Académique Documents
Professionnel Documents
Culture Documents
Abstract— In this paper, a full-vehicle active suspension outside. Due to the nonlinear nature of tyres, the lateral
system is designed to simultaneously improve vehicle ride normal load transfer on the front or rear axle reduces the
comfort and steady-state handling performance. First, a linear lateral force generated on its axle. Thus more roll moment
suspension model of a vehicle and a nonlinear handling model
are described. Next, the link between the suspension model distribution on the front axle will cause the vehicle understeer
and vehicle steady-state handling characteristics is analysed. and more on the rear will cause oversteer. Since the rear
Then, an H-infinity controller for the suspension is designed to suspension of a vehicle is generally designed to be stiffer
achieve integrated ride-comfort and handling control. Finally, than the front one for reduced pitch vibration [5, 6], roughly
the controller is verified by computer simulations. speaking the vehicle unfortunately tends to be oversteered.
I. I NTRODUCTION An active suspension provides an opportunity to change roll
moment distribution on front and rear axles actively and
Although active suspension control has been widely stud- hence the steady-state handling of the vehicle.
ied for decades, most of the research focussed on vehicle ride [1] analysed the nonlinear relationship between lateral and
comfort, with only a few papers [1–3] and some references normal force of a tyre and designed a nonlinear control law
cited therein studying how an active suspension can improve of roll moment distribution. [2] decoupled a full vehicle
vehicle handling. model into four subsystems, one of which was related to
Ride comfort of a vehicle, also known as vibration isola- vehicle handling. A nonlinear controller for the suspension
tion ability, is judged by the level of acceleration, which was then designed in [3] to improve yaw rate. The problem
vehicle passengers are exposed to. Since the manner in of handling control by active suspensions was well studied
which vibration affects comfort is dependent on the vibration in these papers. However, improvement in suspension ride
frequency, frequency weighting functions are used for a fair comfort was not considered and it remains unsure how
judgement. Although vehicle ride comfort is somehow a controllers for ride comfort can work with those nonlinear
subjective issue, it is quantified by the energy of the weighted ones for handling. In this paper, a nonlinear tyre model is
vehicle body acceleration [4]. utilised to transform the nonlinear handling control objective
The most common measure of vehicle steady-state han- into a linear one, which is later incorporated into the H∞
dling performance is the understeer gradient [5], by which suspension control framework for improved ride and steady-
vehicles are categorised into three types: neutral steer, un- state handling.
dersteer and oversteer. In the neutral steer case, the lateral The paper is organised as follows. Section II describes
acceleration at the gravity centre of the vehicle will yield models of vehicle suspension and handling systems and
an identical increase in slip angle at both front and rear Section III derives a linear relationship between the sus-
wheels. In the understeer case, the lateral acceleration will pension model and the steady-state handling performance.
cause more front-wheel slip. Oversteer is the opposite of The relationship is then utilised for controller design in
understeer. An oversteer vehicle could lose its directional Section IV, and the controller performance is evaluated in
stability at the critical speed. Neutral steer is an ideal steering SectionV by computer simulations. Conclusion is finally
condition. However, vehicles are generally designed to be made in Section VI.
understeered for safety since vehicle understeer gradient
varies due to transient manoeuvres. II. M ODELLING OF F ULL V EHICLE WITH ACTIVE
This paper studied the influence of roll moment on vehicle S USPENSIONS
steady-state handling. The roll moment caused by vehicle A. Active Suspension Model
cornering will transfer vehicle weight from its inside to its A schematic diagram of a full-vehicle model with an active
suspension system [7] is shown in Fig. 1.
An extended version of this paper has been submitted to the International
Journal of Vehicle Design. Each quarter of the active suspension consists of a spring,
J. Wang is with the Department of Automation, Tsinghua University, a damping valve and a force generator connected in parallel.
Beijing, 100084, P R China jwang@ieee.org The force generator is regulated by a controller to improve
D. A. Wilson is with the School of Electronic & Electri-
cal Engineering, The University of Leeds, Leeds LS2 9JT, UK vehicle ride and handling, while the spring and damper are
D.A.Wilson@leeds.ac.uk employed to suppress high frequency vibration above the
W. Xu is with the Department of Automation, Tsinghua University, bandwidth of the force generator.
Beijing, 100084, P R China xuwl@mail.tsinghua.edu.cn
D. A. Crolla is with the School of Mechanical Engineering, The Univer- The symbols of the vehicle model is listed in Table I and
sity of Leeds, Leeds LS2 9JT, UK D.A.Crolla@leeds.ac.uk the differential equations are given as follows [7]:
Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on July 22, 2009 at 06:29 from IEEE Xplore. Restrictions apply.
z
2) Wheel (unsprung mass) movements:
mu1 z̈u1 = ks1 z + cs1 ż − lf ks1 θ − lf cs1 θ̇ − wks1 φ
− wcs1 φ̇ − (ks1 + ku )zu1 − cs1 żu1 + ku zr1 − F1
y
φ
x
zu3
θ mu2 z̈u2 = ks2 z + cs2 ż − lf ks2 θ − lf cs2 θ̇ + wks2 φ
zu4
zr3 + wcs2 φ̇ − (ks2 + ku )zu2 − cs2 żu2 + ku zr2 − F2
zu1 Suspension zr4
zu2
Wheel
mu3 z̈u3 = ks3 z + cs3 ż + lr ks3 θ + lr cs1 θ̇ − wks3 φ
zr1
Tyre − wcs3 φ̇ − (ks3 + ku )zu3 − cs3 żu3 + ku zr3 − F3
zr2
1983
Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on July 22, 2009 at 06:29 from IEEE Xplore. Restrictions apply.
i.e. and
lr mt vx2 lf mt vx2 N∆f ε
Fyf = , Fyr = (2) = (7)
lr lr N∆r (1 − ε)
In this paper, we follow [1, 5] and use a second-order Then the normal loads on the wheels are
polynomial, also known as an empirical formula, to describe 1 1
N1 = N10 + N∆f , N2 = N20 − N∆f
the nonlinear relationship between tyre lateral force and 2 2
normal load. The lateral force generated on the front and 1 1
N3 = N30 + N∆r , N4 = N40 − N∆r
rear axles are respectively 2 2
2 2
Fyf = C1 (N10 + N20 ) + C2 (N10 + N20 ) αf
The lateral force generated on the front and the rear axles
1
= C1 Nf 0 + C2 Nf20 αf (3) is respectively
2
1 2
2 2
Fyf = Cf 0 + C2 N∆f αf (8)
Fyr = C1 (N30 + N40 ) + C2 (N30 + N40 ) αr 2
1 2 1 2
= C1 Nr0 + C2 Nr0 αr (4) Fyr = Cr0 + C2 N∆r αr (9)
2 2
where αf and αr are respectively front- and rear-wheel slip From (2), (8) and (9), we have
angles, and C1 and C2 satisfying C1 −C2 > 0 are v2
empirical coefficients decided from tyre experimental data. αf − αr = Kus · x (10)
gr
Define
1 where the understeer gradient of the vehicle is defined as
Cf 0 C1 Nf 0 + C2 Nf20
2 lr lf mt g
1 Kus = − (11)
Cr0 C1 Nr0 + C2 Nr0 2 Cf 0 + 12 C2 N∆f
2 Cr0 + 12 C2 N∆r
2 l
2
Note that Cf 0 and Cr0 are the well-known cornering stiffness Since C1 −C2 > 0 and Nf 0 > N∆f > 0, we have [5]
2
of the front and the rear tyres when there is no lateral load 1 1 C2 N∆f
transfer in the vehicle. The equations (8) and (9) can be ≈ 1−
Cf 0 + 12 C2 N∆f
2 Cf 0 2Cf 0
rewritten as Fyf = Cf 0 αf and Fyr = Cr0 αr .
From (2), (3) and (4), we have Similarly we have
2
lr lf mt g vx2 1 1 C2 N∆r
αf − αr = − · (5) ≈ 1−
Cf 0 Cr0 l gr Cr0 + 12 C2 N∆r
2 Cr0 2Cr0
By the definition given in [5, p.202] and [6, p.287], the Then
2
understeer gradient of the vehicle is lr C2 N∆f 2
lf C2 N∆r mt g
Kus ≈ Kus0 − 2 − 2 (12)
lr lf mt g 2Cf 0 2Cr0 l
Kus0 = − (6)
Cf 0 Cr0 l From (12), we obtain that Kus ≈ Kus0 if
when there is no lateral normal-load transfer. For clarity, N∆f lf Cf 0
Kus0 is also called in this paper the static understeer = √ (13)
N∆r lr Cr0
gradient.
Now we conclude from (7) and (13) that Kus ≈ Kus0 if
III. L INK BETWEEN ACTIVE S USPENSION AND
ε lf Cf 0
H ANDLING = √ (14)
1−ε lr Cr0
When the vehicle is turning with a radius r at a constant
In this paper, the tyre is modelled as a spring and the roll
speed vx , the roll moment due to the inertial centrifugal force
m v2 h moment generated by the front and the rear pairs of tyres is
is Mr = t rx r , where hr is the height of the vehicle roll respectively
centre. The roll moment causes the vehicle load transfer from
its inside to its outside, and the lateral load transfer produces ku (zru1 − zru2 )w = −εMr
roll moment to react against Mr . ku (zru3 − zru4 )w = −(1 − ε)Mr
Suppose the roll moment distributed on the front and the Hence we have
rear axles is εMr and (1 − ε)Mr respectively, where ε ∈ zru1 − zru2 ε
(0, 1). The load transfers on the front and the rear axles then = (15)
zru3 − zru4 1−ε
become
From (14) and (15), we obtain the following equation
εMr εms vx2 hr
N∆f = = lr Cr0 (zru1 − zru2 ) = lf Cf 0 (zru3 − zru4 ) (16)
w wr
(1 − ε)Mr (1 − ε)ms vx2 hr which lays the foundation for integrated active suspension
N∆r = =
w wr control.
1984
Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on July 22, 2009 at 06:29 from IEEE Xplore. Restrictions apply.
IV. I NTEGRATED C ONTROLLER D ESIGN C. Scaling and Weight Selection
In this section, an H∞ controller for the active suspension Before designing an H∞ controller for the active suspen-
is designed to achieve improved ride comfort and handling sion, we scale the state-space model given in (17) and tune
performance simultaneously. weighting functions for the regulated outputs.
Scaling is important in practical applications as it makes
A. Multiple Control Goals
model analysis and controller design much simpler. Scalings
An active suspension system is designed to effectively are chosen to reflect the allowable magnitude of each distur-
isolate the vibration caused by road irregularities while bance signal and the allowable deviation of each regulated
constraining the suspension to move within a limited space output. In this paper, scaling matrices for w and z are
and guaranteeing good road-tyre contact. It can also improve respectively tuned as follows:
handling performance, as shown in [1, 3]. As the understeer
gradient is the most important measure of the steady-state Sw = diag(0.0014, 0.0014, 0.0014, 0.0014, 500)
handling performance, the active suspension is employed Sz = diag(1, 3, 1, 200, 0.0015, 0.0015, 0.0015, 0.0015)
here to keep the vehicle working in slightly understeer and
nearly neutral steer condition when the vehicle is steered at Based on the ISO 2631-1 [4], human sensitivity to vi-
high speed. Thus the control objectives of this design are bration dependent on frequency. Humans are more sensitive
ride and handling performance, and suspension travel and to vertical acceleration at 4∼ 8 Hz and rotational one at
road holding ability are evaluated after the design. 1∼2 Hz. In H∞ control, this frequency-dependent design
specification can be easily satisfied by choosing weighting
B. Output Selection
functions for z̈, θ̈ and φ̈ as follows:
The ride comfort is quantified by the energy of weighted
vehicle body acceleration z̈, θ̈ and φ̈. From (16), we know s2 + 314.2s + 987
Wz̈ =
that Kus ≈ Kus0 if s2 + 43.98s + 987
s2 + 50.27s + 25.27
kf (zru1 − zru2 ) − kr (zru3 − zru4 ) = 0 Wθ̈ = 2 , Wφ̈ = Wθ̈
√ s + 7.037s + 25.27
√ lf C f 0
lr C
√r0 √
where kf = √ and kr = √ . As it is our main concern to minimise the steady-state value
lr Cr0 + lf C f 0 lr Cr0 + lf C f 0
Hence we also need to minimise handling performance index of khpi ,
s + 50
khpi kf (zru1 − zru2 ) − kr (zru3 − zru4 ) Wkhpi =
s+1
so that the understeer characteristics of the vehicle are little is chosen as its weighting function. Since we do not have
affected by steering manoeuvres and external disturbances. In any particular frequency requirements on actuator force,
addition, we need to constrain force generated by actuators. we simply choose Wu = 1 as its weighting function. In
Hence we choose the vector of regulated output variables as summary, the weighting function matrix for the regulated
T
z = z̈ θ̈ φ̈ khpi F1 F2 F3 F4 . output z is
The most common sensors for active suspension control
are accelerometers, gyrometers and suspension defection sen- Wz = diag(Wz̈ , Wθ̈ , Wφ̈ , Wkhpi , Wu , Wu , Wu , Wu )
sors [8]. In this paper, we only choose the accelerometers and D. H∞ Control Design
gyrometers for a simple implementation. To effectively keep
the understeer gradient constant, we also need to measure H∞ control aims to design a controller Kias so that the
the deflection of tyres, which used to be extremely difficult. closed-loop system Gcl achieves the minimal H∞ norm.
Fortunately, recent technology developments [9] have made The closed-loop system is computed by linear operators, i.e.
it possible to measure tyre pressure in production vehicles. Gcl = Wz · Sz · Fl (Ggel , Kias ) · Sw , where Fl stands for a
As a matter of fact, direct tyre-pressure measurements have lower linear fractional transformation. The active suspension
been mandated in the North America for driving safety. In control problem can be formulated as the following standard
this paper, the handling-performance index khpi needs to be one: to compute a controller Kias so that
computed from tyre deflection. In summary, we choose the
T min sup σ̄ Gcl (jω)
vector of measuring signals as y = z̈ θ̇ φ̇ khpi . Kias ω
Combined with the state equation given in (1), a state-
space model G for the integrated active suspension control where σ̄(·) stands for the maximal singular value. Such a
is formulated as follows: problem can be solved by a linear-matrix-inequality method
⎧ given in [10] or the programme ‘hinflmi.m’ provided by LMI
⎨ ẋ = Ax + Bw w + Bu u Control Toolbox for Use with M ATLAB.
z = Cz x + Dzw w + Dzu u (17)
⎩ For performance comparison, an ordinary H∞ control Kas
y = Cy x + Dyw w + Dyu u
is also designed for the active suspension. Kas is computed
where the matrices Cz , Cy , Dzw , Dzu , Dyw and Dyu are in the same procedures as for Kias except that the handling
obtained in a straightforward manner. performance index is not considered as a control objective.
1985
Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on July 22, 2009 at 06:29 from IEEE Xplore. Restrictions apply.
V. S IMULATION R ESULTS AND A NALYSES of IAS is much less than that of AS and PS in a wide
Computer simulations are carried out in this section to low-frequency range. Hence the IAS system is expected to
evaluate the ride and steady-state handling performance of achieve better steady-state handling performance than the
the vehicle. The active suspension with the integrated H∞ AS and PS systems. This conclusion can be more directly
controller Kias (denoted by IAS) is compared to the passive reached in Fig. 5, which shows the time response of Khpi
suspension (denoted by PS) and the active suspension with to a unit step signal Mr . The IAS system has lower steady-
the ordinary H∞ controller Kas (denoted by AS). Simulation state value, less overshoot and shorter settling time, while
parameters are listed in Table II. the AS and PS systems have similar steady-state handling
performance.
TABLE II
S IMULATION PARAMETERS −120
Magnitude (dB)
ms 1500 kg ks1 , ks2 35000 N/m −160
mui 59 kg ks3 , ks4 38000 N/m
Ix 460 kg m2 cs1 , cs2 1000 Ns/m −180
Iy 2160 kg m2 cs3 , cs4 1100 Ns/m −200
IAS
AS
Iz 2500 kg m2 ku 190000 N/m PS
lf 1.4 m δ 5 degree −220 −2 −1 0 1 2
lr 1.7 m C1 13.098 1/rad 10 10 10 10 10
Frequency (Hz)
hr 0.45 m C2 -0.001045 1/(rad N)
Fig. 4. Magnitude of the frequency response of handling performance
Fig. 3 shows the frequency responses of the vehicle index Khpi to external roll moment disturbance Mr
vertical, pitch angular and roll angular acceleration to the
front-left road disturbance. It is clearly observed that the
acceleration of active suspension systems IAS and AS is 0.01
IAS
considerably lower than that of the passive one PS in AS
the frequency range where humans are more sensitive to 0 PS
(rad)
−0.01
50 50
−0.02
0 0.5 1 1.5 2 2.5 3
Magnitude (dB)
Magnitude (dB)
Time (s)
1986
Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on July 22, 2009 at 06:29 from IEEE Xplore. Restrictions apply.
−100 −100
VI. C ONCLUSIONS
−120 −120
This paper designs an H∞ controller for the active suspen-
Magnitude (dB)
Magnitude (dB)
−140 −140 sion system to simultaneously improve vehicle ride comfort
−160 −160 and steady-state handling performance.
IAS The nonlinear effect of roll moment distribution on vehicle
−180 AS −180
PS handling is transformed into a linear control objective, which
−200 −1 −200 −1
10 10
0 1
10
2
10 10 10
0 1
10 10
2 can be easily incorporated into general suspension control
Frequency (Hz) Frequency (Hz) frameworks and achieved by linear control strategies.
(a) (b)
−100 −100 The integrated controller can preserve vehicle handling
−120 −120 characteristics, regardless of lateral acceleration changes due
Magnitude (dB)
Magnitude (dB)
−140 −140
to driving manoeuvres such as forward speed changes. Vehi-
cles in the market are generally designed to be understeer and
−160 −160
sometimes are equipped with anti-roll bars for a sufficient
−180 −180 safety margin although neutral steer is in accord with human
−200 −1 0 1 2
−200 −1 0 1 2
intuition of cornering. The research presented in this paper
10 10 10 10 10 10 10 10
Frequency (Hz) Frequency (Hz) provides the opportunity to design a slightly understeered
(c) (d) and nearly neutral-steered vehicle. The control strategy can
Fig. 6. Magnitude of the frequency response of vehicle suspension improve handling without sacrificing vehicle cornering sta-
deflection to external roll moment disturbance Mr : (a) zus1 ; (b) zus2 ; bility, and it can in fact reinforce vehicle steering safety as
(c) zus3 ; (d) zus4 b shown in the simulations.
In the future, a more generalised case will be considered.
For example, it will be unnecessary for tyres to be identical
and actuator dynamics will be incorporated into the design.
of the frequency. Although the decay rate of PS is larger
than those of IAS and AS, it has little effect on vehicle ACKNOWLEDGEMENTS
tyre deflection dynamics. Hence the IAS, the AS and the PS This research was sponsored in part by Scientific Research
systems have similar road holding ability. Foundation for Returned Overseas Scholars by State Edu-
cation Ministry, Postdoctoral Science Foundation of China
under Grant No. 2004035058, Natural Science Foundation
−100 −100
of China under Grant No. 50275045 and No. 50411130486,
−120 −120 and Royal Society of UK under Grant No. 16558.
Magnitude (dB)
Magnitude (dB)
−140 −140 The authors are grateful to anonymous reviewers for their
−160 −160 insightful comments and valuable suggestions.
IAS
−180 AS −180 R EFERENCES
PS
−200 −1 0 1 2
−200 −1 0 1 2 [1] D. E. Williams and W. M. Haddad, “Nonlinear control of roll moment
10 10 10 10 10 10 10 10 distribution to influence vehicle yam characteristics,” IEEE Transac-
Frequency (Hz) Frequency (Hz)
(a) (b) tions on Control Systems Technology, vol. 3, no. 1, pp. 110–116, March
−100 −100 1995.
[2] M. Lakehal-Ayat, S. Diop, and E. Fenaux, “Development of a
−120 −120
full active suspension system,” in Proceedings of 15th Triennial
Magnitude (dB)
Magnitude (dB)
1987
Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on July 22, 2009 at 06:29 from IEEE Xplore. Restrictions apply.