Escolar Documentos
Profissional Documentos
Cultura Documentos
Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics
State Key Laboratory of Robotics and System (HIT), Harbin Institute of Technology, Harbin 150080, China
Faculty of Science, King Abdulaziz University, Jeddah, Saudi Arabia
a r t i c l e
i n f o
Article history:
Received 20 March 2014
Revised 15 June 2014
Accepted 31 July 2014
Available online 27 August 2014
Keywords:
Active suspension system
Extended state observer
Nonlinear tracking control
Performance constraints
a b s t r a c t
In this paper, a nonlinear tracking control strategy with extended state observer (ESO) is presented for
vehicle active suspensions to improve the ride comfort, where suspension spaces, dynamic tire loads
are considered as time-domain constraints to be guaranteed. The unique characteristic of the proposed
approach lies in the independence on accurate mathematical model. More exactly, the unknown dynamics and external disturbances of the vehicle suspension are regarded as an augmented state of the system
and are estimated using the designed ESO. The stability analysis shows that both the estimation error and
the tracking error of the control output are bounded and that the upper bounds of the errors monotonously decrease with the increase of the observer bandwidth. Finally, a competitive experiment on a
quarter-car suspension prototype is given to demonstrate the effectiveness of the proposed control
schemes.
2014 Elsevier Ltd. All rights reserved.
1. Introduction
Vehicle suspension systems play a crucial role in reasonably well
isolating its passengers from the vibrations generated by irregular
road on the human body, keeping tire contact with the ground
and improving the vehicle maneuverability. Compared with the
passive or semiactive suspension systems, the active suspension
systems are more effective in enhancing the ride comfort and the
road holding capability. Therefore, active suspension control is currently paid considerable interest in both academia and industry for
the issues on how to guarantee the stability of the suspension
systems and achieve the required suspension performances. In
particular, the force actuator in active suspensions is able to both
add and dissipate energy from the system, which has a potential
to control the attitude of the vehicle, to reduce the negative effects
of braking and the vehicle roll during cornering maneuvers.
It is still a challenge to develop an appropriate control strategy
for dealing with the suspension performances, such as ride comfort, which means to isolate passengers from vibration and shock
arising from road roughness; handling performance, which refers
to the uninterrupted contact forces of tires and the road surface;
and suspension travel, which is associated with the displacement
between a sprung mass and an unsprung mass. However, these
Corresponding author.
E-mail address: w.sun@hit.edu.cn (W. Sun).
http://dx.doi.org/10.1016/j.mechatronics.2014.07.006
0957-4158/ 2014 Elsevier Ltd. All rights reserved.
364
consists of the system nonlinearity, external disturbances, structural and parametric uncertainty, sophisticated and multiple
objectives, and performance criteria. In general, a comprehensive
vehicle model is too complicated for use in control system design.
Therefore, by means of an extensively used model in the literature,
a simplied model capturing the essential characteristics of a real
vehicle system is depicted.
2.1. Nonlinear quarter-car model
Here, the dynamics structure diagram of the quarter-vehicle
model is shown in Fig. 1, which is valid for maneuvers involving
a quarter active suspension. The sprung mass ms represents the
car chassis, and the unsprung mass mu represents mass of the
wheel assembly. The passive components of the suspension system
consist of a non-linear stiffening spring with the stiffness coefcients of linear ks and nonlinear terms ksn , and the piece-wise linear
damper with the damping coefcients for the extension be and
compression movements bc . The actuator force between the
sprung and unsprung masses is represented by ut. The tire is
assured contact with the surface of the road when the vehicle is
traveling. It is modeled as a linear spring with a stiffness coefcient
kf and damping coefcient bf of the tire. The zs and zu are the vertical displacements of the sprung and unsprung masses, respectively, and zr is the road displacement input. Derived from
Newtons second law, the dynamic behavior of an active suspension system may be expressed by the following differential
equations:
mszs F d z_ s ; z_ u ; t F s zs ; zu ; t ut;
muzu F d z_ s ; z_ u ; t F s zs ; zu ; t F t zu ; zr ; t F b z_ u ; z_ r ; t ut; 1
where F d and F s denote the forces produced by the springs and
dampers, respectively, and F t ; F b are the elasticity force and damping force of the tires.
Remark 1. It is to be noted that with a change in the number of
passengers or the payload, the vehicle load will easily vary and this
will accordingly change the vehicle mass ms . On the other hand,
the range of change for the sprung mass is determined by the
unload and maximum mass of the vehicle, decreased with the
sprung weight. At the same time, considering the natural environment and articial factors, the stiffness coefcients and damping
coefcients may gradually change over time, which will greatly
affect the controller design. Therefore, in this model, the proposed
strategy requires minimal a priori information of the plant, which
means ms ; mu ; F d ; F s ; F t , and F b are unknown.
zs
ms
Fs
Fd
zu
mu
Ft
Fb
zr
365
jF t F b j < ms mu g;
jzs zu j 6 zmax ;
3. Proposed solution
The existing control design method for vibrational vehicle suspension systems are greatly depended on accurate dynamic model,
or at least depended on the known model structure. However, it
could pose some rather considerable challenges in engineering
practical requirements. The factors such as the internal dynamics
of the vehicle systems and external disturbances which include
the measurement noise, the output disturbance, and unmodel
dynamic terms arising from mechanical imperfections, bring structure uncertainties and modeling errors into the system that cannot
be simply ignored. If the disturbance, representing the discrepancy
between the plant and its model, is estimated in real time, then the
plant-model mismatch can be effectively compensated for, making
the model-based design tolerant of a large amount of uncertainties.
In this circumstance, it is desired to design a observer to estimate
x f x; x
_ gx; xu;
_
x_ 1 x2 ;
_ gx; xu:
_
x_ 2 f x; x
e2 e_ 1 k1 e1 x2 x2eq ;
x2eq , k1 e1 x_ r ;
_ gx; xu
_ x_ 2eq :
e_ 2 f x; x
1
1
_ x_ 2eq e1 k2 e2
_ t;
f x; x
f x; x
_
_
gx; x
gx; x
t xr K 1 xr x1 K 2 x_ r x2 :
The values of K 1 1 k1 k2 ; K 2 k1 k2 being the design parameters are required to be chosen such that the desired tracking
performance is achieved.
Based on the previous designed process, we can conclude the
following lemma.
Lemma 1. For the given SISO system (4), by designing the proposed
feedback linearization control law (8), the closed-loop system is
asymptotically stable, i.e., all the tracking errors will converge to zero
as the time goes to innity.
Proof. Consider a Lyapunov functional candidate as follow
1 2 1 2
e e :
2 1 2 2
10
366
11
Vt
_ s V0 6 V0;
Vd
12
which implies
je1 j 6
p
2V0;
je 2 j 6
p
2V0:
^s ;
z_ As z Bs u Ls ys y
^
ys C s z;
where z z1
13
z2
z3 , Ls b1
18
T
p
p
jx1 j 6 jxr j 2V 2 0 6 kxr k1 2V 2 0;
p
jx2 j 6 kx_ r k1 k1 1 2V 2 0:
14
2 k1 e1 e_ 1 k2 e2 e_ 2
V
15
is bounded. In the end, V_ 2 is uniformly continuous. By using Lyapunov-like lemma [20], we have V_ 2 ! 0 as t ! 1, and then
e1 ! 0; e2 ! 0, which means that the tracking errors e1 ; e2 converge to zero asymptotically. h
Remark 2. If the exact system model is known, i.e., f ; g is known
function, and dx; t 0, then the designed feedback linearization
controller (8) can obtain asymptotic tracking performance. In general, the feedback liberalization requires exact cancellation of nonlinearities. While no matter how accurate the mathematical
models of dynamic nonlinearities and parameter identication
are, it is impossible to capture the entire nonlinear behaviors and
exact parameters of actual nonlinear effects. There always exist
parameter deviation and unmodeled nonlinearities which cannot
be modeled by explicit functions. Therefore, the feedback linearization control laws may not offer satisfactory asymptotic tracking
performance.
19
3 1
1;
2;
3.
Then,
(19)
can
be
rewritten
as
o
e_ x0 Ae e Be
2
hs
x20
3
where Ae 4 3
1
20
;
1
0
0
3
2 3
0
0
1 5, Be 4 0 5.
1
0
zs f zs ; z_ s ; d b0 ut;
16
x_ s As xs Bs u Es hs ;
y s C s xs ;
2
17
3
2 3
2 3
0 1 0
0
0
4
5
4
5
where As 0 0 1 , Bs b0 , C s 1 0 0 , Es 4 0 5 with
0 0 0
1
0
xs3 f being the augmented state and hs is the rate of change of the
1
K 1 xr x1 K 2 x_ r x2 xr f s :
b0
21
367
(18), we can obtain the estimation of uncertainty. With the estimation, the feedback liberalization controller (21) can be rewritten as
1
K 1 xr x1 K 2 x_ r x2 xr z3 :
b0
based
feedback
liberalization
22
controller
is
Proof. Using (25), (26), and (28) and carrying out some simplications, the closed-loop system can be shown as the following tracking error dynamics:
e_ c Af ec Bd ~z3 ;
Lemma 2. The following single-input linear system:
x Ax Bw
23
s!0
t!1
24
x_ Ax Bu Bd f ;
y Cx;
25
T
where x zs z_ s is the state vector, y is the controlled output and
0 1
0
0
, B
, Bd
, C 1 0 , respectively. Then,
A
0 0
b0
1
T
denoting the reference state vector X r xr x_ r , and dening
the state tracking error ec X r x. The controller (22) can be
rewritten as
1
u K s X r K s x xr z3 ;
b0
26
_
e_ c X_ r x:
27
X_ r AX r Bd xr :
Remark 5. This can easily shown that, with plant dynamic being
largely unknown, the tracking error and its derivative are bounded,
and their upper bounds monotonously decrease with the increase
of the observer bandwidths. When the estimate error ~z3 is zero, the
error dynamics (29) is asymptotically stable. A similar result can be
expected if the bound of the estimate error is reasonably small.
s!0
t!1
29
ut F s zs ; zu ; t F d z_ s ; z_ u ; t ms xr r3 :
30
_
xt
Axt wt;
x
xt 3 ,
x4
0
k
where
A
mfu
0
b
wt kf
.
ms
z t mfu z_ r t m
xr r3
mu r
u
1
b
mfu
31
and
_
V_ 0 xt x_ T tPxt xT tPxt
xT tAT P APxt 2xT tPwt:
It is easy to verify that the matrix A has eigenvalues with negative
real parts. Hence, we have AT P AP Q , where Q > 0 is a positive
matrix. Noting that
2xT tPwt 6
1
V_ 0 xt 6 xT tQxt xT tPPxt gwT twt
g
1
1
1
6 kmin P 2 QP2 kmax P V 0 xt gwT twt:
1
1
1
q1 6 kmin P2 QP2 kmax P;
28
368
V_ 0 xt 6 q1 V 0 xt q2 :
32
q
q
V 0 xt 6 V 0 x0 2 eq1 t 2 6 q;
q1
q1
(
where q
jxk tj 6
V 0 x0
2q2
q1
V 0 x0
V 0 x0 P qq21
V 0 x0 < qq21
33
q
q
; k 3; 4. Hence, we obtain that the zero dynamic
kmin P
is stable.
3.5. Performance constraint
Select tunable parameters to guarantee the required performance constraints. From the above analysis, it can be seen that
all the signals are bounded within the known ranges, and the
bound of suspension space can be obtained as
r
q
jx1 x3 j 6 jx1 j jx3 j 6 kxr k1 A1
;
f Bd r3
1
kmin P
34
r
q
1
6 zmax ;
kxr k1 Af Bd r3
kmin P
1
35
Fig. 2. The structure of quarter-car active suspension setup.
jx1 x3 j 6 zmax :
36
jkt bt j
r
q
kt kzr k1 bt kz_ r k1 < ms mu g;
kmin P
37
then
jF t F b j 6 ms mu g;
38
can be guaranteed.
Remark 6. The selection of the design parameters and the initial
values is an important problem to be noted. Based on the previous
analysis, the ESO gain parameters and the initial values should be
chosen to satisfy (35), which implies that the suspension space and
road holding will be guaranteed.
4. Experiment examples
In this section, the proposed ESO-based feedback linearization
controller is implemented on a bench-scale model to emulate a
quarter-car model, as shown in Fig. 2. This active suspension system consists of three masses, or plates. Each mass slides along
stainless steel shafts using linear bearings and is supported by a
set of springs. The upper mass (blue1 plate) represents the vehicle
body supported above the suspension, also known as the sprung
mass. The middle mass (red plate) corresponds to one of the
vehicles tires, or the unsprung mass. The bottom mass (silver plate)
1
For interpretation of color in Fig. 2, the reader is referred to the web version of
this article.
Value
Parameter
Value
ms
mu
2:45 kg
1 kg
900 N/m
10 N/m
2 kg
kf
bf
be
bc
ms max
2500 N/m
1000 N s/m
8 N s/m
7 N s/m
3 kg
ks
ksn
ms min
369
x 10
S1
S2
S3
S1
S2
S3
1.5
3
1
2
0.5
1
0
0.5
2
1
3
1.5
4
5
10
12
10
12
Time (Sec)
Time (Sec)
S2
S3
x 10
S1
S2
S3
3
1
2
1
0
0
1
2
2
3
4
10
12
Time (Sec)
10
12
Time (Sec)
Fig. 4. Control input.
Fig. 6. Suspension spaces.
1.5
s
Z T
RMSx 1=T
xtT xtdt :
0.5
39
x 10
S1
S2
S3
0.5
1
1.5
2
10
Time (Sec)
Fig. 7. The responses of unsprung mass displacement zu .
12
370