Keum-Shik Hong

School of Mechanical Engineering

e-mail: kshong@hyowon.pusan.ac.kr

Hyun-Chul Sohn
Department of Mechanical and Intelligent
Systems Engineering
e-mail: hcson@hyowon.pusan.ac.kr
Pusan National University,
30 Changjeon-dong, Kumjeong-ku,
Pusan 609-735, Korea

J. Karl Hedrick
Department of Mechanical Engineering,
University of California Berkeley,
Berkeley, CA 94720-1740
e-mail: khedrick@me.berkeley.edu

Modified Skyhook Control of

Semi-Active Suspensions: A New
Model, Gain Scheduling, and
Hardware-in-the-Loop Tuning
In this paper, a road adaptive modified skyhook control for the semi-active Macpherson
strut suspension system of hydraulic type is investigated. A new control-oriented model,
which incorporates the rotational motion of the unsprung mass, is introduced. The control
law extends the conventional skyhook-groundhook control scheme and schedules its gains
for various road conditions. Using the vertical acceleration data measured, the road
conditions are estimated by using the linearized new model developed. Two filters for
estimating the absolute velocity of the sprung mass and the relative velocity in the rattle
space are also designed. The hydraulic semi-active actuator dynamics are incorporated in
the hardware-in-the-loop tuning stage of the control algorithm developed. The optimal
gains for the ISO road classes are discussed. Experimental results are included.
DOI: 10.1115/1.1434265
Keywords: Macpherson Strut Suspension, Modeling, Skyhook Control, Semi-Active
Damper, Gain Scheduling, Rapid Control Prototyping, Road Estimation

Introduction

The roles of a suspension system are to support the vehicle

weight, to isolate the vehicle body from road disturbances, and to
maintain the traction force between the tire and the road surface.
Suspension systems are classified into a passive system and an
active system according to the existence of control input. The
active suspension system can be further classified into two types:
a semi-active system and a fully active system according to the
control input generation mechanism. The semi-active suspension
system uses a varying damping force as a control force. For example, a hydraulic semi-active damper varies the size of an orifice
in the hydraulic flow valve to generate desired damping forces. An
electro-rheological ER damper or a magneto-rheological MR
damper applies various levels of electric field or magnetic field to
cause various viscosities of the ER or MR fluids. On the other
hand, the fully active suspension system produces the control
force with a separate hydraulic/pneumatic unit. Therefore, the cost
and the weight of a fully active suspension system are much
higher than those of a semi-active one. Semi-active suspension
systems are getting more attention because of their low cost and
competitive performance to the fully active ones. In this paper, a
road adaptive modified skyhook control for the semi-active
Macpherson strut suspension system of hydraulic type is
investigated.
The performance of a suspension system is characterized by the
ride quality, the handling performance of vehicle, the size of the
rattle space, and the dynamic tire force. The prime purpose of
adopting an active/semi-active suspension system is to improve
the ride quality and the handling performance of vehicle. To improve the ride quality, it is important to isolate the sprung mass
from the road disturbances and to suppress the vertical vibrations
near 5 Hz 4 8 Hz, which is known to be a sensitive frequency
range to human body lateral vibrations at 12 Hz according to
ISO 2631. On the other hand, to improve the handling performance of vehicle, it is important to keep the tire in contact with
Contributed by the Dynamic Systems and Control Division for publication in the
JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT, AND CONTROL. Manuscript
received by the Dynamic Systems and Control Division May 1, 2000. Associate
Editor: S. Sivashankar.

158 Vol. 124, MARCH 2002

the road surface and therefore to decrease the resonant peak near
10 Hz, which is the resonant frequency of the unsprung mass. For
a fixed suspension spring constant, the better isolation of the
sprung mass from the road disturbances can be achieved with a
soft damping by allowing a larger suspension deflection. However, the better road contact can be achieved with a hard damping
by not allowing unnecessary suspension deflections. Therefore,
the ride quality and the handling performance of vehicle are two
conflicting criteria in the control system design of suspension
systems.
Since the skyhook control strategy was introduced by Karnopp
et al. 1, in which a fictitious damper is inserted between the
sprung mass and the stationary sky as a way of suppressing the
vibratory motion of the sprung mass and as a tool to compute the
desired damping force, a number of innovative control methodologies have been proposed to implement this strategy. The skyhook control can reduce the resonant peak of the sprung mass
quite significantly and thus achieves a good ride quality. But, in
order to improve both the ride quality and the handling performance of vehicle, both resonant peaks of the sprung mass and the
unsprung mass need to be reduced. It is known, however, that the
skyhook damper alone cannot reduce both resonant peaks at the
same time. From this point of view, Besinger et al. 2 proposed a
modification of the skyhook control, which includes both a passive damper as well as a skyhook damper, for the computation of
desired control inputs. Novak and Valasek 3 have proposed a
groundhook control, which assumes an additional fictitious
damper between the unsprung mass and the ground, for the purpose of decreasing the dynamic tire force.
Compared to the various control techniques appeared in the
literature, the issues related to the modeling of suspension systems
are rare. Jonsson 4 conducted a finite element analysis for evaluating the deformations of suspension components. Stensson et al.
5 proposed three nonlinear models for the Macpherson strut
wheel suspension for the analysis of motion, force, and deformations. These models would be appropriate for the analysis of mechanics, but are not adequate for control purpose. In the conventional quarter car model Yue et al. 6, only the up-down
movements of the sprung and unsprung masses are assumed. In
the conventional model, the role of the control arm is completely

Copyright 2002 by ASME

Transactions of the ASME

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

ignored. From this point of view, a new control-oriented model,

which includes the rotational motion of the control arm, is
introduced.
As an actuator for generating the semi-active control force, a
continuously variable damper CVD is used. Hence, the control
force is adjusted by changing the size of an orifice of the CVD.
The damping force characteristics of a CVD are highly nonlinear.
Also, the CVD used in this work is designed in such a way that it
produces much larger damping force in the extension motion than
in the compression motion. It is also noted that with a semi-active
suspension the control action is applied only when the control
force is opposite to the direction of suspension relative velocity.
One way of designing a control system for the semi-active suspension is to figure out all the nonlinearities of the plant and
actuator and then to design an appropriate control law based upon
both the vehicle dynamics and the actuator dynamics. For this, the
fluid dynamics of the variable damper has to be investigated.
However, it is not simple to know the complete nonlinear characteristics of the variable damper. Also, considering the fact that the
entire control problem is to stabilize a multiple-degree underactuated mechanical system with one saturating actuator in the presence of unknown road disturbances, see Fig. 3 for control problem
formulation, the efforts to know the complete nonlinear dynamics
of the actuator is not so cost-effective.
Another way is to isolate the most significant nonlinear part of
the actuator, which is the variable damper in our case, and to use
the real one instead of its mathematical model. In this work, the
control law design does not consider the actuator dynamics. However, in the final tuning stage of the control gains, the real actuator
is included in the hardware-in-the-loop simulations.
Recently, the computer aided control system design CACSD
has been the subject of focus in control implementation. The
CACSD is often named as the rapid control prototyping RCP or
the hardware-in-the-loop simulations HILS Hanselmann 7,8.
In the RCP, the plant dynamics and/or the control laws are implemented in a digital signal processing board, which allows an easy
adjustment of various parameters of the plant and/or the controller. If actual hardware is used as a part of simulation loop, the
term HILS is particularly used. Through the CACSD, the total
development time and cost can be much reduced. Also, it is easy
to introduce a new component or a new algorithm because the test
procedure can be easily repeated.
In this paper, assuming the use of a single acceleration sensor
for the quarter-car model and a 16-bit microprocessor, a modified
skyhook control with gain scheduling is investigated. Contributions of this paper are: First, a new control-oriented model for the
semi-active Macpherson suspension system is derived. Second, a
control structure extending the skyhook control and including the
estimation of road conditions is suggested. Third, the optimal
gains for the ISO road classes and the gain scheduling procedure
for estimated roads are developed. Fourth, the procedure for incorporating the actuator dynamics via the hardware-in-the-loop
simulations is discussed. One final comment is that four sensors
three vertical sensors and one lateral sensor for the entire vehicle
control are used, but the issues beyond the quarter-car model are
not discussed in this paper.
This paper is structured as follows. In Section 2, a new model
for the semi-active Macpherson suspension system is derived. In
Section 3, the control problem is formulated, the skyhook control
is modified, road profiles are characterized, optimal gains for the
ISO road classes are determined, the gains for the estimated road
and two filters for controller variables are designed. In Section 4,
the performance of a controlled semi-active system is evaluated
using the hardware-in-the-loop simulations. Conclusions are given
in Section 5.

the unsprung mass. For the brevity of this paper, the detailed
assumptions made for this model are referred to Hong et al. 9. If
the joint between the control arm and the car body is assumed to
be a bushing and the mass of the control arm is included, the
degree-of-freedom of a 1/4-car system is four. The generalized
coordinates in this case are z s , d, 1 , and 2 . However, if the
mass of the control arm is neglected and the bushing is assumed to
be a pin joint, then the degree-of-freedom becomes two, reducing
the generalized coordinates from four to two, z s and as depicted
in Fig. 2. In Hong et al. 9, the equations of motion for this
model are derived. In this section, the derivation in Hong et al. 9
is modified to fit to the control problem of the semi-active suspension system with a continuously variable damper.
The definitions of the parameters and variables in Fig. 2 are:
Parameters Data: m s is the sprung mass 453 Kg; m u is the
unsprung mass 36 Kg; k s is the spring constant of the coil spring
17,658 N/m; k t is the spring constant of the tire 183,887 N/m;
l A is the distance from O to A 0.66 m; l B is the distance from O
to B 0.34 m; l C is the length of the control arm 0.37 m; is the
angle between the y-axis and OA 74 deg; 0 is the angular
displacement of the control arm at a static equilibrium point 2
deg.
Variables: z s is the vertical displacement of the sprung mass;
is the angular displacement of the unsprung mass; l is the relative displacement in the rattle space, and finally f s is the control
damping force generated by the continuously variable damper.

Fig. 1 A schematic diagram of the Macpherson suspension

system: 14-car model

Modeling of 14 Car: A New Model

A schematic diagram of the Macpherson strut suspension system is shown in Fig. 1. This model admits the rotational motion of
Journal of Dynamic Systems, Measurement, and Control

Fig. 2 A new model including the continuously variable damping force

MARCH 2002, Vol. 124 159

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Let T, V, and D denote the kinetic energy, the potential energy

and the damping energy of the system, respectively. Then, they
are

The linearization of 3 at an equilibrium

(x 1e ,x 2e ,x 3e ,x 4e )(0,0, 0 ,0) yields:
x t Ax t B 1 f s B 2 z r t ,

T 21 m s m u z s2 21 m u l C2 2 m u l C cos 0 z s ,
V 2 k s 2a l b l cos cos 2 a l2 a l b l cos
1

k t z s l C sin 0 sin 0 z r ,
1
2

c p b l2 sin2 2
1
c p l 2
,
2
8 a l b l cos

where a l l A2 l B2 , b l 2l A l B , 0 , c l a l2 a l b l cos(),
and d l a l b l b l2 cos().
For the two generalized coordinates q 1 z s and q 2 , the
equations of motion are:

m s m u z s m u l C cos 0 m u l C sin 0 2

k t z s l C sin 0 sin 0 z r 0,
m u l 2c m u l C

b l

(1)

cos 0 z s k t l C cos 0 z s l C sin 0

sin 0 z r

1
2

k s sin

dl
c l d l cos 1/2

l B f s .

(2)

where
f 1

x 2 f 1 x 1 ,x 2 ,x 3 ,x 4 , f s ,z r ,

x 3 x 4 ,

x 4 f 2 x 1 ,x 2 ,x 3 ,x 4 , f s ,z r ,

f1
x1

f1
x2

f1
x3

f1
x4

f2
x1

f2
x2

f2
x3

f2
x4

a 21

a 23

a 41

a 43

0
1

21.177

13796

5105.4

m 2u l C2 sin x 3 0 cos x 3 0 x 24

g 1 x 3 m s l C m u l C sin2 x 3 0 ,

z z x 1 ,x 2 ,z r x 1 l C sin x 3 0 sin 0 z r .
160 Vol. 124, MARCH 2002

T
xx e

0
k t l C sin2 0
m s l C m u l C sin2 0
0
m s k t l C cos 0
m s m u l C2 m 2u l C2

sin 0
2

0
0.494

,
0
13796

cos x 3 0 z m s m u l B f s ,

dl
g 3 x 3 b l
,
c l d l cos x 3 1/2

f1 f2
0
zr zr

m s m u k s sin x 3 g 3 x 3 m s k t l C

g 2 x 3 m s m u l C2 m 2u l C2 sin2 x 3 0 ,

xx e

m s m u l C2 m 2u l C2 sin2 0

l B f s cos x 3 0 ,

1
2

0
l B cos 0
m s l C m u l C sin2 0
0
m s m u l B

B 2 0

sin x 3 cos x 3 0 g 3 x 3 k t l C sin x 3 0 z

f1 f2
0
fs fs

0
0.002

,
0
0.074

(3)

1
g 2 x 3

(4)

xx e

B 1 0

1
1
m u l C2 sin x 3 0 x 24 k s
g 1 x 3
2

f 2

0.49437

Remark 1: The semi-active damping force, f s , in 2 includes

both the passive damping force and the control force. However, in
Eq. 9 of Hong et al. 9, there appear two explicit terms, i.e., a
passive damping coefficient c p as well as a control input term f a .
The equivalence of 2 and Eq. 9 of Hong et al. 9 is shown as
follows: If we set c p 0 and f a f s in 9, 9 and 3 become
identical. Also, if we set c p c s and f a 0 in 9, the application
l/ l c l
of the relationship such that 1/2 D/ c s l
B s
l B f s to 9 yields 3.
Now, define the state variables as x 1 x 2 x 3 x 4 T z s z s T .
Then, 12 can be rewritten as follows:
x 1 x 2 ,

xe

where

cos b l2 cos cos 1/2

x 0 x 0

point

and
a 21

k t l C sin2 0
,
m s l C m u l C sin2 0
Transactions of the ASME

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

a 23

m s l C m u l C sin2 0 2
cos 0

dl
1
k b
2 s l c l d l cos 1/2

d l2 sin
1
k s sin cos 0
2
2 c l d l cos 3/2

k t l C2 sin2 0 cos 0 m s l C m u l C sin2 0

m u k s l C sin sin 0 cos2 0

b l

dl
c l d l cos 1/2

a 41
a 43

m s k t l C cos 0
,
m s m u l C2 m 2u l C2 sin2 0

m s m u l C2 m 2u l C2

sin 0
2

dl
b l

c l d l cos 1/2
sin

d l2

sin

2 c l d l cos 3/2

1
2

1
m m u k s cos
2 s

z s
F 1 z s
.
y f ilter ,
F 2 z s
l

m s m u k s

m s k t l C2 cos 0

m s m u l C2 m 2u l C2 sin2 0
sin sin 0 b l

Fig. 3 Control block diagram of the semi-active suspension

system: a road adaptive modified skyhook control

Figure 3 depicts the entire control structure, in which the filtering

of z s and l from z s and the road estimation using z s are clearly
emphasized. The road estimation will be discussed in Section 3.5
and the filter design will be discussed in Section 3.6.

1
m m u m 2u k s l C2
2 s
dl

c l d l cos 1/2

3 Control Design: A Road Adaptive Modified Skyhook

Control

Remark 2: Comparing the linearized Eq. 4 with the conventional model Yue et al. 6, in which only the vertical movement
of both the sprung and the unsprung masses is considered, the
transfer function of the conventional model and that of 4 become
identical if l B l C , l B l A cos , and 0 0 deg. Therefore, the
conventional model is a special case of this new model in the
sense that the same transfer function can be achieved by restricting l B l C and 0 0. For the detailed comparison, Hong et al. 9
is referred.
Now, because z s is measured, the output equation by defining
yz s is derived as follows:
y t Cx t D 1 f s D 2 z r ,
where
C a 21 0 a 23 0 ,
and
D 2

D 1

(5)

l B cos 0
0.002 ,
m s l C m u l C sin2 0

k t l C sin2 0
0.494 .
m s l C m u l C sin2 0

Remark 3: The variables needed for the computation of control

input are the absolute velocity of the sprung mass, z s , and the
relative velocity of the sprung and unsprung masses l. In Section 3.6, both z s and l are filtered from the signal z s measured.
Recall that the use of only one acceleration sensor is assumed in
this work. However, during the stage of hardware-in-the-loop
simulations of the control laws designed, the following formula,
Eq. 5b of Hong et al. 9, can be used.
l

b l sin
, f , .
2 a l b l cos 1/2

(7)

(6)

Finally, to put the control structure in a feedback form, i.e., the

reference signal is z s l T 0 0 T and its error is regulated, a
filtered output equation is defined as follows:
Journal of Dynamic Systems, Measurement, and Control

A number of papers investigating advanced control techniques

such as nonlinear adaptive control, preview control, multivariable
system decoupling techniques, and robust control have appeared
in the literature Alleyne et al. 10, Yi and Hedrick 11, Alleyne
and Hedrick 12, Hac 13, Kim and Yoon 14, El-Demerdash
and Crolla 15, Lin and Kanellakopoulos 16, Cherry et al. 17,
Choi et al. 18. However, if a state feedback control strategy is
adopted, either sufficient sensors for the whole state variables or
an estimate of the state vector is required Hedrick et al. 19, Yi
and Hedrick 20.
In this paper, the use of only one acceleration sensor for measuring the sprung mass vertical vibrations and a 16-bit microprocessor for implementing the control algorithm is assumed. Therefore, the control law design focuses on the practicality,
implementability, and robustness of the algorithm rather than the
perfection of performance.
The two control objectives are the improvement of both the ride
quality and the handling performance of vehicle. If fixed control
gains are used, these two conflicting objectives cannot be
achieved. However, by adapting road conditions, i.e., by changing
controller gains for various road conditions, both objectives can
be achieved.
3.1 Controller Structure. In this paper, as shown in Fig. 3,
a modified skyhook control with gain scheduling, in the form of
an output feedback control, is proposed. The measured output is
z s . The road disturbance is z r . It is noted that the control law, a
modified skyhook control, is not based upon a mathematical
model of the plant and therefore the control structure is not a
model-based control. However, the mathematical model developed in Section 2 will be used in filtering controller parameters
(z s , l) and estimating road conditions (z r ). The reason for
adopting the structure of Fig. 3 is justified as follows: The entire
control problem is to stabilize a two-degree-of-freedom system
with a single actuator involving saturation. The actuator dynamics
is complicated and its performance is limited. Furthermore, the
two conflicting control objectives cannot be satisfied with fixed
gains. Therefore, a heuristic algorithm rather than a control algorithm based upon a precise mathematical model is suggested. In
this formulation, the desired reference signals are set to z s 0 and
MARCH 2002, Vol. 124 161

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 l0. Now, after a brief discussion on the control law, the issues
related to the road estimation, gain scheduling, and filtering are
described in Sections 3.33.6, respectively.
1) Ideal Skyhook Control. Among the many control methods
developed, the skyhook control introduced by Karnopp et al. 1
is known most effective in terms of the simplicity of the control
algorithm. Their original work uses only one inertia damper between the sprung mass and the inertia frame. The skyhook control
is applicable for both a semi-active system as well as an active
system. However, the original strategy did not pay attention to the
unsprung mass vibrations and therefore might deteriorate the handling performance of vehicle due to the excessive vibrations of the
unsprung mass. In order to overcome the demerits of the original
skyhook control, various modified approaches have been proposed
in the literature Besinger et al. 2, Novak and Valasek 3.
2) Skyhook-Groundhook Control. The groundhook control
by Novak and Valasek 3, which assumes an additional inertia
damper between the unsprung mass and the ground, was proposed
for the purpose of reducing excessive motion of the unsprung
mass. This strategy can compromise the two conflicting criteria,
the ride quality and the handling performance of vehicle. From
Fig. 2, the control force of the skyhook-groundhook model
becomes
uc sky z s c p lc gro z s l C

(8)

where c sky , c p , and c gro are the damping coefficients of the skyhook damper, the passive damper, and the groundhook damper,
respectively. However, because the value of has to be used, this
strategy is not suitable in our case.
3) A Modified Skyhook Control With Gain Scheduling. A
modified skyhook control, which proposes the inclusion of a variable damper besides a skyhook damper, can still achieve both
control objectives by assigning most appropriate damping coefficients at various road conditions. The following law is proposed:
uc sky z r z s c v z r l

(9)

where both the skyhook gain c sky and the variable damper gain c v
depend on the road estimate z r . The scheduling of gains is discussed in Section 3.5. Even though the calculation of control input
is carried out by 9, the generation of control input should be
carried by the variable damper itself. Also, because of saturation,
the actual control force from a CVD is limited as follows:

f s

f s* ,

if f s* u

u,

if f s * u f s*

f s*

if f s * u

(10)

where f s* and f s * denote the maximum and the minimum damping forces available at a given relative velocity. For large control
inputs, the actuator saturation would be unavoidable. Therefore,
the control performance of a semi-active damper would be less
perfect in severe road conditions.
The characteristics of a typical CVD are depicted in Fig. 4. The
solid line in the middle section denotes the damping force characteristics of a typical passive damper. The dotted line of highest
slope denotes the characteristics for 0 Ampere current input,
which is the most hard case. The triangles overlaid on top of the
dotted line represent the lookup table values of the maximum
damping force. The dotted line of lowest slope denotes the minimum damping force characteristics for 1.6 Ampere current input.
Again, the squares denote the lookup table values of the minimum
damping force. The damping forces in the extension region the
first quadrant in Fig. 4 are larger than those in the compression
region the third quadrant.
3.2 Current Generation. The damping force generated in a
hydraulic-type semi-active damper depends on two things: the
size of valve opening, i.e., the current input to the solenoid valve,
and the relative velocity in the rattle space. To determine the current input to the solenoid valve, it is necessary to know the damping force characteristics of the valve versus the current input at a
given relative velocity. For this, two approaches can be pursued.
One is an analytic approach, which investigates a mathematical
expression for the entire hydraulic system including the cylinder
and valve dynamics. However, the mechanism of a semi-active
damper is very complicated and the damping force characteristics
in the expansion and compression strokes are different because the
volumes of compression and reaction chambers are different and
some valves allow the fluid to flow only in one direction. It is also
difficult to measure the parameter values and furthermore the viscosity of oil and gas mixture may vary in temperature and time.
Another approach is an experimental approach, which is more or
less straightforward. The damping forces for various input currents for a specific relative velocity can be measured in a test rig.
In this paper, the experimental approach is adopted.
The experimental data can be either tabulated as a look-up table
or approximated as a polynomial equation by using the least
squares method. After dividing the relative velocity range into
four different sections, the polynomial equations made for individual sections are tabulated in Table 1.
As far as the final results are concerned, as seen in Fig. 4 where
the number of cases for the lookup table is N p 51, there is not
much difference between these two approaches. However, a
smaller access time is possible with the polynomial approach.
This is very important because the entire control algorithm should
be coded in a 16-bit microprocessor. Figure 5 compares the access
time of the two methods for the relative velocity range within
1.4 m/s. It is seen that the access time to the polynomial equation is constant, but the access time to use the look-up table in-

Table 1 Polynomial representation of maximumminimum

damping forces

Fig. 4 Damping force characteristics of a typical CVD: comparison between lookup table and polynomial

162 Vol. 124, MARCH 2002

Transactions of the ASME

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

note various road classes, and N is a dimensionless exponent

Robson, 21, Sharp and Crolla, 22. Table 2 shows five ISO
road classes.
For a given vehicle speed, 12 can be converted to a frequency
domain characterization of a road by utilizing the relationship
between a wave number and a frequency such that f V,
where f is the excitation frequency in Hz and V is the speed of a
vehicle in m/sec, as follows:
S i f i V N1 f N ,

2,

if c

1.5,

if c

(12)

Note that once the vehicle speed V is fixed, the following relationship holds also see Table 2.
Fig. 5 Access time comparison: lookup table and polynomial

S i1 f 4S i f ,

for both N1.5

and

2.

(13)

On the other hand, for a constant i , the relationship between the

spectral density of speed aV and that of speed V becomes
S a f , i aV N1 f N a N1 i V N1 f N a N1 S i f .
(14)
Fig. 6 Current generation

Therefore, comparing 13 and 14 for a

4, the spectral density on the ISO ith road at speed aV is identical to the spectral
density on the ISO (i1)th road at speed V. If 1a N1 4, this
can be considered as the case that the vehicle runs on a road of
roughness between the ISO ith class and the ISO (i1)th class at
speed V. Similarly, if a N1 4, this corresponds to the case that
the vehicle runs on a road of roughness higher that the ISO (i
1)th class at speed V.
All these observations suggest that once the optimal gains for
all ISO roads for a fixed speed are determined, then the gain for an
arbitrary speed can be interpolated from the optimal gains for the
standard ISO roads. Therefore, it is not necessary to make a
2-dimensional lookup table, for the purpose of gain scheduling,
for the vehicle speed and the road classes. Instead, a
1-dimensional lookup table for the road roughness is sufficient. In
this work, V60 Km/h is used.
N1

Table 2 ISO road classes

creases as the number of cases increases. This is because each

case accesses 1000 data. In this work, the polynomial approach is
adopted.
As seen in Fig. 6, the transformation algorithm block determines the duty ratio of a PWM generator by taking two inputs, the
desired control force and the relative velocity of the rattle space.
In our case, the duty ratios that correspond to 1.6 Ampere and 0
Ampere are set to 0.4 and 0, respectively. The natural frequency
of the solenoid valve is about 300 Hz.
3.3 Road Profiles. In this section, the ISO road profiles for
the purpose of gain scheduling are described. The mean square
displacement spectral density of a track, along the length of a
single-sided road surface, defined by the ISO is
S i i / c N ,

2,

if c

1.5,

if c

(11)

where is the wave number in cycles/m, c 1/2 is the critical

wave number, i is the roughness coefficient, iA,B, . . . ,E de-

c sky ,c v o ptimal min

i

w t z r

3.4 Optimal Gains for the ISO Roads. The variations of

c sky and c v in 9 affect both the ride quality and the road holding
ability of vehicle. Figure 7 shows the conflicting behavior between two criteria for various input conditions simulation results, in which the ISO C-class road and a vehicle speed of 60
km/h are used. All plots are generated using the suspension model
developed in Section 2. The big circle marked at 0.86 m/sec2 in
the x-axis and 910 Newton in the y-axis indicates the characteristics of the passive damper in Fig. 4. Because the CVD in Fig. 4
has saturation, its performance is limited to the inside of the dotted rectangle. All curves extending outside the dotted rectangle are
obtained by assuming no saturation. As magnified in the upper
right corner, if c sky is increased while fixing c v , the body acceleration decreases but the dynamic tire force increases. Also, if c v
is increased while fixing c sky , the tire force decreases but the
body acceleration increases.
Now, the procedure for determining optimal gains for various
ISO road classes are discussed. Assume that N d -data of the dynamic tire force and body acceleration, i.e., TF(i), BA(i), i
1, . . . ,N d , have been collected for each ISO road input. Then,
the optimal coefficients are determined using the following
criterion.

TF i TF min 2
BA i BA min
w a z r
TF maxTF min
BA maxBA min

Journal of Dynamic Systems, Measurement, and Control

2 Nd

(15)
i1

MARCH 2002, Vol. 124 163

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

nance frequencies of the suspension system. The eigenvalues of

A are 1.49 j7.60 and 21.38 j66.73, which correspond to
the resonance frequencies of the sprung and unsprung masses at
f bn 1.23 Hz and f wn 11.15 Hz, respectively.
Now, let T and N s be the sampling time and the number of
data collected from 17, respectively. In this work, N s 50 and
T10 msec are used. Let the discrete Fourier transform of Z r
(z r (0),z r (1), . . . ,z r (N s 1)) be
N s 1

Z
rk

z r m e j 2 km/N s A rk jB rk ,

m0

k0,1, . . . ,N s 1
(18)

where
A rk
Fig. 7 Conflicting behavior between the ride comfort and the
handling performance of vehicle

B rk
where TF and BA denote the RMS values of the dynamic tire
force and the body acceleration; w a (z r ) and w t (z r ) are two
weighting factors, respectively. The selection of the weighting
factors is subjective, but the rougher the road is, the larger w t (z r )
is suggested. It is also noted that the gains should be selected in
such a way that the dynamic tire force is smaller than the static
tire force. Finally, the optimal gains determined for all ISO roads
are tabulated in Table 3.
3.5 Road Detection and Gain Tuning. In this section, using the data z s measured, the estimation of road roughness and the
determination of the controller gain for the estimated road are
described. The transfer function from z r (s) to z s (s) is derived
from 4 and 5 as follows:
G rs s ,

(16)
where A AB 1 H, C CD 1 H, (A,B 2 ,C,D 2 ) are from 4
5, and H satisfies f s c p lHx, where H,c p 0 0 l/
l/ ( 0,0, 0 ,0) . It is noted that H has been introduced to
avoid possible roots on the imaginary axis. In this work, H
0 0 0 614 has been used. Therefore, using the transfer function above, the road input can be estimated as follows:
40
1
G rs
s z s s
s40

m0

(17)

z r m cos

N s 1

m0

z r m sin

2
2
S f bn A r1
B r1

2 km
,
Ns

2 km
,
Ns

k1,2, . . . ,

k0,1, . . . ,

Ns
2

Ns
1.
2

and

2
2
S f wn A r5
B r5
.

(19)

While S( f bn ) in 19 can be directly used as a criterion for gain

selection, S( f wn ) should be converted to the equivalent energy
level at the sprung mass resonance frequency as follows:

S f bn

0.5s 4 17212.4s 3 317325s 2 11.5s212.6

s 4 45.7s 3 509765s 2 17212.4s317359

z r s

2
Ns

N s 1

Because N s 50 and T10, the frequency resolution becomes 2

Hz. Therefore, 2 Hz and 10 Hz are assumed to be the resonance
frequencies of the suspension system approximating 1.23 Hz and
11.15 Hz, respectively. For a better approximation, a smaller sampling frequency may be used. But, the selection of sampling time
and data size is related to the microprocessor used. Let S( f bn ) and
S( f wn ) denote the magnitudes at frequencies f bn and f wn ,
respectively, i.e.,

z s s
C sIA 1 B 2 D 2
z r s

2
Ns

where
N b ,N w

S i f bn
f bn N b
S f wn
S f wn
S i f wn
f wn N w
1.5,1.5 ,
2,1.5 ,
2,2 ,

(20)

for 2 f bn V

for 2 f bn V2 f wn .
for 2 f wn V

In deriving 20, 12 has been utilized. Finally, the bigger value

between two is selected as follows:
S f max S f bn ,S f bn .

(21)

Figure 8 shows estimation of three types of the ISO roads,

using the scheme in this section. It is seen that the convergence
requires at least 0.5 second to collect 50 data points.

where a low-pass filter of cutoff frequency 20 Hz has been added

to enhance the stability of the filter.
Now, the gain selection for the estimated road is described. A
real road contains various frequency components. But, the dominant components measured in the sprung mass will be the reso-

Table 3 Optimal gains for the ISO road classes

Fig. 8 Estimation of the ISO B, C, and D road classes simulation results: sampling time T 10 msec

164 Vol. 124, MARCH 2002

Transactions of the ASME

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Finally, let S( f ) satisfy S i ( f bn )S( f )S i1 ( f bn ), where

S i ( f bn ) and S i1 ( f bn ) are the magnitudes of ISO ith and (i
1)th roads, respectively. Then, the control gain g o for the current road estimated is interpolated logarithmically by using the
predetermined optimal gains for the ISO roads as follows:

S i1 f bn o
S f
o
log
g i log
g i1
S
f
S
f

i
bn
g o
log S i1 f bn log S i f bn

3.6 Filter Design. In this section, two filters for estimating

the absolute velocity of the sprung mass and the relative velocity
of the sprung and unsprung masses are designed. It is also remarked that even though the measurement of or is possible, an
estimation method is preferred considering the cost for extra
sensors.
1) Absolute Velocity of the Sprung Mass. The filter to estimate the absolute velocity from the acceleration data of the sprung
mass is suggested as follows:
s
s 2 n s 2n
2

z s

(23)

where 0.707 and n 0.1 Hz. Equation 23 performs as a differentiator below 0.1 Hz and as an integrator above 0.1 Hz. Consequently, this filter will provides a satisfactory absolute velocity
of the sprung mass by excluding a possible DC offset.
2) Relative Velocity of the Sprung and Unsprung Masses.
The relative velocity of two masses is estimated using the model
of suspension dynamics. Assume that 0 0 and is sufficiently
small. Then, the following approximations hold:

hor 0,

z u z s l C ,

1
z u z s
lC

and
lB
ll B z s z u .
lC

(25)
where m s , c p , and k s are the suspension parameters. It is noted
that a low pass filter 1/(1 s) has been added for eliminating
noises, where 1/2 l , 0.707, and l 1520 Hz.

(22)

o
where g io and g i1
are the optimal gains for ISO ith and (i
1)th roads, respectively.

z s F 1 z s

l B z s z u
lB
m ss
1
z s

z
lF 2 z s
lC
z s
l C c p sk s 1 s s

(24)

Therefore, the relative velocity from measurement data is filtered

as follows:

Hardware-in-the-Loop Simulations

The CVD unit has been used for the hardware-in-thesimulations HILS. The purpose of the HILS is to consider the
actuator dynamics neglected during the stages of modeling and
control system design. Therefore, the nonlinear characteristics of
the damping force together with the time-delay and the neglected
dynamics of the solenoid valve can be fully incorporated during
the HILS. Figure 9 shows the configuration for the HILS.
MATLAB/SIMULINK MathWorks Inc. is used as a computeraided control system tool, i.e., a programming environment, and a
dSPACE board digital signal processor board of dSPACE GmbH
Inc. is used as a rapid control prototyping tool for implementing
the plant dynamics and control laws.
The input current, generated by the PWM voltage signal from
TMS320P14, changes the damping force of the CVD. The relative
velocity of the sprung and unsprung masses is made by the MTS
test rig, which consists of a hydraulic actuator, a loadcell, a LVDT,
etc. The procedure for the HILS is summarized as follows:
1 The control algorithm is first of all designed off-line using
MATLAB/SIMULINK.
2 Using the Real-Time-Workshop, a C-code of the control algorithm from its block diagram form is generated. The C-code is
again downloaded to the target DSP board using the Real-TimeInterface. Therefore, the dynamic models developed can be reused
easily for the HILS.
3 The PWM voltage signal related to the damping force and
the relative displacement stroke calculated from the dynamic
model in the computer are transmitted to the damper drive unit
and the MTS control unit, respectively, through D/A converters.
4 The damper drive unit transforms the PWM voltage signal
into a current signal from 0 Ampere to 1.6 Ampere and supplies
the current signal to the CVD. The PWM voltage signal has 2 kHz
carrier frequency and the duty ratio of the PWM voltage signal

Fig. 9 Configuration for the hardware-in-the-loop simulations

Journal of Dynamic Systems, Measurement, and Control

MARCH 2002, Vol. 124 165

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

transferred to the dynamic model in the computer. By varying the

control parameters in real time, the procedures 35 can be
repeated.

Fig. 10 Vertical acceleration comparison between a passive

damper and a CVD with the modified skyhook control experimental results: 1 Hz speed bump

The sampling time is set to 0.01 s. This time step should be

larger than the combined time of the calculation time of the plant
dynamics and the input-output communication time between the
external devices. In experiments when the actual calculation time
was larger than the step size, small oscillations had occurred in a
valve of the hydraulic actuator. If these oscillations are delivered
to the CVD, it will be difficult to measure the exact damping force
from the loadcell. Hence, it may affect the stability of the complete control system.
Finally, using the speed bumper of 1 Hz, the vertical accelerations of the sprung mass for a passive damper and a semi-active
damper are compared in Fig. 10. The modified skyhook controller
with c sky 2000 Ns/m, c v 1960 Ns/m shows a smaller overshoot and a less settling time compared to the passive damper.
Figure 11 shows the current input applied to the CVD. It is shown
that 0 Ampere current input was applied to the periods of high
vertical accelerations. The force tracking performance of the semiactive damper is depicted in Fig. 12.

Fig. 11 Current input applied to the CVD in Fig. 10: 1 Hz speed

bump

Conclusions

In this paper, a new control-oriented model and a road-adaptive

control scheme for the semi-active Macpherson suspension system were investigated. The new model incorporates the rotational
motion of the unsprung mass, giving a better description for the
plant dynamics and keeping the degree-of-freedom of the plant
model by two. Upon the requirement of using only one acceleration sensor, a modified skyhook control with gain scheduling was
developed. The road estimation and the control variables estimation are based upon the linearized model developed. Considering
the complexity of the hydraulic semi-active damper, the actuator
dynamics has been incorporated during the hardware-in-the-loop
simulations. It was demonstrated that the semi-active system
could achieve a competitive control performance by adopting the
road adaptive control laws. Because the control law design, the
road estimation method, the gain scheduling strategy, and the
hardware-in-the-loop simulation method developed in this paper
are not restricted to a particular suspension system, the entire
strategy can be extended to any semi-active system including the
ER damper and the MR damper.

Acknowledgments
This work was supported in part by the Korea Research Foundation under Grant KRF-2000-EA0096. The second author was
supported in part by the Brain Korea 21 Program of the Ministry
of Education and Human Resources, Korea.

References

Fig. 12 Tracking performance of the desired damping force: 1

Hz speed bump

corresponding to 1.6 Ampere is 0.4. The MTS control unit excites

the hydraulic actuator according to the relative displacement
command.
5 The damping force of the CVD and the relative displacement
are measured using a loadcell and a LVDT, respectively, and
166 Vol. 124, MARCH 2002

1 Karnopp, D. C., Crosby, M. J., and Harwood, R. A., 1974, Vibration Control
Using Semi-Active Force Generators, ASME J. Eng. Ind., 96, No. 2, pp.
619 626.
2 Besinger, F. H., Cebon, D., and Cole, D. J., 1995, Force Control of a SemiActive Damper, Veh. Syst. Dyn., 24, pp. 695723.
3 Novak, M., and Valasek, M., 1996, A New Concept of Semi-Active Control
of Trucks Suspension, Proc. of AVEC 96, International Symposium on Advanced Vehicle Control, Aachen University of Technology, pp. 141151.
4 Jonsson, M., 1991, Simulation of Dynamical Behavior of a Front Wheel
Suspension, Veh. Syst. Dyn., 20, pp. 269281.
5 Stensson, A., Asplund, C., and Karlsson, L., 1994, The Nonlinear Behavior
of a Macpherson Strut Wheel Suspension, Veh. Syst. Dyn., 23, pp. 85106.
6 Yue, C., Butsuen, T., and Hedrick, J. K., 1989, Alternative Control Laws for
Automotive Active Suspension, ASME J. Dyn. Syst., Meas., Control, 111, pp.
286 291.
7 Hanselmann, H., 1996, Hardware-in-the-loop Simulation Testing and its Integration into a CACSD Toolset, The IEEE International Symposium on
Computer Aided Control System Design, Dearborn, MI, pp. 152156.
8 Hanselmann, H., 1996, Automotive Control: From Concept to Experiment to
Product, The IEEE International Symposium on Computer Aided Control
System Design, Dearborn, MI, pp. 129134.
9 Hong, K. S., Jeon, D. S., Yoo, W. S., Sunwoo, H., Shin, S. Y., Kim, C. M., and

Transactions of the ASME

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

10
11
12
13
14
15
16

Park, B. S., 1999, A New Model and an Optimal Pole-placement Control of

the Macpherson Suspension System, SAE International Congress and Exposition, Detroit, MI, SAE paper No. 1999011331, pp. 267276.
Alleyne, A., Neuhaus, P. D., and Hedrick, J. K., 1993, Application of Nonlinear Control Theory to Electronically Controlled Suspensions, Veh. Syst.
Dyn., 22, pp. 309320.
Yi, K., and Hedrick, J. K., 1993, Dynamic Tire Force Control by Semi-Active
Suspension, ASME J. Dyn. Syst., Meas., Control, 115, No. 3, pp. 465 474.
Alleyne, A., and Hedrick, J. K., 1995, Nonlinear Adaptive Control of Active
Suspensions, IEEE Trans. Control Syst. Technol., 3, No. 1, pp. 94 101.
Hac, A, 1995, Decentralized Control of Active Vehicle Suspension with Preview, ASME J. Dyn. Syst., Meas., Control, 117, No. 4, pp. 478 483.
Kim, H., and Yoon, Y. S., 1995, Semi-Active Suspension with Preview Using
a Frequency-Shaped Performance Index, Veh. Syst. Dyn., 24, pp. 759780.
El-Demerdash, S. M., and Crolla, D. A., 1996, Hydro-pneumatic Slow-active
Suspension with Preview Control, Veh. Syst. Dyn., 25, pp. 369386.
Lin, J. S., and Kanellakopoulos, I., 1997, Nonlinear Design of Active Sus-

Journal of Dynamic Systems, Measurement, and Control

pensions, IEEE Control Syst. Mag., 17, No. 3, pp. 4559.

17 Cherry, A. S., Jones, R. P., and Potter, T. E. C., 1999, The Use of Multibody
System Modeling and Multivariable System Decoupling Technique in Vehicle
Ride Control, ASME J. Dyn. Syst., Meas., Control, 121, No. 3, pp. 479 486.
18 Choi, S. B., Choi, Y. T., and Park, D. W., 2000, A Sliding Mode Control of a
Full-Car Electrorheological Suspension System via Hardware-in-the-Loop
Simulation, ASME J. Dyn. Syst., Meas., Control, 122, No. 2, pp. 114 121.
19 Hedrick, J. K., Rajamani, R., and Yi, K., 1994, Observer Design for Electronic Suspension Applications, Veh. Syst. Dyn., 23, pp. 413 440.
20 Yi, K., and Hedrick, J. K., 1995, Observer-Based Identification of Nonlinear
System Parameters, ASME J. Dyn. Syst., Meas., Control, 117, No. 2, pp.
175182.
21 Robson, J. D., 1979, Road Surface Description and Vehicle Response, Int. J.
Veh. Des., 1, No. 1, pp. 2535.
22 Sharp, R. S., and Crolla, D. A., 1987, Road Vehicle Suspension System
Designa Review, Veh. Syst. Dyn., 16, pp. 167192.

MARCH 2002, Vol. 124 167

Downloaded From: http://asmedigitalcollection.asme.org/ on 12/16/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use