You are on page 1of 12

Original Article

Proc IMechE Part K:


J Multi-body Dynamics
Managing driving disturbances in 2021, Vol. 235(1) 122–133
! IMechE 2020
lateral vehicle dynamics via adaptive Article reuse guidelines:
sagepub.com/journals-permissions
integrated chassis control DOI: 10.1177/1464419320977374
journals.sagepub.com/home/pik

Narjes Ahmadian1 , Alireza Khosravi1 and Pouria Sarhadi1,2

Abstract
This paper presents a vehicle stability control method based on a multi-input multi-output (MIMO) model reference
adaptive control (MRAC) strategy as an advanced driver assistance system (ADAS) to enhance the handling and yaw
stability of the vehicle lateral dynamics. The corrective yaw moment and additive steering angle are generated using
direct yaw moment control (DYC) and active front steering (AFS) at the upper control level in the hierarchical control
algorithm. A nonlinear term is added to the conventional adaptive control laws to handle parametric uncertainties and
disturbances. The desired yaw moment generated by the upper-level controller is converted to the brake forces and is
distributed to the rear wheels by an optimal procedure at the lower-level. The major contribution of this study is the
introduction of a nonlinear integrated adaptive control method based on a constraint optimization algorithm. To verify
the effectiveness of the proposed control strategy, the nonlinear integrated adaptive controller, and linear time-varying
MRAC are designed and used for comparison. Simulation results are performed for the J-turn and double lane change
(DLC) manoeuvres at high speeds and low tyre-road friction coefficients. The desired performance of the proposed
controller exhibited significant improvement compared to the conventional MRAC in terms of yaw rate tracking and
handling of sideslip limitation.

Keywords
Integrated control, advanced driver assistance system, model reference adaptive control, direct yaw moment control,
active front steering
Date received: 17 April 2020; accepted: 16 October 2020

Introduction can modify the driver’s steering by calculating the


Vehicle handling, stability, safety, and comfort are extra steering angle. This significantly improves the
important challenges in designing chassis control sys- lateral dynamics of the vehicle. However, in critical
tems. These systems include electronic stability control situations where the vehicle enters an unsafe operat-
(ESC),1 anti-lock brake system (ABS),2 active suspen- ing range, due to its nonlinear tyre characteristics and
sion,3 active front steering (AFS),4 traction control saturation, system performance may be disrupted
system (TCS),5 direct yaw moment control (DYC),6 under a large steering wheel. As a result, the DYC
and torque vectoring system,7 which are designed system is developed. The DYC can apply corrective
and manufactured by different suppliers and with var- yaw moment using different longitudinal forces in the
ious control technologies. Performance conflict among left and right directions of the car. It also contributes
different control subsystems such as braking, steering, to vehicle stability and steerability and plays a more
and suspension may occur because of the interaction important role in vehicle stability than ABS and TSC.
between longitudinal, lateral, and vertical dynamics. By applying proper brake torque, DYC adjusts the
To manage the problem and reduce the number of
sensors and actuators, integrated control is introduced. 1
Department of Electrical and Computer Engineering, Babol Noshirvani
Integrated control of vehicle dynamics is an advanced University of Technology, Mazandaran, Iran
2
method for optimizing the use of different vehicle sub- Faculty of Engineering and Physical Sciences, Department of
systems. This method was proposed around 1990 to Mechanical Engineering Sciences, University of Surrey, Guildford, UK
coordinate different chassis control systems in the pres- Corresponding author:
Alireza Khosravi, Department of Electrical and Computer Engineering,
ence of conflict and functional interference.8 Babol Noshirvani University of Technology, Shariati Av, Babol,
Active control systems are used for longitudinal Mazandaran, Iran.
and lateral dynamics control problems. The AFS Email: akhosravi@nit.ac.ir
Ahmadian et al. 123

movement of the vehicle based on information from adding a saturation function based on the tracking
the steering wheel angle, thus ensuring vehicle stabil- error to manage the parametric uncertainties and
ity. However, the external torque produced by DYC external disturbances. The output constraint adaptive
is an expensive control input and should be limited in control problem with constraints on the sideslip angle
emergencies. Overuse of the DYC results in a slower and the yaw rate is solved by defining a barrier func-
speed and consequently driver complaint. Also, the tion. In the second level, an optimization algorithm is
braking force applied by the DYC leads to tyre ero- used to minimize the longitudinal forces and create
sion.9 One effective way to limit the overuse of DYC coordination between the brake and the steering
is to integrate it with AFS. Under such circumstances, actuators. In this paper, to prevent interference with
DYC is activated to offset AFS restrictions. the AFS, it is assumed that the brake force is applied
Numerous papers have discussed the integration of to the rear wheels.4 The major contribution of this
DYC and AFS. He et al.10 and Chokor et al.11 paper is the introduction of an integrated MIMO
designed the AFS and DYC control systems separate- NLMRAC algorithm to coordinate between AFS
ly and incorporated a switching strategy between and DYC based on a constraint optimization
them. However, the internal stability of the system method. There are mainly two contributions; first, a
was questionable due to the switching process, and Lyapunov-based MIMO MRAC method is designed
the discontinuity of the control signals under consec- by adding a nonlinear term to manage parameter
utive switches was undesirable in practical problems. uncertainties such as mass, tyre-road friction coeffi-
The use of gain scheduling theory based on the sta- cient, longitudinal velocity, and angular stiffness and
bility margin of vehicle lateral dynamics has been eliminate external disturbances. For the first time, the
investigated in several articles.4,12,13 In these methods,
stability analysis for the nonlinear controller is per-
the stability domain of the vehicle is first determined
formed using MIMO equations and the tracking error
by drawing the phase plane strategy. The phase plane
convergence is proved without decentralization error
method can be determined using phase portrait draw-
vectors. Second, a two-level integrated strategy is pre-
ing of the sideslip angle and its rate,12,13 yaw rate and
sented to maintain vehicle handling and stability by
lateral velocity,14 or sideslip angle, and tyre-road fric-
reducing the lateral acceleration and sideslip angle.
tion coefficient.15 A stability index is then determined
One of the advantages of the extended method is
based on the permissible area obtained from these
portraits. This index specifies the operating range of that there is no need to define multiple tuning param-
the steering, brake, or both actuators. The use of gain eters in the integrated control process. On the other
scheduling methods to coordinate between steering hand, the application of braking force is reduced by
and brake actuators necessitates several tuning controlling the steering wheel. The simulation results
parameters. Some researchers ignore steering control for the proposed controller are compared to the con-
in lateral dynamics and focus only on DYC.16,17 In ventional MRAC method. Better tracking perfor-
these papers a sliding mode control strategy is designed mance and keeping lateral stability using lower
to generate the required yaw moment. Lack of control control efforts are demonstrated by optimizing the
on the steering wheel leads to using more braking use of brake and steering forces for two lateral
forces in a particular manoeuvre, which is not pleasant manoeuvres.
for the driver. While the integration of active steering The original contribution of this study is the exten-
and braking reduces the requirements of the braking sion of the centralized MRAC method. The extended
force. The use of optimal control for the integration of integrated control strategy has the ability to cope with
AFS and DYC has also been suggested in several different vehicle uncertainties and lateral disturbances
articles.9,18 Where for the problem of path following simultaneously.
and integration between actuators many indexes and This article is organized as follows: Section 2
tuning parameters should be defined. Guo et al.19 explains the lateral vehicle dynamics model. In
designed a controller with input constraints on brake Section 3, the constraint integrated NLMRAC struc-
and steering actuators. They calculated the permissible ture is developed. The simulation results are shown in
range of input changes in lateral dynamics based on Section 4. The conclusions and discussion are pre-
the permissible range for the yaw rate and sideslip sented in Section 5.
angle outputs. On the other hand, Hu et al.20 solved
the problem with output constraints on yaw rate and
System description
sideslip angle outputs by providing a method for path
following problems. This section explains the vehicle lateral dynamics
This paper presents a nonlinear MRAC model and the parameters used therein. Since the
(NLMRAC) method to coordinate AFS and DYC purpose of this study is to control the yaw and lat-
using an optimization algorithm to the optimal distri- eral movements, so the roll, jounce, and rebound
bution of braking forces on rear wheels. A two-level dynamics are ignored. Figure 1 shows the single
control structure is presented. In the first level, a track (ST) 2 degree-of-freedom (DOF) vehicle
model reference adaptive controller is designed by model used for yaw and lateral motion study.
124 Proc IMechE Part K: J Multi-body Dynamics 235(1)

0 1
! L2f Cf þ L2r Cr Lr Cr  Lf Cf
€ B l l C _ !
w B Iz vx Iz C w
¼B
B
C
b_ @ 1 þ l r Cr  Lf Cf
L Cf þ Cr C A b
l
mv2x mvx
0 1
Lf C f 0 1
B I C 1
B z C B C
þB Cd þ @ Iz AMz
@ Cf A
0
mvx
0 1 0 1
1 0
B C
þ @ Iz AMdz þ @ 1 AFdy
Figure 1. ST model of lateral vehicle dynamics. 0 mvx
(4)
The nonlinear yaw and lateral motion equations are
expressed as follows13 where l is the coefficient of friction that depends on
the actual value of the wheel skid and several other
€ ¼ Lf ðFx;f sinðdÞ þ Fy;f cosðdÞÞ
Iz w factors including weather conditions, the material
 Lr Fy;r  DFx;r tr þ Mdz (1) used in the asphalt and wheels, the condition of the
contact point between the wheel and the track (e.g.
mvb_ ¼ Fy;f þ Fy;r þ mvw_ the presence of oil, grease, mud) and slope variations
of the road.
Since the proposed design is achieved in the linear The steering input d ¼ df is obtained from the sum
control framework, the linear ST model, obtained of the two driver command and the AFS controller
from the linearization of (1) for nominal velocity output: d ¼ dd þ d and Mz is the corrective yaw
and assuming, |b | <7 deg, k < 0:1 ( k ¼ Rxvv , where moment control input. The bicycle model is used as
k is the longitudinal slip ratios and x is the angular the reference model with the following assumption,
 
wheel velocity), and cosðdÞ ’ 1. Mz ¼ 0, l ¼ 1, b ¼ tan1 vxy , Mdz ¼ 0, and Fdy ¼ 0.23
v
The vehicle yaw and lateral movement equations
are expressed as follows Note that this model is often used to simplify for
the assessment of different control strategies. And
€ ¼ Lf Fy;f  Lr Fy;r þ M
Iz w that vehicles in lateral motion and cornering are sub-
z
(2) ject to much more complex motions, requiring multi-
mvx ðb_  wÞ
_ ¼ Fy;f þ Fy;r
DOF models.24,25
The index i 2 ff; rg, j 2 fl; rg are used to determine
where w_ is the yaw rate, b is the sideslip angle, vx is the front, rear, left and right positions, respectively.
the longitudinal velocity and is considered to be con-
stant. The Dugoff’s tyre model combined with longi-
tudinal and lateral slips using the above assumptions Hierarchical controller design
of the linearization process is used for the lateral vehi- The structure of the proposed control scheme is
cle dynamics model.4,21,22 Fy;f , Fy;r are the front and shown in Figure 2. The design consists of two upper
rear lateral tyre forces, respectively, and are obtained and lower levels. To facilitate the process of controller
from the following equations. design, it is important to describe the control objec-
tives and design features. The upper-level controller is
w_ a nonlinear MIMO MRAC structure. The goal is to
Fy;f ¼ Cf af ; af ¼ df  b  Lf
vx (3) track the reference yaw rate while the sideslip angle
w_ does not exceed its maximum value. For this purpose,
Fy;r ¼ Cr ar ; ar ¼ dr  b þ Lr
vx two control inputs of ADAS as AFS and DYC in the
first control level are determined using appropriate
af , ar are the front and rear wheel slip angles, adaptive laws. Cornering stiffness, mass, speed, and
respectively. Cf , Cr are the cornering stiffnesses of tyre-road friction coefficient are constantly changing.
the front and rear tyres, respectively. Since AFS is Therefore, a nonlinear term is added to conventional
considered in this paper, so dr ¼ 0. Table 1 summa- adaptive control input to manage these parameter
rizes the parameters used in vehicle equations and uncertainties and external disturbances. This paper
notations. By inserting (2) into (1) and considering assumes that the plant states, reference targets, and
lateral disturbance force Fdy and disturbance plant parameter values are available (the yaw rate w_ is
moment Mdz , equation (1) is rewritten as follows measured through the gyroscope. The sideslip angle b
Ahmadian et al. 125

Table 1. Nomenclature and vehicle parameters.13

Symbols Value Unit Significations

m 1535 kg Vehicle mass


mr 648 kg Vehicle rear mass
Iz 2149 kgm2 Vehicle yaw inertia
Cf 40,000 N/rad Cornering stiffness of front tyres
Cr 40,000 N/rad Cornering stiffness of rear tyres
Lf 1.4 m Distance COG-front axle
Lr 1 m Distance COG-rear axle
tr 1.4 m Rear axle length
l ½0:4 : 1 – Tyre/road contact friction interval
v ½50 : 130 Km/h Vehicle velocity interval
vx ,vy – Km/h Longitudinal and lateral speeds at COG of vehicle respectively
R 0.3 m Effective tyre radius
Fx;f ,Fx;r – N Longitudinal forces on front wheel and rear wheel respectively
Fy;f ,Fy;r – N Lateral forces on front wheel and rear wheel respectively
af ,ar – rad Front and rear wheel sideslip angles respectively
b – rad Sideslip angle at COG of vehicle
bm – rad Desired value of sideslip angle
w_ – rad/s Yaw rate
w_ m – rad/s Desired value of yaw rate
dd – rad Steering angle from driver
d – rad Steering angle of front wheel produced by controller
Mz – Nm Corrective yaw moment produced by controller
Tb – Nm Transmitted wheel torque
Fdy – N lateral disturbance force
Mdz – Nm disturbance moment

Table 2. Abbreviations. laws. The purpose is to minimize the yaw rate track-
ing error between the actual behavior of the vehicle
Abbreviation Description
and the reference model so that the vehicle can bypass
AFS Active front steering a screw well in a specified manoeuvre while maintain-
DYC Direct yaw moment control ing stability. The state-space equations of the refer-
MRAC Model reference adaptive control ence model and the vehicle dynamics in the presence
NLMRAC Nonlinear model reference adaptive control
of uncertainties are as follows
KKT Karush–Kuhn–Tucker
DLC Double lane change
MIMO Multi–input multi–output x_ m ðtÞ ¼ Am xm ðtÞ þ Bm rðtÞ (5)
ESC Electronic stability control
ABS Anti–lock brake system x_ p ðtÞ ¼ ðAp þ DAp Þ xp ðtÞ þ ðBp þ DBp Þ uðtÞ þ GDðtÞ
TSC Traction stability control |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
A p B p
GPS Global positioning system
(6)

can be estimated by designing an observer or using where Ap 2 Rnn and Bp 2 Rnm are known matrices
GPS).13,26,27 and Am 2 Rnn is Hurwitz. Bm 2 Rnm , xp ; xm 2 Rn .
At the lower level, for the coordination between u; r 2 Rm1 are the control inputs and the reference
brake and steering control inputs and the optimal command, respectively and DðtÞ is the perturbation
use of brake forces, a constraint optimization algo- vector. DAp and DBp are time-varying matrices that
rithm based on longitudinal and lateral forces and the contain uncertain parameters of the plant such as lon-
tyre-road friction coefficient is presented. The yaw gitudinal velocity, mass, tyre-road friction coefficient
moment produced in the upper control level is con- and cornering stiffness. For the ST model, described
verted to brake torque and is applied to the wheels. in (4), n ¼ m ¼ 2, and
0 1
MRAC design: Upper-level controller L2f Cf þ L2r Cr Lr Cr  Lf Cf
B l l C
B Iz v x Iz C
In this section, the model reference adaptive control- Ap ¼ B C
@ L r Cr  L f Cf Cf þ Cr A
ler is designed. The upper level consists of a controller 1þl l
in which its parameters are updated through adaptive mv2x mvx
126 Proc IMechE Part K: J Multi-body Dynamics 235(1)

Figure 2. Integrated control structure.

0 1 angle variables from ideal values. Given the model


L f Cf 1 disturbances and parametric uncertainties, the con-
B I Iz C troller must be able to remove disturbances effective-
Bp ¼ B @ Cf
z C, where Bp is invertible. xp ¼
A ly. For this reason, an additional term is added to the
0
mvx adaptive control signal obtained through the match-
 T  T
½ w_ b T , u ¼ d Mz , r ¼ dd 0 , D ¼ ing conditions and adaptive laws. The NLMRAC
 T approach results in improved elimination of pertur-
Mdz Fdy , and G ¼ diagðI1z ; mv1 x Þ. Note that by
bations and convergence and removes large oscilla-
substituting l ¼ 1, Ap ¼ Am , Bp ¼ Bm . tions. Error dynamics are defined as eðtÞ ¼ xp ðtÞ
To manage the parametric uncertainties in the con- xm ðtÞ. Applying rðtÞ, the input signal uðtÞ
troller design, two assumptions are made. should be designed so that the tracking error eðtÞ
Assumption 1. Assume that the lumped distur- converges to zero, therefore (7) will be globally
bance can be written as fðxp ; tÞ ¼ DAp xp ðtÞ þ asymptotic stable system. Here a theorem is utilized
DBp uðtÞ þ GDðtÞ that consists of nonlinear terms for controller design.
resulting from parametric uncertainties and input
disturbances. Theorem 1. A ssuming fðxp ; tÞ is norm bounded, for
Assumption 2. The function fðxp ; tÞ is bounded by t 2 ½t0 ; tf  the plant (7) is globally asymptotic stable if
a vector with positive constants f : kf1 ðxp ; tÞk < f1 , the input vector uðtÞ is designed as follows
kf2 ðxp ; tÞk < f2 .
Thus, equation (6) can be written as follows uðtÞ ¼ uad ðtÞ þ unl ðtÞ (9)

x_ p ðtÞ ¼ Ap xp ðtÞ þ Bp uðtÞ þ fðxp ; tÞ (7) where uad is the MRAC input which is designed as
follows28,29
To determine f , 30% uncertainty for vehicle
parameters described in (4) is considered. uad ðtÞ ¼ KTx xp ðtÞ þ KTr rðtÞ (10)
Furthermore, the maximum values of the yaw rate
and sideslip angle are calculated as follows The unknown control gains Kx and Kr have to satisfy
  0:085 lg the matching conditions
w_ 
max (8)
vx Ap þ Bp KTx ¼ Am ; Bp KTr ¼ Bm (11)
bmax ¼ arctanð0:02 lgÞ

By calculating the saturation range for the brake and unl is the NLMRAC input to handle the system
and steering input actuators, the boundary of fðxp ; tÞ disturbance which is designed as follows
is obtained. As shown in Figure 2, the controller is
designed hierarchically. The upper-level controller unl ðtÞ ¼ B1
p ðgsatðeÞjejÞ (12)
includes the design of the AFS and DYC inputs,
which are designed to track the yaw rate and sideslip where g > 0, is the controller gain, if the following
condition is satisfied
Ahmadian et al. 127

The matrix PðtÞ is chosen to satisfy the Lyapunov If the adaptive laws are selected as
algebraic equation ATm P þ PAm þ Q ¼ P, _ where
_
_ Qb Þ. K^_ x ¼ Cx xp e PBp K^ r ¼ Cr re PBp
1 T 1 T
Q ¼ diagðQw; (20)

Proof : By inserting the input signal into (7) and


using the matching conditions (11), one can write And by inserting (20) into (19), one can write

_ ¼ Am eðtÞ þ Bp ðDKTx xp ðtÞ þ DKTr rðtÞÞ


eðtÞ V_ 1 ¼ eT Qe  0 (21)
(13)
þ Bp unl ðtÞ þ fðxp ; tÞ
V_ 1 becomes globally negative semidefinite. The pro-
where DKx ¼ K^ x  Kx and DKr ¼ K^ r  Kr represents posed NLMRAC can converge the system states to K
the parameter estimation errors. The candidate in finite time under boundary perturbations if
Lyapunov function is defined as (
 1
f þc 2
Vðe; DKx ; DKx Þ ¼ eT P e K¼ e : jej  (22)
(14) g
þ trðDKTx C1
x DKx þ DKTr C1
r DKr Þ

Cx and Cr are positive definite adaptive gain matrices. f > 0 is the upper bound of the lumped disturbance
Taking the derivative of (14) and using (13), gets vector that can be calculated. g > 0 is the controller
gain and positive integer. c ¼ ½c1 ; c2 T > 0 and c1 , c2

V_ ¼ eT ATm P þ PAm þ P_ e are arbitrary small positive values. By inserting
 unl ðtÞ ¼ B1 _
þ 2eT PBp DKTx xp þ eT PBp DKTr r p ðgsatðeÞjejÞ in V 2 , the following equa-
(15) tion can be obtained
þ2trðDKTx C1 ^_ T 1 ^_
x K x þ DKr Cr K r Þ 
 2eT P Bp uns þ f ¼ 2eT PðgsatðeÞjejÞ þ 2eT Pf (23)
þ2eT P Bp unl þ f

Two conditions are then considered


Then (15) can be simplified into the following relation
 Case 1: When kek  1. In this case, the following
V_ 1 ¼ eT ATm P þ PAm þ P_ e inequality can be written

þ 2eT PBp DKTx xp þ eT PBp DKTr r  
^_ _
T 1 ^
(16) V_ 2 ¼ 2geT Pejej þ 2eT Pf  2 gPjej2  Pf jej
þ 2trðDKTx C1x K x þ DKr Cr K r Þ
 (24)
V_ 2 ¼ 2eT P Bp unl þ f
Using the Lyapunov equation ATm P þ PAm þ Q ¼
_ one can write
P, According to Cauchy-Schwarz inequality:30
!  
V_ 1 ¼ eT ATm P þ PAm þ P_ e gPjej2  Pf  kPT k2 gjej2  f (25)
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}
Q
h  i
(17)
þ 2eT PBp DKTx xp þ 2tr DKTx C1 K^_ where P is a positive definite matrix obtained from
x x solving the algebraic Lyapunov equation in
h  i
^_ Assumption 1. Using K in (22), and for optional
þ 2eT PBp DKTr r þ 2tr DKTr C1
r Kr
values in e 2 R  K, there are
and then using the trace operator definition as follows
1
f þ c 2
! jej > (26)
eT PBp DKTx xp ¼ tr DKTx xp eT PBp (18) g
|fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl}
aT b b aT
Therefore, gjej2  f > c. Based on (24), and since
one can end up with c > 0, so for any e 2 R  K, one can write

 h i V_ 2 < cT jej < 0 (27)


V_ 1 ¼ eT Qe þ 2tr DKTx C1 K^_ þ xp eT PBp
x x
 h i In this section, for the first time, the stability analysis
_
1 ^
þ 2tr DKr Cr K r þ re PBp
T T
is performed for the MIMO version of the plant and
(19) the controller is designed as a centralized form. While
Meng et al.31 prove the error convergence as SISO
128 Proc IMechE Part K: J Multi-body Dynamics 235(1)

terms and two decentralized controllers are designed sideslip angle can be kept in the stability domain
for the steering and braking actuators. determined by the phase portrait.
There are two initial modes of eð0Þ. In the first,
eð0Þ 2 K, therefore it is not necessary to prove the
convergence performance. The other case is the
vector of eð0Þ is outside of K. Because for Efficient yaw moment distribution: Lower-level
eðtÞ 2 R  K, V_ 2 < 0, there exists t1 that eðt1 Þ 2 @K algorithm
where @K is the boundary of K. It can be concluded
The direct yaw moment input produced at the upper
that eðtÞ 2 K for t 2 ½t1 ; þ1Þ. By introducing
control level does not apply directly to the vehicle. In
the lower level, this input is converted to braking torque
U ¼ infjej (28)
e2@K and is optimally distributed between the rear wheels.
The goal of optimal control at the lower level is to min-
imize the use of brake forces to prevent driver com-
and
plaints regarding significant vehicle acceleration drop
  in a specified manoeuvre. In addition, the brake force
u ¼ gjej2  f jej (29) is optimized at the lower control level to prevent exces-
sive tyre wear. The yaw moment obtained at the upper
control level can be written as a function of the front
It can be obtained as
and rear tyre forces. To avoid interference between the

 1 front steering, it is assumed that the brake is only
f þc 2 applied to the rear wheels. The brake force should be
U¼ (30)
g applied to the rear left wheel for Mz > 0. Conversely,
and for Mz < 0, only the rear right wheel will feel the brake.
In both cases, the AFS is activated for both the left and

12 right front wheels. With this in mind, the coordination
f þ c
u ¼ cT ; u>0 (31) between the AFS and DYC is accomplished by present-
g ing a constraint optimization algorithm. Conventional
ECS methods based on brake force are often associated
Note that for each eðtÞ 2 K there exists with high costs in longitudinal perturbation, which
results in reduced acceleration and driver complaints.
V_ 2  u < 0 (32) To resolve this issue, the torque is considered as a com-
bination of brake and steering functions. In this case,
the AFS and rear brakes are synchronized together to
Because of the continuity of V2 and u, it can be con-
achieve the best performance.
cluded that there exists a tf > 0 that eðtÞ 2 K for
Here the integrated control algorithm is presented
t 2 ½t1 ; t1 þ tf . And then for any time, eðti Þ 2 K and
for the case of Mz > 0. The procedure can also be
V_ 2  u < 0 8 t ¼ ti . There exists tf > 0 that leads applied for Mz < 0. For simplicity, the front wheels
to having eðtÞ 2 K for t 2 ½ti ; ti þ tf . This means that are assumed to have the same steering angles
when system state variables reach the boundary, the dfl ¼ dfr ¼ df . The correct yaw moment Mz is written
error dynamics eðtÞ converge to the set K for each as a combination of brake and steering forces
time. Thus, for any ½t1 ; þ1Þ, there is



 1 tf
f þc 2 Mz ¼ cosdf  Lf sindf Fx;fl
jej  (33) 2
g

tr tf
þ Fx;rl þ Lf cosdf þ sindf Fy;fl
2 2 (35)


The case with kek > 1 can be proven in a similar tf
fashion. Therefore, the tracking problem for the þ Lf cosdf  sindf Fy;fr
2
system dynamics (7) is solved. To ensure lateral sta-  Lr Fy;rl  Lr Fy;rr
bility, the adaptive adjustment function Qb is defined
as follows
To reduce the optimization variables, the normal
j load ratios are described as
Qb ¼  n (34)
1  bb Fz;rl Fz;fr
Fx;rl ¼ Fx;fl ; Fy;fr ¼ Fy;fl ;
max

Fz;fl Fz;fl
(36)
Where j is a positive adjustment constant and n is a Fz;rl Fz;rr
Fy;rl ¼ Fy;fl ; Fy;rr ¼ Fy;rl
positive constant. Using this barrier function, the Fz;fl Fz;rl
Ahmadian et al. 129

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where Fz;ij are the normal loads on each wheel. Using l2 F2z;rl Mz 2
(36) Mz is rewritten as Fx;rl ¼ 1 þ 12 þ  (44)
1 þ k ð1 þ k Þb 2

!
tf Fz;fl tr a 2 aM Fz;fr
Mz ¼ cosdf  Lf sindf þ Fx;rl Where k ¼ b and 1 ¼ ð1þkÞzb2 , and Fy;fr ¼ Fz;fl Fy;fl .
2 Fz;rl 2
0 1 Therefore, the optimum brake and steering forces
tf are obtained. The corrective yaw moment generated
B Lf cosdf þ 2 sindf C
B
C (37) at the upper control level using NLMRAC is con-
B C
B tf Fz;fr C verted to brake torque at the lower level and applied
þB
B
Lf cosdf  sindf
2
CFy;fl
Fz;fl C to the rear wheels. The same procedure is conceivable
B C
B C for the case of Mz < 0 to obtain Fx;rr .
@  L Fz;rl  L Fz;rr A
r r
Fz;fl Fz;fl
Simulation results
Equation (37) can be rewritten as follows The performance of the proposed method is shown in
this section by studying the simulation results to control
Mz ¼ aFx;rl þ bFy;fl (38) the vehicle lateral dynamics in the J-turn and DLC
manoeuvres. The details of these standard manoeuvres
To design an algorithm for integration between are described in the ISO 3888 and ISO 7401 specifica-
AFS and DYC, the yaw moment distribution is opti- tions, respectively.33,34 To evaluate the effectiveness of
mized via an algorithm to minimize braking force Fx the hierarchical controller described in the previous sec-
on the rear wheels. In this regard, the cost function is tion in the presence of external disturbances and para-
defined as follows metric uncertainties, this method is compared with the
conventional MRAC method. The simulation results
JðFx;rl Þ ¼ F2x;rl (39) are plotted for three uncontrolled cases, controlled by
the conventional MRAC method as well as NLMRAC
with the nonlinear term. The vehicle parameters are
There are two constraints to the optimization described in Table 1. The controller parameters are set
problem. The first is equation (38). The second con- Cx ¼ diagð10; 10Þ, Cr ¼ diagð10; 1Þ, Qw_ ¼ 1, n ¼ 20,
straint, which indicates the physical limitation as well and j ¼ 10. Also Mdz ¼ Asin2pft (Nm) where A ¼
as the vehicle’s lateral stability range for the tyre, is as 1000 and f 2 ½0:1; 3.
follows.23,32
J-turn manoeuvre on a cobblestone road at 98 km/h
F2x;rl þ F2y;fl  l2 F2z;rl  0 (40)
In this scenario, the vehicle performs a J-turn
manoeuvre on a road with l ¼ 0:7. The yaw rate
Using equations (38) to (40), the Lagrange equa- tracking performance is illustrated in Figure 3(a).
tion is defined. The tracking performance of the vehicle using
NLMRAC and MRAC according to this figure is
LðFx;rl þ Fy;fl ; k; qÞ ¼ F2x;rl þ kðaFx;rl þ bFy;fl  Mz Þ much better than the passive vehicle. However, the
þ qðF2x;rl þ F2y;fl  l2 F2z;rl Þ sideslip angle of the controlled car with the extended
(41) method is lower than the uncontrolled and MRAC
cases according to Figure 3(a). Figure 3(b) shows
Where k is the Lagrange multiplier and q is a the AFS and DYC control inputs using conventional
positive quantity. By deriving (41) from its variables MRAC and the proposed nonlinear methods. In this
and applying the Karush-Kuhn-Tucker (KKT) opti- manoeuvre, the NLMRAC strategy can maintain
mality conditions, two acceptable responses are vehicle handling by reducing the sideslip angle and
derived achieving proper tracking performance at lower
Case 1: k ¼ 0; q ¼ 0; F2x;rl þ F2y;fl  l2 F2z;rl < 0 costs. As can be seen in this particular manoeuvre,
the rear brake actuators for the NLMRAC method
Mz
Fx;rl ¼ 0; Fy;fl ¼ (42) do not participate and only AFS is active. Thus, with-
b out deceleration, the car maintains its handling and
Case 2: F2x;rl þ F2y;fl  l2 F2z;rl ¼ 0; q > 0 stability. Whereas the conventional MRAC method
applies brake torque to keep stability. The lateral
Mz a acceleration is shown in Figure 4(a). As shown in
Fy;fl ¼  Fx;rl (43) this figure, this signal that shows slip and unstable
b b
features of the lateral behavior, for the controlled
Substituting (43) in F2x;rl þ F2y;fl  l2 F2z;rl ¼ 0, gets vehicle using NLMRAC is less than those of the pas-
sive and controlled car with MRAC. Sideslip angle
130 Proc IMechE Part K: J Multi-body Dynamics 235(1)

Figure 3. Comparison of the vehicle responses and control Figure 4. Comparison of lateral stability for different control
inputs for different control strategies in the J–turn manoeuvre, strategies in the J–turn manoeuvre, (a) lateral acceleration, (b)
(a) yaw rate and sideslip angle, (b) AFS and corrective yaw phase–plane b_  b.
moment.
reference value with minimum error. The sideslip
and its rate behavior are illustrated in Figure 4(b) in angle also decreased compared to the uncontrolled
the phase plane diagram. As can be seen from this car, indicating a decrease in lateral slip. The additive
figure, the vehicle controlled by the proposed control- steering angle and the yaw moment control signal in
ler, without braking, can stay in a safe range with less the DLC manoeuvre are shown in Figure 5(b). As can
steering input compared to the MRAC and stay away be seen from Figure 5(b), there is less control effort in
from the stability zone boundary, while the passive applying the AFS as well as the continuous control
car and the controlled one using MRAC enter the signal resulting from the extended control method.
unstable area. This is while the control signal in the MRAC
method has undesirable oscillations.
DLC manoeuvre on a dry road at 105 km/h Note that the yaw moment is converted to brake
2M R
In this scenario, the car moves at a desirable speed of torque according to DT ¼ Tb;left  Tb;right ¼ trz in
105 km/h on a road with l ¼ 0:9. Figure 5(a) shows the lower layer and the braking forces are optimally
the yaw rate and the sideslip angle tracking perfor- applied to the wheels according to (44).
mance, respectively, for the uncontrolled case and The lateral acceleration in the DLC manoeuvre is
controlled one using the MRAC and the NLMRAC shown in Figure 6(a). As can be seen from this figure,
method. According to this figure, the uncontrolled the lateral acceleration of the controlled vehicle using
vehicle is far from the reference value, and the con- the proposed method has smaller peaks than the con-
trolled one using conventional MRAC does not have ventional adaptive controller. This means that using
good tracking performance especially when the yaw the proposed integrated controller results in manag-
angle of the vehicle changes (Around 0.6 and ing vehicle handling and stability. Figure 6(b) illus-
1.6 seconds) has oscillations. The magnified version trates the sideslip dynamics changes in the phase
is shown below. While the controlled vehicle using plane. As can be seen from this figure, integrated
the proposed NLMRAC method can track the adaptive control without the use of the extended
Ahmadian et al. 131

Figure 6. Comparison of lateral stability for different control


Figure 5. Comparison of the vehicle responses and control strategies in the DLC manoeuvre, (a) lateral acceleration, (b)
inputs for different control strategies in the DLC manoeuvre, phase–plane b_  b.
(a) yaw rate and sideslip angle, (b) AFS and corrective yaw
moment.
coordination of AFS and DYC individual driving
assistant control modules. A NLMRAC method is
control strategy fails to maintain the lateral stability
designed at the upper control level to generate the
of the vehicle and, like the uncontrolled vehicle, enters
additive steering angle and corrective yaw moment.
the unstable and unsafe area. While the extended
A constraint optimization algorithm using KKT con-
optimal control method has been able to keep the
ditions in the lower control level is utilized to attain
system stable.
As indicated in Figures 3 to 6, it has been shown the optimum control laws as longitudinal and lateral
that optimal tracking performance is achieved by tyre forces. To illustrate the merits of using the non-
reducing the sideslip angle and lateral acceleration. linear controller and efficient yaw moment distribu-
The correct coordination between the brake and tion, a comparative simulation is conducted. The
steering actuators has allowed the controlled car in results show that, in the case of two J-turn and
the simulation scenarios to remain stable. Another DLC manoeuvres, the proposed control system can
achievement seen in the effectiveness of the optimal maintain a better tracking performance with lower
integrated NLMRAC design is its ability to eliminate tracking error compared to the case with no nonlinear
external disturbances, quickly and accurately track terms added and stabilize the vehicle in terms of
the yaw rate from the reference value, and reduce reducing the sideslip angle and lateral acceleration.
the sideslip angle and lateral acceleration by optimal- The proposed integrated control system is a potential
ly applying braking forces and steering. method to effectively control the lateral vehicle
dynamics encountered with external disturbances
and uncertainties. The main finding of this paper is
Conclusion that the extended MRAC method improves handling
A hierarchical control strategy to improve vehicle lat- performance to a large extent and ensure lateral sta-
eral stability and manoeuvreability is presented in this bility. The idea of coordination between AFS and
study. The proposed controller implies optimized DYC and the optimal brake force distribution results
132 Proc IMechE Part K: J Multi-body Dynamics 235(1)

to perform lateral driving manoeuvres with spending 11. Chokor A, Talj R, Doumiati M, et al. A global chassis
less control efforts. The control algorithm proposed control system involving active suspensions, direct yaw
in this study can also be utilized for other vehicle control and active front steering. IFAC-PapersOnLine
models. The practical implementations will be studied 2019; 52: 444–451.
12. Mirzaeinejad H, Mirzaei M and Kazemi R.
and in our future research.
Enhancement of vehicle braking performance on split-
l roads using optimal integrated control of steering and
Declaration of Conflicting Interests
braking systems. Proc IMechE, Part K: J Multi-body
The author(s) declared no potential conflicts of interest with Dynamics 2016; 230: 401–415.
respect to the research, authorship, and/or publication of 13. Ahmadian N, Khosravi A and Sarhadi P. Integrated
this article. model reference adaptive control to coordinate active
front steering and direct yaw moment control. ISA
Funding Trans 2020;
The author(s) received no financial support for the research, 14. Huang Y, and and Chen Y. Integrated AFS and ARS
authorship, and/or publication of this article. control based on estimated vehicle lateral stability
regions. In: 2018 Annual American Control Conference
ORCID iDs (ACC), Wisconsin Center, 27 June 2018, pp. 5516–
5521. Piscataway, NJ: IEEE.
Narjes Ahmadian https://orcid.org/0000-0002-4608-5914
15. Xu W, Fu ZJ, Xie WD, et al. Optimal torque vectoring
Alireza Khosravi https://orcid.org/0000-0003-3342-4144
control for distributed drive electric vehicle. IJMIC
2019; 31: 124–138.
References 16. Abe M. Vehicle dynamics and control for improving
1. Lutz A, Schick B, Holzmann H, et al. Simulation meth- handling and active safety: from four-wheel steering
ods supporting homologation of electronic stability to direct yaw moment control. Proc IMechE, Part K:
control in vehicle variants. Vehicle Syst Dynam 2017; J Multi-body Dynamics 1999; 213: 87–101.
55: 1432–1497. 17. Abe M, Kano Y, Shibahata Y, et al. Improvement of
2. Aghasizade S, Mirzaei M and Rafatnia S. Novel con- vehicle handling safety with vehicle side-slip control by
strained control of active suspension system integrated direct yaw moment. Vehicle Syst Dynam 1999; 33:
with anti-lock braking system based on 14-degree of 665–679.
freedom vehicle model. Proc IMechE, Part K: J 18. Goodarzi A and Ghajar M. Integrating lane-keeping
Multi-body Dynamics 2018; 232: 501–520. system with direct yaw moment control tasks in a
3. Huang D, Chen C, Huang T, et al. An active repetitive novel driver assistance system. Proc IMechE, Part K:
learning control method for lateral suspension systems J Multi-body Dynamics 2015; 229: 16–38.
of high-speed trains. IEEE Trans Neural Netw Learn 19. Guo Y, Guo H, Yin Z, et al. Vehicle lateral stability
Syst 2020; 31: 4094–4103. controller design for critical running conditions using
4. Doumiati M, Sename O, Dugard L, et al. Integrated NMPC based on vehicle dynamics safety envelope. In:
vehicle dynamics control via coordination of active 2019 IEEE international symposium on circuits and sys-
front steering and rear braking. Eur J Control 2013; tems (ISCAS), Sapporo, Hokkaido, Japan, 26 May
19: 121–143. 2019, pp. 1–8. Piscataway, NJ: IEEE.
5. Kim TU, Cheon S and Yang SY. A study on develop- 20. Hu C, Wang R, Yan F, et al. Output constraint control
ment of real-time simulator for electric traction control on path following of four-wheel independently actuated
system. J Drive Control 2019; 16: 67–74. autonomous ground vehicles. IEEE Trans Vehicle
6. Asiabar AN and Kazemi R. A direct yaw moment con- Technol 2016; 65: 4033–4043.
troller for a four in-wheel motor drive electric vehicle 21. Gipser M. FTire—the tyre simulation model for all
using adaptive sliding mode control. Proc IMechE, Part applications related to vehicle dynamics. Vehicle Syst
K: J Multi-body Dynamics 2019 ; 233: 549–567. Dyn 2007; 45: 139–151.
7. Jaafari SM and Heidari Shirazi K. A comparison on 22. Mavros G, Rahnejat H and King P. Analysis of the
optimal torque vectoring strategies in overall perfor- transient handling properties of a tyre, based on the
mance enhancement of a passenger car. Proc IMechE, coupling of a flexible carcass—belt model with a sepa-
Part K: Journal of Multi-Body Dynamics 2016; 230: rate tread incorporating transient viscoelastic frictional
469–488. properties. Vehicle Syst Dynam 2005; 43: 199–208.
8. Fruechte RD, Karmel AM, Rillings JH, et al. 23. Ahmadian N, Khosravi A and Sarhadi P. Adaptive yaw
Integrated vehicle control. In: Proceedings of the 39th stability control by coordination of active steering and
IEEE vehicular technology conference, San Francisco, braking with an optimized lower-level controller. Int J
CA, 1989, Vol. 2, pp.868–877. Piscataway, NJ: IEEE. Adapt Control Signal Process 2020; 34: 1242–1258.
9. Goodarzi A, Sabooteh A and Esmailzadeh E. 24. Hegazy S, Rahnejat H and Hussain K. Multi-body
Automatic path control based on integrated steering dynamics in full-vehicle handling analysis. Proc
and external yaw-moment control. Proc IMechE, Part IMechE, Part K: J Multi-body Dynamics 1999; 213:
K: J Multi-body Dynamics 2008; 222: 189–200. 19–31.
10. He J, Crolla DA, Levesley MC, et al. Coordination of 25. Liao YG and Du HI. Cosimulation of multi-body-
active steering, driveline, and braking for integrated based vehicle dynamics and an electric power steering
vehicle dynamics control. Proc IMechE, Part D: J control system. Proc IMechE, Part K: Journal of Multi-
Automobile Engineering 2006; 220: 1401–1420. Body Dynamics 2001; 215: 141–151.
Ahmadian et al. 133

26. Doumiati M, Charara A, Victorino A, et al. Vehicle sideslip angle estimation using lateral tire force sensors.
dynamics estimation using Kalman filtering: experimen- IEEE Trans Vehicular Technol 2012; 61: 1972–1985.
tal validation. Hoboken, NJ: John Wiley, 2012. 31. Meng Q, Zhao T, Qian C, et al. Integrated stability
27. Yoon JH, Li SE and Ahn C. Estimation of vehicle side- control of AFS and DYC for electric vehicle based on
slip angle and tire-road friction coefficient based on non-smooth control. Int J Syst Sci 2018; 49: 1518–1528.
magnetometer with GPS. Int Automot Technol 2016; 32. Lu S, Cen S, Hu X, et al. J. Integrated control of brak-
17: 427–435. ing and steering subsystems for autonomous vehicle
28. Ahmadian N, Khosravi A and Sarhadi P. Adaptive based on an efficient yaw moment distribution. IEEE
control of a jet turboshaft engine driving a variable Trans Industrial Electron 2017; 99: 1–1.
pitch propeller using multiple models. Mech Syst 33. Jalali K, Uchida T, McPhee J, et al. Development of an
Signal Process 2017; 92: 1–2. advanced fuzzy active steering controller and a novel
29. Ahmadian N, Khosravi A and Sarhadi P. A new method to tune the fuzzy controller. SAE Technical
approach to adaptive control of multi-input multi- Paper 2013-01-0688, 2013.
output systems using multiple models. J Dynam Syst 34. Mastinu G, Gobbi M, and Miano C. Optimal Design of
Meas Control 2015; 137 Complex Mechanical Systems: With Applications to
30. Nam K, Fujimoto H and Hori Y. Lateral stability con- Vehicle Engineering. In: Springer Science & Business
trol of in-wheel-motor-driven electric vehicles based on Media, 2007, pp.191–214.

You might also like