Development of an Electronically Controlled

Pneumatic Suspension for Commercial Vehicles

V. Bhandari, and S. C. Subramanian, Member, IEEE

In a vehicle equipped with an active or a semi-active

Abstract--The suspension system plays a critical role in the suspension system, the sensors continuously measure different
dynamics of an automobile and active force actuators in the parameters such as the displacement or the velocity of the
suspension can result in improved dynamics, be it ride comfort, sprung mass and the unsprung mass. Any variation in the road
better road holding or reduced body roll. In this paper, a control
surface is transmitted through the tyre to the unsprung mass
scheme has been developed for an electronically controlled
pneumatic actuator that can be used in suspensions of and then to the sprung mass of the vehicle when an active
commercial vehicles such as trucks, tractors-trailers and buses. system is absent. In an active suspension, the controller senses
The performance of the suspension system with regards to the disturbance and reacts to counteract it.
different inputs is simulated using the quarter car model and the An extensive amount of work has been done in the field of
half car model. The required active actuation force data obtained optimizing the passive suspension and adopting different
from the simulations are fed to the control scheme which then
controllers to achieve varied suspension performance in active
regulates the pressure of air in the actuator chamber according
to the desired force/pressure. This has been implemented on an suspensions. One of the primary parameters to evaluate the
experimental setup and the performance of the control scheme suspension performance is ride quality. Els [2] has
has been studied for various desired force/pressure profiles. investigated the correlation between the objective methods
used for determining the ride comfort and the subjective
Index Terms--Commercial vehicles, control, suspension. feedback received. It was concluded that any of the four
methods studied, ISO 2631, BS 6841, Average Absorbed
I. INTRODUCTION Power and VDI 2057, can be used to evaluate ride comfort.

T HE suspension of an automobile supports the weight of All the four methods assign weights to vertical acceleration
the vehicle body on the axles. A well designed values based on the frequency at which it is experienced. It
suspension system should provide sufficient compliance was further concluded that even the unweighted RMS values
in the vertical direction so that minimal road roughness is of vertical acceleration correlate well to the subjective
transmitted to the passengers. Road roughness is one of the feedback.
prime excitation sources of vibration experienced in an A detailed discussion on the application of linear optimal
automobile and covers anything from pot holes to the random control theory to semi active and fully active suspension
deviations in the road profile inherited during the construction systems is given in [3, 4]. The effects of using different
and subsequent usage of the road. The traction generated by control techniques such as full state feedback, output feedback
the pneumatic tyres is a non linear function of the normal load and absolute velocity feedback, on the performance of the
that is applied to them. Thus, any variation in the normal load tuned passive, semi-active and fully active suspension have
on the tyre results in a different amount of traction available. been well studied. These control techniques have been used
This leads to lowering of driver confidence as the vehicle is with a higher order half car model to analyze its response to
said to have lost ‘predictability’. A suspension system should statistically described road inputs in [5]. It was shown that full
therefore minimize the dynamic variations in the normal load state feedback can achieve better road isolation than a passive
on the tyres. In addition to the above, the suspension should suspension and can also be tuned to achieve road holding over
also resist roll and pitch motions of the vehicle body when ride comfort if required.
taking a turn or during braking/acceleration [1]. Active suspension systems also play an important role in
Passenger cars and commercial vehicles are usually the safety of an automobile. Vehicles with high center of
equipped with the traditional passive spring and damper gravity like sport utility vehicles (SUVs) and commercial
suspension. Most active and semi-active suspension systems vehicles are prone to roll over at high cornering speeds. This
use a pneumatic/hydraulic force actuator and a variable might be a combination of the high center of gravity and
damper respectively along with a spring. The focus of this vertical road inputs like bumps. Rao [6] has modeled this roll
paper will be on using a pneumatic actuator along with a over problem using a semi car model which attempts to
damper and a spring. differentiate between these two inputs and takes corrective
action. Nieto et al. [7] have developed a model of the
The authors are with the Department of Engineering Design, Indian Institute pneumatic suspension which includes a nonlinear fluid
of Technology Madras, Chennai-600036, India (Ph: +91-44-22574705,Fax:
dynamics model, a model for the suspension stiffness,
+91-
4422574732,email:[email protected],[email protected]). damping factor and transmissibility. It was experimentally

978-1-4244-8542-0/10/$26.00 ©2010 IEEE

 2

found that the sizes of the pipe, the tank and the air spring play where
an important role in determining the suspension’s behavior. ⎡ 0 1 0 −1 ⎤
The authors concluded that the next step to improve the ⎡ z s − zu ⎤ ⎢ −k −bs bs ⎥
suspension’s behavior is to implement a control system that ⎢ z ⎥ ⎢ s 0 ⎥
Ms Ms Ms ⎥
behaves according to the excitation frequency and the value of x= ⎢ s ⎥ ,A = ⎢ ,
⎢ zu − zr ⎥ ⎢ 0 0 0 1 ⎥
the initial sprung mass. ⎢ ⎥
⎢ ⎥
In this study, a control scheme has been developed to ⎣ zu ⎦ ⎢ ks bs − kt
0 ⎥
regulate the active force in a pneumatic suspension and ⎣⎢ M u Mu Mu ⎦⎥
implemented on an experimental setup. The active force to be ⎡ 0 ⎤ ⎡0⎤
applied between the sprung and the unsprung mass is obtained ⎢ 1 ⎥ ⎢0⎥
by regulating compressed air to a chamber using an electro- ⎢ M ⎥
pneumatic regulator. Based on the difference between the B=⎢
s
⎥ and L = ⎢ ⎥.
⎢ 0 ⎥
⎢ −1⎥
desired and actual chamber pressures, the control scheme
⎢ −1 ⎥ ⎢ ⎥
provides an appropriate voltage input to the electro-pneumatic ⎢ Mu ⎥ ⎣0⎦
regulator such that this difference is minimized. ⎣ ⎦
The performance index is chosen as
II. ACTIVE SUSPENSION MODELS ⎡∞ ⎤

A. Quarter Car Model ⎢ ∫

zs2 + ρ1 ( zs − zu )2 + ρ 2 zs2 + ρ3 ( zu − zr )2 + ρ 4 zu2 )dt ⎥ , (4)
J = ⎢ ( 
⎥
⎣0 ⎦
The quarter car model (refer to Fig. 1) is the most where ρ1, ρ2, ρ3 and ρ4 are the weights attached to the different
commonly employed model of a vehicle suspension system. states [4].
The equations of motion of the same are
B. Half Car Model (Roll Model)
Mszs + bs (zs − zu ) + ks (zs − zu ) = Fa , (1)
The half car roll model consists of the sprung mass and the
Muzu + kt (zu − zr ) − bs (zs − zu ) −ks (zs − zu ) = −Fa , (2) unsprung mass connected to each other by the left suspension
where Ms, bs and ks refer to the sprung mass, the suspension and the right suspension. The tyre damping is assumed to be
damping and the suspension stiffness respectively, Mu and kt negligible. This model specifically aims at reducing the body
refer to the unsprung mass and the stiffness of the tyre roll experienced by a vehicle when it corners and hence the
respectively, zs, zu and zr refer to the displacement of the lateral acceleration, and not the road undulations, is taken as
sprung mass, the unsprung mass and the point of contact of the the input to the system.
tyre and the road respectively and Fa refers to the active force.
A dot above a variable refers to the time rate of change of that
variable. Converting the system equations to the state space
form, we get
x = Ax + BFa + Lzr , (3)

Fig. 2. Schematic of half car model (roll model).

Taking moments at the point A (refer to Fig. 2), we get
2 2
(Is + Ms.h22)α2 =−(ksl +ksr )(α2 −α1) s −(bsl +bsr )(α2 −α1) s +
4 4 (5)
Msayh2+Ms gh2α2 +(Fr − Fl ) , s
2
2 2
Fig. 1. Schematic of quarter car model. (Iu + Mu .h12 )α1 = −(ktl + ktr ) t α1 + (ksl + ksr )(α2 −α1) t 4 +
4 (6)
2
(bsl + bsr )(α2 −α1) t + Mu ay h1+Mu gh1α1 − (Fr − Fl ) t ,
4 2
where Is and Iu indicate the moment of inertia of the sprung
mass and the unsprung mass respectively about their center of
mass, which are located at a height of h2 and h1 respectively,
α1 and α2 are the roll angles of the unsprung mass and the
sprung mass respectively. The subscripts tr and tl refer to the
right tyre and the left tyre respectively and the subscripts sr
and sl refer to the right suspension and the left suspension
 3

respectively. The parameter s is the separation between the

( A − BR N) ( ) ( )
T
−1
P + P A − BR−1N + Q − NT R−1N − PBR−1BT P = 0 . (10)
points where the suspension springs have been mounted on the
chassis and t is the track width. The matrices Q, N and R need to be computed for both the
The roll angle of the sprung mass and the unsprung mass models separately by rewriting the performance indices in
and their rate of change are taken as the state variables for this terms of their respective state variables x and inputs F as
system. Converting to state space form, we obtain follows
x = Ax + BF + L a y , (7) ⎡∞ ⎤
J = ⎢ ∫ (xT Qx + 2xT NF + FT RF)dt ⎥ . (11)
⎣0 ⎦
⎡α1 ⎤ ⎡ 0 1 0 0 ⎤
⎢ α ⎥ Table II shows the values of different parameters used.
⎢ A24 ⎥⎥
where x = ⎢ 1 ⎥ , A = ⎢ A21 A22 A23
,
⎢α 2 ⎥ ⎢ 0 0 0 1 ⎥ TABLE II
PARAMETER VALUES (FOR HALF CAR)
⎢ ⎥ ⎢ ⎥
⎣α 2 ⎦ ⎣ A41 A42 A43 A44 ⎦ Ms = 4610 kg Mu= 650 kg
ks = 42 kgf/m kt= 112 kgf/m
⎡ 0 0 ⎤ ⎡ 0 ⎤ Damping coefficient = 0.35
⎢−s ⎥ ⎢m h ⎥
⎢ 2I′ s 2I ′ ⎥ ⎢ u 1 ′⎥
Iu ⎥ ⎡F ⎤ t = 2.014 m s = 0.807 m
B=⎢
u u
⎥, L = ⎢ , F = ⎢ r ⎥, h1 = 0.8 m h2 = 1.6 m
⎢ 0 0 ⎥ ⎢ 0 ⎥ ⎣ Fl ⎦ Is = 6500 kg.m2 Iu = 900 kg.m2
⎢ s ⎥ ⎢ ⎥
−s ⎢ m s h2 ⎥
⎢ 2I ′ ⎥
2 I s′ ⎦
⎣ s
⎣⎢ I s′ ⎦⎥ For the quarter car model the performance of the controller
is analyzed by studying the frequency response of the system.
A21 = − 1 ′ ⎡( ktl + ktr ) t − mu gh1 ⎤
2 2
+ ( k sl + k sr ) s
I u ⎢⎣ 4 4 ⎥⎦ Near the sprung mass natural frequency, the ride quality of the
2 2 active suspension is better than that with the passive
A22 = − 1 ′ ( bsl + bsr ) s , A23 = 1 ′ ( k sl + k sr ) s , suspension (as can be observed from Fig. 3) but this is at the
Iu 4 Iu 4
2 2 cost of slightly increased tyre deformation (see Fig. 4).
A24 = 1 ′ ( bsl + bsr ) s , A41 = 1 ′ ( k sl + k sr ) s ,
Iu 4 Is 4
2 2
A42 = 1 ′ (bsl + bsr ) s , I u′ = Iu + mu h1 ,
Is 4
A43 = − 1 ′ ⎢⎡ ( k sl + k sr ) s − m s gh 2 ⎥⎤ ,
2

Is ⎣ 4 ⎦
I s′ = I s + ms h22 .
2
A44 = − 1 ′ (bsl + bsr ) s ,
Is 4
It can be easily verified that the system is controllable and
observable. The performance index is chosen as
⎡∞ ⎤
J = ⎢ ∫ ( ρ1α 12 + ρ 2α 12 + ρ 3α 22 + ρ 4α 22 + ρ 5 FR2 + ρ 6 FL2 )dt ⎥ , (8)
⎣0 ⎦ Fig. 3. Comparison between sprung mass acceleration of passive and active
where ρ1, ρ2, ρ3 and ρ4 are the weights attached to the states and suspension.
ρ5 and ρ6 are the weights attached to the inputs.
C. Control Scheme and Simulation Results
The objective is to strike a balance between road holding
and ride comfort using the quarter car model and reduce the
body roll (roll of the sprung mass) using the half car model.
This is governed by the choice of the performance weights and
these are chosen as shown in Table I.

TABLE I
PERFORMANCE WEIGHTS
Quarter car model [10000 100 150000 100]
Half car model [16 16 40000 16 10-9 10-9]

The problem is formulated as a H2 optimal control problem Fig. 4. Comparison between tyre deflection of passive and active
suspension.
and the gain matrix K is given by [3]
(
K = R −1 BT P + N , ) (9) To analyze the effect of lateral acceleration on body roll, its
where P is determined by solving the following Riccati variation is studied in the time domain. The input given to the
equation half car model is the real time lateral acceleration experienced
 4

by a car, undergoing a Double Lane Change (DLC) maneuver. system will be referred to as the pneumatic actuator from here
Since this measurement was taken on a car, these values are on. It is assumed that the suspension of a commercial vehicle
considerably higher than what would be experienced by a will have a base spring which will take up the static deflection
commercial vehicle. Hence, the measured lateral acceleration due to the weight of the vehicle body and that this actuator is
is reduced by a factor of 10 to compensate for the same. It can an add-on to the suspension.
be observed from Fig. 5 and Fig. 6 that an active force
actuator can considerably reduce the sprung mass roll.

Fig. 7. A schematic of the experimental facility.

Fig. 5. Comparison between active and passive IV. FORCE ACTUATOR MODEL
sprung mass roll angle for DLC input.
The pneumatic force actuator has been modeled using a
lumped parameter approach. The objective of the model is to
correlate the voltage applied (Vreg) to the electro-pneumatic
regulator to the pressure that is achieved in the chamber and
correspondingly to the force generated.
The area of the electro-pneumatic regulator opening ( AP )
is modeled as

AP =
(
k 90Vreg + Patm − Pb ), (12)
90Vreg + Patm
where k is a constant whose value is obtained from
experiments, Pb is the pressure in the chamber and Patm is the
atmospheric pressure. It is assumed that air behaves as an
ideal gas, the flow across the EPR is isentropic and the fluid
Fig. 6. Comparison between active and passive
unsprung mass roll angle for DLC input.
properties are uniform across any cross-section normal to the
direction of flow. From the balance of mass, the mass flow
rate across the EPR and into the chamber is given by
III. THE EXPERIMENTAL SETUP  b = ρuAP ,
m (13)
The experimental setup (refer to Fig. 7) consists of a where ρ is the density of air and u is the velocity of the air at
compressor connected to a storage reservoir with a capacity of the exit of the regulator.
90 ltrs. The maximum gauge pressure of air supplied by the Using the balance of energy, we obtain
compressor is 900 kPa. The compressed air is transmitted to 1
an Electro Pneumatic Regulator (EPR) via a mist separator. hb + u2 = ho , (14)
The EPR meters out the compressed air from the storage 2
reservoir to the chamber. This regulator is operated by where ho and hb denote the stagnation enthalpy at the entrance
providing an analog voltage between 0-10 V DC. A pressure and the exit of the EPR respectively. Assuming the specific
sensor is mounted at the entry point of the chamber to measure heat to be constant, and using the ideal gas equation of state
the pressure of air. The sensors and the regulator are and the isentropic process relation, we obtain
interfaced with a computer through a PCI (Peripheral ⎛ γ −1 ⎞
Component Interconnect) mounted Data Acquisition board. 2γ Po ⎜ ⎛ Pb ⎞ γ ⎟
u= 1− ⎜ ⎟ , (15)
The chamber consists of an outer shell and a diaphragm. γ −1 ρo ⎜⎜ ⎝ Po ⎠ ⎟⎟
The diaphragm is connected to a linear potentiometer to ⎝ ⎠
measure the stroke length. The metered compressed air from where γ is the ratio of specific heats, Po and Pb denotes the
the EPR enters the chamber and applies pressure on the pressure at the entrance and exit of the EPR respectively.
diaphragm. Thus air pressure is converted into a force and this Substituting (15) into (13), we obtain
 5

⎛ 2 γ +1 ⎞ each instant of time. Then the control scheme computes the

2γ 1 Pb ⎜ γ
Pb γ ⎟ necessary voltage input to the EPR and the same is provided
 b = Po APCD
m ⎜ 2 − γ +1 ⎟, (16) to the EPR through the data acquisition system. The values of
γ −1 RTo ⎜ γ ⎟
⎜ Po Po γ ⎟⎠ the various parameters are provided in Table III.
⎝
where CD is a discharge coefficient used in order to TABLE III
PARAMETER VALUES
compensate for losses that occur during the flow and R is the
To = 298 K Ab = 0.0129 m2
gas constant. The ideal gas equation of state can be written as
γ = 1.4 R = 287 J/(kg K)
PbVb Cd = 0.82 k = 0.00025
mb = . (17)
RTb Patm = 101356 Pa C = 400
Differentiating this equation with respect to time and using the
isentropic process relations, we get
γ −1 γ −1
1
1 V P ⎛ P ⎞ γ P γ P γ V
 b = b b ⎜⎜ o ⎟⎟ + b o b ,
m (18)
γ RTo ⎝ Pb ⎠ RTo
where Vb is the volume of air inside the chamber. Since the
volume of the chamber is the product of the chamber area and
the suspension deflection, we get
Vb = Ab ( zs − zu ) . (19)
This model of the system will now be used to develop a
control scheme to regulate the pressure in the chamber.
Fig. 8. Desired and actual chamber pressure for an
impulse input applied to the quarter car model.
V. CONTROL SCHEME
The difference in the actual chamber pressure and the
desired pressure (e(t)) is given by
e(t) = Pbdes (t) − Pb (t), (20)
where Pbdes(t) is the desired pressure in the chamber.
The evolution of e(t) with respect to time is chosen to
behave according to the following equation:
e(t ) = Pbdes (t) − Pb (t) = −Ce , (21)
where C is a positive real number. This ensures that the error
decays to zero asymptotically. From the above two equations,
we get
Pb (t) = Pbdes (t ) + C(Pbdes (t) − Pb (t)) . (22) Fig. 9. Desired and actual chamber pressure for an
Using this equation and the model equation, we get the desired impulse input applied to the half car model.
area of the regulator opening to be
γ −1 1 γ −1
1 Vb Pb ⎛ Po ⎞ γ γ
(23)
( Pbdes (t ) + C ( Pbdes (t ) − Pb (t ) ) ) + Pb RTPo Ab ( zs − zu )
γ
⎜ ⎟
γ RTo ⎝ Pb ⎠ o
AP = .
⎛ 2 γ +1 ⎞
γ
2γ 1 ⎜ Pb Pb γ ⎟
Po CD ⎜ − γ +1 ⎟
γ − 1 RTo ⎜ 2 γ
⎜ Po Po γ ⎟⎟
⎝ ⎠
Using (12), the voltage input to the EPR corresponding to this
desired area can be calculated and provided to the same.

VI. EXPERIMENTAL IMPLEMENTATION

The control scheme is implemented on the experimental
setup so that the desired chamber pressure is realized Fig. 10. Desired and actual chamber pressure for a
step input applied to the quarter car model.
physically. The desired force from the actuator is converted to
a desired pressure by dividing the former by the area of the
actuator. The actual chamber pressure is measured by the
pressure sensor and is compared with the desired pressure at
 6

a sudden change in the desired chamber pressure is

encountered. It can be observed that the steady state pressure
values are achieved to a good level of accuracy.

VII. CONCLUSION
The vehicle suspension has been analyzed using a quarter
car model and a half car model. While the quarter car model
deals with ride comfort and road holding, the half car model
deals with body roll. This work focused on developing an
active force actuator for commercial vehicle suspension. A
pneumatic actuator was chosen since it can be easily
integrated in current commercial vehicles which use an air
Fig. 11. Desired and actual chamber pressure for a
rectangular input applied to the quarter car model. brake system. The force from this actuator is regulated by
controlling the pressure of air supplied to it. An EPR is used to
regulate the air to the chamber based on the voltage provided
to it. A control scheme was developed to regulate the EPR
based on the difference between the desired and actual
chamber pressures. This scheme was implemented on the
experimental setup for various inputs and the results obtained
are reasonably good.

VIII. ACKNOWLEDGMENT
The authors thank Mr. Dinesh Sonawane for his help in
conducting the experiments and Prof. R. Krishnakumar, Dr. K.
V. N. Rao and Mr. S. Jaiganesh for kindly sharing data from
double lane change maneuver experiments.
Fig. 12. Desired and actual chamber pressure for a
rectangular input applied to the half car model.
IX. REFERENCES
[1] T. D. Gillespie, Fundamentals of Vehicle Dynamics, Warrendale, PA:
Society of Automotive Engineers, Inc.,1992.
[2] P. S. Els, “The applicability of ride comfort standards to off-road
vehicles”, Journal of Terramechanics, 42 , 47–64, 2005.
[3] R. Rajamani, Vehicle Dynamics & Control, New York: Springer, 2006.
[4] T. Butsuen, “The design of semi-active suspensions for Automotive
vehicles”, Ph.D. dissertation, Dept. Mech. Eng., Massachusetts Institute
of Technology, 1989.
[5] M. M. ElMadany and Z. Abduljabbar, “Alternative control laws for
active and semi-active automotive suspensions – A comparative study”,
Computers & Structures, Vol 39, No 6, 623-629, 1990.
[6] N. V. Rao, “An approach to rollover stability in vehicles using
suspension relative position sensors and lateral acceleration sensors”,
M.S. thesis, Dept. Mech. Eng., Texas A&M University, 2005.
[7] A. J. Nieto, A. L. Morales, A. Gonzales, J. M. Chicharro, P. Pintado,
Fig. 13. Desired and actual chamber pressure for a “An analytical model of pneumatic suspensions based on an
DLC input applied to the half car model. experimental characterization”, Journal of Sound and Vibration, 313,
290-307, 2008.
[8] W. L. Brogan, Modern Control Theory, New Jersey: Prentice-Hall, Inc.,
The results obtained from various experimental test runs 1985.
are presented in Figs. 8-13. It can be observed that the actuator [9] A. Tewari, Modern Control Design with MATLAB and SIMULINK,
is able to track the desired pressure reasonably well. The Singapore: John Wiley & Sons, 2005.
effect of the inherent time lag associated with pneumatic
actuation can be observed in all the figures. It is particularly
pronounced when desired pressures corresponding to impulse
inputs are to be tracked (as shown in Fig. 8 and Fig. 9) where