See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/306326006

Modeling of Nevada Sand Behavior Using CHSOIL

Conference Paper · February 2011

CITATIONS READS
3 403

3 authors, including:

Christine Detournay Yanhui Han

Itasca Consulting Group, Inc. Aramco Services Company
33 PUBLICATIONS   450 CITATIONS    82 PUBLICATIONS   596 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Civil Geomechanics View project

Petroleum geomechanics View project

All content following this page was uploaded by Yanhui Han on 20 August 2016.

The user has requested enhancement of the downloaded file.

 Continuum and Distinct Element Numerical Modeling in Geomechanics – 2011 – Sainsbury, Hart, Detournay & Nelson (eds.) Paper: 13-08
©2011 Itasca International Inc., Minneapolis, ISBN 978-0-9767577-2-6

Modeling of Nevada Sand behavior using CHSOIL

C. Detournay
Itasca Consulting Group, Inc., Minneapolis, MN, USA
Z. Wang
AMEC Geomatrix, Inc.
Y. Han
Itasca Consulting Group, Inc., Minneapolis, MN, USA

ABSTRACT: CHSOIL is a constitutive model for soil formulated in the theory of hardening
plasticity. The model, recently available for FLAC (Itasca 2008), is proposed as an alternative to
nonlinear elasticity model (such as Duncan Chang, 1980 hyperbolic model) capable to produce
shear induced volumetric changes. The model features a build-in friction hardening law that
uses parameters similar to the hyperbolic model as direct input. The paper describes the
constitutive model logic and methodology used to calibrate model parameters based on drained
triaxial test results for Nevada Sand. The calibrated model has been used in stability studies of
levee sections built in Nevada Sand, and a generic example is provided as part of the paper to
illustrate model performance.

1 INTRODUCTION

The CHSOIL model is a new elasto-plastic model with friction-hardening for the simulation of
soil behavior. The model is a simplified version of the CYSOIL model that offers built-in laws
but does not have a volumetric cap. Its particular features include (1) a built-in friction-
hardening law that uses model parameters of hyperbolic type as direct input; (2) a cohesive and
frictional Mohr-Coulomb shear envelope; and (3) two built-in dilation laws. The unloading
behavior is elastic. Reloading is elastic up to the outermost yield envelope reached previously.
The model theory is outlined in Section 2, calibration of model properties to fit the behavior
of Nevada Sand in drained triaxial tests is described in Section 3, an example application for a
levee section is provided in Section 4, and conclusions for the work are presented in Section 5.

2 CHSOIL MODEL
2.1 Notation convention
Stress and strain, represented by σ and ε, respectively, are positive in extension. Also, effective
stresses are denoted by a prime ( ' ). The principal effective stresses are σ 1' , σ 2' , σ 3' ; by
convention, σ 1' < σ 2' < σ 3' (i.e. σ 1' is the most compressive stress).

2.2 Elastic behavior

The elastic behavior is expressed using Hooke’s incremental law:
Δσ1' = α1Δε1e + α 2 ( Δε 2e + Δε 3e )
Δσ 2' = α1Δε 2e + α 2 ( Δε1e + Δε 3e ) (1)

Δσ 3' = α1Δε 3e + α 2 ( Δε1e + Δε 2e )

where α1 = Ke + 4 Ge / 3, α2 = Ke – 2 Ge / 3, and Ke and Ge are the tangent elastic bulk and shear
modulus, respectively. In the CHSOIL model, the elastic shear and bulk modulus depend on
initial mean effective stress, pm' . The local pm' value is specified by the user, e.g. using the
expression pm' = – ( σ 1' + σ 2' + σ 3' ) / 3.
The variation of Ge and Ke with pm' is represented by the following equations (see, for
example, Byrne et al. 2003):
n m
⎛ pm' ⎞ ⎛ pm' ⎞
G = Gref ⋅ pref
e
⎜⎜ ⎟⎟ ; K = K ref ⋅ pref
e
⎜⎜ ⎟⎟ (2)
⎝ pref ⎠ ⎝ pref ⎠
The parameter pref is reference pressure, Gref and Kref are shear and bulk modulus number,
respectively (Gref · pref and Kref · pref are tangent elastic shear and bulk modulus at reference
pressure, respectively), and n and m are constant modulus exponents (n ≤ 1, m ≤ 1).
Relations between Kref, Gref, Young’s modulus number, Eref, and Poisson’s ratio at reference
pressure vref are:
Eref Eref
K ref = ; Gref = (3)
3 (1 − 2ν ref ) 2 (1 + ν ref )
The user can specify either Eref and vref or Kref and Gref as input properties for the model. If Eref
and vref are specified, Kref and Gref are calculated internally from (3), and the resulting values are
used with (2).
As noted above, the initial mean effective pressure, pm' , is provided as an input property for
the model. (This value may be evaluated using FISH and stored in the associated zone property
offset). The value of pm' is used to calculate the tangent elastic shear and bulk modulus,
according to (2) and (3). The modulus Ge and Ke are not changed automatically in terms of mean
effective pressure. However the values can be updated using FISH during the course of the
simulation.

2.3 Shear Yield Criterion and Flow Rule

Shear yielding is defined by a Mohr-Coulomb criterion. The yield envelope is expressed as
follows:
f = σ 1' − σ 3' Nφm + 2c Nφm (4)

where φm is the mobilized friction angle, c is cohesion and, by definition Nφm = (1+sin φm) /
(1−sin φm).
The potential function is non-associated, and has the form:
g = σ 1' − σ 3' Nψ m (5)
Where Nψm = (1+sinψm) / (1−sinψm) and ψm is the mobilized dilation angle. It is possible to
specify ψm in terms of plastic shear strain, γp (see Section 2.4), by means of a user-defined table.
Two built-in laws also are available as alternatives: a law based on Rowe’s (1962) dilatancy
theory (see Section 2.7), and a simple step function whereby ψm is zero for φm < φcv, where φcv is
a constant specified by the user, and ψm is equal to the ultimate dilation value ψf for values of
mobilized friction larger than φcv. The additional constraint that ψm cannot exceed φm (to
prevent unwanted generation of energy from taking place) is enforced internally by the code for
all available options.

2.4 Hardening Parameter

The evolution parameter for shear yielding, γp, is defined incrementally as

1⎡
Δγ p =
2 ⎢⎣
( Δε 1 )
dp 2
+ ( Δε 2 )
dp 2
+ ( Δε 3 ) ⎥⎦
dp 2 ⎤
(6)

where Δε idp , i = 1,3 are the principal, deviatoric, plastic shear-strain increments.

2
2.5 Friction Hardening
For most soils, the plot of deviatoric-stress versus axial-strain obtained in a drained triaxial test
may be approximated by a hyperbola. The model incorporates a friction, strain-hardening law to
capture this hyperbolic behavior. The hyperbolic relationship used between stress ratio and
plastic shear strain is similar to that used in the UBCSAND model logic (Byrne et al. 2003), and
also included in the CYSOIL model (see Section 2.4.9 in Theory and Background Manual for
FLAC, Itasca 2008):
⎡ ⎤
sin φ f ⎢⎢ 1
⎥
⎥
sin φm = 1− (7)
Rf ⎢ p G
e
Rf ⎥
⎢ 1 + γ ⎥
⎣ pm sin φ f
'
⎦
where φf , the ultimate friction angle, and Rf (the failure ratio) is a constant, smaller than 1 (0.9
in most cases). Equation (7) is used in the CHSOIL model to calculate mobilized friction in
terms of plastic shear strain. Note however that, in its present stage, the model is not intended to
simulate dynamic liquefaction.

2.6 Over-Consolidation
Normal- or over-consolidation can be prescribed by specifying an initial value of friction that is
either equal or smaller than the normally consolidated value φnc, respectively. The value φnc can
be obtained from (4); in the example of a cohesionless soil, the expression is φnc = arcsin | ( σ 1' −
σ 3' ) / ( σ 1' + σ 3' ) |Initial. The code automatically assigns an initial value of evolution parameter γp
that is consistent with the specified initial value of mobilized friction angle. Also, the initial
value of friction must be smaller than the ultimate value φf, and an upper bound equal to φf is
considered automatically by the code.

2.7 Shear-Induced Compaction and Dilation

A certain amount of irrecoverable volumetric strain, ε vp , is expected to take place as a result of
soil shearing. Also, under small monotonic shear strains, there is a tendency for the soil skeleton
to contract due to grain re-arrangements. For larger shear strains, the soil skeleton may dilate if
the soil is dense, due to grains riding over each other. Dilation strain-hardening laws are
incorporated in the logic to model this behavior.
Rowe dilation law can be used as an option to characterize (mobilized) dilation angle ψm.
According to Rowe stress-dilatancy theory (1962), there is a (constant-volume) friction angle,
φcv below which the material contracts (i.e. for φm < φcv), while, for higher stress ratios (i.e. for
φm > φcv), the material dilates. The equation has the form

sin φm − sin φcv

sinψ m = (8)
1 − sin φm sin φcv
where sin φcv = ( sin φ f − sinψ f ) / (1 − sin φ f sinψ f ) and φf and ψf are (known) ultimate values of
friction and dilation, respectively.
Dilation is evaluated in terms of plastic shear strain, based on the last two equations and the
assumed relation between φm and γp reported in (7). Rowe’s dilation law is selected by
specifying the model property DILAW = 1. A simple dilation law also can be used whereby ψm
= 0 for φm < φcv, and ψm = ψf for φm ≥ φcv; the option is activated by specifying DILAW = 2.
Finally, the user may choose to define an alternative dilation law by specifying an input table of
dilation value versus plastic shear strain. The model property DILAW must be set to zero to
activate this option.

3
2.8 Duncan and CHSOIL model parameters
A parallel can be drawn, for particular cases, between parameters used in the hyperbolic stress-
strain relationships reported in Duncan et al. (1980) and CHSOIL model properties. The
correlations depend on the property input; they are listed in Table 1 below.

Table 1. Relation between Duncan and CHSOIL parameters.

______________________________________________________________________________________
Model m=n No dilation
______________________________________________________________________________________
Duncan et al. 1980 Kur , n Kb , m

9 K ref Gref
CHSOIL input Kref , Gref ,n Kref , m
3K ref + Gref
Eref
CHSOIL input Eref , vref Eref , n ,m
3 (1 − 2vref )
______________________________________________________________________________________

3 CALIBRATION EXAMPLE

Results of drained triaxial CIDC (p' = constant) tests on Nevada Sand are used to give an
example of the methodology that could be used to calibrate the properties of the CHSOIL
model.

3.1 Available data

The test results, plotted in solid lines in Figures 1 to 3, consist of three sets of data:
1 deviatoric stress versus mean effective stress;
2 deviatoric stress versus axial strain (up to 25%); and
3 volumetric strain versus axial strain (up to 25%).
Data are provided for 3 values of mean stress (40, 80 and 160 kPa) and two values of relative
density, Dr (40% and 60%).

3.2 Data characterization

We consider axial strains up to 5% for the calibration because strains are not expected to
develop above that level in the intended (static, drained) application. The material (Nevada
Sand) appears to be purely frictional in character. Also, two main features characterize the data:
(1) the hyperbolic behavior observed in the plot of deviatoric stress versus axial strain, and (2)
the bilinear dilatant behavior exhibited by the plot of volumetric strain versus axial strain. The
CHSOIL constitutive model capabilities, including friction hardening and variable dilation, are
used to capture these features.
The elastic tangent shear and bulk moduli are functions of initial mean effective pressure
according to the laws (2) to (3). For friction hardening, we use the built-in law (7).
The dilation law is bilinear: its value is zero below the stress ratio φ = φcv, and a constant, ψf ,
above it:
ψ = 0 for φ < φcv
(9)
ψ = ψ f for φ ≥ φcv
The failure ratio Rf is 0.99 for this application. The following 8 parameters must be defined:
Eref , ν, pref , φf , ψf , φcv , m, n.

4
3.3 Calibration methodology
Calibration of the model parameters is done in two steps. First, using theoretical considerations
and values recorded in the literature, we derive a first estimate for parameter values from
laboratory results. Second, we improve on the estimates by modeling triaxial experiments
numerically and matching results with those obtained in the laboratory.

3.3.1 Parameter estimates

The value of ultimate friction is derived from a linear fit to plots of deviatoric stress, q, versus
mean effective stress, p', obtained from the CIDC tests. The linear fit to maximum q at given p'
provides a value for the maximum stress ratio, α:

q
=a (10)
p'

From the definition of a purely frictional Mohr-Coulomb criterion, we obtain the following
relation:
3a
sin φ f = (11)
6+a
from which the ultimate friction angle, φf, can be derived. The estimated values are φf = 34°for
Dr = 40% and φf = 39° for Dr = 60%.
The value of ultimate dilation angle is derived from the slope, b, of a linear fit to the
laboratory curves of minus volumetric strain versus axial strain. To relate b to ψf , we use, as a
first approximation, the expression for bilinear idealization of triaxial stress results provided by
Vermeer and de Borst (1984):

2sinψ f
b= (12)
1 − sinψ f
The first estimates are ψf = 8.2° for Dr = 40% and ψf = 14.6° for Dr = 60%.
To estimate the parameter φcv for a given Dr, we first derive the maximum value of axial
strain, ε a* , at which volumetric strain is negligible from the laboratory plots of volumetric strain
versus axial strain at a given p'. From the knowledge of axial strain, ε a* , we estimate q from the
laboratory plot of deviatoric stress versus axial strain at that p', and then the ratio q / p'. Three
values of the ratio are available (one at each p'). The mean value of q / p' is used to calculate the
corresponding friction angle using formula (11), where φf now is replaced by φcv , and α by
average q / p' (see (10)). The estimate is φcv = 27.7° for both Dr = 40% and Dr = 60%.
The first guess for Eref is taken as 1200 for Dr = 40% at atmospheric pressure; this value is
equivalent to that used in the triaxial numerical experiment for dense soil listed in Example 2.9
of the Theory and Background manual for FLAC Version 6.0 (Itasca 2008).
The value of ν for this work is chosen, arbitrarily, to be 0.35. Also, we select n = m = 0.5
for this work, and the elastic constant, Eref, is estimated by matching the initial slopes of q versus
axial strain curves obtained in similar triaxial tests (under constant mean stress) performed
numerically and in the laboratory. These estimates may not be very accurate. More robust
estimates for the elastic constants can be obtained from laboratory results of small unloading-
reloading excursions, but such results were not available for the example.

3.3.2 Triaxial tests using numerical experiments

Triaxial experiments are conducted numerically using FLAC at three levels of mean stress: 40,
80 and 160 kPa, and for two values of Dr: 40% and 60%. The axisymmetric configuration is
selected for the strain-controlled simulations. A servo-control is applied to maintain the mean
stress constant during the numerical experiments.

5
 The estimates for model parameters are used to carry out the numerical experiment, and the
test results are compared to the available laboratory data (imported in FLAC tables). The
parameters are adjusted (in the order listed above), and the numerical experiment is repeated
until a satisfactory curve fitting is obtained. The results of the curve fitting experiment are listed
in Table 2.

Table 2. Calibration results.

_____________________________________________________________________
Dr Eref v φf ψf φcv m n
(°) (°) (°)
_____________________________________________________________________
40 % 1800 0.35 34 7.5 28 0.5 0.5
60 % 2000 0.35 39 13.5 28 0.5 0.5
_____________________________________________________________________

A comparison between numerical prediction and laboratory results using the calibrated
properties for Dr = 40% and Dr = 60%, is shown in Figures 1 through 3. Note that the soil
convention for positive stress/strain in compression is adopted in the plots (Dilation is negative
on the plots.); the convention is opposite to that adopted in FLAC. The results of the calibration
are quite satisfactory overall.

Figure 1. Deviatoric stress versus mean stress for Dr = 40% (left) and Dr = 60% (right) - comparison
between laboratory results (line) and numerical predictions (symbol).

Figure 2. Deviatoric stress versus axial strain in % at three levels of mean stress for Dr = 40% (left) and
Dr = 60% (right) - comparison between laboratory results (line) and numerical predictions (symbol).

6
Figure 3. Volumetric strain in % versus axial strain in % at three levels of mean stress for Dr = 40% (left)
and Dr = 60% (right) - comparison between three laboratory results and three numerical predictions.

4 APPLICATION EXAMPLE

Sheet pile wall analyses were carried out to evaluate the stability of an assumed levee sections
built in Nevada Sand under rising water load. The soil profile included layers of Nevada Sand
modeled using the CHSOIL model with properties derived from model calibration outlined in
Section 3. The water load was increased in one-foot increments from ground surface to 10 feet
high on the flood side of a sheet pile wall. The deformed grid and computed maximum shear
strain are presented in Figure 4. At 10 feet water load, the soil on the protected side approaching
failure and developed more than 10% shear strain. The computed peak horizontal displacements
at the top of the wall are shown in Figure 5. A sharp increase in displacement is observed as the
water level rises above 9 feet.

5 CONCLUSIONS

A hyperbolic type model for simulation of non-cyclic soil behavior has been introduced. The
model, called CHSOIL is formulated based on the theory of plasticity, and is capable to produce
shear induced volumetric changes. The methodology used to calibrate model parameters based
on available drained triaxial test results for Nevada Sand has been outlined. The model
calibration produced a good fit using one set of parameters to simulate available results. The
model has shown good performance in the study of an assumed levee section built in Nevada
Sand.

Figure 4. Sheet pile levee model and computed maximum shear strain (10’ water up to the wall top on the
left side).

7
 Figure 5. Computed wall top horizontal displacement versus water level (feet).

REFERENCES

Byrne, P.M., Park, S.S. & Beaty, M. 2003. Seismic Liquefaction: Centrifuge and Numerical Modeling. In
R. Brummer et al. (eds), FLAC and Numerical Modeling in Geomechanics — 2003 (3rd International
FLAC Symposium, Sudbury, Ontario, Canada, October 2003), pp. 321-331. Lisse: Balkema.
Duncan, J.M., Byrne, P., Wong, K.S. & Mabry, P. 1980. Strength, Stress-Strain and Bulk Modulus
Parameters for Finite Element Analyses of Stresses and Movements in Soil Masses. College of
Engineering, Office of Research Services, University of California, Berkeley, Report No. UCB/GT/80-
01.
Itasca Consulting Group, Inc. 2008. FLAC – Fast Lagrangian Analysis of Continua, Ver. 6.0 User’s
Manual. Minneapolis, Itasca.
Rowe, P.W. 1962. The Stress-Dilatancy Relation for Static Equilibrium of an Assembly of Particles in
Contact. Proc. Roy. Soc. A., 269, 500-527.
Vermeer, P.A., & de Borst, R. 1984. Non-Associated Plasticity for Soils, Concrete and Rock. Heron,
29(3), 3-64.

View publication stats