Professional Documents
Culture Documents
By
JIE FANG
To my parents, Zheng Fang and Pin Zhang, and my friends, Moses and Darsan
ACKNOWLEDGMENTS
I would like to express my special thanks of gratitude to my advisor, Dr. Crane as
well as my PHD instructor, Olugbenga Moses Anubi and Darsan Petal who gave me the
precious opportunity to work on this wonderful project on the topic, active suspension
system of quarter car. I appreciate their patience to me and significant instruction about
the project.
Secondly, I would also like to thank my parents and friends who gave me lots of
support and help in finishing this project within the limited time.
THANKS AGAIN TO ALL WHO HELPED ME.
TABLE OF CONTENTS
page
ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
CHAPTER
1
INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.1 Passive Suspension System . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Active Suspension System . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Semi-active Suspension System . . . . . . . . . . . . . . . . . . . . . . . 15
SYSTEM DESCRIPTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1
2.2
2.3
2.4
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
27
28
32
34
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
38
38
42
43
Dynamic Analysis
Control Design . .
Stability Analysis .
Simulation . . . . .
18
19
20
25
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
48
48
55
59
LIST OF TABLES
Table
page
LIST OF FIGURES
Figure
page
10
first two methods are supported by Lyapunov based stability analysis to prove that the
tracking error will approach zero asymptotically. Simulation results for each method are
also given to show the vehicle performance and tracking performance.
11
CHAPTER 1
INTRODUCTION
A vehicle suspension system is used to separate the car body physically from the
wheels and allow relative motion between the two parts. It is typically rated by its ability
to provide good roadholding, isolate passenger from road disturbance, and improve
passenger comfort. The road disturbance may be caused by various reasons such as
road unevenness, aerodynamics forces, non-uniformity of the tire/wheel assembly, and
even braking force. The quality of roadholding especially during cornering and swerving
determines the active safety of the vehicle. The ability of absorbing vibration from
road disturbance is mainly discussed in this research. Typically, a suspension system
consists of the system of springs, shock absorbers, and linkages that connect a vehicle
to its wheels. The structure of a quarter car model is shown in Figure 1-1.
12
In order to simplify the analysis of the suspension system, a simplified quarter car
model is shown in Figure 1-2.
Based on the simplified quarter car model, a passive suspension system is one
in which the coefficients of the components are constant. The structure of the passive
suspension system is shown in Figure 1-3.
Lots of researchers have worked on the passive suspension system [1-3] including
its structure, performance, and safety. The main components, determined by the
designer of the suspension system, are ks , bs , and kt . ks and bs represent the spring
and damper coefficients between the wheel and the vehicle body while kt represents
the compliance of the tire. r represents the road disturbance. Usually a sine function
is taken into consideration. Vehicle deflection ys , car body velocity ys , acceleration ys ,
wheel deflection yu , wheel velocity yu , and wheel acceleration yu are usually measured
13
suspension system can improve safety and ride comfort of the vehicle significantly. In
an active suspension system, a force actuator is used instead of a passive damper
or both the passive damper and spring. A significant difference between a passive
suspension system and an active suspension system is that active suspension system
can add and dissipate energy from the system via a force actuator, unlike a passive
suspension system which can only dissipate energy via a passive damper. But the
introduction of active force actuator also introduces other problems. The first problem
is the requirement of larger power which decreases the overall performance of the
vehicle. Meanwhile, the force actuator also adds to the complexity of the whole system
which leads to a larger range of control algorithms. Further, most control methods are
based on the accurate knowledge of the suspension system such as the parameters
of the suspension system, and the vehicle system states. Thus the requirement of
sensors also increases the cost of the suspension system. Another problem for active
suspension systems are unacceptable failure modes. It will be dangerous for both the
vehicle and passengers because of the possible failure of an actuator.
1.3
The semi-active system [12-18] retains the conventional spring element of the
passive suspension but uses a controllable damper. External power is required and the
power is used to adjust the damping level, and operate controller and sensors. Thus the
semi-active suspension system has less complexity, cost and more reliability compared
to the active suspension system.
The skyhook control strategy [17] is the most widely used control policy for
semi-active suspension systems. It can reduce the resonant peak of the body mass
and thus reach a good quality of performance. The structure of the skyhook suspension
system is shown in Figure 1-4.
It consists of a linear spring with ks , a damper with bs between the sprung mass
and unsprung mass and a damper with coefficient bsky which is effectively attached to
15
is applied, the particles disperse randomly and the fluid exhibits Newtonian behavior.
When a magnetic field is applied, the rheological behavior changes from Newtonian
to Bingham plastic, which makes the fluid more viscous. Based on this, the amount of
torque transmitted through the device can be controlled by changing the magnetic field.
In the off state, both the electro-rheological (ER) and magneto-rheological (MR)
fluids have similar viscosity, but once switched to the on state, MR fluids shows a
much greater increase in yield strength, therefore viscosity. Meanwhile, it has been
shown that the device based on an ER fluid will have roughly the same overall power
requirement as similar device based on an MR fluid, though the ER device requires
high voltage, low current power, while the MR device requires low voltage, high current
power. The high requirement of voltage for the ER device makes it impractical for most
commercial applications. Moreover, the MR is less sensitive to contaminants, and has
a much broader useful temperature range than ER fluids. Thus according to different
applications, different semi-active suspension systems can be applied to reach desired
performance.
17
CHAPTER 2
SYSTEM DESCRIPTION
A detailed description of the suspension system of the quarter car model used
in this thesis will be given in this chapter. Meanwhile, the comparison of the passive
suspension system, skyhook suspension system, nonlinear energy sink suspension
system, and suspension system combined with a nonlinear energy sink and skyhook
based on car body deflection, velocity and wheel deflection will also be presented.
2.1 Suspension System Model
The suspension analysis is based on the simplified quarter car model which has
been shown in Figure 1-2. The simplified model consists of the sprung mass ms and
unsprung mass mu including the mass of the tire and axles. The tire is modeled as a
linear spring with stiffness kt . The suspension system is controlled by force F which is
designed by the engineer. Typically, F is generated by a linear spring, a damper and
another force actuator which may be an active actuator or a semi-active actuator. r
represents the road disturbance which is treated as a sine function and can be showed
as follows:
r = A1 sin( t)
(21)
where A1 represents the amplitude of the road disturbance signal, for the purpose of
frequency response generation. The values of the parameters used have been given in
Table 2-1. And L0s represents the original length of the spring Ks .
Parameter Value
ms 315 Kg
mu 37.5 Kg
Kt 210000 N/m
Ks 29500 N/m
bs 1500 N/m/s
A1 0.005 m
L0s 0.6 m
Table 2-1. Dynamic System Parameter Values
18
A certain level of performance and safety will be reached by the passive elements
by choosing appropriate value of kt , ks and bs , while the active element will be used to
further improve the roadholding, passenger comfort, responsiveness and safety. Usually
s , suspension deflection Ys Yu , and wheel
one focuses on the car body acceleration Y
deflection Yu r which determine the vehicle performance. The main idea of the active
suspension system is to design an active element to generate force Fa which can adjust
itself continuously to changing road conditions.
2.2 Performance of Suspension System
The items used to judge the performance of the suspension system in this
work are mainly passenger comfort and road holding. The passengers comfort is a
combination of different factors such as the safety of the driver, driving environment
which cant be controlled, and vibration. In this work, vibration is the only factor taken
into consideration. It can be judged by the isolation between the road disturbance and
primary system. The better the isolation is, the better the passengers comfort is. The
s is utilized to represent the passengers comfort. The lower
sprung mass acceleration Y
the sprung mass acceleration is, the better the passengers comfort is.
The second factor, which determine the performance of the suspension system, is
road holding. It is the ability of the vehicle to keep contact with the road and maximize
wheel tracking to road unevenness and to guarantee road contact whatever the road
profile and load transfer situations. In this work, the road holding is determined by the
suspension deflection which can be presented as Ys -Yu . It has been shown that there
is a trade-off between the passengers comfort and road holding. It is impossible to
avoid this for active suspension systems or semi-active suspension systems. Thus it is
necessary for the engineer to design an appropriate suspension system to minimize it
and obtain the desired result.
19
2.3 Suspension System Combined with Nonlinear Energy Sink and Skyhook
A suspension system combined with a nonlinear energy sink and skyhook is a
combined control strategy based on both nonlinear energy sink and skyhook technology.
As shown in Chapter 1, the skyhook control strategy is a semi-active suspension system
which is based on a linear spring. The nonlinear energy sink control strategy [19-30] is
an active suspension system with a highly nonlinear spring.
Recently, it has been shown that a suspension system combined of linear
substructures and strongly nonlinear parts has the abilities of localization and irreversible
transient transferring of energy to prescribed fragments of the structure dependent on
initial conditions and external force. This new active suspension system can react on the
amplitude characteristics of the external force in a wide range for frequencies.
Unlike a passive tuned absorber which can only work in a narrow band of
frequencies and cant absorb multi-frequency transient disturbances, the nonlinear
energy sink suspension system can transfer vibrational energy from a primary system
to the nonlinear energy sink part where the vibrational energy localizes and diminishes
in time due to dissipation. As shown in [28-30], a transient resonance capture on a 1:1
resonance manifold of the system is at the origin of an irreversible and almost all energy
is transferred from the primary system to the nonlinear energy sink part. However, the
transfer is very selective as the two oscillators must be well tuned, and the primary
system must have a specific amount of energy(the nonlinear energy sink is at rest
initially).
The nonlinear energy sink suspension system has be shown in Figure 2-1. The
nonlinear energy sink consists of a linear spring with coefficient K1 and a nonlinear
spring with coefficient K2 Ys2 . Obviously, the stiffness of the nonlinear spring will
change with the car body deflection Ys which, to some extent, determines the passenger
comfort. Thus if Ys is large which means vibration is heavy, then the nonlinear spring
will become hard by increasing the stiffness to decrease the vibration. According to
20
(22)
where L01 ,L02 are the free lengths of the linear and nonlinear springs.
The model of the suspension system with the nonlinear energy sink and skyhook is
shown in Figure 2-2.
Figure 2-2. Suspension System Combined with Nonlinear Energy Sink and Skyhook
Obviously, the combined suspension system has two parts: the nonlinear energy
sink part and skyhook part. Based on the forces generated by the nonlinear energy sink
21
part and skyhook part, the force generated by the combined suspension system can be
presented as
Fa = FNES + Fsky = K1 (L01 Ys ) K2 (L02 Ys )3 bsky Y s
(23)
The values used in this study for the parameters mentioned above are given in Table
2-2.
Parameter Value
K1 5000N/m
K2 15000N/m
L01 50mm
L02 50mm
bsky 2000N/m/s
Table 2-2. Parameter Values for Combined Suspension System
In order to show the advantages of the suspension system consist of the nonlinear
energy sink and skyhook, it is necessary to make a comparison of the passive
suspension, skyhook suspension system, nonlinear energy sink suspension system,
and the suspension system consist of the nonlinear energy sink and skyhook. An
approximate frequency response from the road disturbance r to the sprung mass
s , suspension travel Ys Yu , and wheel deflection Yu r are computed
acceleration Y
by variance gains[31-32]. Usually the bode plot is utilized to analyze the frequency
response for linear system. However, here the system studied here is a nonlinear
suspension system. Thus a new technique (Variance Gain) is used to analyze the
frequency response. The approximate variance gain is given by
v
u
2N
u
Z 2 dt
G (j) = t 2N 0
A2 sin2 (t)dt
0
(24)
22
The variance gains corresponding to different Z from are shown in Figure 2-3 through
2-5:
60
10log10(Gain)
50
40
30
20
Traditional
Skyhook
NES
NES Skyhook
10
0 0
10
10
Frequency (HZ)
10
20
10log10(Gain)
10
0
10
20
Traditional
30
Skyhook
40
NES
50
NES Skyhook
60 0
10
10
Frequency (HZ)
10
20
10log10(Gain)
10
0
10
20
30
Traditional
Skyhook
NES
SkyhookNES
40
50
60 0
10
10
Frequency (HZ)
10
24
hydraulic pressure. The hydraulic cylinder is a double acting cylinder. The position of the
piston can be changed by modulating the oil flow into and out of the cylinder chambers,
which are connected to the spool valve through cylindrical ports. The modulation is
provided by the spool valve. The dynamic function of the hydraulic actuator and the
spool valve are as follows:
PL = Avp PL + xv
Ps sgn(xv )PL
(25)
1
K
x v = xv + u
(26)
F = APL
(27)
where A is the pressure area in the actuator,PL is the load pressure, vp = d is the
actuator piston velocity, and F is the output force generated by the hydraulic actuator.
The parameters , , are determined by actuator pressure area, effective system
oil volume, effective oil bulk modulus, oil density, hydraulic load flow, total leakage
coefficient of the cylinder, discharge coefficient of the cylinder, and servo valve area
gradient. xv is the spool valve position, is the actuator electrical time constant, K = 1
is the DC gain of the four-way spool valve, and u is the input current to the servo valve.
The values of the hydraulic parameters used in this study are shownin Table 2-3.
Parameter Value
4.5151013 N/m5
1sec 1
1.545 109 N/m5/2 Kg 1/2
1/30 sec
Ps 10342500 Pa
A 3.35104 m2
Table 2-3. Hydraulic System Parameter Values
Based on the knowledge of hydraulic system, different control methods are utilized
to design the control input u to let the hydraulic actuator track the force generated by the
suspension system combined with the nonlinear energy sink and skyhook which has
been shown in Figure 2-2.
26
CHAPTER 3
CONTROL DEVELOPMENT BASED ON SINGULAR PERTURBATION
In this chapter, the control method based on singular perturbation technology [36] is
utilized to let the hydraulic system track the ideal force generated from the suspension
system combined with nonlinear energy sink and skyhook. Typically, if the system has
the form shown in Equation 3-1, then singular perturbation technology can be applied to
solve the system.
x = f (t, x, z, ), x(t0 ) = ()
(31)
z = g(t, x, z, ), z(t0 ) = ()
The idea of applying singular perturbation theory is to obtain the knowledge about the
solution of the whole system when is small by using limiting behavior of the system.
3.1
Dynamic Analysis
According to the simplified quarter car model in Figure 1-2, the states and dynamic
system of the simplified quarter car model are given as follows:
x 1 = ys
x2 = ys
x 3 = yu
x4 = yu
x5 = PL actuator load pressure
x6 = xv spool valve position
The dynamic equation is
X 1 = X2
Ks
X 2 = m
X1
s
bs
X
ms 2
Ks
X
ms 3
bs
X
ms 4
A
X
ms 5
X 3 = X4
X 4 =
Ks
X
mu 1
bs
X
mu 2
Kt
(m
u
Ks
)X3
mu
X 6 = 1 (X6 + K u)
27
bs
X
mu 4
A
X
mu 5
(32)
Kt
r
mu
The values of parameter used in this study have been shown in Tables 2-1, 2-2, and 2-3
3.2
Control Design
According to the dynamic system from Equation 3-2 and hydraulic system equation
from Equations 2-5 to 2-7, the force generated by the hydraulic system F can be
obtained by taking a derivative of Equation 2-7 and substituting Equation 2-5 into it
which can be written as
F = F A2 d + A
u
where
(33)
u = xv
F
Ps sgn(xv ) .
A
(34)
(35)
where Fd represents a desired force which can be obtained by analyzing the combined
suspension system. Fd is written as
Fd = FNES + Fsky + Fpassive
(36)
where FNES represents the force generated by the nonlinear energy sink part, Fsky
represents the force generated by the skyhook part, and the Fpassive represents the force
generated by the passive elements which include a linear spring Ks and a damper bs .
Thus
FNES = K1 (L01 Ys ) K2 (L02 Ys )3
(37)
Fsky = bsky Y s
(38)
Fpassive = Ks(L0s d) bs d
(39)
28
(310)
where
d = Ys Yu
(311)
d = Y s Y u
(312)
The desired force Fd is tracked by the F which is exerted by the hydraulic actuator in
Equation 2-7. Taking the derivative of the tracking error e, e can be written as
e = F F d
= F A d + rA
u F d
(313)
Then add and subtract both Fd and Fd , where Fd represents the estimation parameter
of F d . Thus, (3-13) can be written as
e = F A2 d + rA
u F d
= F + Fd A2 d + rA
u F d + Fd Fd Fd .
(314)
ed be the error between the actual value and estimate value of Fd . Thus
Let F
ed = F d Fd .
F
(315)
f
= e F
u Y T )
d + rA(
where
[
Y =
]
d Fd
Fd
[
=
(316)
]
A
r
rA
1
rA
The estimate value of Fd can be obtained by using the high gain observer [20]
2 P = Ahg P bhg Fd
(317)
1 T
Fd = sat( chg
P, a, b)
2
(318)
29
a, if
sat(, a, b) =
, if
b, if
and
<a
ab
(319)
>b
1 1
1
0
Ahg =
, bhg = , chg = , 2 1.
1 0
1
1
(320)
The saturation function can be used to overcome the peaking phenomenon associated
with high gain observers. The stability analysis about the hydraulic suspension system is
is designed as
based on the Lyapunov Function. Thus, the fictitious control U
u = Y T k0 e c0 sgn(e)
(321)
where represents the estimated value of parameter and k0 , c0 are control gains.
Then the closed-loop error can be obtained by substituting (3-21) into (3-16),
Te
f
e = ( + k0 A)e F
d c0 Asgn(e) AY
(322)
where e represents the error between the actual value and estimate value of
e = .
(323)
The control input can be designed by using the singular perturbation technique [21] to
simplify the controller design for the actuators. It is given by
u = Kf xv +
1 + KKf
us
K
(324)
Then by substituting (3-24) into (3-2), the valve psuedo-closed loop dynamics is given by
x v + xv = us
30
(325)
where
.
1 + KKf
(326)
The term is the perturbation constant. Let = 0, then the quasi-steady state solution
(xvi ( = 0)) xv is given by
xv = us .
(327)
The valve dynamics can be decomposed into fast and slow time scales by using the fast
time scale =
(328)
d
=
d
(329)
where () is a boundary layer correction term. According to the (3-29), it is obvious that
(v ) decays exponentially in the fast time scale. Usually, the time constant is designed
to satisfy 0 < 1[22]. Thus, the perturbation constant can be as small as possible,
if large control gain Kf can be chosen. As a consequence, + O() will be negligibly
small. Thus (3-27) exists. And (3-4) can be written as
u = us
F
Ps sgn(us ) .
A
(330)
Assuming sufficient pressure for the hydraulic pump, the term inside the square root
operator is taken as positive. Thus
sgn(
u ) = sgn(us )
Then, (3-30) will be
(331)
u = us
F
Ps sgn(
u) .
A
(332)
31
(333)
At this point, the control design based on singular perturbation for the input to
hydraulic system is completed. In this control design, the tracking error between the
ideal force generated from the suspension system combined with nonlinear energy sink
and skyhook and the force generated from hydraulic system is minimized.
3.3
Stability Analysis
In this section, the Lyapunov based stability analysis is described. The corresponding
adaptation law is also designed in this section to make the actuator force tracking error
approach to zero asymptotically, if the control gains are chosen to meet certain sufficient
conditions.
Theorem 3.1: Given the adaptive update law
= Ye
(334)
where is a positive definite adaptation gain matrix. If the control gain c0 is chosen to
meet the sufficient conditions as follows,
c0
ed |
|F
.
A
(335)
Then the tracking error e in Equation 3-5 will approach to zero as time goes to infinity. i.e
e(t) 0, as t 0.
Proof: Let the Lyapunov function be
V =
1 2 A eT 1 e
e +
.
2
2
(336)
e
V = e e + AeT 1 .
32
(337)
e
Then substituting 3-22, 3-34 into 3-37 yields
Because =
- and = 0. Thus e = .
V = e e AeT 1
Te
f
eT
= e[( + k0 A)e F
d c0 Asgn(e) AY ] + A Ye
f
= ( + k0 A)e 2 + F
d e c0 Aesgn(e).
(338)
f
f
F
d e |Fd ||e|
and
esgn(e) |e|
Equation 3-38 yields
f
V ( + k0 A)e 2 + |F
d ||e| c0 A|e|.
(339)
(340)
According to 3-36 and 3-40, it follows that V(t) is bounded, which means e(t) and e are
also bounded. Meanwhile, from 3-18, it follows that Fd is bounded. Then according to
3-16, e is bounded, which implies that e(t) is uniformly continuous. Then integrating
both sides of 3-40 yields that e(t) L2 . Based on the conditions metioned above, using
Barbalats Lemma [21] yields
e(t) 0, as t .
Thus, based on the stability analysis, the hydraulic system can track the ideal force
generated by suspension system combined with nonlinear energy sink and skyhook
perfectly as time goes to infinity.
33
3.4 Simulation
In this section, the simulation of the behavior of the quarter car suspension system
is done by using Matlab Simmechanics. The Quarter-car Model was modeled in
Solidworks, and then translated to a Simmechanic model (In Figure 3-1). The vertical
strut and tire damping and stiffness used are the ones given in the Renault Megane
Coupe model [39].
4000
3500
Force (N)
3000
2500
2000
1500
F
Fd
1000
15
15.5
16
16.5
17
17.5
Time (s)
34
18
18.5
19
19.5
20
6000
Force (N)
5000
4000
3000
NES
NES Skyhook
2000
Passive
1000
0
15
Skyhook
16
17
Time (s)
18
19
20
6
4
2
0
2
NES
NES Skyhook
6
Passive
Skyhook
8
15
15.5
16
16.5
17
17.5
Time (s)
18
18.5
19
19.5
20
NES
NES Skyhook
0.65
Passive
Skyhook
0.6
0.55
0.5
0.45
0.4
15
15.5
16
16.5
17
17.5
Time (s)
18
18.5
19
19.5
20
36
system with nonlinear energy sink part is larger than that for the other two suspension
systems without the nonlinear energy sink part. However, the degradation in suspension
deflection is not as much as the improvement achieved in the passenger comfort.
Thus if the passenger comfort is the most significant requirement, then the suspension
system with nonlinear energy sink part will be better choice. From Figure 3-3 to 3-5
the advantage for the combined suspension system is obvious. The peak values of
car body acceleration and suspension deflection for the combined suspension system
are smaller than that for the pure nonlinear energy sink part. Thus, the combined
suspension system with nonlinear energy sink and skyhook combined the advantage
of the nonlinear energy sink which can transfer the energy from road disturbance and
dissipate it completely and irreversibly, and the skyhook which can decrease the peak
value of resonance.
37
CHAPTER 4
MULTIPLE SLIDING MODE CONTROL
4.1 Introduction of Sliding Mode Control
In this chapter, multiple sliding mode control strategy [37-40] will be used to design
a control input to the hydraulic system so that the force generated by the hydraulic
actuator can track the ideal force designed from the nonlinear energy sink and skyhook
suspension system perfectly.
Sliding mode control is a class of nonlinear control. The goal of the switching
control law is to drive the nonlinear plants state trajectory onto a designed surface in
the state space and maintain the plants state trajectory on the same surface all the
time. Meanwhile, sliding mode control is also a variable structure control which can
switch from a continuous structure to another. The surface defined here is designed
by engineer which is also called a sliding surface (sliding manifold). A control input
is designed so that the plant state can slide to the surface and stay on the surface.
Moreover, a Lyapunov approach is also utilized to prove the stability of the whole
system.
4.2
Control Design
The dynamic system has been shown in Chapter 3 from Equation 3-2. As discussed
in Chapter 2, the desired dynamics of the active system consisting of a passive
suspension system with a nonlinear energy sink and skyhook has been shown in
Figure 2-2. The ideal force is shown in Equation 3-10.
In order to simplify the problem, the following first order system is taken into
consideration first. Assume the nonlinear system has the form,
x = f (x) + g(x)u
38
(41)
where x is the state of the system and u is the control input to the system. If the control
input is designed based on feedback linearization, then the control input should be
u=
1
(f (x) + v )
g(x)
(42)
Substitute Equation 4-2 into Equation 4-1, then x = v. If we design v = -k*x. Then
asymptotically result can be obtained. The Lyapunov function V is chosen as
V =
1 T
x x.
2
(43)
Then taking the first derivative of the Lyapunov function, asymptotically stable or even
exponentially stable can be achieved by designing an appropriate v. But the feedback
linearization control is based on exact knowledge of the whole system, which means
we need to know f(x) and g(x). In order to design a robust controller, the sliding mode
control method is utilized. Firstly, the first sliding surface which is the error between the
actual state and ideal state can be presented as follows
s1 (x, t) = xactual xdesired .
(44)
1 T
s1 s1 .
2
(45)
If the first derivative of the Lyapunov function satisfies the following unequal equation,
then the system can satisfy the robustness or sliding condition:
d 1
V = ( s 2 ) = s s ks 2
dt 2
(46)
39
larger the k is, the faster the convergence is. But k is also constrained by the control
input which is determined by the physical system.
In the suspension system system, we expect that the force generated by the
hydraulic actuator can track the ideal force generated from the combined suspension
system. And assume the road disturbance of the suspension system is unknown and
bounded by 0.1 m. Thus based on the state space from Equation 3-2, the first sliding
surface can be defined as
s1 (x, t) = Factual Fdesired
(47)
(48)
where A represents the area of the valve, and PL represents the pressure. Thus the
force generated by the hydraulic actuator can be obtained by multiplying the area and
pressure of the valve. Meanwhile, the desired force can also be presented as
Fdesired = K1 (L01 Ys ) K2 (L02 Ys )3 bsky (Y s Y u )
= A x5desired
(49)
(410)
substitute the states which have been shown in dynamic equation, then
F desired = h(x) Hxr
(411)
where xr represents the unknown road disturbance and xr is bounded by 0.1 m and H
is some positive constant determined by parameters bsky , kt , mu . Then according to the
dynamic system, x 5 is
X 5 = X5 A(X2 X4 ) + X6
40
Ps sgn(X6 )X5
(412)
Then let
f (x) = X5 A(X2 X4 )
(413)
(414)
(415)
1
(f (x) + x 5desired k3 s1 k5 sgn(s1))
g(x)
(416)
where f(x), g(x) have been defined in Equation 4-13 and 4-14, and k3 and k5 are some
positive constants. The value of k3 which is designed by the engineer will affect the
convergence rate of the error between the actual force and ideal force. At last x5desired
can be obtained as follows
x 5desired
d( Fdesired
)
A
=
.
dt
(417)
(418)
where x6actual has been defined from Equation 3-2, and x6desired has been defined from the
first sliding surface from Equation 4-15.
Then taking the first derivative of the second sliding surface, gives
s2 = x 6actual x 6desired = 1 (x6actual + u) x 6desired
41
(419)
In order to let the system satisfy robustness and for the nonlinear plants states to slide
along the two surfaces, the control input can be designed as
u = x6actual + (x 6desired k4 s2 )
(420)
where x 6desired can be obtained by taking the derivative of Equation 4-15, and k4 is some
positive constant which is designed by the engineer.
Up to now, the multiple sliding mode control based on defining two sliding surfaces
has been defined which is aimed at letting the actual force generated by the hydraulic
actuator track the ideal force generated by the combined suspension system.
4.3
Stability Analysis
1 2 1 2
s + s .
2 1 2 2
(421)
Then take the first derivative of the Lyapunov function. According to Equation 4-14, 4-15,
4-18, and 4-19, then
V
= s1 s1 + s2 s2
=
k3 s12
k4 s22
Hs1 xr + k5 |s1 |
(422)
(423)
(424)
where C is some positive constant determined by the values of k3 and k4 . After solving
the first derivative equation, V can be presented as
V (t) V (0)e Ct .
42
(425)
Obviously, from Equation 4-25, the V will be exponentially convergent to zero as time
goes to infinity. And because V is defined from Equation 4-21, thus the s1 and s2 should
also converge to zero exponentially as time goes to infinity. If s1 goes to zero as time
goes to infinite, it means the force generated by the hydraulic actuator can track the
ideal force defined by the suspension system combined with the nonlinear energy sink
and skyhook perfectly as time goes to infinity.
4.4 Simulation
In the simulation, assume the vehicle travels at a steady horizontal speed of 40mph
which is same as before. Also there is a road bump with amplitude 0.1 m . Based on the
simulation, several aspects are taken into consideration. The first one is to verify that
the force generated by the hydraulic actuator can track the desired force designed using
sliding mode control. The desired force and actual force are shown in Figure 4-1.
1500
Force (N)
1000
500
0
500
Actual Force
Desired Force
1000
1500
Time (s)
43
0.15
Hydraulic System
Passive System
0.1
0.05
0.05
0
Time (s)
10
15
Hydraulic System
Passive System
2
0
Time (s)
10
15
Hydraulic System
Passive System
0.1
0.05
0.05
0
Time (s)
44
10
15
Hydraulic System
Passive System
2
0
Time (s)
10
15
x 10
2
1
0
1
2
3
0
Time (s)
10
15
0.05
0.05
0
Time (s)
10
15
0.5
0.5
1
0
Time (s)
46
10
15
0.15
Unsprung Mass Position (m)
0.05
0.05
0
Time (s)
10
15
1
Sliding Mode Control
Original System
0.5
0.5
1
0
Time (s)
47
10
15
CHAPTER 5
MODEL PREDICTIVE CONTROL
In this chapter, Model Predictive Control [44] is used to make sure that the actual
force generated by the hydraulic system can track the ideal force defined by the
suspension system combined with the nonlinear energy sink and skyhook. Similarly,
the plots of force tracking, car body position, velocity and other states will be shown
to make a comparison of the passive suspension system, the combined suspension
system by utilizing sliding mode control, and the combined suspension system utilizing
model predictive control.
5.1 Introduction of Model Predictive Control
Model Predictive Control is one of the advanced control methods which has been
worked since the 1980s. The theory behind model predictive control is iterative, finite
horizon optimization. At each time t, the finite prediction of states is calculated and
substituted into a cost function which is designed by the engineer. The best control input
will be calculated to minimize the cost function. After getting the best control input, u,
for current time t, states for time t+1 can be updated and utilized for the calculation of
best control input for next time t+1. The prediction horizon keeps shifting forward, and
the steps repeated. Finally a vector of best control input corresponding to each time is
obtained. Thus model predictive control is also called receding horizon control. Recently,
model predictive control is often used in a majority of existing multivariable control
applications. It can also applied in a system with constraints or controlled variables by
utilizing on-line optimization.
5.2
Control Design
In order to simplify the problem and understand the theory behind model predictive
control, a simple state space system is taken into consideration. Assume the system
can be presented as
x
= Ax + Bu
= Cx + Du
48
(51)
where x represents state, u represents control input, and y represents the output of the
whole system. According to Equation 5-1, the system is a linear system, where matrices
A, B, C, and D are determined by the physical system which are known.
Assume N is the number of points picked during the time T. Np represents the
prediction horizon which means at each time t, states will be predicted from t+1 to t+Np .
Nc presents the control horizon. Usually, Nc Np . Based on these basic concepts, the
first step is discretization. Typically, the system begins analyzed is continuous linear
system. However, the model predictive control usually works on a discrete time system.
By using Matlab, one can use the command c2d which can transfer the continuous
system with state-space form to discrete form. After using the c2d command in Matlab,
the new discrete system can be presented as
x(k + 1) = Am x(k) + Bm u(k)
Y (k)
(52)
= Cm x(k) + Dm u(k)
Am x(k + 1) + Bm u(k + 1)
A2m x(k)
(53)
(54)
+ Am Bm u(k) + Bm u(k + 1)
3
x(k + 3) = Am
x(k) + A2m Bm u(k) + Am Bm u(k) + Bm u(k + 1)
(55)
Np 1
p
p 2
x(k+Np ) = AN
Bm u(k)+AN
Bm u(k+1)+...+Am Bm u(k+Np 2)+Bm u(k+Np 1)
m x(k)+Am
m
(56)
49
Bm
x(k + 1) Am
x(k + 2) A2
A B
Bm
m m
.
.
=
x(k)+
.
.
.
.
N
N 1
N 2
N 3
x(k + Np )
Amp
Amp Bm Amp Bm Amp Bm . . . Bm
Now let
x(k + 1)
x(k + 2)
.
X =
x(k + Np )
,U =
u(k)
u(k + 1)
.
.
.
u(k + Np 1)
(57)
u(k)
u(k + 1)
.
.
.
u(k + Np 1)
(58)
which represents the predictive states at each time, the number of prediction should be
Np which has been determined before. The matrix Equation 5-7 can now be simplified
as
X = E0 x(k) + F0 U
(59)
Y = Ex(k) + FU
(510)
50
where
Am
A2
m
.
E0 =
N
Amp
, F0 =
Bm
Am B m
Bm
.
.
.
N 1
Amp
N 2
Bm Amp
N 3
Bm Amp
(511)
Bm . . . Bm
1
2
J= (
qs ||ek+s || +
rs ||u(k)||2 )
2 s=1
s=0
(512)
where qs and rs are weight parameters which are defined by the engineer. ek+s
represents the error between the output and reference signal at each time t = k.
Typically the value of qs will be large so that the error ek+s will be small which means
the output signal can track the reference signal perfectly. Sometimes it is also necessary
for the engineer to take the value of rs into consideration because it is impossible for the
physical system to provide very large control input. Then based on the matrix form, each
term in the cost function can also be presented as
Yk+1 rk+1
ek+1
Y
k+2 rk+2
ek+2 .
.
e =
. =
ek+Np
Yk+Np rk+Np
51
=Y R
(513)
q I
1
q2 I
Q=
qNp I
,P =
p1 I
p2 I
.
(514)
pN p I
where the matrix I is the identity matrix. Based on these matrices, the cost function can
be presented as
J=
1
1
[Ex(k) + FU R]T Q[Ex(k) + FU R] + U T PU
2
2
(515)
In order to calculate the best control input U to minimize the cost function J, take the first
derivative of the cost function J in terms of control input U, and let it equal to zero.
J
=0
U
(516)
F T Q(Ex(k) + FU R) PU
=0
F T QEx(k) + F T QFU F T QR + PU = 0
(517)
Expanding the equation above and solving for the vector U gives
U = (F T QF + P)1 (F T QR F T QEx(k))
(518)
]
1 0 0. . . 0
(519)
Substituting u(k) into the following equation to obtain x(k+1) and Y(k+1) gives
x(k + 1) = Am x(k) + Bm u(k)
Y (k)
= Cm x(k)
52
(520)
The x(k+1) is obtained by calculating the control input u(k) at time t = k. Repeating the
steps, control input u(k),u(k+1),u(k+2)...u(k+N) can be obtained. Then corresponding
states and outputs can also be obtained.
Based on the theory described above, the suspension system with the hydraulic
system can be taken into consideration. The dynamic equation has been shown in
Equation 3-2. Obviously, the suspension system is a nonlinear system because of the
hydraulic system. Thus it is necessary to simplify the nonlinear system. In order to
simplify the nonlinear dynamic system, a virtual control signal should be introduced,
which is shown as follows
X 1 = X2
Ks
X1
X 2 = m
s
bs
X
ms 2
Ks
X
ms 3
bs
X
ms 4
A
X
ms 5
X 3 = X4
X 4 =
Ks
X
mu 1
(521)
+
bs
X
mu 2
Kt
(m
+
u
Ks
)X3
mu
bs
X
mu 4
A
X
mu 5
Kt
w
mu
X 5 = X5 A(X2 X4 ) + u
Comparing to the previous nonlinear dynamic equation, the virtual control signal is
u = X6
Ps sgn(X6 )X5
(522)
Up to now, the suspension system with the hydraulic actuator is a linear continuous
system. The c2d command can be utilized to discretize the continuous dynamic system.
But in this practical suspension system, a road disturbance w is taken into consideration.
The road disturbance is same as before. Then the dynamic system can be presented as
]
[
u
(523)
X = AX + B Bd
53
where
ks
ms
A=
0
ks
mu
mbss
ks
ms
bs
ms
A
ms
bs
mu
kt
mu
ks
mu
A
0
B = 0 , Bd =
mbsu mAu
(524)
kt
mu
0
(525)
By using the same method mentioned above, the prediction states can be presented as
(527)
X = E0 x(k) + F0 U + G0 W
where
G =
Bdm
Am Bdm
,W =
Bdm
.
.
.
N 1
Amp
N 2
Bdm Amp
N 3
Bdm Amp
Bdm . . . Bdm
w (k)
w ((k + 1)
.
.
.
w (k + np 1)
(528)
1
((Y R)T Q(Y R) + U T PU).
2
54
(529)
Here
(530)
Y = Cm X
And X has been define in Equation 5-27. Meanwhile, in a practical system, the reference
signal is also changing with states. Thus it is necessary to get the states prediction
at each time t = k and substitute into the reference signal equation and update the
reference signal at each time t= k. Then take the derivative of the cost function and let
the derivative of J in terms of U equal 0. Now the equation of U is
U = (F T QF + P)1 (F T QR F T QEx(k) F T QGx(k)).
(531)
Up to now, the control input for each time t = k has been calculated.
5.3 Simulation
Based on the design theorem described above, the simulation of the suspension
system with the hydraulic actuator using model predictive control will be shown as
follows. Assuming the road disturbance is the same as before, the plots for the actuator
force, car body position, velocity, unsprung mass position and velocity are shown in
Figure 5-1 through 5-5.
4000
Reference Force
Output
Force (N)
3000
2000
1000
0
1000
2000
0
Time (s)
55
10
15
According to Figure 5-1, it is not hard to see that the force generated by the
suspension system with hydraulic actuator can track the ideal force. The tracking
performance is not as good as sliding mode control. But as time goes on, the tracking
performance is getting better. Thus model predictive control can still achieve tracking
performance and minimize the tracking error.
1.5
1
0.5
0
0.5
1
1.5
0
Time (s)
10
15
1.5
1
0.5
0
0.5
1
0
Time (s)
10
15
1.5
Model Predictive Control
Passive System
Sliding Mode Control
1
0.5
0
0.5
1
0
Time (s)
10
15
1.5
Model Predictive Control
Passive System
Sliding Mode Control
1
0.5
0
0.5
1
0
Time (s)
10
15
57
which means, the passenger will not feel vibration heavily. Also the passive suspension
system vibrates a lot. Moreover, the suspension system by using model predictive
control vibrates a little longer than the suspension system using sliding mode control.
However the peak value for the model predictive control based suspension system is
smaller. Similarly, the unsprung mass position and velocity also show the same results.
Based on these figures, the advantage of the suspension system with the hydraulic
actuator which can be controlled to generate active force is obvious compared to the
passive suspension system. In addition, for the two control methods, there is another
trade off. If the practical condition requires a smaller peak value, then model predictive
control should be utilized. But if less time of vibration is required, then sliding mode
control should be used.
4
Control Input
4
0
Time (s)
10
15
Figure 5-6 shows the virtual control signal which is determined from Equation 5-22.
The value of control input is also reasonable. But sometimes the system will have some
constraints about the control input, because it is impossible for a practical system to
provide very large control input to get the ideal control result. Thus it is also necessary
to use model predictive control to get the control results with some constraints on the
control input.
58
(532)
Y (k) = Cm x(k)
(533)
(534)
where the values of u1 and u2 are determined by the physical system. Then according to
Equation 5-33, the inequality equation can be represented in matrix form as follows
1
u1
(535)
1
u2
Further let
1
u1
M=
,b =
1
u2
(536)
According to the cost function defined in Equation 5-15, the problem becomes
minU
1
[Ex(k)
2
+ FU R]T Q[Ex(k) + FU R] + 12 U T PU
s.t.Mu b
(537)
(538)
Now the problem become to dual problem. Expanding the cost function gives
J = U T HU + U T F1 + (MU b)
59
(539)
where
H = F T QF + P
F1 = F T QEx(k) F T QR
(540)
Taking the first derivative of the new cost function in terms of u gives
U = H 1 (F1 + M T )
(541)
(542)
1
1
min( T H + T K + F T E 1 F )
0 2
2
(543)
As shown in [43], the minimum problem can be solved by Hildreths quadratic programming.
After solving for , which is defined as a , the U can be presented as
U = E 1 F E 1 MaT a .
(544)
Up to now, the control input for the current time t = k has been obtained. Then the
following steps are similar to the problem without constrains which has been described
above.
Assume the control input is constrained between -1 and 1. The for the same
system, the plots for he states are shown as in Figure 5-7 through 5-10.
From Figure 5-7 to Figure 5-10, the system with control input constraints has a
larger peak value and settles down more slowly. It is because the control input of the
suspension system is bounded between -1 and 1 which means less control input can be
applied. Thus it will definitely take a longer time for the system to settle down. For the
system with constraints, the control input is shown in Figure 5-11.
In Figure 5-11, the control input is bounded between -1 and 1 as expected, which
means every time when the control input is out of the range determined by physical
60
0.2
No Constraint
With Constraint
0.1
0.1
0.2
0
Time (s)
10
15
1.5
No Constraint
With Constraint
1
0.5
0
0.5
1
1.5
0
Time (s)
10
15
No Constraint
With Constraint
0.1
0.05
0.05
0
Time (s)
61
10
15
1.5
No Constraint
With Constraint
1
0.5
0
0.5
1
1.5
0
10
Time (s)
15
Control Input
1
0.5
0
0.5
1
1.5
0
10
20
Time (s)
30
40
50
62
CHAPTER 6
CONCLUSION AND FUTURE WORK
6.1 Conclusion
The design and control of a suspension system is developed in this thesis. In
Chapter 2, a passive suspension system, a suspension system with skyhook, a
suspension system with a nonlinear energy sink, and a suspension system combined
with nonlinear energy and skyhook are described. According to the variable gain for
the car body velocity and acceleration, the combined suspension system with nonlinear
energy sink and skyhook shows the best performance.
Based on the conclusion obtained from Chapter 2, in Chapter 4 the sliding mode
control is applied to let the suspension system with hydraulic actuator track the ideal
force generated from the combined suspension system with nonlinear energy sink and
skyhook. The performance of the four states including sprung mass position, velocity,
unsprung mass position and velocity are worked on. From these plots about the states,
the sliding mode control can indeed control the hydraulic actuator to generate the
expected force.
In Chapter 5, model predictive control is applied to the suspension system with
the hydraulic actuator. The control input is designed based on the model predictive
control principle to minimize the tracking error between the actual force generated by
hydraulic actuator and ideal force obtained by the combined suspension system with
nonlinear energy sink and skyhook. The plots of states are given to show the tracking
performance of model predictive control. Moreover, a comparison between the sliding
model control and model predictive control is also given from these plots. Sliding mode
control and model predictive control show their own advantages. Thus it is necessary
for the designer to take the practical requirements into consideration and then choose
the best control method. Meanwhile the tracking problem with constrained control input
is also taken into consideration. The plots for states with constrained control input show
63
that with less control input, the performance of passenger comfort and road holding
decrease to some extent. However it also shows the advantage of solving the control
system with constraints for model predictive control.
6.2
Future Work
The dynamic system worked on in this thesis is based on the exact knowledge
about the suspension system which means all parameters are known and constant.
However, in practical systems, lots of parameters will change with the condition
and sometimes the parameters in the system are unknown. Thus the future work
will focus on the adaptive model predictive control for the suspension system with
hydraulic actuator. In the adaptive model predictive control, the goal is the same as
beforeminimize the tracking error between the suspension system with hydraulic
actuator and combined suspensions system. But at the same time, some parameters
about the hydraulic system are unknown. Thus, an adaptive control law will be designed
and a new Lyapunov Function will be defined and used to analyze the stability of the
whole suspension system.
64
REFERENCES
[1] O.M.Anubi, D.R.Patel, and C.Crane, A New Variable Stiffness Suspension System:
Passive case, Mechanical Sciences 4(2013),pp. 139-151.
[2] Abdolvahab Agharkali,Ghobad Shafiei Sabet and Armin Barouz, Simulation
and Analysis of Passive and Active Suspension System Using Quarter Car
Model for Different Road Profile, International Journal of Engineering Trends
and Technology-Volume3Issue5-2012
[3] R.Alkhatib,G.N.Jazar, and M.F. Golnaraghi, Optimal design of Passive Linear Suspension Using Genetic Algorithm, Journal of Sound and vinration
275(2004),pp.665-691.
[4] A.C.M.Knaapvan der,A.P.Teerhuis,R.B.G.Tinsel,and R.M.A.F.Vershuren, Active
Suspension Assembly for a Vehicle, Internstional Patent No.2008/049845(2008).
[5] D.Karnopp, Active damping in road vehicle suspension systems, Vehicle System
Dynamics 12(1983),pp.291-316
[6] R.A.Williams, Automotive active suspensions. Part 1:basic principles , in
Proceedings of IMechE,Vol.211,1997.pp.415-426.
[7] A.Alleyne and J.Hedrick, Nonlinear adaptive control of active suspensions, Control
Systems Technology, IEEE Transactions on 3(1995), pp. 94-101.
[8] E. Kim, Nonlinear indirect adaptive control of quarter car active suspension, in
Control Applications, 1996, Proceedings of the 1996 IEEE International Conference
on, 1996,pp.61-66.
[9] R.A. Williams, A. Best, and I.L.Crawford, Refined Low Frequency Active Suspension, In Proceedings of IMechE International Conference, November, 1993.
pp.285-300.
[10] I. Fialho and G.J.Balas, Road Adaptive Active Suspension Design Using Linear
Parameter-Varying Gain-Scheduling, IEEE Transactions on Control Systems
Technology 10.(2002), pp. 43-54
[11] W.J.Evers,A.Teehuis,A.Knaapvan der, I. Besselink, and H.Nijmeijer, The Electromechanical Low-power Active Suspension: Modeling,Control and Prototype Testing,
Journal of Dynamic Systems,Measurement, and Control 133(2011).
[12] D.Karnopp, M.Crosby, and R.Harwood, Vibration control using semi-active force
generators, Journal of Engineering for Industry 96(1974),pp.619-626
[13] M.Valasek and W.Kortum, Nonlinear control of semi-active road-friendly truck
suspension, in Proceedings AVEC 98, Nagoya,1998,pp.275-280.
65
66
67
[41] Faraz Ahmed Andari, RajShree Taparia Modeling, Analysis and Control of Active
Suspension System using Sliding Mode Control and Disturbance Observer, ASME
Vol. 122, December 2000
[42] A. F. Vakakis, L. Manevitch, O. Gendelman, and L. Bergman, Dynamics of Linear
Discrete Systems Connected to Local , Essentially Nonlinear Attachments, Journal
of Sound and Vibration 264(2003),pp. 559-577.
[43] A. Zin, O. Sename, M. Basset, L.Dugard, and G.Gissinger, A nonlinear vehicle bicycle model for suspension and handling control studies, in Proceedings
of the IFAC Conference on Advances in Vehicle Control and Safety(AVCS),
Genova,Italy,October,2004,pp.638-643.
[44] Liuping Wang, Model Predictive Control System Design and Implementation Using
MATLAB, Springer 2009.
[45] Luigi del Re, Frank Allgower, Luigi Glielmo, Carlos Guardiola, LLya Kolmanovsky
Automotive Model Predictive Control: Models, Methods and Applications, Springer
2010.
[46] arlos E. Garcia, David M. Prett, Manfred Morari Model Predictive Control: Theory
and Practicea Survey, Automatica, Vol. 25, No.3, pp. 335-248, 1989.
[47] .Q.Mayne, J.B.Rawlings, C.V.Rao, P.O.M.Scokaert Constrained model Predictive
control: Stability and optimality, Automatica 36 (2000) 789-814.
[48] Christoph Gohrle, Andreas Schindler, Andreas Wagner and Oliver Sawodny Model
Predictive Control of Semi-active and active suspension systems with available
road preview, in 2013 European Control Conference.
[49] Hiroaki Fukushima, Tae-Hyoung Kim, Toshiharu Sugie Adaptive Model Predictive
Control for a class of constrained linear systems based on the comparison model,
Automatica 43 (2007) 301-308
[50] D.Q.Mayne, M.M.Seron, S.V.Rakovic Robust Model Predictive Control of Constrained Linear Systems with Bounded Disturbances, Automatica 41 (2005)
219-224
[51] D.W.Clarke Adaptive Predictive Control, International Federation of Automatic
Control 1997 A.Rev. Control, Vol.20,pp. 83-94,1996
[52] D.Karnopp and G.Heess, Electronically controllable vehicle suspensions, Vehicle
System Dynamics 20(1991),pp.207-217.
[53] Cristiano Spelta, Fabio Previdi, Sergio M. Savaresi, Paolo Bolzern, Maurizio
Cutini, Carlo Bisaglis, Simone A. Bertinotti Performance analysis of semi-active
suspensions with control of variable damping and stiffness, Vehicle System
Dynamics 2010, 1-20,iFirst.
68
[54] Sergio M. Savaresi, Enrico Silani, Sergio Bittanti, Nicola Porciani On performance
evaluation methods and control strategies for semi-active suspension system,
Vehicle System Dynamics 20(1991),pp.207-217.
69
BIOGRAPHICAL SKETCH
Jie Fang received her B.S in Mechanical Engineer from the Shandong University,
Jinan, Shandong in 2012. She is currently completing her Master of Science degree
in Mechanical Engineering at the Center for Intelligent Machines and Robotics, at the
University of Florida, Gainesville, FL. Her research interests are: Vehicle suspension
system, Dynamics, System and Control, Nonlinear Control, Adaptive Control, Model
Predictive Control, Sliding Mode Control.
70