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

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

Two-time-scale redesign for anti-lock braking

systems of ground vehicles
Weichao Sun, Member, IEEE, Jinhua Zhang, Zhiyuan Liu

Abstract—Anti-lock braking system (ABS) is one of the most There have been abundant researches focusing on ABS
effective active safety control systems for ground vehicles, since control scheme design, and most of their conceptions are
it can keep the rotational wheel from locking, and consequently novel and remarkable. In [8], a sliding mode controller is
guarantee the braking safety and handling stability. There have
been a variety of ABS control schemes proposed by many re- employed for ABS, where the typical µ − λ friction model
searchers. However, most results employ sundry tire-road friction developed by Burckhardt and Reimpell is utilized, and a
models, the alleged µ − λ curves (µ is tire-road friction coefficient, three point prediction method to predict the road condition
while λ is tire slip ratio, which is mathematically defined with is proposed. R. Verma, etc. develop a multiple surface sliding
equation (4)), making the ABS controller extremely complicated control with [9], involving a polynomial fitting µ − λ friction
for the highly non-linear characteristics of µ − λ relationship.
Furthermore, the priori knowledge of road conditions for these curve, and the effectiveness is tested via a quarter car ABS
ABS controllers restricts their practicability. To circumvent these bench. The most commonly used Pacejkas magic formula
problems, a two-time-scale ABS control scheme is proposed in µ − λ tire-road adhesion model is utilized in [10], where both
this paper, without considering the intricate µ − λ relationship, linear and nonlinear model predictive control approaches are
making the priori knowledge of road condition no longer a developed. The simulation with high fidelity CarMaker verifies
prerequisite, thus the designed ABS controller is rather simple.
More than that, a modified fast-time-scale estimator is involved the feasibility of the controller. J. Yi, etc. employ an extremely
to estimate the road condition, which is significant in vehicle complicated LuGre friction model to establish an observer-
active dynamics control. The effectiveness of the proposed ABS based braking controller [11]. Although the controller can
controller is verified via numerical simulations and CarSim- maximize the tire-road friction coefficient, the style of both the
Matlab co-simulations. controller and the observer is rather intricate. In [12], existence
Index Terms—Anti-lock braking systems, two-time-scale re- and stability of limit cycles are analyzed via perturbation
design, nonlinear control. theory, and just as [8], the same slightly simple µ − λ friction
model is utilized. Sliding mode controllers are involved in both
I. I NTRODUCTION [1] and [13], but different road friction models are utilized.
EHICLE chassis active control is of great significance The same polynomial fitting µ − λ friction curve is involved
V to improve the vehicle safety performance and to reduce
traffic accidents as well as casualties. As the rapid develop-
with [1], whereas the typical Pacejkas magic formula µ − λ
with a corresponding road friction estimator based on change
ment of electronic control systems in recent decades, vehicle rate of µ is used with [13]. Similarly, [14] and [15] utilize
chassis active control has developed into different kinds of the typical Burckhardt-and-Reimpell developed µ − λ friction
types, including anti-lock braking control[1][2], electronic model to construct a backstepping dynamic surface controller
stability programme[3][4], active suspension system[5][6], etc. and an optimal predictive controller respectively. In addition,
In a braking process of a ground wheeled vehicle, if other control methods, such as such as gain-scheduled control
the maximum braking force is exerted, the wheel is likely [16], switching control [17], etc [18], have also been explored
to be locked, which means the wheel totally slips without for ABS controller design.
any rotation. Under the locked condition, the wheel cannot Comprehensive analysis to the aforementioned studies indi-
execute steering manoeuvre, also the braking distance would cates that the ABS research has achieved tremendous progress
be lengthened. Hence it’s an extremely hazardous condition, for a few decades from the simple on-off valve control
and the anti-lock braking control becomes indispensable. As to adaptive robust control which can address the external
one of the most significant devices for vehicle chassis active uncertainties as well as unknown system parameters. Be-
control[3][7], the main principle of an ABS is to brake the sides, various µ − λ models have emerged, from the most
rotational wheel with proper torque, or force essentially, to simple to the most complicated, covering Burckhardt-and-
keep the wheel from locking, and consequently to decrease the Reimpell developed µ − λ friction model, polynomial fitting
braking distance and maintain the wheels’ steering capability. µ − λ friction model, Pacejkas magic formula µ − λ friction
model and LuGre friction model. Nonetheless, the developed
This work was supported in part by the National Natural Science Foundation µ − λ friction models are extremely complicated. Furthermore,
of China (No. 61773135, 61790564), in part by China Automobile Industry
Innovation and Development Joint Fund (No. U1564213). the identification of the µ − λ model parameters is time-
Weichao Sun and Jinhua Zhang are with the Research Institute of Intelligent consuming, and the priori information of road condition is a
Control and Systems, School of Astronautics, Harbin Institute of Technology, challenge as well. Hence, is there a rather expedient approach
Harbin 150001, China.
Zhiyuan Liu is with the Department of Control Science and Engineering, to construct the ABS controller?
Harbin Institute of Technology, Harbin 150001, China. On the other hand, time-scale separation designs to recover

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

nominal performance of the closed-loop systems with uncer- is braking torque, m is the quarter vehicle mass, v is vehicle
tainties have been studied theoretically. The design philosophy velocity, g is gravitational acceleration, and Fz is the wheel
is to build a high-gain filter to estimate the unknowns over a vertical load respectively.
fast time-scale, then the control evolves over the slow time-
scale and recovers the nominal performance. X. Fan, etc.
[19] design a two-time-scale network congestion controller for
edge-based detection and rectification of uncooperative flows.
Time-scale separation design schemes with dynamic inversion
for non-affine in control systems are given with [20] and [21].
Furthermore, a two-time-scale and three-time-scale redesigns
considering the performance recovery are proposed for nonlin-
ear systems [22]. The corresponding engineering applications
have also been explored. A two-time-scale robust adaptive
control scheme is illustrated for a single-machine infinite-bus Fig. 1: quarter vehicle braking model
power system with an static var compensator in [23]. Also in
power systems, A. Chakrabortty, etc. [24], apply the time-scale The road braking force Fx is essentially the friction between
separation design method to performance recovery of system tire and road, which is given as
with faults and unknown parameters. Furthermore, the two-
Fx = Fz µ (λ , µH , T, v, α ), (3)
time-scale control scheme is also introduced to the tracking
problem of wheeled mobile robots [25]. Apart from these, where µ is the friction coefficient, which is a nonlinear
there exit many other applications in engineering areas, which function with respect to tire slip ratio λ , and the function curve
will not be elaborated. is intricately connected with road condition µH , tire temper-
Since the time-scale separation design can effectively ad- ature T , wheel side slip angle α as well as vehicle velocity
dress unknowns, the design philosophy is employed to settle v simultaneously. Their typical relationship is illustrated in
the road condition uncertainties in this paper. Based on the figure 2. In addition, the tire slip ratio is defined as
aforementioned analysis and the valuable results of these re-
v − ωr
searches, a two-time-scale redesign scheme for ABS controller λ= , (4)
v
of ground vehicles is proposed. The main contributions of this
paper are summarized as follows: i) Theoretically, a novel two- which represents the slip degree of a wheel motion in a vehicle
time-scale ABS controller is proposed, where the complicated braking manoeuvre: λ = 0 indicates there is no slip, and the
highly nonlinear µ − λ tire-road friction model is not involved, wheel is at a pure rotation state; λ = 1 indicates there is no
which saves the time consumption of tedious parameter identi- rotation, and the wheel is at a pure slip state, namely locked.
fication of µ − λ model. Also a modified estimator is develope- Figure 2 indicates to specifically formulate µ (λ , µH , T, v, α )
d to estimate the road condition. ii) Practically, there are only is a challenge. However, we can roughly obtain some valuable
two parameters need to be determined for the controller, which conclusions: i) µ is nonlinearly dependent on λ , but µ has a
facilitates the adjustment of controller parameters. Besides, peak value with λ around 0.15; ii) The derivative of µ with
The developed two-time-scale ABS controller can be adaptive respect to λ is smooth and has an upper bound, namely ∂∂ λµ ≤
to different road conditions, hence the priori knowledge of L1 . Conclusion i) is the reason that the reference value of λ is
road information is not a prerequisite. mostly taken as 0.15, while the conclusion ii) will be utilized
The rest of this paper is organized as follows. The wheel in the following controller design.
braking dynamics is formulated in section II. Then the two- Combining equation (1)∼(4) yields
time-scale ABS controller design and stability analysis are {
λ r2
λ̇ = −( 1−
vm + vJ )Fz µ (λ , µH , T, v, α ) + vJ Tb
r
presented in section III. In section IV-A and IV-B, numerous Fz (5)
numerical simulations and CarSim-Matlab co-simulations are v̇ = − m µ (λ , µH , T, v, α ),
carried out respectively to verify the effectiveness of the
where λ̇ and v̇ represent the time derivative of λ and v
proposed ABS controller, followed by conclusion in section
respectively. Since the vehicle velocity dynamics evolves on a
V.
much slower time scale than the wheel slip dynamics, and the
control target is to regulate wheel slip, only the first equation
II. P ROBLEM F ORMULATION of (5) is taken into consideration. Since m ≫ J, then we can
During the wheel braking process, a quarter vehicle can be have 1/m ≪ r2 /J [17]. Hence the wheel slip dynamics can be
modeled as figure 1, and the corresponding vehicle dynamics simplified and recast as
is formulated as r
λ̇ = (Tb − rFz µ (λ )), (6)
J ω̇ = rFx − Tb (1) vJ
mv̇ = −Fx , (2) where µ (λ ) is abbreviated from µ (λ , µH , T, v, α ) for concise-
ness.
where J is the wheel rotational inertial, ω is wheel angular Notice that the notation µ (λ ) is utilized with the vehicle
speed, r is wheel rolling radius, Fx is road braking force, Tb dynamics formulation, but different from the aforementioned

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

tremely intricate µ − λ model, a faster time scale estimator is

proposed to estimate different road conditions.
Since µ (λ ) in equation (6) is difficult to precisely model,
± the ABS controller design is composed of the following two
procedure: firstly, the nominal controller is designed without
considering the µ (λ ) term; then, an ABS controller redesign
is conducted with estimating the µ (λ ) term via a faster time-
° scale filter.
(a) different road condition [8] Firstly, leaving out term ’-rFz µ (λ )’, the slip dynamics in
equation (6) degrades to
r
λ̇ = Tb . (7)
vJ
Let e = λ − λr , where e is the tracking error of λ to its
±

reference value λr for the degraded slip dynamics (7). Then the
error dynamics can be easily obtained as ė = λ̇ − λ̇r = vJr Tb − λ̇r
with equation (7). Choose V1 = 21 e2 as a Lyapunov candidate,
and design
° vJ
(b) different temperature [16]
Tbnom = (λ̇r − k1 e), (8)
r
which yields V̇1 = −k1 e2 < 0, where k1 > 0 is a parameter to
be designed.
Then, take the term ’-rFz µ (λ )’ into consideration, and
denote it as δ (λ ) = −rFz µ (λ ). To estimate this unmodeled
±

term, an estimator is designed as

vJ 1
δ̂ (λ ) = − (λ̂ − λ ), (9)
r ε
°
with the filter
(c) different side slip [16]
˙ r 1
λ̂ = Tb − (λ̂ − λ ), λ̂ (0) = λ (0), (10)
vJ ε
where ε > 0 is a small positive parameter to be determined.
Claim. 1 The following proposed ABS controller
±

Tb = Tbnom − δ̂ (λ ), (11)

with (8), (9) and (10) can result in an asymptotically tracking

° to its reference value for slip dynamics (6), with all the signals
(d) different velocity [26] involved being available.
Fig. 2: friction coefficient curves with different conditions Proof: Denote l := λ̂ −ε λ , and with (6) and (10), it is easily
to have
r
ε l˙ = −l − δ (λ ), l(0) = 0, (12)
literature, the complicated µ − λ curve is not established. vJ
Hence the controller can be designed in a rather simple ap- which indicates l has a faster time-scale evolution than that of
proach, and consequently the controller parameters are less and λ with a small positive ε and evolves to a small neighborhood
easy to adjust. Furthermore, a modified estimator is given to of the manifold
obtain road conditions. The detailed two-time-scale controller r
l = − δ (λ ). (13)
design to regulate λ to its reference value will be elaborated vJ
in the next section.
Denote the off-manifold variable η := l + vJr δ (λ ), then we can
have δ̂ (λ ) → δ (λ ), if η → 0.
III. TWO - TIME - SCALE ABS CONTROLLER DESIGN Redefine e = λ − λr as the tracking error of λ to its reference
The control objective of the two-time-scale ABS controller value λr for the slip dynamics (6). Accordingly, the error
is to regulate tire slip ratio λ to its reference value λr , thus dynamics can be obtained as
to prevent the rotational wheel from locking and accordingly r
to improve the braking performance. To circumvent the ex- ė = (Tb − rFz µ (λ )) − λr . (14)
vJ

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

Substituting the proposed controller (11) with (9) into the determined, and a sufficient large k1 and a sufficient small
above error dynamics, and noticing the definition of l and ε would be favorable. Nonetheless, some conservativeness
η yields would exist, since the value ε is determined by L2 . By the
r vJ vJ way, the computational complexity is rather low for the simple
ė = ( (λ̇r − k1 e) + l + δ (λ )) − λ̇r wheel braking model and proposed controller.
vJ r r
r
= −k1 e + l + δ (λ )
vJ IV. E FFECTIVENESS V ERIFICATION
= −k1 e + η . (15) A. Numerical Simulation
The dynamics of the off-manifold variable satisfies To verify the effectiveness of the proposed controller, both
∂ ( vJr δ (λ )) numerical simulations and Matlab-CarSim coalition simula-
ε η̇ = ε l˙ + ε λ̇ . (16) tions are performed. The numerical simulations are presented
∂λ
in this subsection, while the Matlab-CarSim coalition simula-
Under the weak condition of λ̇r = 0, the above η dynamics tions will be presented in the next subsection. The parameters
can be rewritten as of a quarter hatchback in the simulations are listed in table
∂ ( −r
2
I, and the parameters of the proposed controller are listed in
vJ Fz µ (λ ))
ε η̇ = −η + ε ė. (17) table II. In this simulation the widely used tire-road friction
∂λ
2
∂ ( −r F µ (λ )) TABLE I: Parameters of a quarter hatchback
Denote σ (λ ) = vJ ∂ λz . According to its physical mean-
ing and the conclusion 2) in section II, we can have |σ (λ )| ≤ Parameter m R J
L2 , where L2 is its upper bound. Thus Value 1230/4 kg 0.31 m 0.6 kg·m2
1
η̇ = − η + σ (λ )ė. (18)
ε
TABLE II: Controller parameters
Choose V2 = V1 + η2 as a Lyapunov candidate, and its
2

derivation along the e dynamics (15) and the η dynamics (18) Parameter Value Parameter Value
k1 1000 ε 0.001
leads to
V̇2 = eė + η η̇ model
1
= −k1 e + eη − η 2 + ησ (λ )(−k1 e + η )
2 µ (λ ) = c1 (1 − e−c2 λ ) − c3 λ (21)
ε
1 2 1 2 1 2 is employed, where c1 , c2 , c3 are parameters corresponding to
≤ −k1 e + e + η − η
2
different road conditions. In addition, the hatchback velocity is
2 2 ε
1 2 1 initialized as v(0) = 10 m/s with ω (0) = 10/0.31 rad/s, while
+ k1 e + k1 L22 η 2 + L2 η 2 the others are taken 0. The sampling time is set 0.001s and λr
2 2
k1 1 2 1 1 1 is taken 0.15 as most ABS controllers. What is noteworthy is
= −( − )e − ( − − k1 L22 − L2 )η 2 , (19) that from the slip dynamics (6), the slip would go infinity if the
2 2 ε 2 2
hatchback velocity approaches 0, hence the ABS is switched
which indicates V̇2 < 0 for k1 > 1 and 0 < ε < 1
1 + 1 k L2 +L .
2 2 1 2 2 off once the the velocity is smaller than 0.2 m/s.
Namely, under the proposed controller, λ can track its refer- 1) Low µ Road Condition: firstly, a braking simulation
ence value λr asymptotically. This completes the proof. on a low µ cobble wet road (c1 = 0.4004, c2 = 33.708,
Under the proposed controller, the priori knowledge of c3 = 0.1204) is carried out. To comparatively demonstrate the
road conditions is not a prerequisite to the two-time-scale effectiveness of the proposed controller, the braking process of
ABS controller, which makes itself adaptive to different road under proposed two-time-scale ABS controller, PID controller
surface. Not only that, the road condition µ can be estimated (with proportion, integration and differentiation coefficient
via a simple modification from the estimator (9) as being kP = 1000, kI = 50, kD = 100 respectively) and no ABS
1 vJ 1 controller, namely the wheel brakes with a constant braking
µ̂ (λ ) = − δ̂ (λ ) = 2 (λ̂ − λ ). (20)
rFz r Fz ε movement of 2000 N·m, are performed comparatively. The
simulation results are presented in figure 3∼7.
Remark 1 Although the proposed controller tactfully circum-
The braking distance is illustrated in figure 3. The station
vent the intricate tire-road modelling, the full state signals
with the proposed ABS controller is 13.45 m, 4.78 m shorter
are necessary, especially for the vehicle velocity. This would
than that of 18.23 m without ABS controller, and 0.17 m
be a limitation to applications on inferior cars without global
shorter than that of 13.62 m with PID controller. Therefore,
position system. Moreover, other control methods, like sliding
an evident decrease of braking distance can be guaranteed with
mode control, fuzzy logic control, can also handle the system
the proposed ABS controller to improve the braking safety.
uncertainties, and corresponding researches would be explored
The wheel velocities are presented in figure 4, where the
further.
notation ω · R in the legend is wheel equivalent linear velocity,
Discussion 1 It should be noticed that the proposed controller representing the wheel rotational speed. From figure 4 we
is rather simple, since only two parameters k1 and ε are to be can see the braking time for proposed ABS controller is 2.6

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

20
no controller
part. On the one hand, k1 is feedback gain for trace error,
PID controller thus a larger k1 can result in a faster convergence rate; on the
15 proposed controller
other hand, ε determines the time scale of estimator 9, thus
station (m)

10 the smaller ε is, the faster the estimated error converges, thus
13.6
a smaller ε can render a preferable trace performance.
5 13.4
2.85 2.9 2.95
0.05
0
0 0.5 1 1.5 2 2.5 3 3.5 4
t (s) 0

-0.05 0
Fig. 3: braking distance (solid: proposed ABS controller;

e
dashed: without ABS controller; dotted: PID controller) -0.1 -0.05
-0.1
-0.15
0.1 0.2 0.3 0.4
-0.2
s, which is 1 s shorter than 3.7 s of no ABS controller, 0 0.5 1 1.5 2 2.5 3 3.5 4
and 0.1 s shorter than 2.7 s of PID controller. Furthermore, t (s)
we can observe the wheel rotational speed without ABS Fig. 6: trace error of slip ratio (solid: determined k1 , ε value;
controller declines rapidly to 0, namely locked, once the dashed: k1 decrease; dotted: ε increase)
braking process starts, which can result in a steering disability,
an extreme hazardous condition. Nonetheless, the designed
As equation (20) indicates, the road condition can be
ABS controller can exempt the wheel from locking.
estimated using the fast-time-scale estimator as in figure 7.
The estimated value of µ is 0.38 with λ = 0.15, which is
10
v (no controller) exactly corresponding to that of figure 2a.
R (no controller)
8 v (PID controller)
R (PID controller)
0.4
v (m/s)

6 v (proposed controller)
R(proposed controller)
4 0.3

2
0.2
0
0 1 2 3 4 5 6 7 0.1
t (s)

Fig. 4: wheel velocity (solid: linear longitudinal velocity of 0

0 0.5 1 1.5 2 2.5 3 3.5 4
wheel center; dashed: angular velocity of wheel; +: without t(s)
ABS controller; ⋆: proposed ABS controller; ◦: PID based Fig. 7: estimation of µ
ABS controller)
2) Opposite µ Road Condition: the following part presents
Figure 5 shows the wheel slip ratio under the braking a braking simulation with opposite µ road condition(0 ∼ 0.5 s:
process. Again, we can see the proposed ABS controller can dry concrete road with c1 = 1.1973, c2 = 25.168, c3 = 0.5373;
regulate λ to its reference value of 0.15 faster than PID 0.5 ∼ 7 s: snow road with c1 = 0.1946, c2 = 94.129, c3 =
controller, while the slip ratio without ABS controller rises 0.0646). The results are presented in figure 8∼ 12. Notice
to 1 and remains, which is corresponding to be locked. that all the parameters of the hatchback and the controller are
kept the same with those of subsection IV-A1, except for the
road conditions.
1 The braking distance is illustrated in figure 8. The braking
0.2
0.8 station under the proposed ABS controller is 9.70 m, which is
no controller
0.6 0.1 PID controller 12.28 m shorter than 21.98 m without ABS controller and 2.36
proposed controller
0.4 0 m shorter than PID controller with 12.06 m, indicating that
0.1 0.2
0.2 the proposed two-time-scale ABS controller can considerably
0
improve the braking safety.
0 0.5 1 1.5 2 2.5 3 3.5 4 The wheel velocity curves are presented in figure 9, from
t (s)
which we can see the braking time with the proposed ABS
Fig. 5: wheel slip ratio (solid: proposed ABS controller; controller is 2.97 s, 2.82 s shorter than the 5.79 s without
dashed: without ABS controller; dotted: PID controller) ABS controller and 0.54 s than PID controller with 3.51 s.
Besides, the wheel rotational velocity decreases to and keeps
The trace error of wheel slip ratio with different control 0 along the braking period, namely locked. In contrast, the
parameters for proposed ABS controller is given in figure 6. proposed controller can keep the wheel from locked. Notice
According to figure 6, the convergence rate of e increases as that the break point at 0.5 s is caused by the change of road
k1 increases and ε decreases, which is in line with theoretical condition.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

25 0.1
no controller
PID controller
20
proposed controller
station (m)
0
15
0

e
10 -0.01
-0.1
-0.02
5
0.05 0.1 0.15
0 -0.2
0 1 2 3 4 5 6 7 0 0.5 1 1.5 2 2.5 3 3.5 4
t (s) t (s)

Fig. 8: braking distance (solid: proposed ABS controller; Fig. 11: trace error of slip ratio (solid: determined k1 , ε value;
dashed: without ABS controller; dotted: PID controller) dashed: k1 decrease; dotted: ε increase)

X: 0.228

10 Y: 1.088

v (no controller) 1
R (proposed controller)
8
v (PID controller)
R (PID controller)
v (m/s)

6 v (proposed controller)
R (proposed controller) 0.5
4
X: 1.658
Y: 0.1848

2
0
0 0 1 2 3 4 5 6 7
0 1 2 3 4 5 6 7 t (s)
t (s)
Fig. 12: estimation of µ
Fig. 9: wheel velocity (solid: linear longitudinal velocity of
wheel center; dashed: angular velocity of wheel; +: without
ABS controller; ⋆: proposed ABS controller; ◦: PID controller)
B. CarSim-Matlab Co-simulation
CarSim can guarantee a high fidelity for vehicle dynamics
and provide abundant of test environments. Thus CarSim-
Figure 10 presents the slip ratio states. Apparently, the pro- Matlab co-simulation is carried out to verify the practicability
posed controller can desirably faster regulate the slip ratio than for the proposed ABS controller. The ABS controller is
PID controller, while that without ABS controller increases to established in Matlab-Simulink, and then the control signal,
1 to be locked rapidly. Similarly, the chattering of the curves wheel baking torque, is feed into CarSim, while after each
at 0.5 s is caused by the change of road condition, while iteration, the vehicle states, including vehicle velocity and
the chattering around 3 s is the result of switch-off of ABS wheel rotational speed, are feedback into Matlab.
controller. As numerical simulations in section IV-A, the same hatch-
back is selected in the CarSim test environment.The vehicle
parameters are entirely the same as table I, except m, since
1 a complete vehicle model is involved in CarSim, namely
0.8
m = 1230 kg. The controller parameters are listed in table
III. Similarly, the reference slip ratio is set 0.15, and the ABS
0.6
is switched off once the vehicle velocity is less than 1 km/h.
0.4
no controller
0.2 PID controller TABLE III: Controller parameters
proposed controller
0
0 1 2 3 4 5 6 7 Parameter Value Parameter Value
t(s) k1 100 ε 0.01
Fig. 10: wheel slip ratio (solid: proposed ABS controller;
dashed: without ABS controller; dotted: PID controller) 1) High µ Road Condition: in this braking test, a high µ
road condition with µH = 0.85 is specified. In this procedure,
the hatchback brakes at an initial speed of 100 km/h. To
The trace error of wheel slip ratio with different control comparatively show the superiority of the two-time-scale ABS
parameters for proposed ABS controller under opposite µ road controller, the test is performed with proposed controller,
is given in figure 11. Similarly we can obtain the convergence PID controller (with proportion, integration and differentiation
rate of e increases as k1 increases and ε decreases. coefficient being kP = 10000, kI = 10, kD = 1 respectively)
Finally, the road condition estimation µ̂ is given in figure and without controller, which means the hatchback brakes at
12. During the time period 0 ∼ 0.5, µ̂ = 1.088; whereas during a constant torque of 2250 N·m for front wheels and 1500 N·m
the period 0.5 ∼ 3, µ̂ = 0.185 with λ = 0.15. It is easy to verify for rear wheels as default. The other parameters for the test
the correctness of the estimator via equation (21), noting the environment are default. The simulation results are presented
road conditions are unknown in advance. in figure 13∼ 17.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

The braking distance is illustrated in figure 13. The station rear right (proposed controller)
without ABS controller is 53.52 m and with PID controller is 1 front right (proposed controller)
rear right (PID controller)
50.29 m, whereas the station with proposed ABS controller is 0.8 front right (PID controller)
46 m, which shortens the braking distance 7.52 m and 4.29 0.6

m respectively. Hence figure 13 can obviously demonstrate 0.4

the effectiveness of the designed controller in terms of safety 0.2

improvement. 0
0 0.5 1 1.5 2 2.5 3 3.5 4
t(s)
60
Fig. 15: wheel slip ratio (solid: front right wheel with proposed
controller; dashed: rear right wheel with proposed controller;
station (m)

40
dotted:rear right wheel with PID controller; dashed and dotted:
front right wheel with PID controller.)
20
no controller
proposed controller
PID controller
0
0 1 2 3 4 5 0
t (s)

Fig. 13: braking distance (solid: proposed ABS controller; -0.1

e
dashed: without ABS controller; dotted: PID controller.)
-0.2
The wheel velocity is given with figure14, where the wheel
linear velocity and rotational velocity of the front left wheel is 0 1 2 3 4 5
t (s)
only presented for concisely illustrating the wheel evolution.
As the same, we can observe that the proposed controller can Fig. 16: tracing error of front right wheel (solid: determined
keep the rotational wheel from being locked, thus shorten the k1 , ε value; dashed: k1 decrease; dotted: ε increase.)
braking period.

100 are reasonable. Furthermore, the inconsistency between front

R(no controller)
v(no controller) and rear wheel could be caused by axle weight transfer in the
80
velocity (km/h)

R(proposed controller)
v(proposed controller)
braking procedure.
60
R(PID controller)
v(PID controller)
40
proposed controller
20 PID controller 1
no controller
0
0 1 2 3 4 5
t (s) 0.5

Fig. 14: velocity of front left wheel (solid: linear longitudinal rear right wheel
velocity of wheel center; dashed: angular velocity of the front right wheel
0
wheel; red: proposed ABS controller; green: PID controller; 0 0.5 1 1.5 2 2.5 3 3.5
t(s)
blue: no controller.)
Fig. 17: estimation of µ (solid: front right wheel; dashed: rear
Figure 15 demonstrates the wheel slip ratio, where only the right wheel.)
right wheels’ curves are given, since the data of right wheels’
and left wheels’ are the same for symmetry. From figure 15, 2) Split µ Road Condition: a rather complicated split µ
the wheel slip ratio is regulated around 0.15 with proposed road condition (left wheels: µH = 0.8; right wheels: µH = 0.5)
controller more precisely than PID controller to guarantee a is involved in this braking test procedure. Similarly, the hatch-
rather road braking force. Notice that the declination of λ back runs both with and without the proposed ABS controller
for proposed ABS controller is caused by singularity when v to comparatively verify the practicability. When the ABS is
approaches 0, and this is also the reason why to switch off ABS switched off, the hatchback brakes with 2250 N·m and 1500
controller if v < 1 km/h mentioned before. Also the tracing N·m as default. In the braking test, the hatchback brakes at an
error for single front right wheel is illustrated in figure 16, initial speed of 65 km/h, and the proposed ABS controller is
and it can be obtained that the convergence error e decreases switched off until v < 1km/h. The other parameters for the test
as k1 increases and ε decreases similarly. environment are default. The simulation results are elaborated
The estimation of road condition µ is presented in figure with figure 18∼24.
17, from which we can see that the estimated value is around 1 The braking distance is illustrated in figure 18. It seems that
with λ = 0.15, bigger than 0.85. An investigation of figure 2a the proposed ABS controller could barely improve the braking
shows µ (λ = 0.15) > µH = µ (λ = 1), Therefore, the results safety, since little distance is reduced under the controller. In

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

fact, the ABS controller can considerably enhance the braking 0.2

safety, which will be explained with following figures.

0.1

30 0
proposed controller
no controller
front left wheel
PID controller
-0.1 rear left wheel
station (m)

20
front right wheel
rear right wheel
26 -0.2
0 0.5 1 1.5 2 2.5 3
10
t(s)
24
2 2.5 3 Fig. 20: wheel slip ratio (solid: front left wheel; dashed: rear
0
0 1 2 3 4 5 left wheel; dotted: rear right wheel; dash-dotted: front right
t (s) wheel.)
Fig. 18: braking distance (solid: proposed ABS controller;
dashed: without ABS controller; dotted: PID controller.) 0.1

Figure 19 demonstrates the evolution of front-left-wheel 0

velocities. Also, the wheel without ABS controller is locked,

-0.1

e
while that with the proposed ABS controller is kept from
locking. Besides, we can find the wheel under PID controller -0.2
is also locked at the end of braking procedure. Furthermore,
we can see the velocity without ABS controller evolves to -0.3
0 1 2 3 4 5
a negative value, which indicates the hatchback undergoes t(s)
a violent spin, causing a terrible heading alteration. This is Fig. 21: tracing error of front right wheel (solid: determined
an extremely dangerous condition, whereas, fortunately the k1 , ε value; dashed: k1 decrease; dotted: ε increase.)
designed ABS controller can help the hatchback avoid this.

60
R(proposed controller)
v(proposed cotroller)
can be observed for the yaw angle without controller, which
further indicates the hatchback has the heading orientation
velocity (km/h)

R(no controller)
v(no cotroller)
40 propopsed controller
R(PID controller) totally reversed. This condition is extremely detrimental, since
v(PID cotroller) the vehicle has lost the directional stability. In contrast, the
20
no controller PID controller proposed controller can keep the yaw angle within a smaller
0 value compared to that under PID controller.
0 1 2 3 4 5
The estimation of µ is given with figure 24, from which
t (s) we can observe that µ̂ is around 0.8 for left wheel and 0.5
for right wheel, which are corresponding to the set of road
Fig. 19: velocity of front left wheel (solid: linear longitudinal
condition. Notice the road condition is conceived unknown in
velocity of wheel center; dashed: angular velocity of the
advance, hence the estimator could approximately estimate the
wheel; red: proposed ABS controller; blue: without ABS
road condition with λ = 0.15.
controller; green: PID controller.)
To visually explain the effectiveness of the proposed con-
The state λ of four wheels and its tracing error e of front troller, the braking performance (braking time and braking
right wheel are given in figure 20 and 21. From figure 20 distance) with proposed controller, PID controller and no
we can observe that the wheel slip ratio is regulated to its controller under different road environment is given in table
reference value 0.15. The drastic decline, when the hatchback IV, where we can intuitively obtain the superiority of two-
approaches to stop, is induced by singularity as mentioned
before, then the ABS controller is switched off. In addition,
1
lateral distance to path (m)

figure 21 indicates the convergence error of e decreases as k1 0.2 target

increases and ε decreases as in figure 16.

proposed controller
0.1 no controller
The lateral distance to path of the hatchback is illustrated 0.5 0
PID controller

in figure 22. The target value is 0, since the test road is 25.9 26 26.1
straight. However, the lateral distance under PID controller
and no ABS controller deviates 0 too much. As a matter of 0
fact, the hatchback has already loses its stability. In contrast,
0 5 10 15 20 25 30
the hatchback with ABS controller can stay around the straight longitudinal distance (m)
target.
Figure 23 exhibits the evolution of the yaw angle during Fig. 22: lateral distance to path (dash-dotted: reference target;
the split µ road test. Because of no steering manoeuvre com- solid: proposed ABS controller; dashed: without ABS con-
manded, its reference value is 0. However an approximate 180◦ troller; dotted: PID controller.)

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

TABLE IV: Braking performance indices

road condition indices no ABS PID ABS proposed ABS improvement (↓ %)
low µ braking time (s) 3.7 2.7 2.6 29.7
braking distance (m) 18.23 13.62 13.45 26.2
high µ braking time (s) 3.85 3.65 3.45 10.4
braking distance (m) 53.52 50.29 46 14.1
opposite µ braking time (s) 5.79 3.51 2.97 50.6
braking distance (m) 21.98 12.06 9.7 55.9
split µ braking time (s) 3.35 3.38 2.9 15.5
braking distance (m) 26.35 26.65 26.1 2.1

200
proposed controller
condition, almost covering all the commonly encountered road
no controller conditions. The proposed ABS controller only have two simple
150 PID controller
yaw angle ( °)

tuning controller parameters, and the priori knowledge for road

100 condition is not a prerequisite, hence the proposed controller
can be easily perform in practice. More than that, a modified
50 estimator is proposed to estimate the road condition, which is
vital to vehicle dynamics control. Since the controller needs
0
0 1 2 3 4 5 full state feedback, the challenging measurement or estimation
t (s)
for some significant signals, such as vehicle velocity, will be
Fig. 23: yaw angle (solid: proposed ABS controller; dashed: further explored in the future work.
without ABS controller; dotted: PID controller)
R EFERENCES
[1] A. Patil, D. Ginoya, P. D. Shendge, and S. B. Phadke,
1 “Uncertainty-estimation-based approach to antilock brak-
0.8 ing systems,” IEEE Transactions on Vehicular Technolo-
0.6 gy, vol. 65, no. 3, pp. 1171–1185, 2016.
0.4 rear left wheel [2] R. Yazdanpanah and M. Mirsalim, “Design of robust
front left wheel
0.2 rear right wheel
speed and slip controllers for a hybrid electromagnetic
0
front right wheel brake system,” Electric Power Applications Iet, vol. 9,
0 0.5 1 1.5 2 2.5 3 3.5 4 no. 4, pp. 307–318, 2015.
t(s)
[3] J. Zhang, W. Sun, and Z. Feng, “Vehicle yaw stability
Fig. 24: estimation of µ (solid: front left wheel; dashed: rear control via H∞ gain scheduling,” Mechanical Systems
left wheel; dotted: rear right wheel; dash-dotted: front right and Signal Processing, vol. 106, pp. 62–75, 2018.
wheel.) [4] J. S. Hu, Y. Wang, H. Fujimoto, and Y. Hori, “Ro-
bust yaw stability control for in-wheel motor electric
vehicles,” IEEE/ASME Transactions on Mechatronics,
time scale ABS controller. Notice that the index improvement vol. 22, no. 3, pp. 1360–1370, 2017.
of proposed controller is compared to that without ABS [5] W. Sun, H. Pan, and H. Gao, “Filter-based adaptive
controller. Notwithstanding the inconspicuous improvement of vibration control for active vehicle suspensions with elec-
braking performance indices under split µ road, the proposed trohydraulic actuators,” IEEE Transactions on Vehicular
controller can significantly improve the braking safety, which Technology, vol. 65, no. 6, pp. 4619–4626, 2016.
is explained in detail with figure 23. [6] W. Sun, H. Gao, and O. Kaynak, “Vibration isola-
Discussion 2 This paper gives a novel ABS control method, tion for active suspensions with performance constraints
employing time-scale separation design to overcome the un- and actuator saturation,” IEEE/ASME Transactions on
certainty of road condition. In fact, other control methods, Mechatronics, vol. 20, no. 2, pp. 675–683, 2014.
such as sliding mode control combining with fuzzy logic [7] W. Sun, Y. Zhang, Y. Huang, H. Gao, and O. Kay-
control[27][28], etc., also have strong robustness for uncertain nak, “Transient-performance-guaranteed robust adaptive
parts of the systems, which would be further explored in future control and its application to precision motion control
research activities. systems,” IEEE Transactions on Industrial Electronics,
vol. 63, no. 10, pp. 6510–6518, 2016.
[8] R. Bhandari, S. Patil, and R. K. Singh, “Surface predic-
V. CONCLUSION
tion and control algorithms for anti-lock brake system,”
In this paper, a two-time-scale redesign scheme for ABS Transportation Research Part C, vol. 21, no. 1, pp. 181–
controller is proposed, and the effectiveness is verified via 195, 2012.
numerical simulation and CarSim-Matlab co-simulation with [9] R. Verma, D. Ginoya, P. D. Shendge, and S. B. Phadke,
various road conditions, including high µ road condition, low “Slip regulation for anti-lock braking systems using
µ road condition, opposite µ road condition and split µ road multiple surface sliding controller combined with inertial

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2864719, IEEE
Transactions on Industrial Electronics

10

delay control,” Vehicle System Dynamics, vol. 53, no. 8, pp. 34–44, 2009.
pp. 1150–1171, 2015. [23] Y. Wan, J. Zhao, and G. M. Dimirovski, “Robust adaptive
[10] M. S. Basrah, E. Siampis, E. Velenis, D. Cao, and control for a single-machine infinite-bus power system
S. Longo, “Wheel slip control with torque blending using with an svc,” Control Engineering Practice, vol. 30,
linear and nonlinear model predictive control,” Vehicle no. SI, pp. 132–139, 2014.
System Dynamics, vol. 55, no. 11, pp. 1–21, 2017. [24] A. Chakrabortty and E. Scholtz, “Time-scale separation
[11] J. Yi, L. Alvarez, X. Claeys, and R. Horowitz, “Emer- designs for performance recovery of power systems with
gency braking control with an observer-based dynamic unknown parameters and faults,” IEEE Transactions on
tire/road friction model and wheel angular velocity mea- Control Systems Technology, vol. 19, no. 2, pp. 382–390,
surement,” Vehicle System Dynamics, vol. 39, no. 2, pp. 2011.
81–97, 2003. [25] W. Sun, S. Tang, H. Gao, and J. Zhao, “Two time-scale
[12] T. Koppen, T. Kupper, and O. Makarenkov, “Existence tracking control of nonholonomic wheeled mobile robot-
and stability of limit cycles in control of anti-lock braking s,” IEEE Transactions on Control Systems Technology,
systems with two boundaries via perturbation theory,” vol. 24, no. 6, pp. 2059–2069, 2016.
International Journal of Control, vol. 90, no. 5, pp. 974– [26] E. Kayacan, Y. Oniz, and O. Kaynak, “A grey system
989, 2016. modeling approach for sliding-mode control of antilock
[13] L. Li, X. Li, X. Wang, Y. Liu, J. Song, and X. Ran, braking system,” IEEE Transactions on Industrial Elec-
“Transient switching control strategy from regenerative tronics, vol. 56, no. 8, pp. 3244–3252, 2009.
braking to anti-lock braking with a semi-brake-by-wire [27] Y. Wang, H. Shen, H. R. Karimi, and D. Duan,
system,” Vehicle System Dynamics, vol. 54, no. 2, pp. “Dissipativity-based fuzzy integral sliding mode control
231–257, 2016. of continuous-time t-s fuzzy systems,” IEEE Transac-
[14] Y. Qiu, X. Liang, and Z. Dai, “Backstepping dynamic tions on Fuzzy Systems, vol. 26, no. 3, pp. 1164–1176,
surface control for an anti-skid braking system,” Control 2018.
Engineering Practice, vol. 42, pp. 140–152, 2015. [28] Y. Wang, Y. Gao, H. R. Karimi, H. Shen, Z. Fang,
[15] B. Wang, X. Huang, J. Wang, X. Guo, and X. Zhu, and Senior, “Sliding mode control of fuzzy singularly
“A robust wheel slip ratio control design combining perturbed systems with application to electric circuit,”
hydraulic and regenerative braking systems for in-wheel- IEEE Transactions on Systems Man and Cybernetics
motors-driven electric vehicles,” Journal of the Franklin Systems, DOI: 10.1109/TSMC.2017.2720968.
Institute, vol. 352, no. 2, pp. 577–602, 2015.
[16] T. A. Johansen, I. Petersen, J. Kalkkuhl, and J. Ldemann,
“Gain-scheduled wheel slip control in automotive brake
systems,” IEEE Transactions on Control Systems Tech- Weichao Sun (M’13) received the Ph.D. degree in
control science and engineering from Harbin Insti-
nology, vol. 11, no. 6, pp. 799–811, 2003. tute of Technology, Harbin, China, in 2013. Since
[17] H. Jing, Z. Liu, and H. Chen, “A switched control July 2013, he has been with the Research Institute
strategy for antilock braking system with on/off valves,” of Intelligent Control and Systems, Harbin Institute
of Technology.
IEEE Transactions on Vehicular Technology, vol. 60,
no. 4, pp. 1470–1484, 2011.
[18] T. Shim, S. Chang, and S. Lee, “Investigation of sliding-
surface design on the performance of sliding mode con-
troller in antilock braking systems,” IEEE Transactions Jinhua Zhang received the B.S. degree in vehicle
on Vehicular Technology, vol. 57, no. 2, pp. 747–759, engineering from Harbin Institute of Technology,
2008. Harbin, China, in 2016. He is currently pursuing the
Ph.D. degree in control science and engineering at
[19] X. Fan, K. Chandrayana, M. Arcak, S. Kalyanaraman, Harbin Institute of Technology, Harbin, China. His
and T. Y. Wen, “A two-time-scale design for edge- research interests include vehicle dynamics control.
based detection and rectification of uncooperative flows,”
IEEE/ACM Transactions on Networking, vol. 14, no. 6,
pp. 1313–1322, 2006.
[20] E. Lavretsky and N. Hovakimyan, “Dynamic inversion
for nonaffine-in-control systems via time-scale separa- Zhiyuan Liu received the B.S. degree from the Zhe-
jiang University of Technology, Hangzhou, China, in
tion. part i,” Journal of Dynamical and Control Systems, 1982, and the Ph.D. degree in control science and
vol. 13, no. 4, pp. 451–465, 2007. engineering from the Harbin Institute of Technology,
[21] E. Lavretsky and N. Hovakimyan, “Adaptive dynamic Harbin, China, in 1992. He was a Postdoctoral Fel-
low with the Department of Mechanical Engineer-
inversion for nonaffine-in-control uncertain systems via ing, Harbin Institute of Technology, from 1992 to
time-scale separation. part ii,” Journal of Dynamical and 1994. Since 1994, he has been with the Department
Control Systems, vol. 14, no. 1, pp. 33–41, 2008. of Control Science and Engineering, Harbin Institute
of Technology, where he became a Professor in
[22] A. Chakrabortty and M. Arcak, “Time-scale separation 1999. He is currently a Professor and the Director
redesigns for stabilization and performance recovery of of the Department of Control Science and Engineering. His current research
uncertain nonlinear systems,” Automatica, vol. 45, no. 1, interests include robotics, robust control, and model predictive control.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.