Validation of A Three-Dimensional Constitutive Model For Nonlinear Site Response and Soil-Structure Interaction Analyses Using Centrifuge Test Data
Validation of A Three-Dimensional Constitutive Model For Nonlinear Site Response and Soil-Structure Interaction Analyses Using Centrifuge Test Data
Validation of A Three-Dimensional Constitutive Model For Nonlinear Site Response and Soil-Structure Interaction Analyses Using Centrifuge Test Data
DOI: 10.1002/nag.2702
RESEARCH ARTICLE
KEYWORDS
buried structures, centrifuge testing, multiaxial soil model, nonlinear site response, soil-structure interac-
tion, soil plasticity
1 INTRODUCTION
Accurate representation of soil behavior under cyclic/dynamic loading is a key capability in most geotechnical earthquake
engineering problems, especially for site response and soil-structure interaction (SSI) analyses. In general, either “equivalent
linear” or fully nonlinear modeling approaches are commonly used in such analyses. The equivalent linear method1 —in which
horizontal shear strain is used as the driving variable that defines the soil's nonlinear shear stress behavior—is more commonly
used because of its simplicity and low computational cost, especially in engineering practice. However, it is recognized that for
Int J Numer Anal Meth Geomech. 2017;1–20. wileyonlinelibrary.com/journal/nag Copyright © 2017 John Wiley & Sons, Ltd. 1
2 ZHANG ET AL.
sites with soft soils, or sites under strong seismic motions, the equivalent linear method offers only limited success in capturing
the real-life behavior.
During the past few decades, a broad range of nonlinear soil models—uniaxial to multiaxial, simplified to advanced—have
been devised.2-6 An example of advanced nonlinear soil models is the multisurface constitutive model devised by Elgamal and
coworkers4 * —aka, the pressure-dependent multi-yield (PDMY) model—which is frequently used in direct simulation of SSI
problems within the research community. The main advantage of this model is that its many hierarchical yield surfaces enable
it to approximate soil behavior within a broad range of strain regimes including post-liquefaction, but this is also a disadvantage
in that the large number of requisite model parameters renders the calibration process formidable.8
A model with a simpler scaffold is that by Borja and Amies,3 which was later extended by Chao and Borja.9 This is also a
multisurface model, but it has only a bounding surface—which can translate in stress-space through the extension proposed
in Chao and Borja9 —in addition to a vanishing elastic region. Incidentally, it needs just a few parameters for calibration. The
validity of this model was previously examined by using the downhole array motions recorded at Lotung, Taiwan, through
one-dimensional nonlinear site response analyses.10 Its sensitivities to statistical parameter variations were also studied,11 and
it was found to offer superior performance when compared with the equivalent linear method.
Because of its mathematical/thermodynamical consistency and relative simplicity, Borja and coworkers' model holds great
promise in capturing the multiaxial behavior of soil deposits in general wave propagation problems. While soil behavior in
one-dimensional settings—especially for for vertically propagating horizontally polarized shear waves—have been well under-
stood, and numerous models have been devised and validated (see, for example, Vucetic and Dobry2 ), models that can capture
soil behavior in more general settings are needed to extend the present capabilities in SSI analyses beyond this simplest config-
uration (for example, to consider inclined waves from a distant source, surface waves, waves emanating from a buried scatterer,
and waves propagating in non-horizontally layered media).
In the present study, we explore the capabilities of Borja and coworkers' model3,9,12 in capturing the dynamic responses
measured during a centrifuge test, wherein multiaxial stress conditions were generated because of the presence of an embedded
scatterer. These tests were conducted by Hushmand et al13 at the University of Colorado Boulder to investigate the seismic
performance of relatively stiff structures buried in dry sand.
In what follows, we first derive the consistent tangent operator of Borja's model for the multiaxial case and implement it as
a user-defined material (UMAT) subroutine in ABAQUS.14 For verification, we make comparisons of one-dimensional (1D)
wave propagation analysis results with those obtained using with DEEPSOIL,15 which is a site-response analysis software that
features well-accepted 1D model(s) of the hysteretic behavior of well-confined soils under low-to-moderate (shear) strains (see,
for example, Vucetic and Matasovic16 ). Next, we explore the capability of this soil model in predicting responses measured
during the tests by Hushmand et al.13 In this, we also make comparisons with numerical results obtained by Deng et al17 and
Esmaeilzadeh Seylabi,18 who used the PDMY4 and equivalent linear soil models, respectively.
2 A N O V E R V I E W O F B O R J A' S S O I L M O D E L
2.1 Formulation
It is well known that soil nonlinearity comes into effect even at very small strain levels19 and that omission of such nonlinearities
may result in significant errors in free-field (and thus, foundation input) motions in SSI analyses.2 In the present study, we
implement and validate a multiaxial plastic with viscous damping soil model with a vanishing elastic region and coupled fully
nonlinear behavior.12 This constitutive model is a minor extension of the model proposed by Borja and Amies3 —in that, it
features a material-point-level stiffness-proportional viscous damping. A summary formulation of this model, which include its
fourth-order consistent tangent material tensor, is presented next.
The total stress tensor 𝝈 of the model consists of 2 major—namely, the inviscid (𝝈 inv ) and the viscous (𝝈 vis )—parts, as
given by
where
𝝈 inv = Ce ∶ (𝝐 − 𝝐 p ),
(2)
̇
𝝈 vis = D ∶ 𝝐,
and Ce and D are elastic stiffness and viscous damping tensors, respectively; 𝝐 is the total strain tensor; 𝝐 p is the plastic strain
tensor, and 𝝐̇ is the total strain rate. In this study, a linear stiffness-proportional damping is adopted,12 which can be devised by
defining D as
2𝜉0 e
D= C, (3)
𝜔0
where 𝜔0 is the frequency at which the small strain damping ratio is equal to 𝜉 0 and the term 𝜔0 can be calibrated to match the
dominant frequency of the input motion.
To achieve an optimal rate of convergence for Newton's method, the consistent tangent moduli are required.20 For their model,
Borja et al21 derived this fourth-order tensor as
d𝝈 inv
n+1 𝜕𝜓
Cinv
ep = = K1 ⊗ 1 + 𝜓Idev + ⊗ Δ𝝐 ′ , (4)
d𝝐 n+1 𝜕𝝐 n+1
where Idev = I − 13 1 ⊗ 1 is the deviatoric identity tensor and K is the bulk modulus. The parameter 𝜓 is defined through the
equation Δ𝝈 = 𝜓Δ𝝐 , where Δ𝝈 and Δ𝝐 denote the deviatoric stress and strain increments, respectively. Further details of
′ ′ ′ ′
derivation are omitted here for brevity and may be found in other works.3,12,21
As seen, the third term on the right-hand side of Equation 4 renders Cinv
ep nonsymmetric in general. However, as demonstrated
by Borja and Wu,21 the symmetric part of this consistent tangent stiffness tensor is often efficient enough to produce accurate
solutions at superlinear convergence rates. Incidentally, a symmetric tangent also facilitates significant savings in memory
requirements as well as in flops for solving the system-level linear equations.
After incorporating the viscous stress and by using the forward-difference method to approximate the total strain rate, we can
derive from Equations 2 and 3
2𝜉0 e 1 2𝜉0 e
𝝈 vis vis
n+1 = Cep ∶ d𝝐 n+1 = C ∶ 𝝐̇ n+1 = C ∶ d𝝐 n+1 (5)
𝜔0 dt 𝜔0
which yields
1 2𝜉0 e
vis
Cep = C. (6)
dt 𝜔0
By only retaining the symmetric part of the consistent tangent of Borja's model, we obtain the total consistent tangent stiffness
moduli as
1 2𝜉0 e
inv
Cep = Cep,symm vis
+ Cep = K1 ⊗ 1 + 𝜓Idev + C, (7)
dt 𝜔0
[ ( √ )m ]−1
G 3
2𝜏0
R∕ 2 + 𝜏0 − 𝜏
=1− h + H0 d𝜏, (8)
Gmax 2𝛾0 ∫0 𝜏
where G = 𝜏 0 ∕𝛾 0 is the secant shear stiffness and R is the radius of the bounding surface. Parameters h, m, and H0 control the
intensity of the hardening. Given this equation, as will be demonstrated subsequently in detail, it is possible to use a nonlinear
solution method—here, Broyden's22 —and an appropriate initial guess, to calibrate the model parameters h and m from 2 points
on the G∕Gmax curve. Alternatively, a nonlinear least-squares regression can be used to determine the optimal values for h, m,
and H0 , if more than 2 points are selected. Here, we used the Matlab23 built-in function lsqnonlin for both cases to obtain the
optimal model parameters. Moreover, to facilitate broad use, we implemented all variants of the model—namely, with and with-
out viscous damping, with symmetric and nonsymmetric tangents, and for plane-strain, axisymmetric, and three-dimensional
cases—in the commonly used commercial finite element analysis software ABAQUS,14 through its UMAT subroutine interface.
3 M O D E L VE R I F I C A T I O N
Here, the basic model is verified, first by examining its output and parameter sensitivities under numerical simple shear tests,
and next by comparing its output for a site-response analysis with that from DEEPSOIL,15 which is a tool specifically designed
for (and limited to) one-dimensional shear wave propagation.
FIGURE 1 A, Hysteresis loops; B, normalized shear modulus degradation; and C, damping ratio curves for
Gmax = 20 MPa, 𝜈 = 0.3, h = Gmax , m = 1, R = 200 kPa, H0 = 0, 𝜔0 = 1 rad/s, and 𝜉 0 = [0, 0.005, 0.01, 0.03, 0.05] [Colour figure can be viewed at
wileyonlinelibrary.com]
ZHANG ET AL. 5
FIGURE 2 1D site response analysis A, acceleration history and B, 5%-damped spectral acceleration subjected to the Ricker wavelet input
[Colour figure can be viewed at wileyonlinelibrary.com]
FIGURE 3 1D site response analysis A, acceleration history and B, 5%-damped spectral acceleration subjected to the earthquake input [Colour
figure can be viewed at wileyonlinelibrary.com]
6 ZHANG ET AL.
indicate, although the nonsymmetric moduli has a higher rate of convergence, the symmetric one still converges successfully,
and its rate of convergence is acceptable. Moreover, the use of viscous part improves the rate of convergence, possibly because
the stress-strain curves are becoming smoother with increased damping.
Hushmand et al13 conducted a series of centrifuge experiments at the University of Colorado Boulder to investigate
the seismic performance of relatively stiff structures buried in dry sand. Three different box-shaped specimens were
designed to represent the characteristics of prototype reinforced concrete reservoir structures with varying stiffnesses. Aspect
ratios and dimensions of these structures, as well as their stiff roofs, restrained excessive rotational movements and pro-
duced significant seismic pressures (and bending strains) on the walls. Three distinct ground motions were applied to
each structure.
Figure 4 shows the centrifuge test layout and instrumentation. The dimensions and properties of the model structures
used for the experiments are provided in Table 2. The material properties of the steel structures were chosen as follows:
density, 𝜌 = 7870 kg/m3 ; Young's modulus, E = 200 GPa; and Poisson's ratio, 𝜈 = 0.29. The 5%-damped spec-
FIGURE 4 Layout and instrumentation of centrifuge experiments in prototype scale [Colour figure can be viewed at wileyonlinelibrary.com]
ZHANG ET AL. 7
tral accelerations and Arias intensity time histories as recorded in the centrifuge test on the flexible structure are shown
in Figure 5.
Deng et al17 explored the capability of the so-called PDMY model, which is a pressure-dependent multi-yield-surface plas-
ticity model, in predicting the response of the specimen structures and the far-field soil. The PDMY model has 20 material
parameters, great majority of which are not directly calibrated but have judiciously chosen “assumed” values. Using the same
test data, Esmaeilzadeh Seylabi explored the range of applicability of calibrated equivalent linear soil models in capturing the
response of the tested structures. Here, we investigate the capability of the implemented nonlinear soil model in predicting the
response of the tested structures and also make comparisons to results obtained with PDMY model by Deng et al17 and with
equivalent linear models by Esmaeilzadeh Seylabi.18
Two-dimensional meshes are used for discretizing tests on the flexible and stiff specimens, and each mesh comprised
18 soil layers with a uniform element size of 0.25 m to capture the soil heterogeneity. It should be noted that the ele-
ment size is selected small enough to resolve wave propagation in the frequency range of interest. To set up the ini-
tial stress condition appropriately, a static analysis under gravity loading is performed prior to each dynamic analysis.
FIGURE 5 A, 5%-damped spectral accelerations and B, Arias intensity time histories of the container base motions recorded for the test on
flexible structure [Colour figure can be viewed at wileyonlinelibrary.com]
FIGURE 6 A, Shear wave velocity and B, stiffness degradation curves [Colour figure can be viewed at wileyonlinelibrary.com]
8 ZHANG ET AL.
During the static analyses, horizontal degrees of freedom (DOFs) at the left and right vertical edges of the domain are
fixed, and the vertical DOFs are left free. For the dynamic analyses, the horizontal DOFs at the said edges are slaved to
each other at every elevation—per the boundary condition imposed by the centrifuge's container—and the vertical DOFs are
considered free.
FIGURE 7 The 5%-damped spectral acceleration at far-field (A2, A3, and A4) and on structure (A12, A13, and A14) for T-Flexible-AL,
Stiff-AL, Flexible-AH, and Stiff-AH obtained numerically and experimentally [Colour figure can be viewed at wileyonlinelibrary.com]
ZHANG ET AL. 9
FIGURE 8 Arias intensity at far-field (A1, A2, A3, and A4) for T-Flexible-AL, Stiff-AL, Flexible-AH, and Stiff-AH obtained numerically and
experimentally [Colour figure can be viewed at wileyonlinelibrary.com]
of approximately 60%, could be achieved. The small-strain shear wave velocity profile of the soil deposit is predicted using the
equation proposed by Bardet et al25
(a − e)2 n
Gmax = A p , (9)
1+e
where A = 8.811, a = 1.935, and n = 0.5 are the 3 constants determined for the Nevada sand.26 The parameters e and p are,
respectively, the void ratio and the mean pressure expressed in kPa. Figure 6A displays the resulting maximum shear wave
velocity profile.
It can be shown that the model by Borja et al9 implies the following relationship between normalized shear modulus, the shear
strain, and other material parameters:
ZHANG ET AL. 11
FIGURE 9 Time-frequency distribution of the energy density of acceleration time series for T-Flexible AH test using experimental and
numerical data [Colour figure can be viewed at wileyonlinelibrary.com]
[ ( √ )m ]−1
G 3
2G𝛾
R∕ 2 + G𝛾 − 𝜏
+ h + H0 d𝜏 − 1 = 0. (10)
Gmax 2𝛾 ∫0 𝜏
The secant stiffness tends to be zero when the amplitude of shear strain is large, which yields H0 = 0. The radius of the bounding
surface can also be simply computed as
√ √
R= 2𝜏max ≈ 2G 𝛾max=5% 𝛾max=5% ≈ 0.003 Gmax . (11)
As such, the material parameter calibration procedure involves the estimation/calibration of 2 parameters only—namely, h and
m in Equation 10.
Here, we use the shear modulus degradation curves proposed by Darendeli27 and pick 2 representative points on that curve
for (𝛾, G∕Gmax ) as (0.003, 0.192) and (1.023E-4, 0.845). Plugging these choices in Equation 10 yields 2 nonlinear equations
in parameters h and m. Using Broyden's method28 eliminates the need to compute an analytical Jacobian for iterative solu-
tion procedure to obtain h and m. Using initial guesses of h0 = 0.1 Gmax and m0 = 1.0 yields the materials parameter as
h = 0.1363 Gmax and m = 1.5477, with a resulting L2 residual norm of 1 × 10−16 .
It is also possible to obtain the optimal h, m, and H0 values when more than 2 experimental data points from the (𝛾, G∕Gmax )
curve are provided. As this would then produce an overdetermined system, nonlinear least-squares procedures are needed to
obtain the optimal material parameter values. One such procedure for the present model is implemented in Matlab by using its
built-in function sqnonlin.
FIGURE 10 Dynamic bending strains and their corresponding Fourier amplitude spectra of the sensor SG8 at the bottom of the north wall
obtained numerically and experimentally [Colour figure can be viewed at wileyonlinelibrary.com]
To examine the performance of the implemented nonlinear soil model, we present the measured and predicted responses for
the centrifuge experiments in terms of accelerations, specimen racking displacements, bending strain, and lateral earth pressure
along the specimen's walls, as well as soil surface settlement. Also, back-calculated stress-strain relationships and the associated
effective stiffness in all 4 tests are presented.
As mentioned above, numerical results are compared with not only the experimental data but also with the numerical results
from Deng et al17 and Esmaeilzadeh Seylabi,18 who, respectively, used the PDMY and equivalent linear soil models (referred
to as EL below). It is important to note that in their analyses, Deng et al used 3 sets of material parameters that were calibrated
from three different G∕Gmax curves to achieve better agreement in a variety of tests. Similarly, Esmaeilzadeh Seylabi used an
optimization-based method to calibrate the shear wave velocity profile and the Rayleigh damping model of a soil column using
the available far-field acceleration data for each test. In the present study, we only use one set of material parameters, which are
calibrated from a single G∕Gmax curve as described in the previous section.
4.2.1 Acceleration
Figure 7 displays the 5%-damped spectral accelerations at different locations for all of the studied test cases. Under the rela-
tively low-amplitude input motion Northridge-L, the equivalent linear (EL) model yields satisfying results at all locations at
ZHANG ET AL. 13
FIGURE 11 The distribution of maximum bending strains along the north wall for T-Flexible-AL, Stiff-AL, Flexible-AH, and Stiff-AH obtained
numerically and experimentally [Colour figure can be viewed at wileyonlinelibrary.com]
FIGURE 12 The distribution of the maximum dynamic and total lateral earth pressure profiles for T-Flexible-AL, Stiff-AL, Flexible-AH, and
Stiff-AH obtained numerically and experimentally [Colour figure can be viewed at wileyonlinelibrary.com]
lower frequencies. However, a better agreement is achieved with experimental data at higher frequencies using the implemented
nonlinear soil model. This observation reveals the nonlinear model's advantages over the equivalent linear model, which expe-
riences high damping at higher frequencies. Under the high-amplitude input motion Northridge-H, where the soil nonlinearity
is more prominent, the EL model always overestimates the spectral accelerations, even at low frequencies. For the nonlinear
model, on the other hand, predictions match the measurements well at deeper locations (this trend is generally true, in fact, for
all models, because the input motions at the bottom of the container are known and prescribed in all of the numerical simula-
tions). However, as we get closer to the soil surface, discrepancies increase and the nonlinear models (NL, NLV) underestimate
the responses.
Figure 8 displays the Arias intensity time series at the far-field (see Figure 4). As seen in Figures 7 and 8, the NLV performs
better than NL. The presently used stiffness-proportional viscous damping in the NLV model increases linearly with frequency,
and thus, it appears possible to improve the NLV's agreement with measurements even further, especially at higher frequencies,
by using a more sophisticated viscous part—a task deferred to a future study.
Figure 9 displays the time-frequency distributions of the signal energy density29 for acceleration time series of T-Flexible-AH
test (for which we expect significant nonlinear behavior) at locations A4 and A14 (cf Figure 4). As seen, both the NLV and
EL models are generally capable of capturing the general patterns of experimental data, and as expected, the NLV model
outperforms EL model at higher frequencies. On the other hand, the time-frequency distributions resulting from the NL model
have spurious energy content at higher frequencies, which are nonexistent in the experimental data, especially towards the end
14 ZHANG ET AL.
FIGURE 13 Racking profiles for T-Flexible-AL, Stiff-AL, Flexible-AH, and Stiff-AH obtained numerically and experimentally [Colour figure
can be viewed at wileyonlinelibrary.com]
of the signal. This ascertains the discrepancies observed in the 5% damped spectral acceleration and Arias Intensity results
presented for the NL model in Figures 7 and 8, respectively.
†
In their study, Deng et al17 did not present the Fourier amplitude spectra of their results for the PDMY model. These calculations are made here using their
reported time series results. Also noted here is that a low-pass Butterworth filter is used to reduce the noise in strain data measured for the stiff structure.
ZHANG ET AL. 15
FIGURE 14 Numerically predicted stress-strain curves for T-Flexible-AL, Stiff-AL, Flexible-AH, and Stiff-AH [Colour figure can be viewed at
wileyonlinelibrary.com]
the soil-structure interface sliding on the distribution of the lateral earth pressure on the structures. The friction coefficient of
the interface is computed using,
tan(𝜙interface ) = 0.7 tan(𝜙soil ) ≈ 0.33, (12)
which agrees with typical values used in engineering practice (see, eg, one study31 ) and was also used by Deng et al.17 Figure 12
displays the distribution of the maximum dynamic (ie, total minus static) and total lateral earth pressures along the north walls
of the specimens. As seen, the NLV and contact models are successful in capturing the experimentally measured lateral earth
pressures,‡ especially for the stiff specimen and strong input motions for which the EL model has a poor performance. The
‡
In general, earth pressures in dynamic centrifuge experiments are difficult to measure reliably because of sensor limitations. Therefore, a higher degree of
uncertainty should be expected in “measured” earth pressures than, for example, direct strain measurements.
16 ZHANG ET AL.
FIGURE 15 Comparisons of shear wave velocity profiles for T-Flexible-AL, Stiff-AL, Flexible-AH, and Stiff-AH obtained from nonlinear and
equivalent linear method [Colour figure can be viewed at wileyonlinelibrary.com]
NLV model also outperforms the PDMY model used by Deng et al,17 who only reported results for low-intensity (Northridge-L)
base motions.
4.2.3 Racking
The racking deformations—ie, the relative displacement between the roof and base—are another critical parameter in the design
procedures for buried structures. Experimental racking deformations are obtained by double-integrating the accelerations at
locations A14 and A12, which are then subtracted from each other (ie, D12−D14). Figure 13 displays the racking deformations
for all the tests. As seen, the numerical and experimental results are in good agreement in all the tests, and NLV displays
significantly better performance than PDMY under high-amplitude base motions.
FIGURE 16 Measured (experimental) and predicted (numerical) surface settlements at the sensor locations LVDT3 (structure) and LVDT6
(free-field) [Colour figure can be viewed at wileyonlinelibrary.com]
where X refers to a given response parameter of interest. Figure 17 shows the range of residuals and variances for each analysis
type and response parameter—namely, PGA profiles for the far-field and the structural walls, surface response spectra, surface
far-field Arias intensity, racking displacement, bending strain and pressure profiles of structural walls, and the amplitude of
surface settlement. As these results indicate, the EL model exhibits satisfactory performance, especially for the low-amplitude
18 ZHANG ET AL.
FIGURE 17 The range of A, residuals and B, variances for each analysis type and response parameters [Colour figure can be viewed at
wileyonlinelibrary.com]
motions, with residuals ranging from about −0.22 to 0.31, if we exclude the pressure residuals (as pressure transducers have yet
unknown reliability). The NLV is superior with residuals ranging from about −0.16 to 0.2, if we exclude the pressure and set-
tlement residuals (as full settlement data is only available for NLV). This performance of NLV is especially impressive, because
unlike EL, the NLV model is calibrated from a single material dataset, whereas the EL model was specifically calibrated 18 to
match—in a weighted least-squares sense—soil behavior in seperate (low- and high-amplitude input) tests.
5 S U M M A R Y A N D C O N C L U S IONS
In this study, we implemented a multiaxial soil constitutive model originally developed by Borja3,12 in ABAQUS. The expression
of its consistent tangent stiffness moduli, when a viscous damping is applied to the soil model, is derived and tested during simple
shear tests with strain-controlled harmonic loading. Furthermore, 1D site response analyses are conducted; the model is verified
by comparing the results with the well-known site-response analysis tool DEEPSOIL. Validation is achieved by calibrating
the material parameters using a theoretical material response curve and by making blind comparisons with measurements
made in several centrifuge tests on embedded structures, including structural strains and deformations, lateral earth pressures,
accelerations, and surface settlements. These results indicated that the model can accurately predict inelastic soil responses
in a plane-strain setting. Comparisons made with results obtained from equivalent linear models and a pressure-dependent
multi-yield surface model suggested that the implemented model is generally superior to them in predicting responses over a
broad range of input frequencies.
ZHANG ET AL. 19
It should be noted here that there are other advanced models that are appropriate for representing multiaxial responses of
well-confined soils. These include models that are different than both the Borja-Amies and PDMY models—that is, they are
either multisurface models (like the PDMY model) but are purely deviatoric (eg, Montáns33 and Caminero and Montáns34 ), or
they are single-surface models with vanishing elastic regions (like the Borja-Amies model) but have volumetric components.
(eg, Pisanò and Jeremić6 ). It appears that a broader comparison of this variety of constitutive models is necessary, and such a
comparison would best be made by using experimental data from dynamic tests on soil-only specimens.
ACKNOWLEDGMENTS
The work presented here was funded by the California Department of Transportation (Caltrans), the United States. (grant no.
65A0561). Any opinions, findings, and conclusions or recommendations expressed in this paper are those of the authors and
do not necessarily reflect the views of the Caltrans.
REFERENCES
1. Seed HB, Idriss IM. Soil moduli and damping factors for dynamic response analysis. Report No. EERC 70 -10, University of California, Berkeley,
CA. 1970.
2. Vucetic M, Dobry R. Effect of soil plasticity on cyclic response. J Geotech Eng. 1991;117(1):89-107.
3. Borja RI, Amies AP. Multiaxial cyclic plasticity model for clays. J Geotech Eng. 1994;120(6):1051-1070.
4. Yang Z, Elgamal A, Parra E. Computational model for cyclic mobility and associated shear deformation. J Geotech Geoenviron Eng.
2003;129(12):1119-1127.
5. Manzari MT, Dafalias YF. A critical state two-surface plasticity model for sands. Géotech. 1997;47(2):255-272.
6. Pisanò F, Jeremić B. Simulating stiffness degradation and damping in soils via a simple visco-elastic–plastic model. Soil Dyn Earthquake Eng.
2014;63:98-109.
7. Prévost JH. A simple plasticity theory for frictional cohesionless soils. Soil Dyn Earthquake Eng. 1985;41:9-17.
8. Karimi Z, Dashti S. Numerical and centrifuge modeling of seismic soil-foundation-structure interaction on liquefiable ground. J Geotech
Geoenviron Eng. 2016;142(1):04015061-1–14.
9. Chao H-Y, Borja RI. Nonlinear dynamic soil-structure interaction analysis and application to Lotung problem, Tech. Rep. J. A. Blume Earthquake
Engrg. Ctr. Tech. Rep. 129, Stanford, CA, Stanford University; 1998.
10. Borja RI, Chao H-Y, Montáns FJ, Lin C-H. Nonlinear ground response at Lotung LSST site. J Geotech Geoenviron Eng. 1999;125(3):187-197.
11. Andrade JE, Borja RI. Quantifying sensitivity of local site response models to statistical variations in soil properties. Acta Geotech.
2006;1(1):3-14.
12. Borja RI, Lin C-H, Sama KM, Masada GM. Modeling nonlinear ground response of non-liquifiable soils. Earthquake Eng Struct Dyn.
2000;29:63-83.
13. Hushmand A, Dashti S, Davis C, et al. Seismic performance of underground reservoir structures: insight from centrifuge modeling on the influence
of structure stiffness. J Geotech Geoenviron Eng. 2016:04016020.
14. Hibbit Karlsson&Sorensen. ABAQUS/Standard Analysis User's Manual. USA: Hibbitt, Karlsson, Sorensen Inc.; 2007.
15. Hashash YMA, Musgrove MI, Harmon JA, Groholski D, Phillips CA, Park D. Deepsoil v6.1, user manual, Tech. rep., Urbana, IL, University of
Illinois at Urbana-Champaign; 2016.
16. Vucetic M, Matasovic N. Cyclic characterization of liquefiable sands. J Geotech Eng. 1993;119(11):1805-1822.
17. Deng YH, Dashti S, Hushmand A, Davis C, Hushmand B. Seismic response of underground reservoir structures in sand: evaluation of class-c
and c1 numerical simulations using centrifuge experiments. Soil Dyn Earthquake Eng. 2016;85:202-216.
18. Esmaeilzadeh Seylabi E. Reduced order modeling of soil structure interaction problems, Ph.D. thesis. University of California, Los Angeles.
2016.
19. Dafalias YF, Popov EP. Cyclic loading for materials with a vanishing elastic region. Nucl Eng Des. 1977;41(2):293-302.
20. Simo J, Hughes T. Computational Inelasticity, Interdisciplinary Applied Mathematics. New York: Springer; 2000. https://books.google.com/
books?id=ftL2AJL8OPYC.
21. Borja RI, Wu W-H. Vibration of foundations on incompressible soils with no elastic region. J Geotech Eng. 1994;120(9):1570-1592.
22. Broyden CG. A class of methods for solving nonlinear simultaneous equations. Math Comput. 1965;19(92):577-593.
23. The MathWorks, Inc., Matlab user's guide (r2016b).
24. Seylabi EE, Jeong C, Taciroglu E. On numerical computation of impedance functions for rigid soil-structure interfaces embedded in heteroge-
neous half-spaces. Comput Geotech. 2016;72:15-27.
25. Bardet JP, Huang Q, Chi S. Numerical prediction for model no. 1. Proceedings of the International Conference on the Verification of Numerical
Procedures for the Analysis of Soil Liquefaction Problems, Vol. 1. Netherlands: A.A. Balkema; 1993:67-86.
26. Hashash YM, Dashti S, Romero MI, Ghayoomi M, Musgrove M. Evaluation of 1-D seismic site response modeling of sand using centrifuge
experiments. Soil Dyn Earthquake Eng. 2015;78:19-31.
27. Darendeli MB. Development of a new family of normalized modulus reduction and material damping curves. Ph.D. Thesis; 2001.
20 ZHANG ET AL.
28. Heath MT. Scientific Computing: An Introductory Survey. 2nd ed. New York: McGraw-Hill; 2002.
29. Auger F, Flandrin P. Improving the readability of time-frequency and time-scale representations by the reassignment method. IEEE Trans Signal
Process. 1995;43(5):1068-1089.
30. Gillis K, Dashti S, Hashash YM. Dynamic calibration of tactile sensors for measurement of soil pressures in centrifuge. ASTM Geotech Test J.
2015;38:261-274.
31. U.S. Naval Facilities Engineering Command. Foundations and Earth Structures: Design Manual 7.2 (NAVFAC DM-7.2), U.S. Navy, U.S.
Government Printing Office, Washington, D.C. 20402; 1986.
32. Kramer S. Geotechnical Earthquake Engineering, Prentice-Hall Civil Engineering and Engineering Mechanics Series: Prentice Hall: Upper
Saddle River, NJ; 1996. 347.
33. Montáns FJ. Implicit multilayer J2-plasticity using Prager's translation rule. Int J Numer Methods Eng. 2001;50(2):347-375.
34. Caminero MÁ, Montáns FJ. An enhanced algorithm for nested surfaces plasticity using the implicit mróz translation rule. Comput Struct.
2006;84(26):1684-1695.
How to cite this article: Zhang W, Esmaeilzadeh Seylabi E, Taciroglu E. Validation of a three-dimensional constitutive
model for nonlinear site response and soil-structure interaction analyses using centrifuge test data. Int J Numer Anal Meth
Geomech. 2017;0:1–20. https://doi.org/10.1002/nag.2702