Aspects of Achievable Performance For Quarter-Car

JOURNAL OF
SOUND AND
VIBRATION
Journal of Sound and Vibration 311 (2008) 440460
Aspects of achievable performance for quarter-car
active suspensions
$
Semiha Tu rkay, Hu seyin Akc- ay
Department of Electrical and Electronics Engineering, Anadolu University, 26470 Eskis -ehir, Turkey
Received 13 November 2006; received in revised form 15 September 2007; accepted 18 September 2007
Available online 25 October 2007
Abstract
In this paper, constraints on the transfer functions from the road disturbance to the vertical acceleration, the suspension
travel, and the tire deection are derived for a quarter-car active suspension system using the vertical acceleration and/or
the suspension travel measurements for feedback. The derived constraints complement the similar constraints in the
literature. By using the factorization approach to feedback stability, it is shown that tire damping couples the motions of
the sprung and unsprung masses; and eliminates a constraint at the wheel-hop frequency. The inuence of tire damping on
the design of an active suspension system for a quarter-car model by a mixture of the LQG methodology and the
interpolation approach is also illustrated.
r 2007 Elsevier Ltd. All rights reserved.
1. Introduction
The suspension system is the main tool to achieve ride comfort and drive safety for a vehicle. Passive
suspension systems have been designed to obtain a good compromise between these objectives, but intrinsic
limitations prevent them from obtaining the best performances for both goals. Compared with passive
suspension systems, active and semi-active suspension systems can achieve a better compromise during various
driving conditions.
Active and semi-active control of vehicle suspensions have been the subject of considerable investigation
since the late 1960s; see, for example Refs. [112] and the references therein. Constraints and trade-offs on
achievable performances have been studied in Refs. [1317]. As put forward in Ref. [16], in a study of
constraints and trade-offs from a control systems point of view, one has to properly address: (i) What can and
cannot be achieved with general dynamic compensation, and (ii) How much freedom is gained by the selection
of measurements for feedback purpose?
In Refs. [13,14], constraints on achievable frequency responses were derived from an invariant point
perspective. A framework using mechanical multi-port networks to study the performance capabilities and
constraints is developed in Ref. [17]. In Ref. [16], for a quarter-car model of an automative suspension a
ARTICLE IN PRESS
www.elsevier.com/locate/jsvi
0022-460X/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jsv.2007.09.014
$
This work was supported in part by the Scientic & Technological Research Council of Turkey under Grant 106E108.
Corresponding author. Tel. +90 222 335 0580; fax: +90 222 323 9501.
E-mail address: [email protected] (H. Akc- ay).
complete set of constraints on several transfer functions of interest from the road and the load disturbances
were determined by making use of the factorization approach to feedback stability and the Youla
parameterization of stabilizing controllers. Roughly speaking, completeness means that from a given set of
constraints, one can identify a quarter-car model within the model class matching the given constraints. Such
an approach reveals the degrees of freedom in shaping the response of the vehicle to disturbances and
determines a minimum set of measurements to exploit this freedom.
In Ref. [16], constraints on the frequency responses of the sprung mass position, the suspension travel, and
the tire deection were derived for various choices of measurements ranging from the suspension travel to a
full set of state variables. These constraints typically arise in the form of nite and nonzero invariant frequency
points and the growth restrictions on the frequency responses and their derivatives at zero and innite
frequencies. The quarter-car model studied in Ref. [16] does not include passive suspension elements spring
and damper; and also tire damping is neglected.
In most works, tire damping is set to zero when modeling automotive active suspension systems. This is
partly due to the fact that tire damping is difcult to estimate. It is generally accepted that damping ratio in a
vehicle tire ranges between 0:03 and 0:10 depending on the size, applied pressure, free or rotating, new or
worn, and the tire type, i.e., all season or snow [1820]. The tire damping by itself has little inuence on the
wheel-hop vibration since this mode is mainly damped by the shock absorber.
The ignorance of damping in tire models compelled misleading conclusions that at the wheel-hop
frequency, no matter what forces are exerted between sprung and unsprung masses, their motion
are uncoupled, and the vertical acceleration of the sprung mass will be unaffected [13,14,16]. It is
pointed out in Ref. [21] that by taking tire damping to be small but nonzero, the motions of the sprung
and unsprung masses are coupled at all frequencies, and control forces can be used to reduce the
sprung mass vertical acceleration at the wheel-hop frequency. The effect of introducing tire damping can be
quite large.
The paper is structured as follows. First, the results in Ref. [16] are complemented assuming that (i) the
sprung mass acceleration measurement instead of the sprung mass position measurement is used for the
parametrization of the stabilizing controllers, (ii) the closed-loop sprung mass acceleration is targeted instead
of the closed-loop sprung mass position for the evaluation of ride comfort, (iii) the passive suspension
elements are included in the vehicle model. The differences and the similarities between the derived results and
Ref. [16] are emphasized. For example, it is demonstrated that employment of the sprung mass acceleration as
a measurement and performance objective in a vehicle model that includes passive suspension elements affects
the parameterizability of the stabilizing controllers.
The reader is reminded that a semi-active suspension consists of in series a spring and damper whose
coefcient is changed in a nonlinear fashion. In a semi-active suspension, ride comfort is taken care of by a
nonlinear damper while safety requirements are met by a xed spring.
Next, the effect of tire damping on the achievable performance is investigated. The results predicate the
conclusions in Refs. [2123] that tire damping couples the motions of the sprung and unsprung masses, and
control forces can be used to reduce the sprung mass vertical acceleration at the wheel-hop frequency without
sacricing road holding.
2. The quarter-car model
A two-degree-of-freedom quarter-car model is shown in Fig. 1. In this model, the sprung and unsprung
masses are denoted, respectively, by m
s
and m
u
. The suspension system is represented by a linear spring of
stiffness k
s
and a linear damper with a damping rate c
s
. The tire is modeled by a linear spring of stiffness k
t
and
a linear damper with a damping rate c
t
. The parameter values, except c
t
, chosen for this study are shown in
Table 1 [24]. They are typical for a lightly damped passenger car. The parameters m
s
; m
u
, and k
t
are xed
throughout the paper while the parameters k
s
; c
s
, c
t
are freely changed.
Assuming that the tire behaves as a point-contact follower that is in contact with the road at all times, the
equations of motion take the form:
m
s
x
1
= k
s
(x
1
x
2
) c
s
( _ x
1
_ x
2
) u,
ARTICLE IN PRESS
S. Tu rkay, H. Akc-ay / Journal of Sound and Vibration 311 (2008) 440460 441
m
u
x
2
= k
s
(x
1
x
2
) c
s
( _ x
1
_ x
2
) u k
t
(x
2
w) c
t
( _ x
2
_ w), (1)
where x
1
and x
2
are, respectively, the displacements of the sprung and unsprung masses, and w is the road
unevenness. The variables x
1
; x
2
, and w are measured with respect to an inertial frame, and the control input u
is a force.
The objective of this paper is to study the performance limits of an actively controlled vehicle imposed by
the road surface unevenness. The vehicle response variables that need to be examined are the vertical
acceleration of the sprung mass as an indicator of the vibration isolation, the suspension travel as a measure of
the rattling space, and the tire deection as an indicator of the road-holding characteristic of the vehicle. These
variables, denoted, respectively by z
1
; z
2
, and z
3
, can be written in terms of the state variables x
1
; x
2
, their
derivatives, and the exogenous input w as follows:
z
1
= x
1
, (2)
z
2
= x
1
x
2
, (3)
z
3
= x
2
w. (4)
Passenger comfort requires z
1
to be as small as possible while compactness of rattle space, good handling
characteristics, and improved road-holding quality require z
2
and z
3
be kept as small as possible.
It is a well-known fact [15] that these objectives cannot be met simultaneously with a passive suspension
system. In a passive suspension system, the only parameter that can be altered in an optimization study is c
s
since k
s
is a priori xed to obtain stiffness against rolling. The conicting three goals can be attained up to a
certain level by replacing passive suspension system with an active or semi-active suspension system
[24,7,8,16,24].
ARTICLE IN PRESS
Fig. 1. The quarter-car model of the vehicle.
Table 1
The vehicle system parameters for the quarter-car model
Sprung mass. m
s
240 kg
Unsprung mass m
u
36 kg
Damping coefcient c
s 980 Ns m
1
Secondary suspension stiffness k
s 16; 000 Nm
1
Primary suspension stiffness k
t 160; 000 Nm
1
S. Tu rkay, H. Akc- ay / Journal of Sound and Vibration 311 (2008) 440460 442
3. Factorization approach to feedback stability
In this section, the factorization approach developed in Ref. [16] for the feedback stability of the quarter-car
model is briey reviewed. The reader is referred to Vidyasagar and Zhou et al. [25,26] for a comprehensive
treatment.
Let Z(s), U(s), and W(s) denote, respectively, the Laplace transforms of the signals z(t) = [z
1
(t) z
2
(t) z
3
(t)]
T
,
u(t), and w(t), where for a given vector b, b
T
denotes the transpose of b. From Eqs. (1)(4),
Z(s) = G
11
(s)W(s) G
12
(s)U(s), (5)
where
G
11
(s) =
1
D(s)
s
2
(c
s
s k
s
)(c
t
s k
t
)
m
s
s
2
(c
t
s k
t
)
s
2
[m
s
m
u
s
2
(m
s
m
u
)c
s
s (m
s
m
u
)k
s
]
2
6
4
3
7
5, (6)
G
12
(s) =
1
D(s)
s
2
(m
u
s
2
c
t
s k
t
)
[(m
s
m
u
)s
2
c
t
s k
t
]
m
s
s
2
2
6
4
3
7
5 (7)
and
D(s) = m
s
m
u
s
4
[(m
s
m
u
)c
s
m
s
c
t
]s
3
[(m
s
m
u
)k
s
m
s
k
t
c
s
c
t
]s
2
(c
s
k
t
c
t
k
s
)s k
s
k
t
. (8)
A polynomial D(s) is said to be Hurwitz if all its zeros lie in the open left-half plane. Note that D(s) is Hurwitz
if k
s
; k
t
40, and c
s
40 or c
t
40.
For the design of a feedback law, consider the measurements:
y
1
= x
1
,
y
2
= x
1
x
2
. (9)
In the study of the constraints, the cases y = y
2
and y = [y
1
y
2
]
T
will be considered. When y = [y
1
y
2
]
T
, from
Eqs. (1)(4),
Y(s) = G
21
(s)W(s) G
22
(s)U(s), (10)
where
G
21
(s) =
1
D(s)
s
2
(c
s
s k
s
)(c
t
s k
t
)
m
s
s
2
(c
t
s k
t
)
" #
, (11)
G
22
(s) =
1
D(s)
s
2
(m
u
s
2
c
t
s k
t
)
(m
s
m
u
)s
2
c
t
s k
t
" #
. (12)
The other case is obtained by simply selecting the second rows of G
21
and G
22
. Hence, the generalized plant
dened by
(13)
maps the pair of inputs [wu]
T
to the pair of outputs [z
T
y
T
]
T
.
Now, let K(s) denote the transfer function of the controller with input y and the output u. The feedback
conguration is shown in Fig. 2. The stabilization problem is to nd a proper feedback transfer function K
such that the closed-loop system in Fig. 2 is internally stable. Assuming that G and G
22
share the same unstable
poles, it is a well-known fact (see, for example, Lemma 12.2 in Ref. [26]) that K internally stabilizes G if and
ARTICLE IN PRESS
only if K internally stabilizes G
22
. Recall that the unstable poles of G are the roots of D(s) in the closed right-
half plane. Assuming that G
22
is internally stabilizable, the set of all compensators which stabilize G can be
parametrized in terms of a coprime factorization of G
22
. This parametrization is called the Youla
parametrization.
The Youla parametrization is obtained as follows. Let RH
pq
o
denote the set of stable p by q
real-rational transfer matrices. (In what follows, superscripts p and q will be dropped and they will be
inferred from the underlying context). Given G
22
, nd matrices N; M;
e
N;
e
M; X; Y;
e
X, and
e
Y in RH
o
such that
G
22
= NM
1
=
e
M
1
e
N, (14)
e
X
e
Y
e
N
e
M
" #
M Y
N X
!
= I, (15)
where I denotes the identity matrix of compatible dimension. The factorization Eq. (14) of G
22
satisfying
Eq. (15) is called double coprime factorization over RH
o
. Then, the Youla parametrization of all stabilizing
controllers takes the form:
K = (Y MQ)(X NQ)
1
; Q c RH
o
det(I X
1
NQ)(o)a0. (16)
With this parametrization, the transfer matrix from w to z denoted by T
zw
(s) takes a particularly convenient
form which is afne in Q:
T
zw
= G
11
G
12
(Y MQ)
e
MG
21
. (17)
As Q varies over RH
o
, Eq. (17) parametrizes all achievable transfer matrices.
4. Achievable performance for quarter car model
In the design of an active suspension system, it is desirable to keep the road response amplitudes
[T
z
k
w
(jo)[; k = 1; 2; 3 as small as possible, at least in the frequency range of interest. The aim of this section is
to investigate the limitations on this goal for the two measurement setups and several assumptions on k
s
; c
s
,
and c
t
. By using the factorization approach, a complete set of constraints on the transfer functions
T
z
k
w
(s); k = 1; 2; 3 will be derived.
The rst case to be treated in the sequel is the case y = y
2
with the assumption that k
s
; c
s
, and c
t
are all
positive. Then, as noted previously, D(s) is a Hurwitz polynomial and a pair of coprime factors for
G
22
= D
1
[(m
s
m
u
)s
2
c
t
s k
t
] (18)
is easily found as
N =
e
N = G
22
; M =
e
M = 1. (19)
Furthermore, the choice given by
X =
e
X = G
22
1; Y =
e
Y = 1 (20)
ARTICLE IN PRESS
Fig. 2. Standart block diagram.
enforces Eq. (15) as can directly be veried. Now, put
b
Q = 1 Q where Q c RH
o
. Then, T
zw
and K in
Eqs. (17) and (16)) take the following forms:
T
z
1
w
= s
2
D
1
(c
t
s k
t
){c
s
s k
s

b
Qm
s
s
2
D
1
(m
u
s
2
c
t
s k
t
)], (21)
T
z
2
w
= m
s
s
2
D
1
(c
t
s k
t
){1
b
QD
1
[(m
s
m
u
)s
2
c
t
s k
t
]], (22)
T
z
3
w
= s
2
D
1
[m
s
m
u
s
2
(m
s
m
u
)c
s
s (m
s
m
u
)k
s
]
b
Qm
2
s
s
4
D
2
(c
t
s k
t
), (23)
K =
b
Q(1 G
22
b
Q)
1
. (24)
The rst constraint is obtained by observing that the coefcient of
b
Q in the expression for T
z
1
w
is O(s
4
) for
all sufciently small complex numbers s and every transfer matrix in RH
o
has elements uniformly bounded
on the closed right half-plane. Here, the notation a(s) = O(s
g
) means that there exist two numbers a; b40
such that
a[s[
g
p[a(s)[pb[s[
g
.
Therefore, a Taylor series expansion of the term s
2
D
1
(c
s
s k
s
)(c
t
s k
t
) in a neighborhood of zero which is
accurate up to the term O(s
4
) will be sufcient to determine the behavior of T
z
1
w
there. By long division,
s
2
D
1
(c
s
s k
s
)(c
t
s k
t
) = s
2
O(s
4
). (25)
Hence, for all small s,
T
z
1
w
= s
2
O(s
4
), (26)
which implies,
T
z
1
w
(0) = T
(1)
z
1
w
(0) = T
(3)
z
1
w
(0) = 0; T
(2)
z
1
w
(0) = 2. (27)
For all large s, observe that the coefcient of
b
Q in the expression for T
z
1
w
is O(s
1
). Thus, a Taylor series
expansion of the term s
2
D
1
(c
s
s k
s
)(c
t
s k
t
) around innity which is accurate up to a term O(s
1
) is
obtained, again by long division, as
s
2
D
1
(c
s
s k
s
)(c
t
s k
t
) =
c
s
c
t
m
s
m
u
O(s
1
). (28)
Hence,
T
z
1
w
=
c
s
c
t
m
s
m
u
O(s
1
). (29)
It remains to show that Eqs. (26) and (29) form a complete set of constraints, i.e., no further constraints on
T
z
1
w
, which are valid for all Q c RH
o
, can be derived. This amounts to showing that given an arbitrary
H
1
c RH
o
subject to constraints Eqs. (26) and (29), there exists a stabilizing controller such that for some
b
Q c RH
o
, Eq. (21) holds for H
1
. In this case, H
1
is said admissible. To this end, from Eq. (21)
b
Q =
[T
z
1
w
s
2
D
1
(c
s
s k
s
)(c
t
s k
t
)]
m
s
s
4
D
2
(m
u
s
2
c
t
s k
t
)(c
t
s k
t
)
.
From Eqs. (28) and (29), the numerator and the denumerator of
b
Q are O(s
1
). Hence,
b
Q is a proper rational
function with a singularity at the origin of multiplicity four. However, the singularity at the origin is
removable from Eqs. (25) and (26). Thus,
b
Q c RH
o
as desired.
It should be noted that as soon as an admissible H
1
is specied, two other admissible functions, be H
2
and H
3
, corresponding to T
z
2
w
and T
z
3
w
are generated via Eqs. (22) and (23). Indeed, elimination of
b
Q in
Eqs. (21)(23) results in the following trade-off relations:
H
2
=
c
t
s k
t
m
u
s
2
c
t
s k
t
(m
s
m
u
)s
2
c
t
s k
t
s
2
(m
u
s
2
c
t
s k
t
)
H
1
, (30)
ARTICLE IN PRESS
H
3
=
m
u
s
2
m
u
s
2
c
t
s k
t
m
s
m
u
s
2
c
t
s k
t
H
1
, (31)
H
3
=
(m
s
m
u
)s
2
(m
s
m
u
)s
2
c
t
s k
t
m
s
s
2
(m
s
m
u
)s
2
c
t
s k
t
H
2
. (32)
The constraints on H
1
and its derivatives at s = 0 are partially recovered by Eqs. (30) and (31). In fact, given
an admissible H
2
, analyticity of H
2
at s = 0 forces the following function:
H
1
(s) s
2
D
1
(c
s
s k
s
)(c
t
s k
t
)
to have at least two zeros there, which implies H
1
(0) = H
/
1
(0) = 0. No results on further derivatives of H
1
at
s = 0 can be deduced from Eqs. (30) and (31).
The above results are captured in the following.
Proposition 4.1. Consider the quarter car model in Eq. (1) with k
s
; c
s
; c
t
40. Assume that y = y
2
and let H
1
be
any function in RH
o
. Then, H
1
= T
z
1
w
for some stabilizing control law if and only if:
(1) H
1
(s) = c
s
c
t
=m
s
m
u
O(s
1
),
(2) H
1
(0) = H
(1)
1
(0) = H
(3)
1
(0) = 0; H
(2)
1
(0) = 2.
A similar derivation to the above can be carried out for H
2
and H
3
or Proposition 4.1 combined with Eqs. (30)
and (31) yields the following results.
s
; c
s
; c
t
2
and let H
2
be
any function in RH
o
. Then, H
2
= T
z
2
w
(1) H
2
(s) = (c
t
=m
u
)s
1
[(m
s
m
u
=m
u
)(c
s
c
t
=m
s
m
u
) (c
2
t
=m
2
u
) (k
t
=m
u
)]s
2
O(s
3
),
(2) H
2
(0) = H
(1)
2
(0) = 0.
s
; c
s
; c
t
2
and let H
3
be
any function in RH
o
. Then, H
3
= T
z
3
w
(1) H
3
(s) = 1 (c
t
=m
u
)s
1
[(k
t
=m
u
) (c
s
c
t
)c
t
=m
2
u
]s
2
O(s
3
),
(2) H
3
(0) = H
(1)
3
(0) = 0, H
(2)
3
(0) = 2(m
s
m
u
)=k
t
, H
(3)
3
(0) = 6(m
s
m
u
)c
t
=k
2
t
.
Now, assume that c
t
= 0 and k
s
; c
s
40. Then, D(s) is still a Hurwitz polynomial, and it sufces to let c
t
= 0 in
Eqs. (21)(23) and Eqs. (30)(32). Two new constraints arise at the frequencies:
o
1
=
k
t
m
s
m
u
s
; o
2
=
k
t
m
u
s
(33)
which have already been observed in Refs. [14,16]. The results for this case are summarized in the following.
s
; c
s
40, and c
t
= 0. Assume that y = y
2
and let
H
1
be any function in RH
o
. Then, H
1
= T
z
1
w
(1) H
1
(s) = (k
t
c
s
=m
s
m
u
)s
1
O(s
2
),
(2) H
1
(0) = H
(1)
1
(0) = H
(3)
1
(0) = 0; H
(2)
1
(0) = 2,
(3) H
1
(jo
2
) = (jo
2
)
2
(m
u
=m
s
).
ARTICLE IN PRESS
s
; c
s
40, and c
t
2
and let
H
2
o
. Then, H
2
= T
z
2
w
(1) H
2
(s) = (k
t
=m
u
)s
2
((m
s
m
u
)c
s
k
t
=m
s
m
2
u
)s
3
O(s
4
),
(2) H
2
(0) = H
(1)
2
(0) = 0,
(3) H
2
(jo
1
) = m
s
m
u
=m
s
.
s
; c
s
40, and c
t
2
and let
H
3
o
. Then, H
3
= T
z
3
w
(1) H
3
(s) = 1 (k
t
=m
u
)s
2
(k
t
c
s
=m
2
u
)s
3
O(s
4
),
(2) H
3
(0) = H
(1)
3
(0) = H
(3)
3
(0) = 0, H
(2)
3
(0) = 2(m
s
m
u
)=k
t
.
Propositions 4.44.6 yield Theorems 13 in Ref. [16] when c
s
= 0. In Ref. [16], T
x
1
w
is constrained instead of
T
z
1
w
. The latter is related to the former by the equation T
z
1
w
(s) = s
2
T
x
1
w
(s). Then, the third formula in
Proposition 4.4 recovers the constraint T
x
1
w
(jo
2
) = m
u
=m
s
derived in Ref. [16].
The appearance of c
s
in the constraints of Propositions 4.44.6 demonstrates that the damper in Fig. 1
cannot be incorporated to u. Otherwise, for a given proper controller K
]
that stabilizes the quarter-car model
in Fig. 1 with c
s
= k
s
= 0, a controller K satisfying Eq. (24) and K
]
= K sc
s
k
s
would be improper since
both controllers have the same input y
2
. The same argument also explains absence of k
s
in the constraints of
Propositions 4.14.6. The reader is cautioned that the rst conclusion drawn above is valid for the quarter-car
model with a suspension consisting of an actuator in parallel with a spring and a damper as shown in Fig. 1.
There are many possibilities to connect passive elements with an actuator, in which the issue of properness
never arises. In hardware implementation of active or semi-active suspensions, parallel connection (without
damper) is a preferred conguration.
Although k
s
does not appear in the constraint formulae above, a given set of measurements may not be
sufcient to parametrize all stabilizing proper controllers if the spring in Fig. 1 is missing. Recall the internal
stabilizability condition: G and G
22
share the same unstable poles. If c
s
40 or c
t
40 and k
s
40, then this
requirement is satised by all elements of G
22
in Eq. (12) and one can also use y
1
for the parametrization of
the stabilizing controllers. However, in general, different measurements lead to different constraint sets.
If c
s
= c
t
= k
s
= 0, then D and G
22
in Eqs. (8) and (12) equal m
s
s
2
(m
u
s
2
k
t
) and
1
m
s
(m
s
m
u
)s
2
k
t
m
s
s
2
(m
u
s
2
k
t
)
2
6
6
6
4
3
7
7
7
5
.
Clearly, y
1
cannot give rise to a parametrization of stabilizing controllers. On the other hand, If k
s
40, then
D(s) = m
s
m
u
s
4
[(m
s
m
u
)k
s
m
s
k
t
]s
2
k
s
k
t
, and since D(jo
1
)a0 and D(jo
2
)a0, no pole-zero cancellation
can happen between D and any component of G
22
. Hence, as a measurement, x
1
or x
1
x
2
is sufcient for the
parametrization of the stabilizing controllers.
The next case to be studied is y = [y
1
y
2
]
T
. Let
b
Q = [
b
Q
1
b
Q
2
] = (Y MQ)
e
M. Since Q c RH
o
is a
two-dimensional row vector, T
zw
in Eq. (17) can be written as
T
zw
= G
11
G
12
G
T
21
b
Q
T
. (34)
Recall that a non-singular matrix is unimodular if its determinant is constant. Now, dene a product of
unimodular matrices by
P =
1 0
m
1
s
c
s
s 1
" #
1 0
m
1
s
k
s
1
" #
(35)
ARTICLE IN PRESS
which is a unimodular matrix with the inverse:
P
1
=
1 0
m
1
s
(c
s
s k
s
) 1
" #
.
The chosen matrix P has the property:
G
T
21
P = m
s
s
2
D
1
(c
t
s k
t
) [0 1]. (36)
It should be clear how to proceed in order to parametrize the stabilizing controllers if more than two
measurements are available for feedback. For example, if y equals [y
1
y
2
x
2
x
2
]
T
, a unimodular matrix P is
constructed such that when premultiplied with G
T
21
, the result is a row vector with the rst three elements being
zero, similar to Eq. (36). Therefore, the two factors of P in Eq. (35) are exactly elementary column operators.
The utility of Eq. (36) is to allow a parametrization of T
zw
in terms of a scalar transfer function.
Since D is Hurwitz, coprime factors of G
22
can be chosen as follows:
N =
e
N = G
22
; M = 1;
e
M = I.
Hence,
T
zw
= G
11
e q G
12
s
2
D
1
(c
t
s k
t
)(c
s
s k
s
),
where
e q =
b
Q
1

b
Q
2
m
s
(c
s
s k
s
)
1
. (37)
It follows that
T
z
1
w
= s
2
D
1
(c
t
s k
t
)(c
s
s k
s
) 1 e qs
2
D
1
(m
u
s
2
c
t
s k
t
)

, (38)
T
z
2
w
= s
2
D
1
(c
t
s k
t
) m
s
e qD
1
[(m
s
m
u
)s
2
c
t
s k
t
](c
s
s k
s
)

, (39)
T
z
3
w
= s
2
D
1
[m
s
m
u
s
2
(m
s
m
u
)c
s
s (m
s
m
u
)k
s
]
e q m
s
s
4
D
2
(c
s
s k
s
)(c
t
s k
t
), (40)
K =
b
Q(I G
22
b
Q)
1
. (41)
The range space of e q equals RH
o
. This means that e q and
b
Q
2
can be used to parametrize the set of the
stabilizing controller. Then,
b
Q
1
is solved from Eq. (37) and plugged in Eq. (41).
The trade-off relations among H
1
; H
2
, and H
3
are the same as Eqs. (30)(32). The constraints on the transfer
functions H
1
; H
2
, and H
3
are expressed in the following results.
s
; c
s
; c
t
40. Assume that y = [y
1
y
2
]
T
and let H
1
o
. Then, H
1
= T
z
1
w
H
1
(0) = H
(1)
1
(0) = H
(3)
1
(0) = 0; H
(2)
1
(0) = 2.
s
; c
s
; c
t
2
and let H
2
be
any function in RH
o
. Then, H
2
= T
z
2
w
(1) H
2
(s) = (c
t
=m
u
)s
1
O(s
2
),
(2) H
2
(0) = H
(1)
2
(0) = 0.
s
; c
s
; c
t
40. Assume that y = [y
1
y
2
]
T
and let H
3
o
. Then, H
3
= T
z
3
w
(1) H
3
(s) = 1 (c
t
=m
u
)s
1
O(s
2
),
(2) H
3
(0) = H
(1)
3
(0) = 0, H
(2)
3
(0) = 2(m
s
m
u
)=k
t
, H
(3)
3
(0) = 6(m
s
m
u
)c
t
=k
2
t
.
ARTICLE IN PRESS
Based on Propositions 4.14.3 and Propositions 4.74.9, the following conclusions can be drawn. First,
extra measurement, i.e., y
1
affects only the Taylor series coefcients of H
1
; H
2
, and H
3
at innity. Second, c
s
does not show up in the formulae of Propositions 4.74.9 in contrast to those of Propositions 4.14.3. Finally,
the trade-off relations are the same for both cases. This implies that a point in the H
2
versus H
1
trade-off curve
determined uniquely by the controller in Eq. (24) corresponds to an innite number of controllers in Eq. (41).
This information can be useful in constraining the controller dynamics. For example, constraining the
1
norm
of K, which is dened as the absolute integral of the impulse response of K, constrains the magnitude of the
input to persistent measurements.
The last case to be studied in this paper is the case k
s
; c
s
40, c
t
= 0, and y = [y
1
y
2
]
T
. As noted before, D is
Hurwitz, and it sufces to let c
t
= 0 in Eqs. (38)(40). The trade-off relations are the same as Eqs. (30)(32)
with c
t
= 0 substituted. The constraints on H
1
, H
2
, and H
3
are captured in the following results.
s
; c
s
40, c
t
= 0. Assume that y = [y
1
y
2
]
T
and
let H
1
o
. Then, H
1
= T
z
1
w
(1) H
1
(s) = O(s
1
),
(2) H
1
(0) = H
(1)
1
(0) = H
(3)
1
(0) = 0; H
(2)
1
(0) = 2,
(3) H
1
(jo
2
) = (jo
2
)
2
m
u
=m
s
.
s
; c
s
40, and c
t
1
y
2
]
T
and let H
2
o
. Then, H
2
= T
z
2
w
(1) H
2
(s) = (k
t
=m
u
)s
2
O(s
3
),
(2) H
2
(0) = H
(1)
2
(0) = 0,
(3) H
2
(jo
1
) = m
s
m
u
=m
s
.
s
; c
s
40, c
t
1
y
2
]
T
and
let H
3
o
. Then, H
3
= T
z
3
w
(1) H
3
(s) = 1 (k
t
=m
u
)s
2
O(s
3
),
(2) H
3
(0) = H
(1)
3
(0) = H
(3)
3
(0) = 0, H
(2)
3
(0) = 2(m
s
m
u
)=k
t
.
As in Propositions 4.74.9,, c
s
does not show up in the above formulae. Thus, whenever y = [y
1
y
2
]
T
, the
damper can be modeled as part of the actuator without affecting the interpolation conditions. The last result is
in contrast with the single measurement case of Propositions 4.14.6. When c
s
= c
t
= 0, D needs not be
Hurwitz; but a slightly more complicated coprime factorization can be performed easily. The reader is referred
to Ref. [16] for more details on this.
The question of controller approximation is in order. More specically, let K and

K be two stabilizing
controllers obtained for the quarter-car model with the measurements y
2
and [y
1
y
2
]
T
and H
k
and

H
k
for
k = 1; 2; 3 denote the corresponding closed-loop transfer functions, respectively. In Ref. [16], when
k
s
; c
s
, and c
t
are all zero, it is shown that the closed-loop transfer functions obtained with a stabilizing
controller that uses the measurements y
2
; _ x
1
; z
3
, and _ x
2
can be approximated within a specied tolerance
by the closed-loop transfer functions of a stabilizing controller that uses only the suspension travel
measurement.
From Eqs. (21) and (38),
H
1

H
1
= (c
t
s k
t
)s
4
D
2
(m
u
s
2
c
t
s k
t
) [m
s
b
Q e q(c
s
s k
s
)]. (42)
Hence,
H
1
(o)

H
1
(o) =
c
s
c
t
m
2
s
m
u
e q(o).
ARTICLE IN PRESS
If e q(o)a0 and c
t
40, there is no way of arbitrarily well approximating

K by a stabilizing compensator K for
the entire range of frequencies. However, if e q(o) = 0, i.e., when e q(s) is a strictly proper transfer function, by
setting
b
Q = e q(sc
s
k
s
)m
1
s
, we get H
1
(s)

H
1
(s) whether c
t
equals zero or not. Note from Eq. (37) that
e q(o) = 0 if and only if
b
Q
1
(o) = 0. The latter equality does not require

K to be strictly proper.
Now, consider the case c
t
= 0. Let
b
Q =
e q(sc
s
k
s
)
m
s
(1 es)
(e40).
As e 0, m
s
b
Q e q(sc
s
k
s
) 0 uniformly on every internal [0; jl), l40 though outside this interval it
diverges as O(s). However, outside the internal the growth is controlled by the factor k
t
s
4
(m
u
s
2
k
t
)D
2
=
O(s
2
). Picking l sufciently large and e sufciently small, the left-hand side of Eq. (42) can be made as small
as desired. The convergences H
2

H
2
and H
3

H
3
as e 0 follow fromthe fact that H
2
and H
3
are continuous
functions of H
1
on the closed right half-plane. The controller approximation result is captured in the following.
Proposition 4.13. Let

H
k
, k = 1; 2; 3 be the closed-loop transfer functions obtained for the quarter car model in
Eq. (1) with k
s
; c
s
40, the measurements y
1
; y
2
, and some stabilizing controller

K. If c
t
= 0 or e q(o) = 0, then for
each e40 a stabilizing controller K that uses only y
2
can be found with the corresponding transfer functions H
k
satisfying [

H
k
(jo) H
k
(jo)[oe for all o and k = 1; 2; 3.
When c
t
= 0, from Propositions 4.6 and 4.12 we have T
z
3
w
(s) = 1 O(s
2
) for all large s, and in this case
Theorem 4 in Ref. [16] reads out
Z
o
0
ln [T
z
3
w
(jo)[ do = p
X
n
k=1
s
k
,
where s
k
, k = 1; . . . ; n denote the real parts of the zeros of T
z
3
w
in the open right half-plane. This is not a
quantitative but a qualitative statement expressing the difculty of controlling the tire deection on a broad
band of frequencies due to the presence of nonminimum phase zeros.
From Propositions 4.14.13, the following conclusions can be drawn:
+ When tire damping is present, utilizing the sprung mass acceleration measurement in addition to the
suspension travel measurement in the feedback law affects only the curvatures of T
z
2
w
and T
z
3
w
at innity,
and T
z
1
w
(o). If tire damping is neglected, one degree higher order terms of the Taylor series expansions of
T
z
k
w
; k = 1; 2; 3 at innity are affected by the additional measurement.
+ Closed-loop performance of any stabilizing feedback law which uses the sprung mass acceleration and the
suspension travel measurements can be obtained within an arbitrary precision by a stabilizing feedback law
relying only on the suspension travel measurement provided that either tire damping is neglected or the
actuator transfer function satises a mild condition in the steady state.
+ No matter how small, tire damping couples the wheel-hop and the heave modes. This coupling eliminates
the constraints of the conventional quarter-car model, which neglects tire damping at the so-called invariant
frequencies o
1
and o
2
. As will be seen in the next section, tire damping improves ride comfort without
sacricing road holding.
+ When the suspension travel is the only available measurement, c
s
inuences T
z
1
w
(o), and the second-order
terms of the Taylor series expansions of T
z
2
w
and T
z
3
w
at innity if c
t
40. If tire damping is neglected, one
degree higher-order terms of the Taylor series expansions of T
z
k
w
; k = 1; 2; 3 at innity are affected by c
s
.
+ If the measurements y
1
and y
2
are both used in the parametrization of the stabilizing controllers, then there
is no need to consider c
s
separately since it can be included in the feedback law.
The analysis of this paper and the results in Ref. [16] show that the constraints on the closed-loop transfer
functions depend on the system parameters as well as the measurements. In the present work, x
1
is taken as a
measured signal instead of x
1
and _ x
1
since in practice, the acceleration is measured, and the velocity and the
position are constructed from the former by integration. The sprung mass acceleration measurement rather
than velocity or position was also considered in Refs. [30, Section 4.5.2]. However, the parameterizability for
ARTICLE IN PRESS
the stabilizing controllers and the constraints on the closed-loop transfer function, in general, depend on
which signal is being used. If x
1
or x
2
is used for the controller parameterization, then one also has to take into
the ltering constraints and trade-offs [29], which are beyond the scope of the current work.
The constraints derived in this paper can be used for the purpose of comparing closed-loop performance of a
proposed controller with benchmark values at specic frequencies. They do not give much information about the
design of an actual controller, which is an involved process, and besides the road disturbance responses many
other factors such as the load disturbance responses and the robustness issues have also to be taken into account.
In Ref. [16], the constraints on the load response functions are derived. Further applications of the controller
parameterization to vehicle active suspension design are reported in Refs. [31,32].
In passing, the constraints in Propositions 4.1 and 4.7 can be viewed as the interpolation constraints. Then,
the problem of nding all stabilizing controller can be cast into a partial realization problem. Many variants of
this formulation, i.e., interpolation with metric and minimal complexity constraints have been considered in
the literature on rational interpolation [27].
5. Active control of the quarter-car model
The purpose of this section is to illustrate the effect of tire damping on the controller design for the quarter-
car model in Fig. 1. The vehicle is assumed to traverse a random road prole with a constant forward velocity
v. Then, the derivative of w(t) is a random process denoted by V
i
(t).
It will be more convenient to dene a new set of state variables in terms of the old state variables in Eq. (1)
as follows:
~ x
1
= x
1
x
2
; ~ x
2
= x
2
w; ~ x
3
= x
3
; ex
4
= x
4
. (43)
Thus,
_
~ x
1
= x
3
x
4
,
_
~ x
2
= x
4
V
i
, and from 1, (2)(4), (9), (43),
_
~ x = Aex B
1
V
i
B
2
u,
z = C
1
ex D
12
u,
y = C
2
ex D
22
u y,
where
A =
0 0 1 1
0 0 0 1
k
s
m
s
0
c
s
m
s
c
s
m
s
k
s
m
u
k
t
m
u
c
s
m
u
c
s
c
t
m
u
2
6
6
6
6
6
6
6
4
3
7
7
7
7
7
7
7
5
; B
1
=
0
1
0
c
t
m
u
2
6
6
6
6
6
4
3
7
7
7
7
7
5
; B
2
=
0
0
1
m
s
1
m
u
2
6
6
6
6
6
6
6
4
3
7
7
7
7
7
7
7
5
,
C
1
=
k
s
m
s
0
c
s
m
s
c
s
m
s
1 0 0 0
0 1 0 0
2
6
6
4
3
7
7
5
; D
12
=
1
m
s
0
0
2
6
6
4
3
7
7
5
,
C
2
=

k
s
m
s
0
c
s
m
s
c
s
m
s
1 0 0 0
2
4
3
5
; D
22
=

1
m
s
0
2
4
3
5
and y is an articially introduced zero-mean white sensor noise uncorrelated with V
i
. Its covariance function
denoted by R
y
satises
R
y
(t) = mId(t).
Here, m40 is a design variable and d(t) is the unit impulse function.
ARTICLE IN PRESS
For simplicity, the random process V
i
is modeled as
V
i
= 2pn
0
kv
_
Z(t), (44)
where Z(t) is a zero-mean white noise process satisfying R
Z
(t) = d(t); and k, n
0
are the road roughness
parameters [28]. In that work, more general road prole models than the integrated white-noise model dened
in Eq. (44) are discussed; and the consequences of the road prole modeling on the random vibration
characterics of the quarter-car model are studied in detail. The roughness parameters in the current study are
set from Ref. [28] as n
0
= 0:15708 cycles per meter and k = 0:76 10
5
. Note the relation T
zV
i
= s
1
T
zw
.
Thus, the Q-parametrization of T
zV
i
can be deduced from the Q-parametrization of T
zw
. In particular, they
share the same invariant frequencies o
1
and o
2
.
The controller will be designed using the linear-quadratic-Gaussian (LQG) design methodology.
Accordingly, u(t) is computed by minimizing
J
LQG
= lim
t
f
o
E
Z
t
f
0
X
3
k=1
r
2
k
z
2
k
r
u
u
2
!
dt
( )
, (45)
where E(c) denotes the expected value of a given random variable c, and r
k
, r
u
are nonnegative weights to be
chosen by the designer. In the simulation, r
u
and r
k
were set, respectively, equal to zero and the root-mean-
square (RMS) values of the open-loop z
k
denoted by RMS
z
k
. Note that even if r
u
were set zero, the control
effort is still penalized in Eq. (45) through the term r
2
1
z
2
1
.
In Figs. 35, the frequency response magnitudes of the passive and the active suspensions using either y
2
only or y
1
and y
2
both as measurements are plotted for the parameter values in Table 1, m = 10
8
, and c
t
= 0.
The RMS values of z
1
; z
2
; z
3
were computed, respectively, as follows: 0.5424, 0.0046, 0.0017 (the passive
suspension); 0.5240, 0.0034, 0.0016 (the active suspension with y
2
measured); 0.5234, 0.0034, 0.0016 (the active
suspension with y
1
and y
2
measured). The frequency responses of the active suspensions for the two
measurement cases are almost identical; and thus, conrming the results in Proposition 4.13 and [16,13].
The natural frequency and the damping ratio of the heave mode are computed as w
h
n
= 1:2507 Hz and
z
h
1
= 0:2178 for the passive suspension. For the wheel-hop mode, they are computed as w
wh
n
= 11:0247 Hz and
z
wh
1
= 0:2013. The invariant frequencies are calculated from Eq. (33) as o
1
= 3:832 Hz and o
2
= 10:610 Hz.
Since o
2
- w
wh
n
, it is difcult to control the wheel-hop mode as clearly seen from Figs. 3 to 5. The 3:5% drop in
the RMS vertical acceleration comes from the suppression of the heave mode vibration. This is possible
ARTICLE IN PRESS
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
2
10
1
10
0
10
1
10
2
Frequency (Hz)
V
e
r
t
i
c
a
l

a
c
c
e
l
e
r
a
t
i
o
n
Fig. 3. The acceleration frequency response magnitude: () passive suspension; (:::) active suspension using the suspension travel
measurement without tire damping; (.) active suspension using the acceleration and the suspension travel measurements without tire
damping.
since the natural frequency of the heave mode given approximately by
k
s
=(m
s
m
u
)
p
is well separated
from o
1
.
Now, let c
t
= 2c
s
. This value is unrealistic for tire damping because it yields w
h
n
= 1:2463 Hz, z
h
1
= 0:2211;
w
wh
n
= 11:0628 Hz, and z
wh
1
= 0:5919. If c
t
is set to 0:1c
s
, then w
h
n
= 1:2504 Hz, z
h
1
= 0:2180; w
wh
n
= 11:0267 Hz,
and z
wh
1
= 0:2209. Hence, the latter seems to be a realistic assumption. In Figs. 68, the counter parts of
Figs. 35 for the same values of the vehicle and the control design parameters but c
t
= 2c
s
are plotted. Clearly,
all the three responses have been improved due to the removal of the invariant frequency at o
2
. For the RMS
values of z
1
; z
2
; z
3
, the following were, respectively, computed: 0.4513, 0.0043, 0.0011 (the passive suspension);
0.2834, 0.0036, 0.0010 (the active suspension with y
2
measured); 0.2724, 0.0037, 0.0010 (the active suspension
ARTICLE IN PRESS
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
6
10
5
10
1
10
2
10
3
10
4
10
0
Frequency (Hz)
S
u
s
p
e
n
s
i
o
n

t
r
a
v
e
l
Fig. 4. The suspension travel frequency response magnitude: () passive suspension; (:::) active suspension using the suspension travel
measurement without tire damping; () active suspension using the acceleration and the suspension travel measurements without tire
damping.
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
5
10
4
10
3
10
2
10
1
Frequency (Hz)
T
i
r
e

d
e
f
l
e
c
t
i
o
n
Fig. 5. The tire deection frequency response magnitude: () passive suspension; (:::) active suspension using the suspension travel
measurement without tire damping; (.) active suspension using the acceleration and the suspension travel measurements without tire
damping.
with y
1
and y
2
measured). Comparison of Figs. 68 with Figs. 35, and the modal natural frequencies and the
damping ratios shows that the improved responses are achieved by suppressing the wheel-hop vibration.
In Fig. 9, the vertical acceleration frequency response magnitude is plotted for the case c
t
= 0:1c
s
. The
suspension travel and the tire deection responses are similar to those in Figs. 4 and 5. The RMS values for
this case are, respectively, 0.5259, 0.0045, 0.0017 (the passive suspension); 0.4895, 0.0034, 0.0016 (the active
suspension with y
2
measured); 0.4900, 0.0034, 0.0016 (the active suspension with y
1
and y
2
measured). The
RMS vertical acceleration is reduced by 6:83% which is about twice of the reduction computed for the case
c
t
= 0. Though as not impressive as the overdamped tire case, the last result shows that the inuence of tire
damping certainly needs to be taken into account in the design of active suspensions to improve ride quality.
ARTICLE IN PRESS
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
2
10
1
10
0
10
1
10
2
Frequency (Hz)
V
e
r
t
i
c
a
l

a
c
c
e
l
e
r
a
t
i
o
n
measurement with tire damping c
t
= 2c
s
: (.) active suspension using the acceleration and the suspension travel measurements with tire
damping c
t
= 2c
s
.
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
5
10
1
10
2
10
3
10
4
10
0
Frequency (Hz)
S
u
s
p
e
n
s
i
o
n

t
r
a
v
e
l
Fig. 7. The suspension travel frequency response magnitude: () passive suspension; (:::) active suspension using the suspension travel
t
= 2c
s
; (.) active suspension using the acceleration and the suspension travel measurements with tire
damping c
t
= 2c
s
.
The rest of this section will be devoted to further enhancement of the closed-loop performance by means of the
interpolation approach of this paper.
In light of the controller approximation result, it is enough to consider the case y = y
2
. Recall that T
zV
i
=
s
1
T
zw
which implies from Eq. (21) that
T
z
1
V
i
= sD
1
(c
t
s k
t
){c
s
s k
s

b
Qm
s
s
2
D
1
(m
u
s
2
c
t
s k
t
)]. (46)
Put c
t
= ac
s
(a40) and H
k
(s; a;
b
Q) = T
z
k
V
i
; k = 1; 2; 3, where we have made the dependence on the parameters
c
t
and
b
Q explicit. Let a
1
= 0:1; a
2
= 2, and Q
and Q
]
denote the
b
Q parameters of the compensators designed
by the above LQG method with c
t
= a
1
c
s
and c
t
= a
2
c
s
, respectively. As far as the closed-loop responses are
ARTICLE IN PRESS
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
5
10
4
10
3
10
2
10
1
Frequency (Hz)
T
i
r
e

d
e
f
l
e
c
t
i
o
n
Fig. 8. The tire deection frequency response magnitude: () passive suspension; (:::) active suspension using the suspension travel
t
= 2c
s
damping c
t
= 2c
s
.
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
2
10
1
10
0
10
1
10
2
Frequency (Hz)
V
e
r
t
i
c
a
l

a
c
c
e
l
e
r
a
t
i
o
n
t
= 0:1c
s
damping c
t
= 0:1c
s
.
concerned, H
k
(s; a
2
; Q
]
); k = 1; 2; 3 are satisfactory while H
k
(s; a
1
; Q
) are not. Thus, the interpolation

problem to be studied is formulated as follows:
Does there exist a
b
Q c RH
o
such that H
1
(s; a
1
;
b
Q) = H
1
(s; a
2
; Q
]
) ?
If there exists a solution to this problem denoted by
b
Q, then the quarter-car model in Fig. 1 with c
t
= 0:1c
s
will have the closed-loop responses H
k
(s; a
2
; Q
]
); k = 1; 2; 3 using the unique controller K corresponding to
this
b
Q. Unfortunately, the formulated problem has no solution. To see this, rst obtain the complete
interpolation conditions for T
z
1
V
i
from Eq. (46) as follows:
(i) H
1
(s) = (c
s
c
t
=m
s
m
u
)s
1
O(s
2
),
(ii) H
1
(0) = H
//
1
(0) = 0; H
/
1
(0) = 1.
Then, from the requirement formulated above and (i):
s H
1
(s; a
1
;
b
Q)[
s=o
= s H
1
(s; a
2
Q
]
)[
o
,
which forces a
1
equal to a
2
. Hence, there does not exist any solution.
Having seen the infeasibility of this interpolation problem, consider now the following variant:
Does there exist any
b
Q c RH
o
such that H
1
(s; a
1
;
b
Q) = H
1
(s; a
2
; Q
]
)C(s) for some C c RH
o
?
Fortunately, there exists a solution to the latter problem. In fact, from the interpolation conditions for T
z
1
V
i
,
it sufces to pick any C c RH
o
satisfying
(iv) C(0) = 1,
(v) C
/
(0) = 0,
(vi) C(o) = a
1
=a
2
.
It is easy to see that the following transfer function:
C(s) =
ss
2
as b
s
2
as b
; a; b40,
where s = a
1
=a
2
has the aforementioned properties. Furthermore, for a given O which is sufciently larger
than the wheel-hop frequency o
wh
n
, if a and b are chosen so that C(jo) is a good approximation to the low-pass
lter:
L
O
(o) =
1; 0popO;
s; o4O
(
on the frequency band [0; O], then a good match to the vertical acceleration response plotted in Fig. 6 by the
solid line is obtained. From the continuity of the trade-off curves, it follows that the other two responses are
satisfactory as well.
It remains to calculate
b
Q. To this end, from Eq. (46),
b
Q =
H
1
(s; a
2
; Q
]
)C(s)D(s; a
1
) s(a
1
c
s
s k
t
)(c
s
s k
s
)
m
s
s
3
(a
1
c
s
s k
t
)(m
u
s
2
a
1
c
s
s k
t
)
D(s : a
1
), (47)
where D(s : a
1
) is calculated from Eq. (8) with c
t
= a
1
c
s
. Since the degree of H
1
(s; a
2
; Q
]
) is 8, the degree of
b
Q(s)
is bounded above by 13. The numerator polynomial of
b
Q before cancellations has order 18. Recall how C(o)
was selected. This drops the order of the numerator polynomial by two. Three more degrees are canceled by
the denumerator factor s
3
. The end result is a proper transfer function
b
Q. Finally, K is calculated from Eq. (24)
with
b
Q in Eq. (47).
It may seem difcult to keep track of pole-zero cancellations. An easy way to circumvent this numerically
ill-conditioned procedure is to evaluate
b
Q(s) and/or its derivatives by computing the right-hand side of
Eq. (47) and/or its-derivatives at a set of sufciently many and arbitrarily selected frequencies s
k
; and from
these evaluations, obtain directly a minimal state-space realization of
b
Q(s). For this purpose, numerically
efcient robust algorithms developed in Refs. [33,34], which deal with multi-variable data as well, can be used.
ARTICLE IN PRESS
In Fig. 10, the acceleration frequency response magnitude of the active suspension designed by using the
suspension travel measurement with the (ctitious) tire damping c
t
= 2c
s
plotted in Fig. 6 is reproduced along
with the acceleration frequency response magnitude of the active suspension designed by the hybrid algorithm
outlined above. The purpose of plotting them together was to show that the approximation H
1
(s; a
1
;
b
Q) -
H
1
(s; a
2
; Q
]
) is very accurate in the bandwidth of interest. This, in turn, implies that the approximations
H
2
(s; a
1
;
b
Q) - H
2
(s; a
2
; Q
]
) and H
3
(s; a
1
;
b
Q) - H
3
(s; a
2
; Q
]
) are also very accurate in the same band of the
frequencies. A comparison of Fig. 9 with Fig. 10 reveals impressive closed-loop performance enhancement by
the interpolation approach. In the simulation, c(s) was chosen as
c(s) =
0:05s
2
500s 10
s
2
500s 10
.
In Fig. 11, the actuator frequency response magnitudes of the active suspensions designed by the LOG
methodology with tire dampings c
t
= 0:1c
s
and c
t
= 2c
s
, and the hybrid algorithm with tire damping c
t
=
0:1 c
s
using the suspension travel measurement are plotted. Fig. 11 shows that the closed-loop performance
enhancement by the interpolation approach is achieved at a reasonable price. Actually, the increase in the
actuator gain is less than 30 decibels for all frequencies. Simulations for the values of tire damping at the
equally spaced 101 points between and including 0:001 c
s
and 0:1 c
s
were also carried out. The numerical
results plotted in Fig. 12 indicate that the actuator frequency response magnitudes of the active suspensions
designed by either the LOG methodology or the hybrid algorithm using the suspension travel measurement
are insentive to changes in tire damping; hence conrming the predication about the efcacy of coupling
between the motions of the sprung and unsprung masses.
Moreover, in Fig. 11, the actuator gain is seen to peak at the heave and the wheel-hop frequencies, which
indicates that a stable inversion of the vehicle transfer function is taking place by canceling these modes. This
feature is reminiscent of the loop-transfer-recovery (LTR) synthesis [35] that applies to square and minimum-
phase plants. Therefore, the hybrid algorithm can be viewed as a loop-shaping method realized in two-stages.
The stages are the minimization of the quadratic criterion in Eq. (45) and the interpolation procedure.
Although, in principle, it is possible to obtain a desired solution in one step by the LQG methodology, it is not
clear how to accomplish this task since the quadratic criterion in Eq. (45) involves nine free weights in its most
general form. Due to the coupling between the modes, it is a non-trivial matter to steer these weights towards a
desired solution. The hybrid algorithm, on the other hand, ignores the interactions among the variables
z
k
; k = 1; 2; 3 and u in the rst stage. Then, the effect of the interactions is taken care of in the interpolation
ARTICLE IN PRESS
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
2
10
1
10
0
10
1
Frequency (Hz)
V
e
r
t
i
c
a
l

a
c
c
e
l
e
r
a
t
i
o
n
Fig. 10. The acceleration frequency response magnitude: () active suspension using the suspension travel measurement with the
(ctitious) tire damping c
t
= 2c
s
; (.) active suspension designed by a mixture of the LQG methodology and the interpolation approach
using the suspension travel measurement with the (actual) tire damping c
t
= 0:1c
s
.
stage. The reader is cautioned not to draw broad conclusions based on this example solely since hardware
limitations and uncertainties, in particular uncertainty in the tire model, may degrade actuator performance.
6. Conclusions
In this paper, the exibility of shaping the closed-loop road frequency responses of a quarter-car model by
feedback control was investigated. The constraints on the achievable responses of the quarter-car active
suspension systems were derived for a wide range of the suspension parameters. The derived constraints
complement the existing results in the literature on vehicle dynamics and control. Also, using the factorization
ARTICLE IN PRESS
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
2
10
3
10
4
10
5
10
6
10
7
Frequency (Hz)
A
c
t
u
a
t
o
r

G
a
i
n
Fig. 11. The actuator frequency response magnitude using the suspension travel measurement: () the LQG design with tire damping
c
t
= 2c
s
; ( ) the LQG design with tire damping c
t
= 0:1c
s
; (.) the hybrid algorithm with tire damping c
t
= 0:1c
s
.
10
3
10
2
10
1
10
0
10
1
10
2
10
3
10
5
10
4
10
3
10
2
10
6
10
7
Frequency (Hz)
A
c
t
u
a
t
o
r

G
a
i
n
Fig. 12. The actuator frequency response magnitude using the suspension travel measurement: () the LQG design with tire damping
c
t
= 2c
s
; ( ) the LQG design with tire damping c
t
at the equally spaced 101 points between and including 0:001c
s
and 0:1c
s
; (.) the hybrid
algorithm with tire damping c
t
at the same 101 points.
approach of the feedback stability, it was shown that tire damping by coupling the motions of the sprung and
unsprung masses eliminates a constraint on the wheel-hop mode. The inuence of tire damping on the design
of an active suspension for a lightly damped quarter-car model by a mixture of the LQG methodology and the
interpolation approach was also illustrated.
The study of the constraints on the achievable performance has remained largely restricted to pointwise
constraints on the frequency responses while ride comfort and safety criteria are mostly expressed in terms of
the RMS values of the related transfer functions. Hence, a study of the constraints on the achievable RMS
responses warrants future research.
References
[1] A.G. Thompson, Design of active suspensions, Proceedings of the Institution of Mechanical Engineers 185 (19701971) 553563.
[2] D. Karnopp, M.J. Crodby, R.A. Harwood, Vibration control using semi-active force generators. Transactions of the ASME, Series B,
Journal of Engineering for Industry 96 (1974) 619626.
[3] A.G. Thompson, An active suspension with optimal linear state feedback, Vehicle System Dynamics 5 (1976) 187192.
[4] D. Hrovat, D.L. Margolis, M. Hubbard, An approach toward the optimal semi-active suspension, Transactions of the ASME, Journal
of Dynamic Systems, Measurement, and Control 110 (1988) 288296.
[5] D. Hrovat, Application of optimal control to advanced automative suspension design, Transactions of the ASME, Journal of Dynamic
Systems, Measurement and Control 115 (1993) 328342.
[6] M. Yamashita, K. Fujimori, K. Hayakawa, H. Kimura, Application of H
o
control to active suspension systems, Automatica 30
(1994) 17171729.
[7] D. Karnopp, Active and semi-active vibration isolation, Transactions of the ASME, Journal of Mechanical Design 117 B (1995)
177185.
[8] D. Hrovat, Survey of advanced suspension developments and related optimal control applications, Automatica 33 (1997)
17811817.
[9] H.D. Tuan, E. Ono, P. Apkarian, Nonlinear H
o
control for an integrated suspension system via parameterized linear matrix
inequality characterizations, IEEE Transactions on Control Systems Technology 9 (2001) 175185.
[10] I. Fialho, G.J. Balas, Road adaptive suspension design using linear parameter-varying gain scheduling, IEEE Transactions on Control
Systems Technology 10 (2002) 4354.
[11] A. Alleyne, J.K. Hedrick, Nonlinear adaptive control of active suspensions, IEEE Transactions on Control Systems Technology 3
(1995) 94101.
[12] H. Chen, K.H. Guo, Constrained H
o
control of active suspensions: an LMI approach, IEEE Transactions on Control Systems
Technology 13 (2005) 412421.
[13] C. Yue, T. Butsuen, J.K. Hedrick, Alternative control laws for automotive active suspensions, Transactions of the ASME, Journal of
Dynamic Systems, Measurement, and Control 111 (1989) 286291.
[14] J.K. Hedrick, T. Butsuen, Invariant properties of automotive suspensions, Proceedings of the Institution of Mechanical Engineers, Part
D, Transport engineering 204 (1990) 2127.
[15] D. Karnopp, Theoretical limitations in active suspensions, Vehicle System Dynamics 15 (1986) 4154.
[16] M.C. Smith, Achievable dynamic response for automotive active suspensions, Vehicle System Dynamics 24 (1995) 134.
[17] M.C. Smith, G.W. Walker, Performance limitations and constraints for active and passive suspensions: a mechanical multi-port
approach, Vehicle System Dynamics 33 (2000) 137168.
[18] G. Jianmin, R. Gall, W. Zuomin, Dynamic damping and stiffness characteristics of the rolling tire, Tire Science and Technology 29
(2001) 258268.
[19] H.B. Pacejka, Tyre and Vehicle Dynamics, second ed., Butterworth-Heinemann Burlington, Massachusetts, 2006.
[20] B.S. Kim, C.H. Chi, T.K. Lee, A study on radial directional natural frequency and damping ratio in a vehicle tire, Applied Acoustics
68 (2007) 538556.
[21] J.A. Levitt, N.G. Zorka, Inuence of tire damping in quarter car active suspension models, Transactions of the ASME, Journal of
Dynamic Systems, Measurement, and Control 113 (1991) 134137.
[22] A.G. Thompson, Effect of tyre damping on the performance of vibration absorbers in active suspension, Journal of Sound and
Vibration 133 (1989) 457465.
[23] A. Zaremba, R. Hampo, D. Hrovat, Optimal active suspension design using constrained optimization, Journal of Sound and Vibration
207 (1997) 351364.
[24] R.M. Chalasani, Ride performance potential of active suspension systemsPart I: Simplied analysis based on a quarter-car model,
American Society of Mechanical Engineers, Applied Mechanics Division AMD 80 (1986) 187204.
[25] M. Vidyasagar, Control System Synthesis: A Factorization Approach, MIT Press, Cambridge, MA, 1985.
[26] K. Zhou, J.C. Doyle, K. Glover, Robust and Optimal Control, Prentice-Hall, Upper Saddle River, 1996.
[27] J.A. Ball, I. Gohberg, L. Rodman, Interpolation of Rational Matrix Functions, Birkha user, Basel, 1990.
[28] S. Tu rkay, H. Akc- ay, A study of random vibration characteristics of the quarter-car model, Journal of Sound and Vibration 282 (2005)
111124.
ARTICLE IN PRESS
[29] G.C. Goodwin, M.M. Seron, Fundamental design tradeoffs in ltering prediction and smoothing, IEEE Transactions on Automatic
Control 42 (1997) 12401251.
[30] G.W. Walker, Constraints upon the Achievable Performance of Vehicle Suspension Systems, PhD Thesis, University of Cambridge,
1997.
[31] M.C. Smith, F.C. Wang, Controller parameterization for disturbance response decoupling: application to vehicle active suspension
control, IEEE Transactions on Control Systems Technology 10 (2002) 393407.
[32] F.C. Wang, M.C. Smith, Disturbance response decoupling and achievable performance with application to vehicle active suspension,
International Journal of Control 75 (2002) 946953.
[33] T. McKelvey, H. Akc-ay, L. Ljung, Subspace-based multivariable system identication from frequency response data, IEEE
Transactions on Automatic Control 41 (1996) 960979.
[34] H. Akc- ay, S. Tu rkay, A subspace-based approach to LagrangeSylvester interpolation of rational matrix functions, SIAM Journal on
Matrix Analysis and Applications 29 (2007) 377395.
[35] G. Stein, M. Athans, The LQG/LTR procedure for multivariable feedback-control design, IEEE Transactions on Automatic Control
32 (1987) 105114.
ARTICLE IN PRESS

Aspects of Achievable Performance For Quarter-Car

Uploaded by

Copyright:

Available Formats

Aspects of Achievable Performance For Quarter-Car

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Aspects of Achievable Performance For Quarter-Car

Uploaded by

Copyright:

Available Formats

JOURNAL OF

) are not. Thus, the interpolation

You might also like