You are on page 1of 19

Cold Regions Science and Technology 40 (2004) 193 – 211

www.elsevier.com/locate/coldregions

A macroscale model for low density snow subjected to


rapid loading
Robert B. Haehnel*, Sally A. Shoop
US Army Engineer Research and Development Center, Cold Regions Research and Development Laboratory (ERDC-CRREL),
Hanover, NH, 03755, Germany
Received 3 September 2003; accepted 4 August 2004

Abstract

A Capped Drucker–Prager (CDP) model was used to simulate the deformation-load response of a low density (150–250 kg/
m3) snow being loaded at high strain rates (i.e., strain rates associated with vehicle passage) in the temperature range of 1 to
10 8C. The range in the appropriate model parameters was determined from experimental data. The model parameters were
refined by running finite-element models of a radially confined uniaxial compression test and a plate sinkage test and comparing
these results with laboratory and field experiments of the same. This effort resulted in the development of two sets of model
parameters for low density snow, one set that is applicable for weak or bsoftQ snow and a second set that is representative of
stronger or bhardQ (aged or sintered) snow. Together, these models provide a prediction of the upper and lower bound of the
macroscale snow response in this density range. Furthermore, the modeled snow compaction density agrees well with measured
data. These models were used to simulate a tire rolling through new fallen snow and showed good agreement with the available
field data over the same depth and density range.
D 2004 Elsevier B.V. All rights reserved.

Keywords: Snow; Low density; Finite element modeling (FEM); Capped Drucker–Prager; Plastic constitutive law; Snow mechanics

1. Introduction (3) Applying to military, such as the ability of snow


to absorb projectile impacts and problems related
Understanding the mechanical properties of snow to snow-covered minefields.
subjected to rapid loading has application to many
areas of interest (Shapiro et al., 1997) including: Field evaluation of these technologies in snow can
be problematic owing to lack of reproducibility of
(1) Designing snow removal equipment.
results in what is seemingly the bsameQ snow (e.g., the
(2) Predicting vehicle performance in snow.
same density and temperature). A typical example is
the difficulty of quantitatively evaluating the relative
* Corresponding author. performance of snow tires, even when each is tested
E-mail address: Robert.B.Haehnel@erdc.usace.army.mil on the same track on the same day. It has long been
(R.B. Haehnel). recognized that this is attributable to the variation in
0165-232X/$ - see front matter D 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.coldregions.2004.08.001
194 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

mechanical properties of the snow with the micro- structures and deposition of water vapor on larger
structure of the pack, which evolves in time because crystal structures (migration from a high surface area/
of changes in air temperature, humidity, sintering of volume grain structure to a low surface area/volume
the snow particles over time, the work done to the grain structure). Concurrently, pressure and temper-
snow (e.g., vehicle passage), etc. ature gradients within the snow pack lead to sintering
Characterization of snow in general, and low of the snow. These mechanisms also lead to consol-
density snow in particular (100–250 kg/m3), is idation of the snow pack, yet can yield a profoundly
problematic because of the range of falling snow different structure and snow strength for the same
types and the rapid metamorphic changes that take density.
place in the snow once it is on the ground. Falling For example, Fig. 1 shows the wide range in
snow can range from a feathery denderitic structure to response of snow that has a density of approximately
pellet-like sub-angular forms. The exact form of the 150 kg/m3 subjected to uniaxial compression. This
snow as it falls depends on the temperature, humidity, plot is for snow collected in its pristine condition from
and wind history that the flakes experience from the the field and then compressed in a radially confined
time of their genesis until they reach the ground. Once uniaxial compression test (Abele and Gow, 1975).
on the ground, the snow particles continue to This kind of scatter is typical for snow. The material
metamorphose due to ambient weather conditions responds differently because of variations in sample
(e.g., wind, humidity, temperature, etc.). For example, temperature, age of sample, snow microstructure, etc.,
a snow cover that is initially dendritic in structure not all of which are easily controlled or readily
(typically density of 100 kg/m3) subjected to wind measured in the field. Of particular interest is the
induced drifting is rapidly transformed to nearly response of tests 1 and 2 in comparison to test 50 (Fig.
spherical particles that have a mean density of 300– 1). Tests 1 and 2 were done at a temperature of 7 8C,
400 kg/m3. In the absence of wind this same fresh while test 50 was conducted at 3 8C—yet tests 1 and
snow cover can undergo temperature-driven meta- 2 are the bsoftestQ samples in this density range, and
morphism that causes sublimation of the small crystal test 50 is among the bhardest.Q In this plot, it is clear

Fig. 1. Stress–strain curve for radially confined uniaxial compression tests of snow with a density range of 140–160 kg/m3 (data from Abele and
Gow, 1975).
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 195

that the strength of the snow does not simply decrease 1994; Fervers, 1994) or critical state models such
with temperature and illustrates that characterization as the Bailey and Johnson (1989) soil compaction
of snow requires far more information than simply model, implemented by Foster et al. (1995), or a new
density or temperature; a knowledge of the snow’s critical state model (similar to Lade and Kim, 1995),
microstructure (the type of snow grains present and implemented by Liu and Wong (1996). Comparisons
the strength of the bond between these grains) and its between the Drucker–Prager and Cam-clay models
influence on the structural properties (compressive for tire–terrain interaction have been published by
strength, elastic modulus, etc.) of the snow pack is Meschke et al. (1996) for snow and by Chi and
required. Tessier (1995) for soil. Although Cam-clay is perhaps
For example, Fukue (1979) showed that the age the most widely used soil model, both of the above
of the snow can have a significant effect on its studies comment on convergence problems when
strength. In unconfined uniaxial compression tests using Cam-Clay. The choice of material model is
using manufactured snow, Fukue showed that, for based on balancing the type of behavior desired in
samples allowed to sinter at constant density, the the model with the information available to determine
strength of the snow increased 10-fold over the model parameters.
course of 6 days. The influence of the snow’s Meschke et al. (1996) and Miyori et al. (2002),
microstructure on its mechanical response is reported respectively, have produced material model parame-
by Armstrong (1980). His study of consolidation of ters for Cam-Clay and Mohr–Coulomb models of
an alpine snow pack showed that a fine-grained, packed snow (snow density of 400–500 kg/m3). In
sintered snow deformed at a rate 10 times greater this work we present a material model that is
than an adjacent layer of depth hoar, though the representative of low density snow (approximately
density of both layers were the same. Finally, 150–250 kg/m3) subjected to short-term loading (i.e.,
Voitkovsky et al. (1975) showed that the cohesion creep response of the snow is not considered).
strength of snow varies widely when plotted as a Recognizing that the strength of snow depends on
function of density, yet the same cohesion data the snow microstructure, and not just density alone;
plotted as a function of specific snow grain contact we have developed material models representative of
surface (net area of inter-grain contacts per unit both weak and strong snow structures using a
volume) revealed a nearly linear relationship. modified Capped Drucker–Prager (CDP) model as
One way to side step the problem of poor implemented in the ABAQUS (HKS, 1998) finite
reproducibility of field results is to use virtual element computer code. Though the microscale
prototyping techniques, in which the snow is repre- processes (e.g., breaking of bonds, sliding of grains)
sented using a computer model and the interaction of are not explicitly modeled, this model is suitable for
the vehicle with the snow surface is simulated. This capturing the large-scale compressive and shear
provides a test bed using reproducible snow con- behavior of natural, fresh snow (Shoop, 2001; Shoop
ditions for a proscribed snow state. However, this et al., 1999a,b, 2001). A description of the model,
requires development of a material model for snow model parameter determination, and validation of the
that is representative of the snow response for an material models with laboratory and field data
appropriate set of conditions (e.g., snow density, age). follows.
Such a model was produced for compacted high
density snow (400–500 kg/m3) by Meschke et al.
(1996). The model presented here is for low density, 2. General concepts for plasticity models
undisturbed snow.
Terrain substrate subjected to wheel loads has The purpose of a plasticity model is to describe the
been represented using a wide variety of material permanent deformation of a plastic material and has
models, including elastic, non-linear elastic, viscoe- the following basic components (Wood, 1990):
lastic (Pi, 1988), and elastic–viscoplastic (Saliba,
1990). Recent studies concentrate on using either (1) Elastic properties to define the recoverable
Capped Drucker–Prager plasticity (Aubel, 1993, deformation.
196 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

(2) A mathematical surface to define the yield the features of a critical state model (i.e., regions of
boundary between elastic behavior and plastic constant volume shear deformation, and compactive–
material behavior. dilatant flow). CDP uses non-associative flow on the
(3) A plastic flow potential to mathematically define shear surface (i.e., the flow potential is not associated
the plastic deformation (also called a plastic flow with the yield surface) and associated flow on the cap
law). surface.
(4) A hardening/softening rule defining the move-
ment (expansion or contraction) of the yield
surface during plastic deformation. 3. Modified cap Drucker–Prager model

A good overview of the development of plasticity The yield surface for the CDP material model is
theory and constitutive modeling of soil is given in described in terms of stress invariant functions of the
Scott (1985). Schofield and Wroth (1968) extended stress tensor, S (a term in bold face denotes a matrix
plasticity theory to the critical state concept, defining or tensor). The particular invariants used for the CDP
either contractile or dilatant deformation of porous model as implemented in ABAQUS (HKS, 1998) are
material as a function of its specific volume or void defined as:
ratio. In critical state theory, this rule is developed (1) The invariant, p, is the equivalent pressure
around the concept of a bcritical state,Q where the stress, or the negative octahedral normal stress, r oct
plastic shearing deformation occurs at a constant (Jaeger and Cook, 1969), which determines uniform
volume. Perhaps the most famous critical state compression or dilation:
model, the Cam-Clay model, was developed based
on the behavior of clays. The concepts are equally 1
p ¼  roct ¼  r : I: ð1Þ
applicable to defining the shearing and volumetric 3
behavior of granular materials, such as granular soils
or snow (Wood, 1990). Although the concepts are where I is the identity matrix and the operator b : Q
applicable for both cohesive and granular materials, denotes a scalar product.
the behavior of the granular materials has not been (2) The invariant, q, is the deviatoric stress, also
explored as thoroughly, particularly regarding the called the Mises equivalent stress, which determines
influence of deviatoric stress on the yield surface, distortion:
which is less clearly defined in soils but may take on rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a much different shape than the yield surface of 3
q¼ ðS:SÞ ð2Þ
metals. 2
Liu (1994) used the Drucker–Prager plasticity
where S is the stress deviator
model for modeling the sliding of rubber blocks on
snow. Mundl et al. (1997) extended Liu’s work
S ¼ r þ pI:
using a multi-surface plasticity model to optimize
the snow behavior under both shear and compres-
This form of the second invariant is analogous—but
sion. Their intent, however, was to simulate
not identical—to the octahedral shear stress put
shearing forces of snow compacted on roads
forward by Jaeger and Cook (1969).
(density of 500 kg/m3), which is a significantly
(3) The invariant, r, is
different material than that encountered during
cross-country mobility on fresh snow (density of ! 13
200–300 kg/m3). For this study, the plasticity 9
r¼ SdS:S : ð3Þ
model used was the well-documented Capped 2
Drucker–Prager.
Cap plasticity models account for both the com- The bd Q operator denotes matrix multiplication.
pression and shearing of an isotropic material. The In the formulation of the CDP model, the second
modified Capped Drucker–Prager (CDP) model has and third invariants are combined to create an
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 197

Fig. 2. (a) Modified Drucker–Prager yield surface in deviatoric space compared with (b) other common yield surfaces (HKS, 1998).
198 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

alternate expression, t, the deviatoric stress measure The following equations define the yield criteria in
(HKS, 1998), each section of the yield surface. For Drucker–Prager
2 ! !3 3 shear or distortional failure
q 1 1 r 5
t ¼ 41 þ  1  ð4Þ Fs ¼ t  ptanb  d ¼ 0 ð5Þ
2 K K q
where d is the Drucker–Prager material cohesion and
where K is the flow stress ratio (ratio of tensile b is the Drucker–Prager material angle of friction.
strength to compressive strength) and describes the These are analogous to Mohr–Coulomb cohesion, c,
yield dependence on the third stress invariant, defin- and the internal angle of friction, /, respectively.
ing the shape of the yield surface in the deviatoric For the cap region of compactive–dilatant failure
plane. For K=1, the yield surface is circular (von vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Mises yield), as shown in Fig. 2a, and the failure u " #2ffi
u Rt
stress is the same in tension and compression. For the Fc ¼ tðp  pa Þ2 þ
surface to remain convex, K is limited to values of ð1 þ a  a=cosbÞ
0.778 to 1. For K=0.778, the CDP model can be used
 Rðd þ pa tanbÞ ¼ 0 ð6Þ
to approximate the Mohr–Coulomb surface shown in
Fig. 2b.
where a is a transition parameter, ranging typically
The subsequent model parameters are defined in
from 0.0 to 0.05, that smoothes the transition between
the pressure-deviatoric plane (also called the meridia-
the shear failure and the cap failure, R is a material
nal, p–q, or p–t plane). For the modified cap Drucker–
parameter controlling the shape of the cap (or the cap
Prager model used in this study, the yield surface is a
eccentricity parameter), and p a is the intersection of
modified von Mises yield (i.e., the material constant
the shear line and the cap (in absence of a transition
K=1.0); hence, Eq. (4) reduces to t=q, eliminating the
surface) and relates to the cap hardening behavior
effects of the invariant, r. In the p–t plane, the yield
according to
surface has two major segments (Fig. 3):
pb  Rd
(1) The Drucker–Prager portion of the curve (anal- pa ¼ ð7Þ
ð1 þ RtanbÞ
ogous to the Mohr–Coulomb line) defines shear
deformation. where the mean hydrostatic pressure, p b, is a function
pl
(2) The cap portion of the surface defines the of the volumetric plastic strain, e vol . This functional
pl
intersection with the pressure axis. relationship, p b=f(qvol ), is the hardening law which

Fig. 3. Modified cap Drucker–Prager yield surface in the p–t plane (HKS, 1998).
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 199

defines the pressure–volume relationship during com- mended by HKS (1998) for a better fit to the data and
pression of the material at the cap failure surface. better model performance; this approach was used in
The transition surface between the shear and the this study.
cap failure is defined as
3.2. Plastic flow
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
" #2
u  
u a The plastic flow is defined by an elliptical shaped,
t 2
Ft ¼ ðp  pa Þ þ t  1  ðd þ pa tanbÞ
cosb flow potential surface. Flow is associative (normal to
the surface) in the cap region; therefore, the equation
 aðd þ pa tanbÞ ¼ 0 ð8Þ for the flow surface is identical to the equation for the
cap yield surface. In the transition and shear region,
the flow is non-associative (flow potential is inde-
3.1. Hardening law pendent of the failure surface), and the flow surface,
G s, is defined as (HKS, 1998)
The pressure–volume relationships define both vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
hardening and softening through volume changes u" #2 " #2ffi
u t
based on how the cap portion of the yield surfaces Gs ¼ t ðpa  pÞtanb þ :
expands and contracts. The cap is generally spherical 1 þ a  a=cosb
or ellipsoidal, and the material either hardens or ð9Þ
softens by expanding or contracting the cap, respec-
tively. This behavior is defined in a pressure–volume
relationship called a hardening law. The pressure–
volume relationship is often exponential and, there- 4. Material parameter determination
fore, can be modeled using an exponential hardening
law. In cases where exponential hardening is not a Extreme changes in snow properties with time,
good fit, as is common in soils, the hardening law can temperature gradients, applied load, and deformation
also be represented in a piecewise linear approach prohibit model parameter acquisition from the same
using the experimental data (e.g., Fig. 4) as a table of snow. Consequently, approximations were made by
pl
p b and e vol pairs. The piecewise approach is recom- estimating parameters using test data from snow of

Fig. 4. Piecewise linear models of the hardening law for low density snow-based on Abele and Gow’s (1975) experimental data—for fresh
(S200) and age hardened snow (H200). The rapid rise in pressure stress for high volumetric strain (N1.5) occurs as the snow is compacted to the
density of ice.
200 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

similar characteristics (density, age and snow type). Acquiring apparent values for cohesion and fric-
The type of snow modeled was a natural snow with tion angle, cV and /V from shear loading of snow
a density of 150–250 kg/m3 at moderate temper- under a vehicle tire and from a ring shear device is
atures (between 10 and 1 8C). Test data were discussed by Blaisdell et al. (1990) and Alger and
gathered from the field and from the literature to Osborne (1989). Both methods were applied to field
match this snow type as closely as possible. Because measurements on undisturbed snow with a density
data were gathered from several studies, the snow ranging from 60 to 250 kg/m3 and temperatures from
was not exactly the same in crystal structure; 2 to 16 8C. From these data, Shoop (2001)
however, the cumulative result can be considered determined that for a snow density of about 200 kg/
an baverageQ snow response taken over all of the m3, c=2.14–43 kPa and /=8.98–148. Because of the
available field and laboratory data. A discussion of nature of the test, these values may be more
selecting the initial values of the material parameters representative of the interface shear. However, similar
follows. values of cohesion (approximately 20–30 kPa) were
reported in Shapiro et al. (1997). These values yield a
4.1. Elastic properties Drucker–Prager d ranging from 3.7 to 72 kPa and b
ranging from 15.28 to 22.58.
For snow deformation, the plastic deformation is Parameters describing the shape of the cap were
much greater than elastic deformation; however, the adjusted based on model response. The flow stress
elastic deformation is not negligible. The basic elastic ratio, K, was held at 1.0. This agrees with data
parameters, Young’s modulus E and Poisson’s ratio m presented in Shapiro et al. (1997), indicating nearly
were estimated based on data compiled by Shapiro et equal values of compressive and tensile strength for
al. (1997). Because the elastic modulus is a strong low-density snow. The cap eccentricity parameter, R,
function of density, varying over 4 orders of was chosen based on typical values for earth materials
magnitude, the elastic modulus should ideally be having a very steep compaction cap (R=0.1 to
modeled as a function of the density as a state 0.0001). The transition surface radius, a, was set to
variable. For this study, however, Young’s modulus 0.0 (no transition). The initial cap yield surface
pl
was held constant at a value equivalent to snow position e vol |0 was arbitrarily set to 0.001 to allow
having a density of 300–400 kg/m3. At snow densities initial softening as needed. These material parameters
below this, the elastic contribution is considered were refined based on model response and compar-
minimal. Based on this, a Young’s modulus in the ison to experimental data.
range of 1–20 MPa and a Poisson’s ratio of 0.3 was
used for initial model parameter selection. 4.3. Hardening law

4.2. Yield surface The hardening law was based on tabular data
obtained from compression test data published in
The parameters defining the shear portion of the Abele and Gow (1975) using three different tests of
yield surface for the Drucker–Prager model can be snow having similar density, along with values
calculated from the Mohr–Coulomb cohesion, c, and calculated from the final pressure–density pairs of
the internal angle of friction, /. Assuming plane strain an additional 21 tests. As the measurements are from
response and non-dilatant flow, we can calculate the one-dimensional compression or consolidation (oed-
Drucker–Prager material cohesion, d, and the ometer) tests, by application of a vertical load to a
Drucker–Prager material angle of friction, b from sample within a rigid cylinder (i.e., radially confined
(HKS, 2001) uniaxial strain), the calculation of the mean hydro-
static pressure, p b, is not straightforward. Two simple
tanb ¼ 1:73sin/ ð10Þ approaches are to assume that r 1=r 2=r 3, which gives
p b=r 1; alternately we can assume r 2=r 3=0, which
gives p b=1/3r 1. These two cases bound the sol-
d ¼ c1:73cos/: ð11Þ ution. An initial hardening table was derived based on
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 201

the average of Abele and Gow tests on warm, 200 kg/ in the data is not explained by Abele and Gow (1975).
m3 snow (tests 10 and 29) using p b=r 1 (H200 curve They mention that b[s]ome samples were stored at
in Fig. 4); this table was modified within the range of constant temperature before testing,Q indicating that
1/3r 1zp bzr 1 during model calibration. aging of the samples indeed occurred, though it was
not documented how long each individual sample sat
at constant temperature prior to testing. Furthermore,
5. Model calibration using laboratory and field there was no attempt to categorize the snow type of
data each individual sample (e.g., grauple, dendritic, etc.);
without such information on sample age and micro-
Finite element models simulating a radially con- structure it is impossible to determine the root cause
fined uniaxial compression test and a plate-sinkage of the scatter. Nevertheless, it is clear that for the same
test were used to refine the material properties given density range snow can behave very differently and
in the previous sections. this can be attributed to the microstructure of the
snow, be it the structure of the snow when it falls or
5.1. Uniaxial compression the time-dependent metamorphic changes in snow
structure; thus, we seek to have the capability to
For the uniaxial compression model, a single model both the soft and hard snow conditions evident
element model was determined to be adequate after in low density snow. This was done by selecting two
multi-element models gave the same results. Fig. 5 sets of model parameters and hardening curves to
gives a plot of all of the data from Abele and Gow capture the range in response of the snow.
(1975) with a density ranging from 140 to 230 kg/m3 Fig. 5 compares the uniaxial experimental data and
and temperature between 7 8C and freezing. finite element model results. The nomenclature used
Unloading was not measured in the laboratory experi- for the CDP material models are bS200Q for weak or
ments. As with Fig. 1, we see a broad range in the soft snow in the density range of 200 kg/m3, and
response of the snow in experimental data; this scatter bH200Q for strong or hard snow in the same density

Fig. 5. Comparison of model and experimental data for uniaxial compression tests on low-density snow. The data are from Abele and Gow
(1975): snow temperature ranges from 0 to 7 8C, snow density ranges from 140 to 230 kg/m3.
202 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

range. The agreement between modeled and measured geometry, with vertical shear surfaces along the
material behavior is reasonably good for the two snow/plate interface. The vertical frictional surfaces
material models, with one model capturing the softer allowed these elements to slide, closely mimicking
response of Abele and Gow snow data and the second the clearly defined shear failure surfaces observed in
model representing harder response of the snow. Table the field. The use of slip planes was a reasonable
1 gives the material parameters used for each of these compromise that allowed us to model this test
snow models. The hardening curves for each of these geometry without having excessive element distortion
models are also plotted in Fig. 4. (in the vicinity of the sharp edge of the plate) that
prevented the model running to completion. The slide
5.2. Plate sinkage planes were modeled using a contact surface with
Coulomb friction of l=0.3. As the cohesion of low
Comparisons with three plate sinkage tests are density snow is very small (5 kPa), we found that
presented: two field tests and one laboratory test. The adding a failure criteria to this slip plane (i.e., the
first field test used, designated bKRC Runway,Q bond between adjacent elements on the slip plane
consisted of pushing a 20-cm (8-in.)-diameter rigid would break if the shear stress at that point exceeded
plate, originally at the surface, into 39.2 cm (15.5 in.) the cohesion strength, d, of the material) did not
of fresh snow with average density of 300 kg/m3 significantly change the simulated load–sinkage
(Alger and Osborne, 1989). This was modeled using curves. This suggests that frictional sliding along this
a 2D axis-symmetric representation of the test failure plane dominates and including only the
frictional resistance along these slide planes is
Table 1 adequate for modeling this case.
Parameters for the modified Capped Drucker–Prager model of low The plate was lowered by constraining the surface
density snow (150–250 kg/m3) nodes radially while displacing them into the snow,
Parameter S200 H200 effectively creating a no-slip contact between the plate
Young’s modulus (MPa), E 1.379 13.79 and snow. Both the S200 and H200 material
Poisson’s ratio, m 0.3 parameters were used to simulate this condition. The
Drucker–Prager cohesion (kPa), d 5 deformed model of the field test is shown in Fig. 6.
Drucker–Prager angle of friction (deg), b 22.538
Cap eccentricity, R 0.02
Fig. 7 gives a comparison of the field data and the
Initial cap yield surface position, epl
vol |0 0.001 model results. Both of the snow models do a fair job
Transition surface radius, a 0.0 of reproducing the field data: the S200 model follows
Flow stress ratio, K 1.0 the low strain region of the curve but under predicts
Average snow density (kg/m3) 200 the peak load; the H200 model does a very good job
Hardening law
of capturing the peak forces but it over predicts the
force at the low end (plate sinkage 0–0.2 m) of the
Mean hydrostatic pressure (Pa) e pl
vol
loading curve.
S200 H200 The second test, designated bKRC Texas field,Q
113.76 113.76 0 used the same size plate as above, but the overall
0.017106 0.05106 0.593 snow depth was 51.4 cm (20.2 in). The upper layer of
0.033106 0.1106 0.669
0.067106 0.2106 0.806
snow was fresh snow with a density of 110 kg/m3, and
0.167106 0.5106 0.944 the lower layer was older snow with a density of 230–
0.333106 1.0106 1.083 280 kg/m3 (Alger and Osborne, 1989). This was
0.667106 2.0106 1.299 modeled using the same basic geometry as the KRC
0.933106 2.8106 1.455 runway described above, except the snow depth used
1.083106 3.25106 1.475
2.0106 6.0106 1.50
in the model matched that of the field, 51.4 cm, and
2.0107 6.0107 1.514 the snow was modeled in three ways:
The softer snow response is modeled using the bS200Q parameters
while the harder snow response is modeled with the bH200Q (1) The entire snow depth was modeled using H200
material parameters. parameters.
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 203

(2) The entire snow depth was modeled using S200


parameters.
(3) The snow was modeled in layers, the upper 13.1
cm of the snow was modeled using the S200
material parameters, and the remaining snow depth
was modeled using the H200 material parameters.

The results of these simulations are compared


with the field data in Fig. 8. Though all three models
provide reasonable agreement, the S200 model tends
under-predict the peak load, while the H200 model
tends to over-predict the forces at low sinkage
depths. The two-layer model does a better job of
capturing the peak forces while still following the
overall force trace. This simulation supports the
notion that there needs to be some consideration in
modeling that takes into account age hardening (or
microstructure) of snow where two samples of snow
that have similar or identical density differ in
strength owing to one being fresh snow and the
other having strengthened over time because of
Fig. 6. Deformed mesh for simulation of plate sinkage tests sintering and metamorphism. In this simulation the
performed on the KRC Runway: H200 model parameters used. upper layer of fresh snow is weaker than the

Fig. 7. Comparison of plate sinkage KRC Runway field test data (Alger and Osborne, 1989) ( S ) to the model results (lines).
204 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

Fig. 8. Comparison of plate sinkage KRC Texas field test data (Alger and Osborne, 1989) (S ) to the model results (lines).

underlying old snow, and the differences in material 606050 cm deep with an average initial density
parameters attempt to reproduce this condition. of 180 kg/m3 (Shoop and Alger, 1998). This same
The laboratory plate sinkage test pushed a 20-cm geometry was captured in the three-dimensional
(8-in.)-diameter rigid plate into a cube of snow model simulation of this test. The snow was modeled

Fig. 9. Comparison of modeled (lines) and measured (Shoop and Alger, 1998) ( S ) plate sinkage results for laboratory experiments.
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 205

Fig. 10. Modeled snow density (kg/m3) of plate sinkage test for Capped Drucker–Prager material (laboratory test simulation at left and field test
simulation at right). Both models shown used the H200 material parameters (Shoop et al., 1999a,b).

using both the S200 and H200 material parameters. good job of capturing the force–displacement curve
The model results are compared with the lab experi- for a penetration of the plate into the snow greater
ment in Fig. 9. The H200 snow simulation does a very than 0.1 m. For plate penetration less than 0.1 m, the

Fig. 11. Measured displacement and snow density in laboratory plate sinkage test (after Shoop and Alger, 1998).
206 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

H200 model over predicts the force; the S200 model The material parameters presented in Table 1 in
in general is too soft and does not accurately some respect dictate that these models would
reproduce the data for these snow conditions. provide bounding conditions. The elastic modulus
In the model, snow density can be calculated from for the two models (S200 and H200) roughly
the volumetric plastic strain. The snow density values follows the bounding range given by Shapiro et
from the models are shown in Fig. 10. These densities al. (1997): 1–20 MPa. By increasing the elastic
can be compared to snow densities measured under modulus within this range. the snow can be made
the plate, as shown in Fig. 11. The maximum harder and btuningQ of this parameter within this
volumetric inelastic strain in the plate sinkage model range can help match specific field data. Likewise,
occurred immediately below the plate at values of the hardening curves given in Table 1 are for the
approximately 0.5, which is equivalent to a snow bounds 1/3r 1zp bzr 1. These hardening tables
density of 360 kg/m3, whereas in the experiment a could also be modified within this range to match a
very dense (440–520 kg/m3) bulb of snow was particular set of experimental data. However, the
observed extending approximately 5 cm below the ability of these models, used together, to capture
plate (Fig. 11). Below this, the modeled snow density the upper and lower bound of the snow response
varies from about 240 kg/m3 under the plate to about can be of great use by providing the expected range
215 kg/m3 near the ground. Yet, the measured density of snow response in this density range, not just the
shown in Fig. 11 is nearly constant at 290–310 kg/m3 average. Capturing the range of snow response,
below the high density bulb—other experimental when so little is known a priori about the age or
results from Shoop and Alger (1998) show that the microstructure of the snow that will be encountered
density under the bulb can vary more widely (e.g., in the field, is far more useful than having a single
270–350 kg/m3) than that shown in Fig. 11. Thus, the model that matches a specific set of snow
model does not clearly capture the high density bulb conditions that may or may not be representative
observed in the laboratory experiments, but rather of the field conditions at another time or location.
appears to smear out the variation in density over the
entire snow depth. Yet, the predicted density of the
snow adjacent to the plate compares quite favorably 7. Wheel on snow application
with the measured lab data with both model and
experiment demonstrating a slight rise in density due Turning now to a prototyping application, we
to lateral compaction. developed a model of a wheel rolling through snow
and compared these results to measurements of the
performance of the CRREL instrumented vehicle
6. Discussion (CIV) rolling through fresh snow. The CIV, originally
a 1977 AMC Jeep Cherokee, is instrumented to
The comparison of the uniaxial data to the model measure vertical, longitudinal, and lateral forces at
results shows that the S200 and H200 models may not the tire interface, wheel speed at each wheel, true
be capable of capturing the specific response of a vehicle speed, and steering angles. In this study we
particular snow microstructure, yet using these two used available data for the CIV documenting the tire
models can be effective at bounding the snow sinkage depth (rut depth), the longitudinal resistance
response in this density range, with S200 capturing force of the wheel as the vehicle moved through the
the lower bound associated with soft snow and H200 fresh snow, and the snow density surrounding the tire
approaching the upper bound (hard snow). This rut (Blaisdell et al., 1990; Green and Blaisdell, 1991;
concept is further reinforced with comparisons to Richmond, 1995).
plate sinkage tests. For both of the field cases (KRC To reduce run time, the tire was modeled as a rigid
Runway and Texas Field) it appears that the two wheel (a rigid analytic surface). Fig. 12 shows the
models do a very good job of bounding the scatter in model geometry and the resulting deformed snow
the snow response, and the two-layer model of the surface for 20-cm-deep snow. Taking advantage of the
KRC Texas Field gives an average through the data. symmetry of the problem, we modeled the simulation
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 207

Fig. 12. Model of a rigid wheel rolling through snow. Material parameters for S200 were used.

in half space. No slip plane was needed to model this From the model we determined the steady state
case, as the rounded transition between the wheel reaction force at the hub (i.e., the resistance force to
btreadQ and bsidewallQ appeared to prevent excessive pull the wheel forward) and the steady state wheel
distortion of the elements in this region, allowing the sinkage depth into the snow. These steady state
model to run to completion. The wheel motion was values were obtained from the portion of the
simulated as follows: simulation at which the hub was moving at a constant
velocity. The resistance force was normalized by the
(1) The wheel was allowed to settle unto the snow vertical load on the hub (the sprung and unsprung
surface under the influence of gravity and its mass carried by that tire). These results, along with
own weight. the field data, are plotted in Fig. 13. The S200 model
(2) The hub of the wheel was accelerated to 2.24 m/ does a very good job of predicting the depth that the
s (5 miles/h); concurrently, an angular velocity tire sinks into the snow (Fig. 13a) over the full range
was applied to the hub to assure zero slip of the available field data, yet the H200 model tends
between the snow surface and the wheel. to under predict the sinkage depth. This is not
(3) Once the hub velocity reached 2.24 m/s, the surprising, as these data were taken in freshly fallen
velocity was maintained constant for the remain- snow and the S200 model should be more represen-
der of the simulation. tative of this condition.
The normalized resistance force (Fig. 13b) pre-
The model was run using both the S200 and H200 dicted by the S200 model is a bit high in comparison
material parameters. The performance of the CIV was to field data, yet it does a reasonable job of capturing
simulated for snow depths ranging from 5 to 50 cm. the upper envelope of the data. As such, it tends to be
The coefficient of friction between the wheel and the conservative with respect to predicting the rolling
snow was set at 0.3. resistance forces on the tire. Again, because the data
208 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

Fig. 13. Comparison of the CDP snow model predictions to field data for (a) sinkage and (b) normalized motion resistance.

were taken in fresh snow, the S200 model provides a measured snow density under the tire. The model does
better estimate of the rolling resistance than the H200 a good job of reproducing the overall snow deforma-
model. tion shape and matches the compaction density of the
Fig. 14 compares the deformation of the snow in snow under the tire. Because the snow density varies
the model and field as well as the modeled and with depth in the field, yet was held constant in the
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 209

Fig. 14. Comparison of (a) the field snow deformation shape (Richmond, 1995) to (b) the model and comparison of (b) the predicted snow
density under a tire to (c) density measurements from Richmond (1995). The snow depth is 19 cm. The model uses the S200 material
parameters. The black lines in the snow shown in (a) are chalk lines used to document the deformation of the snow as the wheel passes
through it.
210 R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211

model, it is more difficult to directly compare the Acknowledgements


variation in density adjacent to the tire side wall. Yet,
it is clear that, in the field, there is some lateral This effort was funded by the U.S. Army High
compaction of the snow adjacent the tire side wall, Fidelity Ground Platform and Terrain Mechanics
evidenced by the lateral deformation of the vertical Science and Technology Objective, Terrain Mechanics
chalk lines in the snow (Fig. 14a); this lateral Models for All-Season Terrain, Work Unit. This work
compaction is predicted by the model as well. We was supported in part by a grant of computer time from
note that the snow density at the corners of the tire is the DOD High Performance Computing Modernization
about 20% higher in the model than that measured in Program at the Engineer Research and Development
the field. Center, Vickburg, MS and the Army Tank-automotive
and Armaments Command, Warren, Michigan.

8. Conclusions
References
The modified Capped Drucker–Prager constitutive
Abele, G., Gow, A., 1975. Compressibility Characteristics of
law was used for modeling the macroscale material Undisturbed Snow. Research Report 336. U.S. Army Cold
behavior of relatively warm (1 to 10 8C) low Regions Research and Engineering Laboratory, Hanover, NH.
density snow (150–250 kg/m3) using the finite Ahlvin, R., Shoop, S.A., 1995. Methodology for predicting for
element method. Available laboratory and field data winter conditions in the NATO reference mobility model. 5th
North American Conf. of the ISTVS. Saskatoon, SK, Canada,
were used to obtain suitable material parameters and
May 1995, pp. 320 – 334.
calibrate the model. Two sets of material parameters Alger, R., 1988. Effect of snow characteristics on shear strength.
were developed, one for describing the behavior of Contract Report No. DACA89-88-K-004. Keweenaw Research
bsoftQ snow, and the second for describing the Center, Michigan Technological University, Houghton, MI.
response of bhardQ snow. The derivation of these Alger, R., Osborne, M., 1989. Snow characterization field data
collection results. Final Report No. ACA89-85-K-002. Kewee-
two cases for describing snow in this density range
naw Research Center, Michigan Technological University,
help to capture the structure-dependent differences in Houghton, MI.
snow strength attributed to snow type and to age Armstrong, R.L., 1980. An analysis of compressive strain in
hardening or sintering of the snow over time. These adjacent temperature-gradient and equi-temperature layers in a
models are not intended to represent a specific snow natural snow cover. J. Glaciol. 26 (94), 283 – 289.
Aubel, T., 1993. FEM simulation of the interaction between elastic
microstructure, but rather provide description of the
tyre and soft soil. 11th International Conference of the ISTVS,
baverageQ snow behavior in this density and temper- Lake Tahoe, NV, vol. 2, pp. 791 – 802.
ature range. Aubel, T., 1994. The interaction between the rolling tyres and soft
The snow models were compared to laboratory and soil FEM Simulation by VENUS and validation. 6th European
field measurements of three load cases: radially Conf. Of the ISTVS, Vienna, vol. 1, pp. 169 – 188.
Bailey, A.D., Johnson, C.E., 1989. A soil compaction model for
confined uniaxial compression, plate sinkage tests,
cylindrical stress states. Trans. ASAE 32 (3), 822 – 825.
and a wheel rolling through new fallen snow. This Blaisdell, G.L., Richmond, P.W., Shoop, S.A., Green, C.E., Alger,
illustrates the application and validation of the R.G., 1990. Wheels and tracks in snow: validation study of the
constitutive model under a wide variety of geometries CRREL shallow snow mobility model. CRREL Report No. 90-
and loading conditions. The agreement for all of these 9. U.S. Army Cold Regions Research and Engineering
Laboratory, Hanover, NH.
geometries is very good. It is clear from a two-layer
Chi, L., Tessier, S., 1995. Finite element analysis of soil compaction
simulation of plate sinkage tests conducted in the field reduction with high flotation tires. 5th North American Conf. of
that the use of the soft snow (S200) parameters for the the ISTVS, Saskatoon, SK, Canada, pp. 167 – 176.
top layer and the hard snow (H200) parameters for the Fervers, C.W., 1994. FE simulations of tyre-profile effects on
bottom layer improves the ability of the simulation to traction on soft soil. 6th European Conf. of the ISTVS, Vienna,
Austria, pp. 618 – 633.
reproduce the measured load trace. The plate sinkage
Foster, W.A., Johnson, C.E., Raper, R.L., Shoop, S.A., 1995. Soil
and wheel-on-snow models illustrate the ability to deformation and stress analysis under a rolling wheel. North
also model the compaction density of the snow as it is American Conf. of the ISTVS. Saskatoon, SK, Canada, May
loaded. 1995, pp. 194 – 203.
R.B. Haehnel, S.A. Shoop / Cold Regions Science and Technology 40 (2004) 193–211 211

Fukue, M., 1979. Mechanical Performance of Snow under Loading. Richmond, P.W., Shoop, S.A., Blaisdell, G.L., 1995. Cold regions
Tokai Univ. Press. (based on author’s thesis, McGill University, mobility models. CRREL Report 95-1. U.S. Army Cold
1977). Regions Research and Engineering Laboratory, Hanover, NH.
Green, C.E., Blaisdell, G.L., 1991. US Army Wheeled Versus Saliba, J.E., 1990. Elastic–viscoplastic finite-element program for
Tracked Vehicle Mobility Performance Test Program. Report no. modeling tire–soil interaction. J. Aircr. 27 (4), 350 – 357.
2: Mobility in shallow snow. U.S. Waterways Experiment Schofield, A.N., Wroth, C.P., 1968. Critical State Soil Mechanics.
Station, Technical Report GL-91-7. McGraw-Hill, London, England.
HKS (Hibbitt, Karlsson and Sorensen), 1998. ABAQUS Theory and Scott, R.F., 1985. Plasticity and constitutive relations in soil
User’s Manuals. Pawtucket, RI. mechanics. J. Geotech. Eng. 11 (5), 563 – 605.
HKS (Hibbitt, Karlsson and Sorensen), 2001. Analysis of Geo- Shapiro, L., Johnson, J., Sturm, M., Blaisdell, G., 1997. Snow
technical problems with ABAQUS. ABAQUS Course Notes. Mechanics: Review of the State of Knowledge and Applica-
Pawtucket, RI. tions. CRREL Report 97-3. U.S. Army Cold Regions Research
Jaeger, J.C., Cook, N.G.W., 1969. Fundamentals of Rock Mechan- and Engineering Laboratory, Hanover, NH.
ics. John Wiley & Sons, New York. Shoop, S.A., 2001. Finite element modeling of tire–terrain
Lade, P.V., Kim, M.K., 1995. Single hardening constitutive model interaction. PhD dissertation, University of Michigan.
for soil, rock and concrete. Int. J. Solids Struct. 32 (14), Shoop, S., Alger, R., 1998. Snow deformation beneath a vertically
1963 – 1978. loaded plate. Proceedings, ASCE Cold Regions Specialty
Liu, C., 1994. Traction Mechanisms of Automobile Tires on Snow. Conference. Duluth, Minnesota.
PhD dissertation, Vienna University of Technology. Vienna, Shoop, S., Haehnel, R., Kestler, K., Stebbins, K., Alger, R., 1999.
Austria. Finite element analysis of a wheel rolling in snow. Proceed-
Liu, C.H., Wong, J.Y., 1996. Numerical simulations of tire–soil ings ASCE Cold Region Engineering Conference, Lincoln, NH,
interaction based on critical state soil mechanics. J. Terramechs. pp. 519 – 530.
33 (5), 209 – 222. Shoop, S., Haehnel, R., Kestler, K., Stebbings, K., Alger, R.,
Meschke, G., Liu, C., Mang, H.A., 1996. Large strain finite-element 1999. Snow–tire FEA. Tire Technology International, June,
analysis of snow. J. Eng. Mech., July, 581 – 591. pp. 20 – 25.
Miyori, A., Shiraishi, M., Yoshinaga, H., Naoaki, I., 2002. Shoop, S., Lacombe, J., Haehnel, R., 2001. Modeling tire perform-
Simulation of tire performance on snow. Proceedings Interna- ance for winter conditions, tire technology international. Annual
tional Tire Exhibition and Conference. Akron, OH. Review of Tire Material and the Tire Manufacturing Technol-
Mundl, R., Meschke, G., Leiderer, W., 1997. Friction mechanism ogy, Dec., pp. 10 – 14.
of tread blocks on snow surfaces. Tire Sci. Technol. 25 (4), Voitkovsky, K.F., Bozhinsky, A.N., Golubev, V.N., Laptev, M.N.,
245 – 264. Zhigulsky, A.A., Slesarenko, Yu.Ye., 1975. Creep induced
Pi, W.S., 1988. Dynamic tire/soil contact surface interaction model changes in structure and density of snow, international
for aircraft ground operations. J. Aircr. 25 (11), 1038 – 1044. symposium on snow mechanics. IAHS–AISH Publication, vol.
Richmond, P.W., 1995. Motion resistance of wheeled vehicles in 114. Grindelwald, Switzerland, pp. 171 – 179.
snow. CRREL Rpt. 95-7. U.S. Army Cold Regions Research Wood, D.M., 1990. Soil Behavior and Critical State Soil Mechanics.
and Engineering Laboratory, Hanover, NH. Cambridge Univ. Press, New York.

You might also like