You are on page 1of 13

Ref: Beaty, M. & Byrne, P. (1998).

An effective stress model for predicting liquefaction behaviour


of sand. In P. Dakoulas, M. Yegian & R.D. Holtz (Eds.), Geotechnical Earthquake Engineering and
Soil Dynamics III, ASCE Geotechnical Special Publication No. 75, Vol. 1, Proceedings of a
Specialty Conference (pp. 766-777). Seattle: ASCE.

(Copy of submitted manuscript)

AN EFFECTIVE STRESS MODEL FOR PREDICTING


LIQUEFACTION BEHAVIOUR OF SAND

Michael Beaty1 and Peter M. Byrne2

Abstract

An elastoplastic constitutive model is presented that simulates the liquefaction


response of sands in a relatively uncomplicated manner. The model, UBCSAND, is
based on the characteristic behaviour of the soil skeleton as observed in laboratory
element tests. The model has several key features, including a hyperbolic relationship
between stress ratio and plastic shear strain, a flow rule for estimating plastic
volumetric strain from plastic shear strain, and the ability to handle anisotropy.

The simple framework which describes the observed soil response and forms the basis
for the model is presented. Monotonic and cyclic results are computed using the
model and shown to be in good agreement with laboratory element tests. The model
is also applied to the Wildlife Site in California and the predictions compared with
field measurements from the 1987 Superstition Hills Earthquake.

1. Introduction

A constitutive model is presented for evaluating the liquefaction response of sands.


The controlling philosophy behind this model is one of simplicity, an avoidance of
unnecessary complexity. It is believed that a model which captures the characteristic
behaviour of granular soil can be a valuable tool in evaluating and understanding its
performance under static and cyclic loading. Although some complexity is
unavoidable in achieving this goal, this relatively simple model is shown to reasonably
represent the soil response as observed in element tests.

1
Graduate Student, Department of Civil Engineering, University of British Columbia, 2324 Main
Mall, Vancouver, British Columbia V6T1Z4, Canada.
2
Professor, Department of Civil Engineering, University of British Columbia, 2324 Main Mall,
Vancouver, British Columbia V6T1Z4, Canada.

766 Beaty and Byrne


2. Description of Constitutive Model

The model can be described as an elastoplastic effective stress formulation based on


an assumed hyperbolic relation between stress ratio and plastic shear strain. The
model represents the behavior of the soil skeleton, with the effect of any pore fluid
introduced through its volumetric stiffness. As a plasticity model it includes such
features as a yield surface, a flow rule, and a definition for loading, unloading, and
hardening. Applications of this model to problems involving monotonic loading have
been previously presented, and these references provide additional detail, equations,
and derivations (Puebla et al., 1997; Byrne et al., 1995).

2.1 Yield Surface

The yield surface is described by a line of constant stress ratio. Stress ratio, defined
as the maximum shear stress divided by the mean normal effective stress, can also be
expressed as the sine of a mobilized friction angle. Since the yield surface is related to
a friction angle, this permits comparison to the classic Mohr-Coulomb definition of
failure. This is demonstrated in Figure 1, where a stress state at yield has a Mohr’s
circle tangent to the developed friction angle, φd.

The primary difference between this model and the Mohr-Coulomb definition is the
mobilized friction angle is made a function of the loading history. For example, the
initial stress ratio of an isotropically consolidated element will equal zero. As any
increase in stress ratio will cause yielding, the initial yield surface must correspond to
a friction angle of zero. If this element is loaded in monotonic drained compression,
there will be a steady increase in stress ratio corresponding to a radially increasing
yield surface. The yield surface will rise until the peak friction angle φf is achieved.
This is shown schematically on Figure 2. The rate at which the yield surface moves
upward is a function of the hyperbolic hardening relation.

Stress Ratio = q/p = sin φ d


Drained Stress Path φf
Shear Stress, q

Failure Envelope
Shear Stress,

(φ d )c
at φ f
Yield Surfaces
Yield Surface
(φ d )b
at φ d
q (φ d )a

p
Effective Normal Stress, σ' Mean Effective Stress, p
FIGURE 1  Mohr’s Circle FIGURE 2  Stress Path versus
at Yielding Stress State Radially Increasing Yield Surface

767 Beaty and Byrne


The model implicitly assumes a second yield surface, extending from the mobilized
stress ratio line down to the q=0 axis. For the sake of simplicity, this surface was
assumed to exist at relatively high confining pressures so that it is not encountered
under typical undrained stress paths. This is not unreasonable for liquefiable sands,
since their tendency to generate pore pressure will generally lead to stress paths away
from this surface. This assumption means that certain stress paths, such as those with
constant stress ratio, will never cause plastic shear or volumetric strains. Future
refinements, or complications, of the model may include this behaviour.

2.2 Hardening of Yield Surface

The stress-strain behaviour of granular soils has long been approximated with a
hyperbolic shape (Duncan and Chang, 1970; Matsuoka and Nakai, 1977). A
formulation similar to that of Duncan and Chang is used in this model, with one
fundamental difference: the hyperbola is revised to represent stress ratio versus
plastic shear strain.

The hyperbolic relation is used as the yield surface hardener in the following manner.
When an increment of plastic shear strain occurs, a corresponding increase in stress
ratio is computed using the plastic shear modulus (Figure 3a). This modulus is found
from the hyperbolic curve, the current stress ratio, and the peak friction angle (Puebla
et al., 1997). The yield surface is then rotated by ∆φ as shown in Figure 3b.

The hyperbolic shape has typically been applied to the total response, but both elastic
and plastic behaviour occur during loading. The elastic portion of this model is based
on classic linear theory, with the stiffness constants derived from small strain
observations. In accordance with elastic theory, elastic strain increments occur in the
same direction as the stress increments. The plastic portion of the response is
governed by the hyperbolic relationship, with the plastic shear strain increment
occurring in the direction of the principal shear stress. Summing the elastic and
plastic components gives a combined response which is slightly modified from a

sin(φ d +∆φ)
φ d+∆φ
= sin

p
sin(φ d ) G
Shear Stress,

φd
1
Stress Ratio,

p p
∆sinφ = ∆γ × G

Plastic Shear Strain, γ p Effective Stress, σ'


(a) (b)
FIGURE 3  Rotation of Yield Surface with Increasing Stress Ratio
(a) Hyperbolic Hardening Relationship (b) Yield Surface Rotation in Stress Space

768 Beaty and Byrne


hyperbolic shape. The amount of difference is a function of the relative stiffness of
each component, and whether the strain increments are coincident or not. In many
cases the elastic response will be much stiffer than the plastic behaviour and have
relatively little impact on the shape of the shear stress-strain response.

Defining a load increment to be a change in stress ratio has a distinct advantage: a


decrease in mean effective confining stress is correctly recognized as a loading
increment even if the shear stress remains constant. This is demonstrated in Figure 4.
When loading is based solely on shear stress, changes in mean effective stress do not
contribute to the plastic response. This effect is particularly important when modeling
the onset of liquefaction due to significant changes in pore pressure.

Using stress ratio to define loading also allows for the strain softening effect seen
during undrained loading. Even though the shear stress may peak and then drop as
the effective stress decreases, the stress ratio is observed to continuously increase.

2.3 Flow Rule

While the hyperbolic relation governs the amount of plastic shear strain, the flow rule
predicts the corresponding volumetric strain. This is critical to the simulation of
liquefaction behaviour, as it is the plastic volumetric strain that generates pore
pressure. The flow rule used in UBCSAND is based on three observations:

1. there is a unique stress ratio, defined by the constant volume friction angle φcv,
for which plastic shear strains do not cause plastic volumetric strains;
2. stress states which lie below sinφcv exhibit contractive behaviour, while stress
states above sinφcv lead to a dilative response; and,
3. the amount of contraction or dilation depends on the difference between the
current stress ratio and the stress ratio at φcv.

A flow rule which satisfies these observations has been derived from energy principles

ηB
(φ d)B
= q/p

ηA B
Shear Stress, q

qA = qB B (φ d)A A
Stress Ratio,

A
∆p ∆γ p

pB pA
Mean Effective Stress, p Plastic Shear Strain, γ p
(a) (b)
FIGURE 4  Effect of Change in Mean Effective Stress
(a) Stress Path (b) Stress Ratio versus Shear Strain

769 Beaty and Byrne


(Puebla et al., 1997). It is similar to derivations already proposed in the literature
(Rowe, 1962; Schofield and Wroth, 1968; and Matsuoka and Nakai, 1977). The
resulting equation is non-associative, and has a simple form:

∆ε pv = (sin φ cv − sin φ d )∆γ p = ( sin ψ ) ∆γ p

where ∆εvp = plastic volumetric strain increment (contraction positive) , ∆γp = plastic
shear strain increment, φcv = constant volume friction angle, φd = developed friction
angle, and ψ = dilation angle. The flow rule is shown schematically in Figure 5.

2.4 Undrained Loading

Since the model is based on the skeleton behaviour of the soil matrix, undrained
conditions are easily included through the stiffness of the pore fluid. With this
stiffness, plastic volumetric strains lead to changes in pore pressure and effective
stress. The constitutive model responds to these changes in effective stress and their
effect on the skeleton. Conditions of partial drainage can be modeled by performing a
combined mechanical-flow analysis.

2.5 Cyclic Loading

One feature needed for cyclic loading is a definition for loading and unloading. All
load increments are assumed linear elastic unless the current stress state is on the yield
surface and the load increment is in an outward direction from that surface. This is a
simplifying assumption, since some curvature is seen in unload-reload loops from
element tests. Since the model assumes linear elastic behaviour for these loops,
hysteretic damping will not occur. The anticipated damping is approximated by
adding a small viscous component to the analysis.

A second requirement is a general definition for stress reversal. This effort has been

φ d > φ cv
Plastic Strain Vectors
Dilative
p

φ d = φ cv
Plastic Shear Strain,
Shear Stress, q

Plastic Potentials
Contractive
Slope = -sinψ
φ d < φ cv

Yield Surfaces

p
Mean Effective Stress, p Plastic Volumetric Strain, εv

FIGURE 5  Flow Rule: Plastic Potentials and Strain Vectors

770 Beaty and Byrne


simplified by defining two shear stress indicators: τxy and (σy - σx). These indicators
create a stress space with four quadrants. A stress reversal is assumed to occur
whenever the stress state changes quadrants. For example, a triaxial test with the
principal axes aligned with x and y results in a stress path along the τxy = 0 axis. A
stress reversal is predicted when the test moves between compression and extension.

Since load reversals typically occur during cyclic loading, it is not always appropriate
to leave the yield surface at the highest stress ratio that was previously attained.
Consider a loading path where the stress ratio is increased to a certain value η1, then
the loading direction is reversed until a stress ratio η2 is achieved in the opposite
direction, where η2 < η1. If the loading is again reversed, it is likely that yielding will
begin at some stress ratio less than the previous maximum, η1.

To account for this, the yield surface hardener was defined to include both isotropic
and kinematic properties. This is shown in a simplified manner on Figure 6, which
depicts a stress reversal for an initial load cycle. If the stress state is in the upper
quadrant, as shown on Figure 6, and the stress ratio drops but no stress reversal
occurs (point 1 to point 2), then the yield surface in the upper quadrant remains at its
peak value (shown as “old yield surface”). Once a stress reversal occurs and yielding
begins in the lower quadrant, the new yield surface will be pushed downwards (point
3). The old yield surface in the upper quadrant is dragged downward at the same rate
the new yield surface is pushed forward in the lower quadrant. If the stresses were to
reverse again (point 3 to point 1), plastic loading would not occur until the previous
yield surface was contacted in its new position (point 4).

Another aspect which must be considered is the increase in stiffness and reduction in
volumetric strains that are observed with each load cycle during drained testing
(Martin et al., 1975). This behaviour is represented through a simple relationship

Upper Quadrant Old Yield Surface


1
Stress Path (Unloading)
Shear Stress

New Yield Surface


θ
4
2
0
θ
Stress Path (Loading)

Lower Quadrant 3 New Yield Surface

Effective Stress
FIGURE 6  Effect of Stress Reversal on Yield Surface

771 Beaty and Byrne


between accumulated plastic volumetric strain and plastic shear modulus, which has
been calibrated to test results. The plastic shear modulus is increased as a function of
the total plastic volumetric strain, which in turn reduces the rate of volumetric strain.

2.6 Anisotropy

Anisotropy is included by making the plastic shear modulus a function of the principal
stress direction. For example, a softened modulus can be used when the major
principal stress direction is horizontal, as in an extension test. The modulus is varied
between the extension and compression directions to match observed behaviour.

3. Application to Monotonic Loading

To demonstrate the utility of the constitutive model, a simulation was made of


undrained monotonic tests performed on loose sand. The analyses were performed
using the finite difference program FLAC (Cundall, 1995). FLAC uses an explicit
solution technique, maintaining dynamic equilibrium at every timestep.

The laboratory tests were performed on samples of Syncrude sand obtained near Fort
McMurray, Alberta (Vaid et al., 1995). Triaxial compression, extension, and simple
shear tests were completed. A brief comparison of the test results and simulations are
shown in Figure 7. The analyses were not performed independently, as each test was
modeled using the same material parameters (Puebla et al., 1997). Very good
agreement was achieved between the laboratory tests and the simulations. The model

p i = 100 kPa Compression p i = 100 kPa


(kPa)

80
Horiz. Shear Stress, xy (kPa)

40
Pore Pressure, u (kPa)
3

60
- 1

20
Deviator Stress,

40

0
20
Extension
-20 0
-5 0 5 0 1 2
Axial Strain, ε1 (%) Shear Strain, γ xy (%)
Model Simulation Model Simulation
Laboratory Results (TxC) Laboratory Pore Pressure
Laboratory Results (TxE) Laboratory Shear Stress
(a) (b)
FIGURE 7  Monotonic Response and Simulation of Syncrude Sand
(a) Undrained Triaxial Extension and Compression (b) Undrained Simple Shear

772 Beaty and Byrne


adequately represents the observed behaviour, including the effects of anisotropy,
strain softening, and pore pressure generation.

A total of 10 model parameters are required for these simulations. The variables
primarily define the elastic moduli, strength, and hyperbolic properties of the material.
A detailed description of these parameters is given by Puebla et al. (1997).

4. Application to Cyclic Loading

A similar comparison was made to laboratory results from cyclic simple shear tests
(Finn, 1985). The tests were performed on samples of Ottawa Sand having a relative
density of approximately 45%. Both drained and undrained tests were performed,
each having an initial vertical effective stress of 200 kPa. Undrained behaviour was
imposed in the laboratory by running a constant volume test with dry sand. The
undrained response in the numerical model was simulated in the same manner. This
technique essentially assumes an infinite stiffness for the pore fluid. Results are
summarized in Figure 8.

The same model parameters were used for both the drained and undrained loading.
The general character of the these simulations is in good agreement with the
observations, although there are some differences in specific details (e.g., the shape of
the cycles versus volumetric strain curve, the area of the drained stress-strain loops).
Some of the difference is due to limitations in the laboratory testing: the effect of
friction within the test equipment is clearly apparent in Figure 8[e]. The simulations
should improve with further refinement of the model and additional calibration of the
input parameters.

5. Wildlife Site

A preliminary application of the model was made to the observed field behaviour at
the Wildlife site. This site, located within the Imperial Valley Wildfowl Management
Area near the southern border of California, has been well instrumented and studied
by the United States Geological Survey (Bennett et al., 1984). Sand boils have been
noted at the site following several earthquakes, including the 1981 Westmorland
event. Instrumentation was installed by USGS to record and study any future events.
Such an occurrence took place during the Superstition Hills earthquake of 1987. This
earthquake had a moment magnitude of 6.6 and an epicentral distance of about 31 km
(Zeghal and Elgamal, 1994). Evidence of liquefaction was found at the site and in the
instrument recordings.

The stratigraphy of the site was characterized by the USGS, and is summarized in
Table 1 for the depths of interest. Unit B was identified as liquefiable and was likely

773 Beaty and Byrne


60 Simulation Simulation
(kPa) Test Results 0.5 Test Results
40

Volumetric Strain (%)


0.4
20
xy
Horiz. Shear Stress,

0.3
0

-20 0.2

-40 0.1
[a] [b]
-60 0
-0.2 -0.1 0 0.1 0.2 -0.2 -0.1 0 0.1 0.2
Shear Strain, γ xy (%) Shear Strain, γ xy (%)

1.5
Volumetric Strain (%)

[c]
1
Simulation
Test Data
0.5

0
0 5 10 15 20 25
Number of Cycles

40 40
Simulation Test Results
30 30
(kPa)

(kPa)

20 20
xy

xy

10 10
Horiz. Shear Stress,

Horiz. Shear Stress,

0 0

-10 -10

-20 -20

-30 [d] -30 [e]

-40 -40
-0.2 -0.1 0 0.1 0.2 -0.2 -0.1 0 0.1 0.2
Shear Strain, γ xy (%) Shear Strain, γ xy (%)

FIGURE 8  Comparison of Simulated and Laboratory Cyclic Simple Shear


[a], [b], [c] Drained [d], [e] Undrained

774 Beaty and Byrne


TABLE 1  Soil Stratigraphy at Wildlife Site
Unit Depth (m) Brief Description
A 0-2.5 Very loose micaceous sandy silt, silt, and clayey silt
B1 2.5-3.5 Very loose to loose sandy silt
B2 3.5-6.8 Loose to medium dense silty sand to very fine sand
C1 6.8-7.5 Medium to stiff clayey silt
C2 7.5-12 Medium to very stiff silty clay

responsible for the observed sand boils. Portions of Unit A might also liquefy. The
installed instrumentation included a surface accelerometer, a downhole accelerometer
at a depth of 7.5 meters, and a suite of 6 piezometers (2 located within the depth
range of unit B1, 3 within unit B2, and 1 at a depth of about 12 meters).

A finite difference model was developed to perform the simulation. The model
consisted of a 1-dimensional column of 7.5 meter depth. The recorded downhole
motion was applied to the model base. The north-south component was used as this
direction experienced the higher peak acceleration. Material parameters were
selected to be consistent with the blowcount values. A brief description of the model
is given in Table 2. Results of the simulation are shown in Figure 9.

TABLE 2  Description of Finite Difference Model


Approximate # of Elements
Unit Depth (m) (N1)60 in Unit
A 0-2.5  4
B1 2.5-3.5 4-6 2
B2 3.5-6.8 10 - 12 6
C1 6.8-7.5  1

The comparison between the observed and simulated response provides a mixed
appraisal of the model. The relative displacement plot indicates that liquefaction was
predicted to occur at approximately the same time as was recorded. The velocity
estimates prior to liquefaction are also quite close to the recorded values. However,
the post liquefaction behaviour is substantially different than observed, especially in
the amplitude of the motion. Although the fundamental response of the model prior
to liquefaction appears to be reasonable, further refinement is needed before it can be
confidently applied to complex loading situations.

775 Beaty and Byrne


0.2
From USGS Recordings
Relative Displac. (m)
0.1

-0.1
Simulation
-0.2
0 5 10 15 20 25 30 35 40 45 50
Time (sec)

0.4
From USGS Recording
Surface Velocity (m/s)

0.3
0.2
0.1
0
-0.1
Simulation
-0.2
-0.3
0 5 10 15 20 25 30 35 40 45 50
Time (sec)
FIGURE 9  Comparison of Simulated and Recorded Response at Wildlife Site

6. Conclusions

An effective stress constitutive model was presented that can be used to simulate the
liquefaction behaviour of sand. The framework of the model was described, and
shown to be relatively simple and fundamental. The model has been applied to
situations of monotonic loading with good results. The current extension to cyclic
loading also shows promise, although further development is needed to capture the
intricacies of observed field behaviour.

7. Acknowledgments

The authors appreciate the substantial contribution of Humberto Puebla, who


provided some of the analyses, assisted with several figures, and has been
instrumental in implementing the constitutive model.

8. References

Bennett, M.J., McLaughlin, P.V., Sarmiento, J.S., and Youd, T.L. (1984).
Geotechnical Investigation of Liquefaction Sites, Imperial Valley, California. Open-
File Report 84-252, U.S. Geological Survey, Menlo Park, California.

776 Beaty and Byrne


Byrne, P.M., Debasis, R., Campanella, R.G., and Hughes, J. (1995). Predicting
Liquefaction Response of Granular Soils from Self-Boring Pressuremeter Tests.
ASCE National Convention, San Diego, Oct. 23-27, ASCE, 56(GSP): 122-135.

Cundall, P.A. (1995). FLAC Manual Version 3.3. ITASCA Consulting Group, Inc.,
Thrasher Square East, 708 South Third Street, Suite 310, Minneapolis, Minnesota.

Duncan, J.M. and Chang, C.Y. (1970). Nonlinear Analysis of Stress and Strain in
Soils. Journal of the Soil Mechanics and Foundations Division, ASCE,
96(SM5):1629-1653.

Finn, W.D.L. (1985). Aspects of Constant Volume Cyclic Simple Shear. Advances
in the Art of Testing Soils Under Cyclic Conditions, ASCE Technical Publication,
October 1985, pp. 74-98.

Martin, G.R., Finn, W.D.L., and Seed, H.B. (1975). Fundamentals of Liquefaction
under Cyclic Loading. Journal of the Geotechnical Engineering Division, ASCE,
Vol. 101, No. GT5, May.

Matsuoka, H., and Nakai, T. (1977). Stress-Strain Relationship of Soil Based on the
SMP. Proceedings, Specialty Session 9, 9th International Conference on Soil
Mechanics and Foundation Engineering, pp. 153-162.

Puebla, H., Byrne, P.M., and Phillips, R. (1997). “Analysis of CANLEX Liquefaction
Embankments: Prototype and Centrifuge Models.” Canadian Geotechnical Journal,
Vol. 34, No. 5, pp. 641-657..

Rowe, P.W. (1962). The Stress-Dilatancy Relation for Static Equilibrium of an


Assembly of Particles in Contact. Proceedings of the Royal Society of London,
Mathematical and Physical Sciences, Series A, 269:500-57.

Schofield, A., and Wroth, P. (1968). Critical State Soil Mechanics. McGraw-Hill
Publishing Company Limited, England.

Vaid, Y.P., Sivathayalan, S., Uthayakumar, M., and Eliadorani, A. (1995).


Liquefaction Potential of Reconstituted Syncrude Sand. Proceedings, 48th Canadian
Geotechnical Conference, Vancouver, B.C., Sept. 25-27, 1995, Vol. 1, pp. 319-329.

Zeghal, M. and Elgamal, A. (1994). Analysis of Site Liquefaction using Earthquake


Records. Journal of the Geotechnical Engineering Division, ASCE, Vol. 120, No.
6, June.

777 Beaty and Byrne


Key Words (10)

constitutive model
dilation
elastoplastic
hyperbolic
liquefaction
plasticity
strain softening
stress ratio
UBCSAND
Wildlife site

Byrne and Beaty

You might also like