Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Research Article

Transportation Research Record


1–6
Ó National Academy of Sciences:
Soil Parameters for Design with the 3D Transportation Research Board 2019
Article reuse guidelines:
PLAXIS Hardening Soil Model sagepub.com/journals-permissions
DOI: 10.1177/0361198119851723
journals.sagepub.com/home/trr

Jeremy T. Bowers1, Mark C. Webb1, and Jesse L. Beaver1

Abstract
The design and analysis of buried structures presents difficulties that cannot often be solved by closed-form solutions. Finite
element methods (FEM) have increasingly become the tool of choice for advanced soil-structure interaction analysis, with
three-dimensional FEM being required for irregular non-plane-strain cases. To accurately capture the stress and deformation
of soils, complex material constitutive models are required. Several input parameters to these models must be deter-
mined from expensive soil testing, which is impractical for most applications. For two-dimensional FEM, good approxima-
tions of these parameters for a wide range of placed backfill soils have been developed and used in practice for many
years in the computer program CANDE. It is the purpose of this paper to take these parameters, developed by Selig for
use in CANDE, and convert them to equivalent parameters for the three-dimensional PLAXIS computer program’s
Hardening Soil model.

The analysis of complex three-dimensional buried struc- Methodology


tures has increasingly become a task accomplished
through soil-structure interaction finite element analysis. The 2D Duncan–Chang (DC) model, the 2D CANDE
In the interaction between soils and structures, the soils Duncan–Selig model, and the 3D PLAXIS HS model
respond in a highly nonlinear manner. The hyperbolic share several common input parameters, as described in
stress-strain model (1, 2) is widely accepted as a good Table 1. Below, we develop the relationships between
approximation of the behavior of coarse-grained, these parameters, with preference given to matching
drained, low moisture sensitivity soils. However, these triaxial behavior over bulk or oedometric behavior.
models require complex input from rigorous soil testing,
which is impractical for design purposes. Selig has pro-
posed typical parameters for a wide range of placed Stiffness Modulus for Primary Triaxial Loading
backfill soil types and compaction levels, which are com- The triaxial stiffness parameters for the HS model (7) as
monly used in the Duncan–Chang (DC) model with the implemented in PLAXIS 3D (6) can be analytically
Selig hyperbolic bulk modulus formulation (called determined from the values provided by Selig (3). The
the DS model here) (3), particularly as implemented in Duncan–Selig model is substantially similar to the DC
the finite element computer program CANDE (4) and model (2), from which the HS model is derived, using the
described in the formulations manual (5). The DS model same formulation for the Young’s modulus but differing
is limited to two-dimensional (2D) analyses, whereas the in the bulk modulus formulation, as described in the fol-
progression of processing power has allowed for the lowing section.
more frequent use of three-dimensional (3D) analyses for From the HS model, the deviatoric stress, q, as a func-
irregular and non-plane-strain cases. Therefore, modern tion of the axial strain, e1, is of the same form as the DS
3D models have become more widely available for use in model and is defined as:
both design and analysis. One such model is the 3D
PLAXIS Hardening Soil (HS) model (6), which shares
many similarities with the DS model. Here, the para- 1
Simpson Gumpertz & Heger Inc, Waltham, MA
meters put forth by Selig to be used with the DS model
are used to calibrate analogous parameters for the HS Corresponding Author:
model. Address correspondence to Jeremy T. Bowers: jtbowers@sgh.com
2 Transportation Research Record 00(0)

Table 1. Primary Input Parameters to Constitutive Models

Parameter description Duncan–Chang model Duncan–Selig model Hardening soil model

Reference tangent Young’s modulus K K na


Reference secant Young’s modulus na na Eref
50
Exponent for calculation of Young’s modulus N n M
Reference tangent bulk modulus Kb Bi na
Reference tangent oedometer modulus na na Eref
oed
Ultimate volumetric strain na eu na
Exponent for calculation of bulk modulus P na na
Exponent for calculation of oedometer modulus na* na M
Reference unloading/reloading modulus na* na Eref
ur
Reference confining pressure Pa Pa pref
Cohesion C C C
Reference friction angle u0 u0 F
Change in friction angle per atmosphere of confining pressure Du Du na
Dilatancy angle na na c
Failure ratio of deviator stress Rf Rf Rf

Note: na = not applicable.

1 q e1 2  Rf
e1 ¼ ! q ¼ 1 e1 ð1Þ E50 = Ei ð6Þ
Ei 1  q=qa Ei þ qa
2

where Thus, for low levels of cohesion (such as for granular


ref
Ei = tangent Young’s modulus when q = 0, materials), the reference secant modulus, E50 , with a ref-
ref
q = s1  s3 , and erence pressure, p , equal to atmospheric, can be
the deviatoric stress asymptote, qa, is related to the devia- approximated by substituting Equation 4 into Equation
toric stress at failure, qf, by the failure ratio, Rf, according 6 and assuming c = 0, yielding Equation 7.
to Equation 2, with qf as defined in Equation 3 in relation
ref 2  Rf
to cohesion, c, friction angle, u, and confining pressure, E50 = KPa ð7Þ
s3, per the DS and HS models, respectively. 2

qf When not considering dilatancy, Equation 7 provides a


qa = ð2Þ good approximation of the triaxial behavior as com-
Rf
pared to the DS model. For soils with a friction angle
2c cosu  2s3 sinu 2 sinu greater than 30°, dilatancy is generally approximated as
qf = = ðc cotu  s3 Þ c = u–30° ø 0°. This results in a stiffer response to
1  sinu 1  sinu
ð3Þ triaxial loading, but has negligible effect on the strength.
For unloading and reloading, the HS introduces a linear
Equations 1–3 show general agreement between the modulus, Eur, with a calculation of the same form as E50.
two models, however the calculation of the initial This implementation is similar to that suggested by Duncan
tangent Young’s modulus, Ei, differs between the models. et al. (8). They suggest that the unloading-reloading stiff-
The DS model calculates the modulus in relation to the ness should be proportional to the initial tangent Young’s
confining pressure normalized by atmospheric pressure, Pa: modulus, Ei, with dense sands being 1.2 times as large and
loose soils being up to 3 times as large as the primary load-
Ei = KPa (s3 =Pa )n ð4Þ ing stiffness. Katona modified the DS model to consider a
similar calculation of the unloading/reloading modulus,
Whereas the HS model modifies this normalized suggesting that Eur should be equal to Ei (9). The PLAXIS
confining pressure by c and u and operates in relation to
Material Models Manual (6) suggests an acceptable default
the secant modulus measured at 50% of the failure ref
reference unloading-reloading stiffness of 3E50 .
strength, E50:
 m
ref c cosu  s3 sinu Oedometric Stiffness Modulus
E50 = E50 ð5Þ
c cosu + pref sinu
The oedometer stiffness, Eoed, or constrained modulus
where for the HS model replaces the bulk modulus, B, input in
Bowers et al 3

Table 2. Soil Parameters for 3D PLAXIS Hardening Soil Model

C j Dj c
Soil type Relative density Eref
50 (psi) Eref
oed (psi) m Rf (psi) (deg) (deg) (deg) K0* Knc
0 ** Kb

SW 95 9,077 3,585 0.6 0.7 0 48 8 18 1.3 0.21 250


90 5,880 2,165 0.43 0.75 0 42 4 12 1.1 0.25 151
85 3,969 1,366 0.35 0.8 0 38 2 8 0.9 0.25 90
80 2,752 1,129 0.35 0.83 0 36 1 6 0.8 0.32 72
60 437 428 0.85 0.9 0 29 0 0 0.5 0.52 35
ML 95 3,396 1,535 0.4 0.95 4 34 0 4 1.2 0.40 110
90 1,632 850 0.26 0.89 3.5 32 0 2 0.9 0.43 55
85 930 561 0.25 0.85 3 30 0 0 0.8 0.50 35
80 662 463 0.25 0.8 2.5 28 0 0 0.7 0.53 28
50 171 171 0.95 0.55 0 23 0 0 0.5 0.61 15
CL 95 882 767 0.45 1 9 15 4 0 0.8 0.71 50
90 584 516 0.54 0.94 7 17 7 0 0.6 0.65 34
85 404 364 0.6 0.9 6 18 8 0 0.5 0.63 25
80 291 276 0.66 0.87 5 19 8.5 0 0.4 0.61 21
45 147 132 0.95 0.75 0 23 11 0 0.3 0.53 15

Note: SW = coarse-grained gravelly sand; ML = sandy silt; CL = silty clay; psi = pounds per square inch.
*
Used to generate Eref
oed , from Selig (3)
**
Input to PLAXIS, selected to allow for use of desired Eref
oed

the DS model. Both are intended to describe the stress Kb, Selig’s formulation incorporates a hyperbolic equa-
response of the soil to changes in volumetric stain. tion based on the mean stress, sm, the initial tangent bulk
Updating the constitutive relationship describing the modulus, Bi, and the ultimate volumetric strain, eu, to
bulk modulus was the primary change that Selig (3) determine the bulk modulus.
implemented to the DC model (2). Thus, this portion of
the calibration is less precise than that for the triaxial B = Bi (1 + sm =ðBi eu Þ)2 ð10Þ
stiffness. Additionally, the DC model allowed for inde-
pendent definition of the exponents input to the power Although Selig’s formulation for bulk modulus was
law models for Young’s modulus (n) and bulk modulus deemed to be more accurate, the DC power law model
(p). The HS restricts the inputs to a single exponent that (Equation 11) is more similar in form to the HS oed-
is shared between the two models (m). ometer stiffness formulation. To provide a better com-
The bulk modulus relates the mean stress, sm, to the parison to the HS model, the power law model is used
volumetric strain, ev, through the approximate relation: here along with Selig’s ‘‘Curve A’’ calibrated values for
the stiffness parameter, Kb, provided in Table 2. When
Dsm Ds1 + Ds2 + Ds3 determining the oedometer stiffness values for HS, the
B= = ð8Þ Curve A results were found to provide a better fit to the
Dev 3Dev
DS model.
For an oedometer test, in which the movement of the
sides of the test specimen is restricted and a vertical load B = Kb Pa (s3 =Pa )p ð11Þ
is applied, the components of stress are related to each
other by the lateral earth pressure coefficient, K0, such The HS model, using Eoed instead of B to define the volu-
that s3 = s2 = K0s1, for which suggested values for K0 metric response of the soil, determines the oedometer
have been provided by Selig and are given in Table 2. stiffness according to:
The volumetric strain is also found to be equal to the ver-  s3 m
tical (first principal) strain, e1. Therefore, the bulk modu- ref
c cosu  K 0
sinu
Eoed = Eoed ð12Þ
lus can be rewritten in relation to the first principal stress c cosu + pref sinu
and strain as:
It is noted that the oedometer stiffness for the HS
Ds1 ð1 + 2K0 Þ ð1 + 2K0 Þ model is essentially calculated using s1 rather than s3.
B= = Eoed ð9Þ For low levels of cohesion, the reference tangent modu-
De1 3 3 ref
lus, Eoed , with a reference pressure equal to atmospheric,
Rather than using the DC power law model based on can be approximated by substituting Equations 11 and
the third principal stress, s3, and the stiffness parameter, 12 into Equation 9 and assuming c = 0. Additionally,
4 Transportation Research Record 00(0)

Figure 1. Comparison of SW soil response in HS and DS soils.

accounting for the conversion from s1 to s3, K0 to the representative confinement level should be selected and
power m is included to produce the conversion Equation the effective friction angle calculated as:
13. Owing to the constraints of the HS model, the expo-
nent m is used rather than p. u = u0  Dulog(s3 =Pa ) ð15Þ

3Kb Pa The dilatancy angle is not included in the DS model and


ref
Eoed = ðK0 Þm ð13Þ affects the relationship of the axial and volumetric
ð1 + 2K0 Þ
strains. Though various correlations exist for the calcula-
Equation 13 provides a decent approximation of the tion of c based on minerology and confinement (e.g.,
volumetric response of the soil as compared to the DS 10), the lack of specific information and a desire for more
model, particularly for well-compacted soils. For soils general applicability disallow the use of these correla-
with relatively low ratios of bulk modulus to Young’s tions. The PLAXIS Material Models Manual (6) suggests
modulus, the calibrated parameters suggested by Selig for quartz sands, c = u–30° gives a rough approxima-
would lead to low oedometer stiffness values for the HS tion. Comparing this approximation against the correla-
model. The combination of these stiffness values with the tion Bolton suggests for quartz and feldspar shows that
input lateral earth pressure coefficient, K0nc , are typically u–30° is conservative, yielding lower values for c (10).
outside the range of what is allowed by the HS model, as
described in the PLAXIS Materials Models Manual (6). Findings
ref
To maintain Eoed as calculated in Equation 13, K0nc is
decreased to fit within the allowable range of the HS The constitutive relationships of the Duncan–Selig and
model. Note that K0nc , which generally follows Jaky’s the 3D PLAXIS HS models are compared below for the
equation, Equation 14, for normally consolidated soils, is various soil types presented by Selig (3). The values for
ref ref
not necessarily the same as K0, which was taken from E50 , Eoed , and u given in Equations 7, 13, and 15 are
used along with the other parameters provided by Selig
Selig to provide the best fit.
as input to the HS model. These values are provided in
K0nc = 1  sinðuÞ ð14Þ Table 2. Built into the HS model are certain constraints
on acceptable combinations of the above parameters. To
maintain the accuracy of triaxial behavior and strength
and the correlation to values published for the DS model,
Strength Parameters the input lateral earth pressure coefficients have been
The HS model describes the soil strength in relation to decreased from the suggested values based on Jaky’s
friction angle, u, cohesion, c, and dilatancy angle, c. It equation in certain cases to conform to the allowable
does not include a parameter to account for the change range.
of friction angle with confining pressure Du, described Figures 1 to 3 provide a comparison of the DS model
by Duncan et al. (8). To best match the DS model, a using Selig’s original parameters (3) to the HS model
Bowers et al 5

Figure 2. Comparison of ML soil response in HS and DS soils.

Figure 3. Comparison of CL soil response in HS and DS soils.

using the parameters listed here. The data for the HS Bi evol
sm = ð16Þ
model were generated using the PLAXIS SoilTest 1  ðevol =eu Þ
Module for triaxial and oedometer tests. The data for
the DS model were calculated using Equations 1 and 16. The plots of the hydrostatic compression tests show a
Plots are given for tests of coarse-grained gravelly sand decent comparison between the models for well-
(SW), sandy silt (ML), and silty clay (CL) and use a fric- compacted soils, but the HS model overpredicts the bulk
tion angle based on 5 pounds per square inch (psi) con- stiffness in some cases, such as in poorly compacted soils.
fining pressure. The plots of the drained triaxial test For these poorly compacted soils, the constraints that
show a good comparison between the two models, par- PLAXIS enforces on the material parameters prevent
ticularly for the SW and ML soils. Note that for some of significant improvement on the values calculated here.
ref
the SW and ML soils, the dilatancy angle results in a stif- For example, for SW80, if E50 is 2,752 psi, the minimum
nc ref
fer response in triaxial loading. value for K0 is 0.254 and for Eoed is 888 psi. This
6 Transportation Research Record 00(0)

and to verify that the differences between the HS and DS


models at the constitutive level do not result in signifi-
cant differences at the global or system level. The devel-
opment of the HS parameters extends the usefulness of
Selig’s original calibration into 3D finite element analy-
ses and allows for the design of more complex buried
structures such as buried bridges and culverts with irre-
gular geometry.

Author Contributions
The authors confirm contribution to the paper as follows: study
conception and design: JLB, MCW, JTB; data collection, anal-
ysis and interpretation of results, draft manuscript preparation:
JTB. All authors reviewed the results and approved the final
version of the manuscript.

References
1. Kondner, R. L. A Hyperbolic Stress Strain Formulation
for Sands. Proc., 2nd Pan American Conference on Soil
Mechanics and Foundation Engineering, Brazil, Vol. 1,
1963, pp. 289–324.
2. Duncan, J. M., and C. Y. Chang. Nonlinear Analysis of
Stress and Strain in Soils. Journal of the Soil Mechanics
Figure 4. Comparison of SW80 bulk behavior in DS model and Foundations Division, Vol. 96, No. SM5, 1970,
against HS model using various values of Eref
oed . pp. 1629–1653.
3. Selig, E. T. Soil Parameters for Design of Buried Pipelines.
combination of parameters represents the closest fit that Proc., Pipeline Infrastructure Conference, ASCE, Univer-
the HS model can achieve in relation to the DS model sity of Massachusetts, Amherst, 1988, pp. 99–116.
for bulk behavior, if triaxial behavior is to be prioritized. 4. Katona, M. G., J. M. Smith, R. S. Odello, and
J. R. Allgood. CANDE – A Modern Approach for Struc-
Figure 4 provides a comparison of the DS model and the
tural Design and Analysis of Buried Culverts. Publication
HS model using the oedometer stiffness from Table 2
FHWA-RD-77-5. FHWA, Naval Civil Engineering
and the oedometer stiffness noted above. The figure Laboratory, Washington, D.C., 1976.
demonstrates that using the most extreme allowed mate- 5. Katona, M. G., M. Mlynarski, and T. J. McGrath.
rial parameters in the HS model results in negligible ben- CANDE-2007 Culvert Analysis and Design Solution Meth-
efit to capturing the behavior of the DS model. If bulk ods and Formulations. Publication NCHRP 15-28. U.S.
stiffness behavior is the primary mode of concern, care Department of Transportation, Washington, D.C., 2008.
should be taken to recalibrate the HS parameters by 6. PLAXIS 3D [Computer Software]. https://www.plaxis.
ref com. Accessed 29 July 2018.
reducing E50 . The heightened bulk stiffnesses presented
here may lead to an overprediction of lateral soil support 7. Schanz, T., P. A. Vermeer, and P. G. Bonnier. The Hard-
to structures embedded in poorly compacted soils. ening Soil Model: Formulation and Verification. Presented
at Beyond 2000 in Computational Geotechnics – 10 Years
Further analyses should be conducted to ensure that this
of PLAXIS, Balkema, Rotterdam, 1999.
does not have a significant impact on global behavior. 8. Duncan, J. M., P. Byrne, K. S. Wong, and P. Mabry.
Strength, Stress-Strain, and Bulk Modulus Parameters for
Finite Element Analyses of Stresses and Movements in Soil
Conclusion Masses. Report No. UCB/GT/80-01. University of Cali-
We have shown that the 3D PLAXIS HS model main- fornia, Berkeley, 1980.
tains a close relation to the Duncan–Selig model, for 9. Katona, M. G. Modifying Duncan-Selig Soil Model for
Plastic-Like Behavior. Transportation Research Record:
which acceptable parameters exist for a range of soil
Journal of the Transportation Research Board, 2015. 2511:
types and compaction levels (3). We selected analogous 53–62.
parameters for the HS model and compared responses in 10. Bolton, M. D. The Strength Dilatancy of Sands. Géotechni-
drained triaxial and hydrostatic compression tests. que, Vol. 36, No. 1, 1986, pp. 65–78.
Further research should be conducted to compare the
performance of 2D and 3D FEM against well- The Standing Committee on Subsurface Soil-Structure
documented field test data to validate the use thereof Interaction (AFS40) peer-reviewed this paper (19-04758).

You might also like