You are on page 1of 13

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 12, NO.

1, FEBRUARY 2007 85
Mathematical Modeling and Nonlinear
Controller Design for a Novel Electrohydraulic
Power-Steering System
Wolfgang Kemmetm uller, Member, IEEE, Steffen M uller, and Andreas Kugi, Member, IEEE
AbstractThis paper is concerned with the mathematical mod-
eling and nonlinear controller design of a novel electrohydraulic
closed-center power-steering system. After a short introduction
to the requirements of modern power-steering systems, the con-
struction of the power-steering system under consideration will
be described. The main advantage of this construction is the en-
hancement of the overall energetic efciency of the steering system
and the possibility of a variable steering assistance while keeping
the good steering feeling of traditional hydraulic steering systems.
Based on a detailed mathematical model of the essential compo-
nents of the system, it is shown how some of the parameters of the
system can be chosen to achieve an optimal dynamic behavior of
the closed-loop system. The controller design proceeds in two steps:
rst, a nonlinear controller for the assistance force based on the
differential atness property of the system is designed, and after-
wards, a steering torque controller using an impedance matching
design is derived. Finally, measurement results of a test stand show
the good performance and the robust behavior of the proposed
control strategy.
Index TermsMathematical modeling, nonlinear control,
power-steering.
I. INTRODUCTION
E
VERY modern passenger car uses a power-steering sys-
tem to assist the driver. Traditionally, these power-steering
systems are constructed in the form of a hydraulic assistance
system comprising a torsion bar and a rotary valve. It is well
known that due to the open-center design of the rotary valve,
the hydraulic assistance system has considerable losses, in case
of driving situations with no steering also. Furthermore, the
performance of the assistance is dened by the construction
and cannot be changed. In the past years, the increasing de-
mands on driving comfort and ride stability, tighter interaction
of various assistance systems in cars, and tighter restrictions
on fuel consumption have necessitated the development of new
power-steering systems [17]. Much effort has been made in the
development of steer-by-wire systems (see, e.g., [1], [18], [19],
and [24]), which are basically characterized by the fact that the
mechanical connection between the steering wheel and the tires
(i.e., the steering column) is replaced by an electric connection.
Manuscript received December 21, 2005; revised June 6, 2006. Recom-
mended by Technical Editor J. R. Wagner.
W. Kemmetm uller and A. Kugi are with the Chair of System Theory
and Automatic Control, Saarland University, D-66041 Saarbr ucken, Germany
(e-mail: andreas.kugi@lsr.uni-saarland.de; wolfgang.kemmetmueller@lsr.uni-
saarland.de).
S. M uller is with BMW-AG, 80335 Munich, Germany (e-mail: steffen.mc.
mueller@bmw.de).
Digital Object Identier 10.1109/TMECH.2006.886257
Even though these approaches will be used in the up-and-coming
steering systems, they still have a drawback that the electric sys-
tem might fail, and the safety issues are not satisfyingly solved.
Therefore, all steering systems incorporated in passenger cars
in near future will still have a mechanical connection in the
form of a steering column. The most important representative,
already introduced in the market in recent years, is the electric
power-steering (EPS) system [6], [10], [22], [23], [25], [26].
This system uses an electric motor to assist the driver. Opposite
to its advantages, namely a high energetic efciency and an easy
possibility to control the system, the EPS systemis known to ex-
tremely stress the electrical supply system of the vehicle, which
is a problem especially for heavy cars like luxury-class cars.
Alternative to the EPS systems, a large number of articles
dealing with the construction and control of electrohydraulic
steering systems have been published in past years. In [1], [19],
and [24], the usage of an electrohydraulic actuator in the steer-
by-wire systems is described. Even though variable assistance
characteristics can be perfectly realized by these systems, ques-
tions concerning the energetic efciency or safety issues in case
of the failure of the actuator are barely discussed. In some pas-
senger cars, electrohydraulic steering systems are used, where
the pump is driven by an electric motor. Although the assistance
characteristics can be adjusted on a limited scale with this mea-
sure, this construction does not provide the variability of, e.g.,
an EPS system. Furthermore, the electric motor again stresses
the electric supply system, and thus, this system may be prob-
lematic when used in the luxury-class cars considered in this
paper. Finally, the energetic efciency of a classical hydraulic
steering system is sometimes improved by using closed-center
rotary valves. Besides the fact that the assistance characteristics
cannot be changed at all, this construction is not capable of pro-
viding a suitable steering feeling. Therefore, this construction
is used only in motor sports applications.
This paper is concerned with the development of a novel
electrohydraulic closed-center power-steering (EHPS) system
by combining the advantages of an EPS system concerning the
energetic efciency and the variable steering assistance with that
of a hydraulic power-steering system concerning the simplicity
to provide the necessary energy via a pump that is driven by
the combustion engine [12], [17]. For this purpose, two essen-
tial modications compared to a conventional hydraulic power
steering system were made: 1) The hydraulic assistance cylin-
der is controlled by two electromagnetic, closed-center valves
that can be controlled independently. This allows to realize the
desired freedom in the choice of the assistance characteristics.
1083-4435/$25.00 2007 IEEE
86 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 12, NO. 1, FEBRUARY 2007
2) The hydraulic supply is built up in the form of a charge
circuit for a hydraulic accumulator that, in combination with
the closed-center construction of the valves, guarantees a high
energetic efciency. Moreover, the electric supply system is not
stressed since the pump is directly driven by the combustion
engine.
The main intention of the development of the new EHPS sys-
tem is to enhance the overall energetic efciency of the steering
system and to enable a variable steering assistance while keep-
ing the good steering feeling of traditional hydraulic steering
systems. Here, we will focus on the mathematical modeling and
control of the actual steering system, whereas the description of
the hydraulic pressure supply will be kept to its necessary min-
imum. In particular, it will be shown that some of the system
parameters can be advantageously used to improve the over-
all system performance. Moreover, it will be emphasized that
the demands on the closed-loop system can only be achieved
by including the essential nonlinearities of the system into the
controller design.
The paper is organized as follows. In Section II, we introduce
the EHPS system and its components. Section III describes the
mathematical modeling of the system, with a focus on the hy-
draulic assistance system and the mechanical steering system.
Section IVincludes the analysis of the model and some informa-
tion on the choice of the parameters of the system. The controller
design, performed in Section V, starts with the denition of the
control objective. The controller has a cascaded structure with a
atness-based controller for the assistance force in the inner loop
and a steering torque controller based on an impedance match-
ing design in the outer loop. Section VI is concerned with the
implementation of the proposed control strategy on a test stand.
Furthermore, measurement results for typical driving situations
are provided. The paper closes with Section VII, where a sum-
mary and a short outlook to further research activities is given.
II. ELECTROHYDRAULIC STEERING SYSTEM
A. Principal Concept
Nowadays, nearly all steering systems incorporated in pas-
senger cars are a combination of a mechanical steering system
with an assistance system. This conguration is mainly used for
safety reasons since the mechanical steering system can be used
as a backup system in case of a failure of the assistance system.
In Fig. 1, a schematic of the electrohydraulic steering system
under consideration is given [12]. It is composed of four sub-
systems: 1) a mechanical steering system comprising a steering
wheel, a steering column, a torsion bar, a steering gear, a steering
rack, and joints in the wheel suspension; 2) an electrohydraulic
assistance system consisting of an assistance cylinder (volume
ows q
a
and q
b
into chambers) and two steering valves (pres-
sure control valves); 3) a hydraulic power supply whose main
components are a belt-driven constant displacement pump, an
electronically controlled inlet orice and bypass valve, and a
hydraulic accumulator; 4) an electronic control unit (ECU) with
related sensors for pressures in the two chambers p
a
and p
b
, as
well as the supply pressure p
s
, the steering wheel angle
sw
,
and the steering torque
sc
.
Fig. 1. Schematic of the electrohydraulic steering system.
The basic functionality of this system can be summarized
as follows. If the driver turns the steering wheel, the torsion
bar is twisted and a steering torque is generated that moves
the steering rack. Depending on the change in the steering rack
position s (velocity w = s), the direction of travel of the vehicle
changes, which causes a change in the rack force f
rack
. With
no servo force applied by the assistance cylinder, the rack force
would be equal to the steering force f
sc
that is equivalent to the
steering torque
sc
= f
sc
/i, with the gear transmission ratio i.
In order to assist the driver and to provide a good steering
feeling, a certain amount of the rack force is compensated by the
servo force f
servo
= (p
a
p
b
)A
ac
generated by the assistance
cylinder, with A
ac
as the effective piston area.
B. Hydraulic Power Supply
The requirements for the volume ow and supply pressure of
the hydraulic power supply dramatically change depending on
the actual driving situation. Typically, high supply pressures are
necessary during parking maneuvers and for low vehicle speed
whereas high volume ows occur for high steering velocities,
e.g., during double change of lanes maneuvers (elk test). On
the other hand, very little volume ow and supply pressure is
necessary for driving situations with high vehicle speed, e.g.,
on a highway. Thus, the hydraulic power supply has to be able
to provide both sufcient volume ow and supply pressure even
in extreme driving situations while also being energetically ef-
cient for driving situations with low volume ows [17].
In this paper, we use a hydraulic power supply based on a
hydraulic accumulator. The accumulator is charged (via a check
valve) by means of a constant displacement radial piston pump.
The basic idea is to use the accumulator as a storage device for
hydraulic energy. The control concept can be summarized as
follows: If the supply pressure sinks beyond a minimum supply
pressure p
s,min
, the pump is activated, and it recharges the accu-
mulator up to a maximum supply pressure p
s,max
. Afterwards,
the pump has to be deactivated, which is done by means of an
electronically controlled bypass valve. Furthermore, the pump
KEMMETM

ULLER et al.: MATHEMATICAL MODELING AND NONLINEAR CONTROLLER DESIGN 87


Fig. 2. Schematic of the steering valve (pressure control valve) [8].
is throttled by means of an electronically controlled inlet orice
to minimize the circulation losses in this situation.
The advantages of this power supply concept are a very high
energetic efciency (see [17]) while peaks in the volume ows
(due to very fast steering maneuvers) can be provided by the
hydraulic accumulator. Thus, the pump can be designed con-
siderably smaller than in conventional hydraulic power-steering
system. Furthermore, the inlet orice and the bypass valve allow
for very smooth transitions between the charging and the deac-
tivation of the pump. Finally, the limits for the supply pressure
p
s,min
and p
s,max
can also depend on other variables represent-
ing the actual driving situation. Thus, for instance, a reduction
of these limits for higher vehicle speed enhances even more the
energetic efciency of the hydraulic power supply.
III. MATHEMATICAL MODEL
A. Assistance Cylinder and Steering Valve
The steering assistance cylinder is integrated into the steering
rack, and therefore, its position and velocity are equal to those of
the steering rack (s, w). The pressures p
a
and p
b
in the chambers
a and b of the double rod cylinder, cf. Fig. 1, are given by
(see, e.g., [2] and [4])
p
a
=

V
ac
+ A
ac
s
(A
ac
w + q
a
) (1a)
p
b
=

V
ac
A
ac
s
(A
ac
w + q
b
) (1b)
where denotes the bulk modulus of the oil, q
a
and q
b
are the
volume ows provided by the steering valves, and V
ac
is the
volume of the cylinder chambers for s = 0.
Fig. 2 shows the schematic of the steering valve being used
[8]. In the following paragraphs, we will only describe valve
a, since the equations for valve b are identical. The body
of the valve comprises the ports for the supply pressure p
s
, the
chamber pressure p
a
, and the tank pressure p
t
as well as a port
for the damping orice that is connected to p
t
.
The port of the chamber pressure p
a
is made up of several
shifted drill holes of different diameters. By means of the valve
spool, which moves inside the valve body (position x
a
, ve-
locity v
a
= x
a
), it is possible to change the effective opening
areas from supply to chamber A
sa
(x
a
) and from chamber to
tank A
at
(x
a
). Since the opening characteristics of the valve
are symmetric, we choose x
a
such that A
sa
(x
a
) = A
at
(x
a
).
During the design process, the number, position, and diameter
of the drill holes were at our disposal. In the next section, it
will be shown how these parameters inuence the behavior of
the system, and how they are chosen in the design process. The
symbolic expression describing the opening characteristics of
the valve is rather complicated, and thus, not suitable for the
controller design. Therefore, we use piecewise dened polyno-
mials of order 3 to approximate the areas A
sa
(x
a
) and A
at
(x
a
).
Under the assumption of a turbulent ow, we can calculate
the volume ow into the chamber in the form q
a
= q
sa
q
at
.
Therein, the volume ows fromsupply to chamber q
sa
and from
chamber to tank q
at
are given in the form [2], [4]
q
sa
=
_
2

A
sa
(x
a
)

p
s
p
a
q
at
=
_
2

A
at
(x
a
)

p
a
p
t
(2)
where denotes the (constant) discharge coefcient and the
density of the oil.
The volume ows q
sa
and q
at
induce jet forces that bring
about a closing of the valve. While jet forces can normally be
neglected in multistage servo valves, they do play an important
role in the pressure control valves under consideration. Due to
the complex geometry of the ow path, an exact description of
this phenomenon is rather complicated. Nonetheless, an approx-
imation of the jet forces of a single control edge with opening
area A
e
and discharge coefcient is given by (see, e.g., [2]
and [4])
f
jet
= q
e
v
e
cos() (3)
where denotes the jet angle and v
e
= q
e
/(A
e
) is the mean
velocity of the uid. For turbulent ows (as in our case), the jet
angle is constant and the entire jet force of valve a due to q
sa
and q
at
can be formulated as [cf. (2)]
f
jet,a
=
_
2(q
sa

p
s
p
a
q
at

p
a
p
t
) cos(). (4)
The valve spool can be controlled by means of the magnetic
force f
mag,a
generated by a solenoid. An inner current con-
trol loop imposes the electric current i
a
on the solenoid and
assures a very fast dynamics of the electrical subsystem of the
valve. Therefore, the magnetic force can be modeled in the form
of a static nonlinear relationship with i
a
as the control input,
i.e., f
mag,a
= f
mag,a
(i
a
). The current control loop consists of
a pulse-width-modulation controlled system. Thereby, the car-
rier frequency is chosen in such a way that the induced current
ripple, and thus, the induced magnetic force ripple prevents the
valve spool from sticking. This is why stick-slip effects do not
play a role in the valve. In addition to the solenoid, the pres-
sure control valve comprises a control spool with the effective
area A
cs
. As it will be shown later, this control spool forms a
mechanically realized proportional controller with the pressure
force p
a
A
cs
acting on the valve spool. Furthermore, the reposi-
tioning of the valve spool is ensured by a spring with the force
f
c,a
= c
v
x
a
+ f
0
, where c
v
denotes the spring stiffness and f
0
the bias force.
In order to damp the motion of the valve spool, a damping
orice is implemented (cf. Fig. 2). The volume ow through the
88 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 12, NO. 1, FEBRUARY 2007
orice can be calculated in the form [4]
q
d,a
=
d
d
2
d

4
_
2

_
|p
d,a
p
t
| sign(p
d,a
p
t
) (5)
where d
d
denotes the diameter of the damping orice,
d
is
the discharge coefcient, and p
d,a
denotes the pressure inside
the damping chamber of the valve. In contrast to (2), the char-
acteristics of the ow through the damping orice varies from
laminar to turbulent. This behavior can be described by means
of the ow number [4]
=
d
d

_
2
_
|p
d,a
p
t
| (6)
with as the dynamic viscosity of the oil. The change from
laminar to turbulent ow is described by a critical ow number

crit
that is directly connected to the Reynolds number [4]. To
incorporate both types of ow into the model, the discharge
coefcient
d
is no longer considered a constant but is given by

d
=
d,max
tanh
_
2

crit
_
(7)
where
d,max
denotes the maximum discharge coefcient at a
fully turbulent ow. If we neglect the compressibility of the oil
inside the damping chamber, the volume owq
d,a
from(5) is di-
rectly related to the velocity of the valve spool in the formq
d,a
=
v
a
A
v
, where A
v
denotes the effective area of the valve spool.
Thus, we can calculate the damping force f
d,a
(v
a
) = p
d,a
A
v
by (numerically) solving (5)(7) for p
d,a
. The friction force due
to the movement of the valve spool in the valve body is mod-
eled in the form f
f,a
= d
v,v
v
a
+ d
c,v
sign(v
a
), where d
v,v
and
d
c,v
denote the viscous and the Coulomb friction coefcients,
respectively.
The overall model of the pressure control valve is completed
by the equation of motion of the valve spool of mass m
v
:
x
a
= v
a
m
v
v
a
= f
mag,a
p
a
A
cs
f
jet,a
f
c,a
f
f,a
f
d,a
. (8)
B. Mechanical Steering System
The mechanical steering system comprises a steering wheel,
a steering torque sensor, a steering column, a torsion bar, a
steering gear, and a steering rack. Furthermore, the suspension
of tires and the tires themselves are also a part of the steering
system. The steering rack includes two cardan joints that have
negligible inuence on the dynamic behavior of the system, and
thus, will be neglected. The steering torque sensor, the steering
column, and the torsion bar can be modeled as linear torsional
springs whose effective stiffness will be summarized in c
sc
. The
moments of inertia of these components are considerably small
compared to that of the steering wheel J
sw
, and can therefore,
be neglected too. With the torque induced by the driver on the
steering wheel
driver
and the steering torque

sc
= c
sc
(
sw

sr
) +
d,sw
(9)
the equation of motion for the steering wheel is given by
J
sw

sw
=
driver

sc
. (10)
Here,
d,sw
= d
v,sw

sw
+ d
c,sw
sign(
sw
) denotes the friction
torque, where d
v,sw
is the viscous and d
c,sw
the Coulomb friction
coefcient.
The systematic determination of the drivers torque
driver
is
a problem that will not be treated within this paper. On the one
hand, the driver clearly applies a torque to the steering wheel
via his/her hands and arms. On the other hand, in normal driving
situations, the driver wants the vehicle to follow a specic path
(i.e., the road) that results in the desired steering wheel angle

sw
. These considerations show that the driver tries to con-
trol the drivers torque in such a way that the desired steering
wheel angle is achieved. A mathematical model of the drivers
behavior is evidently very complex and beyond the scope of
this paper, see, e.g., [14] and references cited therein for more
details on this topic. Nonetheless, for simulation purposes, we
make the assumption that the driver is able to control the steer-
ing wheel angle fast enough. Thus, we can take the (desired)
steering wheel angle

sw
as the drivers input to the steering
system. Then, the drivers torque can simply be calculated by
means of (10). In the following paragraphs, we will design a
steering assistance strategy based on a controller for the steer-
ing torque
sc
instead of the drivers torque
driver
, since these
two quantities are equivalent in the (quasi-)stationary case and
the resulting torque due to an acceleration of the wheel
sw
J
sw
is considered negligible for characteristic steering maneuvers.
The steering torque in the steering column is transmitted to a
steering force
sc
i in the steering rack by means of the steering
gear (gear transmission ratio i). Furthermore, the servo force
f
servo
= (p
a
p
b
)A
ac
, the friction force f
fric
, and the reaction
force of the car and the road f
rack
act on the steering rack. Thus,
the model of the steering rack (mass m
sr
) reads as
s = w
m
sr
w = f
servo
+
sc
i f
fric
f
rack
. (11)
A signicant amount of the friction force arises from the seal-
ings of the assistance cylinder. With increasing pressure inside
the chambers, the sealings are more and more forced against the
piston rod and this increases the friction force. Fig. 3 depicts
the measurement results of the friction force in the assistance
cylinder for different values of the sum of the chamber pres-
sures p

= p
a
+ p
b
. It turned out that a good approximation
of the friction force can be given by a static friction model of
the formf
fric
= d
c,sr
(p

) sign(w) + d
v,sr
wwith a constant vis-
cous friction coefcient d
v,sr
and a Coulomb friction coefcient
d
c,sr
(p

) depending on the sum of the chamber pressures p

.
The dynamic modeling of the reaction force f
rack
of the car
and the road acting on the steering system has been the topic of
many recent works, see, e.g., [14] and references cited therein.
Since we do not need such a model in our control concept, we
will not go into further details here. However, for simulation pur-
poses, a simple single-track model for stable driving situations
and a special parking model for low or zero speed was used.
These models have been calibrated in extensive measurement
campaigns by BMW-AG Munich.
KEMMETM

ULLER et al.: MATHEMATICAL MODELING AND NONLINEAR CONTROLLER DESIGN 89


Fig. 3. Friction force (normalized to the maximumrack force) in the assistance
cylinder.
IV. ANALYSIS OF THE SYSTEM AND PARAMETER
SPECIFICATION
As already mentioned in Section I, we aim at designing an
EHPS system that, in contrast to the conventional hydraulic
power-steering systems, is energetically very efcient. Thereby,
the design of the steering valves plays an important role. In
general, commercially available valves are designed from a sta-
tionary or quasi-stationary point of view and they do not meet
the demands on high dynamic performance and energetic ef-
ciency at the same time. In this section, we will summarize
the basic considerations of the mechatronic design process for
dimensioning of the steering valves by also taking into account
the restrictions concerning fabrication and production costs. All
further investigations are based on the pressure control valve,
as depicted in Fig. 2. Within the design process, the choice of
the following valve parameters is at our disposal: the opening
characteristics of the valve [i.e., A
sa
(x
a
) and A
at
(x
a
), cf. (2)],
the area A
cs
of the control spool, the stiffness c
v
, the bias force
f
0
of the spring, and the diameter d
d
of the damping orice.
Let us rst analyze the inuence of the opening character-
istics A
sa
(x
a
) and A
at
(x
a
). As already mentioned before, the
ports are made up of several shifted drill holes of different
diameters, whereby the number, position, and diameter of the
drill holes are design parameters to be chosen. Clearly, an
open-center design of the valves [i.e., A
sa
(0) = A
at
(0) = 0]
would create a leakage volume ow from supply to the tank that
leads to an energetically inefcient system. From the energetic
efciency point of view, a closed-center design would be the
best choice. But since the position x
a
of the valve spool cannot
be measured, the dead zone due to the overlap areas causes
problems in the controller design. As a compromise, we de-
signed the valve in critical-center construction. It is well known
that due to fabrication tolerances, every position of a drill hole is
inaccurate to a certain degree. Thus, a position error of the drill
hole can cause a considerable leakage volume ow for the valve
spool in the position x
a
= 0, i.e., in driving situations where no
assistance is required. This leakage volume ow is known to
become larger for increasing diameters of the drill hole. On the
other hand, the minimum diameter of the drill holes is limited
Fig. 4. Opening characteristics of the steering valve [8].
due to the manufacturing process. In order to take into account
all these aspects, we designed the critical-center valve in such
a way that the opening characteristics at x
a
= 0 are formed by
small pilot control drill holes. Furthermore, for certain steering
maneuvers (e.g., parking), the volume ows that have to be
provided by the steering valves are very high. Therefore, we
gradually increase the gradient of the opening characteristics
outside the center in order to meet the requirements concerning
the dynamic performance of the steering valve. Fig. 4 shows
the resulting opening characteristics of the valve as being used
in the nal design.
In order to study the inuence of the remaining design pa-
rameters A
cs
, d
d
, and c
v
of the valve, let us reconsider the math-
ematical model for chamber a due to the system (1a), (8) by
neglecting the Coulomb friction (d
c,v
= 0). If we assume zero
velocity of the assistance cylinder, i.e., w = w = 0 and s = s =
constant, as well as a constant magnetic force

f
mag,a
(i.e., a con-
stant input current

i
a
), then the set of equilibriumpoints is given
by
v
= { p
a
, x
a
= 0, v
a
= 0}, with
p
a
=

f
mag,a
f
0
A
cs
. (12)
From (8) and (12), it can be seen that the control spool is
a mechanically implemented proportional controller where the
effective area of the control spool A
cs
corresponds to the pro-
portional gain. Linearizing (1a) and (8) around an equilibrium
point in
v
, we get
x = F x + Gf
mag,a
(13)
with x = x x, x
T
= [p
a
, x
a
, v
a
] and f
mag,a
= f
mag,a

f
mag,a
. It is well known that the eigenvalues of the system
matrix F decide on the stability. Although the conclusions of
this analysis are only valid in a sufciently small neighborhood
around the equilibrium point due to the nonlinear nature of the
system (1a), (8), they make an important contribution to the
dimensioning of the steering valves. Since the valve is built
up symmetrically in critical-center construction (cf. Fig. 4), we
have the relations A
sa
(0) = A
at
(0) = 0 and (A
sa
/x
a
)(0) =
90 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 12, NO. 1, FEBRUARY 2007
(A
at
/x
a
)(0) = k
v
> 0. With this, the system matrix F
from (13) reads as
F =
_
_
0

V
ac
+A
ac
s
q
a
x
a
0
0 0 1

A
cs
m
v

c
v
+
f
jet, a
xa
m
v

d
v , v
+
f
d, a
va
m
v
_
_
x= x
v
(14)
where
q
a
x
a
=
_
2

k
v
(

p
s
p
a
+

p
a
p
t
) > 0 (15)
f
jet,a
x
a
= 2k
v
(p
s
p
t
) cos() > 0 (16)
f
d,a
v
a
=
A
2
v

crit

d,max
d
3
d

> 0. (17)
Here, f
d,a
/v
a
can be found by utilizing the relation
f
d,a
/v
a
= (v
a
/f
d,a
)
1
. By means of the criterion of
RouthHurwitz, it can be easily seen that a necessary and suf-
cient condition for the stability of the linearized system (13) is
given by
__
c
v
+
f
jet,a
x
a
__
d
v,v
+
f
d,a
v
a
__
x= x
v
>
_
A
cs
m
v

V
ac
+ A
ac
s
q
a
x
a
_
x= x
v
. (18)
Clearly, for all possible equilibrium points x
v
, the point
p
A
= (p
s
+ p
t
)/2, where q
a
/x
a
reaches its maximum, con-
stitutes the worst case. In this case, (18) yields
(c
v
+ 2k
v
(p
s
p
t
) cos())
_
d
v,v
+
A
2
v

crit

d,max
d
3
d

_
> A
cs
m
v

V
ac
+ A
ac
s

k
v

p
s
p
t
. (19)
Remember that the area of the control spool A
cs
represents
the gain of a mechanically realized proportional controller. A
large value of A
cs
yields a fast dynamics of the valve. On the
other hand, inequality (19) shows that increasing the value of
A
cs
may lead to an unstable system. Furthermore, as it can be
directly seen from (12), a larger A
cs
requires higher stationary
magnetic forces

f
mag,a
, and thus, a larger solenoid might be-
come necessary. This is undesirable for two reasons: rstly, a
larger solenoid has, in general, a slower dynamics, and secondly,
the space requirements for the valve become larger.
A spring with stiffness c
v
and bias force f
0
provides the
restoration of the valve spool since the solenoid can only produce
positive magnetic forces. The bias force f
0
was chosen in such a
way that we provide enough force reserve for the movement of
the valve spool in both directions over the whole operating range.
For a given solenoid with maximumstationary force

f
mag,a,max
,
we can conclude from (12) that the maximum driving force of
the valve spool takes the value

f
mag,a,max
p
a
A
cs
f
0
and
the maximum restoring force is given by p
a
A
cs
+ f
0
at x
a
= 0.
Here, we have to face two extreme situations where p
a
= p
t
and p
a
= p
s
, respectively. Thus, in order to symmetrize the
force reserve of the valve over the operating range, we choose
the bias force f
0
in the form
f
0
=

f
mag,a,max
(p
s
+ p
t
)A
cs
2
. (20)
The spring stiffness c
v
was selected such that a stable operation
of the valve can also be guaranteed in the case of vanishing jet
forces, cf. (19).
Finally, the design criteria for the diameter d
d
of the damping
orice will be discussed. Obviously, an increase of the damp-
ing (i.e., a decrease of the diameter of the damping orice)
incorporates an enhancement of the stability of the valve due to
(19). On the other hand, the dynamics of the valve is lowered
by this measure. A closer inspection shows that we can make
the diameter of the damping orice considerably high without
getting stability problems. Thus, we obtained a valve with less
hydraulic damping compared to a traditional valve design but
we could improve the dynamic performance.
Summarizing, the quantitative analysis of the valve parame-
ters leads to a steering valve that differs signicantly from the
valve design based on the traditional steady-state considerations.
Within the realms of possibility, in particular, in terms of limi-
tation in the choice of the parameters, we were able to construct
a steering valve that turns out to be an optimum compromise
between the demands on low leakages and on high dynamic
performance, and thus, serves as an appropriate actuator for the
power-steering system.
V. NONLINEAR CONTROLLER DESIGN
This section is concerned with the controller design for the
EHPS. The main goal of the controller design is to provide
a good steering feeling to the driver. In order to ensure this,
it is necessary to give, on the one hand, a sufcient steering
support and, on the other hand, an adequate feedback of the
actual driving condition (road condition) to the driver. The pro-
posed controller is based on a cascaded concept with a atness-
based feedforward controller in the inner loop and a mechanical
impedance controller in the outer loop.
A. Assistance Force Controller
The subsystem (1a), (8) describing the pressure buildup
in chamber a of the assistance cylinder contains essen-
tial nonlinearities due to the opening characteristics A
sa
(x
a
)
and A
at
(x
a
) (cf. Fig. 4), the volume ow q
a
(x
a
, p
a
)
[cf. (2)], the damping force f
d,a
(v
a
) [cf. (5)(7)], the jet force
f
jet,a
(x
a
, p
a
) [cf. (4)], and the magnetic force f
mag,a
(i
a
). In
order to meet the demands of the steering torque controller, it is
mandatory to control the chamber pressure p
a
as fast as possi-
ble. This can be achieved only if the nonlinearities of the system
are systematically included in the controller design. In our case,
we advantageously use the differential atness property of the
system, where
1
= p
a
serves as a possible at output (see,
e.g., [5] and [20] or [21] for an introduction to the notion of
atness). Roughly speaking, differential atness ensures that,
given a at output
1
, the state x
T
= [p
a
, x
a
, v
a
] and the in-
put f
mag,a
can be parametrized by the at output and its time
derivatives. Furthermore, this design allows us to systematically
KEMMETM

ULLER et al.: MATHEMATICAL MODELING AND NONLINEAR CONTROLLER DESIGN 91


include the (known) exogenous variables s(t) and s(t) = w(t)
in the controller design.
Let

1
= p

a
(t) C
3
denote the three times continuously dif-
ferentiable desired trajectory for the chamber pressure p
a
. The
rst derivative with respect to time is given by [cf. (1a)]

1
=

2
=

V
ac
+ A
ac
s
(A
ac
w + q
a
) (21)
with q
a
= q
a
(

1
, x

a
) given by (2). Since the opening character-
istics of the valve according to Fig. 4 are approximated by means
of piecewise dened polynomials of order 3, an analytical so-
lution of (21) for the unknown x

a
is rather unwieldy. Thus, the
unique solution of this equation is calculated numerically in real
time during the control process. For the sake of convenience,
let x

a
=

1
(

1
,

2
, s, w) denote the solution of (21). Further
differentiation gives

2
=

3
=

2
+

2
x

a
x

a
+

2
s
w +

2
w
w (22)
where x

a
= v

a
is the velocity of the valve spool. For a con-
densed notation, we will henceforth use the abbreviations
= [
1
,
2
,
3
] and = [s, w, w, w]. Then, v

a
=

2
(

, s, w, w) denotes the solution of (22). The third


derivative

3
=
2

i=1

i+1
+

3
x

a
v

a
+

3
v

a
v

a
+
3

i=1

i+1
=
_

, f

mag,a
,
_
(23)
nally includes the input, since due to (8), we have
m
v
v

a
= f

mag,a

1
A
cs
f
jet,a
(

1
, x

a
)
c
v
x

a
f
0
f
f,a
(v

a
) f
d,a
(v

a
) . (24)
Combining (23) and (24) and solving for f

mag,a
, we get the
atness-based feedforward control in the form
f

mag,a
=

3
_

3
,
_
. (25)
If the mathematical model is exact and the desired trajectory

1
(t) = p

a
(t) of the at output is consistent with the initial
conditions of the system, then the output p
a
(t) exactly tracks
the desired output. Moreover, by a simple change of coordinates,
the system (1a) and (8) with the feedforward control for valve
a due to (25) can be written in Brunovsky form [9], [21]

1
=
2
,

2
=
3
,

3
=

3
. (26)
In order to cope with the model uncertainties and inconsistent
initial conditions, we have to incorporate a feedback control to
stabilize the tracking error e =

. Therefore, we extend
the feedforward control (25) by a feedback part (e), which
yields
f
mag,a
=

3
_

3
(e),
_
. (27)
Then, the dynamics of the tracking error is given by
e
i
= e
i+1
, i = 1, 2
e
3
=
_
e +

, f
mag,a
_

3
(e),
_
,
_

3
. (28)
In [7], it is shown that a linear state-feedback controller with
integral part given in the form
(e) =
0
_
t
0
e
1
() d +
3

i=1

i
e
i
(29)
with appropriate coefcients
i
> 0, i = 0, . . . , 3, can always
be used to stabilize the error dynamics (28). Obviously, the
feedback control (29) requires the measurement of the full state,
which is not available in our application since only the at output

1
= p
a
can be measured. Nevertheless, subsequently, it will be
shown that a simple integral feedback control
(e) =
0
_
t
0
e
1
() d (30)
sufces to stabilize the error dynamics.
For the proof of stability of the closed-loop system, let us
rst introduce the extended tracking error e
e
= [e
0
, e
1
, e
2
, e
3
]
T
with e
0
=
_
t
0
e
1
() d. Furthermore, the desired trajectory

1
and its time derivatives as well as the exogenous inputs
are combined in (t) = [

1
,

2
,

3
,

3
, s, w, w, w]
T
. Then, the
closed-loop error dynamics (28) extended by e
0
= e
1
reads
as
e
e
= (e
e
, (t)). (31)
The subsequent proof of stability is based on the results of
Kelemen [11], [16], as elaborated in [7]. Summarizing, we can
state that if system (31) meets the following prerequisites:
A1: : R
4
R
8
R
4
is of class C
2
with respect to its
arguments;
A2: there is a bounded set R
8
and a continuously
differentiable function : R
4
such that for each
constant input value

, ((

),

) = 0;
A3: there is a > 0 such that for each

, the eigenvalues
of /e
e
((

),

) have real parts not greater than ;
then the following theorem holds (cf. [16], [11], [7]).
Theorem 1: There is a

> 0 such that for all (0,

] and
T > 0, there exists
1
(),
2
(, T) > 0 for which the following
property holds. If a sufciently continuously differentiable (t)
satises (t) , T t
0
||e
e
(t
0
) ((t
0
)) <
1
(32)
and
1
T
_
t+T
t

()d <
2
, t t
0
(33)
then, the corresponding solution e
e
of (31) satises
e
e
(t) ((t
0
)) < , t t
0
(34)
92 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 12, NO. 1, FEBRUARY 2007
TABLE I
BOUNDARIES FOR THE DESIRED TRAJECTORIES

AND EXOGENOUS INPUTS


that is, system (1a), (8) is stable under the tracking control law
(27), (30). Moreover, if in addition
lim
t

(t) = 0, lim
t
(t) =

(35)
then the corresponding solution of (31) satises
lim
t
e
e
(t) = (

). (36)
Furthermore, if for some T
1
> t
0
, (t) =

for all
t > T
1
, then e
e
(T
1
) is in the domain of attraction of the ex-
ponentially stable equilibrium ((

),

), that is, the system


(1a), (8) is asymptotically stable under the tracking control law
(27), (30).
It can be immediately seen that condition A1 is fullled triv-
ially. If the desired at output
1
is chosen to be sufciently
smooth, then the set is bounded since the boundedness of
the exogenous input obviously follows from physics. Further-
more, in condition A2, the function is given by
(

) = 0. (37)
The prerequisites A1 and A2 together with the smoothness as-
sumption on

1
also imply that (33) is satised for a suitable
choice of
2
. Thus, the closed-loop system is stable if: 1) the
initial tracking error e
e
(t
0
) is sufciently small [cf. (32)] and 2)
the real part of the eigenvalues of

e
e
((

),

) (38)
are all less than 0 for all

. Even though an analytical de-
termination of the eigenvalues of (38) is impossible, we can
numerically calculate the eigenvalues for each

. It turns
out that the choice of
0
= 50 10
9
in (30) yields a stable
closed-loop system for p
s
= 160 10
5
Pa, p
t
= 0 Pa, and all
(t) of practical interest, i.e., for desired trajectories and exoge-
nous inputs keeping the limits summarized in Table I.
Finally, using the last part of Theorem 1, we can show that
for constant desired trajectories and exogenous inputs, i.e.,

1
= constant and s = constant, the closed-loop system is
asymptotically stable.
Remark 1: The position s of the assistance cylinder and its
time derivatives, used in the above control strategy, cannot be
measured in the car. For this reason, we use an estimation of s
based on the steering wheel angle
sw
and the steering torque

sc
in the form s = (
sw

sc
/c
sc
)/i. Due to a high quality of
the sensors used in the car, we are able to derive the velocity
of the assistance cylinder by means of approximate numerical
differentiation. The inuence of the higher derivatives included
in the control law is comparably small in our application, and
can therefore, be neglected in the nal application.
Fig. 5. Trajectories for the chamber pressures p
lim
= 10 bar and p
min
=
6 bar.
The controller for chamber b is completely equivalent to
that of chamber a, if s is replaced by s, and consequently,
w by w. In order to complete the inner control loop, the
desired servo force f

servo
= A
ac
(p

a
p

b
) determined by the
outer control loop has to be subdivided into two trajectories for
the chamber pressures p

a
and p

b
, respectively. It is immediately
clear that by the choice of f

servo
, the chamber pressures are not
xed. Thus, the trajectory planning is based on the following
considerations:
1) The chamber pressures have to be lower than the supply
pressure p
s
and higher than the tank pressure p
t
.
2) In Section III-B, it has been shown that the friction in-
side the assistance cylinder increases with the sum of the
chamber pressures p

(cf. Fig. 3). Thus, in order to keep


the undesired friction as low as possible, p

has to be kept
as small as possible.
3) The resulting trajectories for chamber pressures have to
be sufciently smooth.
A possible solution used in this paper is given by
p

a
=
1
2
_
tanh
_
p

p
lim
_
+ 1
_
p

+ p
min
p

b
=
1
2
_
tanh
_
p

p
lim
_
1
_
p

+ p
min
(39)
where p

= f

servo
/A
ac
denotes the difference pressure and
p
min
and p
lim
serve as parameters to adjust the minimum cham-
ber pressure and the shape of the trajectories at p

= 0. Fig. 5
shows the trajectories for p
lim
= 10 bar and p
min
= 6 bar.
B. Steering Torque Controller
In the outer control loop, the steering torque controller is
responsible for the steering feeling. By the termsteering feeling,
we basically mean the feedback provided by the steering system
to the driver due to the driving condition and his/her steering
maneuvers. A quantitative denition of a good steering feeling
based on the measurement results is extremely difcult. This
KEMMETM

ULLER et al.: MATHEMATICAL MODELING AND NONLINEAR CONTROLLER DESIGN 93


is why the quality of the power-steering system is still mainly
evaluated by the test drivers.
A requirement placed for every power-steering system is to
provide sufcient support to the driver in all driving situations.
This is especially critical for low vehicle velocities or during
parking maneuvers, as here, the highest rack forces do occur.
Via the steering torque, the driver also has to experience suf-
cient information about the driving situation mainly concerning
the road conditions. All this should be ensured without lower-
ing the comfort of the steering system. Thus, in modern steering
systems, the level of assistance and therewith the level of feed-
back is adjusted to the actual driving situation. Furthermore, in
modern cars, a tighter interconnection of different driving assis-
tance systems of the car (e.g., parking assistance) takes place.
Therefore, the power-steering system developed in this paper
should also provide the possibility of an interlink to these assis-
tance systems. Naturally, the most important requirement for the
power-steering system is to guarantee the stability in all driving
situations.
In our paper, we will tackle the steering torque controller
design by means of a mechanical impedance matching task
[3], [15]. This approach has the advantage that the solution of
providing an appropriate steering feeling can be found almost
independently from the stability issue. For this purpose, let us
rst formulate the transformed equations for the steering rack
due to (9)(11) in new coordinates =
sw

sr
,
sr
= is
=
=
sw

i
2
m
sr
(
servo
+
sc

load
) (40)
with the servo torque
servo
= f
servo
/i and the load torque

load
= (f
fric
+ f
rack
)/i including both the rack force f
rack
and
the friction force f
fric
. The basic idea is that (40) is controlled by
the control input
servo
, which is realized by the inner assistance
force controller according to the previous section, such that in
response to a load torque
load
, the controlled closed-loop sys-
tem acts in the same way as a desired mechanical impedance
system. In the rst view, it seems somehow odd that we address
the steering feeling issue by an impedance system taking the
load force
load
and not the steering torque
sc
as an input. But
we will show later how this is linked to the control of the steer-
ing torque. The desired impedance system can, then, be written
as
=
=
i
2
m
sr
(
c
() +
d
()
load
) (41)
where
c
() can be regarded as the desired (nonlinear) spring
characteristics and
d
() as the desired (nonlinear) damp-
ing characteristics. Therein, both the spring and the damping
characteristics have to be sufciently smooth skew symmet-
ric functions strictly monotonic increasing in their arguments.
Thus, the conditions
c
() > 0,
d
() > 0 for all
, R are satised and the functions
c
and
d
are
invertible, which will be utilized later.
Remark 2: Note that the design of the desired impedance
system in the difference angle =
sw

sr
is crucial. If
we would try to enforce an impedance behavior in
sr
, it can be
immediately seen that the drivers demand would be completely
canceled out. If, on the other hand, we would try to design
an impedance system in the steering wheel angle
sw
, all load
forces, and therefore, all information from the road is canceled
out, which is also undesirable.
Comparing (40) and (41), we get the matching condition in
the form

servo
=
m
sr
i
2

sw

sc
+
c
() +
d
(). (42)
It can be seen in (41) that the steering torque
sc
no longer
appears explicitly. The drivers control reenters the system via
the difference angle , which can be written as =
sc
/c
sc
provided that the friction torque of the steering wheel
d,sw
is
assumed to be zero, cf. (9).
If we now assume that the underlying assistance force con-
troller is sufciently fast, we can set
servo
=

servo
and the
closed-loop system (40), (42) exactly matches the desired
impedance system (41). The stability of (41) for
load
= 0 can
be veried easily using the Lyapunov function
V =
1
2
m
sr
i
2

2
+
_

0

c
( ) d . (43)
We, then, have

V =
d
(), which is clearly negative
semidenite, and thus, the closed-loop system is stable in the
sense of Lyapunov. The asymptotic stability directly follows
from the invariance principle of KrassovskiiLaSalle [13].
Let us nowanalyze the overall control systemconsisting of the
steering torque controller and the assistance force controller. As
a matter of fact, in reality, the dynamics of the assistance force
controller is always limited, so that the actual and the desired
values of the assistance torque differ,
servo
=

servo
. In order to
understand the basic features of the cascaded control concept, let
us assume that the dynamics of the assistance torque controller
can be described by means of a rst-order LTI system

servo
=
1
T
servo
(
servo

servo
) (44)
with a time constant T
servo
. The set of equilibrium points
of system (40) and (44) with

servo
equal to the right-
hand side of (42) for
sw
=
sw
= constant and
load
=

load
= constant is given by
i
= { =
1
c
(
load
), = 0,

servo
=
c
()
sc
} with
sc
= c
sc
. Thus, the steering
torque
sc
can also be written in the form

sc
= c
sc

1
c
(
load
). (45)
Without loss of generality, but for simplifying the presentation,
let us consider a linear spring
c
() =
c
, with a stiffness
coefcient
c
and linear damping characteristics
d
() =

d
, with
d
as the damping coefcient. Consequently,
sc
from (45) is, then, given by

sc
=
c
sc

c

load
. (46)
94 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 12, NO. 1, FEBRUARY 2007
Fig. 6. Overall structure of the proposed control strategy.
Under these assumptions, the system (40)(44) becomes linear
and a simple analysis shows that the inequality condition

d
> T
servo
(
c
c
sc
) (47)
is necessary and sufcient for the stability. The values of
d
that can be realized are limited from above since measurement
noise deteriorates the quality of . A critical case for stability
obviously occurs when a high steering support is demanded,
i.e.,
sc

load
, and with this, a high stiffness of the desired
spring is necessary, i.e.,
c
c
sc
. Thus, a high steering support
in combination with the limited dynamics of the assistance force
and measurement noise causes an unstable system (see [6] for
similar observations).
In order to circumvent this problem, let us assume in the
rst step that the load torque
load
(t) were known. The steering
torque controller (42) is, then, augmented by a disturbance feed-
forward part

(
load
(t)), where the condition |

(
load
(t))|
< |
load
(t)| holds for all times t. Hence, we have

servo
=
m
sr
i
2

sw

sc
+
c
() +
d
() +

(
load
).
(48)
If we, now, calculate the set of equilibrium points of the sys-
tem (40) and (44) with

servo
from (48) for
sw
=
sw
= con-
stant and
load
=
load
= constant, then the equilibrium steer-
ing torque
sc
of (45) has to be replaced by

sc
= c
sc

1
c
(
load

(
load
)). (49)
To gain more insight into the consequences of the disturbance
feedforward on the steering torque controller, we again as-
sume linear spring characteristics
c
() =
c
and a lin-
ear disturbance feedforward relationship

(
load
) =

load
,
0 <

< 1. Then, (49) simplies to



sc
= c
sc
(1

c

load
. (50)
Following the arguments mentioned earlier, by comparing (50)
with (46), it can be directly seen that in case of a high steer-
ing support,
c
is considerably smaller in (50) if we choose

sufciently close to 1. Thus, the stability problems due to (47)


occurring from the choice of a very stiff desired spring charac-
teristics can be solved. However, this is due to the fact that the
main part of
load
is compensated by the disturbance feedfor-
ward

(
load
) in (48) and the impedance matching controller
has to cope only with a small difference
load

(
load
). The
price we have to pay is that we additionally need the informa-
tion of the load torque
load
. Clearly, in practice, neither the
load force nor the load torque is measurable. For this reason,
we use a simple observer that determines an estimation
load
of

load
based on the stationary relationship
load
=
sc
+
servo
.
Even though this observer neglects all dynamics of the system,
it produces good results for most driving situations.
In Fig. 6, the principal structure of the resulting control strat-
egy composed of a atness-based assistance force controller,
an impedance-matching-based steering torque controller, and a
load torque observer is depicted.
Remark 3: It is worth mentioning that the impedance match-
ing design methodology is not limited to the simple case of
linear spring and damping characteristics and a linear distur-
bance feedforward, as discussed at the end of this section for
the purpose of illustration. Indeed, we can apply almost arbi-
trary (nonlinear) assistance characteristics that can also be the
functions of other variables, e.g., the vehicle velocity v
car
. Fur-
thermore, it has to be noted that the adjustment of the steering
feeling, i.e., essentially the spring and damping characteristics,
has to be performed mainly by the test drivers. Here, one ad-
ditional advantage of the proposed concept is the capability of
a simple parametrization, in particular because the functions

c
(),
d
(), and

(
load
) can be easily interpreted in
terms of mechanical quantities.
VI. IMPLEMENTATION AND MEASUREMENT RESULTS
For the calibration of the mathematical model and for the
test of the proposed control strategy, a test stand was built up
(see Fig. 7). This test stand includes all components of the EHPS
(cf. Fig. 1), whereby the pump of the pressure supply is driven by
an electric motor instead of a combustion engine. Furthermore,
KEMMETM

ULLER et al.: MATHEMATICAL MODELING AND NONLINEAR CONTROLLER DESIGN 95


Fig. 7. Photograph of the EHPS test stand.
the feedback of the road is simulated by a hydraulic actuator
generating rack forces f
rack
corresponding to a single-track
model of the car. The control concept derived in this paper is
implemented on a real-time hardware of dSPACE Company at a
sample time of 1 ms. Therefore, the whole control strategy was
implementedinCcode s-functions inMatlab/Simulinkthat were
compiled to the real-time hardware of dSPACE using the real-
time workshop of Matlab/Simulink. In addition to the standard
sensors for the steering wheel angle
sw
and the steering torque

sc
, the chamber pressures p
a
and p
b
as well as the supply
pressure p
s
, a number of sensors, e.g., for the steering rack
position s and the rack force f
rack
are included at the test stand.
These additional sensors are not used in the control concept but
serve as a basis to check the mathematical model and the quality
of the proposed control concept.
As already mentioned, the calibration and optimization of the
steering feeling, i.e., the design of the characteristics
c
,
d
,
and

is carried out by the test drivers in the test car. For this
purpose, a test car was built up (BMW5 series) that also includes
all the components of the EHPS systembut does not incorporate
the additional sensors of the test stand. Of course, here, the pump
of the pressure supply is driven by the combustion engine.
In this paper, we will present the results of measurements
performed on the test stand since the additional sensors provide
the possibility of gaining more insight into the behavior of the
steering system. We have chosen four typical driving situations
at a vehicle speed of v
car
= 50 km/h. The choice of the vehicle
speed of 50 km/h was motivated by the following reasons: 1) the
steering assistance necessary at this speed is still high (the need
for steering assistance basically decreases with vehicle speed
and is the highest for zero vehicle speed) and 2) the dynamics
of the steering maneuvers is highest at a vehicle speed in this
range compared to the rather slowdynamics occurring at parking
maneuvers and at a high vehicle speed.
Let us rst consider slow steering maneuvers (see Fig. 8,
phase 1). Such steering movements typically arise during the
driving on highways and are displayed here for rather large
steering angles. In this setting, the slow steering velocities
produce only small volume ows in the steering valves, such
that the difference pressure p = p
a
p
b
almost perfectly
tracks the desired pressure p

= p

a
p

b
. Due to the large
load torque
load
, a large amount has to be compensated by
the servo torque
servo
to guarantee for a low steering torque

sc
. Fig. 8 shows that this specication is fullled very well.
Furthermore, the measurement results for the valve currents i
a
and i
b
illustrate the control inputs.
A driving situation, primarily interesting for the evaluation of
the dynamic properties of the steering system, is the accelerated
sinusoidal steering. Thereby, the test driver tries to enforce a
sinusoidal steering wheel angle with increasing amplitude and
frequency as exact as possible. Optimally, this would result in
an almost constant amplitude of the steering torque
sc
. Phase
2 of Fig. 8 shows that this is the case for low frequencies, but
the steering torque increases with increasing frequency. The
driver experiences this behavior in the form of an increasing
hardening of the steering system, which is, of course, unde-
sirable to a certain extent. An explanation for this behavior is
that the high volume ows cause high jet forces in the steering
valves that cannot be compensated perfectly. This is, as already
mentioned in Section III-A, mainly due to the fact that the mod-
eling of the jet forces is very complicated, and thus, is always
aficted with a certain error. Nonetheless, some potential for
the improvement of performance in this point might arise from
an exacter modeling of the jet forces using, e.g., nite-element
methods.
The double change of lane, also known as the elk test, is a
test that includes very high steering wheel velocities. In this
driving situation, the comfort of the steering system is much
less important than the driving stability. Actually, here, a low
steering torque can cause problems, since then the driver tends
to oversteer. On the other hand, the steering torque must be kept
less than a certain limit to give the driver the possibility of a fast
reaction. Phase 3 of Fig. 9 depicts the results of this test case
and it can be summarized that this situation is handled efciently
with the proposed steering system. Nonetheless, a problem still
exists when the steering torque shows a little overshoot after the
steering velocity reaches its maximum, since the driver does not
expect such a reaction of the system. The reason for this can,
once again, be found in the jet forces and in the high friction of
the assistance cylinder.
The fourth and the last test presented in this paper is the
automatic restoration of the steering wheel. Thereby, the driver
turns the steering wheel to a certain steering wheel angle, and
then, releases the steering wheel. Due to the feedback of the
car, the steering wheel is turned back to a steering wheel angle
of 0

. This test is mainly used as an indicator of the friction


in the system. For example, the friction in some EPS systems
necessitates the inclusion of active restoration strategies. As it
can be seen fromthe measurement results in phase 4 of Fig. 9, in
our case, the restoration behavior is satisfactory without taking
extra measures.
Finally, we want to briey summarize the results of the mea-
surements concerning the energetic efciency of the proposed
EHPS system (see [17] for a detailed description of the speci-
cations and results of the tests performed for the evaluation of
the power consumption of the EHPS system). It turns out that
for typical driving cycles (NEDnew European driving cycle),
the energy consumption of the EHPS system is about 75% less
96 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 12, NO. 1, FEBRUARY 2007
Fig. 8. Measurements performed on the test stand for slow steering and accelerated sinusoidal steering.
Fig. 9. Measurements performed on the test stand for double change of lane and restoration maneuvers.
than the energy used by a traditional hydraulic power-steering
system.
VII. CONCLUSION
In this paper, we presented the mathematical modeling and
the (nonlinear) controller design for a novel EHPS. Starting with
a short presentation of the construction, we derived a mathemat-
ical model for the steering system, whereby the main focus was
laid on a detailed investigation of the hydraulic assistance system
and the mechanical steering system. Afterwards, it was shown
how the design parameters of the steering valves have to be cho-
sen in order to meet the demands on high dynamic performance
KEMMETM

ULLER et al.: MATHEMATICAL MODELING AND NONLINEAR CONTROLLER DESIGN 97


and energetic efciency at the same time. The task of the con-
troller design proceeded in two steps: rstly, a atness-based
(nonlinear) controller for the two chamber pressures of the as-
sistance cylinder combined with a suitable trajectory planning
was designed, and secondly, an impedance matching design
was developed for the steering torque in order to provide a good
steering feeling to the driver. The control concept is character-
ized by its easy way of parametrization by the test drivers. To
conrm the robustness of the proposed control strategy, a large
number of tests, both on the test stand and in the test car at
different vehicle speeds, were performed. It turned out that in
any of the considered test cases, the control strategy showed a
good behavior.
Further work will focus on a detailed analysis of the jet forces
in the steering valve and on alternative valve concepts. Further-
more, the friction in the assistance cylinder and the design of
improved observers for the load torque will be the contents of
the future works. Finally, the transmission of the derived control
concept to EPS systems is going to be analyzed.
ACKNOWLEDGMENT
The authors would like to thank HYDAC Electronic GmbH,
in particular F. Herold and F. Kattler, for the fruitful cooperation
and for the possibility to adjust the design of the pressure
control valve.
REFERENCES
[1] T. Acarman et al., A robust controller design for drive by wire hydraulic
power steering system, in Proc. 2002 Am. Control Conf., Anchorage,
AK, May, pp. 25222527.
[2] J. F. Blackburn, G. Reethof, and J. L. Shearer, Fluid Power Control. New
York: Wiley, 1960.
[3] C. Canudas-de-Wit et al., Fun-to-drive by feedback, Eur. J. Control,
vol. 11, no. 4/5, pp. 353383, 2005.
[4] D. McCloy and H. R. Martin, Control of Fluid Power: Analysis and
Design. New York: Wiley, 1980.
[5] M. J. Fliess et al., Flatness and defect of non-linear systems: Introductory
theory and examples, Int. J. Control, vol. 61, no. 6, pp. 13271361, 1995.
[6] O. Gramann et al., Variable Lenkunterst utzung f ur eine elektromech-
anische Servolenkung, presented at 23rd Conf. Elektronik im KFZ,
Stuttgart, Germany, 2003.
[7] V. Hagenmeyer, Robust Nonlinear Tracking Control Based on Differential
Flatness. D usseldorf, Germany: VDI-Verlag, 2003.
[8] HYDAC Electronic GmbH, Technical documentation for closed center
pressure control valve, internal communication, 2004.
[9] A. Isidori, Nonlinear Control Systems. New York: Springer-Verlag,
1995.
[10] K. Ji-Hoon and S. Jae-Bok, Control logic for an electric power steering
system using assist motor, Mechatronics, vol. 12, pp. 447459, 2002.
[11] M. Kelemen, A stability property, IEEE Trans. Autom. Control,
vol. AC-31, no. 8, pp. 766768, Aug. 1986.
[12] W. Kemmetm uller, A. Kugi, and S. M uller, Modeling and nonlinear
control of an electrohydraulic closed-center power-steering system, in
Proc. 2006 Eur. Control Conf./Conf. Decision Control, Sevilla, Spain,
pp. 50775082.
[13] H. K. Khalil, Nonlinear Systems. Englewood Cliffs, NJ: Prentice Hall,
2002.
[14] U. Kiencke and L. Nielsen, Automotive Control Systems. Berlin,
Germany: Springer-Verlag, 2000.
[15] A. Kugi and W. Kemmetm uller, Impedance control of hydraulic piston
actuators, in 6th IFACSymp. Nonlinear Control Syst., Preprints, Stuttgart,
Germany, 2004, pp. 12411246.
[16] A. L. Lawrence and W. J. Rugh, On a stability theorem for nonlinear
systems with slowly varying inputs, IEEETrans. Autom. Control, vol. 35,
no. 7, pp. 860864, Jul. 1990.
[17] S. M uller, A. Kugi, and W. Kemmetm uller, Analysis of a closed-center
hydraulic power steering for full steer-by-wire functionality and low
fuel consumption, in Proc. 9th Scand. Int. Conf. Fluid Power (CD),
Link oping, Sweden, 2005.
[18] T. J. Park, C. S. Han, and S. H. Lee, Development of the electronic control
unit for the rack-actuating steer-by-wire using the hardware-in-the-loop
simulation system, Mechatronics, vol. 15, pp. 899918, 2005.
[19] H. C. Qiu, Q. Zhang, J. F. Reid, and D. Wu, Modelling and simulation of
an electrohydraulic steering system, Int. J. Veh. Design, vol. 26, no. 2/3,
pp. 161174, 2001.
[20] J. Rudolph, Beitr age zur achheitsbasierten Folgeregelung linearer und
nichtlinearer Systeme endlicher und unendlicher Dimension. Aachen,
Germany: Shaker, 2003.
[21] H. Sira-Ramirez and S. K. Agrawal, Differentially Flat Systems. New
York: Marcel Dekker, 2005.
[22] J. Song et al., Model development and control methodology of a new
electric power steering system, Proc. Inst. Mech. Eng. D, J. Automobile
Eng., vol. 218, no. 9, pp. 967975, 2004.
[23] N. Sugitani et al., Electric power steering with H-innity control de-
signed to obtain road information, in Proc. 1997 Amer. Control Conf.,
Albuquerque, NM, Jun., pp. 29352939.
[24] M. Tai, P. Hingwe, and M. Tomizuka, Modeling and control of steering
system of heavy vehicles for automated highway systems, IEEE/ASME
Trans. Mechatron., vol. 9, no. 4, pp. 609618, Dec. 2004.
[25] D. Tian, G. Yin, and G. Xie, Model and H

robust control design for


electric-power steering system, in Proc. 2004 Int. Conf. Intell. Mechatron.
Autom., Chengdu, P. R. China, 2004, pp. 779783.
[26] A. T. Zaremba, M. K. Liubakka, and R. M. Stuntz, Control and steering
feel issues in the design of an electric power steering system, in Proc.
1998 Am. Control Conf., Philadelphia, PA, Jun., pp. 3640.
Wolfgang Kemmetm uller (M04) received the
Dipl.-Ing. degree in mechatronics from Johannes
Kepler University Linz, Linz, Austria, in 2002.
He is currently a Researcher at Saarland Univer-
sity, Saarbr ucken, Germany. His current research in-
terests include the modeling of mechatronic systems
based on physical models and the nonlinear controller
design, modeling and control of hydraulic systems,
especially in automotive applications and smart u-
ids (electrorheological uids).
Steffen M uller recieved the Masters and Ph.D.
degrees in aeronautics from the Technical Univer-
sity of Berlin, Berlin, Germany, in 1993 and 1998,
respectively.
From 1998 to 2000, he was a Researcher in Me-
chanics and Mechatronic Systems, ABB Corporate
Research Center, Heidelberg, Germany. From 2000
to 2001, he was a Postdoctoral Fellow at University
of California at Berkely, where he was engaged in the
eld of vehicle system dynamics and control. Since
2001, he has been a Project Leader in the Depart-
ment of Advanced Chassis Control Systems, BMW Research Center Munich,
Munich, Germany, where he is engaged in the development of advanced steering
systems.
Andreas Kugi (M96) received the Dipl.-Ing. de-
gree in electrical engineering from Graz University
of Technology, Graz, Austria, in 1992, and the Ph.D.
(Dr.techn.) degree in control engineering and the Ha-
bilitation degree in automatic control and control
theory fromJohannes Kepler University (JKU), Linz,
Austria, in 1995 and 2000, respectively.
From 1995 to 2000, he was an Assistant Profes-
sor, and from 2000 to 2002, an Associate Professor
at JKU. During 2002, he was appointed a Full Pro-
fessor at Saarland University, Saarbr ucken, Germany,
where he is currently the Head of the System Theory and Automatic Control.
His current research interests include the physics-based modeling and control of
(nonlinear) mechatronic systems, differential geometric and algebraic methods
for nonlinear control, and the controller design for innite-dimensional systems.
He is also involved in several industrial research projects in the eld of auto-
motive applications, rolling mills, hydraulic servo-drives, and the design and
optimization of MEMS devices.

You might also like