Documente Academic
Documente Profesional
Documente Cultură
mus &&
xus + ks (xus - xs ) + cs (x&us - x&s ) +kt (xus - r ) = 0 …(3) over the letters indicates differentiation with respect
to time). Therefore,
where, mp = Passenger seat mass (kg), ms = Quarter
1
car sprung body mass (kg), mus = Unsprung mass x& 2 = &x&p = −
mp
[ ]
k p ( x1 − x3 ) + c p ( x 2 − x4 ) …(4)
(kg), kp = Seat stiffness (N/m), ks = Spring stiffness
(N/m), kt = Tire stiffness (N/m), cp = Seat damping
1
coefficient (N-s/m), cs = Suspension damping coef- x&4 = &&
x3 = − k p ( x3 − x1 ) + c p ( x4 − x2 )
ficient (N-s/m), xp = Passenger vertical displacement ms
(m), xs = Sprung mass vertical displacement (m),
+ k3 ( x3 − x5 ) + c p ( x4 − x2 ) …(5)
xus = Unsprung mass vertical displacement (m), and
r = Road profile height (m).
1
x&6 = &&
xus = −
mus
[ k3 ( x5 − x3 )
Using notations (x1 = xp, x2 = x& p , x3 = xs, x4 = x&s ,
x5 = xus, x6 = &x&us ), as in Eq. (6), Eqs (1-3) can be + c3 ( x6 − x4 ) + kt ( x5 − r ) ] …(6)
written in state variable form where Putting Eqs (4-6) in state space representation
x1 , x2 , x3 , x4 , x5 , x6 are taken as the states (the dot form7 ( x&6 = Ax + Gr ), it gives
0 1 0 0 0 0
kp cp kp cp
− − 0 0
0
x&1 mp mp mp mp x1
& x 0
x2 0 0 0 1 0 0 0
2
x&3 x3
…(7)
= + 0 r
x&4 kp cp (ks + k p ) (ks + k p ) ks cs x
4 0
x& ms ms
−
ms
−
ms ms ms
x5
5 kt
x&6 x6
0 0 0 0 0 1 mus
ks cs (ks + k p ) cs
0 0 − −
mus mus mus mus
0 1 0 0 0 0
kp cp kp cp 0
− − 0 0 0
mp mp mp mp
0
where A = 0 0 0 1 0 0 and G =
0
kp cp ( ks + k p ) ( cs + c p ) ks cs 0
− − kt
ms ms ms ms ms ms
m us
0 0 0 0 0 1
k +k
ks cs cs
0 0 − s t −
mus mus mus mus
SENTHIL KUMAR & VIJAYARANGAN: LQR CONTROLLER FOR ACTIVE SUSPENSION SYSTEM OF CAR 215
x p + k p ( x p − xs ) + c p ( x& p − x&s ) = 0
m p && …(8)
xs + k p ( xs − x p ) + c p ( x&s − x& p )
ms &&
+ k s ( xs − xus ) + cs ( x&s − x&us ) − f a = 0 …(9)
+ kt ( xus − r ) + f a = 0 …(10)
where, fa denotes the actuator force (N).
Using notations (x1 = xp, x2 = x& p , x3 = xs, x4 = x&s ,
x5 = xus, x6 = &x&us ) as in Eq. (6), it gives
1
x&2 = &&
xp = − k p ( x1 − x3 ) + c p ( x2 − x4 ) …(11)
mp
1
x&4 = &&
x3 = − k p ( x3 − x1 ) + c p ( x4 − x2 ) + k3 ( x3 − x5 )
ms
1
Active Suspension System (ASS) x&6 = &&
xus = −
mus
[ k3 ( x5 − x3 )
ASS has hydraulic actuator in addition to passive
elements (Fig. 2). Hydraulic actuator is located + c3 ( x6 − x4 ) + kt ( x5 − r ) + f a ] …(13)
parallel to suspension spring and shock absorber.
Putting Eqs (11-13) in state space representation form
Using Newton’s Second Law of Motion and free-
( x& = Ax + Bu + Gr ) , represented as Eq. (14), gives
body diagram concept, following equations are
derived: Eq. 15 as shown below:
0 1 0 0 0 0
kp cp kp cp
− − 0 0
0
x&1 mp mp mp mp x1 0 0
& x
x2 0 0 0 1 0 0
2 0
0
x&3 x3 0 …(15)
= + fa + 0 r
x&4 kp cp (ks + k p ) (ks + k p ) ks cs x
4 1
0
x& ms ms
−
ms
−
ms ms ms
x5 0
5 kt
x&6 x6 −1 m
0 0 0 0 0 1 us
ks cs (ks + k p ) cs
0 0 − −
mus mus mus mus
0
where, B = 0 and u = [ fa].
0
1
0
− 1
216 J SCI IND RES VOL 65 MARCH 2006
∞
J = ∫ ( x T Q x + u T R u )dt …(17)
0
K = R −1 B T P …(18)
where the matrix P is evaluated being the solution of
the Algebraic Riccati Equation,
1
v= [ − k p − c p k p c p 0 0] …(24)
Conventional Method (CM) mp
In CM, J penalizes state variables and inputs. Thus,
it has the standard form as, Eq. (22) can be written as,
SENTHIL KUMAR & VIJAYARANGAN: LQR CONTROLLER FOR ACTIVE SUSPENSION SYSTEM OF CAR 217
∞
J = ∫ ( x T ( Q + v T S v )x + u T R u )dt …(25) Table 1 — Road roughness values classified by ISO
0
Classification Road roughness
S (Ω) K [m2/(cycles/m)]
Thus, Eq. (22) could be written from Eq. (25) as,
Range Average
∞
J = ∫ ( x Qn x + u R u )dt
T T
…(26) A (Very good) 2 × 10-6 ∼8×10-6 4 × 10-6
0
B (Good) 8 × 10-6 ∼32×10-6 16 × 10-6
where, Qn = Q + v T S v . Optimal solution for Eq. (26)
could be found similarly to that of Eq. (17). C (Average) 32 × 10-6 ∼128×10-6 64 × 10-6
Table 2 — Characteristics of random road input under poor class Since, all the state variables, acceleration and
control force have been constrained, the problem of
Random road input
controller design is then a challenge for finding
Spatial, mm PSD [m3/(cycles/m]
suitable weightings that satisfies the design
performances. This is done by choosing arbitrary
Min -56 0 weighting matrices Q and R. Vaughan’s algorithm9 is
used for solving the Ricatti equation. MATLAB
Max 49 2.6 × 10-3
software is used for the simulation. In comparison
Mean 0 3.07 × 10-4 with passive system, the body motion and passenger
accelerations in ASS (Figs 5-16) have reduced
STD 17.7 3.24 × 10-4 significantly, which guarantee better ride comfort.
Moreover, the tire deflection is relatively same.
Therefore, ASS improves the ride comfort while
retaining the road handling characteristics, compared
to passive system.
For the purpose of quantitative comparison, since
the road input to the system is in the form of normal
random distribution, it is expected to have normal
distributed outputs. Therefore, using the concept of
variance for the output signals, useful probability
values can be calculated. For a Gaussian normal
distribution, the probability function of the random
signal x (t) can be written as10:
prob [ λσ ≤ x(t ) ≤ λσ ] =
λσ x2
1 λ
Fig. 3 — Random road surface σ 2π −
∫λσ e 2σ 2
dx = (erf )
2
…(31)
Fig. 10 — Sprung mass displacement for active CM and active ADM vs time
Table 3 — Comparison of 90% probability bounds for passive and This confirms the efficiency of ASS in both ride
active systems comfort and road handling performance (Table 3).
90 % Probability bound
Body motions are lower in CM (Figs 5 & 6).
However, passenger displacement and acceleration
S No States Passive CM ADM
are significantly lower in ADM approach. Passenger
1 xp m 2.20 × 10-2 7.40 × 10-3 3.90 × 10-3 acceleration in ADM approach has reduced (13.5%)
2
than the CM approach (Figs 7 & 8). This is also
x
2 && m/s 2.5745 1.1767 × 10-1 1.572 × 10-2 evident through comparing passenger RMS
p
CM suspension, retaining the road handling (90% probability) is calculated. Body bounce and
performance. passenger acceleration in active case has been found
The actuator forces are well below the applied reduced to more than half of their values in passive
limits and practically implementable. In ADM system and suspension travel is also reduced (25.8%).
approach, gaining better ride comfort is possible by Passenger acceleration in ADM approach has reduced
the cost of higher actuator force (Figs 15 & 16). by more than half to that of CM approach. The
However, optimal controller design could limit the passenger acceleration has to be included in the
actuator forces in some realistic bounds. The performance index to improve ride comfort through
quantitative values (Table 3) could be an effective active suspension system.
tool for the designer to satisfy the required
performance or to compare different designs. References
Conclusions 1 Wong J Y, Theory of Ground Vehicles (John Wiley & Sons,
A methodology was developed to design an active New York) 1998, 436-455.
suspension for a passenger car by designing a 2 Hedrick J K & Butsuen T, Invariant properties of
automotive suspensions, Proc Instn Mech Engrs Part D: J
controller, which improves performance of the system Automobile Engg, 204 (1990) 21-27.
with respect to design goals compared to passive 3 Hrovat D & Hubbard M, Optimal vehicle suspensions
suspension system. Mathematical modelling has been minimizing RMS rattle space, sprung-mass acceleration and
performed using a three degrees-of-freedom model of jerk, Trans ASME, 103 (1981) 228-236.
a quarter car for passive and active suspension system 4 Hrovat D, Survey of advanced suspension development and
considering only bounce motion to evaluate the related optimal control applications, Automatica, 33 (1997)
1781-1817.
performance of the suspension with respect to various
5 Williams R A, Automotive active suspensions Part 1: Basic
contradicting design goals. Two controller design principles, Proc Inst Mech Engrs, Part D, 211 (1997) 415-
approaches (CM & ADM) have been examined for 426.
the active system. Realistic random road surface is 6 Lin J S & Kanellakopoulos I, Nonlinear design of active
modeled for simulation, which according to ISO road suspensions, IEEE Control Syst Mag, 17(3) (1997) 45-59.
surface classification, falls under poor category of 7 Ogata K, Modern Control Engineering (Prentice-Hall, New
road. Also, some limits are assigned for all state York) 1996, 100-105.
variables and passenger acceleration and also actuator 8 ISO, Reporting vehicle road surface irregularities, Tech
Report, ISO/TC108/SC2/WG4 N57, 2000.
forces in order to satisfy the design goals. Almost all
9 Lei Z & Samir A N, Structured H2 optimization of vehicle
variables satisfy the constraints. A stochastical suspensions based on multi-wheel models, Vehicle Syst
technique was used for quantitative comparison Dynamics, 40 (2003) 351-371.
between the two controller design approaches of 10 Esmailzadeh E & Taghirad H D, Active vehicle suspensions
active system with respect to passive system, in which with optimal state-feedback control, Proc of CSME, 2
for each variable the amount of bounding limit (1994) 822-834.