nonlinear site response
analysis. Environ Geol 2006;50:1193200.
[36] Clough R, Penzien J. Dynamics of structures. Berkeley: Computers and
Structures Inc.; 2003.
[37] Lysmer J, Kuhlemeyer RL. Finite dynamic model for infinite media. ASCE EM
1969;90:85977.
[38] Bathe KJ. Finite element procedures in engineering analysis. Upper Saddle
River, NJ: Prentice Hall; 1982.
[39] Haigh SK, Ghosh B, Madabhushi SPG. Importance of time step discretisation
for nonlinear dynamic finite element analysis. Can Geotech J 2005;42:
95763.
[40] Amorosi A, Boldini D, Sasso M. Modellazione numerica del comportamento
dinamico di gallerie superficiali in terreni argillosi. Asterisco, Bologna; 2008.
[published on-line in Alma Mater Digital Library ].
[41] Amorosi A, Boldini D. Numerical modelling of the transverse dynamic
behaviour of circular tunnels in clayey soils. Soil Dyn Earthq Eng
2009;29(6):105972.
528 A. Amorosi et al. / Computers and Geotechnics 37 (2010) 515528
http://amsacta.cib.unibo.it/archive/00002392/http://amsacta.cib.unibo.it/archive/00002392/http://amsacta.cib.unibo.it/archive/00002392/http://amsacta.cib.unibo.it/archive/00002392/