Proceedings of the 26th Chinese Control Conference

July 26-31, 2007, Zhangjiajie, Hunan, China

Time Delay Force Control for Vehicle Active Suspension System

Xuan Dong Ji1, Kim Jin Wan2, Nan Yang Hai2, Kim Young Bae3
1,2 Graduate School of Mechanical Engineering Department, Chonnam National University, Gwangju 500-757
E-mail: djxuan77@hotmail.com
3. School of Mechanical System Engineering, Chonnam National University, Gwangju 500-757
E-mail: ybkim@chonnam.ac.kr

Abstract: This paper presents the simulation results of a force tracking controller for a quarter-car suspension system. The ac-
tive suspension system is decomposed into two loops. At the main loop, the desired force signal is calculated by using a stan-
dard LQ design process. The Time Delay Control (TDC) technique is then used to design the force controller such that the de-
sired force signal is achieved in a robust manner when actuator or other plant uncertainties are present. The ADAMS controls
module is used to render joint simulation between ADAMS and MATLAB, of which the results showed that the TDC strategy
is reasonable and feasible in isolating the road vibration effectively.
Key Words: Time Delay Control, Active suspension, ADAMS/Controls, Hydraulic Actuator

1 INTRODUCTION the TDC is that even if some unknown plant dynamics

The important functionality of the vehicle suspension
system is to support the vehicle body as well as to pro-
vide the comfort driving to the passengers by rejecting
the unpleasant vibratory motion induced from the ir-
regular road inputs. Also, the suspension should main-
tain adequate vertical load to provide the vehicle stabil-
ity when the car turns, brakes or accelerates. The vehicle
stability and riding has mutually adverse effects, there-
fore, simply the passive damper, which is widely used in
the usual vehicle, could not satisfy the riding comfort as
well as driving stability simultaneously. However, as the
recent trends of the vehicle industry is to be luxurious
and driver comfort is more required, the electrical con-
trolled suspension system is now installed and widely
utilized[1, 2]. Especially, the active damper is widely
adopted for the luxurious vehicle because it can over-
come the shortage of the passive damper by providing
the ride comfort and vehicle stability at the same time.
In the active suspension system, the sprung mass can
vary according to the variation of passenger numbers
and cargo as well as fuel loading condition. Also the
actuator itself has the strong non-linear characteristics
and the fluctuation of the hydraulic system’s parameters
are not small (i.e. fluid compressibility, fluid leakage
and electrical servo components) [3]. Therefore, the force
control of the active suspension system is necessitated
by the robust control which satisfies the uncertain pa-
rameters as well as disturbance rejection. Recently Kim
utilized the QFT (Quantitative Feedback Control) for the
robust control for the active hydraulic system[3]. Al-
though, QFT works very well in controlling the hydrau-
lic system but it requires engineer heuristics as well as
experience. Recently, in our lab, Xuan showed the ro-
bust control results in controlling the non-linear mag-
netic bearing system using TDC (Time Delay Control) [4],
therefore, TDC is studied in this paper for the applica-
bility to the non-linear hydraulic system. TDC is widely
understood that it can provide the adequate stability as
well as robust trajectory following characteristics even
though some disturbance exists. The basic idea behind
and external disturbance exist, using the relationship
between system input and output (i.e. transfer function)
during few sampling period, the unknown dynamics
term and external disturbance can be effectively re-
moved or compensated by rendering the system to fol-
low the reference dynamic model. By doing so, even
though some disturbance or uncertainties exist in the
system, the output always follows the reference model
one. The beauty of the TDC is that it always guarantees
the stability within the certain sampling time period with
proper parameter selection[5].
There are many application fields for using TDC to con-
trol the non-linear systems effectively. Lee et al. utilized
TDC to the control of DC servo motor control in order
to study the stability of state estimator and overall sys-
tem[6]. In the vehicle area, Song successfully utilized
TDC in controlling engine torque and automatic trans-
mission system[7].
There are very limited reports in controlling the hydrau-
lic system using TDC. Recently, Alleyne utilized H2/H
control technology for hydraulic control[8], but the con-
trol transfer function has high degree of freedoms in
nature that it is not easy to apply it, if exact mathemati-
cal model is not ready. Meanwhile, Chantranuwatan and
Peng used the ARC (Adaptive Robust Control) for con-
trolling the active suspension system robustly[9-10]. Al-
though, their approaches are challenging, it is quite dif-
ficult to produce adequate control algorithm and many
heuristics should be involved for making a success out-
put. Also, their results show some discrepancies in the
minor force control loop at some frequency range. Also,
they represented the hydraulic system having the first
order differential equation form, which does not fully
describe the real one.
For the present paper, we utilized quarter car model to
design LQR (Linear Quadratic Regulator) controller, and
as a minor loop force controller, TDC is utilized for ac-
curate controlling of the hydraulic force. Instead of real
experiment, we utilized Adams real-car like model for
verifying the effectiveness of the proposed method. The
real virtual vehicle of Adams model is used and Ad-

640
ams/control is interfaced by MATLAB/simulink. where Pl = Pa − Pb . In equation (5), kv is valve coeffi-
cient, xsv is valve opening parameter, Ps is supply
2 MODELING
pressure, Pa , Pb are cylinder inlet and outlet pressures,
2.1 Modeling of active suspension respectively and sgn is conventional sign function. By
considering the fluid compressibility and leakage, the
continuity equation can be obtained as below.
In Fig. 1, the two d.o.f. suspension equation of motion
can be written as
ms 
z s + ks ( zs − zu ) + bs ( zs − zu ) = f a
mu 
zu − k s ( zs − zu ) + kt ( zu − zr ) − bs ( zs − zu ) (1)
+bt ( zu − zr ) = − f a

Zs
ms
Fig.2 Components of an electro-hydraulic actuator

Ks bs d ( zs′ − zu′ ) V dpl

fa q=A + LPl + (6)
dt β dt
Zu In above equation, Ap is the piston area, L is fluid
mu
leakage coefficient, V is volume, β is volumetric elas-
Kt bs tic coefficient, zs′ − zu′ = k as ( zs − zu ) is the piston dis-
placement, and kas is the proportional constant. From
Zr equations (5) and (6), the force relation can be arrived at
B( x, t ) (7)
Fig.1 Quarter car model of fully active vehicle suspension The linearization around the operating point can be
In equation (1), all the masses, dampers and stiffness written as
have the same value of the suspension system using in h+d (8)
Adams quarter car model as shown later. If we convert where, kq is the fluidic gain for the electrical servo, and
above equation into the state space form, then
kc is the fluidic pressure constant. As the vehicle’s in-
X = AX + Lzr + Bf a (2)
put maximum road frequency is usually below 100 Hz,
where X = [ zs − zu zu ] ,
T
zs zu − zr the electrical components of high frequency term can be
−1 safely ignored. Therefore, the
⎡ 0 1 0 ⎤
⎢ k ⎥ simplified equation can be obtained as
⎢− s b bs ⎥ xxv = ksv k a v
− s 0 (9)
⎢ m ms ms ⎥
A=⎢ s ⎥ (3) where, ksv is the valve gain, ka is the servo amplifier
⎢ 0 0 0 1 ⎥ gain and v is the input voltage.
⎢ k bs k (bs + bt ) ⎥
⎢ s − t − ⎥ 3 CONTROLLER DESIGN
⎣⎢ mu mu mu mu ⎦⎥
⎡ 0 ⎤ 3.1 Main loop controller
⎢ 1 ⎥ ⎡ 0 ⎤
⎢ ⎥ ⎢ 0 ⎥ Fig.3 shows the overall control algorithm for the active
⎢ ms ⎥ ⎢ ⎥ suspension system. In order to calculate the active force
B=⎢ ⎥ L = ⎢ −1 ⎥ (4)
⎢ ⎥ amplitude for the hydraulic actuator, LQR algorithm is
⎢ 0 ⎥
⎢ 1 ⎥ ⎢ bt ⎥ utilized, which can reduce the road vibration effectively
⎢− ⎥ ⎢⎣ mu ⎥⎦ by satisfying some optimization constraints. In LQR, all
⎣ mu ⎦ possible sensor signals are assumed to be fed back.
2.2 Modeling odeling of hydraulic system Suspension Speed

Fig.2 shows the brief drawing for the hydraulic actuator Force
Controller
Hydraulic
Actuator
Force Suspension
system
driven by electrical servo system. First, the fluid flow
characteristics between load pressure Pl and flow Road Disturbance

q can be obtained as[3]

Desired Force Main-Loop

q = kv xv Ps − sgn( xv ) Pl (5) Controller

Fig.3 Controller architecture

641
The control input for the active suspension is as follows.
f a = −k1 x1 − k 2 x 2 − k 3 x 3 − k 4 x 4 = − Kx (10)
In order to evaluate the suspension capability, the fol-
lowing performance index is introduced.
1
J = lim E ⎡⎢ ∫ 0 { zs2 + ρ1 ( zs − zu ) 2 + ρ 2 zs2
T

T →∞ T ⎣ (11)
+ ρ3 ( zu − zr ) + ρ 4 zu } dt ⎦
2 2
⎤
where, ρ1 , ρ 2 , ρ3 , ρ4 are weight values for all state vari-
ables.
By using the state variables, above equation can be re- Fig.4 The frequency response of tire deflection
arranged with standard type matrices and vectors as
J = lim E ⎡⎢ ∫ 0 {x T Qx + 2 x T Mf a + rf a2 } dt ⎤⎥
1 T
(12)
T →∞ T ⎣ ⎦
where,
⎡ ks2 bs k s bk ⎤ ⎡ ks ⎤
⎢ 2 + ρ1 2
0 − s 2s ⎥ ⎢− m 2 ⎥
⎢ m m m ⎥
s s s
⎢ s⎥
⎢ bk bs2
bs2 ⎥ ⎢ bs ⎥
⎢ s 2s + ρ2 0 ⎥ ⎢− ⎥
Q = ⎢ ms ms 2
ms ⎥ M = ⎢ ms2 ⎥
2

⎢ 0 0 ρ3 0 ⎥ ⎢ 0 ⎥
⎢ ⎥ ⎢ ⎥
⎢ bs k s bs2
bs2
⎥ ⎢ bs ⎥ Fig.5 The frequency response of sprung mass acceleration
⎢ 0 + ρ4 ⎥ ⎢ m2 ⎥
⎣ s ⎦
2
⎣ ms ms2 ms2 ⎦
1
r= 2
ms
From above relations, the typical Riccati equation can be
derived as shown in equation (13). In equation (13), one
can obtain the optimal feedback gain K after solving P .
( A − Br −1 M T ) T P + P( A − Br −1M T )
+(Q − Mr −1M T ) − PBr −1 B T P = 0
K = r −1 ( B T P + M T ) (13)
Fig.6 Comparison of passive and active at tire deflection
The parameters used in the paper are summarized in
Tab.1, and these values are obtained from the quarter car
vehicle model from Adams as shown in Fig. 9. Also
hydraulic parameters are decided from the results of
other paper [3].
After selecting optimal weighting parameters heuristically,
we can compare the vibratory results induced from ir-
regular road inputs between active suspension and passive
one. The comparison is made by plotting the frequency
response and time trajectory as shown in below figures.
Fig.7 Comparison of passive and active at sprung mass
Tab.1 Suspension system parameters acceleration

ms 450 kg L 1.010-9 m5/Ns As shown in Fig. 4 and 5, if active suspension is in-

volved, the displacement at the tire is superior at lower
mu 60 kg kc 9.43410-11 m5/Ns
frequency range, i.e. near first resonant frequency area.
bs 3000 Nsec/m kq 1.149 m5/N/m At the higher frequency range near second resonant area,
bt 400 Nsec/m Ap 1.3510-3 m2
no vibration improvement was achieved. However, Fig.
5 shows that the significant vibration reduction can be
ks 20000 N/m V 3.42910-4 m3 achieved for all frequency ranges with active suspension
kt 2300000 N/m ka 4mA/V at the location of sprung mass. After referring the results
from other publications [9] and many trial results, it is
Ps 5
21010 N/m 2
ksv 1.207105 m/mA
assumed that the weight values applied in the paper is
P0 3.0105 N/m2 kas 0.7 relevant. In Figs 6 and 7 show the time trajectory of the
β 1.36910 N/m 9 2 suspension (at tire and sprung mass). As shown in these
figures, significant vibration reduction can be achieved
642
using the active suspension. Also as shown in the Figs, hˆ( x (t ), t ) + dˆ (t )
even if the road input signal is random (i.e. it contains
many frequency components up to 100 Hz), active sus- ≈ h( x (t − L ), t − L) + d (t − L)
(19)
pension can effectively isolate the input signal. The = x (t − L) − f ( x(t − L), t − L)
problem is that even though the LQR can offer the ap- − B( x (t − L ), t − L) u (t − L)
propriate force control magnitude, the real situation is In many cases, as the matrix B( x(t ), t ) is unknown or
that the hydraulic actuator should be controlled effec-
tively to control the active suspension in the frequency uncertain, the estimated value of Bˆ (t ) is used to obtain
domain or time one. This is because the hydraulic sys- the following TDC rule
tem is strongly non-linear and has many uncertain dy- u (t ) = Bˆ + (t ){− f ( x (t ), t ) − x (t − L )
namic components, which renders the precise control
difficult. Therefore, it is manifest that the inner loop for + f ( x(t − L), t − L) + Bˆ (t − L )u (t − L) (20)
hydraulic control is compulsory. + xm (t ) − Ae e(t )}
3.2 Design of force control for hydraulic actuator In equation (20), it is common that the time delay L is
decided to have the same time period of the sampling
3.2.1 Time delay law time or its integer multiple. By doing so, discrete system
Let’s consider the following time varying non-linear equation can be easily obtained.
dynamical system[11]. 3.2.2 Design of TDC
x (t ) = f ( x(t ), t ) + h( x(t ), t )
(14) In our system, for easy implementation and computa-
+ B( x (t ), t )u (t ) + d (t ) tional burden, the following first order system is se-
where, x(t ), u (t ), f ( x, t ), h( x, t ), d (t ) represent state lected. Although, the reference has lower order system,
variables, control vector, known dynamics, unknown the outcome is very remarkable as explained below.
dynamics and disturbance, respectively. The objective This is another benefit of using TDC.
of the TDC is to design the non-linear controller to sat- 1 1
xm = − xm + r (21)
isfy desired performance under external disturbances as τ τ
well as system uncertainties. The desired performance The error dynamics can also selected as
can be described with satisfying the following 1
time-invariant reference model ae = − (22)
τ
xm = Am xm (t ) + Bm r (t ) (15) The pseudo inverse of B( x, t ) can be derived from
where, xm (t ) and r (t ) represent the state vector and equation (8) as
reference input vector for the reference model. If we V
Bˆ + = − (23)
consider the error vector as e(t ) = xm (t ) − x(t ) , then 4 β ( kc + L)
wanted error dynamics can be represented as Fig.8 shows overall TDC control block diagram.
e(t ) = Ae e(t ) (16)
d (t )
in which, Ae is the error system matrix. It is manifest
r Re ference xm
r ference + e e u y
Ae − B̂ +
Plant
Plant
that if all the eigen values are located at the left half el
mod el +
− +
plane, the error will converge to zero after time passes. y
Am e − LS
Therefore, using this concept, and by using equations
(14) and (16), the following relation can be obtained. +
x m
B( x, t )u (t ) = − f ( x (t ), t ) − h( x (t ), t ) Bm + +
B̂
(17) −
− d (t ) + xm (t ) − Ae e(t )
As B( x, t ) is not square nor its inverse is not always ob-
Fig. 8 Block diagram of hydraulic servo Control system
tained, the pseudo inverse of B( x, t ) should be utilized using TDC
as B + = ( B T B) −1 B T , then the approximated solution of
equation (17) can be achieved.
u (t ) = B + ( x, t ){− f ( x(t ), t ) − h( x (t ), t )
(18)
− d (t ) + xm (t ) − Ae e(t )}
In equation (18), h( x(t ), t ) + d (t ) is the unknown func-
tion, one can use the estimated value of hˆ( x(t ), t ) + dˆ (t ) .
If the time delay L is sufficiently small, and if h + d is
continuous function, then the approximated function
Fig.9 Active suspension mechanical model
value can be expressed as

643
4 Simulation
Fig.9 shows the quarter car model built with Ad-
ams/View. The suspension model studied in this paper is
the real model with Mcpherson front system. The Ad-
ams/control is interfaced with MATLAB/Simulink.
MATLAB/Simulink provided LQR and TDC controller
to control the suspension effectively. Also, the hydraulic
system was included inside of the MATLAB/simulink
loop. The reference force input was provided by the
output of LQR controller. As mentioned above, the de-
cision of the weighting factors are made heuristically. In
this paper, the weighting parameters were calculated Fig.11 TDC force tracking with random vibration
after normalizing the sprung mass acceleration. The x 10
-4

time delay for the reference model was 1 ms. The ex- 2
TDC
ternal disturbance, which is expressed by the suspension Open Loop

velocity, as shown in equation (8), was calculated from 1

Adams virtual model, and its value is introduced to the
hydraulic system. Fig. 10 shows the overall active sus-
pension system in Matlab/simulink utilized in the paper. 0

In the figure, the dark yellow block represents the Ad-

ams mechanical suspension part.
In order to see the performance of the designed TDC -1

controller, Adams/control, Adams/View and Mat-

lab/Simulink are connected and run simultaneously. Any -2
random vibration was engaged into the tire patch point 0 0.5 1 1.5 2 2.5 3

to see the effectiveness of the active system.

Fig. 11 shows the TDC force trajectory ability with ran- Fig.12 Tire deflection response
dom input, respectively. From the results of Fig.3, it is 1
manifest that TDC has very superior performance ability 0.8
TDC
Open Loop
in tracking the force reference, which means more ac-
0.6
curate control in isolating the road vibration.
0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
0 0.5 1 1.5 2 2.5 3

Fig.13 Sprung mass Acceleration response

5 Conclusion

Fig.10 Control and simulation model
compared the effectiveness of LQR/TDC in time do-
main. The comparison was made when TDC is “ON”
and “OFF”, i.e. no control action. Fig. 4 shows the ran-
dom input response at the tire location. Fig. 5 also
shows the same simulation condition with only measur-
ing the time trajectory at sprung mass location. Both
figures show the acceleration responses at both locations.
Incurring from both figures, it is concluded that
LQR/TDC can offer much optimized vibration isolation
even random vibration is introduced from irregular road.
In this paper, LQR main loop is utilized while TDC
force controller is engaged for minor lop in order to iso-
late the road vibration using the active suspension. The
hydraulic actuator model, which has highly non-linear
characteristics such as fluid compressibility, leakage and
electrical components, is derived and utilized as one of
the system dynamical components. From LQR control
design with adequate weighting parameters selection,
we can obtain the effective vibratory reduction results
across the broad band of input spectrum.
In order to verify the effectiveness of the proposed
method, the real-like vehicle virtual model using Ad-
ams/view was constructed with real vehicle data. Also
Matlab/simulink was interfaced with Adams/control

644
model to simulate the hydraulic actuator and TDC con-
troller. The simulation results show that even though
there might have uncertain parameters from hydraulic
system, vehicle suspension and external disturbances
such as road frequency variations, the proposed
LQR/TDC can effectively reduce the vibration from
random road. Also TDC can offer very accurate force
tracking ability under the broad input frequency range.
From the results of the paper, TDC is the effective
means in controlling the hydraulic system and it can be
utilized as one of the control candidates for the active
vehicle suspension.
REFERENCES
[1] Celniker G W, Hedrick J K. Rail vehicle active suspensions
for lateral ride and stability improvemen[J]. ASME J. of
Dynamic system. Measurement and Control, 1982, 104:
101-106.
[2] Alleyne A, Hedrick J K. Nonlinear adaptive control of active
suspension[J]. IEEE Transactions on Control Systems
Technology, 1995, 3(1): 94-101.
[3] Kim J W, Xuan D J, Kim Y B. Design of Forced Control
System for a Dynamic Load Simulator Using QFT. will ap-
pear at International J. of Automotive Tech., 2007.
[4] Xuan D J. Magnetic Bearing Control using TDC. MS Thesis,
645
Chonnam National Univ., 2006.
[5] Youcef-Toumi K, Fuhlbrigge T. Application of Decentral-
ized Time Delay Controller to Robot Manipulator[C]. Proc.
of IEEE Int. Conf. on Robotics and Automation, 1989:
1786-1791.
[6] Lee J W, Chang P H, Kim S H. A Study on Robust Control
of DC Servo Motor Using Time Delay Control and Ob-
server/Controller Stabilization[J]. KSMEA, 1993, 17(5):
1029-1040.
[7] Song J B, Lee S M. Engine Torque and Engine/Automatic
Transmission Speed Control Systems Using Time Delay
Control[J]. Journal of Control, Automation and Systems En-
gineering, 1996, 2(2).
[8] Alleyne A, Liu R. On the limitations of force tracking con-
trol for hydraulic servo systems[J]. ASME J. of Dynamic
system. Measurement and Control, 1999, 21(2): 184-190.
[9] Chantranuwathana S, Peng H. Adaptive robust force control
for vehicle active suspensions[J]. Int. J. Adapt. Control Sig-
nal Process, 2004, 18: 83-102.
[10] Chantranuwathana S, Peng H. Force tracing control for ac-
tive suspensions –theory and experiments. IEEE Interna-
tional Conference on Control Applications, 1999: 22-27.
[11] Hara T C, Gao L S. Robot Manipulator Control using De-
centralized Time Invariant Time-Delayed Controller[C].
Proc. of IEEE Int. Conf. on Robotics and Automation,
1996: 2070-2075.