IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO.

6, NOVEMBER 2013 2417

Adaptive Robust Vibration Control of Full-Car Active

Suspensions With Electrohydraulic Actuators
Weichao Sun, Huijun Gao, and Bin Yao

Abstract— This paper investigates the problem of vibration and therefore they are attracting the attention of both acad-
suppression in vehicular active suspension systems, whose aim emics and engineers. In [12], a backstepping control design
is to stabilize the attitude of the vehicle and improve the riding is presented for the control of a vehicle active suspension
comfort. A full-car model is adopted, and electrohydraulic actua-
tors with highly nonlinear characteristics are considered to form system, whose controlled plant is a seven degrees-of-freedom
the basis of accurate control. In this paper, the H∞ performance (DOF), nonlinear full-vehicle model. The backstepping control
is introduced to realize the disturbance suppression by selecting employed in [12] achieves stability of the system while at
the actuator forces as virtual inputs, and an adaptive robust the same time improving the riding comfort. Also, various
control technology is further used to design controllers which control techniques are used to improve the performance of
help real force inputs track virtual ones. The resulting controllers
are robust against both actuator parametric uncertainties and active suspension systems, such as H∞ control [13], adaptive
uncertain actuator nonlinearities. The stability analysis for the and backstepping control [14]–[20], sliding-mode control [21],
closed-loop system is given within the Lyapunov framework. preview control [22], [23], fuzzy logic and neural network
Finally, a numerical example is given to illustrate the effectiveness control [24]– [26], and so on.
of the proposed control law, where different road conditions are However, in most of the works treating active suspensions,
considered in order to reveal the closed-loop system performance
in detail. the effect of actuator dynamics is neglected and the actuators
are modeled as ideal force generators. Additionally, studies
Index Terms— Active suspension system, adaptive control, that do consider actuator dynamics in suspension design
full-car model, H∞ control, robust control.
are based on the assumption that the actuator dynamics are
known exactly without parametric uncertainties and unmod-
I. I NTRODUCTION eled nonlinearities. Such an assumption results in inaccu-
racies of controller design in actual engineering [27]– [30].
S USPENSION systems are important parts in vehicles
because of their great contribution to the riding comfort
and safety of passengers. Roughly speaking, vehicle suspen-
In active suspension designs, electrohydraulic systems could
be a good alternative as actuators to isolate the vibrations
sions can be grouped into three types: passive, semiactive, and transmitted to the passengers, because they are more powerful
active [1]–[3]. Although active suspension has not been widely and less bulky compared to other actuators. However, the
used for production vehicles because of high energy demand, main drawback of electrohydraulic systems is their highly
it is a trend in suspension development; many researchers are nonlinear behavior, making it more difficult to design the ideal
focusing on this topic to derive the achievable performance of control law. Furthermore, aside from the nonlinear nature of
suspension systems (for related research, see [4]–[11] and the hydraulic dynamics, hydraulic servo systems are also subject
references therein). In active suspensions, actuators are able to to excessive model uncertainties. Many authors have done
both add and dissipate energy from the system, which enables research on uncertain systems [31]– [38]. Vehicle dynamics
the suspension to control the attitude of the vehicle, to reduce make it difficult to track forces, especially in the presence
the effects of road roughness, and, thus, to increase riding of uncertainties. It is worthwhile to mention that, in [16],
comfort and improve road handling. Since the actuators pull the authors proposed a two-loop design approach where the
down or push up together with the suspension motions, the control architecture contained a force-loop controller and a
limitations arising from this should be taken into account. main-loop controller. In main-loop controller design, the linear
A great number of studies show that active suspensions quadratic Gaussian (LQG) approach was used to achieve the
have the potential to improve the riding comfort significantly, performance of active suspensions, whereas in force-loop con-
troller design the adaptive robust controller was proposed to
Manuscript received May 15, 2011; revised December 1, 2012; accepted track the desired force commanded by the main-loop controller
December 14, 2012. Manuscript received in final form December 27,
2012. Date of publication February 5, 2013; date of current version accurately. The proposed method was effected in applications,
October 15, 2013. This work was supported in part by the Self-Planned Task but the stability analysis of the combined system including
(NO.SKLRS201308B) of the State Key Laboratory of Robotics and System two loops was not given. Based on the above discussions,
(HIT), Key Laboratory Opening Funding of Technology of Micro-Spacecraft
(HIT.KLOF.2009099), and the 973 Project (2009CB320600). Recommended for active suspensions with the nonlinear uncertain actuators,
by Associate Editor J. Lu. it is necessary to develop a systematic control scheme such
W. Sun and H. Gao are with the State Key Laboratory of Robotics and that closed-loop stability is guaranteed. To the best of our
System (HIT), Harbin Institute of Technology, Harbin 150080, China (e-mail:
[email protected]; [email protected]). knowledge, few attempts have been made in this direction,
B. Yao is with the School of Mechanical Engineering, Purdue University, especially in theory, which is also the motivation of this paper.
West Lafayette, IN 47907 USA (e-mail: [email protected]). In this paper, the problem of vibration suppression is
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. investigated to improve the riding comfort of active suspension
Digital Object Identifier 10.1109/TCST.2012.2237174 systems, where the full-car active suspension systems with
1063-6536 © 2013 IEEE
2418 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 6, NOVEMBER 2013

electrohydraulic actuators are considered to cover a wider

range of applications. To do so, an adaptive robust control
(ARC)-based H∞ control technology is proposed in a unified
framework to realize disturbance suppression and to simulta-
neously increase the system robustness against actuator uncer-
tainties. Moreover, the suspension movements can be restricted
within their limits as long as the lumped disturbance energy
satisfies a certain condition. In particular, compared with
existing results, the proposed approach is able to guarantee
rigorous closed-loop stability within the Lyapunov framework.
Besides, the advantages from both H∞ technology and ARC
approach are reflected, which are suitable for active suspension
systems with actuator dynamics. Finally, the effectiveness of
the proposed approach is shown by a design example with
different road conditions.
Fig. 1. Model of full-car active suspension system.
The remainder of this paper is organized as follows. The xvi
problem of vibration control for full-car active suspensions
is formulated in Section II. ARC-based H∞ controllers are
synthesized in Section III. A design example illustrating the Qi
P1i
usefulness and advantage of the proposed methodology is
given in Section IV. Section V shows the conclusion and future Ps yi
Qi P2i
works.
Notation: For a matrix P, P T and P −1 denote its transpose
and inverse, respectively; the notation P > 0 (≥ 0) means that
P is real symmetric and positive definite (semidefinite). For a Spool Valve

vector or matrix, {·} j ( j = 1, 2, . . . ,) represents the j th line Hydraulic Cylinder

of the vector or matrix, and ·∞ , · and ·2 denote the Fig. 2. Structure of an electrohydraulic actuator.
∞-norm, Euclidian norm, and 2-norm of the vector. In sym-
metric block matrices or complex matrix expressions, we use
an asterisk (∗) to represent a term that is induced by symmetry, cylinder dynamics can be written as follows [16]:
and diag{. . .} stands for a block-diagonal matrix. ṖLi = −β PLi − α A ẏi + μx vi gi (·)
where PLi = P1i − P2i is the load pressure of the cylinder,
II. P ROBLEM F ORMULATION A is the ram area of thecylinder, α = (4βe )/Vt , β = αCt p ,
μ = αCd ω ρ1 , gi (·) = Ps − sgn(x vi )PLi , βe is the effective
In this paper, a full-car electrohydraulic suspension model is bulk modulus, Vt is the total actuator volume, Ct p is the
considered [12], which has seven DOFs due to the heave, pitch, total leakage coefficient of the piston, ρ is the hydraulic fluid
and roll motions of the sprung mass and the vertical motions density, ω is the spool valve area gradient, Cd is the discharge
of the unsprung masses, as depicted in Fig. 1. M, Ix , and I y coefficient, and Ps is supply pressure. Noting that the forces
stand for the mass of the vehicle body and mass moment of generated by actuators can be described as f i = A PLi , the
inertia for the roll and pitch motions, respectively, and m i , force dynamics are given as
i = 1, 2, 3, 4 are the unsprung masses of front left, front
right, rear left, and rear right, respectively. Fsi and Fdi denote f˙i = −β f i − α A2  ẏi + Aμx vi gi (·). (1)
the forces produced by the spring and damper, respectively, Here, according to [7], the open-loop natural frequency
and kt i is the stiffness of the tire. For the vehicle body, z, θ ,
 the actuation systems can be estimated as ωn =
of
and ϕ represent the heave, pitch, and roll motions, respectively. ((4βe )/Vt )(A2 /M). By selecting the parameters of the elec-
yi is the unsprung mass displacement, and yoi is the road trohydraulic systems as βe = 700 Mpa (200–800 Mpa),
input to the related wheel. The electrohydraulic actuators are Vt = 1.2 × 10−3 m3 , we have f = ωn /2π = 42.1 Hz.
placed parallel to the suspension springs and dampers, and As we know, the sensitive frequency of the active suspensions
their output forces are denoted by f i . In Fig. 1, a, b, c, and d is usually 4–8 Hz, which implies that the electrohydraulic
show the distances of the suspensions to the center of mass systems can be used as the actuator for the vehicle suspension
of the vehicle body. V is the velocity of the vehicle in the control without worrying about the dynamics of the hydraulic
x-direction. actuation system.
The dynamics of the electrohydraulic actuators are taken Considering the hydraulic fluid’s bulk modulus βe
into consideration as shown in Fig. 2. In this paper, the valve as the
dynamics are neglected and the servo valve opening x vi is  uncertain parameter
T
and defining actuator force
F = f 1 f 2 f 3 f 4 , we can obtain the following dynamics:
directly related to the control input by a known static mapping.
Neglecting the effect of leakage flows in the cylinder, the Ḟ = θ1 (−Ct p F − A2 ż 2 + η) + d0 (2)
SUN et al.: CONTROL OF ACTIVE SUSPENSIONS WITH ELECTROHYDRAULIC ACTUATORS 2419

where θ1 = α = (4βe )/Vt is an unknown parameter, a state. Here, by denoting e1 (t) = F(t) − Fa (t) as the error
d0 denotes the general uncertainties
 arising T from unmod- between F(t) and Fa (t), we obtain
eled √ dynamics, η = η1 η2 η3 η4 , and ηi = ⎧
⎪
⎨ẋ(t) = A p x(t) + B p Fa (t) + B p e1 (t) + (t)
ACd ω (1/ρ)x vi gi (·). θ1 and d0 are assumed to be bounded
˜ p z 1 (t) = C1 x(t) (4)
and their bounds are known as 0 < θ1m < θ1 < θ1M and ⎪
⎩
d0  < d M . z 2 respects the suspension spaces, which will be z 2 (t) = C2 x(t)
defined later.
where (t) = Bw w(t) is the lumped disturbances. The
The dynamic equations of motion for the full vehicle model
following lemma gives the solution to the design of the
with electrohydraulic actuators, based on the assumption that
expected control function Fa (t) = H x(t).
the pitch and roll angles are small, are as follows:
⎧ Lemma 1: Consider the car body system in (4) with
⎪
⎪ z̈ = − M1 4
i=1 (Fdi + Fsi ) + M
1 4
i=1 f i (t)
e1 (t) = 0. If there exist positive scalars γ and σ , and
⎪
⎪ symmetric matrices H, P > 0 satisfying
⎪
⎪θ̈ = − 1
(a 2
(F + F − f (t))
⎪
⎪ Iy i=1 di si i
⎡ T ⎤
⎪
⎪ 4 A p P + H T B Tp P + P A p + P B p H P C1T
⎪
⎪ −b (F
i=3 di + Fsi − f i (t)))
⎪ ⎣
⎨
ϕ̈ = − I1x (d (Fdi + Fsi − f i (t)) ∗ −γ 2 I 0 ⎦ < 0 (5)
I
⎪ i=2,4 ∗ ∗ −I
⎪
⎪ √
⎪
⎪ −c (Fdi + Fsi − fi (t))) −I σ {C2 }i
⎪
⎪ < 0 (6)
⎪
⎪
i=1,3
∗ −yi2max P
⎪
⎪ ÿi = m1 {Fdi + Fsi − kt i (yi − yoi ) − f i (t)}
⎪
⎪
⎩˙ i
and the lumped disturbance energy satisfying 2 ≤ σ/γ 2 ,
fi = θ1 (−Ct p f i − A2  ẏi + ηi ) + di
then a virtual input Fa (t) exists in the form of Fa (t) = H x(t)
where Fsi = ki yi , Fdi = bi  ẏi , ki , bi are the stiffness and (H is a constant matrix), such that:
damping coefficients, respectively, and yi , i = 1, . . . , 4,
1) under zero initial conditions, z 1 2 < γ 2 ;
stand for the suspension deflections.
2) suspension spaces travel within their allowed ranges.
Proof: This proof is omitted, and one can arrive at this
III. ARC-BASED H∞ C ONTROL L AW S YNTHESIS conclusion based on [39].
Step 1: Design a desired virtual force Fa (t), such that if Step 2: Synthesize an actual control law for u = x v , so that
F(t) = Fa (t), then the L 2 gain from disturbances to the output F(t) tracks the desired control function Fa (t) in the presence
z 1 is less than a certain given value γ . of unknown parameter θ1 and general uncertainties d0 .
Defining the state variable x = [z θ ϕ y1 y2 y3 y4 ẏ θ̇ ϕ̇ Differentiating the error dynamics for e1 = F − Fa , we
ẏ1 ẏ2 ẏ3 ẏ4 ]T, output variables z 1 = [z θ ϕ]T, z 2 = [y1 y2 have
y3 y4]T, and disturbance w = [yo1 yo2 yo3 yo4]T, the state-
space form of car body system p can be expressed as ė1 = Ḟ − Ḟa = θ1 (−Ct p F − A2 ż 2 + η) + d0 − Ḟa . (7)
⎧ Design an ARC control law as η = ηa + ηs , where ηa is
⎪
⎨ẋ(t) = A p x(t) + Bw w(t) + B p F(t)
used to achieve an improved adaptive model compensation
p z 1 (t) = C1 x(t) (3)
⎪
⎩ and ηs functions as the robust term. The adaptive part ηa is
z 2 (t) = C2 x(t) designed as
where ηa = Ct p F + A2 C2 (A p x + B p F)
07×7 I7 07×4 07×4 1
Ap = , Bw = , Bp = + (H (A p x + B p F) − 2B Tp Px − ε1 e1 ) (8)
E −1 G 1 E −1 G 2 E −1 F1 E −1 F2 θ̂1
 
E = diag M Iz Ix m 1 m 2 m 3 m 4 , B = diag b1 b2 b3 b4
  where θ̂1 is a bounded projected estimate of θ1 and ε1 is used
Kt = diag kt 1 kt 2 kt 3 kt 4 , K = diag k1 k2 k3 k4
    for tuning the controller. The adaptation law is chosen as the
C1 = I3 03×11 , G 1 = −F2 K T − 07×3 F1 projection type with the following form:
   
G2 = −F2 BT, T = L −I4 , C2 = T 04×7 . ⎧
⎡ ⎤T ⎪
⎨0, if θ̂1 = θ1M and r τ > 0
1 1 1 1 ˙θ̂ = Pr oj (r τ ) =
T 0, if θ̂1 = θ1m and r τ < 0 (9)
L = ⎣ a a −b −b ⎦ , F1 = 03×4 , F2 = L .
1 θ̂1 ⎪
⎩
Kt −I4 r τ, otherwise
−c d −c d
The output z 1 in (3) is defined to describe vertical, pitch, where r > 0 is a tunable gain and τ = e1T (−Ct p F −
and roll motions of the vehicle, and our goal is to stabilize A2 C2 (A p x + B p F) + η).
the output z 1 . Suspension spaces are expressed as output z 2 , The robust control function ηs is now chosen to satisfy the
each component of which should not exceed the allowable following conditions:
maximum yi max , i.e., |yi | ≤ yi max .
condition 1: e1T [θ̂1 ηs − θ̃1 (−Ct p F − A2 C2
Our objective is to synthesize a desired control function
Fa (t) for the virtual control F(t) so that the expected perfor- (A p x + B p F) + η) + d̄] ≤ ε2 ;
mances can be matched. However, F(t) is not a real input, but condition 2: e1T θ̂1 ηs ≤ 0 (10)
2420 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 6, NOVEMBER 2013

where θ̃1 = θ̂1 − θ1 , d̄ = d0 − (θ1 A2 C2 + H ) and ε2 is a Proof of B: Choose a positive function as V = x T Px +

design parameter which can be arbitrarily small. Let h be any (1/2)e1T e1 + (1/2r)θ̃12 , and we have
smooth function satisfying
V̇ (t) = x T (A Tp P + H T B Tp P + P A p + P B p H )x
 2
1  
h ≥ (θ1M − θ1m )2 −Ct p F − A2 C2 (A p x + B p F) + ηa  +2e T B T Px + e T ė + r −1 θ̃ θ̂˙
ε2a 1 p 1 1 1 1

+
1 2
d̄ < −γ −2 T
x P Px Px + ė1 ) + r −1 θ̃1 θ̂˙1
+ e1T (2B Tp
ε2b M
  < −γ −2 x T P Px − ε1 e1T e1 + θ̃1 (r −1 θ̂˙1 − τ ).
and d̄ M a positive number such that d̄  ≤ d̄ M , ε2a , and ε2b
Noticing the property of the projection mapping Pr ojθ̂1 (r τ ) :
are adjustable small positive numbers satisfying ε2a + ε2b =
ε2 . Then, based on [36], the robust control part ηs can be θ̃1 (r −1 Pr ojθ̂1 (r τ ) − τ ) ≤ 0, we have V̇ < −γ −2 x T P Px −
chosen as ε1 e1T e1 < 0, which implies that the state x and the tracking
1 error e1 asymptotically converge to zero.
ηs = − he1 . (11)
4θ1m
Finally, the original control signal u i = x vi can then be IV. D ESIGN E XAMPLE
found from the definition of η This section presents simulation results from the applica-
ηi tion of the control algorithms developed in Section 3. For
ui =   , i = 1, 2, 3, 4. (12) comparison, the backstepping control technology introduced
ACd ω 1
ρ Ps − sgn(ηi )PLi in [30] is also employed to demonstrate the effectiveness of
our proposed method.
Theorem 1: With the ARC law (12) and the projection type The model parameters borrowed from [12] are given as
adaptive law (9), the following results hold: follows: M = 1200 kg, m 1 = m 2 = 25 kg, m 3 = m 4 = 45 kg,
1) in general (i.e., the system is subjected to parametric Ix = 550 kg m2 , I y = 1848 kg m2 , k1 = k2 = 15 000 N/m,
uncertainties, unmodeled uncertainties and external dis- k3 = k4 = 17 000 N/m, bi = 1500 N s/m, kt i = 250 000 N/m,
turbances), all signals in system I are bounded; a = 1.2 m, b = 1.4 m, c = 0.7 m, d = 0.8 m, yi max = 0.1 m,
2) if, after a finite time, the system is subjected to paramet- V = 20 m/s, μ = 2.32×108 N/(m5/2 kg1/2), Ct p = 4×10−13,
ric uncertainties only, then the state x and the tracking A = 0.006 m2 , Ps = 5 × 106 Pa, and α = 2.3 × 1012 N/m5 .
error e1 asymptotically converge to zero. Here we assume that d̄ = (−1)round(t), which implies that we
Proof of Part A: Choose a positive definite function as can choose d̄ M = 1. The controller parameters are given as
V = x T Px + (1/2)e1T e1 , and we have r = 10, σ = 1, ε1 = 100, h = 1000, θ1M = 2.68 × 1012 , and
θ1m = 0.67 × 1012 .
V̇ = x T (A Tp P + H T B Tp P + P A p + P B p H )x + 2x T P
+2e1T B Tp Px + e1T ė1 . (13) A. Bump Road Inputs
Substituting the ARC law (12) into (13) results in Bump road inputs can be generally assumed as shocks, and
according to [12], the corresponding ground displacement is
V̇ (t) ≤ x T (A Tp P + H T B Tp P + P A p + P B p H )x + 2x T P given as

−ε1 e1T e1 + ε2 . (14) h 2 [1−cos(8πt )]
, 1 ≤ t ≤ 1.25
yo = 2 (18)
From (5), we have A Tp P + H T B Tp P + P A p + P B p H < 0, otherwise
−γ −2 P P − C1T C1 , and noting 2x T P ≤ λ11 x T P Px + where h 2 is the height of the bump road input. To validate
λ1 T , where λ1 is a positive value, we have the vibration depression effect of the designed controller, it
is assumed that the bump road input has different magnitudes
V̇ (t) < −(γ −2 − λ−1 T T T
1 )x P Px − ε1 e1 e1 + λ1   + ε2 . (15) for the left and right wheels (3 cm for the right and 4 cm for
the left side) and road conditions for the front and rear wheels
Choosing λ1 so that γ −2 − λ−1 > 0, and defining 0 =
1 are the same but with a time delay of (a + b)/V.
min{λmin (P)(γ −2 −λ−1
1 ), 2ε 1 }, where λmin (P) is the minimal
Fig. 3 shows comparisons among passive systems (dotted
eigenvalue, we have
lines), active suspensions with backstepping-based H∞ con-
V̇ (t) < −0 V + ε3 (16) trollers (dash-dotted lines), and active suspensions with ARC-
based H∞ controllers (solid lines) for the time responses
where ε3 = ε2 + λ1 ∞ , with T  ≤ ∞ . Eq. (16) of the heave motion z, pitch motion θ , and roll motion
shows that the Lyapunov function is bounded by φ with bump road inputs in (18). It is seen from these
ε3 figures that the magnitudes for the heave, pitch, and roll
V (t) < V (0)e−0 t − (1 − e−0 t ) (17) motions are significantly decreased for active suspensions with
0
ARC-based H∞ controllers. The reasons why ARC-based
which implies that the states and force tracking error (e1 ) are H∞ controllers have an advantage over backstepping-based
bounded. H∞ controllers are twofold: adaptive ability (to deal with the
SUN et al.: CONTROL OF ACTIVE SUSPENSIONS WITH ELECTROHYDRAULIC ACTUATORS 2421

Displacement response of the vertical motion −3

x 10
0.04 10
ARC−based Hinf controller M=1200kg
0.02 Backstepping−based Hinf controller 5 M=1100kg
Passive systems M=1300kg
0 0
−0.02 −5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (sec)
Displacement response of the pitch motion 0.01
0.01

0 0

−0.01 −0.01
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (sec) −3
x 10
−3
Displacement response of the roll motion 1
x 10
5

0
0

−1
−5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (sec)
Fig. 5. Responses of the displacements with different body masses
Fig. 3. Displacement responses for heave, pitch, and roll motions. (M = 1200, 1100, and 1300 kg).
Response of the suspension space Δ y Response of the suspension space Δ y2 TABLE I
1
0.04 0.04
RMS VALUES OF B ODY D ISPLACEMENTS
0.02 0.02

0 0 Passive Active Suspensions Active Suspensions

Suspension (B.C.)/Improvements (ARC)/Improvements
−0.02 −0.02
yrms 0.0037 0.0031 (↓ 45.95%) 0.0020 (↓ 56.52%)
4.6163 × 10−4 10−4 10−4
−0.04 −0.04
0 2 4 6 0 2 4 6 θrms 2.2482 × 2.0949 ×
(↓ 51.30%) (↓ 54.62%)
Response of the suspension space Δ y3 Response of the suspension space Δ y4
0.04 0.04 φrms 4.1853 × 10−4 1.7111 × 10−4 1.0783 × 10−4
(↓ 59.12%) (↓ 74.24%)
0.02 0.02

0 0

−0.02 −0.02

−0.04 −0.04
ISO2631 standards, to generate the random road profile. For
0 2 4
Time (sec)
6 0 2 4
Time (sec)
6
the random road disturbances, it is more reasonable to evaluate
the effectiveness of controller design by using the root-mean-
Fig. 4. Responses of the suspension spaces. square (RMS) values. It is well known that the RMS value
of the vehicle body displacement or acceleration is strictly
related to the riding comfort of the passengers, and is often
uncertain parameter θ1 ) and robust ability (to deal with the
used to quantify the amount of displacement or acceleration
uncertain nonlinearity d̄).
transmitted to the vehicle body.
In the active suspension control, the limitations of the
Table I gives the RMS values of the heave, pitch, and roll
suspension spaces should be taken into account, which means
displacements and accelerations. From this table, we can see
the suspension working space must be preserved. It can been
that, compared with backstepping-based H∞ controllers, our
observed from Fig. 4 that the suspension space constraints
proposed controllers can achieve a greater degree of reduction
of the designed systems have been guaranteed, which are all
for the displacements and accelerations under the random road
below the limits yi max = 0.1 m.
inputs, which further verifies the feasibility of the proposed
Additionally, the uncertainties and unmodeled dynamics of
method.
the suspension components themselves are worthy of consid-
eration. In this paper, the simulations for the case of body
uncertainties are performed (Fig. 5), from which we can
see that our designed controllers still have the capacity to V. C ONCLUSION
suppress the disturbances in spite of the existence of parameter
uncertainties in suspension components themselves. This paper presented the theory and methodology of
ARC-based H∞ control, as well as the problem of vibra-
tion suppression, where electrohydraulic actuators with highly
B. Random Road Inputs nonlinear characteristics were considered for accurate control.
When the road disturbances are considered vibrations, they An ARC-based H∞ control was introduced to realize the
are consistent and typically specified as random processes. nonlinear disturbance suppression and simultaneously increase
Consider
√ the road inputs as a sequence of independent N (0, the system robustness against both actuator parameter uncer-
2πn 0 G 0 V ) random variables, where G 0 stands for the road tainties and external disturbances. The stability analysis for the
roughness coefficient and n 0 = 0.1 is the reference spatial closed-loop system was given within the Lyapunov framework,
frequency. Select the road roughness as G q (n 0 ) = 256 × and different road conditions were considered to reveal the
10−6 m3 , which corresponds to D Grade (Poor) according to performance of the controllers.
2422 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 6, NOVEMBER 2013

R EFERENCES [21] S. Huang and H. Chen, “Adaptive sliding controller with self-tuning
fuzzy compensation for vehicle suspension control,” Mechatronics
[1] M. Canale, M. Milanese, and C. Novara, “Semi-active suspension control vol. 16, no. 10, pp. 607–622, 2006.
using ’fast’ model-predictive techniques,” IEEE Trans. Control Syst. [22] A. Akbari and B. Lohmann, “Output feedback H∞ /G H2 pre-
Technol., vol. 14, no. 6, pp. 1034–1046, Nov. 2006. view control of active vehicle suspensions a comparison study of
[2] S. Savaresi and C. Spelta, “A single-sensor control strategy for semi- LQG preview,” Veh. Syst. Dyn., vol. 48, no. 12, pp. 1475–1494,
active suspensions,” IEEE Trans. Control Syst. Technol., vol. 17, no. 1, 2010.
pp. 143–152, Jan. 2009. [23] L. Rao and S. Narayanan, “Preview control of random response of a
[3] V. Sankaranarayanan, M. E. Emekli, B. A. Gilvenc, L. Guvenc, E. S. half-car vehicle model traversing rough road,” J. Sound Vibrat., vol. 310,
Ozturk, E. S. Ersolmaz, I. E. Eyol, and M. Sinal, “Semiactive suspension nos. 1–2, pp. 352–365, Feb. 2008.
control of a light commercial vehicle,” IEEE Trans. Mechatron., vol. 13,
[24] J. Lin and R. Lian, “Intelligent control of active suspension sys-
no. 5, pp. 598–604, Oct. 2008.
tems,” IEEE Trans. Ind. Electron., vol. 58, no. 2, pp. 618–628, Feb.
[4] D. Cao, X. Song, and M. Ahmadian, “Editors’ perspectives: Road
2011.
vehicle suspension design, dynamics, and control,” Veh. Syst. Dyn., vol.
[25] J. Cao, P. Li, and H. Liu, “An interval fuzzy controller for vehicle active
49, nos. 1–2, pp. 3–28, 2011.
suspension systems,” IEEE Trans. Intell. Transport. Syst., vol. 11, no.
[5] D. Hrovat, “Survey of advanced suspension developments and
4, pp. 885–895, Dec. 2010.
related optimal control applications,” Automatica, vol. 33, no. 10,
[26] S. Yildirim, “Vibration control of suspension systems using a proposed
pp. 1781–1817, 1997.
neural network,” J. Sound Vibration, vol. 277, no. 4–5, pp. 1059–1069,
[6] B. Gysen, J. Paulides, J. Janssen, and E. Lomonova, “Active electromag-
2004.
netic suspension system for improved vehicle dynamics,” IEEE Trans.
[27] A. Chamseddine and H. Noura, “Control and sensor fault tolerance of
Veh. Technol., vol. 59, no. 3, pp. 1156–1163, Mar. 2010.
[7] G. Koch, O. Fritsch, and B. Lohmann, “Potential of low bandwidth vehicle active suspension,” IEEE Trans. Control Syst. Technol., vol. 16,
active suspension control with continuously variable damper,” Control no. 3, pp. 416–433, May 2008.
Eng. Pract., vol. 18, no. 11, pp. 1251–1262, Nov. 2010. [28] H. Du and N. Zhang, “Fuzzy control for nonlinear uncertain electro-
[8] S. Lee and W. Kim, “Active suspension control with direct-drive tubular hydraulic active suspensions with input constraint,” IEEE Trans. Fuzzy
linear brushless permanent-magnet motor,” IEEE Trans. Control Syst. Syst., vol. 17, no. 2, pp. 343–356, Apr. 2009.
Technol., vol. 18, no. 4, pp. 859–870, Jul. 2010. [29] T. Fukao, A. Yamawaki, and N. Adachi, “Nonlinear and H∞ control of
[9] R. Amirifar and N. Sadati, “Low-Order H∞ controller design for an active suspension systems with hydraulic actuators,” in Proc. 38th Conf.
active suspension system via LMIs,” IEEE Trans. Ind. Electron., vol. 53, Decision Control, pp. 5125–5128, 1999.
no. 2, pp. 554–560, Apr. 2006. [30] M. Ma and H. Chen, “Disturbance attenuation control of active sus-
[10] M. Smith and F. Wang, “Controller parameterization for disturbance pension with nonlinear actuator dynamics,” Inst. Eng. Technol. Control
response decoupling: Application to vehicle active suspension control,” Theory Appl., vol. 5, no. 1, pp. 112–122, 2011.
IEEE Trans. Control Syst. Technol., vol. 10, no. 3, pp. 393–407, [31] Z. Wang, Y. Liu, and X. Liu, “H∞ filtering for uncertain stochastic time-
May 2002. delay systems with sector-bounded nonlinearities,” Automatica, vol. 44,
[11] C. Huang, J. Lin, and C. Chen, “Road-adaptive algorithm design of no. 5, pp. 1268-1277, 2008.
half-car active suspension system,” Expert Syst. Appl., vol. 37, no. 6, [32] H. Karimi and H. Gao, “New delay-dependent exponential H∞ synchro-
pp. 4392-4402, 2010. nization for uncertain neural networks with mixed time delays,” IEEE
[12] N. Yagiz and Y. Hacioglu, “Backstepping control of a vehicle with active Trans. Syst., Man, Cybern., B, Cybern., vol. 40, no. 1, pp. 173–185,
suspensions,” Control Eng. Pract., vol. 16, no. 12, 1457–1467, 2008. Feb. 2010.
[13] H. Gao, J. Lam, and C. Wang, “Multi-objective control of vehicle active [33] J. Lam, Z. Shu, S. Xu, and E. Boukas, “Robust H∞ control of descriptor
suspension systems via load-dependent controllers,” J. Sound Vibrat., discrete-time markovian jump systems,” Int. J. Control, vol. 80, no. 3,
vol. 290, nos. 3–5, pp. 654–675, 2006. pp. 374–385, 2007.
[14] L. Zuo, J. Slotine, and S. Nayfeh, “Model reaching adaptive control for [34] P. Shi, “Filtering on sampled-data systems with parametric uncertainty,”
vibration isolation,” IEEE Trans. Control Syst. Technol., vol. 13, no. 4, IEEE Trans. Autom. Control, vol. 43, no. 7, pp. 1022–1027, Jul. 1998.
pp. 611–617, Jul. 2005. [35] A. Mohanty and B. Yao, “Indirect adaptive robust control of
[15] A. Alleyne and J. Hedrick, “Nonlinear adaptive control of active hydraulic manipulators with accurate parameter estimates,” IEEE
suspensions,” IEEE Trans. Control Syst. Technol., vol. 3, no. 1, Trans. Control Syst. Technol., vol. 19, no. 3, pp. 567–575,
pp. 94–101, Mar. 1995. May 2011.
[16] S. Chantranuwathana and H. Peng, “Adaptive robust force control [36] B. Yao, G. Chiu, and J. Reedy, “Nonlinear adaptive robust control of
for vehicle active suspensions,” Int J. Adapt. Control Signal Process., one DOF electro-hydraulic servo systems,” in Proc. ASME Int. Mech.
vol. 18, no. 2, pp. 83–102, 2004. Eng. Congr. Exposit. Symp. Fluid Power Syst. Technol., vol. 4. 1997,
[17] S. Chantranuwathana and H. Peng, “Adaptive robust control for active pp. 191–197.
suspensions,” in Proc. Amer. Control Conf., 1999, pp. 1702–1706. [37] B. Yao, F. Bu, J. Reedy, and G. Chiu, “Adaptive robust motion
[18] J. Lin and C. Huang, “Nonlinear backstepping active suspension design control of single-rod hydraulic actuators: Theory and experi-
applied to a half-car model,” Veh. Syst. Dyn., vol. 42, no. 6, pp. 373–393, ments,” IEEE Trans. Mechatron., vol. 5, no. 1, pp. 79–91,
2004. Mar. 2000.
[19] P. Nakkarat and S. Kuntanapreeda, “Observer-based backstepping force [38] B. Yao and M. Tomizuka, “Adaptive robust control of mimo nonlinear
control of an electrohydraulic actuator,” Control Eng. Pract., vol. 17, systems in semi-strict feedback forms,” Automatica, vol. 37, no. 9,
no. 8, pp. 895-902, 2009. pp. 1305–1321, 2001.
[20] C. Kaddissi, J. Kenne, and M. Saad, “Drive by wire control of an electro- [39] W. Sun, H. Gao, and O. Kaynak, “Finite Frequency H∞ control for
hydraulic active suspension a backstepping approach,” in Proc. Conf. vehicle active suspension systems,” IEEE Trans. Control Syst. Technol.,
Control Appl., 2005. pp. 1581–1587. vol. 19, no. 2, pp. 416–422, Mar. 2011.