Int. J. Vehicle Autonomous Systems, Vol. 15, No.

1, 2020 1

A modified extreme seeking-based adaptive fuzzy

sliding mode control scheme for vehicle anti-lock
braking

Wenfei Li* and Haiping Du

School of Electrical, Computer & Telecommunications Engineering,
University of Wollongong,
Wollongong 2522, New South Wales, Australia
Email: wl015@uowmail.edu.au
Email: hdu@uow.edu.au
*Corresponding author

Weihua Li
School of Mechanical, Materials and Mechatronic Engineering,
University of Wollongong,
Wollongong 2522, New South Wales, Australia
Email: weihuali@uow.edu.au

Abstract: Vehicle Anti-lock Braking Systems (ABS) are designed to optimise

vehicle braking performance but there are some challenges for their control.
One of the challenges is that the optimal slip ratio is difficult to obtain in real
time and it can differ under different road conditions. Another challenge is that
ABS are non-linear and many of their parameters are difficult to identify in
advance. To solve these problems, a new extreme seeking-based adaptive fuzzy
sliding mode control strategy is proposed. The proposed Modified Sliding
Mode-based Extreme Seeking (MSMES) algorithm is able to avoid the large
oscillation and automatically search for the optimal slip ratio. An Adaptive
Fuzzy Sliding Mode Controller (AFSMC) is designed to mimic an ideal
controller. The simulation results show that the integrated control scheme
MSMES-AFSMC is able not only to search for the optimal slip ratio under
different road conditions but also has a better braking performance than the
conventional COSR for ABS.

Keywords: anti-lock braking system; extreme seeking; adaptive fuzzy sliding

mode control; adaptive non-linear observer.

Reference to this paper should be made as follows: Li, W., Du, H. and Li, W.
(2020) ‘A modified extreme seeking-based adaptive fuzzy sliding mode control
scheme for vehicle anti-lock braking’, Int. J. Vehicle Autonomous Systems,
Vol. 15, No. 1, pp.1–25.

Biographical notes: Wenfei Li received the BE and MS degrees in Mechanical

Engineering and Automation from the China University of Mining and
Technology, Beijing, in 2011. He is currently working toward his PhD on
brake-by-wire vehicle control at the School of Electrical, Computer and
Telecommunications Engineering at the University of Wollongong, Australia.

Haiping Du received PhD degree in Mechanical Design and Theory from the
Shanghai Jiao Tong University, Shanghai, PR China, in 2002. He was Research

Copyright © 2020 Inderscience Enterprises Ltd.

2 W. Li, H. Du and W. Li

Fellow at the University of Technology, Sydney from 2005–2009, and Post-

Doctoral Research Associate at the Imperial College London (2003–2005) and
the University of Hong Kong (2002–2003). He is currently a Professor of
School of Electrical, Computer and Telecommunications Engineering at the
University of Wollongong. He is an Editorial Advisory Board Member of the
Journal of Sound and Vibration and Associate Editor of IEEE Control System
Society Conference.

Weihua Li received his BE and ME degrees from University of Science and

Technology of China in 1992 and 1995, respectively; and his PhD degree from
Nanyang Technology University, Singapore in 2001. He was with the School
of Mechanical and Production Engineering of Nanyang Technological
University as a research fellow from 2001 to 2003. He was been with the
School of Mechanical Material and Mechatronic Engineering, University
of Wollongong, as Lecturer (2003–2005), Senior Lecturer (2006–2009),
Associate Professor (2010–2012) and Professor (2012) He serves as Associate
Editor or editorial board member for nine international journals.

1 Introduction

Anti-lock Braking Systems (ABS) play an important role in improving vehicle braking
performance during emergency braking. Research shows that the friction coefficient
between road and tyre is related to the wheel slip ratio. The wheel slip ratio is defined as
the normalised difference between the vehicle longitudinal speed and the wheel linear
speed (Nahid and Aliakbar, 2014). The maximum braking force can be obtained when
the wheel slip ratio is the optimal slip ratio. However, the maximum friction coefficient
is different in different road conditions. The optimal slip ratio is also different in different
road conditions. However, the optimal slip ratio is unknown in advance. In order to
obtain the maximum friction coefficient and shorten the braking distance, Extreme
Seeking Algorithm (ESA) is usually used to search for the optimal slip ratio. ABS is used
to make wheel slip ratio track the optimal slip ratio.
Some researchers (Miroslav and Hsin-Hsiung, 2000) and David et al. (2002) treated
the optimal slip ratio value as about 0.2. However, the optimal slip ratio differs under
different road conditions (Haiping et al., 2015; Erkin and Tunc, 2015; Li et al., 2016;
Ehsan et al., 2017; Zhao et al., 2017). Treating the optimal slip ratio as the constant slip
ratio may reduce braking performance. This means that it is necessary to seek the optimal
slip ratio dynamically. In Wang et al. (2014), a three-point prediction method is used to
estimate the road surface. The optimal slip ratio is then obtained based on the estimated
road conditions. This method is a model-based method and its accuracy depends on a
tyre-road friction model. This means that it is not suitable for real application. Recently,
an ESA (Erkin et al., 2012) has been proposed to obtain the optimal slip ratio. It is able to
obtain the optimal slip ratio without the necessity of knowing information about the road
conditions. In Mojtaba et al. (2011), a perturbation based optimal slip seeking algorithm
is proposed. It uses sinusoidal as the excitation signal to get a measure of the gradient
information of     (friction coefficient – slip ratio). In Zhao et al. (2010), a sliding
mode based optimal slip seeking is proposed. However, ESA can cause oscillation near
the optimal slip ratio which it makes difficult for the actuator to track the extreme value
and the oscillation may shorten the actuator life.
 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 3

During emergency braking, it is important to make the wheel slip ratio track the
optimal slip ratio. In order to achieve this, many control algorithms have been proposed.
The most conventional control methods, such as Proportional-Integral-Derivative (PID)
control, have been applied. But the vehicle dynamic is nonlinear, the friction coefficient
between the tyre and the road is a non-linear function and there are many other uncertain
and time-varying parameters (Khatun et al., 2003). This means that it is difficult for
conventional control methods to solve the above challenges. Because intelligent
controllers can identify and adjust model parameters online, it has better performance
than the conventional control in terms of dealing with non-linear and uncertain factors
(Lin and Hsu, 2003). Many researchers have proposed different intelligent control
methods in order to solve these problems. Such solutions include, for example, sliding
mode control (Lee and Stanislaw, 2002; Chianga et al., 2014; Wang et al., 2016a, 2016b;
Hu et al., 2016), adaptive control (Zhao et al., 2017), neural control (Amir, 2009) and
fuzzy control (Ye et al., 2015). Fuzzy control is suitable for solving the problems of non-
linear control (Ye et al., 2015) but a number of problems still need to be solved.
A number of different structures have been proposed and many algorithms have been
applied to fine tune the parameters of the fuzzy controller. For example in Avesta et al.
(2008), a genetic algorithm is applied to automatically fine tune fuzzy membership
functions. In Lin and Hsu (2003), a Self-Learning Fuzzy Sliding-Mode Controller (SLF-
SMC) is used to make wheel slip ratio track the optimal slip ratio. This does not require
an accurate vehicle braking system model but the coefficient of the sliding surface is
chosen as a fixed number. It is hard to achieve a low overshoot and quick response at the
same time. In Selim et al. (2013), a non-linear sliding surface is proposed to achieve
quick response and avoid overshoot but the controller design is based on the assumption
that all the parameters of the system are known.
To overcome the above mentioned problems for both optimal slip ratio seeking and
tracking, a new control scheme is proposed for ABS in this paper. In this new control
scheme, a Modified Sliding Mode-based Extreme Seeking (MSMES) algorithm is
proposed to seek the optimal slip ratio. And an Adaptive Fuzzy Sliding Mode Controller
(AFSMC) is used to make wheel slip ratio track the obtained optimal slip ratio. Based on
the conventional ESA (Nahid and Aliakbar, 2014), a switching condition is added in
MSMES. The switching condition is used to stop seeking process for avoiding oscillation
which is caused by the conventional ESA and to identify changes in the road conditions.
The AFSMC is based on nonlinear sliding surface and the SLF-SMC. The AFSMC is
able to achieve a low overshoot and quick response while the parameters of the system
can be estimated. The effectiveness of the proposed control system is validated by a
number of simulations under various conditions.
The main contributions of this paper are: (1) based on the seeking characteristic of
ESA, an MSMES algorithm is proposed to avoid the oscillation caused by ESA. (2) by
combining the SLF-SMC and non-linear sliding surface, an AFSMC is proposed, which
can automatically adjust the coefficient of the sliding surface in order to achieve low
overshoot and quick response and has more robustness against parameter variation and
uncertain events. The rest of the paper is organised as follows. In Section 2, the vehicle
braking dynamics model is introduced. In Section 3, the innovative control scheme for
the ABS is introduced. It includes the derivation process of MSMES and AFSMC. The
simulation results on a quarter-vehicle braking model are presented in Section 4. Finally,
conclusions are given in Section 5.
4 W. Li, H. Du and W. Li

2 System model

2.1 Vehicle dynamics model

Figure 1 Wheel dynamic model

A quarter-vehicle braking model, which is simple but effective (see Figure 1) is used in
Erkin et al. (2012). The dynamics of the vehicle and wheel models can be expressed as:
J   Tb  Tt (1)

mv   Fx (2)

Tt  RFx (3)

Fx      mg (4)

v  Rv (5)

v  
 (6)
v

where R is the effective wheel radius, J is the moment of inertia of the wheel, Tb is the
braking torque, Fx is the longitudinal tyre force, m is total mass of the quarter vehicle,
v is the longitudinal velocity of the vehicle,  is the angular velocity of the wheel, v
is the angular velocity of the vehicle.     is the longitudinal friction coefficient of the
tyre,  is the wheel slip ratio. Tt is the torque provided by ground friction. In this paper,
the tyre model used is based on the tyre friction model, which is shown below:

     C1 1  e  c    C3
2
(7)

where C1 , C2 and C3 are constants for different road conditions. C1 is the maximum
value of the friction curve, C2 represents the shape of the friction curve and C3 is the
difference between the peak value of the friction curve and the value when the slip ratio
is 1. A general tyre characteristic of the longitudinal coefficient of friction versus the
 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 5

wheel slip ratio is shown in Figure 2. From Figure 2, it can be seen that there is an
optimal slip ratio on any given road surface. When wheel slip ratio is the optimal slip
ratio, the maximum coefficient of friction is obtained.

Figure 2 Longitudinal coefficient of friction versus wheel slip ratio

Taking the time derivative of the wheel slip ratio (6), it is obtained as:

1     v  
  (8)
v
The equation (8) can be further rewritten as:

  Fp     G p u (9)

where

    g       gJ  R 2 mg 
Fp        (10)
 Rv  v RJ

1
Gp  u  Tb (11)
J v

The nominal model of (9) can be represented as:

  Fn     Gn u (12)

where Fn    and Gn are the nominal values of Fp    and G p . If the uncertainties

occur, then (12) can be modified as:

   Fn     ΔFn      Gn  ΔGn  u  Fn     Gn u  w (13)

where
w  ΔFn     ΔGn u (14)
6 W. Li, H. Du and W. Li

with the assumption:

w W (15)

where W is a positive constant.

In this paper, the braking system is a blended system. It combines a Hydraulic
Braking System (HBS) and a Regenerative Braking System (RBS). In the following
section, the model of the HBS is constructed. Then the RBS is described.

2.2 Hydraulic braking system model

In this paper, the HBS is described by the following equation (Wang et al., 2015):
k1Cd Avi k1 kC A k
 pm  pw  i  2 d vd 2  pw  p0  d
 
p w  (16)
V0 V0

where k1 , k2 represent the control signals to simulate the on–off feature of the pressure
modulation for the wheel cylinder, where the states of the HBS can be classified into
three modes: increasing, decreasing and holding. pw , pm , p0 are wheel cylinder
pressure, master cylinder pressure, and reservoir pressure, respectively. Cd , Avi, Avd, k,
and V0 are the flow coefficient of the valves, the cross-sectional area of the inlet valves,
the cross-sectional area of the outlet valves, the bulk modulus of the brake oil, and the
brake oil volume under no pressure, respectively. i and  d are the coefficients for
pressure increasing and decreasing modes, respectively, and they can be calibrated from
experimental data (Wang et al., 2015) as follows:
 37.5416  p  p 0.59 Increasing
 m w

p w   0 Holding (17)

 38.3128  pw  p0 
0.93
Decreasing

The HB torque can then be calculated using the following equation:

Thb  2 pw Aw f hb rhb (18)

where Thb is the hydraulic brake torque, Aw is wheel cylinder cross sectional area, f hb is
the friction coefficient, rhb is the effective brake radius.

2.3 Regenerative braking system

In this paper, the transfer function between the regenerative reference torque and the
executed torque can be described as a first order delay system with dead time as follows
(Wang et al., 2015):
Trb e 1s
Gr   (19)
Trb*  2 s  1
 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 7

where  1 represents the dead time of CAN communication or mechanical response lag;
 2 is the response time of the motor with first order lag; Trb* is the RB system reference
torque; Trb is the real output torque of RB system.
The total braking torque  Tc  is provided by the HBS and the RBS. It is shown as
following equation:
Tb  Trb  Thb (20)

3 A new control scheme for ABS

The whole structure of the new vehicle braking control system is shown in Figure 3. In
this paper, we assume that the angular velocity of the wheel and the vehicle acceleration
can be measured. The role of each part will be introduced below.

Figure 3 Whole control system structure

It is known that there is only one maximum friction value under each road condition (as
shown in Figure 2). The MSMES is used to search for the optimal slip ratio. The input of
the MSMES is the friction coefficient  . The output of the MSMES is the optimal slip
ratio. The MSMES is able to search for the optimal slip ratio for any given road condition
(In most case, the car is driving on one road condition. Occasionally, there will be a pool
of water on the road. Then cars move from one road condition to another road condition.
Such as from dry asphalt to wet asphalt. The given road condition can be a single road
condition or a changing road condition). Even if the road condition changes, the MSMES
is also able to search for the optimal slip ratio quickly. This optimal slip ratio d will be
treated as a reference input or target signal for the AFSMC. Then, the AFSMC is
designed to make the wheel slip ratio track the optimal slip ratio. The input of the
AFSMC is the difference between the optimal slip ratio and the real slip ratio. The output
of the AFSMC is the required braking torque. The required braking torque is then
allocated to the HBS and the RBS by using the defined allocator. Finally, the braking
torques generated from the HBS and the RBS will be applied to stop the vehicle.
8 W. Li, H. Du and W. Li

3.1 Modified sliding mode based extremum seeking

For a given road condition, there exists a maximum  * for the tyre-road friction
coefficient     at  * . And  * will change with the change of road conditions. In
order to search for the maximum friction coefficient, an MSMES algorithm is proposed.
The input of the MSMES is the friction coefficient     . Although     is not
available to be directly measured, the information about     can be obtained from the
vehicle linear acceleration (Wang et al., 2015). In this paper, we assume that the vehicle
linear acceleration is measurable via an accelerometer.
Before introducing the MSMES algorithm, the classic sliding mode based ESA
scheme is given as shown in Figure 4 (Lin and Hsu, 2003).

Figure 4 The classic ESA scheme

In Figure 4, e is defined as
e    g    t (21)
where g is an increasing function satisfying g    0 . g is designed to search for the
extremum value. It is an important part of ESA. Its role can be seen from the following
derivation process. The derivative of e can be written as:
d 
e   (22)
d
For optimal slip ratio seeking,  is designed as:

   e 
  ksign  sin   (23)
  Υ 
where k is a positive constant and sign is the sign function, Y is a positive constant. Then
as long as the condition (24) holds:
d 
 (24)
d k
 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 9

The maximum friction coefficient can be found without a priori knowledge of the road
friction.
Proof: Submitting (23) into (22), one can obtain:
d    e 
e  ksign  sin      (25)
d   Y 
Now, assuming that at the start of seeking the value of e in (21) is between the values of
Y and 2Y such as Y  e  0   2Y , then

   e     e 
sign sin      sign  e  Y  or sign sin     sign  e  2Y  (26)
  Y    Y 
The sliding surface variable s is selected as:
s  e Y (27)
d d
s   ksign  e  Y    or s  ksign  e  2Y    (28)
d d
d 
when  , then s  0 . e will tend to be Y Same to above derivation process, when
d k
d 
0  e  0   Y and  , then s  0 . e will also tend to be Y. Based on the above
d k
d 
analysis, when 0  e  0   2Y and  , as shown in Figure 5, after finite time
d k
interval:
e Y (29)
     t  Y (30)

d 
Same to the above derivation process, when   and r  e  0   3Y , as shown in
d k
Figure 6, after finite time interval:
e  2Y (31)
      t  2Y (32)

Based on the above analysis, for any other initial condition of e  0  , as long as
condition (24) holds, after finite time interval, e will be a constant number.
e  kY (33)
      t  kY (34)

where k  1, 2, 3, 
then
e        (35)
10 W. Li, H. Du and W. Li

d    
Figure 5 Change of e and e if 
d k

d    
Figure 6 Change of e and e if 
d k

Figure 7 Friction coefficient-slip ratio

 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 11

d 
As shown in the above proof, as long as the absolute value of is larger than ,
d k
   will approach peak point in magnitude with the converging slope of  equation
(35). As shown in Figure 7, P1 and P2 characterise the interval points where their
 d 
gradients are smaller than . Below these points,  , the extremum seeking
k d k
algorithm will make slip ratio towards the optimal slip ratio, until the operation hits
P1 or P2 .
For example, when the operation hits P1 , since the gradient is getting smaller and
d 
 , it means e is not a constant anymore and its value decreases with the speed of
d k

d    e 
e  ksgn  sin       0 (36)
d   Υ 

According to (23),  oscillates between the values of k and  k . In this case, we take
d 
0  as an example to explain the process. As shown in Figure 8, e changes its
d k
value when e changes Y. However, the descending speed of e is different at different
intervals. As shown in Figure 8, the length of interval I is equal to the length of interval
II. Because the descending speed of e at interval II is larger than the descending speed of
e at interval I e has a longer duration time at interval I than at interval II. As can be seen
in Figure 9,  takes more negative value k than positive value k . So  will decrease
d 
back to point P1 . When  is at the left side of point P1 ,  , then,  will increase
d k
and toward to point P1 . Based on above analysis, λ will oscillate around the point P1
when the initial value of  is at the left side of point P1 . As shown in above analysis, the
search process of the ESA leads to oscillations.

d 
Figure 8 Change of e and e when 0  
d k
12 W. Li, H. Du and W. Li

d 
Figure 9 Change of  when 0  
d k

Similar to the condition of the initial value of  is at the left side of point P1 ,  will
oscillate around the point P2 when the initial value of  is at the right side of point P2
(the detailed derivation process is same to the derivation process of P1 and is ignored
here). When the initial value of  is between P1 and P2 ,  will oscillate around the point
P1 or P2 according to different initial values of . However, the conventional ESA will
cause oscillation around the optimal slip ratio according the above analysis. In order to
solve above problem, we proposed MSMES algorithm. The MSMES scheme is shown in
Figure 10. By comparing Figure 4 and Figure 10, it can be seen that the MSMES adds the
switching condition on the basis of ESA. Switching condition has two functions. One is
to identify the condition to stop seeking for eliminating oscillation. Another one is to
identify the change of road condition.

Figure 10 MSMES scheme

 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 13

As shown in equation (25), e is a variable. In this paper, we will take e as an indicator

to decide whether to stop seeking or not. Based on equation (25), we can obtain

d    e 
e    ksign sin    (37)
d   Y 

d 
From equation (24), we can get that when  ,     will approach to the peak
d k
d
point. In other words, when k   ( k is positive), it means that the seeking process
d
ends and     has reached its extreme value. Based on the above analysis and equation
(36), the amplitude of e   is less than 1 when     reached its extreme value.
During the process of approaching to the extremum value, we can get that e is  from
equation (35). In order to eliminate interference, we proposed another indicator. When
    reached its extreme value,     will oscillate around the extreme value. For
example, as shown in Figure 7, it will oscillate around point P1 . At the left side of point
d d
P1 , k  1. At the right side of point P1 (not too far away from P1 ), 0  k  1.
d d
According to equation (36), 1  e  0. When e changes more than Y, e changes from
d d d d 
d
k   to 
d
k   or from 
d
k   to
d
k .  e dt  Y , t  e
 0.1s

(in this paper, we make Y  0.1 and k  1 ). It means that the time for the amplitude of
 d 
e to be less than 1   k    is longer than 0.1 s when slip ratio is at the right side
 d  
of point P1 . Based on the above analysis, the time for the amplitude of e to be less
than 1 can be used as an indicator to decide whether the slip ratio is at the optimal slip
ratio or not.
Based on the above analysis, we take two indicators to determine whether    
reached the extreme value or not:
1 The amplitude of e   is less than 1.

2 The time for the amplitude of e to be less than 1 is longer than 0.1 s.

When the above two indicators are met simultaneously, the corresponding slip ratio is
taken as the optimal slip ratio (peak point) and then stop seeking. Because the seeking
d
process stops, it avoids oscillation which caused by ESA. Because is different at the
d
same slip ratio when the road condition is different, e will have a big change when the
14 W. Li, H. Du and W. Li

road condition changes. In our design, we make e bigger than 10 as the road condition
change signal. When the road condition change signal is got, a new extremum seeking
process is activated.
3 When e  10 , restart seeking process.

Based on the above analysis, we make k  0 (it means that seeking stops) when
indicator 1 and indicator 2 are met simultaneously. We make k  1 (it means that seeking
starts) when e  10 .

3.2 Adaptive fuzzy sliding mode controller

A low overshoot can be achieved at the cost of high settling time. However, a low-
settling time is also necessary for a quick response. Thus, most of the design schemes
make a trade-off between these two transient performance indices and it is hard to
achieve a low overshoot and quick response at the same time when coefficient of sliding
surface is a fixed number. In this paper, a non-linear sliding surface is designed to
achieve quick response and avoid overshoot. The coefficient of sliding surface changes
with the changing of difference between the optimal slip ratio and real slip ratio.
As shown in Section 2, the vehicle braking system model is a non-linear system and
some parameters are unknown or uncertain. This means that it is hard to design a
controller based on this model. Based on the above analysis, this controller should be
able to adaptively adjust its parameters according to the relevant environment. In this
paper, an Adaptive Fuzzy Sliding Mode Controller (AFSMC) is designed to make the
wheel slip ratio track the optimal slip ratio. The structure of the AFSMC is shown in
Figure 11.

Figure 11 Structure of the AFSMC

 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 15

The design of the AFSMC is based on the assumption that the wheel speed and vehicle
velocity are measurable. The control objective is to make the wheel slip ratio well track
the optimal slip ratio d which is obtained by the MSMES algorithm. The tracking error
is defined as:
e  d   (38)

e  d   (39)

where  is the real slip ratio.
A sliding surface is designed as:
t
s  e    d ,   e   d (40)
0


Ψ  d ,    (41)
1  e   d
The nonlinear function Ψ  d ,   is used to change convergence rate of the closed loop
system.  ,  are positive numbers.
The sliding mode control law is defined as:
us  ueq  ubt (42)
where ueq is the equivalent controller and is designed as

ueq  Gn1   Fn     d  Ψ  d ,   e  (43)

And ubt is designed to compensate uncertainty as:

ubt  Gn1 Wsgn  s  (44)

then

s  e  Ψ  d ,   e   w  Wsgn  s  (45)

Then a Lyapunov function is chosen as:

1 2
V1  s (46)
2
   sw  s W  s w  s W   s W  w   0
V1  ss (47)

To satisfy the existence condition, a large uncertainty bound W should be chosen. This
will cause big chattering. This chattering will affect the control performance. In order to
solve this problem, an observer is normally needed to estimate this uncertainty bound.
From the above analysis, we can see that system models must be accurately known in
order to design the SMC for ABS. However, accurate mathematical models can be
difficult to obtain. To solve these problems, a fuzzy controller is proposed to control
the ABS.
16 W. Li, H. Du and W. Li

Assume that there are five rules in a fuzzy rule base and each of them has the
following form:
Rule i : if s is si then u is i
where s is the input variable of the fuzzy system; u is the output variable of the fuzzy
system. si are the membership functions; In this paper, five membership functions are
adopted. They are s1  s  exp   s  / 6  / 24   s2  s   exp    s   /12  / 24  
2 2

   
s3  s   exp   s  / 24  
  s4  s   exp    s   / 12   / 24  
  s5  s   exp
2 2

   
    s   / 6   / 24  2  and  are the singleton control actions. The defuzzification
  i

of the AFSMC output is accomplished by the method of centre-of-gravity (Lin and

Hsu, 2003)
5 5
u  i  i  i (48)
i 1 i 1

where i is the weight of i-th rule. And equation (48) can be rewritten as

u  T (49)

where   1 ,2 , ,5  is a parameter vector and   1 , 2 , , 5  is a regressive

T T

vector with i defined as

5
i  i 
i 1
i (50)

an ideal controller can be written as

u*  G p1   Fp     d  Ψ  d ,   e  (51)

Submitting (9) (38) (49) into (43)

e   Ψ  d ,   e (52)

From equation (41), we can see that Ψ  d ,   is a positive number and then
lim e  t   0 . As shown in equation (41), Ψ  d ,   changes with the changing of e .
n 

When the tracking error e is small, Ψ  d ,   is a small value. When the tracking error
e is big, Ψ  d ,   is a big value. Based on the above analysis and equation (52), e is
able to approach to 0 with different speeds, then achieve low overshoot and quick
response. According to the universal approximation theorem, an optimal fuzzy controller
u*fz is used to approximate the ideal controller, that is

u*  u*fz  φ, *      *T    (53)

 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 17

where  is the approximation error and is assumed to be bounded   E . And using a

 
fuzzy controller uˆ fz φ,ˆ to approximate u*fz  φ, *  as

 
uˆ fz φ,ˆ  ˆT  (54)

where ˆ is the estimated value of  * . The controller is designed as following form:

 
usf  uˆ fz φ,ˆ  uˆrb  φ  (55)

 
where the fuzzy controller uˆ fz φ,ˆ is designed to approximate the ideal controller and
the robust controller uˆrb  φ  is designed to compensate for the difference between the
ideal controller and the fuzzy controller.
Then,

  
  Fp     G p uˆ fz φ,ˆ  uˆrb  φ 

(56)

According to (51) and (56), the error equation governing the system can be obtained as
follows:

  
s  e  Ψ  d ,   e  G p u*  uˆ fz φ,ˆ  uˆrb  φ   (57)

Define u fz  u*  uˆ fz ,    *  ˆ, then it is obtained that

u fz   T    (58)
Define a Lyapunov function as
1 G G
 
V2 s, , E  s 2  p  T   p E 2
2 21 22
(59)

where E  E  Eˆ , Ê is the estimation of the approximation error bound, and 1 is

adaptive parameter coefficient, and 2 is adaptive error bound coefficient. They both are
positive constants.
G G
 
˙ ˙
V2 s, , E  ss  p  T   p E E
1 2
G G
 
˙ ˙
 sG p  T     uˆrb  p  T   p E E (60)
1 2
 ˙ 
T    G ˙

 G p s   sG p    uˆrb   p E E
 
1  2
 
For achieving V2  0 , the adaptive laws of the AFSMC are chosen as
˙ ˙
ˆ     1s (61)
18 W. Li, H. Du and W. Li

ˆ
uˆrb  Esgn s (62)
˙ ˙
Ê   E  2 s (63)
Then equation (58) can be rewritten as
 
V2 s, , E   sG p  E s G p   s G p  E     0 (64)

Figure 12 Membership function

3.3 Torque allocation

The braking torque is allocated to RBS and HBS through an allocator. In this paper, we
only consider emergency braking. The torque allocator is designed based on the principle
of maximising energy recovery. Assume that T is the total required braking torque
which is calculated by the AFSMC, Tr max is the maximum braking torque of the RBS. If
T  Tr max , the RBS can meet the required braking torque of the whole brake system. So
only RBS works: Trb*  T . In this case, braking performance can be guaranteed and
energy recovery is maximised. If T  Tr max , RBS and HBS will work together and HBS
is used to provide the required braking torque: Thb*  T , the RBS is used to eliminate the
error between Thb and T caused by slow dynamic characteristics of the hydraulic system
and other uncertain factors.
In summary, the MSMES and AFSMC scheme work as follows:

Step 1: Initialisation with relevant parameters as listed in Tables 1 and 2.

Step 2: Obtain the measurement of the system state: vehicle acceleration and derive the
friction coefficient .
Step 3: Search for optimal slip ratio d by using MSMES.
Step 4: Calculate control signal u by using AFSMC based on the difference between
d and  .
 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 19

Step 4.1: Obtain the fuzzy controller uˆ fz through (54), ˆ is based on (61) and  is
based on (50).
Step 4.2: Obtain the robust controller uˆrb through (62), Ê is based on (63).
Step 4.3: Obtain whole controller usf through (55).
Step 5: Implement the usf to torque allocator and then to the plant.
Step 6: Continue with Step 2 until the end.

4 Simulation results

Simulation studies were conducted to validate the effectiveness of the approach proposed
in this section. Relevant parameters of the braking system are listed in Table 1. For the
simulation test, the tyre-road model should be provided. C1 ~ C3 are parameters of
Burckhardt and Reimpell tyre-road model. When the parameters C1 ~ C3 is given, the
road condition is determined. These parameters are used to simulate road condition. For
the ESA, these parameters and optimal slip ratio are unknown. The ESA is used to search
for the optimal slip ratio.
In the simulations, a defined tyre-road function and the Burckhardt and Reimpell
tyre-road friction mode are used to test the proposed optimal slip ratio seeking algorithm,
respectively. The defined tyre-road function is given as follows:
(  * )
     2 * (65)
( * ) 2   2

and this function has a maximum value at    * with  ( * )   * . Relevant parameters

for the Burckhardt and Reimpell tyre-road friction model are shown in Table 2. The
maximum friction coefficient of cobble wet road is 0.38 and the optimal slip ratio of
cobble wet road is 0.18.
Table 1 Relevant parameters

Symbol Property Value Value

g Acceleration due to gravity 9.8 m/s2 1 Mx  108 Wb = 108 V·s
m Mass of the quarter-car 400 kg 1 G  104 T = 104 Wb/m2
J Moment of inertia of the wheel 1 N/ s2 1 Oe  103/(4) A/m
1 erg/G = 1 emu  103 A·
R Effective wheel radius 0.3 m
m2 = 103 J/T
k Integral coefficient 1 1 erg/(G·g) = 1 emu/g  1 A·m2/kg
 Time constant 20 1 erg/G = 1 emu  4  1010 Wb·m
1 Adaptive parameter coefficient 800 1  4
2 Adaptive error bound coefficient 200 1 cm3/g  4  103 m3/kg
CAN communication or
1 mechanical response lag
0.02 s   r
20 W. Li, H. Du and W. Li

Table 1 Relevant parameters (continued)

Symbol Property Value Value

2 Response time of the motor 0.01 s 1 erg/cm3  101 J/m3
1m Cost function weighting factor 1 1  1/(4)
2m Cost function weighting factor 0 0
 Constant number 0.1
 Constant number 1
 Constant number 100

Table 2 Parameters of the tyre-road model for various surfaces

Surface condition C1 C2 C3
Cobble wet 0.4004 33.708 0.1204

In the first case, slip ratio tracking is compared between SLF-SMC and AFSMC. At this
case, the desired slip ratio is designed as 0.2. The simulation result is shown in Figure 13.

Figure 13 Slip ratio tracking compared between SLF-SMC and AFSMC

As shown in Figure 13, the proposed AFSMC approaches the desired slip ratio faster
than SLF-SMC. It proves that the proposed AFSMC is more effective. In the second
case, the MSMES algorithm is tested on two different road conditions to validate the
effectiveness of the extreme seeking capability. For comparison, the ESA is also test on
the corresponding road conditions.
First, the MSMES and the ESA are tested on defined road conditions. In this test, we
set  *  0.4 and  *  0.6 . As shown in Figure 14, the optimal slip ratio seeking result of
the ESA is around 0.4. The oscillation of the ESA seeking process is caused by seeking
characteristic of the ESA as analysed in Sub-section 3.1. From the Figure 14, we also can
see that the MSMES is able to search for the optimal slip ratio (0.4) and there is no
oscillation after the optimal slip ratio is searched out. It proves that the indicator 1 and
indicator 2 of switching condition works and the seeking process stops when the optimal
slip ratio is searched out. It also proves the effectiveness of the MSMES.
 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 21

Figure 14 Process of MSMES and ESA seeking the optimal slip ratio on defined road

Then the MSMES and ESA are tested on a changing road condition. In the first 5 s, the
road is cobble wet road. Then the road changes to the defined road equation (65) in the
second 5 s. As shown in Figure 15, both the MSMES and the ESA are able to quickly
search for the optimal slip ratio even the road condition changes. Because the seeking
process never stops in the whole process, the ESA is able to search for the optimal slip
ratio when the road condition changes. Different with the ESA, the extremum seeking
capability of the MSMES is based on indicator 3 (as shown in Sub-section 3.1) when
road condition is a changing road condition. When the optimal slip ratio is searched out
on the cobble wet road, the seeking process stops and corresponding slip ratio is treated
as the optimal slip ratio. Compared to the ESA, the MSMES needs an indicator to active
seeking process when the road condition changes. As shown in Figure 15, the MSMES
restart seeking and search out the optimal slip ratio (0.4) when the road condition
changes from cobble wet road to the defined road condition. It proves that the indicator 3
is activated and the MSMES algorithm is able to seek the optimal slip ratio of road even
the road condition changes.

Figure 15 Process of MSMES and ESA seeking the optimal slip ratio on changing road condition

In the third case, the proposed method (MSMES-AFSMC) is tested on different road
conditions for the braking performance in terms of the braking distance. Usually, the
optimal slip ratio is treated as a constant number 0.2. For comparison, the constant
optimal slip ratio (0.2) is also tested on corresponding road conditions. The AFSMC is
also used to track the optimal slip ratio in the constant optimal slip ratio test. The initial
speed of vehicle is 30 m/s.
22 W. Li, H. Du and W. Li

First, the MSMES-AFSMC and the constant optimal slip ratio tracking (COSR-
AFSMC) are tested on a cobble wet road. The simulation results are shown in Figure 16.
As shown in Figure 16, the braking distance of MSMES-AFSMC is 124.4 m and the
braking distance of COSR-AFSMC 127.6 m. Because the optimal slip ratio of cobble wet
road is 0.18 and it is very close to 0.2, the obtained maximum friction coefficient based
on MSMES-AFSMC is close to the obtained maximum friction coefficient based on
COSR- AFSMC. Then the braking distance of MSMES-AFSMC is close to the braking
distance of COSR-AFSMC.

Figure 16 Braking distances of MSMES-AFSMC and COSR-AFSMC on cobble wet road

condition

Then, the MSMES-AFSMC and the COSR-AFSMC are tested on the defined road. As
shown in Figure 17, the braking distance of MSMES-AFSMC is 83.54 m and the braking
distance of COSR-AFSMC is 96.22 m. The MSMES-AFSMC has a better braking
performance than COSR- AFSMC. The optimal slip ratio of defined road is 0.4.
However, 0.2 is treated as the optimal slip ratio in the COSR-AFSMC. The friction
coefficient when wheel slip ratio is 0.2 is less than the maximum friction coefficient on
the defined road condition. And MSMES-AFSMC makes wheel slip ratio work at the
optimal slip ratio (0.4) and obtains the maximum friction coefficient of the defined road.
This makes the MSMES-AFSMC’s braking performance better than COSR-AFSMC’s
braking performance.

Figure 17 Braking distances of MSMES-AFSMC and COSR-AFSMC on defined road condition

And then, the MSMES-AFSMC and the COSR-AFSMC are tested on a changing road
condition. In the first 4 s, the vehicle is on cobble wet road. Then the vehicle is on the
defined road. As shown in Figure 18, the braking distance of the MSMES-AFSMC is
 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 23

115.6 m and the braking distance of the COSR-AFSMC 119.7 m. As shown in Table 3,
the braking performance improvement of MSMES-AFSMC on changing road is better
than cobble wet road and less than the defined road. This is because there is a balance
between small improvement in the first 4 s on the cobble wet road and large
improvement on the defined road.

Figure 18 Braking distances of MSMES-AFSMC and COSR-AFSMC on change road condition

Table 3 Braking distances OF MSMES-AFSMC and COSR-AFSMC

Braking distance (m) Cobble wet road Defined road Changing road
MSMES-AFSMC 124.4 83.54 115.6
COSR-AFSMC 127.6 96.22 119.7
Improvement 2.5% 13.2% 3.4%

As shown in Table 3, the MSMES-AFSMC has better braking performance than the
COSR- AFSMC under all the road conditions. Especially, when the optimal slip ratio of
tyre-road is far away from 0.2, the braking performance improvement of MSMES-
AFSMC will be more obvious. All the simulations validate that the innovative control
scheme (MSMES-AFSMC) makes vehicle have a less braking distance and better
braking performance under different road conditions.

5 Conclusions

In this paper, an innovative control scheme for ABS was proposed. It integrates MSMES
and AFSMC. The MSMES overcomes the oscillation problem which exists in
conventional ESA while keeping the extreme seeking capability. The AFSMC makes
wheel slip ratio track the obtained optimal slip ratio while the parameters of the system
are estimated. The integrated control scheme MSMES-AFSMC is able not only to search
for the optimal slip ratio under different road conditions but also has a better braking
performance than the conventional COSR for ABS.
In the future, energy recovery optimisation will be taken into consideration in this
control algorithm and an actual experimental vehicle test platform will need to be built to
test the control performance of the proposed control method.
24 W. Li, H. Du and W. Li

Acknowledgement

This research was supported under Australian Research Council’s Discovery Projects
funding scheme (project number DP140100303).

References
Amir, P. (2009) ‘Adaptive feedback linearization control of antilock braking systems using neural
networks’, Mechatronics, Vol. 19, pp.767–773.
Avesta, G., Mohammad, B. and Ebrahim, E. (2008) ‘An optimised braking force distribution
strategy for articulated vehicles’, Vehicle System Dynamics, Vol. 46, pp.849–856.
Chianga, W.P., Yin, D. and Hiroshi, S. (2014) ‘Slip-based regenerative ABS control for in-wheel-
motor drive EV’, Journal of the Chinese Institute of Engineers, Vol. 38, No. 2, pp.220–231.
David, M., Bevly, J., Christian, G. and Christopher, W. (2002) ‘The use of GPS based velocity
measurements for measurement of sideslip and wheel slip’, Vehicle System Dynamics,
Vol. 38, No. 2, pp.127–147.
Ehsan, H., Reza, Z., Amir, K., William, M., Alireza, K. and Shih, C. (2017) ‘Real-time estimation
of the road bank and grade angles with unknown input observers’, Vehicle System Dynamics,
Vol. 55, No. 5, pp.648–667.
Erkin, D. and Tunc, A. (2015) ‘Self optimizing ABS control algorithm with application’,
Proceedings of the 6th International Conference on Modeling, Simulation, and Applied
Optimization (ICMSAO), pp.1–6.
Erkin, D., Bilin, A.G. and Tankut, A. (2012) ‘Extremum-seeking control of ABS Braking in road
vehicles with lateral force improvement’, IEEE Transactions on Control Systems Technology,
Vol. 22, No. 1, pp.230–237.
Haiping, D., Weihua, L. and Yongjun, Z. (2015) ‘Tracking control of wheel slip ratio with velocity
estimation for vehicle anti-lock braking system’, Proceedings of the 27th Chinese Control and
Decision Conference, pp.1900–1905.
Hu, C., Wang, R.R., Yan, F.J. (2016) ‘Integral sliding mode-based composite nonlinear feedback
control for path following of four-wheel independently actuated autonomous vehicles’, IEEE
Transactions on Transportation Electrification, Vol. 2, No. 2, pp.221–230.
Khatun, P., Bingham, C.M., Schofield, N. and Mellor, P.H. (2003) ‘Application of fuzzy control
algorithms for electric vehicle antilock braking/traction control systems’, IEEE Transactions
on Vehicular Technology, Vol. 52, No. 5, pp.1356–1364.
Lee, Y.G. and Stanislaw, H. (2002) ‘Designing a genetic neural fuzzy antilock-brake-system
controller’, IEEE Transactions on Evolutionary Computation, Vol. 6, No. 2, pp.198–211.
Li, L., Li, X.J., Wang, X. Y., Liu, Y.H., Song, J. and Ran, X. (2016) ‘Transient switching control
strategy from regenerative braking to anti-lock braking with a semi-brake-by-wire system’,
Vehicle System Dynamics, Vol. 54, No. 2, pp.231–257.
Lin, C.M. and Hsu, C.F. (2003) ‘Self-learning fuzzy sliding-mode control for antilock braking
systems’, IEEE Transactions on Control Systems Technology, Vol. 11, No. 2, pp.273–278.
Miroslav, K. and Hsin-Hsiung, W. (2000) ‘Stability of extremum seeking feedback for general
nonlinear dynamic systems’, Automatica, Vol. 36, pp.595–601.
Mojtaba, A.K., Erdal, K., Mohammad, T. and Okyay, K. (2011) ‘Extended Kalman filter based
learning algorithm for Type-2 fuzzy logic systems and its experimental evaluation’,
IEEE Transactions on Industrial Electronics, Vol. 59, No. 11, pp.4443–4455.
Nahid, E.M. and Aliakbar, G. (2014) ‘Optimization of bang-of-bang TS-Fuzzy based via DARLA
technique for ABS system’, Journal of Advances in Computer Research, Vol. 5, No. 3,
pp.1–12.
 A modified extreme seeking-based adaptive fuzzy sliding mode control scheme 25

Selim, A.O., Keith, J.B., Adam, S., Chris, M., Gary, W., Graham, H., Cian, H. and Ross, M.M.
(2013) ‘Regenerative braking strategies, vehicle safety and stability control systems: critical
use-case proposals’, Vehicle System Dynamics, Vol. 51, No. 5, pp.684–699.
Wang, B., Huang, X.Y., Wang, J.M., Guo, X.X. and Zhu, X.Y. (2015) ‘A robust wheel slip ratio
control design combining hydraulic and regenerative braking systems for in-wheel-motors-
driven electric Vehicles’, Journal of the Franklin Institute, Vol. 352, pp.577–602.
Wang, L.B., Chen, S.L. and Ma, K. (2016) ‘On stability and application of extremum seeking
control without steady-state oscillation’, Automatica, Vol. 68, pp.18–26.
Wang, L.B., Chen, S.L. and Zhao, H. (2014) ‘A novel fast extremum seeking scheme without
steady-state oscillation’, Proceedings of the 33rd Chinese Control Conference, pp.8687–8692.
Wang, R.R., Hu, C., Yan, F.J. and Mohammed, C. (2016) ‘Composite nonlinear feedback control
for path following of four-wheel independently actuated autonomous ground vehicles’, IEEE
Transactions on Intelligent Transportation Systems, Vol. 17, No. 7, pp.2063–2074.
Ye, W.J., Shen, W.X., Zheng, J.C., Demon, H. and Daya, D. (2015) ‘Fuzzy sliding mode control
for logitudinal motion of underground mining electric vehicles’, Proceedings of the IEEE 10th
Conference on Industrial Electronics and Applications (ICIEA), pp.1432–1437.
Zhao, L.H., Liu, Z.Y. and Chen, H. (2010) ‘Design of a nonlinear observer for vehicle velocity
estimation and experiments’, IEEE Transactions on Control Systems Technology, Vol. 19,
No. 3, pp.664–672.
Zhao, Z.G., Zhou, L.J., Zhang, J.T., Zhu, Q., and Hedrick, J.K. (2017) ‘Distributed and self-
adaptive vehicle speed estimation in the composite braking case for four-wheel drive hybrid
electric car’, Vehicle System Dynamics, Vol. 55, No. 5, pp.750–773.