Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Nonlinear Dyn (2022) 108:3467–3484

https://doi.org/10.1007/s11071-022-07210-2

ORIGINAL PAPER

Adaptive constrained antilock braking control under


unknown time-varying slip-friction characteristics
Yanan Qiu · Zhiyong Dai

Received: 8 May 2021 / Accepted: 4 January 2022 / Published online: 10 April 2022
© The Author(s), under exclusive licence to Springer Nature B.V. 2022

Abstract It is clear that the unknown and uncertain mation and the robustness against uncertainties under
slip-friction characteristics arose difficulties in both time-varying friction conditions.
determining the optimal wheel slip and adopting the
accurate friction forces for the ABS. In this study, Keywords Time-varying friction estimation ·
the authors presented a nonlinear control strategy with Nonlinear adaptive control · Antilock braking system
adaptive ability under the time-varying but unknown (ABS) · Constrained control
friction conditions while adopting wheel slip range as
time-domain constraints. Three fundamental issues are
addressed in this study: the judgement of the wheel slip 1 Introduction
region, the estimation of the uncertain friction and the
actuator dynamics. In this research, a range of operat- The anti-lock braking system (ABS) is a safety-critical
ing wheel slip is resolved as a control target area and mechanism that prevents wheel locking during brak-
an asymmetric barrier Lyapunov function is adopted to ing. Two basic objectives of ABS are to maximize
cast constraints while tracking a reference wheel slip. braking forces to achieve efficient performance, and
An observer is developed seamlessly into the backstep- to preserve acceptable lateral forces simultaneously
ping procedure to estimate the unknown friction coeffi- to maintain adequate steerability and stability of the
cient. The motor dynamics is fully considered to inves- vehicle [13]. Hereby, two primary issues are focused
tigate the practical behavior under the physical satura- toward the design of ABS: what is the optimal oper-
tions of the actuation system. The proposed controller ating region with high efficiency, and how to ensure
guarantees the boundedness of output constraints and ABS work robustly subject to unknown and possibly
the asymptotic stability with a thorough mathematical changing road conditions. In this study, instead of reg-
proof. Simulation results confirms the effectiveness of ulating the wheel slip at an optimal point (λ → λ∗ ), a
the constraints on wheel slip, the accuracy of the esti- range of operating wheel slip (λ ∈ [λl , λh ]) is resolved
as a control target area to cast output constraints on
tracking a reference wheel slip, which can ensure the
Y. Qiu (B) wheel slip stay with a safe region to avoid wheel lock-
School of Aerospace Science of Technology, Xidian
ing as well as to guarantee the stability of the system.
University, Xi’an 710126, Shaanxi, China
e-mail: ynqiu@xidian.edu.cn Moreover, incorporating estimation of friction coeffi-
Z. Dai
cient can ensure the robustness of the braking perfor-
School of Mechano-Electronic Engineering, Xidian University, mance against the unknown time-varying friction in the
Xi’an 710071, Shaanxi, China realm.

123
3468 Y. Qiu, Z. Dai

Three fundamental issues that determine the per- at a priori optimal point based on the look-up tabular
formance of ABS are addressed in this study: (1) the approaches. These tables have been calibrated through
judgement of the wheel slip range with consideration iterative laboratory experiments and engineering field
of internal coupling between longitudinal forces and tests under specified road conditions, and were widely
lateral forces; and (2) the robustness against the fric- used in most commercial cars [18]. However, these
tion uncertainty of the tire-road adhesion; (3) the actu- methods may not be adaptive or robust when the road
ator dynamics with the formalization of the demanding conditions are not consistent with the priori model. In
torque. the second type, as given in [1,17,24], the dynamic
Both the first factor and the second factor are related behavior of the tire-road adhesion was considered in
to the unknown and uncertain slip-friction characteris- advance and was resolved with an estimation of friction
tics, which vary widely along with many factors, such coefficient, while the wheel slip have been set to track
as different road conditions and wheel loads. In other one reference value which was appropriate for the road
words, the tire-road adhesion is highly dependent on condition, where both the accuracy and the responding
both the wheel slip and the road conditions [18], in time of the estimation procedure are crucial to the sys-
turn, the optimal wheel slip correspondent to the max- tem. In the third type, as given in [9,10,34,37], both
imum friction is linked to the slip-friction character- optimal wheel slip or road conditions were unknown,
istics. Thus, it is clear that the maximization of tire- the online estimation friction coefficient was utilized to
road adhesion at an optimum wheel slip is not easy seek the peak point of μ − λ curves with respect to the
to achieve and cannot always be satisfactory. Unfortu- unknown and possibly changing road conditions. This
nately, the misjudgment of the optimal wheel slip will group of methods highly relied on the computing unit
bring difficulty in judging the proper braking torques on and required feasible measurements with sensors.
the brake mechanism. Insufficient braking torques will However, issue related to the third factor is relatively
result in a comprised braking efficiency, while exces- new and have recently received attention because elec-
sive braking torques will cause further increase in the tric motors are increasingly used in electric vehicles
wheel slip beyond its optimum value. Moreover, the (EVs) along with the needs of the market and the envi-
latter case will lead to less generation of the longitu- ronment [7,12,29]. The ABS function for full EVs can
dinal forces and dramatic decrease in the lateral force be realized using both traditional friction forces and
[7]. Consequently, the reduction in tire-road adhesion negative torques from motors. So far, the majority of the
can result in the loss of lateral steerability and stability relevant research is focused on exploiting regenerative
via blocking of the tire rotation [9], thereby may lock torques (such as electro motive force, EMF) and also
the wheel. blended composition with frictional braking to improve
Addressing the first factor, this work is paying atten- the performance of ABS [8,13,25], where the opera-
tion to cast output constraints on the wheel slip to ensure tional condition and the braking dynamics determine
that the ABS should only operate within a region, such the involvement of regenerative torques. However, in
that the excessive wheel torques can be controlled to practice, the pure regenerative torques can hardly arrive
decrease to the level where the lateral forces are not the torque demands, while the conjunction of motion
jeopardized as well as the longitudinal forces stay close will complicate the ABS control strategy and risk the
to the maximum values. In addition, the nonlinearities vehicle safety. As such, in almost all existing EVs, it
and the uncertainties in the slip-friction characteristics is common to disengage the regenerative torque dur-
arose challenges to make ABS work robustly under ing the activation of ABS and decelerate the vehicle
changing road conditions. An estimation of the tire- entirely by the frictional braking [6].
road friction coefficient during the braking maneuver The electro-mechanical brakes (EMB) with thor-
is adopted to address the second factor. ough consideration of direct current (DC) motor dynam-
In the last few decades, numerous advanced control ics are investigated in this work. It is clear that the
methods has been investigated to address the aforemen- performance and the stability of the vehicle heavily
tioned two issues primarily in three scopes. In the first depend on the proper operation of the motors. Since
type, as given in [19,20,22,27,31,36], the slip-friction the electric motors can be controlled more precisely
characteristics were considered to be known a priori, and significantly efficient with inexpensive and mature
where ABSs were designed to regulate the wheel slip control methods, the EMBs can continuously achieve

123
Adaptive constrained antilock braking control 3469

fast and accurate torque responses comparing with the to tackle with the unknown slip-friction characteristics
hydraulic-based friction brakes [2,7], and can supply while considering the wheel slip range as time-domain
higher torque comparing with pure electric brakes [8]. constraints based on BLFs and DSC. Our algorithm
This, on the one hand, reduces or even eliminates the presents several new interesting points. First, it copes
coupling between motors and vehicles with its higher with the wheel slip operating region directly by using
bandwidth frequency of motors (above 10 Hz compar- constrained control so as to improve the braking effi-
ing to 3–10 Hz with conventional ones) [7,13], and on ciency in the longitudinal direction as well as to avoid
the other hand allows ABS control to be formulated as the loss of control and stability in the lateral direction.
a classical regulation problem due to the accurate con- Second, the method does not require any knowledge
tinuous adjustment of the clamping forces during the either of the current road conditions or of the instanta-
braking process [34]. Although more research needs neous value of the normal force exerted on the tire.
be carried on to downsize EMB and to reduce its high Third, the global asymptotic stability to the desired
energy consumption, the mentioned advantages pro- set point is proved mathematically with giving precise
vide DC motors to be a good candidate for low-cost bounds on the gains of the control laws. Fourth, the
EVs in this extremely cost-sensitive market [5]. simulation results confirm the interests of the method.
In this study, the first factor is transformed into The outline of this paper is organized as follows: a
nonlinear constrained control philosophy, where the quarter car model together with the actuator dynamics
wheel slip operating region is thoroughly analyzed with are discussed in Sect. 2. The preliminaries of the barrier
respect to different or even changing road conditions, Lyapunov function are formulated in Sect. 3. In Sect. 4,
and both the upper bound and lower bound of wheel the adaptive constrained control strategy incorporating
slip are achieved as constraints. The second factor is with an online estimation of friction coefficient is pre-
resolved to an online estimation by adopting friction sented and its stability is demonstrated. The simulation
coefficient as a time-varying unknown parameter in study under different road conditions and road transi-
a traditional nonlinear system, where the controller tions are given in Sect. 5. The conclusions are drawn
robustness to model uncertainty is investigated through in the last section.
theoretical proof and simulation study. The dynamics
of electrical motors are fully taken into account to han-
dle with the third factor, such as the bandwidth and the 2 Formulation of ABS
physical saturation of the motor in the operating status,
where the implementation of electric motor dynamics A relatively simple but effective mathematical model
construct the braking system as a high-order system. has been built for the system analysis, design and test-
From the perspective of control methods, the bar- ing study of the control law. It is pertinent to dis-
rier Lyapunov function (BLF) [21] has been well- cuss some principal assumptions that are practically
studied to handle with nonlinear constrained sys- considered in this work. First, the modeling and con-
tem control without explicit solutions of the systems trol design are primarily presented under the braking
recently [28,35]. Unlike the original Lyapunov func- maneuver, but extended applications under accelera-
tions, this special Lyapunov candidate grows infin- tion are briefly discussed in Sects. 3 and 4. Second, it
ity whenever any function arguments approach certain is assumed that the control unit has an access to the
limiting values, which ensures the constraints not be system states through sensors or observers, followed
transgressed along the system trajectories. However, by elaboration in each subject.
most of the mentioned methods borrowed backstep-
ping technology [16], where the problem of ‘explosion
of complexity’ grows as the system order n increases. 2.1 Single-corner model
Thereby, dynamic surface control (DSC) [33] is intro-
duced to eliminate repeated differentiation induced The single-corner model is widely used in the ABS
from employing the dynamics of the electric motors design for its simple but proper description of the longi-
in this work. tudinal vehicle braking dynamics [14,31,36]. To appro-
In this paper, the authors develop a low conser- priately describe the model, it is assumed that the ver-
vatism nonlinear control strategy with adaptive ability tical load of the vehicle is uniformly distributed, the

123
3470 Y. Qiu, Z. Dai

Fig. 1 Quarter car forces Table 1 Denotations of DC dynamics


and torques
Name Description

ωm Motor angular velocity


im Motor armature current
e Applied armature voltage
eb Back-EMF voltage
Te Motor electromagnetic torque
TL Motor friction torque
braking torque is symmetrically applied and both the Jm Moment of inertia of motor
lateral forces and the interaction between wheels are R Resistance of motor
neglected. It is also assumed that the vehicle can main- L Inductance of motor
tain the correct heading under asymmetric disturbance ke Back-EMF constant
by using the steering system. In addition, the speed kt Motor electromagnetic torque constant
information is assumed to be available to the control
unit by well-investigated methods in this work, such
as measuring speed of non-driving wheels [7], imple-
menting observers [11] or installing a fifth wheel [15]. 2.2 Actuator dynamics
The linear and the rotational motion of the wheel
during the braking process is shown in Fig. 1, and the The actuation system consists of the DC motor con-
dynamics can be described as Eq. (1) [36]. trolled with a variable armature voltage and a gearbox
transmission to utilize the friction force on the pedal.
m V̇x = − F f
(1) To simplify the model, it is assumed that the motor is at
Jw ω̇w = rw F f − Tb , the idle state or the reverse state. The dynamics of the
where m is the mass of a quarter vehicle, Jw is the wheel DC motor can be described as Eq. (4) with parameters
inertia, rw is the tire rolling effective radius, Vx is the depicted in Table 1.
longitudinal velocity of the vehicle, ωw is the angu-
Jm ω̇m = Te − TL = kt i m − TL
lar velocity of the wheel, F f is the tire-ground adhe- (4)
sion force, Tb is the braking torque generated by the e = L i̇ m + Ri m + eb = L i̇ m + Ri m + ke ωm .
actuation system. Furthermore, the tire-ground adhe- Due to the fluctuation of the friction coefficient
sion force F f is often modeled as: between the brake pad and the brake disk, it is acknowl-
F f = Fn μ(λ, Vx ), (2) edged that a typical displacement versus pressure curve
where Fn is the normal tire reaction force that depends is not linear, where hysteresis is an attribute of the
on the mass, location of the center of gravity, and the majority of EMBs [30]. In this paper, the braking torque
steering and suspension dynamics, and Fn = mg is is assumed to be proportional to brake pressure [23],
denoted in this work; μ is a nonlinear factor defined while the brake pressure is assumed to be proportional
as the ground-contact friction coefficient depending on to the relative angle of the motor θm . Thereby, the brak-
the road condition, the wheel slip λ and other parame- ing torque Tb can be formalized by the relative angle of
ters, such as the longitudinal velocity of the vehicle Vx . the motor θm through a torque constant kω with a func-
It is worthy mentioned here the value of μ is estimated tion Tb = kω θm , where the braking torque is assumed
through an observer developed in the control algorithm. to be linear for small values of the relative angle and
The wheel slip ratio λ during braking is formalized safely saturates at higher levels [3,32]. Thus, the brak-
as: ing torque coupled with the DC motor can be formal-
Vx − ωw rw ized as Equation (5),
λ= . (3)
Vx
Ṫb = kω ωm . (5)
According to Eq. (3), 0 ≤ λ ≤ 1 can be obtained. λ = 0
represents the pure rolling condition (Vx = ωw rw ), where kω is determined by the gear ratio and the tor-
while λ = 1 corresponds to the pure slip condition sional spring constant. Note that an observer-based tor-
(ωw = 0), namely locked. sional spring constant can be adopted to the control unit

123
Adaptive constrained antilock braking control 3471

in a realm. However, a normal spring constant is used where


in this simulation study. x1 = λ; x2 = Tb ; x3 = ωm ; x4 = i m ; u = e
rw
f 1 = φ1 (x1 )μ; g1 =
Vx Jw
2.3 Vehicle braking system dynamics f 2 = 0; g2 = kω
TL kt
Substitute Eq. (2) into (1) to calculate V̇x and ω̇w , f3 = − ; g3 =
Jm Jm
R ke 1
V̇x = −gμ f 4 = − i m − ωm ; g4 = .
(6) L L L
mgμrw Tb
ω̇w = − . Here, x i = [x1 , x2 , . . . , xi ]T , i = 1, . . . , 4 represents
Jw Jw
the states, f i , gi are smooth functions. u and y are the
Taking the time derivative of Eq. (3), we obtain input and output, respectively. μ ∈ Rl is an unknown
time-varying parameter, with its dynamics depicted as
(1 − λ)V̇x − rw ω̇w
λ̇ = . (7) μ̇ = (t); φ1 (x1 ) ∈ Rl is the regressor vector, whose
Vx
elements are previously known functions of the states.
Substituting Eqs. (6) into (7), it follows:
(1 − λ) rw
λ̇ = − gμ − (mgrw μ − Tb ) 3 BLF formulation and main assumptions
Vx Vx Jw
(1 − λ) r2 rw Definition 1 [35] Barrier Lyapunov function is a con-
=−[ g + w mg]μ + Tb (8)
Vx Vx Jw Vx Jw tinuously differentiable and positive definite scalar
rw function V (x), defined with respect to the system ẋ =
= f 1 (λ, μ) + Tb ,
Vx Jw f (x) on an open region D containing the origin. It has
where continuous first-order partial derivatives at every point
of D, has the property V (x) → ∞ as x approaches
  the boundary of D, and satisfies V (x(t)) ≤ b, ∀t ≥ 0
(1 − λ) r2
f 1 (λ, μ) = φ1 (λ)μ, φ1 (λ) = − g + w mg . along the solution of ẋ = f (x) for x(0) ∈ D and some
Vx Vx Jw
positive constant b.
It is clear that φ1 (λ) is a smooth function, and μ is
The following Assumptions and Lemmas are useful for
the unknown time-varying friction coefficient related
establishing constraint satisfaction and performance
to the wheel slip λ and the road conditions.
bounds.
By incorporating the actuation system, we can
acquire the model of the vehicle braking system as Eq. Assumption 1 There exist positive constants k c1 , k c1 ,
(9). Yl0 , Yh0 , Y1 , Y2 , such that the desired trajectory yd (t)
rw and its time derivatives satisfy k c1 ≤ −Yl0 ≤ yd (t) ≤
λ̇ = f 1 (λ, μ) + Tb Yh0 ≤ k c1 and | y˙d (t)| ≤ Y1 , | y¨d (t)| ≤ Y2 , ∀t ≥ 0,
Vx Jw
implying that they are continuous and available in a
Ṫb = kω ωm
compact set  yd := {yd ∈ R : yd2 + ẏd2 + ÿd2 ≤ δ yd } ⊂
kt TL (9) R.
ω̇m = im −
Jm Jm
R ke 1 Assumption 2 The functions gi are positive, and there
i̇ m = − i m − ωm + e. exists a class of positive constants gi min and gi max such
L L L
that 0 < gi min ≤ gi (x i ) ≤ gi max for x1 ∈ (k c1 , k c1 ).
Rewrite Eq. (9) into a standard single-input single-
output (SISO) strict feedback nonlinear system, Assumption 3 The unknown time-varying parameter
μ(t) and its first-order derivative (t) are smooth and
ẋi = f i (x i ) + gi (x i )xi+1 , i = 1, 2, 3 bounded, there exist positive constants and
such
ẋ4 = f 4 (x 4 ) + g4 (x 4 )u (10) that μ(t) ∈ L∞ with
μ(t)
∞ ≤ , (t) ∈ L∞ with
y = x1 ,
(t)
∞ ≤
for all t ≥ 0.

123
3472 Y. Qiu, Z. Dai

Lemma 1 [28,35]. For any positive constants ka1 and


kb1 , let Z := {s1 ∈ R : −ka1 < s1 (t) < kb1 } ⊂ R
and N := Rl × Z ⊂ Rl+1 be open sets. Consider the
system
η̇ = h(t, η),
where η := [ω, s1 ]T ∈ N , and h : R+ × N → Rl+1
is piecewise continuous in t and locally Lipschitz in s1 ,
uniformly in t, on R+ × N . Suppose that there exist
functions U : Rl → R+ and V1 : Z → R+ , con-
tinuously differentiable and positive definite in their
respective domains, such that
V1 (s1 ) → ∞ as s1 (t) → −ka1 or s1 (t) → kb1
Fig. 2 Relationship between μ and λ
γ1 (
ω
) ≤ U (ω) ≤ γ2 (
ω
),
where γ1 and γ2 are class K ∞ functions. Let V (η) :=
V1 (s1 )+U (ω), and s1 (0) ∈ (−ka1 , kb1 ). If the inequal- [5]. In the left zone of the λ∗ (0 ≤ λ ≤ λ∗ ), when the
ity increased braking torque Tb is employed to the wheel,
λ increases to λ∗ , while μ grows to its maximum value
∂V
V̇ = h ≤ −cV + υ μmax . At this point, the friction force generated from the
∂η tire-road adhesion is maximized, which will give birth
holds in the set η ∈ N , and c, υ are positive constants, to the reduction of Vx with high efficiency. Thereby,
then ω remains bounded, and s1 ∈ (−ka1 , kb1 ), ∀t ∈ the braking system operates in the stable region. Once
[0, ∞). the wheel slip falls to the right of the λ∗ (λ∗ ≤ λ ≤ 1),
2
kb1 s12
the friction coefficient μ decreases as the wheel slip
Lemma 2 [28]. The inequality log k 2 −s 2 ≤ 2 −s 2
kb1 λ grows. Unfortunately, any increase of the braking
b1 1 1
holds for all |s1 | < kb1 , . torque in this region would result in further increase
in the wheel slip λ due to the reduction in the longi-
tudinal friction coefficient μ (according to Eq. (8) in
4 Main results Sect. 2.3). This will finally lead to wheel locking when
λ grows to 1. Meanwhile, the dramatical drop of the
4.1 Control objective lateral friction force will be insufficient to sustain the
lateral steering capability and the lateral stability. Thus,
From μ−λ curves in Fig. 2, it is clear that both longitu- the slip dynamics are open loop unstable in this region.
dinal and lateral friction coefficients are highly depen- The control objective in this work is to design an
dent on the wheel slip ratio λ. Moreover, Fig. 2 shows adaptive constrained control law such that: (1) the
a coupling phenomenon that the lateral and the longi- wheel slip can operate within a predetermined bound-
tudinal forces interact with each other : the longitudi- ary (λl ≤ λ ≤ λh as the operating region) to avoid
nal friction coefficient curve has a single peak value severe wheel self-locking and loss of the lateral steer-
μmax and a corresponding optimum wheel slip value ability and the stability; (2) the output of wheel slip
λ∗ , while the lateral friction coefficient decreases con- y(t) = λ can track a desired value yd close to the opti-
siderably to a low magnitude as λ keeps increasing mum value so as to generate longitudinal tire-road fric-
(μ y = 0 when λ = 1). Acknowledging this, it is of tion close to its peak value with efficient braking; (3)
important significance to achieve a reasonable tradeoff ABS has an adaptive ability to work robustly on differ-
between the longitudinal friction μ and the lateral fric- ent road conditions and changing road conditions; (4)
tion μ y by specializing a wheel slip operating region all the closed-loop signals are bounded.
(λl ≤ λ ≤ λh ).
Usually, two regions are identified as the stable Remark 1 Denoted the initial wheel slip ratio as λ(0)
region and the unstable region at each side of the λ∗ and y(0) = λ(0).

123
Adaptive constrained antilock braking control 3473

1. The desired wheel slip yd = λd is chosen for differ- It is clear that the barrier function is smoothly
ent value close to its optimum wheel slip λ∗ accord- extended by a suitable penalty term outside of the cor-
ing to road surface because λ∗ varies corresponding responding constraints set. V (s1 ) is positive, piecewise
to different road conditions. smooth and continuously differentiable in terms with
2. k c1 = 0.3 is selected as λh to be the upper output the fact that lims1 →0+ dV1 /ds1 = lims1 →0− dV1 /ds1 =
constraint on the wheel slip to prevent ABS from 0 in the set −ka1 < s1 (t) < kb1 . Thus, V (s1 ) belongs
operating in deep slip region. to C 1 in the set Z and is a valid Lyapunov Function
3. The initial wheel slip y(0) may be outside the candidate.
operating region at the activation of ABS, such as
y(0) < 0 while the vehicle is under acceleration. To
accommodate these practical applications within 4.3 The observer of the time-varying friction μ(t)
this control framework, the lower output constraint
λl is selected as k c1 = y(0) − 0.05 to be slightly Let μ̂ be the estimation of the unknown friction coef-
less than y(0), which ensures any starting value of ficient μ, such that μ̃ = μ − μ̂ denotes the estimation
y(0) within augmented constraint and extends the error. The following nonlinear algorithm is designed to
effectiveness of the algorithm. observe the unknown parameter μ:
γ kμ
In other words, the objective of this algorithm is to μ̂˙ = γ ϑs1 φ1 (x1 ) + (ẋ1 − fˆ1 − g1 x2 ), (13)
φ1 (x1 )
specify a compact set  y := {y ∈ R : k c1 < y(t) <
k c1 , k c1 > k 0 + Yh0 + |yd (0)|, k c1 < k 0 + Yl0 + in which
|yd (0)|} ⊂ R, ∀t > 0 for any initial compact set to fˆ1 (x1 , μ) = μ̂φ1 (x1 ). (14)
which y(0) belongs 0y := {y ∈ R : k 0 ≤ y(0) ≤
Hence, the error dynamics of the observer can be given
k 0 } ⊂ R. Based on an asymmetric barrier Lyapunov by
function and dynamic surface control technique, an
adaptive constrained control is derived in the follow- μ̃˙ = μ̇ − μ̂˙
ing subsections to address the asymmetric output con- = (t) − γ ϑs1 φ1 (x1 )
straints on wheel slip and the time-varying friction γ kμ (15)
coefficient. − ( f 1 + g1 x2 − fˆ1 − g1 x2 )
φ1 (x1 )
= (t) − γ ϑs1 φ1 (x1 ) − kμ γ μ̃,
4.2 The adoption of asymmetric barrier Lyapunov
function 4.4 Controller synthesis

Denote s1 = x1 − yd as the tracking error, then the fol- Similar to the traditional backsteeping, denote a change
lowing asymmetric barrier Lyapunov function (ABLF) of coordinates: s1 = x1 − yd as the tracking error and
is introduced, si = xi − z i , i = 2, 3, 4 as the virtual errors of step i,
1 − q(s1 ) k2 where z i is the filtering output of the stabilizing func-
V (s1 ) = log 2 a1 2
2 ka1 − s1 tion αi−1 developed at step (i-1). For i = 2, 3, 4, the
following first-order filters with time constant τi are
q(s1 ) k2
+ log 2 b1 2 (11) introduced to generate z i and its derivative ż i :
2 kb1 − s1
τi ż i + z i = αi−1 , z i (0) = αi−1 (0). (16)
where
 We can obtain the dynamic surface errors χi = z i −
1 s1 > 0
q(s1 ) = (12) αi−1 with χi (0) = 0, and ż i = −χi /τi with ż i (0) = 0.
0 s1 ≤ 0.
Consider a Lyapunov candidate comprising the
ka1 and kb1 are the lower and upper error constraints
quadratic Lyapunov functions and an ABLF with
on the tracking error s1 , which are chosen as ka1 =
reduced order.
yd −k c1 , kb1 = k c1 − yd corresponding to the choice of
n
1 2  1 2 μ̃2
n
yd , respectively. q(s1 ) was abbreviated as q throughout V = V (s1 ) + si + χ + , (17)
the paper. 2 2 i 2γ
i=2 i=2

123
3474 Y. Qiu, Z. Dai

1−q q
Denote ϑ = 2 (t)−s 2
ka1
+ 2 (t)−s 2 ,
kb1
and then take the Substitute the observer (13) with its error dynamics
1 1
time derivative of V , (15) , the stabilizing function and control law (19) and
  the derivatives of χi (20) into (18), it follows that
1−q q
V̇ = 2 + 2 s1 ṡ1 V̇ = ϑs1 (−k1 s1 + g1 s2 + g1 χ2 + f 1 − μ̂φ1 ) + χ2 χ̇2
ka1 (t) − s12 kb1 (t) − s12
+s2 (−k2 s2 + g2 s3 + g2 χ3 − ϑg1 s1 ) + χ3 χ̇3
  μ̃μ̃˙
4 4
+ si ṡi + χi χ̇i + +s3 (−k3 s3 + g3 s4 + g3 χ4 − g2 s2 ) + χ4 χ̇4
γ
i=2 i=2 μ̃μ̃˙
+s4 (−k4 s4 − g3 s3 ) +
=ϑs1 [ f 1 + g1 (s2 + χ2 + α1 ) − ẏd ] (18) γ

3 = ϑ(−k1 s12 + g1 s1 χ2 + μ̃φ1 s1 )
+ si [ f i + gi (si+1 + χi+1 + αi ) − ż i ] 
1
i=2 +χ2 − χi + ζ2
τ2
 μ̃μ̃˙
4
1
+ s4 ( f 4 + g4 u − ż 4 ) + χi χ̇i + . +(−k2 s22 + g2 s2 χ3 ) + χ3 (− χ3 + ζ3 )
γ τ3
i=2
1
Design the stabilizing functions and control law in the +(−k3 s3 + g3 s3 χ4 ) + χ4 (− χ4 + ζ4 )
2

modified backstepping procedure as, τ4


μ̃
1 −k4 s42 + ((t) − γ ϑs1 φ1 − kμ γ μ̃)
α1 = (−k1 s1 − μ̂φ1 (x1 ) + ẏd ), γ
g1
 χ2
1 χ2 = ϑ(−k1 s12 + g1 s1 χ2 ) − 2 + χ2 ζ2
α2 = −k2 s2 − f 2 − ϑg1 s1 − , τ2
g2 τ2
 (19) χ 2
1 χ3 −k2 s22 + g2 s2 χ3 − 3 + χ3 ζ3
α3 = −k3 s3 − f 3 − g2 s2 − , τ3
g3 τ3
 χ2
1 χ4 −k3 s32 + g3 s3 χ4 − 4 + χ4 ζ4
u= −k4 s4 − f 4 − g3 s3 − , τ4
g4 τ4
(t)μ̃
where ki , i = 1, . . . , 4 and τi , i = 2, 3, 4 are strictly −k4 s42 + − kμ μ̃2 . (21)
γ
positive constants, μ̂ is the estimate of the time-varying
friction coefficient at time t by adopting the observer Suppose i ∈ R2i−1 be a given compact set
(13). as i := {[s1 , . . . , si−1 , χ2 , . . . , χi , μ̃]T : 1−q
2 log
2
ka1 q
2
kb1
i−1 1 2
i 1 2
Remark 2 By incorporating the DSC technique, the k 2 −s 2
+ 2 log k 2 −s 2 + k=2 2 sk + k=2 2 χk +
a1 1 b1 1
order of the asymmetric barrier Lyapunov function 1
2γ μ̃2 ≤ δi , δi > 0}. Under the Assumptions 1 ,2 and
is reduced from 2n to 2, the repeated differentiations 3 , ζi and (t)have maximum absolute values Mi and
of α1 , α2 , α3 and yd are eliminated, and the control
in a compact set  yd × i , respectively. According
scheme is largely simplified in this high-order system. to Young’s inequality, we obtain these inequalities for
i = 2, 3, 4
For i = 2, 3, 4, take the time derivative of χi
χi2 ζi2 υi M2 υi
χ̇i = ż i − α̇i−1 |χi ζi | ≤ + ≤ i χi2 + , υi > 0
χi 2υi 2 2υ 2
= − + ζi (x i−1 , yd , ẏd , ÿd ),
(20)  i 
τi χi |(t)μ̃|
2
gi−1 si−1 χi ≤ gi−1 max si−1 +
2
where 4 γ
∂αi ∂αi ∂αi (t)2 2 υμ
2 υμ
ζi (x i−1 , yd , ẏd , ÿd ) = − ẋ i−1 − ẏd − ÿd ≤ μ̃ + ≤ μ̃2 + , υμ > 0.
∂ x i−1 ∂ yd ∂ ẏd 2υμ γ 2 2 2υμ γ 2 2
is a continuous function. (22)

123
Adaptive constrained antilock braking control 3475

Table 2 Model parameters of the braking system


Parameter Value Parameter Value

Jm 0.0001 kg m2 m 425 kg
R 0.2  Jw 2 kg m2
L 0.01 H rw 0.325 m
ke 0.05 V/rad/s kw 50 N m/rad
kt 0.1 N m/amp

Table 3 Friction model parameters [4]


Surfaces conditions c1 c2 c3 c4

Dry asphalt 1.2801 23.990 0.52 0.02–0.04


Wet asphalt 0.857 33.822 0.347 0.02–0.04
Snow 0.1946 94.129 0.0646 0.02–0.04

Table 4 Braking distance and MFDD in each scenario


Parameter Method Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5
Dry surface Wet surface Dry to wet Wet to dry Wet to snow
λd = 0.12 λd = 0.1 λd : 0.12–0.1 λd : 0.1–0.12 λd : 0.1–0.05

Braking distance (m) This work 57.3707 80.0519 69.5853 65.6788 231.6998
SMC 59.7144 81.6787 73.2011 67.4081 239.5460
MFDD (m/s2 ) This work 10.019 7.1058 7.7592 9.1817 2.1171
SMC 9.7914 7.0256 7.5245 9.0375 2.0770

Substitute the inequalities (22) into (21), it follows Thereby, we have


   
χ22 χ2 1 ϑ M22
V̇ ≤ ϑ −k1 s1 + g1 max s1 +
2 2
− 2 V̇ ≤ −ϑ(k1 − ϑg1 max )s12
− − g1 max − χ22
4 τ2 τ2 4 2υ2
   
M22 2 υ2 χ32 1 1 M32
+ χ + − k2 s2 + g2 max s2 +
2 2 −(k2 − g2 max )s2 −
2
− g2 max − χ32
2υ2 2 2 4 τ3 4 2υ3
 
χ32 M2 υ3 1 1 M42
−(k3 − g3 max )s3 −
2
− g3 max − χ42
− + 3 χ32 + − k3 s32 τ4 4 2υ4
τ3 2υ3 2
  
χ32 χ2 M2
2  4
υi υμ
+g3 max s3 +
2
− 4 + 4 χ42 −k4 s42 − kμ − μ̃ 2
+ + .
4 τ4 2υ4 2υμ γ 2 2 2
i=2

υ4
2 υμ (24)
+ − k4 s42 + μ̃2 + − kμ μ̃2 . (23)
2 2υμ γ 2 2

123
3476 Y. Qiu, Z. Dai

0.14
1600

0.12 1400

0.1 1200

0.08 1000

0.06 800

600
0.04
400
0.02
200
0
0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 3.5

(a) Slip Ratio (b) Braking Torque

1
100
0.8

0.6

0.4 50

0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5

1
100
0.8

0.6
50
0.4

0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5

(c) Friction Coefficient (d) Vehicle Velocity and Wheel Velocity

Fig. 3 Performances under dry asphalt surface (λd = 0.12)

 
4
Remark 3 In order to guarantee the closed-loop sta-
(t)2 υi υμ
μ = kμ − , υ = + .
bility, the control gains ki should be limited to k1 > 2υμ γ 2 2 2
i=2
ϑg1 max , k2 > g2 max , k3 > g3 max , k4 > 0, the time
(25)
constants τi should be limited to τ12 > ( ϑ4 g1 max +
 
M22 g2 max M32 g3 max M42
2υ2 ), 1
τ3 > 4 + 2υ3 , 1
τ4 > 4 + 2υ4 , It is clear that K i > 0, i = 1, · · · , 4, i > 0, i =
and the observer gain should be limited to kμ γ2 > 2, 3, 4 and μ > 0 by the precise restrictions on

2 gains selections. Then, a positive definite matrix Q is
2υμ ,respectively.
described as Q=diag(K 1 , K 2 , K 3 , K 4 , 2 , 3 , 4 , μ ).
For clarity, denote a change of gain parameters: According to Lemma 2, we can have the following
K 1 = k1 − ϑg1 max , K 2 = k2 − g2 max , K 3 = k3 − g3 max , inequality which holds in the set Z

1 η M22
K 4 = k4 , 2 = − g1 max − ,
τ2 4 2υ2 V̇ ≤ −cV + υ, (26)
 
1 g2 max M32
3 = − − ,
τ3 4 2υ3 where
 
1 g3 max M42
4 = − − ,
τ4 4 2υ4 c = 2λmin (Q) > 0. (27)

123
Adaptive constrained antilock braking control 3477

0.12 1200

0.1
1000

0.08
800

0.06
600

0.04
400
0.02
200
0
0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

(a) Slip Ratio (b) Braking Torque


0.8

100
0.6

0.4
50
0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

0.8

100
0.6

0.4
50
0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

(c) Friction Coefficient (d) Vehicle Velocity and Wheel Velocity

Fig. 4 Performance on wet asphalt surface (λd = 0.1)

4.5 Stability analysis together with V̇ ≤ −cV + υ, it can be shown that


η̇ = h(η, t) satisfy the conditions in Lemma 1, we
Theorem 1 Consider the closed-loop system consist- infer that −ka1 ≤ s1 (t) ≤ kb1 , ∀t > 0. Because
ing of the plant (Eq. (10)), the adaptation law (Eq. y(t) = s1 (t) + yd (t), we obtain that −ka1 + yd <
(13)), the dynamic surface filters (Eq. (16)), the stabi- y(t) < kb1 + yd . From the choices of ka1 , kb1 , we can
lizing functions and the control law (Eq. (19)) under conclude that k c1 < y(t) < k c1 , ∀t ≥ 0.
Assumptions 1–3. For bounded initial conditions, sat- Multiply V̇ ≤ −cV + υ by ect , and then integrate
isfying k c1 ≤ y(0) ≤ k c1 , V (0) ≤ δ, there exist both sides
constants within the selected range for control gains
and time constants depicted in Remark 3, such that υ  −ct υ υ
the close-loop system is semi-globally stable in the 0 ≤ V (t) ≤ V (0) − e + ≤ V (0) + .
c c c
sense that the output constraints are never violated, (28)
i.e., k c1 < y(t) < k c1 , ∀t ≥ 0, all of the closed-loop
signals are bounded, and the output tracking error con-
verges to a neighborhood of zero as t → ∞. Because V (0) ≤ δ, then V (t) ≤ δ + υc , ∀t > 0.
Thus, all the error signals, such as si , i = 1, · · · , 4,
Proof With respect to k c1 ≤ y(0) ≤ k c1 and the z i , i = 2, 3, 4 and μ̃ are semi-globally, uniformly ulti-
choices of ka1 , kb1 , we have ka1 < s1 (0) = y(0)−yd < mately bounded. Furthermore, u(t) and αi , i = 1, 2, 3
kb1 . Denotes η = [s1 , s2 , s3 , s4 , ξ2 , ξ3 , ξ4 , μ̃]T ∈ R8 , are bounded.

123
3478 Y. Qiu, Z. Dai

0.25 1500

0.2

1000
0.15

0.1
500

0.05

0
0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

(a) Slip Ratio (b) Braking Torque

0.8 100

0.6

0.4 50

0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4

0.8 100

0.6

0.4 50

0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

(c) Friction Coefficient (d) Vehicle Velocity and Wheel Velocity

Fig. 5 Performances under road transition from dry asphalt surface (λd = 0.12) to wet asphalt surface (λd = 0.1) at t = 1s

From (28), we have: 5 Simulation results

(1 − q) k2 q k2 The objective of the proposed controller is to maintain


log 2 a1 2 + log 2 b1 2
2 ka1 − s1 2 kb1 − s1 the wheel slip not only at a desired level but also within
υ  −ct υ
≤ V (0) − e + . (29) a predetermined range in the presence of unknown fric-
c c tion. The properties and performances of the proposed
Taking exponentials on both sides of Eq. (29) and control scheme were investigated via a simulation study
rearrange the inequality with respect to the fact that by using MATLAB.
2 − s 2 > 0 and k 2 − s 2 > 0, it leads to the following
ka1 In this simulation, the vehicle velocity and the angu-
1 b1 1
inequality: lar velocity of the wheels are assumed to be measured
by sensors, thus the slip ratio can be calculated from
|s1 (t)| ≤ [(1 − q)ka1 + qkb1 ] these measurements. From Eq. (14) in Sect. 4, these

−ct measurements can be fairly enough to estimate the tire-
1 − e−2[(V (0)−υ/c)e +υ/c] , (30)
  road friction.
ka1 1 − e−2υ/c ≤ s1 (∞) ≤ kb1 1 − e−2υ/c . The initial velocity conditions are given as Vx (0)
=33.33 m/s (120 km/h), ωw (0) =102.56 rad/s, resulting
Therefore, s1 (t) can be made arbitrarily small as t →
in λ(0) = 0. Considering the reality, simulation studies
∞ by appropriate selection of the design parameters.
are conducted up to the point when the vehicle is slowed

123
Adaptive constrained antilock braking control 3479

0.14
1600
0.12 1400

0.1 1200

0.08 1000

0.06 800

600
0.04
400
0.02
200
0
0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 3.5

(a) Slip Ratio (b) Braking Torque

1
100
0.8

0.6

0.4 50

0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5

1
100
0.8

0.6
50
0.4

0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5

(c) Friction Coefficient (d) Vehicle Velocity and Wheel Velocity

Fig. 6 Performances under road transition from wet asphalt surface (λd = 0.1) to dry asphalt surface (λd = 0.12) at t = 1s

to about 1 m/s (3 km/h) as ABS deactivation velocity . Vx and the corresponding road condition vector C as
A 240W DC motor is used in this study, and the Eq. (31), where the parameters c1 , c2 , c3 and c4 of the
state constraints of the motor are fully considered as vector C used in this study are given in Table 3.
i m max = ±20 A, em max = ±24V , ωm max = ±5000 μ(λ, Vx ) = [c1 (1 − e−c2 λ ) − c3 λ]e−c4 λVx . (31)
rpm according to its physical saturations. TL is related
to the viscous friction between the torsional spring and It is clear critical to select the setpoints of wheel
the motor, which is usually proportional to the nor- slip under different road conditions. On the one hand,
mal force. In this simulation, TL is achieved through the optimal wheel slip differs under many factors, such
observers developed from our previous work. In addi- as the characteristics of tire-road contact, slip angle,
tion, the maximum braking torque that can be generated etc., which can be time-varying in some senses. On the
from the actuation system is limited to 2000 Nm. The other hand, a loss of performances (such as decreases of
values of the parameters in the single-corner model and the bandwidth ) may occur if the setpoint of λd moves
the DC motor are given in Table 2. further left of the peak, while the braking system will
A ground-contact model in [4] is introduced to sim- become difficult to be stabilized if the setpoint of λd
ulate the tire-road friction forces for different road con- moves sufficiently far to the right of the peak of μ − λ
ditions in this work. This model provides the longitudi- curves. λ ∈ (0.05, 0.15) is often used as the control
nal friction coefficient μ(λ, Vx ) with a function of the target area by many manufactures, and has been vali-
wheel slip λ, the longitudinal velocity of the vehicle dated as a good compromise for all road condition. In
this study, the desired wheel slip λd is selected as 0.12

123
3480 Y. Qiu, Z. Dai

0.35
1200
out of the stable region
0.3
1000
0.25 0.3
800
0.2 0.2
600
0.15
0.1

0.1 400
0.8 1 1.2 1.4 1.6 1.8
0.05 200

0
0
0 2 4 6 8 10 12 14 0 5 10 15

(a) Slip Ratio (b) Braking Torque

0.8

100
0.6

0.4
50
0.2

0 0
0 5 10 15 0 5 10 15

0.8

100
0.6

0.4
50
0.2

0 0
0 5 10 15 0 5 10 15

(c) Friction Coefficient (d) Vehicle Velocity and Wheel Velocity

Fig. 7 Performances under road transition from wet asphalt surface (λd = 0.1) to snow surface (λd = 0.05) at t = 1s

for dry asphalt surface, 0.1 for wet asphalt surface and employed to replace the utilization of barrier function
0.05 for snow surface, by maximizing the friction force while both the estimation of the friction coefficient
under the tire-road model of Eq. (31). and the DSC are incorporated into SMC in the same
As given in Sect. 4, we choose λh = 0.3 as the way as the proposed controller. Additionally, a satura-
upper constraint on the wheel slip, λl = λ(0) − 0.05 tion is employed in SMC to reduce the chattering phe-
as the lower constraint, respectively. Based limitations nomenon, and the parameters are tuned to minimize
on the gains in Sect. 4 (see Remark 3), the control the tracking error. In this study, SMC is designed as
gains not only determine the convergence of the algo- Eq. (32) with ks = 0.5, ε = 0.01, while other gains
rithm, but also be limited by the physical saturations remain the same as the proposed controller.
of the actuator, which are determined as k1 = 15, 1
k2 = 100, k3 = 80000 and k4 = 30. The time constants α1 = (−ks sε − μ̂φ1 (x1 ) + ẏd ),
g1
are chosen sufficiently small positive real constants as 
1 χ2
τ2 = τ3 = τ4 = 0.001. γ = 0.05 and kμ = 3000 are α2 = −k2 s2 − f 2 − g1 s1 − ,
g2 τ2
chosen for the estimation of the tire-road friction. It is  (32)
1 χ3
worth noting that same controller parameters are used α3 = −k3 s3 − f 3 − g2 s2 − ,
for five different scenarios. g3 τ3

Sliding-mode control (SMC) is adopted for a com- 1 χ4
u= −k4 s4 − f 4 − g3 s3 − ,
parison study. It has to be clarified that SMC is just g4 τ4

123
Adaptive constrained antilock braking control 3481

0.14 1600

0.12 1400

0.1 1200

0.08 1000

0.06 800

600
0.04
400
0.02
200
0
0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 3.5

(a) Slip Ratio (b) Braking Torque

1.2 140
16
14
1 120
12
100 10
0.8
8
2.9 2.95 3 3.05
80
0.6
60
0.4
40

0.2
20

0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5

(c) Friction Coefficient (d) Vehicle Velocity and Wheel Velocity

Fig. 8 Performances on dry asphalt surface (λd = 0.12) with sensor noise added to measurements of Vx and ωw

where time with a small residual set around the origin; at the

⎨1 s1 > ε same time, the wheel slip operates in the predetermined
sε = sε1 |s1 | ≤ ε range shown in Figs. 3a and 4a. Correspondingly, self-
⎩ locking is avoided, which can be observed in Figs. 3d
−1 s1 < −ε.
and 4d showing that the wheel angular velocity follows
Five simulation scenarios under two unchanging road the vehicle velocity closely without oscillations. It also
surface and three road transitions (depicted in Table 4) shows that the braking torque is smooth and stable dur-
are conducted, respectively, with regard to dry asphalt ing the braking process in Figs. 3b and 4b.
surface (Fig. 3), wet asphalt surface (Fig. 4), road tran- The performances under different road transitions
sitions from dry asphalt surface to wet asphalt surface are well-investigated (Figs. 5, 6 and 7). The road tran-
(Fig. 5), from wet asphalt surface to dry asphalt surface sitions are stimulated at t = 1s for each scenario by
(Fig. 6), and from wet asphalt surface to snow surface adopting the μ − λ model (Eq. 31), while the ref-
(Fig. 7). erence wheel slip ratio λd is switched corresponding
Under unchanging road surfaces (Figs. 3, 4), 3c to the changing road surface. Figures 5c, 6c and 7c
and 4c illustrate that the estimated friction coefficient show that an accurate estimation of friction coefficient
remains close to its true value with no oscillations can be achieved in the presence of model uncertain-
during the whole braking maneuver, while the fric- ties regardless of abrupt changes of road surfaces. Fig-
tion coefficient varies wildly and fast along with the ures 5a, 6a and 7a shows that the wheel slip reaches
wheel slip and the velocity of the vehicle under differ- its new reference value in less than 0.3s, the refer-
ent road conditions. With the help of the accurate esti- ence wheel slip ratios have been accurately tracked
mation of the friction coefficient, the adaptive control without producing drastic oscillations. In addition, the
can maintain the desired wheel slip in a short period of

123
3482 Y. Qiu, Z. Dai

transient responses for the road transition have been It can be concluded that satisfactory performance
adapted robustly within the operating region at each and fast convergence are achieved with smooth tran-
scenario even when the road has been changed into sient responses under each scenario no matter how the
the snow surface. The braking torques have changed road condition changes, which can be considered as a
quickly to compensate as the road condition changes. significant enhancement of driving comfort.
Tb decreased significantly ( Figs. 5b and 7b) when mov-
ing from high friction surface to lower friction one to
avoid locking; Tb increased significantly when mov- 6 Conclusion
ing from lower friction surface to higher friction one to
avoid rolling (Fig. 6b ). This advantage can also be seen We in this paper present a simple yet effective adap-
from Figs. 5d, 6d and 7d that both the vehicle velocity tive constrained ABS control for cars equipped with
and the wheel velocity do not experience characteristic electro-mechanical brakes with the following notable
oscillations during the braking maneuvers. features. Firstly, an ABLF incorporating with DSC is
The braking distance and the mean fully developed introduced to ensure wheel slip operate in the prede-
deceleration (MFDD) are depicted in Table 4, where termined region as well as to track the wheel slip ref-
the MFDD is the mean of V̇x in a time interval from erence with a steady state error in a small neighbor-
the vehicle decelerates from 0.9V0 to 0.05V0 [26]: hood. Secondly, an estimation of the friction is devel-
 0.9V oped with excellent accuracy and fast convergences,
MFDD = V̇x 0.05V0 . which provided the system an adaptive capability to
0

Compared with SMC, the proposed method has perform robustly against different road conditions and
achieved faster transient responses with shorter brak- even changing road conditions. Thirdly, the physical
ing distances and higher deceleration. In addition, the characteristics of the actuation system are fully inves-
braking system can operate in the predetermined sta- tigated both in the controller formulation and simula-
ble region with the proposed controller while wheel slip tion study. The boundedness of the output constraints
may move out of the region under a sharp transition, on wheel slip and the asymptotic stability are proven
such as from wet surface to snow surface with SMC mathematically. The simulation results have been pro-
(Fig. 7a), which confirms the effectiveness of casting vided to confirm the sufficient adaptive properties of
constraints on wheel slip under the proposed control the proposed controller.
scheme. It is important to stress that our aim was not to invent
Figure 8 shows the performances with uniform white a class of ABS that would be more efficient than those
noises that added to the measurements of Vx and ωw , that already exist in the literatures. Our aim was to
which can be seen in Fig. 8d. It is observed that the investigate, through a simplified but robust algorithm,
transient response of wheel slip oscillate at low speeds the practical behaviors during the braking maneuver,
due to respective low signal-to-noise ratio, which may such as the driving safety with a thorough consideration
result in an increasingly inaccurate calculation of wheel of wheel slip boundary and the driving adaptivity under
slip (Fig. 8a) . However, the estimation of friction coef- uncertainties. However, some practical issues have not
ficient shows great accuracy regardless of sensor noise been taken into full account at this stage. Thereby, con-
(Fig. 8c) and the controller shows robust ability with trol design based on advanced model subject to sys-
fast response. A set of simulation subject to the five sce- tem uncertainties and physical couplings is the focus of
narios have been completed with similar results, which our future work, such as (1) more sophisticated vehicle
are omitted here for the space limitation. model including its load distribution, latitude dynamics
It is worthy explaining that the braking torque Tb and active suspensions, (2) the coupling from longitu-
increases stably because the friction coefficient μ keeps dinal and latitude dynamics , (3) information fusion to
increasing (according to Eq. 31) as the vehicle deceler- improve data accuracy.
ates if λ is at its desired value (after the wheel slip are
accurately tracked).

123
Adaptive constrained antilock braking control 3483

Acknowledgements This work is supported by the National 14. Johansen, T.A., Petersen, I., Kalkkuhl, J., Ludemann, J.:
Nature Science Foundation of China (NSFC) Under Grant Gain-scheduled wheel slip control in automotive brake sys-
11802217. tems. IEEE Trans. Control Syst. Technol. 11(6), 799–811
(2003)
Data availability Data sharing not applicable to this article as 15. Kim, J., Park, C., Hwang, S., Hori, Y., Kim, H.: Control algo-
no datasets were generated or analysed during the current study. rithm for an independent motor-drive vehicle. IEEE Trans.
Veh. Technol. 59(7), 3213–3222 (2010)
Declarations 16. Krstic, M., Kanellakopoulos, I., Kokotovic, P.V.: Nonlinear
and Adaptive Control Design. Wiley, Hoboken (1995)
17. Lee, C., Hedrick, K., Yi, K.: Real-time slip-based estima-
Conflict of interest The authors declare that they have no con- tion of maximum tire-road friction coefficient. IEEE/ASME
flict of interest. Trans. Mechatron. 9(2), 454–458 (2004)
18. Lin, C.M., Hsu, C.F.: Neural-network hybrid control for
antilock braking systems. IEEE Trans. Neural Networks
14(2), 351–359 (2003)
References 19. Mi, C., Lin, H., Zhang, Y.: Iterative learning control of
antilock braking of electric and hybrid vehicles. IEEE Trans.
1. Ahn, C., Peng, H., Tseng, H.E.: Robust estimation of road Veh. Technol. 54(2), 486–494 (2005)
frictional coefficient. IEEE Trans. Control Syst. Technol. 20. Mirzaeinejad, H., Mirzaei, M.: A novel method for nonlinear
21(1), 1–13 (2011) control of wheel slip in anti-lock braking systems. Control.
2. Ahn, J., Jung, K., Kim, D., Jin, H., Kim, H., Hwang, S.: Eng. Pract. 18(8), 918–926 (2010)
Analysis of a regenerative braking system for hybrid electric 21. Ngo, K.B., Mahony, R., Jiang, Z.P.: Integrator backstepping
vehicles using an electro-mechanical brake. Int. J. Automot. using barrier functions for systems with multiple state con-
Technol. 10(2), 229–234 (2009) straints. In: Decision and control, 2005 and 2005 European
3. Bogdanovic, G., Meurk, A., Rutland, M.W.: Tip friction: tor- Control Conference. CDC-ECC’05. 44th IEEE Conference
sional spring constant determination. Colloids Surf. B 19(4), on, pp. 8306–8312. IEEE (2005)
397–405 (2000) 22. Oniz, Y., Kayacan, E., Kaynak, O.: A dynamic method to
4. Burckhardt, M.: Fahrwerktechnik: Radschlupf- forecast the wheel slip for antilock braking system and its
regelsysteme. Vogel-Verlag, Germany (1993) experimental evaluation. IEEE Trans. Syst. Man Cybern.
5. Dadashnialehi, A., Bab-Hadiashar, A., Cao, Z., Kapoor, Part B 39(2), 551–560 (2008)
A.: Intelligent sensorless abs for in-wheel electric vehicles. 23. Pasillas-Lépine, W.: Hybrid modeling and limit cycle analy-
IEEE Trans. Ind. Electron. 61(4), 1957–1969 (2013) sis for a class of five-phase anti-lock brake algorithms. Veh.
6. Dadashnialehi, A., Bab-Hadiashar, A., Cao, Z., Kapoor, A.: Syst. Dyn. 44(2), 173–188 (2006)
Intelligent sensorless antilock braking system for brushless 24. Patil, A., Ginoya, D., Shendge, P., Phadke, S.: Uncertainty-
in-wheel electric vehicles. IEEE Trans. Ind. Electron. 62(3), estimation-based approach to antilock braking systems.
1629–1638 (2014) IEEE Trans. Veh. Technol. 65(3), 1171–1185 (2015)
7. De Castro, R., Araújo, R.E., Freitas, D.: Wheel slip control 25. Peng, D., Zhang, Y., Yin, C.L., Zhang, J.W.: Combined con-
of EVs based on sliding mode technique with conditional trol of a regenerative braking and antilock braking system
integrators. IEEE Trans. Ind. Electron. 60(8), 3256–3271 for hybrid electric vehicles. Int. J. Automot. Technol. 9(6),
(2013) 749–757 (2008)
8. De Castro, R., Araujo, R.E., Tanelli, M., Savaresi, S.M., 26. Pretagostini, F., Ferranti, L., Berardo, G., Ivanov, V., Shy-
Freitas, D.: Torque blending and wheel slip control in evs rokau, B.: Survey on wheel slip control design strategies,
with in-wheel motors. Veh. Syst. Dyn. 50, 71–94 (2012) evaluation and application to antilock braking systems.
9. Dinçmen, E., Güvenç, B.A., Acarman, T.: Extremum- IEEE Access (2020). https://doi.org/10.1109/ACCESS.
seeking control of abs braking in road vehicles with lat- 2020.2965644
eral force improvement. IEEE Trans. Control Syst. Technol. 27. Qiu, Y., Liang, X., Dai, Z.: Backstepping dynamic surface
22(1), 230–237 (2012) control for an anti-skid braking system. Control. Eng. Pract.
10. Drakunov, S., Ozguner, U., Dix, P., Ashrafi, B.: Abs con- 42, 140–152 (2015)
trol using optimum search via sliding modes. IEEE Trans. 28. Ren, B., Ge, S.S., Tee, K.P., Lee, T.H.: Adaptive neural con-
Control Syst. Technol. 3(1), 79–85 (1995) trol for output feedback nonlinear systems using a barrier
11. Fujii, K., Fujimoto, H.: Traction control based on slip ratio Lyapunov function. IEEE Trans. Neural Networks 21(8),
estimation without detecting vehicle speed for electric vehi- 1339–1345 (2010)
cle. In: 2007 Power conversion conference-Nagoya, pp. 29. Savitski, D., Ivanov, V., Augsburg, K., Emmei, T., Fuse,
688–693. IEEE (2007) H., Fujimoto, H., Fridman, L.M.: Wheel slip control for the
12. Gao, Y., Ehsani, M.: Electronic braking system of ev and electric vehicle with in-wheel motors: variable structure and
hev–integration of regenerative braking, automatic braking sliding mode methods. IEEE Trans. Ind. Electron. 67(10),
force control and abs. SAE Trans. 110, 576–582 (2001) 8535–8544 (2020)
13. Ivanov, V., Savitski, D., Shyrokau, B.: A survey of traction 30. Shyrokau, B., Wang, D., Augsburg, K., Ivanov, V.: Vehicle
control and antilock braking systems of full electric vehicles dynamics with brake hysteresis. Proc. Inst. Mech. Eng. Part
with individually controlled electric motors. IEEE Trans. D J. Autom. Eng. 227(2), 139–150 (2013)
Veh. Technol. 64(9), 3878–3896 (2014)

123
3484 Y. Qiu, Z. Dai

31. Subudhi, B., Ge, S.S.: Sliding-mode-observer-based adap- 36. Unsal, C., Kachroo, P.: Sliding mode measurement feedback
tive slip ratio control for electric and hybrid vehicles. IEEE control for antilock braking systems. IEEE Trans. Control
Trans. Intell. Transp. Syst. 13(4), 1617–1626 (2012) Syst. Technol. 7(2), 271–281 (1999)
32. Sulzer, J.S., Peshkin, M.A., Patton, J.L.: Marionet: An 37. Zhang, C., Ordóñez, R.: Numerical optimization-based
exotendon-driven rotary series elastic actuator for exerting extremum seeking control with application to ABS design.
joint torque. In: Rehabilitation robotics, 2005. ICORR 2005. IEEE Trans. Autom. Control 52(3), 454–467 (2007)
9th International conference on, pp. 103–108. IEEE (2005)
33. Swaroop, D., Hedrick, J.K., Yip, P.P., Gerdes, J.C.: Dynamic
surface control for a class of nonlinear systems. IEEE Trans. Publisher’s Note Springer Nature remains neutral with regard
Autom. Control 45(10), 1893–1899 (2000) to jurisdictional claims in published maps and institutional affil-
34. Tanelli, M., Astolfi, A., Savaresi, S.M.: Robust nonlinear iations.
output feedback control for brake by wire control systems.
Automatica 44(4), 1078–1087 (2008)
35. Tee, K.P., Ge, S.S., Tay, E.H.: Barrier Lyapunov functions for
the control of output-constrained nonlinear systems. Auto-
matica 45(4), 918–927 (2009)

123

You might also like