Professional Documents
Culture Documents
art ic l e i nf o a b s t r a c t
Article history: Wave propagation in soil is dependent on both the stiffness and the material damping of the soil. While
Received 28 May 2014 some researchers have performed finite element modelling of resonant column tests and wave propa-
Received in revised form gation in soil, most do not describe the methodology in detail and there is little or no verification of the
31 May 2016
correctness of the model. Viscoelastic model has been used to model wave propagation in soil. However,
Accepted 14 June 2016
Available online 28 June 2016
the determination of the parameters in the viscoelastic model is complicated and the parameters may
not be related to the soil properties. This paper presents a simplified viscoelastic model with soil
Keywords: parameters obtainable from advanced geotechnical testing to simulate wave propagation in soil medium
Viscoelastic taking into account of material damping. The viscoelastic material model was first calibrated by re-
Resonant column
plicating torsional, longitudinal and flexural modes resonant column tests. The relationships between the
Stiffness
parameters of the simplified viscoelastic model and their corresponding stiffness and damping proper-
Material damping
Finite element ties were investigated. An equation was proposed to correlate the decay constant used in the simplified
viscoelastic model and the material damping ratio obtained through the application of the logarithmic
decrement method on the modelled resonant column test results. The simplified viscoelastic model was
then evaluated by modelling wave propagation in a semi-infinite medium. Results indicated that the
viscoelastic model with parameters as proposed in this paper is able to model wave propagation in soils.
& 2016 Elsevier Ltd. All rights reserved.
1. Introduction Method but no mention was made if damping was used in the
model.
Dynamic testing of soil in the laboratory to obtain small-strain In other cases, the viscoelastic model has been employed to
soil modulus and damping ratio has been facilitated by the ad- investigate dynamic response and wave propagation in soil [8–12].
vancement of test methods and the increasing need to design for While the formulation of the viscoelastic model is well docu-
more challenging situations such as earthquake, wave and impact mented in the literature, they contain lengthy and complicated
loading. In such design situations, strength is not the only con- differential equations with parameters which prove challenging to
sideration as one would also have to consider the effects of the obtain and apply.
propagating wave. However, numerical modelling of dynamic be- Hayner [13], Spanos and Tsavachidis [14] and Lee et al. [15]
havior of soil is not as extensive as experimental tests. Karl et al. used viscoelastic models with complicated configurations of
rheological elements (springs and dashpots) to simulate wave
[1,2] determined numerically the natural frequency of soil speci-
propagation in soil. However, usage of such elements meant that
men in resonant column test under different modes of vibration
numerous input parameters corresponding to the rheological
using ABAQUS. The numerical study, however, did not address the
elements have to be obtained either experimentally or numerically
issue of material damping and did not describe the type of ma-
by using Monte Carlo simulations. To obtain the parameters for the
terial model used. Arroyo et al. [3–5] and Rio [6] employed a mass rheological elements experimentally requires specialized appara-
proportional damping model in FLAC3D to model wave propaga- tus and techniques (e.g. Dynamic Mechanical Analysis) to obtain
tion in soil during bender element tests but did not provide any the strain response of the soil to an applied sinusoidal loading. To
validation. Finally, Hardy et al. [7] modelled bender element tests obtain the parameters for the rheological elements numerically
in a triaxial specimen using Fourier Series Aided Finite Element requires considerable effort to calibrate the numerous unknown
input parameters with known response of the soil. However,
n
Correspondence to: School of Civil & Environmental Engineering, Nanyang
treating the viscoelastic model as a “black box” fails to promote
Technological University, Blk N1, 50 Nanyang Avenue, 639798 Singapore, Singapore. understanding of the viscoelastic model. In some cases, the values
E-mail address: cecleong@ntu.edu.sg (E.C. Leong). of the parameters derived from such “curve-fitting” methods
http://dx.doi.org/10.1016/j.soildyn.2016.06.005
0267-7261/& 2016 Elsevier Ltd. All rights reserved.
208 Z. Cheng, E.C. Leong / Soil Dynamics and Earthquake Engineering 88 (2016) 207–214
Fig. 1. Graphical representation of (a) Kelvin–Voigt model, (b) Maxwell model, (c) Biot hysteretic model, and (d) Standard Linear Solid or Zener model.
might not be unique leading to unreasonable soil parameters used where γo is the amplitude of the sinusoidal strain, τo is the am-
for modelling. plitude of the sinusoidal stress, ϕ is the phase difference between
Thus, the objective of this study aims to provide better un- them, ω is the circular frequency and t is the time.
derstanding of the viscoelastic model which is frequently used to The corresponding shear stiffness can be expressed in terms of
model wave propagation in soil. As such, the viscoelastic model the dynamic complex shear modulus as shown in Eq. (2):
was reviewed. A simplified viscoelastic model calibrated with soil τo
parameters obtained from advanced geotechnical testing was in-
G* = ( cos φ + i cos φ) = G′ + iG′′
γo (2)
troduced. Calibration was conducted by modelling the resonant
column test as both small-strain stiffness and material damping where G* is the complex shear modulus, G′ is the elastic (or sto-
ratio are commonly obtained from the test. These two parameters rage) modulus which measures the ability of the material to store
govern the response of wave propagation. Natural frequencies for energy and G′′ is the loss (or viscous) modulus which measures the
different vibration modes (torsional, flexural and longitudinal) ability of the material to dissipate energy and i is the imaginary
were validated with those derived from well-established analytical number ( ¼ −1 ).
solutions. Material damping ratio derived from the application of Ratio of the loss modulus and elastic modulus gives the loss
the logarithmic decrement method [16] was compared with the factor η which is related to the damping ratio ξ as given below:
input parameters of the simplified viscoelastic model. The study G′′
will also investigate the relationship between the parameters of η= = tan φ = 2ξ
G′ (3)
the simplified viscoelastic model and the stiffness and the
damping responses. Correlations with strong theoretical basis Viscoelastic material exhibits volume change due to loading
were proposed so that the viscoelastic model can be easily em- (elasticity) and time (viscosity) i.e., deformation is time depen-
ployed. Further verification was conducted by modelling wave dent. Phenomenon of viscoelastic materials includes creep (or
propagation in a semi-infinite medium with different damping retardation) and relaxation, which is described mathematically
with the help of constitutive models. These models are described
ratios.
figuratively using different combinations of springs and dashpots.
Examples of these models include the Kelvin–Voigt, Maxwell and
Standard Linear Solid models which will be illustrated in greater
2. Viscoelastic model
detail subsequently.
The Kelvin–Voigt model shown in Fig. 1(a) is used to describe
The difference between elastic and viscoelastic materials cen-
the creep effect. The dashpot which is connected in parallel with
ters on their behavior when subjected to vibratory motion. As-
the spring prevents the spring from reacting immediately to the
suming that a sinusoidal deformation is applied to a specimen, the
applied force [17]. Simply explained, the dashpot provides the
corresponding strain (γ) can be described using Eq. (1a). If the delay in deformation upon application of stress [18]. The de-
material is elastic, the strain and stress will be in phase. However, formation in the Kelvin–Voigt model is represented by [19]:
a viscoelastic material will experience a phase lag in its resulting
τt
stress (τ) as described in Eq. (1b). γt = (1 − e−t / Tc )
G (4)
γ = γo eiωt (1a)
where τt and γt are the shear stress and strain at the time t, G is
the shear modulus and Tc is the retardation time.
The Maxwell model shown in Fig. 1(b) is used to describe the
τ = τo ei (ωt + φ) (1b) relaxation effect. Relaxation is characteristised by the gradual
Z. Cheng, E.C. Leong / Soil Dynamics and Earthquake Engineering 88 (2016) 207–214 209
decrease in stress when the material is subjected to deformation triaxial and resonant column tests. However, neither the combi-
[18]. In this case, the dashpot can be seen to be gradually relaxed nation of Go and G1 with Gmax nor the correlation of β with
upon the application of a sudden deformation. The stress invoked damping ratio were mentioned in the literature [24–26]. The paper
in the spring will therefore decrease slowly, hence the term re- therefore aims to fill this gap in knowledge by investigating the
laxation. The change in stress with time in the Maxwell model is relationship between the above-mentioned parameters and pro-
represented by [19]: vides correlations that can be used for the viscoelastic model.
(
τt = Gγo e−t / Tr ) (5)
where γo is the initial shear strain and Tr is the relaxation time. 3. Input Parameters for LS-DYNA
Expression for retardation and relaxation time is shown in Eq.
(6). While both Tr and Tc is represented by the same equation, they In this paper, the finite element program LS-DYNA [27] was
differ in physical meaning and value [18] through the viscosity used to model the resonant column test on a soil specimen under
coefficient η̂ : different excitation modes. LS-DYNA was chosen as it is able to
perform highly non-linear, transient dynamic analysis using ex-
ηˆ plicit and implicit time integration. The explicit approach involves
T=
G (6) using the central difference method to march forward in time to
where T can be either retardation or relaxation time and η̂ is the solve transient dynamic problem. The implicit approach, on the
viscosity coefficient of the dashpot. other hand, is used to solve static, low strain rate problems and to
Compared to the Kelvin–Voigt model which models the creep extract the corresponding eigenvalue (resonant frequency) and
effect, the Maxwell model is more suited to model damping in soil eigenvector (mode shape). The implicit approach, however, uses
[15,20]. In most cases, the Biot hysteretic model (Fig. 1(c)) is em- much more memory compared to the explicit approach. The si-
ployed to handle more complex dynamic soil behavior [21]. The mulations were run on a 64 bit CPU equipped with Intel Processor
Biot hysteretic model consists of a spring element in parallel with of i7-3610QM and 8GB of RAM. The implicit and explicit ap-
an infinite number of Maxwell models in parallel connection. The proaches take about 15 and 60 min, respectively.
model can be expressed mathematically using Eq. (7). Implicit eigenvalue analysis was used to obtain the funda-
mental frequencies for the various vibration modes (flexural, tor-
N
sional and longitudinal). The fundamental frequency for each vi-
G (t ) = G∞ + ∑ Gi e−t / Tri bration mode was then used to derive the decay constant (β) for
i=1 (7)
use in the explicit time integration. Explicit time integration was
where G(t) is the shear modulus at time t, G1 is the long term or used to obtain the amplitude decay in the time domain upon
infinite shear modulus, Gi and Tri are the shear modulus and the reaching steady state so that material damping ratio can be de-
relaxation time of the ith Maxwell model placed in parallel rived from the logarithmic decrement method.
connection. Before damping can be studied using the Zener model, the
The second term of Eq. (7) is known as the Prony series [17,22] input parameters (K, ρ, Go, G1 and β) need to be correlated to the
and shares a similar expression as the Maxwell model shown by actual parameters (Kmax, ρ, Gmax and ξ). Due to the mismatch in
Eq. (5). The Prony series dictates the decrease in shear modulus the number of parameters available, Gmax is assumed to be Go and
with time. Upon total relaxation, the Biot hysteretic model will be G1 ¼0, thereby reducing the Zener model to a Maxwell model.
left with the long term shear modulus (G1). However, the large From which, the correlation between parameters can be obtained
number of rheological elements present will mean that consider- by reconciling the constitutive model to resemble Eq. (2):
able effort have to be undertaken to calibrate the model. This can In the Maxwell model, the stress in both the spring and
be done experimentally using a Dynamic Mechanical Analyzer dashpot is similar. Total deformation is therefore the summation of
(DMA) or by calibrating with known data. strains which is given as:
The DMA is used to characterize viscoelastic behavior in ma-
dγ 1 dτ τ
terial by recording the response of the material subjected to si- = +
nusoidal stress so that the complex modulus can be determined dt G dt ηˆ (9)
[23]. However, the apparatus is not readily available in most Substituting Eqs. (1a) and (1b) into Eq. (9),
geotechnical testing laboratories. Calibration of the parameters
with Monte Carlo simulation and other optimization algorithm 1 1
(iω) γo ei (ωt ) = (iω) τo ei (ωt + φ) + τo ei (ωt + φ)
will also take significant effort and time [13–15]. Hence, there is a G ηˆ (10)
need for a simplified model where parameters can be obtained Dividing by γo ei (ωt ) throughout gives
from accessible laboratory tests.
Biesen [24], Hu et al. [25] and Smith and Duris [26] used a iω 1 ⎡ iωηˆ + G ⎤
iω = G* + G* = G* ⎢ ⎥
model simplified from the Biot hysteretic model as shown in Fig. 1 G ηˆ ⎣ Gηˆ ⎦
(d) otherwise known as the Standard Linear Solid or Zener model.
This model is available in the finite element program LS-DYNA. ⎡ iωGηˆ ⎤
The time dependent shear modulus [27] is given by:
G* = ⎢ ⎥
⎣ iωηˆ + G ⎦ (11)
G (t ) = G∞ + ( Go − G∞ ) e−βt (8) ⎡ iωηˆ −G ⎤
Multiplying by ⎣ iωηˆ −G ⎦ on both sides of Eq. (11) results in
where Go is the instantaneous or short term shear modulus and β ω2Gηˆ2 ωG2ηˆ
is the decay constant corresponding to the inverse of the relaxa- G* = G′ + iG" = +i 2 2
ω2ηˆ2 +G 2 ω ηˆ + G2 (12)
tion time.
The input parameters for the viscoelastic material model are The above equations allowed the dynamic complex shear
bulk modulus (K), bulk density (ρ), Go, G1 and β. Small-strain bulk modulus to be expressed in terms of the constitutive model
and shear moduli (Kmax and Gmax) can be obtained from advanced parameters. By substituting Eq. (12) into Eq. (3) and simplifying
geotechnical laboratory tests like the bender element, cyclic with the help of Eq. (6), the correlation between β and ξ was
210 Z. Cheng, E.C. Leong / Soil Dynamics and Earthquake Engineering 88 (2016) 207–214
πGD4
KT =
32L (16b)
where Jr,i is the polar moment of inertia of the ith attached units,
1
Jr,s is the polar moment of inertia of the specimen given by 2 mr 2, D
is the diameter of the specimen and r is the radius of the
specimen.
The damping ratio ξ was computed from decaying amplitudes
using the logarithmic decrement method (LDM) described in
ASTM 4015-07 [16]:
δ2
ξ=
4π 2 + δ 2 (17a)
⎛ A ⎞1
δ = ln ⎜ i ⎟
⎝ Ai + N ⎠ N (17b)
Table 3
Comparison of natural frequency obtained from both LS-DYNA and analytical
solution.
L/D¼ 2
LS-DYNA Analytical
L/D ¼3
LS-DYNA Analytical
0.1 0.25
(a)
0.2
0.05
0.15
0
0.1
-0.05 Flex
0.05 Flex-1.1β-GR0
-0.1 Flex-1.2β-GR0
0.0 5.0 10.0 15.0 20.0 0
0 0.05 0.1 0.15 0.2 0.25
Time (ms) (a) Input Damping Ratio
(b) 0
0.25
-1
ln (Displacement)
0.2
Output Damping Ratio
0.2
0.15
0.15
0.1
Long
0.1
Flex 0.05 Long-1.1β-GR0
0.05 Tor Long-1.2β-GR0
0
Long 0 0.05 0.1 0.15 0.2 0.25
0 (c)
0 0.05 0.1 0.15 0.2 0.25 Input Damping Ratio
Input Damping Ratio Fig. 6. Damping ratio obtained from LDM when excited in the (a) flexural
(b) torsional, and (c) longitudinal mode.
Fig. 5. Comparison of output and input damping ratios for flexure, torsional and
longitudinal modes of resonant column test.
damping ratio. From Eq. (13), the damping ratio was observed to
be a function of β . Results from Fig. 6 were then re-plotted to- 0.25
ω
gether to give Fig. 7. Contrary to the data presented in Fig. 6, Fig. 7 1.0β
shows that the damping ratio is independent of β . Results are 0.2 1.1β
ω
shown to have a scatter of 7 20% about the best-fit line (Eq. (18))
Damping Ratio
thereby suggesting that the following equation can be used to 0.15 1.2β
relate the damping ratios to β regardless of the vibration mode.
0.1
β 0.05
ξ = 0.4249 for GR=0
ω (18)
0
0 0.1 0.2 0.3 0.4 0.5 0.6
β/ω
5. Wave propagation in a semi-infinite medium
Fig. 7. Variation of damping ratio with β/ω for GR ¼ 0.
Excitation 6. Conclusion
0
FEM - Elastic
-0.5
FEM - Visco 5% Damping
-1
ln (An/A1)
-1.5
-2
-2.5
y
-3
0 100 200 300
x Non-reflecting boundary condition
Distance from excitation source (mm)
Fig. 9. Model used to simulate vibration in the free field. Fig. 10. Amplitude decrease with distance from excitation source.
214 Z. Cheng, E.C. Leong / Soil Dynamics and Earthquake Engineering 88 (2016) 207–214
Go should be Gmax and K can be obtained from both Gmax and ν. poroelastic soils. J Appl Mech 1992;59(1):128–35.
While the long term shear modulus, G1, does not affect the initial [11] Carcione JM. Seismic modelling in viscoelastic media. Geophysics 1993;58
(1):110–20.
stiffness of the model, inclusion of G1 in the viscoelastic model [12] Dwivedi JP, Singh VP, Lal RK. Dynamic response of partially sealed circular
alters the output damping ratio. It is therefore recommended to tunnel in viscoelastic soil condition. Bull Appl Mech 2011;7(26):37–45.
use G1 ¼0 in the viscoelastic model. The equation proposed to [13] Hayner MA. Optimised finite difference schemes for wave propagation in high
correlate the decay constant (β) in the viscoelastic model with the
loss viscoelastic material [PhD Thesis]. Massachusetts Institute of Technology;
1994.
damping ratio (ξ) was found to be able to produce decaying vi- [14] Spanos PD, Tsavachidis S. Deterministic and stochastic analyses of a nonlinear
bration with the stipulated damping ratio. The simplified viscoe- system with a Biot visco-elastic element. Earthq Eng Struct Dyn 2001;30
(4):595–612.
lastic model was verified to perform as expected by modelling
[15] Lee KH, Szajewski BA, Hah Z, Parker KJ, Maniatty AM. Modelling shear waves
wave propagation in a semi-infinite medium. through a viscoelastic medium induced by acoustic radiation force. Int J Nu-
mer Methods Biomed Eng 2012;28(6–7):678–96.
[16] ASTM Standard D4015-07. Standard test methods for modulus and damping of
soils by resonant-column method, American Society for Testing and Materials,
Acknowledgement West Conshohocken, United States; 2007.
[17] Tschoegl NW. The phenomenological theory of linear viscoelastic behavior.
Berlin Heidelberg: Springer-Verlag; 1989.
The first writer acknowledges the Nanyang President Graduate
[18] Lakes R. Viscoelastic materials.New York: Cambridge University Press; 2009.
Scholarship. The support from Grant MINDEF-NTU-JPP/13/01/02 [19] Vyalov SS. Developments in geotechnical engineering volume 36: rheological
administered by the Protective Technology Research Centre, Na- fundamentals of soil mechanics. New York: Elsevier; 1986.
[20] Stucky P, Lord W. Finite element modeling of transient ultrasonic waves in
nyang Technological University, is gratefully acknowledged.
linear viscoelastic media. IEEE Trans Ultrason Ferroelectr Freq Control 2001;48
(1):6–16.
[21] Muscolino G, Palmeri A, Ricciardelli F. Time-domain response of linear hys-
References teretic systems to deterministic and random excitations. Earthq Eng Struct
Dyn 2005;34(9):1129–47.
[22] Makris N, Zhang J. Time-domain viscoelastic analysis of earth structures.
[1] Karl L, Haegeman W, Degrande G, Dooms D. Measurement of material Earthq Eng Struct Dyn 2000;29(6):745–68.
damping with bender elements in triaxial cell. In: Proceedings of the 3rd In- [23] Menard KP. Dynamic mechanical analysis: a practical introduction.Florida:
ternational Symposium on Deformation Characteristics of Geomaterials, vol. 1. CRC Press; 2008.
Lyon, France; 2003. pp. 3–11. [24] Biesen ED. Prediction of plastic deformation in aluminum softball bats using
[2] Karl L, Haegeman W, Degrande G, Dooms D. Determination of the Material finite element analysis [Master thesis]. Washington State University; 2006.
Damping Ratio with the Bender Element Test. J Geotech Geoenviron Eng [25] Hu J, Jin X, Lee JB, Zhang L, Chaudhary V, Guthikonda M, et al. Intraoperative
2008;134(12):1743–56. brain shift prediction using a 3D inhomogeneous patient-specific finite ele-
[3] Arroyo M, Medina L, Wood DM. Numerical modelling of scale effects in ment model. J Neurosurg 2007;106(1):164–9.
bender-based pulse tests. In: Proceedings of the 8th International Symposium [26] Smith LV, Duris JG. Progress and challenges in numerically modelling solid
Numerical Models in Geomechanics, Rome, Italy; 2002. pp. 589–594. sports ball with application to soft balls. J. Sports Sci 2009;27(4):353–60.
[4] Arroyo M, Wood DM, Greening PD. Source near-field effects and pulse tests in [27] LSTC. LS-DYNA User Manual R7.0. Livermore Software Technology Corpora-
soil samples. Geotechnique 2003;53(3):337–45. tion, California, United States; 2013.
[5] Arroyo M, Wood DM, Greening PD, Medina L, Rio J. Effects of sample size on [28] Kumar J, Madhusudhan BN. On determining the elastic modulus of a cylind-
bender-based axial G(0) measurements. Geotechnique 2006;56(1):39–52. rical sample subjected to flexural excitation in a resonant column apparatus.
[6] Rio JFME. Advances in laboratory geophysics using bender elements [PhD Can Geotech J 2010;47(11):1288–98.
thesis]. University College London; 2006. [29] Cascante G, Santamarina C, Yassir N. Flexural excitation in a standard tor-
[7] Hardy S, Zdravkovi´c L, Potts DM. Numerical interpretation of continuously sional-resonant column device. Can Geotech J 1998;35(3):478–90.
cycled bender element tests. In: Proceedings of the 8th International Sym- [30] Harris CM, Piersol AG. Shock and vibration handbook Harris’. New York:
posium Numerical Models in Geomechanics, Rome, Italy; 2002. pp. 595–600. Mcgraw-Hill Professional; 2002 Fifth ed.
[8] Carcione JM, Kosloff D, Kosloff R. Wave propagation simulation in a linear [31] Timoshenko S. Vibration Problems in Engineering. Second ed.. New York: D
viscoacoustic medium. Geophys J 1988;93(2):393–401. Van Nostrand Company, Inc.; 1937.
[9] Carcione JM, Kosloff D, Kosloff R. Wave propagation simulation in a linear [32] Aung AMW, Leong EC. Discussion of “Near-field effects on array-based surface
viscoelastic medium. Geophys J 1988;95(3):597–611. wave methods with active sources” by S Yoon and GJ Rix. J Geotech Geoen-
[10] Bardet JP. A viscoelastic model for the dynamic behavior of saturated viron Eng 2010;136(5):773–5.