CHINESE JOURNAL OF MECHANICAL ENGINEERING

·486· Vol. 29,aNo. 3,a2016

DOI: 10.3901/CJME.2016.0314.031, available online at www.springerlink.com; www.cjmenet.com

Direct Yaw Moment Control for Distributed Drive Electric Vehicle

Handling Performance Improvement

YU Zhuoping1, LENG Bo1, XIONG Lu1, *, FENG Yuan2, and SHI Fenmiao1
1 School of Automotive Studies, Tongji University, Shanghai 201804, China
2 Pan Asia Technical Automotive Center Co., Ltd., Shanghai 201201, China

Received June 7, 2015; revised August 10, 2015; accepted March 14, 2016

Abstract: For a distributed drive electric vehicle (DDEV) driven by four in-wheel motors, advanced vehicle dynamic control methods
can be realized easily because motors can be controlled independently, quickly and precisely. And direct yaw-moment control (DYC)
has been widely studied and applied to vehicle stability control. Good vehicle handling performance: quick yaw rate transient response,
small overshoot, high steady yaw rate gain, etc, is required by drivers under normal conditions, which is less concerned, however. Based
on the hierarchical control methodology, a novel control system using direct yaw moment control for improving handling performance of
a distributed drive electric vehicle especially under normal driving conditions has been proposed. The upper-loop control system consists
of two parts: a state feedback controller, which aims to realize the ideal transient response of yaw rate, with a vehicle sideslip angle
observer; and a steering wheel angle feedforward controller designed to achieve a desired yaw rate steady gain. Under the restriction of the
effect of poles and zeros in the closed-loop transfer function on the system response and the capacity of in-wheel motors, the integrated
time and absolute error (ITAE) function is utilized as the cost function in the optimal control to calculate the ideal eigen frequency and
damper coefficient of the system and obtain optimal feedback matrix and feedforward matrix. Simulations and experiments with a DDEV
under multiple maneuvers are carried out and show the effectiveness of the proposed method: yaw rate rising time is reduced, steady yaw
rate gain is increased, vehicle steering characteristic is close to neutral steer and drivers burdens are also reduced. The control system
improves vehicle handling performance under normal conditions in both transient and steady response. State feedback control instead of
model following control is introduced in the control system so that the sense of control intervention to drivers is relieved.

Keywords: direct yaw moment control, distributed drive electric vehicle, handling performance improvement, state feedback control

on DDEV the range of DYC is expanded, which makes the

1 Introduction road adhesion utilization of four tires better balanced and
extends the vehicle stability margin[4]; meanwhile, the
Recently the direct yaw-moment control (DYC) has been DDEV-based DYC will not cause a strong sense of
widely applied in order to improve vehicle handling intervention to drivers and is more energy efficient[5].
performance and stability[1]. For conventional internal In a variety of research on DDEV, DYC has focused on
combustion engine drive vehicles (ICVs) equipped with vehicle stability control under critical conditions[6–9], e.g.
anti-lock braking system (ABS) and traction control system low adhesion road, high speed and large lateral acceleration.
(TCS), DYC is usually realized by applying different braking HE and his group proposed a hierarchical control
force to wheels. But the brake system based DYC methodology for stability improvement of 4WD EV in
deteriorates acceleration[2] and the hydraulic unit cannot critical driving conditions[10]. The upper controller was
respond very fast. Compared to an ICV, a distributed drive based on the model following control (MFC) method and
electric vehicle (DDEV), which is driven by four in-wheel the lower controller distributes driving/braking force to
motors, has advantages not only in environmental protection each in-wheel motor to minimize tire utilization according
but also in DYC[3]: Driving or braking torque on each wheel to the generalized force calculated by the upper controller.
can be controlled independently, more quickly, and more Motoki Shino and Masao Nagai researched direct yaw
precisely. Additionally, torque and speed information of the moment control distribution methods to improve handling
in-wheel motor can be obtained easily. Furthermore, based and stability of electric vehicles, in which a model
following controller was also used as an upper
controller[11–12]. In Ref. [13], an adaptive direct yaw
* Corresponding author. E-mail: xiong_lu@tongji.edu.cn
Supported by National Basic Research Program of China (973 Program, moment control method based on identification of yaw rate
Grant No. 2011CB711200), National Science and Technology Support model was proposed so that electric vehicles can track the
Program of China (Grant No. 2015BAG17B00), and National Natural
Science Foundation of China (Grant No. 51475333)
desired dynamic model. Existing DYC research pays less
© Chinese Mechanical Engineering Society and Springer-Verlag Berlin Heidelberg 2016 attention to normal driving conditions. However, according
 CHINESE JOURNAL OF MECHANICAL ENGINEERING ·487·

to statistical results from Germany[14], the lateral acceleration

on a good friction road is less than 4 m/s2 in 95% of
cornering driving conditions. Namely, vehicles work mostly
under normal conditions where good handling performance is
more urgently required by drivers: Vehicle yaw rate responses
to the steering wheel input quickly with small overshoot;
steering characteristics stay close to neutral steer even in high
speeds and acceleration. Normally, the adhesion condition of
the front axle is worse than the rear axle, which means the
front axle gets into a non-linear region more easily and leads
to bad maneuverability. Under normal conditions, tires stay in
a linear region, and the vehicle system can be regarded as a
linear system. According to the modern control theory[15],
closed-loop poles and zeros in the controllable linear system
can be placed at random through state feedback control to
improve dynamic performance and stability margin.
This paper relies on the linear vehicle model by using the
direct yaw moment control to improve the handling Fig. 1. Distributed drive electric vehicle
performance of a distributed drive electric vehicle equipped
with four in-wheel motors especially under normal driving Table 1. DDEV main parameters
conditions. The designed controller consists of two parts: a Parameter Value
state feedback controller, which aims to improve yaw rate Front axle to CG lf/mm 1167
transient response; and a steering wheel angle feedforward Rear axle to CG lr/mm 1233
controller to achieve desired yaw rate steady gain because Wheel base l/mm 2400
the state feedback will decrease the gain of steady-state Wheel track B/mm 1416
yaw rate[16]. The integrated time and absolute error (ITAE) Rolling radius r/mm 292
function[17] was utilized as the cost function in the optimal Vehicle mass m/kg 1022
control to calculate optimum feedback matrix and Yaw moment of inertia Iz/(kg • m2) 1470
feedforward matrix to assign ideal poles and zeros and to Steering system ratio 17.5
Peak power (one motor) P/kW 7.5
obtain enough steady gain of yaw rate. A state estimator
Peak torque (one motor) T/(N • m) 167
based on the extended Kalman filter was adopted to obtain
Note: CG means the center of gravity
vehicle sideslip angle.
The allocation module is responsible for distributing the
generalized force calculated by the upper controller to the 3 Control System Structure
actuators, i.e., the four independent in-wheel motors, on the
premise that the driver’s intention is satisfied. The designed controller is distinguished by the dotted
line shown in Fig. 2. State feedback can compensate the
2 Distributed Drive Electric Vehicle error caused by external disturbance or model uncertainty.
A full dimension state feedback controller was adopted to
A high performance distributed drive electric vehicle assign poles and zeros arbitrarily, namely, both of the
platform developed by Tongji University shown in Fig. 1(a) vehicle sideslip angle  and the yaw rate are fed back.
is equipped with a storage battery as the power source and However, feedback reduces system steady gain and steering
four independent in-wheel motors. The main parameters of sensitivity and results in worse handling performance. To
the DDEV are shown in Table 1 and the measurement solve this problem, a steering wheel angle feedforward
system is shown in Fig. 1(b). controller was designed.

Fig. 2. Control system structure

 YU Zhuoping, et al: Direct Yaw Moment Control for Distributed Drive Electric Vehicle
·488· Handling Performance Improvement

The allocation module is responsible for distributing the ïìïå Fy = Fyf cos  f + Fyr ,
generalized force calculated by the upper controller and í (1)
ïïå M z = l f Fyf cos  f - lr Fyr ,
demanded by the driver to the four independent in-wheel î
motors.
The sideslip angle of the vehicle is an important state where  f is the steering angle of front wheels and is small,
variable to the controller. But it is hard to be measured namely cos  f » 1 . Fyf and Fyr are equivalent lateral forces
directly and a sideslip angle sensor is too expensive to be on front axle and rear axle respectively, equal to the
employed in ordinary cars. Therefore, an estimator is product of tire cornering stiffness and tire slip angle. The
adopted. Simulation results show the accuracy of the equivalent tire slip angles of front and rear axle can be
sideslip angle observer under normal conditions and prove calculated as follows:
that it meets control requirements.
ì
ï
ï l f 
ïf =  + - f ,
4 Control System Design ï
ï V
í
ï
ï l 
ï
ïr =  - r .
4.1 Linear vehicle model ï
î V
In order to make the control system design convenient
and to reflect the main characteristics of vehicle handling, Eq. (1) can be written as
we make some ideal assumptions about the vehicle system.
(1) Drive on a flat road, no vertical road roughness input. ì
ç  + l f  -  ÷÷ - C æç  - lr ö÷÷ ,
ï æ ö
ï
Ignore vertical forces influence and coupling effects related ï
ï å F = - C ç
ççè ÷ rç
ø÷ èç V ø÷
y f f
ï
ï V
to ride dynamics. í (2)
(2) Ignore suspension system; hence, load transfer and ïï æ l f  ö æ l  ö
ï
ï å M z = -l f C f ççç  + -  f ÷÷÷+ lr Cr çç  - r ÷÷ ,
suspension dynamics are not taken into consideration. ï
ï
î èç V ÷
ø ç
è V ø÷
(3) Steering system is rigid, and the transmission ratio
between steering wheel and front wheels is constant.
where  —Vehicle sideslip angle, rad;
(4) Ignore air resistance.
 —Yaw rate, rad/s;
(5) Assure minimal disturbance of vehicle when it is near
V —Longitudinal velocity, m/s;
balance state. The lateral acceleration should be small (less
l f , lr —Distance from front/rear axle to CG, m;
than 0.4g on high friction road). Tires work in linear region,
C f , Cr —Equivalent tire cornering stiffness of front/rear
which means the lateral tire force merely increases
axle, positive, N/rad.
proportionally as tire slip angle increases.
Furthermore,
(6) The longitudinal velocity is constant.
Based on those assumptions, the vehicle is simplified to
ìå Fy = ma y ,
ï
a typical two degrees of freedom model (2DOF model) as ï
í (3)
shown in Fig. 3. ï
ïå M z = I z,
î

where I z is the moment of inertia in yaw motion, kg • m 2 ;

 is yaw acceleration, rad • s-2 . The vehicle lateral
acceleration a y = U + V , m • s-2 , in which U denotes
lateral speed of CG and U =  V . So

a y = (  + )V . (4)

By combining Eqs. (2)–(4), the differential equation of

2DOF vehicle motion are obtained as follows:
Fig. 3. 2DOF vehicle model

Only vehicle sideslip angle and yaw rate are taken into ì
ï 1
ï
ï-(C f + Cr )  + (-l f C f + lr Cr ) +
account as state variables. According to existing vehicle ï
ï V
ï
ï
dynamics research[3, 9], this kind of simplification is ï C  = mV (  + ),
ï
í
f f
(5)
reasonable and effective. In this paper, the 2DOF model is ï
ï 1
the fundamental model for control system design. ï
ï(-l f C f + lr Cr )  + (-l f 2 C f + lr 2 Cr ) +
ï
ï V
Newton's laws of motion are used to establish vehicle ï
î l f C f  f = I z .
ï 
dynamic equations of plane motion. With reference to Fig. ï
3 forces along y axis and torques acting on the center of
gravity are described as The state space can be written as
 CHINESE JOURNAL OF MECHANICAL ENGINEERING ·489·

ì
ï
ï x = A  x + B   f , 4.2.1 Feedback matrix
í (6) The system shown in Eq. (7) can be described as
ï
î y＝C  x ,
ï
ì x = ( A + Q )  x + ( B + P )   f ,
ï
where state variable
T
x = [   ] , output y =  , ï
í (11)
ï
î y = C  x.
ï
system-matrix

é (C + Cr ) (C f l f - Cr lr ) ù The characteristic equation of the system matrix is

ê - f -1 - ú
ê mV mV 2 ú
ê
A= ê ú , s - a11 -a12
ê (C f l f - Cr lr ) (C f l 2f + Cr lr2 ) úú = s 2 - ( a11 + a22 + Q2 ) s +
ê- - ú -( a21 + Q1 ) s - ( a22 + Q2 )
êë Iz I zV úû
a11 (a22 + Q2 )- a12 (a21 + Q1 ) = 0 (12)
and control matrix
Poles of the ideal system are
é Cf ù
ê ú
ê mV ú
B = êê ú. s1,2 = -n  n 1-  2  j. (13)
C l ú
ê f f ú
ê I ú
ë z û Combine Eqs. (12) and (13), get

Under normal conditions, the vehicle lateral acceleration a11 ( a11 + 2n ) + n 2
is no more than 0.4g and the equivalent cornering stiffness Q1 = - - a21 , (14)
a12
C f and Cr can be regarded as constant[14]. It is easy to get

rank[ B, AB ] = 2. Q2 = -2n - a11 - a22 . (15)

The original system is controllable. By substituting Q1 and Q2 in Eq. (9), the feedback
Add extra yaw moment M z . The original state space Eq. matrix can be obtained.
(6) can be rewritten as Eqs. (14) and (15) indicate that the feedback matrix is
determined by parameters of the original system, ideal
ì
ï 1 eigen frequency and damper coefficient.
ï
ï x = A  x + B   f +  M z ,
í I (7)
ï
ï
z
î y＝C  x.
ï 4.2.2 Feedforward matrix
In the controlled system, the transfer function describing
4.2 Control strategy the relationship between the yaw rate and the steering angle
Effects caused by poles and zeros to the system shown in is obtained by substituting Eq. (8) in the state space Eq. (6):
Eq. (7) are considered simultaneously in this paper. In the
control strategy, state feedback control is introduced to b1 (a21 + Q1 ) - a11 (b2 + p )
Gf ( s ) ¢ = 
assign ideal poles, and yaw rate steady gain is ensured by a n 2
steering wheel angle feedforward controller.
1 + Ts s
The extra yaw moment M z can be described in two , (16)
2 1
parts shown in Eq. (8): 1+ s+ s2
n n 2

M z = I z Qx + I z Pu ,
where
é 0 0 ù é ù é 0 ù
Mz = ê ú  ê ú+ê ú  . (8)
ê I z Q1 I z Q2 ú ê ú ê I z p ú f
ë û ë û ë û b2 + p
Ts = ,
b1 (a21 + Q1 ) - a11 (b2 + p)
The feedback matrix can be obtained from Eq. (8):

é 0 0 ù and the gain matrix of the controlled system is

IzQ = ê ú. (9)
ê I z Q1 I z Q2 úû
ë
b1 (a21 + Q1 ) - a11 (b2 + p )
K= .
And the feedforward matrix is n 2

é 0 ù If K is known (K is calculated in section 4.2.3), we can

Iz P = ê ú. (10)
ê I z pú get
ë û
 YU Zhuoping, et al: Direct Yaw Moment Control for Distributed Drive Electric Vehicle
·490· Handling Performance Improvement

b1 (a21 + Q1 ) - n 2 K The steady gain K of the controlled system is 1~(1+kV2)

p= - b2 =
a11 times larger than it was in the original system (if K=1+kV2,
the controlled system becomes neutral steer). K varies with
a11 (a11 + 2n ) + n 2
n 2 K + b1  vehicle velocity as shown in Fig. 4(a) because of the
a12
- - b2 . (17) stability factor k. So does the required yaw moment
a11 deduced from Eqs. (21) and (22) .
In Fig. 4(a), the horizontal line denotes the required yaw
Zeros of the controlled are moment used to change the original system to a
neutral-steer system when velocity varies and lateral
b1 (a21 + Q1 )
s =- + a11 = f (n ,  , K ). (18) acceleration maintains 0.4g. The other three curves show
b2 + p required yaw moment when K equals different multiples of
the original at different speeds and with the same lateral
By substituting Eq. (17) in Eq. (10), the feedforward acceleration (0.4g). As Fig. 4(a) shows, a larger yaw
matrix is obtained as moment is required with the increase of K.
As a fixed-proportional relation between the yaw rate
é0ù and the velocity provides drivers coziness, a constant
Iz P = Iz ê ú =
ê pú steady gain K that can relieve drivers' effort and keep
ë û
é 0 ù vehicle stable should be determined. At the working speed
ê ú
ê æ ö÷ú of the control system, V=30–80 km/h, the ideal K curve
ê çç  2 K + b  a11 (a11 + 2n ) + n
2

ê ç ÷÷úú should stay below the neutral steer line or the controlled
ç
n 1
a ÷.
ê I ç-
ê z çç
12
- b2 ÷÷÷úú (19) system would become an oversteering system and be prone
ê çç a11 ÷÷ú
ê çç ÷÷ú to instability. With reference to Fig. 4(a), the ideal steady
ëê è ø÷ûú gain is

The feedforward matrix varies with the ideal eigen V /l

K ideal = 1.04  ,
frequency and the damper coefficient as the feedback 1 + kV 2
matrix.
The limit of yaw rate  lim is defined as
4.2.3 Steady state gain
The transfer function, which describes the relationship ay 0.6 g
between the yaw rate under steady state and the extra yaw  lim = = .
V V
moment, can be obtained from Eq. (6):
And the limit steady gain is

(C f + Cr )V
GD = . (20)
Mz
(l f + lr ) 2 C f Cr - mV 2 (C f l f - Cr lr )  lim 0.6  g
K lim = =
f V f
Compared with the original system, the extra yaw
moment that needs to be added to the controlled system is To ensure the effectiveness of control and keep the
vehicle system linear, the real gain under steady state is the
D smaller value selected from Kideal and K lim , namely
DM z = 
=  
GD Mz
ìï V /l 0.6  g ü
ï
(l f + lr ) 2 C f Cr - mV 2 (C f l f - Cr lr ) K = min {K ideal , K lim } = min ïí1.04  , ï.
ý
, (21) ïï 1 + kV 2
V  ï
ï
î f þ
(C f + Cr )V

4.2.4 Ideal eigen frequency and damper coefficient

where  is the differential steady-state yaw rate of the The poles and zeros placement is the purpose of the
controlled system to the original system and calculated by: control system design. The poles and zeros of the
controlled system, by Eqs. (13), (17) and (18), are decided
V /l
 =  f ( K -1)  , K Î éê1, 1 + kV 2 ùú . (22) by the eigen frequency and the damper coefficient. So an
1 + kV 2 ë û
optimal control method is proposed to calculate the ideal
eigen frequency and damper coefficient. ITAE is a quality
where k is the stability factor of the original vehicle, and index with good comprehensive dynamic performance and
utilized as the objective function of the optimal control.
m æç l f l ÷÷ö Control variables n and ζ are subject to the following
k= ç - r
2 ç ÷÷.
l çè Cr C f ÷ø rules and conditions.
 CHINESE JOURNAL OF MECHANICAL ENGINEERING ·491·

(1) According to stability condition, all poles must be Based on the above allocation strategy, the maximum
placed in the left half s-plane. yaw moment for control generated by the four in-wheel
From Eq. (13), motors of the DDEV is shown in Fig. 4(b), which is
another constraint to n and ζ.
-n < 0. To solve the ITAE function within those constraints

(2) The system should be small damping or critical ts V /l

ITAE = ò t e(t ) dt，e=  f  K  - act . (23)
damping to ensure the convergence of yaw rate at any 0 1 + kV 2
vehicle speed. So
It is important to point out that although the variables are
 ≤ 1 and n > 0.
only n and ζ, they decide poles directly. And combining
with velocity the two variables can determine zeros as well.
(3) The maximum required yaw moment for control n and ζ obtained by the optimal control method can place
cannot exceed the in-wheel motors’ capacity. the poles and the zeros simultaneously.
As the lower controller is not the emphasis of this paper, The optimization results at different vehicle velocities
a simple rule-based allocation strategy is proposed, namely, are shown in Figs. 4(c) and 4(d).
the yaw moment obtained by the upper controller is By substituting the optimization results in Eqs. (15) and
allocated to the front and rear axle averagely and the left (19), the ideal feedback and feedforward matrixes can be
and right motor of each axle generate opposite and equal obtained.
moment.
The required moments of the front axle Trq , f and the 4.3 Vehicle sideslip angle observer
rear axle Trq , r can be calculated as follows:
4.3.1 Sideslip angle observer design
ì
ïTrq , f = 2Tmax  ia ,
ï
í The sideslip angle is a key state variable of the system
ï
îTrq , r = 2Tmax  ia ,
ï and it cannot be measured directly. So an observer based on
extended Kalman filter was adopted to provide vehicle
where Tmax is the maximum output torque of an in-wheel sideslip angle information to the controller. A diagram
motor, ia Î [0,1] denotes gas pedal input. block of the observer is shown in Fig. 5. Uncertainty during
Trq , f and Trq , r are distributed to the left and right vehicle run time can be due to tires getting into nonlinear
in-wheel motors of an axle with the same method. Take the region. In order to ensure the accuracy of the observer
rear axle as an example: within as vast a scale as possible[18] and avoid a large
If Trq , r < Tr ,max , then number of floating-point computations led by solving
Jacobian matrixes[19], a nonlinear 2DOF vehicle model
Trq , r - Tr Trq , r + Tr based on arctangent tire model is built for the observer.
Trl = , Trr = ,
2 2 The nonlinear 2DOF vehicle system is described as

where Tr ,max is the maximum torque that the rear axle can ì
ï 1 ìï é æ öù
be generated, T is the differential torque calculated by
ï
ï x1 = ïíc atan ê c çç - x - l f x ÷÷ú cos  +
ï ê ççè ÷
V ÷øúúû
f1 f 2 f 1 2 f
ï
ï mV ïï ê
ë
the upper controller, Trl and Trr are the rear-left and the ï î
ï
ï
rear-right in-wheel motor torques respectively. ï é æ l öù ü
ï
ï
ï cr1 atan êcr 2 çç-x1 + r x2 ÷÷÷ú ï ý - x2 ,
If Trq , r > Tr ,max , the in-wheel motor cannot generate ï
ï ê ç
è V øú ï
ï ë û ï
þ
enough traction torque while meeting the requirement of í (24)
ï ì é æ l f öù
ï 1ï
the differential torque. And the traction requirement should ï
ï x2 = ïíl f c f 1 atan êêc f 2 ççç f - x1 - x2 ÷÷÷úú cos  f -
ï I z ïï ç V ÷øûú
be satisfied at first. So ï
ï î ëê è
ï
ï
ï é æ l öù ü
ï
æ Trq , r ÷ö æ T ö ïï
ï l c atan ê cr 2 çç-x1 + r x2 ÷÷ú + uï ý.
Trl = Tmax - çççTmax - ÷÷ - sgn( Tr ) ççTmax - rq , r ÷÷÷ ,
r r 1 ê ç ÷ú ï
ç
ïï
î ë è V øû ïþ
èç 2 ÷ø èç 2 ÷ø
æ Trq , r ÷ö æ T ö where x1 =  , x2 =  , u = Mz.
Trr = Tmax - çççTmax - ÷ + sgn(Tr ) ççTmax - rq , r ÷÷÷.
÷
çè 2 ÷ø ççè 2 ÷ø An extended Kalman filter is based on the traditional
Kalman filter, expands the nonlinear function in Taylor
Characteristics of electrical components of battery or series, omits the second order and finishes linearization.
motors are not taken into account. Regenerative braking The nonlinear Eq. (24), which is estimated by an extended
coefficient equals to 1. Motors that are able to regenerative Kalman filter, can be described as
brake even at low speed have the same external
characteristics during drive and brake. x = f ( x(t ), u (t )), y = h( x(t ), u (t )).
 YU Zhuoping, et al: Direct Yaw Moment Control for Distributed Drive Electric Vehicle
·492· Handling Performance Improvement

Fig. 4. Main parameters of the control system

¶f 2 1é l f c f 1c f 2 cos 
= êê- +
¶x1 J ê 1 + c f 2 ( - x1 - (l f v)  x2 ) 2
2
ë
lr cr1cr 2 ù
ú,
2 2ú
1 + c r 2 (-x1 + (lr v)  x2 ) úû

é
ê lf
ê l f c f 1c f 2 cos 
¶f 2 1ê v
= ê- -
¶x2 Jê æ l ö2
÷
ê ç f
ê 1 + c f 2 çççè - x1 - v x2 ÷÷ø
2
Fig. 5. Diagram block of sideslip angle observer ÷
êë
The Jacobian matrixes are as follows: ù
l ú
lr cr1cr 2 r ú
v ú,
¶f1 1 éê c f 1c f 2 cos  æ lr ÷ö ú
2ú
= ê- - 2 ç
1 + c r 2 ç-x1 + x2 ÷÷ ú
¶x1 mv ê 1 + c f 2 ( - x1 - (l f v)  x2 ) 2
2
èç v ø úû
ë
cr1cr 2 ù
ú, ¶h ¶h
1 + c 2 r 2 (-x1 + (lr v)  x2 ) 2 úúû = 0, = 1.
¶x1 ¶x2

é
ê lf After linearization, the system can be estimated using a
ê c f 1c f 2 cos  traditional Kalman filter approach.
¶f1 1 ê v
= ê- +
¶x2 mv ê æ l ö2
÷
ê 1 + c çç - x - x ÷
2 f
4.3.2 Sideslip angle observer validation
ê v ÷÷ø
f2ç 1 2
êë çè
The precision of the sideslip angle observer was
ù validated through simulations in Carsim®. The vehicle
l ú
cr1cr 2 r ú configuration in Carsim® is set according to Table 1.
v ú -1,
2ú Simulation condition: Double-lane change test with vehicle
æ l ö ú
1 + c 2 r 2 çç-x1 + r x2 ÷÷÷ ú velocity at 40 km/h and tire-road friction coefficient
çè v ø úû =1.0.
 CHINESE JOURNAL OF MECHANICAL ENGINEERING ·493·

The results in Fig. 6 indicate that the observer had For the sideslip angle observer, more details and multiple
satisfactory accuracy when the lateral acceleration was less validations studied by the author’s team can be found in
than 0.6  g , namely, under normal driving conditions. Ref. [20].

Fig. 6. Results of double-lane change test

tire test data.

5 Simulation Results of the Control System
5.1 Steering wheel angle step input test
To check the performance of the proposed control system, Transient and steady-state responses of this test indicate
simulations were carried out based on a Carsim® and time-domain response of vehicle handling stability. In
MATLAB/Simulink joint simulation platform. The vehicle simulations, vehicle velocity was 50 km/h and constant;
configuration in Carsim® was set according to Table 1. A road friction coefficient was 1.0, and steering wheel turned
simplified magic formula tire model was obtained by fitting 60° in 0.2 s. Simulation results are shown in Fig. 7.

Fig. 7. Simulation results in steering wheel angle step input test

 YU Zhuoping, et al: Direct Yaw Moment Control for Distributed Drive Electric Vehicle
·494· Handling Performance Improvement

As shown in Fig. 7(a), it took about 0.5 s for the steering wheel angle at 60°, accelerated vehicle uniformly
uncontrolled vehicle to raise the yaw rate from zero to peak and continuously with the longitudinal acceleration less
but only 0.3 s for the controlled vehicle, which means than 0.25 m/s2 until the lateral acceleration was raised to
improvement in transient response. The overshoot of the 6.5 m/s2. Tire-road friction coefficient was 1.0. Simulation
controlled system was 2.8% while 0.3% in the uncontrolled results are shown in Fig. 8.
system, but still remained within the engineering Fig. 8(a) shows obviously that the yaw rate curve in the
permission scope. controlled system raises faster than in the uncontrolled
system, which means the decrease of understeer
5.2 Steady state turning test characteristics and the increase of yaw-rate steady-state
The purpose of this test is to obtain steady-state yaw rate gain. Drivers’ burdens are reduced.
response and vehicle steer characteristics. Operator kept

Fig. 8. Simulation results in steady state turning test

were all reduced, especially in the latter half of the test,

6 Experimental Results of Control System which means the controlled vehicle had better path tracking
ability. Table 2 shows comparison results of steering wheel
6.1 Slalom test angle in different systems regardless of the first and the last
Slalom tests were designed with a reference to Chinese cones.
National Standard for vehicle handling and stability test
procedure (GB/T 6323-1994[21]). The horizontal distance Table 2. Comparison of steering wheel angle
between two cones in the slalom experiment was set 12 m Steering wheel angle (°)
because of space and vehicle limitations. Velocity was kept No. Change rate (%)
Without control With control
around 35 km/h; tire-road friction coefficient was 0.85.
Cone 2 161 142 –11.80
Experiment results are shown in Fig. 9.
Cone 3 170 213 25.29
Thanks to the control system, the steering wheel angle
Cone 4 343 250 –27.11
was decreased by 93°, from 343° to 250° (absolute values),
Cone 5 215 205 –4.65
reducing handling burdens dramatically as shown in Fig.
Cone 6 298 248 –16.78
9(b). Comparing steering wheel angle in the process of
Cone 7 225 201 –10.67
counter clockwise rotation with control, the peak values
 CHINESE JOURNAL OF MECHANICAL ENGINEERING ·495·

Fig. 9. Experiment results of the slalom test

wheel angle results. The proposed controller reduced

6.2 Obstacle avoidance test steering wheel angles and eased handling burdens.
Obstacle avoidance test, which is a typical test for
vehicle close-loop maneuverability and stability, was Table 3. Comparison of steering wheel angle
designed with a reference to ISO 3888-2:2002[22]. Vehicle Steering wheel angle (°)
Position Change rate (%)
velocity was kept about 40 km/h to 50 km/h; tire-road Without control with control
friction coefficient was 0.85. The experiment results of the First peak 79 56 –29.11
obstacle avoidance test are shown in Fig. 10. Second peak 180 122 –32.22
Table 3 shows a detailed comparison of the steering Third peak 131 60 –54.20
 YU Zhuoping, et al: Direct Yaw Moment Control for Distributed Drive Electric Vehicle
·496· Handling Performance Improvement

Fig. 10. Experiment results of obstacle avoidance

Fig. 10(e) shows hysteresis curves of steer wheel angle based on direct yaw moment control for a distributed drive
and yaw rate with and without control. With the proposed electric vehicle equipped with four in-wheel motors was
control method, the response delay of yaw rate to steer presented under normal driving conditions.
wheel angle input decreased. And the relation of yaw rate The designed controller consists of a state feedback
and steer wheel angle tended to be linear. Namely, the controller and a steering wheel angle feedforward controller.
vehicle was close to neutral steer. The state feedback based control system, which is different
from the model following control widely used in previous
7 Conclusions research, can reduce modeling difficulty and regulate zeros
and poles of the system simultaneously. An observer based
In this paper, a handling improvement control system on extended Kalman filter and nonlinear two degree of
 CHINESE JOURNAL OF MECHANICAL ENGINEERING ·497·

freedom vehicle model was adopted to provide vehicle improving handling and stability[J]. JSAE Review, 2001, 22(4):
sideslip angle information to the controller. 473–480.
[12] SHINO M, NAGAI M. Independent wheel torque control of
The ITAE function was utilized as the objective function
small-scale electric vehicle for handling and stability
in the optimal control with the consideration of motor improvement[J]. JSAE Review, 2003, 24(4): 449–456.
capacity to calculate the ideal eigen frequency and damper [13] KIM D, KIM C, KIM S, et al. Development of adaptive direct
coefficient of the system. Optimal feedback matrix and yaw-moment control method for electric vehicle based on
feedforward matrix were obtained. identification of yaw-rate model[C]//2011 IEEE Intelligent Vehicles
Symposium (IV), Baden-Baden, Germany, June 5–9, 2011:
Simulations were carried out based on a Carsim and
1098–1103.
MATLAB/Simulink co-simulation platform to test the [14] MITSCHKE M, WALLENTOWITZ H. Dynamik der
performance of the control system. Simulation results Kraftfahrzeuge[M]. Berlin: Springer, 2003.
indicated that the yaw rate responded faster, the yaw rate [15] DORF R, BISHOP R. Modern control systems[M]. 11th ed. Beijing:
rise time was reduced by 40%. The steady-state yaw rate Pearson, 2011.
gain increased nearly 20% leading to an approximately [16] FENG Y, YU Z, XIONG L, et al. Torque vectoring control for
distributed drive electric vehicle based on state variable
neutral steer, which also reduced handling burdens.
feedback[C]//SAE 2014 World Congress and Exhibition, Detroit,
Finally, two typical closed-loop experiments, the slalom USA, April 8–10, 2014: SAE Paper 2014-01-0155.
test and the obstacle avoidance test, were carried out based [17] AWOUDA A, MAMAT R. New PID tuning rule using ITAE
on a high performance DDEV for evaluating vehicle criteria[J]. International Journal of Engineering, 2010, 3(6):
handling performance. In the slalom tests, the peak steering 597–608.
[18] GUO K, FU H, DING H. Estimation of CG sideslip angle based on
wheel angle was decreased by 93°, the change rate was
extended Kalman filter[J]. Automobile Technology, 2009(4): 1–3, 44.
more than 27%. Obstacle avoidance tests results showed (in Chinese)
that the vehicle with control was easier to handle, which [19] HIEMER M. Model based detection and reconstruction of road
not only reduced drivers burdens but was significant to traffic accidents[M]. Karlsruhe: Universitätsverlag Karlsruhe, 2004.
vehicle active safety. The experiments results validated the [20] GAO X, YU Z, NEUBECK J, et al. Sideslip angle estimation based
precision and the practicability of the designed control on input-output linearization with tire-road friction adaptation[J].
Vehicle System Dynamics, 2010, 48(2): 217–234.
system.
[21] GB/T 6323-1994. Controllability and stability test procedure for
automobile[S]. Beijing: Standardization Administration of the
References People's Republic of China, 1994. (in Chinese)
[1] SHIBAHATA Y, SHIMADA K, TOMARI T. Improvement of [22] ISO 3888-2:2002. Passenger cars-Test track for severe lane-change
vehicle maneuverability by direct yaw moment control[J]. Vehicle maneuver-Part 2: Obstacle avoidance[S]. London: British
System Dynamics, 1993, 22(516): 465–481. Standards Institution, 2003.
[2] YANG P, XIONG L, YU Z, et al. Motor/hydraulic systems
combined stability control strategy for distributed electric drive Biographical notes
vehicle[C]//AVEC’14 12th International Symposium on Advanced YU Zhuoping, born in 1960, is currently a professor at School of
Vehicle Control, Tokyo, Japan, September 22–26, 2014: 421–424. Automotive Studies, Tongji University, China. His research
[3] HORI Y. Future vehicle driven by electricity and control-research interests include vehicle dynamics and control, intelligent vehicle
on four-wheel-motored" UOT Electric March II"[J]. Industrial and parameter estimation.
Electronics, IEEE Transactions on, 2004, 51(5): 954–962. Tel: +86-21-69589119; E-mail: yuzhuoping@tongji.edu.cn
[4] YU Z, FENG Y, XIONG L. Review on vehicle dynamics control of
distributed drive electric vehicle[J]. Journal of Mechanical LENG Bo, born in 1991, is currently a PhD candidate at School of
Engineering, 2013, 49(8): 105–114. (in Chinese) Automotive Studies, Tongji University, China. He received his
[5] IKUSHIMA, SAWASE: A study on the effect of active yaw moment bachelor degree from Tongji University, China, in 2014. His
control[G]. SAE Paper 950303, 1995. research interests include vehicle dynamics and control.
[6] ONO E, HATTORI Y, MURAGISHI Y, et al. Vehicle dynamics Tel: +86-21-69589124; E-mail: harrisonleng@gmail.com
integrated control for four-wheel-distributed steering and
four-wheel-distributed traction/braking systems[J]. Vehicle System XIONG Lu, born in 1978, is currently an associate professor at
Dynamics, 2006, 44(2): 139–151. School of Automotive Studies, Tongji University, China. His
[7] SHINO M, NAGAI M. Independent wheel torque control of research interests include vehicle dynamics and control,
small-scale electric vehicle for handling and stability unmanned ground vehicle motion control and chassis system
improvement[J]. JSAE Review, 2003, 24(4): 449–456. design and development.
[8] KIM J, PARK C, HWANG S, et al. Control algorithm for an Tel: +86-21-69589124; E-mail: xiong_lu@tongji.edu.cn
independent motor-drive vehicle[J]. IEEE Transactions on
Vehicular Technology, 2010, 59(7): 3213–3222. FENG Yuan, born in 1987, received his doctor degree from
[9] XIONG L, YU Z, WANG Y, et al. Vehicle dynamics control of four Tongji University, China, in 2015. His research interests include
in-wheel motor drive electric vehicle using gain scheduling based vehicle dynamics and control, parameter estimation.
on tyre cornering stiffness estimation[J]. Vehicle System Dynamics, E-mail: ryan_fengyuan@163.com
2012, 50(6): 831–846.
[10] HE P, HORI Y. Optimum traction force distribution for stability SHI Fenmiao, born in 1988, received her master degree from
improvement of 4WD EV in critical driving condition[C]//9th IEEE Tongji University, China, in 2014. Her research interests include
International Workshop on Advanced Motion Control, Istanbul, vehicle dynamics and control.
Turkey, 2006: 596–601. E-mail: juliannesl@163.com
[11] SHINO M, NAGAI M. Yaw-moment control of electric vehicle for