See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224112844

Dual-Mode Switched Control of an Electropneumatic Clutch Actuator

Article in IEEE/ASME Transactions on Mechatronics · January 2011

DOI: 10.1109/TMECH.2009.2036172 · Source: IEEE Xplore

CITATIONS READS
49 852

2 authors:

Hege Langjord Tor Arne Johansen

9 PUBLICATIONS 210 CITATIONS

Norwegian University of Science and Technology
660 PUBLICATIONS 21,794 CITATIONS
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Hege Langjord on 20 January 2020.

The user has requested enhancement of the downloaded file.

TMECH-06-2009-0584 1

Dual-mode switched control of an electropneumatic clutch actuator

Hege Langjord, and Tor Arne Johansen, Member, IEEE

Abstract—This paper proposes a stabilizing switched controller can make accurate control difficult [11]. The nonlinearities
for an electropneumatic clutch actuator. Input restrictions arise arise mainly from the compressibility of air, stiction and high
from the fact that the allocation of air to and from the actuator friction forces. Several papers have been published on the
chamber is governed by on/off valves. The dual-mode controller
is a combination of a local and a global switched controller, topic of position control of pneumatic actuators, see Ahn and
and asymptotic stability in the region of operation of the clutch Yokata [9] and Girin et al [12], and references therein. To
is proven. Asymptotic stability is verified and the controller’s control the air flow to the actuators, where are two main
performance is investigated through experiments in a test vehicle. choices: servo valves or on/off solenoid valves. Servo valves
are the conventional choice, while the much cheaper, but more
Index Terms—Position control, Pneumatic systems, Switched challenging (from a control point of view) on/off solenoid
input, Solenoids valves are increasing in popularity. In addition on/off valves
have advantages as robustness, higher power-to-mass ration
I. I NTRODUCTION and smaller size compared to servo valves [13]. The drawbacks
are mainly the valve’s discrete on/off nature, limitations in
Automated Manual Transmission (AMT) systems consist
response time and a dynamic response which is hard to model
of automated clutch motion during gear shifts and direct
accurately.
transmission through the clutch disc. The gear shifts can either
Pulse Width Modulation (PWM) was first proposed by
be decided by the driver or by a controller based on the
Reeves in the late 1930’s [14] and has since been used
vehicle’s state. One of the advantages of the AMT systems
extensively when dealing with discrete input devices. PWM
over conventional Automated Transmission (AT) systems, is
may allow control laws made for servo valves to be used
that the system can be easily added to existing Manual
for solenoid valves. Varsevald and Bone [8] show how fast,
Transmission (MT) systems. This add-on property is one of the
accurate and inexpensive position control for pneumatic actu-
reasons for the growth in the systems’ popularity over the last
ators can be obtained using on/off solenoid valves and PWM.
decade, especially in the European market [1]. In addition, the
Carducci et al. [15] propose a vision feedback on a robotic
AMT systems have the advantages of lower weight and lower
arm where the on/off solenoid valves are driven by PWM
fuel consumption. The main purpose of AMT systems, like AT
techniques, while Ahn et. al [9] propose a modified PWM
systems, is to increase driving comfort and improve shifting
scheme to improve steady-state error often arising when using
quality [2].
on/off valves. The last decades, sliding mode techniques have
The clutch plates transmit torque from the motor to the
became a more common approach to design controllers for
axle shaft. Smooth engagement is crucial and the clutch
pneumatic actuators, both for systems with servo valves, e.g
is subject to satisfy small friction losses, fast lock-ups and
[6], [16] and [17], and for systems with solenoid valves, e.g
driving comfort [3]. To obtain this, position control of the
[7], [18], [19]. Paul et. al [18] early proposed a position control
clutch actuator is the most important factor when designing
algorithm based on sliding mode control for a pneumatic
AMT systems [4]. AMT systems are especially desired for
cylinder with on/off valves, Shen et. al [19] create at sliding
trucks, because conventional AT systems for such high torque
mode control with equivalent control and when use PWM to
transfer are expensive and have a large power loss [5]. The
produce the control signal, while Nguyen et al. [7] uses a
electropneumatic clutch system for heavy duty trucks, consid-
sliding mode approach to construct a control signal which can
ered in this paper, is well suited as an AMT system.
be directly applied to the solenoid valves.
Pneumatic actuators are common industry actuators [6] and In this paper we consider position control of an electropneu-
are becoming more popular due to recent improvements in matic clutch actuator where on/off solenoid valves are used to
valve technology and reduction of component costs [7]. They control the airflow of the actuator. The same system have been
are especially suited for robot application because of their high considered by Kaasa in his doctoral thesis [5] which concen-
force-to-mass ratio and are in general applied whenever stiff trates on nonlinear feedback control, and by Grancharova and
and lightweight structures are critical [8], [9]. Other advan- Johansen in [20] and [21] which deals with design of explicit
tages of such actuators are clean operation, easy maintenance model predictive controllers. Hydraulic solutions are common
and low cost [10], but they also have a nonlinear behavior that in AMT systems, [2], [1], [22], but pneumatics are chosen in
Manuscript received June 18, 2009; revised September 10, 2009. This this system as pressurized air is already present in trucks. The
work was sponsored by the Norwegian Research Council and Kongsberg on/off valves are not controlled by a PWM signal. Instead of
Automotive ASA. designing a continuous controller and implement this by use of
H. Langjord and T. A. Johansen are with the Department of Engineering
Cybernetics, Norwegian University of Science and Technology, NO-7491 PWM, we choose to take the on/off valves nature into account
Trondheim, Norway (email:[email protected],[email protected]). in the control design. This leads us to the design of a controller
TMECH-06-2009-0584 2

with a restricted number of available inputs levels. One of the [m/s] and x3 is accumulated air [kgm2 /s], where accumu-
main advantages this approach gives us, is that we do not need lated air is proportional to the amount of air in the actuator
to know the on/off valves characteristics, except its response chamber. Parameter values and descriptions corresponding to
time, and may achieve faster closed loop response time since the experimental setup can be found in Appendix A. The state
the controller is not limited by the PWM duty cycle. We have errors are written as x̃i = xi − x∗i , where x∗i are reference
previous published papers on this subject, [23] and [24]. These points,
propose two switched controllers, one there only local stability
is proved and one with better stability properties, but which x∗ = [x∗1 , x∗2 , x∗3 ]T (2)
have poor robustness qualities. The idea behind this paper is to The reference point x∗1 is given, x∗2 = 0 and x∗3 is given by
combine these two controllers, and benefit from the individual
controllers best features. A preliminary version of this paper V (x∗1 ) ∗
x∗3 = (Kl (1 − e−Ll x1 ) − Ml x∗1 + AP0 ).
is presented in [25]. The present paper contributes with addi- A
tional experimental validation, considering different valve sets, The region of operation for the clutch actuator considered
improved tuning and further details on the implementation of in this paper is O = {x1 ∈ [0, 0.025], x2 ∈ R, x3 ∈ h 0, ∞ ]},
the controller. where the position limits correspond to mechanical dead stop.
The only available inputs values using two on/off valves are
II. S YSTEM DESCRIPTION AND MODEL w ∈ {−Umax , 0, Umax } (3)
Figure 1 shows a schematic of the clutch actuator system
under consideration. The Electronic Control Unit (ECU) cal- where Umax is the maximum flow capacity through the valves.
culates the control signals which are sent to the on/off valves. −Umax correspond to open exhaust valve and closed supply
Based on these, the on/off valves control the flow to and from valve, Umax correspond to open supply valve and closed
the actuator chamber. The piston position is a result of the exhaust valve, while the input 0 correspond to both valves
acting forces, the friction, the pressure, the spring and the being closed. Table I show these connection between input
clutch load forces, and this position determines the state of the and valve positions.
clutch plates. The plates can either be engaged, disengaged or TABLE I
slipping. When engaged, the clutch transmits torque from the VALVE POSITION CORRESPONDING TO THE INPUTS
motor to the axle shaft.
Input Supply valve Exhaust valve
−Umax Closed Open
0 Closed Closed
Umax Open Closed

The model (3) is a simplification since the supply and

exhaust capacities in general will be different and depend
on the pressure difference over the valve and the ambient
temperature. This simplification has no major consequence
for the validity of the analysis and design of the controller,
since choked flow can be assumed. In the most important
operational region, the engage/disengage area, the pressure
drop tends to be significant due to a supply pressure of 9.5 bar
and a maximum cylinder pressure of less than 6 bar. The only
region where the choked flow assumption is not valid is when
the clutch is completely disengaged and the cylinder is being
Fig. 1. Electropneumatic clutch actuator system emptied from air, but control accuracy is not very important
in this region.
We use a rather simpler model of the clutch actuator for
control design. III. C ONTROL ALGORITHM DESIGN
The main idea of our control design is at each sampling
ẋ1 = x2 time, choose an available input which ensures a negative
1 Ax3
ẋ2 = (−fl (x1 ) − Dx2 + − AP0 ) (1) definite Lyapunov function time derivative with a well chosen
M V (x1 ) control Lyapunov function (CLF). The question is then how to
ẋ3 = RT0 w find appropriate CLFs. We consider two different strategies:
where fl (x1 ) = Kl (1 − e−Ll x1 ) − Ml x1 is the clutch load • Using backstepping

characteristic and V (x1 ) = V0 + Ax1 is the chamber volume. • Based on knowledge of the system

The variables D, P0 , R and T0 are viscous damping, ambient The following controllers, controller I and II, are presented
pressure, gas constant of air and temperature. The input to the in previous work [23] and [24], while controller III, is the
system is w, while x1 are piston position [m], x2 is velocity combined controller.
TMECH-06-2009-0584 3

A. Controller I - local controller Proposition 1: The equilibrium x∗ of the system (1), with
The following control Lyapunov function is found from the switched control input given by (8) is locally exponentially
backstepping (see Appendix B), with α1 , β1 , λ1 being positive stable.
tuning parameters, See Appendix C for proof.
The region of attraction contains the invariant set
α1 2 β1 2 λ1 2
V1 (z) = z + z2 + z3 (4) Ω̄1 = {z|V1 (z) ≤ ¯} ,
2 1 2 2 (12)
and its time derivative along trajectories is where ¯ is the largest value such that Ω̄1 ⊆ Ω1 and
D 2 β1 Az2 z3
V̇1 = −α1 kz12 − β1 z + + Ω1 = {z ||a(z)| ≤ RT0 Umax } (13)
M 2 M V (z1 + x∗1 )
∗
λ1 z3 (RT0 w − (z2 − kz1 )(Kl (1 − e−Ll (z1 +x1 ) ) The value of ¯ depends on the parameter values, and are
respectively 9.7 · 10−7 and 4.5 · 10−7 for valveset I and II
−Ml (z1 + x∗1 )
− Dkz1 + AP0 − M kz2 + (5)
with the parameter values listed in Appendix A
V (x1 ) ∗
(Kl Ll e−Ll (z1 +x1 ) − Ml − Dk))
A
M kV (z1 + x∗1 ) 2 D B. Controller II - global controller
−kz3 + (k z1 + z2 )).
A M [24] consider the control Lyapunov function
The coordinate transform defining z1 , z2 and z3 as a function Z x̃1
β2 λ2
of x1 , x2 and x3 is given by (23), (27) and (35) in Appendix V2 (x̃) = α2 f (y + x∗1 , x∗3 + x̃3 )dy + x̃22 + x̃23 (14)
0 2 2
B. Notice that
λ1 z3 RT0 w (6) where α2 , β2 , λ2 are positive constants. f (x1 , x3 ) =
1 Ax3
M (fl (x1 ) − V (x1 ) − AP0 ) is chosen based on the function
the only term of V̇1 dependent on the input such that making R x̃1
VC (x̃) = 0 f (y + x∗1 , x∗3 )dy + x̃22 which can be used to
λ1 z3 RT0 w as negative as possible is the best we can do to
show asymptotic stability of the open loop system, w = 0,
get V̇1 negative definite. Since R, T0 and λ1 are constants,
with x3 = x∗3 , see Appendix D-A for proof of this. For
choosing an input which satisfies
the system parameters given in Appendix A, the proposed
sgn(w) = −sgn(z3 ) (7) control Lyapunov function is positive definite if the ratio
α2
λ2 ≤ γ(Kl , Ll , Ml ), see Appendix D-C. The control Lya-
will ensure (6) negative. This gives us the controller punov function time derivative along trajectories is

−Umax sgn(z3 ) 6 0
z3 = α2

V (x1 )

w1 = (8) ˙
V2 = α2 f (x1 , x3 )x̃2 − ln RT0 w
0 z3 = 0 M V (x∗1 )
and D
+β2 x̃2 (−f (x1 , x3 ) − x̃2 ) + λ2 x̃3 RT0 w (15)
V̇1 ≤ −σV1 − λ1 (RT0 Umax |z3 | + z3 a(z)) (9)  M  
β2 D 2 α2 V (x1 )
where = − x̃2 + λ2 x̃3 − ln RT0 w
M M V (x∗1 )
βAz2 β2 D 2
a(z) = −bz3 − + (z2 − kz1 ) = − x̃ + s(x̃1 , x̃3 )RT0 w
λM V (z1 + x∗1 ) M 2
∗
(Kl (1 − e−Ll (z1 +x1 ) ) − Ml (z1 + x∗1 ) (10) where
V (z1 + x∗1 ))
−Dkz1 + AP0 − kz2 + β2 = α2 (16)
A
∗
 
(Kl Ll e−Ll (z1 +x1 ) − Ml − Dk)) − kz3 s(x̃1 , x̃3 ) = λ2 x̃3 −
α2
ln
V (x1 )
(17)
M kV (z1 + x∗1 ) M V (x∗1 )
 
2 D
+ k z1 + z2 .
A M If we choose the following control law
Every switch between w = Umax and w = −Umax is done

−Umax sgn(s(x̃1 , x̃3 )) if s(x̃1 , x̃3 ) 6= 0
when w2 = (18)
0 if s(x̃1 , x̃3 ) = 0
V (x1 )
0 = z3 = x3 − (Kl (1 − e−Ll x1 ) (11) we get a negative semidefinite Lyapunov function time deriva-
A tive
−Ml x1 − Dkx̃1 + AP0 − M k(x2 + kx̃1 ))  
β2 D 2 α2 V (x1 )
which will be the nonlinear switching surface. Such nonlinear V˙2 = − x̃2 − λ2 x̃3 − ln RT0 Umax ≤ 0.
M M V (x∗1 )
switching surfaces are also treated in several papers where slid- (19)
ing mode controllers are designed, as in [26] and [27]. While Proposition 2: The equilibrium point x∗ of the system (1)
we design a backstepping controller first and the switching with the switched controller given in (18) is asymptotically
surface is found from this, the sliding mode case define the stable in the largest invariant region contained in O if αλ22 <
surface first, and uses this to design a controller that prove γ(Kl , Ll , Ml ) given the parameter values in Appendix A.
stability of the system. See Appendix D for proof.
TMECH-06-2009-0584 4

0.018
should in practice be a smoother curve, which would corre-
0.016
spond to smooth clutch release and lower power consumption,
0.014
but the proposed reference is used as a basis to test response
0.012 time and tuning of the controller. It is desired that the
controller makes the system reach the reference point within
Position [m]

0.01

0.1 s, and with a steady state position error of less than 0.2
0.008

Engage/disengage area mm in the area where the clutch engage/disengage in the area
0.006
around y = 5 mm. Outside this area, the requirements can be
0.004 somewhat relaxed. Two on/off solenoid valvesets have been
0.002 tested. Their characteristics are given in Table II.
0
0 0.5 1 1.5 2
Time [s]
2.5 3 3.5 4 TABLE II
VALVESET CHARACTERISTICS
Fig. 2. Clutch sequence
Valveset I Valveset II
Opening time 0.5 ms 0.8 − 1 ms
Closing time 2.5 ms 2.2 − 2.4 ms
Maximum volumetric flow rate, supply 14 l/s 14 l/s
The restriction on αλ22 limits the controller’s proved perfor- Maximum volumetric flow rate, exhaust 16 l/s 16 l/s
mance as it does not allow to weight the position error as
highly as desired. The error in accumulated air will be the main The sampling time, and the controller switching time, are
driving error of the controller, which leads to poor robustness both set to 1 ms. This is the same as the sampling rate of
properties of this controller since x∗3 is subject to model error. the measurements. For valveset II, too fast switching resulted
While it is possible to characterize the largest invariant set in in blown fuses during the experiments, so we impose the
O, we have chosen not to do so since the practical experiment restriction that the valves only can open once in every 20
show there are no practical limit. ms. To avoid chattering of the input signal for both valvesets,
the controller is shut down and w = 0 whenever |x̃1 | < 0.4
C. Controller III - combined controller mm. In addition the controller is shut down when x1 is close
to 0. Experiments have been conducted in a Scania test truck
We combine these two switched controllers into one that
at Kongsberg Automotive ASA, where the control algorithm
make use of the individual controller’s best qualities; the global
run on the dSpace MABX 1401 unit. Figure 3 shows how the
controller is stable in the whole operation region of the clutch
system is connected.
system, and can be used to bring the system close to the
equilibrium point, and the local controller, which depend more
on the position error and has better robustness properties, is
suitable for use close to the equilibrium point. Notice that
x̃ = 0 if and only if z = 0 such that we get a consistent
formulation of a combined controller

w 2 , z ∈ Ω2
wc = (20)
w1, z ∈ Ω̄1
where Ω2 = O\Ω̄1
Proposition 3: The equilibrium point x∗ of the system (1)
with the combined controller given in (20) is asymptotically
stable in the largest invariant region in O.
Proof: In the region Ω̄1 , wc = w1 guarantees local
exponential stability of the systems equilibrium point in the Fig. 3. System configuration
the region Ω̄1 . The equilibrium point lies inside Ω̄1 , and if the
system at t0 is in the region Ω̄1 the exponential stability of
the local controller and invariance of Ω̄1 , will ensure that the Position and pressure are measured, while velocity is cal-
region is not left and that the equilibrium point is asymptotic culated from the position measurement including a second
stable. If the state at t0 is in the invariant region contained order low pass filter with coincident poles in −40 rad/s,
in Ω2 , the asymptotic stability of the global controller ensures accumulated air is calculated from the pressure p and volume
that the system will reach the region Ω̄1 after some finite time, V according to
and the equilibrium point will be asymptotic stable with the x3 = pV (y). (21)
local controller. By assuming v = 0 and v̇ = 0 we have

IV. E XPERIMENTAL RESULTS fl (x1 ) = A(p − P0 ) (22)

A benchmark clutch sequence for evaluating control system from the motion dynamics, which is used to estimate the
performance is shown in Figure 2. This position reference clutch load characteristic. The tuned parameters are shown in
TMECH-06-2009-0584 5

TABLE III TABLE IV

C LUTCH LOAD PARAMETERS AVERAGE DELAYS OF THE VALVE SETS

Load parameter Valveset I Valveset II Valveset I Valveset II

Kl 4000 5300 Input to accumulated air 0.67 ms 5 ms
Ml 20000 20000 Acc. air to change in position 7.2 ms 3.5 ms
Ll 500 700

B. Performance under nominal conditions

Table III, notice that parameters differ for the valvesets as the
Figures 5 and 6 show the performance of the combined
experiments were conducted at different times, and hence with
controller under nominal conditions with the two valvesets.
different degree of wear of the clutch and temperature. Figure
The desired behavior of the controller is shown, the local
4 shows the resulting clutch loads compared to experimental
controller is used close to the equilibrium point, while the
results obtained using (22). We see that the system suffers from
global is used elsewhere. In the region of engagement we see
hysteresis, which is an effect of friction. See the Appendix A
that the position error requirement is only partly fulfilled. This
for tuned values of the other variables. Earlier experiments
is mainly due to the shut down of the controller whenever
|x̃1 | < 0.4 mm, but this is still chosen to reduce chattering.
The unmodelled delays in the valves lead to oscillations around
the reference point as the effect of the last calculated input
signal is not finished before the next is calculated. The tuning
5000 of the controller has been improved between the experiments
with valveset I and valveset II, and the effect of this is
clearly shown if the performance is compared. Also switching
between the local and the global controller, decided by ¯,
4000
have been better tuned for valveset II, resulting in fewer input
switches. The computed velocity is as seen in the figures quite
smooth. This comes from the fact that the second order filter
Clutch load [N]

3000 used in the calculation only is a first assumption, which could

be improved and tuned. The efficiency of the controller would
benefit from a better estimation of velocity.
2000
C. Performance of the local and the global controller
Figure 7 shows the performance of the local and the global
controller with valveset II. The improvement with the com-
1000
bined controller, shown in Figure 6, is evident, better accuracy,
A(p−P0) for the experiments with valveset II
less oscillations and fewer input switches are achieved.
fl with Kl=5300, Ml=20000, Ll=700
A(p−P0) for the experiment with valveset I
0 fl with Kl=4000, Ml=20000, Ll=500 D. Robustness
0 5 10 15 20 25
Position [mm] The effect of parametric errors in the modeled clutch load
characteristics was studied in [25]. Performance of the con-
troller when the load characteristic is detuned to a more worn
clutch show that the controller fails in terms of position error
Fig. 4. Clutch load characteristic.
whenever the global controller is active, while the performance
is much better whenever the local controller is active. The local
with the local controller indicate that the region where this controller is, as expected, more robust to model errors, which
controller performs well is a lot larger than the region Ω̄1 , comes from the fact that the switching surface depend on the
and because of this the rather strict restriction on ¯ have been errors of all three states. The global controller, on the other
lifted in the following experiments. hand, relies only on position and accumulated air error, and
in addition the condition on αλ22 results in higher emphasize
A. Delays in the system on the accumulated air error than the position error. Under
We have two main delays in the system, the delay from nominal conditions a small accumulated air error correlates to
the valve input signal is set to accumulated air starts chang- a small position error, while with model error introduced, the
ing, and the delay from accumulated air changes and until relation between these errors are no longer valid. A practical
change in position is detected. Average values found through implementation must therefore consider adaption of the load
experiments for these delays are given in Table IV. It seems characteristics and possibly other model parameters.
like valveset II have larger capacities then valveset I during
the experiments even though their maximum volumetric flow V. C ONCLUSIONS
rates are the same. This come from the fact that the capacities A stabilizing dual-mode switching controller for an elec-
vary with pressure and temperature. tropneumatic clutch actuator is presented. This controller is
TMECH-06-2009-0584 6

Position error [mm] Position error [mm]

0,8 0,8
0,6 0,6
0,4 0,4
0,2 0,2
−3 0 0
x 10 −0,2 −0,2
−0,4 −0,4
Position [m]

15 −0,6 −0,6
3 3.5 4 8 8.5 9
10
5
0
0 1 2 3 4 5 6 7 8 9
0.2
Velocity [m/s]

0.1
0
−0.1
−0.2
0 1 2 3 4 5 6 7 8 9
5
x 10
6
Pressure [Pa]

0 1 2 3 4 5 6 7 8 9
Input signal [−]

Supply

Closed

Exhaust
0 1 2 3 4 5 6 7 8 9
Control type [−]

Global

Local
0 1 2 3 4 5 6 7 8 9
Time [s]

Fig. 5. Controller performance with valveset I. Dashed curves are reference, solid are commands or measurements.
TMECH-06-2009-0584 7

Position error [m] Position error [m]

0,8 0,8
0,6 0,6
0,4 0,4
0,2 0,2
−3 0 0
x 10 −0,2 −0,2
−0,4 −0,4
Position [m]

15 −0,6 −0,6
3 3.5 4 8 8.5 9
10
5
0
0 1 2 3 4 5 6 7 8 9

0.2
Velocity [m/s]

0.1
0
−0.1
−0.2
0 1 2 3 4 5 6 7 8 9
5
x 10
6
Pressure [Pa]

0 1 2 3 4 5 6 7 8 9
Input signal [−]

Supply

Closed

Exhaust
0 1 2 3 4 5 6 7 8 9
Control type [−]

Global

Local
0 1 2 3 4 5 6 7 8 9
Time [s]

Fig. 6. Controller performance with valveset II. Dashed curves are reference, solid are commands or measurements.
TMECH-06-2009-0584 8

Position error [mm] Position error [mm]

0,8 0,8
0,6 0,6
0,4 0,4
0,2 0,2
−3 0 0
x 10
−0,2 −0,2
−0,4 −0,4
15
Position [m]

−0,6 −0,6
2.5 3 3.5 7.5 8 8.5
10

0
0 1 2 3 4 5 6 7 8 9 10

0.2
Velocity [m/s]

0.1

−0.1

−0.2
0 1 2 3 4 5 6 7 8 9 10
5
x 10
6
Pressure [Pa]

0 1 2 3 4 5 6 7 8 9 10

Supply
Input signal [−]

Closed

Exhaust
0 1 2 3 4 5 6 7 8 9 10
Time [s]

(a) Local controller

Position error [mm] Position error [mm]
0,8 0,8
0,6 0,6
0,4 0,4
0,2 0,2
−3
x 10 0 0
−0,2 −0,2
15 −0,4 −0,4
Position [m]

−0,6 −0,6
10 2.5 3 3.5 7.5 8 8.5

0
0 1 2 3 4 5 6 7 8 9

0.2
Velocity [m/s]

0.1

−0.1

−0.2
0 1 2 3 4 5 6 7 8 9
5
x 10
6
Pressure [Pa]

0 1 2 3 4 5 6 7 8 9

Supply
Input signal [−]

Closed

Exhaust
0 1 2 3 4 5 6 7 8 9
Time [s]

(b) Global controller

Fig. 7. Performance with valveset II. Dashed curves are reference, solid are commands or measurements.
TMECH-06-2009-0584 9

a combination of two switched controllers, and preserves the transforms the system into
best properties of these individual controllers. Experimental re-
sults conducted in a test vehicle, verifies the stability properties ż1 = z2 − kz1
1
of the controller, and show that it is suited for the system. It is ż2 = 8 − Kl (1 − e−Ll x1 ) − D(z2 − kz1 ) (28)
evident that the controller will benefit from online adaption of M
x3
the clutch load, as this is the main uncertainty in the system +Ml x1 + A − AP0 ) + kz2 − k 2 z1
V (x1 )
model. Development of an adaptive observer for this purpose,
and velocity and pressure estimation is a topic for ongoing We now choose x3 as virtual control φ2 (x1 , z1 , z2 ) and take
research. β1 2
VB (z1 , z2 ) = VA (z1 ) + z . (29)
2 2
A PPENDIX A From
PARAMETERS β1
V̇B = α1 z2 z1 − α1 kz12 + z2 (−Kl (1 − e−Ll x1 )
Variable Value Unit Description M
A 12.3 · 10−3 m2 Actuator area +Ml x1 − D(z2 − kz1 ) (30)
P0 1.095 · 105 Pa Ambient pressure φ2 (x1 , z1 , z2 )
+A − AP0 ) + β1 kz22 − β1 k 2 z1 z2
T0 293 K Temperature V (x1 )
J
R 288 kgK Gas constant of air and by setting
M 10 kg Mass of piston α1 = β1 k 2 (31)
PS 9.5 · 105 Pa Supply pressure
Ns we get
D 2000 m Viscous damping
V0 0.148 · 10−3 m3
Volume at y = 0 β1
V̇B = −α1 kz12 + z2 (−Kl (1 − e−Ll x1 )
Kl 4000/5300 N Load char. par. M
Ml 20000 N
Load char. par. +Ml x1 − D(z2 − kz1 ) + (32)
m
Ll 500/700 − Load char. par. φ2 (x1 , z1 , z2 )
A − AP0 ) + β1 kz22 .
k 382/10 − Backstep. parameter V (x1 )
b 10 − Backstep. parameter We choose the virtual control as
α1 400 − Lyapunov func. par.
V (x1 )
β1 0.003/4 − Lyapunov func. par. φ2 (x1 , z1 , z2 ) = (Kl (1 − e−Ll x1 ) (33)
A
λ1 0.1/0.00001 − Lyapunov func. par. −Ml x1 − Dkz1 + AP0 − M kz2 )
α2 3.839/2.726 − Lyapunov func. par.
β2 3.839/2.726 − Lyapunov func. par. to get
λ2 1 − Lyapunov func. par. D 2
V̇B = −α1 kz12 − β1 z . (34)
M 2
Step 3
A PPENDIX B The change of variables
BACKSTEPPING
z3 = x3 − φ2 (x1 , z1 , z2 )
Step 1 V (x1 )
First we define = x3 − (Kl (1 − e−Ll x1 ) (35)
A
z1 = x1 − x∗1 (23) −Ml x1 − Dkz1 + AP0 − M kz2 )
transforms the system into
which gives
ż1 = x2 . (24) ż1 = z2 − kz1
Az3 D
We choose virtual control x2 = φ1 (z1 ) = −kz1 , where k is a ż2 = − k 2 z1 − z2
M V (x1 ) M
positive constant. From the Lyapunov-like function ż3 = RT0 w − (z2 − kz1 )(Kl (1 − e−Ll x1 )
α1 2 −Ml x1 − Dkz1 + AP0 − M kz2 ) (36)
VA (z1 ) = z (25)
2 1 V (x1 )
− ((z2 − kz1 )Kl Ll e−Ll x1
it is easy to show that this virtual control gives A
−Ml (z2 − kz1 ) + Dk(z2 − kz1 ) − M k
V̇A = −α1 kz12 . (26) Az3 D
( − k 2 z1 − z2 ))
M V (x1 ) M
Step 2
The change of variables A new Lyapunov function is chosen
λ1 2
z2 = x2 − φ1 (z1 ) = x2 + kz1 (27) V1 (z) = VB (z1 , z2 ) + z (37)
2 3
TMECH-06-2009-0584 10

and this leads to For all points on the discontinuity surface vector h is defined
D 2 β1 Az2 z3 as  
V̇1 = −α1 kz12 − β1 z + + 0
M 2 M V (Ax1 ) h = f+ − f− =  0  (44)
λ1 z3 ((RT0 w − (z2 − kz1 )(Kl (1 − e−Ll x1 ) −2RT0 Umax
−Ml x1 − Dkz1 + AP0 − M kz2 ) (38) which is along the normal of the discontinuity surface, Ns =
V (x1 ) (0, 0, 1)T . The scalar, hN , defined as the projection of h on
− ((z2 − kz1 )Kl Ll e−Ll x1
A Ns is
−Ml (z2 − kz1 ) + Dk(z2 − kz1 ) hN = Ns h = − 2RT0 Umax < 0 (45)
Az3 D
−M k( − k 2 z1 − z2 )) and will always be negative. According to Lemma 7 in [30],
M V (x1 ) M
the uniqueness of the Filippov solution is then guaranteed.
By choosing the input as Second we consider stability of the solution. The Lyapunov
1 β1 Az2 time derivative is
w = (− + (z2 − kz1 )
RT0 M λ1 V (x1 ) V̇1 ≤ −σV1 − λ1 (RT0 Umax |z3 | + z3 a(z)) (46)
(Kl 1 − e−Ll x1 − Ml x1 + Dkx1 + AP0 (39)


where a(z) is given by (10), and
V (x1 )
((z2 − kz1 )Kl Ll e−Ll x1
 
−M kz2 ) − D
A σ = 2 min k, , b . (47)
Ml (z2 − kz1 ) + Dk(z2 − kz1 ) M
Since a(0) = 0 and a(z) is continuous there must exist an
 
Az3 2 D
−M k − k z1 − z2 ) − bz3 )  > 0 such that for |z| ≤  we have |a(z)| ≤ RT0 Umax . It
M V (x1 ) M
follows that for |z| ≤  we get
where b is a positive constant, we get V̇1 ≤ −σV1 (48)
D 2 and the equilibrium point x∗ is locally exponentially stable.
V̇1 = −α1 kz12 − β1 z − λ1 bz32 (40)
M 2
which shows exponential stability of the system reference A PPENDIX D
equilibrium with backstepping control. P ROOF OF P ROPOSITION 2
Proof: First we prove existence, uniqueness and continu-
A PPENDIX C ity of the solution using Filippov solution theories in the same
P ROOF OF P ROPOSITION 1 way as in the proof of Proposition 1. The discontinuity surface
is described by
Proof: First we prove existence, uniqueness and continu-
ity of the solution using Filippov solution theories as in [28]. S := {x : s(x̃1 , x̃3 ) = 0} (49)
The discontinuity surface can be described by which divides the solution domain Ω into two regions
S := {z : z3 = 0} (41) Ω− : = {x : s(x̃1 , x̃3 ) < 0} (50)
+
and this divides the solution domain Ω into to regions: Ω :=− Ω : = {x : s(x̃1 , x̃3 ) > 0} . (51)
{z : z3 < 0} and Ω+ := {z : z3 > 0} . As the right hand side The right hand side of (1) is defined everywhere in Ω and
of (36) is defined everywhere in Ω and are measurable and are measurable and bounded. This means that the system
bounded, the system (36) satisfy condition B of Filippov’s (1) satisfy condition B of Filippov’s solution theory [29],
solution theory [29]. According to Theorems 4 and 5 in the and according to Theorems 4 and 5 in the same reference,
same reference, we when have local existence and continuity we when have local existence and continuity of a solution.
of a solution. Further, since the right hand of (36) is continuous The right hand side of (1) is also continuous before and
before and after the discontinuity surface, S, and this surface after the discontinuity surface, S, and this surface is smooth
is smooth and independent of time, conditions A, B, C of and independent of time. Hence, the conditions A, B and
Filippov’s solution are satisfied [30]. Following the procedure C of Filippov’s solution [30] are satisfied. By following the
introduced in [29] the vector functions f − and f + are defined procedure introduces in [29] the vector functions f − and f +
as the limiting values of the right-hand sides of the state space can be defined as the limiting values of the right-hand sides
equations in Ω− and Ω+ : of the state space equations in Ω− and Ω+
z2 − kz
   
x2
Az3 D
f− =  M V ((z1 +x∗ − k 2 z1 − M z2  (42) f − =  f (x1 , x3 ) − M
D
x2  (52)
1)
∗ RT0 Umax
RT0 Umax − φ̇((z1 + x1 ), z1 , z2 )
 
z2 − kz x2
 

f+ = 
Az 3
M V (z1 +x∗ − k 2 z1 − M D
z2  (43) f + =  f (x1 , x3 ) − M
D
x2  . (53)
1)
∗
−RT0 Umax − φ̇((z1 + x1 ), z1 , z2 ) −RT0 Umax
TMECH-06-2009-0584 11

Vector h, which is along the normal of the discontinuity By combining (60) and (63)
surface, V (x1 ) V (x∗1 )
 
T − g(x∗1 ) − g(x1 ) (66)
A V (x1 )

1 1
Ns = √ , 0, √ (54) α2

V (x1 )

2 2 = ln
λ2 M V (x∗1 )
is defined as
  and rearranging we get
0
h = f+ − f− = 0  (55) V (x1 )g(x1 ) − V (x∗1 )g(x∗1 ) (67)
−2RT0 Umax
 
α2 A V (x1 )
= ln
for all points on the discontinuity surface. The scalar, hN , λ2 M V (x∗1 )
defined as the projection of h on Ns is From the proof of positive definiteness of the Lyapunov
√ function (Appendix D-C) we have that
hN = Ns h = − 2RT0 Umax < 0 (56)
|V (x1 )g(x1 ) − V (x∗1 )g(x∗1 )| (68)
and will always be negative. According to Lemma 7 in [30], 2Aα2

V (x1 )

the uniqueness of the Filippov solution is then guaranteed. > ln
M λ2 V (x∗1 )
Next we consider the stability properties of the solution.
From (19) we have that V̇2 ≤ 0, and we use LaSalle’s invari- for αλ22 ≤ γ(Kl , Ll , Ml ). From this it follows that the only
ance principle to prove asymptotical stability. From V̇2 = 0 solution of (67) is x̃1 = 0. This show that x̃1 = x̃2 = x̃3 = 0
we get is the only solution and by the LaSalle’s invariance principle
x̃2 = 0 & |s(x̃1 , x̃3 )| = 0 (57) we can show asymptotically stability; n The only solutiono which
can stay identically in the set S = x ∈ O|V˙2 (x) = 0 is the
From this it follows that x̃2 = 0 → x̃˙ 1 = 0 → x̃1 = c1 , reference point (x̃1 , x̃2 , x̃3 ) = 0.
x̃2 = 0 → x̃˙ 2 = 0 → f (x1 , x3 ) = 0 and |s(x̃1 , x̃3 )| =
0 & x̃1 = c1 → x̃3 = c2 . This gives the solution A. Open loop stability
(x̃1 , x̃2 , x̃3 ) = (c1 , 0, c2 ). It then remains to show that c1 = Consider the input w = 0 and assume x3 = x∗3 . We when
c2 = 0 is the only possible solution. The equation can rewrite the system as a simple second order system
|s(x̃1 , x̃3 )| = 0 (58) ẋ1 = x2 (69)
D
have the two possible solutions ẋ2 = −f (x1 , x∗3 ) − x2
  M
α2 V (x1 ) which also can be expressed as
x̃3 = 0 & ln = 0 → x̃1 = 0 (59)
M V (x∗1 )
¨1 + D x̃˙ 1 + f (x1 , x∗3 ) = 0
x̃ (70)
and   M
α2 V (x1 ) If
x̃3 = ln (60)
λ2 M V (x∗1 ) D 2
x̃ > 0 x̃2 6= 0 (71)
The equation M 2
∗
−f (x1 , x3 )x̃1 > 0 x̃1 6= 0
Ax̃3
f (x1 , x3 ) = f (x1 , x∗3 ) + =0 (61) 0
in the region O = {x1 ∈ [0, 0.025], x2 ∈ R}, we can choose
M V (x1 )
Lyapunov function as the sum of potential and kinetic energy
also have two possible solutions Z x̃1
1
Ax̃3 VC = x̃22 + f (y + x∗1 , x∗3 )dy (72)
f (x1 , x∗3 ) = 0 & =0 (62) 2 0
M V (x1 )
and by differentiating it we obtain
x̃1 = 0 & x̃3 = 0
V˙C = x̃2 x̃˙ 2 + f (x1 , x∗3 )x̃˙ 1
and D
Ax̃3 = − x̃22 − f (x1 , x∗3 )x̃2 + f (x1 , x∗3 )x̃2 (73)
f (x1 , x∗3 ) = − (63) M
M V (x1 ) D
= − x̃22 ≤ 0
M V (x1 ) M
x̃3 = − f (x1 , x∗3 ) n 0
o
A Consider S = x ∈ O |V˙C (x̃) = 0 , and the only solution
Defining that can stay identically in S is the reference point (x̃1 , x̃2 ) =
−Lx 0 as
g(x) = Kl (1 − e ) − Ml x + AP0 (64)
x̃2 ≡ 0 =⇒ x̃˙ 2 ≡ 0 =⇒ f (x1 , x∗3 ) ≡ 0 =⇒ x̃1 ≡ 0. (74)
gives
  As the second condition of (71) is satisfied, as shown below,
1 V (x1 )
f (x1 , x∗3 ) = −g(x1 ) + g(x∗1 ) (65) when by LaSalle-Krasovski’s theorem, the origin is proved
M V (x∗1 ) asymptotically stable.
TMECH-06-2009-0584 12

B. Proof of the second condition of (71) is fulfilled. Remember that f (x1 , x3 ) can be written as
We have that Ax̃3
f (x1 , x3 ) = f (x1 , x∗3 ) − (85)
Ax∗3 M V (x1 )
f (x1 , x∗3 ) = Kl (1 − e−Ll x1 ) − Ml x1 − + AP0 )
V (x1 ) P (x1 ) − P (x∗1 ) Ax̃3
= −
= Kl (1 − e−Lx1 ) − Ml x1 (75) M V (x1 ) M V (x1 )
V (x∗
1)
∗
A A (Kl (1− e−Ll x1 ) − Ml x∗1 + AP0 ) As f (x1 , x∗3 )x̃1 > 0 the condition (84) is satisfied if f (x1 , x∗3 )
− + AP0 ) dominates the term Ax̃3 i.e:
V (x1 ) M V (x1 )
V (x∗1 )
 
= g(x1 ) + g(x∗1 ) |P (x1 ) − P (x∗1 )| > |Ax̃3 | (86)
V (x1 )
1 This is the case if
= (P (x1 ) − P (x∗1 ))  
V (x1 ) 2α2 A V (x1 )
|P (x∗1 ) − P (x1 )| > ln (87)
where P (x) = V (x)g(x) is a positive function strictly increas- λ2 M V (x∗1 )
ing in x for x ∈ [0, 0.025] and and this inequality is satisfied if αλ22 ≤ 531.93/619.81 depend-
ing on the parameter values. We call this limit γ(Kl , Ml , Ll ).
g(x) = Kl (1 − e−Lx) ) − Ml x + AP0 > 0 (76)
The proposed Lyapunov function (14) is positive definite when
α2
For ∀ x̃1 > 0 we have λ2 ≤ γ(Kl , Ll , Ml ).

x1 > x∗1 ⇔ P (x1 ) > P (x∗1 ) ⇔ f (x1 , x∗3 ) > 0 (77)

ACKNOWLEDGMENT
and The authors would like to thank Glenn-Ole Kaasa, Christian
f (x1 , x∗3 )x̃1 > 0 (78) Bratli, Sten Roar Snare, Håkon Holm Solberg and João P.
For ∀ x̃1 < 0 we have Hespanha for helpful inputs.

x1 < x∗1 ⇔ P (x1 ) < P (x∗1 ) ⇔ f (x1 , x∗3 ) < 0 (79) R EFERENCES
and [1] L. Glielmo, L. Iannelli, V. Vacca, and F. Vasca, “Gearshift control
for automated manual transmissions,” IEEE/ASME Transactions on
f (x1 , x∗3 )x̃1 > 0 (80) Mechatronics, vol. 11, pp. 17–26, 2006.
[2] G. Lucente, M. Montanari, and C. Rossi, “Modelling of an automated
When we have f (x1 , x∗3 )x̃1 > 0 ∀ x̃1 6= 0 and the second manual transmission system,” Mechatronics 17 (2007), pp. 73–91, 2007,
[3] A. Bemporad, F. Borelli, L. Glielmo, and F. Vasca, “Hybrid control
condition of (71) is satisfied. of dry clutch engagement,” in In proceedings of the European Control
Conference, Porto, Portugal, 2001,
[4] J. Horn, J. Bamberger, P. Michau, and S. Pindl, “Flatness-based clutch
C. Positive definiteness of the Lyapunov function V2 (x) control for automated manual transmissions,” Control Engineering Prac-
We rewrite the Lyapunov function (14) as tice, vol. 11, pp. 1353–1359, 2003,
[5] G. O. Kaasa, “Nonlinear output-feedback control applied to electro-
Z x̃1 pneumatic clutch actuation in heavy-duty trucks,” Ph.D. dissertation,
β2 λ2
V2 = α2 −f (y + x∗1 , x3 )dy + x̃22 + x̃23 NTNU, 2006.
0 2 2 [6] M. Smaoui, X. Brun, and D. Thomasset, “A combined first and second
Z x̃1 order sliding mode approach for position and pressure control of an
β2 λ2
= α2 f (y + x∗1 , x∗3 )dy + x̃22 + x̃23 (81) electropneumatic system,” in Proceedings of the American Control
0 2 2 Conference, Portland, 2005.
Z x̃1 [7] T. Nguyen, J. Leavitt, F. Jabbari, and J. E. Bobrow, “Accurate sliding-
α2 Ax̃3 mode control of pneumatic systems using low-cost solenoid valves,”
− dy
M 0 V (x∗1 + y) IEEE/ASME Transactions on mechatronics, vol. 12, pp. 216–219, 2007,
R x̃ [8] R. B. V. Varseveld and G. M. Bone, “Accurate position control of a
where we have that 0 1 f (y + x∗1 , x∗3 )dy is positive from the pneumatic actuator using on/off solenoid valves,” IEEE/ASME Transac-
tions on mechatronics, vol. 2, pp. 195–204, 1997,
calculations above. When if [9] K. Ahn and S. Yokota, “Intelligent switching control of pneumatic
α2 x̃1 actuator using on/off solenoid valves,” Mechatronics, vol. 15, pp. 683–
Z
Ax̃3 λ2
− dy + x̃23 (82) 702, 2005,
M 0 V (x∗1 + y) 2 [10] S. Aziz and G. M. Bone, “Automatic tuning of an accurate posi-
tion controller for pneumatic actuators,” in Proceedings of the 1998
   
α2 V (x1 ) λ2
= x̃3 − ln + x̃3 > 0 IEEE/RSJ Intl. Conference on Intelligent Robots and System Victoria,
M V (x∗1 ) 2 B.C., Canada, 1998.
[11] A. Ilchmann, O. sawodny, and S. Trenn, “Pneumatic cylinders: mod-
we have V2 > 0, as the other parts of (81) are positive. elling and feedback force-control,” International Journal of Control,
Equation (82) is satisfied if vol. 79, pp. 650–661, 2006.
  [12] A. Girin, F. P. nd X. Brun, and A. Glumineau, “High-order sliding-mode
λ2 α2 V (x1 ) controllers of an electropneumatic actuator: Application to a aeronautic
x̃3 > ln (83)
2 M V (x∗1 ) benchmark,” IEEE Transactions on control system technology, vol. 17,
no. 3, pp. 663–645, 2009.
[13] F. Malaguti and E. Pregnolato, “Proportinal control of on/off solenoid
 
V (x1 )
For x̃3 , |x̃3 | < | λ2α
2M
2
ln V (x ) |, we need to show that
∗ operated hydraulic valve by nonlinear robust controller,” Proceedings of
1
the 2002 IEEE International Symposium on Industrial Electronics, pp.
f (x̃1 , x3 )x̃1 > 0 (84) 415–419, 2002,
TMECH-06-2009-0584 13

[14] O. Keles and Y. Ercan, “Theoretical and experimental investigation of a Tor Arne Johansen (M’98-SM’01) was born in
pulse-width modulated digital hydralic position control system,” Control 1966, received his Siv.Ing. (MSc) degree in 1989
Engineering Practice, vol. 10, pp. 645–654, 2002. and Dr.Ing. (Ph.D) degree in 1994, both in electrical
[15] G. Carducci, M. Foglia, A. Gentile, N. I. Giannoccaro, and A. Messina, and computer engineering from the Norwegian Uni-
“Pneumatic robotic arm controlled by on-off valves for automatic har- versity of Science and Technology, Trondheim. From
vesting based on vision localisation,” in IEEE International Conference 1995 to 1997 he worked at SINTEF Electronics and
on Industrial Technology, vol. 2, 2004, pp. 1017–1022. Cybernetics as a researcher before he was appointed
[16] E. Richer and Y. Hurmuzlu, “A high performance pneumatic force Associated Professor in Engineering Cybernetics at
actuator system: Part ii-nonlinear control design,” Journal of Dynamic the Norwegian University of Science and Technol-
Systems, Measurement and Control, vol. 122, pp. 426–434, 2000. ogy in Trondheim in 1997 and was promoted full
[17] S. Laghroughe, M. Smaoui, F. Plesant, and X. Brun, “Higher order Professor in 2001. He has been a research visitor
sliding mode control based on optimal approach of an electropneumatic at the University of Southern California, Technical University in Delft, and
actuator,” International Journal of Control, vol. 79, no. 2, pp. 119–131, University of California in San Diego. He has published more than 70 articles
2006. in international journals as well as numerous conference articles and book
[18] A. K. Paul, J. K. Mishra, and M. G. Radke, “Reduced order sliding chapters in the areas of nonlinear estimation, optimization, adaptive control,
mode control for pneumatic actuator,” IEEE Transactions on Control model predictive control, nonlinear control and allocation with applications
Systems Technology, vol. 2, pp. 271–276, 1994. in the marine, automotive, biomedical and process industries. He has been
[19] X. Shen, J. Zhang, E. J. Barth, and M. Goldfarb, “Nonlinear averaging supervising more than 10 PhD students, holds several patents, and has been
applied to the control of pulse width modulated (pwm) pneumatic directing numerous research projects. In December 2002 Johansen co-founded
systems,” in Proceedings of the American Control Conference, 2004, the company Marine Cybernetics AS. Prof. Johansen is actively involved in
[20] A. Grancharova and T. Johansen, “Explicit approximate model predictive several IEEE and IFAC technical committees, and received the 2006 Arch T.
control of constrained nonlinear systems with quantized inputs,” in As- Colwell Merit Award of the Society of Automotive Engineers.
sessment and Future Directions of Nonlinear Model Predictive Control.
Springer Lecture Notes in Control and Information Science, L. Mangi,
D. Raimondo, and F. Allgower, Eds. Springer Verlag, 2009.
[21] A. Grancharova and T. A. Johansen, “Explicit model predictive control
of an electropneumtaic clutch actuator using on/off valves and pulse-
width modulation,” in Proceedings of the European Control Conference,
2009.
[22] M. Montanari, F. Ronchi, C. Rossi, A. Tilli, and A. Tonielli, “Control
and performance evaluation of a clutch servo system with hydraulic
actuation,” Control Engineering Practice, vol. 12, pp. 1369–1379, 2004,
[23] H. Sande, T. A. Johansen, G. O. Kaasa, S. R. Snare, and C. Bratli,
“Switched backstepping control of an electropneumatic clutch actuator
using on/off valves,” in Proceedings of the 26th American Control
Conference, New York, 2007, pp. 76–81.
[24] H. Langjord, T. A. Johansen, and J. P. Hespanha, “Switched control of
an electropneumatic clutch actuator using on/off valves,” in Proceedings
of the 27th American Control Conference, Seattle, 2008.
[25] H. Langjord, T. A. Johansen, and C. Bratli, “Dual-mode switched
control of an electropneumatic clutch actuator with input restrictions,”
in Proceedings of the European Control Conference, 2009.
[26] D. S. Lee, M. G. Kim, H. K. Kim, and M. J. Youn, “Controller design
of multivariable variable structure systems with nonlinear switching
surfaces,” Control Theory and Applications, IEEE Proceedings D, vol.
138, pp. 493–499, 1991.
[27] S. E. Lyshevski, “Sliding mode and soft switching control in dynamic
systems,” in Proceedings of the 19th American Control Conference,
Chicago, 2000.
[28] P. Sekhavat, Q. Wu, and N. Sepehri, “Impact control in hydraulic
actuators,” Journal of Dynamic Systems, Measurement and Control, vol.
127, pp. 197–205, 2005.
[29] A. F. Filippov, “Differential equations with discontinuous rigth-hand
sides,” Am. Math. Soc. Transl., vol. 42, pp. 199–231, 1960.
[30] ——, “Differential equations with second member discontinuous on
intersection surface,” Diff. Eq., vol. 15, pp. 1292–1299, 1979.

Hege Langjord was born in 1981. She received

the M.Sc. degree in control engineering from the
Norwegian University of Science and Technology
(NTNU), Trondheim in 2005. She is currently work-
ing towards the Ph.D. degree at the same university.
Her research interests are in nonlinear observer and
control design with application to electropneumatic
clutch actuation.

View publication stats