Research Article

Advances in Mechanical Engineering

2018, Vol. 10(7) 1–13
Ó The Author(s) 2018
Rollover threshold investigation of a DOI: 10.1177/1687814018789505
journals.sagepub.com/home/ade
heavy-duty vehicle during cornering
based on multi-body dynamics

Xiaotong Dong1 , Yi Jiang1, Zhou Zhong2 and Wei Zeng2

Abstract
The objective of this article is to investigate the rollover threshold of a heavy-duty vehicle during cornering. Based on
the multi-body dynamics theory, a high-fidelity model is established, which takes account of the chassis flexibility, the sus-
pension nonlinear characteristics, the tire handling model, and Ackermann steering strategy. Furthermore, by the inverse
fast Fourier transform method, a three-dimensional stochastic road in space domain is employed in the model to
improve the accuracy. A full-size heavy-duty vehicle test was carried out to validate the model. Based on the validated
model, the rollover stability and rollover threshold of the heavy-duty vehicle during cornering are investigated. Lateral
acceleration, yaw rate, roll angle, and vehicle torsional deflection in different cornering conditions are analyzed. The roll-
over threshold is summarized by the response surface methodology for the safe cornering purpose. The result shows
the practical meaning of improving transportation safety of heavy-duty vehicles and also provides useful insights for
developing the rollover warning system.

Keywords
Heavy-duty vehicle, rigid-flexible dynamic model, rollover threshold, cornering, multi-body dynamics, three-dimensional
stochastic road

Date received: 1 February 2018; accepted: 20 June 2018

Handling Editor: Crinela Pislaru

Introduction concluded that a rollover model needs to well represent

Heavy-duty vehicle is a type of heavy equipment trans-
porters designed to execute special projects. They are
more prone to roll over than civilian vehicles during
cornering. In recent years, rollover accidents of heavy-
duty vehicles have comprised a disproportionately large
number of highway fatalities.1–3 Studying the rollover
stability and investigating the rollover threshold of the
heavy-duty vehicles have a profound significance of
reducing rollover accidents and improving the trans-
portation safety.
Many researches have been dedicated to studying
the vehicle rollover stability, aimed to reveal the
mechanism of rollover. Shim and Ghike4 built a 14-
degree-of-freedom vehicle model that included the non-
linear effects due to vehicle geometry changes, and he
the substantial effects such as the suspension nonlinear-
ity, the tire characteristics, and the roll center deviation.
Hegazy et al.5 developed a realistic multi-degree-of-
freedom vehicle model by multi-body dynamics, partic-
ularly suitable for analysis of transient cornering condi-
tions. Malviya and Mishra6 studied how various
conditions of maneuvering, crosswinds, and inclined
1
School of Aerospace Engineering, Beijing Institute of Technology, Beijing,
China
2
China Academy of Launch Vehicle Technology, Beijing, China
Corresponding author:
Yi Jiang, School of Aerospace Engineering, Beijing Institute of Technology,
5 South Zhongguancun Street, Haidian District, Beijing 100081, China.
Email: jy2818@163.com

Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(http://www.creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 Advances in Mechanical Engineering
Figure 1. Topological structure schematic of the whole model.
ground planes affect the vehicle rollover stability for
buses and trucks. Sellami et al.7 developed an empirical
model to substitute the vehicle dynamics and rollover
models and improve the computation effectiveness. He
et al.8 studied the influence of the road friction coeffi-
cient on the rollover stability of a six-wheel truck.
Ikhsan et al.9 investigated the correlation and effect of
the gross vehicle weight on heavy vehicle rollover stabi-
lity during cornering event by the TruckMaker
Software. Zhao et al.10 proposed the predictive vertical
tire force (PVTF) as the rollover index and established
an anti-rollover control system based on active steering.
Mashadi and Mostaghimi11 studied the four-wheel
vehicle dynamic response after the tires lift off, aiming
at presenting a general criterion for the vehicle rollover
threshold. Doecke et al.12 investigated the relationship
between speed limit and fatal crash rate and argued
that the safe speed limits will meet the objectives of a
safe system. Zhang et al.13 studied the contour line of
lateral transfer ratio (LTR) to predict the vehicle roll-
over index threat via the roll dynamics phase plane
analysis.
Most of the research, though, has focused on study-
ing the rollover stability of the civilian vehicles. The
rollover threshold of heavy-duty vehicles during cor-
nering is not clearly addressed yet. In addition, since
the structure of the heavy-duty vehicle is quite different
from other vehicles, a high-fidelity nonlinear model is
indispensable for studying the vehicle rollover.14
Therefore, further investigation is required.
The objective of this article is to establish a high-
fidelity rollover model and to investigate the rollover
threshold of the heavy-duty vehicles during cornering.
First, the rigid-flexible model of the heavy-duty vehicle
is established based on the multi-body dynamics the-
ory.15 A virtual three-dimensional stochastic road is
also employed to simulate the external excitation of the
ground. Second, a full-size vehicle test was carried out
to validate the model, which laid a solid foundation for
the results of the multi-body dynamics simulation.
Then, cornering conditions of different velocities and
different steering wheel angles are analyzed to study
the vehicle dynamic responses. Finally, by means of the
response surface methodology (RSM), the rollover
threshold of the heavy-duty vehicles during cornering is
concluded.
Multi-body dynamic model
Topological structure of model
The whole dynamic model is mainly composed of the
flexible chassis, the payload, the cab, the suspension,
the tires, and the three-dimensional stochastic road.
The topological structure is illustrated in Figure 1. The
payload and the cab are fixed to the flexible chassis.
The suspension is a double A-arm suspension and there
are nonlinear spring forces between the upper control
arms and the chassis to simulate the suspension forces.
The rotation motions are applied on the kingpins
Dong et al. 3

Figure 2. The first mode and the second mode of the vehicle chassis.

Table 1. The modal frequencies of the first 12 modes.

Mode Frequency Mode Frequency Mode Frequency

7 5.28 11 18.87 15 25.59

8 8.37 12 20.55 16 26.13
9 13.39 13 21.16 17 29.83
10 17.99 14 22.19 18 32.13

directly to simulate the steering actions. And the driv-

ing torque moments are exerted on the tires directly.
The modeling of each subsystem will be discussed as
follows. Figure 3. The double A-arm suspension model.

equilibrium air column is Vp =Ap , then, the elastic force

Modeling of chassis and the damping force of one hydro-pneumatic spring
Considering the structural characteristic influence of are given as19,20
the chassis, the chassis is modeled to be flexible. The  
modified Craig–Bampton method16 is implemented to z r
Fk = Pp Ap 1 + ð1Þ
remove the six rigid modes in order to meet the require- zp
ments of multi-body system dynamics calculation. In  2
addition, the reinforcing beams are set as the beam ele- rA3a z_ 2 1
Fc = sgn(_z) Cd Ad + Cc Ac ð1  sgn(_z)Þ ð2Þ
ments with cross-sectional attributes. The flexible part 2 2
comprised 34 attachment points, 43,763 elements, and
62,603 nodes. According to literature,17 modes with the To balance the stress distribution of the chassis,
frequency less than 50 Hz are adopted to ensure the every four hydro-pneumatic springs on the same side
accuracy of the calculation. Table 1 lists the modal fre- share one air chamber. It means the elastic forces of the
quencies of the first 12 modes, which begins from mode four hydro-pneumatic springs are equal. Hence, the
7. And Figure 2 shows the first mode and the second elastic force of one hydro-pneumatic spring is rewritten
mode of the chassis. The result of modal frequency is as
verified by comparing with a similar chassis in  
literature.18 1 ðz1 + z2 + z3 + z4 Þ r
Fk1 =Fk2 =Fk3 =Fk4 = Pp Ap 1 +
4 zp
ð3Þ
Modeling of suspension
The double A-arm suspension consists of upper control where z is the displacement of the hydro-pneumatic
arms, lower control arms, kingpins, and hydro- spring deformation; z_ is the velocity of the hydro-
pneumatic springs, as shown in Figure 3. As the hydro- pneumatic spring deformation; Vp is the equilibrium air
pneumatic springs have strong nonlinear mechanical volume; Pp is the equilibrium air pressure; r is the gas
characteristics, the suspension forces are presented as polytropic exponent, usually ranging from 1.4 to 1.6
nonlinear functions of displacement and velocity of the during the adiabatic process; r is the density of the oil
hydro-pneumatic spring deformation. According to the liquid; and Ap , Aa , Cc , Cd , Ac , and Ad are the dimen-
ideal gas equation, assuming the equivalent height zp of sional parameters of the hydro-pneumatic spring.
4 Advances in Mechanical Engineering
Figure 4. Ackermann steering principle schematic.
The steering strategy of the heavy-duty vehicle is
Ackermann steering,21 which is efficient and depend-
able. Ackermann steering tries to ensure that all tires
can perform pure rolling motion when the vehicle is
cornering, to relieve the bending stress of the chassis
and reduce the tire wear and friction force. As shown
in Figure 4, the vehicle is steering to the left around the
turning center O. According to the Ackermann steering
principle, there is
 11AGY027, based on the template of PAC2002-315/80
cot dil  cot dir = 2B=li
tan di li ð4Þ
tan d1 = l1
When the steering wheel angle dst is positive, the
vehicle is turning left. Thus, the left steering angle and
right steering angle of the first axle are given as
 sity (PSD) function24
d1l = cd  dst , d1r = arc cotðcot(d1l )  2B=l1 Þ when dst .0
d1r = cd  dst , d1l = arc cotðcot(d1r ) + 2B=l1 Þ when dst \0
ð5Þ
The steering angles of other steering axles are
   
li li
dir = arctan tan d1r , dil = arctan tan d1l
l1 l1
ð6Þ
The steering angles of none-steering axles are
dir = dil = 0 ð7Þ
where the coefficient cd = 0:033, B is half of the wheel-
base, li is the distance between axle i and the turning
center, and li is negative when axle i is behind the
Table 2. Fitting results for different road class.
Road class a r SSE
B 9.9e24 0.03201 1.609e210
C 9.9e24 0.0638 3.241e211
D 9.9e24 0.1276 5.149e210
E 9.9e24 0.239 1.389e28
turning center. All the steering angles are given as the
function of the rotation motions that are applied on
the kingpins for steering purpose.
Modeling of tire
There are many types of tire models since tires are com-
plicated mechanical systems. The type of tire model
should be determined by the simulation purpose. The
PAC2002 tire model22 is very suitable for cornering
maneuvers. It is the latest version based on the Magic
Formula with benefits of modeling tire–road interac-
tion forces accurately. In the radial direction, the
PAC2002 tire model considers the tire to behave as a
parallel linear spring and linear damper. And in longi-
tudinal and lateral direction, the tire forces strongly
depend on the slip between the tire patch elements and
the road surface. In addition, the equivalent volume
contact23 is applied between the tire and the road,
which can simulate the vehicle running on rough roads
with high efficiency. The tire model is established
according to the dimensional parameters and the tire
stiffness provided by the manufacturer of Qianjin
R22.5.
Modeling of road
The ISO 8608: 2016 recommended to adopt the follow-
ing equation as the road roughness power spectral den-
Gq (O) = RC  OW ð8Þ
where RC is the coefficient in respect of the class of
road roughness.
On the other side, the rational function form of
equation (8) can be expressed as
2ar2
Gq (O) =  , O.0 ð9Þ
p a 2 + O2
where a and r are the fitting parameters and O is the
spatial frequency. The values of a and r for different
road classes are obtained by the least square method.
Some fitting results along with the sum of squares due
to error (SSE) are given in Table 2.
Dong et al. 5

a
H0 fear g = ð11Þ
(s2 + a2 )3=2
where s is the corresponding variable in the transforma-
tion domain. Hence, the one-side two-dimensional PSD
function Gq (Ox , Oy ) is obtained as follows
 
Gq (Ox , Oy ) = 4F fRq (rx , ry )g = 4H0 Rq (r)
4r2 a ð12Þ
=
3=2
p
O2x + O2y + a2

By the method of the inverse fast Fourier transform

Figure 5. The partial view of road roughness. (IFFT),26 the three-dimensional road in space domain
is given as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Y 1 X
1 X X 1
XY  Gq (Ox , Oy ) j½um, n + 2pðimX + jnYÞ
z(x, y) = e
XY n = 0 m = 0 4Dlx Dly
ð13Þ
where um, n belongs to ½0 2p, following the average dis-
tribution; Dlx and Dly are the space frequency intervals
in x and y directions, respectively; and X and Y are the
numbers of the sampling points in x and y directions,
respectively.
As the regular roads in reality usually belong to B
class or C class, the road model created for simulation
in this article is C class. A three-dimensional, C class
road of 409.6 m 3 102.4 m in size is finally built up.
The partial view of the road roughness is shown in
Figure 4. In this road model, Dlx and Dly are both
0.1 m–1, X is 4096, and Y is 1024. Figure 5 demon-
strates that the road surface is isotropic, stochastic, and
Figure 6. Comparison of the power spectral density (PSD). continuous. Figure 6 shows that the PSD of a random
straight road path is well consistent with the ideal PSD
of C class.
According to the Wiener–Khintchine formula, the
autocorrelation function of equation (8) is
Modeling of full vehicle
r2
Rq (r) = ear ða, r.0Þ ð10Þ By Lagrange’s equation for constrained systems, the
p motion of each part in the system can be described in
Due to the randomness and the isotropy of road terms of the generalized coordinates q, as
roughness, PSD function Gq (Ox , Oy ) and its autocorre- "  #  
lation function Rq (Ox , Oy ) both can be considered as cir- d ∂T T ∂T T
 + FTq l = Q ð14Þ
cular symmetry functions. According to literature,25 the dt ∂q_ ∂q
Fourier transform of a two-dimensional circular sym-
metric function is equivalent to the zero-order Hankel The generalized coordinates q are given by
transformation and it also has q = ðx y z c u fÞT , where the rotational components
are the Euler angles in body 3-1-3 successive rotations.
The constraint equations of the system are given as

ui ðq1 , q2 , . . . , qn , tÞ = 0 i = 1, 2, . . . , r
ð15Þ
uj ðq1 , q2 , . . . , qn , q_ 1 , q_ 2 , . . . , q_ n , tÞ = 0 j = r + 1, r + 2, . . . , K

Multi-body dynamics equations (14) and (15) are

sparsely coupled differential algebraic equations
6 Advances in Mechanical Engineering
Figure 7. Multi-body dynamic model of the heavy-duty vehicle.
(DAEs). To solve the DAEs, the first job is to convert
the algebraic equations into ordinary differential equa-
tions (ODEs). This procedure can be done by differen-
tiating the algebraic equations three times, also called
I3 formulation. To solve the ODEs, the backward dif-
ferential formula (BDF) is applied.
The whole multi-body dynamic model is shown in
Figure 7. The main parameters of the heavy-duty vehi-
cle model are given in Table 3. The measurement coor-
dinate system is fixed at the centroid of the vehicle,
defined as the x-axis pointing to the vehicle front, the
z-axis pointing to the ground, and the y-axis following
the right-hand law. The basic assumptions are as
follows:
1. All constraints are assumed to be ideal (regard-
less of gaps or frictions);
2. Neglect minor influence factors such as wind
loads.
Model validation
Model simulation
The objective velocity and the steering wheel angle are
the two input variables in the simulation. The vehicle
accelerates at the start of the simulation on the three-
dimensional stochastic road and finally maintains a
steady velocity after 15 s. Proportional–integral–deriva-
tive (PID) control is implemented to control the velo-
city in the simulation. The input signal of the PID is the
difference between the measured velocity and the input
velocity (the unit is m/s). The output signal of the PID
is the value of the driving torque moment for one tire.
The parameters of the PID control are P = 2.0 3 106,
I = 2.0 3 104, and D = 0.0, respectively. When the
simulation time reaches 25 s, the steering wheel angle
starts to change as the given input and the vehicle starts
to corner. The vehicle maintains cornering until the end
of the simulation. The total simulation time is 45 s and
the time step is 0.02 s.
Table 3. Main parameters of the heavy-duty vehicle.
Parameter Value Parameter Value
Full-load weight 79 t Tire mass 250 kg
Payload weight 45 t Tire rolling radius 650 mm
Wheelbase 2.5 m Tire stiffness 700 kN/m
Total length 18.6 m Kingpin inclination 8º
Number of tires 16 Camber angle 0.5º
Figure 8. The test field with circularly arranged barricades.
Vehicle test
The aim of the vehicle test is to validate the multi-body
model. The test field is a suburban parking lot, which
has 100 m in length and 70 m in width, with a dry and
flat cement concrete pavement. As the space of test field
is limited, the vehicle is cornering at the same radius.
The circular path was marked by many barricades, as
shown in Figure 8.
The heavy-duty vehicle is fully loaded. Since the
vehicle centroid is inside the payload, it is difficult to
arrange the sensor to the vehicle centroid. Therefore,
the measurement point was located on the upper sur-
face of the payload, right above the centroid of the
vehicle. The GPS/INS device, XW-GI5651 with twin
antenna, was mounted on the measurement point.
The precision of XW-GI5651 is 0.1° for orientation
measurement and 0.01 m/s2 for acceleration measure-
ment. It transmits data to the laptop by RS-232 port.
The output frequency of the measurement devices is
configured to 50 Hz, which meets the experimental
requirements. A steering wheel angle measuring
instrument, HCZ-1A, was installed on the steering
wheel. Figure 9 exhibits the measurement devices
used in the experimental test. The measurement data
mainly include the steering wheel angle, driving velo-
city, lateral acceleration, yaw rate, and roll angle. A
tester was sitting in the cab and monitoring the data
during the test.
The five levels of driving velocities in the test are
shown in Table 4. The steering wheel angles are approx-
imately 630° in all tests.
Dong et al. 7

Figure 9. Measurement devices (HCA-1A and XW-GI5651).

Table 4. Different velocities for the test.

No. Velocity (m/s) Steering wheel angle (°)

1 4.26 637.7
2 4.76 631.4
3 5.10 638.8
4 6.42 628.6
5 7.08 633.2

Figure 11. Roll angle comparison between the simulation and

the test.

Figure 10. Lateral acceleration comparison between the

simulation and the test.

Validation
The root mean square (RMS) of the measured lateral
acceleration, yaw rate, and roll angle during the steady
cornering period are compared with the simulation
results. For the equal time interval data series, the RMS
is defined as
Figure 12. Yaw rate comparison between the simulation and
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N the test.
u1 X
RMS = t x2 ð16Þ
N i=1 i
satisfactory agreement is confirmed. The maximum
The results of the comparison between simulation errors of the lateral acceleration, the yaw rate, and the
and test are shown in Figures 10–12, and the roll angle are 10.1%, 10.5%, and –17.2%, respectively.
8 Advances in Mechanical Engineering
Figure 13. The steering wheel angle input.
Figure 14. Yaw rate versus time.
And the rest of the errors is less than 8%. The reasons
for the errors may be as follows:
1. It is difficult for the driver to keep the driving
velocity to be constant during cornering.
2. The mechanical properties of the tire are differ-
ent between low speed and high speed.
3. Other inevitable unavoidable influence factors
such as the undulation of the road surface.
In conclusion, the multi-body dynamics simulation
demonstrates the satisfying results by comparing with
the test. Hence, the dynamic model is proven to be suf-
ficient to analyze the roll threshold of the heavy-duty
vehicle.
Figure 15. Lateral acceleration versus time.
Cornering simulation and discussion
Cornering process analysis
As the long chassis results in a reduction in vehicle tor-
sional stiffness, there is a roll angle difference between
the cab and the payload during the steady cornering
period, which is denoted as the vehicle torsional deflec-
tion (VTD) in this article. Therefore, the VTD in vehi-
cle roll characteristics needs to be studied.27 Therefore,
four important parameters of the heavy-duty vehicle,
yaw rate, lateral acceleration, roll angle, and VTD will
be discussed in the section.
A case of cornering is analyzed to investigate
dynamic response of the cornering process. The velo-
city in the simulation is 7 m/s, and the steering wheel
angle input is shown in Figure 13. The result is pre-
sented in Figures 14–17, and it demonstrates that a
dynamic cornering of the heavy-duty vehicle can be
divided into four stages:
1. Stage 1: Straight driving period (0 \ t  25 s)
The steering wheel angle is zero and the vehicle is keep-
ing straight driving. Due to the stochastic excitation of
the road, the lateral acceleration, the yaw rate, and the
VTD oscillate around zero, while the roll angle slowly
undulates around zero.
2. Stage 2: Dynamic steering period
(25 s \ t  28 s)
The steering wheel angle starts to increase and the vehi-
cle starts to corner dynamically. The lateral accelera-
tion and the yaw rate decrease rapidly; the roll angle
increases rapidly; and the VTD increases at first and
then decreases.
Dong et al.
Figure 16. Roll angle versus time.
3. Stage 3: Transitional period (28 s \ t  30 s)
The steering wheel angle just reaches the steady value.
But as a result of the inertia effect, the lateral accelera-
tion, the roll angle, and the VTD have some overshoots.
It takes about 2 s to finally get them stabilized.
4. Stage 4: Steady cornering period (t . 30 s)
The steering wheel angle is constant and the vehicle is
cornering steadily. The lateral acceleration, the yaw
rate, the roll angle, and the VTD oscillate around the
steady values. In this case, the RMS of the yaw rate is
11.869°/s, the RMS of the acceleration is 1.019 m/s2,
the RMS of the roll angle is 2.074°, and the RMS of
the VTD is 0.209°.
Overall, the dynamic responses of the multi-body
dynamic model present fine results under the excitation
of the stochastic road. Further investigation can be
achieved with the help of the cornering simulations.
Figure 18. Comparison between the critical cornering and the rollover cornering.
9
Figure 17. VTD versus time.
Analysis of different conditions
When a vehicle is cornering at a high velocity, a roll-
over might occur. Most heavy-duty vehicle drivers fail
to perceive their proximity to rollovers as the time
span of rollover is short and abrupt.28 Figure 18 dis-
plays a comparison between a critical cornering (velo-
city of 9 m/s) and a rollover cornering (velocity of
10 m/s). The states of the two vehicles are quite simi-
lar to each other before 28.5 s, and then, the rollover
happens instantly. Due to the suddenness of the roll-
over, it is beneficial to be aware of the rollover thresh-
old in advance.
Simulations of different driving velocities and steer-
ing wheel angles were performed. Once the roll angle is
larger than 50°, it is considered as a rollover and the
simulation stops. The RMS of the lateral acceleration,
the yaw rate, the roll angle, and the VTD is shown in
Figures 19–22, respectively. The symbols and lines
which are out of the figure boundary indicate that roll-
overs occur, and there are no valid values in the steady
cornering period. The results are discussed as follows:
10
Figure 19. The RMS of lateral acceleration with different
velocities and steering wheel angles.
Figure 20. The RMS of yaw rate with different velocities and
steering wheel angles.
1. The lateral acceleration and the yaw rate
increase with the increase in the velocity and the
steering wheel angle. In addition, the lateral
acceleration has an increasing tendency with the
increase in velocity; the yaw rate increases with
the increase in the velocity and the steering
wheel angle proportionally.
2. The roll angle basically increases with the
increase in the velocity and the steering wheel
angle. However, when the steering wheel angle
is 180°, the roll angle at velocity of 9 m/s is
larger than the roll angle at velocity of 10 m/s;
Advances in Mechanical Engineering
Figure 21. The RMS of roll angle with different velocities and
steering wheel angles.
Figure 22. The RMS of VTD with different velocities and
steering wheel angles.
at a velocity of 5 m/s, the roll angle with steering
wheel angle of 270° is smaller than the roll angle
with steering wheel angle of 180°.
3. For the same steering wheel angle, the VTD
increases first and then decreases below zero
with the increase in the velocity. It also can be
observed that the VTD in the critical conditions
is always negative, which implies the payload is
‘‘taking the initiative’’ to roll the chassis. This
behavior of VTD in the steady cornering period
could indicate whether the vehicle is in the criti-
cal condition or not.
Dong et al. 11

Table 5. LTR of different driving velocities and steering wheel

angles.

Steering wheel Velocity (m/s)

angle (°)
5 7 9 10 11 13

180 0.06 0.14 0.24 0.27 0.34 0.48

270 0.08 0.21 0.34 0.42 0.52 0.73
360 0.14 0.27 0.45 0.57 0.70 1.00
450 0.17 0.32 0.56 0.72 1.00 1.00
540 0.20 0.39 0.69 1.00 1.00 1.00
630 0.23 0.46 0.80 1.00 1.00 1.00
720 0.26 0.51 1.00 1.00 1.00 1.00

Figure 23. Illustration of the response surface model and the

Rollover threshold investigation
The lateral transfer ratio (LTR)29–31 has been consid-
ered as the most commonly used index for evaluating
the vehicle rollover tendency. It is a dimensionless nor-
malized value ranging from –1 to 1. When LTR equals
–1 or 1, it means that one side of tires is totally lift off
from the ground. LTR is defined as
Fz, R  Fz, L
LTR =
Fz, R + Fz, L
sampling points.
Based on Table 4, 29 sampling points are applied to
construct the response surface model, excluding the
sampling points with LTR = 1. The response surface
model and sampling points are shown in Figure 23.
The SSE of the response surface model is 0.0266, which
meets the accuracy requirements of the RSM. The coef-
ð17Þ
ficient b of this RSM model is

where Fz,R is the sum of the right tire vertical forces and
Fz,L is the sum of the left tire vertical forces. Although
the tire forces are non-uniformly distributed and diffi-
cult to measure in reality, they could be easily reckoned
up from the simulation.
The results of LTR of different driving velocities
and steering wheel angles in steady cornering period
are shown in Table 5, where LTR = 1 indicates a roll-
over occurs. It can be seen that the rollovers are more
intended to occur at a velocity larger than 9 m/s or
when the steering wheel angle is bigger than 540°.
To investigate the critical cornering condition, the
relationship between LTR and the driving conditions is
constructed and analyzed by the RSM.32 The basic
form of an RSM model using a second-degree polyno-
mial function with two variables is given as
b0 = 0:2865, b1 = 0:102, b2 = 5:217e  4,
b3 = 6:778e  3, b4 = 3:72e  07, b5 = 2:377e 4
ð20Þ
Since all the LTR in the non-rollover conditions is
no bigger than 0.8 as shown in Table 5, the critical
value of LTR is set to be 0.8. The contour plot of the
response surface model is shown in Figure 24, including
the curve of the critical condition. It offers an intuitive
diagram for evaluating the rollover tendency of the
heavy-duty vehicle in respect of the velocity and the
steering wheel angle.
Based on the response surface model, the rollover
threshold can be fitted by the polynomial interpolating
function given as follows

^y(x) = b0 +b1 x1 +b2 x2 + b3 x21 + b4 x22 + b5 x1 x2 = xT b dst  x 4 V 4 + x 3 V 3 + x 2 V 2 + x1 V + x0 ð21Þ

ð18Þ
where x1 stands for the velocity, x2 stands for the
steering wheel angle, x = ½1, x1 , x2 , x21 , x22 , x1 x2 T , and
b = ½b0 , b1 , . . . , b5 T . With the N sampling points, the
coefficient b could be calculated by least square method
as shown in equation (19)
b = (XT X)1 XT y ð19Þ
where X = ½x1 , x2 , . . . , xN T and y = ½y1 , y2 , . . . , yN T .
where V is the velocity and x4 = 0.1071, x 3 = 26.08,
x 2 = 134, x 1 =21385, and x 0 = 5951.
The cornering conditions which comply with equa-
tion (21) are considered as the safe conditions. The
maximum of the steering wheel angle is 492 deg at a
velocity of 10 m/s, for example, which means that roll-
overs may occur with huge possibility when the steering
wheel angle is bigger than 492° during cornering.
Finally, since the maximum steering wheel angle of the
heavy-duty vehicle is 750°, the safe cornering velocity
12
Figure 24. Contour plot of the response surface model.
of the heavy-duty vehicle is 8.12 m/s according to equa-
tion (21).
Conclusion
This article has demonstrated the application of multi-
body dynamics in investigating the rollover threshold
of heavy-duty vehicles. A rigid-flexible multi-body
dynamic model subjected to a steering input function
with a constant velocity has been presented. The roll-
over threshold of the heavy-duty vehicle is investigated
by means of RSM with two variables: the velocity and
steering wheel angle. The conclusions are summarized
as follows.
1. Based on the multi-body dynamics theory, a
high-fidelity virtual prototype model is estab-
lished, which takes account of the chassis flexi-
bility, the suspension nonlinear characteristics,
the tire handling model, Ackermann steering,
and the three-dimensional stochastic road. The
main vehicle parameters are enclosed in this arti-
cle. The modeling process could be a useful ref-
erence for other studies.
2. The comparison between the vehicle tests and
simulation demonstrates that the multi-body
dynamics simulation can offer efficient and reli-
able numerical results for studying vehicle roll-
over purpose. Based on the simulation, the
vehicle dynamic responses in each stage of the
cornering process are discussed in detail.
3. In terms of the velocity and the steering wheel
angle, the lateral acceleration, the yaw rate, the
roll angle, and the VTD are analyzed. The lat-
eral acceleration and yaw rate both increase
with the increase in the velocity and the steering
Advances in Mechanical Engineering
wheel angle. Road roughness plays a dominant
role in the roll angle when the steering wheel
angle or the velocity is small. The sign of VTD
in steady cornering period corresponds to
whether it is in the critical cornering condition.
The results could contribute to the development
of the rollover warning systems.
4. Based on the RSM model of LTR, the rollover
threshold mathematical expression of the steer-
ing wheel angle and velocity is concluded. The
safe cornering velocity of the heavy-duty vehicle
is 8.12 m/s for all possible steering wheel angles.
Appling RSM to multi-body dynamics simula-
tion has been proven to be an effective solution
to investigate the rollover threshold, which also
shows great potential in studying other hazar-
dous driving maneuvers.
Acknowledgements
Xiaotong Dong contributed the most to this paper. Xiaotong
Dong and Yi Jiang designed and performed the experimental
test, Xiaotong Dong and Wei Zeng analyzed the data,
Xiaotong Dong and Zhou Zhong contributed the analysis
tools, Xiaotong Dong and Yi Jiang wrote the paper, and Yi
Jiang organized and managed the team.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of this
article.
Funding
The author(s) received no financial support for the research,
authorship, and/or publication of this article.
ORCID iD
Xiaotong Dong https://orcid.org/0000-0002-3675-7027
References
1. Jarossi L, Matteson A and Woodrooffe J. Trucks
involved in fatal accidents factbook 2007, http://www.
umtri.umich.edu/content/TIFA2007Factbook.pdf
2. Pollack KM, Yee N, Canham-Chervak M, et al. Narra-
tive text analysis to identify technologies to prevent
motor vehicle crashes: examples from military vehicles. J
Safety Res 2013; 44: 45–49.
3. Viano DC and Parenteau CS. Rollover crash sensing and
safety overview. SAE paper 2004-01-0342, 2004.
4. Shim T and Ghike C. Understanding the limitations of
different vehicle models for roll dynamics studies. Vehicle
Syst Dyn 2007; 45: 191–216.
5. Hegazy S, Rahnejat H and Hussain K. Multi-body
dynamics in full-vehicle handling analysis under transient
maneuver. Vehicle Syst Dyn 2000; 34: 1–24.
Dong et al.
6. Malviya V and Mishra R. Development of an analytical
multi-variable steady-state vehicle stability model for
heavy road vehicles. Appl Math Model 2014; 38:
4756–4777.
7. Sellami Y, Imine H, Boubezoul A, et al. Rollover risk
prediction of heavy vehicles by reliability index and
empirical modeling. Vehicle Syst Dyn 2018; 56: 385–405.
8. He Y, Chu D-F, Wu C-Z, et al. Quantitative analysis of
influence on vehicular lateral stability by road surface
condition. J Wuhan Univ Technol 2014; 38: 784–787 (in
Chinese).
9. Ikhsan N, Abdullah AS and Ramli R. The effect of gross
vehicle weight on the stability of heavy vehicle during
cornering. World Acad Sci Eng Technol Int J Mech
Aerosp Ind Mechatron Manuf Eng 2015; 9: 1251–1256.
10. Zhao W, Ji L and Wang C. HN control of anti-rollover
strategy based on predictive vertical tire force. T I Meas
Control. Epub ahead of print 18 September 2017. DOI:
10.1177/0142331217727581.
11. Mashadi B and Mostaghimi H. Vehicle lift-off modelling
and a new rollover detection criterion. Vehicle Syst Dyn
2017; 55: 704–724.
12. Doecke SD, Kloeden CN, Dutschke JK, et al. Safe speed
limits for a safe system: the relationship between speed
limit and fatal crash rate for different crash types. Traffic
Inj Prev 2018; 19: 404–408.
13. Zhang X, Yang Y, Guo K, et al. Contour line of load
transfer ratio for vehicle rollover prediction. Vehicle Syst
Dyn 2017; 55: 1748–1763.
14. Czechowicz MP and Mavros G. Analysis of vehicle roll-
over dynamics using a high-fidelity model. Vehicle Syst
Dyn 2014; 52: 608–636.
15. Hong Z-J. Computational dynamics of multibody systems.
Beijing, China: Higher Education Press, 1999 (in Chinese).
16. Yin X-G, Chen H and Jian K-L. Sub-structure method for
structural vibration analysis. Beijing, China: China Rail-
way Publishing House, 1991 (in Chinese).
17. Lu Y, Yang S, Li S, et al. Numerical and experimental
investigation on stochastic dynamic load of a heavy duty
vehicle. Appl Math Model 2010; 34: 2698–2710.
18. Wei X, Jiang Y, Zeng W, et al. The coupling effects of the
missile launcher and the ground in vehicle-mounted mis-
sile erecting. Adv Mech Eng 2016; 8: 1–12.
13
19. Tong J-L, Li W and Fu S-L. Effect analysis of main
parameters of hydro-pneumatic spring on suspension sys-
tem performances. J Syst Simul 2008; 20: 2271–2274 (in
Chinese).
20. Zhang P, Li Y and Li P. Dynamic characteristics analysis
of hydro-pneumatic suspension system. J Comput Theor
Nanos 2016; 13: 5794–5800.
21. Yu Z-S. The theory of automobile. Beijing, China: China
Machine Press, 2006 (in Chinese).
22. Pacejka H. Tire and vehicle dynamics. 3rd ed. Oxford:
Butterworth-Heinemann, 2012.
23. MSC Software. Using ADAMS/Tire. ADAMS software
manual. Newport Beach, CA: MSC Software, 2005.
24. ISO 8608:2016. Mechanical vibration—road surface
profiles—reporting of measured data.
25. Piessens R. Chapter 9: the Hankel transform. In: Poulari-
kas AD (ed.) Transforms and applications handbook. 2nd
ed. Boca Raton, FL: CRC Press, 2002, pp.453–457.
26. Liu X, Wang H, Shan Y, et al. Construction of road
roughness in left and right wheel paths based on PSD
and coherence function. Mech Syst Signal Pr 2015; 60–
61: 668–677.
27. Hasegawa S, Kusahara Y and Watanabe Y. Influence of
vehicle body torsional stiffness on vehicle roll characteris-
tics of medium duty trucks. Drive Syst Tech 2016; 30:
33–39.
28. Bernard J, Shannan J and Vanderploeg M. Vehicle roll-
over on smooth surface. SAE paper 891991, 1989.
29. Phanomchoeng G and Rajamani R. Real-time estimation
of rollover index for tripped rollovers with a novel
unknown input nonlinear observer. IEEE ASME Trans
Mechatron 2014; 19: 743–754.
30. Chen BC and Peng H. Differential-braking-based roll-
over prevention for sport utility vehicles with human-
in-the-loop evaluations. Vehicle Syst Dyn 2010; 36:
359–389.
31. Huang H-H, Yedavalli RK and Guenther DA. Active roll
control for rollover prevention of heavy articulated vehi-
cles with multiple-rollover-index minimisation. Vehicle
Syst Dyn 2012; 50: 471–493.
32. Bradley N. The response surface methodology. South
Bend, IN: Indiana University of South Bend, 2007.