Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Journal of Terramechanics 91 (2020) 297–307

Contents lists available at ScienceDirect

Journal of Terramechanics
journal homepage: www.elsevier.com/locate/jterra

Three-dimensional dynamic model for off-road vehicles using discrete


body dynamics q
Yaron Franco ⇑, Michal Shani, Gilad Gat, Itzhak Shmulevich
Research Center of Environmental, Water and Agricultural Engineering, Faculty of Civil and Environmental Engineering, Technion, 3200003, Israel

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a simple, reliable dynamics model of off-road vehicle operation in real-time (RT) on
Received 1 June 2020 terrain with obstacles. The numerical model was formulated by a new method – DBD (Discrete Body
Revised 2 August 2020 Dynamics). The new method is based on a discrete-element method, where the equations of motion
Accepted 12 August 2020
are linear and simple to solve.
Available online 9 September 2020
In this new method, the suspension systems are composed of soft and stiff springs and dampers (in-
stead of suspension arms and joints constrains), to present the kinematics and dynamics of real suspen-
Keywords:
sion. Reduction of the number of bodies and avoidance of constraints significantly improves model
Vehicle dynamics modeling
Off-road
efficiency and simplicity.
Prediction The tires–soil interaction was modeled using Brixius prediction. Specific soil properties were obtained
Multibody from the classification system for each tire–soil interaction, size, and geometric area. The tire–ground
DBD contact was determined by topographic surface and adjustment of the forces and direction acting on
Tire the tires.
The proposed method allows quick and simple definition of a vehicle. The model is written as an inde-
pendent software infrastructure, enabling easy integration with any other software component.
Simulation results were compared with Siemens’ VL commercial multibody dynamics program. The
performance of the proposed method was very similar to the commercial program (R2 > 0.9), with the sig-
nificant advantage of much higher RT performance.
Ó 2020 ISTVS. Published by Elsevier Ltd. All rights reserved.

1. Introduction mens, Adams by MSC, and others. These programs typically use
Lagrange formulation to solve the equations of motion. Consider-
There is strong demand for autonomous off-road vehicles, for ing that dynamic multibody simulations involve the formulation
military maneuvering and navigation and for modern farming in and numerical integration of complicated systems of nonlinear dif-
precision agriculture operations. Building a fully autonomous off- ferential equations, real-time prediction of behavior is a formidable
road vehicle that can reliably navigate and overcome obstacles is challenge for mechanical engineers, numerical analysts, and soft-
a major challenge. Real-time prediction of the vehicle dynamics ware developers. According to Rosenthal and Sherman (1986),
behavior in correspondence with different soil properties and ‘‘the calculation times are excessively high for even cases of aver-
obstacles is crucial for path planning and maneuvering. In addition, age complexity, and there are important formulation and imple-
the vehicle dynamics model is integrated with ground and envi- mentation problems that remain to be solved.” In order to
ronment data systems that can classify the terrain by soil mechan- achieve this objective, it is necessary to optimize the formulation
ics parameters and soil surface topography. A high-fidelity vehicle of the dynamic equations, the algorithms for numerical integra-
dynamics model is a fundamental element in the process of pre- tion, and the computer implementation (Garcia de Jalon et al.,
dicting an optimal driving path. 1991). A myriad of formulations and algorithms have been pro-
There are commercial programs for predicting vehicle dynamics posed by researchers in an effort to develop more efficient, yet gen-
based on multibody dynamics modeling, such as Virtual Lab by Sie- eral simulation algorithms for multibody systems. Evidence of this
may be seen in the ever-increasing calls for a significant reduction
of simulation time in the various applications of multibody algo-
q
Presented at the Proceedings of the 19th International & 14th European-African rithms (Anderson and Duan, 1999). Computational speed is there-
Regional Conference of the ISTVS, Budapest, September 25–27, 2017.
⇑ Corresponding author. fore the most important single consideration in code design (Garcia
E-mail address: agyaron@technion.ac.il (Y. Franco). de Jalon and Bayo, 1994; Jones, 1988; Schwertassek, 1993).

https://doi.org/10.1016/j.jterra.2020.08.003
0022-4898/Ó 2020 ISTVS. Published by Elsevier Ltd. All rights reserved.
298 Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307

Nomenclature

F force vector referred to a local coordinate system that is m body mass


attached to the body at its center of mass g gravity
Ixx; Iyy; Izz moment of inertia at the center of mass Dt time step
M moment vector referred to a local coordinate system t time
that is attached to the body at its center of mass tcr critical time step
H angular momentum vector referred to a local coordinate k spring stiffness
system that is attached to the body at its center of mass c damper constant
XCG; YCG; ZCG center of gravity
x center of gravity, position vector of body
x center of gravity, angular velocity vector of body

In addition, it is very challenging to connect those programs to are based on Bekker’s (1960) approach. The soil properties for Bek-
external systems in order to obtain soil mechanics parameters ker’s constants require heavy equipment for penetration of the soil
and soil surface topography in real-time. by plates.
In order to reduce the simulation time, one may use Newton- Many physics-based models are used in simulations to predict
Euler formulation. In this method, instead of adding an algebraic vehicle performances. Some of them, such as VTIM – Vehicle Ter-
equation for each degree constrained of freedom, we need to rain Interaction Model (Madsen et al., 2012), Nakashima and
reduce one equation. However, the equations are nonlinear. Oida (2004), Wakui and Terumichi (2011), Fervers (2004), and
Pazooki et al. (2012) developed a model of off-road vehicles with Xia (2010), have not been validated experimentally (Taheri et al.,
rear-axle torsion-elastic suspension. The model results agreed very 2015a). Others, such as Smith and Peng (2013), Yong et al.
well with experiments in all respected considered. However, for- (1978), and Grujicic et al. (2009), do not have lateral forces or com-
mulation of this model as a generic model to present suspensions bined slip. Yet other codes, such as CRREL (Shoop, 2001), are based
that include revolute and spherical joints, such as double wish- on FEM, and are therefore unsuitable for real-time.
bone, trailing arm, MacPherson, and the like, is very complex. The tire properties, such as tire stiffness, require special exper-
A similar method to Newton-Euler formulation is the discrete- imental equipment. To simplify construction of the vehicle model,
element method (DEM). Designed to model granular materials it is possible to predict the tire stiffness by the statistical relation-
(Cundall and Strack, 1979; Shmulevich et al., 2007), this method ship between tire dimensions and inflation pressure and the tire
describes every particle or cluster of particles of a material and stiffness as suggested by Lines and Murphy (1991).
the forces exerted on each particle by the other particles in contact The envelope of the lateral and longitudinal forces may be
with it. Simple mechanical formulations describe the relationship assumed to be an ellipse (Janosi et al., 1981). Experiments have
among the particles. It is possible to show an analogy of the inter- shown that the exponential form of the relationship between lat-
actions between the soil particles and the interactions between the eral force and slip angle is reasonably accurate (Crolla and El-
vehicle’s components. The particles may be bodies of the vehicle, Razaz, 1987).
such as the chassis and wheels. The relationship among the parti- In summary, extensive research has been done in an effort to
cles may be springs and dampers, which connect the vehicle’s predict soil–machine interaction for field vehicles, but not to sup-
components. port real-time (RT) vehicle dynamics off road.
To calculate the forces between the soil and the wheel, it is This paper presents a simple, reliable dynamic model for off-
important to know the tire deformation and the footprint. In gen- road vehicle operation in RT on terrain with obstacles. The pro-
eral, tire models can be grouped into three main categories: (1) posed model is named DBD and based on principles of the
empirical models, (2) semi-empirical models, and (3) physics- discrete-element method, where the equations of motion are linear
based models (Taheri et al., 2015a). The most validated and widely and simple to solve rapidly. The suggested model was verified with
used tire model is the empirical model developed by Brixius, which a commercial multibody dynamic program based on qualitative
is accepted as the standard (ASABE D497.4, 1998). It is a two- values and time computation. The comparison was demonstrated
dimensional model, in which the soil is characterized by a single on an ATV TM27 manufactured by Tomcar.
parameter – cone index (CI), according to ASABE EP542 (2013).
Many semi-empirical models are used for simulations to predict
vehicle performances. Some of them, such as STINA, VDANL, SCM, 2. DEM formulation
and FTIRE, have not been validated experimentally (Taheri et al.,
2015a). Some of them, such as NWVPM and AS2TM, are 2D models. Since DEM formulation is the basic approach of the new pre-
According to the HSSTM approach (Taheri et al., 2015b), the inter- sented DBD method, a short explanation of DEM is presented.
action of the tire with the ground is created by a set of massless The discrete element method (DEM), developed by Cundall and
bristles, which are defined as tread-contact elements attached to Strack (1979), is designed to model granular materials. It describes
adjacent belt elements that are distributed all around the tire. every particle or particle cluster of a material and the forces acting
These bristles can represent different deflection and friction char- on each particle by the other particles in contact with it. Simple
acteristics of the tire–ground interaction, using predefined or mechanical formulations describe the relationship among the par-
user-defined functions. The geometrical location of these tread ele- ticles by rheological elements (Fig. 1). The law of motion and inte-
ments can be defined in a way that represents the tread character- gration is similar to Itasca’s PFC program of DEM. The motion of a
istics of the actual tire. Depending on the type of simulation single rigid body is determined by the resultant force and moment
conducted, the number of mass elements should be allowed to vectors acting upon it and can be described in terms of the trans-
vary. Technically, the number of modes in the system is dictated lational motion of a point on the body and the rotational motion
by the number of mass elements. The number of masses also of the body. The translational motion of the center of mass is
depends on the resolution of the ground properties. These models described in terms of its position xi , velocity x_ i and acceleration
Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307 299

ðtþDt Þ ðtþDt Þ ðtþDt Þ


(7) is used to obtain xi . The values of F i and M i , to be
used in the next cycle, are obtained by application of the force-
displacement law.

3. DBD background

The DBD method consists of bodies, which are the particles in


DEM, and forces, which are the rheological elements in DEM. In
DBD, there are no constraints; therefore, the dynamics of each
body can be solved separately. Joints may be built using a set of
stiff springs and dampers, to achieve the kinematics relationship
between bodies. Therefore, this method is suitable for flexible
Fig. 1. Interaction between two particles in DEM model.
multibody systems (where the flexibility is provided by the springs
and dampers). Vehicles are flexible multibody systems, because
€ the joints have bushings and the chassis and suspension arms
xi ; the rotational motion of the body is described in terms of its
are somewhat flexible.
angular velocity x and angular accelerationx _ i.
The bodies in the model include vehicle body, wheels, and axle
The equations of motion can be expressed as two vector equa-
in the case of a solid axle suspension. The body element includes
tions, one of which relates the resultant force to the translational
kinematics information, such as accelerations, velocities, locations
motion, and the other of which relates the resultant moment to
and rotation matrices; dynamics information, such as weights,
the rotational motion. The equation for translational motion can
moments of inertia, and a list of forces and torques acting on the
be written in vector form:
body and the location of their connection to the body. The connec-
F i ¼ mð€xi  gi Þ ð1Þ tion points do not necessarily need to be rigidly attached to the
body; they can also be defined as a relative distance between the
where Fi is the resultant force, the sum of all externally applied
bodies. The body element updates its kinematics information
forces acting on the body; €xis an acceleration vector; m is the total
according to changes in the forces and torques acting on the body.
mass of the body and g is a gravity vector notation [0; 0; 9.81] m/
Force elements connect between two bodies. The forces are
s2.
spring, damper, contact, friction, and tire. The tires are unusual in
The equation for rotational motion can be written in the vector
this respect, since they connect the wheel and the ground, but
form:
the ground is not defined as a body. Force elements update the size
M i ¼ H_ i ð2Þ and direction of the force that passes through the bodies to which
they are connected. The calculation of the forces is based on the
where Mi is the resultant moment acting on the body and H_ i is the distance between the different connection points on the bodies.
angular momentum of the body. This relation is referred to a local The suspension arms, which have low mass and inertia relative
coordinate system that is attached to the body at its center of mass. to the sprung mass, are represented by a set of stiff springs and
The equations of motion are integrated using a centered finite- dampers. These stiff springs and dampers represent the flexibility
difference procedure that involves a time step ofDt. The quantities of the arms and bushings. The methods used to connect the stiff
x_ i and xi are computed at the mid-intervals of t  Dt=2, and the springs and dampers define the kinematic and dynamic suspension
quantities xi, €xi ,Fi and Mi are computed at the primary intervals of behavior. The suspension springs and shock absorbers define the
t  Dt. real spring and shock absorber constants.
The following expressions describe the translational and rota- As mentioned earlier, constraints are impossible. If constraints
tional accelerations at time t in terms of the velocity values at are needed, it is possible to build them using a set of stiff springs
mid-intervals. The accelerations are calculated as and dampers or to ignore the dynamics equation in the constraint
1  ðtþDt=2Þ ðtDt=2Þ
 direction. Fig. 2 shows some connection methods of the stiff
€xði tÞ ¼ x_ i  x_ i ð3Þ springs and dampers set in 2D to define free movement (3 DOF),
Dt
revolute at the body center and revolute outside of the body (2
1  ðtþDt=2Þ  DOF), revolute at the body center (1 DOF), and fix (0 DOF).
x_ ði tÞ ¼ xi ðtDt=2Þ
 xi ð4Þ A significant reduction in the number of bodies in the system
Dt
reduces the calculation time. Since the dynamics of each body
Inserting these expressions into Eqs. (1) and (2) and solving for
are solved separately, the calculation can be done in parallel com-
the velocities at time t  Dt=2 results in
putations. Updating forces is determined according to the equation
ðt Þ
!
Fi for a critical time step – tcr
ðtþDt=2Þ ðtDt=2Þ
x_ i ¼ x_ i þ þ g i Dt ð5Þ rffiffiffiffiffiffi
m mi
tcr ¼ min ð8Þ
! ki
ðt Þ
Mi
x_ ði tþDt=2Þ ¼ x_ ði tDt=2Þ þ þ g i Dt ð6Þ where
m

Finally, the velocities in Eqs. (5) and (6) are used to update the ki – the spring stiffness connected to body i;
position of the body center as mi – mass body i.

ðtþDt Þ ðt Þ ðtþDt=2Þ
xi ¼ xi þ x_ i Dt ð7Þ Time-step size is calculated by dividing tcr by a factor of 3 to be
on the safe side of the Nyquist principle. One can update each of
The calculation cycle for the law of motion can be summarized
ðtDt=2Þ
the system forces in a different time step. In our system, we have
, xi
ðtDt=2Þ ðt Þ ðt Þ ðt Þ
as follows. Given the values of x_ i , xi , F i and M i , three main force components: tire, spring, and construction (stiff
ðtþDt=2Þ
and xi
ðtþDt=2Þ
Eqs. (5) and (6) are used to obtain x_ i . Then, Eq. springs). Each of them has a different stiffness, as shown in Table 1.
300 Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307

Fig. 2. Connection methods of the stiff springs and dampers set in 2D to define free movement (3 DOF), revolute at the body center and revolute outside of the body (2 DOF),
revolute at the body center (1 DOF) and fix (0 DOF).

Table 1 The model of the wheeled vehicle consisted of 5 bodies: a chas-


Time computation complexity, by stiffness – k and critical time step – tcr. sis and 4 wheels. The front suspension was double wishbone, built
Force k tcr Computation complexity of 4 stiff springs and dampers, and the spring and shock absorber of
Tire Low High High the suspension. The rear suspension was a trailing arm, built of 2
Spring Low High Low stiff springs and dampers, and the spring and shock absorber of
Construction High Low Low the suspension. Powertrain and cruise control operated external
forces on the model. The steering control operated a constraint
on the model (therefore, the steering torque could not be calcu-
The critical time step for each component can be calculated using lated). Fig. 3 shows some of these parts on the Tomcar TM27
Eq. (8). The computations for the spring and construction are very side-by-side ATV wheeled vehicle. In this section, modeling of
simple, because they have two connecting points that are defined, some parts of the vehicle will be described.
whereas the connecting points of the tire with ground are numer-
ous and not fixed. This feature of allowing changes in the time step 4.1. The chassis
is important, because, as noted, updating the tire forces requires
large computing resources, and we want to perform this calcula- The chassis is a body with mass of 612.9 kg and inertia of:
tion with the maximum available time step. The different time Ixx = 187.5 kgm2, Iyy = 523.5 kgm2, and Izz = 636 kgm2 at the cen-
steps are synchronized. As a result of these unique features, the ter of mass. A reference coordinate system is located above the
program can work in real time and faster than the vehicle transi- center of the rear suspension at 0.381 m above the ground. The ref-
tion time, without the need for large computing means. erence coordinate’s orientation is same as the axis system of CG
(Fig. 3). The gravity center is located at 0.784 m front to the refer-
4. Wheeled vehicle model ence coordinate system and 0.094 m above the reference coordi-
nate system.
To clarify the new suggested method, we chose to demonstrate Table 2 shows the connection points of the stiff springs and
the method on a simple ATV vehicle. dampers to the body relative to the reference coordinate system.
Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307 301

Fig. 3. Tomcar TM27 side-by-side ATV model.

Table 2
Location of the connection points of the suspensions in the reference coordinate system.

Suspension Connection point X [m] Y [m] Z [m]


Front suspension Lower arm – front connection 2.1755 0.223 0.103
Lower arm – rear connection 1.8335 0.223 0.103
Upper arm – front connection 2.1755 0.209 0.076
Upper arm – rear connection 1.8335 0.209 0.076
Front spring and shock absorber connection 2.0320 0.211 0.240
Rear suspension Trailing arm connection 0.3910 0.547 0.071
Trailing arm connection 0.3910 0.547 0.071
Rear spring and shock absorber connection 0.3640 0.545 0.500

4.2. The front suspension tual stiff spring stiffness is 5  105N/m and the virtual damping
constant is 1000 Ns/m (it is assumed that this is the bushing stiff-
The front suspension is a double wishbone (Fig. 4). The joints ness and damping). The real suspension spring and shock absorber
between the suspension arms and the chassis have bushings. The are defined as linear (k = 122625 [N/m], c = 1285 [Ns/m]).
suspension arms have low mass and inertia relative to the sprung The force from the suspension spring and shock absorber are
mass. Therefore, to reduce the number of equations, virtual stiff distributed among the lower arm connector points according to
springs and virtual dampers are used, instead of bodies, to model the distance between the connection points.
the suspension arms. The virtual model stiff springs and virtual The steering constraint forces on the steering wheel are
dampers are shown in black in Fig. 4b. The real suspension spring neglected at this stage, because we are not interested in the torque
and shock absorber in the vehicle model are shown on Fig. 4b in on the steering wheel.
gray color. The magnitude of the two systems are different. The vir-

Fig. 4. Real double wishbone suspension (a); Virtual DBD double wishbone suspension (b).
302 Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307

4.3. The rear suspension Gear ratio: [3; 2.5; 2; 1.5; 1].)Originally, the vehicle had a CVT
transmission; in our simulation, we used an automatic transmis-
The rear suspension is a trailing arm. The suspension arm has sion where the gear selection is relatively to engine speed).
low mass and inertia relative to the sprung mass. Therefore, to Differential ratio: 6.429.
reduce the number of equations, the suspension arm was modeled
as virtual stiff springs and virtual dampers, instead of bodies.
4.7. The cruise control
The stiff spring stiffness and damping constant of the suspen-
sion arm, as well as the suspension springs and shock absorber,
The cruise control is of the pure pursuit type. The reference is
are assumed to have the same value as reported in Section 4.2.
the desirable ground velocity. The controller variable, the torque
of the engine, is determined by the error and the engine limits.
4.4. Wheels The limits of the engine are: maximum power and torque as func-
tions of engine RPM and maximum and minimum engine RPM.
In the new approach, the wheels are considered as bodies. The
wheel mass consists of: the wheel, tire and knuckle. The wheel
inertia consists of the shafts and gear inertia. Table 3 presents 4.8. The steering control
the connection points of the suspension to the knuckle, relative
to the wheel center and wheel properties (see Table 4). The steering control is of the pure pursuit type, where the con-
troller output is the steering angle instead of the steering torque.
There is a constraint between the steering angles of the front right
4.5. The tire model
and the front left wheels, according to Ackermann steering geom-
etry. The error is the distance between look ahead in the direction
The tire model transfers ground forces from the ground to the
of the vehicle and the desirable path.
wheels according to the topographic surface. Three circumferential
planes, as shown in Fig. 5, represent the 3D tire model. On each
plane, a number of points are sampled to test the penetration of 5. Description of the model
the tire into the ground. The resolution of the sampling points is
determined by the ground resolution data provided. The volume  A three-dimensional dynamic model, based on the discrete
confined between the tire and the ground is calculated by integra- body-dynamics method formulation for real time was devel-
tion of entry points, based on size and the geometrics of the con- oped and tested. The model inputs are: vehicle specification,
fined volume. The normal force direction and magnitude are Terrain and materials map classification, its specification mate-
calculated. Direction and magnitude of the shear forces acting on rial, and the desired path. The model provides predictions of the
the tire from the ground are calculated using Brixius prediction vehicle trafficability, kinematic and dynamic time-dependent
equations, as a function of the normal force. Lines and Murphy’s driving parameters, and more. A general description of the
(1991) model, which is function of rim diameter, tire width, and model is presented in Fig. 7.
inflation pressure, was used to predict the stiffness of the tire. 5.1. Model inputs

4.6. The powertrain  Vehicle specifications


The properties of each vehicle are stored in database
The model of the powertrain system consists of: containing:
Information on the system elements: geometric dimensions,
4.6.1. The Gasoline engine performance curve: torque versus weight, and moment of inertia;
engine speed, obtained as a second-order polynomial; The spring constants information, location in relation to body
4.6.2. The transmission ratios of the various gears; length from a free position;
4.6.3. High/low gear transfer ratio; The damping coefficient of the system components;
4.6.4. The transfer ratio of the differential; Tire data (dimensions and inflation pressure);
4.6.5. The type of propulsion (front drive, rear-wheel drive, or Data transmission (power train differential and engine data);
all-wheel drive); Safety parameters (maximum roll and pitch, angles, maximum
4.6.6. The differential mode (differential in the front, differen- sinkage, and maximum constraint of the vehicle for ground topog-
tial in the rear, or central differential); raphy inclination angles).
4.6.7. The clutch, defined with an exponential delay, to simu-
late the clutch coupling release.  Terrain and materials map classification

The powertrain properties used in this study are: The DBD dynamic vehicle model obtains application-specific
topography and mechanical properties relative to the spatial posi-
RPM to torque [Nm] polynomial coefficients. [0.000003; tion of the tires. The terrain classification can be represented as an
0.0193; 58.378], as presented in Fig. 6. image in which each pixel represents a material. Each material
Max RPM: 3500 rpm. type is connected to a database that stores its mechanical
Min RPM: 1500 rpm. properties.

Table 3
Connection points of the suspension to the knuckle and wheels properties.

Wheel Lower ball location [X,Y,Z] in [m] Upper ball location [X,Y,Z] in [m] Mass [kg] Inertia [kgm2]
Front [0,0.129,0.065] [0,0.167,0.119] 28.55 [1,60,1]
Rear [0,0.207,0.1] [0,0.207,0.1] 40 [1,60,1]
Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307 303

Table 4
The tire properties.

Tire Rim dia. [in] Width [mm] Torus height Tire damper [Ns/m] Tire type Inflation press. [psi]
Tire width

Front 12 203.2 0.8125 500 radial 25


Rear 12 304.8 0.5833 500 radial 25

Fig. 5. Tire model.

Fig. 6. Engine torque curve as a function of engine RPM.


304 Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307

Fig. 7. General description of the model.

The topography may be represented by a regular grid or a TIN 6. Research method


(triangulated irregular network).
The classified image and the topography can reach an accuracy 6.1. Model verification
of a few centimeters per pixel without detriment to the calculation
rate. In order to test the model, the results of simulation of an off-
road vehicle, Tomcar TM27, using the new developed method,
 Desired path were compared with those obtained using the commercial Sie-
mens’ VL multibody software, which is based on the Lagrange
One can specify a travelling path as a string of GPS coordinates. equations.
A steering control based on a proportional controller follows the
desired path. The model provides quantitative predictions of traffi- 6.2. Multibody model
cability values for the specific path. As a default, if no path is
defined, the vehicle will try to move in a straight line according Fig. 9 shows the Tomcar TM27 model that was developed in Sie-
to the initial location with the primary steering control. The speed mens’ VL. The points where the suspensions attach to the vehicle
control, also based on a proportional controller, tries to converge to chassis were modeled using bushing (revolute, spherical, etc.). All
the desired speed. When there is a trafficability problem, the other points of body-to-body connections were modeled using
model automatically reduces the speed of the vehicle. joints (revolute, spherical, etc.). Table 5 presents a list of the model
joints and constraints and the bodies that were connected by each
joint. Table 6 presents a list of the number of bodies and degrees of
5.2. Software freedom of the four subsystems that comprise the model.
The tires were defined as a simple-tire model built into the soft-
The general flowchart of the software developed for the model ware and suitable for road travel. The tire parameters (front: 25x8-
is presented in Fig. 8. It shows the relationship between the body 12; rear: 26x12-12) were obtained from the manufacturer and the
and the forces for a period of one step. The software has two main statistical models by Crolla et al. (1990) and by Lines and Murphy
advantages. The first is that it offers a generic solution. The vehicle (1991).
model is object-oriented, which makes it simple to incorporate
changes such as number of axles, type of suspensions, and more. 6.3. The DBD model
The second advantage is the ability to connect to external systems.
The software is independent, so it can be integrated with any con- The DBD model of Tomcar TM27 was built with the same
trol software. parameters as in the Siemens’ VL software model. The suspensions

Fig. 8. Program flowchart.


Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307 305

Fig. 9. Multibody dynamic model of Tomcar TM27 vehicle (right) and side view of the vehicle passing over the obstacle (left).

Table 5 of bodies, which are canceled, where added separately between the
List of joints and constraints in the multibody dynamics model. bodies that are connected to the canceled bodies. For the purpose
Body 1 Body 2 Joint/Constraint No. of of comparison, the tire model was adapted to fit the tire model in
type joints the multibody software. However, it should be noted that the tire
Lower control arm Chassis Revolute 2 configurations were not the same, and this might have caused
Upper control arm Chassis Revolute 2 some differences in the results obtained from the two different
Knuckle Lower control arm Spherical 2 programs.
Knuckle Upper control arm Spherical 2
Rack Chassis Screw 1
Tie rod Rack Universal 2 6.4. Simulation
Tie rod Knuckle Spherical 2
Front wheel Knuckle Revolute 2
Shock absorber Chassis Universal 4
Simulations were performed to examine the vehicle’s behavior
cylinder when traveling at a constant speed on a road with a single obstacle
Shock absorber Shock absorber Translational 4 that had a rigid radius of 200 mm. Fig. 9 shows the side view of the
piston cylinder vehicle when the front wheels passed over the obstacle.
Lower control arm Shock absorber Spherical 2
piston
Trailing arm Chassis Revolute 2 7. Results
Rear wheel Trailing arm Revolute 2
Trailing arm Shock absorber Spherical 2
piston The results presented here describe the travel of the vehicle at a
Main shaft Chassis Revolute 1 velocity of 5 m/s. In all the illustrations, DBD is compared with
Differential sun Chassis Revolute 1 simulation of multibody dynamics using the Siemens’ VL software.
Differential pinion Differential sun Revolute 1
The kinematics relative to the horizontal position of the vehicle are
Right differential Differential sun Revolute 1
pinion described in Fig. 10. The illustration on the right describes the posi-
Left differential Differential sun Revolute 1 tion of the center of gravity of the chassis on the Z axis, and the
pinion illustration on the left describes the pitch angle (negative value
Main shaft Differential sun Gear 1 describe pitch up according to the reference system as shown in
Differential pinion Right differential Gear 1
pinion
Fig. 3). The figure shows that there was a good match between
Differential pinion Left differential Gear 1 the two computation methods regarding the vehicle’s kinetics,
pinion with a maximum error of less than 4% between the different
Right differential Rear wheel Gear 1 simulations.
pinion
Fig. 11 presents a comparison between the forces vs. time calcu-
Left differential Rear wheel Gear 1
pinion lated using the two methods. The figure on the right depicts the
forces of the rear spring; the figure on the left depicts the emerging
power in one of the bushing front suspensions that connected the
Table 6
vehicle chassis. As seen, a good match was obtained regarding the
Details of subsystems of the Tomcar TM27 vehicle model. former, with a maximum error of less than 3% between the simu-
lations using different methods. In the left-hand figure, the results
Subsystem Name Degrees of Number of Remark
freedom bodies
were less well-matched: the trends were the same, but the rigid-
spring power in our system was restrained more quickly than
Chassis 6 1 Main body
Front suspension 28 14 Double wishbone
the bushing power in the multibody software. The DBD model is
Rear suspension 14 10 Trailing arm different in some aspects from the VL model, such as: tire model
Differential 1 5 Mode: Open and neglecting the mass of light weight bodies. Those differences
Total 49 30 may explain the less well-matched between the two curves.
Fig. 12 shows a comparison of FFT for the linear acceleration of
the Z axis of the vehicle chassis. It shows a good match between
the two methods, and the calculated frequencies had the same val-
were built with a set of stiff springs and dampers, as described in ues. It should be noted that a number of additional speeds that are
Sections 4.2 and 4.3. The Ackermann angles and the differential not presented in this article were also tested; all of them were
were built by an expression instead of bodies and joints. The mass well-matched.
306 Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307

Fig. 10. Comparison of vehicle Z axis vs. position, (right), pitch angle ‘vs. position (left).

Fig. 11. Comparison of forces operating in the system vs. time. On right – rear right spring; on left – bushing that connects the front suspension to the chassis.

A comparison of time computations by the two methods for a plane surface with random amplitudes between 0 and 100 mm.
four similar scenarios can be seen in Table 7. Scenario A is over- The simulation began with 2 sec for stabilization, followed by
coming a semi-round-shaped obstacle with a diameter of 200 mm. acceleration to 2 m/sec with limitation of the powertrain for 8[s].
The simulation began with 2 sec for stabilization, followed by Scenario D was like Scenario C, but with random amplitudes
acceleration to 5 m/sec with limitation of the powertrain. Scenario between 0 and 200 mm.
It can be concluded that the DBD method was faster in all four
B is overcoming 25 m of a sinusoidal path with an amplitude of
scenarios – A, B, C, and D – compared with Siemens’ VL calcula-
100 mm and a wavelength of 5 m. The simulation began with 2
tions, by factors of 1.9, 2.0, 4.7, and 8, respectively, for the specific
sec for stabilization, followed by acceleration to 2 m/sec with lim-
demonstrated vehicle. In addition, in the all scenarios, the DBD cal-
itation of the powertrain. Scenario C is driving in a straight line on culation time is shorter than real time (RT) by factors of 2.2, 2.5, 2.1
and 2.1 respectively to the scenarios. The tire circumference was
divided into 30 points in the DBD method. The DBD calculation
technique had no sensitivity to the rough surface in contrast to
Siemens’ VL program. For all scenarios, the DBD calculation time
was shorter than real time; with multibody dynamics, scenarios C
and D (which were obtained a realistic road) were far from real time.
It can be seen from Table 7 that for the presented vehicle, the
time for computation of the new code was half that for simulation
of a simple road, and 1/8 of the time for a realistic road obtained
from Siemens’ VL commercial program.

8. Summary and conclusions

A new efficient computation dynamic method was developed


for online dynamic prediction of off-road vehicles. The method is
based on DEM formulation. The proposed method was demon-
strated using an ATV TM27 manufactured by Tomcar Israel. The
method was verified using the Siemens’ VL commercial multibody
dynamics program.
The results indicate a good correlation between the results from
the newly DBD simulation codes and the computation by the
Fig. 12. FFT comparison of the system. commercial multibody dynamics program. As noted earlier
Y. Franco et al. / Journal of Terramechanics 91 (2020) 297–307 307

Table 7
Comparison of time results for the two computation methods, in four scenarios.

Comparison Road description Driving speed [m/s] Vehicle transition time [s] Siemens’ VL Calculation time [s] DBD Calculation time [s]
A Semi-round-shaped obstacle 5 11 9.6 5.0
B 5 peaks of a sinusoidal path 2 20 16.0 8.1
C Fine roughness road 2 10 22.0 4.7
D Coarse roughness road 2 10 38.0 4.7

(Section 6.3), the codes for the tire models were not the same; this García de Jalón, J., Avello, A., Cuadrado, J., 1991. An efficient computational method
for real time multibody dynamic simulation in fully cartesian coordinates.
could explain some of the different results indicated by the com-
Comput. Meth. Appl. Mech. Eng. 92, 377–395.
parison. The time computation of the new code was half for simple Garcia de Jalon, J., Bayo, E., 1994. Kinematic and Dynamic Simulation of Multibody
road or 1/8 of the time for a realistic road compared with that Systems: The Real-time Challenge. Springer-Verlag, New York.
obtained from VL, for the presented vehicle. The DBD vehicle Grujicic, M., Marvi, H., Arakere, G., Haque, I., 2009. A finite element analysis of
pneumatic-tire/sand interactions during off-road vehicle travel. Multi Model.
model is simple enough to be used in real-time prediction of path Mat. Struct. 6 (2), 284–308.
navigation for off-road applications. Janosi, Z.J., I.O. Kamm and G. Wray. 1981. Tire turning forces under on and off-road
Further research is highly recommended to verify the findings conditions. Proc. 7th Int. Conf., ISTVS, Calgary, late submission.
Jones, R.E., 1988. Multi-flex body dynamics for control design. In: Proceedings of the
experimentally and extend the method for predicting different sus- Workshop on Multibody Simulation, JPL D-5190, Pasadena, CA, pp. 543–549.
pensions and vehicles. Lines, J.A., Murphy, K., 1991. The stiffness of agricultural tractor tyres. J. Terramech.
28 (1), 49–64.
Madsen, J., Negrut, D., Reid, A., Seidl, A., Ayers, P., Bozdech, G., Freeman, J., O’Kins, J.,
Declaration of Competing Interest 2012. A physics-based vehicle/terrain interaction model for soft soil off-road
vehicle simulations. SAE Int. J. Commer. Veh. 5 (1), 280–290.
The authors declare that they have no known competing finan- Nakashima, H., Oida, A., 2004. Algorithm and implementation of soil–tire contact
analysis code based on dynamic FE–DE method. J. Terramech. 41 (2-3), 127–
cial interests or personal relationships that could have appeared 137.
to influence the work reported in this paper. Pazooki, A., Rakheja, S., Cao, D., 2012. Modeling and validation of off-road vehicle
ride dynamics. Mech. Syst. Sig. Process. 28, 679–695.
Rosenthal, D.E., Sherman, M.A., 1986. High performance multibody simulations via
Acknowledgements
symbolic equation manipulation and Kane’s method. J. Astronaut. Sci. 34 (3),
223–239.
The authors would like to thank the Research Center of Environ- Schwertassek, R., 1993. Reduction of multibody simulation time by appropriate
mental, Water and Agriculture Engineering of the Faculty of Civil formulation of dynamical system equations. In: Proceedings of the NATO
Advanced Study Institute on Computer-Aided Analysis of Rigid and Flexible
and Environmental Engineering at the Technion-Israel Institute of Mechanical Systems, Troia, Portugal, 9, pp. 447–482.
Technology for its support. This research was partially supported Shmulevich, I., Rubinstein, D., Asaf, Z., 2007. Chapter 4: Discrete element method for
by Elbit System Ltd. soil-machine interactions. In: Upadhuyaya, S.K. (Ed.), Soil Tillage. American
Society of Agricultural and Biological Engineers Press.
Shoop, S.A., 2001. Finite Element Modeling of Tire–terrain Interaction. US Army
References Corps of Engineers, Hanover, NH.
Smith, W., Peng, H., 2013. Modeling of wheel–soil interaction over rough terrain
Anderson, K.S., Duan, S., 1999. A hybrid parallelizable low-order for dynamics of using the discrete element method. J. Terramech. 50 (5–6), 277–287.
multi-rigid-body Part I, Chain systems. Math. Comput. Modell. 30, 193–215. Taheri, Sh., Sandu, C., Taheri, S., Pinto, E., Gorsich, D., 2015a. A technical survey on
ASABE D497.4, 1998. Agricultural machinery management data. Terramechanics models for tire–terrain interaction used in modeling and
ASABE EP542, 2013. Procedures for using and reporting data obtained with the soil simulation of wheeled vehicles. J. Terramech. 57, 1–22.
cone penetrometer. Taheri, Sh., Sandu, C., Taheri, S., 2015b. Hybrid soft soil tire model (HSSTM), Part I:
Bekker, M.G., 1960. Off-the-road Locomotion. University of Michigan Press, Ann Tire material and structure modeling. Report by the Department of Mechanical
Arbor. Engineering, Virginia Tech, Blacksburg, VA, USA and by U.S. Army TARDEC, MI,
Crolla, D.A., El-Razaz, A.S.A., 1987. A review of the combined lateral and longitudinal USA. PDF Url : ADA616952, 40p.
force generation of tyres on deformable surfaces. J. Terramech. 24 (3), 199–225. Wakui, F., Terumichi, Y., 2011. Numerical simulation of tire–ground system
Crolla, D.A., Horton, D.N.L., Stayner, R.M., 1990. Effect of tyre modelling on tractor considering soft ground characteristics. J. Syst. Des. Dyn. 5 (8), 486–500.
ride vibration predictions. J. Agric. Eng. Res. 47, 55–77. Xia, K., 2010. Finite element modeling of tire/terrain interaction: application to
Cundall, P.A., Strack, O.D.L., 1979. A discrete numerical model for granular predicting soil compaction and tire mobility. J. Terramech. 48 (2), 113–123.
assemblies. Géotechnique 29 (1), 47–65. Yong, R.N., Fattah, E.A., Boonsinsuk, P., 1978. Analysis and prediction of tyre-soil
Fervers, C.W., 2004. Improved FEM simulation model for tire–soil interaction. J. interaction and performance using finite elements. J. Terramech. 15 (1), 43–63.
Terramech. 41 (2), 87–100.

You might also like