Automatica, Vol.26, No. 5. pp. 845-860. 1990 0005-1098/90$3.00+ 0.

00
Printedin GreatBritain. PergamonPressplc
~) 1990InternationalFederationof AutomaticControl

Optimal Active Suspension Structures for

Quarter-car Vehicle Models*
D. HROVATt

The paper explores the connections between optimal one and two degree-
of-freedom vehicle structures incorporating active suspensions. It is shown
that the resulting, special one DOF structures represent limits of the best
possible performances attainable with two DOF counterparts.
Key Words--Active suspensions; optimal control; singular control; vehicles; vibration control.

Absh-act--The preliminary design of active vehicle suspen- wheels, trucks and other unsprung masses.
sions can be naturally cast into an equivalent linear-quadratic Additional constraints are imposed by the
Gaussian (LQG)-optimization problem, where the perfor-
mance index is an appropriate combination of ride comfort, overall system robustness, reliability and cost
vehicle handling and suspension design constraints. The requirements.
vehicle models vary in complexity with the simplest one Different suspensions satisfy the above re-
being the one degree-of-freedom (DOF) quarter-car model
which neglects wheel dynamics. The present paper explores quirements to differing degrees. Although
the connections between LQG-optimal one DOF and two significant improvements can result from a
DOF models. In a recent study (Hrovat, 1987), it was shown designer's ingenuity, on the average, suspension
that for the optimal two D O F systems, both ride and
handling can be improved by reducing the unsprung mass. performance mainly depends on the type or class
The limiting one DOF case of zero unsprung mass then of suspension used. Here one distinguishes, in an
represented the maximum possible improvement within the ascending order of improved performances,
constraint of a single active actuator. It is shown in the
present paper that in practice these limiting benefits can be between passive, semi-active, and fully active
approached via passive dynamic absorbers attached to the suspensions. Passive suspensions can be found
unsprung mass. Moreover, the structural constraint of a on most of the conventional vehicles. Roughly,
single active actuator has been relaxed to allow for both
secondary suspension and unsprung mass active actuators. they are characterized by the absence of external
Here the maximum possible ride and handling improvements power sources, whereas the active suspensions
for two DOF systems are obtained in the limiting case of require additional energy sources, such as
singular control with zero penalty on unsprung actuator
force. This absolutely optimal structure turns out to be yet compressors or pumps, to achieve superior ride
another special one DOF system, which in practice could be and/or handling performance. More precise
approached with the help of active dynamic absorbers. definitions of active and passive suspensions
were given by Hrovat (1982) following similar
1. INTRODUCTION definitions of passivity and passive operators as
THE DESIGN of ground vehicle suspensions is used, for example, in electrical networks and
influenced by a number of factors. A good mathematics. Finally, the semi-active actuators
vibration isolation is required to secure the (see for example Crosby and Karnopp, 1973;
occupants' ride comfort, whereas good road Hrovat and Margolis, 1981; Morman and
holding is important for vehicle handling and, in Giannopoulos, 1982) as their name implies, fill
general, enhanced safety. Key design constraints the gap between passive and active suspensions
are represented by maximum allowable relative since they offer significant performance im-
displacement between the vehicle body and provements while requiring relatively small
various suspension components, including external power sources.
The possible applications of active and
semi-active suspensions range from improved
maneuverability of tanks and off-road vehicles
* Received 10 September 1987; revised 10 November 1988; (Roley, 1975), to improved ride/handling of
revised 20 October 1989; received in final form 24 November
1989. The original version of this paper was not presented at conventional automotive vehicles and superior
any IFAC meeting. This paper was recommended for ride height and pitch control of championship
publication in revised form by Associate Editor B, Friedland race cars (Metz and Maddock, 1986). The
under the direction of Editor H. Austin Spang III.
t Research Staff, Ford Motor Company, P.O. Box 2053, potential benefits of active suspensions have
Dearborn, MI 48121, U.S.A. been assessed with the help of optimal control
845
846 D. HROVAT

__I ms 2 x4 I r"s l--txs [ ms ]

mus=0

kul

(a) (b) (c)

FIG. 1. (a) Two-degree-of-freedom vehicle model with active suspension S; (b) corresponding limiting case
when mu~ = 0; and (c) two DOF model with conventional passive suspension. In the figure m, and mus
denote vehicle sprung and unsprung mass, respectively.

theory since the early 1960s; see for example the where
survey by Morman and Giannopoulos (1982). Xus = tire deflection from equilibrium
Many of these investigations (e.g. Bender, 1967, X s = suspension deflection (rattlespace)
1968; Karnopp and Trikha, 1969; Hrovat and from equilibrium
Hubbard, 1981, 1987) used the simplest possible, u = sprung mass acceleration
one degree-of-freedom (DOF), quarter-car veh- rt, r2 =weighting parameters; these are the
icle model, where unsprung mass dynamics or tuning parameters such that larger rt
wheel-hop had not been modeled. Strictly, the or 1"2 results in smaller tire or
model represents the dominant dynamics of suspension deflection, respectively.
some advanced air cushion and magnetically The states xus and xs in PI1 correspond to the
levitated vehicles having only primary suspen- state-space description of the two DOF model
sion (Hullender et al., 1972). For other vehicles under the assumptions summarized in Appendix
it is a useful approximation since its simple A. The rms acceleration in (1) is the measure of
structure reveals the dominant trends and, as ride comfort (Smith et al., 1978). The tire
shown below, it establishes the limiting perfor- deflection is an indicator of road holding ability
mances for more complex models. which is relevant for vehicle handling, and the
Some of the aspects of the limiting perfor- suspension deflection represents design con-
mance have been explored in a recent study straints on available suspension stroke. Thus the
(Hrovat, 1987). The study il,vestigates the optimal suspension force which will minimize the
influence of unsprung mass on ride and handling above PI~ will in effect optimize passengers'
quality of active suspension vehicles modeled as comfort while satisfying road holding and
quarter-car, two D O F systems shown in Fig. 1. suspension design constraints. Besides displaying
The results demonstrate that for active suspen- the best possible ride/handling/design tradeoffs,
sions, both ride and handling can be improved the above optimization offers the advantage of
by r e d u c i n g the unsprung mass. In the limit, having to tune only two parameters, the PI~
when the unsprung mass equals zero, the two weights rt and rE.
DOF configuration of Fig. la reduces to a The performance index Pit, or subsets of it
special one D O F structure shown in Fig. lb, (with rt or rE set to zero) has been used by a
with the tire stiffness k~ being the same for both number of researchers, including Bender (1967),
cases. For most road/speed conditions, the Thompson (1984) and Karnopp (1986). In one of
limiting, one D O F model results in the best the first detailed studies, Chalasani (1986) used
performance with respect to given ride/handling/ unsprung mass velocity instead of tire deflection
design tradeoffs specified by equation (1) below. in PI1, to conclude that, within the assumed
Thus the one D O F model serves as a reference constraints, there is only a limited advantage
indicating maximum ride improvements that are offered by active suspensions when compared to
possible with any two D O F structure whose the conventional passive suspensions. Extensive
force actuator is constrained as in Fig. la. studies based on PI~ (Hrovat 1984a, 1987)
Quantitatively, the results from Hrovat (1987) identify substantial potential advantages of
were obtained by minimizing the following active suspensions in their structural properties
(steady-state) performance index and inherent adaptive capabilities, that can be
used for adaptive suspension tuning based on
2 + rtxus.~ms
Pit = U~ms z + r2x; . . . . (1) varying road/speed conditions.
 Optimal active suspension structures 847

iiiii ii iiiii ' . . . . . . . . . . . . "T ........................ ' ............

I0
IO0 i i ' i ! --- ~_~.~I00
I ~[i[i
Z
IIIII IIIIII /
0 IIIll Illlll ]

Illll
: : :1:
! .llllN
r .....
_]
w
J
W [ I ] Ill
.... i !iiii!! ! !

ili!iliii!i
IIIII . ; , ~ ~.....:,.
•

I I ILl ~ Ii |i ]%]N.I I
[ F~,l.. lI
N tillt
1 Illll .5

I1: .I
11111 r¢
O .I
Q
Z .I I I0 z .DOt .or .t i 1o
NORMALIZED RMS SUSPENSIONSTROKE~s [ SECI/~] NORMALIZED RMS TIRE DEFLECTION ~us [ SECz/z]

(a) (b)
FIG. 2. C o m p a r i s o n b e t w e e n a p e r f o r m a n c e of conventional passive s u s p e n s i o n (co s = ~ = 2~rr/s,
p = 10, 0J.~ = 2 zrlOr/s, 0 < ~ < 1), optimal 2 D O F s u s p e n s i o n (oJ., = 2 z 10 r/s, p = 10), and o p t i m a l
limiting case w h e n mus = 0: (a) Li versus .~, curves for r I = I0 4 and r 2 = 10-5; (b) ,7 versus $.~ curves for
r. = 10 -4 and r 2 = 10 -5.

The optimal results from Hrovat (1987) for the already achieved maximum possible improve-
2 DOF model of Fig. la are shown in Fig. 2. The ments for a given structure, the above
results have been normalized by dividing the comparison between one and two D O F models
optimal rms values by 2V~-~AV, where A is the points to the need for fundamental
road roughness coefficient and V is vehicle structural/design changes that will lead to
forward velocity. Figure 2a shows normalized increased effective wheel-hop damping, without
rms sprung mass acceleration versus normalized associated penalty on the ride comfort. One such
rms suspension stroke, and Fig. 2b displays the possibility is offered by the dynamic absorber
normalized rms acceleration versus the normal- (Thompson 1970) which is often used in the
ized rms tire deflection. Only the limiting presence of strong, lightly-damped oscillations.
optimal curves r~ = 10 -4 and r2 = 10 -5 are shown, This will be discussed in Section 2. Further ride
since they enclose most of the solutions of and handling improvements are possible using
practical significance. To put the optimal two active means of unsprung mass vibration control.
DOF active performance in proper perspective Their performance will be discussed in Section 3
the curves for a typical passive two D O F system and compared with the best possible active
(Fig. lc) with bounce natural frequency of one configuration, which, as shown in Section 4,
Hz are also shown in Fig. 2, as well as the turns out to be yet another one D O F system.
limiting performance of the one D O F system
(Fig. lb), which serves as a reference of
2. P A S S I V E D Y N A M I C A B S O R B E R S ( D A ) F O R
maximum possible ride improvements with UNSPRUNG MASS VIBRATION CONTROL
actuator force constrained as in Fig. la.
From Fig. 2 it can be seen that while the 2.1. Dynamic absorber as an add-on to the
optimal active suspension for two D O F systems optimal two D O F systems
significantly outperforms the passive counter- In the past the dynamic absorbers (DA) have
part, it falls short of the optimal one D O F been widely utilized to reduce excessive
performance. For example, at normalized rms vibrations associated with lightly-damped str-
suspension stroke of 0.6 s 1/2, the one DOF case uctural vibration modes. The applications ranged
results in normalized rms acceleration of 3 s -3/2, from electric hair clippers (Den Hartog, 1956) to
whereas the two D O F results in 10.9 s -3/z, which tuned mass dampers for structural control of
amounts to a 263% increase. The main reason buildings (Hrovat et al., 1983). More recently,
for this deterioration stems from the conflicting the absorbers have been used in synergism with
requirements imposed on the active actuator of controls for passive vibration control of helicop-
Fig. la: it should simultaneously provide small ters and large space structures. In automotive
sprung mass acceleration (for smooth ride) and applications dynamic absorbers have often been
unsprung mass damping to reduce the wheel-hop used (many times as an afterthought) to reduce
(for good handling). (unexpected) vibrations. However, it seems that
Recognizing that the optimal control has there was only one widespread, production
848 D. HROVAT

__! =, lt'° ¢..Ous=

i,2

ba
¢.=-- pa = - m u s
2maO)a m a
==.=, ()
where x l = x , s and x3=xs are tire and
suspension deflections from equilibrium, xz and
o, :,. x4 are unsprung and sprung mass velocities,
respectively, x5 and x6 are DA relative
I =°', ]-='=, displacement and velocity, respectively, and w is
the road roughness velocity described by power
spectral density W = A V , with A being road
roughness coefficient and V vehicle speed.
A simple application of the dynamic absorber
would be to append it to an already optimized
FrG. 3. Schematic of a two D O F vehicle model with dynamic two DOF system. As a generic example, the
absorber.
following typical system parameters were cho-
application of dynamic absorbers for vehicle ride sen: sprung-to-unsprung mass ratio P = 10,
improvement. This is the case of a well-known unsprung mass natural frequency ~o,= =
French subcompact, Citroen 2CV, first intro- 2zrl0rads -1, unsprung-to-DA mass ratio pa =
duced in 1949. A superior ride comfort of this m J m a = 10 and DA damping ratio ~a=0.2.
somewhat unique car is achieved through a The corresponding (suboptimal) rms results
special spring arrangement, where an effective were obtained using Matrixx program (Integ-
low spring rate results for the body pitch mode. rated Systems, Inc., 1984, 1986) and are shown
However, such a soft spring rate would reduce in Fig. 4, which also contains the results for the
the wheel-hop damping, thus compromising optimal two DOF system without DA, and the
vehicle handling. The need to improve the latter reference one DOF case. From Fig. 4a it can be
probably led to the introduction of the dynamic seen that, in terms of rms acceleration and
absorber in this car, Similar motivation applies suspension stroke, the system with DA ap-
in the case of active suspensions. proaches the best possible one DOF perfor-
A schematic of a quarter-car vehicle with an mance, and is substantially better, than the
active suspension and a dynamic absorber is original two DOF case. At the same time Fig.
shown in Fig. 3. The absorber consists of a 4b demonstrates that the tire deflection has also
relatively small mass m, (typically 5 to 20 times been significantly reduced. This confirms an
smaller than the unsprung mass m~s), viscous earlier conjecture (Hrovat, 1985) based on
damper b~ and spring k~ which is tuned to, or in preliminary work with DA for the single
the neighbourhood of, the unsprung mass optimization point r t =500 r2= 100 (Hrovat,
natural frequency ~o~ = ~ . The complete 1984b). More precisely, assuming that the tire
mathematical description of the quarter-car deflection is adequate, Fig. 4a shows that for
model with the absorber is given next. xs = 0.6 s 1/2, performance with DA is very close
The equations describing the quarter-car to that of one DOF system; for 0.3 <£'s < 0.6 s t/2
model from Fig. 3 are the DA performance falls between the two DOF
and one DOF performances; and for £s < 0.3 s l/z
i = A x + Bu + r w (2) the DA and two DOF curves almost coincide. It
i
0 1 00 0 0 - 0 should be pointed out that operating points of
2 practical significance for automotive applications
-o~ z - 2 ¢ . w" 0 0 ~°__.!. 2~ °__l
~ P fall in the range of, approximately, 0.2 < £ s <
P~ P~ Pa l s t/z. As a final remark, the required DA
A= 0 -1 0 1 0 0 B= 0 strokes corresponding to Figs 4a and 4b are also
0 0 00 0 0 -1 shown in Fig. 4c.
Illustrative Example 1. As an example of a
. . . . . - Y .... b 6 - b ...... ; 0 possible ride improvement with dynamic
0 2~.~o. 0 0 -w2-2¢.oJ~ ~.0_ absorbers, consider the design case A from
r=[-1 0 0 0 0 0]' Hrovat (1987). There, the vehicle specified via
p = 10, f,~= 10Hz was traversing a medium
U quality road with A = 1 . 6 × 1 0 - 5 f t = 4 . 9 ×
Usa--
ms 10 - 6 m , with the speed of V = 80ft s-t =
 Optimal active suspension structures 849

i l !!iiiiii! ~,i~i i
+..+-. iiii-~~ - i - - . - i i ~ ~ -x~
r~--~T~r~~ .-~.~...-! .......
~ - .~.
. ,. ;.H.~.~ 4 - ; . . 4 .... °<Li ...L...,...~.~LJ..L..-..~ ........ 6~
o "~ ~.._=

...... iL/il :

n~ ........................ rv

. . . . . . . . . . . . . . O

~_ o - -
8 2 - -

~/~_0]S] ~ NOIJ.V~I3"I30~' SIN~I Q3ZI'IVIN~ON [Z~-:33S] ~ NOIJ.V~I3~00V SINI:I a'4ZI'lV~ION

850 D. HROVAT

8 8 . 5 k m h -~. The corresponding optimal two wheel-hop damping introduced via DA. The
D O F performance resulted in the following increased damping is also reflected in the
values: normalized rms tire deflection 2us = frequency response curves between sprung mass
0.3 S 1/2 which implies 2 ...... = 1/3 inch = acceleration and ground velocity as shown in
0.847cm, normalized rms suspension stroke Fig. 6, where a substantial reduction of unsprung
£s = 0.6 s v2 so that x ..... = 2/3 inch = 1.694 cm; mode resonance can be seen in the case of DA.
and the normalized rms acceleration u = For practical implementations, it is often
10.9s -3/2, which implies an rms acceleration of possible to simplify the above optimal control
3.0%g. The p e r f o r m a n c e index weights were law as was done in Hrovat (1987). This is
rl = 1100, r2 = 100, and the closed-loop eigen- accomplished by neglecting the tire deflection
values were e L 2 = 2 . 2 0 + j 2 . 2 6 , and e3.4 = - feedback so that the resulting normalized control
2.75+/'62.9. This design is shown in Fig. 4 as force becomes
point A.
u = -0.5478x2 + 1 0 x 3 + 4.4375X4
The following dynamic absorber has been
attached to the above vehicle model while = 0.5478(x4 - x2) + 10X3 q- 3. 8897X4 (3)
maintaining the same control gains as in the case where the states x~ to x4 are defined via Fig. la.
A. The D A natural frequency was chosen as This control implementation requires only one
wa = kV'~a/m~ = coos = 2srl0, with the unsprung- acceleration m e a s u r e m e n t for sprung mass
t o - D A mass ratio p ~ = m , J m a = 10 and D A velocity and an L V D T for suspension stroke
damping ratio ¢, = bJ2maO)a = 0.2. The result- measurement. The rms normalized acceleration,
ing suboptimal p e r f o r m a n c e was given through: tire deflection and suspension stroke with
£~s = 0.2028 s v2 or x~,rms = 0.225 inch = simplified control law (2) are a = 7 . 5 5 7 5 s -3/2,
0.572 cm, and a = 7.5739 s -3/2 or U~m~= 2.08%g. £ 1 = 0 . 2 0 0 6 s v2, and £ 3 = 0 . 5 6 6 8 s v2, which are
Thus a reduction of 30.7% in rms sprung mass for all practical purposes very close to the values
acceleration has been achieved with the addition based on original four-state feedback where
of the dynamic absorber. From Fig. 4b it can be a = 7.5739 s-3n,£1 = 0.2028 sm and.f3 = 0.5655 s In.
seen that the dynamic absorber also contributes The above results were obtained using a
to a reduction in tire deflection of approximately simple rule for D A tuning where ~Oa = ogus. As
30%. Indeed, with D A the normalized tire shown by Den Hartog (1956), some aspects of
deflection eventually achieves a m a x i m u m value D A performance can be further improved by
which corresponds to the simple case where choosing t O a = t O u s / ( l + l / P a ), ~ = 0 . 1 6 . The
u = 0 , which implies that in this limiting case results for this " o p t i m a l " D A tuning are:
only the unsprung mass and D A are present. £us = 0 -19121/2, £s = 0.5615 s tj2, t~ = 7.3934 s -3/2,
To achieve the above i m p r o v e m e n t s the
£ a = 0 . 4 2 6 3 s ~a. The corresponding frequency
required D A stroke is £, = 0.3496 s v2 or X~,r,~s =
response between the sprung mass acceleration
0.388 inch = 0.986 cm which implies that the D A
and ground velocity is shown in Fig. 7. It
mass will be within +1.164inch ( ± 3 c m ) from
exhibits characteristics similar to those in Fig. 6
its equilibrium position 99.7% of the time. The
with improved balancing of resonant peaks
two D O F vehicle model with D A is represented
around the unsprung mass resonance. For
as design point A ' in Fig. 4. This time, the
simplicity, in the rest of the study only the
closed-loop eigenvalues are at ei.2 = - 2 . 2 0 3 5 ±
simple tuning (/)a = (/)us and ~ = 0.2 will be used,
/'2.2599, e3, 4 = - 4 . 4 9 0 3 +/'54.559, e5.6= - 11.637 ±
leaving the possibility of further tuning to future
/'71.256, from which it can be seen that with the
developments needed for concrete design
help of D A the wheel-hop damping has been
implementation.
increased from 4.4% in case A to more than
16% in case A'. Moreover, as Fig. 4 shows, if
one compares design A with corresponding D A 2.2. The dynamic absorber as a part of three
design with the same level of suspension stroke DOF optimization problem
indicated by point A", then the rms acceleration In the previous section the dynamic absorber
is reduced by as much as 61%. was added to an already optimized two D O F
Figure 5 shows the results of system response structure. This results in suboptimal but
to a unit ground step input. The original two somewhat simpler control laws since only a
D O F case A is presented by the full lines, and maximum of four control gains are required for
the design A ' is presented by the dotted curves. the case of a quarter-car vehicle model. To fully
From Fig. 5 it can be seen that the m a x i m u m investigate potential improvements with dynamic
and minimum values of various traces follow absorbers, the latter will be first appended to the
similar trends as the corresponding rms values. two D O F model and then the optimization will
Evident in Fig. 5 is a significant i m p r o v e m e n t in be p e r f o r m e d for the resulting three D O F
 Optimal active suspension structures 851

.9

.6

0
-.3

~' - . 9

-1.2
0 .z .4 .6 .8 1 1.z 1.4 1.8 1.8
TIME(SEC)
.5

o :'k.7. - x j v

e~ -1
Z
0
M -1.5

-2 , , i , , i . , , t . , , i , , i , , : , , L , :

.2 .4 .s .8 t 1.2 1.4 1.8 1'.8

TIME(SEC)
40

3O
Z
o 2O

~ o

~ -10

-20

-30
o .2 .4 .8 .8 ~ 1.2 1.4 i .6 1.8 2
Trur.(sr~c)
1.5

i- I.

I AAA,
z
c~
.5
rj

eL
/V v.. A v A- __ - _

m
~, -.5

- 1 . 5
.z~ , ,
.4~ . , , .s~ , ,
.sr , ,
1, , '1.2' '-'1
.4 1.6 1.8
TtaE(SEC)
FIG. 5. G r o u n d s t e p r e s p o n s e f o r o p t i m a l t w o D O F s u s p e n s i o n , d e s i g n c a s e A (full l i n e s ) , a n d s u b o p t i m a l
t w o D O F + D A c o n f i g u r a t i o n , d e s i g n c a s e A ' ( d o t t e d lines).

system. The optimization problem in this case solution of the algebraic Riccati equation
consists of finding the optimal control Us,co that
PA + AtP - PBR-tB~P + Q = 0
will minimize the performance index Pit,
equation (1), subject to the dynamic system with
equations (2).
The solution to the above optimization
problem is given in terms of the following Q=
[o 0] r2
feedback law (Bryson and Ho, 1975; Kwaker- The optimal, normalized rms deflections and
naah and Sivan, 1972): forces are given through
Ueo = -Kx (4)
Xi, rms = (~i) 1/2 (5)
K = R-IBtp "~' = (2~W) t/z
where P is the symmetric positive definite with i = 1 for tire deflection, i = 3 for suspension
852 D. HROVAT

iIli
-20

-30

-40
.1 1 i0 100
OM
i00

50

-50

-I00
I
J
-150

-200
.i i i0 i00
OM
FIG. 6. Frequency response between sprung mass acceleration and ground velocity for the optimal two D O F
suspension, design case A (full lines), and suboptimal two D O F + D A configuration, design case A ' (dotted
lines).
stroke and i = 5 for DA stroke, that results from the Lyapunov equation
(A - BK)X + X(A - BK) t = - r r . (7)
U~,LO = (iC~Kt)~,z ' (6)
/2s,LO = ( 2 x W ) 1 : 2
The results, in terms of the optimal normal-
ized rms acceleration and deflections, are
where X is the n o r m a l i z e d covariance matrix s u m m a r i z e d in Fig. 8. For c o m p a r i s o n , also

20

I0 - -

D -i0 /
-20
y
-30

-40
.t 1 10 O0
ON

<
-50
?
l -
-i00

IF F
1
-150 J

-200
.I 1 10 i00
ON
FIG. 7. Frequency response between sprung mass acceleration and ground velocity for the optimal two D O F
suspension, design case A (full lines), and suboptimal two D O F + D A configuration with p a = 10,
to a = 6%J(1 + 1/Pa), ~a =0.16 (dotted lines).
 Optimal active suspension structures 853

shown in Fig. 8 are the corresponding curves for necessarily equal. The relaxation of this
the two DOF case with DA introduced as in constraint has substantial implications on system
Section 2.1. From Fig. 8 it can be seen that, in ride and handling capabilities. For force Us can
general, only small additional improvements are now be used to optimize vehicle ride, while at
possible through the full three DOF system the same time, the force U,s can contain the
optimization. At the same time the three DOF wheel-hop dynamics for superior handling. Here
case results in more complex control laws, with there exist many possibilities. For example, a
six control gains instead of four required by the portion of U,s can be used to effectively reduce
simpler two DOF + DA case. For this reason the the unsprung mass inertia and/or increase the
latter will mostly be used in the following wheel-hop damping, and thus improve both the
sections. vehicle ride and handling.
In practice, however, the generation of an
2.3 The influence of dynamic absorber mass independent force U,s as in Fig. l l a without
and unsprung mass on system performance some form of inertial support is difficult to
In the previous sections it was assumed that achieve. An example of such actuators would be
the ratio of unsprung-to-DA mass p~ = 10. To the usage of the canted vanes as is done for
investigate the influence of this ratio on system structural control on high performance aircrafts.
performance, the optimization process was At least at present this does not appear practical
repeated for the case p~ = 5. The results for both for automotive applications. As shown below,
cases, pa = 5 and p~ = 10 are shown in Fig. 9. the ideal configuration of Fig. l l a can serve as a
From Figs 9a, b and c it can be seen that, while reference of the best possible performance. A
the rms acceleration, suspension and tire more practical implementation of force U.s is
deflection do not change significantly between shown in Fig. 11b where an additional inertial
the two cases, the maximum rms DA strokes mass has been added to the quarter-car vehicle
are about 25% smaller when Pa = 5. Thus the model. As is the case with dynamic absorbers,
main effect of increasing DA mass is in lowering the added mass is typically much smaller than
the overall DA length, which is one of the the unsprung mass. In practice, the additional
critical packaging requirements. absorber mass (m~) would include the unsprung
To investigate the influence of unsprung mass actuator Aus mass itself.
on system performance with the DA, it was The mathematical model of three DOF
assumed that the nominal sprung-to-unsprung vehicle with active sprung and unsprung
mass ratio Pn = 10 has been increased to p = 20, suspensions shown in Fig. 11b is described by
while keeping the tire stiffness and vehicle (2) with eo~ = ¢~ = 0, and
sprung mass consant. Under these conditions,
the unsprung mass natural frequency increases 0 00 _o a] (8)
from the nominal value f,~,. = 10Hz to f,~= p 0 -1 0
f,,,.p/p, = 14.14 I-Iz. The optimization results for The optimization problem now consists of
the two cases are displayed in Fig. 10. From Fig. minimizing
10a, it can be seen that for a given suspension
2 2 2
stroke some acceleration improvements are P I 2 = [Link] s "t- rlXus,rms + r2Xs,rm s
possible by reducing the unsprung mass. On the 2 2
-F r3Uu$,rms + r4xa,rms (9)
other hand, as Figs 10b and c demonstrate, for a
given acceleration, significant reductions in tire where
deflection and DA strokes are possible with a xa=x5 as shown in Fig. l l b . (10)
smaller unsprung mass. It can thus be concluded
that both vehicle handling and DA design/ The last two terms in (9) limit excessive power
packaging requirements can be improved by consumption and stroke of the unsprung
reducing the unsprung mass. actuator,
Illustrative Example 2. The optimization results
3. A C T I V E M E A N S O F U N S P R U N G M A S S with two actuators, for the special case
VIBRATION CONTROL rl = / ' 2 ----r4 ---- 1 0 0 , r 3 = 10 -3, are s u m m a r i z e d in
In the context of the two DOF model, the column 3 of Table 1. Columns 1 and 2 contain
introduction of the dynamic absorber can be results for the optimal 2 DOF case without and
viewed as a special case of a suspension structure with DA, respectively. Column 1 corresponds to
shown in Fig. l l a . In addition, the "only" the design case A from Fig. 4, and column 2
difference between the system of Fig. l l a and corresponds to the design case A" from Fig. 4
the more conventional one of Fig, la is that the where . f s = 0 . 6 S 1/2 and r~=10 -4. Additional
actuator forces Us and Uus in Fig. l l a are not parameters can be found in Table 1.
854 D. HROVAT

+.-+"+" o
o

..,. , +.+~...+~,-. . . . . . . . ~_
• , ...... "+;: !~.'.~,,' . . . . . . . . . ,n : ": ..................

i ....:,+ ~:,<o
.... ~I' ~' ...,..~. . . . :. . . . . . . . . -.-.:... :.. . uJ

: +++:: ++ :+ + + ., o v

i~,J~:"

+
-'T,i
;,..---+.+..-...-.+ ~ ,- + . . . . . . . . . . . . . .
+
...........................

. . . . . . . . . . . . . . . . ++. . . . . . . . . + . . . . . .
;....................

. ............
,, ......

+. . . . . . .
+. . . . . . . . . . . .

:...: .....
bJ
N
--
+-!
..... . ,y
i / .................. ,..o=
- ~:.,~-:.:

O0
.......... -+................................... :..-:- :.............

0
-
_ _. ~
.
z:

~o . ~
o
+.~
~cq
. . . . ~, ~= T
. ~ ............................................... :. . . . . . . . . . . : - " : ~ (,.., I

i
. . . . . . .

< II
.... .................. ~ ,-,, ..- ~x
+,-
Z
i • : i . . : . zx + ~
. . . . ...................

: ::. !., . ? ': ;: . • ~ 0 ~ +++~

i:+ ~ ' ~o,S=::. ; + 0

!i iiii!' : • :!.:.. ! ii ee
@ ,,i+ ~
OlI o
i ~i!i+! : .. . . . : ~ ~ ~ '" <~I
--. e'~

-,;-i-..~;+ ~..+.. -+ ........ ;:-,..: ......... +....... :..... ; -. o ~ ~" W

N
++ ; ! :+::: ::: : " L~J ~
~T-"~-"~ ........ : ........ .,+..,.... ............ : : . . . . . . . . . . . . N ~ II / _1
•:::~..:..:. ~ : .................... ~......... + . ~ ~ ~- < ~+ +..+
e~
0
+,.., e,
..: ........ ~ ~ z
~=
8 o - -~ ~ ~,~

I I H I I I I I
igillll~llltlllllllIlllllll
IlIIlIlIlllllllllIllllllllll
I I l l l l l l l l l l l U
i?i. liIi/ i/il 2~

-0!
11111111k|lilHlllnillllllllilllllllIll ,~
IMIII~IIIIIIIIIIIIIIIIIIIIIIIIII
_ ~: : : : +-.: _---: +~--_ : : : : : : : : : : : : : : : : : : : : : : : : : :
oo ,,<. o

ii iiiiiii;!!iiii iiiiiiiii iiiiii='-i ? :..:: :

IIIl!!!!I~!l!i~I!~_lllll!!!Ilill!!!I Z
I I I I t l I l | I I I iIiiilIl~, IiiiiilII I iiiiiIl~l I
IiIIfllllIIIlllllliI ~llllll, ":" ~IIilllllllll ~lIl~ W ~o
!lllllllllllllllllllIl~h~lllllllll!!lilll 0.

IIIIIIIIIIIIIIIIIIIIlllll!l|illllllllll

iiiiilllil.~mllllliiiiililliiiilliiiiiIi I n-
IlIIIllll~lIIIIIllllIflllllllIIIIlllllII o
IIIl!llillllllllllllllllllllllllllllllll -~

[Z~.33S] n NOI.,LVId3730'OV SINai 0 3 Z I 7 V I A I M O N

 Optimal active suspension structures 855

JX
ms I I_ ms ms ,, ] "-"Jt X •

~Us
@ ,o o11 'u' 5: XO
l Uus Itu "y'-/,o, ,L
_ _ mus l --~,

k1=XUS

• (o) (b) (c)

FIG. I1. (a) Two D O F vehicle model with active suspension capable of producing no necessarily equal
forces Us and U,=, (b) three D O F vehicle model with active sprung mass suspension As and unsprung mass
suspension A,,~, (c) three D O F vehicles model with an active DA.

Common to all columns in Table 1 is the Further improvements with the active D A are
normalized rms suspension stroke of J?, = 0.6 s ~/2. possible at the expense of increasing the
Comparing columns 2 and 3, it can be seen that unsprung actuator force t o 12us=7.6s -3/2 as
for the same level of suspension stroke and tire shown in column 5, Table 1. This time the
deflection, the active unsprung actuator results normalized rms sprung mass acceleration is only
in 15.6% improvement in rms sprung mass a s = 2 . 7 3 s -3/2, which is 34.7% lower than the
acceleration, while at the same time lowering the value with passive DA. It is of interest to note
unsprung actuator stroke requirements by 24% that this value is slightly lower than the
However, the added unsprung actuator does corresponding acceleration, ~= = 2.86 s -3/2, of the
require a normalized rms force t i ~ = 17.1 s -3/2, limiting 1 D O F case at ~, = 0.606 s u2 shown in
leading to an rms acceleration of 4.7 g acting on Fig. 5a.
absorber mass m a on medium quality roads with At this point, it is appropriate to ask what is
A = 1.6 x 10 -5 ft = 4.9 x 10 -6 m at speeds V = the best possible quarter-car structure that
80 ft s -~ = 88.5 km h -1. results in an absolute minimum for performance
The unsprung actuator force requirements can index PI~, and suspension configuration of Fig.
be relaxed by using an active dynamic absorber l l a . As shown in the next section, the answer to
structure as in Fig. 11c. This time the this inquiry leads to a singular, optimal control
optimization problem consists of minimizing PI2 problem since in this case there is no penalty on
with r4 = 0, subject to dynamic system equations the magnitude of unsprung actuator force which
(2) and (8). The representative results are leads: to a partially cheap control problem. In
summarized in columns 4 and 5 of Table 1. In this context the cheap, unsprung control
column 4, the suspension stroke, tire deflection primarily serves as a structure optimizer: it can
and sprung mass acceleration are similar to the reduce/eliminate the unsprung mass and
corresponding values of column 3. However, the add/substitute additional masses and com-
required unsprung actuator force is substantially pliances as needed to minimize Pit. The solution
smaller, with a value of 1.0 vs 17.1, due to the to the above optimization problem can be found
beneficial action of passive D A component. in the next section.

T A B L E 1. C O M P A R I S O N O F V A R I O U S RMS Q U A N T I T I E S FOR Q U A R T E R - C A R V E H I C L E M O D E L S WITH ACTIVE SUSPENSION

2 D O F w/o D A 2 DOF + DA 3 D O F (Fig. l i b ) 2 D O F + Active D A . . . (Fig. 11c)

rl = 1100 r2 = 100 rt = 10-~ r2 = 100 rl=rz=r4=lO0 L~=p=p~=L=lO, C,,= o.2
fu, = P = 10, fus = / 9 = P a = 1 0 r3 = 0.001
to a = w.~ ~a = 0.2 f.s = P = Pa = 10 q = r2 = l O0 q = O r2 = 6 0
(Design A, (Design A" r3= 1 r3 = 0.03
Fig. 4) Fig. 4)

,fus [¢/2] 0.2969 0.2319 0.2307 0.2261 0.1710

is [¢/2] 0.6057 0.6057 0.6134 0.61075 0.6056
3~a Is t/2] -- 0.4181 0.3084 0.40977 0.3967
,72 = 12s Is -3/2] 10.925 4.1757 3.5333 3.4165 2.7301
(xT'2-~z) Is -t/2] 18.629 13.691 14.625 13.291 10.250
t2t = au~ Is -3/2] -- -- 17.1173 1.0175 7.5976
(x6"~-'~4) Is -~'z] -- 30.585 19.722 28.655 20.793
856 D. HROVAT

4. BEST P E R F O R M A N C E : P A R T I A L L Y C H E A P are two inherent integrators between ti2 and £3,

OPTIMAL CONTROL so that the scaled control u2 becomes
To establish the best possible performance for
quarter-car vehicle models, we will consider the u2 = e2t~:. (16)
partially cheap optimal control problem (Koko- The corresponding, rescaled states are
tovic, 1984; Saksena et al.; 1984, Sannuti, 1983)
for the 2 DOF system shown in Fig. 11a, where x3=£3 and x4=e£4. (17)
the penalty on unsprung force U~ is small. More With (16) and (17) the transformed problem
precisely, the optimization problem consists of becomes:
minimizing the following performance index
PI3 = u~ . . . . + FlX3,rms
-2 + r2(x ~ -- X3)rm
- 2 s+ -,
r3U~.rm s minimize PI3 = [u~ + q x ~ + r2(xl - x3) 2
(11)
+u221dt (18)
subject to constraints of dynamic system
subject to
equations
21 = x2 - w (12) 21 = x2 (19)
22 = ut (13) 22 = ul (20)
"1~3-----£4 -- W (14) e23 = x4 (21)
L = - to~d3 - a2 (15) F_~4 = - (,02se2X3- u 2. (22)
where Combining equations (21) and (22) into a single
u t = Usm~ (t2= Uusmus tou,= k~~ffm~. second-order equation

To facilitate subsequent analysis, the unsprung E2jC3 = --(-02sg2X3 -- U 2 (23)

control weighting parameter r3 will be expressed it can be seen that the original cheap control t22
as r3 = e 4, where e is a small quantity. Moreover, has been replaced by an equivalent, conven-
only the corresponding deterministic case will be tional control acting on a small "mass" e 2,
considered, which is formally obtained by supported by a soft "spring" of stiffness OJusE
2 2.
evoking the certainty-equivalence principle and The LQR problem (18-22) is in a standard,
setting w = 0 in (12) and (14). singular-perturbation form (Kokotovic, 1984;
It should be noted that the two-control Saksena et al., 1984). Following Kokotovic and
configuration of Fig. l l a is very flexible and Yackel (1972), the above singular perturbation
powerful. For example, the unsprung control problem can be partitioned into
force (./us can be used to alter both the effective
unsprung mass and tire stiffness without affecting
min PI3 = ( x t Q x + u t R u ) dt (24)
the sprung mass acceleration, which is controlled
via the sprung mass force Us. Of particular
interest here is the limiting singular case when subject to
e = 0, which results in the best possible structure,
that gives an absolute minimum for the original
performance index, PI1 = PI3.e=0. Thus in the 22 [A1 ~ A21 x2 ]B1

present context the potentially unlimited force . ~3

. . . . . . . . LA3 ':A4J
. . . . . . . . . . . x3
. . . + i~ u (25)
Uus can be viewed as some sort of a e.r4 x4
"structure-optimizer." It will be shown below
that this optimal structure is yet another One where
DOF system.
First, we transform the cheap control problem U=
[u,] u2
Q=
[Q,
O~ 03
(11-15) to an equivalent singular perturbation
problem using the special scaling (Saberi and and the remaining matrices can be identified
Sannuti, 1987; Sannuti and Saberi, 1987; Sannuti from (18-22).
and Wason, 1985) based on physical reasoning, The optimal control is then given via
The scaling reflects the number of integrators
between the cheap control t~2 and the cor- u : -- R-IBtl(~x (26)
responding performance index variable x3. One with
should note from Fig. l l a that t22 exerts direct
control upon the "fast" variables ~4 and £3. In
the present case with scaling parameter e there =
r.,i, l;t/;a
K:I (27)
 Optimal active suspension structures 857

being the positive definitive solution of an which, in the present case reduces to
appropriately partitioned Riccati equation.
For the limiting case when e = 0 the K2 = - E2 = - QE(A4 - B2B~K3)- 1.
corresponding submatrices Ka, K2, and K3 can In summary, the original fourth-order, singular-
be obtained by solving the following system of pertrubation problem (24-25) for the limiting
Algebraic Riccati Equations, AREs (Kokotovic case e = 0 transforms to essentially two second-
and Yackel, 1972). The 2 x 2 matrix K3 is order subproblems, one "slow" and the other
determined from "fast", which reflect the standard one DOF LQ
structure, with damping ratio of 0.7 and
- K 3 A a - A~K3 + K3B2R-1BtgK3 - Q3 = 0. (28)
appropriately modified state weighting matrices
It should be noted that this equation is Q and Q3, respectively.
independent from Kt and K2, resulting in only Based on the above Riccati matrices Ka, K2,
second-order ARE. The "fast" subsystem ARE and K3, an approximate optimal control
(28) can be solved analytically for the elements (Kokotovic and Yackel, 1972) for the original
k3i of the matrix K3, problem (24-25) is
K 3 = [ k31 k32] (29) u=-R-t((B~K, + B~K~)[;I ]
I-k32 k33-]
with
~ , = v5 (r~ + r~y '4, + (eB]K2 + B~K3)[x]]} (34)
k32 = V~l +/'2,
k33 = "v5 (rl + ~)"4.
which in the present case reduces to
The submatrix K~ is the solution of
U 1 = - k l 2 X 1 - kl3X 2 (35)
- K t -i~ - -~ktKl + Kll3R-q3tK~ - 0 = 0 (30)
r2
where u2 -- ---k32 xl +k32x3+k33x4. (36)
A =A1 + EtA3 + B1R -I B2E2
t t + EtB1R-1B~E[ This control approximates the optimal one with
= Bt + ElB2 increased accuracy as e---, 0.
The control 42 for the original optimization
= -E2A3- A 3t E 2t - E2B2R-tB~E~+ Q1 problem (11-15) can now be reconstructed
E~ = (B~R-tB~K3 -A2)(A4-B2R-~B~K3) -t through the backsubstitution (16-17) so that

E2=(A~K3 + Q2)(A4-B2R-1B~K3) -1. 1 r2 k32 k33 (37)

a 2 = -'~ U2 = --E2k3----~2X1 "}" E---~.~3-}"T X , .
In the present case
The corresponding linear momentum equation
E1 = 0 for the unsprung mass then becomes
E2 = Q2(A4- B2B~K3)-l
E21~3 2 2
------- E (.OusX3 --
( r2 xt_Fk3223+~.k33~4).
k32
A = A1
(31)
= B1 (38)

Q= -E2B2B[E[ + Q1 =
I rlrtr2
+r2
01 Now as e ~ 0 , the effective unsprung "mass" e 2,
"stiffness" e2co~, and "damping" ek33 approach
0 0 zero, leading to
Since the above "slow" subproblem (30) is also r2 r2
only second-order, it can be solved analytically n2 xl + k32"~'3= - ~ ( x l
-k3 --f3)
for the elements ku of matrix K~, i.e.,
rl
+ ~ 2 3 =0 (39)
K 1=
rkkl2
k,1 k13J
k,21 (32)
or,
where only kt2 and k13 are needed for -r3 r2
(40)
subsequent control analysis, x1 -- ,f3 rt
Equation (39) can be interpreted to imply
k12= ~/ rlr2 k l 3 = W ~ . ~ / . r l - - r2 "
equilibrium of two spring-like forces: one, with
v rt + r2 v r t + r2
stiffness
F i n a l l y g 2 is determined from rt
k p = ~ (41/
K2 = KIE1 - E2 (33)
858 D. HROVAT

is acting across the primary rattlespace, £3, and

the other with stiffness

r2 (42)
ks = V'r~ + r2

is acting across the secondary rattlespace,

(x~-£3). An interpretation of equation (40) is
that, for a given total rattlespace xt, the optimal
ratio (40) minimizes the 1"1,r2-weighted sum of
squares of primary and secondary rattlespaces. mus = 0
The total rattlespace x~ is determined through kI
the "slow" dynamics, resulting from the P

application of sprung mass control force u~,

which, according to equation (35), is independ-
ent of £3 and £4. From a standpoint of sprung FIG. 12. Singular optimal incremental suspension structure.
mass dynamics then, the optimal u~ leads to a
structure consisting of a skyhook damping kl3
and an effective spring of stiffness k~z acting
follows. The sprung mass control Us contains the
across xl.
"slow" mode in accordance with an appropri-
Combining the above facts, it can be seen that
ately modified, "standard" one DOF optimal
identical optimal performance will result for any
control law with damping ratio of 0.7. The cheap
"fast" control t22 with primary and secondary
and fast control Uus accomplishes two tasks. It
stiffnesses kp and ks, distributed according to
eliminates the unsprung mass, and adjusts the
equation (40). In particular, the two can be
primary and secondary incremental stiffnesses
chosen such that their combined serial stiffness
so that the unsprung motion can instantaneously
across x~ equals k~2 which is the stiffness from
adapt to the best possible configuration for a
the sprung mass control u~. This is achieved by
given set of performance index weights, r~ and
rescaling the original limiting stiffnesses kp and
r2. Note also from equation (40) that there is
ks, equations (41) and (42), so that
now a direct relation between the "tuning"
k'p = skp and k£ = sks. (43) parameters r~ and r2, and tire deflection and
suspension rattlespace.
The scaling parameter s is then determined by
Because of the presence of two independent
requiring that
control forces, Us and Uus, the new one DOF
k'pk; optimal structure of Fig. 12 differs substantially
= kl2 (44) from the previously obtained one DOF con-
k'p + k~
figuration of Fig. lb, since the latter retains the
from where same primary suspension stiffness. As such, the
rl+r2 new one DOF optimal structure serves as a mark
s = Vrir2 (45) of best possible performance that can be
obtained with two DOF, quarter-car vehicle
Thus with models. For the illustrative Example 2, with
/9 = 10, fus= 10Hz, r t =22.3, rz= 8.4, this
, ~/-~.r_~,
r2 , . r/~-+ r2 limiting case results in normalized rms sprung
kp = rl and ks = r2 (46)
_ ~ rlr 2 mass acceleration of only 1.17s -3/2 when
the resulting optimal suspension structure for the £~s = 0.23 s 1/2 and £ s = 0.6s ~/2. Thus, this ex-
singular control case e = 0 can be represented as ample illustrates that substantial further
a special one D O F system shown in Fig. 12. It ride/handling improvements are possible with
should be observed that this is an "incremental" high-performance, active unsprung actuators.
optimal structure, where, for example, the While in some cases this superior performance
incremental primary suspension stiffness k'p is may not be practical due to large energy
seen at unsprung mass location but not requirements, it serves as an absolute reference
necessarily at the tire/road interface. This is the against which any practical implementation can
best possible quarter-car configuration, leading be compared.
to the smallest value of the performance index
Pit, for a given set of weights r~ and r2. 5. CONCLUSIONS
With hindsight, the simple one DOF optimal This paper presents a detailed study of
structure of Fig. 12 can be rationalized as possible ride and handling improvements with
 Optimal active suspension structures 859

active suspensions and structural (design) performance within quarter-car vehicle

changes for the case of linear, quarter-car models.
vehicle models. The structural changes include In a sense, the present study completes a more
the introduction of passive dynamic absorbers than two decades old research cycle based on
for unsprung mass vibration control, which is quarter-car vehicle models. The early investiga-
further enhanced with active means. The tions used a one D O F model chiefly for its
following conclusions can be drawn from the simplicity, convenience and transparency of
above work: dominant trends. The introduction of wheel
1. For some road/speed conditions the perfor- dynamics led to two D O F models and
mance of an active suspension can be corresponding reduction in ride/handling bene-
substantially improved by appending a fits achieved from active suspensions. Combining
passive dynamic absorber to the unsprung the two sets of results pointed to the necessary
mass. In particular, up to 83% reduction in structural changes, notably the introduction of
rms sprung mass acceleration was obtained unsprung mass actuators. These recovered the
without the absorber (under the same or lost benefits and led to optimal structures and
reduced suspension and tire deflections). thus the best possible performance. This
However, the improved performance comes completed the research cycle and made it
at the expense of increased hardware possible to again work with simple, but now
complexity and design/packaging require- appropriately modified, one DOF models.
ments.
2. With the help of dynamic absorbers, the Acknowledgement--The author wishes to thank Ian Craw-
ford from Ford of Europe, Dunton, U.K. for providing
performance of the two DOF, quarter-car information on Citroen 2 CV dynamic absorbers and
system approaches that of an optimal one associated hardware.
DOF system. This confirms an earlier
conjecture (Hrovat, 1985) based on the REFERENCES
author's work with absorbers for single Bender, E. K. (1967). Optimization of the random vibration
characteristics of vehicle suspensions using random process
optimization point rt = 5 0 0 r 2 --- 100. Thus, the theory. Sc. D. Thesis, MIT, Cambridge, MA.
two D O F problem can be treated approxim- Bender, E. K. (1968). Optimum linear preview control with
ately as an appropriately modified one DOF application to vehicle suspension. ASME J. Basic Engng,
90, 213-221.
problem, which greatly simplifies control Bryson, A. E. Jr, and Y. C. Ho (1975). Applied Optimal
design, particularly for higher DOF Control. Wiley, New York.
problems. Chalasani, R. M. (1986). Ride performance potential of
active suspension systems---Part I. ASME Monograph
3. The dynamic absorber (DA) performance can AMD Vol. 80, DSC, Vol 2.
be further improved by reducing the un- Crosby, M. and D. C. Karnopp (1973). The active
sprung mass and/or increasing the DA mass. damper--a new concept for shock and vibration control.
43rd Shock and Vibration Bulletin, Part H.
In both cases the main effect is the reduction Den Hartog, J. P. (1956). Mechanical Vibrations.
of tire deflection and D A stroke, for a given McGraw-Hill, New York.
rms sprung mass acceleration. Guenther, C. R. (1973). Application of Stochastic Optimal
Control Theory to high-speed vehicle suspension synthesis.
4. The passive DA performance can be im- Ph.D. Thesis, University of California, Los Angeles.
proved by active and semi-active means. For Hrovat, D. (1982). A class of active LQG optimal actuators.
example, at high speeds on medium quality Automatica, 18, 117-119.
Hrovat, D. (198aa). Performance trade-offs for an
roads, an active absorber can reduce rms LQG-optimal suspension. Handout for Internal Presenta-
sprung mass acceleration an additional 35% tion, Ford Motor Company, Dearborn, MI.
with respect to the passive DA. If this is Hrovat, D. (1984b). Notes on dynamic absorbers for 2 DOF
vehicle models. Ford Motor Company, Dearborn, MI.
compared with conventional passive suspen- Hrovat, D. (1985). A proposal for control theory based
sions, then this translates to a ten-fold design of advanced suspensions. Ford Motor Company,
reduction in rms sprung-mass acceleration, Dearborn, MI.
Hrovat, D. (1987). Influence of unsprung weight on vehicle
under the same or even smaller suspension ride quality. Ford Motor Company Research Report,
and tire deflections. Dearborn, MI; also J. Sound and Vibration, 124,
5. The best possible ride and/or handling 497-516.
Hrovat, D., P. Barak and M. J. Rabins (1983). Semi-active
improvements for the quarter-car vehicle versus passive or active tuned mass dampers for structural
models result in the limiting case of two control of buildings. ASCE J. Engng Mech. Division, 109,
active actuators shown in Fig. l l a , where the 691-705.
Hrovat, D. and M. Hubbard (1981). Optimum vehicle
unsprung actuator possesses unlimited force suspensions minimizing rms rattlespace, sprung-mass
generation capability. Through this limit- acceleration and jerk. ASME J. Dynam. Syst. Meas.
process, the original two D O F system Control, 103, 228-236.
Hrovat, D. and M. Hubbard (1987). A comparison between
reduces to a special one D O F structure which jerk optimal and acceleration optimal vibration isolation.
establishes the mark of the best possible J. Sound and Vibration, 112, 201-210.
860 D. HROVAT

Hrovat, D. and D. L. Margolis (1981). An experimental prediction of passenger riding comfort from acceleratton
comparison between semi-active and passive suspensions data. ASME J. Dynam. Syst., Meas. Control, 100, 34-41~
for air-cushion vehicles. Int. J. Vehicle Des., 2, 308-321. Thompson, A. G. (1970). Design of active suspensions.
Hullender, D. A., D. N. Wormley and H. H. Richardson Proc. Inst. Mech. Engrs. 185, 553-563.
(1972). Active control of vehicle air cushion suspensions. Thompson, A. G. (1984). Optimal and sub-optimal linear
ASME J. Dynam. Syst. Meas. Control, 94, 41-49. active suspensions for road vehicles. Vehicle Syst. Dynam.
Integrated Systems, Inc. (1984, 1986). Matrixx User's Guide. 13, 61-72.
Palo Alto, CA.
Karnopp, D. C. (1986). Theoretical limitations in active
vehicle suspensions. Vehicle Syst. Dynam., 15, 41-54. APPENDIX A: MAIN ASSUMPTIONS
Karnopp, D. C. and A. K. Trikha (1969). Comparative study The following list includes main assumptions used in the
of optimization techniques for shock and vibration present paper.
isolation. A S M E J. Engng Industry, 91, 1128-1132. 1. Road roughness is described by a white-noise velocity
Kokotovic, P. V. (1984). Applications of singular perturba- Gaussian process. Numerous tests (Sevin and Pilkey,
tion techniques to control problems. SIAM Rev., 26, 1971) confirm that this is a good first-order approximation
501-550. of many typical road surfaces.
Kokotovic, P. V. and R. A. Yackel (1972). Singular 2. As is the standard (implied) case in the literature, all
perturbation of linear regulators: Basic theorems. 1EEE relevant variables are assumed detrended from low
Trans. Aut. Control. AC-17, 29-37. frequency components that may result from, for example,
Kwakernaak, H. and R. Sivan (1972). Linear Optimal (large) hills and valleys.
Control Systems. Reinhart and Winston, New York. 3. The simple models used in this work were derived under
Metz, D. and J. Maddock (1986). Optimal ride height and the following assumptions:
pitch control for championship race cars. Automatica, 22, • Tire damping was neglected
509-520. • Suspension dry friction was neglected (proper design
Morman, K. N., Jr and F. Giannopoulos (1982). Recent measures should be taken to insure low friction)
advances in the analytical and computational aspects of • Finite actuator bandwidth effects were neglected
modeling active and passive vehicle suspensions. In • Roll and pitch modes were neglected
Kamal, M. M. and J. A. Wolf Jr (Eds), Computational • Engine/mounts dynamics was neglected
Methods in Ground Transportation Vehicles, pp. 75-115. • Interaction and direct effects on vehicle handling
ASME Publication AMD 50, New York. dynamics (lateral, yaw modes) were neglected
Roley, D. G. (1975). Tractor cab suspension modeling. Ph. • Nonlinear effects of constitutive relations for tire and
D. Thesis, University of California, Davis, CA. suspension elements dynamics were neglected.
Saberi, A. and P. Sannuti (1987). Cheap and singular 4. All states are assumed to be available for control. In
controls for linear quadratic regulators. IEEE Trans. Aut. practice some states will have to be reconstructed from
Control, AC-32, 208-219. available measurements (e.g. sprung mass velocity may
Saksena, V. R., J. O'Reilly and P. V. Kokotovic (1984). be reconstructed from acceleration with appropriate
Singular perturbations and time-scale methods in control filtering). Alternatively, it may be possible to omit certain
theory: Survey 1976-1983. Automatica, 20, 273-293. less influential states from closed-loop implementation.
Sannuti, P. (1983). Direct singular perturbation analysis of This is an important area for future R and D work.
high-gain and cheap control problems. Automatica, 19, 5. In this study, the rms acceleration was used as a measure
41-5l. of passenger comfort. As shown by Smith et al. (1978)
Sannuti, P. and A. Saberi (1987). A special coordinate basis the rms acceleration is strongly correlated with passenger
of multivariable linear systems, finite and infinite zero discomfort. Therefore, for the purpose of the present
structure, squaring-down and decoupling. Technical optimization study which uses a simple model, the rms
Report, Rutgers Univ., Piscataway, NJ. acceleration is the most representative, single comfort
Sannuti, P. and H. Wason (1985). Multiple time-scale parameter. It is straightforward to extend the present
decomposition in cheap control problems-singular control. study toward more complex measures of passenger
lEEE Trans. Aut. Control. AC-30, 633-644. comfort such as the frequency-weighted acceleration
Sevin, E. and W. D. Pilkey (1971). Optimum Shock and spectrum specified by ISO standard (Guenther, 1973).
Vibration Isolation. Shock and Vibration Monograph 6, Further extensions (to parallel higher DOF models) may
Naval Research Lab., Washington, D.C. include more comprehensive ride comfort measures such
Smith, C. C., D. Y. McGehee and A. J. Healey (1978). The as the one developed by NASA.