Vous êtes sur la page 1sur 18

Journal

of
Terramechanics
Journal of Terramechanics 44 (2007) 221–238
www.elsevier.com/locate/jterra

On the impact of cargo weight, vehicle parameters, and terrain


characteristics on the prediction of traction for off-road vehicles
Lin Li *, Corina Sandu
Advanced Vehicle Dynamics Laboratory, Center for Vehicle Systems and Safety, Mechanical Engineering Department,
Virginia Politechnic Institute and State University, Blacksburg, VA 24060, USA

Received 4 August 2006; received in revised form 23 March 2007; accepted 17 April 2007
Available online 12 June 2007

Abstract

A realistic prediction of the traction capacity of vehicles operating in off-road conditions must account for stochastic variations in the
system itself, as well as in the operational environment. Moreover, for mobility studies of wheeled vehicles on deformable soil, the selec-
tion of the tire model used in the simulation influences the degree of confidence in the output. Since the same vehicle may carry various
loads at different times, it is also of interest to analyze the impact of cargo weight on the vehicle’s traction.
This study focuses on the development of an algorithm to calculate the tractive capacity of an off-road vehicle with stochastic vehicle
parameters (such as suspension stiffness, suspension damping coefficient, tire stiffness, and tire inflation pressure), operating on soft soil
with an uncertain level of moisture, and on a terrain topology that induces rapidly changing external excitations on the vehicle. The anal-
ysis of the vehicle–soil dynamics is performed for light cargo and heavy cargo scenarios. The algorithm relies on the comparison of the
ground pressure and the calculated critical pressure to decide if the tire can be approximated as a rigid wheel or if it should be modeled as
a flexible wheel. It also involves using previously-developed vehicle and stochastic terrain models, and computing the vehicle sinkage,
resistance force, tractive force, drawbar pull, and tractive torque.
The vehicle model used as a case study has seven degrees of freedom. Each of the four suspension systems is comprised of a nonlinear
spring and a viscous (linear or magneto-rheological) damper. An off-road terrain profile is simulated as a 2-D random process using a
polynomial chaos approach [Sandu C, Sandu A, Li L. Stochastic modeling of terrain profiles and soil parameters. SAE 2005 transac-
tions. J Commer Vehicles 2005-01-3559]. The soil modeling is concerned with the efficient treatment of the impact of the moisture content
on relationships critical in defining the mobility of an off-road vehicle (such as the pressure–sinkage [Sandu C et al., 2005-01-3559] and
the shear stress–shear displacement relations). The uncertainties in vehicle parameters and in the terrain profile are propagated through
the vehicle model, and the uncertainty in the output of the vehicle model is analyzed [Sandu A, Sandu C, Ahmadian M. Modeling mul-
tibody dynamic systems with uncertainties. Part I: theoretical and computational aspects, Multibody system dynamics. Publisher:
Springer Netherlands; June 29, 2006. p. 1–23 (23), ISSN: 1384-5640 (Paper) 1573-272X (Online). doi:10.1007/s11044-006-9007-5; Sandu
C, Sandu A, Ahmadian M. Modeling multibody dynamic systems with uncertainties. Part II: numerical applications. Multibody system
dynamics, vol. 15, No. 3. Publisher: Springer Netherlands; 2006. p. 241–62 (22). ISSN: 1384-5640 (Paper) 1573-272X (Online).
doi:10.1007/s11044-006-9008-4]. Such simulations can provide the basis for the study of ride performance, handling, and mobility of
the vehicle in rough off-road conditions.
 2007 ISTVS. Published by Elsevier Ltd. All rights reserved.

Keywords: Off-road vehicle; Tire model; Soil properties; Terrain topology; Uncertainties; Stochastic model; Polynomial chaos; Mobility; Traction; Vehicle
dynamics

1. Introduction

The challenges in modeling the vehicle–unprepared


*
Corresponding author. Tel.: +1 540 231 7787; fax: +1 540 231 9100. terrain system are, to a large extent, due to high nonlinear-
E-mail address: lil@vt.edu (L. Li). ities in, and large uncertainties of the vehicle–terrain

0022-4898/$20.00  2007 ISTVS. Published by Elsevier Ltd. All rights reserved.


doi:10.1016/j.jterra.2007.04.002
222 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

Nomenclature

X uncertain parameter hrl vertical road excitation (displacement) on the


Xk kth component of X along the polynomial chaos rear-left wheel [m]
dimension hrr vertical road excitation (displacement) on the
S number of terms in the Karhunen–Loeve expan- rear-right wheel [m]
sion A collocation matrix
W orthogonal polynomial family Wload cargo weight [kg]
n random variable sfl sinkage of the front-left tires [m]
a distance from the front axle to the center of sfr sinkage of front-right tires [m]
mass [m] Wfl front–left wheel load [N]
b distance from the rear axle to the center of mass Wfr front-right wheel load [N]
[m] p normal pressure under the tire [kPa]
B track width [m] bw smallest dimension of the contact patch [m]
L vehicle wheelbase [m] n exponent of deformation
m2 mass of the sprung mass [kg] kc cohesive modulus [kN/mn+1]
J2y moment of inertia of the sprung mass around y k/ friction modulus [kN/mn+2]
axis [kg Æ m2] s vehicle sinkage [m]
J2x moment of inertia of the sprung mass around x s soil shear stress [kPa]
axis [kg Æ m2] j soil shear displacement [m]
m1 unsprung mass [kg] c cohesion of internal shearing resistance of the
k1 tire stiffness [N/m] soil [kPa]
k 01 amplitude of uncertainty in tire stiffness [N/m] ø angle of internal shearing resistance of the soil
c damping of the suspension [N s/m] []
c0 amplitude of uncertainty in damping of the sus- K soil shear deformation modulus [mm]
pension [N s/m] m soil moisture content [%]
k2 spring stiffness of the suspension [N/m] pcr critical pressure under the tire [kPa]
k 02 amplitude of uncertainty in spring stiffness of R radius of the wheel [m]
the suspension [N/m] z0 maximum vehicle sinkage [m]
z2 displacement of the sprung mass [m] h0 entry angle [rad]
h2 pitch angle of the sprung mass [rad] ht exit angle [rad]
/2 roll angle of the sprung mass [rad] 2hf angle corresponding to the deflected region of
Z1fl displacement of the front-left unsprung mass [m] the flexible wheel model [rad]
Z1fr displacement of the front-right unsprung mass i wheel slip
[m] x distance from point A to point B in Fig. 6b [m]
Z1rl displacement of the rear-left unsprung mass [m] Rc resistance force, [N]
Z1rr displacement of the rear-right unsprung mass Nfl calculated wheel load [N]
[m] F tractive force [N]
hfl vertical road excitation (displacement) on the Fd drawbar pull [N]
front-left wheel [m] pagp average ground pressure [kPa]
hfr vertical road excitation (displacement) on the lt contact length of the tire and the soil [m]
front-right wheel [m] d tire deformation [m]

interaction. These interaction features impose restrictions the roll of the chassis, and the vertical motions of the four
on the stochastic modeling techniques to be used. An addi- suspension systems. Each of the four suspension systems
tional modeling difficulty is the inability to collect experi- is comprised of a nonlinear (cubic) spring and a viscous
mental data on the soil and terrain properties for all damper. Both linear dampers and nonlinear magneto-rheo-
future operation environments. logical (MR) dampers are modeled and analyzed. Magneto-
The method previously developed by the authors to han- rheological (MR) fluid dampers have the capability of
dle uncertainties in multibody dynamic systems using a changing their effective damping force depending on the
polynomial chaos approach [1,2] is employed here to model current input to the damper. Uncertainties in key vehicle
the interaction between an off-road vehicle with uncertain parameters, such as the suspension stiffness, the suspension
parameters and unprepared terrain. The vehicle model has damping coefficient, and the tire stiffness are considered and
seven degrees of freedom (DOF): the heave, the pitch, and quantified in the proposed framework.
L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238 223

In addition to the inherent vehicle characteristics, ter- chanics model and the proposed algorithm for evaluating
rain topology and soil parameters also affect the vehicle the tractive capacity. The off-road vehicle running with var-
performance. This study employs the modeling technique ious speeds, traversing on stochastic soil and on different
presented in [6] to simulate a terrain profile as a 2-D ran- terrain profiles while carrying different cargos is simulated.
dom process. In modeling the soil, the moisture content The resistance force, tractive force, drawbar pull, and trac-
is considered a stochastic variable. The effect of uncertain tive torque are calculated and graphed. The relation
soil conditions is incorporated in the simulation via the between the drawbar pull coefficient and the wheel slip is
pressure–sinkage and the shear stress–shear displacement obtained, as well as the relation between the tractive force
relations. and the sinkage. Numerical results and their analysis are
The 2-D terrain model and the soil model are used in presented. The paper ends with conclusions and proposed
this study as operating conditions for the simulation of future work.
the 7-DOF stochastic off-road vehicle model. The uncer-
tainties in vehicle parameters, as well as those in the soil 2. State-of-the-art in modeling vehicles operating off-road
properties and in the terrain profile are propagated through
the vehicle model, and the uncertainties in the outputs of A realistic analysis of vehicle dynamics in off-road
the vehicle model are analyzed. Such simulations allow conditions must account for the statistical variation of
the study of ride performance and mobility of the vehicle intrinsic system parameters, poorly understood soil char-
in realistic off-road conditions. acteristics, and uncertain terrain topology. Going beyond
An important aspect that affects both the efficiency and the deterministic approach and treating these compo-
the accuracy of the terramechanics simulation results is the nents in a stochastic framework is an important step for-
tire model. When choosing a tire model, the decision is ward in bringing the modeling results closer to real-life
based on several factors: vehicle type and wheel/tire char- situations.
acteristics (e.g., agricultural machines, military combat Stochastic modeling has gained recognition and popu-
vehicles, construction equipment, recreational vehicles, larity in the recent years and is not uncommon practice
planetary exploration vehicles); computational restrictions for disciplines such as fluid dynamics and structural engi-
(e.g., real-time simulation, off-line analysis, design-targeted neering. Current methods used to treat systems with uncer-
computation); and operating environment (e.g., rough tainties include Monte Carlo ensembles [1], linear and
hard terrain, soft soil, organic surfaces, fresh snow, icy nonlinear approximations [1,24], perturbation methods
trails). [1,24], Neumann expansion [1,24], and hierarchy closure
To aid the simulation of off-road vehicles, this study approximation [1,24]. While each of these techniques prove
provides an algorithm that takes advantage of both to be valuable in specific areas of application, each is con-
wheel/tire modeling approaches: rigid and flexible. As pre- strained by assumptions that deem the method inefficient
sented earlier, since the model operates under several or inappropriate for treating the uncertainties encountered
uncertainties, it is possible that, during the simulation, in a vehicle-terrain system. The Monte Carlo method is
one tire model is more appropriate than the other type, computationally intense. Approximation methods may fail
due for example to variations in terrain conditions. Thus, to capture essential features of highly nonlinear systems.
we propose an algorithm that uses the comparison of the The perturbation and the Neumann expansion methods
ground pressure and the calculated critical pressure to only produce acceptable results for small fluctuations.
decide – during the simulation – if the tire must be approx- The difficulty in formulating high-order closure approxi-
imated as a rigid wheel, or if it should be modeled as a flex- mations limits the applicability of hierarchy closure
ible wheel. The algorithm calculates the vehicle sinkage, the approximation to cases of small magnitudes of uncertain-
resistance force, the tractive force, the drawbar pull, and ties. An extensive review of literature, presenting the pros
the tractive torque. If the sinkage of the vehicle is signifi- and cons of these and other stochastic methods is found
cantly larger than the deflection of the tire, then a rigid in [1,2].
wheel model is used. The accuracy of the model is not sig- The polynomial chaos approach [1–6,22–28] can simu-
nificantly affected, but the gain in computational efficiency late systems with high nonlinearities and large uncertain
is substantial. If the operating surface is hard and the tire magnitudes in an efficient and accurate manner. It has been
deflection is substantial, a flexible wheel model is used. This applied to structural mechanics [24], fluids mechanics [25],
approach saves computation time and assures realistic plate and beam vibrations [24], biological systems [26], air
results throughout the simulation. quality systems [27], and environment models [28]. Previ-
The paper presents the state-of-the-art in modeling off- ous work published by the authors [1,2] focused on devel-
road vehicles and in addressing uncertainties in multibody oping a computationally-efficient methodology, based on
dynamic systems. A brief review of the polynomial chaos the polynomial chaos approach, to quantify uncertainties
approach which is employed by the authors to address resulting from different sources, and propagate the uncer-
the uncertainties is given next. Following the description tainties through the model of a multibody dynamic system.
of the 7-DOF vehicle model, of the 2-D terrain model, The method was validated against Monte Carlo
and of the stochastic soil model, we present the terrame- simulations, and was shown to be equivalent in terms of
224 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

its accuracy, to require less effort, and to process faster if road profiles are characterized as Markov Chains using
the system’s complexity increases. statistical method. In [6], a more realistic approach is
Because the terrain profile and the soil characteristics taken to facilitate vehicle dynamics simulation by using
play an important part in defining the mobility and perfor- a stochastic 2-D terrain model (i.e. the terrain height is
mance of a vehicle, substantial effort has been placed on a function of the distance in the longitudinal direction
developing experimental techniques to measure them, and and in the lateral direction). The effect on vehicle dynam-
computational methods to model them. From the experi- ics, caused by geometric uncertainties in terrain profile, is
mental point of view, traditional ways to measure soil explored.
mechanics are reviewed in [7,8,13]. Test data of various Methodologies to predict forces on off-road vehicles are
kinds of soil are provided in [7,13,14]. However, the soil reviewed in [12]. The works of Bekker, Reece, Janosi,
models used in those studies are deterministic, and require Hanamoto, and other researchers on tractive capabilities
accurate measurements. From the theoretical point of view, are outlined, and the compatibility of those theoretical
a soil parameter identification algorithm is introduced and and semi-empirical models is explored. Derivations of rigid
tested in [16], by which the cohesion of internal shearing wheel models and flexible wheel models are presented in
resistance of the soil c and the angle of internal shearing [7,8]. Using a deterministic soil model, the vehicle–soil
resistance of the soil / can be estimated. In [6], the stochas- interaction using the rigid wheel model and using the flex-
tic normal pressure–sinkage relationship is studied, which, ible wheel model are studied, respectively, in [10,11,14]. A
for moisture content fluctuations, the model predicts the simplified rigid wheel model for planetary exploration rov-
stochastic normal pressure as a function of sinkage and sto- ers on different soils is described in [10]. A modified flexible
chastic soil parameters. In this study, the stochastic shear wheel model for vehicles on agricultural soil is discussed in
stress–shear displacement relationship is obtained, and [11,14]. In [15], a wheel sinkage and slippage estimation
the vehicle–soil interaction is studied based on the stochas- method for rigid wheels of mobile robots is described and
tic soil model. tested. The tire model choice algorithm is not studied in
Visual technology (e.g., cameras), and advanced sensors those previous works.
(e.g., remote sensing methods, laser, and radar) are used
to detect high-resolution terrain profiles. Such techniques 3. Treating uncertainties using the polynomial chaos method
have been used to aid the mobility of various robots
and wheeled vehicles, especially for planetary exploration The uncertain vehicle or terrain/soil parameters can be
[10,19]. Due to the limitations imposed by the computa- treated as (functional of) random variables and can be rep-
tional ability and the high cost of the equipment, these resented using the polynomial chaos expansion [1]:
real-time methods are appropriate mainly for low-speed X
S
vehicles, such as those performing tasks in space or under- X ðhÞ ¼ b k wk ðnÞ
X ð1Þ
water. One dimensional (1-D) terrain models (i.e. the ter- k¼1
rain height is only a function of the distance in one where wk(n) are orthogonal polynomials, S is the number
direction), while popularly used in vehicle dynamics simu- of terms of the expansion, and n is the random variable.
lation, do not allow the simulation of complex vehicle The superscript k is the polynomial order, and X b k is the
responses, such as roll and yaw. Another way of modeling kth component of X b along the polynomial chaos dimen-
a terrain profile is measuring different typical terrain pro- sion. The state variables of the multibody dynamic system
files and soils, saving the data, classifying the terrain, and are also functionals of the random variables that describe
pre-computing the control maneuverability [17–19]. The the sources of uncertainty.
difficulty in computing the data real-time limits the appli- The dynamics of a deterministic mechanical system with
cation of different control methods on the terrain–vehicle no constraints can be described by a set of simultaneous
interaction. In the program High-fidelity Ground Platform first-order differential equations (ODE):
and Terrain Mechanics Modeling (HGTM) Science and
Technology Objective (STO) [20], spline functions are used y_ ¼ v; vk ¼ F k ðt; y; v; pÞ; yðt0 Þ ¼ y 0 ; t0 6 t 6 tF ð2Þ
to produce higher-resolution terrain surface to overlay The stochastic ODE of the system deterministically
upon lower ones. In [21], a mathematical model of the defined by Eq. (2) is obtained by substituting in it all the
high frequency content of the ground surface is developed uncertain parameters with their Karhunen–Loeve expan-
using a set of Non-uniform Rational B-Splines (NURBS) sion, as given by Eq. (1). The stochastic system obtained
patches which are combined in the terrain query by super- is given in the following equation:
imposing them over the low-frequency polygonal terrain
to reintroduce the missing content. An imagery-based
approach to generating the mathematical surfaces and y_ jk ¼ vjk ;
!
Bump Maps from the database’s decal textures is also pre- X
S X
S X
S X
S

sented. In [22], the Delaunay triangulation and the T- vmk wm ðnÞ ¼ F k t; i i


y w ðnÞ; i i
v w ðnÞ; i i
p w ðnÞ ð3Þ
m¼1 i¼1 i¼1 i¼1
search algorithm are investigated to model the terrain by
searching the resulting triangular polygons. In [23], the yðt0 Þ ¼ y 0 ; t0 6 t 6 tF
L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238 225

In order to derive the evolution equations for the uncer- The loads on the front wheels can be calculated using
tain parameters, in [1,2] we present two approaches: the the following equation:
Galerkin method and the collocation method. The Galerkin  
ðW load þ m2 Þb
method projects Eq. (3) onto the span of basis functions. W fl ¼ k 1 ðhfl  z1fl  sfl þ dÞ þ 9:8  þ m1
2L
The collocation method obtains the evolution equations  
by enforcing Eq. (3) to hold at a given set of collocation vec- ðW load þ m2 Þb
W fr ¼ k 1 ðhfr  z1fr  sfr þ dÞ þ 9:8  þ m1
tors. We also showed that collocation can be regarded as the 2L
Galerkin method, with the ensemble averages evaluated ð5Þ
numerically with a specific choice of numerical quadrature.
In this paper, the uncertain parameters are assumed to
4.2. Terrain model
have either uniform distribution, in which case Legendre
polynomials are used as basis functions, or normal distri-
To simulate the dynamic response of the vehicle, in this
bution, in which case Hermite polynomials are employed.
study we used two different 2-D stochastic terrain profiles,
developed in [6] and illustrated here in Figs. 2 and 3.
4. Modeling elements
4.3. Soil model
4.1. Vehicle model
The mobility of vehicles on soft terrains (e.g., homoge-
A 7-DOF vehicle model is implemented to simulate the
nous, organic, snow), is determined by the pressure–sink-
response of the vehicle on different terrain profiles and to
age and shear stress–shear displacement relations, which
calculate the loads on wheels. The seven degrees of freedom
are established at the contact patch between the tire and
of the vehicle are the heave, pitch, and roll of the sprung
the soil. The normal pressure provides support to the vehi-
mass, and the bounces of the four unsprung masses. The
cle and cargo weight, and the shear stress is the largest con-
vehicle model, as shown in Fig. 1, consists of a body
tributor to the vehicle’s ability to propel, brake and corner.
(sprung mass), four suspension subsystems (unsprung
The pressure–sinkage and shear stress–shear displacement
masses) and four wheels. Each of the suspension includes
relations are very complex, as they depend on intrinsic
a nonlinear (cubic) spring and a viscous damper. The tire
parameters and external environmental factors such as
is modeled as a linear spring. The four suspensions are
the soil type and moisture content. Typically soil parame-
assumed to be identical, having stiffness k2 and damping
ters are measured experimentally. Empirical or semi-empir-
c. All of the four wheels, having stiffness k1 and mass m1,
ical approaches have been used to describe these relations,
are assumed to be identical.
as presented by Wong [7], Muro [8] and Bekker [9]. The
The set of ODE of motions of the vehicle model is
most commonly-used terramechanics relations are
obtained by Newton’s method and provided in [5]. The
uncertainties in the suspension stiffness, the suspension
damping coefficients, and the tire stiffness are quantified
as follows:
8
 0
< k 1 ¼ k 1 þ k 1 n2
>
c ¼ c þ c0 n3 ð4Þ
>
:
k 2 ¼ k 2 þ k 0 n4
2

Numerical simulations were run using the values


k 01 ¼ 80; 000 N=m, c 0 = 600 N s/m, k 02 ¼ 12; 000 N=m.
Fig. 2. Bump terrain [6].

Fig. 1. Seven DOF full-car model [5]. Fig. 3. Sinusoidal terrain [6].
226 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

Table 1
Pressure–sinkage parameters for Michigan sandy loam
Moisture m (%) n kc (kN/mn+1) k/ (kN/mn+2) c (kPa) / ()
11 0.9 52.53 1127.97 4.83 20
23 0.4 11.42 808.96 9.65 35

Bekker’s equation for the pressure–sinkage relation, shown


in Eq. (6), and Janosi–Hanamoto equation for the shear
stress–shear displacement relation, shown in Eq. (7):
 
kc
p¼ þ k / sn ð6Þ
bw
Fig. 4. Deterministic shear stress–shear displacement relation for Mich-
s ¼ smax ð1  ej=K Þ ¼ ðc þ p tan /Þð1  ej=K Þ ð7Þ igan sandy loam for five levels of moisture.
The moisture can be written using the Karhunen–Loeve
expansion as For numerical simulation, the parameters are chosen as
X
S bw = 15 cm and k = 10 mm. The stochastic treatment of
m¼ mj wj ðnÞ ð8Þ the pressure–sinkage relation for the Michigan sandy loam
j¼1
is analyzed in [6]. Here, we present in Fig. 4 the shear
As a case study we used a homogeneous soil type (Mich- stress–shear displacement relation for five distinct levels
igan sandy loam), with the parameters given in Table 1. of moisture content. Fig. 5a illustrates the deterministic
According to [6,7], the soil moisture influences the pres- relation of shear stress–shear displacement at 15% soil
sure–sinkage and shear stress–shear displacement relations. moisture, and the stochastic relationship obtained with
For simplicity, the dependence between the moisture con- using the polynomial chaos method, with Legendre poly-
tent and each of the parameters of these relations is nomials for the basis functions. The error bars represent
assumed to be linear, but the procedure is general and plus/minus one standard deviation. The probability density
can be used for other types of dependences if needed [6]. function of the shear stress versus shear displacement is
Based on the parameter Table 1 [6,7] and the linear shown in Fig. 5b.
models developed in [6] the uncertainty in the soil moisture
is propagated through Eqs. (6) and (7) using the colloca- 5. Ground pressure computation
tion method [1,2]; the stochastic pressure–sinkage model
and the stochastic shear stress–shear displacement model The critical pressure, calculated by Eq. (16) [7], is the
are thus obtained as given by Eqs. (9)–(15): pressure due to wheel loading:
 2nþ1 " #2nþ1
2n
X
S X
S
kc
1
3W fl
k jc wj ðnÞ ¼ 90:2143:426 mj wj ðnÞ ð9Þ pcr ¼ þ k/  pffiffiffiffiffiffi ð16Þ
j¼1 j¼1 bw ð3  nÞbw 2R
X
S X
S
This value plays an important role in determining the
k j/ wj ðnÞ ¼ 1420:426:6 mj wj ðnÞ ð10Þ
j¼1 j¼1 tire model. As illustrated in Fig. 6a, when the critical pres-
sure is less than the average ground pressure (pagp), the
X
S X
S
nj wj ðnÞ ¼ 1:3580:0417 mj wj ðnÞ ð11Þ wheel is modeled as rigid. The value for the average ground
j¼1 j¼1 pressure (pagp) has been set in step 1 of the algorithm at a
X
S X
S value larger than the tire inflation pressure. This is consid-
cj wj ðnÞ ¼ 0:4117þ0:407 mj wj ðnÞ ð12Þ ered appropriate, since, according to [7], the average
j¼1 j¼1 ground pressure is equal to the sum of the tire inflation
X
S X
S pressure plus the terrain pressure due to the carcass stiff-
/j wj ðnÞ ¼ 6:25þ1:25 mj wj ðnÞ ð13Þ ness. Under this criterion, the critical pressure dominates
j¼1 j¼1 the contact patch between the tire and the soil; when this
0 1 
P
S criterion is not met, the wheel is modeled as deformable
Al;j k jc PS
X
S Bj¼1 XS C Al;j nj
and the average ground pressure dominates the tire–soil
B C
Al;j pj ¼ B þ Al;j k j/ Cz j¼1 ð14Þ interaction, as shown in Fig. 6b.
j¼1
@ bw j¼1
A

!! 5.1. Rigid wheel


X
S X
S X
S X
S
Ai;j sj ¼ Ai;j cj þ Ai;j pj tan Ai;j /j ð1ej=K Þ
When the wheel is considered rigid, the contact pressure
j¼1 j¼1 j¼1 j¼1
between the tire and the soil is equal to the critical pressure,
ð15Þ and the maximum sinkage is determined by Eq. (17). For a
L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238 227

a Pressure at moisture 15% b 600


-1.6
400
-1.8
500

Shear Stress [KPa]

Shear Stress [kPa]


-2
300
400 -2.2
-2.4
200 300
-2.6

200 -2.8
100
Deterministic -3
PolyChaos 100 -3.2
0 -3.4
0 0.2 0.4 0.6 0.8 1 0
0 0.2 0.4 0.6 0.8 1
Shear Displacement [m]
Shear Displacement [m]

Fig. 5. Stochastic shear stress–shear displacement relation for Michigan sandy loam for a uniform distribution of the moisture content between 0% and
30%. (a) Stochastic simulation and (b) probability density function.

Fig. 6. Tire models. (a) Rigid wheel and (b) flexible wheel.

2-D tire model, the sinkage at any angle h in Fig. 6a, zh, is 5.2. Flexible wheels model
calculated by Eq. (18), and the shear displacement j by Eq.
(19). The actual load, Nfl, tractive force, F, resistance force, For the 2-D flexible wheel model, the portion of the
Rc, drawbar pull, Fd, and tractive torque on wheels, T, are circumference of the tire that is in contact with the soil
calculated along the contact circumference using Eqs. (20)– can be divided into three parts as shown in Fig. 6b: lead-
(24). ing part BC, flat part AB, and trailing part AD. Along
" #ð2nþ1Þ
2 the section BC, the shear displacement rises, and the nor-
3W fl mal pressure and the shear stress both increase as well.
z0 ¼ pffiffiffiffiffiffi ¼ Rð1  cos h0 Þ ð17Þ
bw ð3  nÞðk c =bw þ k / Þ 2R On the flat section AB, the normal pressure remains con-
stant and equal to the average ground pressure; however,
zh ¼ R cos h  R þ z0 ð18Þ
the shear stress increases with the shear displacement.
j ¼ R½ðh0  hÞ  ð1  iÞðsin h0  sin hÞ ð19Þ Along the section AD, both the normal pressure and
Z h0 Z h0 
the shear stress decrease while the shear displacement
N fl ¼ Rbw sðhÞ sin hdh þ pðhÞ cos hdh ð20Þ increases slightly.
0 0
Z h0 For the leading section BC (hf 6 h 6 h0), the maximum
F¼ sðhÞ cos h  R  bw dh ð21Þ sinkage and the sinkage at any angle are derived by Eqs.
0 (25) and (26), respectively. The shear displacement along
#2nþ2 "
2nþ1 the edge of BC is expressed by Eq. (27). The normal pres-
1 3W fl
Rc ¼ 1=ð2nþ1Þ
p ffiffiffiffiffi
ffi ð22Þ sure and the shear stress are updated using Eqs. (14) and
ðn þ 1Þðk c þ bw k / Þ ð3  nÞ 2R (15). The equations used to calculate for forces and torque
Z h0 Z h0 
are similar to those for the rigid wheel, with the limits of
F d ¼ Rbw sðhÞ cos hdh  pðhÞ sin hdh ð23Þ integration from hf through h0:
0 0
Z h0  1n
pagp
T ¼ bw R 2 sðhÞdh ð24Þ z0 ¼ ¼ Rð1  cos h0 Þ ð25Þ
0 k c =bw þ k /
228 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

zh:BC ¼ R cos h  R þ z0 ¼ Rðcos h  cos h0 Þ ð26Þ the terrain is obtained. Following this, the wheel load is
jBC ¼ R½ðh0  hÞ  ð1  iÞðsin h0  sin hÞ ð27Þ computed by Eq. (5), and the critical pressure calculated
by Eq. (16) is compared with the average ground pressure
For the flat section AB (hf 6 h 6 hf), the shear dis- at each time step to determine which tire model should be
placement increases along AB with distance, as shown in used to compute the forces and torque on the wheel. For
Eq. (28) (x denotes the distance from point A to point the rigid wheel model, the sinkage, normal pressure, shear
B), and the shear stress is updated by calculating it using stress, and contact forces are calculated directly. After
Eq. (15). The contact pressure is equal to the average that, the wheel load obtained by Eq. (20) is compared
ground pressure pagp. In this case, the forces and the torque with the one computed using Eq. (11). If they do not
are integrated from hf to hf by Eqs. 20,21,23 and 24 with agree, or the error exceeds the endurable limit, the sinkage
pagp replacing p. is updated and fed back to Eq. (6). For the flexible wheel
jAB ¼ jB þ ix ¼ R½ðh0  hf Þ  ð1  iÞðsin h0  sin hf Þ model, the tire deformation d and the exit angle ht are ini-
tialized first, then the resultant h0 and hf are obtained by
þ iðsin hf  sin hÞ ð28Þ
Eqs. (32) and (33), and lastly ht is initialized such that
For the trailing section AD (ht 6 h 6 hf), the sinkage h0 6 ht 6 hf:
and shear displacement are calculated by Eqs. (21) and R  z0  d
(23), respectively. The forces and the torque are calculated h0 ¼ cos1 ð32Þ
R pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
using Eqs. 20, 21, 23 and 24 integrating from ht to hf,
with p and s updated by Eqs. (14) and (15): lt 1 2 2Rd  d2
hf ¼ sin1 ¼ sin ð33Þ
2R 2R
zhAD ¼ Rðcos h  cos ht Þ ð29Þ
Z hf The forces in the contact patch are calculated for these
jAD ¼ jA þ R ð1  ð1  iÞ cos hÞ dh angles and sinkage. The calculated wheel load in Eq. (20)
h is checked with the one in Eq. (5). If they do not agree,
¼ R½ðh0  hf Þ  ð1  iÞðsin h0  sin hf Þ þ 2i sin hf  or the error exceeds the endurable limit, new values of d
 R½ðh þ hf Þ  ð1  iÞðsin hf þ sin hÞ ð30Þ and the exit angle ht are assigned, and the iterative compu-
tation is done until the optimal values are found. The algo-
For the whole tire, the tire–soil interaction is the combi- rithm is shown in Fig. 7.
nation of the three parts, DA, AB, and BC.
The resistance force is computed by Eq. (31) [7]: 5.4. Tire–soil interaction on stochastic soil model
 1n
bw kc ðnþ1Þ=n In Section 4.3 we discussed the proposed stochastic soil
Rc ¼  þ k/  pagp ð31Þ
n þ 1 bw model. Since the moisture level of the soil is considered ran-
dom, the soil parameters, the normal pressure, and the
5.3. Algorithm to predict the Vehicle Tractive Capacity shear stress in Eqs. (8)–(15) are used in Eqs. (17)–(24).
The Collocation method was used to construct the tire–soil
Off-road dynamics is affected mainly by the wheel interaction on the stochastic soil model.
model. The tire–soil interaction using a rigid wheel model Since the index i in Eqs. (8)–(15) is from 1 to S, an S · S
is astoundingly different than that using a flexible wheel matrix is obtained for forces and torques on wheels, and
model as the sinkage, the tire deformation, and the contact real values of the forces are recovered by pre-multiplying
patch vary greatly between these two approaches. The tire the S · S matrix by the inverse of the coefficient matrix Ai,j.
model status depends on the comparison of the average
ground pressure and the critical pressure. It has been 5.5. Tire–soil interaction with stochastic average ground
shown [6] that the critical pressure changes with the veloc- pressure
ity, terrain profile, cargo weight, suspension parameters,
and tire mechanics. If the critical pressure is less than the In this case, the soil model is deterministic, but the aver-
average ground pressure (depending on the tire inflation age ground pressure is uncertain due to the tire wear, tem-
and the carcass stiffness), the tire deformation is negligible perature changes, or other unexpected situation. The
with respect to the sinkage, and the wheel is seen as rigid, as deterministic pagp is replaced by the following equation:
shown in Fig. 6a. The critical pressure dominates the con- X
S

tact patch between the tire and the soil. If the average pagp ¼ pjagp wj ðnÞ ð34Þ
j¼1
ground pressure is greater than the critical pressure, the tire
deforms and is separated into three sections as shown in For rigid wheels, pagp is only used for comparison with
Fig. 6b, and the flexible wheel model is used in the vehi- pcr. For flexible wheels, the stochastic pagp replaces the
cle–soil interaction computation. deterministic one in Eqs. (25) and (31). S represents the
In the proposed algorithm, given the initial sinkage, the number of terms in the Karhunen–Loeve expansion illus-
response of the deterministic vehicle model operating on trated in Eq. (34).
L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238 229

Fig. 7. Algorithm to predict the tractive capacity of an off-road vehicle.

6. Numerical results Set A: Stochastic vehicle model, deterministic soil model


and terrain profile:
To analyze the proposed models and traction algorithm, Response of the vehicle on a 2-D deterministic ter-
several scenarios have been simulated. All runs involved rain profile with:
straight forward driving using the seven degree-of-freedom  uncertain magneto-rheological (MR) shock
vehicle model, on Michigan sandy loam. Simulation results absorbers damping characteristics;
are presented for the:  uncertain tire stiffness;
230 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

 uncertain suspension stiffness. response of the heave, pitch, and roll of the sprung mass,
Set B: Stochastic vehicle model, stochastic terrain and the bounces of the four unsprung subsystems are
model, deterministic soil model:Response of the plotted.
full vehicle model on a 2-D stochastic terrain pro- The set A of simulations considers the dynamic evolu-
file with: tion of the vehicle with uncertainties in the suspension stiff-
 uncertain suspension stiffness, suspension damp- ness, k2, the suspension damping, c, or the tire stiffness, k1,
ing, and uncertain tire stiffness. separately and combined. For simplicity, in each of these
Set C: Terramechanics of the deterministic vehicle model simulations the four suspension and tire systems and are
on a 2-D deterministic terrain profile with: assumed to be similar. The vehicle runs straight forward
 uncertain average ground pressure; with a constant speed of 10 m/s. Linear damping and mag-
 uncertain soil moisture. neto-rheological (MR) damping are both modeled and
Set D: Traction analysis: analyzed. Some of the most representative results are pre-
 the relation between the drawbar pull coefficient sented next.
and wheel slip; Fig. 8 illustrates the response of the vehicle under MR
 the relation between the tractive force and the damping uncertainty with uniform distribution in the
sinkage. damping parameter. The collocation method was used to
propagate the uncertainties through the system. We ana-
6.1. Numerical results using a stochastic vehicle model: lyzed two situations: In one of them the suspension damp-
Set A ing had uniform distribution, and in the other it had
normal distribution. (The response of the vehicle does
Numerical simulations are run to analyze the response not show significant differences for the two different types
of the stochastic full-vehicle model running on 2-D of uncertainty distribution.)
deterministic sinusoidal terrain profile. The accuracy of The effect of the tire stiffness uncertainty on vehicle per-
the polynomial chaos method and its efficiency compared formance is shown in Fig. 9.
with the Monte Carlo method have already been The effect caused by suspension stiffness uncertainty on
proved in [1,2], and further illustrated in Table 2. The vehicle performance, was analyzed for the suspension with
linear damping (not shown in the figure), as well as the one
with MR damping. The results for the MR damping case
Table 2
are illustrated in Fig. 10. The ride comfort in the MR dam-
Computational time for the seven DOF vehicle model simulations with
single uncertain parameters, on a 2-D deterministic sinusoidal terrain per case is better than when using a linear damper: the peak
profile value of the heave using the MR damper is less than in the
Uncertainty source Monte Collocation approach linear damper case, and the curve has relatively smaller
Carlo slopes.
Uniform Normal
500 runs
distribution distribution Table 2 summarizes the computation time of the simula-
(Legendre (Hermite tions performed for Set A. The efficiency of the polynomial
polynomials) polynomials) chaos approach is significantly higher than that of the
Linear susp. damper 1440s 9s 10s Monte Carlo approach for both types of uncertainty distri-
MR susp. damper 2522s 15s 16s bution in all cases.
Tire stiffness 1220s (linear 9s 8s
damper)
6.2. Numerical results using a stochastic vehicle model and a
Tire stiffness 2586s (MR 14s 14s
damper) stochastic terrain profile: Set B
Suspension stiffness 1374s (linear 5s 5s
damper) To look at the combined effect of the three parametric
Suspension stiffness 2590s (MR 11s 11s uncertainties (suspension damping, suspension stiffness,
damper)
and tire stiffness) under random external excitation, the

0.1 0.1 0.4

0 0.05 0.2
Heave [m]

Pitch

Roll

-0.1 0 0

-0.2 -0.05 -0.2

-0.3 -0.1 -0.4


0 1 2 3 0 1 2 3 0 1 2 3
Time [s] Time [s] Time [s]

Fig. 8. Chassis performance with MR damping uncertainty with uniform distribution (set A).
L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238 231

0.1 0.1 0.4

0 0.05 0.2

Heave [m]

Pitch

Roll
-0.1 0 0

-0.2 -0.05 -0.2

-0.3 -0.1 -0.4


0 1 2 3 0 1 2 3 0 1 2 3
Time [s] Time [s] Time [s]

Fig. 9. Chassis performance with tire stiffness uncertainty (set A).

0.1 0.1 0.4

0 0.05 0.2
Heave [m]

Pitch

Roll
-0.1 0 0

-0.2 -0.05 -0.2

-0.3 -0.1 -0.4


0 1 2 3 0 1 2 3 0 1 2 3
Time [s] Time [s] Time [s]

Fig. 10. Chassis performance with uncertainty in suspension stiffness (set A).

Rear-Right bounce [m]


0.1 0.1 0.4 0.2

0 0.05 0.2 0.1


Heave [m]

Pitch

Roll

-0.1 0 0 0
MC
-0.2 PolyChaos -0.05 -0.2 -0.1
MC Std
Poly Std -0.1
0 -0.4 -0.2
-0.3 1 2 3 0 1 2 3 0 1 2 3 0 1 2 3
Time [s] Time [s] Time [s] Time [s]

Fig. 11. Vehicle performance under combined parametric and external uncertainties (set B).

Heave Pitch Roll Rear-Right Bounce


40 40 10 50
MC

PC

20 20 5

0 0 0 0
-0.5 0 0.5 -0.5 0 0.5 -0.5 0 0.5 -0.5 0 0.5

Fig. 12. Probability density function of the vehicle performance under combined parametric and external uncertainties (set B).

Table 3
full-vehicle model (with an MR damper) was run on a 2-D Computational time for the seven DOF vehicle model simulations with
stochastic terrain profile. Fig. 11 shows the simulation combined uncertainties on a 2-D stochastic terrain profile
results for the vehicle chassis (sprung mass) and for the Uncertainty Monte Collocation
rear-right suspension assembly (unsprung mass). source Carlo approach
The probability density function of the vehicle perfor- 500 runs
Uniform Normal
mance is illustrated in Fig. 12. The correlation between distribution distribution
the Monte Carlo results and the polynomial chaos results (Legendre (Hermite
is very good. The efficiency of the polynomial chaos polynomials) polynomials)
method can be seen from the computation times given in Combined 1320s (linear damper) 13s 8s
Table 3. Combined 2442s (MR damper) 71s —
232 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

6.3. Terramechanics of the deterministic vehicle model on a Results from these simulations, including the sinkage,
2-D deterministic terrain profile: Set C wheel load, tractive force, resistance force, drawbar pull,
and torque on wheels are shown in Figs. 13–17. Figs.
6.3.1. Numerical results with uncertain average ground 13a,b, 14a,b, 15a,b, 16a,b and 17a,b correspond to the
pressure vehicle with heavy cargo running over a deterministic 2-
To study the effects of uncertain average ground pres- D sinusoidal terrain model at 10 m/s and 20 m/s. Figs.
sure, Michigan Sandy Loam is selected for terrain com- 13c,d, 14c,d, 15c,d, 16c,d and 17c,d show the results for
posure. The influence of velocity and of cargo weight light cargo.
on the tractive capacity of the off-road vehicle is studied. As shown in Fig. 13a, when the vehicle is carrying heavy
An off-road vehicle is simulated moving at 10 m/s and cargo, the mean value of the sinkage remains constant
20 m/s over a 2-D sinusoidal terrain, carrying a heavy almost all the time, except for a short interval around
cargo (2000 kg), and a light cargo (0 kg). The average 3.75 s. in the simulation. The interpretation of this result
ground pressure is chosen to vary between 150 kPa and is as follows: the mean value of the sinkage remains con-
250 kPa (25% uncertainty with mean pressure of stant as long as the wheel is modeled as flexible. For a flex-
200 kPa), as influenced by the tire inflation pressure ible wheel, the average ground pressure is the dominant
and by the pressure from the tire carcass. The average influence in the interaction of wheels and soil. Around
ground pressure range makes it possible for the algorithm 3.75 s, as seen in Fig. 13a, the critical pressure influences
to model the wheel as rigid or flexible depending on the the tire–soil dynamics and the wheel is modeled as rigid,
situation. so the standard deviation decreases to zero.
The dynamic response of the vehicle and the normal The algorithm selected to model the wheel only as flex-
load on the wheels oscillate with changes in the vehicle’s ible when the vehicle is running with 20 m/s. At higher
velocity and fluctuations of the terrain profile, and influ- speeds, the wheel load increases due to the increased heave
ence the critical pressure on the contact patch between and pitch of the sprung mass, as it has been shown in [6].
the tire and the soil. The critical pressure is also affected These higher wheel loads determine the critical pressure
by the cargo weight of the vehicle. If the critical pressure and determine the algorithm to select the flexible wheel
is greater than the average ground pressure, the wheel model.
should be modeled as flexible or deformable; if the critical For the heavy cargo case, the oscillating frequency of
pressure is less than the average ground pressure, the wheel the forces and torques on the wheels increases as the vehicle
is modeled as rigid. The wheel slip is assumed to remain runs faster, and the amplitudes of the resistance and the
constant during the simulation process. drawbar pull increase as well.

a 0.2
b 0.2

0.15 0.15
Sinkage [m]

Sinkage [m]

0.1 0.1

0.05 Deterministic 0.05 Deterministic


PolyChaos PolyChaos
0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, heavy cargo 20 m/s, heavy cargo

c 0.2
d 0.2
Deterministic Deterministic
0.15 PolyChaos 0.15 PolyChaos
Sinkage [m]

Sinkage [m]

0.1 0.1

0.05 0.05

0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, light cargo 20 m/s, light cargo
Fig. 13. Vehicle sinkage at the front-left wheel, sinusoidal terrain (set C).
L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238 233

For the light cargo case, the relation between the wheel oscillate faster at higher vehicle velocities. Unlike the heavy
and soil is dominated by the critical ground pressure. Sim- cargo case, the mean value of the sinkage does not remain
ilar to the heavy cargo, the forces and torques on wheels constant for the light cargo, but rather oscillates, as it can

a 2500 Deterministic b 2500


PolyChaos

Resistance Force [N]


Resistance Force [N]
2000 2000

1500 1500

1000 1000

500 500
Deterministic
PolyChaos
0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, heavy cargo 20 m/s, heavy cargo

c 1500 d 2500
Deterministic

Resistance Force [N]


Deterministic
Resistance Force [N]

PolyChaos 2000 PolyChaos


1000
1500

1000
500
500

0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, light cargo 20 m/s, light cargo
Fig. 14. Resistance force at the front-left wheel (set C).

a 8000
b 8000
Tractive Force [N]

Tractive Force [N]

6000 6000

4000 4000

2000 2000 Deterministic


Deterministic
PolyChaos PolyChaos
0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, heavy cargo 20 m/s, heavy cargo

c 8000
d 8000
Deterministic Deterministic
Tractive Force [N]

Tractive Force [N]

6000 PolyChaos 6000 PolyChaos

4000 4000

2000 2000

0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, light cargo 20 m/s, light cargo
Fig. 15. Tractive forces at the front-left wheel (set C).
234 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

be seen in Fig. 13c,d. The reason for this oscillation is that the vehicle at 10 m/s. Occasionally, for example around
the critical ground pressure is less than the average ground 1.25 s, 2.25 s, 3.25 s, and 4.25 s, for the vehicle running at
pressure, so the wheel is modeled as rigid all the time for 20 m/s, the average ground pressure is greater than the crit-

a 8000 b 8000
Drawbar Pull [N]

Drawbar Pull [N]


6000 6000

4000 4000

2000 2000
Deterministic Deterministic
PolyChaos PolyChaos
0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, heavy cargo 20 m/s, heavy cargo

c 8000
d 8000
Deterministic Deterministic

Drawbar Pull [N]


Drawbar Pull [N]

6000 PolyChaos 6000 PolyChaos

4000 4000

2000 2000

0
1 1.5 2 2.5 3 3.5 4 0
1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, light cargo 20 m/s, light cargo
Fig. 16. Drawbar pull at the front-left wheel (set C).

a 5000 b 5000

4000 4000
Torque [Nm]

Torque [Nm]

3000 3000

2000 2000

Deterministic 1000 Deterministic


1000
PolyChaos PolyChaos
0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, heavy cargo 20 m/s, heavy cargo

c 2000 d 3000
Deterministic Deterministic

1500 PolyChaos PolyChaos


Torque [Nm]

Torque [Nm]

2000

1000

1000
500

0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
10 m/s, light cargo 20 m/s, light cargo
Fig. 17. Torque at the front-left wheel (set C).
L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238 235

ical pressure, and the wheel is modeled as flexible, so Eqn. 19a, 20a and 21a show the sinkage, resistance, tractive
(41) replaces pcr. force, drawbar pull, and the torque of the front-left wheel
The computation time for the vehicle carrying heavy when the vehicle is carrying heavy cargo and running over
cargo was 36 h for 500 Monte Carlo runs, compared with a deterministic 2-D sinusoidal terrain model at 20 m/s.
20 min for the polynomial chaos. Figs. 18b, 19b, 20b and 21b show the results for the light
cargo scenario.
6.3.2. Numerical results with stochastic soil model In contrast to the results of Section 6.3.1, the sinkage,
Results from the case study of the full-vehicle model the forces, and the torques are all influenced by
running on a 2-D deterministic terrain profile with uncer- uncertain soil parameters. Similar to the case of
tain soil moisture are shown in Figs. 18–21. Figs. 18a, uncertain ground pressure, the oscillation frequency

a 0.2 b 0.2
Deterministic
Deterministic
PolyChaos PolyChaos
0.15 0.15
Sinkage [m]

Sinkage [m]
0.1 0.1

0.05 0.05

0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
Heavy cargo Light cargo
Fig. 18. Vehicle sinkage at the front-left wheel, 20 m/s (set C).

a 2500 b 2500
Deterministic
Resistance Force [N]

Resistance Force [N]

2000 2000 PolyChaos

1500 1500

1000 1000

500 500
Deterministic
PolyChaos
0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
Heavy cargo Light cargo
Fig. 19. Resistance forces at the front-left wheel, 20 m/s (set C).

a 8000
b 8000 Deterministic
Tractive Force [N]
Tractive Force [N]

6000 6000 PolyChaos

4000 4000

2000 2000
Deterministic
PolyChaos
0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
Heavy cargo Light cargo
Fig. 20. Tractive forces at the front-left wheel, 20 m/s (set C).
236 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

a 8000
Deterministic b 8000
Deterministic
PolyChaos
PolyChaos

Drawbar Pull [N]


Drawbar Pull [N]
6000 6000

4000 4000

2000 2000

0 0
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
Heavy cargo Light cargo

Fig. 21. Drawbar pull at the front-left wheel, 20 m/s (set C).

increases with increasing vehicle velocity. The standard 8000


Michigan Sandy Loam Soil, 0-30% Moisture, FlatSurface, 20m/s

deviation for the heavy cargo case is larger than for


7000
the light cargo; moreover, it increases with the increase
in vehicle speed. 6000

All plots show the results for the front left wheel.

Tractive Force [N]


5000

4000
6.4. The relation between the drawbar pull coefficient and the
wheel slip, and the tractive force-vehicle sinkage relation: Set 3000

D 2000

1000
Using the proposed algorithm, the drawbar pull coeffi-
cients at different slips in the deterministic soil moisture 0
0.055 0.06 0.065 0.07 0.075 0.08 0.085
Sinkage [m]
case and in the stochastic soil moisture case are calculated.
In the deterministic soil moisture case the drawbar pull Fig. 23. Deterministic tractive force vs. sinkage, 20 m/s (set D).
coefficient curve reaches the peak value at wheel slip 0.3,
and then decreases slightly without losing drawbar pull
7. Conclusions
ability. It can be seen in Fig. 22 that the standard deviation
for the stochastic soil moisture scenario is considerably
This paper presents a comprehensive study for assessing
large.
the tractive capability of vehicles operating in off-road con-
The relation between the tractive force and the wheel
ditions. The underlying idea is to allow the prediction of
sinkage on flat terrain obtained with the proposed algo-
the vehicle performance and mobility in very realistic con-
rithm is shown in Fig. 23. In the deterministic soil case
ditions, while maintaining a good computational efficiency.
the tractive force increases nonlinearly with the increase
Thus, we incorporate uncertainties in essential vehicle com-
in sinkage. From Fig. 23 we can see that the standard devi-
ponents, terrain geometry, and soil parameters, while ana-
ation of the tractive force caused by the stochastic soil
lyzing the effect of various vehicle speeds, cargo weights,
moisture becomes larger with the increase in vehicle
and tire models on the vehicle performance.
sinkage.
To illustrate the proposed approach, a 7-DOF stochas-
tic vehicle model, a stochastic 2-D terrain profile model,
and a stochastic soil model are developed and employed.
1.6
Deterministic
Although in this paper only uncertainties in suspension
1.4 Poly.Chaos stiffness, suspension damping, tire stiffness, terrain geome-
1.2
try, and soil parameters are considered, the approach
described is general, and can be extended as needed to
Drawbar Pull Coefficient

1
accommodate other uncertain parameters or inputs, such
0.8 as slip ratio. The heave, pitch, and roll of the sprung mass,
0.6 and the bounce motions of the unsprung subsystems are
0.4
simulated on 2-D terrain models.
To illustrate the vehicle response under parametric and
0.2
external uncertainty, simulations are performed using
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Monte Carlo method and the polynomial chaos approach.
Slip
The two methods have similar accuracy, with the polyno-
Fig. 22. Drawbar pull coefficient vs. slip, 20 m/s (set D). mial chaos being much more computationally efficient. It
L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238 237

also has the advantage of being able to accommodate high annual symposium on ‘‘Advanced Vehicle Technology’’. Paper
nonlinearities and large uncertain magnitudes. number IMECE2004-60482, Anaheim, CA; November 14–19, 2004.
[4] Sandu C, Sandu A, Chan BJ, Ahmadian M. Treatment of constrained
An algorithm to predict the off-road vehicle tractive multibody dynamic systems with uncertainties. In: Proceedings of the
capacity is presented. It uses the comparison of the ground SAE congress 2005. Paper number 2005-01-0936, Detroit, MI; April
pressure and the calculated critical pressure to decide on- 11–14, 2005.
the-fly if the tire must be approximated as a rigid wheel, [5] Li L, Sandu C, Sandu A. Modeling and simulation of a full vehicle with
or if it should be modeled as a flexible wheel. The algorithm parametric and external uncertainties. In: Proceedings of the 2005
ASME international mechanical engineering congress and exposition,
calculates the vehicle sinkage, the resistance force, the trac- seventh VDC annual symposium on ‘‘Advanced Vehicle Technologies,
tive force, the drawbar pull, and the tractive torque. If the Session 4: Advances in vehicle systems modeling and simulation. Paper
sinkage of the vehicle is significantly larger than the deflec- number IMECE2005-82101, Orlando, FL; November 6–11, 2005.
tion of the tire, then a rigid wheel model is used. If the [6] Sandu C, Sandu A, Li L. Stochastic modeling of terrain profiles and
operating surface is hard and the tire deflection is substan- soil Parameters. SAE 2005 Transactions. J Commer Vehicles, 2005-
01-3559.
tial, a flexible wheel model is used. This approach saves [7] Wong JY. Theory of ground vehicles. 3rd ed. John Wiley & Sons,
computation time and assures realistic results throughout Inc.; 2001.
the simulation. It has also been shown that for the flexible [8] Muro T, O’Brien J. Terramechanics. Land locomotion mechanics.
tire model the mean value of the sinkage remains constant A.A. Balkema Publishers; 2004.
regardless of how the terrain profile and vehicle velocity [9] Bekker MG. Off-the-road locomotion. Ann Arbor: University of
Michigan Press; 1960.
change, while for the rigid tire model it varies according [10] Shibly H, Iagnemma K, Dubowsky S. An equivalent soil mechanics
to the terrain profile. The resistance force, tractive force, formulation for rigid wheels in deformable terrain, with application to
drawbar pull, and torque, however, oscillate with the ter- planetary exploration rovers. J Terramech 2005;42:1–13.
rain profile and depend on the vehicle velocity, regardless [11] Rosca R, Rakosi E, Manolache GH, Cojocariu P. Wheel traction
of the tire model used. prediction on agricultural soil. In: Proceedings of the 30th military
technical academy conference, Bucharest, Romania; 2003 November
The algorithm developed in this study is also used to cal- 6–7. p. 133–9.
culate the relationship between the drawbar pull coefficient [12] Plackett CW. A review of force prediction methods for off-road
and the wheel slip, and the relationship between the sink- wheels. J Agric Eng Res 1985;31(1):1–29.
age and the tractive force on specific terrain profiles. [13] Okello JA. A review of soil strength measurement techniques for
The techniques presented in this study have applications prediction of terrain vehicle performance. J Agric Eng Res
1991;50(2):129–55.
in mobility prediction of off-road vehicles when their per- [14] Okello JA. Prediction of the force distribution between the soil and
formance needs to be investigated in as realistic scenarios pneumatic wheel. J Agric Eng Res 1992;51(4):249–62.
as possible, and the computational efficiency is of concern. [15] Reina G, Ojeda L, Milella A, Borenstein J. Measurement for wheel
Future work will include the development of analytical slippage and sinkage detection in rough-terrain mobile robots. In:
and computational methods to investigate the interaction Proceedings of the 2005 ASME international mechanical engineering
congress and exposition, Paper number IMECE2005-79711, Orlando,
of the vehicle with the stochastic soil model under corner- FL; November 5–11, 2005.
ing and braking conditions. The current paper is intended [16] Iagnemma K, Dubowsky S. ‘‘Terrain Estimation for High-Speed
to build the theoretical foundation of the modeling aspects. Rough-Terrain Autonomous Vehicle Navigation’’. In: Proceedings of
The authors will consider including field data in future the SPIE conference on unmanned ground vehicle technology IV, 2002.
papers. [17] Larson AC, Voyles RM, Demir GK. Terrain classification through
weakly-structured vehicle/terrain interaction. In: Robotics and auto-
mation, Proceedings of ICRA ’04, 2004 IEEE international confer-
Acknowledgement ence, vol. 1; 2004. p. 218–24.
[18] Karlsen RE, Overholt JL. Run-time assessment of vehicle-terrain
This work has been partially supported by NASA Lang- interaction. In: 24th army science conference proceedings; November
ley under the Virginia Institute for Performance and Re- 29–December 2, 2004.
[19] Iagnemma K, Golda D, Spenko M, Dubowsky S. Experimental study
search award. of high-speed rough-terrain mobile robot models for reactive behav-
iors. In: Proceedings of the eighth international symposium on
References experimental robotics, ISER ’02, Sant’ Angelo d’Ischia, Italy; July
2002.
[1] Sandu A, Sandu C, Ahmadian M. Modeling multibody dynamic [20] Nunez P, Jones R, Shoop S. High-fidelity ground platform and
systems with uncertainties. Part I: theoretical and computational terrain mechanics modeling for military applications involving vehicle
aspects. Multibody system dynamics. Publisher: Springer Nether- dynamics and mobility analysis. In: 24th army science conference
lands; 2006 June 29. p. 1–23(23). ISSN: 1384-5640 (Paper) 1573-272X proceedings, Session C, CP-06; November 29–December 2, 2004.
(Online). doi:10.1007/s11044-006-9007-5. [21] Morrison Romano R, Reid A, Gorsich D. High-frequency terrain
[2] Sandu C, Sandu A, Ahmadian M. Modeling multibody dynamic content and surface interactions for off-road simulations. SAE
systems with uncertainties. Part II: numerical applications. Multibody commercial vehicle engineering congress & exhibition technical
system dynamics, vol. 15, No. 3. Publisher: Springer Netherlands; program, Paper 04TB-128.
2006. p. 241–62 (22). ISSN: 1384-5640 (Paper) 1573-272X (Online). [22] Yoon S. A study on terrain-surface modeling and searching
doi:10.1007/s11044-006-9008-4. algorithms for real-time simulation of off-road vehicles. Vehicle Syst
[3] Sandu C, Sandu A, Chan BJ, Ahmadian M. Treating uncertainties in Dyn 2003;39(5):353–63.
multibody dynamic systems using a polynomial chaos spectral [23] Ferris JB. Characterizing road profiles as Markov chains. Int J
decomposition. In: Proceedings of the ASME IMECE 2004, sixth Vehicle Des 2004;36(2/3):103–15.
238 L. Li, C. Sandu / Journal of Terramechanics 44 (2007) 221–238

[24] Ghanem RG, Spanos PD. Stochastic finite elements. Mineola (NY): [27] Isukapalli SS, Georgopoulos PG. Development and application of
Dover Publications Inc.; 2003. methods for assessing uncertainty in photochemical air quality
[25] Xiu D, Lucor D, Su C-H, Karniadakis GE. Stochastic modeling of problems. Interim Report, prepared for the USEPA National
flow–structure interactions using generalized polynomial chaos. J Exposure Research Laboratory, under Cooperative Agreement CR
Fluids Eng 2002;124:51–9. 823467; 1998.
[26] Isukapalli SS, Roy A, Georgopoulos PG. Stochastic response surface [28] Isukapalli SS, Georgopoulos PG. Computationally efficient methods
methods (SRSMs) for uncertainty propogation: application to for uncertainty analysis of environmental models. In: Proceedings of
environmental and biological systems; February 1998. Available the A&WMA specialty conference on Computing in Environmental
from: http://www.ccl.rutgers.edu/~ssi/ srsmreport/srsm.html. Resource Management, A&WMA VIP-68; 1997. p. 656–65.

Vous aimerez peut-être aussi