12th International LS-DYNA Users Conference FSI/ALE(2)
Effect of Soil Material Models on SPH Simulations for Soil-
Structure Interaction
Ronald F. Kulak1
Len Schwer2
1) RFK Engineering Mechanics Consultants
307 Warwick Drive
Naperville, IL 60565, USA
[email protected] 2) Schwer Engineering & Consulting Services
6122 Aaron Court
Windsor CA 95492-8651
[email protected] Abstract
The analysis of bridges and highway structures often involves the interaction between the structural components and
the soil they are embedded in. Modeling and simulation is currently capable of treating soil-structure interaction
problems in which large soil and structural deformations occur. This type of interaction occurs in the analysis of
bridge pier stability during riverbed scour. With the use of parallel computers, large-scale problems can routinely
be solved. For these nonlinear problems, two of the critical steps are the choice of the constitutive models used to
represent the components and the parameters for the models. Mature computer codes such as LS-DYNA have a
large number of material models to choose from. So it is important to know the difference in prognostic capability of
each model. Predictive accuracy cannot be known, only estimated.
This paper presents material parameters for the following models used to represent soil: (1) MAT005, Soil and
Crushable Foam; (2) MAT010, Elastic Plastic Hydro; (3) MAT025, Geological Cap; and (4) MAT079, Hysteretic
Soil. The hydrostatic compression response predicted by each model is compared to experimental data.
The material models were used in the large deformation analysis of a rectangular platen that was pushed into sand
at a rate of 2.54 cm/sec (1 in/sec). The vertical resistance force versus penetration distance was found to be
nonlinear during initial penetration and then transition into a linear response. A hybrid approach that combined
Lagrange and Smoothed Particle Hydrodynamic (SPH) method was used to represent the soil.
Introduction
Currently, civil engineering design practice uses liner elastic soil properties for many soil-
structure interaction problems. However, this approach falls short for problems in which the
structure and soil undergo large deformations. With the availability of the high-performance
computing (HPC) cluster at the United States Department of Transportation’s Transportation
Research and Analysis Computing Center [1], transportation researchers can investigate complex
soil-structure interaction problems.
1-1
FSI/ALE(2) 12th International LS-DYNA Users Conference
One of these problems is the stability of bridge piers during flash floods [2]. For certain riverbed
soils, the high-velocity water washes away the soil covering the bridge piers, and this scour
action can eventually expose the bottom of the pier and footing. In order to simulate this
complex behavior, it is necessary to model the response of the reinforced concrete column –
including concrete material failure – and the nonlinear soil behavior.
This paper addresses some of the issues related to soil modeling. The following four material
models were studied: (1) MAT005, Soil and Crushable Foam; (2) MAT010, Elastic Plastic
Hydro; (3) MAT025, Geological Cap; and (4) MAT079, Hysteretic Soil. The first step in
modeling the soil is to choose an appropriate model. The second step is to obtain the material
parameters required by the specific model chosen. Some basic models require only a few
parameters while the more complex models require many more. For site specific analysis, soil
testing is required and then skilled analysts extract the needed parameters. The parameters
obtained from soil test are given for the above four soil models. The hydrostatic compression
response predicted by each model is compared to experimental data. The MAT005 and MAT025
materials were used in a three-dimensional SPH simulation of a rigid platen being pushed into
sand.
Material Models and Parameters
From the large suite of material models available to the LS-DYNA user, the following four
models were selected: (1) MAT005, Soil and Crushable Foam; (2) MAT010, Elastic Plastic
Hydro; (3) MAT025, Geological Cap; and (4) MAT079, Hysteretic Soil. The reasons for
selecting these four were they encompass a good range from fairly simple to complex and soil
test data was available, which allowed calibration of the material parameters. Descriptions of
these models are not presented here. Note, all of these models are two invariant models.
Standard soil tests for determining mechanical properties include the hydrostatic compression
(HSC), triaxial compression (TXC), triaxial extension (TXE), unconfined compression (UCT)
and uniaxial strain (UXE). The standard test specimen for soil is a right circular cylinder. The
test is performed by applying axial (σ1) and lateral stresses (σ2, σ3) to the cylinder. The
hydrostatic compression test is performed to determine the compaction behavior of soil. The test
is performed by having all three principle stresses equal:
(1)
Triaxial compression tests are used to define the shear strength envelope of the soil. Unlike the
hydrostatic compression test, the axial and lateral stresses are not equal, and thus, the cylindrical
test specimen experiences shear stress, which is equal to the difference between the principal
stresses and is denoted either by SD or :
(2)
The specimen is loaded to a predetermined pressure under hydrostatic compression conditions,
and then the lateral stress is held constant and the axial stress is increased. The unconfined
compression test is performed by having a zero lateral stress. This is the lowest estimate of the
material strength.
1-2
12th International LS-DYNA Users Conference FSI/ALE(2)
The triaxial extension test is performed in a similar manner. First, the specimen is loaded to a
predetermined pressure under hydrostatic compression conditions, and then the axial stress is
held constant and the lateral stress is increased. For soils the uniaxial strain test is usually
performed by placing the soil in a “rigid” cylinder to prevent lateral strain.
Table 1 below list the standard soil tests required by each material model.
Table 1: Soil tests required by the material models
Test Symbol Material Model
005 010 025 079
Hydrostatic Compression HSC
Triaxial Compression TXC
Triaxial Extension TXE
Unconfined Compression UCT
Uniaxial Strain UXE
Using the data from the soil tests, parameters for each model were obtained previously. The
parameters used for each model and their values are listed in Table 2 below. The units were in
the gm-MPa-mm-ms system. It is seen that MAT005 required four parameters, and MAT025
required eight. Note, these models have more parameters available than shown, but the values for
these missing parameters were chosen to be zero. The definition of each parameter can be found
in the LS-DYNA user’s manual [3].
Table 2: Parameter values
Mat 005 MAT 010 MAT 025 MAT 079
Parameter Value Parameter Value Parameter Value Parameter Value
density 1.64e-3 density 1.64e-3 density 1.64e-3 density 1.64e-3
G 136 G 136 Bulk 15000 k0 68.6
kun 4700 PC -0.51 G 5289 p0 -0.0138
a2 0.3736 a1 1.0578 Theta 0.20375 b 0.39
R 2.3 a2 0.373
D 1.6e-3
W 0.49
X0 46.5
Comparison to Hydrostatic Compression Test
With the parameters determined for each material model, a one-hexahedral-element model was
run to simulate hydrostatic compression. Figure 1 shows the results for the four material models
up to 32 percent volumetric strain. It should be noted that the results for MAT005 and MAT010
overlay each other. Furthermore, both MAT005 and MAT010 replicate exactly the experimental
values since the experiment data for the hydrostatic compression test is part of the input and used
directly by these two models. MAT025 over predicts the experimental response and over predicts
1-3
FSI/ALE(2) 12th International LS-DYNA Users Conference
the response of MAT079 up to about 16% volumetric strain. MAT079 under predicts the
response.
Figure 1: Comparison of Model Response to Hydrostatic Compression up to 32 percent volume strain
Figure 2 compares the model responses up to a volumetric strain of 10 percent. At about the 10
percent natural volumetric strain, MAT025b shows a pressure about 80 percent larger than
MAT005.
Figure 2: Comparison of Model Responses to Hydrostatic Compression up to 10 percent natural volumetric
strain
Simulation of Platen Penetration into Soil
To assess the performance of the models, the platen penetration example reported by Schwer [4]
was exercised using MAT005 and MAT025 material models. The example consisted of pressing
1-4
12th International LS-DYNA Users Conference FSI/ALE(2)
a loading platen into soil. The numerical model represented one-quarter of the physical problem:
2439 mm long, 2439 mm wide and 1778 mm deep Figure 3a. The platen was a square with 508
mm sides. Symmetry boundary conditions were applied to the two inside surfaces of the model
to simulate the whole model. The translational degrees-of-freedom were constrained on the
bottom and outer surfaces of the model. To decrease the large computational time required by the
Smooth Particle Hydrodynamics formulation, a hybrid model was developed. For the hybrid
model, the Smooth Particle Hydrodynamics formulation was used in the high deformation region
and the Lagrangian formulation in the small deformation region. The quarter model 508 mm
long by 508 mm wide by 889 mm deep SPH zone was under the platen Figure 3b.
Figure 3: Hybrid model for platen penetration simulation
Because this is considered to be a quasi-static loading soil penetration problem, strain-rate
sensitivity is not a consideration. Also, since explicit time integration can easily handle large
deformation soil-structure interaction problems in which highly nonlinear material response
occurs, the explicit option of LS-DYNA was chosen as the solver. Figure 4 shows the deformed
configuration and contours of the vertical stress under the loading platen.
Figure 4: Deformed configuration for platen penetration into soil
1-5
FSI/ALE(2) 12th International LS-DYNA Users Conference
The platen penetration force for MAT005 and MAT025 are shown in Figure 5. It is seen that a
relatively large difference in response magnitude occurred. The initial near zero force occurred
during gravity loading of the model. Once the platen begins to penetrate into the soil, the platen
force for MAT005 gradually begins to increase to a value of 2 MN at time equal to 5. In contrast
for MAT025, the penetration force initially rises sharply and then gradually increases to a value
of 5.5 MN. This was to be expected since the MAT025 model response to hydrostatic
compression showed similar behavior.
The response labeled MAT025a in Figure 5 was obtained by subtracting the initial sharp rise in
the MAT025 response. This brought the two force histories closer, but the MAT025a response is
still much higher than that of MAT005.
Figure 5: Penetration force history for MAT005 and MAT025
A look at several SPH elements in SD (stress difference)-Pressure space provides a comparison
of the stress state evolution relative to the failure surface, Fe. For MAT025, Figure 6 shows the
stress trajectory of SPH element 351660, which is directly under the platen, and it is seen that the
trajectory starts out and remains under the failure envelop, Fe.
1-6
12th International LS-DYNA Users Conference FSI/ALE(2)
Figure 6: SPH element 351660 stress trajectory in Stress Difference-Pressure space
Figure 7 shows the trajectory for SPH element 36158, which is under the platen but deeper into
the soil. Initially, the trajectory shows that the failure surface is penetrated, but as the platen
continues to penetrate the soil, the trajectory moves below the failure surface.
Figure 7: SPH element 368158 stress trajectory in Stress Difference-Pressure space
The final SPH element studied was element 335130, which is initially just under the outer edge
of the platen. As the platen moves down, element 335130 moves out from under the platen and
around the edge of the platen. Figure 8 shows the trajectory during the initial downward
movement, which is then followed by lateral motion that pulls the element out of the range of
direct platen loading. This effect is shown by the initial increase in pressure and the subsequent
drop in pressure as the element moves laterally. Here also, the failure envelope is slightly
violated.
1-7
FSI/ALE(2) 12th International LS-DYNA Users Conference
From Figure 7 and Figure 8, it is seen that for MAT025 the stress state of several SPH elements
slightly penetrated the failure surface during the loading process. This is probably due to the lack
of strain sub-incrementation in the model.
Figure 8: SPH element 335130 stress trajectory in Stress Difference-Pressure space
Summary and Conclusions
A study was performed to investigate how the choice of material models and their parameters
used to represent nonlinear soil behavior affects the response of a structure undergoing large
displacements within the soil. The following material models were used: (1) MAT005, Soil and
Crushable Foam; (2) MAT010, Elastic Plastic Hydro; (3) MAT025, Geological Cap; and (4)
MAT079, Hysteretic Soil. Parameters for these models were obtained previously, but these
parameters were based on using the model for a different application than the one exercised here.
A one Lagrangian element model was used to compute the hydrostatic compression response for
each material model. MAT005 and MAT010 replicated the experimental hydrostatic
compression response because exact hydrostatic compression data is used as input to these
material models. For the parameters used, MAT025 gave the stiffest response, and MAT079
produced the softest. It should be noted that the MAT025 and MAT079 pressure-volume strain
relationship is difficult to fit to pressure-volume strain data – especially for soil.
The MAT005 and MAT025 models were used in the SPH simulation of a platen being pushed
into soil. Similar to the one Lagrangian element hydrostatic compression simulation, MAT025
required much higher vertical forces to push the platen into the sand. In addition, since MAT005
1-8
12th International LS-DYNA Users Conference FSI/ALE(2)
has a flat cap, MAT025 would encounter the cap sooner than MAT005 for the parts of the soil
whose stress trajectory is further away from hydrostatic compression. It was noted that for
MAT025, several SPH elements slightly penetrated the failure surface during the loading
process. This is probably due to the lack of strain sub-incrementation in the model.
The successful estimation of the response of soil-structure interaction problems requires
calibrated material models for the specific problem. To calibrate the material models, the
analysts should have an idea of the expected strains – for example, volumetric strain – that will
occur in the soil. Soil testing should be done in this range to ensure that the most accurate
material parameters can be obtained. Most likely, this will involve an iterative process between
material model calibration and computer simulations.
Acknowledgement
The first author acknowledges the support of Argonne National Laboratory’s Transportation
Research and Analysis Computing Center (TRACC), which is supported by the U.S. Department
of Transportation. The strong support of TRACC’s Director, Dr. Hubert Ley, is greatly
appreciated.
References
[1] http://web.anl.gov/TRACC/
[2] Kulak R.F. and Bojanowski C., Modeling of Soil-Structure Interaction in Presence of Large
Deformations in Soil, Paper No. 11-2287, 90th Transportation Research Board Annual Meeting,
Washington, D.C., January 23-27, 2011
[3] Livermore Software Technology Corporation (LSTC), LS-DYNA Keyword User’s Manual
[4] Schwer L., Soils and Foams Model Case Study Class Notes: Concrete & Geomaterial
Modeling with LS-DYNA, Schwer Engineering and Consulting Services 2002
1-9
