You are on page 1of 9

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

Fault-tolerant Control for Electric Vehicles with


Independently Driven In-wheel-motors Considering
Individual Driver Steering Characteristics
Han Zhang, Wanzhong Zhao, and Junmin Wang, Senior Member, IEEE

Abstract: This paper presents a fault-tolerant control (FTC) in-wheel-motors (IWMs) have been considered as a
method for electric vehicles (EV) with independently driven promising mobility platform for its noticeable merits
in-wheel-motors (IWM) considering individual human driver comparing with traditional configurations [1]-[2]. The
steering characteristics. Both the human driver steering model application of IWMs can improve the vehicle operational
and the vehicle lateral dynamic model are formed for the FTC
strategy design while taking into account modeling
efficiency because energy consumptions on power
inaccuracies. Based on the driver-vehicle system model, a transmission mechanism such as gearbox, clutch,
dual-loop, sliding-mode controller is designed to deal with the transmission shaft, and other relative devices can be saved.
IWM fault and accomplish vehicle motion control in a Secondly, integrated chassis optimization and control can
customized way with explicit respect to individual driver be accomplished more easily, leading to a better
steering characteristics. The proposed control strategy comprehensive performance. IWMs can be controlled
contains two layers; the first layer calculates a reference through vehicle electronic control unit (ECU)
heading angle aiming to reduce the tracking error induced by independently by x-by-wire technologies, thereby more
the IWM actuator fault; the second layer computes the easily obtaining flexible responses under various
additional yaw moment needed to track the reference heading
angle and suppress the influence of the modeling errors. Such
maneuvering conditions [3]. However, the elimination of
an individualized FTC control method can specifically assist transmission mechanism and the increased number of
the human driver in post-fault vehicle motion control while actuators elevate the requirements on system reliability,
reducing the physical and mental workloads of the human safety, and fault tolerance. The fault of an IWM may not
driver. Simulation results of the proposed FTC method using only cause a loss of driving force, but also jeopardize
Matlab/Simulink-CarSim® demonstrate that it can accomplish vehicle driving stability, leading to a drift or even a rollover,
the expected post-fault vehicle motion control while providing which will affect the driving safety of itself as well as the
appropriate control assistance to different drivers in an vehicles around.
individualized and cooperative way. A substantial amount of research studies on fault-tolerant
Index terms—human-vehicle cooperative control; personalized
control (FTC) for EVs with independently driven IWMs
fault-tolerant control; electric vehicle. has been reported [4]-[6]. The FTC features several aspects,
such as fault detection, fault estimation, and post-fault
NOMENCLATURE treatment. For example, an IWM fault diagnosis approach
Yd Desired lateral displacement
Y Current lateral displacement
for four-wheel independently actuated EV was proposed in
τp Preview time [7], which can deal with tire force modeling inaccuracies
τd1 Pure time delay and absence of tire-road friction coefficient knowledge.
τd2 Delay time constant Based on the fault diagnosis result, a control-allocation
τl Derivative time constant
Gh Steering proportional gain
method was designed to accommodate the IWM fault by
vx Longitudinal velocity automatically allocating the control effort among other
 Yaw angle healthy wheels. In [9], an active fault diagnosis approach
vy Lateral velocity was put forward to explicitly evaluate the estimated control
Γ Yaw rate
Iz Vehicle yaw inertia
gain of the faulty wheel. Then a post-fault accommodation
Fxi Longitudinal force of ith wheel was accomplished by a way in which control efforts of all
Fyi Lateral force of ith wheel the wheels are reconfigured to relieve the torque demand
k1,k2 Cornering stiffness of front and rear tire onto the faulty wheel. Another control strategy involving
αf, αr Slip angle of front and rear tire
Iw Wheel moment of inertia
control allocation was presented in [8]. A closed-loop
Rw Tire rolling radius control architecture was formed as an FTC system, where
C Half of the track width. the vehicle velocity and information about an active fault
L Wheelbase were the states fed back to the controller for generating
A Distance from front axle to center of gravity
B Distance from rear axle to center of gravity
reference vehicle forces and yaw moment. Afterwards, the
Ti Torque of ith IWM reference vehicle forces and yaw moment were processed
according to the principle of control allocation in order to
track a desired trajectory. A hybrid FTC approach
I. INTRODUCTION
combining a linear-quadratic control method and a control

Electric vehicles (EV) with independently driven Lyapunov function technique also investigated the
path-tracking problem for four-wheel-steering and

H. Zhang’s work was partially supported by Chinese Scholar Council four-wheel-driving EVs [10]. By considering the input
(No. 201706830023), Outstanding Doctoral Dissertation in Nanjing constraints, actuator faults, and external disturbances, such
University of Aeronautics and Astronautics (BCXJ17-03), and
Postgraduate Research & Practice Innovation Program of Jiangsu Province. a control method can not only maintain vehicle’s tracking
Han Zhang (e-mail: zhanghanamo@126.com) and Wanzhong Zhao
(e-mail:zhaowanzhong@126.com ) are with the Department of Vehicle
Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing Walker Department of Mechanical Engineering at University of Texas at
210016, China; Junmin Wang (e-mail: JWang@austin.utexas.edu) is with Austin, Austin, Texas, 78712, USA.

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

performance but also reduce the computational burden of considered IWM faults. The individualized and
the fault-tolerant control algorithm. fault-tolerant control method for EVs with independently
The control objectives of all the abovementioned driven IWMs is designed in Section III. Simulation studies
research studies are mainly on EV motion itself. However, of such an individualized, fault-tolerant control method
little attention has been paid to human drivers’ influence on based on a CarSim® vehicle model are presented in Section
the post-fault maneuvers and FTC processes. When a motor IV, followed by some concluding remarks in Section V.
fault occurs, there will be a sudden yaw moment generated II. DRIVER-VEHICLE SYSTEM MODELING
to move the vehicle off its desired trajectory. Then the
driver will intuitively try to correct the trajectory The driver and vehicle models are established and then
displacement error by steering action. The driver’s reaction followed by the modeling of actuator faults.
may vary from person to person since it is affected by many 2.1 Driver model
individual factors, such as driving experience, gender, A driver’s steering action is a dynamic process that can
habits, etc. [11]-[14]. These individual differences have a be described by a driver steering model. Fig. 1 shows the
significant impact on the driving performance particularly schematic view of a preview driver model, where a human
under sudden faulty conditions [15]. For instance, a study driver scans through a future desired road path within a
on driver’s preview time was reported in [16] and the finite distance while driving. The vehicle lateral deviation
experimental results indicate that the driver with a long from the predict position to the preview point can be
preview time performs better under straight-line driving calculated in the Laplace domain as
Y  s   Yd  s  e p  Y  s   L  s  ,
 s
while achieving very poor lane maintenance performance (1)
on curved sections, especially on rough curves. Also, driver
where s is Laplacian, Yd  s  e
 ps
muscular physics was discussed in [17], where it was stands for the Laplace
modeled as a spring-damper system and simulated utilizing transform of the desired vehicle lateral displacement of the
a closed-loop path-following maneuver. The results preview point and τp is the preview time; Y  s  and   s
demonstrate that increment of co-contraction level of
muscles elevates the path-following accuracy and reduces are the Laplace transforms of the current lateral
the oscillations and overshoots of the steering wheel angle displacement and the heading angle of the vehicle,
during the maneuver. The well-known two-point visual respectively; L is the preview distance which can be
driver model was adopted by [18] to characterize the approximated as vxτp and vx denotes the longitudinal
steering behavior of the drivers, and a series of field tests velocity. A first-order driver model with lead-lag elements
were conducted to identify the model parameters and and a time delay is described in [20][21] as
validate the model in real-world scenarios. According to the G 1   L s  e d 1s
sw  s   h Y  s  , (2)
test results, an experienced driver is likely to pay more 1 d 2s
attention to both the anticipatory control and the where θsw(s) is the Laplace transform of the driver’s
compensatory control than a novice driver. Therefore, it is steering wheel angle; Gh is the steering proportional gain;
necessary to understand, characterize, and predict driver τL is the derivative time constant; τd1 is the pure delay time;
steering behavior so as to design a better and more τd2 is the delay time constant representing driver’s response.
proactive FTC controller that can seamlessly work with the
respective driver. Predict prosition
In order to improve the post-fault driving safety of EVs
with independently driven IWMs, an individualized L
fault-tolerant controller considering specific driver’s 
Yp  Y
steering characteristics is developed in this paper to Preview point
accomplish the cooperative motion control in a tailored way.
Preview distance L
An individualized driver-vehicle system with specific
driver parameters such as preview time, delay time, and Y
steering proportional gain is formulated as the control plant.
The driver’s characteristics can be identified during his/her X
daily driving by methods like the one proposed in [19].
Based on the individualized driver-vehicle system, a
dual-loop sliding-mode control is devised to deal with the Fig. 1 Schematic view of preview driver model.
IWM fault and accomplish post-fault vehicle motion Typically, τL and τd1 are significantly less than 1s, and
control in a customized way. The proposed control strategy they are lessly differentiated among different drivers than
contains two layers: the first layer is used to calculate a the other three parameters in Eq. (2). In order to facilitate
reference heading angle to reduce the tracking error caused the design of the controller later, the driver model can be
by actuator fault, and the second layer computes the approximated as
additional yaw moment to track the reference heading angle G
sw    sw  h Yd  s  e  Y   p vx  .
1  s
as well as suppress the influence of the modeling error. The
p
(3)
d2 d2
proposed control strategy can accomplish the following
control objectives: 1) when an IWM fault occurs, the To analyze the driver’s effort in following the trajectory
vehicle can still track the desired path and remain stable; 2) under an IWM fault condition, simulations are conducted
the driver’s workloads against the fault should be reduced using driver models with two different sets of parameters.
as much as possible with the individualized FTC control. The vehicle was set to accelerate from still. The IWM fault
The rest of this paper is organized as follows. Section II occurs when the vehicle longitudinal velocity reaches 15
gives a brief description of system modeling as well as the m/s. Simulation results are shown in Fig. 2.

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

0.05
driver No.1 independently driven IWMs is shown in Fig. 3.
driver No.2 Y

Steering wheel angle (rad)


0
b a
-0.05 Fyfl Fxfl
Fyrl
-0.1 Fxrl f

-0.15
 c
-0.2
0 5 10 15 20 25 30 35 40
t(s) X
CG
(a) Steering wheel angle. Fyfr Fxfr
0.3
Fyrr
driver No.1
c
Fxrr
0.25 driver No.2
Lateral displacement (m)

0.2

0.15 Fig. 3 Schematic diagram of a vehicle model.

0.1
Assuming sin  f  0, cos  f  1 because of the small
0.05 magnitude of front wheel steering angle, the equations of
longitudinal, lateral, and yaw motions of this model are
0
given as follows.
-0.05
 1
vx  v y   Fxfl  Fxfr  Fxrl  Fxrr 
0 5 10 15 20 25 30 35 40
t (s)
 m
(b) Lateral displacement.
 1
0.01
driver No.1 v y  vx   Fyfl  Fyfr  Fyrl  Fyrr  , (4)
 m
driver No.2

0.005  
I
1
a  Fyfl  Fyfr   b  Fyrl  Fyrr   c   Fxfl  Fxfr  Fxrl  Fxrr  

yaw rate (rad/s)

0
where vx, vy and γ are the longitudinal velocity, lateral
velocity, and yaw rate of the vehicle, respectively; m is the
vehicle total mass; Iz is the vehicle yaw inertia; Fxi is the
-0.005
longitudinal force of the ith wheel (i = f l, f r, rl, rr). Fyi (i
-0.01
0 5 10 15 20 25 30 35 40
= f l, f r, rl, rr) represents the lateral force of the ith wheel
t (s) that can be approximated as
(c)Yaw rate.
0.005 Fyfl  Fyfr  k1 f
driver No.1
 , (5)
0 driver No.2
Fyrl  Fyrr  k2r
Lateral velocity (m/s)

-0.005 where k1 and k2 are the cornering stiffness of the front and
-0.01 rear tires, respectively; αf, αr are the slip angles of front and
-0.015
rear tires, which are defined as
 a  v y
-0.02  f   f 
 vx
-0.025  . (6)
  y v  b
-0.03
 r vx
0 5 10 15 20
t (s)
25 30 35 40 
(d)Lateral velocity The wheel rotational dynamics equation is described as
Fig. 2 Simulaiton results with front-left IWM fault. 1 R
It can be seen that vehicle motions have been greatly  i  Ti  w Fxi i  fl , fr , rl , rr  , (7)
Iw Iw
impacted by the actuator fault that occurs to the front-left
IWM at the vehicle speed of 15m/s. A where Ti and ωi are the driving torque and the angular speed
proportional-integral-derivative (PID) control is applied to of the ith wheel; Iw and Rw are the wheel moment of inertia
maintain the target longitudinal velocity by controlling the and tire rolling radius, respectively.
total driving force. Vehicle lateral velocity, lateral Substituting (5)-(7) into (4), and taking δf =θsw /Gp, where
acceleration, and yaw rate are all affected substantially. Gp is the transmission ratio of the steering system, Eq. (4)
Vehicle also deviates from the target path, and can be derived as follows
consequently the driver has to steer to correct it.  1
vx  v y   T  d1
Furthermore, according to the results of steering wheel  mRw
angles, it can be seen that the two different drivers respond  2( k1  k2 )v y 2(ak1  bk2 )
 2k
v y  vx     1  sw , (8)
dissimilarly to the same fault condition. Thus, the  mv x mv x mG p
fault-tolerant control strategy should generate different 
  2(ak1  bk2 ) y  2(a k1  b k2 )   2ak1  sw  c T  d 2
v 2 2

control inputs for different drivers in order to achieve more  Iz vx Iz vx I z G p I z Rw


individualized driving assistance in post-fault maneuvers.
where T  T fl  T fr  Trl  Trr is the differential torque
2.2 Vehicle model
A schematic diagram of the vehicle model [22] with four between two side wheels as the yaw motion control signal,

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

and ΣT = Tfl + Tfr + Trl + Trr is the total driving torque for
longitudinal motion control and is allocated among the four
wheels based on the vertical load, which is shown as:
T . (9)
b b a a
T  T ; T  T ; T  T ; T 
2(a  b) 2(a  b) 2(a  b) 2(a  b)
fld frd rld rrd

Here, Tid (i = f l, f r, rl, rr) is the distributed driving torque


for the longitudinal motion control purpose. The yaw
moment control signal, T , is allocated equally among
other healthy wheels as the commanded torque outputs of
IWMs. For example, if there is a fault in the front-left IWM,
then the wheel torque commands will be
1 1 1
T fl  T fld ; T fr  T frd  T ; Trl  Trld  T ; Trr  Trrd  T .
4 2 4
In equation (8), d1 and d2 are considered as modeling
errors, which can be expressed as
Iw Fig. 4 Block diagram of the proposed control system.
d1   (1   2   3 + 4 ) , (10)
mRw 3. 1 Inner-loop Controller Design
I To realize e  0 , the inner-loop sliding surface is
d 2  w (1   2   3   4 ) . (11)
I z Rw defined as
Based on the previous study [9], the IWM and its power S1  e  K1e , (14)
electronic driver are treated as a unit. Then the motor driver
where K1  0 is the gain to be tuned later.
and IWM unit can be approximately described by a control
gain Km, which is defined as The Lyapunov function candidate of the inner-loop
Ti system is chosen as
K mi  i  fl , fr , rl , rr  , (12) 1
ui V1  S12 . (15)
where Ti is the output torque of the IWM; u i is the control 2
signal to the motor’s driver. The faulty condition of an The derivative of the V1 is V1  S1 S1 . More specifically,
IWM can be expressed as S    K        K  .
1 e 1 e c
(16) 1 e
ui  iudi  ui  i  fl, fr, rl, rr  , (13) The control law is selected as
where udi is the control signal; i   0,1 is the I z Rw 2(ak1  bk2 ) vy 2(a 2 k1  b2 k2 )  2ak1
T  [    sw
loss-of-effectiveness coefficient, which may be caused by c Iz vx Iz vx I z G p (17)
open-circuits in the motor systems; ui represents the  K      sgn  S  +K S ],
1 e e 2 1 2 1

disturbances, which may be caused by short-circuits in the where ρ2 > 0; K1 > 0.


motor systems [23][24] . By substituting the third equation of (8) and (17) into
III. FAULT-TOLERANT CONTROLLER DESIGN USING THE DUAL (16), it can be derived as
CLOSED-LOOP SLIDING MODE CONTROL S1   1 sgn  S1   d 2 . (18)
In order to reduce the influence of the IWM fault on the
And then
drivers, the comprehensive control objective of the
individualized FTC is to make the actual, post-fault, V1  1 sgn  S1  S1  d2 S1  K2 S12 . (19)
driver’s steering angle close to the expected driver’s By choosing 1  max d2 , it can be guaranteed that
steering angle under normal (no IWM fault) driving
conditions, and thus minimize the additional driver’s V1   1 S1  d 2 S1  K 2 S12   K 2 S12 , then V1  2K2V1 ,
steering effort in response to the IWM fault. The driver’s
V1  t0  .
-2 K2  t -t0 

normal steering angle signal can be calculated by his/her


that is V1 (t )  e For K 2  0 , V1
steering model recognized using the corresponding driver’s exponentially converges to 0, rendering e  0 .
daily driving data [19]. 3.2 Outer-loop Controller Design
Wind gust and road interference as well as parametric The integral sliding mode is used to realize the design of
uncertainties are not included in the vehicle model the outer-loop sliding surface, which is defined as
introduced in Section II. To maintain a good system t
S 2   swe  K 3   swe dt , (20)
robustness, a dual-loop sliding-mode control [25] is 0

employed for the personalized FCT system design whose where  swe   swd   sw and  swd is the reference steering
structure is shown in Fig. 4. wheel angle; K 3  0 is the integral gain. By selecting an
To track the expected driver’s normal steering wheel
appropriate gain K 3 , the tracking error of the system can be
angle, a virtual heading angle control signal c is introduced
kept sliding on the sliding surface until reaching zero as
for the outer-loop controller design. With c being the shown in the following analysis.
reference signal that should be tracked by the inner-loop. The derivation of equation (20) is
The error between the references heading angle and the
S2  swe  K 3 swe  swd  sw  K 3 swe . (21)
actual heading angle is e  c   . The design and
Substituting (3) into (21), and taking e  c   , it can be
analysis of this dual-loop control scheme are given in the
following sections. derived as

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

 1  delay, a shorter preview time, and a smaller steering



 d2
G
 d2
 

S2  swd     sw  h Y p  Y   p vx c  e    K 3 swe . (22) proportional gain.
Table I. Parameters of Driver Models.
The derivative of the Lyapunov function candidate for
the outer-loop system V2  (1/ 2)S2 is
2
L  d1 d2 τp Gh
Parameter (0.02~0.3) (0.1~0.34) (0.03~0.3) (0.4~2.5) (0.5~1.2)
V2  S2 S2 . (23) s s s s s

Select the virtual heading angle as Driver A 0.12s 0.05s 0.09s 0.85s 1.0

   Driver B 0.12s 0.05s 0.15s 0.75s 0.8

Gh p vx 
1
d2
G
d2
 p


c  d 2  sw   sw  h Y  Y  K1 sw  2 S2  , (24)
Table II. Parameters of vehicle.
where  2  0 . Substituting (24) into (16) gives Parameter Definition Value
Gh p vx m Vehicle total mass 1259.98 kg
V2   2 S22  e S2
d2 Iz Vehicle yaw moment of inertia 4607 kg-m2
2 2 . (25)
 Gh p vxe   Gh p vxe  a Distance from CG to front axle 1.14 m
=    2 S2    
 2 d 2 2   2 d 2 2  b Distance from CG to rear axle 1.64 m

Rw Tire rolling radius 0.36 m
From Eq. (25), it can be seen that when the bounded
inner-loop tracking error, e , converges to zero, k1 Cornering stiffness of the front tire -143583 N/rad

V    S 2 . Before the inner-loop controller renders  to


2 2 2 e
k2 Cornering stiffness of the rear tire -111200 N/rad

zero, V2 may initially increase with a bounded rate for a Gp Transmission ratio of steering system 17
short period of time while remaining bounded. After e Iw Wheel yaw moment of inertia 1.33 Nm2
quickly converges to zero by the inner-loop control effort, Km IWM control gain 80 Nm/V
V2 will be negative (and zero only when S2  0 ). Thus V2
will be always bounded and eventually converge to zero, Three functions are utilized to evaluate the performance
and so will  swe . and steering work loads of a driver [17]. The integral of a
quadratic function of the vehicle lateral deviation from the
Remark: In the dual-loop control system, the stability
desired path was used to evaluate the vehicle tracking
property of the inner-loop controller has a significant effect
control performance, denoted as ‘J1’. The integral of a
on the outer-loop stability property, consequently
quadratic function of the driver’s steering wheel angle is
influencing the convergence of the whole system [26] [27].
considered as a measure of the driver physical work load as
To achieve satisfactory control performance, it is required
‘J2’. The integral of a quadratic function of change rate of
that the inner-loop convergence speed should be much
the driver steering wheel angle is treated as a measure of
faster than the outer-loop convergence speed. Here, we
the driver’s mental work load as ‘J3’.
choose K1>>K3 and also a value large enough for 2 to
 J  t e 2 d
diminish e quickly, thereby guaranteeing a faster  1 0 y
inner-loop convergence speed and the establishment of  t
 J 2  0  sw d . (26)
2

V2  0 .  t
 J 3   sw2 d
The hyperbolic tangent function tanh(S1/σ) is used to  0

replace the sign switching function sgn(S1) in Eq. (17) to Two maneuvers, straight-line acceleration and
avoid the chattering effects in practical implementation. single-lane change, are conducted in simulations and the
Here, σ is the boundary layer thickness. results are shown in the subsequent sections.
IV. SIMULATION STUDIES 4.1 Straight-line Acceleration
In order to verify the effectiveness of the proposed The first group of simulation studies mainly focuses on
control strategy, simulations are conducted by using the single IWM fault of different levels in straight line
co-simulation platform of CarSim® and Matlab/Simulink. acceleration. Two different driver models, Driver A and
The model is established based on a high-fidelity Driver B, are employed in the simulation. The results are
full-vehicle model in CarSim® and all the powertrain shown as Fig. 5 and Fig. 6.
components are replaced by four IWMs. Two different The vehicle is set to accelerate from still and the fault is
drivers whose steering model parameters are identified in set to occur at the moment when the speed reaches 15m/s.
advance by adopting the method proposed in [19] are Firstly, the simulations are conducted without FTC and the
applied in the simulations. Controllers are designed for the loss-of-effectiveness coefficient λ is set as 25%, 50%, and
two different drivers respectively. The parameters of the 75%, respectively. The smaller the value of λ is, the more
drivers and vehicle are shown as Table I and Table II, yaw moment will be generated. Fig. 5 shows the steering
respectively. wheel angle results when the front-left IWM fault occurs.
The reasonable ranges of the parameters are marked in According to steering wheel angle amplitude changes, the
parentheses in Table I. Driver A is an experienced and drivers make more efforts to correct the vehicle motion
aggressive driver with a smaller time delay, a longer error due to the fault with the exacerbation of the IWM
preview time, and a larger steering proportional gain. driving torque loss.
Driver B is an inexperienced driver with a longer time According to Fig. 6, vehicle lateral displacement also

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

increases with the exacerbation of the IWM driving torque


loss. Moreover, according to the simulation results of
vehicle global trajectories, Driver A performs better in the
post-fault maneuver.
0.05
Steering wheel angle (rad)

-0.05

-0.1
ref erence
driv er A (λ=25%)
-0.15
driv er B (λ=25%)
driv er A (λ=50%)
-0.2 driv er B (λ=50%)
driv er A(λ=75%)
driv er B(λ=75%)
-0.25
0 5 10 15 20 25 30 35 40 (b)Yaw rate.
t (s)
Fig. 5 Simulation results of steering wheel angle.
0.35
ref erence

0.3 driv er A (λ=25%)


driv er B (λ=25%)
0.25 driv er A (λ=50%)
driv er B (λ=50%)
0.2 driv er A (λ=75%)
Global Y (m)

driv er B (λ=75%)
0.15

0.1

0.05

-0.05
0 100 200 300 400 500 600 700 800 900
Global X (m)
Fig. 6 Simulation results of vehicle global trajectory. (c)Vehicle global trajectory.
The second group of simulations were conducted with
the proposed FTC and the loss-of-effectiveness coefficient
λ is also set as 25%, 50%, and 75%, respectively. The
results are shown in Fig. 7.
The red curves represent simulation results of Driver A
and the blue ones represent simulation results of Driver B.
The solid curves indicate simulation results of the driver
model with its matching controller, that is Driver A with
controller A (CA) designed with Driver A’s parameters and
Driver B with controller B (CB) designed with Driver B’s
parameters, while the dash lines represents the mismatched
driver models and controllers. The reference lines represent
simulation results when there is no fault on the vehicle.
(d) Lateral velocity.
Fig. 7 Simulation results under proposed control.
From Fig. 7 (a) and Fig. 5, one can see that the steering
wheel angle amplitudes decrease by several orders of
magnitude under the proposed FTC. The vehicle states,
yaw rate and lateral velocity, are both influenced by the
fault as displayed in the Fig. 7 (b), Fig. 7 (c), Fig. 7 (d). But,
from the vertical axis of each figure, the oscillations of yaw
rate and lateral velocity are very small and gradually
convergence to 0.
Moreover, it can be seen that simulation results of Driver
A are closer to the reference signals than those of Driver B.
Tracking errors have been improved significantly and
differences between Driver A and Driver B with their
(a) Steering wheel angle. respective matching controllers have been reduced
according to vehicle global trajectory results in Fig. 7 (c).
Furthermore, simulation results of matched driver and
controller display a better reference tracking performance,
which means that by considering the specific driver

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

characteristics, the performance of the FTC strategy can be


improved.
To quantitatively illustrate the improvement of the
proposed control strategy more specifically, Eq. (26) is
used to evaluate the performance and the workloads of the
driver. The results are shown as Table III. The numbers in
the parentheses show the improvement of the matched
combinations from the unmatched ones.
Table III. The Quantitative Evaluation Results.
(a)  =25%
J1 J2 J3
driver A 0.5 0.2 0.1
driver A+CA 3.8e-05(153%↑) 1.1e-05 (164%↑) 1.2e-06 (75%↑)
driver A+CB 9.6e-05 2.9e-05 2.1e-06 (b) J2
driver B 0.2 0.2 0.1
driver B+CB 2.5e-05(248%↑) 2.0e-05(280%↑) 3.1e-06(106%↑)
driver B+CA 8.7e-05 7.6e-05 6.4e-06

(b)  =50%
J1 J2 J3
driver A 3.3e-2 5.7e-2 1.4e-2
driver A+CA 6.2e-06(335%↑) 9.2e-06(378%↑) 2.2e-06(159%↑)
driver A+CB 2.7e-05 4.4e-05 5.7e-06
driver B 6.3e-2 5.9e-2 1.6e-2
driver B+CB 8.3e-06(249%↑) 6.9e-06(320%↑) 1.1e-06(110%↑)
driver B+CA 2.9e-05 2.6e-05 2.3e-06

(c)  =75%
J1 J2 J3
(c) J3
driver A 7.1e-3 1.2e-3 2.9e-3
Fig. 8. The evaluations of drivers’ performances and workloads.
driver A+CA 1.2e-06(350%↑) 1.8e-06(383%↑) 4.7e-07(155%↑)
driver A+CB 5.4e-06 8.7e-06 1.2e-06 The tracking errors and drivers’ workloads increase when
driver B 1.3e-2 1.3e-3 3.3e-3 a fault occurs to an IWM, which implies that drivers need
driver B+CB 1.7e-06(247%↑) 1.4e-06(264%↑) 2.2e-07(109%↑) to make more effort to maintain the vehicle’s desired path.
driver B+CA 5.9e-06 5.1e-06 4.6e-07
After implementing the proposed individualized FTC
The evaluation results are obtained based on the strategy, the drivers’ performances are improved and the
aforementioned second group of simulations. It can be workloads are reduced. Also, according to the evaluation
concluded from the results (a) to (c) that with the proposed results, the matching groups outperform the unmatched
control strategy, the drivers’ task performance has been ones, indicating the necessity and benefits of the
improved and physical and mental workloads have been individualized FTC.
reduced. In addition, the designed controller considering In Fig. 9 and Fig. 10, the simulation results with no faults
the specific driver characteristics outperforms the are utilized as the references. Comparing the reference
mismatched one, which demonstrates the effectiveness and signals and the results without control in Fig. 9 and Fig. 10,
superiority of the proposed individualized FTC control when the fault occurs, the drivers need make more efforts
strategy. to maintain the stability of the vehicle and track the target
4.2 Single Lane-Change path by manipulating the steering wheel. According to the
The second case of simulation study is conducted with an simulation results of both drivers with their matching
IWM fault during a single-lane change to simulate the controller, the drivers’ outputs of steering wheel angle are
overtaking maneuver. The vehicle is set to accelerate from very close to the references with the application of the
90km/h to 125km/h during the simulation, and the proposed control strategy. This indicates that the proposed
front-right IWM fault with  =50% occurs when the lateral control strategy can generate a proper control input to the
displacement reaches 0.5m. The simulations last 40 system in a customized way, personally cooperating with
seconds and the evaluations of drivers’ performances and driver to accomplish vehicle FTC motion control.
workloads are also displayed by Fig. 8. The simulation Based on all the simulation results exhibited above, the
results are shown as Fig. 9 and Fig. 10. effectiveness of such an individualized FTC method
considering specific driver steering characteristics are
demonstrated.
ref erence
without control
0.1 driv er A+CA
Steering wheel angle (rad)

driv er A+CB

0.05

-0.05

-0.1

0 5 10 15 20 25 30 35 40
t (s)
(a) Steering wheel angles.
(a) J1

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

0.08 0.05
ref erence ref erence
without control without control
0.06 0.04
driv er A+CA driv er B+CB
driv er A+CB driv er B+CA
0.03
0.04
Lateral velocity (m/s) 0.02
0.02

Yaw rate (rad/s)


0.01
0
0

-0.02 -0.01

-0.04 -0.02

-0.03
-0.06
0 5 10 15 20 25 30 35 40
-0.04
t (s)
(b) Lateral velocities. -0.05
0 5 10 15 20 25 30 35 40
0.05 t (s)
ref erence

0.04 without control (c) Yaw rates.


driv er A+CA
driv er A+CB
0.03

0.02
Yaw rate (rad/s)

0.01

-0.01

-0.02

-0.03

-0.04

-0.05
0 5 10 15 20 25 30 35 40
t (s)
(c) Yaw rates.

(d) Global trajectories.


Fig. 10. Simulation results of driver B.
V. CONCLUSIONS

An FTC strategy for EVs with independently driven


IWMs considering individual driver steering characteristics
is developed in this paper. Driver steering model and
vehicle model are built and modeling inaccuracies are taken
into account in the driver-vehicle system and the controller
design. Based on the driver-vehicle system, a dual
(d) Global trajectories. closed-loop, sliding-mode control scheme is designed to
Fig. 9. Simulation results of Driver A. deal with the IWM fault and accomplish vehicle motion
0.15
ref erence
control in a customized way cooperating with individual
0.1
without control drivers. Simulations using Matlab/Simulink-CarSim® are
driv er B+CB
conducted and multiple groups of comparative results
Steering wheel angle (rad)

driv er B+CA

0.05 demonstrate that the proposed control strategy can


0
cooperate with individual drivers to accomplish vehicle
post-fault motion control by providing personally
-0.05 appropriate control assistance as well as by reducing his/her
-0.1 steering workloads.

0 5 10 15 20 25 30 35 40
REFERENCE
t (s) [1] G. Zhang, H. Zhang, X. Huang, J. Wang, H. Yu and R. Graaf,
(a) Steering wheel angles. "Active Fault-Tolerant Control for Electric Vehicles with
0.08
ref erence Independently Driven Rear In-Wheel Motors Against Certain
0.06
without control
driv er B+CB
Actuator Faults," in IEEE Transactions on Control Systems
driv er B+CA Technology, vol. 24, no. 5, pp. 1557-1572, Sept. 2016
Lateral velocity (m/s)

0.04
[2] L. Zhai, T. Sun and J. Wang, "Electronic Stability Control Based on
0.02 Motor Driving and Braking Torque Distribution for a Four In-Wheel
Motor Drive Electric Vehicle," in IEEE Transactions on Vehicular
0 Technology, vol. 65, no. 6, pp. 4726-4739, June 2016.
-0.02 [3] Y. Chen, X. Li, C. Wiet and J. Wang, "Energy Management and
Driving Strategy for In-Wheel Motor Electric Ground Vehicles With
-0.04 Terrain Profile Preview," in IEEE Transactions on Industrial
-0.06
Informatics, vol. 10, no. 3, pp. 1938-1947, Aug. 2014.
0 5 10 15 20
t (s)
25 30 35 40 [4] C. J. Ifedi, B. C. Mecrow, S. T. M. Brockway, G. S. Boast, G. J.
Atkinson and D. Kostic-Perovic, "Fault-Tolerant In-Wheel Motor
(b) Lateral velocities.
Topologies for High-Performance Electric Vehicles," in IEEE
Transactions on Industry Applications, vol. 49, no. 3, pp. 1249-1257,
May-June 2013.
[5] Y. Fan, W. Zhu, X. Zhang, M. Cheng and K. T. Chau, "Research on a
Single Phase-Loss Fault-Tolerant Control Strategy for a New
Flux-Modulated Permanent-Magnet Compact In-Wheel Motor," in
IEEE Transactions on Energy Conversion, vol. 31, no. 2, pp.

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2904698, IEEE
Transactions on Vehicular Technology

658-666, June 2016. Lyapunov stabilization of nonlinear cascade systems," in IEEE


[6] D. Wanner, M. KreuãŸLein, B. Augusto, et al, "Single Wheel Hub Transactions on Automatic Control, vol. 41, no. 12, pp. 1723-1735,
Motor Failures and Their Impact on Vehicle and Driver Behaviour," Dec 1996.
Vehicle System Dynamics, vol.54, no.10, pp: 1345-1361, 2016. Han Zhang received the B.S. in Vehicle
[7] R. Wang, J. Wang, "Fault-Tolerant Control for Electric Ground engineering from Nanjing University of
Vehicles With Independently-actuated In-wheel Motors," Journal of Aeronautics and Astronautics, Nanjing, China, in
Dynamic Systems Measurement & Control, vol.134, no.2, pp: 2014. She is currently a Ph.D. candidate with the
194-203, 2012. Department of Vehicle Engineering, Nanjing
[8] D. Wanner, M. Nybacka, O. Wallmark, et al, "Experimental University of Aeronautics and Astronautics,
Implementation of a Fault Handling Strategy for Electric Vehicles Nanjing, China. She was a visiting scholar
with Individual-wheel Drives," The Dynamics of Vehicles on Roads September 2017 to September 2018 with the
and Tracks, pp: 147-152, 2015. Department of Mechanical and Aerospace
[9] R. Wang and J. Wang, "Fault-Tolerant Control With Active Fault Engineering at the Ohio State University,
Diagnosis for Four-Wheel Independently Driven Electric Ground Columbus, OH, USA. Her research interests
Vehicles," IEEE Transactions on Vehicular Technology, vol. 60, no. 9, include vehicle system dynamics and control systems.
pp. 4276-4287, Nov. 2011.
[10] H. Yang, V. Cocquempot and B. Jiang, "Optimal Fault-Tolerant Wanzhong Zhao received the B.S. and M.S.
Path-Tracking Control for 4WS4WD Electric Vehicles," in IEEE degrees in vehicle engineering from Jiangsu
Transactions on Intelligent Transportation Systems, vol. 11, no. 1, pp. University, Zhenjiang, China, in 2004 and 2005,
237-243, March 2010. respectively, and Ph.D. degree in vehicle
[11] G. Li, S. Li, B. Cheng, "Field Operational Test of Advanced Driver engineering from Beijing Institute of Technology,
Assistance Systems in Typical Chinese Road Conditions: The Beijing, China, in 2009. He is currently a
Influence of Driver Gender, Age and Aggression," International professor and the director in the Department of
Journal of Automotive Technology, vol. 16, no. 5, pp: 739-750, 2015. Vehicle Engineering, Nanjing University of
[12] M. Flad, C. Trautmann, G. Diehm, et al, "Individual Driver Aeronautics and Astronautics, Nanjing, China. His
Modeling via Optimal Selection of Steering Primitives," IFAC research interests include vehicle system dynamics.
Proceedings Volumes, vol. 47, no. 3, pp: 6276-6282, 2014.
[13] L. Saleh, P. Chevrel, F. Mars, et al, "Human-Like Cybernetic Driver Junmin Wang (SM’14) received the B.E. degree
Model for Lane Keeping," IFAC Proceedings Volumes, vol. 44, no. 1, in automotive engineering and the M.S. degree in
pp: 4368-4373, 2011. power machinery and engineering from Tsinghua
[14] J. M. Choi, Shih-Yuan Liu and J. K. Hedrick, "Human driver model University, Beijing, China, in 1997 and 2000,
and sliding mode control - road tracking capability of the vehicle respectively; the M.S. degrees in electrical
model," 2015 European Control Conference (ECC), Linz, 2015, pp. engineering and mechanical engineering from
2132-2137. University of Minnesota Twin Cities, Minneapolis,
[15] G. Diehm, S. Maier, M. Flad and S. Hohmann, "Online Identification MN, USA, in 2003; and the Ph.D. degree in
of Individual Driver Steering Behaviour and Experimental Results," mechanical engineering from The University of
2013 IEEE International Conference on Systems, Man, and Texas at Austin, Austin, TX, USA, in 2007.
Cybernetics, Manchester, 2013, pp. 221-227. Dr. Junmin Wang is the Accenture Endowed Professor in Mechanical
[16] D. W. Xing, X. S. Li, X. l. Zheng, Y. Y. Ren and Y. Ishiwatari, "Study Engineering at University of Texas at Austin. In 2008, he started his
on driver's preview time based on field tests," 2017 4th International academic career at Ohio State University, where he founded the Vehicle
Conference on Transportation Information and Safety (ICTIS), Banff, Systems and Control Laboratory, was early promoted to Associate
AB, 2017, pp. 575-580. Professor in September 2013 and then very early promoted to Full
[17] P. Bolia, T. Weiskircher, S Müller, "Driver Steering Model for Professor in June 2016. He also gained five years of full-time industrial
Closed-loop Steering Function Analysis," Vehicle System Dynamics, research experience at Southwest Research Institute (San Antonio Texas)
vol. 52, no. sup1, pp: 16-30, 2014. from 2003 to 2008. Prof. Wang has a wide range of research interests
[18] C. You, J. Lu and P. Tsiotras, "Nonlinear Driver Parameter covering control, modeling, estimation, optimization, and diagnosis of
Estimation and Driver Steering Behavior Analysis for ADAS Using dynamical systems, especially for automotive, smart and sustainable
Field Test Data," in IEEE Transactions on Human-Machine Systems, mobility, human-machine, and cyber-physical system applications. Dr.
vol. 47, no. 5, pp. 686-699, Oct. 2017. Wang is the author or co-author of more than 295 peer-reviewed
[19] S. Schnelle, J. Wang, H. J. Su and R. Jagacinski, "A Personalizable publications including 145 journal articles and 13 U.S. patents. Prof.
Driver Steering Model Capable of Predicting Driver Behaviors in Wang is a recipient of numerous international and national honors and
Vehicle Collision Avoidance Maneuvers," in IEEE Transactions on awards. He is an IEEE Vehicular Technology Society Distinguished
Human-Machine Systems, vol. 47, no. 5, pp. 625-635, Oct. 2017. Lecturer, SAE Fellow, and ASME Fellow.
[20] J. Wang, G. Zhang, R. Wang, S. C. Schnelle and J. Wang, "A
Gain-Scheduling Driver Assistance Trajectory-Following Algorithm
Considering Different Driver Steering Characteristics," in IEEE
Transactions on Intelligent Transportation Systems, vol. 18, no. 5, pp.
1097-1108, May 2017.
[21] M. Plöchl, J. Edelmann, "Driver Models in Automobile Dynamics
Application," Vehicle System Dynamics, vol. 45, no. 7-8, pp:
699-741, 2007.
[22] F. Wang, H. Chen and D. Cao, "Nonlinear Coordinated Motion
Control of Road Vehicles After a Tire Blowout," in IEEE
Transactions on Control Systems Technology, vol. 24, no. 3, pp.
956-970, May 2016.
[23] H. Noura, D. Theilliol, J.-C. Ponsart, and A. Chamseddine,
Fault-Tolerant Control Systems: Design and Practical Applications.
Berlin,Germany: Springer-Verlag, 2009.
[24] C. J. Ifedi, B. C. Mecrow, S. T. M. Brockway, G. S. Boast, G. J.
Atkinson and D. Kostic-Perovic, "Fault-Tolerant In-Wheel Motor
Topologies for High-Performance Electric Vehicles," in IEEE
Transactions on Industry Applications, vol. 49, no. 3, pp. 1249-1257,
May-June 2013.
[25] Liu Jinkun. Sliding Mode Control Design and Matlab Simulation.
BeiJing˖ Tsinghai University Press, 2005. (in chinese)
[26] S. Bertrand, N. Guénard, T. Hamel, et al, "Hierarchical Controller for
Miniature VTOL UAVs: Design and Stability Analysis Using
Singular Perturbation Theory," Control Engneering Practice, vol. 19,
no. 10, pp: 1099-1108, 2011.
[27] M. Jankovic, R. Sepulchre and P. V. Kokotovic, "Constructive

0018-9545 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like