You are on page 1of 6

Aleksander Hac*

Department of Mectianical Engineering,


State University of New Yorl<,
Stony Brook, NY 11794

Decentralized Control of Active


Vehicle Suspensions
With Preview
In this paper, decentralized control of active vehicle suspensions with preview of
road irregularities is considered using a two-degree-of-freedom vehicle model. It is
shown that by taking advantage of the separation between the eigenvalues of the
slow subsystem representing the body mode, and the fast subsystem corresponding
to the wheel mode, the design of the preview controller can be decoupled. Since
decentralized preview controllers are synthesized independently for two single-degree-of-freedom systems, analytical solutions are obtained. The results of the analysis
and simulations demonstrate that the performance of the system with the proposed
controller is comparable to that of the optimal preview controller based on a fully
interconnected system.

Introduction

The dynamics of ground vehicles is characterized by the


presence of two dominant oscillatory modes that correspond to
the dynamics of the vehicle body and the wheels. In the design
of active suspensions, adequate control of each of these modes
must be achieved. To provide a stable, comfortable ride, the
vehicle body must be isolated from the vibrations arising from
road irregularities, while vibrations of the wheels must be suppressed to maintain firm, uninterrupted contact between the
wheels and the road for good road holding ability (Hrovat,
1993). The natural frequencies of the body modes are typically
in the range of 0.5-2 Hz, while the wheel hop frequency is in
the neighborhood of 10 Hz. Hence, these modes are well separated and the vehicle suspension can be viewed as a system
composed of slow and fast subsystems. It has been shown by
Chow and Kokotovic (1976) that for linear deterministic systems possessing this time-scale separation property, the design
of feedback controllers for the slow and fast subsystems can be
completely separated. As a result, two control inputs can be
obtained independently for the slow and fast subsystems and
combined into a composite control law. Furthermore, the performance of the closed loop system thus obtained is close to that
of the system with a controller derived by considering a fully
interconnected model. Salman et al. (1990) have shown that
this approach can be applied to the optimal design of active
suspensions when the problem is treated as a deterministic problem, i.e., when the road disturbance is ignored. This methodology simplifies the design procedure and results in a controller
that is easier to implement because the requirements for on-line
data processing and storage are reduced. The main goal of this
paper is to extend the decentralized approach to the synthesis
of controllers for active suspension systems with finite preview
information about the road disturbance. Preview suspensions,
first proposed by Bender (1968), have been recently receiving
a great deal of interest because of their potential for improving
performance and because of recent advances in proximity-sensor technology which make preview implementable. From an
analytical viewpoint, the difference between the problems considered by Salman et al. (1990) or Chow and Kokotovic (1976)

and that analyzed in this paper is the presence of a (partially


measurable) disturbance, which requires modification of the
procedures developed for deterministic regulators. In the presence of preview information, the optimal control law consists
of a feedback part (which is the same as in the corresponding
deterministic problem without preview) and a feedforward part
which depends on the previewed road disturbance ahead of
vehicle (e.g., Hac, 1992). To achieve full separation in the
design of the controller, both parts of the control law must be
decoupled. In the decentralized approach developed in this paper the problem is reduced to two independent optimal preview
control problems for one-degree-of-freedom systems, both of
which can be solved analytically.
2

Formulation of the Problem


A two-degree-of-freedom (2-DOF) vehicle model, which reflects both the low-frequency and the high-frequency dynamics,
is shown in Fig. 1. The massOTIcorresponds to a quarter of the
vehicle body mass andOT2represents the mass of the wheel
with semi-axle and associated linkages; ^2 represents the tire
stiffness. Absolute displacements of the body and the wheel are
denoted by z\ and Zi, respectively. Variable zo denotes the vertical elevation of the road irregularities. It is assumed that the
ground velocity input, za, can be considered a white noise process with zero mean. The time varying portion of the suspension
force, u, is the control variable. The dynamics of this system
is represented by the following system of equations:
m^Z\ = u; m2Z2 + fe(z2 - Zo) =

(1)

In order to take advantage of the time-scale separation property


between the wheel and the body modes, we represent the model
in a singularly perturbed form (Saksena et al., 1984). Introducing the notation

* Present address: Delphi Chassis Systems, Engineering Technical Center, MS


E-520, P.O. Box 1245, Dayton, OH 45401-1245.
Contributed by the Dynamic Systems and Control Division for publication in
the JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT, AND CONTROL . Manuscript

received by the DSCD September 30, 1993. Associate Technical Editor;


R. Rizzoni.

Fig. 1 2-DOF vehicle model with slow and fast subsystems

478 / Vol. 117, DECEMBER 1995

Transactions of the ASME

Copyright 1995 by ASME


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

Xi = [zi - Z2 ZiV,

X2 = [zi - Zo Z2V, w = zo (2)

the system equations may be expressed as


(3a)
(3b)

Xi = A | | X i + A12X2 + BiM
/LiS2 = A22X2 + B2M + D2W

where
AH

"0 r
.0

'0
0

A,2 =

-1"
,
0

0
,

l/mi

B2 = A*

-l/m2.

A22 = M

0
k2/rn2

D2 = At

-1"
0

Optimal Preview Control


The problem of finding a control input that minimizes the
performance index (8) for the system described by the state
Eq. (5) in the presence of preview information about the disturbance has been considered by Hac (1992). Suppose that the
preview time is tp, that is, w(t + T ) , T e [0, tp] is known
deterministically, and for T > f,, ^ ( f + r ) is assumed to be a
zero-mean white noise process. If (A, B) is a stabilizable pair
then the optimal solution is (Hac, 1992):

(4)

Mo(0 = -R~'B'^{K\

+ r)

(10)

where K is a positive definite solution of the algebraic Riccati


equation
KA + A^K - KB/? 'B'^K + Q = 0

(11)

and /.t is a small positive constant (singular perturbation coeffi- and r(r) satisfies the following linear equation
cient). Salman et al. (1990) recommend selecting the value of
r = - A j ' r - KDw, T{t + tp) = 0
(12)
fj, that is close to the ratio between the natural frequencies of
the slow and fast subsystems, i.e., /x >= 0.1. The system (3) can
where A^ = A - B/?''B^K is the closed-loop system matrix.
also be represented as the state equation:
All eigenvalues of A^ have negative real parts if the pair (A,
Q ' " ) is detectable.
X = Ax + BM + Dw
(5)
The control law (10) consists of a feedback part,
-/?^'B'^Kx(0,
which is the same as in the corresponding probwith
lem without preview, and a feedforward part, /J"'B'^r(/)While the (constant) feedback gain matrix can be found off"A A|2
=
, A
line, integration of equation (12) in real time is necessary to
0
-X2_
A22/M_
obtain the preview term. The term r{t) can be expressed as
0
B,
B
D =
(6)
r(f)
KDw(r + T)dT.
(13)
1
.D2/M.

j^

Jo

The purpose of control is to minimize the variance of the body


acceleration (for good ride comfort) and to keep the variation
in the dynamic load of the tire low, under the constraints on
the suspension stroke and the magnitude of the control force.
A suitable performance index to be minimized is therefore:
J

lim

-f

[z1 +

P l ( Z l - Z2)'

+ Piil + p?,u^]dt (7)

2TJO

where pi, p2> and p^ are the weighting coefficients. Although


the variance of the tire deflection is a more straightforward
measure of road holding ability and is more frequently used
than the variance of the wheel vertical velocity, essentially the
same effect, that is, an increase in the damping of the wheel
mode, can be obtained in both cases. Formally, a mean value
operator should be placed in front of the integral in expression
(7) because all variables under the integral are stochastic processes. Since for stationary and ergodic processes, statistical
(ensemble) averages are the same as time averages obtained
from one sufficiently long sample, the mean value operator is
dropped. Since for our model (Eq. (1)) minimization of the
body acceleration is equivalent to minimization of the control
input, u, the weighting coefficient p^ is assumed to be zero.
The index (7) can be expressed in the matrix form:
J = lim
7'-GO

^r^''^^^

X 4- Ru^)dt

(8)

where
Q,
0

Q
Q2

"Pi

0
Q2.

Qi =

'

"0

C) '

p2_

0"

A simplification of this expression would reduce the demand


for on-line data processing.
4

Decentralized Control
Consider the singularly perturbed form of the system representation given by equation ( 3 ) . The difference between this
system and the one considered by Chow and Kokotovic (1976)
or Salman et al. (1990) is the presence of the disturbance w(t).
Assuming that p ^ 0, the slow subsystem can be described by
Xj = AQXJ -t- BcMj + D Q W

where Ao = A
__11, Bo = _
B .i A12AJ2B2 and Do =
-Ai2A22'D2. For the case of the suspension system considered
here A12AJ2 B2 = 0 and hence Bo = B], while Do = [ - 1 0] '^.
The fast subsystem is given by
pXf

= A22X/ + B2M/ -I- D2W.

R =

1/m?.

T-*w 2,1

(9)

Matrices Q, Qi, and Q2 are positive semidefinite and R is


positive.
Journal of Dynamic Systems, Measurement, and Control

(15)

Physically, Eq. (14) represents the 1-DOF vehicle model without the wheel mass and the fast subsystem represents the dynamics of the wheel. These models are shown in Fig. 1. Following
the approach of Chow and Kokotovic (1976) and Salman et al.
(1990), we extract from the total index of performance (8) two
quadratic indices that correspond to the slow and fast subsystems and optimize each of these systems independently by minimizing the appropriate measures of performance.
Bearing in mind that for the slow subsystem (14) Ao = An
and Bo = Bi, we assume that the triple (An, Bi, Qj'^) is
stabilizable and detectable. Then the problem of minimizing the
performance index
hm

(14)

(xjQiX, + Rul)dt

(16)

Jo

subject to the dynamic constraint (14) has the solution


M,(f) = - / ; - ' B [ ( K , x , + r,)

(17)

where Kj and r, satisfy the following equations


DECEMBER 1995, Vol. 1 1 7 / 4 7 9

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

Table 1

Control gains and modal parameters for various active systems

Controller

Control gains

Damped natural frequencies


[rad/s]

Damping ratios

Optimal soft
Decentralized soft
Optimal firm
Decentralized firm

31.62; 7.756; 17.58; -1.686


31.62; 7.953; 0.0; -1.619
31.62; 7.971; -1.655; -4.192
31.62; 7.953; 0.0; -4.195

4.141; 59.49
4.033; 61.02
4.144; 56.26
4.155; 56.10

0.6753; 0.1417
0.6767; 0.1358
0.6746; 0.3520
0.6745; 0.3529

IJK2

(36)

Inserting (36) into (11) and using notation (6) we obtain the
following set of equations:

Wrfz+e ( - ? f

k2Uj{p2 X[i-C,fiof + jio) sin ujjf t,, + ujjf cos ujjf tp ]


(34)
(u;^ - ujlf) + 2C,fUJfUjj
m\u),ij

K,(A - -B.R-'h^Kl)

and
Aijuj)

K,

K =

Zo(;'w)

+ (A - B , ^ - ' B [ K D %

- K,B|i?-'B[K| - KzBz^-'BjKj + Q, = 0
= [(-w'

+ 47,) +

juj^pl"]

K2(A22 - B2R~'BlK,)

X [(m2(xJ^ + ^2 + mivpi) + jwifiiVp2]/m2

(37a)

+ Ki(Ai2 - B.iJ-'BjKj) = 0 (31b)

K3A22 + A2''2K3 - KsBj/J-'BjKj + Q2 = 0.

-(^/^+ia;^/^)(/^^-jwV2pl"). (35)

(37c)

It is seen by comparing Eqs. (37c) with (23) that K3 = K/.


Furthermore, using only a slightly different model Chow and
Kokotovic (1976) have shown that the solution of Eqs. (37a)
and (31b) is

From these expressions, the frequency domain characteristics


for body acceleration, tire deflection, or suspension deflection
can be easily calculated. These expressions can also be used to
obtain the power spectra, and hence the variances, of the system
response to a stochastic road excitation with a given spectrum.

Ki = K j ,

K2 = K,(-Ai2 + B,R-'BlKf)(\22
5 Relationship Between Decentralized and Optimal
Control

~ B2R'^'BlKf)-'.

(38)

It is easy to verify that (38) satisfies Eqs. (37a, b) since

To provide an indication of the degree of deterioration of


performance brought about by the simplified controller, a comparison between the optimal solution obtained for the fully interconnected system and the decentralized solution is considered.
The following theorem is the main result of this section.
Theorem 1. If the triples (A, B, Q " ' ) , ( A , B,, Q | ' ' ) and
(A22, B2, Q^^) are stabilizable and detectable, then the decentralized control law given by equation (26) approaches the optimal control law defined by Eqs. (10) through (12) as /x - 0.
Proof. It is known that the appropriately partitioned matrix K,
which is a positive-definite solution of equation (11), can be
expanded in a power series in the parameter fu,. Retaining only
the first term in the expansion yields

K,

V2pr - i/p^
_{7x~^p\"ip'2

0
0_

(39)

and hence B f K j = B j K j = 0, which implies that Eq. (37a)


reduces to Eq. (18). Furthermore, the first-order approximation
of the optimal control law (10) becomes
Uo(t) = - / ? " ' B [ ( K , x , + r,)
- R-\(ixB\.2

+ BlKj)x2

+ Bj(r2//x)]

(40)

where [ r [ r2]^ = r. Hence, as /x -+ 0 the feedback part of the


optimal control law (40) becomes the same as that of decentral-

Table 2 Relative performance of various suspensions for a random road input; the optimal active suspension without
preview is taken as 100 percent

Type of suspension
Optimal soft

Decentralized soft

Optimal firm

Decentralized firm

Variances of
Preview
time [s]

Body ace.

Susp. defl.

Tire defl.

Total index

r.m.s. power

0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30

100
99.1
74.2
54.5
95.9
94.9
71.9
54.6
100
94.6
52.8
51.1
100.5
94.6
54.1
51.3

100
101.0
65.4
49.7
103.4
103.3
67.4
50.2
100
109.8
96.8
61.4
99.7
105.4
93.4
60.3

100
91.4
43.2
44.6
99.9
96.2
50.8
53.4
100
72.3
46.6
46.3
100.2
81.0
53.6
53.4

100
97.1
62.3
51.4
100.8
98.8
64.0
53.6
100
88.8
54.1
50.5
100.0
90.5
55.4
51.5

100
95.2
36.8
29.4
100.1
97.3
42.0
32.6
100
85.1
31.5
28.5
100.0
88.7
34.7
29.9

DECEMBER 1995, V-'

117/481

Journal of Dynamic Systems, Measurement, and Control

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

Table 1

Control gains and modal parameters for various active systems

Controller

Control gains

Damped natural frequencies


[rad/s]

Damping ratios

Optimal soft
Decentralized soft
Optimal firm
Decentralized firm

31.62; 7.756; 17.58; -1.686


31.62; 7.953; 0.0; -1.619
31.62; 7.971; -1.655; -4.192
31.62; 7.953; 0.0; -4.195

4.141; 59.49
4.033; 61.02
4.144; 56.26
4.155; 56.10

0.6753; 0.1417
0.6767; 0.1358
0.6746; 0.3520
0.6745; 0.3529

IJK2

(36)

Inserting (36) into (11) and using notation (6) we obtain the
following set of equations:

Wrfz+e ( - ? f

k2Uj{p2 X[i-C,fiof + jio) sin ujjf t,, + ujjf cos ujjf tp ]


(34)
(u;^ - ujlf) + 2C,fUJfUjj
m\u),ij

K,(A - -B.R-'h^Kl)

and
Aijuj)

K,

K =

Zo(;'w)

+ (A - B , ^ - ' B [ K D %

- K,B|i?-'B[K| - KzBz^-'BjKj + Q, = 0
= [(-w'

+ 47,) +

juj^pl"]

K2(A22 - B2R~'BlK,)

X [(m2(xJ^ + ^2 + mivpi) + jwifiiVp2]/m2

(37a)

+ Ki(Ai2 - B.iJ-'BjKj) = 0 (31b)

K3A22 + A2''2K3 - KsBj/J-'BjKj + Q2 = 0.

-(^/^+ia;^/^)(/^^-jwV2pl"). (35)

(37c)

It is seen by comparing Eqs. (37c) with (23) that K3 = K/.


Furthermore, using only a slightly different model Chow and
Kokotovic (1976) have shown that the solution of Eqs. (37a)
and (31b) is

From these expressions, the frequency domain characteristics


for body acceleration, tire deflection, or suspension deflection
can be easily calculated. These expressions can also be used to
obtain the power spectra, and hence the variances, of the system
response to a stochastic road excitation with a given spectrum.

Ki = K j ,

K2 = K,(-Ai2 + B,R-'BlKf)(\22
5 Relationship Between Decentralized and Optimal
Control

~ B2R'^'BlKf)-'.

(38)

It is easy to verify that (38) satisfies Eqs. (37a, b) since

To provide an indication of the degree of deterioration of


performance brought about by the simplified controller, a comparison between the optimal solution obtained for the fully interconnected system and the decentralized solution is considered.
The following theorem is the main result of this section.
Theorem 1. If the triples (A, B, Q " ' ) , ( A , B,, Q | ' ' ) and
(A22, B2, Q^^) are stabilizable and detectable, then the decentralized control law given by equation (26) approaches the optimal control law defined by Eqs. (10) through (12) as /x - 0.
Proof. It is known that the appropriately partitioned matrix K,
which is a positive-definite solution of equation (11), can be
expanded in a power series in the parameter fu,. Retaining only
the first term in the expansion yields

K,

V2pr - i/p^
_{7x~^p\"ip'2

0
0_

(39)

and hence B f K j = B j K j = 0, which implies that Eq. (37a)


reduces to Eq. (18). Furthermore, the first-order approximation
of the optimal control law (10) becomes
Uo(t) = - / ? " ' B [ ( K , x , + r,)
- R-\(ixB\.2

+ BlKj)x2

+ Bj(r2//x)]

(40)

where [ r [ r2]^ = r. Hence, as /x -+ 0 the feedback part of the


optimal control law (40) becomes the same as that of decentral-

Table 2 Relative performance of various suspensions for a random road input; the optimal active suspension without
preview is taken as 100 percent

Type of suspension
Optimal soft

Decentralized soft

Optimal firm

Decentralized firm

Variances of
Preview
time [s]

Body ace.

Susp. defl.

Tire defl.

Total index

r.m.s. power

0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30

100
99.1
74.2
54.5
95.9
94.9
71.9
54.6
100
94.6
52.8
51.1
100.5
94.6
54.1
51.3

100
101.0
65.4
49.7
103.4
103.3
67.4
50.2
100
109.8
96.8
61.4
99.7
105.4
93.4
60.3

100
91.4
43.2
44.6
99.9
96.2
50.8
53.4
100
72.3
46.6
46.3
100.2
81.0
53.6
53.4

100
97.1
62.3
51.4
100.8
98.8
64.0
53.6
100
88.8
54.1
50.5
100.0
90.5
55.4
51.5

100
95.2
36.8
29.4
100.1
97.3
42.0
32.6
100
85.1
31.5
28.5
100.0
88.7
34.7
29.9

DECEMBER 1995, V-'

117/481

Journal of Dynamic Systems, Measurement, and Control

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

Table 3 Relative suspension performances for a bump input; the optimal active suspension without preview corresponds
to 100 percent
Variances of
Type of suspension

Preview time [s]

Body ace.

Susp. defl.

Tire defl.

Total index

r.m.s. power

0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30
0.00
0.02
0.10
0.30

100
96.6
77.8
58.2
96.0
94.3
74.7
58.6
100
87.0
64.4
65.9
100.5
89.9
66.0
67.7

100
91.1
56.2
59.5
102.7
96.7
59.3
60.9
100
88.2
69.2
61.0
99.8
89.3
67.5
59.7

100
82.3
33.5
40.3
99.7
89.1
42.2
51.0
100
53.1
32.9
34.6
100.3
64.2
40.6
43.2

100
91.6
59.7
52.9
100.7
95.3
61.4
55.9
100
77.0
50.5
50.2
100.0
81.0
51.7
51.6

100
89.0
39.8
35.8
99.6
92.7
43.9
39.4
100
72.0
33.2
32.0
99.9
77.7
35.3
34.2

Optimal soft

Decentralized soft

Optimal firm

Decentralized firm

law (26). To demonstrate that the feedforward of slow and fast subsystems. When jj,-> 0, it follows from Eq.
same, one needs to show that ri = r^ and Vilfi = (41b) that Tj = -(Ar22)~'[Ari2ri + K^Daiv] and the slow
0 with r^ and r^ defined by Eqs. (19) and (24), subsystem reduces to
It follows from Eqs. (12), (6), and (36) that
(43)
f, = -AJor, - K0D2W, ri(r + t,) = 0
- A j r i - A?2ir2 - K2D2W r,(? + fp) = 0 (41fl)

ized control
parts are the
tf && jjL ^
respectively.
1*1 =

1^2

r^it + f^) = 0

-Ari2r, Ac22r2 ~ K/D2W

(41/.)

where ra/ytt = r2, A^u and Ac22 are the closed-loop system
matrices for the slow and fast subsystem, respectively, Ac2i =
-B2/f~'B[K, and A2 = A^ - B^R^BlKf. Specifically,
OT|

A,2i =

m2

" 0

"

.V^

^p\'\ , Kn = _o v^_

"0 - 1 "

(42)

The system of linear equations (41) can be considered to consist


1

'

<

where AJo = Aj' - Aj2i(A?'22)"'Aj'i2 and Ko = K2 - AJ'21


(A^22)~'K/. The structure of the system matrices makes A^o =
Ac 11 and KQDJ = K,Do. Hence Eq. (43) becomes identical to
Eq. (19), which implies that ri = r,.. For the fast subsystem
(41b), it can be assumed that the slow variables ri(/) are constant during fast transients (Chow and Kokotovic, 1976), and
due to the terminal condition r i ( ; -I- fp) = 0 it can be assumed
that TI(T) f 0 for T G [t, t + tp] for the fast subsystem.
Therefore the term Ajj2ri in Eq. (41a) can be neglected when

A
'I
a

'A'

2.5 '-

0.0

<V

r-^

"

\l

Z| optimal
z, decentralizedZQ (road input)

>

0.2

0.4

0.0

Zi o p t i m a l
z, d e c e n t r a l i z e d
Zo (road input)

'

-5.0

0.6

0,4

time [sec]

2.5

0.6

Z2 optimal
Zj d e c e n t r a l i z e d
z (road i n p u t )

z, optimal
Z2 d e c e n t r a l i z e d
ZQ (road i n p u t )

2.5

o.

t i m e [sec]

0.0

0.4

0.6

time [sec]
Fig. 2 Response of the vehicle model to a bump input. The vehicle Is
equipped with soft suspension and optimal and decentralized controllers
with preview time tp = 0.1 s. (a) Body acceleration; (/>) vertical displacement of the wheel

482 / Vol. 117, DECEMBER 1995

0,4

0.6

time

[sec]

Fig. 3 Response of the vehicle model to a bump input. The vehicle Is


equipped with firm suspension and optimal and decentralized controllers
with preview time tp = 0.1 s. (a) Body acceleration; (b) vertical displacement of the wheel

Transactions of the ASME

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

fj, -^ Q, yielding a differential equation that is identical to Eq.


(24). Hence, r2 = r^. This completes the proof.

6 Results of Simulations
To compare the performances of the suspension systems with
truly optimal and decentralized controllers, the response of the
vehicle model to two types of road disturbances was simulated.
The first input, used to reveal the transient response characteristic, is a single bump described by
{c[l - cos 207r(f - 0.3)]

for t e [0.3; 0.4]

[^ 0

otherwise

zo(0 = ]

(44)

where 2c = 0.05 m is the height of the bump and t is time in


seconds. This bump excites both the wheel and body modes.
The second type of road input, used to evaluate the steady-state
response, is a stationary random process with spectral density
5(a;) =

(45)
TT w^ H- (av)

where u = 30 m/s is the vehicle velocity and a = 0.15 m~' and


a- = 3 X 10"^ m are the road roughness parameters (a^ denotes
the variance of road irregularities). The values of the vehicle
parameters are mi = 1 kg, m2/mi = 0.1, k2lmi = 360 N/(mkg). Hence all values are referred to unit body mass. Two sets
of weighting constants were used: pi = 1000, p2 = 2.62 and
Pi = 1000, p2 = 17.6, which represent soft and firm suspensions,
respectively. The control gains and the modal parameters of the
systems corresponding to the optimal and decentrahzed controllers are shown in Table 1 for both sets of weights. It is seen
that the differences in the values of the control gains (with the
exception of the third gain) and the modal parameters of the
closed loop systems for both controllers are small, especially
for the suspension in the firm mode. For the firm suspension,
better control of the wheel is achieved primarily by increasing
the damping ratio of the wheel mode resulting from the larger
value of p2, while the natural frequencies remain essentially the
same, since the weight pi that affects the suspension stiffness
is the same in both cases.
In Table 2 the results of simulations of the vehicle response
to the random road input are summarized. The vehicle performances in terms of variances of body acceleration, tire deflection, suspension deflection, the total index, and r.m.s. power are
shown for both controllers and several values of the preview
times. In terms of total index of performance the differences
between the two controllers are typically less than 2 percent.
In Table 3 the results obtained for the bump input are shown.
Again the performances of both systems are very similar, but
the discrepancies between the optimal and decentrahzed controllers are slightly larger than for the random road profile. For
both road inputs, significant improvement of road holding ability requires less preview than do improvements in the measures
of the ride comfort or suspension rattle space, because the tire
deflections are determined by the motion of the fast subsystem.
Consequently, short preview times are more effective in reduc-

Journal of Dynamic Systems, Measurement, and Control

ing the total index of performance for firm suspensions that are
designed with more emphasis on road holding properties.
The body acceleration and the wheel vertical displacement
resulting from the bump input are shown in Figs. 2 and 3 for
the soft and firm suspensions, respectively. The preview time
was t,, = 0.1 s. The soft suspension, with a weakly damped
wheel mode, provides isolation of the body that is superior to
that achieved with firm suspension, but the wheel vibrations
reach higher amplitudes and the vibrations persist for a longer
period of time. For the firm suspension, transient responses are
quickly damped but the peak values of the body acceleration
are quite high. The differences between the suspension systems
with the decentralized and optimal controllers are small but
noticable. In both figures, anticipative actions of active suspensions can be observed before the bump is traversed.

7 Conclusions
In this paper, the problem of preview control of an active
vehicle suspension was considered. It was shown that the decentrahzed control strategy developed earlier for deterministic control problems can be applied to the synthesis of preview controllers that utilize preview information about the road disturbances
ahead of the vehicle. By taking advantage of the separation
between the natural frequencies of the body and the wheels,
preview controllers for the body and the wheel can be designed
independently and then combined into one composite controller.
In this approach, an analytical solution can be obtained, from
which the effects of various parameters of the system and the
weighting constants in the cost function on the system dynamics
can be deduced. Since preview control requires on-Une integration to obtain the feedforward part of the control input, the
reduction of data processing requirements brought about by
decentralized control is of particular importance. The results of
both analysis and simulations show that the proposed controller
achieves performance that is close to that of the optimal controller.

References
Bender, E. K., 1968, "Optimum Linear Preview Control with Application to
Vehicle Suspension," ASME Journal of Basic Engineering, Vol. 90, No, 2, pp.
213-221.
Chow, J. H., and Kokotovic, P. V., 1976, "A Decomposition of Near-Optimum
Regulators with Slow and Fast Modes," IEEE Trans, on Automatic Control, Vol.
AC-21, No. 5, pp. 701-705.
Had, A., 1992, "Optimal Linear Preview Control of Active Vehicle Suspension," Vehicle Systems Dynamics, Vol. 21, No. 3, pp. 167-195.
Hrovat, D., 1993, "Applications of Optimal Control to Advanced Automotive
Suspension Design," ASME JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT,

AND CONTROL, Vol. 115, No. 2B, pp. 328-342.


Karnopp, D. C , and Trikha, A. K., 1969, "Comparative Study of Optimization
Techniques for Shock and Vibration Isolation," ASME Journal of Engineering
for Industry, Vol. 91, No. 4, pp. 1128-1132.
Karnopp, D. C , 1989, "Analytical Results for Optimum Actively Damped
Suspensions Under Random Excitation," ASME Journal of Vibration, Acoustics,
Stre.is, and Reliability in Design, Vol. H I , No. 3, pp. 278-281.
Saksena, V. R., O'Reilly, J., and Kokotovic, P. V., 1984, "Singular Perturbations and Time-Scale Methods in Control Theory: Survey 1976-1983,'' Automatica. Vol. 20, No. 3, pp. 273-293.
Salman, M, A., Lee, A. Y., and Boustany, N. M., 1990, "Reduced Order
Design of Active Suspension Control," ASME JOURNAL OF DYNAMIC SYSTEMS,
MEASUREMENT, AND CONTROL, Vol. 112, No. 4, pp. 604-610.

DECEMBER 1995, Vol. 1 1 7 / 4 8 3

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

You might also like