Beruflich Dokumente
Kultur Dokumente
Qadeer Ahmed
June 2009
ii
To my Family
iii
Declaration
I, Qadeer Ahmed, honestly declare that I have worked out my Master of Science thesis
individually and all resources that I have used are mention in the references.
_____________________
Qadeer Ahmed
MT081011
iv
Abstract
This thesis deals with robust control strategies from linear and nonlinear
techniques for helicopter system. This system is prone to highly disturbing
interstate cross-couplings and perturbations in center of gravity that affects
smooth flights. The presented controllers offer solutions for smooth
tracking in presence of strong cross-couplings and disturbing torques
caused by perturbation in center of gravity. The H controller employed
from linear robust control theory involves traditional and Hadamard
weights in controller synthesis process. However the designed controller
works for linear range only and robustness is achieved at the cost of
performance or vice versa. The second attempt involves the sliding mode
controllers from nonlinear theory. This technique delivers solution against
cross-couplings and disturbing torques caused by variation in center of
gravity is solved by using 2-sliding mode controller. These nonlinear
controllers control the system in nonlinear range. Meanwhile an attempt to
design sliding surface from Linear Matrix inequalities algorithms delivers
the solution which is not suitable for practical implementation. The
designed controllers are validated by implementing on helicopter model,
after successful numerical simulations.
Acknowledgments
First and foremost I would like to thank Allah Subhana Wataallahu, who gave me
the courage, guidance and atmosphere to carry on my postgraduate studies in Pakistan.
The perseverance and determination granted by Almighty Allah helped me to bear the
hard times to produce this thesis.
I acknowledge the efforts of my parents who kept me motivated, guided and
focused throughout my MS. Their help in various regards contributed in keeping my
moral high. Apart from this, I would admire my spouse for being cooperative and
supportive during my masters tenure. Her responsible nature made me work free of
deviation and stress.
I consider myself blessed that I found a supervisor like Dr. Aamer Iqbal Bhatti
and mentor like Mr. Sohail Iqbal. The way they developed my skills in control systems
has really contributed in my advance theoretical and practical skills. Their wealth of
ideas, clarity of thoughts, enthusiasm and energy have made my working with them an
exceptional experience. I cannot overstate my gratitude and appreciation for their
encouragement, support and cooperation.
My special thanks to Mr. Nadeem Javaid, who granted me the access to the
helicopter model under his supervision. There I was able to apply my theoretical ideas
and verify my results. That later on contributed in various publications in international
conferences and journals.
I am also grateful to Control and Signal Processing research group members: Ijaz
Kazmi, Mudassar Rizvi, Khubaib Ahmed, Armaghan Mohsin, Muhammad Iqbal, Qudrat
Khan and many others, whose constructive comments and suggestions contributed in
clarifying various concepts.
vi
List of Publications
International Conferences:
Q. Ahmed, A. I. Bhatti, M. A. Rizvi, LMI Based Sliding Mode Control Design for Twin
Rotor System to be presented in SIAM Conference on Control and Its Applications 2009,
Colorado, USA.
Q. Ahmed, A. I. Bhatti, S. Iqbal Nonlinear Robust control design for Decoupling of Twin
Rotor System to be presented in Asian Control Conference (ASCC'09), Hong Kong.
Q. Ahmed , A. I. Bhatti, S. Iqbal, Robust Decoupling Control Design for Twin Rotor
System using Hadamard Weights, to be presented in CCA, MSC 2009, St. Petersburg,
Russia.
Qadeer Ahmed, Aamer Iqbal Bhatti, Sohail Iqbal, Syed Ijaz Kazmi 2-Sliding Mode Based
Robust Control for 2-DOF Helicopter, International workshop on Variable Structure
System VSS 2010, Mexico
vii
Table Of Contents
Abstract
................................................................................................................ v
Acknowledgments ............................................................................................................. vi
List of Publications ........................................................................................................... vii
Table Of Contents ............................................................................................................ viii
List of Figures
............................................................................................................... xi
List of Tables
............................................................................................................. xiv
Chapter 1
Introduction ......................................................................................... - 2 -
Chapter 2
2.1
2.1.1
2.1.1.1
2.1.1.2
2.1.1.3
Gyroscopic Torque............................................................................ - 10 -
2.1.1.4
2.1.2
2.1.3
2.1.4
2.2
Mathematical model...................................................................................... - 13 -
2.2.1
2.2.2
2.2.3
2.2.4
2.2.5
2.2.6
2.3
2.3.1
2.3.2
2.3.3
viii
2.3.4
2.3.5
Controllability ....................................................................................... - 26 -
2.3.6
Observability ......................................................................................... - 26 -
2.4
Control Challenges........................................................................................ - 27 -
Chapter 3
3.1
3.1.1
3.1.1.1
PID Controller................................................................................... - 29 -
3.1.1.2
3.1.2
3.1.3
3.1.4
3.1.5
3.1.6
3.2
3.2.1
3.2.1.1
3.2.1.2
3.2.2
3.2.2.1
3.2.2.2
3.2.3
3.2.3.1
3.2.4
3.3
Near-Decoupling: State Feedback .................................................... - 41 Hadamard Weights in LSDP for Robust Decoupling ........................... - 42 -
Chapter 4
4.1
4.2
4.2.1
4.2.2
4.3
ix
4.3.1
4.3.2
4.3.3
Simulations Results............................................................................... - 55 -
4.4
4.5
4.6
Conclusion .................................................................................................... - 64 -
Chapter 5
5.1
5.2
5.2.1
5.2.2
5.2.2.1
5.2.2.2
5.2.2.3
5.2.3
5.2.3.1
5.3
5.3.1
5.3.1.1
5.3.1.2
Chapter 6
Chapter 7
Appendix
................................................... - 98 -
7.1
7.2
7.3
7.4
7.5
7.6
List of Figures
Figure 2-1: Symbolic representation of helicopter with 2-DOF ..................................... - 7 Figure 2-2: Gravitational and centrifugal forces acting of helicopter in vertical plane .. - 9 Figure 2-3: Main rotor torque caused by 1 in vertical plane ....................................... - 10 Figure 2-4: Gyroscopic torque caused due to rate of change of azimuth in vertical plane.. 10 Figure 2-5: Net torques acting on the helicopter model in vertical plane ..................... - 11 Figure 2-6: Mechanical Torques produced in horizontal plane ................................... - 12 Figure 2-7: Block diagram of nonlinear model of twin rotor system ........................... - 14 Figure 2-8: CE150 HUMUSOFT Helicopter Model .................................................... - 19 Figure 2-9: Schematic diagram of helicopter model..................................................... - 19 Figure 2-10: Linear model validation in vertical plane against an Impulse input ........ - 21 Figure 2-11: Root locus of SISO systems given in G(s) ............................................... - 23 Figure 2-12: Bode plots of SISO systems given in G(s) ............................................... - 24 Figure 2-13: Singular values of MIMO system ............................................................ - 25 Figure 2-14: Phase portrait of Elevation ....................................................................... - 25 Figure 2-15: Phase portrait of Azimuth ........................................................................ - 26 Figure 3-1: General Structure to decouple the system .................................................. - 35 Figure 3-2: Decoupling control system (Boksenbom and Hood) ................................. - 35 Figure 3-3: Non-interacting decoupling control structure (Zalkind &Luyben) ............ - 38 Figure 4-1 General control configuration for H control ............................................. - 45 Figure 4-2: One degree of freedom configuration ........................................................ - 47 Figure 4-3: S/KS mixed sensitivity Plant configuration for tracking control .............. - 48 Figure 4-4 W1, Weighting function for S...................................................................... - 49 Figure 4-5 W2, Weighting function for KS ................................................................... - 49 Figure 4-6 Step response (Mixed sensitivity) ............................................................... - 50 Figure 4-7 Control Signal (Mixed sensitivity) .............................................................. - 51 Figure 4-8: H Robust Stabilization Problem ............................................................... - 51 Figure 4-9 LSDP implementation ................................................................................. - 53 Figure 4-10: Modified Singular Values using Traditional Weights in LSDP .............. - 54 Figure 4-11: Modified Singular Values with Hadamard Weights in LSDP ................. - 55 xi
Figure 4-12: Step response of the system with Traditional Weighted H controller ... - 56 Figure 4-13: Control Effort of Traditional Weighted H controller ............................. - 56 Figure 4-14: Step response of the system with Hadamard Weighted H controller..... - 57 Figure 4-15: Control Effort of Hadamard Weighted H controller .............................. - 57 Figure 4-16: Actual System response with Traditional Weighted H controller, when
exposed to coupling at 32 sec. ...................................................................................... - 58 Figure 4-17: Traditional Weighted H controller effort to over come coupling effects
introduced at 32 sec in azimuth plane. .......................................................................... - 59 Figure 4-18: Actual System response with Traditional weighted H controller to attain
equilibrium position when initialized in nonlinear range ............................................. - 59 Figure 4-19: Traditional Weighted H controller effort to acquire equilibrium position
when actual system was initialized in nonlinear range. ................................................ - 60 Figure 4-20: Traditional Weighted H controller response to coupling when robustness
was compromised with performance ............................................................................ - 60 Figure 4-21: Actual System response with Hadamard Weighted H controller, when
exposed to coupling at 32 sec ....................................................................................... - 61 Figure 4-22: Hadamard Weighted H controller effort to over come coupling effects
introduced at 32 sec in azimuth plane ........................................................................... - 62 Figure 4-23: Undershoots in the multi step response (Minimum phase behavior) of
helicopter system with 2-DOF ...................................................................................... - 62 Figure 5-1: Regulation control of Helicopter outputs ................................................... - 74 Figure 5-2: Phase Portrait for Elevation dynamics ....................................................... - 74 Figure 5-3: Phase portrait for Azimuth Dynamics ........................................................ - 75 Figure 5-4: Sliding mode controller effort for regulation ............................................. - 75 Figure 5-5: Sliding manifolds convergence which ensures states convergence ........... - 76 Figure 5-6: Response of Helicopter system with sliding mode controller when exposed to
coupling at 22 seconds .................................................................................................. - 77 Figure 5-7: Sliding mode Controller effort to decouple when exposed to coupling at 22
seconds. ......................................................................................................................... - 77 Figure 5-8: Response of Actual system with sliding mode control when initialized in
nonlinear range.............................................................................................................. - 78 -
xii
Figure 5-9: Sliding mode Controller effort to reach equilibrium position when released in
nonlinear range.............................................................................................................. - 78 Figure 5-10: Output States response for LMI based Sliding mode control .................. - 83 Figure 5-11: LMI based sliding mode controller effort ................................................ - 84 Figure 5-12: LMI based sliding surface convergence................................................... - 84 Figure 5-13: Regulation response of output states of dynamical model of helicopter . - 88 Figure 5-14: Phase portraits of elevation and azimuth dynamics ................................. - 88 Figure 5-15: Response of helicopter model in Elevation when exposed uncertainty in
center of gravity ............................................................................................................ - 89 Figure 5-16: Variations in center of gravity serving as parametric uncertainty ........... - 90 Figure 5-17: 2-SMCController effort to acquire and maintain equilibrium position in
elevation plane .............................................................................................................. - 90 Figure 5-18: Elevation dynamics sliding manifold convergence ................................. - 91 Figure 5-19 Response of helicopter model with 2-SMC controller in horizontal plane . - 92
Figure 5-20: 2-SMC Controller effort to maintain equilibrium position in azimuth .... - 92 Figure 5-21: Azimuth dynamics sliding surface convergence ...................................... - 93 Figure 7-1 SIMULINK Diagram for H Implementation on Helicopter Model ........ - 104 Figure 7-2 SIMULINK Diagram for H Implementation on Helicopter Model ........ - 105 Figure 7-3 SMC Controller Block .............................................................................. - 106 Figure 7-4 Elevation Controller Block ....................................................................... - 106 Figure 7-5 Azimuth Controller Block ......................................................................... - 107 Figure 7-6 SIMULINK Block Diagram for 2-SMC ................................................... - 108 Figure 7-7 2-SMC controller Block ............................................................................ - 109 -
xiii
List of Tables
Table 2-1 System specifications (HUMUSOFT CE 150 Manual) ............................... - 21 Table 2-2: Eigenvalues of helicopter model ................................................................. - 22 Table 4-1: Performance Indices .................................................................................... - 63 Table 5-1: Performance indices of Elevation Dynamics .............................................. - 79 Table 5-2: Performance indices of Azimuth Dynamics ................................................ - 79 Table 6-1: Comparison between proposed controllers in this thesis ............................ - 96 -
xiv
Chapter 1
Introduction
Chapter Objectives:
Background
Motivation
___________________________________- 2 -_______________________________
Helicopter is an aircraft that is lifted and propelled by one or more horizontal rotors,
each rotor consisting of two or more rotor blades. Helicopters are classified as rotorcraft
or rotary-wing aircraft to distinguish them from fixed-wing aircraft because the helicopter
achieves lift with the rotor blades which rotate around a mast. The primary advantage of a
helicopter is due the rotor which provides lift without the aircraft needing to move
forward, allowing the helicopter take off and land vertically without a runway. For this
reason, helicopters are often used in congested or isolated areas where fixed-wing aircraft
cannot take off or land. The lift from the rotor also allows the helicopter to hover in one
area, more efficiently than other forms of vertical takeoff and landing (VTOL) aircraft,
allowing it to accomplish tasks that fixed-wing aircraft cannot perform [1]
Helicopters are under-actuated an mechanical system that means we have to
perform maneuvers in all six degrees of freedom and available actuators to perform these
tasks are limited to just two in number. This task inherently induces cross-couplings in
the systems dynamics i.e. each actuator must have some of its affect in all of the six
degrees of freedom. However, these cross-couplings should be under control of the
operator so that desired maneuvers can be performed at ease, else uncontrolled crosscoupling can lead to fatal accidents causing human life losses.
Moreover, the disturbance torque caused by perturbations in center of gravity
(CG) also affects helicopter flight adversely, which certainly adds responsibilities for the
on board pilot. This unwanted torque must be compensated for the sake of smooth and
comfortable flight. These disturbing moments can occur as a consequence of weight
variations loaded on board during flight like turbulence in the fuel tank, coolant tanks and
hydraulic fluids tanks. Other causing agents may include the movements of passengers
during flight, wind gusts etc. Similarly relief luggage not loaded about the center of
gravity will continuously cause torque on the helicopter, forcing the nose tip to either
bend down or tilt up thus disturbing the normal helicopter flight. The ideal condition is to
have the helicopter in such perfect balance that the fuselage will remain horizontal in
hovering flight. The fuselage acts as a pendulum suspended from the rotor. Any change
in the center of gravity changes the angle at which it hangs from this point of support and
introduces additional torques that disturbs the flight [2], [3]. This demands a skillful pilot
and adds more responsibilities for the on board pilot.
___________________________________- 3 -_______________________________
The thesis considers solutions for above mentioned problems. The smooth flight
in the presence of cross-coupling and CG perturbations is at stake. However, solutions
from robust control theory may solve the problems and deliver best out of the available
mechanical helicopter structure. The first step involved for above mentioned problems is
modeling of the helicopter system. This modeling will later on contribute in indentifying
the core factors causing disturbance in helicopter flight. The cross-couplings in helicopter
dynamics can be modeled or treated as disturbances; therefore we can attain the solution
to this factor. However, perturbations in CG can be dealt under robust control theory by
declaring it as parametric uncertainty.
The robust controllers employed to handle the above mentioned problems are
from linear and nonlinear theory. H controller [4] from linear control theory has proved
it robustness over the past few decades. This controller offers robustness against the
cross-coupling and parametric uncertainty at the cost of performance. The more the
cross-coupling and parametric uncertainty is catered the more the system looses it
performance. To overcome this problem, Hadamard weighting technique [5] has been
employed that slightly improves the performance and offers robustness at the same time.
Therefore, the two H designing techniques with Traditional and Hadamard weights are
considered for the solution. The traditional weighted H controller offers the control even
in nonlinear domain along with robustness but performance is slightly reduced, however
Hadamard weights do not operate in nonlinear domain but caters cross-coupling along
with desired performance.
Sliding mode control theory [6] has been exploited from the nonlinear control theory.
This technique has also proved its robust nature in the control history. The nonlinear
model of the system is utilized for the development of control algorithm. The control law
evolved as a result is usually not suitable for implementation due to chattering in it. This
problem is resolved by utilizing concepts from Higher Order Sliding Mode. The resultant
controller offers better performance and robustness as compared to its linear counterparts.
The thesis keeps the scope of the problem limited to cross-coupling affecting
horizontal and vertical plane dynamics only, along with uncertainty in CG. H and
___________________________________- 4 -_______________________________
sliding mode controllers have been employed to deliver robust solutions. The verification
of the designed algorithms has been carried out first in simulation and later on Humusoft
Twin rotor system is used to authenticate the control laws. The laboratory helicopter
presents higher coupling between dynamics of the rigid body and dynamics of the rotors
and yields a highly nonlinear, coupled dynamics. Additionally, it can be proved that
characteristic dynamics of the system is non minimum phase, exhibiting unstable zero
dynamics. This system has been extensively investigated yielding a number of control
applications that range from linear robust control techniques to more recent nonlinear
approaches a [7] ~ [17]. Therefore after analyzing the system in detail we will apply few
of the following discussed robust control algorithms to extract the desired performance
from the laboratory helicopter model.
This chapter delivered a brief idea about the overview and motivation for the thesis work.
The rest of the thesis includes: the detailed helicopter modeling and its in-depth analysis
is discussed in Chapter 2, strategies to handle couplings in the dynamical system will be
discussed in Chapter 3 along with literature review of the control algorithms that have
already been implemented to deliver the solutions for the fore-discussed problems.
Finally, proposed robust control algorithms from linear and nonlinear theory will cover
Chapter 4 and Chapter 5. These chapters will include controllers derivation and their
validation based on simulation and implementations carried on the helicopter model
___________________________________- 5 -_______________________________
Chapter 2
Helicopter Modeling
&
Analysis
Chapter Objectives:
Dynamical modeling of helicopter
Model Validation
Analysis of mathematical model
___________________________________- 6 -_______________________________
This chapter deals with mathematical model formulation and its analysis in detail.
The basic physical concepts of moment generation have been utilized to develop
differential equations for helicopter dynamics both in vertical and horizontal planes.
These differential equations are then utilized for dynamical analysis of the system after
developing its state space model and transfer function. The detailed analysis as a result
will formulate the basic objectives for the controller synthesis.
___________________________________- 7 -_______________________________
w l w sin
(2.1)
Where
Eq. (2.2) describes the centrifugal torque produced by centrifugal force during rotation in
horizontal plane.
Where
c = l Fc cos
(2.2)
F ml 2 sin
c
(2.3)
___________________________________- 8 -_______________________________
Pivot Point
Figure 2-2: Gravitational and centrifugal forces acting of helicopter in vertical plane
helicopter body, which will produce angular torque about the pivot point. Therefore we
can say that
1 F1 (1 )
(2.4)
a r 2
F1 mr r 2
Where
(2.5)
1 k1 2
Where
(2.6)
k1 m r l
___________________________________- 9 -_______________________________
(2.7)
Figure 2-4: Gyroscopic torque caused due to rate of change of azimuth in vertical plane
___________________________________- 10 -_______________________________
The Eq. (2.7) is based on the fact that main rotor speed is very high as compared to rate
of change of azimuth i.e. 1 .
f = B1
(2.8)
Where
Considering all the torques produced on the helicopter body as discussed above, the net
torque produced as shown in Figure 2-5 is
I1 1 c G w f
Where
(2.9)
Figure 2-5: Net torques acting on the helicopter model in vertical plane
In calculating the elevation dynamics some influences are neglected, e.g. stabilizing
motor reaction torque and varying air resistance depending on the turnings of the main
propeller. While the influence of the side motor on the elevation angle is almost
negligible, varying damping of body oscillation in elevation is noticeable. The influence
of the speed of the main propeller on friction torque in elevation is hardly to be modeled
analytically and must be evaluated by an experiment and, if significant, nonlinear
coupling must be introduced [7].
___________________________________- 11 -_______________________________
(2.10)
Where
The fictional torque and side rotor torque are calculated similarly to elevation dynamics
as they are proportional to rate of change of angular position and rotor speed respectively.
The main rotor reaction torque acting on azimuth can be estimated by first order transfer
function shown in Eq. (2.11).
T(s)= K r
Tor s +1
Tpr s +1
(2.11)
___________________________________- 12 -_______________________________
M1
1
T1s 1
(2.12)
Where
And the nonlinearity caused by the rotor can be estimated as second order polynomial
whose constants have been indentified as explained in [7].. Finally the torque induced in
helicopter body via motor can be given by the following equation.
1 a 2u1 bu1
(2.13)
Where
(2.14)
y k
(2.15)
Where
___________________________________- 13 -_______________________________
Elevation Angle
x2
x3
Angular speed in Elevation
Side motor speed
x4 2
x
Azimuth
Angle
x7 x7 Angular Moment caused by u1on Azimuth
x Elevation Angle
Y 2
x5 Azimuth Angle
(2.16)
(2.17)
Based on these states and above discussed dynamical equations we can now proceed for
the dynamical model for helicopter model.
1
( x1 u1 )
T1
(2.18)
___________________________________- 14 -_______________________________
x2 = x3
2
x3 = I ((a1 x1 ) +b1 x1 - B1 x3 - Tg sinx2 - K gyro u1 x6 cosx2 )
1
x4 =
1
(-x4 +u2 )
T2
(2.19)
(2.20)
x5 = x6
2
x6 = I ((a2 x4 ) +b2 x4 - B2 x6 +Tpr x7 - K rTor u1 )
(2.21)
x7 = -Tpr x7 K rTor u1
(2.22)
Eq. (2.18) represents the main motor dynamics estimated by 1st order transfer function,
Eq. (2.19) explains the elevation dynamics derived from physical laws, same
phenomenons have been utilized for equation of side motor and azimuth dynamics. Eq.
(2.22) describes the angular momentum caused by first input in horizontal plane. The
constants values can be found in [7]..
f ( x)
x x 0
(2.23)
1
(a1 x12 +b1 x1 - B1 x3 - Tg sinx2 - K gyro u1 x6 cosx2 )
I1
x3
Where
1
[b1
I1
Tg
B1 0 0 0 0] X
(2.24)
X StatesVector
___________________________________- 15 -_______________________________
X AX BU
Y CX DU
(2.25)
X R 71 : StatesVector
Y R 21 : OutputVector
Where
U R 21 : InputVector
A R
7 7
1
- T
1
0
b1
I1
B R 7 2
0
Tg
I1
B1
I1
1
T2
b2
I2
B2
I2
1
T
1
0
-K rTor
K T
r or
0
1
T2
0
0
0
0 1 0 0 0 0 0
C R 27
0 0 0 0 1 0 0
0 0
D R 22
0 0
1
Tpr
I2
Tpr
(2.26)
(2.27)
(2.28)
(2.29)
Based on state space model and the parameter values in Table 2-1, the 2 by 2 transfer
function of the helicopter model can be written as follows.
___________________________________- 16 -_______________________________
g
G ( s ) 11
g 21
g12
g 22
(2.30)
Where
g11 ( s )
g12 ( s )
g 21 ( s )
g 22 ( s )
Mq Cq G
(2.31)
Where
M R 22 : Inertial Matrix
C R 22 : Coriolis Matrix
G R 21 : Gravitational Matrix
q R 21
: State Vector
R 21
: Input TorqueVector
0
I 2
(2.32)
B
C 1
0
0
B2
(2.33)
T sin
G g
(2.34)
___________________________________- 17 -_______________________________
(2.35)
11 12
22 21
(2.36)
11 and 22 are the torques generated by main and side motors and their affect on the
elevation and azimuth respectively. 12 and 21 are the cross coupled torques generated
by side motor on elevation and main motor on azimuth respectively. The torques
equations can be computed from the Eq.(2.18) Eq.(2.22).
___________________________________- 18 -_______________________________
___________________________________- 19 -_______________________________
the helicopter body. The model is interfaced with desktop computer via Humusoft
MF624 data acquisition PCI card which is accessible in MATLAB Simulink environment
through Real-time Toolbox and Real Time Windows Target Toolbox. These toolboxes
provide us the liberty to access the encoder values and issue commands to DC motors and
servo system. The schematic diagram shown in Figure 2-9 gives a brief idea about the
helicopter model interfacing. The system is controlled by changing the angular velocities
of the rotors. This kind of action involves the generation of resultant torque on the body
of double rotor system that makes it to rotate in perpendicular direction of the rotor.
Some of the specifications are shown in Table 2-1 , more details can be found in [7].
System Outputs
50o in elevation
40o in Azimuth
Main Motor 1
Side Motor 2
System Parameters
T1
= 0.3 s
a1
= 0.105 N.m/MU
b1
= 0.00936 N.m/MU2
I1
= 4.37e-3 Kg.m2
B1
= 1.84e-3 Kg.m2/s
Tg
= 3.83e-2 N.m
T2
= 0.25 s
a2
= 0.033 N.m/MU
b2
= 0.0294 N.m/MU2;
Tor
= 2.7 s
Tpr
= 0.75 s
Kr
= 0.00162 N.m/MU
I2
= 4.14e-3 Kg.m2
B2
= 8.69e-3 Kg.m2/s
___________________________________- 20 -_______________________________
Kgyro
= 0.015 Kg.m/s
0.1
0.08
Elevation (rad)
0.06
0.04
0.02
-0.02
-0.04
-0.06
20
25
30
35
40
45
50
55
60
65
Time
Figure 2-10: Linear model validation in vertical plane against an Impulse input
___________________________________- 21 -_______________________________
Eigenvalues
Damping
Freq. (rad/s)
0.00
-1
0.00
-1.33
1.33
-2.10
2.10
-.211 + 2.95i
0.0711
2.96
-.211- 2.95i
0.0711
2.96
-3.33
3.33
-4.00
4.00
It can be seen that one pole of the system is on origin which will affect the stability of
our system and conjugate poles are having less damping ratio which will affect the
performance of the model. Few other techniques have been employed to have through
analysis of the system, so that we have detailed insight of the system behavior.
___________________________________- 22 -_______________________________
Root Locus
10
5
Imaginary Axis
Imaginary Axis
Root Locus
10
-5
-5
-10
-10
-5
-10
-5
Real Axis
10
15
Real Axis
Root Locus
4
Root Locus
x 10
3
0.5
Imaginary Axis
Imaginary Axis
2
1
0
-1
-2
-0.5
-3
-4
-10
-5
-1
-2
-1
Real Axis
1
Real Axis
4
8
x 10
___________________________________- 23 -_______________________________
Bode Diagram
Bode Diagram
-250
Magnitude (dB)
Magnitude (dB)
200
100
0
-100
-350
720
Phase (deg)
Phase (deg)
-200
0
-300
-180
-360
-1
10
10
0
-1
10
10
360
10
Bode Diagram
Bode Diagram
Magnitude (dB)
Magnitude (dB)
10
100
50
0
-50
50
0
-50
-100
360
Phase (deg)
-100
180
Phase (deg)
10
Frequency (rad/sec)
100
90
0
-2
10
10
Frequency (rad/sec)
-1
10
10
10
Frequency (rad/sec)
10
180
0
-2
10
-1
10
10
10
10
Frequency (rad/sec)
___________________________________- 24 -_______________________________
Singular Values
60
40
20
-20
-40
-60
-80
-100
-120
-2
10
10
-1
10
10
10
Frequency (rad/sec)
0.3
0.2
0.1
-0.1
-0.2
-0.3
-0.4
-0.1
-0.08
-0.06
-0.04
-0.02
0
Elevation
0.02
0.04
0.06
0.08
0.1
The azimuth dynamics are stable. It can be observed in Figure 2-15 that the azimuth
reaches to certain value instead of converging to zero. Rate of change of azimuth when
___________________________________- 25 -_______________________________
approaches to zero, determines the current azimuth value i.e. -0.58 radians, as shown in
Figure 2-15.
Azimuth Phase Portrait
1
0.8
0.6
0.4
0.2
-0.2
-0.4
-0.6
-0.8
-1
-0.5
-0.4
-0.3
-0.2
-0.1
0.1
Azimuth
2.3.5 Controllability
A control system is said to be controllable if, for all initial times and all initial
states, there exists some input function that drives the state vector to any final state at
some finite time [19]. The controllability matrix of the LTI system is defined by the pair
(A,B) as follows:
C ( A, B) B
AB
A2 B An 1 B
(2.37)
The LTI system in Eq. (2.25) is said to be controllable if C matrix has rank n. Therefore
for helicopter model if we place the A,B values from Eq.(2.29), the rank of C matrix
comes out to be 7, which concludes that our model is completely controllable.
2.3.6 Observability
A control system is said to be observable if, for all initial times, the state vector can be
determined from the output function [19]. The observability matrix of the LTI system is
defined by the pair (A,C) as follows
___________________________________- 26 -_______________________________
O( A, C ) C CA CA2 CAn 1
(2.38)
The LTI system defined in Eq.(2.25) is said to be observable if O matrix has rank n.
Helicopter model defined in Eq.(2.29) delivers an observable matrix O with rank 7.
Therefore this test verifies that our model is completely observable.
___________________________________- 27 -_______________________________
Chapter 3
Control Algorithms
&
Decoupling Techniques
Chapter Objectives
Linear/Nonlinear Controllers Review
Decoupling Techniques Review
___________________________________- 28 -_______________________________
The objective of this chapter is to indicate the control algorithms that have been
already utilized by the control community to meet the goals formulated in analysis
process for helicopter model. Helicopter control is one of the challenging problems faced
by control engineers. The cross-couplings in its dynamics and parametric uncertainties
lead to foundation of designing robust controllers. Furthermore, this chapter will explore
some decoupling techniques that will be later on utilized to attain decoupling in
helicopter dynamics. The CE 150 helicopter model has been used for validation of
control algorithms, which has MATLAB as working environment. The advantage of this
environment is that the designed controllers are easy to implement but the processing rate
cannot be achieved beyond certain limit. However, one can manipulate the code for
implementation.
can be found in [7]. Following the practice discussed above the control law for the
helicopter model can be written as
U (s) ( K p
Ki
K d s) E ( s)
s
(3.1)
Although the control action from the PID controller is sufficient enough to deliver the
required response but its rise time and settling time is high and one cannot decrease the
response times due to limits applied by movement of poles from left half plane to right
half plane which will cause instability. Similarly this controller is not robust in nature and
is unable to withstand coupling effect when introduced in the system.
(3.2)
Although this control action provides the satisfactory performance results but still the
intentionally introduced coupling was still unhandled. Further details for pole placement
using feedback can be found in [7].
___________________________________- 30 -_______________________________
results show good tracking response but the coupling issue is not addressed in simulation
results. Jun et al [13] presented robust stabilization and H control for class of uncertain
systems. The quadratic stabilizing controllers for uncertain systems are designed by
solving standard H control problem. This method was verified by implementing it on
helicopter model. The analytical analysis of the algorithm after implementing it on
helicopter model tells that tracking has been achieved but the coupling is unaddressed in
this effort. M.Lopez et al [14] suggested H controller for helicopter dynamics. First
feedback linearization was used for decoupling the inputs and outputs, and then the
system was indentified at higher frequencies, as the relative uncertainty increases at
higher frequencies. The controller was designed for the system identified at higher
frequencies which was unable to cater coupling in the final results as shown in the results
given in the paper. M.Lopez et al [15] delivered the non linear H approach for handling
the coupling taken as disturbance. This approach considers a nonlinear H disturbance
rejection procedure on the reduced dynamics of the rotors, which includes integral terms
on the tracking error to cope with persistent disturbances. The resulting controller
exhibits the structure of non-linear PID with time varying constants according to system
dynamics. The implementation of this controller on the model shows remarkable decrease
in coupling caused by vertical plane dynamics on horizontal plane dynamics along with
good tracking results.
___________________________________- 32 -_______________________________
control of helicopter angular positions in vertical and horizontal planes. The sliding mode
controller delivers excellent regulation results, after the authors resolved the chattering
issue for the implementation by introducing the saturation and hyperbolic tangent
function which smooth out the chattering effect. The control law used for the regulation
is derived from the sliding surface which includes the sum of proportional, derivation and
integral components of the error dynamics as shown in Eq. (3.3)
t
2
1
u1 1 e1 e1 2 e1dt
T1 0
T1
(3.3)
2
1
u2 2 e3 e3 2 e3 dt
T2 0
T2
The results shown in are [22] are acceptable but still authors didnt considered coupling
effects on the performance parameters. Gwo R. Yu et al [16] considered sliding mode
control of helicopter model via LQR. The LQR was first applied to control the elevation
and azimuth dynamics and then sliding mode controller is employed to guarantee the
robustness. The results demonstrated are optimal in performance and robust but the
coupling problem is still visible in the results displayed. J.P. Su et al [23] designed
procedure that involves primarily finding an ideal inverse complementary sliding mode
control law for the mechanical subsystem with asserted good tracking performance.
Then, a terminal sliding mode control law is derived for the electrical subsystem to
diminish the error arises from the deviation of the practical inverse control from the idea
of inverse control for the mechanical subsystem
___________________________________- 33 -_______________________________
effect of chattering will be reduced. In other words, HOSM has two important features
that make it a better choice in designing the controller. It improves the accuracy of the
design, which is a very important issue, and may provide a continuous control. This
thesis also contributes towards the 2-SMC design for helicopter model. More details can
be found in Section 5.3.
___________________________________- 34 -_______________________________
(3.4)
___________________________________- 35 -_______________________________
*
*
G21 G22 Gc ,21 Gc ,22 0
0
q2
*
*
*
*
G21Gc ,11 G22Gc ,21 G21Gc ,12 G22Gc ,22 0
*
c
(3.5)
0
q2
(3.6)
(3.7)
0 G11Gc*,12 G12Gc*,22
(3.8)
0 G21Gc*,11 G22Gc*,21
(3.9)
q2 G21Gc*,12 G22Gc*,22
(3.10)
The diagonal component of Gc* are the simple controllers designed for the outputs but the
off diagonal terms of Gc* can be calculated by Eq.(3.8) and Eq. (3.9) respectively as
Gc*,12
G12Gc*,22
Gc*,21
G11
G21Gc*,11
G22
(3.11)
(3.12)
Now the every single is described in the forward path, finally we can say that our over all
system is decoupled for servo control problem as online tuning of the controllers will
change the off diagonal to be recalculated to avoid this retuning we will discuss other
scheme.
___________________________________- 36 -_______________________________
X GGc* diag ( x1 , x2 )
(3.13)
Gc* G 1 X
(3.14)
Where
G 1
G22
1
G11G22 G12G21 G21
G12
G11
(3.15)
G22 x1
1
G11G22 G12G21 G21 x1
G12 x2
G11 x2
(3.16)
The simplest form of this decoupling matrix has unity diagonal elements, i.e.
Gc*,11 Gc*,22 1
(3.17)
Gc*,12
G12
G11
(3.18)
Gc*,21
G21
G22
(3.19)
The above equations show that with this method, the decoupling elements are
independent of the forward path controllers. On-line tuning of the controllers, therefore,
does not require redesign of the decoupling elements; controller modes may be changed
say from PI to PID and either of the forward path controllers may be placed in manual
without loss of decoupling. Note also, that decoupling occurs between the forward path
control signals and the process outputs, and not between set-points and process outputs.
This technique is therefore not restricted to the servo problem [7].
___________________________________- 37 -_______________________________
The above discussed techniques are not applicable on the helicopter model as they
require access to each and every intermediate state of the system however in our system
we have only access to the two inputs and two outputs of the system, which restricts the
application of these techniques to helicopter model.
x Ax Bu
y Cx Du
(3.20)
G ( s ) C ( sI A) 1 B D
(3.21)
The system is said to be completely decoupled if G(s) is diagonal matrix and nonsingular. In order to achieve such a matrix from state-variable feedback control the
following two techniques are considered
___________________________________- 38 -_______________________________
u Kx Fr
(3.22)
H ( s ) [(C DK )( sI A BK ) 1 B D]F
(3.23)
The static decoupling problem by state feedback is solvable if and only if (A,B) is
A B
stabilizable and rank of
n m where n is the number of states and m is the
C D
number of inputs. Assuming these conditions are satisfied the problem is solvable by
proceeding with the following steps.
i.
ii.
iii.
This technique has been implemented in [7] where A,B,C,D are given in Eq. (2.29) and
more details of the technique can be found in [29].
F ( B*) 1
(3.24)
K ( B*) 1 C *
(3.25)
The resultant feedback system has the transfer function in Eq.(3.23) as diagonal matrix.
Where
c1T A0 B
T 1
c AB
*
B 2
T m 1
cm A B
(3.26)
and
___________________________________- 39 -_______________________________
c1T A1
T 2
c A
*
C 2
T m
cm A
(3.27)
c1T
T
c
C 2
T
cm
(3.28)
and
This technique can be implemented on our system by taking A,B,C,D as given in Eq.
(2.29). More details about this technique can be found in [29].
B B1
B2
C1
D11
C
D
2
21
... B p , C , D
C p
D p1
D12
D22
Dp 2
D1 p
D2 p
D pp
(3.29)
Correspondingly, the closed loop transfer function defined in Eq.(3.23) can also split as
___________________________________- 40 -_______________________________
H11 ( s ) H12 ( s )
H ( s) H ( s)
21
22
H (s)
H p1 ( s ) H p 2 ( s )
H1 p ( s )
H 2 p ( s )
H pp ( s )
(3.30)
The described system is said to be nearly decoupled for input and output pairs (ui , yi ) if
the system (3.20) is stable and
( H cl ,ii ( j )) i
H cl ,ij
QAT AQ F T BT BF
BTj
(QCiT F T DiT )T
i {1, 2.....m}
Bj
I
Dij
QCiT F T DiT
DijT
0
(3.31)
(3.32)
i, j {1, 2.....m}, i j
(3.33)
K FQ 1
(3.34)
___________________________________- 41 -_______________________________
( H cl ,ii ( j )) i
H cl ,ij
(3.35)
(3.36)
The robust near decoupling and output feedback decoupling algorithms can be found with
detail in [29].
W21 W22
(3.37)
(3.38)
Even, if we chose W as diagonal matrix, the matrix multiplication procedure will relate
each weight term with each system dynamics. To avoid this relation and to introduce
decoupling in the system through these weights, F.Van Diggelen and K. Glover [5]
proposed element by element weighting which instead to relating weight components
shown in Eq. (3.38) relates the weight components element by element as
G12 W11 W12 G11W11 G12W12
G
W G 11
(3.39)
This procedure gives the liberty to handle each component of our plant independently.
Coupling can be catered by this technique by introducing off diagonal terms as zero or
___________________________________- 42 -_______________________________
very small positive numbers and diagonal terms can be introduced to reshape the plant
dynamics as required. This technique has been applied on helicopter model for
decoupling using LSDP which is explained in Section 4.3 with the design procedure in
detail and its results.
3.3 Conclusion
In view of the above literature survey, the already existing controllers to meet control
demands have been explored in detail. The coupling issue has been addressed by both
methods, one is to handle the coupling directly by introducing decoupling technique and
the second approach is to declare coupling as disturbance and design such a robust
controller that will cope up with desired performance even in the presence of coupling.
The suggested controller design procedures are discussed in next chapters.
___________________________________- 43 -_______________________________
Chapter 4
H
Controller Design
Chapter Objectives
Mixed Sensitivity procedure
Loop Shaping Design procedure
Hadamard Weights
Controllers Evaluations
___________________________________- 44 -_______________________________
(4.1)
The state space realization of the generalized plant [19] can be given as
___________________________________- 45 -_______________________________
A
P C1
C2
B2
D12
D22
B1
D11
D21
(4.2)
The input signals are u the control variable and w the exogenous inputs including the
disturbances and commands. The outputs include v the measured output and exogenous
outputs z which includes error or the signals which are to be minimized. The closed
loop transfer function from w to z can be formulated as
z Fl ( P, K ) w
(4.3)
Where
max ( Fl ( P, K )( j ))
(4.4)
(4.5)
function T I S .For the given closed loop configuration shown in Figure 4-2 in
which d is disturbance, n is noise in the measure sensors and r is the reference input.
In order to impose performance and robustness conditions the following close loop
___________________________________- 46 -_______________________________
relationships between output and the error, the control signal and the generalized
disturbances acting on the system must be taken in account.
y To r So d To n
(4.6)
e So (r d ) To n
(4.7)
u KSo (r n d )
(4.8)
Where So and To are the output sensitivity function and complementary sensitivity
function respectively.
From the above equations, we can conclude that shaping To is desirable for
tracking problems, noise attenuation and robust stability with respect to multiplicative
uncertainty. On the other hand, shaping So will allow to control performance of the
system. In addition the control signal should be attenuated along the frequency.
The stated problem can be solved out by means of mixed sensitivity problem S/KS [[4]].
The optimal control [19] is formulated as finding stabilizing controller K(s) such that
expression in Eq.(4.9) is minimized.
___________________________________- 47 -_______________________________
W1 ( s ) So ( s )
W ( s ) K ( s ) S ( s )
o
2
(4.9)
In the above expression, the W1 and W2 are the weighting function, which are employed
to shape the close loop transfer function So and KSo respectively. Figure 4-3 shows the
new augmented plant configuration.
Figure 4-3: S/KS mixed sensitivity Plant configuration for tracking control
After we have formulated the control problem the generalized plant for helicopter model
can be written as
P
P 11
P21
P12
P22
(4.10)
Where
W
P11 1
0
P21 I
W G
P12 1
W2G
P22 G
Where the values for transfer function G(s) can be taken from Eq. (2.30) and W1 and
W2 are formulated in the coming sections.
lower frequency to reject disturbances at the output and to reduce error tracking.
Therefore to achieve this aim singular values should have high gain at lower frequencies.
Figure 4-4 shows the selected W1 for our system.
Singular Values
-20
-30
-40
-50
-60
-70
-80
-90
-100
-3
10
10
-2
10
-1
10
10
10
10
Frequency (rad/sec)
-80
-100
-120
-140
-160
-180
-2
10
10
-1
10
10
10
10
10
10
10
10
10
Frequency (rad/sec)
___________________________________- 49 -_______________________________
The H controller has been designed using algorithm implemented in [19], which yields a
suboptimal controller. The controller designed delivered the results discussed in next
section.
From: In(2)
1.5
To: Out(1)
Angle (deg)
0.5
0
1
0.9
0.8
To: Out(2)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
0.5
1.5
2.5
3.5
4.5
5 0
0.5
1.5
2.5
3.5
4.5
Time (sec)
___________________________________- 50 -_______________________________
Control input
From: In(1)
From: In(2)
15
To: Out(1)
10
Voltage (V)
-5
0.015
To: Out(2)
0.01
0.005
-0.005
-0.01
0.5
1.5
2.5
3.5
4.5
5 0
0.5
1.5
2.5
3.5
4.5
Time (sec)
(4.11)
The perturbed plant is shown in Figure 4-8. The objective of the robust stabilization is to
stabilize not only the nominal model G but also the perturbed model given as
___________________________________- 51 -_______________________________
G ( M M ) 1 ( N N )
(4.12)
[ N M ]
(4.13)
min
1
max
1 N M
1
2 2
H
(1 ( XZ ))1/2
(4.14)
Where . H denotes Hankel norm, max denotes maximum stability margin [4 and
denotes the spectral radius. For a minimal state space realization (A,B,C,D) of G, Z is the
unique positive definite solution to the algebraic Riccati equation [19].
( A BS 1 DT C ) Z Z ( A BS 1 DT C )T ZC T R 1CZ BS 1 BT 0
(4.15)
Where
R I DDT ,
S I DT D
And X is the unique positive definite solution of the following algebraic Riccati equation.
( A BS 1 DT C )TX X ( A BS 1 DT C ) C T R 1C XBS 1 BT X 0
(4.16)
The plant should be strictly proper in nature to satisfy the above equations. The
controller is designed by the procedure defined in [19] and the controller K which
guarantees
K
1
1
I ( I GK ) M
(4.17)
___________________________________- 52 -_______________________________
The controller is implemented in the way as shown in the Figure 4-9 with pre and
post weighting functions in the series connection. In helicopter model, we have designed
pre weight function and chose post weight function to be identity, as we have simple
digital encoder as sensor and we cannot modify the sensor dynamics like we desire. The
weights are merged with the designed controller at the time of implementation. The new
controller comprises of both the weights and the controller obtained from the procedure
described in [19].
The pre weight function W1 designed for shaping the helicopter dynamics is
s 2 2.2 s 0.4
s 2 1000s 1
W1
s 2 2.2s 0.4
s 2 1000s 1
0
(4.18)
The weights designing procedure described in Eq. (3.38) is the standard practice adopted
in loop shape design procedure. This weight allows us to modify model dynamics by
improving its tracking by increasing the gains at lower frequencies and changing the
slope at cross over frequency to 1 as shown in Figure 4-10.
___________________________________- 53 -_______________________________
Singular Values
150
Gs
G
100
50
-50
-100
-150
-200
-3
10
10
-2
10
-1
10
10
10
10
Frequency (rad/sec)
The re-shaped model of helicopter can be utilized for controller synthesis as described in
[19]. In this practice the coupling is term as disturbance, the controller designed should
be so robust that it handles the coupling very efficiently.
2
s 1000 s 1
W1
1920 s 2 4224 s 768
s 2 1000 s 1
(4.19)
This pre weight function modified the plant singular values as shown in Figure 4-11.
___________________________________- 54 -_______________________________
Singular Values
150
Gs
G
100
50
-50
-100
-150
-3
10
10
-2
10
-1
10
10
10
10
Frequency (rad/sec)
The modified helicopter dynamics are used for controller synthesis as in the previous
case. The element by element weighting gives us the liberty to handle the coupling
directly by making the off diagonal elements of transfer function in Eq. (2.30) equal to
zero by the designed weight, which ensures the decoupling in the system; however this
affects the tracking performance in the consequence. So, we have to trade off between the
decoupling and tracking performance to get optimum results.
___________________________________- 55 -_______________________________
Step response
From: In(1)
From: In(2)
0
5
5
0
5
10
15
20 0
10
15
Figure 4-12: Step response of the system with Traditional Weighted H controller
Control input
From: In(1)
From: In(2)
0
0
10
15
20 0
10
15
Hadamard weights are also examined first on Matlab based simulations. The sub
optimal controller with min 2.3992 designed using the weighting procedure described
in Eq.(3.39). yields results shown in Figure 4-14 and the control effort is shown in Figure
4-15. It can be observed in the results that elevation and azimuth angles achieve the
desired value with in a finite time but the coupling is totally eliminated from system
dynamics. These simulations delivered the ideal results and the controller is now ready
for implementation on actual helicopter model.
___________________________________- 56 -_______________________________
Step response
From: In(1)
From: In(2)
1.5
To: Out(1)
0.5
Angle (deg)
-0.5
-1
1.5
To: Out(2)
0.5
-0.5
-1
10
12
14
16
18
20 0
10
12
14
16
18
20
Time (sec)
Figure 4-14: Step response of the system with Hadamard Weighted H controller
Control input
From: In(1)
From: In(2)
300
250
200
To: Out(1)
150
100
50
0
Voltage (V)
-50
-100
300
250
200
To: Out(2)
150
100
50
0
-50
-100
10
12
14
16
18
20 0
10
12
14
16
18
20
Time (sec)
___________________________________- 57 -_______________________________
Traditional weighting technique is employed. Figure 4-17 gives the idea about the control
effort. It can be observed that rise time is almost 15 to 17 seconds and there is no over
shoot in the response. However when the system settles down at equilibrium state and
second degree of freedom is introduced at 32 second, the coupling is visible both in
elevation and azimuth and with in 5 seconds the controller copes up with coupling and
maintains the equilibrium position. The controller effort to maintain the equilibrium state
can be seen in Figure 4-17 after 32 seconds. Figure 4-18 shows controller response in
when operated in nonlinear domain.
H inf controller (Normal Weights)
25
Elevation (deg)
20
15
10
5
0
-5
-10
10
15
20
25
10
15
20
25
30
35
40
45
30
35
40
45
20
Azimuth (deg)
10
-10
-20
Time (sec)
Figure 4-16: Actual System response with Traditional Weighted H controller, when exposed to
coupling at 32 sec.
___________________________________- 58 -_______________________________
Elevation
0.8
0.6
0.4
0.2
0
10
15
20
25
10
15
20
25
30
35
40
45
30
35
40
45
Azimuth
0.5
-0.5
-1
Time (sec)
Figure 4-17: Traditional Weighted H controller effort to over come coupling effects introduced at 32
sec in azimuth plane.
Output States Response Initialized in Nonlinear Domain
25
Elevation (deg)
20
15
10
5
0
10
15
20
25
10
15
20
25
30
35
40
45
50
30
35
40
45
50
40
Azimuth (deg)
20
-20
-40
Time (sec)
Figure 4-18: Actual System response with Traditional weighted H controller to attain equilibrium
position when initialized in nonlinear range
___________________________________- 59 -_______________________________
Controller Effort
1
Elevation
0.8
0.6
0.4
0.2
10
15
20
25
10
15
20
25
30
35
40
45
50
30
35
40
45
50
2
1.5
Azimuth
1
0.5
0
-0.5
-1
Time (sec)
Figure 4-19: Traditional Weighted H controller effort to acquire equilibrium position when actual
system was initialized in nonlinear range.
The rise time in this case is very high which makes the systems response sluggish,
however if we reduce the robustness of the controller to achieve better performance, the
H controller with min 3.914 implemented on the helicopter yields the results shown
in Figure 4-20 which shows that the rise time has been significantly decreased but
overshoot can be observed. The second degree of freedom is introduced at 31 seconds,
the controller is unable to handle coupling and in azimuth we can see that the helicopter
reaches -40 degrees and then after 20 seconds the controller is able to transport back
helicopter to equilibrium position.
Step resposne (Normal Wt)
25
Elevation (deg)
20
15
10
5
10
20
30
40
50
60
70
10
20
30
40
Time (sec)
50
60
70
40
Azimuth (deg)
20
-20
-40
Figure 4-20: Traditional Weighted H controller response to coupling when robustness was
compromised with performance
___________________________________- 60 -_______________________________
In the second attempt the H with Hadamard weighting technique is implemented on the
helicopter model. Figure 4-21 shows the response given by the helicopter to reach at
equilibrium level and Figure 4-22 shows its respective controller effort. It can be clearly
observed that the rise time is around 5 seconds and there is no over shoot. At the time
when second degree of freedom i.e. azimuth is introduced, the controller successfully
copes up with coupling and with in 2 to 3 seconds it assists the helicopter to gain its
equilibrium position. In the introduction of helicopter dynamics it was said that this
system is minimum phase system and its zero dynamics are unstable. The effect of right
half plane zeros make the helicopter to respond opposite to the command issued which
produces undershoots as a result. This typical response of minimum phase systems can be
seen in Figure 4-23. The H controller with Hadamard weighting technique is employed
for multi step command issued to reach the equilibrium position which delivers the
response shown in Figure 4-23.
H inf Control (Hadamard weights)
25
Elevation (deg)
20
15
10
5
0
-5
-10
10
15
20
10
15
20
25
30
35
40
45
30
35
40
45
40
Azimtuh (deg)
20
-20
-40
25
Time (sec)
Figure 4-21: Actual System response with Hadamard Weighted H controller, when exposed to
coupling at 32 sec
___________________________________- 61 -_______________________________
1
0.8
0.6
0.4
0.2
10
15
20
10
15
20
25
30
35
40
45
30
35
40
45
0.5
-0.5
-1
25
Time (sec)
Figure 4-22: Hadamard Weighted H controller effort to over come coupling effects introduced at 32
sec in azimuth plane
20
Elevation (deg)
15
10
10
15
20
25
30
35
40
45
50
55
60
Time
Figure 4-23: Undershoots in the multi step response (Minimum phase behavior) of helicopter system
with 2-DOF
___________________________________- 62 -_______________________________
ISE e 2 dt
(4.20)
Integral of square of error (ISE) is total sum of square of error that is mostly used to
evaluate practical implementations. Similarly, to reduce the contribution of large initial
error total sum of time multiple of error square (ITSE) or total sum of time multiple of
absolute error (ITAE) are used.
T
ITSE te 2 dt
(4.21)
ITAE t e dt
(4.22)
Total sum of absolute of error (IAE) is generally used for the evaluation of computational
simulations
T
IAE e dt
(4.23)
These performances indices give the idea about the performance calculated from the
error, the more the indices value the more will be low graded performance. The Table 4-1
shows the numeric values of the four performance indices of both horizontal and vertical
plane dynamics with both controllers. Table 4-1 shows indices of H with Traditional
weights and with Hadamard weights. These performance indices give us the idea that
elevation dynamics are well controlled by the H with Hadamard, however the
decoupling at the cost of performance [5] is well depicted as the azimuth dynamics loose
performance.
Performance
Indices
Elevation
(Traditional
Wt)
Elevation
(Hadamard Wt)
Azimuth
(Traditional
Wt)
Azimuth
(Hadamard Wt)
ISE
46.14
24.85
0.4031
3.127
ITSE
287.7
82.97
14.59
107.7
IAE
4.387
2.334
0.3114
0.8696
ITAE
44.92
18.01
11.61
31.51
Table 4-1: Performance Indices comparison between Traditionally and Hadamard Weighted H
___________________________________- 63 -_______________________________
4.6 Conclusion
The simulation results and the experimental tests lead to some very interesting
conclusions. The H controller design with mixed sensitivity procedure is not preferred
due to controller effort which is not as optimized as in LSDP and this technique has not
been explored for decoupling i.e. loop shaping design procedure. Therefore, LSDP
technique is selected for further controller design procedures and implementation.
Loop shaping design procedure provides the liberty to muddle through the
coupling by either ways. In the first attempt, the coupling is taken as disturbance and the
robustness of the controller provides the guarantee to over come its effect but at the cost
of performance which can be seen in the experimental tests results i.e. the more we
reduce the min the more robustness we get but at the same time the response time
increases. However in the LSDP with Hadamard weighting technique optimum
robustness and performance is achieved at the same time. The same conclusion can be
drawn from the controller efforts; the controller in Figure 4-17 has to exert more effort
after the coupling has been introduced as compared to in Figure 4-22 but controller in
vertical plane dynamics has to exert more effort in H control design with Traditional
weighting technique.
The robust control algorithms from linear domain discussed above deliver acceptable
results but the drawbacks of linear range still stick with them like these controllers should
be applied in linear range as the system is linearized around a set point. The linearization
forces to ignore dynamics of the system described non-linearly, which causes to have
loose control over those ignored dynamics. The performance indices showed that vertical
plane dynamics are well handled by H with Hadamard weights and the horizontal plane
dynamics are well handled by H with Traditional weights. So we are not yet having a
single controller that delivers desired results. Therefore we will now attempt for the
robust control algorithms from nonlinear domain to deliver the desired response.
___________________________________- 64 -_______________________________
Chapter 5
Nonlinear
Control Algorithms
Chapter Objectives
Lyapunov Based Control Design
Sliding Mode Control Design
Higher order Sliding Mode Control
Controllers Evaluations
___________________________________- 65 -_______________________________
The shortcomings in control laws generated from linear control guide us to engineer
control laws from nonlinear control algorithms. The nonlinear theory addresses each and
every dynamics of the systems which linear theory fails to address as the linear methods
rely on the key assumption of small range of operation for the linear model to be valid.
The required operating region for twin rotor system is beyond the capacity of linear range
especially in horizontal plane dynamics, which serves as basis to design nonlinear control
for helicopter model. These control algorithms are simpler to design, deeply rooted into
the system physics and are more intuitive than their linear counterparts. The nonlinear
controllers can handle the uncertainties or change in parameter much efficiently which
makes them also robust in nature. Keeping in mind the fore mentioned motivations
various nonlinear control algorithms have been attempted to deliver the desired objective
from the twin rotor system.
Based on this theory we can develop control law from energy equation which will
make helicopter model asymptotically stable. The helicopter model can be decomposed
in elevation and azimuth planes thus allowing us to model two energy functions for the
whole system. Therefore, we can choose the energy function based on state vector in Eq.
(2.16) for elevation dynamics as
1
V ( x) ( x12 x22 x32 )
2
(5.1)
(5.2)
The respective states are defined in Section 2.2. These energy functions are positive
definite and their negative rate of change will help us to calculate respective control laws.
___________________________________- 66 -_______________________________
The control law for elevation dynamics to ensure its stability from Eq. (5.1) can be
derived as
V ( x) x1 x1 x2 x2 x3 x3
(5.3)
x
x12
x2 x3 3 [(a1 x1 ) 2 b1 x1 B1 x3 Tg sin x2 ])
x
I1
( 1 K gyro x6 cos x2 ) T1
T1
(5.5)
From Eq. (5.5) we reach at singularity in the control law as motor angular speed x1 and
rate of change of azimuth x6 will be initially at zero.
Similarly for azimuth the control law can be derived as
V ( x) x4 x4 x5 x5 x6 x6
(5.6)
The control law for horizontal plane dynamics, after plugging Eq.(2.20) and Eq. (2.21),
comes out to be
1
1
V ( x) x4 ( ( x4 u2 )) x5 x6 x6 ( [(a2 x4 ) 2 b2 x4 - B2 x6 k1 x7 - k2u1 ])
T2
I2
u2
T2 1 2
1
( x4 x5 x6 x6 ( [(a2 x4 )2 b2 x4 - B2 x6 k1 x7 - k2u1 ]))
x4 T2
I2
(5.7)
(5.8)
Eq.(5.8) leads to the same conclusion as in elevation dynamics case. The singularity due
to side motor speed x4 0 initially forbids applying this control law.
The same conclusions are drawn from the Lyapunov functions defined as
1
V ( x) ( x22 x32 )
2
(5.9)
1
V ( x) ( x52 x62 )
2
(5.10)
Finally, we can conclude that various Lyapunov functions delivered number of control
laws but due to singularities in the design the laws are not applicable for implementation
on the twin rotor system. Moreover, the system is prone to uncertainties and disturbances,
the control laws achieved as a consequence of Lyapunov functions do not offer
___________________________________- 67 -_______________________________
robustness against them, thus formulating the basis to proceed for robust nonlinear
control synthesis procedures.
___________________________________- 68 -_______________________________
(5.11)
The positive definite Lyapunov function for sliding manifold can be chosen as
V ( s)
1 2
s1
2
(5.12)
(5.13)
s1 ( x) x2 x3
(5.14)
where
1
((a1 x1 ) 2 b1 x1 B1 x3 Tg sin x2 K gyro u1 x6 cos x2 )
I1
(5.15)
(5.16)
1
1
( x3 ( ((a1 x1 ) 2 b1 x1 B1 x3 Tg sin x2 )
K gyro x6 cos x2
I1
(5.17)
And the control law that will ensure the rate of change of Lyapunov function in Eq.(5.12)
to be negative definite can be written as
u1 ueq Ksign( s1 )
(5.18)
___________________________________- 69 -_______________________________
V ( s ) s1 Ksign( s1 )
(5.19)
The above designed control law poses singularity which serves as an obstacle to apply on
the helicopter model. The rate of change of azimuth ( x6 ) is at zero initially which will
make the equivalent control in Eq.(5.17) as undefined.
Similarly for azimuth dynamics the same practice fails to yield control law as in
Eq.(5.24) the control input is not appearing which helps in calculating the equivalent
control. To access the control input we will have to take another derivative of sliding
surface which is out of scope of first order sliding mode control.
s2 ( x) x6 2 x5
1 2
s2
2
(5.21)
V ( s ) s2 s2
(5.22)
s2 ( x) x6 2 x5
(5.23)
V (s)
s2 ( x) (
(5.20)
1
[(a2 x4 ) 2 b2 x4 - B2 x6 k1 x7 - k2u1 ]) 2 x6
I2
(5.24)
(5.25)
where X eq is the vector of desired values of the system states at equilibrium position. The
sliding manifolds in Eq. (5.26) are taken as Hurwitz polynomial of the states defined in
Eq. (2.16)
s e f (e , e )
S 1 3 1 2
s2 e6 f (e4 , e5 )
Where
(5.26)
f (e1 , e2 ) c1e1 c2 e2
f (e4 , e5 ) c4 e4 c5e5
___________________________________- 70 -_______________________________
(5.27)
The system in Eq. (5.27) will be stable if S 0 and the rate of convergence will be
governed by the manifold dynamics. The Lyapunov function for surfaces defined in Eq.
(5.26) can be written as
1 2
s1
2
1
V2 s2 2
2
V1
(5.28)
which are positive definite functions and their time derivative can be written as
V1 s1 s1
V2 s 2 s2
(5.29)
The equivalent control u1eq for elevation dynamics and u2eq for azimuth dynamics on the
manifold s1 c1e1 c2 e2 e3 0 and s2 c4 e4 c5e5 e6 0 respectively can be seen in
Eq. (5.30) and Eq. (5.31).
u1eq
T1 c1
1
[ ( x1 ) ([ ((a1 x1 ) 2 b1 x1 B1 x3 Tg sin x2 )] c2 x3 )]
c1 T1
I1
(5.30)
T2 c4
1
[ (- x4 ) c5 x6 ( ((a2 x4 ) 2 b2 x4 - B2 x6 ))]
c4 T2
I2
(5.31)
u2 eq
The control input vector U that will make the system to converge at S 0 can be
written as in Eq.(5.32), this control law will ensure the system convergence to sliding
manifold along with robustness against the cross-couplings.
u u1eq K1sign( s1 )
U 1
u2 u2 eq K 2 sign( s2 )
(5.32)
To avoid high frequency switching i.e. chattering, implementation of the control laws
have been performed by employing saturation function sat ( S ) defined as
___________________________________- 71 -_______________________________
s1
sat(s1 ) = sign(
sat(s1 ) =
s1
);
if abs(
s1
if abs(
s1
)>1
(5.33)
)<1
b1 b1 b1
For azimuth dynamics the perturbation and bounded uncertainties are taken as
b2 b2 b2 B 2 B2 B2
a2 a2 a2
The cross-coupling affects in the elevation and azimuth dynamics can be taken as
x3 x3 1
(5.34)
x6 x6 2
(5.35)
(5.36)
c
V2 s2 (b2 x4 (a2 x4 )2 B2 x6 4 x4 2 K 2 sat ( s2 ))
(5.37)
T1
T2
Now if
___________________________________- 72 -_______________________________
K1
c1
x B1x3 (a1x1 )2 b1x1 Tg sin x2 1
T1 1
K 2 b2 x4 (a2 x4 ) 2 B2 x6
c4
x4 2
T2
(5.38)
(5.39)
0 x2 0.4363
0 x 0.1
3
x
4 1.112
0.7 x5 0.7
0 x6 0.35
(5.40)
and with 50% perturbation in parameters defined in Table 2-1, we can compute that
K1 0.0011 for elevation dynamics and K 2 0.0018 for azimuth dynamics, that will
ensure
V1 s12
(5.41)
V2 s22
(5.42)
and
V1 in (5.41) and V2 in Eq. (5.42) will always be negative definite for non-zero
manifolds. The above conditions in Eq. (5.38) and Eq. (5.39) assures that the sliding
surface variables reach zero in finite time and once the trajectories are on the sliding
surface they remain on the surface, and approaches to the equilibrium points
asymptotically.
___________________________________- 73 -_______________________________
0.5
0.4
0.3
0.2
0.1
-0.1
5
Time (sec)
10
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
-0.05
0.05
0.1
0.15
0.2
0.25
0.3
Elevation
___________________________________- 74 -_______________________________
Phase Portait II
0.1
-0.1
-0.2
-0.3
-0.4
-0.5
-0.1
0.1
0.2
0.3
Azimuth
0.4
0.5
0.6
0.7
Elevation
0.2
Controller Effort
Azimuth
-0.2
-0.4
-0.6
-0.8
-1
5
Time (sec)
10
___________________________________- 75 -_______________________________
Sliding Surfaces
0.7
Elevation
Azimuth
0.6
0.5
Sliding surface
0.4
0.3
Elevation
0.2
Azimuth
0.1
-0.1
5
Time (sec)
10
___________________________________- 76 -_______________________________
Elevation (deg)
20
15
10
5
0
-5
10
15
20
25
10
15
20
25
30
35
40
45
50
30
35
40
45
50
Azimuth (deg)
20
10
0
-10
-20
Time (sec)
Figure 5-6: Response of Helicopter system with sliding mode controller when exposed to coupling at
22 seconds
Controller Effort
0.9
0.8
Elevation
0.7
0.6
0.5
0.4
0.3
0.2
10
15
20
25
10
15
20
25
30
35
40
45
50
30
35
40
45
50
1.2
Azimuth
1
0.8
0.6
0.4
0.2
Time (sec)
Figure 5-7: Sliding mode Controller effort to decouple when exposed to coupling at 22 seconds.
___________________________________- 77 -_______________________________
Elevation (deg)
20
15
10
5
0
-5
10
15
20
25
10
15
20
25
30
35
40
45
50
30
35
40
45
50
50
40
Azimuth (deg)
30
20
10
0
-10
-20
Time (sec)
Figure 5-8: Response of Actual system with sliding mode control when initialized in nonlinear range
Controller Effort
0.9
0.8
Elevation
0.7
0.6
0.5
0.4
0.3
0.2
10
15
20
25
10
15
20
25
30
35
40
45
50
30
35
40
45
50
1.5
Azimuth
0.5
Time (sec)
Figure 5-9: Sliding mode Controller effort to reach equilibrium position when released in nonlinear
range.
___________________________________- 78 -_______________________________
Elevation
Elevation
(H with
(H with
Traditional Wt)
Hadamard Wt)
ISE
46.14
24.85
24.05
ITSE
287.7
82.97
75.35
IAE
4.387
2.334
2.012
ITAE
44.92
18.01
10.56
Performance
Indices
Elevation
(SMC)
Azimuth
(H with
(H with
Traditional Wt)
Hadamard Wt)
ISE
0.4031
3.127
0.1956
ITSE
14.59
107.7
5.4
IAE
0.3114
0.8696
0.2671
ITAE
11.61
31.51
8.232
Performance
Indices
Azimuth
(SMC)
x Ax Bu
(5.43)
Where
___________________________________- 79 -_______________________________
A12
A
A 11
A21 A22
A11 ( n m )( n m )
0
B
Im
A22 mm
The A and B for helicopter model in Eq.(2.29) can be reformulated for Eq.(5.43) as
-0.9874 -7.1760 12.4544 4.0089 -0.2337 -7.1334 2.9235
0.8188
0
0
-0.4696 0.0209 -0.3296
0
0
0
-1.3330
0
0
0
0
0
0
B 0
0
1
0
0.4094 0
0
C
0
0
0.0105
0
0
0
0
0
0
1
0
0
0
0
-0.0060 -0.4091 0.0086 0
(5.45)
(5.46)
The sliding surface for the above described system can be defined as
x
s ( x) Sx [ S1 S2 ] 1 S1 x1 S 2 x2
x2
(5.47)
s Sx SAx SBu 0
(5.48)
If
Then equivalent control comes out to be
ueq ( SB) 1 SAx
(5.49)
(5.50)
___________________________________- 80 -_______________________________
S S2 M
Im
(5.51)
Where
M m( n m )
S 2 mm
x1 , x 2
(5.52)
x2 Mx1
(5.53)
(5.54)
(5.55)
Thus
(5.56)
Where does not belong to current eigenvalues of the system. Our system in new
coordinates will be given as
A ( I n A) A( I n A) 1
(5.57)
B ( I n A) B
(5.58)
L M
I m ( I n A)( I n A) 1
S M
I m ( I n A) 1
(5.59)
(5.60)
The close loop system resulting in the form of S will have to minimize
J xT ( )Qx( ) u T ( ) Ru ( )d
(5.61)
Where
Q n n
R mm
The Lyapunov stability that will guarantee the system stability in new coordinates can
written as
___________________________________- 81 -_______________________________
)T ( A BL
)) X X (QT Q LT RL ) X
X (( A BL
q q
(5.62)
)T ( A BL
) X XQT Q X XLT RLX
0
X ( A BL
q q
(5.63)
Qq X
R1/2 LX
XQqT
Iq
0
XLT R1/2
0 0
I m
(5.64)
(5.65)
0
0
X2
(5.66)
And
X 1 ( n m )( n m )
X 2 mm
The LMI in Eq.(5.64) has two solution variables therefore we will have to formulated two
LMIs that will deliver solution in two variables. So the new problem LMI can be
formulated as
(5.67)
Minimize Trace( Z )
Subject to
( BN
)T
AX XAT BN
Qq X
R1/2 N
( I n A)
XQqT
Iq
0
N T R1/2
0 0
I m
( I n A)T
0
X
(5.68)
(5.69)
Where
N
N LX
1
X2
(5.70)
___________________________________- 82 -_______________________________
L NX 1 [ N1 X 11
Im ]
(5.71)
M N1 X 11
(5.72)
5.2.3.1
I m ( I n A)
(5.73)
Simulation Results
The feedback controller based on sliding surface delivers the results for regulation
control as shown in Figure 5-10. The parameter values are taken as
-3
Q 30 C
R Im
The controller effort to meet the desired regulation results is shown in Figure 5-11 and
the sliding surface convergence is shown in Figure 5-12.
Helicopter
0.06
Elevation
Azimuth
0.04
0.02
-0.02
-0.04
-0.06
10
Time (sec)
Figure 5-10: Output States response for LMI based Sliding mode control
___________________________________- 83 -_______________________________
Controller Effort
15
Control Effort
10
0
0
5
Time (sec)
10
Sliding Surface
50
40
30
20
10
5
Time (sec)
10
The controller has to exert more effort in order to attain the desired results. This is
because the designed controller is based on input and output energy of the system which
relates to H2 norm and the controller based on H2 norm have exert more effort to deliver
results. This reason forbids us to apply the designed controller on actual system and
restricts us to simulations only.
mode feature and improving its accuracy. Traditional SMC has some intrinsic problems,
such as discontinuous control that often yields chattering [6]. To cope with problem and
achieve higher accuracy, HOSM is proposed [24], [25]. Obviously, kth order HOSM
stabilizes the sliding variable at zero as well as its derivatives. On the other hand, since
the high-frequency switching is hidden in the higher derivative of the sliding variable, the
effect of chattering will be reduced. In other words, HOSM has two important features
that make it a better choice in designing the controller. It improves the accuracy of the
design, which is a very important issue, and may provide a continuous control.
(5.74)
where X eq are the desired values of the system states at equilibrium position The sliding
surface with r 1 for elevation and azimuth can be designed as in Eq.(5.75) and
Eq.(5.76) respectively.
1
e2 1e2 0 e2
(5.75)
2 e5 3e5 2 e5
(5.76)
1
e2 1e2 0 e2
(5.77)
2
e5 3
e5 2 e5
(5.78)
___________________________________- 85 -_______________________________
Y1 21 ; Y1 1 2
Y2 21 ; Y2 1 2
(5.79)
(5.80)
The finite time stabilization problem for the uncertain second order system equivalent to
second order sliding mode control problem which can be written as [].
Y1 Y2
Y2 (t , X ) (t , X )U
(5.81)
21 ; 1 2
Where
21 ; 1 2
U 21 ;U u1 u2
1 ( x, t ) 1 ( x, t ) c2
e3 c1e3
1 ( x, t )
(5.82)
( x u )
1
[{4a1 x1 2a1u1 b1} 1 2 1 B1
x3 (Tg g )( x3 cos x2 x32 sin x2 )] (5.83)
I1
T1
1 ( x, t )
1 2a1 x1 b1
(
)
I1 T1
T1
(5.84)
and
2 ( x, t ) 2 ( x, t ) c2
e6 c1e6
2 ( x, t )
1
I2
({
4a2 x4
2
T2
2 a2
2
T2
2 ( x, t ) (
u2
2 a2
T2
b2
T22
x4
}(- x4 u2 ) B2 x6 )
b2
T2
(5.85)
(5.86)
(5.87)
0 x3 0.1
0 x4 1.112
0.7 x5 0.7
0 x6 0.35
(5.88)
___________________________________- 86 -_______________________________
and all other constants have bounded values which can be found in [7]. Therefore, we can
claim that there exists 1,2 0
0 m1,2 1,2 ( x, t ) M 1,2 are satisfied [31]. The computed bounds of helicopter model
for
elevation
are
1 0.75 ,
m1 7.13 ,
M 1 96.20
and
for
azimuth
i 1, 2
(5.89)
where
ui '(t ) i i sign( i )
u (t )
ui ''(t ) i
i sign( i )
i 1, 2
if
ui 1
if
ui 1
The constants ( i , i ) values [31] are computed keeping in mind the bounds defined as
i
i
mi
2 4 i M i ( i i )
i 2 ( )
mi
i
i
mi
0 0.5
(5.90)
Finally for elevation we have the values as 1 2.31 , 1 0.105 and for azimuth we
have 2 0.053 , 2 0.021 .
The designed control law in Eq.(5.89) and constants bounds computed in
Eq.(5.90) , the sliding variables 1 and 2 will stabilize at zero in finite time assuming
that the uncertainties ( g ) in the parameters are bounded. The chattering effect in the
control law has been filtered out by integrating discontinuous portion caused by the sign
term making control law continuous in nature.
___________________________________- 87 -_______________________________
0.2
0.15
0.1
0.05
10
15
Time (sec)
0.1
-0.1
-0.2
-0.3
-0.4
-0.1
-0.05
0.05
0.1
Angular Position
0.15
0.2
0.25
___________________________________- 88 -_______________________________
20
15
10
10
20
30
40
50
60
Time (sec)
70
80
90
100
110
Figure 5-15: Response of helicopter model in Elevation when exposed uncertainty in center of gravity
___________________________________- 89 -_______________________________
Center of Gravity
1
0.8
0.6
0.4
CoG
0.2
-0.2
-0.4
-0.6
-0.8
-1
10
20
30
40
50
60
Time (sec)
70
80
90
100
110
In the second case when the CG is perturbed 70% behind of its normal position at 69 sec.
the resultant torque forces helicopter model to fly with nose tilted up and the 17%
deviation is nullified by the designed controller in 7 sec. and vice versa when CG again
attains its normal position.
Elevation
1.5
Controller Effort
0.5
0
10
20
30
40
50
60
Time (sec)
70
80
90
100
110
Figure 5-17: 2-SMCController effort in voltage to acquire and maintain equilibrium position in
elevation plane
___________________________________- 90 -_______________________________
0.8
0.6
0.4
Sliding Surface
0.2
-0.2
-0.4
-0.6
-0.8
-1
10
20
30
40
50
60
Time (sec)
70
80
90
100
110
The control law in Eq.(5.89) with 2=1.1, 2=0.5 for azimuth dynamics delivered results
in horizontal plane shown in Figure 5-19 with the respective controller effort in Figure
5-20. The azimuth dynamics sliding surface convergence is shown in Figure 5-21. It can
be observed that the helicopter when released from 35 deg. in azimuth converges to
referred position in 15 sec. and maintains the position afterwards. The horizontal plane
dynamics remain unaffected by the perturbation in center of gravity.
The control inputs for both the elevation and azimuth dynamics are having
chattering in their response, however this phenomena is not really dangerous for the
system as the magnitude are well under the maximum limits of the inputs. However these
responses of control laws signify that the 2-SMC super twisting algorithm certainly
reduces the chattering but fails to remove it. Although, the reduction in chattering can be
achieved by altering in Eq.(5.89) or by boundary layer in the discontinuous control
part [] but as a consequence the performance of the system gets reduced.
___________________________________- 91 -_______________________________
Azimuth
35
Reference
Azimuth
30
25
20
15
10
-5
-10
10
20
30
40
Time (sec)
50
60
70
Figure 5-19 Response of helicopter model with 2-SMC controller in horizontal plane
Azimuth
0.8
0.6
Controller Effort
0.4
0.2
-0.2
-0.4
-0.6
10
20
30
40
Time (sec)
50
60
70
80
___________________________________- 92 -_______________________________
0.4
0.2
Sliding Surface
-0.2
-0.4
-0.6
-0.8
-1
10
20
30
40
Time (sec)
50
60
70
This section presented successful solution for the precise helicopter maneuvers in the
presence of unwanted moments caused by variations in the on board loaded equipments
in the fuselage. The simulations and practical implementation of the control algorithm
guarantees the smooth helicopter flight even with unbalance dynamics caused by the
perturbed CG.
___________________________________- 93 -_______________________________
Chapter 6
Conclusion
&
Future Work
___________________________________- 94 -_______________________________
This thesis contained several attempts to design and implement robust controllers
for helicopter systems which can handle cross-coupling along with parametric uncertainty
and guarantee smooth helicopter flight. In the first phase, H controllers were designed
based on traditional and Hadamard weights with loop shaping design techniques. The
main issue encountered during the design procedure was the selection of weights for loop
shaping of singular values of helicopter system. After number of attempts suitable
weights were selected, which transformed the loop gains such that the singular values at
lower frequencies were with high gains and with low gains at higher frequencies. The
main problem was faced in reduction of the slope of singular values at crossover
frequencies to 1 db/decade. After number of attempts the weights were chosen, but the
validation of the selected weights was carried through practical implementation although
the simulation results were not supportive in terms of performance parameters. The
controller was implemented in SIMULINK environment where the state space model of
the controller was engaged with system to achieve desired results. The controller
provided robustness against the cross-couplings and uncertainty in the CG of the
helicopter model; however the performance declined as a result. Hadamard weighted H
controller failed to operate in nonlinear range, conversely traditional weighted H
successfully guided the helicopter model to equilibrium position when released in
nonlinear range.
The limitations of linear theory founded the base for nonlinear control design,
sliding mode control and its advance algorithms were considered to overcome limitation
of linear domain. Sliding manifold designing involved the availability of rotor speeds,
which was extracted from Luenberger observer. The second reservation for this technique
resides in defining required rotor speeds in control law, as a consequence the coupling
reduction was achieved but robustness against CG uncertainty vanished. This drawback
in the 1-SMC made to move towards higher sliding mode control i.e. 2-SMC, where the
manifolds convergence guaranteed the robustness against CG uncertainty.
The implementation of controllers was carried out in SIMULINK, which made it
easy to design and implement all controllers. As a consequence some of the practical
issues were overshadowed like sampling time issue in SMC and implementation of
difference equations of H controller.
___________________________________- 95 -_______________________________
Controllers Decoupling
Robust
Against
Range
CG
(Azimuth)
(sec)
(10%)
Time (sec)
Variation
Ele
Az
Ele
Az
Ele
Az
PID
No
Yes
Yes
25
Yes
No
30
25
Traditional
Yes
Yes
Yes
20
Yes
No
25
23
Yes
Yes
No
10 N/A Yes
N/A
15 N/A
1-SMC
Yes
No
Yes
10
10
Yes
Yes
10
10
2-SMC
No
Yes
Yes
10
20
Yes
Yes
26
26
Weighted H
Hadamard
Weighted H
After going through the efforts to produce this thesis several other areas also
require attention in future. The areas from controller design may include decoupled
controller using Higher Order Sliding mode control. Although we have attempted to
solve cross-coupling using this technique but the results show that more effort is needed
to put in to get to the solution.
The other area that needs consideration is parameter estimation using nonlinear
techniques. Although, the accompanied manual has these parameters estimated for the
mathematical model using optimization techniques, but these parameters can be
estimated more accurately using techniques from nonlinear control theory like sliding
mode observers, higher order sliding mode observers.
Similarly, the concepts of fault diagnostics can also be validated on this model. The
fault estimation involves the system output and controller effort, therefore fault
diagnostics is an area that can be explored further on the basis of current efforts.
___________________________________- 96 -_______________________________
___________________________________- 97 -_______________________________
Chapter 7
Appendix
___________________________________- 98 -_______________________________
___________________________________- 99 -_______________________________
0 0];
Gss=ss(A,B,C,D);
% Conversion to Transfer function from State Space model
[NUM1,DEN1]=ss2tf(A,B,C,D,1);
[NUM2,DEN2]=ss2tf(A,B,C,D,2);
[r11 c11]=size(NUM1);
[r12 c12]=size(DEN1);
num11=NUM1(1,1:c11);
den11=DEN1(1,1:c12);
num21=NUM1(2,1:c11);
den21=DEN1(1,1:c12);
[r21 c21]=size(NUM2);
[r22 c22]=size(DEN2);
num12=NUM2(1,1:c21);
den12=DEN2(1,1:c22);
num22=NUM2(2,1:c21);
den22=DEN1(1,1:c22);
g11=tf(num11,den11);
g21=tf(num21,den21);
g12=tf(num12,den12);
g22=tf(num22,den22);
G=[g11 g12;g21 g22];
0.001],[1 1000]))
0;0
0.001],[1 1000]))
0;0
num21c=NUM1c(2,1:c11c);
den21c=DEN1c(1,1:c12c);
[r21c c21c]=size(NUM2c);
[r22c c22c]=size(DEN2c);
num12c=NUM2c(1,1:c21c);
den12c=DEN2c(1,1:c22c);
num22c=NUM2c(2,1:c21c);
den22c=DEN1c(1,1:c22c);
g11c=tf(num11c,den11c);
g21c=tf(num21c,den21c);
g12c=tf(num12c,den12c);
g22c=tf(num22c,den22c);
Gc=[g11c g12c;g21c g22c];
Kco=Gc.*W1;
Kcod=c2d(Kco,0.01,'tustin');
[Aco,Bco,Cco,Dco]=ssdata(Kco);
[Acdo,Bcdo,Ccdo,Dcdo]=ssdata(Kcod);
T2=feedback(Kco.*G,eye(2));
S2=1-T2;
figure(5),step(T2,20);grid;title('Step
response');xlabel('Time');ylabel('Angle (deg)');
figure(6),step(Kco*S2,20);grid;title('Control
input');xlabel('Time');ylabel('Voltage (V)');
pi/180
pi/180
Az
-40
Ele
14
Clock
1
G2
G
In1 Out1
Out3
Out2
Out1
Disturbance
Ini
To Workspace4
Controller1
(n)=Cx(n)+Du(n
n+1)=Ax(n)+Bu(
Az1
Scope2
Scope3
To Workspace2
CG3
CG2
To Workspace3
Ele1
Helicopter
Scope1
r2d
-K-
r to d
-K-
To Workspace
Az
Scope
To Workspace1
Ele
r2d
r to d
CG2
-K-
-K-
Scope2
CG3
Ini
Out3
Out2
Out1
u2
x6
Controller
u1
x2
G2
In1 Out1
Disturbance
Scope3
Helicopter
Clock
Scope1
To Workspace4
To Workspace3
Az1
To Workspace
Az
Scope
To Workspace1
Ele
To Workspace2
Ele1
7.5
x2
u1
x7
x2
u1
Elevation
x6
u2
u1
x7
x6
u2
Azimuth
0.34
2
Constant
K Ts
x1
z-1
int
-K+3.33
Sign
-K1
x2
x1
s
Gain1
x2
1
K (z-1)
Ts z
0.001
u1
s/e
-K-
|u|
Abs
Gain
Switch
u1
x3
der
v1
-K3.33
u
x1^2
K-
1/I1
1/2
-K-
Product
1/I2
-Kf(u)
B1
sin
sin(x2)
Fcn
-K1/I4
2
x7
2
K Ts
x5
z-1
int1
-K-
.44
-4
Sign
Constant
x5
s
1
x6
u2
s/e
-K-
x6
K (z-1)
Ts z
Abs
Gain
-K- Gain1
x7
|u|
u2
Switch
2
x7
der
v1
-K2
3.33
u
x1^2
-K-
K-
u1
K Ts x4
-K-
z-1
int
B1
-K-
-K-
1/I1
-1
1/2
1/I2
-K-
-K-
B2
-3
-K-2
Clock
To Workspace4
Controller
x6
x2
1
G2
G
Ini
Out3
Out2
Out1
Disturbance
In1 Out1
CG3
CG2
Scope2
Scope3
Helicopter
r to d
-K-
To Workspace2
r2d
-K-
Az1
To Workspace3
Ele1
Scope1
To Workspace
Az
Scope
To Workspace1
Ele
Sign
1
|u|
sqrt
Abs2
Math
Function
x2
x2
-0.5
-K- -0.2
K (z-1)
Ts z
-Klambda1
sqrt(abs(s1))*sign(s1)
Add4
-K-1
x3
der
u
K Ts
-KSwitch2
-0.01
z-1
int
Sign1
2
x6
|u|
-0.15
sqrt
-K-
x6
Abs1
-K- -0.3
x7
lambda2
sqrt(abs(s1))*sign(s1)1
Math
Function1
Add1
-K-
K (z-1)
Ts z
-.1
der1
K Ts
-K-0.1
Switch1
z-1
int1
BIBLIOGRAPHY
[1]. Gareth D. Padfield, Helicopter Flight Dynamics The Theory and Application of
Flying Qualities and Simulation Modeling Second Edition Blackwell Publishing
[2]. Basic
Helicopter
Handbook,
Federal
Aviation
Handbook,
ISBN
13:9781560270041, 1978
[3]. Markus Kemper and Sergej Fatikow, Impact Of Center Of Gravity In Quadrotor
Helicopter Controller Design, Mechatronic Systems, Vol. 4, Part 1, 2007
and Control and European Control Conference 2005, Seville Spain 2005..
[16]. Gwo-R.Y, H.T.Lui, Sliding mode control of a two degree of freedom helicopter
via Linear Quadratic regulator Proc. of IEEE International Conference on
Systems, Man and Cybernetics, 10-12 Oct. 2005, Waikoloa, Hawaii
[17]. D.W. Gu, P. Hr. Petkov and M. M. Konstantinov, Robust Control Design with
MATLAB Springer, ISBN-10: 1852339837
[18]. Gene Franklin, J.D. Powell, Abbas Emami-Naeini, Feedback Control of Dynamic
Systems, Prentice Hall; 5th edition, ISBN-10: 0131499300.
[20]. Mircea Lazr, Robin De Keyser, Nonlinear Predictive Control, Research Report,
University of Ghent, Department of Control Engineering and Automation
[21]. Arkadiusz S. Dutka, Andrzej W.Ordys, Michael J.Grimble Non-linear Predictive
Control of 2 DOF helicopter model Proc. of the 42nd IEEE Conference on
[27]. S. Laghrouche, M. Smaoui, X. Brun and F. Plestan, Second order sliding mode
controllers for pneumatic actuators, in American Control Conference ACC04,