8th Vienna International Conference on Mathematical Modelling

8th
8th Vienna
Vienna18International
February Conference
- 20, 2015. Vienna
International on
on Mathematical
University
Conference Modelling
of Technology,
Mathematical Vienna,
Modelling
8th Vienna
February 18International
-- 20, 2015. Conference
Vienna on Mathematical
University of Modelling
Technology,
Austria
February 18 20, 2015. Vienna Available
University of online at Vienna,
Technology, www.sciencedirect.com
February 18 - 20, 2015. Vienna University of Technology, Vienna,
Vienna,
Austria
Austria
Austria
ScienceDirect
IFAC-PapersOnLine 48-1 (2015) 802–807
Multi-Variable
Multi-Variable Integral
Integral Sliding
Sliding Mode
Mode
Multi-Variable
Multi-Variable Integral
Integral Sliding
Sliding Mode
Mode
Control
Control of
of a
a Two
Two Degrees
Degrees of
of Freedom
Freedom
Control of a Two
Control of a Two Degrees of Freedom
Degrees of Freedom
Helicopter
Helicopter
Helicopter
Helicopter
Saif S. Butt ∗∗ , Harald Aschemann ∗∗
Saif
Saif S. Butt ∗ , Harald Aschemann ∗
Saif S. S. Butt
Butt ∗ ,, Harald
Harald Aschemann
Aschemann ∗
∗
∗ Chair of Mechatronics, University of Rostock,
∗ Chair of Mechatronics, University of Rostock,
∗ Chair of D-18059Mechatronics, University
Germanyof
Chair of D-18059
Mechatronics,Rostock,
Rostock, University
Germany of Rostock,
Rostock,
(e-mail: {Saif.Butt, D-18059
D-18059 Rostock, Germany
Harald.Aschemann}@uni-rostock.de)
Rostock, Germany
(e-mail: {Saif.Butt, Harald.Aschemann}@uni-rostock.de)
(e-mail: {Saif.Butt,
(e-mail: {Saif.Butt, Harald.Aschemann}@uni-rostock.de)
Harald.Aschemann}@uni-rostock.de)
Abstract: In this paper, a multi-variable nonlinear control-oriented model of a twin rotor
Abstract:
aerodynamicIn
Abstract: In this
this paper, a multi-variable
is presented. nonlinear control-oriented model of of aa twin rotor
Abstract:
aerodynamic this paper,
system
Insystem (TRAS)
paper,
(TRAS)
a
a multi-variable
multi-variable
is presented.
nonlinear
The mathematical
nonlinear
The
control-oriented
control-oriented
mathematical
description
description
model
modelof ofthe
ofthe a twin
twin rotor
multibody
rotor
multibody
aerodynamic
system
aerodynamic is derivedsystem
systemusing (TRAS)
Lagrange’s
(TRAS) is
is presented.
equations.
presented. The
Based
The mathematical
on the
mathematical resulting description
state-space
description of
of the multibody
representation,
the multibody
system
system is derived
is
a multi-variablederivedintegral
using Lagrange’s
using Lagrange’s
sliding mode equations.
equations.
control is Based
Based on the
on
designed the resulting
to resulting
accurately state-space
state-space
track desired representation,
representation,
trajectories
system
a is derivedintegral
multi-variable using Lagrange’s
sliding mode equations.
control Based
is on the
designed to resulting
accurately state-space
track representation,
desired trajectories
a
for
a multi-variable
both the
multi-variable integral
azimuth
integral sliding
angle and
sliding mode
the
mode control
pitch angle.
control is
is designed
Due
designed to to accurately
unmeasurable
to accurately track
states
track desired
and
desired trajectories
uncertainties
trajectories
for
for both
both the azimuth angle and at the pitch angle. as Due
well to asunmeasurable states and uncertainties
for both the
stemming
stemming the azimuth
from
azimuth
from
angle
angle and
simplifications
simplifications and at the pitch
pitch angle.
modelling
themodelling angle. as
Due
Due
well
to
toasunmeasurable
disturbance torques,
unmeasurable
disturbance
states
states and
torques, and a
uncertainties
a discrete-time
uncertainties
discrete-time
stemming
extended
stemmingKalman from
Kalman simplifications
filter
from simplifications (EKF) isat modelling
employed
atemployed
modelling and as well
combined
ascombined as disturbance
with
well as disturbance a torques,
discrete-time a discrete-time
implementation
torques, implementation
a discrete-time
extended
extended Kalman filter
filter (EKF) is and with a discrete-time
of the nonlinear
extended
of the Kalmancontrol
nonlinear filter (EKF)
control law. The
(EKF)
law.
is
is employed
The
proposedand
employed
proposed and combined
control
combined
control
strategy
strategy
with aa discrete-time
allows
withallows discrete-time
for an
implementation
for an excellent
implementation
excellent
tracking
tracking
of
of the
behaviour
the nonlinear
as
nonlinear control
highlighted
control law.
by
law. The proposed
experimental
The proposed control
results.
control strategy
strategy allows
allows for
for an
an excellent
excellent tracking
tracking
behaviour
behaviour as highlighted by experimental results.
behaviour as as highlighted
highlighted by by experimental
experimental results.
results.
© 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: Multivariable systems, helicopter control, sliding mode control, extended Kalman
Keywords:
Keywords:
filter Multivariable
Multivariable systems, systems, helicopter
helicopter control,
control, sliding
sliding mode
mode control,
control, extended
extended Kalman Kalman
Keywords:
filter Multivariable systems, helicopter control, sliding mode control, extended Kalman
filter
filter
1. INTRODUCTION considered as a DOF. An H∞ controller for helicopter
1.
1. INTRODUCTION
INTRODUCTION considered
considered isas aa DOF.
DOF. An
asdescribed An H
H∞ controller for helicopter
1. INTRODUCTION dynamics
considered as a DOF. An H ∞ controller
in Lopez-Martinez for helicopter
et al. (2003,
dynamics is described in ∞ controller for
Lopez-Martinez et helicopter
al. (2003,
In this paper, a twin rotor aerodynamic system (TRAS) dynamics dynamics is
2005). Theisresulting described
described in Lopez-Martinez
controller
in exhibits the et
Lopez-Martinez et al. (2003,
attributes
al. (2003,of
In
In
withthis
this
two paper,
paper,
degrees a
a twin
twin
of rotor
rotor
freedom aerodynamic
aerodynamic
(DOFs) is system
system
considered, (TRAS)
(TRAS)
which 2005).
2005).
a The
The
nonlinear resulting
resulting
PID with controller
controller
time-varying exhibits
exhibits the
the
constants attributes
attributes
according of
of
In
withthis
two paper,
degrees a twin
of rotor
freedom aerodynamic
(DOFs) is system
considered, (TRAS)
which 2005).
a The
nonlinear resulting
PID with controller
time-varying exhibits the
constants attributes
according of
with
was
with two
developed
two degrees
degrees by of
of freedom
INTECO
freedom (DOFs)
(2013),
(DOFs) is considered,
Poland,
is considered,for which
control
which a
to
a nonlinear
the operating
nonlinear PID
PID with
point.
with time-varying
In Ahmed
time-varying et constants
al. (2009),
constants according
a sliding
according
was
was
was
developed
developedThe
experiments.
developed
by INTECO
by first
by INTECO
INTECO
(2013),
(2013), Poland,
DOF characterises
(2013), Poland,
Poland, the for
for control to
control
horizontal
for control to
mode
to
the
the operating
operating
the control
operating
point.
point.
dealing
point.with
In
In Ahmed
In Ahmed
the couplings
Ahmed
et
et al. (2009),
et al.
al. (2009),
(2009),
a
a sliding
a sliding
of a helicopter
sliding
experiments.
experiments.
rotation of the The
The first
first
frame DOF
DOF
using characterises
characterises
the azimuth angle, the
the horizontal
horizontal
whereas mode mode
mode
with two control
control
rotors dealing
dealing with
with
is considered. the
the Ancouplings
couplings of
of a
a
adaptiveof second helicopter
helicopter
order
experiments. The first DOF characterises the horizontal control dealing with the couplings a helicopter
rotation
rotation
the second
rotation
of
of
the
the isframe
of one
the givenusing
frame
frame using
by the
using
the azimuth
thepitch
the angle angle,
azimuth
azimuth angle,
angle,
whereas
whereas
describing
whereasthe with with
sliding
with
two
two
two moderotors
rotors
rotors
is
is
control
is
considered.
has been An
considered.
considered. An
An
adaptive
adaptive
proposed
adaptive in second
Mondalorder
second
second order
and
order
the second one is given by the pitch angle describing the sliding
sliding mode
mode control
control has
has been
been proposed
proposed in
in Mondal
Mondal and
the
the second
inclination
second of
inclination
one
of
onethe is
the given
frame.
is frame.
given by byIn
In
the
a
the
a
pitch
real
pitch
real
angle
helicopter, describing
the
angle describing
helicopter, the
the
aerody- Mahanta
the sliding
aerody- Mahanta mode(2012);
control
(2012);
however,
has been
however,
no
no
experimental
proposed in validation
experimental Mondal and
validation is
and
is
inclination
namic forceof
inclination ofisthe
the frame. by
adjusted
frame. In changing
In a real
a real helicopter,
helicopter,
the anglethe the attack. Mahanta
aerody-
of aerody- presented (2012);
Mahanta therein.however,
(2012); In Butt no
however, et experimental
no al. (2014), a validation
experimental fourth-order
validation is
is
namic
namic
The TRAS, force
force is
is adjusted
adjusted
however, usesby
by changing
changing
achanging
changingthe the
the angle
angle
angular of
of attack.
attack.
velocity presented
presented therein.
therein.
nonlinear control-oriented
of presented In
In Butt
Butt et
et
model al.
al. (2014),
(2014),
of (2014), a
a fourth-order
fourth-order
a TRASamanufactured
namic force is adjusted by angle of attack. therein. In Butt et al. fourth-order
The
The TRAS,
TRAS, however,
the propeller
The TRAS, however, uses
uses a
for this purpose.
however, uses
changing
changing angular
aa changing angular velocity
angular velocity of
velocity of nonlinear
of nonlinear
by INTECO
nonlinear
control-oriented
along with model
control-oriented
control-oriented model of
of
a multi-variable
model
a TRAS flatness-based
of aa TRAS
manufactured
manufactured
TRAS flatness-based
manufactured
the
the
The propeller
propeller
2-DOF for
for this
this
helicopter purpose.
purpose.
system imposes challenging control by
by INTECO
INTECO
control scheme along
along
is with
with
proposed a
a multi-variable
multi-variable
for a 2-DOF flatness-based
helicopter. In this
the propeller for this purpose. by INTECO
control scheme along with
is proposed
proposed a multi-variable
for flatness-based
The
The 2-DOF
2-DOF
problems
The 2-DOF due helicopter
helicopter
to its given
helicopter
system
system
system
imposes
imposes challenging
nonlinearities
imposes as well ascontrol
challenging
challenging signif- control
control
control paper, the
control scheme
scheme same is
is model of for
proposed theaaa 2-DOF
for
2-DOF
2-DOF
helicopter.
TRAS helicopter.
is employed,
helicopter.
In this
In this
and
In this
problems
problems
icant couplingsdue
due to
to its
its
betweengiven
given nonlinearities
nonlinearities
the pitch axis and as
as well
well as
as signif-
signif-
the asazimuth paper,
paper,
an the
the
integral same
same
sliding model
model
mode of
of the
the
control TRAS
TRAS
scheme is
is isemployed,
employed,
designed and
and
problems
icant due
couplings to its
betweengiven nonlinearities
the pitch axis as
and well
the signif-
azimuth paper,
an the
integral same
sliding model
mode of the
control TRAS
scheme is isemployed,
designed and
and
icant couplings between
axis. Moreover,
icant couplings between
not all state the pitch
the pitch axis and
variables
axis and the
are the azimuth an
measurable.
azimuth an integral
validated
integral bysliding
experiments.
sliding mode
mode control
control scheme
scheme is
is designed
designed and
and
axis.
axis. Moreover,
Moreover,
To achieve
axis. Moreover,
not
not all
a satisfactory
not all state
all state
tracking
state
variables
variables
performance
variables
are
are regarding validated
are measurable.
measurable.
measurable. The givenby
validated
validated by experiments.
experiments.
bypaper is structured as follows: In Section 2,
experiments.
To
To achieve
achieve
desired a satisfactory
a satisfactory
satisfactory
trajectories for the tracking
tracking
azimuth performance
performance regarding
regarding
angle and regardingthe pitch The The
The given
given
a nonlinear paper
paper is structured
is structured
control-oriented modelas follows:
as follows:
of the In In
TRASSection
Section 2,
2,
is de-
To achieve
desired a
trajectories for thetracking
azimuth performance
angle and the pitch a given
nonlinear paper is structured
control-oriented modelas follows:
of the In
TRASSection
is 2,
de-
desired trajectories
angle, a trajectories
desired model-basedfor for the
control
the azimuth angle
approachangle
azimuth and
is chosen.
and the
the pitch
pitch a
a nonlinear
rived according
nonlinear control-oriented
to Butt
control-oriented et model
al.
model (2014). of
of the
A
the TRAS is
multi-variable
TRAS is de-
de-
angle,
angle, aa model-based
In several
angle, model-based
contributions,
a model-based
control
both approach
control
control modellingis
approach
approach isand
is experimental rived
chosen.
chosen.
chosen. rived
integral
rived
according
according
sliding mode
according
to
to Butt
to Butt
Butt
et
et
control al.is(2014).
et al.
al.is(2014).
proposed
(2014).
A
A multi-variable
A multi-variable
in Section 3.
multi-variable
In several contributions, both modelling and experimental integral
integral sliding
sliding mode
mode control
control is proposed
proposed in
in Section 3.
In
In several
identification contributions,
of
several contributions,
identification of
similar
similar
both
2-DOF
both
2-DOF
modelling
helicopters
modelling
helicopters
and experimental
set-ups
and experimental
set-ups
have
have
The
integral
The
controller is
sliding ismode
controller
extended
control
extended
by
by
a discrete-time
isa proposed
discrete-time in Section
Extended
Section
Extended
3.
3.
identification
been investigated,see
identification of
of similar
similar Ahmad 2-DOF
2-DOF helicopters
et al. set-ups
(2000b); Rahideh
helicopters set-ups have
and The
have The
Kalman controller
Filter (EKF)
controller is
is extended
extendedfor the by
by a
a discrete-time
estimation
discrete-time Extended
of unmeasured
Extended
been investigated,see Ahmad et al. (2000b); Rahideh and Kalman
Kalman Filter
Filter (EKF)
(EKF) for the estimation of unmeasured
been
Shaheed
been
Shaheed
investigated,see
(2007); Ahmad
investigated,see
(2007); Ahmad
Ahmad
Ahmadet al. et
et al.
et al. (2000b);
(2000a);
al.
(2000a);
(2000b);
Darus
Rahideh
Darus Rahideh
et al.
and states
et al. (2004);
and
(2004); Kalman
states
and
and (EKF) for
disturbance
Filter
disturbance for the
the estimation
torques
torques
in Section 4.of
estimation
in Section 4.
unmeasured
ofThe discrete-
unmeasured
The discrete-
Shaheed
Shaheed (2007);
(2004).
(2007); TheAhmad
The
Ahmad methodset al. (2000a);
proposed
et al. proposed Darus
(2000a); Darus therein et al. (2004);
correspond
al. (2004); states
etcorrespond states
time and disturbance
implementation
and disturbance of torques
both
torques the in Section
control
incontrol
Sectionlaw 4.
law The
and
4. The discrete-
the EKF
discrete-
Shaheed (2004). methods therein time
time implementation
implementation of
of both
both the
the control law and
and the
the EKF
Shaheed
to typical
Shaheed
to typical
(2004).
(2004). The
experimental methods
The methods
experimental
set-ups
set-ups
proposed
from
proposed
from
therein
different correspond
manufac-
therein correspond
different manufac-
are
are
presented
timepresented
implementation in
in
Section
of both
Section
5.
5.
In
the
In
Section
control
Section
6,
law
6, and the EKF
experimental
EKF
experimental
to typical
turers,
to typical experimental
various model-based
experimental set-ups from
approaches
set-ups different
and methods
from different manufac-
em- are
manufac- are
resultspresented
presented in
are giveninthat Section
show the
Section 5.
5. InIn Section
performance
Section 6,of 6, experimental
the tracking
experimental
turers, various model-based approaches and methods em-
em- resultsresults are
are given
given that
wellthat show
show the
the performance of the tracking
turers,
turers,
ploying
various
ployingvarious
artificial
artificial
model-based
model-based
intelligence,
approaches
intelligence,approaches
e.g., radialand
e.g., radial
and methods
basis functions,
methods
basis em-
functions,
controller
results
controller areas given
as well
as the
that
as show
the the performance
estimator.
estimator.performance
Finally,
of
of the
the tracking
Finally, conclusions are
tracking
conclusions are
ploying
neural
ploying artificial
networks intelligence,
and
artificial and genetic
intelligence, e.g., radial
algorithms.
e.g., radial In basis
In functions,
Rahideh
basis and
functions, controller
given in
controller as
as well7.as the estimator. Finally, conclusions are
well
Section as
7. the estimator. Finally, conclusions are
neural
neural networks
networks
Shaheednetworks
(2007), aandand genetic algorithms.
geneticmathematical
complete algorithms. In In Rahideh
Rahideh of
description and
and given
a given in
in Section
Section 7.
neural
Shaheed (2007), genetic algorithms. Rahideh and given in Section 7.
Shaheed (2007), a
2-DOF helicopter
Shaheed (2007),
complete
aa using
complete
using
complete
mathematical
mathematical
different approaches
mathematical
description
description of
of aaa
is described.
description of 2. CONTROL-ORIENTED MODELING
2-DOF helicopter different approaches is described.
described. 2.
2. CONTROL-ORIENTED MODELING
2-DOF
Regarding
2-DOF
Regarding
helicopter
the
helicopter
the
control
control
using
usingof
of
different
the TRAS,
different
the TRAS,
approaches
a simulation
approaches
a simulation
is
is study
described.
study
of a
of a 2. CONTROL-ORIENTED
CONTROL-ORIENTED MODELING MODELING
Regarding
nonlinear
Regarding the
the control
predictive
control of
of the
control
the TRAS,
is
TRAS, a
presented
a simulation
for
simulation a study
ninth-order
study of
of a
a The TRAS manufactured by INTECO is shown in Fig. 1.
nonlinear
model in predictive
nonlinear
nonlinear predictive
Dutka et al.
predictive
control
control
(2003).
control
is
is presented
is presented for
for a
In Lopez-Martinez
presented for et al. The
ninth-order
aa ninth-order
ninth-order The
It
The
TRAS
TRAS
consists
TRASof
manufactured
a beam withby
manufactured
manufactured by
two
by
INTECO
INTECO
propellers
INTECO
is
is shown
is shown
in
in Fig.
in Fig.
– the main
shown Fig.
1.
1.
and
1.
model
model
(2004), inin
in Dutka
Dutka et
et al.
al.
a feedback-linearising (2003).
(2003). In
In
controlLopez-Martinez
Lopez-Martinez
scheme is proposed et
et al.
al. It
It consists
consists
theconsists of
of
tail propeller a
a beam
beam
– at bothwith
with two
two
ends propellers
propellers
of the beam, –
– the
the main
main
driven and
and
by and
DC
model
(2004), a Dutka et al.
feedback-linearising (2003). In
controlLopez-Martinez
scheme is et
proposed al. It
the tail of
propeller a beam
– at with
both two
ends propellers
of the beam, – the main
driven by DC
(2004),
(2004), aa feedback-linearising
for the pitch feedback-linearising
motion only. Therein, controlthe
control scheme
yaw angle
scheme is is not the
is proposed
proposed tail
tail propeller
motors.
the The beam
propeller –– atis both
at pivoted
both ends
endsin of
ofaathe beam,
cardanic
the driven
beam, joint
drivenandby DC
by can
DC
for
for the
the pitch
pitch motion
motion only.
only. Therein,
Therein, the
the yaw
yaw angle
angle is
is not
not motors.
motors. The
The beam
beam is
is pivoted
pivoted in
in a cardanic
cardanic joint
joint and
and can
can
for the pitch motion only. Therein, the yaw angle is not motors. The beam is pivoted in a cardanic joint and can
Copyright
2405-8963 © © 2015,
2015,IFAC
IFAC (International Federation of Automatic Control)
802Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©
Copyright © 2015,
2015, IFAC
IFAC 802
802
Copyright ©under
2015,responsibility
IFAC of International Federation of Automatic
802Control.
10.1016/j.ifacol.2015.05.129
MATHMOD 2015

February 18 - 20, 2015. Vienna, Austria Saif S. Butt et al. / IFAC-PapersOnLine 48-1 (2015) 802–807 803

   
rotate freely both in the horizontal and vertical planes. −lcw sin ψ sin φ lcw sin ψ sin φ
Two counterbalance levers with a weight at their ends rcw1 =  −lcw sin ψ cos φ  , rcw2 =  lcw sin ψ cos φ  .
are fixed to the beam at the pivot. The counterweights
determine the steady-state pitch angle without propeller −lcw cos ψ −lcw cos ψ
actuation. Two velocity sensors are coupled with the (2)
PWM-driven DC motors for the main and tail rotors. The overall kinetic energy of the TRAS is determined
The TRAS has two encoders, mounted at the pivot of with the help of velocity vectors as time derivatives of
the azimuth and pitch axes. The multibody system model the corresponding position vectors. The kinetic energy T
in terms of the generalized coordinates can be expressed
beam encoder main rotor as 
1 2
T = (mm lm + mt lt2 )(φ̇2 cos2 ψ + ψ̇ 2 ) +
2

2 2 2 2 2 2
mcw lcw (φ̇ sin ψ + ψ̇ ) + Jz φ̇ + Jx ψ̇ , (3)

tail rotor dc motor where mm , mt and mcw represent the point masses at
with tacho the main rotor, the tail rotor and at the end of the
counterweight levers, respectively. The parameters Jz and
counter balance Jx are the moments of inertia of the rotating beam w.r.t.
its body-fixed z-axis and x-axis, respectively. Similarly, the
overall potential energy of the point mass system is given
as
1
U = kφ φ 2 (4)
2  
Uφ
 
Fig. 1. TRAS test-rig at the Chair of Mechatronics.
+ g (mm lm − mt lt ) sin ψ + 2mcw lcw (1 − cos ψ) .
for the TRAS consist of the beam, point masses for the
two rotors and point masses for the counterbalances. The Here, g represents the gravitational acceleration. The
modelling starts with assigning a right-hand coordinate restoring energy due to the cable is considered by Uφ .
system with the origin O located at the pivot point of the The Lagrangian for the system is defined as the difference
beam as shown in Fig. 2. between the kinetic energy and the potential energy, i.e.,
L = T − U. (5)
Fψ Finally, Lagrange’s equations in the presence of non-
zI conservativeforces
yI  result in
d ∂L ∂L ∂R
mm − = τq − , q = {φ, ψ} (6)
dt ∂ q̇ ∂q ∂ q̇
pitch
ψ where τψ and τφ are net torques along the pitch and the
lm azimuth axes. The Rayleigh function R is given by
azimuth
φ 1 1
O lcw xI R = cφ φ̇2 + cψ ψ̇ 2 . (7)
Fφ 2 2
lt ψ mcw Substituting Eq. (3), Eq. (4) and Eq. (5) into Eq. (6), the
equations of motion for the system result in
mt Jφ φ̈ = τφ + J φ̇ψ̇ sin(2ψ) − kφ φ − cφ φ̇, (8)
2
φ̇
Fig. 2. Free body diagram of TRAS. Jψ ψ̈ = τψ − J sin(2ψ) − cψ ψ̇ (9)
 2 
The distances of the main rotor and the tail rotor from − g (mm lm − mt lt ) cos ψ + 2mcw lcw sin ψ ,
the origin O are given by lm and lt , respectively. The
generalized coordinates φ and ψ denote the azimuth angle with
and the pitch angle. Moreover, lcw represents the relevant Jφ = (mm lm2
+ mt lt2 ) cos2 ψ + 2mcw lcw
2
sin2 ψ + Jz , (10)
length of the two levers with the counterweights as point 2 2 2
masses mcw at their end. Using a vector notation, the Jψ = mm lm + mt lt + 2mcw lcw + Jx , (11)
2 2 2
corresponding position vectors rm and rt for the main J = mm lm + mt lt − 2mcw lcw . (12)
rotor and the tail rotor, respectively, are given by It can be seen that the differential equations with respect
   
lm cos ψ cos φ lt cos ψ cos φ to the pitch angle and the azimuth angle are highly
rm =  lm cos ψ sin φ  , rt =  lt cos ψ sin φ  . (1) nonlinear and contain coupling terms. The net torques
lm sin ψ −lt sin ψ acting along the azimuth and the pitch axes are given by
τφ = Fφ lt cos ψ , (13)
Likewise, the position vectors rcw1 and rcw2 for the coun-
terweights at both ends of the beam become τψ = Fψ lm . (14)

803
MATHMOD 2015
804
February 18 - 20, 2015. Vienna, Austria Saif S. Butt et al. / IFAC-PapersOnLine 48-1 (2015) 802–807

1 a sliding mode controller is to drive the state trajectory

Fφ
0.8 towards a sliding surface in the state space and to maintain
Fψ
0.6 the state trajectory on this surface afterwards, cf. Slotine
0.4 and Li (1991); Young et al. (1999). The sliding surface is
0.2 independent of the disturbances and parameter variations
F in N

0 and, as a result, overcomes robustness issues. Let eφ and

-0.2 eψ denote the errors between the measured and the desired
-0.4 trajectories (denoted with a subscript d) for the azimuth
-0.6 angle and the pitch angle according to
-0.8
eφ = φ − φd and eψ = ψ − ψd . (17)
-1
-0.9 -0.7 -0.5 -0.3 -0.1 0.1 0.3 0.5 0.7 0.9
u in V Accurate tracking of desired trajectories for the azimuth
angle and the pitch angle can be achieved by the following
Fig. 3. Thrust characteristics as a function of the PWM choice of the sliding surfaces
 
voltage. sφ (x)
s(x) = , (18)
The relationship between the propulsive forces Fφ and sψ (x)
Fψ and the PWM input voltages ut and um for the with
tail rotor and the main rotor, respectively, is identified  t
experimentally, see Fig. 3. An inverse map of the nonlinear si (x) = ėi + αi1 ei + αi0 ei dτ , i ∈ [φ, ψ] . (19)
0
characteristic curve is implemented with the help of a
lookup table that transforms the propulsive forces into Here, the integral actions are introduced to avoid steady-
corresponding PWM inputs. state errors. The integral sliding mode control can be
To handle the parametric uncertainties and unknown divided into two phases: the sliding phase with s(x) =
disturbances, the model is extended with two lumped 0, ṡ(x) = 0, and the reaching phase with s(x) = 0. The
disturbance torques zφ and zψ acting on the azimuth and control input during the ideal sliding mode represents
pitch axes, respectively. Hence, the extended nonlinear the equivalent control ueq = [ueq,φ , ueq,ψ ]T , whereas an
state-space model can be written as additional switching control action usw = [usw,φ , usw,ψ ]T
  provides a finite-time convergence to the sliding surfaces
φ̇ during the reaching phase. Subsequently, the overall con-
   
˙  1  trol input ui can be expressed as the sum of both terms
φ  −(cφ φ̇ + kφ φ) + J φ̇ψ̇ sin(2ψ) + zφ 
 Jφ  ui = ueq,i + usw,i . (20)
 φ̇   
   ψ̇  The differentiation of the sliding surfaces sφ and sψ with
 =  
ψ   2 
 1 −(cψ ψ̇ + kψ ψ) − J φ̇ sin(2ψ) + zψ −  respect to time results in

 
ψ̇  Jψ 2  φ̈ − φ̈d + αφ1 (φ̇ − φ̇d ) + αφ0 (φ − φd )
       ṡ(x) = . (21)
ẋ g (mm lm − mt lt ) cos ψ + 2mcw lcw sin ψ ψ̈ − ψ̈d + αψ1 (ψ̇ − ψ̇d ) + αψ0 (ψ − ψd )
   The sliding condition implies that ∀t ≥ t1 the output
f (x) trajectory remain on the sliding surfaces s(x) = 0. By
 
0 0 choosing strictly positive coefficients of the Hurwitz poly-
 lt cos(ψ)  nomial, it can be ensured that the closed-loop system is
 0    asymptotically stable during the ideal sliding mode. In this
 Jφ  Fφ
+

 . (15) case, the error dynamics is governed by
 0 0 
  Fψ 
 lm  ëi + αi1 ėi + αi0 ei = 0 , i ∈ [φ, ψ] , (22)
0 u
where the strictly positive coefficients αi1 > 0 and αi0 > 0
Jψ
   denote coefficient of Hurwitz polynomials. Substituting
B(x) φ̈ and ψ̈ from Eq. (15) into Eq. (21), and solving for
These disturbance torques account for parametric uncer- the equivalent control vector ueq results in the following
tainty, unmodelled dynamics and all disturbances affecting expressions
the system. Among these are non-modelled effects due to Jφ ν1 − (cφ φ̇ + kφ φ) + J φ̇ψ̇ sin(2ψ) + zφ
the supply cables and gyroscopic torques as well as the ueq,φ = − , (23)
couplings caused by the tail rotor and the main rotor in lt cos ψ

the case of angular accelerations of the propellers. 1 φ̇2
ueq,ψ = − Jψ ν2 − (cψ ψ̇ + kψ ψ) − J sin(2ψ)+
The output vector y is identical to the measurement vector lm 2
y m and contains both generalized coordinates 
T zψ − g(mm lm − mt lt ) cos ψ − 2gmcw lcw sin ψ ,
y(t) = y m (t) = [ φ ψ ] . (16)
(24)
3. INTEGRAL SLIDING MODE CONTROL with
  
ν1 φ̈d − αφ1 (φ̇ − φ̇d ) − αφ0 (φ − φd )
In the case of parameter uncertainties, flatness-based con- = . (25)
ν2 ψ̈d − αψ1 (ψ̇ − ψ̇d ) − αψ0 (ψ − ψd )
trol schemes with linear stabilizing control laws are prone
to robustness issues. Therefore, an integral sliding mode The equivalent control input corresponds to the ideal
control (I-SMC) is proposed in this paper. The aim of motion of the system on the sliding surfaces s(x) = 0.

804
 MATHMOD 2015
February 18 - 20, 2015. Vienna, Austria Saif S. Butt et al. / IFAC-PapersOnLine 48-1 (2015) 802–807 805

However, when the initial state is not located on the following discrete-time state-space representation used for
sliding surfaces, an additional switching control part to the EKF design
be designed in the next step brings the state back to the xe,k+1 = xe,k + Ts f k (xe,k , uk ) +w , (33)
sliding surfaces – the reaching phase – and keeps it on the   
surfaces s(x) = 0 despite disturbances and uncertainties. ϕ (xe,k ,uk )
k
Considering a quadratic Lyapunov V (x) function of the y m,k = C m,e xe,k + v. (34)
form
1 1 1 Here, Ts denotes the sampling time, xe,k the extended
V (x) = s(x)T s(x) = s2φ + s2ψ , (26) state vector, uk the control input vector, and y m,k the
2 2 2
the control input vector usw must be chosen in such a way measured output at discrete-time tk . Furthermore, the
that the time derivative process noise and the measurement noise are given by w
and v, respectively. Both are assumed to be zero-mean
V̇ (x) = sφ ṡφ + sψ ṡψ ≤ −ηφ |sφ | − ηψ |sψ | < 0 (27)
Gaussian white noise processes with zero cross-correlation.
is negative definite. Substituting Eq. (20) and Eq. (24) into The vanishing cross-correlation leads to diagonal covari-
Eq. (21) and solving for the switching control input usw ance matrices Q and R characterising the process noise
leads to w and the measurement noise v, respectively. Fig. 4 shows
lt cos(ψ) lm that the discrete-time EKF implementation can be divided
V̇ (x) = sφ usw,φ +sψ usw,ψ < 0 . (28)
Jφ Jψ into two stages, namely a prediction stage and an innova-
      tion stage Stengel (1994). The error covariance matrix is
−ηφ sgn(sφ ) −ηψ sgn(sψ ) y m,k+1
The parameters ηφ > 0 and ηψ > 0 determine the switch- x̃e,k , P̃ k
ing heights and guarantee that the Lyapunov function
becomes negative definite.
The integral sliding mode control presented so far is sub- Innovation
Prediction
ject to the chattering phenomenon. This involves fast L̃k+1
switching actions introduced by the sgn function, and x̃e,k+1
x̂e,k+1
may lead to the excitation of unmodelled high-frequency P̃ k+1
dynamics. To counteract this effect, smooth switching P̂ k+1
functions tanh(si /) with a strictly positive constant  > 0
– influencing a boundary layer thickness – are utilized. The
chattering reduction depends on value of  at the cost of Fig. 4. Implementation of the discrete-time EKF.
robustness. The larger the value of , the more the chatter-
ing phenomenon is suppressed. At the same time, however, denoted by P k . The algorithm for the discrete-time EKF
the robustness will be reduced and the ideal sliding mode can be summarized at each time tk as follows, cf. Stengel
becomes a real sliding mode with corresponding tracking (1994):
errors. • State prediction
The control laws are implemented in a discrete-time form
with a small sampling time of Ts = 1 ms. The control x̃e,k+1 = ϕk (x̂e,k , uk ) (35)
design, hence, represents a quasi-continuous one. • Prediction of the error covariance matrix P̃ k+1
∂ϕ(x̂e,k , uk ) 
4. DISCRETE-TIME EXTENDED KALMAN FILTER P̃ k+1 = Φk P̂ k ΦTk + Q , with Φk = 
∂xe,k x̂e,k

(36)
To estimate the angular velocities as well as the lumped
disturbance torques, a discrete-time extended Kalman • Update of the gain matrix L̃k+1
filter (EKF) is employed. For this purpose, integrator  −1
disturbance models are introduced according to L̃k+1 = P̃ k+1 C Tm,e C m,e P̃ k+1 C Tm,e + R (37)
• Update of the state vector x̂e,k+1
żφ = 0, and żψ = 0. (29)  
Note that these disturbance models are excited in the case x̂e,k+1 = x̃e,k+1 + L̃k+1 y m,k+1 − C m,e x̃e,k+1 (38)
of output errors between the measured and the estimated • Update of the error covariance matrix for the next
output variables. The state vector of the extended system sampling interval
representation results in  
 T P̂ k+1 = I − L̃k+1 C m,e P̃ k+1 (39)
xe = φ φ̇ ψ ψ̇ zφ zψ , (30)
The closed-loop stability of the overall control structure
and the measurement vector becomes consisting of the multi-variable integral sliding mode con-
 
100000 trol and the discrete-time EKF has been investigated thor-
y m = C m,e xe = x . (31)
001000 e oughly by simulations.
Given the continuous-time state equation of the extended
system 5. IMPLEMENTATION OF THE CONTROL
STRUCTURE
ẋe = f (xe , u), (32)
an explicit Euler time-discretization of Eqs. (32) as well The time-discrete implementation scheme of the multi-
as an introduction of additive noise processes lead to the variable integral sliding mode control combined with a

805
MATHMOD 2015
806
February 18 - 20, 2015. Vienna, Austria Saif S. Butt et al. / IFAC-PapersOnLine 48-1 (2015) 802–807

  5
φd,k
4.5 φd
 
 φ̇d,k  4
φ
φ̈d,k Integral uk 3.5
2-DOF 3
  sliding mode

φ in rad
ψd,k helicopter y m,k 2.5
controller
  2
 ψ̇d,k 
1.5
ψ̈d,k 1
Discrete-
time 0.5
x̂e,k EKF 00 10 20 30 40 50 60 70 80 90 100 110
t in s
Fig. 5. Control implementation.
Fig. 6. Trajectory tracking w.r.t. the azimuth angle.
discrete-time EKF is shown in Fig. 5.
The TRAS is equipped with a dedicated I/O board and a 1.2 ψd
power interface. The control computer communicates with
1.0 ψ
the incremental sensors and motors interfaced by means
of the dedicated I/O board. The I/O board is operated 0.8

ψ in rad
by a real-time software within the MATLAB/Simulink
environment. 0.6

0.4
6. EXPERIMENTAL RESULTS
0.2

The parameter values for the spring and the damping 00 10 20 30 40 50 60 70 80 90 100 110
coefficients identified experimentally with the help of a t in s
nonlinear least-squares minimization are listed in Table 1.
Fig. 7. Trajectory tracking w.r.t. the pitch angle.
Table 1. Parameters of the TRAS 0.2
0.15
Parameter Value Units
0.1
cψ 0.095 Nm · s
0.05
cφ 0.008 Nm · s
e in rad

kφ 0.060 Nm 0
0.05
The counterweights have been adjusted in such a way -0.1
that the the azimuth angle and the pitch angle lies in the -0.15 eφ = φ − φd
following range: -0.2 eψ = ψ − ψd
0 rad ≤ φ ≤ 6.5 rad -0.250 10 20 30 40 50 60 70 80 90 100 110
t in s
0 rad ≤ ψ ≤ 2 rad.
The covariance matrices Q, R and initial error covariance Fig. 8. Tracking error w.r.t. the azimuth angle and the
matrix P 0 for the discrete-time EKF have been chosen pitch angle.
properly to achieve fast error a convergence rate as well as
to minimize the effect of measurement and process noise 0.6
Fφ
on the estimates. The controller is activated at t = 5 s. 0.5
The desired and measured trajectories for the azimuth 0.4 Fψ
angle, cf. Fig. 6, are in good agreement during both the 0.3
transient phase as well as in steady-state.
F in N

0.2
Subsequently, an excellent trajectory tracking behaviour 0.1
is obtained for the pitch angle as depicted in Fig. 7. 0
The corresponding errors in the azimuth angle and the
-0.1
pitch angle are shown in Fig. 8. It can be seen that the
-0.2
maximum steady-state errors are smaller than 0.02 rad
for both the azimuth angle and the pitch angle. On the -0.30 10 20 30 40 50 60 70 80 90 100 110
t in s
other hand, the maximum absolute error during the tran-
sient phase is 0.25 rad and 0.1 rad for the azimuth angle
Fig. 9. Control inputs w.r.t. the pitch axis and azimuth
and the pitch angle, respectively. The proposed controller,
axis.
hence, tracks the desired trajectories well and furthermore,
provides steady-state accuracy. The corresponding control not directly available by measurements. The correspond-
inputs for the pitch axis and azimuth axis are shown in ing estimates of the EKF for the angular velocities are
Fig. 9. depicted in Fig. 10.
The angular velocities of the pitch and azimuth axis are Fig. 11 shows the estimated lumped disturbance torques

806
 MATHMOD 2015
February 18 - 20, 2015. Vienna, Austria Saif S. Butt et al. / IFAC-PapersOnLine 48-1 (2015) 802–807 807

0.8 Butt, S.S., Prabel, R., and Aschemann, H. (2014). Multi-

0.6 φ̇ˆ Variable Flatness-Based Control of a Two Degrees of
Freedom Helicopter. In Control, Decision and Infor-
0.4 ψ̇ˆ
mation Technologies (CoDIT), 2014 International Conf.
ψ̇, φ̇ in rad/s

0.2 on, 321–326.

0 Darus, I.Z.M., Aldebrez, F., and Tokhi, M. (2004). Para-
-0.2 metric Modelling of a Twin Rotor System using Genetic
Algorithms. In Control, Communications and Signal
-0.4
Processing, 2004. First International Symposium on,
-0.6 115–118.
-0.80 10 20 30 40 50 60 70 80 90 100 110
Dutka, A., Ordys, A., and Grimble, M. (2003). Non-
t in s Linear Predictive Control of 2 DOF Helicopter Model.
In Decision and Control, 2003. Proc. 42nd IEEE Conf.
Fig. 10. Estimated angular velocities. on, volume 4, 3954–3959 vol.4.
acting on the pitch and azimuth axis. The magnitudes of INTECO (2013). Two Rotor Aero-Dynamical System
the estimated disturbance torques show that it is vital User’s Manual.
to consider them in the control design for an accurate Lopez-Martinez, M., Diaz, J.M., Ortega, M., and Rubio,
trajectory tracking. F. (2004). Control of a Laboratory Helicopter using
Switched 2-Step Feedback Linearization. In American
0.3
ẑφ
Control Conf., 2004. Proc. of the 2004, volume 5, 4330–
0.25 4335 vol.5.
0.2
ẑψ Lopez-Martinez, M., Ortega, M., and Rubio, F. (2003).
An H∞ Controller for a Double Rotor System. In
z in Nm

0.15 Emerging Technologies and Factory Automation, 2003.

0.1 Proc. ETFA ’03. IEEE Conf., volume 1, 253–259 vol.1.
Lopez-Martinez, M., Vivas, C., and Ortega, M. (2005). A
0.05
Multivariable Nonlinear H∞ Controller for a Laboratory
0 Helicopter. In Decision and Control, 2005 and 2005
-0.05
European Control Conf. CDC-ECC ’05. 44th IEEE
0 10 20 30 40 50 60 70 80 90 100 110 Conf. on, 4065–4070.
t in s
Mondal, S. and Mahanta, C. (2012). Adaptive Second-
Fig. 11. Estimated disturbance torques. order Sliding Mode Controller for a Twin Rotor Multi-
Input-Multi-Output System. Control Theory Ap-
plications, IET, 6(14), 2157–2167. doi:10.1049/iet-
7. CONCLUSION cta.2011.0478.
Rahideh, A. and Shaheed, M.H. (2007). Mathematical
In this paper, a nonlinear control-oriented model of a two
Dynamic Modelling of a Twin-Rotor Multiple Input-
degree of freedom helicopter is derived. The propulsive
Multiple Output System. Proc. of the Institution of
forces acting on the main rotor and the tail rotor are
Mechanical Engineers, Part I: Journal of Systems and
chosen as control inputs. A multi-variable integral sliding
Control Engineering, 221(1), 89–101.
mode control is proposed to track desired trajectories for
Shaheed, M. (2004). Performance Analysis of 4 types of
the azimuth angle as well as the pitch angle. For the esti-
Conjugate Gradient Algorithms in the Nonlinear Dy-
mation of the unmeasured angular velocities and lumped
namic Modelling of a TRMS using Feedforward Neural
disturbance torques acting on the 2-DOF helicopter, a
Networks. In Systems, Man and Cybernetics, 2004 IEEE
discrete-time EKF is employed. Thereby, given process and
International Conf. on, volume 6, 5985–5990 vol.6.
measurement noise can be addressed properly. Experiment
Slotine, J. and Li, W. (1991). Applied Nonlinear Control.
results show that an excellent tracking behaviour with
Prentice-Hall, Inc.
small errors is obtained for both outputs.
Stengel, R. (1994). Optimal Control and Estimation.
REFERENCES Dover Publications, Inc.
Young, K., Utkin, V., and Ozguner, U. (1999). A Control
Ahmad, S., Chipperfield, A., and Tokhi, M. (2000a). Dy- Engineer’s Guide to Sliding Mode Control. Control
namic Modeling and Optimal Control of a Twin Rotor Systems Technology, IEEE Trans. on, 7(3), 328–342.
MIMO System. In National Aerospace and Electronics
Conf., 2000. NAECON 2000. Proc. of the IEEE 2000,
391–398.
Ahmad, S., Shaheed, M., Chipperfield, A., and Tokhi,
M. (2000b). Nonlinear Modelling of a Twin Rotor
MIMO System using Radial Basis Function Networks.
In National Aerospace and Electronics Conf., 2000.
NAECON 2000. Proc. of the IEEE 2000, 313–320.
Ahmed, Q., Bhatti, A., and Iqbal, S. (2009). Nonlinear
Robust Decoupling Control Design for Twin Rotor Sys-
tem. In Asian Control Conf., 2009. ASCC 2009. 7th,
937–942.

807