Direct shear behaviour of dry, granular soils for low

normal stress with application to lightweight robotic

vehicle modelling
C. Senatorea,, K. D. Iagnemmaa
a Department

of Mechanical Engineering, MIT, Cambridge, MA 02139, USA.

Abstract
Modelling of soil shearing behaviour under wheeled or tracked vehicles requires
the knowledge of three soil properties: cohesion, angle of internal friction, and
shear modulus. For lightweight robots it is necessary to characterize the soil for
small normal stress (<15kPa) while most of the data collected in the literature
regards higher stress testing conditions. Soil failure at low stress may diverge
from Mohr-Coulomb envelope invalidating the fundamental assumptions behind
classical terramechanics approach. Through the analysis of direct shearing performance of a dry, granular, soil, this paper address several issues related to off-road
traction mechanics (not necessarily limited to low stress cases). We present an
improved approach for shear modulus calculation that overcome the inaccuracies
introduced by Wong method. We analyze the importance of density in modifying
terrain response. Moreover, we show how erroneous estimation of soil cohesion
and angle of internal friction may limit the applicability of Bekker/Wong theory
to lightweight tracked vehicles.
Keywords: Direct Shear Test, Mojave Mars Simulant, Shear modulus, Low
stress, Lightweight robot

1. Introduction
Traction generation on deformable soil is intimately dependent on soil shearing properties. Wheels and tracks provide traction by inducing shear failure in
Corresponding author

Email addresses: senator@mit.edu (C. Senatore), kdi@mit.edu (K. D. Iagnemma)

c ISTVS 2011
Preprint submitted to

August 31, 2011

Carmine Senatore and Karl D. Iagnemma

the mass of soil beneath them. [14, 16] have visually shown that flow of soil
under running gears follows failure envelope predicted by theory of plastic equilibrium. Several studies by Wong [12] highlighted that good correlation exists
between measured shear stress under a track and direct shear tests results. Soil
shearing fundamental properties are cohesion and angle of internal friction. These
can be obtained through three main testing methodologies: tri-axial compression
tests, direct (or ring) shear test, vane shear tests. Each methodology has its own
strengths and weakness and it is beyond the scope of this study to discuss them.
Direct shear test is chosen here because it is the testing methodology that closely
mimic soil failure behaviour under a running gear. Mohr-Coulomb failure criterion is typically employed to describe soil shear strength. This consists in a linear
envelope that bounds admissible shear-normal stress configurations [3]. However,
this criterion provides a condition for soil failure but it does not provide any information regarding shear vs. displacement trend. Different shearing behaviours
can be summarized under three main categories (also shown in figure 1):
A Shear stress asymptotically reaches residual state.
B Shear stress rises to a maximum, then decreases to reach steady residual
state.
C Shear stress rises to a maximum, then monotonically decreases without
reaching steady residual stat.e
A and B behaviours have been observed for a wide range of mineral terrains
while C behaviour is typical of organic soils (i.e., muskeg) and wont be considered in this study. Janosi and Hanamoto [4] proposed an empirical function to
describe shear vs. displacement for A behaviour. The model is based on MohrCoulomb failure criterion augmented with an additional parameter k, referred as
shear modulus. Bekker [2], Wong [13], and Oida [6] have introduced other fitting curves for type-B curves; these implementations are again based on MohrCoulomb envelope with the addition of some empirically tuned scaling function.
A part from shear stress trend it should be noted that the vast majority of data
available in the literature regards data collected for large normal stress. In the last
15 years, classical terramechanics models have been employed for mobility analysis of lightweight rovers. However, discrepancies noticed by several researchers
warrant for new models dedicated to lightweight vehicles.
The paper will present an improved method for shear modulus calculation in
2, a discussion of type A and B behaviour for granular soils in 3 , and a case study
of lightweight tracked vehicle modelling in 4.

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

Type A

Type B
Type C

\ max

0.8
0.6
0.4
0.2
0
0

0.2

0.4

0.6

0.8

Normalized Displacement
Figure 1: Idealized shear displacement curves for type A, B, and C behaviour.

2. Shear Modulus Calculation

Wong proposed an algorithm for automated calculation of shear modulus through
direct (or ring) shear tests [15]. Method is based on least square minimization and
it was developed for type A, B, and C. Notwithstanding its diffusion among researchers [12, 7], the algorithm introduced some simplifications that ultimately
invalidated fit accuracy at low displacement. This section will focus on type A
behaviour as modeled by Janosi and Hanamoto:


j
k
(1)
= max 1 e
where j is shear displacement and k is the fitting constant labeled shear modulus
(max will be discussed later). Wong rearranged equation 1 and, through least
square minimization, came up with a closed form equation for k:


i 2 2
1 max ji
i=1
k= n 



i 2
i
1 max ji log 1 max
i=1
n

(2)

where subscript i refers to the experimental data points, n is the number of samples
collected, and max is maximum shear measured. Disappointedly, the simplifica3

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

tions introduced by Wong severely deteriorate estimation accuracy. The algorithm

was developed in an era when computational power was limited and model simplification was necessary to make it run quickly. Such approximations can be
avoided now that computers computational power is not a limitation. In particular, equation 2 is sensitive to the chosen shear displacement range. This is shown
in figure 2 where an experimental direct shear dataset is compared with equation
1 fitted with Wong method and exact least square method. It is clear that Wong
approach introduces inaccuracies that will prove to be critical in section 4.
1500

Shear Stress [Pa]

Shear Stress [Pa]

1500

1000

Experimental
Wong fit k = 0.0013 m
Exact fit k = 0.00059 m
500

0
0

Displacement [m]

1000

Experimental
Wong fit k = 0.0172 m
Exact fit k = 0.00063 m
500

0
0

0.02

0.04

0.06

0.08

0.1

x 10

Displacement [m]

Figure 2: Quality of fit of Janosi-Hanamoto equation using Wong approximation and exact least
square approach. Experimental data has been obtained through direct shear testing of a Mars
Mojave Simulant [1, 8] subjected to a normal pressure equal to 2.08kPa.

Problem can be fixed by removing the simplifying assumptions. The objective

function (least square approach) becomes:
n

(k, max ) = i max 1 e

i=1

ji
k

2

(3)

Parameters k and max can be found through standard minimization procedure:

(k, max )
= 0
k
(k, max )
= 0
max

(4)
(5)

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

3. Shear vs. Displacement

Shear Stress [Pa]

Terramechanics is a multi-disciplinary field that combines (principally) mechanical engineering and soil mechanics. This sometimes creates a disconnect
between fundamental knowledge of soil behaviour and its application to applied
problems. In particular here we highlight how in the terramechanics community
has not been clearly recognized that different shear vs. displacement behaviour
is not function of soil type but it is rather a function of soil state (at least for
dry, granular, cohesionless, soils). Behaviour A or B is function of density: same
material may exhibit A or B trend purely depending on soil state. This is a well
know phenomenon to granular physicists and geotechnical engineers but it is often overlooked by terramechanics investigators. In figure 3 direct shear test results
for Mars Mojave Simulant [1, 8] are presented. Data clearly show that type A or
B behaviour purely depends on soil density.

5000

= 2086 Pa
3
= 1.54 g/cm

4000

= 2086 Pa
3
= 1.70 g/cm

3000

= 5340 Pa
3
= 1.54 g/cm

2000

= 5340 Pa
3
= 1.70 g/cm
= 7050 Pa
3
= 1.54 g/cm

1000
0
0

= 7050 Pa
= 1.70 g/cm3
1

Displacement [mm]

Figure 3: Type A and B behaviour for Mojave Mars soil simulant. Same material, at different
density, presents distinct different shearing behaviour.

In section 2, shear modulus calculation for type A behaviour has been discussed. However, it has been shown that type B behaviour may occur for the
same material if density is varied. Bekker [2], Oida [6], and Wong [13] proposed
different methods for description of type B behaviour but none of these methods
was able to capture at the same time type A and B behaviour [5, 13]. Wan and
Guo [11] introduced a constitutive model based on Rowes work [9] which predicted fairly well the behaviour of a granular media under different density levels.
However, the model has only been tested for high stress levels and it requires

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

11 parameters in order to be initialized. Sela [10] hypothesized that shear displacement curve may be described by the sum of two exponential functions, one
accounting for the cohesive component and the other for the frictional component.
Although the distinction between cohesive and frictional behaviour is difficult to
be verified experimentally and it is probably not the correct interpretation of the
phenomena under study, the equation suggested by Sela is an appealing extension
of Janosi-Hanamoto for terramechanics analysis. Here we propose a model for
type A and B characterization based on Selas work, that includes the influence of
density. Total shear stress is calculated as the summation of two components:


k j 1 j 
kj
k
e
(6)
+ 1e
= res
k
First term introduces the parameter k that controls density influence while
second term governs the residual behaviour. Residual stress res is in fact constant
for a given stress level (see figure 3) and thus k can be used as a tuning parameter
to account for density dependence. Figure 4 shows how equation 6 fits experimental results. Holding max and k constant it is possible to model type A and B
behaviour with a single equation. It is interesting to note that the first term in the
modified Selas equation 6 closely follow the rate of dilation measured during the
test. This suggests that k is indeed related to density and equation 6 may represent a reduced order approximation of the model introduced by Wan and Guo [11].
However, this approach has not been fully validated yet and further investigations
are underway.
6000

Shear Stress [Pa]

5000
4000
3000

Experimental @ = 1.70 g/cm

Fit @ k = 0.7

2000

Experimental @ = 1.54 g/cm

1000
0
0

Fit @ k = 0.1

Displacement [mm]

Figure 4: Type A and B behaviour for Mojave Mars soil simulant tested at = 7050 Pa and fitted
through equation 6. Parameters k and res are held constant for both cases.

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

4. Lightweight Tracked Vehicles Case Study

In this section it will be discussed how the aforementioned problems affect
lightweight off-road vehicles modelling. For a given material, shear vs normal
stresses at failure lay on a straight line in the plane, named failure envelope. However, definition of failure, for a granular material, can be vague. In
general, failure is considered to occur when large deformations occur. For type
B behaviour it is common to define failure at maximum shear stress while for
type A there is not clear presence of a maximum shear stress. For this reason
two envelopes can be defined: one at maximum shear stress and one at residual shear stress. In soil mechanics the failure envelope is traditionally known as
Mohr-Coulomb failure criterion and expressed as:

f = c + tan

(7)

This failure criterion is obtained from a series of direct shear tests (or triaxial
tests) where maximum or residual stress are plotted against normal stress and then
a failure envelope is calculated. Figure 5 presents a Mohr-Coulomb envelope obtained from direct shear tests of Mars Mojave Simulant [1, 8]. Each experimental
point correspond to a peak in curves for 1.7 g/cm3 presented in figure 3. Slope
and intersect of the linear envelope correspond to angle of internal friction and
cohesion c respectively. From this procedure it is clear that (apparent) cohesion
is calculated through a minimization procedure that is only loosely related to true
material cohesion. In fact it is not uncommon for the envelope to cross the y-axis
at negative values: in such conditions cohesion is typically assumed to be zero.
7000
6000

Experimental Points
MohrCoulomb Envelope

[Pa]

5000
4000
3000
2000
1000
0
0

1000

2000

3000

4000

5000
[Pa]

6000

7000

8000

9000

10000

Figure 5: An example of Mohr-Coulomb envelope obtained from direct shear tests of Mars Mojave
Simulant at 1.7 g/cm3 .

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

These sort of approximations have been accepted by geotechnical engineer and

large vehicles designers because they usually worked well for high normal stress
applications. However, when vehicle mass is modest, average ground pressure is
reduced. This creates a situation where and c are of the same order of magnitude
and thus, even for low-cohesion soils, it may be an unreasonable approximation
to set c = 0. Apparent cohesion values for dry, granular soils, usually ranges
between 0.5 and 2kPa. For a small rover weighting 15 kg, with 45 cm tracks of
R
) , average ground
12 cm width (typical dimensions of a iRobot 510 Packbot
pressure would be around 1.5 kPa. Clearly, in this case c and are of the same
order of magnitude and crudely assuming c = 0 may invalidate the results. The
concept is visualized in figure 6. Starting from a set of terrain parameters taken
from Wong [12], cohesion is taken equal to zero and equal to nominal, tabulated
value, for a dry sand. Shear prediction is obtained through Janosi-Hanamotos
equation.

Shear Stress [Pa]

2000

1500

1000

500

0
0

Vehicle average ground pressure

Zero cohesion
Nominal cohesion
0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

Track Length [m]

Figure 6: Shear stress profile under the track of a small rover as predicted through JanosiHanamotos equation. Slip is assumed to be 30% and cohesion is set either to zero or nominal
value (960Pa, from Wong [12]).

In 2 and 3 we have shown how shear modulus is typically overestimated and

how density plays a crucial role in defining shearing properties of granular soils.
However, for large vehicles, shear deformation is usually quite large and these
effects may not really influence traction performance estimation. For instance, a
tank with 5 meters long tracks, slipping a 10%, would produce a shear displace8

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

ment of 50 cm, far beyond the few millimeters investigated by direct shear tests
(see figure 1). On the other hand, a 45cm track, slipping ad 10% only displaces
4cm of soil. If any discontinuity occurs (rocks, track compliance, terrain unevenness), soil deformation can easily go down to the order of 1cm. Thus, for this
class of vehicles, what happens in the very first millimeter of soil deformation can
significantly change performance predictions. The concept is presented in figure
7 where experimental shear stress curves are presented over a domain of 1 cm (a)
and 45 cm (b).
2000

2500

Shear Stress [Pa]

Shear Stress [Pa]

2000

1500

1000
3

= 1.55 g/cm

= 1.60 g/cm3

500

1500

1000
3

= 1.55 g/cm

= 1.60 g/cm3
500

= 1.70 g/cm3

= 1.70 g/cm
0
0

Displacement [m]

(a)

6
3

x 10

0
0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Displacement [m]

(b)

Figure 7: (a) Shows shear stress curves at three different densities for 2.86 kPa of normal stress.
(b) presents the same results extended over a 45 cm length. It is obvious that shear modulus and
shear behaviour A, B, and C, have less influence when shear deformation is large.

Figure 8 presents predicted thrust variations for a 45 cm long track that exerts
2.86kPa average ground pressure. 1 point of contact means that the whole track
uniformly distribute its pressure on the soil. 10 points of contact means that the
track looses contact with the soil for 10 times because of uneven terrain profile,
suspension mechanism, rocks, etc. The plot shows that soil density significantly
influences thrust prediction.
5. Conclusions
We have shown that it is possible to calculate with more accuracy shear modulus of granular materials. Moreover, the influence of density on soil shearing
behaviour has been clarified. These aspects have been discussed in details to
highlight the limitations of Bekker/Wong models when applied to lightweight vehicles. It should be remembered that Bekker clearly stated [2] that his models
were scale dependent. However, in the last 15 years, researchers in need of a
9

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

30

Thrust Variation [%]

= 1.55 g/cm3
= 1.60 g/cm3

20

= 1.70 g/cm3
10

10

20
1

10

Number of points of contact

Figure 8: Thrust variation for a 45 cm tracked vehicles that travels over the same soil compacted at
different densities. 1 point of contact means that the whole track uniformly distribute its pressure
on the soil. 10 points of contact means that the track looses contact with the soil for 10 times
because of uneven terrain profile, suspension mechanism, rocks, etc.

model lightweight vehicles have often overlooked that limitations. For lightweight
robots it is necessary to characterize the soil for small normal stress (<15kPa) in
order to have more accurate measure of cohesion and it is crucial to understand
the influence of density on soil behaviour. A unified model for shear stress vs.
displacement curve that combine A and B behaviour has been introduced.
References
[1] L. W. Beegle, G. H. Peters, G. S. Mungas, G. H. Bearman, J. A. Smith, and
R. C. Anderson. Mojave Martian Simulant: A New Martian Soil Simulant.
In Lunar and Planetary Institute Science Conference Abstracts, 2007.
[2] M. G. Bekker. Introduction to Terrain-Vehicle Systems. The University of
Michigan Press, Ann Arbor, 1969.
[3] N. E. Dowling. Mechanical Behavior of Materials. Pearson/Prentice Hall,
Upper Saddle River, N.J, third edition, 2007.
[4] Z. Janosi and B. Hanamoto. Analytical determination of drawbar pull as
a function of slip for tracked vehicles in deformable soils. In Proceedings

10

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

of the 1st International Conference on Terrain-Vehicle Systems, Turin, Italy,

1961.
[5] K. Kogure, H. Yamaguchi, and Y. Ohira. Comparison of strength and soil
thrust characteristics among different soil shear tests. Journal of Terramechanics, 25(3):201 221, 1988.
[6] A. Oida. Study on equation of shear stress-displacement curves. Technical Report Technical Report No. 5, Farm Power and Machinery Laboratory,
Kyoto University, Japan, 1979.
[7] H.A. Oravec, X. Zeng, and V.M. Asnani. Design and characterization of
grc-1: A soil for lunar terramechanics testing in earth-ambient conditions.
Journal of Terramechanics, 47(6):361 377, 2010.
[8] Gregory H. Peters, William Abbey, Gregory H. Bearman, Gregory S.
Mungas, J. Anthony Smith, Robert C. Anderson, Susanne Douglas, and
Luther W. Beegle. Mojave mars simulantcharacterization of a new geologic mars analog. Icarus, 197(2):470 479, 2008.
[9] P.W. Rowe. The stress-dilatancy relation for static equilibrium of an assembly of particles in contact. Proceedings of Royal Society, A(269):500527,
1962.
[10] A.D. Sela. The shear stress-deformation relationship of soils. Journal of
Terramechanics, 1(1):31 37, 1964.
[11] R. G. Wan and P. J. Guo. A simple constitutive model for granular soils:
Modified stress-dilatancy approach. Computers and Geotechnics, 22(2):109
133, 1998.
[12] J. Y. Wong. Terramechanics and Off-Road Vehicle Engineering. Elsevier,
UK, 2nd edition, 2010.
[13] J. Y. Wong and J. Preston-Thomas. On the characterization of the shear
stress-displacement relationship of terrain. Journal of Terramechanics,
19(4):225234, 1983.
[14] J. Y. Wong and A. R. Reece. Prediction of rigid wheel performance based on
analysis of soil-wheel stresses, Part II. Performance of towed rigid wheels.
Journal of Terramechanics, 4(2):725, 1967.
11

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011

Carmine Senatore and Karl D. Iagnemma

[15] J.Y. Wong. Data processing methodology in the characterization of the mechanical properties of terrain. Journal of Terramechanics, 17(1):13 41,
1980.
[16] S.X. Wu, J.H. Hu, and J.Y. Wong. Behaviour of soil under a lugged wheel.
In Proceedings of 8th ISTVS International Conference, Cambridge, U.K.,
1984.

12

Proceedings of the 17th ISTVS International Conference. Blacksburg, VA, USA. September 18-22, 2011