You are on page 1of 13

Nonlinear Dyn

https://doi.org/10.1007/s11071-018-4060-1

ORIGINAL PAPER

Adaptive robust control methodology for active roll control


system with uncertainty
Hao Sun · Ye-Hwa Chen · Han Zhao

Received: 19 January 2017 / Accepted: 6 January 2018


© Springer Science+Business Media B.V., part of Springer Nature 2018

Abstract The active roll control system (ARCS) can converges to zero quickly in 0.3 s from initial error
impose anti-roll moment quickly to prevent the vehicle 0.287 deg, and the final error is of the order of 10−7 .
rolling when the vehicle generates the roll tendency and Thus, the control design renders the ARCS practically
effectively enhance the vehicle dynamic performance stable and achieves constraints following maneuvering.
without sacrificing the ride comfort. In the dynamic
model of the ARCS, the sprung mass of the vehicle Keywords Active roll control system · System
is considered to be the uncertain parameter, which is uncertainty · Initial condition deviation · Adaptive
(possibly) fast-varying. However, what we know about robust control
the uncertainty is just that it is bounded. Furthermore,
the bound is unknown. The target roll angle is regarded
as the constraint when the vehicle equipped with the 1 Introduction
ARCS is running under a given case. Taking the param-
eter uncertainty and possible initial condition deviation When a vehicle is cornering, the vehicle body leans
from the constraint into account, an adaptive robust toward outside due to the centrifugal force and this roll
control scheme based on the Udwadia and Kalaba’s motion degrades the driving comfort that the passen-
approach is proposed to drive the ARCS to follow the ger feels. Most passenger vehicles today are equipped
pre-specified constraint approximately. The adaptive with the torsion bars, called stabilizer bars, which are
law is of leakage type which can adjust itself based attached at the front and rear axles so as to reduce the
on the tracking error. Numerical simulation shows that roll motion. But these conventional passive stabilizer
by using the adaptive robust control scheme, the error bars are difficult to satisfy the requirements of ride com-
between the actual roll angle and the desired roll angle fort and handling stability at the same time, and cannot
adjust the roll stiffness of the suspension in real time.
H. Sun (B) · H. Zhao When the vehicle is cornering in high speed, it is prone
School of Mechanical Engineering, Hefei University of to roll. Too much roll will make the driver feel insecure
Technology, Hefei, Anhui 230009, People’s Republic of
China
and fatigue.
e-mail: sunhao.0806@163.com The active roll control system (ARCS) can impose
anti-roll moment quickly to prevent the vehicle rolling
Y.-H. Chen · H. Sun when the vehicle generates the roll tendency. It can
The George W Woodruff School of Mechanical
Engineering, Georgia Institute of Technology, Atlanta,
also reduce the roll angle and roll angle velocity of
GA 30332, USA the vehicle body greatly, increase the tire normal force
e-mail: yehwa.chen@me.gatech.edu of the independent suspension, and improve the adhe-

123
H. Sun et al.

sion conditions of the wheels and the road. Hence, the is able to deal with nonlinear time-varying systems. (ii)
ARCS can effectively enhance the vehicle dynamic per- The uncertainty could be (possibly fast) time-varying.
formance without sacrificing the ride comfort [1,2]. It is bounded. The bound is, however, unknown. We
There are distinctions that can be made in ARCS use an adaptive law to emulate a constant parameter
according to the actuator types and the actuator loca- vector which may be relevant to the bounding set (so
tions [3]. The most popular actuator in ARCS is hydro- not the uncertainty itself, as in many other adaptive
pneumatic and hydraulic system [4–8]. An active anti- controls). Our control design is not based on a priori
roll bar system called dynamic drive with hydraulic bound, which may be rather conservative and render-
system has been developed by BMW [9]. However, ing excessively large control effort. (iii) The adaptive
the obvious drawbacks of these hydraulic actuators law is of leakage type, which can adjust the magnitude
are manufacturing cost, power consumption, and their of the adaptive parameter based on the system perfor-
slow responses on various roads and steering inputs. mance. This helps to regulate the control effort. As a
As an alternative, electric ARCS [10–15] is more fuel result, the adaptive law renders a modest choice of the
efficient, energy saving, and environmentally friendly adaptive parameter; thereby, the control effort is also
than conventional hydraulic ARCS. And this electric kept moderate. (iv) This control design is applicable to
ARCS is more suitable for today’s new energy vehi- a wide range of tasks (which we summarize as con-
cles, including hybrid electric vehicles (HEVs), fuel- straint following), including trajectory following and
cell electric vehicles (FCEVs), and pure electric vehi- set-point regulation.
cles (EV).
In this paper, we aim to deal with the active roll con-
trol problem in the view of constraint following with 2 Description of constrained mechanical system
system uncertainty. The target roll angle is regarded with no uncertainty
as the constraint when the vehicle equipped with the
ARCS is running under a given case. Unlike many 2.1 Description of mechanical system
other research works, the sprung mass of the vehicle
is considered to be the uncertain parameter, which is First, we consider a mechanical system that can be
(possibly) fast-varying. However, what we know about described by [23,24]:
the uncertainty is just that it is bounded. Furthermore,
M(q(t), t)q̈(t) + C(q(t), q̇(t), t)q̇(t) + G(q(t), t)
the bound is unknown. Since the presence of uncer-
tainty and possible initial condition deviation from the = τ (t), (1)
constraint, it is only realistic to expect the constraint to where t ∈ R is the time, q ∈ Rn is the coordinate,
be approximately followed. q̇ ∈ Rn is the velocity, q̈ ∈ Rn is the acceleration, and
The main contributions of this paper are threefold. τ ∈ Rn is the control input. Furthermore, M(q, t) is
First, the dynamic model of the electric ARCS is estab- the inertia matrix, C(q, q̇, t) is the Coriolis/centrifugal
lished, where the sprung mass is considered to be the force, G(q, t) is the gravitational force. The matri-
uncertain parameter which is (possibly) fast-varying. ces/vector M(q, t), C(q, q̇, t) and G(q, t) are of appro-
Second, a nominal control obtained from the Udwadia– priate dimensions. The functions M(·), C(·), and G(·)
Kalaba equation [16–22] is employed to address the are assumed to be continuous. Here it should be noted
system dynamics of the ARCS when no uncertainty that the coordinate q does not need to be the gener-
is involved. Third, taking the system uncertainty and alized coordinate. It can be chosen by the designer for
possible initial condition deviation from the constraint convenience based on the specifications of the problem.
into account, an adaptive robust control scheme based
on the Udwadia and Kalaba’s approach is proposed to
drive the ARCS to follow the pre-specified constraint 2.2 Description of constraints
(i.e., the target roll angle) approximately. The adaptive
law is of leakage type which can adjust itself based on Then, we introduce two different kinds of constraints in
the tracking error. the mechanical system: the passive constraints and the
The advantages of the adaptive robust control can be servo constraints. The former ones usually depend on
summarized as follows: (i) The adaptive robust control the environment or structure of the mechanical system

123
Adaptive robust control methodology for active roll control…

to supply the required constraint forces. However, in d  ∂cl (q, t)


n
∂cl (q, t)
cl (q, t) = q̇k + . (12)
terms of the latter ones, the control input of the system dt ∂qk ∂t
k=1
could be designed to generate the appropriate constraint
Let
forces to meet the constraint requirements. In this study, n 
 
d d
we shall focus on the latter ones. bl (q, q̇, t) := cl (q, t) − Ali (q, t) q˙i ,
Consider the following servo constraints: dt dt
i=1

n (13)
Bli (q, t) = dl (q, t), l = 1, . . . , m, (2) then Eq. (10) can be written as
i=1
n
where Bli (·) and dl (·) are both C 1 , m ≤ n. The con- Ali (q, t)q¨i = bl (q, q̇, t), (14)
straints in (2) are holonomic in general. The constraints i=1
can be put in the matrix form which is the second-order form of the constraints.
We will represent (9) and (14) in the matrix form as
B(q, t) = d(q, t), (3)
A(q, t)q̇ = c(q, t), (15)
where B = [Bli ]m×n , d = [d1 d2 · · · dm ]T .
Taking the derivative of Eq. (2) with respect to t A(q, t)q̈ = b(q, q̇, t), (16)
yields where A = [Ali ]m×n , c = [c1 c2 · · · cm ]T and b =

n [b1 b2 · · · bm ]T .
d d
Bli (q, t) = dl (q, t), (4) Remark For a given configuration, the first-order form
dt dt
i=1
constraint (15) governs the possible velocity q̇, while
where the second-order form constraint (16) governs the pos-
d  ∂ Bli (q, t)
n
∂ Bli (q, t) sible acceleration q̈. As was demonstrated in [25], var-
Bli (q, t) = q̇k + , (5)
dt ∂qk ∂t ious control problems, including stabilization, trajec-
k=1
tory following, and optimality, can be cast into the form
d  ∂dl (q, t)
n
∂dl (q, t) (16).
dl (q, t) = q̇k + . (6)
dt ∂qk ∂t
k=1
Let 2.3 Constraint force

n
∂ Bli (q, t) 
n
∂dl (q, t)
Ali (q, t) = − , (7) We now introduce the constraint force in mechanical
∂qk ∂qk system which is subject to the d’Alembert’s virtual
k=1 k=1
work principle. Suppose that the following assumptions
∂dl (q, t) ∂ Bli (q, t)
cl (q, t) = − . (8) are imposed.
∂t ∂t
Equation (4) can be rewritten as Assumption 1 For each (q, t) ∈ Rn × R, the inertia
matrix is positive definite: M(q, t) > 0.

n
Ali (q, t)q̇i = cl (q, t), (9) Assumption 2 The constraint Aq̈ = b is consistent
i=1 (i.e., A A+ b = b, where ‘+’ is the Moore-Penrose gen-
which is the first-order form of the constraints. eralized inverse [26]). That is, for given A and b, there
Differentiating the first-order form of constraints (9) exists at least one solution q̈.
with respect to t yields Theorem 1 (Udwadia and Kalaba [17]) Consider the
n   n
d d mechanical system (1) and the constraint (16). If the
Ali (q, t) q˙i + Ali (q, t)q¨i = cl (q, t),
dt dt Assumptions 1, 2 are satisfied, then the constraint force
i=1 i=1
is given by
(10)
Q c (q, q̇, t)
where
= M 1/2 (q, t)(A(q, t)M −1/2 (q, t))+ [b(q, q̇, t)
d  ∂ Ali (q, t)
n
∂ Ali (q, t)
Ali (q, t) = q̇k + , (11) + A(q, t)M −1 (q, t)(C(q, q̇, t)q̇ + G(q, t))].
dt ∂qk ∂t
k=1
(17)

123
H. Sun et al.

Remark Udwadia and Kalaba have verified that the 3 Adaptive robust servo control design with system
constraint force enables all the constraints to be exactly uncertainties
satisfied at every instant of time with minimum control
cost [17]. Based on this theorem, one could apply the We now consider that there exist uncertainties in the
control input τ = Q c to drive the mechanical system mechanical system while designing the control and
to meet the constraint (16). rewrite Eq. (1) as
M(q(t), σ (t), t)q̈(t) + C(q(t), q̇(t), σ (t), t)q̇(t)
2.4 Control input with possible initial condition + G(q(t), σ (t), t) = τ (t), (21)
deviation with
M(q, σ, t) = M(q, t) + M(q, σ, t),
However, it should be noted that the constraints must
C(q, q̇, σ, t) = C(q, q̇, t) + C(q, q̇, σ, t),
be satisfied at each instant of time during the operation
including the initial time (t = 0) when we directly G(q, σ, t) = G(q, t) + G(q, σ, t).
apply the Udwadia–Kalaba approach (i.e., Theorem 1) Here M, C and G are the ‘nominal’ portions with
to design the control input. But in real life, it is usually M > 0, M, C and G are the uncertain
quite difficult to satisfy these constraints at the initial portions. The functions M(·), M(·), C(·), C(·),
time due to various reasons. Hence, in order to deal G(·) and G(·) are all continuous. Let D(q, t) :=
−1 −1
with the possible initial condition deviation from the M (q, t), D(q, σ, t) := M −1 (q, σ, t)− M (q, t),
constraints, the control design needs to be extended. E(q, σ, t) := M(q, t)M −1 (q, σ, t) − I , hence we can
get D(q, σ, t) = D(q, t)E(q, σ, t).
Assumption 3 For each (q, t) ∈ Rn × R, A(q, t) is
of full rank and A(q, t) is bounded. This means Assumption 4 Under the provision of Assumption 3,
A(q, t)AT (q, t) is invertible. for given P ∈ Rm×m , P > 0, let
W (q, σ, t)
Theorem 2 (Chen [27]) Consider the system (1) and
:= P A(q, t)D(q, t)E(q, σ, t)M(q, t)AT (q, t)
the constraint (16) with the possible initial condition
deviation. Under the provision of Assumptions 1, 2, 3, ×(A(q, t)AT (q, t))−1 P −1 . (22)
for given P ∈ Rm×m , P > 0, the control input can be There exists a (possibly unknown) constant ρ E > −1
designed as such that for all (q, t) ∈ Rn × R,
τ (t) = p1 (q(t), q̇(t), t) + p2 (q(t), q̇(t), t), (18) 1
min λm (W (q, σ, t) + W T (q, σ, t)) ≥ ρ E . (23)
with 2 σ∈

p1 (q, q̇, t) = M 1/2 (q, t)(A(q, t)M −1/2 (q, t))+ [b(q, q̇, t) Assumption 5 (1) There exists an unknown constant
vector α ∈ (0, ∞)k and a known function (·) :
+ A(q, t)M −1 (q, t)(C(q, q̇, t)q̇
(0, ∞)k × Rn × Rn × R → R+ such that for all
+ G(q, t))], (19)
(q, q̇, t) ∈ Rn × Rn × R, σ ∈ ,

p2 (q, q̇, t) = − κ M(q, t)AT (q, t)(A(q, t)AT (q, t))−1 P −1


(1 + ρ E )−1 max[P A(q, t)D(q, σ, t)
σ∈
× β(q, q̇, t), (20) ×(− C(q, q̇, σ, t)q̇ − G(q, σ, t) + p1 (q, q̇, t)
where κ is a positive constant. + p2 (q, q̇, t)) − P A(q, t)D(q, t)(C(q, q̇, σ, t)q̇
+ G(q, σ, t))]
Remark The tracking constraint error β(q, q̇, t) can be
derived based on the given constraints. If the constraints ≤ (α, q, q̇, t). (24)
are of the first-order form (i.e., Eq. 15), β(q, q̇, t) can (2) For each (α, q, q̇, t), the function (α, q, q̇, t) can
de defined as β(q, q̇, t) := A(q, t)q̇ − c(q, t). If the be linearly factorized with respect to α: There exists a
constraints are of the zero-order form (i.e., Eq. 3), then function ˜ (·) : Rn × Rn × R → R+ k such that

β(q, q̇, t) can de defined as β(q, q̇, t) := A(q, t)q̇ −


c(q, t) + k(B(q, t) − d(q, t)) = Â(q, t)q̇ − ĉ(q, t). (α, q, q̇, t) = α T ˜ (q, q̇, t). (25)

123
Adaptive robust control methodology for active roll control…

We now propose the control error of the system is large, the adaptive parameter α̃
should increase quickly so as to effectively compensate
τ (t) = p1 (q(t), q̇(t), t) + p2 (q(t), q̇(t), t)
the uncertainty. Hence, the first term on the RHS of Eq.
+ p3 (α̃(t), q(t), q̇(t), t), (26) (32) should be large, while the leakage term should be
with small. When the system performance is satisfactory, the
adaptive rate α̃˙ needs to be negative and α̃ decreases.
p1 (q, q̇, t) This helps to prevent the estimated uncertainty bound
1/2 −1/2
:= M (q, t)(A(q, t)M (q, t))+ [b(q, q̇, t) to be unnecessary large which in turn prevents the con-
−1 trol magnitude to be too large.
+ A(q, t)M (q, t)(C(q, q̇, t) + G(q, t))],
(27) Theorem 3 (Chen [27]) Let δ := [β T (α̃ − α)T ]T ∈
Rm+k . Consider the mechanical system (1) and sup-
p2 (q, q̇, t) pose Assumptions 1, 2, 3, 4, and 5 are met. The control
:= − κ M(q, t)AT (q, t)(A(q, t)AT (q, t))−1 P −1 (26) renders the following performance:
(i) Uniform boundedness: For any r > 0, there is a
× β(q, q̇, t), (28)
d(r ) < ∞ such that if ||δ(t0 )|| ≤ r , then ||δ(t)|| ≤ d(r )
p3 (α̃, q, q̇, t) for all t ≥ t0 .
(ii) Uniform ultimate boundedness: For any r > 0
:= − [M(q, t)AT (q, t)(A(q, t)AT (q, t))−1 P −1 ] with ||δ(t0 )|| ≤ r , there exists a d > 0 such that
× γ (α̃, q, q̇, t)μ(α̃, q, q̇, t) (α̃, q, q̇, t), (29) ||δ(t)|| ≤ d for any d > d as t ≥ t0 + T (d, r ), where
where T (d, r ) < ∞.

⎧ Remark Due to the presence of uncertainty and possi-




1
||μ(α̃,q,q̇,t)|| i f ||μ(α̃, q, q̇, t)|| > ε̂,
ble initial condition deviation from the constraints, it is
γ (α̃, q, q̇, t) = (30)

⎩ only reasonable to expect the constraints to be approx-
1
i f ||μ(α̃, q, q̇, t)|| ≤ ε̂,
ε̂ imately followed. The control τ (t) consists of three
portions. Each of them performs its own function. (i)
μ(α̃, q, q̇, t) = β(q, q̇, t) (α̃, q, q̇, t), (31) The control p1 is the nominal action which gets rid of
and κ is a positive constant. the components that are not constrained in the deter-
The parameter α̃ is governed by the following adap- ministic model so as to prompt the system toward the
tive law: pre-specified constraints. (ii) The control p2 deals with
the possible initial condition deviation from the con-
α̃˙ = k1 ˜ (q, q̇, t)||β(q, q̇, t)|| − k2 α̃, (32) straints and drives the system toward the pre-specified
α̃i (t0 ) > 0 (where α̃i is the i − th component of the constraints. (iii) The control p3 is the adaptive robust
vector α̃, i = 1, . . . , k), k1,2 ∈ R, k1,2 > 0. action which is based on the adaptive parameter α̃. It
aims to tackle the uncertainty in the model and ensure
Remark In this study, the uncertainty could be (pos- that the system does not deviate from the pre-specified
sibly fast) time-varying. It is bounded. The bound is, constraints with the influence of uncertainty.
however, unknown. We use an adaptive law to emulate a
constant parameter vector which may be relevant to the
bounding set (so not the uncertainty itself, as in many
other adaptive controls). Our control design is not based 4 Dynamic model of the active roll control system
on a priori bound, which may be rather conservative
and rendering excessively large control effort. Figure 1 shows the behavior of the active roll control
system. When the vehicle enters a corner, a rolling force
Remark The adaptive law (32) which governs the adap- is generated by the centrifugal force. Consequently, the
tive parameter α̃ is of leakage type. The second term lower arm on one side of the suspension is pulled up,
of the right-hand side (RHS) is the leakage term. Note and the other is pulled down. The active roll control sys-
that if the initial condition α̃i (t0 ) is chosen to be strictly tem aims to control the torsion angle of the stabilizer
positive, then α̃i (t) > 0 for all t ≥ t0 . If the tracking bar so as to offset the roll moment caused by lateral

123
H. Sun et al.

K φ = K φ f + K φr = 2K cφ f + 2K tφ f + 2K cφr
+ 2K tφr . (36)
Remark In this paper, we consider the sprung mass of
the vehicle is uncertain parameter: m s = m s + m s (t),
due to the variation in the vehicle load. Here m s (t) is
unknown, possibly time-varying variation of m s , with
the possible bound of the variation is unknown.
The required compensating roll moment can be dis-
tributed to the front and rear components as given by
Eqs. (38) and (39), respectively, using the roll stiffness
distribution to the front axle ration of Eq. (37)
Maf
α= , (37)
Maf + Mar
Fig. 1 Behavior of the active roll control system
Maf = α(m s h s a y + m s gh s φ − Is φ̈ − Cφ φ̇ − K φ φ),
acceleration. It is realized by utilizing the roll reaction (38)
force generated by the actuators and the roll reaction Mar = (1 − α)(m s h s a y + m s gh s φ − Is φ̈
force obtained from the roll stiffness of the suspension.
− Cφ φ̇ − K φ φ). (39)
Figure 2 illustrates the dynamic model of the ARCS.
Key variables and parameters of the ARCS are listed The active torque m af and m ar for the actuator posi-
in Table 1. Equation (33) expresses the moment equi- tions in the active suspension system can be obtained
librium around the sprung roll center. by conversion using Eqs. (40) and (41)
Is φ̈ + Cφ φ̇ + K φ φ + Maf + Mar m af = γ f Maf , (40)
= m s h s a y + m s gh s φ, (33) m ar = γr Mar . (41)
where Remark γ f and γr are constants related to the geometry
Is = m s h 2s , (34) of the front and rear stabilizer bars, respectively. Due to
Cφ = Cφ f + Cφr = 2Ccφ f the limit of the space layout of the vehicle, the structure
of the front and rear stabilizer bars is usually different.
+ 2Ccφr , (35)
Hence, γ f and γr are correspondingly different.

Fig. 2 Dynamic model of


the active roll control
system

123
Adaptive robust control methodology for active roll control…

Table 1 Nomenclature of the active roll control system with

ms Sprung mass
G(q, t) = (2K cφ f + 2K tφ f + 2K cφr + 2K tφr
Is Moment of inertia of sprung mass − m s gh s )φ − m s h s a y ,
Cφ Total roll damping G(q, t) = (2K cφ f + 2K tφ f + 2K cφr + 2K tφr
Cφ f Front roll damping − m s gh s )φ − m s h s a y ,
Cφr Rear roll damping
and
Ccφ f Front coil spring damping
Ccφr Rear coil spring damping τ = − (Maf + Mar ).
Kφ Total roll stiffness Following the dynamics of the ARCS (i.e., Eq. 42)
Kφ f Front roll stiffness with the parameter uncertainty, the controller design
K φr Rear roll stiffness will be presented in the next section to render the vehi-
K cφ f Front coil spring rate cle equipped with the ARCS to achieve the target roll
K cφr Rear coil spring rate angle in a given driving condition.
K tφ f Front torsion spring rate
K tφr Rear torsion spring rate
φ Roll angle 5 Controller design
φ̇ Roll rate
φ̈ Roll acceleration
5.1 Target roll angle of the ARCS
hs Distance from roll axis to CG of sprung mass
The roll angle is used as an evaluation index in the
g Gravity constant
development of a vehicle equipped with the ARCS.
ay Lateral acceleration
Based on the results of roll attitude angle and a sub-
α Front roll stiffness ratio
jective evaluation, a target roll angle of 1 degree for a
Maf Front active roll moment
lateral acceleration of 5 m/s2 is set, which is half that of
Mar Rear active roll moment
a vehicle equipped with conventional passive stabilizer
γf Front transmission ratio bar system for the same lateral acceleration. Figure 3
γr Rear transmission ratio shows the relationship between the target roll angle
m af Front actuator active torque according to the lateral acceleration.
m ar Rear actuator active torque

5.2 Constraint
Now we define q = φ and rewrite Eq. (33) in the
The vehicle equipped with the ARCS is assumed to run
form of Eq. (21) as
under the given case with a y = 6 sin 0.5t m/s2 . From
M(q(t), σ (t), t)q̈(t)
+ C(q(t), q̇(t), σ (t), t)q̇(t)
+ G(q(t), σ (t), t) = τ (t), (42)
where
M(q, σ, t) = M(q, t) + M(q, σ, t),
with
M(q, t) = m s h 2s , M(q, σ, t) = m s h 2s ,
C(q, q̇, σ, t) = C(q, q̇, t) + C(q, q̇, σ, t),
with
C(q, q̇, t) = 2Ccφ f + 2Ccφr , C(q, q̇, σ, t) = 0,
G(q, σ, t) = G(q, t) + G(q, σ, t), Fig. 3 Target roll angle of the ARCS

123
H. Sun et al.

Fig. 3, we can obtain the target roll angle according to given constraint. The numerical integration throughout
this lateral acceleration as this paper is done in the MATLAB environment [28–
33], using a variable time step ode 15i integrator.
φd = 0.2 ∗ a y = 1.2 sin 0.5t deg
π The nominal system parameters of the vehicle are
= sin 0.5t rad. (43) estimated as follows:
150
Differentiating Eq. (43) with respect to time once m s = 1762 kg, h s = 0.452 m,
and twice, respectively, yields Ccφ f = Ccφr = 3000 Nm/(rad/s),
0.5π K cφ f = K cφr = 10440 Nm/rad,
φ̇d = cos 0.5t rad/s, (44)
150 K tφ f = K tφr = 14760 Nm/rad,
and
g = 9.8 m/s2 , α = 0.52,
0.25π
φ̈d = − sin 0.5t rad/s2 . (45) γ f = 0.301, γr = 0.195. (50)
150
Hence, if Eqs. (43)–(45) are written in the zeroth- The initial conditions of the constraints are given as
order form, first-order form, and second-order form, follows:
respectively, we can get
φ(0) = 0.287deg, φ̇(0) = 0.573deg/s. (51)
A = 1,
π It should be noted that these initial conditions for the
d= sin 0.5t, vehicle do not satisfy the given constraint.
150
0.5π For simulation, the control parameters are chosen as:
c= cos 0.5t, κ = 1, k1 = 4, k2 = 0.2, ε = 0.0001, P = 1 and the
150
0.25π uncertainty is chosen as: m s = 40 sin 0.1t. Note that
b=− sin 0.5t. (46) the terms in M, C q̇ and G are either constant, linear in
150
position, or linear in velocity. As a result, Assumption 5
is met by choosing
5.3 Control inputs of the front and rear actuators
(α, q, q̇, t) = α1 q̇ + α2 q + α3
Considering the sprung mass of the vehicle in the ≤ α(q̇ + q + 1)
dynamic model (i.e., Eq. 42) is the uncertain param- =: α ˜ (q, q̇, t), (52)
eter: m s = m s + m s (t), the adaptive robust control
methodology proposed in Sect. 3 is utilized to deal with where α1,2,3 > 0 are unknown constant parameters,
the constraint following control problem of the ARCS. α = max{α1 , α2 , α3 }.
The control torques of the front electric actuator and For performance demonstrations, the area enclosed
the rear electric actuator can be derived as Eqs. (48) by β(t) and t is addressed:
and (49), respectively, T
S= β(t)dt. (53)
τ = p1 (q, q̇, t) + p2 (q, q̇, t) + p3 (α̃, q, q̇, t), (47) 0
This stands for the accumulative tracking constraint
m af = − αγ f τ, (48) error.
Similarly, the control cost of the front actuator and
m ar = − αγr τ. (49)
rear actuator can be defined as
T
Sf = |m af |dt, (54)
6 Numerical simulations 0
and
In this section, numerical simulations will be done T
to evaluate the performance of the proposed adaptive Sr = |m ar |dt. (55)
robust control methodology to deal with the approxi- 0
mate constraint following of the ARCS with parame- In comparison with [34], the major difference
ter uncertainty and initial condition deviation from the between our works is that our control design is for

123
Adaptive robust control methodology for active roll control…

Fig. 4 Roll angle actual roll angle with adaptive robust controll
2.5 desired roll angle
1.2 actual roll angle with LQR control
2 1.1
1
1.5 0.9

roll angle φ (deg)


14 16 18

0.5

−0.5

−1

−1.5
0 5 10 15 20 25 30 35 40 45 50
times (s)

Fig. 5 Error of roll angle 0.3


error of roll angle with adaptive robust control
error of roll angle with LQR control
0.25
error of roll angle eθ (deg)

0.2

0.15

0.1

0.05

0 5 10 15 20 25 30 35 40 45 50
times (s)

Fig. 6 Roll angle versus 2.5


actual roll angle−lateral accerelation with adaptive robust control
lateral acceleration desired roll angle−lateral acceleration
2 actual roll angle−lateral accerelation with LQR control
1
1.5
0.95
roll angle φ (deg)

1
0.9

0.5
0.85

0 4.5 4.6 4.7 4.8

−0.5

−1

−1.5

−6 −4 −2 0 2 4 6
2
lateral acceleration a (m/s )
y

123
H. Sun et al.

0.25
1.2
adaptive robust control
1 LQR control 100%
0.2
0.8

Area (||β||)
0.15 0.6
34.1%
α̃

0.4
0.1
0.2

0.05 0
1
2
0
0 10 20 30 40 50
times (s) Fig. 9 Comparison of the accumulative tracking constraint error
β
Fig. 7 Adaptive parameter α̃ history

between the actual roll angle and the desired roll angle
nonlinear time-varying systems, while [34] is for lin- with LQR control converges to zero more slowly with
ear time-varying system. To demonstrate the effec- some fluctuations. Figure 6 presents the relationship
tiveness of the adaptive robust control, LQR (linear between the roll angle according to the lateral acceler-
quadratic regulator) control is applied for comparison. ation. Figure 7 shows the adaptive parameter α̃ history.
The robustness measures attributed to the LQR include α̃ increases quickly from the initial value α̃(0) = 0.02
a one-half gain reduction in any input channel, an infi- to the max value α̃max = 0.2074 to effectively com-
nite gain amplification in any input channel, or a phase pensate the uncertainty at first. Then, the adaptive rate
error of plus or minus sixty degrees in any input chan- α̃˙ needs to be negative and α̃ decreases to prevent the
nel. estimated uncertainty bound to be unnecessary large
Figures 4 and 5 show the actual roll angle of the vehi- which in turn prevents the control magnitude to be too
cle and its error by using the adaptive robust control large. Figures 8 and 9 illustrate the comparison of the
methodology. The error between the actual roll angle tracking constraint error β between the use of the
and the desired roll angle that converges to zero quickly adaptive robust control p1 + p2 + p3 and the LQR con-
in 0.3 s as time progresses from initial error 0.287deg trol. The accumulative tracking constraint error with
is shown in Fig. 5, and the final error is of the order LQR control is nearly 3 times of that with adaptive
of 10−7 . From Fig. 5, it is easy to see that the error robust control. Figures 10, 11, 12, and 13 present the

Fig. 8 Comparison of the 0.03


constraint error with adaptive robust control
tracking constraint error −3
x 10 constraint error with LQR control
2
β 0.025
1

0.02 0

−1
||β||

0.015 14 16 18 20 22

0.01

0.005

0
0 5 10 15 20 25 30 35 40 45 50
time (s)

123
Adaptive robust control methodology for active roll control…

Fig. 10 Comparison of the 400


front actuator active torque with adaptive robust control
control input of the front

front actuator active torque maf (Nm)


front actuator active torque with LQR control
actuator 300

200

100

−100

−200

−300

−400
0 5 10 15 20 25 30 35 40 45 50
time(s)

1.2

control cost of the rear actuator


1.2
control cost of the front actuator

adaptive robust control adaptive robust control


LQR control
LQR control
100% 100%
1 1

69.4% 0.8 69.4%


0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
1
1 2
2
Fig. 13 Comparison of the control cost of the rear actuator
Fig. 11 Comparison of the control cost of the front actuator

comparison of the control inputs of the front and rear tor S f with LQR control is nearly 1.45 times of that
electric actuators, respectively. From Fig. 11, we can with adaptive robust control. From Fig. 13, we can find
find that the control cost of the front electric actua- that the control cost of the rear electric actuator Sr with

Fig. 12 Comparison of the 250


rear actuator active torque with adaptive robust control
control input of the rear rear actuator active torque with LQR control
rear actuator active torque mar (Nm)

200
actuator
150

100

50

−50

−100

−150

−200
0 5 10 15 20 25 30 35 40 45 50
time(s)

123
H. Sun et al.

Fig. 14 α̃max versus k1 and 0.8

k2
0.7

0.6
0.8
0.7 0.5

0.6
0.4
0.5

αmax
0.4
0.3
0.3
0.2 0.2
0.1
0.35
0.3 6
5.5
0.25 5
4.5
0.2 4
3.5
0.15 3
k 2.5
2 0.1 2 k
1

LQR control is also nearly 1.45 times of that with adap- can adjust itself based on the tracking error. Numerical
tive robust control. Figure 14 shows the influence of simulations demonstrate that the adaptive robust con-
the adaptive law parameters k1 and k2 on α̃max . α̃max trol renders the ARCS to follow the pre-specified con-
gradually increases, subject to the increasing of k1 and straint better in comparison with LQR control. It should
decreasing of k2 . be noted that the choices of the control parameters are
According to the obtained results, we know that not unique, many options satisfied the assumptions are
when the vehicle equipped with the ARCS is assumed available which is based on the specific of the problem.
to run under the given case with a y = 6 sin 0.5t m/s2 , The flexibility of the adaptive robust control scheme
the proposed adaptive robust control scheme will broadens the applicability in many systems.
ensure the ARCS quickly responds to the desired roll
angle d = 1.2 sin 0.5t deg, which is half that of a Acknowledgements The research is supported by the China
Postdoctoral Science Foundation (No. 2016M590563), the
vehicle equipped with conventional passive stabilizer National Natural Science Foundation of China (No. 51505116),
bar system. Thus, the ARCS could reduce the roll angle and the Fundamental Research Funds for the Central Universities
of the vehicle body quickly to make the passengers feel (No. JZ2016HGTB0716).
comfort.
References
7 Conclusion
1. Cronje, P.H., Els, P.S.: Improving off-road vehicle handling
using an active anti-roll bar. J. Terramechanics 47, 179–189
In this paper, we successfully deal with the active roll (2010)
control problem in the view of constraint following with 2. Gosselin-Brisson, S., Bouazara, M., Richard, M.J.: Design
system uncertainty. The target roll angle is regarded of an active anti-roll bar for off-road vehicles. Shock Vib.
16, 155–174 (2009)
as the constraint to follow when the vehicle equipped
3. Zulkarnain, N., Imaduddin, F., Zamzuri, H., et al.: Applica-
with the ARCS is running under a given case. In the tion of an active anti-roll bar system for enhancing vehicle
dynamics of the ARCS, the sprung mass of the vehi- ride and handling. In: IEEE Colloquium on Humanities, Sci-
cle is considered to be the uncertain parameter, which ence and Engineering (CHUSER), pp. 260–265 (2012)
4. Konik, D.: Development of the dynamic drive for the new
is (possibly) fast-varying. However, what we know 7 series of the BMW group. Int. J. Veh. Des. 28, 131–149
about the uncertainty is just that it is bounded. Fur- (2002)
thermore, the bound is unknown. An adaptive robust 5. Kim, S., Park, K., Song, H.J., et al.: Development of control
control scheme on the basis of Udwadia and Kalaba’s logic for hydraulic active roll control system. Int. J. Automot.
Technol. 13, 87–95 (2012)
approach is proposed to deal with the system uncer- 6. Sename, O., Dugard, L., Gaspar, P.: Active anti-roll bar con-
tainty and possible initial condition deviation from the trol using electronic servo valve hydraulic damper on single
constraint. The adaptive law is of leakage type which unit heavy vehicle. IFAC Pap. Line 49, 418–425 (2016)

123
Adaptive robust control methodology for active roll control…

7. Vu, V.T., Sename, O., Dugard, L., et al.: Enhancing roll sta- 22. Udwadia, F.E., Phohomsiri, P.: Explicit equations of motion
bility of heavy vehicle by LQR active anti-roll bar control for constrained mechanical systems with singular mass
using electronic servo-valve hydraulic actuators. Veh. Syst. matrices and applications to multi-body dynamics. Proc. R.
Dyn. 55, 1405–1429 (2017) Soc. A Math. Phys. Eng. Sci. 462, 2097–2117 (2006)
8. Vu, V.T., Sename, O., Dugard, L., et al.: H∞ active anti-roll 23. Pars, L.A.: A Treatise on Analytical Dynamics. Heinemann
bar control to prevent rollover of heavy vehicles: a robustness Press, London (1965)
analysis. IFAC Pap. Line 49, 99–104 (2016) 24. Rosenberg, R.M.: Analytical Dynamics of Discrete Sys-
9. Strassberger, M., Guldner, J.: BMW’s dynamic drive: an tems. Plenum Press, New York (1977)
active stabilizer bar system. IEEE Control Syst. 24, 28–29 25. Chen, Y.H.: Constraint-following servo control design for
(2004) mechanical systems. J. Vib. Control 15, 369–389 (2009)
10. Kim, H.J., Lee, C.R.: Hybrid roll control using electric ARC 26. Noble, B., Daniel, J.W.: Applied Linear Algebra. Prentice-
system considering limited bandwidth of actuating module. Hall, Englewood Cliffs (1988)
Int. J. Automot. Technol. 3, 123–128 (2002) 27. Chen, Y.H., Zhang, X.R.: Adaptive robust approximate
11. Ohta, Y., Kato, H., Yamada, D., et al.: Development of an constraint-following control for mechanical systems. J.
electric active stabilizer system based on robust design. SAE Frankl. Inst. 347, 69–86 (2010)
Technical Paper, 2006-01-0758 (2006) 28. Valipour, M., Montazar, A.A.: An evaluation of SWDC and
12. Buma, S., Ookuma, Y., Taneda, A., et al.: Design and devel- WinSRFR models to optimize of infiltration parameters in
opment of electric active stabilizer suspension system. J. furrow irrigation. Am. J. Sci. Res. 69, 128–142 (2012)
Syst. Des. Dyn. 4, 61–76 (2010) 29. Viero, D.P., Valipour, M.: Modeling anisotropy in free-
13. Jeon, K., Hwang, H., Choi, S., et al.: Development of an surface overland and shallow inundation flows. Adv. Water
electric active rollcontrol (ARC) algorithm for a SUV. Int. Resour. 104, 1–14 (2017)
J. Automot. Technol. 13, 247–253 (2012) 30. Valipour, M.: Ability of Box–Jenkins models to estimate
14. Muniandy, V., Samin, P.M., Jamaluddin, H.: Application of of reference potential evapotranspiration (a case study:
a self-tuning fuzzy PIPD controller in an active anti-roll bar Mehrabad Synoptic Station, Tehran, Iran). IOSR J. Agric.
system for a passenger car. Veh. Syst. Dyn. 53, 1641–1666 Vet. Sci. 1, 1–11 (2012)
(2015) 31. Valipour, M.: Application of new mass transfer formulae for
15. Kong, Z., Pi, D., Wang, X., et al.: Design and evaluation of a computation of evapotranspiration. J. Appl. Water Eng. Res.
hierarchical control algorithm for an electric active stabilizer 2, 33–46 (2014)
bar system. Stroj. vestn. J. Mech. Eng. 62, 565–576 (2016) 32. Valipour, M., Sefidkouhi, M.A.G., Raeini, M.: Selecting
16. Udwadia, F.E., Kalaba, R.E.: A new perspective on con- the best model to estimate potential evapotranspiration with
strained motion. Proc. R. Soc. 439, 407–410 (1992) respect to climate change and magnitudes of extreme events.
17. Udwadia, F.E., Kalaba, R.E.: Analytical Dynamics: A New Agric. Water Manag. 180, 50–60 (2017)
Approach. Cambridge University Press, Cambridge (1996) 33. Valipour, M.: How much meteorological information is nec-
18. Udwadia, F.E., Kalaba, R.E.: Explicit equations of motion essary to achieve reliable accuracy for rainfall estimations?
for mechanical systems with non-ideal constraints. J. Appl. Agriculture 6, 1–9 (2016)
Mech. 68, 462–467 (2001) 34. Liu, C., Sun, Z., Shi, K., et al.: Robust non-fragile state
19. Udwadia, F.E.: On constrained motion. Appl. Math. Com- feedback attitude control for uncertain spacecraft with input
put. 164, 313–320 (2005) saturation. Proc. Inst. Mech. Eng. Part G J. Aerosp. Eng.
20. Udwadia, F.E., Kalaba, R.E.: On the foundations of ana- (2016). https://doi.org/10.1177/0954410016679194
lytical dynamics. Int. J. Non-linear Mech. 37, 1079–1090
(2002)
21. Udwadia, F.E., Kalaba, R.E.: What is the general form of
the explicit equations of motion for constrained mechanical
systems? J. Appl. Mech. 69, 335–339 (2002)

123

You might also like